@@ -29,17 +29,43 @@ export const SAFE_IMAGE_FILE_EXTENSIONS = new Set([
2929 ".webp" ,
3030] ) ;
3131
32+ // Whether `code` is a character the base64 payload may contain, aside from
33+ // the whitespace handled separately below.
34+ function isBase64Char ( code : number ) : boolean {
35+ return (
36+ ( code >= 0x61 && code <= 0x7a ) || // a-z
37+ ( code >= 0x41 && code <= 0x5a ) || // A-Z
38+ ( code >= 0x30 && code <= 0x39 ) || // 0-9
39+ code === 0x2b || // +
40+ code === 0x2f || // /
41+ code === 0x3d // =
42+ ) ;
43+ }
44+
45+ function isBase64Whitespace ( code : number ) : boolean {
46+ return code === 0x0d || code === 0x0a || code === 0x20 ; // \r \n space
47+ }
48+
49+ // Data URLs carry the full image payload, so this parser must never run a
50+ // regex across the payload: V8's regex engine borrows the JS call stack, and
51+ // matching a multi-megabyte string from a deep call stack (e.g. inside fiber
52+ // execution) throws "Maximum call stack size exceeded".
3253export function parseBase64DataUrl (
3354 dataUrl : string ,
3455) : { readonly mimeType : string ; readonly base64 : string } | null {
35- const match = / ^ d a t a : ( [ ^ , ] + ) , ( [ a - z 0 - 9 + / = \r \n ] + ) $ / i. exec ( dataUrl . trim ( ) ) ;
36- if ( ! match ) return null ;
56+ const trimmed = dataUrl . trim ( ) ;
57+ if ( trimmed . slice ( 0 , 5 ) . toLowerCase ( ) !== "data:" ) return null ;
58+
59+ const commaIndex = trimmed . indexOf ( "," ) ;
60+ if ( commaIndex === - 1 ) return null ;
61+ const header = trimmed . slice ( 5 , commaIndex ) ;
62+ if ( header . length === 0 ) return null ;
3763
3864 const headerParts : Array < string > = [ ] ;
39- for ( const part of ( match [ 1 ] ?? "" ) . split ( ";" ) ) {
40- const trimmed = part . trim ( ) ;
41- if ( trimmed . length > 0 ) {
42- headerParts . push ( trimmed ) ;
65+ for ( const part of header . split ( ";" ) ) {
66+ const partTrimmed = part . trim ( ) ;
67+ if ( partTrimmed . length > 0 ) {
68+ headerParts . push ( partTrimmed ) ;
4369 }
4470 }
4571 if ( headerParts . length < 2 ) {
@@ -51,8 +77,37 @@ export function parseBase64DataUrl(
5177 }
5278
5379 const mimeType = headerParts [ 0 ] ?. toLowerCase ( ) ;
54- const base64 = match [ 2 ] ?. replace ( / \s + / g, "" ) ;
55- if ( ! mimeType || ! base64 ) return null ;
80+ if ( ! mimeType ) return null ;
81+
82+ const payload = trimmed . slice ( commaIndex + 1 ) ;
83+ const runs : Array < string > = [ ] ;
84+ let runStart = - 1 ;
85+ for ( let index = 0 ; index < payload . length ; index += 1 ) {
86+ const code = payload . charCodeAt ( index ) ;
87+ if ( isBase64Char ( code ) ) {
88+ if ( runStart === - 1 ) runStart = index ;
89+ continue ;
90+ }
91+ if ( ! isBase64Whitespace ( code ) ) return null ;
92+ if ( runStart !== - 1 ) {
93+ runs . push ( payload . slice ( runStart , index ) ) ;
94+ runStart = - 1 ;
95+ }
96+ }
97+ if ( runStart !== - 1 ) {
98+ runs . push ( payload . slice ( runStart ) ) ;
99+ }
100+ const base64 = runs . length === 1 ? runs [ 0 ] ! : runs . join ( "" ) ;
101+ if ( base64 . length === 0 || base64 . length % 4 !== 0 ) return null ;
102+ const firstPad = base64 . indexOf ( "=" ) ;
103+ if ( firstPad !== - 1 ) {
104+ // '=' is only valid as one or two trailing padding characters; Node's
105+ // decoder would otherwise silently truncate at the first '='.
106+ if ( base64 . length - firstPad > 2 ) return null ;
107+ for ( let index = firstPad ; index < base64 . length ; index += 1 ) {
108+ if ( base64 . charCodeAt ( index ) !== 0x3d ) return null ;
109+ }
110+ }
56111
57112 return { mimeType, base64 } ;
58113}
0 commit comments