GHSA-4C8J-MGM4-QQVP

Vulnerability from github – Published: 2026-06-26 19:26 – Updated: 2026-06-26 19:26
VLAI
Summary
Remark42: Cross-Site Scripting (XSS) on /api/v1/img via content-type spoofing
Details

Summary

The remark42 image proxy fetches an arbitrary remote URL and re-serves the response from remark42's own origin. The download path decides whether the fetched resource is an image by looking only at the Content-Type header the remote server claims — it never inspects the actual bytes. The serving path then derives the response Content-Type by sniffing those bytes with http.DetectContentType.

An attacker hosts a URL that sets Content-Type to image/png but returns an HTML/JavaScript body:

  • the download check sees image/png → accepts it;
  • the serve path sniffs the body → emits Content-Type: text/html;
  • the browser renders attacker HTML/JS as a document in remark42's origin.

Details

Downloader

backend/app/rest/proxy/image.go — downloadImage(), lines 189-206:

contentType := resp.Header.Get("Content-Type")
if !strings.HasPrefix(contentType, "image/") {
    return nil, fmt.Errorf("invalid content type %s", contentType)
}

maxSize := 5 * 1024 * 1024 // 5MB default
if p.ImageService != nil && p.ImageService.MaxSize > 0 {
    maxSize = p.ImageService.MaxSize
}
lr := io.LimitReader(resp.Body, int64(maxSize)+1)
imgData, err := io.ReadAll(lr)
if err != nil {
    return nil, fmt.Errorf("unable to read image body: %w", err)
}
if len(imgData) > maxSize {
    return nil, fmt.Errorf("image is too large")
}
return imgData, nil          // <-- bytes never validated, returned as-is

Send Content-Type: image/png and the check passes regardless of what the body actually contains.

Server

backend/app/rest/proxy/image.go — Handler(), line 131:

w.Header().Add("Content-Type", p.ImageService.ImgContentType(img))
_, err = io.Copy(w, bytes.NewReader(img))

backend/app/store/image/image.go — ImgContentType(), lines 242-249:

func (s *Service) ImgContentType(img []byte) string {
    contentType := http.DetectContentType(img)
    if contentType == "application/octet-stream" {
        return "image/*"
    }
    return contentType                 // <-- returns text/html for an HTML body
}

PoC

self.send_response(200)
self.send_header("Content-Type", "image/png")
self.send_header("Content-Length", str(len(body)))
self.end_headers()
self.wfile.write(body)            # body = <!DOCTYPE html><script>...</script>

Then have the victim open https://<remark42-host>/api/v1/img?src=<base64(attacker-host)> top-level.

Impact

  • The script can issue authenticated, same-origin API calls with credentials: 'include' — the JWT cookie is sent automatically.
  • The script can read the XSRF-TOKEN cookie and re-send it as the X-XSRF-TOKEN header, defeating CSRF protection. The attacker acts as the victim: delete/edit their comments, change their settings, and — if the victim is admin — perform admin actions.

Triggering requires no remark42 account on the target instance; the attacker only needs to host the malicious upstream URL and deliver the proxy link to a victim by any means (email, DM, link on another site, etc.).

Fix

v1.16.0 adds layered defense to /api/v1/img and /api/v1/picture/{user}/{id}:

  • rest.SafeImgContentType validates sniffed body bytes against a strict allowlist (image/png, image/jpeg, image/gif, image/webp, image/bmp, image/x-icon). Non-image content returns 415 with no body echo. SVG is implicitly excluded.
  • Every response carries Content-Security-Policy: default-src 'none'; sandbox; frame-ancestors 'none', X-Content-Type Options: nosniff, and Content-Disposition: inline; filename="image".
  • The ETag is bumped to "v2:<base64(src)>". Browsers that revalidate cached pre-fix responses get a fresh validated 200 instead of a 304 against the poisoned cached entry.
  • The strict default-src 'none'; sandbox CSP also applies to all /api/v1/* routes as defense-in-depth.

Residual exposure

Browser-local caches that already hold a pre-fix text/html response with Cache-Control: max-age=2592000 keep serving it from local store until the TTL expires or the cache is evicted under memory pressure. The ETag bump only reaches clients that revalidate during the cached lifetime. Operators running a CDN/edge cache in front of remark42 should purge /api/v1/img after deploying v1.16.0.

Severity
8.2 (High)
CVSS:3.0/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:L/A:N
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "Go",
        "name": "github.com/umputun/remark42"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "1.6.0"
            },
            {
              "fixed": "1.16.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-48788"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-436",
      "CWE-79"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-06-26T19:26:43Z",
    "nvd_published_at": "2026-06-17T13:20:43Z",
    "severity": "HIGH"
  },
  "details": "### Summary\nThe remark42 image proxy fetches an arbitrary remote URL and re-serves the response from remark42\u0027s own origin. The download path decides whether the fetched resource is an image by looking only at the `Content-Type` header the remote server claims \u2014 it never inspects the actual bytes. The serving path then derives the response `Content-Type` by sniffing those bytes with `http.DetectContentType`.\n\nAn attacker hosts a URL that sets `Content-Type` to `image/png` but returns an HTML/JavaScript body:\n\n* the download check sees `image/png` \u2192 accepts it;\n* the serve path sniffs the body \u2192 emits `Content-Type: text/html`;\n* the browser renders attacker HTML/JS as a document in remark42\u0027s origin.\n\n### Details\n#### Downloader\n\n`backend/app/rest/proxy/image.go` \u2014 `downloadImage()`, lines 189-206:\n\n```go\ncontentType := resp.Header.Get(\"Content-Type\")\nif !strings.HasPrefix(contentType, \"image/\") {\n    return nil, fmt.Errorf(\"invalid content type %s\", contentType)\n}\n\nmaxSize := 5 * 1024 * 1024 // 5MB default\nif p.ImageService != nil \u0026\u0026 p.ImageService.MaxSize \u003e 0 {\n    maxSize = p.ImageService.MaxSize\n}\nlr := io.LimitReader(resp.Body, int64(maxSize)+1)\nimgData, err := io.ReadAll(lr)\nif err != nil {\n    return nil, fmt.Errorf(\"unable to read image body: %w\", err)\n}\nif len(imgData) \u003e maxSize {\n    return nil, fmt.Errorf(\"image is too large\")\n}\nreturn imgData, nil          // \u003c-- bytes never validated, returned as-is\n```\n\nSend `Content-Type: image/png` and the check passes regardless of what the body actually contains.\n\n#### Server\n\n`backend/app/rest/proxy/image.go` \u2014 `Handler()`, line 131:\n\n```go\nw.Header().Add(\"Content-Type\", p.ImageService.ImgContentType(img))\n_, err = io.Copy(w, bytes.NewReader(img))\n```\n\n`backend/app/store/image/image.go` \u2014 `ImgContentType()`, lines 242-249:\n\n```go\nfunc (s *Service) ImgContentType(img []byte) string {\n    contentType := http.DetectContentType(img)\n    if contentType == \"application/octet-stream\" {\n        return \"image/*\"\n    }\n    return contentType                 // \u003c-- returns text/html for an HTML body\n}\n```\n\n### PoC\n\n```python\nself.send_response(200)\nself.send_header(\"Content-Type\", \"image/png\")\nself.send_header(\"Content-Length\", str(len(body)))\nself.end_headers()\nself.wfile.write(body)            # body = \u003c!DOCTYPE html\u003e\u003cscript\u003e...\u003c/script\u003e\n```\n\nThen have the victim open `https://\u003cremark42-host\u003e/api/v1/img?src=\u003cbase64(attacker-host)\u003e` top-level.\n\n### Impact\n* The script can issue authenticated, same-origin API calls with `credentials: \u0027include\u0027` \u2014 the JWT cookie is sent automatically.\n* The script can read the `XSRF-TOKEN` cookie and re-send it as the `X-XSRF-TOKEN` header, defeating CSRF protection. The attacker acts as the victim: delete/edit their comments, change their settings, and \u2014 if the victim is admin \u2014 perform admin actions.\n\nTriggering requires no remark42 account on the target instance; the attacker only needs to host the malicious upstream URL and deliver the proxy link to a victim by any means (email, DM, link on another site, etc.).\n\n### Fix\n\n`v1.16.0` adds layered defense to `/api/v1/img` and `/api/v1/picture/{user}/{id}`:\n\n* `rest.SafeImgContentType` validates sniffed body bytes against a strict allowlist (`image/png`, `image/jpeg`, `image/gif`, `image/webp`, `image/bmp`, `image/x-icon`). Non-image content returns `415` with no body echo. SVG is implicitly excluded.\n* Every response carries `Content-Security-Policy: default-src \u0027none\u0027; sandbox; frame-ancestors \u0027none\u0027`, `X-Content-Type Options: nosniff`, and `Content-Disposition: inline; filename=\"image\"`.\n* The ETag is bumped to `\"v2:\u003cbase64(src)\u003e\"`. Browsers that revalidate cached pre-fix responses get a fresh validated `200` instead of a `304` against the poisoned cached entry.\n* The strict `default-src \u0027none\u0027; sandbox` CSP also applies to all `/api/v1/*` routes as defense-in-depth.\n\n#### Residual exposure\n\nBrowser-local caches that already hold a pre-fix `text/html` response with `Cache-Control: max-age=2592000` keep serving it from local store until the TTL expires or the cache is evicted under memory pressure. The ETag bump only reaches clients that revalidate during the cached lifetime. Operators running a CDN/edge cache in front of remark42 should purge `/api/v1/img` after deploying `v1.16.0`.",
  "id": "GHSA-4c8j-mgm4-qqvp",
  "modified": "2026-06-26T19:26:44Z",
  "published": "2026-06-26T19:26:43Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/umputun/remark42/security/advisories/GHSA-4c8j-mgm4-qqvp"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-48788"
    },
    {
      "type": "WEB",
      "url": "https://github.com/umputun/remark42/commit/78d6de6bce1e961f023969da3ec8a00dd80c9ae8"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/umputun/remark42"
    },
    {
      "type": "WEB",
      "url": "https://github.com/umputun/remark42/releases/tag/v1.16.0"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.0/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:L/A:N",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Remark42: Cross-Site Scripting (XSS) on /api/v1/img via content-type spoofing"
}


Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.


Detection rules are retrieved from Rulezet.