vSomeIP 3.5.11 Part 2 Release #976
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Wait for unacknowledged data on all TCP endpoints Before calling close() add a wait for all data to be sent and acked. Setting the linger option, on tcp sockets, will not change the TIME_WAIT value. The linger will block the close call, for a maximum of the defined time. This means, if we set linger to 5, the close() call can block up to 5 seconds. Afterwards, the socket can still be in a TIME_WAIT state for 60~120 seconds. The only way to fully remove the TIME_WAIT state is to set the linger to 0, which forces a RST to be sent. The issue with this approach is that when calling the close() the RST will be sent and the buffered data, that is sill to be sent, will be dropped. In order to avoid this, before calling the close(), vsomeip-lib needs to wait until all data is sent.
SOME/IP messages have a 16-byte header, but a significant amount of the codebase (including rmi::on_message) uses a 8-byte definition This is a purposefully small, minimal fix
fcmonteiro
approved these changes
Nov 12, 2025
vvcarvalho-csw
force-pushed
the
master
branch
from
7b8d7e7 to
9335c70
Compare
Returns when receiving messages on invalid port vsomeip sends a StopSubscribe to a service, but still receives a external notification for that service ~15ms after. This leads to the InstanceID be 0xFFFF, and we log the error message Received message on invalid port: [beef.FFFF.0001.0000.0000] from: 192.168.2.57:15000
Modify the multicast group test to detect issues related to UDP packets dispatching. SO_REUSEPORT was added to workaround a TCP issue related to the restart. During the review process, it was added too to UDP but it introduced a regression: UDP packets were no longer being received by the correct socket. Indeed, there are no written rules defining how UDP packets should be distributed by Linux when multiple sockets are bound to the same endpoint. VSOMEIP can create up to three sockets on the same endpoint. It was thought that when the SO_REUSEPORT option was enabled, Linux routed all traffic to the one bound to INADDR_ANY. That is not correct. Therefore, the fix that redirected messages received by the socket bound to INADDR_ANY has been removed from this PR, to avoid any regression. It remains a check that prints a warning if this socket receives a packet addressed to another socket. Therefore, limited to extending the multicast group test in order to detect the problem of unexpected dispatching of UDP packets.
Update the error handling for local communication to synchronize IO operations errors and state. In some scenarios, a race condition between local client endpoints read and write callbacks error handling could force an application to never fully close the socket and be stuck on a re-connection loop. This PR updates the error handling of both IO operations for local tcp and uds communication. The callbacks for a single connection are invoked sequentially and only the first one is processed, moreover, the re-connect decision is forwarded for the routing manager instead of forcing low-level retries.
vvcarvalho-csw
force-pushed
the
master
branch
from
9335c70 to
c4a65bd
Compare
Prevent race condition on insert_subscription and improve logging Consider the following case: Server offers a service, before registering all the events Two applications subscribe said service at the same time, including the unregistered events In this scenario there is a race condition on insert_subscription, where two placeholders are created for the same subscription routing_manager_base::insert_subscription(a524): [cafe.0001.0001.0001] received subscription for unknown (unrequested / unoffered) event. Creating placeholder event holding subscription until event is requested/offered. routing_manager_base::insert_subscription(a52d): [cafe.0001.0001.0001] received subscription for unknown (unrequested / unoffered) event. Creating placeholder event holding subscription until event is requested/offered. However the second placeholder created ends up not being successful and the subscription is not inserted into the eventgroup. This PR fixes this race by protecting the call to create_placeholder_event_and_subscribe Also improve logging
It is possible that an application sees several clients with the same routing info/guest address. It is a transient state, that happens when an application restarts, acquires a new client-id, but binds to the same ip/port There is a long history to the problem: prior to 062d843, it can happen, but it is temporary, as applications would drop client after receiving RIE_DELETE_CLIENT (that of course has other issues...) after 062d843, it can happen, but is no longer temporary 9e5fff5 tries to tackle it by dropping routing, but using the same client connection 176aaef partially reverts 9e5fff5 This commit resolves the issue by dropping old clients, as soon as a new client with the same routing info/guest address is received
fake_tcp_socket_handle::is_connected can extend the lifetime of the underlying socket objects, end up being the one triggering destruction It deadlocks, because destruction invokes socket_manager, and this function is itself invoked by socket_manager. Fix it with a trick
Only close server endpoints at end of suspend process Server endpoints were being closed at the beginning of the suspend process, preventing necessary messages from being sent during suspension and causing them to remain pending until resume starts. One affected case is Stop Offer messages, which were verified as not being sent during suspend, explaining the high number of sdi::check_stopped_services_on_suspend stop offer not sent for service logs in 3.5.11. This change implements the correct flow for the suspend process: close connections only after all cleanup operations are complete.
Update vsomeip-lib to v3.5.11 Only close server endpoints at end of suspend process tests: fix fake_socket_tests deadlock routing: drop old clients with same routing info prevent race on insert_subscription Update error handling for local endpoints extend multigroup test to detect issue with UDP dispatching Return on received message on invalid port/instance rmi: fix buffer overflow Wait for unacknowledged data before closing Client should start Wait Phase when offer TTL expires fix/doc offered_services_info application start-stop-start broken with local TCP Fix reuse address issues in tests manage suspend event log rmc state transitions as a string doc and fix security_test Test cached event feature rm: improve subscription logging Use universal time (availability_handler_test) add logs when unsetting payload misc: add predicate to condition variables Add boardnet initial event test Fix subscribe notify test diff clients Fix restart routing test cei, ltcei, rmc: Trivial changes on logging readability Check ACL and process messages if the address is specified Adds all sanitizers and valgrinds to the check job Fix subscribe notify test one tests emi: remove deferrence of multicast operations remove sessionID from clients_ map cei: trivial changes, removes duplicated code Update cache on change Add basic watchdog for tracking stop offers routing: remove dead code endpoints: set SO_LINGER to 0, force TCP RSTs
vvcarvalho-csw
force-pushed
the
master
branch
from
c4a65bd to
a5e4a50
Compare
duartenfonseca
approved these changes
Nov 14, 2025
anaritarodrigues
approved these changes
Nov 14, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.