Home | History | Annotate | Line # | Download | only in lfs
ulfs_extattr.c revision 1.10
      1 /*	$NetBSD: ulfs_extattr.c,v 1.10 2016/06/20 01:44:05 dholland Exp $	*/
      2 /*  from NetBSD: ufs_extattr.c,v 1.44 2014/11/14 10:09:50 manu  */
      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 1.10 2016/06/20 01:44:05 dholland Exp $");
     53 
     54 #ifdef _KERNEL_OPT
     55 #include "opt_lfs.h"
     56 #endif
     57 
     58 #include <sys/param.h>
     59 #include <sys/systm.h>
     60 #include <sys/reboot.h>
     61 #include <sys/kauth.h>
     62 #include <sys/kernel.h>
     63 #include <sys/namei.h>
     64 #include <sys/kmem.h>
     65 #include <sys/fcntl.h>
     66 #include <sys/lwp.h>
     67 #include <sys/vnode.h>
     68 #include <sys/mount.h>
     69 #include <sys/lock.h>
     70 #include <sys/dirent.h>
     71 #include <sys/extattr.h>
     72 #include <sys/sysctl.h>
     73 
     74 #include <ufs/lfs/ulfs_extattr.h>
     75 #include <ufs/lfs/ulfsmount.h>
     76 #include <ufs/lfs/ulfs_inode.h>
     77 #include <ufs/lfs/ulfs_bswap.h>
     78 #include <ufs/lfs/ulfs_extern.h>
     79 
     80 int ulfs_extattr_sync = 1;
     81 int ulfs_extattr_autocreate = 1024;
     82 
     83 static int	ulfs_extattr_valid_attrname(int attrnamespace,
     84 		    const char *attrname);
     85 static int	ulfs_extattr_enable_with_open(struct ulfsmount *ump,
     86 		    struct vnode *vp, int attrnamespace, const char *attrname,
     87 		    struct lwp *l);
     88 static int	ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
     89 		    const char *attrname, struct vnode *backing_vnode,
     90 		    struct lwp *l);
     91 static int	ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
     92 		    const char *attrname, struct lwp *l);
     93 static int	ulfs_extattr_get(struct vnode *vp, int attrnamespace,
     94 		    const char *name, struct uio *uio, size_t *size,
     95 		    kauth_cred_t cred, struct lwp *l);
     96 static int	ulfs_extattr_list(struct vnode *vp, int attrnamespace,
     97 		    struct uio *uio, size_t *size, int flag,
     98 		    kauth_cred_t cred, struct lwp *l);
     99 static int	ulfs_extattr_set(struct vnode *vp, int attrnamespace,
    100 		    const char *name, struct uio *uio, kauth_cred_t cred,
    101 		    struct lwp *l);
    102 static int	ulfs_extattr_rm(struct vnode *vp, int attrnamespace,
    103 		    const char *name, kauth_cred_t cred, struct lwp *l);
    104 static struct ulfs_extattr_list_entry *ulfs_extattr_find_attr(struct ulfsmount *,
    105 		    int, const char *);
    106 static int	ulfs_extattr_get_header(struct vnode *,
    107 		    struct ulfs_extattr_list_entry *,
    108 		    struct ulfs_extattr_header *, off_t *);
    109 
    110 /*
    111  * Convert a FreeBSD extended attribute and namespace to a consistent string
    112  * representation.
    113  *
    114  * The returned value, if not NULL, is guaranteed to be an allocated object
    115  * of its size as returned by strlen() + 1 and must be freed by the caller.
    116  */
    117 static char *
    118 from_freebsd_extattr(int attrnamespace, const char *attrname)
    119 {
    120 	const char *namespace;
    121 	char *attr;
    122 	size_t len;
    123 
    124 	if (attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
    125 		namespace = "system";
    126 	else if (attrnamespace == EXTATTR_NAMESPACE_USER)
    127 		namespace = "user";
    128 	else
    129 		return NULL;
    130 
    131 	/* <namespace>.<attrname>\0 */
    132 	len = strlen(namespace) + 1 + strlen(attrname) + 1;
    133 
    134 	attr = kmem_alloc(len, KM_SLEEP);
    135 
    136 	snprintf(attr, len, "%s.%s", namespace, attrname);
    137 
    138 	return attr;
    139 }
    140 
    141 /*
    142  * Internal wrapper around a conversion-check-free sequence.
    143  */
    144 static int
    145 internal_extattr_check_cred(vnode_t *vp, int attrnamespace, const char *name,
    146     kauth_cred_t cred, int access_mode)
    147 {
    148 	char *attr;
    149 	int error;
    150 
    151 	attr = from_freebsd_extattr(attrnamespace, name);
    152 	if (attr == NULL)
    153 		return EINVAL;
    154 
    155 	error = extattr_check_cred(vp, attr, cred, access_mode);
    156 
    157 	kmem_free(attr, strlen(attr) + 1);
    158 
    159 	return error;
    160 }
    161 
    162 /*
    163  * Per-FS attribute lock protecting attribute operations.
    164  * XXX Right now there is a lot of lock contention due to having a single
    165  * lock per-FS; really, this should be far more fine-grained.
    166  */
    167 static void
    168 ulfs_extattr_uepm_lock(struct ulfsmount *ump)
    169 {
    170 
    171 	/* XXX Why does this need to be recursive? */
    172 	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
    173 		ump->um_extattr.uepm_lockcnt++;
    174 		return;
    175 	}
    176 	mutex_enter(&ump->um_extattr.uepm_lock);
    177 }
    178 
    179 static void
    180 ulfs_extattr_uepm_unlock(struct ulfsmount *ump)
    181 {
    182 
    183 	if (ump->um_extattr.uepm_lockcnt != 0) {
    184 		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
    185 		ump->um_extattr.uepm_lockcnt--;
    186 		return;
    187 	}
    188 	mutex_exit(&ump->um_extattr.uepm_lock);
    189 }
    190 
    191 /*-
    192  * Determine whether the name passed is a valid name for an actual
    193  * attribute.
    194  *
    195  * Invalid currently consists of:
    196  *	 NULL pointer for attrname
    197  *	 zero-length attrname (used to retrieve application attribute list)
    198  */
    199 static int
    200 ulfs_extattr_valid_attrname(int attrnamespace, const char *attrname)
    201 {
    202 
    203 	if (attrname == NULL)
    204 		return (0);
    205 	if (strlen(attrname) == 0)
    206 		return (0);
    207 	return (1);
    208 }
    209 
    210 /*
    211  * Autocreate an attribute storage
    212  */
    213 static struct ulfs_extattr_list_entry *
    214 ulfs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
    215     const char *attrname, struct lwp *l)
    216 {
    217 	struct mount *mp = vp->v_mount;
    218 	struct ulfsmount *ump = VFSTOULFS(mp);
    219 	struct vnode *backing_vp;
    220 	struct nameidata nd;
    221 	struct pathbuf *pb;
    222 	char *path;
    223 	struct ulfs_extattr_fileheader uef;
    224 	struct ulfs_extattr_list_entry *uele;
    225 	int error;
    226 
    227 	path = PNBUF_GET();
    228 
    229 	/*
    230 	 * We only support system and user namespace autocreation
    231 	 */
    232 	switch (attrnamespace) {
    233 	case EXTATTR_NAMESPACE_SYSTEM:
    234 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    235 			       mp->mnt_stat.f_mntonname,
    236 			       ULFS_EXTATTR_FSROOTSUBDIR,
    237 			       ULFS_EXTATTR_SUBDIR_SYSTEM,
    238 			       attrname);
    239 		break;
    240 	case EXTATTR_NAMESPACE_USER:
    241 		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
    242 			       mp->mnt_stat.f_mntonname,
    243 			       ULFS_EXTATTR_FSROOTSUBDIR,
    244 			       ULFS_EXTATTR_SUBDIR_USER,
    245 			       attrname);
    246 		break;
    247 	default:
    248 		PNBUF_PUT(path);
    249 		return NULL;
    250 		break;
    251 	}
    252 
    253 	/*
    254 	 * XXX unlock/lock should only be done when setting extattr
    255 	 * on backing store or one of its parent directory
    256 	 * including root, but we always do it for now.
    257 	 */
    258 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
    259 	VOP_UNLOCK(vp);
    260 
    261 	pb = pathbuf_create(path);
    262 	NDINIT(&nd, CREATE, LOCKPARENT, pb);
    263 
    264 	error = vn_open(&nd, O_CREAT|O_RDWR, 0600);
    265 
    266 	/*
    267 	 * Reacquire the lock on the vnode
    268 	 */
    269 	KASSERT(VOP_ISLOCKED(vp) == 0);
    270 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    271 
    272 	if (error != 0) {
    273 		pathbuf_destroy(pb);
    274 		PNBUF_PUT(path);
    275 		return NULL;
    276 	}
    277 
    278 	KASSERT(nd.ni_vp != NULL);
    279 	KASSERT(VOP_ISLOCKED(nd.ni_vp) == LK_EXCLUSIVE);
    280 	KASSERT(VOP_ISLOCKED(nd.ni_dvp) == 0);
    281 
    282 	/*
    283  	 * backing_vp is the backing store.
    284 	 */
    285 	backing_vp = nd.ni_vp;
    286 	pathbuf_destroy(pb);
    287 	PNBUF_PUT(path);
    288 
    289 	uef.uef_magic = ULFS_EXTATTR_MAGIC;
    290 	uef.uef_version = ULFS_EXTATTR_VERSION;
    291 	uef.uef_size = ulfs_extattr_autocreate;
    292 
    293 	error = vn_rdwr(UIO_WRITE, backing_vp, &uef, sizeof(uef), 0,
    294 		        UIO_SYSSPACE, IO_NODELOCKED|IO_APPEND,
    295 			l->l_cred, NULL, l);
    296 
    297 	VOP_UNLOCK(backing_vp);
    298 
    299 	if (error != 0) {
    300 		printf("%s: write uef header failed for %s, error = %d\n",
    301 		       __func__, attrname, error);
    302 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    303 		return NULL;
    304 	}
    305 
    306 	/*
    307 	 * Now enable attribute.
    308 	 */
    309 	error = ulfs_extattr_enable(ump,attrnamespace, attrname, backing_vp, l);
    310 	KASSERT(VOP_ISLOCKED(backing_vp) == 0);
    311 
    312 	if (error != 0) {
    313 		printf("%s: enable %s failed, error %d\n",
    314 		       __func__, attrname, error);
    315 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    316 		return NULL;
    317 	}
    318 
    319 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    320 	if (uele == NULL) {
    321 		printf("%s: atttribute %s created but not found!\n",
    322 		       __func__, attrname);
    323 		vn_close(backing_vp, FREAD|FWRITE, l->l_cred);
    324 		return NULL;
    325 	}
    326 
    327 	printf("%s: EA backing store autocreated for %s\n",
    328 	       mp->mnt_stat.f_mntonname, attrname);
    329 
    330 	return uele;
    331 }
    332 
    333 /*
    334  * Locate an attribute given a name and mountpoint.
    335  * Must be holding uepm lock for the mount point.
    336  */
    337 static struct ulfs_extattr_list_entry *
    338 ulfs_extattr_find_attr(struct ulfsmount *ump, int attrnamespace,
    339     const char *attrname)
    340 {
    341 	struct ulfs_extattr_list_entry *search_attribute;
    342 
    343 	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
    344 	    search_attribute != NULL;
    345 	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
    346 		if (!(strncmp(attrname, search_attribute->uele_attrname,
    347 		    ULFS_EXTATTR_MAXEXTATTRNAME)) &&
    348 		    (attrnamespace == search_attribute->uele_attrnamespace)) {
    349 			return (search_attribute);
    350 		}
    351 	}
    352 
    353 	return (0);
    354 }
    355 
    356 /*
    357  * Initialize per-FS structures supporting extended attributes.  Do not
    358  * start extended attributes yet.
    359  */
    360 void
    361 ulfs_extattr_uepm_init(struct ulfs_extattr_per_mount *uepm)
    362 {
    363 
    364 	uepm->uepm_flags = 0;
    365 	uepm->uepm_lockcnt = 0;
    366 
    367 	LIST_INIT(&uepm->uepm_list);
    368 	mutex_init(&uepm->uepm_lock, MUTEX_DEFAULT, IPL_NONE);
    369 	uepm->uepm_flags |= ULFS_EXTATTR_UEPM_INITIALIZED;
    370 }
    371 
    372 /*
    373  * Destroy per-FS structures supporting extended attributes.  Assumes
    374  * that EAs have already been stopped, and will panic if not.
    375  */
    376 void
    377 ulfs_extattr_uepm_destroy(struct ulfs_extattr_per_mount *uepm)
    378 {
    379 
    380 	if (!(uepm->uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    381 		panic("ulfs_extattr_uepm_destroy: not initialized");
    382 
    383 	if ((uepm->uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
    384 		panic("ulfs_extattr_uepm_destroy: called while still started");
    385 
    386 	/*
    387 	 * It's not clear that either order for the next three lines is
    388 	 * ideal, and it should never be a problem if this is only called
    389 	 * during unmount, and with vfs_busy().
    390 	 */
    391 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    392 	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_INITIALIZED;
    393 	mutex_destroy(&uepm->uepm_lock);
    394 }
    395 
    396 /*
    397  * Start extended attribute support on an FS.
    398  */
    399 int
    400 ulfs_extattr_start(struct mount *mp, struct lwp *l)
    401 {
    402 	struct ulfsmount *ump;
    403 	int error = 0;
    404 
    405 	ump = VFSTOULFS(mp);
    406 
    407 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
    408 		ulfs_extattr_uepm_init(&ump->um_extattr);
    409 
    410 	ulfs_extattr_uepm_lock(ump);
    411 
    412 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED)) {
    413 		error = EOPNOTSUPP;
    414 		goto unlock;
    415 	}
    416 	if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
    417 		error = EBUSY;
    418 		goto unlock;
    419 	}
    420 
    421 	ump->um_extattr.uepm_flags |= ULFS_EXTATTR_UEPM_STARTED;
    422 
    423 	ump->um_extattr.uepm_ucred = l->l_cred;
    424 	kauth_cred_hold(ump->um_extattr.uepm_ucred);
    425 
    426  unlock:
    427 	ulfs_extattr_uepm_unlock(ump);
    428 
    429 	return (error);
    430 }
    431 
    432 /*
    433  * Helper routine: given a locked parent directory and filename, return
    434  * the locked vnode of the inode associated with the name.  Will not
    435  * follow symlinks, may return any type of vnode.  Lock on parent will
    436  * be released even in the event of a failure.  In the event that the
    437  * target is the parent (i.e., "."), there will be two references and
    438  * one lock, requiring the caller to possibly special-case.
    439  */
    440 static int
    441 ulfs_extattr_lookup(struct vnode *start_dvp, int lockparent, const char *dirname,
    442     struct vnode **vp, struct lwp *l)
    443 {
    444 	struct vop_lookup_v2_args vargs;
    445 	struct componentname cnp;
    446 	struct vnode *target_vp;
    447 	char *pnbuf;
    448 	int error;
    449 
    450 	KASSERT(VOP_ISLOCKED(start_dvp) == LK_EXCLUSIVE);
    451 
    452 	pnbuf = PNBUF_GET();
    453 
    454 	memset(&cnp, 0, sizeof(cnp));
    455 	cnp.cn_nameiop = LOOKUP;
    456 	cnp.cn_flags = ISLASTCN | lockparent;
    457 	cnp.cn_cred = l->l_cred;
    458 	cnp.cn_nameptr = pnbuf;
    459 	error = copystr(dirname, pnbuf, MAXPATHLEN, &cnp.cn_namelen);
    460 	if (error) {
    461 		if (lockparent == 0) {
    462 			VOP_UNLOCK(start_dvp);
    463 		}
    464 		PNBUF_PUT(pnbuf);
    465 		printf("ulfs_extattr_lookup: copystr failed\n");
    466 		return (error);
    467 	}
    468 	cnp.cn_namelen--;	/* trim nul termination */
    469 	vargs.a_desc = NULL;
    470 	vargs.a_dvp = start_dvp;
    471 	vargs.a_vpp = &target_vp;
    472 	vargs.a_cnp = &cnp;
    473 	error = ulfs_lookup(&vargs);
    474 	PNBUF_PUT(pnbuf);
    475 	if (error) {
    476 		if (lockparent == 0) {
    477 			VOP_UNLOCK(start_dvp);
    478 		}
    479 		return (error);
    480 	}
    481 #if 0
    482 	if (target_vp == start_dvp)
    483 		panic("ulfs_extattr_lookup: target_vp == start_dvp");
    484 #endif
    485 
    486 	if (target_vp != start_dvp) {
    487 		error = vn_lock(target_vp, LK_EXCLUSIVE);
    488 		if (lockparent == 0)
    489 			VOP_UNLOCK(start_dvp);
    490 		if (error) {
    491 			vrele(target_vp);
    492 			return error;
    493 		}
    494 	}
    495 
    496 	KASSERT(VOP_ISLOCKED(target_vp) == LK_EXCLUSIVE);
    497 	*vp = target_vp;
    498 	return (0);
    499 }
    500 
    501 /*
    502  * Enable an EA using the passed filesystem, backing vnode, attribute name,
    503  * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
    504  * to be locked when passed in.  The vnode will be returned unlocked,
    505  * regardless of success/failure of the function.  As a result, the caller
    506  * will always need to vrele(), but not vput().
    507  */
    508 static int
    509 ulfs_extattr_enable_with_open(struct ulfsmount *ump, struct vnode *vp,
    510     int attrnamespace, const char *attrname, struct lwp *l)
    511 {
    512 	int error;
    513 
    514 	error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred);
    515 	if (error) {
    516 		printf("ulfs_extattr_enable_with_open.VOP_OPEN(): failed "
    517 		    "with %d\n", error);
    518 		VOP_UNLOCK(vp);
    519 		return (error);
    520 	}
    521 
    522 	mutex_enter(vp->v_interlock);
    523 	vp->v_writecount++;
    524 	mutex_exit(vp->v_interlock);
    525 
    526 	vref(vp);
    527 
    528 	VOP_UNLOCK(vp);
    529 
    530 	error = ulfs_extattr_enable(ump, attrnamespace, attrname, vp, l);
    531 	if (error != 0)
    532 		vn_close(vp, FREAD|FWRITE, l->l_cred);
    533 	return (error);
    534 }
    535 
    536 /*
    537  * Given a locked directory vnode, iterate over the names in the directory
    538  * and use ulfs_extattr_lookup() to retrieve locked vnodes of potential
    539  * attribute files.  Then invoke ulfs_extattr_enable_with_open() on each
    540  * to attempt to start the attribute.  Leaves the directory locked on
    541  * exit.
    542  */
    543 static int
    544 ulfs_extattr_iterate_directory(struct ulfsmount *ump, struct vnode *dvp,
    545     int attrnamespace, struct lwp *l)
    546 {
    547 	struct vop_readdir_args vargs;
    548 	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
    549 	struct dirent *dp, *edp;
    550 	struct vnode *attr_vp;
    551 	struct uio auio;
    552 	struct iovec aiov;
    553 	char *dirbuf;
    554 	int error, eofflag = 0;
    555 
    556 	if (dvp->v_type != VDIR)
    557 		return (ENOTDIR);
    558 
    559 	dirbuf = kmem_alloc(LFS_DIRBLKSIZ, KM_SLEEP);
    560 
    561 	auio.uio_iov = &aiov;
    562 	auio.uio_iovcnt = 1;
    563 	auio.uio_rw = UIO_READ;
    564 	auio.uio_offset = 0;
    565 	UIO_SETUP_SYSSPACE(&auio);
    566 
    567 	vargs.a_desc = NULL;
    568 	vargs.a_vp = dvp;
    569 	vargs.a_uio = &auio;
    570 	vargs.a_cred = l->l_cred;
    571 	vargs.a_eofflag = &eofflag;
    572 	vargs.a_ncookies = NULL;
    573 	vargs.a_cookies = NULL;
    574 
    575 	while (!eofflag) {
    576 		auio.uio_resid = LFS_DIRBLKSIZ;
    577 		aiov.iov_base = dirbuf;
    578 		aiov.iov_len = LFS_DIRBLKSIZ;
    579 		error = ulfs_readdir(&vargs);
    580 		if (error) {
    581 			printf("ulfs_extattr_iterate_directory: ulfs_readdir "
    582 			    "%d\n", error);
    583 			return (error);
    584 		}
    585 
    586 		/*
    587 		 * XXXRW: While in LFS, we always get LFS_DIRBLKSIZ returns from
    588 		 * the directory code on success, on other file systems this
    589 		 * may not be the case.  For portability, we should check the
    590 		 * read length on return from ulfs_readdir().
    591 		 */
    592 		edp = (struct dirent *)&dirbuf[LFS_DIRBLKSIZ];
    593 		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
    594 			if (dp->d_reclen == 0)
    595 				break;
    596 			/* Skip "." and ".." */
    597 			if (dp->d_name[0] == '.' &&
    598 			    (dp->d_name[1] == '\0' ||
    599 			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
    600 				goto next;
    601 			error = ulfs_extattr_lookup(dvp, LOCKPARENT,
    602 			    dp->d_name, &attr_vp, l);
    603 			if (error == ENOENT) {
    604 				goto next; /* keep silent */
    605 			} else if (error) {
    606 				printf("ulfs_extattr_iterate_directory: lookup "
    607 				    "%s %d\n", dp->d_name, error);
    608 			} else if (attr_vp == dvp) {
    609 				vrele(attr_vp);
    610 			} else if (attr_vp->v_type != VREG) {
    611 				vput(attr_vp);
    612 			} else {
    613 				error = ulfs_extattr_enable_with_open(ump,
    614 				    attr_vp, attrnamespace, dp->d_name, l);
    615 				vrele(attr_vp);
    616 				if (error) {
    617 					printf("ulfs_extattr_iterate_directory: "
    618 					    "enable %s %d\n", dp->d_name,
    619 					    error);
    620 				} else if (bootverbose) {
    621 					printf("%s: EA %s loaded\n",
    622 					       sbp->f_mntonname, dp->d_name);
    623 				}
    624 			}
    625  next:
    626 			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
    627 			if (dp >= edp)
    628 				break;
    629 		}
    630 	}
    631 	kmem_free(dirbuf, LFS_DIRBLKSIZ);
    632 
    633 	return (0);
    634 }
    635 
    636 /*
    637  * Auto-start of extended attributes, to be executed (optionally) at
    638  * mount-time.
    639  */
    640 int
    641 ulfs_extattr_autostart(struct mount *mp, struct lwp *l)
    642 {
    643 	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
    644 	int error;
    645 
    646 	/*
    647 	 * Does ULFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
    648 	 * If so, automatically start EA's.
    649 	 */
    650 	error = VFS_ROOT(mp, &rvp);
    651 	if (error) {
    652 		printf("ulfs_extattr_autostart.VFS_ROOT() returned %d\n",
    653 		    error);
    654 		return (error);
    655 	}
    656 
    657 	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    658 
    659 	error = ulfs_extattr_lookup(rvp, 0,
    660 	    ULFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
    661 	if (error) {
    662 		/* rvp ref'd but now unlocked */
    663 		KASSERT(VOP_ISLOCKED(rvp) == 0);
    664 		vrele(rvp);
    665 		return (error);
    666 	}
    667 	if (rvp == attr_dvp) {
    668 		/* Should never happen. */
    669 		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
    670 		vrele(attr_dvp);
    671 		vput(rvp);
    672 		return (EINVAL);
    673 	}
    674 	KASSERT(VOP_ISLOCKED(rvp) == 0);
    675 	vrele(rvp);
    676 
    677 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    678 
    679 	if (attr_dvp->v_type != VDIR) {
    680 		printf("ulfs_extattr_autostart: %s != VDIR\n",
    681 		    ULFS_EXTATTR_FSROOTSUBDIR);
    682 		goto return_vput_attr_dvp;
    683 	}
    684 
    685 	error = ulfs_extattr_start(mp, l);
    686 	if (error) {
    687 		printf("ulfs_extattr_autostart: ulfs_extattr_start failed (%d)\n",
    688 		    error);
    689 		goto return_vput_attr_dvp;
    690 	}
    691 
    692 	/*
    693 	 * Look for two subdirectories: ULFS_EXTATTR_SUBDIR_SYSTEM,
    694 	 * ULFS_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 = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    700 	    ULFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, l);
    701 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    702 	if (error == 0) {
    703 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    704 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    705 		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, l);
    706 		if (error)
    707 			printf("ulfs_extattr_iterate_directory returned %d\n",
    708 			    error);
    709 		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
    710 		vput(attr_system_dvp);
    711 	}
    712 
    713 	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
    714 	    ULFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, l);
    715 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    716 	if (error == 0) {
    717 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    718 		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
    719 		    attr_user_dvp, EXTATTR_NAMESPACE_USER, l);
    720 		if (error)
    721 			printf("ulfs_extattr_iterate_directory returned %d\n",
    722 			    error);
    723 		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
    724 		vput(attr_user_dvp);
    725 	}
    726 
    727 	/* Mask startup failures in sub-directories. */
    728 	error = 0;
    729 
    730  return_vput_attr_dvp:
    731 	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
    732 	vput(attr_dvp);
    733 
    734 	return (error);
    735 }
    736 
    737 /*
    738  * Stop extended attribute support on an FS.
    739  */
    740 void
    741 ulfs_extattr_stop(struct mount *mp, struct lwp *l)
    742 {
    743 	struct ulfs_extattr_list_entry *uele;
    744 	struct ulfsmount *ump = VFSTOULFS(mp);
    745 
    746 	ulfs_extattr_uepm_lock(ump);
    747 
    748 	/*
    749 	 * If we haven't been started, no big deal.  Just short-circuit
    750 	 * the processing work.
    751 	 */
    752 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    753 		goto unlock;
    754 	}
    755 
    756 	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
    757 		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
    758 		ulfs_extattr_disable(ump, uele->uele_attrnamespace,
    759 		    uele->uele_attrname, l);
    760 	}
    761 
    762 	ump->um_extattr.uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
    763 
    764 	kauth_cred_free(ump->um_extattr.uepm_ucred);
    765 	ump->um_extattr.uepm_ucred = NULL;
    766 
    767  unlock:
    768 	ulfs_extattr_uepm_unlock(ump);
    769 }
    770 
    771 /*
    772  * Enable a named attribute on the specified filesystem; provide an
    773  * unlocked backing vnode to hold the attribute data.
    774  */
    775 static int
    776 ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
    777     const char *attrname, struct vnode *backing_vnode, struct lwp *l)
    778 {
    779 	struct ulfs_extattr_list_entry *attribute;
    780 	struct iovec aiov;
    781 	struct uio auio;
    782 	int error = 0;
    783 
    784 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    785 		return (EINVAL);
    786 	if (backing_vnode->v_type != VREG)
    787 		return (EINVAL);
    788 
    789 	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
    790 
    791 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
    792 		error = EOPNOTSUPP;
    793 		goto free_exit;
    794 	}
    795 
    796 	if (ulfs_extattr_find_attr(ump, attrnamespace, attrname)) {
    797 		error = EEXIST;
    798 		goto free_exit;
    799 	}
    800 
    801 	strncpy(attribute->uele_attrname, attrname,
    802 	    ULFS_EXTATTR_MAXEXTATTRNAME);
    803 	attribute->uele_attrnamespace = attrnamespace;
    804 	memset(&attribute->uele_fileheader, 0,
    805 	    sizeof(struct ulfs_extattr_fileheader));
    806 
    807 	attribute->uele_backing_vnode = backing_vnode;
    808 
    809 	auio.uio_iov = &aiov;
    810 	auio.uio_iovcnt = 1;
    811 	aiov.iov_base = (void *) &attribute->uele_fileheader;
    812 	aiov.iov_len = sizeof(struct ulfs_extattr_fileheader);
    813 	auio.uio_resid = sizeof(struct ulfs_extattr_fileheader);
    814 	auio.uio_offset = (off_t) 0;
    815 	auio.uio_rw = UIO_READ;
    816 	UIO_SETUP_SYSSPACE(&auio);
    817 
    818 	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
    819 	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
    820 	    ump->um_extattr.uepm_ucred);
    821 
    822 	if (error)
    823 		goto unlock_free_exit;
    824 
    825 	if (auio.uio_resid != 0) {
    826 		printf("ulfs_extattr_enable: malformed attribute header\n");
    827 		error = EINVAL;
    828 		goto unlock_free_exit;
    829 	}
    830 
    831 	/*
    832 	 * Try to determine the byte order of the attribute file.
    833 	 */
    834 	if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    835 		attribute->uele_flags |= UELE_F_NEEDSWAP;
    836 		attribute->uele_fileheader.uef_magic =
    837 		    ulfs_rw32(attribute->uele_fileheader.uef_magic,
    838 			     UELE_NEEDSWAP(attribute));
    839 		if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
    840 			printf("ulfs_extattr_enable: invalid attribute header "
    841 			       "magic\n");
    842 			error = EINVAL;
    843 			goto unlock_free_exit;
    844 		}
    845 	}
    846 	attribute->uele_fileheader.uef_version =
    847 	    ulfs_rw32(attribute->uele_fileheader.uef_version,
    848 		     UELE_NEEDSWAP(attribute));
    849 	attribute->uele_fileheader.uef_size =
    850 	    ulfs_rw32(attribute->uele_fileheader.uef_size,
    851 		     UELE_NEEDSWAP(attribute));
    852 
    853 	if (attribute->uele_fileheader.uef_version != ULFS_EXTATTR_VERSION) {
    854 		printf("ulfs_extattr_enable: incorrect attribute header "
    855 		    "version\n");
    856 		error = EINVAL;
    857 		goto unlock_free_exit;
    858 	}
    859 
    860 	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
    861 	    uele_entries);
    862 
    863 	VOP_UNLOCK(backing_vnode);
    864 	return (0);
    865 
    866  unlock_free_exit:
    867 	VOP_UNLOCK(backing_vnode);
    868 
    869  free_exit:
    870 	kmem_free(attribute, sizeof(*attribute));
    871 	return (error);
    872 }
    873 
    874 /*
    875  * Disable extended attribute support on an FS.
    876  */
    877 static int
    878 ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
    879     const char *attrname, struct lwp *l)
    880 {
    881 	struct ulfs_extattr_list_entry *uele;
    882 	int error = 0;
    883 
    884 	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
    885 		return (EINVAL);
    886 
    887 	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
    888 	if (!uele)
    889 		return (ENODATA);
    890 
    891 	LIST_REMOVE(uele, uele_entries);
    892 
    893 	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
    894 	    l->l_cred);
    895 
    896 	kmem_free(uele, sizeof(*uele));
    897 
    898 	return (error);
    899 }
    900 
    901 /*
    902  * VFS call to manage extended attributes in ULFS.  If filename_vp is
    903  * non-NULL, it must be passed in locked, and regardless of errors in
    904  * processing, will be unlocked.
    905  */
    906 int
    907 ulfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    908     int attrnamespace, const char *attrname)
    909 {
    910 	struct lwp *l = curlwp;
    911 	struct ulfsmount *ump = VFSTOULFS(mp);
    912 	int error;
    913 
    914 	/*
    915 	 * Only privileged processes can configure extended attributes.
    916 	 */
    917 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_FS_EXTATTR,
    918 	    0, mp, NULL, NULL);
    919 	if (error) {
    920 		if (filename_vp != NULL)
    921 			VOP_UNLOCK(filename_vp);
    922 		return (error);
    923 	}
    924 
    925 	switch(cmd) {
    926 	case ULFS_EXTATTR_CMD_START:
    927 		if (filename_vp != NULL) {
    928 			VOP_UNLOCK(filename_vp);
    929 			return (EINVAL);
    930 		}
    931 		if (attrname != NULL)
    932 			return (EINVAL);
    933 
    934 		error = ulfs_extattr_autostart(mp, l);
    935 		return (error);
    936 
    937 	case ULFS_EXTATTR_CMD_STOP:
    938 		if (filename_vp != NULL) {
    939 			VOP_UNLOCK(filename_vp);
    940 			return (EINVAL);
    941 		}
    942 		if (attrname != NULL)
    943 			return (EINVAL);
    944 
    945 		ulfs_extattr_stop(mp, l);
    946 		return (0);
    947 
    948 	case ULFS_EXTATTR_CMD_ENABLE:
    949 		if (filename_vp == NULL)
    950 			return (EINVAL);
    951 		if (attrname == NULL) {
    952 			VOP_UNLOCK(filename_vp);
    953 			return (EINVAL);
    954 		}
    955 
    956 		/*
    957 		 * ulfs_extattr_enable_with_open() will always unlock the
    958 		 * vnode, regardless of failure.
    959 		 */
    960 		ulfs_extattr_uepm_lock(ump);
    961 		error = ulfs_extattr_enable_with_open(ump, filename_vp,
    962 		    attrnamespace, attrname, l);
    963 		ulfs_extattr_uepm_unlock(ump);
    964 		return (error);
    965 
    966 	case ULFS_EXTATTR_CMD_DISABLE:
    967 		if (filename_vp != NULL) {
    968 			VOP_UNLOCK(filename_vp);
    969 			return (EINVAL);
    970 		}
    971 		if (attrname == NULL)
    972 			return (EINVAL);
    973 
    974 		ulfs_extattr_uepm_lock(ump);
    975 		error = ulfs_extattr_disable(ump, attrnamespace, attrname, l);
    976 		ulfs_extattr_uepm_unlock(ump);
    977 		return (error);
    978 
    979 	default:
    980 		return (EINVAL);
    981 	}
    982 }
    983 
    984 /*
    985  * Read extended attribute header for a given vnode and attribute.
    986  * Backing vnode should be locked and unlocked by caller.
    987  */
    988 static int
    989 ulfs_extattr_get_header(struct vnode *vp, struct ulfs_extattr_list_entry *uele,
    990     struct ulfs_extattr_header *ueh, off_t *bap)
    991 {
    992 	struct mount *mp = vp->v_mount;
    993 	struct ulfsmount *ump = VFSTOULFS(mp);
    994 	struct inode *ip = VTOI(vp);
    995 	off_t base_offset;
    996 	struct iovec aiov;
    997 	struct uio aio;
    998 	int error;
    999 
   1000 	/*
   1001 	 * Find base offset of header in file based on file header size, and
   1002 	 * data header size + maximum data size, indexed by inode number.
   1003 	 */
   1004 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
   1005 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
   1006 	    uele->uele_fileheader.uef_size);
   1007 
   1008 	/*
   1009 	 * Read in the data header to see if the data is defined, and if so
   1010 	 * how much.
   1011 	 */
   1012 	memset(ueh, 0, sizeof(struct ulfs_extattr_header));
   1013 	aiov.iov_base = ueh;
   1014 	aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1015 	aio.uio_iov = &aiov;
   1016 	aio.uio_iovcnt = 1;
   1017 	aio.uio_rw = UIO_READ;
   1018 	aio.uio_offset = base_offset;
   1019 	aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1020 	UIO_SETUP_SYSSPACE(&aio);
   1021 
   1022 	error = VOP_READ(uele->uele_backing_vnode, &aio,
   1023 	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1024 	if (error)
   1025 		return error;
   1026 
   1027 	/*
   1028 	 * Attribute headers are kept in file system byte order.
   1029 	 * XXX What about the blob of data?
   1030 	 */
   1031 	ueh->ueh_flags = ulfs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
   1032 	ueh->ueh_len   = ulfs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
   1033 	ueh->ueh_i_gen = ulfs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
   1034 
   1035 	/* Defined? */
   1036 	if ((ueh->ueh_flags & ULFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
   1037 		return ENODATA;
   1038 
   1039 	/* Valid for the current inode generation? */
   1040 	if (ueh->ueh_i_gen != ip->i_gen) {
   1041 		/*
   1042 		 * The inode itself has a different generation number
   1043 		 * than the uele data.  For now, the best solution
   1044 		 * is to coerce this to undefined, and let it get cleaned
   1045 		 * up by the next write or extattrctl clean.
   1046 		 */
   1047 		printf("%s (%s): inode gen inconsistency (%u, %jd)\n",
   1048 		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
   1049 		       (intmax_t)ip->i_gen);
   1050 		return ENODATA;
   1051 	}
   1052 
   1053 	/* Local size consistency check. */
   1054 	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
   1055 		return ENXIO;
   1056 
   1057 	/* Return base offset */
   1058 	if (bap != NULL)
   1059 		*bap = base_offset;
   1060 
   1061 	return 0;
   1062 }
   1063 
   1064 /*
   1065  * Vnode operation to retrieve a named extended attribute.
   1066  */
   1067 int
   1068 ulfs_getextattr(struct vop_getextattr_args *ap)
   1069 /*
   1070 vop_getextattr {
   1071 	IN struct vnode *a_vp;
   1072 	IN int a_attrnamespace;
   1073 	IN const char *a_name;
   1074 	INOUT struct uio *a_uio;
   1075 	OUT size_t *a_size;
   1076 	IN kauth_cred_t a_cred;
   1077 };
   1078 */
   1079 {
   1080 	struct mount *mp = ap->a_vp->v_mount;
   1081 	struct ulfsmount *ump = VFSTOULFS(mp);
   1082 	int error;
   1083 
   1084 	ulfs_extattr_uepm_lock(ump);
   1085 
   1086 	error = ulfs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1087 	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
   1088 
   1089 	ulfs_extattr_uepm_unlock(ump);
   1090 
   1091 	return (error);
   1092 }
   1093 
   1094 /*
   1095  * Real work associated with retrieving a named attribute--assumes that
   1096  * the attribute lock has already been grabbed.
   1097  */
   1098 static int
   1099 ulfs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
   1100     struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
   1101 {
   1102 	struct ulfs_extattr_list_entry *attribute;
   1103 	struct ulfs_extattr_header ueh;
   1104 	struct mount *mp = vp->v_mount;
   1105 	struct ulfsmount *ump = VFSTOULFS(mp);
   1106 	off_t base_offset;
   1107 	size_t len, old_len;
   1108 	int error = 0;
   1109 
   1110 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1111 		return (EOPNOTSUPP);
   1112 
   1113 	if (strlen(name) == 0)
   1114 		return (EINVAL);
   1115 
   1116 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1117 	    VREAD);
   1118 	if (error)
   1119 		return (error);
   1120 
   1121 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1122 	if (!attribute)
   1123 		return (ENODATA);
   1124 
   1125 	/*
   1126 	 * Allow only offsets of zero to encourage the read/replace
   1127 	 * extended attribute semantic.  Otherwise we can't guarantee
   1128 	 * atomicity, as we don't provide locks for extended attributes.
   1129 	 */
   1130 	if (uio != NULL && uio->uio_offset != 0)
   1131 		return (ENXIO);
   1132 
   1133 	/*
   1134 	 * Don't need to get a lock on the backing file if the getattr is
   1135 	 * being applied to the backing file, as the lock is already held.
   1136 	 */
   1137 	if (attribute->uele_backing_vnode != vp)
   1138 		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1139 
   1140 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1141 	if (error)
   1142 		goto vopunlock_exit;
   1143 
   1144 	/* Return full data size if caller requested it. */
   1145 	if (size != NULL)
   1146 		*size = ueh.ueh_len;
   1147 
   1148 	/* Return data if the caller requested it. */
   1149 	if (uio != NULL) {
   1150 		/* Allow for offset into the attribute data. */
   1151 		uio->uio_offset = base_offset + sizeof(struct
   1152 		    ulfs_extattr_header);
   1153 
   1154 		/*
   1155 		 * Figure out maximum to transfer -- use buffer size and
   1156 		 * local data limit.
   1157 		 */
   1158 		len = MIN(uio->uio_resid, ueh.ueh_len);
   1159 		old_len = uio->uio_resid;
   1160 		uio->uio_resid = len;
   1161 
   1162 		error = VOP_READ(attribute->uele_backing_vnode, uio,
   1163 		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
   1164 		if (error)
   1165 			goto vopunlock_exit;
   1166 
   1167 		uio->uio_resid = old_len - (len - uio->uio_resid);
   1168 	}
   1169 
   1170  vopunlock_exit:
   1171 
   1172 	if (uio != NULL)
   1173 		uio->uio_offset = 0;
   1174 
   1175 	if (attribute->uele_backing_vnode != vp)
   1176 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1177 
   1178 	return (error);
   1179 }
   1180 
   1181 /*
   1182  * Vnode operation to list extended attribute for a vnode
   1183  */
   1184 int
   1185 ulfs_listextattr(struct vop_listextattr_args *ap)
   1186 /*
   1187 vop_listextattr {
   1188 	IN struct vnode *a_vp;
   1189 	IN int a_attrnamespace;
   1190 	INOUT struct uio *a_uio;
   1191 	OUT size_t *a_size;
   1192 	IN int flag;
   1193 	IN kauth_cred_t a_cred;
   1194 	struct proc *a_p;
   1195 };
   1196 */
   1197 {
   1198 	struct mount *mp = ap->a_vp->v_mount;
   1199 	struct ulfsmount *ump = VFSTOULFS(mp);
   1200 	int error;
   1201 
   1202 	ulfs_extattr_uepm_lock(ump);
   1203 
   1204 	error = ulfs_extattr_list(ap->a_vp, ap->a_attrnamespace,
   1205 	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
   1206 
   1207 	ulfs_extattr_uepm_unlock(ump);
   1208 
   1209 	return (error);
   1210 }
   1211 
   1212 /*
   1213  * Real work associated with retrieving list of attributes--assumes that
   1214  * the attribute lock has already been grabbed.
   1215  */
   1216 static int
   1217 ulfs_extattr_list(struct vnode *vp, int attrnamespace,
   1218     struct uio *uio, size_t *size, int flag,
   1219     kauth_cred_t cred, struct lwp *l)
   1220 {
   1221 	struct ulfs_extattr_list_entry *uele;
   1222 	struct ulfs_extattr_header ueh;
   1223 	struct mount *mp = vp->v_mount;
   1224 	struct ulfsmount *ump = VFSTOULFS(mp);
   1225 	size_t listsize = 0;
   1226 	int error = 0;
   1227 
   1228 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1229 		return (EOPNOTSUPP);
   1230 
   1231 	/*
   1232 	 * XXX: We can move this inside the loop and iterate on individual
   1233 	 *	attributes.
   1234 	 */
   1235 	error = internal_extattr_check_cred(vp, attrnamespace, "", cred,
   1236 	    VREAD);
   1237 	if (error)
   1238 		return (error);
   1239 
   1240 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
   1241 		unsigned char attrnamelen;
   1242 
   1243 		if (uele->uele_attrnamespace != attrnamespace)
   1244 			continue;
   1245 
   1246 		error = ulfs_extattr_get_header(vp, uele, &ueh, NULL);
   1247 		if (error == ENODATA)
   1248 			continue;
   1249 		if (error != 0)
   1250 			return error;
   1251 
   1252 		/*
   1253 		 * Don't need to get a lock on the backing file if
   1254 		 * the listattr is being applied to the backing file,
   1255 		 * as the lock is already held.
   1256 		 */
   1257 		if (uele->uele_backing_vnode != vp)
   1258 			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
   1259 
   1260 		/*
   1261 		 * +1 for trailing NUL (listxattr flavor)
   1262 		 *  or leading name length (extattr_list_file flavor)
   1263 	 	 */
   1264 		attrnamelen = strlen(uele->uele_attrname);
   1265 		listsize += attrnamelen + 1;
   1266 
   1267 		/* Return data if the caller requested it. */
   1268 		if (uio != NULL) {
   1269 			/*
   1270 			 * We support two flavors. Either NUL-terminated
   1271 			 * strings (a la listxattr), or non NUL-terminated,
   1272 			 * one byte length prefixed strings (for
   1273 			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
   1274 		 	 * that second behavior.
   1275 			 */
   1276 			if (flag & EXTATTR_LIST_LENPREFIX) {
   1277 				uint8_t len = (uint8_t)attrnamelen;
   1278 
   1279 				/* Copy leading name length */
   1280 				error = uiomove(&len, sizeof(len), uio);
   1281 				if (error != 0)
   1282 					break;
   1283 			} else {
   1284 				/* Include trailing NULL */
   1285 				attrnamelen++;
   1286 			}
   1287 
   1288 			error = uiomove(uele->uele_attrname,
   1289 					(size_t)attrnamelen, uio);
   1290 			if (error != 0)
   1291 				break;
   1292 		}
   1293 
   1294 		if (uele->uele_backing_vnode != vp)
   1295 			VOP_UNLOCK(uele->uele_backing_vnode);
   1296 
   1297 		if (error != 0)
   1298 			return error;
   1299 	}
   1300 
   1301 	if (uio != NULL)
   1302 		uio->uio_offset = 0;
   1303 
   1304 	/* Return full data size if caller requested it. */
   1305 	if (size != NULL)
   1306 		*size = listsize;
   1307 
   1308 	return 0;
   1309 }
   1310 
   1311 /*
   1312  * Vnode operation to remove a named attribute.
   1313  */
   1314 int
   1315 ulfs_deleteextattr(struct vop_deleteextattr_args *ap)
   1316 /*
   1317 vop_deleteextattr {
   1318 	IN struct vnode *a_vp;
   1319 	IN int a_attrnamespace;
   1320 	IN const char *a_name;
   1321 	IN kauth_cred_t a_cred;
   1322 };
   1323 */
   1324 {
   1325 	struct mount *mp = ap->a_vp->v_mount;
   1326 	struct ulfsmount *ump = VFSTOULFS(mp);
   1327 	int error;
   1328 
   1329 	ulfs_extattr_uepm_lock(ump);
   1330 
   1331 	error = ulfs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1332 	    ap->a_cred, curlwp);
   1333 
   1334 	ulfs_extattr_uepm_unlock(ump);
   1335 
   1336 	return (error);
   1337 }
   1338 
   1339 /*
   1340  * Vnode operation to set a named attribute.
   1341  */
   1342 int
   1343 ulfs_setextattr(struct vop_setextattr_args *ap)
   1344 /*
   1345 vop_setextattr {
   1346 	IN struct vnode *a_vp;
   1347 	IN int a_attrnamespace;
   1348 	IN const char *a_name;
   1349 	INOUT struct uio *a_uio;
   1350 	IN kauth_cred_t a_cred;
   1351 };
   1352 */
   1353 {
   1354 	struct mount *mp = ap->a_vp->v_mount;
   1355 	struct ulfsmount *ump = VFSTOULFS(mp);
   1356 	int error;
   1357 
   1358 	ulfs_extattr_uepm_lock(ump);
   1359 
   1360 	/*
   1361 	 * XXX: No longer a supported way to delete extended attributes.
   1362 	 */
   1363 	if (ap->a_uio == NULL) {
   1364 		ulfs_extattr_uepm_unlock(ump);
   1365 		return (EINVAL);
   1366 	}
   1367 
   1368 	error = ulfs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
   1369 	    ap->a_uio, ap->a_cred, curlwp);
   1370 
   1371 	ulfs_extattr_uepm_unlock(ump);
   1372 
   1373 	return (error);
   1374 }
   1375 
   1376 /*
   1377  * Real work associated with setting a vnode's extended attributes;
   1378  * assumes that the attribute lock has already been grabbed.
   1379  */
   1380 static int
   1381 ulfs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
   1382     struct uio *uio, kauth_cred_t cred, struct lwp *l)
   1383 {
   1384 	struct ulfs_extattr_list_entry *attribute;
   1385 	struct ulfs_extattr_header ueh;
   1386 	struct iovec local_aiov;
   1387 	struct uio local_aio;
   1388 	struct mount *mp = vp->v_mount;
   1389 	struct ulfsmount *ump = VFSTOULFS(mp);
   1390 	struct inode *ip = VTOI(vp);
   1391 	off_t base_offset;
   1392 	int error = 0, ioflag;
   1393 
   1394 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1395 		return (EROFS);
   1396 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1397 		return (EOPNOTSUPP);
   1398 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1399 		return (EINVAL);
   1400 
   1401 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1402 	    VWRITE);
   1403 	if (error)
   1404 		return (error);
   1405 
   1406 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1407 	if (!attribute) {
   1408 		attribute =  ulfs_extattr_autocreate_attr(vp, attrnamespace,
   1409 							 name, l);
   1410 		if  (!attribute)
   1411 			return (ENODATA);
   1412 	}
   1413 
   1414 	/*
   1415 	 * Early rejection of invalid offsets/length.
   1416 	 * Reject: any offset but 0 (replace)
   1417 	 *	 Any size greater than attribute size limit
   1418  	 */
   1419 	if (uio->uio_offset != 0 ||
   1420 	    uio->uio_resid > attribute->uele_fileheader.uef_size)
   1421 		return (ENXIO);
   1422 
   1423 	/*
   1424 	 * Find base offset of header in file based on file header size, and
   1425 	 * data header size + maximum data size, indexed by inode number.
   1426 	 */
   1427 	base_offset = sizeof(struct ulfs_extattr_fileheader) +
   1428 	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
   1429 	    attribute->uele_fileheader.uef_size);
   1430 
   1431 	/*
   1432 	 * Write out a data header for the data.
   1433 	 */
   1434 	ueh.ueh_len = ulfs_rw32((uint32_t) uio->uio_resid,
   1435 	    UELE_NEEDSWAP(attribute));
   1436 	ueh.ueh_flags = ulfs_rw32(ULFS_EXTATTR_ATTR_FLAG_INUSE,
   1437 				 UELE_NEEDSWAP(attribute));
   1438 	ueh.ueh_i_gen = ulfs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
   1439 	local_aiov.iov_base = &ueh;
   1440 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1441 	local_aio.uio_iov = &local_aiov;
   1442 	local_aio.uio_iovcnt = 1;
   1443 	local_aio.uio_rw = UIO_WRITE;
   1444 	local_aio.uio_offset = base_offset;
   1445 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1446 	UIO_SETUP_SYSSPACE(&local_aio);
   1447 
   1448 	/*
   1449 	 * Don't need to get a lock on the backing file if the setattr is
   1450 	 * being applied to the backing file, as the lock is already held.
   1451 	 */
   1452 	if (attribute->uele_backing_vnode != vp)
   1453 		vn_lock(attribute->uele_backing_vnode,
   1454 		    LK_EXCLUSIVE | LK_RETRY);
   1455 
   1456 	ioflag = IO_NODELOCKED;
   1457 	if (ulfs_extattr_sync)
   1458 		ioflag |= IO_SYNC;
   1459 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1460 	    ump->um_extattr.uepm_ucred);
   1461 	if (error)
   1462 		goto vopunlock_exit;
   1463 
   1464 	if (local_aio.uio_resid != 0) {
   1465 		error = ENXIO;
   1466 		goto vopunlock_exit;
   1467 	}
   1468 
   1469 	/*
   1470 	 * Write out user data.
   1471 	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
   1472 	 */
   1473 	uio->uio_offset = base_offset + sizeof(struct ulfs_extattr_header);
   1474 
   1475 	ioflag = IO_NODELOCKED;
   1476 	if (ulfs_extattr_sync)
   1477 		ioflag |= IO_SYNC;
   1478 	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
   1479 	    ump->um_extattr.uepm_ucred);
   1480 
   1481  vopunlock_exit:
   1482 	uio->uio_offset = 0;
   1483 
   1484 	if (attribute->uele_backing_vnode != vp)
   1485 		VOP_UNLOCK(attribute->uele_backing_vnode);
   1486 
   1487 	return (error);
   1488 }
   1489 
   1490 /*
   1491  * Real work associated with removing an extended attribute from a vnode.
   1492  * Assumes the attribute lock has already been grabbed.
   1493  */
   1494 static int
   1495 ulfs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
   1496     kauth_cred_t cred, struct lwp *l)
   1497 {
   1498 	struct ulfs_extattr_list_entry *attribute;
   1499 	struct ulfs_extattr_header ueh;
   1500 	struct mount *mp = vp->v_mount;
   1501 	struct ulfsmount *ump = VFSTOULFS(mp);
   1502 	struct iovec local_aiov;
   1503 	struct uio local_aio;
   1504 	off_t base_offset;
   1505 	int error = 0, ioflag;
   1506 
   1507 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
   1508 		return (EROFS);
   1509 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
   1510 		return (EOPNOTSUPP);
   1511 	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
   1512 		return (EINVAL);
   1513 
   1514 	error = internal_extattr_check_cred(vp, attrnamespace, name, cred,
   1515 	    VWRITE);
   1516 	if (error)
   1517 		return (error);
   1518 
   1519 	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
   1520 	if (!attribute)
   1521 		return (ENODATA);
   1522 
   1523 	/*
   1524 	 * Don't need to get a lock on the backing file if the getattr is
   1525 	 * being applied to the backing file, as the lock is already held.
   1526 	 */
   1527 	if (attribute->uele_backing_vnode != vp)
   1528 		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
   1529 
   1530 	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
   1531 	if (error)
   1532 		goto vopunlock_exit;
   1533 
   1534 	/* Flag it as not in use. */
   1535 	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
   1536 	ueh.ueh_len = 0;		/* ...ditto... */
   1537 
   1538 	local_aiov.iov_base = &ueh;
   1539 	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
   1540 	local_aio.uio_iov = &local_aiov;
   1541 	local_aio.uio_iovcnt = 1;
   1542 	local_aio.uio_rw = UIO_WRITE;
   1543 	local_aio.uio_offset = base_offset;
   1544 	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
   1545 	UIO_SETUP_SYSSPACE(&local_aio);
   1546 
   1547 	ioflag = IO_NODELOCKED;
   1548 	if (ulfs_extattr_sync)
   1549 		ioflag |= IO_SYNC;
   1550 	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
   1551 	    ump->um_extattr.uepm_ucred);
   1552 	if (error)
   1553 		goto vopunlock_exit;
   1554 
   1555 	if (local_aio.uio_resid != 0)
   1556 		error = ENXIO;
   1557 
   1558  vopunlock_exit:
   1559 	VOP_UNLOCK(attribute->uele_backing_vnode);
   1560 
   1561 	return (error);
   1562 }
   1563 
   1564 /*
   1565  * Called by ULFS when an inode is no longer active and should have its
   1566  * attributes stripped.
   1567  */
   1568 void
   1569 ulfs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
   1570 {
   1571 	struct ulfs_extattr_list_entry *uele;
   1572 	struct mount *mp = vp->v_mount;
   1573 	struct ulfsmount *ump = VFSTOULFS(mp);
   1574 
   1575 	/*
   1576 	 * In that case, we cannot lock. We should not have any active vnodes
   1577 	 * on the fs if this is not yet initialized but is going to be, so
   1578 	 * this can go unlocked.
   1579 	 */
   1580 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
   1581 		return;
   1582 
   1583 	ulfs_extattr_uepm_lock(ump);
   1584 
   1585 	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
   1586 		ulfs_extattr_uepm_unlock(ump);
   1587 		return;
   1588 	}
   1589 
   1590 	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
   1591 		ulfs_extattr_rm(vp, uele->uele_attrnamespace,
   1592 		    uele->uele_attrname, lwp0.l_cred, l);
   1593 
   1594 	ulfs_extattr_uepm_unlock(ump);
   1595 }
   1596 
   1597 void
   1598 ulfs_extattr_init(void)
   1599 {
   1600 
   1601 }
   1602 
   1603 void
   1604 ulfs_extattr_done(void)
   1605 {
   1606 
   1607 }
   1608