Merge pull request #1335 from storytold/codex/xml-dtd-review-hardening

fix(xml): address DTD validator review findings

crates/xml/src/dtd.rs

@@ -3,11 +3,19 @@
 //! Document Type Definitions: element declarations (content models) and required attributes, and
 //! validation of an XML document against them (Structure ▸ Validate).
 
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 
 /// How much text parameter entities may add during expansion (a pass count is not an output
 /// bound: each pass replaces whole strings, so references can inflate multiplicatively).
 const MAX_EXPANDED_BYTES: usize = 1024 * 1024;
+const MAX_DTD_BYTES: usize = 1024 * 1024;
+const MAX_VALIDATION_XML_BYTES: usize = 16 * 1024 * 1024;
+const MAX_VALIDATION_NODES: usize = 100_000;
+const MAX_VALIDATION_DEPTH: usize = 256;
+const MAX_VALIDATION_ATTRIBUTES: usize = 1_000_000;
+const MAX_MATCH_WORK: usize = 20_000_000;
+const MAX_MATCH_POSITIONS: usize = 5_000_000;
+const MAX_MATCH_STATES: usize = 100_000;
 
 /// A content particle: a name, a sequence, a choice, or one of those repeated (`?`, `*`, `+`).
 #[derive(Clone, Debug, PartialEq)]
@@ -54,32 +62,38 @@ impl Dtd {
 }
 
 /// Markup declarations (`<!…>`) with comments removed and quoted strings kept whole.
-fn declarations(src: &str) -> Vec<String> {
+fn declarations(src: &str) -> Result<Vec<String>, String> {
     let mut out = Vec::new();
     let mut rest = src;
     while let Some(i) = rest.find("<!") {
         rest = &rest[i..];
         if rest.starts_with("<!--") {
-            rest = rest.find("-->").map_or("", |j| &rest[j + 3..]);
+            let Some(j) = rest.find("-->") else {
+                return Err("unterminated DTD comment".into());
+            };
+            rest = &rest[j + 3..];
             continue;
         }
         let mut quote = None;
-        let mut end = rest.len();
+        let mut end = None;
         for (k, c) in rest.char_indices().skip(2) {
             match (quote, c) {
                 (Some(q), c) if c == q => quote = None,
                 (None, '"' | '\'') => quote = Some(c),
                 (None, '>') => {
-                    end = k;
+                    end = Some(k);
                     break;
                 }
                 _ => {}
             }
         }
+        let Some(end) = end else {
+            return Err("unterminated DTD declaration".into());
+        };
         out.push(rest[2..end].to_string());
-        rest = rest.get(end + 1..).unwrap_or("");
+        rest = &rest[end + 1..];
     }
-    out
+    Ok(out)
 }
 
 fn tokens(s: &str) -> Vec<String> {
@@ -159,8 +173,48 @@ fn parse_model(s: &str) -> Result<Model, String> {
         _ => {}
     }
     if s.contains("#PCDATA") {
-        let t = tokens(&s.replace("#PCDATA", " #PCDATA "));
-        return Ok(Model::Mixed(t.into_iter().filter(|x| !"()|*".contains(x.as_str()) && x != "#PCDATA").collect()));
+        let t = tokens(s);
+        if t.first().map(String::as_str) != Some("(") || t.get(1).map(String::as_str) != Some("#PCDATA") {
+            return Err("malformed mixed content model".into());
+        }
+        let mut i = 2;
+        let mut allowed = Vec::new();
+        if t.get(i).map(String::as_str) == Some(")") {
+            i += 1;
+            if t.get(i).map(String::as_str) == Some("*") {
+                i += 1;
+            }
+            if i != t.len() {
+                return Err("unexpected token after mixed content model".into());
+            }
+            return Ok(Model::Mixed(allowed));
+        }
+        loop {
+            if t.get(i).map(String::as_str) != Some("|") {
+                return Err("expected `|` in mixed content model".into());
+            }
+            i += 1;
+            let Some(name) = t.get(i).filter(|name| !"(),|?*+".contains(name.as_str())) else {
+                return Err("expected element name in mixed content model".into());
+            };
+            if allowed.contains(name) {
+                return Err(format!("duplicate `{name}` in mixed content model"));
+            }
+            allowed.push(name.clone());
+            i += 1;
+            match t.get(i).map(String::as_str) {
+                Some("|") => continue,
+                Some(")") => {
+                    i += 1;
+                    break;
+                }
+                _ => return Err("expected `|` or `)` in mixed content model".into()),
+            }
+        }
+        if t.get(i).map(String::as_str) != Some("*") || i + 1 != t.len() {
+            return Err("mixed content with element names must end in `*`".into());
+        }
+        return Ok(Model::Mixed(allowed));
     }
     let t = tokens(s);
     let mut i = 0;
@@ -196,7 +250,33 @@ fn expand_parameter_entities(src: &str, ents: &[(String, String)]) -> Result<Str
         let mut out = String::with_capacity(text.len());
         let mut rest = text.as_str();
         let mut changed = false;
-        while let Some(i) = rest.find('%') {
+        loop {
+            let mut special = None;
+            for (i, c) in rest.char_indices() {
+                if c == '%' {
+                    special = Some((i, false));
+                    break;
+                }
+                if c == '<' && rest[i..].starts_with("<!--") {
+                    special = Some((i, true));
+                    break;
+                }
+            }
+            let Some((i, is_comment)) = special else {
+                append(&mut out, rest)?;
+                rest = "";
+                break;
+            };
+            if is_comment {
+                append(&mut out, &rest[..i])?;
+                let comment_tail = &rest[i..];
+                let Some(end) = comment_tail.find("-->") else {
+                    return Err("unterminated DTD comment".into());
+                };
+                append(&mut out, &comment_tail[..end + 3])?;
+                rest = &comment_tail[end + 3..];
+                continue;
+            }
             let (prefix, reference) = rest.split_at(i);
             append(&mut out, prefix)?;
             // `find` points to an ASCII `%`, so both split positions are UTF-8 boundaries.
@@ -231,9 +311,12 @@ fn expand_parameter_entities(src: &str, ents: &[(String, String)]) -> Result<Str
 
 /// Parse a DTD (its element and attribute-list declarations; parameter entities are expanded).
 pub fn parse(src: &str) -> Result<Dtd, String> {
+    if src.len() > MAX_DTD_BYTES {
+        return Err(format!("DTD exceeds the {MAX_DTD_BYTES} byte parsing limit"));
+    }
     // Parameter entities first, then expand references and read the declarations again.
     let mut ents: Vec<(String, String)> = Vec::new();
-    for d in declarations(src) {
+    for d in declarations(src)? {
         if let Some(rest) = d.strip_prefix("ENTITY") {
             let rest = rest.trim_start();
             if let Some(rest) = rest.strip_prefix('%') {
@@ -249,7 +332,7 @@ pub fn parse(src: &str) -> Result<Dtd, String> {
     }
     let text = expand_parameter_entities(src, &ents)?;
     let mut dtd = Dtd::default();
-    for d in declarations(&text) {
+    for d in declarations(&text)? {
         if let Some(rest) = d.strip_prefix("ELEMENT") {
             let rest = rest.trim_start();
             let name: String = rest.chars().take_while(|c| !c.is_whitespace()).collect();
@@ -293,13 +376,31 @@ pub fn parse(src: &str) -> Result<Dtd, String> {
             }
             let Some((el, attrs)) = words.split_first() else { continue };
             let mut k = 0;
-            while k + 2 < attrs.len() {
+            while k < attrs.len() {
                 let name = &attrs[k];
-                let default = attrs.get(k + 2).map(String::as_str).unwrap_or("");
+                let Some(ty) = attrs.get(k + 1) else {
+                    return Err(format!("incomplete ATTLIST for `{el}`"));
+                };
+                k += 2;
+                // NOTATION is a two-token type: the keyword followed by its enumeration.
+                if ty == "NOTATION" {
+                    if attrs.get(k).is_none_or(|group| !group.starts_with('(') || !group.ends_with(')')) {
+                        return Err(format!("malformed NOTATION type for `{name}`"));
+                    }
+                    k += 1;
+                }
+                let Some(default) = attrs.get(k).map(String::as_str) else {
+                    return Err(format!("missing default declaration for `{name}`"));
+                };
+                k += 1;
                 if default == "#REQUIRED" {
                     dtd.required.entry(el.clone()).or_default().push(name.clone());
+                } else if default == "#FIXED" {
+                    if k >= attrs.len() {
+                        return Err(format!("missing #FIXED value for `{name}`"));
+                    }
+                    k += 1;
                 }
-                k += if default == "#FIXED" { 4 } else { 3 };
             }
         }
     }
@@ -313,18 +414,61 @@ pub fn parse(src: &str) -> Result<Dtd, String> {
 /// address of its node, which lives as long as the model) from a position.
 type EndsMemo = HashMap<(usize, usize), Vec<usize>>;
 
+#[derive(Default)]
+struct MatchBudget {
+    work: usize,
+    positions: usize,
+    states: usize,
+    exhausted: bool,
+    reported: bool,
+}
+
+impl MatchBudget {
+    fn spend(&mut self, work: usize, positions: usize) -> Result<(), &'static str> {
+        if self.exhausted {
+            return Err("content model matching exceeded its resource budget");
+        }
+        self.work = self.work.saturating_add(work);
+        self.positions = self.positions.saturating_add(positions);
+        if self.work > MAX_MATCH_WORK {
+            self.exhausted = true;
+            return Err("content model matching exceeded its work limit");
+        }
+        if self.positions > MAX_MATCH_POSITIONS {
+            self.exhausted = true;
+            return Err("content model matching exceeded its memory limit");
+        }
+        Ok(())
+    }
+
+    fn retain_state(&mut self) -> Result<(), &'static str> {
+        self.states = self.states.saturating_add(1);
+        if self.states > MAX_MATCH_STATES {
+            self.exhausted = true;
+            return Err("content model matching exceeded its state limit");
+        }
+        Ok(())
+    }
+}
+
 /// End positions after `cp` matches a prefix of `seq[i..]`.
 ///
 /// The positions of each group are worked out once per start position (`memo`): nested repetitions
 /// such as `((a*)*)*` ask the inner group for the same positions again and again, which is
 /// exponential in their depth without it.
-fn ends(cp: &Cp, seq: &[&str], i: usize, memo: &mut EndsMemo) -> Vec<usize> {
+fn ends(cp: &Cp, seq: &[&str], i: usize, memo: &mut EndsMemo, budget: &mut MatchBudget) -> Result<Vec<usize>, &'static str> {
+    budget.spend(1, 0)?;
     if let Cp::Name(n) = cp {
-        return if seq.get(i) == Some(&n.as_str()) { vec![i + 1] } else { vec![] };
+        if seq.get(i) == Some(&n.as_str()) {
+            budget.spend(0, 1)?;
+            return Ok(vec![i + 1]);
+        }
+        return Ok(vec![]);
     }
     let key = (std::ptr::from_ref(cp) as usize, i);
     if let Some(done) = memo.get(&key) {
-        return done.clone();
+        budget.spend(0, done.len())?;
+        return Ok(done.clone());
     }
     let mut out: Vec<usize> = match cp {
         Cp::Name(_) => vec![],
@@ -334,7 +478,9 @@ fn ends(cp: &Cp, seq: &[&str], i: usize, memo: &mut EndsMemo) -> Vec<usize> {
             for c in items {
                 let mut next = Vec::new();
                 for p in at {
-                    next.extend(ends(c, seq, p, memo));
+                    let found = ends(c, seq, p, memo, budget)?;
+                    budget.spend(found.len(), found.len())?;
+                    next.extend(found);
                 }
                 next.sort_unstable();
                 next.dedup();
@@ -345,22 +491,41 @@ fn ends(cp: &Cp, seq: &[&str], i: usize, memo: &mut EndsMemo) -> Vec<usize> {
         Cp::Choice(items) => {
             let mut all = Vec::new();
             for c in items {
-                all.extend(ends(c, seq, i, memo));
+                let found = ends(c, seq, i, memo, budget)?;
+                budget.spend(found.len(), found.len())?;
+                all.extend(found);
             }
             all
         }
         Cp::Rep(c, r) => {
             let mut all = if *r == '+' { vec![] } else { vec![i] };
-            let mut frontier = ends(c, seq, i, memo);
+            let mut seen: HashSet<usize> = all.iter().copied().collect();
+            let mut frontier = ends(c, seq, i, memo, budget)?;
+            budget.spend(frontier.len(), frontier.len())?;
             if *r == '?' {
                 all.extend(frontier);
             } else {
                 while !frontier.is_empty() {
-                    let new: Vec<usize> = frontier.iter().copied().filter(|p| !all.contains(p)).collect();
+                    let mut new = Vec::new();
+                    for p in frontier.iter().copied() {
+                        budget.spend(1, 0)?;
+                        if seen.insert(p) {
+                            budget.spend(0, 1)?;
+                            new.push(p);
+                        }
+                    }
+                    budget.spend(new.len(), new.len())?;
                     all.extend(&new);
                     let mut next = Vec::new();
+                    let mut next_seen = HashSet::new();
                     for p in new.into_iter().filter(|&p| p > i) {
-                        next.extend(ends(c, seq, p, memo).into_iter().filter(|p| !all.contains(p)));
+                        for found in ends(c, seq, p, memo, budget)? {
+                            budget.spend(1, 0)?;
+                            if !seen.contains(&found) && next_seen.insert(found) {
+                                budget.spend(0, 1)?;
+                                next.push(found);
+                            }
+                        }
                     }
                     // Two paths to one position are one position: kept apart they multiply per round.
                     next.sort_unstable();
@@ -373,8 +538,10 @@ fn ends(cp: &Cp, seq: &[&str], i: usize, memo: &mut EndsMemo) -> Vec<usize> {
     };
     out.sort_unstable();
     out.dedup();
+    budget.spend(out.len(), out.len())?;
+    budget.retain_state()?;
     memo.insert(key, out.clone());
-    out
+    Ok(out)
 }
 
 fn show(cp: &Cp) -> String {
@@ -392,60 +559,130 @@ struct Node {
     attrs: Vec<String>,
     children: Vec<Node>,
     text: bool,
+    content: bool,
+}
+
+fn bounded_node(e: &quick_xml::events::BytesStart<'_>, depth: usize, nodes: &mut usize, attributes: &mut usize) -> Result<Node, String> {
+    if depth >= MAX_VALIDATION_DEPTH {
+        return Err(format!("XML nesting exceeds the {MAX_VALIDATION_DEPTH} level validation limit"));
+    }
+    *nodes = nodes.saturating_add(1);
+    if *nodes > MAX_VALIDATION_NODES {
+        return Err(format!("XML exceeds the {MAX_VALIDATION_NODES} node validation limit"));
+    }
+    let mut attrs = Vec::new();
+    for attr in e.attributes().with_checks(true) {
+        let attr = attr.map_err(|err| format!("invalid attribute: {err}"))?;
+        attrs.push(String::from_utf8_lossy(attr.key.as_ref()).to_string());
+    }
+    *attributes = attributes.saturating_add(attrs.len());
+    if *attributes > MAX_VALIDATION_ATTRIBUTES {
+        return Err(format!("XML exceeds the {MAX_VALIDATION_ATTRIBUTES} attribute validation limit"));
+    }
+    Ok(Node { name: String::from_utf8_lossy(e.name().as_ref()).to_string(), attrs, children: vec![], text: false, content: false })
 }
 
 fn read_tree(xml: &str) -> Result<Node, String> {
     use quick_xml::events::Event;
+    if xml.len() > MAX_VALIDATION_XML_BYTES {
+        return Err(format!("XML exceeds the {} byte validation limit", MAX_VALIDATION_XML_BYTES));
+    }
     let mut r = quick_xml::Reader::from_str(xml);
     let mut stack: Vec<Node> = Vec::new();
-    let node = |e: &quick_xml::events::BytesStart| Node {
-        name: String::from_utf8_lossy(e.name().as_ref()).to_string(),
-        attrs: e.attributes().flatten().map(|a| String::from_utf8_lossy(a.key.as_ref()).to_string()).collect(),
-        children: vec![],
-        text: false,
-    };
+    let mut root = None;
+    let mut nodes = 0usize;
+    let mut attributes = 0usize;
     loop {
         match r.read_event().map_err(|e| e.to_string())? {
-            Event::Start(e) => stack.push(node(&e)),
+            Event::Start(e) => {
+                if stack.is_empty() && root.is_some() {
+                    return Err("multiple root elements".into());
+                }
+                let n = bounded_node(&e, stack.len(), &mut nodes, &mut attributes)?;
+                if let Some(parent) = stack.last_mut() {
+                    parent.content = true;
+                }
+                stack.push(n);
+            }
             Event::Empty(e) => {
-                let n = node(&e);
+                if stack.is_empty() && root.is_some() {
+                    return Err("multiple root elements".into());
+                }
+                let n = bounded_node(&e, stack.len(), &mut nodes, &mut attributes)?;
                 match stack.last_mut() {
-                    Some(p) => p.children.push(n),
-                    None => return Ok(n),
+                    Some(p) => {
+                        p.content = true;
+                        p.children.push(n);
+                    }
+                    None => root = Some(n),
                 }
             }
-            Event::End(_) => {
+            Event::End(e) => {
                 let n = stack.pop().ok_or("unbalanced end tag")?;
+                if n.name.as_bytes() != e.name().as_ref() {
+                    return Err("mismatched end tag".into());
+                }
                 match stack.last_mut() {
                     Some(p) => p.children.push(n),
-                    None => return Ok(n),
+                    None => root = Some(n),
                 }
             }
             Event::Text(t) => {
-                if let Some(p) = stack.last_mut()
-                    && !t.iter().all(u8::is_ascii_whitespace)
-                {
-                    p.text = true;
+                if let Some(p) = stack.last_mut() {
+                    p.content = true;
+                    if !t.iter().all(u8::is_ascii_whitespace) {
+                        p.text = true;
+                    }
+                } else if !t.iter().all(u8::is_ascii_whitespace) {
+                    return Err("text outside the root element".into());
                 }
             }
-            Event::CData(_) | Event::GeneralRef(_) => {
+            Event::CData(_) => {
                 if let Some(p) = stack.last_mut() {
                     p.text = true;
+                    p.content = true;
                 }
             }
-            Event::Eof => return Err("no root element".into()),
+            Event::GeneralRef(reference) => {
+                let name = reference.decode().map_err(|e| e.to_string())?;
+                let accepted = if let Some(character) = reference.resolve_char_ref().map_err(|e| e.to_string())? {
+                    crate::is_xml_char(character)
+                } else {
+                    matches!(name.as_ref(), "amp" | "lt" | "gt" | "apos" | "quot")
+                };
+                if !accepted {
+                    return Err(format!("unsupported general entity reference `&{name};`"));
+                }
+                if let Some(p) = stack.last_mut() {
+                    p.text = true;
+                    p.content = true;
+                } else {
+                    return Err("reference outside the root element".into());
+                }
+            }
+            Event::Comment(_) | Event::PI(_) => {
+                if let Some(parent) = stack.last_mut() {
+                    parent.content = true;
+                }
+            }
+            Event::Eof => {
+                if !stack.is_empty() {
+                    return Err("unclosed element".into());
+                }
+                return root.ok_or_else(|| "no root element".into());
+            }
             _ => {}
         }
     }
 }
 
-fn check(dtd: &Dtd, n: &Node, path: &str, out: &mut Vec<Problem>) {
+fn check(dtd: &Dtd, n: &Node, path: &str, out: &mut Vec<Problem>, budget: &mut MatchBudget) {
     let mut problem = |m: String| out.push(Problem { path: path.to_string(), message: m });
     match dtd.model(&n.name) {
         None => problem(format!("`{}` isn't declared in the DTD", n.name)),
         Some(Model::Any) => {}
         Some(Model::Empty) => {
-            if n.text || !n.children.is_empty() {
+            if n.content {
                 problem(format!("`{}` should be empty", n.name));
             }
         }
@@ -461,9 +698,19 @@ fn check(dtd: &Dtd, n: &Node, path: &str, out: &mut Vec<Problem>) {
                 problem(format!("`{}` can't contain text", n.name));
             }
             let seq: Vec<&str> = n.children.iter().map(|c| c.name.as_str()).collect();
-            if !ends(cp, &seq, 0, &mut EndsMemo::new()).contains(&seq.len()) {
-                let got = if seq.is_empty() { "nothing".to_string() } else { seq.join(", ") };
-                problem(format!("`{}` contains {got}; the DTD asks for {}", n.name, show(cp)));
+            let matched = ends(cp, &seq, 0, &mut EndsMemo::new(), budget);
+            match matched {
+                Ok(positions) if positions.contains(&seq.len()) => {}
+                Ok(_) => {
+                    let got = if seq.is_empty() { "nothing".to_string() } else { seq.join(", ") };
+                    problem(format!("`{}` contains {got}; the DTD asks for {}", n.name, show(cp)));
+                }
+                Err(error) => {
+                    if !budget.reported {
+                        problem(format!("`{}`: {error}", n.name));
+                        budget.reported = true;
+                    }
+                }
             }
         }
     }
@@ -476,17 +723,21 @@ fn check(dtd: &Dtd, n: &Node, path: &str, out: &mut Vec<Problem>) {
     for c in &n.children {
         let k = seen.entry(c.name.as_str()).or_default();
         *k += 1;
-        check(dtd, c, &format!("{path}/{}[{k}]", c.name), out);
+        check(dtd, c, &format!("{path}/{}[{k}]", c.name), out, budget);
     }
 }
 
 /// Validate `xml` against `dtd`; an empty list when it's valid.
+///
+/// Validation rejects input above the internal byte, depth, node, attribute and content-model
+/// matching budgets. Only predefined entities and numeric character references are resolved; a
+/// general entity reference is reported as unsupported rather than expanded or fetched.
 pub fn validate(dtd: &Dtd, xml: &str) -> Vec<Problem> {
     match read_tree(xml) {
         Ok(root) => {
             let mut out = Vec::new();
             let path = root.name.clone();
-            check(dtd, &root, &path, &mut out);
+            check(dtd, &root, &path, &mut out, &mut MatchBudget::default());
             out
         }
         Err(e) => vec![Problem { path: String::new(), message: format!("not well-formed XML: {e}") }],
@@ -745,4 +996,68 @@ mod tests {
         assert_eq!(problems.len(), 1, "{problems:?}");
         assert!(problems[0].message.starts_with("`r` contains a, a, a"), "{problems:?}");
     }
+
+    #[test]
+    fn incomplete_declarations_and_malformed_mixed_models_are_rejected() {
+        assert!(parse("<!ELEMENT r EMPTY").is_err());
+        assert!(parse("<!-- unterminated").is_err());
+        for model in ["(#PCDATA | a)", "(#PCDATA, a)*", "(#PCDATA | )*", "(#PCDATA | a)* junk"] {
+            assert!(parse(&format!("<!ELEMENT r {model}>")).is_err(), "{model}");
+        }
+        assert!(parse("<!ELEMENT r (#PCDATA)>").is_ok());
+        assert!(parse("<!ELEMENT r (#PCDATA)*>").is_ok());
+        assert!(parse("<!ELEMENT r (#PCDATA | a)*>").is_ok());
+    }
+
+    #[test]
+    fn parameter_entities_are_not_expanded_inside_comments() {
+        let dtd = parse(
+            r#"<!ENTITY % e "--> <!ELEMENT r ANY> <!--">
+<!-- %e; -->
+<!ELEMENT r EMPTY>
+<!ELEMENT a EMPTY>"#,
+        )
+        .unwrap();
+        assert_eq!(dtd.model("r"), Some(&Model::Empty));
+        assert!(!validate(&dtd, "<r><a/></r>").is_empty());
+    }
+
+    #[test]
+    fn notation_attribute_type_keeps_following_required_attribute_aligned() {
+        let dtd = parse("<!ELEMENT r ANY><!ATTLIST r kind NOTATION (png|jpeg) #IMPLIED id CDATA #REQUIRED>").unwrap();
+        assert!(validate(&dtd, "<r kind='png'/>").iter().any(|p| p.message.contains("required attribute `id`")));
+        assert!(validate(&dtd, "<r id='x'/>").is_empty());
+    }
+
+    #[test]
+    fn malformed_or_unbounded_xml_is_rejected_before_validation() {
+        let dtd = parse("<!ELEMENT r ANY><!ELEMENT a EMPTY>").unwrap();
+        for xml in ["<r/><broken", "<r></r><a/>", "<r a='1' a='2'/>"] {
+            assert!(!validate(&dtd, xml).is_empty(), "{xml}");
+        }
+        let deep = format!("{}{}", "<r>".repeat(MAX_VALIDATION_DEPTH + 1), "</r>".repeat(MAX_VALIDATION_DEPTH + 1));
+        let problems = validate(&dtd, &deep);
+        assert!(problems.iter().any(|p| p.message.contains("nesting exceeds")), "{problems:?}");
+    }
+
+    #[test]
+    fn references_must_be_declared_and_empty_elements_must_have_no_content() {
+        let dtd = parse("<!ELEMENT r ANY><!ELEMENT e EMPTY><!ENTITY known 'value'>").unwrap();
+        assert!(validate(&dtd, "<r>&missing;</r>").iter().any(|p| p.message.contains("unsupported general entity")));
+        assert!(validate(&dtd, "<r>&known;</r>").iter().any(|p| p.message.contains("unsupported general entity")));
+        assert!(validate(&dtd, "<r>&#x41;&amp;</r>").is_empty());
+        assert!(!validate(&dtd, "<r>&#0;</r>").is_empty());
+        for xml in ["<e> </e>", "<e><!--x--></e>", "<e><?work x?></e>", "<e>&amp;</e>"] {
+            assert!(validate(&dtd, xml).iter().any(|p| p.message.contains("should be empty")), "{xml}");
+        }
+        assert!(validate(&dtd, "<e/>").is_empty());
+    }
+
+    #[test]
+    fn ambiguous_content_matching_stops_at_its_budget() {
+        let dtd = parse("<!ELEMENT r ((a*)*)*><!ELEMENT a EMPTY>").unwrap();
+        let xml = format!("<r>{}</r>", "<a/>".repeat(10_000));
+        let problems = validate(&dtd, &xml);
+        assert!(problems.iter().any(|p| p.message.contains("matching exceeded its ")), "{}", problems.len());
+    }
 }