Skip to contentSkip to conversation

Revert "vrf: Remove unnecessary RCU-bh critical section"

PATCH In review [email protected] 3 participants 7 messages Sep 25, 2024 – Sep 27, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Revert "vrf: Remove unnecessary RCU-bh critical section"

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  packet_rcv+0xa33/0x1320
  __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
  __netif_receive_skb_list_core+0x2c9/0x890
  netif_receive_skb_list_internal+0x610/0xcc0
  napi_complete_done+0x1c0/0x7c0
  igb_poll+0x1dbb/0x57e0 [igb]
  __napi_poll.constprop.0+0x99/0x430
  net_rx_action+0x8e7/0xe10
  handle_softirqs+0x1b7/0x800
  __irq_exit_rcu+0x91/0xc0
  irq_exit_rcu+0x5/0x10
  [snip]

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0
       ----
  lock(rlock-AF_PACKET);
  <Interrupt>
    lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
 <TASK>
 dump_stack_lvl+0x73/0xa0
 mark_lock+0x102e/0x16b0
 __lock_acquire+0x9ae/0x6170
 lock_acquire+0x19a/0x4f0
 _raw_spin_lock+0x27/0x40
 tpacket_rcv+0x863/0x3b30
 dev_queue_xmit_nit+0x709/0xa40
 vrf_finish_direct+0x26e/0x340 [vrf]
 vrf_l3_out+0x5f4/0xe80 [vrf]
 __ip_local_out+0x51e/0x7a0

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

---

v2: Edit patch description.

 drivers/net/vrf.c | 2 ++
 net/core/dev.c    | 1 +
 2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
 		eth_zero_addr(eth->h_dest);
 		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
 		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
 		skb_pull(skb, ETH_HLEN);
 	}
net/core/dev.c +1 -0
index cd479f5f22f6..566e69a38eed 100644
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
 /*
  *	Support routine. Sends outgoing frames to any network
  *	taps currently in use.
+ *	BH must be disabled before calling this.
  */
 
 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Signature · 1 line
2.42.0
maintainer
#1 lore raw
Quoted 98 lines · On Wed, Sep 25, 2024 at 11:52:16AM -0700, [email protected] wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  packet_rcv+0xa33/0x1320
  __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
  __netif_receive_skb_list_core+0x2c9/0x890
  netif_receive_skb_list_internal+0x610/0xcc0
  napi_complete_done+0x1c0/0x7c0
  igb_poll+0x1dbb/0x57e0 [igb]
  __napi_poll.constprop.0+0x99/0x430
  net_rx_action+0x8e7/0xe10
  handle_softirqs+0x1b7/0x800
  __irq_exit_rcu+0x91/0xc0
  irq_exit_rcu+0x5/0x10
  [snip]

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0
       ----
  lock(rlock-AF_PACKET);
  <Interrupt>
    lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
 <TASK>
 dump_stack_lvl+0x73/0xa0
 mark_lock+0x102e/0x16b0
 __lock_acquire+0x9ae/0x6170
 lock_acquire+0x19a/0x4f0
 _raw_spin_lock+0x27/0x40
 tpacket_rcv+0x863/0x3b30
 dev_queue_xmit_nit+0x709/0xa40
 vrf_finish_direct+0x26e/0x340 [vrf]
 vrf_l3_out+0x5f4/0xe80 [vrf]
 __ip_local_out+0x51e/0x7a0

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

---

v2: Edit patch description.

 drivers/net/vrf.c | 2 ++
 net/core/dev.c    | 1 +
 2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
 		eth_zero_addr(eth->h_dest);
 		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
 		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
 		skb_pull(skb, ETH_HLEN);
 	}
net/core/dev.c +1 -0
index cd479f5f22f6..566e69a38eed 100644
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
 /*
  *	Support routine. Sends outgoing frames to any network
  *	taps currently in use.
+ *	BH must be disabled before calling this.
  */
 
 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Signature · 1 line
2.42.0

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>

#2 lore raw
In reply to [email protected]
Quoted 54 lines · greearb@ wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  packet_rcv+0xa33/0x1320
  __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
  __netif_receive_skb_list_core+0x2c9/0x890
  netif_receive_skb_list_internal+0x610/0xcc0
  napi_complete_done+0x1c0/0x7c0
  igb_poll+0x1dbb/0x57e0 [igb]
  __napi_poll.constprop.0+0x99/0x430
  net_rx_action+0x8e7/0xe10
  handle_softirqs+0x1b7/0x800
  __irq_exit_rcu+0x91/0xc0
  irq_exit_rcu+0x5/0x10
  [snip]

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0
       ----
  lock(rlock-AF_PACKET);
  <Interrupt>
    lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
 <TASK>
 dump_stack_lvl+0x73/0xa0
 mark_lock+0x102e/0x16b0
 __lock_acquire+0x9ae/0x6170
 lock_acquire+0x19a/0x4f0
 _raw_spin_lock+0x27/0x40
 tpacket_rcv+0x863/0x3b30
 dev_queue_xmit_nit+0x709/0xa40
 vrf_finish_direct+0x26e/0x340 [vrf]
 vrf_l3_out+0x5f4/0xe80 [vrf]
 __ip_local_out+0x51e/0x7a0
Quoted 6 lines

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

Please Cc: all previous reviewers and folks who participated in the discussion. I entirely missed this. No need to add as Cc tags, just --cc in git send-email will do.

Quoted 27 lines

---

v2: Edit patch description.

 drivers/net/vrf.c | 2 ++
 net/core/dev.c    | 1 +
 2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
 		eth_zero_addr(eth->h_dest);
 		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
 		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
 		skb_pull(skb, ETH_HLEN);
 	}
net/core/dev.c +0 -0
index cd479f5f22f6..566e69a38eed 100644

net/core/dev.c EXPORT_SYMBOL_GPL(dev_nit_active); · line 2288

Quoted 4 lines
net/core/dev.c +1 -0
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
 /*
  *	Support routine. Sends outgoing frames to any network
  *	taps currently in use.
+ *	BH must be disabled before calling this.

Unnecessary. Especially patches for stable should be minimal to reduce the chance of conflicts.

author
#3 lore raw
Quoted 86 lines · Willem de Bruijn
Quoted 54 lines · greearb@ wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

   lock_acquire+0x19a/0x4f0
   _raw_spin_lock+0x27/0x40
   packet_rcv+0xa33/0x1320
   __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
   __netif_receive_skb_list_core+0x2c9/0x890
   netif_receive_skb_list_internal+0x610/0xcc0
   napi_complete_done+0x1c0/0x7c0
   igb_poll+0x1dbb/0x57e0 [igb]
   __napi_poll.constprop.0+0x99/0x430
   net_rx_action+0x8e7/0xe10
   handle_softirqs+0x1b7/0x800
   __irq_exit_rcu+0x91/0xc0
   irq_exit_rcu+0x5/0x10
   [snip]

other info that might help us debug this:

  Possible unsafe locking scenario:
        CPU0
        ----
   lock(rlock-AF_PACKET);
   <Interrupt>
     lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
  <TASK>
  dump_stack_lvl+0x73/0xa0
  mark_lock+0x102e/0x16b0
  __lock_acquire+0x9ae/0x6170
  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  tpacket_rcv+0x863/0x3b30
  dev_queue_xmit_nit+0x709/0xa40
  vrf_finish_direct+0x26e/0x340 [vrf]
  vrf_l3_out+0x5f4/0xe80 [vrf]
  __ip_local_out+0x51e/0x7a0
Quoted 6 lines

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

Please Cc: all previous reviewers and folks who participated in the discussion. I entirely missed this. No need to add as Cc tags, just --cc in git send-email will do.

Quoted 20 lines

---

v2: Edit patch description.

  drivers/net/vrf.c | 2 ++
  net/core/dev.c    | 1 +
  2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
  		eth_zero_addr(eth->h_dest);
  		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
  		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
Quoted 14 lines

net/core/dev.c EXPORT_SYMBOL_GPL(dev_nit_active); · line 2288

Quoted 11 lines
drivers/net/vrf.c +0 -0
  		skb_pull(skb, ETH_HLEN);
  	}
net/core/dev.c +1 -0
index cd479f5f22f6..566e69a38eed 100644
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
  /*
   *	Support routine. Sends outgoing frames to any network
   *	taps currently in use.
+ *	BH must be disabled before calling this.

Unnecessary. Especially patches for stable should be minimal to reduce the chance of conflicts.

I was asked to add this, and also asked to CC stable.

Thanks,
Ben

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com
maintainer
#4 lore raw
Quoted 100 lines · Ben Greear
Quoted 90 lines · Willem de Bruijn
Quoted 54 lines · greearb@ wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

   lock_acquire+0x19a/0x4f0
   _raw_spin_lock+0x27/0x40
   packet_rcv+0xa33/0x1320
   __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
   __netif_receive_skb_list_core+0x2c9/0x890
   netif_receive_skb_list_internal+0x610/0xcc0
   napi_complete_done+0x1c0/0x7c0
   igb_poll+0x1dbb/0x57e0 [igb]
   __napi_poll.constprop.0+0x99/0x430
   net_rx_action+0x8e7/0xe10
   handle_softirqs+0x1b7/0x800
   __irq_exit_rcu+0x91/0xc0
   irq_exit_rcu+0x5/0x10
   [snip]

other info that might help us debug this:

  Possible unsafe locking scenario:
        CPU0
        ----
   lock(rlock-AF_PACKET);
   <Interrupt>
     lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
  <TASK>
  dump_stack_lvl+0x73/0xa0
  mark_lock+0x102e/0x16b0
  __lock_acquire+0x9ae/0x6170
  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  tpacket_rcv+0x863/0x3b30
  dev_queue_xmit_nit+0x709/0xa40
  vrf_finish_direct+0x26e/0x340 [vrf]
  vrf_l3_out+0x5f4/0xe80 [vrf]
  __ip_local_out+0x51e/0x7a0
Quoted 6 lines

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

Please Cc: all previous reviewers and folks who participated in the discussion. I entirely missed this. No need to add as Cc tags, just --cc in git send-email will do.

Quoted 24 lines

---

v2: Edit patch description.

  drivers/net/vrf.c | 2 ++
  net/core/dev.c    | 1 +
  2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
  		eth_zero_addr(eth->h_dest);
  		eth->h_proto = skb->protocol;
+		rcu_read_lock_bh();
  		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
  		skb_pull(skb, ETH_HLEN);
  	}
net/core/dev.c +0 -0
index cd479f5f22f6..566e69a38eed 100644
Quoted 8 lines

net/core/dev.c EXPORT_SYMBOL_GPL(dev_nit_active); · line 2288

Quoted 5 lines
net/core/dev.c +1 -0
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
  /*
   *	Support routine. Sends outgoing frames to any network
   *	taps currently in use.
+ *	BH must be disabled before calling this.

Unnecessary. Especially patches for stable should be minimal to reduce the chance of conflicts.

Quoted 1 line

I was asked to add this, and also asked to CC stable.

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>

#5 lore raw
In reply to Ben Greear
Quoted 102 lines · Ben Greear
Quoted 86 lines · Willem de Bruijn
Quoted 54 lines · greearb@ wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

   lock_acquire+0x19a/0x4f0
   _raw_spin_lock+0x27/0x40
   packet_rcv+0xa33/0x1320
   __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
   __netif_receive_skb_list_core+0x2c9/0x890
   netif_receive_skb_list_internal+0x610/0xcc0
   napi_complete_done+0x1c0/0x7c0
   igb_poll+0x1dbb/0x57e0 [igb]
   __napi_poll.constprop.0+0x99/0x430
   net_rx_action+0x8e7/0xe10
   handle_softirqs+0x1b7/0x800
   __irq_exit_rcu+0x91/0xc0
   irq_exit_rcu+0x5/0x10
   [snip]

other info that might help us debug this:

  Possible unsafe locking scenario:
        CPU0
        ----
   lock(rlock-AF_PACKET);
   <Interrupt>
     lock(rlock-AF_PACKET);

*** DEADLOCK ***

Call Trace:
  <TASK>
  dump_stack_lvl+0x73/0xa0
  mark_lock+0x102e/0x16b0
  __lock_acquire+0x9ae/0x6170
  lock_acquire+0x19a/0x4f0
  _raw_spin_lock+0x27/0x40
  tpacket_rcv+0x863/0x3b30
  dev_queue_xmit_nit+0x709/0xa40
  vrf_finish_direct+0x26e/0x340 [vrf]
  vrf_l3_out+0x5f4/0xe80 [vrf]
  __ip_local_out+0x51e/0x7a0
Quoted 6 lines

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

Please Cc: all previous reviewers and folks who participated in the discussion. I entirely missed this. No need to add as Cc tags, just --cc in git send-email will do.

Quoted 20 lines

---

v2: Edit patch description.

  drivers/net/vrf.c | 2 ++
  net/core/dev.c    | 1 +
  2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
  		eth_zero_addr(eth->h_dest);
  		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
  		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
Quoted 14 lines

net/core/dev.c EXPORT_SYMBOL_GPL(dev_nit_active); · line 2288

Quoted 11 lines
drivers/net/vrf.c +0 -0
  		skb_pull(skb, ETH_HLEN);
  	}
net/core/dev.c +1 -0
index cd479f5f22f6..566e69a38eed 100644
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
  /*
   *	Support routine. Sends outgoing frames to any network
   *	taps currently in use.
+ *	BH must be disabled before calling this.

Unnecessary. Especially patches for stable should be minimal to reduce the chance of conflicts.

Quoted 1 line

I was asked to add this, and also asked to CC stable.

Conflicting feedback is annoying, but I disagree with the other comment.

Not only does it potentially complicate backporting, it also is no longer a strict revert. In which case it must not be labeled as such.

Easier is to keep the revert unmodified, and add the comment to the commit message.

Most important is the Link to our earlier thread that explains the history.

If the process appears byzantine, if you prefer I can also send it.

Thanks,

Willem

author
#6 lore raw
Quoted 111 lines · Willem de Bruijn
Quoted 102 lines · Ben Greear
Quoted 86 lines · Willem de Bruijn
Quoted 54 lines · greearb@ wrote:

From: Ben Greear <[email protected]>

This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.

dev_queue_xmit_nit needs to run with bh locking, otherwise it conflicts with packets coming in from a nic in softirq context and packets being transmitted from user context.

================================
WARNING: inconsistent lock state
6.11.0 #1 Tainted: G W
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage. btserver/134819 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff8882da30c118 (rlock-AF_PACKET){+.?.}-{2:2}, at: tpacket_rcv+0x863/0x3b30 {IN-SOFTIRQ-W} state was registered at:

    lock_acquire+0x19a/0x4f0
    _raw_spin_lock+0x27/0x40
    packet_rcv+0xa33/0x1320
    __netif_receive_skb_core.constprop.0+0xcb0/0x3a90
    __netif_receive_skb_list_core+0x2c9/0x890
    netif_receive_skb_list_internal+0x610/0xcc0
    napi_complete_done+0x1c0/0x7c0
    igb_poll+0x1dbb/0x57e0 [igb]
    __napi_poll.constprop.0+0x99/0x430
    net_rx_action+0x8e7/0xe10
    handle_softirqs+0x1b7/0x800
    __irq_exit_rcu+0x91/0xc0
    irq_exit_rcu+0x5/0x10
    [snip]

other info that might help us debug this:

   Possible unsafe locking scenario:
         CPU0
         ----
    lock(rlock-AF_PACKET);
    <Interrupt>
      lock(rlock-AF_PACKET);
   *** DEADLOCK ***
Call Trace:
   <TASK>
   dump_stack_lvl+0x73/0xa0
   mark_lock+0x102e/0x16b0
   __lock_acquire+0x9ae/0x6170
   lock_acquire+0x19a/0x4f0
   _raw_spin_lock+0x27/0x40
   tpacket_rcv+0x863/0x3b30
   dev_queue_xmit_nit+0x709/0xa40
   vrf_finish_direct+0x26e/0x340 [vrf]
   vrf_l3_out+0x5f4/0xe80 [vrf]
   __ip_local_out+0x51e/0x7a0
Quoted 6 lines

[snip]

Fixes 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")

Please Cc: all previous reviewers and folks who participated in the discussion. I entirely missed this. No need to add as Cc tags, just --cc in git send-email will do.

Quoted 20 lines

---

v2: Edit patch description.

   drivers/net/vrf.c | 2 ++
   net/core/dev.c    | 1 +
   2 files changed, 3 insertions(+)
drivers/net/vrf.c +2 -0
index 4d8ccaf9a2b4..4087f72f0d2b 100644
@@ -608,7 +608,9 @@ static void vrf_finish_direct(struct sk_buff *skb)
   		eth_zero_addr(eth->h_dest);
   		eth->h_proto = skb->protocol;
 
+		rcu_read_lock_bh();
   		dev_queue_xmit_nit(skb, vrf_dev);
+		rcu_read_unlock_bh();
 
Quoted 14 lines

net/core/dev.c EXPORT_SYMBOL_GPL(dev_nit_active); · line 2288

Quoted 11 lines
drivers/net/vrf.c +0 -0
   		skb_pull(skb, ETH_HLEN);
   	}
net/core/dev.c +1 -0
index cd479f5f22f6..566e69a38eed 100644
@@ -2285,6 +2285,7 @@ EXPORT_SYMBOL_GPL(dev_nit_active);
   /*
    *	Support routine. Sends outgoing frames to any network
    *	taps currently in use.
+ *	BH must be disabled before calling this.

Unnecessary. Especially patches for stable should be minimal to reduce the chance of conflicts.

Quoted 1 line

I was asked to add this, and also asked to CC stable.

Conflicting feedback is annoying, but I disagree with the other comment.

Not only does it potentially complicate backporting, it also is no longer a strict revert. In which case it must not be labeled as such.

Quoted 7 lines

Easier is to keep the revert unmodified, and add the comment to the commit message.

Most important is the Link to our earlier thread that explains the history.

If the process appears byzantine, if you prefer I can also send it.

I would appreciate it if you can send it. As long as it functions, I will be happy.

Thanks,
Ben

Quoted 7 lines

Thanks,

Willem

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com