[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1486570497.7066.6.camel@haakon3.risingtidesystems.com>
Date: Wed, 08 Feb 2017 08:14:57 -0800
From: "Nicholas A. Bellinger" <nab@...ux-iscsi.org>
To: Christoph Hellwig <hch@...radead.org>
Cc: target-devel <target-devel@...r.kernel.org>,
linux-scsi <linux-scsi@...r.kernel.org>,
lkml <linux-kernel@...r.kernel.org>,
Rob Millner <rlm@...erainc.com>
Subject: Re: [PATCH 4/5] target: Fix multi-session dynamic se_node_acl
double free OOPs
On Tue, 2017-02-07 at 19:46 -0800, Nicholas A. Bellinger wrote:
> On Tue, 2017-02-07 at 15:12 -0800, Christoph Hellwig wrote:
> > And the real patch after compile fixing it is here of course:
> >
>
> Getting rid of the extra se_node_acl->acl_free_comp seems to make sense
> here..
>
> The only potential issue is if returning from configfs syscall
> rmdir /sys/kernel/config/target/$FABRIC/$WWN/$TPGT/acls/$INITIATOR/
> before se_node_acl memory is released has implications when the
> underlying ../$WWN/$TPGT/ is removed immediately after.
>
> In any event, I'll verify this patch with the original test case and use
> it instead if everything looks OK.
Ok, I remember why se_node_acl->acl_free_comp is needed now..
The issue is when se_node_acl is being explicitly removed from configfs
while se_sessions associated with se_node_acl are still active.
The se_node_acl->acl_group configfs_group_operations->drop_item()
callback must block until all associated se_sessions have done their
target_put_nacl() and completed se_node_acl->acl_free_comp, otherwise
there is nothing preventing parent config_groups of se_node_acl from
being removed while the se_sessions are still active.
That said, dropping the unnecessary list_del_init() usage, but otherwise
keeping the patch as-is.
Powered by blists - more mailing lists