Merge pull request #1400 from m-holger/pr1266

Fix parsing of object streams containing objects with no white-space between them

Merge pull request #1400 from m-holger/pr1266
Fix parsing of object streams containing objects with no white-space between them
m-holger · GitHub
2 parents 5eb9a18a b37fc717
Showing 7 changed files with 134 additions and 12 deletions
include/qpdf/QPDF.hh
libqpdf/QPDFParser.cc
libqpdf/QPDF_objects.cc
libqpdf/qpdf/InputSource_private.hh
libqpdf/qpdf/QPDFParser.hh
manual/release-notes.rst
qpdf/qtest/object-stream.test
@@ -45,6 +45,11 @@
 #include <qpdf/QPDFWriter.hh>
 #include <qpdf/QPDFXRefEntry.hh>
+namespace qpdf::is
+{
+    class OffsetBuffer;
+}
+
 class QPDF_Stream;
 class BitStream;
 class BitWriter;
@@ -785,7 +790,7 @@ class QPDF
     QPDFObjectHandle readObject(std::string const& description, QPDFObjGen og);
     void readStream(QPDFObjectHandle& object, QPDFObjGen og, qpdf_offset_t offset);
     void validateStreamLineEnd(QPDFObjectHandle& object, QPDFObjGen og, qpdf_offset_t offset);
-    QPDFObjectHandle readObjectInStream(BufferInputSource& input, int stream_id, int obj_id);
+    QPDFObjectHandle readObjectInStream(qpdf::is::OffsetBuffer& input, int stream_id, int obj_id);
     size_t recoverStreamLength(
         std::shared_ptr<InputSource> input, QPDFObjGen og, qpdf_offset_t stream_offset);
     QPDFTokenizer::Token readToken(InputSource&, size_t max_len = 0);
@@ -12,6 +12,7 @@
 #include <memory>
 using namespace std::literals;
+using namespace qpdf;
 using ObjectPtr = std::shared_ptr<QPDFObject>;
@@ -87,7 +88,7 @@ QPDFParser::parse(
 std::pair<QPDFObjectHandle, bool>
 QPDFParser::parse(
-    BufferInputSource& input, int stream_id, int obj_id, qpdf::Tokenizer& tokenizer, QPDF& context)
+    is::OffsetBuffer& input, int stream_id, int obj_id, qpdf::Tokenizer& tokenizer, QPDF& context)
 {
     bool empty{false};
     auto result = QPDFParser(
@@ -1288,7 +1288,7 @@ QPDF::validateStreamLineEnd(QPDFObjectHandle&amp; object, QPDFObjGen og, qpdf_offset
 }
 QPDFObjectHandle
-QPDF::readObjectInStream(BufferInputSource& input, int stream_id, int obj_id)
+QPDF::readObjectInStream(is::OffsetBuffer& input, int stream_id, int obj_id)
 {
     auto [object, empty] = QPDFParser::parse(input, stream_id, obj_id, m->tokenizer, *this);
     if (empty) {
@@ -1645,12 +1645,26 @@ QPDF::resolveObjectsInStream(int obj_stream_number)
             "object stream " + std::to_string(obj_stream_number) + " has incorrect keys");
     }
-    std::vector<std::pair<int, long long>> offsets;
+    // id, offset, size
+    std::vector<std::tuple<int, qpdf_offset_t, size_t>> offsets;
     auto bp = obj_stream.getStreamData(qpdf_dl_specialized);
+
     BufferInputSource input("", bp.get());
+    const auto b_size = bp->getSize();
+    const auto end_offset = static_cast<qpdf_offset_t>(b_size);
+    auto b_start = bp->getBuffer();
+
+    if (first >= end_offset) {
+        throw damagedPDF(
+            "object " + std::to_string(obj_stream_number) + " 0",
+            "object stream " + std::to_string(obj_stream_number) + " has invalid /First entry");
+    }
+
+    int id = 0;
     long long last_offset = -1;
+    bool is_first = true;
     for (unsigned int i = 0; i < n; ++i) {
         auto tnum = readToken(input);
         auto toffset = readToken(input);
@@ -1682,26 +1696,45 @@ QPDF::resolveObjectsInStream(int obj_stream_number)
                     std::to_string(last_offset) + ")"));
             continue;
         }
-        last_offset = offset;
         if (num > m->xref_table_max_id) {
             continue;
         }
-        offsets.emplace_back(num, offset + first);
+        if (first + offset >= end_offset) {
+            warn(damaged(num, offset, "offset is too large"));
+            continue;
+        }
+
+        if (is_first) {
+            is_first = false;
+        } else {
+            offsets.emplace_back(
+                id, last_offset + first, static_cast<size_t>(offset - last_offset));
+        }
+
+        last_offset = offset;
+        id = num;
+    }
+
+    if (!is_first) {
+        // We found at least one valid entry.
+        offsets.emplace_back(
+            id, last_offset + first, b_size - static_cast<size_t>(last_offset + first));
     }
     // To avoid having to read the object stream multiple times, store all objects that would be
     // found here in the cache.  Remember that some objects stored here might have been overridden
     // by new objects appended to the file, so it is necessary to recheck the xref table and only
     // cache what would actually be resolved here.
-    for (auto const& [id, offset]: offsets) {
-        QPDFObjGen og(id, 0);
+    for (auto const& [obj_id, obj_offset, obj_size]: offsets) {
+        QPDFObjGen og(obj_id, 0);
         auto entry = m->xref_table.find(og);
         if (entry != m->xref_table.end() && entry->second.getType() == 2 &&
             entry->second.getObjStreamNumber() == obj_stream_number) {
-            input.seek(offset, SEEK_SET);
-            QPDFObjectHandle oh = readObjectInStream(input, obj_stream_number, id);
+            Buffer obj_buffer{b_start + obj_offset, obj_size};
+            is::OffsetBuffer in("", &obj_buffer, obj_offset);
+            auto oh = readObjectInStream(in, obj_stream_number, obj_id);
             updateCache(og, oh.getObj(), end_before_space, end_after_space);
         } else {
             QTC::TC("qpdf", "QPDF not caching overridden objstm object");
 #ifndef QPDF_INPUTSOURCE_PRIVATE_HH
 #define QPDF_INPUTSOURCE_PRIVATE_HH
+#include <qpdf/BufferInputSource.hh>
 #include <qpdf/InputSource.hh>
+#include <limits>
+#include <sstream>
+#include <stdexcept>
+
+namespace qpdf::is
+{
+    class OffsetBuffer final: public InputSource
+    {
+      public:
+        OffsetBuffer(std::string const& description, Buffer* buf, qpdf_offset_t global_offset) :
+            proxied(description, buf),
+            global_offset(global_offset)
+        {
+            if (global_offset < 0) {
+                throw std::logic_error("is::OffsetBuffer constructed with negative offset");
+            }
+            last_offset = global_offset;
+        }
+
+        ~OffsetBuffer() final = default;
+
+        qpdf_offset_t
+        findAndSkipNextEOL() final
+        {
+            return proxied.findAndSkipNextEOL() + global_offset;
+        }
+
+        std::string const&
+        getName() const final
+        {
+            return proxied.getName();
+        }
+
+        qpdf_offset_t
+        tell() final
+        {
+            return proxied.tell() + global_offset;
+        }
+
+        void
+        seek(qpdf_offset_t offset, int whence) final
+        {
+            if (whence == SEEK_SET) {
+                proxied.seek(offset - global_offset, whence);
+            } else {
+                proxied.seek(offset, whence);
+            }
+        }
+
+        void
+        rewind() final
+        {
+            seek(0, SEEK_SET);
+        }
+
+        size_t
+        read(char* buffer, size_t length) final
+        {
+            size_t result = proxied.read(buffer, length);
+            setLastOffset(proxied.getLastOffset() + global_offset);
+            return result;
+        }
+
+        void
+        unreadCh(char ch) final
+        {
+            proxied.unreadCh(ch);
+        }
+
+      private:
+        BufferInputSource proxied;
+        qpdf_offset_t global_offset;
+    };
+
+} // namespace qpdf::is
+
 inline size_t
 InputSource::read(std::string& str, size_t count, qpdf_offset_t at)
 {
 #ifndef QPDFPARSER_HH
 #define QPDFPARSER_HH
+#include <qpdf/InputSource_private.hh>
 #include <qpdf/QPDFObjectHandle_private.hh>
 #include <qpdf/QPDFObject_private.hh>
 #include <qpdf/QPDFTokenizer_private.hh>
@@ -38,7 +39,7 @@ class QPDFParser
         QPDF& context);
     static std::pair<QPDFObjectHandle, bool> parse(
-        BufferInputSource& input,
+        qpdf::is::OffsetBuffer& input,
         int stream_id,
         int obj_id,
         qpdf::Tokenizer& tokenizer,
@@ -21,6 +21,11 @@ more detail.
       integer object. Previously the method returned false if the first
       dictionary object was not a linearization parameter dictionary.
+    = Fix parsing of object streams containing objects not seperated by
+      white-space. Pre-2020 editions of the PDF specification incorrectly
+      stated that white-space was required between objects. qpdf relied on this
+      when parsing object streams.
+
     - Fix two object stream error/warning messages that reported the wrong
       object id.
@@ -124,7 +124,7 @@ $td-&gt;runtest(&quot;adjacent compressed objects&quot;,
              {$td->COMMAND => "test_driver 99 no-space-compressed-object.pdf"},
              {$td->FILE => "no-space-compressed-object.out",
               $td->EXIT_STATUS => 0},
-             $td->EXPECT_FAILURE);
+             $td->NORMALIZE_NEWLINES);
 cleanup();
 $td->report(calc_ntests($n_tests, $n_compare_pdfs));