[PATCH] x86/tools: Use POSIX-compliant syntax for empty regex groups

[PATCH] x86/tools: Use POSIX-compliant syntax for empty regex groups

[John Millikin]

The syntax for POSIX regexes technically doesn't allow empty branches in
conditional match groups. GNU libc accepts them as equivalent to `()',
but other libc implementations (e.g. macOS libc) reject them with a
parse error.

Changing to compliant syntax, `(|_sym)' to `(()|_sym)', lets the `relocs'
tool run successfully when using a non-GNU platform as a build host for
cross-compiling the kernel.

Signed-off-by: John Millikin <john@john-millikin.com>
---

Helper binary to verify behavior for the current platform:

    #include <regex.h>
    #include <stdio.h>
    int main(int argc, char **argv) {
        regex_t compiled;
        int err = regcomp(&compiled, argv[1], REG_EXTENDED|REG_NOSUB);
        if (err != 0) {
            char errbuf[128];
            regerror(err, &compiled, errbuf, sizeof(errbuf));
            printf("regcomp(\"%s\"): %s\n", argv[1], errbuf);
           return 1;
        }
        printf("regcomp(\"%s\"): OK\n", argv[1]);
        return 0;
    }

Output from GNU libc 2.28 and a fussier non-GNU libc:

    debian:~$ ./regcomp '^a(|b)$'
    regcomp("^a(|b)$"): OK
    debian:~$ ./regcomp '^a(()|b)$'
    regcomp("^a(()|b)$"): OK

    darwin:~$ ./regcomp '^a(|b)$'
    regcomp("^a(|b)$"): empty (sub)expression
    darwin:~$ ./regcomp '^a(()|b)$'
    regcomp("^a(()|b)$"): OK

 arch/x86/tools/relocs.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/x86/tools/relocs.c b/arch/x86/tools/relocs.c
index ce7188cbdae5..e6f28616a625 100644
--- a/arch/x86/tools/relocs.c
+++ b/arch/x86/tools/relocs.c
@@ -57,12 +57,12 @@ static const char * const sym_regex_kernel[S_NSYMTYPES] = {
     [S_REL] =
     "^(__init_(begin|end)|"
     "__x86_cpu_dev_(start|end)|"
-    "(__parainstructions|__alt_instructions)(|_end)|"
-    "(__iommu_table|__apicdrivers|__smp_locks)(|_end)|"
+    "(__parainstructions|__alt_instructions)(()|_end)|"
+    "(__iommu_table|__apicdrivers|__smp_locks)(()|_end)|"
     "__(start|end)_pci_.*|"
     "__(start|end)_builtin_fw|"
-    "__(start|stop)___ksymtab(|_gpl|_unused|_unused_gpl|_gpl_future)|"
-    "__(start|stop)___kcrctab(|_gpl|_unused|_unused_gpl|_gpl_future)|"
+    "__(start|stop)___ksymtab(()|_gpl|_unused|_unused_gpl|_gpl_future)|"
+    "__(start|stop)___kcrctab(()|_gpl|_unused|_unused_gpl|_gpl_future)|"
     "__(start|stop)___param|"
     "__(start|stop)___modver|"
     "__(start|stop)___bug_table|"
-- 
2.29.2

Re: [PATCH] x86/tools: Use POSIX-compliant syntax for empty regex groups

[Borislav Petkov]

On Fri, Dec 25, 2020 at 10:30:43PM +0900, John Millikin wrote:
> The syntax for POSIX regexes technically doesn't allow empty branches in
> conditional match groups. GNU libc accepts them as equivalent to `()',
> but other libc implementations (e.g. macOS libc) reject them with a
> parse error.
> 
> Changing to compliant syntax, `(|_sym)' to `(()|_sym)', lets the `relocs'
> tool run successfully when using a non-GNU platform as a build host for
> cross-compiling the kernel.
> 
> Signed-off-by: John Millikin <john@john-millikin.com>
> ---
> 
> Helper binary to verify behavior for the current platform:
> 
>     #include <regex.h>
>     #include <stdio.h>
>     int main(int argc, char **argv) {
>         regex_t compiled;
>         int err = regcomp(&compiled, argv[1], REG_EXTENDED|REG_NOSUB);
>         if (err != 0) {
>             char errbuf[128];
>             regerror(err, &compiled, errbuf, sizeof(errbuf));
>             printf("regcomp(\"%s\"): %s\n", argv[1], errbuf);
>            return 1;
>         }
>         printf("regcomp(\"%s\"): OK\n", argv[1]);
>         return 0;
>     }
> 
> Output from GNU libc 2.28 and a fussier non-GNU libc:
> 
>     debian:~$ ./regcomp '^a(|b)$'
>     regcomp("^a(|b)$"): OK
>     debian:~$ ./regcomp '^a(()|b)$'
>     regcomp("^a(()|b)$"): OK
> 
>     darwin:~$ ./regcomp '^a(|b)$'
>     regcomp("^a(|b)$"): empty (sub)expression
>     darwin:~$ ./regcomp '^a(()|b)$'
>     regcomp("^a(()|b)$"): OK
> 

I guess you can add those to the commit message as they explain in
detail what the situation is.

>  arch/x86/tools/relocs.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/x86/tools/relocs.c b/arch/x86/tools/relocs.c
> index ce7188cbdae5..e6f28616a625 100644
> --- a/arch/x86/tools/relocs.c
> +++ b/arch/x86/tools/relocs.c
> @@ -57,12 +57,12 @@ static const char * const sym_regex_kernel[S_NSYMTYPES] = {
>      [S_REL] =

$ test-apply.sh /tmp/build.01
checking file arch/x86/tools/relocs.c
patch: **** malformed patch at line 67:      [S_REL] =

Don't tell me the apple mangles patches...

:-)

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette