1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* Common capabilities, needed by capability.o. 3 */ 4 5 #include <linux/capability.h> 6 #include <linux/audit.h> 7 #include <linux/init.h> 8 #include <linux/kernel.h> 9 #include <linux/lsm_hooks.h> 10 #include <linux/file.h> 11 #include <linux/mm.h> 12 #include <linux/mman.h> 13 #include <linux/pagemap.h> 14 #include <linux/swap.h> 15 #include <linux/skbuff.h> 16 #include <linux/netlink.h> 17 #include <linux/ptrace.h> 18 #include <linux/xattr.h> 19 #include <linux/hugetlb.h> 20 #include <linux/mount.h> 21 #include <linux/sched.h> 22 #include <linux/prctl.h> 23 #include <linux/securebits.h> 24 #include <linux/user_namespace.h> 25 #include <linux/binfmts.h> 26 #include <linux/personality.h> 27 #include <linux/mnt_idmapping.h> 28 #include <uapi/linux/lsm.h> 29 30 #define CREATE_TRACE_POINTS 31 #include <trace/events/capability.h> 32 33 /* 34 * If a non-root user executes a setuid-root binary in 35 * !secure(SECURE_NOROOT) mode, then we raise capabilities. 36 * However if fE is also set, then the intent is for only 37 * the file capabilities to be applied, and the setuid-root 38 * bit is left on either to change the uid (plausible) or 39 * to get full privilege on a kernel without file capabilities 40 * support. So in that case we do not raise capabilities. 41 * 42 * Warn if that happens, once per boot. 43 */ 44 static void warn_setuid_and_fcaps_mixed(const char *fname) 45 { 46 static int warned; 47 if (!warned) { 48 printk(KERN_INFO "warning: `%s' has both setuid-root and" 49 " effective capabilities. Therefore not raising all" 50 " capabilities.\n", fname); 51 warned = 1; 52 } 53 } 54 55 /** 56 * cap_capable_helper - Determine whether a task has a particular effective 57 * capability. 58 * @cred: The credentials to use 59 * @target_ns: The user namespace of the resource being accessed 60 * @cred_ns: The user namespace of the credentials 61 * @cap: The capability to check for 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 * See cap_capable for more details. 67 */ 68 static inline int cap_capable_helper(const struct cred *cred, 69 struct user_namespace *target_ns, 70 const struct user_namespace *cred_ns, 71 int cap) 72 { 73 struct user_namespace *ns = target_ns; 74 75 /* See if cred has the capability in the target user namespace 76 * by examining the target user namespace and all of the target 77 * user namespace's parents. 78 */ 79 for (;;) { 80 /* Do we have the necessary capabilities? */ 81 if (likely(ns == cred_ns)) 82 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM; 83 84 /* 85 * If we're already at a lower level than we're looking for, 86 * we're done searching. 87 */ 88 if (ns->level <= cred_ns->level) 89 return -EPERM; 90 91 /* 92 * The owner of the user namespace in the parent of the 93 * user namespace has all caps. 94 */ 95 if ((ns->parent == cred_ns) && uid_eq(ns->owner, cred->euid)) 96 return 0; 97 98 /* 99 * If you have a capability in a parent user ns, then you have 100 * it over all children user namespaces as well. 101 */ 102 ns = ns->parent; 103 } 104 105 /* We never get here */ 106 } 107 108 /** 109 * cap_capable - Determine whether a task has a particular effective capability 110 * @cred: The credentials to use 111 * @target_ns: The user namespace of the resource being accessed 112 * @cap: The capability to check for 113 * @opts: Bitmask of options defined in include/linux/security.h (unused) 114 * 115 * Determine whether the nominated task has the specified capability amongst 116 * its effective set, returning 0 if it does, -ve if it does not. 117 * 118 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable() 119 * and has_capability() functions. That is, it has the reverse semantics: 120 * cap_has_capability() returns 0 when a task has a capability, but the 121 * kernel's capable() and has_capability() returns 1 for this case. 122 */ 123 int cap_capable(const struct cred *cred, struct user_namespace *target_ns, 124 int cap, unsigned int opts) 125 { 126 const struct user_namespace *cred_ns = cred->user_ns; 127 int ret = cap_capable_helper(cred, target_ns, cred_ns, cap); 128 129 trace_cap_capable(cred, target_ns, cred_ns, cap, ret); 130 return ret; 131 } 132 133 /** 134 * cap_settime - Determine whether the current process may set the system clock 135 * @ts: The time to set 136 * @tz: The timezone to set 137 * 138 * Determine whether the current process may set the system clock and timezone 139 * information, returning 0 if permission granted, -ve if denied. 140 */ 141 int cap_settime(const struct timespec64 *ts, const struct timezone *tz) 142 { 143 if (!capable(CAP_SYS_TIME)) 144 return -EPERM; 145 return 0; 146 } 147 148 /** 149 * cap_ptrace_access_check - Determine whether the current process may access 150 * another 151 * @child: The process to be accessed 152 * @mode: The mode of attachment. 153 * 154 * If we are in the same or an ancestor user_ns and have all the target 155 * task's capabilities, then ptrace access is allowed. 156 * If we have the ptrace capability to the target user_ns, then ptrace 157 * access is allowed. 158 * Else denied. 159 * 160 * Determine whether a process may access another, returning 0 if permission 161 * granted, -ve if denied. 162 */ 163 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) 164 { 165 int ret = 0; 166 const struct cred *cred, *child_cred; 167 const kernel_cap_t *caller_caps; 168 169 rcu_read_lock(); 170 cred = current_cred(); 171 child_cred = __task_cred(child); 172 if (mode & PTRACE_MODE_FSCREDS) 173 caller_caps = &cred->cap_effective; 174 else 175 caller_caps = &cred->cap_permitted; 176 if (cred->user_ns == child_cred->user_ns && 177 cap_issubset(child_cred->cap_permitted, *caller_caps)) 178 goto out; 179 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) 180 goto out; 181 ret = -EPERM; 182 out: 183 rcu_read_unlock(); 184 return ret; 185 } 186 187 /** 188 * cap_ptrace_traceme - Determine whether another process may trace the current 189 * @parent: The task proposed to be the tracer 190 * 191 * If parent is in the same or an ancestor user_ns and has all current's 192 * capabilities, then ptrace access is allowed. 193 * If parent has the ptrace capability to current's user_ns, then ptrace 194 * access is allowed. 195 * Else denied. 196 * 197 * Determine whether the nominated task is permitted to trace the current 198 * process, returning 0 if permission is granted, -ve if denied. 199 */ 200 int cap_ptrace_traceme(struct task_struct *parent) 201 { 202 int ret = 0; 203 const struct cred *cred, *child_cred; 204 205 rcu_read_lock(); 206 cred = __task_cred(parent); 207 child_cred = current_cred(); 208 if (cred->user_ns == child_cred->user_ns && 209 cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) 210 goto out; 211 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) 212 goto out; 213 ret = -EPERM; 214 out: 215 rcu_read_unlock(); 216 return ret; 217 } 218 219 /** 220 * cap_capget - Retrieve a task's capability sets 221 * @target: The task from which to retrieve the capability sets 222 * @effective: The place to record the effective set 223 * @inheritable: The place to record the inheritable set 224 * @permitted: The place to record the permitted set 225 * 226 * This function retrieves the capabilities of the nominated task and returns 227 * them to the caller. 228 */ 229 int cap_capget(const struct task_struct *target, kernel_cap_t *effective, 230 kernel_cap_t *inheritable, kernel_cap_t *permitted) 231 { 232 const struct cred *cred; 233 234 /* Derived from kernel/capability.c:sys_capget. */ 235 rcu_read_lock(); 236 cred = __task_cred(target); 237 *effective = cred->cap_effective; 238 *inheritable = cred->cap_inheritable; 239 *permitted = cred->cap_permitted; 240 rcu_read_unlock(); 241 return 0; 242 } 243 244 /* 245 * Determine whether the inheritable capabilities are limited to the old 246 * permitted set. Returns 1 if they are limited, 0 if they are not. 247 */ 248 static inline int cap_inh_is_capped(void) 249 { 250 /* they are so limited unless the current task has the CAP_SETPCAP 251 * capability 252 */ 253 if (cap_capable(current_cred(), current_cred()->user_ns, 254 CAP_SETPCAP, CAP_OPT_NONE) == 0) 255 return 0; 256 return 1; 257 } 258 259 /** 260 * cap_capset - Validate and apply proposed changes to current's capabilities 261 * @new: The proposed new credentials; alterations should be made here 262 * @old: The current task's current credentials 263 * @effective: A pointer to the proposed new effective capabilities set 264 * @inheritable: A pointer to the proposed new inheritable capabilities set 265 * @permitted: A pointer to the proposed new permitted capabilities set 266 * 267 * This function validates and applies a proposed mass change to the current 268 * process's capability sets. The changes are made to the proposed new 269 * credentials, and assuming no error, will be committed by the caller of LSM. 270 */ 271 int cap_capset(struct cred *new, 272 const struct cred *old, 273 const kernel_cap_t *effective, 274 const kernel_cap_t *inheritable, 275 const kernel_cap_t *permitted) 276 { 277 if (cap_inh_is_capped() && 278 !cap_issubset(*inheritable, 279 cap_combine(old->cap_inheritable, 280 old->cap_permitted))) 281 /* incapable of using this inheritable set */ 282 return -EPERM; 283 284 if (!cap_issubset(*inheritable, 285 cap_combine(old->cap_inheritable, 286 old->cap_bset))) 287 /* no new pI capabilities outside bounding set */ 288 return -EPERM; 289 290 /* verify restrictions on target's new Permitted set */ 291 if (!cap_issubset(*permitted, old->cap_permitted)) 292 return -EPERM; 293 294 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ 295 if (!cap_issubset(*effective, *permitted)) 296 return -EPERM; 297 298 new->cap_effective = *effective; 299 new->cap_inheritable = *inheritable; 300 new->cap_permitted = *permitted; 301 302 /* 303 * Mask off ambient bits that are no longer both permitted and 304 * inheritable. 305 */ 306 new->cap_ambient = cap_intersect(new->cap_ambient, 307 cap_intersect(*permitted, 308 *inheritable)); 309 if (WARN_ON(!cap_ambient_invariant_ok(new))) 310 return -EINVAL; 311 return 0; 312 } 313 314 /** 315 * cap_inode_need_killpriv - Determine if inode change affects privileges 316 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV 317 * 318 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV 319 * affects the security markings on that inode, and if it is, should 320 * inode_killpriv() be invoked or the change rejected. 321 * 322 * Return: 1 if security.capability has a value, meaning inode_killpriv() 323 * is required, 0 otherwise, meaning inode_killpriv() is not required. 324 */ 325 int cap_inode_need_killpriv(struct dentry *dentry) 326 { 327 struct inode *inode = d_backing_inode(dentry); 328 int error; 329 330 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); 331 return error > 0; 332 } 333 334 /** 335 * cap_inode_killpriv - Erase the security markings on an inode 336 * 337 * @idmap: idmap of the mount the inode was found from 338 * @dentry: The inode/dentry to alter 339 * 340 * Erase the privilege-enhancing security markings on an inode. 341 * 342 * If the inode has been found through an idmapped mount the idmap of 343 * the vfsmount must be passed through @idmap. This function will then 344 * take care to map the inode according to @idmap before checking 345 * permissions. On non-idmapped mounts or if permission checking is to be 346 * performed on the raw inode simply pass @nop_mnt_idmap. 347 * 348 * Return: 0 if successful, -ve on error. 349 */ 350 int cap_inode_killpriv(struct mnt_idmap *idmap, struct dentry *dentry) 351 { 352 int error; 353 354 error = __vfs_removexattr(idmap, dentry, XATTR_NAME_CAPS); 355 if (error == -EOPNOTSUPP) 356 error = 0; 357 return error; 358 } 359 360 static bool rootid_owns_currentns(vfsuid_t rootvfsuid) 361 { 362 struct user_namespace *ns; 363 kuid_t kroot; 364 365 if (!vfsuid_valid(rootvfsuid)) 366 return false; 367 368 kroot = vfsuid_into_kuid(rootvfsuid); 369 for (ns = current_user_ns();; ns = ns->parent) { 370 if (from_kuid(ns, kroot) == 0) 371 return true; 372 if (ns == &init_user_ns) 373 break; 374 } 375 376 return false; 377 } 378 379 static __u32 sansflags(__u32 m) 380 { 381 return m & ~VFS_CAP_FLAGS_EFFECTIVE; 382 } 383 384 static bool is_v2header(int size, const struct vfs_cap_data *cap) 385 { 386 if (size != XATTR_CAPS_SZ_2) 387 return false; 388 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2; 389 } 390 391 static bool is_v3header(int size, const struct vfs_cap_data *cap) 392 { 393 if (size != XATTR_CAPS_SZ_3) 394 return false; 395 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3; 396 } 397 398 /* 399 * getsecurity: We are called for security.* before any attempt to read the 400 * xattr from the inode itself. 401 * 402 * This gives us a chance to read the on-disk value and convert it. If we 403 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler. 404 * 405 * Note we are not called by vfs_getxattr_alloc(), but that is only called 406 * by the integrity subsystem, which really wants the unconverted values - 407 * so that's good. 408 */ 409 int cap_inode_getsecurity(struct mnt_idmap *idmap, 410 struct inode *inode, const char *name, void **buffer, 411 bool alloc) 412 { 413 int size; 414 kuid_t kroot; 415 vfsuid_t vfsroot; 416 u32 nsmagic, magic; 417 uid_t root, mappedroot; 418 char *tmpbuf = NULL; 419 struct vfs_cap_data *cap; 420 struct vfs_ns_cap_data *nscap = NULL; 421 struct dentry *dentry; 422 struct user_namespace *fs_ns; 423 424 if (strcmp(name, "capability") != 0) 425 return -EOPNOTSUPP; 426 427 dentry = d_find_any_alias(inode); 428 if (!dentry) 429 return -EINVAL; 430 size = vfs_getxattr_alloc(idmap, dentry, XATTR_NAME_CAPS, &tmpbuf, 431 sizeof(struct vfs_ns_cap_data), GFP_NOFS); 432 dput(dentry); 433 /* gcc11 complains if we don't check for !tmpbuf */ 434 if (size < 0 || !tmpbuf) 435 goto out_free; 436 437 fs_ns = inode->i_sb->s_user_ns; 438 cap = (struct vfs_cap_data *) tmpbuf; 439 if (is_v2header(size, cap)) { 440 root = 0; 441 } else if (is_v3header(size, cap)) { 442 nscap = (struct vfs_ns_cap_data *) tmpbuf; 443 root = le32_to_cpu(nscap->rootid); 444 } else { 445 size = -EINVAL; 446 goto out_free; 447 } 448 449 kroot = make_kuid(fs_ns, root); 450 451 /* If this is an idmapped mount shift the kuid. */ 452 vfsroot = make_vfsuid(idmap, fs_ns, kroot); 453 454 /* If the root kuid maps to a valid uid in current ns, then return 455 * this as a nscap. */ 456 mappedroot = from_kuid(current_user_ns(), vfsuid_into_kuid(vfsroot)); 457 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) { 458 size = sizeof(struct vfs_ns_cap_data); 459 if (alloc) { 460 if (!nscap) { 461 /* v2 -> v3 conversion */ 462 nscap = kzalloc(size, GFP_ATOMIC); 463 if (!nscap) { 464 size = -ENOMEM; 465 goto out_free; 466 } 467 nsmagic = VFS_CAP_REVISION_3; 468 magic = le32_to_cpu(cap->magic_etc); 469 if (magic & VFS_CAP_FLAGS_EFFECTIVE) 470 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; 471 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 472 nscap->magic_etc = cpu_to_le32(nsmagic); 473 } else { 474 /* use allocated v3 buffer */ 475 tmpbuf = NULL; 476 } 477 nscap->rootid = cpu_to_le32(mappedroot); 478 *buffer = nscap; 479 } 480 goto out_free; 481 } 482 483 if (!rootid_owns_currentns(vfsroot)) { 484 size = -EOVERFLOW; 485 goto out_free; 486 } 487 488 /* This comes from a parent namespace. Return as a v2 capability */ 489 size = sizeof(struct vfs_cap_data); 490 if (alloc) { 491 if (nscap) { 492 /* v3 -> v2 conversion */ 493 cap = kzalloc(size, GFP_ATOMIC); 494 if (!cap) { 495 size = -ENOMEM; 496 goto out_free; 497 } 498 magic = VFS_CAP_REVISION_2; 499 nsmagic = le32_to_cpu(nscap->magic_etc); 500 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE) 501 magic |= VFS_CAP_FLAGS_EFFECTIVE; 502 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 503 cap->magic_etc = cpu_to_le32(magic); 504 } else { 505 /* use unconverted v2 */ 506 tmpbuf = NULL; 507 } 508 *buffer = cap; 509 } 510 out_free: 511 kfree(tmpbuf); 512 return size; 513 } 514 515 /** 516 * rootid_from_xattr - translate root uid of vfs caps 517 * 518 * @value: vfs caps value which may be modified by this function 519 * @size: size of @ivalue 520 * @task_ns: user namespace of the caller 521 */ 522 static vfsuid_t rootid_from_xattr(const void *value, size_t size, 523 struct user_namespace *task_ns) 524 { 525 const struct vfs_ns_cap_data *nscap = value; 526 uid_t rootid = 0; 527 528 if (size == XATTR_CAPS_SZ_3) 529 rootid = le32_to_cpu(nscap->rootid); 530 531 return VFSUIDT_INIT(make_kuid(task_ns, rootid)); 532 } 533 534 static bool validheader(size_t size, const struct vfs_cap_data *cap) 535 { 536 return is_v2header(size, cap) || is_v3header(size, cap); 537 } 538 539 /** 540 * cap_convert_nscap - check vfs caps 541 * 542 * @idmap: idmap of the mount the inode was found from 543 * @dentry: used to retrieve inode to check permissions on 544 * @ivalue: vfs caps value which may be modified by this function 545 * @size: size of @ivalue 546 * 547 * User requested a write of security.capability. If needed, update the 548 * xattr to change from v2 to v3, or to fixup the v3 rootid. 549 * 550 * If the inode has been found through an idmapped mount the idmap of 551 * the vfsmount must be passed through @idmap. This function will then 552 * take care to map the inode according to @idmap before checking 553 * permissions. On non-idmapped mounts or if permission checking is to be 554 * performed on the raw inode simply pass @nop_mnt_idmap. 555 * 556 * Return: On success, return the new size; on error, return < 0. 557 */ 558 int cap_convert_nscap(struct mnt_idmap *idmap, struct dentry *dentry, 559 const void **ivalue, size_t size) 560 { 561 struct vfs_ns_cap_data *nscap; 562 uid_t nsrootid; 563 const struct vfs_cap_data *cap = *ivalue; 564 __u32 magic, nsmagic; 565 struct inode *inode = d_backing_inode(dentry); 566 struct user_namespace *task_ns = current_user_ns(), 567 *fs_ns = inode->i_sb->s_user_ns; 568 kuid_t rootid; 569 vfsuid_t vfsrootid; 570 size_t newsize; 571 572 if (!*ivalue) 573 return -EINVAL; 574 if (!validheader(size, cap)) 575 return -EINVAL; 576 if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP)) 577 return -EPERM; 578 if (size == XATTR_CAPS_SZ_2 && (idmap == &nop_mnt_idmap)) 579 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP)) 580 /* user is privileged, just write the v2 */ 581 return size; 582 583 vfsrootid = rootid_from_xattr(*ivalue, size, task_ns); 584 if (!vfsuid_valid(vfsrootid)) 585 return -EINVAL; 586 587 rootid = from_vfsuid(idmap, fs_ns, vfsrootid); 588 if (!uid_valid(rootid)) 589 return -EINVAL; 590 591 nsrootid = from_kuid(fs_ns, rootid); 592 if (nsrootid == -1) 593 return -EINVAL; 594 595 newsize = sizeof(struct vfs_ns_cap_data); 596 nscap = kmalloc(newsize, GFP_ATOMIC); 597 if (!nscap) 598 return -ENOMEM; 599 nscap->rootid = cpu_to_le32(nsrootid); 600 nsmagic = VFS_CAP_REVISION_3; 601 magic = le32_to_cpu(cap->magic_etc); 602 if (magic & VFS_CAP_FLAGS_EFFECTIVE) 603 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; 604 nscap->magic_etc = cpu_to_le32(nsmagic); 605 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 606 607 *ivalue = nscap; 608 return newsize; 609 } 610 611 /* 612 * Calculate the new process capability sets from the capability sets attached 613 * to a file. 614 */ 615 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, 616 struct linux_binprm *bprm, 617 bool *effective, 618 bool *has_fcap) 619 { 620 struct cred *new = bprm->cred; 621 int ret = 0; 622 623 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) 624 *effective = true; 625 626 if (caps->magic_etc & VFS_CAP_REVISION_MASK) 627 *has_fcap = true; 628 629 /* 630 * pP' = (X & fP) | (pI & fI) 631 * The addition of pA' is handled later. 632 */ 633 new->cap_permitted.val = 634 (new->cap_bset.val & caps->permitted.val) | 635 (new->cap_inheritable.val & caps->inheritable.val); 636 637 if (caps->permitted.val & ~new->cap_permitted.val) 638 /* insufficient to execute correctly */ 639 ret = -EPERM; 640 641 /* 642 * For legacy apps, with no internal support for recognizing they 643 * do not have enough capabilities, we return an error if they are 644 * missing some "forced" (aka file-permitted) capabilities. 645 */ 646 return *effective ? ret : 0; 647 } 648 649 /** 650 * get_vfs_caps_from_disk - retrieve vfs caps from disk 651 * 652 * @idmap: idmap of the mount the inode was found from 653 * @dentry: dentry from which @inode is retrieved 654 * @cpu_caps: vfs capabilities 655 * 656 * Extract the on-exec-apply capability sets for an executable file. 657 * 658 * If the inode has been found through an idmapped mount the idmap of 659 * the vfsmount must be passed through @idmap. This function will then 660 * take care to map the inode according to @idmap before checking 661 * permissions. On non-idmapped mounts or if permission checking is to be 662 * performed on the raw inode simply pass @nop_mnt_idmap. 663 */ 664 int get_vfs_caps_from_disk(struct mnt_idmap *idmap, 665 const struct dentry *dentry, 666 struct cpu_vfs_cap_data *cpu_caps) 667 { 668 struct inode *inode = d_backing_inode(dentry); 669 __u32 magic_etc; 670 int size; 671 struct vfs_ns_cap_data data, *nscaps = &data; 672 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data; 673 kuid_t rootkuid; 674 vfsuid_t rootvfsuid; 675 struct user_namespace *fs_ns; 676 677 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); 678 679 if (!inode) 680 return -ENODATA; 681 682 fs_ns = inode->i_sb->s_user_ns; 683 size = __vfs_getxattr((struct dentry *)dentry, inode, 684 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ); 685 if (size == -ENODATA || size == -EOPNOTSUPP) 686 /* no data, that's ok */ 687 return -ENODATA; 688 689 if (size < 0) 690 return size; 691 692 if (size < sizeof(magic_etc)) 693 return -EINVAL; 694 695 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc); 696 697 rootkuid = make_kuid(fs_ns, 0); 698 switch (magic_etc & VFS_CAP_REVISION_MASK) { 699 case VFS_CAP_REVISION_1: 700 if (size != XATTR_CAPS_SZ_1) 701 return -EINVAL; 702 break; 703 case VFS_CAP_REVISION_2: 704 if (size != XATTR_CAPS_SZ_2) 705 return -EINVAL; 706 break; 707 case VFS_CAP_REVISION_3: 708 if (size != XATTR_CAPS_SZ_3) 709 return -EINVAL; 710 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid)); 711 break; 712 713 default: 714 return -EINVAL; 715 } 716 717 rootvfsuid = make_vfsuid(idmap, fs_ns, rootkuid); 718 if (!vfsuid_valid(rootvfsuid)) 719 return -ENODATA; 720 721 /* Limit the caps to the mounter of the filesystem 722 * or the more limited uid specified in the xattr. 723 */ 724 if (!rootid_owns_currentns(rootvfsuid)) 725 return -ENODATA; 726 727 cpu_caps->permitted.val = le32_to_cpu(caps->data[0].permitted); 728 cpu_caps->inheritable.val = le32_to_cpu(caps->data[0].inheritable); 729 730 /* 731 * Rev1 had just a single 32-bit word, later expanded 732 * to a second one for the high bits 733 */ 734 if ((magic_etc & VFS_CAP_REVISION_MASK) != VFS_CAP_REVISION_1) { 735 cpu_caps->permitted.val += (u64)le32_to_cpu(caps->data[1].permitted) << 32; 736 cpu_caps->inheritable.val += (u64)le32_to_cpu(caps->data[1].inheritable) << 32; 737 } 738 739 cpu_caps->permitted.val &= CAP_VALID_MASK; 740 cpu_caps->inheritable.val &= CAP_VALID_MASK; 741 742 cpu_caps->rootid = vfsuid_into_kuid(rootvfsuid); 743 744 return 0; 745 } 746 747 /* 748 * Attempt to get the on-exec apply capability sets for an executable file from 749 * its xattrs and, if present, apply them to the proposed credentials being 750 * constructed by execve(). 751 */ 752 static int get_file_caps(struct linux_binprm *bprm, const struct file *file, 753 bool *effective, bool *has_fcap) 754 { 755 int rc = 0; 756 struct cpu_vfs_cap_data vcaps; 757 758 cap_clear(bprm->cred->cap_permitted); 759 760 if (!file_caps_enabled) 761 return 0; 762 763 if (!mnt_may_suid(file->f_path.mnt)) 764 return 0; 765 766 /* 767 * This check is redundant with mnt_may_suid() but is kept to make 768 * explicit that capability bits are limited to s_user_ns and its 769 * descendants. 770 */ 771 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns)) 772 return 0; 773 774 rc = get_vfs_caps_from_disk(file_mnt_idmap(file), 775 file->f_path.dentry, &vcaps); 776 if (rc < 0) { 777 if (rc == -EINVAL) 778 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n", 779 bprm->filename); 780 else if (rc == -ENODATA) 781 rc = 0; 782 goto out; 783 } 784 785 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap); 786 787 out: 788 if (rc) 789 cap_clear(bprm->cred->cap_permitted); 790 791 return rc; 792 } 793 794 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); } 795 796 static inline bool __is_real(kuid_t uid, struct cred *cred) 797 { return uid_eq(cred->uid, uid); } 798 799 static inline bool __is_eff(kuid_t uid, struct cred *cred) 800 { return uid_eq(cred->euid, uid); } 801 802 static inline bool __is_suid(kuid_t uid, struct cred *cred) 803 { return !__is_real(uid, cred) && __is_eff(uid, cred); } 804 805 /* 806 * handle_privileged_root - Handle case of privileged root 807 * @bprm: The execution parameters, including the proposed creds 808 * @has_fcap: Are any file capabilities set? 809 * @effective: Do we have effective root privilege? 810 * @root_uid: This namespace' root UID WRT initial USER namespace 811 * 812 * Handle the case where root is privileged and hasn't been neutered by 813 * SECURE_NOROOT. If file capabilities are set, they won't be combined with 814 * set UID root and nothing is changed. If we are root, cap_permitted is 815 * updated. If we have become set UID root, the effective bit is set. 816 */ 817 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap, 818 bool *effective, kuid_t root_uid) 819 { 820 const struct cred *old = current_cred(); 821 struct cred *new = bprm->cred; 822 823 if (!root_privileged()) 824 return; 825 /* 826 * If the legacy file capability is set, then don't set privs 827 * for a setuid root binary run by a non-root user. Do set it 828 * for a root user just to cause least surprise to an admin. 829 */ 830 if (has_fcap && __is_suid(root_uid, new)) { 831 warn_setuid_and_fcaps_mixed(bprm->filename); 832 return; 833 } 834 /* 835 * To support inheritance of root-permissions and suid-root 836 * executables under compatibility mode, we override the 837 * capability sets for the file. 838 */ 839 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) { 840 /* pP' = (cap_bset & ~0) | (pI & ~0) */ 841 new->cap_permitted = cap_combine(old->cap_bset, 842 old->cap_inheritable); 843 } 844 /* 845 * If only the real uid is 0, we do not set the effective bit. 846 */ 847 if (__is_eff(root_uid, new)) 848 *effective = true; 849 } 850 851 #define __cap_gained(field, target, source) \ 852 !cap_issubset(target->cap_##field, source->cap_##field) 853 #define __cap_grew(target, source, cred) \ 854 !cap_issubset(cred->cap_##target, cred->cap_##source) 855 #define __cap_full(field, cred) \ 856 cap_issubset(CAP_FULL_SET, cred->cap_##field) 857 858 static inline bool __is_setuid(struct cred *new, const struct cred *old) 859 { return !uid_eq(new->euid, old->uid); } 860 861 static inline bool __is_setgid(struct cred *new, const struct cred *old) 862 { return !gid_eq(new->egid, old->gid); } 863 864 /* 865 * 1) Audit candidate if current->cap_effective is set 866 * 867 * We do not bother to audit if 3 things are true: 868 * 1) cap_effective has all caps 869 * 2) we became root *OR* are were already root 870 * 3) root is supposed to have all caps (SECURE_NOROOT) 871 * Since this is just a normal root execing a process. 872 * 873 * Number 1 above might fail if you don't have a full bset, but I think 874 * that is interesting information to audit. 875 * 876 * A number of other conditions require logging: 877 * 2) something prevented setuid root getting all caps 878 * 3) non-setuid root gets fcaps 879 * 4) non-setuid root gets ambient 880 */ 881 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old, 882 kuid_t root, bool has_fcap) 883 { 884 bool ret = false; 885 886 if ((__cap_grew(effective, ambient, new) && 887 !(__cap_full(effective, new) && 888 (__is_eff(root, new) || __is_real(root, new)) && 889 root_privileged())) || 890 (root_privileged() && 891 __is_suid(root, new) && 892 !__cap_full(effective, new)) || 893 (!__is_setuid(new, old) && 894 ((has_fcap && 895 __cap_gained(permitted, new, old)) || 896 __cap_gained(ambient, new, old)))) 897 898 ret = true; 899 900 return ret; 901 } 902 903 /** 904 * cap_bprm_creds_from_file - Set up the proposed credentials for execve(). 905 * @bprm: The execution parameters, including the proposed creds 906 * @file: The file to pull the credentials from 907 * 908 * Set up the proposed credentials for a new execution context being 909 * constructed by execve(). The proposed creds in @bprm->cred is altered, 910 * which won't take effect immediately. 911 * 912 * Return: 0 if successful, -ve on error. 913 */ 914 int cap_bprm_creds_from_file(struct linux_binprm *bprm, const struct file *file) 915 { 916 /* Process setpcap binaries and capabilities for uid 0 */ 917 const struct cred *old = current_cred(); 918 struct cred *new = bprm->cred; 919 bool effective = false, has_fcap = false, is_setid; 920 int ret; 921 kuid_t root_uid; 922 923 if (WARN_ON(!cap_ambient_invariant_ok(old))) 924 return -EPERM; 925 926 ret = get_file_caps(bprm, file, &effective, &has_fcap); 927 if (ret < 0) 928 return ret; 929 930 root_uid = make_kuid(new->user_ns, 0); 931 932 handle_privileged_root(bprm, has_fcap, &effective, root_uid); 933 934 /* if we have fs caps, clear dangerous personality flags */ 935 if (__cap_gained(permitted, new, old)) 936 bprm->per_clear |= PER_CLEAR_ON_SETID; 937 938 /* Don't let someone trace a set[ug]id/setpcap binary with the revised 939 * credentials unless they have the appropriate permit. 940 * 941 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. 942 */ 943 is_setid = __is_setuid(new, old) || __is_setgid(new, old); 944 945 if ((is_setid || __cap_gained(permitted, new, old)) && 946 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) || 947 !ptracer_capable(current, new->user_ns))) { 948 /* downgrade; they get no more than they had, and maybe less */ 949 if (!ns_capable(new->user_ns, CAP_SETUID) || 950 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { 951 new->euid = new->uid; 952 new->egid = new->gid; 953 } 954 new->cap_permitted = cap_intersect(new->cap_permitted, 955 old->cap_permitted); 956 } 957 958 new->suid = new->fsuid = new->euid; 959 new->sgid = new->fsgid = new->egid; 960 961 /* File caps or setid cancels ambient. */ 962 if (has_fcap || is_setid) 963 cap_clear(new->cap_ambient); 964 965 /* 966 * Now that we've computed pA', update pP' to give: 967 * pP' = (X & fP) | (pI & fI) | pA' 968 */ 969 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); 970 971 /* 972 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, 973 * this is the same as pE' = (fE ? pP' : 0) | pA'. 974 */ 975 if (effective) 976 new->cap_effective = new->cap_permitted; 977 else 978 new->cap_effective = new->cap_ambient; 979 980 if (WARN_ON(!cap_ambient_invariant_ok(new))) 981 return -EPERM; 982 983 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) { 984 ret = audit_log_bprm_fcaps(bprm, new, old); 985 if (ret < 0) 986 return ret; 987 } 988 989 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 990 991 if (WARN_ON(!cap_ambient_invariant_ok(new))) 992 return -EPERM; 993 994 /* Check for privilege-elevated exec. */ 995 if (is_setid || 996 (!__is_real(root_uid, new) && 997 (effective || 998 __cap_grew(permitted, ambient, new)))) 999 bprm->secureexec = 1; 1000 1001 return 0; 1002 } 1003 1004 /** 1005 * cap_inode_setxattr - Determine whether an xattr may be altered 1006 * @dentry: The inode/dentry being altered 1007 * @name: The name of the xattr to be changed 1008 * @value: The value that the xattr will be changed to 1009 * @size: The size of value 1010 * @flags: The replacement flag 1011 * 1012 * Determine whether an xattr may be altered or set on an inode, returning 0 if 1013 * permission is granted, -ve if denied. 1014 * 1015 * This is used to make sure security xattrs don't get updated or set by those 1016 * who aren't privileged to do so. 1017 */ 1018 int cap_inode_setxattr(struct dentry *dentry, const char *name, 1019 const void *value, size_t size, int flags) 1020 { 1021 struct user_namespace *user_ns = dentry->d_sb->s_user_ns; 1022 1023 /* Ignore non-security xattrs */ 1024 if (strncmp(name, XATTR_SECURITY_PREFIX, 1025 XATTR_SECURITY_PREFIX_LEN) != 0) 1026 return 0; 1027 1028 /* 1029 * For XATTR_NAME_CAPS the check will be done in 1030 * cap_convert_nscap(), called by setxattr() 1031 */ 1032 if (strcmp(name, XATTR_NAME_CAPS) == 0) 1033 return 0; 1034 1035 if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 1036 return -EPERM; 1037 return 0; 1038 } 1039 1040 /** 1041 * cap_inode_removexattr - Determine whether an xattr may be removed 1042 * 1043 * @idmap: idmap of the mount the inode was found from 1044 * @dentry: The inode/dentry being altered 1045 * @name: The name of the xattr to be changed 1046 * 1047 * Determine whether an xattr may be removed from an inode, returning 0 if 1048 * permission is granted, -ve if denied. 1049 * 1050 * If the inode has been found through an idmapped mount the idmap of 1051 * the vfsmount must be passed through @idmap. This function will then 1052 * take care to map the inode according to @idmap before checking 1053 * permissions. On non-idmapped mounts or if permission checking is to be 1054 * performed on the raw inode simply pass @nop_mnt_idmap. 1055 * 1056 * This is used to make sure security xattrs don't get removed by those who 1057 * aren't privileged to remove them. 1058 */ 1059 int cap_inode_removexattr(struct mnt_idmap *idmap, 1060 struct dentry *dentry, const char *name) 1061 { 1062 struct user_namespace *user_ns = dentry->d_sb->s_user_ns; 1063 1064 /* Ignore non-security xattrs */ 1065 if (strncmp(name, XATTR_SECURITY_PREFIX, 1066 XATTR_SECURITY_PREFIX_LEN) != 0) 1067 return 0; 1068 1069 if (strcmp(name, XATTR_NAME_CAPS) == 0) { 1070 /* security.capability gets namespaced */ 1071 struct inode *inode = d_backing_inode(dentry); 1072 if (!inode) 1073 return -EINVAL; 1074 if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP)) 1075 return -EPERM; 1076 return 0; 1077 } 1078 1079 if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 1080 return -EPERM; 1081 return 0; 1082 } 1083 1084 /* 1085 * cap_emulate_setxuid() fixes the effective / permitted capabilities of 1086 * a process after a call to setuid, setreuid, or setresuid. 1087 * 1088 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of 1089 * {r,e,s}uid != 0, the permitted and effective capabilities are 1090 * cleared. 1091 * 1092 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective 1093 * capabilities of the process are cleared. 1094 * 1095 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective 1096 * capabilities are set to the permitted capabilities. 1097 * 1098 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 1099 * never happen. 1100 * 1101 * -astor 1102 * 1103 * cevans - New behaviour, Oct '99 1104 * A process may, via prctl(), elect to keep its capabilities when it 1105 * calls setuid() and switches away from uid==0. Both permitted and 1106 * effective sets will be retained. 1107 * Without this change, it was impossible for a daemon to drop only some 1108 * of its privilege. The call to setuid(!=0) would drop all privileges! 1109 * Keeping uid 0 is not an option because uid 0 owns too many vital 1110 * files.. 1111 * Thanks to Olaf Kirch and Peter Benie for spotting this. 1112 */ 1113 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) 1114 { 1115 kuid_t root_uid = make_kuid(old->user_ns, 0); 1116 1117 if ((uid_eq(old->uid, root_uid) || 1118 uid_eq(old->euid, root_uid) || 1119 uid_eq(old->suid, root_uid)) && 1120 (!uid_eq(new->uid, root_uid) && 1121 !uid_eq(new->euid, root_uid) && 1122 !uid_eq(new->suid, root_uid))) { 1123 if (!issecure(SECURE_KEEP_CAPS)) { 1124 cap_clear(new->cap_permitted); 1125 cap_clear(new->cap_effective); 1126 } 1127 1128 /* 1129 * Pre-ambient programs expect setresuid to nonroot followed 1130 * by exec to drop capabilities. We should make sure that 1131 * this remains the case. 1132 */ 1133 cap_clear(new->cap_ambient); 1134 } 1135 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) 1136 cap_clear(new->cap_effective); 1137 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) 1138 new->cap_effective = new->cap_permitted; 1139 } 1140 1141 /** 1142 * cap_task_fix_setuid - Fix up the results of setuid() call 1143 * @new: The proposed credentials 1144 * @old: The current task's current credentials 1145 * @flags: Indications of what has changed 1146 * 1147 * Fix up the results of setuid() call before the credential changes are 1148 * actually applied. 1149 * 1150 * Return: 0 to grant the changes, -ve to deny them. 1151 */ 1152 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) 1153 { 1154 switch (flags) { 1155 case LSM_SETID_RE: 1156 case LSM_SETID_ID: 1157 case LSM_SETID_RES: 1158 /* juggle the capabilities to follow [RES]UID changes unless 1159 * otherwise suppressed */ 1160 if (!issecure(SECURE_NO_SETUID_FIXUP)) 1161 cap_emulate_setxuid(new, old); 1162 break; 1163 1164 case LSM_SETID_FS: 1165 /* juggle the capabilities to follow FSUID changes, unless 1166 * otherwise suppressed 1167 * 1168 * FIXME - is fsuser used for all CAP_FS_MASK capabilities? 1169 * if not, we might be a bit too harsh here. 1170 */ 1171 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 1172 kuid_t root_uid = make_kuid(old->user_ns, 0); 1173 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) 1174 new->cap_effective = 1175 cap_drop_fs_set(new->cap_effective); 1176 1177 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) 1178 new->cap_effective = 1179 cap_raise_fs_set(new->cap_effective, 1180 new->cap_permitted); 1181 } 1182 break; 1183 1184 default: 1185 return -EINVAL; 1186 } 1187 1188 return 0; 1189 } 1190 1191 /* 1192 * Rationale: code calling task_setscheduler, task_setioprio, and 1193 * task_setnice, assumes that 1194 * . if capable(cap_sys_nice), then those actions should be allowed 1195 * . if not capable(cap_sys_nice), but acting on your own processes, 1196 * then those actions should be allowed 1197 * This is insufficient now since you can call code without suid, but 1198 * yet with increased caps. 1199 * So we check for increased caps on the target process. 1200 */ 1201 static int cap_safe_nice(struct task_struct *p) 1202 { 1203 int is_subset, ret = 0; 1204 1205 rcu_read_lock(); 1206 is_subset = cap_issubset(__task_cred(p)->cap_permitted, 1207 current_cred()->cap_permitted); 1208 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) 1209 ret = -EPERM; 1210 rcu_read_unlock(); 1211 1212 return ret; 1213 } 1214 1215 /** 1216 * cap_task_setscheduler - Determine if scheduler policy change is permitted 1217 * @p: The task to affect 1218 * 1219 * Determine if the requested scheduler policy change is permitted for the 1220 * specified task. 1221 * 1222 * Return: 0 if permission is granted, -ve if denied. 1223 */ 1224 int cap_task_setscheduler(struct task_struct *p) 1225 { 1226 return cap_safe_nice(p); 1227 } 1228 1229 /** 1230 * cap_task_setioprio - Determine if I/O priority change is permitted 1231 * @p: The task to affect 1232 * @ioprio: The I/O priority to set 1233 * 1234 * Determine if the requested I/O priority change is permitted for the specified 1235 * task. 1236 * 1237 * Return: 0 if permission is granted, -ve if denied. 1238 */ 1239 int cap_task_setioprio(struct task_struct *p, int ioprio) 1240 { 1241 return cap_safe_nice(p); 1242 } 1243 1244 /** 1245 * cap_task_setnice - Determine if task priority change is permitted 1246 * @p: The task to affect 1247 * @nice: The nice value to set 1248 * 1249 * Determine if the requested task priority change is permitted for the 1250 * specified task. 1251 * 1252 * Return: 0 if permission is granted, -ve if denied. 1253 */ 1254 int cap_task_setnice(struct task_struct *p, int nice) 1255 { 1256 return cap_safe_nice(p); 1257 } 1258 1259 /* 1260 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from 1261 * the current task's bounding set. Returns 0 on success, -ve on error. 1262 */ 1263 static int cap_prctl_drop(unsigned long cap) 1264 { 1265 struct cred *new; 1266 1267 if (!ns_capable(current_user_ns(), CAP_SETPCAP)) 1268 return -EPERM; 1269 if (!cap_valid(cap)) 1270 return -EINVAL; 1271 1272 new = prepare_creds(); 1273 if (!new) 1274 return -ENOMEM; 1275 cap_lower(new->cap_bset, cap); 1276 return commit_creds(new); 1277 } 1278 1279 /** 1280 * cap_task_prctl - Implement process control functions for this security module 1281 * @option: The process control function requested 1282 * @arg2: The argument data for this function 1283 * @arg3: The argument data for this function 1284 * @arg4: The argument data for this function 1285 * @arg5: The argument data for this function 1286 * 1287 * Allow process control functions (sys_prctl()) to alter capabilities; may 1288 * also deny access to other functions not otherwise implemented here. 1289 * 1290 * Return: 0 or +ve on success, -ENOSYS if this function is not implemented 1291 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM 1292 * modules will consider performing the function. 1293 */ 1294 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, 1295 unsigned long arg4, unsigned long arg5) 1296 { 1297 const struct cred *old = current_cred(); 1298 struct cred *new; 1299 1300 switch (option) { 1301 case PR_CAPBSET_READ: 1302 if (!cap_valid(arg2)) 1303 return -EINVAL; 1304 return !!cap_raised(old->cap_bset, arg2); 1305 1306 case PR_CAPBSET_DROP: 1307 return cap_prctl_drop(arg2); 1308 1309 /* 1310 * The next four prctl's remain to assist with transitioning a 1311 * system from legacy UID=0 based privilege (when filesystem 1312 * capabilities are not in use) to a system using filesystem 1313 * capabilities only - as the POSIX.1e draft intended. 1314 * 1315 * Note: 1316 * 1317 * PR_SET_SECUREBITS = 1318 * issecure_mask(SECURE_KEEP_CAPS_LOCKED) 1319 * | issecure_mask(SECURE_NOROOT) 1320 * | issecure_mask(SECURE_NOROOT_LOCKED) 1321 * | issecure_mask(SECURE_NO_SETUID_FIXUP) 1322 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) 1323 * 1324 * will ensure that the current process and all of its 1325 * children will be locked into a pure 1326 * capability-based-privilege environment. 1327 */ 1328 case PR_SET_SECUREBITS: 1329 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) 1330 & (old->securebits ^ arg2)) /*[1]*/ 1331 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ 1332 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ 1333 /* 1334 * [1] no changing of bits that are locked 1335 * [2] no unlocking of locks 1336 * [3] no setting of unsupported bits 1337 */ 1338 ) 1339 /* cannot change a locked bit */ 1340 return -EPERM; 1341 1342 /* 1343 * Doing anything requires privilege (go read about the 1344 * "sendmail capabilities bug"), except for unprivileged bits. 1345 * Indeed, the SECURE_ALL_UNPRIVILEGED bits are not 1346 * restrictions enforced by the kernel but by user space on 1347 * itself. 1348 */ 1349 if (cap_capable(current_cred(), current_cred()->user_ns, 1350 CAP_SETPCAP, CAP_OPT_NONE) != 0) { 1351 const unsigned long unpriv_and_locks = 1352 SECURE_ALL_UNPRIVILEGED | 1353 SECURE_ALL_UNPRIVILEGED << 1; 1354 const unsigned long changed = old->securebits ^ arg2; 1355 1356 /* For legacy reason, denies non-change. */ 1357 if (!changed) 1358 return -EPERM; 1359 1360 /* Denies privileged changes. */ 1361 if (changed & ~unpriv_and_locks) 1362 return -EPERM; 1363 } 1364 1365 new = prepare_creds(); 1366 if (!new) 1367 return -ENOMEM; 1368 new->securebits = arg2; 1369 return commit_creds(new); 1370 1371 case PR_GET_SECUREBITS: 1372 return old->securebits; 1373 1374 case PR_GET_KEEPCAPS: 1375 return !!issecure(SECURE_KEEP_CAPS); 1376 1377 case PR_SET_KEEPCAPS: 1378 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ 1379 return -EINVAL; 1380 if (issecure(SECURE_KEEP_CAPS_LOCKED)) 1381 return -EPERM; 1382 1383 new = prepare_creds(); 1384 if (!new) 1385 return -ENOMEM; 1386 if (arg2) 1387 new->securebits |= issecure_mask(SECURE_KEEP_CAPS); 1388 else 1389 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 1390 return commit_creds(new); 1391 1392 case PR_CAP_AMBIENT: 1393 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { 1394 if (arg3 | arg4 | arg5) 1395 return -EINVAL; 1396 1397 new = prepare_creds(); 1398 if (!new) 1399 return -ENOMEM; 1400 cap_clear(new->cap_ambient); 1401 return commit_creds(new); 1402 } 1403 1404 if (((!cap_valid(arg3)) | arg4 | arg5)) 1405 return -EINVAL; 1406 1407 if (arg2 == PR_CAP_AMBIENT_IS_SET) { 1408 return !!cap_raised(current_cred()->cap_ambient, arg3); 1409 } else if (arg2 != PR_CAP_AMBIENT_RAISE && 1410 arg2 != PR_CAP_AMBIENT_LOWER) { 1411 return -EINVAL; 1412 } else { 1413 if (arg2 == PR_CAP_AMBIENT_RAISE && 1414 (!cap_raised(current_cred()->cap_permitted, arg3) || 1415 !cap_raised(current_cred()->cap_inheritable, 1416 arg3) || 1417 issecure(SECURE_NO_CAP_AMBIENT_RAISE))) 1418 return -EPERM; 1419 1420 new = prepare_creds(); 1421 if (!new) 1422 return -ENOMEM; 1423 if (arg2 == PR_CAP_AMBIENT_RAISE) 1424 cap_raise(new->cap_ambient, arg3); 1425 else 1426 cap_lower(new->cap_ambient, arg3); 1427 return commit_creds(new); 1428 } 1429 1430 default: 1431 /* No functionality available - continue with default */ 1432 return -ENOSYS; 1433 } 1434 } 1435 1436 /** 1437 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted 1438 * @mm: The VM space in which the new mapping is to be made 1439 * @pages: The size of the mapping 1440 * 1441 * Determine whether the allocation of a new virtual mapping by the current 1442 * task is permitted. 1443 * 1444 * Return: 0 if permission granted, negative error code if not. 1445 */ 1446 int cap_vm_enough_memory(struct mm_struct *mm, long pages) 1447 { 1448 return cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN, 1449 CAP_OPT_NOAUDIT); 1450 } 1451 1452 /** 1453 * cap_mmap_addr - check if able to map given addr 1454 * @addr: address attempting to be mapped 1455 * 1456 * If the process is attempting to map memory below dac_mmap_min_addr they need 1457 * CAP_SYS_RAWIO. The other parameters to this function are unused by the 1458 * capability security module. 1459 * 1460 * Return: 0 if this mapping should be allowed or -EPERM if not. 1461 */ 1462 int cap_mmap_addr(unsigned long addr) 1463 { 1464 int ret = 0; 1465 1466 if (addr < dac_mmap_min_addr) { 1467 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, 1468 CAP_OPT_NONE); 1469 /* set PF_SUPERPRIV if it turns out we allow the low mmap */ 1470 if (ret == 0) 1471 current->flags |= PF_SUPERPRIV; 1472 } 1473 return ret; 1474 } 1475 1476 #ifdef CONFIG_SECURITY 1477 1478 static const struct lsm_id capability_lsmid = { 1479 .name = "capability", 1480 .id = LSM_ID_CAPABILITY, 1481 }; 1482 1483 static struct security_hook_list capability_hooks[] __ro_after_init = { 1484 LSM_HOOK_INIT(capable, cap_capable), 1485 LSM_HOOK_INIT(settime, cap_settime), 1486 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), 1487 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), 1488 LSM_HOOK_INIT(capget, cap_capget), 1489 LSM_HOOK_INIT(capset, cap_capset), 1490 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file), 1491 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), 1492 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), 1493 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity), 1494 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), 1495 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), 1496 LSM_HOOK_INIT(task_prctl, cap_task_prctl), 1497 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), 1498 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), 1499 LSM_HOOK_INIT(task_setnice, cap_task_setnice), 1500 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), 1501 }; 1502 1503 static int __init capability_init(void) 1504 { 1505 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks), 1506 &capability_lsmid); 1507 return 0; 1508 } 1509 1510 DEFINE_LSM(capability) = { 1511 .name = "capability", 1512 .order = LSM_ORDER_FIRST, 1513 .init = capability_init, 1514 }; 1515 1516 #endif /* CONFIG_SECURITY */ 1517