xref: /freebsd/sys/fs/ext2fs/ext2_vnops.c (revision 6e660824a82f590542932de52f128db584029893)
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/mount.h>
58 #include <sys/unistd.h>
59 #include <sys/time.h>
60 #include <sys/vnode.h>
61 #include <sys/namei.h>
62 #include <sys/lockf.h>
63 #include <sys/event.h>
64 #include <sys/conf.h>
65 #include <sys/file.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vnode_pager.h>
72 
73 #include "opt_directio.h"
74 
75 #include <ufs/ufs/dir.h>
76 
77 #include <fs/ext2fs/fs.h>
78 #include <fs/ext2fs/inode.h>
79 #include <fs/ext2fs/ext2_extern.h>
80 #include <fs/ext2fs/ext2fs.h>
81 #include <fs/ext2fs/ext2_dinode.h>
82 #include <fs/ext2fs/ext2_dir.h>
83 #include <fs/ext2fs/ext2_mount.h>
84 
85 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
86 static void ext2_itimes_locked(struct vnode *);
87 
88 static vop_access_t	ext2_access;
89 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
90 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
91     struct thread *);
92 static vop_close_t	ext2_close;
93 static vop_create_t	ext2_create;
94 static vop_fsync_t	ext2_fsync;
95 static vop_getattr_t	ext2_getattr;
96 static vop_ioctl_t	ext2_ioctl;
97 static vop_link_t	ext2_link;
98 static vop_mkdir_t	ext2_mkdir;
99 static vop_mknod_t	ext2_mknod;
100 static vop_open_t	ext2_open;
101 static vop_pathconf_t	ext2_pathconf;
102 static vop_print_t	ext2_print;
103 static vop_read_t	ext2_read;
104 static vop_readlink_t	ext2_readlink;
105 static vop_remove_t	ext2_remove;
106 static vop_rename_t	ext2_rename;
107 static vop_rmdir_t	ext2_rmdir;
108 static vop_setattr_t	ext2_setattr;
109 static vop_strategy_t	ext2_strategy;
110 static vop_symlink_t	ext2_symlink;
111 static vop_write_t	ext2_write;
112 static vop_vptofh_t	ext2_vptofh;
113 static vop_close_t	ext2fifo_close;
114 static vop_kqfilter_t	ext2fifo_kqfilter;
115 
116 /* Global vfs data structures for ext2. */
117 struct vop_vector ext2_vnodeops = {
118 	.vop_default =		&default_vnodeops,
119 	.vop_access =		ext2_access,
120 	.vop_bmap =		ext2_bmap,
121 	.vop_cachedlookup =	ext2_lookup,
122 	.vop_close =		ext2_close,
123 	.vop_create =		ext2_create,
124 	.vop_fsync =		ext2_fsync,
125 	.vop_getattr =		ext2_getattr,
126 	.vop_inactive =		ext2_inactive,
127 	.vop_ioctl =		ext2_ioctl,
128 	.vop_link =		ext2_link,
129 	.vop_lookup =		vfs_cache_lookup,
130 	.vop_mkdir =		ext2_mkdir,
131 	.vop_mknod =		ext2_mknod,
132 	.vop_open =		ext2_open,
133 	.vop_pathconf =		ext2_pathconf,
134 	.vop_poll =		vop_stdpoll,
135 	.vop_print =		ext2_print,
136 	.vop_read =		ext2_read,
137 	.vop_readdir =		ext2_readdir,
138 	.vop_readlink =		ext2_readlink,
139 	.vop_reallocblks =	ext2_reallocblks,
140 	.vop_reclaim =		ext2_reclaim,
141 	.vop_remove =		ext2_remove,
142 	.vop_rename =		ext2_rename,
143 	.vop_rmdir =		ext2_rmdir,
144 	.vop_setattr =		ext2_setattr,
145 	.vop_strategy =		ext2_strategy,
146 	.vop_symlink =		ext2_symlink,
147 	.vop_write =		ext2_write,
148 	.vop_vptofh =		ext2_vptofh,
149 };
150 
151 struct vop_vector ext2_fifoops = {
152 	.vop_default =		&fifo_specops,
153 	.vop_access =		ext2_access,
154 	.vop_close =		ext2fifo_close,
155 	.vop_fsync =		ext2_fsync,
156 	.vop_getattr =		ext2_getattr,
157 	.vop_inactive =		ext2_inactive,
158 	.vop_kqfilter =		ext2fifo_kqfilter,
159 	.vop_print =		ext2_print,
160 	.vop_read =		VOP_PANIC,
161 	.vop_reclaim =		ext2_reclaim,
162 	.vop_setattr =		ext2_setattr,
163 	.vop_write =		VOP_PANIC,
164 	.vop_vptofh =		ext2_vptofh,
165 };
166 
167 /*
168  * A virgin directory (no blushing please).
169  * Note that the type and namlen fields are reversed relative to ext2.
170  * Also, we don't use `struct odirtemplate', since it would just cause
171  * endianness problems.
172  */
173 static struct dirtemplate mastertemplate = {
174 	0, 12, 1, EXT2_FT_DIR, ".",
175 	0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
176 };
177 static struct dirtemplate omastertemplate = {
178 	0, 12, 1, EXT2_FT_UNKNOWN, ".",
179 	0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
180 };
181 
182 static void
183 ext2_itimes_locked(struct vnode *vp)
184 {
185 	struct inode *ip;
186 	struct timespec ts;
187 
188 	ASSERT_VI_LOCKED(vp, __func__);
189 
190 	ip = VTOI(vp);
191 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
192 		return;
193 	if ((vp->v_type == VBLK || vp->v_type == VCHR))
194 		ip->i_flag |= IN_LAZYMOD;
195 	else
196 		ip->i_flag |= IN_MODIFIED;
197 	if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
198 		vfs_timestamp(&ts);
199 		if (ip->i_flag & IN_ACCESS) {
200 			ip->i_atime = ts.tv_sec;
201 			ip->i_atimensec = ts.tv_nsec;
202 		}
203 		if (ip->i_flag & IN_UPDATE) {
204 			ip->i_mtime = ts.tv_sec;
205 			ip->i_mtimensec = ts.tv_nsec;
206 			ip->i_modrev++;
207 		}
208 		if (ip->i_flag & IN_CHANGE) {
209 			ip->i_ctime = ts.tv_sec;
210 			ip->i_ctimensec = ts.tv_nsec;
211 		}
212 	}
213 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
214 }
215 
216 void
217 ext2_itimes(struct vnode *vp)
218 {
219 
220 	VI_LOCK(vp);
221 	ext2_itimes_locked(vp);
222 	VI_UNLOCK(vp);
223 }
224 
225 /*
226  * Create a regular file
227  */
228 static int
229 ext2_create(struct vop_create_args *ap)
230 {
231 	int error;
232 
233 	error =
234 	    ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
235 	    ap->a_dvp, ap->a_vpp, ap->a_cnp);
236 	if (error)
237 		return (error);
238 	return (0);
239 }
240 
241 static int
242 ext2_open(struct vop_open_args *ap)
243 {
244 
245 	if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
246 		return (EOPNOTSUPP);
247 
248 	/*
249 	 * Files marked append-only must be opened for appending.
250 	 */
251 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
252 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
253 		return (EPERM);
254 
255 	vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
256 
257 	return (0);
258 }
259 
260 /*
261  * Close called.
262  *
263  * Update the times on the inode.
264  */
265 static int
266 ext2_close(struct vop_close_args *ap)
267 {
268 	struct vnode *vp = ap->a_vp;
269 
270 	VI_LOCK(vp);
271 	if (vp->v_usecount > 1)
272 		ext2_itimes_locked(vp);
273 	VI_UNLOCK(vp);
274 	return (0);
275 }
276 
277 static int
278 ext2_access(struct vop_access_args *ap)
279 {
280 	struct vnode *vp = ap->a_vp;
281 	struct inode *ip = VTOI(vp);
282 	accmode_t accmode = ap->a_accmode;
283 	int error;
284 
285 	if (vp->v_type == VBLK || vp->v_type == VCHR)
286 		return (EOPNOTSUPP);
287 
288 	/*
289 	 * Disallow write attempts on read-only file systems;
290 	 * unless the file is a socket, fifo, or a block or
291 	 * character device resident on the file system.
292 	 */
293 	if (accmode & VWRITE) {
294 		switch (vp->v_type) {
295 		case VDIR:
296 		case VLNK:
297 		case VREG:
298 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
299 				return (EROFS);
300 			break;
301 		default:
302 			break;
303 		}
304 	}
305 
306 	/* If immutable bit set, nobody gets to write it. */
307 	if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
308 		return (EPERM);
309 
310 	error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
311 	    ap->a_accmode, ap->a_cred, NULL);
312 	return (error);
313 }
314 
315 static int
316 ext2_getattr(struct vop_getattr_args *ap)
317 {
318 	struct vnode *vp = ap->a_vp;
319 	struct inode *ip = VTOI(vp);
320 	struct vattr *vap = ap->a_vap;
321 
322 	ext2_itimes(vp);
323 	/*
324 	 * Copy from inode table
325 	 */
326 	vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
327 	vap->va_fileid = ip->i_number;
328 	vap->va_mode = ip->i_mode & ~IFMT;
329 	vap->va_nlink = ip->i_nlink;
330 	vap->va_uid = ip->i_uid;
331 	vap->va_gid = ip->i_gid;
332 	vap->va_rdev = ip->i_rdev;
333 	vap->va_size = ip->i_size;
334 	vap->va_atime.tv_sec = ip->i_atime;
335 	vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
336 	vap->va_mtime.tv_sec = ip->i_mtime;
337 	vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
338 	vap->va_ctime.tv_sec = ip->i_ctime;
339 	vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
340 	if E2DI_HAS_XTIME(ip) {
341 		vap->va_birthtime.tv_sec = ip->i_birthtime;
342 		vap->va_birthtime.tv_nsec = ip->i_birthnsec;
343 	}
344 	vap->va_flags = ip->i_flags;
345 	vap->va_gen = ip->i_gen;
346 	vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
347 	vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
348 	vap->va_type = IFTOVT(ip->i_mode);
349 	vap->va_filerev = ip->i_modrev;
350 	return (0);
351 }
352 
353 /*
354  * Set attribute vnode op. called from several syscalls
355  */
356 static int
357 ext2_setattr(struct vop_setattr_args *ap)
358 {
359 	struct vattr *vap = ap->a_vap;
360 	struct vnode *vp = ap->a_vp;
361 	struct inode *ip = VTOI(vp);
362 	struct ucred *cred = ap->a_cred;
363 	struct thread *td = curthread;
364 	int error;
365 
366 	/*
367 	 * Check for unsettable attributes.
368 	 */
369 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
370 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
371 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
372 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
373 		return (EINVAL);
374 	}
375 	if (vap->va_flags != VNOVAL) {
376 		/* Disallow flags not supported by ext2fs. */
377 		if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
378 			return (EOPNOTSUPP);
379 
380 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
381 			return (EROFS);
382 		/*
383 		 * Callers may only modify the file flags on objects they
384 		 * have VADMIN rights for.
385 		 */
386 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
387 			return (error);
388 		/*
389 		 * Unprivileged processes and privileged processes in
390 		 * jail() are not permitted to unset system flags, or
391 		 * modify flags if any system flags are set.
392 		 * Privileged non-jail processes may not modify system flags
393 		 * if securelevel > 0 and any existing system flags are set.
394 		 */
395 		if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
396 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
397 				error = securelevel_gt(cred, 0);
398 				if (error)
399 					return (error);
400 			}
401 		} else {
402 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
403 			    ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
404 				return (EPERM);
405 		}
406 		ip->i_flags = vap->va_flags;
407 		ip->i_flag |= IN_CHANGE;
408 		if (ip->i_flags & (IMMUTABLE | APPEND))
409 			return (0);
410 	}
411 	if (ip->i_flags & (IMMUTABLE | APPEND))
412 		return (EPERM);
413 	/*
414 	 * Go through the fields and update iff not VNOVAL.
415 	 */
416 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
417 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
418 			return (EROFS);
419 		if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
420 		    td)) != 0)
421 			return (error);
422 	}
423 	if (vap->va_size != VNOVAL) {
424 		/*
425 		 * Disallow write attempts on read-only file systems;
426 		 * unless the file is a socket, fifo, or a block or
427 		 * character device resident on the file system.
428 		 */
429 		switch (vp->v_type) {
430 		case VDIR:
431 			return (EISDIR);
432 		case VLNK:
433 		case VREG:
434 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
435 				return (EROFS);
436 			break;
437 		default:
438 			break;
439 		}
440 		if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
441 			return (error);
442 	}
443 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
444 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
445 			return (EROFS);
446 		/*
447 		 * From utimes(2):
448 		 * If times is NULL, ... The caller must be the owner of
449 		 * the file, have permission to write the file, or be the
450 		 * super-user.
451 		 * If times is non-NULL, ... The caller must be the owner of
452 		 * the file or be the super-user.
453 		 */
454 		if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
455 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
456 		    (error = VOP_ACCESS(vp, VWRITE, cred, td))))
457 			return (error);
458 		if (vap->va_atime.tv_sec != VNOVAL)
459 			ip->i_flag |= IN_ACCESS;
460 		if (vap->va_mtime.tv_sec != VNOVAL)
461 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
462 		ext2_itimes(vp);
463 		if (vap->va_atime.tv_sec != VNOVAL) {
464 			ip->i_atime = vap->va_atime.tv_sec;
465 			ip->i_atimensec = vap->va_atime.tv_nsec;
466 		}
467 		if (vap->va_mtime.tv_sec != VNOVAL) {
468 			ip->i_mtime = vap->va_mtime.tv_sec;
469 			ip->i_mtimensec = vap->va_mtime.tv_nsec;
470 		}
471 		ip->i_birthtime = vap->va_birthtime.tv_sec;
472 		ip->i_birthnsec = vap->va_birthtime.tv_nsec;
473 		error = ext2_update(vp, 0);
474 		if (error)
475 			return (error);
476 	}
477 	error = 0;
478 	if (vap->va_mode != (mode_t)VNOVAL) {
479 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
480 			return (EROFS);
481 		error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
482 	}
483 	return (error);
484 }
485 
486 /*
487  * Change the mode on a file.
488  * Inode must be locked before calling.
489  */
490 static int
491 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
492 {
493 	struct inode *ip = VTOI(vp);
494 	int error;
495 
496 	/*
497 	 * To modify the permissions on a file, must possess VADMIN
498 	 * for that file.
499 	 */
500 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
501 		return (error);
502 	/*
503 	 * Privileged processes may set the sticky bit on non-directories,
504 	 * as well as set the setgid bit on a file with a group that the
505 	 * process is not a member of.
506 	 */
507 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
508 		error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
509 		if (error)
510 			return (EFTYPE);
511 	}
512 	if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
513 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
514 		if (error)
515 			return (error);
516 	}
517 	ip->i_mode &= ~ALLPERMS;
518 	ip->i_mode |= (mode & ALLPERMS);
519 	ip->i_flag |= IN_CHANGE;
520 	return (0);
521 }
522 
523 /*
524  * Perform chown operation on inode ip;
525  * inode must be locked prior to call.
526  */
527 static int
528 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
529     struct thread *td)
530 {
531 	struct inode *ip = VTOI(vp);
532 	uid_t ouid;
533 	gid_t ogid;
534 	int error = 0;
535 
536 	if (uid == (uid_t)VNOVAL)
537 		uid = ip->i_uid;
538 	if (gid == (gid_t)VNOVAL)
539 		gid = ip->i_gid;
540 	/*
541 	 * To modify the ownership of a file, must possess VADMIN
542 	 * for that file.
543 	 */
544 	if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
545 		return (error);
546 	/*
547 	 * To change the owner of a file, or change the group of a file
548 	 * to a group of which we are not a member, the caller must
549 	 * have privilege.
550 	 */
551 	if (uid != ip->i_uid || (gid != ip->i_gid &&
552 	    !groupmember(gid, cred))) {
553 		error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
554 		if (error)
555 			return (error);
556 	}
557 	ogid = ip->i_gid;
558 	ouid = ip->i_uid;
559 	ip->i_gid = gid;
560 	ip->i_uid = uid;
561 	ip->i_flag |= IN_CHANGE;
562 	if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
563 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
564 			ip->i_mode &= ~(ISUID | ISGID);
565 	}
566 	return (0);
567 }
568 
569 /*
570  * Synch an open file.
571  */
572 /* ARGSUSED */
573 static int
574 ext2_fsync(struct vop_fsync_args *ap)
575 {
576 	/*
577 	 * Flush all dirty buffers associated with a vnode.
578 	 */
579 
580 	vop_stdfsync(ap);
581 
582 	return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
583 }
584 
585 /*
586  * Mknod vnode call
587  */
588 /* ARGSUSED */
589 static int
590 ext2_mknod(struct vop_mknod_args *ap)
591 {
592 	struct vattr *vap = ap->a_vap;
593 	struct vnode **vpp = ap->a_vpp;
594 	struct inode *ip;
595 	ino_t ino;
596 	int error;
597 
598 	error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
599 	    ap->a_dvp, vpp, ap->a_cnp);
600 	if (error)
601 		return (error);
602 	ip = VTOI(*vpp);
603 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
604 	if (vap->va_rdev != VNOVAL) {
605 		/*
606 		 * Want to be able to use this to make badblock
607 		 * inodes, so don't truncate the dev number.
608 		 */
609 		ip->i_rdev = vap->va_rdev;
610 	}
611 	/*
612 	 * Remove inode, then reload it through VFS_VGET so it is
613 	 * checked to see if it is an alias of an existing entry in
614 	 * the inode cache.	 XXX I don't believe this is necessary now.
615 	 */
616 	(*vpp)->v_type = VNON;
617 	ino = ip->i_number;	/* Save this before vgone() invalidates ip. */
618 	vgone(*vpp);
619 	vput(*vpp);
620 	error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
621 	if (error) {
622 		*vpp = NULL;
623 		return (error);
624 	}
625 	return (0);
626 }
627 
628 static int
629 ext2_remove(struct vop_remove_args *ap)
630 {
631 	struct inode *ip;
632 	struct vnode *vp = ap->a_vp;
633 	struct vnode *dvp = ap->a_dvp;
634 	int error;
635 
636 	ip = VTOI(vp);
637 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
638 	    (VTOI(dvp)->i_flags & APPEND)) {
639 		error = EPERM;
640 		goto out;
641 	}
642 	error = ext2_dirremove(dvp, ap->a_cnp);
643 	if (error == 0) {
644 		ip->i_nlink--;
645 		ip->i_flag |= IN_CHANGE;
646 	}
647 out:
648 	return (error);
649 }
650 
651 /*
652  * link vnode call
653  */
654 static int
655 ext2_link(struct vop_link_args *ap)
656 {
657 	struct vnode *vp = ap->a_vp;
658 	struct vnode *tdvp = ap->a_tdvp;
659 	struct componentname *cnp = ap->a_cnp;
660 	struct inode *ip;
661 	int error;
662 
663 #ifdef INVARIANTS
664 	if ((cnp->cn_flags & HASBUF) == 0)
665 		panic("ext2_link: no name");
666 #endif
667 	if (tdvp->v_mount != vp->v_mount) {
668 		error = EXDEV;
669 		goto out;
670 	}
671 	ip = VTOI(vp);
672 	if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
673 		error = EMLINK;
674 		goto out;
675 	}
676 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
677 		error = EPERM;
678 		goto out;
679 	}
680 	ip->i_nlink++;
681 	ip->i_flag |= IN_CHANGE;
682 	error = ext2_update(vp, !DOINGASYNC(vp));
683 	if (!error)
684 		error = ext2_direnter(ip, tdvp, cnp);
685 	if (error) {
686 		ip->i_nlink--;
687 		ip->i_flag |= IN_CHANGE;
688 	}
689 out:
690 	return (error);
691 }
692 
693 /*
694  * Rename system call.
695  * 	rename("foo", "bar");
696  * is essentially
697  *	unlink("bar");
698  *	link("foo", "bar");
699  *	unlink("foo");
700  * but ``atomically''.  Can't do full commit without saving state in the
701  * inode on disk which isn't feasible at this time.  Best we can do is
702  * always guarantee the target exists.
703  *
704  * Basic algorithm is:
705  *
706  * 1) Bump link count on source while we're linking it to the
707  *    target.  This also ensure the inode won't be deleted out
708  *    from underneath us while we work (it may be truncated by
709  *    a concurrent `trunc' or `open' for creation).
710  * 2) Link source to destination.  If destination already exists,
711  *    delete it first.
712  * 3) Unlink source reference to inode if still around. If a
713  *    directory was moved and the parent of the destination
714  *    is different from the source, patch the ".." entry in the
715  *    directory.
716  */
717 static int
718 ext2_rename(struct vop_rename_args *ap)
719 {
720 	struct vnode *tvp = ap->a_tvp;
721 	struct vnode *tdvp = ap->a_tdvp;
722 	struct vnode *fvp = ap->a_fvp;
723 	struct vnode *fdvp = ap->a_fdvp;
724 	struct componentname *tcnp = ap->a_tcnp;
725 	struct componentname *fcnp = ap->a_fcnp;
726 	struct inode *ip, *xp, *dp;
727 	struct dirtemplate dirbuf;
728 	int doingdirectory = 0, oldparent = 0, newparent = 0;
729 	int error = 0;
730 	u_char namlen;
731 
732 #ifdef INVARIANTS
733 	if ((tcnp->cn_flags & HASBUF) == 0 ||
734 	    (fcnp->cn_flags & HASBUF) == 0)
735 		panic("ext2_rename: no name");
736 #endif
737 	/*
738 	 * Check for cross-device rename.
739 	 */
740 	if ((fvp->v_mount != tdvp->v_mount) ||
741 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
742 		error = EXDEV;
743 abortit:
744 		if (tdvp == tvp)
745 			vrele(tdvp);
746 		else
747 			vput(tdvp);
748 		if (tvp)
749 			vput(tvp);
750 		vrele(fdvp);
751 		vrele(fvp);
752 		return (error);
753 	}
754 
755 	if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
756 	    (VTOI(tdvp)->i_flags & APPEND))) {
757 		error = EPERM;
758 		goto abortit;
759 	}
760 
761 	/*
762 	 * Renaming a file to itself has no effect.  The upper layers should
763 	 * not call us in that case.  Temporarily just warn if they do.
764 	 */
765 	if (fvp == tvp) {
766 		printf("ext2_rename: fvp == tvp (can't happen)\n");
767 		error = 0;
768 		goto abortit;
769 	}
770 
771 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
772 		goto abortit;
773 	dp = VTOI(fdvp);
774 	ip = VTOI(fvp);
775 	if (ip->i_nlink >= EXT2_LINK_MAX) {
776  		VOP_UNLOCK(fvp, 0);
777  		error = EMLINK;
778  		goto abortit;
779  	}
780 	if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
781 	    || (dp->i_flags & APPEND)) {
782 		VOP_UNLOCK(fvp, 0);
783 		error = EPERM;
784 		goto abortit;
785 	}
786 	if ((ip->i_mode & IFMT) == IFDIR) {
787 		/*
788 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
789 		 */
790 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
791 		    dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
792 		    (ip->i_flag & IN_RENAME)) {
793 			VOP_UNLOCK(fvp, 0);
794 			error = EINVAL;
795 			goto abortit;
796 		}
797 		ip->i_flag |= IN_RENAME;
798 		oldparent = dp->i_number;
799 		doingdirectory++;
800 	}
801 	vrele(fdvp);
802 
803 	/*
804 	 * When the target exists, both the directory
805 	 * and target vnodes are returned locked.
806 	 */
807 	dp = VTOI(tdvp);
808 	xp = NULL;
809 	if (tvp)
810 		xp = VTOI(tvp);
811 
812 	/*
813 	 * 1) Bump link count while we're moving stuff
814 	 *    around.  If we crash somewhere before
815 	 *    completing our work, the link count
816 	 *    may be wrong, but correctable.
817 	 */
818 	ip->i_nlink++;
819 	ip->i_flag |= IN_CHANGE;
820 	if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
821 		VOP_UNLOCK(fvp, 0);
822 		goto bad;
823 	}
824 
825 	/*
826 	 * If ".." must be changed (ie the directory gets a new
827 	 * parent) then the source directory must not be in the
828 	 * directory hierarchy above the target, as this would
829 	 * orphan everything below the source directory. Also
830 	 * the user must have write permission in the source so
831 	 * as to be able to change "..". We must repeat the call
832 	 * to namei, as the parent directory is unlocked by the
833 	 * call to checkpath().
834 	 */
835 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
836 	VOP_UNLOCK(fvp, 0);
837 	if (oldparent != dp->i_number)
838 		newparent = dp->i_number;
839 	if (doingdirectory && newparent) {
840 		if (error)	/* write access check above */
841 			goto bad;
842 		if (xp != NULL)
843 			vput(tvp);
844 		error = ext2_checkpath(ip, dp, tcnp->cn_cred);
845 		if (error)
846 			goto out;
847 		VREF(tdvp);
848 		error = relookup(tdvp, &tvp, tcnp);
849 		if (error)
850 			goto out;
851 		vrele(tdvp);
852 		dp = VTOI(tdvp);
853 		xp = NULL;
854 		if (tvp)
855 			xp = VTOI(tvp);
856 	}
857 	/*
858 	 * 2) If target doesn't exist, link the target
859 	 *    to the source and unlink the source.
860 	 *    Otherwise, rewrite the target directory
861 	 *    entry to reference the source inode and
862 	 *    expunge the original entry's existence.
863 	 */
864 	if (xp == NULL) {
865 		if (dp->i_devvp != ip->i_devvp)
866 			panic("ext2_rename: EXDEV");
867 		/*
868 		 * Account for ".." in new directory.
869 		 * When source and destination have the same
870 		 * parent we don't fool with the link count.
871 		 */
872 		if (doingdirectory && newparent) {
873 			if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
874 				error = EMLINK;
875 				goto bad;
876 			}
877 			dp->i_nlink++;
878 			dp->i_flag |= IN_CHANGE;
879 			error = ext2_update(tdvp, !DOINGASYNC(tdvp));
880 			if (error)
881 				goto bad;
882 		}
883 		error = ext2_direnter(ip, tdvp, tcnp);
884 		if (error) {
885 			if (doingdirectory && newparent) {
886 				dp->i_nlink--;
887 				dp->i_flag |= IN_CHANGE;
888 				(void)ext2_update(tdvp, 1);
889 			}
890 			goto bad;
891 		}
892 		vput(tdvp);
893 	} else {
894 		if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
895 		       panic("ext2_rename: EXDEV");
896 		/*
897 		 * Short circuit rename(foo, foo).
898 		 */
899 		if (xp->i_number == ip->i_number)
900 			panic("ext2_rename: same file");
901 		/*
902 		 * If the parent directory is "sticky", then the user must
903 		 * own the parent directory, or the destination of the rename,
904 		 * otherwise the destination may not be changed (except by
905 		 * root). This implements append-only directories.
906 		 */
907 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
908 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
909 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
910 			error = EPERM;
911 			goto bad;
912 		}
913 		/*
914 		 * Target must be empty if a directory and have no links
915 		 * to it. Also, ensure source and target are compatible
916 		 * (both directories, or both not directories).
917 		 */
918 		if ((xp->i_mode&IFMT) == IFDIR) {
919 			if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
920 			    xp->i_nlink > 2) {
921 				error = ENOTEMPTY;
922 				goto bad;
923 			}
924 			if (!doingdirectory) {
925 				error = ENOTDIR;
926 				goto bad;
927 			}
928 			cache_purge(tdvp);
929 		} else if (doingdirectory) {
930 			error = EISDIR;
931 			goto bad;
932 		}
933 		error = ext2_dirrewrite(dp, ip, tcnp);
934 		if (error)
935 			goto bad;
936 		/*
937 		 * If the target directory is in the same
938 		 * directory as the source directory,
939 		 * decrement the link count on the parent
940 		 * of the target directory.
941 		 */
942 		if (doingdirectory && !newparent) {
943 		       dp->i_nlink--;
944 		       dp->i_flag |= IN_CHANGE;
945 		}
946 		vput(tdvp);
947 		/*
948 		 * Adjust the link count of the target to
949 		 * reflect the dirrewrite above.  If this is
950 		 * a directory it is empty and there are
951 		 * no links to it, so we can squash the inode and
952 		 * any space associated with it.  We disallowed
953 		 * renaming over top of a directory with links to
954 		 * it above, as the remaining link would point to
955 		 * a directory without "." or ".." entries.
956 		 */
957 		xp->i_nlink--;
958 		if (doingdirectory) {
959 			if (--xp->i_nlink != 0)
960 				panic("ext2_rename: linked directory");
961 			error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
962 			    tcnp->cn_cred, tcnp->cn_thread);
963 		}
964 		xp->i_flag |= IN_CHANGE;
965 		vput(tvp);
966 		xp = NULL;
967 	}
968 
969 	/*
970 	 * 3) Unlink the source.
971 	 */
972 	fcnp->cn_flags &= ~MODMASK;
973 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
974 	VREF(fdvp);
975 	error = relookup(fdvp, &fvp, fcnp);
976 	if (error == 0)
977 		vrele(fdvp);
978 	if (fvp != NULL) {
979 		xp = VTOI(fvp);
980 		dp = VTOI(fdvp);
981 	} else {
982 		/*
983 		 * From name has disappeared.
984 		 */
985 		if (doingdirectory)
986 			panic("ext2_rename: lost dir entry");
987 		vrele(ap->a_fvp);
988 		return (0);
989 	}
990 	/*
991 	 * Ensure that the directory entry still exists and has not
992 	 * changed while the new name has been entered. If the source is
993 	 * a file then the entry may have been unlinked or renamed. In
994 	 * either case there is no further work to be done. If the source
995 	 * is a directory then it cannot have been rmdir'ed; its link
996 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
997 	 * The IN_RENAME flag ensures that it cannot be moved by another
998 	 * rename.
999 	 */
1000 	if (xp != ip) {
1001 		if (doingdirectory)
1002 			panic("ext2_rename: lost dir entry");
1003 	} else {
1004 		/*
1005 		 * If the source is a directory with a
1006 		 * new parent, the link count of the old
1007 		 * parent directory must be decremented
1008 		 * and ".." set to point to the new parent.
1009 		 */
1010 		if (doingdirectory && newparent) {
1011 			dp->i_nlink--;
1012 			dp->i_flag |= IN_CHANGE;
1013 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1014 				sizeof(struct dirtemplate), (off_t)0,
1015 				UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1016 				tcnp->cn_cred, NOCRED, NULL, NULL);
1017 			if (error == 0) {
1018 				/* Like ufs little-endian: */
1019 				namlen = dirbuf.dotdot_type;
1020 				if (namlen != 2 ||
1021 				    dirbuf.dotdot_name[0] != '.' ||
1022 				    dirbuf.dotdot_name[1] != '.') {
1023 					ext2_dirbad(xp, (doff_t)12,
1024 					    "rename: mangled dir");
1025 				} else {
1026 					dirbuf.dotdot_ino = newparent;
1027 					(void) vn_rdwr(UIO_WRITE, fvp,
1028 					    (caddr_t)&dirbuf,
1029 					    sizeof(struct dirtemplate),
1030 					    (off_t)0, UIO_SYSSPACE,
1031 					    IO_NODELOCKED | IO_SYNC |
1032 					    IO_NOMACCHECK, tcnp->cn_cred,
1033 					    NOCRED, NULL, NULL);
1034 					cache_purge(fdvp);
1035 				}
1036 			}
1037 		}
1038 		error = ext2_dirremove(fdvp, fcnp);
1039 		if (!error) {
1040 			xp->i_nlink--;
1041 			xp->i_flag |= IN_CHANGE;
1042 		}
1043 		xp->i_flag &= ~IN_RENAME;
1044 	}
1045 	if (dp)
1046 		vput(fdvp);
1047 	if (xp)
1048 		vput(fvp);
1049 	vrele(ap->a_fvp);
1050 	return (error);
1051 
1052 bad:
1053 	if (xp)
1054 		vput(ITOV(xp));
1055 	vput(ITOV(dp));
1056 out:
1057 	if (doingdirectory)
1058 		ip->i_flag &= ~IN_RENAME;
1059 	if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1060 		ip->i_nlink--;
1061 		ip->i_flag |= IN_CHANGE;
1062 		ip->i_flag &= ~IN_RENAME;
1063 		vput(fvp);
1064 	} else
1065 		vrele(fvp);
1066 	return (error);
1067 }
1068 
1069 /*
1070  * Mkdir system call
1071  */
1072 static int
1073 ext2_mkdir(struct vop_mkdir_args *ap)
1074 {
1075 	struct vnode *dvp = ap->a_dvp;
1076 	struct vattr *vap = ap->a_vap;
1077 	struct componentname *cnp = ap->a_cnp;
1078 	struct inode *ip, *dp;
1079 	struct vnode *tvp;
1080 	struct dirtemplate dirtemplate, *dtp;
1081 	int error, dmode;
1082 
1083 #ifdef INVARIANTS
1084 	if ((cnp->cn_flags & HASBUF) == 0)
1085 		panic("ext2_mkdir: no name");
1086 #endif
1087 	dp = VTOI(dvp);
1088 	if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1089 		error = EMLINK;
1090 		goto out;
1091 	}
1092 	dmode = vap->va_mode & 0777;
1093 	dmode |= IFDIR;
1094 	/*
1095 	 * Must simulate part of ext2_makeinode here to acquire the inode,
1096 	 * but not have it entered in the parent directory. The entry is
1097 	 * made later after writing "." and ".." entries.
1098 	 */
1099 	error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1100 	if (error)
1101 		goto out;
1102 	ip = VTOI(tvp);
1103 	ip->i_gid = dp->i_gid;
1104 #ifdef SUIDDIR
1105 	{
1106 		/*
1107 		 * if we are hacking owners here, (only do this where told to)
1108 		 * and we are not giving it TOO root, (would subvert quotas)
1109 		 * then go ahead and give it to the other user.
1110 		 * The new directory also inherits the SUID bit.
1111 		 * If user's UID and dir UID are the same,
1112 		 * 'give it away' so that the SUID is still forced on.
1113 		 */
1114 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1115 		   (dp->i_mode & ISUID) && dp->i_uid) {
1116 			dmode |= ISUID;
1117 			ip->i_uid = dp->i_uid;
1118 		} else {
1119 			ip->i_uid = cnp->cn_cred->cr_uid;
1120 		}
1121 	}
1122 #else
1123 	ip->i_uid = cnp->cn_cred->cr_uid;
1124 #endif
1125 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1126 	ip->i_mode = dmode;
1127 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1128 	ip->i_nlink = 2;
1129 	if (cnp->cn_flags & ISWHITEOUT)
1130 		ip->i_flags |= UF_OPAQUE;
1131 	error = ext2_update(tvp, 1);
1132 
1133 	/*
1134 	 * Bump link count in parent directory
1135 	 * to reflect work done below.  Should
1136 	 * be done before reference is created
1137 	 * so reparation is possible if we crash.
1138 	 */
1139 	dp->i_nlink++;
1140 	dp->i_flag |= IN_CHANGE;
1141 	error = ext2_update(dvp, !DOINGASYNC(dvp));
1142 	if (error)
1143 		goto bad;
1144 
1145 	/* Initialize directory with "." and ".." from static template. */
1146 	if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1147 	    EXT2F_INCOMPAT_FTYPE))
1148 		dtp = &mastertemplate;
1149 	else
1150 		dtp = &omastertemplate;
1151 	dirtemplate = *dtp;
1152 	dirtemplate.dot_ino = ip->i_number;
1153 	dirtemplate.dotdot_ino = dp->i_number;
1154 	/* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1155 	 * so let's just redefine it - for this function only
1156 	 */
1157 #undef  DIRBLKSIZ
1158 #define DIRBLKSIZ  VTOI(dvp)->i_e2fs->e2fs_bsize
1159 	dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1160 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1161 	    sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1162 	    IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1163 	    NULL, NULL);
1164 	if (error) {
1165 		dp->i_nlink--;
1166 		dp->i_flag |= IN_CHANGE;
1167 		goto bad;
1168 	}
1169 	if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1170 		/* XXX should grow with balloc() */
1171 		panic("ext2_mkdir: blksize");
1172 	else {
1173 		ip->i_size = DIRBLKSIZ;
1174 		ip->i_flag |= IN_CHANGE;
1175 	}
1176 
1177 	/* Directory set up, now install its entry in the parent directory. */
1178 	error = ext2_direnter(ip, dvp, cnp);
1179 	if (error) {
1180 		dp->i_nlink--;
1181 		dp->i_flag |= IN_CHANGE;
1182 	}
1183 bad:
1184 	/*
1185 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1186 	 * for us because we set the link count to 0.
1187 	 */
1188 	if (error) {
1189 		ip->i_nlink = 0;
1190 		ip->i_flag |= IN_CHANGE;
1191 		vput(tvp);
1192 	} else
1193 		*ap->a_vpp = tvp;
1194 out:
1195 	return (error);
1196 #undef  DIRBLKSIZ
1197 #define DIRBLKSIZ  DEV_BSIZE
1198 }
1199 
1200 /*
1201  * Rmdir system call.
1202  */
1203 static int
1204 ext2_rmdir(struct vop_rmdir_args *ap)
1205 {
1206 	struct vnode *vp = ap->a_vp;
1207 	struct vnode *dvp = ap->a_dvp;
1208 	struct componentname *cnp = ap->a_cnp;
1209 	struct inode *ip, *dp;
1210 	int error;
1211 
1212 	ip = VTOI(vp);
1213 	dp = VTOI(dvp);
1214 
1215 	/*
1216 	 * Verify the directory is empty (and valid).
1217 	 * (Rmdir ".." won't be valid since
1218 	 *  ".." will contain a reference to
1219 	 *  the current directory and thus be
1220 	 *  non-empty.)
1221 	 */
1222 	error = 0;
1223 	if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1224 		error = ENOTEMPTY;
1225 		goto out;
1226 	}
1227 	if ((dp->i_flags & APPEND)
1228 	    || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1229 		error = EPERM;
1230 		goto out;
1231 	}
1232 	/*
1233 	 * Delete reference to directory before purging
1234 	 * inode.  If we crash in between, the directory
1235 	 * will be reattached to lost+found,
1236 	 */
1237 	error = ext2_dirremove(dvp, cnp);
1238 	if (error)
1239 		goto out;
1240 	dp->i_nlink--;
1241 	dp->i_flag |= IN_CHANGE;
1242 	cache_purge(dvp);
1243 	VOP_UNLOCK(dvp, 0);
1244 	/*
1245 	 * Truncate inode.  The only stuff left
1246 	 * in the directory is "." and "..".  The
1247 	 * "." reference is inconsequential since
1248 	 * we're quashing it.  The ".." reference
1249 	 * has already been adjusted above.  We've
1250 	 * removed the "." reference and the reference
1251 	 * in the parent directory, but there may be
1252 	 * other hard links so decrement by 2 and
1253 	 * worry about them later.
1254 	 */
1255 	ip->i_nlink -= 2;
1256 	error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1257 	    cnp->cn_thread);
1258 	cache_purge(ITOV(ip));
1259 	if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1260 		VOP_UNLOCK(vp, 0);
1261 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1262 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1263 	}
1264 out:
1265 	return (error);
1266 }
1267 
1268 /*
1269  * symlink -- make a symbolic link
1270  */
1271 static int
1272 ext2_symlink(struct vop_symlink_args *ap)
1273 {
1274 	struct vnode *vp, **vpp = ap->a_vpp;
1275 	struct inode *ip;
1276 	int len, error;
1277 
1278 	error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1279 	    vpp, ap->a_cnp);
1280 	if (error)
1281 		return (error);
1282 	vp = *vpp;
1283 	len = strlen(ap->a_target);
1284 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1285 		ip = VTOI(vp);
1286 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1287 		ip->i_size = len;
1288 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1289 	} else
1290 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1291 		    UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1292 		    ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1293 	if (error)
1294 		vput(vp);
1295 	return (error);
1296 }
1297 
1298 /*
1299  * Return target name of a symbolic link
1300  */
1301 static int
1302 ext2_readlink(struct vop_readlink_args *ap)
1303 {
1304 	struct vnode *vp = ap->a_vp;
1305 	struct inode *ip = VTOI(vp);
1306 	int isize;
1307 
1308 	isize = ip->i_size;
1309 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1310 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1311 		return (0);
1312 	}
1313 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1314 }
1315 
1316 /*
1317  * Calculate the logical to physical mapping if not done already,
1318  * then call the device strategy routine.
1319  *
1320  * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1321  * deadlock on memory.  See ext2_bmap() for details.
1322  */
1323 static int
1324 ext2_strategy(struct vop_strategy_args *ap)
1325 {
1326 	struct buf *bp = ap->a_bp;
1327 	struct vnode *vp = ap->a_vp;
1328 	struct inode *ip;
1329 	struct bufobj *bo;
1330 	int32_t blkno;
1331 	int error;
1332 
1333 	ip = VTOI(vp);
1334 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1335 		panic("ext2_strategy: spec");
1336 	if (bp->b_blkno == bp->b_lblkno) {
1337 		error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1338 		bp->b_blkno = blkno;
1339 		if (error) {
1340 			bp->b_error = error;
1341 			bp->b_ioflags |= BIO_ERROR;
1342 			bufdone(bp);
1343 			return (0);
1344 		}
1345 		if ((long)bp->b_blkno == -1)
1346 			vfs_bio_clrbuf(bp);
1347 	}
1348 	if ((long)bp->b_blkno == -1) {
1349 		bufdone(bp);
1350 		return (0);
1351 	}
1352 	bp->b_iooffset = dbtob(bp->b_blkno);
1353 	bo = VFSTOEXT2(vp->v_mount)->um_bo;
1354 	BO_STRATEGY(bo, bp);
1355 	return (0);
1356 }
1357 
1358 /*
1359  * Print out the contents of an inode.
1360  */
1361 static int
1362 ext2_print(struct vop_print_args *ap)
1363 {
1364 	struct vnode *vp = ap->a_vp;
1365 	struct inode *ip = VTOI(vp);
1366 
1367 	vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1368 	if (vp->v_type == VFIFO)
1369 		fifo_printinfo(vp);
1370 	printf("\n");
1371 	return (0);
1372 }
1373 
1374 /*
1375  * Close wrapper for fifos.
1376  *
1377  * Update the times on the inode then do device close.
1378  */
1379 static int
1380 ext2fifo_close(struct vop_close_args *ap)
1381 {
1382 	struct vnode *vp = ap->a_vp;
1383 
1384 	VI_LOCK(vp);
1385 	if (vp->v_usecount > 1)
1386 		ext2_itimes_locked(vp);
1387 	VI_UNLOCK(vp);
1388 	return (fifo_specops.vop_close(ap));
1389 }
1390 
1391 /*
1392  * Kqfilter wrapper for fifos.
1393  *
1394  * Fall through to ext2 kqfilter routines if needed
1395  */
1396 static int
1397 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1398 {
1399 	int error;
1400 
1401 	error = fifo_specops.vop_kqfilter(ap);
1402 	if (error)
1403 		error = vfs_kqfilter(ap);
1404 	return (error);
1405 }
1406 
1407 /*
1408  * Return POSIX pathconf information applicable to ext2 filesystems.
1409  */
1410 static int
1411 ext2_pathconf(struct vop_pathconf_args *ap)
1412 {
1413 	int error = 0;
1414 
1415 	switch (ap->a_name) {
1416 	case _PC_LINK_MAX:
1417 		*ap->a_retval = EXT2_LINK_MAX;
1418 		break;
1419 	case _PC_NAME_MAX:
1420 		*ap->a_retval = NAME_MAX;
1421 		break;
1422 	case _PC_PATH_MAX:
1423 		*ap->a_retval = PATH_MAX;
1424 		break;
1425 	case _PC_PIPE_BUF:
1426 		*ap->a_retval = PIPE_BUF;
1427 		break;
1428 	case _PC_CHOWN_RESTRICTED:
1429 		*ap->a_retval = 1;
1430 		break;
1431 	case _PC_NO_TRUNC:
1432 		*ap->a_retval = 1;
1433 		break;
1434 	case _PC_MIN_HOLE_SIZE:
1435 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1436 		break;
1437 	case _PC_ASYNC_IO:
1438 		/* _PC_ASYNC_IO should have been handled by upper layers. */
1439 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1440 		error = EINVAL;
1441 		break;
1442 	case _PC_PRIO_IO:
1443 		*ap->a_retval = 0;
1444 		break;
1445 	case _PC_SYNC_IO:
1446 		*ap->a_retval = 0;
1447 		break;
1448 	case _PC_ALLOC_SIZE_MIN:
1449 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1450 		break;
1451 	case _PC_FILESIZEBITS:
1452 		*ap->a_retval = 64;
1453 		break;
1454 	case _PC_REC_INCR_XFER_SIZE:
1455 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1456 		break;
1457 	case _PC_REC_MAX_XFER_SIZE:
1458 		*ap->a_retval = -1; /* means ``unlimited'' */
1459 		break;
1460 	case _PC_REC_MIN_XFER_SIZE:
1461 		*ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1462 		break;
1463 	case _PC_REC_XFER_ALIGN:
1464 		*ap->a_retval = PAGE_SIZE;
1465 		break;
1466 	case _PC_SYMLINK_MAX:
1467 		*ap->a_retval = MAXPATHLEN;
1468 		break;
1469 
1470 	default:
1471 		error = EINVAL;
1472 		break;
1473 	}
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Vnode pointer to File handle
1479  */
1480 /* ARGSUSED */
1481 static int
1482 ext2_vptofh(struct vop_vptofh_args *ap)
1483 {
1484 	struct inode *ip;
1485 	struct ufid *ufhp;
1486 
1487 	ip = VTOI(ap->a_vp);
1488 	ufhp = (struct ufid *)ap->a_fhp;
1489 	ufhp->ufid_len = sizeof(struct ufid);
1490 	ufhp->ufid_ino = ip->i_number;
1491 	ufhp->ufid_gen = ip->i_gen;
1492 	return (0);
1493 }
1494 
1495 /*
1496  * Initialize the vnode associated with a new inode, handle aliased
1497  * vnodes.
1498  */
1499 int
1500 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1501 {
1502 	struct inode *ip;
1503 	struct vnode *vp;
1504 
1505 	vp = *vpp;
1506 	ip = VTOI(vp);
1507 	vp->v_type = IFTOVT(ip->i_mode);
1508 	if (vp->v_type == VFIFO)
1509 		vp->v_op = fifoops;
1510 
1511 	if (ip->i_number == EXT2_ROOTINO)
1512 		vp->v_vflag |= VV_ROOT;
1513 	ip->i_modrev = init_va_filerev();
1514 	*vpp = vp;
1515 	return (0);
1516 }
1517 
1518 /*
1519  * Allocate a new inode.
1520  */
1521 static int
1522 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1523     struct componentname *cnp)
1524 {
1525 	struct inode *ip, *pdir;
1526 	struct vnode *tvp;
1527 	int error;
1528 
1529 	pdir = VTOI(dvp);
1530 #ifdef INVARIANTS
1531 	if ((cnp->cn_flags & HASBUF) == 0)
1532 		panic("ext2_makeinode: no name");
1533 #endif
1534 	*vpp = NULL;
1535 	if ((mode & IFMT) == 0)
1536 		mode |= IFREG;
1537 
1538 	error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1539 	if (error) {
1540 		return (error);
1541 	}
1542 	ip = VTOI(tvp);
1543 	ip->i_gid = pdir->i_gid;
1544 #ifdef SUIDDIR
1545 	{
1546 		/*
1547 		 * if we are
1548 		 * not the owner of the directory,
1549 		 * and we are hacking owners here, (only do this where told to)
1550 		 * and we are not giving it TOO root, (would subvert quotas)
1551 		 * then go ahead and give it to the other user.
1552 		 * Note that this drops off the execute bits for security.
1553 		 */
1554 		if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1555 		     (pdir->i_mode & ISUID) &&
1556 		     (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1557 			ip->i_uid = pdir->i_uid;
1558 			mode &= ~07111;
1559 		} else {
1560 			ip->i_uid = cnp->cn_cred->cr_uid;
1561 		}
1562 	}
1563 #else
1564 	ip->i_uid = cnp->cn_cred->cr_uid;
1565 #endif
1566 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1567 	ip->i_mode = mode;
1568 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
1569 	ip->i_nlink = 1;
1570 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1571 		if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1572 			ip->i_mode &= ~ISGID;
1573 	}
1574 
1575 	if (cnp->cn_flags & ISWHITEOUT)
1576 		ip->i_flags |= UF_OPAQUE;
1577 
1578 	/*
1579 	 * Make sure inode goes to disk before directory entry.
1580 	 */
1581 	error = ext2_update(tvp, !DOINGASYNC(tvp));
1582 	if (error)
1583 		goto bad;
1584 	error = ext2_direnter(ip, dvp, cnp);
1585 	if (error)
1586 		goto bad;
1587 
1588 	*vpp = tvp;
1589 	return (0);
1590 
1591 bad:
1592 	/*
1593 	 * Write error occurred trying to update the inode
1594 	 * or the directory so must deallocate the inode.
1595 	 */
1596 	ip->i_nlink = 0;
1597 	ip->i_flag |= IN_CHANGE;
1598 	vput(tvp);
1599 	return (error);
1600 }
1601 
1602 /*
1603  * Vnode op for reading.
1604  */
1605 static int
1606 ext2_read(struct vop_read_args *ap)
1607 {
1608 	struct vnode *vp;
1609 	struct inode *ip;
1610 	struct uio *uio;
1611 	struct m_ext2fs *fs;
1612 	struct buf *bp;
1613 	daddr_t lbn, nextlbn;
1614 	off_t bytesinfile;
1615 	long size, xfersize, blkoffset;
1616 	int error, orig_resid, seqcount;
1617 	int ioflag;
1618 
1619 	vp = ap->a_vp;
1620 	uio = ap->a_uio;
1621 	ioflag = ap->a_ioflag;
1622 
1623 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1624 	ip = VTOI(vp);
1625 
1626 #ifdef INVARIANTS
1627 	if (uio->uio_rw != UIO_READ)
1628 		panic("%s: mode", "ext2_read");
1629 
1630 	if (vp->v_type == VLNK) {
1631 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1632 			panic("%s: short symlink", "ext2_read");
1633 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
1634 		panic("%s: type %d", "ext2_read", vp->v_type);
1635 #endif
1636 	orig_resid = uio->uio_resid;
1637 	KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1638 	if (orig_resid == 0)
1639 		return (0);
1640 	KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1641 	fs = ip->i_e2fs;
1642 	if (uio->uio_offset < ip->i_size &&
1643 	    uio->uio_offset >= fs->e2fs_maxfilesize)
1644 	    	return (EOVERFLOW);
1645 
1646 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1647 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1648 			break;
1649 		lbn = lblkno(fs, uio->uio_offset);
1650 		nextlbn = lbn + 1;
1651 		size = blksize(fs, ip, lbn);
1652 		blkoffset = blkoff(fs, uio->uio_offset);
1653 
1654 		xfersize = fs->e2fs_fsize - blkoffset;
1655 		if (uio->uio_resid < xfersize)
1656 			xfersize = uio->uio_resid;
1657 		if (bytesinfile < xfersize)
1658 			xfersize = bytesinfile;
1659 
1660 		if (lblktosize(fs, nextlbn) >= ip->i_size)
1661 			error = bread(vp, lbn, size, NOCRED, &bp);
1662 		else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
1663 			error = cluster_read(vp, ip->i_size, lbn, size,
1664 			    NOCRED, blkoffset + uio->uio_resid, seqcount,
1665 			    0, &bp);
1666 		} else if (seqcount > 1) {
1667 			int nextsize = blksize(fs, ip, nextlbn);
1668 			error = breadn(vp, lbn,
1669 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1670 		} else
1671 			error = bread(vp, lbn, size, NOCRED, &bp);
1672 		if (error) {
1673 			brelse(bp);
1674 			bp = NULL;
1675 			break;
1676 		}
1677 
1678 		/*
1679 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
1680 		 * will cause us to attempt to release the buffer later on
1681 		 * and will cause the buffer cache to attempt to free the
1682 		 * underlying pages.
1683 		 */
1684 		if (ioflag & IO_DIRECT)
1685 			bp->b_flags |= B_DIRECT;
1686 
1687 		/*
1688 		 * We should only get non-zero b_resid when an I/O error
1689 		 * has occurred, which should cause us to break above.
1690 		 * However, if the short read did not cause an error,
1691 		 * then we want to ensure that we do not uiomove bad
1692 		 * or uninitialized data.
1693 		 */
1694 		size -= bp->b_resid;
1695 		if (size < xfersize) {
1696 			if (size == 0)
1697 				break;
1698 			xfersize = size;
1699 		}
1700 		error = uiomove((char *)bp->b_data + blkoffset,
1701   			(int)xfersize, uio);
1702 		if (error)
1703 			break;
1704 
1705 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1706 			/*
1707 			 * If it's VMIO or direct I/O, then we don't
1708 			 * need the buf, mark it available for
1709 			 * freeing. If it's non-direct VMIO, the VM has
1710 			 * the data.
1711 			 */
1712 			bp->b_flags |= B_RELBUF;
1713 			brelse(bp);
1714 		} else {
1715 			/*
1716 			 * Otherwise let whoever
1717 			 * made the request take care of
1718 			 * freeing it. We just queue
1719 			 * it onto another list.
1720 			 */
1721 			bqrelse(bp);
1722 		}
1723 	}
1724 
1725 	/*
1726 	 * This can only happen in the case of an error
1727 	 * because the loop above resets bp to NULL on each iteration
1728 	 * and on normal completion has not set a new value into it.
1729 	 * so it must have come from a 'break' statement
1730 	 */
1731 	if (bp != NULL) {
1732 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1733 			bp->b_flags |= B_RELBUF;
1734 			brelse(bp);
1735 		} else {
1736 			bqrelse(bp);
1737 		}
1738 	}
1739 
1740 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1741 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1742 		ip->i_flag |= IN_ACCESS;
1743 	return (error);
1744 }
1745 
1746 static int
1747 ext2_ioctl(struct vop_ioctl_args *ap)
1748 {
1749 
1750 	switch (ap->a_command) {
1751 	case FIOSEEKDATA:
1752 	case FIOSEEKHOLE:
1753 		return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1754 		    (off_t *)ap->a_data, ap->a_cred));
1755 	default:
1756 		return (ENOTTY);
1757 	}
1758 }
1759 
1760 /*
1761  * Vnode op for writing.
1762  */
1763 static int
1764 ext2_write(struct vop_write_args *ap)
1765 {
1766 	struct vnode *vp;
1767 	struct uio *uio;
1768 	struct inode *ip;
1769 	struct m_ext2fs *fs;
1770 	struct buf *bp;
1771 	daddr_t lbn;
1772 	off_t osize;
1773 	int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1774 
1775 	ioflag = ap->a_ioflag;
1776 	uio = ap->a_uio;
1777 	vp = ap->a_vp;
1778 
1779 	seqcount = ioflag >> IO_SEQSHIFT;
1780 	ip = VTOI(vp);
1781 
1782 #ifdef INVARIANTS
1783 	if (uio->uio_rw != UIO_WRITE)
1784 		panic("%s: mode", "ext2_write");
1785 #endif
1786 
1787 	switch (vp->v_type) {
1788 	case VREG:
1789 		if (ioflag & IO_APPEND)
1790 			uio->uio_offset = ip->i_size;
1791 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1792 			return (EPERM);
1793 		/* FALLTHROUGH */
1794 	case VLNK:
1795 		break;
1796 	case VDIR:
1797 		/* XXX differs from ffs -- this is called from ext2_mkdir(). */
1798 		if ((ioflag & IO_SYNC) == 0)
1799 		panic("ext2_write: nonsync dir write");
1800 		break;
1801 	default:
1802 		panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1803 		    vp->v_type, (intmax_t)uio->uio_offset,
1804 		    (intmax_t)uio->uio_resid);
1805 	}
1806 
1807 	KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1808 	KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1809 	fs = ip->i_e2fs;
1810 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1811 		return (EFBIG);
1812 	/*
1813 	 * Maybe this should be above the vnode op call, but so long as
1814 	 * file servers have no limits, I don't think it matters.
1815 	 */
1816 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1817 		return (EFBIG);
1818 
1819 	resid = uio->uio_resid;
1820 	osize = ip->i_size;
1821 	if (seqcount > BA_SEQMAX)
1822 		flags = BA_SEQMAX << BA_SEQSHIFT;
1823 	else
1824 		flags = seqcount << BA_SEQSHIFT;
1825 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1826 		flags |= IO_SYNC;
1827 
1828 	for (error = 0; uio->uio_resid > 0;) {
1829 		lbn = lblkno(fs, uio->uio_offset);
1830 		blkoffset = blkoff(fs, uio->uio_offset);
1831 		xfersize = fs->e2fs_fsize - blkoffset;
1832 		if (uio->uio_resid < xfersize)
1833 			xfersize = uio->uio_resid;
1834 		if (uio->uio_offset + xfersize > ip->i_size)
1835 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1836 
1837                 /*
1838 		 * We must perform a read-before-write if the transfer size
1839 		 * does not cover the entire buffer.
1840                  */
1841 		if (fs->e2fs_bsize > xfersize)
1842 			flags |= BA_CLRBUF;
1843 		else
1844 			flags &= ~BA_CLRBUF;
1845 		error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1846 		    ap->a_cred, &bp, flags);
1847 		if (error != 0)
1848 			break;
1849 
1850 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1851 			bp->b_flags |= B_NOCACHE;
1852 		if (uio->uio_offset + xfersize > ip->i_size)
1853 			ip->i_size = uio->uio_offset + xfersize;
1854 		size = blksize(fs, ip, lbn) - bp->b_resid;
1855 		if (size < xfersize)
1856 			xfersize = size;
1857 
1858 		error =
1859 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1860 		/*
1861 		 * If the buffer is not already filled and we encounter an
1862 		 * error while trying to fill it, we have to clear out any
1863 		 * garbage data from the pages instantiated for the buffer.
1864 		 * If we do not, a failed uiomove() during a write can leave
1865 		 * the prior contents of the pages exposed to a userland mmap.
1866 		 *
1867 		 * Note that we need only clear buffers with a transfer size
1868 		 * equal to the block size because buffers with a shorter
1869 		 * transfer size were cleared above by the call to ext2_balloc()
1870 		 * with the BA_CLRBUF flag set.
1871 		 *
1872 		 * If the source region for uiomove identically mmaps the
1873 		 * buffer, uiomove() performed the NOP copy, and the buffer
1874 		 * content remains valid because the page fault handler
1875 		 * validated the pages.
1876 		 */
1877 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
1878 		    fs->e2fs_bsize == xfersize)
1879 			vfs_bio_clrbuf(bp);
1880 		if (ioflag & (IO_VMIO|IO_DIRECT)) {
1881 			bp->b_flags |= B_RELBUF;
1882 		}
1883 
1884 		/*
1885 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1886 		 * if we have a severe page deficiency write the buffer
1887 		 * asynchronously.  Otherwise try to cluster, and if that
1888 		 * doesn't do it then either do an async write (if O_DIRECT),
1889 		 * or a delayed write (if not).
1890 		 */
1891 		if (ioflag & IO_SYNC) {
1892 			(void)bwrite(bp);
1893 		} else if (vm_page_count_severe() ||
1894 		    buf_dirty_count_severe() ||
1895 		    (ioflag & IO_ASYNC)) {
1896 			bp->b_flags |= B_CLUSTEROK;
1897 			bawrite(bp);
1898 		} else if (xfersize + blkoffset == fs->e2fs_fsize) {
1899 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
1900 				bp->b_flags |= B_CLUSTEROK;
1901 				cluster_write(vp, bp, ip->i_size, seqcount, 0);
1902 			} else {
1903 				bawrite(bp);
1904 			}
1905 		} else if (ioflag & IO_DIRECT) {
1906 			bp->b_flags |= B_CLUSTEROK;
1907 			bawrite(bp);
1908 		} else {
1909 			bp->b_flags |= B_CLUSTEROK;
1910 			bdwrite(bp);
1911 		}
1912 		if (error || xfersize == 0)
1913 			break;
1914 	}
1915 	/*
1916 	 * If we successfully wrote any data, and we are not the superuser
1917 	 * we clear the setuid and setgid bits as a precaution against
1918 	 * tampering.
1919 	 */
1920 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1921 	    ap->a_cred) {
1922 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
1923 			ip->i_mode &= ~(ISUID | ISGID);
1924 	}
1925 	if (error) {
1926 		if (ioflag & IO_UNIT) {
1927 			(void)ext2_truncate(vp, osize,
1928 			    ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
1929 			uio->uio_offset -= resid - uio->uio_resid;
1930 			uio->uio_resid = resid;
1931 		}
1932 	}
1933 	if (uio->uio_resid != resid) {
1934                ip->i_flag |= IN_CHANGE | IN_UPDATE;
1935                if (ioflag & IO_SYNC)
1936                        error = ext2_update(vp, 1);
1937        }
1938 	return (error);
1939 }
1940