Bug #10809 “Race condition when writing ~/.bash_history” : Bugs : bash package : Ubuntu

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2004-11-30:

#1

Automatically imported from Debian bug report #283702 http://bugs.debian.org/283702

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2004-11-30:

#2

Message-Id: <email address hidden>
Date: Tue, 30 Nov 2004 21:25:46 +0300
From: Cyril Bouthors <email address hidden>
To: Debian Bug Tracking System <email address hidden>
Subject: Race condition when writing ~/.bash_history

Package: bash
Version: 3.0-10
Severity: critical
Justification: causes serious data loss

Hi,

About once a month, my ~/.bash_history gets empty. This is quite
annoying because I'm using C-r really often.

I get the same problem again and again for quite some time now, maybe
two years.

I think it comes from the fact that I'm not using a typical setup.
I'm running 5 different "bash --noediting -i" under Emacs with
shell-mode buffers and I'm often exiting Emacs without closing all the
shells by hand. The 5 bash processes get killed and I guess they try
to write ~/.bash_history at the exact same time.

I think it's a basic race condition, bash should lock the file before
trying to write it but I haven't checked if it does.

I haven't reproduced the exact same bug where ~/.bash_history gets
empty but a similar unwanted behaviour can be reproduced by doing
massive parallel "history -w" with:

cyb@wide:~$ export HISTFILE=/tmp/history
cyb@wide:~$ export HISTSIZE=100000
cyb@wide:~$ rm -f $HISTFILE
cyb@wide:~$ for a in `seq 5`; do xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i' ; done
cyb@wide:~$ wc -l $HISTFILE
10000 /tmp/history
cyb@wide:~$ rm $HISTFILE
cyb@wide:~$ for a in `seq 5`; do xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i' & done
[1] 4715
[2] 4721
[3] 4722
[4] 4723
[5] 4724
cyb@wide:~$ wait
[1] Done xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[2] Done xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[4]- Done xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[5]+ Done xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[3]+ Done xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
cyb@wide:~$ wc -l $HISTFILE
2794 /tmp/history
cyb@wide:~$ unset HISTFILE
cyb@wide:~$

/tmp/history should be 10k lines long in both cases (5*1000*2).

Regards

-- System Information:
Debian Release: 3.1
APT prefers unstable
APT policy: (500, 'unstable')
Architecture: i386 (i686)
Kernel: Linux 2.6.9-1-686
Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968)

Versions of packages bash depends on:
ii base-files 3.1.0 Debian base system miscellaneous f
ii libc6 2.3.2.ds1-18 GNU C Library: Shared libraries an
ii libncurses5 5.4-4 Shared libraries for terminal hand
ii passwd 1:4.0.3-30.4 Change and administer password and

-- no debconf information

Message-Id: <E1CZCh8-0001FQ-PB@wide.bouthors.org>
Date: Tue, 30 Nov 2004 21:25:46 +0300
From: Cyril Bouthors <cyb@debian.org>
To: Debian Bug Tracking System <submit@bugs.debian.org>
Subject: Race condition when writing ~/.bash_history

Package: bash
Version: 3.0-10
Severity: critical
Justification: causes serious data loss

Hi,

About once a month, my ~/.bash_history gets empty.  This is quite
annoying because I'm using C-r really often.

I get the same problem again and again for quite some time now, maybe
two years.

I think it comes from the fact that I'm not using a typical setup.
I'm running 5 different "bash --noediting -i" under Emacs with
shell-mode buffers and I'm often exiting Emacs without closing all the
shells by hand.  The 5 bash processes get killed and I guess they try
to write ~/.bash_history at the exact same time.

I think it's a basic race condition, bash should lock the file before
trying to write it but I haven't checked if it does.

I haven't reproduced the exact same bug where ~/.bash_history gets
empty but a similar unwanted behaviour can be reproduced by doing
massive parallel "history -w" with:

cyb@wide:~$ export HISTFILE=/tmp/history
cyb@wide:~$ export HISTSIZE=100000
cyb@wide:~$ rm -f $HISTFILE
cyb@wide:~$ for a in `seq 5`; do xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i' ; done
cyb@wide:~$ wc -l $HISTFILE
10000 /tmp/history
cyb@wide:~$ rm $HISTFILE
cyb@wide:~$ for a in `seq 5`; do xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i' & done
[1] 4715
[2] 4721
[3] 4722
[4] 4723
[5] 4724
cyb@wide:~$ wait
[1]   Done                    xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[2]   Done                    xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[4]-  Done                    xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[5]+  Done                    xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
[3]+  Done                    xterm -e 'for a in `seq 1000`; do echo "echo $a; history -w"; done | bash -i'
cyb@wide:~$ wc -l $HISTFILE
2794 /tmp/history
cyb@wide:~$ unset HISTFILE
cyb@wide:~$

/tmp/history should be 10k lines long in both cases (5*1000*2).

Regards

-- System Information:
Debian Release: 3.1
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: i386 (i686)
Kernel: Linux 2.6.9-1-686
Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968)

Versions of packages bash depends on:
ii  base-files                  3.1.0        Debian base system miscellaneous f
ii  libc6                       2.3.2.ds1-18 GNU C Library: Shared libraries an
ii  libncurses5                 5.4-4        Shared libraries for terminal hand
ii  passwd                      1:4.0.3-30.4 Change and administer password and

-- no debconf information

Revision history for this message

In Debian Bug tracker #283702, Peter Walser (peterwalser) wrote on 2004-12-01: bash-history not thread save?

#3

As far as I understand uses bash/history the function readline.

with bash 2.05b.0 :
[... steps described above ...]
$ wc -l $HISTFILE
2805 /tmp/history

So I get quite the same behavior.

I think readline/history is not thread save, but it's not critical.

Regards

Peter

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2004-12-01:

#4

Message-Id: <email address hidden>
Date: Wed, 1 Dec 2004 16:52:11 +0100
From: Peter Walser <email address hidden>
To: <email address hidden>
Subject: bash-history not thread save?

As far as I understand uses bash/history the function readline.

with bash 2.05b.0 :
[... steps described above ...]
$ wc -l $HISTFILE
2805 /tmp/history

So I get quite the same behavior.

I think readline/history is not thread save, but it's not critical.

Regards

Peter

Revision history for this message

In Debian Bug tracker #283702, Matthias Klose (doko-cs) wrote on 2004-12-01: Re: Bug#283702: bash-history not thread save?

#5

severity 283702 important
thanks

Peter Walser writes:
>
> As far as I understand uses bash/history the function readline.
>
> with bash 2.05b.0 :
> [... steps described above ...]
> $ wc -l $HISTFILE
> 2805 /tmp/history
>
> So I get quite the same behavior.
>
> I think readline/history is not thread save, but it's not critical.

agreed.

Revision history for this message

Matthias Klose (doko) wrote on 2004-12-01:

#6

acknowledged, although not RC.

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2004-12-02:

#7

Message-ID: <email address hidden>
Date: Thu, 2 Dec 2004 00:18:54 +0100
From: Matthias Klose <email address hidden>
To: <email address hidden>
Cc: Peter Walser <email address hidden>
Subject: Re: Bug#283702: bash-history not thread save?

severity 283702 important
thanks

Peter Walser writes:
>
> As far as I understand uses bash/history the function readline.
>
> with bash 2.05b.0 :
> [... steps described above ...]
> $ wc -l $HISTFILE
> 2805 /tmp/history
>
> So I get quite the same behavior.
>
> I think readline/history is not thread save, but it's not critical.

agreed.

Revision history for this message

In Debian Bug tracker #283702, Cyril Bouthors (cyril) wrote on 2004-12-09: Forward to upstream?

#8

This is a bug in upstream, maybe we should forward this to upstream
authors, they'll probably have a good reactivity.
--
Cyril Bouthors

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2004-12-09:

#9

Message-ID: <email address hidden>
Date: Thu, 09 Dec 2004 19:17:38 +0300
From: Cyril Bouthors <email address hidden>
To: <email address hidden>
Subject: Forward to upstream?

--=-=-=

This is a bug in upstream, maybe we should forward this to upstream
authors, they'll probably have a good reactivity.
--
Cyril Bouthors

--=-=-=
Content-Type: application/pgp-signature

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQBBuHqiZ1SJHeqsYt8RAkYmAJ0duXAETm8sCoX4NpxUdxPEgIy69ACfdgB8
8AouwFS84rHVomqwbbn5uJk=
=/DAi
-----END PGP SIGNATURE-----
--=-=-=--

Revision history for this message

In Debian Bug tracker #283702, Chet Ramey (chet-ramey) wrote on 2005-09-10: Re: Bug#283702: Race condition when writing ~/.bash_history (forwarded from Cyril Bouthors)

#10

Matthias Klose wrote:
> [severity was lowerd in the Debian BTS]

I'm not inclined to do anything about this.

Chet
--
``The lyf so short, the craft so long to lerne.'' - Chaucer
( ``Discere est Dolere'' -- chet )
Live Strong.
Chet Ramey, ITS, CWRU <email address hidden> http://cnswww.cns.cwru.edu/~chet/

Revision history for this message

In Debian Bug tracker #283702, Matthias Klose (doko-cs) wrote on 2005-09-10:

#11

severity 283702 minor
tags 283702 + wontfix
forwarded 283702 <email address hidden>
thanks

Chet Ramey writes:
>
> I'm not inclined to do anything about this.
>
> Chet
> --
> ``The lyf so short, the craft so long to lerne.'' - Chaucer
> ( ``Discere est Dolere'' -- chet )
> Live Strong.
> Chet Ramey, ITS, CWRU <email address hidden> http://cnswww.cns.cwru.edu/~chet/

Revision history for this message

In Debian Bug tracker #283702, Matthias Klose (doko-cs) wrote on 2005-09-10: bash reports

#12

retitle 316407 [fixed in 3.1] "cd -P" hangs bash on some configurations
forwarded 283702 <email address hidden>
tags 283702 + upstream

forwarded 288319 <email address hidden>
tags 288319 + upstream

forwarded 310598 <email address hidden>
tags 310598 + upstream

retitle 310308 [fixed in 3.1] bash: bash closes fd 2 after an unsuccessful exec
tags 310308 + upstream
tags 310308 + fixed-upstream

tags 313538 + pending

retitle 238226 [fixed in 3.1] bash: segfaults on ia64 when expanding large glob
tags 238226 + upstream
tags 238226 + fixed-upstream

retitle 309654 [fixed in 3.1] Trouble editing with multibyte chars
tags 309654 + upstream
tags 309654 + fixed-upstream

retitle 317324 [fixed in 3.1] bash: double free
tags 317324 + upstream
tags 317324 + fixed-upstream

retitle 310308 [fixed in 3.1] bash closes fd 2 after an unsuccessful exec
tags 310308 + upstream
tags 310308 + fixed-upstream

retitle 297330 [fixed in 3.1] bash3: <Esc>-/ garbages colored prompt in vi-mode
tags 297330 + upstream
tags 297330 + fixed-upstream

retitle 316206 [fixed in 3.1] bash segfaults if execve'd with NULL argv
tags 316206 + upstream
tags 316206 + fixed-upstream

retitle 327035 [fixed in 3.1] bash: manpage macro problem?
tags 327035 + upstream
tags 327035 + fixed-upstream

forwarded 292751 <email address hidden>
tags 292751 + upstream

forwarded 294078 <email address hidden>
tags 294078 + upstream

forwarded 296099 <email address hidden>
tags 296099 + upstream

forwarded 298889 <email address hidden>
tags 298889 + upstream

forwarded #303113 <email address hidden>
tags #303113 + upstream

forwarded 305485 <email address hidden>
tags 305485 + upstream

forwarded 327414 <email address hidden>
tags 327414 + upstream

forwarded 314363 <email address hidden>
tags 314363 + upstream

forwarded 327274 <email address hidden>
tags 327274 + upstream

forwarded 293096 <email address hidden>
tags 293096 + upstream

thanks

Revision history for this message

Matthias Klose (doko) wrote on 2005-09-12:

#13

> I'm not inclined to do anything about this.
>
> Chet

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2005-09-22:

#14

Message-ID: <email address hidden>
Date: Fri, 09 Sep 2005 20:55:40 -0400
From: Chet Ramey <email address hidden>
To: Cyril Bouthors <email address hidden>, <email address hidden>
Subject: Re: Bug#283702: Race condition when writing ~/.bash_history (forwarded from Cyril Bouthors)

Matthias Klose wrote:
> [severity was lowerd in the Debian BTS]

I'm not inclined to do anything about this.

Chet
--
``The lyf so short, the craft so long to lerne.'' - Chaucer
( ``Discere est Dolere'' -- chet )
Live Strong.
Chet Ramey, ITS, CWRU <email address hidden> http://cnswww.cns.cwru.edu/~chet/

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2005-09-22:

#15

Message-ID: <email address hidden>
Date: Sat, 10 Sep 2005 08:04:19 +0200
From: Matthias Klose <email address hidden>
To: <email address hidden>
Subject: Re: Bug#283702: Race condition when writing ~/.bash_history (forwarded from Cyril Bouthors)

severity 283702 minor
tags 283702 + wontfix
forwarded 283702 <email address hidden>
thanks

Chet Ramey writes:
>
> I'm not inclined to do anything about this.
>
> Chet
> --
> ``The lyf so short, the craft so long to lerne.'' - Chaucer
> ( ``Discere est Dolere'' -- chet )
> Live Strong.
> Chet Ramey, ITS, CWRU <email address hidden> http://cnswww.cns.cwru.edu/~chet/

Revision history for this message

Debian Bug Importer (debzilla) wrote on 2005-09-22:

#16

Message-ID: <email address hidden>
Date: Sat, 10 Sep 2005 14:30:24 +0200
From: Matthias Klose <email address hidden>
To: <email address hidden>
Subject: bash reports