git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Valery Yundin]

Hi,

Looks like there might be a bug in SVN import code. Here is the
command that fails

> git svn clone --username anonymous svn://powhegbox.mib.infn.it/trunk/POWHEG-BOX

r217 = 2e6cdb1f4604b2256195590fa8275632971eac42 (refs/remotes/git-svn)
        M       lhefread.f
        M       Z_plus_jet/Born.f
        M       Z_plus_jet/powheg.input
        M       Z_plus_jet/init_processes.f
        D       JJ/madgraph_3_flavours/born/nexternal.inc
write .git/Git_svn_hash_bl5Cj3: Bad file descriptor
 at /usr/lib/perl5/vendor_perl/5.20.1/x86_64-linux-thread-multi/SVN/Ra.pm
line 623.

Tested on:
git-svn version 2.3.0.rc2 (svn 1.8.11) - FAIL
git-svn version 2.2.2 (svn 1.8.10) - FAIL
git-svn version 1.8.4.5 (svn 1.8.11) - WORKS

PS unfortunately I don't have admin access to SVN repository

With best regards, Valery Yundin.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Valery Yundin <yuvalery@gmail.com> wrote:
> Tested on:
> git-svn version 2.3.0.rc2 (svn 1.8.11) - FAIL
> git-svn version 2.2.2 (svn 1.8.10) - FAIL
> git-svn version 1.8.4.5 (svn 1.8.11) - WORKS

Thank you for that info.  Do you think you can bisect which
versions/revisions of git-svn introduced that failure for us?  I don't
have much time this part of the year for git-svn, but maybe it's related
to the performance work we did around fall 2014.

Thanks again.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Valery Yundin]

Hi,

Here you go:
dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit
commit dfa72fdb96befbd790f623bb2909a347176753c2
Author: Eric Wong <normalperson@yhbt.net>
Date:   Fri Oct 24 22:53:52 2014 +0000

    git-svn: reload RA every log-window-size

    Despite attempting to use local memory pools everywhere we can,
    (including our call to SVN::Ra::do_update and all subsequent reporter
    calls), there does not appear to be a way to force the Git::SVN::Fetcher
    callbacks to use a pool other than the per-SVN::Ra pool.
    Git::SVN::Fetcher ends up using the main RA pool which grows
    monotonically in size for the lifetime of the RA object.

    Thus the only way to free that memory appears to be to destroy and
    recreate the RA connection for at every --log-window-size interval.

    This reduces memory usage over the course of fetching 10K revisions
    using a test repository created with the script at the end of this
    commit message.
    .........

Cheers,
Valery
With best regards, Mr. Valery Yundin.


On 30 January 2015 at 00:34, Eric Wong <normalperson@yhbt.net> wrote:
> Valery Yundin <yuvalery@gmail.com> wrote:
>> Tested on:
>> git-svn version 2.3.0.rc2 (svn 1.8.11) - FAIL
>> git-svn version 2.2.2 (svn 1.8.10) - FAIL
>> git-svn version 1.8.4.5 (svn 1.8.11) - WORKS
>
> Thank you for that info.  Do you think you can bisect which
> versions/revisions of git-svn introduced that failure for us?  I don't
> have much time this part of the year for git-svn, but maybe it's related
> to the performance work we did around fall 2014.
>
> Thanks again.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Valery Yundin <yuvalery@gmail.com> wrote:
> Hi,
> 
> Here you go:
> dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit

Thank you.  Can you give the following patch a try?
I have not been able to reproduce the problem on my end.
If it doesn't work out, I might be out of ideas for a bit :/
Increasing --log-window-size will help you run longer without
the error, but that's not ideal as it can also eat memory.

-----------------------8<----------------------
From: Eric Wong <normalperson@yhbt.net>
Subject: [PATCH] git-svn: destroy all tempfiles when reloading RA

This may fix the errors some users are seeing with:
"write .git/Git_svn_hash_XXXXXX: Bad file descriptor"

Thanks to Valery Yundin for helping bisect the problem introduced in
commit dfa72fdb96befbd790f623bb2909a347176753c2
(git-svn: reload RA every log-window-size)

Cc: Valery Yundin <yuvalery@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
---
 perl/Git.pm        | 6 ++++++
 perl/Git/SVN/Ra.pm | 1 +
 2 files changed, 7 insertions(+)

diff --git a/perl/Git.pm b/perl/Git.pm
index b5905ee..698018e 100644
--- a/perl/Git.pm
+++ b/perl/Git.pm
@@ -1347,6 +1347,12 @@ sub temp_path {
 	$TEMP_FILES{$temp_fd}{fname};
 }
 
+sub temp_reset_all {
+	unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
+	%TEMP_FILEMAP = ();
+	%TEMP_FILES = ();
+}
+
 sub END {
 	unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
 }
diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
index 622535e..878679d 100644
--- a/perl/Git/SVN/Ra.pm
+++ b/perl/Git/SVN/Ra.pm
@@ -397,6 +397,7 @@ sub gs_fetch_loop_common {
 		$_[0] = undef;
 		$self = undef;
 		$RA = undef;
+		Git->temp_reset_all;
 		$gpool->clear;
 		$self = Git::SVN::Ra->new($ra_url);
 		$ra_invalid = undef;
-- 
EW

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Valery Yundin]

Hi,

Your patch seems to fix the problem.
I tried several times and I can svn clone the whole repository in one
go without a crash.

Thanks,
Valery

On 30 January 2015 at 01:22, Eric Wong <normalperson@yhbt.net> wrote:
> Valery Yundin <yuvalery@gmail.com> wrote:
>> Hi,
>>
>> Here you go:
>> dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit
>
> Thank you.  Can you give the following patch a try?
> I have not been able to reproduce the problem on my end.
> If it doesn't work out, I might be out of ideas for a bit :/
> Increasing --log-window-size will help you run longer without
> the error, but that's not ideal as it can also eat memory.
>
> -----------------------8<----------------------
> From: Eric Wong <normalperson@yhbt.net>
> Subject: [PATCH] git-svn: destroy all tempfiles when reloading RA
>
> This may fix the errors some users are seeing with:
> "write .git/Git_svn_hash_XXXXXX: Bad file descriptor"
>
> Thanks to Valery Yundin for helping bisect the problem introduced in
> commit dfa72fdb96befbd790f623bb2909a347176753c2
> (git-svn: reload RA every log-window-size)
>
> Cc: Valery Yundin <yuvalery@gmail.com>
> Signed-off-by: Eric Wong <normalperson@yhbt.net>
> ---
>  perl/Git.pm        | 6 ++++++
>  perl/Git/SVN/Ra.pm | 1 +
>  2 files changed, 7 insertions(+)
>
> diff --git a/perl/Git.pm b/perl/Git.pm
> index b5905ee..698018e 100644
> --- a/perl/Git.pm
> +++ b/perl/Git.pm
> @@ -1347,6 +1347,12 @@ sub temp_path {
>         $TEMP_FILES{$temp_fd}{fname};
>  }
>
> +sub temp_reset_all {
> +       unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
> +       %TEMP_FILEMAP = ();
> +       %TEMP_FILES = ();
> +}
> +
>  sub END {
>         unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
>  }
> diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
> index 622535e..878679d 100644
> --- a/perl/Git/SVN/Ra.pm
> +++ b/perl/Git/SVN/Ra.pm
> @@ -397,6 +397,7 @@ sub gs_fetch_loop_common {
>                 $_[0] = undef;
>                 $self = undef;
>                 $RA = undef;
> +               Git->temp_reset_all;
>                 $gpool->clear;
>                 $self = Git::SVN::Ra->new($ra_url);
>                 $ra_invalid = undef;
> --
> EW

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Valery Yundin <yuvalery@gmail.com> wrote:
> Hi,
> 
> Your patch seems to fix the problem.
> I tried several times and I can svn clone the whole repository in one
> go without a crash.

Thanks for the confirmation.  Cc-ing a few other folks who encountered
this problem (and Bcc-ing some folks who emailed me privately).

Can the rest of you give this patch a try on your respective platforms
and confirm the fix?

http://article.gmane.org/gmane.comp.version-control.git/263168/raw
(also: http://mid.gmane.org/20150130002247.GA22519@dcvr.yhbt.net )

Junio: assuming all goes well with testers, can you apply my patch
to the appropriate maintenance tracks with Tested-by:s?
I'm going offline in a few hours and don't think I'll be around
much the next week or so.

Thanks.

> Thanks,
> Valery
> 
> On 30 January 2015 at 01:22, Eric Wong <normalperson@yhbt.net> wrote:
> > Valery Yundin <yuvalery@gmail.com> wrote:
> >> Hi,
> >>
> >> Here you go:
> >> dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit
> >
> > Thank you.  Can you give the following patch a try?
> > I have not been able to reproduce the problem on my end.
> > If it doesn't work out, I might be out of ideas for a bit :/
> > Increasing --log-window-size will help you run longer without
> > the error, but that's not ideal as it can also eat memory.
> >
> > -----------------------8<----------------------
> > From: Eric Wong <normalperson@yhbt.net>
> > Subject: [PATCH] git-svn: destroy all tempfiles when reloading RA
> >
> > This may fix the errors some users are seeing with:
> > "write .git/Git_svn_hash_XXXXXX: Bad file descriptor"
> >
> > Thanks to Valery Yundin for helping bisect the problem introduced in
> > commit dfa72fdb96befbd790f623bb2909a347176753c2
> > (git-svn: reload RA every log-window-size)
> >
> > Cc: Valery Yundin <yuvalery@gmail.com>
> > Signed-off-by: Eric Wong <normalperson@yhbt.net>
> > ---
> >  perl/Git.pm        | 6 ++++++
> >  perl/Git/SVN/Ra.pm | 1 +
> >  2 files changed, 7 insertions(+)
> >
> > diff --git a/perl/Git.pm b/perl/Git.pm
> > index b5905ee..698018e 100644
> > --- a/perl/Git.pm
> > +++ b/perl/Git.pm
> > @@ -1347,6 +1347,12 @@ sub temp_path {
> >         $TEMP_FILES{$temp_fd}{fname};
> >  }
> >
> > +sub temp_reset_all {
> > +       unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
> > +       %TEMP_FILEMAP = ();
> > +       %TEMP_FILES = ();
> > +}
> > +
> >  sub END {
> >         unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
> >  }
> > diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
> > index 622535e..878679d 100644
> > --- a/perl/Git/SVN/Ra.pm
> > +++ b/perl/Git/SVN/Ra.pm
> > @@ -397,6 +397,7 @@ sub gs_fetch_loop_common {
> >                 $_[0] = undef;
> >                 $self = undef;
> >                 $RA = undef;
> > +               Git->temp_reset_all;
> >                 $gpool->clear;
> >                 $self = Git::SVN::Ra->new($ra_url);
> >                 $ra_invalid = undef;
> > --
> > EW
> --

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Nico Schlömer]

I tried the patch and I still get
```
[...]
r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
Unexpected result returned from git cat-file at
/home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
/home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
line 757.

error closing pipe: Bad file descriptor at
/home/nschloe/libexec/git-core/git-svn line 0.
error closing pipe: Bad file descriptor at
/home/nschloe/libexec/git-core/git-svn line 0.
```
when
```
git svn clone https://geuz.org/svn/gmsh/trunk
```

Cheers,
Nico

On Fri, Jan 30, 2015 at 2:30 AM, Eric Wong <normalperson@yhbt.net> wrote:
> Valery Yundin <yuvalery@gmail.com> wrote:
>> Hi,
>>
>> Your patch seems to fix the problem.
>> I tried several times and I can svn clone the whole repository in one
>> go without a crash.
>
> Thanks for the confirmation.  Cc-ing a few other folks who encountered
> this problem (and Bcc-ing some folks who emailed me privately).
>
> Can the rest of you give this patch a try on your respective platforms
> and confirm the fix?
>
> http://article.gmane.org/gmane.comp.version-control.git/263168/raw
> (also: http://mid.gmane.org/20150130002247.GA22519@dcvr.yhbt.net )
>
> Junio: assuming all goes well with testers, can you apply my patch
> to the appropriate maintenance tracks with Tested-by:s?
> I'm going offline in a few hours and don't think I'll be around
> much the next week or so.
>
> Thanks.
>
>> Thanks,
>> Valery
>>
>> On 30 January 2015 at 01:22, Eric Wong <normalperson@yhbt.net> wrote:
>> > Valery Yundin <yuvalery@gmail.com> wrote:
>> >> Hi,
>> >>
>> >> Here you go:
>> >> dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit
>> >
>> > Thank you.  Can you give the following patch a try?
>> > I have not been able to reproduce the problem on my end.
>> > If it doesn't work out, I might be out of ideas for a bit :/
>> > Increasing --log-window-size will help you run longer without
>> > the error, but that's not ideal as it can also eat memory.
>> >
>> > -----------------------8<----------------------
>> > From: Eric Wong <normalperson@yhbt.net>
>> > Subject: [PATCH] git-svn: destroy all tempfiles when reloading RA
>> >
>> > This may fix the errors some users are seeing with:
>> > "write .git/Git_svn_hash_XXXXXX: Bad file descriptor"
>> >
>> > Thanks to Valery Yundin for helping bisect the problem introduced in
>> > commit dfa72fdb96befbd790f623bb2909a347176753c2
>> > (git-svn: reload RA every log-window-size)
>> >
>> > Cc: Valery Yundin <yuvalery@gmail.com>
>> > Signed-off-by: Eric Wong <normalperson@yhbt.net>
>> > ---
>> >  perl/Git.pm        | 6 ++++++
>> >  perl/Git/SVN/Ra.pm | 1 +
>> >  2 files changed, 7 insertions(+)
>> >
>> > diff --git a/perl/Git.pm b/perl/Git.pm
>> > index b5905ee..698018e 100644
>> > --- a/perl/Git.pm
>> > +++ b/perl/Git.pm
>> > @@ -1347,6 +1347,12 @@ sub temp_path {
>> >         $TEMP_FILES{$temp_fd}{fname};
>> >  }
>> >
>> > +sub temp_reset_all {
>> > +       unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
>> > +       %TEMP_FILEMAP = ();
>> > +       %TEMP_FILES = ();
>> > +}
>> > +
>> >  sub END {
>> >         unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
>> >  }
>> > diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
>> > index 622535e..878679d 100644
>> > --- a/perl/Git/SVN/Ra.pm
>> > +++ b/perl/Git/SVN/Ra.pm
>> > @@ -397,6 +397,7 @@ sub gs_fetch_loop_common {
>> >                 $_[0] = undef;
>> >                 $self = undef;
>> >                 $RA = undef;
>> > +               Git->temp_reset_all;
>> >                 $gpool->clear;
>> >                 $self = Git::SVN::Ra->new($ra_url);
>> >                 $ra_invalid = undef;
>> > --
>> > EW
>> --

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Valery Yundin]

Hi,

It seems that the same commit dfa72fdb96 is responsible for the error
in "git svn clone https://geuz.org/svn/gmsh/trunk". But unlike in my
case, the patch doesn't fix it.

Cheers,
Valery


On 31 January 2015 at 13:51, Nico Schlömer <nico.schloemer@gmail.com> wrote:
> I tried the patch and I still get
> ```
> [...]
> r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> Unexpected result returned from git cat-file at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
> Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
> line 757.
>
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> ```
> when
> ```
> git svn clone https://geuz.org/svn/gmsh/trunk
> ```
>
> Cheers,
> Nico
>
> On Fri, Jan 30, 2015 at 2:30 AM, Eric Wong <normalperson@yhbt.net> wrote:
>> Valery Yundin <yuvalery@gmail.com> wrote:
>>> Hi,
>>>
>>> Your patch seems to fix the problem.
>>> I tried several times and I can svn clone the whole repository in one
>>> go without a crash.
>>
>> Thanks for the confirmation.  Cc-ing a few other folks who encountered
>> this problem (and Bcc-ing some folks who emailed me privately).
>>
>> Can the rest of you give this patch a try on your respective platforms
>> and confirm the fix?
>>
>> http://article.gmane.org/gmane.comp.version-control.git/263168/raw
>> (also: http://mid.gmane.org/20150130002247.GA22519@dcvr.yhbt.net )
>>
>> Junio: assuming all goes well with testers, can you apply my patch
>> to the appropriate maintenance tracks with Tested-by:s?
>> I'm going offline in a few hours and don't think I'll be around
>> much the next week or so.
>>
>> Thanks.
>>
>>> Thanks,
>>> Valery
>>>
>>> On 30 January 2015 at 01:22, Eric Wong <normalperson@yhbt.net> wrote:
>>> > Valery Yundin <yuvalery@gmail.com> wrote:
>>> >> Hi,
>>> >>
>>> >> Here you go:
>>> >> dfa72fdb96befbd790f623bb2909a347176753c2 is the first bad commit
>>> >
>>> > Thank you.  Can you give the following patch a try?
>>> > I have not been able to reproduce the problem on my end.
>>> > If it doesn't work out, I might be out of ideas for a bit :/
>>> > Increasing --log-window-size will help you run longer without
>>> > the error, but that's not ideal as it can also eat memory.
>>> >
>>> > -----------------------8<----------------------
>>> > From: Eric Wong <normalperson@yhbt.net>
>>> > Subject: [PATCH] git-svn: destroy all tempfiles when reloading RA
>>> >
>>> > This may fix the errors some users are seeing with:
>>> > "write .git/Git_svn_hash_XXXXXX: Bad file descriptor"
>>> >
>>> > Thanks to Valery Yundin for helping bisect the problem introduced in
>>> > commit dfa72fdb96befbd790f623bb2909a347176753c2
>>> > (git-svn: reload RA every log-window-size)
>>> >
>>> > Cc: Valery Yundin <yuvalery@gmail.com>
>>> > Signed-off-by: Eric Wong <normalperson@yhbt.net>
>>> > ---
>>> >  perl/Git.pm        | 6 ++++++
>>> >  perl/Git/SVN/Ra.pm | 1 +
>>> >  2 files changed, 7 insertions(+)
>>> >
>>> > diff --git a/perl/Git.pm b/perl/Git.pm
>>> > index b5905ee..698018e 100644
>>> > --- a/perl/Git.pm
>>> > +++ b/perl/Git.pm
>>> > @@ -1347,6 +1347,12 @@ sub temp_path {
>>> >         $TEMP_FILES{$temp_fd}{fname};
>>> >  }
>>> >
>>> > +sub temp_reset_all {
>>> > +       unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
>>> > +       %TEMP_FILEMAP = ();
>>> > +       %TEMP_FILES = ();
>>> > +}
>>> > +
>>> >  sub END {
>>> >         unlink values %TEMP_FILEMAP if %TEMP_FILEMAP;
>>> >  }
>>> > diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
>>> > index 622535e..878679d 100644
>>> > --- a/perl/Git/SVN/Ra.pm
>>> > +++ b/perl/Git/SVN/Ra.pm
>>> > @@ -397,6 +397,7 @@ sub gs_fetch_loop_common {
>>> >                 $_[0] = undef;
>>> >                 $self = undef;
>>> >                 $RA = undef;
>>> > +               Git->temp_reset_all;
>>> >                 $gpool->clear;
>>> >                 $self = Git::SVN::Ra->new($ra_url);
>>> >                 $ra_invalid = undef;
>>> > --
>>> > EW
>>> --

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Valery Yundin <yuvalery@gmail.com> wrote:
> On 31 January 2015 at 13:51, Nico Schlömer <nico.schloemer@gmail.com> wrote:
> > I tried the patch and I still get
> > ```
> > [...]
> > r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> > Unexpected result returned from git cat-file at
> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
> > Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
> > line 757.
> >
> > error closing pipe: Bad file descriptor at
> > /home/nschloe/libexec/git-core/git-svn line 0.
> > error closing pipe: Bad file descriptor at
> > /home/nschloe/libexec/git-core/git-svn line 0.
> > ```
> > when
> > ```
> > git svn clone https://geuz.org/svn/gmsh/trunk
> 
> It seems that the same commit dfa72fdb96 is responsible for the error
> in "git svn clone https://geuz.org/svn/gmsh/trunk". But unlike in my
> case, the patch doesn't fix it.

(top-posting corrected)

Odd, I managed to clone that without issues, but I couldn't reproduce
this problem with or without the tempfile clearing patch applied.

   git svn clone --username gmsh https://geuz.org/svn/gmsh/trunk

Anybody else?

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Nico Schlömer]

I just double-checked and I can only produce this issue on one machine
(tested on 3). Apparently, this is has nothing to do with Git itself
then.

Any ideas of what could be wrong?

Cheers,
Nico

On Thu, Feb 12, 2015 at 8:18 PM, Eric Wong <normalperson@yhbt.net> wrote:
> Valery Yundin <yuvalery@gmail.com> wrote:
>> On 31 January 2015 at 13:51, Nico Schlömer <nico.schloemer@gmail.com> wrote:
>> > I tried the patch and I still get
>> > ```
>> > [...]
>> > r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
>> > Unexpected result returned from git cat-file at
>> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
>> > Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
>> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
>> > line 757.
>> >
>> > error closing pipe: Bad file descriptor at
>> > /home/nschloe/libexec/git-core/git-svn line 0.
>> > error closing pipe: Bad file descriptor at
>> > /home/nschloe/libexec/git-core/git-svn line 0.
>> > ```
>> > when
>> > ```
>> > git svn clone https://geuz.org/svn/gmsh/trunk
>>
>> It seems that the same commit dfa72fdb96 is responsible for the error
>> in "git svn clone https://geuz.org/svn/gmsh/trunk". But unlike in my
>> case, the patch doesn't fix it.
>
> (top-posting corrected)
>
> Odd, I managed to clone that without issues, but I couldn't reproduce
> this problem with or without the tempfile clearing patch applied.
>
>    git svn clone --username gmsh https://geuz.org/svn/gmsh/trunk
>
> Anybody else?

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Nico Schlömer <nico.schloemer@gmail.com> wrote:
> I just double-checked and I can only produce this issue on one machine
> (tested on 3). Apparently, this is has nothing to do with Git itself
> then.
> 
> Any ideas of what could be wrong?

What's different about that one machine?
e.g. SVN version, 64 vs 32-bit, Perl version, etc. could all be
factors (assuming identical git versions).

Also, any chance git was misinstalled somehow or your PATH was not
pointing to the correct git installation?

Thanks

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Kyle J. McKay]

I believe I have been able to track down this problem and implement a
fix.  Please report back if this patch fixes the problem for you.

-Kyle

-- 8< --
Subject: [PATCH] Git::SVN::Fetcher: avoid premature 'svn_hash' temp file closure

Since b19138b (git-svn: Make it incrementally faster by minimizing temp
files, v1.6.0), git-svn has been using the Git.pm temp_acquire and
temp_release mechanism to avoid unnecessary temp file churn and provide
a speed boost.

However, that change introduced a call to temp_acquire inside the
Git::SVN::Fetcher::close_file function for an 'svn_hash' temp file.
Because an SVN::Pool is active at the time this function is called, if
the Git::temp_acquire function ends up actually creating a new
FileHandle for the temp file (which it will the first time it's called
with the name 'svn_hash') that FileHandle will end up in the SVN::Pool
and should that pool have SVN::Pool::clear called on it that FileHandle
will be closed out from under Git::temp_acquire.

Since the only call site to Git::temp_acquire with the name 'svn_hash'
is inside the close_file function, if an 'svn_hash' temp file is ever
created its FileHandle is guaranteed to be created in the active
SVN::Pool.

This has not been a problem in the past because the SVN::Pool was not
being cleared.  However, since dfa72fdb (git-svn: reload RA every
log-window-size, v2.2.0) the pool has been getting cleared periodically
at which point the FileHandle for the 'svn_hash' temp file gets closed.
Any subsequent calls to Git::temp_acquire for 'svn_hash', however,
succeed without creating/opening a new temporary file since it still has
the now invalid FileHandle in its cache.  Callers that then attempt to
use that FileHandle fail with an error.

We avoid this problem by making sure the 'svn_hash' temp file is created
in the same place the 'svn_delta_...' and 'git_blob_...' temp files are
(and then temp_release'd) so that it can be safely used inside the
close_file function without having its FileHandle end up in an SVN::Pool
that gets cleared.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
---
 perl/Git/SVN/Fetcher.pm | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/perl/Git/SVN/Fetcher.pm b/perl/Git/SVN/Fetcher.pm
index 10edb277..613055a3 100644
--- a/perl/Git/SVN/Fetcher.pm
+++ b/perl/Git/SVN/Fetcher.pm
@@ -322,6 +322,14 @@ sub apply_textdelta {
 	# (but $base does not,) so dup() it for reading in close_file
 	open my $dup, '<&', $fh or croak $!;
 	my $base = $::_repository->temp_acquire("git_blob_${$}_$suffix");
+	# close_file may call temp_acquire on 'svn_hash', but because of the
+	# call chain, if the temp_acquire call from close_file ends up being the
+	# call that first creates the 'svn_hash' temp file, then the FileHandle
+	# that's created as a result will end up in an SVN::Pool that we clear
+	# in SVN::Ra::gs_fetch_loop_common.  Avoid that by making sure the
+	# 'svn_hash' FileHandle is already created before close_file is called.
+	my $tmp_fh = $::_repository->temp_acquire('svn_hash');
+	$::_repository->temp_release($tmp_fh, 1);
 
 	if ($fb->{blob}) {
 		my ($base_is_link, $size);
--

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Nico Schlömer <nico.schloemer@gmail.com> wrote:
> Thanks Kyle for the patch! I applied it and ran
> ```
> git svn clone https://geuz.org/svn/gmsh/trunk
> ```
> Unfortunately, I'm still getting
> ```
> [...]
> r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> error closing pipe: Bad file descriptor at
> /usr/share/perl5/Git/SVN/Fetcher.pm line 335.
> error closing pipe: Bad file descriptor at
> /usr/share/perl5/Git/SVN/Fetcher.pm line 335.
> out pipe went bad at /usr/share/perl5/Git.pm line 955.

Thanks for testing, is this with or without my other attempt at
fixing this problem applied?

http://mid.gmane.org/20150130002247.GA22519@dcvr.yhbt.net
("git-svn: destroy all tempfiles when reloading RA")

And can you try testing both with/without that patch if you
haven't already?  Thanks again.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Kyle J. McKay]

On Feb 25, 2015, at 07:07, Nico Schlömer wrote:

> Thanks Kyle for the patch! I applied it and ran
> ```
> git svn clone https://geuz.org/svn/gmsh/trunk
> ```
> Unfortunately, I'm still getting
> ```
> [...]
> r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> error closing pipe: Bad file descriptor at /usr/share/perl5/Git/SVN/ 
> Fetcher.pm line 335.
> error closing pipe: Bad file descriptor at /usr/share/perl5/Git/SVN/ 
> Fetcher.pm line 335.
> out pipe went bad at /usr/share/perl5/Git.pm line 955.
> ```

Are you certain you're running with the patch?  I ask because earlier  
you sent this:

On Jan 31, 2015, at 04:51, Nico Schlömer wrote:

> I tried the patch and I still get
> ```
> [...]
> r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> Unexpected result returned from git cat-file at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
> Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
> line 757.
>
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> ```
> when
> ```
> git svn clone https://geuz.org/svn/gmsh/trunk
> ```

And as the patch below applies to Fetcher.pm at line 322 and inserts 8  
lines, I do not see how you could be getting the same error message at  
line 335 if the patch was present.  Even if you manually applied it by  
only inserting the two lines of code, the line numbers would be at  
least 337, not 335 as reported above.  I also notice the path to  
Fetcher.pm is different from your earlier report.

That said, the fact that it happens right after r100 (which is when  
SVN::Pool::clear is getting called) suggests there's another file  
handle getting caught up in the SVN::Pool somehow.  (When I try to  
clone gmsh, I got to r4885 before a server disconnection.)

It could be as simple as the open2 call FileHandle result in later  
perl versions ends up in the SVN::Pool whereas in earlier Perl and/or  
svn versions it does not.

What happens if you add this change on top of the Fetcher.pm change:

diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
index 622535e2..96888228 100644
--- a/perl/Git/SVN/Ra.pm
+++ b/perl/Git/SVN/Ra.pm
@@ -391,6 +391,7 @@ sub longest_common_path {
  sub gs_fetch_loop_common {
  	my ($self, $base, $head, $gsv, $globs) = @_;
  	return if ($base > $head);
+	$::_repository->_open_cat_blob_if_needed;
  	my $gpool = SVN::Pool->new_default;
  	my $ra_url = $self->url;
  	my $reload_ra = sub {

-Kyle

> Cheers,
> Nico
>
> On Wed, Feb 25, 2015 at 11:20 AM Kyle J. McKay <mackyle@gmail.com>  
> wrote:
> I believe I have been able to track down this problem and implement a
> fix.  Please report back if this patch fixes the problem for you.
>
> -Kyle
>
> -- 8< --
> Subject: [PATCH] Git::SVN::Fetcher: avoid premature 'svn_hash' temp  
> file closure
>
> Since b19138b (git-svn: Make it incrementally faster by minimizing  
> temp
> files, v1.6.0), git-svn has been using the Git.pm temp_acquire and
> temp_release mechanism to avoid unnecessary temp file churn and  
> provide
> a speed boost.
>
> However, that change introduced a call to temp_acquire inside the
> Git::SVN::Fetcher::close_file function for an 'svn_hash' temp file.
> Because an SVN::Pool is active at the time this function is called, if
> the Git::temp_acquire function ends up actually creating a new
> FileHandle for the temp file (which it will the first time it's called
> with the name 'svn_hash') that FileHandle will end up in the SVN::Pool
> and should that pool have SVN::Pool::clear called on it that  
> FileHandle
> will be closed out from under Git::temp_acquire.
>
> Since the only call site to Git::temp_acquire with the name 'svn_hash'
> is inside the close_file function, if an 'svn_hash' temp file is ever
> created its FileHandle is guaranteed to be created in the active
> SVN::Pool.
>
> This has not been a problem in the past because the SVN::Pool was not
> being cleared.  However, since dfa72fdb (git-svn: reload RA every
> log-window-size, v2.2.0) the pool has been getting cleared  
> periodically
> at which point the FileHandle for the 'svn_hash' temp file gets  
> closed.
> Any subsequent calls to Git::temp_acquire for 'svn_hash', however,
> succeed without creating/opening a new temporary file since it still  
> has
> the now invalid FileHandle in its cache.  Callers that then attempt to
> use that FileHandle fail with an error.
>
> We avoid this problem by making sure the 'svn_hash' temp file is  
> created
> in the same place the 'svn_delta_...' and 'git_blob_...' temp files  
> are
> (and then temp_release'd) so that it can be safely used inside the
> close_file function without having its FileHandle end up in an  
> SVN::Pool
> that gets cleared.
>
> Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
> ---
>  perl/Git/SVN/Fetcher.pm | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/perl/Git/SVN/Fetcher.pm b/perl/Git/SVN/Fetcher.pm
> index 10edb277..613055a3 100644
> --- a/perl/Git/SVN/Fetcher.pm
> +++ b/perl/Git/SVN/Fetcher.pm
> @@ -322,6 +322,14 @@ sub apply_textdelta {
>         # (but $base does not,) so dup() it for reading in close_file
>         open my $dup, '<&', $fh or croak $!;
>         my $base = $::_repository->temp_acquire("git_blob_${$}_ 
> $suffix");
> +       # close_file may call temp_acquire on 'svn_hash', but  
> because of the
> +       # call chain, if the temp_acquire call from close_file ends  
> up being the
> +       # call that first creates the 'svn_hash' temp file, then the  
> FileHandle
> +       # that's created as a result will end up in an SVN::Pool  
> that we clear
> +       # in SVN::Ra::gs_fetch_loop_common.  Avoid that by making  
> sure the
> +       # 'svn_hash' FileHandle is already created before close_file  
> is called.
> +       my $tmp_fh = $::_repository->temp_acquire('svn_hash');
> +       $::_repository->temp_release($tmp_fh, 1);
>
>         if ($fb->{blob}) {
>                 my ($base_is_link, $size);
> --

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Nico Schlömer]

All,

I applied Kyle's latest patch

> diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
> index 622535e2..96888228 100644
> --- a/perl/Git/SVN/Ra.pm
> +++ b/perl/Git/SVN/Ra.pm
> @@ -391,6 +391,7 @@ sub longest_common_path {
> sub gs_fetch_loop_common {
>         my ($self, $base, $head, $gsv, $globs) = @_;
>         return if ($base > $head);
> +       $::_repository->_open_cat_blob_if_needed;
>         my $gpool = SVN::Pool->new_default;
>         my $ra_url = $self->url;
>         my $reload_ra = sub {

alone and this fixes the bug for me. Thanks a lot, Kyle!

Cheers,
Nico

On Thu, Feb 26, 2015 at 12:37 AM, Nico Schlömer
<nico.schloemer@gmail.com> wrote:
> Kyle,
>
> you are absolutely correct, I used the wrong Git installation in my last
> email. With both your and Eric's patches applied, I'm getting
> ```
> [...]
> r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
> Unexpected result returned from git cat-file at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 344.
> Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
> /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 345, <GEN16> line
> 757.
>
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> error closing pipe: Bad file descriptor at
> /home/nschloe/libexec/git-core/git-svn line 0.
> ```
> where line 344 is
> ```
> $size = $::_repository->cat_blob($fb->{blob}, $base);
> ```
> Adding the line in `perl/Git/SVN/Ra.pm` as per your suggestion does not
> change this.
>
> Cheers,
> Nico
>
>
> On Wed, Feb 25, 2015 at 9:34 PM Kyle J. McKay <mackyle@gmail.com> wrote:
>>
>> On Feb 25, 2015, at 07:07, Nico Schlömer wrote:
>>
>> > Thanks Kyle for the patch! I applied it and ran
>> > ```
>> > git svn clone https://geuz.org/svn/gmsh/trunk
>> > ```
>> > Unfortunately, I'm still getting
>> > ```
>> > [...]
>> > r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
>> > error closing pipe: Bad file descriptor at /usr/share/perl5/Git/SVN/
>> > Fetcher.pm line 335.
>> > error closing pipe: Bad file descriptor at /usr/share/perl5/Git/SVN/
>> > Fetcher.pm line 335.
>> > out pipe went bad at /usr/share/perl5/Git.pm line 955.
>> > ```
>>
>> Are you certain you're running with the patch?  I ask because earlier
>> you sent this:
>>
>> On Jan 31, 2015, at 04:51, Nico Schlömer wrote:
>>
>> > I tried the patch and I still get
>> > ```
>> > [...]
>> > r100 = e2a9b5baa2cebb18591ecb04ff350410d52f36de (refs/remotes/git-svn)
>> > Unexpected result returned from git cat-file at
>> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 335.
>> > Failed to read object 619f9d1d857fb287d06a70c9dac6b8b534d0de6a at
>> > /home/nschloe/share/perl/5.18.2/Git/SVN/Fetcher.pm line 336, <GEN16>
>> > line 757.
>> >
>> > error closing pipe: Bad file descriptor at
>> > /home/nschloe/libexec/git-core/git-svn line 0.
>> > error closing pipe: Bad file descriptor at
>> > /home/nschloe/libexec/git-core/git-svn line 0.
>> > ```
>> > when
>> > ```
>> > git svn clone https://geuz.org/svn/gmsh/trunk
>> > ```
>>
>> And as the patch below applies to Fetcher.pm at line 322 and inserts 8
>> lines, I do not see how you could be getting the same error message at
>> line 335 if the patch was present.  Even if you manually applied it by
>> only inserting the two lines of code, the line numbers would be at
>> least 337, not 335 as reported above.  I also notice the path to
>> Fetcher.pm is different from your earlier report.
>>
>> That said, the fact that it happens right after r100 (which is when
>> SVN::Pool::clear is getting called) suggests there's another file
>> handle getting caught up in the SVN::Pool somehow.  (When I try to
>> clone gmsh, I got to r4885 before a server disconnection.)
>>
>> It could be as simple as the open2 call FileHandle result in later
>> perl versions ends up in the SVN::Pool whereas in earlier Perl and/or
>> svn versions it does not.
>>
>> What happens if you add this change on top of the Fetcher.pm change:
>>
>> diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
>> index 622535e2..96888228 100644
>> --- a/perl/Git/SVN/Ra.pm
>> +++ b/perl/Git/SVN/Ra.pm
>> @@ -391,6 +391,7 @@ sub longest_common_path {
>>   sub gs_fetch_loop_common {
>>         my ($self, $base, $head, $gsv, $globs) = @_;
>>         return if ($base > $head);
>> +       $::_repository->_open_cat_blob_if_needed;
>>         my $gpool = SVN::Pool->new_default;
>>         my $ra_url = $self->url;
>>         my $reload_ra = sub {
>>
>> -Kyle
>>
>> > Cheers,
>> > Nico
>> >
>> > On Wed, Feb 25, 2015 at 11:20 AM Kyle J. McKay <mackyle@gmail.com>
>> > wrote:
>> > I believe I have been able to track down this problem and implement a
>> > fix.  Please report back if this patch fixes the problem for you.
>> >
>> > -Kyle
>> >
>> > -- 8< --
>> > Subject: [PATCH] Git::SVN::Fetcher: avoid premature 'svn_hash' temp
>> > file closure
>> >
>> > Since b19138b (git-svn: Make it incrementally faster by minimizing
>> > temp
>> > files, v1.6.0), git-svn has been using the Git.pm temp_acquire and
>> > temp_release mechanism to avoid unnecessary temp file churn and
>> > provide
>> > a speed boost.
>> >
>> > However, that change introduced a call to temp_acquire inside the
>> > Git::SVN::Fetcher::close_file function for an 'svn_hash' temp file.
>> > Because an SVN::Pool is active at the time this function is called, if
>> > the Git::temp_acquire function ends up actually creating a new
>> > FileHandle for the temp file (which it will the first time it's called
>> > with the name 'svn_hash') that FileHandle will end up in the SVN::Pool
>> > and should that pool have SVN::Pool::clear called on it that
>> > FileHandle
>> > will be closed out from under Git::temp_acquire.
>> >
>> > Since the only call site to Git::temp_acquire with the name 'svn_hash'
>> > is inside the close_file function, if an 'svn_hash' temp file is ever
>> > created its FileHandle is guaranteed to be created in the active
>> > SVN::Pool.
>> >
>> > This has not been a problem in the past because the SVN::Pool was not
>> > being cleared.  However, since dfa72fdb (git-svn: reload RA every
>> > log-window-size, v2.2.0) the pool has been getting cleared
>> > periodically
>> > at which point the FileHandle for the 'svn_hash' temp file gets
>> > closed.
>> > Any subsequent calls to Git::temp_acquire for 'svn_hash', however,
>> > succeed without creating/opening a new temporary file since it still
>> > has
>> > the now invalid FileHandle in its cache.  Callers that then attempt to
>> > use that FileHandle fail with an error.
>> >
>> > We avoid this problem by making sure the 'svn_hash' temp file is
>> > created
>> > in the same place the 'svn_delta_...' and 'git_blob_...' temp files
>> > are
>> > (and then temp_release'd) so that it can be safely used inside the
>> > close_file function without having its FileHandle end up in an
>> > SVN::Pool
>> > that gets cleared.
>> >
>> > Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
>> > ---
>> >  perl/Git/SVN/Fetcher.pm | 8 ++++++++
>> >  1 file changed, 8 insertions(+)
>> >
>> > diff --git a/perl/Git/SVN/Fetcher.pm b/perl/Git/SVN/Fetcher.pm
>> > index 10edb277..613055a3 100644
>> > --- a/perl/Git/SVN/Fetcher.pm
>> > +++ b/perl/Git/SVN/Fetcher.pm
>> > @@ -322,6 +322,14 @@ sub apply_textdelta {
>> >         # (but $base does not,) so dup() it for reading in close_file
>> >         open my $dup, '<&', $fh or croak $!;
>> >         my $base = $::_repository->temp_acquire("git_blob_${$}_
>> > $suffix");
>> > +       # close_file may call temp_acquire on 'svn_hash', but
>> > because of the
>> > +       # call chain, if the temp_acquire call from close_file ends
>> > up being the
>> > +       # call that first creates the 'svn_hash' temp file, then the
>> > FileHandle
>> > +       # that's created as a result will end up in an SVN::Pool
>> > that we clear
>> > +       # in SVN::Ra::gs_fetch_loop_common.  Avoid that by making
>> > sure the
>> > +       # 'svn_hash' FileHandle is already created before close_file
>> > is called.
>> > +       my $tmp_fh = $::_repository->temp_acquire('svn_hash');
>> > +       $::_repository->temp_release($tmp_fh, 1);
>> >
>> >         if ($fb->{blob}) {
>> >                 my ($base_is_link, $size);
>> > --
>>
>

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Kyle J. McKay]

On Feb 26, 2015, at 01:09, Nico Schl=F6mer wrote:
> All,
>
> I applied Kyle's latest patch
>
>> diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
>> index 622535e2..96888228 100644
>> --- a/perl/Git/SVN/Ra.pm
>> +++ b/perl/Git/SVN/Ra.pm
>> @@ -391,6 +391,7 @@ sub longest_common_path {
>> sub gs_fetch_loop_common {
>>        my ($self, $base, $head, $gsv, $globs) =3D @_;
>>        return if ($base > $head);
>> +       $::_repository->_open_cat_blob_if_needed;
>>        my $gpool =3D SVN::Pool->new_default;
>>        my $ra_url =3D $self->url;
>>        my $reload_ra =3D sub {
>
> alone and this fixes the bug for me. Thanks a lot, Kyle!

The updated patch with the additional fix is below.

There are two symptoms it addresses:

1) the 'Git_svn_hash_... bad file descriptor' failure

2) the 'out pipe went bad' failure

While (1) could probably also be addressed by Eric's alternate
"destroy all tempfiles when reloading RA" patch, that would require re-
opening all the temp files every 100 revisions (or whatever the log
window size is changed to).

I noticed that with the default log window size of 100 revisions, each
time the connection was reset at the 100-revision boundary (to reduce
the overall memory usage) it seemed to take approx. 3 seconds to start
up again processing revisions on that gmsh repository.  If you're
cloning 30000 revisions, that adds 15 minutes to the total clone time
already.  Admittedly opening new temp files will be a lot faster and
hardly noticeable, but doing that 300 times over the course of 30000
revisions will probably add at least a little extra delay and since
blowing away all the temp files does not seem to be necessary, why
incur the extra delay?

Also the "destroy all tempfiles when reloading RA" patch isn't going
to fix the cat_blob 'out pipe went bad' problem since that has nothing
to do with the temp files so another change will still be required for
that.

-Kyle

-- 8< --
Subject: [PATCH v2] Git::SVN::*: avoid premature FileHandle closure

Since b19138b (git-svn: Make it incrementally faster by minimizing temp
files, v1.6.0), git-svn has been using the Git.pm temp_acquire and
temp_release mechanism to avoid unnecessary temp file churn and provide
a speed boost.

However, that change introduced a call to temp_acquire inside the
Git::SVN::Fetcher::close_file function for an 'svn_hash' temp file.
Because an SVN::Pool is active at the time this function is called, if
the Git::temp_acquire function ends up actually creating a new
FileHandle for the temp file (which it will the first time it's called
with the name 'svn_hash') that FileHandle will end up in the SVN::Pool
and should that pool have SVN::Pool::clear called on it that FileHandle
will be closed out from under Git::temp_acquire.

Since the only call site to Git::temp_acquire with the name 'svn_hash'
is inside the close_file function, if an 'svn_hash' temp file is ever
created its FileHandle is guaranteed to be created in the active
SVN::Pool.

This has not been a problem in the past because the SVN::Pool was not
being cleared.  However, since dfa72fdb (git-svn: reload RA every
log-window-size, v2.2.0) the pool has been getting cleared periodically
at which point the FileHandle for the 'svn_hash' temp file gets closed.
Any subsequent calls to Git::temp_acquire for 'svn_hash', however,
succeed without creating/opening a new temporary file since it still has
the now invalid FileHandle in its cache.  Callers that then attempt to
use that FileHandle fail with an error.

We avoid this problem by making sure the 'svn_hash' temp file is created
in the same place the 'svn_delta_...' and 'git_blob_...' temp files are
(and then temp_release'd) so that it can be safely used inside the
close_file function without having its FileHandle end up in an SVN::Pool
that gets cleared.

Additionally the Git.pm cat_blob function creates a bidirectional pipe
FileHandle using the IPC::Open2::open2 function.  If that handle is
created too late, it also gets caught up in the SVN::Pool and incorrectly
closed by the SVN::Pool::clear call.  But this only seems to happen with
more recent versions of Perl and svn.

To avoid this problem we add an explicit call to _open_cat_blob_if_needed
before the first call to SVN::Pool->new_default to make sure the open2
handle does not end up in the SVN::Pool.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
---
 perl/Git/SVN/Fetcher.pm | 8 ++++++++
 perl/Git/SVN/Ra.pm      | 3 +++
 2 files changed, 11 insertions(+)

diff --git a/perl/Git/SVN/Fetcher.pm b/perl/Git/SVN/Fetcher.pm
index 10edb277..613055a3 100644
--- a/perl/Git/SVN/Fetcher.pm
+++ b/perl/Git/SVN/Fetcher.pm
@@ -322,6 +322,14 @@ sub apply_textdelta {
 	# (but $base does not,) so dup() it for reading in close_file
 	open my $dup, '<&', $fh or croak $!;
 	my $base = $::_repository->temp_acquire("git_blob_${$}_$suffix");
+	# close_file may call temp_acquire on 'svn_hash', but because of the
+	# call chain, if the temp_acquire call from close_file ends up being the
+	# call that first creates the 'svn_hash' temp file, then the FileHandle
+	# that's created as a result will end up in an SVN::Pool that we clear
+	# in SVN::Ra::gs_fetch_loop_common.  Avoid that by making sure the
+	# 'svn_hash' FileHandle is already created before close_file is called.
+	my $tmp_fh = $::_repository->temp_acquire('svn_hash');
+	$::_repository->temp_release($tmp_fh, 1);
 
 	if ($fb->{blob}) {
 		my ($base_is_link, $size);
diff --git a/perl/Git/SVN/Ra.pm b/perl/Git/SVN/Ra.pm
index 622535e2..32e2f19e 100644
--- a/perl/Git/SVN/Ra.pm
+++ b/perl/Git/SVN/Ra.pm
@@ -391,6 +391,9 @@ sub longest_common_path {
 sub gs_fetch_loop_common {
 	my ($self, $base, $head, $gsv, $globs) = @_;
 	return if ($base > $head);
+	# Make sure the cat_blob open2 FileHandle is created before calling
+	# SVN::Pool::new_default so that it does not incorrectly end up in the pool.
+	$::_repository->_open_cat_blob_if_needed;
 	my $gpool = SVN::Pool->new_default;
 	my $ra_url = $self->url;
 	my $reload_ra = sub {
--

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

"Kyle J. McKay" <mackyle@gmail.com> wrote:
> The updated patch with the additional fix is below.

Thanks, signed-off and pushed to master on git://bogomips.org/git-svn
I've dropped my "destroy all tempfiles..." patch.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Junio C Hamano]

Eric Wong <normalperson@yhbt.net> writes:

> "Kyle J. McKay" <mackyle@gmail.com> wrote:
>> The updated patch with the additional fix is below.
>
> Thanks, signed-off and pushed to master on git://bogomips.org/git-svn
> I've dropped my "destroy all tempfiles..." patch.

I think I missed this exchange.  I see these two patches:

    e0b4cad Git::SVN::*: avoid premature FileHandle closure
    ce1b57b git-svn: fix localtime=true on non-glibc environments

on the 'master' branch of your git://git.bogomips.org/git-svn.git/
repository.  Is this a good time to pull from you, and should I
expect more during this cycle (I am guessing that the answers would
be yes and no, but just to make sure...)?

Thanks.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Junio C Hamano <gitster@pobox.com> wrote:
> Eric Wong <normalperson@yhbt.net> writes:
> > "Kyle J. McKay" <mackyle@gmail.com> wrote:
> >> The updated patch with the additional fix is below.
> >
> > Thanks, signed-off and pushed to master on git://bogomips.org/git-svn
> > I've dropped my "destroy all tempfiles..." patch.
> 
> I think I missed this exchange.  I see these two patches:
> 
>     e0b4cad Git::SVN::*: avoid premature FileHandle closure
>     ce1b57b git-svn: fix localtime=true on non-glibc environments
> 
> on the 'master' branch of your git://git.bogomips.org/git-svn.git/
> repository.  Is this a good time to pull from you, and should I
> expect more during this cycle (I am guessing that the answers would
> be yes and no, but just to make sure...)?

Correct, it's a good time to pull and I don't expect to have more time
to work on new features (lazy load) for a bit.  I was under the
impression you already pulled:

http://mid.gmane.org/xmqq61aol44q.fsf@gitster.dls.corp.google.com

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Junio C Hamano]

Eric Wong <normalperson@yhbt.net> writes:

> Correct, it's a good time to pull and I don't expect to have more time
> to work on new features (lazy load) for a bit.  I was under the
> impression you already pulled:
>
> http://mid.gmane.org/xmqq61aol44q.fsf@gitster.dls.corp.google.com

Indeed I did.  I was fooled by my own rebase X-<.

Thanks, then I am fully up to date with you.

git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Mike]

Similar to the issue mintywalker originally mentioned on Jan 8th 2015, 
during a "git svn clone" I get a Bad File Descriptor error using:

git-svn version 2.2.2 (svn 1.8.8) on Ubuntu 14.04.

<snip>
r460 = 456377de3906d689c56e51af842e18abe086a980 (refs/remotes/origin/trunk)
         A       client/binary/App_Client_v2.1.2.exe
r461 = 36848dbb7f417da2e381b61b68ff7b0d22a5bf7f (refs/remotes/origin/trunk)
write .git/Git_svn_hash_0WWL4a: Bad file descriptor
  at /usr/lib/perl5/SVN/Ra.pm line 623.


$ svn diff --diff-cmd diff -c 461
Index: client/binary/App_Client_v2.1.2.exe
===================================================================
Cannot display: file marked as a binary type.
svn:mime-type = application/octet-stream
Index: client/binary/App_Client_v2.1.2.exe
===================================================================
--- client/binary/App_Client_v2.1.2.exe    (revision 0)
+++ client/binary/App_Client_v2.1.2.exe    (revision 461)

Property changes on: client/binary/App_Client_v2.1.2.exe
___________________________________________________________________
Added: svn:mime-type
## -0,0 +1 ##
+application/octet-stream
\ No newline at end of property


Not sure if it helps or not, but here is the commit immediately after 
that one:

$ svn diff --diff-cmd diff -c 462
Index: interface/help/App_Client.exe
===================================================================
--- interface/help/App_Client.exe  (revision 0)
+++ interface/help/App_Client.exe  (revision 462)
@@ -0,0 +1 @@
+link ../../client/binary/App_Client_v2.1.2.exe
\ No newline at end of file

Property changes on: interface/help/App_Client.exe
___________________________________________________________________
Added: svn:special
## -0,0 +1 ##
+*
\ No newline at end of property

Unfortunately the repository is private, but it seems like a pretty 
simple commit that is causing the problem?

If you need more information please let me know.

Thanks.


-- 
Mike

git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Minty]

I appear to have hit a bug (or have data that the code fails on) while
importing an SVN repo to git.

I'm wondering if there is anything I can do to help find / fix the
cause, given I appear to have a fail-case.

Sadly I can't supply the SVN repo as the code is private.

What I did:

git svn clone https://www.example.com/dshfkjhfsjhsdkjfhsdf/nameofrepo

What I'm running:

MacOS Yosemite 10.10.1 (14B25)

§ git --version
git version 2.2.1

Built via MacPorts using: sudo port install git +svn

and updated today to the latest available.

The process was running for a few minutes and does appear to have
imported a lot.  The last few lines of output, including the error
(with paths/names anonymised)

r869 = 9823c89bbdfa9d51aeb0a16c539049ae96nd5d62 (refs/remotes/git-svn)
    D    path/to/stuff/Example1.java
    D    path/to/stuff/Example2.java
W: -empty_dir: path/to/stuff/Example1.java
W: -empty_dir: path/to/stuff/Example2.java
r870 = b1f06434b0b2f37a11be2ee5dfc6175bjs348545 (refs/remotes/git-svn)
write .git/Git_svn_hash_BmjclS: Bad file descriptor
 at /opt/local/lib/perl5/vendor_perl/5.16.3/darwin-thread-multi-2level/SVN/Ra.pm
line 623.

Any advice / pointers would be welcome -- I'd be happy to run any
tests & I'm reasonably comfortable coding in Perl so happy to poke
around where I can.

I've been using git for year (thanks! it rocks) and hoping I can avoid
having to (re)learn too much about SVN.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Minty]

On Thu, Jan 8, 2015 at 12:43 PM, Minty <mintywalker@gmail.com> wrote:
> git svn clone https://www.example.com/dshfkjhfsjhsdkjfhsdf/nameofrepo
>
> r869 = 9823c89bbdfa9d51aeb0a16c539049ae96nd5d62 (refs/remotes/git-svn)
>     D    path/to/stuff/Example1.java
>     D    path/to/stuff/Example2.java
> W: -empty_dir: path/to/stuff/Example1.java
> W: -empty_dir: path/to/stuff/Example2.java
> r870 = b1f06434b0b2f37a11be2ee5dfc6175bjs348545 (refs/remotes/git-svn)
> write .git/Git_svn_hash_BmjclS: Bad file descriptor
>  at /opt/local/lib/perl5/vendor_perl/5.16.3/darwin-thread-multi-2level/SVN/Ra.pm
> line 623.

fyi - tried the same repo, same command under Ubuntu 14.04, using the
stock git-svn deb & it worked.

happy to help try debugging the problem on MacOS if anyone wants to
give me a pointer, but otherwise I'm good.

Thanks.

Re: git svn import failure : write .git/Git_svn_hash_BmjclS: Bad file descriptor

[Eric Wong]

Minty <mintywalker@gmail.com> wrote:
> § git --version
> git version 2.2.1

What about "git svn --version" ?
(it'll display the SVN binding version, too)

> Any advice / pointers would be welcome -- I'd be happy to run any
> tests & I'm reasonably comfortable coding in Perl so happy to poke
> around where I can.

Great to know.  Unfortunately, I no longer have access to the repo where
I encountered the problem and haven't been able to reproduce it again,
either.

The first step is to find a (preferably) fast way to reproduce the issue
consistently.  If you can isolate which revisions and change patterns in
your private repo cause it, it should be possible to create a public
repo without confidential data in it which reproduces the problem.