Added support for creation of Static Position-Independent Executables (PIE) on i386, x86_64, and arm.

This patch adds the generation of rcrt1.o which is used by gcc when compiling with the --static-pie flag.

rcrt1.o differs from crt1.o and Scrt1.o in that it the executable has a dynamic section but no relocations have been performed prior to _start being called.
crt1.o assumes there to be no dynamic relocations, and Scrt1.o has all relocations performed prior to execution by lsdo.

The new reloc_static_pie function handles parsing the dynamic section, and performing the relocations in a architecture agnostic method.
It also sets _dl_load_base which is used when initalizing TLS to ensure loading from the proper location.
This allows for easier porting of static-pie support to additional architectures as only modifications to crt1.S to find the load address are required.

Signed-off-by: linted <linted@users.noreply.github.com>

3 files changed, 54 insertions, 1 deletions

diff --git a/libc/sysdeps/linux/arm/crt1.S b/libc/sysdeps/linux/arm/crt1.S
index a1d7f0f23..fade1d25c 100644
--- a/libc/sysdeps/linux/arm/crt1.S
+++ b/libc/sysdeps/linux/arm/crt1.S
@@ -246,6 +246,22 @@ _start:
 	mov lr, #0
 
 #ifdef __ARCH_USE_MMU__
+#ifdef L_rcrt1
+	/* We don't need to save a1 since no dynamic linker should have run */
+	ldr a1, .L_GOT          /* Get value at .L_GOT + 0  (offset to GOT)*/
+	adr a2, .L_GOT          /* Get address of .L_GOT */
+	ldr a3, .L_GOT+16       /* Get value of _start(GOT) stored in .L_GOT */
+	adr a4, _start          /* Get address of _start after relocation (changes to pc - ~30 or so) */
+	add a1, a1, a2          /* Calculate where the GOT is */
+	ldr a2, [a1, a3]        /* GOT + _start(GOT) = offset of _start from begin of file */
+	sub a1, a4, a2          /* Current addr of _start - offset from beginning of file = load addr */
+	bl reloc_static_pie
+	mov a1, #0              /* Clean up a1 so that a random address won't get called at the end of program */
+
+	/* Clear the frame pointer and link register again since it might be modified by previous call */
+	mov fp, #0
+	mov lr, #0
+#endif
 	/* Pop argc off the stack and save a pointer to argv */
 	ldr a2, [sp], #4
 	mov a3, sp
@@ -309,6 +325,9 @@ _start:
 	.word _fini(GOT)
 	.word _init(GOT)
 	.word main(GOT)
+#ifdef L_rcrt1
+	.word _start(GOT)
+#endif
 #endif
 #endif
 
diff --git a/libc/sysdeps/linux/i386/crt1.S b/libc/sysdeps/linux/i386/crt1.S
index 35a6552e8..decc68967 100644
--- a/libc/sysdeps/linux/i386/crt1.S
+++ b/libc/sysdeps/linux/i386/crt1.S
@@ -67,6 +67,9 @@
 #endif
 .type   main,%function
 .type   __uClibc_main,%function
+#ifdef L_rcrt1
+.type	reloc_static_pie,%function
+#endif
 _start:
 	/* Clear the frame pointer.  The ABI suggests this be done, to mark
 	   the outermost frame obviously.  */
@@ -100,6 +103,23 @@ _start:
 	pop %ebx
 	addl $_GLOBAL_OFFSET_TABLE_+[.-.L0],%ebx
 
+#ifdef L_rcrt1
+	/* We cannot rely on _DYNAMIC being usable here due to RELRO.
+	   Instead we calculate the load address based off a symbol
+	   that we know will exist, _start. */
+	pushl %ecx                      /* Save ecx so it won't get clobbered */
+	pushl %ebx                      /* Save ebx so it won't get clobbered */
+	xorl %ecx, %ecx                 /* Clear ecx */
+	addl _start@GOT(%ebx), %ecx     /* Get the offset of _start */
+	movl _start@GOT(%ebx), %eax     /* Get the run time address of _start */
+	subl %ecx, %eax                 /* Subtract to find the load address */
+	pushl %eax                      /* Pass the load address */
+	call reloc_static_pie@PLT
+	popl %eax                       /* Clean up from function call */
+	popl %ebx                       /* Restore the GOT address */
+	popl %ecx                       /* restore ecx */
+#endif
+
 	/* Push address of our own entry points to .fini and .init.  */
 	pushl _fini@GOT(%ebx)
 	pushl _init@GOT(%ebx)
diff --git a/libc/sysdeps/linux/x86_64/crt1.S b/libc/sysdeps/linux/x86_64/crt1.S
index 87777dd5d..701cbf2f6 100644
--- a/libc/sysdeps/linux/x86_64/crt1.S
+++ b/libc/sysdeps/linux/x86_64/crt1.S
@@ -80,6 +80,20 @@ _start:
 	   the outermost frame obviously.  */
 	xorl %ebp, %ebp
 
+#ifdef L_rcrt1
+	pushq %rdi                          /* save rdi (but should be 0...) */
+	pushq %rdx                          /* store rdx (rtld_fini) */
+	xorq %rcx, %rcx                     /* ensure rcx is 0 */
+	addq _start@GOTPCREL(%rip), %rcx    /* get offset of _start from beginning of file */
+	movq _start@GOTPCREL(%rip), %rax    /* get run time address of _start */
+	subq %rcx, %rax                     /* calculate run time load offset */
+	movq %rax, %rdi                     /* load offset -> param 1 */
+	call reloc_static_pie               /* relocate dynamic addrs */
+	xorq %rax, %rax                     /* cleanup */
+	popq %rdx
+	popq %rdi
+#endif
+
 	/* Extract the arguments as encoded on the stack and set up
 	   the arguments for __libc_start_main (int (*main) (int, char **, char **),
 		   int argc, char *argv,
@@ -107,7 +121,7 @@ _start:
 	   which grow downwards).  */
 	pushq %rsp
 
-#if defined(L_Scrt1)
+#if defined(L_Scrt1) || defined(L_rcrt1)
 	/* Give address for main() */
 	movq main@GOTPCREL(%rip), %rdi