Re: [PATCH] fdopen in read -e -u N

[Chet Ramey]

On 5/17/24 4:01 AM, Grisha Levit wrote:
> Looks like `read -e -u N' creates (and leaks) a new stream for fd N.
> If a second `read' tries to fdopen the same fd once more, Android's
> FDSAN (on and enforcing defualt) causes the process to abort:

I suppose this has value as a warning, but causing an abort seems ...
overzealous.

> $ bash -c 'read -eu3; read -eu3' 3<&0
> fdsan: failed to exchange ownership of file descriptor:
> fd 3 is owned by FILE* 0x7131a45d18, was expected to be unowned
> Aborted                    bash -c 'read -eu3; read -eu3' 3<&0
>
> I guess when Readline is being used for `read' input, a proper thing
> to do is dup() the fd supplied by -u (thereby defeating the point of
> the option..) and fclose() the resulting stream when done.

Yes, I suppose that avoids the android crash. It would be nice if there
were a way to deallocate the FILE * without touching the underlying file
descriptor (kind of the opposite of fdopen) or if there were some
equivalent of fileno() that could be used as an lvalue, so you could
set the fd in a FILE * to -1 and just call fclose() to deallocate it.

The dup2 is what I was trying to avoid, and accepted the small memory leak
as a consequence. But stdio has only a limited understanding of non-stdio
managed file descriptors.

> I think the current behavior is due to a report[1] about -u being
> silently ignored with -e, though maybe forbidding the combination
> explicitly would be fine too.

No, that would run counter to user expectations.

--
``The lyf so short, the craft so long to lerne.'' - Chaucer
                 ``Ars longa, vita brevis'' - Hippocrates
Chet Ramey, UTech, CWRU    chet@case.edu    http://tiswww.cwru.edu/~chet/

OpenPGP_signature.asc
Description: OpenPGP digital signature