xref: /csrg-svn/sys/ufs/lfs/lfs_vnops.c (revision 26361)
1 /*
2  * Copyright (c) 1982 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)lfs_vnops.c	6.22 (Berkeley) 02/23/86
7  */
8 
9 #include "param.h"
10 #include "systm.h"
11 #include "dir.h"
12 #include "user.h"
13 #include "kernel.h"
14 #include "file.h"
15 #include "stat.h"
16 #include "inode.h"
17 #include "fs.h"
18 #include "buf.h"
19 #include "proc.h"
20 #include "quota.h"
21 #include "uio.h"
22 #include "socket.h"
23 #include "socketvar.h"
24 #include "mount.h"
25 
26 extern	struct fileops inodeops;
27 struct	file *getinode();
28 
29 /*
30  * Change current working directory (``.'').
31  */
32 chdir()
33 {
34 
35 	chdirec(&u.u_cdir);
36 }
37 
38 /*
39  * Change notion of root (``/'') directory.
40  */
41 chroot()
42 {
43 
44 	if (suser())
45 		chdirec(&u.u_rdir);
46 }
47 
48 /*
49  * Common routine for chroot and chdir.
50  */
51 chdirec(ipp)
52 	register struct inode **ipp;
53 {
54 	register struct inode *ip;
55 	struct a {
56 		char	*fname;
57 	} *uap = (struct a *)u.u_ap;
58 	register struct nameidata *ndp = &u.u_nd;
59 
60 	ndp->ni_nameiop = LOOKUP | FOLLOW;
61 	ndp->ni_segflg = UIO_USERSPACE;
62 	ndp->ni_dirp = uap->fname;
63 	ip = namei(ndp);
64 	if (ip == NULL)
65 		return;
66 	if ((ip->i_mode&IFMT) != IFDIR) {
67 		u.u_error = ENOTDIR;
68 		goto bad;
69 	}
70 	if (access(ip, IEXEC))
71 		goto bad;
72 	IUNLOCK(ip);
73 	if (*ipp)
74 		irele(*ipp);
75 	*ipp = ip;
76 	return;
77 
78 bad:
79 	iput(ip);
80 }
81 
82 /*
83  * Open system call.
84  */
85 open()
86 {
87 	struct a {
88 		char	*fname;
89 		int	mode;
90 		int	crtmode;
91 	} *uap = (struct a *) u.u_ap;
92 
93 	copen(uap->mode-FOPEN, uap->crtmode, uap->fname);
94 }
95 
96 /*
97  * Creat system call.
98  */
99 creat()
100 {
101 	struct a {
102 		char	*fname;
103 		int	fmode;
104 	} *uap = (struct a *)u.u_ap;
105 
106 	copen(FWRITE|FCREAT|FTRUNC, uap->fmode, uap->fname);
107 }
108 
109 /*
110  * Common code for open and creat.
111  * Check permissions, allocate an open file structure,
112  * and call the device open routine if any.
113  */
114 copen(mode, arg, fname)
115 	register int mode;
116 	int arg;
117 	caddr_t fname;
118 {
119 	register struct inode *ip;
120 	register struct file *fp;
121 	register struct nameidata *ndp = &u.u_nd;
122 	int indx;
123 
124 	fp = falloc();
125 	if (fp == NULL)
126 		return;
127 	indx = u.u_r.r_val1;
128 	ndp->ni_segflg = UIO_USERSPACE;
129 	ndp->ni_dirp = fname;
130 	if (mode&FCREAT) {
131 		if (mode & FEXCL)
132 			ndp->ni_nameiop = CREATE;
133 		else
134 			ndp->ni_nameiop = CREATE | FOLLOW;
135 		ip = namei(ndp);
136 		if (ip == NULL) {
137 			if (u.u_error)
138 				goto bad1;
139 			ip = maknode(arg&07777&(~ISVTX), ndp);
140 			if (ip == NULL)
141 				goto bad1;
142 			mode &= ~FTRUNC;
143 		} else {
144 			if (mode&FEXCL) {
145 				u.u_error = EEXIST;
146 				goto bad;
147 			}
148 			mode &= ~FCREAT;
149 		}
150 	} else {
151 		ndp->ni_nameiop = LOOKUP | FOLLOW;
152 		ip = namei(ndp);
153 		if (ip == NULL)
154 			goto bad1;
155 	}
156 	if ((ip->i_mode & IFMT) == IFSOCK) {
157 		u.u_error = EOPNOTSUPP;
158 		goto bad;
159 	}
160 	if ((mode&FCREAT) == 0) {
161 		if (mode&FREAD)
162 			if (access(ip, IREAD))
163 				goto bad;
164 		if (mode&(FWRITE|FTRUNC)) {
165 			if (access(ip, IWRITE))
166 				goto bad;
167 			if ((ip->i_mode&IFMT) == IFDIR) {
168 				u.u_error = EISDIR;
169 				goto bad;
170 			}
171 		}
172 	}
173 	if (mode&FTRUNC)
174 		itrunc(ip, (u_long)0);
175 	IUNLOCK(ip);
176 	fp->f_flag = mode&FMASK;
177 	fp->f_type = DTYPE_INODE;
178 	fp->f_ops = &inodeops;
179 	fp->f_data = (caddr_t)ip;
180 	if (setjmp(&u.u_qsave)) {
181 		if (u.u_error == 0)
182 			u.u_error = EINTR;
183 		u.u_ofile[indx] = NULL;
184 		closef(fp);
185 		return;
186 	}
187 	u.u_error = openi(ip, mode);
188 	if (u.u_error == 0)
189 		return;
190 	ILOCK(ip);
191 bad:
192 	iput(ip);
193 bad1:
194 	u.u_ofile[indx] = NULL;
195 	fp->f_count--;
196 }
197 
198 /*
199  * Mknod system call
200  */
201 mknod()
202 {
203 	register struct inode *ip;
204 	register struct a {
205 		char	*fname;
206 		int	fmode;
207 		int	dev;
208 	} *uap = (struct a *)u.u_ap;
209 	register struct nameidata *ndp = &u.u_nd;
210 
211 	if (!suser())
212 		return;
213 	ndp->ni_nameiop = CREATE;
214 	ndp->ni_segflg = UIO_USERSPACE;
215 	ndp->ni_dirp = uap->fname;
216 	ip = namei(ndp);
217 	if (ip != NULL) {
218 		u.u_error = EEXIST;
219 		goto out;
220 	}
221 	if (u.u_error)
222 		return;
223 	ip = maknode(uap->fmode, ndp);
224 	if (ip == NULL)
225 		return;
226 	switch (ip->i_mode & IFMT) {
227 
228 	case IFMT:	/* used by badsect to flag bad sectors */
229 	case IFCHR:
230 	case IFBLK:
231 		if (uap->dev) {
232 			/*
233 			 * Want to be able to use this to make badblock
234 			 * inodes, so don't truncate the dev number.
235 			 */
236 			ip->i_rdev = uap->dev;
237 			ip->i_flag |= IACC|IUPD|ICHG;
238 		}
239 	}
240 
241 out:
242 	iput(ip);
243 }
244 
245 /*
246  * link system call
247  */
248 link()
249 {
250 	register struct inode *ip, *xp;
251 	register struct a {
252 		char	*target;
253 		char	*linkname;
254 	} *uap = (struct a *)u.u_ap;
255 	register struct nameidata *ndp = &u.u_nd;
256 
257 	ndp->ni_nameiop = LOOKUP | FOLLOW;
258 	ndp->ni_segflg = UIO_USERSPACE;
259 	ndp->ni_dirp = uap->target;
260 	ip = namei(ndp);	/* well, this routine is doomed anyhow */
261 	if (ip == NULL)
262 		return;
263 	if ((ip->i_mode&IFMT) == IFDIR && !suser()) {
264 		iput(ip);
265 		return;
266 	}
267 	ip->i_nlink++;
268 	ip->i_flag |= ICHG;
269 	iupdat(ip, &time, &time, 1);
270 	IUNLOCK(ip);
271 	ndp->ni_nameiop = CREATE;
272 	ndp->ni_segflg = UIO_USERSPACE;
273 	ndp->ni_dirp = (caddr_t)uap->linkname;
274 	xp = namei(ndp);
275 	if (xp != NULL) {
276 		u.u_error = EEXIST;
277 		iput(xp);
278 		goto out;
279 	}
280 	if (u.u_error)
281 		goto out;
282 	if (ndp->ni_pdir->i_dev != ip->i_dev) {
283 		iput(ndp->ni_pdir);
284 		u.u_error = EXDEV;
285 		goto out;
286 	}
287 	u.u_error = direnter(ip, ndp);
288 out:
289 	if (u.u_error) {
290 		ip->i_nlink--;
291 		ip->i_flag |= ICHG;
292 	}
293 	irele(ip);
294 }
295 
296 /*
297  * symlink -- make a symbolic link
298  */
299 symlink()
300 {
301 	register struct a {
302 		char	*target;
303 		char	*linkname;
304 	} *uap = (struct a *)u.u_ap;
305 	register struct inode *ip;
306 	register char *tp;
307 	register c, nc;
308 	register struct nameidata *ndp = &u.u_nd;
309 
310 	tp = uap->target;
311 	nc = 0;
312 	while (c = fubyte(tp)) {
313 		if (c < 0) {
314 			u.u_error = EFAULT;
315 			return;
316 		}
317 		tp++;
318 		nc++;
319 	}
320 	ndp->ni_nameiop = CREATE;
321 	ndp->ni_segflg = UIO_USERSPACE;
322 	ndp->ni_dirp = uap->linkname;
323 	ip = namei(ndp);
324 	if (ip) {
325 		iput(ip);
326 		u.u_error = EEXIST;
327 		return;
328 	}
329 	if (u.u_error)
330 		return;
331 	ip = maknode(IFLNK | 0777, ndp);
332 	if (ip == NULL)
333 		return;
334 	u.u_error = rdwri(UIO_WRITE, ip, uap->target, nc, (off_t)0, 0,
335 	    (int *)0);
336 	/* handle u.u_error != 0 */
337 	iput(ip);
338 }
339 
340 /*
341  * Unlink system call.
342  * Hard to avoid races here, especially
343  * in unlinking directories.
344  */
345 unlink()
346 {
347 	struct a {
348 		char	*fname;
349 	} *uap = (struct a *)u.u_ap;
350 	register struct inode *ip, *dp;
351 	register struct nameidata *ndp = &u.u_nd;
352 
353 	ndp->ni_nameiop = DELETE | LOCKPARENT;
354 	ndp->ni_segflg = UIO_USERSPACE;
355 	ndp->ni_dirp = uap->fname;
356 	ip = namei(ndp);
357 	if (ip == NULL)
358 		return;
359 	dp = ndp->ni_pdir;
360 	if ((ip->i_mode&IFMT) == IFDIR && !suser())
361 		goto out;
362 	/*
363 	 * Don't unlink a mounted file.
364 	 */
365 	if (ip->i_dev != dp->i_dev) {
366 		u.u_error = EBUSY;
367 		goto out;
368 	}
369 	if (ip->i_flag&ITEXT)
370 		xrele(ip);	/* try once to free text */
371 	if (dirremove(ndp)) {
372 		ip->i_nlink--;
373 		ip->i_flag |= ICHG;
374 	}
375 out:
376 	if (dp == ip)
377 		irele(ip);
378 	else
379 		iput(ip);
380 	iput(dp);
381 }
382 
383 /*
384  * Seek system call
385  */
386 lseek()
387 {
388 	register struct file *fp;
389 	register struct a {
390 		int	fd;
391 		off_t	off;
392 		int	sbase;
393 	} *uap = (struct a *)u.u_ap;
394 
395 	GETF(fp, uap->fd);
396 	if (fp->f_type != DTYPE_INODE) {
397 		u.u_error = ESPIPE;
398 		return;
399 	}
400 	switch (uap->sbase) {
401 
402 	case L_INCR:
403 		fp->f_offset += uap->off;
404 		break;
405 
406 	case L_XTND:
407 		fp->f_offset = uap->off + ((struct inode *)fp->f_data)->i_size;
408 		break;
409 
410 	case L_SET:
411 		fp->f_offset = uap->off;
412 		break;
413 
414 	default:
415 		u.u_error = EINVAL;
416 		return;
417 	}
418 	u.u_r.r_off = fp->f_offset;
419 }
420 
421 /*
422  * Access system call
423  */
424 saccess()
425 {
426 	register svuid, svgid;
427 	register struct inode *ip;
428 	register struct a {
429 		char	*fname;
430 		int	fmode;
431 	} *uap = (struct a *)u.u_ap;
432 	register struct nameidata *ndp = &u.u_nd;
433 
434 	svuid = u.u_uid;
435 	svgid = u.u_gid;
436 	u.u_uid = u.u_ruid;
437 	u.u_gid = u.u_rgid;
438 	ndp->ni_nameiop = LOOKUP | FOLLOW;
439 	ndp->ni_segflg = UIO_USERSPACE;
440 	ndp->ni_dirp = uap->fname;
441 	ip = namei(ndp);
442 	if (ip != NULL) {
443 		if ((uap->fmode&R_OK) && access(ip, IREAD))
444 			goto done;
445 		if ((uap->fmode&W_OK) && access(ip, IWRITE))
446 			goto done;
447 		if ((uap->fmode&X_OK) && access(ip, IEXEC))
448 			goto done;
449 done:
450 		iput(ip);
451 	}
452 	u.u_uid = svuid;
453 	u.u_gid = svgid;
454 }
455 
456 /*
457  * Stat system call.  This version follows links.
458  */
459 stat()
460 {
461 
462 	stat1(FOLLOW);
463 }
464 
465 /*
466  * Lstat system call.  This version does not follow links.
467  */
468 lstat()
469 {
470 
471 	stat1(NOFOLLOW);
472 }
473 
474 stat1(follow)
475 	int follow;
476 {
477 	register struct inode *ip;
478 	register struct a {
479 		char	*fname;
480 		struct stat *ub;
481 	} *uap = (struct a *)u.u_ap;
482 	struct stat sb;
483 	register struct nameidata *ndp = &u.u_nd;
484 
485 	ndp->ni_nameiop = LOOKUP | follow;
486 	ndp->ni_segflg = UIO_USERSPACE;
487 	ndp->ni_dirp = uap->fname;
488 	ip = namei(ndp);
489 	if (ip == NULL)
490 		return;
491 	(void) ino_stat(ip, &sb);
492 	iput(ip);
493 	u.u_error = copyout((caddr_t)&sb, (caddr_t)uap->ub, sizeof (sb));
494 }
495 
496 /*
497  * Return target name of a symbolic link
498  */
499 readlink()
500 {
501 	register struct inode *ip;
502 	register struct a {
503 		char	*name;
504 		char	*buf;
505 		int	count;
506 	} *uap = (struct a *)u.u_ap;
507 	register struct nameidata *ndp = &u.u_nd;
508 	int resid;
509 
510 	ndp->ni_nameiop = LOOKUP;
511 	ndp->ni_segflg = UIO_USERSPACE;
512 	ndp->ni_dirp = uap->name;
513 	ip = namei(ndp);
514 	if (ip == NULL)
515 		return;
516 	if ((ip->i_mode&IFMT) != IFLNK) {
517 		u.u_error = EINVAL;
518 		goto out;
519 	}
520 	u.u_error = rdwri(UIO_READ, ip, uap->buf, uap->count, (off_t)0, 0,
521 	    &resid);
522 out:
523 	iput(ip);
524 	u.u_r.r_val1 = uap->count - resid;
525 }
526 
527 /*
528  * Change mode of a file given path name.
529  */
530 chmod()
531 {
532 	struct inode *ip;
533 	struct a {
534 		char	*fname;
535 		int	fmode;
536 	} *uap = (struct a *)u.u_ap;
537 
538 	if ((ip = owner(uap->fname, FOLLOW)) == NULL)
539 		return;
540 	u.u_error = chmod1(ip, uap->fmode);
541 	iput(ip);
542 }
543 
544 /*
545  * Change mode of a file given a file descriptor.
546  */
547 fchmod()
548 {
549 	struct a {
550 		int	fd;
551 		int	fmode;
552 	} *uap = (struct a *)u.u_ap;
553 	register struct inode *ip;
554 	register struct file *fp;
555 
556 	fp = getinode(uap->fd);
557 	if (fp == NULL)
558 		return;
559 	ip = (struct inode *)fp->f_data;
560 	if (u.u_uid != ip->i_uid && !suser())
561 		return;
562 	ILOCK(ip);
563 	u.u_error = chmod1(ip, uap->fmode);
564 	IUNLOCK(ip);
565 }
566 
567 /*
568  * Change the mode on a file.
569  * Inode must be locked before calling.
570  */
571 chmod1(ip, mode)
572 	register struct inode *ip;
573 	register int mode;
574 {
575 
576 	if (ip->i_fs->fs_ronly)
577 		return (EROFS);
578 	ip->i_mode &= ~07777;
579 	if (u.u_uid) {
580 		if ((ip->i_mode & IFMT) != IFDIR)
581 			mode &= ~ISVTX;
582 		if (!groupmember(ip->i_gid))
583 			mode &= ~ISGID;
584 	}
585 	ip->i_mode |= mode&07777;
586 	ip->i_flag |= ICHG;
587 	if (ip->i_flag&ITEXT && (ip->i_mode&ISVTX)==0)
588 		xrele(ip);
589 	return (0);
590 }
591 
592 /*
593  * Set ownership given a path name.
594  */
595 chown()
596 {
597 	struct inode *ip;
598 	struct a {
599 		char	*fname;
600 		int	uid;
601 		int	gid;
602 	} *uap = (struct a *)u.u_ap;
603 
604 	if (!suser() || (ip = owner(uap->fname, NOFOLLOW)) == NULL)
605 		return;
606 	u.u_error = chown1(ip, uap->uid, uap->gid);
607 	iput(ip);
608 }
609 
610 /*
611  * Set ownership given a file descriptor.
612  */
613 fchown()
614 {
615 	struct a {
616 		int	fd;
617 		int	uid;
618 		int	gid;
619 	} *uap = (struct a *)u.u_ap;
620 	register struct inode *ip;
621 	register struct file *fp;
622 
623 	fp = getinode(uap->fd);
624 	if (fp == NULL)
625 		return;
626 	ip = (struct inode *)fp->f_data;
627 	if (!suser())
628 		return;
629 	ILOCK(ip);
630 	u.u_error = chown1(ip, uap->uid, uap->gid);
631 	IUNLOCK(ip);
632 }
633 
634 /*
635  * Perform chown operation on inode ip;
636  * inode must be locked prior to call.
637  */
638 chown1(ip, uid, gid)
639 	register struct inode *ip;
640 	int uid, gid;
641 {
642 #ifdef QUOTA
643 	register long change;
644 #endif
645 
646 	if (ip->i_fs->fs_ronly)
647 		return (EROFS);
648 	if (uid == -1)
649 		uid = ip->i_uid;
650 	if (gid == -1)
651 		gid = ip->i_gid;
652 #ifdef QUOTA
653 	if (ip->i_uid == uid)		/* this just speeds things a little */
654 		change = 0;
655 	else
656 		change = ip->i_blocks;
657 	(void) chkdq(ip, -change, 1);
658 	(void) chkiq(ip->i_dev, ip, ip->i_uid, 1);
659 	dqrele(ip->i_dquot);
660 #endif
661 	ip->i_uid = uid;
662 	ip->i_gid = gid;
663 	ip->i_flag |= ICHG;
664 	if (u.u_ruid != 0)
665 		ip->i_mode &= ~(ISUID|ISGID);
666 #ifdef QUOTA
667 	ip->i_dquot = inoquota(ip);
668 	(void) chkdq(ip, change, 1);
669 	(void) chkiq(ip->i_dev, (struct inode *)NULL, (uid_t)uid, 1);
670 	return (u.u_error);		/* should == 0 ALWAYS !! */
671 #else
672 	return (0);
673 #endif
674 }
675 
676 utimes()
677 {
678 	register struct a {
679 		char	*fname;
680 		struct	timeval *tptr;
681 	} *uap = (struct a *)u.u_ap;
682 	register struct inode *ip;
683 	struct timeval tv[2];
684 
685 	if ((ip = owner(uap->fname, FOLLOW)) == NULL)
686 		return;
687 	if (ip->i_fs->fs_ronly) {
688 		u.u_error = EROFS;
689 		iput(ip);
690 		return;
691 	}
692 	u.u_error = copyin((caddr_t)uap->tptr, (caddr_t)tv, sizeof (tv));
693 	if (u.u_error == 0) {
694 		ip->i_flag |= IACC|IUPD|ICHG;
695 		iupdat(ip, &tv[0], &tv[1], 0);
696 	}
697 	iput(ip);
698 }
699 
700 /*
701  * Flush any pending I/O.
702  */
703 sync()
704 {
705 
706 	update();
707 }
708 
709 /*
710  * Truncate a file given its path name.
711  */
712 truncate()
713 {
714 	struct a {
715 		char	*fname;
716 		u_long	length;
717 	} *uap = (struct a *)u.u_ap;
718 	struct inode *ip;
719 	register struct nameidata *ndp = &u.u_nd;
720 
721 	ndp->ni_nameiop = LOOKUP | FOLLOW;
722 	ndp->ni_segflg = UIO_USERSPACE;
723 	ndp->ni_dirp = uap->fname;
724 	ip = namei(ndp);
725 	if (ip == NULL)
726 		return;
727 	if (access(ip, IWRITE))
728 		goto bad;
729 	if ((ip->i_mode&IFMT) == IFDIR) {
730 		u.u_error = EISDIR;
731 		goto bad;
732 	}
733 	itrunc(ip, uap->length);
734 bad:
735 	iput(ip);
736 }
737 
738 /*
739  * Truncate a file given a file descriptor.
740  */
741 ftruncate()
742 {
743 	struct a {
744 		int	fd;
745 		u_long	length;
746 	} *uap = (struct a *)u.u_ap;
747 	struct inode *ip;
748 	struct file *fp;
749 
750 	fp = getinode(uap->fd);
751 	if (fp == NULL)
752 		return;
753 	if ((fp->f_flag&FWRITE) == 0) {
754 		u.u_error = EINVAL;
755 		return;
756 	}
757 	ip = (struct inode *)fp->f_data;
758 	ILOCK(ip);
759 	itrunc(ip, uap->length);
760 	IUNLOCK(ip);
761 }
762 
763 /*
764  * Synch an open file.
765  */
766 fsync()
767 {
768 	struct a {
769 		int	fd;
770 	} *uap = (struct a *)u.u_ap;
771 	struct inode *ip;
772 	struct file *fp;
773 
774 	fp = getinode(uap->fd);
775 	if (fp == NULL)
776 		return;
777 	ip = (struct inode *)fp->f_data;
778 	ILOCK(ip);
779 	syncip(ip);
780 	IUNLOCK(ip);
781 }
782 
783 /*
784  * Rename system call.
785  * 	rename("foo", "bar");
786  * is essentially
787  *	unlink("bar");
788  *	link("foo", "bar");
789  *	unlink("foo");
790  * but ``atomically''.  Can't do full commit without saving state in the
791  * inode on disk which isn't feasible at this time.  Best we can do is
792  * always guarantee the target exists.
793  *
794  * Basic algorithm is:
795  *
796  * 1) Bump link count on source while we're linking it to the
797  *    target.  This also insure the inode won't be deleted out
798  *    from underneath us while we work (it may be truncated by
799  *    a concurrent `trunc' or `open' for creation).
800  * 2) Link source to destination.  If destination already exists,
801  *    delete it first.
802  * 3) Unlink source reference to inode if still around. If a
803  *    directory was moved and the parent of the destination
804  *    is different from the source, patch the ".." entry in the
805  *    directory.
806  *
807  * Source and destination must either both be directories, or both
808  * not be directories.  If target is a directory, it must be empty.
809  */
810 rename()
811 {
812 	struct a {
813 		char	*from;
814 		char	*to;
815 	} *uap = (struct a *)u.u_ap;
816 	register struct inode *ip, *xp, *dp;
817 	struct dirtemplate dirbuf;
818 	int doingdirectory = 0, oldparent = 0, newparent = 0;
819 	register struct nameidata *ndp = &u.u_nd;
820 	int error = 0;
821 
822 	ndp->ni_nameiop = DELETE | LOCKPARENT;
823 	ndp->ni_segflg = UIO_USERSPACE;
824 	ndp->ni_dirp = uap->from;
825 	ip = namei(ndp);
826 	if (ip == NULL)
827 		return;
828 	dp = ndp->ni_pdir;
829 	if ((ip->i_mode&IFMT) == IFDIR) {
830 		register struct direct *d;
831 
832 		d = &ndp->ni_dent;
833 		/*
834 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
835 		 */
836 		if ((d->d_namlen == 1 && d->d_name[0] == '.') ||
837 		    (d->d_namlen == 2 && bcmp(d->d_name, "..", 2) == 0) ||
838 		    (dp == ip) || (ip->i_flag & IRENAME)) {
839 			iput(dp);
840 			if (dp == ip)
841 				irele(ip);
842 			else
843 				iput(ip);
844 			u.u_error = EINVAL;
845 			return;
846 		}
847 		ip->i_flag |= IRENAME;
848 		oldparent = dp->i_number;
849 		doingdirectory++;
850 	}
851 	iput(dp);
852 
853 	/*
854 	 * 1) Bump link count while we're moving stuff
855 	 *    around.  If we crash somewhere before
856 	 *    completing our work, the link count
857 	 *    may be wrong, but correctable.
858 	 */
859 	ip->i_nlink++;
860 	ip->i_flag |= ICHG;
861 	iupdat(ip, &time, &time, 1);
862 	IUNLOCK(ip);
863 
864 	/*
865 	 * When the target exists, both the directory
866 	 * and target inodes are returned locked.
867 	 */
868 	ndp->ni_nameiop = CREATE | LOCKPARENT | NOCACHE;
869 	ndp->ni_dirp = (caddr_t)uap->to;
870 	xp = namei(ndp);
871 	if (u.u_error) {
872 		error = u.u_error;
873 		goto out;
874 	}
875 	dp = ndp->ni_pdir;
876 	/*
877 	 * If ".." must be changed (ie the directory gets a new
878 	 * parent) then the source directory must not be in the
879 	 * directory heirarchy above the target, as this would
880 	 * orphan everything below the source directory. Also
881 	 * the user must have write permission in the source so
882 	 * as to be able to change "..". We must repeat the call
883 	 * to namei, as the parent directory is unlocked by the
884 	 * call to checkpath().
885 	 */
886 	if (oldparent != dp->i_number)
887 		newparent = dp->i_number;
888 	if (doingdirectory && newparent) {
889 		if (access(ip, IWRITE))
890 			goto bad;
891 		do {
892 			dp = ndp->ni_pdir;
893 			if (xp != NULL)
894 				iput(xp);
895 			u.u_error = checkpath(ip, dp);
896 			if (u.u_error)
897 				goto out;
898 			xp = namei(ndp);
899 			if (u.u_error) {
900 				error = u.u_error;
901 				goto out;
902 			}
903 		} while (dp != ndp->ni_pdir);
904 	}
905 	/*
906 	 * 2) If target doesn't exist, link the target
907 	 *    to the source and unlink the source.
908 	 *    Otherwise, rewrite the target directory
909 	 *    entry to reference the source inode and
910 	 *    expunge the original entry's existence.
911 	 */
912 	if (xp == NULL) {
913 		if (dp->i_dev != ip->i_dev) {
914 			error = EXDEV;
915 			goto bad;
916 		}
917 		/*
918 		 * Account for ".." in new directory.
919 		 * When source and destination have the same
920 		 * parent we don't fool with the link count.
921 		 */
922 		if (doingdirectory && newparent) {
923 			dp->i_nlink++;
924 			dp->i_flag |= ICHG;
925 			iupdat(dp, &time, &time, 1);
926 		}
927 		error = direnter(ip, ndp);
928 		if (error)
929 			goto out;
930 	} else {
931 		if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) {
932 			error = EXDEV;
933 			goto bad;
934 		}
935 		/*
936 		 * Short circuit rename(foo, foo).
937 		 */
938 		if (xp->i_number == ip->i_number)
939 			goto bad;
940 		/*
941 		 * If the parent directory is "sticky", then the user must
942 		 * own the parent directory, or the destination of the rename,
943 		 * otherwise the destination may not be changed (except by
944 		 * root). This implements append-only directories.
945 		 */
946 		if ((dp->i_mode & ISVTX) && u.u_uid != 0 &&
947 		    u.u_uid != dp->i_uid && xp->i_uid != u.u_uid) {
948 			error = EPERM;
949 			goto bad;
950 		}
951 		/*
952 		 * Target must be empty if a directory
953 		 * and have no links to it.
954 		 * Also, insure source and target are
955 		 * compatible (both directories, or both
956 		 * not directories).
957 		 */
958 		if ((xp->i_mode&IFMT) == IFDIR) {
959 			if (!dirempty(xp, dp->i_number) || xp->i_nlink > 2) {
960 				error = ENOTEMPTY;
961 				goto bad;
962 			}
963 			if (!doingdirectory) {
964 				error = ENOTDIR;
965 				goto bad;
966 			}
967 			cacheinval(dp);
968 		} else if (doingdirectory) {
969 			error = EISDIR;
970 			goto bad;
971 		}
972 		dirrewrite(dp, ip, ndp);
973 		if (u.u_error) {
974 			error = u.u_error;
975 			goto bad1;
976 		}
977 		/*
978 		 * Adjust the link count of the target to
979 		 * reflect the dirrewrite above.  If this is
980 		 * a directory it is empty and there are
981 		 * no links to it, so we can squash the inode and
982 		 * any space associated with it.  We disallowed
983 		 * renaming over top of a directory with links to
984 		 * it above, as the remaining link would point to
985 		 * a directory without "." or ".." entries.
986 		 */
987 		xp->i_nlink--;
988 		if (doingdirectory) {
989 			if (--xp->i_nlink != 0)
990 				panic("rename: linked directory");
991 			itrunc(xp, (u_long)0);
992 		}
993 		xp->i_flag |= ICHG;
994 		iput(xp);
995 		xp = NULL;
996 	}
997 
998 	/*
999 	 * 3) Unlink the source.
1000 	 */
1001 	ndp->ni_nameiop = DELETE | LOCKPARENT;
1002 	ndp->ni_segflg = UIO_USERSPACE;
1003 	ndp->ni_dirp = uap->from;
1004 	xp = namei(ndp);
1005 	if (xp != NULL)
1006 		dp = ndp->ni_pdir;
1007 	else
1008 		dp = NULL;
1009 	/*
1010 	 * Insure that the directory entry still exists and has not
1011 	 * changed while the new name has been entered. If the source is
1012 	 * a file then the entry may have been unlinked or renamed. In
1013 	 * either case there is no further work to be done. If the source
1014 	 * is a directory then it cannot have been rmdir'ed; its link
1015 	 * count of three would cause a rmdir to fail with ENOTEMPTY.
1016 	 * The IRENAME flag insures that it cannot be moved by another
1017 	 * rename.
1018 	 */
1019 	if (xp != ip) {
1020 		if (doingdirectory)
1021 			panic("rename: lost dir entry");
1022 	} else {
1023 		/*
1024 		 * If the source is a directory with a
1025 		 * new parent, the link count of the old
1026 		 * parent directory must be decremented
1027 		 * and ".." set to point to the new parent.
1028 		 */
1029 		if (doingdirectory && newparent) {
1030 			dp->i_nlink--;
1031 			dp->i_flag |= ICHG;
1032 			error = rdwri(UIO_READ, xp, (caddr_t)&dirbuf,
1033 				sizeof (struct dirtemplate), (off_t)0, 1,
1034 				(int *)0);
1035 			if (error == 0) {
1036 				if (dirbuf.dotdot_namlen != 2 ||
1037 				    dirbuf.dotdot_name[0] != '.' ||
1038 				    dirbuf.dotdot_name[1] != '.') {
1039 					printf("rename: mangled dir\n");
1040 				} else {
1041 					dirbuf.dotdot_ino = newparent;
1042 					(void) rdwri(UIO_WRITE, xp,
1043 					    (caddr_t)&dirbuf,
1044 					    sizeof (struct dirtemplate),
1045 					    (off_t)0, 1, (int *)0);
1046 					cacheinval(dp);
1047 				}
1048 			}
1049 		}
1050 		if (dirremove(ndp)) {
1051 			xp->i_nlink--;
1052 			xp->i_flag |= ICHG;
1053 		}
1054 		xp->i_flag &= ~IRENAME;
1055 		if (error == 0)		/* XXX conservative */
1056 			error = u.u_error;
1057 	}
1058 	if (dp)
1059 		iput(dp);
1060 	if (xp)
1061 		iput(xp);
1062 	irele(ip);
1063 	if (error)
1064 		u.u_error = error;
1065 	return;
1066 
1067 bad:
1068 	iput(dp);
1069 bad1:
1070 	if (xp)
1071 		iput(xp);
1072 out:
1073 	ip->i_nlink--;
1074 	ip->i_flag |= ICHG;
1075 	irele(ip);
1076 	if (error)
1077 		u.u_error = error;
1078 }
1079 
1080 /*
1081  * Make a new file.
1082  */
1083 struct inode *
1084 maknode(mode, ndp)
1085 	int mode;
1086 	register struct nameidata *ndp;
1087 {
1088 	register struct inode *ip;
1089 	register struct inode *pdir = ndp->ni_pdir;
1090 	ino_t ipref;
1091 
1092 	if ((mode & IFMT) == IFDIR)
1093 		ipref = dirpref(pdir->i_fs);
1094 	else
1095 		ipref = pdir->i_number;
1096 	ip = ialloc(pdir, ipref, mode);
1097 	if (ip == NULL) {
1098 		iput(pdir);
1099 		return (NULL);
1100 	}
1101 #ifdef QUOTA
1102 	if (ip->i_dquot != NODQUOT)
1103 		panic("maknode: dquot");
1104 #endif
1105 	ip->i_flag |= IACC|IUPD|ICHG;
1106 	if ((mode & IFMT) == 0)
1107 		mode |= IFREG;
1108 	ip->i_mode = mode & ~u.u_cmask;
1109 	ip->i_nlink = 1;
1110 	ip->i_uid = u.u_uid;
1111 	ip->i_gid = pdir->i_gid;
1112 	if (ip->i_mode & ISGID && !groupmember(ip->i_gid))
1113 		ip->i_mode &= ~ISGID;
1114 #ifdef QUOTA
1115 	ip->i_dquot = inoquota(ip);
1116 #endif
1117 
1118 	/*
1119 	 * Make sure inode goes to disk before directory entry.
1120 	 */
1121 	iupdat(ip, &time, &time, 1);
1122 	u.u_error = direnter(ip, ndp);
1123 	if (u.u_error) {
1124 		/*
1125 		 * Write error occurred trying to update directory
1126 		 * so must deallocate the inode.
1127 		 */
1128 		ip->i_nlink = 0;
1129 		ip->i_flag |= ICHG;
1130 		iput(ip);
1131 		return (NULL);
1132 	}
1133 	return (ip);
1134 }
1135 
1136 /*
1137  * A virgin directory (no blushing please).
1138  */
1139 struct dirtemplate mastertemplate = {
1140 	0, 12, 1, ".",
1141 	0, DIRBLKSIZ - 12, 2, ".."
1142 };
1143 
1144 /*
1145  * Mkdir system call
1146  */
1147 mkdir()
1148 {
1149 	struct a {
1150 		char	*name;
1151 		int	dmode;
1152 	} *uap = (struct a *)u.u_ap;
1153 	register struct inode *ip, *dp;
1154 	struct dirtemplate dirtemplate;
1155 	register struct nameidata *ndp = &u.u_nd;
1156 
1157 	ndp->ni_nameiop = CREATE;
1158 	ndp->ni_segflg = UIO_USERSPACE;
1159 	ndp->ni_dirp = uap->name;
1160 	ip = namei(ndp);
1161 	if (u.u_error)
1162 		return;
1163 	if (ip != NULL) {
1164 		iput(ip);
1165 		u.u_error = EEXIST;
1166 		return;
1167 	}
1168 	dp = ndp->ni_pdir;
1169 	uap->dmode &= 0777;
1170 	uap->dmode |= IFDIR;
1171 	/*
1172 	 * Must simulate part of maknode here
1173 	 * in order to acquire the inode, but
1174 	 * not have it entered in the parent
1175 	 * directory.  The entry is made later
1176 	 * after writing "." and ".." entries out.
1177 	 */
1178 	ip = ialloc(dp, dirpref(dp->i_fs), uap->dmode);
1179 	if (ip == NULL) {
1180 		iput(dp);
1181 		return;
1182 	}
1183 #ifdef QUOTA
1184 	if (ip->i_dquot != NODQUOT)
1185 		panic("mkdir: dquot");
1186 #endif
1187 	ip->i_flag |= IACC|IUPD|ICHG;
1188 	ip->i_mode = uap->dmode & ~u.u_cmask;
1189 	ip->i_nlink = 2;
1190 	ip->i_uid = u.u_uid;
1191 	ip->i_gid = dp->i_gid;
1192 #ifdef QUOTA
1193 	ip->i_dquot = inoquota(ip);
1194 #endif
1195 	iupdat(ip, &time, &time, 1);
1196 
1197 	/*
1198 	 * Bump link count in parent directory
1199 	 * to reflect work done below.  Should
1200 	 * be done before reference is created
1201 	 * so reparation is possible if we crash.
1202 	 */
1203 	dp->i_nlink++;
1204 	dp->i_flag |= ICHG;
1205 	iupdat(dp, &time, &time, 1);
1206 
1207 	/*
1208 	 * Initialize directory with "."
1209 	 * and ".." from static template.
1210 	 */
1211 	dirtemplate = mastertemplate;
1212 	dirtemplate.dot_ino = ip->i_number;
1213 	dirtemplate.dotdot_ino = dp->i_number;
1214 	u.u_error = rdwri(UIO_WRITE, ip, (caddr_t)&dirtemplate,
1215 		sizeof (dirtemplate), (off_t)0, 1, (int *)0);
1216 	if (u.u_error) {
1217 		dp->i_nlink--;
1218 		dp->i_flag |= ICHG;
1219 		goto bad;
1220 	}
1221 	if (DIRBLKSIZ > ip->i_fs->fs_fsize)
1222 		panic("mkdir: blksize");     /* XXX - should grow with bmap() */
1223 	else
1224 		ip->i_size = DIRBLKSIZ;
1225 	/*
1226 	 * Directory all set up, now
1227 	 * install the entry for it in
1228 	 * the parent directory.
1229 	 */
1230 	u.u_error = direnter(ip, ndp);
1231 	dp = NULL;
1232 	if (u.u_error) {
1233 		ndp->ni_nameiop = LOOKUP | NOCACHE;
1234 		ndp->ni_segflg = UIO_USERSPACE;
1235 		ndp->ni_dirp = uap->name;
1236 		dp = namei(ndp);
1237 		if (dp) {
1238 			dp->i_nlink--;
1239 			dp->i_flag |= ICHG;
1240 		}
1241 	}
1242 bad:
1243 	/*
1244 	 * No need to do an explicit itrunc here,
1245 	 * irele will do this for us because we set
1246 	 * the link count to 0.
1247 	 */
1248 	if (u.u_error) {
1249 		ip->i_nlink = 0;
1250 		ip->i_flag |= ICHG;
1251 	}
1252 	if (dp)
1253 		iput(dp);
1254 	iput(ip);
1255 }
1256 
1257 /*
1258  * Rmdir system call.
1259  */
1260 rmdir()
1261 {
1262 	struct a {
1263 		char	*name;
1264 	} *uap = (struct a *)u.u_ap;
1265 	register struct inode *ip, *dp;
1266 	register struct nameidata *ndp = &u.u_nd;
1267 
1268 	ndp->ni_nameiop = DELETE | LOCKPARENT;
1269 	ndp->ni_segflg = UIO_USERSPACE;
1270 	ndp->ni_dirp = uap->name;
1271 	ip = namei(ndp);
1272 	if (ip == NULL)
1273 		return;
1274 	dp = ndp->ni_pdir;
1275 	/*
1276 	 * No rmdir "." please.
1277 	 */
1278 	if (dp == ip) {
1279 		irele(dp);
1280 		iput(ip);
1281 		u.u_error = EINVAL;
1282 		return;
1283 	}
1284 	if ((ip->i_mode&IFMT) != IFDIR) {
1285 		u.u_error = ENOTDIR;
1286 		goto out;
1287 	}
1288 	/*
1289 	 * Don't remove a mounted on directory.
1290 	 */
1291 	if (ip->i_dev != dp->i_dev) {
1292 		u.u_error = EBUSY;
1293 		goto out;
1294 	}
1295 	/*
1296 	 * Verify the directory is empty (and valid).
1297 	 * (Rmdir ".." won't be valid since
1298 	 *  ".." will contain a reference to
1299 	 *  the current directory and thus be
1300 	 *  non-empty.)
1301 	 */
1302 	if (ip->i_nlink != 2 || !dirempty(ip, dp->i_number)) {
1303 		u.u_error = ENOTEMPTY;
1304 		goto out;
1305 	}
1306 	/*
1307 	 * Delete reference to directory before purging
1308 	 * inode.  If we crash in between, the directory
1309 	 * will be reattached to lost+found,
1310 	 */
1311 	if (dirremove(ndp) == 0)
1312 		goto out;
1313 	dp->i_nlink--;
1314 	dp->i_flag |= ICHG;
1315 	cacheinval(dp);
1316 	iput(dp);
1317 	dp = NULL;
1318 	/*
1319 	 * Truncate inode.  The only stuff left
1320 	 * in the directory is "." and "..".  The
1321 	 * "." reference is inconsequential since
1322 	 * we're quashing it.  The ".." reference
1323 	 * has already been adjusted above.  We've
1324 	 * removed the "." reference and the reference
1325 	 * in the parent directory, but there may be
1326 	 * other hard links so decrement by 2 and
1327 	 * worry about them later.
1328 	 */
1329 	ip->i_nlink -= 2;
1330 	itrunc(ip, (u_long)0);
1331 	cacheinval(ip);
1332 out:
1333 	if (dp)
1334 		iput(dp);
1335 	iput(ip);
1336 }
1337 
1338 struct file *
1339 getinode(fdes)
1340 	int fdes;
1341 {
1342 	struct file *fp;
1343 
1344 	if ((unsigned)fdes >= NOFILE || (fp = u.u_ofile[fdes]) == NULL) {
1345 		u.u_error = EBADF;
1346 		return ((struct file *)0);
1347 	}
1348 	if (fp->f_type != DTYPE_INODE) {
1349 		u.u_error = EINVAL;
1350 		return ((struct file *)0);
1351 	}
1352 	return (fp);
1353 }
1354 
1355 /*
1356  * mode mask for creation of files
1357  */
1358 umask()
1359 {
1360 	register struct a {
1361 		int	mask;
1362 	} *uap = (struct a *)u.u_ap;
1363 
1364 	u.u_r.r_val1 = u.u_cmask;
1365 	u.u_cmask = uap->mask & 07777;
1366 }
1367