Sanitizer: validate srcset URLs and fail closed without DOMParser - #42840
Sanitizer: validate srcset URLs and fail closed without DOMParser#42840aljojoby9 wants to merge 5 commits into
Conversation
img srcset is in the allowList but was not treated as a URI attribute, so javascript: (and other unsafe) candidates were never checked. Parse srcset per candidate so commas inside data: URLs are not treated as separators. If DOMParser is missing, clobbered, or throws, return an empty string instead of the original markup. Returning the input would skip sanitization.
Silence no-script-url on the srcset XSS fixtures and bump the gzip caps to cover the srcset parser.
ESLint flags the assertion literal, not the template line, and treats an unused disable-next-line as an error under --report-unused-disable-directives.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 363e0fbaea
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const descriptor = rest.match(SRCSET_DESCRIPTOR) | ||
|
|
||
| if (descriptor) { | ||
| urls.push(rest.slice(0, descriptor.index).trim()) | ||
| rest = rest.slice(descriptor.index + descriptor[0].length).replace(/^,/, '').trim() |
There was a problem hiding this comment.
When a data: candidate omits its descriptor, e.g. srcset="data:image/png;base64,AAAA , javascript:alert(1) 2x", this branch searches the entire remaining string for the next 1x/2x descriptor and treats everything before it as a single data: URL. The javascript: candidate is then never passed to isSafeUrl, so the sanitizer keeps a mixed srcset that the browser parses as separate candidates. Stop the data-URL scan at a real candidate comma, including whitespace-before-comma, or fail closed when the boundary is ambiguous.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 718eba2. A data: payload now ends at the first comma-plus-space after the header comma, so a later candidate is parsed on its own and rejected if it is unsafe.
There was a problem hiding this comment.
Pull request overview
This PR hardens the HTML sanitizer by ensuring img[srcset] is treated as a URI-bearing attribute and that sanitizer operation fails closed when DOMParser is unavailable or throws, preventing legacy DOM-clobbering fail-open behavior.
Changes:
- Validate each
srcsetcandidate URL againstSAFE_URL_PATTERNrather than allowing rawsrcsetthrough. - Fail closed by returning an empty string when
DOMParseris missing, clobbered, or throws during parsing. - Add unit tests covering safe/unsafe
srcsetandDOMParserfailure scenarios.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| js/src/util/sanitizer.js | Adds srcset to URI attributes, parses/validates srcset candidates, and changes sanitizer to return '' when DOMParser can’t be used. |
| js/tests/unit/util/sanitizer.spec.js | Adds unit tests for srcset sanitization behavior and DOMParser failure handling. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
| if (/^data:/i.test(rest)) { | ||
| const descriptor = rest.match(SRCSET_DESCRIPTOR) | ||
|
|
||
| if (descriptor) { | ||
| urls.push(rest.slice(0, descriptor.index).trim()) | ||
| rest = rest.slice(descriptor.index + descriptor[0].length).replace(/^,/, '').trim() | ||
| } else { | ||
| urls.push(rest) | ||
| rest = '' | ||
| } | ||
|
|
||
| continue | ||
| } |
There was a problem hiding this comment.
Fixed in 718eba2. extractSrcsetUrls no longer takes the first 1x/2x/w in the remainder as belonging to the current data: URL.
| it('should keep a data-URI srcset whose commas belong to the payload', () => { | ||
| const dataUrl = 'data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/' | ||
| const template = `<img src="safe.jpg" srcset="${dataUrl} 1x">` | ||
|
|
||
| const result = sanitizeHtml(template, DefaultAllowlist, null) | ||
|
|
||
| expect(result).toContain('srcset=') | ||
| expect(result).toContain('data:image/png;base64,') | ||
| }) | ||
|
|
There was a problem hiding this comment.
Added a regression test for data:..., javascript:alert(1) 2x in 718eba2.
A descriptor on a later candidate must not extend a data: URL that has no descriptor of its own. Split on a comma-plus-space after the data payload so javascript: (and other) candidates are checked separately.
CI measured a few hundred extra gzipped bytes over the previous caps. Raise the limits by a quarter to half kilobyte so a later comment or test does not fail bundlewatch again.
Description
DefaultAllowlistpermitssrcsetonimg, butsrcsetwas not treated as a URI attribute. The sanitizer checkedsrc/hrefagainstSAFE_URL_PATTERNand leftsrcsetuntouched, so ajavascript:candidate insrcsetwas kept.Checking the raw
srcsetstring as one URI is not enough. The value is a list of candidates, anddata:URLs contain commas. This change parses each candidate (without splitting inside adata:payload) and rejects the attribute if any URL failsSAFE_URL_PATTERN.If
DOMParseris missing, clobbered, or throws, return an empty string. Do not return the original markup. That fail-open path is how older Bootstrap skipped sanitization after DOM clobbering.Changes
srcsetcandidate before allowing the attributeDOMParsercannot be usedRelated
data:/vbscript:in single-URL attributes). This PR covers the list-valuedsrcsetpath and parser failure, not scheme blocking onhref/src.