Re: [Bug-wget] gnutls_handshake can return GNUTLS_E_INTERRUPTED
Thanks, Daniel.
I believe, we already handle that case.
If GNUTLS_E_INTERRUPTED occurs, we restart the loop and re-enter
gnutls_handshake(). This happens for all non-fatal errors.
That is the while part
while (err && gnutls_error_is_fatal (err) == 0);
Tim
On Thursday 30 June 2016 15:18:18 Daniel Stenberg wrote:
> Hello,
>
> gnutls_handshake() is documented to possibly return GNUTLS_E_INTERRUPTED as
> well as GNUTLS_E_AGAIN and should probably behave similarly for both return
> codes within wget.
>
> diff --git a/src/gnutls.c b/src/gnutls.c
> index 63c7c33..44c497b 100644
> --- a/src/gnutls.c
> +++ b/src/gnutls.c
> @@ -463,11 +463,11 @@ _do_handshake (gnutls_session_t session, int fd,
> double timeout)
> /* We don't stop the handshake process for non-fatal errors */
> do
> {
> err = gnutls_handshake (session);
>
> - if (timeout && err == GNUTLS_E_AGAIN)
> + if (timeout && ((err == GNUTLS_E_AGAIN) || (err ==
> GNUTLS_E_INTERRUPTED))) {
> if (gnutls_record_get_direction (session))
> {
> /* wait for writeability */
> err = select_fd (fd, timeout, WAIT_FOR_WRITE);
signature.asc
Description: This is a digitally signed message part.
Re: [Bug-wget] gnutls_handshake can return GNUTLS_E_INTERRUPTED
On Thu, 30 Jun 2016, Tim Ruehsen wrote: I believe, we already handle that case. If GNUTLS_E_INTERRUPTED occurs, we restart the loop and re-enter gnutls_handshake(). This happens for all non-fatal errors. Right, but then it won't wait for the socket and just busy-loop. The gnutls_record_get_direction man page[*] seems to suggest that waiting for the socket is the appropriate way: This function provides information about the internals of the record protocol and is only useful if a prior gnutls function call (e.g. gnutls_handshake()) was interrupted for some reason, that is, if a function returned GNUTLS_E_INTERRUPTED or GNUTLS_E_AGAIN. In such a case, you might want to call select() or poll() before calling the interrupted gnutls function again [*] = http://linux.die.net/man/3/gnutls_record_get_direction -- / daniel.haxx.se
Re: [Bug-wget] gnutls_handshake can return GNUTLS_E_INTERRUPTED
On Thursday 30 June 2016 17:18:36 Daniel Stenberg wrote: > On Thu, 30 Jun 2016, Tim Ruehsen wrote: > > I believe, we already handle that case. If GNUTLS_E_INTERRUPTED occurs, we > > restart the loop and re-enter gnutls_handshake(). This happens for all > > non-fatal errors. > > Right, but then it won't wait for the socket and just busy-loop. The > gnutls_record_get_direction man page[*] seems to suggest that waiting for > the socket is the appropriate way: > > This function provides information about the internals of the record > protocol and is only useful if a prior gnutls function call (e.g. > gnutls_handshake()) was interrupted for some reason, that is, if a function > returned GNUTLS_E_INTERRUPTED or GNUTLS_E_AGAIN. In such a case, you might > want to call select() or poll() before calling the interrupted gnutls > function again > > [*] = http://linux.die.net/man/3/gnutls_record_get_direction GNUTLS_E_INTERRUPTED is returned if the process got a signal that interrupts blocking I/O (for wget this is SIGUSR1 and SIGHUP). If it is just *one* signal (very likely), a successive call to gnutls_handshake() would not return again with this value. If the process gets 'hammered' by signals, we have a busy-loop. But with your change, the poll/wait would also immediate return with errno=EINTR - and we have a busy-loop as well. So from what I can judge, your change adds 'complexity' without any gain. Tim signature.asc Description: This is a digitally signed message part.
Re: [Bug-wget] gnutls_handshake can return GNUTLS_E_INTERRUPTED
On Fri, 1 Jul 2016, Tim Ruehsen wrote: GNUTLS_E_INTERRUPTED is returned if the process got a signal that interrupts blocking I/O (for wget this is SIGUSR1 and SIGHUP). If it is just *one* signal (very likely), a successive call to gnutls_handshake() would not return again with this value. If the process gets 'hammered' by signals, we have a busy-loop. But with your change, the poll/wait would also immediate return with errno=EINTR - and we have a busy-loop as well. So from what I can judge, your change adds 'complexity' without any gain. You're free to do what you want with it. I think it follows the spirit of the documentation. -- / daniel.haxx.se
