Subject: fixes for libkern, autoconfig, umount, tar; add rkboot.s (#446)
Index:	 ulrem.s,rk.c,conf.c,rkboot.s,rkauto.c,umount.c,tar.c 2.11BSD

Description:
	1. Standalone mkfs fails to handle a default (<cr>) answer to
	   the 'file sys size' properly, and create a file system with
           a small number of inodes. This is caused by a bug in the unsigned 
	   long modulo (%) routine ulrem in the C runtime system when compiled 
	   for libkern.

        2. Booting a 211BSD system of a RK05 drive fails with 'panic: iinit'

        3. No BOOTDEV option for 'rk' drives available
           No BOOTDEV option for 'xp' drives documented

        4. First access to a RK05 drive causes a hard error when no
           drive 0 attached to the system.

        5. the umount command returns inverted exit() codes in some
           cases, thus 0 when a command failed and 1 when it succeeded.

        6. tar core dumps when a large file system is written to a tarball.

Repeat-By:
	1. Create distribution tape with 'make tmscptape', boot with
           'tm(0,2)' into standalone mkfs. Use defaults for 'file sys size'
           and 'bytes per inode', the resulting isize is much smaller
	   than expected.

        2-3. Observation.

        4. start 211BSD in simh, attach drive rk1 but not rk0 to a file,
           mount the file system on /dev/rk1h. You'll get one hard error:
             rk0b: hard error sn2 er=100000<DRE> ds=4707<SOK,DRY,RWSRDY>
           after that access to rk1h works fine.

        5. Observation, try 'umount' and check exit status

        6. For a full 211BSD system a 'tar c -C / .' ends in a
             Segmentation fault (core dumped)
           a tar.core is written somewhere into the file system.

Fix:	1. mkfs / ulrem:
	   The ulrem routine in the C runtime system is, like several other 
	   long int routines, available in two implementations, with and 
	   without floating point instructions. The version with FPP 
	   instructions is faster, used when libc is created, and thus by 
	   all normal programs. The version without FPP instructions is used 
	   when libkern is created, and intended for kernel and standalone 
	   code usage.

           ulrem as of patch level 445 miscalculates the remainder in some
	   cases when the quotient is zero:
               a=   1; b=    10; ==>  a%b=    0;   Should be 1 !
               a=  99; b= 32766; ==>  a%b=   99;   a/b=0, but ok

           The ltoa() function in /sys/pdpstand/sys.c uses u_long modulo and
	   involved in the processing the default file system size in 
	   standalone mkfs, the ulrem bug leads to a misdetermination of the 
	   file system size when just <cr> is answered.

           To fix this the non-FPP part of ulrem was rewritten. When both 
	   operands are non-negative the new code simply jumps to the lrem 
	   routine (signed long modulo). When one or both operands are 
	   negative, the remainder is calculated quite simplistically by
		rem = lhs-(lhs/rhs)*rhs
           using ldiv and lmul. This is slow, but save. Since this case is in 
	   practice rare, preformance isn't an issue here.

           At this point one wonders how 211BSD worked at all with such 
	   elementary arithmetic problems. The answer is:
           - the FPP version of ulrem works correctly
           - ulrem isn't used by the kernel (unix and netnix)
           - up to patch 434 the standalone programs where mistakenly linked 
	     against libc rather libkern, thus used the FPP version, which 
	     works correctly.
           - /boot does link against ulrem but apparently seems not to be 
	     affected much....
	
	2. Booting 211BSD of a RK05 drive:
	   During kernel startup the root file system and the swap space must
	   be accessed, well before autoconfig is executed. The kernel calls 
	   in /sys/sys/init_main.c
             (void)(*bdevsw[major(bootdev)].d_root)(bootcsr);
           which executes the 'root attach' function of the driver. The rk.c 
	   driver doesn't provide this function, the boot device attach fails,
	   and the kernel panics with 'iinit'.
           
           The fix is simple:

           - rk.c now implements rkroot(), the root attach function
           - the block device switch bdevsw[] entry for rk is updated.

        3a. No BOOTDEV option for 'rk' drives available
            During a 'reboot' the kernel finally calls doboot(). The directory
	    /sys/conf/boot hold implementations for a variety of boot devices, 
	    but not for the admittedly exotic case of rk05 boot device.

            Fix: rkboot.s is added with this patch

        3b. No BOOTDEV option for 'xp' drives documented
            The 'BOOTDEV rm' option can be used in case of a xp boot device.
	    RM and RP drives have different ..uboot.s implementations because
	    in this case knowledge of the disk geometry is needed, but the
	    primary boot block loader ..boot.s only loads lbn 0 and is thus
	    common for RM and RP drives.

            Updated the comment on the 'BOOTDEV rm' line and added
            a 'BOOTDEV rk' line to the GENERIC and non-networking (NONET)
	    configurations shipped with the distribution.

	4. First access to RK drive:
	   The rkauto() function used by autoconfig to probe for RK11
	   controllers executes a 'drive reset' on drive 0:

             stuff(RKCS_IDE | RKCS_DRESET | RKCS_GO, (&(addr->rkcs)));

           If no drive 0 is attached, the RK11 controller will set the hard 
	   error bit in RKER and refuse all further commands until a control
	   reset is executed.

           In a configuration with no drive 0 attached, the autoconfig will 
	   thus detect and properly attach the RK11 controller, but leave it 
	   in an error state. The next request, typically a read caused by a
	   mount, will fail. The rk.c driver does a 'control reset' and retry
	   in this case, so the mount will succeed, fortunately.

           autoconfig should leave devices in operational state, so extra code
	   was added to rkauto() to execute 'control reset' in the case 
	   described above.

	5. umount exit status:
	   The main() of umount.c accumulates the return codes for calls of 
	   umountfs(), intention is to return an exit status of 1 of any of
	   the unmounts fails:

               for (errs = 0; *argv != NULL; ++argv)
	           if (umountfs(*argv) == 0)
	               errs = 1;
	  
	   umountfs returns 0 in case of success, 1 in case of error, so the 
	   if() should use != and not ==.	   

	6. tar memory faults when large tarballs are created:
	   In order to be able to reconstruct files with several hard links 
	   from a tarball tar has to keep track of all ordinary files with a 
	   link count > 1. A struct linkbuf is malloc'ed for each file with
	   additional hard links. Eventually the data segment expands into the
	   stack segment and tar will core dump. As such, this is a principal 
	   limitation of tar  on a 16 bit system.

           It's worth mentioning that the tar man pages state:

             DIAGNOSTICS
               Complaints if enough memory is not available to hold
               the link tables.

           The getmem() function indeed checks whether the malloc() succeeded. 
	   However, the failure mechanism is more subtle. First the data area
	   expands beyond 140000, than the stack area tries to grow below 
	   160000, which is not possible.  In all tests done tar crashed
	   due to a failure to expand the stack.

	   Last but not least: because tar changes the current working 
	   directory to the directory it's currently processing the tar.core 
	   ends up in this directory, and not the cwd where tar was started in.

           Now the good news: struct linkbuf a much larger than needed:
	       struct linkbuf {
	           ino_t  inum;
	           dev_t  devnum;
	           int    count;
	           char	  pathname[NAMSIZ];
	           struct linkbuf *nextp;
                };
           It allocates NAMSIZ bytes for the path name, NAMSIZ is
	   #defined'ed to 100 in tar.c. This is way larger than the typical 
	   file name size, thus wasting precious space.

	   A simple modification is to change this to
	           char	  *pathname;
	   and allocate with a separate malloc a minimally sized buffer for
	   the pathname.

	   This reduces the memory consumption to a point where a tarball with
	   a full 211BSD system disk can be created. The test system had 201 
	   ordinary files with multiple links, the tar needs 11.2 kB for
	   linkbuf's, the data area grows to about 120000. The original tar 
	   needed more than twice as much space for linkbuf structures.

	The patches above where submitted by
	    Walter F.J. Mueller <w.f.j.mueller@gsi.de>

	The following files are affected by this update:

	modified:  /usr/src/lib/libc/pdp/crt/ulrem.s
        modified:  /usr/src/sys/pdpuba/rk.c
	modified:  /usr/src/sys/pdp/conf.c
        added:     /usr/src/sys/conf/boot/rkboot.s
	modified:  /usr/src/sys/conf/boot/rmboot.s
	modified:  /usr/src/sys/conf/GENERIC
	modified:  /usr/src/sys/conf/NONET
	modified:  /usr/src/sys/autoconfig/rkauto.c
	modified:  /usr/src/sbin/umount/umount.c
	modified:  /usr/src/bin/tar.c

	To apply this update save the patch (for example /tmp/446.patch):

	   # The rkboot.s should not exist but rm -f to be certain it is absent
	   rm -f /usr/src/sys/conf/boot/rkboot.s
	   patch -p0 < /tmp/446.patch

           cd /usr/src/lib/libkern
           make clean
           make
           make install
           make clean

           cd /sys/pdpstand
           make clean
           make
           cp boot disklabel toyset /
	   
           cd /usr/src/sys/autoconfig
           make clean
           make
           make install
           make clean

           cd /usr/src/sbin/umount
           make 
           make install
           make clean

           cd /usr/src/bin
           make tar
           install -s -m 751 -g staff tar /bin

       You may wish to recompile the kernel (both the current and the GENERIC)
       but that can be deferred if desired.

---------------------------446.patch cut here----------------------
*** /usr/src/sys/pdpuba/rk.c.old	Sat Jan 18 16:44:45 1997
--- /usr/src/sys/pdpuba/rk.c	Sat Dec 27 17:22:21 2008
***************
*** 43,48 ****
--- 43,55 ----
  					*/
  #endif
  
+ /* added rkroot() to support RK11 based systems (2008-12-14,wfjm) */
+ 
+ rkroot()
+ {
+ 	rkattach((struct rkdevice *)0177400, 0);
+ }
+ 
  rkattach(addr, unit)
  struct rkdevice *addr;
  {
*** /usr/src/sys/conf/boot/rmboot.s.old	Fri Apr 19 21:03:43 1991
--- /usr/src/sys/conf/boot/rmboot.s	Sat Dec 27 17:27:04 2008
***************
*** 50,56 ****
  /  Otherwise, use a BOOT opcode, if available;
  /  if necessary, read in block 0 to location 0 "by hand".
  
! / rm02/3/5 bootstrap - salkind@nyu
  
  WC	= -256.
  READ	= 70
--- 50,56 ----
  /  Otherwise, use a BOOT opcode, if available;
  /  if necessary, read in block 0 to location 0 "by hand".
  
! / rm02/3/5 rp06 bootstrap - salkind@nyu
  
  WC	= -256.
  READ	= 70
*** /usr/src/sys/conf/boot/rkboot.s.old	Sat Dec 27 17:28:29 2008
--- /usr/src/sys/conf/boot/rkboot.s	Fri Dec 26 12:51:47 2008
***************
*** 0 ****
--- 1,77 ----
+ /*
+  *	SCCS id	@(#)rkboot.s	1.0 (2.11BSD)	12/26/2008
+  */
+ #include "localopts.h"
+ 
+ /  The boot options and device are placed in the last SZFLAGS bytes
+ /  at the end of core for the bootstrap.
+ ENDCORE=	160000		/ end of core, mem. management off
+ SZFLAGS=	6		/ size of boot flags
+ BOOTOPTS=	2		/ location of options, bytes below ENDCORE
+ BOOTDEV=	4		/ boot unit
+ CHECKWORD=	6
+ 
+ .globl	_doboot, hardboot, _bootcsr
+ .text
+ _doboot:
+ 	mov	4(sp),r4	/ boot options
+ 	mov	2(sp),r3	/ boot device
+ 
+ #ifndef	KERN_NONSEP
+ /  If running separate I/D, need to turn off memory management.
+ /  Call the routine unmap in low text, after setting up a jump
+ /  in low data where the PC will be pointing.
+ .globl	unmap
+ 	mov	$137,*$unmap+2		/ jmp *$hardboot
+ 	mov	$hardboot,*$unmap+4
+ 	jmp	unmap
+ 	/ "return" from unmap will be to hardboot in data
+ .data
+ #else
+ /  Reset to turn off memory management
+ 	reset
+ #endif
+ 
+ /  On power fail, hardboot is the entry point (map is already off)
+ /  and the args are in r4 (RB_POWRFAIL), r3 (rootdev)
+ 
+ hardboot:
+ 	mov	r4, ENDCORE-BOOTOPTS
+ 	ash	$-3,r3		/ shift out the partition number
+ 	bic	$!7,r3		/ save only the drive number
+ 	mov	r3, ENDCORE-BOOTDEV
+ 	com	r4		/ if CHECKWORD == ~bootopts, flags are believed
+ 	mov	r4, ENDCORE-CHECKWORD
+ 1:
+ 	reset
+ 
+ /  The remainder of the code is dependent on the boot device.
+ /  If you have a bootstrap ROM, just jump to the correct entry.
+ /  Otherwise, use a BOOT opcode, if available;
+ /  if necessary, read in block 0 to location 0 "by hand".
+ 
+ / Bootstrap for rk05 drive - wfjm
+ 
+ WC = -256.
+ 
+ rkcs =  4	/ offset from base csr: control & status
+ rkda = 12	/ desired disk address
+ 
+ / RK05 constants.
+ iocom = 005	/ read + go
+ 
+ / initialize rk
+ 
+ 	mov	_bootcsr,r1	/ bootcsr points to rkcs
+ 	add	$rkda-rkcs,r1	/ r1 -> rkda
+ 	mov	ENDCORE-BOOTDEV,r2	/ drive number
+ 	ash	$13,r2		/ drsel is in bits 15:13
+ 	mov	r2,(r1)		/ setup rkda (disk address)
+ 	clr	-(r1)		/ clear rkba (memory address)
+ 	mov	$WC,-(r1)	/ setup rkwc (transfer length)
+ 	mov	$iocom,-(r1)	/ issue read+go
+ 
+ 1:	tstb	(r1)
+ 	bge	1b		/ wait for iocom to complete
+ 	mov	ENDCORE-BOOTDEV,r0
+ 	clr	pc
*** /usr/src/sys/conf/GENERIC.old	Sat Oct 13 18:20:02 2001
--- /usr/src/sys/conf/GENERIC	Sat Dec 27 17:17:24 2008
***************
*** 52,59 ****
  #BOOTDEV	hk6			# rk06 boot device
  #BOOTDEV	hk7			# rk07 boot device
  #BOOTDEV	ra			# MSCP boot device
  #BOOTDEV	rl			# rl01/02 boot device
! #BOOTDEV	rm			# rm02/03/05 boot device
  #BOOTDEV	br			# Eaton BR1537/BR1711 boot device
  #BOOTDEV	sc11			# Emulex SC11/B boot device
  #BOOTDEV	sc21			# Emulex SC21 boot device
--- 52,60 ----
  #BOOTDEV	hk6			# rk06 boot device
  #BOOTDEV	hk7			# rk07 boot device
  #BOOTDEV	ra			# MSCP boot device
+ #BOOTDEV	rk			# rk05 boot device
  #BOOTDEV	rl			# rl01/02 boot device
! #BOOTDEV	rm			# rm02/03/05 rp06 boot device
  #BOOTDEV	br			# Eaton BR1537/BR1711 boot device
  #BOOTDEV	sc11			# Emulex SC11/B boot device
  #BOOTDEV	sc21			# Emulex SC21 boot device
*** /usr/src/sys/conf/NONET.old	Wed Nov 22 21:26:15 1995
--- /usr/src/sys/conf/NONET	Sat Dec 27 17:19:22 2008
***************
*** 50,57 ****
  #BOOTDEV	hk6			# rk06 boot device
  #BOOTDEV	hk7			# rk07 boot device
  BOOTDEV		ra			# MSCP boot device
  #BOOTDEV	rl			# rl01/02 boot device
! #BOOTDEV	rm			# rm02/03/05 boot device
  #BOOTDEV	br			# Eaton BR1537/BR1711 boot device
  #BOOTDEV	sc11			# Emulex SC11/B boot device
  #BOOTDEV	sc21			# Emulex SC21 boot device
--- 50,58 ----
  #BOOTDEV	hk6			# rk06 boot device
  #BOOTDEV	hk7			# rk07 boot device
  BOOTDEV		ra			# MSCP boot device
+ #BOOTDEV	rk			# rk05 boot device
  #BOOTDEV	rl			# rl01/02 boot device
! #BOOTDEV	rm			# rm02/03/05 rp06 boot device
  #BOOTDEV	br			# Eaton BR1537/BR1711 boot device
  #BOOTDEV	sc11			# Emulex SC11/B boot device
  #BOOTDEV	sc21			# Emulex SC21 boot device
*** /usr/src/sys/autoconfig/rkauto.c.old	Wed Dec 30 17:56:32 1992
--- /usr/src/sys/autoconfig/rkauto.c	Sat Dec 27 17:25:32 2008
***************
*** 16,23 ****
--- 16,39 ----
  	struct rkdevice *addr;
  	int vector;
  {
+ 	extern int errno;
+ 
  	stuff(RKCS_IDE | RKCS_DRESET | RKCS_GO, (&(addr->rkcs)));
  	DELAY(10L);
  	stuff(0, (&(addr->rkcs)));
+ 
+ 	/* in case no drive 0 was attached, the DRESET above will leave
+ 	 * the controller in an error state. The grab below checks for this
+ 	 * and the stuff below executes a RESET (control reset) to clear
+ 	 * the error condition. Without this the first access to any drive
+ 	 * will fail when the system is up.
+ 	 * 2008-12-25 wfjm
+ 	*/
+ 	errno = 0;
+ 	if ((grab(&(addr->rker)) & RKER_DRE) && !errno) {
+ 		stuff(RKCS_RESET | RKCS_GO, (&(addr->rkcs)));
+ 		DELAY(10L);
+ 	}
+ 
  	return(ACP_IFINTR);
  }
*** /usr/src/sys/pdp/conf.c.old	Sat Oct 13 18:20:03 2001
--- /usr/src/sys/pdp/conf.c	Sat Dec 27 17:21:22 2008
***************
*** 23,34 ****
  
  #include "rk.h"
  #if NRK > 0
! int	rkopen(), rkstrategy();
  daddr_t	rksize();
  #define	rkclose		nulldev
  #else
  #define	rkopen		nodev
  #define	rkclose		nodev
  #define	rkstrategy	nodev
  #define	rksize		NULL
  #endif
--- 23,35 ----
  
  #include "rk.h"
  #if NRK > 0
! int	rkopen(), rkroot(), rkstrategy();
  daddr_t	rksize();
  #define	rkclose		nulldev
  #else
  #define	rkopen		nodev
  #define	rkclose		nodev
+ #define rkroot		nulldev
  #define	rkstrategy	nodev
  #define	rksize		NULL
  #endif
***************
*** 192,198 ****
  	raopen,		raclose,	rastrategy,	raroot,		rasize,
  	0,
  /* rk = 6 */
! 	rkopen,		rkclose,	rkstrategy,	nulldev,	rksize,
  	0,
  /* rl = 7 */
  	rlopen,		rlclose,	rlstrategy,	rlroot,		rlsize,
--- 193,199 ----
  	raopen,		raclose,	rastrategy,	raroot,		rasize,
  	0,
  /* rk = 6 */
! 	rkopen,		rkclose,	rkstrategy,	rkroot,		rksize,
  	0,
  /* rl = 7 */
  	rlopen,		rlclose,	rlstrategy,	rlroot,		rlsize,
*** /usr/src/bin/tar.c.old	Tue Oct  3 12:40:03 1989
--- /usr/src/bin/tar.c	Sat Dec 27 17:29:22 2008
***************
*** 55,61 ****
  	ino_t	inum;
  	dev_t	devnum;
  	int	count;
! 	char	pathname[NAMSIZ];
  	struct	linkbuf *nextp;
  };
  
--- 55,61 ----
  	ino_t	inum;
  	dev_t	devnum;
  	int	count;
! 	char	*pathname;
  	struct	linkbuf *nextp;
  };
  
***************
*** 604,615 ****
  			}
  			lp = (struct linkbuf *) getmem(sizeof(*lp));
  			if (lp != NULL) {
! 				lp->nextp = ihead;
! 				ihead = lp;
! 				lp->inum = stbuf.st_ino;
! 				lp->devnum = stbuf.st_dev;
! 				lp->count = stbuf.st_nlink - 1;
! 				strcpy(lp->pathname, longname);
  			}
  		}
  		blocks = (stbuf.st_size + (TBLOCK-1)) / TBLOCK;
--- 604,618 ----
  			}
  			lp = (struct linkbuf *) getmem(sizeof(*lp));
  			if (lp != NULL) {
! 				lp->pathname = getmem(strlen(longname)+1);
! 				if (lp->pathname != NULL) {
! 					lp->nextp = ihead;
! 					ihead = lp;
! 					lp->inum = stbuf.st_ino;
! 					lp->devnum = stbuf.st_dev;
! 					lp->count = stbuf.st_nlink - 1;
! 					strcpy(lp->pathname, longname);
! 				}
  			}
  		}
  		blocks = (stbuf.st_size + (TBLOCK-1)) / TBLOCK;
*** /usr/src/lib/libc/pdp/crt/ulrem.s.old	Sun Jun  6 20:47:00 1993
--- /usr/src/lib/libc/pdp/crt/ulrem.s	Sat Dec 27 17:30:34 2008
***************
*** 6,11 ****
--- 6,12 ----
   *  Version	Date		Modification
   *	0.0	02Feb91		1. Initial inspiration struck.
   *	1.0	05Jun93		2. Released into the Public Domain.
+  *      1.1     23Dec08         Revised, corrected KERNEL (no-FPP) code wfjm
  */
  
  #include "DEFS.h"
***************
*** 56,62 ****
--- 57,105 ----
  #else
  /*
   * ulrem for the kernel (uses only fixed point - no FP)
+  *
+  * uses lrem if lhs and rhs non-negative, otherwise reminder is calculated
+  * 'brute-force' via lhs-(lhs/rhs)*rhs using uldiv and lmul. (wfjm)
  */
+ 
+ 	.globl	ulrem
+ 	.globl	lrem, uldiv, lmul
+ ulrem:
+ ENTRY(ulrem)
+ 	tst	2(sp)		/ hi(lhs)
+ 	bmi	1f		/ if negative, a-(a/b)*b handling
+ 	tst	6(sp)		/ hi(rhs)
+ 	bmi	1f		/ if negative, a-(a/b)*b handling
+ 	jmp	lrem		/ if lhs and rhs >=0, use lrem
+ 
+ 1:	mov	r5,-(sp)	/ need frame pointer
+ 	mov	sp,r5		/ setup frame pointer
+ 	mov	12(r5),-(sp)	/ lo(rhs) for *
+ 	mov	10(r5),-(sp)	/ hi(rhs)
+ 	mov	12(r5),-(sp)	/ lo(rhs) for /
+ 	mov	10(r5),-(sp)	/ hi(rhs)
+ 	mov	6(r5),-(sp)	/ lo(lhs) for /
+ 	mov	4(r5),-(sp)	/ hi(lhs)
+ 	jsr	pc,uldiv	/ do lhs/rhs
+ 	add	$10,sp		/ clean up stack
+ 	mov	r1,-(sp)	/ lo(lhs/rhs)
+ 	mov	r0,-(sp)	/ hi(lhs/rhs)
+ 	jsr	pc,lmul		/ do (lhs/rhs)*rhs
+ 	add	$10,sp		/ clean up stack
+ 	mov	r1,-(sp)	/ lo((lhs/rhs)*rhs))
+ 	mov	r0,-(sp)	/ hi((lhs/rhs)*rhs))
+ 	mov	6(r5),r1	/ lo(lhs)
+ 	mov	4(r5),r0	/ hi(lhs)
+ 	sub	(sp)+,r0	/ hi(lhs-((lhs/rhs)*rhs))
+ 	sub	(sp)+,r1	/ lo(lhs-((lhs/rhs)*rhs))
+ 	sbc	r0		/ handle carry from lo to hi part
+ 	mov	(sp)+,r5	/ restore r5
+ 	rts	pc
+ 
+ #ifdef NEVER
+ /* the old code here for reference and further debugging.
+  * a 1%10 produced a 0 instead of a 1.
+  */ 
  	.globl ulrem
  ulrem:
  ENTRY(ulrem)
***************
*** 131,136 ****
--- 174,180 ----
  	mov	(sp)+,r3
  	mov	(sp)+,r2
  	rts	pc
+ #endif NEVER
  #endif KERNEL
  
  /*
*** /usr/src/sbin/umount/umount.c.old	Tue Jan 16 23:25:05 1996
--- /usr/src/sbin/umount/umount.c	Sat Dec 27 17:29:58 2008
***************
*** 67,73 ****
  	int argc;
  	register char *argv[];
  {
! 	int all, ch, errs;
  
  	/* Start disks transferring immediately. */
  	sync();
--- 67,73 ----
  	int argc;
  	register char *argv[];
  {
! 	int all, ch, errs=0;
  
  	/* Start disks transferring immediately. */
  	sync();
***************
*** 108,114 ****
  		errs = umountall();
  	} else
  		for (errs = 0; *argv != NULL; ++argv)
! 			if (umountfs(*argv) == 0)
  				errs = 1;
  	exit(errs);
  }
--- 108,114 ----
  		errs = umountall();
  	} else
  		for (errs = 0; *argv != NULL; ++argv)
! 			if (umountfs(*argv) != 0)
  				errs = 1;
  	exit(errs);
  }
*** /VERSION.old	Tue Dec 26 14:53:42 2006
--- /VERSION	Sat Dec 27 17:32:05 2008
***************
*** 1,5 ****
! Current Patch Level: 445
! Date: December 26, 2006
  
  2.11 BSD
  ============
--- 1,5 ----
! Current Patch Level: 446
! Date: December 27, 2008
  
  2.11 BSD
  ============