xref: /original-bsd/sys/kern/vfs_syscalls.c (revision b64b0d54)
1 /*	vfs_syscalls.c	4.53	83/03/22	*/
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 	u.u_error = 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 		if (u.u_gid == ip->i_gid)
601 			goto ok;
602 		for (gp = u.u_groups; gp < &u.u_groups[NGROUPS]; gp++)
603 			if (*gp == ip->i_gid)
604 				goto ok;
605 		mode &= ~ISGID;
606 ok:
607 		;
608 #ifdef MUSH
609 		if (u.u_quota->q_syflags & QF_UMASK && u.u_uid != 0 &&
610 		    (ip->i_mode & IFMT) != IFCHR)
611 			mode &= ~u.u_cmask;
612 #endif
613 	}
614 	ip->i_mode |= mode&07777;
615 	ip->i_flag |= ICHG;
616 	if (ip->i_flag&ITEXT && (ip->i_mode&ISVTX)==0)
617 		xrele(ip);
618 }
619 
620 /*
621  * Set ownership given a path name.
622  */
623 chown()
624 {
625 	struct inode *ip;
626 	struct a {
627 		char	*fname;
628 		int	uid;
629 		int	gid;
630 	} *uap;
631 
632 	uap = (struct a *)u.u_ap;
633 	if (!suser() || (ip = owner(0)) == NULL)
634 		return;
635 	chown1(ip, uap->uid, uap->gid);
636 	iput(ip);
637 }
638 
639 /*
640  * Set ownership given a file descriptor.
641  */
642 fchown()
643 {
644 	struct a {
645 		int	fd;
646 		int	uid;
647 		int	gid;
648 	} *uap;
649 	register struct inode *ip;
650 	register struct file *fp;
651 
652 	uap = (struct a *)u.u_ap;
653 	fp = getf(uap->fd);
654 	if (fp == NULL)
655 		return;
656 	if (fp->f_type == DTYPE_SOCKET) {
657 		u.u_error = EINVAL;
658 		return;
659 	}
660 	ip = fp->f_inode;
661 	if (!suser())
662 		return;
663 	ilock(ip);
664 	chown1(ip, uap->uid, uap->gid);
665 	iunlock(ip);
666 }
667 
668 /*
669  * Perform chown operation on inode ip;
670  * inode must be locked prior to call.
671  */
672 chown1(ip, uid, gid)
673 	register struct inode *ip;
674 	int uid, gid;
675 {
676 #ifdef QUOTA
677 	register long change;
678 
679 	/*
680 	 * This doesn't allow for holes in files (which hopefully don't
681 	 * happen often in files that we chown), and is not accurate anyway
682 	 * (eg: it totally ignores 3 level indir blk files - but hopefully
683 	 * noone who can make a file that big will have a quota)
684 	 */
685 	if (ip->i_uid == uid)
686 		change = 0;
687 	else {
688 		register struct fs *fs = ip->i_fs;
689 
690 		if (ip->i_size > (change = NDADDR * fs->fs_bsize)) {
691 			register off_t size;
692 
693 			size = blkroundup(fs, ip->i_size) - change;
694 			change += size;
695 			change += fs->fs_bsize;
696 			/* this assumes NIADDR <= 2 */
697 			if (size > NINDIR(fs) * fs->fs_bsize)
698 				change += fs->fs_bsize;
699 		} else
700 			change = fragroundup(fs, ip->i_size);
701 		change /= DEV_BSIZE;
702 	}
703 	(void)chkdq(ip, -change, 1);
704 	(void)chkiq(ip->i_dev, ip, ip->i_uid, 1);
705 	dqrele(ip->i_dquot);
706 #endif
707 	/*
708 	 * keep uid/gid's in sane range -- no err,
709 	 * so chown(file, uid, -1) will do something useful
710 	 */
711 	if (uid >= 0 && uid <= 32767)	/* should have a constant */
712 		ip->i_uid = uid;
713 	if (gid >= 0 && gid <= 32767)	/* same here */
714 		ip->i_gid = gid;
715 	ip->i_flag |= ICHG;
716 	if (u.u_ruid != 0)
717 		ip->i_mode &= ~(ISUID|ISGID);
718 #ifdef QUOTA
719 	ip->i_dquot = inoquota(ip);
720 	(void)chkdq(ip, change, 1);
721 	(void)chkiq(ip->i_dev, (struct inode *)NULL, uid, 1);
722 #endif
723 }
724 
725 /*
726  * Set IUPD and IACC times on file.
727  * Can't set ICHG.
728  */
729 outime()
730 {
731 	register struct a {
732 		char	*fname;
733 		time_t	*tptr;
734 	} *uap;
735 	register struct inode *ip;
736 	time_t tv[2];
737 	struct timeval tv0, tv1;
738 
739 	uap = (struct a *)u.u_ap;
740 	if ((ip = owner(1)) == NULL)
741 		return;
742 	u.u_error = copyin((caddr_t)uap->tptr, (caddr_t)tv, sizeof(tv));
743 	if (u.u_error == 0) {
744 		ip->i_flag |= IACC|IUPD|ICHG;
745 		tv0.tv_sec = tv[0]; tv0.tv_usec = 0;
746 		tv1.tv_sec = tv[1]; tv1.tv_usec = 0;
747 		iupdat(ip, &tv0, &tv1, 0);
748 	}
749 	iput(ip);
750 }
751 
752 /*
753  * Flush any pending I/O.
754  */
755 sync()
756 {
757 
758 	update();
759 }
760 
761 /*
762  * Apply an advisory lock on a file descriptor.
763  */
764 flock()
765 {
766 	struct a {
767 		int	fd;
768 		int	how;
769 	} *uap;
770 	register struct file *fp;
771 	register int cmd, flags;
772 
773 	uap = (struct a *)u.u_ap;
774 	fp = getf(uap->fd);
775 	if (fp == NULL)
776 		return;
777 	if (fp->f_type == DTYPE_SOCKET) {		/* XXX */
778 		u.u_error = EINVAL;
779 		return;
780 	}
781 	cmd = uap->how;
782 	flags = u.u_pofile[uap->fd] & (UF_SHLOCK|UF_EXLOCK);
783 	if (cmd&FUNLOCK) {
784 		if (flags == 0) {
785 			u.u_error = EINVAL;
786 			return;
787 		}
788 		funlocki(fp->f_inode, flags);
789 		u.u_pofile[uap->fd] &= ~(UF_SHLOCK|UF_EXLOCK);
790 		return;
791 	}
792 	/*
793 	 * No reason to write lock a file we've already
794 	 * write locked, similarly with a read lock.
795 	 */
796 	if ((flags&UF_EXLOCK) && (cmd&FEXLOCK) ||
797 	    (flags&UF_SHLOCK) && (cmd&FSHLOCK))
798 		return;
799 	u.u_pofile[uap->fd] = flocki(fp->f_inode, u.u_pofile[uap->fd], cmd);
800 }
801 
802 /*
803  * Truncate a file given its path name.
804  */
805 truncate()
806 {
807 	struct a {
808 		char	*fname;
809 		u_long	length;
810 	} *uap = (struct a *)u.u_ap;
811 	struct inode *ip;
812 
813 	ip = namei(uchar, LOOKUP, 1);
814 	if (ip == NULL)
815 		return;
816 	if (access(ip, IWRITE))
817 		goto bad;
818 	if ((ip->i_mode&IFMT) == IFDIR) {
819 		u.u_error = EISDIR;
820 		goto bad;
821 	}
822 	itrunc(ip, uap->length);
823 bad:
824 	iput(ip);
825 }
826 
827 /*
828  * Truncate a file given a file descriptor.
829  */
830 ftruncate()
831 {
832 	struct a {
833 		int	fd;
834 		u_long	length;
835 	} *uap = (struct a *)u.u_ap;
836 	struct inode *ip;
837 	struct file *fp;
838 
839 	fp = getf(uap->fd);
840 	if (fp == NULL)
841 		return;
842 	if (fp->f_type == DTYPE_SOCKET) {
843 		u.u_error = EINVAL;
844 		return;
845 	}
846 	if ((fp->f_flag&FWRITE) == 0) {
847 		u.u_error = EINVAL;
848 		return;
849 	}
850 	ip = fp->f_inode;
851 	ilock(ip);
852 	itrunc(ip, uap->length);
853 	iunlock(ip);
854 }
855 
856 /*
857  * Synch an open file.
858  */
859 fsync()
860 {
861 	struct a {
862 		int	fd;
863 	} *uap = (struct a *)u.u_ap;
864 	struct inode *ip;
865 	struct file *fp;
866 
867 	fp = getf(uap->fd);
868 	if (fp == NULL)
869 		return;
870 	if (fp->f_type == DTYPE_SOCKET) {
871 		u.u_error = EINVAL;
872 		return;
873 	}
874 	ip = fp->f_inode;
875 	ilock(ip);
876 	syncip(ip);
877 	iunlock(ip);
878 }
879 
880 /*
881  * Rename system call.
882  * 	rename("foo", "bar");
883  * is essentially
884  *	unlink("bar");
885  *	link("foo", "bar");
886  *	unlink("foo");
887  * but ``atomically''.  Can't do full commit without saving state in the
888  * inode on disk which isn't feasible at this time.  Best we can do is
889  * always guarantee the target exists.
890  *
891  * Basic algorithm is:
892  *
893  * 1) Bump link count on source while we're linking it to the
894  *    target.  This also insure the inode won't be deleted out
895  *    from underneath us while we work.
896  * 2) Link source to destination.  If destination already exists,
897  *    delete it first.
898  * 3) Unlink source reference to inode if still around.
899  * 4) If a directory was moved and the parent of the destination
900  *    is different from the source, patch the ".." entry in the
901  *    directory.
902  *
903  * Source and destination must either both be directories, or both
904  * not be directories.  If target is a directory, it must be empty.
905  */
906 rename()
907 {
908 	struct a {
909 		char	*from;
910 		char	*to;
911 	} *uap;
912 	register struct inode *ip, *xp, *dp;
913 	int oldparent, parentdifferent, doingdirectory;
914 	int error = 0;
915 
916 	uap = (struct a *)u.u_ap;
917 	ip = namei(uchar, DELETE | LOCKPARENT, 0);
918 	if (ip == NULL)
919 		return;
920 	dp = u.u_pdir;
921 	oldparent = 0, doingdirectory = 0;
922 	if ((ip->i_mode&IFMT) == IFDIR) {
923 		register struct direct *d;
924 
925 		d = &u.u_dent;
926 		/*
927 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
928 		 */
929 		if ((d->d_namlen == 1 && d->d_name[0] == '.') ||
930 		    (d->d_namlen == 2 && bcmp(d->d_name, "..", 2) == 0) ||
931 		    (dp == ip)) {
932 			iput(dp);
933 			if (dp == ip)
934 				irele(ip);
935 			else
936 				iput(ip);
937 			u.u_error = EINVAL;
938 			return;
939 		}
940 		oldparent = dp->i_number;
941 		doingdirectory++;
942 	}
943 	iput(dp);
944 
945 	/*
946 	 * 1) Bump link count while we're moving stuff
947 	 *    around.  If we crash somewhere before
948 	 *    completing our work, the link count
949 	 *    may be wrong, but correctable.
950 	 */
951 	ip->i_nlink++;
952 	ip->i_flag |= ICHG;
953 	iupdat(ip, &time, &time, 1);
954 	iunlock(ip);
955 
956 	/*
957 	 * When the target exists, both the directory
958 	 * and target inodes are returned locked.
959 	 */
960 	u.u_dirp = (caddr_t)uap->to;
961 	xp = namei(uchar, CREATE | LOCKPARENT, 0);
962 	if (u.u_error) {
963 		error = u.u_error;
964 		goto out;
965 	}
966 	dp = u.u_pdir;
967 	/*
968 	 * If ".." must be changed (ie the directory gets a new
969 	 * parent) then the user must have write permission.
970 	 */
971 	parentdifferent = oldparent != dp->i_number;
972 	if (doingdirectory && parentdifferent && access(ip, IWRITE))
973 		goto bad;
974 	/*
975 	 * 2) If target doesn't exist, link the target
976 	 *    to the source and unlink the source.
977 	 *    Otherwise, rewrite the target directory
978 	 *    entry to reference the source inode and
979 	 *    expunge the original entry's existence.
980 	 */
981 	if (xp == NULL) {
982 		if (dp->i_dev != ip->i_dev) {
983 			error = EXDEV;
984 			goto bad;
985 		}
986 		/*
987 		 * Disallow rename(foo, foo/bar).
988 		 */
989 		if (dp->i_number == ip->i_number) {
990 			error = EEXIST;
991 			goto bad;
992 		}
993 		/*
994 		 * Account for ".." in directory.
995 		 * When source and destination have the
996 		 * same parent we don't fool with the
997 		 * link count -- this isn't required
998 		 * because we do a similar check below.
999 		 */
1000 		if (doingdirectory && parentdifferent) {
1001 			dp->i_nlink++;
1002 			dp->i_flag |= ICHG;
1003 			iupdat(dp, &time, &time, 1);
1004 		}
1005 		error = direnter(ip);
1006 		if (error)
1007 			goto out;
1008 	} else {
1009 		if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) {
1010 			error = EXDEV;
1011 			goto bad;
1012 		}
1013 		/*
1014 		 * Short circuit rename(foo, foo).
1015 		 */
1016 		if (xp->i_number == ip->i_number)
1017 			goto bad;
1018 		/*
1019 		 * Target must be empty if a directory
1020 		 * and have no links to it.
1021 		 * Also, insure source and target are
1022 		 * compatible (both directories, or both
1023 		 * not directories).
1024 		 */
1025 		if ((xp->i_mode&IFMT) == IFDIR) {
1026 			if (!dirempty(xp) || xp->i_nlink > 2) {
1027 				error = ENOTEMPTY;
1028 				goto bad;
1029 			}
1030 			if (!doingdirectory) {
1031 				error = ENOTDIR;
1032 				goto bad;
1033 			}
1034 		} else if (doingdirectory) {
1035 			error = EISDIR;
1036 			goto bad;
1037 		}
1038 		dirrewrite(dp, ip);
1039 		if (u.u_error) {
1040 			error = u.u_error;
1041 			goto bad1;
1042 		}
1043 		/*
1044 		 * Adjust the link count of the target to
1045 		 * reflect the dirrewrite above.  If this is
1046 		 * a directory it is empty and there are
1047 		 * no links to it, so we can squash the inode and
1048 		 * any space associated with it.  We disallowed
1049 		 * renaming over top of a directory with links to
1050 		 * it above, as we've no way to determine if
1051 		 * we've got a link or the directory itself, and
1052 		 * if we get a link, then ".." will be screwed up.
1053 		 */
1054 		xp->i_nlink--;
1055 		if (doingdirectory) {
1056 			if (--xp->i_nlink != 0)
1057 				panic("rename: linked directory");
1058 			itrunc(xp, (u_long)0);
1059 		}
1060 		xp->i_flag |= ICHG;
1061 		iput(xp);
1062 		xp = NULL;
1063 	}
1064 
1065 	/*
1066 	 * 3) Unlink the source.
1067 	 */
1068 	u.u_dirp = uap->from;
1069 	dp = namei(uchar, DELETE, 0);
1070 	/*
1071 	 * Insure directory entry still exists and
1072 	 * has not changed since the start of all
1073 	 * this.  If either has occured, forget about
1074 	 * about deleting the original entry and just
1075 	 * adjust the link count in the inode.
1076 	 */
1077 	if (dp == NULL || u.u_dent.d_ino != ip->i_number) {
1078 		ip->i_nlink--;
1079 		ip->i_flag |= ICHG;
1080 	} else {
1081 		/*
1082 		 * If source is a directory, must adjust
1083 		 * link count of parent directory also.
1084 		 * If target didn't exist and source and
1085 		 * target have the same parent, then we
1086 		 * needn't touch the link count, it all
1087 		 * balances out in the end.  Otherwise, we
1088 		 * must do so to reflect deletion of ".."
1089 		 * done above.
1090 		 */
1091 		if (doingdirectory && (xp != NULL || parentdifferent)) {
1092 			dp->i_nlink--;
1093 			dp->i_flag |= ICHG;
1094 		}
1095 		if (dirremove()) {
1096 			ip->i_nlink--;
1097 			ip->i_flag |= ICHG;
1098 		}
1099 		if (error == 0)		/* conservative */
1100 			error = u.u_error;
1101 	}
1102 	irele(ip);
1103 	if (dp)
1104 		iput(dp);
1105 
1106 	/*
1107 	 * 4) Renaming a directory with the parent
1108 	 *    different requires ".." to be rewritten.
1109 	 *    The window is still there for ".." to
1110 	 *    be inconsistent, but this is unavoidable,
1111 	 *    and a lot shorter than when it was done
1112 	 *    in a user process.
1113 	 */
1114 	if (doingdirectory && parentdifferent && error == 0) {
1115 		struct dirtemplate dirbuf;
1116 
1117 		u.u_dirp = uap->to;
1118 		ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
1119 		if (ip == NULL) {
1120 			printf("rename: .. went away\n");
1121 			return;
1122 		}
1123 		dp = u.u_pdir;
1124 		if ((ip->i_mode&IFMT) != IFDIR) {
1125 			printf("rename: .. not a directory\n");
1126 			goto stuck;
1127 		}
1128 		error = rdwri(UIO_READ, ip, (caddr_t)&dirbuf,
1129 			sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1130 		if (error == 0) {
1131 			dirbuf.dotdot_ino = dp->i_number;
1132 			(void) rdwri(UIO_WRITE, ip, (caddr_t)&dirbuf,
1133 			  sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
1134 		}
1135 stuck:
1136 		irele(dp);
1137 		iput(ip);
1138 	}
1139 	goto done;
1140 
1141 bad:
1142 	iput(dp);
1143 bad1:
1144 	if (xp)
1145 		iput(xp);
1146 out:
1147 	ip->i_nlink--;
1148 	ip->i_flag |= ICHG;
1149 	irele(ip);
1150 done:
1151 	if (error)
1152 		u.u_error = error;
1153 }
1154 
1155 /*
1156  * Make a new file.
1157  */
1158 struct inode *
1159 maknode(mode)
1160 	int mode;
1161 {
1162 	register struct inode *ip;
1163 	ino_t ipref;
1164 
1165 	if ((mode & IFMT) == IFDIR)
1166 		ipref = dirpref(u.u_pdir->i_fs);
1167 	else
1168 		ipref = u.u_pdir->i_number;
1169 	ip = ialloc(u.u_pdir, ipref, mode);
1170 	if (ip == NULL) {
1171 		iput(u.u_pdir);
1172 		return (NULL);
1173 	}
1174 #ifdef QUOTA
1175 	if (ip->i_dquot != NODQUOT)
1176 		panic("maknode: dquot");
1177 #endif
1178 	ip->i_flag |= IACC|IUPD|ICHG;
1179 	if ((mode & IFMT) == 0)
1180 		mode |= IFREG;
1181 	ip->i_mode = mode & ~u.u_cmask;
1182 	ip->i_nlink = 1;
1183 	ip->i_uid = u.u_uid;
1184 	ip->i_gid = u.u_pdir->i_gid;
1185 #ifdef QUOTA
1186 	ip->i_dquot = inoquota(ip);
1187 #endif
1188 
1189 	/*
1190 	 * Make sure inode goes to disk before directory entry.
1191 	 */
1192 	iupdat(ip, &time, &time, 1);
1193 	u.u_error = direnter(ip);
1194 	if (u.u_error) {
1195 		/*
1196 		 * Write error occurred trying to update directory
1197 		 * so must deallocate the inode.
1198 		 */
1199 		ip->i_nlink = 0;
1200 		ip->i_flag |= ICHG;
1201 		iput(ip);
1202 		return (NULL);
1203 	}
1204 	return (ip);
1205 }
1206