Implement F_GETOWN_EX and F_SETOWN_EX.

Some versions of glibc will convert F_GETOWN fcntl(2) calls into F_GETOWN_EX in
some cases.

PiperOrigin-RevId: 284089373

5 files changed, 255 insertions, 22 deletions

diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go
index f78315ebf..6663a199c 100644
--- a/pkg/abi/linux/fcntl.go
+++ b/pkg/abi/linux/fcntl.go
@@ -16,15 +16,17 @@ package linux
 
 // Commands from linux/fcntl.h.
 const (
-	F_DUPFD         = 0x0
-	F_GETFD         = 0x1
-	F_SETFD         = 0x2
-	F_GETFL         = 0x3
-	F_SETFL         = 0x4
-	F_SETLK         = 0x6
-	F_SETLKW        = 0x7
-	F_SETOWN        = 0x8
-	F_GETOWN        = 0x9
+	F_DUPFD         = 0
+	F_GETFD         = 1
+	F_SETFD         = 2
+	F_GETFL         = 3
+	F_SETFL         = 4
+	F_SETLK         = 6
+	F_SETLKW        = 7
+	F_SETOWN        = 8
+	F_GETOWN        = 9
+	F_SETOWN_EX     = 15
+	F_GETOWN_EX     = 16
 	F_DUPFD_CLOEXEC = 1024 + 6
 	F_SETPIPE_SZ    = 1024 + 7
 	F_GETPIPE_SZ    = 1024 + 8
@@ -32,9 +34,9 @@ const (
 
 // Commands for F_SETLK.
 const (
-	F_RDLCK = 0x0
-	F_WRLCK = 0x1
-	F_UNLCK = 0x2
+	F_RDLCK = 0
+	F_WRLCK = 1
+	F_UNLCK = 2
 )
 
 // Flags for fcntl.
@@ -42,7 +44,7 @@ const (
 	FD_CLOEXEC = 00000001
 )
 
-// Lock structure for F_SETLK.
+// Flock is the lock structure for F_SETLK.
 type Flock struct {
 	Type   int16
 	Whence int16
@@ -52,3 +54,16 @@ type Flock struct {
 	Pid    int32
 	_      [4]byte
 }
+
+// Flags for F_SETOWN_EX and F_GETOWN_EX.
+const (
+	F_OWNER_TID  = 0
+	F_OWNER_PID  = 1
+	F_OWNER_PGRP = 2
+)
+
+// FOwnerEx is the owner structure for F_SETOWN_EX and F_GETOWN_EX.
+type FOwnerEx struct {
+	Type int32
+	PID  int32
+}
diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go
index 3b9181002..9bc2445a5 100644
--- a/pkg/sentry/syscalls/linux/sys_file.go
+++ b/pkg/sentry/syscalls/linux/sys_file.go
@@ -840,25 +840,42 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC
 	return uintptr(newfd), nil, nil
 }
 
-func fGetOwn(t *kernel.Task, file *fs.File) int32 {
+func fGetOwnEx(t *kernel.Task, file *fs.File) linux.FOwnerEx {
 	ma := file.Async(nil)
 	if ma == nil {
-		return 0
+		return linux.FOwnerEx{}
 	}
 	a := ma.(*fasync.FileAsync)
 	ot, otg, opg := a.Owner()
 	switch {
 	case ot != nil:
-		return int32(t.PIDNamespace().IDOfTask(ot))
+		return linux.FOwnerEx{
+			Type: linux.F_OWNER_TID,
+			PID:  int32(t.PIDNamespace().IDOfTask(ot)),
+		}
 	case otg != nil:
-		return int32(t.PIDNamespace().IDOfThreadGroup(otg))
+		return linux.FOwnerEx{
+			Type: linux.F_OWNER_PID,
+			PID:  int32(t.PIDNamespace().IDOfThreadGroup(otg)),
+		}
 	case opg != nil:
-		return int32(-t.PIDNamespace().IDOfProcessGroup(opg))
+		return linux.FOwnerEx{
+			Type: linux.F_OWNER_PGRP,
+			PID:  int32(t.PIDNamespace().IDOfProcessGroup(opg)),
+		}
 	default:
-		return 0
+		return linux.FOwnerEx{}
 	}
 }
 
+func fGetOwn(t *kernel.Task, file *fs.File) int32 {
+	owner := fGetOwnEx(t, file)
+	if owner.Type == linux.F_OWNER_PGRP {
+		return -owner.PID
+	}
+	return owner.PID
+}
+
 // fSetOwn sets the file's owner with the semantics of F_SETOWN in Linux.
 //
 // If who is positive, it represents a PID. If negative, it represents a PGID.
@@ -901,11 +918,13 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 		t.FDTable().SetFlags(fd, kernel.FDFlags{
 			CloseOnExec: flags&linux.FD_CLOEXEC != 0,
 		})
+		return 0, nil, nil
 	case linux.F_GETFL:
 		return uintptr(file.Flags().ToLinux()), nil, nil
 	case linux.F_SETFL:
 		flags := uint(args[2].Uint())
 		file.SetFlags(linuxToFlags(flags).Settable())
+		return 0, nil, nil
 	case linux.F_SETLK, linux.F_SETLKW:
 		// In Linux the file system can choose to provide lock operations for an inode.
 		// Normally pipe and socket types lack lock operations. We diverge and use a heavy
@@ -1008,6 +1027,44 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 	case linux.F_SETOWN:
 		fSetOwn(t, file, args[2].Int())
 		return 0, nil, nil
+	case linux.F_GETOWN_EX:
+		addr := args[2].Pointer()
+		owner := fGetOwnEx(t, file)
+		_, err := t.CopyOut(addr, &owner)
+		return 0, nil, err
+	case linux.F_SETOWN_EX:
+		addr := args[2].Pointer()
+		var owner linux.FOwnerEx
+		n, err := t.CopyIn(addr, &owner)
+		if err != nil {
+			return 0, nil, err
+		}
+		a := file.Async(fasync.New).(*fasync.FileAsync)
+		switch owner.Type {
+		case linux.F_OWNER_TID:
+			task := t.PIDNamespace().TaskWithID(kernel.ThreadID(owner.PID))
+			if task == nil {
+				return 0, nil, syserror.ESRCH
+			}
+			a.SetOwnerTask(t, task)
+			return uintptr(n), nil, nil
+		case linux.F_OWNER_PID:
+			tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID))
+			if tg == nil {
+				return 0, nil, syserror.ESRCH
+			}
+			a.SetOwnerThreadGroup(t, tg)
+			return uintptr(n), nil, nil
+		case linux.F_OWNER_PGRP:
+			pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID))
+			if pg == nil {
+				return 0, nil, syserror.ESRCH
+			}
+			a.SetOwnerProcessGroup(t, pg)
+			return uintptr(n), nil, nil
+		default:
+			return 0, nil, syserror.EINVAL
+		}
 	case linux.F_GET_SEALS:
 		val, err := tmpfs.GetSeals(file.Dirent.Inode)
 		return uintptr(val), nil, err
@@ -1035,7 +1092,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 		// Everything else is not yet supported.
 		return 0, nil, syserror.EINVAL
 	}
-	return 0, nil, nil
 }
 
 const (
diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD
index 7ce2e6270..61f310db9 100644
--- a/test/syscalls/linux/BUILD
+++ b/test/syscalls/linux/BUILD
@@ -753,6 +753,7 @@ cc_binary(
         "//test/util:eventfd_util",
         "//test/util:multiprocess_util",
         "//test/util:posix_error",
+        "//test/util:save_util",
         "//test/util:temp_path",
         "//test/util:test_util",
         "//test/util:timer_util",
diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc
index 8a45be12a..4f3aa81d6 100644
--- a/test/syscalls/linux/fcntl.cc
+++ b/test/syscalls/linux/fcntl.cc
@@ -14,6 +14,7 @@
 
 #include <fcntl.h>
 #include <signal.h>
+#include <sys/types.h>
 #include <syscall.h>
 #include <unistd.h>
 
@@ -32,6 +33,7 @@
 #include "test/util/eventfd_util.h"
 #include "test/util/multiprocess_util.h"
 #include "test/util/posix_error.h"
+#include "test/util/save_util.h"
 #include "test/util/temp_path.h"
 #include "test/util/test_util.h"
 #include "test/util/timer_util.h"
@@ -910,8 +912,166 @@ TEST(FcntlTest, GetOwn) {
   FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
       Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
 
-  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN),
+  EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), 0);
+  MaybeSave();
+}
+
+TEST(FcntlTest, GetOwnEx) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &owner),
+              SyscallSucceedsWithValue(0));
+}
+
+TEST(FcntlTest, SetOwnExInvalidType) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = __pid_type(-1);
+  EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallFailsWithErrno(EINVAL));
+}
+
+TEST(FcntlTest, SetOwnExInvalidTid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_TID;
+  owner.pid = -1;
+
+  EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallFailsWithErrno(ESRCH));
+}
+
+TEST(FcntlTest, SetOwnExInvalidPid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_PID;
+  owner.pid = -1;
+
+  EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallFailsWithErrno(ESRCH));
+}
+
+TEST(FcntlTest, SetOwnExInvalidPgrp) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_PGRP;
+  owner.pid = -1;
+
+  EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallFailsWithErrno(ESRCH));
+}
+
+TEST(FcntlTest, SetOwnExTid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_TID;
+  EXPECT_THAT(owner.pid = syscall(__NR_gettid), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallSucceeds());
+
+  EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid);
+  MaybeSave();
+}
+
+TEST(FcntlTest, SetOwnExPid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_PID;
+  EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallSucceeds());
+
+  EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid);
+  MaybeSave();
+}
+
+TEST(FcntlTest, SetOwnExPgrp) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex owner = {};
+  owner.type = F_OWNER_PGRP;
+  EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner),
+              SyscallSucceeds());
+
+  // NOTE(igudger): I don't understand why, but this is flaky on Linux.
+  // GetOwnExPgrp (below) does not have this issue.
+  SKIP_IF(!IsRunningOnGvisor());
+
+  EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), -owner.pid);
+  MaybeSave();
+}
+
+TEST(FcntlTest, GetOwnExTid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex set_owner = {};
+  set_owner.type = F_OWNER_TID;
+  EXPECT_THAT(set_owner.pid = syscall(__NR_gettid), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner),
+              SyscallSucceeds());
+
+  f_owner_ex got_owner = {};
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner),
+              SyscallSucceedsWithValue(0));
+  EXPECT_EQ(got_owner.type, set_owner.type);
+  EXPECT_EQ(got_owner.pid, set_owner.pid);
+}
+
+TEST(FcntlTest, GetOwnExPid) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex set_owner = {};
+  set_owner.type = F_OWNER_PID;
+  EXPECT_THAT(set_owner.pid = getpid(), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner),
+              SyscallSucceeds());
+
+  f_owner_ex got_owner = {};
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner),
+              SyscallSucceedsWithValue(0));
+  EXPECT_EQ(got_owner.type, set_owner.type);
+  EXPECT_EQ(got_owner.pid, set_owner.pid);
+}
+
+TEST(FcntlTest, GetOwnExPgrp) {
+  FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0));
+
+  f_owner_ex set_owner = {};
+  set_owner.type = F_OWNER_PGRP;
+  EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds());
+
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner),
+              SyscallSucceeds());
+
+  f_owner_ex got_owner = {};
+  ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner),
               SyscallSucceedsWithValue(0));
+  EXPECT_EQ(got_owner.type, set_owner.type);
+  EXPECT_EQ(got_owner.pid, set_owner.pid);
 }
 
 }  // namespace
diff --git a/test/syscalls/linux/ioctl.cc b/test/syscalls/linux/ioctl.cc
index c4f8bff08..b0a07a064 100644
--- a/test/syscalls/linux/ioctl.cc
+++ b/test/syscalls/linux/ioctl.cc
@@ -215,7 +215,8 @@ TEST_F(IoctlTest, FIOASYNCSelfTarget2) {
   auto mask_cleanup =
       ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_UNBLOCK, SIGIO));
 
-  pid_t pid = getpid();
+  pid_t pid = -1;
+  EXPECT_THAT(pid = getpid(), SyscallSucceeds());
   EXPECT_THAT(ioctl(pair->second_fd(), FIOSETOWN, &pid), SyscallSucceeds());
 
   int set = 1;