xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 12a88a3d637e7e7a3726e8dedbcf82cb96cea529)
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  * Copyright (c) 1982, 1986, 1989, 1993
9  *	The Regents of the University of California.  All rights reserved.
10  * (c) UNIX System Laboratories, Inc.
11  * All or some portions of this file are derived from material licensed
12  * to the University of California by American Telephone and Telegraph
13  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14  * the permission of UNIX System Laboratories, Inc.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
41  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42  * $FreeBSD$
43  */
44 
45 #include "opt_suiddir.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
60 #include <sys/time.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 #include <sys/extattr.h>
68 #include <sys/vmmeter.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vnode_pager.h>
77 
78 #include "opt_directio.h"
79 
80 #include <ufs/ufs/dir.h>
81 
82 #include <fs/ext2fs/fs.h>
83 #include <fs/ext2fs/inode.h>
84 #include <fs/ext2fs/ext2_acl.h>
85 #include <fs/ext2fs/ext2_extern.h>
86 #include <fs/ext2fs/ext2fs.h>
87 #include <fs/ext2fs/ext2_dinode.h>
88 #include <fs/ext2fs/ext2_dir.h>
89 #include <fs/ext2fs/ext2_mount.h>
90 #include <fs/ext2fs/ext2_extattr.h>
91 
92 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
93 static void ext2_itimes_locked(struct vnode *);
94 
95 static vop_access_t	ext2_access;
96 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
97 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
98     struct thread *);
99 static vop_close_t	ext2_close;
100 static vop_create_t	ext2_create;
101 static vop_fsync_t	ext2_fsync;
102 static vop_getattr_t	ext2_getattr;
103 static vop_ioctl_t	ext2_ioctl;
104 static vop_link_t	ext2_link;
105 static vop_mkdir_t	ext2_mkdir;
106 static vop_mknod_t	ext2_mknod;
107 static vop_open_t	ext2_open;
108 static vop_pathconf_t	ext2_pathconf;
109 static vop_print_t	ext2_print;
110 static vop_read_t	ext2_read;
111 static vop_readlink_t	ext2_readlink;
112 static vop_remove_t	ext2_remove;
113 static vop_rename_t	ext2_rename;
114 static vop_rmdir_t	ext2_rmdir;
115 static vop_setattr_t	ext2_setattr;
116 static vop_strategy_t	ext2_strategy;
117 static vop_symlink_t	ext2_symlink;
118 static vop_write_t	ext2_write;
119 static vop_deleteextattr_t	ext2_deleteextattr;
120 static vop_getextattr_t	ext2_getextattr;
121 static vop_listextattr_t	ext2_listextattr;
122 static vop_setextattr_t	ext2_setextattr;
123 static vop_vptofh_t	ext2_vptofh;
124 static vop_close_t	ext2fifo_close;
125 static vop_kqfilter_t	ext2fifo_kqfilter;
126 
127 /* Global vfs data structures for ext2. */
128 struct vop_vector ext2_vnodeops = {
129 	.vop_default =		&default_vnodeops,
130 	.vop_access =		ext2_access,
131 	.vop_bmap =		ext2_bmap,
132 	.vop_cachedlookup =	ext2_lookup,
133 	.vop_close =		ext2_close,
134 	.vop_create =		ext2_create,
135 	.vop_fsync =		ext2_fsync,
136 	.vop_getpages =		vnode_pager_local_getpages,
137 	.vop_getpages_async =	vnode_pager_local_getpages_async,
138 	.vop_getattr =		ext2_getattr,
139 	.vop_inactive =		ext2_inactive,
140 	.vop_ioctl =		ext2_ioctl,
141 	.vop_link =		ext2_link,
142 	.vop_lookup =		vfs_cache_lookup,
143 	.vop_mkdir =		ext2_mkdir,
144 	.vop_mknod =		ext2_mknod,
145 	.vop_open =		ext2_open,
146 	.vop_pathconf =		ext2_pathconf,
147 	.vop_poll =		vop_stdpoll,
148 	.vop_print =		ext2_print,
149 	.vop_read =		ext2_read,
150 	.vop_readdir =		ext2_readdir,
151 	.vop_readlink =		ext2_readlink,
152 	.vop_reallocblks =	ext2_reallocblks,
153 	.vop_reclaim =		ext2_reclaim,
154 	.vop_remove =		ext2_remove,
155 	.vop_rename =		ext2_rename,
156 	.vop_rmdir =		ext2_rmdir,
157 	.vop_setattr =		ext2_setattr,
158 	.vop_strategy =		ext2_strategy,
159 	.vop_symlink =		ext2_symlink,
160 	.vop_write =		ext2_write,
161 	.vop_deleteextattr =	ext2_deleteextattr,
162 	.vop_getextattr =	ext2_getextattr,
163 	.vop_listextattr =	ext2_listextattr,
164 	.vop_setextattr =	ext2_setextattr,
165 #ifdef UFS_ACL
166 	.vop_getacl =		ext2_getacl,
167 	.vop_setacl =		ext2_setacl,
168 	.vop_aclcheck =		ext2_aclcheck,
169 #endif /* UFS_ACL */
170 	.vop_vptofh =		ext2_vptofh,
171 };
172 
173 struct vop_vector ext2_fifoops = {
174 	.vop_default =		&fifo_specops,
175 	.vop_access =		ext2_access,
176 	.vop_close =		ext2fifo_close,
177 	.vop_fsync =		ext2_fsync,
178 	.vop_getattr =		ext2_getattr,
179 	.vop_inactive =		ext2_inactive,
180 	.vop_kqfilter =		ext2fifo_kqfilter,
181 	.vop_print =		ext2_print,
182 	.vop_read =		VOP_PANIC,
183 	.vop_reclaim =		ext2_reclaim,
184 	.vop_setattr =		ext2_setattr,
185 	.vop_write =		VOP_PANIC,
186 	.vop_vptofh =		ext2_vptofh,
187 };
188 
189 /*
190  * A virgin directory (no blushing please).
191  * Note that the type and namlen fields are reversed relative to ext2.
192  * Also, we don't use `struct odirtemplate', since it would just cause
193  * endianness problems.
194  */
195 static struct dirtemplate mastertemplate = {
196 	0, 12, 1, EXT2_FT_DIR, ".",
197 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
198 };
199 static struct dirtemplate omastertemplate = {
200 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
201 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
202 };
203 
204 static void
205 ext2_itimes_locked(struct vnode *vp)
206 {
207 	struct inode *ip;
208 	struct timespec ts;
209 
210 	ASSERT_VI_LOCKED(vp, __func__);
211 
212 	ip = VTOI(vp);
213 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
214 		return;
215 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
216 		ip->i_flag |= IN_LAZYMOD;
217 	else
218 		ip->i_flag |= IN_MODIFIED;
219 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
220 		vfs_timestamp(&ts);
221 		if (ip->i_flag & IN_ACCESS) {
222 			ip->i_atime = ts.tv_sec;
223 			ip->i_atimensec = ts.tv_nsec;
224 		}
225 		if (ip->i_flag & IN_UPDATE) {
226 			ip->i_mtime = ts.tv_sec;
227 			ip->i_mtimensec = ts.tv_nsec;
228 			ip->i_modrev++;
229 		}
230 		if (ip->i_flag & IN_CHANGE) {
231 			ip->i_ctime = ts.tv_sec;
232 			ip->i_ctimensec = ts.tv_nsec;
233 		}
234 	}
235 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
236 }
237 
238 void
239 ext2_itimes(struct vnode *vp)
240 {
241 
242 	VI_LOCK(vp);
243 	ext2_itimes_locked(vp);
244 	VI_UNLOCK(vp);
245 }
246 
247 /*
248  * Create a regular file
249  */
250 static int
251 ext2_create(struct vop_create_args *ap)
252 {
253 	int error;
254 
255 	error =
256 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
257 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
258 	if (error != 0)
259 		return (error);
260 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
261 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
262 	return (0);
263 }
264 
265 static int
266 ext2_open(struct vop_open_args *ap)
267 {
268 
269 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
270 		return (EOPNOTSUPP);
271 
272 	/*
273 	 * Files marked append-only must be opened for appending.
274 	 */
275 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
276 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
277 		return (EPERM);
278 
279 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
280 
281 	return (0);
282 }
283 
284 /*
285  * Close called.
286  *
287  * Update the times on the inode.
288  */
289 static int
290 ext2_close(struct vop_close_args *ap)
291 {
292 	struct vnode *vp = ap->a_vp;
293 
294 	VI_LOCK(vp);
295 	if (vp->v_usecount > 1)
296 		ext2_itimes_locked(vp);
297 	VI_UNLOCK(vp);
298 	return (0);
299 }
300 
301 static int
302 ext2_access(struct vop_access_args *ap)
303 {
304 	struct vnode *vp = ap->a_vp;
305 	struct inode *ip = VTOI(vp);
306 	accmode_t accmode = ap->a_accmode;
307 	int error;
308 
309 	if (vp->v_type == VBLK || vp->v_type == VCHR)
310 		return (EOPNOTSUPP);
311 
312 	/*
313 	 * Disallow write attempts on read-only file systems;
314 	 * unless the file is a socket, fifo, or a block or
315 	 * character device resident on the file system.
316 	 */
317 	if (accmode & VWRITE) {
318 		switch (vp->v_type) {
319 		case VDIR:
320 		case VLNK:
321 		case VREG:
322 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
323 				return (EROFS);
324 			break;
325 		default:
326 			break;
327 		}
328 	}
329 
330 	/* If immutable bit set, nobody gets to write it. */
331 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
332 		return (EPERM);
333 
334 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
335 	    ap->a_accmode, ap->a_cred, NULL);
336 	return (error);
337 }
338 
339 static int
340 ext2_getattr(struct vop_getattr_args *ap)
341 {
342 	struct vnode *vp = ap->a_vp;
343 	struct inode *ip = VTOI(vp);
344 	struct vattr *vap = ap->a_vap;
345 
346 	ext2_itimes(vp);
347 	/*
348 	 * Copy from inode table
349 	 */
350 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
351 	vap->va_fileid = ip->i_number;
352 	vap->va_mode = ip->i_mode & ~IFMT;
353 	vap->va_nlink = ip->i_nlink;
354 	vap->va_uid = ip->i_uid;
355 	vap->va_gid = ip->i_gid;
356 	vap->va_rdev = ip->i_rdev;
357 	vap->va_size = ip->i_size;
358 	vap->va_atime.tv_sec = ip->i_atime;
359 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
360 	vap->va_mtime.tv_sec = ip->i_mtime;
361 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
362 	vap->va_ctime.tv_sec = ip->i_ctime;
363 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
364 	if E2DI_HAS_XTIME(ip) {
365 		vap->va_birthtime.tv_sec = ip->i_birthtime;
366 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
367 	}
368 	vap->va_flags = ip->i_flags;
369 	vap->va_gen = ip->i_gen;
370 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
371 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
372 	vap->va_type = IFTOVT(ip->i_mode);
373 	vap->va_filerev = ip->i_modrev;
374 	return (0);
375 }
376 
377 /*
378  * Set attribute vnode op. called from several syscalls
379  */
380 static int
381 ext2_setattr(struct vop_setattr_args *ap)
382 {
383 	struct vattr *vap = ap->a_vap;
384 	struct vnode *vp = ap->a_vp;
385 	struct inode *ip = VTOI(vp);
386 	struct ucred *cred = ap->a_cred;
387 	struct thread *td = curthread;
388 	int error;
389 
390 	/*
391 	 * Check for unsettable attributes.
392 	 */
393 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
394 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
395 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
396 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
397 		return (EINVAL);
398 	}
399 	if (vap->va_flags != VNOVAL) {
400 		/* Disallow flags not supported by ext2fs. */
401 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
402 			return (EOPNOTSUPP);
403 
404 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
405 			return (EROFS);
406 		/*
407 		 * Callers may only modify the file flags on objects they
408 		 * have VADMIN rights for.
409 		 */
410 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
411 			return (error);
412 		/*
413 		 * Unprivileged processes and privileged processes in
414 		 * jail() are not permitted to unset system flags, or
415 		 * modify flags if any system flags are set.
416 		 * Privileged non-jail processes may not modify system flags
417 		 * if securelevel > 0 and any existing system flags are set.
418 		 */
419 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
420 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
421 				error = securelevel_gt(cred, 0);
422 				if (error)
423 					return (error);
424 			}
425 		} else {
426 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
427 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
428 				return (EPERM);
429 		}
430 		ip->i_flags = vap->va_flags;
431 		ip->i_flag |= IN_CHANGE;
432 		if (ip->i_flags & (IMMUTABLE | APPEND))
433 			return (0);
434 	}
435 	if (ip->i_flags & (IMMUTABLE | APPEND))
436 		return (EPERM);
437 	/*
438 	 * Go through the fields and update iff not VNOVAL.
439 	 */
440 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
441 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
442 			return (EROFS);
443 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
444 		    td)) != 0)
445 			return (error);
446 	}
447 	if (vap->va_size != VNOVAL) {
448 		/*
449 		 * Disallow write attempts on read-only file systems;
450 		 * unless the file is a socket, fifo, or a block or
451 		 * character device resident on the file system.
452 		 */
453 		switch (vp->v_type) {
454 		case VDIR:
455 			return (EISDIR);
456 		case VLNK:
457 		case VREG:
458 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
459 				return (EROFS);
460 			break;
461 		default:
462 			break;
463 		}
464 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
465 			return (error);
466 	}
467 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
468 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
469 			return (EROFS);
470 		/*
471 		 * From utimes(2):
472 		 * If times is NULL, ... The caller must be the owner of
473 		 * the file, have permission to write the file, or be the
474 		 * super-user.
475 		 * If times is non-NULL, ... The caller must be the owner of
476 		 * the file or be the super-user.
477 		 */
478 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
479 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
480 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
481 			return (error);
482 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
483 		if (vap->va_atime.tv_sec != VNOVAL) {
484 			ip->i_flag &= ~IN_ACCESS;
485 			ip->i_atime = vap->va_atime.tv_sec;
486 			ip->i_atimensec = vap->va_atime.tv_nsec;
487 		}
488 		if (vap->va_mtime.tv_sec != VNOVAL) {
489 			ip->i_flag &= ~IN_UPDATE;
490 			ip->i_mtime = vap->va_mtime.tv_sec;
491 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
492 		}
493 		ip->i_birthtime = vap->va_birthtime.tv_sec;
494 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
495 		error = ext2_update(vp, 0);
496 		if (error)
497 			return (error);
498 	}
499 	error = 0;
500 	if (vap->va_mode != (mode_t)VNOVAL) {
501 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
502 			return (EROFS);
503 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
504 	}
505 	return (error);
506 }
507 
508 /*
509  * Change the mode on a file.
510  * Inode must be locked before calling.
511  */
512 static int
513 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
514 {
515 	struct inode *ip = VTOI(vp);
516 	int error;
517 
518 	/*
519 	 * To modify the permissions on a file, must possess VADMIN
520 	 * for that file.
521 	 */
522 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
523 		return (error);
524 	/*
525 	 * Privileged processes may set the sticky bit on non-directories,
526 	 * as well as set the setgid bit on a file with a group that the
527 	 * process is not a member of.
528 	 */
529 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
530 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
531 		if (error)
532 			return (EFTYPE);
533 	}
534 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
535 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
536 		if (error)
537 			return (error);
538 	}
539 	ip->i_mode &= ~ALLPERMS;
540 	ip->i_mode |= (mode & ALLPERMS);
541 	ip->i_flag |= IN_CHANGE;
542 	return (0);
543 }
544 
545 /*
546  * Perform chown operation on inode ip;
547  * inode must be locked prior to call.
548  */
549 static int
550 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
551     struct thread *td)
552 {
553 	struct inode *ip = VTOI(vp);
554 	uid_t ouid;
555 	gid_t ogid;
556 	int error = 0;
557 
558 	if (uid == (uid_t)VNOVAL)
559 		uid = ip->i_uid;
560 	if (gid == (gid_t)VNOVAL)
561 		gid = ip->i_gid;
562 	/*
563 	 * To modify the ownership of a file, must possess VADMIN
564 	 * for that file.
565 	 */
566 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
567 		return (error);
568 	/*
569 	 * To change the owner of a file, or change the group of a file
570 	 * to a group of which we are not a member, the caller must
571 	 * have privilege.
572 	 */
573 	if (uid != ip->i_uid || (gid != ip->i_gid &&
574 	    !groupmember(gid, cred))) {
575 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
576 		if (error)
577 			return (error);
578 	}
579 	ogid = ip->i_gid;
580 	ouid = ip->i_uid;
581 	ip->i_gid = gid;
582 	ip->i_uid = uid;
583 	ip->i_flag |= IN_CHANGE;
584 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
585 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
586 			ip->i_mode &= ~(ISUID | ISGID);
587 	}
588 	return (0);
589 }
590 
591 /*
592  * Synch an open file.
593  */
594 /* ARGSUSED */
595 static int
596 ext2_fsync(struct vop_fsync_args *ap)
597 {
598 	/*
599 	 * Flush all dirty buffers associated with a vnode.
600 	 */
601 
602 	vop_stdfsync(ap);
603 
604 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
605 }
606 
607 /*
608  * Mknod vnode call
609  */
610 /* ARGSUSED */
611 static int
612 ext2_mknod(struct vop_mknod_args *ap)
613 {
614 	struct vattr *vap = ap->a_vap;
615 	struct vnode **vpp = ap->a_vpp;
616 	struct inode *ip;
617 	ino_t ino;
618 	int error;
619 
620 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
621 	    ap->a_dvp, vpp, ap->a_cnp);
622 	if (error)
623 		return (error);
624 	ip = VTOI(*vpp);
625 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
626 	if (vap->va_rdev != VNOVAL) {
627 		/*
628 		 * Want to be able to use this to make badblock
629 		 * inodes, so don't truncate the dev number.
630 		 */
631 		if (!(ip->i_flag & IN_E4EXTENTS))
632 			ip->i_rdev = vap->va_rdev;
633 	}
634 	/*
635 	 * Remove inode, then reload it through VFS_VGET so it is
636 	 * checked to see if it is an alias of an existing entry in
637 	 * the inode cache.	 XXX I don't believe this is necessary now.
638 	 */
639 	(*vpp)->v_type = VNON;
640 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
641 	vgone(*vpp);
642 	vput(*vpp);
643 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
644 	if (error) {
645 		*vpp = NULL;
646 		return (error);
647 	}
648 	return (0);
649 }
650 
651 static int
652 ext2_remove(struct vop_remove_args *ap)
653 {
654 	struct inode *ip;
655 	struct vnode *vp = ap->a_vp;
656 	struct vnode *dvp = ap->a_dvp;
657 	int error;
658 
659 	ip = VTOI(vp);
660 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
661 	    (VTOI(dvp)->i_flags & APPEND)) {
662 		error = EPERM;
663 		goto out;
664 	}
665 	error = ext2_dirremove(dvp, ap->a_cnp);
666 	if (error == 0) {
667 		ip->i_nlink--;
668 		ip->i_flag |= IN_CHANGE;
669 	}
670 out:
671 	return (error);
672 }
673 
674 static unsigned short
675 ext2_max_nlink(struct inode *ip)
676 {
677 	struct m_ext2fs *fs;
678 
679 	fs = ip->i_e2fs;
680 
681 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_DIR_NLINK))
682 		return (EXT4_LINK_MAX);
683 	else
684 		return (EXT2_LINK_MAX);
685 }
686 
687 /*
688  * link vnode call
689  */
690 static int
691 ext2_link(struct vop_link_args *ap)
692 {
693 	struct vnode *vp = ap->a_vp;
694 	struct vnode *tdvp = ap->a_tdvp;
695 	struct componentname *cnp = ap->a_cnp;
696 	struct inode *ip;
697 	int error;
698 
699 #ifdef INVARIANTS
700 	if ((cnp->cn_flags & HASBUF) == 0)
701 		panic("ext2_link: no name");
702 #endif
703 	ip = VTOI(vp);
704 	if ((nlink_t)ip->i_nlink >= ext2_max_nlink(ip)) {
705 		error = EMLINK;
706 		goto out;
707 	}
708 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
709 		error = EPERM;
710 		goto out;
711 	}
712 	ip->i_nlink++;
713 	ip->i_flag |= IN_CHANGE;
714 	error = ext2_update(vp, !DOINGASYNC(vp));
715 	if (!error)
716 		error = ext2_direnter(ip, tdvp, cnp);
717 	if (error) {
718 		ip->i_nlink--;
719 		ip->i_flag |= IN_CHANGE;
720 	}
721 out:
722 	return (error);
723 }
724 
725 static int
726 ext2_inc_nlink(struct inode *ip)
727 {
728 
729 	ip->i_nlink++;
730 
731 	if (ext2_htree_has_idx(ip) && ip->i_nlink > 1) {
732 		if (ip->i_nlink >= ext2_max_nlink(ip) || ip->i_nlink == 2)
733 			ip->i_nlink = 1;
734 	} else if (ip->i_nlink > ext2_max_nlink(ip)) {
735 		ip->i_nlink--;
736 		return (EMLINK);
737 	}
738 
739 	return (0);
740 }
741 
742 static void
743 ext2_dec_nlink(struct inode *ip)
744 {
745 
746 	if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
747 		ip->i_nlink--;
748 }
749 
750 /*
751  * Rename system call.
752  * 	rename("foo", "bar");
753  * is essentially
754  *	unlink("bar");
755  *	link("foo", "bar");
756  *	unlink("foo");
757  * but ``atomically''.  Can't do full commit without saving state in the
758  * inode on disk which isn't feasible at this time.  Best we can do is
759  * always guarantee the target exists.
760  *
761  * Basic algorithm is:
762  *
763  * 1) Bump link count on source while we're linking it to the
764  *    target.  This also ensure the inode won't be deleted out
765  *    from underneath us while we work (it may be truncated by
766  *    a concurrent `trunc' or `open' for creation).
767  * 2) Link source to destination.  If destination already exists,
768  *    delete it first.
769  * 3) Unlink source reference to inode if still around. If a
770  *    directory was moved and the parent of the destination
771  *    is different from the source, patch the ".." entry in the
772  *    directory.
773  */
774 static int
775 ext2_rename(struct vop_rename_args *ap)
776 {
777 	struct vnode *tvp = ap->a_tvp;
778 	struct vnode *tdvp = ap->a_tdvp;
779 	struct vnode *fvp = ap->a_fvp;
780 	struct vnode *fdvp = ap->a_fdvp;
781 	struct componentname *tcnp = ap->a_tcnp;
782 	struct componentname *fcnp = ap->a_fcnp;
783 	struct inode *ip, *xp, *dp;
784 	struct dirtemplate dirbuf;
785 	int doingdirectory = 0, oldparent = 0, newparent = 0;
786 	int error = 0;
787 	u_char namlen;
788 
789 #ifdef INVARIANTS
790 	if ((tcnp->cn_flags & HASBUF) == 0 ||
791 	    (fcnp->cn_flags & HASBUF) == 0)
792 		panic("ext2_rename: no name");
793 #endif
794 	/*
795 	 * Check for cross-device rename.
796 	 */
797 	if ((fvp->v_mount != tdvp->v_mount) ||
798 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
799 		error = EXDEV;
800 abortit:
801 		if (tdvp == tvp)
802 			vrele(tdvp);
803 		else
804 			vput(tdvp);
805 		if (tvp)
806 			vput(tvp);
807 		vrele(fdvp);
808 		vrele(fvp);
809 		return (error);
810 	}
811 
812 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
813 	    (VTOI(tdvp)->i_flags & APPEND))) {
814 		error = EPERM;
815 		goto abortit;
816 	}
817 
818 	/*
819 	 * Renaming a file to itself has no effect.  The upper layers should
820 	 * not call us in that case.  Temporarily just warn if they do.
821 	 */
822 	if (fvp == tvp) {
823 		printf("ext2_rename: fvp == tvp (can't happen)\n");
824 		error = 0;
825 		goto abortit;
826 	}
827 
828 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
829 		goto abortit;
830 	dp = VTOI(fdvp);
831 	ip = VTOI(fvp);
832 	if (ip->i_nlink >= ext2_max_nlink(ip) && !ext2_htree_has_idx(ip)) {
833 		VOP_UNLOCK(fvp, 0);
834 		error = EMLINK;
835 		goto abortit;
836 	}
837 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
838 	    || (dp->i_flags & APPEND)) {
839 		VOP_UNLOCK(fvp, 0);
840 		error = EPERM;
841 		goto abortit;
842 	}
843 	if ((ip->i_mode & IFMT) == IFDIR) {
844 		/*
845 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
846 		 */
847 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
848 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
849 		    (ip->i_flag & IN_RENAME)) {
850 			VOP_UNLOCK(fvp, 0);
851 			error = EINVAL;
852 			goto abortit;
853 		}
854 		ip->i_flag |= IN_RENAME;
855 		oldparent = dp->i_number;
856 		doingdirectory++;
857 	}
858 	vrele(fdvp);
859 
860 	/*
861 	 * When the target exists, both the directory
862 	 * and target vnodes are returned locked.
863 	 */
864 	dp = VTOI(tdvp);
865 	xp = NULL;
866 	if (tvp)
867 		xp = VTOI(tvp);
868 
869 	/*
870 	 * 1) Bump link count while we're moving stuff
871 	 *    around.  If we crash somewhere before
872 	 *    completing our work, the link count
873 	 *    may be wrong, but correctable.
874 	 */
875 	ext2_inc_nlink(ip);
876 	ip->i_flag |= IN_CHANGE;
877 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
878 		VOP_UNLOCK(fvp, 0);
879 		goto bad;
880 	}
881 
882 	/*
883 	 * If ".." must be changed (ie the directory gets a new
884 	 * parent) then the source directory must not be in the
885 	 * directory hierarchy above the target, as this would
886 	 * orphan everything below the source directory. Also
887 	 * the user must have write permission in the source so
888 	 * as to be able to change "..". We must repeat the call
889 	 * to namei, as the parent directory is unlocked by the
890 	 * call to checkpath().
891 	 */
892 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
893 	VOP_UNLOCK(fvp, 0);
894 	if (oldparent != dp->i_number)
895 		newparent = dp->i_number;
896 	if (doingdirectory && newparent) {
897 		if (error)	/* write access check above */
898 			goto bad;
899 		if (xp != NULL)
900 			vput(tvp);
901 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
902 		if (error)
903 			goto out;
904 		VREF(tdvp);
905 		error = relookup(tdvp, &tvp, tcnp);
906 		if (error)
907 			goto out;
908 		vrele(tdvp);
909 		dp = VTOI(tdvp);
910 		xp = NULL;
911 		if (tvp)
912 			xp = VTOI(tvp);
913 	}
914 	/*
915 	 * 2) If target doesn't exist, link the target
916 	 *    to the source and unlink the source.
917 	 *    Otherwise, rewrite the target directory
918 	 *    entry to reference the source inode and
919 	 *    expunge the original entry's existence.
920 	 */
921 	if (xp == NULL) {
922 		if (dp->i_devvp != ip->i_devvp)
923 			panic("ext2_rename: EXDEV");
924 		/*
925 		 * Account for ".." in new directory.
926 		 * When source and destination have the same
927 		 * parent we don't fool with the link count.
928 		 */
929 		if (doingdirectory && newparent) {
930 			error = ext2_inc_nlink(dp);
931 			if (error)
932 				goto bad;
933 
934 			dp->i_flag |= IN_CHANGE;
935 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
936 			if (error)
937 				goto bad;
938 		}
939 		error = ext2_direnter(ip, tdvp, tcnp);
940 		if (error) {
941 			if (doingdirectory && newparent) {
942 				ext2_dec_nlink(dp);
943 				dp->i_flag |= IN_CHANGE;
944 				(void)ext2_update(tdvp, 1);
945 			}
946 			goto bad;
947 		}
948 		vput(tdvp);
949 	} else {
950 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
951 			panic("ext2_rename: EXDEV");
952 		/*
953 		 * Short circuit rename(foo, foo).
954 		 */
955 		if (xp->i_number == ip->i_number)
956 			panic("ext2_rename: same file");
957 		/*
958 		 * If the parent directory is "sticky", then the user must
959 		 * own the parent directory, or the destination of the rename,
960 		 * otherwise the destination may not be changed (except by
961 		 * root). This implements append-only directories.
962 		 */
963 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
964 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
965 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
966 			error = EPERM;
967 			goto bad;
968 		}
969 		/*
970 		 * Target must be empty if a directory and have no links
971 		 * to it. Also, ensure source and target are compatible
972 		 * (both directories, or both not directories).
973 		 */
974 		if ((xp->i_mode & IFMT) == IFDIR) {
975 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
976 				error = ENOTEMPTY;
977 				goto bad;
978 			}
979 			if (!doingdirectory) {
980 				error = ENOTDIR;
981 				goto bad;
982 			}
983 			cache_purge(tdvp);
984 		} else if (doingdirectory) {
985 			error = EISDIR;
986 			goto bad;
987 		}
988 		error = ext2_dirrewrite(dp, ip, tcnp);
989 		if (error)
990 			goto bad;
991 		/*
992 		 * If the target directory is in the same
993 		 * directory as the source directory,
994 		 * decrement the link count on the parent
995 		 * of the target directory.
996 		 */
997 		if (doingdirectory && !newparent) {
998 			ext2_dec_nlink(dp);
999 			dp->i_flag |= IN_CHANGE;
1000 		}
1001 		vput(tdvp);
1002 		/*
1003 		 * Adjust the link count of the target to
1004 		 * reflect the dirrewrite above.  If this is
1005 		 * a directory it is empty and there are
1006 		 * no links to it, so we can squash the inode and
1007 		 * any space associated with it.  We disallowed
1008 		 * renaming over top of a directory with links to
1009 		 * it above, as the remaining link would point to
1010 		 * a directory without "." or ".." entries.
1011 		 */
1012 		ext2_dec_nlink(xp);
1013 		if (doingdirectory) {
1014 			if (--xp->i_nlink != 0)
1015 				panic("ext2_rename: linked directory");
1016 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1017 			    tcnp->cn_cred, tcnp->cn_thread);
1018 		}
1019 		xp->i_flag |= IN_CHANGE;
1020 		vput(tvp);
1021 		xp = NULL;
1022 	}
1023 
1024 	/*
1025 	 * 3) Unlink the source.
1026 	 */
1027 	fcnp->cn_flags &= ~MODMASK;
1028 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1029 	VREF(fdvp);
1030 	error = relookup(fdvp, &fvp, fcnp);
1031 	if (error == 0)
1032 		vrele(fdvp);
1033 	if (fvp != NULL) {
1034 		xp = VTOI(fvp);
1035 		dp = VTOI(fdvp);
1036 	} else {
1037 		/*
1038 		 * From name has disappeared.  IN_RENAME is not sufficient
1039 		 * to protect against directory races due to timing windows,
1040 		 * so we can't panic here.
1041 		 */
1042 		vrele(ap->a_fvp);
1043 		return (0);
1044 	}
1045 	/*
1046 	 * Ensure that the directory entry still exists and has not
1047 	 * changed while the new name has been entered. If the source is
1048 	 * a file then the entry may have been unlinked or renamed. In
1049 	 * either case there is no further work to be done. If the source
1050 	 * is a directory then it cannot have been rmdir'ed; its link
1051 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1052 	 * The IN_RENAME flag ensures that it cannot be moved by another
1053 	 * rename.
1054 	 */
1055 	if (xp != ip) {
1056 		/*
1057 		 * From name resolves to a different inode.  IN_RENAME is
1058 		 * not sufficient protection against timing window races
1059 		 * so we can't panic here.
1060 		 */
1061 	} else {
1062 		/*
1063 		 * If the source is a directory with a
1064 		 * new parent, the link count of the old
1065 		 * parent directory must be decremented
1066 		 * and ".." set to point to the new parent.
1067 		 */
1068 		if (doingdirectory && newparent) {
1069 			ext2_dec_nlink(dp);
1070 			dp->i_flag |= IN_CHANGE;
1071 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1072 			    sizeof(struct dirtemplate), (off_t)0,
1073 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1074 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1075 			if (error == 0) {
1076 				/* Like ufs little-endian: */
1077 				namlen = dirbuf.dotdot_type;
1078 				if (namlen != 2 ||
1079 				    dirbuf.dotdot_name[0] != '.' ||
1080 				    dirbuf.dotdot_name[1] != '.') {
1081 					ext2_dirbad(xp, (doff_t)12,
1082 					    "rename: mangled dir");
1083 				} else {
1084 					dirbuf.dotdot_ino = newparent;
1085 					(void)vn_rdwr(UIO_WRITE, fvp,
1086 					    (caddr_t)&dirbuf,
1087 					    sizeof(struct dirtemplate),
1088 					    (off_t)0, UIO_SYSSPACE,
1089 					    IO_NODELOCKED | IO_SYNC |
1090 					    IO_NOMACCHECK, tcnp->cn_cred,
1091 					    NOCRED, NULL, NULL);
1092 					cache_purge(fdvp);
1093 				}
1094 			}
1095 		}
1096 		error = ext2_dirremove(fdvp, fcnp);
1097 		if (!error) {
1098 			ext2_dec_nlink(xp);
1099 			xp->i_flag |= IN_CHANGE;
1100 		}
1101 		xp->i_flag &= ~IN_RENAME;
1102 	}
1103 	if (dp)
1104 		vput(fdvp);
1105 	if (xp)
1106 		vput(fvp);
1107 	vrele(ap->a_fvp);
1108 	return (error);
1109 
1110 bad:
1111 	if (xp)
1112 		vput(ITOV(xp));
1113 	vput(ITOV(dp));
1114 out:
1115 	if (doingdirectory)
1116 		ip->i_flag &= ~IN_RENAME;
1117 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1118 		ext2_dec_nlink(ip);
1119 		ip->i_flag |= IN_CHANGE;
1120 		ip->i_flag &= ~IN_RENAME;
1121 		vput(fvp);
1122 	} else
1123 		vrele(fvp);
1124 	return (error);
1125 }
1126 
1127 #ifdef UFS_ACL
1128 static int
1129 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1130     mode_t dmode, struct ucred *cred, struct thread *td)
1131 {
1132 	int error;
1133 	struct inode *ip = VTOI(tvp);
1134 	struct acl *dacl, *acl;
1135 
1136 	acl = acl_alloc(M_WAITOK);
1137 	dacl = acl_alloc(M_WAITOK);
1138 
1139 	/*
1140 	 * Retrieve default ACL from parent, if any.
1141 	 */
1142 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1143 	switch (error) {
1144 	case 0:
1145 		/*
1146 		 * Retrieved a default ACL, so merge mode and ACL if
1147 		 * necessary.  If the ACL is empty, fall through to
1148 		 * the "not defined or available" case.
1149 		 */
1150 		if (acl->acl_cnt != 0) {
1151 			dmode = acl_posix1e_newfilemode(dmode, acl);
1152 			ip->i_mode = dmode;
1153 			*dacl = *acl;
1154 			ext2_sync_acl_from_inode(ip, acl);
1155 			break;
1156 		}
1157 		/* FALLTHROUGH */
1158 
1159 	case EOPNOTSUPP:
1160 		/*
1161 		 * Just use the mode as-is.
1162 		 */
1163 		ip->i_mode = dmode;
1164 		error = 0;
1165 		goto out;
1166 
1167 	default:
1168 		goto out;
1169 	}
1170 
1171 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1172 	if (error == 0)
1173 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1174 	switch (error) {
1175 	case 0:
1176 		break;
1177 
1178 	case EOPNOTSUPP:
1179 		/*
1180 		 * XXX: This should not happen, as EOPNOTSUPP above
1181 		 * was supposed to free acl.
1182 		 */
1183 #ifdef DEBUG
1184 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1185 #endif	/* DEBUG */
1186 		break;
1187 
1188 	default:
1189 		goto out;
1190 	}
1191 
1192 out:
1193 	acl_free(acl);
1194 	acl_free(dacl);
1195 
1196 	return (error);
1197 }
1198 
1199 static int
1200 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1201     mode_t mode, struct ucred *cred, struct thread *td)
1202 {
1203 	int error;
1204 	struct inode *ip = VTOI(tvp);
1205 	struct acl *acl;
1206 
1207 	acl = acl_alloc(M_WAITOK);
1208 
1209 	/*
1210 	 * Retrieve default ACL for parent, if any.
1211 	 */
1212 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1213 	switch (error) {
1214 	case 0:
1215 		/*
1216 		 * Retrieved a default ACL, so merge mode and ACL if
1217 		 * necessary.
1218 		 */
1219 		if (acl->acl_cnt != 0) {
1220 			/*
1221 			 * Two possible ways for default ACL to not
1222 			 * be present.  First, the EA can be
1223 			 * undefined, or second, the default ACL can
1224 			 * be blank.  If it's blank, fall through to
1225 			 * the it's not defined case.
1226 			 */
1227 			mode = acl_posix1e_newfilemode(mode, acl);
1228 			ip->i_mode = mode;
1229 			ext2_sync_acl_from_inode(ip, acl);
1230 			break;
1231 		}
1232 		/* FALLTHROUGH */
1233 
1234 	case EOPNOTSUPP:
1235 		/*
1236 		 * Just use the mode as-is.
1237 		 */
1238 		ip->i_mode = mode;
1239 		error = 0;
1240 		goto out;
1241 
1242 	default:
1243 		goto out;
1244 	}
1245 
1246 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1247 	switch (error) {
1248 	case 0:
1249 		break;
1250 
1251 	case EOPNOTSUPP:
1252 		/*
1253 		 * XXX: This should not happen, as EOPNOTSUPP above was
1254 		 * supposed to free acl.
1255 		 */
1256 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1257 		    "but no VOP_SETACL()\n");
1258 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1259 		    "but no VOP_SETACL()"); */
1260 		break;
1261 
1262 	default:
1263 		goto out;
1264 	}
1265 
1266 out:
1267 	acl_free(acl);
1268 
1269 	return (error);
1270 }
1271 
1272 #endif /* UFS_ACL */
1273 
1274 /*
1275  * Mkdir system call
1276  */
1277 static int
1278 ext2_mkdir(struct vop_mkdir_args *ap)
1279 {
1280 	struct vnode *dvp = ap->a_dvp;
1281 	struct vattr *vap = ap->a_vap;
1282 	struct componentname *cnp = ap->a_cnp;
1283 	struct inode *ip, *dp;
1284 	struct vnode *tvp;
1285 	struct dirtemplate dirtemplate, *dtp;
1286 	int error, dmode;
1287 
1288 #ifdef INVARIANTS
1289 	if ((cnp->cn_flags & HASBUF) == 0)
1290 		panic("ext2_mkdir: no name");
1291 #endif
1292 	dp = VTOI(dvp);
1293 	if ((nlink_t)dp->i_nlink >= ext2_max_nlink(dp) &&
1294 	    !ext2_htree_has_idx(dp)) {
1295 		error = EMLINK;
1296 		goto out;
1297 	}
1298 	dmode = vap->va_mode & 0777;
1299 	dmode |= IFDIR;
1300 	/*
1301 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1302 	 * but not have it entered in the parent directory. The entry is
1303 	 * made later after writing "." and ".." entries.
1304 	 */
1305 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1306 	if (error)
1307 		goto out;
1308 	ip = VTOI(tvp);
1309 	ip->i_gid = dp->i_gid;
1310 #ifdef SUIDDIR
1311 	{
1312 		/*
1313 		 * if we are hacking owners here, (only do this where told to)
1314 		 * and we are not giving it TOO root, (would subvert quotas)
1315 		 * then go ahead and give it to the other user.
1316 		 * The new directory also inherits the SUID bit.
1317 		 * If user's UID and dir UID are the same,
1318 		 * 'give it away' so that the SUID is still forced on.
1319 		 */
1320 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1321 		    (dp->i_mode & ISUID) && dp->i_uid) {
1322 			dmode |= ISUID;
1323 			ip->i_uid = dp->i_uid;
1324 		} else {
1325 			ip->i_uid = cnp->cn_cred->cr_uid;
1326 		}
1327 	}
1328 #else
1329 	ip->i_uid = cnp->cn_cred->cr_uid;
1330 #endif
1331 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1332 	ip->i_mode = dmode;
1333 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1334 	ip->i_nlink = 2;
1335 	if (cnp->cn_flags & ISWHITEOUT)
1336 		ip->i_flags |= UF_OPAQUE;
1337 	error = ext2_update(tvp, 1);
1338 
1339 	/*
1340 	 * Bump link count in parent directory
1341 	 * to reflect work done below.  Should
1342 	 * be done before reference is created
1343 	 * so reparation is possible if we crash.
1344 	 */
1345 	ext2_inc_nlink(dp);
1346 	dp->i_flag |= IN_CHANGE;
1347 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1348 	if (error)
1349 		goto bad;
1350 
1351 	/* Initialize directory with "." and ".." from static template. */
1352 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1353 	    EXT2F_INCOMPAT_FTYPE))
1354 		dtp = &mastertemplate;
1355 	else
1356 		dtp = &omastertemplate;
1357 	dirtemplate = *dtp;
1358 	dirtemplate.dot_ino = ip->i_number;
1359 	dirtemplate.dotdot_ino = dp->i_number;
1360 	/*
1361 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1362 	 * just redefine it - for this function only
1363 	 */
1364 #undef  DIRBLKSIZ
1365 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1366 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1367 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1368 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1369 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1370 	    NULL, NULL);
1371 	if (error) {
1372 		ext2_dec_nlink(dp);
1373 		dp->i_flag |= IN_CHANGE;
1374 		goto bad;
1375 	}
1376 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1377 		/* XXX should grow with balloc() */
1378 		panic("ext2_mkdir: blksize");
1379 	else {
1380 		ip->i_size = DIRBLKSIZ;
1381 		ip->i_flag |= IN_CHANGE;
1382 	}
1383 
1384 #ifdef UFS_ACL
1385 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1386 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1387 		    cnp->cn_cred, cnp->cn_thread);
1388 		if (error)
1389 			goto bad;
1390 	}
1391 
1392 #endif /* UFS_ACL */
1393 
1394 	/* Directory set up, now install its entry in the parent directory. */
1395 	error = ext2_direnter(ip, dvp, cnp);
1396 	if (error) {
1397 		ext2_dec_nlink(dp);
1398 		dp->i_flag |= IN_CHANGE;
1399 	}
1400 bad:
1401 	/*
1402 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1403 	 * for us because we set the link count to 0.
1404 	 */
1405 	if (error) {
1406 		ip->i_nlink = 0;
1407 		ip->i_flag |= IN_CHANGE;
1408 		vput(tvp);
1409 	} else
1410 		*ap->a_vpp = tvp;
1411 out:
1412 	return (error);
1413 #undef  DIRBLKSIZ
1414 #define DIRBLKSIZ  DEV_BSIZE
1415 }
1416 
1417 /*
1418  * Rmdir system call.
1419  */
1420 static int
1421 ext2_rmdir(struct vop_rmdir_args *ap)
1422 {
1423 	struct vnode *vp = ap->a_vp;
1424 	struct vnode *dvp = ap->a_dvp;
1425 	struct componentname *cnp = ap->a_cnp;
1426 	struct inode *ip, *dp;
1427 	int error;
1428 
1429 	ip = VTOI(vp);
1430 	dp = VTOI(dvp);
1431 
1432 	/*
1433 	 * Verify the directory is empty (and valid).
1434 	 * (Rmdir ".." won't be valid since
1435 	 *  ".." will contain a reference to
1436 	 *  the current directory and thus be
1437 	 *  non-empty.)
1438 	 */
1439 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1440 		error = ENOTEMPTY;
1441 		goto out;
1442 	}
1443 	if ((dp->i_flags & APPEND)
1444 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1445 		error = EPERM;
1446 		goto out;
1447 	}
1448 	/*
1449 	 * Delete reference to directory before purging
1450 	 * inode.  If we crash in between, the directory
1451 	 * will be reattached to lost+found,
1452 	 */
1453 	error = ext2_dirremove(dvp, cnp);
1454 	if (error)
1455 		goto out;
1456 	ext2_dec_nlink(dp);
1457 	dp->i_flag |= IN_CHANGE;
1458 	cache_purge(dvp);
1459 	VOP_UNLOCK(dvp, 0);
1460 	/*
1461 	 * Truncate inode.  The only stuff left
1462 	 * in the directory is "." and "..".
1463 	 */
1464 	ip->i_nlink = 0;
1465 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1466 	    cnp->cn_thread);
1467 	cache_purge(ITOV(ip));
1468 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1469 		VOP_UNLOCK(vp, 0);
1470 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1471 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1472 	}
1473 out:
1474 	return (error);
1475 }
1476 
1477 /*
1478  * symlink -- make a symbolic link
1479  */
1480 static int
1481 ext2_symlink(struct vop_symlink_args *ap)
1482 {
1483 	struct vnode *vp, **vpp = ap->a_vpp;
1484 	struct inode *ip;
1485 	int len, error;
1486 
1487 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1488 	    vpp, ap->a_cnp);
1489 	if (error)
1490 		return (error);
1491 	vp = *vpp;
1492 	len = strlen(ap->a_target);
1493 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1494 		ip = VTOI(vp);
1495 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1496 		ip->i_size = len;
1497 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1498 	} else
1499 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1500 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1501 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1502 	if (error)
1503 		vput(vp);
1504 	return (error);
1505 }
1506 
1507 /*
1508  * Return target name of a symbolic link
1509  */
1510 static int
1511 ext2_readlink(struct vop_readlink_args *ap)
1512 {
1513 	struct vnode *vp = ap->a_vp;
1514 	struct inode *ip = VTOI(vp);
1515 	int isize;
1516 
1517 	isize = ip->i_size;
1518 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1519 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1520 		return (0);
1521 	}
1522 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1523 }
1524 
1525 /*
1526  * Calculate the logical to physical mapping if not done already,
1527  * then call the device strategy routine.
1528  *
1529  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1530  * deadlock on memory.  See ext2_bmap() for details.
1531  */
1532 static int
1533 ext2_strategy(struct vop_strategy_args *ap)
1534 {
1535 	struct buf *bp = ap->a_bp;
1536 	struct vnode *vp = ap->a_vp;
1537 	struct bufobj *bo;
1538 	daddr_t blkno;
1539 	int error;
1540 
1541 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1542 		panic("ext2_strategy: spec");
1543 	if (bp->b_blkno == bp->b_lblkno) {
1544 
1545 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1546 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1547 		else
1548 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1549 
1550 		bp->b_blkno = blkno;
1551 		if (error) {
1552 			bp->b_error = error;
1553 			bp->b_ioflags |= BIO_ERROR;
1554 			bufdone(bp);
1555 			return (0);
1556 		}
1557 		if ((long)bp->b_blkno == -1)
1558 			vfs_bio_clrbuf(bp);
1559 	}
1560 	if ((long)bp->b_blkno == -1) {
1561 		bufdone(bp);
1562 		return (0);
1563 	}
1564 	bp->b_iooffset = dbtob(bp->b_blkno);
1565 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1566 	BO_STRATEGY(bo, bp);
1567 	return (0);
1568 }
1569 
1570 /*
1571  * Print out the contents of an inode.
1572  */
1573 static int
1574 ext2_print(struct vop_print_args *ap)
1575 {
1576 	struct vnode *vp = ap->a_vp;
1577 	struct inode *ip = VTOI(vp);
1578 
1579 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1580 	if (vp->v_type == VFIFO)
1581 		fifo_printinfo(vp);
1582 	printf("\n");
1583 	return (0);
1584 }
1585 
1586 /*
1587  * Close wrapper for fifos.
1588  *
1589  * Update the times on the inode then do device close.
1590  */
1591 static int
1592 ext2fifo_close(struct vop_close_args *ap)
1593 {
1594 	struct vnode *vp = ap->a_vp;
1595 
1596 	VI_LOCK(vp);
1597 	if (vp->v_usecount > 1)
1598 		ext2_itimes_locked(vp);
1599 	VI_UNLOCK(vp);
1600 	return (fifo_specops.vop_close(ap));
1601 }
1602 
1603 /*
1604  * Kqfilter wrapper for fifos.
1605  *
1606  * Fall through to ext2 kqfilter routines if needed
1607  */
1608 static int
1609 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1610 {
1611 	int error;
1612 
1613 	error = fifo_specops.vop_kqfilter(ap);
1614 	if (error)
1615 		error = vfs_kqfilter(ap);
1616 	return (error);
1617 }
1618 
1619 /*
1620  * Return POSIX pathconf information applicable to ext2 filesystems.
1621  */
1622 static int
1623 ext2_pathconf(struct vop_pathconf_args *ap)
1624 {
1625 	int error = 0;
1626 
1627 	switch (ap->a_name) {
1628 	case _PC_LINK_MAX:
1629 		if (ext2_htree_has_idx(VTOI(ap->a_vp)))
1630 			*ap->a_retval = INT_MAX;
1631 		else
1632 			*ap->a_retval = ext2_max_nlink(VTOI(ap->a_vp));
1633 		break;
1634 	case _PC_NO_TRUNC:
1635 		*ap->a_retval = 1;
1636 		break;
1637 
1638 #ifdef UFS_ACL
1639 	case _PC_ACL_EXTENDED:
1640 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1641 			*ap->a_retval = 1;
1642 		else
1643 			*ap->a_retval = 0;
1644 		break;
1645 	case _PC_ACL_PATH_MAX:
1646 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1647 			*ap->a_retval = ACL_MAX_ENTRIES;
1648 		else
1649 			*ap->a_retval = 3;
1650 		break;
1651 #endif /* UFS_ACL */
1652 
1653 	case _PC_MIN_HOLE_SIZE:
1654 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1655 		break;
1656 	case _PC_PRIO_IO:
1657 		*ap->a_retval = 0;
1658 		break;
1659 	case _PC_SYNC_IO:
1660 		*ap->a_retval = 0;
1661 		break;
1662 	case _PC_ALLOC_SIZE_MIN:
1663 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1664 		break;
1665 	case _PC_FILESIZEBITS:
1666 		*ap->a_retval = 64;
1667 		break;
1668 	case _PC_REC_INCR_XFER_SIZE:
1669 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1670 		break;
1671 	case _PC_REC_MAX_XFER_SIZE:
1672 		*ap->a_retval = -1;	/* means ``unlimited'' */
1673 		break;
1674 	case _PC_REC_MIN_XFER_SIZE:
1675 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1676 		break;
1677 	case _PC_REC_XFER_ALIGN:
1678 		*ap->a_retval = PAGE_SIZE;
1679 		break;
1680 	case _PC_SYMLINK_MAX:
1681 		*ap->a_retval = MAXPATHLEN;
1682 		break;
1683 
1684 	default:
1685 		error = vop_stdpathconf(ap);
1686 		break;
1687 	}
1688 	return (error);
1689 }
1690 
1691 /*
1692  * Vnode operation to remove a named attribute.
1693  */
1694 static int
1695 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1696 {
1697 	struct inode *ip;
1698 	struct m_ext2fs *fs;
1699 	int error;
1700 
1701 	ip = VTOI(ap->a_vp);
1702 	fs = ip->i_e2fs;
1703 
1704 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1705 		return (EOPNOTSUPP);
1706 
1707 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1708 		return (EOPNOTSUPP);
1709 
1710 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1711 	    ap->a_cred, ap->a_td, VWRITE);
1712 	if (error)
1713 		return (error);
1714 
1715 	error = ENOATTR;
1716 
1717 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1718 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1719 		if (error != ENOATTR)
1720 			return (error);
1721 	}
1722 
1723 	if (ip->i_facl)
1724 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1725 
1726 	return (error);
1727 }
1728 
1729 /*
1730  * Vnode operation to retrieve a named extended attribute.
1731  */
1732 static int
1733 ext2_getextattr(struct vop_getextattr_args *ap)
1734 {
1735 	struct inode *ip;
1736 	struct m_ext2fs *fs;
1737 	int error;
1738 
1739 	ip = VTOI(ap->a_vp);
1740 	fs = ip->i_e2fs;
1741 
1742 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1743 		return (EOPNOTSUPP);
1744 
1745 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1746 		return (EOPNOTSUPP);
1747 
1748 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1749 	    ap->a_cred, ap->a_td, VREAD);
1750 	if (error)
1751 		return (error);
1752 
1753 	if (ap->a_size != NULL)
1754 		*ap->a_size = 0;
1755 
1756 	error = ENOATTR;
1757 
1758 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1759 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1760 		    ap->a_name, ap->a_uio, ap->a_size);
1761 		if (error != ENOATTR)
1762 			return (error);
1763 	}
1764 
1765 	if (ip->i_facl)
1766 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1767 		    ap->a_name, ap->a_uio, ap->a_size);
1768 
1769 	return (error);
1770 }
1771 
1772 /*
1773  * Vnode operation to retrieve extended attributes on a vnode.
1774  */
1775 static int
1776 ext2_listextattr(struct vop_listextattr_args *ap)
1777 {
1778 	struct inode *ip;
1779 	struct m_ext2fs *fs;
1780 	int error;
1781 
1782 	ip = VTOI(ap->a_vp);
1783 	fs = ip->i_e2fs;
1784 
1785 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1786 		return (EOPNOTSUPP);
1787 
1788 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1789 		return (EOPNOTSUPP);
1790 
1791 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1792 	    ap->a_cred, ap->a_td, VREAD);
1793 	if (error)
1794 		return (error);
1795 
1796 	if (ap->a_size != NULL)
1797 		*ap->a_size = 0;
1798 
1799 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1800 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1801 		    ap->a_uio, ap->a_size);
1802 		if (error)
1803 			return (error);
1804 	}
1805 
1806 	if (ip->i_facl)
1807 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1808 		    ap->a_uio, ap->a_size);
1809 
1810 	return (error);
1811 }
1812 
1813 /*
1814  * Vnode operation to set a named attribute.
1815  */
1816 static int
1817 ext2_setextattr(struct vop_setextattr_args *ap)
1818 {
1819 	struct inode *ip;
1820 	struct m_ext2fs *fs;
1821 	int error;
1822 
1823 	ip = VTOI(ap->a_vp);
1824 	fs = ip->i_e2fs;
1825 
1826 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1827 		return (EOPNOTSUPP);
1828 
1829 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1830 		return (EOPNOTSUPP);
1831 
1832 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1833 	    ap->a_cred, ap->a_td, VWRITE);
1834 	if (error)
1835 		return (error);
1836 
1837 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1838 	if (error)
1839 		return (error);
1840 
1841 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1842 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1843 		    ap->a_name, ap->a_uio);
1844 		if (error != ENOSPC)
1845 			return (error);
1846 	}
1847 
1848 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1849 	    ap->a_name, ap->a_uio);
1850 
1851 	return (error);
1852 }
1853 
1854 /*
1855  * Vnode pointer to File handle
1856  */
1857 /* ARGSUSED */
1858 static int
1859 ext2_vptofh(struct vop_vptofh_args *ap)
1860 {
1861 	struct inode *ip;
1862 	struct ufid *ufhp;
1863 
1864 	ip = VTOI(ap->a_vp);
1865 	ufhp = (struct ufid *)ap->a_fhp;
1866 	ufhp->ufid_len = sizeof(struct ufid);
1867 	ufhp->ufid_ino = ip->i_number;
1868 	ufhp->ufid_gen = ip->i_gen;
1869 	return (0);
1870 }
1871 
1872 /*
1873  * Initialize the vnode associated with a new inode, handle aliased
1874  * vnodes.
1875  */
1876 int
1877 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1878 {
1879 	struct inode *ip;
1880 	struct vnode *vp;
1881 
1882 	vp = *vpp;
1883 	ip = VTOI(vp);
1884 	vp->v_type = IFTOVT(ip->i_mode);
1885 	if (vp->v_type == VFIFO)
1886 		vp->v_op = fifoops;
1887 
1888 	if (ip->i_number == EXT2_ROOTINO)
1889 		vp->v_vflag |= VV_ROOT;
1890 	ip->i_modrev = init_va_filerev();
1891 	*vpp = vp;
1892 	return (0);
1893 }
1894 
1895 /*
1896  * Allocate a new inode.
1897  */
1898 static int
1899 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1900     struct componentname *cnp)
1901 {
1902 	struct inode *ip, *pdir;
1903 	struct vnode *tvp;
1904 	int error;
1905 
1906 	pdir = VTOI(dvp);
1907 #ifdef INVARIANTS
1908 	if ((cnp->cn_flags & HASBUF) == 0)
1909 		panic("ext2_makeinode: no name");
1910 #endif
1911 	*vpp = NULL;
1912 	if ((mode & IFMT) == 0)
1913 		mode |= IFREG;
1914 
1915 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1916 	if (error) {
1917 		return (error);
1918 	}
1919 	ip = VTOI(tvp);
1920 	ip->i_gid = pdir->i_gid;
1921 #ifdef SUIDDIR
1922 	{
1923 		/*
1924 		 * if we are
1925 		 * not the owner of the directory,
1926 		 * and we are hacking owners here, (only do this where told to)
1927 		 * and we are not giving it TOO root, (would subvert quotas)
1928 		 * then go ahead and give it to the other user.
1929 		 * Note that this drops off the execute bits for security.
1930 		 */
1931 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1932 		    (pdir->i_mode & ISUID) &&
1933 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1934 			ip->i_uid = pdir->i_uid;
1935 			mode &= ~07111;
1936 		} else {
1937 			ip->i_uid = cnp->cn_cred->cr_uid;
1938 		}
1939 	}
1940 #else
1941 	ip->i_uid = cnp->cn_cred->cr_uid;
1942 #endif
1943 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1944 	ip->i_mode = mode;
1945 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1946 	ip->i_nlink = 1;
1947 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1948 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1949 			ip->i_mode &= ~ISGID;
1950 	}
1951 
1952 	if (cnp->cn_flags & ISWHITEOUT)
1953 		ip->i_flags |= UF_OPAQUE;
1954 
1955 	/*
1956 	 * Make sure inode goes to disk before directory entry.
1957 	 */
1958 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1959 	if (error)
1960 		goto bad;
1961 
1962 #ifdef UFS_ACL
1963 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1964 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
1965 		    cnp->cn_cred, cnp->cn_thread);
1966 		if (error)
1967 			goto bad;
1968 	}
1969 #endif /* UFS_ACL */
1970 
1971 	error = ext2_direnter(ip, dvp, cnp);
1972 	if (error)
1973 		goto bad;
1974 
1975 	*vpp = tvp;
1976 	return (0);
1977 
1978 bad:
1979 	/*
1980 	 * Write error occurred trying to update the inode
1981 	 * or the directory so must deallocate the inode.
1982 	 */
1983 	ip->i_nlink = 0;
1984 	ip->i_flag |= IN_CHANGE;
1985 	vput(tvp);
1986 	return (error);
1987 }
1988 
1989 /*
1990  * Vnode op for reading.
1991  */
1992 static int
1993 ext2_read(struct vop_read_args *ap)
1994 {
1995 	struct vnode *vp;
1996 	struct inode *ip;
1997 	struct uio *uio;
1998 	struct m_ext2fs *fs;
1999 	struct buf *bp;
2000 	daddr_t lbn, nextlbn;
2001 	off_t bytesinfile;
2002 	long size, xfersize, blkoffset;
2003 	int error, orig_resid, seqcount;
2004 	int ioflag;
2005 
2006 	vp = ap->a_vp;
2007 	uio = ap->a_uio;
2008 	ioflag = ap->a_ioflag;
2009 
2010 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2011 	ip = VTOI(vp);
2012 
2013 #ifdef INVARIANTS
2014 	if (uio->uio_rw != UIO_READ)
2015 		panic("%s: mode", "ext2_read");
2016 
2017 	if (vp->v_type == VLNK) {
2018 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
2019 			panic("%s: short symlink", "ext2_read");
2020 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2021 		panic("%s: type %d", "ext2_read", vp->v_type);
2022 #endif
2023 	orig_resid = uio->uio_resid;
2024 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2025 	if (orig_resid == 0)
2026 		return (0);
2027 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2028 	fs = ip->i_e2fs;
2029 	if (uio->uio_offset < ip->i_size &&
2030 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2031 		return (EOVERFLOW);
2032 
2033 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2034 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2035 			break;
2036 		lbn = lblkno(fs, uio->uio_offset);
2037 		nextlbn = lbn + 1;
2038 		size = blksize(fs, ip, lbn);
2039 		blkoffset = blkoff(fs, uio->uio_offset);
2040 
2041 		xfersize = fs->e2fs_fsize - blkoffset;
2042 		if (uio->uio_resid < xfersize)
2043 			xfersize = uio->uio_resid;
2044 		if (bytesinfile < xfersize)
2045 			xfersize = bytesinfile;
2046 
2047 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2048 			error = bread(vp, lbn, size, NOCRED, &bp);
2049 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2050 			error = cluster_read(vp, ip->i_size, lbn, size,
2051 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2052 			    0, &bp);
2053 		} else if (seqcount > 1) {
2054 			u_int nextsize = blksize(fs, ip, nextlbn);
2055 
2056 			error = breadn(vp, lbn,
2057 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2058 		} else
2059 			error = bread(vp, lbn, size, NOCRED, &bp);
2060 		if (error) {
2061 			brelse(bp);
2062 			bp = NULL;
2063 			break;
2064 		}
2065 
2066 		/*
2067 		 * We should only get non-zero b_resid when an I/O error
2068 		 * has occurred, which should cause us to break above.
2069 		 * However, if the short read did not cause an error,
2070 		 * then we want to ensure that we do not uiomove bad
2071 		 * or uninitialized data.
2072 		 */
2073 		size -= bp->b_resid;
2074 		if (size < xfersize) {
2075 			if (size == 0)
2076 				break;
2077 			xfersize = size;
2078 		}
2079 		error = uiomove((char *)bp->b_data + blkoffset,
2080 		    (int)xfersize, uio);
2081 		if (error)
2082 			break;
2083 		vfs_bio_brelse(bp, ioflag);
2084 	}
2085 
2086 	/*
2087 	 * This can only happen in the case of an error because the loop
2088 	 * above resets bp to NULL on each iteration and on normal
2089 	 * completion has not set a new value into it. so it must have come
2090 	 * from a 'break' statement
2091 	 */
2092 	if (bp != NULL)
2093 		vfs_bio_brelse(bp, ioflag);
2094 
2095 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2096 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2097 		ip->i_flag |= IN_ACCESS;
2098 	return (error);
2099 }
2100 
2101 static int
2102 ext2_ioctl(struct vop_ioctl_args *ap)
2103 {
2104 
2105 	switch (ap->a_command) {
2106 	case FIOSEEKDATA:
2107 	case FIOSEEKHOLE:
2108 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
2109 		    (off_t *)ap->a_data, ap->a_cred));
2110 	default:
2111 		return (ENOTTY);
2112 	}
2113 }
2114 
2115 /*
2116  * Vnode op for writing.
2117  */
2118 static int
2119 ext2_write(struct vop_write_args *ap)
2120 {
2121 	struct vnode *vp;
2122 	struct uio *uio;
2123 	struct inode *ip;
2124 	struct m_ext2fs *fs;
2125 	struct buf *bp;
2126 	daddr_t lbn;
2127 	off_t osize;
2128 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2129 
2130 	ioflag = ap->a_ioflag;
2131 	uio = ap->a_uio;
2132 	vp = ap->a_vp;
2133 
2134 	seqcount = ioflag >> IO_SEQSHIFT;
2135 	ip = VTOI(vp);
2136 
2137 #ifdef INVARIANTS
2138 	if (uio->uio_rw != UIO_WRITE)
2139 		panic("%s: mode", "ext2_write");
2140 #endif
2141 
2142 	switch (vp->v_type) {
2143 	case VREG:
2144 		if (ioflag & IO_APPEND)
2145 			uio->uio_offset = ip->i_size;
2146 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2147 			return (EPERM);
2148 		/* FALLTHROUGH */
2149 	case VLNK:
2150 		break;
2151 	case VDIR:
2152 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2153 		if ((ioflag & IO_SYNC) == 0)
2154 			panic("ext2_write: nonsync dir write");
2155 		break;
2156 	default:
2157 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2158 		    vp->v_type, (intmax_t)uio->uio_offset,
2159 		    (intmax_t)uio->uio_resid);
2160 	}
2161 
2162 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2163 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2164 	fs = ip->i_e2fs;
2165 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2166 		return (EFBIG);
2167 	/*
2168 	 * Maybe this should be above the vnode op call, but so long as
2169 	 * file servers have no limits, I don't think it matters.
2170 	 */
2171 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2172 		return (EFBIG);
2173 
2174 	resid = uio->uio_resid;
2175 	osize = ip->i_size;
2176 	if (seqcount > BA_SEQMAX)
2177 		flags = BA_SEQMAX << BA_SEQSHIFT;
2178 	else
2179 		flags = seqcount << BA_SEQSHIFT;
2180 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2181 		flags |= IO_SYNC;
2182 
2183 	for (error = 0; uio->uio_resid > 0;) {
2184 		lbn = lblkno(fs, uio->uio_offset);
2185 		blkoffset = blkoff(fs, uio->uio_offset);
2186 		xfersize = fs->e2fs_fsize - blkoffset;
2187 		if (uio->uio_resid < xfersize)
2188 			xfersize = uio->uio_resid;
2189 		if (uio->uio_offset + xfersize > ip->i_size)
2190 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2191 
2192 		/*
2193 		 * We must perform a read-before-write if the transfer size
2194 		 * does not cover the entire buffer.
2195 		 */
2196 		if (fs->e2fs_bsize > xfersize)
2197 			flags |= BA_CLRBUF;
2198 		else
2199 			flags &= ~BA_CLRBUF;
2200 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2201 		    ap->a_cred, &bp, flags);
2202 		if (error != 0)
2203 			break;
2204 
2205 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2206 			bp->b_flags |= B_NOCACHE;
2207 		if (uio->uio_offset + xfersize > ip->i_size)
2208 			ip->i_size = uio->uio_offset + xfersize;
2209 		size = blksize(fs, ip, lbn) - bp->b_resid;
2210 		if (size < xfersize)
2211 			xfersize = size;
2212 
2213 		error =
2214 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2215 		/*
2216 		 * If the buffer is not already filled and we encounter an
2217 		 * error while trying to fill it, we have to clear out any
2218 		 * garbage data from the pages instantiated for the buffer.
2219 		 * If we do not, a failed uiomove() during a write can leave
2220 		 * the prior contents of the pages exposed to a userland mmap.
2221 		 *
2222 		 * Note that we need only clear buffers with a transfer size
2223 		 * equal to the block size because buffers with a shorter
2224 		 * transfer size were cleared above by the call to ext2_balloc()
2225 		 * with the BA_CLRBUF flag set.
2226 		 *
2227 		 * If the source region for uiomove identically mmaps the
2228 		 * buffer, uiomove() performed the NOP copy, and the buffer
2229 		 * content remains valid because the page fault handler
2230 		 * validated the pages.
2231 		 */
2232 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2233 		    fs->e2fs_bsize == xfersize)
2234 			vfs_bio_clrbuf(bp);
2235 
2236 		vfs_bio_set_flags(bp, ioflag);
2237 
2238 		/*
2239 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2240 		 * if we have a severe page deficiency write the buffer
2241 		 * asynchronously.  Otherwise try to cluster, and if that
2242 		 * doesn't do it then either do an async write (if O_DIRECT),
2243 		 * or a delayed write (if not).
2244 		 */
2245 		if (ioflag & IO_SYNC) {
2246 			(void)bwrite(bp);
2247 		} else if (vm_page_count_severe() ||
2248 			    buf_dirty_count_severe() ||
2249 		    (ioflag & IO_ASYNC)) {
2250 			bp->b_flags |= B_CLUSTEROK;
2251 			bawrite(bp);
2252 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2253 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2254 				bp->b_flags |= B_CLUSTEROK;
2255 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2256 			} else {
2257 				bawrite(bp);
2258 			}
2259 		} else if (ioflag & IO_DIRECT) {
2260 			bp->b_flags |= B_CLUSTEROK;
2261 			bawrite(bp);
2262 		} else {
2263 			bp->b_flags |= B_CLUSTEROK;
2264 			bdwrite(bp);
2265 		}
2266 		if (error || xfersize == 0)
2267 			break;
2268 	}
2269 	/*
2270 	 * If we successfully wrote any data, and we are not the superuser
2271 	 * we clear the setuid and setgid bits as a precaution against
2272 	 * tampering.
2273 	 */
2274 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2275 	    ap->a_cred) {
2276 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2277 			ip->i_mode &= ~(ISUID | ISGID);
2278 	}
2279 	if (error) {
2280 		if (ioflag & IO_UNIT) {
2281 			(void)ext2_truncate(vp, osize,
2282 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2283 			uio->uio_offset -= resid - uio->uio_resid;
2284 			uio->uio_resid = resid;
2285 		}
2286 	}
2287 	if (uio->uio_resid != resid) {
2288 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2289 		if (ioflag & IO_SYNC)
2290 			error = ext2_update(vp, 1);
2291 	}
2292 	return (error);
2293 }
2294