Re: [PATCH v2] git-compat-util: work around for access(X_OK) under root

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 24, 2019, 13:39 UTC
Message-ID: <xmqqa7gf1q0y.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <AM6PR02MB4950ED66291B16F35BE61A26EA3C0@AM6PR02MB4950.eurprd02.prod.outlook.com>

"CHIGOT, CLEMENT" <clement.chigot@atos.net> writes:

> On AIX, access(X_OK) may success when run under root even if the

s/success/succeed/;

Also perhaps s/under/as/.

> execution isn't possible. This comes from the POSIX specifications

s/comes from/behaviour is allowed by/; I agree with you that AIX behaviour is suboptimal and I do not think we want to give an impression that POSIX encourages such an illogical behaviour. It merely is permitted.

Show 6 quoted lines

> which say:
> "... for a process with appropriate privileges, an implementation
>  may indicate success for X_OK even if execute permission is not
>  granted to any user."
>
> This behavior can lead hook programs to have their execution refused:

OK.

>  "git commit -m content
>   fatal: cannot exec '.git/hooks/pre-commit': Permission denied"

I am not sure what the double-quotes around these two lines are about. Perhaps drop them, and if you want to clarify which part is your word and which part is quoted material, have blank lines to delineate instead, perhap like:

        ... POSIX says:

            ... for a process with appropriate ...
            ... to any user.

        This behaviour can fail the execution of hooks, like so:

            $ git commit
            fatal: cannot exec '.git/hooks/pre-commit': Permission denied

        Add NEED_ACCESS_ROOT_HANDLER in order to...

> is root.
>
> Signed-off-by: Clément Chigot <clement.chigot@atos.net>
> Message-ID: <AM6PR02MB4950BB0152893633FF95DC99EA230@AM6PR02MB4950.eurprd02.prod.outlook.com>

Drop "Message-Id:" from the footer.

Show 9 quoted lines

> ---
> ...
> diff --git a/compat/access.c b/compat/access.c
> new file mode 100644
> index 0000000000..fcfaefb0c0
> --- /dev/null
> +++ b/compat/access.c
> @@ -0,0 +1,30 @@
> +#define COMPAT_CODE_ACCESS

I am torn between just using COMPAT_CODE like the other one does, or introducing the new symbol like this. If we were to do the latter, perhaps we should give the original one a more specific name as well (e.g. COMPAT_CODE_FILENO or something like that).

Show 13 quoted lines

> +#include "../git-compat-util.h"
> +
> +/* Do the same thing access(2) does, but use the effective uid,
> + * and don't make the mistake of telling root that any file is
> + * executable.  This version uses stat(2).
> + */
> +int git_access(const char *path, int mode)
> +{
> +	struct stat st;
> +
> +	/* do not interfere a normal user */
> +	if (geteuid())
> +		return access(path, mode);

OK.

Show 9 quoted lines

> +	if (stat(path, &st) < 0)
> +		return -1;
> +
> +	/* Root can read or write any file. */
> +	if (!(mode & X_OK))
> +		return 0;
> +
> +	/* Root can execute any file that has any one of the execute
> +	   bits set. */

	/*
	 * Our multi-line comment looks like this,
	 * with opening slash-asterisk and closing
	 * asterisk-slash on their own lines.
	 */

Show 6 quoted lines

> +	if (st.st_mode & (S_IXUSR | S_IXGRP | S_IXOTH))
> +		return 0;
> +
> +	errno = EACCES;
> +	return -1;
> +}

OK.

Show 11 quoted lines

> diff --git a/git-compat-util.h b/git-compat-util.h
> index 31b47932bd..d0cb380522 100644
> --- a/git-compat-util.h
> +++ b/git-compat-util.h
> @@ -1242,6 +1242,13 @@ int git_fileno(FILE *stream);
>  # endif
>  #endif
>  
> +#ifdef NEED_ACCESS_ROOT_HANDLER
> +int git_access(const char *path, int mode);
> +# ifndef COMPAT_CODE_ACCESS

Notice that the fileno thing we are trying to mimick protects us from a system header that defines the macro we are about to define, which is a good practice to prevent compilers from complaining against redefinition. We should imitate it like this here:

   #  ifdef access
   #  undef access
   #  endif

> +#  define access(path, mode) git_access(path, mode)
> +# endif
> +#endif
> +

Other than that, looks good to me.