Re: [Qemu-devel] fix ELF loading for 0-length sections

[Peter Maydell <peter.maydell@linaro.org>]

On 21 February 2012 10:42, Damian, Alexandru <alexandru.damian@intel.com> wrote:
> Hi,
>
> I got a problem with QEMU refusing to load an ELF binary with 0-length
> sections,
> while the kernel has no issue doing this.
>
> This patch adds a check that has been in kernel since 2008 at least.

CC'ing Riku (linux-user maintainer).

>
> Cheers,
> Alex
>
> ----------------
> commit a42e5231c1be5f09caeb6c73e34933cd7efa7023
> Author: Alexandru DAMIAN <alexandru.damian@intel.com>
> Date:   Tue Feb 21 12:34:36 2012 +0200
>
>     Do not attempt to map 0-length sections
>
>     Mmap will return an invalid argument, but 0-length sections
>     are valid in any case. The kernel as a similar check
>     when loading elf binaries courtesy of jkosina@suse.cz.
>
>     Signed-off-by: Alexandru Damian <alexandru.damian@intel.com>

Convention is to submit patches as git-format-patch style emails.

> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
> index ea61d0d..71d0ae3 100644
> --- a/linux-user/elfload.c
> +++ b/linux-user/elfload.c
> @@ -918,9 +918,9 @@ static inline void init_thread(struct target_pt_regs
> *regs,
>
>  #define elf_check_arch(x) ( (x) == ELF_ARCH )
>
> -#define ELF_CLASS    ELFCLASS64
> -#define ELF_DATA    ELFDATA2MSB
> -#define ELF_ARCH    EM_S390
> +#define ELF_CLASS        ELFCLASS64
> +#define ELF_DATA        ELFDATA2MSB
> +#define ELF_ARCH        EM_S390

Please don't make unrelated whitespace changes.

>
>  static inline void init_thread(struct target_pt_regs *regs, struct
> image_info *infop)
>  {
> @@ -1565,11 +1565,16 @@ static void load_elf_image(const char *image_name,
> int image_fd,
>              vaddr_po = TARGET_ELF_PAGEOFFSET(vaddr);
>              vaddr_ps = TARGET_ELF_PAGESTART(vaddr);
>
> -            error = target_mmap(vaddr_ps, eppnt->p_filesz + vaddr_po,
> -                                elf_prot, MAP_PRIVATE | MAP_FIXED,
> -                                image_fd, eppnt->p_offset - vaddr_po);
> -            if (error == -1) {
> -                goto exit_perror;
> +            /* Don't attempt to map 0 bytes len sections.
> +               Kernel also has this check.
> +            */

This comment could be better:
   /* mmap() will fail EINVAL if given a zero size, but a
    * segment with zero filesize is perfectly valid (and
    * handled by the kernel's ELF loading code).
    */

> +            if (eppnt->p_filesz + vaddr_po != 0) {
> +                    error = target_mmap(vaddr_ps, eppnt->p_filesz +
> vaddr_po,
> +                                        elf_prot, MAP_PRIVATE | MAP_FIXED,
> +                                        image_fd, eppnt->p_offset -
> vaddr_po);

Something in your patch submission path is wrapping long lines.

> +                    if (error == -1) {
> +                        goto exit_perror;
> +                    }

QEMU coding style is four-space indents, but the indent in your new
code is mostly eight-space.

>              }
>
>              vaddr_ef = vaddr + eppnt->p_filesz;
>

Otherwise looks OK.

-- PMM