xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 5944f899a2519c6321bac3c17cc076418643a088)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * Copyright (c) 1982, 1986, 1989, 1993
9  *	The Regents of the University of California.  All rights reserved.
10  * (c) UNIX System Laboratories, Inc.
11  * All or some portions of this file are derived from material licensed
12  * to the University of California by American Telephone and Telegraph
13  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14  * the permission of UNIX System Laboratories, Inc.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
41  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42  * $FreeBSD$
43  */
44 
45 #include "opt_suiddir.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
60 #include <sys/time.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 #include <sys/extattr.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/vm_extern.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vnode_pager.h>
76 
77 #include "opt_directio.h"
78 
79 #include <ufs/ufs/dir.h>
80 
81 #include <fs/ext2fs/fs.h>
82 #include <fs/ext2fs/inode.h>
83 #include <fs/ext2fs/ext2_extern.h>
84 #include <fs/ext2fs/ext2fs.h>
85 #include <fs/ext2fs/ext2_dinode.h>
86 #include <fs/ext2fs/ext2_dir.h>
87 #include <fs/ext2fs/ext2_mount.h>
88 #include <fs/ext2fs/ext2_extattr.h>
89 
90 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
91 static void ext2_itimes_locked(struct vnode *);
92 static int ext4_ext_read(struct vop_read_args *);
93 static int ext2_ind_read(struct vop_read_args *);
94 
95 static vop_access_t	ext2_access;
96 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
97 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
98     struct thread *);
99 static vop_close_t	ext2_close;
100 static vop_create_t	ext2_create;
101 static vop_fsync_t	ext2_fsync;
102 static vop_getattr_t	ext2_getattr;
103 static vop_ioctl_t	ext2_ioctl;
104 static vop_link_t	ext2_link;
105 static vop_mkdir_t	ext2_mkdir;
106 static vop_mknod_t	ext2_mknod;
107 static vop_open_t	ext2_open;
108 static vop_pathconf_t	ext2_pathconf;
109 static vop_print_t	ext2_print;
110 static vop_read_t	ext2_read;
111 static vop_readlink_t	ext2_readlink;
112 static vop_remove_t	ext2_remove;
113 static vop_rename_t	ext2_rename;
114 static vop_rmdir_t	ext2_rmdir;
115 static vop_setattr_t	ext2_setattr;
116 static vop_strategy_t	ext2_strategy;
117 static vop_symlink_t	ext2_symlink;
118 static vop_write_t	ext2_write;
119 static vop_getextattr_t	ext2_getextattr;
120 static vop_listextattr_t	ext2_listextattr;
121 static vop_vptofh_t	ext2_vptofh;
122 static vop_close_t	ext2fifo_close;
123 static vop_kqfilter_t	ext2fifo_kqfilter;
124 
125 /* Global vfs data structures for ext2. */
126 struct vop_vector ext2_vnodeops = {
127 	.vop_default =		&default_vnodeops,
128 	.vop_access =		ext2_access,
129 	.vop_bmap =		ext2_bmap,
130 	.vop_cachedlookup =	ext2_lookup,
131 	.vop_close =		ext2_close,
132 	.vop_create =		ext2_create,
133 	.vop_fsync =		ext2_fsync,
134 	.vop_getpages =		vnode_pager_local_getpages,
135 	.vop_getpages_async =	vnode_pager_local_getpages_async,
136 	.vop_getattr =		ext2_getattr,
137 	.vop_inactive =		ext2_inactive,
138 	.vop_ioctl =		ext2_ioctl,
139 	.vop_link =		ext2_link,
140 	.vop_lookup =		vfs_cache_lookup,
141 	.vop_mkdir =		ext2_mkdir,
142 	.vop_mknod =		ext2_mknod,
143 	.vop_open =		ext2_open,
144 	.vop_pathconf =		ext2_pathconf,
145 	.vop_poll =		vop_stdpoll,
146 	.vop_print =		ext2_print,
147 	.vop_read =		ext2_read,
148 	.vop_readdir =		ext2_readdir,
149 	.vop_readlink =		ext2_readlink,
150 	.vop_reallocblks =	ext2_reallocblks,
151 	.vop_reclaim =		ext2_reclaim,
152 	.vop_remove =		ext2_remove,
153 	.vop_rename =		ext2_rename,
154 	.vop_rmdir =		ext2_rmdir,
155 	.vop_setattr =		ext2_setattr,
156 	.vop_strategy =		ext2_strategy,
157 	.vop_symlink =		ext2_symlink,
158 	.vop_write =		ext2_write,
159 	.vop_getextattr =	ext2_getextattr,
160 	.vop_listextattr =	ext2_listextattr,
161 	.vop_vptofh =		ext2_vptofh,
162 };
163 
164 struct vop_vector ext2_fifoops = {
165 	.vop_default =		&fifo_specops,
166 	.vop_access =		ext2_access,
167 	.vop_close =		ext2fifo_close,
168 	.vop_fsync =		ext2_fsync,
169 	.vop_getattr =		ext2_getattr,
170 	.vop_inactive =		ext2_inactive,
171 	.vop_kqfilter =		ext2fifo_kqfilter,
172 	.vop_print =		ext2_print,
173 	.vop_read =		VOP_PANIC,
174 	.vop_reclaim =		ext2_reclaim,
175 	.vop_setattr =		ext2_setattr,
176 	.vop_write =		VOP_PANIC,
177 	.vop_vptofh =		ext2_vptofh,
178 };
179 
180 /*
181  * A virgin directory (no blushing please).
182  * Note that the type and namlen fields are reversed relative to ext2.
183  * Also, we don't use `struct odirtemplate', since it would just cause
184  * endianness problems.
185  */
186 static struct dirtemplate mastertemplate = {
187 	0, 12, 1, EXT2_FT_DIR, ".",
188 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
189 };
190 static struct dirtemplate omastertemplate = {
191 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
192 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
193 };
194 
195 static void
196 ext2_itimes_locked(struct vnode *vp)
197 {
198 	struct inode *ip;
199 	struct timespec ts;
200 
201 	ASSERT_VI_LOCKED(vp, __func__);
202 
203 	ip = VTOI(vp);
204 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
205 		return;
206 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
207 		ip->i_flag |= IN_LAZYMOD;
208 	else
209 		ip->i_flag |= IN_MODIFIED;
210 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
211 		vfs_timestamp(&ts);
212 		if (ip->i_flag & IN_ACCESS) {
213 			ip->i_atime = ts.tv_sec;
214 			ip->i_atimensec = ts.tv_nsec;
215 		}
216 		if (ip->i_flag & IN_UPDATE) {
217 			ip->i_mtime = ts.tv_sec;
218 			ip->i_mtimensec = ts.tv_nsec;
219 			ip->i_modrev++;
220 		}
221 		if (ip->i_flag & IN_CHANGE) {
222 			ip->i_ctime = ts.tv_sec;
223 			ip->i_ctimensec = ts.tv_nsec;
224 		}
225 	}
226 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
227 }
228 
229 void
230 ext2_itimes(struct vnode *vp)
231 {
232 
233 	VI_LOCK(vp);
234 	ext2_itimes_locked(vp);
235 	VI_UNLOCK(vp);
236 }
237 
238 /*
239  * Create a regular file
240  */
241 static int
242 ext2_create(struct vop_create_args *ap)
243 {
244 	int error;
245 
246 	error =
247 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
248 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
249 	if (error != 0)
250 		return (error);
251 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
252 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
253 	return (0);
254 }
255 
256 static int
257 ext2_open(struct vop_open_args *ap)
258 {
259 
260 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
261 		return (EOPNOTSUPP);
262 
263 	/*
264 	 * Files marked append-only must be opened for appending.
265 	 */
266 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
267 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
268 		return (EPERM);
269 
270 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
271 
272 	return (0);
273 }
274 
275 /*
276  * Close called.
277  *
278  * Update the times on the inode.
279  */
280 static int
281 ext2_close(struct vop_close_args *ap)
282 {
283 	struct vnode *vp = ap->a_vp;
284 
285 	VI_LOCK(vp);
286 	if (vp->v_usecount > 1)
287 		ext2_itimes_locked(vp);
288 	VI_UNLOCK(vp);
289 	return (0);
290 }
291 
292 static int
293 ext2_access(struct vop_access_args *ap)
294 {
295 	struct vnode *vp = ap->a_vp;
296 	struct inode *ip = VTOI(vp);
297 	accmode_t accmode = ap->a_accmode;
298 	int error;
299 
300 	if (vp->v_type == VBLK || vp->v_type == VCHR)
301 		return (EOPNOTSUPP);
302 
303 	/*
304 	 * Disallow write attempts on read-only file systems;
305 	 * unless the file is a socket, fifo, or a block or
306 	 * character device resident on the file system.
307 	 */
308 	if (accmode & VWRITE) {
309 		switch (vp->v_type) {
310 		case VDIR:
311 		case VLNK:
312 		case VREG:
313 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
314 				return (EROFS);
315 			break;
316 		default:
317 			break;
318 		}
319 	}
320 
321 	/* If immutable bit set, nobody gets to write it. */
322 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
323 		return (EPERM);
324 
325 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
326 	    ap->a_accmode, ap->a_cred, NULL);
327 	return (error);
328 }
329 
330 static int
331 ext2_getattr(struct vop_getattr_args *ap)
332 {
333 	struct vnode *vp = ap->a_vp;
334 	struct inode *ip = VTOI(vp);
335 	struct vattr *vap = ap->a_vap;
336 
337 	ext2_itimes(vp);
338 	/*
339 	 * Copy from inode table
340 	 */
341 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
342 	vap->va_fileid = ip->i_number;
343 	vap->va_mode = ip->i_mode & ~IFMT;
344 	vap->va_nlink = ip->i_nlink;
345 	vap->va_uid = ip->i_uid;
346 	vap->va_gid = ip->i_gid;
347 	vap->va_rdev = ip->i_rdev;
348 	vap->va_size = ip->i_size;
349 	vap->va_atime.tv_sec = ip->i_atime;
350 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
351 	vap->va_mtime.tv_sec = ip->i_mtime;
352 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
353 	vap->va_ctime.tv_sec = ip->i_ctime;
354 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
355 	if E2DI_HAS_XTIME(ip) {
356 		vap->va_birthtime.tv_sec = ip->i_birthtime;
357 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
358 	}
359 	vap->va_flags = ip->i_flags;
360 	vap->va_gen = ip->i_gen;
361 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
362 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
363 	vap->va_type = IFTOVT(ip->i_mode);
364 	vap->va_filerev = ip->i_modrev;
365 	return (0);
366 }
367 
368 /*
369  * Set attribute vnode op. called from several syscalls
370  */
371 static int
372 ext2_setattr(struct vop_setattr_args *ap)
373 {
374 	struct vattr *vap = ap->a_vap;
375 	struct vnode *vp = ap->a_vp;
376 	struct inode *ip = VTOI(vp);
377 	struct ucred *cred = ap->a_cred;
378 	struct thread *td = curthread;
379 	int error;
380 
381 	/*
382 	 * Check for unsettable attributes.
383 	 */
384 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
385 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
386 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
387 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
388 		return (EINVAL);
389 	}
390 	if (vap->va_flags != VNOVAL) {
391 		/* Disallow flags not supported by ext2fs. */
392 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
393 			return (EOPNOTSUPP);
394 
395 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
396 			return (EROFS);
397 		/*
398 		 * Callers may only modify the file flags on objects they
399 		 * have VADMIN rights for.
400 		 */
401 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
402 			return (error);
403 		/*
404 		 * Unprivileged processes and privileged processes in
405 		 * jail() are not permitted to unset system flags, or
406 		 * modify flags if any system flags are set.
407 		 * Privileged non-jail processes may not modify system flags
408 		 * if securelevel > 0 and any existing system flags are set.
409 		 */
410 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
411 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
412 				error = securelevel_gt(cred, 0);
413 				if (error)
414 					return (error);
415 			}
416 		} else {
417 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
418 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
419 				return (EPERM);
420 		}
421 		ip->i_flags = vap->va_flags;
422 		ip->i_flag |= IN_CHANGE;
423 		if (ip->i_flags & (IMMUTABLE | APPEND))
424 			return (0);
425 	}
426 	if (ip->i_flags & (IMMUTABLE | APPEND))
427 		return (EPERM);
428 	/*
429 	 * Go through the fields and update iff not VNOVAL.
430 	 */
431 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
432 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
433 			return (EROFS);
434 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
435 		    td)) != 0)
436 			return (error);
437 	}
438 	if (vap->va_size != VNOVAL) {
439 		/*
440 		 * Disallow write attempts on read-only file systems;
441 		 * unless the file is a socket, fifo, or a block or
442 		 * character device resident on the file system.
443 		 */
444 		switch (vp->v_type) {
445 		case VDIR:
446 			return (EISDIR);
447 		case VLNK:
448 		case VREG:
449 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
450 				return (EROFS);
451 			break;
452 		default:
453 			break;
454 		}
455 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
456 			return (error);
457 	}
458 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
459 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
460 			return (EROFS);
461 		/*
462 		 * From utimes(2):
463 		 * If times is NULL, ... The caller must be the owner of
464 		 * the file, have permission to write the file, or be the
465 		 * super-user.
466 		 * If times is non-NULL, ... The caller must be the owner of
467 		 * the file or be the super-user.
468 		 */
469 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
470 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
471 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
472 			return (error);
473 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
474 		if (vap->va_atime.tv_sec != VNOVAL) {
475 			ip->i_flag &= ~IN_ACCESS;
476 			ip->i_atime = vap->va_atime.tv_sec;
477 			ip->i_atimensec = vap->va_atime.tv_nsec;
478 		}
479 		if (vap->va_mtime.tv_sec != VNOVAL) {
480 			ip->i_flag &= ~IN_UPDATE;
481 			ip->i_mtime = vap->va_mtime.tv_sec;
482 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
483 		}
484 		ip->i_birthtime = vap->va_birthtime.tv_sec;
485 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
486 		error = ext2_update(vp, 0);
487 		if (error)
488 			return (error);
489 	}
490 	error = 0;
491 	if (vap->va_mode != (mode_t)VNOVAL) {
492 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
493 			return (EROFS);
494 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
495 	}
496 	return (error);
497 }
498 
499 /*
500  * Change the mode on a file.
501  * Inode must be locked before calling.
502  */
503 static int
504 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
505 {
506 	struct inode *ip = VTOI(vp);
507 	int error;
508 
509 	/*
510 	 * To modify the permissions on a file, must possess VADMIN
511 	 * for that file.
512 	 */
513 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
514 		return (error);
515 	/*
516 	 * Privileged processes may set the sticky bit on non-directories,
517 	 * as well as set the setgid bit on a file with a group that the
518 	 * process is not a member of.
519 	 */
520 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
521 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
522 		if (error)
523 			return (EFTYPE);
524 	}
525 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
526 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
527 		if (error)
528 			return (error);
529 	}
530 	ip->i_mode &= ~ALLPERMS;
531 	ip->i_mode |= (mode & ALLPERMS);
532 	ip->i_flag |= IN_CHANGE;
533 	return (0);
534 }
535 
536 /*
537  * Perform chown operation on inode ip;
538  * inode must be locked prior to call.
539  */
540 static int
541 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
542     struct thread *td)
543 {
544 	struct inode *ip = VTOI(vp);
545 	uid_t ouid;
546 	gid_t ogid;
547 	int error = 0;
548 
549 	if (uid == (uid_t)VNOVAL)
550 		uid = ip->i_uid;
551 	if (gid == (gid_t)VNOVAL)
552 		gid = ip->i_gid;
553 	/*
554 	 * To modify the ownership of a file, must possess VADMIN
555 	 * for that file.
556 	 */
557 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
558 		return (error);
559 	/*
560 	 * To change the owner of a file, or change the group of a file
561 	 * to a group of which we are not a member, the caller must
562 	 * have privilege.
563 	 */
564 	if (uid != ip->i_uid || (gid != ip->i_gid &&
565 	    !groupmember(gid, cred))) {
566 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
567 		if (error)
568 			return (error);
569 	}
570 	ogid = ip->i_gid;
571 	ouid = ip->i_uid;
572 	ip->i_gid = gid;
573 	ip->i_uid = uid;
574 	ip->i_flag |= IN_CHANGE;
575 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
576 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
577 			ip->i_mode &= ~(ISUID | ISGID);
578 	}
579 	return (0);
580 }
581 
582 /*
583  * Synch an open file.
584  */
585 /* ARGSUSED */
586 static int
587 ext2_fsync(struct vop_fsync_args *ap)
588 {
589 	/*
590 	 * Flush all dirty buffers associated with a vnode.
591 	 */
592 
593 	vop_stdfsync(ap);
594 
595 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
596 }
597 
598 /*
599  * Mknod vnode call
600  */
601 /* ARGSUSED */
602 static int
603 ext2_mknod(struct vop_mknod_args *ap)
604 {
605 	struct vattr *vap = ap->a_vap;
606 	struct vnode **vpp = ap->a_vpp;
607 	struct inode *ip;
608 	ino_t ino;
609 	int error;
610 
611 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
612 	    ap->a_dvp, vpp, ap->a_cnp);
613 	if (error)
614 		return (error);
615 	ip = VTOI(*vpp);
616 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
617 	if (vap->va_rdev != VNOVAL) {
618 		/*
619 		 * Want to be able to use this to make badblock
620 		 * inodes, so don't truncate the dev number.
621 		 */
622 		ip->i_rdev = vap->va_rdev;
623 	}
624 	/*
625 	 * Remove inode, then reload it through VFS_VGET so it is
626 	 * checked to see if it is an alias of an existing entry in
627 	 * the inode cache.	 XXX I don't believe this is necessary now.
628 	 */
629 	(*vpp)->v_type = VNON;
630 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
631 	vgone(*vpp);
632 	vput(*vpp);
633 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
634 	if (error) {
635 		*vpp = NULL;
636 		return (error);
637 	}
638 	return (0);
639 }
640 
641 static int
642 ext2_remove(struct vop_remove_args *ap)
643 {
644 	struct inode *ip;
645 	struct vnode *vp = ap->a_vp;
646 	struct vnode *dvp = ap->a_dvp;
647 	int error;
648 
649 	ip = VTOI(vp);
650 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
651 	    (VTOI(dvp)->i_flags & APPEND)) {
652 		error = EPERM;
653 		goto out;
654 	}
655 	error = ext2_dirremove(dvp, ap->a_cnp);
656 	if (error == 0) {
657 		ip->i_nlink--;
658 		ip->i_flag |= IN_CHANGE;
659 	}
660 out:
661 	return (error);
662 }
663 
664 /*
665  * link vnode call
666  */
667 static int
668 ext2_link(struct vop_link_args *ap)
669 {
670 	struct vnode *vp = ap->a_vp;
671 	struct vnode *tdvp = ap->a_tdvp;
672 	struct componentname *cnp = ap->a_cnp;
673 	struct inode *ip;
674 	int error;
675 
676 #ifdef INVARIANTS
677 	if ((cnp->cn_flags & HASBUF) == 0)
678 		panic("ext2_link: no name");
679 #endif
680 	ip = VTOI(vp);
681 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
682 		error = EMLINK;
683 		goto out;
684 	}
685 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
686 		error = EPERM;
687 		goto out;
688 	}
689 	ip->i_nlink++;
690 	ip->i_flag |= IN_CHANGE;
691 	error = ext2_update(vp, !DOINGASYNC(vp));
692 	if (!error)
693 		error = ext2_direnter(ip, tdvp, cnp);
694 	if (error) {
695 		ip->i_nlink--;
696 		ip->i_flag |= IN_CHANGE;
697 	}
698 out:
699 	return (error);
700 }
701 
702 /*
703  * Rename system call.
704  * 	rename("foo", "bar");
705  * is essentially
706  *	unlink("bar");
707  *	link("foo", "bar");
708  *	unlink("foo");
709  * but ``atomically''.  Can't do full commit without saving state in the
710  * inode on disk which isn't feasible at this time.  Best we can do is
711  * always guarantee the target exists.
712  *
713  * Basic algorithm is:
714  *
715  * 1) Bump link count on source while we're linking it to the
716  *    target.  This also ensure the inode won't be deleted out
717  *    from underneath us while we work (it may be truncated by
718  *    a concurrent `trunc' or `open' for creation).
719  * 2) Link source to destination.  If destination already exists,
720  *    delete it first.
721  * 3) Unlink source reference to inode if still around. If a
722  *    directory was moved and the parent of the destination
723  *    is different from the source, patch the ".." entry in the
724  *    directory.
725  */
726 static int
727 ext2_rename(struct vop_rename_args *ap)
728 {
729 	struct vnode *tvp = ap->a_tvp;
730 	struct vnode *tdvp = ap->a_tdvp;
731 	struct vnode *fvp = ap->a_fvp;
732 	struct vnode *fdvp = ap->a_fdvp;
733 	struct componentname *tcnp = ap->a_tcnp;
734 	struct componentname *fcnp = ap->a_fcnp;
735 	struct inode *ip, *xp, *dp;
736 	struct dirtemplate dirbuf;
737 	int doingdirectory = 0, oldparent = 0, newparent = 0;
738 	int error = 0;
739 	u_char namlen;
740 
741 #ifdef INVARIANTS
742 	if ((tcnp->cn_flags & HASBUF) == 0 ||
743 	    (fcnp->cn_flags & HASBUF) == 0)
744 		panic("ext2_rename: no name");
745 #endif
746 	/*
747 	 * Check for cross-device rename.
748 	 */
749 	if ((fvp->v_mount != tdvp->v_mount) ||
750 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
751 		error = EXDEV;
752 abortit:
753 		if (tdvp == tvp)
754 			vrele(tdvp);
755 		else
756 			vput(tdvp);
757 		if (tvp)
758 			vput(tvp);
759 		vrele(fdvp);
760 		vrele(fvp);
761 		return (error);
762 	}
763 
764 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
765 	    (VTOI(tdvp)->i_flags & APPEND))) {
766 		error = EPERM;
767 		goto abortit;
768 	}
769 
770 	/*
771 	 * Renaming a file to itself has no effect.  The upper layers should
772 	 * not call us in that case.  Temporarily just warn if they do.
773 	 */
774 	if (fvp == tvp) {
775 		printf("ext2_rename: fvp == tvp (can't happen)\n");
776 		error = 0;
777 		goto abortit;
778 	}
779 
780 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
781 		goto abortit;
782 	dp = VTOI(fdvp);
783 	ip = VTOI(fvp);
784 	if (ip->i_nlink >= EXT2_LINK_MAX) {
785 		VOP_UNLOCK(fvp, 0);
786 		error = EMLINK;
787 		goto abortit;
788 	}
789 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
790 	    || (dp->i_flags & APPEND)) {
791 		VOP_UNLOCK(fvp, 0);
792 		error = EPERM;
793 		goto abortit;
794 	}
795 	if ((ip->i_mode & IFMT) == IFDIR) {
796 		/*
797 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
798 		 */
799 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
800 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
801 		    (ip->i_flag & IN_RENAME)) {
802 			VOP_UNLOCK(fvp, 0);
803 			error = EINVAL;
804 			goto abortit;
805 		}
806 		ip->i_flag |= IN_RENAME;
807 		oldparent = dp->i_number;
808 		doingdirectory++;
809 	}
810 	vrele(fdvp);
811 
812 	/*
813 	 * When the target exists, both the directory
814 	 * and target vnodes are returned locked.
815 	 */
816 	dp = VTOI(tdvp);
817 	xp = NULL;
818 	if (tvp)
819 		xp = VTOI(tvp);
820 
821 	/*
822 	 * 1) Bump link count while we're moving stuff
823 	 *    around.  If we crash somewhere before
824 	 *    completing our work, the link count
825 	 *    may be wrong, but correctable.
826 	 */
827 	ip->i_nlink++;
828 	ip->i_flag |= IN_CHANGE;
829 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
830 		VOP_UNLOCK(fvp, 0);
831 		goto bad;
832 	}
833 
834 	/*
835 	 * If ".." must be changed (ie the directory gets a new
836 	 * parent) then the source directory must not be in the
837 	 * directory hierarchy above the target, as this would
838 	 * orphan everything below the source directory. Also
839 	 * the user must have write permission in the source so
840 	 * as to be able to change "..". We must repeat the call
841 	 * to namei, as the parent directory is unlocked by the
842 	 * call to checkpath().
843 	 */
844 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
845 	VOP_UNLOCK(fvp, 0);
846 	if (oldparent != dp->i_number)
847 		newparent = dp->i_number;
848 	if (doingdirectory && newparent) {
849 		if (error)	/* write access check above */
850 			goto bad;
851 		if (xp != NULL)
852 			vput(tvp);
853 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
854 		if (error)
855 			goto out;
856 		VREF(tdvp);
857 		error = relookup(tdvp, &tvp, tcnp);
858 		if (error)
859 			goto out;
860 		vrele(tdvp);
861 		dp = VTOI(tdvp);
862 		xp = NULL;
863 		if (tvp)
864 			xp = VTOI(tvp);
865 	}
866 	/*
867 	 * 2) If target doesn't exist, link the target
868 	 *    to the source and unlink the source.
869 	 *    Otherwise, rewrite the target directory
870 	 *    entry to reference the source inode and
871 	 *    expunge the original entry's existence.
872 	 */
873 	if (xp == NULL) {
874 		if (dp->i_devvp != ip->i_devvp)
875 			panic("ext2_rename: EXDEV");
876 		/*
877 		 * Account for ".." in new directory.
878 		 * When source and destination have the same
879 		 * parent we don't fool with the link count.
880 		 */
881 		if (doingdirectory && newparent) {
882 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
883 				error = EMLINK;
884 				goto bad;
885 			}
886 			dp->i_nlink++;
887 			dp->i_flag |= IN_CHANGE;
888 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
889 			if (error)
890 				goto bad;
891 		}
892 		error = ext2_direnter(ip, tdvp, tcnp);
893 		if (error) {
894 			if (doingdirectory && newparent) {
895 				dp->i_nlink--;
896 				dp->i_flag |= IN_CHANGE;
897 				(void)ext2_update(tdvp, 1);
898 			}
899 			goto bad;
900 		}
901 		vput(tdvp);
902 	} else {
903 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
904 			panic("ext2_rename: EXDEV");
905 		/*
906 		 * Short circuit rename(foo, foo).
907 		 */
908 		if (xp->i_number == ip->i_number)
909 			panic("ext2_rename: same file");
910 		/*
911 		 * If the parent directory is "sticky", then the user must
912 		 * own the parent directory, or the destination of the rename,
913 		 * otherwise the destination may not be changed (except by
914 		 * root). This implements append-only directories.
915 		 */
916 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
917 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
918 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
919 			error = EPERM;
920 			goto bad;
921 		}
922 		/*
923 		 * Target must be empty if a directory and have no links
924 		 * to it. Also, ensure source and target are compatible
925 		 * (both directories, or both not directories).
926 		 */
927 		if ((xp->i_mode & IFMT) == IFDIR) {
928 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
929 			    xp->i_nlink > 2) {
930 				error = ENOTEMPTY;
931 				goto bad;
932 			}
933 			if (!doingdirectory) {
934 				error = ENOTDIR;
935 				goto bad;
936 			}
937 			cache_purge(tdvp);
938 		} else if (doingdirectory) {
939 			error = EISDIR;
940 			goto bad;
941 		}
942 		error = ext2_dirrewrite(dp, ip, tcnp);
943 		if (error)
944 			goto bad;
945 		/*
946 		 * If the target directory is in the same
947 		 * directory as the source directory,
948 		 * decrement the link count on the parent
949 		 * of the target directory.
950 		 */
951 		if (doingdirectory && !newparent) {
952 			dp->i_nlink--;
953 			dp->i_flag |= IN_CHANGE;
954 		}
955 		vput(tdvp);
956 		/*
957 		 * Adjust the link count of the target to
958 		 * reflect the dirrewrite above.  If this is
959 		 * a directory it is empty and there are
960 		 * no links to it, so we can squash the inode and
961 		 * any space associated with it.  We disallowed
962 		 * renaming over top of a directory with links to
963 		 * it above, as the remaining link would point to
964 		 * a directory without "." or ".." entries.
965 		 */
966 		xp->i_nlink--;
967 		if (doingdirectory) {
968 			if (--xp->i_nlink != 0)
969 				panic("ext2_rename: linked directory");
970 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
971 			    tcnp->cn_cred, tcnp->cn_thread);
972 		}
973 		xp->i_flag |= IN_CHANGE;
974 		vput(tvp);
975 		xp = NULL;
976 	}
977 
978 	/*
979 	 * 3) Unlink the source.
980 	 */
981 	fcnp->cn_flags &= ~MODMASK;
982 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
983 	VREF(fdvp);
984 	error = relookup(fdvp, &fvp, fcnp);
985 	if (error == 0)
986 		vrele(fdvp);
987 	if (fvp != NULL) {
988 		xp = VTOI(fvp);
989 		dp = VTOI(fdvp);
990 	} else {
991 		/*
992 		 * From name has disappeared.  IN_RENAME is not sufficient
993 		 * to protect against directory races due to timing windows,
994 		 * so we can't panic here.
995 		 */
996 		vrele(ap->a_fvp);
997 		return (0);
998 	}
999 	/*
1000 	 * Ensure that the directory entry still exists and has not
1001 	 * changed while the new name has been entered. If the source is
1002 	 * a file then the entry may have been unlinked or renamed. In
1003 	 * either case there is no further work to be done. If the source
1004 	 * is a directory then it cannot have been rmdir'ed; its link
1005 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1006 	 * The IN_RENAME flag ensures that it cannot be moved by another
1007 	 * rename.
1008 	 */
1009 	if (xp != ip) {
1010 		/*
1011 		 * From name resolves to a different inode.  IN_RENAME is
1012 		 * not sufficient protection against timing window races
1013 		 * so we can't panic here.
1014 		 */
1015 	} else {
1016 		/*
1017 		 * If the source is a directory with a
1018 		 * new parent, the link count of the old
1019 		 * parent directory must be decremented
1020 		 * and ".." set to point to the new parent.
1021 		 */
1022 		if (doingdirectory && newparent) {
1023 			dp->i_nlink--;
1024 			dp->i_flag |= IN_CHANGE;
1025 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1026 			    sizeof(struct dirtemplate), (off_t)0,
1027 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1028 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1029 			if (error == 0) {
1030 				/* Like ufs little-endian: */
1031 				namlen = dirbuf.dotdot_type;
1032 				if (namlen != 2 ||
1033 				    dirbuf.dotdot_name[0] != '.' ||
1034 				    dirbuf.dotdot_name[1] != '.') {
1035 					ext2_dirbad(xp, (doff_t)12,
1036 					    "rename: mangled dir");
1037 				} else {
1038 					dirbuf.dotdot_ino = newparent;
1039 					(void)vn_rdwr(UIO_WRITE, fvp,
1040 					    (caddr_t)&dirbuf,
1041 					    sizeof(struct dirtemplate),
1042 					    (off_t)0, UIO_SYSSPACE,
1043 					    IO_NODELOCKED | IO_SYNC |
1044 					    IO_NOMACCHECK, tcnp->cn_cred,
1045 					    NOCRED, NULL, NULL);
1046 					cache_purge(fdvp);
1047 				}
1048 			}
1049 		}
1050 		error = ext2_dirremove(fdvp, fcnp);
1051 		if (!error) {
1052 			xp->i_nlink--;
1053 			xp->i_flag |= IN_CHANGE;
1054 		}
1055 		xp->i_flag &= ~IN_RENAME;
1056 	}
1057 	if (dp)
1058 		vput(fdvp);
1059 	if (xp)
1060 		vput(fvp);
1061 	vrele(ap->a_fvp);
1062 	return (error);
1063 
1064 bad:
1065 	if (xp)
1066 		vput(ITOV(xp));
1067 	vput(ITOV(dp));
1068 out:
1069 	if (doingdirectory)
1070 		ip->i_flag &= ~IN_RENAME;
1071 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1072 		ip->i_nlink--;
1073 		ip->i_flag |= IN_CHANGE;
1074 		ip->i_flag &= ~IN_RENAME;
1075 		vput(fvp);
1076 	} else
1077 		vrele(fvp);
1078 	return (error);
1079 }
1080 
1081 /*
1082  * Mkdir system call
1083  */
1084 static int
1085 ext2_mkdir(struct vop_mkdir_args *ap)
1086 {
1087 	struct vnode *dvp = ap->a_dvp;
1088 	struct vattr *vap = ap->a_vap;
1089 	struct componentname *cnp = ap->a_cnp;
1090 	struct inode *ip, *dp;
1091 	struct vnode *tvp;
1092 	struct dirtemplate dirtemplate, *dtp;
1093 	int error, dmode;
1094 
1095 #ifdef INVARIANTS
1096 	if ((cnp->cn_flags & HASBUF) == 0)
1097 		panic("ext2_mkdir: no name");
1098 #endif
1099 	dp = VTOI(dvp);
1100 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1101 		error = EMLINK;
1102 		goto out;
1103 	}
1104 	dmode = vap->va_mode & 0777;
1105 	dmode |= IFDIR;
1106 	/*
1107 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1108 	 * but not have it entered in the parent directory. The entry is
1109 	 * made later after writing "." and ".." entries.
1110 	 */
1111 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1112 	if (error)
1113 		goto out;
1114 	ip = VTOI(tvp);
1115 	ip->i_gid = dp->i_gid;
1116 #ifdef SUIDDIR
1117 	{
1118 		/*
1119 		 * if we are hacking owners here, (only do this where told to)
1120 		 * and we are not giving it TOO root, (would subvert quotas)
1121 		 * then go ahead and give it to the other user.
1122 		 * The new directory also inherits the SUID bit.
1123 		 * If user's UID and dir UID are the same,
1124 		 * 'give it away' so that the SUID is still forced on.
1125 		 */
1126 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1127 		    (dp->i_mode & ISUID) && dp->i_uid) {
1128 			dmode |= ISUID;
1129 			ip->i_uid = dp->i_uid;
1130 		} else {
1131 			ip->i_uid = cnp->cn_cred->cr_uid;
1132 		}
1133 	}
1134 #else
1135 	ip->i_uid = cnp->cn_cred->cr_uid;
1136 #endif
1137 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1138 	ip->i_mode = dmode;
1139 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1140 	ip->i_nlink = 2;
1141 	if (cnp->cn_flags & ISWHITEOUT)
1142 		ip->i_flags |= UF_OPAQUE;
1143 	error = ext2_update(tvp, 1);
1144 
1145 	/*
1146 	 * Bump link count in parent directory
1147 	 * to reflect work done below.  Should
1148 	 * be done before reference is created
1149 	 * so reparation is possible if we crash.
1150 	 */
1151 	dp->i_nlink++;
1152 	dp->i_flag |= IN_CHANGE;
1153 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1154 	if (error)
1155 		goto bad;
1156 
1157 	/* Initialize directory with "." and ".." from static template. */
1158 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1159 	    EXT2F_INCOMPAT_FTYPE))
1160 		dtp = &mastertemplate;
1161 	else
1162 		dtp = &omastertemplate;
1163 	dirtemplate = *dtp;
1164 	dirtemplate.dot_ino = ip->i_number;
1165 	dirtemplate.dotdot_ino = dp->i_number;
1166 	/*
1167 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1168 	 * just redefine it - for this function only
1169 	 */
1170 #undef  DIRBLKSIZ
1171 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1172 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1173 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1174 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1175 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1176 	    NULL, NULL);
1177 	if (error) {
1178 		dp->i_nlink--;
1179 		dp->i_flag |= IN_CHANGE;
1180 		goto bad;
1181 	}
1182 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1183 		/* XXX should grow with balloc() */
1184 		panic("ext2_mkdir: blksize");
1185 	else {
1186 		ip->i_size = DIRBLKSIZ;
1187 		ip->i_flag |= IN_CHANGE;
1188 	}
1189 
1190 	/* Directory set up, now install its entry in the parent directory. */
1191 	error = ext2_direnter(ip, dvp, cnp);
1192 	if (error) {
1193 		dp->i_nlink--;
1194 		dp->i_flag |= IN_CHANGE;
1195 	}
1196 bad:
1197 	/*
1198 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1199 	 * for us because we set the link count to 0.
1200 	 */
1201 	if (error) {
1202 		ip->i_nlink = 0;
1203 		ip->i_flag |= IN_CHANGE;
1204 		vput(tvp);
1205 	} else
1206 		*ap->a_vpp = tvp;
1207 out:
1208 	return (error);
1209 #undef  DIRBLKSIZ
1210 #define DIRBLKSIZ  DEV_BSIZE
1211 }
1212 
1213 /*
1214  * Rmdir system call.
1215  */
1216 static int
1217 ext2_rmdir(struct vop_rmdir_args *ap)
1218 {
1219 	struct vnode *vp = ap->a_vp;
1220 	struct vnode *dvp = ap->a_dvp;
1221 	struct componentname *cnp = ap->a_cnp;
1222 	struct inode *ip, *dp;
1223 	int error;
1224 
1225 	ip = VTOI(vp);
1226 	dp = VTOI(dvp);
1227 
1228 	/*
1229 	 * Verify the directory is empty (and valid).
1230 	 * (Rmdir ".." won't be valid since
1231 	 *  ".." will contain a reference to
1232 	 *  the current directory and thus be
1233 	 *  non-empty.)
1234 	 */
1235 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1236 		error = ENOTEMPTY;
1237 		goto out;
1238 	}
1239 	if ((dp->i_flags & APPEND)
1240 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1241 		error = EPERM;
1242 		goto out;
1243 	}
1244 	/*
1245 	 * Delete reference to directory before purging
1246 	 * inode.  If we crash in between, the directory
1247 	 * will be reattached to lost+found,
1248 	 */
1249 	error = ext2_dirremove(dvp, cnp);
1250 	if (error)
1251 		goto out;
1252 	dp->i_nlink--;
1253 	dp->i_flag |= IN_CHANGE;
1254 	cache_purge(dvp);
1255 	VOP_UNLOCK(dvp, 0);
1256 	/*
1257 	 * Truncate inode.  The only stuff left
1258 	 * in the directory is "." and "..".  The
1259 	 * "." reference is inconsequential since
1260 	 * we're quashing it.  The ".." reference
1261 	 * has already been adjusted above.  We've
1262 	 * removed the "." reference and the reference
1263 	 * in the parent directory, but there may be
1264 	 * other hard links so decrement by 2 and
1265 	 * worry about them later.
1266 	 */
1267 	ip->i_nlink -= 2;
1268 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1269 	    cnp->cn_thread);
1270 	cache_purge(ITOV(ip));
1271 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1272 		VOP_UNLOCK(vp, 0);
1273 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1274 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1275 	}
1276 out:
1277 	return (error);
1278 }
1279 
1280 /*
1281  * symlink -- make a symbolic link
1282  */
1283 static int
1284 ext2_symlink(struct vop_symlink_args *ap)
1285 {
1286 	struct vnode *vp, **vpp = ap->a_vpp;
1287 	struct inode *ip;
1288 	int len, error;
1289 
1290 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1291 	    vpp, ap->a_cnp);
1292 	if (error)
1293 		return (error);
1294 	vp = *vpp;
1295 	len = strlen(ap->a_target);
1296 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1297 		ip = VTOI(vp);
1298 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1299 		ip->i_size = len;
1300 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1301 	} else
1302 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1303 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1304 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1305 	if (error)
1306 		vput(vp);
1307 	return (error);
1308 }
1309 
1310 /*
1311  * Return target name of a symbolic link
1312  */
1313 static int
1314 ext2_readlink(struct vop_readlink_args *ap)
1315 {
1316 	struct vnode *vp = ap->a_vp;
1317 	struct inode *ip = VTOI(vp);
1318 	int isize;
1319 
1320 	isize = ip->i_size;
1321 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1322 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1323 		return (0);
1324 	}
1325 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1326 }
1327 
1328 /*
1329  * Calculate the logical to physical mapping if not done already,
1330  * then call the device strategy routine.
1331  *
1332  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1333  * deadlock on memory.  See ext2_bmap() for details.
1334  */
1335 static int
1336 ext2_strategy(struct vop_strategy_args *ap)
1337 {
1338 	struct buf *bp = ap->a_bp;
1339 	struct vnode *vp = ap->a_vp;
1340 	struct bufobj *bo;
1341 	daddr_t blkno;
1342 	int error;
1343 
1344 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1345 		panic("ext2_strategy: spec");
1346 	if (bp->b_blkno == bp->b_lblkno) {
1347 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1348 		bp->b_blkno = blkno;
1349 		if (error) {
1350 			bp->b_error = error;
1351 			bp->b_ioflags |= BIO_ERROR;
1352 			bufdone(bp);
1353 			return (0);
1354 		}
1355 		if ((long)bp->b_blkno == -1)
1356 			vfs_bio_clrbuf(bp);
1357 	}
1358 	if ((long)bp->b_blkno == -1) {
1359 		bufdone(bp);
1360 		return (0);
1361 	}
1362 	bp->b_iooffset = dbtob(bp->b_blkno);
1363 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1364 	BO_STRATEGY(bo, bp);
1365 	return (0);
1366 }
1367 
1368 /*
1369  * Print out the contents of an inode.
1370  */
1371 static int
1372 ext2_print(struct vop_print_args *ap)
1373 {
1374 	struct vnode *vp = ap->a_vp;
1375 	struct inode *ip = VTOI(vp);
1376 
1377 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1378 	if (vp->v_type == VFIFO)
1379 		fifo_printinfo(vp);
1380 	printf("\n");
1381 	return (0);
1382 }
1383 
1384 /*
1385  * Close wrapper for fifos.
1386  *
1387  * Update the times on the inode then do device close.
1388  */
1389 static int
1390 ext2fifo_close(struct vop_close_args *ap)
1391 {
1392 	struct vnode *vp = ap->a_vp;
1393 
1394 	VI_LOCK(vp);
1395 	if (vp->v_usecount > 1)
1396 		ext2_itimes_locked(vp);
1397 	VI_UNLOCK(vp);
1398 	return (fifo_specops.vop_close(ap));
1399 }
1400 
1401 /*
1402  * Kqfilter wrapper for fifos.
1403  *
1404  * Fall through to ext2 kqfilter routines if needed
1405  */
1406 static int
1407 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1408 {
1409 	int error;
1410 
1411 	error = fifo_specops.vop_kqfilter(ap);
1412 	if (error)
1413 		error = vfs_kqfilter(ap);
1414 	return (error);
1415 }
1416 
1417 /*
1418  * Return POSIX pathconf information applicable to ext2 filesystems.
1419  */
1420 static int
1421 ext2_pathconf(struct vop_pathconf_args *ap)
1422 {
1423 	int error = 0;
1424 
1425 	switch (ap->a_name) {
1426 	case _PC_LINK_MAX:
1427 		*ap->a_retval = EXT2_LINK_MAX;
1428 		break;
1429 	case _PC_NAME_MAX:
1430 		*ap->a_retval = NAME_MAX;
1431 		break;
1432 	case _PC_PATH_MAX:
1433 		*ap->a_retval = PATH_MAX;
1434 		break;
1435 	case _PC_PIPE_BUF:
1436 		*ap->a_retval = PIPE_BUF;
1437 		break;
1438 	case _PC_CHOWN_RESTRICTED:
1439 		*ap->a_retval = 1;
1440 		break;
1441 	case _PC_NO_TRUNC:
1442 		*ap->a_retval = 1;
1443 		break;
1444 	case _PC_MIN_HOLE_SIZE:
1445 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1446 		break;
1447 	case _PC_ASYNC_IO:
1448 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1449 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1450 		error = EINVAL;
1451 		break;
1452 	case _PC_PRIO_IO:
1453 		*ap->a_retval = 0;
1454 		break;
1455 	case _PC_SYNC_IO:
1456 		*ap->a_retval = 0;
1457 		break;
1458 	case _PC_ALLOC_SIZE_MIN:
1459 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1460 		break;
1461 	case _PC_FILESIZEBITS:
1462 		*ap->a_retval = 64;
1463 		break;
1464 	case _PC_REC_INCR_XFER_SIZE:
1465 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1466 		break;
1467 	case _PC_REC_MAX_XFER_SIZE:
1468 		*ap->a_retval = -1;	/* means ``unlimited'' */
1469 		break;
1470 	case _PC_REC_MIN_XFER_SIZE:
1471 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1472 		break;
1473 	case _PC_REC_XFER_ALIGN:
1474 		*ap->a_retval = PAGE_SIZE;
1475 		break;
1476 	case _PC_SYMLINK_MAX:
1477 		*ap->a_retval = MAXPATHLEN;
1478 		break;
1479 
1480 	default:
1481 		error = EINVAL;
1482 		break;
1483 	}
1484 	return (error);
1485 }
1486 
1487 /*
1488  * Vnode operation to retrieve a named extended attribute.
1489  */
1490 static int
1491 ext2_getextattr(struct vop_getextattr_args *ap)
1492 {
1493 	struct inode *ip;
1494 	struct m_ext2fs *fs;
1495 	int error;
1496 
1497 	ip = VTOI(ap->a_vp);
1498 	fs = ip->i_e2fs;
1499 
1500 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1501 		return (EOPNOTSUPP);
1502 
1503 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1504 		return (EOPNOTSUPP);
1505 
1506 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1507 	    ap->a_cred, ap->a_td, VREAD);
1508 	if (error)
1509 		return (error);
1510 
1511 	if (ap->a_size != NULL)
1512 		*ap->a_size = 0;
1513 
1514 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1515 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1516 		    ap->a_name, ap->a_uio, ap->a_size);
1517 		if (error)
1518 			return (error);
1519 	}
1520 
1521 	if (ip->i_facl)
1522 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1523 		    ap->a_name, ap->a_uio, ap->a_size);
1524 
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Vnode operation to retrieve extended attributes on a vnode.
1530  */
1531 static int
1532 ext2_listextattr(struct vop_listextattr_args *ap)
1533 {
1534 	struct inode *ip;
1535 	struct m_ext2fs *fs;
1536 	int error;
1537 
1538 	ip = VTOI(ap->a_vp);
1539 	fs = ip->i_e2fs;
1540 
1541 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1542 		return (EOPNOTSUPP);
1543 
1544 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1545 		return (EOPNOTSUPP);
1546 
1547 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1548 	    ap->a_cred, ap->a_td, VREAD);
1549 	if (error)
1550 		return (error);
1551 
1552 	if (ap->a_size != NULL)
1553 		*ap->a_size = 0;
1554 
1555 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1556 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1557 		    ap->a_uio, ap->a_size);
1558 		if(error)
1559 			return (error);
1560 	}
1561 
1562 	if(ip->i_facl)
1563 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1564 		    ap->a_uio, ap->a_size);
1565 
1566 	return (error);
1567 }
1568 
1569 /*
1570  * Vnode pointer to File handle
1571  */
1572 /* ARGSUSED */
1573 static int
1574 ext2_vptofh(struct vop_vptofh_args *ap)
1575 {
1576 	struct inode *ip;
1577 	struct ufid *ufhp;
1578 
1579 	ip = VTOI(ap->a_vp);
1580 	ufhp = (struct ufid *)ap->a_fhp;
1581 	ufhp->ufid_len = sizeof(struct ufid);
1582 	ufhp->ufid_ino = ip->i_number;
1583 	ufhp->ufid_gen = ip->i_gen;
1584 	return (0);
1585 }
1586 
1587 /*
1588  * Initialize the vnode associated with a new inode, handle aliased
1589  * vnodes.
1590  */
1591 int
1592 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1593 {
1594 	struct inode *ip;
1595 	struct vnode *vp;
1596 
1597 	vp = *vpp;
1598 	ip = VTOI(vp);
1599 	vp->v_type = IFTOVT(ip->i_mode);
1600 	if (vp->v_type == VFIFO)
1601 		vp->v_op = fifoops;
1602 
1603 	if (ip->i_number == EXT2_ROOTINO)
1604 		vp->v_vflag |= VV_ROOT;
1605 	ip->i_modrev = init_va_filerev();
1606 	*vpp = vp;
1607 	return (0);
1608 }
1609 
1610 /*
1611  * Allocate a new inode.
1612  */
1613 static int
1614 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1615     struct componentname *cnp)
1616 {
1617 	struct inode *ip, *pdir;
1618 	struct vnode *tvp;
1619 	int error;
1620 
1621 	pdir = VTOI(dvp);
1622 #ifdef INVARIANTS
1623 	if ((cnp->cn_flags & HASBUF) == 0)
1624 		panic("ext2_makeinode: no name");
1625 #endif
1626 	*vpp = NULL;
1627 	if ((mode & IFMT) == 0)
1628 		mode |= IFREG;
1629 
1630 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1631 	if (error) {
1632 		return (error);
1633 	}
1634 	ip = VTOI(tvp);
1635 	ip->i_gid = pdir->i_gid;
1636 #ifdef SUIDDIR
1637 	{
1638 		/*
1639 		 * if we are
1640 		 * not the owner of the directory,
1641 		 * and we are hacking owners here, (only do this where told to)
1642 		 * and we are not giving it TOO root, (would subvert quotas)
1643 		 * then go ahead and give it to the other user.
1644 		 * Note that this drops off the execute bits for security.
1645 		 */
1646 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1647 		    (pdir->i_mode & ISUID) &&
1648 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1649 			ip->i_uid = pdir->i_uid;
1650 			mode &= ~07111;
1651 		} else {
1652 			ip->i_uid = cnp->cn_cred->cr_uid;
1653 		}
1654 	}
1655 #else
1656 	ip->i_uid = cnp->cn_cred->cr_uid;
1657 #endif
1658 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1659 	ip->i_mode = mode;
1660 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1661 	ip->i_nlink = 1;
1662 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1663 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1664 			ip->i_mode &= ~ISGID;
1665 	}
1666 
1667 	if (cnp->cn_flags & ISWHITEOUT)
1668 		ip->i_flags |= UF_OPAQUE;
1669 
1670 	/*
1671 	 * Make sure inode goes to disk before directory entry.
1672 	 */
1673 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1674 	if (error)
1675 		goto bad;
1676 	error = ext2_direnter(ip, dvp, cnp);
1677 	if (error)
1678 		goto bad;
1679 
1680 	*vpp = tvp;
1681 	return (0);
1682 
1683 bad:
1684 	/*
1685 	 * Write error occurred trying to update the inode
1686 	 * or the directory so must deallocate the inode.
1687 	 */
1688 	ip->i_nlink = 0;
1689 	ip->i_flag |= IN_CHANGE;
1690 	vput(tvp);
1691 	return (error);
1692 }
1693 
1694 /*
1695  * Vnode op for reading.
1696  */
1697 static int
1698 ext2_read(struct vop_read_args *ap)
1699 {
1700 	struct vnode *vp;
1701 	struct inode *ip;
1702 	int error;
1703 
1704 	vp = ap->a_vp;
1705 	ip = VTOI(vp);
1706 
1707 	/* EXT4_EXT_LOCK(ip); */
1708 	if (ip->i_flag & IN_E4EXTENTS)
1709 		error = ext4_ext_read(ap);
1710 	else
1711 		error = ext2_ind_read(ap);
1712 	/* EXT4_EXT_UNLOCK(ip); */
1713 	return (error);
1714 }
1715 
1716 /*
1717  * Vnode op for reading.
1718  */
1719 static int
1720 ext2_ind_read(struct vop_read_args *ap)
1721 {
1722 	struct vnode *vp;
1723 	struct inode *ip;
1724 	struct uio *uio;
1725 	struct m_ext2fs *fs;
1726 	struct buf *bp;
1727 	daddr_t lbn, nextlbn;
1728 	off_t bytesinfile;
1729 	long size, xfersize, blkoffset;
1730 	int error, orig_resid, seqcount;
1731 	int ioflag;
1732 
1733 	vp = ap->a_vp;
1734 	uio = ap->a_uio;
1735 	ioflag = ap->a_ioflag;
1736 
1737 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1738 	ip = VTOI(vp);
1739 
1740 #ifdef INVARIANTS
1741 	if (uio->uio_rw != UIO_READ)
1742 		panic("%s: mode", "ext2_read");
1743 
1744 	if (vp->v_type == VLNK) {
1745 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1746 			panic("%s: short symlink", "ext2_read");
1747 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1748 		panic("%s: type %d", "ext2_read", vp->v_type);
1749 #endif
1750 	orig_resid = uio->uio_resid;
1751 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1752 	if (orig_resid == 0)
1753 		return (0);
1754 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1755 	fs = ip->i_e2fs;
1756 	if (uio->uio_offset < ip->i_size &&
1757 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1758 		return (EOVERFLOW);
1759 
1760 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1761 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1762 			break;
1763 		lbn = lblkno(fs, uio->uio_offset);
1764 		nextlbn = lbn + 1;
1765 		size = blksize(fs, ip, lbn);
1766 		blkoffset = blkoff(fs, uio->uio_offset);
1767 
1768 		xfersize = fs->e2fs_fsize - blkoffset;
1769 		if (uio->uio_resid < xfersize)
1770 			xfersize = uio->uio_resid;
1771 		if (bytesinfile < xfersize)
1772 			xfersize = bytesinfile;
1773 
1774 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1775 			error = bread(vp, lbn, size, NOCRED, &bp);
1776 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1777 			error = cluster_read(vp, ip->i_size, lbn, size,
1778 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1779 			    0, &bp);
1780 		} else if (seqcount > 1) {
1781 			u_int nextsize = blksize(fs, ip, nextlbn);
1782 
1783 			error = breadn(vp, lbn,
1784 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1785 		} else
1786 			error = bread(vp, lbn, size, NOCRED, &bp);
1787 		if (error) {
1788 			brelse(bp);
1789 			bp = NULL;
1790 			break;
1791 		}
1792 
1793 		/*
1794 		 * We should only get non-zero b_resid when an I/O error
1795 		 * has occurred, which should cause us to break above.
1796 		 * However, if the short read did not cause an error,
1797 		 * then we want to ensure that we do not uiomove bad
1798 		 * or uninitialized data.
1799 		 */
1800 		size -= bp->b_resid;
1801 		if (size < xfersize) {
1802 			if (size == 0)
1803 				break;
1804 			xfersize = size;
1805 		}
1806 		error = uiomove((char *)bp->b_data + blkoffset,
1807 		    (int)xfersize, uio);
1808 		if (error)
1809 			break;
1810 		vfs_bio_brelse(bp, ioflag);
1811 	}
1812 
1813 	/*
1814 	 * This can only happen in the case of an error because the loop
1815 	 * above resets bp to NULL on each iteration and on normal
1816 	 * completion has not set a new value into it. so it must have come
1817 	 * from a 'break' statement
1818 	 */
1819 	if (bp != NULL)
1820 		vfs_bio_brelse(bp, ioflag);
1821 
1822 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1823 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1824 		ip->i_flag |= IN_ACCESS;
1825 	return (error);
1826 }
1827 
1828 static int
1829 ext2_ioctl(struct vop_ioctl_args *ap)
1830 {
1831 
1832 	switch (ap->a_command) {
1833 	case FIOSEEKDATA:
1834 	case FIOSEEKHOLE:
1835 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1836 		    (off_t *)ap->a_data, ap->a_cred));
1837 	default:
1838 		return (ENOTTY);
1839 	}
1840 }
1841 
1842 /*
1843  * this function handles ext4 extents block mapping
1844  */
1845 static int
1846 ext4_ext_read(struct vop_read_args *ap)
1847 {
1848 	static unsigned char zeroes[EXT2_MAX_BLOCK_SIZE];
1849 	struct vnode *vp;
1850 	struct inode *ip;
1851 	struct uio *uio;
1852 	struct m_ext2fs *fs;
1853 	struct buf *bp;
1854 	struct ext4_extent nex, *ep;
1855 	struct ext4_extent_path path;
1856 	daddr_t lbn, newblk;
1857 	off_t bytesinfile;
1858 	int cache_type;
1859 	ssize_t orig_resid;
1860 	int error;
1861 	long size, xfersize, blkoffset;
1862 
1863 	vp = ap->a_vp;
1864 	ip = VTOI(vp);
1865 	uio = ap->a_uio;
1866 	memset(&path, 0, sizeof(path));
1867 
1868 	orig_resid = uio->uio_resid;
1869 	KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1870 	if (orig_resid == 0)
1871 		return (0);
1872 	KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1873 	fs = ip->i_e2fs;
1874 	if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1875 		return (EOVERFLOW);
1876 
1877 	while (uio->uio_resid > 0) {
1878 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1879 			break;
1880 		lbn = lblkno(fs, uio->uio_offset);
1881 		size = blksize(fs, ip, lbn);
1882 		blkoffset = blkoff(fs, uio->uio_offset);
1883 
1884 		xfersize = fs->e2fs_fsize - blkoffset;
1885 		xfersize = MIN(xfersize, uio->uio_resid);
1886 		xfersize = MIN(xfersize, bytesinfile);
1887 
1888 		/* get block from ext4 extent cache */
1889 		cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1890 		switch (cache_type) {
1891 		case EXT4_EXT_CACHE_NO:
1892 			ext4_ext_find_extent(fs, ip, lbn, &path);
1893 			if (path.ep_is_sparse)
1894 				ep = &path.ep_sparse_ext;
1895 			else
1896 				ep = path.ep_ext;
1897 			if (ep == NULL)
1898 				return (EIO);
1899 
1900 			ext4_ext_put_cache(ip, ep,
1901 			    path.ep_is_sparse ? EXT4_EXT_CACHE_GAP : EXT4_EXT_CACHE_IN);
1902 
1903 			newblk = lbn - ep->e_blk + (ep->e_start_lo |
1904 			    (daddr_t)ep->e_start_hi << 32);
1905 
1906 			if (path.ep_bp != NULL) {
1907 				brelse(path.ep_bp);
1908 				path.ep_bp = NULL;
1909 			}
1910 			break;
1911 
1912 		case EXT4_EXT_CACHE_GAP:
1913 			/* block has not been allocated yet */
1914 			break;
1915 
1916 		case EXT4_EXT_CACHE_IN:
1917 			newblk = lbn - nex.e_blk + (nex.e_start_lo |
1918 			    (daddr_t)nex.e_start_hi << 32);
1919 			break;
1920 
1921 		default:
1922 			panic("%s: invalid cache type", __func__);
1923 		}
1924 
1925 		if (cache_type == EXT4_EXT_CACHE_GAP ||
1926 		    (cache_type == EXT4_EXT_CACHE_NO && path.ep_is_sparse)) {
1927 			if (xfersize > sizeof(zeroes))
1928 				xfersize = sizeof(zeroes);
1929 			error = uiomove(zeroes, xfersize, uio);
1930 			if (error)
1931 				return (error);
1932 		} else {
1933 			error = bread(ip->i_devvp, fsbtodb(fs, newblk), size,
1934 			    NOCRED, &bp);
1935 			if (error) {
1936 				brelse(bp);
1937 				return (error);
1938 			}
1939 
1940 			size -= bp->b_resid;
1941 			if (size < xfersize) {
1942 				if (size == 0) {
1943 					bqrelse(bp);
1944 					break;
1945 				}
1946 				xfersize = size;
1947 			}
1948 			error = uiomove(bp->b_data + blkoffset, xfersize, uio);
1949 			bqrelse(bp);
1950 			if (error)
1951 				return (error);
1952 		}
1953 	}
1954 
1955 	return (0);
1956 }
1957 
1958 /*
1959  * Vnode op for writing.
1960  */
1961 static int
1962 ext2_write(struct vop_write_args *ap)
1963 {
1964 	struct vnode *vp;
1965 	struct uio *uio;
1966 	struct inode *ip;
1967 	struct m_ext2fs *fs;
1968 	struct buf *bp;
1969 	daddr_t lbn;
1970 	off_t osize;
1971 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1972 
1973 	ioflag = ap->a_ioflag;
1974 	uio = ap->a_uio;
1975 	vp = ap->a_vp;
1976 
1977 	seqcount = ioflag >> IO_SEQSHIFT;
1978 	ip = VTOI(vp);
1979 
1980 #ifdef INVARIANTS
1981 	if (uio->uio_rw != UIO_WRITE)
1982 		panic("%s: mode", "ext2_write");
1983 #endif
1984 
1985 	switch (vp->v_type) {
1986 	case VREG:
1987 		if (ioflag & IO_APPEND)
1988 			uio->uio_offset = ip->i_size;
1989 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1990 			return (EPERM);
1991 		/* FALLTHROUGH */
1992 	case VLNK:
1993 		break;
1994 	case VDIR:
1995 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
1996 		if ((ioflag & IO_SYNC) == 0)
1997 			panic("ext2_write: nonsync dir write");
1998 		break;
1999 	default:
2000 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2001 		    vp->v_type, (intmax_t)uio->uio_offset,
2002 		    (intmax_t)uio->uio_resid);
2003 	}
2004 
2005 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2006 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2007 	fs = ip->i_e2fs;
2008 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2009 		return (EFBIG);
2010 	/*
2011 	 * Maybe this should be above the vnode op call, but so long as
2012 	 * file servers have no limits, I don't think it matters.
2013 	 */
2014 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2015 		return (EFBIG);
2016 
2017 	resid = uio->uio_resid;
2018 	osize = ip->i_size;
2019 	if (seqcount > BA_SEQMAX)
2020 		flags = BA_SEQMAX << BA_SEQSHIFT;
2021 	else
2022 		flags = seqcount << BA_SEQSHIFT;
2023 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2024 		flags |= IO_SYNC;
2025 
2026 	for (error = 0; uio->uio_resid > 0;) {
2027 		lbn = lblkno(fs, uio->uio_offset);
2028 		blkoffset = blkoff(fs, uio->uio_offset);
2029 		xfersize = fs->e2fs_fsize - blkoffset;
2030 		if (uio->uio_resid < xfersize)
2031 			xfersize = uio->uio_resid;
2032 		if (uio->uio_offset + xfersize > ip->i_size)
2033 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2034 
2035 		/*
2036 		 * We must perform a read-before-write if the transfer size
2037 		 * does not cover the entire buffer.
2038 		 */
2039 		if (fs->e2fs_bsize > xfersize)
2040 			flags |= BA_CLRBUF;
2041 		else
2042 			flags &= ~BA_CLRBUF;
2043 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2044 		    ap->a_cred, &bp, flags);
2045 		if (error != 0)
2046 			break;
2047 
2048 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2049 			bp->b_flags |= B_NOCACHE;
2050 		if (uio->uio_offset + xfersize > ip->i_size)
2051 			ip->i_size = uio->uio_offset + xfersize;
2052 		size = blksize(fs, ip, lbn) - bp->b_resid;
2053 		if (size < xfersize)
2054 			xfersize = size;
2055 
2056 		error =
2057 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2058 		/*
2059 		 * If the buffer is not already filled and we encounter an
2060 		 * error while trying to fill it, we have to clear out any
2061 		 * garbage data from the pages instantiated for the buffer.
2062 		 * If we do not, a failed uiomove() during a write can leave
2063 		 * the prior contents of the pages exposed to a userland mmap.
2064 		 *
2065 		 * Note that we need only clear buffers with a transfer size
2066 		 * equal to the block size because buffers with a shorter
2067 		 * transfer size were cleared above by the call to ext2_balloc()
2068 		 * with the BA_CLRBUF flag set.
2069 		 *
2070 		 * If the source region for uiomove identically mmaps the
2071 		 * buffer, uiomove() performed the NOP copy, and the buffer
2072 		 * content remains valid because the page fault handler
2073 		 * validated the pages.
2074 		 */
2075 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2076 		    fs->e2fs_bsize == xfersize)
2077 			vfs_bio_clrbuf(bp);
2078 
2079 		vfs_bio_set_flags(bp, ioflag);
2080 
2081 		/*
2082 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2083 		 * if we have a severe page deficiency write the buffer
2084 		 * asynchronously.  Otherwise try to cluster, and if that
2085 		 * doesn't do it then either do an async write (if O_DIRECT),
2086 		 * or a delayed write (if not).
2087 		 */
2088 		if (ioflag & IO_SYNC) {
2089 			(void)bwrite(bp);
2090 		} else if (vm_page_count_severe() ||
2091 			    buf_dirty_count_severe() ||
2092 		    (ioflag & IO_ASYNC)) {
2093 			bp->b_flags |= B_CLUSTEROK;
2094 			bawrite(bp);
2095 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2096 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2097 				bp->b_flags |= B_CLUSTEROK;
2098 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2099 			} else {
2100 				bawrite(bp);
2101 			}
2102 		} else if (ioflag & IO_DIRECT) {
2103 			bp->b_flags |= B_CLUSTEROK;
2104 			bawrite(bp);
2105 		} else {
2106 			bp->b_flags |= B_CLUSTEROK;
2107 			bdwrite(bp);
2108 		}
2109 		if (error || xfersize == 0)
2110 			break;
2111 	}
2112 	/*
2113 	 * If we successfully wrote any data, and we are not the superuser
2114 	 * we clear the setuid and setgid bits as a precaution against
2115 	 * tampering.
2116 	 */
2117 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2118 	    ap->a_cred) {
2119 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2120 			ip->i_mode &= ~(ISUID | ISGID);
2121 	}
2122 	if (error) {
2123 		if (ioflag & IO_UNIT) {
2124 			(void)ext2_truncate(vp, osize,
2125 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2126 			uio->uio_offset -= resid - uio->uio_resid;
2127 			uio->uio_resid = resid;
2128 		}
2129 	}
2130 	if (uio->uio_resid != resid) {
2131 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2132 		if (ioflag & IO_SYNC)
2133 			error = ext2_update(vp, 1);
2134 	}
2135 	return (error);
2136 }
2137