1/*-
2 * Copyright (c) 1999-2002 Robert N. M. Watson
3 * Copyright (c) 2002-2003 Networks Associates Technology, Inc.
4 * All rights reserved.
5 *
6 * This software was developed by Robert Watson for the TrustedBSD Project.
7 *
8 * This software was developed for the FreeBSD Project in part by Network
9 * Associates Laboratories, the Security Research Division of Network
10 * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"),
11 * as part of the DARPA CHATS research program.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 *    notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 *    notice, this list of conditions and the following disclaimer in the
20 *    documentation and/or other materials provided with the distribution.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 */
35
36/*
37 * Support for filesystem extended attribute: UFS-specific support functions.
38 */
39
40#include <sys/cdefs.h>
41__FBSDID("$FreeBSD$");
42
43#include "opt_ufs.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/kernel.h>
48#include <sys/namei.h>
49#include <sys/malloc.h>
50#include <sys/fcntl.h>
51#include <sys/priv.h>
52#include <sys/proc.h>
53#include <sys/vnode.h>
54#include <sys/mount.h>
55#include <sys/lock.h>
56#include <sys/dirent.h>
57#include <sys/extattr.h>
58#include <sys/sx.h>
59#include <sys/sysctl.h>
60
61#include <vm/uma.h>
62
63#include <ufs/ufs/dir.h>
64#include <ufs/ufs/extattr.h>
65#include <ufs/ufs/quota.h>
66#include <ufs/ufs/ufsmount.h>
67#include <ufs/ufs/inode.h>
68#include <ufs/ufs/ufs_extern.h>
69
70#ifdef UFS_EXTATTR
71
72static MALLOC_DEFINE(M_UFS_EXTATTR, "ufs_extattr", "ufs extended attribute");
73
74static int ufs_extattr_sync = 0;
75SYSCTL_INT(_debug, OID_AUTO, ufs_extattr_sync, CTLFLAG_RW, &ufs_extattr_sync,
76    0, "");
77
78static int	ufs_extattr_valid_attrname(int attrnamespace,
79		    const char *attrname);
80static int	ufs_extattr_enable_with_open(struct ufsmount *ump,
81		    struct vnode *vp, int attrnamespace, const char *attrname,
82		    struct thread *td);
83static int	ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
84		    const char *attrname, struct vnode *backing_vnode,
85		    struct thread *td);
86static int	ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
87		    const char *attrname, struct thread *td);
88static int	ufs_extattr_get(struct vnode *vp, int attrnamespace,
89		    const char *name, struct uio *uio, size_t *size,
90		    struct ucred *cred, struct thread *td);
91static int	ufs_extattr_set(struct vnode *vp, int attrnamespace,
92		    const char *name, struct uio *uio, struct ucred *cred,
93		    struct thread *td);
94static int	ufs_extattr_rm(struct vnode *vp, int attrnamespace,
95		    const char *name, struct ucred *cred, struct thread *td);
96#ifdef UFS_EXTATTR_AUTOSTART
97static int	ufs_extattr_autostart_locked(struct mount *mp,
98		    struct thread *td);
99#endif
100static int	ufs_extattr_start_locked(struct ufsmount *ump,
101		    struct thread *td);
102
103/*
104 * Per-FS attribute lock protecting attribute operations.
105 *
106 * XXXRW: Perhaps something more fine-grained would be appropriate, but at
107 * the end of the day we're going to contend on the vnode lock for the
108 * backing file anyway.
109 */
110static void
111ufs_extattr_uepm_lock(struct ufsmount *ump, struct thread *td)
112{
113
114	sx_xlock(&ump->um_extattr.uepm_lock);
115}
116
117static void
118ufs_extattr_uepm_unlock(struct ufsmount *ump, struct thread *td)
119{
120
121	sx_xunlock(&ump->um_extattr.uepm_lock);
122}
123
124/*-
125 * Determine whether the name passed is a valid name for an actual
126 * attribute.
127 *
128 * Invalid currently consists of:
129 *	 NULL pointer for attrname
130 *	 zero-length attrname (used to retrieve application attribute list)
131 */
132static int
133ufs_extattr_valid_attrname(int attrnamespace, const char *attrname)
134{
135
136	if (attrname == NULL)
137		return (0);
138	if (strlen(attrname) == 0)
139		return (0);
140	return (1);
141}
142
143/*
144 * Locate an attribute given a name and mountpoint.
145 * Must be holding uepm lock for the mount point.
146 */
147static struct ufs_extattr_list_entry *
148ufs_extattr_find_attr(struct ufsmount *ump, int attrnamespace,
149    const char *attrname)
150{
151	struct ufs_extattr_list_entry *search_attribute;
152
153	sx_assert(&ump->um_extattr.uepm_lock, SA_XLOCKED);
154
155	for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
156	    search_attribute != NULL;
157	    search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
158		if (!(strncmp(attrname, search_attribute->uele_attrname,
159		    UFS_EXTATTR_MAXEXTATTRNAME)) &&
160		    (attrnamespace == search_attribute->uele_attrnamespace)) {
161			return (search_attribute);
162		}
163	}
164
165	return (0);
166}
167
168/*
169 * Initialize per-FS structures supporting extended attributes.  Do not
170 * start extended attributes yet.
171 */
172void
173ufs_extattr_uepm_init(struct ufs_extattr_per_mount *uepm)
174{
175
176	uepm->uepm_flags = 0;
177	LIST_INIT(&uepm->uepm_list);
178	sx_init(&uepm->uepm_lock, "ufs_extattr_sx");
179	uepm->uepm_flags |= UFS_EXTATTR_UEPM_INITIALIZED;
180}
181
182/*
183 * Destroy per-FS structures supporting extended attributes.  Assumes
184 * that EAs have already been stopped, and will panic if not.
185 */
186void
187ufs_extattr_uepm_destroy(struct ufs_extattr_per_mount *uepm)
188{
189
190	if (!(uepm->uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
191		panic("ufs_extattr_uepm_destroy: not initialized");
192
193	if ((uepm->uepm_flags & UFS_EXTATTR_UEPM_STARTED))
194		panic("ufs_extattr_uepm_destroy: called while still started");
195
196	/*
197	 * It's not clear that either order for the next two lines is
198	 * ideal, and it should never be a problem if this is only called
199	 * during unmount, and with vfs_busy().
200	 */
201	uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_INITIALIZED;
202	sx_destroy(&uepm->uepm_lock);
203}
204
205/*
206 * Start extended attribute support on an FS.
207 */
208int
209ufs_extattr_start(struct mount *mp, struct thread *td)
210{
211	struct ufsmount *ump;
212	int error = 0;
213
214	ump = VFSTOUFS(mp);
215
216	ufs_extattr_uepm_lock(ump, td);
217	error = ufs_extattr_start_locked(ump, td);
218	ufs_extattr_uepm_unlock(ump, td);
219	return (error);
220}
221
222static int
223ufs_extattr_start_locked(struct ufsmount *ump, struct thread *td)
224{
225	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
226		return (EOPNOTSUPP);
227	if (ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)
228		return (EBUSY);
229
230	ump->um_extattr.uepm_flags |= UFS_EXTATTR_UEPM_STARTED;
231	ump->um_extattr.uepm_ucred = crhold(td->td_ucred);
232	return (0);
233}
234
235#ifdef UFS_EXTATTR_AUTOSTART
236/*
237 * Helper routine: given a locked parent directory and filename, return
238 * the locked vnode of the inode associated with the name.  Will not
239 * follow symlinks, may return any type of vnode.  Lock on parent will
240 * be released even in the event of a failure.  In the event that the
241 * target is the parent (i.e., "."), there will be two references and
242 * one lock, requiring the caller to possibly special-case.
243 */
244#define	UE_GETDIR_LOCKPARENT	1
245#define	UE_GETDIR_LOCKPARENT_DONT	2
246static int
247ufs_extattr_lookup(struct vnode *start_dvp, int lockparent, char *dirname,
248    struct vnode **vp, struct thread *td)
249{
250	struct vop_cachedlookup_args vargs;
251	struct componentname cnp;
252	struct vnode *target_vp;
253	int error;
254
255	bzero(&cnp, sizeof(cnp));
256	cnp.cn_nameiop = LOOKUP;
257	cnp.cn_flags = ISLASTCN;
258	if (lockparent == UE_GETDIR_LOCKPARENT)
259		cnp.cn_flags |= LOCKPARENT;
260	cnp.cn_lkflags = LK_EXCLUSIVE;
261	cnp.cn_thread = td;
262	cnp.cn_cred = td->td_ucred;
263	cnp.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
264	cnp.cn_nameptr = cnp.cn_pnbuf;
265	error = copystr(dirname, cnp.cn_pnbuf, MAXPATHLEN,
266	    (size_t *) &cnp.cn_namelen);
267	if (error) {
268		if (lockparent == UE_GETDIR_LOCKPARENT_DONT) {
269			VOP_UNLOCK(start_dvp, 0);
270		}
271		uma_zfree(namei_zone, cnp.cn_pnbuf);
272		printf("ufs_extattr_lookup: copystr failed\n");
273		return (error);
274	}
275	cnp.cn_namelen--;	/* trim nul termination */
276	vargs.a_gen.a_desc = NULL;
277	vargs.a_dvp = start_dvp;
278	vargs.a_vpp = &target_vp;
279	vargs.a_cnp = &cnp;
280	error = ufs_lookup(&vargs);
281	uma_zfree(namei_zone, cnp.cn_pnbuf);
282	if (error) {
283		/*
284		 * Error condition, may have to release the lock on the parent
285		 * if ufs_lookup() didn't.
286		 */
287		if (lockparent == UE_GETDIR_LOCKPARENT_DONT)
288			VOP_UNLOCK(start_dvp, 0);
289
290		/*
291		 * Check that ufs_lookup() didn't release the lock when we
292		 * didn't want it to.
293		 */
294		if (lockparent == UE_GETDIR_LOCKPARENT)
295			ASSERT_VOP_LOCKED(start_dvp, "ufs_extattr_lookup");
296
297		return (error);
298	}
299/*
300	if (target_vp == start_dvp)
301		panic("ufs_extattr_lookup: target_vp == start_dvp");
302*/
303
304	if (target_vp != start_dvp && lockparent == UE_GETDIR_LOCKPARENT_DONT)
305		VOP_UNLOCK(start_dvp, 0);
306
307	if (lockparent == UE_GETDIR_LOCKPARENT)
308		ASSERT_VOP_LOCKED(start_dvp, "ufs_extattr_lookup");
309
310	/* printf("ufs_extattr_lookup: success\n"); */
311	*vp = target_vp;
312	return (0);
313}
314#endif /* !UFS_EXTATTR_AUTOSTART */
315
316/*
317 * Enable an EA using the passed filesystem, backing vnode, attribute name,
318 * namespace, and proc.  Will perform a VOP_OPEN() on the vp, so expects vp
319 * to be locked when passed in.  The vnode will be returned unlocked,
320 * regardless of success/failure of the function.  As a result, the caller
321 * will always need to vrele(), but not vput().
322 */
323static int
324ufs_extattr_enable_with_open(struct ufsmount *ump, struct vnode *vp,
325    int attrnamespace, const char *attrname, struct thread *td)
326{
327	int error;
328
329	error = VOP_OPEN(vp, FREAD|FWRITE, td->td_ucred, td, NULL);
330	if (error) {
331		printf("ufs_extattr_enable_with_open.VOP_OPEN(): failed "
332		    "with %d\n", error);
333		VOP_UNLOCK(vp, 0);
334		return (error);
335	}
336
337	VOP_ADD_WRITECOUNT(vp, 1);
338
339	vref(vp);
340
341	VOP_UNLOCK(vp, 0);
342
343	error = ufs_extattr_enable(ump, attrnamespace, attrname, vp, td);
344	if (error != 0)
345		vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
346	return (error);
347}
348
349#ifdef UFS_EXTATTR_AUTOSTART
350/*
351 * Given a locked directory vnode, iterate over the names in the directory
352 * and use ufs_extattr_lookup() to retrieve locked vnodes of potential
353 * attribute files.  Then invoke ufs_extattr_enable_with_open() on each
354 * to attempt to start the attribute.  Leaves the directory locked on
355 * exit.
356 */
357static int
358ufs_extattr_iterate_directory(struct ufsmount *ump, struct vnode *dvp,
359    int attrnamespace, struct thread *td)
360{
361	struct vop_readdir_args vargs;
362	struct dirent *dp, *edp;
363	struct vnode *attr_vp;
364	struct uio auio;
365	struct iovec aiov;
366	char *dirbuf;
367	int error, eofflag = 0;
368
369	if (dvp->v_type != VDIR)
370		return (ENOTDIR);
371
372	dirbuf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK);
373
374	auio.uio_iov = &aiov;
375	auio.uio_iovcnt = 1;
376	auio.uio_rw = UIO_READ;
377	auio.uio_segflg = UIO_SYSSPACE;
378	auio.uio_td = td;
379	auio.uio_offset = 0;
380
381	vargs.a_gen.a_desc = NULL;
382	vargs.a_vp = dvp;
383	vargs.a_uio = &auio;
384	vargs.a_cred = td->td_ucred;
385	vargs.a_eofflag = &eofflag;
386	vargs.a_ncookies = NULL;
387	vargs.a_cookies = NULL;
388
389	while (!eofflag) {
390		auio.uio_resid = DIRBLKSIZ;
391		aiov.iov_base = dirbuf;
392		aiov.iov_len = DIRBLKSIZ;
393		error = ufs_readdir(&vargs);
394		if (error) {
395			printf("ufs_extattr_iterate_directory: ufs_readdir "
396			    "%d\n", error);
397			return (error);
398		}
399
400		/*
401		 * XXXRW: While in UFS, we always get DIRBLKSIZ returns from
402		 * the directory code on success, on other file systems this
403		 * may not be the case.  For portability, we should check the
404		 * read length on return from ufs_readdir().
405		 */
406		edp = (struct dirent *)&dirbuf[DIRBLKSIZ];
407		for (dp = (struct dirent *)dirbuf; dp < edp; ) {
408#if (BYTE_ORDER == LITTLE_ENDIAN)
409			dp->d_type = dp->d_namlen;
410			dp->d_namlen = 0;
411#else
412			dp->d_type = 0;
413#endif
414			if (dp->d_reclen == 0)
415				break;
416			error = ufs_extattr_lookup(dvp, UE_GETDIR_LOCKPARENT,
417			    dp->d_name, &attr_vp, td);
418			if (error) {
419				printf("ufs_extattr_iterate_directory: lookup "
420				    "%s %d\n", dp->d_name, error);
421			} else if (attr_vp == dvp) {
422				vrele(attr_vp);
423			} else if (attr_vp->v_type != VREG) {
424				vput(attr_vp);
425			} else {
426				error = ufs_extattr_enable_with_open(ump,
427				    attr_vp, attrnamespace, dp->d_name, td);
428				vrele(attr_vp);
429				if (error) {
430					printf("ufs_extattr_iterate_directory: "
431					    "enable %s %d\n", dp->d_name,
432					    error);
433				} else if (bootverbose) {
434					printf("UFS autostarted EA %s\n",
435					    dp->d_name);
436				}
437			}
438			dp = (struct dirent *) ((char *)dp + dp->d_reclen);
439			if (dp >= edp)
440				break;
441		}
442	}
443	free(dirbuf, M_TEMP);
444
445	return (0);
446}
447
448/*
449 * Auto-start of extended attributes, to be executed (optionally) at
450 * mount-time.
451 */
452int
453ufs_extattr_autostart(struct mount *mp, struct thread *td)
454{
455	struct ufsmount *ump;
456	int error;
457
458	ump = VFSTOUFS(mp);
459	ufs_extattr_uepm_lock(ump, td);
460	error = ufs_extattr_autostart_locked(mp, td);
461	ufs_extattr_uepm_unlock(ump, td);
462	return (error);
463}
464
465static int
466ufs_extattr_autostart_locked(struct mount *mp, struct thread *td)
467{
468	struct vnode *rvp, *attr_dvp, *attr_system_dvp, *attr_user_dvp;
469	struct ufsmount *ump = VFSTOUFS(mp);
470	int error;
471
472	/*
473	 * UFS_EXTATTR applies only to UFS1, as UFS2 uses native extended
474	 * attributes, so don't autostart.
475	 */
476	if (ump->um_fstype != UFS1)
477		return (0);
478
479	/*
480	 * Does UFS_EXTATTR_FSROOTSUBDIR exist off the filesystem root?
481	 * If so, automatically start EA's.
482	 */
483	error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp);
484	if (error) {
485		printf("ufs_extattr_autostart.VFS_ROOT() returned %d\n",
486		    error);
487		return (error);
488	}
489
490	error = ufs_extattr_lookup(rvp, UE_GETDIR_LOCKPARENT_DONT,
491	    UFS_EXTATTR_FSROOTSUBDIR, &attr_dvp, td);
492	if (error) {
493		/* rvp ref'd but now unlocked */
494		vrele(rvp);
495		return (error);
496	}
497	if (rvp == attr_dvp) {
498		/* Should never happen. */
499		vput(rvp);
500		vrele(attr_dvp);
501		return (EINVAL);
502	}
503	vrele(rvp);
504
505	if (attr_dvp->v_type != VDIR) {
506		printf("ufs_extattr_autostart: %s != VDIR\n",
507		    UFS_EXTATTR_FSROOTSUBDIR);
508		goto return_vput_attr_dvp;
509	}
510
511	error = ufs_extattr_start_locked(ump, td);
512	if (error) {
513		printf("ufs_extattr_autostart: ufs_extattr_start failed (%d)\n",
514		    error);
515		goto return_vput_attr_dvp;
516	}
517
518	/*
519	 * Look for two subdirectories: UFS_EXTATTR_SUBDIR_SYSTEM,
520	 * UFS_EXTATTR_SUBDIR_USER.  For each, iterate over the sub-directory,
521	 * and start with appropriate type.  Failures in either don't
522	 * result in an over-all failure.  attr_dvp is left locked to
523	 * be cleaned up on exit.
524	 */
525	error = ufs_extattr_lookup(attr_dvp, UE_GETDIR_LOCKPARENT,
526	    UFS_EXTATTR_SUBDIR_SYSTEM, &attr_system_dvp, td);
527	if (!error) {
528		error = ufs_extattr_iterate_directory(VFSTOUFS(mp),
529		    attr_system_dvp, EXTATTR_NAMESPACE_SYSTEM, td);
530		if (error)
531			printf("ufs_extattr_iterate_directory returned %d\n",
532			    error);
533		vput(attr_system_dvp);
534	}
535
536	error = ufs_extattr_lookup(attr_dvp, UE_GETDIR_LOCKPARENT,
537	    UFS_EXTATTR_SUBDIR_USER, &attr_user_dvp, td);
538	if (!error) {
539		error = ufs_extattr_iterate_directory(VFSTOUFS(mp),
540		    attr_user_dvp, EXTATTR_NAMESPACE_USER, td);
541		if (error)
542			printf("ufs_extattr_iterate_directory returned %d\n",
543			    error);
544		vput(attr_user_dvp);
545	}
546
547	/* Mask startup failures in sub-directories. */
548	error = 0;
549
550return_vput_attr_dvp:
551	vput(attr_dvp);
552
553	return (error);
554}
555#endif /* !UFS_EXTATTR_AUTOSTART */
556
557/*
558 * Stop extended attribute support on an FS.
559 */
560int
561ufs_extattr_stop(struct mount *mp, struct thread *td)
562{
563	struct ufs_extattr_list_entry *uele;
564	struct ufsmount *ump = VFSTOUFS(mp);
565	int error = 0;
566
567	ufs_extattr_uepm_lock(ump, td);
568
569	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
570		error = EOPNOTSUPP;
571		goto unlock;
572	}
573
574	while ((uele = LIST_FIRST(&ump->um_extattr.uepm_list)) != NULL) {
575		ufs_extattr_disable(ump, uele->uele_attrnamespace,
576		    uele->uele_attrname, td);
577	}
578
579	ump->um_extattr.uepm_flags &= ~UFS_EXTATTR_UEPM_STARTED;
580
581	crfree(ump->um_extattr.uepm_ucred);
582	ump->um_extattr.uepm_ucred = NULL;
583
584unlock:
585	ufs_extattr_uepm_unlock(ump, td);
586
587	return (error);
588}
589
590/*
591 * Enable a named attribute on the specified filesystem; provide an
592 * unlocked backing vnode to hold the attribute data.
593 */
594static int
595ufs_extattr_enable(struct ufsmount *ump, int attrnamespace,
596    const char *attrname, struct vnode *backing_vnode, struct thread *td)
597{
598	struct ufs_extattr_list_entry *attribute;
599	struct iovec aiov;
600	struct uio auio;
601	int error = 0;
602
603	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
604		return (EINVAL);
605	if (backing_vnode->v_type != VREG)
606		return (EINVAL);
607
608	attribute = malloc(sizeof(struct ufs_extattr_list_entry),
609	    M_UFS_EXTATTR, M_WAITOK);
610	if (attribute == NULL)
611		return (ENOMEM);
612
613	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
614		error = EOPNOTSUPP;
615		goto free_exit;
616	}
617
618	if (ufs_extattr_find_attr(ump, attrnamespace, attrname)) {
619		error = EEXIST;
620		goto free_exit;
621	}
622
623	strncpy(attribute->uele_attrname, attrname,
624	    UFS_EXTATTR_MAXEXTATTRNAME);
625	attribute->uele_attrnamespace = attrnamespace;
626	bzero(&attribute->uele_fileheader,
627	    sizeof(struct ufs_extattr_fileheader));
628
629	attribute->uele_backing_vnode = backing_vnode;
630
631	auio.uio_iov = &aiov;
632	auio.uio_iovcnt = 1;
633	aiov.iov_base = (caddr_t) &attribute->uele_fileheader;
634	aiov.iov_len = sizeof(struct ufs_extattr_fileheader);
635	auio.uio_resid = sizeof(struct ufs_extattr_fileheader);
636	auio.uio_offset = (off_t) 0;
637	auio.uio_segflg = UIO_SYSSPACE;
638	auio.uio_rw = UIO_READ;
639	auio.uio_td = td;
640
641	vn_lock(backing_vnode, LK_SHARED | LK_RETRY);
642	error = VOP_READ(backing_vnode, &auio, IO_NODELOCKED,
643	    ump->um_extattr.uepm_ucred);
644
645	if (error)
646		goto unlock_free_exit;
647
648	if (auio.uio_resid != 0) {
649		printf("ufs_extattr_enable: malformed attribute header\n");
650		error = EINVAL;
651		goto unlock_free_exit;
652	}
653
654	if (attribute->uele_fileheader.uef_magic != UFS_EXTATTR_MAGIC) {
655		printf("ufs_extattr_enable: invalid attribute header magic\n");
656		error = EINVAL;
657		goto unlock_free_exit;
658	}
659
660	if (attribute->uele_fileheader.uef_version != UFS_EXTATTR_VERSION) {
661		printf("ufs_extattr_enable: incorrect attribute header "
662		    "version\n");
663		error = EINVAL;
664		goto unlock_free_exit;
665	}
666
667	ASSERT_VOP_LOCKED(backing_vnode, "ufs_extattr_enable");
668	LIST_INSERT_HEAD(&ump->um_extattr.uepm_list, attribute,
669	    uele_entries);
670
671	VOP_UNLOCK(backing_vnode, 0);
672	return (0);
673
674unlock_free_exit:
675	VOP_UNLOCK(backing_vnode, 0);
676
677free_exit:
678	free(attribute, M_UFS_EXTATTR);
679	return (error);
680}
681
682/*
683 * Disable extended attribute support on an FS.
684 */
685static int
686ufs_extattr_disable(struct ufsmount *ump, int attrnamespace,
687    const char *attrname, struct thread *td)
688{
689	struct ufs_extattr_list_entry *uele;
690	int error = 0;
691
692	if (!ufs_extattr_valid_attrname(attrnamespace, attrname))
693		return (EINVAL);
694
695	uele = ufs_extattr_find_attr(ump, attrnamespace, attrname);
696	if (!uele)
697		return (ENOATTR);
698
699	LIST_REMOVE(uele, uele_entries);
700
701	vn_lock(uele->uele_backing_vnode, LK_SHARED | LK_RETRY);
702	ASSERT_VOP_LOCKED(uele->uele_backing_vnode, "ufs_extattr_disable");
703	VOP_UNLOCK(uele->uele_backing_vnode, 0);
704	error = vn_close(uele->uele_backing_vnode, FREAD|FWRITE,
705	    td->td_ucred, td);
706
707	free(uele, M_UFS_EXTATTR);
708
709	return (error);
710}
711
712/*
713 * VFS call to manage extended attributes in UFS.  If filename_vp is
714 * non-NULL, it must be passed in locked, and regardless of errors in
715 * processing, will be unlocked.
716 */
717int
718ufs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
719    int attrnamespace, const char *attrname)
720{
721	struct ufsmount *ump = VFSTOUFS(mp);
722	struct thread *td = curthread;
723	int error;
724
725	/*
726	 * Processes with privilege, but in jail, are not allowed to
727	 * configure extended attributes.
728	 */
729	error = priv_check(td, PRIV_UFS_EXTATTRCTL);
730	if (error) {
731		if (filename_vp != NULL)
732			VOP_UNLOCK(filename_vp, 0);
733		return (error);
734	}
735
736	/*
737	 * We only allow extattrctl(2) on UFS1 file systems, as UFS2 uses
738	 * native extended attributes.
739	 */
740	if (ump->um_fstype != UFS1) {
741		if (filename_vp != NULL)
742			VOP_UNLOCK(filename_vp, 0);
743		return (EOPNOTSUPP);
744	}
745
746	switch(cmd) {
747	case UFS_EXTATTR_CMD_START:
748		if (filename_vp != NULL) {
749			VOP_UNLOCK(filename_vp, 0);
750			return (EINVAL);
751		}
752		if (attrname != NULL)
753			return (EINVAL);
754
755		error = ufs_extattr_start(mp, td);
756
757		return (error);
758
759	case UFS_EXTATTR_CMD_STOP:
760		if (filename_vp != NULL) {
761			VOP_UNLOCK(filename_vp, 0);
762			return (EINVAL);
763		}
764		if (attrname != NULL)
765			return (EINVAL);
766
767		error = ufs_extattr_stop(mp, td);
768
769		return (error);
770
771	case UFS_EXTATTR_CMD_ENABLE:
772
773		if (filename_vp == NULL)
774			return (EINVAL);
775		if (attrname == NULL) {
776			VOP_UNLOCK(filename_vp, 0);
777			return (EINVAL);
778		}
779
780		/*
781		 * ufs_extattr_enable_with_open() will always unlock the
782		 * vnode, regardless of failure.
783		 */
784		ufs_extattr_uepm_lock(ump, td);
785		error = ufs_extattr_enable_with_open(ump, filename_vp,
786		    attrnamespace, attrname, td);
787		ufs_extattr_uepm_unlock(ump, td);
788
789		return (error);
790
791	case UFS_EXTATTR_CMD_DISABLE:
792
793		if (filename_vp != NULL) {
794			VOP_UNLOCK(filename_vp, 0);
795			return (EINVAL);
796		}
797		if (attrname == NULL)
798			return (EINVAL);
799
800		ufs_extattr_uepm_lock(ump, td);
801		error = ufs_extattr_disable(ump, attrnamespace, attrname,
802		    td);
803		ufs_extattr_uepm_unlock(ump, td);
804
805		return (error);
806
807	default:
808		return (EINVAL);
809	}
810}
811
812/*
813 * Vnode operating to retrieve a named extended attribute.
814 */
815int
816ufs_getextattr(struct vop_getextattr_args *ap)
817/*
818vop_getextattr {
819	IN struct vnode *a_vp;
820	IN int a_attrnamespace;
821	IN const char *a_name;
822	INOUT struct uio *a_uio;
823	OUT size_t *a_size;
824	IN struct ucred *a_cred;
825	IN struct thread *a_td;
826};
827*/
828{
829	struct mount *mp = ap->a_vp->v_mount;
830	struct ufsmount *ump = VFSTOUFS(mp);
831	int error;
832
833	ufs_extattr_uepm_lock(ump, ap->a_td);
834
835	error = ufs_extattr_get(ap->a_vp, ap->a_attrnamespace, ap->a_name,
836	    ap->a_uio, ap->a_size, ap->a_cred, ap->a_td);
837
838	ufs_extattr_uepm_unlock(ump, ap->a_td);
839
840	return (error);
841}
842
843/*
844 * Real work associated with retrieving a named attribute--assumes that
845 * the attribute lock has already been grabbed.
846 */
847static int
848ufs_extattr_get(struct vnode *vp, int attrnamespace, const char *name,
849    struct uio *uio, size_t *size, struct ucred *cred, struct thread *td)
850{
851	struct ufs_extattr_list_entry *attribute;
852	struct ufs_extattr_header ueh;
853	struct iovec local_aiov;
854	struct uio local_aio;
855	struct mount *mp = vp->v_mount;
856	struct ufsmount *ump = VFSTOUFS(mp);
857	struct inode *ip = VTOI(vp);
858	off_t base_offset;
859	size_t len, old_len;
860	int error = 0;
861
862	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
863		return (EOPNOTSUPP);
864
865	if (strlen(name) == 0)
866		return (EINVAL);
867
868	error = extattr_check_cred(vp, attrnamespace, cred, td, VREAD);
869	if (error)
870		return (error);
871
872	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
873	if (!attribute)
874		return (ENOATTR);
875
876	/*
877	 * Allow only offsets of zero to encourage the read/replace
878	 * extended attribute semantic.  Otherwise we can't guarantee
879	 * atomicity, as we don't provide locks for extended attributes.
880	 */
881	if (uio != NULL && uio->uio_offset != 0)
882		return (ENXIO);
883
884	/*
885	 * Find base offset of header in file based on file header size, and
886	 * data header size + maximum data size, indexed by inode number.
887	 */
888	base_offset = sizeof(struct ufs_extattr_fileheader) +
889	    ip->i_number * (sizeof(struct ufs_extattr_header) +
890	    attribute->uele_fileheader.uef_size);
891
892	/*
893	 * Read in the data header to see if the data is defined, and if so
894	 * how much.
895	 */
896	bzero(&ueh, sizeof(struct ufs_extattr_header));
897	local_aiov.iov_base = (caddr_t) &ueh;
898	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
899	local_aio.uio_iov = &local_aiov;
900	local_aio.uio_iovcnt = 1;
901	local_aio.uio_rw = UIO_READ;
902	local_aio.uio_segflg = UIO_SYSSPACE;
903	local_aio.uio_td = td;
904	local_aio.uio_offset = base_offset;
905	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
906
907	/*
908	 * Acquire locks.
909	 *
910	 * Don't need to get a lock on the backing file if the getattr is
911	 * being applied to the backing file, as the lock is already held.
912	 */
913	if (attribute->uele_backing_vnode != vp)
914		vn_lock(attribute->uele_backing_vnode, LK_SHARED | LK_RETRY);
915
916	error = VOP_READ(attribute->uele_backing_vnode, &local_aio,
917	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
918	if (error)
919		goto vopunlock_exit;
920
921	/* Defined? */
922	if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) {
923		error = ENOATTR;
924		goto vopunlock_exit;
925	}
926
927	/* Valid for the current inode generation? */
928	if (ueh.ueh_i_gen != ip->i_gen) {
929		/*
930		 * The inode itself has a different generation number
931		 * than the attribute data.  For now, the best solution
932		 * is to coerce this to undefined, and let it get cleaned
933		 * up by the next write or extattrctl clean.
934		 */
935		printf("ufs_extattr_get (%s): inode number inconsistency (%d, %ju)\n",
936		    mp->mnt_stat.f_mntonname, ueh.ueh_i_gen, (uintmax_t)ip->i_gen);
937		error = ENOATTR;
938		goto vopunlock_exit;
939	}
940
941	/* Local size consistency check. */
942	if (ueh.ueh_len > attribute->uele_fileheader.uef_size) {
943		error = ENXIO;
944		goto vopunlock_exit;
945	}
946
947	/* Return full data size if caller requested it. */
948	if (size != NULL)
949		*size = ueh.ueh_len;
950
951	/* Return data if the caller requested it. */
952	if (uio != NULL) {
953		/* Allow for offset into the attribute data. */
954		uio->uio_offset = base_offset + sizeof(struct
955		    ufs_extattr_header);
956
957		/*
958		 * Figure out maximum to transfer -- use buffer size and
959		 * local data limit.
960		 */
961		len = MIN(uio->uio_resid, ueh.ueh_len);
962		old_len = uio->uio_resid;
963		uio->uio_resid = len;
964
965		error = VOP_READ(attribute->uele_backing_vnode, uio,
966		    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
967		if (error)
968			goto vopunlock_exit;
969
970		uio->uio_resid = old_len - (len - uio->uio_resid);
971	}
972
973vopunlock_exit:
974
975	if (uio != NULL)
976		uio->uio_offset = 0;
977
978	if (attribute->uele_backing_vnode != vp)
979		VOP_UNLOCK(attribute->uele_backing_vnode, 0);
980
981	return (error);
982}
983
984/*
985 * Vnode operation to remove a named attribute.
986 */
987int
988ufs_deleteextattr(struct vop_deleteextattr_args *ap)
989/*
990vop_deleteextattr {
991	IN struct vnode *a_vp;
992	IN int a_attrnamespace;
993	IN const char *a_name;
994	IN struct ucred *a_cred;
995	IN struct thread *a_td;
996};
997*/
998{
999	struct mount *mp = ap->a_vp->v_mount;
1000	struct ufsmount *ump = VFSTOUFS(mp);
1001	int error;
1002
1003	ufs_extattr_uepm_lock(ump, ap->a_td);
1004
1005	error = ufs_extattr_rm(ap->a_vp, ap->a_attrnamespace, ap->a_name,
1006	    ap->a_cred, ap->a_td);
1007
1008
1009	ufs_extattr_uepm_unlock(ump, ap->a_td);
1010
1011	return (error);
1012}
1013
1014/*
1015 * Vnode operation to set a named attribute.
1016 */
1017int
1018ufs_setextattr(struct vop_setextattr_args *ap)
1019/*
1020vop_setextattr {
1021	IN struct vnode *a_vp;
1022	IN int a_attrnamespace;
1023	IN const char *a_name;
1024	INOUT struct uio *a_uio;
1025	IN struct ucred *a_cred;
1026	IN struct thread *a_td;
1027};
1028*/
1029{
1030	struct mount *mp = ap->a_vp->v_mount;
1031	struct ufsmount *ump = VFSTOUFS(mp);
1032	int error;
1033
1034	/*
1035	 * XXX: No longer a supported way to delete extended attributes.
1036	 */
1037	if (ap->a_uio == NULL)
1038		return (EINVAL);
1039
1040	ufs_extattr_uepm_lock(ump, ap->a_td);
1041
1042	error = ufs_extattr_set(ap->a_vp, ap->a_attrnamespace, ap->a_name,
1043	    ap->a_uio, ap->a_cred, ap->a_td);
1044
1045	ufs_extattr_uepm_unlock(ump, ap->a_td);
1046
1047	return (error);
1048}
1049
1050/*
1051 * Real work associated with setting a vnode's extended attributes;
1052 * assumes that the attribute lock has already been grabbed.
1053 */
1054static int
1055ufs_extattr_set(struct vnode *vp, int attrnamespace, const char *name,
1056    struct uio *uio, struct ucred *cred, struct thread *td)
1057{
1058	struct ufs_extattr_list_entry *attribute;
1059	struct ufs_extattr_header ueh;
1060	struct iovec local_aiov;
1061	struct uio local_aio;
1062	struct mount *mp = vp->v_mount;
1063	struct ufsmount *ump = VFSTOUFS(mp);
1064	struct inode *ip = VTOI(vp);
1065	off_t base_offset;
1066	int error = 0, ioflag;
1067
1068	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1069		return (EROFS);
1070	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
1071		return (EOPNOTSUPP);
1072	if (!ufs_extattr_valid_attrname(attrnamespace, name))
1073		return (EINVAL);
1074
1075	error = extattr_check_cred(vp, attrnamespace, cred, td, VWRITE);
1076	if (error)
1077		return (error);
1078
1079	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
1080	if (!attribute)
1081		return (ENOATTR);
1082
1083	/*
1084	 * Early rejection of invalid offsets/length.
1085	 * Reject: any offset but 0 (replace)
1086	 *	 Any size greater than attribute size limit
1087 	 */
1088	if (uio->uio_offset != 0 ||
1089	    uio->uio_resid > attribute->uele_fileheader.uef_size)
1090		return (ENXIO);
1091
1092	/*
1093	 * Find base offset of header in file based on file header size, and
1094	 * data header size + maximum data size, indexed by inode number.
1095	 */
1096	base_offset = sizeof(struct ufs_extattr_fileheader) +
1097	    ip->i_number * (sizeof(struct ufs_extattr_header) +
1098	    attribute->uele_fileheader.uef_size);
1099
1100	/*
1101	 * Write out a data header for the data.
1102	 */
1103	ueh.ueh_len = uio->uio_resid;
1104	ueh.ueh_flags = UFS_EXTATTR_ATTR_FLAG_INUSE;
1105	ueh.ueh_i_gen = ip->i_gen;
1106	local_aiov.iov_base = (caddr_t) &ueh;
1107	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
1108	local_aio.uio_iov = &local_aiov;
1109	local_aio.uio_iovcnt = 1;
1110	local_aio.uio_rw = UIO_WRITE;
1111	local_aio.uio_segflg = UIO_SYSSPACE;
1112	local_aio.uio_td = td;
1113	local_aio.uio_offset = base_offset;
1114	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
1115
1116	/*
1117	 * Acquire locks.
1118	 *
1119	 * Don't need to get a lock on the backing file if the setattr is
1120	 * being applied to the backing file, as the lock is already held.
1121	 */
1122	if (attribute->uele_backing_vnode != vp)
1123		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
1124
1125	ioflag = IO_NODELOCKED;
1126	if (ufs_extattr_sync)
1127		ioflag |= IO_SYNC;
1128	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
1129	    ump->um_extattr.uepm_ucred);
1130	if (error)
1131		goto vopunlock_exit;
1132
1133	if (local_aio.uio_resid != 0) {
1134		error = ENXIO;
1135		goto vopunlock_exit;
1136	}
1137
1138	/*
1139	 * Write out user data.
1140	 */
1141	uio->uio_offset = base_offset + sizeof(struct ufs_extattr_header);
1142
1143	ioflag = IO_NODELOCKED;
1144	if (ufs_extattr_sync)
1145		ioflag |= IO_SYNC;
1146	error = VOP_WRITE(attribute->uele_backing_vnode, uio, ioflag,
1147	    ump->um_extattr.uepm_ucred);
1148
1149vopunlock_exit:
1150	uio->uio_offset = 0;
1151
1152	if (attribute->uele_backing_vnode != vp)
1153		VOP_UNLOCK(attribute->uele_backing_vnode, 0);
1154
1155	return (error);
1156}
1157
1158/*
1159 * Real work associated with removing an extended attribute from a vnode.
1160 * Assumes the attribute lock has already been grabbed.
1161 */
1162static int
1163ufs_extattr_rm(struct vnode *vp, int attrnamespace, const char *name,
1164    struct ucred *cred, struct thread *td)
1165{
1166	struct ufs_extattr_list_entry *attribute;
1167	struct ufs_extattr_header ueh;
1168	struct iovec local_aiov;
1169	struct uio local_aio;
1170	struct mount *mp = vp->v_mount;
1171	struct ufsmount *ump = VFSTOUFS(mp);
1172	struct inode *ip = VTOI(vp);
1173	off_t base_offset;
1174	int error = 0, ioflag;
1175
1176	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1177		return (EROFS);
1178	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED))
1179		return (EOPNOTSUPP);
1180	if (!ufs_extattr_valid_attrname(attrnamespace, name))
1181		return (EINVAL);
1182
1183	error = extattr_check_cred(vp, attrnamespace, cred, td, VWRITE);
1184	if (error)
1185		return (error);
1186
1187	attribute = ufs_extattr_find_attr(ump, attrnamespace, name);
1188	if (!attribute)
1189		return (ENOATTR);
1190
1191	/*
1192	 * Find base offset of header in file based on file header size, and
1193	 * data header size + maximum data size, indexed by inode number.
1194	 */
1195	base_offset = sizeof(struct ufs_extattr_fileheader) +
1196	    ip->i_number * (sizeof(struct ufs_extattr_header) +
1197	    attribute->uele_fileheader.uef_size);
1198
1199	/*
1200	 * Check to see if currently defined.
1201	 */
1202	bzero(&ueh, sizeof(struct ufs_extattr_header));
1203
1204	local_aiov.iov_base = (caddr_t) &ueh;
1205	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
1206	local_aio.uio_iov = &local_aiov;
1207	local_aio.uio_iovcnt = 1;
1208	local_aio.uio_rw = UIO_READ;
1209	local_aio.uio_segflg = UIO_SYSSPACE;
1210	local_aio.uio_td = td;
1211	local_aio.uio_offset = base_offset;
1212	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
1213
1214	/*
1215	 * Don't need to get the lock on the backing vnode if the vnode we're
1216	 * modifying is it, as we already hold the lock.
1217	 */
1218	if (attribute->uele_backing_vnode != vp)
1219		vn_lock(attribute->uele_backing_vnode, LK_EXCLUSIVE | LK_RETRY);
1220
1221	error = VOP_READ(attribute->uele_backing_vnode, &local_aio,
1222	    IO_NODELOCKED, ump->um_extattr.uepm_ucred);
1223	if (error)
1224		goto vopunlock_exit;
1225
1226	/* Defined? */
1227	if ((ueh.ueh_flags & UFS_EXTATTR_ATTR_FLAG_INUSE) == 0) {
1228		error = ENOATTR;
1229		goto vopunlock_exit;
1230	}
1231
1232	/* Valid for the current inode generation? */
1233	if (ueh.ueh_i_gen != ip->i_gen) {
1234		/*
1235		 * The inode itself has a different generation number than
1236		 * the attribute data.  For now, the best solution is to
1237		 * coerce this to undefined, and let it get cleaned up by
1238		 * the next write or extattrctl clean.
1239		 */
1240		printf("ufs_extattr_rm (%s): inode number inconsistency (%d, %jd)\n",
1241		    mp->mnt_stat.f_mntonname, ueh.ueh_i_gen, (intmax_t)ip->i_gen);
1242		error = ENOATTR;
1243		goto vopunlock_exit;
1244	}
1245
1246	/* Flag it as not in use. */
1247	ueh.ueh_flags = 0;
1248	ueh.ueh_len = 0;
1249
1250	local_aiov.iov_base = (caddr_t) &ueh;
1251	local_aiov.iov_len = sizeof(struct ufs_extattr_header);
1252	local_aio.uio_iov = &local_aiov;
1253	local_aio.uio_iovcnt = 1;
1254	local_aio.uio_rw = UIO_WRITE;
1255	local_aio.uio_segflg = UIO_SYSSPACE;
1256	local_aio.uio_td = td;
1257	local_aio.uio_offset = base_offset;
1258	local_aio.uio_resid = sizeof(struct ufs_extattr_header);
1259
1260	ioflag = IO_NODELOCKED;
1261	if (ufs_extattr_sync)
1262		ioflag |= IO_SYNC;
1263	error = VOP_WRITE(attribute->uele_backing_vnode, &local_aio, ioflag,
1264	    ump->um_extattr.uepm_ucred);
1265	if (error)
1266		goto vopunlock_exit;
1267
1268	if (local_aio.uio_resid != 0)
1269		error = ENXIO;
1270
1271vopunlock_exit:
1272	VOP_UNLOCK(attribute->uele_backing_vnode, 0);
1273
1274	return (error);
1275}
1276
1277/*
1278 * Called by UFS when an inode is no longer active and should have its
1279 * attributes stripped.
1280 */
1281void
1282ufs_extattr_vnode_inactive(struct vnode *vp, struct thread *td)
1283{
1284	struct ufs_extattr_list_entry *uele;
1285	struct mount *mp = vp->v_mount;
1286	struct ufsmount *ump = VFSTOUFS(mp);
1287
1288	/*
1289	 * In that case, we cannot lock. We should not have any active vnodes
1290	 * on the fs if this is not yet initialized but is going to be, so
1291	 * this can go unlocked.
1292	 */
1293	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_INITIALIZED))
1294		return;
1295
1296	ufs_extattr_uepm_lock(ump, td);
1297
1298	if (!(ump->um_extattr.uepm_flags & UFS_EXTATTR_UEPM_STARTED)) {
1299		ufs_extattr_uepm_unlock(ump, td);
1300		return;
1301	}
1302
1303	LIST_FOREACH(uele, &ump->um_extattr.uepm_list, uele_entries)
1304		ufs_extattr_rm(vp, uele->uele_attrnamespace,
1305		    uele->uele_attrname, NULL, td);
1306
1307	ufs_extattr_uepm_unlock(ump, td);
1308}
1309
1310#endif /* !UFS_EXTATTR */
1311