Re: Let StartServiceCtrlDispatcher handle services started other than with StartService (with patch)

[Alexander Yaworsky]

Hello

In fact, the existing implementation of service subsystem in advapi32 is very poor and is sufficient only for very simple services
or test programs. I tried to launch oracle 7 database under wine and that trial has been failed. The main problem was inability of
parameter passing. Not only oracle -- most services require that. I did some improvements. Although I still cannot start database,
because its launcher, oradim, requires CreateRemoteThread, the listener service works well (although, it cannot be fully tested
because no database instance exists, but there is nothing suspicious in trace logs).
I even tried to submit a patch but it was silently ignored. Let me remind about this patch --
http://www.winehq.org/hypermail/wine-patches/2004/06/0174.html maybe it will be useful. If you have suggestions, I promise to
rewrite it as required ;))

Also, maybe the diagram shown below will be useful.

Thanks.

This simplified diagram shows interaction between three main
components of win32 service subsystem.

Service Control |   Service Control Manager   |   Service Program
    Program     |                             |
----------------+-----------------------------+-----------------------------
                |                             |
 StartService-----> LockServiceDatabase       |
                |   IF OWN_PROCES or          |
                |      (SHARE_PROCESS and Service
                |      Program is not running)|
                |   THEN                      |
                |                             |
                |   CreateProcess===============> ...
                |   wait for call to          |   initialization
                |   StartServiceCtrlDispatcher|   must not be longer
                |   timeout=30s               |   than 30 seconds
                |                             |   ...
                |   if wait timed out then    |   ...
                |   TerminateProcess          |   ...
      failure <---- UnlockServiceDatabase     |   ...
                |                             |   ...
                |   wait satisfied <============= StartServiceCtrlDispatcher
                |                             |
                |   END IF                    |
                |                             |
                |   CreateThread ***************> ServiceMain
                |   (or CreateRemoteThread    |   ...
                |    for SHARE_PROCESS;       |    *
                |    if SHARE_PROCESS then    |    *
                |    find entry in service    |    **********************
                |    table by service name;   |                         *
                |    if OWN_PROCESS then use  |                         *
                |    first entry)             |                         *
      success <---- UnlockServiceDatabase     |                         *
                |                             |                         *
                |   accept control requests   |                         *
                |                             |                         *
 ControlService---> request accepted ===========> ServiceControlHandler *
                |                             |   ...                   *
      status <----- return status <============== SetServiceStatus      *
                |                             |                         *
                |                             |                         *
                |                             |    **********************
                |                             |    *
                |                             |   ...
                |                             |   SetServiceStatus
                |   ExitThread <***************** (STOPPED)
                |   if Service Program has    |
                |   now only one thread       |
                |   return control from       |
                |   StartServiceCtrlDispatcher==> ...
                |                             |   ExitProcess

legend:

--- Service Control Program's thread
=== Service Program's main thread (control thread)
*** Service Program's service thread
OWN_PROCESS == SERVICE_WIN32_OWN_PROCESS
SHARE_PROCESS == SERVICE_WIN32_SHARE_PROCESS

Implementation.

completely win32-based

Two basic things:
a) The only time the service control manager requests ownership of the lock is when it is starting a service.
   (C) M$DN;
b) When StartService creates a new service process, it must pass arguments, or at least
   argv[0] - the service name - for ServiceMain across process boundary.

So, use the existence of shared memory object as lock indicator and the content of object for
passing service name.

Also we need a wait object for monitoring process startup. The call to StartServiceCtrlDispatcher will satisfy
wait condition.

A dedicated shared memory object is used for each service thread. It holds other arguments,
control requests and status.

Also we need mutex object for serialization of control requests and two wait objects for
serialization of service control program thread and service control handler thread.

Let StartServiceCtrlDispatcher handle services started other than with StartService (with patch)

[Steven Edwards]

Hi Alexander,

--- Alexander Yaworsky  wrote:
> I even tried to submit a patch but it was silently ignored. Let me
> remind about this patch --
> http://www.winehq.org/hypermail/wine-patches/2004/06/0174.html maybe
> it will be useful. If you have suggestions, I promise to
> rewrite it as required ;))

Dont take offence if your patch goes to the void. Just resubmit after a
a week to give it a chance to filter through Alexandre's tree. Its
possible either

1. He is going to merge it and its just taken him a while
2. He rejected it and just forgot to send a note due to the constrants
of time.
3. It went to the void in space where the lost patches go.

Just resubmit and you should hear something back from someone.

Thanks
Steven

__________________________________
Do you Yahoo!?
Yahoo! Mail Address AutoComplete - You start. We finish.
http://promotions.yahoo.com/new_mail

Let StartServiceCtrlDispatcher handle services started other than with StartService (with patch)

[Shachar Shemesh]

Steven Edwards wrote:
> Hi Alexander,
>
> --- Alexander Yaworsky  wrote:
> > I even tried to submit a patch but it was silently ignored. Let me
> > remind about this patch --
> > http://www.winehq.org/hypermail/wine-patches/2004/06/0174.html maybe
> > it will be useful. If you have suggestions, I promise to
> > rewrite it as required ;))
>
> Dont take offence if your patch goes to the void. Just resubmit after a
> a week to give it a chance to filter through Alexandre's tree. Its
> possible either
> - He is going to merge it and its just taken him a while
> - He rejected it and just forgot to send a note due to the constrants
>
> of time.
> 3. It went to the void in space where the lost patches go.
>
> Just resubmit and you should hear something back from someone.
>
> Thanks
> Steven

Actually, during the last wineconf, "There's nothing obviously wrong 
with it, but it makes Alexandre uneasy" for the usual reason patches go 
to void

Shachar

Let StartServiceCtrlDispatcher handle services started other than with StartService (with patch)

[Alexander Yaworsky]

Hello
> Dont take offence if your patch goes to the void. Just resubmit after a
> a week to give it a chance to filter through Alexandre's tree. Its
> possible either

I merely meant that maybe something wrong with implementation. Well, I will resubmit it soon.

Let StartServiceCtrlDispatcher handle services started other than with StartService (with patch)

[Mike McCormack]

Alexander Yaworsky wrote:
> I even tried to submit a patch but it was silently ignored. Let me remind about this patch --
> http://www.winehq.org/hypermail/wine-patches/2004/06/0174.html maybe it will be useful. If you have suggestions, I promise to
> rewrite it as required ;))

Hi Alexandre,

Your patch is very large.  It's difficult to find regressions in large 
patches, and it's difficult to review as so much code is changed.  It 
may be all perfectly correct, but it's not obviously perfectly correct :)

I'd suggest that you resubmit the patch and break it up into smaller 
chunks, so that as much as possible, each one is easy to review and 
makes a small understandable, atomic and necessary change.  Test cases 
also help, if you can write them.

From the looks of things your patch is well researched, and you 
understand what you're doing.  If you submit your work in smaller 
chunks, I'm pretty sure it will be accepted.

Mike