Debian Bug report logs - #563726
tar: futimens() with a bad file descriptor (AT_FDCWD)

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Guillaume Ayoub <guillaume.ayoub@kozea.fr>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: libc6-2.10.2-3 may break utime on folders

Date: Mon, 4 Jan 2010 22:28:06 +0100

Package: libc6
Version: 2.10.2-3
Severity: important


Hello,

Just after upgrading libc6 from 2.10.2-2 to 2.10.2-3, utime calls fail for
folders (not for regular files). This breaks main programs such as tar, dpkg,
touch and postfix.

Here are 2 examples:

$ ls -l
drwxr-xr-x 2 user users 4096 jan.  4 17:22 test_directory
-rw-r--r-- 1 user users    0 jan.  4 17:22 test_file
$touch test_file
$touch test_directory
touch: setting times of `test_directory': Bad file descriptor

$aptitude upgrade
(...)
tar: ./postrm: Cannot utime: Bad file descriptor
tar: ./postinst: Cannot utime: Bad file descriptor
tar: ./md5sums: Cannot utime: Bad file descriptor
tar: ./shlibs: Cannot utime: Bad file descriptor
tar: ./control: Cannot utime: Bad file descriptor
tar: ./symbols: Cannot utime: Bad file descriptor
tar: .: Cannot utime: Bad file descriptor
tar: Exiting with failure status due to previous errors
dpkg-deb: subprocess tar returned error exit status 2

Downgrading /lib/libc-2.10.2.so to version 2.10.2-2 fixes the bug.

I have an ext3 partition with a lot of free space. Note that I have an old
kernel (2.6.21.1).


-- System Information:
Debian Release: squeeze/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: i386 (i686)

Kernel: Linux 2.6.21.1dedibox-r7
Locale: LANG=fr_FR.UTF-8, LC_CTYPE=fr_FR.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Versions of packages libc6 depends on:
ii  libc-bin                      2.10.2-3   Embedded GNU C Library: Binaries
ii  libgcc1                       1:4.4.2-8  GCC support library

Versions of packages libc6 recommends:
pn  libc6-i686                    <none>     (no description available)

Versions of packages libc6 suggests:
ii  debconf [debconf-2.0]         1.5.28     Debian configuration management sy
pn  glibc-doc                     <none>     (no description available)
ii  locales                       2.10.2-3   Embedded GNU C Library: National L

-- debconf information excluded

Message #10 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Guillaume Ayoub <guillaume.ayoub@kozea.fr>

To: 563726@bugs.debian.org

Subject: More information

Date: Mon, 4 Jan 2010 23:49:09 +0100

More information can be found at:
http://www.mail-archive.com/debian-user@lists.debian.org/msg562576.html

As Sven says, the bug is caused by debian/patches/any/cvs-futimens.diff,
reporting the 'bad file descriptor' error. Without this patch, futimens()
failed silently, as utimensat() has been added in kernel 2.6.22.

Useful links:
- eglibc 2.10 futimens.c original code:
  http://www.eglibc.org/cgi-bin/viewcvs.cgi/branches/eglibc-2_10/libc/sysdeps/unix/sysv/linux/futimens.c?view=markup
- upstream patch in eglibc trunk backported in Debian:
  http://www.eglibc.org/cgi-bin/viewcvs.cgi/trunk/libc/sysdeps/unix/sysv/linux/futimens.c?r1=text&tr1=2244&r2=text&tr2=9570&diff_format=h

Message #15 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Guillaume Ayoub <guillaume.ayoub@kozea.fr>, 563726@bugs.debian.org

Subject: Re: Bug#563726: More information

Date: Tue, 5 Jan 2010 00:31:34 +0100

On Mon, Jan 04, 2010 at 11:49:09PM +0100, Guillaume Ayoub wrote:
> More information can be found at:
> http://www.mail-archive.com/debian-user@lists.debian.org/msg562576.html
> 
> As Sven says, the bug is caused by debian/patches/any/cvs-futimens.diff,
> reporting the 'bad file descriptor' error. Without this patch, futimens()
> failed silently, as utimensat() has been added in kernel 2.6.22.

The patch is actually correct, it makes futimens() conform to POSIX [1],
and reject bad descriptors. The fact that the kernel has utimensat() or 
not should not change anything here. I'll investigate that more in
details tomorrow.

[1] http://www.opengroup.org/onlinepubs/9699919799/functions/utimensat.html

-- 
Aurelien Jarno	                        GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Message #22 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Guillaume Ayoub <guillaume.ayoub@kozea.fr>

To: 563726@bugs.debian.org

Subject: More information

Date: Mon, 4 Jan 2010 23:33:32 +0100

More information can be found at:
http://www.mail-archive.com/debian-user@lists.debian.org/msg562576.html

As Sven says, the bug is caused by debian/patches/any/cvs-futimens.diff,
reporting the 'bad file descriptor' error. Without this patch, futimens()
failed silently, as utimensat() has been added in kernel 2.6.22.

Useful links:
- eglibc 2.10 futimens.c original code:
  http://www.eglibc.org/cgi-bin/viewcvs.cgi/branches/eglibc-2_10/libc/sysdeps/unix/sysv/linux/futimens.c?view=markup
- upstream patch in eglibc trunk backported in Debian:
  http://www.eglibc.org/cgi-bin/viewcvs.cgi/trunk/libc/sysdeps/unix/sysv/linux/futimens.c?r1=text&tr1=2244&r2=text&tr2=9570&diff_format=h

Message #29 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Christian Marillat <marillat@debian.org>

To: 563726@bugs.debian.org

Subject: libc6-2.10.2-3 may break utime on folders

Date: Tue, 05 Jan 2010 08:41:26 +0100

Hi,

I'm also able to reproduce this bug on a Sparc machine and a 2.6.18
kernel (latest kernel doesn't boot).

Christian

Message #66 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: 563726@bugs.debian.org, Bdale Garbee <bdale@gag.com>

Cc: control@bugs.debian.org

Subject: Patch

Date: Tue, 5 Jan 2010 12:41:33 +0100

tag 563726 + patch
thanks

This patch makes sure that futimens is only called with a valid file
descriptor. It may looks a lot of changes, but they are only due to
indentation. The only added line is the "if (fd >= 0)", which makes sure
that futimens() is not called with AT_FDCWD as the first argument. In
that case POSIX 2008 mandates that -1 is returned with errno set to 
EBADF.

--- tar-1.22.orig/lib/utimens.c
+++ tar-1.22/lib/utimens.c
@@ -120,16 +120,17 @@
     }
 #endif
 #if HAVE_FUTIMENS
-  {
-    int result = futimens (fd, timespec);
+  if (fd >= 0)
+    {
+      int result = futimens (fd, timespec);
 # ifdef __linux__
-    /* Work around the same bug as above.  */
-    if (0 < result)
-      errno = ENOSYS;
+      /* Work around the same bug as above.  */
+      if (0 < result)
+        errno = ENOSYS;
 # endif
-    if (result == 0 || errno != ENOSYS)
-      return result;
-  }
+      if (result == 0 || errno != ENOSYS)
+        return result;
+    }
 #endif
 
   /* The platform lacks an interface to set file timestamps with

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Message #73 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Guillaume Ayoub <guillaume.ayoub@kozea.fr>

To: 563726@bugs.debian.org

Cc: Bdale Garbee <bdale@gag.com>, Aurelien Jarno <aurelien@aurel32.net>

Subject: Patch test

Date: Tue, 5 Jan 2010 16:15:48 +0100

I tested the patch on my machine.

- applied on tar, it fixed tar;
- applied on coreutils, it fixed touch.

Everything seems to work now (postfix, apt-get, dpkg, etc.), but other
softwares calling futimens() may be affected.

Thanks a lot for your work,
-- 
Guillaume Ayoub

Message #83 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Guillaume Ayoub <guillaume.ayoub@kozea.fr>

Cc: 563726@bugs.debian.org, Bdale Garbee <bdale@gag.com>

Subject: Re: Patch test

Date: Tue, 5 Jan 2010 16:36:22 +0100

On Tue, Jan 05, 2010 at 04:15:48PM +0100, Guillaume Ayoub wrote:
> I tested the patch on my machine.
> 
> - applied on tar, it fixed tar;
> - applied on coreutils, it fixed touch.
> 
> Everything seems to work now (postfix, apt-get, dpkg, etc.), but other
> softwares calling futimens() may be affected.
> 

This is a bug, and they will need to be fixed too. Even with the current
code, no error is returned, but it still means a file that should have
had the timestamp update actually does not have it.

-- 
Aurelien Jarno	                        GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Message #88 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Jim Meyering <jim@meyering.net>

To: Ondřej Vašík <ondrej.vasik@gmail.com>

Cc: 563726@bugs.debian.org, bug-gnulib@gnu.org

Subject: Re: [Bug 552320] Handling AT_FDCWD as EBADF in glibc's utimens() causes tar failures

Date: Tue, 05 Jan 2010 22:07:34 +0100

bugzilla@redhat.com wrote:
> https://bugzilla.redhat.com/show_bug.cgi?id=552320
>
> Ondrej Vasik <ovasik@redhat.com> changed:
>
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jim@meyering.net
>             Summary|new tar is unable to open   |Handling AT_FDCWD as EBADF
>                    |archives created with old   |in glibc's utimens() causes
>                    |tar                         |tar failures
>
> --- Comment #5 from Ondrej Vasik <ovasik@redhat.com>  2010-01-05 14:29:13 EDT ---
> Ah - even the latest gnulib doesn't handle this new recent change - so touch
> from coreutils might be affected as well as mentioned in debian bugzilla from
> comment #4. Therefore adding coreutils/gnulib upstream maintainer into CC.
>
> Jim, what do you think? Is the patch from
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=563726#66 good way to go and
> should be applied to gnulib/coreutils-8.3 to prevent touch (or better said
> futimens() with AT_FDCWD file descriptor) failures after glibc update?

Hi Ondrej,

Thanks for the heads up.  Good timing.
That does indeed look like a bug, and Aurelien Jarno's fix seems right.

Here's "git diff -w" output:

diff --git a/lib/utimens.c b/lib/utimens.c
index 4df1646..6b7716f 100644
--- a/lib/utimens.c
+++ b/lib/utimens.c
@@ -264,6 +264,7 @@ fdutimens (char const *file, int fd, struct timespec const timespec[2])
         }
 # endif /* HAVE_UTIMENSAT */
 # if HAVE_FUTIMENS
+      if (0 <= fd)
       {
         result = futimens (fd, ts);
 #  ifdef __linux__





I expect to push the following to gnulib in the morning:

From e7d83f123f5eed312523252c066510b5de0f7cd1 Mon Sep 17 00:00:00 2001
From: Aurelien Jarno <aurelien@aurel32.net>
Date: Tue, 5 Jan 2010 21:31:41 +0100
Subject: [PATCH] utimens (fdutimens): ignore a negative FD, per contract

* lib/utimens.c (fdutimens) [HAVE_FUTIMENS]: Call futimens only
when we have a valid file descriptor.  Otherwise, using a brand
new glibc (with just-patched futimes that now fails with EBADF)
would cause this function to fail with ENOSYS.
Reported by Guillaume Ayoub in http://bugs.debian.org/563726.
See also http://bugzilla.redhat.com/552320
---
 ChangeLog     |   10 ++++++++++
 lib/utimens.c |   23 ++++++++++++-----------
 2 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 2857d3d..80ba5cd 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,13 @@
+2010-01-05  Aurelien Jarno  <aurelien@aurel32.net>
+
+	utimens (fdutimens): ignore a negative FD, per contract
+	* lib/utimens.c (fdutimens) [HAVE_FUTIMENS]: Call futimens only
+	when we have a valid file descriptor.  Otherwise, using a brand
+	new glibc (with just-patched futimes that now fails with EBADF)
+	would cause this function to fail with ENOSYS.
+	Reported by Guillaume Ayoub in http://bugs.debian.org/563726.
+	See also http://bugzilla.redhat.com/552320
+
 2010-01-05  Jim Meyering  <meyering@redhat.com>

 	maint: remove useless inclusions of "xalloc.h"
diff --git a/lib/utimens.c b/lib/utimens.c
index 4df1646..6b7716f 100644
--- a/lib/utimens.c
+++ b/lib/utimens.c
@@ -264,19 +264,20 @@ fdutimens (char const *file, int fd, struct timespec const timespec[2])
         }
 # endif /* HAVE_UTIMENSAT */
 # if HAVE_FUTIMENS
-      {
-        result = futimens (fd, ts);
+      if (0 <= fd)
+	{
+	  result = futimens (fd, ts);
 #  ifdef __linux__
-        /* Work around the same bug as above.  */
-        if (0 < result)
-          errno = ENOSYS;
+	  /* Work around the same bug as above.  */
+	  if (0 < result)
+	    errno = ENOSYS;
 #  endif /* __linux__ */
-        if (result == 0 || errno != ENOSYS)
-          {
-            utimensat_works_really = 1;
-            return result;
-          }
-      }
+	  if (result == 0 || errno != ENOSYS)
+	    {
+	      utimensat_works_really = 1;
+	      return result;
+	    }
+	}
 # endif /* HAVE_FUTIMENS */
     }
   utimensat_works_really = -1;
--
1.6.6.387.g2649b1

Message #93 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Eric Blake <ebb9@byu.net>

To: Jim Meyering <jim@meyering.net>

Cc: Ondr(ej Vas(ík <ondrej.vasik@gmail.com>, 563726@bugs.debian.org, bug-gnulib@gnu.org

Subject: Re: [Bug 552320] Handling AT_FDCWD as EBADF in glibc's utimens() causes tar failures

Date: Tue, 05 Jan 2010 20:15:31 -0700

[Message part 1 (text/plain, inline)]

According to Jim Meyering on 1/5/2010 2:07 PM:
>> https://bugzilla.redhat.com/show_bug.cgi?id=552320
> Thanks for the heads up.  Good timing.
> That does indeed look like a bug, and Aurelien Jarno's fix seems right.

I can confirm that the fix is correct.

> 
> I expect to push the following to gnulib in the morning:
> 
>>From e7d83f123f5eed312523252c066510b5de0f7cd1 Mon Sep 17 00:00:00 2001
> From: Aurelien Jarno <aurelien@aurel32.net>
> Date: Tue, 5 Jan 2010 21:31:41 +0100
> Subject: [PATCH] utimens (fdutimens): ignore a negative FD, per contract
> 
> * lib/utimens.c (fdutimens) [HAVE_FUTIMENS]: Call futimens only
> when we have a valid file descriptor.  Otherwise, using a brand
> new glibc (with just-patched futimes that now fails with EBADF)

s/futimes/futimens/ (ChangeLog and commit message)

> +++ b/lib/utimens.c
> @@ -264,19 +264,20 @@ fdutimens (char const *file, int fd, struct timespec const timespec[2])
>          }
>  # endif /* HAVE_UTIMENSAT */
>  # if HAVE_FUTIMENS
> -      {
> -        result = futimens (fd, ts);
> +      if (0 <= fd)
> +	{

Why the TAB?  We just got rid of those in gnulib.

-- 
Don't work too hard, make some time for fun as well!

Eric Blake             ebb9@byu.net

[signature.asc (application/pgp-signature, attachment)]

Message #98 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Jim Meyering <jim@meyering.net>

To: Eric Blake <ebb9@byu.net>

Cc: Ondr(ej Vas(ík <ondrej.vasik@gmail.com>, 563726@bugs.debian.org, bug-gnulib@gnu.org

Subject: Re: [Bug 552320] Handling AT_FDCWD as EBADF in glibc's utimens() causes tar failures

Date: Wed, 06 Jan 2010 07:47:42 +0100

Eric Blake wrote:
...
> s/futimes/futimens/ (ChangeLog and commit message)

Thanks!

>> +++ b/lib/utimens.c
>> @@ -264,19 +264,20 @@ fdutimens (char const *file, int fd, struct timespec const timespec[2])
>>          }
>>  # endif /* HAVE_UTIMENSAT */
>>  # if HAVE_FUTIMENS
>> -      {
>> -        result = futimens (fd, ts);
>> +      if (0 <= fd)
>> +	{
>
> Why the TAB?  We just got rid of those in gnulib.

My emacs code had not been informed about gnulib's change,
so I actually inserted them as I wrote that patch.

Message #103 received at 563726@bugs.debian.org (full text, mbox, reply):

From: Goswin von Brederlow <goswin-v-b@web.de>

To: Debian Bug Tracking System <563726@bugs.debian.org>

Subject: Should open file if fd < 0 and futimesat() is ENOSYS

Date: Wed, 10 Feb 2010 14:19:25 +0100

Package: tar
Version: 1.22-2
Severity: normal

Hi,

first let me say that this bug is pretty anoying as it makes it
impossible to bootstrap a recent Debian or Ubuntu here. Please make
sure this gets fixed for Squeeze and Ubuntu Lucid.

Now, looking at the source and strace I see that tar first tries
utimensat(AT_FDCWD, file, timespec, 0) and recieves ENOSYS.

Then it falls through to calling futimens (fd, timespec) recieving
EBADFD and failing.


The patch in the bugreport adds

+      if (0 <= fd)

That limits the fallback to cases that do have an FD, which is not the
case here. For otheres, which here is the case, it then falls back to
utimes which has less resolution for the timestamp.

Why not open the file, call futimens(new_fd, timespec) and close it
again?

MfG
	Goswin

-- System Information:
Debian Release: squeeze/sid
  APT prefers unstable
  APT policy: (500, 'unstable'), (500, 'stable'), (1, 'experimental')
Architecture: amd64 (x86_64)

Kernel: Linux 2.6.31.5-book-1 (SMP w/2 CPU cores)
Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968)
Shell: /bin/sh linked to /bin/bash

Versions of packages tar depends on:
ii  libc6                         2.10.2-5   Embedded GNU C Library: Shared lib

tar recommends no packages.

Versions of packages tar suggests:
ii  bzip2                         1.0.5-4    high-quality block-sorting file co
pn  ncompress                     <none>     (no description available)

-- no debconf information

Message #108 received at 563726-done@bugs.debian.org (full text, mbox, reply):

From: bdale@gag.com (Bdale Garbee)

To: 563726-done@bugs.debian.org

Subject: fixed

Date: Tue, 16 Mar 2010 13:48:53 -0600 (MDT)

This bug is confirmed to be fixed in 1.23-1.

Bdale

Send a report that this bug log contains spam.