netdev - Re: [RFC][PATCH -mm take5 4/7] using symlink for the net

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <a781481a0706131203g405a4407y1212bb397f14e8eb@mail.gmail.com>
Date:	Thu, 14 Jun 2007 00:33:41 +0530
From:	"Satyam Sharma" <satyam.sharma@...il.com>
To:	"Keiichi KII" <k-keiichi@...jp.nec.com>
Cc:	"Matt Mackall" <mpm@...enic.com>,
	"Andrew Morton" <akpm@...ux-foundation.org>,
	"David Miller" <davem@...emloft.net>, linux-kernel@...r.kernel.org,
	netdev@...r.kernel.org
Subject: Re: [RFC][PATCH -mm take5 4/7] using symlink for the net_device

Hi again,

Ok, so sysfs_create_link() would be illegal from inside spin_lock_irqsave(),
and this is why we have to use the dual-list mechanism to react to the net
device rename. This isn't so obvious, a comment at the point where you
declare modify_target_list would be nice? (BTW temporary_list would be
a better name for that, IMO)

> On 6/13/07, Keiichi KII <k-keiichi@...jp.nec.com> wrote:
> > [...]
> > +static DECLARE_MUTEX(netdev_change_sem);
>
> The preferred style these days is to use a DEFINE_MUTEX
> (and the struct mutex primitives) for such locks that are used
> as binary semaphores.
>
> BTW, a comment here to note what this lock protects is required.
> [ You don't really need to give a comment for the target_list_lock
> because it's defined just below the "target_list". It's not equally obvious
> at first glance what is protected by the netdev_change_sem, however. ]

Ok, so reading through the code makes it obvious that this mutex is used
to protect against the following race:

Thread #1                               Thread #2
=========                               =========

[ NETDEV_CHANGENAME notifier ]          [ ioctl(NETCON_REMOVE_TARGET) ]

netconsole_event()
move from target_list to temp list
work on temp list
                                        kobject_unregister()
                                        -> release_target()
                                           -> remove_target()
move back to target_list

Which would mean a deleted/removed target added back => *boom*

But, the race still hasn't been closed properly!

You're taking the mutex only around "work on temp list" which is
insufficient, you need to ensure atomicity inside netconsole_event()
_completely_ like this (renaming netdev_change_sem to
netdev_changename_mtx):

> > +static int netconsole_event(struct notifier_block *this, unsigned long event,
> > +                           void *ptr)
> > +{
> > +       int error = 0;
> > +       unsigned long flags;
> > +       char *old_link_name = NULL, *new_link_name = NULL;
> > +       struct netconsole_target *nt, *tmp;
> > +       struct net_device *dev = ptr;
> > +       LIST_HEAD(modify_target_list);
> > +
> > +       if (event == NETDEV_CHANGENAME) {

mutex_lock(netdev_changename_mtx) here.

> > +               spin_lock_irqsave(&target_list_lock, flags);
> > +               list_for_each_entry_safe(nt, tmp, &target_list, list)
> > +                       if (nt->np.dev == dev)
> > +                               list_move(&nt->list, &modify_target_list);
> > +               spin_unlock_irqrestore(&target_list_lock, flags);

> > +               down(&netdev_change_sem);

This goes away.

> > +               list_for_each_entry(nt, &modify_target_list, list) {
> > + [...]
> > +               }

> > +               up(&netdev_change_sem);

So does this.

> > +               spin_lock_irqsave(&target_list_lock, flags);
> > +               list_for_each_entry_safe(nt, tmp, &modify_target_list, list)
> > +                       list_move(&nt->list, &target_list);
> > +               spin_unlock_irqrestore(&target_list_lock, flags);

mutex_unlock(netdev_changename_mtx) comes here.

> > +       }
> > +
> > +       return NOTIFY_DONE;
> > +}

> @@ -239,12 +240,14 @@ static void remove_target(struct netcons
>  {
>         unsigned long flags;
>
> +       down(&netdev_change_sem);
>         spin_lock_irqsave(&target_list_lock, flags);
>         list_del(&nt->list);
>         if (list_empty(&target_list))
>                 netpoll_cleanup(&nt->np);
>         spin_unlock_irqrestore(&target_list_lock, flags);
>         kfree(nt);
> +       up(&netdev_change_sem);
>  }

As I said earlier, the target_list_lock spin-locking needs to be
pushed out from here to the callers of remove_target.
=> mutex_lock(netdev_changename_mtx) must also be done
by them.

> +static char *make_netdev_class_name(char *netdev_name)
> +{
> +       char *name;
> +
> +       name = kasprintf(GFP_KERNEL, "net:%s", netdev_name);

Why the "net:" prefix in the filename?

> +       if (!name) {
> +               printk(KERN_ERR "netconsole: kmalloc() failed!\n");
> +               return NULL;
> +       }
> +
> +       return name;
> +}

And this doesn't want to be a separate function either.

>  static int setup_target_sysfs(struct netconsole_target *nt)
>  {
> +       int retval = 0;
> +       char *name;
> +
>         kobject_set_name(&nt->obj, "port%d", nt->id);
>         nt->obj.parent = &netconsole_miscdev.this_device->kobj;
>         nt->obj.ktype = &target_ktype;
> -       return kobject_register(&nt->obj);
> +       retval = kobject_register(&nt->obj);
> +       name = make_netdev_class_name(nt->np.dev_name);
> +       if (!name)
> +               return -ENOMEM;

Just call kasprintf() directly, why the obfuscation?

Satyam
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html