internal/uevent: separate recvmsg helper

This enables messages to be received separately.

Signed-off-by: Ophestra <cat@gensokyo.uk>

container/dispatcher.go

@@ -179,7 +179,7 @@ func (direct) mustLoopback(ctx context.Context, msg message.Msg) {
 		lo = ifi.Index
 	}
 
-	c, err := netlink.DialRoute()
+	c, err := netlink.DialRoute(0)
 	if err != nil {
 		msg.GetLogger().Fatalln(err)
 	}

internal/netlink/netlink.go

@@ -46,7 +46,11 @@ type Conn struct {
 
 // Dial returns the address of a newly connected generic netlink connection of
 // specified family and groups.
-func Dial(family int, groups uint32) (*Conn, error) {
+//
+// For a nonzero rcvbuf, the socket receive buffer size is set to its absolute
+// value via SO_RCVBUF for a positive value, or SO_RCVBUFFORCE for a negative
+// value.
+func Dial(family int, groups uint32, rcvbuf int64) (*Conn, error) {
 	var c Conn
 	if fd, err := syscall.Socket(
 		syscall.AF_NETLINK,
@@ -75,6 +79,23 @@ func Dial(family int, groups uint32) (*Conn, error) {
 			return nil, syscall.ENOTRECOVERABLE
 		}
 
+		if rcvbuf != 0 {
+			opt := syscall.SO_RCVBUF
+			if rcvbuf < 0 {
+				opt = syscall.SO_RCVBUFFORCE
+				rcvbuf = -rcvbuf
+			}
+			if err = syscall.SetsockoptInt(
+				fd,
+				syscall.SOL_SOCKET,
+				opt,
+				int(rcvbuf),
+			); err != nil {
+				_ = syscall.Close(fd)
+				return nil, os.NewSyscallError("setsockopt", err)
+			}
+		}
+
 		c.family = family
 		c.f = os.NewFile(uintptr(fd), "netlink")
 		if c.raw, err = c.f.SyscallConn(); err != nil {
@@ -101,13 +122,13 @@ func (c *Conn) Close() error {
 	return c.f.Close()
 }
 
-// Recvfrom wraps recv(2) with nonblocking behaviour via the runtime network poller.
+// Recvmsg wraps recv(2) with nonblocking behaviour via the runtime network poller.
 //
-// The returned slice is valid until the next call to Recvfrom.
-func (c *Conn) Recvfrom(
+// The returned slice is valid until the next call to Recvmsg.
+func (c *Conn) Recvmsg(
 	ctx context.Context,
 	flags int,
-) (data []byte, from syscall.Sockaddr, err error) {
+) (data []byte, recvflags int, from syscall.Sockaddr, err error) {
 	if err = c.f.SetReadDeadline(time.Time{}); err != nil {
 		return
 	}
@@ -117,7 +138,7 @@ func (c *Conn) Recvfrom(
 	done := make(chan error, 1)
 	go func() {
 		rcErr := c.raw.Read(func(fd uintptr) (done bool) {
-			n, from, err = syscall.Recvfrom(int(fd), data, flags)
+			n, _, recvflags, from, err = syscall.Recvmsg(int(fd), data, nil, flags)
 			return err != syscall.EWOULDBLOCK
 		})
 		if n >= 0 {
@@ -129,7 +150,7 @@ func (c *Conn) Recvfrom(
 	select {
 	case rcErr := <-done:
 		if err != nil {
-			err = os.NewSyscallError("recvfrom", err)
+			err = os.NewSyscallError("recvmsg", err)
 		} else {
 			err = rcErr
 		}
@@ -147,12 +168,12 @@ func (c *Conn) Recvfrom(
 	}
 }
 
-// Sendto wraps send(2) with nonblocking behaviour via the runtime network poller.
-func (c *Conn) Sendto(
+// Sendmsg wraps send(2) with nonblocking behaviour via the runtime network poller.
+func (c *Conn) Sendmsg(
 	ctx context.Context,
 	p []byte,
-	flags int,
 	to syscall.Sockaddr,
+	flags int,
 ) (err error) {
 	if err = c.f.SetWriteDeadline(time.Time{}); err != nil {
 		return
@@ -161,7 +182,7 @@ func (c *Conn) Sendto(
 	done := make(chan error, 1)
 	go func() {
 		done <- c.raw.Write(func(fd uintptr) (done bool) {
-			err = syscall.Sendto(int(fd), p, flags, to)
+			err = syscall.Sendmsg(int(fd), p, nil, to, flags)
 			return err != syscall.EWOULDBLOCK
 		})
 	}()
@@ -169,7 +190,7 @@ func (c *Conn) Sendto(
 	select {
 	case rcErr := <-done:
 		if err != nil {
-			err = os.NewSyscallError("sendto", err)
+			err = os.NewSyscallError("sendmsg", err)
 		} else {
 			err = rcErr
 		}
@@ -278,7 +299,7 @@ type HandlerFunc func(resp []syscall.NetlinkMessage) error
 func (c *Conn) receive(ctx context.Context, f HandlerFunc, flags int) error {
 	for {
 		var resp []syscall.NetlinkMessage
-		if data, _, err := c.Recvfrom(ctx, flags); err != nil {
+		if data, _, _, err := c.Recvmsg(ctx, flags); err != nil {
 			return err
 		} else if len(data) < syscall.NLMSG_HDRLEN {
 			return syscall.EBADE
@@ -302,9 +323,9 @@ func (c *Conn) Roundtrip(ctx context.Context, f HandlerFunc) error {
 	}
 	defer func() { c.seq++ }()
 
-	if err := c.Sendto(ctx, c.pending(), 0, &syscall.SockaddrNetlink{
+	if err := c.Sendmsg(ctx, c.pending(), &syscall.SockaddrNetlink{
 		Family: syscall.AF_NETLINK,
-	}); err != nil {
+	}, 0); err != nil {
 		return err
 	}
 

internal/netlink/rtnl.go

@@ -13,8 +13,8 @@ type RouteConn struct{ conn *Conn }
 func (c *RouteConn) Close() error { return c.conn.Close() }
 
 // DialRoute returns the address of a newly connected [RouteConn].
-func DialRoute() (*RouteConn, error) {
-	c, err := Dial(syscall.NETLINK_ROUTE, 0)
+func DialRoute(rcvbuf int64) (*RouteConn, error) {
+	c, err := Dial(syscall.NETLINK_ROUTE, 0, rcvbuf)
 	if err != nil {
 		return nil, err
 	}

internal/uevent/action.go

@@ -19,10 +19,6 @@ const (
 	KOBJ_OFFLINE
 	KOBJ_BIND
 	KOBJ_UNBIND
-
-	// Synthetic denotes a [Message] that originates from outside the kernel. It
-	// is not valid in the wire format and is only meaningful within this package.
-	Synthetic KobjectAction = 0xfeed
 )
 
 // lib/kobject_uevent.c
@@ -42,10 +38,6 @@ func (act KobjectAction) Valid() bool { return int(act) < len(kobject_actions) }
 
 // String returns the corresponding string sent over netlink.
 func (act KobjectAction) String() string {
-	if act == Synthetic {
-		return "synthetic"
-	}
-
 	if !act.Valid() {
 		return "unsupported kobject_action " + strconv.Itoa(int(act))
 	}
@@ -53,7 +45,7 @@ func (act KobjectAction) String() string {
 }
 
 func (act KobjectAction) AppendText(b []byte) ([]byte, error) {
-	if !act.Valid() && act != Synthetic {
+	if !act.Valid() {
 		return b, syscall.EINVAL
 	}
 	return append(b, act.String()...), nil

internal/uevent/action_test.go

@@ -29,15 +29,4 @@ func TestKobjectAction(t *testing.T) {
 			t.Errorf("String: %q, want %q", got, want)
 		}
 	})
-
-	adeT(t, "synthetic", uevent.Synthetic, "synthetic",
-		uevent.UnsupportedActionError("synthetic"), nil)
-
-	t.Run("validate synthetic", func(t *testing.T) {
-		t.Parallel()
-
-		if uevent.Synthetic.Valid() {
-			t.Errorf("Valid unexpectedly succeeded")
-		}
-	})
 }

internal/uevent/message_test.go

@@ -110,12 +110,6 @@ SEQNUM=780`},
 		}, "move", uevent.MissingHeaderError(
 			"move",
 		), syscall.EINVAL, "unsupported kobject_action 2989 event:"},
-
-		{"synthetic", uevent.Message{
-			Action: uevent.Synthetic,
-		}, "synthetic@\x00", uevent.UnsupportedActionError(
-			"synthetic",
-		), nil, "synthetic event:"},
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {

internal/uevent/sysfs.go

@@ -1,87 +0,0 @@
-package uevent
-
-import (
-	"bytes"
-	"errors"
-	"io/fs"
-	"log"
-	"path/filepath"
-	"unsafe"
-)
-
-// Enumerate scans sysfs and emits [Synthetic] events. It returns the first
-// error it encounters.
-//
-// The specified filesystem must present the sysfs root.
-func Enumerate(
-	sysfs fs.FS,
-	handleWalkErr func(error) error,
-	events chan<- *Message,
-) error {
-	if handleWalkErr == nil {
-		handleWalkErr = func(err error) error {
-			if errors.Is(err, fs.ErrNotExist) {
-				log.Println("enumerate", err)
-				return nil
-			}
-			return err
-		}
-	}
-
-	return fs.WalkDir(sysfs, "devices", func(
-		path string,
-		d fs.DirEntry,
-		err error,
-	) error {
-		if err != nil {
-			return handleWalkErr(err)
-		}
-
-		if d.IsDir() || d.Name() != "uevent" {
-			return nil
-		}
-
-		msg := Message{
-			Action: Synthetic,
-
-			// cleans path, appears to be compatible with kernel behaviour
-			DevPath: filepath.Dir(path),
-		}
-
-		var target string
-		if target, err = fs.ReadLink(
-			sysfs,
-			filepath.Join(msg.DevPath, "subsystem"),
-		); err != nil {
-			if err = handleWalkErr(err); err != nil {
-				return err
-			}
-		} else {
-			msg.Env = append(msg.Env, "SUBSYSTEM="+filepath.Base(target))
-		}
-
-		// read entire file: slicing does not copy
-		var env []byte
-		if env, err = fs.ReadFile(sysfs, path); err != nil {
-			return handleWalkErr(err)
-		}
-
-		for _, s := range bytes.Split(env, []byte{'\n'}) {
-			if len(s) == 0 {
-				continue
-			}
-			msg.Env = append(msg.Env, unsafe.String(unsafe.SliceData(s), len(s)))
-		}
-
-		if len(msg.Env) == 0 {
-			// this implies absent subsystem, its error is already handled
-			return nil
-		}
-
-		if msg.DevPath != "" && msg.DevPath[0] != '/' {
-			msg.DevPath = "/" + msg.DevPath
-		}
-		events <- &msg
-		return nil
-	})
-}

internal/uevent/sysfs_test.go

@@ -1,28 +0,0 @@
-package uevent_test
-
-import (
-	"os"
-	"sync"
-	"testing"
-
-	"hakurei.app/internal/uevent"
-)
-
-func TestEnumerate(t *testing.T) {
-	t.Parallel()
-
-	var wg sync.WaitGroup
-	defer wg.Wait()
-
-	events := make(chan *uevent.Message, 1<<10)
-	wg.Go(func() {
-		for msg := range events {
-			t.Log(msg)
-		}
-	})
-
-	if err := uevent.Enumerate(os.DirFS("/sys"), nil, events); err != nil {
-		t.Fatalf("Enumerate: error = %v", err)
-	}
-	close(events)
-}

internal/uevent/uevent.go

@@ -18,6 +18,19 @@ type (
 	Recoverable interface{ recoverable() }
 	// Nontrivial is satisfied by errors preferring a JSON encoding.
 	Nontrivial interface{ nontrivial() }
+
+	// NeedsColdboot is satisfied by errors indicating divergence of local state
+	// from the kernel, usually from lost uevent data.
+	NeedsColdboot interface {
+		Recoverable
+		coldboot()
+	}
+)
+
+const (
+	exclConsume = iota
+
+	_exclLen
 )
 
 // Conn represents a NETLINK_KOBJECT_UEVENT socket.
@@ -25,27 +38,27 @@ type Conn struct {
 	conn *netlink.Conn
 
 	// Whether currently between a call to enterExcl and exitExcl.
-	excl atomic.Bool
+	excl [_exclLen]atomic.Bool
 }
 
 // enterExcl must be called entering a critical section that interacts with conn.
-func (c *Conn) enterExcl() error {
-	if !c.excl.CompareAndSwap(false, true) {
+func (c *Conn) enterExcl(k int) error {
+	if !c.excl[k].CompareAndSwap(false, true) {
 		return syscall.EAGAIN
 	}
 	return nil
 }
 
 // exitExcl must be called exiting a critical section that interacts with conn.
-func (c *Conn) exitExcl() { c.excl.Store(false) }
+func (c *Conn) exitExcl(k int) { c.excl[k].Store(false) }
 
 // Close closes the underlying socket.
 func (c *Conn) Close() error { return c.conn.Close() }
 
 // Dial returns the address of a newly connected [Conn].
-func Dial() (*Conn, error) {
+func Dial(rcvbuf int64) (*Conn, error) {
 	// kernel group is hard coded in lib/kobject_uevent.c, undocumented
-	c, err := netlink.Dial(syscall.NETLINK_KOBJECT_UEVENT, 1)
+	c, err := netlink.Dial(syscall.NETLINK_KOBJECT_UEVENT, 1, rcvbuf)
 	if err != nil {
 		return nil, err
 	}
@@ -58,6 +71,18 @@ var (
 	ErrBadSocket = errors.New("unexpected socket address")
 )
 
+// ReceiveBufferError indicates one or more [Message] being lost due to the
+// socket receive buffer filling up. This is usually caused by epoll waking the
+// receiving program up too late.
+type ReceiveBufferError struct{ _ [0]*ReceiveBufferError }
+
+var _ NeedsColdboot = ReceiveBufferError{}
+
+func (ReceiveBufferError) recoverable()    {}
+func (ReceiveBufferError) coldboot()       {}
+func (ReceiveBufferError) Unwrap() error   { return syscall.ENOBUFS }
+func (e ReceiveBufferError) Error() string { return syscall.ENOBUFS.Error() }
+
 // BadPortError is returned by [Conn.Consume] upon receiving a message that did
 // not come from the kernel.
 type BadPortError syscall.SockaddrNetlink
@@ -70,36 +95,48 @@ func (e *BadPortError) Error() string {
 		" on NETLINK_KOBJECT_UEVENT"
 }
 
+// receiveEvent receives a single event and returns the address of its [Message].
+func (c *Conn) receiveEvent(ctx context.Context) (*Message, error) {
+	data, _, from, err := c.conn.Recvmsg(ctx, 0)
+	if err != nil {
+		if errors.Is(err, syscall.ENOBUFS) {
+			return nil, ReceiveBufferError{}
+		}
+		return nil, err
+	}
+
+	// lib/kobject_uevent.c:
+	// set portid 0 to inform userspace message comes from kernel
+	if v, ok := from.(*syscall.SockaddrNetlink); !ok {
+		return nil, ErrBadSocket
+	} else if v.Pid != 0 {
+		return nil, (*BadPortError)(v)
+
+	}
+
+	var msg Message
+	if err = msg.UnmarshalBinary(data); err != nil {
+		return nil, err
+	}
+	return &msg, err
+}
+
 // Consume continuously receives and parses events from the kernel. It returns
 // the first error it encounters.
 //
 // Callers must not restart event processing after a non-nil error that does not
 // satisfy [Recoverable] is returned.
 func (c *Conn) Consume(ctx context.Context, events chan<- *Message) error {
-	if err := c.enterExcl(); err != nil {
+	if err := c.enterExcl(exclConsume); err != nil {
 		return err
 	}
-	defer c.exitExcl()
+	defer c.exitExcl(exclConsume)
 
 	for {
-		data, from, err := c.conn.Recvfrom(ctx, 0)
+		msg, err := c.receiveEvent(ctx)
 		if err != nil {
 			return err
 		}
-
-		// lib/kobject_uevent.c:
-		// set portid 0 to inform userspace message comes from kernel
-		if v, ok := from.(*syscall.SockaddrNetlink); !ok {
-			return ErrBadSocket
-		} else if v.Pid != 0 {
-			return (*BadPortError)(v)
-
-		}
-
-		var msg Message
-		if err = msg.UnmarshalBinary(data); err != nil {
-			return err
-		}
-		events <- &msg
+		events <- msg
 	}
 }

internal/uevent/uevent_test.go

@@ -116,7 +116,7 @@ func adeB[V any, S interface {
 func TestDialConsume(t *testing.T) {
 	t.Parallel()
 
-	c, err := uevent.Dial()
+	c, err := uevent.Dial(0)
 	if err != nil {
 		t.Fatalf("Dial: error = %v", err)
 	}
@@ -127,7 +127,7 @@ func TestDialConsume(t *testing.T) {
 	})
 
 	// check kernel-assigned port id
-	c0, err0 := uevent.Dial()
+	c0, err0 := uevent.Dial(0)
 	if err0 != nil {
 		t.Fatalf("Dial: error = %v", err)
 	}