| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This fixes the RTO doubling on timeout according to Karn/Partridge.
Exponentially increasing RTO when it times out (e.g. doubling)
fixes the problem that a sudden increase in real RTT starves the sRTT
updates by never getting out of backoff as retransmitted packets can't
update RTT.
Added an parameter to make it less aggressive, default is doubling.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
There was an unused struct timerwheel * lingering in the application
instance.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
Growing pains.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
| |
The delayed ACK was wrongly measuring the delay against the receiver
activity instead of the sender activity. Also fixed receiver activity
not being updated for non-data packets (and duplicates and other
dropped traffic).
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds the option to use the Round-Trip-Time (RTT) estimation
algorithm as it is implemented in the TCP implementation in Linux. It
looks like it outperforms the TCP default algorithm, so I enabled this
one by default. Also adds the option to change the RTO timeout
calculation to include more (or less) than 4 times the mdev (specified
as a power of 2. Left the default value to 2 (so, 4 mdevs), but 3 (8
mdevs) gives better results in my tests.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
If a flow was deallocated while there were still unprocessed events in
an fqueue, it would cause a SEGV in fqueue_next because it was not
checking the validity of the returned flow descriptor.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
| |
Delayed ACKs are now sent after twice the internal tick time. Fixes
initial ACK record (rcv_cr.seqno) being uninitialized (0) when the
first ACK was to be sent. Adds some FRCT metrics for number of
received delayed (bare) ACKs and the RTT estimator.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
The fqueues were relying on the fact that the portevent were two
integers. This cleans that up a bit.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The protobuf message was free'd before usage in flow_init.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
If not passed a value for the last parameter, oping would SEGV.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Unidirectional traffic has one of the peers only send bare FRCT
packets. These never set a DRF, since they have no sequence number.
At the receiver, all these ACKs and window updates were always dropped
as the receiver connection record was timed out.
Also fixes a SEGV if flow control kicks in (passing NULL timeout to
pthread_cond_timedwait).
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The application will now handle incoming FRCT packets even if the
application never reads data from the flow (for instance servers). To
do this, it reserves an fset_t (id 0). When an FRCT-enabled flow is
created, it is automatically added to this fset. An rx thread will
listen for incoming events and perform necessary actions on the flow
if needed. If the FRCT flow is added to another user fset, it will be
handled by that user fset (and if the flow is removed from a user
fset, it will be re-added to the set with id 0 to be handled by the
rx_flow thread. The flow monitoring is handled by the same thread,
replacing the previous monitoring thread.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
| |
Now the instance keeps all flows for an application in a linked list
to easily iterate over all allocated flows, which is needed by the
keepalive monitoring. This is more efficient that tracking min and max
fd.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
We don't need to iterate fsets anymore since the removal of fset_keepalive.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The frcti_filter was reading raw data from the buffers, causing the
frcti_rcv to operate directly on encrypted packets. It decrypt and
filter for invalid packets. I moved the function from frct to the
fqueue implementation and renamed it fqueue_filter as it filters
fqueues. Should be extended to filter out keepalives on non-FRCT
flows, as these will now still cause spurious wakeups.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
This adds a monitoring thread to handle flow keepalive management in
the application and removes the thread interruptions to schedule FRCT
calls within the regular IPC calls.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reading/writing to (N + 1)-flows from the IPCP was using a raw QoS flow
to bypass some functions in the ipcp_flow_read call. But this call was
broken for keepalive packets. Fixing the ipcp_flow_read call for
(N - 1) flows causes the IPCPs to drop 0-byte keepalive packets coming from
(N + 1) client flows.
>From now on, there is a dedicated call for (N + 1) reads/writes from
the IPCPs that's more efficient and cleaner. The (N + 1) flow internal
QoS is now also defaulted to a qos_np1 qosspec, instead of tampering
with the qosspec requested by the (N + 1) client.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
This allows setting the FLOWPEER state on a flow to signal a peer is
unresponsive.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the first step moving away from scheduling the FRCT and flow
monitoring functions as part of the IPC calls (flow_read / flow_write
/ fevent) and towards the more scalable (and far less complicated)
implementation to take care of these functions in separate threads.
If a process creates the first flow that requires FRCT, it will spin
up a thread to process events on the timerwheel (retransmissions and
delayed ACKs). This single thread lives until the last flow with FRCT
is deallocated.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
The creation of FRCT instances (if needed) is now part of flow_init()
call instead of an addition after the flow is initialized.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
Writing valid packets to the rbuff (add crc check, encrypt) is now
extracted into a function.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
Prog name is not used anymore, probably a remnant from the early days,
when we were passing rina_name_t tuples all over the place.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reading packets from the rbuff and checking their validity (non-zero
size, pass crc check, pass decryption) is now extracted into a
function.
Also adds a function to get the length of an sdu_du_buff instead of
subtracting the tail and head pointers.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
The fset add function was notifying for each packet already stored in
the rx rbuff, which isn't needed.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
The oping server will not print receiving packets when the --quiet
(-Q) flag is passed, like the client.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It doesn't really make sense to manually and one-sidedly configure the
timeout of delayed acknowledgements, as setting it too high upsets the
peer's sRTT estimates. Even worse, it also causes a lot of spurious
retransmissions if it exceeds the sRTT mean deviation calculated by
the receiver. Compensating on bare acknowledgment for the ack delay
could improve the RTT estimate deviation, but not the spurious
retransmissions if it was set too high. This sets the delayed ack to
wait for a single RTT mean deviation. Probably needs more tweaking to
further reduce differences between the RTT estimates at the sender and
receiver, e.g. compensate the RTT estimate for delayed acks, or
increase the RTO to add 8 mdevs to sRTT instead of 4. However, it
looks like the mdev estimate is the trickiest one to get to sync, not
the RTT average. Linux reduces the sample weight for mdev from 1/4 to
1/32 in some cases, will give that a shot some day too to see if that
further align sRTT estimates. In any case, this patch already improves
things a lot.
Also fixes a bug where the sender was sending acknowlegments on the
first packets in flight for the 0 sequence number. The receiver
activity was measured in seconds but compared to a timeout value in
nanoseconds.
There's still a lot of spurious retransmissions that start after
actual packet loss occurs, I'm still investigating what causes it.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
| |
This exposes some additional metrics relating to FRCT / Flow control:
the number of duplicate packets received, number of packets received
out of the flow control window and / or reordering queue, and the
number of rendez-vous messages sent.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There still were a couple of bugs in the timerwheel. If the future
schedule was coinciding with the slot currently being processed
(i.e. exactly RXMQ_SLOTS in the future), the list_add_tail caused an
infinite loop. Another bug was causing the slots at higher levels to
be processed too soon.
Retransmissions should now schedule correctly.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The timerwheel was retransmitting packets and the error check for
negative values of the rbuff allocation was instead checking for
non-zero values, causing a buffer allocation to succeed but the
program to continue down the unhappy path leaving that packet stuck in
the buffer unattended.
Also fixes wrongly scheduled retransmissions that cause packet storms.
FRCP is much more stable now. Still needs some work for high
bandwidth-delay products (fast-retransmit).
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
If there is no piggyback data, memcpy was passed a NULL pointer in
memcpy(buf, NULL, 0) calls, which is undefined behaviour.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
A small refactor of the kad_req_create function's cleanup code.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The maximum packet lifetime (MPL) is a property of the flow that needs
to be passed to the reliable transmission protocol (FRCP) for its
correct operation. Previously, the value of MPL was set fixed as one
of the (fixed) Delta-t parameters. This patch makes the MPL a property
of the layer, and it can now be set per layer-type at build time.
This is a step towards a proper MPL estimator in the flow allocator.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
The parameters were set directly from the build configs. A first step
to making FRCP configurable at runtime, is to pass the parameters to
the frcti_create() function.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The notorious off-by-one hit again.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
If the keepalive would underflow if set to 1-3 ms.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On exit of the IRMd all flows will now be flagged as down, so external
applications will not hang anymore. Note: reads keep work from flows
that are down until there are no more remaining packets in the buffer,
but no more packets can be written.
When the RIB is used, the external application may exit a bit later
than the IRMd, so I added a brief sleep before the IRMd tries to
remove the fuse main directory.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
There was a lock reversal in the timerwheel. There still is a thorough
revision needed of the locking in dev.c after the FRCP logic is
completed and tuned.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We cancel the thread, so the SO_RCVTIMEO is not needed anymore (it
dated from when we checked the state every so often.
The address sanitizer is complaining about the the cleanup handlers in
the acceptloops after the thread gets cancelled in the read(). I've
tried to resolve it, but no avail. Pretty convinced it's a
false-positive, so ASan will ignore these functions for now.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
| |
IPCPs would call rib_fini() twice, once after cleaning up their
managed RIB, and once again for the program-generic RIB, which is not
initialized for IPCPs. rib_fini() checked if the mount name was valid,
but it didn't unset it after execution.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The rib_init return value wasn't checked.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
Bare FRCP messages (ACKs without data, Rendez-vous packets) were not
encrypted on encrypted flows, causing the receiver to fail decryption.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
The qosspec_t now has a timeout value that sets the timeout value of
the flow. Flows with a peer that has timed out will now return
-EFLOWPEER on flow_read() or flow_write().
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The ocbr tool was returning 0 on error.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The irm_list_ipcps function can return negative values.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
The checked condition can't happen.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
| |
The ret_msg variable can leak in the main loop of the irmd in this
failure path.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
| |
Fixes some unchecked and wrongly checked return values.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds flow liveness monitoring for flows, with a fixed timeout of
120s. I will make it configurable at flow allocation later on (timeout
needs to be communicated to the peer). If one peer dies, or doesn't
call any IPC calls (flow_write/flow_read/fevent) it will stop sending
keepalives and the other peer's read/writes will error on an
-EFLOWDOWN after the timeout expires.
Packets without a payload (0 length packets) are interpreted as
keepalive packets for the flow. They can be sent from any application,
but they will not trigger a message read at the receiver side (0 as a
return value on flow_read indicates a previous partial read has
completed at exactly the buffer size).
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
The flow_set will now keep a list of the flows in the set, this makes
it more efficient to iterate over the flows. Extending the public API
for fset_t with an iterator will also be useful.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|
|
|
|
|
|
|
|
|
| |
When handling management frames, there was a cancellation point after
the unlock, which would cause the cleanup handler to attempt a double
unlock if the thread was cancelled at that point.
Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>
|