security(compose): bound dense working sets and nested Blend If evaluation (#8949)

* security(compose): bound dense working sets and repeated group evaluation Smart-filter mixing and stroke fitting work tile by tile instead of over the union of sparse surfaces; render_layer admits its rect like render; hidden layers no longer disable the proxy and uniform-area shortcuts; Blend If groups nested deeper than two are judged by their result, so nesting costs polynomial instead of exponential work; effect builds past an absolute byte cap are skipped. * CodeRabbit #8949: exact tile skips in filter_mask, allocated-tile scan in fit_stroke, contrasting backdrop in the nested-group test - filter_mask mix/unmix: bare tiles (no tiles on either side) skip for non-alpha formats too when the two surfaces' default pixels are equal; differing defaults still get processed, so the clone's default pixel is never left wrong under a non-unit mask - shape_split fit_stroke: scan the union of the three surfaces' allocated tiles (exact skip when all three defaults are transparent) instead of every tile of a 300k-canvas document; non-transparent defaults fall back to the full scan - tests: the nested-groups assertion now uses a black backdrop so the asserted white can only come from rendering through the group chain fixes CodeRabbit review comments (major/minor) on filter_mask.rs:27,29, shape_split.rs:126, tests.rs:2354 * compose: skip a BTreeSet in fit_stroke — a sorted Vec compiles under the wasm toolchain (E0599 on ::drain there)
2026-10-24 01:47:08 -06:00 · 2026-10-23 15:33:01 -04:00
parent ecd1a692a3
commit 6ca5bf359d
6 changed files with 222 additions and 52 deletions
@@ -21,27 +21,38 @@ pub fn mix(base: &Surface, filtered: &Surface, mask: &LayerMask) -> Surface {
        return out;
    }
    let n = fmt.channels();
-    let (b, f) = (base.read_region(area), filtered.read_region(area));
-    let mut k = Vec::new();
-    mask.values_into(area, &mut k);
-    let mut o = vec![0.0f32; b.len()];
-    for (i, ((bp, fp), op)) in b.chunks_exact(n).zip(f.chunks_exact(n)).zip(o.chunks_exact_mut(n)).enumerate() {
-        let m = k.get(i).copied().unwrap_or(1.0).clamp(0.0, 1.0);
-        if !fmt.alpha {
-            for c in 0..n {
-                op[c] = bp[c] + (fp[c] - bp[c]) * m;
-            }
+    // Tile by tile: the union of two sparse surfaces far apart is a rectangle of mostly nothing,
+    // and transparent pixels in tiles neither surface has stay as they are.
+    let whole = area;
+    let same_default = base.default_pixel() == filtered.default_pixel();
+    for t in whole.tiles() {
+        let area = t.rect().intersect(&whole);
+        let bare = !base.has_tiles_in(area) && !filtered.has_tiles_in(area);
+        if area.is_empty() || (bare && (fmt.alpha || same_default)) {
            continue;
        }
-        let a = n - 1;
-        let oa = bp[a] + (fp[a] - bp[a]) * m;
-        op[a] = oa;
-        for c in 0..a {
-            let pm = bp[c] * bp[a] + (fp[c] * fp[a] - bp[c] * bp[a]) * m;
-            op[c] = if oa > 0.0 { (pm / oa).clamp(0.0, 1.0) } else { 0.0 };
+        let (b, f) = (base.read_region(area), filtered.read_region(area));
+        let mut k = Vec::new();
+        mask.values_into(area, &mut k);
+        let mut o = vec![0.0f32; b.len()];
+        for (i, ((bp, fp), op)) in b.chunks_exact(n).zip(f.chunks_exact(n)).zip(o.chunks_exact_mut(n)).enumerate() {
+            let m = k.get(i).copied().unwrap_or(1.0).clamp(0.0, 1.0);
+            if !fmt.alpha {
+                for c in 0..n {
+                    op[c] = bp[c] + (fp[c] - bp[c]) * m;
+                }
+                continue;
+            }
+            let a = n - 1;
+            let oa = bp[a] + (fp[a] - bp[a]) * m;
+            op[a] = oa;
+            for c in 0..a {
+                let pm = bp[c] * bp[a] + (fp[c] * fp[a] - bp[c] * bp[a]) * m;
+                op[c] = if oa > 0.0 { (pm / oa).clamp(0.0, 1.0) } else { 0.0 };
+            }
        }
+        out.write_region(area, &o);
    }
-    out.write_region(area, &o);
    out.prune();
    out
 }
@@ -61,38 +72,51 @@ pub fn unmix(base: &Surface, mixed: &Surface, mask: &LayerMask) -> Surface {
        return out;
    }
    let n = fmt.channels();
-    let (b, x) = (base.read_region(area), mixed.read_region(area));
-    let mut k = Vec::new();
-    mask.values_into(area, &mut k);
-    let mut o = vec![0.0f32; b.len()];
-    for (i, ((bp, xp), op)) in b.chunks_exact(n).zip(x.chunks_exact(n)).zip(o.chunks_exact_mut(n)).enumerate() {
-        let m = k.get(i).copied().unwrap_or(1.0).clamp(0.0, 1.0);
-        if m < HIDDEN {
-            // A transparent base pixel (Photoshop's cache keeps white there) stays empty, so the
-            // stored pixels don't grow past the content.
-            if !fmt.alpha || bp[n - 1] > 0.0 {
-                op.copy_from_slice(bp);
+    // Tile by tile: the union of two sparse surfaces far apart is a rectangle of mostly nothing,
+    // and transparent pixels in tiles neither surface has stay as they are.
+    let whole = area;
+    // Non-alpha still needs every tile whose defaults differ (an unmix through a non-unit mask
+    // would change that default, and the clone already holds the old one).
+    let same_default = base.default_pixel() == mixed.default_pixel();
+    for t in whole.tiles() {
+        let area = t.rect().intersect(&whole);
+        let bare = !base.has_tiles_in(area) && !mixed.has_tiles_in(area);
+        if area.is_empty() || (bare && (fmt.alpha || same_default)) {
+            continue;
+        }
+        let (b, x) = (base.read_region(area), mixed.read_region(area));
+        let mut k = Vec::new();
+        mask.values_into(area, &mut k);
+        let mut o = vec![0.0f32; b.len()];
+        for (i, ((bp, xp), op)) in b.chunks_exact(n).zip(x.chunks_exact(n)).zip(o.chunks_exact_mut(n)).enumerate() {
+            let m = k.get(i).copied().unwrap_or(1.0).clamp(0.0, 1.0);
+            if m < HIDDEN {
+                // A transparent base pixel (Photoshop's cache keeps white there) stays empty, so the
+                // stored pixels don't grow past the content.
+                if !fmt.alpha || bp[n - 1] > 0.0 {
+                    op.copy_from_slice(bp);
+                }
+                continue;
            }
-            continue;
-        }
-        if !fmt.alpha {
-            for c in 0..n {
-                op[c] = (bp[c] + (xp[c] - bp[c]) / m).clamp(0.0, 1.0);
+            if !fmt.alpha {
+                for c in 0..n {
+                    op[c] = (bp[c] + (xp[c] - bp[c]) / m).clamp(0.0, 1.0);
+                }
+                continue;
+            }
+            let a = n - 1;
+            let fa = (bp[a] + (xp[a] - bp[a]) / m).clamp(0.0, 1.0);
+            if fa <= 0.0 {
+                continue;
+            }
+            op[a] = fa;
+            for c in 0..a {
+                let pm = bp[c] * bp[a] + (xp[c] * xp[a] - bp[c] * bp[a]) / m;
+                op[c] = (pm / fa).clamp(0.0, 1.0);
            }
-            continue;
-        }
-        let a = n - 1;
-        let fa = (bp[a] + (xp[a] - bp[a]) / m).clamp(0.0, 1.0);
-        if fa <= 0.0 {
-            continue;
-        }
-        op[a] = fa;
-        for c in 0..a {
-            let pm = bp[c] * bp[a] + (xp[c] * xp[a] - bp[c] * bp[a]) / m;
-            op[c] = (pm / fa).clamp(0.0, 1.0);
        }
+        out.write_region(area, &o);
    }
-    out.write_region(area, &o);
    out.prune();
    out
 }
@@ -122,6 +146,19 @@ mod tests {
        m
    }

+    #[test]
+    fn mixing_sparse_surfaces_far_apart_does_not_build_their_union() {
+        // 10¹⁰ px between two single pixels would be 160 GB of floats read densely.
+        let fmt = PixelFormat::new(ColorMode::Rgb, SampleType::F32, true);
+        let (mut base, mut filtered) = (Surface::new(fmt), Surface::new(fmt));
+        base.write_pixel(0, 0, &[0.2, 0.4, 0.6, 1.0]);
+        filtered.write_pixel(100_000, 100_000, &[1.0, 0.0, 0.0, 1.0]);
+        let mixed = mix(&base, &filtered, &LayerMask::reveal_all());
+        assert_eq!(mixed.pixel(100_000, 100_000), vec![1.0, 0.0, 0.0, 1.0]);
+        let back = unmix(&base, &mixed, &LayerMask::reveal_all());
+        assert_eq!(back.pixel(100_000, 100_000), vec![1.0, 0.0, 0.0, 1.0]);
+    }
+
    #[test]
    fn unmix_inverts_mix_wherever_the_mask_shows_the_filter() {
        for fmt in [PixelFormat::new(ColorMode::Rgb, SampleType::F32, true), PixelFormat::new(ColorMode::Rgb, SampleType::F32, false)] {
@@ -150,6 +150,22 @@ fn scratch_reuse_preserves_pixels_and_bounds_retention() {
    });
 }

+/// Every layer the compositor draws, groups before their children: hidden layers and everything
+/// inside hidden groups are left out.
+pub(crate) fn visible_layers(layers: &[Layer]) -> Vec<&Layer> {
+    fn rec<'a>(layers: &'a [Layer], out: &mut Vec<&'a Layer>) {
+        for l in layers.iter().filter(|l| l.visible) {
+            out.push(l);
+            if let Some(children) = l.children() {
+                rec(children, out);
+            }
+        }
+    }
+    let mut out = Vec::new();
+    rec(layers, &mut out);
+    out
+}
+
 /// Composite the whole document over `rect`, in parallel 256² tiles on
 /// native targets (single-threaded on wasm).
 pub fn render(doc: &Document, rect: Rect) -> Buffer {
@@ -383,7 +399,11 @@ pub fn flatten(doc: &Document) -> Buffer {
 }

 /// Render an arbitrary subset: a single layer (e.g. for thumbnails), isolated.
+/// Rects past [`MAX_RENDER_PIXELS`] give an empty buffer.
 pub fn render_layer(layer: &Layer, rect: Rect) -> Buffer {
+    if u64::from(rect.width()) * u64::from(rect.height()) > MAX_RENDER_PIXELS {
+        return Buffer::transparent(Rect::EMPTY);
+    }
    let mut buf = Buffer::transparent(rect);
    let patterns = pattern::PreparedPatterns::new(&[], pattern::PREPARED_PATTERN_BYTES);
    composite_stack_at(
@@ -1720,14 +1740,34 @@ fn blend_if_level(x: f32) -> f32 {
    if (v - r).abs() < 1e-4 { r } else { v }
 }

+/// Blend If groups inside one another whose own colour is still rendered separately. Each extra
+/// level multiplies the work by the tree's depth, so the cost is polynomial, not exponential.
+const MAX_BLEND_IF_GROUP_DEPTH: u32 = 2;
+
+thread_local! {
+    /// Blend If groups being rendered for their own colour on this thread.
+    static BLEND_IF_GROUPS: std::cell::Cell<u32> = const { std::cell::Cell::new(0) };
+}
+
 /// Apply `layer`'s Blend If to a finished composite: `out` (the backdrop with the layer drawn)
 /// is mixed back towards `before` (the backdrop without it) where the ranges hide the layer.
 /// Mixing premultiplied colour by `k` equals compositing the layer at `k` × its alpha, since
 /// every blend mode's source-over result is linear in the source alpha.
 fn apply_blend_if(layer: &Layer, before: &Buffer, out: &mut Buffer, cx: &Ctx) {
    // "This Layer" is the layer's own colour; adjustment layers (no content of their own) are
-    // judged by their result.
-    let own = if matches!(layer.content, LayerContent::Adjustment(_)) { None } else { render_content(layer, out.rect, cx) };
+    // judged by their result. So are groups nested deeper than `MAX_BLEND_IF_GROUP_DEPTH` Blend
+    // Ifs: a group's own colour re-composites its children, which doubles the work at every
+    // level of Blend If groups inside one another.
+    let is_group = matches!(layer.content, LayerContent::Group(_));
+    let nested = BLEND_IF_GROUPS.with(std::cell::Cell::get);
+    let own = if matches!(layer.content, LayerContent::Adjustment(_)) || (is_group && nested >= MAX_BLEND_IF_GROUP_DEPTH) {
+        None
+    } else {
+        BLEND_IF_GROUPS.with(|d| d.set(nested + u32::from(is_group)));
+        let own = render_content(layer, out.rect, cx);
+        BLEND_IF_GROUPS.with(|d| d.set(nested));
+        own
+    };
    for (i, (p, b)) in out.px.iter_mut().zip(&before.px).enumerate() {
        let this = match &own {
            Some(o) => Some(o.px[i]).filter(|q| q[3] > 0.0),
@@ -2226,12 +2266,21 @@ fn fx_region_bytes(layers: &[Layer], canvas: Rect) -> u64 {
    bytes(layers, canvas, 0)
 }

+/// Most bytes one layer's effect maps and source may take to build, whatever the cache budget.
+const MAX_FX_BUILD_BYTES: u64 = if cfg!(target_pointer_width = "64") { 4 << 30 } else { 512 << 20 };
+
 /// The layer's effect maps over its region (layer bounds grown by the effect reach, within the
 /// canvas grown likewise; see [`effect_region`]), built once per layer state and shared by every
 /// tile. `rect` is what the caller composites the layer over (see [`effect_region`]).
 fn effect_maps(layer: &Layer, rect: Rect, cx: &Ctx) -> std::sync::Arc<effects::FxMaps> {
    use std::hash::{Hash, Hasher};
-    let (region, clipped) = effect_region(layer, rect, cx);
+    let (mut region, clipped) = effect_region(layer, rect, cx);
+    // Non-local effects keep their whole source region, which a few distant pixels stretch over
+    // the canvas: past the cap the effects are left out instead of allocating it.
+    let region_px = u64::from(region.width()) * u64::from(region.height());
+    if (effects::map_count(layer) as u64 + 1).saturating_mul(region_px).saturating_mul(4) > MAX_FX_BUILD_BYTES {
+        region = Rect::EMPTY;
+    }
    static NO_CACHE: std::sync::OnceLock<bool> = std::sync::OnceLock::new();
    if *NO_CACHE.get_or_init(|| std::env::var_os("PHOTOCRAFT_FX_NOCACHE").is_some()) {
        let shape = if region.is_empty() { Vec::new() } else { effect_shape(layer, region, cx) };
@@ -112,7 +112,8 @@ pub fn proxy_document(doc: &Document, k: u32) -> Document {
 /// with it. Empty vector masks have constant coverage, including inversion and density.
 /// Nonempty paths can still anchor fill layouts even when their mask density is zero.
 pub fn proxy_faithful(doc: &Document) -> bool {
-    doc.walk().iter().all(|(_, _, l)| !crate::effects::has_effects(l) && l.vector_mask.as_ref().is_none_or(|v| !v.enabled || v.path.is_empty()))
+    // Hidden layers (and what they hold) are not composited, so they don't count.
+    crate::visible_layers(&doc.layers).iter().all(|l| !crate::effects::has_effects(l) && l.vector_mask.as_ref().is_none_or(|v| !v.enabled || v.path.is_empty()))
 }

 #[cfg(test)]
@@ -263,6 +264,25 @@ mod tests {
        assert!(proxy_faithful(&doc));
    }

+    #[test]
+    fn a_hidden_layer_with_effects_does_not_rule_out_the_proxy_or_the_uniform_shortcut() {
+        let mut doc = two_colors(Size::new(32, 16), SampleType::U8);
+        doc.layers[1].effects.items.push(Effect::default_drop_shadow());
+        doc.layers[1].effects.enabled = true;
+        assert!(!proxy_faithful(&doc));
+        assert!(crate::uniform::UniformAreas::new(&doc).is_none());
+        doc.layers[1].visible = false;
+        assert!(proxy_faithful(&doc));
+        assert!(crate::uniform::UniformAreas::new(&doc).is_some());
+        // Inside a hidden group too.
+        doc.layers[1].visible = true;
+        let group = Layer::group("g", vec![doc.layers.remove(1)]);
+        doc.layers.push(group);
+        assert!(!proxy_faithful(&doc));
+        doc.layers[1].visible = false;
+        assert!(proxy_faithful(&doc));
+    }
+
    #[test]
    fn empty_vector_masks_keep_constant_coverage_at_every_depth() {
        for depth in [SampleType::U8, SampleType::U16, SampleType::F32] {
@@ -8,7 +8,7 @@ use std::sync::{Mutex, OnceLock};

 use photocraft_color::{PixelFormat, SampleType};
 use photocraft_doc::vector::ShapeLayer;
-use photocraft_geom::Rect;
+use photocraft_geom::{Rect, TileCoord};
 use photocraft_raster::Surface;

 /// (fill, stroke), last use, and the shape's pixels the stroke was fitted to (kept alive so
@@ -118,13 +118,42 @@ fn fit_stroke(fill: &Surface, stroke: &mut Surface, cache: &Surface, canvas: Rec
    if r.is_empty() {
        return;
    }
+    let float = stroke.format().sample == SampleType::F32;
+    // Tiles none of the three reaches read as the defaults; with transparent defaults the fit
+    // leaves the stroke's default pixel there, so those tiles are exact skips and we can scan
+    // just the union of the three surfaces' allocated tiles instead of every canvas tile (two
+    // sparse regions on a 300k×300k document would otherwise walk ~1.4M empty tiles). A
+    // non-transparent default is everywhere, so the scan falls back to the full rectangle.
+    let whole = r;
+    let bare_transparent = [fill, &*stroke, cache].iter().all(|s| s.default_pixel().iter().all(|v| *v == 0.0));
+    if bare_transparent {
+        let mut coords: Vec<TileCoord> = fill.tiles().map(|(c, _)| *c).chain(stroke.tiles().map(|(c, _)| *c)).chain(cache.tiles().map(|(c, _)| *c)).collect();
+        coords.sort_unstable();
+        coords.dedup();
+        for c in coords {
+            let r = c.rect().intersect(&whole);
+            if !r.is_empty() {
+                fit_stroke_rect(fill, stroke, cache, r, float);
+            }
+        }
+    } else {
+        for t in whole.tiles() {
+            let r = t.rect().intersect(&whole);
+            if !r.is_empty() {
+                fit_stroke_rect(fill, stroke, cache, r, float);
+            }
+        }
+    }
+}
+
+/// [`fit_stroke`] over one rectangle.
+fn fit_stroke_rect(fill: &Surface, stroke: &mut Surface, cache: &Surface, r: Rect, float: bool) {
    let n = r.width() as usize * r.height() as usize;
    let (mut f, mut s, mut c) = (vec![[0.0f32; 4]; n], vec![[0.0f32; 4]; n], vec![[0.0f32; 4]; n]);
    fill.read_rgba_into(r, &mut f);
    stroke.read_rgba_into(r, &mut s);
    cache.read_rgba_into(r, &mut c);
    let mut out = Vec::with_capacity(n * 4);
-    let float = stroke.format().sample == SampleType::F32;
    for ((fp, sp), cp) in f.iter().zip(&s).zip(&c) {
        let mut p = *sp;
        if fp[3] < 1.0 - 1e-3 {
@@ -1737,6 +1737,20 @@ fn vector_mask_combines_with_pixel_mask() {
    assert!(close4(px(&d, 6, 1), [1.0; 4]));
 }

+#[test]
+fn effect_maps_over_distant_pixels_are_not_built_over_the_canvas() {
+    // Two pixels 190000 apart: the effect source region spans 3.6·10¹⁰ px.
+    let mut doc = Document::new("big", Size::new(200_000, 200_000), ColorMode::Rgb, SampleType::U8);
+    let mut l = solid_layer("fx", Rect::new(0, 0, 1, 1), [1.0, 0.0, 0.0, 1.0]);
+    l.surface_mut().unwrap().fill_rect(Rect::new(190_000, 190_000, 190_001, 190_001), &[1.0, 0.0, 0.0, 1.0]);
+    l.effects.items.push(photocraft_doc::Effect::default_drop_shadow());
+    doc.layers.push(l);
+    let t = std::time::Instant::now();
+    let px = render(&doc, Rect::new(0, 0, 4, 4));
+    assert_eq!(px.px.len(), 16);
+    assert!(t.elapsed().as_secs() < 20, "{:?}", t.elapsed());
+}
+
 #[test]
 fn effect_maps_are_cached_and_invalidated_by_pixel_changes() {
    let mut doc = doc_white(64, 64);
@@ -2326,6 +2340,26 @@ fn blend_if_this_layer_hides_by_the_layers_own_value() {
    assert!(close4(px(&d, 1, 1), [0.0, 0.0, 0.0, 1.0]));
 }

+#[test]
+fn nested_blend_if_groups_do_not_double_the_work_at_every_level() {
+    // Each Blend If group re-renders its children for its own colour; 40 levels would be 2^40.
+    let mut inner = solid_layer("dot", Rect::new(0, 0, 4, 4), [1.0, 1.0, 1.0, 1.0]);
+    inner.blend_if.set(0, [range([10, 10], [255, 255]), FULL]);
+    let mut layers = vec![inner];
+    for _ in 0..40 {
+        let mut g = Layer::group("g", layers);
+        g.blend_if.set(0, [range([10, 10], [255, 255]), FULL]);
+        layers = vec![g];
+    }
+    // A black backdrop, not doc_white's white: the asserted white can then only come from the
+    // inner layer rendered through the whole group chain, not from the backdrop itself.
+    let mut d = Document::with_background("t", Size::new(4, 4), ColorMode::Rgb, SampleType::U8, Color::BLACK);
+    d.layers.extend(layers);
+    let t = std::time::Instant::now();
+    assert!(close4(px(&d, 1, 1), [1.0, 1.0, 1.0, 1.0]));
+    assert!(t.elapsed().as_secs() < 10, "{:?}", t.elapsed());
+}
+
 #[test]
 fn blend_if_underlying_layer_hides_by_the_backdrop_value() {
    // Backdrop black on the left, white on the right; a red layer over all of it.
@@ -32,7 +32,8 @@ impl<'a> UniformAreas<'a> {
            return None;
        }
        let mut surfaces = Vec::new();
-        for (_, _, l) in doc.walk() {
+        // Hidden layers are not composited: one with an effect must not rule the shortcut out.
+        for l in crate::visible_layers(&doc.layers) {
            let effects = l.effects.enabled && !l.effects.items.is_empty();
            if effects || l.vector_mask.is_some() || l.video.is_some() || l.blend == BlendMode::Dissolve {
                return None;