Home | History | Annotate | Line # | Download | only in lfs
ulfs_extattr.c revision 1.16
      1 /*	$NetBSD: ulfs_extattr.c,v 1.16 2020/05/16 18:31:53 christos Exp $	*/
      2 /*  from NetBSD: ulfs_extattr.c,v 1.48 2016/11/09 05:08:35 dholland 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 1.16 2020/05/16 18:31:53 christos 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  * Per-FS attribute lock protecting attribute operations.
    112  * XXX Right now there is a lot of lock contention due to having a single
    113  * lock per-FS; really, this should be far more fine-grained.
    114  */
    115 static void
    116 ulfs_extattr_uepm_lock(struct ulfsmount *ump)
    117 {
    118 
    119 	/*
    120 	 * XXX This needs to be recursive for the following reasons:
    121 	 *   - it is taken in ulfs_extattr_vnode_inactive
    122 	 *   - which is called from VOP_INACTIVE
    123 	 *   - which can be triggered by any vrele, vput, or vn_close
    124 	 *   - several of these can happen while it's held
    125 	 */
    126 	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
    127 		ump->um_extattr.uepm_lockcnt++;
    128 		return;
    129 	}
    130 	mutex_enter(&ump->um_extattr.uepm_lock);
    131 }
    132 
    133 static void
    134 ulfs_extattr_uepm_unlock(struct ulfsmount *ump)
    135 {
    136 
    137 	if (ump->um_extattr.uepm_lockcnt != 0) {
    138 		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
    139 		ump->um_extattr.uepm_lockcnt--;
    140 		return;
    141 	}
    142 	mutex_exit(&ump->um_extattr.uepm_lock);
    143 }
    144 
    145 /*-
    146  * Determine whether the name passed is a valid name for an actual
    147  * attribute.
    148  *
    149  * Invalid currently consists of:
    150  *	 NULL pointer for attrname
    151  *	 zero-length attrname (used to retrieve application attribute list)
    152  */
    153 static int
    154 ulfs_extattr_valid_attrname(int attrnamespace, const char *attrname)
    155 {
    156 
    157 	if (attrname == NULL)
    158 		return (0);
    159 	if (strlen(attrname) == 0)
    160 		return (0);
    161 	return (1);
    162 }
    163 
    164 /*
    165  * Autocreate an attribute storage
    166  */
    167 static int
    168 ulfs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
    169     const char *attrname, struct lwp *l, struct ulfs_extattr_list_entry **uelep)
    170 {
    171 	struct mount *mp = vp->v_mount;
    172 	struct ulfsmount *ump = VFSTOULFS(mp);
    173 	struct vnode *backing_vp;
    174 	struct nameidata nd;
    175 	struct pathbuf *pb;
    176 	char *path;
    177 	struct ulfs_extattr_fileheader uef;
    178 	struct ulfs_extattr_list_entry *uele;
    179 	int error;
    180 
    181 	path = PNBUF_GET();
    182 
    183 	/*
    184 	 * We only support system and user namespace autocreation
    185 	 */
    186 	switch (attrnamespace) {
    187 	case EXTATTR_NAMESPACE_SYSTEM:
    188 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    189 			       mp->mnt_stat.f_mntonname,
    190 			       ULFS_EXTATTR_FSROOTSUBDIR,
    191 			       ULFS_EXTATTR_SUBDIR_SYSTEM,
    192 			       attrname);
    193 		break;
    194 	case EXTATTR_NAMESPACE_USER:
    195 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    196 			       mp->mnt_stat.f_mntonname,
    197 			       ULFS_EXTATTR_FSROOTSUBDIR,
    198 			       ULFS_EXTATTR_SUBDIR_USER,
    199 			       attrname);
    200 		break;
    201 	default:
    202 		PNBUF_PUT(path);
    203 		*uelep = NULL;
    204 		return EINVAL;
    205 		break;
    206 	}
    207 
    208 	/*
    209 	 * Release extended attribute mount lock, otherwise
    210 	 * we can deadlock with another thread that would lock
    211 	 * vp after we unlock it below, and call
    212 	 * ulfs_extattr_uepm_lock(ump), for instance
    213 	 * in ulfs_getextattr().
    214 	 */
    215 	ulfs_extattr_uepm_unlock(ump);
    216 
    217 	/*
    218 	 * XXX unlock/lock should only be done when setting extattr
    219 	 * on backing store or one of its parent directory
    220 	 * including root, but we always do it for now.
    221 	 */
    222 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    223 	VOP_UNLOCK(vp);
    224 
    225 	pb = pathbuf_create(path);
    226 	NDINIT(&nd, CREATE, LOCKPARENT, pb);
    227 
    228 	/*
    229 	 * Since we do not hold ulfs_extattr_uepm_lock anymore,
    230 	 * another thread may race with us for backend creation,
    231 	 * but only one can succeed here thanks to O_EXCL
    232 	 */
    233 	error = vn_open(&nd, O_CREAT|O_EXCL|O_RDWR, 0600);
    234 
    235 	/*
    236 	 * Reacquire the lock on the vnode
    237 	 */
    238 	KASSERT(VOP_ISLOCKED(vp) == 0);
    239 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    240 
    241 	ulfs_extattr_uepm_lock(ump);
    242 
    243 	if (error != 0) {
    244 		pathbuf_destroy(pb);
    245 		PNBUF_PUT(path);
    246 		*uelep = NULL;
    247 		return error;
    248 	}
    249 
    250 	KASSERT(nd.ni_vp != NULL);
    251 	KASSERT(VOP_ISLOCKED(nd.ni_vp) == LK_EXCLUSIVE);
    252 	KASSERT(VOP_ISLOCKED(nd.ni_dvp) == 0);
    253 
    254 	/*
    255  	 * backing_vp is the backing store.
    256 	 */
    257 	backing_vp = nd.ni_vp;
    258 	pathbuf_destroy(pb);
    259 	PNBUF_PUT(path);
    260 
    261 	uef.uef_magic = ULFS_EXTATTR_MAGIC;
    262 	uef.uef_version = ULFS_EXTATTR_VERSION;
    263 	uef.uef_size = ulfs_extattr_autocreate;
    264 
    265 	error = vn_rdwr(UIO_WRITE, backing_vp, &uef, sizeof(uef), 0,
    266 		        UIO_SYSSPACE, IO_NODELOCKED|IO_APPEND,
    267 			l->l_cred, NULL, l);
    268 
    269 	VOP_UNLOCK(backing_vp);
    270 
    271 	if (error != 0) {
    272 		printf("%s: write uef header failed for %s, error = %d\n",
    273 		       __func__, attrname, error);
    274 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    275 		*uelep = NULL;
    276 		return error;
    277 	}
    278 
    279 	/*
    280 	 * Now enable attribute.
    281 	 */
    282 	error = ulfs_extattr_enable(ump,attrnamespace, attrname, backing_vp, l);
    283 	KASSERT(VOP_ISLOCKED(backing_vp) == 0);
    284 
    285 	if (error != 0) {
    286 		printf("%s: enable %s failed, error %d\n",
    287 		       __func__, attrname, error);
    288 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    289 		*uelep = NULL;
    290 		return error;
    291 	}
    292 
    293 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    294 	if (uele == NULL) {
    295 		printf("%s: atttribute %s created but not found!\n",
    296 		       __func__, attrname);
    297 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    298 		*uelep = NULL;
    299 		return ESRCH; /* really internal error */
    300 	}
    301 
    302 	printf("%s: EA backing store autocreated for %s\n",
    303 	       mp->mnt_stat.f_mntonname, attrname);
    304 
    305 	*uelep = uele;
    306 	return 0;
    307 }
    308 
    309 /*
    310  * Locate an attribute given a name and mountpoint.
    311  * Must be holding uepm lock for the mount point.
    312  */
    313 static struct ulfs_extattr_list_entry *
    314 ulfs_extattr_find_attr(struct ulfsmount *ump, int attrnamespace,
    315     const char *attrname)
    316 {
    317 	struct ulfs_extattr_list_entry *search_attribute;
    318 
    319 	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
    320 	    search_attribute != NULL;
    321 	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
    322 		if (!(strncmp(attrname, search_attribute->uele_attrname,
    323 		    ULFS_EXTATTR_MAXEXTATTRNAME)) &&
    324 		    (attrnamespace == search_attribute->uele_attrnamespace)) {
    325 			return (search_attribute);
    326 		}
    327 	}
    328 
    329 	return (0);
    330 }
    331 
    332 /*
    333  * Initialize per-FS structures supporting extended attributes.  Do not
    334  * start extended attributes yet.
    335  */
    336 void
    337 ulfs_extattr_uepm_init(struct ulfs_extattr_per_mount *uepm)
    338 {
    339 
    340 	uepm->uepm_flags = 0;
    341 	uepm->uepm_lockcnt = 0;
    342 
    343 	LIST_INIT(&uepm->uepm_list);
    344 	mutex_init(&uepm->uepm_lock, MUTEX_DEFAULT, IPL_NONE);
    345 	uepm->uepm_flags |= ULFS_EXTATTR_UEPM_INITIALIZED;
    346 }
    347 
    348 /*
    349  * Destroy per-FS structures supporting extended attributes.  Assumes
    350  * that EAs have already been stopped, and will panic if not.
    351  */
    352 void
    353 ulfs_extattr_uepm_destroy(struct ulfs_extattr_per_mount *uepm)
    354 {
    355 
    356 	if (!(uepm->uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    357 		panic("ulfs_extattr_uepm_destroy: not initialized");
    358 
    359 	if ((uepm->uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
    360 		panic("ulfs_extattr_uepm_destroy: called while still started");
    361 
    362 	/*
    363 	 * It's not clear that either order for the next three lines is
    364 	 * ideal, and it should never be a problem if this is only called
    365 	 * during unmount, and with vfs_busy().
    366 	 */
    367 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    368 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_INITIALIZED;
    369 	mutex_destroy(&uepm->uepm_lock);
    370 }
    371 
    372 /*
    373  * Start extended attribute support on an FS.
    374  */
    375 int
    376 ulfs_extattr_start(struct mount *mp, struct lwp *l)
    377 {
    378 	struct ulfsmount *ump;
    379 	int error = 0;
    380 
    381 	ump = VFSTOULFS(mp);
    382 
    383 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    384 		ulfs_extattr_uepm_init(&ump->um_extattr);
    385 
    386 	ulfs_extattr_uepm_lock(ump);
    387 
    388 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED)) {
    389 		error = EOPNOTSUPP;
    390 		goto unlock;
    391 	}
    392 	if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
    393 		error = EBUSY;
    394 		goto unlock;
    395 	}
    396 
    397 	ump->um_extattr.uepm_flags |= ULFS_EXTATTR_UEPM_STARTED;
    398 
    399 	ump->um_extattr.uepm_ucred = l->l_cred;
    400 	kauth_cred_hold(ump->um_extattr.uepm_ucred);
    401 
    402  unlock:
    403 	ulfs_extattr_uepm_unlock(ump);
    404 
    405 	return (error);
    406 }
    407 
    408 /*
    409  * Helper routine: given a locked parent directory and filename, return
    410  * the locked vnode of the inode associated with the name.  Will not
    411  * follow symlinks, may return any type of vnode.  Lock on parent will
    412  * be released even in the event of a failure.  In the event that the
    413  * target is the parent (i.e., "."), there will be two references and
    414  * one lock, requiring the caller to possibly special-case.
    415  */
    416 static int
    417 ulfs_extattr_lookup(struct vnode *start_dvp, int lockparent, const char *dirname,
    418     struct vnode **vp, struct lwp *l)
    419 {
    420 	struct vop_lookup_v2_args vargs;
    421 	struct componentname cnp;
    422 	struct vnode *target_vp;
    423 	char *pnbuf;
    424 	int error;
    425 
    426 	KASSERT(VOP_ISLOCKED(start_dvp) == LK_EXCLUSIVE);
    427 
    428 	pnbuf = PNBUF_GET();
    429 
    430 	memset(&cnp, 0, sizeof(cnp));
    431 	cnp.cn_nameiop = LOOKUP;
    432 	cnp.cn_flags = ISLASTCN | lockparent;
    433 	cnp.cn_cred = l->l_cred;
    434 	cnp.cn_nameptr = pnbuf;
    435 	error = copystr(dirname, pnbuf, MAXPATHLEN, &cnp.cn_namelen);
    436 	if (error) {
    437 		if (lockparent == 0) {
    438 			VOP_UNLOCK(start_dvp);
    439 		}
    440 		PNBUF_PUT(pnbuf);
    441 		printf("ulfs_extattr_lookup: copystr failed\n");
    442 		return (error);
    443 	}
    444 	cnp.cn_namelen--;	/* trim nul termination */
    445 	vargs.a_desc = NULL;
    446 	vargs.a_dvp = start_dvp;
    447 	vargs.a_vpp = &target_vp;
    448 	vargs.a_cnp = &cnp;
    449 	error = ulfs_lookup(&vargs);
    450 	PNBUF_PUT(pnbuf);
    451 	if (error) {
    452 		if (lockparent == 0) {
    453 			VOP_UNLOCK(start_dvp);
    454 		}
    455 		return (error);
    456 	}
    457 #if 0
    458 	if (target_vp == start_dvp)
    459 		panic("ulfs_extattr_lookup: target_vp == start_dvp");
    460 #endif
    461 
    462 	if (target_vp != start_dvp) {
    463 		error = vn_lock(target_vp, LK_EXCLUSIVE);
    464 		if (lockparent == 0)
    465 			VOP_UNLOCK(start_dvp);
    466 		if (error) {
    467 			vrele(target_vp);
    468 			return error;
    469 		}
    470 	}
    471 
    472 	KASSERT(VOP_ISLOCKED(target_vp) == LK_EXCLUSIVE);
    473 	*vp = target_vp;
    474 	return (0);
    475 }
    476 
    477 /*
    478  * Enable an EA using the passed filesystem, backing vnode, attribute name,
    479  * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
    480  * to be locked when passed in.  The vnode will be returned unlocked,
    481  * regardless of success/failure of the function.  As a result, the caller
    482  * will always need to vrele(), but not vput().
    483  */
    484 static int
    485 ulfs_extattr_enable_with_open(struct ulfsmount *ump, struct vnode *vp,
    486     int attrnamespace, const char *attrname, struct lwp *l)
    487 {
    488 	int error;
    489 
    490 	error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred);
    491 	if (error) {
    492 		printf("ulfs_extattr_enable_with_open.VOP_OPEN(): failed "
    493 		    "with %d\n", error);
    494 		VOP_UNLOCK(vp);
    495 		return (error);
    496 	}
    497 
    498 	mutex_enter(vp->v_interlock);
    499 	vp->v_writecount++;
    500 	mutex_exit(vp->v_interlock);
    501 
    502 	vref(vp);
    503 
    504 	VOP_UNLOCK(vp);
    505 
    506 	error = ulfs_extattr_enable(ump, attrnamespace, attrname, vp, l);
    507 	if (error != 0)
    508 		vn_close(vp, FREAD|FWRITE, l->l_cred);
    509 	return (error);
    510 }
    511 
    512 /*
    513  * Given a locked directory vnode, iterate over the names in the directory
    514  * and use ulfs_extattr_lookup() to retrieve locked vnodes of potential
    515  * attribute files.  Then invoke ulfs_extattr_enable_with_open() on each
    516  * to attempt to start the attribute.  Leaves the directory locked on
    517  * exit.
    518  */
    519 static int
    520 ulfs_extattr_iterate_directory(struct ulfsmount *ump, struct vnode *dvp,
    521     int attrnamespace, struct lwp *l)
    522 {
    523 	struct vop_readdir_args vargs;
    524 	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
    525 	struct dirent *dp, *edp;
    526 	struct vnode *attr_vp;
    527 	struct uio auio;
    528 	struct iovec aiov;
    529 	char *dirbuf;
    530 	int error, eofflag = 0;
    531 
    532 	if (dvp->v_type != VDIR)
    533 		return (ENOTDIR);
    534 
    535 	dirbuf = kmem_alloc(LFS_DIRBLKSIZ, KM_SLEEP);
    536 
    537 	auio.uio_iov = &aiov;
    538 	auio.uio_iovcnt = 1;
    539 	auio.uio_rw = UIO_READ;
    540 	auio.uio_offset = 0;
    541 	UIO_SETUP_SYSSPACE(&auio);
    542 
    543 	vargs.a_desc = NULL;
    544 	vargs.a_vp = dvp;
    545 	vargs.a_uio = &auio;
    546 	vargs.a_cred = l->l_cred;
    547 	vargs.a_eofflag = &eofflag;
    548 	vargs.a_ncookies = NULL;
    549 	vargs.a_cookies = NULL;
    550 
    551 	while (!eofflag) {
    552 		auio.uio_resid = LFS_DIRBLKSIZ;
    553 		aiov.iov_base = dirbuf;
    554 		aiov.iov_len = LFS_DIRBLKSIZ;
    555 		error = ulfs_readdir(&vargs);
    556 		if (error) {
    557 			printf("ulfs_extattr_iterate_directory: ulfs_readdir "
    558 			    "%d\n", error);
    559 			return (error);
    560 		}
    561 
    562 		/*
    563 		 * XXXRW: While in LFS, we always get LFS_DIRBLKSIZ returns from
    564 		 * the directory code on success, on other file systems this
    565 		 * may not be the case.  For portability, we should check the
    566 		 * read length on return from ulfs_readdir().
    567 		 */
    568 		edp = (struct dirent *)&dirbuf[LFS_DIRBLKSIZ];
    569 		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
    570 			if (dp->d_reclen == 0)
    571 				break;
    572 			/* Skip "." and ".." */
    573 			if (dp->d_name[0] == '.' &&
    574 			    (dp->d_name[1] == '\0' ||
    575 			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
    576 				goto next;
    577 			error = ulfs_extattr_lookup(dvp, LOCKPARENT,
    578 			    dp->d_name, &attr_vp, l);
    579 			if (error == ENOENT) {
    580 				goto next; /* keep silent */
    581 			} else if (error) {
    582 				printf("ulfs_extattr_iterate_directory: lookup "
    583 				    "%s %d\n", dp->d_name, error);
    584 			} else if (attr_vp == dvp) {
    585 				vrele(attr_vp);
    586 			} else if (attr_vp->v_type != VREG) {
    587 				vput(attr_vp);
    588 			} else {
    589 				error = ulfs_extattr_enable_with_open(ump,
    590 				    attr_vp, attrnamespace, dp->d_name, l);
    591 				vrele(attr_vp);
    592 				if (error) {
    593 					printf("ulfs_extattr_iterate_directory: "
    594 					    "enable %s %d\n", dp->d_name,
    595 					    error);
    596 				} else if (bootverbose) {
    597 					printf("%s: EA %s loaded\n",
    598 					       sbp->f_mntonname, dp->d_name);
    599 				}
    600 			}
    601  next:
    602 			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
    603 			if (dp >= edp)
    604 				break;
    605 		}
    606 	}
    607 	kmem_free(dirbuf, LFS_DIRBLKSIZ);
    608 
    609 	return (0);
    610 }
    611 
    612 /*
    613  * Auto-start of extended attributes, to be executed (optionally) at
    614  * mount-time.
    615  */
    616 int
    617 ulfs_extattr_autostart(struct mount *mp, struct lwp *l)
    618 {
    619 	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
    620 	int error;
    621 
    622 	/*
    623 	 * Does ULFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
    624 	 * If so, automatically start EA's.
    625 	 */
    626 	error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp);
    627 	if (error) {
    628 		printf("ulfs_extattr_autostart.VFS_ROOT() returned %d\n",
    629 		    error);
    630 		return (error);
    631 	}
    632 
    633 	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    634 
    635 	error = ulfs_extattr_lookup(rvp, 0,
    636 	    ULFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
    637 	if (error) {
    638 		/* rvp ref'd but now unlocked */
    639 		KASSERT(VOP_ISLOCKED(rvp) == 0);
    640 		vrele(rvp);
    641 		return (error);
    642 	}
    643 	if (rvp == attr_dvp) {
    644 		/* Should never happen. */
    645 		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    646 		vrele(attr_dvp);
    647 		vput(rvp);
    648 		return (EINVAL);
    649 	}
    650 	KASSERT(VOP_ISLOCKED(rvp) == 0);
    651 	vrele(rvp);
    652 
    653 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    654 
    655 	if (attr_dvp->v_type != VDIR) {
    656 		printf("ulfs_extattr_autostart: %s != VDIR\n",
    657 		    ULFS_EXTATTR_FSROOTSUBDIR);
    658 		goto return_vput_attr_dvp;
    659 	}
    660 
    661 	error = ulfs_extattr_start(mp, l);
    662 	if (error) {
    663 		printf("ulfs_extattr_autostart: ulfs_extattr_start failed (%d)\n",
    664 		    error);
    665 		goto return_vput_attr_dvp;
    666 	}
    667 
    668 	/*
    669 	 * Look for two subdirectories: ULFS_EXTATTR_SUBDIR_SYSTEM,
    670 	 * ULFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
    671 	 * and start with appropriate type.  Failures in either don't
    672 	 * result in an over-all failure.  attr_dvp is left locked to
    673 	 * be cleaned up on exit.
    674 	 */
    675 	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    676 	    ULFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, l);
    677 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    678 	if (error == 0) {
    679 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    680 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    681 		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, l);
    682 		if (error)
    683 			printf("ulfs_extattr_iterate_directory returned %d\n",
    684 			    error);
    685 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    686 		vput(attr_system_dvp);
    687 	}
    688 
    689 	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    690 	    ULFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, l);
    691 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    692 	if (error == 0) {
    693 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    694 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    695 		    attr_user_dvp, EXTATTR_NAMESPACE_USER, l);
    696 		if (error)
    697 			printf("ulfs_extattr_iterate_directory returned %d\n",
    698 			    error);
    699 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    700 		vput(attr_user_dvp);
    701 	}
    702 
    703 	/* Mask startup failures in sub-directories. */
    704 	error = 0;
    705 
    706  return_vput_attr_dvp:
    707 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    708 	vput(attr_dvp);
    709 
    710 	return (error);
    711 }
    712 
    713 /*
    714  * Stop extended attribute support on an FS.
    715  */
    716 void
    717 ulfs_extattr_stop(struct mount *mp, struct lwp *l)
    718 {
    719 	struct ulfs_extattr_list_entry *uele;
    720 	struct ulfsmount *ump = VFSTOULFS(mp);
    721 
    722 	ulfs_extattr_uepm_lock(ump);
    723 
    724 	/*
    725 	 * If we haven't been started, no big deal.  Just short-circuit
    726 	 * the processing work.
    727 	 */
    728 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    729 		goto unlock;
    730 	}
    731 
    732 	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
    733 		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
    734 		ulfs_extattr_disable(ump, uele->uele_attrnamespace,
    735 		    uele->uele_attrname, l);
    736 	}
    737 
    738 	ump->um_extattr.uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    739 
    740 	kauth_cred_free(ump->um_extattr.uepm_ucred);
    741 	ump->um_extattr.uepm_ucred = NULL;
    742 
    743  unlock:
    744 	ulfs_extattr_uepm_unlock(ump);
    745 }
    746 
    747 /*
    748  * Enable a named attribute on the specified filesystem; provide an
    749  * unlocked backing vnode to hold the attribute data.
    750  */
    751 static int
    752 ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
    753     const char *attrname, struct vnode *backing_vnode, struct lwp *l)
    754 {
    755 	struct ulfs_extattr_list_entry *attribute;
    756 	struct iovec aiov;
    757 	struct uio auio;
    758 	int error = 0;
    759 
    760 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    761 		return (EINVAL);
    762 	if (backing_vnode->v_type != VREG)
    763 		return (EINVAL);
    764 
    765 	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
    766 
    767 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    768 		error = EOPNOTSUPP;
    769 		goto free_exit;
    770 	}
    771 
    772 	if (ulfs_extattr_find_attr(ump, attrnamespace, attrname)) {
    773 		error = EEXIST;
    774 		goto free_exit;
    775 	}
    776 
    777 	strncpy(attribute->uele_attrname, attrname,
    778 	    ULFS_EXTATTR_MAXEXTATTRNAME);
    779 	attribute->uele_attrnamespace = attrnamespace;
    780 	memset(&attribute->uele_fileheader, 0,
    781 	    sizeof(struct ulfs_extattr_fileheader));
    782 
    783 	attribute->uele_backing_vnode = backing_vnode;
    784 
    785 	auio.uio_iov = &aiov;
    786 	auio.uio_iovcnt = 1;
    787 	aiov.iov_base = (void *) &attribute->uele_fileheader;
    788 	aiov.iov_len = sizeof(struct ulfs_extattr_fileheader);
    789 	auio.uio_resid = sizeof(struct ulfs_extattr_fileheader);
    790 	auio.uio_offset = (off_t) 0;
    791 	auio.uio_rw = UIO_READ;
    792 	UIO_SETUP_SYSSPACE(&auio);
    793 
    794 	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
    795 	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
    796 	    ump->um_extattr.uepm_ucred);
    797 
    798 	if (error)
    799 		goto unlock_free_exit;
    800 
    801 	if (auio.uio_resid != 0) {
    802 		printf("ulfs_extattr_enable: malformed attribute header\n");
    803 		error = EINVAL;
    804 		goto unlock_free_exit;
    805 	}
    806 
    807 	/*
    808 	 * Try to determine the byte order of the attribute file.
    809 	 */
    810 	if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    811 		attribute->uele_flags |= UELE_F_NEEDSWAP;
    812 		attribute->uele_fileheader.uef_magic =
    813 		    ulfs_rw32(attribute->uele_fileheader.uef_magic,
    814 			     UELE_NEEDSWAP(attribute));
    815 		if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    816 			printf("ulfs_extattr_enable: invalid attribute header "
    817 			       "magic\n");
    818 			error = EINVAL;
    819 			goto unlock_free_exit;
    820 		}
    821 	}
    822 	attribute->uele_fileheader.uef_version =
    823 	    ulfs_rw32(attribute->uele_fileheader.uef_version,
    824 		     UELE_NEEDSWAP(attribute));
    825 	attribute->uele_fileheader.uef_size =
    826 	    ulfs_rw32(attribute->uele_fileheader.uef_size,
    827 		     UELE_NEEDSWAP(attribute));
    828 
    829 	if (attribute->uele_fileheader.uef_version != ULFS_EXTATTR_VERSION) {
    830 		printf("ulfs_extattr_enable: incorrect attribute header "
    831 		    "version\n");
    832 		error = EINVAL;
    833 		goto unlock_free_exit;
    834 	}
    835 
    836 	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
    837 	    uele_entries);
    838 
    839 	VOP_UNLOCK(backing_vnode);
    840 	return (0);
    841 
    842  unlock_free_exit:
    843 	VOP_UNLOCK(backing_vnode);
    844 
    845  free_exit:
    846 	kmem_free(attribute, sizeof(*attribute));
    847 	return (error);
    848 }
    849 
    850 /*
    851  * Disable extended attribute support on an FS.
    852  */
    853 static int
    854 ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
    855     const char *attrname, struct lwp *l)
    856 {
    857 	struct ulfs_extattr_list_entry *uele;
    858 	int error = 0;
    859 
    860 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    861 		return (EINVAL);
    862 
    863 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    864 	if (!uele)
    865 		return (ENODATA);
    866 
    867 	LIST_REMOVE(uele, uele_entries);
    868 
    869 	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
    870 	    l->l_cred);
    871 
    872 	kmem_free(uele, sizeof(*uele));
    873 
    874 	return (error);
    875 }
    876 
    877 /*
    878  * VFS call to manage extended attributes in ULFS.  If filename_vp is
    879  * non-NULL, it must be passed in locked, and regardless of errors in
    880  * processing, will be unlocked.
    881  */
    882 int
    883 ulfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    884     int attrnamespace, const char *attrname)
    885 {
    886 	struct lwp *l = curlwp;
    887 	struct ulfsmount *ump = VFSTOULFS(mp);
    888 	int error;
    889 
    890 	/*
    891 	 * Only privileged processes can configure extended attributes.
    892 	 */
    893 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_FS_EXTATTR,
    894 	    0, mp, NULL, NULL);
    895 	if (error) {
    896 		if (filename_vp != NULL)
    897 			VOP_UNLOCK(filename_vp);
    898 		return (error);
    899 	}
    900 
    901 	switch(cmd) {
    902 	case ULFS_EXTATTR_CMD_START:
    903 		if (filename_vp != NULL) {
    904 			VOP_UNLOCK(filename_vp);
    905 			return (EINVAL);
    906 		}
    907 		if (attrname != NULL)
    908 			return (EINVAL);
    909 
    910 		error = ulfs_extattr_autostart(mp, l);
    911 		return (error);
    912 
    913 	case ULFS_EXTATTR_CMD_STOP:
    914 		if (filename_vp != NULL) {
    915 			VOP_UNLOCK(filename_vp);
    916 			return (EINVAL);
    917 		}
    918 		if (attrname != NULL)
    919 			return (EINVAL);
    920 
    921 		ulfs_extattr_stop(mp, l);
    922 		return (0);
    923 
    924 	case ULFS_EXTATTR_CMD_ENABLE:
    925 		if (filename_vp == NULL)
    926 			return (EINVAL);
    927 		if (attrname == NULL) {
    928 			VOP_UNLOCK(filename_vp);
    929 			return (EINVAL);
    930 		}
    931 
    932 		/*
    933 		 * ulfs_extattr_enable_with_open() will always unlock the
    934 		 * vnode, regardless of failure.
    935 		 */
    936 		ulfs_extattr_uepm_lock(ump);
    937 		error = ulfs_extattr_enable_with_open(ump, filename_vp,
    938 		    attrnamespace, attrname, l);
    939 		ulfs_extattr_uepm_unlock(ump);
    940 		return (error);
    941 
    942 	case ULFS_EXTATTR_CMD_DISABLE:
    943 		if (filename_vp != NULL) {
    944 			VOP_UNLOCK(filename_vp);
    945 			return (EINVAL);
    946 		}
    947 		if (attrname == NULL)
    948 			return (EINVAL);
    949 
    950 		ulfs_extattr_uepm_lock(ump);
    951 		error = ulfs_extattr_disable(ump, attrnamespace, attrname, l);
    952 		ulfs_extattr_uepm_unlock(ump);
    953 		return (error);
    954 
    955 	default:
    956 		return (EINVAL);
    957 	}
    958 }
    959 
    960 /*
    961  * Read extended attribute header for a given vnode and attribute.
    962  * Backing vnode should be locked and unlocked by caller.
    963  */
    964 static int
    965 ulfs_extattr_get_header(struct vnode *vp, struct ulfs_extattr_list_entry *uele,
    966     struct ulfs_extattr_header *ueh, off_t *bap)
    967 {
    968 	struct mount *mp = vp->v_mount;
    969 	struct ulfsmount *ump = VFSTOULFS(mp);
    970 	struct inode *ip = VTOI(vp);
    971 	off_t base_offset;
    972 	struct iovec aiov;
    973 	struct uio aio;
    974 	int error;
    975 
    976 	/*
    977 	 * Find base offset of header in file based on file header size, and
    978 	 * data header size + maximum data size, indexed by inode number.
    979 	 */
    980 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
    981 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
    982 	    uele->uele_fileheader.uef_size);
    983 
    984 	/*
    985 	 * Read in the data header to see if the data is defined, and if so
    986 	 * how much.
    987 	 */
    988 	memset(ueh, 0, sizeof(struct ulfs_extattr_header));
    989 	aiov.iov_base = ueh;
    990 	aiov.iov_len = sizeof(struct ulfs_extattr_header);
    991 	aio.uio_iov = &aiov;
    992 	aio.uio_iovcnt = 1;
    993 	aio.uio_rw = UIO_READ;
    994 	aio.uio_offset = base_offset;
    995 	aio.uio_resid = sizeof(struct ulfs_extattr_header);
    996 	UIO_SETUP_SYSSPACE(&aio);
    997 
    998 	error = VOP_READ(uele->uele_backing_vnode, &aio,
    999 	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1000 	if (error)
   1001 		return error;
   1002 
   1003 	/*
   1004 	 * Attribute headers are kept in file system byte order.
   1005 	 * XXX What about the blob of data?
   1006 	 */
   1007 	ueh->ueh_flags = ulfs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
   1008 	ueh->ueh_len   = ulfs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
   1009 	ueh->ueh_i_gen = ulfs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
   1010 
   1011 	/* Defined? */
   1012 	if ((ueh->ueh_flags & ULFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
   1013 		return ENODATA;
   1014 
   1015 	/* Valid for the current inode generation? */
   1016 	if (ueh->ueh_i_gen != ip->i_gen) {
   1017 		/*
   1018 		 * The inode itself has a different generation number
   1019 		 * than the uele data.  For now, the best solution
   1020 		 * is to coerce this to undefined, and let it get cleaned
   1021 		 * up by the next write or extattrctl clean.
   1022 		 */
   1023 		printf("%s (%s): inode gen inconsistency (%u, %jd)\n",
   1024 		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
   1025 		       (intmax_t)ip->i_gen);
   1026 		return ENODATA;
   1027 	}
   1028 
   1029 	/* Local size consistency check. */
   1030 	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
   1031 		return ENXIO;
   1032 
   1033 	/* Return base offset */
   1034 	if (bap != NULL)
   1035 		*bap = base_offset;
   1036 
   1037 	return 0;
   1038 }
   1039 
   1040 /*
   1041  * Vnode operation to retrieve a named extended attribute.
   1042  */
   1043 int
   1044 ulfs_getextattr(struct vop_getextattr_args *ap)
   1045 /*
   1046 vop_getextattr {
   1047 	IN struct vnode *a_vp;
   1048 	IN int a_attrnamespace;
   1049 	IN const char *a_name;
   1050 	INOUT struct uio *a_uio;
   1051 	OUT size_t *a_size;
   1052 	IN kauth_cred_t a_cred;
   1053 };
   1054 */
   1055 {
   1056 	struct mount *mp = ap->a_vp->v_mount;
   1057 	struct ulfsmount *ump = VFSTOULFS(mp);
   1058 	int error;
   1059 
   1060 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1061 		return (EOPNOTSUPP);
   1062 
   1063 	ulfs_extattr_uepm_lock(ump);
   1064 
   1065 	error = ulfs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1066 	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
   1067 
   1068 	ulfs_extattr_uepm_unlock(ump);
   1069 
   1070 	return (error);
   1071 }
   1072 
   1073 /*
   1074  * Real work associated with retrieving a named attribute--assumes that
   1075  * the attribute lock has already been grabbed.
   1076  */
   1077 static int
   1078 ulfs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
   1079     struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
   1080 {
   1081 	struct ulfs_extattr_list_entry *attribute;
   1082 	struct ulfs_extattr_header ueh;
   1083 	struct mount *mp = vp->v_mount;
   1084 	struct ulfsmount *ump = VFSTOULFS(mp);
   1085 	off_t base_offset;
   1086 	size_t len, old_len;
   1087 	int error = 0;
   1088 
   1089 	if (strlen(name) == 0)
   1090 		return (EINVAL);
   1091 
   1092 	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
   1093 	if (error)
   1094 		return (error);
   1095 
   1096 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1097 	if (!attribute)
   1098 		return (ENODATA);
   1099 
   1100 	/*
   1101 	 * Allow only offsets of zero to encourage the read/replace
   1102 	 * extended attribute semantic.  Otherwise we can't guarantee
   1103 	 * atomicity, as we don't provide locks for extended attributes.
   1104 	 */
   1105 	if (uio != NULL && uio->uio_offset != 0)
   1106 		return (ENXIO);
   1107 
   1108 	/*
   1109 	 * Don't need to get a lock on the backing file if the getattr is
   1110 	 * being applied to the backing file, as the lock is already held.
   1111 	 */
   1112 	if (attribute->uele_backing_vnode != vp)
   1113 		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1114 
   1115 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1116 	if (error)
   1117 		goto vopunlock_exit;
   1118 
   1119 	/* Return full data size if caller requested it. */
   1120 	if (size != NULL)
   1121 		*size = ueh.ueh_len;
   1122 
   1123 	/* Return data if the caller requested it. */
   1124 	if (uio != NULL) {
   1125 		/* Allow for offset into the attribute data. */
   1126 		uio->uio_offset = base_offset + sizeof(struct
   1127 		    ulfs_extattr_header);
   1128 
   1129 		/*
   1130 		 * Figure out maximum to transfer -- use buffer size and
   1131 		 * local data limit.
   1132 		 */
   1133 		len = MIN(uio->uio_resid, ueh.ueh_len);
   1134 		old_len = uio->uio_resid;
   1135 		uio->uio_resid = len;
   1136 
   1137 		error = VOP_READ(attribute->uele_backing_vnode, uio,
   1138 		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1139 		if (error)
   1140 			goto vopunlock_exit;
   1141 
   1142 		uio->uio_resid = old_len - (len - uio->uio_resid);
   1143 	}
   1144 
   1145  vopunlock_exit:
   1146 
   1147 	if (uio != NULL)
   1148 		uio->uio_offset = 0;
   1149 
   1150 	if (attribute->uele_backing_vnode != vp)
   1151 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1152 
   1153 	return (error);
   1154 }
   1155 
   1156 /*
   1157  * Vnode operation to list extended attribute for a vnode
   1158  */
   1159 int
   1160 ulfs_listextattr(struct vop_listextattr_args *ap)
   1161 /*
   1162 vop_listextattr {
   1163 	IN struct vnode *a_vp;
   1164 	IN int a_attrnamespace;
   1165 	INOUT struct uio *a_uio;
   1166 	OUT size_t *a_size;
   1167 	IN int flag;
   1168 	IN kauth_cred_t a_cred;
   1169 	struct proc *a_p;
   1170 };
   1171 */
   1172 {
   1173 	struct mount *mp = ap->a_vp->v_mount;
   1174 	struct ulfsmount *ump = VFSTOULFS(mp);
   1175 	int error;
   1176 
   1177 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1178 		return (EOPNOTSUPP);
   1179 
   1180 	ulfs_extattr_uepm_lock(ump);
   1181 
   1182 	error = ulfs_extattr_list(ap->a_vp, ap->a_attrnamespace,
   1183 	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
   1184 
   1185 	ulfs_extattr_uepm_unlock(ump);
   1186 
   1187 	return (error);
   1188 }
   1189 
   1190 /*
   1191  * Real work associated with retrieving list of attributes--assumes that
   1192  * the attribute lock has already been grabbed.
   1193  */
   1194 static int
   1195 ulfs_extattr_list(struct vnode *vp, int attrnamespace,
   1196     struct uio *uio, size_t *size, int flag,
   1197     kauth_cred_t cred, struct lwp *l)
   1198 {
   1199 	struct ulfs_extattr_list_entry *uele;
   1200 	struct ulfs_extattr_header ueh;
   1201 	struct mount *mp = vp->v_mount;
   1202 	struct ulfsmount *ump = VFSTOULFS(mp);
   1203 	size_t listsize = 0;
   1204 	int error = 0;
   1205 
   1206 	/*
   1207 	 * XXX: We can move this inside the loop and iterate on individual
   1208 	 *	attributes.
   1209 	 */
   1210 	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
   1211 	if (error)
   1212 		return (error);
   1213 
   1214 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
   1215 		unsigned char attrnamelen;
   1216 
   1217 		if (uele->uele_attrnamespace != attrnamespace)
   1218 			continue;
   1219 
   1220 		error = ulfs_extattr_get_header(vp, uele, &ueh, NULL);
   1221 		if (error == ENODATA)
   1222 			continue;
   1223 		if (error != 0)
   1224 			return error;
   1225 
   1226 		/*
   1227 		 * Don't need to get a lock on the backing file if
   1228 		 * the listattr is being applied to the backing file,
   1229 		 * as the lock is already held.
   1230 		 */
   1231 		if (uele->uele_backing_vnode != vp)
   1232 			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1233 
   1234 		/*
   1235 		 * +1 for trailing NUL (listxattr flavor)
   1236 		 *  or leading name length (extattr_list_file flavor)
   1237 	 	 */
   1238 		attrnamelen = strlen(uele->uele_attrname);
   1239 		listsize += attrnamelen + 1;
   1240 
   1241 		/* Return data if the caller requested it. */
   1242 		if (uio != NULL) {
   1243 			/*
   1244 			 * We support two flavors. Either NUL-terminated
   1245 			 * strings (a la listxattr), or non NUL-terminated,
   1246 			 * one byte length prefixed strings (for
   1247 			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
   1248 		 	 * that second behavior.
   1249 			 */
   1250 			if (flag & EXTATTR_LIST_LENPREFIX) {
   1251 				uint8_t len = (uint8_t)attrnamelen;
   1252 
   1253 				/* Copy leading name length */
   1254 				error = uiomove(&len, sizeof(len), uio);
   1255 				if (error != 0)
   1256 					break;
   1257 			} else {
   1258 				/* Include trailing NULL */
   1259 				attrnamelen++;
   1260 			}
   1261 
   1262 			error = uiomove(uele->uele_attrname,
   1263 					(size_t)attrnamelen, uio);
   1264 			if (error != 0)
   1265 				break;
   1266 		}
   1267 
   1268 		if (uele->uele_backing_vnode != vp)
   1269 			VOP_UNLOCK(uele->uele_backing_vnode);
   1270 
   1271 		if (error != 0)
   1272 			return error;
   1273 	}
   1274 
   1275 	if (uio != NULL)
   1276 		uio->uio_offset = 0;
   1277 
   1278 	/* Return full data size if caller requested it. */
   1279 	if (size != NULL)
   1280 		*size = listsize;
   1281 
   1282 	return 0;
   1283 }
   1284 
   1285 /*
   1286  * Vnode operation to remove a named attribute.
   1287  */
   1288 int
   1289 ulfs_deleteextattr(struct vop_deleteextattr_args *ap)
   1290 /*
   1291 vop_deleteextattr {
   1292 	IN struct vnode *a_vp;
   1293 	IN int a_attrnamespace;
   1294 	IN const char *a_name;
   1295 	IN kauth_cred_t a_cred;
   1296 };
   1297 */
   1298 {
   1299 	struct mount *mp = ap->a_vp->v_mount;
   1300 	struct ulfsmount *ump = VFSTOULFS(mp);
   1301 	int error;
   1302 
   1303 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1304 		return (EOPNOTSUPP);
   1305 
   1306 	ulfs_extattr_uepm_lock(ump);
   1307 
   1308 	error = ulfs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1309 	    ap->a_cred, curlwp);
   1310 
   1311 	ulfs_extattr_uepm_unlock(ump);
   1312 
   1313 	return (error);
   1314 }
   1315 
   1316 /*
   1317  * Vnode operation to set a named attribute.
   1318  */
   1319 int
   1320 ulfs_setextattr(struct vop_setextattr_args *ap)
   1321 /*
   1322 vop_setextattr {
   1323 	IN struct vnode *a_vp;
   1324 	IN int a_attrnamespace;
   1325 	IN const char *a_name;
   1326 	INOUT struct uio *a_uio;
   1327 	IN kauth_cred_t a_cred;
   1328 };
   1329 */
   1330 {
   1331 	struct mount *mp = ap->a_vp->v_mount;
   1332 	struct ulfsmount *ump = VFSTOULFS(mp);
   1333 	int error;
   1334 
   1335 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1336 		return (EOPNOTSUPP);
   1337 
   1338 	ulfs_extattr_uepm_lock(ump);
   1339 
   1340 	/*
   1341 	 * XXX: No longer a supported way to delete extended attributes.
   1342 	 */
   1343 	if (ap->a_uio == NULL) {
   1344 		ulfs_extattr_uepm_unlock(ump);
   1345 		return (EINVAL);
   1346 	}
   1347 
   1348 	error = ulfs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1349 	    ap->a_uio, ap->a_cred, curlwp);
   1350 
   1351 	ulfs_extattr_uepm_unlock(ump);
   1352 
   1353 	return (error);
   1354 }
   1355 
   1356 /*
   1357  * Real work associated with setting a vnode's extended attributes;
   1358  * assumes that the attribute lock has already been grabbed.
   1359  */
   1360 static int
   1361 ulfs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
   1362     struct uio *uio, kauth_cred_t cred, struct lwp *l)
   1363 {
   1364 	struct ulfs_extattr_list_entry *attribute;
   1365 	struct ulfs_extattr_header ueh;
   1366 	struct iovec local_aiov;
   1367 	struct uio local_aio;
   1368 	struct mount *mp = vp->v_mount;
   1369 	struct ulfsmount *ump = VFSTOULFS(mp);
   1370 	struct inode *ip = VTOI(vp);
   1371 	off_t base_offset;
   1372 	int error = 0, ioflag;
   1373 
   1374 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1375 		return (EROFS);
   1376 
   1377 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1378 		return (EINVAL);
   1379 
   1380 	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
   1381 	if (error)
   1382 		return (error);
   1383 
   1384 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1385 	if (!attribute) {
   1386 		error = ulfs_extattr_autocreate_attr(vp, attrnamespace,
   1387 						    name, l, &attribute);
   1388 		if (error == EEXIST) {
   1389 			/* Another thread raced us for backend creation */
   1390 			error = 0;
   1391 			attribute =
   1392 			    ulfs_extattr_find_attr(ump, attrnamespace, name);
   1393 		}
   1394 
   1395 		if (error || !attribute)
   1396 			return ENODATA;
   1397 	}
   1398 
   1399 	/*
   1400 	 * Early rejection of invalid offsets/length.
   1401 	 * Reject: any offset but 0 (replace)
   1402 	 *	 Any size greater than attribute size limit
   1403  	 */
   1404 	if (uio->uio_offset != 0 ||
   1405 	    uio->uio_resid > attribute->uele_fileheader.uef_size)
   1406 		return (ENXIO);
   1407 
   1408 	/*
   1409 	 * Find base offset of header in file based on file header size, and
   1410 	 * data header size + maximum data size, indexed by inode number.
   1411 	 */
   1412 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
   1413 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
   1414 	    attribute->uele_fileheader.uef_size);
   1415 
   1416 	/*
   1417 	 * Write out a data header for the data.
   1418 	 */
   1419 	ueh.ueh_len = ulfs_rw32((uint32_t) uio->uio_resid,
   1420 	    UELE_NEEDSWAP(attribute));
   1421 	ueh.ueh_flags = ulfs_rw32(ULFS_EXTATTR_ATTR_FLAG_INUSE,
   1422 				 UELE_NEEDSWAP(attribute));
   1423 	ueh.ueh_i_gen = ulfs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
   1424 	local_aiov.iov_base = &ueh;
   1425 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1426 	local_aio.uio_iov = &local_aiov;
   1427 	local_aio.uio_iovcnt = 1;
   1428 	local_aio.uio_rw = UIO_WRITE;
   1429 	local_aio.uio_offset = base_offset;
   1430 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1431 	UIO_SETUP_SYSSPACE(&local_aio);
   1432 
   1433 	/*
   1434 	 * Don't need to get a lock on the backing file if the setattr is
   1435 	 * being applied to the backing file, as the lock is already held.
   1436 	 */
   1437 	if (attribute->uele_backing_vnode != vp)
   1438 		vn_lock(attribute->uele_backing_vnode,
   1439 		    LK_EXCLUSIVE | LK_RETRY);
   1440 
   1441 	ioflag = IO_NODELOCKED;
   1442 	if (ulfs_extattr_sync)
   1443 		ioflag |= IO_SYNC;
   1444 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1445 	    ump->um_extattr.uepm_ucred);
   1446 	if (error)
   1447 		goto vopunlock_exit;
   1448 
   1449 	if (local_aio.uio_resid != 0) {
   1450 		error = ENXIO;
   1451 		goto vopunlock_exit;
   1452 	}
   1453 
   1454 	/*
   1455 	 * Write out user data.
   1456 	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
   1457 	 */
   1458 	uio->uio_offset = base_offset + sizeof(struct ulfs_extattr_header);
   1459 
   1460 	ioflag = IO_NODELOCKED;
   1461 	if (ulfs_extattr_sync)
   1462 		ioflag |= IO_SYNC;
   1463 	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
   1464 	    ump->um_extattr.uepm_ucred);
   1465 
   1466  vopunlock_exit:
   1467 	uio->uio_offset = 0;
   1468 
   1469 	if (attribute->uele_backing_vnode != vp)
   1470 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1471 
   1472 	return (error);
   1473 }
   1474 
   1475 /*
   1476  * Real work associated with removing an extended attribute from a vnode.
   1477  * Assumes the attribute lock has already been grabbed.
   1478  */
   1479 static int
   1480 ulfs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
   1481     kauth_cred_t cred, struct lwp *l)
   1482 {
   1483 	struct ulfs_extattr_list_entry *attribute;
   1484 	struct ulfs_extattr_header ueh;
   1485 	struct mount *mp = vp->v_mount;
   1486 	struct ulfsmount *ump = VFSTOULFS(mp);
   1487 	struct iovec local_aiov;
   1488 	struct uio local_aio;
   1489 	off_t base_offset;
   1490 	int error = 0, ioflag;
   1491 
   1492 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1493 		return (EROFS);
   1494 
   1495 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1496 		return (EINVAL);
   1497 
   1498 	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
   1499 	if (error)
   1500 		return (error);
   1501 
   1502 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1503 	if (!attribute)
   1504 		return (ENODATA);
   1505 
   1506 	/*
   1507 	 * Don't need to get a lock on the backing file if the getattr is
   1508 	 * being applied to the backing file, as the lock is already held.
   1509 	 */
   1510 	if (attribute->uele_backing_vnode != vp)
   1511 		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
   1512 
   1513 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1514 	if (error)
   1515 		goto vopunlock_exit;
   1516 
   1517 	/* Flag it as not in use. */
   1518 	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
   1519 	ueh.ueh_len = 0;		/* ...ditto... */
   1520 
   1521 	local_aiov.iov_base = &ueh;
   1522 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1523 	local_aio.uio_iov = &local_aiov;
   1524 	local_aio.uio_iovcnt = 1;
   1525 	local_aio.uio_rw = UIO_WRITE;
   1526 	local_aio.uio_offset = base_offset;
   1527 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1528 	UIO_SETUP_SYSSPACE(&local_aio);
   1529 
   1530 	ioflag = IO_NODELOCKED;
   1531 	if (ulfs_extattr_sync)
   1532 		ioflag |= IO_SYNC;
   1533 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1534 	    ump->um_extattr.uepm_ucred);
   1535 	if (error)
   1536 		goto vopunlock_exit;
   1537 
   1538 	if (local_aio.uio_resid != 0)
   1539 		error = ENXIO;
   1540 
   1541  vopunlock_exit:
   1542 	VOP_UNLOCK(attribute->uele_backing_vnode);
   1543 
   1544 	return (error);
   1545 }
   1546 
   1547 /*
   1548  * Called by ULFS when an inode is no longer active and should have its
   1549  * attributes stripped.
   1550  */
   1551 void
   1552 ulfs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
   1553 {
   1554 	struct ulfs_extattr_list_entry *uele;
   1555 	struct mount *mp = vp->v_mount;
   1556 	struct ulfsmount *ump = VFSTOULFS(mp);
   1557 
   1558 	/*
   1559 	 * In that case, we cannot lock. We should not have any active vnodes
   1560 	 * on the fs if this is not yet initialized but is going to be, so
   1561 	 * this can go unlocked.
   1562 	 */
   1563 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
   1564 		return;
   1565 
   1566 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1567 		return;
   1568 
   1569 	ulfs_extattr_uepm_lock(ump);
   1570 
   1571 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
   1572 		ulfs_extattr_rm(vp, uele->uele_attrnamespace,
   1573 		    uele->uele_attrname, lwp0.l_cred, l);
   1574 
   1575 	ulfs_extattr_uepm_unlock(ump);
   1576 }
   1577 
   1578 void
   1579 ulfs_extattr_init(void)
   1580 {
   1581 
   1582 }
   1583 
   1584 void
   1585 ulfs_extattr_done(void)
   1586 {
   1587 
   1588 }
   1589