Commit Graph

973 Commits

Author SHA1 Message Date
Rachel Chen
fd2c345152 sendQueue: ignore "datagram too large" error (#3328)
This commit introduces additional platform-dependent checking when the
kernel returns an error. Previously, the session is terminated when
PingFrame sends a discovery packet larger than the limit. With this
commit, the error is checked, and if it is "datagram too large", the
error is ignored.

Additionally,
1. This commit re-enables MTU discovery on Windows unless it
is disabled explicitly by user (Undo #3276),
2. Set IP_DONTFRAGMENT and IPV6_DONTFRAG with error checking on Windows,
   and
3. Set IP_MTU_DISCOVERY to PMTUDISC_DO for both IPv4 and IPv6 on Linux
   so that the kernel will return "message too long".

Fixes #3273 #3327
2022-02-20 00:21:32 -08:00
Marten Seemann
cb6db58fc1 use a uint64 in the AllowConnectionWindowIncrease callback 2022-01-18 09:23:56 +04:00
Marten Seemann
bfd685faf2 add a AllowConnectionWindowIncrease config option 2022-01-14 12:04:04 +04:00
Marten Seemann
f9904c7c45 add a callback to block window increases to the connection flow controller 2022-01-14 12:03:30 +04:00
Marten Seemann
e6572e3d0a Merge pull request #3276 from lucas-clemente/windows-disable-pmtud
disable Path MTU Discovery on Windows
2021-10-13 08:55:42 +01:00
Marten Seemann
d410b590e9 disable Path MTU Discovery on Windows 2021-09-17 18:24:26 +02:00
Marten Seemann
a54b00e770 enter the regular run loop if no undecryptable packet was processed 2021-09-08 15:35:53 +01:00
Marten Seemann
afaaaf97d0 Merge pull request #3215 from lucas-clemente/fix-errors-is
fix incorrect usage of errors.Is
2021-06-28 17:53:14 -07:00
Marten Seemann
dbb517858e fix incorrect usage of errors.Is
errors.Is is supposed to used for equality of errors, not for type
assertions. That's what errors.As is there for.
2021-06-28 15:29:52 -07:00
Mathis Engelbart
75d2276430 Return error on SendMessage when session is closed 2021-06-28 11:15:37 +02:00
Marten Seemann
1613809ba2 use the new error types to log the reason why a connection is closed 2021-05-03 14:00:41 +07:00
Marten Seemann
93cfef57ca expose a StatelessResetError 2021-05-01 09:39:51 +07:00
Marten Seemann
1ce572228b expose a VersionNegoationError 2021-05-01 09:39:24 +07:00
Marten Seemann
42b61729bd expose the TransportError and the ApplicationError 2021-05-01 09:38:49 +07:00
Marten Seemann
f5238bf7b1 move the ApplicationErrorCdoe to the qerr package 2021-05-01 09:38:49 +07:00
Marten Seemann
592fb9cad9 introduce a dedicated qerr.TransportError and qerr.ApplicationError 2021-05-01 09:38:48 +07:00
Marten Seemann
d2739abd74 Merge pull request #3148 from lucas-clemente/fix-on-loss-detection-busy-looping
declare the handshake confirmed when receiving an ACK for a 1-RTT packet
2021-04-30 23:57:17 +07:00
Marten Seemann
0413afd615 Merge pull request #3153 from lucas-clemente/trace-version-selection
trace and qlog version selection / negotiation
2021-04-28 18:13:18 +07:00
Marten Seemann
e7c4e756ad trace and qlog version selection / negotiation 2021-04-19 11:38:10 +07:00
Marten Seemann
17d9860db6 declare the handshake confirmed when receiving an ACK for a 1-RTT packet
... on the client side. Both the receipt of HANDSHAKE_DONE and the
receipt of an ACK for a 1-RTT packet are sufficient for declaring
confirmation of the handshake.
2021-04-14 22:33:45 +07:00
Marten Seemann
4917760726 attach a tracing ID to the session context 2021-04-14 16:39:41 +07:00
Marten Seemann
e3f36af2e2 Merge pull request #3109 from lucas-clemente/version-negotiation-logging
don't pass the QUIC version to the StartedConnection event
2021-04-02 19:29:55 +07:00
Marten Seemann
4b10e67bf1 Merge pull request #3132 from lucas-clemente/amplification-limit-fixes
various amplification limit fixes
2021-04-02 18:32:38 +07:00
Marten Seemann
431a1fa650 Merge pull request #3120 from lucas-clemente/fix-handshake-idle-timeout
fix calculation of the handshake idle timeout
2021-04-02 18:32:24 +07:00
Marten Seemann
63e810f0d8 fix calculation of the handshake idle timeout 2021-04-02 18:19:38 +07:00
Marten Seemann
7f55fc7107 Merge pull request #3138 from lucas-clemente/pmtud-after-handshake-confirmation
only start PMTUD after handshake confirmation
2021-04-02 18:02:40 +07:00
Marten Seemann
22daaf4536 don't call ClosedConnection when closing for version negotiation 2021-04-02 17:35:20 +07:00
Marten Seemann
f60306c4bf Merge pull request #3107 from lucas-clemente/fix-cubic-initialization
initialize the congestion controller with the actual max datagram size
2021-04-02 17:35:13 +07:00
Marten Seemann
c30a45ef6f don't pass the QUIC version to the StartedConnection event
The version might change in response to a Version Negotiation packet.
2021-04-02 17:33:49 +07:00
Marten Seemann
8c388e0aaf Merge pull request #3112 from lucas-clemente/fix-timeout-error-logging
fix duplicate logging of errors when the first error was a timeout error
2021-04-02 17:24:45 +07:00
Marten Seemann
6c0142cb4c notify the sent packet handler about all received bytes 2021-04-02 17:20:53 +07:00
Marten Seemann
91629cd979 only start PMTUD after handshake confirmation
Our PMTUD packets are never coalesced packets. Since we try packing
coalesced packets until the handshake is confirmed, it makes sense to start
PMTUD then.
2021-04-02 17:07:00 +07:00
Ari Mattila
e098ccd2b3 fix calculation of the time for the next keep alive 2021-03-26 16:28:13 +07:00
Marten Seemann
04074fbea3 remove special treatment of timeout error when logging 2021-03-22 14:29:07 +08:00
Marten Seemann
154b55b50a introduce a qerr.Err{Idle,Handshake}Timeout 2021-03-22 14:22:36 +08:00
Marten Seemann
7feda789d2 initialize the congestion controller with the actual max datagram size 2021-03-20 14:02:28 +08:00
Marten Seemann
7ea53e6c2c rename the Config values for Max{Stream, Connection}ReceiveWindow 2021-03-16 23:15:18 +08:00
Marten Seemann
354bbb0e2e rename the Config values for Initial{Stream, Connection}ReceiveWindow 2021-03-16 23:15:18 +08:00
Marten Seemann
f51eff48e9 Merge pull request #3093 from lucas-clemente/improve-packet-info
cache the serialized OOB in the conn, not in the packet info
2021-03-16 22:18:25 +08:00
Marten Seemann
5073e300a6 Merge pull request #3092 from lucas-clemente/version-negotiation-failure-logging
make it possible to detect version negotiation failures in logging, fix qlogging of those
2021-03-16 22:17:34 +08:00
Marten Seemann
59eb3ed2fb Merge pull request #3083 from lucas-clemente/configure-initial-flow-control-window
make the initial stream / connection flow control windows configurable
2021-03-16 22:17:08 +08:00
Marten Seemann
1d2756f9f7 Merge pull request #3085 from lucas-clemente/client-hold-tp-until-handshake-completion
only apply server's transport parameters after handshake completion
2021-03-16 22:16:46 +08:00
Marten Seemann
6102a9b308 Merge pull request #3088 from lucas-clemente/update-flow-control-for-0rtt-streams
update the flow control windows of streams opened in 0-RTT
2021-03-16 22:15:53 +08:00
Marten Seemann
162cb16b31 cache the serialized OOB in the conn, not in the packet info
We're allocating a lot of packetInfo structs during the lifetime of a
connection. It's better to keep that struct as small as possible.
2021-03-16 14:04:15 +08:00
Olivier Poitrey
eb6bdfdfc1 Use the correct source IP when binding multiple IPs
When the server is listening on multiple interfaces or interfaces with
multiple IPs, the outgoing datagrams are sometime delivered with the
wrong source IP address.

In order to fix that, each quic connection needs to extract the
destination IP (and optionally interface id) of the received datagrams,
and set it as source IP (and interface) on the sent datagrams.

On most platforms, this can be done using ancillary data with recvmsg()
and sendmsg(). Some of the machinery for this is already there for ECN,
this change extends it to read the destination IP info and write it to
the outgoing packets.

Fix #1736
2021-03-16 00:50:05 +01:00
Marten Seemann
5208845191 introduce a logging.CloseReason for version negotiation errors 2021-03-15 13:36:45 +08:00
Marten Seemann
c27f5a55b3 rename sendStream.handleMaxStreamDataFrame to updateSendWindow 2021-03-12 13:36:23 +08:00
Marten Seemann
d970a1027d only apply server's transport parameters after handshake completion
On the client side, 0-RTT packets have to be sent using the old
transport parameters. We're only allowed to use the updated values for
1-RTT packets.
2021-03-11 18:58:22 +08:00
Marten Seemann
a15cfd6871 make the initial connection flow control window configurable 2021-03-10 15:04:38 +08:00
Marten Seemann
d9434f523e make the initial stream flow control window configurable 2021-03-10 14:58:15 +08:00