ASan: Run with verify_asan_link_order=0.

[bernhardu]

Avoids error message:

  ASan runtime does not come first in initial library list; you should either link runtime to your application or manually preload it with LD_PRELOAD.

Affected tests:

  nested_detach_kill
  nested_detach_kill_stuck
  nested_detach_wait

(These tests may also need the ASan: Copy just used pages from ANONYMOUS|NORESERVE mappings. patch too.)

[Keno]

I feel like this conflates the issues of rr itself running under asan and rr tracing something running under asan (which itself may be rr). If it's just a matter of fixing the tests, I think this should be in __asan_default_options so it doesn't propagate to children.

[bernhardu]

Thanks, let me try how it works when I put it into the default options.

[bernhardu]

For the tests nested_detach_kill, nested_detach_kill_stuck and nested_detach_wait following diff would be sufficient.

diff --git a/src/main.cc b/src/main.cc
index 128fcbe3..f7ea5c34 100644
--- a/src/main.cc
+++ b/src/main.cc
@@ -231,6 +231,13 @@ size_t saved_argv0_space() {
 
 } // namespace rr
 
+#ifdef __SANITIZE_ADDRESS__
+#include <sanitizer/asan_interface.h>
+const char *__asan_default_options() {
+  return "verify_asan_link_order=0";
+}
+#endif
+
 using namespace rr;
 
 int main(int argc, char* argv[]) {

But the test nested_detach still fails. Following is the content of an example recording:

$ bin/rr record bin/rr record --nested=detach bin/simple
rr: Saving execution to trace directory `/home/bernhard/.local/share/rr/rr-23'.
rr: Saving execution to trace directory `/home/bernhard/.local/share/rr/simple-34'.
==1247576==ASan runtime does not come first in initial library list; you should either link runtime to your application or manually preload it with LD_PRELOAD.

$ bin/rr ps rr-23
PID     PPID    EXIT    CMD
1247550 --      detach  bin/rr record --nested=detach bin/simple

$ bin/rr replay --debugger-option=-q rr-23
(rr) b main
(rr) cont
(rr) print *environ@100
... "LD_PRELOAD=libasan.so.8:/home/bernhard/x86_64_asan/obj/bin/../lib/rr/librrpreload.so::::libdl.so.2", ...

$ bin/rr ps simple-34
PID     PPID    EXIT    CMD
1247576 --      1       bin/simple

$ bin/rr replay --debugger-option=-q simple-34
(rr) cont
(rr) print *environ@100
... "LD_PRELOAD=/home/bernhard/x86_64_asan/obj/bin/../lib/rr/librrpreload.so::::libdl.so.2:libasan.so.8:libdl.so.2", ...

So I guess the question is here if the simple binary should have the "libasan.so.8" in LD_PRELOAD?

To make it even more complicated the nested_detach 32-bit is also subject of another patch bf7a42acea which I used for above test.

[bernhardu]

I pushed a new version with no longer setting the environment, instead using __asan_default_options.

[bernhardu]

Changes:

• Rebased to current tip.
• Do not try to add libasan to LD_PRELOAD below ASan-rr.

[bernhardu]

Just a rebase to current tip.