Re: [PATCH] function dispatch should return if mds session does not exist

[Ilya Dryomov]

On Mon, Oct 14, 2019 at 11:01 AM Yanhu Cao <gmayyyha@xxxxxxxxx> wrote:
>
> we shouldn't call ceph_msg_put, otherwise libceph will pass
> invalid pointer to mm.
>
> kernel panic - not syncing: fatal exception
>     [5452201.213885] ------------[ cut here ]------------
>     [5452201.213889] kernel BUG at mm/slub.c:3901!
>     [5452201.213938] invalid opcode: 0000 [#1] SMP PTI
>     [5452201.213971] CPU: 35 PID: 3037447 Comm: kworker/35:1 Kdump: loaded Not tainted 4.19.15 #1
>     [5452201.214020] Hardware name: HP ProLiant DL380 Gen9/ProLiant DL380 Gen9, BIOS P89 01/22/2018
>     [5452201.214088] Workqueue: ceph-msgr ceph_con_workfn [libceph]
>     [5452201.214129] RIP: 0010:kfree+0x15b/0x170
>     [5452201.214156] Code: 8b 02 f6 c4 80 75 08 49 8b 42 08 a8 01 74 1b 49 8b 02 31 f6 f6 c4 80 74 05 41 0f b6 72 51 5b 5d 41 5c 4c 89 d7 e9 95 03 f9 ff <0f> 0b 48 83 e8 01 e9 01 ff ff ff 49 83 ea 01 e9 e9 fe ff ff 90 0f
>     [5452201.214262] RSP: 0018:ffffb8c3a0607cb0 EFLAGS: 00010246
>     [5452201.214296] RAX: ffffeee840000008 RBX: ffff9130c0000000 RCX: 0000000080200016
>     [5452201.214339] RDX: 00006f0ec0000000 RSI: 0000000000000000 RDI: ffff9130c0000000
>     [5452201.214383] RBP: ffff91107f823970 R08: 0000000000000001 R09: 0000000000000000
>     [5452201.214426] R10: ffffeee840000000 R11: 0000000000000001 R12: ffffffffc076c45d
>     [5452201.214469] R13: ffff91107f823970 R14: ffff91107f8239e0 R15: ffff91107f823900
>     [5452201.214513] FS:  0000000000000000(0000) GS:ffff9110bfbc0000(0000) knlGS:0000000000000000
>     [5452201.214562] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>     [5452201.214598] CR2: 000055993ab29620 CR3: 0000003a1e00a003 CR4: 00000000003606e0
>     [5452201.214641] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>     [5452201.214685] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>     [5452201.214728] Call Trace:
>     [5452201.214759]  ceph_msg_release+0x15d/0x190 [libceph]
>     [5452201.214811]  dispatch+0x66/0xa50 [ceph]
>     [5452201.214846]  try_read+0x7f3/0x11d0 [libceph]
>     [5452201.214878]  ? dequeue_entity+0x37e/0x7e0
>     [5452201.214907]  ? pick_next_task_fair+0x291/0x610
>     [5452201.214937]  ? dequeue_task_fair+0x5d/0x700
>     [5452201.214966]  ? __switch_to+0x8c/0x470
>     [5452201.214999]  ceph_con_workfn+0xa2/0x5b0 [libceph]
>     [5452201.215033]  process_one_work+0x16b/0x370
>     [5452201.215062]  worker_thread+0x49/0x3f0
>     [5452201.215089]  kthread+0xf5/0x130
>     [5452201.215112]  ? max_active_store+0x80/0x80
>     [5452201.215139]  ? kthread_bind+0x10/0x10
>     [5452201.215167]  ret_from_fork+0x1f/0x30
>
> Link: https://tracker.ceph.com/issues/42288
>
> Signed-off-by: Yanhu Cao <gmayyyha@xxxxxxxxx>
> ---
>  fs/ceph/mds_client.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index a8a8f84f3bbf..066358fea347 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -4635,7 +4635,7 @@ static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
>         mutex_lock(&mdsc->mutex);
>         if (__verify_registered_session(mdsc, s) < 0) {
>                 mutex_unlock(&mdsc->mutex);
> -               goto out;
> +               return;
>         }
>         mutex_unlock(&mdsc->mutex);
>
> @@ -4672,7 +4672,6 @@ static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
>                 pr_err("received unknown message type %d %s\n", type,
>                        ceph_msg_type_name(type));
>         }
> -out:
>         ceph_msg_put(msg);
>  }
>

Hi Yanhu,

This doesn't look right to me.  The messenger hands its reference to
the dispatch function, the dispatch function is responsible for putting
it.  Even if the session isn't registered, the message should still be
valid and should still be freed.  The bug is somewhere else...

Thanks,

                Ilya