| Message ID | 20201120054036.15199-3-ms@dev.tdt.de |
|---|---|
| State | Superseded |
| Headers | show |
| Series | net/x25: netdev event handling | expand |
Should we also handle the NETDEV_UP event here? In previous versions of this patch series you seemed to want to establish the L2 connection on device-up. But in this patch, you didn't handle NETDEV_UP. Maybe on device-up, we need to check if the carrier is up, and if it is, we do the same thing as we do on carrier-up.
On Fri, Nov 20, 2020 at 3:11 PM Xie He <xie.he.0141@gmail.com> wrote: > > Should we also handle the NETDEV_UP event here? In previous versions > of this patch series you seemed to want to establish the L2 connection > on device-up. But in this patch, you didn't handle NETDEV_UP. > > Maybe on device-up, we need to check if the carrier is up, and if it > is, we do the same thing as we do on carrier-up. Are the device up/down status and the carrier up/down status independent of each other? If they are, on device-up or carrier-up, we only need to try establishing the L2 connection if we see both are up. On NETDEV_GOING_DOWN, we can also check the carrier status first and if it is down, we don't need to call lapb_disconnect_request.
On 2020-11-21 00:50, Xie He wrote: > On Fri, Nov 20, 2020 at 3:11 PM Xie He <xie.he.0141@gmail.com> wrote: >> >> Should we also handle the NETDEV_UP event here? In previous versions >> of this patch series you seemed to want to establish the L2 connection >> on device-up. But in this patch, you didn't handle NETDEV_UP. >> >> Maybe on device-up, we need to check if the carrier is up, and if it >> is, we do the same thing as we do on carrier-up. > > Are the device up/down status and the carrier up/down status > independent of each other? If they are, on device-up or carrier-up, we > only need to try establishing the L2 connection if we see both are up. No, they aren't independent. The carrier can only be up if the device / interface is UP. And as far as I can see a NETDEV_CHANGE event will also only be generated on interfaces that are UP. So you can be sure, that if there is a NETDEV_CHANGE event then the device is UP. I removed the NETDEV_UP handling because I don't think it makes sense to implicitly try to establish layer2 (LAPB) if there is no carrier. And with the first X.25 connection request on that interface, it will be established anyway by x25_transmit_link(). I've tested it here with an HDLC WAN Adapter and it works as expected. These are also the ideal conditions for the already mentioned "on demand" scenario. The only necessary change would be to call x25_terminate_link() on an interface after clearing the last X.25 session. > On NETDEV_GOING_DOWN, we can also check the carrier status first and > if it is down, we don't need to call lapb_disconnect_request. This is not necessary because lapb_disconnect_request() checks the current state. And if the carrier is DOWN then the state should also be LAPB_STATE_0 and so lapb_disconnect_request() does nothing.
On Sun, Nov 22, 2020 at 10:55 PM Martin Schiller <ms@dev.tdt.de> wrote: > > No, they aren't independent. The carrier can only be up if the device / > interface is UP. And as far as I can see a NETDEV_CHANGE event will also > only be generated on interfaces that are UP. > > So you can be sure, that if there is a NETDEV_CHANGE event then the > device is UP. OK. Thanks for your explanation! > I removed the NETDEV_UP handling because I don't think it makes sense > to implicitly try to establish layer2 (LAPB) if there is no carrier. As I understand, when the device goes up, the carrier can be either down or up. Right? If this is true, when a device goes up and the carrier then goes up after that, L2 will automatically connect, but if a device goes up and the carrier is already up, L2 will not automatically connect. I think it might be better to eliminate this difference in handling. It might be better to make it automatically connect in both situations, or in neither situations. If you want to go with the second way (auto connect in neither situations), the next (3rd) patch of this series might be also not needed. I just want to make the behavior of LAPB more consistent. I think we should either make LAPB auto-connect in all situations, or make LAPB wait for L3's instruction to connect in all situations. > And with the first X.25 connection request on that interface, it will > be established anyway by x25_transmit_link(). > > I've tested it here with an HDLC WAN Adapter and it works as expected. > > These are also the ideal conditions for the already mentioned "on > demand" scenario. The only necessary change would be to call > x25_terminate_link() on an interface after clearing the last X.25 > session. > > > On NETDEV_GOING_DOWN, we can also check the carrier status first and > > if it is down, we don't need to call lapb_disconnect_request. > > This is not necessary because lapb_disconnect_request() checks the > current state. And if the carrier is DOWN then the state should also be > LAPB_STATE_0 and so lapb_disconnect_request() does nothing. Yes, I understand. I just thought adding this check might make the code cleaner. But you are right.
On 2020-11-23 09:31, Xie He wrote: > On Sun, Nov 22, 2020 at 10:55 PM Martin Schiller <ms@dev.tdt.de> wrote: >> >> No, they aren't independent. The carrier can only be up if the device >> / >> interface is UP. And as far as I can see a NETDEV_CHANGE event will >> also >> only be generated on interfaces that are UP. >> >> So you can be sure, that if there is a NETDEV_CHANGE event then the >> device is UP. > > OK. Thanks for your explanation! > >> I removed the NETDEV_UP handling because I don't think it makes sense >> to implicitly try to establish layer2 (LAPB) if there is no carrier. > > As I understand, when the device goes up, the carrier can be either > down or up. Right? > > If this is true, when a device goes up and the carrier then goes up > after that, L2 will automatically connect, but if a device goes up and > the carrier is already up, L2 will not automatically connect. I think > it might be better to eliminate this difference in handling. It might > be better to make it automatically connect in both situations, or in > neither situations. AFAIK the carrier can't be up before the device is up. Therefore, there will be a NETDEV_CHANGE event after the NETDEV_UP event. This is what I can see in my tests (with the HDLC interface). Is the behaviour different for e.g. lapbether? > > If you want to go with the second way (auto connect in neither > situations), the next (3rd) patch of this series might be also not > needed. > > I just want to make the behavior of LAPB more consistent. I think we > should either make LAPB auto-connect in all situations, or make LAPB > wait for L3's instruction to connect in all situations. > >> And with the first X.25 connection request on that interface, it will >> be established anyway by x25_transmit_link(). >> >> I've tested it here with an HDLC WAN Adapter and it works as expected. >> >> These are also the ideal conditions for the already mentioned "on >> demand" scenario. The only necessary change would be to call >> x25_terminate_link() on an interface after clearing the last X.25 >> session. >> >> > On NETDEV_GOING_DOWN, we can also check the carrier status first and >> > if it is down, we don't need to call lapb_disconnect_request. >> >> This is not necessary because lapb_disconnect_request() checks the >> current state. And if the carrier is DOWN then the state should also >> be >> LAPB_STATE_0 and so lapb_disconnect_request() does nothing. > > Yes, I understand. I just thought adding this check might make the > code cleaner. But you are right.
On Mon, Nov 23, 2020 at 1:00 AM Martin Schiller <ms@dev.tdt.de> wrote: > > AFAIK the carrier can't be up before the device is up. Therefore, there > will be a NETDEV_CHANGE event after the NETDEV_UP event. > > This is what I can see in my tests (with the HDLC interface). > > Is the behaviour different for e.g. lapbether? Some drivers don't support carrier status and will never change it. Their carrier status will always be UP. There will not be a NETDEV_CHANGE event. lapbether doesn't change carrier status. I also have my own virtual HDLC WAN driver (for testing) which also doesn't change carrier status. I just tested with lapbether. When I bring up the interface, there will only be NETDEV_PRE_UP and then NETDEV_UP. There will not be NETDEV_CHANGE. The carrier status is alway UP. I haven't tested whether a device can receive NETDEV_CHANGE when it is down. It's possible for a device driver to call netif_carrier_on when the interface is down. Do you know what will happen if a device driver calls netif_carrier_on when the interface is down?
On Mon, Nov 23, 2020 at 1:36 AM Xie He <xie.he.0141@gmail.com> wrote: > > Some drivers don't support carrier status and will never change it. > Their carrier status will always be UP. There will not be a > NETDEV_CHANGE event. > > lapbether doesn't change carrier status. I also have my own virtual > HDLC WAN driver (for testing) which also doesn't change carrier > status. > > I just tested with lapbether. When I bring up the interface, there > will only be NETDEV_PRE_UP and then NETDEV_UP. There will not be > NETDEV_CHANGE. The carrier status is alway UP. > > I haven't tested whether a device can receive NETDEV_CHANGE when it is > down. It's possible for a device driver to call netif_carrier_on when > the interface is down. Do you know what will happen if a device driver > calls netif_carrier_on when the interface is down? I just did a test on lapbether and saw there would be no NETDEV_CHANGE event when the netif is down, even if netif_carrier_on/off is called. So we can rest assured of this part.
On 2020-11-23 11:08, Xie He wrote: > On Mon, Nov 23, 2020 at 1:36 AM Xie He <xie.he.0141@gmail.com> wrote: >> >> Some drivers don't support carrier status and will never change it. >> Their carrier status will always be UP. There will not be a >> NETDEV_CHANGE event. Well, one could argue that we would have to repair these drivers, but I don't think that will get us anywhere. From this point of view it will be the best to handle the NETDEV_UP in the lapb event handler and establish the link analog to the NETDEV_CHANGE event if the carrier is UP. >> >> lapbether doesn't change carrier status. I also have my own virtual >> HDLC WAN driver (for testing) which also doesn't change carrier >> status. >> >> I just tested with lapbether. When I bring up the interface, there >> will only be NETDEV_PRE_UP and then NETDEV_UP. There will not be >> NETDEV_CHANGE. The carrier status is alway UP. >> >> I haven't tested whether a device can receive NETDEV_CHANGE when it is >> down. It's possible for a device driver to call netif_carrier_on when >> the interface is down. Do you know what will happen if a device driver >> calls netif_carrier_on when the interface is down? > > I just did a test on lapbether and saw there would be no NETDEV_CHANGE > event when the netif is down, even if netif_carrier_on/off is called. > So we can rest assured of this part.
On Mon, Nov 23, 2020 at 2:38 AM Martin Schiller <ms@dev.tdt.de> wrote: > > Well, one could argue that we would have to repair these drivers, but I > don't think that will get us anywhere. Yeah... One problem I see with the Linux project is the lack of docs/specs. Often we don't know what is right and what is wrong. > From this point of view it will be the best to handle the NETDEV_UP in > the lapb event handler and establish the link analog to the > NETDEV_CHANGE event if the carrier is UP. Thanks! This way we can make sure LAPB would automatically connect in all situations. Since we'll have a netif_carrier_ok check in NETDEV_UP handing, it might make the code look prettier to also have a netif_carrier_ok check in NETDEV_GOING_DOWN handing (for symmetry). Just a suggestion. You can do whatever looks good to you :) Thanks!
On Mon, 23 Nov 2020 03:17:54 -0800 Xie He wrote: > On Mon, Nov 23, 2020 at 2:38 AM Martin Schiller <ms@dev.tdt.de> wrote: > > Well, one could argue that we would have to repair these drivers, but I > > don't think that will get us anywhere. > > Yeah... One problem I see with the Linux project is the lack of > docs/specs. Often we don't know what is right and what is wrong. More of a historic thing than a requirement AFAIK. Some software devices, e.g. loopback will not generate carrier events. But in this case looks like all the devices Martin wants to handle are lapb. > > From this point of view it will be the best to handle the NETDEV_UP in > > the lapb event handler and establish the link analog to the > > NETDEV_CHANGE event if the carrier is UP. > > Thanks! This way we can make sure LAPB would automatically connect in > all situations. > > Since we'll have a netif_carrier_ok check in NETDEV_UP handing, it > might make the code look prettier to also have a netif_carrier_ok > check in NETDEV_GOING_DOWN handing (for symmetry). Just a suggestion. > You can do whatever looks good to you :) Xie other than this the patches look good to you? Martin should I expect a respin to follow Xie's suggestion or should I apply v4?
On Mon, Nov 23, 2020 at 11:36 AM Jakub Kicinski <kuba@kernel.org> wrote: > > > > From this point of view it will be the best to handle the NETDEV_UP in > > > the lapb event handler and establish the link analog to the > > > NETDEV_CHANGE event if the carrier is UP. > > > > Thanks! This way we can make sure LAPB would automatically connect in > > all situations. > > > > Since we'll have a netif_carrier_ok check in NETDEV_UP handing, it > > might make the code look prettier to also have a netif_carrier_ok > > check in NETDEV_GOING_DOWN handing (for symmetry). Just a suggestion. > > You can do whatever looks good to you :) > > Xie other than this the patches look good to you? > > Martin should I expect a respin to follow Xie's suggestion > or should I apply v4? There should be a respin because we need to handle the NETDEV_UP event. The lapbether driver (and possibly some HDLC WAN drivers) doesn't generate carrier events so we need to do auto-connect in the NETDEV_UP event.
On 2020-11-23 23:09, Xie He wrote: > On Mon, Nov 23, 2020 at 11:36 AM Jakub Kicinski <kuba@kernel.org> > wrote: >> >> > > From this point of view it will be the best to handle the NETDEV_UP in >> > > the lapb event handler and establish the link analog to the >> > > NETDEV_CHANGE event if the carrier is UP. >> > >> > Thanks! This way we can make sure LAPB would automatically connect in >> > all situations. >> > >> > Since we'll have a netif_carrier_ok check in NETDEV_UP handing, it >> > might make the code look prettier to also have a netif_carrier_ok >> > check in NETDEV_GOING_DOWN handing (for symmetry). Just a suggestion. >> > You can do whatever looks good to you :) >> >> Xie other than this the patches look good to you? >> >> Martin should I expect a respin to follow Xie's suggestion >> or should I apply v4? > > There should be a respin because we need to handle the NETDEV_UP > event. The lapbether driver (and possibly some HDLC WAN drivers) > doesn't generate carrier events so we need to do auto-connect in the > NETDEV_UP event. I'll send a v5 with the appropriate change.
diff --git a/net/lapb/lapb_iface.c b/net/lapb/lapb_iface.c index 3c03f6512c5f..52d59984fbe6 100644 --- a/net/lapb/lapb_iface.c +++ b/net/lapb/lapb_iface.c @@ -418,14 +418,86 @@ int lapb_data_transmit(struct lapb_cb *lapb, struct sk_buff *skb) return used; } +/* Handle device status changes. */ +static int lapb_device_event(struct notifier_block *this, unsigned long event, + void *ptr) +{ + struct net_device *dev = netdev_notifier_info_to_dev(ptr); + struct lapb_cb *lapb; + + if (!net_eq(dev_net(dev), &init_net)) + return NOTIFY_DONE; + + if (dev->type == ARPHRD_X25) { + switch (event) { + case NETDEV_GOING_DOWN: + lapb_disconnect_request(dev); + break; + case NETDEV_DOWN: + lapb = lapb_devtostruct(dev); + if (!lapb) + break; + + lapb_dbg(0, "(%p) Interface down: %s\n", dev, + dev->name); + lapb_dbg(0, "(%p) S%d -> S0\n", dev, + lapb->state); + lapb_clear_queues(lapb); + lapb->state = LAPB_STATE_0; + lapb->n2count = 0; + lapb_stop_t1timer(lapb); + lapb_stop_t2timer(lapb); + break; + case NETDEV_CHANGE: + lapb = lapb_devtostruct(dev); + if (!lapb) + break; + + if (!netif_carrier_ok(dev)) { + lapb_dbg(0, "(%p) Carrier lost: %s\n", dev, + dev->name); + lapb_dbg(0, "(%p) S%d -> S0\n", dev, + lapb->state); + lapb_clear_queues(lapb); + lapb->state = LAPB_STATE_0; + lapb->n2count = 0; + lapb_stop_t1timer(lapb); + lapb_stop_t2timer(lapb); + } else { + lapb_dbg(0, "(%p): Carrier detected: %s\n", + dev, dev->name); + if (lapb->mode & LAPB_DCE) { + lapb_start_t1timer(lapb); + } else { + if (lapb->state == LAPB_STATE_0) { + lapb->state = LAPB_STATE_1; + lapb_establish_data_link(lapb); + } + } + } + break; + } + } + + return NOTIFY_DONE; +} + +static struct notifier_block lapb_dev_notifier = { + .notifier_call = lapb_device_event, +}; + static int __init lapb_init(void) { + register_netdevice_notifier(&lapb_dev_notifier); + return 0; } static void __exit lapb_exit(void) { WARN_ON(!list_empty(&lapb_list)); + + unregister_netdevice_notifier(&lapb_dev_notifier); } MODULE_AUTHOR("Jonathan Naylor <g4klx@g4klx.demon.co.uk>");
This patch allows layer2 (LAPB) to react to netdev events itself and avoids the detour via layer3 (X.25). 1. Call lapb_disconnect_request() on NETDEV_GOING_DOWN events. 2. When a NETDEV_DOWN event occur, clear all queues, enter state LAPB_STATE_0 and stop all timers. 3. The NETDEV_CHANGE event makes it possible to handle carrier loss and detection. In case of Carrier Loss, clear all queues, enter state LAPB_STATE_0 and stop all timers. In case of Carrier Detection, we start timer t1 on a DCE interface, and on a DTE interface we change to state LAPB_STATE_1 and start sending SABM(E). Signed-off-by: Martin Schiller <ms@dev.tdt.de> --- net/lapb/lapb_iface.c | 72 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 72 insertions(+)