1/*	$NetBSD: ulfs_extattr.c,v 1.18 2024/02/10 18:43:53 andvar Exp $	*/
2/*  from NetBSD: ulfs_extattr.c,v 1.48 2016/11/09 05:08:35 dholland Exp  */
3
4/*-
5 * Copyright (c) 1999-2002 Robert N. M. Watson
6 * Copyright (c) 2002-2003 Networks Associates Technology, Inc.
7 * All rights reserved.
8 *
9 * This software was developed by Robert Watson for the TrustedBSD Project.
10 *
11 * This software was developed for the FreeBSD Project in part by Network
12 * Associates Laboratories, the Security Research Division of Network
13 * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"),
14 * as part of the DARPA CHATS research program.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 *    notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 *    notice, this list of conditions and the following disclaimer in the
23 *    documentation and/or other materials provided with the distribution.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 */
38
39/*
40 * Support for file system extended attributes on the ULFS1 file system.
41 *
42 * Extended attributes are defined in the form name=value, where name is
43 * a nul-terminated string in the style of a file name, and value is a
44 * binary blob of zero or more bytes.  The ULFS1 extended attribute service
45 * layers support for extended attributes onto a backing file, in the style
46 * of the quota implementation, meaning that it requires no underlying format
47 * changes to the file system.  This design choice exchanges simplicity,
48 * usability, and easy deployment for performance.
49 */
50
51#include <sys/cdefs.h>
52__KERNEL_RCSID(0, "$NetBSD: ulfs_extattr.c,v 1.18 2024/02/10 18:43:53 andvar 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
80int ulfs_extattr_sync = 1;
81int ulfs_extattr_autocreate = 1024;
82
83static int	ulfs_extattr_valid_attrname(int attrnamespace,
84		    const char *attrname);
85static int	ulfs_extattr_enable_with_open(struct ulfsmount *ump,
86		    struct vnode *vp, int attrnamespace, const char *attrname,
87		    struct lwp *l);
88static int	ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
89		    const char *attrname, struct vnode *backing_vnode,
90		    struct lwp *l);
91static int	ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
92		    const char *attrname, struct lwp *l);
93static 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);
96static 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);
99static int	ulfs_extattr_set(struct vnode *vp, int attrnamespace,
100		    const char *name, struct uio *uio, kauth_cred_t cred,
101		    struct lwp *l);
102static int	ulfs_extattr_rm(struct vnode *vp, int attrnamespace,
103		    const char *name, kauth_cred_t cred, struct lwp *l);
104static struct ulfs_extattr_list_entry *ulfs_extattr_find_attr(struct ulfsmount *,
105		    int, const char *);
106static int	ulfs_extattr_get_header(struct vnode *,
107		    struct ulfs_extattr_list_entry *,
108		    struct ulfs_extattr_header *, off_t *);
109
110/*
111 * Per-FS attribute lock protecting attribute operations.
112 * XXX Right now there is a lot of lock contention due to having a single
113 * lock per-FS; really, this should be far more fine-grained.
114 */
115static void
116ulfs_extattr_uepm_lock(struct ulfsmount *ump)
117{
118
119	/*
120	 * XXX This needs to be recursive for the following reasons:
121	 *   - it is taken in ulfs_extattr_vnode_inactive
122	 *   - which is called from VOP_INACTIVE
123	 *   - which can be triggered by any vrele, vput, or vn_close
124	 *   - several of these can happen while it's held
125	 */
126	if (mutex_owned(&ump->um_extattr.uepm_lock)) {
127		ump->um_extattr.uepm_lockcnt++;
128		return;
129	}
130	mutex_enter(&ump->um_extattr.uepm_lock);
131}
132
133static void
134ulfs_extattr_uepm_unlock(struct ulfsmount *ump)
135{
136
137	if (ump->um_extattr.uepm_lockcnt != 0) {
138		KASSERT(mutex_owned(&ump->um_extattr.uepm_lock));
139		ump->um_extattr.uepm_lockcnt--;
140		return;
141	}
142	mutex_exit(&ump->um_extattr.uepm_lock);
143}
144
145/*-
146 * Determine whether the name passed is a valid name for an actual
147 * attribute.
148 *
149 * Invalid currently consists of:
150 *	 NULL pointer for attrname
151 *	 zero-length attrname (used to retrieve application attribute list)
152 */
153static int
154ulfs_extattr_valid_attrname(int attrnamespace, const char *attrname)
155{
156
157	if (attrname == NULL)
158		return (0);
159	if (strlen(attrname) == 0)
160		return (0);
161	return (1);
162}
163
164/*
165 * Autocreate an attribute storage
166 */
167static int
168ulfs_extattr_autocreate_attr(struct vnode *vp, int attrnamespace,
169    const char *attrname, struct lwp *l, struct ulfs_extattr_list_entry **uelep)
170{
171	struct mount *mp = vp->v_mount;
172	struct ulfsmount *ump = VFSTOULFS(mp);
173	struct vnode *backing_vp;
174	struct pathbuf *pb;
175	char *path;
176	struct ulfs_extattr_fileheader uef;
177	struct ulfs_extattr_list_entry *uele;
178	int error;
179
180	path = PNBUF_GET();
181
182	/*
183	 * We only support system and user namespace autocreation
184	 */
185	switch (attrnamespace) {
186	case EXTATTR_NAMESPACE_SYSTEM:
187		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
188			       mp->mnt_stat.f_mntonname,
189			       ULFS_EXTATTR_FSROOTSUBDIR,
190			       ULFS_EXTATTR_SUBDIR_SYSTEM,
191			       attrname);
192		break;
193	case EXTATTR_NAMESPACE_USER:
194		(void)snprintf(path, PATH_MAX, "%s/%s/%s/%s",
195			       mp->mnt_stat.f_mntonname,
196			       ULFS_EXTATTR_FSROOTSUBDIR,
197			       ULFS_EXTATTR_SUBDIR_USER,
198			       attrname);
199		break;
200	default:
201		PNBUF_PUT(path);
202		*uelep = NULL;
203		return 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	 * ulfs_extattr_uepm_lock(ump), for instance
212	 * in ulfs_getextattr().
213	 */
214	ulfs_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 ulfs_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	ulfs_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 = ULFS_EXTATTR_MAGIC;
258	uef.uef_version = ULFS_EXTATTR_VERSION;
259	uef.uef_size = ulfs_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, error = %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 = ulfs_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, error %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 = ulfs_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 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 */
309static struct ulfs_extattr_list_entry *
310ulfs_extattr_find_attr(struct ulfsmount *ump, int attrnamespace,
311    const char *attrname)
312{
313	struct ulfs_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		    ULFS_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 */
332void
333ulfs_extattr_uepm_init(struct ulfs_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 |= ULFS_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 */
348void
349ulfs_extattr_uepm_destroy(struct ulfs_extattr_per_mount *uepm)
350{
351
352	if (!(uepm->uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
353		panic("ulfs_extattr_uepm_destroy: not initialized");
354
355	if ((uepm->uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
356		panic("ulfs_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 &= ~ULFS_EXTATTR_UEPM_STARTED;
364	uepm->uepm_flags &= ~ULFS_EXTATTR_UEPM_INITIALIZED;
365	mutex_destroy(&uepm->uepm_lock);
366}
367
368/*
369 * Start extended attribute support on an FS.
370 */
371int
372ulfs_extattr_start(struct mount *mp, struct lwp *l)
373{
374	struct ulfsmount *ump;
375	int error = 0;
376
377	ump = VFSTOULFS(mp);
378
379	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
380		ulfs_extattr_uepm_init(&ump->um_extattr);
381
382	ulfs_extattr_uepm_lock(ump);
383
384	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED)) {
385		error = EOPNOTSUPP;
386		goto unlock;
387	}
388	if (ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED) {
389		error = EBUSY;
390		goto unlock;
391	}
392
393	ump->um_extattr.uepm_flags |= ULFS_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	ulfs_extattr_uepm_unlock(ump);
400
401	return (error);
402}
403
404/*
405 * Helper routine: given a locked parent directory and filename, return
406 * the locked vnode of the inode associated with the name.  Will not
407 * follow symlinks, may return any type of vnode.  Lock on parent will
408 * be released even in the event of a failure.  In the event that the
409 * target is the parent (i.e., "."), there will be two references and
410 * one lock, requiring the caller to possibly special-case.
411 */
412static int
413ulfs_extattr_lookup(struct vnode *start_dvp, int lockparent, 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("ulfs_extattr_lookup: copystr failed\n");
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 = ulfs_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("ulfs_extattr_lookup: target_vp == start_dvp");
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 */
480static int
481ulfs_extattr_enable_with_open(struct ulfsmount *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("ulfs_extattr_enable_with_open.VOP_OPEN(): failed "
489		    "with %d\n", error);
490		VOP_UNLOCK(vp);
491		return (error);
492	}
493
494	mutex_enter(vp->v_interlock);
495	vp->v_writecount++;
496	mutex_exit(vp->v_interlock);
497
498	vref(vp);
499
500	VOP_UNLOCK(vp);
501
502	error = ulfs_extattr_enable(ump, attrnamespace, attrname, vp, l);
503	if (error != 0)
504		vn_close(vp, FREAD|FWRITE, l->l_cred);
505	return (error);
506}
507
508/*
509 * Given a locked directory vnode, iterate over the names in the directory
510 * and use ulfs_extattr_lookup() to retrieve locked vnodes of potential
511 * attribute files.  Then invoke ulfs_extattr_enable_with_open() on each
512 * to attempt to start the attribute.  Leaves the directory locked on
513 * exit.
514 */
515static int
516ulfs_extattr_iterate_directory(struct ulfsmount *ump, struct vnode *dvp,
517    int attrnamespace, struct lwp *l)
518{
519	struct vop_readdir_args vargs;
520	struct statvfs *sbp = &ump->um_mountp->mnt_stat;
521	struct dirent *dp, *edp;
522	struct vnode *attr_vp;
523	struct uio auio;
524	struct iovec aiov;
525	char *dirbuf;
526	int error, eofflag = 0;
527
528	if (dvp->v_type != VDIR)
529		return (ENOTDIR);
530
531	dirbuf = kmem_alloc(LFS_DIRBLKSIZ, KM_SLEEP);
532
533	auio.uio_iov = &aiov;
534	auio.uio_iovcnt = 1;
535	auio.uio_rw = UIO_READ;
536	auio.uio_offset = 0;
537	UIO_SETUP_SYSSPACE(&auio);
538
539	vargs.a_desc = NULL;
540	vargs.a_vp = dvp;
541	vargs.a_uio = &auio;
542	vargs.a_cred = l->l_cred;
543	vargs.a_eofflag = &eofflag;
544	vargs.a_ncookies = NULL;
545	vargs.a_cookies = NULL;
546
547	while (!eofflag) {
548		auio.uio_resid = LFS_DIRBLKSIZ;
549		aiov.iov_base = dirbuf;
550		aiov.iov_len = LFS_DIRBLKSIZ;
551		error = ulfs_readdir(&vargs);
552		if (error) {
553			printf("ulfs_extattr_iterate_directory: ulfs_readdir "
554			    "%d\n", error);
555			return (error);
556		}
557
558		/*
559		 * XXXRW: While in LFS, we always get LFS_DIRBLKSIZ returns from
560		 * the directory code on success, on other file systems this
561		 * may not be the case.  For portability, we should check the
562		 * read length on return from ulfs_readdir().
563		 */
564		edp = (struct dirent *)&dirbuf[LFS_DIRBLKSIZ];
565		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
566			if (dp->d_reclen == 0)
567				break;
568			/* Skip "." and ".." */
569			if (dp->d_name[0] == '.' &&
570			    (dp->d_name[1] == '\0' ||
571			     (dp->d_name[1] == '.' && dp->d_name[2] == '\0')))
572				goto next;
573			error = ulfs_extattr_lookup(dvp, LOCKPARENT,
574			    dp->d_name, &attr_vp, l);
575			if (error == ENOENT) {
576				goto next; /* keep silent */
577			} else if (error) {
578				printf("ulfs_extattr_iterate_directory: lookup "
579				    "%s %d\n", dp->d_name, error);
580			} else if (attr_vp == dvp) {
581				vrele(attr_vp);
582			} else if (attr_vp->v_type != VREG) {
583				vput(attr_vp);
584			} else {
585				error = ulfs_extattr_enable_with_open(ump,
586				    attr_vp, attrnamespace, dp->d_name, l);
587				vrele(attr_vp);
588				if (error) {
589					printf("ulfs_extattr_iterate_directory: "
590					    "enable %s %d\n", dp->d_name,
591					    error);
592				} else if (bootverbose) {
593					printf("%s: EA %s loaded\n",
594					       sbp->f_mntonname, dp->d_name);
595				}
596			}
597 next:
598			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
599			if (dp >= edp)
600				break;
601		}
602	}
603	kmem_free(dirbuf, LFS_DIRBLKSIZ);
604
605	return (0);
606}
607
608/*
609 * Auto-start of extended attributes, to be executed (optionally) at
610 * mount-time.
611 */
612int
613ulfs_extattr_autostart(struct mount *mp, struct lwp *l)
614{
615	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
616	int error;
617
618	/*
619	 * Does ULFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
620	 * If so, automatically start EA's.
621	 */
622	error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp);
623	if (error) {
624		printf("ulfs_extattr_autostart.VFS_ROOT() returned %d\n",
625		    error);
626		return (error);
627	}
628
629	KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
630
631	error = ulfs_extattr_lookup(rvp, 0,
632	    ULFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, l);
633	if (error) {
634		/* rvp ref'd but now unlocked */
635		KASSERT(VOP_ISLOCKED(rvp) == 0);
636		vrele(rvp);
637		return (error);
638	}
639	if (rvp == attr_dvp) {
640		/* Should never happen. */
641		KASSERT(VOP_ISLOCKED(rvp) == LK_EXCLUSIVE);
642		vrele(attr_dvp);
643		vput(rvp);
644		return (EINVAL);
645	}
646	KASSERT(VOP_ISLOCKED(rvp) == 0);
647	vrele(rvp);
648
649	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
650
651	if (attr_dvp->v_type != VDIR) {
652		printf("ulfs_extattr_autostart: %s != VDIR\n",
653		    ULFS_EXTATTR_FSROOTSUBDIR);
654		goto return_vput_attr_dvp;
655	}
656
657	error = ulfs_extattr_start(mp, l);
658	if (error) {
659		printf("ulfs_extattr_autostart: ulfs_extattr_start failed (%d)\n",
660		    error);
661		goto return_vput_attr_dvp;
662	}
663
664	/*
665	 * Look for two subdirectories: ULFS_EXTATTR_SUBDIR_SYSTEM,
666	 * ULFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
667	 * and start with appropriate type.  Failures in either don't
668	 * result in an over-all failure.  attr_dvp is left locked to
669	 * be cleaned up on exit.
670	 */
671	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
672	    ULFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, l);
673	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
674	if (error == 0) {
675		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
676		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
677		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, l);
678		if (error)
679			printf("ulfs_extattr_iterate_directory returned %d\n",
680			    error);
681		KASSERT(VOP_ISLOCKED(attr_system_dvp) == LK_EXCLUSIVE);
682		vput(attr_system_dvp);
683	}
684
685	error = ulfs_extattr_lookup(attr_dvp, LOCKPARENT,
686	    ULFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, l);
687	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
688	if (error == 0) {
689		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
690		error = ulfs_extattr_iterate_directory(VFSTOULFS(mp),
691		    attr_user_dvp, EXTATTR_NAMESPACE_USER, l);
692		if (error)
693			printf("ulfs_extattr_iterate_directory returned %d\n",
694			    error);
695		KASSERT(VOP_ISLOCKED(attr_user_dvp) == LK_EXCLUSIVE);
696		vput(attr_user_dvp);
697	}
698
699	/* Mask startup failures in sub-directories. */
700	error = 0;
701
702 return_vput_attr_dvp:
703	KASSERT(VOP_ISLOCKED(attr_dvp) == LK_EXCLUSIVE);
704	vput(attr_dvp);
705
706	return (error);
707}
708
709/*
710 * Stop extended attribute support on an FS.
711 */
712void
713ulfs_extattr_stop(struct mount *mp, struct lwp *l)
714{
715	struct ulfs_extattr_list_entry *uele;
716	struct ulfsmount *ump = VFSTOULFS(mp);
717
718	ulfs_extattr_uepm_lock(ump);
719
720	/*
721	 * If we haven't been started, no big deal.  Just short-circuit
722	 * the processing work.
723	 */
724	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
725		goto unlock;
726	}
727
728	while (LIST_FIRST(&ump->um_extattr.uepm_list) != NULL) {
729		uele = LIST_FIRST(&ump->um_extattr.uepm_list);
730		ulfs_extattr_disable(ump, uele->uele_attrnamespace,
731		    uele->uele_attrname, l);
732	}
733
734	ump->um_extattr.uepm_flags &= ~ULFS_EXTATTR_UEPM_STARTED;
735
736	kauth_cred_free(ump->um_extattr.uepm_ucred);
737	ump->um_extattr.uepm_ucred = NULL;
738
739 unlock:
740	ulfs_extattr_uepm_unlock(ump);
741}
742
743/*
744 * Enable a named attribute on the specified filesystem; provide an
745 * unlocked backing vnode to hold the attribute data.
746 */
747static int
748ulfs_extattr_enable(struct ulfsmount *ump, int attrnamespace,
749    const char *attrname, struct vnode *backing_vnode, struct lwp *l)
750{
751	struct ulfs_extattr_list_entry *attribute;
752	struct iovec aiov;
753	struct uio auio;
754	int error = 0;
755
756	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
757		return (EINVAL);
758	if (backing_vnode->v_type != VREG)
759		return (EINVAL);
760
761	attribute = kmem_zalloc(sizeof(*attribute), KM_SLEEP);
762
763	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED)) {
764		error = EOPNOTSUPP;
765		goto free_exit;
766	}
767
768	if (ulfs_extattr_find_attr(ump, attrnamespace, attrname)) {
769		error = EEXIST;
770		goto free_exit;
771	}
772
773	strncpy(attribute->uele_attrname, attrname,
774	    ULFS_EXTATTR_MAXEXTATTRNAME);
775	attribute->uele_attrnamespace = attrnamespace;
776	memset(&attribute->uele_fileheader, 0,
777	    sizeof(struct ulfs_extattr_fileheader));
778
779	attribute->uele_backing_vnode = backing_vnode;
780
781	auio.uio_iov = &aiov;
782	auio.uio_iovcnt = 1;
783	aiov.iov_base = (void *) &attribute->uele_fileheader;
784	aiov.iov_len = sizeof(struct ulfs_extattr_fileheader);
785	auio.uio_resid = sizeof(struct ulfs_extattr_fileheader);
786	auio.uio_offset = (off_t) 0;
787	auio.uio_rw = UIO_READ;
788	UIO_SETUP_SYSSPACE(&auio);
789
790	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
791	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
792	    ump->um_extattr.uepm_ucred);
793
794	if (error)
795		goto unlock_free_exit;
796
797	if (auio.uio_resid != 0) {
798		printf("ulfs_extattr_enable: malformed attribute header\n");
799		error = EINVAL;
800		goto unlock_free_exit;
801	}
802
803	/*
804	 * Try to determine the byte order of the attribute file.
805	 */
806	if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
807		attribute->uele_flags |= UELE_F_NEEDSWAP;
808		attribute->uele_fileheader.uef_magic =
809		    ulfs_rw32(attribute->uele_fileheader.uef_magic,
810			     UELE_NEEDSWAP(attribute));
811		if (attribute->uele_fileheader.uef_magic != ULFS_EXTATTR_MAGIC) {
812			printf("ulfs_extattr_enable: invalid attribute header "
813			       "magic\n");
814			error = EINVAL;
815			goto unlock_free_exit;
816		}
817	}
818	attribute->uele_fileheader.uef_version =
819	    ulfs_rw32(attribute->uele_fileheader.uef_version,
820		     UELE_NEEDSWAP(attribute));
821	attribute->uele_fileheader.uef_size =
822	    ulfs_rw32(attribute->uele_fileheader.uef_size,
823		     UELE_NEEDSWAP(attribute));
824
825	if (attribute->uele_fileheader.uef_version != ULFS_EXTATTR_VERSION) {
826		printf("ulfs_extattr_enable: incorrect attribute header "
827		    "version\n");
828		error = EINVAL;
829		goto unlock_free_exit;
830	}
831
832	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
833	    uele_entries);
834
835	VOP_UNLOCK(backing_vnode);
836	return (0);
837
838 unlock_free_exit:
839	VOP_UNLOCK(backing_vnode);
840
841 free_exit:
842	kmem_free(attribute, sizeof(*attribute));
843	return (error);
844}
845
846/*
847 * Disable extended attribute support on an FS.
848 */
849static int
850ulfs_extattr_disable(struct ulfsmount *ump, int attrnamespace,
851    const char *attrname, struct lwp *l)
852{
853	struct ulfs_extattr_list_entry *uele;
854	int error = 0;
855
856	if (!ulfs_extattr_valid_attrname(attrnamespace, attrname))
857		return (EINVAL);
858
859	uele = ulfs_extattr_find_attr(ump, attrnamespace, attrname);
860	if (!uele)
861		return (ENODATA);
862
863	LIST_REMOVE(uele, uele_entries);
864
865	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
866	    l->l_cred);
867
868	kmem_free(uele, sizeof(*uele));
869
870	return (error);
871}
872
873/*
874 * VFS call to manage extended attributes in ULFS.  If filename_vp is
875 * non-NULL, it must be passed in locked, and regardless of errors in
876 * processing, will be unlocked.
877 */
878int
879ulfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
880    int attrnamespace, const char *attrname)
881{
882	struct lwp *l = curlwp;
883	struct ulfsmount *ump = VFSTOULFS(mp);
884	int error;
885
886	/*
887	 * Only privileged processes can configure extended attributes.
888	 */
889	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_FS_EXTATTR,
890	    0, mp, NULL, NULL);
891	if (error) {
892		if (filename_vp != NULL)
893			VOP_UNLOCK(filename_vp);
894		return (error);
895	}
896
897	switch(cmd) {
898	case ULFS_EXTATTR_CMD_START:
899		if (filename_vp != NULL) {
900			VOP_UNLOCK(filename_vp);
901			return (EINVAL);
902		}
903		if (attrname != NULL)
904			return (EINVAL);
905
906		error = ulfs_extattr_autostart(mp, l);
907		return (error);
908
909	case ULFS_EXTATTR_CMD_STOP:
910		if (filename_vp != NULL) {
911			VOP_UNLOCK(filename_vp);
912			return (EINVAL);
913		}
914		if (attrname != NULL)
915			return (EINVAL);
916
917		ulfs_extattr_stop(mp, l);
918		return (0);
919
920	case ULFS_EXTATTR_CMD_ENABLE:
921		if (filename_vp == NULL)
922			return (EINVAL);
923		if (attrname == NULL) {
924			VOP_UNLOCK(filename_vp);
925			return (EINVAL);
926		}
927
928		/*
929		 * ulfs_extattr_enable_with_open() will always unlock the
930		 * vnode, regardless of failure.
931		 */
932		ulfs_extattr_uepm_lock(ump);
933		error = ulfs_extattr_enable_with_open(ump, filename_vp,
934		    attrnamespace, attrname, l);
935		ulfs_extattr_uepm_unlock(ump);
936		return (error);
937
938	case ULFS_EXTATTR_CMD_DISABLE:
939		if (filename_vp != NULL) {
940			VOP_UNLOCK(filename_vp);
941			return (EINVAL);
942		}
943		if (attrname == NULL)
944			return (EINVAL);
945
946		ulfs_extattr_uepm_lock(ump);
947		error = ulfs_extattr_disable(ump, attrnamespace, attrname, l);
948		ulfs_extattr_uepm_unlock(ump);
949		return (error);
950
951	default:
952		return (EINVAL);
953	}
954}
955
956/*
957 * Read extended attribute header for a given vnode and attribute.
958 * Backing vnode should be locked and unlocked by caller.
959 */
960static int
961ulfs_extattr_get_header(struct vnode *vp, struct ulfs_extattr_list_entry *uele,
962    struct ulfs_extattr_header *ueh, off_t *bap)
963{
964	struct mount *mp = vp->v_mount;
965	struct ulfsmount *ump = VFSTOULFS(mp);
966	struct inode *ip = VTOI(vp);
967	off_t base_offset;
968	struct iovec aiov;
969	struct uio aio;
970	int error;
971
972	/*
973	 * Find base offset of header in file based on file header size, and
974	 * data header size + maximum data size, indexed by inode number.
975	 */
976	base_offset = sizeof(struct ulfs_extattr_fileheader) +
977	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
978	    uele->uele_fileheader.uef_size);
979
980	/*
981	 * Read in the data header to see if the data is defined, and if so
982	 * how much.
983	 */
984	memset(ueh, 0, sizeof(struct ulfs_extattr_header));
985	aiov.iov_base = ueh;
986	aiov.iov_len = sizeof(struct ulfs_extattr_header);
987	aio.uio_iov = &aiov;
988	aio.uio_iovcnt = 1;
989	aio.uio_rw = UIO_READ;
990	aio.uio_offset = base_offset;
991	aio.uio_resid = sizeof(struct ulfs_extattr_header);
992	UIO_SETUP_SYSSPACE(&aio);
993
994	error = VOP_READ(uele->uele_backing_vnode, &aio,
995	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
996	if (error)
997		return error;
998
999	/*
1000	 * Attribute headers are kept in file system byte order.
1001	 * XXX What about the blob of data?
1002	 */
1003	ueh->ueh_flags = ulfs_rw32(ueh->ueh_flags, UELE_NEEDSWAP(uele));
1004	ueh->ueh_len   = ulfs_rw32(ueh->ueh_len, UELE_NEEDSWAP(uele));
1005	ueh->ueh_i_gen = ulfs_rw32(ueh->ueh_i_gen, UELE_NEEDSWAP(uele));
1006
1007	/* Defined? */
1008	if ((ueh->ueh_flags & ULFS_EXTATTR_ATTR_FLAG_INUSE) == 0)
1009		return ENODATA;
1010
1011	/* Valid for the current inode generation? */
1012	if (ueh->ueh_i_gen != ip->i_gen) {
1013		/*
1014		 * The inode itself has a different generation number
1015		 * than the uele data.  For now, the best solution
1016		 * is to coerce this to undefined, and let it get cleaned
1017		 * up by the next write or extattrctl clean.
1018		 */
1019		printf("%s (%s): inode gen inconsistency (%u, %jd)\n",
1020		       __func__,  mp->mnt_stat.f_mntonname, ueh->ueh_i_gen,
1021		       (intmax_t)ip->i_gen);
1022		return ENODATA;
1023	}
1024
1025	/* Local size consistency check. */
1026	if (ueh->ueh_len > uele->uele_fileheader.uef_size)
1027		return ENXIO;
1028
1029	/* Return base offset */
1030	if (bap != NULL)
1031		*bap = base_offset;
1032
1033	return 0;
1034}
1035
1036/*
1037 * Vnode operation to retrieve a named extended attribute.
1038 */
1039int
1040ulfs_getextattr(struct vop_getextattr_args *ap)
1041/*
1042vop_getextattr {
1043	IN struct vnode *a_vp;
1044	IN int a_attrnamespace;
1045	IN const char *a_name;
1046	INOUT struct uio *a_uio;
1047	OUT size_t *a_size;
1048	IN kauth_cred_t a_cred;
1049};
1050*/
1051{
1052	struct mount *mp = ap->a_vp->v_mount;
1053	struct ulfsmount *ump = VFSTOULFS(mp);
1054	int error;
1055
1056	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
1057		return (EOPNOTSUPP);
1058
1059	ulfs_extattr_uepm_lock(ump);
1060
1061	error = ulfs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
1062	    ap->a_uio, ap->a_size, ap->a_cred, curlwp);
1063
1064	ulfs_extattr_uepm_unlock(ump);
1065
1066	return (error);
1067}
1068
1069/*
1070 * Real work associated with retrieving a named attribute--assumes that
1071 * the attribute lock has already been grabbed.
1072 */
1073static int
1074ulfs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
1075    struct uio *uio, size_t *size, kauth_cred_t cred, struct lwp *l)
1076{
1077	struct ulfs_extattr_list_entry *attribute;
1078	struct ulfs_extattr_header ueh;
1079	struct mount *mp = vp->v_mount;
1080	struct ulfsmount *ump = VFSTOULFS(mp);
1081	off_t base_offset;
1082	size_t len, old_len;
1083	int error = 0;
1084
1085	if (strlen(name) == 0)
1086		return (EINVAL);
1087
1088	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
1089	if (error)
1090		return (error);
1091
1092	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
1093	if (!attribute)
1094		return (ENODATA);
1095
1096	/*
1097	 * Allow only offsets of zero to encourage the read/replace
1098	 * extended attribute semantic.  Otherwise we can't guarantee
1099	 * atomicity, as we don't provide locks for extended attributes.
1100	 */
1101	if (uio != NULL && uio->uio_offset != 0)
1102		return (ENXIO);
1103
1104	/*
1105	 * Don't need to get a lock on the backing file if the getattr is
1106	 * being applied to the backing file, as the lock is already held.
1107	 */
1108	if (attribute->uele_backing_vnode != vp)
1109		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
1110
1111	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
1112	if (error)
1113		goto vopunlock_exit;
1114
1115	/* Return full data size if caller requested it. */
1116	if (size != NULL)
1117		*size = ueh.ueh_len;
1118
1119	/* Return data if the caller requested it. */
1120	if (uio != NULL) {
1121		/* Allow for offset into the attribute data. */
1122		uio->uio_offset = base_offset + sizeof(struct
1123		    ulfs_extattr_header);
1124
1125		/*
1126		 * Figure out maximum to transfer -- use buffer size and
1127		 * local data limit.
1128		 */
1129		len = MIN(uio->uio_resid, ueh.ueh_len);
1130		old_len = uio->uio_resid;
1131		uio->uio_resid = len;
1132
1133		error = VOP_READ(attribute->uele_backing_vnode, uio,
1134		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
1135		if (error)
1136			goto vopunlock_exit;
1137
1138		uio->uio_resid = old_len - (len - uio->uio_resid);
1139	}
1140
1141 vopunlock_exit:
1142
1143	if (uio != NULL)
1144		uio->uio_offset = 0;
1145
1146	if (attribute->uele_backing_vnode != vp)
1147		VOP_UNLOCK(attribute->uele_backing_vnode);
1148
1149	return (error);
1150}
1151
1152/*
1153 * Vnode operation to list extended attribute for a vnode
1154 */
1155int
1156ulfs_listextattr(struct vop_listextattr_args *ap)
1157/*
1158vop_listextattr {
1159	IN struct vnode *a_vp;
1160	IN int a_attrnamespace;
1161	INOUT struct uio *a_uio;
1162	OUT size_t *a_size;
1163	IN int flag;
1164	IN kauth_cred_t a_cred;
1165	struct proc *a_p;
1166};
1167*/
1168{
1169	struct mount *mp = ap->a_vp->v_mount;
1170	struct ulfsmount *ump = VFSTOULFS(mp);
1171	int error;
1172
1173	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
1174		return (EOPNOTSUPP);
1175
1176	ulfs_extattr_uepm_lock(ump);
1177
1178	error = ulfs_extattr_list(ap->a_vp, ap->a_attrnamespace,
1179	    ap->a_uio, ap->a_size, ap->a_flag, ap->a_cred, curlwp);
1180
1181	ulfs_extattr_uepm_unlock(ump);
1182
1183	return (error);
1184}
1185
1186/*
1187 * Real work associated with retrieving list of attributes--assumes that
1188 * the attribute lock has already been grabbed.
1189 */
1190static int
1191ulfs_extattr_list(struct vnode *vp, int attrnamespace,
1192    struct uio *uio, size_t *size, int flag,
1193    kauth_cred_t cred, struct lwp *l)
1194{
1195	struct ulfs_extattr_list_entry *uele;
1196	struct ulfs_extattr_header ueh;
1197	struct mount *mp = vp->v_mount;
1198	struct ulfsmount *ump = VFSTOULFS(mp);
1199	size_t listsize = 0;
1200	int error = 0;
1201
1202	/*
1203	 * XXX: We can move this inside the loop and iterate on individual
1204	 *	attributes.
1205	 */
1206	error = extattr_check_cred(vp, attrnamespace, cred, VREAD);
1207	if (error)
1208		return (error);
1209
1210	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries) {
1211		unsigned char attrnamelen;
1212
1213		if (uele->uele_attrnamespace != attrnamespace)
1214			continue;
1215
1216		error = ulfs_extattr_get_header(vp, uele, &ueh, NULL);
1217		if (error == ENODATA)
1218			continue;
1219		if (error != 0)
1220			return error;
1221
1222		/*
1223		 * Don't need to get a lock on the backing file if
1224		 * the listattr is being applied to the backing file,
1225		 * as the lock is already held.
1226		 */
1227		if (uele->uele_backing_vnode != vp)
1228			vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
1229
1230		/*
1231		 * +1 for trailing NUL (listxattr flavor)
1232		 *  or leading name length (extattr_list_file flavor)
1233	 	 */
1234		attrnamelen = strlen(uele->uele_attrname);
1235		listsize += attrnamelen + 1;
1236
1237		/* Return data if the caller requested it. */
1238		if (uio != NULL) {
1239			/*
1240			 * We support two flavors. Either NUL-terminated
1241			 * strings (a la listxattr), or non NUL-terminated,
1242			 * one byte length prefixed strings (for
1243			 * extattr_list_file). EXTATTR_LIST_LENPREFIX switches
1244		 	 * that second behavior.
1245			 */
1246			if (flag & EXTATTR_LIST_LENPREFIX) {
1247				uint8_t len = (uint8_t)attrnamelen;
1248
1249				/* Copy leading name length */
1250				error = uiomove(&len, sizeof(len), uio);
1251				if (error != 0)
1252					break;
1253			} else {
1254				/* Include trailing NULL */
1255				attrnamelen++;
1256			}
1257
1258			error = uiomove(uele->uele_attrname,
1259					(size_t)attrnamelen, uio);
1260			if (error != 0)
1261				break;
1262		}
1263
1264		if (uele->uele_backing_vnode != vp)
1265			VOP_UNLOCK(uele->uele_backing_vnode);
1266
1267		if (error != 0)
1268			return error;
1269	}
1270
1271	if (uio != NULL)
1272		uio->uio_offset = 0;
1273
1274	/* Return full data size if caller requested it. */
1275	if (size != NULL)
1276		*size = listsize;
1277
1278	return 0;
1279}
1280
1281/*
1282 * Vnode operation to remove a named attribute.
1283 */
1284int
1285ulfs_deleteextattr(struct vop_deleteextattr_args *ap)
1286/*
1287vop_deleteextattr {
1288	IN struct vnode *a_vp;
1289	IN int a_attrnamespace;
1290	IN const char *a_name;
1291	IN kauth_cred_t a_cred;
1292};
1293*/
1294{
1295	struct mount *mp = ap->a_vp->v_mount;
1296	struct ulfsmount *ump = VFSTOULFS(mp);
1297	int error;
1298
1299	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
1300		return (EOPNOTSUPP);
1301
1302	ulfs_extattr_uepm_lock(ump);
1303
1304	error = ulfs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
1305	    ap->a_cred, curlwp);
1306
1307	ulfs_extattr_uepm_unlock(ump);
1308
1309	return (error);
1310}
1311
1312/*
1313 * Vnode operation to set a named attribute.
1314 */
1315int
1316ulfs_setextattr(struct vop_setextattr_args *ap)
1317/*
1318vop_setextattr {
1319	IN struct vnode *a_vp;
1320	IN int a_attrnamespace;
1321	IN const char *a_name;
1322	INOUT struct uio *a_uio;
1323	IN kauth_cred_t a_cred;
1324};
1325*/
1326{
1327	struct mount *mp = ap->a_vp->v_mount;
1328	struct ulfsmount *ump = VFSTOULFS(mp);
1329	int error;
1330
1331	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
1332		return (EOPNOTSUPP);
1333
1334	ulfs_extattr_uepm_lock(ump);
1335
1336	/*
1337	 * XXX: No longer a supported way to delete extended attributes.
1338	 */
1339	if (ap->a_uio == NULL) {
1340		ulfs_extattr_uepm_unlock(ump);
1341		return (EINVAL);
1342	}
1343
1344	error = ulfs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
1345	    ap->a_uio, ap->a_cred, curlwp);
1346
1347	ulfs_extattr_uepm_unlock(ump);
1348
1349	return (error);
1350}
1351
1352/*
1353 * Real work associated with setting a vnode's extended attributes;
1354 * assumes that the attribute lock has already been grabbed.
1355 */
1356static int
1357ulfs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
1358    struct uio *uio, kauth_cred_t cred, struct lwp *l)
1359{
1360	struct ulfs_extattr_list_entry *attribute;
1361	struct ulfs_extattr_header ueh;
1362	struct iovec local_aiov;
1363	struct uio local_aio;
1364	struct mount *mp = vp->v_mount;
1365	struct ulfsmount *ump = VFSTOULFS(mp);
1366	struct inode *ip = VTOI(vp);
1367	off_t base_offset;
1368	int error = 0, ioflag;
1369
1370	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1371		return (EROFS);
1372
1373	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
1374		return (EINVAL);
1375
1376	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
1377	if (error)
1378		return (error);
1379
1380	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
1381	if (!attribute) {
1382		error = ulfs_extattr_autocreate_attr(vp, attrnamespace,
1383						    name, l, &attribute);
1384		if (error == EEXIST) {
1385			/* Another thread raced us for backend creation */
1386			error = 0;
1387			attribute =
1388			    ulfs_extattr_find_attr(ump, attrnamespace, name);
1389		}
1390
1391		if (error || !attribute)
1392			return ENODATA;
1393	}
1394
1395	/*
1396	 * Early rejection of invalid offsets/length.
1397	 * Reject: any offset but 0 (replace)
1398	 *	 Any size greater than attribute size limit
1399 	 */
1400	if (uio->uio_offset != 0 ||
1401	    uio->uio_resid > attribute->uele_fileheader.uef_size)
1402		return (ENXIO);
1403
1404	/*
1405	 * Find base offset of header in file based on file header size, and
1406	 * data header size + maximum data size, indexed by inode number.
1407	 */
1408	base_offset = sizeof(struct ulfs_extattr_fileheader) +
1409	    ip->i_number * (sizeof(struct ulfs_extattr_header) +
1410	    attribute->uele_fileheader.uef_size);
1411
1412	/*
1413	 * Write out a data header for the data.
1414	 */
1415	ueh.ueh_len = ulfs_rw32((uint32_t) uio->uio_resid,
1416	    UELE_NEEDSWAP(attribute));
1417	ueh.ueh_flags = ulfs_rw32(ULFS_EXTATTR_ATTR_FLAG_INUSE,
1418				 UELE_NEEDSWAP(attribute));
1419	ueh.ueh_i_gen = ulfs_rw32(ip->i_gen, UELE_NEEDSWAP(attribute));
1420	local_aiov.iov_base = &ueh;
1421	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
1422	local_aio.uio_iov = &local_aiov;
1423	local_aio.uio_iovcnt = 1;
1424	local_aio.uio_rw = UIO_WRITE;
1425	local_aio.uio_offset = base_offset;
1426	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
1427	UIO_SETUP_SYSSPACE(&local_aio);
1428
1429	/*
1430	 * Don't need to get a lock on the backing file if the setattr is
1431	 * being applied to the backing file, as the lock is already held.
1432	 */
1433	if (attribute->uele_backing_vnode != vp)
1434		vn_lock(attribute->uele_backing_vnode,
1435		    LK_EXCLUSIVE | LK_RETRY);
1436
1437	ioflag = IO_NODELOCKED;
1438	if (ulfs_extattr_sync)
1439		ioflag |= IO_SYNC;
1440	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
1441	    ump->um_extattr.uepm_ucred);
1442	if (error)
1443		goto vopunlock_exit;
1444
1445	if (local_aio.uio_resid != 0) {
1446		error = ENXIO;
1447		goto vopunlock_exit;
1448	}
1449
1450	/*
1451	 * Write out user data.
1452	 * XXX NOT ATOMIC WITH RESPECT TO THE HEADER.
1453	 */
1454	uio->uio_offset = base_offset + sizeof(struct ulfs_extattr_header);
1455
1456	ioflag = IO_NODELOCKED;
1457	if (ulfs_extattr_sync)
1458		ioflag |= IO_SYNC;
1459	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
1460	    ump->um_extattr.uepm_ucred);
1461
1462 vopunlock_exit:
1463	uio->uio_offset = 0;
1464
1465	if (attribute->uele_backing_vnode != vp)
1466		VOP_UNLOCK(attribute->uele_backing_vnode);
1467
1468	return (error);
1469}
1470
1471/*
1472 * Real work associated with removing an extended attribute from a vnode.
1473 * Assumes the attribute lock has already been grabbed.
1474 */
1475static int
1476ulfs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
1477    kauth_cred_t cred, struct lwp *l)
1478{
1479	struct ulfs_extattr_list_entry *attribute;
1480	struct ulfs_extattr_header ueh;
1481	struct mount *mp = vp->v_mount;
1482	struct ulfsmount *ump = VFSTOULFS(mp);
1483	struct iovec local_aiov;
1484	struct uio local_aio;
1485	off_t base_offset;
1486	int error = 0, ioflag;
1487
1488	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1489		return (EROFS);
1490
1491	if (!ulfs_extattr_valid_attrname(attrnamespace, name))
1492		return (EINVAL);
1493
1494	error = extattr_check_cred(vp, attrnamespace, cred, VWRITE);
1495	if (error)
1496		return (error);
1497
1498	attribute = ulfs_extattr_find_attr(ump, attrnamespace, name);
1499	if (!attribute)
1500		return (ENODATA);
1501
1502	/*
1503	 * Don't need to get a lock on the backing file if the getattr is
1504	 * being applied to the backing file, as the lock is already held.
1505	 */
1506	if (attribute->uele_backing_vnode != vp)
1507		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
1508
1509	error = ulfs_extattr_get_header(vp, attribute, &ueh, &base_offset);
1510	if (error)
1511		goto vopunlock_exit;
1512
1513	/* Flag it as not in use. */
1514	ueh.ueh_flags = 0;		/* No need to byte swap 0 */
1515	ueh.ueh_len = 0;		/* ...ditto... */
1516
1517	local_aiov.iov_base = &ueh;
1518	local_aiov.iov_len = sizeof(struct ulfs_extattr_header);
1519	local_aio.uio_iov = &local_aiov;
1520	local_aio.uio_iovcnt = 1;
1521	local_aio.uio_rw = UIO_WRITE;
1522	local_aio.uio_offset = base_offset;
1523	local_aio.uio_resid = sizeof(struct ulfs_extattr_header);
1524	UIO_SETUP_SYSSPACE(&local_aio);
1525
1526	ioflag = IO_NODELOCKED;
1527	if (ulfs_extattr_sync)
1528		ioflag |= IO_SYNC;
1529	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
1530	    ump->um_extattr.uepm_ucred);
1531	if (error)
1532		goto vopunlock_exit;
1533
1534	if (local_aio.uio_resid != 0)
1535		error = ENXIO;
1536
1537 vopunlock_exit:
1538	VOP_UNLOCK(attribute->uele_backing_vnode);
1539
1540	return (error);
1541}
1542
1543/*
1544 * Called by ULFS when an inode is no longer active and should have its
1545 * attributes stripped.
1546 */
1547void
1548ulfs_extattr_vnode_inactive(struct vnode *vp, struct lwp *l)
1549{
1550	struct ulfs_extattr_list_entry *uele;
1551	struct mount *mp = vp->v_mount;
1552	struct ulfsmount *ump = VFSTOULFS(mp);
1553
1554	/*
1555	 * In that case, we cannot lock. We should not have any active vnodes
1556	 * on the fs if this is not yet initialized but is going to be, so
1557	 * this can go unlocked.
1558	 */
1559	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_INITIALIZED))
1560		return;
1561
1562	if (!(ump->um_extattr.uepm_flags & ULFS_EXTATTR_UEPM_STARTED))
1563		return;
1564
1565	ulfs_extattr_uepm_lock(ump);
1566
1567	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
1568		ulfs_extattr_rm(vp, uele->uele_attrnamespace,
1569		    uele->uele_attrname, lwp0.l_cred, l);
1570
1571	ulfs_extattr_uepm_unlock(ump);
1572}
1573
1574void
1575ulfs_extattr_init(void)
1576{
1577
1578}
1579
1580void
1581ulfs_extattr_done(void)
1582{
1583
1584}
1585