xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 488ab515d6cc02f6f743f0badfc8e94eb553cd30)
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/ext2fs.h>
88 #include <fs/ext2fs/ext2_extern.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 			dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1075 			if (!dirbuf) {
1076 				error = ENOMEM;
1077 				goto bad;
1078 			}
1079 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1080 			    ip->i_e2fs->e2fs_bsize, (off_t)0,
1081 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1082 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1083 			if (error == 0) {
1084 				/* Like ufs little-endian: */
1085 				namlen = dirbuf->dotdot_type;
1086 				if (namlen != 2 ||
1087 				    dirbuf->dotdot_name[0] != '.' ||
1088 				    dirbuf->dotdot_name[1] != '.') {
1089 					ext2_dirbad(xp, (doff_t)12,
1090 					    "rename: mangled dir");
1091 				} else {
1092 					dirbuf->dotdot_ino = newparent;
1093 					ext2_dir_blk_csum_set_mem(ip,
1094 					    (char *)dirbuf,
1095 					    ip->i_e2fs->e2fs_bsize);
1096 					(void)vn_rdwr(UIO_WRITE, fvp,
1097 					    (caddr_t)dirbuf,
1098 					    ip->i_e2fs->e2fs_bsize,
1099 					    (off_t)0, UIO_SYSSPACE,
1100 					    IO_NODELOCKED | IO_SYNC |
1101 					    IO_NOMACCHECK, tcnp->cn_cred,
1102 					    NOCRED, NULL, NULL);
1103 					cache_purge(fdvp);
1104 				}
1105 			}
1106 			free(dirbuf, M_TEMP);
1107 		}
1108 		error = ext2_dirremove(fdvp, fcnp);
1109 		if (!error) {
1110 			ext2_dec_nlink(xp);
1111 			xp->i_flag |= IN_CHANGE;
1112 		}
1113 		xp->i_flag &= ~IN_RENAME;
1114 	}
1115 	if (dp)
1116 		vput(fdvp);
1117 	if (xp)
1118 		vput(fvp);
1119 	vrele(ap->a_fvp);
1120 	return (error);
1121 
1122 bad:
1123 	if (xp)
1124 		vput(ITOV(xp));
1125 	vput(ITOV(dp));
1126 out:
1127 	if (doingdirectory)
1128 		ip->i_flag &= ~IN_RENAME;
1129 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1130 		ext2_dec_nlink(ip);
1131 		ip->i_flag |= IN_CHANGE;
1132 		ip->i_flag &= ~IN_RENAME;
1133 		vput(fvp);
1134 	} else
1135 		vrele(fvp);
1136 	return (error);
1137 }
1138 
1139 #ifdef UFS_ACL
1140 static int
1141 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1142     mode_t dmode, struct ucred *cred, struct thread *td)
1143 {
1144 	int error;
1145 	struct inode *ip = VTOI(tvp);
1146 	struct acl *dacl, *acl;
1147 
1148 	acl = acl_alloc(M_WAITOK);
1149 	dacl = acl_alloc(M_WAITOK);
1150 
1151 	/*
1152 	 * Retrieve default ACL from parent, if any.
1153 	 */
1154 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1155 	switch (error) {
1156 	case 0:
1157 		/*
1158 		 * Retrieved a default ACL, so merge mode and ACL if
1159 		 * necessary.  If the ACL is empty, fall through to
1160 		 * the "not defined or available" case.
1161 		 */
1162 		if (acl->acl_cnt != 0) {
1163 			dmode = acl_posix1e_newfilemode(dmode, acl);
1164 			ip->i_mode = dmode;
1165 			*dacl = *acl;
1166 			ext2_sync_acl_from_inode(ip, acl);
1167 			break;
1168 		}
1169 		/* FALLTHROUGH */
1170 
1171 	case EOPNOTSUPP:
1172 		/*
1173 		 * Just use the mode as-is.
1174 		 */
1175 		ip->i_mode = dmode;
1176 		error = 0;
1177 		goto out;
1178 
1179 	default:
1180 		goto out;
1181 	}
1182 
1183 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1184 	if (error == 0)
1185 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1186 	switch (error) {
1187 	case 0:
1188 		break;
1189 
1190 	case EOPNOTSUPP:
1191 		/*
1192 		 * XXX: This should not happen, as EOPNOTSUPP above
1193 		 * was supposed to free acl.
1194 		 */
1195 #ifdef DEBUG
1196 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1197 #endif	/* DEBUG */
1198 		break;
1199 
1200 	default:
1201 		goto out;
1202 	}
1203 
1204 out:
1205 	acl_free(acl);
1206 	acl_free(dacl);
1207 
1208 	return (error);
1209 }
1210 
1211 static int
1212 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1213     mode_t mode, struct ucred *cred, struct thread *td)
1214 {
1215 	int error;
1216 	struct inode *ip = VTOI(tvp);
1217 	struct acl *acl;
1218 
1219 	acl = acl_alloc(M_WAITOK);
1220 
1221 	/*
1222 	 * Retrieve default ACL for parent, if any.
1223 	 */
1224 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1225 	switch (error) {
1226 	case 0:
1227 		/*
1228 		 * Retrieved a default ACL, so merge mode and ACL if
1229 		 * necessary.
1230 		 */
1231 		if (acl->acl_cnt != 0) {
1232 			/*
1233 			 * Two possible ways for default ACL to not
1234 			 * be present.  First, the EA can be
1235 			 * undefined, or second, the default ACL can
1236 			 * be blank.  If it's blank, fall through to
1237 			 * the it's not defined case.
1238 			 */
1239 			mode = acl_posix1e_newfilemode(mode, acl);
1240 			ip->i_mode = mode;
1241 			ext2_sync_acl_from_inode(ip, acl);
1242 			break;
1243 		}
1244 		/* FALLTHROUGH */
1245 
1246 	case EOPNOTSUPP:
1247 		/*
1248 		 * Just use the mode as-is.
1249 		 */
1250 		ip->i_mode = mode;
1251 		error = 0;
1252 		goto out;
1253 
1254 	default:
1255 		goto out;
1256 	}
1257 
1258 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1259 	switch (error) {
1260 	case 0:
1261 		break;
1262 
1263 	case EOPNOTSUPP:
1264 		/*
1265 		 * XXX: This should not happen, as EOPNOTSUPP above was
1266 		 * supposed to free acl.
1267 		 */
1268 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1269 		    "but no VOP_SETACL()\n");
1270 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1271 		    "but no VOP_SETACL()"); */
1272 		break;
1273 
1274 	default:
1275 		goto out;
1276 	}
1277 
1278 out:
1279 	acl_free(acl);
1280 
1281 	return (error);
1282 }
1283 
1284 #endif /* UFS_ACL */
1285 
1286 static void
1287 ext2_init_dirent_tail(struct ext2fs_direct_tail *tp)
1288 {
1289 	memset(tp, 0, sizeof(struct ext2fs_direct_tail));
1290 	tp->e2dt_rec_len = sizeof(struct ext2fs_direct_tail);
1291 	tp->e2dt_reserved_ft = EXT2_FT_DIR_CSUM;
1292 }
1293 
1294 /*
1295  * Mkdir system call
1296  */
1297 static int
1298 ext2_mkdir(struct vop_mkdir_args *ap)
1299 {
1300 	struct m_ext2fs *fs;
1301 	struct vnode *dvp = ap->a_dvp;
1302 	struct vattr *vap = ap->a_vap;
1303 	struct componentname *cnp = ap->a_cnp;
1304 	struct inode *ip, *dp;
1305 	struct vnode *tvp;
1306 	struct dirtemplate dirtemplate, *dtp;
1307 	char *buf = NULL;
1308 	int error, dmode;
1309 
1310 #ifdef INVARIANTS
1311 	if ((cnp->cn_flags & HASBUF) == 0)
1312 		panic("ext2_mkdir: no name");
1313 #endif
1314 	dp = VTOI(dvp);
1315 	if ((nlink_t)dp->i_nlink >= ext2_max_nlink(dp) &&
1316 	    !ext2_htree_has_idx(dp)) {
1317 		error = EMLINK;
1318 		goto out;
1319 	}
1320 	dmode = vap->va_mode & 0777;
1321 	dmode |= IFDIR;
1322 	/*
1323 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1324 	 * but not have it entered in the parent directory. The entry is
1325 	 * made later after writing "." and ".." entries.
1326 	 */
1327 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1328 	if (error)
1329 		goto out;
1330 	ip = VTOI(tvp);
1331 	fs = ip->i_e2fs;
1332 	ip->i_gid = dp->i_gid;
1333 #ifdef SUIDDIR
1334 	{
1335 		/*
1336 		 * if we are hacking owners here, (only do this where told to)
1337 		 * and we are not giving it TOO root, (would subvert quotas)
1338 		 * then go ahead and give it to the other user.
1339 		 * The new directory also inherits the SUID bit.
1340 		 * If user's UID and dir UID are the same,
1341 		 * 'give it away' so that the SUID is still forced on.
1342 		 */
1343 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1344 		    (dp->i_mode & ISUID) && dp->i_uid) {
1345 			dmode |= ISUID;
1346 			ip->i_uid = dp->i_uid;
1347 		} else {
1348 			ip->i_uid = cnp->cn_cred->cr_uid;
1349 		}
1350 	}
1351 #else
1352 	ip->i_uid = cnp->cn_cred->cr_uid;
1353 #endif
1354 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1355 	ip->i_mode = dmode;
1356 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1357 	ip->i_nlink = 2;
1358 	if (cnp->cn_flags & ISWHITEOUT)
1359 		ip->i_flags |= UF_OPAQUE;
1360 	error = ext2_update(tvp, 1);
1361 
1362 	/*
1363 	 * Bump link count in parent directory
1364 	 * to reflect work done below.  Should
1365 	 * be done before reference is created
1366 	 * so reparation is possible if we crash.
1367 	 */
1368 	ext2_inc_nlink(dp);
1369 	dp->i_flag |= IN_CHANGE;
1370 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1371 	if (error)
1372 		goto bad;
1373 
1374 	/* Initialize directory with "." and ".." from static template. */
1375 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1376 	    EXT2F_INCOMPAT_FTYPE))
1377 		dtp = &mastertemplate;
1378 	else
1379 		dtp = &omastertemplate;
1380 	dirtemplate = *dtp;
1381 	dirtemplate.dot_ino = ip->i_number;
1382 	dirtemplate.dotdot_ino = dp->i_number;
1383 	/*
1384 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1385 	 * just redefine it - for this function only
1386 	 */
1387 #undef  DIRBLKSIZ
1388 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1389 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1390 	buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1391 	if (!buf) {
1392 		error = ENOMEM;
1393 		ext2_dec_nlink(dp);
1394 		dp->i_flag |= IN_CHANGE;
1395 		goto bad;
1396 	}
1397 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1398 		dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail);
1399 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1400 	}
1401 	memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1402 	ext2_dir_blk_csum_set_mem(ip, buf, DIRBLKSIZ);
1403 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1404 	    DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1405 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1406 	    NULL, NULL);
1407 	if (error) {
1408 		ext2_dec_nlink(dp);
1409 		dp->i_flag |= IN_CHANGE;
1410 		goto bad;
1411 	}
1412 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1413 		/* XXX should grow with balloc() */
1414 		panic("ext2_mkdir: blksize");
1415 	else {
1416 		ip->i_size = DIRBLKSIZ;
1417 		ip->i_flag |= IN_CHANGE;
1418 	}
1419 
1420 #ifdef UFS_ACL
1421 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1422 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1423 		    cnp->cn_cred, cnp->cn_thread);
1424 		if (error)
1425 			goto bad;
1426 	}
1427 
1428 #endif /* UFS_ACL */
1429 
1430 	/* Directory set up, now install its entry in the parent directory. */
1431 	error = ext2_direnter(ip, dvp, cnp);
1432 	if (error) {
1433 		ext2_dec_nlink(dp);
1434 		dp->i_flag |= IN_CHANGE;
1435 	}
1436 bad:
1437 	/*
1438 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1439 	 * for us because we set the link count to 0.
1440 	 */
1441 	if (error) {
1442 		ip->i_nlink = 0;
1443 		ip->i_flag |= IN_CHANGE;
1444 		vput(tvp);
1445 	} else
1446 		*ap->a_vpp = tvp;
1447 out:
1448 	free(buf, M_TEMP);
1449 	return (error);
1450 #undef  DIRBLKSIZ
1451 #define DIRBLKSIZ  DEV_BSIZE
1452 }
1453 
1454 /*
1455  * Rmdir system call.
1456  */
1457 static int
1458 ext2_rmdir(struct vop_rmdir_args *ap)
1459 {
1460 	struct vnode *vp = ap->a_vp;
1461 	struct vnode *dvp = ap->a_dvp;
1462 	struct componentname *cnp = ap->a_cnp;
1463 	struct inode *ip, *dp;
1464 	int error;
1465 
1466 	ip = VTOI(vp);
1467 	dp = VTOI(dvp);
1468 
1469 	/*
1470 	 * Verify the directory is empty (and valid).
1471 	 * (Rmdir ".." won't be valid since
1472 	 *  ".." will contain a reference to
1473 	 *  the current directory and thus be
1474 	 *  non-empty.)
1475 	 */
1476 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1477 		error = ENOTEMPTY;
1478 		goto out;
1479 	}
1480 	if ((dp->i_flags & APPEND)
1481 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1482 		error = EPERM;
1483 		goto out;
1484 	}
1485 	/*
1486 	 * Delete reference to directory before purging
1487 	 * inode.  If we crash in between, the directory
1488 	 * will be reattached to lost+found,
1489 	 */
1490 	error = ext2_dirremove(dvp, cnp);
1491 	if (error)
1492 		goto out;
1493 	ext2_dec_nlink(dp);
1494 	dp->i_flag |= IN_CHANGE;
1495 	cache_purge(dvp);
1496 	VOP_UNLOCK(dvp, 0);
1497 	/*
1498 	 * Truncate inode.  The only stuff left
1499 	 * in the directory is "." and "..".
1500 	 */
1501 	ip->i_nlink = 0;
1502 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1503 	    cnp->cn_thread);
1504 	cache_purge(ITOV(ip));
1505 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1506 		VOP_UNLOCK(vp, 0);
1507 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1508 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1509 	}
1510 out:
1511 	return (error);
1512 }
1513 
1514 /*
1515  * symlink -- make a symbolic link
1516  */
1517 static int
1518 ext2_symlink(struct vop_symlink_args *ap)
1519 {
1520 	struct vnode *vp, **vpp = ap->a_vpp;
1521 	struct inode *ip;
1522 	int len, error;
1523 
1524 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1525 	    vpp, ap->a_cnp);
1526 	if (error)
1527 		return (error);
1528 	vp = *vpp;
1529 	len = strlen(ap->a_target);
1530 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1531 		ip = VTOI(vp);
1532 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1533 		ip->i_size = len;
1534 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1535 	} else
1536 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1537 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1538 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1539 	if (error)
1540 		vput(vp);
1541 	return (error);
1542 }
1543 
1544 /*
1545  * Return target name of a symbolic link
1546  */
1547 static int
1548 ext2_readlink(struct vop_readlink_args *ap)
1549 {
1550 	struct vnode *vp = ap->a_vp;
1551 	struct inode *ip = VTOI(vp);
1552 	int isize;
1553 
1554 	isize = ip->i_size;
1555 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1556 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1557 		return (0);
1558 	}
1559 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1560 }
1561 
1562 /*
1563  * Calculate the logical to physical mapping if not done already,
1564  * then call the device strategy routine.
1565  *
1566  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1567  * deadlock on memory.  See ext2_bmap() for details.
1568  */
1569 static int
1570 ext2_strategy(struct vop_strategy_args *ap)
1571 {
1572 	struct buf *bp = ap->a_bp;
1573 	struct vnode *vp = ap->a_vp;
1574 	struct bufobj *bo;
1575 	daddr_t blkno;
1576 	int error;
1577 
1578 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1579 		panic("ext2_strategy: spec");
1580 	if (bp->b_blkno == bp->b_lblkno) {
1581 
1582 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1583 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1584 		else
1585 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1586 
1587 		bp->b_blkno = blkno;
1588 		if (error) {
1589 			bp->b_error = error;
1590 			bp->b_ioflags |= BIO_ERROR;
1591 			bufdone(bp);
1592 			return (0);
1593 		}
1594 		if ((long)bp->b_blkno == -1)
1595 			vfs_bio_clrbuf(bp);
1596 	}
1597 	if ((long)bp->b_blkno == -1) {
1598 		bufdone(bp);
1599 		return (0);
1600 	}
1601 	bp->b_iooffset = dbtob(bp->b_blkno);
1602 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1603 	BO_STRATEGY(bo, bp);
1604 	return (0);
1605 }
1606 
1607 /*
1608  * Print out the contents of an inode.
1609  */
1610 static int
1611 ext2_print(struct vop_print_args *ap)
1612 {
1613 	struct vnode *vp = ap->a_vp;
1614 	struct inode *ip = VTOI(vp);
1615 
1616 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1617 	if (vp->v_type == VFIFO)
1618 		fifo_printinfo(vp);
1619 	printf("\n");
1620 	return (0);
1621 }
1622 
1623 /*
1624  * Close wrapper for fifos.
1625  *
1626  * Update the times on the inode then do device close.
1627  */
1628 static int
1629 ext2fifo_close(struct vop_close_args *ap)
1630 {
1631 	struct vnode *vp = ap->a_vp;
1632 
1633 	VI_LOCK(vp);
1634 	if (vp->v_usecount > 1)
1635 		ext2_itimes_locked(vp);
1636 	VI_UNLOCK(vp);
1637 	return (fifo_specops.vop_close(ap));
1638 }
1639 
1640 /*
1641  * Kqfilter wrapper for fifos.
1642  *
1643  * Fall through to ext2 kqfilter routines if needed
1644  */
1645 static int
1646 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1647 {
1648 	int error;
1649 
1650 	error = fifo_specops.vop_kqfilter(ap);
1651 	if (error)
1652 		error = vfs_kqfilter(ap);
1653 	return (error);
1654 }
1655 
1656 /*
1657  * Return POSIX pathconf information applicable to ext2 filesystems.
1658  */
1659 static int
1660 ext2_pathconf(struct vop_pathconf_args *ap)
1661 {
1662 	int error = 0;
1663 
1664 	switch (ap->a_name) {
1665 	case _PC_LINK_MAX:
1666 		if (ext2_htree_has_idx(VTOI(ap->a_vp)))
1667 			*ap->a_retval = INT_MAX;
1668 		else
1669 			*ap->a_retval = ext2_max_nlink(VTOI(ap->a_vp));
1670 		break;
1671 	case _PC_NAME_MAX:
1672 		*ap->a_retval = NAME_MAX;
1673 		break;
1674 	case _PC_PIPE_BUF:
1675 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1676 			*ap->a_retval = PIPE_BUF;
1677 		else
1678 			error = EINVAL;
1679 		break;
1680 	case _PC_CHOWN_RESTRICTED:
1681 		*ap->a_retval = 1;
1682 		break;
1683 	case _PC_NO_TRUNC:
1684 		*ap->a_retval = 1;
1685 		break;
1686 
1687 #ifdef UFS_ACL
1688 	case _PC_ACL_EXTENDED:
1689 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1690 			*ap->a_retval = 1;
1691 		else
1692 			*ap->a_retval = 0;
1693 		break;
1694 	case _PC_ACL_PATH_MAX:
1695 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1696 			*ap->a_retval = ACL_MAX_ENTRIES;
1697 		else
1698 			*ap->a_retval = 3;
1699 		break;
1700 #endif /* UFS_ACL */
1701 
1702 	case _PC_MIN_HOLE_SIZE:
1703 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1704 		break;
1705 	case _PC_PRIO_IO:
1706 		*ap->a_retval = 0;
1707 		break;
1708 	case _PC_SYNC_IO:
1709 		*ap->a_retval = 0;
1710 		break;
1711 	case _PC_ALLOC_SIZE_MIN:
1712 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1713 		break;
1714 	case _PC_FILESIZEBITS:
1715 		*ap->a_retval = 64;
1716 		break;
1717 	case _PC_REC_INCR_XFER_SIZE:
1718 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1719 		break;
1720 	case _PC_REC_MAX_XFER_SIZE:
1721 		*ap->a_retval = -1;	/* means ``unlimited'' */
1722 		break;
1723 	case _PC_REC_MIN_XFER_SIZE:
1724 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1725 		break;
1726 	case _PC_REC_XFER_ALIGN:
1727 		*ap->a_retval = PAGE_SIZE;
1728 		break;
1729 	case _PC_SYMLINK_MAX:
1730 		*ap->a_retval = MAXPATHLEN;
1731 		break;
1732 
1733 	default:
1734 		error = vop_stdpathconf(ap);
1735 		break;
1736 	}
1737 	return (error);
1738 }
1739 
1740 /*
1741  * Vnode operation to remove a named attribute.
1742  */
1743 static int
1744 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1745 {
1746 	struct inode *ip;
1747 	struct m_ext2fs *fs;
1748 	int error;
1749 
1750 	ip = VTOI(ap->a_vp);
1751 	fs = ip->i_e2fs;
1752 
1753 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1754 		return (EOPNOTSUPP);
1755 
1756 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1757 		return (EOPNOTSUPP);
1758 
1759 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1760 	    ap->a_cred, ap->a_td, VWRITE);
1761 	if (error)
1762 		return (error);
1763 
1764 	error = ENOATTR;
1765 
1766 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1767 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1768 		if (error != ENOATTR)
1769 			return (error);
1770 	}
1771 
1772 	if (ip->i_facl)
1773 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1774 
1775 	return (error);
1776 }
1777 
1778 /*
1779  * Vnode operation to retrieve a named extended attribute.
1780  */
1781 static int
1782 ext2_getextattr(struct vop_getextattr_args *ap)
1783 {
1784 	struct inode *ip;
1785 	struct m_ext2fs *fs;
1786 	int error;
1787 
1788 	ip = VTOI(ap->a_vp);
1789 	fs = ip->i_e2fs;
1790 
1791 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1792 		return (EOPNOTSUPP);
1793 
1794 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1795 		return (EOPNOTSUPP);
1796 
1797 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1798 	    ap->a_cred, ap->a_td, VREAD);
1799 	if (error)
1800 		return (error);
1801 
1802 	if (ap->a_size != NULL)
1803 		*ap->a_size = 0;
1804 
1805 	error = ENOATTR;
1806 
1807 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1808 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1809 		    ap->a_name, ap->a_uio, ap->a_size);
1810 		if (error != ENOATTR)
1811 			return (error);
1812 	}
1813 
1814 	if (ip->i_facl)
1815 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1816 		    ap->a_name, ap->a_uio, ap->a_size);
1817 
1818 	return (error);
1819 }
1820 
1821 /*
1822  * Vnode operation to retrieve extended attributes on a vnode.
1823  */
1824 static int
1825 ext2_listextattr(struct vop_listextattr_args *ap)
1826 {
1827 	struct inode *ip;
1828 	struct m_ext2fs *fs;
1829 	int error;
1830 
1831 	ip = VTOI(ap->a_vp);
1832 	fs = ip->i_e2fs;
1833 
1834 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1835 		return (EOPNOTSUPP);
1836 
1837 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1838 		return (EOPNOTSUPP);
1839 
1840 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1841 	    ap->a_cred, ap->a_td, VREAD);
1842 	if (error)
1843 		return (error);
1844 
1845 	if (ap->a_size != NULL)
1846 		*ap->a_size = 0;
1847 
1848 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1849 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1850 		    ap->a_uio, ap->a_size);
1851 		if (error)
1852 			return (error);
1853 	}
1854 
1855 	if (ip->i_facl)
1856 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1857 		    ap->a_uio, ap->a_size);
1858 
1859 	return (error);
1860 }
1861 
1862 /*
1863  * Vnode operation to set a named attribute.
1864  */
1865 static int
1866 ext2_setextattr(struct vop_setextattr_args *ap)
1867 {
1868 	struct inode *ip;
1869 	struct m_ext2fs *fs;
1870 	int error;
1871 
1872 	ip = VTOI(ap->a_vp);
1873 	fs = ip->i_e2fs;
1874 
1875 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1876 		return (EOPNOTSUPP);
1877 
1878 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1879 		return (EOPNOTSUPP);
1880 
1881 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1882 	    ap->a_cred, ap->a_td, VWRITE);
1883 	if (error)
1884 		return (error);
1885 
1886 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1887 	if (error)
1888 		return (error);
1889 
1890 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1891 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1892 		    ap->a_name, ap->a_uio);
1893 		if (error != ENOSPC)
1894 			return (error);
1895 	}
1896 
1897 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1898 	    ap->a_name, ap->a_uio);
1899 
1900 	return (error);
1901 }
1902 
1903 /*
1904  * Vnode pointer to File handle
1905  */
1906 /* ARGSUSED */
1907 static int
1908 ext2_vptofh(struct vop_vptofh_args *ap)
1909 {
1910 	struct inode *ip;
1911 	struct ufid *ufhp;
1912 
1913 	ip = VTOI(ap->a_vp);
1914 	ufhp = (struct ufid *)ap->a_fhp;
1915 	ufhp->ufid_len = sizeof(struct ufid);
1916 	ufhp->ufid_ino = ip->i_number;
1917 	ufhp->ufid_gen = ip->i_gen;
1918 	return (0);
1919 }
1920 
1921 /*
1922  * Initialize the vnode associated with a new inode, handle aliased
1923  * vnodes.
1924  */
1925 int
1926 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1927 {
1928 	struct inode *ip;
1929 	struct vnode *vp;
1930 
1931 	vp = *vpp;
1932 	ip = VTOI(vp);
1933 	vp->v_type = IFTOVT(ip->i_mode);
1934 	if (vp->v_type == VFIFO)
1935 		vp->v_op = fifoops;
1936 
1937 	if (ip->i_number == EXT2_ROOTINO)
1938 		vp->v_vflag |= VV_ROOT;
1939 	ip->i_modrev = init_va_filerev();
1940 	*vpp = vp;
1941 	return (0);
1942 }
1943 
1944 /*
1945  * Allocate a new inode.
1946  */
1947 static int
1948 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1949     struct componentname *cnp)
1950 {
1951 	struct inode *ip, *pdir;
1952 	struct vnode *tvp;
1953 	int error;
1954 
1955 	pdir = VTOI(dvp);
1956 #ifdef INVARIANTS
1957 	if ((cnp->cn_flags & HASBUF) == 0)
1958 		panic("ext2_makeinode: no name");
1959 #endif
1960 	*vpp = NULL;
1961 	if ((mode & IFMT) == 0)
1962 		mode |= IFREG;
1963 
1964 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1965 	if (error) {
1966 		return (error);
1967 	}
1968 	ip = VTOI(tvp);
1969 	ip->i_gid = pdir->i_gid;
1970 #ifdef SUIDDIR
1971 	{
1972 		/*
1973 		 * if we are
1974 		 * not the owner of the directory,
1975 		 * and we are hacking owners here, (only do this where told to)
1976 		 * and we are not giving it TOO root, (would subvert quotas)
1977 		 * then go ahead and give it to the other user.
1978 		 * Note that this drops off the execute bits for security.
1979 		 */
1980 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1981 		    (pdir->i_mode & ISUID) &&
1982 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1983 			ip->i_uid = pdir->i_uid;
1984 			mode &= ~07111;
1985 		} else {
1986 			ip->i_uid = cnp->cn_cred->cr_uid;
1987 		}
1988 	}
1989 #else
1990 	ip->i_uid = cnp->cn_cred->cr_uid;
1991 #endif
1992 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1993 	ip->i_mode = mode;
1994 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1995 	ip->i_nlink = 1;
1996 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1997 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1998 			ip->i_mode &= ~ISGID;
1999 	}
2000 
2001 	if (cnp->cn_flags & ISWHITEOUT)
2002 		ip->i_flags |= UF_OPAQUE;
2003 
2004 	/*
2005 	 * Make sure inode goes to disk before directory entry.
2006 	 */
2007 	error = ext2_update(tvp, !DOINGASYNC(tvp));
2008 	if (error)
2009 		goto bad;
2010 
2011 #ifdef UFS_ACL
2012 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2013 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2014 		    cnp->cn_cred, cnp->cn_thread);
2015 		if (error)
2016 			goto bad;
2017 	}
2018 #endif /* UFS_ACL */
2019 
2020 	error = ext2_direnter(ip, dvp, cnp);
2021 	if (error)
2022 		goto bad;
2023 
2024 	*vpp = tvp;
2025 	return (0);
2026 
2027 bad:
2028 	/*
2029 	 * Write error occurred trying to update the inode
2030 	 * or the directory so must deallocate the inode.
2031 	 */
2032 	ip->i_nlink = 0;
2033 	ip->i_flag |= IN_CHANGE;
2034 	vput(tvp);
2035 	return (error);
2036 }
2037 
2038 /*
2039  * Vnode op for reading.
2040  */
2041 static int
2042 ext2_read(struct vop_read_args *ap)
2043 {
2044 	struct vnode *vp;
2045 	struct inode *ip;
2046 	struct uio *uio;
2047 	struct m_ext2fs *fs;
2048 	struct buf *bp;
2049 	daddr_t lbn, nextlbn;
2050 	off_t bytesinfile;
2051 	long size, xfersize, blkoffset;
2052 	int error, orig_resid, seqcount;
2053 	int ioflag;
2054 
2055 	vp = ap->a_vp;
2056 	uio = ap->a_uio;
2057 	ioflag = ap->a_ioflag;
2058 
2059 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2060 	ip = VTOI(vp);
2061 
2062 #ifdef INVARIANTS
2063 	if (uio->uio_rw != UIO_READ)
2064 		panic("%s: mode", "ext2_read");
2065 
2066 	if (vp->v_type == VLNK) {
2067 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
2068 			panic("%s: short symlink", "ext2_read");
2069 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2070 		panic("%s: type %d", "ext2_read", vp->v_type);
2071 #endif
2072 	orig_resid = uio->uio_resid;
2073 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2074 	if (orig_resid == 0)
2075 		return (0);
2076 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2077 	fs = ip->i_e2fs;
2078 	if (uio->uio_offset < ip->i_size &&
2079 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2080 		return (EOVERFLOW);
2081 
2082 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2083 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2084 			break;
2085 		lbn = lblkno(fs, uio->uio_offset);
2086 		nextlbn = lbn + 1;
2087 		size = blksize(fs, ip, lbn);
2088 		blkoffset = blkoff(fs, uio->uio_offset);
2089 
2090 		xfersize = fs->e2fs_fsize - blkoffset;
2091 		if (uio->uio_resid < xfersize)
2092 			xfersize = uio->uio_resid;
2093 		if (bytesinfile < xfersize)
2094 			xfersize = bytesinfile;
2095 
2096 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2097 			error = bread(vp, lbn, size, NOCRED, &bp);
2098 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2099 			error = cluster_read(vp, ip->i_size, lbn, size,
2100 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2101 			    0, &bp);
2102 		} else if (seqcount > 1) {
2103 			u_int nextsize = blksize(fs, ip, nextlbn);
2104 
2105 			error = breadn(vp, lbn,
2106 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2107 		} else
2108 			error = bread(vp, lbn, size, NOCRED, &bp);
2109 		if (error) {
2110 			brelse(bp);
2111 			bp = NULL;
2112 			break;
2113 		}
2114 
2115 		/*
2116 		 * We should only get non-zero b_resid when an I/O error
2117 		 * has occurred, which should cause us to break above.
2118 		 * However, if the short read did not cause an error,
2119 		 * then we want to ensure that we do not uiomove bad
2120 		 * or uninitialized data.
2121 		 */
2122 		size -= bp->b_resid;
2123 		if (size < xfersize) {
2124 			if (size == 0)
2125 				break;
2126 			xfersize = size;
2127 		}
2128 		error = uiomove((char *)bp->b_data + blkoffset,
2129 		    (int)xfersize, uio);
2130 		if (error)
2131 			break;
2132 		vfs_bio_brelse(bp, ioflag);
2133 	}
2134 
2135 	/*
2136 	 * This can only happen in the case of an error because the loop
2137 	 * above resets bp to NULL on each iteration and on normal
2138 	 * completion has not set a new value into it. so it must have come
2139 	 * from a 'break' statement
2140 	 */
2141 	if (bp != NULL)
2142 		vfs_bio_brelse(bp, ioflag);
2143 
2144 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2145 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2146 		ip->i_flag |= IN_ACCESS;
2147 	return (error);
2148 }
2149 
2150 static int
2151 ext2_ioctl(struct vop_ioctl_args *ap)
2152 {
2153 
2154 	switch (ap->a_command) {
2155 	case FIOSEEKDATA:
2156 	case FIOSEEKHOLE:
2157 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
2158 		    (off_t *)ap->a_data, ap->a_cred));
2159 	default:
2160 		return (ENOTTY);
2161 	}
2162 }
2163 
2164 /*
2165  * Vnode op for writing.
2166  */
2167 static int
2168 ext2_write(struct vop_write_args *ap)
2169 {
2170 	struct vnode *vp;
2171 	struct uio *uio;
2172 	struct inode *ip;
2173 	struct m_ext2fs *fs;
2174 	struct buf *bp;
2175 	daddr_t lbn;
2176 	off_t osize;
2177 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2178 
2179 	ioflag = ap->a_ioflag;
2180 	uio = ap->a_uio;
2181 	vp = ap->a_vp;
2182 
2183 	seqcount = ioflag >> IO_SEQSHIFT;
2184 	ip = VTOI(vp);
2185 
2186 #ifdef INVARIANTS
2187 	if (uio->uio_rw != UIO_WRITE)
2188 		panic("%s: mode", "ext2_write");
2189 #endif
2190 
2191 	switch (vp->v_type) {
2192 	case VREG:
2193 		if (ioflag & IO_APPEND)
2194 			uio->uio_offset = ip->i_size;
2195 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2196 			return (EPERM);
2197 		/* FALLTHROUGH */
2198 	case VLNK:
2199 		break;
2200 	case VDIR:
2201 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2202 		if ((ioflag & IO_SYNC) == 0)
2203 			panic("ext2_write: nonsync dir write");
2204 		break;
2205 	default:
2206 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2207 		    vp->v_type, (intmax_t)uio->uio_offset,
2208 		    (intmax_t)uio->uio_resid);
2209 	}
2210 
2211 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2212 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2213 	fs = ip->i_e2fs;
2214 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2215 		return (EFBIG);
2216 	/*
2217 	 * Maybe this should be above the vnode op call, but so long as
2218 	 * file servers have no limits, I don't think it matters.
2219 	 */
2220 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2221 		return (EFBIG);
2222 
2223 	resid = uio->uio_resid;
2224 	osize = ip->i_size;
2225 	if (seqcount > BA_SEQMAX)
2226 		flags = BA_SEQMAX << BA_SEQSHIFT;
2227 	else
2228 		flags = seqcount << BA_SEQSHIFT;
2229 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2230 		flags |= IO_SYNC;
2231 
2232 	for (error = 0; uio->uio_resid > 0;) {
2233 		lbn = lblkno(fs, uio->uio_offset);
2234 		blkoffset = blkoff(fs, uio->uio_offset);
2235 		xfersize = fs->e2fs_fsize - blkoffset;
2236 		if (uio->uio_resid < xfersize)
2237 			xfersize = uio->uio_resid;
2238 		if (uio->uio_offset + xfersize > ip->i_size)
2239 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2240 
2241 		/*
2242 		 * We must perform a read-before-write if the transfer size
2243 		 * does not cover the entire buffer.
2244 		 */
2245 		if (fs->e2fs_bsize > xfersize)
2246 			flags |= BA_CLRBUF;
2247 		else
2248 			flags &= ~BA_CLRBUF;
2249 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2250 		    ap->a_cred, &bp, flags);
2251 		if (error != 0)
2252 			break;
2253 
2254 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2255 			bp->b_flags |= B_NOCACHE;
2256 		if (uio->uio_offset + xfersize > ip->i_size)
2257 			ip->i_size = uio->uio_offset + xfersize;
2258 		size = blksize(fs, ip, lbn) - bp->b_resid;
2259 		if (size < xfersize)
2260 			xfersize = size;
2261 
2262 		error =
2263 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2264 		/*
2265 		 * If the buffer is not already filled and we encounter an
2266 		 * error while trying to fill it, we have to clear out any
2267 		 * garbage data from the pages instantiated for the buffer.
2268 		 * If we do not, a failed uiomove() during a write can leave
2269 		 * the prior contents of the pages exposed to a userland mmap.
2270 		 *
2271 		 * Note that we need only clear buffers with a transfer size
2272 		 * equal to the block size because buffers with a shorter
2273 		 * transfer size were cleared above by the call to ext2_balloc()
2274 		 * with the BA_CLRBUF flag set.
2275 		 *
2276 		 * If the source region for uiomove identically mmaps the
2277 		 * buffer, uiomove() performed the NOP copy, and the buffer
2278 		 * content remains valid because the page fault handler
2279 		 * validated the pages.
2280 		 */
2281 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2282 		    fs->e2fs_bsize == xfersize)
2283 			vfs_bio_clrbuf(bp);
2284 
2285 		vfs_bio_set_flags(bp, ioflag);
2286 
2287 		/*
2288 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2289 		 * if we have a severe page deficiency write the buffer
2290 		 * asynchronously.  Otherwise try to cluster, and if that
2291 		 * doesn't do it then either do an async write (if O_DIRECT),
2292 		 * or a delayed write (if not).
2293 		 */
2294 		if (ioflag & IO_SYNC) {
2295 			(void)bwrite(bp);
2296 		} else if (vm_page_count_severe() ||
2297 			    buf_dirty_count_severe() ||
2298 		    (ioflag & IO_ASYNC)) {
2299 			bp->b_flags |= B_CLUSTEROK;
2300 			bawrite(bp);
2301 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2302 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2303 				bp->b_flags |= B_CLUSTEROK;
2304 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2305 			} else {
2306 				bawrite(bp);
2307 			}
2308 		} else if (ioflag & IO_DIRECT) {
2309 			bp->b_flags |= B_CLUSTEROK;
2310 			bawrite(bp);
2311 		} else {
2312 			bp->b_flags |= B_CLUSTEROK;
2313 			bdwrite(bp);
2314 		}
2315 		if (error || xfersize == 0)
2316 			break;
2317 	}
2318 	/*
2319 	 * If we successfully wrote any data, and we are not the superuser
2320 	 * we clear the setuid and setgid bits as a precaution against
2321 	 * tampering.
2322 	 */
2323 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2324 	    ap->a_cred) {
2325 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2326 			ip->i_mode &= ~(ISUID | ISGID);
2327 	}
2328 	if (error) {
2329 		if (ioflag & IO_UNIT) {
2330 			(void)ext2_truncate(vp, osize,
2331 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2332 			uio->uio_offset -= resid - uio->uio_resid;
2333 			uio->uio_resid = resid;
2334 		}
2335 	}
2336 	if (uio->uio_resid != resid) {
2337 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2338 		if (ioflag & IO_SYNC)
2339 			error = ext2_update(vp, 1);
2340 	}
2341 	return (error);
2342 }
2343