Thank you very much!
Ger Hobbelt wrote:
On Thu, Aug 27, 2009 at 2:24 PM, Laura
Arhire<laura.arh...@endion-software.com> wrote:
Hey
I'm using the BIO abstraction for reading/writing to sockets - a small part
of the BIO_read method is unclear: For a blocking socket, will the BIO_read
call block until the length provided in the call is filled in the buffer,
or will it return as soon as it managed to read anything from the socket?
I see that the underlying implementation uses recv() - recv blocks until
something is available: when something is available for reading, read up to
len bytes and copy it to the provided buffer, then return, i.e. the length
is used as a maximum value. From reading the openssl code I think that
BIO_read behaves in the same way as recv - which means that if I want to
read a specific number of bytes (no more, no less), I need to loop until I
get everything I need. Can anyone confirm this ?
Confirmed.
BIO_read won't wait/guarantee that the requested number of bytes are
delivered all, so you'll indeed need to loop to get them all (or an
error due to connection issues) if you want to receive a predetermined
number of bytes.
For completeness / additive: note that 'officially', you should check
the BIO_should_retry() call on error (negative error code return),
i.e.
---snip[stripped]---
...
i = BIO_read(bio,buf,bufsize);
if (i < 0)
{
// error condition - possibly
if (BIO_should_retry(bio))
{
if (BIO_should_read(bio))
...
if (BIO_should_write(bio))
...
}
else
{
// real error. process the error the way you want/need
ERR_print_errors(bio_err);
got fail_dramatically;
}
}
else if (i == 0)
{
// blocking --> conn termination
...
break;
}
else
{
// collect the returned bytes; shift buf, etc., then wait for the rest
...
}
---snip---
as this is what it should look like for any and all BIOs, i.e. when
your BIO-using code layer should not be specifically aware of the
chain/device peculiarities accessed through the BIO abstraction layer.
In the case of raw socket I/O, the 'should_retry' won't fire, at least
not with the current implementation on UNIX. Which might be a bother
regarding testability versus 'correct' use/implementation of the i/o
abstraction. Once you're stacking, say, and SSL BIO into this BIO
chain however, things will certainly look /very/ different and you'll
surely need that should_retry/etc. code in there.
Take care,
Ger
--
Laura
