[glibc/arm/gcs] aarch64: Enable GCS in dynamic linked exe

[Szabolcs Nagy <nsz@sourceware.org>]

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=7ea8526a50e6867b154d2bb4fbe9de4ff2fc9468

commit 7ea8526a50e6867b154d2bb4fbe9de4ff2fc9468
Author: Szabolcs Nagy <szabolcs.nagy@arm.com>
Date:   Fri Jul 14 15:49:11 2023 +0100

    aarch64: Enable GCS in dynamic linked exe
    
    Use the dynamic linker start code to enable GCS in the dynamic linked
    case after _dl_start returns and before _dl_start_user which marks
    the point after which user code may run.
    
    Like in the static linked case this ensures that GCS is enabled on a
    top level stack frame.

Diff:
---
 sysdeps/aarch64/Makefile                |  4 +++-
 sysdeps/aarch64/dl-start.S              | 23 +++++++++++++++++++++--
 sysdeps/aarch64/rtld-global-offsets.sym |  5 +++++
 3 files changed, 29 insertions(+), 3 deletions(-)

diff --git a/sysdeps/aarch64/Makefile b/sysdeps/aarch64/Makefile
index 141d7d9cc2..ca8b96f550 100644
--- a/sysdeps/aarch64/Makefile
+++ b/sysdeps/aarch64/Makefile
@@ -35,7 +35,9 @@ endif
 ifeq ($(subdir),elf)
 sysdep-rtld-routines += dl-start
 sysdep-dl-routines += tlsdesc dl-tlsdesc
-gen-as-const-headers += dl-link.sym
+gen-as-const-headers += \
+  dl-link.sym \
+  rtld-global-offsets.sym
 
 tests-internal += tst-ifunc-arg-1 tst-ifunc-arg-2
 
diff --git a/sysdeps/aarch64/dl-start.S b/sysdeps/aarch64/dl-start.S
index d645484e79..271bd5bf00 100644
--- a/sysdeps/aarch64/dl-start.S
+++ b/sysdeps/aarch64/dl-start.S
@@ -18,6 +18,7 @@
    <https://www.gnu.org/licenses/>.  */
 
 #include <sysdep.h>
+#include <rtld-global-offsets.h>
 
 ENTRY (_start)
 	/* Create an initial frame with 0 LR and FP */
@@ -25,11 +26,30 @@ ENTRY (_start)
 	mov	x29, #0
 	mov	x30, #0
 
+	/* Load and relocate all library dependencies.  */
 	mov	x0, sp
 	PTR_ARG (0)
 	bl	_dl_start
 	/* Returns user entry point in x0.  */
 	mov	PTR_REG (21), PTR_REG (0)
+
+	/* Use GL(dl_aarch64_gcs) to set the shadow stack status.  */
+	adrp	x16, _rtld_local
+	add	PTR_REG (16), PTR_REG (16), :lo12:_rtld_local
+	ldr	x1, [x16, GL_DL_AARCH64_GCS_OFFSET]
+	cbz	x1, L(skip_gcs_enable)
+
+	/* Enable GCS before user code runs.  Note that IFUNC resolvers and
+	   LD_AUDIT hooks may run before, but should not create threads.  */
+#define PR_SET_SHADOW_STACK_STATUS     72
+	mov	x0, PR_SET_SHADOW_STACK_STATUS
+	mov	x2, 0
+	mov	x3, 0
+	mov	x4, 0
+	mov	x8, #SYS_ify(prctl)
+	svc	0x0
+L(skip_gcs_enable):
+
 .globl _dl_start_user
 .type _dl_start_user, %function
 _dl_start_user:
@@ -40,8 +60,7 @@ _dl_start_user:
 	/* Compute envp.  */
 	add	PTR_REG (3), PTR_REG (2), PTR_REG (1), lsl PTR_LOG_SIZE
 	add	PTR_REG (3), PTR_REG (3), PTR_SIZE
-	adrp	x16, _rtld_local
-	add	PTR_REG (16), PTR_REG (16), :lo12:_rtld_local
+	/* Run the init functions of the loaded modules.  */
 	ldr	PTR_REG (0), [x16]
 	bl	_dl_init
 	/* Load the finalizer function.  */
diff --git a/sysdeps/aarch64/rtld-global-offsets.sym b/sysdeps/aarch64/rtld-global-offsets.sym
index 23cdaf7d9e..6c0690bb95 100644
--- a/sysdeps/aarch64/rtld-global-offsets.sym
+++ b/sysdeps/aarch64/rtld-global-offsets.sym
@@ -3,8 +3,13 @@
 #include <ldsodefs.h>
 
 #define GLRO_offsetof(name) offsetof (struct rtld_global_ro, _##name)
+#define GL_offsetof(name) offsetof (struct rtld_global, _##name)
 
 -- Offsets of _rtld_global_ro in libc.so
 
 GLRO_DL_HWCAP_OFFSET	GLRO_offsetof (dl_hwcap)
 GLRO_DL_HWCAP2_OFFSET	GLRO_offsetof (dl_hwcap2)
+
+-- Offsets of _rtld_global in libc.so
+
+GL_DL_AARCH64_GCS_OFFSET	GL_offsetof (dl_aarch64_gcs)