DebugNamesDWARFIndex: fix handling of compressed sections

This fixes a silly bug where we were accidentally freeing the memory
used to store the decompressed .debug_names data. I had actually
considered this scenario when writing the class and put appropriate
precautions in place -- I just failed to wire it all up correctly.

This was only an issue for compressed sections because in case of
uncompressed ones we would access the data straight out of the mmapped
object file.

llvm-svn: 334717
This commit is contained in:
Pavel Labath 2018-06-14 14:41:30 +00:00
parent 1ee741bdb6
commit 234c681686
2 changed files with 15 additions and 0 deletions

View File

@ -0,0 +1,14 @@
// Test for a bug where we crashed while processing a compressed debug_names
// section (use after free).
// REQUIRES: lld, zlib
// RUN: clang -g -c -o %t.o --target=x86_64-pc-linux -mllvm -accel-tables=Dwarf %s
// RUN: ld.lld %t.o -o %t --compress-debug-sections=zlib
// RUN: lldb-test symbols --find=variable --name=foo %t | FileCheck %s
// CHECK: Found 1 variables:
int foo;
// ONE-DAG: name = "foo", type = {{.*}} (int), {{.*}} decl = debug-names-compressed.cpp:[[@LINE-1]]
extern "C" void _start() {}

View File

@ -53,6 +53,7 @@ private:
DWARFDataExtractor debug_str_data,
DWARFDebugInfo &debug_info)
: DWARFIndex(module), m_debug_info(debug_info),
m_debug_names_data(debug_names_data), m_debug_str_data(debug_str_data),
m_debug_names_up(std::move(debug_names_up)),
m_fallback(module, &debug_info, GetUnits(*m_debug_names_up)) {}