Re: [PATCH] util/oslib-posix : qemu_init_exec_dir implementation for MacOS

[Peter Maydell]

On Wed, 3 Jun 2020 at 15:23, David CARLIER <devnexen@gmail.com> wrote:
>
> Good point even tough the libproc api is here in this form since quite a time.
>
> From d23bf60961ee036f8298794f879d1b8b9bd717dc Mon Sep 17 00:00:00 2001
> From: David Carlier <devnexen@gmail.com>
> Date: Tue, 26 May 2020 21:35:27 +0100
> Subject: [PATCH] util/oslib: current process full path resolution on MacOS
>
> Using existing libproc to fill the path.

Could you send new versions of the patch as their own emails
(ie not replies to the first one) with "[PATCH v2]" "[PATCH v3]"
etc in the subject line, please? Otherwise it gets tricky to
figure out which is the most recent version of the patch.

> Signed-off-by: David Carlier <devnexen@gmail.com>
> ---
>  util/oslib-posix.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
>
> diff --git a/util/oslib-posix.c b/util/oslib-posix.c
> index 062236a1ab..9dd1e1a18b 100644
> --- a/util/oslib-posix.c
> +++ b/util/oslib-posix.c
> @@ -55,6 +55,10 @@
>  #include <sys/sysctl.h>
>  #endif
>
> +#ifdef __APPLE__
> +#include <libproc.h>
> +#endif
> +
>  #include "qemu/mmap-alloc.h"
>
>  #ifdef CONFIG_DEBUG_STACK_USAGE
> @@ -366,6 +370,15 @@ void qemu_init_exec_dir(const char *argv0)
>              p = buf;
>          }
>      }
> +#elif defined(__APPLE__)
> +    {
> +        int len;
> +        len = proc_pidpath(getpid(), buf, sizeof(buf) - 1);
> +        if (len <= 0) {
> +            return;

I think that if proc_pidpath() doesn't work we should fall back
to the "try argv[0]" code below, not return without initializing
exec_dir[]. This is what the existing Linux and BSD codepaths do.

> +        }
> +        p = buf;
> +    }
>  #endif
>      /* If we don't have any way of figuring out the actual executable
>         location then try argv[0].  */

I did a bit of searching to see whether there were any alternatives
to proc_pidpath(), and I found _NSGetExecutablePath(). This
function *is* documented:
 
https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man3/dyld.3.html

Will it do what we want here? You'll need to call
realpath() on the results, and we should test that it
actually does better than just looking at argv[0] -- ie
that if you start QEMU via execve(/path/to/qemu,
argv_with_bogus_argv0, ...) then we get the /path/to/qemu,
not whatever the bogus argv0 value was.

There's also the "get the ui/cocoa.m code to find the path
via the AppKit API" approach that Roman suggested, though I
think that is more awkward than _NSGetExecutablePath() as
you'd need to get the ui/cocoa.m code to save the path for you.

If that doesn't work then I guess we can use proc_pidpath(),
but I'd rather avoid that if we can. If we do have to go that
route we should have a comment noting that it's an undocumented
and unsupported API but that it's also used by well-known
applications X, Y...

thanks
-- PMM