In the last article, I looked at the influence that Microsoft has on the BSD world. Two or three people disagreed with this, calling it Microsoft bashing. They weren't very specific, and none answered my question about what they objected to. Still, it isn't my intention to make other software look bad, I just want BSD to look good. Despite the title, my discussion last month wasn't really about Microsoft: it was about the way the industry is catering only for one software vendor.
This month, I'll look at a more comparable operating system: UNIX® System V. Most of the comparisons between BSD and System V that I have seen have looked at BSD from a System V perspective. In this article, I'll look at System V from a BSD perspective.
We had some discussion whether we should use PCs or Sun 3 workstations. In any case, we compared the prices and decided we could get better value for money out of PCs than we could from Sun, so we bought 20 MHz 386s with 8 MB of memory and 20" monitors, and installed Interactive UNIX version 2.2, a release of System V.3.2, along with the X window system 11.3.
It would be incorrect to say that we were delighted by the machines in all respects. It took us experienced software professionals an average of two days each to install the software. Once the base system was installed, we needed to install TCP/IP and X, along with options like NFS. Each package came on its own set of floppies and a little slip of paper with two-part ``activation keys'' like ltd005269 and timzdfuz. Type one of them wrong, and you can restart the installation. I don't recall how many packages we had, but I have a list of the packages on a slightly later installation of SCO UNIX 2.2: there were a total of 77 floppies in 14 different sets.
Once we had the software installed, we had a quite reasonable system at our disposal. Sure, there were some rough edges, and it didn't quite measure up to Tandem's own NonStop-UX version of System V.3.2, but it ran rings round MS-DOS. And there were some strange little problems, but we found workarounds for them after we discovered that phone support cost additional money, and what they offered didn't help anyway.
I was pleasantly surprised when the installation went smoothly. Well, much more smoothly than the System V installation, anyway. After all, there was just one tape to install from, and I only needed to decide which components to install. No multiple passes, no activation keys, but I did need a pocket calculator to work out my disk partitioning.
Using BSD/386 was even more of a revelation: it just worked. None of the messing around I had needed with Interactive. I was very impressed, and almost immediately wrote an article for the German magazine iX, which in those days specialized in UNIX. They had just published a ``review'' of two of the first Intel-based System V.4 offerings. The review was relatively short: they hadn't been able to install either of them.
System V is descended from the Sixth Edition of Research UNIX. At the time it diverged from Research UNIX, BSD was primarily a distribution of user programs, rather like a Linux distribution nowadays, and it had little influence on the direction of kernel development. In the early 80s, that changed. BSD started to develop serious kernel improvements, in particular virtual memory, the fast file system (now known as the UNIX file system or ufs) and integrated Internet software. This even caused a break in the development of Research UNIX: the First to the Seventh Editions were an uninterrupted development, but the Eighth Edition was based on 4.1cBSD.
System V did not include any of the BSD kernel features. The BSD tapes were available, of course, and a majority of vendors included some of these features, typically calling them ``Berkeley extensions''. In the meantime, though, System V was developing its own version of networking support, based on Dennis Ritchie's STREAMS concept. By contrast, Berkeley networking was called sockets, since they were the most obvious difference from the user standpoint. Many implementations of System V have both STREAMS and sockets network interfaces, though nowadays there is a tendency to implement sockets as a library interfacing to STREAMS.
Another issue, a very political one about 12 years ago, was AT&T's attempts to woo Sun Microsystems to use System V. At the time, SunOS was firmly based on 4.2BSD, and AT&T saw this as serious competition. They had already come to an agreement with Microsoft to merge the functionality of XENIX into System V.3, giving rise to System V.3.2. The agreement which was finally reached with Sun involved merging all BSD functionality into System V.4. In fact, the term ``merge'' is misleading; System V.4 kernels were about the same size as a System V.3 kernel and a 4.3BSD kernel put together, so maybe the term ``add'' would have made more sense.
The first time I used a Sun workstation (running SunOS 4) I had no manuals handy, and it drove me crazy that I couldn't even use the ps command, since it had different option letters. I typed ps -ef and got the message
ps: f: unknown option ps: usage: ps [-acCegjklnrStuvwxU] [num] [kernel_name] [c_dump_file] [swap_file]
Why this gratuitous difference?
The Seventh Edition ps command had exactly four option letters: a (show all processes), l (produce long listing), x (include dæmons) and k (get information from a kernel core dump instead of the running system). 4.4BSD has the first three of these, but with the advent of virtual memory the k flag changed to M. It also adds a u flag for additional information which the Seventh Edition didn't show. System V has also dispensed with the k flag, and also with the x flag: instead of writing ax, you write -e. Instead of writing u, you write -f. Oh yes, and System V insists on a hyphen (-) before the commands, so if on a System V machine (here SGI's IRIX 5.3) you write ps aux, you'll get the message
Usage: ps [ -edalfcjM ] [ -t termlist ] [ -u uidlist ]
[ -p proclist ] [ -g grplist ] [ -s sidlist ]
OK, you might think, big deal. Right, but some people do think it's a big deal. In fact, of course, the ps command is one of the commands most intimately related with the kernel internal structure. Here's an example looking at the status of process 1, init:
SunOS 4.1.3:
$ ps uax
USER PID %CPU %MEM SZ RSS TT STAT START TIME COMMAND
root 1 0.0 0.0 52 0 ? IW Feb 16 0:00 /sbin/init -
$ ps lax
F UID PID PPID CP PRI NI SZ RSS WCHAN STAT TT TIME COMMAND
20088000 0 1 0 0 5 0 52 0 child IW ? 0:00 /sbin/init -
FreeBSD 4.0:
$ ps up1
USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND
root 1 0.0 0.0 496 72 ?? Is 28Mar99 0:01.79 /sbin/init --
$ ps lp1
UID PID PPID CPU PRI NI VSZ RSS WCHAN STAT TT TIME COMMAND
0 1 0 0 10 0 496 72 wait Is ?? 0:01.79 /sbin/init --
IRIX 5.3:
$ ps -fp1
UID PID PPID C STIME TTY TIME COMD
root 1 0 0 13:00:36 ? 0:00 /etc/init
$ ps -lp1
F S UID PID PPID C PRI NI P SZ:RSS WCHAN TTY TIME COMD
30 S 0 1 0 0 39 20 * 68:40 801752c0 ? 0:00 init
You'll see a lot of differences between all three. Here's a summary:
System V and SunOS also have an additional field, F, which provides additional information, conveniently in hexadecimal. The value 30 for IRIX shows that both the process and its user area are in core. The value 20088000 for SunOS shows that the process has completed an exec system call, that it is orphaned (in other words, it has no parent, which is always the case for init) and, to quote the SunOS man page, ``init data space on demand, from vnode''. This last flag would probably require access to the sources to interpret correctly.
This is only one example, of course. At the beginning, it looked as if FreeBSD and SunOS are strongly related, and that System V is rather different. By now it should be apparent that the differences are more complicated than they appeared.
SunOS 4.1.3:
$ ifconfig -a
le0: flags=63<UP,BROADCAST,NOTRAILERS,RUNNING>
inet 192.109.197.145 netmask ffffff00 broadcast 192.109.197.255
ether 8:0:20:e:2c:98
lo0: flags=49<UP,LOOPBACK,RUNNING>
inet 127.0.0.1 netmask ff000000
FreeBSD 4.0:
$ ifconfig -a
ed2: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
inet 192.109.197.137 netmask 0xffffff00 broadcast 192.109.197.255
ether 00:80:48:e6:a0:61
ppp0: flags=8051<UP,POINTOPOINT,RUNNING,MULTICAST> mtu 1500
inet 139.130.136.133 --> 139.130.136.129 netmask 0xffff0000
lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> mtu 16384
inet 127.0.0.1 netmask 0xff000000
IRIX 5.3:
$ ifconfig -a
ifconfig: ioctl (SIOCGIFFLAGS): no such interface
usage: ifconfig interface
[ af [ address [ dest_addr ] ] [ up ] [ down ][ netmask mask ] ]
[ metric n ]
[ primary ]
[ arp | -arp ]
$ ifconfig ec0
ec0: flags=c63<UP,BROADCAST,NOTRAILERS,RUNNING,FILTMULTI,MULTICAST>
inet 192.109.197.146 netmask 0xffffff00 broadcast 192.109.197.255
$ ifconfig lo0
lo0: flags=1849<UP,LOOPBACK,RUNNING,MULTICAST,CKSUM>
inet 127.0.0.1 netmask 0xff000000
Again, the most obvious thing we see here is that the output is same, but different. IRIX doesn't understand the -a flag to ifconfig, so we have to ask for them piecemeal. When we do, the output looks pretty much the same as for the BSD systems. The only difference is that it doesn't show the Ethernet address for the Ethernet interface. FreeBSD has an additional interface, ppp0, which is the gateway to the Internet, but that's not a feature of the operating system. In addition, the interface flags are similar, but not the same; it's clear that even most of the bit positions in the representation are the same for all three.
How about the routing tables? Here they are:
SunOS 4.1.3:
$ netstat -r Routing tables Destination Gateway Flags Refcnt Use Interface loopback loopback UH 1 1390 lo0 192.109.197.0 iskra U 9 2872 le0
FreeBSD 4.0:
$ netstat -r Routing tables Internet: Destination Gateway Flags Refs Use Netif Expire default Cont0.way3.Adelaid UGSc 19 37438 ppp0 localhost localhost UH 11 7508 lo0 Cont0.way3.Adelaid freebie UH 15 0 ppp0 widecast ff:ff:ff:ff:ff:ff UHLWb 0 19 ed2 => widecast link#1 UC 0 0 ed2 allegro 0:0:c0:44:a5:68 UHLW 9 27709597 ed2 662 freebie 0:80:48:e6:a0:61 UHLW 12 1005213 lo0 yana 0:0:b4:33:6d:a2 UHLW 4 106587923 ed2 333 iskra 8:0:20:e:2c:98 UHLW 3 1794 ed2 531 raptor 8:0:69:1:7:7 UHLW 5 2136 ed2 363 panic 0:a0:24:37:c:bd UHLW 12 6827834 ed2 998 papillon link#1 UHLW 1 71551886 ed2 broadcast ff:ff:ff:ff:ff:ff UHLWb 2 84577 ed2
IRIX 5.3:
$ netstat -r Routing tables Destination Gateway Flags MTU RTT RTTvar Use Interface localhost.lemis. localhost.lemis.co UH 0 0 0 0 lo0 raptor.lemis.com localhost.lemis.co UH 0 0 0 62 lo0 default freebie.lemis.com UG 0 0 0 0 ec0 192.109.197 raptor.lemis.com U 0 0 0 2002 ec0 BASE-ADDRESS.MCA raptor.lemis.com U 0 0 0 0 ec0
The biggest difference here is FreeBSD, not IRIX. IRIX has a number of different fields, and it shows a path to itself (raptor.lemis.com) via the local host, and the names are fully qualified. This latter may be a matter of configuration.
The difference between the SunOS routing table and the FreeBSD routing table is a question of evolution. FreeBSD includes information on link-level routing (flagged with L in the display). In this example, it's all Ethernet (interface ed2), and the Gateway value is the Ethernet address. In the case of system papillon, arp has timed out, so it shows link#1 instead. The next time a packet goes to or from papillon, this value will be replaced by the Ethernet address. broadcast and widecast are broadcast addresses (widecast is address 0, required by some brain damage in SunOS 4). As a result, the Ethernet address is the Ethernet broadcast address.
SunOS iskra IRIX raptor FreeBSD freebie.lemis.com
Well, there was a difference, anyway. Let's see what IRIX's man page on hostname says:
hostname(5) hostname(5)
NAME
hostname - hostname resolution description
DESCRIPTION
Hostnames are domains, where a domain is a hierarchical, dot-separated
list of subdomains; for example, the machine monet, in the Berkeley
subdomain of the EDU subdomain of the Internet would be represented as
monet.Berkeley.EDU
(with no trailing dot).
Does this look like System V? What does it look like in FreeBSD?
HOSTNAME(7) FreeBSD Miscellaneous Information Manual HOSTNAME(7)
NAME
hostname - host name resolution description
DESCRIPTION
Hostnames are domains, where a domain is a hierarchical, dot-separated
list of subdomains; for example, the machine monet, in the Berkeley sub-
domain of the EDU subdomain of the Internet would be represented as
monet.Berkeley.EDU
(with no trailing dot).
In other words, nothing has changed except for the format and the manual section. Even the name of the system shouts ``BSD''. Interestingly, I have other man pages, from the Lachman implementation of System V STREAMS TCP/IP, which have changed the name: monet.Berkeley.EDU has become laiter.Lachman.COM. The rest of the text appears identical.
Anyway, I tried this on the IRIX machine:
$ hostname -s raptor.lemis.com $ netstat -r Routing tables Destination Gateway Flags MTU RTT RTTvar Use Interface localhost localhost UH 0 0 0 0 lo0 raptor localhost UH 0 0 0 62 lo0 default freebie UG 0 0 0 0 ec0 192.109.197 raptor U 0 0 0 2903 ec0 BASE-ADDRESS.MCA raptor U 0 0 0 0 ec0So yes, the fully qualified name is due to the host name. I can see the same effect in FreeBSD, but interestingly, not in SunOS.