Fix several problems with our ELF filters implementation.

Do not relocate twice an object which happens to be needed by loaded
binary (or dso) and some filtee opened due to symbol resolution when
relocating need objects.  Record the state of the relocation
processing in Obj_Entry and short-circuit relocate_objects() if
current object already processed.

Do not call constructors for filtees loaded during the early
relocation processing before image is initialized enough to run
user-provided code.  Filtees are loaded using dlopen_object(), which
normally performs relocation and initialization.  If filtee is
lazy-loaded during the relocation of dso needed by the main object,
dlopen_object() runs too earlier, when most runtime services are not
yet ready.

Postpone the constructors call to the time when main binary and
depended libraries constructors are run, passing the new flag
RTLD_LO_EARLY to dlopen_object().  Symbol lookups callers inform
symlook_* functions about early stage of initialization with
SYMLOOK_EARLY.  Pass flags through all functions participating in
object relocation.

Use the opportunity and fix flags argument to find_symdef() in
arch-specific reloc.c to use proper name SYMLOOK_IN_PLT instead of
true, which happen to have the same numeric value.

Reported and tested by:	theraven
Reviewed by:	kan
MFC after:	2 weeks

10 files changed, 179 insertions, 118 deletions

diff --git a/libexec/rtld-elf/amd64/reloc.c b/libexec/rtld-elf/amd64/reloc.c
index 5f40847..9c755bc 100644
--- a/libexec/rtld-elf/amd64/reloc.c
+++ b/libexec/rtld-elf/amd64/reloc.c
@@ -82,6 +82,7 @@ do_copy_relocations(Obj_Entry *dstobj)
 	    size = dstsym->st_size;
 	    symlook_init(&req, name);
 	    req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
+	    req.flags = SYMLOOK_EARLY;
 
 	    for (srcobj = dstobj->next;  srcobj != NULL;  srcobj = srcobj->next) {
 		res = symlook_obj(&req, srcobj);
@@ -118,7 +119,8 @@ init_pltgot(Obj_Entry *obj)
 
 /* Process the non-PLT relocations. */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rela *relalim;
 	const Elf_Rela *rela;
@@ -151,7 +153,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -170,7 +172,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -200,7 +202,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -214,7 +216,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -245,7 +247,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -277,7 +279,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -291,7 +293,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -305,7 +307,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -330,7 +332,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 done:
 	if (cache != NULL)
 	    free(cache);
-	return(r);
+	return (r);
 }
 
 /* Process the PLT relocations. */
@@ -366,7 +368,7 @@ reloc_plt(Obj_Entry *obj)
 
 /* Relocate the jump slots in an object. */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
     const Elf_Rela *relalim;
     const Elf_Rela *rela;
@@ -382,8 +384,8 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 	switch (ELF_R_TYPE(rela->r_info)) {
 	case R_X86_64_JMP_SLOT:
 	  where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
-	  def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, true, NULL,
-	      lockstate);
+	  def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
+		SYMLOOK_IN_PLT | flags, NULL, lockstate);
 	  if (def == NULL)
 	      return (-1);
 	  if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC) {
@@ -438,7 +440,7 @@ reloc_iresolve(Obj_Entry *obj, RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
     const Elf_Rela *relalim;
     const Elf_Rela *rela;
@@ -454,8 +456,8 @@ reloc_gnu_ifunc(Obj_Entry *obj, RtldLockState *lockstate)
 	switch (ELF_R_TYPE(rela->r_info)) {
 	case R_X86_64_JMP_SLOT:
 	  where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
-	  def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, true, NULL,
-	      lockstate);
+	  def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
+		SYMLOOK_IN_PLT | flags, NULL, lockstate);
 	  if (def == NULL)
 	      return (-1);
 	  if (ELF_ST_TYPE(def->st_info) != STT_GNU_IFUNC)
diff --git a/libexec/rtld-elf/arm/reloc.c b/libexec/rtld-elf/arm/reloc.c
index 11b962a..aaad2da 100644
--- a/libexec/rtld-elf/arm/reloc.c
+++ b/libexec/rtld-elf/arm/reloc.c
@@ -54,6 +54,8 @@ do_copy_relocations(Obj_Entry *dstobj)
 			symlook_init(&req, name);
 			req.ventry = fetch_ventry(dstobj,
 			    ELF_R_SYM(rel->r_info));
+			req.flags = SYMLOOK_EARLY;
+
 			for (srcobj = dstobj->next;  srcobj != NULL; 
 			     srcobj = srcobj->next) {
 				res = symlook_obj(&req, srcobj);
@@ -135,7 +137,7 @@ store_ptr(void *where, Elf_Addr val)
 
 static int
 reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
-    RtldLockState *lockstate)
+    int flags, RtldLockState *lockstate)
 {
 	Elf_Addr        *where;
 	const Elf_Sym   *def;
@@ -161,7 +163,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
 		if (addend & 0x00800000)
 			addend |= 0xff000000;
 		
-		def = find_symdef(symnum, obj, &defobj, false, cache,
+		def = find_symdef(symnum, obj, &defobj, flags, cache,
 		    lockstate);
 		if (def == NULL)
 				return -1;
@@ -188,7 +190,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
 
 		case R_ARM_ABS32:	/* word32 B + S + A */
 		case R_ARM_GLOB_DAT:	/* word32 B + S */
-			def = find_symdef(symnum, obj, &defobj, false, cache,
+			def = find_symdef(symnum, obj, &defobj, flags, cache,
 			    lockstate);
 			if (def == NULL)
 				return -1;
@@ -237,7 +239,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
 			break;
 
 		case R_ARM_TLS_DTPOFF32:
-			def = find_symdef(symnum, obj, &defobj, false, cache,
+			def = find_symdef(symnum, obj, &defobj, flags, cache,
 			    lockstate);
 			if (def == NULL)
 				return -1;
@@ -254,7 +256,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
 
 			break;
 		case R_ARM_TLS_DTPMOD32:
-			def = find_symdef(symnum, obj, &defobj, false, cache,
+			def = find_symdef(symnum, obj, &defobj, flags, cache,
 			    lockstate);
 			if (def == NULL)
 				return -1;
@@ -272,7 +274,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
 			break;
 
 		case R_ARM_TLS_TPOFF32:
-			def = find_symdef(symnum, obj, &defobj, false, cache,
+			def = find_symdef(symnum, obj, &defobj, flags, cache,
 			    lockstate);
 			if (def == NULL)
 				return -1;
@@ -311,7 +313,8 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rel *rel, SymCache *cache,
  *  * Process non-PLT relocations
  *   */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rel *rellim;
 	const Elf_Rel *rel;
@@ -330,7 +333,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 
 	rellim = (const Elf_Rel *)((caddr_t)obj->rel + obj->relsize);
 	for (rel = obj->rel; rel < rellim; rel++) {
-		if (reloc_nonplt_object(obj, rel, cache, lockstate) < 0)
+		if (reloc_nonplt_object(obj, rel, cache, flags, lockstate) < 0)
 			goto done;
 	}
 	r = 0;
@@ -367,7 +370,7 @@ reloc_plt(Obj_Entry *obj)
  *  * LD_BIND_NOW was set - force relocation for all jump slots
  *   */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	const Obj_Entry *defobj;
 	const Elf_Rel *rellim;
@@ -381,7 +384,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 		assert(ELF_R_TYPE(rel->r_info) == R_ARM_JUMP_SLOT);
 		where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
 		def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		    true, NULL, lockstate);
+		    SYMLOOK_IN_PLT | flags, NULL, lockstate);
 		if (def == NULL) {
 			dbg("reloc_jmpslots: sym not found");
 			return (-1);
@@ -406,7 +409,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */
diff --git a/libexec/rtld-elf/i386/reloc.c b/libexec/rtld-elf/i386/reloc.c
index a2c8c98..5f4e247 100644
--- a/libexec/rtld-elf/i386/reloc.c
+++ b/libexec/rtld-elf/i386/reloc.c
@@ -83,6 +83,7 @@ do_copy_relocations(Obj_Entry *dstobj)
 	    size = dstsym->st_size;
 	    symlook_init(&req, name);
 	    req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rel->r_info));
+	    req.flags = SYMLOOK_EARLY;
 
 	    for (srcobj = dstobj->next;  srcobj != NULL;  srcobj = srcobj->next) {
 		res = symlook_obj(&req, srcobj);
@@ -119,7 +120,8 @@ init_pltgot(Obj_Entry *obj)
 
 /* Process the non-PLT relocations. */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rel *rellim;
 	const Elf_Rel *rel;
@@ -151,7 +153,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -170,7 +172,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -200,7 +202,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -220,7 +222,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    Elf_Addr add;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -253,7 +255,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -267,7 +269,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		    const Obj_Entry *defobj;
 
 		    def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
-		      false, cache, lockstate);
+		      flags, cache, lockstate);
 		    if (def == NULL)
 			goto done;
 
@@ -286,7 +288,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 done:
 	if (cache != NULL)
 	    free(cache);
-	return(r);
+	return (r);
 }
 
 /* Process the PLT relocations. */
@@ -322,7 +324,7 @@ reloc_plt(Obj_Entry *obj)
 
 /* Relocate the jump slots in an object. */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
     const Elf_Rel *rellim;
     const Elf_Rel *rel;
@@ -338,8 +340,8 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 	switch (ELF_R_TYPE(rel->r_info)) {
 	case R_386_JMP_SLOT:
 	  where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
-	  def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, true, NULL,
-	      lockstate);
+	  def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
+		SYMLOOK_IN_PLT | flags, NULL, lockstate);
 	  if (def == NULL)
 	      return (-1);
 	  if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC) {
@@ -390,7 +392,7 @@ reloc_iresolve(Obj_Entry *obj, RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
     const Elf_Rel *rellim;
     const Elf_Rel *rel;
@@ -406,8 +408,8 @@ reloc_gnu_ifunc(Obj_Entry *obj, RtldLockState *lockstate)
 	switch (ELF_R_TYPE(rel->r_info)) {
 	case R_386_JMP_SLOT:
 	  where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
-	  def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, true, NULL,
-	      lockstate);
+	  def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj,
+		SYMLOOK_IN_PLT | flags, NULL, lockstate);
 	  if (def == NULL)
 	      return (-1);
 	  if (ELF_ST_TYPE(def->st_info) != STT_GNU_IFUNC)
diff --git a/libexec/rtld-elf/ia64/reloc.c b/libexec/rtld-elf/ia64/reloc.c
index d5b45b7..1afcecf 100644
--- a/libexec/rtld-elf/ia64/reloc.c
+++ b/libexec/rtld-elf/ia64/reloc.c
@@ -151,7 +151,7 @@ free_fptrs(Obj_Entry *obj, bool mapped)
 /* Relocate a non-PLT object with addend. */
 static int
 reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
-    SymCache *cache, RtldLockState *lockstate)
+    SymCache *cache, int flags, RtldLockState *lockstate)
 {
 	struct fptr **fptrs;
 	Elf_Addr *where = (Elf_Addr *) (obj->relocbase + rela->r_offset);
@@ -172,7 +172,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		Elf_Addr target;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return -1;
 
@@ -195,7 +195,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		int sym_index;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    true, cache, lockstate);
+		    SYMLOOK_IN_PLT | flags, cache, lockstate);
 		if (def == NULL) {
 			/*
 			 * XXX r_debug_state is problematic and find_symdef()
@@ -254,7 +254,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		Elf_Addr target, gp;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return -1;
 
@@ -277,7 +277,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		const Obj_Entry *defobj;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return -1;
 
@@ -290,7 +290,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		const Obj_Entry *defobj;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return -1;
 
@@ -303,7 +303,7 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		const Obj_Entry *defobj;
 
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return -1;
 
@@ -342,7 +342,8 @@ reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 
 /* Process the non-PLT relocations. */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rel *rellim;
 	const Elf_Rel *rel;
@@ -368,7 +369,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 		locrela.r_info = rel->r_info;
 		locrela.r_offset = rel->r_offset;
 		locrela.r_addend = 0;
-		if (reloc_non_plt_obj(obj_rtld, obj, &locrela, cache,
+		if (reloc_non_plt_obj(obj_rtld, obj, &locrela, cache, flags,
 		    lockstate))
 			goto done;
 	}
@@ -376,7 +377,8 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 	/* Perform relocations with addend if there are any: */
 	relalim = (const Elf_Rela *) ((caddr_t) obj->rela + obj->relasize);
 	for (rela = obj->rela;  obj->rela != NULL && rela < relalim;  rela++) {
-		if (reloc_non_plt_obj(obj_rtld, obj, rela, cache, lockstate))
+		if (reloc_non_plt_obj(obj_rtld, obj, rela, cache, flags,
+		    lockstate))
 			goto done;
 	}
 
@@ -444,7 +446,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */
@@ -453,7 +456,7 @@ reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 
 /* Relocate the jump slots in an object. */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	if (obj->jmpslots_done)
 		return 0;
@@ -472,7 +475,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 			assert(ELF_R_TYPE(rel->r_info) == R_IA_64_IPLTLSB);
 			where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
 			def = find_symdef(ELF_R_SYM(rel->r_info), obj,
-			    &defobj, true, NULL, lockstate);
+			    &defobj, SYMLOOK_IN_PLT | flags, NULL, lockstate);
 			if (def == NULL)
 				return -1;
 			reloc_jmpslot(where,
@@ -493,7 +496,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 
 			where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 			def = find_symdef(ELF_R_SYM(rela->r_info), obj,
-			    &defobj, true, NULL, lockstate);
+			    &defobj, SYMLOOK_IN_PLT | flags, NULL, lockstate);
 			if (def == NULL)
 				return -1;
 			reloc_jmpslot(where,
diff --git a/libexec/rtld-elf/mips/reloc.c b/libexec/rtld-elf/mips/reloc.c
index fd108f9..24e56ce 100644
--- a/libexec/rtld-elf/mips/reloc.c
+++ b/libexec/rtld-elf/mips/reloc.c
@@ -258,7 +258,8 @@ _mips_rtld_bind(Obj_Entry *obj, Elf_Size reloff)
 }
 
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rel *rel;
 	const Elf_Rel *rellim;
@@ -317,7 +318,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 			 * to 0 if there are non-PLT references, but older
 			 * versions of GNU ld do not do this.
 			 */
-			def = find_symdef(i, obj, &defobj, false, NULL,
+			def = find_symdef(i, obj, &defobj, flags, NULL,
 			    lockstate);
 			if (def == NULL)
 				return -1;
@@ -359,7 +360,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 			}
 		} else {
 			/* TODO: add cache here */
-			def = find_symdef(i, obj, &defobj, false, NULL,
+			def = find_symdef(i, obj, &defobj, flags, NULL,
 			    lockstate);
 			if (def == NULL) {
 				dbg("Warning4, can't find symbole %d", i);
@@ -458,7 +459,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 			Elf_Addr old = load_ptr(where, rlen);
 			Elf_Addr val = old;
 
-        		def = find_symdef(r_symndx, obj, &defobj, false, NULL,
+        		def = find_symdef(r_symndx, obj, &defobj, flags, NULL,
 	    			lockstate);
 			if (def == NULL)
 				return -1;
@@ -482,7 +483,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 			Elf_Addr old = load_ptr(where, rlen);
 			Elf_Addr val = old;
 
-        		def = find_symdef(r_symndx, obj, &defobj, false, NULL,
+        		def = find_symdef(r_symndx, obj, &defobj, flags, NULL,
 	    			lockstate);
 			if (def == NULL)
 				return -1;
@@ -509,7 +510,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 			Elf_Addr old = load_ptr(where, rlen);
 			Elf_Addr val = old;
 
-        		def = find_symdef(r_symndx, obj, &defobj, false, NULL,
+        		def = find_symdef(r_symndx, obj, &defobj, flags, NULL,
 	    			lockstate);
 
 			if (def == NULL)
@@ -577,7 +578,7 @@ reloc_plt(Obj_Entry *obj)
  * LD_BIND_NOW was set - force relocation for all jump slots
  */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	/* Do nothing */
 	obj->jmpslots_done = true;
@@ -594,7 +595,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */
diff --git a/libexec/rtld-elf/powerpc/reloc.c b/libexec/rtld-elf/powerpc/reloc.c
index d0637f4..979e43e 100644
--- a/libexec/rtld-elf/powerpc/reloc.c
+++ b/libexec/rtld-elf/powerpc/reloc.c
@@ -92,6 +92,7 @@ do_copy_relocations(Obj_Entry *dstobj)
 		size = dstsym->st_size;
 		symlook_init(&req, name);
 		req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
+		req.flags = SYMLOOK_EARLY;
 
 		for (srcobj = dstobj->next;  srcobj != NULL;
 		     srcobj = srcobj->next) {
@@ -159,7 +160,7 @@ reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
  */
 static int
 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
-    SymCache *cache, RtldLockState *lockstate)
+    SymCache *cache, int flags, RtldLockState *lockstate)
 {
 	Elf_Addr        *where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 	const Elf_Sym   *def;
@@ -174,7 +175,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
         case R_PPC_ADDR32:    /* word32 S + A */
         case R_PPC_GLOB_DAT:  /* word32 S + A */
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL) {
 			return (-1);
 		}
@@ -221,7 +222,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 
 	case R_PPC_DTPMOD32:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -232,7 +233,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 
 	case R_PPC_TPREL32:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -261,7 +262,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		
 	case R_PPC_DTPREL32:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -285,7 +286,8 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
  * Process non-PLT relocations
  */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rela *relalim;
 	const Elf_Rela *rela;
@@ -309,8 +311,8 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 	 */
 	relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
 	for (rela = obj->rela; rela < relalim; rela++) {
-		if (reloc_nonplt_object(obj_rtld, obj, rela, cache, lockstate)
-		    < 0)
+		if (reloc_nonplt_object(obj_rtld, obj, rela, cache, flags,
+		    lockstate) < 0)
 			goto done;
 	}
 	r = 0;
@@ -416,7 +418,7 @@ reloc_plt(Obj_Entry *obj)
  * LD_BIND_NOW was set - force relocation for all jump slots
  */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	const Obj_Entry *defobj;
 	const Elf_Rela *relalim;
@@ -430,7 +432,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 		assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
 		where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    true, NULL, lockstate);
+		    SYMLOOK_IN_PLT | flags, NULL, lockstate);
 		if (def == NULL) {
 			dbg("reloc_jmpslots: sym not found");
 			return (-1);
@@ -525,7 +527,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */
diff --git a/libexec/rtld-elf/powerpc64/reloc.c b/libexec/rtld-elf/powerpc64/reloc.c
index 47368a8..46e2c62 100644
--- a/libexec/rtld-elf/powerpc64/reloc.c
+++ b/libexec/rtld-elf/powerpc64/reloc.c
@@ -86,6 +86,7 @@ do_copy_relocations(Obj_Entry *dstobj)
 		size = dstsym->st_size;
 		symlook_init(&req, name);
 		req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
+		req.flags = SYMLOOK_EARLY;
 
 		for (srcobj = dstobj->next;  srcobj != NULL;
 		     srcobj = srcobj->next) {
@@ -153,7 +154,7 @@ reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
  */
 static int
 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
-    SymCache *cache, RtldLockState *lockstate)
+    SymCache *cache, int flags, RtldLockState *lockstate)
 {
 	Elf_Addr        *where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 	const Elf_Sym   *def;
@@ -169,7 +170,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
         case R_PPC64_ADDR64:
         case R_PPC_GLOB_DAT:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL) {
 			return (-1);
 		}
@@ -216,7 +217,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 
 	case R_PPC64_DTPMOD64:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -227,7 +228,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 
 	case R_PPC64_TPREL64:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -256,7 +257,7 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
 		
 	case R_PPC64_DTPREL64:
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 
 		if (def == NULL)
 			return (-1);
@@ -280,7 +281,8 @@ reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
  * Process non-PLT relocations
  */
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rela *relalim;
 	const Elf_Rela *rela;
@@ -307,8 +309,8 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 	 */
 	relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
 	for (rela = obj->rela; rela < relalim; rela++) {
-		if (reloc_nonplt_object(obj_rtld, obj, rela, cache, lockstate)
-		    < 0)
+		if (reloc_nonplt_object(obj_rtld, obj, rela, cache, flags,
+		    lockstate) < 0)
 			goto done;
 	}
 	r = 0;
@@ -379,7 +381,7 @@ reloc_plt(Obj_Entry *obj)
  * LD_BIND_NOW was set - force relocation for all jump slots
  */
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	const Obj_Entry *defobj;
 	const Elf_Rela *relalim;
@@ -393,7 +395,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 		assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
 		where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    true, NULL, lockstate);
+		    SYMLOOK_IN_PLT | flags, NULL, lockstate);
 		if (def == NULL) {
 			dbg("reloc_jmpslots: sym not found");
 			return (-1);
@@ -468,7 +470,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */
diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
index 3432232..f2ad554 100644
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -116,9 +116,10 @@ static void objlist_push_head(Objlist *, Obj_Entry *);
 static void objlist_push_tail(Objlist *, Obj_Entry *);
 static void objlist_remove(Objlist *, Obj_Entry *);
 static void *path_enumerate(const char *, path_enum_proc, void *);
-static int relocate_objects(Obj_Entry *, bool, Obj_Entry *, RtldLockState *);
+static int relocate_objects(Obj_Entry *, bool, Obj_Entry *, int,
+    RtldLockState *);
 static int resolve_objects_ifunc(Obj_Entry *first, bool bind_now,
-    RtldLockState *lockstate);
+    int flags, RtldLockState *lockstate);
 static int rtld_dirname(const char *, char *);
 static int rtld_dirname_abs(const char *, char *);
 static void *rtld_dlopen(const char *name, int fd, int mode);
@@ -545,7 +546,8 @@ _rtld(Elf_Addr *sp, func_ptr_type *exit_proc, Obj_Entry **objp)
     }
 
     if (relocate_objects(obj_main,
-      ld_bind_now != NULL && *ld_bind_now != '\0', &obj_rtld, NULL) == -1)
+      ld_bind_now != NULL && *ld_bind_now != '\0',
+      &obj_rtld, SYMLOOK_EARLY, NULL) == -1)
 	die();
 
     dbg("doing copy relocations");
@@ -580,7 +582,8 @@ _rtld(Elf_Addr *sp, func_ptr_type *exit_proc, Obj_Entry **objp)
 
     dbg("resolving ifuncs");
     if (resolve_objects_ifunc(obj_main,
-      ld_bind_now != NULL && *ld_bind_now != '\0', NULL) == -1)
+      ld_bind_now != NULL && *ld_bind_now != '\0', SYMLOOK_EARLY,
+      NULL) == -1)
 	die();
 
     if (!obj_main->crt_no_init) {
@@ -1552,7 +1555,7 @@ init_rtld(caddr_t mapbase, Elf_Auxinfo **aux_info)
 	 * that symbols can be found.
 	 */
 
-	relocate_objects(&objtmp, true, &objtmp, NULL);
+	relocate_objects(&objtmp, true, &objtmp, 0, NULL);
     }
 
     /* Initialize the object list. */
@@ -1605,6 +1608,7 @@ initlist_add_neededs(Needed_Entry *needed, Objlist *list)
 static void
 initlist_add_objects(Obj_Entry *obj, Obj_Entry **tail, Objlist *list)
 {
+
     if (obj->init_scanned || obj->init_done)
 	return;
     obj->init_scanned = true;
@@ -1616,6 +1620,10 @@ initlist_add_objects(Obj_Entry *obj, Obj_Entry **tail, Objlist *list)
     /* Recursively process the needed objects. */
     if (obj->needed != NULL)
 	initlist_add_neededs(obj->needed, list);
+    if (obj->needed_filtees != NULL)
+	initlist_add_neededs(obj->needed_filtees, list);
+    if (obj->needed_aux_filtees != NULL)
+	initlist_add_neededs(obj->needed_aux_filtees, list);
 
     /* Add the object to the init list. */
     if (obj->preinit_array != (Elf_Addr)NULL || obj->init != (Elf_Addr)NULL ||
@@ -2144,13 +2152,17 @@ objlist_remove(Objlist *list, Obj_Entry *obj)
  */
 static int
 relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj,
-    RtldLockState *lockstate)
+    int flags, RtldLockState *lockstate)
 {
     Obj_Entry *obj;
 
     for (obj = first;  obj != NULL;  obj = obj->next) {
+	if (obj->relocated)
+	    continue;
+	obj->relocated = true;
 	if (obj != rtldobj)
 	    dbg("relocating \"%s\"", obj->path);
+
 	if (obj->nbuckets == 0 || obj->nchains == 0 || obj->buckets == NULL ||
 	    obj->symtab == NULL || obj->strtab == NULL) {
 	    _rtld_error("%s: Shared object has no run-time symbol table",
@@ -2169,7 +2181,7 @@ relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj,
 	}
 
 	/* Process the non-PLT relocations. */
-	if (reloc_non_plt(obj, rtldobj, lockstate))
+	if (reloc_non_plt(obj, rtldobj, flags, lockstate))
 		return -1;
 
 	if (obj->textrel) {	/* Re-protected the text segment. */
@@ -2190,7 +2202,7 @@ relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj,
 	    return -1;
 	/* Relocate the jump slots if we are doing immediate binding. */
 	if (obj->bind_now || bind_now)
-	    if (reloc_jmpslots(obj, lockstate) == -1)
+	    if (reloc_jmpslots(obj, flags, lockstate) == -1)
 		return -1;
 
 	if (obj->relro_size > 0) {
@@ -2225,35 +2237,39 @@ relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj,
  * consistent with how GNU does it.
  */
 static int
-resolve_object_ifunc(Obj_Entry *obj, bool bind_now, RtldLockState *lockstate)
+resolve_object_ifunc(Obj_Entry *obj, bool bind_now, int flags,
+    RtldLockState *lockstate)
 {
 	if (obj->irelative && reloc_iresolve(obj, lockstate) == -1)
 		return (-1);
 	if ((obj->bind_now || bind_now) && obj->gnu_ifunc &&
-	    reloc_gnu_ifunc(obj, lockstate) == -1)
+	    reloc_gnu_ifunc(obj, flags, lockstate) == -1)
 		return (-1);
 	return (0);
 }
 
 static int
-resolve_objects_ifunc(Obj_Entry *first, bool bind_now, RtldLockState *lockstate)
+resolve_objects_ifunc(Obj_Entry *first, bool bind_now, int flags,
+    RtldLockState *lockstate)
 {
 	Obj_Entry *obj;
 
 	for (obj = first;  obj != NULL;  obj = obj->next) {
-		if (resolve_object_ifunc(obj, bind_now, lockstate) == -1)
+		if (resolve_object_ifunc(obj, bind_now, flags, lockstate) == -1)
 			return (-1);
 	}
 	return (0);
 }
 
 static int
-initlist_objects_ifunc(Objlist *list, bool bind_now, RtldLockState *lockstate)
+initlist_objects_ifunc(Objlist *list, bool bind_now, int flags,
+    RtldLockState *lockstate)
 {
 	Objlist_Entry *elm;
 
 	STAILQ_FOREACH(elm, list, link) {
-		if (resolve_object_ifunc(elm->obj, bind_now, lockstate) == -1)
+		if (resolve_object_ifunc(elm->obj, bind_now, flags,
+		    lockstate) == -1)
 			return (-1);
 	}
 	return (0);
@@ -2515,17 +2531,30 @@ dlopen_object(const char *name, int fd, Obj_Entry *refobj, int lo_flags,
 	    objlist_push_tail(&list_global, obj);
 	if (*old_obj_tail != NULL) {		/* We loaded something new. */
 	    assert(*old_obj_tail == obj);
-	    result = load_needed_objects(obj, lo_flags & RTLD_LO_DLOPEN);
+	    result = load_needed_objects(obj,
+		lo_flags & (RTLD_LO_DLOPEN | RTLD_LO_EARLY));
 	    init_dag(obj);
 	    ref_dag(obj);
 	    if (result != -1)
 		result = rtld_verify_versions(&obj->dagmembers);
 	    if (result != -1 && ld_tracing)
 		goto trace;
-	    if (result == -1 || (relocate_objects(obj, (mode & RTLD_MODEMASK)
-	      == RTLD_NOW, &obj_rtld, &lockstate)) == -1) {
+	    if (result == -1 || (relocate_objects(obj,
+	     (mode & RTLD_MODEMASK) == RTLD_NOW, &obj_rtld,
+	      (lo_flags & RTLD_LO_EARLY) ? SYMLOOK_EARLY : 0,
+	      &lockstate)) == -1) {
 		dlopen_cleanup(obj);
 		obj = NULL;
+	    } else if (lo_flags & RTLD_LO_EARLY) {
+		/*
+		 * Do not call the init functions for early loaded
+		 * filtees.  The image is still not initialized enough
+		 * for them to work.
+		 *
+		 * Our object is found by the global object list and
+		 * will be ordered among all init calls done right
+		 * before transferring control to main.
+		 */
 	    } else {
 		/* Make list of init functions to call. */
 		initlist_add_objects(obj, &obj->next, &initlist);
@@ -2559,6 +2588,7 @@ dlopen_object(const char *name, int fd, Obj_Entry *refobj, int lo_flags,
     map_stacks_exec(&lockstate);
 
     if (initlist_objects_ifunc(&initlist, (mode & RTLD_MODEMASK) == RTLD_NOW,
+      (lo_flags & RTLD_LO_EARLY) ? SYMLOOK_EARLY : 0,
       &lockstate) == -1) {
 	objlist_clear(&initlist);
 	dlopen_cleanup(obj);
@@ -2566,8 +2596,10 @@ dlopen_object(const char *name, int fd, Obj_Entry *refobj, int lo_flags,
 	return (NULL);
     }
 
-    /* Call the init functions. */
-    objlist_call_init(&initlist, &lockstate);
+    if (!(lo_flags & RTLD_LO_EARLY)) {
+	/* Call the init functions. */
+	objlist_call_init(&initlist, &lockstate);
+    }
     objlist_clear(&initlist);
     lock_release(rtld_bind_lock, &lockstate);
     return obj;
@@ -3354,12 +3386,13 @@ symlook_obj(SymLook *req, const Obj_Entry *obj)
 {
     DoneList donelist;
     SymLook req1;
-    int res, mres;
+    int flags, res, mres;
 
     mres = symlook_obj1(req, obj);
     if (mres == 0) {
 	if (obj->needed_filtees != NULL) {
-	    load_filtees(__DECONST(Obj_Entry *, obj), 0, req->lockstate);
+	    flags = (req->flags & SYMLOOK_EARLY) ? RTLD_LO_EARLY : 0;
+	    load_filtees(__DECONST(Obj_Entry *, obj), flags, req->lockstate);
 	    donelist_init(&donelist);
 	    symlook_init_from_req(&req1, req);
 	    res = symlook_needed(&req1, obj->needed_filtees, &donelist);
@@ -3370,7 +3403,8 @@ symlook_obj(SymLook *req, const Obj_Entry *obj)
 	    return (res);
 	}
 	if (obj->needed_aux_filtees != NULL) {
-	    load_filtees(__DECONST(Obj_Entry *, obj), 0, req->lockstate);
+	    flags = (req->flags & SYMLOOK_EARLY) ? RTLD_LO_EARLY : 0;
+	    load_filtees(__DECONST(Obj_Entry *, obj), flags, req->lockstate);
 	    donelist_init(&donelist);
 	    symlook_init_from_req(&req1, req);
 	    res = symlook_needed(&req1, obj->needed_aux_filtees, &donelist);
diff --git a/libexec/rtld-elf/rtld.h b/libexec/rtld-elf/rtld.h
index 9dfa0a2..2aabffe 100644
--- a/libexec/rtld-elf/rtld.h
+++ b/libexec/rtld-elf/rtld.h
@@ -229,6 +229,7 @@ typedef struct Struct_Obj_Entry {
 
     bool mainprog : 1;		/* True if this is the main program */
     bool rtld : 1;		/* True if this is the dynamic linker */
+    bool relocated : 1;		/* True if processed by relocate_objects() */
     bool textrel : 1;		/* True if there are relocations to text seg */
     bool symbolic : 1;		/* True if generated with "-Bsymbolic" */
     bool bind_now : 1;		/* True if all relocations should be made first */
@@ -267,6 +268,7 @@ typedef struct Struct_Obj_Entry {
 #define SYMLOOK_IN_PLT	0x01	/* Lookup for PLT symbol */
 #define SYMLOOK_DLSYM	0x02	/* Return newest versioned symbol. Used by
 				   dlsym. */
+#define	SYMLOOK_EARLY	0x04	/* Symlook is done during initialization. */
 
 /* Flags for load_object(). */
 #define	RTLD_LO_NOLOAD	0x01	/* dlopen() specified RTLD_NOLOAD. */
@@ -274,6 +276,8 @@ typedef struct Struct_Obj_Entry {
 #define	RTLD_LO_TRACE	0x04	/* Only tracing. */
 #define	RTLD_LO_NODELETE 0x08	/* Loaded object cannot be closed. */
 #define	RTLD_LO_FILTEES 0x10	/* Loading filtee. */
+#define	RTLD_LO_EARLY	0x20	/* Do not call ctors, postpone it to the
+				   initialization during the image start. */
 
 /*
  * Symbol cache entry used during relocation to avoid multiple lookups
@@ -353,11 +357,12 @@ const Ver_Entry *fetch_ventry(const Obj_Entry *obj, unsigned long);
  * MD function declarations.
  */
 int do_copy_relocations(Obj_Entry *);
-int reloc_non_plt(Obj_Entry *, Obj_Entry *, struct Struct_RtldLockState *);
+int reloc_non_plt(Obj_Entry *, Obj_Entry *, int flags,
+    struct Struct_RtldLockState *);
 int reloc_plt(Obj_Entry *);
-int reloc_jmpslots(Obj_Entry *, struct Struct_RtldLockState *);
+int reloc_jmpslots(Obj_Entry *, int flags, struct Struct_RtldLockState *);
 int reloc_iresolve(Obj_Entry *, struct Struct_RtldLockState *);
-int reloc_gnu_ifunc(Obj_Entry *, struct Struct_RtldLockState *);
+int reloc_gnu_ifunc(Obj_Entry *, int flags, struct Struct_RtldLockState *);
 void allocate_initial_tls(Obj_Entry *);
 
 #endif /* } */
diff --git a/libexec/rtld-elf/sparc64/reloc.c b/libexec/rtld-elf/sparc64/reloc.c
index 4ab0eae..ce0d945 100644
--- a/libexec/rtld-elf/sparc64/reloc.c
+++ b/libexec/rtld-elf/sparc64/reloc.c
@@ -231,7 +231,7 @@ static const long reloc_target_bitmask[] = {
 	__asm __volatile("flush %0 + %1" : : "r" (va), "I" (offs));
 
 static int reloc_nonplt_object(Obj_Entry *obj, const Elf_Rela *rela,
-    SymCache *cache, RtldLockState *lockstate);
+    SymCache *cache, int flags, RtldLockState *lockstate);
 static void install_plt(Elf_Word *pltgot, Elf_Addr proc);
 
 extern char _rtld_bind_start_0[];
@@ -264,6 +264,7 @@ do_copy_relocations(Obj_Entry *dstobj)
 			symlook_init(&req, name);
 			req.ventry = fetch_ventry(dstobj,
 			    ELF_R_SYM(rela->r_info));
+			req.flags = SYMLOOK_EARLY;
 
 			for (srcobj = dstobj->next; srcobj != NULL;
 			    srcobj = srcobj->next) {
@@ -291,7 +292,8 @@ do_copy_relocations(Obj_Entry *dstobj)
 }
 
 int
-reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
+reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
+    RtldLockState *lockstate)
 {
 	const Elf_Rela *relalim;
 	const Elf_Rela *rela;
@@ -310,7 +312,7 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate)
 
 	relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
 	for (rela = obj->rela; rela < relalim; rela++) {
-		if (reloc_nonplt_object(obj, rela, cache, lockstate) < 0)
+		if (reloc_nonplt_object(obj, rela, cache, flags, lockstate) < 0)
 			goto done;
 	}
 	r = 0;
@@ -322,7 +324,7 @@ done:
 
 static int
 reloc_nonplt_object(Obj_Entry *obj, const Elf_Rela *rela, SymCache *cache,
-    RtldLockState *lockstate)
+    int flags, RtldLockState *lockstate)
 {
 	const Obj_Entry *defobj;
 	const Elf_Sym *def;
@@ -385,7 +387,7 @@ reloc_nonplt_object(Obj_Entry *obj, const Elf_Rela *rela, SymCache *cache,
 	if (RELOC_RESOLVE_SYMBOL(type)) {
 		/* Find the symbol. */
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    false, cache, lockstate);
+		    flags, cache, lockstate);
 		if (def == NULL)
 			return (-1);
 
@@ -526,7 +528,7 @@ reloc_plt(Obj_Entry *obj)
 #define	LOVAL(v)	((v) & 0x000003ff)
 
 int
-reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
+reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	const Obj_Entry *defobj;
 	const Elf_Rela *relalim;
@@ -540,7 +542,7 @@ reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate)
 		assert(ELF64_R_TYPE_ID(rela->r_info) == R_SPARC_JMP_SLOT);
 		where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
-		    true, NULL, lockstate);
+		    SYMLOOK_IN_PLT | flags, NULL, lockstate);
 		if (def == NULL)
 			return -1;
 		target = (Elf_Addr)(defobj->relocbase + def->st_value);
@@ -559,7 +561,8 @@ reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
 }
 
 int
-reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
+reloc_gnu_ifunc(Obj_Entry *obj, int flags,
+    struct Struct_RtldLockState *lockstate)
 {
 
 	/* XXX not implemented */