os/user: speed up Current on Windows

[This is a roll-forward of CL 597255, which had to be rolled back
because it broke the windows-arm64 builder, whose current user display
name is unavailable. This new CL fixes the issue by reintroducing the
historical behavior of falling back to the user name instead of
returning an error].

user.Current is slow on Windows sessions connected to an Active
Directory domain. This is because it uses Windows APIs that do RPC
calls to the domain controller, such as TranslateAccountW and
NetUserGetInfo.

This change speeds up user.Current by using the GetUserNameEx API
instead, which is already optimized for retrieving the current user
name in different formats.

These are the improvements I see with the new implementation:

goos: windows
goarch: amd64
pkg: os/user
cpu: Intel(R) Core(TM) i7-10850H CPU @ 2.70GHz
           │   old.txt   │               new.txt                │
           │   sec/op    │    sec/op     vs base                │
Current-12   501.8µ ± 7%   118.6µ ± 11%  -76.36% (p=0.000 n=10)

           │  old.txt   │              new.txt              │
           │    B/op    │    B/op     vs base               │
Current-12   888.0 ± 0%   832.0 ± 0%  -6.31% (p=0.000 n=10)

           │  old.txt   │              new.txt               │
           │ allocs/op  │ allocs/op   vs base                │
Current-12   15.00 ± 0%   11.00 ± 0%  -26.67% (p=0.000 n=10)

Updates #5298
Fixes #21867
Fixes #68312

Cq-Include-Trybots: luci.golang.try:gotip-windows-amd64-longtest,gotip-windows-arm64
Change-Id: Ib7f77086d389cccb9d91cb77ea688d438a0ee5fd
Reviewed-on: https://go-review.googlesource.com/c/go/+/605135


Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: Mauri de Souza Meneguzzo <mauri870@gmail.com>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

src/internal/syscall/windows/security_windows.go

@@ -156,3 +156,22 @@ type UserInfo4 struct {
 //
 //go:linkname GetSystemDirectory
 func GetSystemDirectory() string // Implemented in runtime package.
+
+// GetUserName retrieves the user name of the current thread
+// in the specified format.
+func GetUserName(format uint32) (string, error) {
+	n := uint32(50)
+	for {
+		b := make([]uint16, n)
+		e := syscall.GetUserNameEx(format, &b[0], &n)
+		if e == nil {
+			return syscall.UTF16ToString(b[:n]), nil
+		}
+		if e != syscall.ERROR_MORE_DATA {
+			return "", e
+		}
+		if n <= uint32(len(b)) {
+			return "", e
+		}
+	}
+}

src/os/user/lookup_windows.go

@@ -232,12 +232,24 @@ func current() (*User, error) {
 		if e != nil {
 			return e
 		}
-		username, domain, e := lookupUsernameAndDomain(u.User.Sid)
+		username, e := windows.GetUserName(syscall.NameSamCompatible)
 		if e != nil {
 			return e
 		}
-		usr, e = newUser(uid, gid, dir, username, domain)
-		return e
+		displayName, e := windows.GetUserName(syscall.NameDisplay)
+		if e != nil {
+			// Historically, the username is used as fallback
+			// when the display name can't be retrieved.
+			displayName = username
+		}
+		usr = &User{
+			Uid:      uid,
+			Gid:      gid,
+			Username: username,
+			Name:     displayName,
+			HomeDir:  dir,
+		}
+		return nil
 	})
 	return usr, err
 }

src/os/user/user_test.go

@@ -45,8 +45,9 @@ func TestCurrent(t *testing.T) {
 }
 
 func BenchmarkCurrent(b *testing.B) {
+	// Benchmark current instead of Current because Current caches the result.
 	for i := 0; i < b.N; i++ {
-		Current()
+		current()
 	}
 }