Re: select() support for NR_OPEN>256
Andi Kleen (andi@mlm.extern.lrz-muenchen.de)
Wed, 05 Feb 1997 21:18:45 +0100
In message <199702051923.UAA08601@sunsite.ms.mff.cuni.cz>you write:
>> - * We do a VERIFY_WRITE here even though we are only reading this time:
>> - * we'll write to it eventually..
>> - *
>> * Use "int" accesses to let user-mode fd_set's be int-aligned.
>> */
>> static int __get_fd_set(unsigned long nr, int * fs_pointer, int * fdset)
>> @@ -209,16 +205,14 @@
>> /* round up nr to nearest "int" */
>> nr = (nr + 8*sizeof(int)-1) / (8*sizeof(int));
>> if (fs_pointer) {
>> - int error = verify_area(VERIFY_WRITE,fs_pointer,nr*sizeof(int))
>;
>> - if (!error) {
>> - while (nr) {
>> - get_user(*fdset, fs_pointer);
>> - nr--;
>> - fs_pointer++;
>> - fdset++;
>> - }
>> + while (nr) {
>> + if (get_user(*fdset, fs_pointer))
>> + return -EFAULT;
>> + nr--;
>> + fs_pointer++;
>> + fdset++;
>> }
>> - return error;
>> + return 0;
>
>Isn't it better to leave there the verify_area and just use __get_user
>instead of get_user? This will generate much nicer code (at least on the
>sparc)...
>
>Like:
>
> int error = verify_area(VERIFY_WRITE,fs_pointer,nr*sizeof(int));
> if (!error) {
> while (nr) {
> if (__get_user(*fdset, fs_pointer))
> return -EFAULT;
> nr--;
> fs_pointer++;
> fdset++;
> }
> }
> return error;
Yup, you're right. My original code replaced the whole function
with a copy_from_user() but I got some strange errors (I suspected
a gcc problem) so I reverted back to this simple-minded implementation.
The copy_from_user() macro also expanded to unneeded byte-access code.
I'll change my code to keep the verify_area().
-Andi