Here are a couple of fixes and an extra tracepoint for AF_RXRPC:
(1) Calculate the RTO pretty much as TCP does, rather than making
something up, including an initial 4s timeout (which causes return
probes from the fileserver to fail if a packet goes missing), and add
backoff.
(2) Fix the discarding of out-of-order received ACKs. We mustn't let the
hard-ACK point regress, nor do we want to do unnecessary
retransmission because the soft-ACK list regresses. This is not
trivial, however, due to some loose wording in various old protocol
specs, the ACK field that should be used for this sometimes has the
wrong information in it.
(3) Add a tracepoint to log a discarded ACK.
[V2] Fixed a "Fixes" line in a commit message.
The patches are tagged here:
git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git
rxrpc-fixes-20200520
and can also be found on the following branch:
http://git.kernel.org/cgit/linux/kernel/git/dhowells/linux-fs.git/log/?h=rxrpc-fixes
David
---
David Howells (1):
rxrpc: Fix ack discard
fs/afs/fs_probe.c | 18 ++--
fs/afs/vl_probe.c | 18 ++--
include/net/af_rxrpc.h | 2 +-
include/trace/events/rxrpc.h | 52 +++++++++---
net/rxrpc/Makefile | 1 +
net/rxrpc/ar-internal.h | 25 ++++--
net/rxrpc/call_accept.c | 2 +-
net/rxrpc/call_event.c | 22 ++---
net/rxrpc/input.c | 44 ++++++++--
net/rxrpc/misc.c | 5 --
net/rxrpc/output.c | 9 +-
net/rxrpc/peer_event.c | 46 ----------
net/rxrpc/peer_object.c | 12 +--
net/rxrpc/proc.c | 8 +-
net/rxrpc/rtt.c | 195 +++++++++++++++++++++++++++++++++++++++++++
net/rxrpc/sendmsg.c | 26 ++----
net/rxrpc/sysctl.c | 9 --
17 files changed, 335 insertions(+), 159 deletions(-)
create mode 100644 net/rxrpc/rtt.c
