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 29 /* 30 * If a non-root user executes a setuid-root binary in 31 * !secure(SECURE_NOROOT) mode, then we raise capabilities. 32 * However if fE is also set, then the intent is for only 33 * the file capabilities to be applied, and the setuid-root 34 * bit is left on either to change the uid (plausible) or 35 * to get full privilege on a kernel without file capabilities 36 * support. So in that case we do not raise capabilities. 37 * 38 * Warn if that happens, once per boot. 39 */ 40 static void warn_setuid_and_fcaps_mixed(const char *fname) 41 { 42 static int warned; 43 if (!warned) { 44 printk(KERN_INFO "warning: `%s' has both setuid-root and" 45 " effective capabilities. Therefore not raising all" 46 " capabilities.\n", fname); 47 warned = 1; 48 } 49 } 50 51 /** 52 * cap_capable - Determine whether a task has a particular effective capability 53 * @cred: The credentials to use 54 * @targ_ns: The user namespace in which we need the capability 55 * @cap: The capability to check for 56 * @opts: Bitmask of options defined in include/linux/security.h 57 * 58 * Determine whether the nominated task has the specified capability amongst 59 * its effective set, returning 0 if it does, -ve if it does not. 60 * 61 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable() 62 * and has_capability() functions. That is, it has the reverse semantics: 63 * cap_has_capability() returns 0 when a task has a capability, but the 64 * kernel's capable() and has_capability() returns 1 for this case. 65 */ 66 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns, 67 int cap, unsigned int opts) 68 { 69 struct user_namespace *ns = targ_ns; 70 71 /* See if cred has the capability in the target user namespace 72 * by examining the target user namespace and all of the target 73 * user namespace's parents. 74 */ 75 for (;;) { 76 /* Do we have the necessary capabilities? */ 77 if (ns == cred->user_ns) 78 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM; 79 80 /* 81 * If we're already at a lower level than we're looking for, 82 * we're done searching. 83 */ 84 if (ns->level <= cred->user_ns->level) 85 return -EPERM; 86 87 /* 88 * The owner of the user namespace in the parent of the 89 * user namespace has all caps. 90 */ 91 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid)) 92 return 0; 93 94 /* 95 * If you have a capability in a parent user ns, then you have 96 * it over all children user namespaces as well. 97 */ 98 ns = ns->parent; 99 } 100 101 /* We never get here */ 102 } 103 104 /** 105 * cap_settime - Determine whether the current process may set the system clock 106 * @ts: The time to set 107 * @tz: The timezone to set 108 * 109 * Determine whether the current process may set the system clock and timezone 110 * information, returning 0 if permission granted, -ve if denied. 111 */ 112 int cap_settime(const struct timespec64 *ts, const struct timezone *tz) 113 { 114 if (!capable(CAP_SYS_TIME)) 115 return -EPERM; 116 return 0; 117 } 118 119 /** 120 * cap_ptrace_access_check - Determine whether the current process may access 121 * another 122 * @child: The process to be accessed 123 * @mode: The mode of attachment. 124 * 125 * If we are in the same or an ancestor user_ns and have all the target 126 * task's capabilities, then ptrace access is allowed. 127 * If we have the ptrace capability to the target user_ns, then ptrace 128 * access is allowed. 129 * Else denied. 130 * 131 * Determine whether a process may access another, returning 0 if permission 132 * granted, -ve if denied. 133 */ 134 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) 135 { 136 int ret = 0; 137 const struct cred *cred, *child_cred; 138 const kernel_cap_t *caller_caps; 139 140 rcu_read_lock(); 141 cred = current_cred(); 142 child_cred = __task_cred(child); 143 if (mode & PTRACE_MODE_FSCREDS) 144 caller_caps = &cred->cap_effective; 145 else 146 caller_caps = &cred->cap_permitted; 147 if (cred->user_ns == child_cred->user_ns && 148 cap_issubset(child_cred->cap_permitted, *caller_caps)) 149 goto out; 150 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) 151 goto out; 152 ret = -EPERM; 153 out: 154 rcu_read_unlock(); 155 return ret; 156 } 157 158 /** 159 * cap_ptrace_traceme - Determine whether another process may trace the current 160 * @parent: The task proposed to be the tracer 161 * 162 * If parent is in the same or an ancestor user_ns and has all current's 163 * capabilities, then ptrace access is allowed. 164 * If parent has the ptrace capability to current's user_ns, then ptrace 165 * access is allowed. 166 * Else denied. 167 * 168 * Determine whether the nominated task is permitted to trace the current 169 * process, returning 0 if permission is granted, -ve if denied. 170 */ 171 int cap_ptrace_traceme(struct task_struct *parent) 172 { 173 int ret = 0; 174 const struct cred *cred, *child_cred; 175 176 rcu_read_lock(); 177 cred = __task_cred(parent); 178 child_cred = current_cred(); 179 if (cred->user_ns == child_cred->user_ns && 180 cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) 181 goto out; 182 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) 183 goto out; 184 ret = -EPERM; 185 out: 186 rcu_read_unlock(); 187 return ret; 188 } 189 190 /** 191 * cap_capget - Retrieve a task's capability sets 192 * @target: The task from which to retrieve the capability sets 193 * @effective: The place to record the effective set 194 * @inheritable: The place to record the inheritable set 195 * @permitted: The place to record the permitted set 196 * 197 * This function retrieves the capabilities of the nominated task and returns 198 * them to the caller. 199 */ 200 int cap_capget(struct task_struct *target, kernel_cap_t *effective, 201 kernel_cap_t *inheritable, kernel_cap_t *permitted) 202 { 203 const struct cred *cred; 204 205 /* Derived from kernel/capability.c:sys_capget. */ 206 rcu_read_lock(); 207 cred = __task_cred(target); 208 *effective = cred->cap_effective; 209 *inheritable = cred->cap_inheritable; 210 *permitted = cred->cap_permitted; 211 rcu_read_unlock(); 212 return 0; 213 } 214 215 /* 216 * Determine whether the inheritable capabilities are limited to the old 217 * permitted set. Returns 1 if they are limited, 0 if they are not. 218 */ 219 static inline int cap_inh_is_capped(void) 220 { 221 /* they are so limited unless the current task has the CAP_SETPCAP 222 * capability 223 */ 224 if (cap_capable(current_cred(), current_cred()->user_ns, 225 CAP_SETPCAP, CAP_OPT_NONE) == 0) 226 return 0; 227 return 1; 228 } 229 230 /** 231 * cap_capset - Validate and apply proposed changes to current's capabilities 232 * @new: The proposed new credentials; alterations should be made here 233 * @old: The current task's current credentials 234 * @effective: A pointer to the proposed new effective capabilities set 235 * @inheritable: A pointer to the proposed new inheritable capabilities set 236 * @permitted: A pointer to the proposed new permitted capabilities set 237 * 238 * This function validates and applies a proposed mass change to the current 239 * process's capability sets. The changes are made to the proposed new 240 * credentials, and assuming no error, will be committed by the caller of LSM. 241 */ 242 int cap_capset(struct cred *new, 243 const struct cred *old, 244 const kernel_cap_t *effective, 245 const kernel_cap_t *inheritable, 246 const kernel_cap_t *permitted) 247 { 248 if (cap_inh_is_capped() && 249 !cap_issubset(*inheritable, 250 cap_combine(old->cap_inheritable, 251 old->cap_permitted))) 252 /* incapable of using this inheritable set */ 253 return -EPERM; 254 255 if (!cap_issubset(*inheritable, 256 cap_combine(old->cap_inheritable, 257 old->cap_bset))) 258 /* no new pI capabilities outside bounding set */ 259 return -EPERM; 260 261 /* verify restrictions on target's new Permitted set */ 262 if (!cap_issubset(*permitted, old->cap_permitted)) 263 return -EPERM; 264 265 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ 266 if (!cap_issubset(*effective, *permitted)) 267 return -EPERM; 268 269 new->cap_effective = *effective; 270 new->cap_inheritable = *inheritable; 271 new->cap_permitted = *permitted; 272 273 /* 274 * Mask off ambient bits that are no longer both permitted and 275 * inheritable. 276 */ 277 new->cap_ambient = cap_intersect(new->cap_ambient, 278 cap_intersect(*permitted, 279 *inheritable)); 280 if (WARN_ON(!cap_ambient_invariant_ok(new))) 281 return -EINVAL; 282 return 0; 283 } 284 285 /** 286 * cap_inode_need_killpriv - Determine if inode change affects privileges 287 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV 288 * 289 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV 290 * affects the security markings on that inode, and if it is, should 291 * inode_killpriv() be invoked or the change rejected. 292 * 293 * Return: 1 if security.capability has a value, meaning inode_killpriv() 294 * is required, 0 otherwise, meaning inode_killpriv() is not required. 295 */ 296 int cap_inode_need_killpriv(struct dentry *dentry) 297 { 298 struct inode *inode = d_backing_inode(dentry); 299 int error; 300 301 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); 302 return error > 0; 303 } 304 305 /** 306 * cap_inode_killpriv - Erase the security markings on an inode 307 * 308 * @mnt_userns: user namespace of the mount the inode was found from 309 * @dentry: The inode/dentry to alter 310 * 311 * Erase the privilege-enhancing security markings on an inode. 312 * 313 * If the inode has been found through an idmapped mount the user namespace of 314 * the vfsmount must be passed through @mnt_userns. This function will then 315 * take care to map the inode according to @mnt_userns before checking 316 * permissions. On non-idmapped mounts or if permission checking is to be 317 * performed on the raw inode simply passs init_user_ns. 318 * 319 * Return: 0 if successful, -ve on error. 320 */ 321 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry) 322 { 323 int error; 324 325 error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS); 326 if (error == -EOPNOTSUPP) 327 error = 0; 328 return error; 329 } 330 331 static bool rootid_owns_currentns(kuid_t kroot) 332 { 333 struct user_namespace *ns; 334 335 if (!uid_valid(kroot)) 336 return false; 337 338 for (ns = current_user_ns(); ; ns = ns->parent) { 339 if (from_kuid(ns, kroot) == 0) 340 return true; 341 if (ns == &init_user_ns) 342 break; 343 } 344 345 return false; 346 } 347 348 static __u32 sansflags(__u32 m) 349 { 350 return m & ~VFS_CAP_FLAGS_EFFECTIVE; 351 } 352 353 static bool is_v2header(size_t size, const struct vfs_cap_data *cap) 354 { 355 if (size != XATTR_CAPS_SZ_2) 356 return false; 357 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2; 358 } 359 360 static bool is_v3header(size_t size, const struct vfs_cap_data *cap) 361 { 362 if (size != XATTR_CAPS_SZ_3) 363 return false; 364 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3; 365 } 366 367 /* 368 * getsecurity: We are called for security.* before any attempt to read the 369 * xattr from the inode itself. 370 * 371 * This gives us a chance to read the on-disk value and convert it. If we 372 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler. 373 * 374 * Note we are not called by vfs_getxattr_alloc(), but that is only called 375 * by the integrity subsystem, which really wants the unconverted values - 376 * so that's good. 377 */ 378 int cap_inode_getsecurity(struct user_namespace *mnt_userns, 379 struct inode *inode, const char *name, void **buffer, 380 bool alloc) 381 { 382 int size, ret; 383 kuid_t kroot; 384 u32 nsmagic, magic; 385 uid_t root, mappedroot; 386 char *tmpbuf = NULL; 387 struct vfs_cap_data *cap; 388 struct vfs_ns_cap_data *nscap = NULL; 389 struct dentry *dentry; 390 struct user_namespace *fs_ns; 391 392 if (strcmp(name, "capability") != 0) 393 return -EOPNOTSUPP; 394 395 dentry = d_find_any_alias(inode); 396 if (!dentry) 397 return -EINVAL; 398 399 size = sizeof(struct vfs_ns_cap_data); 400 ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS, 401 &tmpbuf, size, GFP_NOFS); 402 dput(dentry); 403 404 if (ret < 0 || !tmpbuf) { 405 size = ret; 406 goto out_free; 407 } 408 409 fs_ns = inode->i_sb->s_user_ns; 410 cap = (struct vfs_cap_data *) tmpbuf; 411 if (is_v2header((size_t) ret, cap)) { 412 root = 0; 413 } else if (is_v3header((size_t) ret, cap)) { 414 nscap = (struct vfs_ns_cap_data *) tmpbuf; 415 root = le32_to_cpu(nscap->rootid); 416 } else { 417 size = -EINVAL; 418 goto out_free; 419 } 420 421 kroot = make_kuid(fs_ns, root); 422 423 /* If this is an idmapped mount shift the kuid. */ 424 kroot = mapped_kuid_fs(mnt_userns, fs_ns, kroot); 425 426 /* If the root kuid maps to a valid uid in current ns, then return 427 * this as a nscap. */ 428 mappedroot = from_kuid(current_user_ns(), kroot); 429 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) { 430 size = sizeof(struct vfs_ns_cap_data); 431 if (alloc) { 432 if (!nscap) { 433 /* v2 -> v3 conversion */ 434 nscap = kzalloc(size, GFP_ATOMIC); 435 if (!nscap) { 436 size = -ENOMEM; 437 goto out_free; 438 } 439 nsmagic = VFS_CAP_REVISION_3; 440 magic = le32_to_cpu(cap->magic_etc); 441 if (magic & VFS_CAP_FLAGS_EFFECTIVE) 442 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; 443 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 444 nscap->magic_etc = cpu_to_le32(nsmagic); 445 } else { 446 /* use allocated v3 buffer */ 447 tmpbuf = NULL; 448 } 449 nscap->rootid = cpu_to_le32(mappedroot); 450 *buffer = nscap; 451 } 452 goto out_free; 453 } 454 455 if (!rootid_owns_currentns(kroot)) { 456 size = -EOVERFLOW; 457 goto out_free; 458 } 459 460 /* This comes from a parent namespace. Return as a v2 capability */ 461 size = sizeof(struct vfs_cap_data); 462 if (alloc) { 463 if (nscap) { 464 /* v3 -> v2 conversion */ 465 cap = kzalloc(size, GFP_ATOMIC); 466 if (!cap) { 467 size = -ENOMEM; 468 goto out_free; 469 } 470 magic = VFS_CAP_REVISION_2; 471 nsmagic = le32_to_cpu(nscap->magic_etc); 472 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE) 473 magic |= VFS_CAP_FLAGS_EFFECTIVE; 474 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 475 cap->magic_etc = cpu_to_le32(magic); 476 } else { 477 /* use unconverted v2 */ 478 tmpbuf = NULL; 479 } 480 *buffer = cap; 481 } 482 out_free: 483 kfree(tmpbuf); 484 return size; 485 } 486 487 /** 488 * rootid_from_xattr - translate root uid of vfs caps 489 * 490 * @value: vfs caps value which may be modified by this function 491 * @size: size of @ivalue 492 * @task_ns: user namespace of the caller 493 * @mnt_userns: user namespace of the mount the inode was found from 494 * @fs_userns: user namespace of the filesystem 495 * 496 * If the inode has been found through an idmapped mount the user namespace of 497 * the vfsmount must be passed through @mnt_userns. This function will then 498 * take care to map the inode according to @mnt_userns before checking 499 * permissions. On non-idmapped mounts or if permission checking is to be 500 * performed on the raw inode simply passs init_user_ns. 501 */ 502 static kuid_t rootid_from_xattr(const void *value, size_t size, 503 struct user_namespace *task_ns, 504 struct user_namespace *mnt_userns, 505 struct user_namespace *fs_userns) 506 { 507 const struct vfs_ns_cap_data *nscap = value; 508 kuid_t rootkid; 509 uid_t rootid = 0; 510 511 if (size == XATTR_CAPS_SZ_3) 512 rootid = le32_to_cpu(nscap->rootid); 513 514 rootkid = make_kuid(task_ns, rootid); 515 return mapped_kuid_user(mnt_userns, fs_userns, rootkid); 516 } 517 518 static bool validheader(size_t size, const struct vfs_cap_data *cap) 519 { 520 return is_v2header(size, cap) || is_v3header(size, cap); 521 } 522 523 /** 524 * cap_convert_nscap - check vfs caps 525 * 526 * @mnt_userns: user namespace of the mount the inode was found from 527 * @dentry: used to retrieve inode to check permissions on 528 * @ivalue: vfs caps value which may be modified by this function 529 * @size: size of @ivalue 530 * 531 * User requested a write of security.capability. If needed, update the 532 * xattr to change from v2 to v3, or to fixup the v3 rootid. 533 * 534 * If the inode has been found through an idmapped mount the user namespace of 535 * the vfsmount must be passed through @mnt_userns. This function will then 536 * take care to map the inode according to @mnt_userns before checking 537 * permissions. On non-idmapped mounts or if permission checking is to be 538 * performed on the raw inode simply passs init_user_ns. 539 * 540 * Return: On success, return the new size; on error, return < 0. 541 */ 542 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry, 543 const void **ivalue, size_t size) 544 { 545 struct vfs_ns_cap_data *nscap; 546 uid_t nsrootid; 547 const struct vfs_cap_data *cap = *ivalue; 548 __u32 magic, nsmagic; 549 struct inode *inode = d_backing_inode(dentry); 550 struct user_namespace *task_ns = current_user_ns(), 551 *fs_ns = inode->i_sb->s_user_ns; 552 kuid_t rootid; 553 size_t newsize; 554 555 if (!*ivalue) 556 return -EINVAL; 557 if (!validheader(size, cap)) 558 return -EINVAL; 559 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP)) 560 return -EPERM; 561 if (size == XATTR_CAPS_SZ_2 && (mnt_userns == fs_ns)) 562 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP)) 563 /* user is privileged, just write the v2 */ 564 return size; 565 566 rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns, fs_ns); 567 if (!uid_valid(rootid)) 568 return -EINVAL; 569 570 nsrootid = from_kuid(fs_ns, rootid); 571 if (nsrootid == -1) 572 return -EINVAL; 573 574 newsize = sizeof(struct vfs_ns_cap_data); 575 nscap = kmalloc(newsize, GFP_ATOMIC); 576 if (!nscap) 577 return -ENOMEM; 578 nscap->rootid = cpu_to_le32(nsrootid); 579 nsmagic = VFS_CAP_REVISION_3; 580 magic = le32_to_cpu(cap->magic_etc); 581 if (magic & VFS_CAP_FLAGS_EFFECTIVE) 582 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; 583 nscap->magic_etc = cpu_to_le32(nsmagic); 584 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 585 586 *ivalue = nscap; 587 return newsize; 588 } 589 590 /* 591 * Calculate the new process capability sets from the capability sets attached 592 * to a file. 593 */ 594 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, 595 struct linux_binprm *bprm, 596 bool *effective, 597 bool *has_fcap) 598 { 599 struct cred *new = bprm->cred; 600 unsigned i; 601 int ret = 0; 602 603 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) 604 *effective = true; 605 606 if (caps->magic_etc & VFS_CAP_REVISION_MASK) 607 *has_fcap = true; 608 609 CAP_FOR_EACH_U32(i) { 610 __u32 permitted = caps->permitted.cap[i]; 611 __u32 inheritable = caps->inheritable.cap[i]; 612 613 /* 614 * pP' = (X & fP) | (pI & fI) 615 * The addition of pA' is handled later. 616 */ 617 new->cap_permitted.cap[i] = 618 (new->cap_bset.cap[i] & permitted) | 619 (new->cap_inheritable.cap[i] & inheritable); 620 621 if (permitted & ~new->cap_permitted.cap[i]) 622 /* insufficient to execute correctly */ 623 ret = -EPERM; 624 } 625 626 /* 627 * For legacy apps, with no internal support for recognizing they 628 * do not have enough capabilities, we return an error if they are 629 * missing some "forced" (aka file-permitted) capabilities. 630 */ 631 return *effective ? ret : 0; 632 } 633 634 /** 635 * get_vfs_caps_from_disk - retrieve vfs caps from disk 636 * 637 * @mnt_userns: user namespace of the mount the inode was found from 638 * @dentry: dentry from which @inode is retrieved 639 * @cpu_caps: vfs capabilities 640 * 641 * Extract the on-exec-apply capability sets for an executable file. 642 * 643 * If the inode has been found through an idmapped mount the user namespace of 644 * the vfsmount must be passed through @mnt_userns. This function will then 645 * take care to map the inode according to @mnt_userns before checking 646 * permissions. On non-idmapped mounts or if permission checking is to be 647 * performed on the raw inode simply passs init_user_ns. 648 */ 649 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns, 650 const struct dentry *dentry, 651 struct cpu_vfs_cap_data *cpu_caps) 652 { 653 struct inode *inode = d_backing_inode(dentry); 654 __u32 magic_etc; 655 unsigned tocopy, i; 656 int size; 657 struct vfs_ns_cap_data data, *nscaps = &data; 658 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data; 659 kuid_t rootkuid; 660 struct user_namespace *fs_ns; 661 662 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); 663 664 if (!inode) 665 return -ENODATA; 666 667 fs_ns = inode->i_sb->s_user_ns; 668 size = __vfs_getxattr((struct dentry *)dentry, inode, 669 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ); 670 if (size == -ENODATA || size == -EOPNOTSUPP) 671 /* no data, that's ok */ 672 return -ENODATA; 673 674 if (size < 0) 675 return size; 676 677 if (size < sizeof(magic_etc)) 678 return -EINVAL; 679 680 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc); 681 682 rootkuid = make_kuid(fs_ns, 0); 683 switch (magic_etc & VFS_CAP_REVISION_MASK) { 684 case VFS_CAP_REVISION_1: 685 if (size != XATTR_CAPS_SZ_1) 686 return -EINVAL; 687 tocopy = VFS_CAP_U32_1; 688 break; 689 case VFS_CAP_REVISION_2: 690 if (size != XATTR_CAPS_SZ_2) 691 return -EINVAL; 692 tocopy = VFS_CAP_U32_2; 693 break; 694 case VFS_CAP_REVISION_3: 695 if (size != XATTR_CAPS_SZ_3) 696 return -EINVAL; 697 tocopy = VFS_CAP_U32_3; 698 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid)); 699 break; 700 701 default: 702 return -EINVAL; 703 } 704 /* Limit the caps to the mounter of the filesystem 705 * or the more limited uid specified in the xattr. 706 */ 707 rootkuid = mapped_kuid_fs(mnt_userns, fs_ns, rootkuid); 708 if (!rootid_owns_currentns(rootkuid)) 709 return -ENODATA; 710 711 CAP_FOR_EACH_U32(i) { 712 if (i >= tocopy) 713 break; 714 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted); 715 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable); 716 } 717 718 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 719 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 720 721 cpu_caps->rootid = rootkuid; 722 723 return 0; 724 } 725 726 /* 727 * Attempt to get the on-exec apply capability sets for an executable file from 728 * its xattrs and, if present, apply them to the proposed credentials being 729 * constructed by execve(). 730 */ 731 static int get_file_caps(struct linux_binprm *bprm, struct file *file, 732 bool *effective, bool *has_fcap) 733 { 734 int rc = 0; 735 struct cpu_vfs_cap_data vcaps; 736 737 cap_clear(bprm->cred->cap_permitted); 738 739 if (!file_caps_enabled) 740 return 0; 741 742 if (!mnt_may_suid(file->f_path.mnt)) 743 return 0; 744 745 /* 746 * This check is redundant with mnt_may_suid() but is kept to make 747 * explicit that capability bits are limited to s_user_ns and its 748 * descendants. 749 */ 750 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns)) 751 return 0; 752 753 rc = get_vfs_caps_from_disk(file_mnt_user_ns(file), 754 file->f_path.dentry, &vcaps); 755 if (rc < 0) { 756 if (rc == -EINVAL) 757 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n", 758 bprm->filename); 759 else if (rc == -ENODATA) 760 rc = 0; 761 goto out; 762 } 763 764 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap); 765 766 out: 767 if (rc) 768 cap_clear(bprm->cred->cap_permitted); 769 770 return rc; 771 } 772 773 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); } 774 775 static inline bool __is_real(kuid_t uid, struct cred *cred) 776 { return uid_eq(cred->uid, uid); } 777 778 static inline bool __is_eff(kuid_t uid, struct cred *cred) 779 { return uid_eq(cred->euid, uid); } 780 781 static inline bool __is_suid(kuid_t uid, struct cred *cred) 782 { return !__is_real(uid, cred) && __is_eff(uid, cred); } 783 784 /* 785 * handle_privileged_root - Handle case of privileged root 786 * @bprm: The execution parameters, including the proposed creds 787 * @has_fcap: Are any file capabilities set? 788 * @effective: Do we have effective root privilege? 789 * @root_uid: This namespace' root UID WRT initial USER namespace 790 * 791 * Handle the case where root is privileged and hasn't been neutered by 792 * SECURE_NOROOT. If file capabilities are set, they won't be combined with 793 * set UID root and nothing is changed. If we are root, cap_permitted is 794 * updated. If we have become set UID root, the effective bit is set. 795 */ 796 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap, 797 bool *effective, kuid_t root_uid) 798 { 799 const struct cred *old = current_cred(); 800 struct cred *new = bprm->cred; 801 802 if (!root_privileged()) 803 return; 804 /* 805 * If the legacy file capability is set, then don't set privs 806 * for a setuid root binary run by a non-root user. Do set it 807 * for a root user just to cause least surprise to an admin. 808 */ 809 if (has_fcap && __is_suid(root_uid, new)) { 810 warn_setuid_and_fcaps_mixed(bprm->filename); 811 return; 812 } 813 /* 814 * To support inheritance of root-permissions and suid-root 815 * executables under compatibility mode, we override the 816 * capability sets for the file. 817 */ 818 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) { 819 /* pP' = (cap_bset & ~0) | (pI & ~0) */ 820 new->cap_permitted = cap_combine(old->cap_bset, 821 old->cap_inheritable); 822 } 823 /* 824 * If only the real uid is 0, we do not set the effective bit. 825 */ 826 if (__is_eff(root_uid, new)) 827 *effective = true; 828 } 829 830 #define __cap_gained(field, target, source) \ 831 !cap_issubset(target->cap_##field, source->cap_##field) 832 #define __cap_grew(target, source, cred) \ 833 !cap_issubset(cred->cap_##target, cred->cap_##source) 834 #define __cap_full(field, cred) \ 835 cap_issubset(CAP_FULL_SET, cred->cap_##field) 836 837 static inline bool __is_setuid(struct cred *new, const struct cred *old) 838 { return !uid_eq(new->euid, old->uid); } 839 840 static inline bool __is_setgid(struct cred *new, const struct cred *old) 841 { return !gid_eq(new->egid, old->gid); } 842 843 /* 844 * 1) Audit candidate if current->cap_effective is set 845 * 846 * We do not bother to audit if 3 things are true: 847 * 1) cap_effective has all caps 848 * 2) we became root *OR* are were already root 849 * 3) root is supposed to have all caps (SECURE_NOROOT) 850 * Since this is just a normal root execing a process. 851 * 852 * Number 1 above might fail if you don't have a full bset, but I think 853 * that is interesting information to audit. 854 * 855 * A number of other conditions require logging: 856 * 2) something prevented setuid root getting all caps 857 * 3) non-setuid root gets fcaps 858 * 4) non-setuid root gets ambient 859 */ 860 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old, 861 kuid_t root, bool has_fcap) 862 { 863 bool ret = false; 864 865 if ((__cap_grew(effective, ambient, new) && 866 !(__cap_full(effective, new) && 867 (__is_eff(root, new) || __is_real(root, new)) && 868 root_privileged())) || 869 (root_privileged() && 870 __is_suid(root, new) && 871 !__cap_full(effective, new)) || 872 (!__is_setuid(new, old) && 873 ((has_fcap && 874 __cap_gained(permitted, new, old)) || 875 __cap_gained(ambient, new, old)))) 876 877 ret = true; 878 879 return ret; 880 } 881 882 /** 883 * cap_bprm_creds_from_file - Set up the proposed credentials for execve(). 884 * @bprm: The execution parameters, including the proposed creds 885 * @file: The file to pull the credentials from 886 * 887 * Set up the proposed credentials for a new execution context being 888 * constructed by execve(). The proposed creds in @bprm->cred is altered, 889 * which won't take effect immediately. 890 * 891 * Return: 0 if successful, -ve on error. 892 */ 893 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file) 894 { 895 /* Process setpcap binaries and capabilities for uid 0 */ 896 const struct cred *old = current_cred(); 897 struct cred *new = bprm->cred; 898 bool effective = false, has_fcap = false, is_setid; 899 int ret; 900 kuid_t root_uid; 901 902 if (WARN_ON(!cap_ambient_invariant_ok(old))) 903 return -EPERM; 904 905 ret = get_file_caps(bprm, file, &effective, &has_fcap); 906 if (ret < 0) 907 return ret; 908 909 root_uid = make_kuid(new->user_ns, 0); 910 911 handle_privileged_root(bprm, has_fcap, &effective, root_uid); 912 913 /* if we have fs caps, clear dangerous personality flags */ 914 if (__cap_gained(permitted, new, old)) 915 bprm->per_clear |= PER_CLEAR_ON_SETID; 916 917 /* Don't let someone trace a set[ug]id/setpcap binary with the revised 918 * credentials unless they have the appropriate permit. 919 * 920 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. 921 */ 922 is_setid = __is_setuid(new, old) || __is_setgid(new, old); 923 924 if ((is_setid || __cap_gained(permitted, new, old)) && 925 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) || 926 !ptracer_capable(current, new->user_ns))) { 927 /* downgrade; they get no more than they had, and maybe less */ 928 if (!ns_capable(new->user_ns, CAP_SETUID) || 929 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { 930 new->euid = new->uid; 931 new->egid = new->gid; 932 } 933 new->cap_permitted = cap_intersect(new->cap_permitted, 934 old->cap_permitted); 935 } 936 937 new->suid = new->fsuid = new->euid; 938 new->sgid = new->fsgid = new->egid; 939 940 /* File caps or setid cancels ambient. */ 941 if (has_fcap || is_setid) 942 cap_clear(new->cap_ambient); 943 944 /* 945 * Now that we've computed pA', update pP' to give: 946 * pP' = (X & fP) | (pI & fI) | pA' 947 */ 948 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); 949 950 /* 951 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, 952 * this is the same as pE' = (fE ? pP' : 0) | pA'. 953 */ 954 if (effective) 955 new->cap_effective = new->cap_permitted; 956 else 957 new->cap_effective = new->cap_ambient; 958 959 if (WARN_ON(!cap_ambient_invariant_ok(new))) 960 return -EPERM; 961 962 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) { 963 ret = audit_log_bprm_fcaps(bprm, new, old); 964 if (ret < 0) 965 return ret; 966 } 967 968 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 969 970 if (WARN_ON(!cap_ambient_invariant_ok(new))) 971 return -EPERM; 972 973 /* Check for privilege-elevated exec. */ 974 if (is_setid || 975 (!__is_real(root_uid, new) && 976 (effective || 977 __cap_grew(permitted, ambient, new)))) 978 bprm->secureexec = 1; 979 980 return 0; 981 } 982 983 /** 984 * cap_inode_setxattr - Determine whether an xattr may be altered 985 * @dentry: The inode/dentry being altered 986 * @name: The name of the xattr to be changed 987 * @value: The value that the xattr will be changed to 988 * @size: The size of value 989 * @flags: The replacement flag 990 * 991 * Determine whether an xattr may be altered or set on an inode, returning 0 if 992 * permission is granted, -ve if denied. 993 * 994 * This is used to make sure security xattrs don't get updated or set by those 995 * who aren't privileged to do so. 996 */ 997 int cap_inode_setxattr(struct dentry *dentry, const char *name, 998 const void *value, size_t size, int flags) 999 { 1000 struct user_namespace *user_ns = dentry->d_sb->s_user_ns; 1001 1002 /* Ignore non-security xattrs */ 1003 if (strncmp(name, XATTR_SECURITY_PREFIX, 1004 XATTR_SECURITY_PREFIX_LEN) != 0) 1005 return 0; 1006 1007 /* 1008 * For XATTR_NAME_CAPS the check will be done in 1009 * cap_convert_nscap(), called by setxattr() 1010 */ 1011 if (strcmp(name, XATTR_NAME_CAPS) == 0) 1012 return 0; 1013 1014 if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 1015 return -EPERM; 1016 return 0; 1017 } 1018 1019 /** 1020 * cap_inode_removexattr - Determine whether an xattr may be removed 1021 * 1022 * @mnt_userns: User namespace of the mount the inode was found from 1023 * @dentry: The inode/dentry being altered 1024 * @name: The name of the xattr to be changed 1025 * 1026 * Determine whether an xattr may be removed from an inode, returning 0 if 1027 * permission is granted, -ve if denied. 1028 * 1029 * If the inode has been found through an idmapped mount the user namespace of 1030 * the vfsmount must be passed through @mnt_userns. This function will then 1031 * take care to map the inode according to @mnt_userns before checking 1032 * permissions. On non-idmapped mounts or if permission checking is to be 1033 * performed on the raw inode simply passs init_user_ns. 1034 * 1035 * This is used to make sure security xattrs don't get removed by those who 1036 * aren't privileged to remove them. 1037 */ 1038 int cap_inode_removexattr(struct user_namespace *mnt_userns, 1039 struct dentry *dentry, const char *name) 1040 { 1041 struct user_namespace *user_ns = dentry->d_sb->s_user_ns; 1042 1043 /* Ignore non-security xattrs */ 1044 if (strncmp(name, XATTR_SECURITY_PREFIX, 1045 XATTR_SECURITY_PREFIX_LEN) != 0) 1046 return 0; 1047 1048 if (strcmp(name, XATTR_NAME_CAPS) == 0) { 1049 /* security.capability gets namespaced */ 1050 struct inode *inode = d_backing_inode(dentry); 1051 if (!inode) 1052 return -EINVAL; 1053 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP)) 1054 return -EPERM; 1055 return 0; 1056 } 1057 1058 if (!ns_capable(user_ns, CAP_SYS_ADMIN)) 1059 return -EPERM; 1060 return 0; 1061 } 1062 1063 /* 1064 * cap_emulate_setxuid() fixes the effective / permitted capabilities of 1065 * a process after a call to setuid, setreuid, or setresuid. 1066 * 1067 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of 1068 * {r,e,s}uid != 0, the permitted and effective capabilities are 1069 * cleared. 1070 * 1071 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective 1072 * capabilities of the process are cleared. 1073 * 1074 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective 1075 * capabilities are set to the permitted capabilities. 1076 * 1077 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 1078 * never happen. 1079 * 1080 * -astor 1081 * 1082 * cevans - New behaviour, Oct '99 1083 * A process may, via prctl(), elect to keep its capabilities when it 1084 * calls setuid() and switches away from uid==0. Both permitted and 1085 * effective sets will be retained. 1086 * Without this change, it was impossible for a daemon to drop only some 1087 * of its privilege. The call to setuid(!=0) would drop all privileges! 1088 * Keeping uid 0 is not an option because uid 0 owns too many vital 1089 * files.. 1090 * Thanks to Olaf Kirch and Peter Benie for spotting this. 1091 */ 1092 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) 1093 { 1094 kuid_t root_uid = make_kuid(old->user_ns, 0); 1095 1096 if ((uid_eq(old->uid, root_uid) || 1097 uid_eq(old->euid, root_uid) || 1098 uid_eq(old->suid, root_uid)) && 1099 (!uid_eq(new->uid, root_uid) && 1100 !uid_eq(new->euid, root_uid) && 1101 !uid_eq(new->suid, root_uid))) { 1102 if (!issecure(SECURE_KEEP_CAPS)) { 1103 cap_clear(new->cap_permitted); 1104 cap_clear(new->cap_effective); 1105 } 1106 1107 /* 1108 * Pre-ambient programs expect setresuid to nonroot followed 1109 * by exec to drop capabilities. We should make sure that 1110 * this remains the case. 1111 */ 1112 cap_clear(new->cap_ambient); 1113 } 1114 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) 1115 cap_clear(new->cap_effective); 1116 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) 1117 new->cap_effective = new->cap_permitted; 1118 } 1119 1120 /** 1121 * cap_task_fix_setuid - Fix up the results of setuid() call 1122 * @new: The proposed credentials 1123 * @old: The current task's current credentials 1124 * @flags: Indications of what has changed 1125 * 1126 * Fix up the results of setuid() call before the credential changes are 1127 * actually applied. 1128 * 1129 * Return: 0 to grant the changes, -ve to deny them. 1130 */ 1131 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) 1132 { 1133 switch (flags) { 1134 case LSM_SETID_RE: 1135 case LSM_SETID_ID: 1136 case LSM_SETID_RES: 1137 /* juggle the capabilities to follow [RES]UID changes unless 1138 * otherwise suppressed */ 1139 if (!issecure(SECURE_NO_SETUID_FIXUP)) 1140 cap_emulate_setxuid(new, old); 1141 break; 1142 1143 case LSM_SETID_FS: 1144 /* juggle the capabilties to follow FSUID changes, unless 1145 * otherwise suppressed 1146 * 1147 * FIXME - is fsuser used for all CAP_FS_MASK capabilities? 1148 * if not, we might be a bit too harsh here. 1149 */ 1150 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 1151 kuid_t root_uid = make_kuid(old->user_ns, 0); 1152 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) 1153 new->cap_effective = 1154 cap_drop_fs_set(new->cap_effective); 1155 1156 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) 1157 new->cap_effective = 1158 cap_raise_fs_set(new->cap_effective, 1159 new->cap_permitted); 1160 } 1161 break; 1162 1163 default: 1164 return -EINVAL; 1165 } 1166 1167 return 0; 1168 } 1169 1170 /* 1171 * Rationale: code calling task_setscheduler, task_setioprio, and 1172 * task_setnice, assumes that 1173 * . if capable(cap_sys_nice), then those actions should be allowed 1174 * . if not capable(cap_sys_nice), but acting on your own processes, 1175 * then those actions should be allowed 1176 * This is insufficient now since you can call code without suid, but 1177 * yet with increased caps. 1178 * So we check for increased caps on the target process. 1179 */ 1180 static int cap_safe_nice(struct task_struct *p) 1181 { 1182 int is_subset, ret = 0; 1183 1184 rcu_read_lock(); 1185 is_subset = cap_issubset(__task_cred(p)->cap_permitted, 1186 current_cred()->cap_permitted); 1187 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) 1188 ret = -EPERM; 1189 rcu_read_unlock(); 1190 1191 return ret; 1192 } 1193 1194 /** 1195 * cap_task_setscheduler - Detemine if scheduler policy change is permitted 1196 * @p: The task to affect 1197 * 1198 * Detemine if the requested scheduler policy change is permitted for the 1199 * specified task. 1200 * 1201 * Return: 0 if permission is granted, -ve if denied. 1202 */ 1203 int cap_task_setscheduler(struct task_struct *p) 1204 { 1205 return cap_safe_nice(p); 1206 } 1207 1208 /** 1209 * cap_task_setioprio - Detemine if I/O priority change is permitted 1210 * @p: The task to affect 1211 * @ioprio: The I/O priority to set 1212 * 1213 * Detemine if the requested I/O priority change is permitted for the specified 1214 * task. 1215 * 1216 * Return: 0 if permission is granted, -ve if denied. 1217 */ 1218 int cap_task_setioprio(struct task_struct *p, int ioprio) 1219 { 1220 return cap_safe_nice(p); 1221 } 1222 1223 /** 1224 * cap_task_setnice - Detemine if task priority change is permitted 1225 * @p: The task to affect 1226 * @nice: The nice value to set 1227 * 1228 * Detemine if the requested task priority change is permitted for the 1229 * specified task. 1230 * 1231 * Return: 0 if permission is granted, -ve if denied. 1232 */ 1233 int cap_task_setnice(struct task_struct *p, int nice) 1234 { 1235 return cap_safe_nice(p); 1236 } 1237 1238 /* 1239 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from 1240 * the current task's bounding set. Returns 0 on success, -ve on error. 1241 */ 1242 static int cap_prctl_drop(unsigned long cap) 1243 { 1244 struct cred *new; 1245 1246 if (!ns_capable(current_user_ns(), CAP_SETPCAP)) 1247 return -EPERM; 1248 if (!cap_valid(cap)) 1249 return -EINVAL; 1250 1251 new = prepare_creds(); 1252 if (!new) 1253 return -ENOMEM; 1254 cap_lower(new->cap_bset, cap); 1255 return commit_creds(new); 1256 } 1257 1258 /** 1259 * cap_task_prctl - Implement process control functions for this security module 1260 * @option: The process control function requested 1261 * @arg2: The argument data for this function 1262 * @arg3: The argument data for this function 1263 * @arg4: The argument data for this function 1264 * @arg5: The argument data for this function 1265 * 1266 * Allow process control functions (sys_prctl()) to alter capabilities; may 1267 * also deny access to other functions not otherwise implemented here. 1268 * 1269 * Return: 0 or +ve on success, -ENOSYS if this function is not implemented 1270 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM 1271 * modules will consider performing the function. 1272 */ 1273 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, 1274 unsigned long arg4, unsigned long arg5) 1275 { 1276 const struct cred *old = current_cred(); 1277 struct cred *new; 1278 1279 switch (option) { 1280 case PR_CAPBSET_READ: 1281 if (!cap_valid(arg2)) 1282 return -EINVAL; 1283 return !!cap_raised(old->cap_bset, arg2); 1284 1285 case PR_CAPBSET_DROP: 1286 return cap_prctl_drop(arg2); 1287 1288 /* 1289 * The next four prctl's remain to assist with transitioning a 1290 * system from legacy UID=0 based privilege (when filesystem 1291 * capabilities are not in use) to a system using filesystem 1292 * capabilities only - as the POSIX.1e draft intended. 1293 * 1294 * Note: 1295 * 1296 * PR_SET_SECUREBITS = 1297 * issecure_mask(SECURE_KEEP_CAPS_LOCKED) 1298 * | issecure_mask(SECURE_NOROOT) 1299 * | issecure_mask(SECURE_NOROOT_LOCKED) 1300 * | issecure_mask(SECURE_NO_SETUID_FIXUP) 1301 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) 1302 * 1303 * will ensure that the current process and all of its 1304 * children will be locked into a pure 1305 * capability-based-privilege environment. 1306 */ 1307 case PR_SET_SECUREBITS: 1308 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) 1309 & (old->securebits ^ arg2)) /*[1]*/ 1310 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ 1311 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ 1312 || (cap_capable(current_cred(), 1313 current_cred()->user_ns, 1314 CAP_SETPCAP, 1315 CAP_OPT_NONE) != 0) /*[4]*/ 1316 /* 1317 * [1] no changing of bits that are locked 1318 * [2] no unlocking of locks 1319 * [3] no setting of unsupported bits 1320 * [4] doing anything requires privilege (go read about 1321 * the "sendmail capabilities bug") 1322 */ 1323 ) 1324 /* cannot change a locked bit */ 1325 return -EPERM; 1326 1327 new = prepare_creds(); 1328 if (!new) 1329 return -ENOMEM; 1330 new->securebits = arg2; 1331 return commit_creds(new); 1332 1333 case PR_GET_SECUREBITS: 1334 return old->securebits; 1335 1336 case PR_GET_KEEPCAPS: 1337 return !!issecure(SECURE_KEEP_CAPS); 1338 1339 case PR_SET_KEEPCAPS: 1340 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ 1341 return -EINVAL; 1342 if (issecure(SECURE_KEEP_CAPS_LOCKED)) 1343 return -EPERM; 1344 1345 new = prepare_creds(); 1346 if (!new) 1347 return -ENOMEM; 1348 if (arg2) 1349 new->securebits |= issecure_mask(SECURE_KEEP_CAPS); 1350 else 1351 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 1352 return commit_creds(new); 1353 1354 case PR_CAP_AMBIENT: 1355 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { 1356 if (arg3 | arg4 | arg5) 1357 return -EINVAL; 1358 1359 new = prepare_creds(); 1360 if (!new) 1361 return -ENOMEM; 1362 cap_clear(new->cap_ambient); 1363 return commit_creds(new); 1364 } 1365 1366 if (((!cap_valid(arg3)) | arg4 | arg5)) 1367 return -EINVAL; 1368 1369 if (arg2 == PR_CAP_AMBIENT_IS_SET) { 1370 return !!cap_raised(current_cred()->cap_ambient, arg3); 1371 } else if (arg2 != PR_CAP_AMBIENT_RAISE && 1372 arg2 != PR_CAP_AMBIENT_LOWER) { 1373 return -EINVAL; 1374 } else { 1375 if (arg2 == PR_CAP_AMBIENT_RAISE && 1376 (!cap_raised(current_cred()->cap_permitted, arg3) || 1377 !cap_raised(current_cred()->cap_inheritable, 1378 arg3) || 1379 issecure(SECURE_NO_CAP_AMBIENT_RAISE))) 1380 return -EPERM; 1381 1382 new = prepare_creds(); 1383 if (!new) 1384 return -ENOMEM; 1385 if (arg2 == PR_CAP_AMBIENT_RAISE) 1386 cap_raise(new->cap_ambient, arg3); 1387 else 1388 cap_lower(new->cap_ambient, arg3); 1389 return commit_creds(new); 1390 } 1391 1392 default: 1393 /* No functionality available - continue with default */ 1394 return -ENOSYS; 1395 } 1396 } 1397 1398 /** 1399 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted 1400 * @mm: The VM space in which the new mapping is to be made 1401 * @pages: The size of the mapping 1402 * 1403 * Determine whether the allocation of a new virtual mapping by the current 1404 * task is permitted. 1405 * 1406 * Return: 1 if permission is granted, 0 if not. 1407 */ 1408 int cap_vm_enough_memory(struct mm_struct *mm, long pages) 1409 { 1410 int cap_sys_admin = 0; 1411 1412 if (cap_capable(current_cred(), &init_user_ns, 1413 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0) 1414 cap_sys_admin = 1; 1415 1416 return cap_sys_admin; 1417 } 1418 1419 /** 1420 * cap_mmap_addr - check if able to map given addr 1421 * @addr: address attempting to be mapped 1422 * 1423 * If the process is attempting to map memory below dac_mmap_min_addr they need 1424 * CAP_SYS_RAWIO. The other parameters to this function are unused by the 1425 * capability security module. 1426 * 1427 * Return: 0 if this mapping should be allowed or -EPERM if not. 1428 */ 1429 int cap_mmap_addr(unsigned long addr) 1430 { 1431 int ret = 0; 1432 1433 if (addr < dac_mmap_min_addr) { 1434 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, 1435 CAP_OPT_NONE); 1436 /* set PF_SUPERPRIV if it turns out we allow the low mmap */ 1437 if (ret == 0) 1438 current->flags |= PF_SUPERPRIV; 1439 } 1440 return ret; 1441 } 1442 1443 int cap_mmap_file(struct file *file, unsigned long reqprot, 1444 unsigned long prot, unsigned long flags) 1445 { 1446 return 0; 1447 } 1448 1449 #ifdef CONFIG_SECURITY 1450 1451 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = { 1452 LSM_HOOK_INIT(capable, cap_capable), 1453 LSM_HOOK_INIT(settime, cap_settime), 1454 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), 1455 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), 1456 LSM_HOOK_INIT(capget, cap_capget), 1457 LSM_HOOK_INIT(capset, cap_capset), 1458 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file), 1459 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), 1460 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), 1461 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity), 1462 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), 1463 LSM_HOOK_INIT(mmap_file, cap_mmap_file), 1464 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), 1465 LSM_HOOK_INIT(task_prctl, cap_task_prctl), 1466 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), 1467 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), 1468 LSM_HOOK_INIT(task_setnice, cap_task_setnice), 1469 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), 1470 }; 1471 1472 static int __init capability_init(void) 1473 { 1474 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks), 1475 "capability"); 1476 return 0; 1477 } 1478 1479 DEFINE_LSM(capability) = { 1480 .name = "capability", 1481 .order = LSM_ORDER_FIRST, 1482 .init = capability_init, 1483 }; 1484 1485 #endif /* CONFIG_SECURITY */ 1486