Move /proc/net to /proc/PID/net, and make /proc/net -> /proc/self/net.

Issue #1833

PiperOrigin-RevId: 299998105

9 files changed, 68 insertions, 36 deletions

diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go
index 95d5817ff..bd18177d4 100644
--- a/pkg/sentry/fs/proc/net.go
+++ b/pkg/sentry/fs/proc/net.go
@@ -40,47 +40,48 @@ import (
 
 // LINT.IfChange
 
-// newNet creates a new proc net entry.
-func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode {
+// newNetDir creates a new proc net entry.
+func newNetDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode {
+	k := t.Kernel()
+
 	var contents map[string]*fs.Inode
-	// TODO(gvisor.dev/issue/1833): Support for using the network stack in the
-	// network namespace of the calling process. We should make this per-process,
-	// a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net.
-	if s := p.k.RootNetworkNamespace().Stack(); s != nil {
+	if s := t.NetworkNamespace().Stack(); s != nil {
+		// TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task
+		// network namespace.
 		contents = map[string]*fs.Inode{
-			"dev":  seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc),
-			"snmp": seqfile.NewSeqFileInode(ctx, &netSnmp{s: s}, msrc),
+			"dev":  seqfile.NewSeqFileInode(t, &netDev{s: s}, msrc),
+			"snmp": seqfile.NewSeqFileInode(t, &netSnmp{s: s}, msrc),
 
 			// The following files are simple stubs until they are
 			// implemented in netstack, if the file contains a
 			// header the stub is just the header otherwise it is
 			// an empty file.
-			"arp": newStaticProcInode(ctx, msrc, []byte("IP address       HW type     Flags       HW address            Mask     Device\n")),
+			"arp": newStaticProcInode(t, msrc, []byte("IP address       HW type     Flags       HW address            Mask     Device\n")),
 
-			"netlink":   newStaticProcInode(ctx, msrc, []byte("sk       Eth Pid    Groups   Rmem     Wmem     Dump     Locks     Drops     Inode\n")),
-			"netstat":   newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n")),
-			"packet":    newStaticProcInode(ctx, msrc, []byte("sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n")),
-			"protocols": newStaticProcInode(ctx, msrc, []byte("protocol  size sockets  memory press maxhdr  slab module     cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n")),
+			"netlink":   newStaticProcInode(t, msrc, []byte("sk       Eth Pid    Groups   Rmem     Wmem     Dump     Locks     Drops     Inode\n")),
+			"netstat":   newStaticProcInode(t, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n")),
+			"packet":    newStaticProcInode(t, msrc, []byte("sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n")),
+			"protocols": newStaticProcInode(t, msrc, []byte("protocol  size sockets  memory press maxhdr  slab module     cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n")),
 			// Linux sets psched values to: nsec per usec, psched
 			// tick in ns, 1000000, high res timer ticks per sec
 			// (ClockGetres returns 1ns resolution).
-			"psched": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))),
-			"ptype":  newStaticProcInode(ctx, msrc, []byte("Type Device      Function\n")),
-			"route":  seqfile.NewSeqFileInode(ctx, &netRoute{s: s}, msrc),
-			"tcp":    seqfile.NewSeqFileInode(ctx, &netTCP{k: k}, msrc),
-			"udp":    seqfile.NewSeqFileInode(ctx, &netUDP{k: k}, msrc),
-			"unix":   seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc),
+			"psched": newStaticProcInode(t, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))),
+			"ptype":  newStaticProcInode(t, msrc, []byte("Type Device      Function\n")),
+			"route":  seqfile.NewSeqFileInode(t, &netRoute{s: s}, msrc),
+			"tcp":    seqfile.NewSeqFileInode(t, &netTCP{k: k}, msrc),
+			"udp":    seqfile.NewSeqFileInode(t, &netUDP{k: k}, msrc),
+			"unix":   seqfile.NewSeqFileInode(t, &netUnix{k: k}, msrc),
 		}
 
 		if s.SupportsIPv6() {
-			contents["if_inet6"] = seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc)
-			contents["ipv6_route"] = newStaticProcInode(ctx, msrc, []byte(""))
-			contents["tcp6"] = seqfile.NewSeqFileInode(ctx, &netTCP6{k: k}, msrc)
-			contents["udp6"] = newStaticProcInode(ctx, msrc, []byte("  sl  local_address                         remote_address                        st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode\n"))
+			contents["if_inet6"] = seqfile.NewSeqFileInode(t, &ifinet6{s: s}, msrc)
+			contents["ipv6_route"] = newStaticProcInode(t, msrc, []byte(""))
+			contents["tcp6"] = seqfile.NewSeqFileInode(t, &netTCP6{k: k}, msrc)
+			contents["udp6"] = newStaticProcInode(t, msrc, []byte("  sl  local_address                         remote_address                        st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode\n"))
 		}
 	}
-	d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555))
-	return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil)
+	d := ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0555))
+	return newProcInode(t, d, msrc, fs.SpecialDirectory, t)
 }
 
 // ifinet6 implements seqfile.SeqSource for /proc/net/if_inet6.
@@ -837,4 +838,4 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se
 	return data, 0
 }
 
-// LINT.ThenChange(../../fsimpl/proc/tasks_net.go)
+// LINT.ThenChange(../../fsimpl/proc/task_net.go)
diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go
index c8abb5052..c659224a7 100644
--- a/pkg/sentry/fs/proc/proc.go
+++ b/pkg/sentry/fs/proc/proc.go
@@ -70,6 +70,7 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string
 		"loadavg":     seqfile.NewSeqFileInode(ctx, &loadavgData{}, msrc),
 		"meminfo":     seqfile.NewSeqFileInode(ctx, &meminfoData{k}, msrc),
 		"mounts":      newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/mounts"), msrc, fs.Symlink, nil),
+		"net":         newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/net"), msrc, fs.Symlink, nil),
 		"self":        newSelf(ctx, pidns, msrc),
 		"stat":        seqfile.NewSeqFileInode(ctx, &statData{k}, msrc),
 		"thread-self": newThreadSelf(ctx, pidns, msrc),
@@ -86,7 +87,6 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string
 	}
 
 	// Add more contents that need proc to be initialized.
-	p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc))
 	p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc))
 
 	return newProcInode(ctx, p, msrc, fs.SpecialDirectory, nil), nil
diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go
index 4e9b0fc00..03cc788c8 100644
--- a/pkg/sentry/fs/proc/task.go
+++ b/pkg/sentry/fs/proc/task.go
@@ -84,6 +84,7 @@ func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bo
 		"maps":          newMaps(t, msrc),
 		"mountinfo":     seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
 		"mounts":        seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
+		"net":           newNetDir(t, msrc),
 		"ns":            newNamespaceDir(t, msrc),
 		"oom_score":     newOOMScore(t, msrc),
 		"oom_score_adj": newOOMScoreAdj(t, msrc),
diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD
index a83245866..bb609a305 100644
--- a/pkg/sentry/fsimpl/proc/BUILD
+++ b/pkg/sentry/fsimpl/proc/BUILD
@@ -9,9 +9,9 @@ go_library(
         "subtasks.go",
         "task.go",
         "task_files.go",
+        "task_net.go",
         "tasks.go",
         "tasks_files.go",
-        "tasks_net.go",
         "tasks_sys.go",
     ],
     visibility = ["//pkg/sentry:internal"],
diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go
index c0d643f51..493acbd1b 100644
--- a/pkg/sentry/fsimpl/proc/task.go
+++ b/pkg/sentry/fsimpl/proc/task.go
@@ -57,6 +57,7 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames
 		"maps":    newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}),
 		//"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc),
 		//"mounts":    seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc),
+		"net": newTaskNetDir(task, inoGen),
 		"ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{
 			"net":  newNamespaceSymlink(task, inoGen.NextIno(), "net"),
 			"pid":  newNamespaceSymlink(task, inoGen.NextIno(), "pid"),
diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/task_net.go
index d4e1812d8..373a7b17d 100644
--- a/pkg/sentry/fsimpl/proc/tasks_net.go
+++ b/pkg/sentry/fsimpl/proc/task_net.go
@@ -37,12 +37,13 @@ import (
 	"gvisor.dev/gvisor/pkg/usermem"
 )
 
-func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry {
+func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry {
+	k := task.Kernel()
+	pidns := task.PIDNamespace()
+	root := auth.NewRootCredentials(pidns.UserNamespace())
+
 	var contents map[string]*kernfs.Dentry
-	// TODO(gvisor.dev/issue/1833): Support for using the network stack in the
-	// network namespace of the calling process. We should make this per-process,
-	// a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net.
-	if stack := k.RootNetworkNamespace().Stack(); stack != nil {
+	if stack := task.NetworkNamespace().Stack(); stack != nil {
 		const (
 			arp       = "IP address       HW type     Flags       HW address            Mask     Device\n"
 			netlink   = "sk       Eth Pid    Groups   Rmem     Wmem     Dump     Locks     Drops     Inode\n"
@@ -53,6 +54,8 @@ func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k
 		)
 		psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond))
 
+		// TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task
+		// network namespace.
 		contents = map[string]*kernfs.Dentry{
 			"dev":  newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}),
 			"snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}),
@@ -84,7 +87,7 @@ func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k
 		}
 	}
 
-	return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents)
+	return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents)
 }
 
 // ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6.
diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go
index b1e39c82f..d203cebd4 100644
--- a/pkg/sentry/fsimpl/proc/tasks.go
+++ b/pkg/sentry/fsimpl/proc/tasks.go
@@ -72,7 +72,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames
 		"sys":     newSysDir(root, inoGen, k),
 		"meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}),
 		"mounts":  kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"),
-		"net":     newNetDir(root, inoGen, k),
+		"net":     kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"),
 		"stat":    newDentry(root, inoGen.NextIno(), 0444, &statData{k: k}),
 		"uptime":  newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}),
 		"version": newDentry(root, inoGen.NextIno(), 0444, &versionData{k: k}),
diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go
index 0eb401619..1bb9430c0 100644
--- a/pkg/sentry/fsimpl/proc/tasks_test.go
+++ b/pkg/sentry/fsimpl/proc/tasks_test.go
@@ -50,7 +50,7 @@ var (
 		"loadavg":     linux.DT_REG,
 		"meminfo":     linux.DT_REG,
 		"mounts":      linux.DT_LNK,
-		"net":         linux.DT_DIR,
+		"net":         linux.DT_LNK,
 		"self":        linux.DT_LNK,
 		"stat":        linux.DT_REG,
 		"sys":         linux.DT_DIR,
@@ -71,6 +71,7 @@ var (
 		"gid_map":       linux.DT_REG,
 		"io":            linux.DT_REG,
 		"maps":          linux.DT_REG,
+		"net":           linux.DT_DIR,
 		"ns":            linux.DT_DIR,
 		"oom_score":     linux.DT_REG,
 		"oom_score_adj": linux.DT_REG,
diff --git a/test/syscalls/linux/proc_net.cc b/test/syscalls/linux/proc_net.cc
index 3a611a86f..05c952b99 100644
--- a/test/syscalls/linux/proc_net.cc
+++ b/test/syscalls/linux/proc_net.cc
@@ -33,6 +33,31 @@ namespace gvisor {
 namespace testing {
 namespace {
 
+constexpr const char kProcNet[] = "/proc/net";
+
+TEST(ProcNetSymlinkTarget, FileMode) {
+  struct stat s;
+  ASSERT_THAT(stat(kProcNet, &s), SyscallSucceeds());
+  EXPECT_EQ(s.st_mode & S_IFMT, S_IFDIR);
+  EXPECT_EQ(s.st_mode & 0777, 0555);
+}
+
+TEST(ProcNetSymlink, FileMode) {
+  struct stat s;
+  ASSERT_THAT(lstat(kProcNet, &s), SyscallSucceeds());
+  EXPECT_EQ(s.st_mode & S_IFMT, S_IFLNK);
+  EXPECT_EQ(s.st_mode & 0777, 0777);
+}
+
+TEST(ProcNetSymlink, Contents) {
+  char buf[40] = {};
+  int n = readlink(kProcNet, buf, sizeof(buf));
+  ASSERT_THAT(n, SyscallSucceeds());
+
+  buf[n] = 0;
+  EXPECT_STREQ(buf, "self/net");
+}
+
 TEST(ProcNetIfInet6, Format) {
   auto ifinet6 = ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/if_inet6"));
   EXPECT_THAT(ifinet6,