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"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_alloc_btree.h"
28 #include "xfs_ialloc_btree.h"
29 #include "xfs_dir_sf.h"
30 #include "xfs_dir2_sf.h"
31 #include "xfs_attr_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
39 #include <linux/capability.h>
40 #include <linux/posix_acl_xattr.h>
42 STATIC int xfs_acl_setmode(vnode_t *, xfs_acl_t *, int *);
43 STATIC void xfs_acl_filter_mode(mode_t, xfs_acl_t *);
44 STATIC void xfs_acl_get_endian(xfs_acl_t *);
45 STATIC int xfs_acl_access(uid_t, gid_t, xfs_acl_t *, mode_t, cred_t *);
46 STATIC int xfs_acl_invalid(xfs_acl_t *);
47 STATIC void xfs_acl_sync_mode(mode_t, xfs_acl_t *);
48 STATIC void xfs_acl_get_attr(vnode_t *, xfs_acl_t *, int, int, int *);
49 STATIC void xfs_acl_set_attr(vnode_t *, xfs_acl_t *, int, int *);
50 STATIC int xfs_acl_allow_set(vnode_t *, int);
52 kmem_zone_t *xfs_acl_zone;
56 * Test for existence of access ACL attribute as efficiently as possible.
59 xfs_acl_vhasacl_access(
64 xfs_acl_get_attr(vp, NULL, _ACL_TYPE_ACCESS, ATTR_KERNOVAL, &error);
69 * Test for existence of default ACL attribute as efficiently as possible.
72 xfs_acl_vhasacl_default(
79 xfs_acl_get_attr(vp, NULL, _ACL_TYPE_DEFAULT, ATTR_KERNOVAL, &error);
84 * Convert from extended attribute representation to in-memory for XFS.
87 posix_acl_xattr_to_xfs(
88 posix_acl_xattr_header *src,
92 posix_acl_xattr_entry *src_entry;
93 xfs_acl_entry_t *dest_entry;
99 if (size < sizeof(posix_acl_xattr_header))
102 if (src->a_version != cpu_to_le32(POSIX_ACL_XATTR_VERSION))
105 memset(dest, 0, sizeof(xfs_acl_t));
106 dest->acl_cnt = posix_acl_xattr_count(size);
107 if (dest->acl_cnt < 0 || dest->acl_cnt > XFS_ACL_MAX_ENTRIES)
111 * acl_set_file(3) may request that we set default ACLs with
112 * zero length -- defend (gracefully) against that here.
117 src_entry = (posix_acl_xattr_entry *)((char *)src + sizeof(*src));
118 dest_entry = &dest->acl_entry[0];
120 for (n = 0; n < dest->acl_cnt; n++, src_entry++, dest_entry++) {
121 dest_entry->ae_perm = le16_to_cpu(src_entry->e_perm);
122 if (_ACL_PERM_INVALID(dest_entry->ae_perm))
124 dest_entry->ae_tag = le16_to_cpu(src_entry->e_tag);
125 switch(dest_entry->ae_tag) {
128 dest_entry->ae_id = le32_to_cpu(src_entry->e_id);
134 dest_entry->ae_id = ACL_UNDEFINED_ID;
140 if (xfs_acl_invalid(dest))
147 * Comparison function called from xfs_sort().
148 * Primary key is ae_tag, secondary key is ae_id.
151 xfs_acl_entry_compare(
155 xfs_acl_entry_t *a = (xfs_acl_entry_t *)va,
156 *b = (xfs_acl_entry_t *)vb;
158 if (a->ae_tag == b->ae_tag)
159 return (a->ae_id - b->ae_id);
160 return (a->ae_tag - b->ae_tag);
164 * Convert from in-memory XFS to extended attribute representation.
167 posix_acl_xfs_to_xattr(
169 posix_acl_xattr_header *dest,
173 size_t new_size = posix_acl_xattr_size(src->acl_cnt);
174 posix_acl_xattr_entry *dest_entry;
175 xfs_acl_entry_t *src_entry;
180 /* Need to sort src XFS ACL by <ae_tag,ae_id> */
181 xfs_sort(src->acl_entry, src->acl_cnt, sizeof(src->acl_entry[0]),
182 xfs_acl_entry_compare);
184 dest->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
185 dest_entry = &dest->a_entries[0];
186 src_entry = &src->acl_entry[0];
187 for (n = 0; n < src->acl_cnt; n++, dest_entry++, src_entry++) {
188 dest_entry->e_perm = cpu_to_le16(src_entry->ae_perm);
189 if (_ACL_PERM_INVALID(src_entry->ae_perm))
191 dest_entry->e_tag = cpu_to_le16(src_entry->ae_tag);
192 switch (src_entry->ae_tag) {
195 dest_entry->e_id = cpu_to_le32(src_entry->ae_id);
201 dest_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
218 xfs_acl_t *xfs_acl = NULL;
219 posix_acl_xattr_header *ext_acl = acl;
224 if (!(_ACL_ALLOC(xfs_acl))) {
228 memset(xfs_acl, 0, sizeof(xfs_acl_t));
230 flags = ATTR_KERNOVAL;
232 xfs_acl_get_attr(vp, xfs_acl, kind, flags, &error);
237 error = -posix_acl_xattr_size(XFS_ACL_MAX_ENTRIES);
239 if (xfs_acl_invalid(xfs_acl)) {
243 if (kind == _ACL_TYPE_ACCESS) {
246 va.va_mask = XFS_AT_MODE;
247 VOP_GETATTR(vp, &va, 0, sys_cred, error);
250 xfs_acl_sync_mode(va.va_mode, xfs_acl);
252 error = -posix_acl_xfs_to_xattr(xfs_acl, ext_acl, size);
269 error = xfs_acl_allow_set(vp, kind);
271 VOP_ATTR_REMOVE(vp, kind == _ACL_TYPE_DEFAULT?
272 SGI_ACL_DEFAULT: SGI_ACL_FILE,
273 ATTR_ROOT, sys_cred, error);
274 if (error == ENOATTR)
275 error = 0; /* 'scool */
288 posix_acl_xattr_header *ext_acl = acl;
291 int basicperms = 0; /* more than std unix perms? */
296 if (!(_ACL_ALLOC(xfs_acl)))
299 error = posix_acl_xattr_to_xfs(ext_acl, size, xfs_acl);
304 if (!xfs_acl->acl_cnt) {
310 error = xfs_acl_allow_set(vp, kind);
314 /* Incoming ACL exists, set file mode based on its value */
315 if (kind == _ACL_TYPE_ACCESS)
316 xfs_acl_setmode(vp, xfs_acl, &basicperms);
319 * If we have more than std unix permissions, set up the actual attr.
320 * Otherwise, delete any existing attr. This prevents us from
321 * having actual attrs for permissions that can be stored in the
322 * standard permission bits.
325 xfs_acl_set_attr(vp, xfs_acl, kind, &error);
327 xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
345 if (!(_ACL_ALLOC(acl)))
348 /* If the file has no ACL return -1. */
349 rval = sizeof(xfs_acl_t);
350 if (xfs_attr_fetch(ip, SGI_ACL_FILE, SGI_ACL_FILE_SIZE,
351 (char *)acl, &rval, ATTR_ROOT | ATTR_KERNACCESS, cr)) {
355 xfs_acl_get_endian(acl);
357 /* If the file has an empty ACL return -1. */
358 if (acl->acl_cnt == XFS_ACL_NOT_PRESENT) {
363 /* Synchronize ACL with mode bits */
364 xfs_acl_sync_mode(ip->i_d.di_mode, acl);
366 rval = xfs_acl_access(ip->i_d.di_uid, ip->i_d.di_gid, acl, mode, cr);
379 if (vp->v_inode.i_flags & (S_IMMUTABLE|S_APPEND))
381 if (kind == _ACL_TYPE_DEFAULT && !VN_ISDIR(vp))
383 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
385 va.va_mask = XFS_AT_UID;
386 VOP_GETATTR(vp, &va, 0, NULL, error);
389 if (va.va_uid != current->fsuid && !capable(CAP_FOWNER))
395 * The access control process to determine the access permission:
396 * if uid == file owner id, use the file owner bits.
397 * if gid == file owner group id, use the file group bits.
398 * scan ACL for a maching user or group, and use matched entry
399 * permission. Use total permissions of all matching group entries,
400 * until all acl entries are exhausted. The final permission produced
401 * by matching acl entry or entries needs to be & with group permission.
402 * if not owner, owning group, or matching entry in ACL, use file
406 xfs_acl_capability_check(
410 if ((mode & ACL_READ) && !capable_cred(cr, CAP_DAC_READ_SEARCH))
412 if ((mode & ACL_WRITE) && !capable_cred(cr, CAP_DAC_OVERRIDE))
414 if ((mode & ACL_EXECUTE) && !capable_cred(cr, CAP_DAC_OVERRIDE))
421 * Note: cr is only used here for the capability check if the ACL test fails.
422 * It is not used to find out the credentials uid or groups etc, as was
423 * done in IRIX. It is assumed that the uid and groups for the current
424 * thread are taken from "current" instead of the cr parameter.
434 xfs_acl_entry_t matched;
436 int maskallows = -1; /* true, but not 1, either */
437 int seen_userobj = 0;
439 matched.ae_tag = 0; /* Invalid type */
441 md >>= 6; /* Normalize the bits for comparison */
443 for (i = 0; i < fap->acl_cnt; i++) {
445 * Break out if we've got a user_obj entry or
446 * a user entry and the mask (and have processed USER_OBJ)
448 if (matched.ae_tag == ACL_USER_OBJ)
450 if (matched.ae_tag == ACL_USER) {
451 if (maskallows != -1 && seen_userobj)
453 if (fap->acl_entry[i].ae_tag != ACL_MASK &&
454 fap->acl_entry[i].ae_tag != ACL_USER_OBJ)
457 /* True if this entry allows the requested access */
458 allows = ((fap->acl_entry[i].ae_perm & md) == md);
460 switch (fap->acl_entry[i].ae_tag) {
463 if (fuid != current->fsuid)
465 matched.ae_tag = ACL_USER_OBJ;
466 matched.ae_perm = allows;
469 if (fap->acl_entry[i].ae_id != current->fsuid)
471 matched.ae_tag = ACL_USER;
472 matched.ae_perm = allows;
475 if ((matched.ae_tag == ACL_GROUP_OBJ ||
476 matched.ae_tag == ACL_GROUP) && !allows)
478 if (!in_group_p(fgid))
480 matched.ae_tag = ACL_GROUP_OBJ;
481 matched.ae_perm = allows;
484 if ((matched.ae_tag == ACL_GROUP_OBJ ||
485 matched.ae_tag == ACL_GROUP) && !allows)
487 if (!in_group_p(fap->acl_entry[i].ae_id))
489 matched.ae_tag = ACL_GROUP;
490 matched.ae_perm = allows;
496 if (matched.ae_tag != 0)
498 matched.ae_tag = ACL_OTHER;
499 matched.ae_perm = allows;
504 * First possibility is that no matched entry allows access.
505 * The capability to override DAC may exist, so check for it.
507 switch (matched.ae_tag) {
516 if (maskallows && matched.ae_perm)
523 return xfs_acl_capability_check(md, cr);
527 * ACL validity checker.
528 * This acl validation routine checks each ACL entry read in makes sense.
534 xfs_acl_entry_t *entry, *e;
535 int user = 0, group = 0, other = 0, mask = 0;
536 int mask_required = 0;
542 if (aclp->acl_cnt > XFS_ACL_MAX_ENTRIES)
545 for (i = 0; i < aclp->acl_cnt; i++) {
546 entry = &aclp->acl_entry[i];
547 switch (entry->ae_tag) {
562 for (j = i + 1; j < aclp->acl_cnt; j++) {
563 e = &aclp->acl_entry[j];
564 if (e->ae_id == entry->ae_id &&
565 e->ae_tag == entry->ae_tag)
578 if (!user || !group || !other || (mask_required && !mask))
587 * Do ACL endian conversion.
593 xfs_acl_entry_t *ace, *end;
595 INT_SET(aclp->acl_cnt, ARCH_CONVERT, aclp->acl_cnt);
596 end = &aclp->acl_entry[0]+aclp->acl_cnt;
597 for (ace = &aclp->acl_entry[0]; ace < end; ace++) {
598 INT_SET(ace->ae_tag, ARCH_CONVERT, ace->ae_tag);
599 INT_SET(ace->ae_id, ARCH_CONVERT, ace->ae_id);
600 INT_SET(ace->ae_perm, ARCH_CONVERT, ace->ae_perm);
605 * Get the ACL from the EA and do endian conversion.
615 int len = sizeof(xfs_acl_t);
617 ASSERT((flags & ATTR_KERNOVAL) ? (aclp == NULL) : 1);
620 kind == _ACL_TYPE_ACCESS ? SGI_ACL_FILE : SGI_ACL_DEFAULT,
621 (char *)aclp, &len, flags, sys_cred, *error);
622 if (*error || (flags & ATTR_KERNOVAL))
624 xfs_acl_get_endian(aclp);
628 * Set the EA with the ACL and do endian conversion.
637 xfs_acl_entry_t *ace, *newace, *end;
641 if (!(_ACL_ALLOC(newacl))) {
646 len = sizeof(xfs_acl_t) -
647 (sizeof(xfs_acl_entry_t) * (XFS_ACL_MAX_ENTRIES - aclp->acl_cnt));
648 end = &aclp->acl_entry[0]+aclp->acl_cnt;
649 for (ace = &aclp->acl_entry[0], newace = &newacl->acl_entry[0];
652 INT_SET(newace->ae_tag, ARCH_CONVERT, ace->ae_tag);
653 INT_SET(newace->ae_id, ARCH_CONVERT, ace->ae_id);
654 INT_SET(newace->ae_perm, ARCH_CONVERT, ace->ae_perm);
656 INT_SET(newacl->acl_cnt, ARCH_CONVERT, aclp->acl_cnt);
658 kind == _ACL_TYPE_ACCESS ? SGI_ACL_FILE: SGI_ACL_DEFAULT,
659 (char *)newacl, len, ATTR_ROOT, sys_cred, *error);
666 xfs_acl_t *access_acl,
667 xfs_acl_t *default_acl)
674 * Get the Access ACL and the mode. If either cannot
675 * be obtained for some reason, invalidate the access ACL.
677 xfs_acl_get_attr(vp, access_acl, _ACL_TYPE_ACCESS, 0, &error);
679 /* Got the ACL, need the mode... */
680 va.va_mask = XFS_AT_MODE;
681 VOP_GETATTR(vp, &va, 0, sys_cred, error);
685 access_acl->acl_cnt = XFS_ACL_NOT_PRESENT;
686 else /* We have a good ACL and the file mode, synchronize. */
687 xfs_acl_sync_mode(va.va_mode, access_acl);
691 xfs_acl_get_attr(vp, default_acl, _ACL_TYPE_DEFAULT, 0, &error);
693 default_acl->acl_cnt = XFS_ACL_NOT_PRESENT;
699 * This function retrieves the parent directory's acl, processes it
700 * and lets the child inherit the acl(s) that it should.
713 * If the parent does not have a default ACL, or it's an
714 * invalid ACL, we're done.
718 if (!pdaclp || xfs_acl_invalid(pdaclp))
722 * Copy the default ACL of the containing directory to
723 * the access ACL of the new file and use the mode that
724 * was passed in to set up the correct initial values for
725 * the u::,g::[m::], and o:: entries. This is what makes
726 * umask() "work" with ACL's.
729 if (!(_ACL_ALLOC(cacl)))
732 memcpy(cacl, pdaclp, sizeof(xfs_acl_t));
733 xfs_acl_filter_mode(vap->va_mode, cacl);
734 xfs_acl_setmode(vp, cacl, &basicperms);
737 * Set the Default and Access ACL on the file. The mode is already
738 * set on the file, so we don't need to worry about that.
740 * If the new file is a directory, its default ACL is a copy of
741 * the containing directory's default ACL.
744 xfs_acl_set_attr(vp, pdaclp, _ACL_TYPE_DEFAULT, &error);
745 if (!error && !basicperms)
746 xfs_acl_set_attr(vp, cacl, _ACL_TYPE_ACCESS, &error);
752 * Set up the correct mode on the file based on the supplied ACL. This
753 * makes sure that the mode on the file reflects the state of the
754 * u::,g::[m::], and o:: entries in the ACL. Since the mode is where
755 * the ACL is going to get the permissions for these entries, we must
756 * synchronize the mode whenever we set the ACL on a file.
766 xfs_acl_entry_t *gap = NULL;
767 int i, error, nomask = 1;
771 if (acl->acl_cnt == XFS_ACL_NOT_PRESENT)
775 * Copy the u::, g::, o::, and m:: bits from the ACL into the
776 * mode. The m:: bits take precedence over the g:: bits.
778 va.va_mask = XFS_AT_MODE;
779 VOP_GETATTR(vp, &va, 0, sys_cred, error);
783 va.va_mask = XFS_AT_MODE;
784 va.va_mode &= ~(S_IRWXU|S_IRWXG|S_IRWXO);
786 for (i = 0; i < acl->acl_cnt; ++i) {
787 switch (ap->ae_tag) {
789 va.va_mode |= ap->ae_perm << 6;
794 case ACL_MASK: /* more than just standard modes */
796 va.va_mode |= ap->ae_perm << 3;
800 va.va_mode |= ap->ae_perm;
802 default: /* more than just standard modes */
809 /* Set the group bits from ACL_GROUP_OBJ if there's no ACL_MASK */
811 va.va_mode |= gap->ae_perm << 3;
813 VOP_SETATTR(vp, &va, 0, sys_cred, error);
818 * The permissions for the special ACL entries (u::, g::[m::], o::) are
819 * actually stored in the file mode (if there is both a group and a mask,
820 * the group is stored in the ACL entry and the mask is stored on the file).
821 * This allows the mode to remain automatically in sync with the ACL without
822 * the need for a call-back to the ACL system at every point where the mode
823 * could change. This function takes the permissions from the specified mode
824 * and places it in the supplied ACL.
826 * This implementation draws its validity from the fact that, when the ACL
827 * was assigned, the mode was copied from the ACL.
828 * If the mode did not change, therefore, the mode remains exactly what was
829 * taken from the special ACL entries at assignment.
830 * If a subsequent chmod() was done, the POSIX spec says that the change in
831 * mode must cause an update to the ACL seen at user level and used for
832 * access checks. Before and after a mode change, therefore, the file mode
833 * most accurately reflects what the special ACL entries should permit/deny.
835 * CAVEAT: If someone sets the SGI_ACL_FILE attribute directly,
836 * the existing mode bits will override whatever is in the
837 * ACL. Similarly, if there is a pre-existing ACL that was
838 * never in sync with its mode (owing to a bug in 6.5 and
839 * before), it will now magically (or mystically) be
840 * synchronized. This could cause slight astonishment, but
841 * it is better than inconsistent permissions.
843 * The supplied ACL is a template that may contain any combination
844 * of special entries. These are treated as place holders when we fill
845 * out the ACL. This routine does not add or remove special entries, it
846 * simply unites each special entry with its associated set of permissions.
855 xfs_acl_entry_t *gap = NULL;
858 * Set ACL entries. POSIX1003.1eD16 requires that the MASK
859 * be set instead of the GROUP entry, if there is a MASK.
861 for (ap = acl->acl_entry, i = 0; i < acl->acl_cnt; ap++, i++) {
862 switch (ap->ae_tag) {
864 ap->ae_perm = (mode >> 6) & 0x7;
871 ap->ae_perm = (mode >> 3) & 0x7;
874 ap->ae_perm = mode & 0x7;
880 /* Set the ACL_GROUP_OBJ if there's no ACL_MASK */
882 gap->ae_perm = (mode >> 3) & 0x7;
886 * When inheriting an Access ACL from a directory Default ACL,
887 * the ACL bits are set to the intersection of the ACL default
888 * permission bits and the file permission bits in mode. If there
889 * are no permission bits on the file then we must not give them
890 * the ACL. This is what what makes umask() work with ACLs.
899 xfs_acl_entry_t *gap = NULL;
902 * Set ACL entries. POSIX1003.1eD16 requires that the MASK
903 * be merged with GROUP entry, if there is a MASK.
905 for (ap = acl->acl_entry, i = 0; i < acl->acl_cnt; ap++, i++) {
906 switch (ap->ae_tag) {
908 ap->ae_perm &= (mode >> 6) & 0x7;
915 ap->ae_perm &= (mode >> 3) & 0x7;
918 ap->ae_perm &= mode & 0x7;
924 /* Set the ACL_GROUP_OBJ if there's no ACL_MASK */
926 gap->ae_perm &= (mode >> 3) & 0x7;