RE: FW: [cygwin] Cygwin's git says "error: failed to read delta-pack base object"

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



TLDR = Cygwin remote filesystem sometimes has strange failures -> workaround is to use rename, not link/unlink; 
see https://github.com/pdinc-oss/git/commit/5a36824ed01d4335148ca3846e75cc99c11650e2
> -----Original Message-----
> From: Jason Pyeron
> Sent: Friday, December 05, 2014 10:30
> 
> > -----Original Message-----
> > From: Jason Pyeron
> > Sent: Thursday, December 04, 2014 16:34
> > 
> > > -----Original Message-----
> > > From: brian m. carlson
> > > Sent: Wednesday, December 03, 2014 19:55
> > > 
> > > On Wed, Dec 03, 2014 at 06:31:18PM -0500, Jason Pyeron wrote:
> > > > I remember hitting this a while ago, but just gave up.
> > > > 
> > > > It seems to be a problem for others too.
> > > > 
> > > > Any ideas on how to debug this so it can be patched?
> > > > 
> > > > -----Original Message-----
> > > > From: Dave Lindbergh
> > > > Sent: Wednesday, December 03, 2014 18:07
> > > > To: cygwin
> > > > 
> > > > Aha - you're right.
> > > > 
> > > > It works fine on a local NTFS volume.
> > > > 
> > > > I get the error when I do it on Z:, which is mapped to a 
> > > network drive
> > > > (on another Windows box).
> > > > 
> > > > Is there a workaround? Why does this happen?

I have a really hacky workaround, commit 5a36824ed01d4335148ca3846e75cc99c11650e2 comments out the logic and forces a rename instead of link and unlink.

https://github.com/pdinc-oss/git/tree/cygwin-issue-remoteCIFS-rename

<snip/>

> Pseudo code and observations
> ./sha1_file.c:write_loose_object(sha1)
> {
>  filename=sha1_file_name(sha1)
>  (fd,tmp_file)=create_tmpfile(filename)
>  write_buffer(fd)
>  close_sha1_file(fd)
>  return move_temp_to_file(tmp_file, filename)
> }
> 
> move_temp_to_file(tmpfile, filename)
> {
>  // I am thinking about forcing renames to see if the problem 
> exists then as well
>  // if that "works" then a per repo config option allowing 
> for forced renames
>  if (OBJECT_CREATION_USES_RENAMES) goto try_rename
>  else if link(tmpfile,filename)
> 

Dave has tested a build I made for him on 64 bit Cygwin and it works. I no longer have access to the environment I was having this problem in last February. I will try to investigate this further, but I am not hopeful, maybe Corinna will have luck on the issue. But I was in a secure corporate environment and I thought the host based security system (AV), coupled with the remote file system was causing the problem, namely files created are not available instantly.

I do think that we should have a config option for this, as most users who could encounter such a problem are not likely to be able (or allowed) to rebuild the git executable themselves.

<snip/>

> > -----Original Message-----
> > From: Corinna Vinschen
> > Sent: Friday, December 05, 2014 6:35
> > To: cygwin@xxxxxxxxxx
> <snip/>
> > What I found in the strace is this:
> > 
> > - Create file Z:\pic32mx-bmf\.git\objects\30\tmp_obj_YljwNZ
> > 
> > - open file, write something, close file.
> > 
> > - link (Z:\pic32mx-bmf\.git\objects\30\tmp_obj_YljwNZ,
> > 	
> > 
> Z:\pic32mx-bmf\.git\objects\30\0bdeb2fd209d24afb865584da10b78aa8fefc4)
> >   succeeds.
> > 
> > - unlink (Z:\pic32mx-bmf\.git\objects\30\tmp_obj_YljwNZ) succeeds
> > 
> > - Trying to open
> >   
> > 
> Z:\pic32mx-bmf\.git\objects\30\0bdeb2fd209d24afb865584da10b78aa8fefc4
> >   but the file doesn't exist and NtCreateFile fails with status
> >   0xC0000034, STATUS_OBJECT_NAME_NOT_FOUND --> ENOENT.
> > 
> > - Subsequent unlink (Z:\pic32mx-bmf\.git\objects\30) fails with a
> >   STATUS_DIRECTORY_NOT_EMPTY --> ENOTEMPTY.
> > 
> > - git seems to be prepared for such a case, the parent process calls
> >   opendir/readdir on the directory.  Enumerating the files in
> >   Z:\pic32mx-bmf\.git\objects\30 shows the entries ".", ".." and
> >   "0bdeb2fd209d24afb865584da10b78aa8fefc4".
> > 
> > - Then git calls lstat on the file, which results in NtOpenFile
> >   returning status STATUS_OBJECT_NAME_NOT_FOUND again.
> > 
> > - From a POSIX POV this means "somebody else" deleted the file,
> >   so the dir is empty now.  Git tries to delete the directory again,
> >   which again results in STATUS_DIRECTORY_NOT_EMPTY --> ENOTEMPTY
> >   and, internally, a sharing violation which disallows to move the
> >   directory out of the way.
> > 
> > This looks suspiciously like a bug in the remote filesystem.  Link
> > succeeded, so there are two links to the same file in the directory.
> > Unlinking link 1 succeeds, so there's still one link to the 
> > file in the
> > directory, but link 2 is inaccessible as if the file has 
> been deleted
> > completely.  Thus, a full POSIX git on this drive is broken.
> > 
> > Can you please run
> > 
> >   /usr/lib/csih/getVolInfo /cygdrive/z
> > 
> > and paste the output here?  Maybe I can workaround this in the next
> > Cygwin version.

Dave's response: https://www.cygwin.com/ml/cygwin/2014-12/msg00066.html

--
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
-                                                               -
- Jason Pyeron                      PD Inc. http://www.pdinc.us -
- Principal Consultant              10 West 24th Street #100    -
- +1 (443) 269-1555 x333            Baltimore, Maryland 21218   -
-                                                               -
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
This message is copyright PD Inc, subject to license 20080407P00. 

--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [Linux Kernel Development]     [Gcc Help]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [V4L]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Fedora Users]