2 * Copyright (c) 2001-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_alloc_btree.h"
27 #include "xfs_ialloc_btree.h"
28 #include "xfs_dir2_sf.h"
29 #include "xfs_attr_sf.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
37 #include <linux/capability.h>
38 #include <linux/posix_acl_xattr.h>
40 STATIC int xfs_acl_setmode(bhv_vnode_t *, xfs_acl_t *, int *);
41 STATIC void xfs_acl_filter_mode(mode_t, xfs_acl_t *);
42 STATIC void xfs_acl_get_endian(xfs_acl_t *);
43 STATIC int xfs_acl_access(uid_t, gid_t, xfs_acl_t *, mode_t, cred_t *);
44 STATIC int xfs_acl_invalid(xfs_acl_t *);
45 STATIC void xfs_acl_sync_mode(mode_t, xfs_acl_t *);
46 STATIC void xfs_acl_get_attr(bhv_vnode_t *, xfs_acl_t *, int, int, int *);
47 STATIC void xfs_acl_set_attr(bhv_vnode_t *, xfs_acl_t *, int, int *);
48 STATIC int xfs_acl_allow_set(bhv_vnode_t *, int);
50 kmem_zone_t *xfs_acl_zone;
54 * Test for existence of access ACL attribute as efficiently as possible.
57 xfs_acl_vhasacl_access(
62 xfs_acl_get_attr(vp, NULL, _ACL_TYPE_ACCESS, ATTR_KERNOVAL, &error);
67 * Test for existence of default ACL attribute as efficiently as possible.
70 xfs_acl_vhasacl_default(
77 xfs_acl_get_attr(vp, NULL, _ACL_TYPE_DEFAULT, ATTR_KERNOVAL, &error);
82 * Convert from extended attribute representation to in-memory for XFS.
85 posix_acl_xattr_to_xfs(
86 posix_acl_xattr_header *src,
90 posix_acl_xattr_entry *src_entry;
91 xfs_acl_entry_t *dest_entry;
97 if (size < sizeof(posix_acl_xattr_header))
100 if (src->a_version != cpu_to_le32(POSIX_ACL_XATTR_VERSION))
103 memset(dest, 0, sizeof(xfs_acl_t));
104 dest->acl_cnt = posix_acl_xattr_count(size);
105 if (dest->acl_cnt < 0 || dest->acl_cnt > XFS_ACL_MAX_ENTRIES)
109 * acl_set_file(3) may request that we set default ACLs with
110 * zero length -- defend (gracefully) against that here.
115 src_entry = (posix_acl_xattr_entry *)((char *)src + sizeof(*src));
116 dest_entry = &dest->acl_entry[0];
118 for (n = 0; n < dest->acl_cnt; n++, src_entry++, dest_entry++) {
119 dest_entry->ae_perm = le16_to_cpu(src_entry->e_perm);
120 if (_ACL_PERM_INVALID(dest_entry->ae_perm))
122 dest_entry->ae_tag = le16_to_cpu(src_entry->e_tag);
123 switch(dest_entry->ae_tag) {
126 dest_entry->ae_id = le32_to_cpu(src_entry->e_id);
132 dest_entry->ae_id = ACL_UNDEFINED_ID;
138 if (xfs_acl_invalid(dest))
145 * Comparison function called from xfs_sort().
146 * Primary key is ae_tag, secondary key is ae_id.
149 xfs_acl_entry_compare(
153 xfs_acl_entry_t *a = (xfs_acl_entry_t *)va,
154 *b = (xfs_acl_entry_t *)vb;
156 if (a->ae_tag == b->ae_tag)
157 return (a->ae_id - b->ae_id);
158 return (a->ae_tag - b->ae_tag);
162 * Convert from in-memory XFS to extended attribute representation.
165 posix_acl_xfs_to_xattr(
167 posix_acl_xattr_header *dest,
171 size_t new_size = posix_acl_xattr_size(src->acl_cnt);
172 posix_acl_xattr_entry *dest_entry;
173 xfs_acl_entry_t *src_entry;
178 /* Need to sort src XFS ACL by <ae_tag,ae_id> */
179 xfs_sort(src->acl_entry, src->acl_cnt, sizeof(src->acl_entry[0]),
180 xfs_acl_entry_compare);
182 dest->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
183 dest_entry = &dest->a_entries[0];
184 src_entry = &src->acl_entry[0];
185 for (n = 0; n < src->acl_cnt; n++, dest_entry++, src_entry++) {
186 dest_entry->e_perm = cpu_to_le16(src_entry->ae_perm);
187 if (_ACL_PERM_INVALID(src_entry->ae_perm))
189 dest_entry->e_tag = cpu_to_le16(src_entry->ae_tag);
190 switch (src_entry->ae_tag) {
193 dest_entry->e_id = cpu_to_le32(src_entry->ae_id);
199 dest_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
216 xfs_acl_t *xfs_acl = NULL;
217 posix_acl_xattr_header *ext_acl = acl;
222 if (!(_ACL_ALLOC(xfs_acl))) {
226 memset(xfs_acl, 0, sizeof(xfs_acl_t));
228 flags = ATTR_KERNOVAL;
230 xfs_acl_get_attr(vp, xfs_acl, kind, flags, &error);
235 error = -posix_acl_xattr_size(XFS_ACL_MAX_ENTRIES);
237 if (xfs_acl_invalid(xfs_acl)) {
241 if (kind == _ACL_TYPE_ACCESS) {
244 va.va_mask = XFS_AT_MODE;
245 error = bhv_vop_getattr(vp, &va, 0, sys_cred);
248 xfs_acl_sync_mode(va.va_mode, xfs_acl);
250 error = -posix_acl_xfs_to_xattr(xfs_acl, ext_acl, size);
267 error = xfs_acl_allow_set(vp, kind);
269 error = bhv_vop_attr_remove(vp, kind == _ACL_TYPE_DEFAULT?
270 SGI_ACL_DEFAULT: SGI_ACL_FILE,
271 ATTR_ROOT, sys_cred);
272 if (error == ENOATTR)
273 error = 0; /* 'scool */
286 posix_acl_xattr_header *ext_acl = acl;
289 int basicperms = 0; /* more than std unix perms? */
294 if (!(_ACL_ALLOC(xfs_acl)))
297 error = posix_acl_xattr_to_xfs(ext_acl, size, xfs_acl);
302 if (!xfs_acl->acl_cnt) {
308 error = xfs_acl_allow_set(vp, kind);
312 /* Incoming ACL exists, set file mode based on its value */
313 if (kind == _ACL_TYPE_ACCESS)
314 xfs_acl_setmode(vp, xfs_acl, &basicperms);
317 * If we have more than std unix permissions, set up the actual attr.
318 * Otherwise, delete any existing attr. This prevents us from
319 * having actual attrs for permissions that can be stored in the
320 * standard permission bits.
323 xfs_acl_set_attr(vp, xfs_acl, kind, &error);
325 xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
343 if (!(_ACL_ALLOC(acl)))
346 /* If the file has no ACL return -1. */
347 rval = sizeof(xfs_acl_t);
348 if (xfs_attr_fetch(ip, SGI_ACL_FILE, SGI_ACL_FILE_SIZE,
349 (char *)acl, &rval, ATTR_ROOT | ATTR_KERNACCESS, cr)) {
353 xfs_acl_get_endian(acl);
355 /* If the file has an empty ACL return -1. */
356 if (acl->acl_cnt == XFS_ACL_NOT_PRESENT) {
361 /* Synchronize ACL with mode bits */
362 xfs_acl_sync_mode(ip->i_d.di_mode, acl);
364 rval = xfs_acl_access(ip->i_d.di_uid, ip->i_d.di_gid, acl, mode, cr);
377 if (vp->v_inode.i_flags & (S_IMMUTABLE|S_APPEND))
379 if (kind == _ACL_TYPE_DEFAULT && !VN_ISDIR(vp))
381 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
383 va.va_mask = XFS_AT_UID;
384 error = bhv_vop_getattr(vp, &va, 0, NULL);
387 if (va.va_uid != current->fsuid && !capable(CAP_FOWNER))
393 * The access control process to determine the access permission:
394 * if uid == file owner id, use the file owner bits.
395 * if gid == file owner group id, use the file group bits.
396 * scan ACL for a matching user or group, and use matched entry
397 * permission. Use total permissions of all matching group entries,
398 * until all acl entries are exhausted. The final permission produced
399 * by matching acl entry or entries needs to be & with group permission.
400 * if not owner, owning group, or matching entry in ACL, use file
404 xfs_acl_capability_check(
408 if ((mode & ACL_READ) && !capable_cred(cr, CAP_DAC_READ_SEARCH))
410 if ((mode & ACL_WRITE) && !capable_cred(cr, CAP_DAC_OVERRIDE))
412 if ((mode & ACL_EXECUTE) && !capable_cred(cr, CAP_DAC_OVERRIDE))
419 * Note: cr is only used here for the capability check if the ACL test fails.
420 * It is not used to find out the credentials uid or groups etc, as was
421 * done in IRIX. It is assumed that the uid and groups for the current
422 * thread are taken from "current" instead of the cr parameter.
432 xfs_acl_entry_t matched;
434 int maskallows = -1; /* true, but not 1, either */
435 int seen_userobj = 0;
437 matched.ae_tag = 0; /* Invalid type */
439 md >>= 6; /* Normalize the bits for comparison */
441 for (i = 0; i < fap->acl_cnt; i++) {
443 * Break out if we've got a user_obj entry or
444 * a user entry and the mask (and have processed USER_OBJ)
446 if (matched.ae_tag == ACL_USER_OBJ)
448 if (matched.ae_tag == ACL_USER) {
449 if (maskallows != -1 && seen_userobj)
451 if (fap->acl_entry[i].ae_tag != ACL_MASK &&
452 fap->acl_entry[i].ae_tag != ACL_USER_OBJ)
455 /* True if this entry allows the requested access */
456 allows = ((fap->acl_entry[i].ae_perm & md) == md);
458 switch (fap->acl_entry[i].ae_tag) {
461 if (fuid != current->fsuid)
463 matched.ae_tag = ACL_USER_OBJ;
464 matched.ae_perm = allows;
467 if (fap->acl_entry[i].ae_id != current->fsuid)
469 matched.ae_tag = ACL_USER;
470 matched.ae_perm = allows;
473 if ((matched.ae_tag == ACL_GROUP_OBJ ||
474 matched.ae_tag == ACL_GROUP) && !allows)
476 if (!in_group_p(fgid))
478 matched.ae_tag = ACL_GROUP_OBJ;
479 matched.ae_perm = allows;
482 if ((matched.ae_tag == ACL_GROUP_OBJ ||
483 matched.ae_tag == ACL_GROUP) && !allows)
485 if (!in_group_p(fap->acl_entry[i].ae_id))
487 matched.ae_tag = ACL_GROUP;
488 matched.ae_perm = allows;
494 if (matched.ae_tag != 0)
496 matched.ae_tag = ACL_OTHER;
497 matched.ae_perm = allows;
502 * First possibility is that no matched entry allows access.
503 * The capability to override DAC may exist, so check for it.
505 switch (matched.ae_tag) {
514 if (maskallows && matched.ae_perm)
521 return xfs_acl_capability_check(md, cr);
525 * ACL validity checker.
526 * This acl validation routine checks each ACL entry read in makes sense.
532 xfs_acl_entry_t *entry, *e;
533 int user = 0, group = 0, other = 0, mask = 0;
534 int mask_required = 0;
540 if (aclp->acl_cnt > XFS_ACL_MAX_ENTRIES)
543 for (i = 0; i < aclp->acl_cnt; i++) {
544 entry = &aclp->acl_entry[i];
545 switch (entry->ae_tag) {
560 for (j = i + 1; j < aclp->acl_cnt; j++) {
561 e = &aclp->acl_entry[j];
562 if (e->ae_id == entry->ae_id &&
563 e->ae_tag == entry->ae_tag)
576 if (!user || !group || !other || (mask_required && !mask))
585 * Do ACL endian conversion.
591 xfs_acl_entry_t *ace, *end;
593 INT_SET(aclp->acl_cnt, ARCH_CONVERT, aclp->acl_cnt);
594 end = &aclp->acl_entry[0]+aclp->acl_cnt;
595 for (ace = &aclp->acl_entry[0]; ace < end; ace++) {
596 INT_SET(ace->ae_tag, ARCH_CONVERT, ace->ae_tag);
597 INT_SET(ace->ae_id, ARCH_CONVERT, ace->ae_id);
598 INT_SET(ace->ae_perm, ARCH_CONVERT, ace->ae_perm);
603 * Get the ACL from the EA and do endian conversion.
613 int len = sizeof(xfs_acl_t);
615 ASSERT((flags & ATTR_KERNOVAL) ? (aclp == NULL) : 1);
617 *error = bhv_vop_attr_get(vp, kind == _ACL_TYPE_ACCESS ?
618 SGI_ACL_FILE : SGI_ACL_DEFAULT,
619 (char *)aclp, &len, flags, sys_cred);
620 if (*error || (flags & ATTR_KERNOVAL))
622 xfs_acl_get_endian(aclp);
626 * Set the EA with the ACL and do endian conversion.
635 xfs_acl_entry_t *ace, *newace, *end;
639 if (!(_ACL_ALLOC(newacl))) {
644 len = sizeof(xfs_acl_t) -
645 (sizeof(xfs_acl_entry_t) * (XFS_ACL_MAX_ENTRIES - aclp->acl_cnt));
646 end = &aclp->acl_entry[0]+aclp->acl_cnt;
647 for (ace = &aclp->acl_entry[0], newace = &newacl->acl_entry[0];
650 INT_SET(newace->ae_tag, ARCH_CONVERT, ace->ae_tag);
651 INT_SET(newace->ae_id, ARCH_CONVERT, ace->ae_id);
652 INT_SET(newace->ae_perm, ARCH_CONVERT, ace->ae_perm);
654 INT_SET(newacl->acl_cnt, ARCH_CONVERT, aclp->acl_cnt);
655 *error = bhv_vop_attr_set(vp, kind == _ACL_TYPE_ACCESS ?
656 SGI_ACL_FILE: SGI_ACL_DEFAULT,
657 (char *)newacl, len, ATTR_ROOT, sys_cred);
664 xfs_acl_t *access_acl,
665 xfs_acl_t *default_acl)
672 * Get the Access ACL and the mode. If either cannot
673 * be obtained for some reason, invalidate the access ACL.
675 xfs_acl_get_attr(vp, access_acl, _ACL_TYPE_ACCESS, 0, &error);
677 /* Got the ACL, need the mode... */
678 va.va_mask = XFS_AT_MODE;
679 error = bhv_vop_getattr(vp, &va, 0, sys_cred);
683 access_acl->acl_cnt = XFS_ACL_NOT_PRESENT;
684 else /* We have a good ACL and the file mode, synchronize. */
685 xfs_acl_sync_mode(va.va_mode, access_acl);
689 xfs_acl_get_attr(vp, default_acl, _ACL_TYPE_DEFAULT, 0, &error);
691 default_acl->acl_cnt = XFS_ACL_NOT_PRESENT;
697 * This function retrieves the parent directory's acl, processes it
698 * and lets the child inherit the acl(s) that it should.
711 * If the parent does not have a default ACL, or it's an
712 * invalid ACL, we're done.
716 if (!pdaclp || xfs_acl_invalid(pdaclp))
720 * Copy the default ACL of the containing directory to
721 * the access ACL of the new file and use the mode that
722 * was passed in to set up the correct initial values for
723 * the u::,g::[m::], and o:: entries. This is what makes
724 * umask() "work" with ACL's.
727 if (!(_ACL_ALLOC(cacl)))
730 memcpy(cacl, pdaclp, sizeof(xfs_acl_t));
731 xfs_acl_filter_mode(vap->va_mode, cacl);
732 xfs_acl_setmode(vp, cacl, &basicperms);
735 * Set the Default and Access ACL on the file. The mode is already
736 * set on the file, so we don't need to worry about that.
738 * If the new file is a directory, its default ACL is a copy of
739 * the containing directory's default ACL.
742 xfs_acl_set_attr(vp, pdaclp, _ACL_TYPE_DEFAULT, &error);
743 if (!error && !basicperms)
744 xfs_acl_set_attr(vp, cacl, _ACL_TYPE_ACCESS, &error);
750 * Set up the correct mode on the file based on the supplied ACL. This
751 * makes sure that the mode on the file reflects the state of the
752 * u::,g::[m::], and o:: entries in the ACL. Since the mode is where
753 * the ACL is going to get the permissions for these entries, we must
754 * synchronize the mode whenever we set the ACL on a file.
764 xfs_acl_entry_t *gap = NULL;
765 int i, error, nomask = 1;
769 if (acl->acl_cnt == XFS_ACL_NOT_PRESENT)
773 * Copy the u::, g::, o::, and m:: bits from the ACL into the
774 * mode. The m:: bits take precedence over the g:: bits.
776 va.va_mask = XFS_AT_MODE;
777 error = bhv_vop_getattr(vp, &va, 0, sys_cred);
781 va.va_mask = XFS_AT_MODE;
782 va.va_mode &= ~(S_IRWXU|S_IRWXG|S_IRWXO);
784 for (i = 0; i < acl->acl_cnt; ++i) {
785 switch (ap->ae_tag) {
787 va.va_mode |= ap->ae_perm << 6;
792 case ACL_MASK: /* more than just standard modes */
794 va.va_mode |= ap->ae_perm << 3;
798 va.va_mode |= ap->ae_perm;
800 default: /* more than just standard modes */
807 /* Set the group bits from ACL_GROUP_OBJ if there's no ACL_MASK */
809 va.va_mode |= gap->ae_perm << 3;
811 return bhv_vop_setattr(vp, &va, 0, sys_cred);
815 * The permissions for the special ACL entries (u::, g::[m::], o::) are
816 * actually stored in the file mode (if there is both a group and a mask,
817 * the group is stored in the ACL entry and the mask is stored on the file).
818 * This allows the mode to remain automatically in sync with the ACL without
819 * the need for a call-back to the ACL system at every point where the mode
820 * could change. This function takes the permissions from the specified mode
821 * and places it in the supplied ACL.
823 * This implementation draws its validity from the fact that, when the ACL
824 * was assigned, the mode was copied from the ACL.
825 * If the mode did not change, therefore, the mode remains exactly what was
826 * taken from the special ACL entries at assignment.
827 * If a subsequent chmod() was done, the POSIX spec says that the change in
828 * mode must cause an update to the ACL seen at user level and used for
829 * access checks. Before and after a mode change, therefore, the file mode
830 * most accurately reflects what the special ACL entries should permit/deny.
832 * CAVEAT: If someone sets the SGI_ACL_FILE attribute directly,
833 * the existing mode bits will override whatever is in the
834 * ACL. Similarly, if there is a pre-existing ACL that was
835 * never in sync with its mode (owing to a bug in 6.5 and
836 * before), it will now magically (or mystically) be
837 * synchronized. This could cause slight astonishment, but
838 * it is better than inconsistent permissions.
840 * The supplied ACL is a template that may contain any combination
841 * of special entries. These are treated as place holders when we fill
842 * out the ACL. This routine does not add or remove special entries, it
843 * simply unites each special entry with its associated set of permissions.
852 xfs_acl_entry_t *gap = NULL;
855 * Set ACL entries. POSIX1003.1eD16 requires that the MASK
856 * be set instead of the GROUP entry, if there is a MASK.
858 for (ap = acl->acl_entry, i = 0; i < acl->acl_cnt; ap++, i++) {
859 switch (ap->ae_tag) {
861 ap->ae_perm = (mode >> 6) & 0x7;
868 ap->ae_perm = (mode >> 3) & 0x7;
871 ap->ae_perm = mode & 0x7;
877 /* Set the ACL_GROUP_OBJ if there's no ACL_MASK */
879 gap->ae_perm = (mode >> 3) & 0x7;
883 * When inheriting an Access ACL from a directory Default ACL,
884 * the ACL bits are set to the intersection of the ACL default
885 * permission bits and the file permission bits in mode. If there
886 * are no permission bits on the file then we must not give them
887 * the ACL. This is what what makes umask() work with ACLs.
896 xfs_acl_entry_t *gap = NULL;
899 * Set ACL entries. POSIX1003.1eD16 requires that the MASK
900 * be merged with GROUP entry, if there is a MASK.
902 for (ap = acl->acl_entry, i = 0; i < acl->acl_cnt; ap++, i++) {
903 switch (ap->ae_tag) {
905 ap->ae_perm &= (mode >> 6) & 0x7;
912 ap->ae_perm &= (mode >> 3) & 0x7;
915 ap->ae_perm &= mode & 0x7;
921 /* Set the ACL_GROUP_OBJ if there's no ACL_MASK */
923 gap->ae_perm &= (mode >> 3) & 0x7;