A suggested patch for rejecting AVDTP CLOSE and START requests when a
CLOSE request has been initiated. The check is done by testing the
close_int flag in the avdtp_stream struct. The flag is reset when
receiving a timeout or rejection from the remote.
Changes in unix.c were made to call avdtp_unref() in case an AVDTP
session was marked by avdtp_ref(), but no SEP was found. This could
happen if START is called on a session being disconnected.
We found that these changes help to improve behavior both in case the
audio streaming application tries to start the stream during a
disconnection procedure, and if avdtp_close() is called twice.
Br,
Daniel ?rstadius
------------------------
audio/avdtp.c | 19 +++++++++++++++++--
audio/unix.c | 11 ++++++++++-
2 files changed, 27 insertions(+), 3 deletions(-)
diff --git a/audio/avdtp.c b/audio/avdtp.c
index 8046dda..839f035 100644
--- a/audio/avdtp.c
+++ b/audio/avdtp.c
@@ -2223,9 +2223,12 @@ static gboolean request_timeout(gpointer user_data)
break;
case AVDTP_CLOSE:
error("Close request timed out");
- if (lsep && lsep->cfm && lsep->cfm->close)
+ if (lsep && lsep->cfm && lsep->cfm->close) {
lsep->cfm->close(session, lsep, stream, &err,
lsep->user_data);
+ if (stream)
+ stream->close_int = FALSE;
+ }
break;
case AVDTP_SET_CONFIGURATION:
error("SetConfiguration request timed out");
@@ -2675,9 +2678,11 @@ static gboolean avdtp_parse_rej(struct avdtp *session,
return FALSE;
error("CLOSE request rejected: %s (%d)",
avdtp_strerror(&err), err.err.error_code);
- if (sep && sep->cfm && sep->cfm->close)
+ if (sep && sep->cfm && sep->cfm->close) {
sep->cfm->close(session, sep, stream, &err,
sep->user_data);
+ stream->close_int = FALSE;
+ }
return TRUE;
case AVDTP_ABORT:
if (!stream_rej_to_err(buf, size, &err, &acp_seid))
@@ -3138,6 +3143,11 @@ int avdtp_start(struct avdtp *session, struct
avdtp_stream *stream)
if (stream->lsep->state != AVDTP_STATE_OPEN)
return -EINVAL;
+ if (stream->close_int == TRUE) {
+ error("avdtp_start: rejecting start since close is initiated");
+ return -EINVAL;
+ }
+
memset(&req, 0, sizeof(req));
req.first_seid.seid = stream->rseid;
@@ -3156,6 +3166,11 @@ int avdtp_close(struct avdtp *session, struct
avdtp_stream *stream)
if (stream->lsep->state < AVDTP_STATE_OPEN)
return -EINVAL;
+ if (stream->close_int == TRUE) {
+ error("avdtp_close: rejecting close since it is already inititated");
+ return -EINVAL;
+ }
+
memset(&req, 0, sizeof(req));
req.acp_seid = stream->rseid;
diff --git a/audio/unix.c b/audio/unix.c
index 0829630..61cc369 100644
--- a/audio/unix.c
+++ b/audio/unix.c
@@ -1051,14 +1051,17 @@ static void start_resume(struct audio_device
*dev, struct unix_client *client)
struct a2dp_data *a2dp;
struct headset_data *hs;
unsigned int id;
+ gboolean ref_session = FALSE;
switch (client->type) {
case TYPE_SINK:
case TYPE_SOURCE:
a2dp = &client->d.a2dp;
- if (!a2dp->session)
+ if (!a2dp->session) {
a2dp->session = avdtp_get(&dev->src, &dev->dst);
+ ref_session = TRUE;
+ }
if (!a2dp->session) {
error("Unable to get a session");
@@ -1067,6 +1070,12 @@ static void start_resume(struct audio_device
*dev, struct unix_client *client)
if (!a2dp->sep) {
error("seid not opened");
+
+ if (ref_session) {
+ avdtp_unref(a2dp->session);
+ a2dp->session = NULL;
+ }
+
goto failed;
}
Hi Daniel,
On Thu, Aug 27, 2009 at 12:24 PM, Daniel
?rstadius<[email protected]> wrote:
> A suggested patch for rejecting AVDTP CLOSE and START requests when a
> CLOSE request has been initiated. The check is done by testing the
> close_int flag in the avdtp_stream struct. The flag is reset when
> receiving a timeout or rejection from the remote.
>
> Changes in unix.c were made to call avdtp_unref() in case an AVDTP
> session was marked by avdtp_ref(), but no SEP was found. This could
> happen if START is called on a session being disconnected.
>
> We found that these changes help to improve behavior both in case the
> audio streaming application tries to start the stream during a
> disconnection procedure, and if avdtp_close() is called twice.
It probably works better if you place it on avdtp struct so it works
even if we don't have a stream configured, I also prefer closing
instead of close_int.
The other parts looks good except this:
> diff --git a/audio/unix.c b/audio/unix.c
> index 0829630..61cc369 100644
> --- a/audio/unix.c
> +++ b/audio/unix.c
> @@ -1051,14 +1051,17 @@ static void start_resume(struct audio_device
> *dev, struct unix_client *client)
> ? ? ? ?struct a2dp_data *a2dp;
> ? ? ? ?struct headset_data *hs;
> ? ? ? ?unsigned int id;
> + ? ? ? gboolean ref_session = FALSE;
>
> ? ? ? ?switch (client->type) {
> ? ? ? ?case TYPE_SINK:
> ? ? ? ?case TYPE_SOURCE:
> ? ? ? ? ? ? ? ?a2dp = &client->d.a2dp;
>
> - ? ? ? ? ? ? ? if (!a2dp->session)
> + ? ? ? ? ? ? ? if (!a2dp->session) {
> ? ? ? ? ? ? ? ? ? ? ? ?a2dp->session = avdtp_get(&dev->src, &dev->dst);
> + ? ? ? ? ? ? ? ? ? ? ? ref_session = TRUE;
> + ? ? ? ? ? ? ? }
>
> ? ? ? ? ? ? ? ?if (!a2dp->session) {
> ? ? ? ? ? ? ? ? ? ? ? ?error("Unable to get a session");
> @@ -1067,6 +1070,12 @@ static void start_resume(struct audio_device
> *dev, struct unix_client *client)
>
> ? ? ? ? ? ? ? ?if (!a2dp->sep) {
> ? ? ? ? ? ? ? ? ? ? ? ?error("seid not opened");
> +
> + ? ? ? ? ? ? ? ? ? ? ? if (ref_session) {
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? avdtp_unref(a2dp->session);
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? a2dp->session = NULL;
> + ? ? ? ? ? ? ? ? ? ? ? }
> +
> ? ? ? ? ? ? ? ? ? ? ? ?goto failed;
> ? ? ? ? ? ? ? ?}
Normally we should wait for the client to disconnect then we release
its avdtp session, if it is closing/closed we shouldn't even attempt
to create another session to destroy a few lines ahead. So in this
case the client probably need fixing so it don't even try to start
while closing/closed.
ps: There is a patch to pulseaudio already that should fix this:
http://git.0pointer.de/?p=pulseaudio.git;a=commitdiff;h=8169a6a6c921215c1353e8a34fccbdc4e2e20440
--
Luiz Augusto von Dentz
Engenheiro de Computa??o
Hi Daniel,
On Tue, Sep 01, 2009, Daniel ?rstadius wrote:
> From 627850090c93152b8faa3e4a724df24dbc556d4e Mon Sep 17 00:00:00 2001
> From: Daniel Orstadius <[email protected]>
> Date: Tue, 1 Sep 2009 13:55:44 +0300
> Subject: [PATCH] Reject AVDTP START/STOP when disconnecting
>
> ---
> audio/avdtp.c | 19 +++++++++++++++++--
> 1 files changed, 17 insertions(+), 2 deletions(-)
The patch has been pushed upstream. Thanks.
Johan
On Tue, Sep 1, 2009 at 10:02 AM, Johan Hedberg<[email protected]> wrote:
> So Daniel, could you split out the first part of your patch and resend it?
>From 627850090c93152b8faa3e4a724df24dbc556d4e Mon Sep 17 00:00:00 2001
From: Daniel Orstadius <[email protected]>
Date: Tue, 1 Sep 2009 13:55:44 +0300
Subject: [PATCH] Reject AVDTP START/STOP when disconnecting
---
audio/avdtp.c | 19 +++++++++++++++++--
1 files changed, 17 insertions(+), 2 deletions(-)
diff --git a/audio/avdtp.c b/audio/avdtp.c
index 8046dda..839f035 100644
--- a/audio/avdtp.c
+++ b/audio/avdtp.c
@@ -2223,9 +2223,12 @@ static gboolean request_timeout(gpointer user_data)
break;
case AVDTP_CLOSE:
error("Close request timed out");
- if (lsep && lsep->cfm && lsep->cfm->close)
+ if (lsep && lsep->cfm && lsep->cfm->close) {
lsep->cfm->close(session, lsep, stream, &err,
lsep->user_data);
+ if (stream)
+ stream->close_int = FALSE;
+ }
break;
case AVDTP_SET_CONFIGURATION:
error("SetConfiguration request timed out");
@@ -2675,9 +2678,11 @@ static gboolean avdtp_parse_rej(struct avdtp *session,
return FALSE;
error("CLOSE request rejected: %s (%d)",
avdtp_strerror(&err), err.err.error_code);
- if (sep && sep->cfm && sep->cfm->close)
+ if (sep && sep->cfm && sep->cfm->close) {
sep->cfm->close(session, sep, stream, &err,
sep->user_data);
+ stream->close_int = FALSE;
+ }
return TRUE;
case AVDTP_ABORT:
if (!stream_rej_to_err(buf, size, &err, &acp_seid))
@@ -3138,6 +3143,11 @@ int avdtp_start(struct avdtp *session, struct
avdtp_stream *stream)
if (stream->lsep->state != AVDTP_STATE_OPEN)
return -EINVAL;
+ if (stream->close_int == TRUE) {
+ error("avdtp_start: rejecting start since close is initiated");
+ return -EINVAL;
+ }
+
memset(&req, 0, sizeof(req));
req.first_seid.seid = stream->rseid;
@@ -3156,6 +3166,11 @@ int avdtp_close(struct avdtp *session, struct
avdtp_stream *stream)
if (stream->lsep->state < AVDTP_STATE_OPEN)
return -EINVAL;
+ if (stream->close_int == TRUE) {
+ error("avdtp_close: rejecting close since it is already inititated");
+ return -EINVAL;
+ }
+
memset(&req, 0, sizeof(req));
req.acp_seid = stream->rseid;
--
1.6.0.4
Thanks for the feedback.
/Daniel
Hi Daniel & Luiz,
On Thu, Aug 27, 2009, Luiz Augusto von Dentz wrote:
> On Thu, Aug 27, 2009 at 12:24 PM, Daniel
> ?rstadius<[email protected]> wrote:
> > A suggested patch for rejecting AVDTP CLOSE and START requests when a
> > CLOSE request has been initiated. The check is done by testing the
> > close_int flag in the avdtp_stream struct. The flag is reset when
> > receiving a timeout or rejection from the remote.
> >
> > Changes in unix.c were made to call avdtp_unref() in case an AVDTP
> > session was marked by avdtp_ref(), but no SEP was found. This could
> > happen if START is called on a session being disconnected.
> >
> > We found that these changes help to improve behavior both in case the
> > audio streaming application tries to start the stream during a
> > disconnection procedure, and if avdtp_close() is called twice.
>
> It probably works better if you place it on avdtp struct so it works
> even if we don't have a stream configured,
But a stream can't be in OPEN state if it hasn't been configured (i.e. we
have a stream struct for it). So this part of the patch looks good to me.
> I also prefer closing instead of close_int.
That would have a different meaning though. To me closing means we're
closing but doesn't say anything about who initiated the close. close_int
otoh tells us that we initiated the close. Also, since this variable
already exists and is needed I think reusing it is fine.
So Daniel, could you split out the first part of your patch and resend it?
Thanks.
Johan