[openssl-users] Openssl asynchronous operation in real network

Ananthu Unnikrishnan ananthuu741 at gmail.com
Wed Jan 16 05:36:08 UTC 2019


Hi Matt,

        Thanks for your update. Will try to implement one scheme using the
same.

        I have one doubt regarding the points which we have discussed
earlier.

        I am now facing an issue of getting an invalid waitctx address
while trying to get it using ASYNC_get_wait_ctx(). I have checked using
gdb, in which got memory inaccessible print when I tried to dump
the waitctx structure. If I put some delay in the engine, things are
working fine, ie. no corruption of waitctx structure address.

        But I am just copying the address of job structure before the
ASYNC_pause().  After the HW operation is completed I will use this address
to get waitctx and fd using ASYNC_get_wait_ctx() and
ASYNC_WAIT_CTX_get_fd() from the thread which  I have created from the
engine. I haven't access the job address directly from the thread.

        I know you have already told me that I should be careful while
handling ASYNC_JOB structure from outside threads. But I have only a little
experience on handling these ASYNC_JOB structures. If you can provide a
solution for this, it would be helpful.

On 14 Jan 2019 5:33 pm, "Matt Caswell" <matt at openssl.org> wrote:

>
>
> On 13/01/2019 04:36, Ananthu Unnikrishnan wrote:
> > Hi Matt,
> >
> >             Thanks a lot for the response. The alternative method to
> resume job
> > operation is a good one.
> >
> >             I need some more clarifications regarding the async job
> operation.
> > Please help.
> >             In ssl/ss_lib.c, ssl_start_async_job() function found. In
> which
> > ASYNC_start_job function is calling but no mechanism found for polling
> the fd's.
> > Is it written elsewhere?.
>
> You are supposed to write the fd polling code yourself, e.g. by using
> SSL_get_all_async_fds() or SSL_get_changed_async_fds() to get hold of the
> fds,
> and then followed by "select" or similar.
>
> See:
>
> https://www.openssl.org/docs/man1.1.1/man3/SSL_get_all_async_fds.html
>
> mATT
>
>
>
> >
> >
> > On Fri, Jan 11, 2019 at 3:12 PM Matt Caswell <matt at openssl.org
> > <mailto:matt at openssl.org>> wrote:
> >
> >
> >
> >     On 10/01/2019 18:09, Ananthu Unnikrishnan wrote:
> >     > Hi Matt,
> >     >
> >     >         Thanks a lot for the reply.
> >     >
> >     >        After calling ASYNC_pause_job() from the engine, control
> will
> >     transfer to
> >     > the place where we start the ASYNC_start_job right? So how can we
> write
> >     the code
> >     > to put a trigger on fd in the same thread?
> >
> >     I'm not saying you can't use threads to do this - only that you need
> to be
> >     careful about using them and must obey the rules.
> >
> >
> >     > If I am wrong  please correct me.
> >     > Also if u can suggest where resume operation should be done, it
> would be
> >     helpful.
> >
> >     Typically the calling application will wait on the fd (e.g. using
> "select" or
> >     similar) until it is readable. Once it sees the fd as readable then
> it calls
> >     ASYNC_start_job() again.
> >
> >
> >     As an aside you may also be interested in this PR that is currently
> being
> >     reviewed:
> >
> >     https://github.com/openssl/openssl/pull/7573
> >
> >     This adds an alternative mechanism for signalling other than using
> fds (i.e. to
> >     use a callback instead) and will be available in OpenSSL 3.0 when it
> gets
> >     released.
> >
> >     Matt
> >
> >
> >
> >     >
> >     > On Thu, Jan 10, 2019 at 10:11 PM Matt Caswell <matt at openssl.org
> >     <mailto:matt at openssl.org>
> >     > <mailto:matt at openssl.org <mailto:matt at openssl.org>>> wrote:
> >     >
> >     >
> >     >
> >     >     On 10/01/2019 09:39, Ananthu Unnikrishnan wrote:
> >     >     > Hi all,
> >     >     >
> >     >     >         We are not able to access the waitctx address from
> the job
> >     address
> >     >     using
> >     >     > ASYNC_get_wait_ctx(job) from a thread which starts in the
> bind
> >     section of the
> >     >     > dynamic engine. The job address is the same as that we got
> >     >     > using ASYNC_get_current_job.  Can anyone help on this?
> >     >
> >     >     It's very unclear from the your various emails what you are
> trying to
> >     achieve
> >     >     and the problem that you are experiencing.
> >     >
> >     >     From the above it sounds like you are starting threads from
> inside
> >     your dynamic
> >     >     engine? ASYNC_JOBs are always local to a thread. When a job is
> started
> >     it is
> >     >     associated with the current thread. So ASYNC_get_current_job()
> returns
> >     the job
> >     >     associated with the current thread. Similarly if you pause a
> job,
> >     >     ASYNC_pause_job() must be called from that same thread.
> Finally if you
> >     restart a
> >     >     paused job it must also be restarted on the same thread.
> >     >
> >     >     If you are starting new threads from within your engine then
> extreme
> >     care must
> >     >     be taken to obey the above rules - otherwise you are likely to
> get strange
> >     >     results.
> >     >
> >     >     Matt
> >     >
> >     >
> >     >     >
> >     >     >
> >     >     >
> >     >     > On Tue, Jan 8, 2019 at 11:26 AM Ananthu Unnikrishnan
> >     >     <ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>
> >     >     > <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>>>
> wrote:
> >     >     >
> >     >     >     Can anyone please help on this? If u need any additional
> information
> >     >     please
> >     >     >     let me know.
> >     >     >
> >     >     >     On Mon, Jan 7, 2019 at 6:25 PM Ananthu Unnikrishnan
> >     >     <ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>
> >     >     >     <mailto:ananthuu741 at gmail.com <mailto:
> ananthuu741 at gmail.com>
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>>>
> wrote:
> >     >     >
> >     >     >         Hi all,
> >     >     >
> >     >     >         Adding more details to the previous mail. We have
> edited the
> >     OpenSSL
> >     >     >         code for implementing the polling for changed
> fd's as in
> >     OpenSSL speed
> >     >     >         command. Attached the code snippet of the same along
> with
> >     this mail.
> >     >     >         Mentioned below some observations which found
> doubtful:
> >     >     >
> >     >     >         1) We have got prints in ASYNC_FINISH case, before
> getting
> >     print in
> >     >     >         ASYNC_PAUSE case. Attaching the log also, so you can
> understand
> >     >     easily.
> >     >     >         2) Shown below the code snippet which we have
> written to
> >     provide a
> >     >     write
> >     >     >         on fd.
> >     >     >
> >     >     >         if ((waitctx = ASYNC_get_wait_ctx((ASYNC_JOB *)job))
> == NULL) {
> >     >     >                 printf("In dynamic engine | waitctx == NULL
> : %d\n",
> >     >     __LINE__);
> >     >     >                 return ret;
> >     >     >             }
> >     >     >
> >     >     >             printf("\n----- In dynamic engine | After
> pausing | job
> >     is %lx |
> >     >     >         waitctx in resume job %lx |-----\n", job, waitctx);
> >     >     >
> >     >     >                 if ((ret = ASYNC_WAIT_CTX_get_fd(waitctx,
> engine_id,
> >     &efd,
> >     >     >                                       &custom)) > 0) {
> >     >     >                 if (write(efd, &buf, sizeof(uint64_t)) ==
> -1) {
> >     >     >                     printf("\nFailed to write\n");
> >     >     >                 }
> >     >     >             }
> >     >     >
> >     >     >         Here waitctx is getting NULL when we tried to fetch
> waitctx
> >     using
> >     >     >         ASYNC_get_wait_ctx(). We have printed data of
> sizeof(ASYNC_JOB)
> >     >     ie.1176
> >     >     >         bytes contained in the job address from the engine,
> we got
> >     >     correct data
> >     >     >         before calling ASYNC_pause() and but got half data
> zero data
> >     when we
> >     >     >         tried to access the same address after
> ASYNC_pause().
> >     >     >
> >     >     >         Attaching both job structure contents before and
> after calling
> >     >     >         ASYNC_pause(). Also, we got the correct data in the
> same address
> >     >     when we
> >     >     >         have printed from ssl_start_async_job() before
> polling starts,
> >     >     where we
> >     >     >         have started the async_job
> >     >     >         .
> >     >     >         The prints starting with "In dynamic engine" are
> prints
> >     inside the
> >     >     >         engine, rest prints are from ss/ssl_lib.c. Kindly
> check this and
> >     >     please
> >     >     >         point out if anything is wrong somewhere. Thanks in
> advance.
> >     >     >
> >     >     >
> >     >     >         On Sun, Jan 6, 2019 at 10:30 AM Ananthu Unnikrishnan
> >     >     >         <ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com
> >
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>
> >     >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>
> >     <mailto:ananthuu741 at gmail.com <mailto:ananthuu741 at gmail.com>>>>
> wrote:
> >     >     >
> >     >     >             Hi all,
> >     >     >
> >     >     >                    We have implemented a dynamic engine and
> tested
> >     in the
> >     >     async
> >     >     >             mode using OpenSSL speed command. But in a real
> network
> >     >     scenario, we
> >     >     >             have seen only starting the async_job(in file
> ssl/ssl_lib.c,
> >     >     >             function: ssl_start_async_job) in the OpenSSL. We
> >     haven't seen
> >     >     >             polling async_fd's for resuming the job as in
> >     apps/speed.c(in
> >     >     speed
> >     >     >             command case).
> >     >     >                     So can anyone please suggest any
> solution for
> >     resuming the
> >     >     >             async_jobs in real network scenario? Thanks in
> advance.
> >     >     >
> >     >     >
> >     >     >
> >     >     >
> >     >     >
> >     >     >         --
> >     >     >         With best Regards,
> >     >     >
> >     >     >                  Ananthu
> >     >     >
> >     >     >
> >     >     >
> >     >     >
> >     >     >     --
> >     >     >     With best Regards,
> >     >     >
> >     >     >              Ananthu
> >     >     >
> >     >     >
> >     >     >
> >     >     >
> >     >     > --
> >     >     > With best Regards,
> >     >     >
> >     >     >          Ananthu
> >     >     >
> >     >     >
> >     >     >
> >     >     --
> >     >     openssl-users mailing list
> >     >     To unsubscribe:
> https://mta.openssl.org/mailman/listinfo/openssl-users
> >     >
> >     >
> >     >
> >     > --
> >     > With best Regards,
> >     >
> >     >          Ananthu
> >     >
> >     >
> >     >
> >     --
> >     openssl-users mailing list
> >     To unsubscribe:
> https://mta.openssl.org/mailman/listinfo/openssl-users
> >
> >
> >
> > --
> > With best Regards,
> >
> >          Ananthu
> >
> >
> >
> --
> openssl-users mailing list
> To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mta.openssl.org/pipermail/openssl-users/attachments/20190116/a1991443/attachment-0001.html>


More information about the openssl-users mailing list