Exit rather than abort if asked to run a non-PIE executable.

Each release we're asked to investigate tombstones from code that hasn't
been allowed to run on Android since L. This is just wasting our time,
and clearly the "obviousness" of aborting rather than exiting hasn't ensured
that all app developers rebuild their old binaries. In some cases it seems
like they run them "just in case" and don't care if they fail.

Bug: http://b/34112178
Test: ran libsupervisor.so from com.ss.android.article.news
Change-Id: I8a3f196c4755601a3888281566fbb7b817f01dca

1 files changed, 14 insertions, 2 deletions

diff --git a/linker/linker_main.cpp b/linker/linker_main.cpp
index 13edfe1f7..a5abdfffa 100644
--- a/linker/linker_main.cpp
+++ b/linker/linker_main.cpp
@@ -306,9 +306,21 @@ static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(
   si->dynamic = nullptr;
 
   ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
+
+  // We haven't supported non-PIE since Lollipop for security reasons.
   if (elf_hdr->e_type != ET_DYN) {
-    __libc_fatal("\"%s\": error: only position independent executables (PIE) are supported.",
-                 g_argv[0]);
+    // We don't use __libc_fatal here because we don't want a tombstone: it's
+    // been several years now but we still find ourselves on app compatibility
+    // investigations because some app's trying to launch an executable that
+    // hasn't worked in at least three years, and we've "helpfully" dropped a
+    // tombstone for them. The tombstone never provided any detail relevant to
+    // fixing the problem anyway, and the utility of drawing extra attention
+    // to the problem is non-existent at this late date.
+    __libc_format_fd(STDOUT_FILENO,
+                     "\"%s\": error: Android 5.0 and later only support "
+                     "position-independent executables (-fPIE).",
+                     g_argv[0]);
+    exit(0);
   }
 
   // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).