Fix for IPv6 Dual Stack support #12791
Conversation
|
@morser499 can you review this changes. |
|
@cy-arsm, thank you for your changes. |
Issue: The problem is that there is a race condition introduced in that the LWIP thread is relying on the interface as it is taken down by a application thread while calling disconnect. In disconnect api called from application context, whd_emac_wifi_link_state_changed() will refer to netif interface structure in its callback api netif_link_irq(netif). This netif will be cleared by remove_etherent_interface(). whd_emac_wifi_link_state_changed will post message to tcpip_thread. tcpip_thread will process the message and call the callback api netif_link_irq(netif) Calling sequence is whd_emac_wifi_link_state_changed -> remove_etherent_interface(). Hence there is a timing issue that netif might be cleared first before tcpip thread process the message netif_link_irq(netif) Fix: remove_etherent_interface() will post message to tcpip thread and tcpip thread process the message delete_interface() which will actually remove the inferface from the netif_list. Calling sequence is whd_emac_wifi_link_state_changed() message post -> remove_etherent_interface() message post. message processing order netif_link_irq(netif) -> delete_interface(). Since both the processing is handled in single thread, processing of message is handled sequentially.
|
Changes seem reasonable. Is there a GreenTea test report for the Cypress boards? I want to confirm that there are no issues in the various network related tests (tests-netsocket-,tests-network-,tests-integration-*). |
|
@morser499 , Yes Greentea test's for cypress boards are available and i ran GT locally for CY8CPROTO_062_4343W. |
|
@AnttiKauppila @0xc0170 - Can this be merged for 5.15.2 release. We need this fix for our upcoming software release. |
This test report does not seem to include networking tests that would be impacted by this change. Do you have reports for the following suites? |
|
CI started |
|
@cy-arsm @sathishm6 @AnttiKauppila for this to even be considered for a release off 5.15 there needs to be a corresponding PR directly against that branch.....It is way past the code freeze for 5.15.2 which is literally about to be made. Checking with @andypowers and @bulislaw . |
|
@cy-arsm @sathishm6 Having spoken to @andypowers I'm afraid this has missed 5.15.2. If you would like this considered for a future release from 5.15 please raise the PR as mentioned above. |
Test run: SUCCESSSummary: 6 of 6 test jobs passed |
0xc0170
added
needs: review
release-version: 6.0.0-beta-1
and removed
ready for merge
labels
Modified LwIP mbed-os wrapper LWIPInterface.cpp to add an api remove interface which will be called as part of WhdSoftAPInterface::stop() to support sending UDP IPv6 packet is sent over link-local interface
Issue: udp_sendto() Fails for multicast IPV6 packet when interface is switched from SoftAP to STA mode.
When SoftAP start is called, add_ethernet_interface() will be called internally to add interface details into netif_list.
When switching from SoftAP to STA mode, add_ethernet_interface() will be called again to append the interace details into netif_list.
When udp_sendto() is called, ip6_route() will return interface as NULL since it consider device as single interface.
Summary of changes
LWIPInterface.cpp
LWIP::remove_ethernet_interface() api is added to remove the interface from the netif_list.
WhdSoftAPInterface.cpp
WhdSoftAPInterface::stop(), calling remove_ethernet_interface() from the network list which was added as part of WhdSoftAPInterface::start()
WhdSTAInterface.cpp
WhdSTAInterface::stop(), calling remove_ethernet_interface() from the network list which was added as part of WhdSTAInterface::start()
Impact of changes
Migration actions required
Documentation
--------------------------------------------------------------------------------------------------------------### Pull request type
Test results
Reviewers