xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
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/limits.h>
55 #include <sys/sdt.h>
56 #include <sys/stat.h>
57 #include <sys/bio.h>
58 #include <sys/buf.h>
59 #include <sys/endian.h>
60 #include <sys/priv.h>
61 #include <sys/rwlock.h>
62 #include <sys/mount.h>
63 #include <sys/unistd.h>
64 #include <sys/time.h>
65 #include <sys/vnode.h>
66 #include <sys/namei.h>
67 #include <sys/lockf.h>
68 #include <sys/event.h>
69 #include <sys/conf.h>
70 #include <sys/file.h>
71 #include <sys/extattr.h>
72 #include <sys/vmmeter.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
81 
82 #include "opt_directio.h"
83 
84 #include <ufs/ufs/dir.h>
85 
86 #include <fs/ext2fs/fs.h>
87 #include <fs/ext2fs/inode.h>
88 #include <fs/ext2fs/ext2_acl.h>
89 #include <fs/ext2fs/ext2fs.h>
90 #include <fs/ext2fs/ext2_extern.h>
91 #include <fs/ext2fs/ext2_dinode.h>
92 #include <fs/ext2fs/ext2_dir.h>
93 #include <fs/ext2fs/ext2_mount.h>
94 #include <fs/ext2fs/ext2_extattr.h>
95 #include <fs/ext2fs/ext2_extents.h>
96 
97 SDT_PROVIDER_DECLARE(ext2fs);
98 /*
99  * ext2fs trace probe:
100  * arg0: verbosity. Higher numbers give more verbose messages
101  * arg1: Textual message
102  */
103 SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*");
104 
105 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
106 static void ext2_itimes_locked(struct vnode *);
107 
108 static vop_access_t	ext2_access;
109 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
110 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
111     struct thread *);
112 static vop_close_t	ext2_close;
113 static vop_create_t	ext2_create;
114 static vop_fsync_t	ext2_fsync;
115 static vop_getattr_t	ext2_getattr;
116 static vop_ioctl_t	ext2_ioctl;
117 static vop_link_t	ext2_link;
118 static vop_mkdir_t	ext2_mkdir;
119 static vop_mknod_t	ext2_mknod;
120 static vop_open_t	ext2_open;
121 static vop_pathconf_t	ext2_pathconf;
122 static vop_print_t	ext2_print;
123 static vop_read_t	ext2_read;
124 static vop_readlink_t	ext2_readlink;
125 static vop_remove_t	ext2_remove;
126 static vop_rename_t	ext2_rename;
127 static vop_rmdir_t	ext2_rmdir;
128 static vop_setattr_t	ext2_setattr;
129 static vop_strategy_t	ext2_strategy;
130 static vop_symlink_t	ext2_symlink;
131 static vop_write_t	ext2_write;
132 static vop_deleteextattr_t	ext2_deleteextattr;
133 static vop_getextattr_t	ext2_getextattr;
134 static vop_listextattr_t	ext2_listextattr;
135 static vop_setextattr_t	ext2_setextattr;
136 static vop_vptofh_t	ext2_vptofh;
137 static vop_close_t	ext2fifo_close;
138 
139 /* Global vfs data structures for ext2. */
140 struct vop_vector ext2_vnodeops = {
141 	.vop_default =		&default_vnodeops,
142 	.vop_access =		ext2_access,
143 	.vop_bmap =		ext2_bmap,
144 	.vop_cachedlookup =	ext2_lookup,
145 	.vop_close =		ext2_close,
146 	.vop_create =		ext2_create,
147 	.vop_fsync =		ext2_fsync,
148 	.vop_getpages =		vnode_pager_local_getpages,
149 	.vop_getpages_async =	vnode_pager_local_getpages_async,
150 	.vop_getattr =		ext2_getattr,
151 	.vop_inactive =		ext2_inactive,
152 	.vop_ioctl =		ext2_ioctl,
153 	.vop_link =		ext2_link,
154 	.vop_lookup =		vfs_cache_lookup,
155 	.vop_mkdir =		ext2_mkdir,
156 	.vop_mknod =		ext2_mknod,
157 	.vop_open =		ext2_open,
158 	.vop_pathconf =		ext2_pathconf,
159 	.vop_poll =		vop_stdpoll,
160 	.vop_print =		ext2_print,
161 	.vop_read =		ext2_read,
162 	.vop_readdir =		ext2_readdir,
163 	.vop_readlink =		ext2_readlink,
164 	.vop_reallocblks =	ext2_reallocblks,
165 	.vop_reclaim =		ext2_reclaim,
166 	.vop_remove =		ext2_remove,
167 	.vop_rename =		ext2_rename,
168 	.vop_rmdir =		ext2_rmdir,
169 	.vop_setattr =		ext2_setattr,
170 	.vop_strategy =		ext2_strategy,
171 	.vop_symlink =		ext2_symlink,
172 	.vop_write =		ext2_write,
173 	.vop_deleteextattr =	ext2_deleteextattr,
174 	.vop_getextattr =	ext2_getextattr,
175 	.vop_listextattr =	ext2_listextattr,
176 	.vop_setextattr =	ext2_setextattr,
177 #ifdef UFS_ACL
178 	.vop_getacl =		ext2_getacl,
179 	.vop_setacl =		ext2_setacl,
180 	.vop_aclcheck =		ext2_aclcheck,
181 #endif /* UFS_ACL */
182 	.vop_vptofh =		ext2_vptofh,
183 };
184 VFS_VOP_VECTOR_REGISTER(ext2_vnodeops);
185 
186 struct vop_vector ext2_fifoops = {
187 	.vop_default =		&fifo_specops,
188 	.vop_access =		ext2_access,
189 	.vop_close =		ext2fifo_close,
190 	.vop_fsync =		ext2_fsync,
191 	.vop_getattr =		ext2_getattr,
192 	.vop_inactive =		ext2_inactive,
193 	.vop_pathconf =		ext2_pathconf,
194 	.vop_print =		ext2_print,
195 	.vop_read =		VOP_PANIC,
196 	.vop_reclaim =		ext2_reclaim,
197 	.vop_setattr =		ext2_setattr,
198 	.vop_write =		VOP_PANIC,
199 	.vop_vptofh =		ext2_vptofh,
200 };
201 VFS_VOP_VECTOR_REGISTER(ext2_fifoops);
202 
203 /*
204  * A virgin directory (no blushing please).
205  * Note that the type and namlen fields are reversed relative to ext2.
206  * Also, we don't use `struct odirtemplate', since it would just cause
207  * endianness problems.
208  */
209 static struct dirtemplate mastertemplate = {
210 	0, htole16(12), 1, EXT2_FT_DIR, ".",
211 	0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_DIR, ".."
212 };
213 static struct dirtemplate omastertemplate = {
214 	0, htole16(12), 1, EXT2_FT_UNKNOWN, ".",
215 	0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_UNKNOWN, ".."
216 };
217 
218 static void
219 ext2_itimes_locked(struct vnode *vp)
220 {
221 	struct inode *ip;
222 	struct timespec ts;
223 
224 	ASSERT_VI_LOCKED(vp, __func__);
225 
226 	ip = VTOI(vp);
227 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
228 		return;
229 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
230 		ip->i_flag |= IN_LAZYMOD;
231 	else
232 		ip->i_flag |= IN_MODIFIED;
233 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
234 		vfs_timestamp(&ts);
235 		if (ip->i_flag & IN_ACCESS) {
236 			ip->i_atime = ts.tv_sec;
237 			ip->i_atimensec = ts.tv_nsec;
238 		}
239 		if (ip->i_flag & IN_UPDATE) {
240 			ip->i_mtime = ts.tv_sec;
241 			ip->i_mtimensec = ts.tv_nsec;
242 			ip->i_modrev++;
243 		}
244 		if (ip->i_flag & IN_CHANGE) {
245 			ip->i_ctime = ts.tv_sec;
246 			ip->i_ctimensec = ts.tv_nsec;
247 		}
248 	}
249 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
250 }
251 
252 void
253 ext2_itimes(struct vnode *vp)
254 {
255 
256 	VI_LOCK(vp);
257 	ext2_itimes_locked(vp);
258 	VI_UNLOCK(vp);
259 }
260 
261 /*
262  * Create a regular file
263  */
264 static int
265 ext2_create(struct vop_create_args *ap)
266 {
267 	int error;
268 
269 	error =
270 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
271 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
272 	if (error != 0)
273 		return (error);
274 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
275 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
276 	return (0);
277 }
278 
279 static int
280 ext2_open(struct vop_open_args *ap)
281 {
282 
283 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
284 		return (EOPNOTSUPP);
285 
286 	/*
287 	 * Files marked append-only must be opened for appending.
288 	 */
289 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
290 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
291 		return (EPERM);
292 
293 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
294 
295 	return (0);
296 }
297 
298 /*
299  * Close called.
300  *
301  * Update the times on the inode.
302  */
303 static int
304 ext2_close(struct vop_close_args *ap)
305 {
306 	struct vnode *vp = ap->a_vp;
307 
308 	VI_LOCK(vp);
309 	if (vp->v_usecount > 1)
310 		ext2_itimes_locked(vp);
311 	VI_UNLOCK(vp);
312 	return (0);
313 }
314 
315 static int
316 ext2_access(struct vop_access_args *ap)
317 {
318 	struct vnode *vp = ap->a_vp;
319 	struct inode *ip = VTOI(vp);
320 	accmode_t accmode = ap->a_accmode;
321 	int error;
322 
323 	/*
324 	 * Disallow write attempts on read-only file systems;
325 	 * unless the file is a socket, fifo, or a block or
326 	 * character device resident on the file system.
327 	 */
328 	if (accmode & VWRITE) {
329 		switch (vp->v_type) {
330 		case VDIR:
331 		case VLNK:
332 		case VREG:
333 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
334 				return (EROFS);
335 			break;
336 		default:
337 			break;
338 		}
339 	}
340 
341 	/* If immutable bit set, nobody gets to write it. */
342 	if ((accmode & VWRITE) && (ip->i_flags & SF_IMMUTABLE))
343 		return (EPERM);
344 
345 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
346 	    ap->a_accmode, ap->a_cred);
347 	return (error);
348 }
349 
350 static int
351 ext2_getattr(struct vop_getattr_args *ap)
352 {
353 	struct vnode *vp = ap->a_vp;
354 	struct inode *ip = VTOI(vp);
355 	struct vattr *vap = ap->a_vap;
356 
357 	ext2_itimes(vp);
358 	/*
359 	 * Copy from inode table
360 	 */
361 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
362 	vap->va_fileid = ip->i_number;
363 	vap->va_mode = ip->i_mode & ~IFMT;
364 	vap->va_nlink = ip->i_nlink;
365 	vap->va_uid = ip->i_uid;
366 	vap->va_gid = ip->i_gid;
367 	vap->va_rdev = ip->i_rdev;
368 	vap->va_size = ip->i_size;
369 	vap->va_atime.tv_sec = ip->i_atime;
370 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
371 	vap->va_mtime.tv_sec = ip->i_mtime;
372 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
373 	vap->va_ctime.tv_sec = ip->i_ctime;
374 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
375 	if (E2DI_HAS_XTIME(ip)) {
376 		vap->va_birthtime.tv_sec = ip->i_birthtime;
377 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
378 	}
379 	vap->va_flags = ip->i_flags;
380 	vap->va_gen = ip->i_gen;
381 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
382 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
383 	vap->va_type = IFTOVT(ip->i_mode);
384 	vap->va_filerev = ip->i_modrev;
385 	return (0);
386 }
387 
388 /*
389  * Set attribute vnode op. called from several syscalls
390  */
391 static int
392 ext2_setattr(struct vop_setattr_args *ap)
393 {
394 	struct vattr *vap = ap->a_vap;
395 	struct vnode *vp = ap->a_vp;
396 	struct inode *ip = VTOI(vp);
397 	struct ucred *cred = ap->a_cred;
398 	struct thread *td = curthread;
399 	int error;
400 
401 	/*
402 	 * Check for unsettable attributes.
403 	 */
404 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
405 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
406 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
407 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
408 		return (EINVAL);
409 	}
410 	if (vap->va_flags != VNOVAL) {
411 		/* Disallow flags not supported by ext2fs. */
412 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
413 			return (EOPNOTSUPP);
414 
415 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
416 			return (EROFS);
417 		/*
418 		 * Callers may only modify the file flags on objects they
419 		 * have VADMIN rights for.
420 		 */
421 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
422 			return (error);
423 		/*
424 		 * Unprivileged processes and privileged processes in
425 		 * jail() are not permitted to unset system flags, or
426 		 * modify flags if any system flags are set.
427 		 * Privileged non-jail processes may not modify system flags
428 		 * if securelevel > 0 and any existing system flags are set.
429 		 */
430 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
431 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
432 				error = securelevel_gt(cred, 0);
433 				if (error)
434 					return (error);
435 			}
436 		} else {
437 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
438 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
439 				return (EPERM);
440 		}
441 		ip->i_flags = vap->va_flags;
442 		ip->i_flag |= IN_CHANGE;
443 		if (ip->i_flags & (IMMUTABLE | APPEND))
444 			return (0);
445 	}
446 	if (ip->i_flags & (IMMUTABLE | APPEND))
447 		return (EPERM);
448 	/*
449 	 * Go through the fields and update iff not VNOVAL.
450 	 */
451 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
452 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
453 			return (EROFS);
454 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
455 		    td)) != 0)
456 			return (error);
457 	}
458 	if (vap->va_size != VNOVAL) {
459 		/*
460 		 * Disallow write attempts on read-only file systems;
461 		 * unless the file is a socket, fifo, or a block or
462 		 * character device resident on the file system.
463 		 */
464 		switch (vp->v_type) {
465 		case VDIR:
466 			return (EISDIR);
467 		case VLNK:
468 		case VREG:
469 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
470 				return (EROFS);
471 			break;
472 		default:
473 			break;
474 		}
475 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
476 			return (error);
477 	}
478 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
479 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
480 			return (EROFS);
481 		/*
482 		 * From utimes(2):
483 		 * If times is NULL, ... The caller must be the owner of
484 		 * the file, have permission to write the file, or be the
485 		 * super-user.
486 		 * If times is non-NULL, ... The caller must be the owner of
487 		 * the file or be the super-user.
488 		 */
489 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
490 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
491 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
492 			return (error);
493 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
494 		if (vap->va_atime.tv_sec != VNOVAL) {
495 			ip->i_flag &= ~IN_ACCESS;
496 			ip->i_atime = vap->va_atime.tv_sec;
497 			ip->i_atimensec = vap->va_atime.tv_nsec;
498 		}
499 		if (vap->va_mtime.tv_sec != VNOVAL) {
500 			ip->i_flag &= ~IN_UPDATE;
501 			ip->i_mtime = vap->va_mtime.tv_sec;
502 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
503 		}
504 		if (E2DI_HAS_XTIME(ip) && vap->va_birthtime.tv_sec != VNOVAL) {
505 			ip->i_birthtime = vap->va_birthtime.tv_sec;
506 			ip->i_birthnsec = vap->va_birthtime.tv_nsec;
507 		}
508 		error = ext2_update(vp, 0);
509 		if (error)
510 			return (error);
511 	}
512 	error = 0;
513 	if (vap->va_mode != (mode_t)VNOVAL) {
514 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
515 			return (EROFS);
516 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
517 	}
518 	return (error);
519 }
520 
521 /*
522  * Change the mode on a file.
523  * Inode must be locked before calling.
524  */
525 static int
526 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
527 {
528 	struct inode *ip = VTOI(vp);
529 	int error;
530 
531 	/*
532 	 * To modify the permissions on a file, must possess VADMIN
533 	 * for that file.
534 	 */
535 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
536 		return (error);
537 	/*
538 	 * Privileged processes may set the sticky bit on non-directories,
539 	 * as well as set the setgid bit on a file with a group that the
540 	 * process is not a member of.
541 	 */
542 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
543 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE);
544 		if (error)
545 			return (EFTYPE);
546 	}
547 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
548 		error = priv_check_cred(cred, PRIV_VFS_SETGID);
549 		if (error)
550 			return (error);
551 	}
552 	ip->i_mode &= ~ALLPERMS;
553 	ip->i_mode |= (mode & ALLPERMS);
554 	ip->i_flag |= IN_CHANGE;
555 	return (0);
556 }
557 
558 /*
559  * Perform chown operation on inode ip;
560  * inode must be locked prior to call.
561  */
562 static int
563 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
564     struct thread *td)
565 {
566 	struct inode *ip = VTOI(vp);
567 	uid_t ouid;
568 	gid_t ogid;
569 	int error = 0;
570 
571 	if (uid == (uid_t)VNOVAL)
572 		uid = ip->i_uid;
573 	if (gid == (gid_t)VNOVAL)
574 		gid = ip->i_gid;
575 	/*
576 	 * To modify the ownership of a file, must possess VADMIN
577 	 * for that file.
578 	 */
579 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
580 		return (error);
581 	/*
582 	 * To change the owner of a file, or change the group of a file
583 	 * to a group of which we are not a member, the caller must
584 	 * have privilege.
585 	 */
586 	if (uid != ip->i_uid || (gid != ip->i_gid &&
587 	    !groupmember(gid, cred))) {
588 		error = priv_check_cred(cred, PRIV_VFS_CHOWN);
589 		if (error)
590 			return (error);
591 	}
592 	ogid = ip->i_gid;
593 	ouid = ip->i_uid;
594 	ip->i_gid = gid;
595 	ip->i_uid = uid;
596 	ip->i_flag |= IN_CHANGE;
597 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
598 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0)
599 			ip->i_mode &= ~(ISUID | ISGID);
600 	}
601 	return (0);
602 }
603 
604 /*
605  * Synch an open file.
606  */
607 /* ARGSUSED */
608 static int
609 ext2_fsync(struct vop_fsync_args *ap)
610 {
611 	/*
612 	 * Flush all dirty buffers associated with a vnode.
613 	 */
614 
615 	vop_stdfsync(ap);
616 
617 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
618 }
619 
620 static int
621 ext2_check_mknod_limits(dev_t dev)
622 {
623 	unsigned maj = major(dev);
624 	unsigned min = minor(dev);
625 
626 	if (maj > EXT2_MAJOR_MAX || min > EXT2_MINOR_MAX)
627 		return (EINVAL);
628 
629 	return (0);
630 }
631 
632 /*
633  * Mknod vnode call
634  */
635 /* ARGSUSED */
636 static int
637 ext2_mknod(struct vop_mknod_args *ap)
638 {
639 	struct vattr *vap = ap->a_vap;
640 	struct vnode **vpp = ap->a_vpp;
641 	struct inode *ip;
642 	ino_t ino;
643 	int error;
644 
645 	if (vap->va_rdev != VNOVAL) {
646 		error = ext2_check_mknod_limits(vap->va_rdev);
647 		if (error)
648 			return (error);
649 	}
650 
651 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
652 	    ap->a_dvp, vpp, ap->a_cnp);
653 	if (error)
654 		return (error);
655 	ip = VTOI(*vpp);
656 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
657 	if (vap->va_rdev != VNOVAL)
658 		ip->i_rdev = vap->va_rdev;
659 
660 	/*
661 	 * Remove inode, then reload it through VFS_VGET so it is
662 	 * checked to see if it is an alias of an existing entry in
663 	 * the inode cache.	 XXX I don't believe this is necessary now.
664 	 */
665 	(*vpp)->v_type = VNON;
666 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
667 	vgone(*vpp);
668 	vput(*vpp);
669 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
670 	if (error) {
671 		*vpp = NULL;
672 		return (error);
673 	}
674 	return (0);
675 }
676 
677 static int
678 ext2_remove(struct vop_remove_args *ap)
679 {
680 	struct inode *ip;
681 	struct vnode *vp = ap->a_vp;
682 	struct vnode *dvp = ap->a_dvp;
683 	int error;
684 
685 	ip = VTOI(vp);
686 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
687 	    (VTOI(dvp)->i_flags & APPEND)) {
688 		error = EPERM;
689 		goto out;
690 	}
691 	error = ext2_dirremove(dvp, ap->a_cnp);
692 	if (error == 0) {
693 		ip->i_nlink--;
694 		ip->i_flag |= IN_CHANGE;
695 	}
696 out:
697 	return (error);
698 }
699 
700 /*
701  * link vnode call
702  */
703 static int
704 ext2_link(struct vop_link_args *ap)
705 {
706 	struct vnode *vp = ap->a_vp;
707 	struct vnode *tdvp = ap->a_tdvp;
708 	struct componentname *cnp = ap->a_cnp;
709 	struct inode *ip;
710 	int error;
711 
712 	ip = VTOI(vp);
713 	if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) {
714 		error = EMLINK;
715 		goto out;
716 	}
717 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
718 		error = EPERM;
719 		goto out;
720 	}
721 	ip->i_nlink++;
722 	ip->i_flag |= IN_CHANGE;
723 	error = ext2_update(vp, !DOINGASYNC(vp));
724 	if (!error)
725 		error = ext2_direnter(ip, tdvp, cnp);
726 	if (error) {
727 		ip->i_nlink--;
728 		ip->i_flag |= IN_CHANGE;
729 	}
730 out:
731 	return (error);
732 }
733 
734 static int
735 ext2_inc_nlink(struct inode *ip)
736 {
737 
738 	ip->i_nlink++;
739 
740 	if (S_ISDIR(ip->i_mode) &&
741 	    EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) &&
742 	    ip->i_nlink > 1) {
743 		if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2)
744 			ip->i_nlink = 1;
745 	} else if (ip->i_nlink > EXT4_LINK_MAX) {
746 		ip->i_nlink--;
747 		return (EMLINK);
748 	}
749 
750 	return (0);
751 }
752 
753 static void
754 ext2_dec_nlink(struct inode *ip)
755 {
756 
757 	if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
758 		ip->i_nlink--;
759 }
760 
761 /*
762  * Rename system call.
763  * 	rename("foo", "bar");
764  * is essentially
765  *	unlink("bar");
766  *	link("foo", "bar");
767  *	unlink("foo");
768  * but ``atomically''.  Can't do full commit without saving state in the
769  * inode on disk which isn't feasible at this time.  Best we can do is
770  * always guarantee the target exists.
771  *
772  * Basic algorithm is:
773  *
774  * 1) Bump link count on source while we're linking it to the
775  *    target.  This also ensure the inode won't be deleted out
776  *    from underneath us while we work (it may be truncated by
777  *    a concurrent `trunc' or `open' for creation).
778  * 2) Link source to destination.  If destination already exists,
779  *    delete it first.
780  * 3) Unlink source reference to inode if still around. If a
781  *    directory was moved and the parent of the destination
782  *    is different from the source, patch the ".." entry in the
783  *    directory.
784  */
785 static int
786 ext2_rename(struct vop_rename_args *ap)
787 {
788 	struct vnode *tvp = ap->a_tvp;
789 	struct vnode *tdvp = ap->a_tdvp;
790 	struct vnode *fvp = ap->a_fvp;
791 	struct vnode *fdvp = ap->a_fdvp;
792 	struct componentname *tcnp = ap->a_tcnp;
793 	struct componentname *fcnp = ap->a_fcnp;
794 	struct inode *ip, *xp, *dp;
795 	struct dirtemplate *dirbuf;
796 	int doingdirectory = 0, oldparent = 0, newparent = 0;
797 	int error = 0;
798 	u_char namlen;
799 
800 	/*
801 	 * Check for cross-device rename.
802 	 */
803 	if ((fvp->v_mount != tdvp->v_mount) ||
804 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
805 		error = EXDEV;
806 abortit:
807 		if (tdvp == tvp)
808 			vrele(tdvp);
809 		else
810 			vput(tdvp);
811 		if (tvp)
812 			vput(tvp);
813 		vrele(fdvp);
814 		vrele(fvp);
815 		return (error);
816 	}
817 
818 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
819 	    (VTOI(tdvp)->i_flags & APPEND))) {
820 		error = EPERM;
821 		goto abortit;
822 	}
823 
824 	/*
825 	 * Renaming a file to itself has no effect.  The upper layers should
826 	 * not call us in that case.  Temporarily just warn if they do.
827 	 */
828 	if (fvp == tvp) {
829 		SDT_PROBE2(ext2fs, , vnops, trace, 1,
830 		    "rename: fvp == tvp (can't happen)");
831 		error = 0;
832 		goto abortit;
833 	}
834 
835 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
836 		goto abortit;
837 	dp = VTOI(fdvp);
838 	ip = VTOI(fvp);
839 	if (ip->i_nlink >= EXT4_LINK_MAX &&
840 	    !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
841 		VOP_UNLOCK(fvp);
842 		error = EMLINK;
843 		goto abortit;
844 	}
845 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
846 	    || (dp->i_flags & APPEND)) {
847 		VOP_UNLOCK(fvp);
848 		error = EPERM;
849 		goto abortit;
850 	}
851 	if ((ip->i_mode & IFMT) == IFDIR) {
852 		/*
853 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
854 		 */
855 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
856 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
857 		    (ip->i_flag & IN_RENAME)) {
858 			VOP_UNLOCK(fvp);
859 			error = EINVAL;
860 			goto abortit;
861 		}
862 		ip->i_flag |= IN_RENAME;
863 		oldparent = dp->i_number;
864 		doingdirectory++;
865 	}
866 	vrele(fdvp);
867 
868 	/*
869 	 * When the target exists, both the directory
870 	 * and target vnodes are returned locked.
871 	 */
872 	dp = VTOI(tdvp);
873 	xp = NULL;
874 	if (tvp)
875 		xp = VTOI(tvp);
876 
877 	/*
878 	 * 1) Bump link count while we're moving stuff
879 	 *    around.  If we crash somewhere before
880 	 *    completing our work, the link count
881 	 *    may be wrong, but correctable.
882 	 */
883 	ext2_inc_nlink(ip);
884 	ip->i_flag |= IN_CHANGE;
885 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
886 		VOP_UNLOCK(fvp);
887 		goto bad;
888 	}
889 
890 	/*
891 	 * If ".." must be changed (ie the directory gets a new
892 	 * parent) then the source directory must not be in the
893 	 * directory hierarchy above the target, as this would
894 	 * orphan everything below the source directory. Also
895 	 * the user must have write permission in the source so
896 	 * as to be able to change "..". We must repeat the call
897 	 * to namei, as the parent directory is unlocked by the
898 	 * call to checkpath().
899 	 */
900 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
901 	VOP_UNLOCK(fvp);
902 	if (oldparent != dp->i_number)
903 		newparent = dp->i_number;
904 	if (doingdirectory && newparent) {
905 		if (error)	/* write access check above */
906 			goto bad;
907 		if (xp != NULL)
908 			vput(tvp);
909 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
910 		if (error)
911 			goto out;
912 		VREF(tdvp);
913 		error = vfs_relookup(tdvp, &tvp, tcnp, true);
914 		if (error)
915 			goto out;
916 		vrele(tdvp);
917 		dp = VTOI(tdvp);
918 		xp = NULL;
919 		if (tvp)
920 			xp = VTOI(tvp);
921 	}
922 	/*
923 	 * 2) If target doesn't exist, link the target
924 	 *    to the source and unlink the source.
925 	 *    Otherwise, rewrite the target directory
926 	 *    entry to reference the source inode and
927 	 *    expunge the original entry's existence.
928 	 */
929 	if (xp == NULL) {
930 		if (dp->i_devvp != ip->i_devvp)
931 			panic("ext2_rename: EXDEV");
932 		/*
933 		 * Account for ".." in new directory.
934 		 * When source and destination have the same
935 		 * parent we don't fool with the link count.
936 		 */
937 		if (doingdirectory && newparent) {
938 			error = ext2_inc_nlink(dp);
939 			if (error)
940 				goto bad;
941 
942 			dp->i_flag |= IN_CHANGE;
943 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
944 			if (error)
945 				goto bad;
946 		}
947 		error = ext2_direnter(ip, tdvp, tcnp);
948 		if (error) {
949 			if (doingdirectory && newparent) {
950 				ext2_dec_nlink(dp);
951 				dp->i_flag |= IN_CHANGE;
952 				(void)ext2_update(tdvp, 1);
953 			}
954 			goto bad;
955 		}
956 		vput(tdvp);
957 	} else {
958 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
959 			panic("ext2_rename: EXDEV");
960 		/*
961 		 * Short circuit rename(foo, foo).
962 		 */
963 		if (xp->i_number == ip->i_number)
964 			panic("ext2_rename: same file");
965 		/*
966 		 * If the parent directory is "sticky", then the user must
967 		 * own the parent directory, or the destination of the rename,
968 		 * otherwise the destination may not be changed (except by
969 		 * root). This implements append-only directories.
970 		 */
971 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
972 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
973 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
974 			error = EPERM;
975 			goto bad;
976 		}
977 		/*
978 		 * Target must be empty if a directory and have no links
979 		 * to it. Also, ensure source and target are compatible
980 		 * (both directories, or both not directories).
981 		 */
982 		if ((xp->i_mode & IFMT) == IFDIR) {
983 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
984 				error = ENOTEMPTY;
985 				goto bad;
986 			}
987 			if (!doingdirectory) {
988 				error = ENOTDIR;
989 				goto bad;
990 			}
991 			cache_purge(tdvp);
992 		} else if (doingdirectory) {
993 			error = EISDIR;
994 			goto bad;
995 		}
996 		error = ext2_dirrewrite(dp, ip, tcnp);
997 		if (error)
998 			goto bad;
999 		/*
1000 		 * If the target directory is in the same
1001 		 * directory as the source directory,
1002 		 * decrement the link count on the parent
1003 		 * of the target directory.
1004 		 */
1005 		if (doingdirectory && !newparent) {
1006 			ext2_dec_nlink(dp);
1007 			dp->i_flag |= IN_CHANGE;
1008 		}
1009 		vput(tdvp);
1010 		/*
1011 		 * Adjust the link count of the target to
1012 		 * reflect the dirrewrite above.  If this is
1013 		 * a directory it is empty and there are
1014 		 * no links to it, so we can squash the inode and
1015 		 * any space associated with it.  We disallowed
1016 		 * renaming over top of a directory with links to
1017 		 * it above, as the remaining link would point to
1018 		 * a directory without "." or ".." entries.
1019 		 */
1020 		ext2_dec_nlink(xp);
1021 		if (doingdirectory) {
1022 			if (xp->i_nlink > 2)
1023 				panic("ext2_rename: linked directory");
1024 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1025 			    tcnp->cn_cred, curthread);
1026 			xp->i_nlink = 0;
1027 		}
1028 		xp->i_flag |= IN_CHANGE;
1029 		vput(tvp);
1030 		xp = NULL;
1031 	}
1032 
1033 	/*
1034 	 * 3) Unlink the source.
1035 	 */
1036 	fcnp->cn_flags &= ~MODMASK;
1037 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1038 	VREF(fdvp);
1039 	error = vfs_relookup(fdvp, &fvp, fcnp, true);
1040 	if (error == 0)
1041 		vrele(fdvp);
1042 	if (fvp != NULL) {
1043 		xp = VTOI(fvp);
1044 		dp = VTOI(fdvp);
1045 	} else {
1046 		/*
1047 		 * From name has disappeared.  IN_RENAME is not sufficient
1048 		 * to protect against directory races due to timing windows,
1049 		 * so we can't panic here.
1050 		 */
1051 		vrele(ap->a_fvp);
1052 		return (0);
1053 	}
1054 	/*
1055 	 * Ensure that the directory entry still exists and has not
1056 	 * changed while the new name has been entered. If the source is
1057 	 * a file then the entry may have been unlinked or renamed. In
1058 	 * either case there is no further work to be done. If the source
1059 	 * is a directory then it cannot have been rmdir'ed; its link
1060 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1061 	 * The IN_RENAME flag ensures that it cannot be moved by another
1062 	 * rename.
1063 	 */
1064 	if (xp != ip) {
1065 		/*
1066 		 * From name resolves to a different inode.  IN_RENAME is
1067 		 * not sufficient protection against timing window races
1068 		 * so we can't panic here.
1069 		 */
1070 	} else {
1071 		/*
1072 		 * If the source is a directory with a
1073 		 * new parent, the link count of the old
1074 		 * parent directory must be decremented
1075 		 * and ".." set to point to the new parent.
1076 		 */
1077 		if (doingdirectory && newparent) {
1078 			ext2_dec_nlink(dp);
1079 			dp->i_flag |= IN_CHANGE;
1080 			dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1081 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1082 			    ip->i_e2fs->e2fs_bsize, (off_t)0,
1083 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1084 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1085 			if (error == 0) {
1086 				/* Like ufs little-endian: */
1087 				namlen = dirbuf->dotdot_type;
1088 				if (namlen != 2 ||
1089 				    dirbuf->dotdot_name[0] != '.' ||
1090 				    dirbuf->dotdot_name[1] != '.') {
1091 					ext2_dirbad(xp, (doff_t)12,
1092 					    "rename: mangled dir");
1093 				} else {
1094 					dirbuf->dotdot_ino = htole32(newparent);
1095 					/*
1096 					 * dirblock 0 could be htree root,
1097 					 * try both csum update functions.
1098 					 */
1099 					ext2_dirent_csum_set(ip,
1100 					    (struct ext2fs_direct_2 *)dirbuf);
1101 					ext2_dx_csum_set(ip,
1102 					    (struct ext2fs_direct_2 *)dirbuf);
1103 					(void)vn_rdwr(UIO_WRITE, fvp,
1104 					    (caddr_t)dirbuf,
1105 					    ip->i_e2fs->e2fs_bsize,
1106 					    (off_t)0, UIO_SYSSPACE,
1107 					    IO_NODELOCKED | IO_SYNC |
1108 					    IO_NOMACCHECK, tcnp->cn_cred,
1109 					    NOCRED, NULL, NULL);
1110 					cache_purge(fdvp);
1111 				}
1112 			}
1113 			free(dirbuf, M_TEMP);
1114 		}
1115 		error = ext2_dirremove(fdvp, fcnp);
1116 		if (!error) {
1117 			ext2_dec_nlink(xp);
1118 			xp->i_flag |= IN_CHANGE;
1119 		}
1120 		xp->i_flag &= ~IN_RENAME;
1121 	}
1122 	if (dp)
1123 		vput(fdvp);
1124 	if (xp)
1125 		vput(fvp);
1126 	vrele(ap->a_fvp);
1127 	return (error);
1128 
1129 bad:
1130 	if (xp)
1131 		vput(ITOV(xp));
1132 	vput(ITOV(dp));
1133 out:
1134 	if (doingdirectory)
1135 		ip->i_flag &= ~IN_RENAME;
1136 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1137 		ext2_dec_nlink(ip);
1138 		ip->i_flag |= IN_CHANGE;
1139 		ip->i_flag &= ~IN_RENAME;
1140 		vput(fvp);
1141 	} else
1142 		vrele(fvp);
1143 	return (error);
1144 }
1145 
1146 #ifdef UFS_ACL
1147 static int
1148 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1149     mode_t dmode, struct ucred *cred, struct thread *td)
1150 {
1151 	int error;
1152 	struct inode *ip = VTOI(tvp);
1153 	struct acl *dacl, *acl;
1154 
1155 	acl = acl_alloc(M_WAITOK);
1156 	dacl = acl_alloc(M_WAITOK);
1157 
1158 	/*
1159 	 * Retrieve default ACL from parent, if any.
1160 	 */
1161 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1162 	switch (error) {
1163 	case 0:
1164 		/*
1165 		 * Retrieved a default ACL, so merge mode and ACL if
1166 		 * necessary.  If the ACL is empty, fall through to
1167 		 * the "not defined or available" case.
1168 		 */
1169 		if (acl->acl_cnt != 0) {
1170 			dmode = acl_posix1e_newfilemode(dmode, acl);
1171 			ip->i_mode = dmode;
1172 			*dacl = *acl;
1173 			ext2_sync_acl_from_inode(ip, acl);
1174 			break;
1175 		}
1176 		/* FALLTHROUGH */
1177 
1178 	case EOPNOTSUPP:
1179 		/*
1180 		 * Just use the mode as-is.
1181 		 */
1182 		ip->i_mode = dmode;
1183 		error = 0;
1184 		goto out;
1185 
1186 	default:
1187 		goto out;
1188 	}
1189 
1190 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1191 	if (error == 0)
1192 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1193 	switch (error) {
1194 	case 0:
1195 		break;
1196 
1197 	case EOPNOTSUPP:
1198 		/*
1199 		 * XXX: This should not happen, as EOPNOTSUPP above
1200 		 * was supposed to free acl.
1201 		 */
1202 #ifdef DEBUG
1203 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1204 #endif	/* DEBUG */
1205 		break;
1206 
1207 	default:
1208 		goto out;
1209 	}
1210 
1211 out:
1212 	acl_free(acl);
1213 	acl_free(dacl);
1214 
1215 	return (error);
1216 }
1217 
1218 static int
1219 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1220     mode_t mode, struct ucred *cred, struct thread *td)
1221 {
1222 	int error;
1223 	struct inode *ip = VTOI(tvp);
1224 	struct acl *acl;
1225 
1226 	acl = acl_alloc(M_WAITOK);
1227 
1228 	/*
1229 	 * Retrieve default ACL for parent, if any.
1230 	 */
1231 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1232 	switch (error) {
1233 	case 0:
1234 		/*
1235 		 * Retrieved a default ACL, so merge mode and ACL if
1236 		 * necessary.
1237 		 */
1238 		if (acl->acl_cnt != 0) {
1239 			/*
1240 			 * Two possible ways for default ACL to not
1241 			 * be present.  First, the EA can be
1242 			 * undefined, or second, the default ACL can
1243 			 * be blank.  If it's blank, fall through to
1244 			 * the it's not defined case.
1245 			 */
1246 			mode = acl_posix1e_newfilemode(mode, acl);
1247 			ip->i_mode = mode;
1248 			ext2_sync_acl_from_inode(ip, acl);
1249 			break;
1250 		}
1251 		/* FALLTHROUGH */
1252 
1253 	case EOPNOTSUPP:
1254 		/*
1255 		 * Just use the mode as-is.
1256 		 */
1257 		ip->i_mode = mode;
1258 		error = 0;
1259 		goto out;
1260 
1261 	default:
1262 		goto out;
1263 	}
1264 
1265 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1266 	switch (error) {
1267 	case 0:
1268 		break;
1269 
1270 	case EOPNOTSUPP:
1271 		/*
1272 		 * XXX: This should not happen, as EOPNOTSUPP above was
1273 		 * supposed to free acl.
1274 		 */
1275 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1276 		    "but no VOP_SETACL()\n");
1277 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1278 		    "but no VOP_SETACL()"); */
1279 		break;
1280 
1281 	default:
1282 		goto out;
1283 	}
1284 
1285 out:
1286 	acl_free(acl);
1287 
1288 	return (error);
1289 }
1290 
1291 #endif /* UFS_ACL */
1292 
1293 /*
1294  * Mkdir system call
1295  */
1296 static int
1297 ext2_mkdir(struct vop_mkdir_args *ap)
1298 {
1299 	struct m_ext2fs *fs;
1300 	struct vnode *dvp = ap->a_dvp;
1301 	struct vattr *vap = ap->a_vap;
1302 	struct componentname *cnp = ap->a_cnp;
1303 	struct inode *ip, *dp;
1304 	struct vnode *tvp;
1305 	struct dirtemplate dirtemplate, *dtp;
1306 	char *buf = NULL;
1307 	int error, dmode;
1308 
1309 	dp = VTOI(dvp);
1310 	if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX &&
1311 	    !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
1312 		error = EMLINK;
1313 		goto out;
1314 	}
1315 	dmode = vap->va_mode & 0777;
1316 	dmode |= IFDIR;
1317 	/*
1318 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1319 	 * but not have it entered in the parent directory. The entry is
1320 	 * made later after writing "." and ".." entries.
1321 	 */
1322 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1323 	if (error)
1324 		goto out;
1325 	ip = VTOI(tvp);
1326 	fs = ip->i_e2fs;
1327 	ip->i_gid = dp->i_gid;
1328 #ifdef SUIDDIR
1329 	{
1330 		/*
1331 		 * if we are hacking owners here, (only do this where told to)
1332 		 * and we are not giving it TOO root, (would subvert quotas)
1333 		 * then go ahead and give it to the other user.
1334 		 * The new directory also inherits the SUID bit.
1335 		 * If user's UID and dir UID are the same,
1336 		 * 'give it away' so that the SUID is still forced on.
1337 		 */
1338 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1339 		    (dp->i_mode & ISUID) && dp->i_uid) {
1340 			dmode |= ISUID;
1341 			ip->i_uid = dp->i_uid;
1342 		} else {
1343 			ip->i_uid = cnp->cn_cred->cr_uid;
1344 		}
1345 	}
1346 #else
1347 	ip->i_uid = cnp->cn_cred->cr_uid;
1348 #endif
1349 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1350 	ip->i_mode = dmode;
1351 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1352 	ip->i_nlink = 2;
1353 	if (cnp->cn_flags & ISWHITEOUT)
1354 		ip->i_flags |= UF_OPAQUE;
1355 	error = ext2_update(tvp, 1);
1356 
1357 	/*
1358 	 * Bump link count in parent directory
1359 	 * to reflect work done below.  Should
1360 	 * be done before reference is created
1361 	 * so reparation is possible if we crash.
1362 	 */
1363 	ext2_inc_nlink(dp);
1364 	dp->i_flag |= IN_CHANGE;
1365 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1366 	if (error)
1367 		goto bad;
1368 
1369 	/* Initialize directory with "." and ".." from static template. */
1370 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1371 	    EXT2F_INCOMPAT_FTYPE))
1372 		dtp = &mastertemplate;
1373 	else
1374 		dtp = &omastertemplate;
1375 	dirtemplate = *dtp;
1376 	dirtemplate.dot_ino = htole32(ip->i_number);
1377 	dirtemplate.dotdot_ino = htole32(dp->i_number);
1378 	/*
1379 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1380 	 * just redefine it - for this function only
1381 	 */
1382 #undef  DIRBLKSIZ
1383 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1384 	dirtemplate.dotdot_reclen = htole16(DIRBLKSIZ - 12);
1385 	buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1386 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1387 		dirtemplate.dotdot_reclen =
1388 		    htole16(le16toh(dirtemplate.dotdot_reclen) -
1389 		    sizeof(struct ext2fs_direct_tail));
1390 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1391 	}
1392 	memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1393 	ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf);
1394 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1395 	    DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1396 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1397 	    NULL, NULL);
1398 	if (error) {
1399 		ext2_dec_nlink(dp);
1400 		dp->i_flag |= IN_CHANGE;
1401 		goto bad;
1402 	}
1403 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1404 		/* XXX should grow with balloc() */
1405 		panic("ext2_mkdir: blksize");
1406 	else {
1407 		ip->i_size = DIRBLKSIZ;
1408 		ip->i_flag |= IN_CHANGE;
1409 	}
1410 
1411 #ifdef UFS_ACL
1412 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1413 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1414 		    cnp->cn_cred, curthread);
1415 		if (error)
1416 			goto bad;
1417 	}
1418 
1419 #endif /* UFS_ACL */
1420 
1421 	/* Directory set up, now install its entry in the parent directory. */
1422 	error = ext2_direnter(ip, dvp, cnp);
1423 	if (error) {
1424 		ext2_dec_nlink(dp);
1425 		dp->i_flag |= IN_CHANGE;
1426 	}
1427 bad:
1428 	/*
1429 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1430 	 * for us because we set the link count to 0.
1431 	 */
1432 	if (error) {
1433 		ip->i_nlink = 0;
1434 		ip->i_flag |= IN_CHANGE;
1435 		vput(tvp);
1436 	} else
1437 		*ap->a_vpp = tvp;
1438 out:
1439 	free(buf, M_TEMP);
1440 	return (error);
1441 #undef  DIRBLKSIZ
1442 #define DIRBLKSIZ  DEV_BSIZE
1443 }
1444 
1445 /*
1446  * Rmdir system call.
1447  */
1448 static int
1449 ext2_rmdir(struct vop_rmdir_args *ap)
1450 {
1451 	struct vnode *vp = ap->a_vp;
1452 	struct vnode *dvp = ap->a_dvp;
1453 	struct componentname *cnp = ap->a_cnp;
1454 	struct inode *ip, *dp;
1455 	int error;
1456 
1457 	ip = VTOI(vp);
1458 	dp = VTOI(dvp);
1459 
1460 	/*
1461 	 * Verify the directory is empty (and valid).
1462 	 * (Rmdir ".." won't be valid since
1463 	 *  ".." will contain a reference to
1464 	 *  the current directory and thus be
1465 	 *  non-empty.)
1466 	 */
1467 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1468 		error = ENOTEMPTY;
1469 		goto out;
1470 	}
1471 	if ((dp->i_flags & APPEND)
1472 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1473 		error = EPERM;
1474 		goto out;
1475 	}
1476 	/*
1477 	 * Delete reference to directory before purging
1478 	 * inode.  If we crash in between, the directory
1479 	 * will be reattached to lost+found,
1480 	 */
1481 	error = ext2_dirremove(dvp, cnp);
1482 	if (error)
1483 		goto out;
1484 	ext2_dec_nlink(dp);
1485 	dp->i_flag |= IN_CHANGE;
1486 	cache_purge(dvp);
1487 	VOP_UNLOCK(dvp);
1488 	/*
1489 	 * Truncate inode.  The only stuff left
1490 	 * in the directory is "." and "..".
1491 	 */
1492 	ip->i_nlink = 0;
1493 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1494 	    curthread);
1495 	cache_purge(ITOV(ip));
1496 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1497 		VOP_UNLOCK(vp);
1498 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1499 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1500 	}
1501 out:
1502 	return (error);
1503 }
1504 
1505 /*
1506  * symlink -- make a symbolic link
1507  */
1508 static int
1509 ext2_symlink(struct vop_symlink_args *ap)
1510 {
1511 	struct vnode *vp, **vpp = ap->a_vpp;
1512 	struct inode *ip;
1513 	int len, error;
1514 
1515 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1516 	    vpp, ap->a_cnp);
1517 	if (error)
1518 		return (error);
1519 	vp = *vpp;
1520 	len = strlen(ap->a_target);
1521 	if (len < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1522 		ip = VTOI(vp);
1523 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1524 		ip->i_size = len;
1525 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1526 	} else
1527 		error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
1528 		    len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1529 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1530 	if (error)
1531 		vput(vp);
1532 	return (error);
1533 }
1534 
1535 /*
1536  * Return target name of a symbolic link
1537  */
1538 static int
1539 ext2_readlink(struct vop_readlink_args *ap)
1540 {
1541 	struct vnode *vp = ap->a_vp;
1542 	struct inode *ip = VTOI(vp);
1543 	int isize;
1544 
1545 	isize = ip->i_size;
1546 	if (isize < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1547 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1548 		return (0);
1549 	}
1550 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1551 }
1552 
1553 /*
1554  * Calculate the logical to physical mapping if not done already,
1555  * then call the device strategy routine.
1556  *
1557  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1558  * deadlock on memory.  See ext2_bmap() for details.
1559  */
1560 static int
1561 ext2_strategy(struct vop_strategy_args *ap)
1562 {
1563 	struct buf *bp = ap->a_bp;
1564 	struct vnode *vp = ap->a_vp;
1565 	struct bufobj *bo;
1566 	daddr_t blkno;
1567 	int error;
1568 
1569 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1570 		panic("ext2_strategy: spec");
1571 	if (bp->b_blkno == bp->b_lblkno) {
1572 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1573 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1574 		else
1575 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1576 
1577 		bp->b_blkno = blkno;
1578 		if (error) {
1579 			bp->b_error = error;
1580 			bp->b_ioflags |= BIO_ERROR;
1581 			bufdone(bp);
1582 			return (0);
1583 		}
1584 		if ((long)bp->b_blkno == -1)
1585 			vfs_bio_clrbuf(bp);
1586 	}
1587 	if ((long)bp->b_blkno == -1) {
1588 		bufdone(bp);
1589 		return (0);
1590 	}
1591 	bp->b_iooffset = dbtob(bp->b_blkno);
1592 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1593 	BO_STRATEGY(bo, bp);
1594 	return (0);
1595 }
1596 
1597 /*
1598  * Print out the contents of an inode.
1599  */
1600 static int
1601 ext2_print(struct vop_print_args *ap)
1602 {
1603 	struct vnode *vp = ap->a_vp;
1604 	struct inode *ip = VTOI(vp);
1605 
1606 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1607 	if (vp->v_type == VFIFO)
1608 		fifo_printinfo(vp);
1609 	printf("\n");
1610 	return (0);
1611 }
1612 
1613 /*
1614  * Close wrapper for fifos.
1615  *
1616  * Update the times on the inode then do device close.
1617  */
1618 static int
1619 ext2fifo_close(struct vop_close_args *ap)
1620 {
1621 	struct vnode *vp = ap->a_vp;
1622 
1623 	VI_LOCK(vp);
1624 	if (vp->v_usecount > 1)
1625 		ext2_itimes_locked(vp);
1626 	VI_UNLOCK(vp);
1627 	return (fifo_specops.vop_close(ap));
1628 }
1629 
1630 /*
1631  * Return POSIX pathconf information applicable to ext2 filesystems.
1632  */
1633 static int
1634 ext2_pathconf(struct vop_pathconf_args *ap)
1635 {
1636 	int error = 0;
1637 
1638 	switch (ap->a_name) {
1639 	case _PC_LINK_MAX:
1640 		if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs,
1641 		    EXT2F_ROCOMPAT_DIR_NLINK))
1642 			*ap->a_retval = INT_MAX;
1643 		else
1644 			*ap->a_retval = EXT4_LINK_MAX;
1645 		break;
1646 	case _PC_NAME_MAX:
1647 		*ap->a_retval = NAME_MAX;
1648 		break;
1649 	case _PC_PIPE_BUF:
1650 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1651 			*ap->a_retval = PIPE_BUF;
1652 		else
1653 			error = EINVAL;
1654 		break;
1655 	case _PC_CHOWN_RESTRICTED:
1656 		*ap->a_retval = 1;
1657 		break;
1658 	case _PC_NO_TRUNC:
1659 		*ap->a_retval = 1;
1660 		break;
1661 
1662 #ifdef UFS_ACL
1663 	case _PC_ACL_EXTENDED:
1664 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1665 			*ap->a_retval = 1;
1666 		else
1667 			*ap->a_retval = 0;
1668 		break;
1669 	case _PC_ACL_PATH_MAX:
1670 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1671 			*ap->a_retval = ACL_MAX_ENTRIES;
1672 		else
1673 			*ap->a_retval = 3;
1674 		break;
1675 #endif /* UFS_ACL */
1676 
1677 	case _PC_MIN_HOLE_SIZE:
1678 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1679 		break;
1680 	case _PC_PRIO_IO:
1681 		*ap->a_retval = 0;
1682 		break;
1683 	case _PC_SYNC_IO:
1684 		*ap->a_retval = 0;
1685 		break;
1686 	case _PC_ALLOC_SIZE_MIN:
1687 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1688 		break;
1689 	case _PC_FILESIZEBITS:
1690 		*ap->a_retval = 64;
1691 		break;
1692 	case _PC_REC_INCR_XFER_SIZE:
1693 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1694 		break;
1695 	case _PC_REC_MAX_XFER_SIZE:
1696 		*ap->a_retval = -1;	/* means ``unlimited'' */
1697 		break;
1698 	case _PC_REC_MIN_XFER_SIZE:
1699 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1700 		break;
1701 	case _PC_REC_XFER_ALIGN:
1702 		*ap->a_retval = PAGE_SIZE;
1703 		break;
1704 	case _PC_SYMLINK_MAX:
1705 		*ap->a_retval = MAXPATHLEN;
1706 		break;
1707 
1708 	default:
1709 		error = vop_stdpathconf(ap);
1710 		break;
1711 	}
1712 	return (error);
1713 }
1714 
1715 /*
1716  * Vnode operation to remove a named attribute.
1717  */
1718 static int
1719 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1720 {
1721 	struct inode *ip;
1722 	struct m_ext2fs *fs;
1723 	int error;
1724 
1725 	ip = VTOI(ap->a_vp);
1726 	fs = ip->i_e2fs;
1727 
1728 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1729 		return (EOPNOTSUPP);
1730 
1731 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1732 		return (EOPNOTSUPP);
1733 
1734 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1735 	    ap->a_cred, ap->a_td, VWRITE);
1736 	if (error)
1737 		return (error);
1738 
1739 	error = ENOATTR;
1740 
1741 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1742 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1743 		if (error != ENOATTR)
1744 			return (error);
1745 	}
1746 
1747 	if (ip->i_facl)
1748 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1749 
1750 	return (error);
1751 }
1752 
1753 /*
1754  * Vnode operation to retrieve a named extended attribute.
1755  */
1756 static int
1757 ext2_getextattr(struct vop_getextattr_args *ap)
1758 {
1759 	struct inode *ip;
1760 	struct m_ext2fs *fs;
1761 	int error;
1762 
1763 	ip = VTOI(ap->a_vp);
1764 	fs = ip->i_e2fs;
1765 
1766 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1767 		return (EOPNOTSUPP);
1768 
1769 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1770 		return (EOPNOTSUPP);
1771 
1772 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1773 	    ap->a_cred, ap->a_td, VREAD);
1774 	if (error)
1775 		return (error);
1776 
1777 	if (ap->a_size != NULL)
1778 		*ap->a_size = 0;
1779 
1780 	error = ENOATTR;
1781 
1782 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1783 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1784 		    ap->a_name, ap->a_uio, ap->a_size);
1785 		if (error != ENOATTR)
1786 			return (error);
1787 	}
1788 
1789 	if (ip->i_facl)
1790 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1791 		    ap->a_name, ap->a_uio, ap->a_size);
1792 
1793 	return (error);
1794 }
1795 
1796 /*
1797  * Vnode operation to retrieve extended attributes on a vnode.
1798  */
1799 static int
1800 ext2_listextattr(struct vop_listextattr_args *ap)
1801 {
1802 	struct inode *ip;
1803 	struct m_ext2fs *fs;
1804 	int error;
1805 
1806 	ip = VTOI(ap->a_vp);
1807 	fs = ip->i_e2fs;
1808 
1809 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1810 		return (EOPNOTSUPP);
1811 
1812 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1813 		return (EOPNOTSUPP);
1814 
1815 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1816 	    ap->a_cred, ap->a_td, VREAD);
1817 	if (error)
1818 		return (error);
1819 
1820 	if (ap->a_size != NULL)
1821 		*ap->a_size = 0;
1822 
1823 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1824 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1825 		    ap->a_uio, ap->a_size);
1826 		if (error)
1827 			return (error);
1828 	}
1829 
1830 	if (ip->i_facl)
1831 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1832 		    ap->a_uio, ap->a_size);
1833 
1834 	return (error);
1835 }
1836 
1837 /*
1838  * Vnode operation to set a named attribute.
1839  */
1840 static int
1841 ext2_setextattr(struct vop_setextattr_args *ap)
1842 {
1843 	struct inode *ip;
1844 	struct m_ext2fs *fs;
1845 	int error;
1846 
1847 	ip = VTOI(ap->a_vp);
1848 	fs = ip->i_e2fs;
1849 
1850 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1851 		return (EOPNOTSUPP);
1852 
1853 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1854 		return (EOPNOTSUPP);
1855 
1856 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1857 	    ap->a_cred, ap->a_td, VWRITE);
1858 	if (error)
1859 		return (error);
1860 
1861 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1862 	if (error)
1863 		return (error);
1864 
1865 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1866 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1867 		    ap->a_name, ap->a_uio);
1868 		if (error != ENOSPC)
1869 			return (error);
1870 	}
1871 
1872 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1873 	    ap->a_name, ap->a_uio);
1874 
1875 	return (error);
1876 }
1877 
1878 /*
1879  * Vnode pointer to File handle
1880  */
1881 /* ARGSUSED */
1882 static int
1883 ext2_vptofh(struct vop_vptofh_args *ap)
1884 {
1885 	struct inode *ip;
1886 	struct ufid *ufhp;
1887 
1888 	ip = VTOI(ap->a_vp);
1889 	ufhp = (struct ufid *)ap->a_fhp;
1890 	ufhp->ufid_len = sizeof(struct ufid);
1891 	ufhp->ufid_ino = ip->i_number;
1892 	ufhp->ufid_gen = ip->i_gen;
1893 	return (0);
1894 }
1895 
1896 /*
1897  * Initialize the vnode associated with a new inode, handle aliased
1898  * vnodes.
1899  */
1900 int
1901 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1902 {
1903 	struct inode *ip;
1904 	struct vnode *vp;
1905 
1906 	vp = *vpp;
1907 	ip = VTOI(vp);
1908 	vp->v_type = IFTOVT(ip->i_mode);
1909 	/*
1910 	 * Only unallocated inodes should be of type VNON.
1911 	 */
1912 	if (ip->i_mode != 0 && vp->v_type == VNON)
1913 		return (EINVAL);
1914 	if (vp->v_type == VFIFO)
1915 		vp->v_op = fifoops;
1916 
1917 	if (ip->i_number == EXT2_ROOTINO)
1918 		vp->v_vflag |= VV_ROOT;
1919 	ip->i_modrev = init_va_filerev();
1920 	*vpp = vp;
1921 	return (0);
1922 }
1923 
1924 /*
1925  * Allocate a new inode.
1926  */
1927 static int
1928 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1929     struct componentname *cnp)
1930 {
1931 	struct inode *ip, *pdir;
1932 	struct vnode *tvp;
1933 	int error;
1934 
1935 	pdir = VTOI(dvp);
1936 	*vpp = NULL;
1937 	if ((mode & IFMT) == 0)
1938 		mode |= IFREG;
1939 
1940 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1941 	if (error) {
1942 		return (error);
1943 	}
1944 	ip = VTOI(tvp);
1945 	ip->i_gid = pdir->i_gid;
1946 #ifdef SUIDDIR
1947 	{
1948 		/*
1949 		 * if we are
1950 		 * not the owner of the directory,
1951 		 * and we are hacking owners here, (only do this where told to)
1952 		 * and we are not giving it TOO root, (would subvert quotas)
1953 		 * then go ahead and give it to the other user.
1954 		 * Note that this drops off the execute bits for security.
1955 		 */
1956 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1957 		    (pdir->i_mode & ISUID) &&
1958 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1959 			ip->i_uid = pdir->i_uid;
1960 			mode &= ~07111;
1961 		} else {
1962 			ip->i_uid = cnp->cn_cred->cr_uid;
1963 		}
1964 	}
1965 #else
1966 	ip->i_uid = cnp->cn_cred->cr_uid;
1967 #endif
1968 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1969 	ip->i_mode = mode;
1970 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1971 	ip->i_nlink = 1;
1972 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1973 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID))
1974 			ip->i_mode &= ~ISGID;
1975 	}
1976 
1977 	if (cnp->cn_flags & ISWHITEOUT)
1978 		ip->i_flags |= UF_OPAQUE;
1979 
1980 	/*
1981 	 * Make sure inode goes to disk before directory entry.
1982 	 */
1983 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1984 	if (error)
1985 		goto bad;
1986 
1987 #ifdef UFS_ACL
1988 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1989 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
1990 		    cnp->cn_cred, curthread);
1991 		if (error)
1992 			goto bad;
1993 	}
1994 #endif /* UFS_ACL */
1995 
1996 	error = ext2_direnter(ip, dvp, cnp);
1997 	if (error)
1998 		goto bad;
1999 
2000 	*vpp = tvp;
2001 	return (0);
2002 
2003 bad:
2004 	/*
2005 	 * Write error occurred trying to update the inode
2006 	 * or the directory so must deallocate the inode.
2007 	 */
2008 	ip->i_nlink = 0;
2009 	ip->i_flag |= IN_CHANGE;
2010 	vput(tvp);
2011 	return (error);
2012 }
2013 
2014 /*
2015  * Vnode op for reading.
2016  */
2017 static int
2018 ext2_read(struct vop_read_args *ap)
2019 {
2020 	struct vnode *vp;
2021 	struct inode *ip;
2022 	struct uio *uio;
2023 	struct m_ext2fs *fs;
2024 	struct buf *bp;
2025 	daddr_t lbn, nextlbn;
2026 	off_t bytesinfile;
2027 	long size, xfersize, blkoffset;
2028 	int error, orig_resid, seqcount;
2029 	int ioflag;
2030 
2031 	vp = ap->a_vp;
2032 	uio = ap->a_uio;
2033 	ioflag = ap->a_ioflag;
2034 
2035 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2036 	ip = VTOI(vp);
2037 
2038 #ifdef INVARIANTS
2039 	if (uio->uio_rw != UIO_READ)
2040 		panic("%s: mode", "ext2_read");
2041 
2042 	if (vp->v_type == VLNK) {
2043 		if ((int)ip->i_size <
2044 		    VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen)
2045 			panic("%s: short symlink", "ext2_read");
2046 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2047 		panic("%s: type %d", "ext2_read", vp->v_type);
2048 #endif
2049 	orig_resid = uio->uio_resid;
2050 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2051 	if (orig_resid == 0)
2052 		return (0);
2053 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2054 	fs = ip->i_e2fs;
2055 	if (uio->uio_offset < ip->i_size &&
2056 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2057 		return (EOVERFLOW);
2058 
2059 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2060 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2061 			break;
2062 		lbn = lblkno(fs, uio->uio_offset);
2063 		nextlbn = lbn + 1;
2064 		size = blksize(fs, ip, lbn);
2065 		blkoffset = blkoff(fs, uio->uio_offset);
2066 
2067 		xfersize = fs->e2fs_fsize - blkoffset;
2068 		if (uio->uio_resid < xfersize)
2069 			xfersize = uio->uio_resid;
2070 		if (bytesinfile < xfersize)
2071 			xfersize = bytesinfile;
2072 
2073 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2074 			error = bread(vp, lbn, size, NOCRED, &bp);
2075 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2076 			error = cluster_read(vp, ip->i_size, lbn, size,
2077 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2078 			    0, &bp);
2079 		} else if (seqcount > 1) {
2080 			u_int nextsize = blksize(fs, ip, nextlbn);
2081 
2082 			error = breadn(vp, lbn,
2083 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2084 		} else
2085 			error = bread(vp, lbn, size, NOCRED, &bp);
2086 		if (error) {
2087 			brelse(bp);
2088 			bp = NULL;
2089 			break;
2090 		}
2091 
2092 		/*
2093 		 * We should only get non-zero b_resid when an I/O error
2094 		 * has occurred, which should cause us to break above.
2095 		 * However, if the short read did not cause an error,
2096 		 * then we want to ensure that we do not uiomove bad
2097 		 * or uninitialized data.
2098 		 */
2099 		size -= bp->b_resid;
2100 		if (size < xfersize) {
2101 			if (size == 0)
2102 				break;
2103 			xfersize = size;
2104 		}
2105 		error = uiomove((char *)bp->b_data + blkoffset,
2106 		    (int)xfersize, uio);
2107 		if (error)
2108 			break;
2109 		vfs_bio_brelse(bp, ioflag);
2110 	}
2111 
2112 	/*
2113 	 * This can only happen in the case of an error because the loop
2114 	 * above resets bp to NULL on each iteration and on normal
2115 	 * completion has not set a new value into it. so it must have come
2116 	 * from a 'break' statement
2117 	 */
2118 	if (bp != NULL)
2119 		vfs_bio_brelse(bp, ioflag);
2120 
2121 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2122 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2123 		ip->i_flag |= IN_ACCESS;
2124 	return (error);
2125 }
2126 
2127 static int
2128 ext2_ioctl(struct vop_ioctl_args *ap)
2129 {
2130 	struct vnode *vp;
2131 	int error;
2132 
2133 	vp = ap->a_vp;
2134 	switch (ap->a_command) {
2135 	case FIOSEEKDATA:
2136 		if (!(VTOI(vp)->i_flag & IN_E4EXTENTS)) {
2137 			error = vn_lock(vp, LK_SHARED);
2138 			if (error == 0) {
2139 				error = ext2_bmap_seekdata(vp,
2140 				    (off_t *)ap->a_data);
2141 				VOP_UNLOCK(vp);
2142 			} else
2143 				error = EBADF;
2144 			return (error);
2145 		}
2146 	case FIOSEEKHOLE:
2147 		return (vn_bmap_seekhole(vp, ap->a_command,
2148 		    (off_t *)ap->a_data, ap->a_cred));
2149 	default:
2150 		return (ENOTTY);
2151 	}
2152 }
2153 
2154 /*
2155  * Vnode op for writing.
2156  */
2157 static int
2158 ext2_write(struct vop_write_args *ap)
2159 {
2160 	struct vnode *vp;
2161 	struct uio *uio;
2162 	struct inode *ip;
2163 	struct m_ext2fs *fs;
2164 	struct buf *bp;
2165 	daddr_t lbn;
2166 	off_t osize;
2167 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2168 
2169 	ioflag = ap->a_ioflag;
2170 	uio = ap->a_uio;
2171 	vp = ap->a_vp;
2172 
2173 	seqcount = ioflag >> IO_SEQSHIFT;
2174 	ip = VTOI(vp);
2175 
2176 #ifdef INVARIANTS
2177 	if (uio->uio_rw != UIO_WRITE)
2178 		panic("%s: mode", "ext2_write");
2179 #endif
2180 
2181 	switch (vp->v_type) {
2182 	case VREG:
2183 		if (ioflag & IO_APPEND)
2184 			uio->uio_offset = ip->i_size;
2185 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2186 			return (EPERM);
2187 		/* FALLTHROUGH */
2188 	case VLNK:
2189 		break;
2190 	case VDIR:
2191 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2192 		if ((ioflag & IO_SYNC) == 0)
2193 			panic("ext2_write: nonsync dir write");
2194 		break;
2195 	default:
2196 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2197 		    vp->v_type, (intmax_t)uio->uio_offset,
2198 		    (intmax_t)uio->uio_resid);
2199 	}
2200 
2201 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2202 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2203 	fs = ip->i_e2fs;
2204 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2205 		return (EFBIG);
2206 	/*
2207 	 * Maybe this should be above the vnode op call, but so long as
2208 	 * file servers have no limits, I don't think it matters.
2209 	 */
2210 	error = vn_rlimit_fsize(vp, uio, uio->uio_td);
2211 	if (error != 0)
2212 		return (error);
2213 
2214 	resid = uio->uio_resid;
2215 	osize = ip->i_size;
2216 	if (seqcount > BA_SEQMAX)
2217 		flags = BA_SEQMAX << BA_SEQSHIFT;
2218 	else
2219 		flags = seqcount << BA_SEQSHIFT;
2220 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2221 		flags |= IO_SYNC;
2222 
2223 	for (error = 0; uio->uio_resid > 0;) {
2224 		lbn = lblkno(fs, uio->uio_offset);
2225 		blkoffset = blkoff(fs, uio->uio_offset);
2226 		xfersize = fs->e2fs_fsize - blkoffset;
2227 		if (uio->uio_resid < xfersize)
2228 			xfersize = uio->uio_resid;
2229 		if (uio->uio_offset + xfersize > ip->i_size)
2230 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2231 
2232 		/*
2233 		 * We must perform a read-before-write if the transfer size
2234 		 * does not cover the entire buffer.
2235 		 */
2236 		if (fs->e2fs_bsize > xfersize)
2237 			flags |= BA_CLRBUF;
2238 		else
2239 			flags &= ~BA_CLRBUF;
2240 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2241 		    ap->a_cred, &bp, flags);
2242 		if (error != 0)
2243 			break;
2244 
2245 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2246 			bp->b_flags |= B_NOCACHE;
2247 		if (uio->uio_offset + xfersize > ip->i_size)
2248 			ip->i_size = uio->uio_offset + xfersize;
2249 		size = blksize(fs, ip, lbn) - bp->b_resid;
2250 		if (size < xfersize)
2251 			xfersize = size;
2252 
2253 		error =
2254 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2255 		/*
2256 		 * If the buffer is not already filled and we encounter an
2257 		 * error while trying to fill it, we have to clear out any
2258 		 * garbage data from the pages instantiated for the buffer.
2259 		 * If we do not, a failed uiomove() during a write can leave
2260 		 * the prior contents of the pages exposed to a userland mmap.
2261 		 *
2262 		 * Note that we need only clear buffers with a transfer size
2263 		 * equal to the block size because buffers with a shorter
2264 		 * transfer size were cleared above by the call to ext2_balloc()
2265 		 * with the BA_CLRBUF flag set.
2266 		 *
2267 		 * If the source region for uiomove identically mmaps the
2268 		 * buffer, uiomove() performed the NOP copy, and the buffer
2269 		 * content remains valid because the page fault handler
2270 		 * validated the pages.
2271 		 */
2272 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2273 		    fs->e2fs_bsize == xfersize)
2274 			vfs_bio_clrbuf(bp);
2275 
2276 		vfs_bio_set_flags(bp, ioflag);
2277 
2278 		/*
2279 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2280 		 * if we have a severe page deficiency write the buffer
2281 		 * asynchronously.  Otherwise try to cluster, and if that
2282 		 * doesn't do it then either do an async write (if O_DIRECT),
2283 		 * or a delayed write (if not).
2284 		 */
2285 		if (ioflag & IO_SYNC) {
2286 			(void)bwrite(bp);
2287 		} else if (vm_page_count_severe() ||
2288 			    buf_dirty_count_severe() ||
2289 		    (ioflag & IO_ASYNC)) {
2290 			bp->b_flags |= B_CLUSTEROK;
2291 			bawrite(bp);
2292 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2293 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2294 				bp->b_flags |= B_CLUSTEROK;
2295 				cluster_write(vp, &ip->i_clusterw, bp,
2296 				    ip->i_size, seqcount, 0);
2297 			} else {
2298 				bawrite(bp);
2299 			}
2300 		} else if (ioflag & IO_DIRECT) {
2301 			bp->b_flags |= B_CLUSTEROK;
2302 			bawrite(bp);
2303 		} else {
2304 			bp->b_flags |= B_CLUSTEROK;
2305 			bdwrite(bp);
2306 		}
2307 		if (error || xfersize == 0)
2308 			break;
2309 	}
2310 	/*
2311 	 * If we successfully wrote any data, and we are not the superuser
2312 	 * we clear the setuid and setgid bits as a precaution against
2313 	 * tampering.
2314 	 */
2315 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2316 	    ap->a_cred) {
2317 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID))
2318 			ip->i_mode &= ~(ISUID | ISGID);
2319 	}
2320 	if (error) {
2321 		if (ioflag & IO_UNIT) {
2322 			(void)ext2_truncate(vp, osize,
2323 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2324 			uio->uio_offset -= resid - uio->uio_resid;
2325 			uio->uio_resid = resid;
2326 		}
2327 	}
2328 	if (uio->uio_resid != resid) {
2329 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2330 		if (ioflag & IO_SYNC)
2331 			error = ext2_update(vp, 1);
2332 	}
2333 	return (error);
2334 }
2335