lkml.org 
[lkml]   [2009]   [May]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/5] Split wait_noreap_copyout()

* Vitaly Mayatskikh <v.mayatskih@gmail.com> wrote:

> At Mon, 11 May 2009 15:25:50 +0200, Vitaly Mayatskikh wrote:
> >
> > Move getrusage() and put_user() code from wait_noreap_copyout()
> > to copy_wait_opts_to_user(). The same code is spreaded across all
> > wait_task_*() routines, it's better to reuse one copy.
> >
> > Signed-off-by: Vitaly Mayatskikh <v.mayatskih@gmail.com>
> > ---
> > kernel/exit.c | 39 +++++++++++++++++++++++----------------
> > 1 files changed, 23 insertions(+), 16 deletions(-)
> >
> > diff --git a/kernel/exit.c b/kernel/exit.c
> > index 25782da..9546362 100644
> > --- a/kernel/exit.c
> > +++ b/kernel/exit.c
> > @@ -1123,27 +1123,34 @@ static int eligible_child(struct wait_opts *wo, struct task_struct *p)
> > return 1;
> > }
> >
> > -static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
> > - pid_t pid, uid_t uid, int why, int status)
> > +static int copy_wait_opts_to_user(struct wait_opts *wo, struct task_struct *p,
> > + pid_t pid, uid_t uid, int why, int status, int signal)
> > {
> > - struct siginfo __user *infop;
> > + struct siginfo __user *infop = wo->wo_info;
> > int retval = wo->wo_rusage
> > ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
> >
> > + if (!retval && infop) {
> > + retval = put_user(signal, &infop->si_signo);
> ...
> > +static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
> > + pid_t pid, uid_t uid, int why, int status)
> > +{
> > + int retval = copy_wait_opts_to_user(wo, p, pid, uid, why, status, SIGCHLD);
> > put_task_struct(p);
> > - infop = wo->wo_info;
> > - if (!retval)
> > - retval = put_user(SIGCHLD, &infop->si_signo);
> ...
>
> Oleg has pointed me to broken behaviour here. Previously
> wait_noreap_copyout was doing unconditional put_user and was returning
> EFAULT when infop is NULL. Now it uses copy_wait_opts_to_user, which
> checks infop and return NULL in the same case. This change is visible
> from userspace in waitid() function.
>
> There're 2 opportunities how to deal with new behaviour:
>
> 1. Assume wait_task_zombie had a bug previously, and let this patch go.
> 2. Fix copy_wait_opts_to_user to old behaviour by something like:
>
> if (!retval && (infop || WNOWAIT)) {
>
> What's your opinion?

I'd suggest a variant of 2: keep this large-ish patch an equivalent
transformation - i.e. an impact: cleanup type of change.

Then queue up a patch that removes this quirk. Should this change
break any user-space, the quirk can be reinstated promptly, without
affecting the cleanups. It will also be a very clear, easy target to
bisect to - not obscured by clean-up details.

Does this sound good to you?

Ingo


\
 
 \ /
  Last update: 2009-05-20 20:25    [W:0.131 / U:5.176 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site