Home | History | Annotate | Line # | Download | only in lfs
ulfs_extattr.c revision 1.1
      1 /*	$NetBSD: ulfs_extattr.c,v 1.1 2013/06/06 00:40:55 dholland Exp $	*/
      2 /*  from NetBSD: ufs_extattr.c,v 1.41 2012/12/08 13:42:36 manu Exp  */
      3 
      4 /*-
      5  * Copyright (c) 1999-2002 Robert N. M. Watson
      6  * Copyright (c) 2002-2003 Networks Associates Technology, Inc.
      7  * All rights reserved.
      8  *
      9  * This software was developed by Robert Watson for the TrustedBSD Project.
     10  *
     11  * This software was developed for the FreeBSD Project in part by Network
     12  * Associates Laboratories, the Security Research Division of Network
     13  * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"),
     14  * as part of the DARPA CHATS research program.
     15  *
     16  * Redistribution and use in source and binary forms, with or without
     17  * modification, are permitted provided that the following conditions
     18  * are met:
     19  * 1. Redistributions of source code must retain the above copyright
     20  *    notice, this list of conditions and the following disclaimer.
     21  * 2. Redistributions in binary form must reproduce the above copyright
     22  *    notice, this list of conditions and the following disclaimer in the
     23  *    documentation and/or other materials provided with the distribution.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     35  * SUCH DAMAGE.
     36  *
     37  */
     38 
     39 /*
     40  * Support for file system extended attributes on the UFS1 file system.
     41  *
     42  * Extended attributes are defined in the form name=value, where name is
     43  * a nul-terminated string in the style of a file name, and value is a
     44  * binary blob of zero or more bytes.  The UFS1 extended attribute service
     45  * layers support for extended attributes onto a backing file, in the style
     46  * of the quota implementation, meaning that it requires no underlying format
     47  * changes to the file system.  This design choice exchanges simplicity,
     48  * usability, and easy deployment for performance.
     49  */
     50 
     51 #include <sys/cdefs.h>
     52 __KERNEL_RCSID(0, "$NetBSD: ulfs_extattr.c,v 1.1 2013/06/06 00:40:55 dholland Exp $");
     53 
     54 #ifdef _KERNEL_OPT
     55 #include "opt_ffs.h"
     56 #endif
     57 
     58 #include <sys/param.h>
     59 #include <sys/systm.h>
     60 #include <sys/reboot.h>
     61 #include <sys/kauth.h>
     62 #include <sys/kernel.h>
     63 #include <sys/namei.h>
     64 #include <sys/kmem.h>
     65 #include <sys/fcntl.h>
     66 #include <sys/lwp.h>
     67 #include <sys/vnode.h>
     68 #include <sys/mount.h>
     69 #include <sys/lock.h>
     70 #include <sys/dirent.h>
     71 #include <sys/extattr.h>
     72 #include <sys/sysctl.h>
     73 
     74 #include <ufs/ufs/dir.h>
     75 #include <ufs/ufs/extattr.h>
     76 #include <ufs/ufs/ufsmount.h>
     77 #include <ufs/ufs/inode.h>
     78 #include <ufs/ufs/ufs_bswap.h>
     79 #include <ufs/ufs/ufs_extern.h>
     80 
     81 int ufs_extattr_sync = 1;
     82 int ufs_extattr_autocreate = 1024;
     83 
     84 static int	ufs_extattr_valid_attrname(int attrnamespace,
     85 		    const char *attrname);
     86 static int	ufs_extattr_enable_with_open(struct ufsmount *ump,
     87 		    struct vnode *vp, int attrnamespace, const char *attrname,
     88 		    struct lwp *l);
     89 static int	ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
     90 		    const char *attrname, struct vnode *backing_vnode,
     91 		    struct lwp *l);
     92 static int	ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
     93 		    const char *attrname, struct lwp *l);
     94 static int	ufs_extattr_get(struct vnode *vp, int attrnamespace,
     95 		    const char *name, struct uio *uio, size_t *size,
     96 		    kauth_cred_t cred, struct lwp *l);
     97 static int	ufs_extattr_list(struct vnode *vp, int attrnamespace,
     98 		    struct uio *uio, size_t *size, int flag,
     99 		    kauth_cred_t cred, struct lwp *l);
    100 static int	ufs_extattr_set(struct vnode *vp, int attrnamespace,
    101 		    const char *name, struct uio *uio, kauth_cred_t cred,
    102 		    struct lwp *l);
    103 static int	ufs_extattr_rm(struct vnode *vp, int attrnamespace,
    104 		    const char *name, kauth_cred_t cred, struct lwp *l);
    105 static struct ufs_extattr_list_entry *ufs_extattr_find_attr(struct ufsmount *,
    106 		    int, const char *);
    107 static int	ufs_extattr_get_header(struct vnode *,
    108 		    struct ufs_extattr_list_entry *,
    109 		    struct ufs_extattr_header *, off_t *);
    110 
    111 /*
    112  * Convert a FreeBSD extended attribute and namespace to a consistent string
    113  * representation.
    114  *
    115  * The returned value, if not NULL, is guaranteed to be an allocated object
    116  * of its size as returned by strlen() + 1 and must be freed by the caller.
    117  */
    118 static char *
    119 from_freebsd_extattr(int attrnamespace, const char *attrname)
    120 {
    121 	const char *namespace;
    122 	char *attr;
    123 	size_t len;
    124 
    125 	if (attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
    126 		namespace = "system";
    127 	else if (attrnamespace == EXTATTR_NAMESPACE_USER)
    128 		namespace = "user";
    129 	else
    130 		return NULL;
    131 
    132 	/* <namespace>.<attrname>\0 */
    133 	len = strlen(namespace) + 1 + strlen(attrname) + 1;
    134 
    135 	attr = kmem_alloc(len, KM_SLEEP);
    136 
    137 	snprintf(attr, len, "%s.%s", namespace, attrname);
    138 
    139 	return attr;
    140 }
    141 
    142 /*
    143  * Internal wrapper around a conversion-check-free sequence.
    144  */
    145 static int
    146 internal_extattr_check_cred(vnode_t *vp, int attrnamespace, const char *name,
    147     kauth_cred_t cred, int access_mode)
    148 {
    149 	char *attr;
    150 	int error;
    151 
    152 	attr = from_freebsd_extattr(attrnamespace, name);
    153 	if (attr == NULL)
    154 		return EINVAL;
    155 
    156 	error = extattr_check_cred(vp, attr, cred, access_mode);
    157 
    158 	kmem_free(attr, strlen(attr) + 1);
    159 
    160 	return error;
    161 }
    162 
    163 /*
    164  * Per-FS attribute lock protecting attribute operations.
    165  * XXX Right now there is a lot of lock contention due to having a single
    166  * lock per-FS; really, this should be far more fine-grained.
    167  */
    168 static void
    169 ufs_extattr_uepm_lock(struct ufsmount *ump)
    170 {
    171 
    172 	/* XXX Why does this need to be recursive? */
    173 	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
    174 		ump->um_extattr.uepm_lockcnt++;
    175 		return;
    176 	}
    177 	mutex_enter(&ump->um_extattr.uepm_lock);
    178 }
    179 
    180 static void
    181 ufs_extattr_uepm_unlock(struct ufsmount *ump)
    182 {
    183 
    184 	if (ump->um_extattr.uepm_lockcnt != 0) {
    185 		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
    186 		ump->um_extattr.uepm_lockcnt--;
    187 		return;
    188 	}
    189 	mutex_exit(&ump->um_extattr.uepm_lock);
    190 }
    191 
    192 /*-
    193  * Determine whether the name passed is a valid name for an actual
    194  * attribute.
    195  *
    196  * Invalid currently consists of:
    197  *	 NULL pointer for attrname
    198  *	 zero-length attrname (used to retrieve application attribute list)
    199  */
    200 static int
    201 ufs_extattr_valid_attrname(int attrnamespace, const char *attrname)
    202 {
    203 
    204 	if (attrname == NULL)
    205 		return (0);
    206 	if (strlen(attrname) == 0)
    207 		return (0);
    208 	return (1);
    209 }
    210 
    211 /*
    212  * Autocreate an attribute storage
    213  */
    214 static struct ufs_extattr_list_entry *
    215 ufs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
    216     const char *attrname, struct lwp *l)
    217 {
    218 	struct mount *mp = vp->v_mount;
    219 	struct ufsmount *ump = VFSTOUFS(mp);
    220 	struct vnode *backing_vp;
    221 	struct nameidata nd;
    222 	struct pathbuf *pb;
    223 	char *path;
    224 	struct ufs_extattr_fileheader uef;
    225 	struct ufs_extattr_list_entry *uele;
    226 	int error;
    227 
    228 	path = PNBUF_GET();
    229 
    230 	/*
    231 	 * We only support system and user namespace autocreation
    232 	 */
    233 	switch (attrnamespace) {
    234 	case EXTATTR_NAMESPACE_SYSTEM:
    235 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    236 			       mp->mnt_stat.f_mntonname,
    237 			       UFS_EXTATTR_FSROOTSUBDIR,
    238 			       UFS_EXTATTR_SUBDIR_SYSTEM,
    239 			       attrname);
    240 		break;
    241 	case EXTATTR_NAMESPACE_USER:
    242 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    243 			       mp->mnt_stat.f_mntonname,
    244 			       UFS_EXTATTR_FSROOTSUBDIR,
    245 			       UFS_EXTATTR_SUBDIR_USER,
    246 			       attrname);
    247 		break;
    248 	default:
    249 		PNBUF_PUT(path);
    250 		return NULL;
    251 		break;
    252 	}
    253 
    254 	/*
    255 	 * XXX unlock/lock should only be done when setting extattr
    256 	 * on backing store or one of its parent directory
    257 	 * including root, but we always do it for now.
    258 	 */
    259 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    260 	VOP_UNLOCK(vp);
    261 
    262 	pb = pathbuf_create(path);
    263 	NDINIT(&nd, CREATE, LOCKPARENT, pb);
    264 
    265 	error = vn_open(&nd, O_CREAT|O_RDWR, 0600);
    266 
    267 	/*
    268 	 * Reacquire the lock on the vnode
    269 	 */
    270 	KASSERT(VOP_ISLOCKED(vp) == 0);
    271 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    272 
    273 	if (error != 0) {
    274 		pathbuf_destroy(pb);
    275 		PNBUF_PUT(path);
    276 		return NULL;
    277 	}
    278 
    279 	KASSERT(nd.ni_vp != NULL);
    280 	KASSERT(VOP_ISLOCKED(nd.ni_vp) == LK_EXCLUSIVE);
    281 	KASSERT(VOP_ISLOCKED(nd.ni_dvp) == 0);
    282 
    283 	/*
    284  	 * backing_vp is the backing store.
    285 	 */
    286 	backing_vp = nd.ni_vp;
    287 	pathbuf_destroy(pb);
    288 	PNBUF_PUT(path);
    289 
    290 	uef.uef_magic = UFS_EXTATTR_MAGIC;
    291 	uef.uef_version = UFS_EXTATTR_VERSION;
    292 	uef.uef_size = ufs_extattr_autocreate;
    293 
    294 	error = vn_rdwr(UIO_WRITE, backing_vp, &uef, sizeof(uef), 0,
    295 		        UIO_SYSSPACE, IO_NODELOCKED|IO_APPEND,
    296 			l->l_cred, NULL, l);
    297 
    298 	VOP_UNLOCK(backing_vp);
    299 
    300 	if (error != 0) {
    301 		printf("%s: write uef header failed for %s, error = %d\n",
    302 		       __func__, attrname, error);
    303 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    304 		return NULL;
    305 	}
    306 
    307 	/*
    308 	 * Now enable attribute.
    309 	 */
    310 	error = ufs_extattr_enable(ump,attrnamespace, attrname, backing_vp, l);
    311 	KASSERT(VOP_ISLOCKED(backing_vp) == 0);
    312 
    313 	if (error != 0) {
    314 		printf("%s: enable %s failed, error %d\n",
    315 		       __func__, attrname, error);
    316 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    317 		return NULL;
    318 	}
    319 
    320 	uele = ufs_extattr_find_attr(ump, attrnamespace, attrname);
    321 	if (uele == NULL) {
    322 		printf("%s: atttribute %s created but not found!\n",
    323 		       __func__, attrname);
    324 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    325 		return NULL;
    326 	}
    327 
    328 	printf("%s: EA backing store autocreated for %s\n",
    329 	       mp->mnt_stat.f_mntonname, attrname);
    330 
    331 	return uele;
    332 }
    333 
    334 /*
    335  * Locate an attribute given a name and mountpoint.
    336  * Must be holding uepm lock for the mount point.
    337  */
    338 static struct ufs_extattr_list_entry *
    339 ufs_extattr_find_attr(struct ufsmount *ump, int attrnamespace,
    340     const char *attrname)
    341 {
    342 	struct ufs_extattr_list_entry *search_attribute;
    343 
    344 	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
    345 	    search_attribute != NULL;
    346 	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
    347 		if (!(strncmp(attrname, search_attribute->uele_attrname,
    348 		    UFS_EXTATTR_MAXEXTATTRNAME)) &&
    349 		    (attrnamespace == search_attribute->uele_attrnamespace)) {
    350 			return (search_attribute);
    351 		}
    352 	}
    353 
    354 	return (0);
    355 }
    356 
    357 /*
    358  * Initialize per-FS structures supporting extended attributes.  Do not
    359  * start extended attributes yet.
    360  */
    361 void
    362 ufs_extattr_uepm_init(struct ufs_extattr_per_mount *uepm)
    363 {
    364 
    365 	uepm->uepm_flags = 0;
    366 	uepm->uepm_lockcnt = 0;
    367 
    368 	LIST_INIT(&uepm->uepm_list);
    369 	mutex_init(&uepm->uepm_lock, MUTEX_DEFAULT, IPL_NONE);
    370 	uepm->uepm_flags |= UFS_EXTATTR_UEPM_INITIALIZED;
    371 }
    372 
    373 /*
    374  * Destroy per-FS structures supporting extended attributes.  Assumes
    375  * that EAs have already been stopped, and will panic if not.
    376  */
    377 void
    378 ufs_extattr_uepm_destroy(struct ufs_extattr_per_mount *uepm)
    379 {
    380 
    381 	if (!(uepm->uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
    382 		panic("ufs_extattr_uepm_destroy: not initialized");
    383 
    384 	if ((uepm->uepm_flags & UFS_EXTATTR_UEPM_STARTED))
    385 		panic("ufs_extattr_uepm_destroy: called while still started");
    386 
    387 	/*
    388 	 * It's not clear that either order for the next two lines is
    389 	 * ideal, and it should never be a problem if this is only called
    390 	 * during unmount, and with vfs_busy().
    391 	 */
    392 	uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_INITIALIZED;
    393 	mutex_destroy(&uepm->uepm_lock);
    394 }
    395 
    396 /*
    397  * Start extended attribute support on an FS.
    398  */
    399 int
    400 ufs_extattr_start(struct mount *mp, struct lwp *l)
    401 {
    402 	struct ufsmount *ump;
    403 	int error = 0;
    404 
    405 	ump = VFSTOUFS(mp);
    406 
    407 	ufs_extattr_uepm_lock(ump);
    408 
    409 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED)) {
    410 		error = EOPNOTSUPP;
    411 		goto unlock;
    412 	}
    413 	if (ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED) {
    414 		error = EBUSY;
    415 		goto unlock;
    416 	}
    417 
    418 	ump->um_extattr.uepm_flags |= UFS_EXTATTR_UEPM_STARTED;
    419 
    420 	ump->um_extattr.uepm_ucred = l->l_cred;
    421 	kauth_cred_hold(ump->um_extattr.uepm_ucred);
    422 
    423  unlock:
    424 	ufs_extattr_uepm_unlock(ump);
    425 
    426 	return (error);
    427 }
    428 
    429 /*
    430  * Helper routine: given a locked parent directory and filename, return
    431  * the locked vnode of the inode associated with the name.  Will not
    432  * follow symlinks, may return any type of vnode.  Lock on parent will
    433  * be released even in the event of a failure.  In the event that the
    434  * target is the parent (i.e., "."), there will be two references and
    435  * one lock, requiring the caller to possibly special-case.
    436  */
    437 static int
    438 ufs_extattr_lookup(struct vnode *start_dvp, int lockparent, const char *dirname,
    439     struct vnode **vp, struct lwp *l)
    440 {
    441 	struct vop_lookup_args vargs;
    442 	struct componentname cnp;
    443 	struct vnode *target_vp;
    444 	char *pnbuf;
    445 	int error;
    446 
    447 	KASSERT(VOP_ISLOCKED(start_dvp) == LK_EXCLUSIVE);
    448 
    449 	pnbuf = PNBUF_GET();
    450 
    451 	memset(&cnp, 0, sizeof(cnp));
    452 	cnp.cn_nameiop = LOOKUP;
    453 	cnp.cn_flags = ISLASTCN | lockparent;
    454 	cnp.cn_cred = l->l_cred;
    455 	cnp.cn_nameptr = pnbuf;
    456 	error = copystr(dirname, pnbuf, MAXPATHLEN, &cnp.cn_namelen);
    457 	if (error) {
    458 		if (lockparent == 0) {
    459 			VOP_UNLOCK(start_dvp);
    460 		}
    461 		PNBUF_PUT(pnbuf);
    462 		printf("ufs_extattr_lookup: copystr failed\n");
    463 		return (error);
    464 	}
    465 	cnp.cn_namelen--;	/* trim nul termination */
    466 	vargs.a_desc = NULL;
    467 	vargs.a_dvp = start_dvp;
    468 	vargs.a_vpp = &target_vp;
    469 	vargs.a_cnp = &cnp;
    470 	error = ufs_lookup(&vargs);
    471 	PNBUF_PUT(pnbuf);
    472 	if (error) {
    473 		if (lockparent == 0) {
    474 			VOP_UNLOCK(start_dvp);
    475 		}
    476 		return (error);
    477 	}
    478 #if 0
    479 	if (target_vp == start_dvp)
    480 		panic("ufs_extattr_lookup: target_vp == start_dvp");
    481 #endif
    482 
    483 	if ((target_vp != start_dvp) && (lockparent == 0))
    484 		 VOP_UNLOCK(start_dvp);
    485 
    486 	KASSERT(VOP_ISLOCKED(target_vp) == LK_EXCLUSIVE);
    487 	*vp = target_vp;
    488 	return (0);
    489 }
    490 
    491 /*
    492  * Enable an EA using the passed filesystem, backing vnode, attribute name,
    493  * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
    494  * to be locked when passed in.  The vnode will be returned unlocked,
    495  * regardless of success/failure of the function.  As a result, the caller
    496  * will always need to vrele(), but not vput().
    497  */
    498 static int
    499 ufs_extattr_enable_with_open(struct ufsmount *ump, struct vnode *vp,
    500     int attrnamespace, const char *attrname, struct lwp *l)
    501 {
    502 	int error;
    503 
    504 	error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred);
    505 	if (error) {
    506 		printf("ufs_extattr_enable_with_open.VOP_OPEN(): failed "
    507 		    "with %d\n", error);
    508 		VOP_UNLOCK(vp);
    509 		return (error);
    510 	}
    511 
    512 	mutex_enter(vp->v_interlock);
    513 	vp->v_writecount++;
    514 	mutex_exit(vp->v_interlock);
    515 
    516 	vref(vp);
    517 
    518 	VOP_UNLOCK(vp);
    519 
    520 	error = ufs_extattr_enable(ump, attrnamespace, attrname, vp, l);
    521 	if (error != 0)
    522 		vn_close(vp, FREAD|FWRITE, l->l_cred);
    523 	return (error);
    524 }
    525 
    526 /*
    527  * Given a locked directory vnode, iterate over the names in the directory
    528  * and use ufs_extattr_lookup() to retrieve locked vnodes of potential
    529  * attribute files.  Then invoke ufs_extattr_enable_with_open() on each
    530  * to attempt to start the attribute.  Leaves the directory locked on
    531  * exit.
    532  */
    533 static int
    534 ufs_extattr_iterate_directory(struct ufsmount *ump, struct vnode *dvp,
    535     int attrnamespace, struct lwp *l)
    536 {
    537 	struct vop_readdir_args vargs;
    538 	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
    539 	struct dirent *dp, *edp;
    540 	struct vnode *attr_vp;
    541 	struct uio auio;
    542 	struct iovec aiov;
    543 	char *dirbuf;
    544 	int error, eofflag = 0;
    545 
    546 	if (dvp->v_type != VDIR)
    547 		return (ENOTDIR);
    548 
    549 	dirbuf = kmem_alloc(DIRBLKSIZ, KM_SLEEP);
    550 
    551 	auio.uio_iov = &aiov;
    552 	auio.uio_iovcnt = 1;
    553 	auio.uio_rw = UIO_READ;
    554 	auio.uio_offset = 0;
    555 	UIO_SETUP_SYSSPACE(&auio);
    556 
    557 	vargs.a_desc = NULL;
    558 	vargs.a_vp = dvp;
    559 	vargs.a_uio = &auio;
    560 	vargs.a_cred = l->l_cred;
    561 	vargs.a_eofflag = &eofflag;
    562 	vargs.a_ncookies = NULL;
    563 	vargs.a_cookies = NULL;
    564 
    565 	while (!eofflag) {
    566 		auio.uio_resid = DIRBLKSIZ;
    567 		aiov.iov_base = dirbuf;
    568 		aiov.iov_len = DIRBLKSIZ;
    569 		error = ufs_readdir(&vargs);
    570 		if (error) {
    571 			printf("ufs_extattr_iterate_directory: ufs_readdir "
    572 			    "%d\n", error);
    573 			return (error);
    574 		}
    575 
    576 		/*
    577 		 * XXXRW: While in UFS, we always get DIRBLKSIZ returns from
    578 		 * the directory code on success, on other file systems this
    579 		 * may not be the case.  For portability, we should check the
    580 		 * read length on return from ufs_readdir().
    581 		 */
    582 		edp = (struct dirent *)&dirbuf[DIRBLKSIZ];
    583 		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
    584 			if (dp->d_reclen == 0)
    585 				break;
    586 			/* Skip "." and ".." */
    587 			if (dp->d_name[0] == '.' &&
    588 			    (dp->d_name[1] == '\0' ||
    589 			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
    590 				goto next;
    591 			error = ufs_extattr_lookup(dvp, LOCKPARENT,
    592 			    dp->d_name, &attr_vp, l);
    593 			if (error == ENOENT) {
    594 				goto next; /* keep silent */
    595 			} else if (error) {
    596 				printf("ufs_extattr_iterate_directory: lookup "
    597 				    "%s %d\n", dp->d_name, error);
    598 			} else if (attr_vp == dvp) {
    599 				vrele(attr_vp);
    600 			} else if (attr_vp->v_type != VREG) {
    601 				vput(attr_vp);
    602 			} else {
    603 				error = ufs_extattr_enable_with_open(ump,
    604 				    attr_vp, attrnamespace, dp->d_name, l);
    605 				vrele(attr_vp);
    606 				if (error) {
    607 					printf("ufs_extattr_iterate_directory: "
    608 					    "enable %s %d\n", dp->d_name,
    609 					    error);
    610 				} else if (bootverbose) {
    611 					printf("%s: EA %s loaded\n",
    612 					       sbp->f_mntonname, dp->d_name);
    613 				}
    614 			}
    615  next:
    616 			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
    617 			if (dp >= edp)
    618 				break;
    619 		}
    620 	}
    621 	kmem_free(dirbuf, DIRBLKSIZ);
    622 
    623 	return (0);
    624 }
    625 
    626 /*
    627  * Auto-start of extended attributes, to be executed (optionally) at
    628  * mount-time.
    629  */
    630 int
    631 ufs_extattr_autostart(struct mount *mp, struct lwp *l)
    632 {
    633 	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
    634 	int error;
    635 
    636 	/*
    637 	 * Does UFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
    638 	 * If so, automatically start EA's.
    639 	 */
    640 	error = VFS_ROOT(mp, &rvp);
    641 	if (error) {
    642 		printf("ufs_extattr_autostart.VFS_ROOT() returned %d\n",
    643 		    error);
    644 		return (error);
    645 	}
    646 
    647 	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    648 
    649 	error = ufs_extattr_lookup(rvp, 0,
    650 	    UFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
    651 	if (error) {
    652 		/* rvp ref'd but now unlocked */
    653 		KASSERT(VOP_ISLOCKED(rvp) == 0);
    654 		vrele(rvp);
    655 		return (error);
    656 	}
    657 	if (rvp == attr_dvp) {
    658 		/* Should never happen. */
    659 		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    660 		vrele(attr_dvp);
    661 		vput(rvp);
    662 		return (EINVAL);
    663 	}
    664 	KASSERT(VOP_ISLOCKED(rvp) == 0);
    665 	vrele(rvp);
    666 
    667 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    668 
    669 	if (attr_dvp->v_type != VDIR) {
    670 		printf("ufs_extattr_autostart: %s != VDIR\n",
    671 		    UFS_EXTATTR_FSROOTSUBDIR);
    672 		goto return_vput_attr_dvp;
    673 	}
    674 
    675 	error = ufs_extattr_start(mp, l);
    676 	if (error) {
    677 		printf("ufs_extattr_autostart: ufs_extattr_start failed (%d)\n",
    678 		    error);
    679 		goto return_vput_attr_dvp;
    680 	}
    681 
    682 	/*
    683 	 * Look for two subdirectories: UFS_EXTATTR_SUBDIR_SYSTEM,
    684 	 * UFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
    685 	 * and start with appropriate type.  Failures in either don't
    686 	 * result in an over-all failure.  attr_dvp is left locked to
    687 	 * be cleaned up on exit.
    688 	 */
    689 	error = ufs_extattr_lookup(attr_dvp, LOCKPARENT,
    690 	    UFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, l);
    691 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    692 	if (error == 0) {
    693 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    694 		error = ufs_extattr_iterate_directory(VFSTOUFS(mp),
    695 		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, l);
    696 		if (error)
    697 			printf("ufs_extattr_iterate_directory returned %d\n",
    698 			    error);
    699 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    700 		vput(attr_system_dvp);
    701 	}
    702 
    703 	error = ufs_extattr_lookup(attr_dvp, LOCKPARENT,
    704 	    UFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, l);
    705 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    706 	if (error == 0) {
    707 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    708 		error = ufs_extattr_iterate_directory(VFSTOUFS(mp),
    709 		    attr_user_dvp, EXTATTR_NAMESPACE_USER, l);
    710 		if (error)
    711 			printf("ufs_extattr_iterate_directory returned %d\n",
    712 			    error);
    713 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    714 		vput(attr_user_dvp);
    715 	}
    716 
    717 	/* Mask startup failures in sub-directories. */
    718 	error = 0;
    719 
    720  return_vput_attr_dvp:
    721 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    722 	vput(attr_dvp);
    723 
    724 	return (error);
    725 }
    726 
    727 /*
    728  * Stop extended attribute support on an FS.
    729  */
    730 void
    731 ufs_extattr_stop(struct mount *mp, struct lwp *l)
    732 {
    733 	struct ufs_extattr_list_entry *uele;
    734 	struct ufsmount *ump = VFSTOUFS(mp);
    735 
    736 	ufs_extattr_uepm_lock(ump);
    737 
    738 	/*
    739 	 * If we haven't been started, no big deal.  Just short-circuit
    740 	 * the processing work.
    741 	 */
    742 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
    743 		goto unlock;
    744 	}
    745 
    746 	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
    747 		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
    748 		ufs_extattr_disable(ump, uele->uele_attrnamespace,
    749 		    uele->uele_attrname, l);
    750 	}
    751 
    752 	ump->um_extattr.uepm_flags &= ~UFS_EXTATTR_UEPM_STARTED;
    753 
    754 	kauth_cred_free(ump->um_extattr.uepm_ucred);
    755 	ump->um_extattr.uepm_ucred = NULL;
    756 
    757  unlock:
    758 	ufs_extattr_uepm_unlock(ump);
    759 }
    760 
    761 /*
    762  * Enable a named attribute on the specified filesystem; provide an
    763  * unlocked backing vnode to hold the attribute data.
    764  */
    765 static int
    766 ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
    767     const char *attrname, struct vnode *backing_vnode, struct lwp *l)
    768 {
    769 	struct ufs_extattr_list_entry *attribute;
    770 	struct iovec aiov;
    771 	struct uio auio;
    772 	int error = 0;
    773 
    774 	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
    775 		return (EINVAL);
    776 	if (backing_vnode->v_type != VREG)
    777 		return (EINVAL);
    778 
    779 	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
    780 
    781 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
    782 		error = EOPNOTSUPP;
    783 		goto free_exit;
    784 	}
    785 
    786 	if (ufs_extattr_find_attr(ump, attrnamespace, attrname)) {
    787 		error = EEXIST;
    788 		goto free_exit;
    789 	}
    790 
    791 	strncpy(attribute->uele_attrname, attrname,
    792 	    UFS_EXTATTR_MAXEXTATTRNAME);
    793 	attribute->uele_attrnamespace = attrnamespace;
    794 	memset(&attribute->uele_fileheader, 0,
    795 	    sizeof(struct ufs_extattr_fileheader));
    796 
    797 	attribute->uele_backing_vnode = backing_vnode;
    798 
    799 	auio.uio_iov = &aiov;
    800 	auio.uio_iovcnt = 1;
    801 	aiov.iov_base = (void *) &attribute->uele_fileheader;
    802 	aiov.iov_len = sizeof(struct ufs_extattr_fileheader);
    803 	auio.uio_resid = sizeof(struct ufs_extattr_fileheader);
    804 	auio.uio_offset = (off_t) 0;
    805 	auio.uio_rw = UIO_READ;
    806 	UIO_SETUP_SYSSPACE(&auio);
    807 
    808 	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
    809 	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
    810 	    ump->um_extattr.uepm_ucred);
    811 
    812 	if (error)
    813 		goto unlock_free_exit;
    814 
    815 	if (auio.uio_resid != 0) {
    816 		printf("ufs_extattr_enable: malformed attribute header\n");
    817 		error = EINVAL;
    818 		goto unlock_free_exit;
    819 	}
    820 
    821 	/*
    822 	 * Try to determine the byte order of the attribute file.
    823 	 */
    824 	if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) {
    825 		attribute->uele_flags |= UELE_F_NEEDSWAP;
    826 		attribute->uele_fileheader.uef_magic =
    827 		    ufs_rw32(attribute->uele_fileheader.uef_magic,
    828 			     UELE_NEEDSWAP(attribute));
    829 		if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) {
    830 			printf("ufs_extattr_enable: invalid attribute header "
    831 			       "magic\n");
    832 			error = EINVAL;
    833 			goto unlock_free_exit;
    834 		}
    835 	}
    836 	attribute->uele_fileheader.uef_version =
    837 	    ufs_rw32(attribute->uele_fileheader.uef_version,
    838 		     UELE_NEEDSWAP(attribute));
    839 	attribute->uele_fileheader.uef_size =
    840 	    ufs_rw32(attribute->uele_fileheader.uef_size,
    841 		     UELE_NEEDSWAP(attribute));
    842 
    843 	if (attribute->uele_fileheader.uef_version != UFS_EXTATTR_VERSION) {
    844 		printf("ufs_extattr_enable: incorrect attribute header "
    845 		    "version\n");
    846 		error = EINVAL;
    847 		goto unlock_free_exit;
    848 	}
    849 
    850 	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
    851 	    uele_entries);
    852 
    853 	VOP_UNLOCK(backing_vnode);
    854 	return (0);
    855 
    856  unlock_free_exit:
    857 	VOP_UNLOCK(backing_vnode);
    858 
    859  free_exit:
    860 	kmem_free(attribute, sizeof(*attribute));
    861 	return (error);
    862 }
    863 
    864 /*
    865  * Disable extended attribute support on an FS.
    866  */
    867 static int
    868 ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
    869     const char *attrname, struct lwp *l)
    870 {
    871 	struct ufs_extattr_list_entry *uele;
    872 	int error = 0;
    873 
    874 	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
    875 		return (EINVAL);
    876 
    877 	uele = ufs_extattr_find_attr(ump, attrnamespace, attrname);
    878 	if (!uele)
    879 		return (ENODATA);
    880 
    881 	LIST_REMOVE(uele, uele_entries);
    882 
    883 	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
    884 	    l->l_cred);
    885 
    886 	kmem_free(uele, sizeof(*uele));
    887 
    888 	return (error);
    889 }
    890 
    891 /*
    892  * VFS call to manage extended attributes in UFS.  If filename_vp is
    893  * non-NULL, it must be passed in locked, and regardless of errors in
    894  * processing, will be unlocked.
    895  */
    896 int
    897 ufs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    898     int attrnamespace, const char *attrname)
    899 {
    900 	struct lwp *l = curlwp;
    901 	struct ufsmount *ump = VFSTOUFS(mp);
    902 	int error;
    903 
    904 	/*
    905 	 * Only privileged processes can configure extended attributes.
    906 	 */
    907 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_FS_EXTATTR,
    908 	    0, mp, NULL, NULL);
    909 	if (error) {
    910 		if (filename_vp != NULL)
    911 			VOP_UNLOCK(filename_vp);
    912 		return (error);
    913 	}
    914 
    915 	switch(cmd) {
    916 	case UFS_EXTATTR_CMD_START:
    917 		if (filename_vp != NULL) {
    918 			VOP_UNLOCK(filename_vp);
    919 			return (EINVAL);
    920 		}
    921 		if (attrname != NULL)
    922 			return (EINVAL);
    923 
    924 		error = ufs_extattr_autostart(mp, l);
    925 		return (error);
    926 
    927 	case UFS_EXTATTR_CMD_STOP:
    928 		if (filename_vp != NULL) {
    929 			VOP_UNLOCK(filename_vp);
    930 			return (EINVAL);
    931 		}
    932 		if (attrname != NULL)
    933 			return (EINVAL);
    934 
    935 		ufs_extattr_stop(mp, l);
    936 		return (0);
    937 
    938 	case UFS_EXTATTR_CMD_ENABLE:
    939 		if (filename_vp == NULL)
    940 			return (EINVAL);
    941 		if (attrname == NULL) {
    942 			VOP_UNLOCK(filename_vp);
    943 			return (EINVAL);
    944 		}
    945 
    946 		/*
    947 		 * ufs_extattr_enable_with_open() will always unlock the
    948 		 * vnode, regardless of failure.
    949 		 */
    950 		ufs_extattr_uepm_lock(ump);
    951 		error = ufs_extattr_enable_with_open(ump, filename_vp,
    952 		    attrnamespace, attrname, l);
    953 		ufs_extattr_uepm_unlock(ump);
    954 		return (error);
    955 
    956 	case UFS_EXTATTR_CMD_DISABLE:
    957 		if (filename_vp != NULL) {
    958 			VOP_UNLOCK(filename_vp);
    959 			return (EINVAL);
    960 		}
    961 		if (attrname == NULL)
    962 			return (EINVAL);
    963 
    964 		ufs_extattr_uepm_lock(ump);
    965 		error = ufs_extattr_disable(ump, attrnamespace, attrname, l);
    966 		ufs_extattr_uepm_unlock(ump);
    967 		return (error);
    968 
    969 	default:
    970 		return (EINVAL);
    971 	}
    972 }
    973 
    974 /*
    975  * Read extended attribute header for a given vnode and attribute.
    976  * Backing vnode should be locked and unlocked by caller.
    977  */
    978 static int
    979 ufs_extattr_get_header(struct vnode *vp, struct ufs_extattr_list_entry *uele,
    980     struct ufs_extattr_header *ueh, off_t *bap)
    981 {
    982 	struct mount *mp = vp->v_mount;
    983 	struct ufsmount *ump = VFSTOUFS(mp);
    984 	struct inode *ip = VTOI(vp);
    985 	off_t base_offset;
    986 	struct iovec aiov;
    987 	struct uio aio;
    988 	int error;
    989 
    990 	/*
    991 	 * Find base offset of header in file based on file header size, and
    992 	 * data header size + maximum data size, indexed by inode number.
    993 	 */
    994 	base_offset = sizeof(struct ufs_extattr_fileheader) +
    995 	    ip->i_number * (sizeof(struct ufs_extattr_header) +
    996 	    uele->uele_fileheader.uef_size);
    997 
    998 	/*
    999 	 * Read in the data header to see if the data is defined, and if so
   1000 	 * how much.
   1001 	 */
   1002 	memset(ueh, 0, sizeof(struct ufs_extattr_header));
   1003 	aiov.iov_base = ueh;
   1004 	aiov.iov_len = sizeof(struct ufs_extattr_header);
   1005 	aio.uio_iov = &aiov;
   1006 	aio.uio_iovcnt = 1;
   1007 	aio.uio_rw = UIO_READ;
   1008 	aio.uio_offset = base_offset;
   1009 	aio.uio_resid = sizeof(struct ufs_extattr_header);
   1010 	UIO_SETUP_SYSSPACE(&aio);
   1011 
   1012 	error = VOP_READ(uele->uele_backing_vnode, &aio,
   1013 	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1014 	if (error)
   1015 		return error;
   1016 
   1017 	/*
   1018 	 * Attribute headers are kept in file system byte order.
   1019 	 * XXX What about the blob of data?
   1020 	 */
   1021 	ueh->ueh_flags = ufs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
   1022 	ueh->ueh_len   = ufs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
   1023 	ueh->ueh_i_gen = ufs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
   1024 
   1025 	/* Defined? */
   1026 	if ((ueh->ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
   1027 		return ENODATA;
   1028 
   1029 	/* Valid for the current inode generation? */
   1030 	if (ueh->ueh_i_gen != ip->i_gen) {
   1031 		/*
   1032 		 * The inode itself has a different generation number
   1033 		 * than the uele data.  For now, the best solution
   1034 		 * is to coerce this to undefined, and let it get cleaned
   1035 		 * up by the next write or extattrctl clean.
   1036 		 */
   1037 		printf("%s (%s): inode gen inconsistency (%u, %jd)\n",
   1038 		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
   1039 		       (intmax_t)ip->i_gen);
   1040 		return ENODATA;
   1041 	}
   1042 
   1043 	/* Local size consistency check. */
   1044 	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
   1045 		return ENXIO;
   1046 
   1047 	/* Return base offset */
   1048 	if (bap != NULL)
   1049 		*bap = base_offset;
   1050 
   1051 	return 0;
   1052 }
   1053 
   1054 /*
   1055  * Vnode operation to retrieve a named extended attribute.
   1056  */
   1057 int
   1058 ufs_getextattr(struct vop_getextattr_args *ap)
   1059 /*
   1060 vop_getextattr {
   1061 	IN struct vnode *a_vp;
   1062 	IN int a_attrnamespace;
   1063 	IN const char *a_name;
   1064 	INOUT struct uio *a_uio;
   1065 	OUT size_t *a_size;
   1066 	IN kauth_cred_t a_cred;
   1067 };
   1068 */
   1069 {
   1070 	struct mount *mp = ap->a_vp->v_mount;
   1071 	struct ufsmount *ump = VFSTOUFS(mp);
   1072 	int error;
   1073 
   1074 	ufs_extattr_uepm_lock(ump);
   1075 
   1076 	error = ufs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1077 	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
   1078 
   1079 	ufs_extattr_uepm_unlock(ump);
   1080 
   1081 	return (error);
   1082 }
   1083 
   1084 /*
   1085  * Real work associated with retrieving a named attribute--assumes that
   1086  * the attribute lock has already been grabbed.
   1087  */
   1088 static int
   1089 ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
   1090     struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
   1091 {
   1092 	struct ufs_extattr_list_entry *attribute;
   1093 	struct ufs_extattr_header ueh;
   1094 	struct mount *mp = vp->v_mount;
   1095 	struct ufsmount *ump = VFSTOUFS(mp);
   1096 	off_t base_offset;
   1097 	size_t len, old_len;
   1098 	int error = 0;
   1099 
   1100 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1101 		return (EOPNOTSUPP);
   1102 
   1103 	if (strlen(name) == 0)
   1104 		return (EINVAL);
   1105 
   1106 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1107 	    VREAD);
   1108 	if (error)
   1109 		return (error);
   1110 
   1111 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1112 	if (!attribute)
   1113 		return (ENODATA);
   1114 
   1115 	/*
   1116 	 * Allow only offsets of zero to encourage the read/replace
   1117 	 * extended attribute semantic.  Otherwise we can't guarantee
   1118 	 * atomicity, as we don't provide locks for extended attributes.
   1119 	 */
   1120 	if (uio != NULL && uio->uio_offset != 0)
   1121 		return (ENXIO);
   1122 
   1123 	/*
   1124 	 * Don't need to get a lock on the backing file if the getattr is
   1125 	 * being applied to the backing file, as the lock is already held.
   1126 	 */
   1127 	if (attribute->uele_backing_vnode != vp)
   1128 		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1129 
   1130 	error = ufs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1131 	if (error)
   1132 		goto vopunlock_exit;
   1133 
   1134 	/* Return full data size if caller requested it. */
   1135 	if (size != NULL)
   1136 		*size = ueh.ueh_len;
   1137 
   1138 	/* Return data if the caller requested it. */
   1139 	if (uio != NULL) {
   1140 		/* Allow for offset into the attribute data. */
   1141 		uio->uio_offset = base_offset + sizeof(struct
   1142 		    ufs_extattr_header);
   1143 
   1144 		/*
   1145 		 * Figure out maximum to transfer -- use buffer size and
   1146 		 * local data limit.
   1147 		 */
   1148 		len = MIN(uio->uio_resid, ueh.ueh_len);
   1149 		old_len = uio->uio_resid;
   1150 		uio->uio_resid = len;
   1151 
   1152 		error = VOP_READ(attribute->uele_backing_vnode, uio,
   1153 		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1154 		if (error)
   1155 			goto vopunlock_exit;
   1156 
   1157 		uio->uio_resid = old_len - (len - uio->uio_resid);
   1158 	}
   1159 
   1160  vopunlock_exit:
   1161 
   1162 	if (uio != NULL)
   1163 		uio->uio_offset = 0;
   1164 
   1165 	if (attribute->uele_backing_vnode != vp)
   1166 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1167 
   1168 	return (error);
   1169 }
   1170 
   1171 /*
   1172  * Vnode operation to list extended attribute for a vnode
   1173  */
   1174 int
   1175 ufs_listextattr(struct vop_listextattr_args *ap)
   1176 /*
   1177 vop_listextattr {
   1178 	IN struct vnode *a_vp;
   1179 	IN int a_attrnamespace;
   1180 	INOUT struct uio *a_uio;
   1181 	OUT size_t *a_size;
   1182 	IN int flag;
   1183 	IN kauth_cred_t a_cred;
   1184 	struct proc *a_p;
   1185 };
   1186 */
   1187 {
   1188 	struct mount *mp = ap->a_vp->v_mount;
   1189 	struct ufsmount *ump = VFSTOUFS(mp);
   1190 	int error;
   1191 
   1192 	ufs_extattr_uepm_lock(ump);
   1193 
   1194 	error = ufs_extattr_list(ap->a_vp, ap->a_attrnamespace,
   1195 	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
   1196 
   1197 	ufs_extattr_uepm_unlock(ump);
   1198 
   1199 	return (error);
   1200 }
   1201 
   1202 /*
   1203  * Real work associated with retrieving list of attributes--assumes that
   1204  * the attribute lock has already been grabbed.
   1205  */
   1206 static int
   1207 ufs_extattr_list(struct vnode *vp, int attrnamespace,
   1208     struct uio *uio, size_t *size, int flag,
   1209     kauth_cred_t cred, struct lwp *l)
   1210 {
   1211 	struct ufs_extattr_list_entry *uele;
   1212 	struct ufs_extattr_header ueh;
   1213 	struct mount *mp = vp->v_mount;
   1214 	struct ufsmount *ump = VFSTOUFS(mp);
   1215 	size_t listsize = 0;
   1216 	int error = 0;
   1217 
   1218 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1219 		return (EOPNOTSUPP);
   1220 
   1221 	/*
   1222 	 * XXX: We can move this inside the loop and iterate on individual
   1223 	 *	attributes.
   1224 	 */
   1225 	error = internal_extattr_check_cred(vp, attrnamespace, "", cred,
   1226 	    VREAD);
   1227 	if (error)
   1228 		return (error);
   1229 
   1230 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
   1231 		unsigned char attrnamelen;
   1232 
   1233 		if (uele->uele_attrnamespace != attrnamespace)
   1234 			continue;
   1235 
   1236 		error = ufs_extattr_get_header(vp, uele, &ueh, NULL);
   1237 		if (error == ENODATA)
   1238 			continue;
   1239 		if (error != 0)
   1240 			return error;
   1241 
   1242 		/*
   1243 		 * Don't need to get a lock on the backing file if
   1244 		 * the listattr is being applied to the backing file,
   1245 		 * as the lock is already held.
   1246 		 */
   1247 		if (uele->uele_backing_vnode != vp)
   1248 			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1249 
   1250 		/*
   1251 		 * +1 for trailing NUL (listxattr flavor)
   1252 		 *  or leading name length (extattr_list_file flavor)
   1253 	 	 */
   1254 		attrnamelen = strlen(uele->uele_attrname);
   1255 		listsize += attrnamelen + 1;
   1256 
   1257 		/* Return data if the caller requested it. */
   1258 		if (uio != NULL) {
   1259 			/*
   1260 			 * We support two flavors. Either NUL-terminated
   1261 			 * strings (a la listxattr), or non NUL-terminated,
   1262 			 * one byte length prefixed strings (for
   1263 			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
   1264 		 	 * that second behavior.
   1265 			 */
   1266 			if (flag & EXTATTR_LIST_LENPREFIX) {
   1267 				uint8_t len = (uint8_t)attrnamelen;
   1268 
   1269 				/* Copy leading name length */
   1270 				error = uiomove(&len, sizeof(len), uio);
   1271 				if (error != 0)
   1272 					break;
   1273 			} else {
   1274 				/* Include trailing NULL */
   1275 				attrnamelen++;
   1276 			}
   1277 
   1278 			error = uiomove(uele->uele_attrname,
   1279 					(size_t)attrnamelen, uio);
   1280 			if (error != 0)
   1281 				break;
   1282 		}
   1283 
   1284 		if (uele->uele_backing_vnode != vp)
   1285 			VOP_UNLOCK(uele->uele_backing_vnode);
   1286 
   1287 		if (error != 0)
   1288 			return error;
   1289 	}
   1290 
   1291 	if (uio != NULL)
   1292 		uio->uio_offset = 0;
   1293 
   1294 	/* Return full data size if caller requested it. */
   1295 	if (size != NULL)
   1296 		*size = listsize;
   1297 
   1298 	return 0;
   1299 }
   1300 
   1301 /*
   1302  * Vnode operation to remove a named attribute.
   1303  */
   1304 int
   1305 ufs_deleteextattr(struct vop_deleteextattr_args *ap)
   1306 /*
   1307 vop_deleteextattr {
   1308 	IN struct vnode *a_vp;
   1309 	IN int a_attrnamespace;
   1310 	IN const char *a_name;
   1311 	IN kauth_cred_t a_cred;
   1312 };
   1313 */
   1314 {
   1315 	struct mount *mp = ap->a_vp->v_mount;
   1316 	struct ufsmount *ump = VFSTOUFS(mp);
   1317 	int error;
   1318 
   1319 	ufs_extattr_uepm_lock(ump);
   1320 
   1321 	error = ufs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1322 	    ap->a_cred, curlwp);
   1323 
   1324 	ufs_extattr_uepm_unlock(ump);
   1325 
   1326 	return (error);
   1327 }
   1328 
   1329 /*
   1330  * Vnode operation to set a named attribute.
   1331  */
   1332 int
   1333 ufs_setextattr(struct vop_setextattr_args *ap)
   1334 /*
   1335 vop_setextattr {
   1336 	IN struct vnode *a_vp;
   1337 	IN int a_attrnamespace;
   1338 	IN const char *a_name;
   1339 	INOUT struct uio *a_uio;
   1340 	IN kauth_cred_t a_cred;
   1341 };
   1342 */
   1343 {
   1344 	struct mount *mp = ap->a_vp->v_mount;
   1345 	struct ufsmount *ump = VFSTOUFS(mp);
   1346 	int error;
   1347 
   1348 	ufs_extattr_uepm_lock(ump);
   1349 
   1350 	/*
   1351 	 * XXX: No longer a supported way to delete extended attributes.
   1352 	 */
   1353 	if (ap->a_uio == NULL) {
   1354 		ufs_extattr_uepm_unlock(ump);
   1355 		return (EINVAL);
   1356 	}
   1357 
   1358 	error = ufs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1359 	    ap->a_uio, ap->a_cred, curlwp);
   1360 
   1361 	ufs_extattr_uepm_unlock(ump);
   1362 
   1363 	return (error);
   1364 }
   1365 
   1366 /*
   1367  * Real work associated with setting a vnode's extended attributes;
   1368  * assumes that the attribute lock has already been grabbed.
   1369  */
   1370 static int
   1371 ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
   1372     struct uio *uio, kauth_cred_t cred, struct lwp *l)
   1373 {
   1374 	struct ufs_extattr_list_entry *attribute;
   1375 	struct ufs_extattr_header ueh;
   1376 	struct iovec local_aiov;
   1377 	struct uio local_aio;
   1378 	struct mount *mp = vp->v_mount;
   1379 	struct ufsmount *ump = VFSTOUFS(mp);
   1380 	struct inode *ip = VTOI(vp);
   1381 	off_t base_offset;
   1382 	int error = 0, ioflag;
   1383 
   1384 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1385 		return (EROFS);
   1386 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1387 		return (EOPNOTSUPP);
   1388 	if (!ufs_extattr_valid_attrname(attrnamespace, name))
   1389 		return (EINVAL);
   1390 
   1391 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1392 	    VWRITE);
   1393 	if (error)
   1394 		return (error);
   1395 
   1396 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1397 	if (!attribute) {
   1398 		attribute =  ufs_extattr_autocreate_attr(vp, attrnamespace,
   1399 							 name, l);
   1400 		if  (!attribute)
   1401 			return (ENODATA);
   1402 	}
   1403 
   1404 	/*
   1405 	 * Early rejection of invalid offsets/length.
   1406 	 * Reject: any offset but 0 (replace)
   1407 	 *	 Any size greater than attribute size limit
   1408  	 */
   1409 	if (uio->uio_offset != 0 ||
   1410 	    uio->uio_resid > attribute->uele_fileheader.uef_size)
   1411 		return (ENXIO);
   1412 
   1413 	/*
   1414 	 * Find base offset of header in file based on file header size, and
   1415 	 * data header size + maximum data size, indexed by inode number.
   1416 	 */
   1417 	base_offset = sizeof(struct ufs_extattr_fileheader) +
   1418 	    ip->i_number * (sizeof(struct ufs_extattr_header) +
   1419 	    attribute->uele_fileheader.uef_size);
   1420 
   1421 	/*
   1422 	 * Write out a data header for the data.
   1423 	 */
   1424 	ueh.ueh_len = ufs_rw32((uint32_t) uio->uio_resid,
   1425 	    UELE_NEEDSWAP(attribute));
   1426 	ueh.ueh_flags = ufs_rw32(UFS_EXTATTR_ATTR_FLAG_INUSE,
   1427 				 UELE_NEEDSWAP(attribute));
   1428 	ueh.ueh_i_gen = ufs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
   1429 	local_aiov.iov_base = &ueh;
   1430 	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
   1431 	local_aio.uio_iov = &local_aiov;
   1432 	local_aio.uio_iovcnt = 1;
   1433 	local_aio.uio_rw = UIO_WRITE;
   1434 	local_aio.uio_offset = base_offset;
   1435 	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
   1436 	UIO_SETUP_SYSSPACE(&local_aio);
   1437 
   1438 	/*
   1439 	 * Don't need to get a lock on the backing file if the setattr is
   1440 	 * being applied to the backing file, as the lock is already held.
   1441 	 */
   1442 	if (attribute->uele_backing_vnode != vp)
   1443 		vn_lock(attribute->uele_backing_vnode,
   1444 		    LK_EXCLUSIVE | LK_RETRY);
   1445 
   1446 	ioflag = IO_NODELOCKED;
   1447 	if (ufs_extattr_sync)
   1448 		ioflag |= IO_SYNC;
   1449 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1450 	    ump->um_extattr.uepm_ucred);
   1451 	if (error)
   1452 		goto vopunlock_exit;
   1453 
   1454 	if (local_aio.uio_resid != 0) {
   1455 		error = ENXIO;
   1456 		goto vopunlock_exit;
   1457 	}
   1458 
   1459 	/*
   1460 	 * Write out user data.
   1461 	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
   1462 	 */
   1463 	uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header);
   1464 
   1465 	ioflag = IO_NODELOCKED;
   1466 	if (ufs_extattr_sync)
   1467 		ioflag |= IO_SYNC;
   1468 	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
   1469 	    ump->um_extattr.uepm_ucred);
   1470 
   1471  vopunlock_exit:
   1472 	uio->uio_offset = 0;
   1473 
   1474 	if (attribute->uele_backing_vnode != vp)
   1475 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1476 
   1477 	return (error);
   1478 }
   1479 
   1480 /*
   1481  * Real work associated with removing an extended attribute from a vnode.
   1482  * Assumes the attribute lock has already been grabbed.
   1483  */
   1484 static int
   1485 ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
   1486     kauth_cred_t cred, struct lwp *l)
   1487 {
   1488 	struct ufs_extattr_list_entry *attribute;
   1489 	struct ufs_extattr_header ueh;
   1490 	struct mount *mp = vp->v_mount;
   1491 	struct ufsmount *ump = VFSTOUFS(mp);
   1492 	struct iovec local_aiov;
   1493 	struct uio local_aio;
   1494 	off_t base_offset;
   1495 	int error = 0, ioflag;
   1496 
   1497 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1498 		return (EROFS);
   1499 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1500 		return (EOPNOTSUPP);
   1501 	if (!ufs_extattr_valid_attrname(attrnamespace, name))
   1502 		return (EINVAL);
   1503 
   1504 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1505 	    VWRITE);
   1506 	if (error)
   1507 		return (error);
   1508 
   1509 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1510 	if (!attribute)
   1511 		return (ENODATA);
   1512 
   1513 	/*
   1514 	 * Don't need to get a lock on the backing file if the getattr is
   1515 	 * being applied to the backing file, as the lock is already held.
   1516 	 */
   1517 	if (attribute->uele_backing_vnode != vp)
   1518 		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
   1519 
   1520 	error = ufs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1521 	if (error)
   1522 		goto vopunlock_exit;
   1523 
   1524 	/* Flag it as not in use. */
   1525 	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
   1526 	ueh.ueh_len = 0;		/* ...ditto... */
   1527 
   1528 	local_aiov.iov_base = &ueh;
   1529 	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
   1530 	local_aio.uio_iov = &local_aiov;
   1531 	local_aio.uio_iovcnt = 1;
   1532 	local_aio.uio_rw = UIO_WRITE;
   1533 	local_aio.uio_offset = base_offset;
   1534 	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
   1535 	UIO_SETUP_SYSSPACE(&local_aio);
   1536 
   1537 	ioflag = IO_NODELOCKED;
   1538 	if (ufs_extattr_sync)
   1539 		ioflag |= IO_SYNC;
   1540 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1541 	    ump->um_extattr.uepm_ucred);
   1542 	if (error)
   1543 		goto vopunlock_exit;
   1544 
   1545 	if (local_aio.uio_resid != 0)
   1546 		error = ENXIO;
   1547 
   1548  vopunlock_exit:
   1549 	VOP_UNLOCK(attribute->uele_backing_vnode);
   1550 
   1551 	return (error);
   1552 }
   1553 
   1554 /*
   1555  * Called by UFS when an inode is no longer active and should have its
   1556  * attributes stripped.
   1557  */
   1558 void
   1559 ufs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
   1560 {
   1561 	struct ufs_extattr_list_entry *uele;
   1562 	struct mount *mp = vp->v_mount;
   1563 	struct ufsmount *ump = VFSTOUFS(mp);
   1564 
   1565 	/*
   1566 	 * In that case, we cannot lock. We should not have any active vnodes
   1567 	 * on the fs if this is not yet initialized but is going to be, so
   1568 	 * this can go unlocked.
   1569 	 */
   1570 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
   1571 		return;
   1572 
   1573 	ufs_extattr_uepm_lock(ump);
   1574 
   1575 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
   1576 		ufs_extattr_uepm_unlock(ump);
   1577 		return;
   1578 	}
   1579 
   1580 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
   1581 		ufs_extattr_rm(vp, uele->uele_attrnamespace,
   1582 		    uele->uele_attrname, lwp0.l_cred, l);
   1583 
   1584 	ufs_extattr_uepm_unlock(ump);
   1585 }
   1586 
   1587 void
   1588 ufs_extattr_init(void)
   1589 {
   1590 
   1591 }
   1592 
   1593 void
   1594 ufs_extattr_done(void)
   1595 {
   1596 
   1597 }
   1598