Re: [PATCH BlueZ v3 06/14] core: mutually exclude concurrent connections
From: Joao Paulo Rechi Vita <hidden>
Date: 2012-08-30 22:08:28
On Fri, Aug 17, 2012 at 4:51 AM, Johan Hedberg [off-list ref] wrote:
Hi, On Thu, Aug 16, 2012, João Paulo Rechi Vita wrote:quoted
@@ -133,8 +133,10 @@ struct btd_adapter { + GSList *connecting_list; /* Pending connects */ + gboolean connecting; /* Connect active */What's the relationship with the above two? Isn't connecting always true when g_slist_length(connecting_list) > 0 and false when the list is empty?
The connecting_list keeps track of devices that have been found but for which a connect has not been issued yet. This was mainly used for 1) not trying to connect twice to the same device (on two subsequent device_found events), and 2) to know if there are still any device waiting for a connect when one connect finishes, to device if we should restart scanning or not. Reviewing this logic I've been able to simplify it a little bit, using a counter (adapter->waiting_to_connect) to address problem #2, and #1 could be handled only using the adapter->connect_list.
quoted
static gboolean connect_pending_cb(gpointer user_data) { struct btd_device *device = user_data; struct btd_adapter *adapter = device_get_adapter(device); + GIOChannel *io; /* in the future we may want to check here if the controller supports * scanning and connecting at the same time */ if (adapter->discovering) return TRUE; - device_att_connect(device); + if (adapter->connecting) + return TRUE; + + adapter->connecting = TRUE; + io = device_att_connect(device); + g_io_add_watch(io, G_IO_OUT | G_IO_ERR, clean_connecting_state, + btd_device_ref(device)); return FALSE; }Since you're not storing "io" after the function returns you should be doing a g_io_channel_unref() after calling g_io_add_watch(). Looking at device_att_connect it seems like it's missing a g_io_channel_ref (either when assigning to device->att_io or when returning from the function).quoted
+ adapter->connecting_list = g_slist_append( + adapter->connecting_list, device);Missing btd_device_unref()?
The whole ref counting has been reviewed as well, with the new logic modifications mentioned above. -- João Paulo Rechi Vita Openbossa Labs - INdT