On Tue, 2016-11-08 at 06:53 -0500, Jeff Layton wrote:
> On Mon, 2016-11-07 at 22:42 -0700, Ross Zwisler wrote:
> >
> > I've got a virtual machine that has some NFS mounts, and with a newly compiled
> > kernel based on v4.9-rc3 I see the following warning/info message:
> >
> > [ 42.750181] ===============================
> > [ 42.750192] [ INFO: suspicious RCU usage. ]
> > [ 42.750203] 4.9.0-rc3-00002-g7b6e7de #3 Not tainted
> > [ 42.750213] -------------------------------
> > [ 42.750225] net/sunrpc/clnt.c:2773 suspicious rcu_dereference_check() usage!
> > [ 42.750235]
> > [ 42.750235] other info that might help us debug this:
> > [ 42.750235]
> > [ 42.750246]
> > [ 42.750246] rcu_scheduler_active = 1, debug_locks = 0
> > [ 42.750257] 1 lock held by mount.nfs4/6440:
> > [ 42.750278] #0:
> > [ 42.750299] (
> > [ 42.750319] &(&nn->nfs_client_lock)->rlock
> > [ 42.750340] ){+.+...}
> > [ 42.750362] , at:
> > [ 42.750372] [<ffffffff813012b5>] nfs_get_client+0x105/0x5e0
> > [ 42.750383]
> > [ 42.750383] stack backtrace:
> > [ 42.750394] CPU: 0 PID: 6440 Comm: mount.nfs4 Not tainted 4.9.0-rc3-00002-g7b6e7de #3
> > [ 42.750406] Hardware name: Intel Corporation PURLEY/PURLEY, BIOS PLYDCRB1.MBH.0096.D23.1608240105 08/24/2016
> > [ 42.750429] ffffc9000092fa68 ffffffff8150730f ffff88014ec8da40 0000000000000001
> > [ 42.750452] ffffc9000092fa98 ffffffff810bc3f7 ffff880150b0b228 ffff88015068dbb0
> > [ 42.750475] ffffc9000092fb38 ffff88014fc99180 ffffc9000092fac0 ffffffff81b243e5
> > [ 42.750486] Call Trace:
> > [ 42.750498] [<ffffffff8150730f>] dump_stack+0x67/0x98
> > [ 42.750511] [<ffffffff810bc3f7>] lockdep_rcu_suspicious+0xe7/0x120
> > [ 42.750524] [<ffffffff81b243e5>] rpc_clnt_xprt_switch_has_addr+0x115/0x150
> > [ 42.750536] [<ffffffff813013f4>] nfs_get_client+0x244/0x5e0
> > [ 42.750549] [<ffffffff813012ac>] ? nfs_get_client+0xfc/0x5e0
> > [ 42.750561] [<ffffffff813568f8>] nfs4_set_client+0x98/0x130
> > [ 42.750574] [<ffffffff8135872e>] nfs4_create_server+0x13e/0x390
> > [ 42.750588] [<ffffffff8134cd0e>] nfs4_remote_mount+0x2e/0x60
> > [ 42.750600] [<ffffffff811f3a29>] mount_fs+0x39/0x170
> > [ 42.750614] [<ffffffff81214a0b>] vfs_kern_mount+0x6b/0x150
> > [ 42.750626] [<ffffffff8134cbec>] ? nfs_do_root_mount+0x3c/0xc0
> > [ 42.750639] [<ffffffff8134cc36>] nfs_do_root_mount+0x86/0xc0
> > [ 42.750652] [<ffffffff8134d014>] nfs4_try_mount+0x44/0xc0
> > [ 42.750664] [<ffffffff81302097>] ? get_nfs_version+0x27/0x90
> > [ 42.750677] [<ffffffff81310f8c>] nfs_fs_mount+0x4ac/0xd80
> > [ 42.750689] [<ffffffff810bb938>] ? lockdep_init_map+0x88/0x1f0
> > [ 42.750701] [<ffffffff81311ac0>] ? nfs_clone_super+0x130/0x130
> > [ 42.750713] [<ffffffff8130f300>] ? param_set_portnr+0x70/0x70
> > [ 42.750726] [<ffffffff811f3a29>] mount_fs+0x39/0x170
> > [ 42.750740] [<ffffffff81214a0b>] vfs_kern_mount+0x6b/0x150
> > [ 42.750752] [<ffffffff812176f1>] do_mount+0x1f1/0xd10
> > [ 42.750765] [<ffffffff81217441>] ? copy_mount_options+0xa1/0x140
> > [ 42.750777] [<ffffffff81218543>] SyS_mount+0x83/0xd0
> > [ 42.750790] [<ffffffff81002abc>] do_syscall_64+0x5c/0x130
> > [ 42.750802] [<ffffffff81c479a4>] entry_SYSCALL64_slow_path+0x25/0x25
> >
> > This rcu_dereference_check() was introduced by the following commit:
> >
> > commit 39e5d2df959dd4aea81fa33d765d2a5cc67a0512
> > Author: Andy Adamson <andros@xxxxxxxxxx>
> > Date: Fri Sep 9 09:22:25 2016 -0400
> >
> > SUNRPC search xprt switch for sockaddr
> >
> > Signed-off-by: Andy Adamson <andros@xxxxxxxxxx>
> > Signed-off-by: Anna Schumaker <Anna.Schumaker@xxxxxxxxxx>
> >
> > Thanks,
> > - Ross
>
> Thanks Ross,
>
> ----------------------8<----------------------
> bool rpc_clnt_xprt_switch_has_addr(struct rpc_clnt *clnt,
> const struct sockaddr *sap)
> {
> struct rpc_xprt_switch *xps;
> bool ret;
>
> xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch);
>
> rcu_read_lock();
> ret = rpc_xprt_switch_has_addr(xps, sap);
> rcu_read_unlock();
> return ret;
> }
> ----------------------8<----------------------
>
> Looks like the simple fix is to just move that rcu_dereference call
> inside the rcu_read_lock there.
>
Hmm...that said though, there are some other suspicious accesses
of xpi_xpswitch. Looks like these are called without the rcu_read_lock
clearly being held:
rpc_clnt_xprt_switch_add_xprt
rpc_clnt_xprt_switch_put
...though it's possible I missed something there.
--
Jeff Layton <jlayton@xxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
