Re: [PATCH bpf v2 14/15] selftests/bpf: Check BPFTOOL env var in detect_bpftool_path()

[Alexei Starovoitov]

On Wed, Feb 18, 2026 at 5:13 PM Ihor Solodrai <ihor.solodrai@xxxxxxxxx> wrote:
>
> On 2/18/26 10:17 AM, Ihor Solodrai wrote:
> > On 2/18/26 9:38 AM, Alexei Starovoitov wrote:
> >> On Tue, Feb 17, 2026 at 4:31 PM Ihor Solodrai <ihor.solodrai@xxxxxxxxx> wrote:
> >>>
> >>> [...]
> >>>
> >>>         /* Check default bpftool location (will work if we are running the
> >>>          * default flavor of test_progs)
> >>>          */
> >>>         snprintf(tmp, BPFTOOL_PATH_MAX_LEN, "./%s", BPFTOOL_DEFAULT_PATH);
> >>>         if (access(tmp, X_OK) == 0) {
> >>> -               strncpy(buffer, tmp, BPFTOOL_PATH_MAX_LEN);
> >>> +               snprintf(buffer, BPFTOOL_PATH_MAX_LEN, "%s", tmp);
> >>
> >> I guess it's ok for user space, but
> >>
> >> git log --oneline|grep "snprintf with strscpy"
> >> ad789a85b163 mm/cma: replace snprintf with strscpy in cma_new_area
> >> 674fb053e95d sparc: vio: Replace snprintf with strscpy in vio_create_one
> >> 2dfc417414c6 genirq/proc: Replace snprintf with strscpy in register_handler_proc
> >> f46ebb910989 block: Replace snprintf with strscpy in check_partition
> >> b66215e7b780 media: verisilicon: replace snprintf with strscpy+strlcat
> >> a86028f8e3ee staging: most: sound: replace snprintf with strscpy
> >>
> >> and many others...
> >> So.. should we introduce strscpy() in selftests/bpf ?
> >
> > I guess we can, but:
> >
> > $ grep -r 'snprintf(' --include="*.[ch]" tools/testing/selftests/bpf/ | wc -l
> > 238
> >
> > The reason to prefer strscpy() is speed, right?
> > Can we use kernel implementation in userspace directly?
> >
> > In tools/include I only see this:
> >
> >       #define strscpy strcpy
> >
>
> I think we can add a simple implementation to tools/lib/string.c
> See a diff below: essentially a copy from arch/s390/boot/string.c [1]
>
> I suppose it's faster than snprintf(). However I am not sure about
> going through all 238 usages, as that would touch many files. But at
> least we can make actual strscpy available in selftests.
>
> I can add this in v3. Alexei, wdyt?
>
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/s390/boot/string.c?h=v6.19#n32
>
> diff --git a/tools/include/linux/string.h b/tools/include/linux/string.h
> index 51ad3cf4fa82..ca183a0e846a 100644
> --- a/tools/include/linux/string.h
> +++ b/tools/include/linux/string.h
> @@ -3,6 +3,7 @@
>  #define _TOOLS_LINUX_STRING_H_
>
>  #include <linux/types.h>       /* for size_t */
> +#include <sys/types.h>         /* for ssize_t */
>  #include <string.h>
>
>  void *memdup(const void *src, size_t len);
> @@ -12,7 +13,7 @@ void argv_free(char **argv);
>
>  int strtobool(const char *s, bool *res);
>
> -#define strscpy strcpy
> +ssize_t strscpy(char *dest, const char *src, size_t count);
>
>  /*
>   * glibc based builds needs the extern while uClibc doesn't.
> diff --git a/tools/lib/string.c b/tools/lib/string.c
> index 3126d2cff716..54c9163fa2b0 100644
> --- a/tools/lib/string.c
> +++ b/tools/lib/string.c
> @@ -239,3 +239,17 @@ void *memchr_inv(const void *start, int c, size_t bytes)
>
>         return check_bytes8(start, value, bytes % 8);
>  }
> +
> +ssize_t strscpy(char *dst, const char *src, size_t count)
> +{
> +       size_t len;
> +
> +       if (count == 0)
> +               return -E2BIG;
> +
> +       len = strnlen(src, count - 1);
> +       memcpy(dst, src, len);
> +       dst[len] = '\0';
> +
> +       return src[len] ? -E2BIG : len;
> +}

Hmm. it looks a bit like strlcpy() which is also not recommended.
See Documentation/process/deprecated.rst
that explains the reasons why strscpy() is the only "approved" version.

Also tools/lib/ is too scrutinized.
Let's take sized_strscpy() from lib/string.c
drop all optimizations like word a time, page boundary tricks, etc
and only keep essentially this:
        while (count > 1) {
                char c;

                c = src[res];
                dest[res] = c;
                if (!c)
                        return res;
                res++;
                count--;
        }

        /* Force NUL-termination. */
        dest[res] = '\0';

        /* Return E2BIG if the source didn't stop */
        return src[res] ? -E2BIG : res;

and add it somewhere in selftests/bpf..

and only use it in these couple cases where we need to fix strncpy.
Further snprintf() cleanup can be dealt with later.

I mainly want to avoid more churn when people will inevitably
start sending patches to replace snprintf with strscpy.