Make json filter safe

Previously the built-in json filter had an issue that made it unsafe to
use in HTML data. When used in HTML attributes an attacker who is able
to supply an arbitrary string that should be JSON encoded could close
the containing HTML element e.g. with `"</div>"`, and write arbitrary
HTML code afterwards as long as they use apostrophes instead of
quotation marks. The programmer could make this use case safe by
explicitly escaping the JSON result: `{{data|json|escape}}`.

In a `<script>` context the json filter was not usable at all, because
in scripts HTML escaped entities are not parsed outside of XHTML
documents. Without using the safe filter an attacker could close the
current script using `"</script>"`.

This PR fixes the problem by always escaping less-than, greater-than,
ampersand, and apostrophe characters using their JSON unicode escape
sequence `\u00xx`. Unless the programmer explicitly uses the safe
filter, quotation marks are HTML encoded as `&quot`. In scripts the
programmer should use the safe filter, otherwise not.

9 files changed, 191 insertions, 49 deletions

diff --git a/askama_escape/Cargo.toml b/askama_escape/Cargo.toml
index ed7148c..643f81a 100644
--- a/askama_escape/Cargo.toml
+++ b/askama_escape/Cargo.toml
@@ -17,6 +17,9 @@ maintenance = { status = "actively-developed" }
 [dev-dependencies]
 criterion = "0.3"
 
+[features]
+json = []
+
 [[bench]]
 name = "all"
 harness = false
diff --git a/askama_escape/src/lib.rs b/askama_escape/src/lib.rs
index ad08e55..1788843 100644
--- a/askama_escape/src/lib.rs
+++ b/askama_escape/src/lib.rs
@@ -1,10 +1,7 @@
-#![no_std]
+#![cfg_attr(not(any(feature = "json", test)), no_std)]
 #![deny(elided_lifetimes_in_paths)]
 #![deny(unreachable_pub)]
 
-#[cfg(test)]
-extern crate std;
-
 use core::fmt::{self, Display, Formatter, Write};
 use core::str;
 
@@ -175,6 +172,57 @@ pub trait Escaper {
 
 const FLAG: u8 = b'>' - b'"';
 
+/// Escape chevrons, ampersand and apostrophes for use in JSON
+#[cfg(feature = "json")]
+#[derive(Debug, Clone, Default)]
+pub struct JsonEscapeBuffer(Vec<u8>);
+
+#[cfg(feature = "json")]
+impl JsonEscapeBuffer {
+    pub fn new() -> Self {
+        Self(Vec::new())
+    }
+
+    pub fn finish(self) -> String {
+        unsafe { String::from_utf8_unchecked(self.0) }
+    }
+}
+
+#[cfg(feature = "json")]
+impl std::io::Write for JsonEscapeBuffer {
+    fn write(&mut self, bytes: &[u8]) -> std::io::Result<usize> {
+        macro_rules! push_esc_sequence {
+            ($start:ident, $i:ident, $self:ident, $bytes:ident, $quote:expr) => {{
+                if $start < $i {
+                    $self.0.extend_from_slice(&$bytes[$start..$i]);
+                }
+                $self.0.extend_from_slice($quote);
+                $start = $i + 1;
+            }};
+        }
+
+        self.0.reserve(bytes.len());
+        let mut start = 0;
+        for (i, b) in bytes.iter().enumerate() {
+            match *b {
+                b'&' => push_esc_sequence!(start, i, self, bytes, br#"\u0026"#),
+                b'\'' => push_esc_sequence!(start, i, self, bytes, br#"\u0027"#),
+                b'<' => push_esc_sequence!(start, i, self, bytes, br#"\u003c"#),
+                b'>' => push_esc_sequence!(start, i, self, bytes, br#"\u003e"#),
+                _ => (),
+            }
+        }
+        if start < bytes.len() {
+            self.0.extend_from_slice(&bytes[start..]);
+        }
+        Ok(bytes.len())
+    }
+
+    fn flush(&mut self) -> std::io::Result<()> {
+        Ok(())
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/askama_shared/Cargo.toml b/askama_shared/Cargo.toml
index e19ad35..ea9e968 100644
--- a/askama_shared/Cargo.toml
+++ b/askama_shared/Cargo.toml
@@ -12,7 +12,7 @@ edition = "2018"
 [features]
 default = ["config", "humansize", "num-traits", "percent-encoding"]
 config = ["serde", "toml"]
-json = ["serde", "serde_json"]
+json = ["serde", "serde_json", "askama_escape/json"]
 markdown = ["comrak"]
 yaml = ["serde", "serde_yaml"]
 
diff --git a/askama_shared/src/filters/json.rs b/askama_shared/src/filters/json.rs
index c0df707..e94e50c 100644
--- a/askama_shared/src/filters/json.rs
+++ b/askama_shared/src/filters/json.rs
@@ -1,33 +1,40 @@
 use crate::error::{Error, Result};
-use askama_escape::{Escaper, MarkupDisplay};
+use askama_escape::JsonEscapeBuffer;
 use serde::Serialize;
+use serde_json::to_writer_pretty;
 
-/// Serialize to JSON (requires `serde_json` feature)
+/// Serialize to JSON (requires `json` feature)
 ///
-/// ## Errors
+/// The generated string does not contain ampersands `&`, chevrons `< >`, or apostrophes `'`.
+/// To use it in a `<script>` you can combine it with the safe filter:
 ///
-/// This will panic if `S`'s implementation of `Serialize` decides to fail,
-/// or if `T` contains a map with non-string keys.
-pub fn json<E: Escaper, S: Serialize>(e: E, s: S) -> Result<MarkupDisplay<E, String>> {
-    match serde_json::to_string_pretty(&s) {
-        Ok(s) => Ok(MarkupDisplay::new_safe(s, e)),
-        Err(e) => Err(Error::from(e)),
-    }
+/// ``` html
+/// <script>
+/// var data = {{data|json|safe}};
+/// </script>
+/// ```
+///
+/// To use it in HTML attributes, you can either use it in quotation marks `"{{data|json}}"` as is,
+/// or in apostrophes with the (optional) safe filter `'{{data|json|safe}}'`.
+/// In HTML texts the output of e.g. `<pre>{{data|json|safe}}</pre>` is safe, too.
+pub fn json<S: Serialize>(s: S) -> Result<String> {
+    let mut writer = JsonEscapeBuffer::new();
+    to_writer_pretty(&mut writer, &s).map_err(Error::from)?;
+    Ok(writer.finish())
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
-    use askama_escape::Html;
 
     #[test]
     fn test_json() {
-        assert_eq!(json(Html, true).unwrap().to_string(), "true");
-        assert_eq!(json(Html, "foo").unwrap().to_string(), r#""foo""#);
-        assert_eq!(json(Html, &true).unwrap().to_string(), "true");
-        assert_eq!(json(Html, &"foo").unwrap().to_string(), r#""foo""#);
+        assert_eq!(json(true).unwrap(), "true");
+        assert_eq!(json("foo").unwrap(), r#""foo""#);
+        assert_eq!(json(&true).unwrap(), "true");
+        assert_eq!(json(&"foo").unwrap(), r#""foo""#);
         assert_eq!(
-            json(Html, &vec!["foo", "bar"]).unwrap().to_string(),
+            json(&vec!["foo", "bar"]).unwrap(),
             r#"[
   "foo",
   "bar"
diff --git a/askama_shared/src/generator.rs b/askama_shared/src/generator.rs
index 6024436..ea22a83 100644
--- a/askama_shared/src/generator.rs
+++ b/askama_shared/src/generator.rs
@@ -1171,7 +1171,7 @@ impl<'a, S: std::hash::BuildHasher> Generator<'a, S> {
             return Err("the `yaml` filter requires the `serde-yaml` feature to be enabled".into());
         }
 
-        const FILTERS: [&str; 3] = ["safe", "json", "yaml"];
+        const FILTERS: [&str; 2] = ["safe", "yaml"];
         if FILTERS.contains(&name) {
             buf.write(&format!(
                 "::askama::filters::{}({}, ",
diff --git a/book/src/filters.md b/book/src/filters.md
index d00d778..c24a94c 100644
--- a/book/src/filters.md
+++ b/book/src/filters.md
@@ -287,6 +287,53 @@ Output:
 5
 ```
 
+## Optional / feature gated filters
+
+The following filters can be enabled by requesting the respective feature in the Cargo.toml
+[dependencies section](https://doc.rust-lang.org/cargo/reference/specifying-dependencies.html), e.g.
+
+```
+[dependencies]
+askama = { version = "0.11.0", features = "serde-json" }
+```
+
+### `json` | `tojson`
+
+Enabling the `serde-json` feature will enable the use of the `json` filter.
+This will output formatted JSON for any value that implements the required
+[`Serialize`](https://docs.rs/serde/1.*/serde/trait.Serialize.html) trait.
+The generated string does not contain ampersands `&`, chevrons `< >`, or apostrophes `'`.
+
+To use it in a `<script>` you can combine it with the safe filter.
+In HTML attributes, you can either use it in quotation marks `"{{data|json}}"` as is,
+or in apostrophes with the (optional) safe filter `'{{data|json|safe}}'`.
+In HTML texts the output of e.g. `<pre>{{data|json|safe}}</pre>` is safe, too.
+
+```
+Good: <li data-extra="{{data|json}}">…</li>
+Good: <li data-extra='{{data|json|safe}}'>…</li>
+Good: <pre>{{data|json|safe}}</pre>
+Good: <script>var data = {{data|json|safe}};</script>
+
+Bad:  <li data-extra="{{data|json|safe}}">…</li>
+Bad:  <script>var data = {{data|json}};</script>
+Bad:  <script>var data = "{{data|json|safe}}";</script>
+
+Ugly: <script>var data = "{{data|json}}";</script>
+Ugly: <script>var data = '{{data|json|safe}}';</script>
+```
+
+### `yaml`
+
+Enabling the `serde-yaml` feature will enable the use of the `yaml` filter.
+This will output formatted YAML for any value that implements the required
+[`Serialize`](https://docs.rs/serde/1.*/serde/trait.Serialize.html) trait.
+
+```jinja
+{{ foo|yaml }}
+```
+
+
 ## Custom Filters
 
 To define your own filters, simply have a module named filters in scope of the context deriving a `Template` impl.
@@ -311,28 +358,3 @@ fn main() {
     assert_eq!(t.render().unwrap(), "faa");
 }
 ```
-
-## The `json` filter
-
-Enabling the `serde-json` feature will enable the use of the `json` filter.
-This will output formatted JSON for any value that implements the required
-`Serialize` trait.
-
-```jinja
-{
-  "foo": "{{ foo }}",
-  "bar": {{ bar|json }}
-}
-```
-
-For compatibility with Jinja, `tojson` can be used in place of `json`.
-
-## The `yaml` filter
-
-Enabling the `serde-yaml` feature will enable the use of the `yaml` filter.
-This will output formatted JSON for any value that implements the required
-`Serialize` trait.
-
-```
-{{ foo|yaml }}
-```
diff --git a/testing/templates/json.html b/testing/templates/json.html
index 250b7be..a1855ca 100644
--- a/testing/templates/json.html
+++ b/testing/templates/json.html
@@ -1,4 +1,4 @@
 {
   "foo": "{{ foo }}",
-  "bar": {{ bar|json }}
+  "bar": {{ bar|json|safe }}
 }
diff --git a/testing/tests/filters.rs b/testing/tests/filters.rs
index be3e0ab..7973f45 100644
--- a/testing/tests/filters.rs
+++ b/testing/tests/filters.rs
@@ -250,3 +250,63 @@ fn test_filter_truncate() {
     };
     assert_eq!(t.render().unwrap(), "alpha baralpha...");
 }
+
+#[cfg(feature = "serde-json")]
+#[derive(Template)]
+#[template(source = r#"<li data-name="{{name|json}}"></li>"#, ext = "html")]
+struct JsonAttributeTemplate<'a> {
+    name: &'a str,
+}
+
+#[cfg(feature = "serde-json")]
+#[test]
+fn test_json_attribute() {
+    let t = JsonAttributeTemplate {
+        name: r#""><button>Hacked!</button>"#,
+    };
+    assert_eq!(
+        t.render().unwrap(),
+        r#"<li data-name="&quot;\&quot;\u003e\u003cbutton\u003eHacked!\u003c/button\u003e&quot;"></li>"#
+    );
+}
+
+#[cfg(feature = "serde-json")]
+#[derive(Template)]
+#[template(source = r#"<li data-name='{{name|json|safe}}'></li>"#, ext = "html")]
+struct JsonAttribute2Template<'a> {
+    name: &'a str,
+}
+
+#[cfg(feature = "serde-json")]
+#[test]
+fn test_json_attribute2() {
+    let t = JsonAttribute2Template {
+        name: r#"'><button>Hacked!</button>"#,
+    };
+    assert_eq!(
+        t.render().unwrap(),
+        r#"<li data-name='"\u0027\u003e\u003cbutton\u003eHacked!\u003c/button\u003e"'></li>"#
+    );
+}
+
+#[cfg(feature = "serde-json")]
+#[derive(Template)]
+#[template(
+    source = r#"<script>var user = {{name|json|safe}}</script>"#,
+    ext = "html"
+)]
+struct JsonScriptTemplate<'a> {
+    name: &'a str,
+}
+
+#[cfg(feature = "serde-json")]
+#[test]
+fn test_json_script() {
+    let t = JsonScriptTemplate {
+        name: r#"</script><button>Hacked!</button>"#,
+    };
+    assert_eq!(
+        t.render().unwrap(),
+        r#"<script>var user = "\u003c/script\u003e\u003cbutton\u003eHacked!\u003c/button\u003e"</script>"#
+    );
+}
diff --git a/testing/tests/whitespace.rs b/testing/tests/whitespace.rs
index ca72b23..cbcddd7 100644
--- a/testing/tests/whitespace.rs
+++ b/testing/tests/whitespace.rs
@@ -1,3 +1,5 @@
+#![cfg(feature = "serde-json")]
+
 use askama::Template;
 
 #[derive(askama::Template, Default)]
@@ -37,5 +39,5 @@ fn test_extra_whitespace() {
     let mut template = AllowWhitespaces::default();
     template.nested_1.nested_2.array = &["a0", "a1", "a2", "a3"];
     template.nested_1.nested_2.hash.insert("key", "value");
-    assert_eq!(template.render().unwrap(), "\n0\n0\n0\n0\n\n\n\n0\n0\n0\n0\n0\n\na0\na1\nvalue\n\n\n\n\n\n[\n  \"a0\",\n  \"a1\",\n  \"a2\",\n  \"a3\"\n]\n[\n  \"a0\",\n  \"a1\",\n  \"a2\",\n  \"a3\"\n][\n  \"a0\",\n  \"a1\",\n  \"a2\",\n  \"a3\"\n]\n[\n  \"a1\"\n][\n  \"a1\"\n]\n[\n  \"a1\",\n  \"a2\"\n][\n  \"a1\",\n  \"a2\"\n]\n[\n  \"a1\"\n][\n  \"a1\"\n]1-1-1\n3333 3\n2222 2\n0000 0\n3333 3\n\ntruefalse\nfalsefalsefalse\n\n\n\n\n\n\n\n\n\n\n\n\n\n");
+    assert_eq!(template.render().unwrap(), "\n0\n0\n0\n0\n\n\n\n0\n0\n0\n0\n0\n\na0\na1\nvalue\n\n\n\n\n\n[\n  &quot;a0&quot;,\n  &quot;a1&quot;,\n  &quot;a2&quot;,\n  &quot;a3&quot;\n]\n[\n  &quot;a0&quot;,\n  &quot;a1&quot;,\n  &quot;a2&quot;,\n  &quot;a3&quot;\n][\n  &quot;a0&quot;,\n  &quot;a1&quot;,\n  &quot;a2&quot;,\n  &quot;a3&quot;\n]\n[\n  &quot;a1&quot;\n][\n  &quot;a1&quot;\n]\n[\n  &quot;a1&quot;,\n  &quot;a2&quot;\n][\n  &quot;a1&quot;,\n  &quot;a2&quot;\n]\n[\n  &quot;a1&quot;\n][\n  &quot;a1&quot;\n]1-1-1\n3333 3\n2222 2\n0000 0\n3333 3\n\ntruefalse\nfalsefalsefalse\n\n\n\n\n\n\n\n\n\n\n\n\n\n");
 }