2015-07-13 13:18:46

by Kirill A. Shutemov

[permalink] [raw]
Subject: mmap()ed AF_NETLINK: lockdep and sleep-in-atomic warnings

Hi,

This simple test-case trigers few locking asserts in kernel:

#define _GNU_SOURCE
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <linux/netlink.h>

#define SOL_NETLINK 270

int main(int argc, char **argv)
{
unsigned int block_size = 16 * 4096;
struct nl_mmap_req req = {
.nm_block_size = block_size,
.nm_block_nr = 64,
.nm_frame_size = 16384,
.nm_frame_nr = 64 * block_size / 16384,
};
unsigned int ring_size;
int fd;

fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
if (setsockopt(fd, SOL_NETLINK, NETLINK_RX_RING, &req, sizeof(req)) < 0)
exit(1);
if (setsockopt(fd, SOL_NETLINK, NETLINK_TX_RING, &req, sizeof(req)) < 0)
exit(1);

ring_size = req.nm_block_nr * req.nm_block_size;
mmap(NULL, 2 * ring_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
return 0;
}

+++ exited with 0 +++
[ 2.500126] BUG: sleeping function called from invalid context at /home/kas/git/public/linux-mm/kernel/locking/mutex.c:616
[ 2.501328] in_atomic(): 1, irqs_disabled(): 0, pid: 1, name: init
[ 2.501997] 3 locks held by init/1:
[ 2.502380] #0: (reboot_mutex){+.+...}, at: [<ffffffff81080959>] SyS_reboot+0xa9/0x220
[ 2.503328] #1: ((reboot_notifier_list).rwsem){.+.+..}, at: [<ffffffff8107f379>] __blocking_notifier_call_chain+0x39/0x70
[ 2.504659] #2: (rcu_callback){......}, at: [<ffffffff810d32e0>] rcu_do_batch.isra.49+0x160/0x10c0
[ 2.505724] Preemption disabled at:[<ffffffff8145365f>] __delay+0xf/0x20
[ 2.506443]
[ 2.506612] CPU: 1 PID: 1 Comm: init Not tainted 4.1.0-00009-gbddf4c4818e0 #253
[ 2.507378] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS Debian-1.8.2-1 04/01/2014
[ 2.508386] ffff88017b3d8000 ffff88027bc03c38 ffffffff81929ceb 0000000000000102
[ 2.509233] 0000000000000000 ffff88027bc03c68 ffffffff81085a9d 0000000000000002
[ 2.510057] ffffffff81ca2a20 0000000000000268 0000000000000000 ffff88027bc03c98
[ 2.510882] Call Trace:
[ 2.511146] <IRQ> [<ffffffff81929ceb>] dump_stack+0x4f/0x7b
[ 2.511763] [<ffffffff81085a9d>] ___might_sleep+0x16d/0x270
[ 2.512476] [<ffffffff81085bed>] __might_sleep+0x4d/0x90
[ 2.513071] [<ffffffff8192e96f>] mutex_lock_nested+0x2f/0x430
[ 2.513683] [<ffffffff81932fed>] ? _raw_spin_unlock_irqrestore+0x5d/0x80
[ 2.514385] [<ffffffff81464143>] ? __this_cpu_preempt_check+0x13/0x20
[ 2.515066] [<ffffffff8182fc3d>] netlink_set_ring+0x1ed/0x350
[ 2.515694] [<ffffffff8182e000>] ? netlink_undo_bind+0x70/0x70
[ 2.516411] [<ffffffff8182fe20>] netlink_sock_destruct+0x80/0x150
[ 2.517070] [<ffffffff817e484d>] __sk_free+0x1d/0x160
[ 2.517607] [<ffffffff817e49a9>] sk_free+0x19/0x20
[ 2.518118] [<ffffffff8182e020>] deferred_put_nlk_sk+0x20/0x30
[ 2.518735] [<ffffffff810d391c>] rcu_do_batch.isra.49+0x79c/0x10c0
[ 2.519386] [<ffffffff810d32e0>] ? rcu_do_batch.isra.49+0x160/0x10c0
[ 2.520101] [<ffffffff810d787b>] rcu_process_callbacks+0xdb/0x6d0
[ 2.520790] [<ffffffff8105dd52>] __do_softirq+0x152/0x630
[ 2.521370] [<ffffffff8105e3be>] irq_exit+0x8e/0xb0
[ 2.521895] [<ffffffff81936366>] smp_apic_timer_interrupt+0x46/0x60
[ 2.522558] [<ffffffff8145365f>] ? __delay+0xf/0x20
[ 2.523079] [<ffffffff81934a00>] apic_timer_interrupt+0x70/0x80
[ 2.523705] <EOI> [<ffffffff8145365f>] ? __delay+0xf/0x20
[ 2.524366] [<ffffffff810b26ab>] ? in_lock_functions+0x1b/0x20
[ 2.524995] [<ffffffff8108ab81>] get_parent_ip+0x11/0x50
[ 2.525562] [<ffffffff8108ad1f>] preempt_count_sub+0x9f/0xf0
[ 2.526179] [<ffffffff81453778>] delay_tsc+0x68/0xc0
[ 2.526706] [<ffffffff8145365f>] __delay+0xf/0x20
[ 2.527207] [<ffffffff8145369a>] __const_udelay+0x2a/0x30
[ 2.527781] [<ffffffff8172d05a>] md_notify_reboot+0xea/0x100
[ 2.528489] [<ffffffff8107f379>] ? __blocking_notifier_call_chain+0x39/0x70
[ 2.529236] [<ffffffff8107efc6>] notifier_call_chain+0x66/0x90
[ 2.529856] [<ffffffff8107f391>] __blocking_notifier_call_chain+0x51/0x70
[ 2.530570] [<ffffffff810ae8c6>] ? __lock_acquire+0x606/0xf50
[ 2.531172] [<ffffffff8107f3c6>] blocking_notifier_call_chain+0x16/0x20
[ 2.531869] [<ffffffff8108061d>] kernel_restart_prepare+0x1d/0x40
[ 2.532593] [<ffffffff810806e6>] kernel_restart+0x16/0x60
[ 2.533183] [<ffffffff81080a07>] SyS_reboot+0x157/0x220
[ 2.533738] [<ffffffff81010778>] ? __restore_xstate_sig+0xf8/0x720
[ 2.534390] [<ffffffff81464127>] ? debug_smp_processor_id+0x17/0x20
[ 2.535051] [<ffffffff810a836e>] ? put_lock_stats.isra.19+0xe/0x30
[ 2.535707] [<ffffffff81933040>] ? _raw_spin_unlock_irq+0x30/0x60
[ 2.536446] [<ffffffff8108ad2b>] ? preempt_count_sub+0xab/0xf0
[ 2.537112] [<ffffffff81933daa>] ? syscall_return+0x11/0x54
[ 2.537709] [<ffffffff81464143>] ? __this_cpu_preempt_check+0x13/0x20
[ 2.538399] [<ffffffff810abf73>] ? trace_hardirqs_on_caller+0xf3/0x240
[ 2.539094] [<ffffffff81454b98>] ? trace_hardirqs_on_thunk+0x17/0x19
[ 2.539764] [<ffffffff81933b97>] system_call_fastpath+0x12/0x6f
[ 2.540523]
[ 2.540695] =================================
[ 2.541161] [ INFO: inconsistent lock state ]
[ 2.541618] 4.1.0-00009-gbddf4c4818e0 #253 Not tainted
[ 2.542154] ---------------------------------
[ 2.542610] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[ 2.543236] init/1 [HC0[0]:SC1[1]:HE1:SE0] takes:
[ 2.543729] (&nlk->pg_vec_lock){+.?.+.}, at: [<ffffffff8182fc3d>] netlink_set_ring+0x1ed/0x350
[ 2.544503] {SOFTIRQ-ON-W} state was registered at:
[ 2.544503] [<ffffffff810aedce>] __lock_acquire+0xb0e/0xf50
[ 2.544503] [<ffffffff810afb22>] lock_acquire+0xd2/0x2b0
[ 2.544503] [<ffffffff8192e9b1>] mutex_lock_nested+0x71/0x430
[ 2.544503] [<ffffffff8182fc3d>] netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff81832010>] netlink_setsockopt+0x1b0/0x300
[ 2.544503] [<ffffffff817e0133>] SyS_setsockopt+0x63/0xc0
[ 2.544503] [<ffffffff81933cbc>] tracesys_phase2+0x84/0x89
[ 2.544503] irq event stamp: 1159264
[ 2.544503] hardirqs last enabled at (1159264): [<ffffffff810c3495>] vprintk_emit+0x2c5/0x600
[ 2.544503] hardirqs last disabled at (1159263): [<ffffffff810c321e>] vprintk_emit+0x4e/0x600
[ 2.544503] softirqs last enabled at (1159014): [<ffffffff8105de2e>] __do_softirq+0x22e/0x630
[ 2.544503] softirqs last disabled at (1159017): [<ffffffff8105e3be>] irq_exit+0x8e/0xb0
[ 2.544503]
[ 2.544503] other info that might help us debug this:
[ 2.544503] Possible unsafe locking scenario:
[ 2.544503]
[ 2.544503] CPU0
[ 2.544503] ----
[ 2.544503] lock(&nlk->pg_vec_lock);
[ 2.544503] <Interrupt>
[ 2.544503] lock(&nlk->pg_vec_lock);
[ 2.544503]
[ 2.544503] *** DEADLOCK ***
[ 2.544503]
[ 2.544503] 3 locks held by init/1:
[ 2.544503] #0: (reboot_mutex){+.+...}, at: [<ffffffff81080959>] SyS_reboot+0xa9/0x220
[ 2.544503] #1: ((reboot_notifier_list).rwsem){.+.+..}, at: [<ffffffff8107f379>] __blocking_notifier_call_chain+0x39/0x70
[ 2.544503] #2: (rcu_callback){......}, at: [<ffffffff810d32e0>] rcu_do_batch.isra.49+0x160/0x10c0
[ 2.544503]
[ 2.544503] stack backtrace:
[ 2.544503] CPU: 1 PID: 1 Comm: init Not tainted 4.1.0-00009-gbddf4c4818e0 #253
[ 2.544503] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS Debian-1.8.2-1 04/01/2014
[ 2.544503] ffff88017b3d8000 ffff88027bc03af8 ffffffff81929ceb 0000000000000103
[ 2.544503] ffffffff82ad2ed0 ffff88027bc03b38 ffffffff819279f8 0000000000000001
[ 2.544503] ffff88017b3d8810 0000000000000004 0000000000000004 0000000000000006
[ 2.544503] Call Trace:
[ 2.544503] <IRQ> [<ffffffff81929ceb>] dump_stack+0x4f/0x7b
[ 2.544503] [<ffffffff819279f8>] print_usage_bug.part.35+0x281/0x290
[ 2.544503] [<ffffffff810ab89c>] mark_lock+0x22c/0x760
[ 2.544503] [<ffffffff810a97b0>] ? check_usage_backwards+0x150/0x150
[ 2.544503] [<ffffffff810aed77>] __lock_acquire+0xab7/0xf50
[ 2.544503] [<ffffffff810afb22>] lock_acquire+0xd2/0x2b0
[ 2.544503] [<ffffffff8182fc3d>] ? netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff81085a9d>] ? ___might_sleep+0x16d/0x270
[ 2.544503] [<ffffffff8182fc3d>] ? netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff8192e9b1>] mutex_lock_nested+0x71/0x430
[ 2.544503] [<ffffffff8182fc3d>] ? netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff8182fc3d>] ? netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff81464143>] ? __this_cpu_preempt_check+0x13/0x20
[ 2.544503] [<ffffffff8182fc3d>] netlink_set_ring+0x1ed/0x350
[ 2.544503] [<ffffffff8182e000>] ? netlink_undo_bind+0x70/0x70
[ 2.544503] [<ffffffff8182fe20>] netlink_sock_destruct+0x80/0x150
[ 2.544503] [<ffffffff817e484d>] __sk_free+0x1d/0x160
[ 2.544503] [<ffffffff817e49a9>] sk_free+0x19/0x20
[ 2.544503] [<ffffffff8182e020>] deferred_put_nlk_sk+0x20/0x30
[ 2.544503] [<ffffffff810d391c>] rcu_do_batch.isra.49+0x79c/0x10c0
[ 2.544503] [<ffffffff810d32e0>] ? rcu_do_batch.isra.49+0x160/0x10c0
[ 2.544503] [<ffffffff810d787b>] rcu_process_callbacks+0xdb/0x6d0
[ 2.544503] [<ffffffff8105dd52>] __do_softirq+0x152/0x630
[ 2.544503] [<ffffffff8105e3be>] irq_exit+0x8e/0xb0
[ 2.544503] [<ffffffff81936366>] smp_apic_timer_interrupt+0x46/0x60
[ 2.544503] [<ffffffff8145365f>] ? __delay+0xf/0x20
[ 2.544503] [<ffffffff81934a00>] apic_timer_interrupt+0x70/0x80
[ 2.544503] <EOI> [<ffffffff8145365f>] ? __delay+0xf/0x20
[ 2.544503] [<ffffffff810b26ab>] ? in_lock_functions+0x1b/0x20
[ 2.544503] [<ffffffff8108ab81>] get_parent_ip+0x11/0x50
[ 2.544503] [<ffffffff8108ad1f>] preempt_count_sub+0x9f/0xf0
[ 2.544503] [<ffffffff81453778>] delay_tsc+0x68/0xc0
[ 2.544503] [<ffffffff8145365f>] __delay+0xf/0x20
[ 2.544503] [<ffffffff8145369a>] __const_udelay+0x2a/0x30
[ 2.544503] [<ffffffff8172d05a>] md_notify_reboot+0xea/0x100
[ 2.544503] [<ffffffff8107f379>] ? __blocking_notifier_call_chain+0x39/0x70
[ 2.544503] [<ffffffff8107efc6>] notifier_call_chain+0x66/0x90
[ 2.544503] [<ffffffff8107f391>] __blocking_notifier_call_chain+0x51/0x70
[ 2.544503] [<ffffffff810ae8c6>] ? __lock_acquire+0x606/0xf50
[ 2.544503] [<ffffffff8107f3c6>] blocking_notifier_call_chain+0x16/0x20
[ 2.544503] [<ffffffff8108061d>] kernel_restart_prepare+0x1d/0x40
[ 2.544503] [<ffffffff810806e6>] kernel_restart+0x16/0x60
[ 2.544503] [<ffffffff81080a07>] SyS_reboot+0x157/0x220
[ 2.544503] [<ffffffff81010778>] ? __restore_xstate_sig+0xf8/0x720
[ 2.544503] [<ffffffff81464127>] ? debug_smp_processor_id+0x17/0x20
[ 2.544503] [<ffffffff810a836e>] ? put_lock_stats.isra.19+0xe/0x30
[ 2.544503] [<ffffffff81933040>] ? _raw_spin_unlock_irq+0x30/0x60
[ 2.544503] [<ffffffff8108ad2b>] ? preempt_count_sub+0xab/0xf0
[ 2.544503] [<ffffffff81933daa>] ? syscall_return+0x11/0x54
[ 2.544503] [<ffffffff81464143>] ? __this_cpu_preempt_check+0x13/0x20
[ 2.544503] [<ffffffff810abf73>] ? trace_hardirqs_on_caller+0xf3/0x240
[ 2.544503] [<ffffffff81454b98>] ? trace_hardirqs_on_thunk+0x17/0x19
[ 2.544503] [<ffffffff81933b97>] system_call_fastpath+0x12/0x6f

--
Kirill A. Shutemov


2015-07-14 05:11:29

by Cong Wang

[permalink] [raw]
Subject: Re: mmap()ed AF_NETLINK: lockdep and sleep-in-atomic warnings

On Mon, Jul 13, 2015 at 6:18 AM, Kirill A. Shutemov
<[email protected]> wrote:
> Hi,
>
> This simple test-case trigers few locking asserts in kernel:
>
> #define _GNU_SOURCE
> #include <stdlib.h>
> #include <stdio.h>
> #include <string.h>
> #include <sys/mman.h>
> #include <sys/socket.h>
> #include <sys/types.h>
> #include <linux/netlink.h>
>
> #define SOL_NETLINK 270
>
> int main(int argc, char **argv)
> {
> unsigned int block_size = 16 * 4096;
> struct nl_mmap_req req = {
> .nm_block_size = block_size,
> .nm_block_nr = 64,
> .nm_frame_size = 16384,
> .nm_frame_nr = 64 * block_size / 16384,
> };
> unsigned int ring_size;
> int fd;
>
> fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
> if (setsockopt(fd, SOL_NETLINK, NETLINK_RX_RING, &req, sizeof(req)) < 0)
> exit(1);
> if (setsockopt(fd, SOL_NETLINK, NETLINK_TX_RING, &req, sizeof(req)) < 0)
> exit(1);
>
> ring_size = req.nm_block_nr * req.nm_block_size;
> mmap(NULL, 2 * ring_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
> return 0;
> }
>
> +++ exited with 0 +++
> [ 2.500126] BUG: sleeping function called from invalid context at /home/kas/git/public/linux-mm/kernel/locking/mutex.c:616
> [ 2.501328] in_atomic(): 1, irqs_disabled(): 0, pid: 1, name: init
> [ 2.501997] 3 locks held by init/1:
> [ 2.502380] #0: (reboot_mutex){+.+...}, at: [<ffffffff81080959>] SyS_reboot+0xa9/0x220
> [ 2.503328] #1: ((reboot_notifier_list).rwsem){.+.+..}, at: [<ffffffff8107f379>] __blocking_notifier_call_chain+0x39/0x70
> [ 2.504659] #2: (rcu_callback){......}, at: [<ffffffff810d32e0>] rcu_do_batch.isra.49+0x160/0x10c0
> [ 2.505724] Preemption disabled at:[<ffffffff8145365f>] __delay+0xf/0x20
> [ 2.506443]
> [ 2.506612] CPU: 1 PID: 1 Comm: init Not tainted 4.1.0-00009-gbddf4c4818e0 #253
> [ 2.507378] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS Debian-1.8.2-1 04/01/2014
> [ 2.508386] ffff88017b3d8000 ffff88027bc03c38 ffffffff81929ceb 0000000000000102
> [ 2.509233] 0000000000000000 ffff88027bc03c68 ffffffff81085a9d 0000000000000002
> [ 2.510057] ffffffff81ca2a20 0000000000000268 0000000000000000 ffff88027bc03c98
> [ 2.510882] Call Trace:
> [ 2.511146] <IRQ> [<ffffffff81929ceb>] dump_stack+0x4f/0x7b
> [ 2.511763] [<ffffffff81085a9d>] ___might_sleep+0x16d/0x270
> [ 2.512476] [<ffffffff81085bed>] __might_sleep+0x4d/0x90
> [ 2.513071] [<ffffffff8192e96f>] mutex_lock_nested+0x2f/0x430
> [ 2.513683] [<ffffffff81932fed>] ? _raw_spin_unlock_irqrestore+0x5d/0x80
> [ 2.514385] [<ffffffff81464143>] ? __this_cpu_preempt_check+0x13/0x20
> [ 2.515066] [<ffffffff8182fc3d>] netlink_set_ring+0x1ed/0x350
> [ 2.515694] [<ffffffff8182e000>] ? netlink_undo_bind+0x70/0x70
> [ 2.516411] [<ffffffff8182fe20>] netlink_sock_destruct+0x80/0x150
> [ 2.517070] [<ffffffff817e484d>] __sk_free+0x1d/0x160
> [ 2.517607] [<ffffffff817e49a9>] sk_free+0x19/0x20
> [ 2.518118] [<ffffffff8182e020>] deferred_put_nlk_sk+0x20/0x30
> [ 2.518735] [<ffffffff810d391c>] rcu_do_batch.isra.49+0x79c/0x10c0

Caused by:

commit 21e4902aea80ef35afc00ee8d2abdea4f519b7f7
Author: Thomas Graf <[email protected]>
Date: Fri Jan 2 23:00:22 2015 +0100

netlink: Lockless lookup with RCU grace period in socket release

Defers the release of the socket reference using call_rcu() to
allow using an RCU read-side protected call to rhashtable_lookup()

This restores behaviour and performance gains as previously
introduced by e341694 ("netlink: Convert netlink_lookup() to use
RCU protected hash table") without the side effect of severely
delayed socket destruction.

Signed-off-by: Thomas Graf <[email protected]>
Signed-off-by: David S. Miller <[email protected]>


We can't hold mutex lock in a rcu callback, perhaps we could
defer the mmap ring cleanup to a workqueue.

2015-07-14 09:38:30

by Thomas Graf

[permalink] [raw]
Subject: Re: mmap()ed AF_NETLINK: lockdep and sleep-in-atomic warnings

On 07/13/15 at 10:11pm, Cong Wang wrote:
> Caused by:
>
> commit 21e4902aea80ef35afc00ee8d2abdea4f519b7f7
> Author: Thomas Graf <[email protected]>
> Date: Fri Jan 2 23:00:22 2015 +0100
>
> netlink: Lockless lookup with RCU grace period in socket release
>
> Defers the release of the socket reference using call_rcu() to
> allow using an RCU read-side protected call to rhashtable_lookup()
>
> This restores behaviour and performance gains as previously
> introduced by e341694 ("netlink: Convert netlink_lookup() to use
> RCU protected hash table") without the side effect of severely
> delayed socket destruction.
>
> Signed-off-by: Thomas Graf <[email protected]>
> Signed-off-by: David S. Miller <[email protected]>
>
>
> We can't hold mutex lock in a rcu callback, perhaps we could
> defer the mmap ring cleanup to a workqueue.

The socket should be dead at this point. It might be simpler to
add a netlink_release_ring() function which doesn't require
locking at all. It would also get rid of all the special handling
for closing vs. non-closing.

2015-07-14 09:50:44

by Florian Westphal

[permalink] [raw]
Subject: Re: mmap()ed AF_NETLINK: lockdep and sleep-in-atomic warnings

Thomas Graf <[email protected]> wrote:
> On 07/13/15 at 10:11pm, Cong Wang wrote:
> > Caused by:
> >
> > commit 21e4902aea80ef35afc00ee8d2abdea4f519b7f7
> > Author: Thomas Graf <[email protected]>
> > Date: Fri Jan 2 23:00:22 2015 +0100
> >
> > netlink: Lockless lookup with RCU grace period in socket release
> >
> > Defers the release of the socket reference using call_rcu() to
> > allow using an RCU read-side protected call to rhashtable_lookup()
> >
> > This restores behaviour and performance gains as previously
> > introduced by e341694 ("netlink: Convert netlink_lookup() to use
> > RCU protected hash table") without the side effect of severely
> > delayed socket destruction.
> >
> > Signed-off-by: Thomas Graf <[email protected]>
> > Signed-off-by: David S. Miller <[email protected]>
> >
> >
> > We can't hold mutex lock in a rcu callback, perhaps we could
> > defer the mmap ring cleanup to a workqueue.
>
> The socket should be dead at this point. It might be simpler to
> add a netlink_release_ring() function which doesn't require
> locking at all. It would also get rid of all the special handling
> for closing vs. non-closing.

Do you know of any users of mmap'd netlink?
Support for this was never added to libmnl; given the fact that netlink
mmap doesn't have any advantages vs. socket interface i'd just revert it
(or mark CONFIG_NETLINK_MMAP broken -- although I don't see
how the unconditional copy on tx could ever be avoided).

2015-07-14 18:40:31

by Stephen Hemminger

[permalink] [raw]
Subject: Re: mmap()ed AF_NETLINK: lockdep and sleep-in-atomic warnings

On Tue, 14 Jul 2015 11:50:37 +0200
Florian Westphal <[email protected]> wrote:

> Thomas Graf <[email protected]> wrote:
> > On 07/13/15 at 10:11pm, Cong Wang wrote:
> > > Caused by:
> > >
> > > commit 21e4902aea80ef35afc00ee8d2abdea4f519b7f7
> > > Author: Thomas Graf <[email protected]>
> > > Date: Fri Jan 2 23:00:22 2015 +0100
> > >
> > > netlink: Lockless lookup with RCU grace period in socket release
> > >
> > > Defers the release of the socket reference using call_rcu() to
> > > allow using an RCU read-side protected call to rhashtable_lookup()
> > >
> > > This restores behaviour and performance gains as previously
> > > introduced by e341694 ("netlink: Convert netlink_lookup() to use
> > > RCU protected hash table") without the side effect of severely
> > > delayed socket destruction.
> > >
> > > Signed-off-by: Thomas Graf <[email protected]>
> > > Signed-off-by: David S. Miller <[email protected]>
> > >
> > >
> > > We can't hold mutex lock in a rcu callback, perhaps we could
> > > defer the mmap ring cleanup to a workqueue.
> >
> > The socket should be dead at this point. It might be simpler to
> > add a netlink_release_ring() function which doesn't require
> > locking at all. It would also get rid of all the special handling
> > for closing vs. non-closing.
>
> Do you know of any users of mmap'd netlink?
> Support for this was never added to libmnl; given the fact that netlink
> mmap doesn't have any advantages vs. socket interface i'd just revert it
> (or mark CONFIG_NETLINK_MMAP broken -- although I don't see
> how the unconditional copy on tx could ever be avoided).

We use netlink mmap because it gives slightly better performance
when processing a route flap.