|
|
FreeBSD Remote Upgrades
Purpose
For:
- Experts remote upgrading many hosts in parallel,
- Using mouse cut & pasting commands from remote
xterms
- Keeping a table of hosts & numeric action points
completed,
- To avoid needing human memory & de-link from real
life interrupts, eg breaks for going off line, to lunch,
home network & power outages & customers calling
for other support etc).
Not For:
Newbies learning how to upgrade single domestic systems:
For that first learn from standard FreeBSD
documentation.
Disclaimer: Total !
Matrix
To copy then edit as you upgrade.
Horizontal Top: Action Numbers below
Vertical Left: Host names.
XXXX 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3
XXXX 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
aaaa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
bbbb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
cccc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
dddd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Keys: R = Running, 0 = Not tried, 1 =
Done, X = Failed, N = Next, - = Skip
(perhaps cos not enough space), W = Waiting on something.
Colour Scheme Documenting Action Numbers
| Blue |
Source Host as normal user.
|
| Blue Bold |
Source Host as Root
User |
| Orange |
Target Host as normal
user |
| Orange Bold |
Target Host as Root
User |
| Green |
Comment |
HOST=TARGET
Action Numbers
-
|
If this is an Internet live
server, some days before, you have already changed
timeouts for your domains, & patched out the IP
of this box from your cluster.
Check Assumptions:
- AMD running.
- You are using a JHS
standard AMD host mount path (not a FreeBSD default
one).
- Your shell is csh with a
built in echo (not eg sh needing
/bin/echo).
- Set pwd of root & toor
& a normal login user to know values & You
have already manually saved all user data & any
source hacks, creating diffs on another
system.
/bin/csh
|
-
Backup /etc/ configs to other
hosts in case of catastrophe
cd / && rdist6 -P
/usr/bin/ssh -f /etc/Distfile |
-
Make sure processor type of
target can use what we
have generated.
dmesg | grep CPU:
|
HOST=SOURCE
-
I have proven one can boot a
cross installed kernel, where SOURCE cflags has 686
& TARGET cpu=586, however, it would produce
unusable src/ binaries, & I'm not sure of /boot
stuff, so best avoid the risk.
grep CFLAGS
/etc/make.conf |
-
Ensure we have all binaries
ready
cd /usr/src && make
world
Or
cd /usr/src && make
all |
-
Build GENERIC kernel, in case
our specific kernel fails to configure or boot.
|
-
Build host specific
kernel.
cd /sys/i386/conf &&
config $TARGET.small
cd /sys/compile/$TARGET.small && make depend
&& make |
HOST=TARGET
-
Make space: Remove src ports doc
obj
rm /usr/src/.c*
/usr/ports/.c*
Cope with possible symbolic
links of src ports doc obj"
foreach i ( src ports doc
obj )
echo
"Emptying /usr/$i"
( rm -rf /usr/$i/* )
end
|
-
Check enough space free for new
binaries (145M for 4.11, 370M for 7.2-RELEASE ) where
/usr/destdir will target
df / /usr*
If not enough space try
eg
cd /usr && mv local
/usr1/ && ln -s ../usr1/local
/usr/local
Either
mkdir
/usr/NEW.ALL
Else
(Which won't work so well, as
moving across file systems when binaries are not in
findable paths, amp; will fail & need manual
intervention.
mkdir /usr1/NEW.ALL
ln -s -f ../usr1/NEW.ALL /usr/NEW.ALL
|
HOST=SOURCE
-
Copy across (but do not install)
binaries & sources before booting new kernel, (in
case reboot fails & we have to do local login).
.
setenv target host
setenv TARGET `echo $target | dd conv=ucase`
setenv DESTDIR /host/$target/usr/NEW.ALL
mkdir /host/$target/usr/NEW.ALL
Can't do a direct install to
root of target cos of lib problems over NFS so stage
it via /usr/NEW.ALL
cd /usr/src
pushd etc
make distrib-dirs
make distribution
make etc-examples
make install
popd
make install
cd /sys/i386/compile/GENERIC.small ; make
install
cd /sys/i386/compile/HOST.small ; make
install
HOST=TARGET
Test Binaries (make sure eg you
haven't installed 586 bins on a 486).
(The test
above is not adequate, it doesn't blow up if eg on a
486 target you run a 586 mount binary.)
cd /usr/NEW.ALL/sbin
&& ./mount
|
HOST=TARGET
-
Install src ports doc obj
We did the binaries before this in case we ran out of
space.
cd /usr && mkdir src
ports doc obj
cd /usr && ls -al src ports doc obj
cd /usr && chown -R -H jhs:staff src ports doc
obj |
LOGIN=jhs
-
If $TARGET is allowed AMD mounts
by $SOURCE:
foreach i ( src ports doc
)
cd /usr/$i && tar
zxf
/pri/FreeBSD/releases/7.2-RELEASE/tars/$i.tgz
echo done $i
end
Else if $TARGET is a gate that
$SOURCE does not allow AMD mounts from:
HOST=SOURCE
foreach i ( src ports doc
)
cd
/host/$target/usr/$i && tar zxf
/pri/FreeBSD/releases/`uname
-r`/tars/$i.tgz
echo done $i
end
|
-
cd /usr/ports && ln
-s -f ../distfiles
cd /usr/ports && ln -s -f ../packages
cd /usr/ports/.. && ln -s -f
/pub/FreeBSD/dists/7.2-RELEASE distfiles
cd /usr/ports/.. && ln -s -f
/pub/FreeBSD/ports/i386/packages-`echo
7.2-RELEASE|dd conv=lcase` packages
|
-
chflags noschg
/boot/kernel/kernel*
cp /boot/kernel/kernel /kernel.OLD
|
HOST=SOURCE
-
Optionally Install new live
GENERIC kernel as fallback.
Don't do this if eg
- Your network card is not
configured for a GENERIC config.
- Your host is a HP Network
ScanJet 5 & a Generic kernel will hang
during boot on aic probe & never
complete.
setenv DESTDIR
/host/$target/NEW
cd /sys/i386/conf/../../compile/GENERIC &&
make install
|
-
Install new live custom kernel
(old release or new release Generic becomes
fallback)
setenv DESTDIR
/host/$target/NEW
cd /sys/i386/conf/../../compile/$TARGET.small
&& make install |
HOST=TARGET
-
Do you have special
/boot/loader.conf /boot/kernel.conf that might be
essential to boot.
find /boot -name \*.conf |
grep -v /defaults/ |
HOST=SOURCE
-
Install new boot & kernel
modules
setenv DESTDIR /host/$target
cd /sys && make install |
HOST=TARGET
-
cd /etc/mail ; make
stop
vi -c/sendmail_enable /etc/rc.conf #
sendmail_enable="NO" to
avoid start on after reboot, before new configs in
place
cd /usr/local/etc/rc.d ; ./apache stop
/etc/rc.d/named stop # kill -9 `cat
/var/run/named/pid`
ps -laxxww # look for
anything else to kill
echo "Upgrading now, kicking users off" |
wall
rsh local fetchmail # Drain mail |
-
Last chance to Think
!
rm -f /var/run/nologin
/etc/nologin # (Path
in /etc/login.conf & man sshd )
Ensure no top directories lack
permissions (eg from inherited umask from mv / cp)
else eg if /bin is drwx------ you will get /bin/csh:
Permission denied
cd / ; ls -l | grep
rwx---
Ensure alternate rescue
entries.
grep "^PermitRootLogin"
/etc/ssh/sshd_config
Look for "yes", else if normal
user login fails, you'll be stuck.
grep toor /etc/passwd ;
vipw
Ensure rlogin & telnet also allow login, because
after upgrading from 6.1 to 6.3 I got:
ssh_exchange_identification: Connection closed by
remote host
mv /etc/pam.conf
/etc/pam.conf.MV
Ensure no obsolete /etc/pam.conf
might lock you out.
Optional sanity check febore
upfrade of binaries: reboot
|
-
Move new binary trees to right
file systems.
mkdir /OLD /usr/OLD
/var/OLD
mv /usr/NEW.ALL/var /var/NEW
mv /usr/NEW.ALL/usr /usr/NEW
Next line complex to
avoid just moving a possible symbolic link
You cant just do this
cd /usr/NEW.ALL/../ &&
mv NEW.ALL /NEW
because /rescue contains 121 sym
links of 4 Meg, so do this:
( cd /usr/NEW.ALL && tar cf - . ) | ( cd
/NEW && tar xf - )
|
-
Installing /var
( rmdir /var/NEW/* )
cd /var/NEW && foreach i ( * )
echo "Starting $i"
mv /var/$i /var/OLD/$i
mv $i /var/$i
echo "Finished $i"
end
|
-
Installing /usr
( rmdir /usr/NEW/* )
cd /usr/NEW && foreach i ( * )
echo "Starting $i"
mv /usr/$i /usr/OLD/$i
mv $i /usr/$i
echo "Finished $i"
end
|
-
Installing Root
find /boot -name \*.conf |
grep -v /defaults/
( cp /boot/loader.conf
/NEW/boot/ )
( cp /boot/kernel.conf /NEW/boot/ )
( rmdir /NEW/* )
rm /NEW/sys
foreach i ( .cshrc .profile dev etc root
)
mv /NEW/$i /$i.NEW
end
Copy tools to known location,
normally only mv is needed, but if you get anything
wrong, eg on the wrong file system, mv will fail to
fork /bin/cp
(cd /bin && file cp
ls mv pwd rm | grep "statically linked" ;
Ensure they are static, not
dynamic.
cd /bin && cp cp ls mv pwd rm / )
cd /NEW && foreach i ( *
)
echo "Starting
$i"
/mv /$i /OLD/$i
/mv /NEW/$i /$i
echo "Finished $i"
end
(cd / && rm mv cp rm
pwd ls )
rm /etc/objformat /usr/bin/objformat
cmp /usr/src/etc/rc
/etc/rc # See if mergemaster has failed to update,
(when that happened next reboot had no net &
hostname Amnesiac)
|
-
Check system is bootable, &
load new shared libraries etc.
reboot |
-
Delete old binaries.
rmdir /NEW /usr/NEW
cd / && ls /OLD /usr/OLD
chflags -R noschg /OLD /usr/OLD /kernel.Old
rm -rf /mv /OLD /usr/OLD /kernel.Old
ls /var/NEW
/bin/csh |
-
Optionally fix some rc.conf
& rc.local type errors, (visible after booting with
X off). But some errors not worth fixing as we haven't
upgraded /etc/ scripts yet.
/bin/csh |
-
Manually also install
/usr/local if a major release upgrade, where old shared
libs wont work any more.
/bin/csh |
LOGIN=jhs
-
cd tmp
&&
script
Optionally, don't customise if
anything might break world, eg SASL on a
gate
|
-
LOGIN=root
-
Rebuild binaries with local
/etc/make.conf CFLAGS
cd /usr/src
sh
nohup make world &
(nohup allows build
(running for hours) to continue, even if you later
reboot or halt or drop the link of the controlling
screen.
At any time monitoring is resumable with
tail -f
/host/$target/usr/src/nohup.out
|
-
Compare & upgrade /etc
with/from /usr/src/etc
cd /etc && tar zcf
../etc.`date +%Y_%m_%d_%H_%M_%S`.tgz .
mergemaster -sicv |
-
You need a new
/etc/mail/sendmail.cf by now else you wont get error
logs mailed
cd /usr/obj/`cd
/usr/src;/bin/pwd`/etc/sendmail;
cp `hostname`.cf /etc/mail
cp *submit* /etc/mail
cd /etc/mail && csh |
-
Optionally reinstall packages, or
mandatory if a major release upgrade. cd /var/db/pkg && ls
pkg_info -a
You may want to upgrade with
some of /usr/ports/*/pkg*
rlogin #source ; cd
/host/$target/usr/local (cd /usr/local && tar
cf - . ) | tar xf -
rlogin #source ; cd /host/$target/var/db/pkg ( cd
/var/db/pkg && tar cf - . ) | tar xf
-
|
-
Optionally remake ports, or
mandatory if a major release.
portsupgrade |
-
Detect binaries that failed to
rebuild.
foreach i ( /*bin /usr/*bin /usr/local/bin
)
echo Listing $i
ls -ltr $i
echo " "
end
|
-
Look to see errors in
/var/log/, especially messages.
cd /var/log ; more
messages |
-
If this is a live internet
server, check:
- ftpd: Ensure no incoming
etc.
- popd Enable befor
smtpd
- smtp Enable before
named
- httpd Enable &
check
more
/site/domain/this/usr/local/etc/apache/httpd.conf
- named: re-enable the IP
into domain cluster, & reset DNS timeouts to
longer defaults.
vi .... rdist6 ....
/site/domain/init
-
ssh: Check ssh allows login,
then Optionally:
vi /etc/inetd.conf #
comment out rlogin & telnet kill -HUP `cat
/var/run/inetd.pid` to make machine safer, (unless all
users are aware rlogin & telnet are
dangerous, can be packet sniffed to steal
passwords, so normal users should use
ssh.)
- SASL:grep CFLAGS
/etc.make.conf | grep i sasl #
- cd /usr/src ; make
world
- customise
- make ; make
install
- Upgrade
/etc/mail/sendmail.cf & re-enable
|
| |