darling-xnu/bsd/vfs/vfs_xattr.c
2023-05-16 21:41:14 -07:00

3468 lines
93 KiB
C

/*
* Copyright (c) 2004-2012 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* compliance with the License. The rights granted to you under the License
* may not be used to create, or enable the creation or redistribution of,
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
*
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
*
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
* support for mandatory and extensible security protections. This notice
* is included in support of clause 2.2 (b) of the Apple Public License,
* Version 2.0.
*/
#include <sys/param.h>
#include <sys/fcntl.h>
#include <sys/fsevents.h>
#include <sys/kernel.h>
#include <sys/kauth.h>
#include <kern/kalloc.h>
#include <sys/mount_internal.h>
#include <sys/namei.h>
#include <sys/proc_internal.h>
#include <sys/stat.h>
#include <sys/uio.h>
#include <sys/utfconv.h>
#include <sys/vnode.h>
#include <sys/vnode_internal.h>
#include <sys/xattr.h>
#include <libkern/OSByteOrder.h>
#include <vm/vm_kern.h>
#if CONFIG_MACF
#include <security/mac_framework.h>
#endif
#if NAMEDSTREAMS
static int shadow_sequence;
/*
* We use %p to prevent loss of precision for pointers on varying architectures.
*/
#define SHADOW_NAME_FMT ".vfs_rsrc_stream_%p%08x%p"
#define SHADOW_DIR_FMT ".vfs_rsrc_streams_%p%x"
#define SHADOW_DIR_CONTAINER "/var/run"
#define MAKE_SHADOW_NAME(VP, NAME) \
snprintf((NAME), sizeof((NAME)), (SHADOW_NAME_FMT), \
((void*)(VM_KERNEL_ADDRPERM(VP))), \
(VP)->v_id, \
((void*)(VM_KERNEL_ADDRPERM((VP)->v_data))))
/* The full path to the shadow directory */
#define MAKE_SHADOW_DIRNAME(VP, NAME) \
snprintf((NAME), sizeof((NAME)), (SHADOW_DIR_CONTAINER "/" SHADOW_DIR_FMT), \
((void*)(VM_KERNEL_ADDRPERM(VP))), shadow_sequence)
/* The shadow directory as a 'leaf' entry */
#define MAKE_SHADOW_DIR_LEAF(VP, NAME) \
snprintf((NAME), sizeof((NAME)), (SHADOW_DIR_FMT), \
((void*)(VM_KERNEL_ADDRPERM(VP))), shadow_sequence)
static int default_getnamedstream(vnode_t vp, vnode_t *svpp, const char *name, enum nsoperation op, vfs_context_t context);
static int default_makenamedstream(vnode_t vp, vnode_t *svpp, const char *name, vfs_context_t context);
static int default_removenamedstream(vnode_t vp, const char *name, vfs_context_t context);
static int getshadowfile(vnode_t vp, vnode_t *svpp, int makestream, size_t *rsrcsize, int *creator, vfs_context_t context);
static int get_shadow_dir(vnode_t *sdvpp);
#endif /* NAMEDSTREAMS */
/*
* Default xattr support routines.
*/
static int default_getxattr(vnode_t vp, const char *name, uio_t uio, size_t *size, int options,
vfs_context_t context);
static int default_setxattr(vnode_t vp, const char *name, uio_t uio, int options,
vfs_context_t context);
static int default_listxattr(vnode_t vp, uio_t uio, size_t *size, int options,
vfs_context_t context);
static int default_removexattr(vnode_t vp, const char *name, int options,
vfs_context_t context);
/*
* Retrieve the data of an extended attribute.
*/
int
vn_getxattr(vnode_t vp, const char *name, uio_t uio, size_t *size,
int options, vfs_context_t context)
{
int error;
if (!XATTR_VNODE_SUPPORTED(vp)) {
return EPERM;
}
#if NAMEDSTREAMS
/* getxattr calls are not allowed for streams. */
if (vp->v_flag & VISNAMEDSTREAM) {
error = EPERM;
goto out;
}
#endif
/*
* Non-kernel request need extra checks performed.
*
* The XATTR_NOSECURITY flag implies a kernel request.
*/
if (!(options & XATTR_NOSECURITY)) {
#if CONFIG_MACF
error = mac_vnode_check_getextattr(context, vp, name, uio);
if (error) {
goto out;
}
#endif /* MAC */
if ((error = xattr_validatename(name))) {
goto out;
}
if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_EXTATTRIBUTES, context))) {
goto out;
}
/* The offset can only be non-zero for resource forks. */
if (uio != NULL && uio_offset(uio) != 0 &&
strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
error = EINVAL;
goto out;
}
}
/* The offset can only be non-zero for resource forks. */
if (uio != NULL && uio_offset(uio) != 0 &&
strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
error = EINVAL;
goto out;
}
error = VNOP_GETXATTR(vp, name, uio, size, options, context);
if (error == ENOTSUP && !(options & XATTR_NODEFAULT)) {
/*
* A filesystem may keep some EAs natively and return ENOTSUP for others.
*/
error = default_getxattr(vp, name, uio, size, options, context);
}
out:
return error;
}
/*
* Set the data of an extended attribute.
*/
int
vn_setxattr(vnode_t vp, const char *name, uio_t uio, int options, vfs_context_t context)
{
int error;
if (!XATTR_VNODE_SUPPORTED(vp)) {
return EPERM;
}
#if NAMEDSTREAMS
/* setxattr calls are not allowed for streams. */
if (vp->v_flag & VISNAMEDSTREAM) {
error = EPERM;
goto out;
}
#endif
if ((options & (XATTR_REPLACE | XATTR_CREATE)) == (XATTR_REPLACE | XATTR_CREATE)) {
return EINVAL;
}
if ((error = xattr_validatename(name))) {
return error;
}
if (!(options & XATTR_NOSECURITY)) {
#if CONFIG_MACF
error = mac_vnode_check_setextattr(context, vp, name, uio);
if (error) {
goto out;
}
#endif /* MAC */
error = vnode_authorize(vp, NULL, KAUTH_VNODE_WRITE_EXTATTRIBUTES, context);
if (error) {
goto out;
}
}
/* The offset can only be non-zero for resource forks. */
if (uio_offset(uio) != 0 &&
strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
error = EINVAL;
goto out;
}
error = VNOP_SETXATTR(vp, name, uio, options, context);
#ifdef DUAL_EAS
/*
* An EJUSTRETURN is from a filesystem which keeps this xattr
* natively as well as in a dot-underscore file. In this case the
* EJUSTRETURN means the filesytem has done nothing, but identifies the
* EA as one which may be represented natively and/or in a DU, and
* since XATTR_CREATE or XATTR_REPLACE was specified, only up here in
* in vn_setxattr can we do the getxattrs needed to ascertain whether
* the XATTR_{CREATE,REPLACE} should yield an error.
*/
if (error == EJUSTRETURN) {
int native = 0, dufile = 0;
size_t sz; /* not used */
native = VNOP_GETXATTR(vp, name, NULL, &sz, 0, context) ? 0 : 1;
dufile = default_getxattr(vp, name, NULL, &sz, 0, context) ? 0 : 1;
if (options & XATTR_CREATE && (native || dufile)) {
error = EEXIST;
goto out;
}
if (options & XATTR_REPLACE && !(native || dufile)) {
error = ENOATTR;
goto out;
}
/*
* Having determined no CREATE/REPLACE error should result, we
* zero those bits, so both backing stores get written to.
*/
options &= ~(XATTR_CREATE | XATTR_REPLACE);
error = VNOP_SETXATTR(vp, name, uio, options, context);
/* the mainline path here is to have error==ENOTSUP ... */
}
#endif /* DUAL_EAS */
if (error == ENOTSUP && !(options & XATTR_NODEFAULT)) {
/*
* A filesystem may keep some EAs natively and return ENOTSUP for others.
*/
error = default_setxattr(vp, name, uio, options, context);
}
#if CONFIG_MACF
if ((error == 0) && !(options & XATTR_NOSECURITY)) {
mac_vnode_notify_setextattr(context, vp, name, uio);
if (vfs_flags(vnode_mount(vp)) & MNT_MULTILABEL) {
mac_vnode_label_update_extattr(vnode_mount(vp), vp, name);
}
}
#endif
out:
return error;
}
/*
* Remove an extended attribute.
*/
int
vn_removexattr(vnode_t vp, const char * name, int options, vfs_context_t context)
{
int error;
if (!XATTR_VNODE_SUPPORTED(vp)) {
return EPERM;
}
#if NAMEDSTREAMS
/* removexattr calls are not allowed for streams. */
if (vp->v_flag & VISNAMEDSTREAM) {
error = EPERM;
goto out;
}
#endif
if ((error = xattr_validatename(name))) {
return error;
}
if (!(options & XATTR_NOSECURITY)) {
#if CONFIG_MACF
error = mac_vnode_check_deleteextattr(context, vp, name);
if (error) {
goto out;
}
#endif /* MAC */
error = vnode_authorize(vp, NULL, KAUTH_VNODE_WRITE_EXTATTRIBUTES, context);
if (error) {
goto out;
}
}
error = VNOP_REMOVEXATTR(vp, name, options, context);
if (error == ENOTSUP && !(options & XATTR_NODEFAULT)) {
/*
* A filesystem may keep some EAs natively and return ENOTSUP for others.
*/
error = default_removexattr(vp, name, options, context);
#ifdef DUAL_EAS
} else if (error == EJUSTRETURN) {
/*
* EJUSTRETURN is from a filesystem which keeps this xattr natively as well
* as in a dot-underscore file. EJUSTRETURN means the filesytem did remove
* a native xattr, so failure to find it in a DU file during
* default_removexattr should not be considered an error.
*/
error = default_removexattr(vp, name, options, context);
if (error == ENOATTR) {
error = 0;
}
#endif /* DUAL_EAS */
}
#if CONFIG_MACF
if ((error == 0) && !(options & XATTR_NOSECURITY)) {
mac_vnode_notify_deleteextattr(context, vp, name);
if (vfs_flags(vnode_mount(vp)) & MNT_MULTILABEL) {
mac_vnode_label_update_extattr(vnode_mount(vp), vp, name);
}
}
#endif
out:
return error;
}
/*
* Retrieve the list of extended attribute names.
*/
int
vn_listxattr(vnode_t vp, uio_t uio, size_t *size, int options, vfs_context_t context)
{
int error;
if (!XATTR_VNODE_SUPPORTED(vp)) {
return EPERM;
}
#if NAMEDSTREAMS
/* listxattr calls are not allowed for streams. */
if (vp->v_flag & VISNAMEDSTREAM) {
return EPERM;
}
#endif
if (!(options & XATTR_NOSECURITY)) {
#if CONFIG_MACF
error = mac_vnode_check_listextattr(context, vp);
if (error) {
goto out;
}
#endif /* MAC */
error = vnode_authorize(vp, NULL, KAUTH_VNODE_READ_EXTATTRIBUTES, context);
if (error) {
goto out;
}
}
error = VNOP_LISTXATTR(vp, uio, size, options, context);
if (error == ENOTSUP && !(options & XATTR_NODEFAULT)) {
/*
* A filesystem may keep some but not all EAs natively, in which case
* the native EA names will have been uiomove-d out (or *size updated)
* and the default_listxattr here will finish the job.
*/
error = default_listxattr(vp, uio, size, options, context);
}
out:
return error;
}
int
xattr_validatename(const char *name)
{
size_t namelen;
if (name == NULL || name[0] == '\0') {
return EINVAL;
}
namelen = strlen(name);
if (utf8_validatestr((const unsigned char *)name, namelen) != 0) {
return EINVAL;
}
return 0;
}
/*
* Determine whether an EA is a protected system attribute.
*/
int
xattr_protected(const char *attrname)
{
return !strncmp(attrname, "com.apple.system.", 17);
}
static void
vnode_setasnamedstream_internal(vnode_t vp, vnode_t svp)
{
uint32_t streamflags = VISNAMEDSTREAM;
if ((vp->v_mount->mnt_kern_flag & MNTK_NAMED_STREAMS) == 0) {
streamflags |= VISSHADOW;
}
/* Tag the vnode. */
vnode_lock_spin(svp);
svp->v_flag |= streamflags;
vnode_unlock(svp);
/* Tag the parent so we know to flush credentials for streams on setattr */
vnode_lock_spin(vp);
vp->v_lflag |= VL_HASSTREAMS;
vnode_unlock(vp);
/* Make the file it's parent.
* Note: This parent link helps us distinguish vnodes for
* shadow stream files from vnodes for resource fork on file
* systems that support namedstream natively (both have
* VISNAMEDSTREAM set) by allowing access to mount structure
* for checking MNTK_NAMED_STREAMS bit at many places in the
* code.
*/
vnode_update_identity(svp, vp, NULL, 0, 0, VNODE_UPDATE_NAMEDSTREAM_PARENT);
if (vnode_isdyldsharedcache(vp)) {
vnode_lock_spin(svp);
svp->v_flag |= VSHARED_DYLD;
vnode_unlock(svp);
}
return;
}
errno_t
vnode_setasnamedstream(vnode_t vp, vnode_t svp)
{
if ((vp->v_mount->mnt_kern_flag & MNTK_NAMED_STREAMS) == 0) {
return EINVAL;
}
vnode_setasnamedstream_internal(vp, svp);
return 0;
}
#if NAMEDSTREAMS
/*
* Obtain a named stream from vnode vp.
*/
errno_t
vnode_getnamedstream(vnode_t vp, vnode_t *svpp, const char *name, enum nsoperation op, int flags, vfs_context_t context)
{
int error;
if (vp->v_mount->mnt_kern_flag & MNTK_NAMED_STREAMS) {
error = VNOP_GETNAMEDSTREAM(vp, svpp, name, op, flags, context);
} else {
if (flags) {
error = ENOTSUP;
} else {
error = default_getnamedstream(vp, svpp, name, op, context);
}
}
if (error == 0) {
vnode_setasnamedstream_internal(vp, *svpp);
}
return error;
}
/*
* Make a named stream for vnode vp.
*/
errno_t
vnode_makenamedstream(vnode_t vp, vnode_t *svpp, const char *name, int flags, vfs_context_t context)
{
int error;
if (vp->v_mount->mnt_kern_flag & MNTK_NAMED_STREAMS) {
error = VNOP_MAKENAMEDSTREAM(vp, svpp, name, flags, context);
} else {
error = default_makenamedstream(vp, svpp, name, context);
}
if (error == 0) {
vnode_setasnamedstream_internal(vp, *svpp);
}
return error;
}
/*
* Remove a named stream from vnode vp.
*/
errno_t
vnode_removenamedstream(vnode_t vp, vnode_t svp, const char *name, int flags, vfs_context_t context)
{
int error;
if (vp->v_mount->mnt_kern_flag & MNTK_NAMED_STREAMS) {
error = VNOP_REMOVENAMEDSTREAM(vp, svp, name, flags, context);
} else {
error = default_removenamedstream(vp, name, context);
}
return error;
}
#define NS_IOBUFSIZE (128 * 1024)
/*
* Release a named stream shadow file.
*
* Note: This function is called from two places where we do not need
* to check if the vnode has any references held before deleting the
* shadow file. Once from vclean() when the vnode is being reclaimed
* and we do not hold any references on the vnode. Second time from
* default_getnamedstream() when we get an error during shadow stream
* file initialization so that other processes who are waiting for the
* shadow stream file initialization by the creator will get opportunity
* to create and initialize the file again.
*/
errno_t
vnode_relenamedstream(vnode_t vp, vnode_t svp)
{
vnode_t dvp;
struct componentname cn;
char tmpname[80];
errno_t err;
/*
* We need to use the kernel context here. If we used the supplied
* VFS context we have no clue whether or not it originated from userland
* where it could be subject to a chroot jail. We need to ensure that all
* filesystem access to shadow files is done on the same FS regardless of
* userland process restrictions.
*/
vfs_context_t kernelctx = vfs_context_kernel();
cache_purge(svp);
vnode_lock(svp);
MAKE_SHADOW_NAME(vp, tmpname);
vnode_unlock(svp);
cn.cn_nameiop = DELETE;
cn.cn_flags = ISLASTCN;
cn.cn_context = kernelctx;
cn.cn_pnbuf = tmpname;
cn.cn_pnlen = sizeof(tmpname);
cn.cn_nameptr = cn.cn_pnbuf;
cn.cn_namelen = (int)strlen(tmpname);
/*
* Obtain the vnode for the shadow files directory. Make sure to
* use the kernel ctx as described above.
*/
err = get_shadow_dir(&dvp);
if (err != 0) {
return err;
}
(void) VNOP_REMOVE(dvp, svp, &cn, 0, kernelctx);
vnode_put(dvp);
return 0;
}
/*
* Flush a named stream shadow file.
*
* 'vp' represents the AppleDouble file.
* 'svp' represents the shadow file.
*/
errno_t
vnode_flushnamedstream(vnode_t vp, vnode_t svp, vfs_context_t context)
{
struct vnode_attr va;
uio_t auio = NULL;
caddr_t bufptr = NULL;
size_t bufsize = 0;
size_t offset;
size_t iosize;
size_t datasize;
int error;
/*
* The kernel context must be used for all I/O to the shadow file
* and its namespace operations
*/
vfs_context_t kernelctx = vfs_context_kernel();
/* The supplied context is used for access to the AD file itself */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_data_size);
if (VNOP_GETATTR(svp, &va, context) != 0 ||
!VATTR_IS_SUPPORTED(&va, va_data_size)) {
return 0;
}
if (va.va_data_size > UINT32_MAX) {
return EINVAL;
}
datasize = (size_t)va.va_data_size;
if (datasize == 0) {
(void) default_removexattr(vp, XATTR_RESOURCEFORK_NAME, 0, context);
return 0;
}
iosize = bufsize = MIN(datasize, NS_IOBUFSIZE);
if (kmem_alloc(kernel_map, (vm_offset_t *)&bufptr, bufsize, VM_KERN_MEMORY_FILE)) {
return ENOMEM;
}
auio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
offset = 0;
/*
* Copy the shadow stream file data into the resource fork.
*/
error = VNOP_OPEN(svp, 0, kernelctx);
if (error) {
printf("vnode_flushnamedstream: err %d opening file\n", error);
goto out;
}
while (offset < datasize) {
iosize = MIN(datasize - offset, iosize);
uio_reset(auio, offset, UIO_SYSSPACE, UIO_READ);
uio_addiov(auio, (uintptr_t)bufptr, iosize);
error = VNOP_READ(svp, auio, 0, kernelctx);
if (error) {
break;
}
/* Since there's no truncate xattr we must remove the resource fork. */
if (offset == 0) {
error = default_removexattr(vp, XATTR_RESOURCEFORK_NAME, 0, context);
if ((error != 0) && (error != ENOATTR)) {
break;
}
}
uio_reset(auio, offset, UIO_SYSSPACE, UIO_WRITE);
uio_addiov(auio, (uintptr_t)bufptr, iosize);
error = vn_setxattr(vp, XATTR_RESOURCEFORK_NAME, auio, XATTR_NOSECURITY, context);
if (error) {
break;
}
offset += iosize;
}
/* close shadowfile */
(void) VNOP_CLOSE(svp, 0, kernelctx);
out:
if (bufptr) {
kmem_free(kernel_map, (vm_offset_t)bufptr, bufsize);
}
if (auio) {
uio_free(auio);
}
return error;
}
/*
* Verify that the vnode 'vp' is a vnode that lives in the shadow
* directory. We can't just query the parent pointer directly since
* the shadowfile is hooked up to the actual file it's a stream for.
*/
errno_t
vnode_verifynamedstream(vnode_t vp)
{
int error;
struct vnode *shadow_dvp = NULL;
struct vnode *shadowfile = NULL;
struct componentname cn;
/*
* We need to use the kernel context here. If we used the supplied
* VFS context we have no clue whether or not it originated from userland
* where it could be subject to a chroot jail. We need to ensure that all
* filesystem access to shadow files is done on the same FS regardless of
* userland process restrictions.
*/
vfs_context_t kernelctx = vfs_context_kernel();
char tmpname[80];
/* Get the shadow directory vnode */
error = get_shadow_dir(&shadow_dvp);
if (error) {
return error;
}
/* Re-generate the shadow name in the buffer */
MAKE_SHADOW_NAME(vp, tmpname);
/* Look up item in shadow dir */
bzero(&cn, sizeof(cn));
cn.cn_nameiop = LOOKUP;
cn.cn_flags = ISLASTCN | CN_ALLOWRSRCFORK;
cn.cn_context = kernelctx;
cn.cn_pnbuf = tmpname;
cn.cn_pnlen = sizeof(tmpname);
cn.cn_nameptr = cn.cn_pnbuf;
cn.cn_namelen = (int)strlen(tmpname);
if (VNOP_LOOKUP(shadow_dvp, &shadowfile, &cn, kernelctx) == 0) {
/* is the pointer the same? */
if (shadowfile == vp) {
error = 0;
} else {
error = EPERM;
}
/* drop the iocount acquired */
vnode_put(shadowfile);
}
/* Drop iocount on shadow dir */
vnode_put(shadow_dvp);
return error;
}
/*
* Access or create the shadow file as needed.
*
* 'makestream' with non-zero value means that we need to guarantee we were the
* creator of the shadow file.
*
* 'context' is the user supplied context for the original VFS operation that
* caused us to need a shadow file.
*
* int pointed to by 'creator' is nonzero if we created the shadowfile.
*/
static int
getshadowfile(vnode_t vp, vnode_t *svpp, int makestream, size_t *rsrcsize,
int *creator, vfs_context_t context)
{
vnode_t dvp = NULLVP;
vnode_t svp = NULLVP;
struct componentname cn;
struct vnode_attr va;
char tmpname[80];
size_t datasize = 0;
int error = 0;
int retries = 0;
vfs_context_t kernelctx = vfs_context_kernel();
retry_create:
*creator = 0;
/* Establish a unique file name. */
MAKE_SHADOW_NAME(vp, tmpname);
bzero(&cn, sizeof(cn));
cn.cn_nameiop = LOOKUP;
cn.cn_flags = ISLASTCN;
cn.cn_context = context;
cn.cn_pnbuf = tmpname;
cn.cn_pnlen = sizeof(tmpname);
cn.cn_nameptr = cn.cn_pnbuf;
cn.cn_namelen = (int)strlen(tmpname);
/* Pick up uid, gid, mode and date from original file. */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_uid);
VATTR_WANTED(&va, va_gid);
VATTR_WANTED(&va, va_mode);
VATTR_WANTED(&va, va_create_time);
VATTR_WANTED(&va, va_modify_time);
if (VNOP_GETATTR(vp, &va, context) != 0 ||
!VATTR_IS_SUPPORTED(&va, va_uid) ||
!VATTR_IS_SUPPORTED(&va, va_gid) ||
!VATTR_IS_SUPPORTED(&va, va_mode)) {
va.va_uid = KAUTH_UID_NONE;
va.va_gid = KAUTH_GID_NONE;
va.va_mode = S_IRUSR | S_IWUSR;
}
va.va_vaflags = VA_EXCLUSIVE;
VATTR_SET(&va, va_type, VREG);
/* We no longer change the access, but we still hide it. */
VATTR_SET(&va, va_flags, UF_HIDDEN);
/* Obtain the vnode for the shadow files directory. */
if (get_shadow_dir(&dvp) != 0) {
error = ENOTDIR;
goto out;
}
if (!makestream) {
/* See if someone else already has it open. */
if (VNOP_LOOKUP(dvp, &svp, &cn, kernelctx) == 0) {
/* Double check existence by asking for size. */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_data_size);
if (VNOP_GETATTR(svp, &va, context) == 0 &&
VATTR_IS_SUPPORTED(&va, va_data_size)) {
goto out; /* OK to use. */
}
}
/*
* Otherwise make sure the resource fork data exists.
* Use the supplied context for accessing the AD file.
*/
error = vn_getxattr(vp, XATTR_RESOURCEFORK_NAME, NULL, &datasize,
XATTR_NOSECURITY, context);
/*
* To maintain binary compatibility with legacy Carbon
* emulated resource fork support, if the resource fork
* doesn't exist but the Finder Info does, then act as
* if an empty resource fork is present (see 4724359).
*/
if ((error == ENOATTR) &&
(vn_getxattr(vp, XATTR_FINDERINFO_NAME, NULL, &datasize,
XATTR_NOSECURITY, context) == 0)) {
datasize = 0;
error = 0;
} else {
if (error) {
goto out;
}
/* If the resource fork exists, its size is expected to be non-zero. */
if (datasize == 0) {
error = ENOATTR;
goto out;
}
}
}
/* Create the shadow stream file. */
error = VNOP_CREATE(dvp, &svp, &cn, &va, kernelctx);
if (error == 0) {
vnode_recycle(svp);
*creator = 1;
} else if ((error == EEXIST) && !makestream) {
error = VNOP_LOOKUP(dvp, &svp, &cn, kernelctx);
} else if ((error == ENOENT) && !makestream) {
/*
* We could have raced with a rmdir on the shadow directory
* post-lookup. Retry from the beginning, 1x only, to
* try and see if we need to re-create the shadow directory
* in get_shadow_dir.
*/
if (retries == 0) {
retries++;
if (dvp) {
vnode_put(dvp);
dvp = NULLVP;
}
if (svp) {
vnode_put(svp);
svp = NULLVP;
}
goto retry_create;
}
/* Otherwise, just error out normally below */
}
out:
if (dvp) {
vnode_put(dvp);
}
if (error) {
/* On errors, clean up shadow stream file. */
if (svp) {
vnode_put(svp);
svp = NULLVP;
}
}
*svpp = svp;
if (rsrcsize) {
*rsrcsize = datasize;
}
return error;
}
static int
default_getnamedstream(vnode_t vp, vnode_t *svpp, const char *name, enum nsoperation op, vfs_context_t context)
{
vnode_t svp = NULLVP;
uio_t auio = NULL;
caddr_t bufptr = NULL;
size_t bufsize = 0;
size_t datasize = 0;
int creator;
int error;
/* need the kernel context for accessing the shadowfile */
vfs_context_t kernelctx = vfs_context_kernel();
/*
* Only the "com.apple.ResourceFork" stream is supported here.
*/
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
*svpp = NULLVP;
return ENOATTR;
}
retry:
/*
* Obtain a shadow file for the resource fork I/O.
*
* Need to pass along the supplied context so that getshadowfile
* can access the AD file as needed, using it.
*/
error = getshadowfile(vp, &svp, 0, &datasize, &creator, context);
if (error) {
*svpp = NULLVP;
return error;
}
/*
* The creator of the shadow file provides its file data,
* all other threads should wait until its ready. In order to
* prevent a deadlock during error codepaths, we need to check if the
* vnode is being created, or if it has failed out. Regardless of success or
* failure, we set the VISSHADOW bit on the vnode, so we check that
* if the vnode's flags don't have VISNAMEDSTREAM set. If it doesn't,
* then we can infer the creator isn't done yet. If it's there, but
* VISNAMEDSTREAM is not set, then we can infer it errored out and we should
* try again.
*/
if (!creator) {
vnode_lock(svp);
if (svp->v_flag & VISNAMEDSTREAM) {
/* data is ready, go use it */
vnode_unlock(svp);
goto out;
} else {
/* It's not ready, wait for it (sleep using v_parent as channel) */
if ((svp->v_flag & VISSHADOW)) {
/*
* No VISNAMEDSTREAM, but we did see VISSHADOW, indicating that the other
* thread is done with this vnode. Just unlock the vnode and try again
*/
vnode_unlock(svp);
} else {
/* Otherwise, sleep if the shadow file is not created yet */
msleep((caddr_t)&svp->v_parent, &svp->v_lock, PINOD | PDROP,
"getnamedstream", NULL);
}
vnode_put(svp);
svp = NULLVP;
goto retry;
}
}
/*
* Copy the real resource fork data into shadow stream file.
*/
if (op == NS_OPEN && datasize != 0) {
size_t offset;
size_t iosize;
iosize = bufsize = MIN(datasize, NS_IOBUFSIZE);
if (kmem_alloc(kernel_map, (vm_offset_t *)&bufptr, bufsize, VM_KERN_MEMORY_FILE)) {
error = ENOMEM;
goto out;
}
auio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
offset = 0;
/* open the shadow file */
error = VNOP_OPEN(svp, 0, kernelctx);
if (error) {
goto out;
}
while (offset < datasize) {
size_t tmpsize;
iosize = MIN(datasize - offset, iosize);
uio_reset(auio, offset, UIO_SYSSPACE, UIO_READ);
uio_addiov(auio, (uintptr_t)bufptr, iosize);
/* use supplied ctx for AD file */
error = vn_getxattr(vp, XATTR_RESOURCEFORK_NAME, auio, &tmpsize,
XATTR_NOSECURITY, context);
if (error) {
break;
}
uio_reset(auio, offset, UIO_SYSSPACE, UIO_WRITE);
uio_addiov(auio, (uintptr_t)bufptr, iosize);
/* kernel context for writing shadowfile */
error = VNOP_WRITE(svp, auio, 0, kernelctx);
if (error) {
break;
}
offset += iosize;
}
/* close shadow file */
(void) VNOP_CLOSE(svp, 0, kernelctx);
}
out:
/* Wake up anyone waiting for svp file content */
if (creator) {
if (error == 0) {
vnode_lock(svp);
/* VISSHADOW would be set later on anyway, so we set it now */
svp->v_flag |= (VISNAMEDSTREAM | VISSHADOW);
wakeup((caddr_t)&svp->v_parent);
vnode_unlock(svp);
} else {
/* On post create errors, get rid of the shadow file. This
* way if there is another process waiting for initialization
* of the shadowfile by the current process will wake up and
* retry by creating and initializing the shadow file again.
* Also add the VISSHADOW bit here to indicate we're done operating
* on this vnode.
*/
(void)vnode_relenamedstream(vp, svp);
vnode_lock(svp);
svp->v_flag |= VISSHADOW;
wakeup((caddr_t)&svp->v_parent);
vnode_unlock(svp);
}
}
if (bufptr) {
kmem_free(kernel_map, (vm_offset_t)bufptr, bufsize);
}
if (auio) {
uio_free(auio);
}
if (error) {
/* On errors, clean up shadow stream file. */
if (svp) {
vnode_put(svp);
svp = NULLVP;
}
}
*svpp = svp;
return error;
}
static int
default_makenamedstream(vnode_t vp, vnode_t *svpp, const char *name, vfs_context_t context)
{
int creator;
int error;
/*
* Only the "com.apple.ResourceFork" stream is supported here.
*/
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
*svpp = NULLVP;
return ENOATTR;
}
/* Supply the context to getshadowfile so it can manipulate the AD file */
error = getshadowfile(vp, svpp, 1, NULL, &creator, context);
/*
* Wake up any waiters over in default_getnamedstream().
*/
if ((error == 0) && (*svpp != NULL) && creator) {
vnode_t svp = *svpp;
vnode_lock(svp);
/* If we're the creator, mark it as a named stream */
svp->v_flag |= (VISNAMEDSTREAM | VISSHADOW);
/* Wakeup any waiters on the v_parent channel */
wakeup((caddr_t)&svp->v_parent);
vnode_unlock(svp);
}
return error;
}
static int
default_removenamedstream(vnode_t vp, const char *name, vfs_context_t context)
{
/*
* Only the "com.apple.ResourceFork" stream is supported here.
*/
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) != 0) {
return ENOATTR;
}
/*
* XXX - what about other opened instances?
*/
return default_removexattr(vp, XATTR_RESOURCEFORK_NAME, 0, context);
}
static int
get_shadow_dir(vnode_t *sdvpp)
{
vnode_t dvp = NULLVP;
vnode_t sdvp = NULLVP;
struct componentname cn;
struct vnode_attr va;
char tmpname[80];
uint32_t tmp_fsid;
int error;
vfs_context_t kernelctx = vfs_context_kernel();
bzero(tmpname, sizeof(tmpname));
MAKE_SHADOW_DIRNAME(rootvnode, tmpname);
/*
* Look up the shadow directory to ensure that it still exists.
* By looking it up, we get an iocounted dvp to use, and avoid some coherency issues
* in caching it when multiple threads may be trying to manipulate the pointers.
*
* Make sure to use the kernel context. We want a singular view of
* the shadow dir regardless of chrooted processes.
*/
error = vnode_lookup(tmpname, 0, &sdvp, kernelctx);
if (error == 0) {
/*
* If we get here, then we have successfully looked up the shadow dir,
* and it has an iocount from the lookup. Return the vp in the output argument.
*/
*sdvpp = sdvp;
return 0;
}
/* In the failure case, no iocount is acquired */
sdvp = NULLVP;
bzero(tmpname, sizeof(tmpname));
/*
* Obtain the vnode for "/var/run" directory using the kernel
* context.
*
* This is defined in the SHADOW_DIR_CONTAINER macro
*/
if (vnode_lookup(SHADOW_DIR_CONTAINER, 0, &dvp, kernelctx) != 0) {
error = ENOTSUP;
goto out;
}
/*
* Create the shadow stream directory.
* 'dvp' below suggests the parent directory so
* we only need to provide the leaf entry name
*/
MAKE_SHADOW_DIR_LEAF(rootvnode, tmpname);
bzero(&cn, sizeof(cn));
cn.cn_nameiop = LOOKUP;
cn.cn_flags = ISLASTCN;
cn.cn_context = kernelctx;
cn.cn_pnbuf = tmpname;
cn.cn_pnlen = sizeof(tmpname);
cn.cn_nameptr = cn.cn_pnbuf;
cn.cn_namelen = (int)strlen(tmpname);
/*
* owned by root, only readable by root, hidden
*/
VATTR_INIT(&va);
VATTR_SET(&va, va_uid, 0);
VATTR_SET(&va, va_gid, 0);
VATTR_SET(&va, va_mode, S_IRUSR | S_IXUSR);
VATTR_SET(&va, va_type, VDIR);
VATTR_SET(&va, va_flags, UF_HIDDEN);
va.va_vaflags = VA_EXCLUSIVE;
error = VNOP_MKDIR(dvp, &sdvp, &cn, &va, kernelctx);
/*
* There can be only one winner for an exclusive create.
*/
if (error == EEXIST) {
/* loser has to look up directory */
error = VNOP_LOOKUP(dvp, &sdvp, &cn, kernelctx);
if (error == 0) {
/* Make sure its in fact a directory */
if (sdvp->v_type != VDIR) {
goto baddir;
}
/* Obtain the fsid for /var/run directory */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_fsid);
if (VNOP_GETATTR(dvp, &va, kernelctx) != 0 ||
!VATTR_IS_SUPPORTED(&va, va_fsid)) {
goto baddir;
}
tmp_fsid = va.va_fsid;
VATTR_INIT(&va);
VATTR_WANTED(&va, va_uid);
VATTR_WANTED(&va, va_gid);
VATTR_WANTED(&va, va_mode);
VATTR_WANTED(&va, va_fsid);
VATTR_WANTED(&va, va_dirlinkcount);
VATTR_WANTED(&va, va_acl);
/* Provide defaults for attrs that may not be supported */
va.va_dirlinkcount = 1;
va.va_acl = (kauth_acl_t) KAUTH_FILESEC_NONE;
if (VNOP_GETATTR(sdvp, &va, kernelctx) != 0 ||
!VATTR_IS_SUPPORTED(&va, va_uid) ||
!VATTR_IS_SUPPORTED(&va, va_gid) ||
!VATTR_IS_SUPPORTED(&va, va_mode) ||
!VATTR_IS_SUPPORTED(&va, va_fsid)) {
goto baddir;
}
/*
* Make sure its what we want:
* - owned by root
* - not writable by anyone
* - on same file system as /var/run
* - not a hard-linked directory
* - no ACLs (they might grant write access)
*/
if ((va.va_uid != 0) || (va.va_gid != 0) ||
(va.va_mode & (S_IWUSR | S_IRWXG | S_IRWXO)) ||
(va.va_fsid != tmp_fsid) ||
(va.va_dirlinkcount != 1) ||
(va.va_acl != (kauth_acl_t) KAUTH_FILESEC_NONE)) {
goto baddir;
}
}
}
out:
if (dvp) {
vnode_put(dvp);
}
if (error) {
/* On errors, clean up shadow stream directory. */
if (sdvp) {
vnode_put(sdvp);
sdvp = NULLVP;
}
}
*sdvpp = sdvp;
return error;
baddir:
/* This is not the dir we're looking for, move along */
++shadow_sequence; /* try something else next time */
error = ENOTDIR;
goto out;
}
#endif /* NAMEDSTREAMS */
#if CONFIG_APPLEDOUBLE
/*
* Default Implementation (Non-native EA)
*/
/*
* Typical "._" AppleDouble Header File layout:
* ------------------------------------------------------------
* MAGIC 0x00051607
* VERSION 0x00020000
* FILLER 0
* COUNT 2
* .-- AD ENTRY[0] Finder Info Entry (must be first)
* .--+-- AD ENTRY[1] Resource Fork Entry (must be last)
* | '-> FINDER INFO
* | ///////////// Fixed Size Data (32 bytes)
* | EXT ATTR HDR
* | /////////////
* | ATTR ENTRY[0] --.
* | ATTR ENTRY[1] --+--.
* | ATTR ENTRY[2] --+--+--.
* | ... | | |
* | ATTR ENTRY[N] --+--+--+--.
* | ATTR DATA 0 <-' | | |
* | //////////// | | |
* | ATTR DATA 1 <----' | |
* | ///////////// | |
* | ATTR DATA 2 <-------' |
* | ///////////// |
* | ... |
* | ATTR DATA N <----------'
* | /////////////
* | Attribute Free Space
* |
* '----> RESOURCE FORK
* ///////////// Variable Sized Data
* /////////////
* /////////////
* /////////////
* /////////////
* /////////////
* ...
* /////////////
*
* ------------------------------------------------------------
*
* NOTE: The EXT ATTR HDR, ATTR ENTRY's and ATTR DATA's are
* stored as part of the Finder Info. The length in the Finder
* Info AppleDouble entry includes the length of the extended
* attribute header, attribute entries, and attribute data.
*/
/*
* On Disk Data Structures
*
* Note: Motorola 68K alignment and big-endian.
*
* See RFC 1740 for additional information about the AppleDouble file format.
*
*/
#define ADH_MAGIC 0x00051607
#define ADH_VERSION 0x00020000
#define ADH_MACOSX "Mac OS X "
/*
* AppleDouble Entry ID's
*/
#define AD_DATA 1 /* Data fork */
#define AD_RESOURCE 2 /* Resource fork */
#define AD_REALNAME 3 /* File's name on home file system */
#define AD_COMMENT 4 /* Standard Mac comment */
#define AD_ICONBW 5 /* Mac black & white icon */
#define AD_ICONCOLOR 6 /* Mac color icon */
#define AD_UNUSED 7 /* Not used */
#define AD_FILEDATES 8 /* File dates; create, modify, etc */
#define AD_FINDERINFO 9 /* Mac Finder info & extended info */
#define AD_MACINFO 10 /* Mac file info, attributes, etc */
#define AD_PRODOSINFO 11 /* Pro-DOS file info, attrib., etc */
#define AD_MSDOSINFO 12 /* MS-DOS file info, attributes, etc */
#define AD_AFPNAME 13 /* Short name on AFP server */
#define AD_AFPINFO 14 /* AFP file info, attrib., etc */
#define AD_AFPDIRID 15 /* AFP directory ID */
#define AD_ATTRIBUTES AD_FINDERINFO
#define ATTR_FILE_PREFIX "._"
#define ATTR_HDR_MAGIC 0x41545452 /* 'ATTR' */
#define ATTR_BUF_SIZE 4096 /* default size of the attr file and how much we'll grow by */
/* Implementation Limits */
#define ATTR_MAX_SIZE AD_XATTR_MAXSIZE
#define ATTR_MAX_HDR_SIZE 65536
/*
* Note: ATTR_MAX_HDR_SIZE is the largest attribute header
* size supported (including the attribute entries). All of
* the attribute entries must reside within this limit. If
* any of the attribute data crosses the ATTR_MAX_HDR_SIZE
* boundry, then all of the attribute data I/O is performed
* separately from the attribute header I/O.
*
* In particular, all of the attr_entry structures must lie
* completely within the first ATTR_MAX_HDR_SIZE bytes of the
* AppleDouble file. However, the attribute data (i.e. the
* contents of the extended attributes) may extend beyond the
* first ATTR_MAX_HDR_SIZE bytes of the file. Note that this
* limit is to allow the implementation to optimize by reading
* the first ATTR_MAX_HDR_SIZE bytes of the file.
*/
#define FINDERINFOSIZE 32
typedef struct apple_double_entry {
u_int32_t type; /* entry type: see list, 0 invalid */
u_int32_t offset; /* entry data offset from the beginning of the file. */
u_int32_t length; /* entry data length in bytes. */
} __attribute__((aligned(2), packed)) apple_double_entry_t;
typedef struct apple_double_header {
u_int32_t magic; /* == ADH_MAGIC */
u_int32_t version; /* format version: 2 = 0x00020000 */
u_int32_t filler[4];
u_int16_t numEntries; /* number of entries which follow */
apple_double_entry_t entries[2]; /* 'finfo' & 'rsrc' always exist */
u_int8_t finfo[FINDERINFOSIZE]; /* Must start with Finder Info (32 bytes) */
u_int8_t pad[2]; /* get better alignment inside attr_header */
} __attribute__((aligned(2), packed)) apple_double_header_t;
#define ADHDRSIZE (4+4+16+2)
/* Entries are aligned on 4 byte boundaries */
typedef struct attr_entry {
u_int32_t offset; /* file offset to data */
u_int32_t length; /* size of attribute data */
u_int16_t flags;
u_int8_t namelen;
u_int8_t name[1]; /* NULL-terminated UTF-8 name (up to 128 bytes max) */
} __attribute__((aligned(2), packed)) attr_entry_t;
/* Header + entries must fit into 64K. Data may extend beyond 64K. */
typedef struct attr_header {
apple_double_header_t appledouble;
u_int32_t magic; /* == ATTR_HDR_MAGIC */
u_int32_t debug_tag; /* for debugging == file id of owning file */
u_int32_t total_size; /* file offset of end of attribute header + entries + data */
u_int32_t data_start; /* file offset to attribute data area */
u_int32_t data_length; /* length of attribute data area */
u_int32_t reserved[3];
u_int16_t flags;
u_int16_t num_attrs;
} __attribute__((aligned(2), packed)) attr_header_t;
/* Empty Resource Fork Header */
typedef struct rsrcfork_header {
u_int32_t fh_DataOffset;
u_int32_t fh_MapOffset;
u_int32_t fh_DataLength;
u_int32_t fh_MapLength;
u_int8_t systemData[112];
u_int8_t appData[128];
u_int32_t mh_DataOffset;
u_int32_t mh_MapOffset;
u_int32_t mh_DataLength;
u_int32_t mh_MapLength;
u_int32_t mh_Next;
u_int16_t mh_RefNum;
u_int8_t mh_Attr;
u_int8_t mh_InMemoryAttr;
u_int16_t mh_Types;
u_int16_t mh_Names;
u_int16_t typeCount;
} __attribute__((aligned(2), packed)) rsrcfork_header_t;
#define RF_FIRST_RESOURCE 256
#define RF_NULL_MAP_LENGTH 30
#define RF_EMPTY_TAG "This resource fork intentionally left blank "
/* Runtime information about the attribute file. */
typedef struct attr_info {
vfs_context_t context;
vnode_t filevp;
size_t filesize;
size_t iosize;
u_int8_t *rawdata;
size_t rawsize; /* minimum of filesize or ATTR_MAX_HDR_SIZE */
apple_double_header_t *filehdr;
apple_double_entry_t *finderinfo;
apple_double_entry_t *rsrcfork;
attr_header_t *attrhdr;
attr_entry_t *attr_entry;
u_int8_t readonly;
u_int8_t emptyfinderinfo;
} attr_info_t;
#define ATTR_SETTING 1
#define ATTR_ALIGN 3L /* Use four-byte alignment */
#define ATTR_ENTRY_LENGTH(namelen) \
((sizeof(attr_entry_t) - 1 + (namelen) + ATTR_ALIGN) & (~ATTR_ALIGN))
#define ATTR_NEXT(ae) \
(attr_entry_t *)((u_int8_t *)(ae) + ATTR_ENTRY_LENGTH((ae)->namelen))
#define ATTR_VALID(ae, ai) \
((&(ae)->namelen < ((ai).rawdata + (ai).rawsize)) && \
(u_int8_t *)ATTR_NEXT(ae) <= ((ai).rawdata + (ai).rawsize))
#define SWAP16(x) OSSwapBigToHostInt16((x))
#define SWAP32(x) OSSwapBigToHostInt32((x))
#define SWAP64(x) OSSwapBigToHostInt64((x))
static u_int32_t emptyfinfo[8] = {0};
/*
* Local support routines
*/
static void close_xattrfile(vnode_t xvp, int fileflags, vfs_context_t context);
static int open_xattrfile(vnode_t vp, int fileflags, vnode_t *xvpp, vfs_context_t context);
static int create_xattrfile(vnode_t xvp, u_int32_t fileid, vfs_context_t context);
static int remove_xattrfile(vnode_t xvp, vfs_context_t context);
static int get_xattrinfo(vnode_t xvp, int setting, attr_info_t *ainfop, vfs_context_t context);
static void rel_xattrinfo(attr_info_t *ainfop);
static int write_xattrinfo(attr_info_t *ainfop);
static void init_empty_resource_fork(rsrcfork_header_t * rsrcforkhdr);
static int lock_xattrfile(vnode_t xvp, short locktype, vfs_context_t context);
static int unlock_xattrfile(vnode_t xvp, vfs_context_t context);
#if BYTE_ORDER == LITTLE_ENDIAN
static void swap_adhdr(apple_double_header_t *adh);
static void swap_attrhdr(attr_header_t *ah, attr_info_t* info);
#else
#define swap_adhdr(x)
#define swap_attrhdr(x, y)
#endif
static int check_and_swap_attrhdr(attr_header_t *ah, attr_info_t* ainfop);
static int shift_data_down(vnode_t xvp, off_t start, size_t len, off_t delta, vfs_context_t context);
static int shift_data_up(vnode_t xvp, off_t start, size_t len, off_t delta, vfs_context_t context);
/*
* Sanity check and swap the header of an AppleDouble file. Assumes the buffer
* is in big endian (as it would exist on disk). Verifies the following:
* - magic field
* - version field
* - number of entries
* - that each entry fits within the file size
*
* If the header is invalid, ENOATTR is returned.
*
* NOTE: Does not attempt to validate the extended attributes header that
* may be embedded in the Finder Info entry.
*/
static int
check_and_swap_apple_double_header(attr_info_t *ainfop)
{
int i, j;
u_int32_t header_end;
u_int32_t entry_end;
size_t rawsize;
apple_double_header_t *header;
rawsize = ainfop->rawsize;
header = (apple_double_header_t *) ainfop->rawdata;
/* Is the file big enough to contain an AppleDouble header? */
if (rawsize < offsetof(apple_double_header_t, entries)) {
return ENOATTR;
}
/* Swap the AppleDouble header fields to native order */
header->magic = SWAP32(header->magic);
header->version = SWAP32(header->version);
header->numEntries = SWAP16(header->numEntries);
/* Sanity check the AppleDouble header fields */
if (header->magic != ADH_MAGIC ||
header->version != ADH_VERSION ||
header->numEntries < 1 ||
header->numEntries > 15) {
return ENOATTR;
}
/* Calculate where the entries[] array ends */
header_end = offsetof(apple_double_header_t, entries) +
header->numEntries * sizeof(apple_double_entry_t);
/* Is the file big enough to contain the AppleDouble entries? */
if (rawsize < header_end) {
return ENOATTR;
}
/* Swap and sanity check each AppleDouble entry */
for (i = 0; i < header->numEntries; i++) {
/* Swap the per-entry fields to native order */
header->entries[i].type = SWAP32(header->entries[i].type);
header->entries[i].offset = SWAP32(header->entries[i].offset);
header->entries[i].length = SWAP32(header->entries[i].length);
entry_end = header->entries[i].offset + header->entries[i].length;
/*
* Does the entry's content start within the header itself,
* did the addition overflow, or does the entry's content
* extend past the end of the file?
*/
if (header->entries[i].offset < header_end ||
entry_end < header->entries[i].offset ||
entry_end > ainfop->filesize) {
return ENOATTR;
}
/*
* Does the current entry's content overlap with a previous
* entry's content?
*
* Yes, this is O(N**2), and there are more efficient algorithms
* for testing pairwise overlap of N ranges when N is large.
* But we have already ensured N < 16, and N is almost always 2.
* So there's no point in using a more complex algorithm.
*/
for (j = 0; j < i; j++) {
if (entry_end > header->entries[j].offset &&
header->entries[j].offset + header->entries[j].length > header->entries[i].offset) {
return ENOATTR;
}
}
}
return 0;
}
/*
* Retrieve the data of an extended attribute.
*/
static int
default_getxattr(vnode_t vp, const char *name, uio_t uio, size_t *size,
__unused int options, vfs_context_t context)
{
vnode_t xvp = NULL;
attr_info_t ainfo;
attr_header_t *header;
attr_entry_t *entry;
u_int8_t *attrdata;
u_int32_t datalen;
size_t namelen;
int isrsrcfork;
int fileflags;
int i;
int error;
fileflags = FREAD;
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) == 0) {
isrsrcfork = 1;
/*
* Open the file locked (shared) since the Carbon
* File Manager may have the Apple Double file open
* and could be changing the resource fork.
*/
fileflags |= O_SHLOCK;
} else {
isrsrcfork = 0;
}
if ((error = open_xattrfile(vp, fileflags, &xvp, context))) {
return error;
}
if ((error = get_xattrinfo(xvp, 0, &ainfo, context))) {
close_xattrfile(xvp, fileflags, context);
return error;
}
/* Get the Finder Info. */
if (strncmp(name, XATTR_FINDERINFO_NAME, sizeof(XATTR_FINDERINFO_NAME)) == 0) {
if (ainfo.finderinfo == NULL || ainfo.emptyfinderinfo) {
error = ENOATTR;
} else if (uio == NULL) {
*size = FINDERINFOSIZE;
error = 0;
} else if (uio_offset(uio) != 0) {
error = EINVAL;
} else if (uio_resid(uio) < FINDERINFOSIZE) {
error = ERANGE;
} else {
attrdata = (u_int8_t*)ainfo.filehdr + ainfo.finderinfo->offset;
error = uiomove((caddr_t)attrdata, FINDERINFOSIZE, uio);
}
goto out;
}
/* Read the Resource Fork. */
if (isrsrcfork) {
if (!vnode_isreg(vp)) {
error = EPERM;
} else if (ainfo.rsrcfork == NULL) {
error = ENOATTR;
} else if (uio == NULL) {
*size = (size_t)ainfo.rsrcfork->length;
} else {
uio_setoffset(uio, uio_offset(uio) + ainfo.rsrcfork->offset);
error = VNOP_READ(xvp, uio, 0, context);
if (error == 0) {
uio_setoffset(uio, uio_offset(uio) - ainfo.rsrcfork->offset);
}
}
goto out;
}
if (ainfo.attrhdr == NULL || ainfo.attr_entry == NULL) {
error = ENOATTR;
goto out;
}
if (uio_offset(uio) != 0) {
error = EINVAL;
goto out;
}
error = ENOATTR;
namelen = strlen(name) + 1;
header = ainfo.attrhdr;
entry = ainfo.attr_entry;
/*
* Search for attribute name in the header.
*/
for (i = 0; i < header->num_attrs && ATTR_VALID(entry, ainfo); i++) {
if (strncmp((const char *)entry->name, name, namelen) == 0) {
datalen = entry->length;
if (uio == NULL) {
*size = datalen;
error = 0;
break;
}
if (uio_resid(uio) < (user_ssize_t)datalen) {
error = ERANGE;
break;
}
if (entry->offset + datalen < ATTR_MAX_HDR_SIZE) {
attrdata = ((u_int8_t *)header + entry->offset);
error = uiomove((caddr_t)attrdata, datalen, uio);
} else {
uio_setoffset(uio, entry->offset);
error = VNOP_READ(xvp, uio, 0, context);
uio_setoffset(uio, 0);
}
break;
}
entry = ATTR_NEXT(entry);
}
out:
rel_xattrinfo(&ainfo);
close_xattrfile(xvp, fileflags, context);
return error;
}
/*
* Set the data of an extended attribute.
*/
static int
default_setxattr(vnode_t vp, const char *name, uio_t uio, int options, vfs_context_t context)
{
vnode_t xvp = NULL;
attr_info_t ainfo;
attr_header_t *header;
attr_entry_t *entry;
attr_entry_t *lastentry;
u_int8_t *attrdata;
size_t datalen;
size_t entrylen;
size_t datafreespace;
int namelen;
int found = 0;
int i;
int splitdata;
int fileflags;
int error;
char finfo[FINDERINFOSIZE];
datalen = uio_resid(uio);
if (datalen > XATTR_MAXSIZE) {
return EINVAL;
}
namelen = (int)strlen(name) + 1;
if (namelen > UINT8_MAX) {
return EINVAL;
}
entrylen = ATTR_ENTRY_LENGTH(namelen);
/*
* By convention, Finder Info that is all zeroes is equivalent to not
* having a Finder Info EA. So if we're trying to set the Finder Info
* to all zeroes, then delete it instead. If a file didn't have an
* AppleDouble file before, this prevents creating an AppleDouble file
* with no useful content.
*
* If neither XATTR_CREATE nor XATTR_REPLACE were specified, we check
* for all zeroes Finder Info before opening the AppleDouble file.
* But if either of those options were specified, we need to open the
* AppleDouble file to see whether there was already Finder Info (so we
* can return an error if needed); this case is handled further below.
*
* NOTE: this copies the Finder Info data into the "finfo" local.
*/
if (strncmp(name, XATTR_FINDERINFO_NAME, sizeof(XATTR_FINDERINFO_NAME)) == 0) {
/*
* TODO: check the XATTR_CREATE and XATTR_REPLACE flags.
* That means we probably have to open_xattrfile and get_xattrinfo.
*/
if (uio_offset(uio) != 0 || datalen != FINDERINFOSIZE) {
return EINVAL;
}
error = uiomove(finfo, (int)datalen, uio);
if (error) {
return error;
}
if ((options & (XATTR_CREATE | XATTR_REPLACE)) == 0 &&
bcmp(finfo, emptyfinfo, FINDERINFOSIZE) == 0) {
error = default_removexattr(vp, name, 0, context);
if (error == ENOATTR) {
error = 0;
}
return error;
}
}
start:
/*
* Open the file locked since setting an attribute
* can change the layout of the Apple Double file.
*/
fileflags = FREAD | FWRITE | O_EXLOCK;
if ((error = open_xattrfile(vp, O_CREAT | fileflags, &xvp, context))) {
return error;
}
if ((error = get_xattrinfo(xvp, ATTR_SETTING, &ainfo, context))) {
close_xattrfile(xvp, fileflags, context);
return error;
}
/* Set the Finder Info. */
if (strncmp(name, XATTR_FINDERINFO_NAME, sizeof(XATTR_FINDERINFO_NAME)) == 0) {
if (ainfo.finderinfo && !ainfo.emptyfinderinfo) {
/* attr exists and "create" was specified? */
if (options & XATTR_CREATE) {
error = EEXIST;
goto out;
}
} else {
/* attr doesn't exists and "replace" was specified? */
if (options & XATTR_REPLACE) {
error = ENOATTR;
goto out;
}
}
if (options != 0 && bcmp(finfo, emptyfinfo, FINDERINFOSIZE) == 0) {
/*
* Setting the Finder Info to all zeroes is equivalent to
* removing it. Close the xattr file and let
* default_removexattr do the work (including deleting
* the xattr file if there are no other xattrs).
*
* Note that we have to handle the case where the
* Finder Info was already all zeroes, and we ignore
* ENOATTR.
*
* The common case where options == 0 was handled above.
*/
rel_xattrinfo(&ainfo);
close_xattrfile(xvp, fileflags, context);
error = default_removexattr(vp, name, 0, context);
if (error == ENOATTR) {
error = 0;
}
return error;
}
if (ainfo.finderinfo) {
attrdata = (u_int8_t *)ainfo.filehdr + ainfo.finderinfo->offset;
bcopy(finfo, attrdata, datalen);
ainfo.iosize = sizeof(attr_header_t);
error = write_xattrinfo(&ainfo);
goto out;
}
error = ENOATTR;
goto out;
}
/* Write the Resource Fork. */
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) == 0) {
off_t endoffset;
if (!vnode_isreg(vp)) {
error = EPERM;
goto out;
}
/* Make sure we have a rsrc fork pointer.. */
if (ainfo.rsrcfork == NULL) {
error = ENOATTR;
goto out;
}
if (ainfo.rsrcfork) {
if (ainfo.rsrcfork->length != 0) {
if (options & XATTR_CREATE) {
/* attr exists, and create specified ? */
error = EEXIST;
goto out;
}
} else {
/* Zero length AD rsrc fork */
if (options & XATTR_REPLACE) {
/* attr doesn't exist (0-length), but replace specified ? */
error = ENOATTR;
goto out;
}
}
} else {
/* We can't do much if we somehow didn't get an AD rsrc pointer */
error = ENOATTR;
goto out;
}
endoffset = uio_resid(uio) + uio_offset(uio); /* new size */
if (endoffset > UINT32_MAX || endoffset < 0) {
error = EINVAL;
goto out;
}
uio_setoffset(uio, uio_offset(uio) + ainfo.rsrcfork->offset);
error = VNOP_WRITE(xvp, uio, 0, context);
if (error) {
goto out;
}
uio_setoffset(uio, uio_offset(uio) - ainfo.rsrcfork->offset);
if (endoffset > ainfo.rsrcfork->length) {
ainfo.rsrcfork->length = (u_int32_t)endoffset;
ainfo.iosize = sizeof(attr_header_t);
error = write_xattrinfo(&ainfo);
goto out;
}
goto out;
}
if (datalen > ATTR_MAX_SIZE) {
return E2BIG; /* EINVAL instead ? */
}
if (ainfo.attrhdr == NULL) {
error = ENOATTR;
goto out;
}
header = ainfo.attrhdr;
entry = ainfo.attr_entry;
/* Check if data area crosses the maximum header size. */
if ((header->data_start + header->data_length + entrylen + datalen) > ATTR_MAX_HDR_SIZE) {
splitdata = 1; /* do data I/O separately */
} else {
splitdata = 0;
}
/*
* See if attribute already exists.
*/
for (i = 0; i < header->num_attrs && ATTR_VALID(entry, ainfo); i++) {
if (strncmp((const char *)entry->name, name, namelen) == 0) {
found = 1;
break;
}
entry = ATTR_NEXT(entry);
}
if (found) {
if (options & XATTR_CREATE) {
error = EEXIST;
goto out;
}
if (datalen == entry->length) {
if (splitdata) {
uio_setoffset(uio, entry->offset);
error = VNOP_WRITE(xvp, uio, 0, context);
uio_setoffset(uio, 0);
if (error) {
printf("setxattr: VNOP_WRITE error %d\n", error);
}
} else {
attrdata = (u_int8_t *)header + entry->offset;
error = uiomove((caddr_t)attrdata, (int)datalen, uio);
if (error) {
goto out;
}
ainfo.iosize = ainfo.attrhdr->data_start + ainfo.attrhdr->data_length;
error = write_xattrinfo(&ainfo);
if (error) {
printf("setxattr: write_xattrinfo error %d\n", error);
}
}
goto out;
} else {
/*
* Brute force approach - just remove old entry and set new entry.
*/
found = 0;
rel_xattrinfo(&ainfo);
close_xattrfile(xvp, fileflags, context);
error = default_removexattr(vp, name, options, context);
if (error) {
return error;
}
/* Clear XATTR_REPLACE option since we just removed the attribute. */
options &= ~XATTR_REPLACE;
goto start; /* start over */
}
} else {
if (!ATTR_VALID(entry, ainfo)) {
error = ENOSPC;
goto out;
}
}
if (options & XATTR_REPLACE) {
error = ENOATTR; /* nothing there to replace */
goto out;
}
/* Check if header size limit has been reached. */
if ((header->data_start + entrylen) > ATTR_MAX_HDR_SIZE) {
error = ENOSPC;
goto out;
}
datafreespace = header->total_size - (header->data_start + header->data_length);
/* Check if we need more space. */
if ((datalen + entrylen) > datafreespace) {
size_t growsize;
growsize = roundup((datalen + entrylen) - datafreespace, ATTR_BUF_SIZE);
/* Clip roundup size when we can still fit in ATTR_MAX_HDR_SIZE. */
if (!splitdata && (header->total_size + growsize) > ATTR_MAX_HDR_SIZE) {
growsize = ATTR_MAX_HDR_SIZE - header->total_size;
}
ainfo.filesize += growsize;
error = vnode_setsize(xvp, ainfo.filesize, 0, context);
if (error) {
printf("setxattr: VNOP_TRUNCATE error %d\n", error);
}
if (error) {
goto out;
}
/*
* Move the resource fork out of the way.
*/
if (ainfo.rsrcfork) {
if (ainfo.rsrcfork->length != 0) {
shift_data_down(xvp,
ainfo.rsrcfork->offset,
ainfo.rsrcfork->length,
growsize, context);
}
ainfo.rsrcfork->offset += growsize;
}
ainfo.finderinfo->length += growsize;
header->total_size += growsize;
}
/* Make space for a new entry. */
if (splitdata) {
shift_data_down(xvp,
header->data_start,
header->data_length,
entrylen, context);
} else {
bcopy((u_int8_t *)header + header->data_start,
(u_int8_t *)header + header->data_start + entrylen,
header->data_length);
}
header->data_start += entrylen;
/* Fix up entry data offsets. */
lastentry = entry;
for (entry = ainfo.attr_entry; entry != lastentry && ATTR_VALID(entry, ainfo); entry = ATTR_NEXT(entry)) {
entry->offset += entrylen;
}
/*
* If the attribute data area is entirely within
* the header buffer, then just update the buffer,
* otherwise we'll write it separately to the file.
*/
if (splitdata) {
off_t offset;
/* Write new attribute data after the end of existing data. */
offset = header->data_start + header->data_length;
uio_setoffset(uio, offset);
error = VNOP_WRITE(xvp, uio, 0, context);
uio_setoffset(uio, 0);
if (error) {
printf("setxattr: VNOP_WRITE error %d\n", error);
goto out;
}
} else {
attrdata = (u_int8_t *)header + header->data_start + header->data_length;
error = uiomove((caddr_t)attrdata, (int)datalen, uio);
if (error) {
printf("setxattr: uiomove error %d\n", error);
goto out;
}
}
/* Create the attribute entry. */
lastentry->length = (u_int32_t)datalen;
lastentry->offset = header->data_start + header->data_length;
lastentry->namelen = (u_int8_t)namelen;
lastentry->flags = 0;
bcopy(name, &lastentry->name[0], namelen);
/* Update the attributes header. */
header->num_attrs++;
header->data_length += datalen;
if (splitdata) {
/* Only write the entries, since the data was written separately. */
ainfo.iosize = ainfo.attrhdr->data_start;
} else {
/* The entry and data are both in the header; write them together. */
ainfo.iosize = ainfo.attrhdr->data_start + ainfo.attrhdr->data_length;
}
error = write_xattrinfo(&ainfo);
if (error) {
printf("setxattr: write_xattrinfo error %d\n", error);
}
out:
rel_xattrinfo(&ainfo);
close_xattrfile(xvp, fileflags, context);
/* Touch the change time if we changed an attribute. */
if (error == 0) {
struct vnode_attr va;
/* Re-write the mtime to cause a ctime change. */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_modify_time);
if (vnode_getattr(vp, &va, context) == 0) {
VATTR_INIT(&va);
VATTR_SET(&va, va_modify_time, va.va_modify_time);
(void) vnode_setattr(vp, &va, context);
}
}
post_event_if_success(vp, error, NOTE_ATTRIB);
return error;
}
/*
* Remove an extended attribute.
*/
static int
default_removexattr(vnode_t vp, const char *name, __unused int options, vfs_context_t context)
{
vnode_t xvp = NULL;
attr_info_t ainfo;
attr_header_t *header;
attr_entry_t *entry;
attr_entry_t *oldslot;
u_int8_t *attrdata;
u_int32_t dataoff;
size_t datalen;
size_t entrylen;
int namelen;
int found = 0, lastone = 0;
int i;
int splitdata;
int attrcount = 0;
int isrsrcfork;
int fileflags;
int error;
fileflags = FREAD | FWRITE;
if (strncmp(name, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME)) == 0) {
isrsrcfork = 1;
/*
* Open the file locked (exclusive) since the Carbon
* File Manager may have the Apple Double file open
* and could be changing the resource fork.
*/
fileflags |= O_EXLOCK;
} else {
isrsrcfork = 0;
}
if ((error = open_xattrfile(vp, fileflags, &xvp, context))) {
return error;
}
if ((error = get_xattrinfo(xvp, 0, &ainfo, context))) {
close_xattrfile(xvp, fileflags, context);
return error;
}
if (ainfo.attrhdr) {
attrcount += ainfo.attrhdr->num_attrs;
}
if (ainfo.rsrcfork) {
++attrcount;
}
if (ainfo.finderinfo && !ainfo.emptyfinderinfo) {
++attrcount;
}
/* Clear the Finder Info. */
if (strncmp(name, XATTR_FINDERINFO_NAME, sizeof(XATTR_FINDERINFO_NAME)) == 0) {
if (ainfo.finderinfo == NULL || ainfo.emptyfinderinfo) {
error = ENOATTR;
goto out;
}
/* On removal of last attribute the ._ file is removed. */
if (--attrcount == 0) {
goto out;
}
attrdata = (u_int8_t *)ainfo.filehdr + ainfo.finderinfo->offset;
bzero((caddr_t)attrdata, FINDERINFOSIZE);
error = write_xattrinfo(&ainfo);
goto out;
}
/* Clear the Resource Fork. */
if (isrsrcfork) {
if (!vnode_isreg(vp)) {
error = EPERM;
goto out;
}
if (ainfo.rsrcfork == NULL || ainfo.rsrcfork->length == 0) {
error = ENOATTR;
goto out;
}
/* On removal of last attribute the ._ file is removed. */
if (--attrcount == 0) {
goto out;
}
/*
* XXX
* If the resource fork isn't the last AppleDouble
* entry then the space needs to be reclaimed by
* shifting the entries after the resource fork.
*/
if ((ainfo.rsrcfork->offset + ainfo.rsrcfork->length) == ainfo.filesize) {
ainfo.filesize -= ainfo.rsrcfork->length;
error = vnode_setsize(xvp, ainfo.filesize, 0, context);
}
if (error == 0) {
ainfo.rsrcfork->length = 0;
ainfo.iosize = sizeof(attr_header_t);
error = write_xattrinfo(&ainfo);
}
goto out;
}
if (ainfo.attrhdr == NULL) {
error = ENOATTR;
goto out;
}
namelen = (int)strlen(name) + 1;
header = ainfo.attrhdr;
entry = ainfo.attr_entry;
/*
* See if this attribute exists.
*/
for (i = 0; i < header->num_attrs && ATTR_VALID(entry, ainfo); i++) {
if (strncmp((const char *)entry->name, name, namelen) == 0) {
found = 1;
if ((i + 1) == header->num_attrs) {
lastone = 1;
}
break;
}
entry = ATTR_NEXT(entry);
}
if (!found) {
error = ENOATTR;
goto out;
}
/* On removal of last attribute the ._ file is removed. */
if (--attrcount == 0) {
goto out;
}
datalen = entry->length;
dataoff = entry->offset;
entrylen = ATTR_ENTRY_LENGTH(namelen);
if ((header->data_start + header->data_length) > ATTR_MAX_HDR_SIZE) {
splitdata = 1;
} else {
splitdata = 0;
}
/* Remove the attribute entry. */
if (!lastone) {
bcopy((u_int8_t *)entry + entrylen, (u_int8_t *)entry,
((size_t)header + header->data_start) - ((size_t)entry + entrylen));
}
/* Adjust the attribute data. */
if (splitdata) {
shift_data_up(xvp,
header->data_start,
dataoff - header->data_start,
entrylen,
context);
if (!lastone) {
shift_data_up(xvp,
dataoff + datalen,
(header->data_start + header->data_length) - (dataoff + datalen),
datalen + entrylen,
context);
}
/* XXX write zeros to freed space ? */
ainfo.iosize = ainfo.attrhdr->data_start - entrylen;
} else {
bcopy((u_int8_t *)header + header->data_start,
(u_int8_t *)header + header->data_start - entrylen,
dataoff - header->data_start);
if (!lastone) {
bcopy((u_int8_t *)header + dataoff + datalen,
(u_int8_t *)header + dataoff - entrylen,
(header->data_start + header->data_length) - (dataoff + datalen));
}
bzero(((u_int8_t *)header + header->data_start + header->data_length) - (datalen + entrylen), (datalen + entrylen));
ainfo.iosize = ainfo.attrhdr->data_start + ainfo.attrhdr->data_length;
}
/* Adjust the header values and entry offsets. */
header->num_attrs--;
header->data_start -= entrylen;
header->data_length -= datalen;
oldslot = entry;
entry = ainfo.attr_entry;
for (i = 0; i < header->num_attrs && ATTR_VALID(entry, ainfo); i++) {
entry->offset -= entrylen;
if (entry >= oldslot) {
entry->offset -= datalen;
}
entry = ATTR_NEXT(entry);
}
error = write_xattrinfo(&ainfo);
if (error) {
printf("removexattr: write_xattrinfo error %d\n", error);
}
out:
rel_xattrinfo(&ainfo);
/* When there are no more attributes remove the ._ file. */
if (attrcount == 0) {
if (fileflags & O_EXLOCK) {
(void) unlock_xattrfile(xvp, context);
}
VNOP_CLOSE(xvp, fileflags, context);
vnode_rele(xvp);
error = remove_xattrfile(xvp, context);
vnode_put(xvp);
} else {
close_xattrfile(xvp, fileflags, context);
}
/* Touch the change time if we changed an attribute. */
if (error == 0) {
struct vnode_attr va;
/* Re-write the mtime to cause a ctime change. */
VATTR_INIT(&va);
VATTR_WANTED(&va, va_modify_time);
if (vnode_getattr(vp, &va, context) == 0) {
VATTR_INIT(&va);
VATTR_SET(&va, va_modify_time, va.va_modify_time);
(void) vnode_setattr(vp, &va, context);
}
}
post_event_if_success(vp, error, NOTE_ATTRIB);
return error;
}
/*
* Retrieve the list of extended attribute names.
*/
static int
default_listxattr(vnode_t vp, uio_t uio, size_t *size, __unused int options, vfs_context_t context)
{
vnode_t xvp = NULL;
attr_info_t ainfo;
attr_entry_t *entry;
int i, count;
int error;
/*
* We do not zero "*size" here as we don't want to stomp a size set when
* VNOP_LISTXATTR processed any native EAs. That size is initially zeroed by the
* system call layer, up in listxattr or flistxattr.
*/
if ((error = open_xattrfile(vp, FREAD, &xvp, context))) {
if (error == ENOATTR) {
error = 0;
}
return error;
}
if ((error = get_xattrinfo(xvp, 0, &ainfo, context))) {
if (error == ENOATTR) {
error = 0;
}
close_xattrfile(xvp, FREAD, context);
return error;
}
/* Check for Finder Info. */
if (ainfo.finderinfo && !ainfo.emptyfinderinfo) {
if (uio == NULL) {
*size += sizeof(XATTR_FINDERINFO_NAME);
} else if (uio_resid(uio) < (user_ssize_t)sizeof(XATTR_FINDERINFO_NAME)) {
error = ERANGE;
goto out;
} else {
error = uiomove(XATTR_FINDERINFO_NAME,
sizeof(XATTR_FINDERINFO_NAME), uio);
if (error) {
error = ERANGE;
goto out;
}
}
}
/* Check for Resource Fork. */
if (vnode_isreg(vp) && ainfo.rsrcfork) {
if (uio == NULL) {
*size += sizeof(XATTR_RESOURCEFORK_NAME);
} else if (uio_resid(uio) < (user_ssize_t)sizeof(XATTR_RESOURCEFORK_NAME)) {
error = ERANGE;
goto out;
} else {
error = uiomove(XATTR_RESOURCEFORK_NAME,
sizeof(XATTR_RESOURCEFORK_NAME), uio);
if (error) {
error = ERANGE;
goto out;
}
}
}
/* Check for attributes. */
if (ainfo.attrhdr) {
count = ainfo.attrhdr->num_attrs;
for (i = 0, entry = ainfo.attr_entry; i < count && ATTR_VALID(entry, ainfo); i++) {
if (xattr_protected((const char *)entry->name) ||
((entry->namelen < XATTR_MAXNAMELEN) &&
(entry->name[entry->namelen] == '\0') &&
(xattr_validatename((const char *)entry->name) != 0))) {
entry = ATTR_NEXT(entry);
continue;
}
if (uio == NULL) {
*size += entry->namelen;
entry = ATTR_NEXT(entry);
continue;
}
if (uio_resid(uio) < entry->namelen) {
error = ERANGE;
break;
}
error = uiomove((caddr_t) entry->name, entry->namelen, uio);
if (error) {
if (error != EFAULT) {
error = ERANGE;
}
break;
}
entry = ATTR_NEXT(entry);
}
}
out:
rel_xattrinfo(&ainfo);
close_xattrfile(xvp, FREAD, context);
return error;
}
static int
open_xattrfile(vnode_t vp, int fileflags, vnode_t *xvpp, vfs_context_t context)
{
vnode_t xvp = NULLVP;
vnode_t dvp = NULLVP;
struct vnode_attr *va = NULL;
struct nameidata *nd = NULL;
char smallname[64];
char *filename = NULL;
const char *basename = NULL;
size_t alloc_len;
size_t copy_len;
errno_t error;
int opened = 0;
int referenced = 0;
if (vnode_isvroot(vp) && vnode_isdir(vp)) {
/*
* For the root directory use "._." to hold the attributes.
*/
filename = &smallname[0];
snprintf(filename, sizeof(smallname), "%s%s", ATTR_FILE_PREFIX, ".");
dvp = vp; /* the "._." file resides in the root dir */
goto lookup;
}
if ((dvp = vnode_getparent(vp)) == NULLVP) {
error = ENOATTR;
goto out;
}
if ((basename = vnode_getname(vp)) == NULL) {
error = ENOATTR;
goto out;
}
/* "._" Attribute files cannot have attributes */
if (vp->v_type == VREG && strlen(basename) > 2 &&
basename[0] == '.' && basename[1] == '_') {
error = EPERM;
goto out;
}
filename = &smallname[0];
alloc_len = snprintf(filename, sizeof(smallname), "%s%s", ATTR_FILE_PREFIX, basename);
if (alloc_len >= sizeof(smallname)) {
alloc_len++; /* snprintf result doesn't include '\0' */
filename = kheap_alloc(KHEAP_TEMP, alloc_len, Z_WAITOK);
copy_len = snprintf(filename, alloc_len, "%s%s", ATTR_FILE_PREFIX, basename);
}
/*
* Note that the lookup here does not authorize. Since we are looking
* up in the same directory that we already have the file vnode in,
* we must have been given the file vnode legitimately. Read/write
* access has already been authorized in layers above for calls from
* userspace, and the authorization code using this path to read
* file security from the EA must always get access
*/
lookup:
nd = kheap_alloc(KHEAP_TEMP, sizeof(struct nameidata), Z_WAITOK);
NDINIT(nd, LOOKUP, OP_OPEN, LOCKLEAF | NOFOLLOW | USEDVP | DONOTAUTH,
UIO_SYSSPACE, CAST_USER_ADDR_T(filename), context);
nd->ni_dvp = dvp;
va = kheap_alloc(KHEAP_TEMP, sizeof(struct vnode_attr), Z_WAITOK);
if (fileflags & O_CREAT) {
nd->ni_cnd.cn_nameiop = CREATE;
#if CONFIG_TRIGGERS
nd->ni_op = OP_LINK;
#endif
if (dvp != vp) {
nd->ni_cnd.cn_flags |= LOCKPARENT;
}
if ((error = namei(nd))) {
nd->ni_dvp = NULLVP;
error = ENOATTR;
goto out;
}
if ((xvp = nd->ni_vp) == NULLVP) {
uid_t uid;
gid_t gid;
mode_t umode;
/*
* Pick up uid/gid/mode from target file.
*/
VATTR_INIT(va);
VATTR_WANTED(va, va_uid);
VATTR_WANTED(va, va_gid);
VATTR_WANTED(va, va_mode);
if (VNOP_GETATTR(vp, va, context) == 0 &&
VATTR_IS_SUPPORTED(va, va_uid) &&
VATTR_IS_SUPPORTED(va, va_gid) &&
VATTR_IS_SUPPORTED(va, va_mode)) {
uid = va->va_uid;
gid = va->va_gid;
umode = va->va_mode & (S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
} else { /* fallback values */
uid = KAUTH_UID_NONE;
gid = KAUTH_GID_NONE;
umode = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH;
}
VATTR_INIT(va);
VATTR_SET(va, va_type, VREG);
VATTR_SET(va, va_mode, umode);
if (uid != KAUTH_UID_NONE) {
VATTR_SET(va, va_uid, uid);
}
if (gid != KAUTH_GID_NONE) {
VATTR_SET(va, va_gid, gid);
}
error = vn_create(dvp, &nd->ni_vp, nd, va,
VN_CREATE_NOAUTH | VN_CREATE_NOINHERIT | VN_CREATE_NOLABEL,
0, NULL,
context);
if (error) {
error = ENOATTR;
} else {
xvp = nd->ni_vp;
}
}
nameidone(nd);
if (dvp != vp) {
vnode_put(dvp); /* drop iocount from LOCKPARENT request above */
}
if (error) {
goto out;
}
} else {
if ((error = namei(nd))) {
nd->ni_dvp = NULLVP;
error = ENOATTR;
goto out;
}
xvp = nd->ni_vp;
nameidone(nd);
}
nd->ni_dvp = NULLVP;
if (xvp->v_type != VREG) {
error = ENOATTR;
goto out;
}
/*
* Owners must match.
*/
VATTR_INIT(va);
VATTR_WANTED(va, va_uid);
if (VNOP_GETATTR(vp, va, context) == 0 && VATTR_IS_SUPPORTED(va, va_uid)) {
uid_t owner = va->va_uid;
VATTR_INIT(va);
VATTR_WANTED(va, va_uid);
if (VNOP_GETATTR(xvp, va, context) == 0 && (owner != va->va_uid)) {
error = ENOATTR; /* don't use this "._" file */
goto out;
}
}
if ((error = VNOP_OPEN(xvp, fileflags & ~(O_EXLOCK | O_SHLOCK), context))) {
error = ENOATTR;
goto out;
}
opened = 1;
if ((error = vnode_ref(xvp))) {
goto out;
}
referenced = 1;
/* If create was requested, make sure file header exists. */
if (fileflags & O_CREAT) {
VATTR_INIT(va);
VATTR_WANTED(va, va_data_size);
VATTR_WANTED(va, va_fileid);
VATTR_WANTED(va, va_nlink);
if ((error = vnode_getattr(xvp, va, context)) != 0) {
error = EPERM;
goto out;
}
/* If the file is empty then add a default header. */
if (va->va_data_size == 0) {
/* Don't adopt hard-linked "._" files. */
if (VATTR_IS_SUPPORTED(va, va_nlink) && va->va_nlink > 1) {
error = EPERM;
goto out;
}
if ((error = create_xattrfile(xvp, (u_int32_t)va->va_fileid, context))) {
goto out;
}
}
}
/* Apply file locking if requested. */
if (fileflags & (O_EXLOCK | O_SHLOCK)) {
short locktype;
locktype = (fileflags & O_EXLOCK) ? F_WRLCK : F_RDLCK;
error = lock_xattrfile(xvp, locktype, context);
if (error) {
error = ENOATTR;
}
}
out:
if (error) {
if (xvp != NULLVP) {
if (opened) {
(void) VNOP_CLOSE(xvp, fileflags, context);
}
if (fileflags & O_CREAT) {
/* Delete the xattr file if we encountered any errors */
(void) remove_xattrfile(xvp, context);
}
if (referenced) {
(void) vnode_rele(xvp);
}
(void) vnode_put(xvp);
xvp = NULLVP;
}
if ((error == ENOATTR) && (fileflags & O_CREAT)) {
error = EPERM;
}
}
/* Release resources after error-handling */
kheap_free(KHEAP_TEMP, nd, sizeof(struct nameidata));
kheap_free(KHEAP_TEMP, va, sizeof(struct vnode_attr));
if (dvp && (dvp != vp)) {
vnode_put(dvp);
}
if (basename) {
vnode_putname(basename);
}
if (filename && filename != &smallname[0]) {
kheap_free(KHEAP_TEMP, filename, alloc_len);
}
*xvpp = xvp; /* return a referenced vnode */
return error;
}
static void
close_xattrfile(vnode_t xvp, int fileflags, vfs_context_t context)
{
// if (fileflags & FWRITE)
// (void) VNOP_FSYNC(xvp, MNT_WAIT, context);
if (fileflags & (O_EXLOCK | O_SHLOCK)) {
(void) unlock_xattrfile(xvp, context);
}
(void) VNOP_CLOSE(xvp, fileflags, context);
(void) vnode_rele(xvp);
(void) vnode_put(xvp);
}
static int
remove_xattrfile(vnode_t xvp, vfs_context_t context)
{
vnode_t dvp;
struct nameidata nd;
char *path = NULL;
int pathlen;
int error = 0;
path = zalloc(ZV_NAMEI);
pathlen = MAXPATHLEN;
error = vn_getpath(xvp, path, &pathlen);
if (error) {
zfree(ZV_NAMEI, path);
return error;
}
NDINIT(&nd, DELETE, OP_UNLINK, LOCKPARENT | NOFOLLOW | DONOTAUTH,
UIO_SYSSPACE, CAST_USER_ADDR_T(path), context);
error = namei(&nd);
zfree(ZV_NAMEI, path);
if (error) {
return error;
}
dvp = nd.ni_dvp;
xvp = nd.ni_vp;
error = VNOP_REMOVE(dvp, xvp, &nd.ni_cnd, 0, context);
nameidone(&nd);
vnode_put(dvp);
vnode_put(xvp);
return error;
}
/*
* Read in and parse the AppleDouble header and entries, and the extended
* attribute header and entries if any. Populates the fields of ainfop
* based on the headers and entries found.
*
* The basic idea is to:
* - Read in up to ATTR_MAX_HDR_SIZE bytes of the start of the file. All
* AppleDouble entries, the extended attribute header, and extended
* attribute entries must lie within this part of the file; the rest of
* the AppleDouble handling code assumes this. Plus it allows us to
* somewhat optimize by doing a smaller number of larger I/Os.
* - Swap and sanity check the AppleDouble header (including the AppleDouble
* entries).
* - Find the Finder Info and Resource Fork entries, if any.
* - If we're going to be writing, try to make sure the Finder Info entry has
* room to store the extended attribute header, plus some space for extended
* attributes.
* - Swap and sanity check the extended attribute header and entries (if any).
*/
static int
get_xattrinfo(vnode_t xvp, int setting, attr_info_t *ainfop, vfs_context_t context)
{
uio_t auio = NULL;
void * buffer = NULL;
apple_double_header_t *filehdr;
struct vnode_attr va;
size_t iosize = 0;
int i;
int error;
bzero(ainfop, sizeof(attr_info_t));
ainfop->filevp = xvp;
ainfop->context = context;
VATTR_INIT(&va);
VATTR_WANTED(&va, va_data_size);
VATTR_WANTED(&va, va_fileid);
if ((error = vnode_getattr(xvp, &va, context))) {
goto bail;
}
ainfop->filesize = va.va_data_size;
/* When setting attributes, allow room for the header to grow. */
if (setting) {
iosize = ATTR_MAX_HDR_SIZE;
} else {
iosize = MIN(ATTR_MAX_HDR_SIZE, ainfop->filesize);
}
if (iosize == 0 || iosize < sizeof(apple_double_header_t)) {
error = ENOATTR;
goto bail;
}
ainfop->iosize = iosize;
buffer = kheap_alloc(KHEAP_DATA_BUFFERS, iosize, Z_WAITOK | Z_ZERO);
if (buffer == NULL) {
error = ENOMEM;
goto bail;
}
auio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
uio_addiov(auio, (uintptr_t)buffer, iosize);
/* Read the file header. */
error = VNOP_READ(xvp, auio, 0, context);
if (error) {
goto bail;
}
ainfop->rawsize = iosize - uio_resid(auio);
ainfop->rawdata = (u_int8_t *)buffer;
filehdr = (apple_double_header_t *)buffer;
error = check_and_swap_apple_double_header(ainfop);
if (error) {
goto bail;
}
ainfop->filehdr = filehdr; /* valid AppleDouble header */
/* rel_xattrinfo is responsible for freeing the header buffer */
buffer = NULL;
/* Find the Finder Info and Resource Fork entries, if any */
for (i = 0; i < filehdr->numEntries; ++i) {
if (filehdr->entries[i].type == AD_FINDERINFO &&
filehdr->entries[i].length >= FINDERINFOSIZE) {
/* We found the Finder Info entry. */
ainfop->finderinfo = &filehdr->entries[i];
/* At this point check_and_swap_apple_double_header() call above
* verified that all apple double entires are valid:
* they point somewhere within the file.
*
* Now for finderinfo make sure that the fixed portion
* is within the buffer we read in.
*/
if (((ainfop->finderinfo->offset + FINDERINFOSIZE) > ainfop->finderinfo->offset) &&
((ainfop->finderinfo->offset + FINDERINFOSIZE) <= ainfop->rawsize)) {
/*
* Is the Finder Info "empty" (all zeroes)? If so,
* we'll pretend like the Finder Info extended attribute
* does not exist.
*/
if (bcmp((u_int8_t*)ainfop->filehdr + ainfop->finderinfo->offset, emptyfinfo, sizeof(emptyfinfo)) == 0) {
ainfop->emptyfinderinfo = 1;
}
} else {
error = ENOATTR;
goto bail;
}
}
if (filehdr->entries[i].type == AD_RESOURCE) {
/*
* Ignore zero-length resource forks when getting. If setting,
* we need to remember the resource fork entry so it can be
* updated once the new content has been written.
*/
if (filehdr->entries[i].length == 0 && !setting) {
continue;
}
/*
* Check to see if any "empty" resource fork is ours (i.e. is ignorable).
*
* The "empty" resource headers we created have a system data tag of:
* "This resource fork intentionally left blank "
*/
if (filehdr->entries[i].length == sizeof(rsrcfork_header_t) && !setting) {
uio_t rf_uio;
u_int8_t systemData[64];
int rf_err;
/* Read the system data which starts at byte 16 */
rf_uio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
uio_addiov(rf_uio, (uintptr_t)systemData, sizeof(systemData));
uio_setoffset(rf_uio, filehdr->entries[i].offset + 16);
rf_err = VNOP_READ(xvp, rf_uio, 0, context);
uio_free(rf_uio);
if (rf_err != 0 ||
bcmp(systemData, RF_EMPTY_TAG, sizeof(RF_EMPTY_TAG)) == 0) {
continue; /* skip this resource fork */
}
}
ainfop->rsrcfork = &filehdr->entries[i];
if (i != (filehdr->numEntries - 1)) {
printf("get_xattrinfo: resource fork not last entry\n");
ainfop->readonly = 1;
}
continue;
}
}
/*
* See if this file looks like it is laid out correctly to contain
* extended attributes. If so, then do the following:
*
* - If we're going to be writing, try to make sure the Finder Info
* entry has room to store the extended attribute header, plus some
* space for extended attributes.
*
* - Swap and sanity check the extended attribute header and entries
* (if any).
*/
if (filehdr->numEntries == 2 &&
ainfop->finderinfo == &filehdr->entries[0] &&
ainfop->rsrcfork == &filehdr->entries[1] &&
ainfop->finderinfo->offset == offsetof(apple_double_header_t, finfo)) {
attr_header_t *attrhdr;
attrhdr = (attr_header_t *)filehdr;
/*
* If we're going to be writing, try to make sure the Finder
* Info entry has room to store the extended attribute header,
* plus some space for extended attributes.
*/
if (setting && ainfop->finderinfo->length == FINDERINFOSIZE) {
size_t delta;
size_t writesize;
delta = ATTR_BUF_SIZE - (filehdr->entries[0].offset + FINDERINFOSIZE);
if (ainfop->rsrcfork && filehdr->entries[1].length) {
/* Make some room before existing resource fork. */
shift_data_down(xvp,
filehdr->entries[1].offset,
filehdr->entries[1].length,
delta, context);
writesize = sizeof(attr_header_t);
} else {
/* Create a new, empty resource fork. */
rsrcfork_header_t *rsrcforkhdr;
vnode_setsize(xvp, filehdr->entries[1].offset + delta, 0, context);
/* Steal some space for an empty RF header. */
delta -= sizeof(rsrcfork_header_t);
bzero(&attrhdr->appledouble.pad[0], delta);
rsrcforkhdr = (rsrcfork_header_t *)((char *)filehdr + filehdr->entries[1].offset + delta);
/* Fill in Empty Resource Fork Header. */
init_empty_resource_fork(rsrcforkhdr);
filehdr->entries[1].length = sizeof(rsrcfork_header_t);
writesize = ATTR_BUF_SIZE;
}
filehdr->entries[0].length += delta;
filehdr->entries[1].offset += delta;
/* Fill in Attribute Header. */
attrhdr->magic = ATTR_HDR_MAGIC;
attrhdr->debug_tag = (u_int32_t)va.va_fileid;
attrhdr->total_size = filehdr->entries[1].offset;
attrhdr->data_start = sizeof(attr_header_t);
attrhdr->data_length = 0;
attrhdr->reserved[0] = 0;
attrhdr->reserved[1] = 0;
attrhdr->reserved[2] = 0;
attrhdr->flags = 0;
attrhdr->num_attrs = 0;
/* Push out new header */
uio_reset(auio, 0, UIO_SYSSPACE, UIO_WRITE);
uio_addiov(auio, (uintptr_t)filehdr, writesize);
swap_adhdr(filehdr); /* to big endian */
swap_attrhdr(attrhdr, ainfop); /* to big endian */
error = VNOP_WRITE(xvp, auio, 0, context);
swap_adhdr(filehdr); /* back to native */
/* The attribute header gets swapped below. */
}
}
/*
* Swap and sanity check the extended attribute header and
* entries (if any). The Finder Info content must be big enough
* to include the extended attribute header; if not, we just
* ignore it.
*
* Note that we're passing the offset + length (i.e. the end)
* of the Finder Info instead of rawsize to validate_attrhdr.
* This ensures that all extended attributes lie within the
* Finder Info content according to the AppleDouble entry.
*
* Sets ainfop->attrhdr and ainfop->attr_entry if a valid
* header was found.
*/
if (ainfop->finderinfo &&
ainfop->finderinfo == &filehdr->entries[0] &&
ainfop->finderinfo->length >= (sizeof(attr_header_t) - sizeof(apple_double_header_t))) {
attr_header_t *attrhdr = (attr_header_t*)filehdr;
if (ainfop->finderinfo->offset != offsetof(apple_double_header_t, finfo)) {
error = ENOATTR;
goto bail;
}
if ((error = check_and_swap_attrhdr(attrhdr, ainfop)) == 0) {
ainfop->attrhdr = attrhdr; /* valid attribute header */
/* First attr_entry starts immediately following attribute header */
ainfop->attr_entry = (attr_entry_t *)&attrhdr[1];
}
}
error = 0;
bail:
if (auio != NULL) {
uio_free(auio);
}
kheap_free(KHEAP_DATA_BUFFERS, buffer, iosize);
return error;
}
static int
create_xattrfile(vnode_t xvp, u_int32_t fileid, vfs_context_t context)
{
attr_header_t *xah;
rsrcfork_header_t *rsrcforkhdr;
void * buffer;
uio_t auio;
int rsrcforksize;
int error;
buffer = kheap_alloc(KHEAP_TEMP, ATTR_BUF_SIZE, Z_WAITOK);
bzero(buffer, ATTR_BUF_SIZE);
xah = (attr_header_t *)buffer;
auio = uio_create(1, 0, UIO_SYSSPACE, UIO_WRITE);
uio_addiov(auio, (uintptr_t)buffer, ATTR_BUF_SIZE);
rsrcforksize = sizeof(rsrcfork_header_t);
rsrcforkhdr = (rsrcfork_header_t *) ((char *)buffer + ATTR_BUF_SIZE - rsrcforksize);
/* Fill in Apple Double Header. */
xah->appledouble.magic = SWAP32(ADH_MAGIC);
xah->appledouble.version = SWAP32(ADH_VERSION);
xah->appledouble.numEntries = SWAP16(2);
xah->appledouble.entries[0].type = SWAP32(AD_FINDERINFO);
xah->appledouble.entries[0].offset = SWAP32(offsetof(apple_double_header_t, finfo));
xah->appledouble.entries[0].length = SWAP32(ATTR_BUF_SIZE - offsetof(apple_double_header_t, finfo) - rsrcforksize);
xah->appledouble.entries[1].type = SWAP32(AD_RESOURCE);
xah->appledouble.entries[1].offset = SWAP32(ATTR_BUF_SIZE - rsrcforksize);
xah->appledouble.entries[1].length = SWAP32(rsrcforksize);
bcopy(ADH_MACOSX, xah->appledouble.filler, sizeof(xah->appledouble.filler));
/* Fill in Attribute Header. */
xah->magic = SWAP32(ATTR_HDR_MAGIC);
xah->debug_tag = SWAP32(fileid);
xah->total_size = SWAP32(ATTR_BUF_SIZE - rsrcforksize);
xah->data_start = SWAP32(sizeof(attr_header_t));
/* Fill in Empty Resource Fork Header. */
init_empty_resource_fork(rsrcforkhdr);
/* Push it out. */
error = VNOP_WRITE(xvp, auio, IO_UNIT, context);
/* Did we write out the full uio? */
if (uio_resid(auio) > 0) {
error = ENOSPC;
}
uio_free(auio);
kheap_free(KHEAP_TEMP, buffer, ATTR_BUF_SIZE);
return error;
}
static void
init_empty_resource_fork(rsrcfork_header_t * rsrcforkhdr)
{
bzero(rsrcforkhdr, sizeof(rsrcfork_header_t));
rsrcforkhdr->fh_DataOffset = SWAP32(RF_FIRST_RESOURCE);
rsrcforkhdr->fh_MapOffset = SWAP32(RF_FIRST_RESOURCE);
rsrcforkhdr->fh_MapLength = SWAP32(RF_NULL_MAP_LENGTH);
rsrcforkhdr->mh_DataOffset = SWAP32(RF_FIRST_RESOURCE);
rsrcforkhdr->mh_MapOffset = SWAP32(RF_FIRST_RESOURCE);
rsrcforkhdr->mh_MapLength = SWAP32(RF_NULL_MAP_LENGTH);
rsrcforkhdr->mh_Types = SWAP16(RF_NULL_MAP_LENGTH - 2 );
rsrcforkhdr->mh_Names = SWAP16(RF_NULL_MAP_LENGTH);
rsrcforkhdr->typeCount = SWAP16(-1);
bcopy(RF_EMPTY_TAG, rsrcforkhdr->systemData, sizeof(RF_EMPTY_TAG));
}
static void
rel_xattrinfo(attr_info_t *ainfop)
{
kheap_free_addr(KHEAP_DATA_BUFFERS, ainfop->filehdr);
bzero(ainfop, sizeof(attr_info_t));
}
static int
write_xattrinfo(attr_info_t *ainfop)
{
uio_t auio;
int error;
auio = uio_create(1, 0, UIO_SYSSPACE, UIO_WRITE);
uio_addiov(auio, (uintptr_t)ainfop->filehdr, ainfop->iosize);
swap_adhdr(ainfop->filehdr);
if (ainfop->attrhdr != NULL) {
swap_attrhdr(ainfop->attrhdr, ainfop);
}
error = VNOP_WRITE(ainfop->filevp, auio, 0, ainfop->context);
swap_adhdr(ainfop->filehdr);
if (ainfop->attrhdr != NULL) {
swap_attrhdr(ainfop->attrhdr, ainfop);
}
uio_free(auio);
return error;
}
#if BYTE_ORDER == LITTLE_ENDIAN
/*
* Endian swap apple double header
*/
static void
swap_adhdr(apple_double_header_t *adh)
{
int count;
int i;
count = (adh->magic == ADH_MAGIC) ? adh->numEntries : SWAP16(adh->numEntries);
adh->magic = SWAP32(adh->magic);
adh->version = SWAP32(adh->version);
adh->numEntries = SWAP16(adh->numEntries);
for (i = 0; i < count; i++) {
adh->entries[i].type = SWAP32(adh->entries[i].type);
adh->entries[i].offset = SWAP32(adh->entries[i].offset);
adh->entries[i].length = SWAP32(adh->entries[i].length);
}
}
/*
* Endian swap extended attributes header
*/
static void
swap_attrhdr(attr_header_t *ah, attr_info_t* info)
{
attr_entry_t *ae;
int count;
int i;
count = (ah->magic == ATTR_HDR_MAGIC) ? ah->num_attrs : SWAP16(ah->num_attrs);
ah->magic = SWAP32(ah->magic);
ah->debug_tag = SWAP32(ah->debug_tag);
ah->total_size = SWAP32(ah->total_size);
ah->data_start = SWAP32(ah->data_start);
ah->data_length = SWAP32(ah->data_length);
ah->flags = SWAP16(ah->flags);
ah->num_attrs = SWAP16(ah->num_attrs);
ae = (attr_entry_t *)(&ah[1]);
for (i = 0; i < count && ATTR_VALID(ae, *info); i++, ae = ATTR_NEXT(ae)) {
ae->offset = SWAP32(ae->offset);
ae->length = SWAP32(ae->length);
ae->flags = SWAP16(ae->flags);
}
}
#endif
/*
* Validate and swap the attributes header contents, and each attribute's
* attr_entry_t.
*
* Note: Assumes the caller has verified that the Finder Info content is large
* enough to contain the attr_header structure itself. Therefore, we can
* swap the header fields before sanity checking them.
*/
static int
check_and_swap_attrhdr(attr_header_t *ah, attr_info_t *ainfop)
{
attr_entry_t *ae;
u_int8_t *buf_end;
u_int32_t end;
int count;
int i;
uint32_t total_header_size;
if (ah == NULL) {
return EINVAL;
}
if (SWAP32(ah->magic) != ATTR_HDR_MAGIC) {
return EINVAL;
}
/* Swap the basic header fields */
ah->magic = SWAP32(ah->magic);
ah->debug_tag = SWAP32(ah->debug_tag);
ah->total_size = SWAP32(ah->total_size);
ah->data_start = SWAP32(ah->data_start);
ah->data_length = SWAP32(ah->data_length);
ah->flags = SWAP16(ah->flags);
ah->num_attrs = SWAP16(ah->num_attrs);
/*
* Make sure the total_size fits within the Finder Info area, and the
* extended attribute data area fits within total_size.
*/
end = ah->data_start + ah->data_length;
if (ah->total_size > ainfop->finderinfo->offset + ainfop->finderinfo->length ||
ah->data_start < sizeof(attr_header_t) ||
end < ah->data_start ||
end > ah->total_size) {
return EINVAL;
}
/*
* Make sure each of the attr_entry_t's fits within total_size.
*/
buf_end = ainfop->rawdata + ah->data_start;
if (buf_end > ainfop->rawdata + ainfop->rawsize) {
return EINVAL;
}
count = ah->num_attrs;
if (count > 256) {
return EINVAL;
}
ae = (attr_entry_t *)(&ah[1]);
total_header_size = sizeof(attr_header_t);
for (i = 0; i < count; i++) {
/* Make sure the fixed-size part of this attr_entry_t fits. */
if ((u_int8_t *) &ae[1] > buf_end) {
return EINVAL;
}
/* Make sure the variable-length name fits (+1 is for NUL terminator) */
if (&ae->name[ae->namelen + 1] > buf_end) {
return EINVAL;
}
/* Make sure that namelen is matching name's real length, namelen included NUL */
if (strnlen((const char *)ae->name, ae->namelen) != ae->namelen - 1) {
return EINVAL;
}
/* Swap the attribute entry fields */
ae->offset = SWAP32(ae->offset);
ae->length = SWAP32(ae->length);
ae->flags = SWAP16(ae->flags);
/* Make sure the attribute content fits and points to the data part */
end = ae->offset + ae->length;
if (end < ae->offset || end > ah->total_size) {
return EINVAL;
}
/* Make sure entry points to data section and not header */
if (ae->offset < ah->data_start) {
return EINVAL;
}
/* We verified namelen is ok above, so add this entry's size to a total */
total_header_size += ATTR_ENTRY_LENGTH(ae->namelen);
ae = ATTR_NEXT(ae);
}
/* make sure data_start is actually after all the xattr key entries */
if (ah->data_start < total_header_size) {
return EINVAL;
}
return 0;
}
//
// "start" & "end" are byte offsets in the file.
// "to" is the byte offset we want to move the
// data to. "to" should be > "start".
//
// we do the copy backwards to avoid problems if
// there's an overlap.
//
static int
shift_data_down(vnode_t xvp, off_t start, size_t len, off_t delta, vfs_context_t context)
{
int ret, iolen;
size_t chunk, orig_chunk;
char *buff;
off_t pos;
kauth_cred_t ucred = vfs_context_ucred(context);
proc_t p = vfs_context_proc(context);
if (delta == 0 || len == 0) {
return 0;
}
chunk = 4096;
if (len < chunk) {
chunk = len;
}
orig_chunk = chunk;
if (kmem_alloc(kernel_map, (vm_offset_t *)&buff, chunk, VM_KERN_MEMORY_FILE)) {
return ENOMEM;
}
for (pos = start + len - chunk; pos >= start; pos -= chunk) {
ret = vn_rdwr(UIO_READ, xvp, buff, (int)chunk, pos, UIO_SYSSPACE, IO_NODELOCKED | IO_NOAUTH, ucred, &iolen, p);
if (iolen != 0) {
printf("xattr:shift_data: error reading data @ %lld (read %d of %lu) (%d)\n",
pos, ret, chunk, ret);
break;
}
ret = vn_rdwr(UIO_WRITE, xvp, buff, (int)chunk, pos + delta, UIO_SYSSPACE, IO_NODELOCKED | IO_NOAUTH, ucred, &iolen, p);
if (iolen != 0) {
printf("xattr:shift_data: error writing data @ %lld (wrote %d of %lu) (%d)\n",
pos + delta, ret, chunk, ret);
break;
}
if ((pos - (off_t)chunk) < start) {
chunk = pos - start;
if (chunk == 0) { // we're all done
break;
}
}
}
kmem_free(kernel_map, (vm_offset_t)buff, orig_chunk);
return 0;
}
static int
shift_data_up(vnode_t xvp, off_t start, size_t len, off_t delta, vfs_context_t context)
{
int ret, iolen;
size_t chunk, orig_chunk;
char *buff;
off_t pos;
off_t end;
kauth_cred_t ucred = vfs_context_ucred(context);
proc_t p = vfs_context_proc(context);
if (delta == 0 || len == 0) {
return 0;
}
chunk = 4096;
if (len < chunk) {
chunk = len;
}
orig_chunk = chunk;
end = start + len;
if (kmem_alloc(kernel_map, (vm_offset_t *)&buff, chunk, VM_KERN_MEMORY_FILE)) {
return ENOMEM;
}
for (pos = start; pos < end; pos += chunk) {
ret = vn_rdwr(UIO_READ, xvp, buff, (int)chunk, pos, UIO_SYSSPACE, IO_NODELOCKED | IO_NOAUTH, ucred, &iolen, p);
if (iolen != 0) {
printf("xattr:shift_data: error reading data @ %lld (read %d of %lu) (%d)\n",
pos, ret, chunk, ret);
break;
}
ret = vn_rdwr(UIO_WRITE, xvp, buff, (int)chunk, pos - delta, UIO_SYSSPACE, IO_NODELOCKED | IO_NOAUTH, ucred, &iolen, p);
if (iolen != 0) {
printf("xattr:shift_data: error writing data @ %lld (wrote %d of %lu) (%d)\n",
pos + delta, ret, chunk, ret);
break;
}
if ((pos + (off_t)chunk) > end) {
chunk = end - pos;
if (chunk == 0) { // we're all done
break;
}
}
}
kmem_free(kernel_map, (vm_offset_t)buff, orig_chunk);
return 0;
}
static int
lock_xattrfile(vnode_t xvp, short locktype, vfs_context_t context)
{
struct flock lf;
int error;
lf.l_whence = SEEK_SET;
lf.l_start = 0;
lf.l_len = 0;
lf.l_type = locktype; /* F_WRLCK or F_RDLCK */
/* Note: id is just a kernel address that's not a proc */
error = VNOP_ADVLOCK(xvp, (caddr_t)xvp, F_SETLK, &lf, F_FLOCK | F_WAIT, context, NULL);
return error == ENOTSUP ? 0 : error;
}
int
unlock_xattrfile(vnode_t xvp, vfs_context_t context)
{
struct flock lf;
int error;
lf.l_whence = SEEK_SET;
lf.l_start = 0;
lf.l_len = 0;
lf.l_type = F_UNLCK;
/* Note: id is just a kernel address that's not a proc */
error = VNOP_ADVLOCK(xvp, (caddr_t)xvp, F_UNLCK, &lf, F_FLOCK, context, NULL);
return error == ENOTSUP ? 0 : error;
}
#else /* CONFIG_APPLEDOUBLE */
static int
default_getxattr(__unused vnode_t vp, __unused const char *name,
__unused uio_t uio, __unused size_t *size, __unused int options,
__unused vfs_context_t context)
{
return ENOTSUP;
}
static int
default_setxattr(__unused vnode_t vp, __unused const char *name,
__unused uio_t uio, __unused int options, __unused vfs_context_t context)
{
return ENOTSUP;
}
static int
default_listxattr(__unused vnode_t vp,
__unused uio_t uio, __unused size_t *size, __unused int options,
__unused vfs_context_t context)
{
return ENOTSUP;
}
static int
default_removexattr(__unused vnode_t vp, __unused const char *name,
__unused int options, __unused vfs_context_t context)
{
return ENOTSUP;
}
#endif /* CONFIG_APPLEDOUBLE */