This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH][BZ #13601] Retry read in ld.so if the entire ELF header is not read in
On 20 Oct 2012, Rich Felker spake thusly:
> On Sat, Oct 20, 2012 at 10:17:44AM -0700, Roland McGrath wrote:
>> The bug actually doesn't make it clear whether the bug was observed or
>> purely theoretical. But it is entirely kosher for any filesystem to behave
>> this way, so the fix is correct even if the bug has yet to be observed in
>> practice.
>
> Agreed. I didn't intend to imply that it shouldn't be fixed, just that
> it's probably purely-theoretical, and that some trickery would be
> required to test the fix.
I've seen it in the real world -- not with ld.so, true, but with both
read() and write(), on NFS with intr set, and with Solaris NFS clients
even without anything analogous set. (Yes, I know, NFS with intr set is
considered bad, but it's only bad because of the existence of software
that assumes that disk I/O will never -EINTR or return short reads.
I now consider this to be clearly an incorrect assumption.)
--
NULL && (void)