fs/posix_acl.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2002,2003 by Andreas Gruenbacher <a.gruenbacher@computer.org>
   4  *
   5  * Fixes from William Schumacher incorporated on 15 March 2001.
   6  *    (Reported by Charles Bertsch, <CBertsch@microtest.com>).
   7  */
   8
   9 /*
  10  *  This file contains generic functions for manipulating
  11  *  POSIX 1003.1e draft standard 17 ACLs.
  12  */
  13
  14 #include <linux/kernel.h>
  15 #include <linux/slab.h>
  16 #include <linux/atomic.h>
  17 #include <linux/fs.h>
  18 #include <linux/sched.h>
  19 #include <linux/cred.h>
  20 #include <linux/posix_acl.h>
  21 #include <linux/posix_acl_xattr.h>
  22 #include <linux/xattr.h>
  23 #include <linux/export.h>
  24 #include <linux/user_namespace.h>
  25 #include <linux/namei.h>
  26 #include <linux/mnt_idmapping.h>
  27 #include <linux/iversion.h>
  28 #include <linux/security.h>
  29 #include <linux/fsnotify.h>
  30 #include <linux/filelock.h>
  31
  32 #include "internal.h"
  33
  34 static struct posix_acl **acl_by_type(struct inode *inode, int type)
  35 {
  36         switch (type) {
  37         case ACL_TYPE_ACCESS:
  38                 return &inode->i_acl;
  39         case ACL_TYPE_DEFAULT:
  40                 return &inode->i_default_acl;
  41         default:
  42                 BUG();
  43         }
  44 }
  45
  46 struct posix_acl *get_cached_acl(struct inode *inode, int type)
  47 {
  48         struct posix_acl **p = acl_by_type(inode, type);
  49         struct posix_acl *acl;
  50
  51         for (;;) {
  52                 rcu_read_lock();
  53                 acl = rcu_dereference(*p);
  54                 if (!acl || is_uncached_acl(acl) ||
  55                     refcount_inc_not_zero(&acl->a_refcount))
  56                         break;
  57                 rcu_read_unlock();
  58                 cpu_relax();
  59         }
  60         rcu_read_unlock();
  61         return acl;
  62 }
  63 EXPORT_SYMBOL(get_cached_acl);
  64
  65 struct posix_acl *get_cached_acl_rcu(struct inode *inode, int type)
  66 {
  67         struct posix_acl *acl = rcu_dereference(*acl_by_type(inode, type));
  68
  69         if (acl == ACL_DONT_CACHE) {
  70                 struct posix_acl *ret;
  71
  72                 ret = inode->i_op->get_inode_acl(inode, type, LOOKUP_RCU);
  73                 if (!IS_ERR(ret))
  74                         acl = ret;
  75         }
  76
  77         return acl;
  78 }
  79 EXPORT_SYMBOL(get_cached_acl_rcu);
  80
  81 void set_cached_acl(struct inode *inode, int type, struct posix_acl *acl)
  82 {
  83         struct posix_acl **p = acl_by_type(inode, type);
  84         struct posix_acl *old;
  85
  86         old = xchg(p, posix_acl_dup(acl));
  87         if (!is_uncached_acl(old))
  88                 posix_acl_release(old);
  89 }
  90 EXPORT_SYMBOL(set_cached_acl);
  91
  92 static void __forget_cached_acl(struct posix_acl **p)
  93 {
  94         struct posix_acl *old;
  95
  96         old = xchg(p, ACL_NOT_CACHED);
  97         if (!is_uncached_acl(old))
  98                 posix_acl_release(old);
  99 }
 100
 101 void forget_cached_acl(struct inode *inode, int type)
 102 {
 103         __forget_cached_acl(acl_by_type(inode, type));
 104 }
 105 EXPORT_SYMBOL(forget_cached_acl);
 106
 107 void forget_all_cached_acls(struct inode *inode)
 108 {
 109         __forget_cached_acl(&inode->i_acl);
 110         __forget_cached_acl(&inode->i_default_acl);
 111 }
 112 EXPORT_SYMBOL(forget_all_cached_acls);
 113
 114 static struct posix_acl *__get_acl(struct mnt_idmap *idmap,
 115                                    struct dentry *dentry, struct inode *inode,
 116                                    int type)
 117 {
 118         struct posix_acl *sentinel;
 119         struct posix_acl **p;
 120         struct posix_acl *acl;
 121
 122         /*
 123          * The sentinel is used to detect when another operation like
 124          * set_cached_acl() or forget_cached_acl() races with get_inode_acl().
 125          * It is guaranteed that is_uncached_acl(sentinel) is true.
 126          */
 127
 128         acl = get_cached_acl(inode, type);
 129         if (!is_uncached_acl(acl))
 130                 return acl;
 131
 132         if (!IS_POSIXACL(inode))
 133                 return NULL;
 134
 135         sentinel = uncached_acl_sentinel(current);
 136         p = acl_by_type(inode, type);
 137
 138         /*
 139          * If the ACL isn't being read yet, set our sentinel.  Otherwise, the
 140          * current value of the ACL will not be ACL_NOT_CACHED and so our own
 141          * sentinel will not be set; another task will update the cache.  We
 142          * could wait for that other task to complete its job, but it's easier
 143          * to just call ->get_inode_acl to fetch the ACL ourself.  (This is
 144          * going to be an unlikely race.)
 145          */
 146         cmpxchg(p, ACL_NOT_CACHED, sentinel);
 147
 148         /*
 149          * Normally, the ACL returned by ->get{_inode}_acl will be cached.
 150          * A filesystem can prevent that by calling
 151          * forget_cached_acl(inode, type) in ->get{_inode}_acl.
 152          *
 153          * If the filesystem doesn't have a get{_inode}_ acl() function at all,
 154          * we'll just create the negative cache entry.
 155          */
 156         if (dentry && inode->i_op->get_acl) {
 157                 acl = inode->i_op->get_acl(idmap, dentry, type);
 158         } else if (inode->i_op->get_inode_acl) {
 159                 acl = inode->i_op->get_inode_acl(inode, type, false);
 160         } else {
 161                 set_cached_acl(inode, type, NULL);
 162                 return NULL;
 163         }
 164         if (IS_ERR(acl)) {
 165                 /*
 166                  * Remove our sentinel so that we don't block future attempts
 167                  * to cache the ACL.
 168                  */
 169                 cmpxchg(p, sentinel, ACL_NOT_CACHED);
 170                 return acl;
 171         }
 172
 173         /*
 174          * Cache the result, but only if our sentinel is still in place.
 175          */
 176         posix_acl_dup(acl);
 177         if (unlikely(!try_cmpxchg(p, &sentinel, acl)))
 178                 posix_acl_release(acl);
 179         return acl;
 180 }
 181
 182 struct posix_acl *get_inode_acl(struct inode *inode, int type)
 183 {
 184         return __get_acl(&nop_mnt_idmap, NULL, inode, type);
 185 }
 186 EXPORT_SYMBOL(get_inode_acl);
 187
 188 /*
 189  * Init a fresh posix_acl
 190  */
 191 void
 192 posix_acl_init(struct posix_acl *acl, int count)
 193 {
 194         refcount_set(&acl->a_refcount, 1);
 195         acl->a_count = count;
 196 }
 197 EXPORT_SYMBOL(posix_acl_init);
 198
 199 /*
 200  * Allocate a new ACL with the specified number of entries.
 201  */
 202 struct posix_acl *
 203 posix_acl_alloc(unsigned int count, gfp_t flags)
 204 {
 205         struct posix_acl *acl;
 206
 207         acl = kmalloc(struct_size(acl, a_entries, count), flags);
 208         if (acl)
 209                 posix_acl_init(acl, count);
 210         return acl;
 211 }
 212 EXPORT_SYMBOL(posix_acl_alloc);
 213
 214 /*
 215  * Clone an ACL.
 216  */
 217 struct posix_acl *
 218 posix_acl_clone(const struct posix_acl *acl, gfp_t flags)
 219 {
 220         struct posix_acl *clone = NULL;
 221
 222         if (acl) {
 223                 clone = kmemdup(acl, struct_size(acl, a_entries, acl->a_count),
 224                                 flags);
 225                 if (clone)
 226                         refcount_set(&clone->a_refcount, 1);
 227         }
 228         return clone;
 229 }
 230 EXPORT_SYMBOL_GPL(posix_acl_clone);
 231
 232 /*
 233  * Check if an acl is valid. Returns 0 if it is, or -E... otherwise.
 234  */
 235 int
 236 posix_acl_valid(struct user_namespace *user_ns, const struct posix_acl *acl)
 237 {
 238         const struct posix_acl_entry *pa, *pe;
 239         int state = ACL_USER_OBJ;
 240         int needs_mask = 0;
 241
 242         FOREACH_ACL_ENTRY(pa, acl, pe) {
 243                 if (pa->e_perm & ~(ACL_READ|ACL_WRITE|ACL_EXECUTE))
 244                         return -EINVAL;
 245                 switch (pa->e_tag) {
 246                         case ACL_USER_OBJ:
 247                                 if (state == ACL_USER_OBJ) {
 248                                         state = ACL_USER;
 249                                         break;
 250                                 }
 251                                 return -EINVAL;
 252
 253                         case ACL_USER:
 254                                 if (state != ACL_USER)
 255                                         return -EINVAL;
 256                                 if (!kuid_has_mapping(user_ns, pa->e_uid))
 257                                         return -EINVAL;
 258                                 needs_mask = 1;
 259                                 break;
 260
 261                         case ACL_GROUP_OBJ:
 262                                 if (state == ACL_USER) {
 263                                         state = ACL_GROUP;
 264                                         break;
 265                                 }
 266                                 return -EINVAL;
 267
 268                         case ACL_GROUP:
 269                                 if (state != ACL_GROUP)
 270                                         return -EINVAL;
 271                                 if (!kgid_has_mapping(user_ns, pa->e_gid))
 272                                         return -EINVAL;
 273                                 needs_mask = 1;
 274                                 break;
 275
 276                         case ACL_MASK:
 277                                 if (state != ACL_GROUP)
 278                                         return -EINVAL;
 279                                 state = ACL_OTHER;
 280                                 break;
 281
 282                         case ACL_OTHER:
 283                                 if (state == ACL_OTHER ||
 284                                     (state == ACL_GROUP && !needs_mask)) {
 285                                         state = 0;
 286                                         break;
 287                                 }
 288                                 return -EINVAL;
 289
 290                         default:
 291                                 return -EINVAL;
 292                 }
 293         }
 294         if (state == 0)
 295                 return 0;
 296         return -EINVAL;
 297 }
 298 EXPORT_SYMBOL(posix_acl_valid);
 299
 300 /*
 301  * Returns 0 if the acl can be exactly represented in the traditional
 302  * file mode permission bits, or else 1. Returns -E... on error.
 303  */
 304 int
 305 posix_acl_equiv_mode(const struct posix_acl *acl, umode_t *mode_p)
 306 {
 307         const struct posix_acl_entry *pa, *pe;
 308         umode_t mode = 0;
 309         int not_equiv = 0;
 310
 311         /*
 312          * A null ACL can always be presented as mode bits.
 313          */
 314         if (!acl)
 315                 return 0;
 316
 317         FOREACH_ACL_ENTRY(pa, acl, pe) {
 318                 switch (pa->e_tag) {
 319                         case ACL_USER_OBJ:
 320                                 mode |= (pa->e_perm & S_IRWXO) << 6;
 321                                 break;
 322                         case ACL_GROUP_OBJ:
 323                                 mode |= (pa->e_perm & S_IRWXO) << 3;
 324                                 break;
 325                         case ACL_OTHER:
 326                                 mode |= pa->e_perm & S_IRWXO;
 327                                 break;
 328                         case ACL_MASK:
 329                                 mode = (mode & ~S_IRWXG) |
 330                                        ((pa->e_perm & S_IRWXO) << 3);
 331                                 not_equiv = 1;
 332                                 break;
 333                         case ACL_USER:
 334                         case ACL_GROUP:
 335                                 not_equiv = 1;
 336                                 break;
 337                         default:
 338                                 return -EINVAL;
 339                 }
 340         }
 341         if (mode_p)
 342                 *mode_p = (*mode_p & ~S_IRWXUGO) | mode;
 343         return not_equiv;
 344 }
 345 EXPORT_SYMBOL(posix_acl_equiv_mode);
 346
 347 /*
 348  * Create an ACL representing the file mode permission bits of an inode.
 349  */
 350 struct posix_acl *
 351 posix_acl_from_mode(umode_t mode, gfp_t flags)
 352 {
 353         struct posix_acl *acl = posix_acl_alloc(3, flags);
 354         if (!acl)
 355                 return ERR_PTR(-ENOMEM);
 356
 357         acl->a_entries[0].e_tag  = ACL_USER_OBJ;
 358         acl->a_entries[0].e_perm = (mode & S_IRWXU) >> 6;
 359
 360         acl->a_entries[1].e_tag  = ACL_GROUP_OBJ;
 361         acl->a_entries[1].e_perm = (mode & S_IRWXG) >> 3;
 362
 363         acl->a_entries[2].e_tag  = ACL_OTHER;
 364         acl->a_entries[2].e_perm = (mode & S_IRWXO);
 365         return acl;
 366 }
 367 EXPORT_SYMBOL(posix_acl_from_mode);
 368
 369 /*
 370  * Return 0 if current is granted want access to the inode
 371  * by the acl. Returns -E... otherwise.
 372  */
 373 int
 374 posix_acl_permission(struct mnt_idmap *idmap, struct inode *inode,
 375                      const struct posix_acl *acl, int want)
 376 {
 377         const struct posix_acl_entry *pa, *pe, *mask_obj;
 378         struct user_namespace *fs_userns = i_user_ns(inode);
 379         int found = 0;
 380         vfsuid_t vfsuid;
 381         vfsgid_t vfsgid;
 382
 383         want &= MAY_READ | MAY_WRITE | MAY_EXEC;
 384
 385         FOREACH_ACL_ENTRY(pa, acl, pe) {
 386                 switch(pa->e_tag) {
 387                         case ACL_USER_OBJ:
 388                                 /* (May have been checked already) */
 389                                 vfsuid = i_uid_into_vfsuid(idmap, inode);
 390                                 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
 391                                         goto check_perm;
 392                                 break;
 393                         case ACL_USER:
 394                                 vfsuid = make_vfsuid(idmap, fs_userns,
 395                                                      pa->e_uid);
 396                                 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
 397                                         goto mask;
 398                                 break;
 399                         case ACL_GROUP_OBJ:
 400                                 vfsgid = i_gid_into_vfsgid(idmap, inode);
 401                                 if (vfsgid_in_group_p(vfsgid)) {
 402                                         found = 1;
 403                                         if ((pa->e_perm & want) == want)
 404                                                 goto mask;
 405                                 }
 406                                 break;
 407                         case ACL_GROUP:
 408                                 vfsgid = make_vfsgid(idmap, fs_userns,
 409                                                      pa->e_gid);
 410                                 if (vfsgid_in_group_p(vfsgid)) {
 411                                         found = 1;
 412                                         if ((pa->e_perm & want) == want)
 413                                                 goto mask;
 414                                 }
 415                                 break;
 416                         case ACL_MASK:
 417                                 break;
 418                         case ACL_OTHER:
 419                                 if (found)
 420                                         return -EACCES;
 421                                 else
 422                                         goto check_perm;
 423                         default:
 424                                 return -EIO;
 425                 }
 426         }
 427         return -EIO;
 428
 429 mask:
 430         for (mask_obj = pa+1; mask_obj != pe; mask_obj++) {
 431                 if (mask_obj->e_tag == ACL_MASK) {
 432                         if ((pa->e_perm & mask_obj->e_perm & want) == want)
 433                                 return 0;
 434                         return -EACCES;
 435                 }
 436         }
 437
 438 check_perm:
 439         if ((pa->e_perm & want) == want)
 440                 return 0;
 441         return -EACCES;
 442 }
 443
 444 /*
 445  * Modify acl when creating a new inode. The caller must ensure the acl is
 446  * only referenced once.
 447  *
 448  * mode_p initially must contain the mode parameter to the open() / creat()
 449  * system calls. All permissions that are not granted by the acl are removed.
 450  * The permissions in the acl are changed to reflect the mode_p parameter.
 451  */
 452 static int posix_acl_create_masq(struct posix_acl *acl, umode_t *mode_p)
 453 {
 454         struct posix_acl_entry *pa, *pe;
 455         struct posix_acl_entry *group_obj = NULL, *mask_obj = NULL;
 456         umode_t mode = *mode_p;
 457         int not_equiv = 0;
 458
 459         /* assert(atomic_read(acl->a_refcount) == 1); */
 460
 461         FOREACH_ACL_ENTRY(pa, acl, pe) {
 462                 switch(pa->e_tag) {
 463                         case ACL_USER_OBJ:
 464                                 pa->e_perm &= (mode >> 6) | ~S_IRWXO;
 465                                 mode &= (pa->e_perm << 6) | ~S_IRWXU;
 466                                 break;
 467
 468                         case ACL_USER:
 469                         case ACL_GROUP:
 470                                 not_equiv = 1;
 471                                 break;
 472
 473                         case ACL_GROUP_OBJ:
 474                                 group_obj = pa;
 475                                 break;
 476
 477                         case ACL_OTHER:
 478                                 pa->e_perm &= mode | ~S_IRWXO;
 479                                 mode &= pa->e_perm | ~S_IRWXO;
 480                                 break;
 481
 482                         case ACL_MASK:
 483                                 mask_obj = pa;
 484                                 not_equiv = 1;
 485                                 break;
 486
 487                         default:
 488                                 return -EIO;
 489                 }
 490         }
 491
 492         if (mask_obj) {
 493                 mask_obj->e_perm &= (mode >> 3) | ~S_IRWXO;
 494                 mode &= (mask_obj->e_perm << 3) | ~S_IRWXG;
 495         } else {
 496                 if (!group_obj)
 497                         return -EIO;
 498                 group_obj->e_perm &= (mode >> 3) | ~S_IRWXO;
 499                 mode &= (group_obj->e_perm << 3) | ~S_IRWXG;
 500         }
 501
 502         *mode_p = (*mode_p & ~S_IRWXUGO) | mode;
 503         return not_equiv;
 504 }
 505
 506 /*
 507  * Modify the ACL for the chmod syscall.
 508  */
 509 static int __posix_acl_chmod_masq(struct posix_acl *acl, umode_t mode)
 510 {
 511         struct posix_acl_entry *group_obj = NULL, *mask_obj = NULL;
 512         struct posix_acl_entry *pa, *pe;
 513
 514         /* assert(atomic_read(acl->a_refcount) == 1); */
 515
 516         FOREACH_ACL_ENTRY(pa, acl, pe) {
 517                 switch(pa->e_tag) {
 518                         case ACL_USER_OBJ:
 519                                 pa->e_perm = (mode & S_IRWXU) >> 6;
 520                                 break;
 521
 522                         case ACL_USER:
 523                         case ACL_GROUP:
 524                                 break;
 525
 526                         case ACL_GROUP_OBJ:
 527                                 group_obj = pa;
 528                                 break;
 529
 530                         case ACL_MASK:
 531                                 mask_obj = pa;
 532                                 break;
 533
 534                         case ACL_OTHER:
 535                                 pa->e_perm = (mode & S_IRWXO);
 536                                 break;
 537
 538                         default:
 539                                 return -EIO;
 540                 }
 541         }
 542
 543         if (mask_obj) {
 544                 mask_obj->e_perm = (mode & S_IRWXG) >> 3;
 545         } else {
 546                 if (!group_obj)
 547                         return -EIO;
 548                 group_obj->e_perm = (mode & S_IRWXG) >> 3;
 549         }
 550
 551         return 0;
 552 }
 553
 554 int
 555 __posix_acl_create(struct posix_acl **acl, gfp_t gfp, umode_t *mode_p)
 556 {
 557         struct posix_acl *clone = posix_acl_clone(*acl, gfp);
 558         int err = -ENOMEM;
 559         if (clone) {
 560                 err = posix_acl_create_masq(clone, mode_p);
 561                 if (err < 0) {
 562                         posix_acl_release(clone);
 563                         clone = NULL;
 564                 }
 565         }
 566         posix_acl_release(*acl);
 567         *acl = clone;
 568         return err;
 569 }
 570 EXPORT_SYMBOL(__posix_acl_create);
 571
 572 int
 573 __posix_acl_chmod(struct posix_acl **acl, gfp_t gfp, umode_t mode)
 574 {
 575         struct posix_acl *clone = posix_acl_clone(*acl, gfp);
 576         int err = -ENOMEM;
 577         if (clone) {
 578                 err = __posix_acl_chmod_masq(clone, mode);
 579                 if (err) {
 580                         posix_acl_release(clone);
 581                         clone = NULL;
 582                 }
 583         }
 584         posix_acl_release(*acl);
 585         *acl = clone;
 586         return err;
 587 }
 588 EXPORT_SYMBOL(__posix_acl_chmod);
 589
 590 /**
 591  * posix_acl_chmod - chmod a posix acl
 592  *
 593  * @idmap:      idmap of the mount @inode was found from
 594  * @dentry:     dentry to check permissions on
 595  * @mode:       the new mode of @inode
 596  *
 597  * If the dentry has been found through an idmapped mount the idmap of
 598  * the vfsmount must be passed through @idmap. This function will then
 599  * take care to map the inode according to @idmap before checking
 600  * permissions. On non-idmapped mounts or if permission checking is to be
 601  * performed on the raw inode simply pass @nop_mnt_idmap.
 602  */
 603 int
 604  posix_acl_chmod(struct mnt_idmap *idmap, struct dentry *dentry,
 605                     umode_t mode)
 606 {
 607         struct inode *inode = d_inode(dentry);
 608         struct posix_acl *acl;
 609         int ret = 0;
 610
 611         if (!IS_POSIXACL(inode))
 612                 return 0;
 613         if (!inode->i_op->set_acl)
 614                 return -EOPNOTSUPP;
 615
 616         acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
 617         if (IS_ERR_OR_NULL(acl)) {
 618                 if (acl == ERR_PTR(-EOPNOTSUPP))
 619                         return 0;
 620                 return PTR_ERR(acl);
 621         }
 622
 623         ret = __posix_acl_chmod(&acl, GFP_KERNEL, mode);
 624         if (ret)
 625                 return ret;
 626         ret = inode->i_op->set_acl(idmap, dentry, acl, ACL_TYPE_ACCESS);
 627         posix_acl_release(acl);
 628         return ret;
 629 }
 630 EXPORT_SYMBOL(posix_acl_chmod);
 631
 632 int
 633 posix_acl_create(struct inode *dir, umode_t *mode,
 634                 struct posix_acl **default_acl, struct posix_acl **acl)
 635 {
 636         struct posix_acl *p;
 637         struct posix_acl *clone;
 638         int ret;
 639
 640         *acl = NULL;
 641         *default_acl = NULL;
 642
 643         if (S_ISLNK(*mode) || !IS_POSIXACL(dir))
 644                 return 0;
 645
 646         p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
 647         if (!p || p == ERR_PTR(-EOPNOTSUPP)) {
 648                 *mode &= ~current_umask();
 649                 return 0;
 650         }
 651         if (IS_ERR(p))
 652                 return PTR_ERR(p);
 653
 654         ret = -ENOMEM;
 655         clone = posix_acl_clone(p, GFP_NOFS);
 656         if (!clone)
 657                 goto err_release;
 658
 659         ret = posix_acl_create_masq(clone, mode);
 660         if (ret < 0)
 661                 goto err_release_clone;
 662
 663         if (ret == 0)
 664                 posix_acl_release(clone);
 665         else
 666                 *acl = clone;
 667
 668         if (!S_ISDIR(*mode))
 669                 posix_acl_release(p);
 670         else
 671                 *default_acl = p;
 672
 673         return 0;
 674
 675 err_release_clone:
 676         posix_acl_release(clone);
 677 err_release:
 678         posix_acl_release(p);
 679         return ret;
 680 }
 681 EXPORT_SYMBOL_GPL(posix_acl_create);
 682
 683 /**
 684  * posix_acl_update_mode  -  update mode in set_acl
 685  * @idmap:      idmap of the mount @inode was found from
 686  * @inode:      target inode
 687  * @mode_p:     mode (pointer) for update
 688  * @acl:        acl pointer
 689  *
 690  * Update the file mode when setting an ACL: compute the new file permission
 691  * bits based on the ACL.  In addition, if the ACL is equivalent to the new
 692  * file mode, set *@acl to NULL to indicate that no ACL should be set.
 693  *
 694  * As with chmod, clear the setgid bit if the caller is not in the owning group
 695  * or capable of CAP_FSETID (see inode_change_ok).
 696  *
 697  * If the inode has been found through an idmapped mount the idmap of
 698  * the vfsmount must be passed through @idmap. This function will then
 699  * take care to map the inode according to @idmap before checking
 700  * permissions. On non-idmapped mounts or if permission checking is to be
 701  * performed on the raw inode simply pass @nop_mnt_idmap.
 702  *
 703  * Called from set_acl inode operations.
 704  */
 705 int posix_acl_update_mode(struct mnt_idmap *idmap,
 706                           struct inode *inode, umode_t *mode_p,
 707                           struct posix_acl **acl)
 708 {
 709         umode_t mode = inode->i_mode;
 710         int error;
 711
 712         error = posix_acl_equiv_mode(*acl, &mode);
 713         if (error < 0)
 714                 return error;
 715         if (error == 0)
 716                 *acl = NULL;
 717         if (!in_group_or_capable(idmap, inode,
 718                                  i_gid_into_vfsgid(idmap, inode)))
 719                 mode &= ~S_ISGID;
 720         *mode_p = mode;
 721         return 0;
 722 }
 723 EXPORT_SYMBOL(posix_acl_update_mode);
 724
 725 /*
 726  * Fix up the uids and gids in posix acl extended attributes in place.
 727  */
 728 static int posix_acl_fix_xattr_common(const void *value, size_t size)
 729 {
 730         const struct posix_acl_xattr_header *header = value;
 731         int count;
 732
 733         if (!header)
 734                 return -EINVAL;
 735         if (size < sizeof(struct posix_acl_xattr_header))
 736                 return -EINVAL;
 737         if (header->a_version != cpu_to_le32(POSIX_ACL_XATTR_VERSION))
 738                 return -EOPNOTSUPP;
 739
 740         count = posix_acl_xattr_count(size);
 741         if (count < 0)
 742                 return -EINVAL;
 743         if (count == 0)
 744                 return 0;
 745
 746         return count;
 747 }
 748
 749 /**
 750  * posix_acl_from_xattr - convert POSIX ACLs from backing store to VFS format
 751  * @userns: the filesystem's idmapping
 752  * @value: the uapi representation of POSIX ACLs
 753  * @size: the size of @void
 754  *
 755  * Filesystems that store POSIX ACLs in the unaltered uapi format should use
 756  * posix_acl_from_xattr() when reading them from the backing store and
 757  * converting them into the struct posix_acl VFS format. The helper is
 758  * specifically intended to be called from the acl inode operation.
 759  *
 760  * The posix_acl_from_xattr() function will map the raw {g,u}id values stored
 761  * in ACL_{GROUP,USER} entries into idmapping in @userns.
 762  *
 763  * Note that posix_acl_from_xattr() does not take idmapped mounts into account.
 764  * If it did it calling it from the get acl inode operation would return POSIX
 765  * ACLs mapped according to an idmapped mount which would mean that the value
 766  * couldn't be cached for the filesystem. Idmapped mounts are taken into
 767  * account on the fly during permission checking or right at the VFS -
 768  * userspace boundary before reporting them to the user.
 769  *
 770  * Return: Allocated struct posix_acl on success, NULL for a valid header but
 771  *         without actual POSIX ACL entries, or ERR_PTR() encoded error code.
 772  */
 773 struct posix_acl *posix_acl_from_xattr(struct user_namespace *userns,
 774                                        const void *value, size_t size)
 775 {
 776         const struct posix_acl_xattr_header *header = value;
 777         const struct posix_acl_xattr_entry *entry = (const void *)(header + 1), *end;
 778         int count;
 779         struct posix_acl *acl;
 780         struct posix_acl_entry *acl_e;
 781
 782         count = posix_acl_fix_xattr_common(value, size);
 783         if (count < 0)
 784                 return ERR_PTR(count);
 785         if (count == 0)
 786                 return NULL;
 787
 788         acl = posix_acl_alloc(count, GFP_NOFS);
 789         if (!acl)
 790                 return ERR_PTR(-ENOMEM);
 791         acl_e = acl->a_entries;
 792
 793         for (end = entry + count; entry != end; acl_e++, entry++) {
 794                 acl_e->e_tag  = le16_to_cpu(entry->e_tag);
 795                 acl_e->e_perm = le16_to_cpu(entry->e_perm);
 796
 797                 switch(acl_e->e_tag) {
 798                         case ACL_USER_OBJ:
 799                         case ACL_GROUP_OBJ:
 800                         case ACL_MASK:
 801                         case ACL_OTHER:
 802                                 break;
 803
 804                         case ACL_USER:
 805                                 acl_e->e_uid = make_kuid(userns,
 806                                                 le32_to_cpu(entry->e_id));
 807                                 if (!uid_valid(acl_e->e_uid))
 808                                         goto fail;
 809                                 break;
 810                         case ACL_GROUP:
 811                                 acl_e->e_gid = make_kgid(userns,
 812                                                 le32_to_cpu(entry->e_id));
 813                                 if (!gid_valid(acl_e->e_gid))
 814                                         goto fail;
 815                                 break;
 816
 817                         default:
 818                                 goto fail;
 819                 }
 820         }
 821         return acl;
 822
 823 fail:
 824         posix_acl_release(acl);
 825         return ERR_PTR(-EINVAL);
 826 }
 827 EXPORT_SYMBOL (posix_acl_from_xattr);
 828
 829 /*
 830  * Convert from in-memory to extended attribute representation.
 831  */
 832 int
 833 posix_acl_to_xattr(struct user_namespace *user_ns, const struct posix_acl *acl,
 834                    void *buffer, size_t size)
 835 {
 836         struct posix_acl_xattr_header *ext_acl = buffer;
 837         struct posix_acl_xattr_entry *ext_entry;
 838         int real_size, n;
 839
 840         real_size = posix_acl_xattr_size(acl->a_count);
 841         if (!buffer)
 842                 return real_size;
 843         if (real_size > size)
 844                 return -ERANGE;
 845
 846         ext_entry = (void *)(ext_acl + 1);
 847         ext_acl->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
 848
 849         for (n=0; n < acl->a_count; n++, ext_entry++) {
 850                 const struct posix_acl_entry *acl_e = &acl->a_entries[n];
 851                 ext_entry->e_tag  = cpu_to_le16(acl_e->e_tag);
 852                 ext_entry->e_perm = cpu_to_le16(acl_e->e_perm);
 853                 switch(acl_e->e_tag) {
 854                 case ACL_USER:
 855                         ext_entry->e_id =
 856                                 cpu_to_le32(from_kuid(user_ns, acl_e->e_uid));
 857                         break;
 858                 case ACL_GROUP:
 859                         ext_entry->e_id =
 860                                 cpu_to_le32(from_kgid(user_ns, acl_e->e_gid));
 861                         break;
 862                 default:
 863                         ext_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
 864                         break;
 865                 }
 866         }
 867         return real_size;
 868 }
 869 EXPORT_SYMBOL (posix_acl_to_xattr);
 870
 871 /**
 872  * vfs_posix_acl_to_xattr - convert from kernel to userspace representation
 873  * @idmap: idmap of the mount
 874  * @inode: inode the posix acls are set on
 875  * @acl: the posix acls as represented by the vfs
 876  * @buffer: the buffer into which to convert @acl
 877  * @size: size of @buffer
 878  *
 879  * This converts @acl from the VFS representation in the filesystem idmapping
 880  * to the uapi form reportable to userspace. And mount and caller idmappings
 881  * are handled appropriately.
 882  *
 883  * Return: On success, the size of the stored uapi posix acls, on error a
 884  * negative errno.
 885  */
 886 static ssize_t vfs_posix_acl_to_xattr(struct mnt_idmap *idmap,
 887                                       struct inode *inode,
 888                                       const struct posix_acl *acl, void *buffer,
 889                                       size_t size)
 890
 891 {
 892         struct posix_acl_xattr_header *ext_acl = buffer;
 893         struct posix_acl_xattr_entry *ext_entry;
 894         struct user_namespace *fs_userns, *caller_userns;
 895         ssize_t real_size, n;
 896         vfsuid_t vfsuid;
 897         vfsgid_t vfsgid;
 898
 899         real_size = posix_acl_xattr_size(acl->a_count);
 900         if (!buffer)
 901                 return real_size;
 902         if (real_size > size)
 903                 return -ERANGE;
 904
 905         ext_entry = (void *)(ext_acl + 1);
 906         ext_acl->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
 907
 908         fs_userns = i_user_ns(inode);
 909         caller_userns = current_user_ns();
 910         for (n=0; n < acl->a_count; n++, ext_entry++) {
 911                 const struct posix_acl_entry *acl_e = &acl->a_entries[n];
 912                 ext_entry->e_tag  = cpu_to_le16(acl_e->e_tag);
 913                 ext_entry->e_perm = cpu_to_le16(acl_e->e_perm);
 914                 switch(acl_e->e_tag) {
 915                 case ACL_USER:
 916                         vfsuid = make_vfsuid(idmap, fs_userns, acl_e->e_uid);
 917                         ext_entry->e_id = cpu_to_le32(from_kuid(
 918                                 caller_userns, vfsuid_into_kuid(vfsuid)));
 919                         break;
 920                 case ACL_GROUP:
 921                         vfsgid = make_vfsgid(idmap, fs_userns, acl_e->e_gid);
 922                         ext_entry->e_id = cpu_to_le32(from_kgid(
 923                                 caller_userns, vfsgid_into_kgid(vfsgid)));
 924                         break;
 925                 default:
 926                         ext_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
 927                         break;
 928                 }
 929         }
 930         return real_size;
 931 }
 932
 933 int
 934 set_posix_acl(struct mnt_idmap *idmap, struct dentry *dentry,
 935               int type, struct posix_acl *acl)
 936 {
 937         struct inode *inode = d_inode(dentry);
 938
 939         if (!IS_POSIXACL(inode))
 940                 return -EOPNOTSUPP;
 941         if (!inode->i_op->set_acl)
 942                 return -EOPNOTSUPP;
 943
 944         if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
 945                 return acl ? -EACCES : 0;
 946         if (!inode_owner_or_capable(idmap, inode))
 947                 return -EPERM;
 948
 949         if (acl) {
 950                 int ret = posix_acl_valid(inode->i_sb->s_user_ns, acl);
 951                 if (ret)
 952                         return ret;
 953         }
 954         return inode->i_op->set_acl(idmap, dentry, acl, type);
 955 }
 956 EXPORT_SYMBOL(set_posix_acl);
 957
 958 int posix_acl_listxattr(struct inode *inode, char **buffer,
 959                         ssize_t *remaining_size)
 960 {
 961         int err;
 962
 963         if (!IS_POSIXACL(inode))
 964                 return 0;
 965
 966         if (inode->i_acl) {
 967                 err = xattr_list_one(buffer, remaining_size,
 968                                      XATTR_NAME_POSIX_ACL_ACCESS);
 969                 if (err)
 970                         return err;
 971         }
 972
 973         if (inode->i_default_acl) {
 974                 err = xattr_list_one(buffer, remaining_size,
 975                                      XATTR_NAME_POSIX_ACL_DEFAULT);
 976                 if (err)
 977                         return err;
 978         }
 979
 980         return 0;
 981 }
 982
 983 static bool
 984 posix_acl_xattr_list(struct dentry *dentry)
 985 {
 986         return IS_POSIXACL(d_backing_inode(dentry));
 987 }
 988
 989 /*
 990  * nop_posix_acl_access - legacy xattr handler for access POSIX ACLs
 991  *
 992  * This is the legacy POSIX ACL access xattr handler. It is used by some
 993  * filesystems to implement their ->listxattr() inode operation. New code
 994  * should never use them.
 995  */
 996 const struct xattr_handler nop_posix_acl_access = {
 997         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 998         .list = posix_acl_xattr_list,
 999 };
1000 EXPORT_SYMBOL_GPL(nop_posix_acl_access);
1001
1002 /*
1003  * nop_posix_acl_default - legacy xattr handler for default POSIX ACLs
1004  *
1005  * This is the legacy POSIX ACL default xattr handler. It is used by some
1006  * filesystems to implement their ->listxattr() inode operation. New code
1007  * should never use them.
1008  */
1009 const struct xattr_handler nop_posix_acl_default = {
1010         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1011         .list = posix_acl_xattr_list,
1012 };
1013 EXPORT_SYMBOL_GPL(nop_posix_acl_default);
1014
1015 int simple_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1016                    struct posix_acl *acl, int type)
1017 {
1018         int error;
1019         struct inode *inode = d_inode(dentry);
1020
1021         if (type == ACL_TYPE_ACCESS) {
1022                 error = posix_acl_update_mode(idmap, inode,
1023                                 &inode->i_mode, &acl);
1024                 if (error)
1025                         return error;
1026         }
1027
1028         inode_set_ctime_current(inode);
1029         if (IS_I_VERSION(inode))
1030                 inode_inc_iversion(inode);
1031         set_cached_acl(inode, type, acl);
1032         return 0;
1033 }
1034
1035 int simple_acl_create(struct inode *dir, struct inode *inode)
1036 {
1037         struct posix_acl *default_acl, *acl;
1038         int error;
1039
1040         error = posix_acl_create(dir, &inode->i_mode, &default_acl, &acl);
1041         if (error)
1042                 return error;
1043
1044         set_cached_acl(inode, ACL_TYPE_DEFAULT, default_acl);
1045         set_cached_acl(inode, ACL_TYPE_ACCESS, acl);
1046
1047         if (default_acl)
1048                 posix_acl_release(default_acl);
1049         if (acl)
1050                 posix_acl_release(acl);
1051         return 0;
1052 }
1053
1054 static int vfs_set_acl_idmapped_mnt(struct mnt_idmap *idmap,
1055                                     struct user_namespace *fs_userns,
1056                                     struct posix_acl *acl)
1057 {
1058         for (int n = 0; n < acl->a_count; n++) {
1059                 struct posix_acl_entry *acl_e = &acl->a_entries[n];
1060
1061                 switch (acl_e->e_tag) {
1062                 case ACL_USER:
1063                         acl_e->e_uid = from_vfsuid(idmap, fs_userns,
1064                                                    VFSUIDT_INIT(acl_e->e_uid));
1065                         break;
1066                 case ACL_GROUP:
1067                         acl_e->e_gid = from_vfsgid(idmap, fs_userns,
1068                                                    VFSGIDT_INIT(acl_e->e_gid));
1069                         break;
1070                 }
1071         }
1072
1073         return 0;
1074 }
1075
1076 /**
1077  * vfs_set_acl - set posix acls
1078  * @idmap: idmap of the mount
1079  * @dentry: the dentry based on which to set the posix acls
1080  * @acl_name: the name of the posix acl
1081  * @kacl: the posix acls in the appropriate VFS format
1082  *
1083  * This function sets @kacl. The caller must all posix_acl_release() on @kacl
1084  * afterwards.
1085  *
1086  * Return: On success 0, on error negative errno.
1087  */
1088 int vfs_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1089                 const char *acl_name, struct posix_acl *kacl)
1090 {
1091         int acl_type;
1092         int error;
1093         struct inode *inode = d_inode(dentry);
1094         struct inode *delegated_inode = NULL;
1095
1096         acl_type = posix_acl_type(acl_name);
1097         if (acl_type < 0)
1098                 return -EINVAL;
1099
1100         if (kacl) {
1101                 /*
1102                  * If we're on an idmapped mount translate from mount specific
1103                  * vfs{g,u}id_t into global filesystem k{g,u}id_t.
1104                  * Afterwards we can cache the POSIX ACLs filesystem wide and -
1105                  * if this is a filesystem with a backing store - ultimately
1106                  * translate them to backing store values.
1107                  */
1108                 error = vfs_set_acl_idmapped_mnt(idmap, i_user_ns(inode), kacl);
1109                 if (error)
1110                         return error;
1111         }
1112
1113 retry_deleg:
1114         inode_lock(inode);
1115
1116         /*
1117          * We only care about restrictions the inode struct itself places upon
1118          * us otherwise POSIX ACLs aren't subject to any VFS restrictions.
1119          */
1120         error = may_write_xattr(idmap, inode);
1121         if (error)
1122                 goto out_inode_unlock;
1123
1124         error = security_inode_set_acl(idmap, dentry, acl_name, kacl);
1125         if (error)
1126                 goto out_inode_unlock;
1127
1128         error = try_break_deleg(inode, &delegated_inode);
1129         if (error)
1130                 goto out_inode_unlock;
1131
1132         if (likely(!is_bad_inode(inode)))
1133                 error = set_posix_acl(idmap, dentry, acl_type, kacl);
1134         else
1135                 error = -EIO;
1136         if (!error) {
1137                 fsnotify_xattr(dentry);
1138                 security_inode_post_set_acl(dentry, acl_name, kacl);
1139         }
1140
1141 out_inode_unlock:
1142         inode_unlock(inode);
1143
1144         if (delegated_inode) {
1145                 error = break_deleg_wait(&delegated_inode);
1146                 if (!error)
1147                         goto retry_deleg;
1148         }
1149
1150         return error;
1151 }
1152 EXPORT_SYMBOL_GPL(vfs_set_acl);
1153
1154 /**
1155  * vfs_get_acl - get posix acls
1156  * @idmap: idmap of the mount
1157  * @dentry: the dentry based on which to retrieve the posix acls
1158  * @acl_name: the name of the posix acl
1159  *
1160  * This function retrieves @kacl from the filesystem. The caller must all
1161  * posix_acl_release() on @kacl.
1162  *
1163  * Return: On success POSIX ACLs in VFS format, on error negative errno.
1164  */
1165 struct posix_acl *vfs_get_acl(struct mnt_idmap *idmap,
1166                               struct dentry *dentry, const char *acl_name)
1167 {
1168         struct inode *inode = d_inode(dentry);
1169         struct posix_acl *acl;
1170         int acl_type, error;
1171
1172         acl_type = posix_acl_type(acl_name);
1173         if (acl_type < 0)
1174                 return ERR_PTR(-EINVAL);
1175
1176         /*
1177          * The VFS has no restrictions on reading POSIX ACLs so calling
1178          * something like xattr_permission() isn't needed. Only LSMs get a say.
1179          */
1180         error = security_inode_get_acl(idmap, dentry, acl_name);
1181         if (error)
1182                 return ERR_PTR(error);
1183
1184         if (!IS_POSIXACL(inode))
1185                 return ERR_PTR(-EOPNOTSUPP);
1186         if (S_ISLNK(inode->i_mode))
1187                 return ERR_PTR(-EOPNOTSUPP);
1188
1189         acl = __get_acl(idmap, dentry, inode, acl_type);
1190         if (IS_ERR(acl))
1191                 return acl;
1192         if (!acl)
1193                 return ERR_PTR(-ENODATA);
1194
1195         return acl;
1196 }
1197 EXPORT_SYMBOL_GPL(vfs_get_acl);
1198
1199 /**
1200  * vfs_remove_acl - remove posix acls
1201  * @idmap: idmap of the mount
1202  * @dentry: the dentry based on which to retrieve the posix acls
1203  * @acl_name: the name of the posix acl
1204  *
1205  * This function removes posix acls.
1206  *
1207  * Return: On success 0, on error negative errno.
1208  */
1209 int vfs_remove_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1210                    const char *acl_name)
1211 {
1212         int acl_type;
1213         int error;
1214         struct inode *inode = d_inode(dentry);
1215         struct inode *delegated_inode = NULL;
1216
1217         acl_type = posix_acl_type(acl_name);
1218         if (acl_type < 0)
1219                 return -EINVAL;
1220
1221 retry_deleg:
1222         inode_lock(inode);
1223
1224         /*
1225          * We only care about restrictions the inode struct itself places upon
1226          * us otherwise POSIX ACLs aren't subject to any VFS restrictions.
1227          */
1228         error = may_write_xattr(idmap, inode);
1229         if (error)
1230                 goto out_inode_unlock;
1231
1232         error = security_inode_remove_acl(idmap, dentry, acl_name);
1233         if (error)
1234                 goto out_inode_unlock;
1235
1236         error = try_break_deleg(inode, &delegated_inode);
1237         if (error)
1238                 goto out_inode_unlock;
1239
1240         if (likely(!is_bad_inode(inode)))
1241                 error = set_posix_acl(idmap, dentry, acl_type, NULL);
1242         else
1243                 error = -EIO;
1244         if (!error) {
1245                 fsnotify_xattr(dentry);
1246                 security_inode_post_remove_acl(idmap, dentry, acl_name);
1247         }
1248
1249 out_inode_unlock:
1250         inode_unlock(inode);
1251
1252         if (delegated_inode) {
1253                 error = break_deleg_wait(&delegated_inode);
1254                 if (!error)
1255                         goto retry_deleg;
1256         }
1257
1258         return error;
1259 }
1260 EXPORT_SYMBOL_GPL(vfs_remove_acl);
1261
1262 int do_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1263                const char *acl_name, const void *kvalue, size_t size)
1264 {
1265         int error;
1266         struct posix_acl *acl = NULL;
1267
1268         if (size) {
1269                 /*
1270                  * Note that posix_acl_from_xattr() uses GFP_NOFS when it
1271                  * probably doesn't need to here.
1272                  */
1273                 acl = posix_acl_from_xattr(current_user_ns(), kvalue, size);
1274                 if (IS_ERR(acl))
1275                         return PTR_ERR(acl);
1276         }
1277
1278         error = vfs_set_acl(idmap, dentry, acl_name, acl);
1279         posix_acl_release(acl);
1280         return error;
1281 }
1282
1283 ssize_t do_get_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1284                    const char *acl_name, void *kvalue, size_t size)
1285 {
1286         ssize_t error;
1287         struct posix_acl *acl;
1288
1289         acl = vfs_get_acl(idmap, dentry, acl_name);
1290         if (IS_ERR(acl))
1291                 return PTR_ERR(acl);
1292
1293         error = vfs_posix_acl_to_xattr(idmap, d_inode(dentry),
1294                                        acl, kvalue, size);
1295         posix_acl_release(acl);
1296         return error;
1297 }