Merge pull request #3810 from Silvertree2010/fix/3567-audit-eps-import-postscript-filt

Charge PostScript filter output to the EPS interpreter budget

crates/eps/src/import/data.rs

@@ -731,6 +731,12 @@ impl Interp<'_> {
         self.push_interval(Obj::Str(s), 0, got.len())
     }
 
+    /// Count `n` decoded bytes against the memory budget, and the time to decode them.
+    fn charge_decoded(&mut self, n: usize) -> Res {
+        self.alloc(n)?;
+        self.spend((n / 4096) as u64)
+    }
+
     /// `filter`: a file that decodes its source. Over the program's own data it decodes when
     /// first read (the data follows the operator that reads it); over other data, at once.
     fn filter(&mut self) -> Res {
@@ -775,6 +781,7 @@ impl Interp<'_> {
                         let mut b = f.borrow_mut();
                         let Stream::Buf { data, pos, .. } = &mut *b else { return io("filter") };
                         let (out, used, rgb) = decode(&spec, data.get(*pos..).unwrap_or_default())?;
+                        self.charge_decoded(out.len())?;
                         *pos = pos.saturating_add(used);
                         Stream::Buf { data: out, pos: 0, rgb }
                     }
@@ -783,11 +790,13 @@ impl Interp<'_> {
             Obj::Str(s) => {
                 self.alloc(s.borrow().len())?;
                 let (data, _, rgb) = decode(&spec, &s.borrow())?;
+                self.charge_decoded(data.len())?;
                 Stream::Buf { data, pos: 0, rgb }
             }
             p @ Obj::Array { .. } => {
                 let raw = self.gather(&p, MAX_DATA)?;
                 let (data, _, rgb) = decode(&spec, &raw)?;
+                self.charge_decoded(data.len())?;
                 Stream::Buf { data, pos: 0, rgb }
             }
             _ => return ps_err("typecheck", "filter"),
@@ -830,13 +839,13 @@ impl Interp<'_> {
                 None => {
                     let raw = self.lex.data();
                     let (data, used, rgb) = decode(spec, raw)?;
-                    self.alloc(data.len())?;
+                    self.charge_decoded(data.len())?;
                     self.lex.advance(used);
                     (data, rgb)
                 }
                 Some((prev, was)) => {
                     let (data, _, rgb) = decode(spec, &prev)?;
-                    self.alloc(data.len())?;
+                    self.charge_decoded(data.len())?;
                     (data, rgb.or(was))
                 }
             });

crates/eps/src/import/tests_generators.rs

@@ -1103,3 +1103,19 @@ fn restore_puts_dictionaries_back() {
     // take, is positive (chapter 8), and programs divide by it.
     super::tests::check("cachestatus 7 1 roll 6 { pop } repeat 0 gt");
 }
+
+/// #3567: what a `filter` decodes counts against the memory budget, not just its input. A
+/// small string that inflates to 16 MiB, decoded over and over (each result dropped), ends
+/// the program; a single one does not.
+#[test]
+fn filter_output_counts_against_the_memory_budget() {
+    fn hex(data: &[u8]) -> String {
+        data.iter().map(|b| format!("{b:02x}")).collect()
+    }
+    let z = hex(&deflate(&vec![0u8; 16 << 20]));
+    let once = import(&file("t", &format!("/z <{z}> def z /FlateDecode filter pop 0 0 1 1 rectfill"))).unwrap();
+    assert!(once.warnings.iter().all(|w| !w.contains("memory")), "{:?}", once.warnings);
+    let many = import(&file("t", &format!("/z <{z}> def 100 {{ z /FlateDecode filter pop }} repeat 0 0 1 1 rectfill")));
+    let Err(err) = many else { panic!("the budget ends the program") };
+    assert!(err.contains("too much memory"), "{err}");
+}