Re: dbghelp: Fix stabs_pts_read_type_def() for when typename is NULL

[Eric Pouech]

Andrew Talbot a écrit :
> Changelog:
>     dbghelp: Fix stabs_pts_read_type_def() for when typename is NULL.
>
> diff --git a/dlls/dbghelp/stabs.c b/dlls/dbghelp/stabs.c
> index d550633..3c69eec 100644
> --- a/dlls/dbghelp/stabs.c
> +++ b/dlls/dbghelp/stabs.c
> @@ -896,10 +896,10 @@ static int stabs_pts_read_type_def(struct ParseTypedefData* ptd, const char* typ
>                       * removed
>                       */
>                      l1 = strlen(udt->hash_elt.name);
> -
>                l2 = strlen(typename);
> -
>                if (l1 > l2 || strcmp(udt->hash_elt.name, typename + l2 - l1))
> -
>                l2 = typename ? strlen(typename) : 0;
> -
>                if (l1 > l2 || (typename && strcmp(udt->hash_elt.name, typename + l2 - l1)))
>                    ERR("Forward declaration name mismatch %s <> %s\n",
> -
>                        udt->hash_elt.name, typename);
> -
>                        udt->hash_elt.name, typename ? typename : "(null)");
>                new_dt = &udt->symt;
>            }
>            PTS_ABORTIF(ptd, stabs_pts_read_aggregate(ptd, udt) == -1);
>
> @@ -1003,7 +1003,7 @@ static int stabs_pts_read_type_def(struct ParseTypedefData* ptd, const char* typ
>  *stabs_find_ref(filenr1, subnr1) = *ret_dt = new_dt;
>
> - TRACE("Adding (%ld,%ld) %s\n", filenr1, subnr1, typename);
> -
>
> TRACE("Adding (%ld,%ld) %s\n", filenr1, subnr1, typename ? typename : "(null)");
>
> return 0;
>
> }

looks like a bit strange to me that you get a null typename here
can you send me the DLL/.so file on which you get the seg fault
A+

dbghelp: Fix stabs_pts_read_type_def() for when typename is NULL

[Andrew Talbot]

Eric Pouech wrote:
> looks like a bit strange to me that you get a null typename here
> can you send me the DLL/.so file on which you get the seg fault
> A+

No known segfaults; I'm just doing static analysis. But
stabs_pts_read_type_def() is called several times within stabs.c passing
NULL as a typename argument: e.g., line 494:

PTS_ABORTIF(ptd, stabs_pts_read_type_def(ptd, NULL, &ref) == -1);

And, with case 'u' (i.e., *ptd->ptr == 'u') and udt != NULL, it would be
passed to strlen(). I didn't establish whether this will ever occur with
typename == NULL, I just sought to make the function more bullet proof.

dbghelp: Fix stabs_pts_read_type_def() for when typename is NULL

[Eric Pouech]

Andrew Talbot a écrit :
> Eric Pouech wrote:
> > looks like a bit strange to me that you get a null typename here
> > can you send me the DLL/.so file on which you get the seg fault
> > A+
>
> No known segfaults; I'm just doing static analysis. But
> stabs_pts_read_type_def() is called several times within stabs.c passing
> NULL as a typename argument: e.g., line 494:
> PTS_ABORTIF(ptd, stabs_pts_read_type_def(ptd, NULL, &ref) == -1);
>
> And, with case 'u' (i.e., *ptd->ptr == 'u') and udt != NULL, it would be
> passed to strlen(). I didn't establish whether this will ever occur with
> typename == NULL, I just sought to make the function more bullet proof.

this code path shouldn't need to be protected
however the last trace should be protected (and debugstr_a is a better 
choice than testing for a null string)
A+

dbghelp: Fix stabs_pts_read_type_def() for when typename is NULL

[Andrew Talbot]

Eric Pouech wrote:
> however the last trace should be protected (and debugstr_a is a better
> choice than testing for a null string)
> A+

Thanks, Eric. I've sent a replacement patch (with a revised title).