switch stacks

arch/sysv_x86-64/loader-trampoline.asm

@@ -0,0 +1,55 @@
+
+bits 64
+
+section .text
+global ld_stack_trampoline
+
+; DESCRIPTION:
+;   This function copies the current stack down from RSP to the top of the stack
+;   into a memory region provided in parameter
+;   After the stack has been copied, we jump into ld_stage3_entry
+; PARAMS:
+;   RDI (u64) - dst stack base
+;   RSI (u64) - src stack base
+;   RDX (u64) - dst stack size
+;   RCX (u64) - src stack size
+;   R8  (u64) - function to call
+;   R9  (u64) - param to pass to function
+; RETURNS:
+;   guess what, it doesn't
+ld_stack_trampoline:
+    ; Calculate the amount of memory we need to move to the new stack
+    add rcx, rsi
+    sub rcx, rsp
+    ; Calculate the base of the source memory
+    mov rsi, rsp
+    ; Check to see if the amount we're about to move is less than dst stack size
+    cmp rcx, rdx
+    jl .continue
+    xor rdi, rdi
+    mov rsi, r9
+    mov rax, r8
+    call rax
+.continue:
+    ; Save rsi, rdi, rcx on stack
+    push rsi
+    push rdi
+    push rcx
+    ; copy stack
+    shr rcx, 3
+    rep movsq
+    ; get back our regs
+    pop rcx
+    pop rdi
+    pop rsi
+    ; We copied the stack at this point, lets switch
+    add rsp, rdi
+    sub rsp, rsi
+    add rsp, rdx
+    sub rsp, rcx
+    ; We should have the new stack now, which is an exact copy of the old stack
+    and rsp, -16
+    mov rdi, 1
+    mov rsi, r9
+    mov rax, r8
+    call rax

build.py

@@ -144,7 +144,8 @@ if loader_debugging_enabled:
     cc_defines.append('_CIA_LD_DEBUG')
 print_step("Building lib/ld-cia.so\n")
 assemble_obj('bin/loader-entry.o', [f'arch/{target_abi}_{target_arch}/loader-entry.asm'], ['-f "elf64"'])
-compile_shared('lib/ld-cia.so', ['bin/loader-entry.o','loader/loader-self-reloc.c','loader/loader.c'])
+assemble_obj('bin/loader-trampoline.o', [f'arch/{target_abi}_{target_arch}/loader-trampoline.asm'], ['-f "elf64"'])
+compile_shared('lib/ld-cia.so', ['bin/loader-entry.o','loader/loader-self-reloc.c','loader/loader.c','bin/loader-trampoline.o'])
 cc_defines_pop()
 cc_flags_pop()
 

loader/loader-self-reloc.c

@@ -14,18 +14,27 @@
 // are fine with gcc and clang (I think))
 
 #include <cia/def.h>
+#include <stdarg.h>
 #include <bin/elf.h>
 #include <sys/syscall.h>
-#include <stdarg.h>
+#include <sys/mman.h>
+#include <sched.h>
+#include <errno.h>
+#include <fcntl.h>
+
 #include "loader.h"
 
-extern void loader_entry(Loader_Info *ld_info);
+extern void ld_stage2_entry(Loader_Info *ld_info);
 
-void _dlstart_reloc_c(u64 *sp, Elf64_Dyn *dynv) {
+void ld_stage1_entry(u64 *sp, Elf64_Dyn *dynv) {
     _dbg_print_string("Entered dynamic loader\n");
     // Find argc, argv in stack
     int argc = *sp;
     char **argv = (void *)(sp+1);
+    _dbg_printf("ARGV:\n");
+    for(int i = 0; i < argc; ++i) {
+        _dbg_printf("[%d]: %s\n", (i64)i, argv[i]);
+    }
     // Skip over environment
     char **envp = argv+argc+1;
     int envc = 0;
@@ -173,7 +182,6 @@ void _dlstart_reloc_c(u64 *sp, Elf64_Dyn *dynv) {
             }
             rela_offs += rela_ent;
         }
-        
     }
     _mfence();
     Loader_Info ld_info;
@@ -182,7 +190,7 @@ void _dlstart_reloc_c(u64 *sp, Elf64_Dyn *dynv) {
     ld_info.dyn = dyn;
     ld_info.aux = aux;
     _dbg_printf("Self-relocation finished. Entering the loader\n");
-    loader_entry(&ld_info);
+    ld_stage2_entry(&ld_info);
     sys_exit(0);
 }
 

loader/loader.c

@@ -39,6 +39,7 @@ struct Elf_Image {
 #define elf_addr(elf, off) (void *)((elf)->base + (u64)off)
 
 static Cia_Pool image_pool;
+static Cia_Arena tmp_arena;
 
 static u32 elf_sym_gnu_hash(char *name) {
     unsigned char *s = (void *)name;
@@ -91,7 +92,24 @@ static Elf64_Sym *elf_symbol_by_name(Elf_Image *image, char *name) {
     return NULL;
 }
 
-void loader_entry(Loader_Info *ld_info) {
+struct Stage3_Info_Struct typedef Stage3_Info_Struct;
+struct Stage3_Info_Struct {
+    Elf_Image *app;
+    Elf_Image *ldso;
+};
+
+static void ld_stage3_entry(u64 has_new_stack, void *ctx);
+
+void ld_stack_trampoline(
+    void *stack_base
+    , void *old_stack_base
+    , u64 stack_size
+    , u64 old_stack_size
+    , void (*fn)(u64 has_new_stack, void *ctx)
+    , void *ctx
+);
+
+void ld_stage2_entry(Loader_Info *ld_info) {
     _dbg_printf("Loader entry point reached!\n");
     // Get our loader data back
     u64 *sp = ld_info->sp;
@@ -101,7 +119,11 @@ void loader_entry(Loader_Info *ld_info) {
     cia_pool_create(&image_pool, cia_allocator_pages(), 1*KB, sizeof(Elf_Image), 0x10);
     Elf_Image *ldso = cia_pool_alloc(&image_pool);
     Elf_Image *app  = cia_pool_alloc(&image_pool);
-    cia_arena_create(&app->arena, cia_allocator_pages(), 1*MB);
+    // Save some stuff for stage3 to eat
+    cia_arena_create(&tmp_arena, cia_allocator_pages(), 1*MB);
+    Stage3_Info_Struct *stage3 = cia_arena_alloc(&tmp_arena, sizeof(Stage3_Info_Struct));
+    stage3->app = app;
+    stage3->ldso = ldso;
     ldso->base = ldso_base;
     ldso->dyn = dyn;
     // Read ldso elf header
@@ -150,14 +172,6 @@ void loader_entry(Loader_Info *ld_info) {
     if(fd != 0) {
         sys_close(fd);
     }
-    // Get the information about the main thread stack
-    if(linux_read_stack_info()) {
-        printf("ERROR: failed to read /proc/self/maps to get the stack info\n");
-        sys_exit(1);
-    }
-    _dbg_printf("Received stack: %x-%x\n", stack_info.start_addr, stack_info.end_addr);
-    u64 *ptr = (void *)stack_info.start_addr;
-    // *ptr = 1245;
     // Find .dynamic section
     {
         u8 *phdr = (void *)aux[AT_PHDR];
@@ -281,9 +295,32 @@ void loader_entry(Loader_Info *ld_info) {
         // Get the app main
         // Elf64_Sym *app_main = elf_symbol_by_name(app, "main");
         // _dbg_printf("app main: %x\n", app_main);
-        void (*crt_entry)() = elf_addr(app, eh->e_entry);
-        _dbg_printf("Exiting the dynamic loader, trying to enter the main app\n");
-        crt_entry();
+        // Get the information about the main thread stack
+        if(linux_read_stack_info()) {
+            printf("ERROR: failed to read /proc/self/maps to get the stack info\n");
+            sys_exit(1);
+        }
+        _dbg_printf("Found default stack at: %x-%x\n", stack_info.start_addr, stack_info.end_addr);
+        void *old_stack_base = (void *)stack_info.start_addr;
+        u64 old_stack_size = (u64)stack_info.end_addr - (u64)stack_info.start_addr;
+        u64 stack_size = 0x10000;
+        void *stack_base = sys_mmap(0, stack_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0);
+        _dbg_printf("stage3 info struct: %x\n", stage3);
+        // Will jump to ld_stage3_entry
+        ld_stack_trampoline(stack_base, old_stack_base, stack_size, old_stack_size, &ld_stage3_entry, stage3);
     }
     sys_exit(0);
 }
+
+static void ld_stage3_entry(u64 has_new_stack, void *ctx) {
+    if(!has_new_stack) {
+        printf("ERROR: failed to switch the stack\n");
+        sys_exit(1);
+    }
+    Stage3_Info_Struct *info = ctx;
+    _dbg_printf("Entered loader stage 3. Try entering main executable\n");
+    _dbg_printf("stage3 info struct: %x\n", info);
+    void (*crt_entry)() = elf_addr(info->app, ((Elf64_Ehdr *)info->app->base)->e_entry);
+    _dbg_printf("Entry at: %x\n", crt_entry);
+    crt_entry();
+}