xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision d66820f2c8c6cdc3d0359abf1004dac0a466a01d)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * SPDX-License-Identifier: BSD-3-Clause
9  *
10  * Copyright (c) 1982, 1986, 1989, 1993
11  *	The Regents of the University of California.  All rights reserved.
12  * (c) UNIX System Laboratories, Inc.
13  * All or some portions of this file are derived from material licensed
14  * to the University of California by American Telephone and Telegraph
15  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16  * the permission of UNIX System Laboratories, Inc.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
43  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
44  * $FreeBSD$
45  */
46 
47 #include "opt_suiddir.h"
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/stat.h>
55 #include <sys/bio.h>
56 #include <sys/buf.h>
57 #include <sys/endian.h>
58 #include <sys/priv.h>
59 #include <sys/rwlock.h>
60 #include <sys/mount.h>
61 #include <sys/unistd.h>
62 #include <sys/time.h>
63 #include <sys/vnode.h>
64 #include <sys/namei.h>
65 #include <sys/lockf.h>
66 #include <sys/event.h>
67 #include <sys/conf.h>
68 #include <sys/file.h>
69 #include <sys/extattr.h>
70 #include <sys/vmmeter.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vnode_pager.h>
79 
80 #include "opt_directio.h"
81 
82 #include <ufs/ufs/dir.h>
83 
84 #include <fs/ext2fs/fs.h>
85 #include <fs/ext2fs/inode.h>
86 #include <fs/ext2fs/ext2_acl.h>
87 #include <fs/ext2fs/ext2fs.h>
88 #include <fs/ext2fs/ext2_extern.h>
89 #include <fs/ext2fs/ext2_dinode.h>
90 #include <fs/ext2fs/ext2_dir.h>
91 #include <fs/ext2fs/ext2_mount.h>
92 #include <fs/ext2fs/ext2_extattr.h>
93 #include <fs/ext2fs/ext2_extents.h>
94 
95 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
96 static void ext2_itimes_locked(struct vnode *);
97 
98 static vop_access_t	ext2_access;
99 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
100 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
101     struct thread *);
102 static vop_close_t	ext2_close;
103 static vop_create_t	ext2_create;
104 static vop_fsync_t	ext2_fsync;
105 static vop_getattr_t	ext2_getattr;
106 static vop_ioctl_t	ext2_ioctl;
107 static vop_link_t	ext2_link;
108 static vop_mkdir_t	ext2_mkdir;
109 static vop_mknod_t	ext2_mknod;
110 static vop_open_t	ext2_open;
111 static vop_pathconf_t	ext2_pathconf;
112 static vop_print_t	ext2_print;
113 static vop_read_t	ext2_read;
114 static vop_readlink_t	ext2_readlink;
115 static vop_remove_t	ext2_remove;
116 static vop_rename_t	ext2_rename;
117 static vop_rmdir_t	ext2_rmdir;
118 static vop_setattr_t	ext2_setattr;
119 static vop_strategy_t	ext2_strategy;
120 static vop_symlink_t	ext2_symlink;
121 static vop_write_t	ext2_write;
122 static vop_deleteextattr_t	ext2_deleteextattr;
123 static vop_getextattr_t	ext2_getextattr;
124 static vop_listextattr_t	ext2_listextattr;
125 static vop_setextattr_t	ext2_setextattr;
126 static vop_vptofh_t	ext2_vptofh;
127 static vop_close_t	ext2fifo_close;
128 static vop_kqfilter_t	ext2fifo_kqfilter;
129 
130 /* Global vfs data structures for ext2. */
131 struct vop_vector ext2_vnodeops = {
132 	.vop_default =		&default_vnodeops,
133 	.vop_access =		ext2_access,
134 	.vop_bmap =		ext2_bmap,
135 	.vop_cachedlookup =	ext2_lookup,
136 	.vop_close =		ext2_close,
137 	.vop_create =		ext2_create,
138 	.vop_fsync =		ext2_fsync,
139 	.vop_getpages =		vnode_pager_local_getpages,
140 	.vop_getpages_async =	vnode_pager_local_getpages_async,
141 	.vop_getattr =		ext2_getattr,
142 	.vop_inactive =		ext2_inactive,
143 	.vop_ioctl =		ext2_ioctl,
144 	.vop_link =		ext2_link,
145 	.vop_lookup =		vfs_cache_lookup,
146 	.vop_mkdir =		ext2_mkdir,
147 	.vop_mknod =		ext2_mknod,
148 	.vop_open =		ext2_open,
149 	.vop_pathconf =		ext2_pathconf,
150 	.vop_poll =		vop_stdpoll,
151 	.vop_print =		ext2_print,
152 	.vop_read =		ext2_read,
153 	.vop_readdir =		ext2_readdir,
154 	.vop_readlink =		ext2_readlink,
155 	.vop_reallocblks =	ext2_reallocblks,
156 	.vop_reclaim =		ext2_reclaim,
157 	.vop_remove =		ext2_remove,
158 	.vop_rename =		ext2_rename,
159 	.vop_rmdir =		ext2_rmdir,
160 	.vop_setattr =		ext2_setattr,
161 	.vop_strategy =		ext2_strategy,
162 	.vop_symlink =		ext2_symlink,
163 	.vop_write =		ext2_write,
164 	.vop_deleteextattr =	ext2_deleteextattr,
165 	.vop_getextattr =	ext2_getextattr,
166 	.vop_listextattr =	ext2_listextattr,
167 	.vop_setextattr =	ext2_setextattr,
168 #ifdef UFS_ACL
169 	.vop_getacl =		ext2_getacl,
170 	.vop_setacl =		ext2_setacl,
171 	.vop_aclcheck =		ext2_aclcheck,
172 #endif /* UFS_ACL */
173 	.vop_vptofh =		ext2_vptofh,
174 };
175 
176 struct vop_vector ext2_fifoops = {
177 	.vop_default =		&fifo_specops,
178 	.vop_access =		ext2_access,
179 	.vop_close =		ext2fifo_close,
180 	.vop_fsync =		ext2_fsync,
181 	.vop_getattr =		ext2_getattr,
182 	.vop_inactive =		ext2_inactive,
183 	.vop_kqfilter =		ext2fifo_kqfilter,
184 	.vop_pathconf =		ext2_pathconf,
185 	.vop_print =		ext2_print,
186 	.vop_read =		VOP_PANIC,
187 	.vop_reclaim =		ext2_reclaim,
188 	.vop_setattr =		ext2_setattr,
189 	.vop_write =		VOP_PANIC,
190 	.vop_vptofh =		ext2_vptofh,
191 };
192 
193 /*
194  * A virgin directory (no blushing please).
195  * Note that the type and namlen fields are reversed relative to ext2.
196  * Also, we don't use `struct odirtemplate', since it would just cause
197  * endianness problems.
198  */
199 static struct dirtemplate mastertemplate = {
200 	0, 12, 1, EXT2_FT_DIR, ".",
201 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
202 };
203 static struct dirtemplate omastertemplate = {
204 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
205 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
206 };
207 
208 static void
209 ext2_itimes_locked(struct vnode *vp)
210 {
211 	struct inode *ip;
212 	struct timespec ts;
213 
214 	ASSERT_VI_LOCKED(vp, __func__);
215 
216 	ip = VTOI(vp);
217 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
218 		return;
219 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
220 		ip->i_flag |= IN_LAZYMOD;
221 	else
222 		ip->i_flag |= IN_MODIFIED;
223 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
224 		vfs_timestamp(&ts);
225 		if (ip->i_flag & IN_ACCESS) {
226 			ip->i_atime = ts.tv_sec;
227 			ip->i_atimensec = ts.tv_nsec;
228 		}
229 		if (ip->i_flag & IN_UPDATE) {
230 			ip->i_mtime = ts.tv_sec;
231 			ip->i_mtimensec = ts.tv_nsec;
232 			ip->i_modrev++;
233 		}
234 		if (ip->i_flag & IN_CHANGE) {
235 			ip->i_ctime = ts.tv_sec;
236 			ip->i_ctimensec = ts.tv_nsec;
237 		}
238 	}
239 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
240 }
241 
242 void
243 ext2_itimes(struct vnode *vp)
244 {
245 
246 	VI_LOCK(vp);
247 	ext2_itimes_locked(vp);
248 	VI_UNLOCK(vp);
249 }
250 
251 /*
252  * Create a regular file
253  */
254 static int
255 ext2_create(struct vop_create_args *ap)
256 {
257 	int error;
258 
259 	error =
260 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
261 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
262 	if (error != 0)
263 		return (error);
264 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
265 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
266 	return (0);
267 }
268 
269 static int
270 ext2_open(struct vop_open_args *ap)
271 {
272 
273 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
274 		return (EOPNOTSUPP);
275 
276 	/*
277 	 * Files marked append-only must be opened for appending.
278 	 */
279 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
280 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
281 		return (EPERM);
282 
283 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
284 
285 	return (0);
286 }
287 
288 /*
289  * Close called.
290  *
291  * Update the times on the inode.
292  */
293 static int
294 ext2_close(struct vop_close_args *ap)
295 {
296 	struct vnode *vp = ap->a_vp;
297 
298 	VI_LOCK(vp);
299 	if (vp->v_usecount > 1)
300 		ext2_itimes_locked(vp);
301 	VI_UNLOCK(vp);
302 	return (0);
303 }
304 
305 static int
306 ext2_access(struct vop_access_args *ap)
307 {
308 	struct vnode *vp = ap->a_vp;
309 	struct inode *ip = VTOI(vp);
310 	accmode_t accmode = ap->a_accmode;
311 	int error;
312 
313 	if (vp->v_type == VBLK || vp->v_type == VCHR)
314 		return (EOPNOTSUPP);
315 
316 	/*
317 	 * Disallow write attempts on read-only file systems;
318 	 * unless the file is a socket, fifo, or a block or
319 	 * character device resident on the file system.
320 	 */
321 	if (accmode & VWRITE) {
322 		switch (vp->v_type) {
323 		case VDIR:
324 		case VLNK:
325 		case VREG:
326 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
327 				return (EROFS);
328 			break;
329 		default:
330 			break;
331 		}
332 	}
333 
334 	/* If immutable bit set, nobody gets to write it. */
335 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
336 		return (EPERM);
337 
338 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
339 	    ap->a_accmode, ap->a_cred, NULL);
340 	return (error);
341 }
342 
343 static int
344 ext2_getattr(struct vop_getattr_args *ap)
345 {
346 	struct vnode *vp = ap->a_vp;
347 	struct inode *ip = VTOI(vp);
348 	struct vattr *vap = ap->a_vap;
349 
350 	ext2_itimes(vp);
351 	/*
352 	 * Copy from inode table
353 	 */
354 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
355 	vap->va_fileid = ip->i_number;
356 	vap->va_mode = ip->i_mode & ~IFMT;
357 	vap->va_nlink = ip->i_nlink;
358 	vap->va_uid = ip->i_uid;
359 	vap->va_gid = ip->i_gid;
360 	vap->va_rdev = ip->i_rdev;
361 	vap->va_size = ip->i_size;
362 	vap->va_atime.tv_sec = ip->i_atime;
363 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
364 	vap->va_mtime.tv_sec = ip->i_mtime;
365 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
366 	vap->va_ctime.tv_sec = ip->i_ctime;
367 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
368 	if E2DI_HAS_XTIME(ip) {
369 		vap->va_birthtime.tv_sec = ip->i_birthtime;
370 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
371 	}
372 	vap->va_flags = ip->i_flags;
373 	vap->va_gen = ip->i_gen;
374 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
375 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
376 	vap->va_type = IFTOVT(ip->i_mode);
377 	vap->va_filerev = ip->i_modrev;
378 	return (0);
379 }
380 
381 /*
382  * Set attribute vnode op. called from several syscalls
383  */
384 static int
385 ext2_setattr(struct vop_setattr_args *ap)
386 {
387 	struct vattr *vap = ap->a_vap;
388 	struct vnode *vp = ap->a_vp;
389 	struct inode *ip = VTOI(vp);
390 	struct ucred *cred = ap->a_cred;
391 	struct thread *td = curthread;
392 	int error;
393 
394 	/*
395 	 * Check for unsettable attributes.
396 	 */
397 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
398 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
399 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
400 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
401 		return (EINVAL);
402 	}
403 	if (vap->va_flags != VNOVAL) {
404 		/* Disallow flags not supported by ext2fs. */
405 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
406 			return (EOPNOTSUPP);
407 
408 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
409 			return (EROFS);
410 		/*
411 		 * Callers may only modify the file flags on objects they
412 		 * have VADMIN rights for.
413 		 */
414 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
415 			return (error);
416 		/*
417 		 * Unprivileged processes and privileged processes in
418 		 * jail() are not permitted to unset system flags, or
419 		 * modify flags if any system flags are set.
420 		 * Privileged non-jail processes may not modify system flags
421 		 * if securelevel > 0 and any existing system flags are set.
422 		 */
423 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
424 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
425 				error = securelevel_gt(cred, 0);
426 				if (error)
427 					return (error);
428 			}
429 		} else {
430 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
431 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
432 				return (EPERM);
433 		}
434 		ip->i_flags = vap->va_flags;
435 		ip->i_flag |= IN_CHANGE;
436 		if (ip->i_flags & (IMMUTABLE | APPEND))
437 			return (0);
438 	}
439 	if (ip->i_flags & (IMMUTABLE | APPEND))
440 		return (EPERM);
441 	/*
442 	 * Go through the fields and update iff not VNOVAL.
443 	 */
444 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
445 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
446 			return (EROFS);
447 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
448 		    td)) != 0)
449 			return (error);
450 	}
451 	if (vap->va_size != VNOVAL) {
452 		/*
453 		 * Disallow write attempts on read-only file systems;
454 		 * unless the file is a socket, fifo, or a block or
455 		 * character device resident on the file system.
456 		 */
457 		switch (vp->v_type) {
458 		case VDIR:
459 			return (EISDIR);
460 		case VLNK:
461 		case VREG:
462 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
463 				return (EROFS);
464 			break;
465 		default:
466 			break;
467 		}
468 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
469 			return (error);
470 	}
471 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
472 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
473 			return (EROFS);
474 		/*
475 		 * From utimes(2):
476 		 * If times is NULL, ... The caller must be the owner of
477 		 * the file, have permission to write the file, or be the
478 		 * super-user.
479 		 * If times is non-NULL, ... The caller must be the owner of
480 		 * the file or be the super-user.
481 		 */
482 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
483 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
484 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
485 			return (error);
486 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
487 		if (vap->va_atime.tv_sec != VNOVAL) {
488 			ip->i_flag &= ~IN_ACCESS;
489 			ip->i_atime = vap->va_atime.tv_sec;
490 			ip->i_atimensec = vap->va_atime.tv_nsec;
491 		}
492 		if (vap->va_mtime.tv_sec != VNOVAL) {
493 			ip->i_flag &= ~IN_UPDATE;
494 			ip->i_mtime = vap->va_mtime.tv_sec;
495 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
496 		}
497 		ip->i_birthtime = vap->va_birthtime.tv_sec;
498 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
499 		error = ext2_update(vp, 0);
500 		if (error)
501 			return (error);
502 	}
503 	error = 0;
504 	if (vap->va_mode != (mode_t)VNOVAL) {
505 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
506 			return (EROFS);
507 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
508 	}
509 	return (error);
510 }
511 
512 /*
513  * Change the mode on a file.
514  * Inode must be locked before calling.
515  */
516 static int
517 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
518 {
519 	struct inode *ip = VTOI(vp);
520 	int error;
521 
522 	/*
523 	 * To modify the permissions on a file, must possess VADMIN
524 	 * for that file.
525 	 */
526 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
527 		return (error);
528 	/*
529 	 * Privileged processes may set the sticky bit on non-directories,
530 	 * as well as set the setgid bit on a file with a group that the
531 	 * process is not a member of.
532 	 */
533 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
534 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
535 		if (error)
536 			return (EFTYPE);
537 	}
538 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
539 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
540 		if (error)
541 			return (error);
542 	}
543 	ip->i_mode &= ~ALLPERMS;
544 	ip->i_mode |= (mode & ALLPERMS);
545 	ip->i_flag |= IN_CHANGE;
546 	return (0);
547 }
548 
549 /*
550  * Perform chown operation on inode ip;
551  * inode must be locked prior to call.
552  */
553 static int
554 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
555     struct thread *td)
556 {
557 	struct inode *ip = VTOI(vp);
558 	uid_t ouid;
559 	gid_t ogid;
560 	int error = 0;
561 
562 	if (uid == (uid_t)VNOVAL)
563 		uid = ip->i_uid;
564 	if (gid == (gid_t)VNOVAL)
565 		gid = ip->i_gid;
566 	/*
567 	 * To modify the ownership of a file, must possess VADMIN
568 	 * for that file.
569 	 */
570 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
571 		return (error);
572 	/*
573 	 * To change the owner of a file, or change the group of a file
574 	 * to a group of which we are not a member, the caller must
575 	 * have privilege.
576 	 */
577 	if (uid != ip->i_uid || (gid != ip->i_gid &&
578 	    !groupmember(gid, cred))) {
579 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
580 		if (error)
581 			return (error);
582 	}
583 	ogid = ip->i_gid;
584 	ouid = ip->i_uid;
585 	ip->i_gid = gid;
586 	ip->i_uid = uid;
587 	ip->i_flag |= IN_CHANGE;
588 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
589 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
590 			ip->i_mode &= ~(ISUID | ISGID);
591 	}
592 	return (0);
593 }
594 
595 /*
596  * Synch an open file.
597  */
598 /* ARGSUSED */
599 static int
600 ext2_fsync(struct vop_fsync_args *ap)
601 {
602 	/*
603 	 * Flush all dirty buffers associated with a vnode.
604 	 */
605 
606 	vop_stdfsync(ap);
607 
608 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
609 }
610 
611 /*
612  * Mknod vnode call
613  */
614 /* ARGSUSED */
615 static int
616 ext2_mknod(struct vop_mknod_args *ap)
617 {
618 	struct vattr *vap = ap->a_vap;
619 	struct vnode **vpp = ap->a_vpp;
620 	struct inode *ip;
621 	ino_t ino;
622 	int error;
623 
624 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
625 	    ap->a_dvp, vpp, ap->a_cnp);
626 	if (error)
627 		return (error);
628 	ip = VTOI(*vpp);
629 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
630 	if (vap->va_rdev != VNOVAL) {
631 		/*
632 		 * Want to be able to use this to make badblock
633 		 * inodes, so don't truncate the dev number.
634 		 */
635 		if (!(ip->i_flag & IN_E4EXTENTS))
636 			ip->i_rdev = vap->va_rdev;
637 	}
638 	/*
639 	 * Remove inode, then reload it through VFS_VGET so it is
640 	 * checked to see if it is an alias of an existing entry in
641 	 * the inode cache.	 XXX I don't believe this is necessary now.
642 	 */
643 	(*vpp)->v_type = VNON;
644 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
645 	vgone(*vpp);
646 	vput(*vpp);
647 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
648 	if (error) {
649 		*vpp = NULL;
650 		return (error);
651 	}
652 	return (0);
653 }
654 
655 static int
656 ext2_remove(struct vop_remove_args *ap)
657 {
658 	struct inode *ip;
659 	struct vnode *vp = ap->a_vp;
660 	struct vnode *dvp = ap->a_dvp;
661 	int error;
662 
663 	ip = VTOI(vp);
664 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
665 	    (VTOI(dvp)->i_flags & APPEND)) {
666 		error = EPERM;
667 		goto out;
668 	}
669 	error = ext2_dirremove(dvp, ap->a_cnp);
670 	if (error == 0) {
671 		ip->i_nlink--;
672 		ip->i_flag |= IN_CHANGE;
673 	}
674 out:
675 	return (error);
676 }
677 
678 /*
679  * link vnode call
680  */
681 static int
682 ext2_link(struct vop_link_args *ap)
683 {
684 	struct vnode *vp = ap->a_vp;
685 	struct vnode *tdvp = ap->a_tdvp;
686 	struct componentname *cnp = ap->a_cnp;
687 	struct inode *ip;
688 	int error;
689 
690 #ifdef INVARIANTS
691 	if ((cnp->cn_flags & HASBUF) == 0)
692 		panic("ext2_link: no name");
693 #endif
694 	ip = VTOI(vp);
695 	if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) {
696 		error = EMLINK;
697 		goto out;
698 	}
699 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
700 		error = EPERM;
701 		goto out;
702 	}
703 	ip->i_nlink++;
704 	ip->i_flag |= IN_CHANGE;
705 	error = ext2_update(vp, !DOINGASYNC(vp));
706 	if (!error)
707 		error = ext2_direnter(ip, tdvp, cnp);
708 	if (error) {
709 		ip->i_nlink--;
710 		ip->i_flag |= IN_CHANGE;
711 	}
712 out:
713 	return (error);
714 }
715 
716 static int
717 ext2_inc_nlink(struct inode *ip)
718 {
719 
720 	ip->i_nlink++;
721 
722 	if (S_ISDIR(ip->i_mode) &&
723 	    EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) &&
724 	    ip->i_nlink > 1) {
725 		if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2)
726 			ip->i_nlink = 1;
727 	} else if (ip->i_nlink > EXT4_LINK_MAX) {
728 		ip->i_nlink--;
729 		return (EMLINK);
730 	}
731 
732 	return (0);
733 }
734 
735 static void
736 ext2_dec_nlink(struct inode *ip)
737 {
738 
739 	if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
740 		ip->i_nlink--;
741 }
742 
743 /*
744  * Rename system call.
745  * 	rename("foo", "bar");
746  * is essentially
747  *	unlink("bar");
748  *	link("foo", "bar");
749  *	unlink("foo");
750  * but ``atomically''.  Can't do full commit without saving state in the
751  * inode on disk which isn't feasible at this time.  Best we can do is
752  * always guarantee the target exists.
753  *
754  * Basic algorithm is:
755  *
756  * 1) Bump link count on source while we're linking it to the
757  *    target.  This also ensure the inode won't be deleted out
758  *    from underneath us while we work (it may be truncated by
759  *    a concurrent `trunc' or `open' for creation).
760  * 2) Link source to destination.  If destination already exists,
761  *    delete it first.
762  * 3) Unlink source reference to inode if still around. If a
763  *    directory was moved and the parent of the destination
764  *    is different from the source, patch the ".." entry in the
765  *    directory.
766  */
767 static int
768 ext2_rename(struct vop_rename_args *ap)
769 {
770 	struct vnode *tvp = ap->a_tvp;
771 	struct vnode *tdvp = ap->a_tdvp;
772 	struct vnode *fvp = ap->a_fvp;
773 	struct vnode *fdvp = ap->a_fdvp;
774 	struct componentname *tcnp = ap->a_tcnp;
775 	struct componentname *fcnp = ap->a_fcnp;
776 	struct inode *ip, *xp, *dp;
777 	struct dirtemplate *dirbuf;
778 	int doingdirectory = 0, oldparent = 0, newparent = 0;
779 	int error = 0;
780 	u_char namlen;
781 
782 #ifdef INVARIANTS
783 	if ((tcnp->cn_flags & HASBUF) == 0 ||
784 	    (fcnp->cn_flags & HASBUF) == 0)
785 		panic("ext2_rename: no name");
786 #endif
787 	/*
788 	 * Check for cross-device rename.
789 	 */
790 	if ((fvp->v_mount != tdvp->v_mount) ||
791 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
792 		error = EXDEV;
793 abortit:
794 		if (tdvp == tvp)
795 			vrele(tdvp);
796 		else
797 			vput(tdvp);
798 		if (tvp)
799 			vput(tvp);
800 		vrele(fdvp);
801 		vrele(fvp);
802 		return (error);
803 	}
804 
805 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
806 	    (VTOI(tdvp)->i_flags & APPEND))) {
807 		error = EPERM;
808 		goto abortit;
809 	}
810 
811 	/*
812 	 * Renaming a file to itself has no effect.  The upper layers should
813 	 * not call us in that case.  Temporarily just warn if they do.
814 	 */
815 	if (fvp == tvp) {
816 		printf("ext2_rename: fvp == tvp (can't happen)\n");
817 		error = 0;
818 		goto abortit;
819 	}
820 
821 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
822 		goto abortit;
823 	dp = VTOI(fdvp);
824 	ip = VTOI(fvp);
825 	if (ip->i_nlink >= EXT4_LINK_MAX &&
826 	    !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
827 		VOP_UNLOCK(fvp, 0);
828 		error = EMLINK;
829 		goto abortit;
830 	}
831 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
832 	    || (dp->i_flags & APPEND)) {
833 		VOP_UNLOCK(fvp, 0);
834 		error = EPERM;
835 		goto abortit;
836 	}
837 	if ((ip->i_mode & IFMT) == IFDIR) {
838 		/*
839 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
840 		 */
841 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
842 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
843 		    (ip->i_flag & IN_RENAME)) {
844 			VOP_UNLOCK(fvp, 0);
845 			error = EINVAL;
846 			goto abortit;
847 		}
848 		ip->i_flag |= IN_RENAME;
849 		oldparent = dp->i_number;
850 		doingdirectory++;
851 	}
852 	vrele(fdvp);
853 
854 	/*
855 	 * When the target exists, both the directory
856 	 * and target vnodes are returned locked.
857 	 */
858 	dp = VTOI(tdvp);
859 	xp = NULL;
860 	if (tvp)
861 		xp = VTOI(tvp);
862 
863 	/*
864 	 * 1) Bump link count while we're moving stuff
865 	 *    around.  If we crash somewhere before
866 	 *    completing our work, the link count
867 	 *    may be wrong, but correctable.
868 	 */
869 	ext2_inc_nlink(ip);
870 	ip->i_flag |= IN_CHANGE;
871 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
872 		VOP_UNLOCK(fvp, 0);
873 		goto bad;
874 	}
875 
876 	/*
877 	 * If ".." must be changed (ie the directory gets a new
878 	 * parent) then the source directory must not be in the
879 	 * directory hierarchy above the target, as this would
880 	 * orphan everything below the source directory. Also
881 	 * the user must have write permission in the source so
882 	 * as to be able to change "..". We must repeat the call
883 	 * to namei, as the parent directory is unlocked by the
884 	 * call to checkpath().
885 	 */
886 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
887 	VOP_UNLOCK(fvp, 0);
888 	if (oldparent != dp->i_number)
889 		newparent = dp->i_number;
890 	if (doingdirectory && newparent) {
891 		if (error)	/* write access check above */
892 			goto bad;
893 		if (xp != NULL)
894 			vput(tvp);
895 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
896 		if (error)
897 			goto out;
898 		VREF(tdvp);
899 		error = relookup(tdvp, &tvp, tcnp);
900 		if (error)
901 			goto out;
902 		vrele(tdvp);
903 		dp = VTOI(tdvp);
904 		xp = NULL;
905 		if (tvp)
906 			xp = VTOI(tvp);
907 	}
908 	/*
909 	 * 2) If target doesn't exist, link the target
910 	 *    to the source and unlink the source.
911 	 *    Otherwise, rewrite the target directory
912 	 *    entry to reference the source inode and
913 	 *    expunge the original entry's existence.
914 	 */
915 	if (xp == NULL) {
916 		if (dp->i_devvp != ip->i_devvp)
917 			panic("ext2_rename: EXDEV");
918 		/*
919 		 * Account for ".." in new directory.
920 		 * When source and destination have the same
921 		 * parent we don't fool with the link count.
922 		 */
923 		if (doingdirectory && newparent) {
924 			error = ext2_inc_nlink(dp);
925 			if (error)
926 				goto bad;
927 
928 			dp->i_flag |= IN_CHANGE;
929 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
930 			if (error)
931 				goto bad;
932 		}
933 		error = ext2_direnter(ip, tdvp, tcnp);
934 		if (error) {
935 			if (doingdirectory && newparent) {
936 				ext2_dec_nlink(dp);
937 				dp->i_flag |= IN_CHANGE;
938 				(void)ext2_update(tdvp, 1);
939 			}
940 			goto bad;
941 		}
942 		vput(tdvp);
943 	} else {
944 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
945 			panic("ext2_rename: EXDEV");
946 		/*
947 		 * Short circuit rename(foo, foo).
948 		 */
949 		if (xp->i_number == ip->i_number)
950 			panic("ext2_rename: same file");
951 		/*
952 		 * If the parent directory is "sticky", then the user must
953 		 * own the parent directory, or the destination of the rename,
954 		 * otherwise the destination may not be changed (except by
955 		 * root). This implements append-only directories.
956 		 */
957 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
958 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
959 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
960 			error = EPERM;
961 			goto bad;
962 		}
963 		/*
964 		 * Target must be empty if a directory and have no links
965 		 * to it. Also, ensure source and target are compatible
966 		 * (both directories, or both not directories).
967 		 */
968 		if ((xp->i_mode & IFMT) == IFDIR) {
969 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
970 				error = ENOTEMPTY;
971 				goto bad;
972 			}
973 			if (!doingdirectory) {
974 				error = ENOTDIR;
975 				goto bad;
976 			}
977 			cache_purge(tdvp);
978 		} else if (doingdirectory) {
979 			error = EISDIR;
980 			goto bad;
981 		}
982 		error = ext2_dirrewrite(dp, ip, tcnp);
983 		if (error)
984 			goto bad;
985 		/*
986 		 * If the target directory is in the same
987 		 * directory as the source directory,
988 		 * decrement the link count on the parent
989 		 * of the target directory.
990 		 */
991 		if (doingdirectory && !newparent) {
992 			ext2_dec_nlink(dp);
993 			dp->i_flag |= IN_CHANGE;
994 		}
995 		vput(tdvp);
996 		/*
997 		 * Adjust the link count of the target to
998 		 * reflect the dirrewrite above.  If this is
999 		 * a directory it is empty and there are
1000 		 * no links to it, so we can squash the inode and
1001 		 * any space associated with it.  We disallowed
1002 		 * renaming over top of a directory with links to
1003 		 * it above, as the remaining link would point to
1004 		 * a directory without "." or ".." entries.
1005 		 */
1006 		ext2_dec_nlink(xp);
1007 		if (doingdirectory) {
1008 			if (--xp->i_nlink != 0)
1009 				panic("ext2_rename: linked directory");
1010 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1011 			    tcnp->cn_cred, tcnp->cn_thread);
1012 		}
1013 		xp->i_flag |= IN_CHANGE;
1014 		vput(tvp);
1015 		xp = NULL;
1016 	}
1017 
1018 	/*
1019 	 * 3) Unlink the source.
1020 	 */
1021 	fcnp->cn_flags &= ~MODMASK;
1022 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1023 	VREF(fdvp);
1024 	error = relookup(fdvp, &fvp, fcnp);
1025 	if (error == 0)
1026 		vrele(fdvp);
1027 	if (fvp != NULL) {
1028 		xp = VTOI(fvp);
1029 		dp = VTOI(fdvp);
1030 	} else {
1031 		/*
1032 		 * From name has disappeared.  IN_RENAME is not sufficient
1033 		 * to protect against directory races due to timing windows,
1034 		 * so we can't panic here.
1035 		 */
1036 		vrele(ap->a_fvp);
1037 		return (0);
1038 	}
1039 	/*
1040 	 * Ensure that the directory entry still exists and has not
1041 	 * changed while the new name has been entered. If the source is
1042 	 * a file then the entry may have been unlinked or renamed. In
1043 	 * either case there is no further work to be done. If the source
1044 	 * is a directory then it cannot have been rmdir'ed; its link
1045 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1046 	 * The IN_RENAME flag ensures that it cannot be moved by another
1047 	 * rename.
1048 	 */
1049 	if (xp != ip) {
1050 		/*
1051 		 * From name resolves to a different inode.  IN_RENAME is
1052 		 * not sufficient protection against timing window races
1053 		 * so we can't panic here.
1054 		 */
1055 	} else {
1056 		/*
1057 		 * If the source is a directory with a
1058 		 * new parent, the link count of the old
1059 		 * parent directory must be decremented
1060 		 * and ".." set to point to the new parent.
1061 		 */
1062 		if (doingdirectory && newparent) {
1063 			ext2_dec_nlink(dp);
1064 			dp->i_flag |= IN_CHANGE;
1065 			dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1066 			if (!dirbuf) {
1067 				error = ENOMEM;
1068 				goto bad;
1069 			}
1070 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1071 			    ip->i_e2fs->e2fs_bsize, (off_t)0,
1072 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1073 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1074 			if (error == 0) {
1075 				/* Like ufs little-endian: */
1076 				namlen = dirbuf->dotdot_type;
1077 				if (namlen != 2 ||
1078 				    dirbuf->dotdot_name[0] != '.' ||
1079 				    dirbuf->dotdot_name[1] != '.') {
1080 					ext2_dirbad(xp, (doff_t)12,
1081 					    "rename: mangled dir");
1082 				} else {
1083 					dirbuf->dotdot_ino = newparent;
1084 					/*
1085 					 * dirblock 0 could be htree root,
1086 					 * try both csum update functions.
1087 					 */
1088 					ext2_dirent_csum_set(ip,
1089 					    (struct ext2fs_direct_2 *)dirbuf);
1090 					ext2_dx_csum_set(ip,
1091 					    (struct ext2fs_direct_2 *)dirbuf);
1092 					(void)vn_rdwr(UIO_WRITE, fvp,
1093 					    (caddr_t)dirbuf,
1094 					    ip->i_e2fs->e2fs_bsize,
1095 					    (off_t)0, UIO_SYSSPACE,
1096 					    IO_NODELOCKED | IO_SYNC |
1097 					    IO_NOMACCHECK, tcnp->cn_cred,
1098 					    NOCRED, NULL, NULL);
1099 					cache_purge(fdvp);
1100 				}
1101 			}
1102 			free(dirbuf, M_TEMP);
1103 		}
1104 		error = ext2_dirremove(fdvp, fcnp);
1105 		if (!error) {
1106 			ext2_dec_nlink(xp);
1107 			xp->i_flag |= IN_CHANGE;
1108 		}
1109 		xp->i_flag &= ~IN_RENAME;
1110 	}
1111 	if (dp)
1112 		vput(fdvp);
1113 	if (xp)
1114 		vput(fvp);
1115 	vrele(ap->a_fvp);
1116 	return (error);
1117 
1118 bad:
1119 	if (xp)
1120 		vput(ITOV(xp));
1121 	vput(ITOV(dp));
1122 out:
1123 	if (doingdirectory)
1124 		ip->i_flag &= ~IN_RENAME;
1125 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1126 		ext2_dec_nlink(ip);
1127 		ip->i_flag |= IN_CHANGE;
1128 		ip->i_flag &= ~IN_RENAME;
1129 		vput(fvp);
1130 	} else
1131 		vrele(fvp);
1132 	return (error);
1133 }
1134 
1135 #ifdef UFS_ACL
1136 static int
1137 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1138     mode_t dmode, struct ucred *cred, struct thread *td)
1139 {
1140 	int error;
1141 	struct inode *ip = VTOI(tvp);
1142 	struct acl *dacl, *acl;
1143 
1144 	acl = acl_alloc(M_WAITOK);
1145 	dacl = acl_alloc(M_WAITOK);
1146 
1147 	/*
1148 	 * Retrieve default ACL from parent, if any.
1149 	 */
1150 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1151 	switch (error) {
1152 	case 0:
1153 		/*
1154 		 * Retrieved a default ACL, so merge mode and ACL if
1155 		 * necessary.  If the ACL is empty, fall through to
1156 		 * the "not defined or available" case.
1157 		 */
1158 		if (acl->acl_cnt != 0) {
1159 			dmode = acl_posix1e_newfilemode(dmode, acl);
1160 			ip->i_mode = dmode;
1161 			*dacl = *acl;
1162 			ext2_sync_acl_from_inode(ip, acl);
1163 			break;
1164 		}
1165 		/* FALLTHROUGH */
1166 
1167 	case EOPNOTSUPP:
1168 		/*
1169 		 * Just use the mode as-is.
1170 		 */
1171 		ip->i_mode = dmode;
1172 		error = 0;
1173 		goto out;
1174 
1175 	default:
1176 		goto out;
1177 	}
1178 
1179 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1180 	if (error == 0)
1181 		error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1182 	switch (error) {
1183 	case 0:
1184 		break;
1185 
1186 	case EOPNOTSUPP:
1187 		/*
1188 		 * XXX: This should not happen, as EOPNOTSUPP above
1189 		 * was supposed to free acl.
1190 		 */
1191 #ifdef DEBUG
1192 		printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1193 #endif	/* DEBUG */
1194 		break;
1195 
1196 	default:
1197 		goto out;
1198 	}
1199 
1200 out:
1201 	acl_free(acl);
1202 	acl_free(dacl);
1203 
1204 	return (error);
1205 }
1206 
1207 static int
1208 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1209     mode_t mode, struct ucred *cred, struct thread *td)
1210 {
1211 	int error;
1212 	struct inode *ip = VTOI(tvp);
1213 	struct acl *acl;
1214 
1215 	acl = acl_alloc(M_WAITOK);
1216 
1217 	/*
1218 	 * Retrieve default ACL for parent, if any.
1219 	 */
1220 	error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1221 	switch (error) {
1222 	case 0:
1223 		/*
1224 		 * Retrieved a default ACL, so merge mode and ACL if
1225 		 * necessary.
1226 		 */
1227 		if (acl->acl_cnt != 0) {
1228 			/*
1229 			 * Two possible ways for default ACL to not
1230 			 * be present.  First, the EA can be
1231 			 * undefined, or second, the default ACL can
1232 			 * be blank.  If it's blank, fall through to
1233 			 * the it's not defined case.
1234 			 */
1235 			mode = acl_posix1e_newfilemode(mode, acl);
1236 			ip->i_mode = mode;
1237 			ext2_sync_acl_from_inode(ip, acl);
1238 			break;
1239 		}
1240 		/* FALLTHROUGH */
1241 
1242 	case EOPNOTSUPP:
1243 		/*
1244 		 * Just use the mode as-is.
1245 		 */
1246 		ip->i_mode = mode;
1247 		error = 0;
1248 		goto out;
1249 
1250 	default:
1251 		goto out;
1252 	}
1253 
1254 	error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1255 	switch (error) {
1256 	case 0:
1257 		break;
1258 
1259 	case EOPNOTSUPP:
1260 		/*
1261 		 * XXX: This should not happen, as EOPNOTSUPP above was
1262 		 * supposed to free acl.
1263 		 */
1264 		printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1265 		    "but no VOP_SETACL()\n");
1266 		/* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1267 		    "but no VOP_SETACL()"); */
1268 		break;
1269 
1270 	default:
1271 		goto out;
1272 	}
1273 
1274 out:
1275 	acl_free(acl);
1276 
1277 	return (error);
1278 }
1279 
1280 #endif /* UFS_ACL */
1281 
1282 /*
1283  * Mkdir system call
1284  */
1285 static int
1286 ext2_mkdir(struct vop_mkdir_args *ap)
1287 {
1288 	struct m_ext2fs *fs;
1289 	struct vnode *dvp = ap->a_dvp;
1290 	struct vattr *vap = ap->a_vap;
1291 	struct componentname *cnp = ap->a_cnp;
1292 	struct inode *ip, *dp;
1293 	struct vnode *tvp;
1294 	struct dirtemplate dirtemplate, *dtp;
1295 	char *buf = NULL;
1296 	int error, dmode;
1297 
1298 #ifdef INVARIANTS
1299 	if ((cnp->cn_flags & HASBUF) == 0)
1300 		panic("ext2_mkdir: no name");
1301 #endif
1302 	dp = VTOI(dvp);
1303 	if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX &&
1304 	    !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
1305 		error = EMLINK;
1306 		goto out;
1307 	}
1308 	dmode = vap->va_mode & 0777;
1309 	dmode |= IFDIR;
1310 	/*
1311 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1312 	 * but not have it entered in the parent directory. The entry is
1313 	 * made later after writing "." and ".." entries.
1314 	 */
1315 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1316 	if (error)
1317 		goto out;
1318 	ip = VTOI(tvp);
1319 	fs = ip->i_e2fs;
1320 	ip->i_gid = dp->i_gid;
1321 #ifdef SUIDDIR
1322 	{
1323 		/*
1324 		 * if we are hacking owners here, (only do this where told to)
1325 		 * and we are not giving it TOO root, (would subvert quotas)
1326 		 * then go ahead and give it to the other user.
1327 		 * The new directory also inherits the SUID bit.
1328 		 * If user's UID and dir UID are the same,
1329 		 * 'give it away' so that the SUID is still forced on.
1330 		 */
1331 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1332 		    (dp->i_mode & ISUID) && dp->i_uid) {
1333 			dmode |= ISUID;
1334 			ip->i_uid = dp->i_uid;
1335 		} else {
1336 			ip->i_uid = cnp->cn_cred->cr_uid;
1337 		}
1338 	}
1339 #else
1340 	ip->i_uid = cnp->cn_cred->cr_uid;
1341 #endif
1342 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1343 	ip->i_mode = dmode;
1344 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1345 	ip->i_nlink = 2;
1346 	if (cnp->cn_flags & ISWHITEOUT)
1347 		ip->i_flags |= UF_OPAQUE;
1348 	error = ext2_update(tvp, 1);
1349 
1350 	/*
1351 	 * Bump link count in parent directory
1352 	 * to reflect work done below.  Should
1353 	 * be done before reference is created
1354 	 * so reparation is possible if we crash.
1355 	 */
1356 	ext2_inc_nlink(dp);
1357 	dp->i_flag |= IN_CHANGE;
1358 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1359 	if (error)
1360 		goto bad;
1361 
1362 	/* Initialize directory with "." and ".." from static template. */
1363 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1364 	    EXT2F_INCOMPAT_FTYPE))
1365 		dtp = &mastertemplate;
1366 	else
1367 		dtp = &omastertemplate;
1368 	dirtemplate = *dtp;
1369 	dirtemplate.dot_ino = ip->i_number;
1370 	dirtemplate.dotdot_ino = dp->i_number;
1371 	/*
1372 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1373 	 * just redefine it - for this function only
1374 	 */
1375 #undef  DIRBLKSIZ
1376 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1377 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1378 	buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1379 	if (!buf) {
1380 		error = ENOMEM;
1381 		ext2_dec_nlink(dp);
1382 		dp->i_flag |= IN_CHANGE;
1383 		goto bad;
1384 	}
1385 	if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1386 		dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail);
1387 		ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1388 	}
1389 	memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1390 	ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf);
1391 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1392 	    DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1393 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1394 	    NULL, NULL);
1395 	if (error) {
1396 		ext2_dec_nlink(dp);
1397 		dp->i_flag |= IN_CHANGE;
1398 		goto bad;
1399 	}
1400 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1401 		/* XXX should grow with balloc() */
1402 		panic("ext2_mkdir: blksize");
1403 	else {
1404 		ip->i_size = DIRBLKSIZ;
1405 		ip->i_flag |= IN_CHANGE;
1406 	}
1407 
1408 #ifdef UFS_ACL
1409 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1410 		error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1411 		    cnp->cn_cred, cnp->cn_thread);
1412 		if (error)
1413 			goto bad;
1414 	}
1415 
1416 #endif /* UFS_ACL */
1417 
1418 	/* Directory set up, now install its entry in the parent directory. */
1419 	error = ext2_direnter(ip, dvp, cnp);
1420 	if (error) {
1421 		ext2_dec_nlink(dp);
1422 		dp->i_flag |= IN_CHANGE;
1423 	}
1424 bad:
1425 	/*
1426 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1427 	 * for us because we set the link count to 0.
1428 	 */
1429 	if (error) {
1430 		ip->i_nlink = 0;
1431 		ip->i_flag |= IN_CHANGE;
1432 		vput(tvp);
1433 	} else
1434 		*ap->a_vpp = tvp;
1435 out:
1436 	free(buf, M_TEMP);
1437 	return (error);
1438 #undef  DIRBLKSIZ
1439 #define DIRBLKSIZ  DEV_BSIZE
1440 }
1441 
1442 /*
1443  * Rmdir system call.
1444  */
1445 static int
1446 ext2_rmdir(struct vop_rmdir_args *ap)
1447 {
1448 	struct vnode *vp = ap->a_vp;
1449 	struct vnode *dvp = ap->a_dvp;
1450 	struct componentname *cnp = ap->a_cnp;
1451 	struct inode *ip, *dp;
1452 	int error;
1453 
1454 	ip = VTOI(vp);
1455 	dp = VTOI(dvp);
1456 
1457 	/*
1458 	 * Verify the directory is empty (and valid).
1459 	 * (Rmdir ".." won't be valid since
1460 	 *  ".." will contain a reference to
1461 	 *  the current directory and thus be
1462 	 *  non-empty.)
1463 	 */
1464 	if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1465 		error = ENOTEMPTY;
1466 		goto out;
1467 	}
1468 	if ((dp->i_flags & APPEND)
1469 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1470 		error = EPERM;
1471 		goto out;
1472 	}
1473 	/*
1474 	 * Delete reference to directory before purging
1475 	 * inode.  If we crash in between, the directory
1476 	 * will be reattached to lost+found,
1477 	 */
1478 	error = ext2_dirremove(dvp, cnp);
1479 	if (error)
1480 		goto out;
1481 	ext2_dec_nlink(dp);
1482 	dp->i_flag |= IN_CHANGE;
1483 	cache_purge(dvp);
1484 	VOP_UNLOCK(dvp, 0);
1485 	/*
1486 	 * Truncate inode.  The only stuff left
1487 	 * in the directory is "." and "..".
1488 	 */
1489 	ip->i_nlink = 0;
1490 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1491 	    cnp->cn_thread);
1492 	cache_purge(ITOV(ip));
1493 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1494 		VOP_UNLOCK(vp, 0);
1495 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1496 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1497 	}
1498 out:
1499 	return (error);
1500 }
1501 
1502 /*
1503  * symlink -- make a symbolic link
1504  */
1505 static int
1506 ext2_symlink(struct vop_symlink_args *ap)
1507 {
1508 	struct vnode *vp, **vpp = ap->a_vpp;
1509 	struct inode *ip;
1510 	int len, error;
1511 
1512 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1513 	    vpp, ap->a_cnp);
1514 	if (error)
1515 		return (error);
1516 	vp = *vpp;
1517 	len = strlen(ap->a_target);
1518 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1519 		ip = VTOI(vp);
1520 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1521 		ip->i_size = len;
1522 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1523 	} else
1524 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1525 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1526 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1527 	if (error)
1528 		vput(vp);
1529 	return (error);
1530 }
1531 
1532 /*
1533  * Return target name of a symbolic link
1534  */
1535 static int
1536 ext2_readlink(struct vop_readlink_args *ap)
1537 {
1538 	struct vnode *vp = ap->a_vp;
1539 	struct inode *ip = VTOI(vp);
1540 	int isize;
1541 
1542 	isize = ip->i_size;
1543 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1544 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1545 		return (0);
1546 	}
1547 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1548 }
1549 
1550 /*
1551  * Calculate the logical to physical mapping if not done already,
1552  * then call the device strategy routine.
1553  *
1554  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1555  * deadlock on memory.  See ext2_bmap() for details.
1556  */
1557 static int
1558 ext2_strategy(struct vop_strategy_args *ap)
1559 {
1560 	struct buf *bp = ap->a_bp;
1561 	struct vnode *vp = ap->a_vp;
1562 	struct bufobj *bo;
1563 	daddr_t blkno;
1564 	int error;
1565 
1566 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1567 		panic("ext2_strategy: spec");
1568 	if (bp->b_blkno == bp->b_lblkno) {
1569 
1570 		if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1571 			error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1572 		else
1573 			error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1574 
1575 		bp->b_blkno = blkno;
1576 		if (error) {
1577 			bp->b_error = error;
1578 			bp->b_ioflags |= BIO_ERROR;
1579 			bufdone(bp);
1580 			return (0);
1581 		}
1582 		if ((long)bp->b_blkno == -1)
1583 			vfs_bio_clrbuf(bp);
1584 	}
1585 	if ((long)bp->b_blkno == -1) {
1586 		bufdone(bp);
1587 		return (0);
1588 	}
1589 	bp->b_iooffset = dbtob(bp->b_blkno);
1590 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1591 	BO_STRATEGY(bo, bp);
1592 	return (0);
1593 }
1594 
1595 /*
1596  * Print out the contents of an inode.
1597  */
1598 static int
1599 ext2_print(struct vop_print_args *ap)
1600 {
1601 	struct vnode *vp = ap->a_vp;
1602 	struct inode *ip = VTOI(vp);
1603 
1604 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1605 	if (vp->v_type == VFIFO)
1606 		fifo_printinfo(vp);
1607 	printf("\n");
1608 	return (0);
1609 }
1610 
1611 /*
1612  * Close wrapper for fifos.
1613  *
1614  * Update the times on the inode then do device close.
1615  */
1616 static int
1617 ext2fifo_close(struct vop_close_args *ap)
1618 {
1619 	struct vnode *vp = ap->a_vp;
1620 
1621 	VI_LOCK(vp);
1622 	if (vp->v_usecount > 1)
1623 		ext2_itimes_locked(vp);
1624 	VI_UNLOCK(vp);
1625 	return (fifo_specops.vop_close(ap));
1626 }
1627 
1628 /*
1629  * Kqfilter wrapper for fifos.
1630  *
1631  * Fall through to ext2 kqfilter routines if needed
1632  */
1633 static int
1634 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1635 {
1636 	int error;
1637 
1638 	error = fifo_specops.vop_kqfilter(ap);
1639 	if (error)
1640 		error = vfs_kqfilter(ap);
1641 	return (error);
1642 }
1643 
1644 /*
1645  * Return POSIX pathconf information applicable to ext2 filesystems.
1646  */
1647 static int
1648 ext2_pathconf(struct vop_pathconf_args *ap)
1649 {
1650 	int error = 0;
1651 
1652 	switch (ap->a_name) {
1653 	case _PC_LINK_MAX:
1654 		if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs,
1655 		    EXT2F_ROCOMPAT_DIR_NLINK))
1656 			*ap->a_retval = INT_MAX;
1657 		else
1658 			*ap->a_retval = EXT4_LINK_MAX;
1659 		break;
1660 	case _PC_NAME_MAX:
1661 		*ap->a_retval = NAME_MAX;
1662 		break;
1663 	case _PC_PIPE_BUF:
1664 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1665 			*ap->a_retval = PIPE_BUF;
1666 		else
1667 			error = EINVAL;
1668 		break;
1669 	case _PC_CHOWN_RESTRICTED:
1670 		*ap->a_retval = 1;
1671 		break;
1672 	case _PC_NO_TRUNC:
1673 		*ap->a_retval = 1;
1674 		break;
1675 
1676 #ifdef UFS_ACL
1677 	case _PC_ACL_EXTENDED:
1678 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1679 			*ap->a_retval = 1;
1680 		else
1681 			*ap->a_retval = 0;
1682 		break;
1683 	case _PC_ACL_PATH_MAX:
1684 		if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1685 			*ap->a_retval = ACL_MAX_ENTRIES;
1686 		else
1687 			*ap->a_retval = 3;
1688 		break;
1689 #endif /* UFS_ACL */
1690 
1691 	case _PC_MIN_HOLE_SIZE:
1692 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1693 		break;
1694 	case _PC_PRIO_IO:
1695 		*ap->a_retval = 0;
1696 		break;
1697 	case _PC_SYNC_IO:
1698 		*ap->a_retval = 0;
1699 		break;
1700 	case _PC_ALLOC_SIZE_MIN:
1701 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1702 		break;
1703 	case _PC_FILESIZEBITS:
1704 		*ap->a_retval = 64;
1705 		break;
1706 	case _PC_REC_INCR_XFER_SIZE:
1707 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1708 		break;
1709 	case _PC_REC_MAX_XFER_SIZE:
1710 		*ap->a_retval = -1;	/* means ``unlimited'' */
1711 		break;
1712 	case _PC_REC_MIN_XFER_SIZE:
1713 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1714 		break;
1715 	case _PC_REC_XFER_ALIGN:
1716 		*ap->a_retval = PAGE_SIZE;
1717 		break;
1718 	case _PC_SYMLINK_MAX:
1719 		*ap->a_retval = MAXPATHLEN;
1720 		break;
1721 
1722 	default:
1723 		error = vop_stdpathconf(ap);
1724 		break;
1725 	}
1726 	return (error);
1727 }
1728 
1729 /*
1730  * Vnode operation to remove a named attribute.
1731  */
1732 static int
1733 ext2_deleteextattr(struct vop_deleteextattr_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, VWRITE);
1750 	if (error)
1751 		return (error);
1752 
1753 	error = ENOATTR;
1754 
1755 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1756 		error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1757 		if (error != ENOATTR)
1758 			return (error);
1759 	}
1760 
1761 	if (ip->i_facl)
1762 		error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1763 
1764 	return (error);
1765 }
1766 
1767 /*
1768  * Vnode operation to retrieve a named extended attribute.
1769  */
1770 static int
1771 ext2_getextattr(struct vop_getextattr_args *ap)
1772 {
1773 	struct inode *ip;
1774 	struct m_ext2fs *fs;
1775 	int error;
1776 
1777 	ip = VTOI(ap->a_vp);
1778 	fs = ip->i_e2fs;
1779 
1780 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1781 		return (EOPNOTSUPP);
1782 
1783 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1784 		return (EOPNOTSUPP);
1785 
1786 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1787 	    ap->a_cred, ap->a_td, VREAD);
1788 	if (error)
1789 		return (error);
1790 
1791 	if (ap->a_size != NULL)
1792 		*ap->a_size = 0;
1793 
1794 	error = ENOATTR;
1795 
1796 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1797 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1798 		    ap->a_name, ap->a_uio, ap->a_size);
1799 		if (error != ENOATTR)
1800 			return (error);
1801 	}
1802 
1803 	if (ip->i_facl)
1804 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1805 		    ap->a_name, ap->a_uio, ap->a_size);
1806 
1807 	return (error);
1808 }
1809 
1810 /*
1811  * Vnode operation to retrieve extended attributes on a vnode.
1812  */
1813 static int
1814 ext2_listextattr(struct vop_listextattr_args *ap)
1815 {
1816 	struct inode *ip;
1817 	struct m_ext2fs *fs;
1818 	int error;
1819 
1820 	ip = VTOI(ap->a_vp);
1821 	fs = ip->i_e2fs;
1822 
1823 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1824 		return (EOPNOTSUPP);
1825 
1826 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1827 		return (EOPNOTSUPP);
1828 
1829 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1830 	    ap->a_cred, ap->a_td, VREAD);
1831 	if (error)
1832 		return (error);
1833 
1834 	if (ap->a_size != NULL)
1835 		*ap->a_size = 0;
1836 
1837 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1838 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1839 		    ap->a_uio, ap->a_size);
1840 		if (error)
1841 			return (error);
1842 	}
1843 
1844 	if (ip->i_facl)
1845 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1846 		    ap->a_uio, ap->a_size);
1847 
1848 	return (error);
1849 }
1850 
1851 /*
1852  * Vnode operation to set a named attribute.
1853  */
1854 static int
1855 ext2_setextattr(struct vop_setextattr_args *ap)
1856 {
1857 	struct inode *ip;
1858 	struct m_ext2fs *fs;
1859 	int error;
1860 
1861 	ip = VTOI(ap->a_vp);
1862 	fs = ip->i_e2fs;
1863 
1864 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1865 		return (EOPNOTSUPP);
1866 
1867 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1868 		return (EOPNOTSUPP);
1869 
1870 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1871 	    ap->a_cred, ap->a_td, VWRITE);
1872 	if (error)
1873 		return (error);
1874 
1875 	error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1876 	if (error)
1877 		return (error);
1878 
1879 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1880 		error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1881 		    ap->a_name, ap->a_uio);
1882 		if (error != ENOSPC)
1883 			return (error);
1884 	}
1885 
1886 	error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1887 	    ap->a_name, ap->a_uio);
1888 
1889 	return (error);
1890 }
1891 
1892 /*
1893  * Vnode pointer to File handle
1894  */
1895 /* ARGSUSED */
1896 static int
1897 ext2_vptofh(struct vop_vptofh_args *ap)
1898 {
1899 	struct inode *ip;
1900 	struct ufid *ufhp;
1901 
1902 	ip = VTOI(ap->a_vp);
1903 	ufhp = (struct ufid *)ap->a_fhp;
1904 	ufhp->ufid_len = sizeof(struct ufid);
1905 	ufhp->ufid_ino = ip->i_number;
1906 	ufhp->ufid_gen = ip->i_gen;
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Initialize the vnode associated with a new inode, handle aliased
1912  * vnodes.
1913  */
1914 int
1915 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1916 {
1917 	struct inode *ip;
1918 	struct vnode *vp;
1919 
1920 	vp = *vpp;
1921 	ip = VTOI(vp);
1922 	vp->v_type = IFTOVT(ip->i_mode);
1923 	if (vp->v_type == VFIFO)
1924 		vp->v_op = fifoops;
1925 
1926 	if (ip->i_number == EXT2_ROOTINO)
1927 		vp->v_vflag |= VV_ROOT;
1928 	ip->i_modrev = init_va_filerev();
1929 	*vpp = vp;
1930 	return (0);
1931 }
1932 
1933 /*
1934  * Allocate a new inode.
1935  */
1936 static int
1937 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1938     struct componentname *cnp)
1939 {
1940 	struct inode *ip, *pdir;
1941 	struct vnode *tvp;
1942 	int error;
1943 
1944 	pdir = VTOI(dvp);
1945 #ifdef INVARIANTS
1946 	if ((cnp->cn_flags & HASBUF) == 0)
1947 		panic("ext2_makeinode: no name");
1948 #endif
1949 	*vpp = NULL;
1950 	if ((mode & IFMT) == 0)
1951 		mode |= IFREG;
1952 
1953 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1954 	if (error) {
1955 		return (error);
1956 	}
1957 	ip = VTOI(tvp);
1958 	ip->i_gid = pdir->i_gid;
1959 #ifdef SUIDDIR
1960 	{
1961 		/*
1962 		 * if we are
1963 		 * not the owner of the directory,
1964 		 * and we are hacking owners here, (only do this where told to)
1965 		 * and we are not giving it TOO root, (would subvert quotas)
1966 		 * then go ahead and give it to the other user.
1967 		 * Note that this drops off the execute bits for security.
1968 		 */
1969 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1970 		    (pdir->i_mode & ISUID) &&
1971 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1972 			ip->i_uid = pdir->i_uid;
1973 			mode &= ~07111;
1974 		} else {
1975 			ip->i_uid = cnp->cn_cred->cr_uid;
1976 		}
1977 	}
1978 #else
1979 	ip->i_uid = cnp->cn_cred->cr_uid;
1980 #endif
1981 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1982 	ip->i_mode = mode;
1983 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1984 	ip->i_nlink = 1;
1985 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1986 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1987 			ip->i_mode &= ~ISGID;
1988 	}
1989 
1990 	if (cnp->cn_flags & ISWHITEOUT)
1991 		ip->i_flags |= UF_OPAQUE;
1992 
1993 	/*
1994 	 * Make sure inode goes to disk before directory entry.
1995 	 */
1996 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1997 	if (error)
1998 		goto bad;
1999 
2000 #ifdef UFS_ACL
2001 	if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2002 		error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2003 		    cnp->cn_cred, cnp->cn_thread);
2004 		if (error)
2005 			goto bad;
2006 	}
2007 #endif /* UFS_ACL */
2008 
2009 	error = ext2_direnter(ip, dvp, cnp);
2010 	if (error)
2011 		goto bad;
2012 
2013 	*vpp = tvp;
2014 	return (0);
2015 
2016 bad:
2017 	/*
2018 	 * Write error occurred trying to update the inode
2019 	 * or the directory so must deallocate the inode.
2020 	 */
2021 	ip->i_nlink = 0;
2022 	ip->i_flag |= IN_CHANGE;
2023 	vput(tvp);
2024 	return (error);
2025 }
2026 
2027 /*
2028  * Vnode op for reading.
2029  */
2030 static int
2031 ext2_read(struct vop_read_args *ap)
2032 {
2033 	struct vnode *vp;
2034 	struct inode *ip;
2035 	struct uio *uio;
2036 	struct m_ext2fs *fs;
2037 	struct buf *bp;
2038 	daddr_t lbn, nextlbn;
2039 	off_t bytesinfile;
2040 	long size, xfersize, blkoffset;
2041 	int error, orig_resid, seqcount;
2042 	int ioflag;
2043 
2044 	vp = ap->a_vp;
2045 	uio = ap->a_uio;
2046 	ioflag = ap->a_ioflag;
2047 
2048 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2049 	ip = VTOI(vp);
2050 
2051 #ifdef INVARIANTS
2052 	if (uio->uio_rw != UIO_READ)
2053 		panic("%s: mode", "ext2_read");
2054 
2055 	if (vp->v_type == VLNK) {
2056 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
2057 			panic("%s: short symlink", "ext2_read");
2058 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
2059 		panic("%s: type %d", "ext2_read", vp->v_type);
2060 #endif
2061 	orig_resid = uio->uio_resid;
2062 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2063 	if (orig_resid == 0)
2064 		return (0);
2065 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2066 	fs = ip->i_e2fs;
2067 	if (uio->uio_offset < ip->i_size &&
2068 	    uio->uio_offset >= fs->e2fs_maxfilesize)
2069 		return (EOVERFLOW);
2070 
2071 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2072 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2073 			break;
2074 		lbn = lblkno(fs, uio->uio_offset);
2075 		nextlbn = lbn + 1;
2076 		size = blksize(fs, ip, lbn);
2077 		blkoffset = blkoff(fs, uio->uio_offset);
2078 
2079 		xfersize = fs->e2fs_fsize - blkoffset;
2080 		if (uio->uio_resid < xfersize)
2081 			xfersize = uio->uio_resid;
2082 		if (bytesinfile < xfersize)
2083 			xfersize = bytesinfile;
2084 
2085 		if (lblktosize(fs, nextlbn) >= ip->i_size)
2086 			error = bread(vp, lbn, size, NOCRED, &bp);
2087 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2088 			error = cluster_read(vp, ip->i_size, lbn, size,
2089 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
2090 			    0, &bp);
2091 		} else if (seqcount > 1) {
2092 			u_int nextsize = blksize(fs, ip, nextlbn);
2093 
2094 			error = breadn(vp, lbn,
2095 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2096 		} else
2097 			error = bread(vp, lbn, size, NOCRED, &bp);
2098 		if (error) {
2099 			brelse(bp);
2100 			bp = NULL;
2101 			break;
2102 		}
2103 
2104 		/*
2105 		 * We should only get non-zero b_resid when an I/O error
2106 		 * has occurred, which should cause us to break above.
2107 		 * However, if the short read did not cause an error,
2108 		 * then we want to ensure that we do not uiomove bad
2109 		 * or uninitialized data.
2110 		 */
2111 		size -= bp->b_resid;
2112 		if (size < xfersize) {
2113 			if (size == 0)
2114 				break;
2115 			xfersize = size;
2116 		}
2117 		error = uiomove((char *)bp->b_data + blkoffset,
2118 		    (int)xfersize, uio);
2119 		if (error)
2120 			break;
2121 		vfs_bio_brelse(bp, ioflag);
2122 	}
2123 
2124 	/*
2125 	 * This can only happen in the case of an error because the loop
2126 	 * above resets bp to NULL on each iteration and on normal
2127 	 * completion has not set a new value into it. so it must have come
2128 	 * from a 'break' statement
2129 	 */
2130 	if (bp != NULL)
2131 		vfs_bio_brelse(bp, ioflag);
2132 
2133 	if ((error == 0 || uio->uio_resid != orig_resid) &&
2134 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2135 		ip->i_flag |= IN_ACCESS;
2136 	return (error);
2137 }
2138 
2139 static int
2140 ext2_ioctl(struct vop_ioctl_args *ap)
2141 {
2142 
2143 	switch (ap->a_command) {
2144 	case FIOSEEKDATA:
2145 	case FIOSEEKHOLE:
2146 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
2147 		    (off_t *)ap->a_data, ap->a_cred));
2148 	default:
2149 		return (ENOTTY);
2150 	}
2151 }
2152 
2153 /*
2154  * Vnode op for writing.
2155  */
2156 static int
2157 ext2_write(struct vop_write_args *ap)
2158 {
2159 	struct vnode *vp;
2160 	struct uio *uio;
2161 	struct inode *ip;
2162 	struct m_ext2fs *fs;
2163 	struct buf *bp;
2164 	daddr_t lbn;
2165 	off_t osize;
2166 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2167 
2168 	ioflag = ap->a_ioflag;
2169 	uio = ap->a_uio;
2170 	vp = ap->a_vp;
2171 
2172 	seqcount = ioflag >> IO_SEQSHIFT;
2173 	ip = VTOI(vp);
2174 
2175 #ifdef INVARIANTS
2176 	if (uio->uio_rw != UIO_WRITE)
2177 		panic("%s: mode", "ext2_write");
2178 #endif
2179 
2180 	switch (vp->v_type) {
2181 	case VREG:
2182 		if (ioflag & IO_APPEND)
2183 			uio->uio_offset = ip->i_size;
2184 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2185 			return (EPERM);
2186 		/* FALLTHROUGH */
2187 	case VLNK:
2188 		break;
2189 	case VDIR:
2190 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
2191 		if ((ioflag & IO_SYNC) == 0)
2192 			panic("ext2_write: nonsync dir write");
2193 		break;
2194 	default:
2195 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2196 		    vp->v_type, (intmax_t)uio->uio_offset,
2197 		    (intmax_t)uio->uio_resid);
2198 	}
2199 
2200 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2201 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2202 	fs = ip->i_e2fs;
2203 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2204 		return (EFBIG);
2205 	/*
2206 	 * Maybe this should be above the vnode op call, but so long as
2207 	 * file servers have no limits, I don't think it matters.
2208 	 */
2209 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2210 		return (EFBIG);
2211 
2212 	resid = uio->uio_resid;
2213 	osize = ip->i_size;
2214 	if (seqcount > BA_SEQMAX)
2215 		flags = BA_SEQMAX << BA_SEQSHIFT;
2216 	else
2217 		flags = seqcount << BA_SEQSHIFT;
2218 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2219 		flags |= IO_SYNC;
2220 
2221 	for (error = 0; uio->uio_resid > 0;) {
2222 		lbn = lblkno(fs, uio->uio_offset);
2223 		blkoffset = blkoff(fs, uio->uio_offset);
2224 		xfersize = fs->e2fs_fsize - blkoffset;
2225 		if (uio->uio_resid < xfersize)
2226 			xfersize = uio->uio_resid;
2227 		if (uio->uio_offset + xfersize > ip->i_size)
2228 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2229 
2230 		/*
2231 		 * We must perform a read-before-write if the transfer size
2232 		 * does not cover the entire buffer.
2233 		 */
2234 		if (fs->e2fs_bsize > xfersize)
2235 			flags |= BA_CLRBUF;
2236 		else
2237 			flags &= ~BA_CLRBUF;
2238 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2239 		    ap->a_cred, &bp, flags);
2240 		if (error != 0)
2241 			break;
2242 
2243 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2244 			bp->b_flags |= B_NOCACHE;
2245 		if (uio->uio_offset + xfersize > ip->i_size)
2246 			ip->i_size = uio->uio_offset + xfersize;
2247 		size = blksize(fs, ip, lbn) - bp->b_resid;
2248 		if (size < xfersize)
2249 			xfersize = size;
2250 
2251 		error =
2252 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2253 		/*
2254 		 * If the buffer is not already filled and we encounter an
2255 		 * error while trying to fill it, we have to clear out any
2256 		 * garbage data from the pages instantiated for the buffer.
2257 		 * If we do not, a failed uiomove() during a write can leave
2258 		 * the prior contents of the pages exposed to a userland mmap.
2259 		 *
2260 		 * Note that we need only clear buffers with a transfer size
2261 		 * equal to the block size because buffers with a shorter
2262 		 * transfer size were cleared above by the call to ext2_balloc()
2263 		 * with the BA_CLRBUF flag set.
2264 		 *
2265 		 * If the source region for uiomove identically mmaps the
2266 		 * buffer, uiomove() performed the NOP copy, and the buffer
2267 		 * content remains valid because the page fault handler
2268 		 * validated the pages.
2269 		 */
2270 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2271 		    fs->e2fs_bsize == xfersize)
2272 			vfs_bio_clrbuf(bp);
2273 
2274 		vfs_bio_set_flags(bp, ioflag);
2275 
2276 		/*
2277 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2278 		 * if we have a severe page deficiency write the buffer
2279 		 * asynchronously.  Otherwise try to cluster, and if that
2280 		 * doesn't do it then either do an async write (if O_DIRECT),
2281 		 * or a delayed write (if not).
2282 		 */
2283 		if (ioflag & IO_SYNC) {
2284 			(void)bwrite(bp);
2285 		} else if (vm_page_count_severe() ||
2286 			    buf_dirty_count_severe() ||
2287 		    (ioflag & IO_ASYNC)) {
2288 			bp->b_flags |= B_CLUSTEROK;
2289 			bawrite(bp);
2290 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2291 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2292 				bp->b_flags |= B_CLUSTEROK;
2293 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2294 			} else {
2295 				bawrite(bp);
2296 			}
2297 		} else if (ioflag & IO_DIRECT) {
2298 			bp->b_flags |= B_CLUSTEROK;
2299 			bawrite(bp);
2300 		} else {
2301 			bp->b_flags |= B_CLUSTEROK;
2302 			bdwrite(bp);
2303 		}
2304 		if (error || xfersize == 0)
2305 			break;
2306 	}
2307 	/*
2308 	 * If we successfully wrote any data, and we are not the superuser
2309 	 * we clear the setuid and setgid bits as a precaution against
2310 	 * tampering.
2311 	 */
2312 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2313 	    ap->a_cred) {
2314 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2315 			ip->i_mode &= ~(ISUID | ISGID);
2316 	}
2317 	if (error) {
2318 		if (ioflag & IO_UNIT) {
2319 			(void)ext2_truncate(vp, osize,
2320 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2321 			uio->uio_offset -= resid - uio->uio_resid;
2322 			uio->uio_resid = resid;
2323 		}
2324 	}
2325 	if (uio->uio_resid != resid) {
2326 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2327 		if (ioflag & IO_SYNC)
2328 			error = ext2_update(vp, 1);
2329 	}
2330 	return (error);
2331 }
2332