linux-kernel - Re: [PATCH] fs/ceph/mds_client: ignore responses for waiting requests

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 for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <c2f9e0d3-0242-1304-26ea-04f25c3cdee4@redhat.com>
Date:   Wed, 8 Mar 2023 11:42:22 +0800
From:   Xiubo Li <xiubli@...hat.com>
To:     Max Kellermann <max.kellermann@...os.com>, idryomov@...il.com,
        jlayton@...nel.org, ceph-devel@...r.kernel.org
Cc:     linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
        stable@...r.kernel.org
Subject: Re: [PATCH] fs/ceph/mds_client: ignore responses for waiting requests

Hi Max,

Sorry for late.

On 02/03/2023 21:06, Max Kellermann wrote:
> If a request is put on the waiting list, its submission is postponed
> until the session becomes ready (e.g. via `mdsc->waiting_for_map` or
> `session->s_waiting`).  If a `CEPH_MSG_CLIENT_REPLY` happens to be
> received before `CEPH_MSG_MDS_MAP`, the request gets freed, and then
> this assertion fails:

How could this happen ?

Since the req hasn't been submitted yet, how could it receive a reply 
normally ?

>   WARN_ON_ONCE(!list_empty(&req->r_wait));
>
> This occurred on a server after the Ceph MDS connection failed, and a
> corrupt reply packet was received for a waiting request:
>
>   libceph: mds1 (1)10.0.0.10:6801 socket error on write
>   libceph: mds1 (1)10.0.0.10:6801 session reset
>   ceph: mds1 closed our session
>   ceph: mds1 reconnect start
>   ceph: mds1 reconnect success
>   ceph: problem parsing mds trace -5
>   ceph: mds parse_reply err -5
>   ceph: mdsc_handle_reply got corrupt reply mds1(tid:5530222)

It should be a corrupted reply and it lead us to get a incorrect req, 
which hasn't been submitted yet.

BTW, do you have the dump of the corrupted msg by 'ceph_msg_dump(msg)' ?

We can check what have corrupted exactly.

Thanks

- Xiubo

>   [...]
>   ------------[ cut here ]------------
>   WARNING: CPU: 9 PID: 229180 at fs/ceph/mds_client.c:966 ceph_mdsc_release_request+0x17a/0x180
>   Modules linked in:
>   CPU: 9 PID: 229180 Comm: kworker/9:3 Not tainted 6.1.8-cm4all1 #45
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-2 04/01/2014
>   Workqueue: ceph-msgr ceph_con_workfn
>   RIP: 0010:ceph_mdsc_release_request+0x17a/0x180
>   Code: 39 d8 75 26 5b 48 89 ee 48 8b 3d f9 2d 04 02 5d e9 fb 01 c9 ff e8 56 85 ab ff eb 9c 48 8b 7f 58 e8 db 4d ff ff e9 a4 fe ff ff <0f> 0b eb d6 66 90 0f 1f 44 00 00 41 54 48 8d 86 b8 03 00 00 55 4c
>   RSP: 0018:ffffa6f2c0e2bd20 EFLAGS: 00010287
>   RAX: ffff8f58b93687f8 RBX: ffff8f592f6374a8 RCX: 0000000000000aed
>   RDX: 0000000000000ac0 RSI: 0000000000000000 RDI: 0000000000000000
>   RBP: ffff8f592f637148 R08: 0000000000000001 R09: ffffffffa901de00
>   R10: 0000000000000001 R11: ffffd630ad09dfc8 R12: ffff8f58b9368000
>   R13: ffff8f5806b33800 R14: ffff8f58894f6780 R15: 000000000054626e
>   FS:  0000000000000000(0000) GS:ffff8f630f040000(0000) knlGS:0000000000000000
>   CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>   CR2: 00007ffc2926df68 CR3: 0000000218dce002 CR4: 00000000001706e0
>   DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>   DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>   Call Trace:
>    <TASK>
>    mds_dispatch+0xec5/0x1460
>    ? inet_recvmsg+0x4d/0xf0
>    ceph_con_process_message+0x6b/0x80
>    ceph_con_v1_try_read+0xb92/0x1400
>    ceph_con_workfn+0x383/0x4e0
>    process_one_work+0x1da/0x370
>    ? process_one_work+0x370/0x370
>    worker_thread+0x4d/0x3c0
>    ? process_one_work+0x370/0x370
>    kthread+0xbb/0xe0
>    ? kthread_complete_and_exit+0x20/0x20
>    ret_from_fork+0x22/0x30
>    </TASK>
>   ---[ end trace 0000000000000000 ]---
>   ceph: mds1 caps renewed
>
> If we know that a request has not yet been submitted, we should ignore
> all responses for it, just like we ignore responses for unknown TIDs.
>
> To: ceph-devel@...r.kernel.org
> Cc: stable@...r.kernel.org
> Signed-off-by: Max Kellermann <max.kellermann@...os.com>
> ---
>   fs/ceph/mds_client.c | 7 +++++++
>   1 file changed, 7 insertions(+)
>
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index 27a245d959c0..fa74fdb2cbfb 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -3275,6 +3275,13 @@ static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
>   	}
>   	dout("handle_reply %p\n", req);
>   
> +	/* waiting, not yet submitted? */
> +	if (!list_empty(&req->r_wait)) {
> +		pr_err("mdsc_handle_reply on waiting request tid %llu\n", tid);
> +		mutex_unlock(&mdsc->mutex);
> +		goto out;
> +	}
> +
>   	/* correct session? */
>   	if (req->r_session != session) {
>   		pr_err("mdsc_handle_reply got %llu on session mds%d"

-- 
Best Regards,

Xiubo Li (李秀波)

Email: xiubli@...hat.com/xiubli@....com
Slack: @Xiubo Li