setpriority typo

setpriority typo

[Justin Cormack]

[-- Attachment #1: Type: text/plain, Size: 241 bytes --]

cat ./src/misc/setpriority.c
#include <sys/resource.h>
#include "syscall.h"

int setpriority(int which, id_t who, int prio)
{
return syscall(SYS_getpriority, which, who, prio);
}

                                    ^ should be set!

Justin

[-- Attachment #2: Type: text/html, Size: 520 bytes --]

Re: setpriority typo

[Rich Felker]

On Mon, Apr 01, 2013 at 02:54:18PM +0100, Justin Cormack wrote:
> cat ./src/misc/setpriority.c
> #include <sys/resource.h>
> #include "syscall.h"
> 
> int setpriority(int which, id_t who, int prio)
> {
> return syscall(SYS_getpriority, which, who, prio);
> }
> 
>                                     ^ should be set!

This is definitely a bug, but I need to look at whether changing it
will fix the problem. getpriority is doing some transformation on the
prio value and I don't remember why, but it seems likely setpriority
might need to do the same.

Rich

Re: setpriority typo

[Justin Cormack]

[-- Attachment #1: Type: text/plain, Size: 1228 bytes --]

On Mon, Apr 1, 2013 at 3:05 PM, Rich Felker <dalias@aerifal.cx> wrote:

> On Mon, Apr 01, 2013 at 02:54:18PM +0100, Justin Cormack wrote:
> > cat ./src/misc/setpriority.c
> > #include <sys/resource.h>
> > #include "syscall.h"
> >
> > int setpriority(int which, id_t who, int prio)
> > {
> > return syscall(SYS_getpriority, which, who, prio);
> > }
> >
> >                                     ^ should be set!
>
> This is definitely a bug, but I need to look at whether changing it
> will fix the problem. getpriority is doing some transformation on the
> prio value and I don't remember why, but it seems likely setpriority
> might need to do the same.
>
> Rich
>

From the man page:

Within the kernel, nice values are actually represented using the
corresponding range 40..1
       (since negative numbers are error codes) and these are the values
employed by the setpriority() and  getpriority()  system
       calls... The glibc wrapper functions for these system calls handle
the translations between the user-land and kernel repre‐
       sentations of the nice value according to the formula unice = 20 -
knice.

So to be compatible with glibc there should be an adjustment.

Justin

[-- Attachment #2: Type: text/html, Size: 2195 bytes --]

Re: setpriority typo

[Rich Felker]

On Mon, Apr 01, 2013 at 03:30:06PM +0100, Justin Cormack wrote:
> On Mon, Apr 1, 2013 at 3:05 PM, Rich Felker <dalias@aerifal.cx> wrote:
> 
> > On Mon, Apr 01, 2013 at 02:54:18PM +0100, Justin Cormack wrote:
> > > cat ./src/misc/setpriority.c
> > > #include <sys/resource.h>
> > > #include "syscall.h"
> > >
> > > int setpriority(int which, id_t who, int prio)
> > > {
> > > return syscall(SYS_getpriority, which, who, prio);
> > > }
> > >
> > >                                     ^ should be set!
> >
> > This is definitely a bug, but I need to look at whether changing it
> > will fix the problem. getpriority is doing some transformation on the
> > prio value and I don't remember why, but it seems likely setpriority
> > might need to do the same.
> >
> > Rich
> >
> 
> >From the man page:
> 
> Within the kernel, nice values are actually represented using the
> corresponding range 40..1
>        (since negative numbers are error codes) and these are the values
> employed by the setpriority() and  getpriority()  system
>        calls... The glibc wrapper functions for these system calls handle
> the translations between the user-land and kernel repre‐
>        sentations of the nice value according to the formula unice = 20 -
> knice.
> 
> So to be compatible with glibc there should be an adjustment.

Reading the kernel source, apparently the adjustment is just required
with getpriority to avoid it returning negative values on success.
setpriority takes the unmodified value as an argument.

BTW, these interfaces are badly nonconforming on Linux; they're
affecting threads, not processes... I'm not clear what the conforming
behavior would be on a system like Linux that does not support system
scheduling scope, but unfortunately it's probably not what anybody
wants or expects...

Anyway I'll fix the bug for now, then we can discuss whether these
interfaces need changing.

Rich