Home | History | Annotate | Line # | Download | only in lfs
ulfs_extattr.c revision 1.13
      1 /*	$NetBSD: ulfs_extattr.c,v 1.13 2016/07/07 06:55:44 msaitoh Exp $	*/
      2 /*  from NetBSD: ufs_extattr.c,v 1.46 2014/11/19 16:26:47 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 ULFS1 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 ULFS1 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 1.13 2016/07/07 06:55:44 msaitoh Exp $");
     53 
     54 #ifdef _KERNEL_OPT
     55 #include "opt_lfs.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/lfs/ulfs_extattr.h>
     75 #include <ufs/lfs/ulfsmount.h>
     76 #include <ufs/lfs/ulfs_inode.h>
     77 #include <ufs/lfs/ulfs_bswap.h>
     78 #include <ufs/lfs/ulfs_extern.h>
     79 
     80 int ulfs_extattr_sync = 1;
     81 int ulfs_extattr_autocreate = 1024;
     82 
     83 static int	ulfs_extattr_valid_attrname(int attrnamespace,
     84 		    const char *attrname);
     85 static int	ulfs_extattr_enable_with_open(struct ulfsmount *ump,
     86 		    struct vnode *vp, int attrnamespace, const char *attrname,
     87 		    struct lwp *l);
     88 static int	ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
     89 		    const char *attrname, struct vnode *backing_vnode,
     90 		    struct lwp *l);
     91 static int	ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
     92 		    const char *attrname, struct lwp *l);
     93 static int	ulfs_extattr_get(struct vnode *vp, int attrnamespace,
     94 		    const char *name, struct uio *uio, size_t *size,
     95 		    kauth_cred_t cred, struct lwp *l);
     96 static int	ulfs_extattr_list(struct vnode *vp, int attrnamespace,
     97 		    struct uio *uio, size_t *size, int flag,
     98 		    kauth_cred_t cred, struct lwp *l);
     99 static int	ulfs_extattr_set(struct vnode *vp, int attrnamespace,
    100 		    const char *name, struct uio *uio, kauth_cred_t cred,
    101 		    struct lwp *l);
    102 static int	ulfs_extattr_rm(struct vnode *vp, int attrnamespace,
    103 		    const char *name, kauth_cred_t cred, struct lwp *l);
    104 static struct ulfs_extattr_list_entry *ulfs_extattr_find_attr(struct ulfsmount *,
    105 		    int, const char *);
    106 static int	ulfs_extattr_get_header(struct vnode *,
    107 		    struct ulfs_extattr_list_entry *,
    108 		    struct ulfs_extattr_header *, off_t *);
    109 
    110 /*
    111  * Convert a FreeBSD extended attribute and namespace to a consistent string
    112  * representation.
    113  *
    114  * The returned value, if not NULL, is guaranteed to be an allocated object
    115  * of its size as returned by strlen() + 1 and must be freed by the caller.
    116  */
    117 static char *
    118 from_freebsd_extattr(int attrnamespace, const char *attrname)
    119 {
    120 	const char *namespace;
    121 	char *attr;
    122 	size_t len;
    123 
    124 	if (attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
    125 		namespace = "system";
    126 	else if (attrnamespace == EXTATTR_NAMESPACE_USER)
    127 		namespace = "user";
    128 	else
    129 		return NULL;
    130 
    131 	/* <namespace>.<attrname>\0 */
    132 	len = strlen(namespace) + 1 + strlen(attrname) + 1;
    133 
    134 	attr = kmem_alloc(len, KM_SLEEP);
    135 
    136 	snprintf(attr, len, "%s.%s", namespace, attrname);
    137 
    138 	return attr;
    139 }
    140 
    141 /*
    142  * Internal wrapper around a conversion-check-free sequence.
    143  */
    144 static int
    145 internal_extattr_check_cred(vnode_t *vp, int attrnamespace, const char *name,
    146     kauth_cred_t cred, int access_mode)
    147 {
    148 	char *attr;
    149 	int error;
    150 
    151 	attr = from_freebsd_extattr(attrnamespace, name);
    152 	if (attr == NULL)
    153 		return EINVAL;
    154 
    155 	error = extattr_check_cred(vp, attr, cred, access_mode);
    156 
    157 	kmem_free(attr, strlen(attr) + 1);
    158 
    159 	return error;
    160 }
    161 
    162 /*
    163  * Per-FS attribute lock protecting attribute operations.
    164  * XXX Right now there is a lot of lock contention due to having a single
    165  * lock per-FS; really, this should be far more fine-grained.
    166  */
    167 static void
    168 ulfs_extattr_uepm_lock(struct ulfsmount *ump)
    169 {
    170 
    171 	/* XXX Why does this need to be recursive? */
    172 	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
    173 		ump->um_extattr.uepm_lockcnt++;
    174 		return;
    175 	}
    176 	mutex_enter(&ump->um_extattr.uepm_lock);
    177 }
    178 
    179 static void
    180 ulfs_extattr_uepm_unlock(struct ulfsmount *ump)
    181 {
    182 
    183 	if (ump->um_extattr.uepm_lockcnt != 0) {
    184 		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
    185 		ump->um_extattr.uepm_lockcnt--;
    186 		return;
    187 	}
    188 	mutex_exit(&ump->um_extattr.uepm_lock);
    189 }
    190 
    191 /*-
    192  * Determine whether the name passed is a valid name for an actual
    193  * attribute.
    194  *
    195  * Invalid currently consists of:
    196  *	 NULL pointer for attrname
    197  *	 zero-length attrname (used to retrieve application attribute list)
    198  */
    199 static int
    200 ulfs_extattr_valid_attrname(int attrnamespace, const char *attrname)
    201 {
    202 
    203 	if (attrname == NULL)
    204 		return (0);
    205 	if (strlen(attrname) == 0)
    206 		return (0);
    207 	return (1);
    208 }
    209 
    210 /*
    211  * Autocreate an attribute storage
    212  */
    213 static int
    214 ulfs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
    215     const char *attrname, struct lwp *l, struct ulfs_extattr_list_entry **uelep)
    216 {
    217 	struct mount *mp = vp->v_mount;
    218 	struct ulfsmount *ump = VFSTOULFS(mp);
    219 	struct vnode *backing_vp;
    220 	struct nameidata nd;
    221 	struct pathbuf *pb;
    222 	char *path;
    223 	struct ulfs_extattr_fileheader uef;
    224 	struct ulfs_extattr_list_entry *uele;
    225 	int error;
    226 
    227 	path = PNBUF_GET();
    228 
    229 	/*
    230 	 * We only support system and user namespace autocreation
    231 	 */
    232 	switch (attrnamespace) {
    233 	case EXTATTR_NAMESPACE_SYSTEM:
    234 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    235 			       mp->mnt_stat.f_mntonname,
    236 			       ULFS_EXTATTR_FSROOTSUBDIR,
    237 			       ULFS_EXTATTR_SUBDIR_SYSTEM,
    238 			       attrname);
    239 		break;
    240 	case EXTATTR_NAMESPACE_USER:
    241 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    242 			       mp->mnt_stat.f_mntonname,
    243 			       ULFS_EXTATTR_FSROOTSUBDIR,
    244 			       ULFS_EXTATTR_SUBDIR_USER,
    245 			       attrname);
    246 		break;
    247 	default:
    248 		PNBUF_PUT(path);
    249 		*uelep = NULL;
    250 		return EINVAL;
    251 		break;
    252 	}
    253 
    254 	/*
    255 	 * Release extended attribute mount lock, otherwise
    256 	 * we can deadlock with another thread that would lock
    257 	 * vp after we unlock it below, and call
    258 	 * ulfs_extattr_uepm_lock(ump), for instance
    259 	 * in ulfs_getextattr().
    260 	 */
    261 	ulfs_extattr_uepm_unlock(ump);
    262 
    263 	/*
    264 	 * XXX unlock/lock should only be done when setting extattr
    265 	 * on backing store or one of its parent directory
    266 	 * including root, but we always do it for now.
    267 	 */
    268 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    269 	VOP_UNLOCK(vp);
    270 
    271 	pb = pathbuf_create(path);
    272 	NDINIT(&nd, CREATE, LOCKPARENT, pb);
    273 
    274 	/*
    275 	 * Since we do not hold ulfs_extattr_uepm_lock anymore,
    276 	 * another thread may race with us for backend creation,
    277 	 * but only one can succeed here thanks to O_EXCL
    278 	 */
    279 	error = vn_open(&nd, O_CREAT|O_EXCL|O_RDWR, 0600);
    280 
    281 	/*
    282 	 * Reacquire the lock on the vnode
    283 	 */
    284 	KASSERT(VOP_ISLOCKED(vp) == 0);
    285 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    286 
    287 	ulfs_extattr_uepm_lock(ump);
    288 
    289 	if (error != 0) {
    290 		pathbuf_destroy(pb);
    291 		PNBUF_PUT(path);
    292 		*uelep = NULL;
    293 		return error;
    294 	}
    295 
    296 	KASSERT(nd.ni_vp != NULL);
    297 	KASSERT(VOP_ISLOCKED(nd.ni_vp) == LK_EXCLUSIVE);
    298 	KASSERT(VOP_ISLOCKED(nd.ni_dvp) == 0);
    299 
    300 	/*
    301  	 * backing_vp is the backing store.
    302 	 */
    303 	backing_vp = nd.ni_vp;
    304 	pathbuf_destroy(pb);
    305 	PNBUF_PUT(path);
    306 
    307 	uef.uef_magic = ULFS_EXTATTR_MAGIC;
    308 	uef.uef_version = ULFS_EXTATTR_VERSION;
    309 	uef.uef_size = ulfs_extattr_autocreate;
    310 
    311 	error = vn_rdwr(UIO_WRITE, backing_vp, &uef, sizeof(uef), 0,
    312 		        UIO_SYSSPACE, IO_NODELOCKED|IO_APPEND,
    313 			l->l_cred, NULL, l);
    314 
    315 	VOP_UNLOCK(backing_vp);
    316 
    317 	if (error != 0) {
    318 		printf("%s: write uef header failed for %s, error = %d\n",
    319 		       __func__, attrname, error);
    320 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    321 		*uelep = NULL;
    322 		return error;
    323 	}
    324 
    325 	/*
    326 	 * Now enable attribute.
    327 	 */
    328 	error = ulfs_extattr_enable(ump,attrnamespace, attrname, backing_vp, l);
    329 	KASSERT(VOP_ISLOCKED(backing_vp) == 0);
    330 
    331 	if (error != 0) {
    332 		printf("%s: enable %s failed, error %d\n",
    333 		       __func__, attrname, error);
    334 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    335 		*uelep = NULL;
    336 		return error;
    337 	}
    338 
    339 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    340 	if (uele == NULL) {
    341 		printf("%s: atttribute %s created but not found!\n",
    342 		       __func__, attrname);
    343 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    344 		*uelep = NULL;
    345 		return ESRCH; /* really internal error */
    346 	}
    347 
    348 	printf("%s: EA backing store autocreated for %s\n",
    349 	       mp->mnt_stat.f_mntonname, attrname);
    350 
    351 	*uelep = uele;
    352 	return 0;
    353 }
    354 
    355 /*
    356  * Locate an attribute given a name and mountpoint.
    357  * Must be holding uepm lock for the mount point.
    358  */
    359 static struct ulfs_extattr_list_entry *
    360 ulfs_extattr_find_attr(struct ulfsmount *ump, int attrnamespace,
    361     const char *attrname)
    362 {
    363 	struct ulfs_extattr_list_entry *search_attribute;
    364 
    365 	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
    366 	    search_attribute != NULL;
    367 	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
    368 		if (!(strncmp(attrname, search_attribute->uele_attrname,
    369 		    ULFS_EXTATTR_MAXEXTATTRNAME)) &&
    370 		    (attrnamespace == search_attribute->uele_attrnamespace)) {
    371 			return (search_attribute);
    372 		}
    373 	}
    374 
    375 	return (0);
    376 }
    377 
    378 /*
    379  * Initialize per-FS structures supporting extended attributes.  Do not
    380  * start extended attributes yet.
    381  */
    382 void
    383 ulfs_extattr_uepm_init(struct ulfs_extattr_per_mount *uepm)
    384 {
    385 
    386 	uepm->uepm_flags = 0;
    387 	uepm->uepm_lockcnt = 0;
    388 
    389 	LIST_INIT(&uepm->uepm_list);
    390 	mutex_init(&uepm->uepm_lock, MUTEX_DEFAULT, IPL_NONE);
    391 	uepm->uepm_flags |= ULFS_EXTATTR_UEPM_INITIALIZED;
    392 }
    393 
    394 /*
    395  * Destroy per-FS structures supporting extended attributes.  Assumes
    396  * that EAs have already been stopped, and will panic if not.
    397  */
    398 void
    399 ulfs_extattr_uepm_destroy(struct ulfs_extattr_per_mount *uepm)
    400 {
    401 
    402 	if (!(uepm->uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    403 		panic("ulfs_extattr_uepm_destroy: not initialized");
    404 
    405 	if ((uepm->uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
    406 		panic("ulfs_extattr_uepm_destroy: called while still started");
    407 
    408 	/*
    409 	 * It's not clear that either order for the next three lines is
    410 	 * ideal, and it should never be a problem if this is only called
    411 	 * during unmount, and with vfs_busy().
    412 	 */
    413 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    414 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_INITIALIZED;
    415 	mutex_destroy(&uepm->uepm_lock);
    416 }
    417 
    418 /*
    419  * Start extended attribute support on an FS.
    420  */
    421 int
    422 ulfs_extattr_start(struct mount *mp, struct lwp *l)
    423 {
    424 	struct ulfsmount *ump;
    425 	int error = 0;
    426 
    427 	ump = VFSTOULFS(mp);
    428 
    429 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    430 		ulfs_extattr_uepm_init(&ump->um_extattr);
    431 
    432 	ulfs_extattr_uepm_lock(ump);
    433 
    434 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED)) {
    435 		error = EOPNOTSUPP;
    436 		goto unlock;
    437 	}
    438 	if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
    439 		error = EBUSY;
    440 		goto unlock;
    441 	}
    442 
    443 	ump->um_extattr.uepm_flags |= ULFS_EXTATTR_UEPM_STARTED;
    444 
    445 	ump->um_extattr.uepm_ucred = l->l_cred;
    446 	kauth_cred_hold(ump->um_extattr.uepm_ucred);
    447 
    448  unlock:
    449 	ulfs_extattr_uepm_unlock(ump);
    450 
    451 	return (error);
    452 }
    453 
    454 /*
    455  * Helper routine: given a locked parent directory and filename, return
    456  * the locked vnode of the inode associated with the name.  Will not
    457  * follow symlinks, may return any type of vnode.  Lock on parent will
    458  * be released even in the event of a failure.  In the event that the
    459  * target is the parent (i.e., "."), there will be two references and
    460  * one lock, requiring the caller to possibly special-case.
    461  */
    462 static int
    463 ulfs_extattr_lookup(struct vnode *start_dvp, int lockparent, const char *dirname,
    464     struct vnode **vp, struct lwp *l)
    465 {
    466 	struct vop_lookup_v2_args vargs;
    467 	struct componentname cnp;
    468 	struct vnode *target_vp;
    469 	char *pnbuf;
    470 	int error;
    471 
    472 	KASSERT(VOP_ISLOCKED(start_dvp) == LK_EXCLUSIVE);
    473 
    474 	pnbuf = PNBUF_GET();
    475 
    476 	memset(&cnp, 0, sizeof(cnp));
    477 	cnp.cn_nameiop = LOOKUP;
    478 	cnp.cn_flags = ISLASTCN | lockparent;
    479 	cnp.cn_cred = l->l_cred;
    480 	cnp.cn_nameptr = pnbuf;
    481 	error = copystr(dirname, pnbuf, MAXPATHLEN, &cnp.cn_namelen);
    482 	if (error) {
    483 		if (lockparent == 0) {
    484 			VOP_UNLOCK(start_dvp);
    485 		}
    486 		PNBUF_PUT(pnbuf);
    487 		printf("ulfs_extattr_lookup: copystr failed\n");
    488 		return (error);
    489 	}
    490 	cnp.cn_namelen--;	/* trim nul termination */
    491 	vargs.a_desc = NULL;
    492 	vargs.a_dvp = start_dvp;
    493 	vargs.a_vpp = &target_vp;
    494 	vargs.a_cnp = &cnp;
    495 	error = ulfs_lookup(&vargs);
    496 	PNBUF_PUT(pnbuf);
    497 	if (error) {
    498 		if (lockparent == 0) {
    499 			VOP_UNLOCK(start_dvp);
    500 		}
    501 		return (error);
    502 	}
    503 #if 0
    504 	if (target_vp == start_dvp)
    505 		panic("ulfs_extattr_lookup: target_vp == start_dvp");
    506 #endif
    507 
    508 	if (target_vp != start_dvp) {
    509 		error = vn_lock(target_vp, LK_EXCLUSIVE);
    510 		if (lockparent == 0)
    511 			VOP_UNLOCK(start_dvp);
    512 		if (error) {
    513 			vrele(target_vp);
    514 			return error;
    515 		}
    516 	}
    517 
    518 	KASSERT(VOP_ISLOCKED(target_vp) == LK_EXCLUSIVE);
    519 	*vp = target_vp;
    520 	return (0);
    521 }
    522 
    523 /*
    524  * Enable an EA using the passed filesystem, backing vnode, attribute name,
    525  * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
    526  * to be locked when passed in.  The vnode will be returned unlocked,
    527  * regardless of success/failure of the function.  As a result, the caller
    528  * will always need to vrele(), but not vput().
    529  */
    530 static int
    531 ulfs_extattr_enable_with_open(struct ulfsmount *ump, struct vnode *vp,
    532     int attrnamespace, const char *attrname, struct lwp *l)
    533 {
    534 	int error;
    535 
    536 	error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred);
    537 	if (error) {
    538 		printf("ulfs_extattr_enable_with_open.VOP_OPEN(): failed "
    539 		    "with %d\n", error);
    540 		VOP_UNLOCK(vp);
    541 		return (error);
    542 	}
    543 
    544 	mutex_enter(vp->v_interlock);
    545 	vp->v_writecount++;
    546 	mutex_exit(vp->v_interlock);
    547 
    548 	vref(vp);
    549 
    550 	VOP_UNLOCK(vp);
    551 
    552 	error = ulfs_extattr_enable(ump, attrnamespace, attrname, vp, l);
    553 	if (error != 0)
    554 		vn_close(vp, FREAD|FWRITE, l->l_cred);
    555 	return (error);
    556 }
    557 
    558 /*
    559  * Given a locked directory vnode, iterate over the names in the directory
    560  * and use ulfs_extattr_lookup() to retrieve locked vnodes of potential
    561  * attribute files.  Then invoke ulfs_extattr_enable_with_open() on each
    562  * to attempt to start the attribute.  Leaves the directory locked on
    563  * exit.
    564  */
    565 static int
    566 ulfs_extattr_iterate_directory(struct ulfsmount *ump, struct vnode *dvp,
    567     int attrnamespace, struct lwp *l)
    568 {
    569 	struct vop_readdir_args vargs;
    570 	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
    571 	struct dirent *dp, *edp;
    572 	struct vnode *attr_vp;
    573 	struct uio auio;
    574 	struct iovec aiov;
    575 	char *dirbuf;
    576 	int error, eofflag = 0;
    577 
    578 	if (dvp->v_type != VDIR)
    579 		return (ENOTDIR);
    580 
    581 	dirbuf = kmem_alloc(LFS_DIRBLKSIZ, KM_SLEEP);
    582 
    583 	auio.uio_iov = &aiov;
    584 	auio.uio_iovcnt = 1;
    585 	auio.uio_rw = UIO_READ;
    586 	auio.uio_offset = 0;
    587 	UIO_SETUP_SYSSPACE(&auio);
    588 
    589 	vargs.a_desc = NULL;
    590 	vargs.a_vp = dvp;
    591 	vargs.a_uio = &auio;
    592 	vargs.a_cred = l->l_cred;
    593 	vargs.a_eofflag = &eofflag;
    594 	vargs.a_ncookies = NULL;
    595 	vargs.a_cookies = NULL;
    596 
    597 	while (!eofflag) {
    598 		auio.uio_resid = LFS_DIRBLKSIZ;
    599 		aiov.iov_base = dirbuf;
    600 		aiov.iov_len = LFS_DIRBLKSIZ;
    601 		error = ulfs_readdir(&vargs);
    602 		if (error) {
    603 			printf("ulfs_extattr_iterate_directory: ulfs_readdir "
    604 			    "%d\n", error);
    605 			return (error);
    606 		}
    607 
    608 		/*
    609 		 * XXXRW: While in LFS, we always get LFS_DIRBLKSIZ returns from
    610 		 * the directory code on success, on other file systems this
    611 		 * may not be the case.  For portability, we should check the
    612 		 * read length on return from ulfs_readdir().
    613 		 */
    614 		edp = (struct dirent *)&dirbuf[LFS_DIRBLKSIZ];
    615 		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
    616 			if (dp->d_reclen == 0)
    617 				break;
    618 			/* Skip "." and ".." */
    619 			if (dp->d_name[0] == '.' &&
    620 			    (dp->d_name[1] == '\0' ||
    621 			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
    622 				goto next;
    623 			error = ulfs_extattr_lookup(dvp, LOCKPARENT,
    624 			    dp->d_name, &attr_vp, l);
    625 			if (error == ENOENT) {
    626 				goto next; /* keep silent */
    627 			} else if (error) {
    628 				printf("ulfs_extattr_iterate_directory: lookup "
    629 				    "%s %d\n", dp->d_name, error);
    630 			} else if (attr_vp == dvp) {
    631 				vrele(attr_vp);
    632 			} else if (attr_vp->v_type != VREG) {
    633 				vput(attr_vp);
    634 			} else {
    635 				error = ulfs_extattr_enable_with_open(ump,
    636 				    attr_vp, attrnamespace, dp->d_name, l);
    637 				vrele(attr_vp);
    638 				if (error) {
    639 					printf("ulfs_extattr_iterate_directory: "
    640 					    "enable %s %d\n", dp->d_name,
    641 					    error);
    642 				} else if (bootverbose) {
    643 					printf("%s: EA %s loaded\n",
    644 					       sbp->f_mntonname, dp->d_name);
    645 				}
    646 			}
    647  next:
    648 			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
    649 			if (dp >= edp)
    650 				break;
    651 		}
    652 	}
    653 	kmem_free(dirbuf, LFS_DIRBLKSIZ);
    654 
    655 	return (0);
    656 }
    657 
    658 /*
    659  * Auto-start of extended attributes, to be executed (optionally) at
    660  * mount-time.
    661  */
    662 int
    663 ulfs_extattr_autostart(struct mount *mp, struct lwp *l)
    664 {
    665 	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
    666 	int error;
    667 
    668 	/*
    669 	 * Does ULFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
    670 	 * If so, automatically start EA's.
    671 	 */
    672 	error = VFS_ROOT(mp, &rvp);
    673 	if (error) {
    674 		printf("ulfs_extattr_autostart.VFS_ROOT() returned %d\n",
    675 		    error);
    676 		return (error);
    677 	}
    678 
    679 	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    680 
    681 	error = ulfs_extattr_lookup(rvp, 0,
    682 	    ULFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
    683 	if (error) {
    684 		/* rvp ref'd but now unlocked */
    685 		KASSERT(VOP_ISLOCKED(rvp) == 0);
    686 		vrele(rvp);
    687 		return (error);
    688 	}
    689 	if (rvp == attr_dvp) {
    690 		/* Should never happen. */
    691 		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    692 		vrele(attr_dvp);
    693 		vput(rvp);
    694 		return (EINVAL);
    695 	}
    696 	KASSERT(VOP_ISLOCKED(rvp) == 0);
    697 	vrele(rvp);
    698 
    699 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    700 
    701 	if (attr_dvp->v_type != VDIR) {
    702 		printf("ulfs_extattr_autostart: %s != VDIR\n",
    703 		    ULFS_EXTATTR_FSROOTSUBDIR);
    704 		goto return_vput_attr_dvp;
    705 	}
    706 
    707 	error = ulfs_extattr_start(mp, l);
    708 	if (error) {
    709 		printf("ulfs_extattr_autostart: ulfs_extattr_start failed (%d)\n",
    710 		    error);
    711 		goto return_vput_attr_dvp;
    712 	}
    713 
    714 	/*
    715 	 * Look for two subdirectories: ULFS_EXTATTR_SUBDIR_SYSTEM,
    716 	 * ULFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
    717 	 * and start with appropriate type.  Failures in either don't
    718 	 * result in an over-all failure.  attr_dvp is left locked to
    719 	 * be cleaned up on exit.
    720 	 */
    721 	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    722 	    ULFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, l);
    723 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    724 	if (error == 0) {
    725 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    726 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    727 		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, l);
    728 		if (error)
    729 			printf("ulfs_extattr_iterate_directory returned %d\n",
    730 			    error);
    731 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    732 		vput(attr_system_dvp);
    733 	}
    734 
    735 	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    736 	    ULFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, l);
    737 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    738 	if (error == 0) {
    739 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    740 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    741 		    attr_user_dvp, EXTATTR_NAMESPACE_USER, l);
    742 		if (error)
    743 			printf("ulfs_extattr_iterate_directory returned %d\n",
    744 			    error);
    745 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    746 		vput(attr_user_dvp);
    747 	}
    748 
    749 	/* Mask startup failures in sub-directories. */
    750 	error = 0;
    751 
    752  return_vput_attr_dvp:
    753 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    754 	vput(attr_dvp);
    755 
    756 	return (error);
    757 }
    758 
    759 /*
    760  * Stop extended attribute support on an FS.
    761  */
    762 void
    763 ulfs_extattr_stop(struct mount *mp, struct lwp *l)
    764 {
    765 	struct ulfs_extattr_list_entry *uele;
    766 	struct ulfsmount *ump = VFSTOULFS(mp);
    767 
    768 	ulfs_extattr_uepm_lock(ump);
    769 
    770 	/*
    771 	 * If we haven't been started, no big deal.  Just short-circuit
    772 	 * the processing work.
    773 	 */
    774 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    775 		goto unlock;
    776 	}
    777 
    778 	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
    779 		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
    780 		ulfs_extattr_disable(ump, uele->uele_attrnamespace,
    781 		    uele->uele_attrname, l);
    782 	}
    783 
    784 	ump->um_extattr.uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    785 
    786 	kauth_cred_free(ump->um_extattr.uepm_ucred);
    787 	ump->um_extattr.uepm_ucred = NULL;
    788 
    789  unlock:
    790 	ulfs_extattr_uepm_unlock(ump);
    791 }
    792 
    793 /*
    794  * Enable a named attribute on the specified filesystem; provide an
    795  * unlocked backing vnode to hold the attribute data.
    796  */
    797 static int
    798 ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
    799     const char *attrname, struct vnode *backing_vnode, struct lwp *l)
    800 {
    801 	struct ulfs_extattr_list_entry *attribute;
    802 	struct iovec aiov;
    803 	struct uio auio;
    804 	int error = 0;
    805 
    806 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    807 		return (EINVAL);
    808 	if (backing_vnode->v_type != VREG)
    809 		return (EINVAL);
    810 
    811 	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
    812 
    813 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    814 		error = EOPNOTSUPP;
    815 		goto free_exit;
    816 	}
    817 
    818 	if (ulfs_extattr_find_attr(ump, attrnamespace, attrname)) {
    819 		error = EEXIST;
    820 		goto free_exit;
    821 	}
    822 
    823 	strncpy(attribute->uele_attrname, attrname,
    824 	    ULFS_EXTATTR_MAXEXTATTRNAME);
    825 	attribute->uele_attrnamespace = attrnamespace;
    826 	memset(&attribute->uele_fileheader, 0,
    827 	    sizeof(struct ulfs_extattr_fileheader));
    828 
    829 	attribute->uele_backing_vnode = backing_vnode;
    830 
    831 	auio.uio_iov = &aiov;
    832 	auio.uio_iovcnt = 1;
    833 	aiov.iov_base = (void *) &attribute->uele_fileheader;
    834 	aiov.iov_len = sizeof(struct ulfs_extattr_fileheader);
    835 	auio.uio_resid = sizeof(struct ulfs_extattr_fileheader);
    836 	auio.uio_offset = (off_t) 0;
    837 	auio.uio_rw = UIO_READ;
    838 	UIO_SETUP_SYSSPACE(&auio);
    839 
    840 	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
    841 	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
    842 	    ump->um_extattr.uepm_ucred);
    843 
    844 	if (error)
    845 		goto unlock_free_exit;
    846 
    847 	if (auio.uio_resid != 0) {
    848 		printf("ulfs_extattr_enable: malformed attribute header\n");
    849 		error = EINVAL;
    850 		goto unlock_free_exit;
    851 	}
    852 
    853 	/*
    854 	 * Try to determine the byte order of the attribute file.
    855 	 */
    856 	if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    857 		attribute->uele_flags |= UELE_F_NEEDSWAP;
    858 		attribute->uele_fileheader.uef_magic =
    859 		    ulfs_rw32(attribute->uele_fileheader.uef_magic,
    860 			     UELE_NEEDSWAP(attribute));
    861 		if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    862 			printf("ulfs_extattr_enable: invalid attribute header "
    863 			       "magic\n");
    864 			error = EINVAL;
    865 			goto unlock_free_exit;
    866 		}
    867 	}
    868 	attribute->uele_fileheader.uef_version =
    869 	    ulfs_rw32(attribute->uele_fileheader.uef_version,
    870 		     UELE_NEEDSWAP(attribute));
    871 	attribute->uele_fileheader.uef_size =
    872 	    ulfs_rw32(attribute->uele_fileheader.uef_size,
    873 		     UELE_NEEDSWAP(attribute));
    874 
    875 	if (attribute->uele_fileheader.uef_version != ULFS_EXTATTR_VERSION) {
    876 		printf("ulfs_extattr_enable: incorrect attribute header "
    877 		    "version\n");
    878 		error = EINVAL;
    879 		goto unlock_free_exit;
    880 	}
    881 
    882 	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
    883 	    uele_entries);
    884 
    885 	VOP_UNLOCK(backing_vnode);
    886 	return (0);
    887 
    888  unlock_free_exit:
    889 	VOP_UNLOCK(backing_vnode);
    890 
    891  free_exit:
    892 	kmem_free(attribute, sizeof(*attribute));
    893 	return (error);
    894 }
    895 
    896 /*
    897  * Disable extended attribute support on an FS.
    898  */
    899 static int
    900 ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
    901     const char *attrname, struct lwp *l)
    902 {
    903 	struct ulfs_extattr_list_entry *uele;
    904 	int error = 0;
    905 
    906 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    907 		return (EINVAL);
    908 
    909 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    910 	if (!uele)
    911 		return (ENODATA);
    912 
    913 	LIST_REMOVE(uele, uele_entries);
    914 
    915 	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
    916 	    l->l_cred);
    917 
    918 	kmem_free(uele, sizeof(*uele));
    919 
    920 	return (error);
    921 }
    922 
    923 /*
    924  * VFS call to manage extended attributes in ULFS.  If filename_vp is
    925  * non-NULL, it must be passed in locked, and regardless of errors in
    926  * processing, will be unlocked.
    927  */
    928 int
    929 ulfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    930     int attrnamespace, const char *attrname)
    931 {
    932 	struct lwp *l = curlwp;
    933 	struct ulfsmount *ump = VFSTOULFS(mp);
    934 	int error;
    935 
    936 	/*
    937 	 * Only privileged processes can configure extended attributes.
    938 	 */
    939 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_FS_EXTATTR,
    940 	    0, mp, NULL, NULL);
    941 	if (error) {
    942 		if (filename_vp != NULL)
    943 			VOP_UNLOCK(filename_vp);
    944 		return (error);
    945 	}
    946 
    947 	switch(cmd) {
    948 	case ULFS_EXTATTR_CMD_START:
    949 		if (filename_vp != NULL) {
    950 			VOP_UNLOCK(filename_vp);
    951 			return (EINVAL);
    952 		}
    953 		if (attrname != NULL)
    954 			return (EINVAL);
    955 
    956 		error = ulfs_extattr_autostart(mp, l);
    957 		return (error);
    958 
    959 	case ULFS_EXTATTR_CMD_STOP:
    960 		if (filename_vp != NULL) {
    961 			VOP_UNLOCK(filename_vp);
    962 			return (EINVAL);
    963 		}
    964 		if (attrname != NULL)
    965 			return (EINVAL);
    966 
    967 		ulfs_extattr_stop(mp, l);
    968 		return (0);
    969 
    970 	case ULFS_EXTATTR_CMD_ENABLE:
    971 		if (filename_vp == NULL)
    972 			return (EINVAL);
    973 		if (attrname == NULL) {
    974 			VOP_UNLOCK(filename_vp);
    975 			return (EINVAL);
    976 		}
    977 
    978 		/*
    979 		 * ulfs_extattr_enable_with_open() will always unlock the
    980 		 * vnode, regardless of failure.
    981 		 */
    982 		ulfs_extattr_uepm_lock(ump);
    983 		error = ulfs_extattr_enable_with_open(ump, filename_vp,
    984 		    attrnamespace, attrname, l);
    985 		ulfs_extattr_uepm_unlock(ump);
    986 		return (error);
    987 
    988 	case ULFS_EXTATTR_CMD_DISABLE:
    989 		if (filename_vp != NULL) {
    990 			VOP_UNLOCK(filename_vp);
    991 			return (EINVAL);
    992 		}
    993 		if (attrname == NULL)
    994 			return (EINVAL);
    995 
    996 		ulfs_extattr_uepm_lock(ump);
    997 		error = ulfs_extattr_disable(ump, attrnamespace, attrname, l);
    998 		ulfs_extattr_uepm_unlock(ump);
    999 		return (error);
   1000 
   1001 	default:
   1002 		return (EINVAL);
   1003 	}
   1004 }
   1005 
   1006 /*
   1007  * Read extended attribute header for a given vnode and attribute.
   1008  * Backing vnode should be locked and unlocked by caller.
   1009  */
   1010 static int
   1011 ulfs_extattr_get_header(struct vnode *vp, struct ulfs_extattr_list_entry *uele,
   1012     struct ulfs_extattr_header *ueh, off_t *bap)
   1013 {
   1014 	struct mount *mp = vp->v_mount;
   1015 	struct ulfsmount *ump = VFSTOULFS(mp);
   1016 	struct inode *ip = VTOI(vp);
   1017 	off_t base_offset;
   1018 	struct iovec aiov;
   1019 	struct uio aio;
   1020 	int error;
   1021 
   1022 	/*
   1023 	 * Find base offset of header in file based on file header size, and
   1024 	 * data header size + maximum data size, indexed by inode number.
   1025 	 */
   1026 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
   1027 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
   1028 	    uele->uele_fileheader.uef_size);
   1029 
   1030 	/*
   1031 	 * Read in the data header to see if the data is defined, and if so
   1032 	 * how much.
   1033 	 */
   1034 	memset(ueh, 0, sizeof(struct ulfs_extattr_header));
   1035 	aiov.iov_base = ueh;
   1036 	aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1037 	aio.uio_iov = &aiov;
   1038 	aio.uio_iovcnt = 1;
   1039 	aio.uio_rw = UIO_READ;
   1040 	aio.uio_offset = base_offset;
   1041 	aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1042 	UIO_SETUP_SYSSPACE(&aio);
   1043 
   1044 	error = VOP_READ(uele->uele_backing_vnode, &aio,
   1045 	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1046 	if (error)
   1047 		return error;
   1048 
   1049 	/*
   1050 	 * Attribute headers are kept in file system byte order.
   1051 	 * XXX What about the blob of data?
   1052 	 */
   1053 	ueh->ueh_flags = ulfs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
   1054 	ueh->ueh_len   = ulfs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
   1055 	ueh->ueh_i_gen = ulfs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
   1056 
   1057 	/* Defined? */
   1058 	if ((ueh->ueh_flags & ULFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
   1059 		return ENODATA;
   1060 
   1061 	/* Valid for the current inode generation? */
   1062 	if (ueh->ueh_i_gen != ip->i_gen) {
   1063 		/*
   1064 		 * The inode itself has a different generation number
   1065 		 * than the uele data.  For now, the best solution
   1066 		 * is to coerce this to undefined, and let it get cleaned
   1067 		 * up by the next write or extattrctl clean.
   1068 		 */
   1069 		printf("%s (%s): inode gen inconsistency (%u, %jd)\n",
   1070 		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
   1071 		       (intmax_t)ip->i_gen);
   1072 		return ENODATA;
   1073 	}
   1074 
   1075 	/* Local size consistency check. */
   1076 	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
   1077 		return ENXIO;
   1078 
   1079 	/* Return base offset */
   1080 	if (bap != NULL)
   1081 		*bap = base_offset;
   1082 
   1083 	return 0;
   1084 }
   1085 
   1086 /*
   1087  * Vnode operation to retrieve a named extended attribute.
   1088  */
   1089 int
   1090 ulfs_getextattr(struct vop_getextattr_args *ap)
   1091 /*
   1092 vop_getextattr {
   1093 	IN struct vnode *a_vp;
   1094 	IN int a_attrnamespace;
   1095 	IN const char *a_name;
   1096 	INOUT struct uio *a_uio;
   1097 	OUT size_t *a_size;
   1098 	IN kauth_cred_t a_cred;
   1099 };
   1100 */
   1101 {
   1102 	struct mount *mp = ap->a_vp->v_mount;
   1103 	struct ulfsmount *ump = VFSTOULFS(mp);
   1104 	int error;
   1105 
   1106 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1107 		return (EOPNOTSUPP);
   1108 
   1109 	ulfs_extattr_uepm_lock(ump);
   1110 
   1111 	error = ulfs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1112 	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
   1113 
   1114 	ulfs_extattr_uepm_unlock(ump);
   1115 
   1116 	return (error);
   1117 }
   1118 
   1119 /*
   1120  * Real work associated with retrieving a named attribute--assumes that
   1121  * the attribute lock has already been grabbed.
   1122  */
   1123 static int
   1124 ulfs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
   1125     struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
   1126 {
   1127 	struct ulfs_extattr_list_entry *attribute;
   1128 	struct ulfs_extattr_header ueh;
   1129 	struct mount *mp = vp->v_mount;
   1130 	struct ulfsmount *ump = VFSTOULFS(mp);
   1131 	off_t base_offset;
   1132 	size_t len, old_len;
   1133 	int error = 0;
   1134 
   1135 	if (strlen(name) == 0)
   1136 		return (EINVAL);
   1137 
   1138 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1139 	    VREAD);
   1140 	if (error)
   1141 		return (error);
   1142 
   1143 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1144 	if (!attribute)
   1145 		return (ENODATA);
   1146 
   1147 	/*
   1148 	 * Allow only offsets of zero to encourage the read/replace
   1149 	 * extended attribute semantic.  Otherwise we can't guarantee
   1150 	 * atomicity, as we don't provide locks for extended attributes.
   1151 	 */
   1152 	if (uio != NULL && uio->uio_offset != 0)
   1153 		return (ENXIO);
   1154 
   1155 	/*
   1156 	 * Don't need to get a lock on the backing file if the getattr is
   1157 	 * being applied to the backing file, as the lock is already held.
   1158 	 */
   1159 	if (attribute->uele_backing_vnode != vp)
   1160 		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1161 
   1162 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1163 	if (error)
   1164 		goto vopunlock_exit;
   1165 
   1166 	/* Return full data size if caller requested it. */
   1167 	if (size != NULL)
   1168 		*size = ueh.ueh_len;
   1169 
   1170 	/* Return data if the caller requested it. */
   1171 	if (uio != NULL) {
   1172 		/* Allow for offset into the attribute data. */
   1173 		uio->uio_offset = base_offset + sizeof(struct
   1174 		    ulfs_extattr_header);
   1175 
   1176 		/*
   1177 		 * Figure out maximum to transfer -- use buffer size and
   1178 		 * local data limit.
   1179 		 */
   1180 		len = MIN(uio->uio_resid, ueh.ueh_len);
   1181 		old_len = uio->uio_resid;
   1182 		uio->uio_resid = len;
   1183 
   1184 		error = VOP_READ(attribute->uele_backing_vnode, uio,
   1185 		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1186 		if (error)
   1187 			goto vopunlock_exit;
   1188 
   1189 		uio->uio_resid = old_len - (len - uio->uio_resid);
   1190 	}
   1191 
   1192  vopunlock_exit:
   1193 
   1194 	if (uio != NULL)
   1195 		uio->uio_offset = 0;
   1196 
   1197 	if (attribute->uele_backing_vnode != vp)
   1198 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1199 
   1200 	return (error);
   1201 }
   1202 
   1203 /*
   1204  * Vnode operation to list extended attribute for a vnode
   1205  */
   1206 int
   1207 ulfs_listextattr(struct vop_listextattr_args *ap)
   1208 /*
   1209 vop_listextattr {
   1210 	IN struct vnode *a_vp;
   1211 	IN int a_attrnamespace;
   1212 	INOUT struct uio *a_uio;
   1213 	OUT size_t *a_size;
   1214 	IN int flag;
   1215 	IN kauth_cred_t a_cred;
   1216 	struct proc *a_p;
   1217 };
   1218 */
   1219 {
   1220 	struct mount *mp = ap->a_vp->v_mount;
   1221 	struct ulfsmount *ump = VFSTOULFS(mp);
   1222 	int error;
   1223 
   1224 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1225 		return (EOPNOTSUPP);
   1226 
   1227 	ulfs_extattr_uepm_lock(ump);
   1228 
   1229 	error = ulfs_extattr_list(ap->a_vp, ap->a_attrnamespace,
   1230 	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
   1231 
   1232 	ulfs_extattr_uepm_unlock(ump);
   1233 
   1234 	return (error);
   1235 }
   1236 
   1237 /*
   1238  * Real work associated with retrieving list of attributes--assumes that
   1239  * the attribute lock has already been grabbed.
   1240  */
   1241 static int
   1242 ulfs_extattr_list(struct vnode *vp, int attrnamespace,
   1243     struct uio *uio, size_t *size, int flag,
   1244     kauth_cred_t cred, struct lwp *l)
   1245 {
   1246 	struct ulfs_extattr_list_entry *uele;
   1247 	struct ulfs_extattr_header ueh;
   1248 	struct mount *mp = vp->v_mount;
   1249 	struct ulfsmount *ump = VFSTOULFS(mp);
   1250 	size_t listsize = 0;
   1251 	int error = 0;
   1252 
   1253 	/*
   1254 	 * XXX: We can move this inside the loop and iterate on individual
   1255 	 *	attributes.
   1256 	 */
   1257 	error = internal_extattr_check_cred(vp, attrnamespace, "", cred,
   1258 	    VREAD);
   1259 	if (error)
   1260 		return (error);
   1261 
   1262 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
   1263 		unsigned char attrnamelen;
   1264 
   1265 		if (uele->uele_attrnamespace != attrnamespace)
   1266 			continue;
   1267 
   1268 		error = ulfs_extattr_get_header(vp, uele, &ueh, NULL);
   1269 		if (error == ENODATA)
   1270 			continue;
   1271 		if (error != 0)
   1272 			return error;
   1273 
   1274 		/*
   1275 		 * Don't need to get a lock on the backing file if
   1276 		 * the listattr is being applied to the backing file,
   1277 		 * as the lock is already held.
   1278 		 */
   1279 		if (uele->uele_backing_vnode != vp)
   1280 			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1281 
   1282 		/*
   1283 		 * +1 for trailing NUL (listxattr flavor)
   1284 		 *  or leading name length (extattr_list_file flavor)
   1285 	 	 */
   1286 		attrnamelen = strlen(uele->uele_attrname);
   1287 		listsize += attrnamelen + 1;
   1288 
   1289 		/* Return data if the caller requested it. */
   1290 		if (uio != NULL) {
   1291 			/*
   1292 			 * We support two flavors. Either NUL-terminated
   1293 			 * strings (a la listxattr), or non NUL-terminated,
   1294 			 * one byte length prefixed strings (for
   1295 			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
   1296 		 	 * that second behavior.
   1297 			 */
   1298 			if (flag & EXTATTR_LIST_LENPREFIX) {
   1299 				uint8_t len = (uint8_t)attrnamelen;
   1300 
   1301 				/* Copy leading name length */
   1302 				error = uiomove(&len, sizeof(len), uio);
   1303 				if (error != 0)
   1304 					break;
   1305 			} else {
   1306 				/* Include trailing NULL */
   1307 				attrnamelen++;
   1308 			}
   1309 
   1310 			error = uiomove(uele->uele_attrname,
   1311 					(size_t)attrnamelen, uio);
   1312 			if (error != 0)
   1313 				break;
   1314 		}
   1315 
   1316 		if (uele->uele_backing_vnode != vp)
   1317 			VOP_UNLOCK(uele->uele_backing_vnode);
   1318 
   1319 		if (error != 0)
   1320 			return error;
   1321 	}
   1322 
   1323 	if (uio != NULL)
   1324 		uio->uio_offset = 0;
   1325 
   1326 	/* Return full data size if caller requested it. */
   1327 	if (size != NULL)
   1328 		*size = listsize;
   1329 
   1330 	return 0;
   1331 }
   1332 
   1333 /*
   1334  * Vnode operation to remove a named attribute.
   1335  */
   1336 int
   1337 ulfs_deleteextattr(struct vop_deleteextattr_args *ap)
   1338 /*
   1339 vop_deleteextattr {
   1340 	IN struct vnode *a_vp;
   1341 	IN int a_attrnamespace;
   1342 	IN const char *a_name;
   1343 	IN kauth_cred_t a_cred;
   1344 };
   1345 */
   1346 {
   1347 	struct mount *mp = ap->a_vp->v_mount;
   1348 	struct ulfsmount *ump = VFSTOULFS(mp);
   1349 	int error;
   1350 
   1351 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1352 		return (EOPNOTSUPP);
   1353 
   1354 	ulfs_extattr_uepm_lock(ump);
   1355 
   1356 	error = ulfs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1357 	    ap->a_cred, curlwp);
   1358 
   1359 	ulfs_extattr_uepm_unlock(ump);
   1360 
   1361 	return (error);
   1362 }
   1363 
   1364 /*
   1365  * Vnode operation to set a named attribute.
   1366  */
   1367 int
   1368 ulfs_setextattr(struct vop_setextattr_args *ap)
   1369 /*
   1370 vop_setextattr {
   1371 	IN struct vnode *a_vp;
   1372 	IN int a_attrnamespace;
   1373 	IN const char *a_name;
   1374 	INOUT struct uio *a_uio;
   1375 	IN kauth_cred_t a_cred;
   1376 };
   1377 */
   1378 {
   1379 	struct mount *mp = ap->a_vp->v_mount;
   1380 	struct ulfsmount *ump = VFSTOULFS(mp);
   1381 	int error;
   1382 
   1383 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1384 		return (EOPNOTSUPP);
   1385 
   1386 	ulfs_extattr_uepm_lock(ump);
   1387 
   1388 	/*
   1389 	 * XXX: No longer a supported way to delete extended attributes.
   1390 	 */
   1391 	if (ap->a_uio == NULL) {
   1392 		ulfs_extattr_uepm_unlock(ump);
   1393 		return (EINVAL);
   1394 	}
   1395 
   1396 	error = ulfs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1397 	    ap->a_uio, ap->a_cred, curlwp);
   1398 
   1399 	ulfs_extattr_uepm_unlock(ump);
   1400 
   1401 	return (error);
   1402 }
   1403 
   1404 /*
   1405  * Real work associated with setting a vnode's extended attributes;
   1406  * assumes that the attribute lock has already been grabbed.
   1407  */
   1408 static int
   1409 ulfs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
   1410     struct uio *uio, kauth_cred_t cred, struct lwp *l)
   1411 {
   1412 	struct ulfs_extattr_list_entry *attribute;
   1413 	struct ulfs_extattr_header ueh;
   1414 	struct iovec local_aiov;
   1415 	struct uio local_aio;
   1416 	struct mount *mp = vp->v_mount;
   1417 	struct ulfsmount *ump = VFSTOULFS(mp);
   1418 	struct inode *ip = VTOI(vp);
   1419 	off_t base_offset;
   1420 	int error = 0, ioflag;
   1421 
   1422 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1423 		return (EROFS);
   1424 
   1425 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1426 		return (EINVAL);
   1427 
   1428 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1429 	    VWRITE);
   1430 	if (error)
   1431 		return (error);
   1432 
   1433 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1434 	if (!attribute) {
   1435 		error = ulfs_extattr_autocreate_attr(vp, attrnamespace,
   1436 						    name, l, &attribute);
   1437 		if (error == EEXIST) {
   1438 			/* Another thread raced us for backend creation */
   1439 			error = 0;
   1440 			attribute =
   1441 			    ulfs_extattr_find_attr(ump, attrnamespace, name);
   1442 		}
   1443 
   1444 		if (error || !attribute)
   1445 			return ENODATA;
   1446 	}
   1447 
   1448 	/*
   1449 	 * Early rejection of invalid offsets/length.
   1450 	 * Reject: any offset but 0 (replace)
   1451 	 *	 Any size greater than attribute size limit
   1452  	 */
   1453 	if (uio->uio_offset != 0 ||
   1454 	    uio->uio_resid > attribute->uele_fileheader.uef_size)
   1455 		return (ENXIO);
   1456 
   1457 	/*
   1458 	 * Find base offset of header in file based on file header size, and
   1459 	 * data header size + maximum data size, indexed by inode number.
   1460 	 */
   1461 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
   1462 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
   1463 	    attribute->uele_fileheader.uef_size);
   1464 
   1465 	/*
   1466 	 * Write out a data header for the data.
   1467 	 */
   1468 	ueh.ueh_len = ulfs_rw32((uint32_t) uio->uio_resid,
   1469 	    UELE_NEEDSWAP(attribute));
   1470 	ueh.ueh_flags = ulfs_rw32(ULFS_EXTATTR_ATTR_FLAG_INUSE,
   1471 				 UELE_NEEDSWAP(attribute));
   1472 	ueh.ueh_i_gen = ulfs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
   1473 	local_aiov.iov_base = &ueh;
   1474 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1475 	local_aio.uio_iov = &local_aiov;
   1476 	local_aio.uio_iovcnt = 1;
   1477 	local_aio.uio_rw = UIO_WRITE;
   1478 	local_aio.uio_offset = base_offset;
   1479 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1480 	UIO_SETUP_SYSSPACE(&local_aio);
   1481 
   1482 	/*
   1483 	 * Don't need to get a lock on the backing file if the setattr is
   1484 	 * being applied to the backing file, as the lock is already held.
   1485 	 */
   1486 	if (attribute->uele_backing_vnode != vp)
   1487 		vn_lock(attribute->uele_backing_vnode,
   1488 		    LK_EXCLUSIVE | LK_RETRY);
   1489 
   1490 	ioflag = IO_NODELOCKED;
   1491 	if (ulfs_extattr_sync)
   1492 		ioflag |= IO_SYNC;
   1493 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1494 	    ump->um_extattr.uepm_ucred);
   1495 	if (error)
   1496 		goto vopunlock_exit;
   1497 
   1498 	if (local_aio.uio_resid != 0) {
   1499 		error = ENXIO;
   1500 		goto vopunlock_exit;
   1501 	}
   1502 
   1503 	/*
   1504 	 * Write out user data.
   1505 	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
   1506 	 */
   1507 	uio->uio_offset = base_offset + sizeof(struct ulfs_extattr_header);
   1508 
   1509 	ioflag = IO_NODELOCKED;
   1510 	if (ulfs_extattr_sync)
   1511 		ioflag |= IO_SYNC;
   1512 	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
   1513 	    ump->um_extattr.uepm_ucred);
   1514 
   1515  vopunlock_exit:
   1516 	uio->uio_offset = 0;
   1517 
   1518 	if (attribute->uele_backing_vnode != vp)
   1519 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1520 
   1521 	return (error);
   1522 }
   1523 
   1524 /*
   1525  * Real work associated with removing an extended attribute from a vnode.
   1526  * Assumes the attribute lock has already been grabbed.
   1527  */
   1528 static int
   1529 ulfs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
   1530     kauth_cred_t cred, struct lwp *l)
   1531 {
   1532 	struct ulfs_extattr_list_entry *attribute;
   1533 	struct ulfs_extattr_header ueh;
   1534 	struct mount *mp = vp->v_mount;
   1535 	struct ulfsmount *ump = VFSTOULFS(mp);
   1536 	struct iovec local_aiov;
   1537 	struct uio local_aio;
   1538 	off_t base_offset;
   1539 	int error = 0, ioflag;
   1540 
   1541 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1542 		return (EROFS);
   1543 
   1544 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1545 		return (EINVAL);
   1546 
   1547 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1548 	    VWRITE);
   1549 	if (error)
   1550 		return (error);
   1551 
   1552 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1553 	if (!attribute)
   1554 		return (ENODATA);
   1555 
   1556 	/*
   1557 	 * Don't need to get a lock on the backing file if the getattr is
   1558 	 * being applied to the backing file, as the lock is already held.
   1559 	 */
   1560 	if (attribute->uele_backing_vnode != vp)
   1561 		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
   1562 
   1563 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1564 	if (error)
   1565 		goto vopunlock_exit;
   1566 
   1567 	/* Flag it as not in use. */
   1568 	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
   1569 	ueh.ueh_len = 0;		/* ...ditto... */
   1570 
   1571 	local_aiov.iov_base = &ueh;
   1572 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1573 	local_aio.uio_iov = &local_aiov;
   1574 	local_aio.uio_iovcnt = 1;
   1575 	local_aio.uio_rw = UIO_WRITE;
   1576 	local_aio.uio_offset = base_offset;
   1577 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1578 	UIO_SETUP_SYSSPACE(&local_aio);
   1579 
   1580 	ioflag = IO_NODELOCKED;
   1581 	if (ulfs_extattr_sync)
   1582 		ioflag |= IO_SYNC;
   1583 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1584 	    ump->um_extattr.uepm_ucred);
   1585 	if (error)
   1586 		goto vopunlock_exit;
   1587 
   1588 	if (local_aio.uio_resid != 0)
   1589 		error = ENXIO;
   1590 
   1591  vopunlock_exit:
   1592 	VOP_UNLOCK(attribute->uele_backing_vnode);
   1593 
   1594 	return (error);
   1595 }
   1596 
   1597 /*
   1598  * Called by ULFS when an inode is no longer active and should have its
   1599  * attributes stripped.
   1600  */
   1601 void
   1602 ulfs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
   1603 {
   1604 	struct ulfs_extattr_list_entry *uele;
   1605 	struct mount *mp = vp->v_mount;
   1606 	struct ulfsmount *ump = VFSTOULFS(mp);
   1607 
   1608 	/*
   1609 	 * In that case, we cannot lock. We should not have any active vnodes
   1610 	 * on the fs if this is not yet initialized but is going to be, so
   1611 	 * this can go unlocked.
   1612 	 */
   1613 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
   1614 		return;
   1615 
   1616 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1617 		return;
   1618 
   1619 	ulfs_extattr_uepm_lock(ump);
   1620 
   1621 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
   1622 		ulfs_extattr_rm(vp, uele->uele_attrnamespace,
   1623 		    uele->uele_attrname, lwp0.l_cred, l);
   1624 
   1625 	ulfs_extattr_uepm_unlock(ump);
   1626 }
   1627 
   1628 void
   1629 ulfs_extattr_init(void)
   1630 {
   1631 
   1632 }
   1633 
   1634 void
   1635 ulfs_extattr_done(void)
   1636 {
   1637 
   1638 }
   1639