xref: /original-bsd/sys/ufs/ffs/ffs_vnops.c (revision d201b27a)
1 /*	ffs_vnops.c	4.47	83/01/01	*/
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_spare1 = 0;
500 	ds.st_mtime = ip->i_mtime;
501 	ds.st_spare2 = 0;
502 	ds.st_ctime = ip->i_ctime;
503 	ds.st_spare3 = 0;
504 	/* this doesn't belong here */
505 	if ((ip->i_mode&IFMT) == IFBLK)
506 		ds.st_blksize = BLKDEV_IOSIZE;
507 	else if ((ip->i_mode&IFMT) == IFCHR)
508 		ds.st_blksize = MAXBSIZE;
509 	else
510 		ds.st_blksize = ip->i_fs->fs_bsize;
511 	ds.st_spare4[0] = ds.st_spare4[1] = ds.st_spare4[2] = 0;
512 	u.u_error = copyout((caddr_t)&ds, (caddr_t)ub, sizeof(ds));
513 }
514 
515 /*
516  * Return target name of a symbolic link
517  */
518 readlink()
519 {
520 	register struct inode *ip;
521 	register struct a {
522 		char	*name;
523 		char	*buf;
524 		int	count;
525 	} *uap = (struct a *)u.u_ap;
526 	int resid;
527 
528 	ip = namei(uchar, LOOKUP, 0);
529 	if (ip == NULL)
530 		return;
531 	if ((ip->i_mode&IFMT) != IFLNK) {
532 		u.u_error = ENXIO;
533 		goto out;
534 	}
535 	u.u_error = rdwri(UIO_READ, ip, uap->buf, uap->count, 0, 0, &resid);
536 out:
537 	iput(ip);
538 	u.u_r.r_val1 = uap->count - resid;
539 }
540 
541 /*
542  * Change mode of a file given path name.
543  */
544 chmod()
545 {
546 	struct inode *ip;
547 	struct a {
548 		char	*fname;
549 		int	fmode;
550 	} *uap;
551 
552 	uap = (struct a *)u.u_ap;
553 	if ((ip = owner(1)) == NULL)
554 		return;
555 	chmod1(ip, uap->fmode);
556 	iput(ip);
557 }
558 
559 /*
560  * Change mode of a file given a file descriptor.
561  */
562 fchmod()
563 {
564 	struct a {
565 		int	fd;
566 		int	fmode;
567 	} *uap;
568 	register struct inode *ip;
569 	register struct file *fp;
570 
571 	uap = (struct a *)u.u_ap;
572 	fp = getf(uap->fd);
573 	if (fp == NULL)
574 		return;
575 	if (fp->f_type == DTYPE_SOCKET) {
576 		u.u_error = EINVAL;
577 		return;
578 	}
579 	ip = fp->f_inode;
580 	if (u.u_uid != ip->i_uid && !suser())
581 		return;
582 	ilock(ip);
583 	chmod1(ip, uap->fmode);
584 	iunlock(ip);
585 }
586 
587 /*
588  * Change the mode on a file.
589  * Inode must be locked before calling.
590  */
591 chmod1(ip, mode)
592 	register struct inode *ip;
593 	register int mode;
594 {
595 	register int *gp;
596 
597 	ip->i_mode &= ~07777;
598 	if (u.u_uid) {
599 		mode &= ~ISVTX;
600 		for (gp = u.u_groups; gp < &u.u_groups[NGROUPS]; gp++)
601 			if (*gp == ip->i_gid)
602 				goto ok;
603 		mode &= ~ISGID;
604 ok:
605 		;
606 #ifdef MUSH
607 		if (u.u_quota->q_syflags & QF_UMASK && u.u_uid != 0 &&
608 		    (ip->i_mode & IFMT) != IFCHR)
609 			mode &= ~u.u_cmask;
610 #endif
611 	}
612 	ip->i_mode |= mode&07777;
613 	ip->i_flag |= ICHG;
614 	if (ip->i_flag&ITEXT && (ip->i_mode&ISVTX)==0)
615 		xrele(ip);
616 }
617 
618 /*
619  * Set ownership given a path name.
620  */
621 chown()
622 {
623 	struct inode *ip;
624 	struct a {
625 		char	*fname;
626 		int	uid;
627 		int	gid;
628 	} *uap;
629 
630 	uap = (struct a *)u.u_ap;
631 	if (!suser() || (ip = owner(0)) == NULL)
632 		return;
633 	chown1(ip, uap->uid, uap->gid);
634 	iput(ip);
635 }
636 
637 /*
638  * Set ownership given a file descriptor.
639  */
640 fchown()
641 {
642 	struct a {
643 		int	fd;
644 		int	uid;
645 		int	gid;
646 	} *uap;
647 	register struct inode *ip;
648 	register struct file *fp;
649 
650 	uap = (struct a *)u.u_ap;
651 	fp = getf(uap->fd);
652 	if (fp == NULL)
653 		return;
654 	if (fp->f_type == DTYPE_SOCKET) {
655 		u.u_error = EINVAL;
656 		return;
657 	}
658 	ip = fp->f_inode;
659 	if (!suser())
660 		return;
661 	ilock(ip);
662 	chown1(ip, uap->uid, uap->gid);
663 	iunlock(ip);
664 }
665 
666 /*
667  * Perform chown operation on inode ip;
668  * inode must be locked prior to call.
669  */
670 chown1(ip, uid, gid)
671 	register struct inode *ip;
672 	int uid, gid;
673 {
674 #ifdef QUOTA
675 	register long change;
676 
677 	/*
678 	 * This doesn't allow for holes in files (which hopefully don't
679 	 * happen often in files that we chown), and is not accurate anyway
680 	 * (eg: it totally ignores 3 level indir blk files - but hopefully
681 	 * noone who can make a file that big will have a quota)
682 	 */
683 	if (ip->i_uid == uid)
684 		change = 0;
685 	else {
686 		register struct fs *fs = ip->i_fs;
687 
688 		if (ip->i_size > (change = NDADDR * fs->fs_bsize)) {
689 			register off_t size;
690 
691 			size = blkroundup(fs, ip->i_size) - change;
692 			change += size;
693 			change += fs->fs_bsize;
694 			/* this assumes NIADDR <= 2 */
695 			if (size > NINDIR(fs) * fs->fs_bsize)
696 				change += fs->fs_bsize;
697 		} else
698 			change = fragroundup(fs, ip->i_size);
699 		change /= DEV_BSIZE;
700 	}
701 	(void)chkdq(ip, -change, 1);
702 	(void)chkiq(ip->i_dev, ip, ip->i_uid, 1);
703 	dqrele(ip->i_dquot);
704 #endif
705 	/*
706 	 * keep uid/gid's in sane range -- no err,
707 	 * so chown(file, uid, -1) will do something useful
708 	 */
709 	if (uid >= 0 && uid <= 32767)	/* should have a constant */
710 		ip->i_uid = uid;
711 	if (gid >= 0 && gid <= 32767)	/* same here */
712 		ip->i_gid = gid;
713 	ip->i_flag |= ICHG;
714 	if (u.u_ruid != 0)
715 		ip->i_mode &= ~(ISUID|ISGID);
716 #ifdef QUOTA
717 	ip->i_dquot = inoquota(ip);
718 	(void)chkdq(ip, change, 1);
719 	(void)chkiq(ip->i_dev, (struct inode *)NULL, uid, 1);
720 #endif
721 }
722 
723 /*
724  * Set IUPD and IACC times on file.
725  * Can't set ICHG.
726  */
727 outime()
728 {
729 	register struct a {
730 		char	*fname;
731 		time_t	*tptr;
732 	} *uap;
733 	register struct inode *ip;
734 	time_t tv[2];
735 	struct timeval tv0, tv1;
736 
737 	uap = (struct a *)u.u_ap;
738 	if ((ip = owner(1)) == NULL)
739 		return;
740 	u.u_error = copyin((caddr_t)uap->tptr, (caddr_t)tv, sizeof(tv));
741 	if (u.u_error == 0) {
742 		ip->i_flag |= IACC|IUPD|ICHG;
743 		tv0.tv_sec = tv[0]; tv0.tv_usec = 0;
744 		tv1.tv_sec = tv[1]; tv1.tv_usec = 0;
745 		iupdat(ip, &tv0, &tv1, 0);
746 	}
747 	iput(ip);
748 }
749 
750 /*
751  * Flush any pending I/O.
752  */
753 sync()
754 {
755 
756 	update();
757 }
758 
759 /*
760  * Apply an advisory lock on a file descriptor.
761  */
762 flock()
763 {
764 	struct a {
765 		int	fd;
766 		int	how;
767 	} *uap;
768 	register struct file *fp;
769 	register int cmd, flags;
770 
771 	uap = (struct a *)u.u_ap;
772 	fp = getf(uap->fd);
773 	if (fp == NULL)
774 		return;
775 	if (fp->f_type == DTYPE_SOCKET) {		/* XXX */
776 		u.u_error = EINVAL;
777 		return;
778 	}
779 	cmd = uap->how;
780 	flags = u.u_pofile[uap->fd] & (UF_SHLOCK|UF_EXLOCK);
781 	if (cmd&FUNLOCK) {
782 		if (flags == 0) {
783 			u.u_error = EINVAL;
784 			return;
785 		}
786 		funlocki(fp->f_inode, flags);
787 		u.u_pofile[uap->fd] &= ~(UF_SHLOCK|UF_EXLOCK);
788 		return;
789 	}
790 	/*
791 	 * No reason to write lock a file we've already
792 	 * write locked, similarly with a read lock.
793 	 */
794 	if ((flags&UF_EXLOCK) && (cmd&FEXLOCK) ||
795 	    (flags&UF_SHLOCK) && (cmd&FSHLOCK))
796 		return;
797 	u.u_pofile[uap->fd] = flocki(fp->f_inode, u.u_pofile[uap->fd], cmd);
798 }
799 
800 /*
801  * Truncate a file given its path name.
802  */
803 truncate()
804 {
805 	struct a {
806 		char	*fname;
807 		u_long	length;
808 	} *uap = (struct a *)u.u_ap;
809 	struct inode *ip;
810 
811 	ip = namei(uchar, LOOKUP, 1);
812 	if (ip == NULL)
813 		return;
814 	if (access(ip, IWRITE))
815 		goto bad;
816 	if ((ip->i_mode&IFMT) == IFDIR) {
817 		u.u_error = EISDIR;
818 		goto bad;
819 	}
820 	itrunc(ip, uap->length);
821 bad:
822 	iput(ip);
823 }
824 
825 /*
826  * Truncate a file given a file descriptor.
827  */
828 ftruncate()
829 {
830 	struct a {
831 		int	fd;
832 		u_long	length;
833 	} *uap = (struct a *)u.u_ap;
834 	struct inode *ip;
835 	struct file *fp;
836 
837 	fp = getf(uap->fd);
838 	if (fp == NULL)
839 		return;
840 	if (fp->f_type == DTYPE_SOCKET) {
841 		u.u_error = EINVAL;
842 		return;
843 	}
844 	if ((fp->f_flag&FWRITE) == 0) {
845 		u.u_error = EINVAL;
846 		return;
847 	}
848 	ip = fp->f_inode;
849 	ilock(ip);
850 	itrunc(ip, uap->length);
851 	iunlock(ip);
852 }
853 
854 /*
855  * Synch an open file.
856  */
857 fsync()
858 {
859 	struct a {
860 		int	fd;
861 	} *uap = (struct a *)u.u_ap;
862 	struct inode *ip;
863 	struct file *fp;
864 
865 	fp = getf(uap->fd);
866 	if (fp == NULL)
867 		return;
868 	if (fp->f_type == DTYPE_SOCKET) {
869 		u.u_error = EINVAL;
870 		return;
871 	}
872 	ip = fp->f_inode;
873 	ilock(ip);
874 	syncip(ip);
875 	iunlock(ip);
876 }
877 
878 /*
879  * Rename system call.
880  * 	rename("foo", "bar");
881  * is essentially
882  *	unlink("bar");
883  *	link("foo", "bar");
884  *	unlink("foo");
885  * but ``atomically''.  Can't do full commit without saving state in the
886  * inode on disk which isn't feasible at this time.  Best we can do is
887  * always guarantee the target exists.
888  *
889  * Basic algorithm is:
890  *
891  * 1) Bump link count on source while we're linking it to the
892  *    target.  This also insure the inode won't be deleted out
893  *    from underneath us while we work.
894  * 2) Link source to destination.  If destination already exists,
895  *    delete it first.
896  * 3) Unlink source reference to inode if still around.
897  * 4) If a directory was moved and the parent of the destination
898  *    is different from the source, patch the ".." entry in the
899  *    directory.
900  *
901  * Source and destination must either both be directories, or both
902  * not be directories.  If target is a directory, it must be empty.
903  */
904 rename()
905 {
906 	struct a {
907 		char	*from;
908 		char	*to;
909 	} *uap;
910 	register struct inode *ip, *xp, *dp;
911 	int oldparent, parentdifferent, doingdirectory;
912 	int error = 0;
913 
914 	uap = (struct a *)u.u_ap;
915 	ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
916 	if (ip == NULL)
917 		return;
918 	dp = u.u_pdir;
919 	oldparent = 0, doingdirectory = 0;
920 	if ((ip->i_mode&IFMT) == IFDIR) {
921 		register struct direct *d;
922 
923 		d = &u.u_dent;
924 		/*
925 		 * Avoid "." and ".." for obvious reasons.
926 		 */
927 		if (d->d_name[0] == '.') {
928 			if (d->d_namlen == 1 ||
929 			    (d->d_namlen == 2 && d->d_name[1] == '.')) {
930 				iput(ip);
931 				u.u_error = EINVAL;
932 				return;
933 			}
934 		}
935 		oldparent = dp->i_number;
936 		doingdirectory++;
937 	}
938 	irele(dp);
939 
940 	/*
941 	 * 1) Bump link count while we're moving stuff
942 	 *    around.  If we crash somewhere before
943 	 *    completing our work, the link count
944 	 *    may be wrong, but correctable.
945 	 */
946 	ip->i_nlink++;
947 	ip->i_flag |= ICHG;
948 	iupdat(ip, &time, &time, 1);
949 	iunlock(ip);
950 
951 	/*
952 	 * When the target exists, both the directory
953 	 * and target inodes are returned locked.
954 	 */
955 	u.u_dirp = (caddr_t)uap->to;
956 	xp = namei(uchar, CREATE | LOCKPARENT, 0);
957 	if (u.u_error) {
958 		error = u.u_error;
959 		goto out;
960 	}
961 	dp = u.u_pdir;
962 	/*
963 	 * 2) If target doesn't exist, link the target
964 	 *    to the source and unlink the source.
965 	 *    Otherwise, rewrite the target directory
966 	 *    entry to reference the source inode and
967 	 *    expunge the original entry's existence.
968 	 */
969 	parentdifferent = oldparent != dp->i_number;
970 	if (xp == NULL) {
971 		if (dp->i_dev != ip->i_dev) {
972 			error = EXDEV;
973 			goto bad;
974 		}
975 		/*
976 		 * Account for ".." in directory.
977 		 * When source and destination have the
978 		 * same parent we don't fool with the
979 		 * link count -- this isn't required
980 		 * because we do a similar check below.
981 		 */
982 		if (doingdirectory && parentdifferent) {
983 			dp->i_nlink++;
984 			dp->i_flag |= ICHG;
985 			iupdat(dp, &time, &time, 1);
986 		}
987 		direnter(ip);
988 		if (u.u_error) {
989 			error = u.u_error;
990 			goto out;
991 		}
992 	} else {
993 		if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) {
994 			error = EXDEV;
995 			goto bad;
996 		}
997 		/*
998 		 * Target must be empty if a directory
999 		 * and have no links to it.
1000 		 * Also, insure source and target are
1001 		 * compatible (both directories, or both
1002 		 * not directories).
1003 		 */
1004 		if ((xp->i_mode&IFMT) == IFDIR) {
1005 			if (!dirempty(xp) || xp->i_nlink > 2) {
1006 				error = ENOTEMPTY;
1007 				goto bad;
1008 			}
1009 			if (!doingdirectory) {
1010 				error = ENOTDIR;
1011 				goto bad;
1012 			}
1013 		} else if (doingdirectory) {
1014 			error = EISDIR;
1015 			goto bad;
1016 		}
1017 		dirrewrite(dp, ip);
1018 		if (u.u_error) {
1019 			error = u.u_error;
1020 			goto bad1;
1021 		}
1022 		/*
1023 		 * Adjust the link count of the target to
1024 		 * reflect the dirrewrite above.  If this is
1025 		 * a directory it is empty and there are
1026 		 * no links to it, so we can squash the inode and
1027 		 * any space associated with it.  We disallowed
1028 		 * renaming over top of a directory with links to
1029 		 * it above, as we've no way to determine if
1030 		 * we've got a link or the directory itself, and
1031 		 * if we get a link, then ".." will be screwed up.
1032 		 */
1033 		xp->i_nlink--;
1034 		if (doingdirectory) {
1035 			if (--xp->i_nlink != 0)
1036 				panic("rename: linked directory");
1037 			itrunc(xp, (u_long)0);
1038 		}
1039 		xp->i_flag |= ICHG;
1040 		iput(xp);
1041 	}
1042 
1043 	/*
1044 	 * 3) Unlink the source.
1045 	 */
1046 	u.u_dirp = uap->from;
1047 	dp = namei(uchar, DELETE, 0);
1048 	/*
1049 	 * Insure directory entry still exists and
1050 	 * has not changed since the start of all
1051 	 * this.  If either has occured, forget about
1052 	 * about deleting the original entry and just
1053 	 * adjust the link count in the inode.
1054 	 */
1055 	if (dp == NULL || u.u_dent.d_ino != ip->i_number) {
1056 		ip->i_nlink--;
1057 		ip->i_flag |= ICHG;
1058 	} else {
1059 		/*
1060 		 * If source is a directory, must adjust
1061 		 * link count of parent directory also.
1062 		 * If target didn't exist and source and
1063 		 * target have the same parent, then we
1064 		 * needn't touch the link count, it all
1065 		 * balances out in the end.  Otherwise, we
1066 		 * must do so to reflect deletion of ".."
1067 		 * done above.
1068 		 */
1069 		if (doingdirectory && (xp != NULL || parentdifferent)) {
1070 			dp->i_nlink--;
1071 			dp->i_flag |= ICHG;
1072 		}
1073 		if (dirremove()) {
1074 			ip->i_nlink--;
1075 			ip->i_flag |= ICHG;
1076 		}
1077 		if (error == 0)		/* conservative */
1078 			error = u.u_error;
1079 	}
1080 	irele(ip);
1081 	if (dp)
1082 		iput(dp);
1083 
1084 	/*
1085 	 * 4) Renaming a directory with the parent
1086 	 *    different requires ".." to be rewritten.
1087 	 *    The window is still there for ".." to
1088 	 *    be inconsistent, but this is unavoidable,
1089 	 *    and a lot shorter than when it was done
1090 	 *    in a user process.
1091 	 */
1092 	if (doingdirectory && parentdifferent && error == 0) {
1093 		struct dirtemplate dirbuf;
1094 
1095 		u.u_dirp = uap->to;
1096 		ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
1097 		if (ip == NULL) {
1098 			printf("rename: .. went away\n");
1099 			return;
1100 		}
1101 		dp = u.u_pdir;
1102 		if ((ip->i_mode&IFMT) != IFDIR) {
1103 			printf("rename: .. not a directory\n");
1104 			goto stuck;
1105 		}
1106 		error = rdwri(UIO_READ, ip, (caddr_t)&dirbuf,
1107 			sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1108 		if (error == 0) {
1109 			dirbuf.dotdot_ino = dp->i_number;
1110 			(void) rdwri(UIO_WRITE, ip, (caddr_t)&dirbuf,
1111 			  sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1112 		}
1113 stuck:
1114 		irele(dp);
1115 		iput(ip);
1116 	}
1117 	goto done;
1118 
1119 bad:
1120 	iput(u.u_pdir);
1121 bad1:
1122 	if (xp)
1123 		irele(xp);
1124 out:
1125 	ip->i_nlink--;
1126 	ip->i_flag |= ICHG;
1127 	irele(ip);
1128 done:
1129 	if (error)
1130 		u.u_error = error;
1131 }
1132 
1133 /*
1134  * Make a new file.
1135  */
1136 struct inode *
1137 maknode(mode)
1138 	int mode;
1139 {
1140 	register struct inode *ip;
1141 	ino_t ipref;
1142 
1143 	if ((mode & IFMT) == IFDIR)
1144 		ipref = dirpref(u.u_pdir->i_fs);
1145 	else
1146 		ipref = u.u_pdir->i_number;
1147 	ip = ialloc(u.u_pdir, ipref, mode);
1148 	if (ip == NULL) {
1149 		iput(u.u_pdir);
1150 		return (NULL);
1151 	}
1152 #ifdef QUOTA
1153 	if (ip->i_dquot != NODQUOT)
1154 		panic("maknode: dquot");
1155 #endif
1156 	ip->i_flag |= IACC|IUPD|ICHG;
1157 	if ((mode & IFMT) == 0)
1158 		mode |= IFREG;
1159 	ip->i_mode = mode & ~u.u_cmask;
1160 	ip->i_nlink = 1;
1161 	ip->i_uid = u.u_uid;
1162 	ip->i_gid = u.u_pdir->i_gid;
1163 #ifdef QUOTA
1164 	ip->i_dquot = inoquota(ip);
1165 #endif
1166 
1167 	/*
1168 	 * Make sure inode goes to disk before directory entry.
1169 	 */
1170 	iupdat(ip, &time, &time, 1);
1171 	direnter(ip);
1172 	if (u.u_error) {
1173 		/*
1174 		 * write error occurred trying to update directory
1175 		 * so must deallocate the inode
1176 		 */
1177 		ip->i_nlink = 0;
1178 		ip->i_flag |= ICHG;
1179 		iput(ip);
1180 		return (NULL);
1181 	}
1182 	return (ip);
1183 }
1184