xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision e3514747256465c52c3b2aedc9795f52c0d3efe9)
1 /*-
2  *  modified for EXT2FS support in Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*-
8  * Copyright (c) 1982, 1986, 1989, 1993
9  *	The Regents of the University of California.  All rights reserved.
10  * (c) UNIX System Laboratories, Inc.
11  * All or some portions of this file are derived from material licensed
12  * to the University of California by American Telephone and Telegraph
13  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14  * the permission of UNIX System Laboratories, Inc.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)ufs_vnops.c	8.7 (Berkeley) 2/3/94
41  *	@(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42  * $FreeBSD$
43  */
44 
45 #include "opt_suiddir.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/rwlock.h>
58 #include <sys/mount.h>
59 #include <sys/unistd.h>
60 #include <sys/time.h>
61 #include <sys/vnode.h>
62 #include <sys/namei.h>
63 #include <sys/lockf.h>
64 #include <sys/event.h>
65 #include <sys/conf.h>
66 #include <sys/file.h>
67 #include <sys/extattr.h>
68 #include <sys/vmmeter.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vnode_pager.h>
77 
78 #include "opt_directio.h"
79 
80 #include <ufs/ufs/dir.h>
81 
82 #include <fs/ext2fs/fs.h>
83 #include <fs/ext2fs/inode.h>
84 #include <fs/ext2fs/ext2_extern.h>
85 #include <fs/ext2fs/ext2fs.h>
86 #include <fs/ext2fs/ext2_dinode.h>
87 #include <fs/ext2fs/ext2_dir.h>
88 #include <fs/ext2fs/ext2_mount.h>
89 #include <fs/ext2fs/ext2_extattr.h>
90 
91 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
92 static void ext2_itimes_locked(struct vnode *);
93 static int ext4_ext_read(struct vop_read_args *);
94 static int ext2_ind_read(struct vop_read_args *);
95 
96 static vop_access_t	ext2_access;
97 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
98 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
99     struct thread *);
100 static vop_close_t	ext2_close;
101 static vop_create_t	ext2_create;
102 static vop_fsync_t	ext2_fsync;
103 static vop_getattr_t	ext2_getattr;
104 static vop_ioctl_t	ext2_ioctl;
105 static vop_link_t	ext2_link;
106 static vop_mkdir_t	ext2_mkdir;
107 static vop_mknod_t	ext2_mknod;
108 static vop_open_t	ext2_open;
109 static vop_pathconf_t	ext2_pathconf;
110 static vop_print_t	ext2_print;
111 static vop_read_t	ext2_read;
112 static vop_readlink_t	ext2_readlink;
113 static vop_remove_t	ext2_remove;
114 static vop_rename_t	ext2_rename;
115 static vop_rmdir_t	ext2_rmdir;
116 static vop_setattr_t	ext2_setattr;
117 static vop_strategy_t	ext2_strategy;
118 static vop_symlink_t	ext2_symlink;
119 static vop_write_t	ext2_write;
120 static vop_getextattr_t	ext2_getextattr;
121 static vop_listextattr_t	ext2_listextattr;
122 static vop_vptofh_t	ext2_vptofh;
123 static vop_close_t	ext2fifo_close;
124 static vop_kqfilter_t	ext2fifo_kqfilter;
125 
126 /* Global vfs data structures for ext2. */
127 struct vop_vector ext2_vnodeops = {
128 	.vop_default =		&default_vnodeops,
129 	.vop_access =		ext2_access,
130 	.vop_bmap =		ext2_bmap,
131 	.vop_cachedlookup =	ext2_lookup,
132 	.vop_close =		ext2_close,
133 	.vop_create =		ext2_create,
134 	.vop_fsync =		ext2_fsync,
135 	.vop_getpages =		vnode_pager_local_getpages,
136 	.vop_getpages_async =	vnode_pager_local_getpages_async,
137 	.vop_getattr =		ext2_getattr,
138 	.vop_inactive =		ext2_inactive,
139 	.vop_ioctl =		ext2_ioctl,
140 	.vop_link =		ext2_link,
141 	.vop_lookup =		vfs_cache_lookup,
142 	.vop_mkdir =		ext2_mkdir,
143 	.vop_mknod =		ext2_mknod,
144 	.vop_open =		ext2_open,
145 	.vop_pathconf =		ext2_pathconf,
146 	.vop_poll =		vop_stdpoll,
147 	.vop_print =		ext2_print,
148 	.vop_read =		ext2_read,
149 	.vop_readdir =		ext2_readdir,
150 	.vop_readlink =		ext2_readlink,
151 	.vop_reallocblks =	ext2_reallocblks,
152 	.vop_reclaim =		ext2_reclaim,
153 	.vop_remove =		ext2_remove,
154 	.vop_rename =		ext2_rename,
155 	.vop_rmdir =		ext2_rmdir,
156 	.vop_setattr =		ext2_setattr,
157 	.vop_strategy =		ext2_strategy,
158 	.vop_symlink =		ext2_symlink,
159 	.vop_write =		ext2_write,
160 	.vop_getextattr =	ext2_getextattr,
161 	.vop_listextattr =	ext2_listextattr,
162 	.vop_vptofh =		ext2_vptofh,
163 };
164 
165 struct vop_vector ext2_fifoops = {
166 	.vop_default =		&fifo_specops,
167 	.vop_access =		ext2_access,
168 	.vop_close =		ext2fifo_close,
169 	.vop_fsync =		ext2_fsync,
170 	.vop_getattr =		ext2_getattr,
171 	.vop_inactive =		ext2_inactive,
172 	.vop_kqfilter =		ext2fifo_kqfilter,
173 	.vop_print =		ext2_print,
174 	.vop_read =		VOP_PANIC,
175 	.vop_reclaim =		ext2_reclaim,
176 	.vop_setattr =		ext2_setattr,
177 	.vop_write =		VOP_PANIC,
178 	.vop_vptofh =		ext2_vptofh,
179 };
180 
181 /*
182  * A virgin directory (no blushing please).
183  * Note that the type and namlen fields are reversed relative to ext2.
184  * Also, we don't use `struct odirtemplate', since it would just cause
185  * endianness problems.
186  */
187 static struct dirtemplate mastertemplate = {
188 	0, 12, 1, EXT2_FT_DIR, ".",
189 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
190 };
191 static struct dirtemplate omastertemplate = {
192 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
193 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
194 };
195 
196 static void
197 ext2_itimes_locked(struct vnode *vp)
198 {
199 	struct inode *ip;
200 	struct timespec ts;
201 
202 	ASSERT_VI_LOCKED(vp, __func__);
203 
204 	ip = VTOI(vp);
205 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
206 		return;
207 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
208 		ip->i_flag |= IN_LAZYMOD;
209 	else
210 		ip->i_flag |= IN_MODIFIED;
211 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
212 		vfs_timestamp(&ts);
213 		if (ip->i_flag & IN_ACCESS) {
214 			ip->i_atime = ts.tv_sec;
215 			ip->i_atimensec = ts.tv_nsec;
216 		}
217 		if (ip->i_flag & IN_UPDATE) {
218 			ip->i_mtime = ts.tv_sec;
219 			ip->i_mtimensec = ts.tv_nsec;
220 			ip->i_modrev++;
221 		}
222 		if (ip->i_flag & IN_CHANGE) {
223 			ip->i_ctime = ts.tv_sec;
224 			ip->i_ctimensec = ts.tv_nsec;
225 		}
226 	}
227 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
228 }
229 
230 void
231 ext2_itimes(struct vnode *vp)
232 {
233 
234 	VI_LOCK(vp);
235 	ext2_itimes_locked(vp);
236 	VI_UNLOCK(vp);
237 }
238 
239 /*
240  * Create a regular file
241  */
242 static int
243 ext2_create(struct vop_create_args *ap)
244 {
245 	int error;
246 
247 	error =
248 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
249 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
250 	if (error != 0)
251 		return (error);
252 	if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
253 		cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
254 	return (0);
255 }
256 
257 static int
258 ext2_open(struct vop_open_args *ap)
259 {
260 
261 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
262 		return (EOPNOTSUPP);
263 
264 	/*
265 	 * Files marked append-only must be opened for appending.
266 	 */
267 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
268 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
269 		return (EPERM);
270 
271 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
272 
273 	return (0);
274 }
275 
276 /*
277  * Close called.
278  *
279  * Update the times on the inode.
280  */
281 static int
282 ext2_close(struct vop_close_args *ap)
283 {
284 	struct vnode *vp = ap->a_vp;
285 
286 	VI_LOCK(vp);
287 	if (vp->v_usecount > 1)
288 		ext2_itimes_locked(vp);
289 	VI_UNLOCK(vp);
290 	return (0);
291 }
292 
293 static int
294 ext2_access(struct vop_access_args *ap)
295 {
296 	struct vnode *vp = ap->a_vp;
297 	struct inode *ip = VTOI(vp);
298 	accmode_t accmode = ap->a_accmode;
299 	int error;
300 
301 	if (vp->v_type == VBLK || vp->v_type == VCHR)
302 		return (EOPNOTSUPP);
303 
304 	/*
305 	 * Disallow write attempts on read-only file systems;
306 	 * unless the file is a socket, fifo, or a block or
307 	 * character device resident on the file system.
308 	 */
309 	if (accmode & VWRITE) {
310 		switch (vp->v_type) {
311 		case VDIR:
312 		case VLNK:
313 		case VREG:
314 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
315 				return (EROFS);
316 			break;
317 		default:
318 			break;
319 		}
320 	}
321 
322 	/* If immutable bit set, nobody gets to write it. */
323 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
324 		return (EPERM);
325 
326 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
327 	    ap->a_accmode, ap->a_cred, NULL);
328 	return (error);
329 }
330 
331 static int
332 ext2_getattr(struct vop_getattr_args *ap)
333 {
334 	struct vnode *vp = ap->a_vp;
335 	struct inode *ip = VTOI(vp);
336 	struct vattr *vap = ap->a_vap;
337 
338 	ext2_itimes(vp);
339 	/*
340 	 * Copy from inode table
341 	 */
342 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
343 	vap->va_fileid = ip->i_number;
344 	vap->va_mode = ip->i_mode & ~IFMT;
345 	vap->va_nlink = ip->i_nlink;
346 	vap->va_uid = ip->i_uid;
347 	vap->va_gid = ip->i_gid;
348 	vap->va_rdev = ip->i_rdev;
349 	vap->va_size = ip->i_size;
350 	vap->va_atime.tv_sec = ip->i_atime;
351 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
352 	vap->va_mtime.tv_sec = ip->i_mtime;
353 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
354 	vap->va_ctime.tv_sec = ip->i_ctime;
355 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
356 	if E2DI_HAS_XTIME(ip) {
357 		vap->va_birthtime.tv_sec = ip->i_birthtime;
358 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
359 	}
360 	vap->va_flags = ip->i_flags;
361 	vap->va_gen = ip->i_gen;
362 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
363 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
364 	vap->va_type = IFTOVT(ip->i_mode);
365 	vap->va_filerev = ip->i_modrev;
366 	return (0);
367 }
368 
369 /*
370  * Set attribute vnode op. called from several syscalls
371  */
372 static int
373 ext2_setattr(struct vop_setattr_args *ap)
374 {
375 	struct vattr *vap = ap->a_vap;
376 	struct vnode *vp = ap->a_vp;
377 	struct inode *ip = VTOI(vp);
378 	struct ucred *cred = ap->a_cred;
379 	struct thread *td = curthread;
380 	int error;
381 
382 	/*
383 	 * Check for unsettable attributes.
384 	 */
385 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
386 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
387 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
388 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
389 		return (EINVAL);
390 	}
391 	if (vap->va_flags != VNOVAL) {
392 		/* Disallow flags not supported by ext2fs. */
393 		if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
394 			return (EOPNOTSUPP);
395 
396 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
397 			return (EROFS);
398 		/*
399 		 * Callers may only modify the file flags on objects they
400 		 * have VADMIN rights for.
401 		 */
402 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
403 			return (error);
404 		/*
405 		 * Unprivileged processes and privileged processes in
406 		 * jail() are not permitted to unset system flags, or
407 		 * modify flags if any system flags are set.
408 		 * Privileged non-jail processes may not modify system flags
409 		 * if securelevel > 0 and any existing system flags are set.
410 		 */
411 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
412 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
413 				error = securelevel_gt(cred, 0);
414 				if (error)
415 					return (error);
416 			}
417 		} else {
418 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
419 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
420 				return (EPERM);
421 		}
422 		ip->i_flags = vap->va_flags;
423 		ip->i_flag |= IN_CHANGE;
424 		if (ip->i_flags & (IMMUTABLE | APPEND))
425 			return (0);
426 	}
427 	if (ip->i_flags & (IMMUTABLE | APPEND))
428 		return (EPERM);
429 	/*
430 	 * Go through the fields and update iff not VNOVAL.
431 	 */
432 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
433 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
434 			return (EROFS);
435 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
436 		    td)) != 0)
437 			return (error);
438 	}
439 	if (vap->va_size != VNOVAL) {
440 		/*
441 		 * Disallow write attempts on read-only file systems;
442 		 * unless the file is a socket, fifo, or a block or
443 		 * character device resident on the file system.
444 		 */
445 		switch (vp->v_type) {
446 		case VDIR:
447 			return (EISDIR);
448 		case VLNK:
449 		case VREG:
450 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
451 				return (EROFS);
452 			break;
453 		default:
454 			break;
455 		}
456 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
457 			return (error);
458 	}
459 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
460 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
461 			return (EROFS);
462 		/*
463 		 * From utimes(2):
464 		 * If times is NULL, ... The caller must be the owner of
465 		 * the file, have permission to write the file, or be the
466 		 * super-user.
467 		 * If times is non-NULL, ... The caller must be the owner of
468 		 * the file or be the super-user.
469 		 */
470 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
471 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
472 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
473 			return (error);
474 		ip->i_flag |= IN_CHANGE | IN_MODIFIED;
475 		if (vap->va_atime.tv_sec != VNOVAL) {
476 			ip->i_flag &= ~IN_ACCESS;
477 			ip->i_atime = vap->va_atime.tv_sec;
478 			ip->i_atimensec = vap->va_atime.tv_nsec;
479 		}
480 		if (vap->va_mtime.tv_sec != VNOVAL) {
481 			ip->i_flag &= ~IN_UPDATE;
482 			ip->i_mtime = vap->va_mtime.tv_sec;
483 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
484 		}
485 		ip->i_birthtime = vap->va_birthtime.tv_sec;
486 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
487 		error = ext2_update(vp, 0);
488 		if (error)
489 			return (error);
490 	}
491 	error = 0;
492 	if (vap->va_mode != (mode_t)VNOVAL) {
493 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
494 			return (EROFS);
495 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
496 	}
497 	return (error);
498 }
499 
500 /*
501  * Change the mode on a file.
502  * Inode must be locked before calling.
503  */
504 static int
505 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
506 {
507 	struct inode *ip = VTOI(vp);
508 	int error;
509 
510 	/*
511 	 * To modify the permissions on a file, must possess VADMIN
512 	 * for that file.
513 	 */
514 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
515 		return (error);
516 	/*
517 	 * Privileged processes may set the sticky bit on non-directories,
518 	 * as well as set the setgid bit on a file with a group that the
519 	 * process is not a member of.
520 	 */
521 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
522 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
523 		if (error)
524 			return (EFTYPE);
525 	}
526 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
527 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
528 		if (error)
529 			return (error);
530 	}
531 	ip->i_mode &= ~ALLPERMS;
532 	ip->i_mode |= (mode & ALLPERMS);
533 	ip->i_flag |= IN_CHANGE;
534 	return (0);
535 }
536 
537 /*
538  * Perform chown operation on inode ip;
539  * inode must be locked prior to call.
540  */
541 static int
542 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
543     struct thread *td)
544 {
545 	struct inode *ip = VTOI(vp);
546 	uid_t ouid;
547 	gid_t ogid;
548 	int error = 0;
549 
550 	if (uid == (uid_t)VNOVAL)
551 		uid = ip->i_uid;
552 	if (gid == (gid_t)VNOVAL)
553 		gid = ip->i_gid;
554 	/*
555 	 * To modify the ownership of a file, must possess VADMIN
556 	 * for that file.
557 	 */
558 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
559 		return (error);
560 	/*
561 	 * To change the owner of a file, or change the group of a file
562 	 * to a group of which we are not a member, the caller must
563 	 * have privilege.
564 	 */
565 	if (uid != ip->i_uid || (gid != ip->i_gid &&
566 	    !groupmember(gid, cred))) {
567 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
568 		if (error)
569 			return (error);
570 	}
571 	ogid = ip->i_gid;
572 	ouid = ip->i_uid;
573 	ip->i_gid = gid;
574 	ip->i_uid = uid;
575 	ip->i_flag |= IN_CHANGE;
576 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
577 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
578 			ip->i_mode &= ~(ISUID | ISGID);
579 	}
580 	return (0);
581 }
582 
583 /*
584  * Synch an open file.
585  */
586 /* ARGSUSED */
587 static int
588 ext2_fsync(struct vop_fsync_args *ap)
589 {
590 	/*
591 	 * Flush all dirty buffers associated with a vnode.
592 	 */
593 
594 	vop_stdfsync(ap);
595 
596 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
597 }
598 
599 /*
600  * Mknod vnode call
601  */
602 /* ARGSUSED */
603 static int
604 ext2_mknod(struct vop_mknod_args *ap)
605 {
606 	struct vattr *vap = ap->a_vap;
607 	struct vnode **vpp = ap->a_vpp;
608 	struct inode *ip;
609 	ino_t ino;
610 	int error;
611 
612 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
613 	    ap->a_dvp, vpp, ap->a_cnp);
614 	if (error)
615 		return (error);
616 	ip = VTOI(*vpp);
617 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
618 	if (vap->va_rdev != VNOVAL) {
619 		/*
620 		 * Want to be able to use this to make badblock
621 		 * inodes, so don't truncate the dev number.
622 		 */
623 		ip->i_rdev = vap->va_rdev;
624 	}
625 	/*
626 	 * Remove inode, then reload it through VFS_VGET so it is
627 	 * checked to see if it is an alias of an existing entry in
628 	 * the inode cache.	 XXX I don't believe this is necessary now.
629 	 */
630 	(*vpp)->v_type = VNON;
631 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
632 	vgone(*vpp);
633 	vput(*vpp);
634 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
635 	if (error) {
636 		*vpp = NULL;
637 		return (error);
638 	}
639 	return (0);
640 }
641 
642 static int
643 ext2_remove(struct vop_remove_args *ap)
644 {
645 	struct inode *ip;
646 	struct vnode *vp = ap->a_vp;
647 	struct vnode *dvp = ap->a_dvp;
648 	int error;
649 
650 	ip = VTOI(vp);
651 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
652 	    (VTOI(dvp)->i_flags & APPEND)) {
653 		error = EPERM;
654 		goto out;
655 	}
656 	error = ext2_dirremove(dvp, ap->a_cnp);
657 	if (error == 0) {
658 		ip->i_nlink--;
659 		ip->i_flag |= IN_CHANGE;
660 	}
661 out:
662 	return (error);
663 }
664 
665 /*
666  * link vnode call
667  */
668 static int
669 ext2_link(struct vop_link_args *ap)
670 {
671 	struct vnode *vp = ap->a_vp;
672 	struct vnode *tdvp = ap->a_tdvp;
673 	struct componentname *cnp = ap->a_cnp;
674 	struct inode *ip;
675 	int error;
676 
677 #ifdef INVARIANTS
678 	if ((cnp->cn_flags & HASBUF) == 0)
679 		panic("ext2_link: no name");
680 #endif
681 	ip = VTOI(vp);
682 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
683 		error = EMLINK;
684 		goto out;
685 	}
686 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
687 		error = EPERM;
688 		goto out;
689 	}
690 	ip->i_nlink++;
691 	ip->i_flag |= IN_CHANGE;
692 	error = ext2_update(vp, !DOINGASYNC(vp));
693 	if (!error)
694 		error = ext2_direnter(ip, tdvp, cnp);
695 	if (error) {
696 		ip->i_nlink--;
697 		ip->i_flag |= IN_CHANGE;
698 	}
699 out:
700 	return (error);
701 }
702 
703 /*
704  * Rename system call.
705  * 	rename("foo", "bar");
706  * is essentially
707  *	unlink("bar");
708  *	link("foo", "bar");
709  *	unlink("foo");
710  * but ``atomically''.  Can't do full commit without saving state in the
711  * inode on disk which isn't feasible at this time.  Best we can do is
712  * always guarantee the target exists.
713  *
714  * Basic algorithm is:
715  *
716  * 1) Bump link count on source while we're linking it to the
717  *    target.  This also ensure the inode won't be deleted out
718  *    from underneath us while we work (it may be truncated by
719  *    a concurrent `trunc' or `open' for creation).
720  * 2) Link source to destination.  If destination already exists,
721  *    delete it first.
722  * 3) Unlink source reference to inode if still around. If a
723  *    directory was moved and the parent of the destination
724  *    is different from the source, patch the ".." entry in the
725  *    directory.
726  */
727 static int
728 ext2_rename(struct vop_rename_args *ap)
729 {
730 	struct vnode *tvp = ap->a_tvp;
731 	struct vnode *tdvp = ap->a_tdvp;
732 	struct vnode *fvp = ap->a_fvp;
733 	struct vnode *fdvp = ap->a_fdvp;
734 	struct componentname *tcnp = ap->a_tcnp;
735 	struct componentname *fcnp = ap->a_fcnp;
736 	struct inode *ip, *xp, *dp;
737 	struct dirtemplate dirbuf;
738 	int doingdirectory = 0, oldparent = 0, newparent = 0;
739 	int error = 0;
740 	u_char namlen;
741 
742 #ifdef INVARIANTS
743 	if ((tcnp->cn_flags & HASBUF) == 0 ||
744 	    (fcnp->cn_flags & HASBUF) == 0)
745 		panic("ext2_rename: no name");
746 #endif
747 	/*
748 	 * Check for cross-device rename.
749 	 */
750 	if ((fvp->v_mount != tdvp->v_mount) ||
751 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
752 		error = EXDEV;
753 abortit:
754 		if (tdvp == tvp)
755 			vrele(tdvp);
756 		else
757 			vput(tdvp);
758 		if (tvp)
759 			vput(tvp);
760 		vrele(fdvp);
761 		vrele(fvp);
762 		return (error);
763 	}
764 
765 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
766 	    (VTOI(tdvp)->i_flags & APPEND))) {
767 		error = EPERM;
768 		goto abortit;
769 	}
770 
771 	/*
772 	 * Renaming a file to itself has no effect.  The upper layers should
773 	 * not call us in that case.  Temporarily just warn if they do.
774 	 */
775 	if (fvp == tvp) {
776 		printf("ext2_rename: fvp == tvp (can't happen)\n");
777 		error = 0;
778 		goto abortit;
779 	}
780 
781 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
782 		goto abortit;
783 	dp = VTOI(fdvp);
784 	ip = VTOI(fvp);
785 	if (ip->i_nlink >= EXT2_LINK_MAX) {
786 		VOP_UNLOCK(fvp, 0);
787 		error = EMLINK;
788 		goto abortit;
789 	}
790 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
791 	    || (dp->i_flags & APPEND)) {
792 		VOP_UNLOCK(fvp, 0);
793 		error = EPERM;
794 		goto abortit;
795 	}
796 	if ((ip->i_mode & IFMT) == IFDIR) {
797 		/*
798 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
799 		 */
800 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
801 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
802 		    (ip->i_flag & IN_RENAME)) {
803 			VOP_UNLOCK(fvp, 0);
804 			error = EINVAL;
805 			goto abortit;
806 		}
807 		ip->i_flag |= IN_RENAME;
808 		oldparent = dp->i_number;
809 		doingdirectory++;
810 	}
811 	vrele(fdvp);
812 
813 	/*
814 	 * When the target exists, both the directory
815 	 * and target vnodes are returned locked.
816 	 */
817 	dp = VTOI(tdvp);
818 	xp = NULL;
819 	if (tvp)
820 		xp = VTOI(tvp);
821 
822 	/*
823 	 * 1) Bump link count while we're moving stuff
824 	 *    around.  If we crash somewhere before
825 	 *    completing our work, the link count
826 	 *    may be wrong, but correctable.
827 	 */
828 	ip->i_nlink++;
829 	ip->i_flag |= IN_CHANGE;
830 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
831 		VOP_UNLOCK(fvp, 0);
832 		goto bad;
833 	}
834 
835 	/*
836 	 * If ".." must be changed (ie the directory gets a new
837 	 * parent) then the source directory must not be in the
838 	 * directory hierarchy above the target, as this would
839 	 * orphan everything below the source directory. Also
840 	 * the user must have write permission in the source so
841 	 * as to be able to change "..". We must repeat the call
842 	 * to namei, as the parent directory is unlocked by the
843 	 * call to checkpath().
844 	 */
845 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
846 	VOP_UNLOCK(fvp, 0);
847 	if (oldparent != dp->i_number)
848 		newparent = dp->i_number;
849 	if (doingdirectory && newparent) {
850 		if (error)	/* write access check above */
851 			goto bad;
852 		if (xp != NULL)
853 			vput(tvp);
854 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
855 		if (error)
856 			goto out;
857 		VREF(tdvp);
858 		error = relookup(tdvp, &tvp, tcnp);
859 		if (error)
860 			goto out;
861 		vrele(tdvp);
862 		dp = VTOI(tdvp);
863 		xp = NULL;
864 		if (tvp)
865 			xp = VTOI(tvp);
866 	}
867 	/*
868 	 * 2) If target doesn't exist, link the target
869 	 *    to the source and unlink the source.
870 	 *    Otherwise, rewrite the target directory
871 	 *    entry to reference the source inode and
872 	 *    expunge the original entry's existence.
873 	 */
874 	if (xp == NULL) {
875 		if (dp->i_devvp != ip->i_devvp)
876 			panic("ext2_rename: EXDEV");
877 		/*
878 		 * Account for ".." in new directory.
879 		 * When source and destination have the same
880 		 * parent we don't fool with the link count.
881 		 */
882 		if (doingdirectory && newparent) {
883 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
884 				error = EMLINK;
885 				goto bad;
886 			}
887 			dp->i_nlink++;
888 			dp->i_flag |= IN_CHANGE;
889 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
890 			if (error)
891 				goto bad;
892 		}
893 		error = ext2_direnter(ip, tdvp, tcnp);
894 		if (error) {
895 			if (doingdirectory && newparent) {
896 				dp->i_nlink--;
897 				dp->i_flag |= IN_CHANGE;
898 				(void)ext2_update(tdvp, 1);
899 			}
900 			goto bad;
901 		}
902 		vput(tdvp);
903 	} else {
904 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
905 			panic("ext2_rename: EXDEV");
906 		/*
907 		 * Short circuit rename(foo, foo).
908 		 */
909 		if (xp->i_number == ip->i_number)
910 			panic("ext2_rename: same file");
911 		/*
912 		 * If the parent directory is "sticky", then the user must
913 		 * own the parent directory, or the destination of the rename,
914 		 * otherwise the destination may not be changed (except by
915 		 * root). This implements append-only directories.
916 		 */
917 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
918 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
919 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
920 			error = EPERM;
921 			goto bad;
922 		}
923 		/*
924 		 * Target must be empty if a directory and have no links
925 		 * to it. Also, ensure source and target are compatible
926 		 * (both directories, or both not directories).
927 		 */
928 		if ((xp->i_mode & IFMT) == IFDIR) {
929 			if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
930 			    xp->i_nlink > 2) {
931 				error = ENOTEMPTY;
932 				goto bad;
933 			}
934 			if (!doingdirectory) {
935 				error = ENOTDIR;
936 				goto bad;
937 			}
938 			cache_purge(tdvp);
939 		} else if (doingdirectory) {
940 			error = EISDIR;
941 			goto bad;
942 		}
943 		error = ext2_dirrewrite(dp, ip, tcnp);
944 		if (error)
945 			goto bad;
946 		/*
947 		 * If the target directory is in the same
948 		 * directory as the source directory,
949 		 * decrement the link count on the parent
950 		 * of the target directory.
951 		 */
952 		if (doingdirectory && !newparent) {
953 			dp->i_nlink--;
954 			dp->i_flag |= IN_CHANGE;
955 		}
956 		vput(tdvp);
957 		/*
958 		 * Adjust the link count of the target to
959 		 * reflect the dirrewrite above.  If this is
960 		 * a directory it is empty and there are
961 		 * no links to it, so we can squash the inode and
962 		 * any space associated with it.  We disallowed
963 		 * renaming over top of a directory with links to
964 		 * it above, as the remaining link would point to
965 		 * a directory without "." or ".." entries.
966 		 */
967 		xp->i_nlink--;
968 		if (doingdirectory) {
969 			if (--xp->i_nlink != 0)
970 				panic("ext2_rename: linked directory");
971 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
972 			    tcnp->cn_cred, tcnp->cn_thread);
973 		}
974 		xp->i_flag |= IN_CHANGE;
975 		vput(tvp);
976 		xp = NULL;
977 	}
978 
979 	/*
980 	 * 3) Unlink the source.
981 	 */
982 	fcnp->cn_flags &= ~MODMASK;
983 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
984 	VREF(fdvp);
985 	error = relookup(fdvp, &fvp, fcnp);
986 	if (error == 0)
987 		vrele(fdvp);
988 	if (fvp != NULL) {
989 		xp = VTOI(fvp);
990 		dp = VTOI(fdvp);
991 	} else {
992 		/*
993 		 * From name has disappeared.  IN_RENAME is not sufficient
994 		 * to protect against directory races due to timing windows,
995 		 * so we can't panic here.
996 		 */
997 		vrele(ap->a_fvp);
998 		return (0);
999 	}
1000 	/*
1001 	 * Ensure that the directory entry still exists and has not
1002 	 * changed while the new name has been entered. If the source is
1003 	 * a file then the entry may have been unlinked or renamed. In
1004 	 * either case there is no further work to be done. If the source
1005 	 * is a directory then it cannot have been rmdir'ed; its link
1006 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1007 	 * The IN_RENAME flag ensures that it cannot be moved by another
1008 	 * rename.
1009 	 */
1010 	if (xp != ip) {
1011 		/*
1012 		 * From name resolves to a different inode.  IN_RENAME is
1013 		 * not sufficient protection against timing window races
1014 		 * so we can't panic here.
1015 		 */
1016 	} else {
1017 		/*
1018 		 * If the source is a directory with a
1019 		 * new parent, the link count of the old
1020 		 * parent directory must be decremented
1021 		 * and ".." set to point to the new parent.
1022 		 */
1023 		if (doingdirectory && newparent) {
1024 			dp->i_nlink--;
1025 			dp->i_flag |= IN_CHANGE;
1026 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1027 			    sizeof(struct dirtemplate), (off_t)0,
1028 			    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1029 			    tcnp->cn_cred, NOCRED, NULL, NULL);
1030 			if (error == 0) {
1031 				/* Like ufs little-endian: */
1032 				namlen = dirbuf.dotdot_type;
1033 				if (namlen != 2 ||
1034 				    dirbuf.dotdot_name[0] != '.' ||
1035 				    dirbuf.dotdot_name[1] != '.') {
1036 					ext2_dirbad(xp, (doff_t)12,
1037 					    "rename: mangled dir");
1038 				} else {
1039 					dirbuf.dotdot_ino = newparent;
1040 					(void)vn_rdwr(UIO_WRITE, fvp,
1041 					    (caddr_t)&dirbuf,
1042 					    sizeof(struct dirtemplate),
1043 					    (off_t)0, UIO_SYSSPACE,
1044 					    IO_NODELOCKED | IO_SYNC |
1045 					    IO_NOMACCHECK, tcnp->cn_cred,
1046 					    NOCRED, NULL, NULL);
1047 					cache_purge(fdvp);
1048 				}
1049 			}
1050 		}
1051 		error = ext2_dirremove(fdvp, fcnp);
1052 		if (!error) {
1053 			xp->i_nlink--;
1054 			xp->i_flag |= IN_CHANGE;
1055 		}
1056 		xp->i_flag &= ~IN_RENAME;
1057 	}
1058 	if (dp)
1059 		vput(fdvp);
1060 	if (xp)
1061 		vput(fvp);
1062 	vrele(ap->a_fvp);
1063 	return (error);
1064 
1065 bad:
1066 	if (xp)
1067 		vput(ITOV(xp));
1068 	vput(ITOV(dp));
1069 out:
1070 	if (doingdirectory)
1071 		ip->i_flag &= ~IN_RENAME;
1072 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1073 		ip->i_nlink--;
1074 		ip->i_flag |= IN_CHANGE;
1075 		ip->i_flag &= ~IN_RENAME;
1076 		vput(fvp);
1077 	} else
1078 		vrele(fvp);
1079 	return (error);
1080 }
1081 
1082 /*
1083  * Mkdir system call
1084  */
1085 static int
1086 ext2_mkdir(struct vop_mkdir_args *ap)
1087 {
1088 	struct vnode *dvp = ap->a_dvp;
1089 	struct vattr *vap = ap->a_vap;
1090 	struct componentname *cnp = ap->a_cnp;
1091 	struct inode *ip, *dp;
1092 	struct vnode *tvp;
1093 	struct dirtemplate dirtemplate, *dtp;
1094 	int error, dmode;
1095 
1096 #ifdef INVARIANTS
1097 	if ((cnp->cn_flags & HASBUF) == 0)
1098 		panic("ext2_mkdir: no name");
1099 #endif
1100 	dp = VTOI(dvp);
1101 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1102 		error = EMLINK;
1103 		goto out;
1104 	}
1105 	dmode = vap->va_mode & 0777;
1106 	dmode |= IFDIR;
1107 	/*
1108 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1109 	 * but not have it entered in the parent directory. The entry is
1110 	 * made later after writing "." and ".." entries.
1111 	 */
1112 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1113 	if (error)
1114 		goto out;
1115 	ip = VTOI(tvp);
1116 	ip->i_gid = dp->i_gid;
1117 #ifdef SUIDDIR
1118 	{
1119 		/*
1120 		 * if we are hacking owners here, (only do this where told to)
1121 		 * and we are not giving it TOO root, (would subvert quotas)
1122 		 * then go ahead and give it to the other user.
1123 		 * The new directory also inherits the SUID bit.
1124 		 * If user's UID and dir UID are the same,
1125 		 * 'give it away' so that the SUID is still forced on.
1126 		 */
1127 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1128 		    (dp->i_mode & ISUID) && dp->i_uid) {
1129 			dmode |= ISUID;
1130 			ip->i_uid = dp->i_uid;
1131 		} else {
1132 			ip->i_uid = cnp->cn_cred->cr_uid;
1133 		}
1134 	}
1135 #else
1136 	ip->i_uid = cnp->cn_cred->cr_uid;
1137 #endif
1138 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1139 	ip->i_mode = dmode;
1140 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1141 	ip->i_nlink = 2;
1142 	if (cnp->cn_flags & ISWHITEOUT)
1143 		ip->i_flags |= UF_OPAQUE;
1144 	error = ext2_update(tvp, 1);
1145 
1146 	/*
1147 	 * Bump link count in parent directory
1148 	 * to reflect work done below.  Should
1149 	 * be done before reference is created
1150 	 * so reparation is possible if we crash.
1151 	 */
1152 	dp->i_nlink++;
1153 	dp->i_flag |= IN_CHANGE;
1154 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1155 	if (error)
1156 		goto bad;
1157 
1158 	/* Initialize directory with "." and ".." from static template. */
1159 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1160 	    EXT2F_INCOMPAT_FTYPE))
1161 		dtp = &mastertemplate;
1162 	else
1163 		dtp = &omastertemplate;
1164 	dirtemplate = *dtp;
1165 	dirtemplate.dot_ino = ip->i_number;
1166 	dirtemplate.dotdot_ino = dp->i_number;
1167 	/*
1168 	 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1169 	 * just redefine it - for this function only
1170 	 */
1171 #undef  DIRBLKSIZ
1172 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1173 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1174 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1175 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1176 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1177 	    NULL, NULL);
1178 	if (error) {
1179 		dp->i_nlink--;
1180 		dp->i_flag |= IN_CHANGE;
1181 		goto bad;
1182 	}
1183 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1184 		/* XXX should grow with balloc() */
1185 		panic("ext2_mkdir: blksize");
1186 	else {
1187 		ip->i_size = DIRBLKSIZ;
1188 		ip->i_flag |= IN_CHANGE;
1189 	}
1190 
1191 	/* Directory set up, now install its entry in the parent directory. */
1192 	error = ext2_direnter(ip, dvp, cnp);
1193 	if (error) {
1194 		dp->i_nlink--;
1195 		dp->i_flag |= IN_CHANGE;
1196 	}
1197 bad:
1198 	/*
1199 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1200 	 * for us because we set the link count to 0.
1201 	 */
1202 	if (error) {
1203 		ip->i_nlink = 0;
1204 		ip->i_flag |= IN_CHANGE;
1205 		vput(tvp);
1206 	} else
1207 		*ap->a_vpp = tvp;
1208 out:
1209 	return (error);
1210 #undef  DIRBLKSIZ
1211 #define DIRBLKSIZ  DEV_BSIZE
1212 }
1213 
1214 /*
1215  * Rmdir system call.
1216  */
1217 static int
1218 ext2_rmdir(struct vop_rmdir_args *ap)
1219 {
1220 	struct vnode *vp = ap->a_vp;
1221 	struct vnode *dvp = ap->a_dvp;
1222 	struct componentname *cnp = ap->a_cnp;
1223 	struct inode *ip, *dp;
1224 	int error;
1225 
1226 	ip = VTOI(vp);
1227 	dp = VTOI(dvp);
1228 
1229 	/*
1230 	 * Verify the directory is empty (and valid).
1231 	 * (Rmdir ".." won't be valid since
1232 	 *  ".." will contain a reference to
1233 	 *  the current directory and thus be
1234 	 *  non-empty.)
1235 	 */
1236 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1237 		error = ENOTEMPTY;
1238 		goto out;
1239 	}
1240 	if ((dp->i_flags & APPEND)
1241 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1242 		error = EPERM;
1243 		goto out;
1244 	}
1245 	/*
1246 	 * Delete reference to directory before purging
1247 	 * inode.  If we crash in between, the directory
1248 	 * will be reattached to lost+found,
1249 	 */
1250 	error = ext2_dirremove(dvp, cnp);
1251 	if (error)
1252 		goto out;
1253 	dp->i_nlink--;
1254 	dp->i_flag |= IN_CHANGE;
1255 	cache_purge(dvp);
1256 	VOP_UNLOCK(dvp, 0);
1257 	/*
1258 	 * Truncate inode.  The only stuff left
1259 	 * in the directory is "." and "..".  The
1260 	 * "." reference is inconsequential since
1261 	 * we're quashing it.  The ".." reference
1262 	 * has already been adjusted above.  We've
1263 	 * removed the "." reference and the reference
1264 	 * in the parent directory, but there may be
1265 	 * other hard links so decrement by 2 and
1266 	 * worry about them later.
1267 	 */
1268 	ip->i_nlink -= 2;
1269 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1270 	    cnp->cn_thread);
1271 	cache_purge(ITOV(ip));
1272 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1273 		VOP_UNLOCK(vp, 0);
1274 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1275 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1276 	}
1277 out:
1278 	return (error);
1279 }
1280 
1281 /*
1282  * symlink -- make a symbolic link
1283  */
1284 static int
1285 ext2_symlink(struct vop_symlink_args *ap)
1286 {
1287 	struct vnode *vp, **vpp = ap->a_vpp;
1288 	struct inode *ip;
1289 	int len, error;
1290 
1291 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1292 	    vpp, ap->a_cnp);
1293 	if (error)
1294 		return (error);
1295 	vp = *vpp;
1296 	len = strlen(ap->a_target);
1297 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1298 		ip = VTOI(vp);
1299 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1300 		ip->i_size = len;
1301 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1302 	} else
1303 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1304 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1305 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1306 	if (error)
1307 		vput(vp);
1308 	return (error);
1309 }
1310 
1311 /*
1312  * Return target name of a symbolic link
1313  */
1314 static int
1315 ext2_readlink(struct vop_readlink_args *ap)
1316 {
1317 	struct vnode *vp = ap->a_vp;
1318 	struct inode *ip = VTOI(vp);
1319 	int isize;
1320 
1321 	isize = ip->i_size;
1322 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1323 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1324 		return (0);
1325 	}
1326 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1327 }
1328 
1329 /*
1330  * Calculate the logical to physical mapping if not done already,
1331  * then call the device strategy routine.
1332  *
1333  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1334  * deadlock on memory.  See ext2_bmap() for details.
1335  */
1336 static int
1337 ext2_strategy(struct vop_strategy_args *ap)
1338 {
1339 	struct buf *bp = ap->a_bp;
1340 	struct vnode *vp = ap->a_vp;
1341 	struct bufobj *bo;
1342 	daddr_t blkno;
1343 	int error;
1344 
1345 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1346 		panic("ext2_strategy: spec");
1347 	if (bp->b_blkno == bp->b_lblkno) {
1348 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1349 		bp->b_blkno = blkno;
1350 		if (error) {
1351 			bp->b_error = error;
1352 			bp->b_ioflags |= BIO_ERROR;
1353 			bufdone(bp);
1354 			return (0);
1355 		}
1356 		if ((long)bp->b_blkno == -1)
1357 			vfs_bio_clrbuf(bp);
1358 	}
1359 	if ((long)bp->b_blkno == -1) {
1360 		bufdone(bp);
1361 		return (0);
1362 	}
1363 	bp->b_iooffset = dbtob(bp->b_blkno);
1364 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1365 	BO_STRATEGY(bo, bp);
1366 	return (0);
1367 }
1368 
1369 /*
1370  * Print out the contents of an inode.
1371  */
1372 static int
1373 ext2_print(struct vop_print_args *ap)
1374 {
1375 	struct vnode *vp = ap->a_vp;
1376 	struct inode *ip = VTOI(vp);
1377 
1378 	vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1379 	if (vp->v_type == VFIFO)
1380 		fifo_printinfo(vp);
1381 	printf("\n");
1382 	return (0);
1383 }
1384 
1385 /*
1386  * Close wrapper for fifos.
1387  *
1388  * Update the times on the inode then do device close.
1389  */
1390 static int
1391 ext2fifo_close(struct vop_close_args *ap)
1392 {
1393 	struct vnode *vp = ap->a_vp;
1394 
1395 	VI_LOCK(vp);
1396 	if (vp->v_usecount > 1)
1397 		ext2_itimes_locked(vp);
1398 	VI_UNLOCK(vp);
1399 	return (fifo_specops.vop_close(ap));
1400 }
1401 
1402 /*
1403  * Kqfilter wrapper for fifos.
1404  *
1405  * Fall through to ext2 kqfilter routines if needed
1406  */
1407 static int
1408 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1409 {
1410 	int error;
1411 
1412 	error = fifo_specops.vop_kqfilter(ap);
1413 	if (error)
1414 		error = vfs_kqfilter(ap);
1415 	return (error);
1416 }
1417 
1418 /*
1419  * Return POSIX pathconf information applicable to ext2 filesystems.
1420  */
1421 static int
1422 ext2_pathconf(struct vop_pathconf_args *ap)
1423 {
1424 	int error = 0;
1425 
1426 	switch (ap->a_name) {
1427 	case _PC_LINK_MAX:
1428 		*ap->a_retval = EXT2_LINK_MAX;
1429 		break;
1430 	case _PC_NAME_MAX:
1431 		*ap->a_retval = NAME_MAX;
1432 		break;
1433 	case _PC_PATH_MAX:
1434 		*ap->a_retval = PATH_MAX;
1435 		break;
1436 	case _PC_PIPE_BUF:
1437 		*ap->a_retval = PIPE_BUF;
1438 		break;
1439 	case _PC_CHOWN_RESTRICTED:
1440 		*ap->a_retval = 1;
1441 		break;
1442 	case _PC_NO_TRUNC:
1443 		*ap->a_retval = 1;
1444 		break;
1445 	case _PC_MIN_HOLE_SIZE:
1446 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1447 		break;
1448 	case _PC_ASYNC_IO:
1449 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1450 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1451 		error = EINVAL;
1452 		break;
1453 	case _PC_PRIO_IO:
1454 		*ap->a_retval = 0;
1455 		break;
1456 	case _PC_SYNC_IO:
1457 		*ap->a_retval = 0;
1458 		break;
1459 	case _PC_ALLOC_SIZE_MIN:
1460 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1461 		break;
1462 	case _PC_FILESIZEBITS:
1463 		*ap->a_retval = 64;
1464 		break;
1465 	case _PC_REC_INCR_XFER_SIZE:
1466 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1467 		break;
1468 	case _PC_REC_MAX_XFER_SIZE:
1469 		*ap->a_retval = -1;	/* means ``unlimited'' */
1470 		break;
1471 	case _PC_REC_MIN_XFER_SIZE:
1472 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1473 		break;
1474 	case _PC_REC_XFER_ALIGN:
1475 		*ap->a_retval = PAGE_SIZE;
1476 		break;
1477 	case _PC_SYMLINK_MAX:
1478 		*ap->a_retval = MAXPATHLEN;
1479 		break;
1480 
1481 	default:
1482 		error = EINVAL;
1483 		break;
1484 	}
1485 	return (error);
1486 }
1487 
1488 /*
1489  * Vnode operation to retrieve a named extended attribute.
1490  */
1491 static int
1492 ext2_getextattr(struct vop_getextattr_args *ap)
1493 {
1494 	struct inode *ip;
1495 	struct m_ext2fs *fs;
1496 	int error;
1497 
1498 	ip = VTOI(ap->a_vp);
1499 	fs = ip->i_e2fs;
1500 
1501 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1502 		return (EOPNOTSUPP);
1503 
1504 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1505 		return (EOPNOTSUPP);
1506 
1507 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1508 	    ap->a_cred, ap->a_td, VREAD);
1509 	if (error)
1510 		return (error);
1511 
1512 	if (ap->a_size != NULL)
1513 		*ap->a_size = 0;
1514 
1515 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1516 		error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1517 		    ap->a_name, ap->a_uio, ap->a_size);
1518 		if (error)
1519 			return (error);
1520 	}
1521 
1522 	if (ip->i_facl)
1523 		error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1524 		    ap->a_name, ap->a_uio, ap->a_size);
1525 
1526 	return (error);
1527 }
1528 
1529 /*
1530  * Vnode operation to retrieve extended attributes on a vnode.
1531  */
1532 static int
1533 ext2_listextattr(struct vop_listextattr_args *ap)
1534 {
1535 	struct inode *ip;
1536 	struct m_ext2fs *fs;
1537 	int error;
1538 
1539 	ip = VTOI(ap->a_vp);
1540 	fs = ip->i_e2fs;
1541 
1542 	if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1543 		return (EOPNOTSUPP);
1544 
1545 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1546 		return (EOPNOTSUPP);
1547 
1548 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1549 	    ap->a_cred, ap->a_td, VREAD);
1550 	if (error)
1551 		return (error);
1552 
1553 	if (ap->a_size != NULL)
1554 		*ap->a_size = 0;
1555 
1556 	if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1557 		error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1558 		    ap->a_uio, ap->a_size);
1559 		if(error)
1560 			return (error);
1561 	}
1562 
1563 	if(ip->i_facl)
1564 		error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1565 		    ap->a_uio, ap->a_size);
1566 
1567 	return (error);
1568 }
1569 
1570 /*
1571  * Vnode pointer to File handle
1572  */
1573 /* ARGSUSED */
1574 static int
1575 ext2_vptofh(struct vop_vptofh_args *ap)
1576 {
1577 	struct inode *ip;
1578 	struct ufid *ufhp;
1579 
1580 	ip = VTOI(ap->a_vp);
1581 	ufhp = (struct ufid *)ap->a_fhp;
1582 	ufhp->ufid_len = sizeof(struct ufid);
1583 	ufhp->ufid_ino = ip->i_number;
1584 	ufhp->ufid_gen = ip->i_gen;
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Initialize the vnode associated with a new inode, handle aliased
1590  * vnodes.
1591  */
1592 int
1593 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1594 {
1595 	struct inode *ip;
1596 	struct vnode *vp;
1597 
1598 	vp = *vpp;
1599 	ip = VTOI(vp);
1600 	vp->v_type = IFTOVT(ip->i_mode);
1601 	if (vp->v_type == VFIFO)
1602 		vp->v_op = fifoops;
1603 
1604 	if (ip->i_number == EXT2_ROOTINO)
1605 		vp->v_vflag |= VV_ROOT;
1606 	ip->i_modrev = init_va_filerev();
1607 	*vpp = vp;
1608 	return (0);
1609 }
1610 
1611 /*
1612  * Allocate a new inode.
1613  */
1614 static int
1615 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1616     struct componentname *cnp)
1617 {
1618 	struct inode *ip, *pdir;
1619 	struct vnode *tvp;
1620 	int error;
1621 
1622 	pdir = VTOI(dvp);
1623 #ifdef INVARIANTS
1624 	if ((cnp->cn_flags & HASBUF) == 0)
1625 		panic("ext2_makeinode: no name");
1626 #endif
1627 	*vpp = NULL;
1628 	if ((mode & IFMT) == 0)
1629 		mode |= IFREG;
1630 
1631 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1632 	if (error) {
1633 		return (error);
1634 	}
1635 	ip = VTOI(tvp);
1636 	ip->i_gid = pdir->i_gid;
1637 #ifdef SUIDDIR
1638 	{
1639 		/*
1640 		 * if we are
1641 		 * not the owner of the directory,
1642 		 * and we are hacking owners here, (only do this where told to)
1643 		 * and we are not giving it TOO root, (would subvert quotas)
1644 		 * then go ahead and give it to the other user.
1645 		 * Note that this drops off the execute bits for security.
1646 		 */
1647 		if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1648 		    (pdir->i_mode & ISUID) &&
1649 		    (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1650 			ip->i_uid = pdir->i_uid;
1651 			mode &= ~07111;
1652 		} else {
1653 			ip->i_uid = cnp->cn_cred->cr_uid;
1654 		}
1655 	}
1656 #else
1657 	ip->i_uid = cnp->cn_cred->cr_uid;
1658 #endif
1659 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1660 	ip->i_mode = mode;
1661 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1662 	ip->i_nlink = 1;
1663 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1664 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1665 			ip->i_mode &= ~ISGID;
1666 	}
1667 
1668 	if (cnp->cn_flags & ISWHITEOUT)
1669 		ip->i_flags |= UF_OPAQUE;
1670 
1671 	/*
1672 	 * Make sure inode goes to disk before directory entry.
1673 	 */
1674 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1675 	if (error)
1676 		goto bad;
1677 	error = ext2_direnter(ip, dvp, cnp);
1678 	if (error)
1679 		goto bad;
1680 
1681 	*vpp = tvp;
1682 	return (0);
1683 
1684 bad:
1685 	/*
1686 	 * Write error occurred trying to update the inode
1687 	 * or the directory so must deallocate the inode.
1688 	 */
1689 	ip->i_nlink = 0;
1690 	ip->i_flag |= IN_CHANGE;
1691 	vput(tvp);
1692 	return (error);
1693 }
1694 
1695 /*
1696  * Vnode op for reading.
1697  */
1698 static int
1699 ext2_read(struct vop_read_args *ap)
1700 {
1701 	struct vnode *vp;
1702 	struct inode *ip;
1703 	int error;
1704 
1705 	vp = ap->a_vp;
1706 	ip = VTOI(vp);
1707 
1708 	/* EXT4_EXT_LOCK(ip); */
1709 	if (ip->i_flag & IN_E4EXTENTS)
1710 		error = ext4_ext_read(ap);
1711 	else
1712 		error = ext2_ind_read(ap);
1713 	/* EXT4_EXT_UNLOCK(ip); */
1714 	return (error);
1715 }
1716 
1717 /*
1718  * Vnode op for reading.
1719  */
1720 static int
1721 ext2_ind_read(struct vop_read_args *ap)
1722 {
1723 	struct vnode *vp;
1724 	struct inode *ip;
1725 	struct uio *uio;
1726 	struct m_ext2fs *fs;
1727 	struct buf *bp;
1728 	daddr_t lbn, nextlbn;
1729 	off_t bytesinfile;
1730 	long size, xfersize, blkoffset;
1731 	int error, orig_resid, seqcount;
1732 	int ioflag;
1733 
1734 	vp = ap->a_vp;
1735 	uio = ap->a_uio;
1736 	ioflag = ap->a_ioflag;
1737 
1738 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1739 	ip = VTOI(vp);
1740 
1741 #ifdef INVARIANTS
1742 	if (uio->uio_rw != UIO_READ)
1743 		panic("%s: mode", "ext2_read");
1744 
1745 	if (vp->v_type == VLNK) {
1746 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1747 			panic("%s: short symlink", "ext2_read");
1748 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1749 		panic("%s: type %d", "ext2_read", vp->v_type);
1750 #endif
1751 	orig_resid = uio->uio_resid;
1752 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1753 	if (orig_resid == 0)
1754 		return (0);
1755 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1756 	fs = ip->i_e2fs;
1757 	if (uio->uio_offset < ip->i_size &&
1758 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1759 		return (EOVERFLOW);
1760 
1761 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1762 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1763 			break;
1764 		lbn = lblkno(fs, uio->uio_offset);
1765 		nextlbn = lbn + 1;
1766 		size = blksize(fs, ip, lbn);
1767 		blkoffset = blkoff(fs, uio->uio_offset);
1768 
1769 		xfersize = fs->e2fs_fsize - blkoffset;
1770 		if (uio->uio_resid < xfersize)
1771 			xfersize = uio->uio_resid;
1772 		if (bytesinfile < xfersize)
1773 			xfersize = bytesinfile;
1774 
1775 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1776 			error = bread(vp, lbn, size, NOCRED, &bp);
1777 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1778 			error = cluster_read(vp, ip->i_size, lbn, size,
1779 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1780 			    0, &bp);
1781 		} else if (seqcount > 1) {
1782 			u_int nextsize = blksize(fs, ip, nextlbn);
1783 
1784 			error = breadn(vp, lbn,
1785 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1786 		} else
1787 			error = bread(vp, lbn, size, NOCRED, &bp);
1788 		if (error) {
1789 			brelse(bp);
1790 			bp = NULL;
1791 			break;
1792 		}
1793 
1794 		/*
1795 		 * We should only get non-zero b_resid when an I/O error
1796 		 * has occurred, which should cause us to break above.
1797 		 * However, if the short read did not cause an error,
1798 		 * then we want to ensure that we do not uiomove bad
1799 		 * or uninitialized data.
1800 		 */
1801 		size -= bp->b_resid;
1802 		if (size < xfersize) {
1803 			if (size == 0)
1804 				break;
1805 			xfersize = size;
1806 		}
1807 		error = uiomove((char *)bp->b_data + blkoffset,
1808 		    (int)xfersize, uio);
1809 		if (error)
1810 			break;
1811 		vfs_bio_brelse(bp, ioflag);
1812 	}
1813 
1814 	/*
1815 	 * This can only happen in the case of an error because the loop
1816 	 * above resets bp to NULL on each iteration and on normal
1817 	 * completion has not set a new value into it. so it must have come
1818 	 * from a 'break' statement
1819 	 */
1820 	if (bp != NULL)
1821 		vfs_bio_brelse(bp, ioflag);
1822 
1823 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1824 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1825 		ip->i_flag |= IN_ACCESS;
1826 	return (error);
1827 }
1828 
1829 static int
1830 ext2_ioctl(struct vop_ioctl_args *ap)
1831 {
1832 
1833 	switch (ap->a_command) {
1834 	case FIOSEEKDATA:
1835 	case FIOSEEKHOLE:
1836 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1837 		    (off_t *)ap->a_data, ap->a_cred));
1838 	default:
1839 		return (ENOTTY);
1840 	}
1841 }
1842 
1843 /*
1844  * this function handles ext4 extents block mapping
1845  */
1846 static int
1847 ext4_ext_read(struct vop_read_args *ap)
1848 {
1849 	static unsigned char zeroes[EXT2_MAX_BLOCK_SIZE];
1850 	struct vnode *vp;
1851 	struct inode *ip;
1852 	struct uio *uio;
1853 	struct m_ext2fs *fs;
1854 	struct buf *bp;
1855 	struct ext4_extent nex, *ep;
1856 	struct ext4_extent_path path;
1857 	daddr_t lbn, newblk;
1858 	off_t bytesinfile;
1859 	int cache_type;
1860 	ssize_t orig_resid;
1861 	int error;
1862 	long size, xfersize, blkoffset;
1863 
1864 	vp = ap->a_vp;
1865 	ip = VTOI(vp);
1866 	uio = ap->a_uio;
1867 	memset(&path, 0, sizeof(path));
1868 
1869 	orig_resid = uio->uio_resid;
1870 	KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1871 	if (orig_resid == 0)
1872 		return (0);
1873 	KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1874 	fs = ip->i_e2fs;
1875 	if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1876 		return (EOVERFLOW);
1877 
1878 	while (uio->uio_resid > 0) {
1879 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1880 			break;
1881 		lbn = lblkno(fs, uio->uio_offset);
1882 		size = blksize(fs, ip, lbn);
1883 		blkoffset = blkoff(fs, uio->uio_offset);
1884 
1885 		xfersize = fs->e2fs_fsize - blkoffset;
1886 		xfersize = MIN(xfersize, uio->uio_resid);
1887 		xfersize = MIN(xfersize, bytesinfile);
1888 
1889 		/* get block from ext4 extent cache */
1890 		cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1891 		switch (cache_type) {
1892 		case EXT4_EXT_CACHE_NO:
1893 			ext4_ext_find_extent(fs, ip, lbn, &path);
1894 			if (path.ep_is_sparse)
1895 				ep = &path.ep_sparse_ext;
1896 			else
1897 				ep = path.ep_ext;
1898 			if (ep == NULL)
1899 				return (EIO);
1900 
1901 			ext4_ext_put_cache(ip, ep,
1902 			    path.ep_is_sparse ? EXT4_EXT_CACHE_GAP : EXT4_EXT_CACHE_IN);
1903 
1904 			newblk = lbn - ep->e_blk + (ep->e_start_lo |
1905 			    (daddr_t)ep->e_start_hi << 32);
1906 
1907 			if (path.ep_bp != NULL) {
1908 				brelse(path.ep_bp);
1909 				path.ep_bp = NULL;
1910 			}
1911 			break;
1912 
1913 		case EXT4_EXT_CACHE_GAP:
1914 			/* block has not been allocated yet */
1915 			break;
1916 
1917 		case EXT4_EXT_CACHE_IN:
1918 			newblk = lbn - nex.e_blk + (nex.e_start_lo |
1919 			    (daddr_t)nex.e_start_hi << 32);
1920 			break;
1921 
1922 		default:
1923 			panic("%s: invalid cache type", __func__);
1924 		}
1925 
1926 		if (cache_type == EXT4_EXT_CACHE_GAP ||
1927 		    (cache_type == EXT4_EXT_CACHE_NO && path.ep_is_sparse)) {
1928 			if (xfersize > sizeof(zeroes))
1929 				xfersize = sizeof(zeroes);
1930 			error = uiomove(zeroes, xfersize, uio);
1931 			if (error)
1932 				return (error);
1933 		} else {
1934 			error = bread(ip->i_devvp, fsbtodb(fs, newblk), size,
1935 			    NOCRED, &bp);
1936 			if (error) {
1937 				brelse(bp);
1938 				return (error);
1939 			}
1940 
1941 			size -= bp->b_resid;
1942 			if (size < xfersize) {
1943 				if (size == 0) {
1944 					bqrelse(bp);
1945 					break;
1946 				}
1947 				xfersize = size;
1948 			}
1949 			error = uiomove(bp->b_data + blkoffset, xfersize, uio);
1950 			bqrelse(bp);
1951 			if (error)
1952 				return (error);
1953 		}
1954 	}
1955 
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Vnode op for writing.
1961  */
1962 static int
1963 ext2_write(struct vop_write_args *ap)
1964 {
1965 	struct vnode *vp;
1966 	struct uio *uio;
1967 	struct inode *ip;
1968 	struct m_ext2fs *fs;
1969 	struct buf *bp;
1970 	daddr_t lbn;
1971 	off_t osize;
1972 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1973 
1974 	ioflag = ap->a_ioflag;
1975 	uio = ap->a_uio;
1976 	vp = ap->a_vp;
1977 
1978 	seqcount = ioflag >> IO_SEQSHIFT;
1979 	ip = VTOI(vp);
1980 
1981 #ifdef INVARIANTS
1982 	if (uio->uio_rw != UIO_WRITE)
1983 		panic("%s: mode", "ext2_write");
1984 #endif
1985 
1986 	switch (vp->v_type) {
1987 	case VREG:
1988 		if (ioflag & IO_APPEND)
1989 			uio->uio_offset = ip->i_size;
1990 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1991 			return (EPERM);
1992 		/* FALLTHROUGH */
1993 	case VLNK:
1994 		break;
1995 	case VDIR:
1996 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
1997 		if ((ioflag & IO_SYNC) == 0)
1998 			panic("ext2_write: nonsync dir write");
1999 		break;
2000 	default:
2001 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2002 		    vp->v_type, (intmax_t)uio->uio_offset,
2003 		    (intmax_t)uio->uio_resid);
2004 	}
2005 
2006 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2007 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2008 	fs = ip->i_e2fs;
2009 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2010 		return (EFBIG);
2011 	/*
2012 	 * Maybe this should be above the vnode op call, but so long as
2013 	 * file servers have no limits, I don't think it matters.
2014 	 */
2015 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2016 		return (EFBIG);
2017 
2018 	resid = uio->uio_resid;
2019 	osize = ip->i_size;
2020 	if (seqcount > BA_SEQMAX)
2021 		flags = BA_SEQMAX << BA_SEQSHIFT;
2022 	else
2023 		flags = seqcount << BA_SEQSHIFT;
2024 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2025 		flags |= IO_SYNC;
2026 
2027 	for (error = 0; uio->uio_resid > 0;) {
2028 		lbn = lblkno(fs, uio->uio_offset);
2029 		blkoffset = blkoff(fs, uio->uio_offset);
2030 		xfersize = fs->e2fs_fsize - blkoffset;
2031 		if (uio->uio_resid < xfersize)
2032 			xfersize = uio->uio_resid;
2033 		if (uio->uio_offset + xfersize > ip->i_size)
2034 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2035 
2036 		/*
2037 		 * We must perform a read-before-write if the transfer size
2038 		 * does not cover the entire buffer.
2039 		 */
2040 		if (fs->e2fs_bsize > xfersize)
2041 			flags |= BA_CLRBUF;
2042 		else
2043 			flags &= ~BA_CLRBUF;
2044 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2045 		    ap->a_cred, &bp, flags);
2046 		if (error != 0)
2047 			break;
2048 
2049 		if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2050 			bp->b_flags |= B_NOCACHE;
2051 		if (uio->uio_offset + xfersize > ip->i_size)
2052 			ip->i_size = uio->uio_offset + xfersize;
2053 		size = blksize(fs, ip, lbn) - bp->b_resid;
2054 		if (size < xfersize)
2055 			xfersize = size;
2056 
2057 		error =
2058 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2059 		/*
2060 		 * If the buffer is not already filled and we encounter an
2061 		 * error while trying to fill it, we have to clear out any
2062 		 * garbage data from the pages instantiated for the buffer.
2063 		 * If we do not, a failed uiomove() during a write can leave
2064 		 * the prior contents of the pages exposed to a userland mmap.
2065 		 *
2066 		 * Note that we need only clear buffers with a transfer size
2067 		 * equal to the block size because buffers with a shorter
2068 		 * transfer size were cleared above by the call to ext2_balloc()
2069 		 * with the BA_CLRBUF flag set.
2070 		 *
2071 		 * If the source region for uiomove identically mmaps the
2072 		 * buffer, uiomove() performed the NOP copy, and the buffer
2073 		 * content remains valid because the page fault handler
2074 		 * validated the pages.
2075 		 */
2076 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2077 		    fs->e2fs_bsize == xfersize)
2078 			vfs_bio_clrbuf(bp);
2079 
2080 		vfs_bio_set_flags(bp, ioflag);
2081 
2082 		/*
2083 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2084 		 * if we have a severe page deficiency write the buffer
2085 		 * asynchronously.  Otherwise try to cluster, and if that
2086 		 * doesn't do it then either do an async write (if O_DIRECT),
2087 		 * or a delayed write (if not).
2088 		 */
2089 		if (ioflag & IO_SYNC) {
2090 			(void)bwrite(bp);
2091 		} else if (vm_page_count_severe() ||
2092 			    buf_dirty_count_severe() ||
2093 		    (ioflag & IO_ASYNC)) {
2094 			bp->b_flags |= B_CLUSTEROK;
2095 			bawrite(bp);
2096 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2097 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2098 				bp->b_flags |= B_CLUSTEROK;
2099 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2100 			} else {
2101 				bawrite(bp);
2102 			}
2103 		} else if (ioflag & IO_DIRECT) {
2104 			bp->b_flags |= B_CLUSTEROK;
2105 			bawrite(bp);
2106 		} else {
2107 			bp->b_flags |= B_CLUSTEROK;
2108 			bdwrite(bp);
2109 		}
2110 		if (error || xfersize == 0)
2111 			break;
2112 	}
2113 	/*
2114 	 * If we successfully wrote any data, and we are not the superuser
2115 	 * we clear the setuid and setgid bits as a precaution against
2116 	 * tampering.
2117 	 */
2118 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2119 	    ap->a_cred) {
2120 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2121 			ip->i_mode &= ~(ISUID | ISGID);
2122 	}
2123 	if (error) {
2124 		if (ioflag & IO_UNIT) {
2125 			(void)ext2_truncate(vp, osize,
2126 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2127 			uio->uio_offset -= resid - uio->uio_resid;
2128 			uio->uio_resid = resid;
2129 		}
2130 	}
2131 	if (uio->uio_resid != resid) {
2132 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2133 		if (ioflag & IO_SYNC)
2134 			error = ext2_update(vp, 1);
2135 	}
2136 	return (error);
2137 }
2138