[Bug dynamic-link/31662] New: Stack not aligned at _start when calling ld.so explicitly

[Bug dynamic-link/31662] New: Stack not aligned at _start when calling ld.so explicitly

[mail@felix-potthast.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

            Bug ID: 31662
           Summary: Stack not aligned at _start when calling ld.so
                    explicitly
           Product: glibc
           Version: 2.35
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: dynamic-link
          Assignee: unassigned at sourceware dot org
          Reporter: mail@felix-potthast.de
  Target Milestone: ---

Created attachment 15474
  --> https://sourceware.org/bugzilla/attachment.cgi?id=15474&action=edit
Test case

Up to glibc version 2.35 the stack is not 16-byte aligned when the process is
entered at _start when ld.so was called explicitly.

The x86-64 SYS V ABI explicitly states it has to be aligned at that point,
though.
It also works fine when ld.so is invoked as elf interpreter.

This doesn't result in any issue most of the time because the stack
is aligned between _start and main in `sysdeps/x86_64/start.S`, for legacy
reasons probably.

I think the right place to do this alignment is at the end of ld.so.


I attached a simple example that can be build with
`gcc -nostartfiles test.S -o test`.

It runs fine when running `./test` but segfaults when running with
`/lib64/ld-linux-x86-64.so.2 ./test` due to missing stack alignment.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

--- Comment #1 from Andreas Schwab <schwab@linux-m68k.org> ---
*** Bug 31663 has been marked as a duplicate of this bug. ***

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

--- Comment #2 from Andreas Schwab <schwab@linux-m68k.org> ---
At the start of a normal function, the stack pointer is always 8 mod 16, due to
the return address on the stack.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
The assembly code doesn't align stack when calling memcpy:

[hjl@gnu-cfl-3 tmp]$ cat x.c
extern void *dest;
extern void *src;
extern void copy (void *, void *, long);

int
foo (void)
{
  copy (dest, src, 5);
  return 0;
}
[hjl@gnu-cfl-3 tmp]$ gcc -S -O2 x.c
[hjl@gnu-cfl-3 tmp]$ cat x.s
        .file   "x.c"
        .text
        .p2align 4
        .globl  foo
        .type   foo, @function
foo:
.LFB0:
        .cfi_startproc
        subq    $8, %rsp   <<<<<<<<<< Align stack to 16 bytes before calling
copy.
        .cfi_def_cfa_offset 16
        movq    src(%rip), %rsi
        movq    dest(%rip), %rdi
        movl    $5, %edx
        call    copy
        xorl    %eax, %eax
        addq    $8, %rsp
        .cfi_def_cfa_offset 8
        ret
        .cfi_endproc
.LFE0:
        .size   foo, .-foo
        .ident  "GCC: (GNU) 14.0.1 20240411 (Red Hat 14.0.1-0)"
        .section        .note.GNU-stack,"",@progbits
[hjl@gnu-cfl-3 tmp]$

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[mail@felix-potthast.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

--- Comment #4 from Felix Potthast <mail@felix-potthast.de> ---
@Andreas Schwab:
Sure, but the elf entry point is not a normal function and there is no return
address on the stack. It is not called but jumped to at the end of
_dl_start_user.

@H.J.Lu:
Yes, this is intentional. The point is: the stack pointer is guaranteed to be
16-byte aligned at process entry by the ABI. So i should'nt have to align it
before jumping into memcpy.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

--- Comment #5 from Andreas Schwab <schwab@linux-m68k.org> ---
Of course, that's why I wrote normal function.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

Andreas Schwab <schwab@linux-m68k.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |2.36
         Resolution|INVALID                     |FIXED

--- Comment #6 from Andreas Schwab <schwab@linux-m68k.org> ---
This has been fixed as a side effect of commit ec7bc492b6.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/31662] Stack not aligned at _start when calling ld.so explicitly

[mail@felix-potthast.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=31662

--- Comment #7 from Felix Potthast <mail@felix-potthast.de> ---
Yes, this is not occurring anymore since 2.36.
2.35 is still around, though. (Ubuntu 22.04 for example)

Do you have any plans on preventing this from occurring again?
This seems to have been unnoticed.

I would suggest removing the align from `sysdeps/x86_64/start.S`. (Lines 88,89)
It goes back to 2001 when the ABI didn't explicitly state the alignment at
process
entry. It shouldn't be needed anymore and just masks bugs like this.

-- 
You are receiving this mail because:
You are on the CC list for the bug.