mirror of
https://github.com/nodejs/node.git
synced 2024-11-21 10:59:27 +00:00
3800d60c66
This reverts commit dd8eeec3f0
.
PR-URL: https://github.com/nodejs/node/pull/54588/
Refs: https://github.com/nodejs/node/pull/54524
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
137 lines
4.4 KiB
JavaScript
137 lines
4.4 KiB
JavaScript
'use strict';
|
|
|
|
const common = require('../common');
|
|
const assert = require('assert');
|
|
|
|
[-1, 10].forEach((offset) => {
|
|
assert.throws(
|
|
() => Buffer.alloc(9).write('foo', offset),
|
|
{
|
|
code: 'ERR_OUT_OF_RANGE',
|
|
name: 'RangeError',
|
|
message: 'The value of "offset" is out of range. ' +
|
|
`It must be >= 0 && <= 9. Received ${offset}`
|
|
}
|
|
);
|
|
});
|
|
|
|
const resultMap = new Map([
|
|
['utf8', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['ucs2', Buffer.from([102, 0, 111, 0, 111, 0, 0, 0, 0])],
|
|
['ascii', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['latin1', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['binary', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['utf16le', Buffer.from([102, 0, 111, 0, 111, 0, 0, 0, 0])],
|
|
['base64', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['base64url', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
['hex', Buffer.from([102, 111, 111, 0, 0, 0, 0, 0, 0])],
|
|
]);
|
|
|
|
// utf8, ucs2, ascii, latin1, utf16le
|
|
const encodings = ['utf8', 'utf-8', 'ucs2', 'ucs-2', 'ascii', 'latin1',
|
|
'binary', 'utf16le', 'utf-16le'];
|
|
|
|
encodings
|
|
.reduce((es, e) => es.concat(e, e.toUpperCase()), [])
|
|
.forEach((encoding) => {
|
|
const buf = Buffer.alloc(9);
|
|
const len = Buffer.byteLength('foo', encoding);
|
|
assert.strictEqual(buf.write('foo', 0, len, encoding), len);
|
|
|
|
if (encoding.includes('-'))
|
|
encoding = encoding.replace('-', '');
|
|
|
|
assert.deepStrictEqual(buf, resultMap.get(encoding.toLowerCase()));
|
|
});
|
|
|
|
// base64
|
|
['base64', 'BASE64', 'base64url', 'BASE64URL'].forEach((encoding) => {
|
|
const buf = Buffer.alloc(9);
|
|
const len = Buffer.byteLength('Zm9v', encoding);
|
|
|
|
assert.strictEqual(buf.write('Zm9v', 0, len, encoding), len);
|
|
assert.deepStrictEqual(buf, resultMap.get(encoding.toLowerCase()));
|
|
});
|
|
|
|
// hex
|
|
['hex', 'HEX'].forEach((encoding) => {
|
|
const buf = Buffer.alloc(9);
|
|
const len = Buffer.byteLength('666f6f', encoding);
|
|
|
|
assert.strictEqual(buf.write('666f6f', 0, len, encoding), len);
|
|
assert.deepStrictEqual(buf, resultMap.get(encoding.toLowerCase()));
|
|
});
|
|
|
|
// Invalid encodings
|
|
for (let i = 1; i < 10; i++) {
|
|
const encoding = String(i).repeat(i);
|
|
const error = common.expectsError({
|
|
code: 'ERR_UNKNOWN_ENCODING',
|
|
name: 'TypeError',
|
|
message: `Unknown encoding: ${encoding}`
|
|
});
|
|
|
|
assert.ok(!Buffer.isEncoding(encoding));
|
|
assert.throws(() => Buffer.alloc(9).write('foo', encoding), error);
|
|
}
|
|
|
|
// UCS-2 overflow CVE-2018-12115
|
|
for (let i = 1; i < 4; i++) {
|
|
// Allocate two Buffers sequentially off the pool. Run more than once in case
|
|
// we hit the end of the pool and don't get sequential allocations
|
|
const x = Buffer.allocUnsafe(4).fill(0);
|
|
const y = Buffer.allocUnsafe(4).fill(1);
|
|
// Should not write anything, pos 3 doesn't have enough room for a 16-bit char
|
|
assert.strictEqual(x.write('ыыыыыы', 3, 'ucs2'), 0);
|
|
// CVE-2018-12115 experienced via buffer overrun to next block in the pool
|
|
assert.strictEqual(Buffer.compare(y, Buffer.alloc(4, 1)), 0);
|
|
}
|
|
|
|
// Should not write any data when there is no space for 16-bit chars
|
|
const z = Buffer.alloc(4, 0);
|
|
assert.strictEqual(z.write('\u0001', 3, 'ucs2'), 0);
|
|
assert.strictEqual(Buffer.compare(z, Buffer.alloc(4, 0)), 0);
|
|
// Make sure longer strings are written up to the buffer end.
|
|
assert.strictEqual(z.write('abcd', 2), 2);
|
|
assert.deepStrictEqual([...z], [0, 0, 0x61, 0x62]);
|
|
|
|
// Large overrun could corrupt the process
|
|
assert.strictEqual(Buffer.alloc(4)
|
|
.write('ыыыыыы'.repeat(100), 3, 'utf16le'), 0);
|
|
|
|
{
|
|
// .write() does not affect the byte after the written-to slice of the Buffer.
|
|
// Refs: https://github.com/nodejs/node/issues/26422
|
|
const buf = Buffer.alloc(8);
|
|
assert.strictEqual(buf.write('ыы', 1, 'utf16le'), 4);
|
|
assert.deepStrictEqual([...buf], [0, 0x4b, 0x04, 0x4b, 0x04, 0, 0, 0]);
|
|
}
|
|
|
|
{
|
|
const buf = Buffer.alloc(1);
|
|
assert.strictEqual(buf.write('ww'), 1);
|
|
assert.strictEqual(buf.toString(), 'w');
|
|
}
|
|
|
|
assert.throws(() => {
|
|
const buf = Buffer.alloc(1);
|
|
assert.strictEqual(buf.asciiWrite('ww', 0, -1));
|
|
assert.strictEqual(buf.latin1Write('ww', 0, -1));
|
|
assert.strictEqual(buf.utf8Write('ww', 0, -1));
|
|
}, common.expectsError({
|
|
code: 'ERR_BUFFER_OUT_OF_BOUNDS',
|
|
}));
|
|
|
|
|
|
assert.throws(() => {
|
|
Buffer.alloc(1).asciiWrite('ww', 0, 2);
|
|
}, common.expectsError({
|
|
code: 'ERR_BUFFER_OUT_OF_BOUNDS',
|
|
}));
|
|
|
|
assert.throws(() => {
|
|
Buffer.alloc(1).asciiWrite('ww', 1, 1);
|
|
}, common.expectsError({
|
|
code: 'ERR_BUFFER_OUT_OF_BOUNDS',
|
|
}));
|