xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision cbb3ec25236ba72f91cbdf23f8b78b9d1af0cedf)
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  */
45 
46 #include "opt_suiddir.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/fcntl.h>
52 #include <sys/filio.h>
53 #include <sys/limits.h>
54 #include <sys/sdt.h>
55 #include <sys/stat.h>
56 #include <sys/bio.h>
57 #include <sys/buf.h>
58 #include <sys/endian.h>
59 #include <sys/priv.h>
60 #include <sys/rwlock.h>
61 #include <sys/mount.h>
62 #include <sys/unistd.h>
63 #include <sys/time.h>
64 #include <sys/vnode.h>
65 #include <sys/namei.h>
66 #include <sys/lockf.h>
67 #include <sys/event.h>
68 #include <sys/conf.h>
69 #include <sys/file.h>
70 #include <sys/extattr.h>
71 #include <sys/vmmeter.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pager.h>
79 #include <vm/vnode_pager.h>
80 
81 #include "opt_directio.h"
82 
83 #include <ufs/ufs/dir.h>
84 
85 #include <fs/ext2fs/fs.h>
86 #include <fs/ext2fs/inode.h>
87 #include <fs/ext2fs/ext2_acl.h>
88 #include <fs/ext2fs/ext2fs.h>
89 #include <fs/ext2fs/ext2_extern.h>
90 #include <fs/ext2fs/ext2_dinode.h>
91 #include <fs/ext2fs/ext2_dir.h>
92 #include <fs/ext2fs/ext2_mount.h>
93 #include <fs/ext2fs/ext2_extattr.h>
94 #include <fs/ext2fs/ext2_extents.h>
95 
96 SDT_PROVIDER_DECLARE(ext2fs);
97 /*
98  * ext2fs trace probe:
99  * arg0: verbosity. Higher numbers give more verbose messages
100  * arg1: Textual message
101  */
102 SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*");
103 
104 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
105 static void ext2_itimes_locked(struct vnode *);
106 
107 static vop_access_t	ext2_access;
108 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
109 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
110     struct thread *);
111 static vop_close_t	ext2_close;
112 static vop_create_t	ext2_create;
113 static vop_fsync_t	ext2_fsync;
114 static vop_getattr_t	ext2_getattr;
115 static vop_ioctl_t	ext2_ioctl;
116 static vop_link_t	ext2_link;
117 static vop_mkdir_t	ext2_mkdir;
118 static vop_mknod_t	ext2_mknod;
119 static vop_open_t	ext2_open;
120 static vop_pathconf_t	ext2_pathconf;
121 static vop_print_t	ext2_print;
122 static vop_read_t	ext2_read;
123 static vop_readlink_t	ext2_readlink;
124 static vop_remove_t	ext2_remove;
125 static vop_rename_t	ext2_rename;
126 static vop_rmdir_t	ext2_rmdir;
127 static vop_setattr_t	ext2_setattr;
128 static vop_strategy_t	ext2_strategy;
129 static vop_symlink_t	ext2_symlink;
130 static vop_write_t	ext2_write;
131 static vop_deleteextattr_t	ext2_deleteextattr;
132 static vop_getextattr_t	ext2_getextattr;
133 static vop_listextattr_t	ext2_listextattr;
134 static vop_setextattr_t	ext2_setextattr;
135 static vop_vptofh_t	ext2_vptofh;
136 static vop_close_t	ext2fifo_close;
137 
138 /* Global vfs data structures for ext2. */
139 struct vop_vector ext2_vnodeops = {
140 	.vop_default =		&default_vnodeops,
141 	.vop_access =		ext2_access,
142 	.vop_bmap =		ext2_bmap,
143 	.vop_cachedlookup =	ext2_lookup,
144 	.vop_close =		ext2_close,
145 	.vop_create =		ext2_create,
146 	.vop_fsync =		ext2_fsync,
147 	.vop_getpages =		vnode_pager_local_getpages,
148 	.vop_getpages_async =	vnode_pager_local_getpages_async,
149 	.vop_getattr =		ext2_getattr,
150 	.vop_inactive =		ext2_inactive,
151 	.vop_ioctl =		ext2_ioctl,
152 	.vop_link =		ext2_link,
153 	.vop_lookup =		vfs_cache_lookup,
154 	.vop_mkdir =		ext2_mkdir,
155 	.vop_mknod =		ext2_mknod,
156 	.vop_open =		ext2_open,
157 	.vop_pathconf =		ext2_pathconf,
158 	.vop_poll =		vop_stdpoll,
159 	.vop_print =		ext2_print,
160 	.vop_read =		ext2_read,
161 	.vop_readdir =		ext2_readdir,
162 	.vop_readlink =		ext2_readlink,
163 	.vop_reallocblks =	ext2_reallocblks,
164 	.vop_reclaim =		ext2_reclaim,
165 	.vop_remove =		ext2_remove,
166 	.vop_rename =		ext2_rename,
167 	.vop_rmdir =		ext2_rmdir,
168 	.vop_setattr =		ext2_setattr,
169 	.vop_strategy =		ext2_strategy,
170 	.vop_symlink =		ext2_symlink,
171 	.vop_write =		ext2_write,
172 	.vop_deleteextattr =	ext2_deleteextattr,
173 	.vop_getextattr =	ext2_getextattr,
174 	.vop_listextattr =	ext2_listextattr,
175 	.vop_setextattr =	ext2_setextattr,
176 #ifdef UFS_ACL
177 	.vop_getacl =		ext2_getacl,
178 	.vop_setacl =		ext2_setacl,
179 	.vop_aclcheck =		ext2_aclcheck,
180 #endif /* UFS_ACL */
181 	.vop_vptofh =		ext2_vptofh,
182 };
183 VFS_VOP_VECTOR_REGISTER(ext2_vnodeops);
184 
185 struct vop_vector ext2_fifoops = {
186 	.vop_default =		&fifo_specops,
187 	.vop_access =		ext2_access,
188 	.vop_close =		ext2fifo_close,
189 	.vop_fsync =		ext2_fsync,
190 	.vop_getattr =		ext2_getattr,
191 	.vop_inactive =		ext2_inactive,
192 	.vop_pathconf =		ext2_pathconf,
193 	.vop_print =		ext2_print,
194 	.vop_read =		VOP_PANIC,
195 	.vop_reclaim =		ext2_reclaim,
196 	.vop_setattr =		ext2_setattr,
197 	.vop_write =		VOP_PANIC,
198 	.vop_vptofh =		ext2_vptofh,
199 };
200 VFS_VOP_VECTOR_REGISTER(ext2_fifoops);
201 
202 /*
203  * A virgin directory (no blushing please).
204  * Note that the type and namlen fields are reversed relative to ext2.
205  * Also, we don't use `struct odirtemplate', since it would just cause
206  * endianness problems.
207  */
208 static struct dirtemplate mastertemplate = {
209 	0, htole16(12), 1, EXT2_FT_DIR, ".",
210 	0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_DIR, ".."
211 };
212 static struct dirtemplate omastertemplate = {
213 	0, htole16(12), 1, EXT2_FT_UNKNOWN, ".",
214 	0, htole16(DIRBLKSIZ - 12), 2, EXT2_FT_UNKNOWN, ".."
215 };
216 
217 static void
218 ext2_itimes_locked(struct vnode *vp)
219 {
220 	struct inode *ip;
221 	struct timespec ts;
222 
223 	ASSERT_VI_LOCKED(vp, __func__);
224 
225 	ip = VTOI(vp);
226 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
227 		return;
228 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
229 		ip->i_flag |= IN_LAZYMOD;
230 	else
231 		ip->i_flag |= IN_MODIFIED;
232 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
233 		vfs_timestamp(&ts);
234 		if (ip->i_flag & IN_ACCESS) {
235 			ip->i_atime = ts.tv_sec;
236 			ip->i_atimensec = ts.tv_nsec;
237 		}
238 		if (ip->i_flag & IN_UPDATE) {
239 			ip->i_mtime = ts.tv_sec;
240 			ip->i_mtimensec = ts.tv_nsec;
241 			ip->i_modrev++;
242 		}
243 		if (ip->i_flag & IN_CHANGE) {
244 			ip->i_ctime = ts.tv_sec;
245 			ip->i_ctimensec = ts.tv_nsec;
246 		}
247 	}
248 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
249 }
250 
251 void
252 ext2_itimes(struct vnode *vp)
253 {
254 
255 	VI_LOCK(vp);
256 	ext2_itimes_locked(vp);
257 	VI_UNLOCK(vp);
258 }
259 
260 /*
261  * Create a regular file
262  */
263 static int
264 ext2_create(struct vop_create_args *ap)
265 {
266 	int error;
267 
268 	error =
269 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
270 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
271 	if (error != 0)
272 		return (error);
273 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
274 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
275 	return (0);
276 }
277 
278 static int
279 ext2_open(struct vop_open_args *ap)
280 {
281 
282 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
283 		return (EOPNOTSUPP);
284 
285 	/*
286 	 * Files marked append-only must be opened for appending.
287 	 */
288 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
289 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
290 		return (EPERM);
291 
292 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
293 
294 	return (0);
295 }
296 
297 /*
298  * Close called.
299  *
300  * Update the times on the inode.
301  */
302 static int
303 ext2_close(struct vop_close_args *ap)
304 {
305 	struct vnode *vp = ap->a_vp;
306 
307 	VI_LOCK(vp);
308 	if (vp->v_usecount > 1)
309 		ext2_itimes_locked(vp);
310 	VI_UNLOCK(vp);
311 	return (0);
312 }
313 
314 static int
315 ext2_access(struct vop_access_args *ap)
316 {
317 	struct vnode *vp = ap->a_vp;
318 	struct inode *ip = VTOI(vp);
319 	accmode_t accmode = ap->a_accmode;
320 	int error;
321 
322 	/*
323 	 * Disallow write attempts on read-only file systems;
324 	 * unless the file is a socket, fifo, or a block or
325 	 * character device resident on the file system.
326 	 */
327 	if (accmode & VWRITE) {
328 		switch (vp->v_type) {
329 		case VDIR:
330 		case VLNK:
331 		case VREG:
332 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
333 				return (EROFS);
334 			break;
335 		default:
336 			break;
337 		}
338 	}
339 
340 	/* If immutable bit set, nobody gets to write it. */
341 	if ((accmode & VWRITE) && (ip->i_flags & SF_IMMUTABLE))
342 		return (EPERM);
343 
344 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
345 	    ap->a_accmode, ap->a_cred);
346 	return (error);
347 }
348 
349 static int
350 ext2_getattr(struct vop_getattr_args *ap)
351 {
352 	struct vnode *vp = ap->a_vp;
353 	struct inode *ip = VTOI(vp);
354 	struct vattr *vap = ap->a_vap;
355 
356 	ext2_itimes(vp);
357 	/*
358 	 * Copy from inode table
359 	 */
360 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
361 	vap->va_fileid = ip->i_number;
362 	vap->va_mode = ip->i_mode & ~IFMT;
363 	vap->va_nlink = ip->i_nlink;
364 	vap->va_uid = ip->i_uid;
365 	vap->va_gid = ip->i_gid;
366 	vap->va_rdev = ip->i_rdev;
367 	vap->va_size = ip->i_size;
368 	vap->va_atime.tv_sec = ip->i_atime;
369 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
370 	vap->va_mtime.tv_sec = ip->i_mtime;
371 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
372 	vap->va_ctime.tv_sec = ip->i_ctime;
373 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
374 	if (E2DI_HAS_XTIME(ip)) {
375 		vap->va_birthtime.tv_sec = ip->i_birthtime;
376 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
377 	}
378 	vap->va_flags = ip->i_flags;
379 	vap->va_gen = ip->i_gen;
380 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
381 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
382 	vap->va_type = IFTOVT(ip->i_mode);
383 	vap->va_filerev = ip->i_modrev;
384 	return (0);
385 }
386 
387 /*
388  * Set attribute vnode op. called from several syscalls
389  */
390 static int
391 ext2_setattr(struct vop_setattr_args *ap)
392 {
393 	struct vattr *vap = ap->a_vap;
394 	struct vnode *vp = ap->a_vp;
395 	struct inode *ip = VTOI(vp);
396 	struct ucred *cred = ap->a_cred;
397 	struct thread *td = curthread;
398 	int error;
399 
400 	/*
401 	 * Check for unsettable attributes.
402 	 */
403 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
404 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
405 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
406 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
407 		return (EINVAL);
408 	}
409 	if (vap->va_flags != VNOVAL) {
410 		/* Disallow flags not supported by ext2fs. */
411 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
412 			return (EOPNOTSUPP);
413 
414 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
415 			return (EROFS);
416 		/*
417 		 * Callers may only modify the file flags on objects they
418 		 * have VADMIN rights for.
419 		 */
420 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
421 			return (error);
422 		/*
423 		 * Unprivileged processes and privileged processes in
424 		 * jail() are not permitted to unset system flags, or
425 		 * modify flags if any system flags are set.
426 		 * Privileged non-jail processes may not modify system flags
427 		 * if securelevel > 0 and any existing system flags are set.
428 		 */
429 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
430 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
431 				error = securelevel_gt(cred, 0);
432 				if (error)
433 					return (error);
434 			}
435 		} else {
436 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
437 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
438 				return (EPERM);
439 		}
440 		ip->i_flags = vap->va_flags;
441 		ip->i_flag |= IN_CHANGE;
442 		if (ip->i_flags & (IMMUTABLE | APPEND))
443 			return (0);
444 	}
445 	if (ip->i_flags & (IMMUTABLE | APPEND))
446 		return (EPERM);
447 	/*
448 	 * Go through the fields and update iff not VNOVAL.
449 	 */
450 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
451 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
452 			return (EROFS);
453 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
454 		    td)) != 0)
455 			return (error);
456 	}
457 	if (vap->va_size != VNOVAL) {
458 		/*
459 		 * Disallow write attempts on read-only file systems;
460 		 * unless the file is a socket, fifo, or a block or
461 		 * character device resident on the file system.
462 		 */
463 		switch (vp->v_type) {
464 		case VDIR:
465 			return (EISDIR);
466 		case VLNK:
467 		case VREG:
468 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
469 				return (EROFS);
470 			break;
471 		default:
472 			break;
473 		}
474 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
475 			return (error);
476 	}
477 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
478 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
479 			return (EROFS);
480 		/*
481 		 * From utimes(2):
482 		 * If times is NULL, ... The caller must be the owner of
483 		 * the file, have permission to write the file, or be the
484 		 * super-user.
485 		 * If times is non-NULL, ... The caller must be the owner of
486 		 * the file or be the super-user.
487 		 */
488 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
489 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
490 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
491 			return (error);
492 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
493 		if (vap->va_atime.tv_sec != VNOVAL) {
494 			ip->i_flag &= ~IN_ACCESS;
495 			ip->i_atime = vap->va_atime.tv_sec;
496 			ip->i_atimensec = vap->va_atime.tv_nsec;
497 		}
498 		if (vap->va_mtime.tv_sec != VNOVAL) {
499 			ip->i_flag &= ~IN_UPDATE;
500 			ip->i_mtime = vap->va_mtime.tv_sec;
501 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
502 		}
503 		if (E2DI_HAS_XTIME(ip) && vap->va_birthtime.tv_sec != VNOVAL) {
504 			ip->i_birthtime = vap->va_birthtime.tv_sec;
505 			ip->i_birthnsec = vap->va_birthtime.tv_nsec;
506 		}
507 		error = ext2_update(vp, 0);
508 		if (error)
509 			return (error);
510 	}
511 	error = 0;
512 	if (vap->va_mode != (mode_t)VNOVAL) {
513 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
514 			return (EROFS);
515 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
516 	}
517 	return (error);
518 }
519 
520 /*
521  * Change the mode on a file.
522  * Inode must be locked before calling.
523  */
524 static int
525 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
526 {
527 	struct inode *ip = VTOI(vp);
528 	int error;
529 
530 	/*
531 	 * To modify the permissions on a file, must possess VADMIN
532 	 * for that file.
533 	 */
534 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
535 		return (error);
536 	/*
537 	 * Privileged processes may set the sticky bit on non-directories,
538 	 * as well as set the setgid bit on a file with a group that the
539 	 * process is not a member of.
540 	 */
541 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
542 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE);
543 		if (error)
544 			return (EFTYPE);
545 	}
546 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
547 		error = priv_check_cred(cred, PRIV_VFS_SETGID);
548 		if (error)
549 			return (error);
550 	}
551 	ip->i_mode &= ~ALLPERMS;
552 	ip->i_mode |= (mode & ALLPERMS);
553 	ip->i_flag |= IN_CHANGE;
554 	return (0);
555 }
556 
557 /*
558  * Perform chown operation on inode ip;
559  * inode must be locked prior to call.
560  */
561 static int
562 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
563     struct thread *td)
564 {
565 	struct inode *ip = VTOI(vp);
566 	uid_t ouid;
567 	gid_t ogid;
568 	int error = 0;
569 
570 	if (uid == (uid_t)VNOVAL)
571 		uid = ip->i_uid;
572 	if (gid == (gid_t)VNOVAL)
573 		gid = ip->i_gid;
574 	/*
575 	 * To modify the ownership of a file, must possess VADMIN
576 	 * for that file.
577 	 */
578 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
579 		return (error);
580 	/*
581 	 * To change the owner of a file, or change the group of a file
582 	 * to a group of which we are not a member, the caller must
583 	 * have privilege.
584 	 */
585 	if (uid != ip->i_uid || (gid != ip->i_gid &&
586 	    !groupmember(gid, cred))) {
587 		error = priv_check_cred(cred, PRIV_VFS_CHOWN);
588 		if (error)
589 			return (error);
590 	}
591 	ogid = ip->i_gid;
592 	ouid = ip->i_uid;
593 	ip->i_gid = gid;
594 	ip->i_uid = uid;
595 	ip->i_flag |= IN_CHANGE;
596 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
597 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID) != 0)
598 			ip->i_mode &= ~(ISUID | ISGID);
599 	}
600 	return (0);
601 }
602 
603 /*
604  * Synch an open file.
605  */
606 /* ARGSUSED */
607 static int
608 ext2_fsync(struct vop_fsync_args *ap)
609 {
610 	/*
611 	 * Flush all dirty buffers associated with a vnode.
612 	 */
613 
614 	vop_stdfsync(ap);
615 
616 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
617 }
618 
619 static int
620 ext2_check_mknod_limits(dev_t dev)
621 {
622 	unsigned maj = major(dev);
623 	unsigned min = minor(dev);
624 
625 	if (maj > EXT2_MAJOR_MAX || min > EXT2_MINOR_MAX)
626 		return (EINVAL);
627 
628 	return (0);
629 }
630 
631 /*
632  * Mknod vnode call
633  */
634 /* ARGSUSED */
635 static int
636 ext2_mknod(struct vop_mknod_args *ap)
637 {
638 	struct vattr *vap = ap->a_vap;
639 	struct vnode **vpp = ap->a_vpp;
640 	struct inode *ip;
641 	ino_t ino;
642 	int error;
643 
644 	if (vap->va_rdev != VNOVAL) {
645 		error = ext2_check_mknod_limits(vap->va_rdev);
646 		if (error)
647 			return (error);
648 	}
649 
650 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
651 	    ap->a_dvp, vpp, ap->a_cnp);
652 	if (error)
653 		return (error);
654 	ip = VTOI(*vpp);
655 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
656 	if (vap->va_rdev != VNOVAL)
657 		ip->i_rdev = vap->va_rdev;
658 
659 	/*
660 	 * Remove inode, then reload it through VFS_VGET so it is
661 	 * checked to see if it is an alias of an existing entry in
662 	 * the inode cache.	 XXX I don't believe this is necessary now.
663 	 */
664 	(*vpp)->v_type = VNON;
665 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
666 	vgone(*vpp);
667 	vput(*vpp);
668 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
669 	if (error) {
670 		*vpp = NULL;
671 		return (error);
672 	}
673 	return (0);
674 }
675 
676 static int
677 ext2_remove(struct vop_remove_args *ap)
678 {
679 	struct inode *ip;
680 	struct vnode *vp = ap->a_vp;
681 	struct vnode *dvp = ap->a_dvp;
682 	int error;
683 
684 	ip = VTOI(vp);
685 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
686 	    (VTOI(dvp)->i_flags & APPEND)) {
687 		error = EPERM;
688 		goto out;
689 	}
690 	error = ext2_dirremove(dvp, ap->a_cnp);
691 	if (error == 0) {
692 		ip->i_nlink--;
693 		ip->i_flag |= IN_CHANGE;
694 	}
695 out:
696 	return (error);
697 }
698 
699 /*
700  * link vnode call
701  */
702 static int
703 ext2_link(struct vop_link_args *ap)
704 {
705 	struct vnode *vp = ap->a_vp;
706 	struct vnode *tdvp = ap->a_tdvp;
707 	struct componentname *cnp = ap->a_cnp;
708 	struct inode *ip;
709 	int error;
710 
711 	ip = VTOI(vp);
712 	if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) {
713 		error = EMLINK;
714 		goto out;
715 	}
716 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
717 		error = EPERM;
718 		goto out;
719 	}
720 	ip->i_nlink++;
721 	ip->i_flag |= IN_CHANGE;
722 	error = ext2_update(vp, !DOINGASYNC(vp));
723 	if (!error)
724 		error = ext2_direnter(ip, tdvp, cnp);
725 	if (error) {
726 		ip->i_nlink--;
727 		ip->i_flag |= IN_CHANGE;
728 	}
729 out:
730 	return (error);
731 }
732 
733 static int
734 ext2_inc_nlink(struct inode *ip)
735 {
736 
737 	ip->i_nlink++;
738 
739 	if (S_ISDIR(ip->i_mode) &&
740 	    EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) &&
741 	    ip->i_nlink > 1) {
742 		if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2)
743 			ip->i_nlink = 1;
744 	} else if (ip->i_nlink > EXT4_LINK_MAX) {
745 		ip->i_nlink--;
746 		return (EMLINK);
747 	}
748 
749 	return (0);
750 }
751 
752 static void
753 ext2_dec_nlink(struct inode *ip)
754 {
755 
756 	if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
757 		ip->i_nlink--;
758 }
759 
760 /*
761  * Rename system call.
762  * 	rename("foo", "bar");
763  * is essentially
764  *	unlink("bar");
765  *	link("foo", "bar");
766  *	unlink("foo");
767  * but ``atomically''.  Can't do full commit without saving state in the
768  * inode on disk which isn't feasible at this time.  Best we can do is
769  * always guarantee the target exists.
770  *
771  * Basic algorithm is:
772  *
773  * 1) Bump link count on source while we're linking it to the
774  *    target.  This also ensure the inode won't be deleted out
775  *    from underneath us while we work (it may be truncated by
776  *    a concurrent `trunc' or `open' for creation).
777  * 2) Link source to destination.  If destination already exists,
778  *    delete it first.
779  * 3) Unlink source reference to inode if still around. If a
780  *    directory was moved and the parent of the destination
781  *    is different from the source, patch the ".." entry in the
782  *    directory.
783  */
784 static int
785 ext2_rename(struct vop_rename_args *ap)
786 {
787 	struct vnode *tvp = ap->a_tvp;
788 	struct vnode *tdvp = ap->a_tdvp;
789 	struct vnode *fvp = ap->a_fvp;
790 	struct vnode *fdvp = ap->a_fdvp;
791 	struct componentname *tcnp = ap->a_tcnp;
792 	struct componentname *fcnp = ap->a_fcnp;
793 	struct inode *ip, *xp, *dp;
794 	struct dirtemplate *dirbuf;
795 	int doingdirectory = 0, oldparent = 0, newparent = 0;
796 	int error = 0;
797 	u_char namlen;
798 
799 	/*
800 	 * Check for cross-device rename.
801 	 */
802 	if ((fvp->v_mount != tdvp->v_mount) ||
803 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
804 		error = EXDEV;
805 abortit:
806 		if (tdvp == tvp)
807 			vrele(tdvp);
808 		else
809 			vput(tdvp);
810 		if (tvp)
811 			vput(tvp);
812 		vrele(fdvp);
813 		vrele(fvp);
814 		return (error);
815 	}
816 
817 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
818 	    (VTOI(tdvp)->i_flags & APPEND))) {
819 		error = EPERM;
820 		goto abortit;
821 	}
822 
823 	/*
824 	 * Renaming a file to itself has no effect.  The upper layers should
825 	 * not call us in that case.  Temporarily just warn if they do.
826 	 */
827 	if (fvp == tvp) {
828 		SDT_PROBE2(ext2fs, , vnops, trace, 1,
829 		    "rename: fvp == tvp (can't happen)");
830 		error = 0;
831 		goto abortit;
832 	}
833 
834 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
835 		goto abortit;
836 	dp = VTOI(fdvp);
837 	ip = VTOI(fvp);
838 	if (ip->i_nlink >= EXT4_LINK_MAX &&
839 	    !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
840 		VOP_UNLOCK(fvp);
841 		error = EMLINK;
842 		goto abortit;
843 	}
844 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
845 	    || (dp->i_flags & APPEND)) {
846 		VOP_UNLOCK(fvp);
847 		error = EPERM;
848 		goto abortit;
849 	}
850 	if ((ip->i_mode & IFMT) == IFDIR) {
851 		/*
852 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
853 		 */
854 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
855 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
856 		    (ip->i_flag & IN_RENAME)) {
857 			VOP_UNLOCK(fvp);
858 			error = EINVAL;
859 			goto abortit;
860 		}
861 		ip->i_flag |= IN_RENAME;
862 		oldparent = dp->i_number;
863 		doingdirectory++;
864 	}
865 	vrele(fdvp);
866 
867 	/*
868 	 * When the target exists, both the directory
869 	 * and target vnodes are returned locked.
870 	 */
871 	dp = VTOI(tdvp);
872 	xp = NULL;
873 	if (tvp)
874 		xp = VTOI(tvp);
875 
876 	/*
877 	 * 1) Bump link count while we're moving stuff
878 	 *    around.  If we crash somewhere before
879 	 *    completing our work, the link count
880 	 *    may be wrong, but correctable.
881 	 */
882 	ext2_inc_nlink(ip);
883 	ip->i_flag |= IN_CHANGE;
884 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
885 		VOP_UNLOCK(fvp);
886 		goto bad;
887 	}
888 
889 	/*
890 	 * If ".." must be changed (ie the directory gets a new
891 	 * parent) then the source directory must not be in the
892 	 * directory hierarchy above the target, as this would
893 	 * orphan everything below the source directory. Also
894 	 * the user must have write permission in the source so
895 	 * as to be able to change "..". We must repeat the call
896 	 * to namei, as the parent directory is unlocked by the
897 	 * call to checkpath().
898 	 */
899 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
900 	VOP_UNLOCK(fvp);
901 	if (oldparent != dp->i_number)
902 		newparent = dp->i_number;
903 	if (doingdirectory && newparent) {
904 		if (error)	/* write access check above */
905 			goto bad;
906 		if (xp != NULL)
907 			vput(tvp);
908 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
909 		if (error)
910 			goto out;
911 		VREF(tdvp);
912 		error = vfs_relookup(tdvp, &tvp, tcnp, true);
913 		if (error)
914 			goto out;
915 		vrele(tdvp);
916 		dp = VTOI(tdvp);
917 		xp = NULL;
918 		if (tvp)
919 			xp = VTOI(tvp);
920 	}
921 	/*
922 	 * 2) If target doesn't exist, link the target
923 	 *    to the source and unlink the source.
924 	 *    Otherwise, rewrite the target directory
925 	 *    entry to reference the source inode and
926 	 *    expunge the original entry's existence.
927 	 */
928 	if (xp == NULL) {
929 		if (dp->i_devvp != ip->i_devvp)
930 			panic("ext2_rename: EXDEV");
931 		/*
932 		 * Account for ".." in new directory.
933 		 * When source and destination have the same
934 		 * parent we don't fool with the link count.
935 		 */
936 		if (doingdirectory && newparent) {
937 			error = ext2_inc_nlink(dp);
938 			if (error)
939 				goto bad;
940 
941 			dp->i_flag |= IN_CHANGE;
942 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
943 			if (error)
944 				goto bad;
945 		}
946 		error = ext2_direnter(ip, tdvp, tcnp);
947 		if (error) {
948 			if (doingdirectory && newparent) {
949 				ext2_dec_nlink(dp);
950 				dp->i_flag |= IN_CHANGE;
951 				(void)ext2_update(tdvp, 1);
952 			}
953 			goto bad;
954 		}
955 		vput(tdvp);
956 	} else {
957 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
958 			panic("ext2_rename: EXDEV");
959 		/*
960 		 * Short circuit rename(foo, foo).
961 		 */
962 		if (xp->i_number == ip->i_number)
963 			panic("ext2_rename: same file");
964 		/*
965 		 * If the parent directory is "sticky", then the user must
966 		 * own the parent directory, or the destination of the rename,
967 		 * otherwise the destination may not be changed (except by
968 		 * root). This implements append-only directories.
969 		 */
970 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
971 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
972 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
973 			error = EPERM;
974 			goto bad;
975 		}
976 		/*
977 		 * Target must be empty if a directory and have no links
978 		 * to it. Also, ensure source and target are compatible
979 		 * (both directories, or both not directories).
980 		 */
981 		if ((xp->i_mode & IFMT) == IFDIR) {
982 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
983 				error = ENOTEMPTY;
984 				goto bad;
985 			}
986 			if (!doingdirectory) {
987 				error = ENOTDIR;
988 				goto bad;
989 			}
990 			cache_purge(tdvp);
991 		} else if (doingdirectory) {
992 			error = EISDIR;
993 			goto bad;
994 		}
995 		error = ext2_dirrewrite(dp, ip, tcnp);
996 		if (error)
997 			goto bad;
998 		/*
999 		 * If the target directory is in the same
1000 		 * directory as the source directory,
1001 		 * decrement the link count on the parent
1002 		 * of the target directory.
1003 		 */
1004 		if (doingdirectory && !newparent) {
1005 			ext2_dec_nlink(dp);
1006 			dp->i_flag |= IN_CHANGE;
1007 		}
1008 		vput(tdvp);
1009 		/*
1010 		 * Adjust the link count of the target to
1011 		 * reflect the dirrewrite above.  If this is
1012 		 * a directory it is empty and there are
1013 		 * no links to it, so we can squash the inode and
1014 		 * any space associated with it.  We disallowed
1015 		 * renaming over top of a directory with links to
1016 		 * it above, as the remaining link would point to
1017 		 * a directory without "." or ".." entries.
1018 		 */
1019 		ext2_dec_nlink(xp);
1020 		if (doingdirectory) {
1021 			if (xp->i_nlink > 2)
1022 				panic("ext2_rename: linked directory");
1023 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1024 			    tcnp->cn_cred, curthread);
1025 			xp->i_nlink = 0;
1026 		}
1027 		xp->i_flag |= IN_CHANGE;
1028 		vput(tvp);
1029 		xp = NULL;
1030 	}
1031 
1032 	/*
1033 	 * 3) Unlink the source.
1034 	 */
1035 	fcnp->cn_flags &= ~MODMASK;
1036 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1037 	VREF(fdvp);
1038 	error = vfs_relookup(fdvp, &fvp, fcnp, true);
1039 	if (error == 0)
1040 		vrele(fdvp);
1041 	if (fvp != NULL) {
1042 		xp = VTOI(fvp);
1043 		dp = VTOI(fdvp);
1044 	} else {
1045 		/*
1046 		 * From name has disappeared.  IN_RENAME is not sufficient
1047 		 * to protect against directory races due to timing windows,
1048 		 * so we can't panic here.
1049 		 */
1050 		vrele(ap->a_fvp);
1051 		return (0);
1052 	}
1053 	/*
1054 	 * Ensure that the directory entry still exists and has not
1055 	 * changed while the new name has been entered. If the source is
1056 	 * a file then the entry may have been unlinked or renamed. In
1057 	 * either case there is no further work to be done. If the source
1058 	 * is a directory then it cannot have been rmdir'ed; its link
1059 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1060 	 * The IN_RENAME flag ensures that it cannot be moved by another
1061 	 * rename.
1062 	 */
1063 	if (xp != ip) {
1064 		/*
1065 		 * From name resolves to a different inode.  IN_RENAME is
1066 		 * not sufficient protection against timing window races
1067 		 * so we can't panic here.
1068 		 */
1069 	} else {
1070 		/*
1071 		 * If the source is a directory with a
1072 		 * new parent, the link count of the old
1073 		 * parent directory must be decremented
1074 		 * and ".." set to point to the new parent.
1075 		 */
1076 		if (doingdirectory && newparent) {
1077 			ext2_dec_nlink(dp);
1078 			dp->i_flag |= IN_CHANGE;
1079 			dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1080 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1081 			    ip->i_e2fs->e2fs_bsize, (off_t)0,
1082 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1083 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1084 			if (error == 0) {
1085 				/* Like ufs little-endian: */
1086 				namlen = dirbuf->dotdot_type;
1087 				if (namlen != 2 ||
1088 				    dirbuf->dotdot_name[0] != '.' ||
1089 				    dirbuf->dotdot_name[1] != '.') {
1090 					/*
1091 					 * The filesystem is in corrupted state,
1092 					 * need to run fsck to fix mangled dir
1093 					 * entry. From other side this error
1094 					 * need to be ignored because it is
1095 					 * too difficult to revert directories
1096 					 * to state before rename from this
1097 					 * point.
1098 					 */
1099 					ext2_dirbad(xp, (doff_t)12,
1100 					    "rename: mangled dir");
1101 				} else {
1102 					dirbuf->dotdot_ino = htole32(newparent);
1103 					/*
1104 					 * dirblock 0 could be htree root,
1105 					 * try both csum update functions.
1106 					 */
1107 					ext2_dirent_csum_set(ip,
1108 					    (struct ext2fs_direct_2 *)dirbuf);
1109 					ext2_dx_csum_set(ip,
1110 					    (struct ext2fs_direct_2 *)dirbuf);
1111 					(void)vn_rdwr(UIO_WRITE, fvp,
1112 					    (caddr_t)dirbuf,
1113 					    ip->i_e2fs->e2fs_bsize,
1114 					    (off_t)0, UIO_SYSSPACE,
1115 					    IO_NODELOCKED | IO_SYNC |
1116 					    IO_NOMACCHECK, tcnp->cn_cred,
1117 					    NOCRED, NULL, NULL);
1118 					cache_purge(fdvp);
1119 				}
1120 			}
1121 			free(dirbuf, M_TEMP);
1122 		}
1123 		error = ext2_dirremove(fdvp, fcnp);
1124 		if (!error) {
1125 			ext2_dec_nlink(xp);
1126 			xp->i_flag |= IN_CHANGE;
1127 		}
1128 		xp->i_flag &= ~IN_RENAME;
1129 	}
1130 	if (dp)
1131 		vput(fdvp);
1132 	if (xp)
1133 		vput(fvp);
1134 	vrele(ap->a_fvp);
1135 	return (error);
1136 
1137 bad:
1138 	if (xp)
1139 		vput(ITOV(xp));
1140 	vput(ITOV(dp));
1141 out:
1142 	if (doingdirectory)
1143 		ip->i_flag &= ~IN_RENAME;
1144 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1145 		ext2_dec_nlink(ip);
1146 		ip->i_flag |= IN_CHANGE;
1147 		ip->i_flag &= ~IN_RENAME;
1148 		vput(fvp);
1149 	} else
1150 		vrele(fvp);
1151 	return (error);
1152 }
1153 
1154 #ifdef UFS_ACL
1155 static int
1156 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1157     mode_t dmode, struct ucred *cred, struct thread *td)
1158 {
1159 	int error;
1160 	struct inode *ip = VTOI(tvp);
1161 	struct acl *dacl, *acl;
1162 
1163 	acl = acl_alloc(M_WAITOK);
1164 	dacl = acl_alloc(M_WAITOK);
1165 
1166 	/*
1167 	 * Retrieve default ACL from parent, if any.
1168 	 */
1169 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1170 	switch (error) {
1171 	case 0:
1172 		/*
1173 		 * Retrieved a default ACL, so merge mode and ACL if
1174 		 * necessary.  If the ACL is empty, fall through to
1175 		 * the "not defined or available" case.
1176 		 */
1177 		if (acl->acl_cnt != 0) {
1178 			dmode = acl_posix1e_newfilemode(dmode, acl);
1179 			ip->i_mode = dmode;
1180 			*dacl = *acl;
1181 			ext2_sync_acl_from_inode(ip, acl);
1182 			break;
1183 		}
1184 		/* FALLTHROUGH */
1185 
1186 	case EOPNOTSUPP:
1187 		/*
1188 		 * Just use the mode as-is.
1189 		 */
1190 		ip->i_mode = dmode;
1191 		error = 0;
1192 		goto out;
1193 
1194 	default:
1195 		goto out;
1196 	}
1197 
1198 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1199 	if (error == 0)
1200 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1201 	switch (error) {
1202 	case 0:
1203 		break;
1204 
1205 	case EOPNOTSUPP:
1206 		/*
1207 		 * XXX: This should not happen, as EOPNOTSUPP above
1208 		 * was supposed to free acl.
1209 		 */
1210 #ifdef DEBUG
1211 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1212 #endif	/* DEBUG */
1213 		break;
1214 
1215 	default:
1216 		goto out;
1217 	}
1218 
1219 out:
1220 	acl_free(acl);
1221 	acl_free(dacl);
1222 
1223 	return (error);
1224 }
1225 
1226 static int
1227 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1228     mode_t mode, struct ucred *cred, struct thread *td)
1229 {
1230 	int error;
1231 	struct inode *ip = VTOI(tvp);
1232 	struct acl *acl;
1233 
1234 	acl = acl_alloc(M_WAITOK);
1235 
1236 	/*
1237 	 * Retrieve default ACL for parent, if any.
1238 	 */
1239 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1240 	switch (error) {
1241 	case 0:
1242 		/*
1243 		 * Retrieved a default ACL, so merge mode and ACL if
1244 		 * necessary.
1245 		 */
1246 		if (acl->acl_cnt != 0) {
1247 			/*
1248 			 * Two possible ways for default ACL to not
1249 			 * be present.  First, the EA can be
1250 			 * undefined, or second, the default ACL can
1251 			 * be blank.  If it's blank, fall through to
1252 			 * the it's not defined case.
1253 			 */
1254 			mode = acl_posix1e_newfilemode(mode, acl);
1255 			ip->i_mode = mode;
1256 			ext2_sync_acl_from_inode(ip, acl);
1257 			break;
1258 		}
1259 		/* FALLTHROUGH */
1260 
1261 	case EOPNOTSUPP:
1262 		/*
1263 		 * Just use the mode as-is.
1264 		 */
1265 		ip->i_mode = mode;
1266 		error = 0;
1267 		goto out;
1268 
1269 	default:
1270 		goto out;
1271 	}
1272 
1273 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1274 	switch (error) {
1275 	case 0:
1276 		break;
1277 
1278 	case EOPNOTSUPP:
1279 		/*
1280 		 * XXX: This should not happen, as EOPNOTSUPP above was
1281 		 * supposed to free acl.
1282 		 */
1283 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1284 		    "but no VOP_SETACL()\n");
1285 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1286 		    "but no VOP_SETACL()"); */
1287 		break;
1288 
1289 	default:
1290 		goto out;
1291 	}
1292 
1293 out:
1294 	acl_free(acl);
1295 
1296 	return (error);
1297 }
1298 
1299 #endif /* UFS_ACL */
1300 
1301 /*
1302  * Mkdir system call
1303  */
1304 static int
1305 ext2_mkdir(struct vop_mkdir_args *ap)
1306 {
1307 	struct m_ext2fs *fs;
1308 	struct vnode *dvp = ap->a_dvp;
1309 	struct vattr *vap = ap->a_vap;
1310 	struct componentname *cnp = ap->a_cnp;
1311 	struct inode *ip, *dp;
1312 	struct vnode *tvp;
1313 	struct dirtemplate dirtemplate, *dtp;
1314 	char *buf = NULL;
1315 	int error, dmode;
1316 
1317 	dp = VTOI(dvp);
1318 	if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX &&
1319 	    !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
1320 		error = EMLINK;
1321 		goto out;
1322 	}
1323 	dmode = vap->va_mode & 0777;
1324 	dmode |= IFDIR;
1325 	/*
1326 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1327 	 * but not have it entered in the parent directory. The entry is
1328 	 * made later after writing "." and ".." entries.
1329 	 */
1330 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1331 	if (error)
1332 		goto out;
1333 	ip = VTOI(tvp);
1334 	fs = ip->i_e2fs;
1335 	ip->i_gid = dp->i_gid;
1336 #ifdef SUIDDIR
1337 	{
1338 		/*
1339 		 * if we are hacking owners here, (only do this where told to)
1340 		 * and we are not giving it TOO root, (would subvert quotas)
1341 		 * then go ahead and give it to the other user.
1342 		 * The new directory also inherits the SUID bit.
1343 		 * If user's UID and dir UID are the same,
1344 		 * 'give it away' so that the SUID is still forced on.
1345 		 */
1346 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1347 		    (dp->i_mode & ISUID) && dp->i_uid) {
1348 			dmode |= ISUID;
1349 			ip->i_uid = dp->i_uid;
1350 		} else {
1351 			ip->i_uid = cnp->cn_cred->cr_uid;
1352 		}
1353 	}
1354 #else
1355 	ip->i_uid = cnp->cn_cred->cr_uid;
1356 #endif
1357 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1358 	ip->i_mode = dmode;
1359 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1360 	ip->i_nlink = 2;
1361 	if (cnp->cn_flags & ISWHITEOUT)
1362 		ip->i_flags |= UF_OPAQUE;
1363 	error = ext2_update(tvp, 1);
1364 
1365 	/*
1366 	 * Bump link count in parent directory
1367 	 * to reflect work done below.  Should
1368 	 * be done before reference is created
1369 	 * so reparation is possible if we crash.
1370 	 */
1371 	ext2_inc_nlink(dp);
1372 	dp->i_flag |= IN_CHANGE;
1373 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1374 	if (error)
1375 		goto bad;
1376 
1377 	/* Initialize directory with "." and ".." from static template. */
1378 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1379 	    EXT2F_INCOMPAT_FTYPE))
1380 		dtp = &mastertemplate;
1381 	else
1382 		dtp = &omastertemplate;
1383 	dirtemplate = *dtp;
1384 	dirtemplate.dot_ino = htole32(ip->i_number);
1385 	dirtemplate.dotdot_ino = htole32(dp->i_number);
1386 	/*
1387 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1388 	 * just redefine it - for this function only
1389 	 */
1390 #undef  DIRBLKSIZ
1391 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1392 	dirtemplate.dotdot_reclen = htole16(DIRBLKSIZ - 12);
1393 	buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1394 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1395 		dirtemplate.dotdot_reclen =
1396 		    htole16(le16toh(dirtemplate.dotdot_reclen) -
1397 		    sizeof(struct ext2fs_direct_tail));
1398 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1399 	}
1400 	memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1401 	ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf);
1402 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1403 	    DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1404 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1405 	    NULL, NULL);
1406 	if (error) {
1407 		ext2_dec_nlink(dp);
1408 		dp->i_flag |= IN_CHANGE;
1409 		goto bad;
1410 	}
1411 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1412 		/* XXX should grow with balloc() */
1413 		panic("ext2_mkdir: blksize");
1414 	else {
1415 		ip->i_size = DIRBLKSIZ;
1416 		ip->i_flag |= IN_CHANGE;
1417 	}
1418 
1419 #ifdef UFS_ACL
1420 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1421 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1422 		    cnp->cn_cred, curthread);
1423 		if (error)
1424 			goto bad;
1425 	}
1426 
1427 #endif /* UFS_ACL */
1428 
1429 	/* Directory set up, now install its entry in the parent directory. */
1430 	error = ext2_direnter(ip, dvp, cnp);
1431 	if (error) {
1432 		ext2_dec_nlink(dp);
1433 		dp->i_flag |= IN_CHANGE;
1434 	}
1435 bad:
1436 	/*
1437 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1438 	 * for us because we set the link count to 0.
1439 	 */
1440 	if (error) {
1441 		ip->i_nlink = 0;
1442 		ip->i_flag |= IN_CHANGE;
1443 		vput(tvp);
1444 	} else
1445 		*ap->a_vpp = tvp;
1446 out:
1447 	free(buf, M_TEMP);
1448 	return (error);
1449 #undef  DIRBLKSIZ
1450 #define DIRBLKSIZ  DEV_BSIZE
1451 }
1452 
1453 /*
1454  * Rmdir system call.
1455  */
1456 static int
1457 ext2_rmdir(struct vop_rmdir_args *ap)
1458 {
1459 	struct vnode *vp = ap->a_vp;
1460 	struct vnode *dvp = ap->a_dvp;
1461 	struct componentname *cnp = ap->a_cnp;
1462 	struct inode *ip, *dp;
1463 	int error;
1464 
1465 	ip = VTOI(vp);
1466 	dp = VTOI(dvp);
1467 
1468 	/*
1469 	 * Verify the directory is empty (and valid).
1470 	 * (Rmdir ".." won't be valid since
1471 	 *  ".." will contain a reference to
1472 	 *  the current directory and thus be
1473 	 *  non-empty.)
1474 	 */
1475 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1476 		error = ENOTEMPTY;
1477 		goto out;
1478 	}
1479 	if ((dp->i_flags & APPEND)
1480 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1481 		error = EPERM;
1482 		goto out;
1483 	}
1484 	/*
1485 	 * Delete reference to directory before purging
1486 	 * inode.  If we crash in between, the directory
1487 	 * will be reattached to lost+found,
1488 	 */
1489 	error = ext2_dirremove(dvp, cnp);
1490 	if (error)
1491 		goto out;
1492 	ext2_dec_nlink(dp);
1493 	dp->i_flag |= IN_CHANGE;
1494 	cache_purge(dvp);
1495 	VOP_UNLOCK(dvp);
1496 	/*
1497 	 * Truncate inode.  The only stuff left
1498 	 * in the directory is "." and "..".
1499 	 */
1500 	ip->i_nlink = 0;
1501 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1502 	    curthread);
1503 	cache_purge(ITOV(ip));
1504 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1505 		VOP_UNLOCK(vp);
1506 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1507 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1508 	}
1509 out:
1510 	return (error);
1511 }
1512 
1513 /*
1514  * symlink -- make a symbolic link
1515  */
1516 static int
1517 ext2_symlink(struct vop_symlink_args *ap)
1518 {
1519 	struct vnode *vp, **vpp = ap->a_vpp;
1520 	struct inode *ip;
1521 	int len, error;
1522 
1523 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1524 	    vpp, ap->a_cnp);
1525 	if (error)
1526 		return (error);
1527 	vp = *vpp;
1528 	len = strlen(ap->a_target);
1529 	if (len < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1530 		ip = VTOI(vp);
1531 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1532 		ip->i_size = len;
1533 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1534 	} else
1535 		error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
1536 		    len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1537 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1538 	if (error)
1539 		vput(vp);
1540 	return (error);
1541 }
1542 
1543 /*
1544  * Return target name of a symbolic link
1545  */
1546 static int
1547 ext2_readlink(struct vop_readlink_args *ap)
1548 {
1549 	struct vnode *vp = ap->a_vp;
1550 	struct inode *ip = VTOI(vp);
1551 	int isize;
1552 
1553 	isize = ip->i_size;
1554 	if (isize < VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen) {
1555 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1556 		return (0);
1557 	}
1558 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1559 }
1560 
1561 /*
1562  * Calculate the logical to physical mapping if not done already,
1563  * then call the device strategy routine.
1564  *
1565  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1566  * deadlock on memory.  See ext2_bmap() for details.
1567  */
1568 static int
1569 ext2_strategy(struct vop_strategy_args *ap)
1570 {
1571 	struct buf *bp = ap->a_bp;
1572 	struct vnode *vp = ap->a_vp;
1573 	struct bufobj *bo;
1574 	daddr_t blkno;
1575 	int error;
1576 
1577 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1578 		panic("ext2_strategy: spec");
1579 	if (bp->b_blkno == bp->b_lblkno) {
1580 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1581 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1582 		else
1583 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1584 
1585 		bp->b_blkno = blkno;
1586 		if (error) {
1587 			bp->b_error = error;
1588 			bp->b_ioflags |= BIO_ERROR;
1589 			bufdone(bp);
1590 			return (0);
1591 		}
1592 		if ((long)bp->b_blkno == -1)
1593 			vfs_bio_clrbuf(bp);
1594 	}
1595 	if ((long)bp->b_blkno == -1) {
1596 		bufdone(bp);
1597 		return (0);
1598 	}
1599 	bp->b_iooffset = dbtob(bp->b_blkno);
1600 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1601 	BO_STRATEGY(bo, bp);
1602 	return (0);
1603 }
1604 
1605 /*
1606  * Print out the contents of an inode.
1607  */
1608 static int
1609 ext2_print(struct vop_print_args *ap)
1610 {
1611 	struct vnode *vp = ap->a_vp;
1612 	struct inode *ip = VTOI(vp);
1613 
1614 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1615 	if (vp->v_type == VFIFO)
1616 		fifo_printinfo(vp);
1617 	printf("\n");
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Close wrapper for fifos.
1623  *
1624  * Update the times on the inode then do device close.
1625  */
1626 static int
1627 ext2fifo_close(struct vop_close_args *ap)
1628 {
1629 	struct vnode *vp = ap->a_vp;
1630 
1631 	VI_LOCK(vp);
1632 	if (vp->v_usecount > 1)
1633 		ext2_itimes_locked(vp);
1634 	VI_UNLOCK(vp);
1635 	return (fifo_specops.vop_close(ap));
1636 }
1637 
1638 /*
1639  * Return POSIX pathconf information applicable to ext2 filesystems.
1640  */
1641 static int
1642 ext2_pathconf(struct vop_pathconf_args *ap)
1643 {
1644 	int error = 0;
1645 
1646 	switch (ap->a_name) {
1647 	case _PC_LINK_MAX:
1648 		if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs,
1649 		    EXT2F_ROCOMPAT_DIR_NLINK))
1650 			*ap->a_retval = INT_MAX;
1651 		else
1652 			*ap->a_retval = EXT4_LINK_MAX;
1653 		break;
1654 	case _PC_NAME_MAX:
1655 		*ap->a_retval = NAME_MAX;
1656 		break;
1657 	case _PC_PIPE_BUF:
1658 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1659 			*ap->a_retval = PIPE_BUF;
1660 		else
1661 			error = EINVAL;
1662 		break;
1663 	case _PC_CHOWN_RESTRICTED:
1664 		*ap->a_retval = 1;
1665 		break;
1666 	case _PC_NO_TRUNC:
1667 		*ap->a_retval = 1;
1668 		break;
1669 
1670 #ifdef UFS_ACL
1671 	case _PC_ACL_EXTENDED:
1672 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1673 			*ap->a_retval = 1;
1674 		else
1675 			*ap->a_retval = 0;
1676 		break;
1677 	case _PC_ACL_PATH_MAX:
1678 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1679 			*ap->a_retval = ACL_MAX_ENTRIES;
1680 		else
1681 			*ap->a_retval = 3;
1682 		break;
1683 #endif /* UFS_ACL */
1684 
1685 	case _PC_MIN_HOLE_SIZE:
1686 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1687 		break;
1688 	case _PC_PRIO_IO:
1689 		*ap->a_retval = 0;
1690 		break;
1691 	case _PC_SYNC_IO:
1692 		*ap->a_retval = 0;
1693 		break;
1694 	case _PC_ALLOC_SIZE_MIN:
1695 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1696 		break;
1697 	case _PC_FILESIZEBITS:
1698 		*ap->a_retval = 64;
1699 		break;
1700 	case _PC_REC_INCR_XFER_SIZE:
1701 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1702 		break;
1703 	case _PC_REC_MAX_XFER_SIZE:
1704 		*ap->a_retval = -1;	/* means ``unlimited'' */
1705 		break;
1706 	case _PC_REC_MIN_XFER_SIZE:
1707 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1708 		break;
1709 	case _PC_REC_XFER_ALIGN:
1710 		*ap->a_retval = PAGE_SIZE;
1711 		break;
1712 	case _PC_SYMLINK_MAX:
1713 		*ap->a_retval = MAXPATHLEN;
1714 		break;
1715 
1716 	default:
1717 		error = vop_stdpathconf(ap);
1718 		break;
1719 	}
1720 	return (error);
1721 }
1722 
1723 /*
1724  * Vnode operation to remove a named attribute.
1725  */
1726 static int
1727 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1728 {
1729 	struct inode *ip;
1730 	struct m_ext2fs *fs;
1731 	int error;
1732 
1733 	ip = VTOI(ap->a_vp);
1734 	fs = ip->i_e2fs;
1735 
1736 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1737 		return (EOPNOTSUPP);
1738 
1739 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1740 		return (EOPNOTSUPP);
1741 
1742 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1743 	    ap->a_cred, ap->a_td, VWRITE);
1744 	if (error)
1745 		return (error);
1746 
1747 	error = ENOATTR;
1748 
1749 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1750 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1751 		if (error != ENOATTR)
1752 			return (error);
1753 	}
1754 
1755 	if (ip->i_facl)
1756 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1757 
1758 	return (error);
1759 }
1760 
1761 /*
1762  * Vnode operation to retrieve a named extended attribute.
1763  */
1764 static int
1765 ext2_getextattr(struct vop_getextattr_args *ap)
1766 {
1767 	struct inode *ip;
1768 	struct m_ext2fs *fs;
1769 	int error;
1770 
1771 	ip = VTOI(ap->a_vp);
1772 	fs = ip->i_e2fs;
1773 
1774 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1775 		return (EOPNOTSUPP);
1776 
1777 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1778 		return (EOPNOTSUPP);
1779 
1780 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1781 	    ap->a_cred, ap->a_td, VREAD);
1782 	if (error)
1783 		return (error);
1784 
1785 	if (ap->a_size != NULL)
1786 		*ap->a_size = 0;
1787 
1788 	error = ENOATTR;
1789 
1790 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1791 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1792 		    ap->a_name, ap->a_uio, ap->a_size);
1793 		if (error != ENOATTR)
1794 			return (error);
1795 	}
1796 
1797 	if (ip->i_facl)
1798 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1799 		    ap->a_name, ap->a_uio, ap->a_size);
1800 
1801 	return (error);
1802 }
1803 
1804 /*
1805  * Vnode operation to retrieve extended attributes on a vnode.
1806  */
1807 static int
1808 ext2_listextattr(struct vop_listextattr_args *ap)
1809 {
1810 	struct inode *ip;
1811 	struct m_ext2fs *fs;
1812 	int error;
1813 
1814 	ip = VTOI(ap->a_vp);
1815 	fs = ip->i_e2fs;
1816 
1817 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1818 		return (EOPNOTSUPP);
1819 
1820 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1821 		return (EOPNOTSUPP);
1822 
1823 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1824 	    ap->a_cred, ap->a_td, VREAD);
1825 	if (error)
1826 		return (error);
1827 
1828 	if (ap->a_size != NULL)
1829 		*ap->a_size = 0;
1830 
1831 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1832 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1833 		    ap->a_uio, ap->a_size);
1834 		if (error)
1835 			return (error);
1836 	}
1837 
1838 	if (ip->i_facl)
1839 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1840 		    ap->a_uio, ap->a_size);
1841 
1842 	return (error);
1843 }
1844 
1845 /*
1846  * Vnode operation to set a named attribute.
1847  */
1848 static int
1849 ext2_setextattr(struct vop_setextattr_args *ap)
1850 {
1851 	struct inode *ip;
1852 	struct m_ext2fs *fs;
1853 	int error;
1854 
1855 	ip = VTOI(ap->a_vp);
1856 	fs = ip->i_e2fs;
1857 
1858 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1859 		return (EOPNOTSUPP);
1860 
1861 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1862 		return (EOPNOTSUPP);
1863 
1864 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1865 	    ap->a_cred, ap->a_td, VWRITE);
1866 	if (error)
1867 		return (error);
1868 
1869 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1870 	if (error)
1871 		return (error);
1872 
1873 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1874 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1875 		    ap->a_name, ap->a_uio);
1876 		if (error != ENOSPC)
1877 			return (error);
1878 	}
1879 
1880 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1881 	    ap->a_name, ap->a_uio);
1882 
1883 	return (error);
1884 }
1885 
1886 /*
1887  * Vnode pointer to File handle
1888  */
1889 /* ARGSUSED */
1890 static int
1891 ext2_vptofh(struct vop_vptofh_args *ap)
1892 {
1893 	struct inode *ip;
1894 	struct ufid *ufhp;
1895 
1896 	ip = VTOI(ap->a_vp);
1897 	ufhp = (struct ufid *)ap->a_fhp;
1898 	ufhp->ufid_len = sizeof(struct ufid);
1899 	ufhp->ufid_ino = ip->i_number;
1900 	ufhp->ufid_gen = ip->i_gen;
1901 	return (0);
1902 }
1903 
1904 /*
1905  * Initialize the vnode associated with a new inode, handle aliased
1906  * vnodes.
1907  */
1908 int
1909 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1910 {
1911 	struct inode *ip;
1912 	struct vnode *vp;
1913 
1914 	vp = *vpp;
1915 	ip = VTOI(vp);
1916 	vp->v_type = IFTOVT(ip->i_mode);
1917 	/*
1918 	 * Only unallocated inodes should be of type VNON.
1919 	 */
1920 	if (ip->i_mode != 0 && vp->v_type == VNON)
1921 		return (EINVAL);
1922 	if (vp->v_type == VFIFO)
1923 		vp->v_op = fifoops;
1924 
1925 	if (ip->i_number == EXT2_ROOTINO)
1926 		vp->v_vflag |= VV_ROOT;
1927 	ip->i_modrev = init_va_filerev();
1928 	*vpp = vp;
1929 	return (0);
1930 }
1931 
1932 /*
1933  * Allocate a new inode.
1934  */
1935 static int
1936 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1937     struct componentname *cnp)
1938 {
1939 	struct inode *ip, *pdir;
1940 	struct vnode *tvp;
1941 	int error;
1942 
1943 	pdir = VTOI(dvp);
1944 	*vpp = NULL;
1945 	if ((mode & IFMT) == 0)
1946 		mode |= IFREG;
1947 
1948 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1949 	if (error) {
1950 		return (error);
1951 	}
1952 	ip = VTOI(tvp);
1953 	ip->i_gid = pdir->i_gid;
1954 #ifdef SUIDDIR
1955 	{
1956 		/*
1957 		 * if we are
1958 		 * not the owner of the directory,
1959 		 * and we are hacking owners here, (only do this where told to)
1960 		 * and we are not giving it TOO root, (would subvert quotas)
1961 		 * then go ahead and give it to the other user.
1962 		 * Note that this drops off the execute bits for security.
1963 		 */
1964 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1965 		    (pdir->i_mode & ISUID) &&
1966 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1967 			ip->i_uid = pdir->i_uid;
1968 			mode &= ~07111;
1969 		} else {
1970 			ip->i_uid = cnp->cn_cred->cr_uid;
1971 		}
1972 	}
1973 #else
1974 	ip->i_uid = cnp->cn_cred->cr_uid;
1975 #endif
1976 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1977 	ip->i_mode = mode;
1978 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1979 	ip->i_nlink = 1;
1980 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1981 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID))
1982 			ip->i_mode &= ~ISGID;
1983 	}
1984 
1985 	if (cnp->cn_flags & ISWHITEOUT)
1986 		ip->i_flags |= UF_OPAQUE;
1987 
1988 	/*
1989 	 * Make sure inode goes to disk before directory entry.
1990 	 */
1991 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1992 	if (error)
1993 		goto bad;
1994 
1995 #ifdef UFS_ACL
1996 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1997 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
1998 		    cnp->cn_cred, curthread);
1999 		if (error)
2000 			goto bad;
2001 	}
2002 #endif /* UFS_ACL */
2003 
2004 	error = ext2_direnter(ip, dvp, cnp);
2005 	if (error)
2006 		goto bad;
2007 
2008 	*vpp = tvp;
2009 	return (0);
2010 
2011 bad:
2012 	/*
2013 	 * Write error occurred trying to update the inode
2014 	 * or the directory so must deallocate the inode.
2015 	 */
2016 	ip->i_nlink = 0;
2017 	ip->i_flag |= IN_CHANGE;
2018 	vput(tvp);
2019 	return (error);
2020 }
2021 
2022 /*
2023  * Vnode op for reading.
2024  */
2025 static int
2026 ext2_read(struct vop_read_args *ap)
2027 {
2028 	struct vnode *vp;
2029 	struct inode *ip;
2030 	struct uio *uio;
2031 	struct m_ext2fs *fs;
2032 	struct buf *bp;
2033 	daddr_t lbn, nextlbn;
2034 	off_t bytesinfile;
2035 	long size, xfersize, blkoffset;
2036 	int error, orig_resid, seqcount;
2037 	int ioflag;
2038 
2039 	vp = ap->a_vp;
2040 	uio = ap->a_uio;
2041 	ioflag = ap->a_ioflag;
2042 
2043 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2044 	ip = VTOI(vp);
2045 
2046 #ifdef INVARIANTS
2047 	if (uio->uio_rw != UIO_READ)
2048 		panic("%s: mode", "ext2_read");
2049 
2050 	if (vp->v_type == VLNK) {
2051 		if ((int)ip->i_size <
2052 		    VFSTOEXT2(vp->v_mount)->um_e2fs->e2fs_maxsymlinklen)
2053 			panic("%s: short symlink", "ext2_read");
2054 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2055 		panic("%s: type %d", "ext2_read", vp->v_type);
2056 #endif
2057 	orig_resid = uio->uio_resid;
2058 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2059 	if (orig_resid == 0)
2060 		return (0);
2061 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2062 	fs = ip->i_e2fs;
2063 	if (uio->uio_offset < ip->i_size &&
2064 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2065 		return (EOVERFLOW);
2066 
2067 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2068 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2069 			break;
2070 		lbn = lblkno(fs, uio->uio_offset);
2071 		nextlbn = lbn + 1;
2072 		size = blksize(fs, ip, lbn);
2073 		blkoffset = blkoff(fs, uio->uio_offset);
2074 
2075 		xfersize = fs->e2fs_fsize - blkoffset;
2076 		if (uio->uio_resid < xfersize)
2077 			xfersize = uio->uio_resid;
2078 		if (bytesinfile < xfersize)
2079 			xfersize = bytesinfile;
2080 
2081 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2082 			error = bread(vp, lbn, size, NOCRED, &bp);
2083 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2084 			error = cluster_read(vp, ip->i_size, lbn, size,
2085 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2086 			    0, &bp);
2087 		} else if (seqcount > 1) {
2088 			u_int nextsize = blksize(fs, ip, nextlbn);
2089 
2090 			error = breadn(vp, lbn,
2091 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2092 		} else
2093 			error = bread(vp, lbn, size, NOCRED, &bp);
2094 		if (error) {
2095 			brelse(bp);
2096 			bp = NULL;
2097 			break;
2098 		}
2099 
2100 		/*
2101 		 * We should only get non-zero b_resid when an I/O error
2102 		 * has occurred, which should cause us to break above.
2103 		 * However, if the short read did not cause an error,
2104 		 * then we want to ensure that we do not uiomove bad
2105 		 * or uninitialized data.
2106 		 */
2107 		size -= bp->b_resid;
2108 		if (size < xfersize) {
2109 			if (size == 0)
2110 				break;
2111 			xfersize = size;
2112 		}
2113 		error = uiomove((char *)bp->b_data + blkoffset,
2114 		    (int)xfersize, uio);
2115 		if (error)
2116 			break;
2117 		vfs_bio_brelse(bp, ioflag);
2118 	}
2119 
2120 	/*
2121 	 * This can only happen in the case of an error because the loop
2122 	 * above resets bp to NULL on each iteration and on normal
2123 	 * completion has not set a new value into it. so it must have come
2124 	 * from a 'break' statement
2125 	 */
2126 	if (bp != NULL)
2127 		vfs_bio_brelse(bp, ioflag);
2128 
2129 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2130 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2131 		ip->i_flag |= IN_ACCESS;
2132 	return (error);
2133 }
2134 
2135 static int
2136 ext2_ioctl(struct vop_ioctl_args *ap)
2137 {
2138 	struct vnode *vp;
2139 	int error;
2140 
2141 	vp = ap->a_vp;
2142 	switch (ap->a_command) {
2143 	case FIOSEEKDATA:
2144 		if (!(VTOI(vp)->i_flag & IN_E4EXTENTS)) {
2145 			error = vn_lock(vp, LK_SHARED);
2146 			if (error == 0) {
2147 				error = ext2_bmap_seekdata(vp,
2148 				    (off_t *)ap->a_data);
2149 				VOP_UNLOCK(vp);
2150 			} else
2151 				error = EBADF;
2152 			return (error);
2153 		}
2154 	case FIOSEEKHOLE:
2155 		return (vn_bmap_seekhole(vp, ap->a_command,
2156 		    (off_t *)ap->a_data, ap->a_cred));
2157 	default:
2158 		return (ENOTTY);
2159 	}
2160 }
2161 
2162 /*
2163  * Vnode op for writing.
2164  */
2165 static int
2166 ext2_write(struct vop_write_args *ap)
2167 {
2168 	struct vnode *vp;
2169 	struct uio *uio;
2170 	struct inode *ip;
2171 	struct m_ext2fs *fs;
2172 	struct buf *bp;
2173 	daddr_t lbn;
2174 	off_t osize;
2175 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2176 
2177 	ioflag = ap->a_ioflag;
2178 	uio = ap->a_uio;
2179 	vp = ap->a_vp;
2180 
2181 	seqcount = ioflag >> IO_SEQSHIFT;
2182 	ip = VTOI(vp);
2183 
2184 #ifdef INVARIANTS
2185 	if (uio->uio_rw != UIO_WRITE)
2186 		panic("%s: mode", "ext2_write");
2187 #endif
2188 
2189 	switch (vp->v_type) {
2190 	case VREG:
2191 		if (ioflag & IO_APPEND)
2192 			uio->uio_offset = ip->i_size;
2193 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2194 			return (EPERM);
2195 		/* FALLTHROUGH */
2196 	case VLNK:
2197 		break;
2198 	case VDIR:
2199 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2200 		if ((ioflag & IO_SYNC) == 0)
2201 			panic("ext2_write: nonsync dir write");
2202 		break;
2203 	default:
2204 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2205 		    vp->v_type, (intmax_t)uio->uio_offset,
2206 		    (intmax_t)uio->uio_resid);
2207 	}
2208 
2209 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2210 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2211 	fs = ip->i_e2fs;
2212 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2213 		return (EFBIG);
2214 	/*
2215 	 * Maybe this should be above the vnode op call, but so long as
2216 	 * file servers have no limits, I don't think it matters.
2217 	 */
2218 	error = vn_rlimit_fsize(vp, uio, uio->uio_td);
2219 	if (error != 0)
2220 		return (error);
2221 
2222 	resid = uio->uio_resid;
2223 	osize = ip->i_size;
2224 	if (seqcount > BA_SEQMAX)
2225 		flags = BA_SEQMAX << BA_SEQSHIFT;
2226 	else
2227 		flags = seqcount << BA_SEQSHIFT;
2228 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2229 		flags |= IO_SYNC;
2230 
2231 	for (error = 0; uio->uio_resid > 0;) {
2232 		lbn = lblkno(fs, uio->uio_offset);
2233 		blkoffset = blkoff(fs, uio->uio_offset);
2234 		xfersize = fs->e2fs_fsize - blkoffset;
2235 		if (uio->uio_resid < xfersize)
2236 			xfersize = uio->uio_resid;
2237 		if (uio->uio_offset + xfersize > ip->i_size)
2238 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2239 
2240 		/*
2241 		 * We must perform a read-before-write if the transfer size
2242 		 * does not cover the entire buffer.
2243 		 */
2244 		if (fs->e2fs_bsize > xfersize)
2245 			flags |= BA_CLRBUF;
2246 		else
2247 			flags &= ~BA_CLRBUF;
2248 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2249 		    ap->a_cred, &bp, flags);
2250 		if (error != 0)
2251 			break;
2252 
2253 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2254 			bp->b_flags |= B_NOCACHE;
2255 		if (uio->uio_offset + xfersize > ip->i_size)
2256 			ip->i_size = uio->uio_offset + xfersize;
2257 		size = blksize(fs, ip, lbn) - bp->b_resid;
2258 		if (size < xfersize)
2259 			xfersize = size;
2260 
2261 		error =
2262 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2263 		/*
2264 		 * If the buffer is not already filled and we encounter an
2265 		 * error while trying to fill it, we have to clear out any
2266 		 * garbage data from the pages instantiated for the buffer.
2267 		 * If we do not, a failed uiomove() during a write can leave
2268 		 * the prior contents of the pages exposed to a userland mmap.
2269 		 *
2270 		 * Note that we need only clear buffers with a transfer size
2271 		 * equal to the block size because buffers with a shorter
2272 		 * transfer size were cleared above by the call to ext2_balloc()
2273 		 * with the BA_CLRBUF flag set.
2274 		 *
2275 		 * If the source region for uiomove identically mmaps the
2276 		 * buffer, uiomove() performed the NOP copy, and the buffer
2277 		 * content remains valid because the page fault handler
2278 		 * validated the pages.
2279 		 */
2280 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2281 		    fs->e2fs_bsize == xfersize)
2282 			vfs_bio_clrbuf(bp);
2283 
2284 		vfs_bio_set_flags(bp, ioflag);
2285 
2286 		/*
2287 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2288 		 * if we have a severe page deficiency write the buffer
2289 		 * asynchronously.  Otherwise try to cluster, and if that
2290 		 * doesn't do it then either do an async write (if O_DIRECT),
2291 		 * or a delayed write (if not).
2292 		 */
2293 		if (ioflag & IO_SYNC) {
2294 			(void)bwrite(bp);
2295 		} else if (vm_page_count_severe() ||
2296 			    buf_dirty_count_severe() ||
2297 		    (ioflag & IO_ASYNC)) {
2298 			bp->b_flags |= B_CLUSTEROK;
2299 			bawrite(bp);
2300 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2301 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2302 				bp->b_flags |= B_CLUSTEROK;
2303 				cluster_write(vp, &ip->i_clusterw, bp,
2304 				    ip->i_size, seqcount, 0);
2305 			} else {
2306 				bawrite(bp);
2307 			}
2308 		} else if (ioflag & IO_DIRECT) {
2309 			bp->b_flags |= B_CLUSTEROK;
2310 			bawrite(bp);
2311 		} else {
2312 			bp->b_flags |= B_CLUSTEROK;
2313 			bdwrite(bp);
2314 		}
2315 		if (error || xfersize == 0)
2316 			break;
2317 	}
2318 	/*
2319 	 * If we successfully wrote any data, and we are not the superuser
2320 	 * we clear the setuid and setgid bits as a precaution against
2321 	 * tampering.
2322 	 */
2323 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2324 	    ap->a_cred) {
2325 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID))
2326 			ip->i_mode &= ~(ISUID | ISGID);
2327 	}
2328 	if (error) {
2329 		if (ioflag & IO_UNIT) {
2330 			(void)ext2_truncate(vp, osize,
2331 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2332 			uio->uio_offset -= resid - uio->uio_resid;
2333 			uio->uio_resid = resid;
2334 		}
2335 	}
2336 	if (uio->uio_resid != resid) {
2337 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2338 		if (ioflag & IO_SYNC)
2339 			error = ext2_update(vp, 1);
2340 	}
2341 	return (error);
2342 }
2343