[Tarantool-patches] [PATCH luajit 1/2] symtab: fix static symtab dump

[Sergey Kaplun]

Hi, Maxim!

Thanks for the patch!
LGTM, with several comments below.

On 04.07.22, Maxim Kokryashkin wrote:
> The `dl_iterate_phdr` returns an empty string as a name for
> the executable from which it was called.

Don't get: why it is the problem? I.e. why it is problem to obtain the
main symbol table? Is it sysprof parser restrictions?

>                                          It is still possible
> to access its dynamic symbol table, but it is vital for
> sysprof to obtain the main symbol table for the LuaJIT
> executable. To do so, we need a valid path to the executable.
> 
> Since there is no way to obtain the path to a running executable
> using the C standard library, this commit adds call to
> `readlink` to gather the symbolic link from `/proc/self/exe`.
> Most of the UNIX-based systems have procfs, so it is not
> a problem.
> 
> Needed for tarantool/tarantool#7244
> ---
>  src/lj_symtab.c | 28 ++++++++++++++++++++++++++++
>  1 file changed, 28 insertions(+)
> 
> diff --git a/src/lj_symtab.c b/src/lj_symtab.c
> index 2b2b205b..23696401 100644
> --- a/src/lj_symtab.c
> +++ b/src/lj_symtab.c
> @@ -16,10 +16,12 @@
>  
>  #if LJ_HASRESOLVER
>  
> +#include <linux/limits.h>
>  #include <elf.h>
>  #include <link.h>
>  #include <stdio.h>
>  #include <sys/auxv.h>
> +#include <unistd.h>
>  #include "lj_gc.h"
>  #endif
>  
> @@ -352,6 +354,7 @@ static int resolve_symbolnames(struct dl_phdr_info *info, size_t info_size,
>    struct lj_wbuf *buf = conf->buf;
>    lua_State *L = conf->L;
>    const uint8_t header = conf->header;
> +  char executable_path[PATH_MAX] = "";

Why do we need this initialization?   ^

>  
>    uint32_t lib_cnt = 0;
>  
> @@ -387,6 +390,31 @@ static int resolve_symbolnames(struct dl_phdr_info *info, size_t info_size,
>    ** Main way: try to open ELF and read SHT_SYMTAB, SHT_STRTAB and SHT_HASH
>    ** sections from it.
>    */
> +
> +  /*
> +  ** The `dl_iterate_phdr` returns an empty string as a name for
> +  ** the executable from which it was called. It is still possible to
> +  ** access its dynamic symbol table, but it is vital for sysprof to obtain
> +  ** the main symbol table for the LuaJIT executable. To do so, we need a
> +  ** valid path to the executable. Since there is no way to obtain the
> +  ** path to a running executable using the C standard library, the only
> +  ** more or less reliable way to do this is by reading the symbolic link
> +  ** from `/proc/self/exe`. Most of the UNIX-based systems have procfs, so
> +  ** it is not a problem.
> +  */
> +  if (*info->dlpi_name == 0) {

Nit: s/0/'\0'/ to emphasize that we check that string contains only NULL
char.

> +    if (-1 != readlink("/proc/self/exe", executable_path, PATH_MAX))

Nit: `readlink("/proc/self/exe", executable_path, PATH_MAX) != -1`

> +      info->dlpi_name = executable_path;
> +    else
> +      /*
> +      ** It is impossible for sysprof to function properly without the
> +      ** LuaJIT's .symtab section present. The assertion below
> +      ** is unlikely to be triggered on any system supported by sysprof,
> +      ** unless someone have deleted the LuaJIT binary right after the
> +      ** start.
> +      */
> +      lua_assert(0);
> +  }
>    if (dump_sht_symtab(info->dlpi_name, buf, L, header, info->dlpi_addr) == 0) {
>      ++conf->cur_lib;
>    }
> -- 
> 2.36.1
> 

-- 
Best regards,
Sergey Kaplun