unix: avoid allocations for common uses of Readv, Writev, etc.

Fixes golang/go#57296

Change-Id: Ifd57487122a590df467e97e2d35f388a58cc080d
Reviewed-on: https://go-review.googlesource.com/c/sys/+/457815
Run-TryBot: Ian Lance Taylor <iant@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Reviewed-by: Benny Siegert <bsiegert@gmail.com>
Reviewed-by: Matt Layher <mdlayher@gmail.com>

unix/syscall_linux.go

@@ -1973,17 +1973,27 @@ func Signalfd(fd int, sigmask *Sigset_t, flags int) (newfd int, err error) {
 //sys	preadv2(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr, flags int) (n int, err error) = SYS_PREADV2
 //sys	pwritev2(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr, flags int) (n int, err error) = SYS_PWRITEV2
 
-func bytes2iovec(bs [][]byte) []Iovec {
-	iovecs := make([]Iovec, len(bs))
-	for i, b := range bs {
-		iovecs[i].SetLen(len(b))
+// minIovec is the size of the small initial allocation used by
+// Readv, Writev, etc.
+//
+// This small allocation gets stack allocated, which lets the
+// common use case of len(iovs) <= minIovs avoid more expensive
+// heap allocations.
+const minIovec = 8
+
+// appendBytes converts bs to Iovecs and appends them to vecs.
+func appendBytes(vecs []Iovec, bs [][]byte) []Iovec {
+	for _, b := range bs {
+		var v Iovec
+		v.SetLen(len(b))
 		if len(b) > 0 {
-			iovecs[i].Base = &b[0]
+			v.Base = &b[0]
 		} else {
-			iovecs[i].Base = (*byte)(unsafe.Pointer(&_zero))
+			v.Base = (*byte)(unsafe.Pointer(&_zero))
 		}
+		vecs = append(vecs, v)
 	}
-	return iovecs
+	return vecs
 }
 
 // offs2lohi splits offs into its low and high order bits.
@@ -1993,22 +2003,25 @@ func offs2lohi(offs int64) (lo, hi uintptr) {
 }
 
 func Readv(fd int, iovs [][]byte) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	n, err = readv(fd, iovecs)
 	readvRacedetect(iovecs, n, err)
 	return n, err
 }
 
 func Preadv(fd int, iovs [][]byte, offset int64) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	lo, hi := offs2lohi(offset)
 	n, err = preadv(fd, iovecs, lo, hi)
 	readvRacedetect(iovecs, n, err)
 	return n, err
 }
 
 func Preadv2(fd int, iovs [][]byte, offset int64, flags int) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	lo, hi := offs2lohi(offset)
 	n, err = preadv2(fd, iovecs, lo, hi, flags)
 	readvRacedetect(iovecs, n, err)
@@ -2035,7 +2048,8 @@ func readvRacedetect(iovecs []Iovec, n int, err error) {
 }
 
 func Writev(fd int, iovs [][]byte) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	if raceenabled {
 		raceReleaseMerge(unsafe.Pointer(&ioSync))
 	}
@@ -2045,7 +2059,8 @@ func Writev(fd int, iovs [][]byte) (n int, err error) {
 }
 
 func Pwritev(fd int, iovs [][]byte, offset int64) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	if raceenabled {
 		raceReleaseMerge(unsafe.Pointer(&ioSync))
 	}
@@ -2056,7 +2071,8 @@ func Pwritev(fd int, iovs [][]byte, offset int64) (n int, err error) {
 }
 
 func Pwritev2(fd int, iovs [][]byte, offset int64, flags int) (n int, err error) {
-	iovecs := bytes2iovec(iovs)
+	iovecs := make([]Iovec, 0, minIovec)
+	iovecs = appendBytes(iovecs, iovs)
 	if raceenabled {
 		raceReleaseMerge(unsafe.Pointer(&ioSync))
 	}

unix/syscall_linux_test.go

@@ -1132,3 +1132,44 @@ func TestPwritevOffsets(t *testing.T) {
 		t.Fatalf("expected size to be %d, got %d", want, info.Size())
 	}
 }
+
+func TestReadvAllocate(t *testing.T) {
+	f, err := os.Create(filepath.Join(t.TempDir(), "test"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { f.Close() })
+
+	test := func(name string, fn func(fd int)) {
+		n := int(testing.AllocsPerRun(100, func() {
+			fn(int(f.Fd()))
+		}))
+		if n != 0 {
+			t.Errorf("%q got %d allocations, want 0", name, n)
+		}
+	}
+
+	iovs := make([][]byte, 8)
+	for i := range iovs {
+		iovs[i] = []byte{'A'}
+	}
+
+	test("Writev", func(fd int) {
+		unix.Writev(fd, iovs)
+	})
+	test("Pwritev", func(fd int) {
+		unix.Pwritev(fd, iovs, 0)
+	})
+	test("Pwritev2", func(fd int) {
+		unix.Pwritev2(fd, iovs, 0, 0)
+	})
+	test("Readv", func(fd int) {
+		unix.Readv(fd, iovs)
+	})
+	test("Preadv", func(fd int) {
+		unix.Preadv(fd, iovs, 0)
+	})
+	test("Preadv2", func(fd int) {
+		unix.Preadv2(fd, iovs, 0, 0)
+	})
+}