Re: [PATCH 4.14 022/284] SUNRPC: avoid race between mod_timer() and del_timer_sync()

From: Eugeniu Rosca
Date: Wed Sep 07 2022 - 10:28:10 EST

Hello all,

On Mo, Apr 18, 2022 at 02:10:03 +0200, Greg Kroah-Hartman wrote:
> From: NeilBrown <neilb@xxxxxxx>
>
> commit 3848e96edf4788f772d83990022fa7023a233d83 upstream.
>
> xprt_destory() claims XPRT_LOCKED and then calls del_timer_sync().
> Both xprt_unlock_connect() and xprt_release() call
> ->release_xprt()
> which drops XPRT_LOCKED and *then* xprt_schedule_autodisconnect()
> which calls mod_timer().
>
> This may result in mod_timer() being called *after* del_timer_sync().
> When this happens, the timer may fire long after the xprt has been freed,
> and run_timer_softirq() will probably crash.
>
> The pairing of ->release_xprt() and xprt_schedule_autodisconnect() is
> always called under ->transport_lock. So if we take ->transport_lock to
> call del_timer_sync(), we can be sure that mod_timer() will run first
> (if it runs at all).
>
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: NeilBrown <neilb@xxxxxxx>
> Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> ---
> net/sunrpc/xprt.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> --- a/net/sunrpc/xprt.c
> +++ b/net/sunrpc/xprt.c
> @@ -1520,7 +1520,14 @@ static void xprt_destroy(struct rpc_xprt
> */
> wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
>
> + /*
> + * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
> + * is cleared. We use ->transport_lock to ensure the mod_timer()
> + * can only run *before* del_time_sync(), never after.
> + */
> + spin_lock(&xprt->transport_lock);
> del_timer_sync(&xprt->timer);
> + spin_unlock(&xprt->transport_lock);
>
> /*
> * Destroy sockets etc from the system workqueue so they can

This commit introduced the following warning [1][2] on the
v4.9, v4.14 and v4.19 stable trees, when booting Renesas H3ULCB
(and potentially other HW targets) from NFS.

Once in a while, the issue leads to the real freeze of the target.

The culprit commits have been identified via git bisecting (see [1]).

Additionally, it has been determined that what's missing for fixing
the issue on the stable trees are the two v5.3-rc1 mainline commits:

4f8943f8088348 ("SUNRPC: Replace direct task wakeups from softirq context")
b5e924191f8723 ("SUNRPC: Remove the bh-safe lock requirement on xprt->transport_lock")

However, attempting to port them to the stable trees leads to
significant amount of conflicts. Any idea if the culprit commit(s)
should better be reverted?

[1] https://gist.github.com/erosca/7b5f1dadd4172b38461478d38c1040b8
[2] Excerpt from [1]

[ 7.548549] ================================
[ 7.552827] WARNING: inconsistent lock state
[ 7.557112] 4.14.292 #35 Not tainted
[ 7.560694] --------------------------------
[ 7.564973] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[ 7.570994] swapper/0/0 [HC0[0]:SC1[3]:HE1:SE0] takes:
[ 7.576142] (&(&xprt->transport_lock)->rlock){+.?.}, at: [<ffff2000095a6a48>] xs_tcp_state_change+0x740/0xb24
[ 7.586196] {SOFTIRQ-ON-W} state was registered at:
[ 7.591093] lock_acquire+0x724/0x790
[ 7.594857] _raw_spin_lock+0xec/0x12c
[ 7.598706] xprt_destroy+0xc8/0x214
[ 7.602379] xprt_put+0x34/0x40
[ 7.605618] rpc_new_client+0x1a8/0x8e8
[ 7.609552] rpc_create_xprt+0x124/0x300
[ 7.613573] rpc_create+0x234/0x410
[ 7.617162] nfs_create_rpc_client+0x33c/0x38c
[ 7.621709] nfs4_alloc_client+0x8f4/0xb5c
[ 7.625904] nfs_get_client+0x10d4/0x10e8
[ 7.630013] nfs4_set_client+0x1b0/0x254
[ 7.634034] nfs4_create_server+0x4c0/0x92c
[ 7.638317] nfs4_remote_mount+0x74/0xac
[ 7.642339] mount_fs+0x80/0x27c
[ 7.645666] vfs_kern_mount+0xe0/0x3b4
[ 7.649514] nfs_do_root_mount+0x8c/0xc8
[ 7.653534] nfs4_try_mount+0xdc/0x120
[ 7.657382] nfs_fs_mount+0x1b6c/0x2038
[ 7.661315] mount_fs+0x80/0x27c
[ 7.664639] vfs_kern_mount+0xe0/0x3b4
[ 7.668488] do_mount+0x1324/0x1c90
[ 7.672074] SyS_mount+0xc0/0xd0
[ 7.675403] mount_root+0xe4/0x1c0
[ 7.678903] prepare_namespace+0x174/0x184
[ 7.683098] kernel_init_freeable+0x5b4/0x674
[ 7.687556] kernel_init+0x18/0x138
[ 7.691144] ret_from_fork+0x10/0x18
[ 7.694814] irq event stamp: 24322
[ 7.698228] hardirqs last enabled at (24322): [<ffff200009679f8c>] _raw_spin_unlock_irqrestore+0x100/0x108
[ 7.707990] hardirqs last disabled at (24321): [<ffff200009679a38>] _raw_spin_lock_irqsave+0x30/0x138
[ 7.717233] softirqs last enabled at (24244): [<ffff2000080e79d8>] _local_bh_enable+0x78/0x84
[ 7.725865] softirqs last disabled at (24245): [<ffff2000080e9ed4>] irq_exit+0x350/0x4e0
[ 7.733969]
[ 7.733969] other info that might help us debug this:
[ 7.740511] Possible unsafe locking scenario:
[ 7.740511]
[ 7.746442] CPU0
[ 7.748893] ----
[ 7.751343] lock(&(&xprt->transport_lock)->rlock);
[ 7.756323] <Interrupt>
[ 7.758947] lock(&(&xprt->transport_lock)->rlock);
[ 7.764100]
[ 7.764100] *** DEADLOCK ***
[ 7.764100]
[ 7.770038] 4 locks held by swapper/0/0:
[ 7.773967] #0: (rcu_read_lock){....}, at: [<ffff200009318d1c>] netif_receive_skb_internal+0x1b0/0x99c
[ 7.783487] #1: (rcu_read_lock){....}, at: [<ffff2000093e0b10>] ip_local_deliver_finish+0x1e4/0xae0
[ 7.792743] #2: (slock-AF_INET-RPC/1){+.-.}, at: [<ffff2000094757cc>] tcp_v4_rcv+0x1a04/0x2694
[ 7.801567] #3: (k-clock-AF_INET){++.-}, at: [<ffff2000095a633c>] xs_tcp_state_change+0x34/0xb24
[ 7.810559]
[ 7.810559] stack backtrace:
[ 7.814931] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.14.292 #35
[ 7.821123] Hardware name: Renesas H3ULCB Kingfisher board based on r8a7795 ES2.0+ (DT)
[ 7.829142] Call trace:
[ 7.831599] dump_backtrace+0x0/0x320
[ 7.835272] show_stack+0x24/0x30
[ 7.838598] dump_stack+0x150/0x1b8
[ 7.842097] print_usage_bug.part.23+0x5c4/0x724
[ 7.846725] mark_lock+0x940/0x113c
[ 7.850223] __lock_acquire+0x9a8/0x2e70
[ 7.854156] lock_acquire+0x724/0x790
[ 7.857827] _raw_spin_lock+0xec/0x12c
[ 7.861586] xs_tcp_state_change+0x740/0xb24
[ 7.865869] tcp_rcv_state_process+0x1330/0x3a34
[ 7.870498] tcp_v4_do_rcv+0x9c8/0x9dc
[ 7.874256] tcp_v4_rcv+0x1c5c/0x2694
[ 7.877930] ip_local_deliver_finish+0x770/0xae0
[ 7.882560] ip_local_deliver+0x1c0/0x528
[ 7.886581] ip_rcv_finish+0x770/0x1024
[ 7.890427] ip_rcv+0xa50/0xe5c
[ 7.893579] __netif_receive_skb_core+0x1adc/0x2164
[ 7.898470] __netif_receive_skb+0x1e0/0x1e8
[ 7.902752] netif_receive_skb_internal+0x6c8/0x99c
[ 7.907642] napi_gro_receive+0x79c/0x7dc
[ 7.911666] ravb_poll+0xc98/0x1594
[ 7.915165] napi_poll+0x260/0xb6c
[ 7.918577] net_rx_action+0x2fc/0x668
[ 7.922336] __do_softirq+0xec8/0x1620
[ 7.926095] irq_exit+0x350/0x4e0
[ 7.929421] __handle_domain_irq+0x124/0x1c0
[ 7.933701] gic_handle_irq+0x70/0xb0
[ 7.937372] el1_irq+0xb4/0x140
[ 7.940523] arch_cpu_idle+0x17c/0x7fc
[ 7.944282] default_idle_call+0x74/0x8c
[ 7.948216] do_idle+0x250/0x344
[ 7.951453] cpu_startup_entry+0x28/0x38
[ 7.955387] rest_init+0x5f0/0x604
[ 7.958799] start_kernel+0x5dc/0x60c

Best Regards,
Eugeniu Rosca