go/parser: limit recursion depth

Limit nested parsing to 100,000, which prevents stack exhaustion when
parsing deeply nested statements, types, and expressions. Also limit
the scope depth to 1,000 during object resolution.

Thanks to Juho Nurminen of Mattermost for reporting this issue.

Fixes #53616
Fixes CVE-2022-1962

Change-Id: I4d7b86c1d75d0bf3c7af1fdea91582aa74272c64
Reviewed-on: https://team-review.git.corp.google.com/c/golang/go-private/+/1491025
Reviewed-by: Russ Cox <rsc@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-on: https://go-review.googlesource.com/c/go/+/417063
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>

src/go/parser/interface.go

@@ -94,8 +94,11 @@ func ParseFile(fset *token.FileSet, filename string, src any, mode Mode) (f *ast
 	defer func() {
 		if e := recover(); e != nil {
 			// resume same panic if it's not a bailout
-			if _, ok := e.(bailout); !ok {
+			bail, ok := e.(bailout)
+			if !ok {
 				panic(e)
+			} else if bail.msg != "" {
+				p.errors.Add(p.file.Position(bail.pos), bail.msg)
 			}
 		}
 
@@ -198,8 +201,11 @@ func ParseExprFrom(fset *token.FileSet, filename string, src any, mode Mode) (ex
 	defer func() {
 		if e := recover(); e != nil {
 			// resume same panic if it's not a bailout
-			if _, ok := e.(bailout); !ok {
+			bail, ok := e.(bailout)
+			if !ok {
 				panic(e)
+			} else if bail.msg != "" {
+				p.errors.Add(p.file.Position(bail.pos), bail.msg)
 			}
 		}
 		p.errors.Sort()

src/go/parser/parser.go

@@ -59,6 +59,10 @@ type parser struct {
 	inRhs   bool // if set, the parser is parsing a rhs expression
 
 	imports []*ast.ImportSpec // list of imports
+
+	// nestLev is used to track and limit the recursion depth
+	// during parsing.
+	nestLev int
 }
 
 func (p *parser) init(fset *token.FileSet, filename string, src []byte, mode Mode) {
@@ -108,6 +112,24 @@ func un(p *parser) {
 	p.printTrace(")")
 }
 
+// maxNestLev is the deepest we're willing to recurse during parsing
+const maxNestLev int = 1e5
+
+func incNestLev(p *parser) *parser {
+	p.nestLev++
+	if p.nestLev > maxNestLev {
+		p.error(p.pos, "exceeded max nesting depth")
+		panic(bailout{})
+	}
+	return p
+}
+
+// decNestLev is used to track nesting depth during parsing to prevent stack exhaustion.
+// It is used along with incNestLev in a similar fashion to how un and trace are used.
+func decNestLev(p *parser) {
+	p.nestLev--
+}
+
 // Advance to the next token.
 func (p *parser) next0() {
 	// Because of one-token look-ahead, print the previous token
@@ -218,8 +240,12 @@ func (p *parser) next() {
 	}
 }
 
-// A bailout panic is raised to indicate early termination.
+// A bailout panic is raised to indicate early termination. pos and msg are
-type bailout struct{}
+// only populated when bailing out of object resolution.
+type bailout struct {
+	pos token.Pos
+	msg string
+}
 
 func (p *parser) error(pos token.Pos, msg string) {
 	if p.trace {
@@ -1247,6 +1273,8 @@ func (p *parser) parseTypeInstance(typ ast.Expr) ast.Expr {
 }
 
 func (p *parser) tryIdentOrType() ast.Expr {
+	defer decNestLev(incNestLev(p))
+
 	switch p.tok {
 	case token.IDENT:
 		typ := p.parseTypeName(nil)
@@ -1657,7 +1685,13 @@ func (p *parser) parsePrimaryExpr(x ast.Expr) ast.Expr {
 	if x == nil {
 		x = p.parseOperand()
 	}
-	for {
+	// We track the nesting here rather than at the entry for the function,
+	// since it can iteratively produce a nested output, and we want to
+	// limit how deep a structure we generate.
+	var n int
+	defer func() { p.nestLev -= n }()
+	for n = 1; ; n++ {
+		incNestLev(p)
 		switch p.tok {
 		case token.PERIOD:
 			p.next()
@@ -1717,6 +1751,8 @@ func (p *parser) parsePrimaryExpr(x ast.Expr) ast.Expr {
 }
 
 func (p *parser) parseUnaryExpr() ast.Expr {
+	defer decNestLev(incNestLev(p))
+
 	if p.trace {
 		defer un(trace(p, "UnaryExpr"))
 	}
@@ -1806,7 +1842,13 @@ func (p *parser) parseBinaryExpr(x ast.Expr, prec1 int, check bool) ast.Expr {
 	if x == nil {
 		x = p.parseUnaryExpr()
 	}
-	for {
+	// We track the nesting here rather than at the entry for the function,
+	// since it can iteratively produce a nested output, and we want to
+	// limit how deep a structure we generate.
+	var n int
+	defer func() { p.nestLev -= n }()
+	for n = 1; ; n++ {
+		incNestLev(p)
 		op, oprec := p.tokPrec()
 		if oprec < prec1 {
 			return x
@@ -2099,6 +2141,8 @@ func (p *parser) parseIfHeader() (init ast.Stmt, cond ast.Expr) {
 }
 
 func (p *parser) parseIfStmt() *ast.IfStmt {
+	defer decNestLev(incNestLev(p))
+
 	if p.trace {
 		defer un(trace(p, "IfStmt"))
 	}
@@ -2402,6 +2446,8 @@ func (p *parser) parseForStmt() ast.Stmt {
 }
 
 func (p *parser) parseStmt() (s ast.Stmt) {
+	defer decNestLev(incNestLev(p))
+
 	if p.trace {
 		defer un(trace(p, "Statement"))
 	}

src/go/parser/parser_test.go

@@ -10,6 +10,7 @@ import (
 	"go/ast"
 	"go/token"
 	"io/fs"
+	"runtime"
 	"strings"
 	"testing"
 )
@@ -577,3 +578,171 @@ type x int // comment
 		t.Errorf("got %q, want %q", comment, "// comment")
 	}
 }
+
+var parseDepthTests = []struct {
+	name   string
+	format string
+	// multipler is used when a single statement may result in more than one
+	// change in the depth level, for instance "1+(..." produces a BinaryExpr
+	// followed by a UnaryExpr, which increments the depth twice. The test
+	// case comment explains which nodes are triggering the multiple depth
+	// changes.
+	parseMultiplier int
+	// scope is true if we should also test the statement for the resolver scope
+	// depth limit.
+	scope bool
+	// scopeMultiplier does the same as parseMultiplier, but for the scope
+	// depths.
+	scopeMultiplier int
+}{
+	// The format expands the part inside « » many times.
+	// A second set of brackets nested inside the first stops the repetition,
+	// so that for example «(«1»)» expands to (((...((((1))))...))).
+	{name: "array", format: "package main; var x «[1]»int"},
+	{name: "slice", format: "package main; var x «[]»int"},
+	{name: "struct", format: "package main; var x «struct { X «int» }»", scope: true},
+	{name: "pointer", format: "package main; var x «*»int"},
+	{name: "func", format: "package main; var x «func()»int", scope: true},
+	{name: "chan", format: "package main; var x «chan »int"},
+	{name: "chan2", format: "package main; var x «<-chan »int"},
+	{name: "interface", format: "package main; var x «interface { M() «int» }»", scope: true, scopeMultiplier: 2}, // Scopes: InterfaceType, FuncType
+	{name: "map", format: "package main; var x «map[int]»int"},
+	{name: "slicelit", format: "package main; var x = «[]any{«»}»", parseMultiplier: 2},             // Parser nodes: UnaryExpr, CompositeLit
+	{name: "arraylit", format: "package main; var x = «[1]any{«nil»}»", parseMultiplier: 2},         // Parser nodes: UnaryExpr, CompositeLit
+	{name: "structlit", format: "package main; var x = «struct{x any}{«nil»}»", parseMultiplier: 2}, // Parser nodes: UnaryExpr, CompositeLit
+	{name: "maplit", format: "package main; var x = «map[int]any{1:«nil»}»", parseMultiplier: 2},    // Parser nodes: CompositeLit, KeyValueExpr
+	{name: "dot", format: "package main; var x = «x.»x"},
+	{name: "index", format: "package main; var x = x«[1]»"},
+	{name: "slice", format: "package main; var x = x«[1:2]»"},
+	{name: "slice3", format: "package main; var x = x«[1:2:3]»"},
+	{name: "dottype", format: "package main; var x = x«.(any)»"},
+	{name: "callseq", format: "package main; var x = x«()»"},
+	{name: "methseq", format: "package main; var x = x«.m()»", parseMultiplier: 2}, // Parser nodes: SelectorExpr, CallExpr
+	{name: "binary", format: "package main; var x = «1+»1"},
+	{name: "binaryparen", format: "package main; var x = «1+(«1»)»", parseMultiplier: 2}, // Parser nodes: BinaryExpr, ParenExpr
+	{name: "unary", format: "package main; var x = «^»1"},
+	{name: "addr", format: "package main; var x = «& »x"},
+	{name: "star", format: "package main; var x = «*»x"},
+	{name: "recv", format: "package main; var x = «<-»x"},
+	{name: "call", format: "package main; var x = «f(«1»)»", parseMultiplier: 2},    // Parser nodes: Ident, CallExpr
+	{name: "conv", format: "package main; var x = «(*T)(«1»)»", parseMultiplier: 2}, // Parser nodes: ParenExpr, CallExpr
+	{name: "label", format: "package main; func main() { «Label:» }"},
+	{name: "if", format: "package main; func main() { «if true { «» }»}", parseMultiplier: 2, scope: true, scopeMultiplier: 2}, // Parser nodes: IfStmt, BlockStmt. Scopes: IfStmt, BlockStmt
+	{name: "ifelse", format: "package main; func main() { «if true {} else » {} }", scope: true},
+	{name: "switch", format: "package main; func main() { «switch { default: «» }»}", scope: true, scopeMultiplier: 2},               // Scopes: TypeSwitchStmt, CaseClause
+	{name: "typeswitch", format: "package main; func main() { «switch x.(type) { default: «» }» }", scope: true, scopeMultiplier: 2}, // Scopes: TypeSwitchStmt, CaseClause
+	{name: "for0", format: "package main; func main() { «for { «» }» }", scope: true, scopeMultiplier: 2},                            // Scopes: ForStmt, BlockStmt
+	{name: "for1", format: "package main; func main() { «for x { «» }» }", scope: true, scopeMultiplier: 2},                          // Scopes: ForStmt, BlockStmt
+	{name: "for3", format: "package main; func main() { «for f(); g(); h() { «» }» }", scope: true, scopeMultiplier: 2},              // Scopes: ForStmt, BlockStmt
+	{name: "forrange0", format: "package main; func main() { «for range x { «» }» }", scope: true, scopeMultiplier: 2},               // Scopes: RangeStmt, BlockStmt
+	{name: "forrange1", format: "package main; func main() { «for x = range z { «» }» }", scope: true, scopeMultiplier: 2},           // Scopes: RangeStmt, BlockStmt
+	{name: "forrange2", format: "package main; func main() { «for x, y = range z { «» }» }", scope: true, scopeMultiplier: 2},        // Scopes: RangeStmt, BlockStmt
+	{name: "go", format: "package main; func main() { «go func() { «» }()» }", parseMultiplier: 2, scope: true},                      // Parser nodes: GoStmt, FuncLit
+	{name: "defer", format: "package main; func main() { «defer func() { «» }()» }", parseMultiplier: 2, scope: true},                // Parser nodes: DeferStmt, FuncLit
+	{name: "select", format: "package main; func main() { «select { default: «» }» }", scope: true},
+}
+
+// split splits pre«mid»post into pre, mid, post.
+// If the string does not have that form, split returns x, "", "".
+func split(x string) (pre, mid, post string) {
+	start, end := strings.Index(x, "«"), strings.LastIndex(x, "»")
+	if start < 0 || end < 0 {
+		return x, "", ""
+	}
+	return x[:start], x[start+len("«") : end], x[end+len("»"):]
+}
+
+func TestParseDepthLimit(t *testing.T) {
+	if runtime.GOARCH == "wasm" {
+		t.Skip("causes call stack exhaustion on js/wasm")
+	}
+	for _, tt := range parseDepthTests {
+		for _, size := range []string{"small", "big"} {
+			t.Run(tt.name+"/"+size, func(t *testing.T) {
+				n := maxNestLev + 1
+				if tt.parseMultiplier > 0 {
+					n /= tt.parseMultiplier
+				}
+				if size == "small" {
+					// Decrease the number of statements by 10, in order to check
+					// that we do not fail when under the limit. 10 is used to
+					// provide some wiggle room for cases where the surrounding
+					// scaffolding syntax adds some noise to the depth that changes
+					// on a per testcase basis.
+					n -= 10
+				}
+
+				pre, mid, post := split(tt.format)
+				if strings.Contains(mid, "«") {
+					left, base, right := split(mid)
+					mid = strings.Repeat(left, n) + base + strings.Repeat(right, n)
+				} else {
+					mid = strings.Repeat(mid, n)
+				}
+				input := pre + mid + post
+
+				fset := token.NewFileSet()
+				_, err := ParseFile(fset, "", input, ParseComments|SkipObjectResolution)
+				if size == "small" {
+					if err != nil {
+						t.Errorf("ParseFile(...): %v (want success)", err)
+					}
+				} else {
+					expected := "exceeded max nesting depth"
+					if err == nil || !strings.HasSuffix(err.Error(), expected) {
+						t.Errorf("ParseFile(...) = _, %v, want %q", err, expected)
+					}
+				}
+			})
+		}
+	}
+}
+
+func TestScopeDepthLimit(t *testing.T) {
+	if runtime.GOARCH == "wasm" {
+		t.Skip("causes call stack exhaustion on js/wasm")
+	}
+	for _, tt := range parseDepthTests {
+		if !tt.scope {
+			continue
+		}
+		for _, size := range []string{"small", "big"} {
+			t.Run(tt.name+"/"+size, func(t *testing.T) {
+				n := maxScopeDepth + 1
+				if tt.scopeMultiplier > 0 {
+					n /= tt.scopeMultiplier
+				}
+				if size == "small" {
+					// Decrease the number of statements by 10, in order to check
+					// that we do not fail when under the limit. 10 is used to
+					// provide some wiggle room for cases where the surrounding
+					// scaffolding syntax adds some noise to the depth that changes
+					// on a per testcase basis.
+					n -= 10
+				}
+
+				pre, mid, post := split(tt.format)
+				if strings.Contains(mid, "«") {
+					left, base, right := split(mid)
+					mid = strings.Repeat(left, n) + base + strings.Repeat(right, n)
+				} else {
+					mid = strings.Repeat(mid, n)
+				}
+				input := pre + mid + post
+
+				fset := token.NewFileSet()
+				_, err := ParseFile(fset, "", input, DeclarationErrors)
+				if size == "small" {
+					if err != nil {
+						t.Errorf("ParseFile(...): %v (want success)", err)
+					}
+				} else {
+					expected := "exceeded max scope depth during object resolution"
+					if err == nil || !strings.HasSuffix(err.Error(), expected) {
+						t.Errorf("ParseFile(...) = _, %v, want %q", err, expected)
+					}
+				}
+			})
+		}
+	}
+}

src/go/parser/resolver.go

@@ -54,6 +54,8 @@ func resolveFile(file *ast.File, handle *token.File, declErr func(token.Pos, str
 	file.Unresolved = r.unresolved[0:i]
 }
 
+const maxScopeDepth int = 1e3
+
 type resolver struct {
 	handle  *token.File
 	declErr func(token.Pos, string)
@@ -85,16 +87,19 @@ func (r *resolver) sprintf(format string, args ...any) string {
 }
 
 func (r *resolver) openScope(pos token.Pos) {
+	r.depth++
+	if r.depth > maxScopeDepth {
+		panic(bailout{pos: pos, msg: "exceeded max scope depth during object resolution"})
+	}
 	if debugResolve {
 		r.trace("opening scope @%v", pos)
-		r.depth++
 	}
 	r.topScope = ast.NewScope(r.topScope)
 }
 
 func (r *resolver) closeScope() {
+	r.depth--
 	if debugResolve {
-		r.depth--
 		r.trace("closing scope")
 	}
 	r.topScope = r.topScope.Outer