1.7 installation failed (on network drive?)

[Thomas Wolff]

Corinna Vinschen schrieb:
> Thomas,
>
> On Dec  7 17:09, Thomas Wolff wrote:
>   
>> I've tested the network installation with last Saturday's update of
>> setup-1.7.exe and unfortunately, the problem remains.
>> However, I can add a screen log after calling setup from mintty for
>> the case that setup cannot even store its log anywhere (attached).
>> Don't know if that helps. Apparently setup-1-7.exe cannot create
>> directories (including the top package directory) on that network
>> NFTS drive, although all other tools can easily create them.
>> Despite cgf's point this should not delay the 1.7.1 release, may I
>> suggest once more (and I'm not adding this to the "1.7.1 release
>> date" thread) that while the release is being delayed anyway due to
>> other discussions, we might as well try to resolve this. Personally,
>> I can do with a workaround; I'm just trying to help avoid trouble
>> with other users complaining after "the release"...
>> Thomas
>>     
>
> I just had another look into this and unfortunately I still can't
> reproduce it.  What I see in your log output is the fact that ...
> hang on ...
>
>   
>> io_stream_cygfile: fopen(/etc/setup/timestamp) failed 13 Permission denied
>> io_stream_cygfile: fopen(/var/run/utmp) failed 13 Permission denied
>> io_stream_cygfile: fopen(/etc/setup/alternatives.lst.gz) failed 13 Permission denied
>> [etc]
>>     
>
> So the fopen calls fail.  fopen in setup is actually a call to nt_wfopen
> in filemanip.cc.  If you could take a look, it's not a very complicated
> function.  The general idea is to call NtCreateFile with
> FILE_OPEN_FOR_BACKUP_INTENT rights to allow an admin to install without
> permission trouble.
>
> As you can see, the error handling (lines 466ff) is somewhat
> oversimplified.  You get a "Permission denied" as a fallback error
> if none of the path-related errors is triggered.  Unfortunately
> we don't know the exact status code returned by NtCreateFile and
> I have no real idea what the problem might be.
>
> What we need is this:  Build a debug version of setup (you need gcc-3
> for that since the -mno-cygwin option is still used), start it under
> GDB, and set a breakpoint to filemanip.cc:468.  GDB will break there if
> an error gets triggered.  Examine what status codes are returned.
> See /usr/include/w32api/ddk/ntstatus.h what status code that is.
>   
OK; since setup is not a cygwin package, where would I get the sources 
in the first place? And is there a README telling me how to build a 
debug version (make debug?)?
Then I assume I'd use the gdb commands break and print (sorry, not 
familiar with gdb)?

In case I don't get it compiled, maybe you could prepare two debug 
versions (with/w/o the change below) and send them to me? I could take 
an extra trip to that lab tomorrow for testing.

Thomas

> That might give us some clue.  The only vague idea what you could try
> else is to remove the FILE_OPEN_REPARSE_POINT flag from the NtCreateFile
> call and see if it works without that.
>