xref: /original-bsd/sys/ufs/ffs/ufs_vnops.c (revision 27393bdf)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * %sccs.include.redist.c%
11  *
12  *	@(#)ufs_vnops.c	8.11 (Berkeley) 06/04/94
13  */
14 
15 #include <sys/param.h>
16 #include <sys/systm.h>
17 #include <sys/namei.h>
18 #include <sys/resourcevar.h>
19 #include <sys/kernel.h>
20 #include <sys/file.h>
21 #include <sys/stat.h>
22 #include <sys/buf.h>
23 #include <sys/proc.h>
24 #include <sys/conf.h>
25 #include <sys/mount.h>
26 #include <sys/vnode.h>
27 #include <sys/malloc.h>
28 #include <sys/dirent.h>
29 
30 #include <vm/vm.h>
31 
32 #include <miscfs/specfs/specdev.h>
33 
34 #include <ufs/ufs/lockf.h>
35 #include <ufs/ufs/quota.h>
36 #include <ufs/ufs/inode.h>
37 #include <ufs/ufs/dir.h>
38 #include <ufs/ufs/ufsmount.h>
39 #include <ufs/ufs/ufs_extern.h>
40 
41 static int ufs_chmod __P((struct vnode *, int, struct ucred *, struct proc *));
42 static int ufs_chown
43 	__P((struct vnode *, uid_t, gid_t, struct ucred *, struct proc *));
44 
45 union _qcvt {
46 	quad_t qcvt;
47 	long val[2];
48 };
49 #define SETHIGH(q, h) { \
50 	union _qcvt tmp; \
51 	tmp.qcvt = (q); \
52 	tmp.val[_QUAD_HIGHWORD] = (h); \
53 	(q) = tmp.qcvt; \
54 }
55 #define SETLOW(q, l) { \
56 	union _qcvt tmp; \
57 	tmp.qcvt = (q); \
58 	tmp.val[_QUAD_LOWWORD] = (l); \
59 	(q) = tmp.qcvt; \
60 }
61 
62 /*
63  * Create a regular file
64  */
65 int
66 ufs_create(ap)
67 	struct vop_create_args /* {
68 		struct vnode *a_dvp;
69 		struct vnode **a_vpp;
70 		struct componentname *a_cnp;
71 		struct vattr *a_vap;
72 	} */ *ap;
73 {
74 	int error;
75 
76 	if (error =
77 	    ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
78 	    ap->a_dvp, ap->a_vpp, ap->a_cnp))
79 		return (error);
80 	return (0);
81 }
82 
83 /*
84  * Mknod vnode call
85  */
86 /* ARGSUSED */
87 int
88 ufs_mknod(ap)
89 	struct vop_mknod_args /* {
90 		struct vnode *a_dvp;
91 		struct vnode **a_vpp;
92 		struct componentname *a_cnp;
93 		struct vattr *a_vap;
94 	} */ *ap;
95 {
96 	register struct vattr *vap = ap->a_vap;
97 	register struct vnode **vpp = ap->a_vpp;
98 	register struct inode *ip;
99 	int error;
100 
101 	if (error =
102 	    ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
103 	    ap->a_dvp, vpp, ap->a_cnp))
104 		return (error);
105 	ip = VTOI(*vpp);
106 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
107 	if (vap->va_rdev != VNOVAL) {
108 		/*
109 		 * Want to be able to use this to make badblock
110 		 * inodes, so don't truncate the dev number.
111 		 */
112 		ip->i_rdev = vap->va_rdev;
113 	}
114 	/*
115 	 * Remove inode so that it will be reloaded by VFS_VGET and
116 	 * checked to see if it is an alias of an existing entry in
117 	 * the inode cache.
118 	 */
119 	vput(*vpp);
120 	(*vpp)->v_type = VNON;
121 	vgone(*vpp);
122 	*vpp = 0;
123 	return (0);
124 }
125 
126 /*
127  * Open called.
128  *
129  * Nothing to do.
130  */
131 /* ARGSUSED */
132 int
133 ufs_open(ap)
134 	struct vop_open_args /* {
135 		struct vnode *a_vp;
136 		int  a_mode;
137 		struct ucred *a_cred;
138 		struct proc *a_p;
139 	} */ *ap;
140 {
141 
142 	/*
143 	 * Files marked append-only must be opened for appending.
144 	 */
145 	if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
146 	    (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
147 		return (EPERM);
148 	return (0);
149 }
150 
151 /*
152  * Close called.
153  *
154  * Update the times on the inode.
155  */
156 /* ARGSUSED */
157 int
158 ufs_close(ap)
159 	struct vop_close_args /* {
160 		struct vnode *a_vp;
161 		int  a_fflag;
162 		struct ucred *a_cred;
163 		struct proc *a_p;
164 	} */ *ap;
165 {
166 	register struct vnode *vp = ap->a_vp;
167 	register struct inode *ip = VTOI(vp);
168 
169 	if (vp->v_usecount > 1 && !(ip->i_flag & IN_LOCKED))
170 		ITIMES(ip, &time, &time);
171 	return (0);
172 }
173 
174 int
175 ufs_access(ap)
176 	struct vop_access_args /* {
177 		struct vnode *a_vp;
178 		int  a_mode;
179 		struct ucred *a_cred;
180 		struct proc *a_p;
181 	} */ *ap;
182 {
183 	register struct vnode *vp = ap->a_vp;
184 	register struct inode *ip = VTOI(vp);
185 	register struct ucred *cred = ap->a_cred;
186 	mode_t mask, mode = ap->a_mode;
187 	register gid_t *gp;
188 	int i, error;
189 
190 #ifdef DIAGNOSTIC
191 	if (!VOP_ISLOCKED(vp)) {
192 		vprint("ufs_access: not locked", vp);
193 		panic("ufs_access: not locked");
194 	}
195 #endif
196 #ifdef QUOTA
197 	if (mode & VWRITE)
198 		switch (vp->v_type) {
199 		case VDIR:
200 		case VLNK:
201 		case VREG:
202 			if (error = getinoquota(ip))
203 				return (error);
204 			break;
205 		}
206 #endif
207 
208 	/* If immutable bit set, nobody gets to write it. */
209 	if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
210 		return (EPERM);
211 
212 	/* Otherwise, user id 0 always gets access. */
213 	if (cred->cr_uid == 0)
214 		return (0);
215 
216 	mask = 0;
217 
218 	/* Otherwise, check the owner. */
219 	if (cred->cr_uid == ip->i_uid) {
220 		if (mode & VEXEC)
221 			mask |= S_IXUSR;
222 		if (mode & VREAD)
223 			mask |= S_IRUSR;
224 		if (mode & VWRITE)
225 			mask |= S_IWUSR;
226 		return ((ip->i_mode & mask) == mask ? 0 : EACCES);
227 	}
228 
229 	/* Otherwise, check the groups. */
230 	for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
231 		if (ip->i_gid == *gp) {
232 			if (mode & VEXEC)
233 				mask |= S_IXGRP;
234 			if (mode & VREAD)
235 				mask |= S_IRGRP;
236 			if (mode & VWRITE)
237 				mask |= S_IWGRP;
238 			return ((ip->i_mode & mask) == mask ? 0 : EACCES);
239 		}
240 
241 	/* Otherwise, check everyone else. */
242 	if (mode & VEXEC)
243 		mask |= S_IXOTH;
244 	if (mode & VREAD)
245 		mask |= S_IROTH;
246 	if (mode & VWRITE)
247 		mask |= S_IWOTH;
248 	return ((ip->i_mode & mask) == mask ? 0 : EACCES);
249 }
250 
251 /* ARGSUSED */
252 int
253 ufs_getattr(ap)
254 	struct vop_getattr_args /* {
255 		struct vnode *a_vp;
256 		struct vattr *a_vap;
257 		struct ucred *a_cred;
258 		struct proc *a_p;
259 	} */ *ap;
260 {
261 	register struct vnode *vp = ap->a_vp;
262 	register struct inode *ip = VTOI(vp);
263 	register struct vattr *vap = ap->a_vap;
264 
265 	ITIMES(ip, &time, &time);
266 	/*
267 	 * Copy from inode table
268 	 */
269 	vap->va_fsid = ip->i_dev;
270 	vap->va_fileid = ip->i_number;
271 	vap->va_mode = ip->i_mode & ~IFMT;
272 	vap->va_nlink = ip->i_nlink;
273 	vap->va_uid = ip->i_uid;
274 	vap->va_gid = ip->i_gid;
275 	vap->va_rdev = (dev_t)ip->i_rdev;
276 	vap->va_size = ip->i_din.di_size;
277 	vap->va_atime = ip->i_atime;
278 	vap->va_mtime = ip->i_mtime;
279 	vap->va_ctime = ip->i_ctime;
280 	vap->va_flags = ip->i_flags;
281 	vap->va_gen = ip->i_gen;
282 	/* this doesn't belong here */
283 	if (vp->v_type == VBLK)
284 		vap->va_blocksize = BLKDEV_IOSIZE;
285 	else if (vp->v_type == VCHR)
286 		vap->va_blocksize = MAXBSIZE;
287 	else
288 		vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
289 	vap->va_bytes = dbtob(ip->i_blocks);
290 	vap->va_type = vp->v_type;
291 	vap->va_filerev = ip->i_modrev;
292 	return (0);
293 }
294 
295 /*
296  * Set attribute vnode op. called from several syscalls
297  */
298 int
299 ufs_setattr(ap)
300 	struct vop_setattr_args /* {
301 		struct vnode *a_vp;
302 		struct vattr *a_vap;
303 		struct ucred *a_cred;
304 		struct proc *a_p;
305 	} */ *ap;
306 {
307 	register struct vattr *vap = ap->a_vap;
308 	register struct vnode *vp = ap->a_vp;
309 	register struct inode *ip = VTOI(vp);
310 	register struct ucred *cred = ap->a_cred;
311 	register struct proc *p = ap->a_p;
312 	struct timeval atimeval, mtimeval;
313 	int error;
314 
315 	/*
316 	 * Check for unsettable attributes.
317 	 */
318 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
319 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
320 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
321 	    ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
322 		return (EINVAL);
323 	}
324 	if (vap->va_flags != VNOVAL) {
325 		if (cred->cr_uid != ip->i_uid &&
326 		    (error = suser(cred, &p->p_acflag)))
327 			return (error);
328 		if (cred->cr_uid == 0) {
329 			if ((ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) &&
330 			    securelevel > 0)
331 				return (EPERM);
332 			ip->i_flags = vap->va_flags;
333 		} else {
334 			if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND))
335 				return (EPERM);
336 			ip->i_flags &= SF_SETTABLE;
337 			ip->i_flags |= (vap->va_flags & UF_SETTABLE);
338 		}
339 		ip->i_flag |= IN_CHANGE;
340 		if (vap->va_flags & (IMMUTABLE | APPEND))
341 			return (0);
342 	}
343 	if (ip->i_flags & (IMMUTABLE | APPEND))
344 		return (EPERM);
345 	/*
346 	 * Go through the fields and update iff not VNOVAL.
347 	 */
348 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL)
349 		if (error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred, p))
350 			return (error);
351 	if (vap->va_size != VNOVAL) {
352 		if (vp->v_type == VDIR)
353 			return (EISDIR);
354 		if (error = VOP_TRUNCATE(vp, vap->va_size, 0, cred, p))
355 			return (error);
356 	}
357 	ip = VTOI(vp);
358 	if (vap->va_atime.ts_sec != VNOVAL || vap->va_mtime.ts_sec != VNOVAL) {
359 		if (cred->cr_uid != ip->i_uid &&
360 		    (error = suser(cred, &p->p_acflag)) &&
361 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
362 		    (error = VOP_ACCESS(vp, VWRITE, cred, p))))
363 			return (error);
364 		if (vap->va_atime.ts_sec != VNOVAL)
365 			ip->i_flag |= IN_ACCESS;
366 		if (vap->va_mtime.ts_sec != VNOVAL)
367 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
368 		atimeval.tv_sec = vap->va_atime.ts_sec;
369 		atimeval.tv_usec = vap->va_atime.ts_nsec / 1000;
370 		mtimeval.tv_sec = vap->va_mtime.ts_sec;
371 		mtimeval.tv_usec = vap->va_mtime.ts_nsec / 1000;
372 		if (error = VOP_UPDATE(vp, &atimeval, &mtimeval, 1))
373 			return (error);
374 	}
375 	error = 0;
376 	if (vap->va_mode != (mode_t)VNOVAL)
377 		error = ufs_chmod(vp, (int)vap->va_mode, cred, p);
378 	return (error);
379 }
380 
381 /*
382  * Change the mode on a file.
383  * Inode must be locked before calling.
384  */
385 static int
386 ufs_chmod(vp, mode, cred, p)
387 	register struct vnode *vp;
388 	register int mode;
389 	register struct ucred *cred;
390 	struct proc *p;
391 {
392 	register struct inode *ip = VTOI(vp);
393 	int error;
394 
395 	if (cred->cr_uid != ip->i_uid &&
396 	    (error = suser(cred, &p->p_acflag)))
397 		return (error);
398 	if (cred->cr_uid) {
399 		if (vp->v_type != VDIR && (mode & S_ISTXT))
400 			return (EFTYPE);
401 		if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
402 			return (EPERM);
403 	}
404 	ip->i_mode &= ~ALLPERMS;
405 	ip->i_mode |= (mode & ALLPERMS);
406 	ip->i_flag |= IN_CHANGE;
407 	if ((vp->v_flag & VTEXT) && (ip->i_mode & S_ISTXT) == 0)
408 		(void) vnode_pager_uncache(vp);
409 	return (0);
410 }
411 
412 /*
413  * Perform chown operation on inode ip;
414  * inode must be locked prior to call.
415  */
416 static int
417 ufs_chown(vp, uid, gid, cred, p)
418 	register struct vnode *vp;
419 	uid_t uid;
420 	gid_t gid;
421 	struct ucred *cred;
422 	struct proc *p;
423 {
424 	register struct inode *ip = VTOI(vp);
425 	uid_t ouid;
426 	gid_t ogid;
427 	int error = 0;
428 #ifdef QUOTA
429 	register int i;
430 	long change;
431 #endif
432 
433 	if (uid == (uid_t)VNOVAL)
434 		uid = ip->i_uid;
435 	if (gid == (gid_t)VNOVAL)
436 		gid = ip->i_gid;
437 	/*
438 	 * If we don't own the file, are trying to change the owner
439 	 * of the file, or are not a member of the target group,
440 	 * the caller must be superuser or the call fails.
441 	 */
442 	if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
443 	    !groupmember((gid_t)gid, cred)) &&
444 	    (error = suser(cred, &p->p_acflag)))
445 		return (error);
446 	ogid = ip->i_gid;
447 	ouid = ip->i_uid;
448 #ifdef QUOTA
449 	if (error = getinoquota(ip))
450 		return (error);
451 	if (ouid == uid) {
452 		dqrele(vp, ip->i_dquot[USRQUOTA]);
453 		ip->i_dquot[USRQUOTA] = NODQUOT;
454 	}
455 	if (ogid == gid) {
456 		dqrele(vp, ip->i_dquot[GRPQUOTA]);
457 		ip->i_dquot[GRPQUOTA] = NODQUOT;
458 	}
459 	change = ip->i_blocks;
460 	(void) chkdq(ip, -change, cred, CHOWN);
461 	(void) chkiq(ip, -1, cred, CHOWN);
462 	for (i = 0; i < MAXQUOTAS; i++) {
463 		dqrele(vp, ip->i_dquot[i]);
464 		ip->i_dquot[i] = NODQUOT;
465 	}
466 #endif
467 	ip->i_gid = gid;
468 	ip->i_uid = uid;
469 #ifdef QUOTA
470 	if ((error = getinoquota(ip)) == 0) {
471 		if (ouid == uid) {
472 			dqrele(vp, ip->i_dquot[USRQUOTA]);
473 			ip->i_dquot[USRQUOTA] = NODQUOT;
474 		}
475 		if (ogid == gid) {
476 			dqrele(vp, ip->i_dquot[GRPQUOTA]);
477 			ip->i_dquot[GRPQUOTA] = NODQUOT;
478 		}
479 		if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
480 			if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
481 				goto good;
482 			else
483 				(void) chkdq(ip, -change, cred, CHOWN|FORCE);
484 		}
485 		for (i = 0; i < MAXQUOTAS; i++) {
486 			dqrele(vp, ip->i_dquot[i]);
487 			ip->i_dquot[i] = NODQUOT;
488 		}
489 	}
490 	ip->i_gid = ogid;
491 	ip->i_uid = ouid;
492 	if (getinoquota(ip) == 0) {
493 		if (ouid == uid) {
494 			dqrele(vp, ip->i_dquot[USRQUOTA]);
495 			ip->i_dquot[USRQUOTA] = NODQUOT;
496 		}
497 		if (ogid == gid) {
498 			dqrele(vp, ip->i_dquot[GRPQUOTA]);
499 			ip->i_dquot[GRPQUOTA] = NODQUOT;
500 		}
501 		(void) chkdq(ip, change, cred, FORCE|CHOWN);
502 		(void) chkiq(ip, 1, cred, FORCE|CHOWN);
503 		(void) getinoquota(ip);
504 	}
505 	return (error);
506 good:
507 	if (getinoquota(ip))
508 		panic("chown: lost quota");
509 #endif /* QUOTA */
510 	if (ouid != uid || ogid != gid)
511 		ip->i_flag |= IN_CHANGE;
512 	if (ouid != uid && cred->cr_uid != 0)
513 		ip->i_mode &= ~ISUID;
514 	if (ogid != gid && cred->cr_uid != 0)
515 		ip->i_mode &= ~ISGID;
516 	return (0);
517 }
518 
519 /* ARGSUSED */
520 int
521 ufs_ioctl(ap)
522 	struct vop_ioctl_args /* {
523 		struct vnode *a_vp;
524 		int  a_command;
525 		caddr_t  a_data;
526 		int  a_fflag;
527 		struct ucred *a_cred;
528 		struct proc *a_p;
529 	} */ *ap;
530 {
531 
532 	return (ENOTTY);
533 }
534 
535 /* ARGSUSED */
536 int
537 ufs_select(ap)
538 	struct vop_select_args /* {
539 		struct vnode *a_vp;
540 		int  a_which;
541 		int  a_fflags;
542 		struct ucred *a_cred;
543 		struct proc *a_p;
544 	} */ *ap;
545 {
546 
547 	/*
548 	 * We should really check to see if I/O is possible.
549 	 */
550 	return (1);
551 }
552 
553 /*
554  * Mmap a file
555  *
556  * NB Currently unsupported.
557  */
558 /* ARGSUSED */
559 int
560 ufs_mmap(ap)
561 	struct vop_mmap_args /* {
562 		struct vnode *a_vp;
563 		int  a_fflags;
564 		struct ucred *a_cred;
565 		struct proc *a_p;
566 	} */ *ap;
567 {
568 
569 	return (EINVAL);
570 }
571 
572 /*
573  * Seek on a file
574  *
575  * Nothing to do, so just return.
576  */
577 /* ARGSUSED */
578 int
579 ufs_seek(ap)
580 	struct vop_seek_args /* {
581 		struct vnode *a_vp;
582 		off_t  a_oldoff;
583 		off_t  a_newoff;
584 		struct ucred *a_cred;
585 	} */ *ap;
586 {
587 
588 	return (0);
589 }
590 
591 int
592 ufs_remove(ap)
593 	struct vop_remove_args /* {
594 		struct vnode *a_dvp;
595 		struct vnode *a_vp;
596 		struct componentname *a_cnp;
597 	} */ *ap;
598 {
599 	register struct inode *ip;
600 	register struct vnode *vp = ap->a_vp;
601 	register struct vnode *dvp = ap->a_dvp;
602 	int error;
603 
604 	ip = VTOI(vp);
605 	if ((ip->i_flags & (IMMUTABLE | APPEND)) ||
606 	    (VTOI(dvp)->i_flags & APPEND)) {
607 		error = EPERM;
608 		goto out;
609 	}
610 	if ((error = ufs_dirremove(dvp, ap->a_cnp)) == 0) {
611 		ip->i_nlink--;
612 		ip->i_flag |= IN_CHANGE;
613 	}
614 out:
615 	if (dvp == vp)
616 		vrele(vp);
617 	else
618 		vput(vp);
619 	vput(dvp);
620 	return (error);
621 }
622 
623 /*
624  * link vnode call
625  */
626 int
627 ufs_link(ap)
628 	struct vop_link_args /* {
629 		struct vnode *a_vp;
630 		struct vnode *a_tdvp;
631 		struct componentname *a_cnp;
632 	} */ *ap;
633 {
634 	register struct vnode *vp = ap->a_vp;
635 	register struct vnode *tdvp = ap->a_tdvp;
636 	register struct componentname *cnp = ap->a_cnp;
637 	register struct inode *ip;
638 	struct timeval tv;
639 	int error;
640 
641 #ifdef DIAGNOSTIC
642 	if ((cnp->cn_flags & HASBUF) == 0)
643 		panic("ufs_link: no name");
644 #endif
645 	if (vp->v_mount != tdvp->v_mount) {
646 		VOP_ABORTOP(vp, cnp);
647 		error = EXDEV;
648 		goto out2;
649 	}
650 	if (vp != tdvp && (error = VOP_LOCK(tdvp))) {
651 		VOP_ABORTOP(vp, cnp);
652 		goto out2;
653 	}
654 	ip = VTOI(tdvp);
655 	if ((nlink_t)ip->i_nlink >= LINK_MAX) {
656 		VOP_ABORTOP(vp, cnp);
657 		error = EMLINK;
658 		goto out1;
659 	}
660 	if (ip->i_flags & (IMMUTABLE | APPEND)) {
661 		VOP_ABORTOP(vp, cnp);
662 		error = EPERM;
663 		goto out1;
664 	}
665 	ip->i_nlink++;
666 	ip->i_flag |= IN_CHANGE;
667 	tv = time;
668 	error = VOP_UPDATE(tdvp, &tv, &tv, 1);
669 	if (!error)
670 		error = ufs_direnter(ip, vp, cnp);
671 	if (error) {
672 		ip->i_nlink--;
673 		ip->i_flag |= IN_CHANGE;
674 	}
675 	FREE(cnp->cn_pnbuf, M_NAMEI);
676 out1:
677 	if (vp != tdvp)
678 		VOP_UNLOCK(tdvp);
679 out2:
680 	vput(vp);
681 	return (error);
682 }
683 
684 
685 
686 /*
687  * relookup - lookup a path name component
688  *    Used by lookup to re-aquire things.
689  */
690 int
691 relookup(dvp, vpp, cnp)
692 	struct vnode *dvp, **vpp;
693 	struct componentname *cnp;
694 {
695 	register struct vnode *dp = 0;	/* the directory we are searching */
696 	int docache;			/* == 0 do not cache last component */
697 	int wantparent;			/* 1 => wantparent or lockparent flag */
698 	int rdonly;			/* lookup read-only flag bit */
699 	int error = 0;
700 #ifdef NAMEI_DIAGNOSTIC
701 	int newhash;			/* DEBUG: check name hash */
702 	char *cp;			/* DEBUG: check name ptr/len */
703 #endif
704 
705 	/*
706 	 * Setup: break out flag bits into variables.
707 	 */
708 	wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
709 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
710 	if (cnp->cn_nameiop == DELETE ||
711 	    (wantparent && cnp->cn_nameiop != CREATE))
712 		docache = 0;
713 	rdonly = cnp->cn_flags & RDONLY;
714 	cnp->cn_flags &= ~ISSYMLINK;
715 	dp = dvp;
716 	VOP_LOCK(dp);
717 
718 /* dirloop: */
719 	/*
720 	 * Search a new directory.
721 	 *
722 	 * The cn_hash value is for use by vfs_cache.
723 	 * The last component of the filename is left accessible via
724 	 * cnp->cn_nameptr for callers that need the name. Callers needing
725 	 * the name set the SAVENAME flag. When done, they assume
726 	 * responsibility for freeing the pathname buffer.
727 	 */
728 #ifdef NAMEI_DIAGNOSTIC
729 	for (newhash = 0, cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
730 		newhash += (unsigned char)*cp;
731 	if (newhash != cnp->cn_hash)
732 		panic("relookup: bad hash");
733 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
734 		panic ("relookup: bad len");
735 	if (*cp != 0)
736 		panic("relookup: not last component");
737 	printf("{%s}: ", cnp->cn_nameptr);
738 #endif
739 
740 	/*
741 	 * Check for degenerate name (e.g. / or "")
742 	 * which is a way of talking about a directory,
743 	 * e.g. like "/." or ".".
744 	 */
745 	if (cnp->cn_nameptr[0] == '\0') {
746 		if (cnp->cn_nameiop != LOOKUP || wantparent) {
747 			error = EISDIR;
748 			goto bad;
749 		}
750 		if (dp->v_type != VDIR) {
751 			error = ENOTDIR;
752 			goto bad;
753 		}
754 		if (!(cnp->cn_flags & LOCKLEAF))
755 			VOP_UNLOCK(dp);
756 		*vpp = dp;
757 		if (cnp->cn_flags & SAVESTART)
758 			panic("lookup: SAVESTART");
759 		return (0);
760 	}
761 
762 	if (cnp->cn_flags & ISDOTDOT)
763 		panic ("relookup: lookup on dot-dot");
764 
765 	/*
766 	 * We now have a segment name to search for, and a directory to search.
767 	 */
768 	if (error = VOP_LOOKUP(dp, vpp, cnp)) {
769 #ifdef DIAGNOSTIC
770 		if (*vpp != NULL)
771 			panic("leaf should be empty");
772 #endif
773 		if (error != EJUSTRETURN)
774 			goto bad;
775 		/*
776 		 * If creating and at end of pathname, then can consider
777 		 * allowing file to be created.
778 		 */
779 		if (rdonly || (dvp->v_mount->mnt_flag & MNT_RDONLY)) {
780 			error = EROFS;
781 			goto bad;
782 		}
783 		/* ASSERT(dvp == ndp->ni_startdir) */
784 		if (cnp->cn_flags & SAVESTART)
785 			VREF(dvp);
786 		/*
787 		 * We return with ni_vp NULL to indicate that the entry
788 		 * doesn't currently exist, leaving a pointer to the
789 		 * (possibly locked) directory inode in ndp->ni_dvp.
790 		 */
791 		return (0);
792 	}
793 	dp = *vpp;
794 
795 #ifdef DIAGNOSTIC
796 	/*
797 	 * Check for symbolic link
798 	 */
799 	if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
800 		panic ("relookup: symlink found.\n");
801 #endif
802 
803 	/*
804 	 * Check for read-only file systems.
805 	 */
806 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME) {
807 		/*
808 		 * Disallow directory write attempts on read-only
809 		 * file systems.
810 		 */
811 		if (rdonly || (dp->v_mount->mnt_flag & MNT_RDONLY) ||
812 		    (wantparent &&
813 		     (dvp->v_mount->mnt_flag & MNT_RDONLY))) {
814 			error = EROFS;
815 			goto bad2;
816 		}
817 	}
818 	/* ASSERT(dvp == ndp->ni_startdir) */
819 	if (cnp->cn_flags & SAVESTART)
820 		VREF(dvp);
821 
822 	if (!wantparent)
823 		vrele(dvp);
824 	if ((cnp->cn_flags & LOCKLEAF) == 0)
825 		VOP_UNLOCK(dp);
826 	return (0);
827 
828 bad2:
829 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
830 		VOP_UNLOCK(dvp);
831 	vrele(dvp);
832 bad:
833 	vput(dp);
834 	*vpp = NULL;
835 	return (error);
836 }
837 
838 
839 /*
840  * Rename system call.
841  * 	rename("foo", "bar");
842  * is essentially
843  *	unlink("bar");
844  *	link("foo", "bar");
845  *	unlink("foo");
846  * but ``atomically''.  Can't do full commit without saving state in the
847  * inode on disk which isn't feasible at this time.  Best we can do is
848  * always guarantee the target exists.
849  *
850  * Basic algorithm is:
851  *
852  * 1) Bump link count on source while we're linking it to the
853  *    target.  This also ensure the inode won't be deleted out
854  *    from underneath us while we work (it may be truncated by
855  *    a concurrent `trunc' or `open' for creation).
856  * 2) Link source to destination.  If destination already exists,
857  *    delete it first.
858  * 3) Unlink source reference to inode if still around. If a
859  *    directory was moved and the parent of the destination
860  *    is different from the source, patch the ".." entry in the
861  *    directory.
862  */
863 int
864 ufs_rename(ap)
865 	struct vop_rename_args  /* {
866 		struct vnode *a_fdvp;
867 		struct vnode *a_fvp;
868 		struct componentname *a_fcnp;
869 		struct vnode *a_tdvp;
870 		struct vnode *a_tvp;
871 		struct componentname *a_tcnp;
872 	} */ *ap;
873 {
874 	struct vnode *tvp = ap->a_tvp;
875 	register struct vnode *tdvp = ap->a_tdvp;
876 	struct vnode *fvp = ap->a_fvp;
877 	register struct vnode *fdvp = ap->a_fdvp;
878 	register struct componentname *tcnp = ap->a_tcnp;
879 	register struct componentname *fcnp = ap->a_fcnp;
880 	register struct inode *ip, *xp, *dp;
881 	struct dirtemplate dirbuf;
882 	struct timeval tv;
883 	int doingdirectory = 0, oldparent = 0, newparent = 0;
884 	int error = 0;
885 	u_char namlen;
886 
887 #ifdef DIAGNOSTIC
888 	if ((tcnp->cn_flags & HASBUF) == 0 ||
889 	    (fcnp->cn_flags & HASBUF) == 0)
890 		panic("ufs_rename: no name");
891 #endif
892 	/*
893 	 * Check for cross-device rename.
894 	 */
895 	if ((fvp->v_mount != tdvp->v_mount) ||
896 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
897 		error = EXDEV;
898 abortit:
899 		VOP_ABORTOP(tdvp, tcnp); /* XXX, why not in NFS? */
900 		if (tdvp == tvp)
901 			vrele(tdvp);
902 		else
903 			vput(tdvp);
904 		if (tvp)
905 			vput(tvp);
906 		VOP_ABORTOP(fdvp, fcnp); /* XXX, why not in NFS? */
907 		vrele(fdvp);
908 		vrele(fvp);
909 		return (error);
910 	}
911 
912 	/*
913 	 * Check if just deleting a link name.
914 	 */
915 	if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) ||
916 	    (VTOI(tdvp)->i_flags & APPEND))) {
917 		error = EPERM;
918 		goto abortit;
919 	}
920 	if (fvp == tvp) {
921 		if (fvp->v_type == VDIR) {
922 			error = EINVAL;
923 			goto abortit;
924 		}
925 		VOP_ABORTOP(fdvp, fcnp);
926 		vrele(fdvp);
927 		vrele(fvp);
928 		vput(tdvp);
929 		vput(tvp);
930 		tcnp->cn_flags &= ~MODMASK;
931 		tcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
932 		if ((tcnp->cn_flags & SAVESTART) == 0)
933 			panic("ufs_rename: lost from startdir");
934 		tcnp->cn_nameiop = DELETE;
935 		(void) relookup(tdvp, &tvp, tcnp);
936 		return (VOP_REMOVE(tdvp, tvp, tcnp));
937 	}
938 	if (error = VOP_LOCK(fvp))
939 		goto abortit;
940 	dp = VTOI(fdvp);
941 	ip = VTOI(fvp);
942 	if ((ip->i_flags & (IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) {
943 		VOP_UNLOCK(fvp);
944 		error = EPERM;
945 		goto abortit;
946 	}
947 	if ((ip->i_mode & IFMT) == IFDIR) {
948 		/*
949 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
950 		 */
951 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
952 		    dp == ip || (fcnp->cn_flags&ISDOTDOT) ||
953 		    (ip->i_flag & IN_RENAME)) {
954 			VOP_UNLOCK(fvp);
955 			error = EINVAL;
956 			goto abortit;
957 		}
958 		ip->i_flag |= IN_RENAME;
959 		oldparent = dp->i_number;
960 		doingdirectory++;
961 	}
962 	vrele(fdvp);
963 
964 	/*
965 	 * When the target exists, both the directory
966 	 * and target vnodes are returned locked.
967 	 */
968 	dp = VTOI(tdvp);
969 	xp = NULL;
970 	if (tvp)
971 		xp = VTOI(tvp);
972 
973 	/*
974 	 * 1) Bump link count while we're moving stuff
975 	 *    around.  If we crash somewhere before
976 	 *    completing our work, the link count
977 	 *    may be wrong, but correctable.
978 	 */
979 	ip->i_nlink++;
980 	ip->i_flag |= IN_CHANGE;
981 	tv = time;
982 	if (error = VOP_UPDATE(fvp, &tv, &tv, 1)) {
983 		VOP_UNLOCK(fvp);
984 		goto bad;
985 	}
986 
987 	/*
988 	 * If ".." must be changed (ie the directory gets a new
989 	 * parent) then the source directory must not be in the
990 	 * directory heirarchy above the target, as this would
991 	 * orphan everything below the source directory. Also
992 	 * the user must have write permission in the source so
993 	 * as to be able to change "..". We must repeat the call
994 	 * to namei, as the parent directory is unlocked by the
995 	 * call to checkpath().
996 	 */
997 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_proc);
998 	VOP_UNLOCK(fvp);
999 	if (oldparent != dp->i_number)
1000 		newparent = dp->i_number;
1001 	if (doingdirectory && newparent) {
1002 		if (error)	/* write access check above */
1003 			goto bad;
1004 		if (xp != NULL)
1005 			vput(tvp);
1006 		if (error = ufs_checkpath(ip, dp, tcnp->cn_cred))
1007 			goto out;
1008 		if ((tcnp->cn_flags & SAVESTART) == 0)
1009 			panic("ufs_rename: lost to startdir");
1010 		if (error = relookup(tdvp, &tvp, tcnp))
1011 			goto out;
1012 		dp = VTOI(tdvp);
1013 		xp = NULL;
1014 		if (tvp)
1015 			xp = VTOI(tvp);
1016 	}
1017 	/*
1018 	 * 2) If target doesn't exist, link the target
1019 	 *    to the source and unlink the source.
1020 	 *    Otherwise, rewrite the target directory
1021 	 *    entry to reference the source inode and
1022 	 *    expunge the original entry's existence.
1023 	 */
1024 	if (xp == NULL) {
1025 		if (dp->i_dev != ip->i_dev)
1026 			panic("rename: EXDEV");
1027 		/*
1028 		 * Account for ".." in new directory.
1029 		 * When source and destination have the same
1030 		 * parent we don't fool with the link count.
1031 		 */
1032 		if (doingdirectory && newparent) {
1033 			if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1034 				error = EMLINK;
1035 				goto bad;
1036 			}
1037 			dp->i_nlink++;
1038 			dp->i_flag |= IN_CHANGE;
1039 			if (error = VOP_UPDATE(tdvp, &tv, &tv, 1))
1040 				goto bad;
1041 		}
1042 		if (error = ufs_direnter(ip, tdvp, tcnp)) {
1043 			if (doingdirectory && newparent) {
1044 				dp->i_nlink--;
1045 				dp->i_flag |= IN_CHANGE;
1046 				(void)VOP_UPDATE(tdvp, &tv, &tv, 1);
1047 			}
1048 			goto bad;
1049 		}
1050 		vput(tdvp);
1051 	} else {
1052 		if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1053 			panic("rename: EXDEV");
1054 		/*
1055 		 * Short circuit rename(foo, foo).
1056 		 */
1057 		if (xp->i_number == ip->i_number)
1058 			panic("rename: same file");
1059 		/*
1060 		 * If the parent directory is "sticky", then the user must
1061 		 * own the parent directory, or the destination of the rename,
1062 		 * otherwise the destination may not be changed (except by
1063 		 * root). This implements append-only directories.
1064 		 */
1065 		if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1066 		    tcnp->cn_cred->cr_uid != dp->i_uid &&
1067 		    xp->i_uid != tcnp->cn_cred->cr_uid) {
1068 			error = EPERM;
1069 			goto bad;
1070 		}
1071 		/*
1072 		 * Target must be empty if a directory and have no links
1073 		 * to it. Also, ensure source and target are compatible
1074 		 * (both directories, or both not directories).
1075 		 */
1076 		if ((xp->i_mode&IFMT) == IFDIR) {
1077 			if (!ufs_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
1078 			    xp->i_nlink > 2) {
1079 				error = ENOTEMPTY;
1080 				goto bad;
1081 			}
1082 			if (!doingdirectory) {
1083 				error = ENOTDIR;
1084 				goto bad;
1085 			}
1086 			cache_purge(tdvp);
1087 		} else if (doingdirectory) {
1088 			error = EISDIR;
1089 			goto bad;
1090 		}
1091 		if (error = ufs_dirrewrite(dp, ip, tcnp))
1092 			goto bad;
1093 		/*
1094 		 * If the target directory is in the same
1095 		 * directory as the source directory,
1096 		 * decrement the link count on the parent
1097 		 * of the target directory.
1098 		 */
1099 		 if (doingdirectory && !newparent) {
1100 			dp->i_nlink--;
1101 			dp->i_flag |= IN_CHANGE;
1102 		}
1103 		vput(tdvp);
1104 		/*
1105 		 * Adjust the link count of the target to
1106 		 * reflect the dirrewrite above.  If this is
1107 		 * a directory it is empty and there are
1108 		 * no links to it, so we can squash the inode and
1109 		 * any space associated with it.  We disallowed
1110 		 * renaming over top of a directory with links to
1111 		 * it above, as the remaining link would point to
1112 		 * a directory without "." or ".." entries.
1113 		 */
1114 		xp->i_nlink--;
1115 		if (doingdirectory) {
1116 			if (--xp->i_nlink != 0)
1117 				panic("rename: linked directory");
1118 			error = VOP_TRUNCATE(tvp, (off_t)0, IO_SYNC,
1119 			    tcnp->cn_cred, tcnp->cn_proc);
1120 		}
1121 		xp->i_flag |= IN_CHANGE;
1122 		vput(tvp);
1123 		xp = NULL;
1124 	}
1125 
1126 	/*
1127 	 * 3) Unlink the source.
1128 	 */
1129 	fcnp->cn_flags &= ~MODMASK;
1130 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1131 	if ((fcnp->cn_flags & SAVESTART) == 0)
1132 		panic("ufs_rename: lost from startdir");
1133 	(void) relookup(fdvp, &fvp, fcnp);
1134 	if (fvp != NULL) {
1135 		xp = VTOI(fvp);
1136 		dp = VTOI(fdvp);
1137 	} else {
1138 		/*
1139 		 * From name has disappeared.
1140 		 */
1141 		if (doingdirectory)
1142 			panic("rename: lost dir entry");
1143 		vrele(ap->a_fvp);
1144 		return (0);
1145 	}
1146 	/*
1147 	 * Ensure that the directory entry still exists and has not
1148 	 * changed while the new name has been entered. If the source is
1149 	 * a file then the entry may have been unlinked or renamed. In
1150 	 * either case there is no further work to be done. If the source
1151 	 * is a directory then it cannot have been rmdir'ed; its link
1152 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1153 	 * The IRENAME flag ensures that it cannot be moved by another
1154 	 * rename.
1155 	 */
1156 	if (xp != ip) {
1157 		if (doingdirectory)
1158 			panic("rename: lost dir entry");
1159 	} else {
1160 		/*
1161 		 * If the source is a directory with a
1162 		 * new parent, the link count of the old
1163 		 * parent directory must be decremented
1164 		 * and ".." set to point to the new parent.
1165 		 */
1166 		if (doingdirectory && newparent) {
1167 			dp->i_nlink--;
1168 			dp->i_flag |= IN_CHANGE;
1169 			error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1170 				sizeof (struct dirtemplate), (off_t)0,
1171 				UIO_SYSSPACE, IO_NODELOCKED,
1172 				tcnp->cn_cred, (int *)0, (struct proc *)0);
1173 			if (error == 0) {
1174 #				if (BYTE_ORDER == LITTLE_ENDIAN)
1175 					if (fvp->v_mount->mnt_maxsymlinklen <= 0)
1176 						namlen = dirbuf.dotdot_type;
1177 					else
1178 						namlen = dirbuf.dotdot_namlen;
1179 #				else
1180 					namlen = dirbuf.dotdot_namlen;
1181 #				endif
1182 				if (namlen != 2 ||
1183 				    dirbuf.dotdot_name[0] != '.' ||
1184 				    dirbuf.dotdot_name[1] != '.') {
1185 					ufs_dirbad(xp, (doff_t)12,
1186 					    "rename: mangled dir");
1187 				} else {
1188 					dirbuf.dotdot_ino = newparent;
1189 					(void) vn_rdwr(UIO_WRITE, fvp,
1190 					    (caddr_t)&dirbuf,
1191 					    sizeof (struct dirtemplate),
1192 					    (off_t)0, UIO_SYSSPACE,
1193 					    IO_NODELOCKED|IO_SYNC,
1194 					    tcnp->cn_cred, (int *)0,
1195 					    (struct proc *)0);
1196 					cache_purge(fdvp);
1197 				}
1198 			}
1199 		}
1200 		error = ufs_dirremove(fdvp, fcnp);
1201 		if (!error) {
1202 			xp->i_nlink--;
1203 			xp->i_flag |= IN_CHANGE;
1204 		}
1205 		xp->i_flag &= ~IN_RENAME;
1206 	}
1207 	if (dp)
1208 		vput(fdvp);
1209 	if (xp)
1210 		vput(fvp);
1211 	vrele(ap->a_fvp);
1212 	return (error);
1213 
1214 bad:
1215 	if (xp)
1216 		vput(ITOV(xp));
1217 	vput(ITOV(dp));
1218 out:
1219 	if (VOP_LOCK(fvp) == 0) {
1220 		ip->i_nlink--;
1221 		ip->i_flag |= IN_CHANGE;
1222 		vput(fvp);
1223 	} else
1224 		vrele(fvp);
1225 	return (error);
1226 }
1227 
1228 /*
1229  * A virgin directory (no blushing please).
1230  */
1231 static struct dirtemplate mastertemplate = {
1232 	0, 12, DT_DIR, 1, ".",
1233 	0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
1234 };
1235 static struct odirtemplate omastertemplate = {
1236 	0, 12, 1, ".",
1237 	0, DIRBLKSIZ - 12, 2, ".."
1238 };
1239 
1240 /*
1241  * Mkdir system call
1242  */
1243 int
1244 ufs_mkdir(ap)
1245 	struct vop_mkdir_args /* {
1246 		struct vnode *a_dvp;
1247 		struct vnode **a_vpp;
1248 		struct componentname *a_cnp;
1249 		struct vattr *a_vap;
1250 	} */ *ap;
1251 {
1252 	register struct vnode *dvp = ap->a_dvp;
1253 	register struct vattr *vap = ap->a_vap;
1254 	register struct componentname *cnp = ap->a_cnp;
1255 	register struct inode *ip, *dp;
1256 	struct vnode *tvp;
1257 	struct dirtemplate dirtemplate, *dtp;
1258 	struct timeval tv;
1259 	int error, dmode;
1260 
1261 #ifdef DIAGNOSTIC
1262 	if ((cnp->cn_flags & HASBUF) == 0)
1263 		panic("ufs_mkdir: no name");
1264 #endif
1265 	dp = VTOI(dvp);
1266 	if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1267 		error = EMLINK;
1268 		goto out;
1269 	}
1270 	dmode = vap->va_mode & 0777;
1271 	dmode |= IFDIR;
1272 	/*
1273 	 * Must simulate part of ufs_makeinode here to acquire the inode,
1274 	 * but not have it entered in the parent directory. The entry is
1275 	 * made later after writing "." and ".." entries.
1276 	 */
1277 	if (error = VOP_VALLOC(dvp, dmode, cnp->cn_cred, &tvp))
1278 		goto out;
1279 	ip = VTOI(tvp);
1280 	ip->i_uid = cnp->cn_cred->cr_uid;
1281 	ip->i_gid = dp->i_gid;
1282 #ifdef QUOTA
1283 	if ((error = getinoquota(ip)) ||
1284 	    (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
1285 		free(cnp->cn_pnbuf, M_NAMEI);
1286 		VOP_VFREE(tvp, ip->i_number, dmode);
1287 		vput(tvp);
1288 		vput(dvp);
1289 		return (error);
1290 	}
1291 #endif
1292 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1293 	ip->i_mode = dmode;
1294 	tvp->v_type = VDIR;	/* Rest init'd in getnewvnode(). */
1295 	ip->i_nlink = 2;
1296 	tv = time;
1297 	error = VOP_UPDATE(tvp, &tv, &tv, 1);
1298 
1299 	/*
1300 	 * Bump link count in parent directory
1301 	 * to reflect work done below.  Should
1302 	 * be done before reference is created
1303 	 * so reparation is possible if we crash.
1304 	 */
1305 	dp->i_nlink++;
1306 	dp->i_flag |= IN_CHANGE;
1307 	if (error = VOP_UPDATE(dvp, &tv, &tv, 1))
1308 		goto bad;
1309 
1310 	/* Initialize directory with "." and ".." from static template. */
1311 	if (dvp->v_mount->mnt_maxsymlinklen > 0)
1312 		dtp = &mastertemplate;
1313 	else
1314 		dtp = (struct dirtemplate *)&omastertemplate;
1315 	dirtemplate = *dtp;
1316 	dirtemplate.dot_ino = ip->i_number;
1317 	dirtemplate.dotdot_ino = dp->i_number;
1318 	error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1319 	    sizeof (dirtemplate), (off_t)0, UIO_SYSSPACE,
1320 	    IO_NODELOCKED|IO_SYNC, cnp->cn_cred, (int *)0, (struct proc *)0);
1321 	if (error) {
1322 		dp->i_nlink--;
1323 		dp->i_flag |= IN_CHANGE;
1324 		goto bad;
1325 	}
1326 	if (DIRBLKSIZ > VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1327 		panic("ufs_mkdir: blksize"); /* XXX should grow with balloc() */
1328 	else {
1329 		ip->i_size = DIRBLKSIZ;
1330 		ip->i_flag |= IN_CHANGE;
1331 	}
1332 
1333 	/* Directory set up, now install it's entry in the parent directory. */
1334 	if (error = ufs_direnter(ip, dvp, cnp)) {
1335 		dp->i_nlink--;
1336 		dp->i_flag |= IN_CHANGE;
1337 	}
1338 bad:
1339 	/*
1340 	 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1341 	 * for us because we set the link count to 0.
1342 	 */
1343 	if (error) {
1344 		ip->i_nlink = 0;
1345 		ip->i_flag |= IN_CHANGE;
1346 		vput(tvp);
1347 	} else
1348 		*ap->a_vpp = tvp;
1349 out:
1350 	FREE(cnp->cn_pnbuf, M_NAMEI);
1351 	vput(dvp);
1352 	return (error);
1353 }
1354 
1355 /*
1356  * Rmdir system call.
1357  */
1358 int
1359 ufs_rmdir(ap)
1360 	struct vop_rmdir_args /* {
1361 		struct vnode *a_dvp;
1362 		struct vnode *a_vp;
1363 		struct componentname *a_cnp;
1364 	} */ *ap;
1365 {
1366 	register struct vnode *vp = ap->a_vp;
1367 	register struct vnode *dvp = ap->a_dvp;
1368 	register struct componentname *cnp = ap->a_cnp;
1369 	register struct inode *ip, *dp;
1370 	int error;
1371 
1372 	ip = VTOI(vp);
1373 	dp = VTOI(dvp);
1374 	/*
1375 	 * No rmdir "." please.
1376 	 */
1377 	if (dp == ip) {
1378 		vrele(dvp);
1379 		vput(vp);
1380 		return (EINVAL);
1381 	}
1382 	/*
1383 	 * Verify the directory is empty (and valid).
1384 	 * (Rmdir ".." won't be valid since
1385 	 *  ".." will contain a reference to
1386 	 *  the current directory and thus be
1387 	 *  non-empty.)
1388 	 */
1389 	error = 0;
1390 	if (ip->i_nlink != 2 ||
1391 	    !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1392 		error = ENOTEMPTY;
1393 		goto out;
1394 	}
1395 	if ((dp->i_flags & APPEND) || (ip->i_flags & (IMMUTABLE | APPEND))) {
1396 		error = EPERM;
1397 		goto out;
1398 	}
1399 	/*
1400 	 * Delete reference to directory before purging
1401 	 * inode.  If we crash in between, the directory
1402 	 * will be reattached to lost+found,
1403 	 */
1404 	if (error = ufs_dirremove(dvp, cnp))
1405 		goto out;
1406 	dp->i_nlink--;
1407 	dp->i_flag |= IN_CHANGE;
1408 	cache_purge(dvp);
1409 	vput(dvp);
1410 	dvp = NULL;
1411 	/*
1412 	 * Truncate inode.  The only stuff left
1413 	 * in the directory is "." and "..".  The
1414 	 * "." reference is inconsequential since
1415 	 * we're quashing it.  The ".." reference
1416 	 * has already been adjusted above.  We've
1417 	 * removed the "." reference and the reference
1418 	 * in the parent directory, but there may be
1419 	 * other hard links so decrement by 2 and
1420 	 * worry about them later.
1421 	 */
1422 	ip->i_nlink -= 2;
1423 	error = VOP_TRUNCATE(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1424 	    cnp->cn_proc);
1425 	cache_purge(ITOV(ip));
1426 out:
1427 	if (dvp)
1428 		vput(dvp);
1429 	vput(vp);
1430 	return (error);
1431 }
1432 
1433 /*
1434  * symlink -- make a symbolic link
1435  */
1436 int
1437 ufs_symlink(ap)
1438 	struct vop_symlink_args /* {
1439 		struct vnode *a_dvp;
1440 		struct vnode **a_vpp;
1441 		struct componentname *a_cnp;
1442 		struct vattr *a_vap;
1443 		char *a_target;
1444 	} */ *ap;
1445 {
1446 	register struct vnode *vp, **vpp = ap->a_vpp;
1447 	register struct inode *ip;
1448 	int len, error;
1449 
1450 	if (error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1451 	    vpp, ap->a_cnp))
1452 		return (error);
1453 	vp = *vpp;
1454 	len = strlen(ap->a_target);
1455 	if (len < vp->v_mount->mnt_maxsymlinklen) {
1456 		ip = VTOI(vp);
1457 		bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1458 		ip->i_size = len;
1459 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1460 	} else
1461 		error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1462 		    UIO_SYSSPACE, IO_NODELOCKED, ap->a_cnp->cn_cred, (int *)0,
1463 		    (struct proc *)0);
1464 	vput(vp);
1465 	return (error);
1466 }
1467 
1468 /*
1469  * Vnode op for reading directories.
1470  *
1471  * The routine below assumes that the on-disk format of a directory
1472  * is the same as that defined by <sys/dirent.h>. If the on-disk
1473  * format changes, then it will be necessary to do a conversion
1474  * from the on-disk format that read returns to the format defined
1475  * by <sys/dirent.h>.
1476  */
1477 int
1478 ufs_readdir(ap)
1479 	struct vop_readdir_args /* {
1480 		struct vnode *a_vp;
1481 		struct uio *a_uio;
1482 		struct ucred *a_cred;
1483 		int *a_eofflag;
1484 		u_long *a_cookies;
1485 		int ncookies;
1486 	} */ *ap;
1487 {
1488 	register struct uio *uio = ap->a_uio;
1489 	int error;
1490 	size_t count, lost;
1491 	off_t off = uio->uio_offset;
1492 
1493 	count = uio->uio_resid;
1494 	/* Make sure we don't return partial entries. */
1495 	count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
1496 	if (count <= 0)
1497 		return (EINVAL);
1498 	lost = uio->uio_resid - count;
1499 	uio->uio_resid = count;
1500 	uio->uio_iov->iov_len = count;
1501 #	if (BYTE_ORDER == LITTLE_ENDIAN)
1502 		if (ap->a_vp->v_mount->mnt_maxsymlinklen > 0) {
1503 			error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1504 		} else {
1505 			struct dirent *dp, *edp;
1506 			struct uio auio;
1507 			struct iovec aiov;
1508 			caddr_t dirbuf;
1509 			int readcnt;
1510 			u_char tmp;
1511 
1512 			auio = *uio;
1513 			auio.uio_iov = &aiov;
1514 			auio.uio_iovcnt = 1;
1515 			auio.uio_segflg = UIO_SYSSPACE;
1516 			aiov.iov_len = count;
1517 			MALLOC(dirbuf, caddr_t, count, M_TEMP, M_WAITOK);
1518 			aiov.iov_base = dirbuf;
1519 			error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
1520 			if (error == 0) {
1521 				readcnt = count - auio.uio_resid;
1522 				edp = (struct dirent *)&dirbuf[readcnt];
1523 				for (dp = (struct dirent *)dirbuf; dp < edp; ) {
1524 					tmp = dp->d_namlen;
1525 					dp->d_namlen = dp->d_type;
1526 					dp->d_type = tmp;
1527 					if (dp->d_reclen > 0) {
1528 						dp = (struct dirent *)
1529 						    ((char *)dp + dp->d_reclen);
1530 					} else {
1531 						error = EIO;
1532 						break;
1533 					}
1534 				}
1535 				if (dp >= edp)
1536 					error = uiomove(dirbuf, readcnt, uio);
1537 			}
1538 			FREE(dirbuf, M_TEMP);
1539 		}
1540 #	else
1541 		error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1542 #	endif
1543 	if (!error && ap->a_ncookies) {
1544 		register struct dirent *dp;
1545 		register u_long *cookies = ap->a_cookies;
1546 		register int ncookies = ap->a_ncookies;
1547 
1548 		/*
1549 		 * Only the NFS server uses cookies, and it loads the
1550 		 * directory block into system space, so we can just look at
1551 		 * it directly.
1552 		 */
1553 		if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1554 			panic("ufs_readdir: lost in space");
1555 		dp = (struct dirent *)
1556 		     (uio->uio_iov->iov_base - (uio->uio_offset - off));
1557 		while (ncookies-- && off < uio->uio_offset) {
1558 			if (dp->d_reclen == 0)
1559 				break;
1560 			off += dp->d_reclen;
1561 			*(cookies++) = off;
1562 			dp = (struct dirent *)((caddr_t)dp + dp->d_reclen);
1563 		}
1564 		lost += uio->uio_offset - off;
1565 		uio->uio_offset = off;
1566 	}
1567 	uio->uio_resid += lost;
1568 	*ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
1569 	return (error);
1570 }
1571 
1572 /*
1573  * Return target name of a symbolic link
1574  */
1575 int
1576 ufs_readlink(ap)
1577 	struct vop_readlink_args /* {
1578 		struct vnode *a_vp;
1579 		struct uio *a_uio;
1580 		struct ucred *a_cred;
1581 	} */ *ap;
1582 {
1583 	register struct vnode *vp = ap->a_vp;
1584 	register struct inode *ip = VTOI(vp);
1585 	int isize;
1586 
1587 	isize = ip->i_size;
1588 	if (isize < vp->v_mount->mnt_maxsymlinklen) {
1589 		uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1590 		return (0);
1591 	}
1592 	return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1593 }
1594 
1595 /*
1596  * Ufs abort op, called after namei() when a CREATE/DELETE isn't actually
1597  * done. If a buffer has been saved in anticipation of a CREATE, delete it.
1598  */
1599 /* ARGSUSED */
1600 int
1601 ufs_abortop(ap)
1602 	struct vop_abortop_args /* {
1603 		struct vnode *a_dvp;
1604 		struct componentname *a_cnp;
1605 	} */ *ap;
1606 {
1607 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
1608 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
1609 	return (0);
1610 }
1611 
1612 /*
1613  * Lock an inode. If its already locked, set the WANT bit and sleep.
1614  */
1615 int
1616 ufs_lock(ap)
1617 	struct vop_lock_args /* {
1618 		struct vnode *a_vp;
1619 	} */ *ap;
1620 {
1621 	register struct vnode *vp = ap->a_vp;
1622 	register struct inode *ip;
1623 	struct proc *p = curproc;	/* XXX */
1624 
1625 start:
1626 	while (vp->v_flag & VXLOCK) {
1627 		vp->v_flag |= VXWANT;
1628 		sleep((caddr_t)vp, PINOD);
1629 	}
1630 	if (vp->v_tag == VT_NON)
1631 		return (ENOENT);
1632 	ip = VTOI(vp);
1633 	if (ip->i_flag & IN_LOCKED) {
1634 		ip->i_flag |= IN_WANTED;
1635 #ifdef DIAGNOSTIC
1636 		if (p) {
1637 			if (p->p_pid == ip->i_lockholder)
1638 				panic("locking against myself");
1639 			ip->i_lockwaiter = p->p_pid;
1640 		} else
1641 			ip->i_lockwaiter = -1;
1642 #endif
1643 		(void) sleep((caddr_t)ip, PINOD);
1644 		goto start;
1645 	}
1646 #ifdef DIAGNOSTIC
1647 	ip->i_lockwaiter = 0;
1648 	if (ip->i_lockholder != 0)
1649 		panic("lockholder (%d) != 0", ip->i_lockholder);
1650 	if (p && p->p_pid == 0)
1651 		printf("locking by process 0\n");
1652 	if (p)
1653 		ip->i_lockholder = p->p_pid;
1654 	else
1655 		ip->i_lockholder = -1;
1656 #endif
1657 	ip->i_flag |= IN_LOCKED;
1658 	return (0);
1659 }
1660 
1661 /*
1662  * Unlock an inode.  If WANT bit is on, wakeup.
1663  */
1664 int lockcount = 90;
1665 int
1666 ufs_unlock(ap)
1667 	struct vop_unlock_args /* {
1668 		struct vnode *a_vp;
1669 	} */ *ap;
1670 {
1671 	register struct inode *ip = VTOI(ap->a_vp);
1672 	struct proc *p = curproc;	/* XXX */
1673 
1674 #ifdef DIAGNOSTIC
1675 	if ((ip->i_flag & IN_LOCKED) == 0) {
1676 		vprint("ufs_unlock: unlocked inode", ap->a_vp);
1677 		panic("ufs_unlock NOT LOCKED");
1678 	}
1679 	if (p && p->p_pid != ip->i_lockholder && p->p_pid > -1 &&
1680 	    ip->i_lockholder > -1 && lockcount++ < 100)
1681 		panic("unlocker (%d) != lock holder (%d)",
1682 		    p->p_pid, ip->i_lockholder);
1683 	ip->i_lockholder = 0;
1684 #endif
1685 	ip->i_flag &= ~IN_LOCKED;
1686 	if (ip->i_flag & IN_WANTED) {
1687 		ip->i_flag &= ~IN_WANTED;
1688 		wakeup((caddr_t)ip);
1689 	}
1690 	return (0);
1691 }
1692 
1693 /*
1694  * Check for a locked inode.
1695  */
1696 int
1697 ufs_islocked(ap)
1698 	struct vop_islocked_args /* {
1699 		struct vnode *a_vp;
1700 	} */ *ap;
1701 {
1702 
1703 	if (VTOI(ap->a_vp)->i_flag & IN_LOCKED)
1704 		return (1);
1705 	return (0);
1706 }
1707 
1708 /*
1709  * Calculate the logical to physical mapping if not done already,
1710  * then call the device strategy routine.
1711  */
1712 int
1713 ufs_strategy(ap)
1714 	struct vop_strategy_args /* {
1715 		struct buf *a_bp;
1716 	} */ *ap;
1717 {
1718 	register struct buf *bp = ap->a_bp;
1719 	register struct vnode *vp = bp->b_vp;
1720 	register struct inode *ip;
1721 	int error;
1722 
1723 	ip = VTOI(vp);
1724 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1725 		panic("ufs_strategy: spec");
1726 	if (bp->b_blkno == bp->b_lblkno) {
1727 		if (error =
1728 		    VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL)) {
1729 			bp->b_error = error;
1730 			bp->b_flags |= B_ERROR;
1731 			biodone(bp);
1732 			return (error);
1733 		}
1734 		if ((long)bp->b_blkno == -1)
1735 			clrbuf(bp);
1736 	}
1737 	if ((long)bp->b_blkno == -1) {
1738 		biodone(bp);
1739 		return (0);
1740 	}
1741 	vp = ip->i_devvp;
1742 	bp->b_dev = vp->v_rdev;
1743 	VOCALL (vp->v_op, VOFFSET(vop_strategy), ap);
1744 	return (0);
1745 }
1746 
1747 /*
1748  * Print out the contents of an inode.
1749  */
1750 int
1751 ufs_print(ap)
1752 	struct vop_print_args /* {
1753 		struct vnode *a_vp;
1754 	} */ *ap;
1755 {
1756 	register struct vnode *vp = ap->a_vp;
1757 	register struct inode *ip = VTOI(vp);
1758 
1759 	printf("tag VT_UFS, ino %d, on dev %d, %d", ip->i_number,
1760 		major(ip->i_dev), minor(ip->i_dev));
1761 #ifdef FIFO
1762 	if (vp->v_type == VFIFO)
1763 		fifo_printinfo(vp);
1764 #endif /* FIFO */
1765 	printf("%s\n", (ip->i_flag & IN_LOCKED) ? " (LOCKED)" : "");
1766 	if (ip->i_lockholder == 0)
1767 		return (0);
1768 	printf("\towner pid %d", ip->i_lockholder);
1769 	if (ip->i_lockwaiter)
1770 		printf(" waiting pid %d", ip->i_lockwaiter);
1771 	printf("\n");
1772 	return (0);
1773 }
1774 
1775 /*
1776  * Read wrapper for special devices.
1777  */
1778 int
1779 ufsspec_read(ap)
1780 	struct vop_read_args /* {
1781 		struct vnode *a_vp;
1782 		struct uio *a_uio;
1783 		int  a_ioflag;
1784 		struct ucred *a_cred;
1785 	} */ *ap;
1786 {
1787 
1788 	/*
1789 	 * Set access flag.
1790 	 */
1791 	VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1792 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_read), ap));
1793 }
1794 
1795 /*
1796  * Write wrapper for special devices.
1797  */
1798 int
1799 ufsspec_write(ap)
1800 	struct vop_write_args /* {
1801 		struct vnode *a_vp;
1802 		struct uio *a_uio;
1803 		int  a_ioflag;
1804 		struct ucred *a_cred;
1805 	} */ *ap;
1806 {
1807 
1808 	/*
1809 	 * Set update and change flags.
1810 	 */
1811 	VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1812 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_write), ap));
1813 }
1814 
1815 /*
1816  * Close wrapper for special devices.
1817  *
1818  * Update the times on the inode then do device close.
1819  */
1820 int
1821 ufsspec_close(ap)
1822 	struct vop_close_args /* {
1823 		struct vnode *a_vp;
1824 		int  a_fflag;
1825 		struct ucred *a_cred;
1826 		struct proc *a_p;
1827 	} */ *ap;
1828 {
1829 	register struct inode *ip = VTOI(ap->a_vp);
1830 
1831 	if (ap->a_vp->v_usecount > 1 && !(ip->i_flag & IN_LOCKED))
1832 		ITIMES(ip, &time, &time);
1833 	return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
1834 }
1835 
1836 #ifdef FIFO
1837 /*
1838  * Read wrapper for fifo's
1839  */
1840 int
1841 ufsfifo_read(ap)
1842 	struct vop_read_args /* {
1843 		struct vnode *a_vp;
1844 		struct uio *a_uio;
1845 		int  a_ioflag;
1846 		struct ucred *a_cred;
1847 	} */ *ap;
1848 {
1849 	extern int (**fifo_vnodeop_p)();
1850 
1851 	/*
1852 	 * Set access flag.
1853 	 */
1854 	VTOI(ap->a_vp)->i_flag |= IN_ACCESS;
1855 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_read), ap));
1856 }
1857 
1858 /*
1859  * Write wrapper for fifo's.
1860  */
1861 int
1862 ufsfifo_write(ap)
1863 	struct vop_write_args /* {
1864 		struct vnode *a_vp;
1865 		struct uio *a_uio;
1866 		int  a_ioflag;
1867 		struct ucred *a_cred;
1868 	} */ *ap;
1869 {
1870 	extern int (**fifo_vnodeop_p)();
1871 
1872 	/*
1873 	 * Set update and change flags.
1874 	 */
1875 	VTOI(ap->a_vp)->i_flag |= IN_CHANGE | IN_UPDATE;
1876 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_write), ap));
1877 }
1878 
1879 /*
1880  * Close wrapper for fifo's.
1881  *
1882  * Update the times on the inode then do device close.
1883  */
1884 ufsfifo_close(ap)
1885 	struct vop_close_args /* {
1886 		struct vnode *a_vp;
1887 		int  a_fflag;
1888 		struct ucred *a_cred;
1889 		struct proc *a_p;
1890 	} */ *ap;
1891 {
1892 	extern int (**fifo_vnodeop_p)();
1893 	register struct inode *ip = VTOI(ap->a_vp);
1894 
1895 	if (ap->a_vp->v_usecount > 1 && !(ip->i_flag & IN_LOCKED))
1896 		ITIMES(ip, &time, &time);
1897 	return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
1898 }
1899 #endif /* FIFO */
1900 
1901 /*
1902  * Return POSIX pathconf information applicable to ufs filesystems.
1903  */
1904 ufs_pathconf(ap)
1905 	struct vop_pathconf_args /* {
1906 		struct vnode *a_vp;
1907 		int a_name;
1908 		int *a_retval;
1909 	} */ *ap;
1910 {
1911 
1912 	switch (ap->a_name) {
1913 	case _PC_LINK_MAX:
1914 		*ap->a_retval = LINK_MAX;
1915 		return (0);
1916 	case _PC_NAME_MAX:
1917 		*ap->a_retval = NAME_MAX;
1918 		return (0);
1919 	case _PC_PATH_MAX:
1920 		*ap->a_retval = PATH_MAX;
1921 		return (0);
1922 	case _PC_PIPE_BUF:
1923 		*ap->a_retval = PIPE_BUF;
1924 		return (0);
1925 	case _PC_CHOWN_RESTRICTED:
1926 		*ap->a_retval = 1;
1927 		return (0);
1928 	case _PC_NO_TRUNC:
1929 		*ap->a_retval = 1;
1930 		return (0);
1931 	default:
1932 		return (EINVAL);
1933 	}
1934 	/* NOTREACHED */
1935 }
1936 
1937 /*
1938  * Advisory record locking support
1939  */
1940 int
1941 ufs_advlock(ap)
1942 	struct vop_advlock_args /* {
1943 		struct vnode *a_vp;
1944 		caddr_t  a_id;
1945 		int  a_op;
1946 		struct flock *a_fl;
1947 		int  a_flags;
1948 	} */ *ap;
1949 {
1950 	register struct inode *ip = VTOI(ap->a_vp);
1951 	register struct flock *fl = ap->a_fl;
1952 	register struct lockf *lock;
1953 	off_t start, end;
1954 	int error;
1955 
1956 	/*
1957 	 * Avoid the common case of unlocking when inode has no locks.
1958 	 */
1959 	if (ip->i_lockf == (struct lockf *)0) {
1960 		if (ap->a_op != F_SETLK) {
1961 			fl->l_type = F_UNLCK;
1962 			return (0);
1963 		}
1964 	}
1965 	/*
1966 	 * Convert the flock structure into a start and end.
1967 	 */
1968 	switch (fl->l_whence) {
1969 
1970 	case SEEK_SET:
1971 	case SEEK_CUR:
1972 		/*
1973 		 * Caller is responsible for adding any necessary offset
1974 		 * when SEEK_CUR is used.
1975 		 */
1976 		start = fl->l_start;
1977 		break;
1978 
1979 	case SEEK_END:
1980 		start = ip->i_size + fl->l_start;
1981 		break;
1982 
1983 	default:
1984 		return (EINVAL);
1985 	}
1986 	if (start < 0)
1987 		return (EINVAL);
1988 	if (fl->l_len == 0)
1989 		end = -1;
1990 	else
1991 		end = start + fl->l_len - 1;
1992 	/*
1993 	 * Create the lockf structure
1994 	 */
1995 	MALLOC(lock, struct lockf *, sizeof *lock, M_LOCKF, M_WAITOK);
1996 	lock->lf_start = start;
1997 	lock->lf_end = end;
1998 	lock->lf_id = ap->a_id;
1999 	lock->lf_inode = ip;
2000 	lock->lf_type = fl->l_type;
2001 	lock->lf_next = (struct lockf *)0;
2002 	lock->lf_block = (struct lockf *)0;
2003 	lock->lf_flags = ap->a_flags;
2004 	/*
2005 	 * Do the requested operation.
2006 	 */
2007 	switch(ap->a_op) {
2008 	case F_SETLK:
2009 		return (lf_setlock(lock));
2010 
2011 	case F_UNLCK:
2012 		error = lf_clearlock(lock);
2013 		FREE(lock, M_LOCKF);
2014 		return (error);
2015 
2016 	case F_GETLK:
2017 		error = lf_getlock(lock, fl);
2018 		FREE(lock, M_LOCKF);
2019 		return (error);
2020 
2021 	default:
2022 		free(lock, M_LOCKF);
2023 		return (EINVAL);
2024 	}
2025 	/* NOTREACHED */
2026 }
2027 
2028 /*
2029  * Initialize the vnode associated with a new inode, handle aliased
2030  * vnodes.
2031  */
2032 int
2033 ufs_vinit(mntp, specops, fifoops, vpp)
2034 	struct mount *mntp;
2035 	int (**specops)();
2036 	int (**fifoops)();
2037 	struct vnode **vpp;
2038 {
2039 	struct inode *ip;
2040 	struct vnode *vp, *nvp;
2041 
2042 	vp = *vpp;
2043 	ip = VTOI(vp);
2044 	switch(vp->v_type = IFTOVT(ip->i_mode)) {
2045 	case VCHR:
2046 	case VBLK:
2047 		vp->v_op = specops;
2048 		if (nvp = checkalias(vp, ip->i_rdev, mntp)) {
2049 			/*
2050 			 * Discard unneeded vnode, but save its inode.
2051 			 */
2052 			ufs_ihashrem(ip);
2053 			VOP_UNLOCK(vp);
2054 			nvp->v_data = vp->v_data;
2055 			vp->v_data = NULL;
2056 			vp->v_op = spec_vnodeop_p;
2057 			vrele(vp);
2058 			vgone(vp);
2059 			/*
2060 			 * Reinitialize aliased inode.
2061 			 */
2062 			vp = nvp;
2063 			ip->i_vnode = vp;
2064 			ufs_ihashins(ip);
2065 		}
2066 		break;
2067 	case VFIFO:
2068 #ifdef FIFO
2069 		vp->v_op = fifoops;
2070 		break;
2071 #else
2072 		return (EOPNOTSUPP);
2073 #endif
2074 	}
2075 	if (ip->i_number == ROOTINO)
2076                 vp->v_flag |= VROOT;
2077 	/*
2078 	 * Initialize modrev times
2079 	 */
2080 	SETHIGH(ip->i_modrev, mono_time.tv_sec);
2081 	SETLOW(ip->i_modrev, mono_time.tv_usec * 4294);
2082 	*vpp = vp;
2083 	return (0);
2084 }
2085 
2086 /*
2087  * Allocate a new inode.
2088  */
2089 int
2090 ufs_makeinode(mode, dvp, vpp, cnp)
2091 	int mode;
2092 	struct vnode *dvp;
2093 	struct vnode **vpp;
2094 	struct componentname *cnp;
2095 {
2096 	register struct inode *ip, *pdir;
2097 	struct timeval tv;
2098 	struct vnode *tvp;
2099 	int error;
2100 
2101 	pdir = VTOI(dvp);
2102 #ifdef DIAGNOSTIC
2103 	if ((cnp->cn_flags & HASBUF) == 0)
2104 		panic("ufs_makeinode: no name");
2105 #endif
2106 	*vpp = NULL;
2107 	if ((mode & IFMT) == 0)
2108 		mode |= IFREG;
2109 
2110 	if (error = VOP_VALLOC(dvp, mode, cnp->cn_cred, &tvp)) {
2111 		free(cnp->cn_pnbuf, M_NAMEI);
2112 		vput(dvp);
2113 		return (error);
2114 	}
2115 	ip = VTOI(tvp);
2116 	ip->i_gid = pdir->i_gid;
2117 	if ((mode & IFMT) == IFLNK)
2118 		ip->i_uid = pdir->i_uid;
2119 	else
2120 		ip->i_uid = cnp->cn_cred->cr_uid;
2121 #ifdef QUOTA
2122 	if ((error = getinoquota(ip)) ||
2123 	    (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2124 		free(cnp->cn_pnbuf, M_NAMEI);
2125 		VOP_VFREE(tvp, ip->i_number, mode);
2126 		vput(tvp);
2127 		vput(dvp);
2128 		return (error);
2129 	}
2130 #endif
2131 	ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2132 	ip->i_mode = mode;
2133 	tvp->v_type = IFTOVT(mode);	/* Rest init'd in getnewvnode(). */
2134 	ip->i_nlink = 1;
2135 	if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2136 	    suser(cnp->cn_cred, NULL))
2137 		ip->i_mode &= ~ISGID;
2138 
2139 	/*
2140 	 * Make sure inode goes to disk before directory entry.
2141 	 */
2142 	tv = time;
2143 	if (error = VOP_UPDATE(tvp, &tv, &tv, 1))
2144 		goto bad;
2145 	if (error = ufs_direnter(ip, dvp, cnp))
2146 		goto bad;
2147 	if ((cnp->cn_flags & SAVESTART) == 0)
2148 		FREE(cnp->cn_pnbuf, M_NAMEI);
2149 	vput(dvp);
2150 	*vpp = tvp;
2151 	return (0);
2152 
2153 bad:
2154 	/*
2155 	 * Write error occurred trying to update the inode
2156 	 * or the directory so must deallocate the inode.
2157 	 */
2158 	free(cnp->cn_pnbuf, M_NAMEI);
2159 	vput(dvp);
2160 	ip->i_nlink = 0;
2161 	ip->i_flag |= IN_CHANGE;
2162 	vput(tvp);
2163 	return (error);
2164 }
2165