[Spice-devel] [PATCH v2 9/9] Change RedCharDevicePrivate::clients to GList
Jonathon Jongsma
jjongsma at redhat.com
Thu Sep 15 19:44:31 UTC 2016
On Thu, 2016-09-15 at 11:16 -0400, Frediano Ziglio wrote:
> >
> >
> > More Ring cleanup. At the moment, we only support a single client,
> > so
> > this is only a one-element list
> > ---
> > Changes in v2:
> > - simplified red_char_device_finalize() to avoid using
> > g_list_last()
> >
> > server/char-device.c | 68
> > ++++++++++++++++++++--------------------------------
> > 1 file changed, 26 insertions(+), 42 deletions(-)
> >
> > diff --git a/server/char-device.c b/server/char-device.c
> > index f826524..1b99aa9 100644
> > --- a/server/char-device.c
> > +++ b/server/char-device.c
> > @@ -32,7 +32,6 @@
> >
> > typedef struct RedCharDeviceClient RedCharDeviceClient;
> > struct RedCharDeviceClient {
> > - RingItem link;
> > RedCharDevice *dev;
> > RedClient *client;
> > int do_flow_control;
> > @@ -58,8 +57,7 @@ struct RedCharDevicePrivate {
> > SpiceTimer *write_to_dev_timer;
> > uint64_t num_self_tokens;
> >
> > - Ring clients; /* list of RedCharDeviceClient */
> > - uint32_t num_clients;
> > + GList *clients; /* list of RedCharDeviceClient */
> >
> > uint64_t client_tokens_interval; /* frequency of returning
> > tokens to the
> > client */
> > SpiceCharDeviceInstance *sin;
> > @@ -207,8 +205,7 @@ static void
> > red_char_device_client_free(RedCharDevice
> > *dev,
> > dev->priv->cur_write_buf->client = NULL;
> > }
> >
> > - dev->priv->num_clients--;
> > - ring_remove(&dev_client->link);
> > + dev->priv->clients = g_list_remove(dev->priv->clients,
> > dev_client);
> > free(dev_client);
> > }
> >
> > @@ -222,12 +219,11 @@ static void
> > red_char_device_handle_client_overflow(RedCharDeviceClient
> > *dev_clie
> > static RedCharDeviceClient
> > *red_char_device_client_find(RedCharDevice *dev,
> > RedClient
> > *client)
> > {
> > - RingItem *item;
> > + GList *item;
> >
> > - RING_FOREACH(item, &dev->priv->clients) {
> > - RedCharDeviceClient *dev_client;
> > + for (item = dev->priv->clients; item != NULL; item = item-
> > >next) {
> > + RedCharDeviceClient *dev_client = item->data;
> >
> > - dev_client = SPICE_CONTAINEROF(item, RedCharDeviceClient,
> > link);
> > if (dev_client->client == client) {
> > return dev_client;
> > }
> > @@ -253,13 +249,11 @@ static int
> > red_char_device_can_send_to_client(RedCharDeviceClient *dev_client)
> >
> > static uint64_t red_char_device_max_send_tokens(RedCharDevice
> > *dev)
> > {
> > - RingItem *item;
> > + GList *item;
> > uint64_t max = 0;
> >
> > - RING_FOREACH(item, &dev->priv->clients) {
> > - RedCharDeviceClient *dev_client;
> > -
> > - dev_client = SPICE_CONTAINEROF(item, RedCharDeviceClient,
> > link);
> > + for (item = dev->priv->clients; item != NULL; item = item-
> > >next) {
> > + RedCharDeviceClient *dev_client = item->data;
> >
> > if (!dev_client->do_flow_control) {
> > max = ~0;
> > @@ -295,12 +289,11 @@ static void
> > red_char_device_add_msg_to_client_queue(RedCharDeviceClient
> > *dev_cli
> > static void red_char_device_send_msg_to_clients(RedCharDevice
> > *dev,
> > RedPipeItem *msg)
> > {
> > - RingItem *item, *next;
> > + GList *l;
> >
> > - RING_FOREACH_SAFE(item, next, &dev->priv->clients) {
> > - RedCharDeviceClient *dev_client;
> > + for (l = dev->priv->clients; l != NULL; l = l->next) {
> > + RedCharDeviceClient *dev_client = l->data;
> >
> > - dev_client = SPICE_CONTAINEROF(item, RedCharDeviceClient,
> > link);
> > if (red_char_device_can_send_to_client(dev_client)) {
> > dev_client->num_send_tokens--;
> > spice_assert(g_queue_is_empty(dev_client-
> > >send_queue));
>
> The _SAFE loop here was necessary as there are combinations were
> the client is removed from the list during the loop.
> This will make l vanish before l->next is read.
> Even if there is only a client this can happen.
Hmm, I couldn't find a code path where this happened. But there is a
comment within the loop suggesting that it can happen, so it's better
to use the safer loop style. Sorry for changing it without enough
diligence.
jonathon
More information about the Spice-devel
mailing list