Age | Commit message (Collapse) | Author |
|
RACK (Recent Acknowledgement) is a new loss detection
algorithm in TCP. These are the fields which should be
stored on connections to implement RACK algorithm.
PiperOrigin-RevId: 324948703
|
|
Envoy (#170) uses this to get the original destination of redirected
packets.
|
|
CurrentConnected counter is incorrectly decremented on close of an
endpoint which is still not connected.
Fixes #3443
PiperOrigin-RevId: 324155171
|
|
In
https://github.com/google/gvisor/commit/ca6bded95dbce07f9683904b4b768dfc2d4a09b2
we reduced the default buffer size to 32KB. This mostly works fine except at
high throughput where we hit zero window very quickly and the TCP receive
buffer moderation is not able to grow the window. This can be seen in the
benchmarks where with a 32KB buffer and 100 connections downloading a 10MB
file we get about 30 requests/s vs the 1MB buffer gives us about 53 requests/s.
A proper fix requires a few changes to when we send a zero window as well as
when we decide to send a zero window update. Today we consider available space
below 1MSS as zero and send an update when it crosses 1MSS of available space.
This is way too low and results in the window staying very small once we hit
a zero window condition as we keep sending updates with size barely over 1MSS.
Linux and BSD are smarter about this and use different thresholds. We should
separately update our logic to match linux or BSD so that we don't send
window updates that are really tiny or wait until we drop below 1MSS to
advertise a zero window.
PiperOrigin-RevId: 324087019
|
|
When sending packets to a known network's broadcast address, use the
broadcast MAC address.
Test:
- stack_test.TestOutgoingSubnetBroadcast
- udp_test.TestOutgoingSubnetBroadcast
PiperOrigin-RevId: 324062407
|
|
PiperOrigin-RevId: 323715260
|
|
TCP now tracks the overhead of the segment structure itself in it's out-of-order
queue (pending). This is required to ensure that a malicious sender sending 1
byte out-of-order segments cannot queue like 1000's of segments which bloat up
memory usage.
We also reduce the default receive window to 32KB. With TCP moderation there is
no need to keep this window at 1MB which means that for new connections the
default out-of-order queue will be small unless the application actually reads
the data that is being sent. This prevents a sender from just maliciously
filling up pending buf with lots of tiny out-of-order segments.
PiperOrigin-RevId: 323450913
|
|
Changes the API of tcpip.Clock to also provide a method for scheduling and
rescheduling work after a specified duration. This change also implements the
AfterFunc method for existing implementations of tcpip.Clock.
This is the groundwork required to mock time within tests. All references to
CancellableTimer has been replaced with the tcpip.Job interface, allowing for
custom implementations of scheduling work.
This is a BREAKING CHANGE for clients that implement their own tcpip.Clock or
use tcpip.CancellableTimer. Migration plan:
1. Add AfterFunc(d, f) to tcpip.Clock
2. Replace references of tcpip.CancellableTimer with tcpip.Job
3. Replace calls to tcpip.CancellableTimer#StopLocked with tcpip.Job#Cancel
4. Replace calls to tcpip.CancellableTimer#Reset with tcpip.Job#Schedule
5. Replace calls to tcpip.NewCancellableTimer with tcpip.NewJob.
PiperOrigin-RevId: 322906897
|
|
PiperOrigin-RevId: 322853192
|
|
Fixes #3334
PiperOrigin-RevId: 322846384
|
|
Previously, ICMP destination unreachable datagrams were ignored by TCP
endpoints. This caused connect to hang when an intermediate router
couldn't find a route to the host.
This manifested as a Kokoro error when Docker IPv6 was enabled. The Ruby
image test would try to install the sinatra gem and hang indefinitely
attempting to use an IPv6 address.
Fixes #3079.
|
|
Updates #173
PiperOrigin-RevId: 322665518
|
|
Updates #173
PiperOrigin-RevId: 321690756
|
|
Packet sockets also seem to allow double binding and do not return an error on
linux. This was tested by running the syscall test in a linux namespace as root
and the current test DoubleBind fails@HEAD.
Passes after this change.
Updates #173
PiperOrigin-RevId: 321445137
|
|
As in Linux, we must periodically clean up unused connections.
PiperOrigin-RevId: 321003353
|
|
Updates #2746
PiperOrigin-RevId: 320757963
|
|
RFC-1122 (and others) specify that UDP should not receive
datagrams that have a source address that is a multicast address.
Packets should never be received FROM a multicast address.
See also, RFC 768: 'User Datagram Protocol'
J. Postel, ISI, 28 August 1980
A UDP datagram received with an invalid IP source address
(e.g., a broadcast or multicast address) must be discarded
by UDP or by the IP layer (see rfc 1122 Section 3.2.1.3).
This CL does not address TCP or broadcast which is more complicated.
Also adds a test for both ipv6 and ipv4 UDP.
Fixes #3154
PiperOrigin-RevId: 320547674
|
|
Updates #2746
Fixes #3158
PiperOrigin-RevId: 320497190
|
|
PiperOrigin-RevId: 320250773
|
|
RFC 6864 imposes various restrictions on the uniqueness of the IPv4
Identification field for non-atomic datagrams, defined as an IP datagram that
either can be fragmented (DF=0) or is already a fragment (MF=1 or positive
fragment offset). In order to be compliant, the ID field is assigned for all
non-atomic datagrams.
Add a TCP unit test that induces retransmissions and checks that the IPv4
ID field is unique every time. Add basic handling of the IP_MTU_DISCOVER
socket option so that the option can be used to disable PMTU discovery,
effectively setting DF=0. Attempting to set the sockopt to anything other
than disabled will fail because PMTU discovery is currently not implemented,
and the default behavior matches that of disabled.
PiperOrigin-RevId: 320081842
|
|
The current convention is when a header is set to pkt.XxxHeader field, it
gets removed from pkt.Data. ICMP does not currently follow this convention.
PiperOrigin-RevId: 320078606
|
|
Updates #2746
PiperOrigin-RevId: 319887810
|
|
stack_x_test: 2m -> 20s
tcp_x_test: 80s -> 25s
PiperOrigin-RevId: 319828101
|
|
PiperOrigin-RevId: 319770124
|
|
Avoid a race where an arbitrary goroutine scheduling delay can cause the
processor to miss events and hang indefinitely.
Reduce allocations by storing processors by-value in the dispatcher, and
by using a single WaitGroup rather than one per processor.
PiperOrigin-RevId: 319665861
|
|
The application can choose to initiate a non-blocking connect and
later block on a read, when the endpoint is still in SYN-SENT state.
PiperOrigin-RevId: 319311016
|
|
a) When GSO is in use we should not cap the segment to maxPayloadSize in
sender.maybeSendSegment as the GSO logic will cap the segment to the correct
size. Without this the host GSO is not used as we end up breaking up large
segments into small MSS sized segments before writing the packets to the
host.
b) The check to not split a segment due to it not fitting in the receiver window
when there are pending segments is incorrect as segments in writeList can be
really large as we just take the write call's buffer size and create a single
large segment. So a write of say 128KB will just be 1 segment in the
writeList.
The linux code checks if 1 MSS sized segments fits in the receiver's window
and if not then does not split the current segment. gVisor's check was
incorrect that it was checking if the whole segment which could be >>> 1 MSS
would fit in the receiver's window. This was causing us to prematurely stop
sending and falling back to retransmit timer/probe from the other end to send
data.
This was seen when running HTTPD benchmarks where @ HEAD when sending large
files the benchmark was taking forever to run.
The tcp_splitseg_mss_test.go is being deleted as the test as written doesn't
test what is intended correctly. This is because GSO is enabled by default and
the reason the MSS+1 sized segment is sent is because GSO is in use. A proper
test will require disabling GSO on linux and netstack which is going to take a
bit of work in packetimpact to do it correctly.
Separately a new test probably should be written that verifies that a segment >
availableWindow is not split if the availableWindow is < 1 MSS.
Fixes #3107
PiperOrigin-RevId: 319172089
|
|
...by calling (*tcp.endpoint).EndpointState only once when possible.
Avoid wrapping (*sleep.Waker).Assert in a useless func while I'm here.
PiperOrigin-RevId: 319074149
|
|
IPv6 raw sockets never include the IPv6 header.
PiperOrigin-RevId: 318582989
|
|
SO_NO_CHECK is used to skip the UDP checksum generation on a TX socket
(UDP checksum is optional on IPv4).
Test:
- TestNoChecksum
- SoNoCheckOffByDefault (UdpSocketTest)
- SoNoCheck (UdpSocketTest)
Fixes #3055
PiperOrigin-RevId: 318575215
|
|
Linux controls socket send/receive buffers using a few sysctl variables
- net.core.rmem_default
- net.core.rmem_max
- net.core.wmem_max
- net.core.wmem_default
- net.ipv4.tcp_rmem
- net.ipv4.tcp_wmem
The first 4 control the default socket buffer sizes for all sockets
raw/packet/tcp/udp and also the maximum permitted socket buffer that can be
specified in setsockopt(SOL_SOCKET, SO_(RCV|SND)BUF,...).
The last two control the TCP auto-tuning limits and override the default
specified in rmem_default/wmem_default as well as the max limits.
Netstack today only implements tcp_rmem/tcp_wmem and incorrectly uses it
to limit the maximum size in setsockopt() as well as uses it for raw/udp
sockets.
This changelist introduces the other 4 and updates the udp/raw sockets to use
the newly introduced variables. The values for min/max match the current
tcp_rmem/wmem values and the default value buffers for UDP/RAW sockets is
updated to match the linux value of 212KiB up from the really low current value
of 32 KiB.
Updates #3043
Fixes #3043
PiperOrigin-RevId: 318089805
|
|
For TCP sockets, SO_REUSEADDR relaxes the rules for binding addresses.
gVisor/netstack already supported a behavior similar to SO_REUSEADDR, but did
not allow disabling it. This change brings the SO_REUSEADDR behavior closer to
the behavior implemented by Linux and adds a new SO_REUSEADDR disabled
behavior. Like Linux, SO_REUSEADDR is now disabled by default.
PiperOrigin-RevId: 317984380
|
|
Test:
- TestIncrementChecksumErrors
Fixes #2943
PiperOrigin-RevId: 317348158
|
|
Updates #173,#6
Fixes #2888
PiperOrigin-RevId: 317087652
|
|
When a tcp.timer or tcpip.Route is no longer used, clean up its
resources so that unused memory may be released.
PiperOrigin-RevId: 317046582
|
|
Ensure that CurrentConnected stat is updated on any errors and cleanups
during connected state processing.
Fixes #2968
PiperOrigin-RevId: 316919426
|
|
PiperOrigin-RevId: 316767969
|
|
In passive open cases, we transition to Established state after
initializing endpoint's sender and receiver. With this we lose out
on any updates coming from the ACK that completes the handshake.
This change ensures that we uniformly transition to Established in all
cases and does minor cleanups.
Fixes #2938
PiperOrigin-RevId: 316567014
|
|
I am not really sure what the point of this is, but someone filed a bug about
it, so I assume something relies on it.
PiperOrigin-RevId: 316225127
|
|
On UDP sockets, SO_REUSEADDR allows multiple sockets to bind to the same
address, but only delivers packets to the most recently bound socket. This
differs from the behavior of SO_REUSEADDR on TCP sockets. SO_REUSEADDR for TCP
sockets will likely need an almost completely independent implementation.
SO_REUSEADDR has some odd interactions with the similar SO_REUSEPORT. These
interactions are tested fairly extensively and all but one particularly odd
one (that honestly seems like a bug) behave the same on gVisor and Linux.
PiperOrigin-RevId: 315844832
|
|
After this change e.mu is only promoted to exclusively locked during
route.Resolve. It downgrades back to read-lock afterwards.
This prevents the second RLock() call gets stuck later in the stack.
https://syzkaller.appspot.com/bug?id=065b893bd8d1d04a4e0a1d53c578537cde1efe99
Syzkaller logs does not contain interesting stack traces.
The following stack trace is obtained by running repro locally.
goroutine 53 [semacquire, 3 minutes]:
runtime.gopark(0xfd4278, 0x1896320, 0xc000301912, 0x4)
GOROOT/src/runtime/proc.go:304 +0xe0 fp=0xc0000e25f8 sp=0xc0000e25d8 pc=0x437170
runtime.goparkunlock(...)
GOROOT/src/runtime/proc.go:310
runtime.semacquire1(0xc0001220b0, 0xc00000a300, 0x1, 0x0)
GOROOT/src/runtime/sema.go:144 +0x1c0 fp=0xc0000e2660 sp=0xc0000e25f8 pc=0x4484e0
sync.runtime_Semacquire(0xc0001220b0)
GOROOT/src/runtime/sema.go:56 +0x42 fp=0xc0000e2690 sp=0xc0000e2660 pc=0x448132
gvisor.dev/gvisor/pkg/sync.(*RWMutex).RLock(...)
pkg/sync/rwmutex_unsafe.go:76
gvisor.dev/gvisor/pkg/tcpip/transport/udp.(*endpoint).HandleControlPacket(0xc000122000, 0x7ee5, 0xc00053c16c, 0x4, 0x5e21, 0xc00053c224, 0x4, 0x1, 0x0, 0xc00007ed00)
pkg/tcpip/transport/udp/endpoint.go:1345 +0x169 fp=0xc0000e26d8 sp=0xc0000e2690 pc=0x9843f9
......
gvisor.dev/gvisor/pkg/tcpip/transport/udp.(*protocol).HandleUnknownDestinationPacket(0x18bb5a0, 0xc000556540, 0x5e21, 0xc00053c16c, 0x4, 0x7ee5, 0xc00053c1ec, 0x4, 0xc00007e680, 0x4)
pkg/tcpip/transport/udp/protocol.go:143 +0xb9a fp=0xc0000e8260 sp=0xc0000e7510 pc=0x9859ba
......
gvisor.dev/gvisor/pkg/tcpip/transport/udp.sendUDP(0xc0001220d0, 0xc00053ece0, 0x1, 0x1, 0x883, 0x1405e217ee5, 0x11100a0, 0xc000592000, 0xf88780)
pkg/tcpip/transport/udp/endpoint.go:924 +0x3b0 fp=0xc0000ed390 sp=0xc0000ec750 pc=0x981af0
gvisor.dev/gvisor/pkg/tcpip/transport/udp.(*endpoint).write(0xc000122000, 0x11104e0, 0xc00020a460, 0x0, 0x0, 0x0, 0x0, 0x0)
pkg/tcpip/transport/udp/endpoint.go:510 +0x4ad fp=0xc0000ed658 sp=0xc0000ed390 pc=0x97f2dd
PiperOrigin-RevId: 315590041
|
|
Netstack has traditionally parsed headers on-demand as a packet moves up the
stack. This is conceptually simple and convenient, but incompatible with
iptables, where headers can be inspected and mangled before even a routing
decision is made.
This changes header parsing to happen early in the incoming packet path, as soon
as the NIC gets the packet from a link endpoint. Even if an invalid packet is
found (e.g. a TCP header of insufficient length), the packet is passed up the
stack for proper stats bookkeeping.
PiperOrigin-RevId: 315179302
|
|
PiperOrigin-RevId: 315018295
|
|
PiperOrigin-RevId: 314996457
|
|
For TCP sockets gVisor incorrectly returns EAGAIN when no ephemeral ports are
available to bind during a connect. Linux returns EADDRNOTAVAIL. This change
fixes gVisor to return the correct code and adds a test for the same.
This change also fixes a minor bug for ping sockets where connect() would fail
with EINVAL unless the socket was bound first.
Also added tests for testing UDP Port exhaustion and Ping socket port
exhaustion.
PiperOrigin-RevId: 314988525
|
|
IPTables.connections contains a sync.RWMutex. Copying it will trigger copylocks
analysis. Tested by manually enabling nogo tests.
sync.RWMutex is added to IPTables for the additional race condition discovered.
PiperOrigin-RevId: 314817019
|
|
- Always split segments larger than MSS.
Currently, we base the segment split decision as a function of the
send congestion window and MSS, which could be greater than the MSS
advertised by remote.
- While splitting segments, ensure the PSH flag is reset when there
are segments that are queued to be sent.
- With TCP_CORK, hold up segments up until MSS. Fix a bug in computing
available send space before attempting to coalesce segments.
Fixes #2832
PiperOrigin-RevId: 314802928
|
|
Historically we've been passing PacketBuffer by shallow copying through out
the stack. Right now, this is only correct as the caller would not use
PacketBuffer after passing into the next layer in netstack.
With new buffer management effort in gVisor/netstack, PacketBuffer will
own a Buffer (to be added). Internally, both PacketBuffer and Buffer may
have pointers and shallow copying shouldn't be used.
Updates #2404.
PiperOrigin-RevId: 314610879
|
|
If the entire segment cannot be accommodated in the receiver advertised
window and if there are still unacknowledged pending segments, skip
splitting the segment. The segment transmit would get retried by the
retransmit handler.
PiperOrigin-RevId: 314538523
|
|
RST handling is broken when the TCP state transitions
from SYN-SENT to SYN-RCVD in case of simultaneous open.
An incoming RST should trigger cleanup of the endpoint.
RFC793, section 3.9, page 70.
Fixes #2814
PiperOrigin-RevId: 313828777
|