fix: prevent markdown injection in notifications (#28340)

First of two PRs.

Notification title and body templates are Markdown authored by Coder,
but the label values interpolated into them are user-controlled and were
substituted through `text/template`, which does no escaping. Those
values arrive from user profile fields and from OIDC/GitHub name claims.

This PR:

- Neutralizes Markdown structure in label, data, and `UserName` values
before they reach the template. Applied in `notifier.prepare`, so
already-queued messages are covered and the stored payload keeps its
original values for webhook consumers. Nested `.Data` map keys are
escaped too: one shipped template prints a key, and those keys are
Terraform resource addresses.
- Narrows the notification Markdown grammar to what templates actually
use. `CommonExtensions` enabled Tables, DefinitionLists and MathJax,
each openable from a value and used by no template. Autolink stays off
so a URL in a value cannot become an anchor.
- Enables `html.Safelink`, restricting generated hrefs to safe schemes,
and guards the panic it exposes: `parser.IsSafeURL` slices a destination
before bounds-checking it, so `[docs]()` crashed both renderers.
- Folds line breaks out of the `Subject:` header and encodes it, fixing
a pre-existing RFC 2047 violation for non-ASCII subjects, a forged
encoded-word that let a value choose the displayed subject, and headers
running past RFC 5322's 998-octet line limit.

Escaping is narrow on purpose, split by where each character carries
meaning:

- `` \[]()!<` `` everywhere. Backtick is in this group because a fenced
block's info string is an HTML sink: gomarkdown writes it into
`class="language-..."` unescaped, and `SkipHTML` does not apply to a
`CodeBlock` node, so a value that closes the attribute and the tag
injects live markup.
- `#-+.>|` only in leading position, so values like `bobby-workspace`
and `1.5` are untouched.
- `=`, `~` and `:` are not escapable by both renderers, so the preceding
line break is folded instead. `:` opens a definition list and a GFM
table delimiter row that escaping `|` cannot reach. A value's *first*
line has no preceding break to fold, so a real tilde fence or `===`
underline is escaped there instead, accepting a visible backslash: an
unterminated `~~~` at the start of a title otherwise renders the
Subject, `<title>` and heading empty.
- Leading indentation is truncated to three spaces. Four open an
indented code block and a space has no escape.
- Emphasis characters are left alone. Escaping `_` corrupts label values
such as `user_override` that body templates compare with `eq`, which
silently drops content from the rendered email.

One golden file changes: the resource replacements `body_markdown` now
reads `docker_container\[0\]`, from the map-key escaping above. Every
other golden is byte-identical.

**One known residual**, pinned by a test that fails if it closes:
CommonMark does not process escapes inside a code span, so where a
template wraps a value in one, as the workspace out-of-disk body does,
the escaper's own backslashes reach the reader. That depends on where
the value lands rather than what it contains, which a pre-render escaper
cannot see. This narrows the class rather than closing it.

**#28397 completes the fix and is stacked on this branch. This PR should
not merge without it.**

Escaping here cannot reach the SMTP HTML template's sinks, by design
rather than by oversight: the subject is produced by
`PlaintextFromMarkdown`, which strips exactly the backslashes added
here, and `html.gotmpl` then interpolated the result through
`text/template`. On this branch alone, a label value still reaches the
Subject, `<title>` and `<h1>` as live markup. #28397 escapes at those
sinks with `| html`, which is the only place the information needed to
escape correctly exists.
This commit is contained in:
Bobby Ho
2026-08-25 08:12:00 -07:00
committed by GitHub
parent be9f8592eb
commit 9e8075db0f
13 changed files with 1489 additions and 7 deletions
+188
View File
@@ -0,0 +1,188 @@
package render
import "strings"
// Character classes for EscapeMarkdown, split by where each character carries
// structural meaning. The split is what keeps escaping away from the enum-like
// label values that body templates compare with `eq`, such as "user_override",
// "bobby-workspace" and "1.5": escaping one changes template control flow.
const (
// inlineCritical characters can produce a link, an image, an angle autolink,
// or forge an escape from anywhere in a value, so they are always escaped.
//
// Backtick is here, not in blockStart, because a fence's info string is an
// HTML sink: gomarkdown writes it into class="language-..." unescaped, and
// SkipHTML does not apply to a CodeBlock node. Escaping only the leading
// backtick would leave two, which open an inline code span.
inlineCritical = "\\[]()!<`"
// blockStart characters carry structural meaning only as the first
// non-space character of a line, so they are escaped only there.
blockStart = `#-+.>|`
// leadingEmphasis characters carry inline meaning anywhere but also open a
// block construct in leading position: "* " starts a bullet list, and three
// or more of either character starts a thematic break. Escaping them only
// there costs emphasis that begins on a line boundary.
leadingEmphasis = `*_`
// foldStart characters also carry meaning only at the start of a line, but
// glamour does not honor a backslash before them, so escaping would leave a
// literal backslash in the plaintext part. The preceding line break becomes
// a space instead, denying them the line-start position.
//
// ":" is here for the GFM delimiter row ":-- | --:" as well as definition
// lists. Escaping "|" does not reach that row: its pipes are mid-line.
foldStart = `=~:`
// maxLeadingSpaces is the widest indentation a line may keep, since four
// spaces open an indented code block and a space cannot be escaped.
maxLeadingSpaces = 3
)
// EscapeMarkdown neutralizes Markdown structure in an untrusted value so that it
// renders as literal text through both HTMLFromNotificationMarkdown and
// PlaintextFromMarkdown. Line breaks are preserved so multi-line values keep
// their shape. Other control characters are dropped, being the carrier for SMTP
// header injection.
//
// Known residual: a template that wraps the value in a code span. CommonMark
// does not process escapes inside one, so the backslashes emitted here reach
// the reader. Nothing here can detect that, since the sink is decided after
// this runs.
func EscapeMarkdown(s string) string {
if s == "" {
return s
}
lines := strings.Split(stripControl(s), "\n")
var b strings.Builder
b.Grow(len(s) + len(s)/8)
for i, line := range lines {
if i > 0 {
// Joining a fold-start line to the previous one takes it out of
// leading position.
if opensFoldConstruct(line) {
_ = b.WriteByte(' ')
} else {
_ = b.WriteByte('\n')
}
}
// The first line has no preceding break to fold, so escaping is the only
// lever left there.
_, _ = b.WriteString(escapeLine(line, i == 0 && isLeadingFoldConstruct(line)))
}
return b.String()
}
// stripControl keeps line breaks, turns the other whitespace controls into
// spaces and drops the rest. Carriage returns are folded rather than kept so a
// value cannot terminate an SMTP header.
func stripControl(s string) string {
if strings.IndexFunc(s, isStrippable) < 0 {
return s
}
var b strings.Builder
b.Grow(len(s))
for _, r := range s {
switch {
case r == '\n':
_, _ = b.WriteRune(r)
case r == '\r' || r == '\t' || r == '\v' || r == '\f':
_, _ = b.WriteRune(' ')
case r < 0x20 || r == 0x7f:
// Dropped.
default:
_, _ = b.WriteRune(r)
}
}
return b.String()
}
func isStrippable(r rune) bool {
return r != '\n' && (r < 0x20 || r == 0x7f)
}
// escapeLine escapes one line's structural characters and truncates its
// indentation to maxLeadingSpaces.
//
// escapeFold additionally escapes a leading "=" or "~". Only EscapeMarkdown's
// first line passes it, and only when that line really is a fold construct.
func escapeLine(line string, escapeFold bool) string {
var b strings.Builder
b.Grow(len(line))
leading := true
spaces := 0
// digitRun reports whether the line so far is nothing but indentation and
// digits, which is the only position where "." opens an ordered list.
digitRun := false
for i, r := range line {
switch {
case r < 0x80 && strings.ContainsRune(inlineCritical, r):
_ = b.WriteByte('\\')
_, _ = b.WriteRune(r)
case leading && r == ' ':
// Indentation keeps the next character in leading position.
if spaces < maxLeadingSpaces {
_, _ = b.WriteRune(r)
spaces++
}
continue
case leading && r < 0x80 && strings.ContainsRune(blockStart+leadingEmphasis, r):
_ = b.WriteByte('\\')
_, _ = b.WriteRune(r)
case leading && escapeFold && (r == '=' || r == '~'):
_ = b.WriteByte('\\')
_, _ = b.WriteRune(r)
case digitRun && r == '.' && closesMarker(line, i):
// The "1." of an ordered list. Its sibling "1)" needs no case
// because ")" is inlineCritical and is always escaped.
_ = b.WriteByte('\\')
_, _ = b.WriteRune(r)
default:
_, _ = b.WriteRune(r)
}
digitRun = (leading || digitRun) && r >= '0' && r <= '9'
leading = false
}
return b.String()
}
// closesMarker reports whether the single-byte list-marker delimiter at i is
// followed by a space or ends the line, as CommonMark requires of a marker.
// That requirement is what keeps a value such as "1.5" out of the escaped set.
// Tabs need no handling: stripControl has already folded them into spaces.
func closesMarker(line string, i int) bool {
return i+1 == len(line) || line[i+1] == ' '
}
// opensFoldConstruct reports whether a line's first non-space character is a
// foldStart character. Approximate on purpose: it only decides whether to drop
// a line break, which costs nothing.
func opensFoldConstruct(line string) bool {
t := strings.TrimLeft(line, " ")
if t == "" {
return false
}
return strings.ContainsRune(foldStart, rune(t[0]))
}
// isLeadingFoldConstruct reports whether a line is itself a tilde fence opener
// or a Setext "=" underline, rather than merely starting with one of those
// characters. Exact, because it governs escaping and the backslash is visible:
// "=> next" must not acquire one, while "~~~" must, since an unterminated fence
// at the start of a title renders the Subject, <title> and heading empty.
//
// Indentation is ignored: escapeLine truncates it to maxLeadingSpaces, which
// still leaves the line able to open a block. ":" is excluded because a
// definition list or table needs a preceding line that a first line lacks.
func isLeadingFoldConstruct(line string) bool {
t := strings.TrimLeft(line, " ")
if strings.HasPrefix(t, "~~~") {
return true
}
t = strings.TrimRight(t, " ")
return t != "" && strings.Trim(t, "=") == ""
}
+313
View File
@@ -0,0 +1,313 @@
package render
import (
"fmt"
"strings"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// asciiPunctuation is every ASCII punctuation character, the set CommonMark
// declares escapable.
const asciiPunctuation = "!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~"
// TestEscapableSet pins the escapes each renderer honors, which is what
// EscapeMarkdown's character classes are derived from.
func TestEscapableSet(t *testing.T) {
t.Parallel()
// Measured against gomarkdown and glamour as vendored today.
const (
wantHTML = "!#$&()*+-.:<>[\\]^_`{|}~"
wantPlain = "!#()*+-.<>[\\]_`{|}"
)
var gotHTML, gotPlain strings.Builder
for _, r := range asciiPunctuation {
escaped := `X\` + string(r) + `Y`
// HTML escapes markup characters, so compare the entity form.
wantLiteral := "X" + string(r) + "Y"
switch r {
case '<':
wantLiteral = "X&lt;Y"
case '>':
wantLiteral = "X&gt;Y"
case '&':
wantLiteral = "X&amp;Y"
case '"':
wantLiteral = "X&quot;Y"
}
html := strings.TrimSuffix(strings.TrimPrefix(HTMLFromMarkdown(escaped), "<p>"), "</p>")
if html == wantLiteral {
_, _ = gotHTML.WriteRune(r)
}
plain, err := PlaintextFromMarkdown(escaped)
require.NoError(t, err)
if plain == "X"+string(r)+"Y" {
_, _ = gotPlain.WriteRune(r)
}
}
require.Equal(t, wantHTML, gotHTML.String(),
"the set of characters gomarkdown honors as escapes has changed; re-derive EscapeMarkdown's character classes")
require.Equal(t, wantPlain, gotPlain.String(),
"the set of characters glamour honors as escapes has changed; re-derive EscapeMarkdown's character classes")
// Every character EscapeMarkdown escapes must be honored by both renderers.
for _, r := range inlineCritical + blockStart + leadingEmphasis {
assert.Contains(t, wantHTML, string(r), "gomarkdown does not honor \\%s", string(r))
assert.Contains(t, wantPlain, string(r), "glamour does not honor \\%s", string(r))
}
// foldStart characters are folded precisely because they are not escapable.
for _, r := range foldStart {
assert.NotContains(t, wantPlain, string(r),
"glamour now honors \\%s, so it could be escaped instead of folded", string(r))
}
}
// TestEscapeMarkdownControlValues guards the label values body templates compare
// with `eq`. Escaping one silently changes template control flow.
func TestEscapeMarkdownControlValues(t *testing.T) {
t.Parallel()
for _, v := range []string{
"user_override", // migrations/000553, {{if eq .Labels.limit_source "user_override"}}
"service", // migrations/000568, {{if eq .Labels.account_type "service"}}
"0", // migrations/000480, {{if eq .Data.retention_days "0"}}
"autobuild",
"initiator",
"user-override",
"1.5",
"10.0.0.1",
"bobby-workspace",
} {
require.Equal(t, v, EscapeMarkdown(v), "escaping changed a control value")
}
}
func TestEscapeMarkdown(t *testing.T) {
t.Parallel()
// Emphasis and code tags are accepted residuals: no destination.
structuralTags := []string{
"<a ", "<img ", "<h1", "<h2", "<h3", "<h4", "<h5", "<h6",
"<ul", "<ol", "<hr", "<blockquote", "<table",
}
t.Run("NeutralisesStructure", func(t *testing.T) {
t.Parallel()
type structureCase struct {
name string
value string
// inertRaw marks a value neutralized by something other than marker
// escaping. Leaving it unset on one fails the liveness check below.
inertRaw bool
}
for _, tc := range []structureCase{
{name: "DisclosurePayload", value: "Eve\n## URGENT: SSO certificate expiring\n[Re-authenticate now](https://coder-sso.attacker.example/login)"},
{name: "InlineLink", value: "[Re-authenticate now](https://attacker.example/login)"},
// A link reference definition is not recognized mid-paragraph.
{name: "ReferenceLink", value: "[Re-auth][1]\n\n[1]: https://attacker.example", inertRaw: true},
{name: "Image", value: "![px](https://tracker.attacker.example/p.gif)"},
{name: "AngleAutolink", value: "Eve <https://attacker.example>"},
// Neutralized by autolinking being off. See
// TestEscapeMarkdownNoAutolink.
{name: "BareURL", value: "Eve https://attacker.example/login", inertRaw: true},
{name: "Mailto", value: "Eve mailto:eve@attacker.example", inertRaw: true},
{name: "ATXHeading", value: "Eve\n## URGENT"},
{name: "SetextH1", value: "URGENT: re-auth required\n===\nx"},
{name: "SetextH1Spaced", value: "Eve\n=== \n#### x"},
{name: "SetextH1Repeated", value: "Eve\n===\n===\nx"},
{name: "SetextH2", value: "Eve\n---\nx"},
{name: "ThematicBreak", value: "Eve\n----\nx"},
{name: "ThematicBreakStars", value: "Eve\n***\nnext"},
{name: "ThematicBreakUnderscores", value: "Eve\n___\nnext"},
{name: "ThematicBreakSpacedStars", value: "Eve\n* * *\nnext"},
{name: "ThematicBreakSpacedUnderscores", value: "Eve\n_ _ _\nnext"},
{name: "BulletList", value: "Eve\n- one\n- two"},
{name: "BulletListStar", value: "Eve\n* one\n* two"},
{name: "BulletListStarIndented", value: "Eve\n * one\n * two"},
{name: "OrderedList", value: "Eve\n1. one\n2. two"},
{name: "OrderedListMultiDigit", value: "Eve\n99. one\n100. two"},
{name: "OrderedListParen", value: "Eve\n1) one\n2) two"},
{name: "Blockquote", value: "Eve\n> quoted"},
// Neutralized by the Tables extension being off; the escaper's own
// handling is covered by TestEscapeMarkdownColon.
{name: "Table", value: "a | b\n--- | ---\nc | d", inertRaw: true},
// Neutralized by the safelink policy. See
// TestEscapeMarkdownNoAutolink/SafelinkRejectsUnsafeSchemes.
{name: "JavascriptScheme", value: "[click](javascript:alert(1))", inertRaw: true},
// Asserts escaping does not re-enable the value's own backslashes.
{name: "EscapeForging", value: `Eve \[Re-auth\](https://attacker.example)`, inertRaw: true},
} {
t.Run(tc.name, func(t *testing.T) {
t.Parallel()
// Rendered twice, the second with line breaks doubled: only a
// blank line lets a block construct interrupt a paragraph.
rawProducedTag := false
for _, value := range []string{tc.value, strings.ReplaceAll(tc.value, "\n", "\n\n")} {
escaped := EscapeMarkdown(value)
html := HTMLFromNotificationMarkdown(suspendedBody(escaped))
plain, err := PlaintextFromMarkdown(suspendedBody(escaped))
require.NoError(t, err)
raw := HTMLFromNotificationMarkdown(suspendedBody(value))
for _, tag := range structuralTags {
assert.NotContains(t, html, tag, "value %q rendered HTML: %s", value, html)
rawProducedTag = rawProducedTag || strings.Contains(raw, tag)
}
// A backslash the value did not contain means a character the
// renderer does not honor was escaped.
if !strings.Contains(value, `\`) {
assert.NotContains(t, html, `\`, "value %q leaked a literal backslash into HTML", value)
assert.NotContains(t, plain, `\`, "value %q leaked a literal backslash into plaintext", value)
}
}
// Without this, a row whose value can never reach a line-start
// position passes whether or not EscapeMarkdown runs.
if !tc.inertRaw {
assert.True(t, rawProducedTag,
"vacuous row: %q produces no structural tag even unescaped, so the assertions above guard nothing; fix the value or set inertRaw with a reason",
tc.value)
}
})
}
})
t.Run("PreservesBenignValues", func(t *testing.T) {
t.Parallel()
// Also why this change leaves the notification golden files untouched.
for _, value := range []string{
"William Tables",
"bobby-workspace",
"Bobby's Template",
"O'Brien-Smith (Eng) 100%",
"autodeleted due to dormancy (autobuild)",
"José Müller 日本語",
// Documented multi-line custom notification, see
// docs/admin/monitoring/notifications/index.md.
"Test results:\n • ✅ success",
"Test results:\n • ❌ failed (3 tests failed)",
} {
t.Run(value, func(t *testing.T) {
t.Parallel()
escaped := EscapeMarkdown(value)
require.Equal(t, HTMLFromNotificationMarkdown(suspendedBody(value)), HTMLFromNotificationMarkdown(suspendedBody(escaped)))
wantPlain, err := PlaintextFromMarkdown(suspendedBody(value))
require.NoError(t, err)
gotPlain, err := PlaintextFromMarkdown(suspendedBody(escaped))
require.NoError(t, err)
require.Equal(t, wantPlain, gotPlain)
})
}
})
t.Run("ControlCharacters", func(t *testing.T) {
t.Parallel()
for _, tc := range []struct {
name string
value string
want string
}{
{"KeepsNewlines", "a\nb", "a\nb"},
{"FoldsCarriageReturn", "a\r\nb", "a \nb"},
{"FoldsTab", "a\tb", "a b"},
{"FoldsVerticalTab", "a\vb", "a b"},
{"DropsNul", "a\x00b", "ab"},
{"DropsBell", "a\x07b", "ab"},
{"DropsDelete", "a\x7fb", "ab"},
{"Empty", "", ""},
} {
t.Run(tc.name, func(t *testing.T) {
t.Parallel()
require.Equal(t, tc.want, EscapeMarkdown(tc.value))
})
}
})
t.Run("AngleBracketsAreNeutralised", func(t *testing.T) {
t.Parallel()
// "Ops <ops@example.com>" used to render as a mailto anchor; turning it
// into text is a deliberate behavior change.
html := HTMLFromNotificationMarkdown(suspendedBody(EscapeMarkdown("Ops <ops@example.com>")))
require.NotContains(t, html, "<a ")
require.Contains(t, html, "&lt;ops@example.com&gt;")
})
t.Run("EmphasisIsNotEscaped", func(t *testing.T) {
t.Parallel()
// Mid-line "*" and "_" carry no destination and escaping "_" corrupts
// control values. Backtick reaches the info-string sink, so it is escaped.
require.Equal(t, "Eve *_\\`~", EscapeMarkdown("Eve *_`~"))
// Leading "*" and "_" open a list or thematic break. "~" is denied the
// line-start position by the fold, glamour not honoring "\~".
require.Equal(t, "\\*_\\`~", EscapeMarkdown("*_`~"))
require.Equal(t, "\\_*\\`~", EscapeMarkdown("_*`~"))
})
}
// TestEscapeMarkdownNoAutolink: a URL in an untrusted value must not become an
// anchor, while links in the trusted template markdown keep working.
func TestEscapeMarkdownNoAutolink(t *testing.T) {
t.Parallel()
t.Run("UntrustedValueProducesNoAnchor", func(t *testing.T) {
t.Parallel()
for _, value := range []string{
"Eve https://attacker.example/login",
"Eve [Re-auth](https://attacker.example/login)",
"Eve <https://attacker.example>",
"Eve mailto:eve@attacker.example",
"Eve http://attacker.example",
} {
html := HTMLFromNotificationMarkdown("Account **" + EscapeMarkdown(value) + "** suspended.")
assert.NotContains(t, html, "<a ", "value %q produced an anchor", value)
}
})
t.Run("TrustedTemplateLinksStillRender", func(t *testing.T) {
t.Parallel()
// Shapes taken from shipped notification body templates.
for _, markdown := range []string{
"marked as [**dormant**](https://coder.com/docs/templates/schedule#dormancy-threshold-enterprise) because of x",
"See [the docs](https://coder.com/docs/admin/templates/troubleshooting).",
} {
html := HTMLFromNotificationMarkdown(markdown)
assert.Contains(t, html, `<a href="https://coder.com/docs/`, "trusted link did not render: %s", html)
}
})
t.Run("HTMLFromMarkdownStillAutolinks", func(t *testing.T) {
t.Parallel()
// The shared renderer keeps Autolink, so the OIDC signups-disabled page
// still linkifies. Safelink does now apply; see TestHTMLFromMarkdownSafelink.
require.Contains(t, HTMLFromMarkdown("see https://coder.com/docs"), `<a href="https://coder.com/docs"`)
})
t.Run("SafelinkRejectsUnsafeSchemes", func(t *testing.T) {
t.Parallel()
for _, dest := range []string{"javascript:alert(1)", "data:text/html;base64,PHNjcmlwdD4="} {
html := HTMLFromNotificationMarkdown(fmt.Sprintf("[click](%s)", dest))
assert.NotContains(t, html, "<a ", "unsafe scheme %q was linked", dest)
}
})
}
+321
View File
@@ -0,0 +1,321 @@
package render
import (
"strings"
"testing"
"github.com/gomarkdown/markdown/parser"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
xhtml "golang.org/x/net/html"
)
// permissive is the grammar notificationExtensions used to enable. Tests that
// must exercise the escaper rather than the allowlist render against it.
const permissive = parser.CommonExtensions | parser.HardLineBreak
// suspendedBody mirrors the live TemplateUserAccountSuspended body: the
// untrusted value sits mid-paragraph with trusted text on both sides.
func suspendedBody(value string) string {
return "The account belongs to **" + value + "** and it was suspended by **rob**."
}
// TestEscapeMarkdownFenceInfo covers the info-string sink that made backtick
// inlineCritical: a `"` closes the class attribute and a `>` closes the tag.
func TestEscapeMarkdownFenceInfo(t *testing.T) {
t.Parallel()
const info = `"><a/href="https://attacker.example/login">Click here`
// Each payload needs a line after the closing fence, or the template's
// trailing text lands on it and the fence stops being one.
for name, value := range map[string]string{
"Anchor": "Eve\n\n```" + info + "\nhidden\n```\nmore",
"Image": "Eve\n\n```\"><img/src=x/onerror=alert(1)>\nhidden\n```\nmore",
"Tilde": "Eve\n\n~~~" + info + "\nhidden\n~~~\nmore",
"AtStart": "```" + info + "\nhidden\n```\nmore",
} {
t.Run(name, func(t *testing.T) {
t.Parallel()
html := HTMLFromNotificationMarkdown(suspendedBody(EscapeMarkdown(value)))
assert.NotContains(t, html, `class="language-`,
"an info string reached the class attribute: %s", html)
assert.NotContains(t, html, "<a/", "the info string produced an anchor: %s", html)
assert.NotContains(t, html, "<img/", "the info string produced an image: %s", html)
})
}
// Liveness: unescaped, the anchor payload must reach the sink.
raw := HTMLFromNotificationMarkdown(suspendedBody("Eve\n\n```" + info + "\nhidden\n```\nmore"))
require.Contains(t, raw, `class="language-`,
"vacuous test: the payload no longer reaches the info-string sink even unescaped")
}
// TestEscapeMarkdownColon covers ":", rendered under CommonExtensions so the
// shipped allowlist, which kills these constructs anyway, cannot carry the test.
func TestEscapeMarkdownColon(t *testing.T) {
t.Parallel()
for name, value := range map[string]string{
"DefinitionList": "Term\n: definition",
"TableColonBoth": "a | b\n:-- | --:\nc | d",
"TableColonCentre": "a | b\n:-: | :-:\nc | d",
} {
t.Run(name, func(t *testing.T) {
t.Parallel()
escaped := EscapeMarkdown(value)
html := renderHTML(suspendedBody(escaped), permissive)
plain, err := PlaintextFromMarkdown(suspendedBody(escaped))
require.NoError(t, err)
for _, tag := range []string{"<table", "<dl", "<dt", "<dd"} {
assert.NotContains(t, html, tag, "value %q rendered %s", value, html)
}
assert.NotContains(t, plain, `\`, "folding ':' should not leak a backslash: %q", plain)
// Liveness: unescaped, each value must produce one of those tags.
raw := renderHTML(suspendedBody(value), permissive)
assert.True(t,
strings.Contains(raw, "<table") || strings.Contains(raw, "<dl"),
"vacuous row: %q produces no table or definition list even unescaped", value)
})
}
}
// TestNotificationExtensionsDropUnusedGrammar pins the allowlist: each construct
// is openable from an untrusted value and used by no template.
func TestNotificationExtensionsDropUnusedGrammar(t *testing.T) {
t.Parallel()
for name, tc := range map[string]struct{ markdown, tag string }{
"Tables": {"a | b\n:-- | --:\nc | d", "<table"},
"DefinitionLists": {"Term\n: definition", "<dl"},
"MathJax": {"Eve $x^2$ end", `class="math`},
} {
t.Run(name, func(t *testing.T) {
t.Parallel()
require.Contains(t, renderHTML(tc.markdown, permissive), tc.tag,
"the construct is no longer reachable under CommonExtensions, so this test guards nothing")
assert.NotContains(t, HTMLFromNotificationMarkdown(tc.markdown), tc.tag,
"notificationExtensions still enables %s", name)
})
}
// What shipped templates do use must keep rendering.
for _, tc := range []struct{ markdown, want string }{
{"see [the docs](https://coder.com/docs/x).", `<a href="https://coder.com/docs/x"`},
{"Your workspace **foo** was suspended.", "<strong>foo</strong>"},
{"Resources:\n\n- one\n- two\n", "<li>"},
{"marked as [**dormant**](https://coder.com/docs/y) because", `<a href="https://coder.com/docs/y"`},
} {
assert.Contains(t, HTMLFromNotificationMarkdown(tc.markdown), tc.want)
}
}
// TestEscapeMarkdownIndentedCode covers the one construct with no escape: a
// space cannot be escaped, so the indent run is truncated instead.
func TestEscapeMarkdownIndentedCode(t *testing.T) {
t.Parallel()
for name, value := range map[string]string{
"FourSpaces": "Eve\n\n hidden",
"EightSpaces": "Eve\n\n hidden",
"SingleBreak": "Eve\n hidden",
"DeepInList": "Eve\n\n - hidden",
"OnlyIndented": " hidden",
} {
t.Run(name, func(t *testing.T) {
t.Parallel()
html := HTMLFromNotificationMarkdown(suspendedBody(EscapeMarkdown(value)))
assert.NotContains(t, html, "<pre", "value %q produced a code block: %s", value, html)
})
}
// Indentation up to the cap is preserved, so documented multi-line custom
// notification values keep their shape.
require.Equal(t, "Test results:\n • ok", EscapeMarkdown("Test results:\n • ok"))
require.Equal(t, "a\n b", EscapeMarkdown("a\n b"))
require.Equal(t, "a\n b", EscapeMarkdown("a\n b"))
}
// TestEscapeMarkdownEmptyLinkDestination covers the panic html.Safelink
// introduced: parser.IsSafeURL slices a destination before bounds-checking it.
func TestEscapeMarkdownEmptyLinkDestination(t *testing.T) {
t.Parallel()
for _, md := range []string{
"[our docs]()", "![px]()", "[a]( )", "[](https://coder.com)", "[a](x)",
} {
assert.NotPanics(t, func() { _ = HTMLFromNotificationMarkdown(md) }, "markdown %q", md)
// Reachable outside notifications, via OIDCConfig.SignupsDisabledText.
assert.NotPanics(t, func() { _ = HTMLFromMarkdown(md) }, "markdown %q", md)
}
// Destinations Safelink still permits must keep rendering.
for _, tc := range []struct{ md, want string }{
{"[a](https://coder.com)", `<a href="https://coder.com"`},
{"[a](/path)", `<a href="/path"`},
{"[a](./p)", `<a href="./p"`},
{"[a](mailto:x@y.z)", `<a href="mailto:x@y.z"`},
} {
assert.Contains(t, HTMLFromNotificationMarkdown(tc.md), tc.want)
}
// And the ones it rejects must stay rejected.
for _, md := range []string{"[a](javascript:alert(1))", "[a](data:text/html;base64,eA==)"} {
assert.NotContains(t, HTMLFromNotificationMarkdown(md), "<a ", "unsafe scheme linked: %s", md)
}
// Safelink also drops these two, a silent loss for a template author rather
// than a security property. Pinned so renderHTML's comment cannot drift.
for _, md := range []string{"[a](#anchor)", "[a](docs/x.md)"} {
assert.NotContains(t, HTMLFromNotificationMarkdown(md), "<a ",
"destination %q now renders an anchor; update the comment on renderHTML", md)
}
}
// TestEscapeMarkdownLeadingFoldConstruct covers a first line the fold cannot
// reach. Escaping costs a visible backslash, so the untouched cases matter too.
func TestEscapeMarkdownLeadingFoldConstruct(t *testing.T) {
t.Parallel()
t.Run("TitleKeepsItsTrustedText", func(t *testing.T) {
t.Parallel()
for _, value := range []string{"~~~", "~~~~", "~~~x", "~~~ ", " ~~~"} {
subject, err := PlaintextFromMarkdown(EscapeMarkdown(value) + " shared a chat with you")
require.NoError(t, err)
assert.Contains(t, subject, "shared a chat with you",
"value %q swallowed the trusted subject text", value)
}
// The backtick spelling is closed by backtick being inlineCritical.
subject, err := PlaintextFromMarkdown(EscapeMarkdown("```") + " shared a chat with you")
require.NoError(t, err)
require.Equal(t, "``` shared a chat with you", subject)
})
t.Run("SetextCannotPromoteATrustedLine", func(t *testing.T) {
t.Parallel()
for _, value := range []string{"===", "=", "===\nx", " === "} {
html := HTMLFromNotificationMarkdown(
"Trusted line\n" + EscapeMarkdown(value) + "\nTrusted trailer.")
assert.NotContains(t, html, "<h1", "value %q promoted a heading: %s", value, html)
}
})
t.Run("NonConstructsAreUntouched", func(t *testing.T) {
t.Parallel()
// These begin with a fold character without being a construct; escaping
// them would put a backslash in front of an ordinary display name.
for _, value := range []string{
"=> next", "~tilde name", "= x", "~~strike~~", "=?utf-8?q?x?=",
"~", "~~", "=== and more", "a\n===",
} {
plain, err := PlaintextFromMarkdown(EscapeMarkdown(value) + " end")
require.NoError(t, err)
assert.NotContains(t, plain, `\`,
"value %q was escaped when it is not a fold construct", value)
}
})
}
// TestRecoverToEscapedSource drives renderHTML's panic guard directly: safeURL
// closed the only input known to panic it.
func TestRecoverToEscapedSource(t *testing.T) {
t.Parallel()
const src = `<script>alert(1)</script> & "quoted"`
got := recoverToEscapedSource(src, func() string { panic("boom") })
assert.Equal(t, xhtml.EscapeString(src), got)
// The point of escaping rather than returning the source: no markup escapes.
assert.NotContains(t, got, "<script>")
assert.Equal(t, "rendered",
recoverToEscapedSource(src, func() string { return "rendered" }))
}
// TestHTMLFromMarkdownSafelink pins the behavior change Safelink brought to the
// shared renderer, called by OIDCConfig.SignupsDisabledText.
func TestHTMLFromMarkdownSafelink(t *testing.T) {
t.Parallel()
// Unsafe schemes stopped linking here, not just in notifications.
for _, md := range []string{"[a](javascript:alert(1))", "[a](data:text/html;base64,eA==)"} {
assert.NotContains(t, HTMLFromMarkdown(md), "<a ", "unsafe scheme linked: %s", md)
}
// Fragment and bare relative destinations stopped linking too, a silent loss
// rather than a security property. See renderHTML.
for _, md := range []string{"[a](#anchor)", "[a](docs/x.md)"} {
assert.NotContains(t, HTMLFromMarkdown(md), "<a ", "destination %q now links", md)
}
// What the signups-disabled text actually uses must keep working.
for _, tc := range []struct{ md, want string }{
{"see https://coder.com/docs", `<a href="https://coder.com/docs"`},
{"[docs](https://coder.com/docs)", `<a href="https://coder.com/docs"`},
{"contact [us](mailto:support@coder.com)", `<a href="mailto:support@coder.com"`},
{"**bold** and _italic_", "<strong>bold</strong>"},
} {
assert.Contains(t, HTMLFromMarkdown(tc.md), tc.want)
}
}
// TestEscapeMarkdownResiduals pins the one gap EscapeMarkdown cannot close.
func TestEscapeMarkdownResiduals(t *testing.T) {
t.Parallel()
t.Run("CodeSpanSwallowsEscapes", func(t *testing.T) {
t.Parallel()
// CommonMark does not process escapes inside a code span, and the
// workspace out-of-disk body wraps a value in one.
html := HTMLFromNotificationMarkdown("The volume `" + EscapeMarkdown("config[0]") + "` is full.")
require.Contains(t, html, `config\[0\]`,
"if this no longer leaks, the residual is closed and this test should become an assertion that it stays closed")
})
}
// TestEscapeMarkdownNoStrayBackslash asserts the no-stray-backslash invariant
// across every interpolation position a shipped template provides.
func TestEscapeMarkdownNoStrayBackslash(t *testing.T) {
t.Parallel()
positions := map[string]func(string) string{
"midline": suspendedBody,
"linestart": func(v string) string { return v + " shared a chat with you." },
"afterblank": func(v string) string { return "Hi.\n\n" + v + "\n\nRegards." },
"listitem": func(v string) string { return "Resources:\n\n- " + v + "\n" },
"trailing": func(v string) string { return "The account belongs to **" + v + "**" },
}
for _, value := range []string{
"William Tables", "bobby-workspace", "Bobby's Template",
"O'Brien-Smith (Eng) 100%", "José Müller 日本語",
"config[0]", "vol(1)", "1.5", "user_override",
} {
for name, pos := range positions {
t.Run(name+"/"+value, func(t *testing.T) {
t.Parallel()
md := pos(EscapeMarkdown(value))
html := HTMLFromNotificationMarkdown(md)
plain, err := PlaintextFromMarkdown(md)
require.NoError(t, err)
assert.NotContains(t, html, `\`, "stray backslash in HTML: %s", html)
assert.NotContains(t, plain, `\`, "stray backslash in plaintext: %q", plain)
assert.False(t, strings.Contains(html, `class="language-`),
"benign value reached the info-string sink: %s", html)
})
}
}
}
+76 -2
View File
@@ -113,12 +113,86 @@ func PlaintextFromMarkdown(markdown string) (string, error) {
return strings.TrimSpace(output), nil
}
// notificationExtensions is an allowlist. Shipped templates use only core
// CommonMark, so Tables, DefinitionLists, MathJax and Autolink are absent; each
// is openable from an untrusted label value. Adding one back means revisiting
// EscapeMarkdown.
const notificationExtensions = parser.NoIntraEmphasis | parser.HardLineBreak
func HTMLFromMarkdown(markdown string) string {
p := parser.NewWithExtensions(parser.CommonExtensions | parser.HardLineBreak) // Added HardLineBreak.
return renderHTML(markdown, parser.CommonExtensions|parser.HardLineBreak) // Added HardLineBreak.
}
// HTMLFromNotificationMarkdown converts a rendered notification body to HTML.
// Unlike HTMLFromMarkdown it does not autolink bare URLs, because notification
// bodies interpolate attacker-controlled label values.
func HTMLFromNotificationMarkdown(markdown string) string {
return renderHTML(markdown, notificationExtensions)
}
// longestURLPath is the longest relative-path prefix parser.IsSafeURL compares
// against. Derived so a dependency bump that adds a longer one stays correct.
var longestURLPath = func() int {
longest := 0
for _, p := range parser.Paths {
if len(p) > longest {
longest = len(p)
}
}
return longest
}()
// safeURL wraps parser.IsSafeURL, which slices a destination to each candidate
// prefix length before checking it is that long, and so panics on a short one
// with no spare capacity, as "[docs]()" produces. Padding the capacity keeps
// the slice in bounds; IsSafeURL's own guards still decide the result.
func safeURL(url []byte) bool {
if cap(url) < longestURLPath {
padded := make([]byte, len(url), longestURLPath)
copy(padded, url)
url = padded
}
return parser.IsSafeURL(url)
}
// recoverToEscapedSource runs render and, if it panics, returns the source
// HTML-escaped instead: the notification still arrives, showing Markdown
// source, and no markup escapes. Kept separate from renderHTML so the recovery
// is testable, safeURL having closed the only input known to panic.
func recoverToEscapedSource(markdown string, render func() string) (out string) {
defer func() {
if r := recover(); r != nil {
out = xhtml.EscapeString(markdown)
}
}()
return render()
}
// renderHTML converts Markdown to HTML. Input is untrusted, so a parser panic
// is recovered rather than taking down the dispatcher.
//
// Safelink silently drops fragment and bare relative destinations, so [a](#x)
// and [a](docs/x.md) render without an anchor. /path, ./path, mailto: and
// http(s):// still work.
func renderHTML(markdown string, extensions parser.Extensions) string {
return recoverToEscapedSource(markdown, func() string {
return renderHTMLUnsafe(markdown, extensions)
})
}
// renderHTMLUnsafe is renderHTML without the panic guard.
func renderHTMLUnsafe(markdown string, extensions parser.Extensions) string {
p := parser.NewWithExtensions(extensions)
p.IsSafeURLOverride = safeURL
doc := p.Parse([]byte(markdown))
renderer := html.NewRenderer(html.RendererOptions{
Flags: html.CommonFlags | html.SkipHTML,
// Safelink restricts generated hrefs to trusted schemes, which keeps
// javascript: and data: out of rendered output.
Flags: html.CommonFlags | html.SkipHTML | html.Safelink,
})
// Safelink routes every destination through parser.IsSafeURL, which panics
// on a short one. The hook lives on the renderer, not on its options.
renderer.IsSafeURLOverride = safeURL
return string(bytes.TrimSpace(gomarkdown.Render(doc, renderer)))
}