git name-rev segfault

git name-rev segfault

[Tamas Papp]

hi All,


We ran into a segfault with all version of command line git.

We are able to reproduce this issue by this way:


Generate 100k file into a repository:

#!/bin/bash

rm -rf .git test.file
git init
git config user.email a@b
git config user.name c

time for i in {1..100000}
do
   [ $((i % 2)) -eq 1 ] && echo $i>test.file || echo 0 >test.file
   git add test.file

   git commit -m "$i committed"

done


Run git on it:

$ git name-rev a20f6989b75fa63ec6259a988e38714e1f5328a0



Could you coment on it?


Thank you,

tamas

Re: git name-rev segfault

[Jeff King]

On Mon, Jul 29, 2019 at 04:19:47PM +0200, Tamas Papp wrote:

> Generate 100k file into a repository:
> 
> #!/bin/bash
> 
> rm -rf .git test.file
> git init
> git config user.email a@b
> git config user.name c
> 
> time for i in {1..100000}
> do
>   [ $((i % 2)) -eq 1 ] && echo $i>test.file || echo 0 >test.file
>   git add test.file
> 
>   git commit -m "$i committed"
> 
> done

I lost patience kicking off two hundred thousand processes. Try this:

  for i in {1..100000}
  do
    echo "commit HEAD"
    echo "committer c <a@b> $i +0000"
    echo "data <<EOF"
    echo "$i committed"
    echo "EOF"
    echo
  done | git fast-import

which runs much faster. This doesn't change any files in each commit,
but I don't think it's necessary for what you're showing (name-rev
wouldn't ever look at the trees).

> Run git on it:
> 
> $ git name-rev a20f6989b75fa63ec6259a988e38714e1f5328a0

Anybody who runs your script will get a different sha1 because of the
change in timestamps. I guess this is HEAD, though. I also needed to
have an actual tag to find. So:

  git tag old-tag HEAD~99999
  git name-rev HEAD

segfaults for me.

> Could you coment on it?

This is a known issue. The algorithm used by name-rev is recursive, and
you can run out of stack space in some deep cases. There's more
discussion this thread:

  https://public-inbox.org/git/6a4cbbee-ffc6-739b-d649-079ba01439ca@grubix.eu/

including some patches that document the problem with an expected
failure in our test suite. Nobody has actually rewritten the C code yet,
though.

-Peff

Re: git name-rev segfault

[Tamas Papp]

Thanks for letting me know and for the corrections too.


Cheers,

tamas


On 7/29/19 9:50 PM, Jeff King wrote:
> On Mon, Jul 29, 2019 at 04:19:47PM +0200, Tamas Papp wrote:
>
>> Generate 100k file into a repository:
>>
>> #!/bin/bash
>>
>> rm -rf .git test.file
>> git init
>> git config user.email a@b
>> git config user.name c
>>
>> time for i in {1..100000}
>> do
>>    [ $((i % 2)) -eq 1 ] && echo $i>test.file || echo 0 >test.file
>>    git add test.file
>>
>>    git commit -m "$i committed"
>>
>> done
> I lost patience kicking off two hundred thousand processes. Try this:
>
>    for i in {1..100000}
>    do
>      echo "commit HEAD"
>      echo "committer c <a@b> $i +0000"
>      echo "data <<EOF"
>      echo "$i committed"
>      echo "EOF"
>      echo
>    done | git fast-import
>
> which runs much faster. This doesn't change any files in each commit,
> but I don't think it's necessary for what you're showing (name-rev
> wouldn't ever look at the trees).
>
>> Run git on it:
>>
>> $ git name-rev a20f6989b75fa63ec6259a988e38714e1f5328a0
> Anybody who runs your script will get a different sha1 because of the
> change in timestamps. I guess this is HEAD, though. I also needed to
> have an actual tag to find. So:
>
>    git tag old-tag HEAD~99999
>    git name-rev HEAD
>
> segfaults for me.
>
>> Could you coment on it?
> This is a known issue. The algorithm used by name-rev is recursive, and
> you can run out of stack space in some deep cases. There's more
> discussion this thread:
>
>    https://public-inbox.org/git/6a4cbbee-ffc6-739b-d649-079ba01439ca@grubix.eu/
>
> including some patches that document the problem with an expected
> failure in our test suite. Nobody has actually rewritten the C code yet,
> though.
>
> -Peff