[ELF2] Symbol Versioning: part 1, VERSION() directive parsing

[davide]

Summary

So, this is a WIP. But it's enough that I can parse non-trivial trees, e.g.

VERSION {

 VERS_1.1 {
     global:
       x;
       *;
     local:
       zelda;
 };
VERS_1.2 {
  local:
    z;
};
VERS_1.3 {
  global:
    y;
} VERS_1.1;
VERS_1.4 {
  x;
} VERS_1.3;

}

There are a couple of issues with the current code:

1. It's not able to parse 'extern' : extern "C++" { ns::*; "f(int, double)"; };
2. It currently thinks ':' is a token (which I'm not completely sure it's right). So, when parsing namespaces (see example above) it will fail.

Still, this is expressive enough to parse almost every single version script FreeBSD base system ships with (modulo bugs). I hope to collect feedback, implement semantic action, and only after that commit.

[ruiu]

Does GNU linker accepts both 'global :' and 'global:' as a label? If not, we can handle ':' as part of the name rather than an independent token.

[davide]

In D13960#272578, @ruiu wrote:

Does GNU linker accepts both 'global :' and 'global:' as a label? If not, we can handle ':' as part of the name rather than an independent token.

I originally thought about this, but , alas, they do accept both.

% cat version3.script
VERSION {

VERS_1.1 {
    global :
     x;
};

}

% /usr/local/bin/ld.bfd -T version3.script
/usr/local/bin/ld.bfd: no input files

% /usr/local/bin/ld.gold -T version3.script
/usr/local/bin/ld.gold: fatal error: no input files

They also support things like:
global:

global;

for added fun. It looks like for GNU ld "global" is a keyword but they have special cases recognizing "global" (and "local" and "extern") in the yacc grammar.

FWIW, I noticed that GNU gold refuses the example above (but ld.bdf accepts it) and after a conversation with Ian Taylor he agreed this is a gold bug but also pointed out that in some cases there's no right/wrong answer for linker scripts/version scripts because there's no "serious" documentation.

[ruiu]

Oh yeah, that's very unfortunate, but that's true.

I took a look at gold's source code and found that their lexer behaves differently depending on the context -- if it's outside of version script, ':' is a token character, but inside version script, only '::' are handled as part of a token. I don't know if we really want to have this craziness.

I'd really like to always handle ':' as a token character, and let users to update their linker script if not compatible with that behavior. Davide, could you grep all linker scripts in FreeBSD which has a space between 'global' or 'local' and the following semicolon and count that number?

[emaste]

Davide, could you grep all linker scripts in FreeBSD which has a space between 'global' or 'local' and the following semicolon and count that number?

I'm not sure that's a very useful metric since the FreeBSD base system version scripts are bespoke and likely copied from each other. These aren't representative of 3rd-party scripts. We should be able to update these in FreeBSD pretty trivially.

[davide]

New patch after discussion with Rui/Ed.
Also, test added.

[ruiu]

You want to plug this into something so that the code added here will actually do a task it supposed to do.

[davide]

Superseded, close.

ELF/LinkerScript.cpp

 
 #include "Config.h"
 #include "Driver.h"
 #include "SymbolTable.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/StringSaver.h"
+#include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
 using namespace lld;
 using namespace lld::elf2;
 
 namespace {
 class LinkerScript {
 public:
private:
   void readEntry();
   void readExtern();
   void readGroup();
   void readInclude();
   void readOutput();
   void readOutputArch();
   void readOutputFormat();
   void readSearchDir();
+  void readVersion();
 
   StringSaver Saver;
   std::vector<StringRef> Tokens;
   size_t Pos = 0;
 };
 }
 
 void LinkerScript::run() {
while (!atEOF()) {
     } else if (Tok == "OUTPUT") {
       readOutput();
     } else if (Tok == "OUTPUT_ARCH") {
       readOutputArch();
     } else if (Tok == "OUTPUT_FORMAT") {
       readOutputFormat();
     } else if (Tok == "SEARCH_DIR") {
       readSearchDir();
+    } else if (Tok == "VERSION") {
+      readVersion();
     } else {
       error("unknown directive: " + Tok);
     }
   }
 }
 
 // Split S into linker script tokens.
 std::vector<StringRef> LinkerScript::tokenize(StringRef S) {

 }
 
 void LinkerScript::readSearchDir() {
   expect("(");
   Config->SearchPaths.push_back(next());
   expect(")");
 }
 
+void LinkerScript::readVersion() {
+  StringRef Tok, TokNext;
+  expect("{");
+  for (;;) {
+    Tok = next();
+    if (Tok == "}")
+      break;
+    expect("{");
+    Tok = next();
+
+    // Neither global or local is specified.
+    // Just parse the list of the symbols.
+    if (Tok != "global:" && Tok != "local:") {
+      expect(";");
+      while ((Tok = next()) != "}")
+        TokNext = next();
+      TokNext = next();
+    } else {
+      TokNext = next();
+
+      // Parse global symbols.
+      if (Tok == "global:")
+        do {
+          Tok = next(); // Symbol name
+          TokNext = next();
+        } while (Tok != "}");
+
+      // Parse local symbols.
+      if (Tok == "local:")
+        do {
+          Tok = next(); // Symbol name
+          TokNext = next();
+        } while (Tok != "}");
+    }
+    if (TokNext != ";")
+      // Dependency on another version.
+      expect(";");
+  }
+}
+
 // Entry point. The other functions or classes are private to this file.
 void lld::elf2::readLinkerScript(BumpPtrAllocator *A, MemoryBufferRef MB) {
   LinkerScript(A, MB.getBuffer()).run();
 }

test/elf2/Inputs/version.script

+VERSION {
+   VERS_1.1 {
+       global:
+         x;
+         *;
+       local:
+         q;
+   };
+  VERS_1.2 {
+    local:
+      z;
+  };
+  VERS_1.3 {
+    global:
+      y;
+  } VERS_1.1;
+  VERS_1.4 {
+    x;
+  } VERS_1.3;
+}

test/elf2/linkerscript-version.s

+// RUN: llvm-mc %s -filetype=obj -triple=x86_64-unknown-freebsd -o %s.o
+// RUN: ld.lld2 -shared -T %p/Inputs/version.script %s.o -o %t.so
+// RUN: llvm-objdump %t > /dev/null