Re: [PATCH] wininet: Close sockets open in http requests

[Misha Koshelev]

> n 10/13/07, Nigel Liang <ncliang at gmail.com> wrote:
> > Hi,
> >
> > Http requests open sockets and forget to close them. Left running over a long
> > period, the socket file descriptor starts getting large. Once it exceeds
> > FD_SETSIZE, the behavior becomes unpredictable and results in funcky crashes.
> >
> > -Nigel
> >
> > ---
> >  dlls/wininet/internet.c |    3 +++
> >  1 files changed, 3 insertions(+), 0 deletions(-)
> >
> > diff --git a/dlls/wininet/internet.c b/dlls/wininet/internet.c
> > index 0edca74..8e3a98e 100644
> > --- a/dlls/wininet/internet.c
> > +++ b/dlls/wininet/internet.c
> > @@ -1014,6 +1014,9 @@ BOOL WINAPI InternetCloseHandle(HINTERNE
> >          return FALSE;
> >      }
> >
> > +    if (lpwh->htype == WH_HHTTPREQ)
> > +        NETCON_close(&((LPWININETHTTPREQW)lpwh)->netConnection);
> > +
> >      WININET_Release( lpwh );
> >      WININET_FreeHandle( hInternet );
> >
> > --
> > 1.4.1
>
> Hi,
>
> This patch was sent 4 days ago and it hasn't gone in yet. Could anyone
> give me some comments on it? Dan filed bug 10032 which is related to
> this. While we shouldn't be using select() on high FDs for networking,
> this patch closes the sockets that were open by http requests to avoid
> getting to high FDs...
>
> Thanks,
> -Nigel

Hi, well looking over this very quickly right off the bat it does not seem 
correct to me. WININET_Release will call the close_connection method of the 
handle if it is defined, and for WH_HHTTPREQ this is HTTP_CloseConnection,
which does call NETCON_Close along with sending all the proper status notifications,
etc.

So if in fact NETCON_Close is _not_ being called for some reason, you need to track
down where involving those functions either why the proper one is not
getting called (if that is the case) or if that function somehow needs to call
NETCON_Close when it doesn't.

Misha

[PATCH] wininet: Close sockets open in http requests

[Misha Koshelev]

On Wed, 2007-10-17 at 17:10 -0500, Misha Koshelev wrote:
> > n 10/13/07, Nigel Liang <ncliang at gmail.com> wrote:
> > > Hi,
> > >
> > > Http requests open sockets and forget to close them. Left running over a long
> > > period, the socket file descriptor starts getting large. Once it exceeds
> > > FD_SETSIZE, the behavior becomes unpredictable and results in funcky crashes.
> > >
> > > -Nigel
> > >
> > > ---
> > >  dlls/wininet/internet.c |    3 +++
> > >  1 files changed, 3 insertions(+), 0 deletions(-)
> > >
> > > diff --git a/dlls/wininet/internet.c b/dlls/wininet/internet.c
> > > index 0edca74..8e3a98e 100644
> > > --- a/dlls/wininet/internet.c
> > > +++ b/dlls/wininet/internet.c
> > > @@ -1014,6 +1014,9 @@ BOOL WINAPI InternetCloseHandle(HINTERNE
> > >          return FALSE;
> > >      }
> > >
> > > +    if (lpwh->htype == WH_HHTTPREQ)
> > > +        NETCON_close(&((LPWININETHTTPREQW)lpwh)->netConnection);
> > > +
> > >      WININET_Release( lpwh );
> > >      WININET_FreeHandle( hInternet );
> > >
> > > --
> > > 1.4.1
> >
> > Hi,
> >
> > This patch was sent 4 days ago and it hasn't gone in yet. Could anyone
> > give me some comments on it? Dan filed bug 10032 which is related to
> > this. While we shouldn't be using select() on high FDs for networking,
> > this patch closes the sockets that were open by http requests to avoid
> > getting to high FDs...
> >
> > Thanks,
> > -Nigel
>
> Hi, well looking over this very quickly right off the bat it does not seem 
> correct to me. WININET_Release will call the close_connection method of the 
> handle if it is defined, and for WH_HHTTPREQ this is HTTP_CloseConnection,
> which does call NETCON_Close along with sending all the proper status notifications,
> etc.
>
> So if in fact NETCON_Close is _not_ being called for some reason, you need to track
> down where involving those functions either why the proper one is not
> getting called (if that is the case) or if that function somehow needs to call
> NETCON_Close when it doesn't.
>
> Misha

Actually, thinking about this a little more, if this is indeed happening
(NETCON_close not getting called) then likely this is a reference count
problem for the object, and thus the proper fix would be to find where
the reference count is being incremented but not appropriately
decremented, as info->close_connection only gets called if the reference
count of the object is zero. In my own experience, there have been quite
a few places where reference counts were incremented and not decremented
appropriately in wininet (the ones I ran into I believe are patched now
though).

Misha

[PATCH] wininet: Close sockets open in http requests

[Nigel Liang]

On 10/17/07, Misha Koshelev <mk144210(a)bcm.edu> wrote:
> On Wed, 2007-10-17 at 17:10 -0500, Misha Koshelev wrote:
> > > n 10/13/07, Nigel Liang <ncliang at gmail.com> wrote:
> > > > Hi,
> > > >
> > > > Http requests open sockets and forget to close them. Left running over a long
> > > > period, the socket file descriptor starts getting large. Once it exceeds
> > > > FD_SETSIZE, the behavior becomes unpredictable and results in funcky crashes.
> > > >
> > > > -Nigel
> > > >
> > > > ---
> > > >  dlls/wininet/internet.c |    3 +++
> > > >  1 files changed, 3 insertions(+), 0 deletions(-)
> > > >
> > > > diff --git a/dlls/wininet/internet.c b/dlls/wininet/internet.c
> > > > index 0edca74..8e3a98e 100644
> > > > --- a/dlls/wininet/internet.c
> > > > +++ b/dlls/wininet/internet.c
> > > > @@ -1014,6 +1014,9 @@ BOOL WINAPI InternetCloseHandle(HINTERNE
> > > >          return FALSE;
> > > >      }
> > > >
> > > > +    if (lpwh->htype == WH_HHTTPREQ)
> > > > +        NETCON_close(&((LPWININETHTTPREQW)lpwh)->netConnection);
> > > > +
> > > >      WININET_Release( lpwh );
> > > >      WININET_FreeHandle( hInternet );
> > > >
> > > > --
> > > > 1.4.1
> > >
> > > Hi,
> > >
> > > This patch was sent 4 days ago and it hasn't gone in yet. Could anyone
> > > give me some comments on it? Dan filed bug 10032 which is related to
> > > this. While we shouldn't be using select() on high FDs for networking,
> > > this patch closes the sockets that were open by http requests to avoid
> > > getting to high FDs...
> > >
> > > Thanks,
> > > -Nigel
> >
> > Hi, well looking over this very quickly right off the bat it does not seem
> > correct to me. WININET_Release will call the close_connection method of the
> > handle if it is defined, and for WH_HHTTPREQ this is HTTP_CloseConnection,
> > which does call NETCON_Close along with sending all the proper status notifications,
> > etc.
> >
> > So if in fact NETCON_Close is _not_ being called for some reason, you need to track
> > down where involving those functions either why the proper one is not
> > getting called (if that is the case) or if that function somehow needs to call
> > NETCON_Close when it doesn't.
> >
> > Misha
>
> Actually, thinking about this a little more, if this is indeed happening
> (NETCON_close not getting called) then likely this is a reference count
> problem for the object, and thus the proper fix would be to find where
> the reference count is being incremented but not appropriately
> decremented, as info->close_connection only gets called if the reference
> count of the object is zero. In my own experience, there have been quite
> a few places where reference counts were incremented and not decremented
> appropriately in wininet (the ones I ran into I believe are patched now
> though).

Hi,

Yep, you were right. There was a missing call to WININET_Release.
Found it and submitted a new patch:
http://www.winehq.org/pipermail/wine-patches/2007-October/045337.html

Thanks!
-Nigel

[PATCH] wininet: Close sockets open in http requests

[Misha Koshelev]

On Wed, 2007-10-17 at 17:10 -0700, Nigel Liang wrote:
> On 10/17/07, Misha Koshelev <mk144210(a)bcm.edu> wrote:
> > On Wed, 2007-10-17 at 17:10 -0500, Misha Koshelev wrote:
> > > > n 10/13/07, Nigel Liang <ncliang at gmail.com> wrote:
> > > > > Hi,
> > > > >
> > > > > Http requests open sockets and forget to close them. Left running over a long
> > > > > period, the socket file descriptor starts getting large. Once it exceeds
> > > > > FD_SETSIZE, the behavior becomes unpredictable and results in funcky crashes.
> > > > >
> > > > > -Nigel
> > > > >
> > > > > ---
> > > > >  dlls/wininet/internet.c |    3 +++
> > > > >  1 files changed, 3 insertions(+), 0 deletions(-)
> > > > >
> > > > > diff --git a/dlls/wininet/internet.c b/dlls/wininet/internet.c
> > > > > index 0edca74..8e3a98e 100644
> > > > > --- a/dlls/wininet/internet.c
> > > > > +++ b/dlls/wininet/internet.c
> > > > > @@ -1014,6 +1014,9 @@ BOOL WINAPI InternetCloseHandle(HINTERNE
> > > > >          return FALSE;
> > > > >      }
> > > > >
> > > > > +    if (lpwh->htype == WH_HHTTPREQ)
> > > > > +        NETCON_close(&((LPWININETHTTPREQW)lpwh)->netConnection);
> > > > > +
> > > > >      WININET_Release( lpwh );
> > > > >      WININET_FreeHandle( hInternet );
> > > > >
> > > > > --
> > > > > 1.4.1
> > > >
> > > > Hi,
> > > >
> > > > This patch was sent 4 days ago and it hasn't gone in yet. Could anyone
> > > > give me some comments on it? Dan filed bug 10032 which is related to
> > > > this. While we shouldn't be using select() on high FDs for networking,
> > > > this patch closes the sockets that were open by http requests to avoid
> > > > getting to high FDs...
> > > >
> > > > Thanks,
> > > > -Nigel
> > >
> > > Hi, well looking over this very quickly right off the bat it does not seem
> > > correct to me. WININET_Release will call the close_connection method of the
> > > handle if it is defined, and for WH_HHTTPREQ this is HTTP_CloseConnection,
> > > which does call NETCON_Close along with sending all the proper status notifications,
> > > etc.
> > >
> > > So if in fact NETCON_Close is _not_ being called for some reason, you need to track
> > > down where involving those functions either why the proper one is not
> > > getting called (if that is the case) or if that function somehow needs to call
> > > NETCON_Close when it doesn't.
> > >
> > > Misha
> >
> > Actually, thinking about this a little more, if this is indeed happening
> > (NETCON_close not getting called) then likely this is a reference count
> > problem for the object, and thus the proper fix would be to find where
> > the reference count is being incremented but not appropriately
> > decremented, as info->close_connection only gets called if the reference
> > count of the object is zero. In my own experience, there have been quite
> > a few places where reference counts were incremented and not decremented
> > appropriately in wininet (the ones I ran into I believe are patched now
> > though).
>
> Hi,
>
> Yep, you were right. There was a missing call to WININET_Release.
> Found it and submitted a new patch:
> http://www.winehq.org/pipermail/wine-patches/2007-October/045337.html
>
> Thanks!
> -Nigel

Glad my hours of staring at wininet led to something useful :) I looked
at your new patch and it looks good to me. Now if I could just figure
out bug 9922 (or even knew if it still occurs since I can't reproduce
it)...

Misha