This is the mail archive of the libc-help@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

access call not calling faccessat correctly


I was seeing a failure in the chmod_1.f90 test that is part of gcc's validation testsuite and am pretty sure it is caused by some bad code in glibc.

In sysdeps/unix/sysv/linux/access.c, a call to "access()" is implemented by either calling the access syscall or the faccessat syscall, depending on whether __NR_access is #defined.  If __NR_access is not #defined and the faccessat syscall is used, it is only called with 3 arguments - no value for the flags argument is passed.  This causes EINVAL to be returned in some cases, depending on whatever happens to be in the register that would be used to pass that argument.

I think the change is simple:

diff --git a/sysdeps/unix/sysv/linux/access.c b/sysdeps/unix/sysv/linux/access.c
index 524594ac79..706f242adb 100644
--- a/sysdeps/unix/sysv/linux/access.c
+++ b/sysdeps/unix/sysv/linux/access.c
@@ -26,7 +26,7 @@ __access (const char *file, int type)
#ifdef __NR_access
   return INLINE_SYSCALL_CALL (access, file, type);
#else
-  return INLINE_SYSCALL_CALL (faccessat, AT_FDCWD, file, type);
+  return INLINE_SYSCALL_CALL (faccessat, AT_FDCWD, file, type, 0);
#endif
}
weak_alias (__access, access)

I haven't filed a bug yet, but can do so.

  Michael


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]