lkml.org 
[lkml]   [2018]   [Jan]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] staging: lustre: Fix avoid intensive reconnecting for ko2iblnd patch
On Tue, Jan 16, 2018 at 06:02:07PM +0000, Eremin, Dmitry wrote:
> The logic of the original commit 4d99b2581eff ("staging: lustre: avoid intensive reconnecting for ko2iblnd")
> was assumed conditional free of struct kib_conn if the second argument free_conn in function
> kiblnd_destroy_conn(struct kib_conn *conn, bool free_conn) is true. But this hunk of code was dropped
> from original commit. As result the logic works wrong and current code use struct kib_conn after
> free.
>
> > drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
> > 3317 kiblnd_destroy_conn(conn, !peer);
> > ^^^^ Freed always (but should be conditionally)
> > 3318
> > 3319 spin_lock_irqsave(lock, flags);
> > 3320 if (!peer)
> > 3321 continue;
> > 3322
> > 3323 conn->ibc_peer = peer;
> > ^^^^^^^^^^^^^ Use after free
> > 3324 if (peer->ibp_reconnected < KIB_RECONN_HIGH_RACE)
> > 3325 list_add_tail(&conn->ibc_list,
> > ^^^^^^^^^^^^
> > 3326 &kiblnd_data.kib_reconn_list);
> > 3327 else
> > 3328 list_add_tail(&conn->ibc_list,
> > ^^^^^^^^^^^^
> > 3329 &kiblnd_data.kib_reconn_wait);
>
> After attached patch this code will use struct kib_conn only when it was not freed.
>
> Signed-off-by: Dmitry Eremin <Dmitry.Eremin@intel.com>
> ---
> drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
> index 2ebc484..a15a625 100644
> --- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
> +++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
> @@ -890,7 +890,8 @@ void kiblnd_destroy_conn(struct kib_conn *conn, bool free_conn)
> atomic_dec(&net->ibn_nconns);
> }
>
> - kfree(conn);
> + if (free_conn)
> + kfree(conn);
> }
>
> int kiblnd_close_peer_conns_locked(struct kib_peer *peer, int why)

This patch needs a real "Fixes:" tag, right?

Also, as this was from 4.6, it should go to the stable trees, right?
Can you resend this with that info, and then send a follow-on patch to
fix this up the way I recommended so that no one is confused in the
future?

thanks,

greg k-h

\
 
 \ /
  Last update: 2018-01-22 11:39    [W:0.041 / U:12.180 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site