xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 0e97acdf58fe27b09c4824a474b0344daf997c5f)
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  * 4. 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 
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vnode_pager.h>
75 
76 #include "opt_directio.h"
77 
78 #include <ufs/ufs/dir.h>
79 
80 #include <fs/ext2fs/fs.h>
81 #include <fs/ext2fs/inode.h>
82 #include <fs/ext2fs/ext2_extern.h>
83 #include <fs/ext2fs/ext2fs.h>
84 #include <fs/ext2fs/ext2_dinode.h>
85 #include <fs/ext2fs/ext2_dir.h>
86 #include <fs/ext2fs/ext2_mount.h>
87 
88 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
89 static void ext2_itimes_locked(struct vnode *);
90 static int ext4_ext_read(struct vop_read_args *);
91 static int ext2_ind_read(struct vop_read_args *);
92 
93 static vop_access_t	ext2_access;
94 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
95 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
96     struct thread *);
97 static vop_close_t	ext2_close;
98 static vop_create_t	ext2_create;
99 static vop_fsync_t	ext2_fsync;
100 static vop_getattr_t	ext2_getattr;
101 static vop_ioctl_t	ext2_ioctl;
102 static vop_link_t	ext2_link;
103 static vop_mkdir_t	ext2_mkdir;
104 static vop_mknod_t	ext2_mknod;
105 static vop_open_t	ext2_open;
106 static vop_pathconf_t	ext2_pathconf;
107 static vop_print_t	ext2_print;
108 static vop_read_t	ext2_read;
109 static vop_readlink_t	ext2_readlink;
110 static vop_remove_t	ext2_remove;
111 static vop_rename_t	ext2_rename;
112 static vop_rmdir_t	ext2_rmdir;
113 static vop_setattr_t	ext2_setattr;
114 static vop_strategy_t	ext2_strategy;
115 static vop_symlink_t	ext2_symlink;
116 static vop_write_t	ext2_write;
117 static vop_vptofh_t	ext2_vptofh;
118 static vop_close_t	ext2fifo_close;
119 static vop_kqfilter_t	ext2fifo_kqfilter;
120 
121 /* Global vfs data structures for ext2. */
122 struct vop_vector ext2_vnodeops = {
123 	.vop_default =		&default_vnodeops,
124 	.vop_access =		ext2_access,
125 	.vop_bmap =		ext2_bmap,
126 	.vop_cachedlookup =	ext2_lookup,
127 	.vop_close =		ext2_close,
128 	.vop_create =		ext2_create,
129 	.vop_fsync =		ext2_fsync,
130 	.vop_getpages =		vnode_pager_local_getpages,
131 	.vop_getattr =		ext2_getattr,
132 	.vop_inactive =		ext2_inactive,
133 	.vop_ioctl =		ext2_ioctl,
134 	.vop_link =		ext2_link,
135 	.vop_lookup =		vfs_cache_lookup,
136 	.vop_mkdir =		ext2_mkdir,
137 	.vop_mknod =		ext2_mknod,
138 	.vop_open =		ext2_open,
139 	.vop_pathconf =		ext2_pathconf,
140 	.vop_poll =		vop_stdpoll,
141 	.vop_print =		ext2_print,
142 	.vop_read =		ext2_read,
143 	.vop_readdir =		ext2_readdir,
144 	.vop_readlink =		ext2_readlink,
145 	.vop_reallocblks =	ext2_reallocblks,
146 	.vop_reclaim =		ext2_reclaim,
147 	.vop_remove =		ext2_remove,
148 	.vop_rename =		ext2_rename,
149 	.vop_rmdir =		ext2_rmdir,
150 	.vop_setattr =		ext2_setattr,
151 	.vop_strategy =		ext2_strategy,
152 	.vop_symlink =		ext2_symlink,
153 	.vop_write =		ext2_write,
154 	.vop_vptofh =		ext2_vptofh,
155 };
156 
157 struct vop_vector ext2_fifoops = {
158 	.vop_default =		&fifo_specops,
159 	.vop_access =		ext2_access,
160 	.vop_close =		ext2fifo_close,
161 	.vop_fsync =		ext2_fsync,
162 	.vop_getattr =		ext2_getattr,
163 	.vop_inactive =		ext2_inactive,
164 	.vop_kqfilter =		ext2fifo_kqfilter,
165 	.vop_print =		ext2_print,
166 	.vop_read =		VOP_PANIC,
167 	.vop_reclaim =		ext2_reclaim,
168 	.vop_setattr =		ext2_setattr,
169 	.vop_write =		VOP_PANIC,
170 	.vop_vptofh =		ext2_vptofh,
171 };
172 
173 /*
174  * A virgin directory (no blushing please).
175  * Note that the type and namlen fields are reversed relative to ext2.
176  * Also, we don't use `struct odirtemplate', since it would just cause
177  * endianness problems.
178  */
179 static struct dirtemplate mastertemplate = {
180 	0, 12, 1, EXT2_FT_DIR, ".",
181 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
182 };
183 static struct dirtemplate omastertemplate = {
184 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
185 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
186 };
187 
188 static void
189 ext2_itimes_locked(struct vnode *vp)
190 {
191 	struct inode *ip;
192 	struct timespec ts;
193 
194 	ASSERT_VI_LOCKED(vp, __func__);
195 
196 	ip = VTOI(vp);
197 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
198 		return;
199 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
200 		ip->i_flag |= IN_LAZYMOD;
201 	else
202 		ip->i_flag |= IN_MODIFIED;
203 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
204 		vfs_timestamp(&ts);
205 		if (ip->i_flag & IN_ACCESS) {
206 			ip->i_atime = ts.tv_sec;
207 			ip->i_atimensec = ts.tv_nsec;
208 		}
209 		if (ip->i_flag & IN_UPDATE) {
210 			ip->i_mtime = ts.tv_sec;
211 			ip->i_mtimensec = ts.tv_nsec;
212 			ip->i_modrev++;
213 		}
214 		if (ip->i_flag & IN_CHANGE) {
215 			ip->i_ctime = ts.tv_sec;
216 			ip->i_ctimensec = ts.tv_nsec;
217 		}
218 	}
219 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
220 }
221 
222 void
223 ext2_itimes(struct vnode *vp)
224 {
225 
226 	VI_LOCK(vp);
227 	ext2_itimes_locked(vp);
228 	VI_UNLOCK(vp);
229 }
230 
231 /*
232  * Create a regular file
233  */
234 static int
235 ext2_create(struct vop_create_args *ap)
236 {
237 	int error;
238 
239 	error =
240 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
241 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
242 	if (error)
243 		return (error);
244 	return (0);
245 }
246 
247 static int
248 ext2_open(struct vop_open_args *ap)
249 {
250 
251 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
252 		return (EOPNOTSUPP);
253 
254 	/*
255 	 * Files marked append-only must be opened for appending.
256 	 */
257 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
258 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
259 		return (EPERM);
260 
261 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
262 
263 	return (0);
264 }
265 
266 /*
267  * Close called.
268  *
269  * Update the times on the inode.
270  */
271 static int
272 ext2_close(struct vop_close_args *ap)
273 {
274 	struct vnode *vp = ap->a_vp;
275 
276 	VI_LOCK(vp);
277 	if (vp->v_usecount > 1)
278 		ext2_itimes_locked(vp);
279 	VI_UNLOCK(vp);
280 	return (0);
281 }
282 
283 static int
284 ext2_access(struct vop_access_args *ap)
285 {
286 	struct vnode *vp = ap->a_vp;
287 	struct inode *ip = VTOI(vp);
288 	accmode_t accmode = ap->a_accmode;
289 	int error;
290 
291 	if (vp->v_type == VBLK || vp->v_type == VCHR)
292 		return (EOPNOTSUPP);
293 
294 	/*
295 	 * Disallow write attempts on read-only file systems;
296 	 * unless the file is a socket, fifo, or a block or
297 	 * character device resident on the file system.
298 	 */
299 	if (accmode & VWRITE) {
300 		switch (vp->v_type) {
301 		case VDIR:
302 		case VLNK:
303 		case VREG:
304 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
305 				return (EROFS);
306 			break;
307 		default:
308 			break;
309 		}
310 	}
311 
312 	/* If immutable bit set, nobody gets to write it. */
313 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
314 		return (EPERM);
315 
316 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
317 	    ap->a_accmode, ap->a_cred, NULL);
318 	return (error);
319 }
320 
321 static int
322 ext2_getattr(struct vop_getattr_args *ap)
323 {
324 	struct vnode *vp = ap->a_vp;
325 	struct inode *ip = VTOI(vp);
326 	struct vattr *vap = ap->a_vap;
327 
328 	ext2_itimes(vp);
329 	/*
330 	 * Copy from inode table
331 	 */
332 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
333 	vap->va_fileid = ip->i_number;
334 	vap->va_mode = ip->i_mode & ~IFMT;
335 	vap->va_nlink = ip->i_nlink;
336 	vap->va_uid = ip->i_uid;
337 	vap->va_gid = ip->i_gid;
338 	vap->va_rdev = ip->i_rdev;
339 	vap->va_size = ip->i_size;
340 	vap->va_atime.tv_sec = ip->i_atime;
341 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
342 	vap->va_mtime.tv_sec = ip->i_mtime;
343 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
344 	vap->va_ctime.tv_sec = ip->i_ctime;
345 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
346 	if E2DI_HAS_XTIME(ip) {
347 		vap->va_birthtime.tv_sec = ip->i_birthtime;
348 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
349 	}
350 	vap->va_flags = ip->i_flags;
351 	vap->va_gen = ip->i_gen;
352 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
353 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
354 	vap->va_type = IFTOVT(ip->i_mode);
355 	vap->va_filerev = ip->i_modrev;
356 	return (0);
357 }
358 
359 /*
360  * Set attribute vnode op. called from several syscalls
361  */
362 static int
363 ext2_setattr(struct vop_setattr_args *ap)
364 {
365 	struct vattr *vap = ap->a_vap;
366 	struct vnode *vp = ap->a_vp;
367 	struct inode *ip = VTOI(vp);
368 	struct ucred *cred = ap->a_cred;
369 	struct thread *td = curthread;
370 	int error;
371 
372 	/*
373 	 * Check for unsettable attributes.
374 	 */
375 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
376 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
377 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
378 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
379 		return (EINVAL);
380 	}
381 	if (vap->va_flags != VNOVAL) {
382 		/* Disallow flags not supported by ext2fs. */
383 		if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
384 			return (EOPNOTSUPP);
385 
386 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
387 			return (EROFS);
388 		/*
389 		 * Callers may only modify the file flags on objects they
390 		 * have VADMIN rights for.
391 		 */
392 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
393 			return (error);
394 		/*
395 		 * Unprivileged processes and privileged processes in
396 		 * jail() are not permitted to unset system flags, or
397 		 * modify flags if any system flags are set.
398 		 * Privileged non-jail processes may not modify system flags
399 		 * if securelevel > 0 and any existing system flags are set.
400 		 */
401 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
402 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
403 				error = securelevel_gt(cred, 0);
404 				if (error)
405 					return (error);
406 			}
407 		} else {
408 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
409 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
410 				return (EPERM);
411 		}
412 		ip->i_flags = vap->va_flags;
413 		ip->i_flag |= IN_CHANGE;
414 		if (ip->i_flags & (IMMUTABLE | APPEND))
415 			return (0);
416 	}
417 	if (ip->i_flags & (IMMUTABLE | APPEND))
418 		return (EPERM);
419 	/*
420 	 * Go through the fields and update iff not VNOVAL.
421 	 */
422 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
423 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
424 			return (EROFS);
425 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
426 		    td)) != 0)
427 			return (error);
428 	}
429 	if (vap->va_size != VNOVAL) {
430 		/*
431 		 * Disallow write attempts on read-only file systems;
432 		 * unless the file is a socket, fifo, or a block or
433 		 * character device resident on the file system.
434 		 */
435 		switch (vp->v_type) {
436 		case VDIR:
437 			return (EISDIR);
438 		case VLNK:
439 		case VREG:
440 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
441 				return (EROFS);
442 			break;
443 		default:
444 			break;
445 		}
446 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
447 			return (error);
448 	}
449 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
450 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
451 			return (EROFS);
452 		/*
453 		 * From utimes(2):
454 		 * If times is NULL, ... The caller must be the owner of
455 		 * the file, have permission to write the file, or be the
456 		 * super-user.
457 		 * If times is non-NULL, ... The caller must be the owner of
458 		 * the file or be the super-user.
459 		 */
460 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
461 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
462 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
463 			return (error);
464 		if (vap->va_atime.tv_sec != VNOVAL)
465 			ip->i_flag |= IN_ACCESS;
466 		if (vap->va_mtime.tv_sec != VNOVAL)
467 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
468 		ext2_itimes(vp);
469 		if (vap->va_atime.tv_sec != VNOVAL) {
470 			ip->i_atime = vap->va_atime.tv_sec;
471 			ip->i_atimensec = vap->va_atime.tv_nsec;
472 		}
473 		if (vap->va_mtime.tv_sec != VNOVAL) {
474 			ip->i_mtime = vap->va_mtime.tv_sec;
475 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
476 		}
477 		ip->i_birthtime = vap->va_birthtime.tv_sec;
478 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
479 		error = ext2_update(vp, 0);
480 		if (error)
481 			return (error);
482 	}
483 	error = 0;
484 	if (vap->va_mode != (mode_t)VNOVAL) {
485 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
486 			return (EROFS);
487 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
488 	}
489 	return (error);
490 }
491 
492 /*
493  * Change the mode on a file.
494  * Inode must be locked before calling.
495  */
496 static int
497 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
498 {
499 	struct inode *ip = VTOI(vp);
500 	int error;
501 
502 	/*
503 	 * To modify the permissions on a file, must possess VADMIN
504 	 * for that file.
505 	 */
506 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
507 		return (error);
508 	/*
509 	 * Privileged processes may set the sticky bit on non-directories,
510 	 * as well as set the setgid bit on a file with a group that the
511 	 * process is not a member of.
512 	 */
513 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
514 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
515 		if (error)
516 			return (EFTYPE);
517 	}
518 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
519 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
520 		if (error)
521 			return (error);
522 	}
523 	ip->i_mode &= ~ALLPERMS;
524 	ip->i_mode |= (mode & ALLPERMS);
525 	ip->i_flag |= IN_CHANGE;
526 	return (0);
527 }
528 
529 /*
530  * Perform chown operation on inode ip;
531  * inode must be locked prior to call.
532  */
533 static int
534 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
535     struct thread *td)
536 {
537 	struct inode *ip = VTOI(vp);
538 	uid_t ouid;
539 	gid_t ogid;
540 	int error = 0;
541 
542 	if (uid == (uid_t)VNOVAL)
543 		uid = ip->i_uid;
544 	if (gid == (gid_t)VNOVAL)
545 		gid = ip->i_gid;
546 	/*
547 	 * To modify the ownership of a file, must possess VADMIN
548 	 * for that file.
549 	 */
550 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
551 		return (error);
552 	/*
553 	 * To change the owner of a file, or change the group of a file
554 	 * to a group of which we are not a member, the caller must
555 	 * have privilege.
556 	 */
557 	if (uid != ip->i_uid || (gid != ip->i_gid &&
558 	    !groupmember(gid, cred))) {
559 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
560 		if (error)
561 			return (error);
562 	}
563 	ogid = ip->i_gid;
564 	ouid = ip->i_uid;
565 	ip->i_gid = gid;
566 	ip->i_uid = uid;
567 	ip->i_flag |= IN_CHANGE;
568 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
569 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
570 			ip->i_mode &= ~(ISUID | ISGID);
571 	}
572 	return (0);
573 }
574 
575 /*
576  * Synch an open file.
577  */
578 /* ARGSUSED */
579 static int
580 ext2_fsync(struct vop_fsync_args *ap)
581 {
582 	/*
583 	 * Flush all dirty buffers associated with a vnode.
584 	 */
585 
586 	vop_stdfsync(ap);
587 
588 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
589 }
590 
591 /*
592  * Mknod vnode call
593  */
594 /* ARGSUSED */
595 static int
596 ext2_mknod(struct vop_mknod_args *ap)
597 {
598 	struct vattr *vap = ap->a_vap;
599 	struct vnode **vpp = ap->a_vpp;
600 	struct inode *ip;
601 	ino_t ino;
602 	int error;
603 
604 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
605 	    ap->a_dvp, vpp, ap->a_cnp);
606 	if (error)
607 		return (error);
608 	ip = VTOI(*vpp);
609 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
610 	if (vap->va_rdev != VNOVAL) {
611 		/*
612 		 * Want to be able to use this to make badblock
613 		 * inodes, so don't truncate the dev number.
614 		 */
615 		ip->i_rdev = vap->va_rdev;
616 	}
617 	/*
618 	 * Remove inode, then reload it through VFS_VGET so it is
619 	 * checked to see if it is an alias of an existing entry in
620 	 * the inode cache.	 XXX I don't believe this is necessary now.
621 	 */
622 	(*vpp)->v_type = VNON;
623 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
624 	vgone(*vpp);
625 	vput(*vpp);
626 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
627 	if (error) {
628 		*vpp = NULL;
629 		return (error);
630 	}
631 	return (0);
632 }
633 
634 static int
635 ext2_remove(struct vop_remove_args *ap)
636 {
637 	struct inode *ip;
638 	struct vnode *vp = ap->a_vp;
639 	struct vnode *dvp = ap->a_dvp;
640 	int error;
641 
642 	ip = VTOI(vp);
643 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
644 	    (VTOI(dvp)->i_flags & APPEND)) {
645 		error = EPERM;
646 		goto out;
647 	}
648 	error = ext2_dirremove(dvp, ap->a_cnp);
649 	if (error == 0) {
650 		ip->i_nlink--;
651 		ip->i_flag |= IN_CHANGE;
652 	}
653 out:
654 	return (error);
655 }
656 
657 /*
658  * link vnode call
659  */
660 static int
661 ext2_link(struct vop_link_args *ap)
662 {
663 	struct vnode *vp = ap->a_vp;
664 	struct vnode *tdvp = ap->a_tdvp;
665 	struct componentname *cnp = ap->a_cnp;
666 	struct inode *ip;
667 	int error;
668 
669 #ifdef INVARIANTS
670 	if ((cnp->cn_flags & HASBUF) == 0)
671 		panic("ext2_link: no name");
672 #endif
673 	ip = VTOI(vp);
674 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
675 		error = EMLINK;
676 		goto out;
677 	}
678 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
679 		error = EPERM;
680 		goto out;
681 	}
682 	ip->i_nlink++;
683 	ip->i_flag |= IN_CHANGE;
684 	error = ext2_update(vp, !DOINGASYNC(vp));
685 	if (!error)
686 		error = ext2_direnter(ip, tdvp, cnp);
687 	if (error) {
688 		ip->i_nlink--;
689 		ip->i_flag |= IN_CHANGE;
690 	}
691 out:
692 	return (error);
693 }
694 
695 /*
696  * Rename system call.
697  * 	rename("foo", "bar");
698  * is essentially
699  *	unlink("bar");
700  *	link("foo", "bar");
701  *	unlink("foo");
702  * but ``atomically''.  Can't do full commit without saving state in the
703  * inode on disk which isn't feasible at this time.  Best we can do is
704  * always guarantee the target exists.
705  *
706  * Basic algorithm is:
707  *
708  * 1) Bump link count on source while we're linking it to the
709  *    target.  This also ensure the inode won't be deleted out
710  *    from underneath us while we work (it may be truncated by
711  *    a concurrent `trunc' or `open' for creation).
712  * 2) Link source to destination.  If destination already exists,
713  *    delete it first.
714  * 3) Unlink source reference to inode if still around. If a
715  *    directory was moved and the parent of the destination
716  *    is different from the source, patch the ".." entry in the
717  *    directory.
718  */
719 static int
720 ext2_rename(struct vop_rename_args *ap)
721 {
722 	struct vnode *tvp = ap->a_tvp;
723 	struct vnode *tdvp = ap->a_tdvp;
724 	struct vnode *fvp = ap->a_fvp;
725 	struct vnode *fdvp = ap->a_fdvp;
726 	struct componentname *tcnp = ap->a_tcnp;
727 	struct componentname *fcnp = ap->a_fcnp;
728 	struct inode *ip, *xp, *dp;
729 	struct dirtemplate dirbuf;
730 	int doingdirectory = 0, oldparent = 0, newparent = 0;
731 	int error = 0;
732 	u_char namlen;
733 
734 #ifdef INVARIANTS
735 	if ((tcnp->cn_flags & HASBUF) == 0 ||
736 	    (fcnp->cn_flags & HASBUF) == 0)
737 		panic("ext2_rename: no name");
738 #endif
739 	/*
740 	 * Check for cross-device rename.
741 	 */
742 	if ((fvp->v_mount != tdvp->v_mount) ||
743 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
744 		error = EXDEV;
745 abortit:
746 		if (tdvp == tvp)
747 			vrele(tdvp);
748 		else
749 			vput(tdvp);
750 		if (tvp)
751 			vput(tvp);
752 		vrele(fdvp);
753 		vrele(fvp);
754 		return (error);
755 	}
756 
757 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
758 	    (VTOI(tdvp)->i_flags & APPEND))) {
759 		error = EPERM;
760 		goto abortit;
761 	}
762 
763 	/*
764 	 * Renaming a file to itself has no effect.  The upper layers should
765 	 * not call us in that case.  Temporarily just warn if they do.
766 	 */
767 	if (fvp == tvp) {
768 		printf("ext2_rename: fvp == tvp (can't happen)\n");
769 		error = 0;
770 		goto abortit;
771 	}
772 
773 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
774 		goto abortit;
775 	dp = VTOI(fdvp);
776 	ip = VTOI(fvp);
777 	if (ip->i_nlink >= EXT2_LINK_MAX) {
778 		VOP_UNLOCK(fvp, 0);
779 		error = EMLINK;
780 		goto abortit;
781 	}
782 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
783 	    || (dp->i_flags & APPEND)) {
784 		VOP_UNLOCK(fvp, 0);
785 		error = EPERM;
786 		goto abortit;
787 	}
788 	if ((ip->i_mode & IFMT) == IFDIR) {
789 		/*
790 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
791 		 */
792 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
793 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
794 		    (ip->i_flag & IN_RENAME)) {
795 			VOP_UNLOCK(fvp, 0);
796 			error = EINVAL;
797 			goto abortit;
798 		}
799 		ip->i_flag |= IN_RENAME;
800 		oldparent = dp->i_number;
801 		doingdirectory++;
802 	}
803 	vrele(fdvp);
804 
805 	/*
806 	 * When the target exists, both the directory
807 	 * and target vnodes are returned locked.
808 	 */
809 	dp = VTOI(tdvp);
810 	xp = NULL;
811 	if (tvp)
812 		xp = VTOI(tvp);
813 
814 	/*
815 	 * 1) Bump link count while we're moving stuff
816 	 *    around.  If we crash somewhere before
817 	 *    completing our work, the link count
818 	 *    may be wrong, but correctable.
819 	 */
820 	ip->i_nlink++;
821 	ip->i_flag |= IN_CHANGE;
822 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
823 		VOP_UNLOCK(fvp, 0);
824 		goto bad;
825 	}
826 
827 	/*
828 	 * If ".." must be changed (ie the directory gets a new
829 	 * parent) then the source directory must not be in the
830 	 * directory hierarchy above the target, as this would
831 	 * orphan everything below the source directory. Also
832 	 * the user must have write permission in the source so
833 	 * as to be able to change "..". We must repeat the call
834 	 * to namei, as the parent directory is unlocked by the
835 	 * call to checkpath().
836 	 */
837 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
838 	VOP_UNLOCK(fvp, 0);
839 	if (oldparent != dp->i_number)
840 		newparent = dp->i_number;
841 	if (doingdirectory && newparent) {
842 		if (error)	/* write access check above */
843 			goto bad;
844 		if (xp != NULL)
845 			vput(tvp);
846 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
847 		if (error)
848 			goto out;
849 		VREF(tdvp);
850 		error = relookup(tdvp, &tvp, tcnp);
851 		if (error)
852 			goto out;
853 		vrele(tdvp);
854 		dp = VTOI(tdvp);
855 		xp = NULL;
856 		if (tvp)
857 			xp = VTOI(tvp);
858 	}
859 	/*
860 	 * 2) If target doesn't exist, link the target
861 	 *    to the source and unlink the source.
862 	 *    Otherwise, rewrite the target directory
863 	 *    entry to reference the source inode and
864 	 *    expunge the original entry's existence.
865 	 */
866 	if (xp == NULL) {
867 		if (dp->i_devvp != ip->i_devvp)
868 			panic("ext2_rename: EXDEV");
869 		/*
870 		 * Account for ".." in new directory.
871 		 * When source and destination have the same
872 		 * parent we don't fool with the link count.
873 		 */
874 		if (doingdirectory && newparent) {
875 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
876 				error = EMLINK;
877 				goto bad;
878 			}
879 			dp->i_nlink++;
880 			dp->i_flag |= IN_CHANGE;
881 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
882 			if (error)
883 				goto bad;
884 		}
885 		error = ext2_direnter(ip, tdvp, tcnp);
886 		if (error) {
887 			if (doingdirectory && newparent) {
888 				dp->i_nlink--;
889 				dp->i_flag |= IN_CHANGE;
890 				(void)ext2_update(tdvp, 1);
891 			}
892 			goto bad;
893 		}
894 		vput(tdvp);
895 	} else {
896 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
897 		       panic("ext2_rename: EXDEV");
898 		/*
899 		 * Short circuit rename(foo, foo).
900 		 */
901 		if (xp->i_number == ip->i_number)
902 			panic("ext2_rename: same file");
903 		/*
904 		 * If the parent directory is "sticky", then the user must
905 		 * own the parent directory, or the destination of the rename,
906 		 * otherwise the destination may not be changed (except by
907 		 * root). This implements append-only directories.
908 		 */
909 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
910 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
911 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
912 			error = EPERM;
913 			goto bad;
914 		}
915 		/*
916 		 * Target must be empty if a directory and have no links
917 		 * to it. Also, ensure source and target are compatible
918 		 * (both directories, or both not directories).
919 		 */
920 		if ((xp->i_mode&IFMT) == IFDIR) {
921 			if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
922 			    xp->i_nlink > 2) {
923 				error = ENOTEMPTY;
924 				goto bad;
925 			}
926 			if (!doingdirectory) {
927 				error = ENOTDIR;
928 				goto bad;
929 			}
930 			cache_purge(tdvp);
931 		} else if (doingdirectory) {
932 			error = EISDIR;
933 			goto bad;
934 		}
935 		error = ext2_dirrewrite(dp, ip, tcnp);
936 		if (error)
937 			goto bad;
938 		/*
939 		 * If the target directory is in the same
940 		 * directory as the source directory,
941 		 * decrement the link count on the parent
942 		 * of the target directory.
943 		 */
944 		if (doingdirectory && !newparent) {
945 			dp->i_nlink--;
946 			dp->i_flag |= IN_CHANGE;
947 		}
948 		vput(tdvp);
949 		/*
950 		 * Adjust the link count of the target to
951 		 * reflect the dirrewrite above.  If this is
952 		 * a directory it is empty and there are
953 		 * no links to it, so we can squash the inode and
954 		 * any space associated with it.  We disallowed
955 		 * renaming over top of a directory with links to
956 		 * it above, as the remaining link would point to
957 		 * a directory without "." or ".." entries.
958 		 */
959 		xp->i_nlink--;
960 		if (doingdirectory) {
961 			if (--xp->i_nlink != 0)
962 				panic("ext2_rename: linked directory");
963 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
964 			    tcnp->cn_cred, tcnp->cn_thread);
965 		}
966 		xp->i_flag |= IN_CHANGE;
967 		vput(tvp);
968 		xp = NULL;
969 	}
970 
971 	/*
972 	 * 3) Unlink the source.
973 	 */
974 	fcnp->cn_flags &= ~MODMASK;
975 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
976 	VREF(fdvp);
977 	error = relookup(fdvp, &fvp, fcnp);
978 	if (error == 0)
979 		vrele(fdvp);
980 	if (fvp != NULL) {
981 		xp = VTOI(fvp);
982 		dp = VTOI(fdvp);
983 	} else {
984 		/*
985 		 * From name has disappeared.
986 		 */
987 		if (doingdirectory)
988 			panic("ext2_rename: lost dir entry");
989 		vrele(ap->a_fvp);
990 		return (0);
991 	}
992 	/*
993 	 * Ensure that the directory entry still exists and has not
994 	 * changed while the new name has been entered. If the source is
995 	 * a file then the entry may have been unlinked or renamed. In
996 	 * either case there is no further work to be done. If the source
997 	 * is a directory then it cannot have been rmdir'ed; its link
998 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
999 	 * The IN_RENAME flag ensures that it cannot be moved by another
1000 	 * rename.
1001 	 */
1002 	if (xp != ip) {
1003 		if (doingdirectory)
1004 			panic("ext2_rename: lost dir entry");
1005 	} else {
1006 		/*
1007 		 * If the source is a directory with a
1008 		 * new parent, the link count of the old
1009 		 * parent directory must be decremented
1010 		 * and ".." set to point to the new parent.
1011 		 */
1012 		if (doingdirectory && newparent) {
1013 			dp->i_nlink--;
1014 			dp->i_flag |= IN_CHANGE;
1015 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1016 				sizeof(struct dirtemplate), (off_t)0,
1017 				UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1018 				tcnp->cn_cred, NOCRED, NULL, NULL);
1019 			if (error == 0) {
1020 				/* Like ufs little-endian: */
1021 				namlen = dirbuf.dotdot_type;
1022 				if (namlen != 2 ||
1023 				    dirbuf.dotdot_name[0] != '.' ||
1024 				    dirbuf.dotdot_name[1] != '.') {
1025 					ext2_dirbad(xp, (doff_t)12,
1026 					    "rename: mangled dir");
1027 				} else {
1028 					dirbuf.dotdot_ino = newparent;
1029 					(void) vn_rdwr(UIO_WRITE, fvp,
1030 					    (caddr_t)&dirbuf,
1031 					    sizeof(struct dirtemplate),
1032 					    (off_t)0, UIO_SYSSPACE,
1033 					    IO_NODELOCKED | IO_SYNC |
1034 					    IO_NOMACCHECK, tcnp->cn_cred,
1035 					    NOCRED, NULL, NULL);
1036 					cache_purge(fdvp);
1037 				}
1038 			}
1039 		}
1040 		error = ext2_dirremove(fdvp, fcnp);
1041 		if (!error) {
1042 			xp->i_nlink--;
1043 			xp->i_flag |= IN_CHANGE;
1044 		}
1045 		xp->i_flag &= ~IN_RENAME;
1046 	}
1047 	if (dp)
1048 		vput(fdvp);
1049 	if (xp)
1050 		vput(fvp);
1051 	vrele(ap->a_fvp);
1052 	return (error);
1053 
1054 bad:
1055 	if (xp)
1056 		vput(ITOV(xp));
1057 	vput(ITOV(dp));
1058 out:
1059 	if (doingdirectory)
1060 		ip->i_flag &= ~IN_RENAME;
1061 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1062 		ip->i_nlink--;
1063 		ip->i_flag |= IN_CHANGE;
1064 		ip->i_flag &= ~IN_RENAME;
1065 		vput(fvp);
1066 	} else
1067 		vrele(fvp);
1068 	return (error);
1069 }
1070 
1071 /*
1072  * Mkdir system call
1073  */
1074 static int
1075 ext2_mkdir(struct vop_mkdir_args *ap)
1076 {
1077 	struct vnode *dvp = ap->a_dvp;
1078 	struct vattr *vap = ap->a_vap;
1079 	struct componentname *cnp = ap->a_cnp;
1080 	struct inode *ip, *dp;
1081 	struct vnode *tvp;
1082 	struct dirtemplate dirtemplate, *dtp;
1083 	int error, dmode;
1084 
1085 #ifdef INVARIANTS
1086 	if ((cnp->cn_flags & HASBUF) == 0)
1087 		panic("ext2_mkdir: no name");
1088 #endif
1089 	dp = VTOI(dvp);
1090 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1091 		error = EMLINK;
1092 		goto out;
1093 	}
1094 	dmode = vap->va_mode & 0777;
1095 	dmode |= IFDIR;
1096 	/*
1097 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1098 	 * but not have it entered in the parent directory. The entry is
1099 	 * made later after writing "." and ".." entries.
1100 	 */
1101 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1102 	if (error)
1103 		goto out;
1104 	ip = VTOI(tvp);
1105 	ip->i_gid = dp->i_gid;
1106 #ifdef SUIDDIR
1107 	{
1108 		/*
1109 		 * if we are hacking owners here, (only do this where told to)
1110 		 * and we are not giving it TOO root, (would subvert quotas)
1111 		 * then go ahead and give it to the other user.
1112 		 * The new directory also inherits the SUID bit.
1113 		 * If user's UID and dir UID are the same,
1114 		 * 'give it away' so that the SUID is still forced on.
1115 		 */
1116 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1117 		   (dp->i_mode & ISUID) && dp->i_uid) {
1118 			dmode |= ISUID;
1119 			ip->i_uid = dp->i_uid;
1120 		} else {
1121 			ip->i_uid = cnp->cn_cred->cr_uid;
1122 		}
1123 	}
1124 #else
1125 	ip->i_uid = cnp->cn_cred->cr_uid;
1126 #endif
1127 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1128 	ip->i_mode = dmode;
1129 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1130 	ip->i_nlink = 2;
1131 	if (cnp->cn_flags & ISWHITEOUT)
1132 		ip->i_flags |= UF_OPAQUE;
1133 	error = ext2_update(tvp, 1);
1134 
1135 	/*
1136 	 * Bump link count in parent directory
1137 	 * to reflect work done below.  Should
1138 	 * be done before reference is created
1139 	 * so reparation is possible if we crash.
1140 	 */
1141 	dp->i_nlink++;
1142 	dp->i_flag |= IN_CHANGE;
1143 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1144 	if (error)
1145 		goto bad;
1146 
1147 	/* Initialize directory with "." and ".." from static template. */
1148 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1149 	    EXT2F_INCOMPAT_FTYPE))
1150 		dtp = &mastertemplate;
1151 	else
1152 		dtp = &omastertemplate;
1153 	dirtemplate = *dtp;
1154 	dirtemplate.dot_ino = ip->i_number;
1155 	dirtemplate.dotdot_ino = dp->i_number;
1156 	/* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1157 	 * so let's just redefine it - for this function only
1158 	 */
1159 #undef  DIRBLKSIZ
1160 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1161 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1162 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1163 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1164 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1165 	    NULL, NULL);
1166 	if (error) {
1167 		dp->i_nlink--;
1168 		dp->i_flag |= IN_CHANGE;
1169 		goto bad;
1170 	}
1171 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1172 		/* XXX should grow with balloc() */
1173 		panic("ext2_mkdir: blksize");
1174 	else {
1175 		ip->i_size = DIRBLKSIZ;
1176 		ip->i_flag |= IN_CHANGE;
1177 	}
1178 
1179 	/* Directory set up, now install its entry in the parent directory. */
1180 	error = ext2_direnter(ip, dvp, cnp);
1181 	if (error) {
1182 		dp->i_nlink--;
1183 		dp->i_flag |= IN_CHANGE;
1184 	}
1185 bad:
1186 	/*
1187 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1188 	 * for us because we set the link count to 0.
1189 	 */
1190 	if (error) {
1191 		ip->i_nlink = 0;
1192 		ip->i_flag |= IN_CHANGE;
1193 		vput(tvp);
1194 	} else
1195 		*ap->a_vpp = tvp;
1196 out:
1197 	return (error);
1198 #undef  DIRBLKSIZ
1199 #define DIRBLKSIZ  DEV_BSIZE
1200 }
1201 
1202 /*
1203  * Rmdir system call.
1204  */
1205 static int
1206 ext2_rmdir(struct vop_rmdir_args *ap)
1207 {
1208 	struct vnode *vp = ap->a_vp;
1209 	struct vnode *dvp = ap->a_dvp;
1210 	struct componentname *cnp = ap->a_cnp;
1211 	struct inode *ip, *dp;
1212 	int error;
1213 
1214 	ip = VTOI(vp);
1215 	dp = VTOI(dvp);
1216 
1217 	/*
1218 	 * Verify the directory is empty (and valid).
1219 	 * (Rmdir ".." won't be valid since
1220 	 *  ".." will contain a reference to
1221 	 *  the current directory and thus be
1222 	 *  non-empty.)
1223 	 */
1224 	error = 0;
1225 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1226 		error = ENOTEMPTY;
1227 		goto out;
1228 	}
1229 	if ((dp->i_flags & APPEND)
1230 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1231 		error = EPERM;
1232 		goto out;
1233 	}
1234 	/*
1235 	 * Delete reference to directory before purging
1236 	 * inode.  If we crash in between, the directory
1237 	 * will be reattached to lost+found,
1238 	 */
1239 	error = ext2_dirremove(dvp, cnp);
1240 	if (error)
1241 		goto out;
1242 	dp->i_nlink--;
1243 	dp->i_flag |= IN_CHANGE;
1244 	cache_purge(dvp);
1245 	VOP_UNLOCK(dvp, 0);
1246 	/*
1247 	 * Truncate inode.  The only stuff left
1248 	 * in the directory is "." and "..".  The
1249 	 * "." reference is inconsequential since
1250 	 * we're quashing it.  The ".." reference
1251 	 * has already been adjusted above.  We've
1252 	 * removed the "." reference and the reference
1253 	 * in the parent directory, but there may be
1254 	 * other hard links so decrement by 2 and
1255 	 * worry about them later.
1256 	 */
1257 	ip->i_nlink -= 2;
1258 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1259 	    cnp->cn_thread);
1260 	cache_purge(ITOV(ip));
1261 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1262 		VOP_UNLOCK(vp, 0);
1263 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1264 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1265 	}
1266 out:
1267 	return (error);
1268 }
1269 
1270 /*
1271  * symlink -- make a symbolic link
1272  */
1273 static int
1274 ext2_symlink(struct vop_symlink_args *ap)
1275 {
1276 	struct vnode *vp, **vpp = ap->a_vpp;
1277 	struct inode *ip;
1278 	int len, error;
1279 
1280 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1281 	    vpp, ap->a_cnp);
1282 	if (error)
1283 		return (error);
1284 	vp = *vpp;
1285 	len = strlen(ap->a_target);
1286 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1287 		ip = VTOI(vp);
1288 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1289 		ip->i_size = len;
1290 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1291 	} else
1292 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1293 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1294 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1295 	if (error)
1296 		vput(vp);
1297 	return (error);
1298 }
1299 
1300 /*
1301  * Return target name of a symbolic link
1302  */
1303 static int
1304 ext2_readlink(struct vop_readlink_args *ap)
1305 {
1306 	struct vnode *vp = ap->a_vp;
1307 	struct inode *ip = VTOI(vp);
1308 	int isize;
1309 
1310 	isize = ip->i_size;
1311 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1312 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1313 		return (0);
1314 	}
1315 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1316 }
1317 
1318 /*
1319  * Calculate the logical to physical mapping if not done already,
1320  * then call the device strategy routine.
1321  *
1322  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1323  * deadlock on memory.  See ext2_bmap() for details.
1324  */
1325 static int
1326 ext2_strategy(struct vop_strategy_args *ap)
1327 {
1328 	struct buf *bp = ap->a_bp;
1329 	struct vnode *vp = ap->a_vp;
1330 	struct inode *ip;
1331 	struct bufobj *bo;
1332 	daddr_t blkno;
1333 	int error;
1334 
1335 	ip = VTOI(vp);
1336 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1337 		panic("ext2_strategy: spec");
1338 	if (bp->b_blkno == bp->b_lblkno) {
1339 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1340 		bp->b_blkno = blkno;
1341 		if (error) {
1342 			bp->b_error = error;
1343 			bp->b_ioflags |= BIO_ERROR;
1344 			bufdone(bp);
1345 			return (0);
1346 		}
1347 		if ((long)bp->b_blkno == -1)
1348 			vfs_bio_clrbuf(bp);
1349 	}
1350 	if ((long)bp->b_blkno == -1) {
1351 		bufdone(bp);
1352 		return (0);
1353 	}
1354 	bp->b_iooffset = dbtob(bp->b_blkno);
1355 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1356 	BO_STRATEGY(bo, bp);
1357 	return (0);
1358 }
1359 
1360 /*
1361  * Print out the contents of an inode.
1362  */
1363 static int
1364 ext2_print(struct vop_print_args *ap)
1365 {
1366 	struct vnode *vp = ap->a_vp;
1367 	struct inode *ip = VTOI(vp);
1368 
1369 	vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1370 	if (vp->v_type == VFIFO)
1371 		fifo_printinfo(vp);
1372 	printf("\n");
1373 	return (0);
1374 }
1375 
1376 /*
1377  * Close wrapper for fifos.
1378  *
1379  * Update the times on the inode then do device close.
1380  */
1381 static int
1382 ext2fifo_close(struct vop_close_args *ap)
1383 {
1384 	struct vnode *vp = ap->a_vp;
1385 
1386 	VI_LOCK(vp);
1387 	if (vp->v_usecount > 1)
1388 		ext2_itimes_locked(vp);
1389 	VI_UNLOCK(vp);
1390 	return (fifo_specops.vop_close(ap));
1391 }
1392 
1393 /*
1394  * Kqfilter wrapper for fifos.
1395  *
1396  * Fall through to ext2 kqfilter routines if needed
1397  */
1398 static int
1399 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1400 {
1401 	int error;
1402 
1403 	error = fifo_specops.vop_kqfilter(ap);
1404 	if (error)
1405 		error = vfs_kqfilter(ap);
1406 	return (error);
1407 }
1408 
1409 /*
1410  * Return POSIX pathconf information applicable to ext2 filesystems.
1411  */
1412 static int
1413 ext2_pathconf(struct vop_pathconf_args *ap)
1414 {
1415 	int error = 0;
1416 
1417 	switch (ap->a_name) {
1418 	case _PC_LINK_MAX:
1419 		*ap->a_retval = EXT2_LINK_MAX;
1420 		break;
1421 	case _PC_NAME_MAX:
1422 		*ap->a_retval = NAME_MAX;
1423 		break;
1424 	case _PC_PATH_MAX:
1425 		*ap->a_retval = PATH_MAX;
1426 		break;
1427 	case _PC_PIPE_BUF:
1428 		*ap->a_retval = PIPE_BUF;
1429 		break;
1430 	case _PC_CHOWN_RESTRICTED:
1431 		*ap->a_retval = 1;
1432 		break;
1433 	case _PC_NO_TRUNC:
1434 		*ap->a_retval = 1;
1435 		break;
1436 	case _PC_MIN_HOLE_SIZE:
1437 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1438 		break;
1439 	case _PC_ASYNC_IO:
1440 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1441 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1442 		error = EINVAL;
1443 		break;
1444 	case _PC_PRIO_IO:
1445 		*ap->a_retval = 0;
1446 		break;
1447 	case _PC_SYNC_IO:
1448 		*ap->a_retval = 0;
1449 		break;
1450 	case _PC_ALLOC_SIZE_MIN:
1451 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1452 		break;
1453 	case _PC_FILESIZEBITS:
1454 		*ap->a_retval = 64;
1455 		break;
1456 	case _PC_REC_INCR_XFER_SIZE:
1457 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1458 		break;
1459 	case _PC_REC_MAX_XFER_SIZE:
1460 		*ap->a_retval = -1; /* means ``unlimited'' */
1461 		break;
1462 	case _PC_REC_MIN_XFER_SIZE:
1463 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1464 		break;
1465 	case _PC_REC_XFER_ALIGN:
1466 		*ap->a_retval = PAGE_SIZE;
1467 		break;
1468 	case _PC_SYMLINK_MAX:
1469 		*ap->a_retval = MAXPATHLEN;
1470 		break;
1471 
1472 	default:
1473 		error = EINVAL;
1474 		break;
1475 	}
1476 	return (error);
1477 }
1478 
1479 /*
1480  * Vnode pointer to File handle
1481  */
1482 /* ARGSUSED */
1483 static int
1484 ext2_vptofh(struct vop_vptofh_args *ap)
1485 {
1486 	struct inode *ip;
1487 	struct ufid *ufhp;
1488 
1489 	ip = VTOI(ap->a_vp);
1490 	ufhp = (struct ufid *)ap->a_fhp;
1491 	ufhp->ufid_len = sizeof(struct ufid);
1492 	ufhp->ufid_ino = ip->i_number;
1493 	ufhp->ufid_gen = ip->i_gen;
1494 	return (0);
1495 }
1496 
1497 /*
1498  * Initialize the vnode associated with a new inode, handle aliased
1499  * vnodes.
1500  */
1501 int
1502 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1503 {
1504 	struct inode *ip;
1505 	struct vnode *vp;
1506 
1507 	vp = *vpp;
1508 	ip = VTOI(vp);
1509 	vp->v_type = IFTOVT(ip->i_mode);
1510 	if (vp->v_type == VFIFO)
1511 		vp->v_op = fifoops;
1512 
1513 	if (ip->i_number == EXT2_ROOTINO)
1514 		vp->v_vflag |= VV_ROOT;
1515 	ip->i_modrev = init_va_filerev();
1516 	*vpp = vp;
1517 	return (0);
1518 }
1519 
1520 /*
1521  * Allocate a new inode.
1522  */
1523 static int
1524 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1525     struct componentname *cnp)
1526 {
1527 	struct inode *ip, *pdir;
1528 	struct vnode *tvp;
1529 	int error;
1530 
1531 	pdir = VTOI(dvp);
1532 #ifdef INVARIANTS
1533 	if ((cnp->cn_flags & HASBUF) == 0)
1534 		panic("ext2_makeinode: no name");
1535 #endif
1536 	*vpp = NULL;
1537 	if ((mode & IFMT) == 0)
1538 		mode |= IFREG;
1539 
1540 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1541 	if (error) {
1542 		return (error);
1543 	}
1544 	ip = VTOI(tvp);
1545 	ip->i_gid = pdir->i_gid;
1546 #ifdef SUIDDIR
1547 	{
1548 		/*
1549 		 * if we are
1550 		 * not the owner of the directory,
1551 		 * and we are hacking owners here, (only do this where told to)
1552 		 * and we are not giving it TOO root, (would subvert quotas)
1553 		 * then go ahead and give it to the other user.
1554 		 * Note that this drops off the execute bits for security.
1555 		 */
1556 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1557 		     (pdir->i_mode & ISUID) &&
1558 		     (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1559 			ip->i_uid = pdir->i_uid;
1560 			mode &= ~07111;
1561 		} else {
1562 			ip->i_uid = cnp->cn_cred->cr_uid;
1563 		}
1564 	}
1565 #else
1566 	ip->i_uid = cnp->cn_cred->cr_uid;
1567 #endif
1568 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1569 	ip->i_mode = mode;
1570 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1571 	ip->i_nlink = 1;
1572 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1573 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1574 			ip->i_mode &= ~ISGID;
1575 	}
1576 
1577 	if (cnp->cn_flags & ISWHITEOUT)
1578 		ip->i_flags |= UF_OPAQUE;
1579 
1580 	/*
1581 	 * Make sure inode goes to disk before directory entry.
1582 	 */
1583 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1584 	if (error)
1585 		goto bad;
1586 	error = ext2_direnter(ip, dvp, cnp);
1587 	if (error)
1588 		goto bad;
1589 
1590 	*vpp = tvp;
1591 	return (0);
1592 
1593 bad:
1594 	/*
1595 	 * Write error occurred trying to update the inode
1596 	 * or the directory so must deallocate the inode.
1597 	 */
1598 	ip->i_nlink = 0;
1599 	ip->i_flag |= IN_CHANGE;
1600 	vput(tvp);
1601 	return (error);
1602 }
1603 
1604 /*
1605  * Vnode op for reading.
1606  */
1607 static int
1608 ext2_read(struct vop_read_args *ap)
1609 {
1610 	struct vnode *vp;
1611 	struct inode *ip;
1612 	int error;
1613 
1614 	vp = ap->a_vp;
1615 	ip = VTOI(vp);
1616 
1617 	/*EXT4_EXT_LOCK(ip);*/
1618 	if (ip->i_flag & IN_E4EXTENTS)
1619 		error = ext4_ext_read(ap);
1620 	else
1621 		error = ext2_ind_read(ap);
1622 	/*EXT4_EXT_UNLOCK(ip);*/
1623 	return (error);
1624 }
1625 
1626 /*
1627  * Vnode op for reading.
1628  */
1629 static int
1630 ext2_ind_read(struct vop_read_args *ap)
1631 {
1632 	struct vnode *vp;
1633 	struct inode *ip;
1634 	struct uio *uio;
1635 	struct m_ext2fs *fs;
1636 	struct buf *bp;
1637 	daddr_t lbn, nextlbn;
1638 	off_t bytesinfile;
1639 	long size, xfersize, blkoffset;
1640 	int error, orig_resid, seqcount;
1641 	int ioflag;
1642 
1643 	vp = ap->a_vp;
1644 	uio = ap->a_uio;
1645 	ioflag = ap->a_ioflag;
1646 
1647 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1648 	ip = VTOI(vp);
1649 
1650 #ifdef INVARIANTS
1651 	if (uio->uio_rw != UIO_READ)
1652 		panic("%s: mode", "ext2_read");
1653 
1654 	if (vp->v_type == VLNK) {
1655 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1656 			panic("%s: short symlink", "ext2_read");
1657 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1658 		panic("%s: type %d", "ext2_read", vp->v_type);
1659 #endif
1660 	orig_resid = uio->uio_resid;
1661 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1662 	if (orig_resid == 0)
1663 		return (0);
1664 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1665 	fs = ip->i_e2fs;
1666 	if (uio->uio_offset < ip->i_size &&
1667 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1668 	    	return (EOVERFLOW);
1669 
1670 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1671 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1672 			break;
1673 		lbn = lblkno(fs, uio->uio_offset);
1674 		nextlbn = lbn + 1;
1675 		size = blksize(fs, ip, lbn);
1676 		blkoffset = blkoff(fs, uio->uio_offset);
1677 
1678 		xfersize = fs->e2fs_fsize - blkoffset;
1679 		if (uio->uio_resid < xfersize)
1680 			xfersize = uio->uio_resid;
1681 		if (bytesinfile < xfersize)
1682 			xfersize = bytesinfile;
1683 
1684 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1685 			error = bread(vp, lbn, size, NOCRED, &bp);
1686 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1687 			error = cluster_read(vp, ip->i_size, lbn, size,
1688 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1689 			    0, &bp);
1690 		} else if (seqcount > 1) {
1691 			u_int nextsize = blksize(fs, ip, nextlbn);
1692 			error = breadn(vp, lbn,
1693 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1694 		} else
1695 			error = bread(vp, lbn, size, NOCRED, &bp);
1696 		if (error) {
1697 			brelse(bp);
1698 			bp = NULL;
1699 			break;
1700 		}
1701 
1702 		/*
1703 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
1704 		 * will cause us to attempt to release the buffer later on
1705 		 * and will cause the buffer cache to attempt to free the
1706 		 * underlying pages.
1707 		 */
1708 		if (ioflag & IO_DIRECT)
1709 			bp->b_flags |= B_DIRECT;
1710 
1711 		/*
1712 		 * We should only get non-zero b_resid when an I/O error
1713 		 * has occurred, which should cause us to break above.
1714 		 * However, if the short read did not cause an error,
1715 		 * then we want to ensure that we do not uiomove bad
1716 		 * or uninitialized data.
1717 		 */
1718 		size -= bp->b_resid;
1719 		if (size < xfersize) {
1720 			if (size == 0)
1721 				break;
1722 			xfersize = size;
1723 		}
1724 		error = uiomove((char *)bp->b_data + blkoffset,
1725 			(int)xfersize, uio);
1726 		if (error)
1727 			break;
1728 
1729 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1730 			/*
1731 			 * If it's VMIO or direct I/O, then we don't
1732 			 * need the buf, mark it available for
1733 			 * freeing. If it's non-direct VMIO, the VM has
1734 			 * the data.
1735 			 */
1736 			bp->b_flags |= B_RELBUF;
1737 			brelse(bp);
1738 		} else {
1739 			/*
1740 			 * Otherwise let whoever
1741 			 * made the request take care of
1742 			 * freeing it. We just queue
1743 			 * it onto another list.
1744 			 */
1745 			bqrelse(bp);
1746 		}
1747 	}
1748 
1749 	/*
1750 	 * This can only happen in the case of an error
1751 	 * because the loop above resets bp to NULL on each iteration
1752 	 * and on normal completion has not set a new value into it.
1753 	 * so it must have come from a 'break' statement
1754 	 */
1755 	if (bp != NULL) {
1756 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1757 			bp->b_flags |= B_RELBUF;
1758 			brelse(bp);
1759 		} else {
1760 			bqrelse(bp);
1761 		}
1762 	}
1763 
1764 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1765 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
1766 		ip->i_flag |= IN_ACCESS;
1767 	return (error);
1768 }
1769 
1770 static int
1771 ext2_ioctl(struct vop_ioctl_args *ap)
1772 {
1773 
1774 	switch (ap->a_command) {
1775 	case FIOSEEKDATA:
1776 	case FIOSEEKHOLE:
1777 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1778 		    (off_t *)ap->a_data, ap->a_cred));
1779 	default:
1780 		return (ENOTTY);
1781 	}
1782 }
1783 
1784 /*
1785  * this function handles ext4 extents block mapping
1786  */
1787 static int
1788 ext4_ext_read(struct vop_read_args *ap)
1789 {
1790 	struct vnode *vp;
1791 	struct inode *ip;
1792 	struct uio *uio;
1793 	struct m_ext2fs *fs;
1794 	struct buf *bp;
1795 	struct ext4_extent nex, *ep;
1796 	struct ext4_extent_path path;
1797 	daddr_t lbn, newblk;
1798 	off_t bytesinfile;
1799 	int cache_type;
1800 	ssize_t orig_resid;
1801 	int error;
1802 	long size, xfersize, blkoffset;
1803 
1804 	vp = ap->a_vp;
1805 	ip = VTOI(vp);
1806 	uio = ap->a_uio;
1807 	memset(&path, 0, sizeof(path));
1808 
1809 	orig_resid = uio->uio_resid;
1810 	KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1811 	if (orig_resid == 0)
1812 		return (0);
1813 	KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1814 	fs = ip->i_e2fs;
1815 	if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1816 		return (EOVERFLOW);
1817 
1818 	while (uio->uio_resid > 0) {
1819 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1820 			break;
1821 		lbn = lblkno(fs, uio->uio_offset);
1822 		size = blksize(fs, ip, lbn);
1823 		blkoffset = blkoff(fs, uio->uio_offset);
1824 
1825 		xfersize = fs->e2fs_fsize - blkoffset;
1826 		xfersize = MIN(xfersize, uio->uio_resid);
1827 		xfersize = MIN(xfersize, bytesinfile);
1828 
1829 		/* get block from ext4 extent cache */
1830 		cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1831 		switch (cache_type) {
1832 		case EXT4_EXT_CACHE_NO:
1833 			ext4_ext_find_extent(fs, ip, lbn, &path);
1834 			ep = path.ep_ext;
1835 			if (ep == NULL)
1836 				return (EIO);
1837 
1838 			ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN);
1839 
1840 			newblk = lbn - ep->e_blk + (ep->e_start_lo |
1841 			    (daddr_t)ep->e_start_hi << 32);
1842 
1843 			if (path.ep_bp != NULL) {
1844 				brelse(path.ep_bp);
1845 				path.ep_bp = NULL;
1846 			}
1847 			break;
1848 
1849 		case EXT4_EXT_CACHE_GAP:
1850 			/* block has not been allocated yet */
1851 			return (0);
1852 
1853 		case EXT4_EXT_CACHE_IN:
1854 			newblk = lbn - nex.e_blk + (nex.e_start_lo |
1855 			    (daddr_t)nex.e_start_hi << 32);
1856 			break;
1857 
1858 		default:
1859 			panic("%s: invalid cache type", __func__);
1860 		}
1861 
1862 		error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp);
1863 		if (error) {
1864 			brelse(bp);
1865 			return (error);
1866 		}
1867 
1868 		size -= bp->b_resid;
1869 		if (size < xfersize) {
1870 			if (size == 0) {
1871 				bqrelse(bp);
1872 				break;
1873 			}
1874 			xfersize = size;
1875 		}
1876 		error = uiomove(bp->b_data + blkoffset, (int)xfersize, uio);
1877 		bqrelse(bp);
1878 		if (error)
1879 			return (error);
1880 	}
1881 
1882 	return (0);
1883 }
1884 
1885 /*
1886  * Vnode op for writing.
1887  */
1888 static int
1889 ext2_write(struct vop_write_args *ap)
1890 {
1891 	struct vnode *vp;
1892 	struct uio *uio;
1893 	struct inode *ip;
1894 	struct m_ext2fs *fs;
1895 	struct buf *bp;
1896 	daddr_t lbn;
1897 	off_t osize;
1898 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1899 
1900 	ioflag = ap->a_ioflag;
1901 	uio = ap->a_uio;
1902 	vp = ap->a_vp;
1903 
1904 	seqcount = ioflag >> IO_SEQSHIFT;
1905 	ip = VTOI(vp);
1906 
1907 #ifdef INVARIANTS
1908 	if (uio->uio_rw != UIO_WRITE)
1909 		panic("%s: mode", "ext2_write");
1910 #endif
1911 
1912 	switch (vp->v_type) {
1913 	case VREG:
1914 		if (ioflag & IO_APPEND)
1915 			uio->uio_offset = ip->i_size;
1916 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1917 			return (EPERM);
1918 		/* FALLTHROUGH */
1919 	case VLNK:
1920 		break;
1921 	case VDIR:
1922 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
1923 		if ((ioflag & IO_SYNC) == 0)
1924 		panic("ext2_write: nonsync dir write");
1925 		break;
1926 	default:
1927 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1928 		    vp->v_type, (intmax_t)uio->uio_offset,
1929 		    (intmax_t)uio->uio_resid);
1930 	}
1931 
1932 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1933 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1934 	fs = ip->i_e2fs;
1935 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1936 		return (EFBIG);
1937 	/*
1938 	 * Maybe this should be above the vnode op call, but so long as
1939 	 * file servers have no limits, I don't think it matters.
1940 	 */
1941 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1942 		return (EFBIG);
1943 
1944 	resid = uio->uio_resid;
1945 	osize = ip->i_size;
1946 	if (seqcount > BA_SEQMAX)
1947 		flags = BA_SEQMAX << BA_SEQSHIFT;
1948 	else
1949 		flags = seqcount << BA_SEQSHIFT;
1950 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1951 		flags |= IO_SYNC;
1952 
1953 	for (error = 0; uio->uio_resid > 0;) {
1954 		lbn = lblkno(fs, uio->uio_offset);
1955 		blkoffset = blkoff(fs, uio->uio_offset);
1956 		xfersize = fs->e2fs_fsize - blkoffset;
1957 		if (uio->uio_resid < xfersize)
1958 			xfersize = uio->uio_resid;
1959 		if (uio->uio_offset + xfersize > ip->i_size)
1960 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1961 
1962 		/*
1963 		 * We must perform a read-before-write if the transfer size
1964 		 * does not cover the entire buffer.
1965 		 */
1966 		if (fs->e2fs_bsize > xfersize)
1967 			flags |= BA_CLRBUF;
1968 		else
1969 			flags &= ~BA_CLRBUF;
1970 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1971 		    ap->a_cred, &bp, flags);
1972 		if (error != 0)
1973 			break;
1974 
1975 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1976 			bp->b_flags |= B_NOCACHE;
1977 		if (uio->uio_offset + xfersize > ip->i_size)
1978 			ip->i_size = uio->uio_offset + xfersize;
1979 		size = blksize(fs, ip, lbn) - bp->b_resid;
1980 		if (size < xfersize)
1981 			xfersize = size;
1982 
1983 		error =
1984 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1985 		/*
1986 		 * If the buffer is not already filled and we encounter an
1987 		 * error while trying to fill it, we have to clear out any
1988 		 * garbage data from the pages instantiated for the buffer.
1989 		 * If we do not, a failed uiomove() during a write can leave
1990 		 * the prior contents of the pages exposed to a userland mmap.
1991 		 *
1992 		 * Note that we need only clear buffers with a transfer size
1993 		 * equal to the block size because buffers with a shorter
1994 		 * transfer size were cleared above by the call to ext2_balloc()
1995 		 * with the BA_CLRBUF flag set.
1996 		 *
1997 		 * If the source region for uiomove identically mmaps the
1998 		 * buffer, uiomove() performed the NOP copy, and the buffer
1999 		 * content remains valid because the page fault handler
2000 		 * validated the pages.
2001 		 */
2002 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2003 		    fs->e2fs_bsize == xfersize)
2004 			vfs_bio_clrbuf(bp);
2005 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
2006 			bp->b_flags |= B_RELBUF;
2007 		}
2008 
2009 		/*
2010 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
2011 		 * if we have a severe page deficiency write the buffer
2012 		 * asynchronously.  Otherwise try to cluster, and if that
2013 		 * doesn't do it then either do an async write (if O_DIRECT),
2014 		 * or a delayed write (if not).
2015 		 */
2016 		if (ioflag & IO_SYNC) {
2017 			(void)bwrite(bp);
2018 		} else if (vm_page_count_severe() ||
2019 		    buf_dirty_count_severe() ||
2020 		    (ioflag & IO_ASYNC)) {
2021 			bp->b_flags |= B_CLUSTEROK;
2022 			bawrite(bp);
2023 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
2024 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2025 				bp->b_flags |= B_CLUSTEROK;
2026 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
2027 			} else {
2028 				bawrite(bp);
2029 			}
2030 		} else if (ioflag & IO_DIRECT) {
2031 			bp->b_flags |= B_CLUSTEROK;
2032 			bawrite(bp);
2033 		} else {
2034 			bp->b_flags |= B_CLUSTEROK;
2035 			bdwrite(bp);
2036 		}
2037 		if (error || xfersize == 0)
2038 			break;
2039 	}
2040 	/*
2041 	 * If we successfully wrote any data, and we are not the superuser
2042 	 * we clear the setuid and setgid bits as a precaution against
2043 	 * tampering.
2044 	 */
2045 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2046 	    ap->a_cred) {
2047 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2048 			ip->i_mode &= ~(ISUID | ISGID);
2049 	}
2050 	if (error) {
2051 		if (ioflag & IO_UNIT) {
2052 			(void)ext2_truncate(vp, osize,
2053 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2054 			uio->uio_offset -= resid - uio->uio_resid;
2055 			uio->uio_resid = resid;
2056 		}
2057 	}
2058 	if (uio->uio_resid != resid) {
2059 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
2060 		if (ioflag & IO_SYNC)
2061 			error = ext2_update(vp, 1);
2062 	}
2063 	return (error);
2064 }
2065