klibc - Aug 2004 - [PATCH] fcntl takes unsigned long as third arg

The syscall defines only a signed long, but the kernel expects unsigned.
No idea if this is intentional or if it breaks anything. I just had to
look why gcc complained about pointer to int conversions.

diff -p -purN klibc-0.168/klibc/SYSCALLS.def
klibc-0.168.fcntl/klibc/SYSCALLS.def
--- klibc-0.168/klibc/SYSCALLS.def      2004-08-03 03:04:30.000000000 +0200
+++ klibc-0.168.fcntl/klibc/SYSCALLS.def        2004-08-25 15:32:30.000000000
+0200
@@ -120,7 +120,7 @@ int close(int)
 <32> int _llseek::__llseek(int, unsigned long, unsigned long, off_t *,
int)
 int dup(int)
 int dup2(int, int)
-int fcntl(int, int, long)
+int fcntl(int, int, unsigned long)
 int ioctl(int, int, void *)
 int flock(int, int)
 int _newselect,select::select(int, fd_set *, fd_set *, fd_set *, struct timeval
*)


-- 
USB is for mice, FireWire is for men!

sUse lINUX ag, n?RNBERG

Olaf Hering wrote:
> The syscall defines only a signed long, but the kernel expects unsigned.
> No idea if this is intentional or if it breaks anything. I just had to
> look why gcc complained about pointer to int conversions.
> 
> diff -p -purN klibc-0.168/klibc/SYSCALLS.def
klibc-0.168.fcntl/klibc/SYSCALLS.def
> --- klibc-0.168/klibc/SYSCALLS.def      2004-08-03 03:04:30.000000000 +0200
> +++ klibc-0.168.fcntl/klibc/SYSCALLS.def        2004-08-25
15:32:30.000000000 +0200
> @@ -120,7 +120,7 @@ int close(int)
>  <32> int _llseek::__llseek(int, unsigned long, unsigned long, off_t
*, int)
>  int dup(int)
>  int dup2(int, int)
> -int fcntl(int, int, long)
> +int fcntl(int, int, unsigned long)
>  int ioctl(int, int, void *)
>  int flock(int, int)
>  int _newselect,select::select(int, fd_set *, fd_set *, fd_set *, struct
timeval *)
> 
Hm.  POSIX wants signed long; glibc has int fcntl(int, int, ...), since 
it needs to take either a signed long or a struct flock.  Probably we 
need to do something similar to the hack we did for open(2).

	-hpa

Olaf Hering wrote:
> The syscall defines only a signed long, but the kernel expects unsigned.
> No idea if this is intentional or if it breaks anything. I just had to
> look why gcc complained about pointer to int conversions.
> 
> diff -p -purN klibc-0.168/klibc/SYSCALLS.def
klibc-0.168.fcntl/klibc/SYSCALLS.def
> --- klibc-0.168/klibc/SYSCALLS.def      2004-08-03 03:04:30.000000000 +0200
> +++ klibc-0.168.fcntl/klibc/SYSCALLS.def        2004-08-25
15:32:30.000000000 +0200
> @@ -120,7 +120,7 @@ int close(int)
>  <32> int _llseek::__llseek(int, unsigned long, unsigned long, off_t
*, int)
>  int dup(int)
>  int dup2(int, int)
> -int fcntl(int, int, long)
> +int fcntl(int, int, unsigned long)
>  int ioctl(int, int, void *)
>  int flock(int, int)
>  int _newselect,select::select(int, fd_set *, fd_set *, fd_set *, struct
timeval *)
Bleah.  fcntl() is a can of worms.  Not only is it varadic, but it has a 
whole different set of operations for "struct flock64".

I really hate multiplexed system calls.  I *really, really* hate 
multiplexed API calls.

	-hpa

On Wed, Aug 25, 2004 at 12:16:49PM -0700, H. Peter Anvin
wrote:
> Olaf Hering wrote:
> >-int fcntl(int, int, long)
> >+int fcntl(int, int, unsigned long)
> 
> Hm.  POSIX wants signed long;
Just FYI, the Single Unix Specification (v2 and v3) says:

  int fcntl(int fildes, int cmd, ...);

http://www.unix.org/version3/online.html
                                                  -Dave Dodge

Dave Dodge wrote:
> 
> Just FYI, the Single Unix Specification (v2 and v3) says:
> 
>   int fcntl(int fildes, int cmd, ...);
> 
Yup, a multiplexed, varadic syscall, visible to apps.  GAG BARF PUKE 
VOMIT :(

I implemented that yesterday (in 0.170); it should now also call 
fcntl64() with F_GETLK64 et al where appropriate, too.

	-hpa