Patchwork spice-qemu-char: Generate chardev open/close events (v3)

login
register
mail settings
Submitter Hans de Goede
Date July 22, 2011, 9:40 a.m.
Message ID <1311327655-5728-1-git-send-email-hdegoede@redhat.com>
Download mbox | patch
Permalink /patch/106247/
State New
Headers show

Comments

Hans de Goede - July 22, 2011, 9:40 a.m.
Define a state callback and make that generate chardev open/close events when
called by the spice-server. Note the code ignores these events for a spicevmc
with a subtypem of vdagent, this subtype specific knowledge is undesirable,
but unavoidable, see:
http://lists.freedesktop.org/archives/spice-devel/2011-July/004837.html

Changes in v2:
-Only ignore the state callback for spicevmc chardevs with a subtype of
 vdagent, instead of only allowing them for a subtype of usbredir

Changes in v3:
-This version actually compiles, sorry about that!

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
 spice-qemu-char.c |   42 +++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 41 insertions(+), 1 deletions(-)
Alon Levy - July 22, 2011, 12:25 p.m.
On Fri, Jul 22, 2011 at 11:40:55AM +0200, Hans de Goede wrote:
> Define a state callback and make that generate chardev open/close events when
> called by the spice-server. Note the code ignores these events for a spicevmc
> with a subtypem of vdagent, this subtype specific knowledge is undesirable,
> but unavoidable, see:
> http://lists.freedesktop.org/archives/spice-devel/2011-July/004837.html
> 
> Changes in v2:
> -Only ignore the state callback for spicevmc chardevs with a subtype of
>  vdagent, instead of only allowing them for a subtype of usbredir
> 
> Changes in v3:
> -This version actually compiles, sorry about that!
> 

ACK.

Another option would be to not call the state callback for the vdagent
subtype from spice-server, because like you say it ignores the continued
communication between the agent and the server. Then you could remove
the subtype check.

> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> ---
>  spice-qemu-char.c |   42 +++++++++++++++++++++++++++++++++++++++++-
>  1 files changed, 41 insertions(+), 1 deletions(-)
> 
> diff --git a/spice-qemu-char.c b/spice-qemu-char.c
> index ce75e91..a472f7e 100644
> --- a/spice-qemu-char.c
> +++ b/spice-qemu-char.c
> @@ -89,11 +89,48 @@ static int vmc_read(SpiceCharDeviceInstance *sin, uint8_t *buf, int len)
>      return bytes;
>  }
>  
> +static void vmc_state(SpiceCharDeviceInstance *sin, int connected)
> +{
> +    SpiceCharDriver *scd = container_of(sin, SpiceCharDriver, sin);
> +    int event;
> +
> +    /*
> +     * spice-server calls the state callback for the agent channel when the
> +     * spice client connects / disconnects. Given that not the client but
> +     * the server is doing the parsing of the messages this is wrong as the
> +     * server is still listening. Worse, this causes the parser in the server
> +     * to go out of sync, so we ignore state calls for subtype vdagent
> +     * spicevmc chardevs. For the full story see:
> +     * http://lists.freedesktop.org/archives/spice-devel/2011-July/004837.html
> +     */
> +    if (strcmp(sin->subtype, "vdagent") == 0) {
> +        return;
> +    }
> +
> +    if ((scd->chr->opened && connected) ||
> +        (!scd->chr->opened && !connected)) {
> +        return;
> +    }
> +
> +    if (connected) {
> +        scd->chr->opened = 1;
> +        event = CHR_EVENT_OPENED;
> +    } else {
> +        scd->chr->opened = 0;
> +        event = CHR_EVENT_CLOSED;
> +    }
> +
> +    if (scd->chr->chr_event) {
> +        scd->chr->chr_event(scd->chr->handler_opaque, event);
> +    }
> +}
> +
>  static SpiceCharDeviceInterface vmc_interface = {
>      .base.type          = SPICE_INTERFACE_CHAR_DEVICE,
>      .base.description   = "spice virtual channel char device",
>      .base.major_version = SPICE_INTERFACE_CHAR_DEVICE_MAJOR,
>      .base.minor_version = SPICE_INTERFACE_CHAR_DEVICE_MINOR,
> +    .state              = vmc_state,
>      .write              = vmc_write,
>      .read               = vmc_read,
>  };
> @@ -222,7 +259,10 @@ CharDriverState *qemu_chr_open_spice(QemuOpts *opts)
>      chr->chr_guest_close = spice_chr_guest_close;
>      s->unblock_timer = qemu_new_timer_ms(vm_clock, spice_chr_unblock, s);
>  
> -    qemu_chr_generic_open(chr);
> +    /* See comment in vmc_state() */
> +    if (strcmp(subtype, "vdagent") == 0) {
> +        qemu_chr_generic_open(chr);
> +    }
>  
>      return chr;
>  }
> -- 
> 1.7.5.1
> 
>
Gerd Hoffmann - July 22, 2011, 12:40 p.m.
> ACK.
>
> Another option would be to not call the state callback for the vdagent
> subtype from spice-server, because like you say it ignores the continued
> communication between the agent and the server. Then you could remove
> the subtype check.

We can even #ifdef the check using SPICE_SERVER_VERSION once the 
spice-server change is committed, and in a year or so we can raise the 
minimum required spice-server version and remove some #ifdef clutter.

cheers,
   Gerd
Hans de Goede - July 24, 2011, 6:15 p.m.
Hi,

On 07/22/2011 02:40 PM, Gerd Hoffmann wrote:
>> ACK.
>>
>> Another option would be to not call the state callback for the vdagent
>> subtype from spice-server, because like you say it ignores the continued
>> communication between the agent and the server. Then you could remove
>> the subtype check.
>
> We can even #ifdef the check using SPICE_SERVER_VERSION once the spice-server change is committed, and in a year or so we can raise the minimum required spice-server version and remove some #ifdef clutter.
>

Yeah that is more or less what I proposed in the last mail in the spice-devel
mailinglist thread, I'm glad you agree. I'll prepare a patch to remove
the state calls from the spice-server vdagent handling.

Regards,

Hans

Patch

diff --git a/spice-qemu-char.c b/spice-qemu-char.c
index ce75e91..a472f7e 100644
--- a/spice-qemu-char.c
+++ b/spice-qemu-char.c
@@ -89,11 +89,48 @@  static int vmc_read(SpiceCharDeviceInstance *sin, uint8_t *buf, int len)
     return bytes;
 }
 
+static void vmc_state(SpiceCharDeviceInstance *sin, int connected)
+{
+    SpiceCharDriver *scd = container_of(sin, SpiceCharDriver, sin);
+    int event;
+
+    /*
+     * spice-server calls the state callback for the agent channel when the
+     * spice client connects / disconnects. Given that not the client but
+     * the server is doing the parsing of the messages this is wrong as the
+     * server is still listening. Worse, this causes the parser in the server
+     * to go out of sync, so we ignore state calls for subtype vdagent
+     * spicevmc chardevs. For the full story see:
+     * http://lists.freedesktop.org/archives/spice-devel/2011-July/004837.html
+     */
+    if (strcmp(sin->subtype, "vdagent") == 0) {
+        return;
+    }
+
+    if ((scd->chr->opened && connected) ||
+        (!scd->chr->opened && !connected)) {
+        return;
+    }
+
+    if (connected) {
+        scd->chr->opened = 1;
+        event = CHR_EVENT_OPENED;
+    } else {
+        scd->chr->opened = 0;
+        event = CHR_EVENT_CLOSED;
+    }
+
+    if (scd->chr->chr_event) {
+        scd->chr->chr_event(scd->chr->handler_opaque, event);
+    }
+}
+
 static SpiceCharDeviceInterface vmc_interface = {
     .base.type          = SPICE_INTERFACE_CHAR_DEVICE,
     .base.description   = "spice virtual channel char device",
     .base.major_version = SPICE_INTERFACE_CHAR_DEVICE_MAJOR,
     .base.minor_version = SPICE_INTERFACE_CHAR_DEVICE_MINOR,
+    .state              = vmc_state,
     .write              = vmc_write,
     .read               = vmc_read,
 };
@@ -222,7 +259,10 @@  CharDriverState *qemu_chr_open_spice(QemuOpts *opts)
     chr->chr_guest_close = spice_chr_guest_close;
     s->unblock_timer = qemu_new_timer_ms(vm_clock, spice_chr_unblock, s);
 
-    qemu_chr_generic_open(chr);
+    /* See comment in vmc_state() */
+    if (strcmp(subtype, "vdagent") == 0) {
+        qemu_chr_generic_open(chr);
+    }
 
     return chr;
 }