| Age | Commit message (Collapse) | Author |
|---|
|
This code path is for outgoing packets, and we don't currently do memory
accounting on this path. So it wasn't breaking anything.
This change did not add a test for ref-counting issue fixed, but will switch to
the leak-checking ref-counter later when all ref-counting issues are fixed.
PiperOrigin-RevId: 373447913
|
|
On receiving an ICMP error during handshake, the error is propagated
by reading `endpoint.lastError`. This can race with the socket layer
invoking getsockopt() with SO_ERROR where the same value is read and
cleared, causing the handshake to bail out with a non-error state.
Fix the race by checking for lastError state and failing the
handshake with ErrConnectionAborted if the lastError was read and
cleared by say SO_ERROR.
The race mentioned in the bug, is caught only with the newly added
tcp_test unit test, where we have control over stopping/resuming
protocol loop. Adding a packetimpact test as well for sanity testing
of ICMP error handling during handshake.
Fixes #5922
PiperOrigin-RevId: 372135662
|
|
Listen backlog value is 1 more than what is configured by the socket
layer listen call. TestListenBacklogFull expects this behavior which
is incorrect as it directly invokes endpoint Listen and with
cl/369974744, backlog++ logic is moved to the callers of Listen().
This test passes sometimes, because the handshakes could overlap causing
the last SYN to arrive at the listener before the previous handshake is
enqueued to the accept queue. In such a case the accept queue is still
not full and the SYN is replied to. The final ACK of this last handshake
would get dropped eventually.
PiperOrigin-RevId: 372041827
|
|
PiperOrigin-RevId: 372021039
|
|
...as all GSO capable endpoints must implement GSOEndpoint.
PiperOrigin-RevId: 371804175
|
|
PiperOrigin-RevId: 371231148
|
|
In https://github.com/google/gvisor/commit/f075522849fa a check to increase zero
to a minimum backlog length was removed from sys_socket.go to bring it in parity
with linux and then in tcp/endpoint.go we bump backlog by 1. But this broke
calling listen on a AF_UNIX socket w/ a zero backlog as in linux it does allow 1
connection even with a zero backlog.
This was caught by a php runtime test socket_abstract_path.phpt.
PiperOrigin-RevId: 369974744
|
|
With this change, GSO options no longer needs to be passed around as
a function argument in the write path.
This change is done in preparation for a later change that defers
segmentation, and may change GSO options for a packet as it flows
down the stack.
Updates #170.
PiperOrigin-RevId: 369774872
|
|
Fixes #2926, #674
PiperOrigin-RevId: 369457123
|
|
This is done for IPv4, UDP and TCP headers.
This also changes the packet checkers used in tests to error on
zero-checksum, not sure why it was allowed before.
And while I'm here, make comments' case consistent.
RELNOTES: n/a
Fixes #5049
PiperOrigin-RevId: 369383862
|
|
This change replaces individual private members in tcp.endpoint with a single
private TCPEndpointState member.
Some internal substructures within endpoint (receiver, sender) have been broken
into a public substructure (which is then copied into the TCPEndpointState
returned from completeState()) alongside other private fields.
Fixes #4466
PiperOrigin-RevId: 369329514
|
|
- Added delay to increase the RTT: In DSACK tests with RACK enabled and low
RTT, TLP can be detected before sending ACK and the tests flake. Increasing
the RTT will ensure that TLP does not happen before the ACK is sent.
- Fix TestRACKOnePacketTailLoss: The ACK does not contain DSACK, which means
either the original or retransmission (probe) was lost and SACKRecovery count
must be incremented.
Before: http://sponge2/c9bd51de-f72f-481c-a7f3-e782e7524883
After: http://sponge2/1307a796-103a-4a45-b699-e8d239220ed1
PiperOrigin-RevId: 369305720
|
|
Also count failed TCP port allocations
PiperOrigin-RevId: 368939619
|
|
Reduce the ephemeral port range, which decreases the calls to makeEP.
PiperOrigin-RevId: 368748379
|
|
This was semi-automated -- there are many addresses that were not replaced.
Future commits should clean those up.
Parse4 and Parse6 were given their own package because //pkg/test can introduce
dependency cycles, as it depends transitively on //pkg/tcpip and some other
netstack packages.
PiperOrigin-RevId: 368726528
|
|
Fix a race where the ACK completing the handshake can be dropped by
a closing listener without RST to the peer. The listener close would
reset the accepted queue and that causes the connecting endpoint
in SYNRCVD state to drop the ACK thinking the queue if filled up.
PiperOrigin-RevId: 368165509
|
|
Holding this lock can cause the user's callback to deadlock if it
attempts to inspect the accept queue.
PiperOrigin-RevId: 368068334
|
|
Some other cleanup while I'm here:
- Remove unused arguments
- Handle some unhandled errors
- Remove redundant casts
- Remove redundant parens
- Avoid shadowing `hash` package name
PiperOrigin-RevId: 367816161
|
|
Use a linked list with cached length and capacity. The current channel
is already composed with a mutex and condition variable, and is never
used for its channel-like properties. Channels also require eager
allocation equal to their capacity, which a linked list does not.
PiperOrigin-RevId: 367766626
|
|
Move maxListenBacklog check to the caller of endpoint Listen so that it
is applicable to Unix domain sockets as well.
This was changed in cl/366935921.
Reported-by: syzbot+a35ae7cdfdde0c41cf7a@syzkaller.appspotmail.com
PiperOrigin-RevId: 367728052
|
|
Both code paths perform this check; extract it and remove the comment
that suggests it is unique to one of the paths.
PiperOrigin-RevId: 367666160
|
|
Both callers of this function still drop this error on the floor, but
progress is progress.
Updates #4690.
PiperOrigin-RevId: 367604788
|
|
- Change the accept queue full condition for a listening endpoint
to only honor completed (and delivered) connections.
- Use syncookies if the number of incomplete connections is beyond
listen backlog. This also cleans up the SynThreshold option code
as that is no longer used with this change.
- Added a new stack option to unconditionally generate syncookies.
Similar to sysctl -w net.ipv4.tcp_syncookies=2 on Linux.
- Enable keeping of incomplete connections beyond listen backlog.
- Drop incoming SYNs only if the accept queue is filled up.
- Drop incoming ACKs that complete handshakes when accept queue is full
- Enable the stack to accept one more connection than programmed by
listen backlog.
- Handle backlog argument being zero, negative for listen, as Linux.
- Add syscall and packetimpact tests to reflect the changes above.
- Remove TCPConnectBacklog test which is polling for completed
connections on the client side which is not reflective of whether
the accept queue is filled up by the test. The modified syscall test
in this CL addresses testing of connecting sockets.
Fixes #3153
PiperOrigin-RevId: 366935921
|
|
On Linux these are meant to be equivalent to POLLIN/POLLOUT. Rather
than hack these on in sys_poll etc it felt cleaner to just cleanup
the call sites to notify for both events. This is what linux does
as well.
Fixes #5544
PiperOrigin-RevId: 364859977
|
|
This change sets the inner `routeInfo` struct to be a named private member
and replaces direct access with access through getters. Note that direct
access to the fields of `routeInfo` is still possible through the `RouteInfo`
struct.
Fixes #4902
PiperOrigin-RevId: 364822872
|
|
PiperOrigin-RevId: 364596526
|
|
Transport demuxer and UDP tests should not use a loopback address as the
source address for packets injected into the stack as martian loopback
packets will be dropped in a later change.
PiperOrigin-RevId: 363479681
|
|
Netstack does not check ACK number for FIN-ACK packets and goes into TIMEWAIT
unconditionally. Fixing the state machine will give us back the retransmission
of FIN.
PiperOrigin-RevId: 363301883
|
|
There is a race in handling new incoming connections on a listening
endpoint that causes the endpoint to reply to more incoming SYNs than
what is permitted by the listen backlog.
The race occurs when there is a successful passive connection handshake
and the synRcvdCount counter is decremented, followed by the endpoint
delivered to the accept queue. In the window of time between
synRcvdCount decrementing and the endpoint being enqueued for accept,
new incoming SYNs can be handled without honoring the listen backlog
value, as the backlog could be perceived not full.
Fixes #5637
PiperOrigin-RevId: 363279372
|
|
Doing so involved breaking dependencies between //pkg/tcpip and the rest
of gVisor, which are discouraged anyways.
Tested on the Go branch via:
gvisor.dev/gvisor/pkg/tcpip/...
Addresses #1446.
PiperOrigin-RevId: 363081778
|
|
Lots of small changes:
- simplify package API via Reservation type
- rename some single-letter variable names that were hard to follow
- rename some types
PiperOrigin-RevId: 362442366
|
|
- Implement Stringer for it so that we can improve error messages.
- Use TCPFlags through the code base. There used to be a mixed usage of byte,
uint8 and int as TCP flags.
PiperOrigin-RevId: 361940150
|
|
Speeds up the socket stress tests by a couple orders of magnitude.
PiperOrigin-RevId: 361721050
|
|
Updates #5597
PiperOrigin-RevId: 361252003
|
|
The only user is in (*handshake).complete and it specifies MaxRTO, so there is
no behavior changes.
PiperOrigin-RevId: 360954447
|
|
One of the preparation to decouple underlying buffer implementation.
There are still some methods that tie to VectorisedView, and they will be
changed gradually in later CLs.
This CL also introduce a new ICMPv6ChecksumParams to replace long list of
parameters when calling ICMPv6Checksum, aiming to be more descriptive.
PiperOrigin-RevId: 360778149
|
|
This validates that struct fields if annotated with "// checklocks:mu" where
"mu" is a mutex field in the same struct then access to the field is only
done with "mu" locked.
All types that are guarded by a mutex must be annotated with
// +checklocks:<mutex field name>
For more details please refer to README.md.
PiperOrigin-RevId: 360729328
|
|
io.Reader.ReadFull returns the number of bytes copied and an error if fewer
bytes were read.
PiperOrigin-RevId: 360247614
|
|
There is a short race where in Write an endpoint can transition from writable
to non-writable state due to say an incoming RST during the time we release
the endpoint lock and reacquire after copying the payload. In such a case
if the write happens to be a zero sized write we end up trying to call
sendData() even though nothing was queued.
This can panic when trying to enable/disable TCP timers if the endpoint had
already transitioned to a CLOSED/ERROR state due to the incoming RST as we
cleanup timers when the protocol goroutine terminates.
Sadly the race window is small enough that my attempts at reproducing the panic
in a syscall test has not been successful.
PiperOrigin-RevId: 359887905
|
|
Also increase refcount of raw.endpoint.route while in use.
Avoid allocating an array of size zero.
PiperOrigin-RevId: 359797788
|
|
Use maybeSendSegment while sending segments in RACK recovery which checks if
the receiver has space and splits the segments when the segment size is
greater than MSS.
PiperOrigin-RevId: 359641097
|
|
Prevents the following deadlock:
- Raw packet is sent via e.Write(), which read locks e.mu
- Connect() is called, blocking on write locking e.mu
- The packet is routed to loopback and back to e.HandlePacket(), which read
locks e.mu
Per the atomic.RWMutex documentation, this deadlocks:
"If a goroutine holds a RWMutex for reading and another goroutine might call
Lock, no goroutine should expect to be able to acquire a read lock until the
initial read lock is released. In particular, this prohibits recursive read
locking. This is to ensure that the lock eventually becomes available; a blocked
Lock call excludes new readers from acquiring the lock."
Also, release eps.mu earlier in deliverRawPacket.
PiperOrigin-RevId: 359600926
|
|
This change implements TLP details enumerated in
https://tools.ietf.org/html/draft-ietf-tcpm-rack-08#section-7.5.3
Fixes #5085
PiperOrigin-RevId: 357125037
|
|
Entry check:
- Earlier implementation was preventing us from entering recovery even if
SND.UNA is lost but dupAckCount is still below threshold. Fixed that.
- We should only enter recovery when at least one more byte of data beyond the
highest byte that was outstanding when fast retransmit was last entered is
acked. Added that check.
Exit check:
- Earlier we were checking if SEG.ACK is in range [SND.UNA, SND.NXT]. The
intention was to check if any unacknowledged data was ACKed. Note that
(SEG.ACK - 1) is actually the sequence number which was ACKed. So we were
incorrectly including (SND.UNA - 1) in the range. Fixed the check to now be
(SEG.ACK - 1) in range [SND.UNA, SND.NXT).
Additionally, moved a RACK specific test to the rack tests file.
Added tests for the changes I made.
PiperOrigin-RevId: 357091322
|
|
TestRACKWithDuplicateACK is flaky as the reorder window can expire before
receiving three duplicate ACKs which will result in sending the first
unacknowledged segment twice: when reorder timer expired and again after
receiving the third duplicate ACK.
This CL will fix this behavior and will not resend the segment again if it was
already re-transmittted when reorder timer expired.
Update the TestRACKWithDuplicateACK to test that the first segment is
considered as lost and is re-transmitted.
PiperOrigin-RevId: 356855168
|
|
The limits for snd/rcv buffers for unix domain socket is controlled by the
following sysctls on linux
- net.core.rmem_default
- net.core.rmem_max
- net.core.wmem_default
- net.core.wmem_max
Today in gVisor we do not expose these sysctls but we do support setting the
equivalent in netstack via stack.Options() method. But AF_UNIX sockets in gVisor
can be used without netstack, with hostinet or even without any networking stack
at all. Which means ideally these sysctls need to live as globals in gVisor.
But rather than make this a big change for now we hardcode the limits in the
AF_UNIX implementation itself (which in itself is better than where we were
before) where it SO_SNDBUF was hardcoded to 16KiB. Further we bump the initial
limit to a default value of 208 KiB to match linux from the paltry 16 KiB we use
today.
Updates #5132
PiperOrigin-RevId: 356665498
|
|
We previously return EINVAL when connecting to port 0, however this is not the
observed behavior on Linux. One of the observable effects after connecting to
port 0 on Linux is that getpeername() will fail with ENOTCONN.
PiperOrigin-RevId: 356413451
|
|
Detect packet loss using reorder window and re-transmit them after the reorder
timer expires.
PiperOrigin-RevId: 356321786
|
|
- Adds a function to enable RACK in tests.
- RACK update functions are guarded behind the flag tcpRecovery.
PiperOrigin-RevId: 355435973
|
|
Rename HandleNDupAcks() to HandleLossDetected() as it will enter this when
is detected after:
- reorder window expires and TLP (in case of RACK)
- dupAckCount >= 3
PiperOrigin-RevId: 355237858
|
