Implement /proc/net/unix.

PiperOrigin-RevId: 232948478
Change-Id: Ib830121e5e79afaf5d38d17aeef5a1ef97913d23

6 files changed, 195 insertions, 6 deletions

diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD
index f6bc90634..666b0ab3a 100644
--- a/pkg/sentry/fs/proc/BUILD
+++ b/pkg/sentry/fs/proc/BUILD
@@ -30,6 +30,7 @@ go_library(
     visibility = ["//pkg/sentry:internal"],
     deps = [
         "//pkg/abi/linux",
+        "//pkg/log",
         "//pkg/sentry/context",
         "//pkg/sentry/fs",
         "//pkg/sentry/fs/fsutil",
@@ -43,6 +44,8 @@ go_library(
         "//pkg/sentry/kernel/time",
         "//pkg/sentry/mm",
         "//pkg/sentry/socket/rpcinet",
+        "//pkg/sentry/socket/unix",
+        "//pkg/sentry/socket/unix/transport",
         "//pkg/sentry/usage",
         "//pkg/sentry/usermem",
         "//pkg/syserror",
diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go
index 219eea7f8..55a958f9e 100644
--- a/pkg/sentry/fs/proc/net.go
+++ b/pkg/sentry/fs/proc/net.go
@@ -15,19 +15,24 @@
 package proc
 
 import (
+	"bytes"
 	"fmt"
 	"time"
 
 	"gvisor.googlesource.com/gvisor/pkg/abi/linux"
+	"gvisor.googlesource.com/gvisor/pkg/log"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/context"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/fs"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/inet"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/kernel"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix"
+	"gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport"
 )
 
 // newNet creates a new proc net entry.
-func (p *proc) newNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode {
+func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode {
 	var contents map[string]*fs.Inode
 	if s := p.k.NetworkStack(); s != nil {
 		contents = map[string]*fs.Inode{
@@ -52,6 +57,8 @@ func (p *proc) newNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode {
 			"tcp":    newStaticProcInode(ctx, msrc, []byte("  sl  local_address rem_address   st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode")),
 
 			"udp": newStaticProcInode(ctx, msrc, []byte("  sl  local_address rem_address   st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode ref pointer drops")),
+
+			"unix": seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc),
 		}
 
 		if s.SupportsIPv6() {
@@ -182,3 +189,120 @@ func (n *netDev) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se
 
 	return data, 0
 }
+
+// netUnix implements seqfile.SeqSource for /proc/net/unix.
+//
+// +stateify savable
+type netUnix struct {
+	k *kernel.Kernel
+}
+
+// NeedsUpdate implements seqfile.SeqSource.NeedsUpdate.
+func (*netUnix) NeedsUpdate(generation int64) bool {
+	return true
+}
+
+// ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData.
+func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) {
+	if h != nil {
+		return []seqfile.SeqData{}, 0
+	}
+
+	var buf bytes.Buffer
+	// Header
+	fmt.Fprintf(&buf, "Num       RefCount Protocol Flags    Type St Inode Path\n")
+
+	// Entries
+	for _, sref := range n.k.ListSockets(linux.AF_UNIX) {
+		s := sref.Get()
+		if s == nil {
+			log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", sref)
+			continue
+		}
+		sfile := s.(*fs.File)
+		sops, ok := sfile.FileOperations.(*unix.SocketOperations)
+		if !ok {
+			panic(fmt.Sprintf("Found non-unix socket file in unix socket table: %+v", sfile))
+		}
+
+		addr, err := sops.Endpoint().GetLocalAddress()
+		if err != nil {
+			log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err)
+			addr.Addr = "<unknown>"
+		}
+
+		sockFlags := 0
+		if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok {
+			if ce.Listening() {
+				// For unix domain sockets, linux reports a single flag
+				// value if the socket is listening, of __SO_ACCEPTCON.
+				sockFlags = linux.SO_ACCEPTCON
+			}
+		}
+
+		var sockState int
+		switch sops.Endpoint().Type() {
+		case linux.SOCK_DGRAM:
+			sockState = linux.SS_CONNECTING
+			// Unlike Linux, we don't have unbound connection-less sockets,
+			// so no SS_DISCONNECTING.
+
+		case linux.SOCK_SEQPACKET:
+			fallthrough
+		case linux.SOCK_STREAM:
+			// Connectioned.
+			if sops.Endpoint().(transport.ConnectingEndpoint).Connected() {
+				sockState = linux.SS_CONNECTED
+			} else {
+				sockState = linux.SS_UNCONNECTED
+			}
+		}
+
+		// In the socket entry below, the value for the 'Num' field requires
+		// some consideration. Linux prints the address to the struct
+		// unix_sock representing a socket in the kernel, but may redact the
+		// value for unprivileged users depending on the kptr_restrict
+		// sysctl.
+		//
+		// One use for this field is to allow a privileged user to
+		// introspect into the kernel memory to determine information about
+		// a socket not available through procfs, such as the socket's peer.
+		//
+		// On gvisor, returning a pointer to our internal structures would
+		// be pointless, as it wouldn't match the memory layout for struct
+		// unix_sock, making introspection difficult. We could populate a
+		// struct unix_sock with the appropriate data, but even that
+		// requires consideration for which kernel version to emulate, as
+		// the definition of this struct changes over time.
+		//
+		// For now, we always redact this pointer.
+		fmt.Fprintf(&buf, "%#016p: %08X %08X %08X %04X %02X %5d",
+			(*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct.
+			sfile.ReadRefs()-1,            // RefCount, don't count our own ref.
+			0,                             // Protocol, always 0 for UDS.
+			sockFlags,                     // Flags.
+			sops.Endpoint().Type(),        // Type.
+			sockState,                     // State.
+			sfile.InodeID(),               // Inode.
+		)
+
+		// Path
+		if len(addr.Addr) != 0 {
+			if addr.Addr[0] == 0 {
+				// Abstract path.
+				fmt.Fprintf(&buf, " @%s", string(addr.Addr[1:]))
+			} else {
+				fmt.Fprintf(&buf, " %s", string(addr.Addr))
+			}
+		}
+		fmt.Fprintf(&buf, "\n")
+
+		sfile.DecRef()
+	}
+
+	data := []seqfile.SeqData{{
+		Buf:    buf.Bytes(),
+		Handle: (*netUnix)(nil),
+	}}
+	return data, 0
+}
diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go
index be04f94af..88018e707 100644
--- a/pkg/sentry/fs/proc/proc.go
+++ b/pkg/sentry/fs/proc/proc.go
@@ -85,7 +85,7 @@ func New(ctx context.Context, msrc *fs.MountSource) (*fs.Inode, error) {
 	if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok {
 		p.AddChild(ctx, "net", newRPCInetProcNet(ctx, msrc))
 	} else {
-		p.AddChild(ctx, "net", p.newNetDir(ctx, msrc))
+		p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc))
 	}
 
 	return newProcInode(p, msrc, fs.SpecialDirectory, nil), nil
diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go
index 43e9823cb..e7e5ff777 100644
--- a/pkg/sentry/kernel/kernel.go
+++ b/pkg/sentry/kernel/kernel.go
@@ -43,6 +43,7 @@ import (
 	"gvisor.googlesource.com/gvisor/pkg/cpuid"
 	"gvisor.googlesource.com/gvisor/pkg/eventchannel"
 	"gvisor.googlesource.com/gvisor/pkg/log"
+	"gvisor.googlesource.com/gvisor/pkg/refs"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/arch"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/context"
 	"gvisor.googlesource.com/gvisor/pkg/sentry/fs"
@@ -164,7 +165,7 @@ type Kernel struct {
 	// nextInotifyCookie is a monotonically increasing counter used for
 	// generating unique inotify event cookies.
 	//
-	// nextInotifyCookie is mutable, and is accesed using atomic memory
+	// nextInotifyCookie is mutable, and is accessed using atomic memory
 	// operations.
 	nextInotifyCookie uint32
 
@@ -177,6 +178,10 @@ type Kernel struct {
 
 	// danglingEndpoints is used to save / restore tcpip.DanglingEndpoints.
 	danglingEndpoints struct{} `state:".([]tcpip.Endpoint)"`
+
+	// socketTable is used to track all sockets on the system. Protected by
+	// extMu.
+	socketTable map[int]map[*refs.WeakRef]struct{}
 }
 
 // InitKernelArgs holds arguments to Init.
@@ -266,6 +271,7 @@ func (k *Kernel) Init(args InitKernelArgs) error {
 	k.monotonicClock = &timekeeperClock{tk: args.Timekeeper, c: sentrytime.Monotonic}
 	k.futexes = futex.NewManager()
 	k.netlinkPorts = port.New()
+	k.socketTable = make(map[int]map[*refs.WeakRef]struct{})
 
 	return nil
 }
@@ -1051,6 +1057,56 @@ func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) {
 	})
 }
 
+// socketEntry represents a socket recorded in Kernel.socketTable. It implements
+// refs.WeakRefUser for sockets stored in the socket table.
+//
+// +stateify savable
+type socketEntry struct {
+	k      *Kernel
+	sock   *refs.WeakRef
+	family int
+}
+
+// WeakRefGone implements refs.WeakRefUser.WeakRefGone.
+func (s *socketEntry) WeakRefGone() {
+	s.k.extMu.Lock()
+	// k.socketTable is guaranteed to point to a valid socket table for s.family
+	// at this point, since we made sure of the fact when we created this
+	// socketEntry, and we never delete socket tables.
+	delete(s.k.socketTable[s.family], s.sock)
+	s.k.extMu.Unlock()
+}
+
+// RecordSocket adds a socket to the system-wide socket table for tracking.
+//
+// Precondition: Caller must hold a reference to sock.
+func (k *Kernel) RecordSocket(sock *fs.File, family int) {
+	k.extMu.Lock()
+	table, ok := k.socketTable[family]
+	if !ok {
+		table = make(map[*refs.WeakRef]struct{})
+		k.socketTable[family] = table
+	}
+	se := socketEntry{k: k, family: family}
+	se.sock = refs.NewWeakRef(sock, &se)
+	table[se.sock] = struct{}{}
+	k.extMu.Unlock()
+}
+
+// ListSockets returns a snapshot of all sockets of a given family.
+func (k *Kernel) ListSockets(family int) []*refs.WeakRef {
+	k.extMu.Lock()
+	socks := []*refs.WeakRef{}
+	if table, ok := k.socketTable[family]; ok {
+		socks = make([]*refs.WeakRef, 0, len(table))
+		for s, _ := range table {
+			socks = append(socks, s)
+		}
+	}
+	k.extMu.Unlock()
+	return socks
+}
+
 type supervisorContext struct {
 	context.NoopSleeper
 	log.Logger
diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go
index e28d2c4fa..5ab423f3c 100644
--- a/pkg/sentry/socket/socket.go
+++ b/pkg/sentry/socket/socket.go
@@ -147,6 +147,7 @@ func New(t *kernel.Task, family int, stype transport.SockType, protocol int) (*f
 			return nil, err
 		}
 		if s != nil {
+			t.Kernel().RecordSocket(s, family)
 			return s, nil
 		}
 	}
@@ -163,12 +164,15 @@ func Pair(t *kernel.Task, family int, stype transport.SockType, protocol int) (*
 	}
 
 	for _, p := range providers {
-		s, t, err := p.Pair(t, stype, protocol)
+		s1, s2, err := p.Pair(t, stype, protocol)
 		if err != nil {
 			return nil, nil, err
 		}
-		if s != nil && t != nil {
-			return s, t, nil
+		if s1 != nil && s2 != nil {
+			k := t.Kernel()
+			k.RecordSocket(s1, family)
+			k.RecordSocket(s2, family)
+			return s1, s2, nil
 		}
 	}
 
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 19258e692..c857a0f33 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -219,6 +219,8 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int,
 		return 0, nil, 0, syserr.FromError(e)
 	}
 
+	t.Kernel().RecordSocket(ns, linux.AF_UNIX)
+
 	return fd, addr, addrLen, nil
 }