Re: [PATCH] rpcrdma: don't decref EP if a ESTABLISHED did not happen

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sun, May 05, 2024 at 03:49:10PM +0300, Dan Aloni wrote:
> We found a case where `RDMA_CM_EVENT_DEVICE_REMOVAL` causes a refcount
> underflow.
> 
> The specific scenario that caused this to happen is IB device bonding,
> when bringing down one of the ports, or all ports. The situation is not
> just a print - it also causes a non-recoverable state it is not even
> possible to complete the disconnect and shut it down the mount,
> requiring a reboot, suggesting that tear-down is also incomplete in this
> state.
> 
> The trivial fix seems to work as such - if we did not receive a
> `RDMA_CM_EVENT_ESTABLISHED`, we should not decref the EP, otherwise
> `rpcrdma_xprt_drain` kills the EP prematurely in from the context of
> `rpcrdma_xprt_disconnect`.
> 
> Fixes: 2acc5cae2923 ('xprtrdma: Prevent dereferencing r_xprt->rx_ep after it is freed')

Hi Dan, thanks for the bug report!

In the future, note that the Fixes: tag needs to go down by the
Signed-off-by: tag. But I wonder if 2acc5 introduced this issue --
ep reference counting seems to have been added before this commit.

Second, I wonder if, when bonding interfaces, there's an opportunity
for the verbs consumer to take an additional transport reference.
Cc'ing linux-rdma for input on that issue. Can you show a brief
diagram of when the ep reference count changes when bonding?

Also, I note that the WARNING below happened on a RHEL9 kernel:

   5.14.0-284.11.1.el9_2.x86_64

Can you confirm that this issue reproduces on v6.8 and newer ?


> Example crash:
> 
> rpcrdma: removing device mlx5_3 for 172.21.208.2:20049
> ------------[ cut here ]------------
> refcount_t: underflow; use-after-free.
> WARNING: CPU: 60 PID: 19700 at lib/refcount.c:28 refcount_warn_saturate+0xba/0x110
> Modules linked in: mst_pciconf(OE) nfsv3(OE) nfs_acl(OE) rpcsec_gss_krb5(OE) auth_rpcgss(OE) nfsv4(OE) dns_resolver rpcrdma(OE) nfs(OE) lockd(OE) grace compat_nfs_ssc(OE) snd_seq_dummy snd_hrtimer snd_seq snd_timer snd_seq_device snd soundcore uio_pci_generic uio vfio_pci vfio_pci_core vfio_virqfd vfio_iommu_>
> isst_if_mmio mei isst_if_common i2c_smbus intel_pch_thermal intel_vsec ipmi_msghandler acpi_power_meter xfs libcrc32c mlx5_ib(OE) ib_uverbs(OE) ib_core(OE) sd_mod t10_pi sg mgag200 i2c_algo_bit drm_shmem_helper drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops ahci libahci drm crct10dif_pclmul mlx>
> CPU: 60 PID: 19700 Comm: kworker/u132:4 Kdump: loaded Tainted: G        W  OE    --------  ---  5.14.0-284.11.1.el9_2.x86_64 #1
> Hardware name: Dell Inc. PowerEdge C6520/0TY3YW, BIOS 1.8.2 09/14/2022
> Workqueue: xprtiod xprt_rdma_connect_worker [rpcrdma]
> RIP: 0010:refcount_warn_saturate+0xba/0x110
> Code: 01 01 e8 27 e1 56 00 0f 0b c3 cc cc cc cc 80 3d b8 29 9b 01 00 75 85 48 c7 c7 38 ec 04 93 c6 05 a8 29 9b 01 01 e8 04 e1 56 00 <0f> 0b c3 cc cc cc cc 80 3d 93 29 9b 01 00 0f 85 5e ff ff ff 48 c7
> RSP: 0018:ff34fa4968cafe10 EFLAGS: 00010282
> RAX: 0000000000000000 RBX: ff1210404a15e000 RCX: 0000000000000027
> RDX: ff12103f803998a8 RSI: 0000000000000001 RDI: ff12103f803998a0
> RBP: ff1210404a15e648 R08: 0000000000000000 R09: 00000000ffff7fff
> R10: ff34fa4968cafcb0 R11: ffffffff939e9608 R12: 0000000000000000
> R13: dead000000000122 R14: dead000000000100 R15: ff1210404a15e928
> FS:  0000000000000000(0000) GS:ff12103f80380000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007f170f8a5000 CR3: 00000001c3adc002 CR4: 0000000000771ee0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> PKRU: 55555554
> Call Trace:
> <TASK>
> rpcrdma_ep_put+0x42/0x50 [rpcrdma]
> rpcrdma_xprt_disconnect+0x303/0x3b0 [rpcrdma]
> xprt_rdma_connect_worker+0xc8/0xd0 [rpcrdma]
> process_one_work+0x1e5/0x3c0
> ? rescuer_thread+0x3a0/0x3a0
> worker_thread+0x50/0x3b0
> ? rescuer_thread+0x3a0/0x3a0
> kthread+0xd6/0x100
> ? kthread_complete_and_exit+0x20/0x20
> ret_from_fork+0x1f/0x30
> </TASK>
> 
> Signed-off-by: Dan Aloni <dan.aloni@xxxxxxxxxxxx>
> ---
>  net/sunrpc/xprtrdma/verbs.c     | 5 ++++-
>  net/sunrpc/xprtrdma/xprt_rdma.h | 1 +
>  2 files changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
> index 4f8d7efa469f..19996515da94 100644
> --- a/net/sunrpc/xprtrdma/verbs.c
> +++ b/net/sunrpc/xprtrdma/verbs.c
> @@ -250,6 +250,7 @@ rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
>  		goto disconnected;
>  	case RDMA_CM_EVENT_ESTABLISHED:
>  		rpcrdma_ep_get(ep);
> +               ep->re_connect_ref = true;
>  		ep->re_connect_status = 1;
>  		rpcrdma_update_cm_private(ep, &event->param.conn);
>  		trace_xprtrdma_inline_thresh(ep);
> @@ -272,7 +273,9 @@ rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
>  		ep->re_connect_status = -ECONNABORTED;
>  disconnected:
>  		rpcrdma_force_disconnect(ep);
> -		return rpcrdma_ep_put(ep);
> +		if (ep->re_connect_ref)
> +			return rpcrdma_ep_put(ep);
> +		return 0;
>  	default:
>  		break;
>  	}
> diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
> index da409450dfc0..1553ef69a844 100644
> --- a/net/sunrpc/xprtrdma/xprt_rdma.h
> +++ b/net/sunrpc/xprtrdma/xprt_rdma.h
> @@ -84,6 +84,7 @@ struct rpcrdma_ep {
>  	unsigned int		re_max_inline_recv;
>  	int			re_async_rc;
>  	int			re_connect_status;
> +	bool			re_connect_ref;
>  	atomic_t		re_receiving;
>  	atomic_t		re_force_disconnect;
>  	struct ib_qp_init_attr	re_attr;
> -- 
> 2.39.3
> 

-- 
Chuck Lever




[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Photo]     [Yosemite News]     [Yosemite Photos]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux