linux-stable/net/core/dev_addr_lists.c

1051 lines
26 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* net/core/dev_addr_lists.c - Functions for handling net device lists
* Copyright (c) 2010 Jiri Pirko <jpirko@redhat.com>
*
* This file contains functions for working with unicast, multicast and device
* addresses lists.
*/
#include <linux/netdevice.h>
#include <linux/rtnetlink.h>
#include <linux/export.h>
#include <linux/list.h>
#include "dev.h"
/*
* General list handling functions
*/
static int __hw_addr_insert(struct netdev_hw_addr_list *list,
struct netdev_hw_addr *new, int addr_len)
{
struct rb_node **ins_point = &list->tree.rb_node, *parent = NULL;
struct netdev_hw_addr *ha;
while (*ins_point) {
int diff;
ha = rb_entry(*ins_point, struct netdev_hw_addr, node);
diff = memcmp(new->addr, ha->addr, addr_len);
if (diff == 0)
diff = memcmp(&new->type, &ha->type, sizeof(new->type));
parent = *ins_point;
if (diff < 0)
ins_point = &parent->rb_left;
else if (diff > 0)
ins_point = &parent->rb_right;
else
return -EEXIST;
}
rb_link_node_rcu(&new->node, parent, ins_point);
rb_insert_color(&new->node, &list->tree);
return 0;
}
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
static struct netdev_hw_addr*
__hw_addr_create(const unsigned char *addr, int addr_len,
unsigned char addr_type, bool global, bool sync)
{
struct netdev_hw_addr *ha;
int alloc_size;
alloc_size = sizeof(*ha);
if (alloc_size < L1_CACHE_BYTES)
alloc_size = L1_CACHE_BYTES;
ha = kmalloc(alloc_size, GFP_ATOMIC);
if (!ha)
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
return NULL;
memcpy(ha->addr, addr, addr_len);
ha->type = addr_type;
ha->refcount = 1;
ha->global_use = global;
ha->synced = sync ? 1 : 0;
ha->sync_cnt = 0;
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
return ha;
}
static int __hw_addr_add_ex(struct netdev_hw_addr_list *list,
const unsigned char *addr, int addr_len,
unsigned char addr_type, bool global, bool sync,
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
int sync_count, bool exclusive)
{
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
struct rb_node **ins_point = &list->tree.rb_node, *parent = NULL;
struct netdev_hw_addr *ha;
if (addr_len > MAX_ADDR_LEN)
return -EINVAL;
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
while (*ins_point) {
int diff;
ha = rb_entry(*ins_point, struct netdev_hw_addr, node);
diff = memcmp(addr, ha->addr, addr_len);
if (diff == 0)
diff = memcmp(&addr_type, &ha->type, sizeof(addr_type));
parent = *ins_point;
if (diff < 0) {
ins_point = &parent->rb_left;
} else if (diff > 0) {
ins_point = &parent->rb_right;
} else {
if (exclusive)
return -EEXIST;
if (global) {
/* check if addr is already used as global */
if (ha->global_use)
return 0;
else
ha->global_use = true;
}
if (sync) {
if (ha->synced && sync_count)
return -EEXIST;
else
ha->synced++;
}
ha->refcount++;
return 0;
}
}
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
ha = __hw_addr_create(addr, addr_len, addr_type, global, sync);
if (!ha)
return -ENOMEM;
rb_link_node(&ha->node, parent, ins_point);
rb_insert_color(&ha->node, &list->tree);
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
list_add_tail_rcu(&ha->list, &list->list);
list->count++;
return 0;
}
static int __hw_addr_add(struct netdev_hw_addr_list *list,
const unsigned char *addr, int addr_len,
unsigned char addr_type)
{
return __hw_addr_add_ex(list, addr, addr_len, addr_type, false, false,
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
0, false);
}
static int __hw_addr_del_entry(struct netdev_hw_addr_list *list,
struct netdev_hw_addr *ha, bool global,
bool sync)
{
if (global && !ha->global_use)
return -ENOENT;
if (sync && !ha->synced)
return -ENOENT;
if (global)
ha->global_use = false;
if (sync)
ha->synced--;
if (--ha->refcount)
return 0;
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
rb_erase(&ha->node, &list->tree);
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
list_del_rcu(&ha->list);
kfree_rcu(ha, rcu_head);
list->count--;
return 0;
}
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
static struct netdev_hw_addr *__hw_addr_lookup(struct netdev_hw_addr_list *list,
const unsigned char *addr, int addr_len,
unsigned char addr_type)
{
struct rb_node *node;
node = list->tree.rb_node;
while (node) {
struct netdev_hw_addr *ha = rb_entry(node, struct netdev_hw_addr, node);
int diff = memcmp(addr, ha->addr, addr_len);
if (diff == 0 && addr_type)
diff = memcmp(&addr_type, &ha->type, sizeof(addr_type));
if (diff < 0)
node = node->rb_left;
else if (diff > 0)
node = node->rb_right;
else
return ha;
}
return NULL;
}
static int __hw_addr_del_ex(struct netdev_hw_addr_list *list,
const unsigned char *addr, int addr_len,
unsigned char addr_type, bool global, bool sync)
{
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
struct netdev_hw_addr *ha = __hw_addr_lookup(list, addr, addr_len, addr_type);
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
if (!ha)
return -ENOENT;
return __hw_addr_del_entry(list, ha, global, sync);
}
static int __hw_addr_del(struct netdev_hw_addr_list *list,
const unsigned char *addr, int addr_len,
unsigned char addr_type)
{
return __hw_addr_del_ex(list, addr, addr_len, addr_type, false, false);
}
static int __hw_addr_sync_one(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr *ha,
int addr_len)
{
int err;
err = __hw_addr_add_ex(to_list, ha->addr, addr_len, ha->type,
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
false, true, ha->sync_cnt, false);
if (err && err != -EEXIST)
return err;
if (!err) {
ha->sync_cnt++;
ha->refcount++;
}
return 0;
}
static void __hw_addr_unsync_one(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list,
struct netdev_hw_addr *ha,
int addr_len)
{
int err;
err = __hw_addr_del_ex(to_list, ha->addr, addr_len, ha->type,
false, true);
if (err)
return;
ha->sync_cnt--;
/* address on from list is not marked synced */
__hw_addr_del_entry(from_list, ha, false, false);
}
static int __hw_addr_sync_multiple(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list,
int addr_len)
{
int err = 0;
struct netdev_hw_addr *ha, *tmp;
list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
if (ha->sync_cnt == ha->refcount) {
__hw_addr_unsync_one(to_list, from_list, ha, addr_len);
} else {
err = __hw_addr_sync_one(to_list, ha, addr_len);
if (err)
break;
}
}
return err;
}
/* This function only works where there is a strict 1-1 relationship
* between source and destionation of they synch. If you ever need to
* sync addresses to more then 1 destination, you need to use
* __hw_addr_sync_multiple().
*/
int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list,
int addr_len)
{
int err = 0;
struct netdev_hw_addr *ha, *tmp;
list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
if (!ha->sync_cnt) {
err = __hw_addr_sync_one(to_list, ha, addr_len);
if (err)
break;
} else if (ha->refcount == 1)
__hw_addr_unsync_one(to_list, from_list, ha, addr_len);
}
return err;
}
EXPORT_SYMBOL(__hw_addr_sync);
void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
struct netdev_hw_addr_list *from_list,
int addr_len)
{
struct netdev_hw_addr *ha, *tmp;
list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
if (ha->sync_cnt)
__hw_addr_unsync_one(to_list, from_list, ha, addr_len);
}
}
EXPORT_SYMBOL(__hw_addr_unsync);
/**
* __hw_addr_sync_dev - Synchonize device's multicast list
* @list: address list to syncronize
* @dev: device to sync
* @sync: function to call if address should be added
* @unsync: function to call if address should be removed
*
* This function is intended to be called from the ndo_set_rx_mode
* function of devices that require explicit address add/remove
* notifications. The unsync function may be NULL in which case
* the addresses requiring removal will simply be removed without
* any notification to the device.
**/
int __hw_addr_sync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*sync)(struct net_device *, const unsigned char *),
int (*unsync)(struct net_device *,
const unsigned char *))
{
struct netdev_hw_addr *ha, *tmp;
int err;
/* first go through and flush out any stale entries */
list_for_each_entry_safe(ha, tmp, &list->list, list) {
if (!ha->sync_cnt || ha->refcount != 1)
continue;
/* if unsync is defined and fails defer unsyncing address */
if (unsync && unsync(dev, ha->addr))
continue;
ha->sync_cnt--;
__hw_addr_del_entry(list, ha, false, false);
}
/* go through and sync new entries to the list */
list_for_each_entry_safe(ha, tmp, &list->list, list) {
if (ha->sync_cnt)
continue;
err = sync(dev, ha->addr);
if (err)
return err;
ha->sync_cnt++;
ha->refcount++;
}
return 0;
}
EXPORT_SYMBOL(__hw_addr_sync_dev);
/**
* __hw_addr_ref_sync_dev - Synchronize device's multicast address list taking
* into account references
* @list: address list to synchronize
* @dev: device to sync
* @sync: function to call if address or reference on it should be added
* @unsync: function to call if address or some reference on it should removed
*
* This function is intended to be called from the ndo_set_rx_mode
* function of devices that require explicit address or references on it
* add/remove notifications. The unsync function may be NULL in which case
* the addresses or references on it requiring removal will simply be
* removed without any notification to the device. That is responsibility of
* the driver to identify and distribute address or references on it between
* internal address tables.
**/
int __hw_addr_ref_sync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*sync)(struct net_device *,
const unsigned char *, int),
int (*unsync)(struct net_device *,
const unsigned char *, int))
{
struct netdev_hw_addr *ha, *tmp;
int err, ref_cnt;
/* first go through and flush out any unsynced/stale entries */
list_for_each_entry_safe(ha, tmp, &list->list, list) {
/* sync if address is not used */
if ((ha->sync_cnt << 1) <= ha->refcount)
continue;
/* if fails defer unsyncing address */
ref_cnt = ha->refcount - ha->sync_cnt;
if (unsync && unsync(dev, ha->addr, ref_cnt))
continue;
ha->refcount = (ref_cnt << 1) + 1;
ha->sync_cnt = ref_cnt;
__hw_addr_del_entry(list, ha, false, false);
}
/* go through and sync updated/new entries to the list */
list_for_each_entry_safe(ha, tmp, &list->list, list) {
/* sync if address added or reused */
if ((ha->sync_cnt << 1) >= ha->refcount)
continue;
ref_cnt = ha->refcount - ha->sync_cnt;
err = sync(dev, ha->addr, ref_cnt);
if (err)
return err;
ha->refcount = ref_cnt << 1;
ha->sync_cnt = ref_cnt;
}
return 0;
}
EXPORT_SYMBOL(__hw_addr_ref_sync_dev);
/**
* __hw_addr_ref_unsync_dev - Remove synchronized addresses and references on
* it from device
* @list: address list to remove synchronized addresses (references on it) from
* @dev: device to sync
* @unsync: function to call if address and references on it should be removed
*
* Remove all addresses that were added to the device by
* __hw_addr_ref_sync_dev(). This function is intended to be called from the
* ndo_stop or ndo_open functions on devices that require explicit address (or
* references on it) add/remove notifications. If the unsync function pointer
* is NULL then this function can be used to just reset the sync_cnt for the
* addresses in the list.
**/
void __hw_addr_ref_unsync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *, int))
{
struct netdev_hw_addr *ha, *tmp;
list_for_each_entry_safe(ha, tmp, &list->list, list) {
if (!ha->sync_cnt)
continue;
/* if fails defer unsyncing address */
if (unsync && unsync(dev, ha->addr, ha->sync_cnt))
continue;
ha->refcount -= ha->sync_cnt - 1;
ha->sync_cnt = 0;
__hw_addr_del_entry(list, ha, false, false);
}
}
EXPORT_SYMBOL(__hw_addr_ref_unsync_dev);
/**
* __hw_addr_unsync_dev - Remove synchronized addresses from device
* @list: address list to remove synchronized addresses from
* @dev: device to sync
* @unsync: function to call if address should be removed
*
* Remove all addresses that were added to the device by __hw_addr_sync_dev().
* This function is intended to be called from the ndo_stop or ndo_open
* functions on devices that require explicit address add/remove
* notifications. If the unsync function pointer is NULL then this function
* can be used to just reset the sync_cnt for the addresses in the list.
**/
void __hw_addr_unsync_dev(struct netdev_hw_addr_list *list,
struct net_device *dev,
int (*unsync)(struct net_device *,
const unsigned char *))
{
struct netdev_hw_addr *ha, *tmp;
list_for_each_entry_safe(ha, tmp, &list->list, list) {
if (!ha->sync_cnt)
continue;
/* if unsync is defined and fails defer unsyncing address */
if (unsync && unsync(dev, ha->addr))
continue;
ha->sync_cnt--;
__hw_addr_del_entry(list, ha, false, false);
}
}
EXPORT_SYMBOL(__hw_addr_unsync_dev);
static void __hw_addr_flush(struct netdev_hw_addr_list *list)
{
struct netdev_hw_addr *ha, *tmp;
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
list->tree = RB_ROOT;
list_for_each_entry_safe(ha, tmp, &list->list, list) {
list_del_rcu(&ha->list);
kfree_rcu(ha, rcu_head);
}
list->count = 0;
}
void __hw_addr_init(struct netdev_hw_addr_list *list)
{
INIT_LIST_HEAD(&list->list);
list->count = 0;
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
list->tree = RB_ROOT;
}
EXPORT_SYMBOL(__hw_addr_init);
/*
* Device addresses handling functions
*/
/* Check that netdev->dev_addr is not written to directly as this would
* break the rbtree layout. All changes should go thru dev_addr_set() and co.
* Remove this check in mid-2024.
*/
void dev_addr_check(struct net_device *dev)
{
if (!memcmp(dev->dev_addr, dev->dev_addr_shadow, MAX_ADDR_LEN))
return;
netdev_warn(dev, "Current addr: %*ph\n", MAX_ADDR_LEN, dev->dev_addr);
netdev_warn(dev, "Expected addr: %*ph\n",
MAX_ADDR_LEN, dev->dev_addr_shadow);
netdev_WARN(dev, "Incorrect netdev->dev_addr\n");
}
/**
* dev_addr_flush - Flush device address list
* @dev: device
*
* Flush device address list and reset ->dev_addr.
*
* The caller must hold the rtnl_mutex.
*/
void dev_addr_flush(struct net_device *dev)
{
/* rtnl_mutex must be held here */
dev_addr_check(dev);
__hw_addr_flush(&dev->dev_addrs);
dev->dev_addr = NULL;
}
/**
* dev_addr_init - Init device address list
* @dev: device
*
* Init device address list and create the first element,
* used by ->dev_addr.
*
* The caller must hold the rtnl_mutex.
*/
int dev_addr_init(struct net_device *dev)
{
unsigned char addr[MAX_ADDR_LEN];
struct netdev_hw_addr *ha;
int err;
/* rtnl_mutex must be held here */
__hw_addr_init(&dev->dev_addrs);
memset(addr, 0, sizeof(addr));
err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
NETDEV_HW_ADDR_T_LAN);
if (!err) {
/*
* Get the first (previously created) address from the list
* and set dev_addr pointer to this location.
*/
ha = list_first_entry(&dev->dev_addrs.list,
struct netdev_hw_addr, list);
dev->dev_addr = ha->addr;
}
return err;
}
void dev_addr_mod(struct net_device *dev, unsigned int offset,
const void *addr, size_t len)
{
struct netdev_hw_addr *ha;
dev_addr_check(dev);
ha = container_of(dev->dev_addr, struct netdev_hw_addr, addr[0]);
rb_erase(&ha->node, &dev->dev_addrs.tree);
memcpy(&ha->addr[offset], addr, len);
memcpy(&dev->dev_addr_shadow[offset], addr, len);
WARN_ON(__hw_addr_insert(&dev->dev_addrs, ha, dev->addr_len));
}
EXPORT_SYMBOL(dev_addr_mod);
/**
* dev_addr_add - Add a device address
* @dev: device
* @addr: address to add
* @addr_type: address type
*
* Add a device address to the device or increase the reference count if
* it already exists.
*
* The caller must hold the rtnl_mutex.
*/
int dev_addr_add(struct net_device *dev, const unsigned char *addr,
unsigned char addr_type)
{
int err;
ASSERT_RTNL();
err = dev_pre_changeaddr_notify(dev, addr, NULL);
if (err)
return err;
err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
if (!err)
call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
return err;
}
EXPORT_SYMBOL(dev_addr_add);
/**
* dev_addr_del - Release a device address.
* @dev: device
* @addr: address to delete
* @addr_type: address type
*
* Release reference to a device address and remove it from the device
* if the reference count drops to zero.
*
* The caller must hold the rtnl_mutex.
*/
int dev_addr_del(struct net_device *dev, const unsigned char *addr,
unsigned char addr_type)
{
int err;
struct netdev_hw_addr *ha;
ASSERT_RTNL();
/*
* We can not remove the first address from the list because
* dev->dev_addr points to that.
*/
ha = list_first_entry(&dev->dev_addrs.list,
struct netdev_hw_addr, list);
if (!memcmp(ha->addr, addr, dev->addr_len) &&
ha->type == addr_type && ha->refcount == 1)
return -ENOENT;
err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
addr_type);
if (!err)
call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
return err;
}
EXPORT_SYMBOL(dev_addr_del);
/*
* Unicast list handling functions
*/
/**
* dev_uc_add_excl - Add a global secondary unicast address
* @dev: device
* @addr: address to add
*/
int dev_uc_add_excl(struct net_device *dev, const unsigned char *addr)
{
int err;
netif_addr_lock_bh(dev);
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
err = __hw_addr_add_ex(&dev->uc, addr, dev->addr_len,
NETDEV_HW_ADDR_T_UNICAST, true, false,
0, true);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
EXPORT_SYMBOL(dev_uc_add_excl);
/**
* dev_uc_add - Add a secondary unicast address
* @dev: device
* @addr: address to add
*
* Add a secondary unicast address to the device or increase
* the reference count if it already exists.
*/
int dev_uc_add(struct net_device *dev, const unsigned char *addr)
{
int err;
netif_addr_lock_bh(dev);
err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
NETDEV_HW_ADDR_T_UNICAST);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
EXPORT_SYMBOL(dev_uc_add);
/**
* dev_uc_del - Release secondary unicast address.
* @dev: device
* @addr: address to delete
*
* Release reference to a secondary unicast address and remove it
* from the device if the reference count drops to zero.
*/
int dev_uc_del(struct net_device *dev, const unsigned char *addr)
{
int err;
netif_addr_lock_bh(dev);
err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
NETDEV_HW_ADDR_T_UNICAST);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
EXPORT_SYMBOL(dev_uc_del);
/**
* dev_uc_sync - Synchronize device's unicast list to another device
* @to: destination device
* @from: source device
*
* Add newly added addresses to the destination device and release
* addresses that have no users left. The source device must be
* locked by netif_addr_lock_bh.
*
* This function is intended to be called from the dev->set_rx_mode
* function of layered software devices. This function assumes that
* addresses will only ever be synced to the @to devices and no other.
*/
int dev_uc_sync(struct net_device *to, struct net_device *from)
{
int err = 0;
if (to->addr_len != from->addr_len)
return -EINVAL;
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
if (!err)
__dev_set_rx_mode(to);
netif_addr_unlock(to);
return err;
}
EXPORT_SYMBOL(dev_uc_sync);
/**
* dev_uc_sync_multiple - Synchronize device's unicast list to another
* device, but allow for multiple calls to sync to multiple devices.
* @to: destination device
* @from: source device
*
* Add newly added addresses to the destination device and release
* addresses that have been deleted from the source. The source device
* must be locked by netif_addr_lock_bh.
*
* This function is intended to be called from the dev->set_rx_mode
* function of layered software devices. It allows for a single source
* device to be synced to multiple destination devices.
*/
int dev_uc_sync_multiple(struct net_device *to, struct net_device *from)
{
int err = 0;
if (to->addr_len != from->addr_len)
return -EINVAL;
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
err = __hw_addr_sync_multiple(&to->uc, &from->uc, to->addr_len);
if (!err)
__dev_set_rx_mode(to);
netif_addr_unlock(to);
return err;
}
EXPORT_SYMBOL(dev_uc_sync_multiple);
/**
* dev_uc_unsync - Remove synchronized addresses from the destination device
* @to: destination device
* @from: source device
*
* Remove all addresses that were added to the destination device by
* dev_uc_sync(). This function is intended to be called from the
* dev->stop function of layered software devices.
*/
void dev_uc_unsync(struct net_device *to, struct net_device *from)
{
if (to->addr_len != from->addr_len)
return;
/* netif_addr_lock_bh() uses lockdep subclass 0, this is okay for two
* reasons:
* 1) This is always called without any addr_list_lock, so as the
* outermost one here, it must be 0.
* 2) This is called by some callers after unlinking the upper device,
* so the dev->lower_level becomes 1 again.
* Therefore, the subclass for 'from' is 0, for 'to' is either 1 or
* larger.
*/
netif_addr_lock_bh(from);
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
__hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
__dev_set_rx_mode(to);
netif_addr_unlock(to);
netif_addr_unlock_bh(from);
}
EXPORT_SYMBOL(dev_uc_unsync);
/**
* dev_uc_flush - Flush unicast addresses
* @dev: device
*
* Flush unicast addresses.
*/
void dev_uc_flush(struct net_device *dev)
{
netif_addr_lock_bh(dev);
__hw_addr_flush(&dev->uc);
netif_addr_unlock_bh(dev);
}
EXPORT_SYMBOL(dev_uc_flush);
/**
* dev_uc_init - Init unicast address list
* @dev: device
*
* Init unicast address list.
*/
void dev_uc_init(struct net_device *dev)
{
__hw_addr_init(&dev->uc);
}
EXPORT_SYMBOL(dev_uc_init);
/*
* Multicast list handling functions
*/
/**
* dev_mc_add_excl - Add a global secondary multicast address
* @dev: device
* @addr: address to add
*/
int dev_mc_add_excl(struct net_device *dev, const unsigned char *addr)
{
int err;
netif_addr_lock_bh(dev);
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
err = __hw_addr_add_ex(&dev->mc, addr, dev->addr_len,
NETDEV_HW_ADDR_T_MULTICAST, true, false,
0, true);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
EXPORT_SYMBOL(dev_mc_add_excl);
static int __dev_mc_add(struct net_device *dev, const unsigned char *addr,
bool global)
{
int err;
netif_addr_lock_bh(dev);
err = __hw_addr_add_ex(&dev->mc, addr, dev->addr_len,
net-next: When a bond have a massive amount of VLANs with IPv6 addresses, performance of changing link state, attaching a VRF, changing an IPv6 address, etc. go down dramtically. The source of most of the slow down is the `dev_addr_lists.c` module, which mainatins a linked list of HW addresses. When using IPv6, this list grows for each IPv6 address added on a VLAN, since each IPv6 address has a multicast HW address associated with it. When performing any modification to the involved links, this list is traversed many times, often for nothing, all while holding the RTNL lock. Instead, this patch adds an auxilliary rbtree which cuts down traversal time significantly. Performance can be seen with the following script: #!/bin/bash ip netns del test || true 2>/dev/null ip netns add test echo 1 | ip netns exec test tee /proc/sys/net/ipv6/conf/all/keep_addr_on_down > /dev/null set -e ip -n test link add foo type veth peer name bar ip -n test link add b1 type bond ip -n test link add florp type vrf table 10 ip -n test link set bar master b1 ip -n test link set foo up ip -n test link set bar up ip -n test link set b1 up ip -n test link set florp up VLAN_COUNT=1500 BASE_DEV=b1 echo Creating vlans ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link add link $BASE_DEV name foo.\$i type vlan id \$i; done" echo Bringing them up ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i up; done" echo Assiging IPv6 Addresses ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test address add dev foo.\$i 2000::\$i/64; done" echo Attaching to VRF ip netns exec test time -p bash -c "for i in \$(seq 1 $VLAN_COUNT); do ip -n test link set foo.\$i master florp; done" On an Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz machine, the performance before the patch is (truncated): Creating vlans real 108.35 Bringing them up real 4.96 Assiging IPv6 Addresses real 19.22 Attaching to VRF real 458.84 After the patch: Creating vlans real 5.59 Bringing them up real 5.07 Assiging IPv6 Addresses real 5.64 Attaching to VRF real 25.37 Cc: David S. Miller <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Lu Wei <luwei32@huawei.com> Cc: Xiongfeng Wang <wangxiongfeng2@huawei.com> Cc: Taehee Yoo <ap420073@gmail.com> Signed-off-by: Gilad Naaman <gnaaman@drivenets.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-08-19 07:17:27 +00:00
NETDEV_HW_ADDR_T_MULTICAST, global, false,
0, false);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
/**
* dev_mc_add - Add a multicast address
* @dev: device
* @addr: address to add
*
* Add a multicast address to the device or increase
* the reference count if it already exists.
*/
int dev_mc_add(struct net_device *dev, const unsigned char *addr)
{
return __dev_mc_add(dev, addr, false);
}
EXPORT_SYMBOL(dev_mc_add);
/**
* dev_mc_add_global - Add a global multicast address
* @dev: device
* @addr: address to add
*
* Add a global multicast address to the device.
*/
int dev_mc_add_global(struct net_device *dev, const unsigned char *addr)
{
return __dev_mc_add(dev, addr, true);
}
EXPORT_SYMBOL(dev_mc_add_global);
static int __dev_mc_del(struct net_device *dev, const unsigned char *addr,
bool global)
{
int err;
netif_addr_lock_bh(dev);
err = __hw_addr_del_ex(&dev->mc, addr, dev->addr_len,
NETDEV_HW_ADDR_T_MULTICAST, global, false);
if (!err)
__dev_set_rx_mode(dev);
netif_addr_unlock_bh(dev);
return err;
}
/**
* dev_mc_del - Delete a multicast address.
* @dev: device
* @addr: address to delete
*
* Release reference to a multicast address and remove it
* from the device if the reference count drops to zero.
*/
int dev_mc_del(struct net_device *dev, const unsigned char *addr)
{
return __dev_mc_del(dev, addr, false);
}
EXPORT_SYMBOL(dev_mc_del);
/**
* dev_mc_del_global - Delete a global multicast address.
* @dev: device
* @addr: address to delete
*
* Release reference to a multicast address and remove it
* from the device if the reference count drops to zero.
*/
int dev_mc_del_global(struct net_device *dev, const unsigned char *addr)
{
return __dev_mc_del(dev, addr, true);
}
EXPORT_SYMBOL(dev_mc_del_global);
/**
* dev_mc_sync - Synchronize device's multicast list to another device
* @to: destination device
* @from: source device
*
* Add newly added addresses to the destination device and release
* addresses that have no users left. The source device must be
* locked by netif_addr_lock_bh.
*
* This function is intended to be called from the ndo_set_rx_mode
* function of layered software devices.
*/
int dev_mc_sync(struct net_device *to, struct net_device *from)
{
int err = 0;
if (to->addr_len != from->addr_len)
return -EINVAL;
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
err = __hw_addr_sync(&to->mc, &from->mc, to->addr_len);
if (!err)
__dev_set_rx_mode(to);
netif_addr_unlock(to);
return err;
}
EXPORT_SYMBOL(dev_mc_sync);
/**
* dev_mc_sync_multiple - Synchronize device's multicast list to another
* device, but allow for multiple calls to sync to multiple devices.
* @to: destination device
* @from: source device
*
* Add newly added addresses to the destination device and release
* addresses that have no users left. The source device must be
* locked by netif_addr_lock_bh.
*
* This function is intended to be called from the ndo_set_rx_mode
* function of layered software devices. It allows for a single
* source device to be synced to multiple destination devices.
*/
int dev_mc_sync_multiple(struct net_device *to, struct net_device *from)
{
int err = 0;
if (to->addr_len != from->addr_len)
return -EINVAL;
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
err = __hw_addr_sync_multiple(&to->mc, &from->mc, to->addr_len);
if (!err)
__dev_set_rx_mode(to);
netif_addr_unlock(to);
return err;
}
EXPORT_SYMBOL(dev_mc_sync_multiple);
/**
* dev_mc_unsync - Remove synchronized addresses from the destination device
* @to: destination device
* @from: source device
*
* Remove all addresses that were added to the destination device by
* dev_mc_sync(). This function is intended to be called from the
* dev->stop function of layered software devices.
*/
void dev_mc_unsync(struct net_device *to, struct net_device *from)
{
if (to->addr_len != from->addr_len)
return;
/* See the above comments inside dev_uc_unsync(). */
netif_addr_lock_bh(from);
net: core: add nested_level variable in net_device This patch is to add a new variable 'nested_level' into the net_device structure. This variable will be used as a parameter of spin_lock_nested() of dev->addr_list_lock. netif_addr_lock() can be called recursively so spin_lock_nested() is used instead of spin_lock() and dev->lower_level is used as a parameter of spin_lock_nested(). But, dev->lower_level value can be updated while it is being used. So, lockdep would warn a possible deadlock scenario. When a stacked interface is deleted, netif_{uc | mc}_sync() is called recursively. So, spin_lock_nested() is called recursively too. At this moment, the dev->lower_level variable is used as a parameter of it. dev->lower_level value is updated when interfaces are being unlinked/linked immediately. Thus, After unlinking, dev->lower_level shouldn't be a parameter of spin_lock_nested(). A (macvlan) | B (vlan) | C (bridge) | D (macvlan) | E (vlan) | F (bridge) A->lower_level : 6 B->lower_level : 5 C->lower_level : 4 D->lower_level : 3 E->lower_level : 2 F->lower_level : 1 When an interface 'A' is removed, it releases resources. At this moment, netif_addr_lock() would be called. Then, netdev_upper_dev_unlink() is called recursively. Then dev->lower_level is updated. There is no problem. But, when the bridge module is removed, 'C' and 'F' interfaces are removed at once. If 'F' is removed first, a lower_level value is like below. A->lower_level : 5 B->lower_level : 4 C->lower_level : 3 D->lower_level : 2 E->lower_level : 1 F->lower_level : 1 Then, 'C' is removed. at this moment, netif_addr_lock() is called recursively. The ordering is like this. C(3)->D(2)->E(1)->F(1) At this moment, the lower_level value of 'E' and 'F' are the same. So, lockdep warns a possible deadlock scenario. In order to avoid this problem, a new variable 'nested_level' is added. This value is the same as dev->lower_level - 1. But this value is updated in rtnl_unlock(). So, this variable can be used as a parameter of spin_lock_nested() safely in the rtnl context. Test commands: ip link add br0 type bridge vlan_filtering 1 ip link add vlan1 link br0 type vlan id 10 ip link add macvlan2 link vlan1 type macvlan ip link add br3 type bridge vlan_filtering 1 ip link set macvlan2 master br3 ip link add vlan4 link br3 type vlan id 10 ip link add macvlan5 link vlan4 type macvlan ip link add br6 type bridge vlan_filtering 1 ip link set macvlan5 master br6 ip link add vlan7 link br6 type vlan id 10 ip link add macvlan8 link vlan7 type macvlan ip link set br0 up ip link set vlan1 up ip link set macvlan2 up ip link set br3 up ip link set vlan4 up ip link set macvlan5 up ip link set br6 up ip link set vlan7 up ip link set macvlan8 up modprobe -rv bridge Splat looks like: [ 36.057436][ T744] WARNING: possible recursive locking detected [ 36.058848][ T744] 5.9.0-rc6+ #728 Not tainted [ 36.059959][ T744] -------------------------------------------- [ 36.061391][ T744] ip/744 is trying to acquire lock: [ 36.062590][ T744] ffff8c4767509280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_set_rx_mode+0x19/0x30 [ 36.064922][ T744] [ 36.064922][ T744] but task is already holding lock: [ 36.066626][ T744] ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.068851][ T744] [ 36.068851][ T744] other info that might help us debug this: [ 36.070731][ T744] Possible unsafe locking scenario: [ 36.070731][ T744] [ 36.072497][ T744] CPU0 [ 36.073238][ T744] ---- [ 36.074007][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.075290][ T744] lock(&vlan_netdev_addr_lock_key); [ 36.076590][ T744] [ 36.076590][ T744] *** DEADLOCK *** [ 36.076590][ T744] [ 36.078515][ T744] May be due to missing lock nesting notation [ 36.078515][ T744] [ 36.080491][ T744] 3 locks held by ip/744: [ 36.081471][ T744] #0: ffffffff98571df0 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x236/0x490 [ 36.083614][ T744] #1: ffff8c4767769280 (&vlan_netdev_addr_lock_key){+...}-{2:2}, at: dev_uc_add+0x1e/0x60 [ 36.085942][ T744] #2: ffff8c476c8da280 (&bridge_netdev_addr_lock_key/4){+...}-{2:2}, at: dev_uc_sync+0x39/0x80 [ 36.088400][ T744] [ 36.088400][ T744] stack backtrace: [ 36.089772][ T744] CPU: 6 PID: 744 Comm: ip Not tainted 5.9.0-rc6+ #728 [ 36.091364][ T744] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014 [ 36.093630][ T744] Call Trace: [ 36.094416][ T744] dump_stack+0x77/0x9b [ 36.095385][ T744] __lock_acquire+0xbc3/0x1f40 [ 36.096522][ T744] lock_acquire+0xb4/0x3b0 [ 36.097540][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.098657][ T744] ? rtmsg_ifinfo+0x1f/0x30 [ 36.099711][ T744] ? __dev_notify_flags+0xa5/0xf0 [ 36.100874][ T744] ? rtnl_is_locked+0x11/0x20 [ 36.101967][ T744] ? __dev_set_promiscuity+0x7b/0x1a0 [ 36.103230][ T744] _raw_spin_lock_bh+0x38/0x70 [ 36.104348][ T744] ? dev_set_rx_mode+0x19/0x30 [ 36.105461][ T744] dev_set_rx_mode+0x19/0x30 [ 36.106532][ T744] dev_set_promiscuity+0x36/0x50 [ 36.107692][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.108929][ T744] dev_set_promiscuity+0x1e/0x50 [ 36.110093][ T744] br_port_set_promisc+0x1f/0x40 [bridge] [ 36.111415][ T744] br_manage_promisc+0x8b/0xe0 [bridge] [ 36.112728][ T744] __dev_set_promiscuity+0x123/0x1a0 [ 36.113967][ T744] ? __hw_addr_sync_one+0x23/0x50 [ 36.115135][ T744] __dev_set_rx_mode+0x68/0x90 [ 36.116249][ T744] dev_uc_sync+0x70/0x80 [ 36.117244][ T744] dev_uc_add+0x50/0x60 [ 36.118223][ T744] macvlan_open+0x18e/0x1f0 [macvlan] [ 36.119470][ T744] __dev_open+0xd6/0x170 [ 36.120470][ T744] __dev_change_flags+0x181/0x1d0 [ 36.121644][ T744] dev_change_flags+0x23/0x60 [ 36.122741][ T744] do_setlink+0x30a/0x11e0 [ 36.123778][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.124929][ T744] ? __nla_validate_parse.part.6+0x45/0x8e0 [ 36.126309][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.127457][ T744] __rtnl_newlink+0x546/0x8e0 [ 36.128560][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.129623][ T744] ? deactivate_slab.isra.85+0x6a1/0x850 [ 36.130946][ T744] ? __lock_acquire+0x92c/0x1f40 [ 36.132102][ T744] ? lock_acquire+0xb4/0x3b0 [ 36.133176][ T744] ? is_bpf_text_address+0x5/0xe0 [ 36.134364][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.135445][ T744] ? rcu_read_lock_sched_held+0x32/0x60 [ 36.136771][ T744] ? kmem_cache_alloc_trace+0x2d8/0x380 [ 36.138070][ T744] ? rtnl_newlink+0x2e/0x70 [ 36.139164][ T744] rtnl_newlink+0x47/0x70 [ ... ] Fixes: 845e0ebb4408 ("net: change addr_list_lock back to static key") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2020-09-25 18:13:29 +00:00
netif_addr_lock(to);
__hw_addr_unsync(&to->mc, &from->mc, to->addr_len);
__dev_set_rx_mode(to);
netif_addr_unlock(to);
netif_addr_unlock_bh(from);
}
EXPORT_SYMBOL(dev_mc_unsync);
/**
* dev_mc_flush - Flush multicast addresses
* @dev: device
*
* Flush multicast addresses.
*/
void dev_mc_flush(struct net_device *dev)
{
netif_addr_lock_bh(dev);
__hw_addr_flush(&dev->mc);
netif_addr_unlock_bh(dev);
}
EXPORT_SYMBOL(dev_mc_flush);
/**
* dev_mc_init - Init multicast address list
* @dev: device
*
* Init multicast address list.
*/
void dev_mc_init(struct net_device *dev)
{
__hw_addr_init(&dev->mc);
}
EXPORT_SYMBOL(dev_mc_init);