Complete the ld.so boot cleanup: move most libc initialization from

_csu_finish() to _libc_preinit(), which is an .init_array function
in shared libc (and mark it INITFIRST) or a .preinit_array function
in static libc, grabbing the _dl_cb callback there from ld.so.  Then
in _csu_finish(), invoke the dl_clean_boot() callback to free ld.so's
startup bits before main() gets control.

Other cleanups this permits:
 - move various startup structures into .data.rel.ro
 - the dl* stubs in libc can be strong and call the callbacks provided
   via _dl_cb
 - no longer need to conditionalize dlctl() calls on presence of _DYNAMIC

original concept and full diff ok kettenis@
ok deraadt@

2 files changed, 86 insertions, 40 deletions

diff --git a/lib/libc/dlfcn/dlfcn_stubs.c b/lib/libc/dlfcn/dlfcn_stubs.c
index e04a2aa0706..78d728f66cb 100644
--- a/lib/libc/dlfcn/dlfcn_stubs.c
+++ b/lib/libc/dlfcn/dlfcn_stubs.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: dlfcn_stubs.c,v 1.14 2016/09/06 18:49:34 guenther Exp $	*/
+/*	$OpenBSD: dlfcn_stubs.c,v 1.15 2019/06/02 01:03:01 guenther Exp $	*/
 
 /*
  * Copyright (c) 1998 Per Fogelstrom, Opsycon AB
@@ -28,34 +28,19 @@
 
 #include <sys/types.h>
 #include <stddef.h>
+#include <stdio.h>
 #include <link.h>
+#include <dlfcn.h>
+#include <tib.h>
 
 #include "init.h"
 
-/*
- *	All functions here are just stubs that will be overridden
- *	by the real functions in ld.so when dynamic loading is
- *	performed at exec. The symbols here are provided as a link
- *	helper so we can link a program using the dl functions
- *	without getting any unresolved references.
- */
-
-void	*dlopen(const char *libname, int how) __attribute__((weak));
-int	 dlclose(void *handle) __attribute__((weak));
-void	*dlsym(void *handle, const char *name) __attribute__((weak));
-int	 dlctl(void *handle, int command, void *data) __attribute__((weak));
-char 	*dlerror(void) __attribute__((weak));
-
-struct dl_info;
-int	dladdr(const void *addr, struct dl_info *info) __attribute__((weak));
-
-int	 dl_iterate_phdr(int (*callback)(struct dl_phdr_info *, size_t, void *),	    void *date) __attribute__((weak));
-
-#include <stdio.h>
 
 void *
 dlopen(const char *libname, int how)
 {
+	if (_dl_cb != NULL && _dl_cb->dlopen != NULL)
+		return _dl_cb->dlopen(libname, how);
 	printf("Wrong dl symbols!\n");
 	return NULL;
 }
@@ -63,6 +48,8 @@ dlopen(const char *libname, int how)
 int
 dlclose(void *handle)
 {
+	if (_dl_cb != NULL && _dl_cb->dlclose != NULL)
+		return _dl_cb->dlclose(handle);
 	printf("Wrong dl symbols!\n");
 	return 0;
 }
@@ -70,6 +57,8 @@ dlclose(void *handle)
 void *
 dlsym(void *handle, const char *name)
 {
+	if (_dl_cb != NULL && _dl_cb->dlsym != NULL)
+		return _dl_cb->dlsym(handle, name);
 	printf("Wrong dl symbols!\n");
 	return NULL;
 }
@@ -77,12 +66,17 @@ dlsym(void *handle, const char *name)
 int
 dlctl(void *handle, int command, void *data)
 {
+	if (_dl_cb != NULL && _dl_cb->dlctl != NULL)
+		return _dl_cb->dlctl(handle, command, data);
 	return -1;
 }
+DEF_WEAK(dlctl);
 
 char *
 dlerror(void)
 {
+	if (_dl_cb != NULL && _dl_cb->dlerror != NULL)
+		return _dl_cb->dlerror();
 	return "Wrong dl symbols!\n";
 }
 
@@ -90,6 +84,8 @@ int
 dl_iterate_phdr(int (*callback)(struct dl_phdr_info *, size_t, void *),
 	void *data)
 {
+	if (_dl_cb != NULL && _dl_cb->dl_iterate_phdr != NULL)
+		return _dl_cb->dl_iterate_phdr(callback, data);
 #ifndef PIC
 	if (_static_phdr_info.dlpi_phdr != NULL)
 		return callback(&_static_phdr_info, sizeof(_static_phdr_info),
@@ -97,14 +93,18 @@ dl_iterate_phdr(int (*callback)(struct dl_phdr_info *, size_t, void *),
 #endif /* !PIC */
 	return -1;
 }
+DEF_WEAK(dl_iterate_phdr);
 
 int
 dladdr(const void *addr, struct dl_info *info)
 {
+	if (_dl_cb != NULL && _dl_cb->dladdr != NULL)
+		return _dl_cb->dladdr(addr, info);
 	printf("Wrong dl symbols!\n");
 	return -1;
 }
 
+#if 0
 /* Thread Local Storage argument structure */
 typedef struct {
 	unsigned long int ti_module;
@@ -131,3 +131,4 @@ ___tls_get_addr(tls_index *ti)
 }
 #endif /* __i386 */
 #endif /* arch with TLS support enabled */
+#endif
diff --git a/lib/libc/dlfcn/init.c b/lib/libc/dlfcn/init.c
index 3c387a60d6b..270f54aada5 100644
--- a/lib/libc/dlfcn/init.c
+++ b/lib/libc/dlfcn/init.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: init.c,v 1.6 2017/12/01 23:30:05 guenther Exp $ */
+/*	$OpenBSD: init.c,v 1.7 2019/06/02 01:03:01 guenther Exp $ */
 /*
  * Copyright (c) 2014,2015 Philip Guenther <guenther@openbsd.org>
  *
@@ -57,7 +57,7 @@ char	*__progname __attribute__((weak)) = NULL;
 
 
 #ifndef PIC
-struct dl_phdr_info	_static_phdr_info = { .dlpi_name = "a.out" };
+struct dl_phdr_info	_static_phdr_info __relro = { .dlpi_name = "a.out" };
 
 static inline void early_static_init(char **_argv, char **_envp);
 static inline void setup_static_tib(Elf_Phdr *_phdr, int _phnum);
@@ -66,13 +66,12 @@ static inline void setup_static_tib(Elf_Phdr *_phdr, int _phnum);
 extern Elf_Ehdr __executable_start[] __attribute__((weak));
 #endif /* PIC */
 
-/*
- * extract useful bits from the auxiliary vector and either
- * a) register ld.so's cleanup in dynamic links, or
- * b) init __progname, environ, and the TIB in static links.
- */
-char ***
-_csu_finish(char **argv, char **envp, void (*cleanup)(void))
+/* provide definitions for these */
+const dl_cb *_dl_cb __relro = NULL;
+
+void _libc_preinit(int, char **, char **, dl_cb_cb *) __dso_hidden;
+void
+_libc_preinit(int argc, char **argv, char **envp, dl_cb_cb *cb)
 {
 	AuxInfo	*aux;
 #ifndef PIC
@@ -80,10 +79,13 @@ _csu_finish(char **argv, char **envp, void (*cleanup)(void))
 	int phnum = 0;
 
 	/* static libc in a static link? */
-	if (cleanup == NULL)
+	if (cb == NULL)
 		early_static_init(argv, envp);
 #endif /* !PIC */
 
+	if (cb != NULL)
+		_dl_cb = cb(DL_CB_CUR);
+
 	/* Extract useful bits from the auxiliary vector */
 	while (*envp++ != NULL)
 		;
@@ -98,30 +100,73 @@ _csu_finish(char **argv, char **envp, void (*cleanup)(void))
 			break;
 		case AUX_phdr:
 			phdr = (void *)aux->au_v;
-			_static_phdr_info.dlpi_phdr = phdr;
 			break;
 		case AUX_phnum:
 			phnum = aux->au_v;
-			_static_phdr_info.dlpi_phnum = phnum;
 			break;
 #endif /* !PIC */
 		}
 	}
 
 #ifndef PIC
-	if (cleanup == NULL && phdr == NULL && __executable_start != NULL) {
+	if (cb == NULL && phdr == NULL && __executable_start != NULL) {
 		/*
 		 * Static non-PIE processes don't get an AUX vector,
 		 * so find the phdrs through the ELF header
 		 */
+		_static_phdr_info.dlpi_addr = (Elf_Addr)__executable_start;
 		phdr = (void *)((char *)__executable_start +
 		    __executable_start->e_phoff);
 		phnum = __executable_start->e_phnum;
 	}
+	_static_phdr_info.dlpi_phdr = phdr;
+	_static_phdr_info.dlpi_phnum = phnum;
+
 	/* static libc in a static link? */
-	if (cleanup == NULL)
+	if (cb == NULL)
 		setup_static_tib(phdr, phnum);
 #endif /* !PIC */
+}
+
+/* ARM just had to be different... */
+#ifndef __arm__
+# define TYPE	"@"
+#else
+# define TYPE	"%"
+#endif
+
+#ifdef __LP64__
+# define VALUE_ALIGN		".balign 8"
+# define VALUE_DIRECTIVE	".quad"
+#else
+# define VALUE_ALIGN		".balign 4"
+# ifdef __hppa__
+   /* hppa just had to be different: func pointers prefix with 'P%' */
+#  define VALUE_DIRECTIVE	".int P%"
+# else
+#  define VALUE_DIRECTIVE	".int"
+# endif
+#endif
+
+#define ADD_TO_ARRAY(func, which) \
+	__asm(	" .section ."#which",\"a\","TYPE#which"\n " \
+	VALUE_ALIGN"\n "VALUE_DIRECTIVE" "#func"\n .previous")
+
+#ifdef PIC
+ADD_TO_ARRAY(_libc_preinit, init_array);
+#else
+ADD_TO_ARRAY(_libc_preinit, preinit_array);
+#endif
+
+/*
+ * In dynamic links, invoke ld.so's dl_clean_boot() callback, if any,
+ * and register its cleanup.
+ */
+char ***
+_csu_finish(char **argv, char **envp, void (*cleanup)(void))
+{
+	if (_dl_cb != NULL && _dl_cb->dl_clean_boot != NULL)
+		_dl_cb->dl_clean_boot();
 
 	if (cleanup != NULL)
 		atexit(cleanup);
@@ -163,12 +208,12 @@ early_static_init(char **argv, char **envp)
 #define ELF_ROUND(x,malign)	(((x) + (malign)-1) & ~((malign)-1))
 
 /* for static binaries, the location and size of the TLS image */
-static void		*static_tls;
-static size_t		static_tls_fsize;
+static void		*static_tls __relro;
+static size_t		static_tls_fsize __relro;
 
-size_t			_static_tls_size = 0;
-int			_static_tls_align;
-int			_static_tls_align_offset;
+size_t			_static_tls_size __relro = 0;
+int			_static_tls_align __relro;
+int			_static_tls_align_offset __relro;
 
 static inline void
 setup_static_tib(Elf_Phdr *phdr, int phnum)