Add support for setting SO_SNDBUF for unix domain sockets.

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

1 files changed, 13 insertions, 5 deletions

diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go
index 72aa535f8..6763f5b0c 100644
--- a/pkg/sentry/fsimpl/host/socket.go
+++ b/pkg/sentry/fsimpl/host/socket.go
@@ -16,6 +16,7 @@ package host
 
 import (
 	"fmt"
+	"sync/atomic"
 	"syscall"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
@@ -111,7 +112,7 @@ func (c *ConnectedEndpoint) init() *syserr.Error {
 	}
 
 	c.stype = linux.SockType(stype)
-	c.sndbuf = int64(sndbuf)
+	atomic.StoreInt64(&c.sndbuf, int64(sndbuf))
 
 	return nil
 }
@@ -150,7 +151,7 @@ func (c *ConnectedEndpoint) Send(ctx context.Context, data [][]byte, controlMess
 	// only as much of the message as fits in the send buffer.
 	truncate := c.stype == linux.SOCK_STREAM
 
-	n, totalLen, err := fdWriteVec(c.fd, data, c.sndbuf, truncate)
+	n, totalLen, err := fdWriteVec(c.fd, data, c.SendMaxQueueSize(), truncate)
 	if n < totalLen && err == nil {
 		// The host only returns a short write if it would otherwise
 		// block (and only for stream sockets).
@@ -226,7 +227,7 @@ func (c *ConnectedEndpoint) Recv(ctx context.Context, data [][]byte, creds bool,
 
 	// N.B. Unix sockets don't have a receive buffer, the send buffer
 	// serves both purposes.
-	rl, ml, cl, cTrunc, err := fdReadVec(c.fd, data, []byte(cm), peek, c.sndbuf)
+	rl, ml, cl, cTrunc, err := fdReadVec(c.fd, data, []byte(cm), peek, c.RecvMaxQueueSize())
 	if rl > 0 && err != nil {
 		// We got some data, so all we need to do on error is return
 		// the data that we got. Short reads are fine, no need to
@@ -300,14 +301,14 @@ func (c *ConnectedEndpoint) RecvQueuedSize() int64 {
 
 // SendMaxQueueSize implements transport.Receiver.SendMaxQueueSize.
 func (c *ConnectedEndpoint) SendMaxQueueSize() int64 {
-	return int64(c.sndbuf)
+	return atomic.LoadInt64(&c.sndbuf)
 }
 
 // RecvMaxQueueSize implements transport.Receiver.RecvMaxQueueSize.
 func (c *ConnectedEndpoint) RecvMaxQueueSize() int64 {
 	// N.B. Unix sockets don't use the receive buffer. We'll claim it is
 	// the same size as the send buffer.
-	return int64(c.sndbuf)
+	return atomic.LoadInt64(&c.sndbuf)
 }
 
 func (c *ConnectedEndpoint) destroyLocked() {
@@ -327,6 +328,13 @@ func (c *ConnectedEndpoint) Release(ctx context.Context) {
 // CloseUnread implements transport.ConnectedEndpoint.CloseUnread.
 func (c *ConnectedEndpoint) CloseUnread() {}
 
+// SetSendBufferSize implements transport.ConnectedEndpoint.SetSendBufferSize.
+func (c *ConnectedEndpoint) SetSendBufferSize(v int64) (newSz int64) {
+	// gVisor does not permit setting of SO_SNDBUF for host backed unix domain
+	// sockets.
+	return atomic.LoadInt64(&c.sndbuf)
+}
+
 // SCMConnectedEndpoint represents an endpoint backed by a host fd that was
 // passed through a gofer Unix socket. It resembles ConnectedEndpoint, with the
 // following differences: