[email protected]
dd if=/dev/hda8 of=/etc/dosswap
gzip -9 /etc/dosswap
mkswap /dev/hda8 XXXXX
swapon -av
swapoff -av
zcat /etc/dosswap.gz | dd of=/dev/hda8 bs=1k count=100
>> What are the pros and cons of doing this?
Pros: you save a substantial amount of disk space.
Cons: if step 5 is not automatic, you have to remember to do it by hand, and it slows the reboot process by a nanosecond :-)
[email protected]
Here's a trick I've had to use a few times.
Desperate person's text file undelete.
If you accidentally remove a text file, for example, some email, or the results of a late night programming session, all may not be lost. If the file ever made it to disk, ie it was around for more than 30 seconds, its contents may still be in the disk partition.
You can use the grep command to search the raw disk partition for the contents of file.
For example, recently, I accidentally deleted a piece of email. So I immediately ceased any activity that could modify that partition: in this case I just refrained from saving any files or doing any compiles etc. On other occasions, I've actually gone to the trouble of bring the system down to single user mode, and unmounted the filesystem.
I then used the egrep command on the disk partition: in my case the email message was in /usr/local/home/michael/, so from the output from df, I could see this was in /dev/hdb5
sputnik3:~ % df
Filesystem 1024-blocks Used Available Capacity Mounted on
/dev/hda3 18621 9759 7901 55% /
/dev/hdb3 308852 258443 34458 88% /usr
/dev/hdb5 466896 407062 35720 92% /usr/local
sputnik3:~ % su
Password:
[michael@sputnik3 michael]# egrep -50 'ftp.+COL' /dev/hdb5 > /tmp/x
I then used strings to help me inspect the output
strings /tmp/x | less
This method can't be relied on, all, or some, of the disk space may have already been re-used.
This trick is probably only useful on single user systems. On multi-users systems with high disk activity, the space you free'ed up may have already been reused. And most of use can't just rip the box out from under our users when ever we need to recover a file.
On my home system this trick has come in handy on about three occasions in the past few years - usually when I accidentally trash some of the days work. If what I'm working survives to a point where I feel I made significant progress, it get's backed up onto floppy, so I haven't needed this trick very often.
[email protected]
Use the Immutable Flag
Right after you install and configure your system go through the /bin, /sbin/, /usr/bin, /usr/sbin and /usr/lib (and a few of the other usual suspects and make liberal use of the 'chattr +i command'. Also add that to the the kernel files in root. Now 'mkdir /etc/.dist/' copy everything from /etc/ on down (I do this in two steps using /tmp/etcdist.tar to avoid recursion) into that directory. (Optionally you can just create /etc/.dist.tar.gz) -- and mark that as immutable.
The reason for all of this is to limit the damage that you can do when logged in as root. You won't overwrite files with a stray redirection operator, and you won't make the system unusable with a stray space in an 'rm -fr' command (you might still do alot of damage to your data -- but your libs and bins will be safer.
This also makes a variety of security and denial of service exploits either impossible or more difficult (since many of them rely on overwriting a file through the actions of some SUID program that *isn't providing an arbitrary shell command*).
The only inconvenience of this is when building and doing your 'make install' on various sorts of system binaries. On the other hand it also prevents the 'make install' from over-writing the files. When you forget to read the Makefile and chattr -i the files that are to be overwritten (and the directories to which you want to add files) -- the make fails, you just use the chattr command and rerun it. You can also take that opportunity to move your old bin's, libs, or whatever into a .old/ directory or rename or tar them or whatever.
[email protected]All new stuff starts under /usr/local! or /usr/local/`hostname`
If your distribution is one that leaves /usr/local empty then just create your /usr/local/src, /usr/local/bin etc and use that. If your distribution puts things in the /usr/local tree than you may want to 'mkdir /usr/local/`hostname`' and give the 'wheel' group +w to it (I also make it SUID and SGID to insure that each member of the wheel group can only mess with their own files thereunder, and that all files created will belong to the 'wheel' group.
Now discipline yourself to *ALWAYS! ALWAYS! ALWAYS!* put new packages under /usr/local/src/.from/$WHEREVER_I_GOT_IT/ (for the .tar or whatever files) and build them under /usr/local/src (or .../$HOSTNAME/src). Make sure that it installs under the local hierarchy. If it *absolutely must* be installed back in /bin or /usr/bin or somewhere else -- put a symlink from the local heirarchy to each element that when anywhere else.
The reason for this -- even though it's more work -- is that it helps isolate what has to be backed up and restored or reinstalled in the event of a full re-install from the distribution medio (usually CD these days). By using a /usr/local/.from directory you also keep an informal log of where your sources are coming from -- which helps when you're looking for new updates -- and may be critical when monitoring the security announcement lists.
One of my systems at home (the one I'm calling from) was put together before I adopted these policies for myself. I still don't "know" all the ways that it differs from the stock "as installed" system. This is despite the fact that I've done very little with my home system's configuration and I'm the *only* person who ever uses it.
By contrast the systems I've set up at work (when I was thrust into the role of system administrator there) have all been configured this way -- have been administered by many contractors and other MIS people, and have had a large number of upgrades and package installations. Nonetheless I have a very good idea which precise elements were put in *after* the initial installation and configuration.
[email protected]
I noticed a few overly difficult or unnecessary procedures recommended in the 2c tips section of Issue 12. Since there is more than one, I'm sending it to you:
#!/bin/sh
# lowerit
# convert all file names in the current directory to lower case
# only operates on plain files--does not change the name of directories
# will ask for verification before overwriting an existing file
for x in `ls`
do
if [ ! -f $x ]; then
continue
fi
lc=`echo $x | tr '[A-Z]' '[a-z]'`
if [ $lc != $x ]; then
mv -i $x $lc
fi
done
for i in * ; do [ -f $i ] && mv -i $i `echo $i | tr '[A-Z]' '[a-z]'`;
done;
The contributor says he wrote the script how he did for understandability (see below).
On the next tip, this one about adding and removing users, Geoff is doing fine until that last step. Reboot? Boy, I hope he doesn't reboot every time he removes a user. All you have to do is the first two steps. What sort of processes would that user have going, anyway? An irc bot? Killing the processes with a simple
kill -9 `ps -aux |grep ^<username> |tr -s " " |cut -d " " -f2`
kill -9 `ps -aux |grep ^foo |tr -s " " |cut -d " " -f2`
The solution given in the Gazette is the most universal one, but not the easiest one. With both LILO and loadlin, one may provide the boot parameter "single" to boot directly into the default shell with no login or password prompt. From there, one may change or remove any passwords before typing "init 3" to start multiuser mode. Number of reboots: 1 The other way Number of reboots: 2
Justin Dossey
[email protected]We're starting from raw, clean source. First, obtain the sendmail source code. I've d/led version 8.9.0, which is, as you will notice, bleeding edge. I grabbed it from ftp.sendmail.org:/pub/sendmail/sendmail.8.9.0.tar.gz
It's about 1Meg, and considering I'm running 8.7.6, I think it's worth the effort. If this works, you'll undoubtedly hear about it, elsewise I can't get the new HOWTO versions out without e-mail:)
Now, once you've got the source d/led, unpack it. It'll create a dir called
sendmail-8.9.0 in the current directory. Change into that directory,
read the README and RELEASE_NOTES files(and be amazed at the updates they've
done). Now, cd in src. This is where most of your work will be done.
A quick note: Sendmail is a small, powerful and well-written program. The sendmail binary itself compiled in less than 5 minutes on my 5x86 133 with 32Megs RAM! The entire compile and install(sans config) took under 15 minutes!
I don't normally run BIND on my system, so I found the lines:
# ifndef NAMED_BIND
# define NAMED_BIND 1 /* use Berkeley Internet Domain Server */
# endif
and changed the 1 to a 0, ala:
# ifndef NAMED_BIND
# define NAMED_BIND 0 /* use Berkeley Internet Domain Server */
# endif
On Debian 1.3.1, db.h is by default installed in /usr/include/db, instead of /usr/include, where sendmail hopes to find it. Change to the src, mailstats, makemap, praliases, rmail and smrsh directories and execute the following command:
./Build -I/usr/include/db
Once you've done that, cd .. and type make install. There! Sendmail version 8.9.0 should now be installed! This is, of course, assuming you already have your original configuration. For everything to work smoothly on my system, since I host free mailing lists for people using majordomo, I had to add the following to the beginning of my /etc/sendmail.cf:
O DontBlameSendmail=forwardfileinunsafedirpath, forwardfileinunsafedirpathsafe
Sendmail 8.9.0 is rather pedantic about directory and file permissions these days, and will complain about dirs and files in aliases or .forward files that are group or world writeable. While it's not a good idea to disable this pedantry, I am only running with a single person at the console and I felt it was okay to allow this minor security hole. YMMV.
[email protected]Create and maintain a /README.`hostname` and/or a /etc/README.`hostname` [Or possibly /usr/local/etc/README.`hostname` -Maint. ]
Absolutely, from *day one* of administering a system take notes in an online log file. You might make "vi /README.$(hostname)" a line in root's /bash_logout. Another way to do this is to write an su or a sudo script that does something like:
function exit \
{ unset exit; exit; \
cat ~/tmp/session.$(date +%y%m%d) \
>> /README.$(hostname) && \
vi /README.$(hostname)
}
script -a ~/tmp/session.$(date +%y%m%d)
/bin/su.org -
(use the typescript command to create a session log and create a function to automate appending and updating the log).
I'll admit that I haven't implemented this automation of policy -- I've just relied on self-discipline so far. However I have been toying with the idea (even to the point of prototyping the scripts and shell functions as you see them). One thing that holds me back on this is the 'script' command itself. I think I'll have to grab the sources and add a couple of command line parameters (to pause/stop the script recording from the command line) before I commit to using this).
My last suggestion (for this round):
Root's path should consist of 'PATH= /bin'
That's it. Nothing else on root's path. Everything root does is provided by a symlink from /bin or by an alias or shell function, or is a script or binary in /bin, or is typed out with an explicit path.
This makes anyone running as root aware (sometimes painfully so) of how he or she is trusting binaries. The wise admin of a multi-user host will periodically look through his or here /bin and /.*history files to look for patterns and loopholes.
The really motivated admin will spot sequences that can be automated, places where sanity checks can be inserted, and tasks for which "root" privileges should be temporarily eschewed (launching editors, MTA's and other large interactive programs with elaborate scripting features that *might* be embedded in transparent or data files -- like the infamous vi ./.exrc and emacs ./.emacs and the even more insidous $EXINIT and the embedded header/footer macros). Naturally those sorts of commands can be run with something like:
cp $data $some_users_home/tmp
su -c $origcommand $whatever_switches
cp $some_users_home/tmp $data
Mostly these last sorts of precautions are overboard for the home or "single" user workstation -- but they are very good policy the admin of a multi-user -- particular a publicly exposed system (like the one's at netcom).
[email protected]
/usr/bin/X11/xdm
exec /usr/bin/X11/X -indirect hostname
I add this because when I was, desperately, trying to set it up for my own subnet over here it took me about a week to suss out all the problems.
Caveat: with old SLS (1.1.1) for some reason you can leave a -nodaemon after the xdm line -- this does NOT work for later releases.