Merge branch 'master' into pr_syscall_linux

9 files changed, 194 insertions, 86 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 2f77d587b..72c383537 100644
--- a/pkg/sentry/syscalls/linux/linux64.go
+++ b/pkg/sentry/syscalls/linux/linux64.go
@@ -19,4 +19,4 @@ const (
 	_LINUX_SYSNAME = "Linux"
 	_LINUX_RELEASE = "4.4"
 	_LINUX_VERSION = "#1 SMP Sun Jan 10 15:06:54 PST 2016"
-)
+)
\ No newline at end of file
diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go
index 2e00a91ce..b9a8e3e21 100644
--- a/pkg/sentry/syscalls/linux/sys_file.go
+++ b/pkg/sentry/syscalls/linux/sys_file.go
@@ -1423,9 +1423,6 @@ func unlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error {
 	if err != nil {
 		return err
 	}
-	if dirPath {
-		return syserror.ENOENT
-	}
 
 	return fileOpAt(t, dirFD, path, func(root *fs.Dirent, d *fs.Dirent, name string, _ uint) error {
 		if !fs.IsDir(d.Inode.StableAttr) {
@@ -1436,7 +1433,7 @@ func unlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error {
 			return err
 		}
 
-		return d.Remove(t, root, name)
+		return d.Remove(t, root, name, dirPath)
 	})
 }
 
diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go
index 3ab54271c..cd31e0649 100644
--- a/pkg/sentry/syscalls/linux/sys_read.go
+++ b/pkg/sentry/syscalls/linux/sys_read.go
@@ -72,6 +72,39 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC
 	return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "read", file)
 }
 
+// Readahead implements readahead(2).
+func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	fd := args[0].Int()
+	offset := args[1].Int64()
+	size := args[2].SizeT()
+
+	file := t.GetFile(fd)
+	if file == nil {
+		return 0, nil, syserror.EBADF
+	}
+	defer file.DecRef()
+
+	// Check that the file is readable.
+	if !file.Flags().Read {
+		return 0, nil, syserror.EBADF
+	}
+
+	// Check that the size is valid.
+	if int(size) < 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	// Check that the offset is legitimate.
+	if offset < 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	// Return EINVAL; if the underlying file type does not support readahead,
+	// then Linux will return EINVAL to indicate as much. In the future, we
+	// may extend this function to actually support readahead hints.
+	return 0, nil, syserror.EINVAL
+}
+
 // Pread64 implements linux syscall pread64(2).
 func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
 	fd := args[0].Int()
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)
+}
diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go
index 3bac4d90d..b5a72ce63 100644
--- a/pkg/sentry/syscalls/linux/sys_socket.go
+++ b/pkg/sentry/syscalls/linux/sys_socket.go
@@ -531,7 +531,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
 		return 0, nil, syserror.ENOTSOCK
 	}
 
-	if optLen <= 0 {
+	if optLen < 0 {
 		return 0, nil, syserror.EINVAL
 	}
 	if optLen > maxOptLen {
diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go
index 17e3dde1f..9f705ebca 100644
--- a/pkg/sentry/syscalls/linux/sys_splice.go
+++ b/pkg/sentry/syscalls/linux/sys_splice.go
@@ -29,9 +29,8 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB
 		total int64
 		n     int64
 		err   error
-		ch    chan struct{}
-		inW   bool
-		outW  bool
+		inCh  chan struct{}
+		outCh chan struct{}
 	)
 	for opts.Length > 0 {
 		n, err = fs.Splice(t, outFile, inFile, opts)
@@ -43,35 +42,33 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB
 			break
 		}
 
-		// Are we a registered waiter?
-		if ch == nil {
-			ch = make(chan struct{}, 1)
-		}
-		if !inW && !inFile.Flags().NonBlocking {
-			w, _ := waiter.NewChannelEntry(ch)
-			inFile.EventRegister(&w, EventMaskRead)
-			defer inFile.EventUnregister(&w)
-			inW = true // Registered.
-		} else if !outW && !outFile.Flags().NonBlocking {
-			w, _ := waiter.NewChannelEntry(ch)
-			outFile.EventRegister(&w, EventMaskWrite)
-			defer outFile.EventUnregister(&w)
-			outW = true // Registered.
-		}
-
-		// Was anything registered? If no, everything is non-blocking.
-		if !inW && !outW {
-			break
-		}
-
-		if (!inW || inFile.Readiness(EventMaskRead) != 0) && (!outW || outFile.Readiness(EventMaskWrite) != 0) {
-			// Something became ready, try again without blocking.
-			continue
+		// Note that the blocking behavior here is a bit different than the
+		// normal pattern. Because we need to have both data to read and data
+		// to write simultaneously, we actually explicitly block on both of
+		// these cases in turn before returning to the splice operation.
+		if inFile.Readiness(EventMaskRead) == 0 {
+			if inCh == nil {
+				inCh = make(chan struct{}, 1)
+				inW, _ := waiter.NewChannelEntry(inCh)
+				inFile.EventRegister(&inW, EventMaskRead)
+				defer inFile.EventUnregister(&inW)
+				continue // Need to refresh readiness.
+			}
+			if err = t.Block(inCh); err != nil {
+				break
+			}
 		}
-
-		// Block until there's data.
-		if err = t.Block(ch); err != nil {
-			break
+		if outFile.Readiness(EventMaskWrite) == 0 {
+			if outCh == nil {
+				outCh = make(chan struct{}, 1)
+				outW, _ := waiter.NewChannelEntry(outCh)
+				outFile.EventRegister(&outW, EventMaskWrite)
+				defer outFile.EventUnregister(&outW)
+				continue // Need to refresh readiness.
+			}
+			if err = t.Block(outCh); err != nil {
+				break
+			}
 		}
 	}
 
@@ -91,22 +88,29 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
 	}
 
 	// Get files.
+	inFile := t.GetFile(inFD)
+	if inFile == nil {
+		return 0, nil, syserror.EBADF
+	}
+	defer inFile.DecRef()
+
+	if !inFile.Flags().Read {
+		return 0, nil, syserror.EBADF
+	}
+
 	outFile := t.GetFile(outFD)
 	if outFile == nil {
 		return 0, nil, syserror.EBADF
 	}
 	defer outFile.DecRef()
 
-	inFile := t.GetFile(inFD)
-	if inFile == nil {
+	if !outFile.Flags().Write {
 		return 0, nil, syserror.EBADF
 	}
-	defer inFile.DecRef()
 
-	// Verify that the outfile Append flag is not set. Note that fs.Splice
-	// itself validates that the output file is writable.
+	// Verify that the outfile Append flag is not set.
 	if outFile.Flags().Append {
-		return 0, nil, syserror.EBADF
+		return 0, nil, syserror.EINVAL
 	}
 
 	// Verify that we have a regular infile. This is a requirement; the
@@ -142,7 +146,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
 			Length:    count,
 			SrcOffset: true,
 			SrcStart:  offset,
-		}, false)
+		}, outFile.Flags().NonBlocking)
 
 		// Copy out the new offset.
 		if _, err := t.CopyOut(offsetAddr, n+offset); err != nil {
@@ -152,7 +156,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
 		// Send data using splice.
 		n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{
 			Length: count,
-		}, false)
+		}, outFile.Flags().NonBlocking)
 	}
 
 	// We can only pass a single file to handleIOError, so pick inFile
@@ -174,12 +178,6 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
 		return 0, nil, syserror.EINVAL
 	}
 
-	// Only non-blocking is meaningful. Note that unlike in Linux, this
-	// flag is applied consistently. We will have either fully blocking or
-	// non-blocking behavior below, regardless of the underlying files
-	// being spliced to. It's unclear if this is a bug or not yet.
-	nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0
-
 	// Get files.
 	outFile := t.GetFile(outFD)
 	if outFile == nil {
@@ -193,6 +191,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
 	}
 	defer inFile.DecRef()
 
+	// The operation is non-blocking if anything is non-blocking.
+	//
+	// N.B. This is a rather simplistic heuristic that avoids some
+	// poor edge case behavior since the exact semantics here are
+	// underspecified and vary between versions of Linux itself.
+	nonBlock := inFile.Flags().NonBlocking || outFile.Flags().NonBlocking || (flags&linux.SPLICE_F_NONBLOCK != 0)
+
 	// Construct our options.
 	//
 	// Note that exactly one of the underlying buffers must be a pipe. We
@@ -240,17 +245,17 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
 		if inOffset != 0 || outOffset != 0 {
 			return 0, nil, syserror.ESPIPE
 		}
-	default:
-		return 0, nil, syserror.EINVAL
-	}
 
-	// We may not refer to the same pipe; otherwise it's a continuous loop.
-	if inFile.Dirent.Inode.StableAttr.InodeID == outFile.Dirent.Inode.StableAttr.InodeID {
+		// We may not refer to the same pipe; otherwise it's a continuous loop.
+		if inFile.Dirent.Inode.StableAttr.InodeID == outFile.Dirent.Inode.StableAttr.InodeID {
+			return 0, nil, syserror.EINVAL
+		}
+	default:
 		return 0, nil, syserror.EINVAL
 	}
 
 	// Splice data.
-	n, err := doSplice(t, outFile, inFile, opts, nonBlocking)
+	n, err := doSplice(t, outFile, inFile, opts, nonBlock)
 
 	// See above; inFile is chosen arbitrarily here.
 	return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "splice", inFile)
@@ -268,9 +273,6 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo
 		return 0, nil, syserror.EINVAL
 	}
 
-	// Only non-blocking is meaningful.
-	nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0
-
 	// Get files.
 	outFile := t.GetFile(outFD)
 	if outFile == nil {
@@ -294,11 +296,14 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo
 		return 0, nil, syserror.EINVAL
 	}
 
+	// The operation is non-blocking if anything is non-blocking.
+	nonBlock := inFile.Flags().NonBlocking || outFile.Flags().NonBlocking || (flags&linux.SPLICE_F_NONBLOCK != 0)
+
 	// Splice data.
 	n, err := doSplice(t, outFile, inFile, fs.SpliceOpts{
 		Length: count,
 		Dup:    true,
-	}, nonBlocking)
+	}, nonBlock)
 
 	// See above; inFile is chosen arbitrarily here.
 	return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "tee", inFile)
diff --git a/pkg/sentry/syscalls/linux/sys_time.go b/pkg/sentry/syscalls/linux/sys_time.go
index 4b3f043a2..b887fa9d7 100644
--- a/pkg/sentry/syscalls/linux/sys_time.go
+++ b/pkg/sentry/syscalls/linux/sys_time.go
@@ -15,6 +15,7 @@
 package linux
 
 import (
+	"fmt"
 	"time"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
@@ -228,41 +229,35 @@ func clockNanosleepFor(t *kernel.Task, c ktime.Clock, dur time.Duration, rem use
 
 	timer.Destroy()
 
-	var remaining time.Duration
-	// Did we just block for the entire duration?
-	if err == syserror.ETIMEDOUT {
-		remaining = 0
-	} else {
-		remaining = dur - after.Sub(start)
+	switch err {
+	case syserror.ETIMEDOUT:
+		// Slept for entire timeout.
+		return nil
+	case syserror.ErrInterrupted:
+		// Interrupted.
+		remaining := dur - after.Sub(start)
 		if remaining < 0 {
 			remaining = time.Duration(0)
 		}
-	}
 
-	// Copy out remaining time.
-	if err != nil && rem != usermem.Addr(0) {
-		timeleft := linux.NsecToTimespec(remaining.Nanoseconds())
-		if err := copyTimespecOut(t, rem, &timeleft); err != nil {
-			return err
+		// Copy out remaining time.
+		if rem != 0 {
+			timeleft := linux.NsecToTimespec(remaining.Nanoseconds())
+			if err := copyTimespecOut(t, rem, &timeleft); err != nil {
+				return err
+			}
 		}
-	}
-
-	// Did we just block for the entire duration?
-	if err == syserror.ETIMEDOUT {
-		return nil
-	}
 
-	// If interrupted, arrange for a restart with the remaining duration.
-	if err == syserror.ErrInterrupted {
+		// Arrange for a restart with the remaining duration.
 		t.SetSyscallRestartBlock(&clockNanosleepRestartBlock{
 			c:        c,
 			duration: remaining,
 			rem:      rem,
 		})
 		return kernel.ERESTART_RESTARTBLOCK
+	default:
+		panic(fmt.Sprintf("Impossible BlockWithTimer error %v", err))
 	}
-
-	return err
 }
 
 // Nanosleep implements linux syscall Nanosleep(2).
diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go
index 271ace08e..748e8dd8d 100644
--- a/pkg/sentry/syscalls/linux/sys_utsname.go
+++ b/pkg/sentry/syscalls/linux/sys_utsname.go
@@ -79,11 +79,11 @@ func Sethostname(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S
 		return 0, nil, syserror.EINVAL
 	}
 
-	name, err := t.CopyInString(nameAddr, int(size))
-	if err != nil {
+	name := make([]byte, size)
+	if _, err := t.CopyInBytes(nameAddr, name); err != nil {
 		return 0, nil, err
 	}
 
-	utsns.SetHostName(name)
+	utsns.SetHostName(string(name))
 	return 0, nil, nil
 }