Skip to contentSkip to conversation

Bluetooth: Cleanups and LE advertising support

PATCH In review [email protected] 3 participants 11 messages Sep 25, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v2 cover Bluetooth: Cleanups and LE advertising support

Hi,

Here's an updated set based on feedback on the first one. The mgmt_valid_hdev patch has been dropped since it's a trivial one and I didn't want the debate over it to slow down the overall progress of this set.

The main difference to the initial set_le and set_advertising code is that I now let the hci_event.c handlers take care of setting the right values of the dev_flags bits. This works fine since these handlers are always executed before the async request callbacks which then get called with dev_flags already having the correct value. The benefit of using the event handlers like this is that we retain at least partial support for raw HCI access (hciconfig) by ensuring that the kernel state flags are always correct.

Johan

----------------------------------------------------------------
Johan Hedberg (7):
Bluetooth: Add clarifying comment to bt_sock_wait_state()
Bluetooth: Clean up socket locking in l2cap_sock_recvmsg
Bluetooth: Fix busy return for mgmt_set_powered in some cases
Bluetooth: Move mgmt response convenience functions to a better location
Bluetooth: Use async request for LE enable/disable
Bluetooth: Add new mgmt setting for LE advertising
Bluetooth: Add new mgmt_set_advertising command

 include/net/bluetooth/hci_core.h |   1 -
 include/net/bluetooth/mgmt.h     |   3 +
 net/bluetooth/af_bluetooth.c     |   1 +
 net/bluetooth/hci_event.c        |  12 +-
 net/bluetooth/l2cap_sock.c       |   4 +-
 net/bluetooth/mgmt.c             | 257 ++++++++++++++++++++++++++------------
 6 files changed, 191 insertions(+), 87 deletions(-)
#1 lore raw

PATCH v2 1/7 Bluetooth: Add clarifying comment to bt_sock_wait_state()

From: Johan Hedberg <[email protected]>

The bt_sock_wait_state requires the sk lock to be held (through lock_sock) so document it clearly in the code.

---
 net/bluetooth/af_bluetooth.c | 1 +
 1 file changed, 1 insertion(+)
net/bluetooth/af_bluetooth.c +1 -0
index c600631..e6e1278 100644
@@ -490,6 +490,7 @@ int bt_sock_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 }
 EXPORT_SYMBOL(bt_sock_ioctl);
 
+/* This function expects the sk lock to be held when called */
 int bt_sock_wait_state(struct sock *sk, int state, unsigned long timeo)
 {
 	DECLARE_WAITQUEUE(wait, current);
Signature · 1 line
1.8.3.1
#2 lore raw

PATCH v2 2/7 Bluetooth: Clean up socket locking in l2cap_sock_recvmsg

In reply to [email protected]

From: Johan Hedberg <[email protected]>

This patch cleans up the locking login in l2cap_sock_recvmsg by pairing up each lock_sock call with a release_sock call. The function already has a "done" label that handles releasing the socket and returning from the function so the fix is rather simple.

---
 net/bluetooth/l2cap_sock.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/bluetooth/l2cap_sock.c +2 -2
index ad95b42..c85537c 100644
@@ -805,8 +805,8 @@ static int l2cap_sock_recvmsg(struct kiocb *iocb, struct socket *sock,
 		pi->chan->state = BT_CONFIG;
 
 		__l2cap_connect_rsp_defer(pi->chan);
-		release_sock(sk);
-		return 0;
+		err = 0;
+		goto done;
 	}
 
 	release_sock(sk);
Signature · 1 line
1.8.3.1
maintainer Acked-by on v2 2/7
#3 lore raw

Re: [PATCH v2 2/7] Bluetooth: Clean up socket locking in l2cap_sock_recvmsg

Hi Johan,

Quoted 9 lines

This patch cleans up the locking login in l2cap_sock_recvmsg by pairing up each lock_sock call with a release_sock call. The function already has a "done" label that handles releasing the socket and returning from the function so the fix is rather simple.

---
net/bluetooth/l2cap_sock.c | 4 ++--

1 file changed, 2 insertions(+), 2 deletions(-)

Regards

Marcel

#4 lore raw

PATCH v2 3/7 Bluetooth: Fix busy return for mgmt_set_powered in some cases

In reply to [email protected]

From: Johan Hedberg <[email protected]>

We should return a "busy" error always when there is another mgmt_set_powered operation in progress. Previously when powering on while the auto off timer was still set the code could have let two or more pending power on commands to be queued. This patch fixes the issue by moving the check for duplicate commands to an earlier point in the set_powered handler.

---
 net/bluetooth/mgmt.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
net/bluetooth/mgmt.c +6 -6
index 3070e77..85bfa21 100644
@@ -807,6 +807,12 @@ static int set_powered(struct sock *sk, struct hci_dev *hdev, void *data,
 
 	hci_dev_lock(hdev);
 
+	if (mgmt_pending_find(MGMT_OP_SET_POWERED, hdev)) {
+		err = cmd_status(sk, hdev->id, MGMT_OP_SET_POWERED,
+				 MGMT_STATUS_BUSY);
+		goto failed;
+	}
+
 	if (test_and_clear_bit(HCI_AUTO_OFF, &hdev->dev_flags)) {
 		cancel_delayed_work(&hdev->power_off);
 
@@ -823,12 +829,6 @@ static int set_powered(struct sock *sk, struct hci_dev *hdev, void *data,
 		goto failed;
 	}
 
-	if (mgmt_pending_find(MGMT_OP_SET_POWERED, hdev)) {
-		err = cmd_status(sk, hdev->id, MGMT_OP_SET_POWERED,
-				 MGMT_STATUS_BUSY);
-		goto failed;
-	}
-
 	cmd = mgmt_pending_add(sk, MGMT_OP_SET_POWERED, hdev, data, len);
 	if (!cmd) {
 		err = -ENOMEM;
Signature · 1 line
1.8.3.1
#5 lore raw

PATCH v2 4/7 Bluetooth: Move mgmt response convenience functions to a better location

In reply to [email protected]

From: Johan Hedberg <[email protected]>

The settings_rsp and cmd_status_rsp functions can be useful for all mgmt command handlers when asynchronous request callbacks are used. They will e.g. be used by subsequent patches to change set_le to use an async request as well as a new set_advertising command. Therefore, move them higher up in the mgmt.c file to avoid unnecessary forward declarations or mixing this trivial change with other patches.

---
 net/bluetooth/mgmt.c | 60 ++++++++++++++++++++++++++--------------------------
 1 file changed, 30 insertions(+), 30 deletions(-)
net/bluetooth/mgmt.c +30 -30
index 85bfa21..61d4b19 100644
@@ -886,6 +886,36 @@ static int new_settings(struct hci_dev *hdev, struct sock *skip)
 	return mgmt_event(MGMT_EV_NEW_SETTINGS, hdev, &ev, sizeof(ev), skip);
 }
 
+struct cmd_lookup {
+	struct sock *sk;
+	struct hci_dev *hdev;
+	u8 mgmt_status;
+};
+
+static void settings_rsp(struct pending_cmd *cmd, void *data)
+{
+	struct cmd_lookup *match = data;
+
+	send_settings_rsp(cmd->sk, cmd->opcode, match->hdev);
+
+	list_del(&cmd->list);
+
+	if (match->sk == NULL) {
+		match->sk = cmd->sk;
+		sock_hold(match->sk);
+	}
+
+	mgmt_pending_free(cmd);
+}
+
+static void cmd_status_rsp(struct pending_cmd *cmd, void *data)
+{
+	u8 *status = data;
+
+	cmd_status(cmd->sk, cmd->index, cmd->opcode, *status);
+	mgmt_pending_remove(cmd);
+}
+
 static int set_discoverable(struct sock *sk, struct hci_dev *hdev, void *data,
 			    u16 len)
 {
@@ -3374,14 +3404,6 @@ done:
 	return err;
 }
 
-static void cmd_status_rsp(struct pending_cmd *cmd, void *data)
-{
-	u8 *status = data;
-
-	cmd_status(cmd->sk, cmd->index, cmd->opcode, *status);
-	mgmt_pending_remove(cmd);
-}
-
 int mgmt_index_added(struct hci_dev *hdev)
 {
 	if (!mgmt_valid_hdev(hdev))
@@ -3402,28 +3424,6 @@ int mgmt_index_removed(struct hci_dev *hdev)
 	return mgmt_event(MGMT_EV_INDEX_REMOVED, hdev, NULL, 0, NULL);
 }
 
-struct cmd_lookup {
-	struct sock *sk;
-	struct hci_dev *hdev;
-	u8 mgmt_status;
-};
-
-static void settings_rsp(struct pending_cmd *cmd, void *data)
-{
-	struct cmd_lookup *match = data;
-
-	send_settings_rsp(cmd->sk, cmd->opcode, match->hdev);
-
-	list_del(&cmd->list);
-
-	if (match->sk == NULL) {
-		match->sk = cmd->sk;
-		sock_hold(match->sk);
-	}
-
-	mgmt_pending_free(cmd);
-}
-
 static void set_bredr_scan(struct hci_request *req)
 {
 	struct hci_dev *hdev = req->hdev;
Signature · 1 line
1.8.3.1
#6 lore raw

PATCH v2 5/7 Bluetooth: Use async request for LE enable/disable

In reply to [email protected]

From: Johan Hedberg <[email protected]>

This patch updates the code to use an asynchronous request for handling the enabling and disabling of LE support. This refactoring is necessary as a preparation for adding advertising support, since when LE is disabled we should also disable advertising, and the cleanest way to do this is to perform the two respective HCI commands in the same asynchronous request.

---
 include/net/bluetooth/hci_core.h |  1 -
 net/bluetooth/hci_event.c        | 11 +++----
 net/bluetooth/mgmt.c             | 67 ++++++++++++++++------------------------
 3 files changed, 32 insertions(+), 47 deletions(-)
include/net/bluetooth/hci_core.h +0 -1
index 3ede820..26cc9f7 100644
@@ -1168,7 +1168,6 @@ int mgmt_set_class_of_dev_complete(struct hci_dev *hdev, u8 *dev_class,
 int mgmt_set_local_name_complete(struct hci_dev *hdev, u8 *name, u8 status);
 int mgmt_read_local_oob_data_reply_complete(struct hci_dev *hdev, u8 *hash,
 					    u8 *randomizer, u8 status);
-int mgmt_le_enable_complete(struct hci_dev *hdev, u8 enable, u8 status);
 int mgmt_device_found(struct hci_dev *hdev, bdaddr_t *bdaddr, u8 link_type,
 		      u8 addr_type, u8 *dev_class, s8 rssi, u8 cfm_name,
 		      u8 ssp, u8 *eir, u16 eir_len);
net/bluetooth/hci_event.c +5 -6
index 94aab73..48db81f 100644
@@ -994,20 +994,19 @@ static void hci_cc_write_le_host_supported(struct hci_dev *hdev,
 		return;
 
 	if (!status) {
-		if (sent->le)
+		if (sent->le) {
 			hdev->features[1][0] |= LMP_HOST_LE;
-		else
+			set_bit(HCI_LE_ENABLED, &hdev->dev_flags);
+		} else {
 			hdev->features[1][0] &= ~LMP_HOST_LE;
+			clear_bit(HCI_LE_ENABLED, &hdev->dev_flags);
+		}
 
 		if (sent->simul)
 			hdev->features[1][0] |= LMP_HOST_LE_BREDR;
 		else
 			hdev->features[1][0] &= ~LMP_HOST_LE_BREDR;
 	}
-
-	if (test_bit(HCI_MGMT, &hdev->dev_flags) &&
-	    !test_bit(HCI_INIT, &hdev->flags))
-		mgmt_le_enable_complete(hdev, sent->le, status);
 }
 
 static void hci_cc_write_remote_amp_assoc(struct hci_dev *hdev,
net/bluetooth/mgmt.c +27 -40

90 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#7 lore raw

PATCH v2 6/7 Bluetooth: Add new mgmt setting for LE advertising

In reply to [email protected]

From: Johan Hedberg <[email protected]>

This patch adds a new mgmt setting for LE advertising and hooks up the necessary places in the mgmt code to operate on the HCI_LE_PERIPHERAL flag (which corresponds to this setting). This patch does not yet add any new command for enabling the setting - that is left for a subsequent patch.

---
 include/net/bluetooth/mgmt.h |  1 +
 net/bluetooth/hci_event.c    |  1 +
 net/bluetooth/mgmt.c         | 21 ++++++++++++++++++++-
 3 files changed, 22 insertions(+), 1 deletion(-)
include/net/bluetooth/mgmt.h +1 -0
index 9944c3e..6cc72b6 100644
@@ -93,6 +93,7 @@ struct mgmt_rp_read_index_list {
 #define MGMT_SETTING_BREDR		0x00000080
 #define MGMT_SETTING_HS			0x00000100
 #define MGMT_SETTING_LE			0x00000200
+#define MGMT_SETTING_ADVERTISING	0x00000400
 
 #define MGMT_OP_READ_INFO		0x0004
 #define MGMT_READ_INFO_SIZE		0
net/bluetooth/hci_event.c +1 -0
index 48db81f..917c7c8 100644
@@ -1000,6 +1000,7 @@ static void hci_cc_write_le_host_supported(struct hci_dev *hdev,
 		} else {
 			hdev->features[1][0] &= ~LMP_HOST_LE;
 			clear_bit(HCI_LE_ENABLED, &hdev->dev_flags);
+			clear_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags);
 		}
 
 		if (sent->simul)
net/bluetooth/mgmt.c +20 -1
index 4c3984e..9a2faa3 100644
@@ -384,8 +384,10 @@ static u32 get_supported_settings(struct hci_dev *hdev)
 	if (enable_hs)
 		settings |= MGMT_SETTING_HS;
 
-	if (lmp_le_capable(hdev))
+	if (lmp_le_capable(hdev)) {
 		settings |= MGMT_SETTING_LE;
+		settings |= MGMT_SETTING_ADVERTISING;
+	}
 
 	return settings;
 }
@@ -424,6 +426,9 @@ static u32 get_current_settings(struct hci_dev *hdev)
 	if (test_bit(HCI_HS_ENABLED, &hdev->dev_flags))
 		settings |= MGMT_SETTING_HS;
 
+	if (test_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags))
+		settings |= MGMT_SETTING_ADVERTISING;
+
 	return settings;
 }
 
@@ -1411,6 +1416,11 @@ static int set_le(struct sock *sk, struct hci_dev *hdev, void *data, u16 len)
 			changed = true;
 		}
 
+		if (!val && test_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags)) {
+			clear_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags);
+			changed = true;
+		}
+
 		err = send_settings_rsp(sk, MGMT_OP_SET_LE, hdev);
 		if (err < 0)
 			goto unlock;
@@ -1442,6 +1452,9 @@ static int set_le(struct sock *sk, struct hci_dev *hdev, void *data, u16 len)
 
 	hci_req_init(&req, hdev);
 
+	if (test_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags) && !val)
+		hci_req_add(&req, HCI_OP_LE_SET_ADV_ENABLE, sizeof(val), &val);
+
 	hci_req_add(&req, HCI_OP_WRITE_LE_HOST_SUPPORTED, sizeof(hci_cp),
 		    &hci_cp);
 
@@ -3517,6 +3530,12 @@ static int powered_update_hci(struct hci_dev *hdev)
 				    sizeof(cp), &cp);
 	}
 
+	if (test_bit(HCI_LE_PERIPHERAL, &hdev->dev_flags)) {
+		u8 adv = 0x01;
+
+		hci_req_add(&req, HCI_OP_LE_SET_ADV_ENABLE, sizeof(adv), &adv);
+	}
+
 	link_sec = test_bit(HCI_LINK_SECURITY, &hdev->dev_flags);
 	if (link_sec != test_bit(HCI_AUTH, &hdev->flags))
 		hci_req_add(&req, HCI_OP_WRITE_AUTH_ENABLE,
Signature · 1 line
1.8.3.1
#8 lore raw

PATCH v2 7/7 Bluetooth: Add new mgmt_set_advertising command

In reply to [email protected]

From: Johan Hedberg <[email protected]>

This patch adds a new mgmt command for enabling and disabling LE advertising. The command depends on the LE setting being enabled first and will return a "rejected" response otherwise. The patch also adds safeguards so that there will ever only be one set_le or set_advertising command pending per adapter.

The response handling and new_settings event sending is done in an asynchronous request callback, meaning raw HCI access from user space to enable advertising (e.g. hciconfig leadv) will not trigger the new_settings event. This is intentional since trying to support mixed raw HCI and mgmt access would mean adding extra state tracking or new helper functions, essentially negating the benefit of using the asynchronous request framework. The HCI_LE_ENABLED and HCI_LE_PERIPHERAL flags however are updated correctly even with raw HCI access so this will not completely break subsequent access over mgmt.

---
 include/net/bluetooth/mgmt.h |  2 +
 net/bluetooth/mgmt.c         | 97 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 98 insertions(+), 1 deletion(-)
include/net/bluetooth/mgmt.h +2 -0
index 6cc72b6..421d763 100644
@@ -352,6 +352,8 @@ struct mgmt_cp_set_device_id {
 } __packed;
 #define MGMT_SET_DEVICE_ID_SIZE		8
 
+#define MGMT_OP_SET_ADVERTISING		0x0029
+
 #define MGMT_EV_CMD_COMPLETE		0x0001
 struct mgmt_ev_cmd_complete {
 	__le16	opcode;
net/bluetooth/mgmt.c +96 -1

121 lines not on this page · raw message

Signature · 1 line
1.8.3.1
maintainer Acked-by on v2 7/7
#9 lore raw

Re: [PATCH v2 7/7] Bluetooth: Add new mgmt_set_advertising command

Hi Johan,

Quoted 16 lines

This patch adds a new mgmt command for enabling and disabling LE advertising. The command depends on the LE setting being enabled first and will return a "rejected" response otherwise. The patch also adds safeguards so that there will ever only be one set_le or set_advertising command pending per adapter.

The response handling and new_settings event sending is done in an asynchronous request callback, meaning raw HCI access from user space to enable advertising (e.g. hciconfig leadv) will not trigger the new_settings event. This is intentional since trying to support mixed raw HCI and mgmt access would mean adding extra state tracking or new helper functions, essentially negating the benefit of using the asynchronous request framework. The HCI_LE_ENABLED and HCI_LE_PERIPHERAL flags however are updated correctly even with raw HCI access so this will not completely break subsequent access over mgmt.

Quoted 5 lines

---
include/net/bluetooth/mgmt.h | 2 +
net/bluetooth/mgmt.c | 97 +++++++++++++++++++++++++++++++++++++++++++-

2 files changed, 98 insertions(+), 1 deletion(-)

Regards

Marcel

#10 lore raw

Re: [PATCH v2 7/7] Bluetooth: Add new mgmt_set_advertising command

In reply to [email protected]

Hi Johan,

2013-09-25 [email protected] <[email protected]>:

Quoted 18 lines

From: Johan Hedberg <[email protected]>

This patch adds a new mgmt command for enabling and disabling LE advertising. The command depends on the LE setting being enabled first and will return a "rejected" response otherwise. The patch also adds safeguards so that there will ever only be one set_le or set_advertising command pending per adapter.

The response handling and new_settings event sending is done in an asynchronous request callback, meaning raw HCI access from user space to enable advertising (e.g. hciconfig leadv) will not trigger the new_settings event. This is intentional since trying to support mixed raw HCI and mgmt access would mean adding extra state tracking or new helper functions, essentially negating the benefit of using the asynchronous request framework. The HCI_LE_ENABLED and HCI_LE_PERIPHERAL flags however are updated correctly even with raw HCI access so this will not completely break subsequent access over mgmt.

Quoted 5 lines
---
 include/net/bluetooth/mgmt.h |  2 +
 net/bluetooth/mgmt.c         | 97 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 98 insertions(+), 1 deletion(-)

All patches have been applied to bluetooth-next. Thanks.

Gustavo