| Message ID | 20220810164627.1.Id730b98f188a504d9835b96ddcbc83d49a70bb36@changeid (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | Bluetooth: Honor name resolve evt regardless of discov state | expand |
Hi Archie, On Wed, Aug 10, 2022 at 1:47 AM Archie Pusaka <apusaka@google.com> wrote: > > From: Archie Pusaka <apusaka@chromium.org> > > Currently, we don't update the name resolving cache when receiving > a name resolve event if the discovery phase is not in the resolving > stage. > > However, if the user connect to a device while we are still resolving > remote name for another device, discovery will be stopped, and because > we are no longer in the discovery resolving phase, the corresponding > remote name event will be ignored, and thus the device being resolved > will stuck in NAME_PENDING state. > > If discovery is then restarted and then stopped, this will cause us to > try cancelling the name resolve of the same device again, which is > incorrect and might upset the controller. Please add the Fixes tag. > Signed-off-by: Archie Pusaka <apusaka@chromium.org> > Reviewed-by: Ying Hsu <yinghsu@chromium.org> > > --- > The following steps are performed: > (1) Prepare 2 classic peer devices that needs RNR. Put device A > closer to DUT and device B (much) farther from DUT. > (2) Remove all cache and previous connection from DUT > (3) Put both peers into pairing mode, then start scanning on DUT > (4) After ~8 sec, turn off peer B. > *This is done so DUT can discover peer B (discovery time is 10s), > but it hasn't started RNR. Peer is turned off to buy us the max > time in the RNR phase (5s). > (5) Immediately as device A is shown on UI, click to connect. > *We thus know that the DUT is in the RNR phase and trying to > resolve the name of peer B when we initiate connection to peer A. > (6) Forget peer A. > (7) Restart scan and stop scan. > *Before the CL, stop scan is broken because we will try to cancel > a nonexistent RNR > (8) Restart scan again. Observe DUT can scan normally. > > > net/bluetooth/hci_event.c | 17 ++++++++++------- > 1 file changed, 10 insertions(+), 7 deletions(-) > > diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c > index 395c6479456f..95e145e278c9 100644 > --- a/net/bluetooth/hci_event.c > +++ b/net/bluetooth/hci_event.c > @@ -2453,6 +2453,16 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > !test_and_set_bit(HCI_CONN_MGMT_CONNECTED, &conn->flags)) > mgmt_device_connected(hdev, conn, name, name_len); > > + e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > + > + if (e) { > + list_del(&e->list); > + > + e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > + mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > + name, name_len); > + } > + > if (discov->state == DISCOVERY_STOPPED) > return; > > @@ -2462,7 +2472,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > if (discov->state != DISCOVERY_RESOLVING) > return; > > - e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > /* If the device was not found in a list of found devices names of which > * are pending. there is no need to continue resolving a next name as it > * will be done upon receiving another Remote Name Request Complete > @@ -2470,12 +2479,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > if (!e) > return; > > - list_del(&e->list); > - > - e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > - mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > - name, name_len); > - > if (hci_resolve_next_name(hdev)) > return; > > -- > 2.37.1.595.g718a3a8f04-goog >
Hi Luiz, On Thu, 11 Aug 2022 at 03:58, Luiz Augusto von Dentz <luiz.dentz@gmail.com> wrote: > > Hi Archie, > > On Wed, Aug 10, 2022 at 1:47 AM Archie Pusaka <apusaka@google.com> wrote: > > > > From: Archie Pusaka <apusaka@chromium.org> > > > > Currently, we don't update the name resolving cache when receiving > > a name resolve event if the discovery phase is not in the resolving > > stage. > > > > However, if the user connect to a device while we are still resolving > > remote name for another device, discovery will be stopped, and because > > we are no longer in the discovery resolving phase, the corresponding > > remote name event will be ignored, and thus the device being resolved > > will stuck in NAME_PENDING state. > > > > If discovery is then restarted and then stopped, this will cause us to > > try cancelling the name resolve of the same device again, which is > > incorrect and might upset the controller. > > Please add the Fixes tag. Unfortunately I don't know when was the issue introduced, I don't even know whether this is a recent issue or an old one. Looking back, this part of the code has stayed this way since 2012. Do I still need to add the fixes tag? If so, does it have to be accurate? > > > Signed-off-by: Archie Pusaka <apusaka@chromium.org> > > Reviewed-by: Ying Hsu <yinghsu@chromium.org> > > > > --- > > The following steps are performed: > > (1) Prepare 2 classic peer devices that needs RNR. Put device A > > closer to DUT and device B (much) farther from DUT. > > (2) Remove all cache and previous connection from DUT > > (3) Put both peers into pairing mode, then start scanning on DUT > > (4) After ~8 sec, turn off peer B. > > *This is done so DUT can discover peer B (discovery time is 10s), > > but it hasn't started RNR. Peer is turned off to buy us the max > > time in the RNR phase (5s). > > (5) Immediately as device A is shown on UI, click to connect. > > *We thus know that the DUT is in the RNR phase and trying to > > resolve the name of peer B when we initiate connection to peer A. > > (6) Forget peer A. > > (7) Restart scan and stop scan. > > *Before the CL, stop scan is broken because we will try to cancel > > a nonexistent RNR > > (8) Restart scan again. Observe DUT can scan normally. > > > > > > net/bluetooth/hci_event.c | 17 ++++++++++------- > > 1 file changed, 10 insertions(+), 7 deletions(-) > > > > diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c > > index 395c6479456f..95e145e278c9 100644 > > --- a/net/bluetooth/hci_event.c > > +++ b/net/bluetooth/hci_event.c > > @@ -2453,6 +2453,16 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > !test_and_set_bit(HCI_CONN_MGMT_CONNECTED, &conn->flags)) > > mgmt_device_connected(hdev, conn, name, name_len); > > > > + e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > > + > > + if (e) { > > + list_del(&e->list); > > + > > + e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > > + mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > > + name, name_len); > > + } > > + > > if (discov->state == DISCOVERY_STOPPED) > > return; > > > > @@ -2462,7 +2472,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > if (discov->state != DISCOVERY_RESOLVING) > > return; > > > > - e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > > /* If the device was not found in a list of found devices names of which > > * are pending. there is no need to continue resolving a next name as it > > * will be done upon receiving another Remote Name Request Complete > > @@ -2470,12 +2479,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > if (!e) > > return; > > > > - list_del(&e->list); > > - > > - e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > > - mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > > - name, name_len); > > - > > if (hci_resolve_next_name(hdev)) > > return; > > > > -- > > 2.37.1.595.g718a3a8f04-goog > > > > > -- > Luiz Augusto von Dentz Thanks, Archie
Hi Archie, On Thu, Aug 11, 2022 at 12:00 AM Archie Pusaka <apusaka@google.com> wrote: > > Hi Luiz, > > On Thu, 11 Aug 2022 at 03:58, Luiz Augusto von Dentz > <luiz.dentz@gmail.com> wrote: > > > > Hi Archie, > > > > On Wed, Aug 10, 2022 at 1:47 AM Archie Pusaka <apusaka@google.com> wrote: > > > > > > From: Archie Pusaka <apusaka@chromium.org> > > > > > > Currently, we don't update the name resolving cache when receiving > > > a name resolve event if the discovery phase is not in the resolving > > > stage. > > > > > > However, if the user connect to a device while we are still resolving > > > remote name for another device, discovery will be stopped, and because > > > we are no longer in the discovery resolving phase, the corresponding > > > remote name event will be ignored, and thus the device being resolved > > > will stuck in NAME_PENDING state. > > > > > > If discovery is then restarted and then stopped, this will cause us to > > > try cancelling the name resolve of the same device again, which is > > > incorrect and might upset the controller. > > > > Please add the Fixes tag. > > Unfortunately I don't know when was the issue introduced, I don't even > know whether this is a recent issue or an old one. > Looking back, this part of the code has stayed this way since 2012. > Do I still need to add the fixes tag? If so, does it have to be accurate? Hmm I thought this was related to some recent changes of RNR, we will need to trace back with git blame, note it important to have the fixes tag so we can add this change to stable kernels and if that is really since 2012 that have this bug it probably even more important since it might be applied to more stable versions. > > > > > Signed-off-by: Archie Pusaka <apusaka@chromium.org> > > > Reviewed-by: Ying Hsu <yinghsu@chromium.org> > > > > > > --- > > > The following steps are performed: > > > (1) Prepare 2 classic peer devices that needs RNR. Put device A > > > closer to DUT and device B (much) farther from DUT. > > > (2) Remove all cache and previous connection from DUT > > > (3) Put both peers into pairing mode, then start scanning on DUT > > > (4) After ~8 sec, turn off peer B. > > > *This is done so DUT can discover peer B (discovery time is 10s), > > > but it hasn't started RNR. Peer is turned off to buy us the max > > > time in the RNR phase (5s). > > > (5) Immediately as device A is shown on UI, click to connect. > > > *We thus know that the DUT is in the RNR phase and trying to > > > resolve the name of peer B when we initiate connection to peer A. > > > (6) Forget peer A. > > > (7) Restart scan and stop scan. > > > *Before the CL, stop scan is broken because we will try to cancel > > > a nonexistent RNR > > > (8) Restart scan again. Observe DUT can scan normally. > > > > > > > > > net/bluetooth/hci_event.c | 17 ++++++++++------- > > > 1 file changed, 10 insertions(+), 7 deletions(-) > > > > > > diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c > > > index 395c6479456f..95e145e278c9 100644 > > > --- a/net/bluetooth/hci_event.c > > > +++ b/net/bluetooth/hci_event.c > > > @@ -2453,6 +2453,16 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > > !test_and_set_bit(HCI_CONN_MGMT_CONNECTED, &conn->flags)) > > > mgmt_device_connected(hdev, conn, name, name_len); > > > > > > + e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > > > + > > > + if (e) { > > > + list_del(&e->list); > > > + > > > + e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > > > + mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > > > + name, name_len); > > > + } > > > + > > > if (discov->state == DISCOVERY_STOPPED) > > > return; > > > > > > @@ -2462,7 +2472,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > > if (discov->state != DISCOVERY_RESOLVING) > > > return; > > > > > > - e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); > > > /* If the device was not found in a list of found devices names of which > > > * are pending. there is no need to continue resolving a next name as it > > > * will be done upon receiving another Remote Name Request Complete > > > @@ -2470,12 +2479,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, > > > if (!e) > > > return; > > > > > > - list_del(&e->list); > > > - > > > - e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; > > > - mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, > > > - name, name_len); > > > - > > > if (hci_resolve_next_name(hdev)) > > > return; > > > > > > -- > > > 2.37.1.595.g718a3a8f04-goog > > > > > > > > > -- > > Luiz Augusto von Dentz > > Thanks, > Archie
diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c index 395c6479456f..95e145e278c9 100644 --- a/net/bluetooth/hci_event.c +++ b/net/bluetooth/hci_event.c @@ -2453,6 +2453,16 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, !test_and_set_bit(HCI_CONN_MGMT_CONNECTED, &conn->flags)) mgmt_device_connected(hdev, conn, name, name_len); + e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); + + if (e) { + list_del(&e->list); + + e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; + mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, + name, name_len); + } + if (discov->state == DISCOVERY_STOPPED) return; @@ -2462,7 +2472,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, if (discov->state != DISCOVERY_RESOLVING) return; - e = hci_inquiry_cache_lookup_resolve(hdev, bdaddr, NAME_PENDING); /* If the device was not found in a list of found devices names of which * are pending. there is no need to continue resolving a next name as it * will be done upon receiving another Remote Name Request Complete @@ -2470,12 +2479,6 @@ static void hci_check_pending_name(struct hci_dev *hdev, struct hci_conn *conn, if (!e) return; - list_del(&e->list); - - e->name_state = name ? NAME_KNOWN : NAME_NOT_KNOWN; - mgmt_remote_name(hdev, bdaddr, ACL_LINK, 0x00, e->data.rssi, - name, name_len); - if (hci_resolve_next_name(hdev)) return;