Thread: Re: [PATCHES] Bad bug in fopen() wrapper code

"Magnus Hagander" <mha@sollentuna.net> writes:
> Attached patch sets the  O_CREAT option when appending to files.

That looks correct, but I went looking to see if there were any other
mistakes of the same ilk, and I'm wondering what the sense is in
openFlagsToCreateFileFlags ... seems like it's ignoring O_EXCL in
some combinations and not others.  Is that right?

            regards, tom lane

> > Attached patch sets the  O_CREAT option when appending to files.
>
> That looks correct, but I went looking to see if there were
> any other mistakes of the same ilk, and I'm wondering what
> the sense is in openFlagsToCreateFileFlags ... seems like
> it's ignoring O_EXCL in some combinations and not others.  Is
> that right?

That is part of the original open() code that Claudio did back for 8.0,
so it has definitly been working since then. I haven't really read into
the code, though... But a qiuck look doesn't show me any place wher eit
does ignore O_EXCL - which combination would that be?

//Magnus

"Magnus Hagander" <mha@sollentuna.net> writes:
> That is part of the original open() code that Claudio did back for 8.0,
> so it has definitly been working since then.

Hm, maybe best not to touch it, but still...

> I haven't really read into
> the code, though... But a qiuck look doesn't show me any place wher eit
> does ignore O_EXCL - which combination would that be?

What's bugging me is that 0 and O_EXCL give the same answer, and
O_TRUNC and O_TRUNC | O_EXCL give the same answer, but O_CREAT and
O_CREAT | O_EXCL give different answers, as do O_CREAT | O_TRUNC
and O_CREAT | O_TRUNC | O_EXCL.  I'm also pretty suspicious of
both O_CREAT | O_EXCL and O_CREAT | O_TRUNC | O_EXCL giving the
same answer.  However, I have no idea what the semantics are of
the symbols the function is mapping into, so maybe it's OK.

            regards, tom lane

Tom Lane wrote:
> "Magnus Hagander" <mha@sollentuna.net> writes:
> > That is part of the original open() code that Claudio did back for 8.0,
> > so it has definitly been working since then.
> 
> Hm, maybe best not to touch it, but still...
> 
> > I haven't really read into
> > the code, though... But a qiuck look doesn't show me any place wher eit
> > does ignore O_EXCL - which combination would that be?
> 
> What's bugging me is that 0 and O_EXCL give the same answer, and
> O_TRUNC and O_TRUNC | O_EXCL give the same answer, but O_CREAT and
> O_CREAT | O_EXCL give different answers, as do O_CREAT | O_TRUNC
> and O_CREAT | O_TRUNC | O_EXCL.  I'm also pretty suspicious of
> both O_CREAT | O_EXCL and O_CREAT | O_TRUNC | O_EXCL giving the
> same answer.  However, I have no idea what the semantics are of
> the symbols the function is mapping into, so maybe it's OK.

I am CC'ing Claudio Natoli on this question about
open.c::openFlagsToCreateFileFlags(), but in looking at the code, it
seems OK because:
o O_EXCL doesn't have any meaning unless O_CREAT is used
o O_TRUNC has no meaning when O_CREAT | O_EXCL are usedbecause you are guaranteed to be creating a new file

Claudio, the function is here, at the top of the file:
http://developer.postgresql.org/cvsweb.cgi/pgsql/src/port/open.c?rev=1.18;content-type=text%2Fx-cvsweb-markup

--  Bruce Momjian  <bruce@momjian.us>          http://momjian.us EnterpriseDB
http://www.enterprisedb.com
 + If your life is a hard drive, Christ can be your backup. +