Signalfd support

Note that the exact semantics for these signalfds are slightly different from
Linux. These signalfds are bound to the process at creation time. Reads, polls,
etc. are all associated with signals directed at that task. In Linux, all
signalfd operations are associated with current, regardless of where the
signalfd originated.

In practice, this should not be an issue given how signalfds are used. In order
to fix this however, we will need to plumb the context through all the event
APIs. This gets complicated really quickly, because the waiter APIs are all
netstack-specific, and not generally exposed to the context.  Probably not
worthwhile fixing immediately.

PiperOrigin-RevId: 269901749

3 files changed, 80 insertions, 2 deletions

diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD
index 33a40b9c6..e76ee27d2 100644
--- a/pkg/sentry/syscalls/linux/BUILD
+++ b/pkg/sentry/syscalls/linux/BUILD
@@ -74,6 +74,7 @@ go_library(
         "//pkg/sentry/kernel/pipe",
         "//pkg/sentry/kernel/sched",
         "//pkg/sentry/kernel/shm",
+        "//pkg/sentry/kernel/signalfd",
         "//pkg/sentry/kernel/time",
         "//pkg/sentry/limits",
         "//pkg/sentry/memmap",
diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go
index 150999fb8..18d24ab61 100644
--- a/pkg/sentry/syscalls/linux/linux64.go
+++ b/pkg/sentry/syscalls/linux/linux64.go
@@ -327,14 +327,14 @@ var AMD64 = &kernel.SyscallTable{
 		279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil),                               // requires cap_sys_nice (mostly)
 		280: syscalls.Supported("utimensat", Utimensat),
 		281: syscalls.Supported("epoll_pwait", EpollPwait),
-		282: syscalls.ErrorWithEvent("signalfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426)
+		282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}),
 		283: syscalls.Supported("timerfd_create", TimerfdCreate),
 		284: syscalls.Supported("eventfd", Eventfd),
 		285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil),
 		286: syscalls.Supported("timerfd_settime", TimerfdSettime),
 		287: syscalls.Supported("timerfd_gettime", TimerfdGettime),
 		288: syscalls.Supported("accept4", Accept4),
-		289: syscalls.ErrorWithEvent("signalfd4", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426)
+		289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}),
 		290: syscalls.Supported("eventfd2", Eventfd2),
 		291: syscalls.Supported("epoll_create1", EpollCreate1),
 		292: syscalls.Supported("dup3", Dup3),
diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go
index 0104a94c0..fb6efd5d8 100644
--- a/pkg/sentry/syscalls/linux/sys_signal.go
+++ b/pkg/sentry/syscalls/linux/sys_signal.go
@@ -20,7 +20,10 @@ import (
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/signalfd"
+	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserror"
 )
 
@@ -506,3 +509,77 @@ func RestartSyscall(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne
 	t.Debugf("Restart block missing in restart_syscall(2). Did ptrace inject a return value of ERESTART_RESTARTBLOCK?")
 	return 0, nil, syserror.EINTR
 }
+
+// sharedSignalfd is shared between the two calls.
+func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize uint, flags int32) (uintptr, *kernel.SyscallControl, error) {
+	// Copy in the signal mask.
+	mask, err := copyInSigSet(t, sigset, sigsetsize)
+	if err != nil {
+		return 0, nil, err
+	}
+
+	// Always check for valid flags, even if not creating.
+	if flags&^(linux.SFD_NONBLOCK|linux.SFD_CLOEXEC) != 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	// Is this a change to an existing signalfd?
+	//
+	// The spec indicates that this should adjust the mask.
+	if fd != -1 {
+		file := t.GetFile(fd)
+		if file == nil {
+			return 0, nil, syserror.EBADF
+		}
+		defer file.DecRef()
+
+		// Is this a signalfd?
+		if s, ok := file.FileOperations.(*signalfd.SignalOperations); ok {
+			s.SetMask(mask)
+			return 0, nil, nil
+		}
+
+		// Not a signalfd.
+		return 0, nil, syserror.EINVAL
+	}
+
+	// Create a new file.
+	file, err := signalfd.New(t, mask)
+	if err != nil {
+		return 0, nil, err
+	}
+	defer file.DecRef()
+
+	// Set appropriate flags.
+	file.SetFlags(fs.SettableFileFlags{
+		NonBlocking: flags&linux.SFD_NONBLOCK != 0,
+	})
+
+	// Create a new descriptor.
+	fd, err = t.NewFDFrom(0, file, kernel.FDFlags{
+		CloseOnExec: flags&linux.SFD_CLOEXEC != 0,
+	})
+	if err != nil {
+		return 0, nil, err
+	}
+
+	// Done.
+	return uintptr(fd), nil, nil
+}
+
+// Signalfd implements the linux syscall signalfd(2).
+func Signalfd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	fd := args[0].Int()
+	sigset := args[1].Pointer()
+	sigsetsize := args[2].SizeT()
+	return sharedSignalfd(t, fd, sigset, sigsetsize, 0)
+}
+
+// Signalfd4 implements the linux syscall signalfd4(2).
+func Signalfd4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	fd := args[0].Int()
+	sigset := args[1].Pointer()
+	sigsetsize := args[2].SizeT()
+	flags := args[3].Int()
+	return sharedSignalfd(t, fd, sigset, sigsetsize, flags)
+}