You made yourself clear on that I removed all struct MHD_Connections in my
structures and never use MHD_close_connection.
I am running the program on Linux. I tested a lot of scenarios with 0.9.13,
closed and reopened 20 connections with 2 different clients etc... and
things work fine.
Whereas with the 0.9.14 I followed the Segmentation fault in debug with
Anjuta and as I told you it appears :
int MHD_connection_handle_write (struct MHD_Connection *connection)
|
|_switch (connection->state)
|
|_case MHD_CONNECTION_CHUNKED_BODY_READY:
|
|_check_write_done (connection,
(connection->response->total_size ==
connection->response_write_position) ?
MHD_CONNECTION_BODY_SENT :
MHD_CONNECTION_CHUNKED_BODY_UNREADY);
Valgrind says :
==24566== Thread 4:
==24566== Invalid read of size 8
==24566== at 0x4E33DD3: MHD_connection_handle_write (connection.c:1899)
==24566== by 0x4E36E78: MHD_handle_connection (daemon.c:624)
==24566== by 0x626CD8B: start_thread (pthread_create.c:304)
==24566== by 0x5A9D04C: clone (clone.S:112)
==24566== Address 0x50 is not stack'd, malloc'd or (recently) free'd
On Fri, Sep 16, 2011 at 1:41 PM, Christian Grothoff
<[email protected]>wrote:
> On Friday, September 16, 2011 11:46:15 AM Regis Louge wrote:
> > Christian,
> >
> > The purpose of the Web Server is to manage Services, as it is now it
> > provides an interface for the clients for them to interact with those
> > services, for example a performing a HTTP GET request on a Service will
> > return its value and performing a HTTP PUT request will set this value to
> > the given one. The Web Server can contain a lot of Services and have a
> lot
> > of clients.
> >
> > In addition to that, one of the feature that I want to add is to be able
> to
> > receive notifications on value changes of certain services. To do so,
> each
> > client should be able to perform a GET on the URI "/events" in order to
> > engage the connection to these notifications. Then, whenever a Service
> > changes value, the server checks if the client has subscribed to that
> > service's notifications, and if so, it sends him a notification of the
> > value change with the new value.
> >
> > For now on, this works well thanks to MHD_create_response_from_callback.
> > All clients receive notifications of the services that they subscribed to
> > etc...
> >
> > I have to admit that the choice of defining clients with IP addresses was
> > not well thought,indeed I used IP address as unique identifiers for each
> > client and didn't think of all the scenarios (even though the web server
> > aims at being run in a Local Area Network) I remodeled the notifications
> > part on the web server using Sessions, thank you for the advice !
> >
> > Nevertheless, the segmentation fault still appears when I close and
> reopen
> > another connection, MHD tells me that it is impossible to send data due
> to
> > a "Broken Pipe", and goes into the request_completed callback. (the same
> > segmentation fault in my previous mail)
> >
> > I just downgraded my MHD to 0.9.13 and it works like a charm ! (tried
> with
> > 0.9.14 and SVN HEAD) Any idea why ?
>
> I doubt it is the different version, you might just have gotten lucky
> (memory
> corruption can cause non-deterministic bugs...). In any case, let me again
> say that if your code still does (as you wrote earlier):
>
> if(client->connection != NULL)
> MHD_connection_close(client->connection,
> MHD_REQUEST_TERMINATED_WITH_ERROR);
>
> you will have problems. MHD_connection_close is not in microhttpd.h and not
> an
> exported symbol so client code must not use it or face the music (i.e.
> crashes).
>
> As for 'broken pipe' -- did you ever read this?
>
> http://www.gnu.org/s/libmicrohttpd/microhttpd.html#SIGPIPE
>
> If not, this might be your problem (even though it is not a SIGSEGV in that
> case, but a SIGPIPE), unless you're on GNU/Linux. If this is not the
> issue,
> try using valgrind to find out more about your segfault...
>
> Happy hacking,
>
> Christian
>
> >
> > On Thu, Sep 15, 2011 at 2:57 PM, Christian Grothoff
> >
> > <[email protected]>wrote:
> > > Eh, MHD_connection_close is NOT part of the public API for good
> reasons.
> > >
> > > You MUST not use it, the symbol is not even exported by the library.
> > >
> > > Getting segmentation faults from using it yourself is expected. Do not
> > > do that.
> > >
> > > Also, I am getting more an more confused about what you're trying to
> do.
> > > Naturally you CAN get two connections from the same source IP address
> at
> > > the same time, and your server should support that (otherwise it is
> > > simply broken). MHD has no problem handling more than one connection
> > > per IP at a time either.
> > >
> > >
> > > I currently can only conclude that you're trying to do something with
> IP
> > > addresses that you should manage with sessions/cookies and that your
> > > application logic is most likely severely broken (based on not
> > > understanding TCP/IP and/or HTTP) and thus you're trying to do these
> > > strange things and come up with odd requirements like the 'instant'
> > > cleanup on connection closure.
> > >
> > > TCP-FIN is not 'instant' and a browser may send us a TCP-SYN before we
> > > get the FIN, right? Not to mention, what would your system do with two
> > > users behind NAT using the same IP address?
> > >
> > > With SVN HEAD, MHD is now calling the response destroy handler (as well
> > > as the 'connection completed' handlers) as soon as the OS tells us that
> > > the connection is dead. More cannot be done, and mucking around with
> > > MHD internals is not going to improve the situation. Also, you should
> > > NEVER EVER store a 'struct MHD_Connection' in your own data structures,
> > > that's virtually always a bug. Instead, store a pointer to your data
> > > structures in the 'void**' you're provided by MHD.
> > >
> > > Happy hacking!
> > >
> > > Christian
> > >
> > > On 09/15/2011 09:15 AM, Regis Louge wrote:
> > >> Hi,
> > >>
> > >> I have updated my version to the 0.9.14 that contains this patch
> > >> (without compilation error :) ). I managed to handle the disconnection
> > >> reconnection of the client by storing the MHD_Connection in my
> structure
> > >> and closing it manually whenever a call on "/push" is made from the
> same
> > >> IP address and a previous connection was stored :
> > >>
> > >> if(client->connection != NULL)
> > >> MHD_connection_close(client->**connection,
> > >> MHD_REQUEST_TERMINATED_WITH_**ERROR);
> > >> client->connection = connection;
> > >>
> > >> Now I encounter something really weird, indeed, if a client
> disconnects,
> > >> I can "send" him a push notification without anything happening on my
> > >> web server, but when I "send" him a second one I have this weird
> > >> segmentation fault in :
> > >>
> > >> int MHD_connection_handle_write (struct MHD_Connection *connection)
> > >>
> > >> |_switch (connection->state)
> > >> |
> > >> |_case MHD_CONNECTION_CHUNKED_BODY_**READY:
> > >> |_check_write_done (connection,
> > >> |
> > >> (connection->response->total_**size ==
> > >>
> > >> connection->response_write_**position) ?
> > >>
> > >> MHD_CONNECTION_BODY_SENT :
> > >> MHD_CONNECTION_CHUNKED_BODY_**UNREADY);
> > >>
> > >> It is right after the connection closes so connection->state is
> supposed
> > >> to be MHD_CONNECTION_CLOSED ?
> > >>
> > >> Any idea what that would happen ? And why does the first push
> > >> notification attempt don't fail ?
> > >>
> > >> Thanks a lot
> > >>
>
--
Régis LOUGE
Eleve Ingénieur 5° année - ECE Paris - Aalborg Universitet
Mail : [email protected]
Tel : +33 6 62 51 40 91
