xref: /original-bsd/sys/kern/vfs_syscalls.c (revision f0fd5f8a)
1 /*	vfs_syscalls.c	4.44	82/12/19	*/
2 
3 #include "../h/param.h"
4 #include "../h/systm.h"
5 #include "../h/dir.h"
6 #include "../h/user.h"
7 #include "../h/kernel.h"
8 #include "../h/file.h"
9 #include "../h/stat.h"
10 #include "../h/inode.h"
11 #include "../h/fs.h"
12 #include "../h/buf.h"
13 #include "../h/proc.h"
14 #include "../h/quota.h"
15 #include "../h/descrip.h"
16 #include "../h/uio.h"
17 #include "../h/socket.h"
18 #include "../h/socketvar.h"
19 #include "../h/nami.h"
20 
21 /*
22  * Change current working directory (``.'').
23  */
24 chdir()
25 {
26 
27 	chdirec(&u.u_cdir);
28 }
29 
30 /*
31  * Change notion of root (``/'') directory.
32  */
33 chroot()
34 {
35 
36 	if (suser())
37 		chdirec(&u.u_rdir);
38 }
39 
40 /*
41  * Common routine for chroot and chdir.
42  */
43 chdirec(ipp)
44 	register struct inode **ipp;
45 {
46 	register struct inode *ip;
47 	struct a {
48 		char	*fname;
49 	};
50 
51 	ip = namei(uchar, LOOKUP, 1);
52 	if (ip == NULL)
53 		return;
54 	if ((ip->i_mode&IFMT) != IFDIR) {
55 		u.u_error = ENOTDIR;
56 		goto bad;
57 	}
58 	if (access(ip, IEXEC))
59 		goto bad;
60 	iunlock(ip);
61 	if (*ipp)
62 		irele(*ipp);
63 	*ipp = ip;
64 	return;
65 
66 bad:
67 	iput(ip);
68 }
69 
70 /*
71  * Open system call.
72  */
73 open()
74 {
75 	register struct inode *ip;
76 	register struct a {
77 		char	*fname;
78 		int	flags;
79 		int	mode;
80 	} *uap;
81 	int checkpermissions = 1, flags;
82 
83 	uap = (struct a *)u.u_ap;
84 	flags = uap->flags + 1;
85 	if ((flags&FTRUNCATE) && (flags&FWRITE) == 0) {
86 		u.u_error = EINVAL;
87 		return;
88 	}
89 	if (flags&FCREATE) {
90 		ip = namei(uchar, CREATE, 1);
91 		if (ip == NULL) {
92 			if (u.u_error)
93 				return;
94 			ip = maknode(uap->mode&07777&(~ISVTX));
95 			checkpermissions = 0;
96 			flags &= ~FTRUNCATE;
97 		}
98 	} else
99 		ip = namei(uchar, LOOKUP, 1);
100 	if (ip == NULL)
101 		return;
102 	open1(ip, flags, checkpermissions);
103 }
104 
105 #ifndef NOCOMPAT
106 /*
107  * Creat system call.
108  */
109 ocreat()
110 {
111 	register struct inode *ip;
112 	register struct a {
113 		char	*fname;
114 		int	fmode;
115 	} *uap;
116 
117 	uap = (struct a *)u.u_ap;
118 	ip = namei(uchar, CREATE, 1);
119 	if (ip == NULL) {
120 		if (u.u_error)
121 			return;
122 		ip = maknode(uap->fmode&07777&(~ISVTX));
123 		if (ip == NULL)
124 			return;
125 		open1(ip, FWRITE, 0);
126 	} else
127 		open1(ip, FWRITE|FTRUNCATE, 1);
128 }
129 #endif
130 
131 /*
132  * Common code for open and creat.
133  * Check permissions (if we haven't done so already),
134  * allocate an open file structure, and call
135  * the device open routine, if any.
136  */
137 open1(ip, mode, checkpermissions)
138 	register struct inode *ip;
139 	register mode;
140 {
141 	register struct file *fp;
142 	int i, flags;
143 
144 	if (checkpermissions) {
145 		if (mode&FREAD)
146 			if (access(ip, IREAD))
147 				goto bad;
148 		if (mode&FWRITE) {
149 			if (access(ip, IWRITE))
150 				goto bad;
151 			if ((ip->i_mode&IFMT) == IFDIR) {
152 				u.u_error = EISDIR;
153 				goto bad;
154 			}
155 		}
156 	}
157 
158 	/*
159 	 * Check locking on inode.  Release "inode lock"
160 	 * while doing so in case we block inside flocki.
161 	 */
162 	flags = 0;
163 	if (mode&(FSHLOCK|FEXLOCK)) {
164 		iunlock(ip);
165 		flags = flocki(ip, 0, mode);
166 		ilock(ip);
167 		if (u.u_error)
168 			goto bad;
169 	}
170 	if (mode&FTRUNCATE)
171 		itrunc(ip, (u_long)0);
172 	iunlock(ip);
173 	if ((fp = falloc()) == NULL)
174 		goto out;
175 	fp->f_flag = mode & FMODES;
176 	fp->f_type = DTYPE_FILE;
177 	i = u.u_r.r_val1;
178 	fp->f_inode = ip;
179 	u.u_error = openi(ip, mode);
180 	if (u.u_error == 0) {
181 		u.u_pofile[i] = flags;
182 		return;
183 	}
184 	u.u_ofile[i] = NULL;
185 	fp->f_count--;
186 out:
187 	irele(ip);
188 	return;
189 bad:
190 	iput(ip);
191 }
192 
193 /*
194  * Mknod system call
195  */
196 mknod()
197 {
198 	register struct inode *ip;
199 	register struct a {
200 		char	*fname;
201 		int	fmode;
202 		int	dev;
203 	} *uap;
204 
205 	uap = (struct a *)u.u_ap;
206 	if (suser()) {
207 		ip = namei(uchar, CREATE, 0);
208 		if (ip != NULL) {
209 			u.u_error = EEXIST;
210 			goto out;
211 		}
212 	}
213 	if (u.u_error)
214 		return;
215 	ip = maknode(uap->fmode);
216 	if (ip == NULL)
217 		return;
218 	if (uap->dev) {
219 		/*
220 		 * Want to be able to use this to make badblock
221 		 * inodes, so don't truncate the dev number.
222 		 */
223 		ip->i_rdev = uap->dev;
224 		ip->i_flag |= IACC|IUPD|ICHG;
225 	}
226 
227 out:
228 	iput(ip);
229 }
230 
231 /*
232  * link system call
233  */
234 link()
235 {
236 	register struct inode *ip, *xp;
237 	register struct a {
238 		char	*target;
239 		char	*linkname;
240 	} *uap;
241 
242 	uap = (struct a *)u.u_ap;
243 	ip = namei(uchar, LOOKUP, 1); /* well, this routine is doomed anyhow */
244 	if (ip == NULL)
245 		return;
246 	if ((ip->i_mode&IFMT) == IFDIR && !suser()) {
247 		iput(ip);
248 		return;
249 	}
250 	ip->i_nlink++;
251 	ip->i_flag |= ICHG;
252 	iupdat(ip, &time, &time, 1);
253 	iunlock(ip);
254 	u.u_dirp = (caddr_t)uap->linkname;
255 	xp = namei(uchar, CREATE, 0);
256 	if (xp != NULL) {
257 		u.u_error = EEXIST;
258 		iput(xp);
259 		goto out;
260 	}
261 	if (u.u_error)
262 		goto out;
263 	if (u.u_pdir->i_dev != ip->i_dev) {
264 		iput(u.u_pdir);
265 		u.u_error = EXDEV;
266 		goto out;
267 	}
268 	direnter(ip);
269 out:
270 	if (u.u_error) {
271 		ip->i_nlink--;
272 		ip->i_flag |= ICHG;
273 	}
274 	irele(ip);
275 }
276 
277 /*
278  * symlink -- make a symbolic link
279  */
280 symlink()
281 {
282 	register struct a {
283 		char	*target;
284 		char	*linkname;
285 	} *uap;
286 	register struct inode *ip;
287 	register char *tp;
288 	register c, nc;
289 
290 	uap = (struct a *)u.u_ap;
291 	tp = uap->target;
292 	nc = 0;
293 	while (c = fubyte(tp)) {
294 		if (c < 0) {
295 			u.u_error = EFAULT;
296 			return;
297 		}
298 		tp++;
299 		nc++;
300 	}
301 	u.u_dirp = uap->linkname;
302 	ip = namei(uchar, CREATE, 0);
303 	if (ip) {
304 		iput(ip);
305 		u.u_error = EEXIST;
306 		return;
307 	}
308 	if (u.u_error)
309 		return;
310 	ip = maknode(IFLNK | 0777);
311 	if (ip == NULL)
312 		return;
313 	u.u_error = rdwri(UIO_WRITE, ip, uap->target, nc, 0, 0, (int *)0);
314 	/* handle u.u_error != 0 */
315 	iput(ip);
316 }
317 
318 /*
319  * Unlink system call.
320  * Hard to avoid races here, especially
321  * in unlinking directories.
322  */
323 unlink()
324 {
325 	struct a {
326 		char	*fname;
327 	};
328 	register struct inode *ip, *dp;
329 
330 	ip = namei(uchar, DELETE | LOCKPARENT, 0);
331 	if (ip == NULL)
332 		return;
333 	dp = u.u_pdir;
334 	if ((ip->i_mode&IFMT) == IFDIR && !suser())
335 		goto out;
336 	/*
337 	 * Don't unlink a mounted file.
338 	 */
339 	if (ip->i_dev != dp->i_dev) {
340 		u.u_error = EBUSY;
341 		goto out;
342 	}
343 	if (ip->i_flag&ITEXT)
344 		xrele(ip);	/* try once to free text */
345 	if (dirremove()) {
346 		ip->i_nlink--;
347 		ip->i_flag |= ICHG;
348 	}
349 out:
350 	if (dp == ip)
351 		irele(ip);
352 	else
353 		iput(ip);
354 	iput(dp);
355 }
356 
357 /*
358  * Seek system call
359  */
360 lseek()
361 {
362 	register struct file *fp;
363 	register struct a {
364 		int	fd;
365 		off_t	off;
366 		int	sbase;
367 	} *uap;
368 
369 	uap = (struct a *)u.u_ap;
370 	fp = getf(uap->fd);
371 	if (fp == NULL)
372 		return;
373 	if (fp->f_type == DTYPE_SOCKET) {
374 		u.u_error = ESPIPE;
375 		return;
376 	}
377 	if (uap->sbase == FSEEK_RELATIVE)
378 		uap->off += fp->f_offset;
379 	else if (uap->sbase == FSEEK_EOF)
380 		uap->off += fp->f_inode->i_size;
381 	fp->f_offset = uap->off;
382 	u.u_r.r_off = uap->off;
383 }
384 
385 /*
386  * Access system call
387  */
388 saccess()
389 {
390 	register svuid, svgid;
391 	register struct inode *ip;
392 	register struct a {
393 		char	*fname;
394 		int	fmode;
395 	} *uap;
396 
397 	uap = (struct a *)u.u_ap;
398 	svuid = u.u_uid;
399 	svgid = u.u_gid;
400 	u.u_uid = u.u_ruid;
401 	u.u_gid = u.u_rgid;
402 	ip = namei(uchar, LOOKUP, 1);
403 	if (ip != NULL) {
404 		if ((uap->fmode&FACCESS_READ) && access(ip, IREAD))
405 			goto done;
406 		if ((uap->fmode&FACCESS_WRITE) && access(ip, IWRITE))
407 			goto done;
408 		if ((uap->fmode&FACCESS_EXECUTE) && access(ip, IEXEC))
409 			goto done;
410 done:
411 		iput(ip);
412 	}
413 	u.u_uid = svuid;
414 	u.u_gid = svgid;
415 }
416 
417 /*
418  * the fstat system call.
419  */
420 fstat()
421 {
422 	register struct file *fp;
423 	register struct a {
424 		int	fd;
425 		struct stat *sb;
426 	} *uap;
427 
428 	uap = (struct a *)u.u_ap;
429 	fp = getf(uap->fd);
430 	if (fp == NULL)
431 		return;
432 	if (fp->f_type == DTYPE_SOCKET)
433 		u.u_error = sostat(fp->f_socket, uap->sb);
434 	else
435 		stat1(fp->f_inode, uap->sb);
436 }
437 
438 /*
439  * Stat system call.  This version follows links.
440  */
441 stat()
442 {
443 	register struct inode *ip;
444 	register struct a {
445 		char	*fname;
446 		struct stat *sb;
447 	} *uap;
448 
449 	uap = (struct a *)u.u_ap;
450 	ip = namei(uchar, LOOKUP, 1);
451 	if (ip == NULL)
452 		return;
453 	stat1(ip, uap->sb);
454 	iput(ip);
455 }
456 
457 /*
458  * Lstat system call.  This version does not follow links.
459  */
460 lstat()
461 {
462 	register struct inode *ip;
463 	register struct a {
464 		char	*fname;
465 		struct stat *sb;
466 	} *uap;
467 
468 	uap = (struct a *)u.u_ap;
469 	ip = namei(uchar, LOOKUP, 0);
470 	if (ip == NULL)
471 		return;
472 	stat1(ip, uap->sb);
473 	iput(ip);
474 }
475 
476 /*
477  * The basic routine for fstat and stat:
478  * get the inode and pass appropriate parts back.
479  */
480 stat1(ip, ub)
481 	register struct inode *ip;
482 	struct stat *ub;
483 {
484 	struct stat ds;
485 
486 	IUPDAT(ip, &time, &time, 0);
487 	/*
488 	 * Copy from inode table
489 	 */
490 	ds.st_dev = ip->i_dev;
491 	ds.st_ino = ip->i_number;
492 	ds.st_mode = ip->i_mode;
493 	ds.st_nlink = ip->i_nlink;
494 	ds.st_uid = ip->i_uid;
495 	ds.st_gid = ip->i_gid;
496 	ds.st_rdev = (dev_t)ip->i_rdev;
497 	ds.st_size = ip->i_size;
498 	ds.st_atime = ip->i_atime;
499 	ds.st_mtime = ip->i_mtime;
500 	ds.st_ctime = ip->i_ctime;
501 	/* this doesn't belong here */
502 	if ((ip->i_mode&IFMT) == IFBLK)
503 		ds.st_blksize = BLKDEV_IOSIZE;
504 	else if ((ip->i_mode&IFMT) == IFCHR)
505 		ds.st_blksize = MAXBSIZE;
506 	else
507 		ds.st_blksize = ip->i_fs->fs_bsize;
508 	if (copyout((caddr_t)&ds, (caddr_t)ub, sizeof(ds)) < 0)
509 		u.u_error = EFAULT;
510 }
511 
512 /*
513  * Return target name of a symbolic link
514  */
515 readlink()
516 {
517 	register struct inode *ip;
518 	register struct a {
519 		char	*name;
520 		char	*buf;
521 		int	count;
522 	} *uap = (struct a *)u.u_ap;
523 	int resid;
524 
525 	ip = namei(uchar, LOOKUP, 0);
526 	if (ip == NULL)
527 		return;
528 	if ((ip->i_mode&IFMT) != IFLNK) {
529 		u.u_error = ENXIO;
530 		goto out;
531 	}
532 	u.u_error = rdwri(UIO_READ, ip, uap->buf, uap->count, 0, 0, &resid);
533 out:
534 	iput(ip);
535 	u.u_r.r_val1 = uap->count - resid;
536 }
537 
538 /*
539  * Change mode of a file given path name.
540  */
541 chmod()
542 {
543 	struct inode *ip;
544 	struct a {
545 		char	*fname;
546 		int	fmode;
547 	} *uap;
548 
549 	uap = (struct a *)u.u_ap;
550 	if ((ip = owner(1)) == NULL)
551 		return;
552 	chmod1(ip, uap->fmode);
553 	iput(ip);
554 }
555 
556 /*
557  * Change mode of a file given a file descriptor.
558  */
559 fchmod()
560 {
561 	struct a {
562 		int	fd;
563 		int	fmode;
564 	} *uap;
565 	register struct inode *ip;
566 	register struct file *fp;
567 
568 	uap = (struct a *)u.u_ap;
569 	fp = getf(uap->fd);
570 	if (fp == NULL)
571 		return;
572 	if (fp->f_type == DTYPE_SOCKET) {
573 		u.u_error = EINVAL;
574 		return;
575 	}
576 	ip = fp->f_inode;
577 	if (u.u_uid != ip->i_uid && !suser())
578 		return;
579 	ilock(ip);
580 	chmod1(ip, uap->fmode);
581 	iunlock(ip);
582 }
583 
584 /*
585  * Change the mode on a file.
586  * Inode must be locked before calling.
587  */
588 chmod1(ip, mode)
589 	register struct inode *ip;
590 	register int mode;
591 {
592 	register int *gp;
593 
594 	ip->i_mode &= ~07777;
595 	if (u.u_uid) {
596 		mode &= ~ISVTX;
597 		for (gp = u.u_groups; gp < &u.u_groups[NGROUPS]; gp++)
598 			if (*gp == ip->i_gid)
599 				goto ok;
600 		mode &= ~ISGID;
601 ok:
602 		;
603 #ifdef MUSH
604 		if (u.u_quota->q_syflags & QF_UMASK && u.u_uid != 0 &&
605 		    (ip->i_mode & IFMT) != IFCHR)
606 			mode &= ~u.u_cmask;
607 #endif
608 	}
609 	ip->i_mode |= mode&07777;
610 	ip->i_flag |= ICHG;
611 	if (ip->i_flag&ITEXT && (ip->i_mode&ISVTX)==0)
612 		xrele(ip);
613 }
614 
615 /*
616  * Set ownership given a path name.
617  */
618 chown()
619 {
620 	struct inode *ip;
621 	struct a {
622 		char	*fname;
623 		int	uid;
624 		int	gid;
625 	} *uap;
626 
627 	uap = (struct a *)u.u_ap;
628 	if (!suser() || (ip = owner(0)) == NULL)
629 		return;
630 	chown1(ip, uap->uid, uap->gid);
631 	iput(ip);
632 }
633 
634 /*
635  * Set ownership given a file descriptor.
636  */
637 fchown()
638 {
639 	struct a {
640 		int	fd;
641 		int	uid;
642 		int	gid;
643 	} *uap;
644 	register struct inode *ip;
645 	register struct file *fp;
646 
647 	uap = (struct a *)u.u_ap;
648 	fp = getf(uap->fd);
649 	if (fp == NULL)
650 		return;
651 	if (fp->f_type == DTYPE_SOCKET) {
652 		u.u_error = EINVAL;
653 		return;
654 	}
655 	ip = fp->f_inode;
656 	if (!suser())
657 		return;
658 	ilock(ip);
659 	chown1(ip, uap->uid, uap->gid);
660 	iunlock(ip);
661 }
662 
663 /*
664  * Perform chown operation on inode ip;
665  * inode must be locked prior to call.
666  */
667 chown1(ip, uid, gid)
668 	register struct inode *ip;
669 	int uid, gid;
670 {
671 #ifdef QUOTA
672 	register long change;
673 
674 	/*
675 	 * This doesn't allow for holes in files (which hopefully don't
676 	 * happen often in files that we chown), and is not accurate anyway
677 	 * (eg: it totally ignores 3 level indir blk files - but hopefully
678 	 * noone who can make a file that big will have a quota)
679 	 */
680 	if (ip->i_uid == uid)
681 		change = 0;
682 	else {
683 		register struct fs *fs = ip->i_fs;
684 
685 		if (ip->i_size > (change = NDADDR * fs->fs_bsize)) {
686 			register off_t size;
687 
688 			size = blkroundup(fs, ip->i_size) - change;
689 			change += size;
690 			change += fs->fs_bsize;
691 			/* this assumes NIADDR <= 2 */
692 			if (size > NINDIR(fs) * fs->fs_bsize)
693 				change += fs->fs_bsize;
694 		} else
695 			change = fragroundup(fs, ip->i_size);
696 		change /= DEV_BSIZE;
697 	}
698 	(void)chkdq(ip, -change, 1);
699 	(void)chkiq(ip->i_dev, ip, ip->i_uid, 1);
700 	dqrele(ip->i_dquot);
701 #endif
702 	/*
703 	 * keep uid/gid's in sane range -- no err,
704 	 * so chown(file, uid, -1) will do something useful
705 	 */
706 	if (uid >= 0 && uid <= 32767)	/* should have a constant */
707 		ip->i_uid = uid;
708 	if (gid >= 0 && gid <= 32767)	/* same here */
709 		ip->i_gid = gid;
710 	ip->i_flag |= ICHG;
711 	if (u.u_ruid != 0)
712 		ip->i_mode &= ~(ISUID|ISGID);
713 #ifdef QUOTA
714 	ip->i_dquot = inoquota(ip);
715 	(void)chkdq(ip, change, 1);
716 	(void)chkiq(ip->i_dev, (struct inode *)NULL, uid, 1);
717 #endif
718 }
719 
720 /*
721  * Set IUPD and IACC times on file.
722  * Can't set ICHG.
723  */
724 outime()
725 {
726 	register struct a {
727 		char	*fname;
728 		time_t	*tptr;
729 	} *uap;
730 	register struct inode *ip;
731 	time_t tv[2];
732 	struct timeval tv0, tv1;
733 
734 	uap = (struct a *)u.u_ap;
735 	if ((ip = owner(1)) == NULL)
736 		return;
737 	if (copyin((caddr_t)uap->tptr, (caddr_t)tv, sizeof(tv))) {
738 		u.u_error = EFAULT;
739 	} else {
740 		ip->i_flag |= IACC|IUPD|ICHG;
741 		tv0.tv_sec = tv[0]; tv0.tv_usec = 0;
742 		tv1.tv_sec = tv[1]; tv1.tv_usec = 0;
743 		iupdat(ip, &tv0, &tv1, 0);
744 	}
745 	iput(ip);
746 }
747 
748 /*
749  * Flush any pending I/O.
750  */
751 sync()
752 {
753 
754 	update();
755 }
756 
757 /*
758  * Apply an advisory lock on a file descriptor.
759  */
760 flock()
761 {
762 	struct a {
763 		int	fd;
764 		int	how;
765 	} *uap;
766 	register struct file *fp;
767 	register int cmd, flags;
768 
769 	uap = (struct a *)u.u_ap;
770 	fp = getf(uap->fd);
771 	if (fp == NULL)
772 		return;
773 	if (fp->f_type == DTYPE_SOCKET) {		/* XXX */
774 		u.u_error = EINVAL;
775 		return;
776 	}
777 	cmd = uap->how;
778 	flags = u.u_pofile[uap->fd] & (UF_SHLOCK|UF_EXLOCK);
779 	if (cmd&FUNLOCK) {
780 		if (flags == 0) {
781 			u.u_error = EINVAL;
782 			return;
783 		}
784 		funlocki(fp->f_inode, flags);
785 		u.u_pofile[uap->fd] &= ~(UF_SHLOCK|UF_EXLOCK);
786 		return;
787 	}
788 	/*
789 	 * No reason to write lock a file we've already
790 	 * write locked, similarly with a read lock.
791 	 */
792 	if ((flags&UF_EXLOCK) && (cmd&FEXLOCK) ||
793 	    (flags&UF_SHLOCK) && (cmd&FSHLOCK))
794 		return;
795 	u.u_pofile[uap->fd] = flocki(fp->f_inode, u.u_pofile[uap->fd], cmd);
796 }
797 
798 /*
799  * Truncate a file given its path name.
800  */
801 truncate()
802 {
803 	struct a {
804 		char	*fname;
805 		u_long	length;
806 	} *uap = (struct a *)u.u_ap;
807 	struct inode *ip;
808 
809 	ip = namei(uchar, LOOKUP, 1);
810 	if (ip == NULL)
811 		return;
812 	if (access(ip, IWRITE))
813 		goto bad;
814 	if ((ip->i_mode&IFMT) == IFDIR) {
815 		u.u_error = EISDIR;
816 		goto bad;
817 	}
818 	itrunc(ip, uap->length);
819 bad:
820 	iput(ip);
821 }
822 
823 /*
824  * Truncate a file given a file descriptor.
825  */
826 ftruncate()
827 {
828 	struct a {
829 		int	fd;
830 		u_long	length;
831 	} *uap = (struct a *)u.u_ap;
832 	struct inode *ip;
833 	struct file *fp;
834 
835 	fp = getf(uap->fd);
836 	if (fp == NULL)
837 		return;
838 	if (fp->f_type == DTYPE_SOCKET) {
839 		u.u_error = EINVAL;
840 		return;
841 	}
842 	if ((fp->f_flag&FWRITE) == 0) {
843 		u.u_error = EINVAL;
844 		return;
845 	}
846 	ip = fp->f_inode;
847 	ilock(ip);
848 	itrunc(ip, uap->length);
849 	iunlock(ip);
850 }
851 
852 /*
853  * Synch an open file.
854  */
855 fsync()
856 {
857 	struct a {
858 		int	fd;
859 	} *uap = (struct a *)u.u_ap;
860 	struct inode *ip;
861 	struct file *fp;
862 
863 	fp = getf(uap->fd);
864 	if (fp == NULL)
865 		return;
866 	if (fp->f_type == DTYPE_SOCKET) {
867 		u.u_error = EINVAL;
868 		return;
869 	}
870 	ip = fp->f_inode;
871 	ilock(ip);
872 	syncip(ip);
873 	iunlock(ip);
874 }
875 
876 /*
877  * Rename system call.
878  * 	rename("foo", "bar");
879  * is essentially
880  *	unlink("bar");
881  *	link("foo", "bar");
882  *	unlink("foo");
883  * but ``atomically''.  Can't do full commit without saving state in the
884  * inode on disk which isn't feasible at this time.  Best we can do is
885  * always guarantee the target exists.
886  *
887  * Basic algorithm is:
888  *
889  * 1) Bump link count on source while we're linking it to the
890  *    target.  This also insure the inode won't be deleted out
891  *    from underneath us while we work.
892  * 2) Link source to destination.  If destination already exists,
893  *    delete it first.
894  * 3) Unlink source reference to inode if still around.
895  * 4) If a directory was moved and the parent of the destination
896  *    is different from the source, patch the ".." entry in the
897  *    directory.
898  *
899  * Source and destination must either both be directories, or both
900  * not be directories.  If target is a directory, it must be empty.
901  */
902 rename()
903 {
904 	struct a {
905 		char	*from;
906 		char	*to;
907 	} *uap;
908 	register struct inode *ip, *xp, *dp;
909 	int oldparent, parentdifferent, doingdirectory;
910 
911 	uap = (struct a *)u.u_ap;
912 	ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
913 	if (ip == NULL)
914 		return;
915 	dp = u.u_pdir;
916 	oldparent = 0, doingdirectory = 0;
917 	if ((ip->i_mode&IFMT) == IFDIR) {
918 		register struct direct *d;
919 
920 		d = &u.u_dent;
921 		/*
922 		 * Avoid "." and ".." for obvious reasons.
923 		 */
924 		if (d->d_name[0] == '.') {
925 			if (d->d_namlen == 1 ||
926 			    (d->d_namlen == 2 && d->d_name[1] == '.')) {
927 				u.u_error = EINVAL;
928 				iput(ip);
929 				return;
930 			}
931 		}
932 		oldparent = dp->i_number;
933 		doingdirectory++;
934 	}
935 	irele(dp);
936 
937 	/*
938 	 * 1) Bump link count while we're moving stuff
939 	 *    around.  If we crash somewhere before
940 	 *    completing our work, the link count
941 	 *    may be wrong, but correctable.
942 	 */
943 	ip->i_nlink++;
944 	ip->i_flag |= ICHG;
945 	iupdat(ip, &time, &time, 1);
946 	iunlock(ip);
947 
948 	/*
949 	 * When the target exists, both the directory
950 	 * and target inodes are returned locked.
951 	 */
952 	u.u_dirp = (caddr_t)uap->to;
953 	xp = namei(uchar, CREATE | LOCKPARENT, 0);
954 	if (u.u_error)
955 		goto out;
956 	dp = u.u_pdir;
957 	/*
958 	 * 2) If target doesn't exist, link the target
959 	 *    to the source and unlink the source.
960 	 *    Otherwise, rewrite the target directory
961 	 *    entry to reference the source inode and
962 	 *    expunge the original entry's existence.
963 	 */
964 	parentdifferent = oldparent != dp->i_number;
965 	if (xp == NULL) {
966 		if (dp->i_dev != ip->i_dev) {
967 			u.u_error = EXDEV;
968 			goto bad;
969 		}
970 		/*
971 		 * Account for ".." in directory.
972 		 * When source and destination have the
973 		 * same parent we don't fool with the
974 		 * link count -- this isn't required
975 		 * because we do a similar check below.
976 		 */
977 		if (doingdirectory && parentdifferent) {
978 			dp->i_nlink++;
979 			dp->i_flag |= ICHG;
980 			iupdat(dp, &time, &time, 1);
981 		}
982 		direnter(ip);
983 		if (u.u_error)
984 			goto out;
985 	} else {
986 		if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) {
987 			u.u_error = EXDEV;
988 			goto bad;
989 		}
990 		/*
991 		 * Target must be empty if a directory.
992 		 * Also, insure source and target are
993 		 * compatible (both directories, or both
994 		 * not directories).
995 		 */
996 		if ((xp->i_mode&IFMT) == IFDIR) {
997 			if (!dirempty(xp)) {
998 				u.u_error = ENOTEMPTY;
999 				goto bad;
1000 			}
1001 			if (!doingdirectory) {
1002 				u.u_error = ENOTDIR;
1003 				goto bad;
1004 			}
1005 		} else if (doingdirectory) {
1006 			u.u_error = EISDIR;
1007 			goto bad;
1008 		}
1009 		dirrewrite(dp, ip);
1010 		if (u.u_error)
1011 			goto bad1;
1012 		/*
1013 		 * If this is a directory we know it is
1014 		 * empty and we can squash the inode and
1015 		 * any space associated with it.  Otherwise,
1016 		 * we've got a plain file and the link count
1017 		 * simply needs to be adjusted.
1018 		 */
1019 		if (doingdirectory) {
1020 			xp->i_nlink = 0;
1021 			itrunc(xp, (u_long)0);
1022 		} else
1023 			xp->i_nlink--;
1024 		xp->i_flag |= ICHG;
1025 		iput(xp);
1026 	}
1027 
1028 	/*
1029 	 * 3) Unlink the source.
1030 	 */
1031 	u.u_dirp = uap->from;
1032 	dp = namei(uchar, DELETE, 0);
1033 	/*
1034 	 * Insure directory entry still exists and
1035 	 * has not changed since the start of all
1036 	 * this.  If either has occured, forget about
1037 	 * about deleting the original entry and just
1038 	 * adjust the link count in the inode.
1039 	 */
1040 	if (dp == NULL || u.u_dent.d_ino != ip->i_number) {
1041 		ip->i_nlink--;
1042 		ip->i_flag |= ICHG;
1043 	} else {
1044 		/*
1045 		 * If source is a directory, must adjust
1046 		 * link count of parent directory also.
1047 		 * If target didn't exist and source and
1048 		 * target have the same parent, then we
1049 		 * needn't touch the link count, it all
1050 		 * balances out in the end.  Otherwise, we
1051 		 * must do so to reflect deletion of ".."
1052 		 * done above.
1053 		 */
1054 		if (doingdirectory && (xp != NULL || parentdifferent)) {
1055 			dp->i_nlink--;
1056 			dp->i_flag |= ICHG;
1057 		}
1058 		if (dirremove()) {
1059 			ip->i_nlink--;
1060 			ip->i_flag |= ICHG;
1061 		}
1062 	}
1063 	irele(ip);
1064 	if (dp)
1065 		iput(dp);
1066 
1067 	/*
1068 	 * 4) Renaming a directory with the parent
1069 	 *    different requires ".." to be rewritten.
1070 	 *    The window is still there for ".." to
1071 	 *    be inconsistent, but this is unavoidable,
1072 	 *    and a lot shorter than when it was done
1073 	 *    in a user process.
1074 	 */
1075 	if (doingdirectory && parentdifferent && u.u_error == 0) {
1076 		struct dirtemplate dirbuf;
1077 
1078 		u.u_dirp = uap->to;
1079 		ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
1080 		if (ip == NULL) {
1081 			printf("rename: .. went away\n");
1082 			return;
1083 		}
1084 		dp = u.u_pdir;
1085 		if ((ip->i_mode&IFMT) != IFDIR) {
1086 			printf("rename: .. not a directory\n");
1087 			goto stuck;
1088 		}
1089 		u.u_error = rdwri(UIO_READ, ip, (caddr_t)&dirbuf,
1090 			sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1091 		if (u.u_error == 0) {
1092 			dirbuf.dotdot_ino = dp->i_number;
1093 			(void) rdwri(UIO_WRITE, ip, (caddr_t)&dirbuf,
1094 			  sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1095 		}
1096 stuck:
1097 		irele(dp);
1098 		iput(ip);
1099 	}
1100 	return;
1101 bad:
1102 	iput(u.u_pdir);
1103 bad1:
1104 	if (xp)
1105 		irele(xp);
1106 out:
1107 	ip->i_nlink--;
1108 	ip->i_flag |= ICHG;
1109 	irele(ip);
1110 }
1111 
1112 /*
1113  * Make a new file.
1114  */
1115 struct inode *
1116 maknode(mode)
1117 	int mode;
1118 {
1119 	register struct inode *ip;
1120 	ino_t ipref;
1121 
1122 	if ((mode & IFMT) == IFDIR)
1123 		ipref = dirpref(u.u_pdir->i_fs);
1124 	else
1125 		ipref = u.u_pdir->i_number;
1126 	ip = ialloc(u.u_pdir, ipref, mode);
1127 	if (ip == NULL) {
1128 		iput(u.u_pdir);
1129 		return (NULL);
1130 	}
1131 #ifdef QUOTA
1132 	if (ip->i_dquot != NODQUOT)
1133 		panic("maknode: dquot");
1134 #endif
1135 	ip->i_flag |= IACC|IUPD|ICHG;
1136 	if ((mode & IFMT) == 0)
1137 		mode |= IFREG;
1138 	ip->i_mode = mode & ~u.u_cmask;
1139 	ip->i_nlink = 1;
1140 	ip->i_uid = u.u_uid;
1141 	ip->i_gid = u.u_pdir->i_gid;
1142 #ifdef QUOTA
1143 	ip->i_dquot = inoquota(ip);
1144 #endif
1145 
1146 	/*
1147 	 * Make sure inode goes to disk before directory entry.
1148 	 */
1149 	iupdat(ip, &time, &time, 1);
1150 	direnter(ip);
1151 	if (u.u_error) {
1152 		/*
1153 		 * write error occurred trying to update directory
1154 		 * so must deallocate the inode
1155 		 */
1156 		ip->i_nlink = 0;
1157 		ip->i_flag |= ICHG;
1158 		iput(ip);
1159 		return (NULL);
1160 	}
1161 	return (ip);
1162 }
1163