forked from 1b-agent/opencadstudio-upstream
fix(pdf): clamp display rasters and bound the print raster
Bug: The PDF underlay renderer sized its pixmap from page size x requested DPI with no ceiling: a large page at a high raster DPI (display or print) computes a multi-gigabyte image - testing reached a 17 GB pixmap and a 543-second render attempt before the allocator gave up. Reproduce: 1. Attach a large-page PDF; set a high raster DPI (print above 150 DPI, or display with DPI raised). 2. Before: minutes of rendering and a multi-GB allocation, then abort. 3. After: display rasters are clamped so the longest side is at most 3072 (like DWF/DGN, the effective DPI is stored); print rasters over 268,435,456 pixels are refused with an error instead of rendered. Fix: Display (at or below 150 DPI): the scale is clamped to RASTER_SIDE/longest before the pixmap is allocated, and the DPI actually used is stored for downstream math. Print (above 150 DPI): refused when width x height exceeds MAX_PRINT_RASTER_PIXELS (256 Mpx) or is non-finite - print DPI is never silently rescaled because the sheet-mm-from-exact-DPI conversion must stay exact; the user gets an error instead of a hang. A regression test covers the hostile page x DPI case (about 1 second, previously 543 seconds and 17 GB).
This commit is contained in:
1 parent
21c4d6ab41
commit
86642bb639
1 file changed
+61
-2
@@ -18,6 +18,13 @@ pub struct PdfPage {
|
|||||||
|
|
||||||
const RASTER_DPI: f32 = 150.0;
|
const RASTER_DPI: f32 = 150.0;
|
||||||
|
|
||||||
|
/// Ceiling on the print raster's pixel count (RGBA ⇒ ×4 bytes): an ANSI E
|
||||||
|
/// sheet at 300 DPI is ~135 M px and stays under it, while a crafted
|
||||||
|
/// MediaBox cannot ask hayro for a multi-gigabyte pixmap — its `u16`
|
||||||
|
/// dimensions saturate at 65 535, i.e. a 17 GB allocation (abort or a
|
||||||
|
/// multi-minute freeze).
|
||||||
|
const MAX_PRINT_RASTER_PIXELS: f64 = 268_435_456.0;
|
||||||
|
|
||||||
/// Source path, page, DPI bits and whether the page background is transparent.
|
/// Source path, page, DPI bits and whether the page background is transparent.
|
||||||
type PageKey = (String, String, u32, bool);
|
type PageKey = (String, String, u32, bool);
|
||||||
|
|
||||||
@@ -115,7 +122,33 @@ fn rasterize_uncached(path: &str, page: &str, dpi: f32, transparent: bool) -> Op
|
|||||||
let pdf = Pdf::new(bytes).ok()?;
|
let pdf = Pdf::new(bytes).ok()?;
|
||||||
let page_no = page.trim().parse::<usize>().unwrap_or(1).max(1);
|
let page_no = page.trim().parse::<usize>().unwrap_or(1).max(1);
|
||||||
let page = pdf.pages().get(page_no - 1)?;
|
let page = pdf.pages().get(page_no - 1)?;
|
||||||
let scale = dpi / 72.0;
|
let mut scale = dpi / 72.0;
|
||||||
|
let mut effective_dpi = dpi;
|
||||||
|
// Page size in points, straight from the file's MediaBox/CropBox.
|
||||||
|
let (page_w, page_h) = page.render_dimensions();
|
||||||
|
let longest = (page_w as f64).max(page_h as f64);
|
||||||
|
if longest.is_finite() && longest > 0.0 {
|
||||||
|
if dpi <= DISPLAY_DPI {
|
||||||
|
// Screen rasters get the same longest-side clamp the DWF/DGN
|
||||||
|
// underlays receive (`underlay_vector::RASTER_SIDE`); the page
|
||||||
|
// maps onto the same world rect regardless of pixel count.
|
||||||
|
if longest * f64::from(dpi) / 72.0 > super::underlay_vector::RASTER_SIDE {
|
||||||
|
let clamped = super::underlay_vector::RASTER_SIDE / longest;
|
||||||
|
scale = clamped as f32;
|
||||||
|
effective_dpi = (clamped * 72.0) as f32;
|
||||||
|
}
|
||||||
|
} else {
|
||||||
|
// Print keeps the exact DPI — the sheet size in millimetres is
|
||||||
|
// derived from it (`page.width / dpi`) — so an oversized page
|
||||||
|
// is refused instead of rescaled (a rescale would print the
|
||||||
|
// wrong physical size).
|
||||||
|
let pw = page_w as f64 * f64::from(dpi) / 72.0;
|
||||||
|
let ph = page_h as f64 * f64::from(dpi) / 72.0;
|
||||||
|
if !(pw.is_finite() && ph.is_finite()) || pw * ph > MAX_PRINT_RASTER_PIXELS {
|
||||||
|
return None;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
let pixmap = hayro::render(
|
let pixmap = hayro::render(
|
||||||
page,
|
page,
|
||||||
&RenderCache::new(),
|
&RenderCache::new(),
|
||||||
@@ -137,7 +170,7 @@ fn rasterize_uncached(path: &str, page: &str, dpi: f32, transparent: bool) -> Op
|
|||||||
pixels: Arc::new(pixels),
|
pixels: Arc::new(pixels),
|
||||||
width,
|
width,
|
||||||
height,
|
height,
|
||||||
dpi,
|
dpi: effective_dpi,
|
||||||
}))
|
}))
|
||||||
}
|
}
|
||||||
|
|
||||||
@@ -347,4 +380,30 @@ mod tests {
|
|||||||
assert_eq!((page.width, page.height), (150, 150));
|
assert_eq!((page.width, page.height), (150, 150));
|
||||||
assert_eq!(page.pixels.len(), 150 * 150 * 4);
|
assert_eq!(page.pixels.len(), 150 * 150 * 4);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/// A 15 m square MediaBox at 150 DPI would be 88 583 px a side: hayro
|
||||||
|
/// saturates its `u16` dimensions to 65 535 and `Pixmap::new` asks for
|
||||||
|
/// 65 535² × 4 = **17 GB** — the report's allocation-failure abort.
|
||||||
|
/// The display raster must clamp to `RASTER_SIDE` instead, and the
|
||||||
|
/// print raster (exact DPI, physical size derived from it) must be
|
||||||
|
/// refused rather than rescaled.
|
||||||
|
#[test]
|
||||||
|
fn an_enormous_page_is_clamped_or_refused_never_a_17gb_pixmap() {
|
||||||
|
let mut document = PdfDocument::new("PDF underlay huge-page test");
|
||||||
|
document
|
||||||
|
.pages
|
||||||
|
.push(OutputPage::new(Mm(15_000.0), Mm(15_000.0), Vec::new()));
|
||||||
|
let bytes = document.save(&PdfSaveOptions::default(), &mut Vec::new());
|
||||||
|
let path = "memory://pdf-underlay-huge-page.pdf";
|
||||||
|
|
||||||
|
register_source(path, Arc::new(bytes));
|
||||||
|
let page = rasterize_page_display(path, "1").expect("display raster is clamped");
|
||||||
|
assert_eq!((page.width, page.height), (3072, 3072));
|
||||||
|
assert!(page.dpi < DISPLAY_DPI, "clamped raster reports its effective DPI");
|
||||||
|
|
||||||
|
assert!(
|
||||||
|
rasterize_page_at_dpi(path, "1", 300.0).is_none(),
|
||||||
|
"a print raster over the pixel budget is refused, not rescaled"
|
||||||
|
);
|
||||||
|
}
|
||||||
}
|
}
|
||||||
Reference in new issue
Block a user