xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision d06955f9bdb1416d9196043ed781f9b36dae9adc)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * SPDX-License-Identifier: BSD-3-Clause
9  *
10  * Copyright (c) 1982, 1986, 1989, 1993
11  *	The Regents of the University of California.  All rights reserved.
12  * (c) UNIX System Laboratories, Inc.
13  * All or some portions of this file are derived from material licensed
14  * to the University of California by American Telephone and Telegraph
15  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16  * the permission of UNIX System Laboratories, Inc.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
43  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
44  * $FreeBSD$
45  */
46 
47 #include "opt_suiddir.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/stat.h>
55 #include <sys/bio.h>
56 #include <sys/buf.h>
57 #include <sys/endian.h>
58 #include <sys/priv.h>
59 #include <sys/rwlock.h>
60 #include <sys/mount.h>
61 #include <sys/unistd.h>
62 #include <sys/time.h>
63 #include <sys/vnode.h>
64 #include <sys/namei.h>
65 #include <sys/lockf.h>
66 #include <sys/event.h>
67 #include <sys/conf.h>
68 #include <sys/file.h>
69 #include <sys/extattr.h>
70 #include <sys/vmmeter.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vnode_pager.h>
79 
80 #include "opt_directio.h"
81 
82 #include <ufs/ufs/dir.h>
83 
84 #include <fs/ext2fs/fs.h>
85 #include <fs/ext2fs/inode.h>
86 #include <fs/ext2fs/ext2_acl.h>
87 #include <fs/ext2fs/ext2_extern.h>
88 #include <fs/ext2fs/ext2fs.h>
89 #include <fs/ext2fs/ext2_dinode.h>
90 #include <fs/ext2fs/ext2_dir.h>
91 #include <fs/ext2fs/ext2_mount.h>
92 #include <fs/ext2fs/ext2_extattr.h>
93 
94 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
95 static void ext2_itimes_locked(struct vnode *);
96 
97 static vop_access_t	ext2_access;
98 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
99 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
100     struct thread *);
101 static vop_close_t	ext2_close;
102 static vop_create_t	ext2_create;
103 static vop_fsync_t	ext2_fsync;
104 static vop_getattr_t	ext2_getattr;
105 static vop_ioctl_t	ext2_ioctl;
106 static vop_link_t	ext2_link;
107 static vop_mkdir_t	ext2_mkdir;
108 static vop_mknod_t	ext2_mknod;
109 static vop_open_t	ext2_open;
110 static vop_pathconf_t	ext2_pathconf;
111 static vop_print_t	ext2_print;
112 static vop_read_t	ext2_read;
113 static vop_readlink_t	ext2_readlink;
114 static vop_remove_t	ext2_remove;
115 static vop_rename_t	ext2_rename;
116 static vop_rmdir_t	ext2_rmdir;
117 static vop_setattr_t	ext2_setattr;
118 static vop_strategy_t	ext2_strategy;
119 static vop_symlink_t	ext2_symlink;
120 static vop_write_t	ext2_write;
121 static vop_deleteextattr_t	ext2_deleteextattr;
122 static vop_getextattr_t	ext2_getextattr;
123 static vop_listextattr_t	ext2_listextattr;
124 static vop_setextattr_t	ext2_setextattr;
125 static vop_vptofh_t	ext2_vptofh;
126 static vop_close_t	ext2fifo_close;
127 static vop_kqfilter_t	ext2fifo_kqfilter;
128 
129 /* Global vfs data structures for ext2. */
130 struct vop_vector ext2_vnodeops = {
131 	.vop_default =		&default_vnodeops,
132 	.vop_access =		ext2_access,
133 	.vop_bmap =		ext2_bmap,
134 	.vop_cachedlookup =	ext2_lookup,
135 	.vop_close =		ext2_close,
136 	.vop_create =		ext2_create,
137 	.vop_fsync =		ext2_fsync,
138 	.vop_getpages =		vnode_pager_local_getpages,
139 	.vop_getpages_async =	vnode_pager_local_getpages_async,
140 	.vop_getattr =		ext2_getattr,
141 	.vop_inactive =		ext2_inactive,
142 	.vop_ioctl =		ext2_ioctl,
143 	.vop_link =		ext2_link,
144 	.vop_lookup =		vfs_cache_lookup,
145 	.vop_mkdir =		ext2_mkdir,
146 	.vop_mknod =		ext2_mknod,
147 	.vop_open =		ext2_open,
148 	.vop_pathconf =		ext2_pathconf,
149 	.vop_poll =		vop_stdpoll,
150 	.vop_print =		ext2_print,
151 	.vop_read =		ext2_read,
152 	.vop_readdir =		ext2_readdir,
153 	.vop_readlink =		ext2_readlink,
154 	.vop_reallocblks =	ext2_reallocblks,
155 	.vop_reclaim =		ext2_reclaim,
156 	.vop_remove =		ext2_remove,
157 	.vop_rename =		ext2_rename,
158 	.vop_rmdir =		ext2_rmdir,
159 	.vop_setattr =		ext2_setattr,
160 	.vop_strategy =		ext2_strategy,
161 	.vop_symlink =		ext2_symlink,
162 	.vop_write =		ext2_write,
163 	.vop_deleteextattr =	ext2_deleteextattr,
164 	.vop_getextattr =	ext2_getextattr,
165 	.vop_listextattr =	ext2_listextattr,
166 	.vop_setextattr =	ext2_setextattr,
167 #ifdef UFS_ACL
168 	.vop_getacl =		ext2_getacl,
169 	.vop_setacl =		ext2_setacl,
170 	.vop_aclcheck =		ext2_aclcheck,
171 #endif /* UFS_ACL */
172 	.vop_vptofh =		ext2_vptofh,
173 };
174 
175 struct vop_vector ext2_fifoops = {
176 	.vop_default =		&fifo_specops,
177 	.vop_access =		ext2_access,
178 	.vop_close =		ext2fifo_close,
179 	.vop_fsync =		ext2_fsync,
180 	.vop_getattr =		ext2_getattr,
181 	.vop_inactive =		ext2_inactive,
182 	.vop_kqfilter =		ext2fifo_kqfilter,
183 	.vop_pathconf =		ext2_pathconf,
184 	.vop_print =		ext2_print,
185 	.vop_read =		VOP_PANIC,
186 	.vop_reclaim =		ext2_reclaim,
187 	.vop_setattr =		ext2_setattr,
188 	.vop_write =		VOP_PANIC,
189 	.vop_vptofh =		ext2_vptofh,
190 };
191 
192 /*
193  * A virgin directory (no blushing please).
194  * Note that the type and namlen fields are reversed relative to ext2.
195  * Also, we don't use `struct odirtemplate', since it would just cause
196  * endianness problems.
197  */
198 static struct dirtemplate mastertemplate = {
199 	0, 12, 1, EXT2_FT_DIR, ".",
200 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
201 };
202 static struct dirtemplate omastertemplate = {
203 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
204 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
205 };
206 
207 static void
208 ext2_itimes_locked(struct vnode *vp)
209 {
210 	struct inode *ip;
211 	struct timespec ts;
212 
213 	ASSERT_VI_LOCKED(vp, __func__);
214 
215 	ip = VTOI(vp);
216 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
217 		return;
218 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
219 		ip->i_flag |= IN_LAZYMOD;
220 	else
221 		ip->i_flag |= IN_MODIFIED;
222 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
223 		vfs_timestamp(&ts);
224 		if (ip->i_flag & IN_ACCESS) {
225 			ip->i_atime = ts.tv_sec;
226 			ip->i_atimensec = ts.tv_nsec;
227 		}
228 		if (ip->i_flag & IN_UPDATE) {
229 			ip->i_mtime = ts.tv_sec;
230 			ip->i_mtimensec = ts.tv_nsec;
231 			ip->i_modrev++;
232 		}
233 		if (ip->i_flag & IN_CHANGE) {
234 			ip->i_ctime = ts.tv_sec;
235 			ip->i_ctimensec = ts.tv_nsec;
236 		}
237 	}
238 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
239 }
240 
241 void
242 ext2_itimes(struct vnode *vp)
243 {
244 
245 	VI_LOCK(vp);
246 	ext2_itimes_locked(vp);
247 	VI_UNLOCK(vp);
248 }
249 
250 /*
251  * Create a regular file
252  */
253 static int
254 ext2_create(struct vop_create_args *ap)
255 {
256 	int error;
257 
258 	error =
259 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
260 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
261 	if (error != 0)
262 		return (error);
263 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
264 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
265 	return (0);
266 }
267 
268 static int
269 ext2_open(struct vop_open_args *ap)
270 {
271 
272 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
273 		return (EOPNOTSUPP);
274 
275 	/*
276 	 * Files marked append-only must be opened for appending.
277 	 */
278 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
279 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
280 		return (EPERM);
281 
282 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
283 
284 	return (0);
285 }
286 
287 /*
288  * Close called.
289  *
290  * Update the times on the inode.
291  */
292 static int
293 ext2_close(struct vop_close_args *ap)
294 {
295 	struct vnode *vp = ap->a_vp;
296 
297 	VI_LOCK(vp);
298 	if (vp->v_usecount > 1)
299 		ext2_itimes_locked(vp);
300 	VI_UNLOCK(vp);
301 	return (0);
302 }
303 
304 static int
305 ext2_access(struct vop_access_args *ap)
306 {
307 	struct vnode *vp = ap->a_vp;
308 	struct inode *ip = VTOI(vp);
309 	accmode_t accmode = ap->a_accmode;
310 	int error;
311 
312 	if (vp->v_type == VBLK || vp->v_type == VCHR)
313 		return (EOPNOTSUPP);
314 
315 	/*
316 	 * Disallow write attempts on read-only file systems;
317 	 * unless the file is a socket, fifo, or a block or
318 	 * character device resident on the file system.
319 	 */
320 	if (accmode & VWRITE) {
321 		switch (vp->v_type) {
322 		case VDIR:
323 		case VLNK:
324 		case VREG:
325 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
326 				return (EROFS);
327 			break;
328 		default:
329 			break;
330 		}
331 	}
332 
333 	/* If immutable bit set, nobody gets to write it. */
334 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
335 		return (EPERM);
336 
337 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
338 	    ap->a_accmode, ap->a_cred, NULL);
339 	return (error);
340 }
341 
342 static int
343 ext2_getattr(struct vop_getattr_args *ap)
344 {
345 	struct vnode *vp = ap->a_vp;
346 	struct inode *ip = VTOI(vp);
347 	struct vattr *vap = ap->a_vap;
348 
349 	ext2_itimes(vp);
350 	/*
351 	 * Copy from inode table
352 	 */
353 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
354 	vap->va_fileid = ip->i_number;
355 	vap->va_mode = ip->i_mode & ~IFMT;
356 	vap->va_nlink = ip->i_nlink;
357 	vap->va_uid = ip->i_uid;
358 	vap->va_gid = ip->i_gid;
359 	vap->va_rdev = ip->i_rdev;
360 	vap->va_size = ip->i_size;
361 	vap->va_atime.tv_sec = ip->i_atime;
362 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
363 	vap->va_mtime.tv_sec = ip->i_mtime;
364 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
365 	vap->va_ctime.tv_sec = ip->i_ctime;
366 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
367 	if E2DI_HAS_XTIME(ip) {
368 		vap->va_birthtime.tv_sec = ip->i_birthtime;
369 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
370 	}
371 	vap->va_flags = ip->i_flags;
372 	vap->va_gen = ip->i_gen;
373 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
374 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
375 	vap->va_type = IFTOVT(ip->i_mode);
376 	vap->va_filerev = ip->i_modrev;
377 	return (0);
378 }
379 
380 /*
381  * Set attribute vnode op. called from several syscalls
382  */
383 static int
384 ext2_setattr(struct vop_setattr_args *ap)
385 {
386 	struct vattr *vap = ap->a_vap;
387 	struct vnode *vp = ap->a_vp;
388 	struct inode *ip = VTOI(vp);
389 	struct ucred *cred = ap->a_cred;
390 	struct thread *td = curthread;
391 	int error;
392 
393 	/*
394 	 * Check for unsettable attributes.
395 	 */
396 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
397 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
398 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
399 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
400 		return (EINVAL);
401 	}
402 	if (vap->va_flags != VNOVAL) {
403 		/* Disallow flags not supported by ext2fs. */
404 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
405 			return (EOPNOTSUPP);
406 
407 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
408 			return (EROFS);
409 		/*
410 		 * Callers may only modify the file flags on objects they
411 		 * have VADMIN rights for.
412 		 */
413 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
414 			return (error);
415 		/*
416 		 * Unprivileged processes and privileged processes in
417 		 * jail() are not permitted to unset system flags, or
418 		 * modify flags if any system flags are set.
419 		 * Privileged non-jail processes may not modify system flags
420 		 * if securelevel > 0 and any existing system flags are set.
421 		 */
422 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
423 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
424 				error = securelevel_gt(cred, 0);
425 				if (error)
426 					return (error);
427 			}
428 		} else {
429 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
430 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
431 				return (EPERM);
432 		}
433 		ip->i_flags = vap->va_flags;
434 		ip->i_flag |= IN_CHANGE;
435 		if (ip->i_flags & (IMMUTABLE | APPEND))
436 			return (0);
437 	}
438 	if (ip->i_flags & (IMMUTABLE | APPEND))
439 		return (EPERM);
440 	/*
441 	 * Go through the fields and update iff not VNOVAL.
442 	 */
443 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
444 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
445 			return (EROFS);
446 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
447 		    td)) != 0)
448 			return (error);
449 	}
450 	if (vap->va_size != VNOVAL) {
451 		/*
452 		 * Disallow write attempts on read-only file systems;
453 		 * unless the file is a socket, fifo, or a block or
454 		 * character device resident on the file system.
455 		 */
456 		switch (vp->v_type) {
457 		case VDIR:
458 			return (EISDIR);
459 		case VLNK:
460 		case VREG:
461 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
462 				return (EROFS);
463 			break;
464 		default:
465 			break;
466 		}
467 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
468 			return (error);
469 	}
470 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
471 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
472 			return (EROFS);
473 		/*
474 		 * From utimes(2):
475 		 * If times is NULL, ... The caller must be the owner of
476 		 * the file, have permission to write the file, or be the
477 		 * super-user.
478 		 * If times is non-NULL, ... The caller must be the owner of
479 		 * the file or be the super-user.
480 		 */
481 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
482 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
483 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
484 			return (error);
485 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
486 		if (vap->va_atime.tv_sec != VNOVAL) {
487 			ip->i_flag &= ~IN_ACCESS;
488 			ip->i_atime = vap->va_atime.tv_sec;
489 			ip->i_atimensec = vap->va_atime.tv_nsec;
490 		}
491 		if (vap->va_mtime.tv_sec != VNOVAL) {
492 			ip->i_flag &= ~IN_UPDATE;
493 			ip->i_mtime = vap->va_mtime.tv_sec;
494 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
495 		}
496 		ip->i_birthtime = vap->va_birthtime.tv_sec;
497 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
498 		error = ext2_update(vp, 0);
499 		if (error)
500 			return (error);
501 	}
502 	error = 0;
503 	if (vap->va_mode != (mode_t)VNOVAL) {
504 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
505 			return (EROFS);
506 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
507 	}
508 	return (error);
509 }
510 
511 /*
512  * Change the mode on a file.
513  * Inode must be locked before calling.
514  */
515 static int
516 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
517 {
518 	struct inode *ip = VTOI(vp);
519 	int error;
520 
521 	/*
522 	 * To modify the permissions on a file, must possess VADMIN
523 	 * for that file.
524 	 */
525 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
526 		return (error);
527 	/*
528 	 * Privileged processes may set the sticky bit on non-directories,
529 	 * as well as set the setgid bit on a file with a group that the
530 	 * process is not a member of.
531 	 */
532 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
533 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
534 		if (error)
535 			return (EFTYPE);
536 	}
537 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
538 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
539 		if (error)
540 			return (error);
541 	}
542 	ip->i_mode &= ~ALLPERMS;
543 	ip->i_mode |= (mode & ALLPERMS);
544 	ip->i_flag |= IN_CHANGE;
545 	return (0);
546 }
547 
548 /*
549  * Perform chown operation on inode ip;
550  * inode must be locked prior to call.
551  */
552 static int
553 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
554     struct thread *td)
555 {
556 	struct inode *ip = VTOI(vp);
557 	uid_t ouid;
558 	gid_t ogid;
559 	int error = 0;
560 
561 	if (uid == (uid_t)VNOVAL)
562 		uid = ip->i_uid;
563 	if (gid == (gid_t)VNOVAL)
564 		gid = ip->i_gid;
565 	/*
566 	 * To modify the ownership of a file, must possess VADMIN
567 	 * for that file.
568 	 */
569 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
570 		return (error);
571 	/*
572 	 * To change the owner of a file, or change the group of a file
573 	 * to a group of which we are not a member, the caller must
574 	 * have privilege.
575 	 */
576 	if (uid != ip->i_uid || (gid != ip->i_gid &&
577 	    !groupmember(gid, cred))) {
578 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
579 		if (error)
580 			return (error);
581 	}
582 	ogid = ip->i_gid;
583 	ouid = ip->i_uid;
584 	ip->i_gid = gid;
585 	ip->i_uid = uid;
586 	ip->i_flag |= IN_CHANGE;
587 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
588 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
589 			ip->i_mode &= ~(ISUID | ISGID);
590 	}
591 	return (0);
592 }
593 
594 /*
595  * Synch an open file.
596  */
597 /* ARGSUSED */
598 static int
599 ext2_fsync(struct vop_fsync_args *ap)
600 {
601 	/*
602 	 * Flush all dirty buffers associated with a vnode.
603 	 */
604 
605 	vop_stdfsync(ap);
606 
607 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
608 }
609 
610 /*
611  * Mknod vnode call
612  */
613 /* ARGSUSED */
614 static int
615 ext2_mknod(struct vop_mknod_args *ap)
616 {
617 	struct vattr *vap = ap->a_vap;
618 	struct vnode **vpp = ap->a_vpp;
619 	struct inode *ip;
620 	ino_t ino;
621 	int error;
622 
623 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
624 	    ap->a_dvp, vpp, ap->a_cnp);
625 	if (error)
626 		return (error);
627 	ip = VTOI(*vpp);
628 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
629 	if (vap->va_rdev != VNOVAL) {
630 		/*
631 		 * Want to be able to use this to make badblock
632 		 * inodes, so don't truncate the dev number.
633 		 */
634 		if (!(ip->i_flag & IN_E4EXTENTS))
635 			ip->i_rdev = vap->va_rdev;
636 	}
637 	/*
638 	 * Remove inode, then reload it through VFS_VGET so it is
639 	 * checked to see if it is an alias of an existing entry in
640 	 * the inode cache.	 XXX I don't believe this is necessary now.
641 	 */
642 	(*vpp)->v_type = VNON;
643 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
644 	vgone(*vpp);
645 	vput(*vpp);
646 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
647 	if (error) {
648 		*vpp = NULL;
649 		return (error);
650 	}
651 	return (0);
652 }
653 
654 static int
655 ext2_remove(struct vop_remove_args *ap)
656 {
657 	struct inode *ip;
658 	struct vnode *vp = ap->a_vp;
659 	struct vnode *dvp = ap->a_dvp;
660 	int error;
661 
662 	ip = VTOI(vp);
663 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
664 	    (VTOI(dvp)->i_flags & APPEND)) {
665 		error = EPERM;
666 		goto out;
667 	}
668 	error = ext2_dirremove(dvp, ap->a_cnp);
669 	if (error == 0) {
670 		ip->i_nlink--;
671 		ip->i_flag |= IN_CHANGE;
672 	}
673 out:
674 	return (error);
675 }
676 
677 static unsigned short
678 ext2_max_nlink(struct inode *ip)
679 {
680 	struct m_ext2fs *fs;
681 
682 	fs = ip->i_e2fs;
683 
684 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_DIR_NLINK))
685 		return (EXT4_LINK_MAX);
686 	else
687 		return (EXT2_LINK_MAX);
688 }
689 
690 /*
691  * link vnode call
692  */
693 static int
694 ext2_link(struct vop_link_args *ap)
695 {
696 	struct vnode *vp = ap->a_vp;
697 	struct vnode *tdvp = ap->a_tdvp;
698 	struct componentname *cnp = ap->a_cnp;
699 	struct inode *ip;
700 	int error;
701 
702 #ifdef INVARIANTS
703 	if ((cnp->cn_flags & HASBUF) == 0)
704 		panic("ext2_link: no name");
705 #endif
706 	ip = VTOI(vp);
707 	if ((nlink_t)ip->i_nlink >= ext2_max_nlink(ip)) {
708 		error = EMLINK;
709 		goto out;
710 	}
711 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
712 		error = EPERM;
713 		goto out;
714 	}
715 	ip->i_nlink++;
716 	ip->i_flag |= IN_CHANGE;
717 	error = ext2_update(vp, !DOINGASYNC(vp));
718 	if (!error)
719 		error = ext2_direnter(ip, tdvp, cnp);
720 	if (error) {
721 		ip->i_nlink--;
722 		ip->i_flag |= IN_CHANGE;
723 	}
724 out:
725 	return (error);
726 }
727 
728 static int
729 ext2_inc_nlink(struct inode *ip)
730 {
731 
732 	ip->i_nlink++;
733 
734 	if (ext2_htree_has_idx(ip) && ip->i_nlink > 1) {
735 		if (ip->i_nlink >= ext2_max_nlink(ip) || ip->i_nlink == 2)
736 			ip->i_nlink = 1;
737 	} else if (ip->i_nlink > ext2_max_nlink(ip)) {
738 		ip->i_nlink--;
739 		return (EMLINK);
740 	}
741 
742 	return (0);
743 }
744 
745 static void
746 ext2_dec_nlink(struct inode *ip)
747 {
748 
749 	if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
750 		ip->i_nlink--;
751 }
752 
753 /*
754  * Rename system call.
755  * 	rename("foo", "bar");
756  * is essentially
757  *	unlink("bar");
758  *	link("foo", "bar");
759  *	unlink("foo");
760  * but ``atomically''.  Can't do full commit without saving state in the
761  * inode on disk which isn't feasible at this time.  Best we can do is
762  * always guarantee the target exists.
763  *
764  * Basic algorithm is:
765  *
766  * 1) Bump link count on source while we're linking it to the
767  *    target.  This also ensure the inode won't be deleted out
768  *    from underneath us while we work (it may be truncated by
769  *    a concurrent `trunc' or `open' for creation).
770  * 2) Link source to destination.  If destination already exists,
771  *    delete it first.
772  * 3) Unlink source reference to inode if still around. If a
773  *    directory was moved and the parent of the destination
774  *    is different from the source, patch the ".." entry in the
775  *    directory.
776  */
777 static int
778 ext2_rename(struct vop_rename_args *ap)
779 {
780 	struct vnode *tvp = ap->a_tvp;
781 	struct vnode *tdvp = ap->a_tdvp;
782 	struct vnode *fvp = ap->a_fvp;
783 	struct vnode *fdvp = ap->a_fdvp;
784 	struct componentname *tcnp = ap->a_tcnp;
785 	struct componentname *fcnp = ap->a_fcnp;
786 	struct inode *ip, *xp, *dp;
787 	struct dirtemplate dirbuf;
788 	int doingdirectory = 0, oldparent = 0, newparent = 0;
789 	int error = 0;
790 	u_char namlen;
791 
792 #ifdef INVARIANTS
793 	if ((tcnp->cn_flags & HASBUF) == 0 ||
794 	    (fcnp->cn_flags & HASBUF) == 0)
795 		panic("ext2_rename: no name");
796 #endif
797 	/*
798 	 * Check for cross-device rename.
799 	 */
800 	if ((fvp->v_mount != tdvp->v_mount) ||
801 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
802 		error = EXDEV;
803 abortit:
804 		if (tdvp == tvp)
805 			vrele(tdvp);
806 		else
807 			vput(tdvp);
808 		if (tvp)
809 			vput(tvp);
810 		vrele(fdvp);
811 		vrele(fvp);
812 		return (error);
813 	}
814 
815 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
816 	    (VTOI(tdvp)->i_flags & APPEND))) {
817 		error = EPERM;
818 		goto abortit;
819 	}
820 
821 	/*
822 	 * Renaming a file to itself has no effect.  The upper layers should
823 	 * not call us in that case.  Temporarily just warn if they do.
824 	 */
825 	if (fvp == tvp) {
826 		printf("ext2_rename: fvp == tvp (can't happen)\n");
827 		error = 0;
828 		goto abortit;
829 	}
830 
831 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
832 		goto abortit;
833 	dp = VTOI(fdvp);
834 	ip = VTOI(fvp);
835 	if (ip->i_nlink >= ext2_max_nlink(ip) && !ext2_htree_has_idx(ip)) {
836 		VOP_UNLOCK(fvp, 0);
837 		error = EMLINK;
838 		goto abortit;
839 	}
840 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
841 	    || (dp->i_flags & APPEND)) {
842 		VOP_UNLOCK(fvp, 0);
843 		error = EPERM;
844 		goto abortit;
845 	}
846 	if ((ip->i_mode & IFMT) == IFDIR) {
847 		/*
848 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
849 		 */
850 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
851 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
852 		    (ip->i_flag & IN_RENAME)) {
853 			VOP_UNLOCK(fvp, 0);
854 			error = EINVAL;
855 			goto abortit;
856 		}
857 		ip->i_flag |= IN_RENAME;
858 		oldparent = dp->i_number;
859 		doingdirectory++;
860 	}
861 	vrele(fdvp);
862 
863 	/*
864 	 * When the target exists, both the directory
865 	 * and target vnodes are returned locked.
866 	 */
867 	dp = VTOI(tdvp);
868 	xp = NULL;
869 	if (tvp)
870 		xp = VTOI(tvp);
871 
872 	/*
873 	 * 1) Bump link count while we're moving stuff
874 	 *    around.  If we crash somewhere before
875 	 *    completing our work, the link count
876 	 *    may be wrong, but correctable.
877 	 */
878 	ext2_inc_nlink(ip);
879 	ip->i_flag |= IN_CHANGE;
880 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
881 		VOP_UNLOCK(fvp, 0);
882 		goto bad;
883 	}
884 
885 	/*
886 	 * If ".." must be changed (ie the directory gets a new
887 	 * parent) then the source directory must not be in the
888 	 * directory hierarchy above the target, as this would
889 	 * orphan everything below the source directory. Also
890 	 * the user must have write permission in the source so
891 	 * as to be able to change "..". We must repeat the call
892 	 * to namei, as the parent directory is unlocked by the
893 	 * call to checkpath().
894 	 */
895 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
896 	VOP_UNLOCK(fvp, 0);
897 	if (oldparent != dp->i_number)
898 		newparent = dp->i_number;
899 	if (doingdirectory && newparent) {
900 		if (error)	/* write access check above */
901 			goto bad;
902 		if (xp != NULL)
903 			vput(tvp);
904 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
905 		if (error)
906 			goto out;
907 		VREF(tdvp);
908 		error = relookup(tdvp, &tvp, tcnp);
909 		if (error)
910 			goto out;
911 		vrele(tdvp);
912 		dp = VTOI(tdvp);
913 		xp = NULL;
914 		if (tvp)
915 			xp = VTOI(tvp);
916 	}
917 	/*
918 	 * 2) If target doesn't exist, link the target
919 	 *    to the source and unlink the source.
920 	 *    Otherwise, rewrite the target directory
921 	 *    entry to reference the source inode and
922 	 *    expunge the original entry's existence.
923 	 */
924 	if (xp == NULL) {
925 		if (dp->i_devvp != ip->i_devvp)
926 			panic("ext2_rename: EXDEV");
927 		/*
928 		 * Account for ".." in new directory.
929 		 * When source and destination have the same
930 		 * parent we don't fool with the link count.
931 		 */
932 		if (doingdirectory && newparent) {
933 			error = ext2_inc_nlink(dp);
934 			if (error)
935 				goto bad;
936 
937 			dp->i_flag |= IN_CHANGE;
938 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
939 			if (error)
940 				goto bad;
941 		}
942 		error = ext2_direnter(ip, tdvp, tcnp);
943 		if (error) {
944 			if (doingdirectory && newparent) {
945 				ext2_dec_nlink(dp);
946 				dp->i_flag |= IN_CHANGE;
947 				(void)ext2_update(tdvp, 1);
948 			}
949 			goto bad;
950 		}
951 		vput(tdvp);
952 	} else {
953 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
954 			panic("ext2_rename: EXDEV");
955 		/*
956 		 * Short circuit rename(foo, foo).
957 		 */
958 		if (xp->i_number == ip->i_number)
959 			panic("ext2_rename: same file");
960 		/*
961 		 * If the parent directory is "sticky", then the user must
962 		 * own the parent directory, or the destination of the rename,
963 		 * otherwise the destination may not be changed (except by
964 		 * root). This implements append-only directories.
965 		 */
966 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
967 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
968 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
969 			error = EPERM;
970 			goto bad;
971 		}
972 		/*
973 		 * Target must be empty if a directory and have no links
974 		 * to it. Also, ensure source and target are compatible
975 		 * (both directories, or both not directories).
976 		 */
977 		if ((xp->i_mode & IFMT) == IFDIR) {
978 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
979 				error = ENOTEMPTY;
980 				goto bad;
981 			}
982 			if (!doingdirectory) {
983 				error = ENOTDIR;
984 				goto bad;
985 			}
986 			cache_purge(tdvp);
987 		} else if (doingdirectory) {
988 			error = EISDIR;
989 			goto bad;
990 		}
991 		error = ext2_dirrewrite(dp, ip, tcnp);
992 		if (error)
993 			goto bad;
994 		/*
995 		 * If the target directory is in the same
996 		 * directory as the source directory,
997 		 * decrement the link count on the parent
998 		 * of the target directory.
999 		 */
1000 		if (doingdirectory && !newparent) {
1001 			ext2_dec_nlink(dp);
1002 			dp->i_flag |= IN_CHANGE;
1003 		}
1004 		vput(tdvp);
1005 		/*
1006 		 * Adjust the link count of the target to
1007 		 * reflect the dirrewrite above.  If this is
1008 		 * a directory it is empty and there are
1009 		 * no links to it, so we can squash the inode and
1010 		 * any space associated with it.  We disallowed
1011 		 * renaming over top of a directory with links to
1012 		 * it above, as the remaining link would point to
1013 		 * a directory without "." or ".." entries.
1014 		 */
1015 		ext2_dec_nlink(xp);
1016 		if (doingdirectory) {
1017 			if (--xp->i_nlink != 0)
1018 				panic("ext2_rename: linked directory");
1019 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1020 			    tcnp->cn_cred, tcnp->cn_thread);
1021 		}
1022 		xp->i_flag |= IN_CHANGE;
1023 		vput(tvp);
1024 		xp = NULL;
1025 	}
1026 
1027 	/*
1028 	 * 3) Unlink the source.
1029 	 */
1030 	fcnp->cn_flags &= ~MODMASK;
1031 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1032 	VREF(fdvp);
1033 	error = relookup(fdvp, &fvp, fcnp);
1034 	if (error == 0)
1035 		vrele(fdvp);
1036 	if (fvp != NULL) {
1037 		xp = VTOI(fvp);
1038 		dp = VTOI(fdvp);
1039 	} else {
1040 		/*
1041 		 * From name has disappeared.  IN_RENAME is not sufficient
1042 		 * to protect against directory races due to timing windows,
1043 		 * so we can't panic here.
1044 		 */
1045 		vrele(ap->a_fvp);
1046 		return (0);
1047 	}
1048 	/*
1049 	 * Ensure that the directory entry still exists and has not
1050 	 * changed while the new name has been entered. If the source is
1051 	 * a file then the entry may have been unlinked or renamed. In
1052 	 * either case there is no further work to be done. If the source
1053 	 * is a directory then it cannot have been rmdir'ed; its link
1054 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1055 	 * The IN_RENAME flag ensures that it cannot be moved by another
1056 	 * rename.
1057 	 */
1058 	if (xp != ip) {
1059 		/*
1060 		 * From name resolves to a different inode.  IN_RENAME is
1061 		 * not sufficient protection against timing window races
1062 		 * so we can't panic here.
1063 		 */
1064 	} else {
1065 		/*
1066 		 * If the source is a directory with a
1067 		 * new parent, the link count of the old
1068 		 * parent directory must be decremented
1069 		 * and ".." set to point to the new parent.
1070 		 */
1071 		if (doingdirectory && newparent) {
1072 			ext2_dec_nlink(dp);
1073 			dp->i_flag |= IN_CHANGE;
1074 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1075 			    sizeof(struct dirtemplate), (off_t)0,
1076 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1077 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1078 			if (error == 0) {
1079 				/* Like ufs little-endian: */
1080 				namlen = dirbuf.dotdot_type;
1081 				if (namlen != 2 ||
1082 				    dirbuf.dotdot_name[0] != '.' ||
1083 				    dirbuf.dotdot_name[1] != '.') {
1084 					ext2_dirbad(xp, (doff_t)12,
1085 					    "rename: mangled dir");
1086 				} else {
1087 					dirbuf.dotdot_ino = newparent;
1088 					(void)vn_rdwr(UIO_WRITE, fvp,
1089 					    (caddr_t)&dirbuf,
1090 					    sizeof(struct dirtemplate),
1091 					    (off_t)0, UIO_SYSSPACE,
1092 					    IO_NODELOCKED | IO_SYNC |
1093 					    IO_NOMACCHECK, tcnp->cn_cred,
1094 					    NOCRED, NULL, NULL);
1095 					cache_purge(fdvp);
1096 				}
1097 			}
1098 		}
1099 		error = ext2_dirremove(fdvp, fcnp);
1100 		if (!error) {
1101 			ext2_dec_nlink(xp);
1102 			xp->i_flag |= IN_CHANGE;
1103 		}
1104 		xp->i_flag &= ~IN_RENAME;
1105 	}
1106 	if (dp)
1107 		vput(fdvp);
1108 	if (xp)
1109 		vput(fvp);
1110 	vrele(ap->a_fvp);
1111 	return (error);
1112 
1113 bad:
1114 	if (xp)
1115 		vput(ITOV(xp));
1116 	vput(ITOV(dp));
1117 out:
1118 	if (doingdirectory)
1119 		ip->i_flag &= ~IN_RENAME;
1120 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1121 		ext2_dec_nlink(ip);
1122 		ip->i_flag |= IN_CHANGE;
1123 		ip->i_flag &= ~IN_RENAME;
1124 		vput(fvp);
1125 	} else
1126 		vrele(fvp);
1127 	return (error);
1128 }
1129 
1130 #ifdef UFS_ACL
1131 static int
1132 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1133     mode_t dmode, struct ucred *cred, struct thread *td)
1134 {
1135 	int error;
1136 	struct inode *ip = VTOI(tvp);
1137 	struct acl *dacl, *acl;
1138 
1139 	acl = acl_alloc(M_WAITOK);
1140 	dacl = acl_alloc(M_WAITOK);
1141 
1142 	/*
1143 	 * Retrieve default ACL from parent, if any.
1144 	 */
1145 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1146 	switch (error) {
1147 	case 0:
1148 		/*
1149 		 * Retrieved a default ACL, so merge mode and ACL if
1150 		 * necessary.  If the ACL is empty, fall through to
1151 		 * the "not defined or available" case.
1152 		 */
1153 		if (acl->acl_cnt != 0) {
1154 			dmode = acl_posix1e_newfilemode(dmode, acl);
1155 			ip->i_mode = dmode;
1156 			*dacl = *acl;
1157 			ext2_sync_acl_from_inode(ip, acl);
1158 			break;
1159 		}
1160 		/* FALLTHROUGH */
1161 
1162 	case EOPNOTSUPP:
1163 		/*
1164 		 * Just use the mode as-is.
1165 		 */
1166 		ip->i_mode = dmode;
1167 		error = 0;
1168 		goto out;
1169 
1170 	default:
1171 		goto out;
1172 	}
1173 
1174 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1175 	if (error == 0)
1176 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1177 	switch (error) {
1178 	case 0:
1179 		break;
1180 
1181 	case EOPNOTSUPP:
1182 		/*
1183 		 * XXX: This should not happen, as EOPNOTSUPP above
1184 		 * was supposed to free acl.
1185 		 */
1186 #ifdef DEBUG
1187 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1188 #endif	/* DEBUG */
1189 		break;
1190 
1191 	default:
1192 		goto out;
1193 	}
1194 
1195 out:
1196 	acl_free(acl);
1197 	acl_free(dacl);
1198 
1199 	return (error);
1200 }
1201 
1202 static int
1203 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1204     mode_t mode, struct ucred *cred, struct thread *td)
1205 {
1206 	int error;
1207 	struct inode *ip = VTOI(tvp);
1208 	struct acl *acl;
1209 
1210 	acl = acl_alloc(M_WAITOK);
1211 
1212 	/*
1213 	 * Retrieve default ACL for parent, if any.
1214 	 */
1215 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1216 	switch (error) {
1217 	case 0:
1218 		/*
1219 		 * Retrieved a default ACL, so merge mode and ACL if
1220 		 * necessary.
1221 		 */
1222 		if (acl->acl_cnt != 0) {
1223 			/*
1224 			 * Two possible ways for default ACL to not
1225 			 * be present.  First, the EA can be
1226 			 * undefined, or second, the default ACL can
1227 			 * be blank.  If it's blank, fall through to
1228 			 * the it's not defined case.
1229 			 */
1230 			mode = acl_posix1e_newfilemode(mode, acl);
1231 			ip->i_mode = mode;
1232 			ext2_sync_acl_from_inode(ip, acl);
1233 			break;
1234 		}
1235 		/* FALLTHROUGH */
1236 
1237 	case EOPNOTSUPP:
1238 		/*
1239 		 * Just use the mode as-is.
1240 		 */
1241 		ip->i_mode = mode;
1242 		error = 0;
1243 		goto out;
1244 
1245 	default:
1246 		goto out;
1247 	}
1248 
1249 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1250 	switch (error) {
1251 	case 0:
1252 		break;
1253 
1254 	case EOPNOTSUPP:
1255 		/*
1256 		 * XXX: This should not happen, as EOPNOTSUPP above was
1257 		 * supposed to free acl.
1258 		 */
1259 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1260 		    "but no VOP_SETACL()\n");
1261 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1262 		    "but no VOP_SETACL()"); */
1263 		break;
1264 
1265 	default:
1266 		goto out;
1267 	}
1268 
1269 out:
1270 	acl_free(acl);
1271 
1272 	return (error);
1273 }
1274 
1275 #endif /* UFS_ACL */
1276 
1277 /*
1278  * Mkdir system call
1279  */
1280 static int
1281 ext2_mkdir(struct vop_mkdir_args *ap)
1282 {
1283 	struct vnode *dvp = ap->a_dvp;
1284 	struct vattr *vap = ap->a_vap;
1285 	struct componentname *cnp = ap->a_cnp;
1286 	struct inode *ip, *dp;
1287 	struct vnode *tvp;
1288 	struct dirtemplate dirtemplate, *dtp;
1289 	int error, dmode;
1290 
1291 #ifdef INVARIANTS
1292 	if ((cnp->cn_flags & HASBUF) == 0)
1293 		panic("ext2_mkdir: no name");
1294 #endif
1295 	dp = VTOI(dvp);
1296 	if ((nlink_t)dp->i_nlink >= ext2_max_nlink(dp) &&
1297 	    !ext2_htree_has_idx(dp)) {
1298 		error = EMLINK;
1299 		goto out;
1300 	}
1301 	dmode = vap->va_mode & 0777;
1302 	dmode |= IFDIR;
1303 	/*
1304 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1305 	 * but not have it entered in the parent directory. The entry is
1306 	 * made later after writing "." and ".." entries.
1307 	 */
1308 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1309 	if (error)
1310 		goto out;
1311 	ip = VTOI(tvp);
1312 	ip->i_gid = dp->i_gid;
1313 #ifdef SUIDDIR
1314 	{
1315 		/*
1316 		 * if we are hacking owners here, (only do this where told to)
1317 		 * and we are not giving it TOO root, (would subvert quotas)
1318 		 * then go ahead and give it to the other user.
1319 		 * The new directory also inherits the SUID bit.
1320 		 * If user's UID and dir UID are the same,
1321 		 * 'give it away' so that the SUID is still forced on.
1322 		 */
1323 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1324 		    (dp->i_mode & ISUID) && dp->i_uid) {
1325 			dmode |= ISUID;
1326 			ip->i_uid = dp->i_uid;
1327 		} else {
1328 			ip->i_uid = cnp->cn_cred->cr_uid;
1329 		}
1330 	}
1331 #else
1332 	ip->i_uid = cnp->cn_cred->cr_uid;
1333 #endif
1334 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1335 	ip->i_mode = dmode;
1336 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1337 	ip->i_nlink = 2;
1338 	if (cnp->cn_flags & ISWHITEOUT)
1339 		ip->i_flags |= UF_OPAQUE;
1340 	error = ext2_update(tvp, 1);
1341 
1342 	/*
1343 	 * Bump link count in parent directory
1344 	 * to reflect work done below.  Should
1345 	 * be done before reference is created
1346 	 * so reparation is possible if we crash.
1347 	 */
1348 	ext2_inc_nlink(dp);
1349 	dp->i_flag |= IN_CHANGE;
1350 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1351 	if (error)
1352 		goto bad;
1353 
1354 	/* Initialize directory with "." and ".." from static template. */
1355 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1356 	    EXT2F_INCOMPAT_FTYPE))
1357 		dtp = &mastertemplate;
1358 	else
1359 		dtp = &omastertemplate;
1360 	dirtemplate = *dtp;
1361 	dirtemplate.dot_ino = ip->i_number;
1362 	dirtemplate.dotdot_ino = dp->i_number;
1363 	/*
1364 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1365 	 * just redefine it - for this function only
1366 	 */
1367 #undef  DIRBLKSIZ
1368 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1369 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1370 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1371 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1372 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1373 	    NULL, NULL);
1374 	if (error) {
1375 		ext2_dec_nlink(dp);
1376 		dp->i_flag |= IN_CHANGE;
1377 		goto bad;
1378 	}
1379 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1380 		/* XXX should grow with balloc() */
1381 		panic("ext2_mkdir: blksize");
1382 	else {
1383 		ip->i_size = DIRBLKSIZ;
1384 		ip->i_flag |= IN_CHANGE;
1385 	}
1386 
1387 #ifdef UFS_ACL
1388 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1389 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1390 		    cnp->cn_cred, cnp->cn_thread);
1391 		if (error)
1392 			goto bad;
1393 	}
1394 
1395 #endif /* UFS_ACL */
1396 
1397 	/* Directory set up, now install its entry in the parent directory. */
1398 	error = ext2_direnter(ip, dvp, cnp);
1399 	if (error) {
1400 		ext2_dec_nlink(dp);
1401 		dp->i_flag |= IN_CHANGE;
1402 	}
1403 bad:
1404 	/*
1405 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1406 	 * for us because we set the link count to 0.
1407 	 */
1408 	if (error) {
1409 		ip->i_nlink = 0;
1410 		ip->i_flag |= IN_CHANGE;
1411 		vput(tvp);
1412 	} else
1413 		*ap->a_vpp = tvp;
1414 out:
1415 	return (error);
1416 #undef  DIRBLKSIZ
1417 #define DIRBLKSIZ  DEV_BSIZE
1418 }
1419 
1420 /*
1421  * Rmdir system call.
1422  */
1423 static int
1424 ext2_rmdir(struct vop_rmdir_args *ap)
1425 {
1426 	struct vnode *vp = ap->a_vp;
1427 	struct vnode *dvp = ap->a_dvp;
1428 	struct componentname *cnp = ap->a_cnp;
1429 	struct inode *ip, *dp;
1430 	int error;
1431 
1432 	ip = VTOI(vp);
1433 	dp = VTOI(dvp);
1434 
1435 	/*
1436 	 * Verify the directory is empty (and valid).
1437 	 * (Rmdir ".." won't be valid since
1438 	 *  ".." will contain a reference to
1439 	 *  the current directory and thus be
1440 	 *  non-empty.)
1441 	 */
1442 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1443 		error = ENOTEMPTY;
1444 		goto out;
1445 	}
1446 	if ((dp->i_flags & APPEND)
1447 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1448 		error = EPERM;
1449 		goto out;
1450 	}
1451 	/*
1452 	 * Delete reference to directory before purging
1453 	 * inode.  If we crash in between, the directory
1454 	 * will be reattached to lost+found,
1455 	 */
1456 	error = ext2_dirremove(dvp, cnp);
1457 	if (error)
1458 		goto out;
1459 	ext2_dec_nlink(dp);
1460 	dp->i_flag |= IN_CHANGE;
1461 	cache_purge(dvp);
1462 	VOP_UNLOCK(dvp, 0);
1463 	/*
1464 	 * Truncate inode.  The only stuff left
1465 	 * in the directory is "." and "..".
1466 	 */
1467 	ip->i_nlink = 0;
1468 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1469 	    cnp->cn_thread);
1470 	cache_purge(ITOV(ip));
1471 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1472 		VOP_UNLOCK(vp, 0);
1473 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1474 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1475 	}
1476 out:
1477 	return (error);
1478 }
1479 
1480 /*
1481  * symlink -- make a symbolic link
1482  */
1483 static int
1484 ext2_symlink(struct vop_symlink_args *ap)
1485 {
1486 	struct vnode *vp, **vpp = ap->a_vpp;
1487 	struct inode *ip;
1488 	int len, error;
1489 
1490 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1491 	    vpp, ap->a_cnp);
1492 	if (error)
1493 		return (error);
1494 	vp = *vpp;
1495 	len = strlen(ap->a_target);
1496 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1497 		ip = VTOI(vp);
1498 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1499 		ip->i_size = len;
1500 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1501 	} else
1502 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1503 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1504 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1505 	if (error)
1506 		vput(vp);
1507 	return (error);
1508 }
1509 
1510 /*
1511  * Return target name of a symbolic link
1512  */
1513 static int
1514 ext2_readlink(struct vop_readlink_args *ap)
1515 {
1516 	struct vnode *vp = ap->a_vp;
1517 	struct inode *ip = VTOI(vp);
1518 	int isize;
1519 
1520 	isize = ip->i_size;
1521 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1522 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1523 		return (0);
1524 	}
1525 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1526 }
1527 
1528 /*
1529  * Calculate the logical to physical mapping if not done already,
1530  * then call the device strategy routine.
1531  *
1532  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1533  * deadlock on memory.  See ext2_bmap() for details.
1534  */
1535 static int
1536 ext2_strategy(struct vop_strategy_args *ap)
1537 {
1538 	struct buf *bp = ap->a_bp;
1539 	struct vnode *vp = ap->a_vp;
1540 	struct bufobj *bo;
1541 	daddr_t blkno;
1542 	int error;
1543 
1544 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1545 		panic("ext2_strategy: spec");
1546 	if (bp->b_blkno == bp->b_lblkno) {
1547 
1548 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1549 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1550 		else
1551 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1552 
1553 		bp->b_blkno = blkno;
1554 		if (error) {
1555 			bp->b_error = error;
1556 			bp->b_ioflags |= BIO_ERROR;
1557 			bufdone(bp);
1558 			return (0);
1559 		}
1560 		if ((long)bp->b_blkno == -1)
1561 			vfs_bio_clrbuf(bp);
1562 	}
1563 	if ((long)bp->b_blkno == -1) {
1564 		bufdone(bp);
1565 		return (0);
1566 	}
1567 	bp->b_iooffset = dbtob(bp->b_blkno);
1568 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1569 	BO_STRATEGY(bo, bp);
1570 	return (0);
1571 }
1572 
1573 /*
1574  * Print out the contents of an inode.
1575  */
1576 static int
1577 ext2_print(struct vop_print_args *ap)
1578 {
1579 	struct vnode *vp = ap->a_vp;
1580 	struct inode *ip = VTOI(vp);
1581 
1582 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1583 	if (vp->v_type == VFIFO)
1584 		fifo_printinfo(vp);
1585 	printf("\n");
1586 	return (0);
1587 }
1588 
1589 /*
1590  * Close wrapper for fifos.
1591  *
1592  * Update the times on the inode then do device close.
1593  */
1594 static int
1595 ext2fifo_close(struct vop_close_args *ap)
1596 {
1597 	struct vnode *vp = ap->a_vp;
1598 
1599 	VI_LOCK(vp);
1600 	if (vp->v_usecount > 1)
1601 		ext2_itimes_locked(vp);
1602 	VI_UNLOCK(vp);
1603 	return (fifo_specops.vop_close(ap));
1604 }
1605 
1606 /*
1607  * Kqfilter wrapper for fifos.
1608  *
1609  * Fall through to ext2 kqfilter routines if needed
1610  */
1611 static int
1612 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1613 {
1614 	int error;
1615 
1616 	error = fifo_specops.vop_kqfilter(ap);
1617 	if (error)
1618 		error = vfs_kqfilter(ap);
1619 	return (error);
1620 }
1621 
1622 /*
1623  * Return POSIX pathconf information applicable to ext2 filesystems.
1624  */
1625 static int
1626 ext2_pathconf(struct vop_pathconf_args *ap)
1627 {
1628 	int error = 0;
1629 
1630 	switch (ap->a_name) {
1631 	case _PC_LINK_MAX:
1632 		if (ext2_htree_has_idx(VTOI(ap->a_vp)))
1633 			*ap->a_retval = INT_MAX;
1634 		else
1635 			*ap->a_retval = ext2_max_nlink(VTOI(ap->a_vp));
1636 		break;
1637 	case _PC_NAME_MAX:
1638 		*ap->a_retval = NAME_MAX;
1639 		break;
1640 	case _PC_PIPE_BUF:
1641 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1642 			*ap->a_retval = PIPE_BUF;
1643 		else
1644 			error = EINVAL;
1645 		break;
1646 	case _PC_CHOWN_RESTRICTED:
1647 		*ap->a_retval = 1;
1648 		break;
1649 	case _PC_NO_TRUNC:
1650 		*ap->a_retval = 1;
1651 		break;
1652 
1653 #ifdef UFS_ACL
1654 	case _PC_ACL_EXTENDED:
1655 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1656 			*ap->a_retval = 1;
1657 		else
1658 			*ap->a_retval = 0;
1659 		break;
1660 	case _PC_ACL_PATH_MAX:
1661 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1662 			*ap->a_retval = ACL_MAX_ENTRIES;
1663 		else
1664 			*ap->a_retval = 3;
1665 		break;
1666 #endif /* UFS_ACL */
1667 
1668 	case _PC_MIN_HOLE_SIZE:
1669 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1670 		break;
1671 	case _PC_PRIO_IO:
1672 		*ap->a_retval = 0;
1673 		break;
1674 	case _PC_SYNC_IO:
1675 		*ap->a_retval = 0;
1676 		break;
1677 	case _PC_ALLOC_SIZE_MIN:
1678 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1679 		break;
1680 	case _PC_FILESIZEBITS:
1681 		*ap->a_retval = 64;
1682 		break;
1683 	case _PC_REC_INCR_XFER_SIZE:
1684 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1685 		break;
1686 	case _PC_REC_MAX_XFER_SIZE:
1687 		*ap->a_retval = -1;	/* means ``unlimited'' */
1688 		break;
1689 	case _PC_REC_MIN_XFER_SIZE:
1690 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1691 		break;
1692 	case _PC_REC_XFER_ALIGN:
1693 		*ap->a_retval = PAGE_SIZE;
1694 		break;
1695 	case _PC_SYMLINK_MAX:
1696 		*ap->a_retval = MAXPATHLEN;
1697 		break;
1698 
1699 	default:
1700 		error = vop_stdpathconf(ap);
1701 		break;
1702 	}
1703 	return (error);
1704 }
1705 
1706 /*
1707  * Vnode operation to remove a named attribute.
1708  */
1709 static int
1710 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1711 {
1712 	struct inode *ip;
1713 	struct m_ext2fs *fs;
1714 	int error;
1715 
1716 	ip = VTOI(ap->a_vp);
1717 	fs = ip->i_e2fs;
1718 
1719 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1720 		return (EOPNOTSUPP);
1721 
1722 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1723 		return (EOPNOTSUPP);
1724 
1725 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1726 	    ap->a_cred, ap->a_td, VWRITE);
1727 	if (error)
1728 		return (error);
1729 
1730 	error = ENOATTR;
1731 
1732 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1733 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1734 		if (error != ENOATTR)
1735 			return (error);
1736 	}
1737 
1738 	if (ip->i_facl)
1739 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1740 
1741 	return (error);
1742 }
1743 
1744 /*
1745  * Vnode operation to retrieve a named extended attribute.
1746  */
1747 static int
1748 ext2_getextattr(struct vop_getextattr_args *ap)
1749 {
1750 	struct inode *ip;
1751 	struct m_ext2fs *fs;
1752 	int error;
1753 
1754 	ip = VTOI(ap->a_vp);
1755 	fs = ip->i_e2fs;
1756 
1757 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1758 		return (EOPNOTSUPP);
1759 
1760 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1761 		return (EOPNOTSUPP);
1762 
1763 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1764 	    ap->a_cred, ap->a_td, VREAD);
1765 	if (error)
1766 		return (error);
1767 
1768 	if (ap->a_size != NULL)
1769 		*ap->a_size = 0;
1770 
1771 	error = ENOATTR;
1772 
1773 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1774 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1775 		    ap->a_name, ap->a_uio, ap->a_size);
1776 		if (error != ENOATTR)
1777 			return (error);
1778 	}
1779 
1780 	if (ip->i_facl)
1781 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1782 		    ap->a_name, ap->a_uio, ap->a_size);
1783 
1784 	return (error);
1785 }
1786 
1787 /*
1788  * Vnode operation to retrieve extended attributes on a vnode.
1789  */
1790 static int
1791 ext2_listextattr(struct vop_listextattr_args *ap)
1792 {
1793 	struct inode *ip;
1794 	struct m_ext2fs *fs;
1795 	int error;
1796 
1797 	ip = VTOI(ap->a_vp);
1798 	fs = ip->i_e2fs;
1799 
1800 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1801 		return (EOPNOTSUPP);
1802 
1803 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1804 		return (EOPNOTSUPP);
1805 
1806 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1807 	    ap->a_cred, ap->a_td, VREAD);
1808 	if (error)
1809 		return (error);
1810 
1811 	if (ap->a_size != NULL)
1812 		*ap->a_size = 0;
1813 
1814 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1815 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1816 		    ap->a_uio, ap->a_size);
1817 		if (error)
1818 			return (error);
1819 	}
1820 
1821 	if (ip->i_facl)
1822 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1823 		    ap->a_uio, ap->a_size);
1824 
1825 	return (error);
1826 }
1827 
1828 /*
1829  * Vnode operation to set a named attribute.
1830  */
1831 static int
1832 ext2_setextattr(struct vop_setextattr_args *ap)
1833 {
1834 	struct inode *ip;
1835 	struct m_ext2fs *fs;
1836 	int error;
1837 
1838 	ip = VTOI(ap->a_vp);
1839 	fs = ip->i_e2fs;
1840 
1841 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1842 		return (EOPNOTSUPP);
1843 
1844 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1845 		return (EOPNOTSUPP);
1846 
1847 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1848 	    ap->a_cred, ap->a_td, VWRITE);
1849 	if (error)
1850 		return (error);
1851 
1852 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1853 	if (error)
1854 		return (error);
1855 
1856 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1857 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1858 		    ap->a_name, ap->a_uio);
1859 		if (error != ENOSPC)
1860 			return (error);
1861 	}
1862 
1863 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1864 	    ap->a_name, ap->a_uio);
1865 
1866 	return (error);
1867 }
1868 
1869 /*
1870  * Vnode pointer to File handle
1871  */
1872 /* ARGSUSED */
1873 static int
1874 ext2_vptofh(struct vop_vptofh_args *ap)
1875 {
1876 	struct inode *ip;
1877 	struct ufid *ufhp;
1878 
1879 	ip = VTOI(ap->a_vp);
1880 	ufhp = (struct ufid *)ap->a_fhp;
1881 	ufhp->ufid_len = sizeof(struct ufid);
1882 	ufhp->ufid_ino = ip->i_number;
1883 	ufhp->ufid_gen = ip->i_gen;
1884 	return (0);
1885 }
1886 
1887 /*
1888  * Initialize the vnode associated with a new inode, handle aliased
1889  * vnodes.
1890  */
1891 int
1892 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1893 {
1894 	struct inode *ip;
1895 	struct vnode *vp;
1896 
1897 	vp = *vpp;
1898 	ip = VTOI(vp);
1899 	vp->v_type = IFTOVT(ip->i_mode);
1900 	if (vp->v_type == VFIFO)
1901 		vp->v_op = fifoops;
1902 
1903 	if (ip->i_number == EXT2_ROOTINO)
1904 		vp->v_vflag |= VV_ROOT;
1905 	ip->i_modrev = init_va_filerev();
1906 	*vpp = vp;
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Allocate a new inode.
1912  */
1913 static int
1914 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1915     struct componentname *cnp)
1916 {
1917 	struct inode *ip, *pdir;
1918 	struct vnode *tvp;
1919 	int error;
1920 
1921 	pdir = VTOI(dvp);
1922 #ifdef INVARIANTS
1923 	if ((cnp->cn_flags & HASBUF) == 0)
1924 		panic("ext2_makeinode: no name");
1925 #endif
1926 	*vpp = NULL;
1927 	if ((mode & IFMT) == 0)
1928 		mode |= IFREG;
1929 
1930 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1931 	if (error) {
1932 		return (error);
1933 	}
1934 	ip = VTOI(tvp);
1935 	ip->i_gid = pdir->i_gid;
1936 #ifdef SUIDDIR
1937 	{
1938 		/*
1939 		 * if we are
1940 		 * not the owner of the directory,
1941 		 * and we are hacking owners here, (only do this where told to)
1942 		 * and we are not giving it TOO root, (would subvert quotas)
1943 		 * then go ahead and give it to the other user.
1944 		 * Note that this drops off the execute bits for security.
1945 		 */
1946 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1947 		    (pdir->i_mode & ISUID) &&
1948 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1949 			ip->i_uid = pdir->i_uid;
1950 			mode &= ~07111;
1951 		} else {
1952 			ip->i_uid = cnp->cn_cred->cr_uid;
1953 		}
1954 	}
1955 #else
1956 	ip->i_uid = cnp->cn_cred->cr_uid;
1957 #endif
1958 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1959 	ip->i_mode = mode;
1960 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1961 	ip->i_nlink = 1;
1962 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1963 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1964 			ip->i_mode &= ~ISGID;
1965 	}
1966 
1967 	if (cnp->cn_flags & ISWHITEOUT)
1968 		ip->i_flags |= UF_OPAQUE;
1969 
1970 	/*
1971 	 * Make sure inode goes to disk before directory entry.
1972 	 */
1973 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1974 	if (error)
1975 		goto bad;
1976 
1977 #ifdef UFS_ACL
1978 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1979 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
1980 		    cnp->cn_cred, cnp->cn_thread);
1981 		if (error)
1982 			goto bad;
1983 	}
1984 #endif /* UFS_ACL */
1985 
1986 	error = ext2_direnter(ip, dvp, cnp);
1987 	if (error)
1988 		goto bad;
1989 
1990 	*vpp = tvp;
1991 	return (0);
1992 
1993 bad:
1994 	/*
1995 	 * Write error occurred trying to update the inode
1996 	 * or the directory so must deallocate the inode.
1997 	 */
1998 	ip->i_nlink = 0;
1999 	ip->i_flag |= IN_CHANGE;
2000 	vput(tvp);
2001 	return (error);
2002 }
2003 
2004 /*
2005  * Vnode op for reading.
2006  */
2007 static int
2008 ext2_read(struct vop_read_args *ap)
2009 {
2010 	struct vnode *vp;
2011 	struct inode *ip;
2012 	struct uio *uio;
2013 	struct m_ext2fs *fs;
2014 	struct buf *bp;
2015 	daddr_t lbn, nextlbn;
2016 	off_t bytesinfile;
2017 	long size, xfersize, blkoffset;
2018 	int error, orig_resid, seqcount;
2019 	int ioflag;
2020 
2021 	vp = ap->a_vp;
2022 	uio = ap->a_uio;
2023 	ioflag = ap->a_ioflag;
2024 
2025 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2026 	ip = VTOI(vp);
2027 
2028 #ifdef INVARIANTS
2029 	if (uio->uio_rw != UIO_READ)
2030 		panic("%s: mode", "ext2_read");
2031 
2032 	if (vp->v_type == VLNK) {
2033 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
2034 			panic("%s: short symlink", "ext2_read");
2035 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2036 		panic("%s: type %d", "ext2_read", vp->v_type);
2037 #endif
2038 	orig_resid = uio->uio_resid;
2039 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2040 	if (orig_resid == 0)
2041 		return (0);
2042 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2043 	fs = ip->i_e2fs;
2044 	if (uio->uio_offset < ip->i_size &&
2045 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2046 		return (EOVERFLOW);
2047 
2048 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2049 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2050 			break;
2051 		lbn = lblkno(fs, uio->uio_offset);
2052 		nextlbn = lbn + 1;
2053 		size = blksize(fs, ip, lbn);
2054 		blkoffset = blkoff(fs, uio->uio_offset);
2055 
2056 		xfersize = fs->e2fs_fsize - blkoffset;
2057 		if (uio->uio_resid < xfersize)
2058 			xfersize = uio->uio_resid;
2059 		if (bytesinfile < xfersize)
2060 			xfersize = bytesinfile;
2061 
2062 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2063 			error = bread(vp, lbn, size, NOCRED, &bp);
2064 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2065 			error = cluster_read(vp, ip->i_size, lbn, size,
2066 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2067 			    0, &bp);
2068 		} else if (seqcount > 1) {
2069 			u_int nextsize = blksize(fs, ip, nextlbn);
2070 
2071 			error = breadn(vp, lbn,
2072 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2073 		} else
2074 			error = bread(vp, lbn, size, NOCRED, &bp);
2075 		if (error) {
2076 			brelse(bp);
2077 			bp = NULL;
2078 			break;
2079 		}
2080 
2081 		/*
2082 		 * We should only get non-zero b_resid when an I/O error
2083 		 * has occurred, which should cause us to break above.
2084 		 * However, if the short read did not cause an error,
2085 		 * then we want to ensure that we do not uiomove bad
2086 		 * or uninitialized data.
2087 		 */
2088 		size -= bp->b_resid;
2089 		if (size < xfersize) {
2090 			if (size == 0)
2091 				break;
2092 			xfersize = size;
2093 		}
2094 		error = uiomove((char *)bp->b_data + blkoffset,
2095 		    (int)xfersize, uio);
2096 		if (error)
2097 			break;
2098 		vfs_bio_brelse(bp, ioflag);
2099 	}
2100 
2101 	/*
2102 	 * This can only happen in the case of an error because the loop
2103 	 * above resets bp to NULL on each iteration and on normal
2104 	 * completion has not set a new value into it. so it must have come
2105 	 * from a 'break' statement
2106 	 */
2107 	if (bp != NULL)
2108 		vfs_bio_brelse(bp, ioflag);
2109 
2110 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2111 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2112 		ip->i_flag |= IN_ACCESS;
2113 	return (error);
2114 }
2115 
2116 static int
2117 ext2_ioctl(struct vop_ioctl_args *ap)
2118 {
2119 
2120 	switch (ap->a_command) {
2121 	case FIOSEEKDATA:
2122 	case FIOSEEKHOLE:
2123 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
2124 		    (off_t *)ap->a_data, ap->a_cred));
2125 	default:
2126 		return (ENOTTY);
2127 	}
2128 }
2129 
2130 /*
2131  * Vnode op for writing.
2132  */
2133 static int
2134 ext2_write(struct vop_write_args *ap)
2135 {
2136 	struct vnode *vp;
2137 	struct uio *uio;
2138 	struct inode *ip;
2139 	struct m_ext2fs *fs;
2140 	struct buf *bp;
2141 	daddr_t lbn;
2142 	off_t osize;
2143 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2144 
2145 	ioflag = ap->a_ioflag;
2146 	uio = ap->a_uio;
2147 	vp = ap->a_vp;
2148 
2149 	seqcount = ioflag >> IO_SEQSHIFT;
2150 	ip = VTOI(vp);
2151 
2152 #ifdef INVARIANTS
2153 	if (uio->uio_rw != UIO_WRITE)
2154 		panic("%s: mode", "ext2_write");
2155 #endif
2156 
2157 	switch (vp->v_type) {
2158 	case VREG:
2159 		if (ioflag & IO_APPEND)
2160 			uio->uio_offset = ip->i_size;
2161 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2162 			return (EPERM);
2163 		/* FALLTHROUGH */
2164 	case VLNK:
2165 		break;
2166 	case VDIR:
2167 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2168 		if ((ioflag & IO_SYNC) == 0)
2169 			panic("ext2_write: nonsync dir write");
2170 		break;
2171 	default:
2172 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2173 		    vp->v_type, (intmax_t)uio->uio_offset,
2174 		    (intmax_t)uio->uio_resid);
2175 	}
2176 
2177 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2178 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2179 	fs = ip->i_e2fs;
2180 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2181 		return (EFBIG);
2182 	/*
2183 	 * Maybe this should be above the vnode op call, but so long as
2184 	 * file servers have no limits, I don't think it matters.
2185 	 */
2186 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2187 		return (EFBIG);
2188 
2189 	resid = uio->uio_resid;
2190 	osize = ip->i_size;
2191 	if (seqcount > BA_SEQMAX)
2192 		flags = BA_SEQMAX << BA_SEQSHIFT;
2193 	else
2194 		flags = seqcount << BA_SEQSHIFT;
2195 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2196 		flags |= IO_SYNC;
2197 
2198 	for (error = 0; uio->uio_resid > 0;) {
2199 		lbn = lblkno(fs, uio->uio_offset);
2200 		blkoffset = blkoff(fs, uio->uio_offset);
2201 		xfersize = fs->e2fs_fsize - blkoffset;
2202 		if (uio->uio_resid < xfersize)
2203 			xfersize = uio->uio_resid;
2204 		if (uio->uio_offset + xfersize > ip->i_size)
2205 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2206 
2207 		/*
2208 		 * We must perform a read-before-write if the transfer size
2209 		 * does not cover the entire buffer.
2210 		 */
2211 		if (fs->e2fs_bsize > xfersize)
2212 			flags |= BA_CLRBUF;
2213 		else
2214 			flags &= ~BA_CLRBUF;
2215 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2216 		    ap->a_cred, &bp, flags);
2217 		if (error != 0)
2218 			break;
2219 
2220 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2221 			bp->b_flags |= B_NOCACHE;
2222 		if (uio->uio_offset + xfersize > ip->i_size)
2223 			ip->i_size = uio->uio_offset + xfersize;
2224 		size = blksize(fs, ip, lbn) - bp->b_resid;
2225 		if (size < xfersize)
2226 			xfersize = size;
2227 
2228 		error =
2229 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2230 		/*
2231 		 * If the buffer is not already filled and we encounter an
2232 		 * error while trying to fill it, we have to clear out any
2233 		 * garbage data from the pages instantiated for the buffer.
2234 		 * If we do not, a failed uiomove() during a write can leave
2235 		 * the prior contents of the pages exposed to a userland mmap.
2236 		 *
2237 		 * Note that we need only clear buffers with a transfer size
2238 		 * equal to the block size because buffers with a shorter
2239 		 * transfer size were cleared above by the call to ext2_balloc()
2240 		 * with the BA_CLRBUF flag set.
2241 		 *
2242 		 * If the source region for uiomove identically mmaps the
2243 		 * buffer, uiomove() performed the NOP copy, and the buffer
2244 		 * content remains valid because the page fault handler
2245 		 * validated the pages.
2246 		 */
2247 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2248 		    fs->e2fs_bsize == xfersize)
2249 			vfs_bio_clrbuf(bp);
2250 
2251 		vfs_bio_set_flags(bp, ioflag);
2252 
2253 		/*
2254 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2255 		 * if we have a severe page deficiency write the buffer
2256 		 * asynchronously.  Otherwise try to cluster, and if that
2257 		 * doesn't do it then either do an async write (if O_DIRECT),
2258 		 * or a delayed write (if not).
2259 		 */
2260 		if (ioflag & IO_SYNC) {
2261 			(void)bwrite(bp);
2262 		} else if (vm_page_count_severe() ||
2263 			    buf_dirty_count_severe() ||
2264 		    (ioflag & IO_ASYNC)) {
2265 			bp->b_flags |= B_CLUSTEROK;
2266 			bawrite(bp);
2267 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2268 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2269 				bp->b_flags |= B_CLUSTEROK;
2270 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2271 			} else {
2272 				bawrite(bp);
2273 			}
2274 		} else if (ioflag & IO_DIRECT) {
2275 			bp->b_flags |= B_CLUSTEROK;
2276 			bawrite(bp);
2277 		} else {
2278 			bp->b_flags |= B_CLUSTEROK;
2279 			bdwrite(bp);
2280 		}
2281 		if (error || xfersize == 0)
2282 			break;
2283 	}
2284 	/*
2285 	 * If we successfully wrote any data, and we are not the superuser
2286 	 * we clear the setuid and setgid bits as a precaution against
2287 	 * tampering.
2288 	 */
2289 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2290 	    ap->a_cred) {
2291 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2292 			ip->i_mode &= ~(ISUID | ISGID);
2293 	}
2294 	if (error) {
2295 		if (ioflag & IO_UNIT) {
2296 			(void)ext2_truncate(vp, osize,
2297 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2298 			uio->uio_offset -= resid - uio->uio_resid;
2299 			uio->uio_resid = resid;
2300 		}
2301 	}
2302 	if (uio->uio_resid != resid) {
2303 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2304 		if (ioflag & IO_SYNC)
2305 			error = ext2_update(vp, 1);
2306 	}
2307 	return (error);
2308 }
2309