On Mon, Feb 11, 2013 at 5:15 PM, Dave Thompson <dthomp...@prinpay.com> wrote:
>> From: owner-openssl-us...@openssl.org On Behalf Of Jeffrey Walton
>> Sent: Monday, 11 February, 2013 01:12
>
>> I'm trying to memory map a file. Does OpenSSL BIO's allow this?
>>
>> I seem to be failing on BIO_read_filename with a `ret` of 0. The docs
>> state I should expect `ret` of 1. When I look for the problem, I'm
>> getting an error code of 0 (`err`, success).
>>
> (BIO_read_filename = BIO_ctrl _SET_FILENAME on a mem-BIO)
> That doesn't make sense. OpenSSL doesn't map a file itself;
> that isn't portable.
I gave up on trying to pipeline the data using a source/sink paradigm.
For what its worth, it does not matter to me how the library
internally handles sources/sinks. As a dumb user, I just expect them
to work :)
> If you *already* have a file mapped into
> memory by e.g. mmap in Unix and something different in Windows,
> you can create a *readonly* mem-BIO on that memory with
> BIO_new_mem_buf. I don't think you can do this for write;
> writing to a mem-BIO must be able to re-allocate the underlying
> buffer when it thinks necessary, and that can't work for a file
> mapped to a fixed area of memory.
Thanks. Once I had to turn to C for some of the stuff, I just used C
for all of the stuff.
The deal breaker was lack of a way to determine a file size when using
a file BIO. I could not seek to the end and then do a tell() via
BIO_C_GET_FILE_PTR because seek() with {0, SEEK_END} was missing. I
even tried sending a BIO_ctrl (BIO_C_SET_FILE_PTR) with {0, SEEK_END}
parameters, which resulted in a crash.
> BIO_ctrl of an option inapplicable to mem-BIO returns 0 but doesn't
> set anything in the error queue. Yes, this isn't too helpful.
Gotcha. I'm used to having error information available on failure
(GetLastError(), errno, etc). I assumed error information was/is
because that's what APIs usually do.
Perhaps the docs should state error information on failure is not available.
>> I ask because the doc's appear to be incorrect on the return type of
>> BIO_read_filename. When I try and use an `int`, I get a size warning
>> (long instead of int). So I suspect the docs may be wrong.
>> http://www.openssl.org/docs/crypto/BIO_s_file.html.
>>
> I suspect at some point in the distant past BIO_read_filename was
> a separate routine and did return int. When it was combined into
> the more general BIO_ctrl/file_ctrl mechanism, that handles many
> options some of which need to return long, and the manpage probably
> wasn't updated.
Gotcha, thanks.
I wish I had "Network Security with OpenSSL" with me. It would have
saved me a lot of time since I/O is covered in detail. It would have
saved you too :)
Jeff
______________________________________________________________________
OpenSSL Project http://www.openssl.org
User Support Mailing List openssl-users@openssl.org
Automated List Manager majord...@openssl.org
