Fix a JBIG2 decoding regression.

When trying to sanity check SBNUMINSTANCES for JBIG2 text regions for
https://crbug.com/837192, the check did not take data encoding into
account. Fix this by assuming there is a potential for 16:1 compression
in the lossless encoding.

BUG=chromium:871848

Change-Id: I991dacff9fbefa4e302f5ce92a355902ef94308a
Reviewed-on: https://pdfium-review.googlesource.com/39790
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Commit-Queue: Lei Zhang <thestig@chromium.org>

core/fxcodec/jbig2/JBig2_Context.cpp

diff --git a/core/fxcodec/jbig2/JBig2_Context.cpp b/core/fxcodec/jbig2/JBig2_Context.cpp
index ec81990..5bf9df9 100644
--- a/core/fxcodec/jbig2/JBig2_Context.cpp
+++ b/core/fxcodec/jbig2/JBig2_Context.cpp
@@ -649,13 +649,13 @@
   if (m_pStream->readInteger(&pTRD->SBNUMINSTANCES) != 0)
     return JBig2_Result::kFailure;
 
-  // Assume each instance takes at least 4 bits. That means for a stream of
-  // length N, there can be at most 2N instances. This is an extremely
+  // Assume each instance takes at least 0.25 bits when encoded. That means for
+  // a stream of length N bytes, there can be at most 32N instances. This is a
   // conservative estimate just to sanitize the |SBNUMINSTANCES| value.
   // Use FX_SAFE_INT32 to be safe, though it should never overflow because PDFs
   // have a maximum size of roughly 11 GB.
   FX_SAFE_INT32 nMaxStripInstances = m_pStream->getLength();
-  nMaxStripInstances *= 2;
+  nMaxStripInstances *= 32;
   if (pTRD->SBNUMINSTANCES > nMaxStripInstances.ValueOrDie())
     return JBig2_Result::kFailure;