Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-04-21 Thread GOTO Masanori
tags 288710 confirmed fixed-upstream
thanks

At Wed, 6 Apr 2005 07:48:16 +1000,
Nathan Scott wrote:
> > Maybe should this bug be untagged "upstream" then ?
> 
> Since its working on other versions of glibc (ie SLES9), odds are
> its an upstream bug anyway (maybe not, but probably).  At this
> stage, I think we should leave it for the glibc guys to determine.

Yes, this bug was already fixed in the recent glibc - it's affected
only for sarge and sid.  Note that the current glibc 2.3.5-1 in
experimental does not have this problem.

Regards,
-- gotom



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-04-05 Thread Nathan Scott
On Tue, Apr 05, 2005 at 10:32:49AM +0200, Xavier Hienne wrote:
> Nathan Scott a écrit :
> > Let's reassign this bug to glibc, and ask those folks to take
> > a look at the problem.
> 
> Maybe should this bug be untagged "upstream" then ?

Since its working on other versions of glibc (ie SLES9), odds are
its an upstream bug anyway (maybe not, but probably).  At this
stage, I think we should leave it for the glibc guys to determine.

cheers.

-- 
Nathan



Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-04-05 Thread Xavier Hienne
Nathan Scott a écrit :
On Mon, Apr 04, 2005 at 10:58:45PM +0200, Xavier Hienne wrote:
Hi,
I too am encountering the very same bug on a 1900+ entry /home 
directory. The bug is easily reproducible :

Yep, just tried it and it is indeed easy to hit with that test
An even easier way to reproduce it could be :
$ sh
$ ulimit -n 5
$ getfacl / / /

case, thanks.  I've also tried this on a SLES9 machine I had
handy, and the same getfacl code doesn't fail there, so I guess
you're right in suggesting this is a glibc problem.
Let's reassign this bug to glibc, and ask those folks to take
a look at the problem.
Maybe should this bug be untagged "upstream" then ?
Regards,
Xavier


Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-04-04 Thread Nathan Scott
On Mon, Apr 04, 2005 at 10:58:45PM +0200, Xavier Hienne wrote:
> Hi,
> 
> I too am encountering the very same bug on a 1900+ entry /home 
> directory. The bug is easily reproducible :

Yep, just tried it and it is indeed easy to hit with that test
case, thanks.  I've also tried this on a SLES9 machine I had
handy, and the same getfacl code doesn't fail there, so I guess
you're right in suggesting this is a glibc problem.

Let's reassign this bug to glibc, and ask those folks to take
a look at the problem.

thanks.

-- 
Nathan


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-04-04 Thread Xavier Hienne
Hi,
I too am encountering the very same bug on a 1900+ entry /home 
directory. The bug is easily reproducible :

$ mkdir /tmp/dir
$ cd /tmp/dir
$ for ((i=0; i<2000; i++)); do mkdir $i; done
$ getfacl *
[...]
user::rwx
group::r-x
other::r-x
getfacl: Too many open files
getfacl: Too many open files
getfacl: Too many open files
[...]
The bug only occurs with a great number of directories, not files. Both 
Woody and Sid (glibc 2.3.2) are affected. It seems that this is not an 
acl bug but a glibc bug. For example, it also occurs with the getfattr 
from the attr package.

After a bit of investigation, I suspect the nftw(3) function (file 
io/ftw.c in the glibc sources).

Here is an excerpt of "ltrace -S -n4 getfattr *" :
<... nftw64 resumed> )= 1
nftw64(0xbfffec85, 0x08049500, 0, 0, 0x40136e48 
SYS_access(0xbfffec85, 4, 0x40136e48, 0, 0xbfffb8c4) = 0
SYS_stat64(0x0804d178, 0xbfffb768, 0x40137aa0, 0x4013b4ec, 
0x0804d178) = 0
SYS_open("1916", 100352, 00)  = 1023
SYS_fstat64(1023, 0xbfffb53c, 0x40137aa0, 0x4013b4ec, 1023) = 0
SYS_fcntl64(1023, 2, 1, 0x0804d178, 1023) = 0
listxattr(0x0804d178, 0, 0, 0x4002a2f8, 0x400235d8 
SYS_listxattr(0x0804d178, 0, 0, 0x40136e48, 0x08458098) = 0
<... listxattr resumed> ) = 0
free(0x0804b368)  = 

At each inner loop of nftw64, the result of SYS_open increments by one 
until the kernel returns -EMFILE (Too many open files). I suspect that 
the opendir() call in ftw.c is not followed by a corresponding closedir().

I have no time to look further and I hope that someone else will easily 
find what's going wrong exactly.

Regards,
Xavier
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]


Bug#288710: setfacl with many files at cmdline -> Too many open files

2005-01-10 Thread Dirk Pape
Hello Nathan,
--Am 10. Januar 2005 09:35:54 +1100 schrieb Nathan Scott <[EMAIL PROTECTED]>:
Hi Dirk,
I sent this onto the maintainer, he's asked a couple of questions:
On Thu, Jan 06, 2005 at 07:18:35PM +0100, Andreas Gruenbacher wrote:
On Thursday 06 January 2005 04:43, Nathan Scott wrote:
> FYI - wonder if we have a file descriptor leak in setfacl?
> Or perhaps the nftw in there is holding too many fds open?
What's in /proc/sys/fs/file-max?
/proc/sys/fs/file-max is 64000
What does this give?
find . -print0 | xargs -0 sh -c "cat /proc/sys/fs/file-nr"
many lines ;-) with
2232146264000
2nd number varies between 1460 an 1467
Regards,
Dirk.

--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]