Repository navigation
Harden PDF/A and PDF/X conformance, memory safety, and Rust bounds checks #295
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changes from all commits
1438f71
ad860c0
4dabdf0
17921c2
ca4d927
7be3394
853966b
4ab2737
b4dfa5b
12dede4
097a8a2
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -66,14 +66,17 @@ public final class PdfCompressor { | |
| private static final byte[][] PDFA_MARKERS = { | ||
| "pdfaid:part>".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:part=\"".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:part='".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:conformance>".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:conformance=\"".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:conformance='".getBytes(StandardCharsets.US_ASCII), | ||
| "GTS_PDFA1".getBytes(StandardCharsets.US_ASCII), | ||
| }; | ||
| // PDF/A-1 forbids object streams and cross-reference streams; later parts allow them. | ||
| private static final byte[][] PDFA_PART1_MARKERS = { | ||
| "pdfaid:part>1<".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:part=\"1\"".getBytes(StandardCharsets.US_ASCII), | ||
| "pdfaid:part='1'".getBytes(StandardCharsets.US_ASCII), | ||
| "GTS_PDFA1".getBytes(StandardCharsets.US_ASCII), | ||
| }; | ||
| // Catalog /Metadata reference. XMP streams are often Flate-compressed, so a | ||
|
|
@@ -112,18 +115,21 @@ public final class PdfCompressor { | |
| // instead of a full pass per marker. | ||
| private static final int N_PDFA_PART = 0; | ||
| private static final int N_PDFA_PART_ATTR = 1; | ||
| private static final int N_PDFA_CONF = 2; | ||
| private static final int N_PDFA_CONF_ATTR = 3; | ||
| private static final int N_GTS_PDFA1 = 4; | ||
| private static final int N_PDFA1_PART = 5; | ||
| private static final int N_PDFA1_PART_ATTR = 6; | ||
| private static final int N_METADATA = 7; | ||
| private static final int N_BYTE_RANGE = 8; | ||
| private static final int N_TYPE_SIG = 9; | ||
| private static final int N_TYPE_SPACE_SIG = 10; | ||
| private static final int N_SUBFILTER = 11; | ||
| private static final int N_OBJSTM = 12; | ||
| private static final int N_NEEDLES = 13; | ||
| private static final int N_PDFA_PART_SQ = 2; | ||
| private static final int N_PDFA_CONF = 3; | ||
| private static final int N_PDFA_CONF_ATTR = 4; | ||
| private static final int N_PDFA_CONF_SQ = 5; | ||
| private static final int N_GTS_PDFA1 = 6; | ||
| private static final int N_PDFA1_PART = 7; | ||
| private static final int N_PDFA1_PART_ATTR = 8; | ||
| private static final int N_PDFA1_PART_SQ = 9; | ||
| private static final int N_METADATA = 10; | ||
| private static final int N_BYTE_RANGE = 11; | ||
| private static final int N_TYPE_SIG = 12; | ||
| private static final int N_TYPE_SPACE_SIG = 13; | ||
| private static final int N_SUBFILTER = 14; | ||
| private static final int N_OBJSTM = 15; | ||
| private static final int N_NEEDLES = 16; | ||
|
|
||
| // The /ID key is matched byte-by-byte and its range is packed into a long, so the | ||
| // signed fast path performs no allocation of its own. | ||
|
|
@@ -140,8 +146,11 @@ public final class PdfCompressor { | |
| PDFA_MARKERS[2], | ||
| PDFA_MARKERS[3], | ||
| PDFA_MARKERS[4], | ||
| PDFA_MARKERS[5], | ||
| PDFA_MARKERS[6], | ||
| PDFA_PART1_MARKERS[0], | ||
| PDFA_PART1_MARKERS[1], | ||
| PDFA_PART1_MARKERS[2], | ||
| METADATA_MARKER, | ||
| BYTE_RANGE, | ||
| SIG_MARKERS[0], | ||
|
|
@@ -426,8 +435,9 @@ && serializesIdentically(doc, sourceBytes, openStateBytes)) { | |
| // PDF/A-1 forbids object streams and cross-reference streams, so a | ||
| // generated-xref result would break the conformance it claims. | ||
| byte[] opt = pdfA1 | ||
| ? PdfOptimizer.optimize(resultBytes, PdfOptimizer.DEFAULT, 9, | ||
| PdfOptimizer.OBJECT_STREAMS_PRESERVE, | ||
| ? PdfOptimizer.optimize(resultBytes, | ||
| PdfOptimizer.RECOMPRESS_FLATE | PdfOptimizer.COMPRESS_STREAMS, 9, | ||
| PdfOptimizer.OBJECT_STREAMS_DISABLE, | ||
| PdfOptimizer.STREAM_DATA_COMPRESS, | ||
| PdfOptimizer.DECODE_LEVEL_GENERALIZED) | ||
| : PdfOptimizer.optimize(resultBytes, | ||
|
|
@@ -443,21 +453,22 @@ && serializesIdentically(doc, sourceBytes, openStateBytes)) { | |
| resultBytes = opt; | ||
| streamsOptimized = true; | ||
| actions.add(pdfA1 | ||
| ? "qpdf: recompressed streams (object streams preserved for PDF/A-1)" | ||
| ? "qpdf: recompressed streams (object streams disabled for PDF/A-1)" | ||
| : "qpdf: recompressed streams and generated object streams"); | ||
| } else { | ||
| warnings.add("qpdf stream pass did not reduce size; kept original bytes"); | ||
| } | ||
| } else if (!signed && opts.removeUnusedObjects()) { | ||
| // Compact without restructuring object streams. Explicit flags are | ||
| // required: the all-DEFAULT form measured worse than the source on | ||
| // some inputs (inflating streams instead of recompressing them). | ||
| // Compact without restructuring object streams (disable for PDF/A-1). | ||
| // Explicit flags are required: the all-DEFAULT form measured worse than the | ||
| // source on some inputs (inflating streams instead of recompressing them). | ||
| byte[] opt = PdfOptimizer.optimize(resultBytes, | ||
| PdfOptimizer.RECOMPRESS_FLATE | PdfOptimizer.COMPRESS_STREAMS | ||
| | PdfOptimizer.DEDUPLICATE_IMAGES | ||
| | (jbig2 ? PdfOptimizer.JBIG2 : 0) | ||
| | (opts.ccittG4() ? PdfOptimizer.CCITT_G4 : 0), 9, | ||
| PdfOptimizer.OBJECT_STREAMS_PRESERVE, | ||
| pdfA1 ? PdfOptimizer.OBJECT_STREAMS_DISABLE | ||
| : PdfOptimizer.OBJECT_STREAMS_PRESERVE, | ||
| PdfOptimizer.STREAM_DATA_COMPRESS, | ||
| PdfOptimizer.DECODE_LEVEL_GENERALIZED); | ||
| if (opt != null && opt.length < resultBytes.length) { | ||
|
|
@@ -646,7 +657,8 @@ private static int pdfaPart(byte[] pdf, Markers m) { | |
| return -1; | ||
| } | ||
| boolean rawPdfa = m.first[N_PDFA_PART] >= 0 || m.first[N_PDFA_PART_ATTR] >= 0 | ||
| || m.first[N_PDFA_CONF] >= 0 || m.first[N_PDFA_CONF_ATTR] >= 0 | ||
| || m.first[N_PDFA_PART_SQ] >= 0 || m.first[N_PDFA_CONF] >= 0 | ||
| || m.first[N_PDFA_CONF_ATTR] >= 0 || m.first[N_PDFA_CONF_SQ] >= 0 | ||
|
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Resolve the PDF/A part after a conformance-only match. If compressed catalog XMP claims part 1, an Info string containing When the raw match does not establish the part, inspect the catalog metadata before selecting the structural policy. Add a regression with this combination. |
||
| || m.first[N_GTS_PDFA1] >= 0; | ||
| byte[] xmp = null; | ||
| if (!rawPdfa) { | ||
|
|
@@ -656,7 +668,7 @@ private static int pdfaPart(byte[] pdf, Markers m) { | |
| } | ||
| } | ||
| boolean part1 = m.first[N_PDFA1_PART] >= 0 || m.first[N_PDFA1_PART_ATTR] >= 0 | ||
| || m.first[N_GTS_PDFA1] >= 0 | ||
| || m.first[N_PDFA1_PART_SQ] >= 0 || m.first[N_GTS_PDFA1] >= 0 | ||
| || (xmp != null && containsAny(xmp, PDFA_PART1_MARKERS)); | ||
| return part1 ? 1 : 2; | ||
| } | ||
|
|
||
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -5,6 +5,7 @@ | |
| import stirling.software.jpdfium.exception.JPDFiumException; | ||
| import stirling.software.jpdfium.panama.AnnotationBindings; | ||
| import stirling.software.jpdfium.panama.NativeRuntime; | ||
| import stirling.software.jpdfium.panama.AnnotationBindings; | ||
| import stirling.software.jpdfium.panama.PageEditBindings; | ||
| import stirling.software.jpdfium.panama.QpdfLib; | ||
| import stirling.software.jpdfium.util.NativeJsonParser; | ||
|
|
@@ -574,6 +575,167 @@ | |
| } | ||
| } | ||
|
|
||
| /** | ||
| * Font base names used by text objects that are actually painted. Text in | ||
| * rendering mode 3 (invisible OCR layers) is excluded. Nested form XObjects | ||
| * are traversed when the native build supports it. | ||
| */ | ||
| public static Set<String> visibleFontKeys(PdfDocument doc) { | ||
| Set<String> keys = new LinkedHashSet<>(); | ||
| for (int i = 0; i < doc.pageCount(); i++) { | ||
| try (PdfPage page = doc.page(i)) { | ||
| MemorySegment raw = page.rawHandleForRead(); | ||
| int annotations = | ||
| (int) AnnotationBindings.FPDFPage_GetAnnotCount.invokeExact(raw); | ||
| // Annotation appearance streams are not walked here, so the | ||
| // visible set would be incomplete. | ||
| if (annotations > 0) return null; | ||
| int count = (int) PageEditBindings.FPDFPage_CountObjects.invokeExact(raw); | ||
| for (int j = 0; j < count; j++) { | ||
| MemorySegment obj = | ||
| (MemorySegment) PageEditBindings.FPDFPage_GetObject.invokeExact(raw, j); | ||
| collectVisibleFonts(obj, keys, 0); | ||
| } | ||
| } catch (Throwable t) { | ||
| NativeRuntime.rethrowFatal(t); | ||
| // A page that cannot be walked tells us nothing; signal an | ||
| // incomplete result instead of hiding fonts. | ||
| return null; | ||
| } | ||
|
Comment on lines
+599
to
+604
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Discard incomplete visibility results. If page 0 contributes font A and enumeration of page 1 fails, this returns Return an explicit incomplete result, or disable suppression for the whole document after any scan failure. Keep suppression enabled only after a complete scan. |
||
| } | ||
| return keys; | ||
| } | ||
|
|
||
| /** | ||
| * Embedded font base names whose stored program is empty, unreadable, or | ||
| * does not start with a recognized font magic. Such a font renders with a | ||
| * fallback and should be replaced when an embeddable substitute exists. | ||
| */ | ||
| public static Set<String> corruptEmbeddedKeys(PdfDocument doc) { | ||
| Set<String> keys = new LinkedHashSet<>(); | ||
| for (int i = 0; i < doc.pageCount(); i++) { | ||
| try (PdfPage page = doc.page(i)) { | ||
| MemorySegment raw = page.rawHandleForRead(); | ||
| int count = (int) PageEditBindings.FPDFPage_CountObjects.invokeExact(raw); | ||
| for (int j = 0; j < count; j++) { | ||
| MemorySegment obj = | ||
| (MemorySegment) PageEditBindings.FPDFPage_GetObject.invokeExact(raw, j); | ||
| collectCorruptFonts(obj, keys, 0); | ||
| } | ||
| } catch (Throwable t) { | ||
| NativeRuntime.rethrowFatal(t); | ||
| // Best effort; leave unknown fonts alone. | ||
| } | ||
| } | ||
| return keys; | ||
| } | ||
|
|
||
| /** | ||
| * Names of fonts whose descriptor points at an embedded program that is not a | ||
| * recognizable font program. Uses the qpdf inventory so a program PDFium refuses to expose | ||
| * is still detected. | ||
| */ | ||
| static List<String> corruptEmbeddedKeys(byte[] pdf) { | ||
| if (pdf == null || pdf.length == 0 || !QpdfLib.isFontInventorySupported()) return List.of(); | ||
| String json = QpdfLib.inspectFonts(pdf); | ||
| if (json == null || json.length() <= 2) return List.of(); | ||
| Set<String> names = new LinkedHashSet<>(); | ||
| for (Map<String, String> field : NativeJsonParser.parseArray(json)) { | ||
| if ("true".equals(field.get("program")) | ||
| && "false".equals(field.get("programValid"))) { | ||
| String name = field.getOrDefault("baseFont", ""); | ||
| if (!name.isEmpty()) names.add(name); | ||
| } | ||
| } | ||
| return List.copyOf(names); | ||
| } | ||
|
|
||
| private static void collectCorruptFonts(MemorySegment obj, Set<String> keys, int depth) { | ||
| if (depth > 64 || obj == null || obj.equals(MemorySegment.NULL)) return; | ||
| try { | ||
| int type = (int) PageEditBindings.FPDFPageObj_GetType.invokeExact(obj); | ||
| if (type == 1) { | ||
| MemorySegment font = | ||
| (MemorySegment) PageEditBindings.FPDFTextObj_GetFont.invokeExact(obj); | ||
| String name = getFontBaseName(font); | ||
| if (name.isEmpty() || "<unknown>".equals(name)) return; | ||
| if ((int) PageEditBindings.FPDFFont_GetIsEmbedded.invokeExact(font) == 1) { | ||
| byte[] data = getFontProgram(font); | ||
| if (!looksLikeFontProgram(data)) { | ||
| keys.add(name); | ||
| } | ||
| } | ||
| } else if (type == 5 && PageEditBindings.formTraversalAvailable()) { | ||
| int children = | ||
| (int) PageEditBindings.FPDFFormObj_CountObjects.invokeExact(obj); | ||
| for (int i = 0; i < children; i++) { | ||
| collectCorruptFonts(PageEditBindings.formObject(obj, i), keys, depth + 1); | ||
| } | ||
| } | ||
| } catch (Throwable t) { | ||
| NativeRuntime.rethrowFatal(t); | ||
| // Best effort. | ||
| } | ||
| } | ||
|
|
||
| private static byte[] getFontProgram(MemorySegment font) { | ||
| try { | ||
| long needed = getFontDataSize(font); | ||
| if (needed <= 0 || needed > 64L * 1024 * 1024) return null; | ||
| try (Arena arena = Arena.ofConfined()) { | ||
| MemorySegment buf = arena.allocate(needed); | ||
| MemorySegment outLen = arena.allocate(ValueLayout.JAVA_LONG); | ||
| int ok = (int) PageEditBindings.FPDFFont_GetFontData.invokeExact( | ||
| font, buf, needed, outLen); | ||
| if (ok == 0) return null; | ||
| long len = Math.min(needed, outLen.get(ValueLayout.JAVA_LONG, 0)); | ||
| if (len <= 0) return null; | ||
| return buf.asSlice(0, len).toArray(ValueLayout.JAVA_BYTE); | ||
| } | ||
| } catch (Throwable t) { | ||
| NativeRuntime.rethrowFatal(t); | ||
| return null; | ||
| } | ||
| } | ||
|
|
||
| static boolean looksLikeFontProgram(byte[] data) { | ||
| if (data == null || data.length < 4) return false; | ||
| if (data[0] == 0x00 && data[1] == 0x01 && data[2] == 0x00 && data[3] == 0x00) return true; | ||
| if (data[0] == 0x01 && data[1] == 0x00 && data[2] == 0x04 && data[3] == 0x04) return true; | ||
| String head = new String(data, 0, Math.min(4, data.length), StandardCharsets.ISO_8859_1); | ||
| if ("OTTO".equals(head) || "true".equals(head) || "ttcf".equals(head)) return true; | ||
| if (data.length >= 4 && data[0] == '%' && data[1] == '!') return true; | ||
|
|
||
| return data.length >= 2 && (data[0] & 0xFF) == 0x80 && data[1] == 0x01; | ||
|
|
||
| } | ||
|
|
||
| private static void collectVisibleFonts(MemorySegment obj, Set<String> keys, int depth) { | ||
| if (depth > 64 || obj == null || obj.equals(MemorySegment.NULL)) return; | ||
| try { | ||
| int type = (int) PageEditBindings.FPDFPageObj_GetType.invokeExact(obj); | ||
| if (type == 1) { | ||
| int renderMode = | ||
| (int) PageEditBindings.FPDFTextObj_GetTextRenderMode.invokeExact(obj); | ||
| if (renderMode != 3) { | ||
| MemorySegment font = | ||
| (MemorySegment) PageEditBindings.FPDFTextObj_GetFont.invokeExact(obj); | ||
| String name = getFontBaseName(font); | ||
| if (!name.isEmpty() && !"<unknown>".equals(name)) { | ||
| keys.add(PdfFontEmbedder.key(name)); | ||
| } | ||
| } | ||
| } else if (type == 5 && PageEditBindings.formTraversalAvailable()) { | ||
| int children = | ||
| (int) PageEditBindings.FPDFFormObj_CountObjects.invokeExact(obj); | ||
| for (int i = 0; i < children; i++) { | ||
| MemorySegment child = PageEditBindings.formObject(obj, i); | ||
| collectVisibleFonts(child, keys, depth + 1); | ||
| } | ||
| } | ||
| } catch (Throwable _) { | ||
| // Best-effort scan; keep whatever was collected. | ||
| } | ||
| } | ||
|
|
||
| private static String getFontBaseName(MemorySegment font) { | ||
| try (Arena arena = Arena.ofConfined()) { | ||
| // First call: get required buffer size | ||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Script executed:
rg -n -i 'XStep|YStep|applyDocumentOptions|clamp' native/bridge/src/jpdfium_qpdf.cpp sed -n 425,475p jpdfium/src/main/java/stirling/software/jpdfium/doc/PdfCompressor.javaRepository: Stirling-Tools/JPDFium
Length of output: 4477
Keep tiling-pattern steps unchanged in the PDF/A-1 stream pass.
The stream pass calls
PdfOptimizer.optimizewithRECOMPRESS_FLATE | COMPRESS_STREAMS. NativeapplyDocumentOptionsrunsclampPatternStepsbefore it checks any flag. That clamp replaces a tiling-pattern/XStepor/YStepwith an absolute value below1.0by1.0. A pattern with/XStep 0.5therefore renders with different tile spacing after this pass.The Java code adopts the output whenever it is smaller (
opt.length < resultBytes.length). It does not compare rendering, so the changed pattern is accepted without a check.This pass must leave tiling-pattern steps unchanged, or reject its output when they change. Keep the overflow clamp for non-finite and out-of-range steps, because the native source documents that protection.
Add a regression. Render a PDF/A-1 page that contains a
/XStep 0.5tiling pattern before and after this pass, and assert that the rendering matches.