disallow pseudo terminal in ssh

Some Oracle documentation wants you to setup ssh with no password and no passphrase.

Configuring ssh

This is not really something your security admin will like.

ssh-keygen -t dsa
First, using DSA, which is deprecated and disabled by default in OpenSSH 7.0, is a pretty dump instruction
OpenSSH 7.0 and greater similarly disable the ssh-dss (DSA) public key algorithm. It too is weak and we recommend against its use.
The two recommended key types are rsa and ecdsa. You should not use dsa

Second, ssh-key without passphrase is a huge security hole. If one get access to your key, for instance on a disk, a tape backup, etc, she’ll get access as oracle to all your database nodes. Best practice to use a pass phrase. Depending on your setup, it is sufficient to get ssh keys at installation/upgrade time only.

Third, providing interactive ssh-login as Oracle is against best practice for tracability. You better use SUDO or another elevation mechanism.

Let’s try:

First, use a recommended algoryhtm and key-length.
ssh-keygen -t rsa -b 4096
ssh-keygen -t ecdsa -b 521

Then, use a passphrase

Enter passphrase (empty for no passphrase): ***
Enter same passphrase again: ***

Then, when creating you authorized key, disable unwanted features, like pseudo terminal



ecdsa-sha2-nistp521 AAAABBBB/cccc== oracle@srv001


no-agent-forwarding,no-port-forwarding,no-pty,no-user-rc,no-x11-forwarding ecdsa-sha2-nistp521 AAAABBBB/cccc== oracle@srv001

Also, you could deactivate some features on the client config


This could also be done one the server sshd_config, but if you are not the sysadmin, don’t mess up with it.

Because you have a passphrase, you need to use an agent before starting your installation. Because pseudo-terminal (no-pty) is disabled, you cannot get a prompt. Because x11 is disabled (no-x11-forwarding), you cannot start an xterm

$ ssh srv002
Permission denied
$ eval $(ssh-agent)
Agent pid 12345
$ ps -fp 12345
  oracle 123451 0    ssh-agent
$ ssh-add ~/.ssh/id_ecdsa
Enter passphrase for ~/.ssh/id_ecdsa: 
Identity added: ~/.ssh/id_ecdsa (~/.ssh/id_ecdsa)
$ ssh -t srv002
PTY allocation request failed on channel 0
$ ssh -Y srv002 aixterm
X11 forwarding request failed on channel 0
1363-008  X server named  was not found.
$ ssh srv002 date
Fri Jul 13 12:50:22 CEST 2018

Those are basic steps to make your ssh less unsecure.

Untrusted X11 forwarding

I wrote a while ago about my security concerns regarding

xhost +
xterm -display mypc:0

Way back then, I suggested ssh tunnel. SSH is pretty easy to set up, by enabling the X11Forwarding option.

In OpenSSH 3.8 release note, 2004, there was a new default .

ssh(1) now uses untrusted cookies for X11-Forwarding

In the man ssh_config page, it’s still documented as being the default

ForwardX11Trusted The default is ‘no’

But it actually isn’t on most *ix derivates, e.g. RedHat /etc/ssh/ssh_config

# If this option is set to yes then
# remote X11 clients will have full access
# to the original X11 display. As virtually
# no X11 client supports the untrusted
# mode correctly we set this to yes.
ForwardX11Trusted yes

Who is we?

Okay, let’s go back.

If you use the unsafest method, xhost + and xterm -display pc:0, then you grant everybody the right to manipulate X.

If you use trusted ssh, which is the _undocumented_ default in Linux, then you grant this right only to anyone with access to your authority, most probably located in the file $HOME/.Xauthority. So root and yourself, at least.

If you trust neither yourself nor root, you could restrict access to your resource, preventing one hacker from switching your mouse buttons or doing a screenshot. But this is probably going to prevent most of your applications from working. Also, it probably won’t work at all if you use putty, reflection and (virtually any?) other client tools.

If you want to force Trusted mode, use -Y or -o ForwardX11Trusted=yes.

If you want to force Untrusted mode, use -X and -o ForwardX11Trusted=no.

If you use only -X, it may transparently defaults to the more convenient but less secure -Y. Sometimes. At least on Linux OpenSSH. But if you use different Unix / SSH flavours, the -X may ends with an error message like connection to “localhost:10.0” refused by server. In that case, simply use -Y. Actually, always use -Y if you want Trusted.

run sudo, ssh, password, su in simulated interactive mode

Some commands do not like non-interactive mode

$ passwd <<EOF
> oldpassword
> newpassword
> newpassword
Changing password for user lsc.
Current password for lsc@example.com: passwd: Authentication token manipulation error
$ echo oraclepassword | su - oracle
standard in must be a tty
$ echo sudopassword | sudo su - oracle
[sudo] password for lsc:
sudo: sorry, you must have a tty to run sudo

But ok, if you really want to run those in a script, you have plenty of clean (no hack there) ways of doing this.

For instance, let’s use a screen called BAR.

$ xterm -e "screen -S BAR" &
[1]     31732

Now we have an interactive terminal. It could be redirected to a frame buffer device if no x server is started.

Not really a human device, but an interactive terminal.

Now let’s send stuff

$ CR="$(echo '\r')"
$ screen -S BAR -X stuff "sudo su - oracle$CR"
$ screen -S BAR -X stuff "sudopassword$CR"
$ screen -S BAR -X stuff "id > /tmp/xxx$CR"
$ screen -S BAR -X stuff "exit$CR"
$ screen -S BAR -X stuff "exit$CR"
[1] +  Done                    xterm -e "screen -S BAR" &
$ cat /tmp/xxx
uid=100(oracle) gid=100(dba) groups=100(dba)

Usual disclaimer: it is a bad security practice to hardcode your passwords in scripts. Use this only if you really understand security. Read man openssl about how to use openssl to encrypt your password. Ask your security friends before trying

How to *really* send a script to the background

Let’s check this small script


echo foo.1:`date` | tee $HOME/tmp/foo.txt
sleep 3
echo foo.2:`date` | tee -a $HOME/tmp/foo.txt

$ $HOME/tmp/foo.sh
foo.1:Thu Nov 27 17:34:53 CET 2014
foo.2:Thu Nov 27 17:34:56 CET 2014

Very obvious, I write to the console, wait three seconds, then write to the console.

Ok, let’s take another script that would call this script in the background using &


echo bar.1:`date`
$HOME/tmp/foo.sh &
echo bar.2:`date`

$ $HOME/tmp/bar.sh
bar.1:Thu Nov 27 17:36:32 CET 2014
bar.2:Thu Nov 27 17:36:32 CET 2014
foo.1:Thu Nov 27 17:36:32 CET 2014
foo.2:Thu Nov 27 17:36:35 CET 2014

bar is printing the date, calling foo in the background, then printing the date, then it returns to you, and foo is still running.

BUT this is only in a relative background …

Let’s try this

$ time $HOME/tmp/bar.sh > /dev/null

real    0m0.01s
user    0m0.00s
sys     0m0.00s

So it takes no time to run bar you believe ?

Let’s try, for instance, over ssh (or cron or whatever)

$ time ssh localhost $HOME/tmp/bar.sh > /dev/null
real    0m3.81s
user    0m0.01s
sys     0m0.01s

running bar suddenly waits 3 seconds for foo to finish.

To be sure the script is sent to the farest background, you need to close the file descriptors, stdin, stdout, stderr

I rewrote it as


echo bar.1:`date`
$HOME/tmp/foo.sh <&- >&- 2>&- &
echo bar.2:`date`

$ time ssh localhost $HOME/tmp/baz.sh >/dev/null
real    0m0.44s
user    0m0.00s
sys     0m0.00s

Now the script baz is immediately finished and does not wait for foo to complete

fun with cron

Today I find out that my scheduler was too busy to execute all jobs in my crontab !?

* * * * * (while :;do ssh example.com :; done)
59 23 19 06 * touch /tmp/bang

my while loop is going to produce so much hangs on the cron deamon that it may not be able to read the crontab once a minute. If it reads it at 23:58 and at 00:00, the 23:59 won’t be run.

This is actually the first time I see this behaviour. And -believe me- it’s annoying!

Testing for (non-)empty string in shell

One way to test for (non-)empty string is to use test and -z (-n)

$ x=foo
$ test -z "$x"
$ echo $?

This is mostly seen with an if and [ -z … ] syntax

$ y=bar
$ if [ -n "$y" ];
  then echo non-empty; 

Instead of a variable, it could be the output of a script.


if [ -n "$(grep ORA- alertDB01.log)" ] 
  echo there is an error in the alert log
  echo "fine :)"

This will work for years until one day you get :

ksh: no space

Why that? This is the way the shell works. Your shell (here ksh on AIX) starts having errors as soon as your subshell (here the grep) is exhausting the space.

$ wc -l alertDB01.log
       2 alertDB01.log
$ if [ -n "$(grep ORA- alertDB01.log)" ]; 
  then echo non-empty; 
  else echo "fine :)"; 
$ wc -l alertDB01.log
   75025 alertDB01.log
$ if [ -n "$(grep ORA- alertDB01.log)" ]; 
  then echo non-empty; 
  else echo "fine :)"; 
ksh: no space

You got a memory error, how the shell will react is random (core dump, errors, continue, crashes). It will just bug and you do not want this.

There is more than one to circumvent this. For instance you could use the return code of grep

$ if grep ORA- alertDB01.log >/dev/null;
  then echo non-empty;
  else echo "fine :)";

Different shells (Bash / Bourne) and different OSs (Linux / AIX / HPUX) may react differently. If AIX crashed with a 50’000 lines, it may scale up to millions of lines in recent Linux’s – but still use trucks of memory

hot to bypass requiretty in sudo

You can execute it a command without password from the commande line

$ sudo -l
User lsc may run the following commands on this host:
    (root) NOPASSWD: /usr/local/bin/helloworld
$ sudo /usr/local/bin/helloworld
Hello World!

Now you try to run it via cron and you get

sudo: sorry, you must have a tty to run sudo

The message is clear, you need a terminal.

Either you edit your sudoers files to disable requiretty, or you just get yourself a terminal.

Maybe you tried to assign a pseudo terminal with ssh -t, but you may get an error if ssh has no local tty

Pseudo-terminal will not be allocated because stdin is not a terminal.

Don’t despair, read man ssh

Multiple -t options force tty allocation, even if ssh has no local tty.

Let’s try

* * * * * ssh -t -t sudo /usr/local/bin/helloworld >> /tmp/txt

This should work, providing you configured ssh keys 馃檪

use cron to schedule a job only once

I wrote about not using DAY OF MONTH and DAY OF WEEK simultanously in how to cron

The correct method is to use
15 14 15 05 * /tmp/run-my-job

But… I wrote this five years ago. Hmmm ! Not that correct then since it would run every year 馃槈

Ok, periodically I check for jobs are scheduled to run a specific date only

$ crontab -l|awk '$1!~/#/&&$3*$4'
15 14 15 05 * /tmp/run-my-job

I have 9 more days to remove this before it runs for the fifth time 馃檪

remove the current directory

Can I remove the current directory?

I used to believe you cannot.


$ uname -s
$ mkdir /tmp/bla
$ cd /tmp/bla
$ rm -r /tmp/bla
rm: Cannot remove any directory in the path of the current working directory


$ uname -s
$ mkdir /tmp/bla
$ cd /tmp/bla
$ rm -r /tmp/bla
rm: Cannot remove the current directory /tmp/bla.

Today I did a rm that I expected to fail, but …

$ uname -s
$ mkdir /tmp/bla
$ cd /tmp/bla
$ rm -r /tmp/bla

Wait, did it work?

$ cd /tmp/bla
$ pwd
$ cd /tmp/bla
$ ls -lad /tmp/bla
ls: /tmp/bla: No such file or directory
$ cd /tmp/bla

Somehow I am still there, in /tmp/bla, but /tmp/bla has been removed. What a strange operating system 馃槈

How to quit crontab -e without overwritting cron

Imagine my crontab

* * * * * /usr/bin/date > /tmp/foo

I am writing the date to /tmp/foo every minute
$  cat /tmp/foo
Thu Jul  5 08:45:01 CEST 2012

Now I want to view my crontab in my EDITOR (vi).

$ crontab -e

I do not quit yet.

In the meantime, my colleague modify the crontab.

* * * * * /usr/bin/date > /tmp/bar

Later, I quit vi with :q!

O Surprise, the crontab is * * * * * /usr/bin/date > /tmp/foo again

According to the doc :

When you finish creating entries and exit the file, the crontab command
copies it into the /var/spool/cron/crontabs directory

Even if you did not make change, you overwrite the content of your crontab !

If you want to exit your crontab editor really without overwritting the crontab, you need to kill yourself.


[2] + Stopped (SIGTSTP)        crontab -e
$ kill %2
[2] + Stopped (SIGTTOU)        crontab -e

Thanks to Colin comment, I realized I could not kill with kill, let’s kill with -9

$ kill -9 %2
[2] + Killed                   crontab -e

Check mount option in linux

I did not find a clean way to check the mount option in Linux.

For instance wsize=32768

On AIX, I simply type “mount” and see the mount option…

For some reasons, my Linux does not show me the complete mount options !

$ mount
precision:/nfsserver on /nfsclient type nfs (rw,bg,addr=
$ grep nfsclient /proc/mounts
precision:/nfsserver /nfsclient nfs rw,vers=3,rsize=32768,wsize=32768,hard,proto=tcp,timeo=600,retrans=2,sec=sys,addr=precision 0 0

To lookup my NFS mount was using the mount option specified by Oracle for creating a tablespace that uses NFS, I had to grep in /proc… unreal!

Check if it a program is already running in Unix

There is more than one way to do it, the safe is probably to check if /home/lsc/OH_YES_I_AM_RUNNING exists and believe it. This is called the file.PID method and is widely used (Apache used to use it since a long long time). It needs file. It needs cleanup if you reboot your server in the middle of something (and surely you do not want to delete old pid files yourself)

Ok, often you see this :

ps -ef | grep program

There you list all processes and check the lines that contain program. So some does a vi program or anything worse (emacs?), you will get more rows than needed.

Maybe it is fine to run program with different arguments, this must be decided.

Well, take a simple test case :
x1.sh and x2.sh :

while :
  date  > /dev/null

let’s try to use ps

$ nohup ./x1.sh &
$ nohup ./x2.sh &
$ jobs
[2] +  Running                 nohup ./x2.sh &
[1] -  Running                 nohup ./x1.sh &
$ ps -ef | egrep 'x[12]'
  u22  9240796  6226164  30 14:56:52  pts/2  0:00 /bin/ksh ./x2.sh
  u22 20840608  6226164  31 14:56:48  pts/2  0:01 /bin/ksh ./x1.sh

So fine so good, I see I have one instance of each program.

Let’s try to see if the results are consistent over time :

 $ n=9999;while :
    ps -ef | 
      egrep 'x[12].sh'>f
    if [ $(wc -l <f) != $n ]
      n=$(wc -l <f)
      cat f
      echo "==> $n"

Fri Oct 28 15:01:01 CEST 2011
  u22  9240796  6226164  32 14:56:52  pts/2  0:14 /bin/ksh ./x2.sh
  u22 20840608  6226164  28 14:56:48  pts/2  0:14 /bin/ksh ./x1.sh
==>        2

Fri Oct 28 15:01:08 CEST 2011
  u22  9240796  6226164  50 14:56:52  pts/2  0:14 /bin/ksh ./x2.sh
==>        1

Fri Oct 28 15:01:09 CEST 2011
  u22  9240796  6226164  52 14:56:52  pts/2  0:14 /bin/ksh ./x2.sh
  u22 20840608  6226164  53 14:56:48  pts/2  0:15 /bin/ksh ./x1.sh
==>        2

Fri Oct 28 15:01:17 CEST 2011
  u22  9240796  6226164  40 14:56:52  pts/2  0:15 /bin/ksh ./x2.sh
  u22 10944520  9240796   0 15:01:17  pts/2  0:00 /bin/ksh ./x2.sh
  u22 20840608  6226164  31 14:56:48  pts/2  0:16 /bin/ksh ./x1.sh
==>        3

the fact that a subshell (pid 10944520 ) of x2 appear is not a problem for me. I have much more of a problem at 15:01:08 where x1 disappeared !

Conclusion : you cannot trust ps

shell and list of files

How do you loop thru a list of files?

For instance you want to archive than delete all pdf documents in the current directory :

Bad practice :

tar cvf f.tar *.pdf
rm *.pdf

There are multiple issue with the command above

1) new files could come during the tar, so the rm will delete files that have not been archived

filelist=$(ls *.pdf)
tar cvf f.tar $filelist
rm $filelist

2) if there is no file, tar and rm will return an error

filelist=$(ls|grep '\.pdf')
if [ -n "$filelist" ]
  tar cvf f.tar $filelist
  rm $filelist

3) this will not work for long list (above 100k documents)

filelist=/tmp/filelist.$(date "+%Y%m%d%H%M%S").$$.$RANDOM
ls|grep '\.pdf' > $filelist
if [ -s "$filelist" ]
  tar cvfL f.tar $filelist
  for f in $(<filelist)
    rm $f

As you see, this require special handling. tar for instance use the -L option to accept a list of files, rm could delete files one by one (or in bunches with xargs -L).

This 100’000 limit (the limit may vary for your shell/os) is something that often gets forgotten.

Typical error that could occur are

ksh: no space
bash: Arg list too long

Generate network graph from command line

I recently wrote on gnuplot, today I tried another command line utility to generate graphs, graphviz, version 2.24.0 on AIX5L.

Pretty straightforward syntax :

  echo "digraph Emp {"
    sqlplus -s -L scott/tiger << EOF
      set pages 0 lin 120 hea off feed off
        ename ||'->'|| 
        (select ename from emp where empno=e.mgr) || ';' 
      from emp e where mgr is not null;
  echo "}"
)| neato -Tpng | uuencode Emp.png | mailx laurentschneider@example.com

(or neato -Tpng -o Emp.png to save locally, or -Tps|lp, etc…)

send graph per mail from sqlplus

How to send a graph with a single command from your database to your mail in Unix?

I tried this (gnuplot is available for Solaris, AIX and most Unix derivates) :

echo '
  set hea off pages 0 feed off
  prom set title "salaries of EMP"
  prom unset key
  prom unset xtics
  prom unset xlabel
  prom set term png
  prom plot "-" with circle 
  select row_number() over (order by sal),sal from emp;
  prom e
  prom quit
' | 
  sqlplus -s scott/tiger | 
    gnuplot | 
      uuencode emp.png | 
        mailx email@example.com

The picture is sent as attachment.

If you want to embed your png in an HTML mail, use sendmail

/usr/sbin/sendmail -t <<EOF
To: email@example.com
From: email@example.com
Subject: EMP
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="FILEBOUNDARY"

Content-Type: multipart/alternative; boundary="MSGBOUNDARY"

Content-Type: text/html
Content-Disposition: inline

<html><body>Below a graph...<br/>
<img src="cid:png"/><br/></body></html>
Content-Type: image/png
Content-Disposition: inline; filename="png.png"
Content-Transfer-Encoding: base64
Content-Id: <png>


Sendmail syntax and more tips on : Sending Emails With Sendmail – Part 3

scp tuning

I twitted yesterday :

copying 1TB over ssh sucks. How do you fastcopy in Unix without installing Software and without root privilege?

I got plenty of expert answers. I have not gone to far in recompile ssh and I did not try plain ftp.

Ok, let’s try first to transfer 10 files of 100M from srv001 to srv002 with scp :

time scp 100M* srv002:
100M1    100%   95MB   4.5MB/s   00:21
100M10   100%   95MB   6.4MB/s   00:15
100M2    100%   95MB   6.0MB/s   00:16
100M3    100%   95MB   4.2MB/s   00:23
100M4    100%   95MB   3.4MB/s   00:28
100M5    100%   95MB   4.2MB/s   00:23
100M6    100%   95MB   6.4MB/s   00:15
100M7    100%   95MB   6.8MB/s   00:14
100M8    100%   95MB   6.8MB/s   00:14
100M9    100%   95MB   6.4MB/s   00:15

real    3m4.50s
user    0m27.07s
sys     0m21.56s

more than 3 minutes for 1G.

I got hints about the buffer size, about SFTP, about the cipher algorythm, and about parallelizing. I did not install new software and I have a pretty old openssh client (3.8). Thanks to all my contributors tmuth, Ik_zelf, TanelPoder, fritshoogland, jcnars, aejes, surachart, and the ones the will answer after the writting of this blog post…

Ok, let’s try a faster algorythm, with sftp (instead of scp), a higher buffer and in parallel

$ cat batch.ksh
echo "progress\nput 100M1" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M2" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M3" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M4" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M5" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M6" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M7" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M8" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M9" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
echo "progress\nput 100M10" | sftp -B 260000 -o Ciphers=arcfour -R 512 srv002&
$ time batch.ksh
real    0m19.07s
user    0m12.08s
sys     0m5.86s

This is a 1000% speed enhancement 馃檪

What is the current setting of NLS_LANG in sqlplus?

I just learnt a neat trick from Oracle Support.

How do you see the current value of NLS_LANG in SQLPLUS ?

HOST is not the right answer.


SQL> host echo $NLS_LANG



The correct setting is revealed by @.[%NLS_LANG%]

SP2-0310: unable to open file ".[AMERICAN_AMERICA.WE8ISO8859P1]"


SQL>  @.[%NLS_LANG%]
SP2-0310: unable to open file ".[AMERICAN_AMERICA.WE8ISO8859P1]"

It could well be that both return the same answer, but not necessarly, as shown above.

The unix discrepancy is related to the subshell created by HOST. The subshell may read some .profile and overwrite the value of NLS_LANG

In Windows, the NLS_LANG setting may be set by sqlplus according to some registry entries

Time offset in Unix

What is the time offset of the current date in Unix?

perl -e '
print ($d."\n")'

Am I in summer (DST)?

perl -e 'if((localtime)[8]){print"yes"}else{print "no"}'

last access time of a file

I was reading http://blogs.oracle.com/myoraclediary and there was a command about printing the modification details of a file.

In Linux / Cygwin “stat” exists as a command

$ stat /etc/hosts
Access: 2010-08-25 15:20:49.782522200 +0200
Modify: 2010-08-18 14:04:25.868114200 +0200
Change: 2010-08-18 14:04:26.072413100 +0200

Or use the one-liner perl below

### st_atime;         /* Time of last access */
$ perl -e 'use POSIX;[-f"/etc/hosts"]&&print ctime((stat(_))[8])'
Wed Aug 25 15:20:08 2010
### st_mtime;         /* Time of last data modification */
$ perl -e 'use POSIX;[-f"/etc/hosts"]&&print ctime((stat(_))[9])'
Wed Jun 10 11:36:40 2009
### st_ctime;         /* Time of last file status change */
$ perl -e 'use POSIX;[-f"/etc/hosts"]&&print ctime((stat(_))[10])' 
Wed Aug 25 01:00:07 2010

to cvs or to subversion

First surprise, after migration, the size of my subversion folder is double the size of my cvs folder. With a bunch of 2Gb disks shared amoung dozens of unix persons, and regular reminders the current usage reached 100%, you will feel the pain of having each developers doublesizing its home directory…
The reason is a .svn/test-base directory containing a duplicate of your local copy.
The benefit there is that it reduces network usage, for instance when diffing with the working version.

Second surprise, not sure if it is a generic issue or related to the one above, a full checkout take ages 馃檨 .

svn and cvs command line interfaces almost have the same options, a bit like vi and vim -vi is my favorite- but the “tagging” massively differs. The is no cvs tag command in subversion, you use a different syntax.
cvs :

$ cd /home/lsc/CVS/proj1/source/pkg
$ cvs tag MYTAG helloworld.pkb helloworld.pks

now enjoy the power of subversion! [updated as I found the –parents option]

$ cd /home/lsc/SVN/proj1/source/pkg
$ svn copy --parents helloworld.pkb http://myserver.domain.com/svn/REPOS1/proj1/tags/MYTAG/proj1/source/pkg -m "my first svn tag"
$ svn copy --parents helloworld.pks http://myserver.domain.com/svn/REPOS1/proj1/tags/MYTAG/proj1/source/pkg -m "my first svn tag"

I always loved the cvs rename command

$ cvs rename pkg1.pkb pkg2.pkb
cvs [rename aborted]: Remote server does not support rename

Joke apart, in CVS I used to physically logon to the server and manually move the pkg1.pkb,v to pkg2.pkb,v
Renaming works in subversion :

$ svn move pkg1.pkb pkg2.pkb
A         pkg2.pkb
D         pkg1.pkb

Update: One more annoyance, you cannot checkout a single file 馃檨
subversion faq

extract xml from the command line

I just discovered this morning this cool utility in my /bin directory : xmllint

You can use it to extract values from your xml files within your shell scripts

$ cat foo.xml
  <emp no="1">
  <emp no="2">
$ echo 'cat //emplist/emp[@no="1"]/ename/text()'|
  xmllint --shell foo.xml |
  sed -n 3p

I like this !

read without Enter

A small unix tip today.

Do you want to continue ?

If you are expecting “y” or “n” but do not want to enforce the user to type y[Enter] but simply y, you can use the -n option in bash.

Within a ksh script:

yorn=$(bash -c 'read -p "Do you want to continue ? " -n 1 ans;echo "$ans"')

Within bash:

read -n 1 yorn


Do you know cd ? I thought I did until this afternoon …

OK, let’s start some basic.

I create two directories

$ echo $SHELL
$ mkdir /tmp/foo
$ mkdir /tmp/bar

create a symlink /tmp/bar/baz pointing to /tmp/foo

$ ln -s /tmp/foo /tmp/bar/baz

create a file foo1 in foo

$ touch /tmp/foo/foo1

change to the symbolic link baz

$ cd /tmp/bar/baz

Ok, so far so good. Let’s check what is in ../foo

$ ls ../foo

From the symbolic baz, .. point to /tmp/foo. This is because ls and most command line utilities use the physical path.

To print the Logical [default] and Physical working directories, use pwd -L and pwd -P

$ pwd -L
$ pwd -P

to change directory relatively to the logical path, use cd -L … (default), for physical, use cd -P … !

$ pwd -L
$ cd -L ../foo
ksh: ../foo:  not found

Obviously /tmp/bar/foo does not exist

$ pwd -L
$ pwd -P
$ cd -P ../foo
$ pwd

Obviously /tmp/foo/../foo is /tmp/foo

So far so good, some of you may know that already.

Let’s bring some devil element in play

$ bash

Arghh!!! 脭么么么么么么么么么么么么么 r芒ge, 脭么么么么么么么么么么么么么 d茅sespoir, I switched to a non-working shell!

$ cd /tmp/bar/baz
$ cd -L ../foo
$ pwd -L

Even if I switched to a not working directory, bash cd -L weirdly decided to switch to the physical path instead of the logical path.

Let’s retry

$ cd /tmp/bar/baz
$ mkdir /tmp/bar/foo
$ cd -L ../foo
$ pwd

This time bash cd -L changed to the logical path. So if you use bash and cd, you cannot possibly know where you are landing without checking first if the directory exist !

BTW, I just discovered Digger HD , unrelated to this post of course …


Something I like in ksh is to change from /my/old/directory/path to /my/new/directory/path by typing cd old new. This does not work in bash

So I had to find a workaround 馃槈

$ ksh
$ cd /app/oracle/product/
$ cd 6 7
$ pwd 
$ bash
$ cd 7 6
bash: cd: 7: No such file or directory
$ c() { cd ${PWD/$1/$2}; }
$ c 7 6
$ pwd

chmod -R 777 .

This is one of the thing I hate to see, recursively changing everything to 777 馃懣

If you want to give read access to all, then 644 is enough for files and 755 for directories. If you want to give execution permission too, you could give 755 to executable files.

Also sometimes you have files with special permission like suid or guid bit (to run as the owner instead of to run as the run user), so it would be best to use relative (go+r) to make the file readable to group and others.

Therefore I prefer relative change. Still there is one thing I do not want, is making every file executable…

Ok, here it is

chmod -R a+rX .

note the big X 馃檪 we change recursively all files and dirs to be readable, and we set the executable flag ONLY IF the file is executable for the owner !

to ftp or to sftp

Ftp is seen as an old-time unsecure protocol. Many shops nowadays have switched or are switching to sftp. I will try to point out some differences :

Compatibility: none. the protocol is completly different. Multiple graphical clients however do support both mode. But the basic “ftp” client will not work with sftp.

Ascii mode: only in ftp. In sftp, it is always binary so there will be no conversion. Also no blocksize, recordlength or primary/secondary space for your OS/390 connections.

Interactive mode: similar. you enter your username and password, do cd, put and get. But to quit, by will not work in sftp 馃槈 Use quit or exit instead

Batch mode: different. Most probably you will end up setting a private/public key infrastructure for your ssh connection and use scp (secure copy). If you are using a ssh client like putty, it is possible to do something like pscp -l user -pw password server:file .

Security: sftp is secure, ftp is not.

Speed: ftp is fast, sftp is slow 馃檨 !

Oh NOOOOOOO!!!!! What’s the point is bringing something new if it is slower !!!

Ok, let’s try to download a 100m file:

$ time (echo "open dbsrv01
user oracle secret
get 100m"|ftp -n )

real    0m24.673s
user    0m0.030s
sys     0m0.016s
$ time scp -q oracle@dbsrv01:100m .

real    1m46.978s
user    0m0.108s
sys     0m0.202s

it is about 4x slower! Is there anything we could do about it?

Well, maybe :

$ time scp -q -o Compression=yes oracle@dbsrv01:100m .

real    0m18.634s
user    0m0.748s
sys     0m0.452s

ssh/scp/sftp have a compression mode. If you are transferring your large files across a slow network, this may be an interesting option to consider !

OpenSSH homepage : http://www.openssh.org

return code before grep

In my previous post hide-password-from-ps-output-sql-loader I mentioned a way to pass the password to the loader thru a parameter file. As correctly suggested by Brian Tkatch, the password could be passed as standard input

sqlldr control=x.ctl silent=header,feedback <<EOF

The Username: prompt is displayed 馃檨   馃檨

How do we get rid of this ?

sqlldr control=x.ctl silent=header,feedback <<EOF | grep -v "^Username:"

There is no output. But what’s the error code

echo $?

The return code is 1 馃檨

This is not the error code from sqlldr, but the error code from grep !

Ok, here is the trick, a bit cryptic if you are not familiar with file descriptors

( ( (sqlldr control=x <<EOF;echo $? >&3) |grep -v "^Username:" >&4 ) 3>&1 |(read x;exit $x) )4>&1
echo $?

The return code is 0 馃檪

To bash or not to bash

I have been inspired by Chen to talk about bash…

I have been using ksh for many years, and I mean ksh88 not ksh93. The main reason is, I want my script to run the same way in any Unix flavor.

ksh93 has never been too much popular. I used it a few time to sleep half a second

echo sleep 0.5| /usr/dt/bin/dtksh

ksh has a lot of nice features. I just used one of them in my script :

$ typeset -u name
$ read name?"Enter your name : "
Enter your name : Laurent
$ echo $name

Way easier to force a variable to be uppercase rather than using echo|tr etc

Bash has some nice features too, but unfortunately every OS release come with a different bash version, which is the same pain as perl when you want to write a script that last for a decade or two.

Ok, just4fun

$ mkdir -p {a..z}/{1..9}
... create directories a/1 a/2 ... z/8 z /9
$ [[ text =~ t..t ]] 
... check if text matches regular expression t..t
$ echo ${text/pattern/string}
... replace pattern by string

The first two commands require bash3, the last is just fine with bash2.

Have fun shell-scripting 馃檪

echo does not accept end of arguments operator

Let’s start with an example :

$ cat AI
while :
  echo "What's your name ?"
  read a
  if [ ! $a ]
  echo "Your name is :"
  echo $a
echo "Bye"

$ ./AI
What's your name ?
Your name is :

What's your name ?
Your name is :

What's your name ?
Your name is :

What's your name ?


This artificial intelligence is not very intelligent, it cannot recognize me if I am called “-e” (it is Friday, have a look at Little Bobby Tables ).

Most unix tools consider
-- signals the end of options and disables further option processing

But not echo 馃檨

$ touch -e
touch: invalid option -- e
Try `touch --help' for more information.
$ touch -- -e
$ ls -l -e
ls: invalid option -- e
Try `ls --help' for more information.
$ ls -l -- -e
-rw-r--r-- 1 lsc dba 0 Oct 31 15:44 -e
$ rm -e
rm: invalid option -- e
Try `rm ./-e' to remove the file `-e'.
Try `rm --help' for more information.
$ rm -- -e
$ echo -e

$ echo -- -e
-- -e

So, what’s the solution? well, probably not using “echo”, for example printf

$ (echo “What’s your name ?”
read a
echo “Your name is :”
printf “%s\n” “$a”)
What’s your name ?
Your name is :

updatedb does not work with samba drives and cygwin

updatedb and locate are parts of the findutils package, which exists since a long time on most unix / linux flavors.

Something I never get worked is to search for files on network drives with Cygwin, I always got :

find: /cygdrive/y/Favorites/Links/del changed during execution of find (old inode number -474324984, new inode number -44545478
4, filesystem type is system) [ref 1114]
find: /cygdrive/y/Favorites/Links/del changed during execution of find (old inode number -513303800, new inode number -47432498
4, filesystem type is system) [ref 1114]

which is because the file system is mounted and unmounted on demand and get new inodes.

So I wrote a hack in /usr/bin/updatedb

# lschnei2
for d in c: y:
cd $d
/cygdrive/c/OS/system32/attrib /s
) |
sed ‘s,………..\(.\):,/cygdrive/\1,;y,\\,/,’ |
tr ‘\r’ ‘\0’ | tr -d ‘\n’

Then I can search efficiently

$ time updatedb

real 0m18.273s
user 0m3.806s
sys 0m7.332s

$ time locate hosts

real 0m0.261s
user 0m0.265s
sys 0m0.015s

How to cron?

RTFM is not the best answer …

man crontab
SunOS 5.10          Last change: 10 Nov 2005

User Commands                                          crontab(1)
     A crontab file consists of lines of  six  fields  each.  The
     fields  are  separated by spaces or tabs. The first five are
     integer patterns that specify the following:

     minute (0-59),
     hour (0-23),
     day of the month (1-31),
     month of the year (1-12),
     day of the week (0-6 with 0=Sunday).

so far so good. But read this

     Each of these patterns can be either  an  asterisk  (meaning
     all legal values) or a list of elements separated by commas.
     An element is either a number or two numbers separated by  a
     minus sign (meaning an inclusive range). Time specified here
     is interpreted in the timezone of the cron(1M) daemon, which
     is  set system-wide in /etc/default/init. Entries do not use
     the invoking user's timezone. The specification of days  can
     be  made  by  two  fields  (day  of the month and day of the
     week). Both are adhered to if specified as a  list  of  ele-
     ments. See .

See .

Let’s imagine you want to run a job today, Thursday May 15, 2008 at 2:15pm
You write:
15 14 15 05 4 /tmp/run-my-job

The job will be run Today at 2:15, next year Friday May 15, 2009 at 2:15pm and next week Thursday May 22, 2008 at 2:15pm…

The correct method is to use
15 14 15 05 * /tmp/run-my-job

Specifying both day of week and day of month is not what I expected 馃槇