Home | History | Annotate | Line # | Download | only in ufs
      1 /*	$NetBSD: ufs_extattr.c,v 1.57 2026/01/22 03:24:19 riastradh Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999-2002 Robert N. M. Watson
      5  * Copyright (c) 2002-2003 Networks Associates Technology, Inc.
      6  * All rights reserved.
      7  *
      8  * This software was developed by Robert Watson for the TrustedBSD Project.
      9  *
     10  * This software was developed for the FreeBSD Project in part by Network
     11  * Associates Laboratories, the Security Research Division of Network
     12  * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"),
     13  * as part of the DARPA CHATS research program.
     14  *
     15  * Redistribution and use in source and binary forms, with or without
     16  * modification, are permitted provided that the following conditions
     17  * are met:
     18  * 1. Redistributions of source code must retain the above copyright
     19  *    notice, this list of conditions and the following disclaimer.
     20  * 2. Redistributions in binary form must reproduce the above copyright
     21  *    notice, this list of conditions and the following disclaimer in the
     22  *    documentation and/or other materials provided with the distribution.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  */
     37 
     38 /*
     39  * Support for file system extended attributes on the UFS1 file system.
     40  *
     41  * Extended attributes are defined in the form name=value, where name is
     42  * a nul-terminated string in the style of a file name, and value is a
     43  * binary blob of zero or more bytes.  The UFS1 extended attribute service
     44  * layers support for extended attributes onto a backing file, in the style
     45  * of the quota implementation, meaning that it requires no underlying format
     46  * changes to the file system.  This design choice exchanges simplicity,
     47  * usability, and easy deployment for performance.
     48  */
     49 
     50 #include <sys/cdefs.h>
     51 __KERNEL_RCSID(0, "$NetBSD: ufs_extattr.c,v 1.57 2026/01/22 03:24:19 riastradh Exp $");
     52 
     53 #ifdef _KERNEL_OPT
     54 #include "opt_ffs.h"
     55 #endif
     56 
     57 #include <sys/param.h>
     58 #include <sys/types.h>
     59 
     60 #include <sys/dirent.h>
     61 #include <sys/extattr.h>
     62 #include <sys/fcntl.h>
     63 #include <sys/kauth.h>
     64 #include <sys/kernel.h>
     65 #include <sys/kmem.h>
     66 #include <sys/lock.h>
     67 #include <sys/lwp.h>
     68 #include <sys/mount.h>
     69 #include <sys/namei.h>
     70 #include <sys/reboot.h>
     71 #include <sys/sdt.h>
     72 #include <sys/sysctl.h>
     73 #include <sys/systm.h>
     74 #include <sys/vnode.h>
     75 
     76 #include <ufs/ufs/dir.h>
     77 #include <ufs/ufs/extattr.h>
     78 #include <ufs/ufs/inode.h>
     79 #include <ufs/ufs/ufs_bswap.h>
     80 #include <ufs/ufs/ufs_extern.h>
     81 #include <ufs/ufs/ufsmount.h>
     82 
     83 int ufs_extattr_sync = 1;
     84 int ufs_extattr_autocreate = 1024;
     85 
     86 static int	ufs_extattr_valid_attrname(int attrnamespace,
     87 		    const char *attrname);
     88 static int	ufs_extattr_enable_with_open(struct ufsmount *ump,
     89 		    struct vnode *vp, int attrnamespace, const char *attrname,
     90 		    struct lwp *l);
     91 static int	ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
     92 		    const char *attrname, struct vnode *backing_vnode,
     93 		    struct lwp *l);
     94 static int	ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
     95 		    const char *attrname, struct lwp *l);
     96 static int	ufs_extattr_get(struct vnode *vp, int attrnamespace,
     97 		    const char *name, struct uio *uio, size_t *size,
     98 		    kauth_cred_t cred, struct lwp *l);
     99 static int	ufs_extattr_list(struct vnode *vp, int attrnamespace,
    100 		    struct uio *uio, size_t *size, int flag,
    101 		    kauth_cred_t cred, struct lwp *l);
    102 static int	ufs_extattr_set(struct vnode *vp, int attrnamespace,
    103 		    const char *name, struct uio *uio, kauth_cred_t cred,
    104 		    struct lwp *l);
    105 static int	ufs_extattr_rm(struct vnode *vp, int attrnamespace,
    106 		    const char *name, kauth_cred_t cred, struct lwp *l);
    107 static struct ufs_extattr_list_entry *ufs_extattr_find_attr(struct ufsmount *,
    108 		    int, const char *);
    109 static int	ufs_extattr_get_header(struct vnode *,
    110 		    struct ufs_extattr_list_entry *,
    111 		    struct ufs_extattr_header *, off_t *);
    112 
    113 
    114 /*
    115  * Per-FS attribute lock protecting attribute operations.
    116  * XXX Right now there is a lot of lock contention due to having a single
    117  * lock per-FS; really, this should be far more fine-grained.
    118  */
    119 static void
    120 ufs_extattr_uepm_lock(struct ufsmount *ump)
    121 {
    122 
    123 	/*
    124 	 * XXX This needs to be recursive for the following reasons:
    125 	 *   - it is taken in ufs_extattr_vnode_inactive
    126 	 *   - which is called from VOP_INACTIVE
    127 	 *   - which can be triggered by any vrele, vput, or vn_close
    128 	 *   - several of these can happen while it's held
    129 	 */
    130 	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
    131 		ump->um_extattr.uepm_lockcnt++;
    132 		return;
    133 	}
    134 	mutex_enter(&ump->um_extattr.uepm_lock);
    135 }
    136 
    137 static void
    138 ufs_extattr_uepm_unlock(struct ufsmount *ump)
    139 {
    140 
    141 	if (ump->um_extattr.uepm_lockcnt != 0) {
    142 		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
    143 		ump->um_extattr.uepm_lockcnt--;
    144 		return;
    145 	}
    146 	mutex_exit(&ump->um_extattr.uepm_lock);
    147 }
    148 
    149 /*-
    150  * Determine whether the name passed is a valid name for an actual
    151  * attribute.
    152  *
    153  * Invalid currently consists of:
    154  *	 NULL pointer for attrname
    155  *	 zero-length attrname (used to retrieve application attribute list)
    156  */
    157 static int
    158 ufs_extattr_valid_attrname(int attrnamespace, const char *attrname)
    159 {
    160 
    161 	if (attrname == NULL)
    162 		return 0;
    163 	if (strlen(attrname) == 0)
    164 		return 0;
    165 	return 1;
    166 }
    167 
    168 /*
    169  * Autocreate an attribute storage
    170  */
    171 static int
    172 ufs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
    173     const char *attrname, struct lwp *l, struct ufs_extattr_list_entry **uelep)
    174 {
    175 	struct mount *mp = vp->v_mount;
    176 	struct ufsmount *ump = VFSTOUFS(mp);
    177 	struct vnode *backing_vp;
    178 	struct pathbuf *pb;
    179 	char *path;
    180 	struct ufs_extattr_fileheader uef;
    181 	struct ufs_extattr_list_entry *uele;
    182 	int error;
    183 
    184 	path = PNBUF_GET();
    185 
    186 	/*
    187 	 * We only support system and user namespace autocreation
    188 	 */
    189 	switch (attrnamespace) {
    190 	case EXTATTR_NAMESPACE_SYSTEM:
    191 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    192 		    mp->mnt_stat.f_mntonname, UFS_EXTATTR_FSROOTSUBDIR,
    193 		    UFS_EXTATTR_SUBDIR_SYSTEM, attrname);
    194 		break;
    195 	case EXTATTR_NAMESPACE_USER:
    196 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    197 		    mp->mnt_stat.f_mntonname, UFS_EXTATTR_FSROOTSUBDIR,
    198 		    UFS_EXTATTR_SUBDIR_USER, attrname);
    199 		break;
    200 	default:
    201 		PNBUF_PUT(path);
    202 		*uelep = NULL;
    203 		return SET_ERROR(EINVAL);
    204 		break;
    205 	}
    206 
    207 	/*
    208 	 * Release extended attribute mount lock, otherwise
    209 	 * we can deadlock with another thread that would lock
    210 	 * vp after we unlock it below, and call
    211 	 * ufs_extattr_uepm_lock(ump), for instance
    212 	 * in ufs_getextattr().
    213 	 */
    214 	ufs_extattr_uepm_unlock(ump);
    215 
    216 	/*
    217 	 * XXX unlock/lock should only be done when setting extattr
    218 	 * on backing store or one of its parent directory
    219 	 * including root, but we always do it for now.
    220 	 */
    221 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    222 	VOP_UNLOCK(vp);
    223 
    224 	pb = pathbuf_create(path);
    225 
    226 	/*
    227 	 * Since we do not hold ufs_extattr_uepm_lock anymore,
    228 	 * another thread may race with us for backend creation,
    229 	 * but only one can succeed here thanks to O_EXCL.
    230 	 *
    231  	 * backing_vp is the backing store.
    232 	 */
    233 	error = vn_open(NULL, pb, 0, O_CREAT|O_EXCL|O_RDWR, 0600,
    234 	    &backing_vp, NULL, NULL);
    235 
    236 	/*
    237 	 * Reacquire the lock on the vnode
    238 	 */
    239 	KASSERT(VOP_ISLOCKED(vp) == 0);
    240 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    241 
    242 	ufs_extattr_uepm_lock(ump);
    243 
    244 	if (error != 0) {
    245 		pathbuf_destroy(pb);
    246 		PNBUF_PUT(path);
    247 		*uelep = NULL;
    248 		return error;
    249 	}
    250 
    251 	KASSERT(backing_vp != NULL);
    252 	KASSERT(VOP_ISLOCKED(backing_vp) == LK_EXCLUSIVE);
    253 
    254 	pathbuf_destroy(pb);
    255 	PNBUF_PUT(path);
    256 
    257 	uef.uef_magic = UFS_EXTATTR_MAGIC;
    258 	uef.uef_version = UFS_EXTATTR_VERSION;
    259 	uef.uef_size = ufs_extattr_autocreate;
    260 
    261 	error = vn_rdwr(UIO_WRITE, backing_vp, &uef, sizeof(uef), 0,
    262 		        UIO_SYSSPACE, IO_NODELOCKED|IO_APPEND,
    263 			l->l_cred, NULL, l);
    264 
    265 	VOP_UNLOCK(backing_vp);
    266 
    267 	if (error != 0) {
    268 		printf("%s: write uef header failed for `%s' (%d)\n",
    269 		    __func__, attrname, error);
    270 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    271 		*uelep = NULL;
    272 		return error;
    273 	}
    274 
    275 	/*
    276 	 * Now enable attribute.
    277 	 */
    278 	error = ufs_extattr_enable(ump,attrnamespace, attrname, backing_vp, l);
    279 	KASSERT(VOP_ISLOCKED(backing_vp) == 0);
    280 
    281 	if (error != 0) {
    282 		printf("%s: enable `%s' failed (%d)\n",
    283 		    __func__, attrname, error);
    284 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    285 		*uelep = NULL;
    286 		return error;
    287 	}
    288 
    289 	uele = ufs_extattr_find_attr(ump, attrnamespace, attrname);
    290 	if (uele == NULL) {
    291 		printf("%s: attribute `%s' created but not found!\n",
    292 		    __func__, attrname);
    293 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    294 		*uelep = NULL;
    295 		return SET_ERROR(ESRCH); /* really internal error */
    296 	}
    297 
    298 	printf("%s: EA backing store autocreated for %s\n",
    299 	    mp->mnt_stat.f_mntonname, attrname);
    300 
    301 	*uelep = uele;
    302 	return 0;
    303 }
    304 
    305 /*
    306  * Locate an attribute given a name and mountpoint.
    307  * Must be holding uepm lock for the mount point.
    308  */
    309 static struct ufs_extattr_list_entry *
    310 ufs_extattr_find_attr(struct ufsmount *ump, int attrnamespace,
    311     const char *attrname)
    312 {
    313 	struct ufs_extattr_list_entry *search_attribute;
    314 
    315 	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
    316 	    search_attribute != NULL;
    317 	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
    318 		if (!(strncmp(attrname, search_attribute->uele_attrname,
    319 			    UFS_EXTATTR_MAXEXTATTRNAME)) &&
    320 		    (attrnamespace == search_attribute->uele_attrnamespace)) {
    321 			return search_attribute;
    322 		}
    323 	}
    324 
    325 	return 0;
    326 }
    327 
    328 /*
    329  * Initialize per-FS structures supporting extended attributes.  Do not
    330  * start extended attributes yet.
    331  */
    332 void
    333 ufs_extattr_uepm_init(struct ufs_extattr_per_mount *uepm)
    334 {
    335 
    336 	uepm->uepm_flags = 0;
    337 	uepm->uepm_lockcnt = 0;
    338 
    339 	LIST_INIT(&uepm->uepm_list);
    340 	mutex_init(&uepm->uepm_lock, MUTEX_DEFAULT, IPL_NONE);
    341 	uepm->uepm_flags |= UFS_EXTATTR_UEPM_INITIALIZED;
    342 }
    343 
    344 /*
    345  * Destroy per-FS structures supporting extended attributes.  Assumes
    346  * that EAs have already been stopped, and will panic if not.
    347  */
    348 void
    349 ufs_extattr_uepm_destroy(struct ufs_extattr_per_mount *uepm)
    350 {
    351 
    352 	if (!(uepm->uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
    353 		panic("ufs_extattr_uepm_destroy: not initialized");
    354 
    355 	if ((uepm->uepm_flags & UFS_EXTATTR_UEPM_STARTED))
    356 		panic("ufs_extattr_uepm_destroy: called while still started");
    357 
    358 	/*
    359 	 * It's not clear that either order for the next three lines is
    360 	 * ideal, and it should never be a problem if this is only called
    361 	 * during unmount, and with vfs_busy().
    362 	 */
    363 	uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_STARTED;
    364 	uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_INITIALIZED;
    365 	mutex_destroy(&uepm->uepm_lock);
    366 }
    367 
    368 /*
    369  * Start extended attribute support on an FS.
    370  */
    371 int
    372 ufs_extattr_start(struct mount *mp, struct lwp *l)
    373 {
    374 	struct ufsmount *ump;
    375 	int error = 0;
    376 
    377 	ump = VFSTOUFS(mp);
    378 
    379 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
    380 		ufs_extattr_uepm_init(&ump->um_extattr);
    381 
    382 	ufs_extattr_uepm_lock(ump);
    383 
    384 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED)) {
    385 		error = SET_ERROR(EOPNOTSUPP);
    386 		goto unlock;
    387 	}
    388 	if (ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED) {
    389 		error = SET_ERROR(EBUSY);
    390 		goto unlock;
    391 	}
    392 
    393 	ump->um_extattr.uepm_flags |= UFS_EXTATTR_UEPM_STARTED;
    394 
    395 	ump->um_extattr.uepm_ucred = l->l_cred;
    396 	kauth_cred_hold(ump->um_extattr.uepm_ucred);
    397 
    398  unlock:
    399 	ufs_extattr_uepm_unlock(ump);
    400 	return error;
    401 }
    402 
    403 /*
    404  * Helper routine: given a locked parent directory and filename, return
    405  * the locked vnode of the inode associated with the name.  Will not
    406  * follow symlinks, may return any type of vnode.  Lock on parent will
    407  * be released even in the event of a failure.  In the event that the
    408  * target is the parent (i.e., "."), there will be two references and
    409  * one lock, requiring the caller to possibly special-case.
    410  */
    411 static int
    412 ufs_extattr_lookup(struct vnode *start_dvp, int lockparent,
    413     const char *dirname,
    414     struct vnode **vp, struct lwp *l)
    415 {
    416 	struct vop_lookup_v2_args vargs;
    417 	struct componentname cnp;
    418 	struct vnode *target_vp;
    419 	char *pnbuf;
    420 	int error;
    421 
    422 	KASSERT(VOP_ISLOCKED(start_dvp) == LK_EXCLUSIVE);
    423 
    424 	pnbuf = PNBUF_GET();
    425 
    426 	memset(&cnp, 0, sizeof(cnp));
    427 	cnp.cn_nameiop = LOOKUP;
    428 	cnp.cn_flags = ISLASTCN | lockparent;
    429 	cnp.cn_cred = l->l_cred;
    430 	cnp.cn_nameptr = pnbuf;
    431 	error = copystr(dirname, pnbuf, MAXPATHLEN, &cnp.cn_namelen);
    432 	if (error) {
    433 		if (lockparent == 0) {
    434 			VOP_UNLOCK(start_dvp);
    435 		}
    436 		PNBUF_PUT(pnbuf);
    437 		printf("%s: copystr failed (%d)\n", __func__, error);
    438 		return error;
    439 	}
    440 	cnp.cn_namelen--;	/* trim nul termination */
    441 	vargs.a_desc = NULL;
    442 	vargs.a_dvp = start_dvp;
    443 	vargs.a_vpp = &target_vp;
    444 	vargs.a_cnp = &cnp;
    445 	error = ufs_lookup(&vargs);
    446 	PNBUF_PUT(pnbuf);
    447 	if (error) {
    448 		if (lockparent == 0) {
    449 			VOP_UNLOCK(start_dvp);
    450 		}
    451 		return error;
    452 	}
    453 #if 0
    454 	if (target_vp == start_dvp)
    455 		panic("%s: target_vp == start_dvp", __func__);
    456 #endif
    457 
    458 	if (target_vp != start_dvp) {
    459 		error = vn_lock(target_vp, LK_EXCLUSIVE);
    460 		if (lockparent == 0)
    461 			VOP_UNLOCK(start_dvp);
    462 		if (error) {
    463 			vrele(target_vp);
    464 			return error;
    465 		}
    466 	}
    467 
    468 	KASSERT(VOP_ISLOCKED(target_vp) == LK_EXCLUSIVE);
    469 	*vp = target_vp;
    470 	return 0;
    471 }
    472 
    473 /*
    474  * Enable an EA using the passed filesystem, backing vnode, attribute name,
    475  * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
    476  * to be locked when passed in.  The vnode will be returned unlocked,
    477  * regardless of success/failure of the function.  As a result, the caller
    478  * will always need to vrele(), but not vput().
    479  */
    480 static int
    481 ufs_extattr_enable_with_open(struct ufsmount *ump, struct vnode *vp,
    482     int attrnamespace, const char *attrname, struct lwp *l)
    483 {
    484 	int error;
    485 
    486 	error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred);
    487 	if (error) {
    488 		printf("%s: VOP_OPEN(): failed (%d)\n", __func__, error);
    489 		VOP_UNLOCK(vp);
    490 		return error;
    491 	}
    492 
    493 	mutex_enter(vp->v_interlock);
    494 	vp->v_writecount++;
    495 	mutex_exit(vp->v_interlock);
    496 
    497 	vref(vp);
    498 
    499 	VOP_UNLOCK(vp);
    500 
    501 	error = ufs_extattr_enable(ump, attrnamespace, attrname, vp, l);
    502 	if (error != 0)
    503 		vn_close(vp, FREAD|FWRITE, l->l_cred);
    504 	return error;
    505 }
    506 
    507 /*
    508  * Given a locked directory vnode, iterate over the names in the directory
    509  * and use ufs_extattr_lookup() to retrieve locked vnodes of potential
    510  * attribute files.  Then invoke ufs_extattr_enable_with_open() on each
    511  * to attempt to start the attribute.  Leaves the directory locked on
    512  * exit.
    513  */
    514 static int
    515 ufs_extattr_iterate_directory(struct ufsmount *ump, struct vnode *dvp,
    516     int attrnamespace, struct lwp *l)
    517 {
    518 	struct vop_readdir_args vargs;
    519 	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
    520 	struct dirent *dp, *edp;
    521 	struct vnode *attr_vp;
    522 	struct uio auio;
    523 	struct iovec aiov;
    524 	char *dirbuf;
    525 	int error, eofflag = 0;
    526 
    527 	if (dvp->v_type != VDIR)
    528 		return SET_ERROR(ENOTDIR);
    529 
    530 	dirbuf = kmem_alloc(UFS_DIRBLKSIZ, KM_SLEEP);
    531 
    532 	auio.uio_iov = &aiov;
    533 	auio.uio_iovcnt = 1;
    534 	auio.uio_rw = UIO_READ;
    535 	auio.uio_offset = 0;
    536 	UIO_SETUP_SYSSPACE(&auio);
    537 
    538 	vargs.a_desc = NULL;
    539 	vargs.a_vp = dvp;
    540 	vargs.a_uio = &auio;
    541 	vargs.a_cred = l->l_cred;
    542 	vargs.a_eofflag = &eofflag;
    543 	vargs.a_ncookies = NULL;
    544 	vargs.a_cookies = NULL;
    545 
    546 	while (!eofflag) {
    547 		auio.uio_resid = UFS_DIRBLKSIZ;
    548 		aiov.iov_base = dirbuf;
    549 		aiov.iov_len = UFS_DIRBLKSIZ;
    550 		error = ufs_readdir(&vargs);
    551 		if (error) {
    552 			printf("%s: ufs_readdir (%d)\n", __func__, error);
    553 			return error;
    554 		}
    555 
    556 		/*
    557 		 * XXXRW: While in UFS, we always get UFS_DIRBLKSIZ returns from
    558 		 * the directory code on success, on other file systems this
    559 		 * may not be the case.  For portability, we should check the
    560 		 * read length on return from ufs_readdir().
    561 		 */
    562 		edp = (struct dirent *)&dirbuf[UFS_DIRBLKSIZ];
    563 		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
    564 			if (dp->d_reclen == 0)
    565 				break;
    566 			/* Skip "." and ".." */
    567 			if (dp->d_name[0] == '.' &&
    568 			    (dp->d_name[1] == '\0' ||
    569 			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
    570 				goto next;
    571 			error = ufs_extattr_lookup(dvp, LOCKPARENT,
    572 			    dp->d_name, &attr_vp, l);
    573 			if (error == ENOENT) {
    574 				goto next; /* keep silent */
    575 			} else if (error) {
    576 				printf("%s: lookup `%s' (%d)\n", __func__,
    577 				    dp->d_name, error);
    578 			} else if (attr_vp == dvp) {
    579 				vrele(attr_vp);
    580 			} else if (attr_vp->v_type != VREG) {
    581 				vput(attr_vp);
    582 			} else {
    583 				error = ufs_extattr_enable_with_open(ump,
    584 				    attr_vp, attrnamespace, dp->d_name, l);
    585 				vrele(attr_vp);
    586 				if (error) {
    587 					printf("%s: enable `%s' (%d)\n",
    588 					    __func__, dp->d_name, error);
    589 				} else if (bootverbose) {
    590 					printf("%s: EA %s loaded\n",
    591 					    sbp->f_mntonname, dp->d_name);
    592 				}
    593 			}
    594  next:
    595 			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
    596 			if (dp >= edp)
    597 				break;
    598 		}
    599 	}
    600 	kmem_free(dirbuf, UFS_DIRBLKSIZ);
    601 
    602 	return 0;
    603 }
    604 
    605 static int
    606 ufs_extattr_subdir(struct lwp *l, struct mount *mp, struct vnode *attr_dvp,
    607     const char *subdir, int namespace)
    608 {
    609 	int error;
    610 	struct vnode *attr_sub;
    611 	error = ufs_extattr_lookup(attr_dvp, LOCKPARENT, subdir, &attr_sub, l);
    612 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    613 	if (error) {
    614 		printf("%s: Can't find `%s/%s/%s' (%d)\n",
    615 		    __func__, mp->mnt_stat.f_mntonname,
    616 		    UFS_EXTATTR_FSROOTSUBDIR, subdir, error);
    617 		return error;
    618 	}
    619 	KASSERT(VOP_ISLOCKED(attr_sub) == LK_EXCLUSIVE);
    620 	error = ufs_extattr_iterate_directory(VFSTOUFS(mp),
    621 	    attr_sub, namespace, l);
    622 	if (error) {
    623 		printf("%s: ufs_extattr_iterate_directory `%s/%s/%s' (%d)\n",
    624 		    __func__, mp->mnt_stat.f_mntonname,
    625 		    UFS_EXTATTR_FSROOTSUBDIR, subdir, error);
    626 	}
    627 	KASSERT(VOP_ISLOCKED(attr_sub) == LK_EXCLUSIVE);
    628 	vput(attr_sub);
    629 	return error;
    630 }
    631 
    632 /*
    633  * Auto-start of extended attributes, to be executed (optionally) at
    634  * mount-time.
    635  */
    636 int
    637 ufs_extattr_autostart(struct mount *mp, struct lwp *l)
    638 {
    639 	struct vnode *rvp, *attr_dvp;
    640 	int error;
    641 
    642 	/*
    643 	 * Does UFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
    644 	 * If so, automatically start EA's.
    645 	 */
    646 	error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp);
    647 	if (error) {
    648 		printf("%s: VFS_ROOT() (%d)\n", __func__, error);
    649 		return error;
    650 	}
    651 
    652 	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    653 
    654 	error = ufs_extattr_lookup(rvp, 0,
    655 	    UFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
    656 	if (error) {
    657 		/* rvp ref'd but now unlocked */
    658 		KASSERT(VOP_ISLOCKED(rvp) == 0);
    659 		vrele(rvp);
    660 		printf("%s: lookup `%s/%s' (%d)\n", __func__,
    661 		    mp->mnt_stat.f_mntonname, UFS_EXTATTR_FSROOTSUBDIR, error);
    662 		return error;
    663 	}
    664 	if (rvp == attr_dvp) {
    665 		/* Should never happen. */
    666 		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    667 		vrele(attr_dvp);
    668 		vput(rvp);
    669 		printf("%s: `/' == `%s/%s' (%d)\n", __func__,
    670 		    mp->mnt_stat.f_mntonname, UFS_EXTATTR_FSROOTSUBDIR, EINVAL);
    671 		return SET_ERROR(EINVAL);
    672 	}
    673 	KASSERT(VOP_ISLOCKED(rvp) == 0);
    674 	vrele(rvp);
    675 
    676 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    677 
    678 	if (attr_dvp->v_type != VDIR) {
    679 		printf("%s: `%s/%s' is not a directory\n",
    680 		    __func__, mp->mnt_stat.f_mntonname,
    681 		    UFS_EXTATTR_FSROOTSUBDIR);
    682 		goto return_vput_attr_dvp;
    683 	}
    684 
    685 	error = ufs_extattr_start(mp, l);
    686 	if (error) {
    687 		printf("%s: ufs_extattr_start failed (%d)\n", __func__,
    688 		    error);
    689 		goto return_vput_attr_dvp;
    690 	}
    691 
    692 	/*
    693 	 * Look for two subdirectories: UFS_EXTATTR_SUBDIR_SYSTEM,
    694 	 * UFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
    695 	 * and start with appropriate type.  Failures in either don't
    696 	 * result in an over-all failure.  attr_dvp is left locked to
    697 	 * be cleaned up on exit.
    698 	 */
    699 	error = ufs_extattr_subdir(l, mp, attr_dvp, UFS_EXTATTR_SUBDIR_SYSTEM,
    700 	    EXTATTR_NAMESPACE_SYSTEM);
    701 	error = ufs_extattr_subdir(l, mp, attr_dvp, UFS_EXTATTR_SUBDIR_USER,
    702 	    EXTATTR_NAMESPACE_USER);
    703 
    704 	/* Mask startup failures in sub-directories. */
    705 	error = 0;
    706 
    707  return_vput_attr_dvp:
    708 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    709 	vput(attr_dvp);
    710 
    711 	return error;
    712 }
    713 
    714 /*
    715  * Stop extended attribute support on an FS.
    716  */
    717 void
    718 ufs_extattr_stop(struct mount *mp, struct lwp *l)
    719 {
    720 	struct ufs_extattr_list_entry *uele;
    721 	struct ufsmount *ump = VFSTOUFS(mp);
    722 
    723 	ufs_extattr_uepm_lock(ump);
    724 
    725 	/*
    726 	 * If we haven't been started, no big deal.  Just short-circuit
    727 	 * the processing work.
    728 	 */
    729 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
    730 		goto unlock;
    731 	}
    732 
    733 	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
    734 		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
    735 		ufs_extattr_disable(ump, uele->uele_attrnamespace,
    736 		    uele->uele_attrname, l);
    737 	}
    738 
    739 	ump->um_extattr.uepm_flags &= ~UFS_EXTATTR_UEPM_STARTED;
    740 
    741 	kauth_cred_free(ump->um_extattr.uepm_ucred);
    742 	ump->um_extattr.uepm_ucred = NULL;
    743 
    744  unlock:
    745 	ufs_extattr_uepm_unlock(ump);
    746 }
    747 
    748 /*
    749  * Enable a named attribute on the specified filesystem; provide an
    750  * unlocked backing vnode to hold the attribute data.
    751  */
    752 static int
    753 ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
    754     const char *attrname, struct vnode *backing_vnode, struct lwp *l)
    755 {
    756 	struct ufs_extattr_list_entry *attribute;
    757 	struct iovec aiov;
    758 	struct uio auio;
    759 	int error = 0;
    760 
    761 	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
    762 		return SET_ERROR(EINVAL);
    763 	if (backing_vnode->v_type != VREG)
    764 		return SET_ERROR(EINVAL);
    765 
    766 	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
    767 
    768 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
    769 		error = SET_ERROR(EOPNOTSUPP);
    770 		goto free_exit;
    771 	}
    772 
    773 	if (ufs_extattr_find_attr(ump, attrnamespace, attrname)) {
    774 		error = SET_ERROR(EEXIST);
    775 		goto free_exit;
    776 	}
    777 
    778 	strncpy(attribute->uele_attrname, attrname,
    779 	    UFS_EXTATTR_MAXEXTATTRNAME);
    780 	attribute->uele_attrnamespace = attrnamespace;
    781 	memset(&attribute->uele_fileheader, 0,
    782 	    sizeof(struct ufs_extattr_fileheader));
    783 
    784 	attribute->uele_backing_vnode = backing_vnode;
    785 
    786 	auio.uio_iov = &aiov;
    787 	auio.uio_iovcnt = 1;
    788 	aiov.iov_base = (void *) &attribute->uele_fileheader;
    789 	aiov.iov_len = sizeof(struct ufs_extattr_fileheader);
    790 	auio.uio_resid = sizeof(struct ufs_extattr_fileheader);
    791 	auio.uio_offset = (off_t) 0;
    792 	auio.uio_rw = UIO_READ;
    793 	UIO_SETUP_SYSSPACE(&auio);
    794 
    795 	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
    796 	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
    797 	    ump->um_extattr.uepm_ucred);
    798 
    799 	if (error)
    800 		goto unlock_free_exit;
    801 
    802 	if (auio.uio_resid != 0) {
    803 		printf("%s: malformed attribute header\n", __func__);
    804 		error = SET_ERROR(EINVAL);
    805 		goto unlock_free_exit;
    806 	}
    807 
    808 	/*
    809 	 * Try to determine the byte order of the attribute file.
    810 	 */
    811 	if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) {
    812 		attribute->uele_flags |= UELE_F_NEEDSWAP;
    813 		attribute->uele_fileheader.uef_magic =
    814 		    ufs_rw32(attribute->uele_fileheader.uef_magic,
    815 			UELE_NEEDSWAP(attribute));
    816 		if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) {
    817 			printf("%s: invalid attribute header magic\n",
    818 			    __func__);
    819 			error = SET_ERROR(EINVAL);
    820 			goto unlock_free_exit;
    821 		}
    822 	}
    823 	attribute->uele_fileheader.uef_version =
    824 	    ufs_rw32(attribute->uele_fileheader.uef_version,
    825 		UELE_NEEDSWAP(attribute));
    826 	attribute->uele_fileheader.uef_size =
    827 	    ufs_rw32(attribute->uele_fileheader.uef_size,
    828 		UELE_NEEDSWAP(attribute));
    829 
    830 	if (attribute->uele_fileheader.uef_version != UFS_EXTATTR_VERSION) {
    831 		printf("%s: incorrect attribute header version %d != %d\n",
    832 		    __func__, attribute->uele_fileheader.uef_version,
    833 		    UFS_EXTATTR_VERSION);
    834 		error = SET_ERROR(EINVAL);
    835 		goto unlock_free_exit;
    836 	}
    837 
    838 	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute, uele_entries);
    839 
    840 	VOP_UNLOCK(backing_vnode);
    841 	return 0;
    842 
    843  unlock_free_exit:
    844 	VOP_UNLOCK(backing_vnode);
    845 
    846  free_exit:
    847 	kmem_free(attribute, sizeof(*attribute));
    848 	return error;
    849 }
    850 
    851 /*
    852  * Disable extended attribute support on an FS.
    853  */
    854 static int
    855 ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
    856     const char *attrname, struct lwp *l)
    857 {
    858 	struct ufs_extattr_list_entry *uele;
    859 	int error = 0;
    860 
    861 	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
    862 		return SET_ERROR(EINVAL);
    863 
    864 	uele = ufs_extattr_find_attr(ump, attrnamespace, attrname);
    865 	if (!uele)
    866 		return SET_ERROR(ENODATA);
    867 
    868 	LIST_REMOVE(uele, uele_entries);
    869 
    870 	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE, 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 UFS.  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 ufs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    884     int attrnamespace, const char *attrname)
    885 {
    886 	struct lwp *l = curlwp;
    887 	struct ufsmount *ump = VFSTOUFS(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 UFS_EXTATTR_CMD_START:
    903 	case UFS_EXTATTR_CMD_STOP:
    904 	case UFS_EXTATTR_CMD_ENABLE:
    905 	case UFS_EXTATTR_CMD_DISABLE:
    906 		if (filename_vp != NULL) {
    907 			VOP_UNLOCK(filename_vp);
    908 			return SET_ERROR(EINVAL);
    909 		}
    910 		if (attrname != NULL)
    911 			return SET_ERROR(EINVAL);
    912 		break;
    913 	default:
    914 		return SET_ERROR(EINVAL);
    915 	}
    916 
    917 	switch(cmd) {
    918 	case UFS_EXTATTR_CMD_START:
    919 		error = ufs_extattr_autostart(mp, l);
    920 		return error;
    921 
    922 	case UFS_EXTATTR_CMD_STOP:
    923 		ufs_extattr_stop(mp, l);
    924 		return 0;
    925 
    926 	case UFS_EXTATTR_CMD_ENABLE:
    927 		/*
    928 		 * ufs_extattr_enable_with_open() will always unlock the
    929 		 * vnode, regardless of failure.
    930 		 */
    931 		ufs_extattr_uepm_lock(ump);
    932 		error = ufs_extattr_enable_with_open(ump, filename_vp,
    933 		    attrnamespace, attrname, l);
    934 		ufs_extattr_uepm_unlock(ump);
    935 		return error;
    936 
    937 	case UFS_EXTATTR_CMD_DISABLE:
    938 		ufs_extattr_uepm_lock(ump);
    939 		error = ufs_extattr_disable(ump, attrnamespace, attrname, l);
    940 		ufs_extattr_uepm_unlock(ump);
    941 		return error;
    942 
    943 	default:
    944 		return SET_ERROR(EINVAL);
    945 	}
    946 }
    947 
    948 /*
    949  * Read extended attribute header for a given vnode and attribute.
    950  * Backing vnode should be locked and unlocked by caller.
    951  */
    952 static int
    953 ufs_extattr_get_header(struct vnode *vp, struct ufs_extattr_list_entry *uele,
    954     struct ufs_extattr_header *ueh, off_t *bap)
    955 {
    956 	struct mount *mp = vp->v_mount;
    957 	struct ufsmount *ump = VFSTOUFS(mp);
    958 	struct inode *ip = VTOI(vp);
    959 	off_t base_offset;
    960 	struct iovec aiov;
    961 	struct uio aio;
    962 	int error;
    963 
    964 	/*
    965 	 * Find base offset of header in file based on file header size, and
    966 	 * data header size + maximum data size, indexed by inode number.
    967 	 */
    968 	base_offset = sizeof(struct ufs_extattr_fileheader) +
    969 	    ip->i_number * (sizeof(struct ufs_extattr_header) +
    970 	    uele->uele_fileheader.uef_size);
    971 
    972 	/*
    973 	 * Read in the data header to see if the data is defined, and if so
    974 	 * how much.
    975 	 */
    976 	memset(ueh, 0, sizeof(struct ufs_extattr_header));
    977 	aiov.iov_base = ueh;
    978 	aiov.iov_len = sizeof(struct ufs_extattr_header);
    979 	aio.uio_iov = &aiov;
    980 	aio.uio_iovcnt = 1;
    981 	aio.uio_rw = UIO_READ;
    982 	aio.uio_offset = base_offset;
    983 	aio.uio_resid = sizeof(struct ufs_extattr_header);
    984 	UIO_SETUP_SYSSPACE(&aio);
    985 
    986 	error = VOP_READ(uele->uele_backing_vnode, &aio,
    987 	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
    988 	if (error)
    989 		return error;
    990 
    991 	/*
    992 	 * Attribute headers are kept in file system byte order.
    993 	 * XXX What about the blob of data?
    994 	 */
    995 	ueh->ueh_flags = ufs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
    996 	ueh->ueh_len   = ufs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
    997 	ueh->ueh_i_gen = ufs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
    998 
    999 	/* Defined? */
   1000 	if ((ueh->ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
   1001 		return SET_ERROR(ENODATA);
   1002 
   1003 	/* Valid for the current inode generation? */
   1004 	if (ueh->ueh_i_gen != ip->i_gen) {
   1005 		/*
   1006 		 * The inode itself has a different generation number
   1007 		 * than the uele data.  For now, the best solution
   1008 		 * is to coerce this to undefined, and let it get cleaned
   1009 		 * up by the next write or extattrctl clean.
   1010 		 */
   1011 		printf("%s: %s: inode gen inconsistency (%u, %jd)\n",
   1012 		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
   1013 		       (intmax_t)ip->i_gen);
   1014 		return SET_ERROR(ENODATA);
   1015 	}
   1016 
   1017 	/* Local size consistency check. */
   1018 	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
   1019 		return SET_ERROR(ENXIO);
   1020 
   1021 	/* Return base offset */
   1022 	if (bap != NULL)
   1023 		*bap = base_offset;
   1024 
   1025 	return 0;
   1026 }
   1027 
   1028 /*
   1029  * Vnode operation to retrieve a named extended attribute.
   1030  */
   1031 int
   1032 ufs_getextattr(struct vop_getextattr_args *ap)
   1033 /*
   1034 vop_getextattr {
   1035 	IN struct vnode *a_vp;
   1036 	IN int a_attrnamespace;
   1037 	IN const char *a_name;
   1038 	INOUT struct uio *a_uio;
   1039 	OUT size_t *a_size;
   1040 	IN kauth_cred_t a_cred;
   1041 };
   1042 */
   1043 {
   1044 	struct mount *mp = ap->a_vp->v_mount;
   1045 	struct ufsmount *ump = VFSTOUFS(mp);
   1046 	int error;
   1047 
   1048 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1049 		return SET_ERROR(EOPNOTSUPP);
   1050 
   1051 	ufs_extattr_uepm_lock(ump);
   1052 
   1053 	error = ufs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1054 	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
   1055 
   1056 	ufs_extattr_uepm_unlock(ump);
   1057 
   1058 	return error;
   1059 }
   1060 
   1061 /*
   1062  * Real work associated with retrieving a named attribute--assumes that
   1063  * the attribute lock has already been grabbed.
   1064  */
   1065 static int
   1066 ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
   1067     struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
   1068 {
   1069 	struct ufs_extattr_list_entry *attribute;
   1070 	struct ufs_extattr_header ueh;
   1071 	struct mount *mp = vp->v_mount;
   1072 	struct ufsmount *ump = VFSTOUFS(mp);
   1073 	off_t base_offset;
   1074 	size_t len, old_len;
   1075 	int error = 0;
   1076 
   1077 	if (strlen(name) == 0)
   1078 		return SET_ERROR(EINVAL);
   1079 
   1080 	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
   1081 	if (error)
   1082 		return error;
   1083 
   1084 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1085 	if (!attribute)
   1086 		return SET_ERROR(ENODATA);
   1087 
   1088 	/*
   1089 	 * Allow only offsets of zero to encourage the read/replace
   1090 	 * extended attribute semantic.  Otherwise we can't guarantee
   1091 	 * atomicity, as we don't provide locks for extended attributes.
   1092 	 */
   1093 	if (uio != NULL && uio->uio_offset != 0)
   1094 		return SET_ERROR(ENXIO);
   1095 
   1096 	/*
   1097 	 * Don't need to get a lock on the backing file if the getattr is
   1098 	 * being applied to the backing file, as the lock is already held.
   1099 	 */
   1100 	if (attribute->uele_backing_vnode != vp)
   1101 		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1102 
   1103 	error = ufs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1104 	if (error)
   1105 		goto vopunlock_exit;
   1106 
   1107 	/* Return full data size if caller requested it. */
   1108 	if (size != NULL)
   1109 		*size = ueh.ueh_len;
   1110 
   1111 	/* Return data if the caller requested it. */
   1112 	if (uio != NULL) {
   1113 		/* Allow for offset into the attribute data. */
   1114 		uio->uio_offset = base_offset + sizeof(struct
   1115 		    ufs_extattr_header);
   1116 
   1117 		/*
   1118 		 * Figure out maximum to transfer -- use buffer size and
   1119 		 * local data limit.
   1120 		 */
   1121 		len = MIN(uio->uio_resid, ueh.ueh_len);
   1122 		old_len = uio->uio_resid;
   1123 		uio->uio_resid = len;
   1124 
   1125 		error = VOP_READ(attribute->uele_backing_vnode, uio,
   1126 		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1127 		if (error)
   1128 			goto vopunlock_exit;
   1129 
   1130 		uio->uio_resid = old_len - (len - uio->uio_resid);
   1131 	}
   1132 
   1133  vopunlock_exit:
   1134 
   1135 	if (uio != NULL)
   1136 		uio->uio_offset = 0;
   1137 
   1138 	if (attribute->uele_backing_vnode != vp)
   1139 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1140 
   1141 	return error;
   1142 }
   1143 
   1144 /*
   1145  * Vnode operation to list extended attribute for a vnode
   1146  */
   1147 int
   1148 ufs_listextattr(struct vop_listextattr_args *ap)
   1149 /*
   1150 vop_listextattr {
   1151 	IN struct vnode *a_vp;
   1152 	IN int a_attrnamespace;
   1153 	INOUT struct uio *a_uio;
   1154 	OUT size_t *a_size;
   1155 	IN int flag;
   1156 	IN kauth_cred_t a_cred;
   1157 	struct proc *a_p;
   1158 };
   1159 */
   1160 {
   1161 	struct mount *mp = ap->a_vp->v_mount;
   1162 	struct ufsmount *ump = VFSTOUFS(mp);
   1163 	int error;
   1164 
   1165 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1166 		return SET_ERROR(EOPNOTSUPP);
   1167 
   1168 	ufs_extattr_uepm_lock(ump);
   1169 
   1170 	error = ufs_extattr_list(ap->a_vp, ap->a_attrnamespace,
   1171 	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
   1172 
   1173 	ufs_extattr_uepm_unlock(ump);
   1174 
   1175 	return error;
   1176 }
   1177 
   1178 /*
   1179  * Real work associated with retrieving list of attributes--assumes that
   1180  * the attribute lock has already been grabbed.
   1181  */
   1182 static int
   1183 ufs_extattr_list(struct vnode *vp, int attrnamespace,
   1184     struct uio *uio, size_t *size, int flag,
   1185     kauth_cred_t cred, struct lwp *l)
   1186 {
   1187 	struct ufs_extattr_list_entry *uele;
   1188 	struct ufs_extattr_header ueh;
   1189 	struct mount *mp = vp->v_mount;
   1190 	struct ufsmount *ump = VFSTOUFS(mp);
   1191 	size_t listsize = 0;
   1192 	int error = 0;
   1193 
   1194 	/*
   1195 	 * XXX: We can move this inside the loop and iterate on individual
   1196 	 *	attributes.
   1197 	 */
   1198 	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
   1199 	if (error)
   1200 		return error;
   1201 
   1202 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
   1203 		unsigned char attrnamelen;
   1204 
   1205 		if (uele->uele_attrnamespace != attrnamespace)
   1206 			continue;
   1207 
   1208 		error = ufs_extattr_get_header(vp, uele, &ueh, NULL);
   1209 		if (error == ENODATA)
   1210 			continue;
   1211 		if (error != 0)
   1212 			return error;
   1213 
   1214 		/*
   1215 		 * Don't need to get a lock on the backing file if
   1216 		 * the listattr is being applied to the backing file,
   1217 		 * as the lock is already held.
   1218 		 */
   1219 		if (uele->uele_backing_vnode != vp)
   1220 			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1221 
   1222 		/*
   1223 		 * +1 for trailing NUL (listxattr flavor)
   1224 		 *  or leading name length (extattr_list_file flavor)
   1225 	 	 */
   1226 		attrnamelen = strlen(uele->uele_attrname);
   1227 		listsize += attrnamelen + 1;
   1228 
   1229 		/* Return data if the caller requested it. */
   1230 		if (uio != NULL) {
   1231 			/*
   1232 			 * We support two flavors. Either NUL-terminated
   1233 			 * strings (a la listxattr), or non NUL-terminated,
   1234 			 * one byte length prefixed strings (for
   1235 			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
   1236 		 	 * that second behavior.
   1237 			 */
   1238 			if (flag & EXTATTR_LIST_LENPREFIX) {
   1239 				uint8_t len = (uint8_t)attrnamelen;
   1240 
   1241 				/* Copy leading name length */
   1242 				error = uiomove(&len, sizeof(len), uio);
   1243 				if (error != 0)
   1244 					break;
   1245 			} else {
   1246 				/* Include trailing NULL */
   1247 				attrnamelen++;
   1248 			}
   1249 
   1250 			error = uiomove(uele->uele_attrname,
   1251 			    (size_t)attrnamelen, uio);
   1252 			if (error != 0)
   1253 				break;
   1254 		}
   1255 
   1256 		if (uele->uele_backing_vnode != vp)
   1257 			VOP_UNLOCK(uele->uele_backing_vnode);
   1258 
   1259 		if (error != 0)
   1260 			return error;
   1261 	}
   1262 
   1263 	if (uio != NULL)
   1264 		uio->uio_offset = 0;
   1265 
   1266 	/* Return full data size if caller requested it. */
   1267 	if (size != NULL)
   1268 		*size = listsize;
   1269 
   1270 	return 0;
   1271 }
   1272 
   1273 /*
   1274  * Vnode operation to remove a named attribute.
   1275  */
   1276 int
   1277 ufs_deleteextattr(struct vop_deleteextattr_args *ap)
   1278 /*
   1279 vop_deleteextattr {
   1280 	IN struct vnode *a_vp;
   1281 	IN int a_attrnamespace;
   1282 	IN const char *a_name;
   1283 	IN kauth_cred_t a_cred;
   1284 };
   1285 */
   1286 {
   1287 	struct mount *mp = ap->a_vp->v_mount;
   1288 	struct ufsmount *ump = VFSTOUFS(mp);
   1289 	int error;
   1290 
   1291 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1292 		return SET_ERROR(EOPNOTSUPP);
   1293 
   1294 	ufs_extattr_uepm_lock(ump);
   1295 
   1296 	error = ufs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1297 	    ap->a_cred, curlwp);
   1298 
   1299 	ufs_extattr_uepm_unlock(ump);
   1300 
   1301 	return error;
   1302 }
   1303 
   1304 /*
   1305  * Vnode operation to set a named attribute.
   1306  */
   1307 int
   1308 ufs_setextattr(struct vop_setextattr_args *ap)
   1309 /*
   1310 vop_setextattr {
   1311 	IN struct vnode *a_vp;
   1312 	IN int a_attrnamespace;
   1313 	IN const char *a_name;
   1314 	INOUT struct uio *a_uio;
   1315 	IN kauth_cred_t a_cred;
   1316 };
   1317 */
   1318 {
   1319 	struct mount *mp = ap->a_vp->v_mount;
   1320 	struct ufsmount *ump = VFSTOUFS(mp);
   1321 	int error;
   1322 
   1323 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1324 		return SET_ERROR(EOPNOTSUPP);
   1325 
   1326 	ufs_extattr_uepm_lock(ump);
   1327 
   1328 	/*
   1329 	 * XXX: No longer a supported way to delete extended attributes.
   1330 	 */
   1331 	if (ap->a_uio == NULL) {
   1332 		ufs_extattr_uepm_unlock(ump);
   1333 		return SET_ERROR(EINVAL);
   1334 	}
   1335 
   1336 	error = ufs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1337 	    ap->a_uio, ap->a_cred, curlwp);
   1338 
   1339 	ufs_extattr_uepm_unlock(ump);
   1340 
   1341 	return error;
   1342 }
   1343 
   1344 /*
   1345  * Real work associated with setting a vnode's extended attributes;
   1346  * assumes that the attribute lock has already been grabbed.
   1347  */
   1348 static int
   1349 ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
   1350     struct uio *uio, kauth_cred_t cred, struct lwp *l)
   1351 {
   1352 	struct ufs_extattr_list_entry *attribute;
   1353 	struct ufs_extattr_header ueh;
   1354 	struct iovec local_aiov;
   1355 	struct uio local_aio;
   1356 	struct mount *mp = vp->v_mount;
   1357 	struct ufsmount *ump = VFSTOUFS(mp);
   1358 	struct inode *ip = VTOI(vp);
   1359 	off_t base_offset;
   1360 	int error = 0, ioflag;
   1361 
   1362 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1363 		return SET_ERROR(EROFS);
   1364 
   1365 	if (!ufs_extattr_valid_attrname(attrnamespace, name))
   1366 		return SET_ERROR(EINVAL);
   1367 
   1368 	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
   1369 	if (error)
   1370 		return error;
   1371 
   1372 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1373 	if (!attribute) {
   1374 		error = ufs_extattr_autocreate_attr(vp, attrnamespace,
   1375 		    name, l, &attribute);
   1376 		if (error == EEXIST) {
   1377 			/* Another thread raced us for backend creation */
   1378 			error = 0;
   1379 			attribute =
   1380 			    ufs_extattr_find_attr(ump, attrnamespace, name);
   1381 		}
   1382 
   1383 		if (error || !attribute)
   1384 			return SET_ERROR(ENODATA);
   1385 	}
   1386 
   1387 	/*
   1388 	 * Early rejection of invalid offsets/length.
   1389 	 * Reject: any offset but 0 (replace)
   1390 	 *	 Any size greater than attribute size limit
   1391  	 */
   1392 	if (uio->uio_offset != 0 ||
   1393 	    uio->uio_resid > attribute->uele_fileheader.uef_size)
   1394 		return SET_ERROR(ENXIO);
   1395 
   1396 	/*
   1397 	 * Find base offset of header in file based on file header size, and
   1398 	 * data header size + maximum data size, indexed by inode number.
   1399 	 */
   1400 	base_offset = sizeof(struct ufs_extattr_fileheader) +
   1401 	    ip->i_number * (sizeof(struct ufs_extattr_header) +
   1402 	    attribute->uele_fileheader.uef_size);
   1403 
   1404 	/*
   1405 	 * Write out a data header for the data.
   1406 	 */
   1407 	ueh.ueh_len = ufs_rw32((uint32_t) uio->uio_resid,
   1408 	    UELE_NEEDSWAP(attribute));
   1409 	ueh.ueh_flags = ufs_rw32(UFS_EXTATTR_ATTR_FLAG_INUSE,
   1410 	    UELE_NEEDSWAP(attribute));
   1411 	ueh.ueh_i_gen = ufs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
   1412 	local_aiov.iov_base = &ueh;
   1413 	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
   1414 	local_aio.uio_iov = &local_aiov;
   1415 	local_aio.uio_iovcnt = 1;
   1416 	local_aio.uio_rw = UIO_WRITE;
   1417 	local_aio.uio_offset = base_offset;
   1418 	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
   1419 	UIO_SETUP_SYSSPACE(&local_aio);
   1420 
   1421 	/*
   1422 	 * Don't need to get a lock on the backing file if the setattr is
   1423 	 * being applied to the backing file, as the lock is already held.
   1424 	 */
   1425 	if (attribute->uele_backing_vnode != vp)
   1426 		vn_lock(attribute->uele_backing_vnode,
   1427 		    LK_EXCLUSIVE | LK_RETRY);
   1428 
   1429 	ioflag = IO_NODELOCKED;
   1430 	if (ufs_extattr_sync)
   1431 		ioflag |= IO_SYNC;
   1432 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1433 	    ump->um_extattr.uepm_ucred);
   1434 	if (error)
   1435 		goto vopunlock_exit;
   1436 
   1437 	if (local_aio.uio_resid != 0) {
   1438 		error = SET_ERROR(ENXIO);
   1439 		goto vopunlock_exit;
   1440 	}
   1441 
   1442 	/*
   1443 	 * Write out user data.
   1444 	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
   1445 	 */
   1446 	uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header);
   1447 
   1448 	ioflag = IO_NODELOCKED;
   1449 	if (ufs_extattr_sync)
   1450 		ioflag |= IO_SYNC;
   1451 	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
   1452 	    ump->um_extattr.uepm_ucred);
   1453 
   1454  vopunlock_exit:
   1455 	uio->uio_offset = 0;
   1456 
   1457 	if (attribute->uele_backing_vnode != vp)
   1458 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1459 
   1460 	return error;
   1461 }
   1462 
   1463 /*
   1464  * Real work associated with removing an extended attribute from a vnode.
   1465  * Assumes the attribute lock has already been grabbed.
   1466  */
   1467 static int
   1468 ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
   1469     kauth_cred_t cred, struct lwp *l)
   1470 {
   1471 	struct ufs_extattr_list_entry *attribute;
   1472 	struct ufs_extattr_header ueh;
   1473 	struct mount *mp = vp->v_mount;
   1474 	struct ufsmount *ump = VFSTOUFS(mp);
   1475 	struct iovec local_aiov;
   1476 	struct uio local_aio;
   1477 	off_t base_offset;
   1478 	int error = 0, ioflag;
   1479 
   1480 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1481 		return SET_ERROR(EROFS);
   1482 
   1483 	if (!ufs_extattr_valid_attrname(attrnamespace, name))
   1484 		return SET_ERROR(EINVAL);
   1485 
   1486 	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
   1487 	if (error)
   1488 		return error;
   1489 
   1490 	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
   1491 	if (!attribute)
   1492 		return SET_ERROR(ENODATA);
   1493 
   1494 	/*
   1495 	 * Don't need to get a lock on the backing file if the getattr is
   1496 	 * being applied to the backing file, as the lock is already held.
   1497 	 */
   1498 	if (attribute->uele_backing_vnode != vp)
   1499 		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
   1500 
   1501 	error = ufs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1502 	if (error)
   1503 		goto vopunlock_exit;
   1504 
   1505 	/* Flag it as not in use. */
   1506 	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
   1507 	ueh.ueh_len = 0;		/* ...ditto... */
   1508 
   1509 	local_aiov.iov_base = &ueh;
   1510 	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
   1511 	local_aio.uio_iov = &local_aiov;
   1512 	local_aio.uio_iovcnt = 1;
   1513 	local_aio.uio_rw = UIO_WRITE;
   1514 	local_aio.uio_offset = base_offset;
   1515 	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
   1516 	UIO_SETUP_SYSSPACE(&local_aio);
   1517 
   1518 	ioflag = IO_NODELOCKED;
   1519 	if (ufs_extattr_sync)
   1520 		ioflag |= IO_SYNC;
   1521 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1522 	    ump->um_extattr.uepm_ucred);
   1523 	if (error)
   1524 		goto vopunlock_exit;
   1525 
   1526 	if (local_aio.uio_resid != 0)
   1527 		error = SET_ERROR(ENXIO);
   1528 
   1529  vopunlock_exit:
   1530 	VOP_UNLOCK(attribute->uele_backing_vnode);
   1531 
   1532 	return error;
   1533 }
   1534 
   1535 /*
   1536  * Called by UFS when an inode is no longer active and should have its
   1537  * attributes stripped.
   1538  */
   1539 void
   1540 ufs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
   1541 {
   1542 	struct ufs_extattr_list_entry *uele;
   1543 	struct mount *mp = vp->v_mount;
   1544 	struct ufsmount *ump = VFSTOUFS(mp);
   1545 
   1546 	/*
   1547 	 * In that case, we cannot lock. We should not have any active vnodes
   1548 	 * on the fs if this is not yet initialized but is going to be, so
   1549 	 * this can go unlocked.
   1550 	 */
   1551 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
   1552 		return;
   1553 
   1554 	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
   1555 		return;
   1556 
   1557 	ufs_extattr_uepm_lock(ump);
   1558 
   1559 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
   1560 		ufs_extattr_rm(vp, uele->uele_attrnamespace,
   1561 		    uele->uele_attrname, lwp0.l_cred, l);
   1562 
   1563 	ufs_extattr_uepm_unlock(ump);
   1564 }
   1565 
   1566 void
   1567 ufs_extattr_init(void)
   1568 {
   1569 
   1570 }
   1571 
   1572 void
   1573 ufs_extattr_done(void)
   1574 {
   1575 
   1576 }
   1577