2019-05-24 21:43:04 +00:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
/* Generic nexthop implementation
|
|
|
|
*
|
|
|
|
* Copyright (c) 2017-19 Cumulus Networks
|
|
|
|
* Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com>
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/nexthop.h>
|
|
|
|
#include <linux/rtnetlink.h>
|
|
|
|
#include <linux/slab.h>
|
2021-12-29 00:49:13 +00:00
|
|
|
#include <linux/vmalloc.h>
|
2019-05-24 21:43:08 +00:00
|
|
|
#include <net/arp.h>
|
2019-05-24 21:43:06 +00:00
|
|
|
#include <net/ipv6_stubs.h>
|
2019-05-24 21:43:07 +00:00
|
|
|
#include <net/lwtunnel.h>
|
2019-05-24 21:43:08 +00:00
|
|
|
#include <net/ndisc.h>
|
2019-05-24 21:43:04 +00:00
|
|
|
#include <net/nexthop.h>
|
2019-05-24 21:43:05 +00:00
|
|
|
#include <net/route.h>
|
2019-05-24 21:43:04 +00:00
|
|
|
#include <net/sock.h>
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
#define NH_RES_DEFAULT_IDLE_TIMER (120 * HZ)
|
|
|
|
#define NH_RES_DEFAULT_UNBALANCED_TIMER 0 /* No forced rebalancing. */
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static void remove_nexthop(struct net *net, struct nexthop *nh,
|
|
|
|
struct nl_info *nlinfo);
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
#define NH_DEV_HASHBITS 8
|
|
|
|
#define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS)
|
|
|
|
|
2021-01-20 15:44:12 +00:00
|
|
|
static const struct nla_policy rtm_nh_policy_new[] = {
|
2019-05-24 21:43:04 +00:00
|
|
|
[NHA_ID] = { .type = NLA_U32 },
|
|
|
|
[NHA_GROUP] = { .type = NLA_BINARY },
|
|
|
|
[NHA_GROUP_TYPE] = { .type = NLA_U16 },
|
|
|
|
[NHA_BLACKHOLE] = { .type = NLA_FLAG },
|
|
|
|
[NHA_OIF] = { .type = NLA_U32 },
|
|
|
|
[NHA_GATEWAY] = { .type = NLA_BINARY },
|
|
|
|
[NHA_ENCAP_TYPE] = { .type = NLA_U16 },
|
|
|
|
[NHA_ENCAP] = { .type = NLA_NESTED },
|
2020-05-22 05:26:13 +00:00
|
|
|
[NHA_FDB] = { .type = NLA_FLAG },
|
2021-03-11 18:03:21 +00:00
|
|
|
[NHA_RES_GROUP] = { .type = NLA_NESTED },
|
2019-05-24 21:43:04 +00:00
|
|
|
};
|
|
|
|
|
2021-01-20 15:44:10 +00:00
|
|
|
static const struct nla_policy rtm_nh_policy_get[] = {
|
|
|
|
[NHA_ID] = { .type = NLA_U32 },
|
|
|
|
};
|
|
|
|
|
2021-01-20 15:44:11 +00:00
|
|
|
static const struct nla_policy rtm_nh_policy_dump[] = {
|
|
|
|
[NHA_OIF] = { .type = NLA_U32 },
|
|
|
|
[NHA_GROUPS] = { .type = NLA_FLAG },
|
|
|
|
[NHA_MASTER] = { .type = NLA_U32 },
|
|
|
|
[NHA_FDB] = { .type = NLA_FLAG },
|
|
|
|
};
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
static const struct nla_policy rtm_nh_res_policy_new[] = {
|
|
|
|
[NHA_RES_GROUP_BUCKETS] = { .type = NLA_U16 },
|
|
|
|
[NHA_RES_GROUP_IDLE_TIMER] = { .type = NLA_U32 },
|
|
|
|
[NHA_RES_GROUP_UNBALANCED_TIMER] = { .type = NLA_U32 },
|
|
|
|
};
|
|
|
|
|
2021-03-11 18:03:22 +00:00
|
|
|
static const struct nla_policy rtm_nh_policy_dump_bucket[] = {
|
|
|
|
[NHA_ID] = { .type = NLA_U32 },
|
|
|
|
[NHA_OIF] = { .type = NLA_U32 },
|
|
|
|
[NHA_MASTER] = { .type = NLA_U32 },
|
|
|
|
[NHA_RES_BUCKET] = { .type = NLA_NESTED },
|
|
|
|
};
|
|
|
|
|
|
|
|
static const struct nla_policy rtm_nh_res_bucket_policy_dump[] = {
|
|
|
|
[NHA_RES_BUCKET_NH_ID] = { .type = NLA_U32 },
|
|
|
|
};
|
|
|
|
|
2021-03-11 18:03:23 +00:00
|
|
|
static const struct nla_policy rtm_nh_policy_get_bucket[] = {
|
|
|
|
[NHA_ID] = { .type = NLA_U32 },
|
|
|
|
[NHA_RES_BUCKET] = { .type = NLA_NESTED },
|
|
|
|
};
|
|
|
|
|
|
|
|
static const struct nla_policy rtm_nh_res_bucket_policy_get[] = {
|
|
|
|
[NHA_RES_BUCKET_INDEX] = { .type = NLA_U16 },
|
|
|
|
};
|
|
|
|
|
2020-11-04 13:30:25 +00:00
|
|
|
static bool nexthop_notifiers_is_empty(struct net *net)
|
|
|
|
{
|
|
|
|
return !net->nexthop.notifier_chain.head;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
__nh_notifier_single_info_init(struct nh_notifier_single_info *nh_info,
|
2021-03-11 18:03:13 +00:00
|
|
|
const struct nh_info *nhi)
|
2020-11-04 13:30:25 +00:00
|
|
|
{
|
|
|
|
nh_info->dev = nhi->fib_nhc.nhc_dev;
|
|
|
|
nh_info->gw_family = nhi->fib_nhc.nhc_gw_family;
|
|
|
|
if (nh_info->gw_family == AF_INET)
|
|
|
|
nh_info->ipv4 = nhi->fib_nhc.nhc_gw.ipv4;
|
|
|
|
else if (nh_info->gw_family == AF_INET6)
|
|
|
|
nh_info->ipv6 = nhi->fib_nhc.nhc_gw.ipv6;
|
|
|
|
|
|
|
|
nh_info->is_reject = nhi->reject_nh;
|
|
|
|
nh_info->is_fdb = nhi->fdb_nh;
|
|
|
|
nh_info->has_encap = !!nhi->fib_nhc.nhc_lwtstate;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_notifier_single_info_init(struct nh_notifier_info *info,
|
|
|
|
const struct nexthop *nh)
|
|
|
|
{
|
2021-03-11 18:03:13 +00:00
|
|
|
struct nh_info *nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
|
2021-01-28 12:49:17 +00:00
|
|
|
info->type = NH_NOTIFIER_INFO_TYPE_SINGLE;
|
2020-11-04 13:30:25 +00:00
|
|
|
info->nh = kzalloc(sizeof(*info->nh), GFP_KERNEL);
|
|
|
|
if (!info->nh)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2021-03-11 18:03:13 +00:00
|
|
|
__nh_notifier_single_info_init(info->nh, nhi);
|
2020-11-04 13:30:25 +00:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_notifier_single_info_fini(struct nh_notifier_info *info)
|
|
|
|
{
|
|
|
|
kfree(info->nh);
|
|
|
|
}
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
static int nh_notifier_mpath_info_init(struct nh_notifier_info *info,
|
|
|
|
struct nh_group *nhg)
|
2020-11-04 13:30:25 +00:00
|
|
|
{
|
|
|
|
u16 num_nh = nhg->num_nh;
|
|
|
|
int i;
|
|
|
|
|
2021-01-28 12:49:17 +00:00
|
|
|
info->type = NH_NOTIFIER_INFO_TYPE_GRP;
|
2020-11-04 13:30:25 +00:00
|
|
|
info->nh_grp = kzalloc(struct_size(info->nh_grp, nh_entries, num_nh),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!info->nh_grp)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
info->nh_grp->num_nh = num_nh;
|
|
|
|
info->nh_grp->is_fdb = nhg->fdb_nh;
|
|
|
|
|
|
|
|
for (i = 0; i < num_nh; i++) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
2021-03-11 18:03:13 +00:00
|
|
|
struct nh_info *nhi;
|
2020-11-04 13:30:25 +00:00
|
|
|
|
2021-03-11 18:03:13 +00:00
|
|
|
nhi = rtnl_dereference(nhge->nh->nh_info);
|
2020-11-04 13:30:25 +00:00
|
|
|
info->nh_grp->nh_entries[i].id = nhge->nh->id;
|
|
|
|
info->nh_grp->nh_entries[i].weight = nhge->weight;
|
|
|
|
__nh_notifier_single_info_init(&info->nh_grp->nh_entries[i].nh,
|
2021-03-11 18:03:13 +00:00
|
|
|
nhi);
|
2020-11-04 13:30:25 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
static int nh_notifier_res_table_info_init(struct nh_notifier_info *info,
|
|
|
|
struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
u16 num_nh_buckets = res_table->num_nh_buckets;
|
|
|
|
unsigned long size;
|
|
|
|
u16 i;
|
|
|
|
|
|
|
|
info->type = NH_NOTIFIER_INFO_TYPE_RES_TABLE;
|
|
|
|
size = struct_size(info->nh_res_table, nhs, num_nh_buckets);
|
|
|
|
info->nh_res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO |
|
|
|
|
__GFP_NOWARN);
|
|
|
|
if (!info->nh_res_table)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
info->nh_res_table->num_nh_buckets = num_nh_buckets;
|
|
|
|
|
|
|
|
for (i = 0; i < num_nh_buckets; i++) {
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
nhge = rtnl_dereference(bucket->nh_entry);
|
|
|
|
nhi = rtnl_dereference(nhge->nh->nh_info);
|
|
|
|
__nh_notifier_single_info_init(&info->nh_res_table->nhs[i],
|
|
|
|
nhi);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:18 +00:00
|
|
|
static int nh_notifier_grp_info_init(struct nh_notifier_info *info,
|
|
|
|
const struct nexthop *nh)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold)
|
|
|
|
return nh_notifier_mpath_info_init(info, nhg);
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
else if (nhg->resilient)
|
|
|
|
return nh_notifier_res_table_info_init(info, nhg);
|
2021-01-28 12:49:18 +00:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_notifier_grp_info_fini(struct nh_notifier_info *info,
|
|
|
|
const struct nexthop *nh)
|
2020-11-04 13:30:25 +00:00
|
|
|
{
|
2021-01-28 12:49:18 +00:00
|
|
|
struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold)
|
2021-01-28 12:49:18 +00:00
|
|
|
kfree(info->nh_grp);
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
else if (nhg->resilient)
|
|
|
|
vfree(info->nh_res_table);
|
2020-11-04 13:30:25 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_notifier_info_init(struct nh_notifier_info *info,
|
|
|
|
const struct nexthop *nh)
|
|
|
|
{
|
|
|
|
info->id = nh->id;
|
|
|
|
|
2021-01-28 12:49:17 +00:00
|
|
|
if (nh->is_group)
|
2020-11-04 13:30:25 +00:00
|
|
|
return nh_notifier_grp_info_init(info, nh);
|
|
|
|
else
|
|
|
|
return nh_notifier_single_info_init(info, nh);
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:17 +00:00
|
|
|
static void nh_notifier_info_fini(struct nh_notifier_info *info,
|
|
|
|
const struct nexthop *nh)
|
2020-11-04 13:30:25 +00:00
|
|
|
{
|
2021-01-28 12:49:17 +00:00
|
|
|
if (nh->is_group)
|
2021-01-28 12:49:18 +00:00
|
|
|
nh_notifier_grp_info_fini(info, nh);
|
2020-11-04 13:30:25 +00:00
|
|
|
else
|
|
|
|
nh_notifier_single_info_fini(info);
|
|
|
|
}
|
|
|
|
|
2020-05-22 05:26:15 +00:00
|
|
|
static int call_nexthop_notifiers(struct net *net,
|
2020-05-27 08:00:20 +00:00
|
|
|
enum nexthop_event_type event_type,
|
2020-11-04 13:30:24 +00:00
|
|
|
struct nexthop *nh,
|
|
|
|
struct netlink_ext_ack *extack)
|
2020-05-22 05:26:15 +00:00
|
|
|
{
|
2020-11-04 13:30:25 +00:00
|
|
|
struct nh_notifier_info info = {
|
|
|
|
.net = net,
|
|
|
|
.extack = extack,
|
|
|
|
};
|
2020-05-22 05:26:15 +00:00
|
|
|
int err;
|
|
|
|
|
2020-11-04 13:30:25 +00:00
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (nexthop_notifiers_is_empty(net))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err = nh_notifier_info_init(&info, nh);
|
|
|
|
if (err) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2020-09-15 11:41:01 +00:00
|
|
|
err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
|
2020-11-04 13:30:26 +00:00
|
|
|
event_type, &info);
|
2021-01-28 12:49:17 +00:00
|
|
|
nh_notifier_info_fini(&info, nh);
|
2020-11-04 13:30:25 +00:00
|
|
|
|
2020-05-22 05:26:15 +00:00
|
|
|
return notifier_to_errno(err);
|
|
|
|
}
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
static int
|
|
|
|
nh_notifier_res_bucket_idle_timer_get(const struct nh_notifier_info *info,
|
|
|
|
bool force, unsigned int *p_idle_timer_ms)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
/* When 'force' is false, nexthop bucket replacement is performed
|
|
|
|
* because the bucket was deemed to be idle. In this case, capable
|
|
|
|
* listeners can choose to perform an atomic replacement: The bucket is
|
|
|
|
* only replaced if it is inactive. However, if the idle timer interval
|
|
|
|
* is smaller than the interval in which a listener is querying
|
|
|
|
* buckets' activity from the device, then atomic replacement should
|
|
|
|
* not be tried. Pass the idle timer value to listeners, so that they
|
|
|
|
* could determine which type of replacement to perform.
|
|
|
|
*/
|
|
|
|
if (force) {
|
|
|
|
*p_idle_timer_ms = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
nh = nexthop_find_by_id(info->net, info->id);
|
|
|
|
if (!nh) {
|
|
|
|
err = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
nhg = rcu_dereference(nh->nh_grp);
|
|
|
|
res_table = rcu_dereference(nhg->res_table);
|
|
|
|
*p_idle_timer_ms = jiffies_to_msecs(res_table->idle_timer);
|
|
|
|
|
|
|
|
out:
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_notifier_res_bucket_info_init(struct nh_notifier_info *info,
|
|
|
|
u16 bucket_index, bool force,
|
|
|
|
struct nh_info *oldi,
|
|
|
|
struct nh_info *newi)
|
|
|
|
{
|
|
|
|
unsigned int idle_timer_ms;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nh_notifier_res_bucket_idle_timer_get(info, force,
|
|
|
|
&idle_timer_ms);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
info->type = NH_NOTIFIER_INFO_TYPE_RES_BUCKET;
|
|
|
|
info->nh_res_bucket = kzalloc(sizeof(*info->nh_res_bucket),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!info->nh_res_bucket)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
info->nh_res_bucket->bucket_index = bucket_index;
|
|
|
|
info->nh_res_bucket->idle_timer_ms = idle_timer_ms;
|
|
|
|
info->nh_res_bucket->force = force;
|
|
|
|
__nh_notifier_single_info_init(&info->nh_res_bucket->old_nh, oldi);
|
|
|
|
__nh_notifier_single_info_init(&info->nh_res_bucket->new_nh, newi);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_notifier_res_bucket_info_fini(struct nh_notifier_info *info)
|
|
|
|
{
|
|
|
|
kfree(info->nh_res_bucket);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id,
|
|
|
|
u16 bucket_index, bool force,
|
|
|
|
struct nh_info *oldi,
|
|
|
|
struct nh_info *newi,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_notifier_info info = {
|
|
|
|
.net = net,
|
|
|
|
.extack = extack,
|
|
|
|
.id = nhg_id,
|
|
|
|
};
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (nexthop_notifiers_is_empty(net))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err = nh_notifier_res_bucket_info_init(&info, bucket_index, force,
|
|
|
|
oldi, newi);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
|
|
|
|
NEXTHOP_EVENT_BUCKET_REPLACE, &info);
|
|
|
|
nh_notifier_res_bucket_info_fini(&info);
|
|
|
|
|
|
|
|
return notifier_to_errno(err);
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
/* There are three users of RES_TABLE, and NHs etc. referenced from there:
|
|
|
|
*
|
|
|
|
* 1) a collection of callbacks for NH maintenance. This operates under
|
|
|
|
* RTNL,
|
|
|
|
* 2) the delayed work that gradually balances the resilient table,
|
|
|
|
* 3) and nexthop_select_path(), operating under RCU.
|
|
|
|
*
|
|
|
|
* Both the delayed work and the RTNL block are writers, and need to
|
|
|
|
* maintain mutual exclusion. Since there are only two and well-known
|
|
|
|
* writers for each table, the RTNL code can make sure it has exclusive
|
|
|
|
* access thus:
|
|
|
|
*
|
|
|
|
* - Have the DW operate without locking;
|
|
|
|
* - synchronously cancel the DW;
|
|
|
|
* - do the writing;
|
|
|
|
* - if the write was not actually a delete, call upkeep, which schedules
|
|
|
|
* DW again if necessary.
|
|
|
|
*
|
|
|
|
* The functions that are always called from the RTNL context use
|
|
|
|
* rtnl_dereference(). The functions that can also be called from the DW do
|
|
|
|
* a raw dereference and rely on the above mutual exclusion scheme.
|
|
|
|
*/
|
|
|
|
#define nh_res_dereference(p) (rcu_dereference_raw(p))
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
static int call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id,
|
|
|
|
u16 bucket_index, bool force,
|
|
|
|
struct nexthop *old_nh,
|
|
|
|
struct nexthop *new_nh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_info *oldi = nh_res_dereference(old_nh->nh_info);
|
|
|
|
struct nh_info *newi = nh_res_dereference(new_nh->nh_info);
|
|
|
|
|
|
|
|
return __call_nexthop_res_bucket_notifiers(net, nhg_id, bucket_index,
|
|
|
|
force, oldi, newi, extack);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int call_nexthop_res_table_notifiers(struct net *net, struct nexthop *nh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_notifier_info info = {
|
|
|
|
.net = net,
|
|
|
|
.extack = extack,
|
|
|
|
};
|
|
|
|
struct nh_group *nhg;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
if (nexthop_notifiers_is_empty(net))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* At this point, the nexthop buckets are still not populated. Only
|
|
|
|
* emit a notification with the logical nexthops, so that a listener
|
|
|
|
* could potentially veto it in case of unsupported configuration.
|
|
|
|
*/
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
2021-03-26 13:20:22 +00:00
|
|
|
err = nh_notifier_mpath_info_init(&info, nhg);
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
if (err) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
|
|
|
|
NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE,
|
|
|
|
&info);
|
|
|
|
kfree(info.nh_grp);
|
|
|
|
|
|
|
|
return notifier_to_errno(err);
|
|
|
|
}
|
|
|
|
|
2020-11-04 13:30:35 +00:00
|
|
|
static int call_nexthop_notifier(struct notifier_block *nb, struct net *net,
|
|
|
|
enum nexthop_event_type event_type,
|
|
|
|
struct nexthop *nh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_notifier_info info = {
|
|
|
|
.net = net,
|
|
|
|
.extack = extack,
|
|
|
|
};
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nh_notifier_info_init(&info, nh);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = nb->notifier_call(nb, event_type, &info);
|
2021-01-28 12:49:17 +00:00
|
|
|
nh_notifier_info_fini(&info, nh);
|
2020-11-04 13:30:35 +00:00
|
|
|
|
|
|
|
return notifier_to_errno(err);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
static unsigned int nh_dev_hashfn(unsigned int val)
|
|
|
|
{
|
|
|
|
unsigned int mask = NH_DEV_HASHSIZE - 1;
|
|
|
|
|
|
|
|
return (val ^
|
|
|
|
(val >> NH_DEV_HASHBITS) ^
|
|
|
|
(val >> (NH_DEV_HASHBITS * 2))) & mask;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
|
|
|
|
{
|
|
|
|
struct net_device *dev = nhi->fib_nhc.nhc_dev;
|
|
|
|
struct hlist_head *head;
|
|
|
|
unsigned int hash;
|
|
|
|
|
|
|
|
WARN_ON(!dev);
|
|
|
|
|
|
|
|
hash = nh_dev_hashfn(dev->ifindex);
|
|
|
|
head = &net->nexthop.devhash[hash];
|
|
|
|
hlist_add_head(&nhi->dev_hash, head);
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:13 +00:00
|
|
|
static void nexthop_free_group(struct nexthop *nh)
|
2019-05-24 21:43:08 +00:00
|
|
|
{
|
|
|
|
struct nh_group *nhg;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
nhg = rcu_dereference_raw(nh->nh_grp);
|
2020-05-26 18:56:15 +00:00
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
WARN_ON(!list_empty(&nhge->nh_list));
|
|
|
|
nexthop_put(nhge->nh);
|
|
|
|
}
|
|
|
|
|
|
|
|
WARN_ON(nhg->spare == nhg);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (nhg->resilient)
|
|
|
|
vfree(rcu_dereference_raw(nhg->res_table));
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
kfree(nhg->spare);
|
2019-05-24 21:43:08 +00:00
|
|
|
kfree(nhg);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nexthop_free_single(struct nexthop *nh)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
nhi = rcu_dereference_raw(nh->nh_info);
|
2019-05-24 21:43:05 +00:00
|
|
|
switch (nhi->family) {
|
|
|
|
case AF_INET:
|
|
|
|
fib_nh_release(nh->net, &nhi->fib_nh);
|
|
|
|
break;
|
2019-05-24 21:43:06 +00:00
|
|
|
case AF_INET6:
|
|
|
|
ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
|
|
|
|
break;
|
2019-05-24 21:43:05 +00:00
|
|
|
}
|
2019-05-24 21:43:04 +00:00
|
|
|
kfree(nhi);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void nexthop_free_rcu(struct rcu_head *head)
|
|
|
|
{
|
|
|
|
struct nexthop *nh = container_of(head, struct nexthop, rcu);
|
|
|
|
|
|
|
|
if (nh->is_group)
|
2021-01-28 12:49:13 +00:00
|
|
|
nexthop_free_group(nh);
|
2019-05-24 21:43:08 +00:00
|
|
|
else
|
|
|
|
nexthop_free_single(nh);
|
2019-05-24 21:43:04 +00:00
|
|
|
|
|
|
|
kfree(nh);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(nexthop_free_rcu);
|
|
|
|
|
|
|
|
static struct nexthop *nexthop_alloc(void)
|
|
|
|
{
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nh) {
|
2019-06-04 03:19:51 +00:00
|
|
|
INIT_LIST_HEAD(&nh->fi_list);
|
2019-06-04 03:19:52 +00:00
|
|
|
INIT_LIST_HEAD(&nh->f6i_list);
|
2019-05-24 21:43:08 +00:00
|
|
|
INIT_LIST_HEAD(&nh->grp_list);
|
2020-05-22 05:26:13 +00:00
|
|
|
INIT_LIST_HEAD(&nh->fdb_list);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
2019-05-24 21:43:04 +00:00
|
|
|
return nh;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static struct nh_group *nexthop_grp_alloc(u16 num_nh)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
2020-08-26 16:48:51 +00:00
|
|
|
nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL);
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nhg)
|
|
|
|
nhg->num_nh = num_nh;
|
|
|
|
|
|
|
|
return nhg;
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
static void nh_res_table_upkeep_dw(struct work_struct *work);
|
|
|
|
|
|
|
|
static struct nh_res_table *
|
|
|
|
nexthop_res_table_alloc(struct net *net, u32 nhg_id, struct nh_config *cfg)
|
|
|
|
{
|
|
|
|
const u16 num_nh_buckets = cfg->nh_grp_res_num_buckets;
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
unsigned long size;
|
|
|
|
|
|
|
|
size = struct_size(res_table, nh_buckets, num_nh_buckets);
|
|
|
|
res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN);
|
|
|
|
if (!res_table)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
res_table->net = net;
|
|
|
|
res_table->nhg_id = nhg_id;
|
|
|
|
INIT_DELAYED_WORK(&res_table->upkeep_dw, &nh_res_table_upkeep_dw);
|
|
|
|
INIT_LIST_HEAD(&res_table->uw_nh_entries);
|
|
|
|
res_table->idle_timer = cfg->nh_grp_res_idle_timer;
|
|
|
|
res_table->unbalanced_timer = cfg->nh_grp_res_unbalanced_timer;
|
|
|
|
res_table->num_nh_buckets = num_nh_buckets;
|
|
|
|
return res_table;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static void nh_base_seq_inc(struct net *net)
|
|
|
|
{
|
|
|
|
while (++net->nexthop.seq == 0)
|
|
|
|
;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* no reference taken; rcu lock or rtnl must be held */
|
|
|
|
struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
|
|
|
|
{
|
|
|
|
struct rb_node **pp, *parent = NULL, *next;
|
|
|
|
|
|
|
|
pp = &net->nexthop.rb_root.rb_node;
|
|
|
|
while (1) {
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
next = rcu_dereference_raw(*pp);
|
|
|
|
if (!next)
|
|
|
|
break;
|
|
|
|
parent = next;
|
|
|
|
|
|
|
|
nh = rb_entry(parent, struct nexthop, rb_node);
|
|
|
|
if (id < nh->id)
|
|
|
|
pp = &next->rb_left;
|
|
|
|
else if (id > nh->id)
|
|
|
|
pp = &next->rb_right;
|
|
|
|
else
|
|
|
|
return nh;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(nexthop_find_by_id);
|
|
|
|
|
|
|
|
/* used for auto id allocation; called with rtnl held */
|
|
|
|
static u32 nh_find_unused_id(struct net *net)
|
|
|
|
{
|
|
|
|
u32 id_start = net->nexthop.last_id_allocated;
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
net->nexthop.last_id_allocated++;
|
|
|
|
if (net->nexthop.last_id_allocated == id_start)
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
|
|
|
|
return net->nexthop.last_id_allocated;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
static void nh_res_time_set_deadline(unsigned long next_time,
|
|
|
|
unsigned long *deadline)
|
|
|
|
{
|
|
|
|
if (time_before(next_time, *deadline))
|
|
|
|
*deadline = next_time;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
static clock_t nh_res_table_unbalanced_time(struct nh_res_table *res_table)
|
|
|
|
{
|
|
|
|
if (list_empty(&res_table->uw_nh_entries))
|
|
|
|
return 0;
|
|
|
|
return jiffies_delta_to_clock_t(jiffies - res_table->unbalanced_since);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
struct nlattr *nest;
|
|
|
|
|
|
|
|
nest = nla_nest_start(skb, NHA_RES_GROUP);
|
|
|
|
if (!nest)
|
|
|
|
return -EMSGSIZE;
|
|
|
|
|
|
|
|
if (nla_put_u16(skb, NHA_RES_GROUP_BUCKETS,
|
|
|
|
res_table->num_nh_buckets) ||
|
|
|
|
nla_put_u32(skb, NHA_RES_GROUP_IDLE_TIMER,
|
|
|
|
jiffies_to_clock_t(res_table->idle_timer)) ||
|
|
|
|
nla_put_u32(skb, NHA_RES_GROUP_UNBALANCED_TIMER,
|
|
|
|
jiffies_to_clock_t(res_table->unbalanced_timer)) ||
|
|
|
|
nla_put_u64_64bit(skb, NHA_RES_GROUP_UNBALANCED_TIME,
|
|
|
|
nh_res_table_unbalanced_time(res_table),
|
|
|
|
NHA_RES_GROUP_PAD))
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
|
|
|
nla_nest_end(skb, nest);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nla_put_failure:
|
|
|
|
nla_nest_cancel(skb, nest);
|
|
|
|
return -EMSGSIZE;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
struct nexthop_grp *p;
|
|
|
|
size_t len = nhg->num_nh * sizeof(*p);
|
|
|
|
struct nlattr *nla;
|
|
|
|
u16 group_type = 0;
|
|
|
|
int i;
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold)
|
2019-05-24 21:43:08 +00:00
|
|
|
group_type = NEXTHOP_GRP_TYPE_MPATH;
|
2021-03-11 18:03:21 +00:00
|
|
|
else if (nhg->resilient)
|
|
|
|
group_type = NEXTHOP_GRP_TYPE_RES;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
|
|
|
nla = nla_reserve(skb, NHA_GROUP, len);
|
|
|
|
if (!nla)
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
|
|
|
p = nla_data(nla);
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
p->id = nhg->nh_entries[i].nh->id;
|
|
|
|
p->weight = nhg->nh_entries[i].weight - 1;
|
|
|
|
p += 1;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
if (nhg->resilient && nla_put_nh_group_res(skb, nhg))
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
nla_put_failure:
|
|
|
|
return -EMSGSIZE;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
|
|
|
|
int event, u32 portid, u32 seq, unsigned int nlflags)
|
|
|
|
{
|
2019-05-24 21:43:06 +00:00
|
|
|
struct fib6_nh *fib6_nh;
|
2019-05-24 21:43:05 +00:00
|
|
|
struct fib_nh *fib_nh;
|
2019-05-24 21:43:04 +00:00
|
|
|
struct nlmsghdr *nlh;
|
|
|
|
struct nh_info *nhi;
|
|
|
|
struct nhmsg *nhm;
|
|
|
|
|
|
|
|
nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
|
|
|
|
if (!nlh)
|
|
|
|
return -EMSGSIZE;
|
|
|
|
|
|
|
|
nhm = nlmsg_data(nlh);
|
|
|
|
nhm->nh_family = AF_UNSPEC;
|
|
|
|
nhm->nh_flags = nh->nh_flags;
|
|
|
|
nhm->nh_protocol = nh->protocol;
|
|
|
|
nhm->nh_scope = 0;
|
|
|
|
nhm->resvd = 0;
|
|
|
|
|
|
|
|
if (nla_put_u32(skb, NHA_ID, nh->id))
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nh->is_group) {
|
|
|
|
struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB))
|
|
|
|
goto nla_put_failure;
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nla_put_nh_group(skb, nhg))
|
|
|
|
goto nla_put_failure;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
nhm->nh_family = nhi->family;
|
|
|
|
if (nhi->reject_nh) {
|
|
|
|
if (nla_put_flag(skb, NHA_BLACKHOLE))
|
|
|
|
goto nla_put_failure;
|
|
|
|
goto out;
|
2020-06-09 02:54:43 +00:00
|
|
|
} else if (nhi->fdb_nh) {
|
|
|
|
if (nla_put_flag(skb, NHA_FDB))
|
|
|
|
goto nla_put_failure;
|
|
|
|
} else {
|
2019-05-24 21:43:05 +00:00
|
|
|
const struct net_device *dev;
|
|
|
|
|
|
|
|
dev = nhi->fib_nhc.nhc_dev;
|
|
|
|
if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
|
|
|
|
goto nla_put_failure;
|
|
|
|
}
|
|
|
|
|
|
|
|
nhm->nh_scope = nhi->fib_nhc.nhc_scope;
|
|
|
|
switch (nhi->family) {
|
|
|
|
case AF_INET:
|
|
|
|
fib_nh = &nhi->fib_nh;
|
|
|
|
if (fib_nh->fib_nh_gw_family &&
|
2020-08-26 16:48:52 +00:00
|
|
|
nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
|
2019-05-24 21:43:05 +00:00
|
|
|
goto nla_put_failure;
|
|
|
|
break;
|
2019-05-24 21:43:06 +00:00
|
|
|
|
|
|
|
case AF_INET6:
|
|
|
|
fib6_nh = &nhi->fib6_nh;
|
|
|
|
if (fib6_nh->fib_nh_gw_family &&
|
|
|
|
nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
|
|
|
|
goto nla_put_failure;
|
|
|
|
break;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:07 +00:00
|
|
|
if (nhi->fib_nhc.nhc_lwtstate &&
|
|
|
|
lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
|
|
|
|
NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
out:
|
|
|
|
nlmsg_end(skb, nlh);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nla_put_failure:
|
2020-05-20 01:57:12 +00:00
|
|
|
nlmsg_cancel(skb, nlh);
|
2019-05-24 21:43:04 +00:00
|
|
|
return -EMSGSIZE;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
static size_t nh_nlmsg_size_grp_res(struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
return nla_total_size(0) + /* NHA_RES_GROUP */
|
|
|
|
nla_total_size(2) + /* NHA_RES_GROUP_BUCKETS */
|
|
|
|
nla_total_size(4) + /* NHA_RES_GROUP_IDLE_TIMER */
|
|
|
|
nla_total_size(4) + /* NHA_RES_GROUP_UNBALANCED_TIMER */
|
|
|
|
nla_total_size_64bit(8);/* NHA_RES_GROUP_UNBALANCED_TIME */
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static size_t nh_nlmsg_size_grp(struct nexthop *nh)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
|
2021-03-11 18:03:21 +00:00
|
|
|
size_t tot = nla_total_size(sz) +
|
|
|
|
nla_total_size(2); /* NHA_GROUP_TYPE */
|
|
|
|
|
|
|
|
if (nhg->resilient)
|
|
|
|
tot += nh_nlmsg_size_grp_res(nhg);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
return tot;
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static size_t nh_nlmsg_size_single(struct nexthop *nh)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
2019-05-24 21:43:05 +00:00
|
|
|
struct nh_info *nhi = rtnl_dereference(nh->nh_info);
|
2019-05-24 21:43:08 +00:00
|
|
|
size_t sz;
|
2019-05-24 21:43:04 +00:00
|
|
|
|
|
|
|
/* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
|
|
|
|
* are mutually exclusive
|
|
|
|
*/
|
2019-05-24 21:43:08 +00:00
|
|
|
sz = nla_total_size(4); /* NHA_OIF */
|
2019-05-24 21:43:04 +00:00
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
switch (nhi->family) {
|
|
|
|
case AF_INET:
|
|
|
|
if (nhi->fib_nh.fib_nh_gw_family)
|
|
|
|
sz += nla_total_size(4); /* NHA_GATEWAY */
|
|
|
|
break;
|
2019-05-24 21:43:06 +00:00
|
|
|
|
|
|
|
case AF_INET6:
|
|
|
|
/* NHA_GATEWAY */
|
|
|
|
if (nhi->fib6_nh.fib_nh_gw_family)
|
|
|
|
sz += nla_total_size(sizeof(const struct in6_addr));
|
|
|
|
break;
|
2019-05-24 21:43:05 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:07 +00:00
|
|
|
if (nhi->fib_nhc.nhc_lwtstate) {
|
|
|
|
sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
|
|
|
|
sz += nla_total_size(2); /* NHA_ENCAP_TYPE */
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
return sz;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static size_t nh_nlmsg_size(struct nexthop *nh)
|
|
|
|
{
|
2020-01-24 21:53:27 +00:00
|
|
|
size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
|
|
|
|
|
|
|
|
sz += nla_total_size(4); /* NHA_ID */
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
if (nh->is_group)
|
|
|
|
sz += nh_nlmsg_size_grp(nh);
|
|
|
|
else
|
|
|
|
sz += nh_nlmsg_size_single(nh);
|
|
|
|
|
|
|
|
return sz;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
|
|
|
|
{
|
|
|
|
unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
|
|
|
|
u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
|
|
|
|
struct sk_buff *skb;
|
|
|
|
int err = -ENOBUFS;
|
|
|
|
|
|
|
|
skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
|
|
|
|
if (!skb)
|
|
|
|
goto errout;
|
|
|
|
|
|
|
|
err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
|
|
|
|
if (err < 0) {
|
|
|
|
/* -EMSGSIZE implies BUG in nh_nlmsg_size() */
|
|
|
|
WARN_ON(err == -EMSGSIZE);
|
|
|
|
kfree_skb(skb);
|
|
|
|
goto errout;
|
|
|
|
}
|
|
|
|
|
|
|
|
rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
|
|
|
|
info->nlh, gfp_any());
|
|
|
|
return;
|
|
|
|
errout:
|
|
|
|
if (err < 0)
|
|
|
|
rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
static unsigned long nh_res_bucket_used_time(const struct nh_res_bucket *bucket)
|
|
|
|
{
|
|
|
|
return (unsigned long)atomic_long_read(&bucket->used_time);
|
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned long
|
|
|
|
nh_res_bucket_idle_point(const struct nh_res_table *res_table,
|
|
|
|
const struct nh_res_bucket *bucket,
|
|
|
|
unsigned long now)
|
|
|
|
{
|
|
|
|
unsigned long time = nh_res_bucket_used_time(bucket);
|
|
|
|
|
|
|
|
/* Bucket was not used since it was migrated. The idle time is now. */
|
|
|
|
if (time == bucket->migrated_time)
|
|
|
|
return now;
|
|
|
|
|
|
|
|
return time + res_table->idle_timer;
|
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned long
|
|
|
|
nh_res_table_unb_point(const struct nh_res_table *res_table)
|
|
|
|
{
|
|
|
|
return res_table->unbalanced_since + res_table->unbalanced_timer;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_bucket_set_idle(const struct nh_res_table *res_table,
|
|
|
|
struct nh_res_bucket *bucket)
|
|
|
|
{
|
|
|
|
unsigned long now = jiffies;
|
|
|
|
|
|
|
|
atomic_long_set(&bucket->used_time, (long)now);
|
|
|
|
bucket->migrated_time = now;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_bucket_set_busy(struct nh_res_bucket *bucket)
|
|
|
|
{
|
|
|
|
atomic_long_set(&bucket->used_time, (long)jiffies);
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:22 +00:00
|
|
|
static clock_t nh_res_bucket_idle_time(const struct nh_res_bucket *bucket)
|
|
|
|
{
|
|
|
|
unsigned long used_time = nh_res_bucket_used_time(bucket);
|
|
|
|
|
|
|
|
return jiffies_delta_to_clock_t(jiffies - used_time);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_fill_res_bucket(struct sk_buff *skb, struct nexthop *nh,
|
|
|
|
struct nh_res_bucket *bucket, u16 bucket_index,
|
|
|
|
int event, u32 portid, u32 seq,
|
|
|
|
unsigned int nlflags,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry);
|
|
|
|
struct nlmsghdr *nlh;
|
|
|
|
struct nlattr *nest;
|
|
|
|
struct nhmsg *nhm;
|
|
|
|
|
|
|
|
nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
|
|
|
|
if (!nlh)
|
|
|
|
return -EMSGSIZE;
|
|
|
|
|
|
|
|
nhm = nlmsg_data(nlh);
|
|
|
|
nhm->nh_family = AF_UNSPEC;
|
|
|
|
nhm->nh_flags = bucket->nh_flags;
|
|
|
|
nhm->nh_protocol = nh->protocol;
|
|
|
|
nhm->nh_scope = 0;
|
|
|
|
nhm->resvd = 0;
|
|
|
|
|
|
|
|
if (nla_put_u32(skb, NHA_ID, nh->id))
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
|
|
|
nest = nla_nest_start(skb, NHA_RES_BUCKET);
|
|
|
|
if (!nest)
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
|
|
|
if (nla_put_u16(skb, NHA_RES_BUCKET_INDEX, bucket_index) ||
|
|
|
|
nla_put_u32(skb, NHA_RES_BUCKET_NH_ID, nhge->nh->id) ||
|
|
|
|
nla_put_u64_64bit(skb, NHA_RES_BUCKET_IDLE_TIME,
|
|
|
|
nh_res_bucket_idle_time(bucket),
|
|
|
|
NHA_RES_BUCKET_PAD))
|
|
|
|
goto nla_put_failure_nest;
|
|
|
|
|
|
|
|
nla_nest_end(skb, nest);
|
|
|
|
nlmsg_end(skb, nlh);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nla_put_failure_nest:
|
|
|
|
nla_nest_cancel(skb, nest);
|
|
|
|
nla_put_failure:
|
|
|
|
nlmsg_cancel(skb, nlh);
|
|
|
|
return -EMSGSIZE;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:24 +00:00
|
|
|
static void nexthop_bucket_notify(struct nh_res_table *res_table,
|
|
|
|
u16 bucket_index)
|
|
|
|
{
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index];
|
|
|
|
struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry);
|
|
|
|
struct nexthop *nh = nhge->nh_parent;
|
|
|
|
struct sk_buff *skb;
|
|
|
|
int err = -ENOBUFS;
|
|
|
|
|
|
|
|
skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
|
|
|
if (!skb)
|
|
|
|
goto errout;
|
|
|
|
|
|
|
|
err = nh_fill_res_bucket(skb, nh, bucket, bucket_index,
|
|
|
|
RTM_NEWNEXTHOPBUCKET, 0, 0, NLM_F_REPLACE,
|
|
|
|
NULL);
|
|
|
|
if (err < 0) {
|
|
|
|
kfree_skb(skb);
|
|
|
|
goto errout;
|
|
|
|
}
|
|
|
|
|
|
|
|
rtnl_notify(skb, nh->net, 0, RTNLGRP_NEXTHOP, NULL, GFP_KERNEL);
|
|
|
|
return;
|
|
|
|
errout:
|
|
|
|
if (err < 0)
|
|
|
|
rtnl_set_sk_err(nh->net, RTNLGRP_NEXTHOP, err);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
|
2020-06-09 02:54:43 +00:00
|
|
|
bool *is_fdb, struct netlink_ext_ack *extack)
|
2019-05-24 21:43:05 +00:00
|
|
|
{
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nh->is_group) {
|
|
|
|
struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
|
2019-05-24 21:43:05 +00:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
/* Nesting groups within groups is not supported. */
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold) {
|
2019-05-24 21:43:08 +00:00
|
|
|
NL_SET_ERR_MSG(extack,
|
2021-03-26 13:20:22 +00:00
|
|
|
"Hash-threshold group can not be a nexthop within a group");
|
2019-05-24 21:43:08 +00:00
|
|
|
return false;
|
|
|
|
}
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (nhg->resilient) {
|
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"Resilient group can not be a nexthop within a group");
|
|
|
|
return false;
|
|
|
|
}
|
2020-06-09 02:54:43 +00:00
|
|
|
*is_fdb = nhg->fdb_nh;
|
2019-05-24 21:43:08 +00:00
|
|
|
} else {
|
|
|
|
struct nh_info *nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
|
|
|
|
if (nhi->reject_nh && npaths > 1) {
|
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"Blackhole nexthop can not be used in a group with more than 1 path");
|
|
|
|
return false;
|
|
|
|
}
|
2020-06-09 02:54:43 +00:00
|
|
|
*is_fdb = nhi->fdb_nh;
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
|
|
|
|
if (!nhi->fdb_nh) {
|
2020-05-22 05:26:13 +00:00
|
|
|
NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (*nh_family == AF_UNSPEC) {
|
|
|
|
*nh_family = nhi->family;
|
|
|
|
} else if (*nh_family != nhi->family) {
|
|
|
|
NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-01-20 15:44:12 +00:00
|
|
|
static int nh_check_attr_group(struct net *net,
|
|
|
|
struct nlattr *tb[], size_t tb_size,
|
2021-03-11 18:03:21 +00:00
|
|
|
u16 nh_grp_type, struct netlink_ext_ack *extack)
|
2019-05-24 21:43:08 +00:00
|
|
|
{
|
|
|
|
unsigned int len = nla_len(tb[NHA_GROUP]);
|
2020-05-22 05:26:13 +00:00
|
|
|
u8 nh_family = AF_UNSPEC;
|
2019-05-24 21:43:08 +00:00
|
|
|
struct nexthop_grp *nhg;
|
|
|
|
unsigned int i, j;
|
2020-05-22 05:26:13 +00:00
|
|
|
u8 nhg_fdb = 0;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-08-22 12:06:36 +00:00
|
|
|
if (!len || len & (sizeof(struct nexthop_grp) - 1)) {
|
2019-05-24 21:43:08 +00:00
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"Invalid length for nexthop group attribute");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* convert len to number of nexthop ids */
|
|
|
|
len /= sizeof(*nhg);
|
|
|
|
|
|
|
|
nhg = nla_data(tb[NHA_GROUP]);
|
|
|
|
for (i = 0; i < len; ++i) {
|
|
|
|
if (nhg[i].resvd1 || nhg[i].resvd2) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
if (nhg[i].weight > 254) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid value for weight");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
for (j = i + 1; j < len; ++j) {
|
|
|
|
if (nhg[i].id == nhg[j].id) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (tb[NHA_FDB])
|
|
|
|
nhg_fdb = 1;
|
2019-05-24 21:43:08 +00:00
|
|
|
nhg = nla_data(tb[NHA_GROUP]);
|
|
|
|
for (i = 0; i < len; ++i) {
|
|
|
|
struct nexthop *nh;
|
2020-06-09 02:54:43 +00:00
|
|
|
bool is_fdb_nh;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
nh = nexthop_find_by_id(net, nhg[i].id);
|
|
|
|
if (!nh) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid nexthop id");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2020-06-09 02:54:43 +00:00
|
|
|
if (!valid_group_nh(nh, len, &is_fdb_nh, extack))
|
2019-05-24 21:43:08 +00:00
|
|
|
return -EINVAL;
|
2020-05-22 05:26:13 +00:00
|
|
|
|
|
|
|
if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
if (!nhg_fdb && is_fdb_nh) {
|
2020-05-22 05:26:13 +00:00
|
|
|
NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
2021-01-20 15:44:12 +00:00
|
|
|
for (i = NHA_GROUP_TYPE + 1; i < tb_size; ++i) {
|
2019-05-24 21:43:08 +00:00
|
|
|
if (!tb[i])
|
|
|
|
continue;
|
2021-03-11 18:03:21 +00:00
|
|
|
switch (i) {
|
|
|
|
case NHA_FDB:
|
2020-05-22 05:26:13 +00:00
|
|
|
continue;
|
2021-03-11 18:03:21 +00:00
|
|
|
case NHA_RES_GROUP:
|
|
|
|
if (nh_grp_type == NEXTHOP_GRP_TYPE_RES)
|
|
|
|
continue;
|
|
|
|
break;
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"No other attributes can be set in nexthop groups");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool ipv6_good_nh(const struct fib6_nh *nh)
|
|
|
|
{
|
|
|
|
int state = NUD_REACHABLE;
|
|
|
|
struct neighbour *n;
|
|
|
|
|
|
|
|
rcu_read_lock_bh();
|
|
|
|
|
|
|
|
n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
|
|
|
|
if (n)
|
|
|
|
state = n->nud_state;
|
|
|
|
|
|
|
|
rcu_read_unlock_bh();
|
|
|
|
|
|
|
|
return !!(state & NUD_VALID);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool ipv4_good_nh(const struct fib_nh *nh)
|
|
|
|
{
|
|
|
|
int state = NUD_REACHABLE;
|
|
|
|
struct neighbour *n;
|
|
|
|
|
|
|
|
rcu_read_lock_bh();
|
|
|
|
|
|
|
|
n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
|
|
|
|
(__force u32)nh->fib_nh_gw4);
|
|
|
|
if (n)
|
|
|
|
state = n->nud_state;
|
|
|
|
|
|
|
|
rcu_read_unlock_bh();
|
|
|
|
|
|
|
|
return !!(state & NUD_VALID);
|
|
|
|
}
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash)
|
2019-05-24 21:43:08 +00:00
|
|
|
{
|
|
|
|
struct nexthop *rc = NULL;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (hash > atomic_read(&nhge->hthr.upper_bound))
|
2019-05-24 21:43:08 +00:00
|
|
|
continue;
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
nhi = rcu_dereference(nhge->nh->nh_info);
|
|
|
|
if (nhi->fdb_nh)
|
2020-05-22 05:26:13 +00:00
|
|
|
return nhge->nh;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
/* nexthops always check if it is good and does
|
|
|
|
* not rely on a sysctl for this behavior
|
|
|
|
*/
|
|
|
|
switch (nhi->family) {
|
|
|
|
case AF_INET:
|
|
|
|
if (ipv4_good_nh(&nhi->fib_nh))
|
|
|
|
return nhge->nh;
|
|
|
|
break;
|
|
|
|
case AF_INET6:
|
|
|
|
if (ipv6_good_nh(&nhi->fib6_nh))
|
|
|
|
return nhge->nh;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!rc)
|
|
|
|
rc = nhge->nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
return rc;
|
|
|
|
}
|
2021-01-28 12:49:14 +00:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table = rcu_dereference(nhg->res_table);
|
|
|
|
u16 bucket_index = hash % res_table->num_nh_buckets;
|
|
|
|
struct nh_res_bucket *bucket;
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
/* nexthop_select_path() is expected to return a non-NULL value, so
|
|
|
|
* skip protocol validation and just hand out whatever there is.
|
|
|
|
*/
|
|
|
|
bucket = &res_table->nh_buckets[bucket_index];
|
|
|
|
nh_res_bucket_set_busy(bucket);
|
|
|
|
nhge = rcu_dereference(bucket->nh_entry);
|
|
|
|
return nhge->nh;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:14 +00:00
|
|
|
struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
|
|
|
if (!nh->is_group)
|
|
|
|
return nh;
|
|
|
|
|
|
|
|
nhg = rcu_dereference(nh->nh_grp);
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold)
|
|
|
|
return nexthop_select_path_hthr(nhg, hash);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
else if (nhg->resilient)
|
|
|
|
return nexthop_select_path_res(nhg, hash);
|
2021-01-28 12:49:14 +00:00
|
|
|
|
|
|
|
/* Unreachable. */
|
|
|
|
return NULL;
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
EXPORT_SYMBOL_GPL(nexthop_select_path);
|
|
|
|
|
2019-06-08 21:53:22 +00:00
|
|
|
int nexthop_for_each_fib6_nh(struct nexthop *nh,
|
|
|
|
int (*cb)(struct fib6_nh *nh, void *arg),
|
|
|
|
void *arg)
|
|
|
|
{
|
|
|
|
struct nh_info *nhi;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (nh->is_group) {
|
|
|
|
struct nh_group *nhg;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
nhg = rcu_dereference_rtnl(nh->nh_grp);
|
|
|
|
for (i = 0; i < nhg->num_nh; i++) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
|
|
|
|
nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
|
|
|
|
err = cb(&nhi->fib6_nh, arg);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
nhi = rcu_dereference_rtnl(nh->nh_info);
|
|
|
|
err = cb(&nhi->fib6_nh, arg);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
static int check_src_addr(const struct in6_addr *saddr,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
if (!ipv6_addr_any(saddr)) {
|
|
|
|
NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-06-04 03:19:52 +00:00
|
|
|
int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_info *nhi;
|
2020-06-09 02:54:43 +00:00
|
|
|
bool is_fdb_nh;
|
2020-05-22 05:26:13 +00:00
|
|
|
|
2019-06-04 03:19:52 +00:00
|
|
|
/* fib6_src is unique to a fib6_info and limits the ability to cache
|
|
|
|
* routes in fib6_nh within a nexthop that is potentially shared
|
|
|
|
* across multiple fib entries. If the config wants to use source
|
|
|
|
* routing it can not use nexthop objects. mlxsw also does not allow
|
|
|
|
* fib6_src on routes.
|
|
|
|
*/
|
2019-06-08 21:53:35 +00:00
|
|
|
if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
|
2019-06-04 03:19:52 +00:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (nh->is_group) {
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
if (nhg->has_v4)
|
|
|
|
goto no_v4_nh;
|
2020-06-09 02:54:43 +00:00
|
|
|
is_fdb_nh = nhg->fdb_nh;
|
2019-06-04 03:19:52 +00:00
|
|
|
} else {
|
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
if (nhi->family == AF_INET)
|
|
|
|
goto no_v4_nh;
|
2020-06-09 02:54:43 +00:00
|
|
|
is_fdb_nh = nhi->fdb_nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_fdb_nh) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
|
|
|
|
return -EINVAL;
|
2019-06-04 03:19:52 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
no_v4_nh:
|
|
|
|
NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(fib6_check_nexthop);
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
/* if existing nexthop has ipv6 routes linked to it, need
|
|
|
|
* to verify this new spec works with ipv6
|
|
|
|
*/
|
|
|
|
static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct fib6_info *f6i;
|
|
|
|
|
|
|
|
if (list_empty(&old->f6i_list))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
list_for_each_entry(f6i, &old->f6i_list, nh_list) {
|
|
|
|
if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return fib6_check_nexthop(new, NULL, extack);
|
|
|
|
}
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
static int nexthop_check_scope(struct nh_info *nhi, u8 scope,
|
2019-06-04 03:19:51 +00:00
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
|
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"Route with host scope can not have a gateway");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Invoked by fib add code to verify nexthop by id is ok with
|
|
|
|
* config for prefix; parts of fib_check_nh not done when nexthop
|
|
|
|
* object is used.
|
|
|
|
*/
|
|
|
|
int fib_check_nexthop(struct nexthop *nh, u8 scope,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
2020-06-09 02:54:43 +00:00
|
|
|
struct nh_info *nhi;
|
2019-06-04 03:19:51 +00:00
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
if (nh->is_group) {
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
if (nhg->fdb_nh) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
|
|
|
|
err = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2019-06-04 03:19:51 +00:00
|
|
|
if (scope == RT_SCOPE_HOST) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
|
|
|
|
err = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* all nexthops in a group have the same scope */
|
2020-06-09 02:54:43 +00:00
|
|
|
nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info);
|
|
|
|
err = nexthop_check_scope(nhi, scope, extack);
|
2019-06-04 03:19:51 +00:00
|
|
|
} else {
|
2020-06-09 02:54:43 +00:00
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
if (nhi->fdb_nh) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
|
|
|
|
err = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
err = nexthop_check_scope(nhi, scope, extack);
|
2019-06-04 03:19:51 +00:00
|
|
|
}
|
2020-06-09 02:54:43 +00:00
|
|
|
|
2019-06-04 03:19:51 +00:00
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct fib_info *fi;
|
|
|
|
|
|
|
|
list_for_each_entry(fi, &old->fi_list, nh_list) {
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = fib_check_nexthop(new, fi->fib_scope, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
static bool nh_res_nhge_is_balanced(const struct nh_grp_entry *nhge)
|
|
|
|
{
|
|
|
|
return nhge->res.count_buckets == nhge->res.wants_buckets;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool nh_res_nhge_is_ow(const struct nh_grp_entry *nhge)
|
|
|
|
{
|
|
|
|
return nhge->res.count_buckets > nhge->res.wants_buckets;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool nh_res_nhge_is_uw(const struct nh_grp_entry *nhge)
|
|
|
|
{
|
|
|
|
return nhge->res.count_buckets < nhge->res.wants_buckets;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool nh_res_table_is_balanced(const struct nh_res_table *res_table)
|
|
|
|
{
|
|
|
|
return list_empty(&res_table->uw_nh_entries);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_bucket_unset_nh(struct nh_res_bucket *bucket)
|
|
|
|
{
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
if (bucket->occupied) {
|
|
|
|
nhge = nh_res_dereference(bucket->nh_entry);
|
|
|
|
nhge->res.count_buckets--;
|
|
|
|
bucket->occupied = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_bucket_set_nh(struct nh_res_bucket *bucket,
|
|
|
|
struct nh_grp_entry *nhge)
|
|
|
|
{
|
|
|
|
nh_res_bucket_unset_nh(bucket);
|
|
|
|
|
|
|
|
bucket->occupied = true;
|
|
|
|
rcu_assign_pointer(bucket->nh_entry, nhge);
|
|
|
|
nhge->res.count_buckets++;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool nh_res_bucket_should_migrate(struct nh_res_table *res_table,
|
|
|
|
struct nh_res_bucket *bucket,
|
|
|
|
unsigned long *deadline, bool *force)
|
|
|
|
{
|
|
|
|
unsigned long now = jiffies;
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
unsigned long idle_point;
|
|
|
|
|
|
|
|
if (!bucket->occupied) {
|
|
|
|
/* The bucket is not occupied, its NHGE pointer is either
|
|
|
|
* NULL or obsolete. We _have to_ migrate: set force.
|
|
|
|
*/
|
|
|
|
*force = true;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
nhge = nh_res_dereference(bucket->nh_entry);
|
|
|
|
|
|
|
|
/* If the bucket is populated by an underweight or balanced
|
|
|
|
* nexthop, do not migrate.
|
|
|
|
*/
|
|
|
|
if (!nh_res_nhge_is_ow(nhge))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* At this point we know that the bucket is populated with an
|
|
|
|
* overweight nexthop. It needs to be migrated to a new nexthop if
|
|
|
|
* the idle timer of unbalanced timer expired.
|
|
|
|
*/
|
|
|
|
|
|
|
|
idle_point = nh_res_bucket_idle_point(res_table, bucket, now);
|
|
|
|
if (time_after_eq(now, idle_point)) {
|
|
|
|
/* The bucket is idle. We _can_ migrate: unset force. */
|
|
|
|
*force = false;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Unbalanced timer of 0 means "never force". */
|
|
|
|
if (res_table->unbalanced_timer) {
|
|
|
|
unsigned long unb_point;
|
|
|
|
|
|
|
|
unb_point = nh_res_table_unb_point(res_table);
|
|
|
|
if (time_after(now, unb_point)) {
|
|
|
|
/* The bucket is not idle, but the unbalanced timer
|
|
|
|
* expired. We _can_ migrate, but set force anyway,
|
|
|
|
* so that drivers know to ignore activity reports
|
|
|
|
* from the HW.
|
|
|
|
*/
|
|
|
|
*force = true;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
nh_res_time_set_deadline(unb_point, deadline);
|
|
|
|
}
|
|
|
|
|
|
|
|
nh_res_time_set_deadline(idle_point, deadline);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool nh_res_bucket_migrate(struct nh_res_table *res_table,
|
2021-03-11 18:03:24 +00:00
|
|
|
u16 bucket_index, bool notify,
|
|
|
|
bool notify_nl, bool force)
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
{
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index];
|
|
|
|
struct nh_grp_entry *new_nhge;
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
struct netlink_ext_ack extack;
|
|
|
|
int err;
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
|
|
|
|
new_nhge = list_first_entry_or_null(&res_table->uw_nh_entries,
|
|
|
|
struct nh_grp_entry,
|
|
|
|
res.uw_nh_entry);
|
|
|
|
if (WARN_ON_ONCE(!new_nhge))
|
|
|
|
/* If this function is called, "bucket" is either not
|
|
|
|
* occupied, or it belongs to a next hop that is
|
|
|
|
* overweight. In either case, there ought to be a
|
|
|
|
* corresponding underweight next hop.
|
|
|
|
*/
|
|
|
|
return false;
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
if (notify) {
|
|
|
|
struct nh_grp_entry *old_nhge;
|
|
|
|
|
|
|
|
old_nhge = nh_res_dereference(bucket->nh_entry);
|
|
|
|
err = call_nexthop_res_bucket_notifiers(res_table->net,
|
|
|
|
res_table->nhg_id,
|
|
|
|
bucket_index, force,
|
|
|
|
old_nhge->nh,
|
|
|
|
new_nhge->nh, &extack);
|
|
|
|
if (err) {
|
|
|
|
pr_err_ratelimited("%s\n", extack._msg);
|
|
|
|
if (!force)
|
|
|
|
return false;
|
|
|
|
/* It is not possible to veto a forced replacement, so
|
|
|
|
* just clear the hardware flags from the nexthop
|
|
|
|
* bucket to indicate to user space that this bucket is
|
|
|
|
* not correctly populated in hardware.
|
|
|
|
*/
|
|
|
|
bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
nh_res_bucket_set_nh(bucket, new_nhge);
|
|
|
|
nh_res_bucket_set_idle(res_table, bucket);
|
|
|
|
|
2021-03-11 18:03:24 +00:00
|
|
|
if (notify_nl)
|
|
|
|
nexthop_bucket_notify(res_table, bucket_index);
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (nh_res_nhge_is_balanced(new_nhge))
|
|
|
|
list_del(&new_nhge->res.uw_nh_entry);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define NH_RES_UPKEEP_DW_MINIMUM_INTERVAL (HZ / 2)
|
|
|
|
|
2021-03-11 18:03:24 +00:00
|
|
|
static void nh_res_table_upkeep(struct nh_res_table *res_table,
|
|
|
|
bool notify, bool notify_nl)
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
{
|
|
|
|
unsigned long now = jiffies;
|
|
|
|
unsigned long deadline;
|
|
|
|
u16 i;
|
|
|
|
|
|
|
|
/* Deadline is the next time that upkeep should be run. It is the
|
|
|
|
* earliest time at which one of the buckets might be migrated.
|
|
|
|
* Start at the most pessimistic estimate: either unbalanced_timer
|
|
|
|
* from now, or if there is none, idle_timer from now. For each
|
|
|
|
* encountered time point, call nh_res_time_set_deadline() to
|
|
|
|
* refine the estimate.
|
|
|
|
*/
|
|
|
|
if (res_table->unbalanced_timer)
|
|
|
|
deadline = now + res_table->unbalanced_timer;
|
|
|
|
else
|
|
|
|
deadline = now + res_table->idle_timer;
|
|
|
|
|
|
|
|
for (i = 0; i < res_table->num_nh_buckets; i++) {
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
|
|
|
|
bool force;
|
|
|
|
|
|
|
|
if (nh_res_bucket_should_migrate(res_table, bucket,
|
|
|
|
&deadline, &force)) {
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
if (!nh_res_bucket_migrate(res_table, i, notify,
|
2021-03-11 18:03:24 +00:00
|
|
|
notify_nl, force)) {
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
unsigned long idle_point;
|
|
|
|
|
|
|
|
/* A driver can override the migration
|
|
|
|
* decision if the HW reports that the
|
|
|
|
* bucket is actually not idle. Therefore
|
|
|
|
* remark the bucket as busy again and
|
|
|
|
* update the deadline.
|
|
|
|
*/
|
|
|
|
nh_res_bucket_set_busy(bucket);
|
|
|
|
idle_point = nh_res_bucket_idle_point(res_table,
|
|
|
|
bucket,
|
|
|
|
now);
|
|
|
|
nh_res_time_set_deadline(idle_point, &deadline);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If the group is still unbalanced, schedule the next upkeep to
|
|
|
|
* either the deadline computed above, or the minimum deadline,
|
|
|
|
* whichever comes later.
|
|
|
|
*/
|
|
|
|
if (!nh_res_table_is_balanced(res_table)) {
|
|
|
|
unsigned long now = jiffies;
|
|
|
|
unsigned long min_deadline;
|
|
|
|
|
|
|
|
min_deadline = now + NH_RES_UPKEEP_DW_MINIMUM_INTERVAL;
|
|
|
|
if (time_before(deadline, min_deadline))
|
|
|
|
deadline = min_deadline;
|
|
|
|
|
|
|
|
queue_delayed_work(system_power_efficient_wq,
|
|
|
|
&res_table->upkeep_dw, deadline - now);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_table_upkeep_dw(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct delayed_work *dw = to_delayed_work(work);
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
|
|
|
|
res_table = container_of(dw, struct nh_res_table, upkeep_dw);
|
2021-03-11 18:03:24 +00:00
|
|
|
nh_res_table_upkeep(res_table, true, true);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_table_cancel_upkeep(struct nh_res_table *res_table)
|
|
|
|
{
|
|
|
|
cancel_delayed_work_sync(&res_table->upkeep_dw);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nh_res_group_rebalance(struct nh_group *nhg,
|
|
|
|
struct nh_res_table *res_table)
|
|
|
|
{
|
|
|
|
int prev_upper_bound = 0;
|
|
|
|
int total = 0;
|
|
|
|
int w = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&res_table->uw_nh_entries);
|
|
|
|
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i)
|
|
|
|
total += nhg->nh_entries[i].weight;
|
|
|
|
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
int upper_bound;
|
|
|
|
|
|
|
|
w += nhge->weight;
|
|
|
|
upper_bound = DIV_ROUND_CLOSEST(res_table->num_nh_buckets * w,
|
|
|
|
total);
|
|
|
|
nhge->res.wants_buckets = upper_bound - prev_upper_bound;
|
|
|
|
prev_upper_bound = upper_bound;
|
|
|
|
|
|
|
|
if (nh_res_nhge_is_uw(nhge)) {
|
|
|
|
if (list_empty(&res_table->uw_nh_entries))
|
|
|
|
res_table->unbalanced_since = jiffies;
|
|
|
|
list_add(&nhge->res.uw_nh_entry,
|
|
|
|
&res_table->uw_nh_entries);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Migrate buckets in res_table so that they reference NHGE's from NHG with
|
|
|
|
* the right NH ID. Set those buckets that do not have a corresponding NHGE
|
|
|
|
* entry in NHG as not occupied.
|
|
|
|
*/
|
|
|
|
static void nh_res_table_migrate_buckets(struct nh_res_table *res_table,
|
|
|
|
struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
u16 i;
|
|
|
|
|
|
|
|
for (i = 0; i < res_table->num_nh_buckets; i++) {
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
|
|
|
|
u32 id = rtnl_dereference(bucket->nh_entry)->nh->id;
|
|
|
|
bool found = false;
|
|
|
|
int j;
|
|
|
|
|
|
|
|
for (j = 0; j < nhg->num_nh; j++) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[j];
|
|
|
|
|
|
|
|
if (nhge->nh->id == id) {
|
|
|
|
nh_res_bucket_set_nh(bucket, nhge);
|
|
|
|
found = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!found)
|
|
|
|
nh_res_bucket_unset_nh(bucket);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void replace_nexthop_grp_res(struct nh_group *oldg,
|
|
|
|
struct nh_group *newg)
|
|
|
|
{
|
|
|
|
/* For NH group replacement, the new NHG might only have a stub
|
|
|
|
* hash table with 0 buckets, because the number of buckets was not
|
|
|
|
* specified. For NH removal, oldg and newg both reference the same
|
|
|
|
* res_table. So in any case, in the following, we want to work
|
|
|
|
* with oldg->res_table.
|
|
|
|
*/
|
|
|
|
struct nh_res_table *old_res_table = rtnl_dereference(oldg->res_table);
|
|
|
|
unsigned long prev_unbalanced_since = old_res_table->unbalanced_since;
|
|
|
|
bool prev_has_uw = !list_empty(&old_res_table->uw_nh_entries);
|
|
|
|
|
|
|
|
nh_res_table_cancel_upkeep(old_res_table);
|
|
|
|
nh_res_table_migrate_buckets(old_res_table, newg);
|
|
|
|
nh_res_group_rebalance(newg, old_res_table);
|
|
|
|
if (prev_has_uw && !list_empty(&old_res_table->uw_nh_entries))
|
|
|
|
old_res_table->unbalanced_since = prev_unbalanced_since;
|
2021-03-11 18:03:24 +00:00
|
|
|
nh_res_table_upkeep(old_res_table, true, false);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
}
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
static void nh_hthr_group_rebalance(struct nh_group *nhg)
|
2019-05-24 21:43:08 +00:00
|
|
|
{
|
|
|
|
int total = 0;
|
|
|
|
int w = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i)
|
|
|
|
total += nhg->nh_entries[i].weight;
|
|
|
|
|
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
int upper_bound;
|
|
|
|
|
|
|
|
w += nhge->weight;
|
|
|
|
upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
|
2021-03-26 13:20:22 +00:00
|
|
|
atomic_set(&nhge->hthr.upper_bound, upper_bound);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-26 18:56:14 +00:00
|
|
|
static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
|
2019-05-24 21:43:08 +00:00
|
|
|
struct nl_info *nlinfo)
|
|
|
|
{
|
2020-05-26 18:56:15 +00:00
|
|
|
struct nh_grp_entry *nhges, *new_nhges;
|
2020-05-26 18:56:14 +00:00
|
|
|
struct nexthop *nhp = nhge->nh_parent;
|
2020-11-04 13:30:33 +00:00
|
|
|
struct netlink_ext_ack extack;
|
2019-05-24 21:43:08 +00:00
|
|
|
struct nexthop *nh = nhge->nh;
|
2020-05-26 18:56:15 +00:00
|
|
|
struct nh_group *nhg, *newg;
|
2020-11-04 13:30:33 +00:00
|
|
|
int i, j, err;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
WARN_ON(!nh);
|
|
|
|
|
2020-05-26 18:56:14 +00:00
|
|
|
nhg = rtnl_dereference(nhp->nh_grp);
|
2020-05-26 18:56:15 +00:00
|
|
|
newg = nhg->spare;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
/* last entry, keep it visible and remove the parent */
|
|
|
|
if (nhg->num_nh == 1) {
|
|
|
|
remove_nexthop(net, nhp, nlinfo);
|
2019-05-24 21:43:08 +00:00
|
|
|
return;
|
2020-05-26 18:56:15 +00:00
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-08-26 16:48:54 +00:00
|
|
|
newg->has_v4 = false;
|
2021-03-11 18:03:14 +00:00
|
|
|
newg->is_multipath = nhg->is_multipath;
|
2021-03-26 13:20:22 +00:00
|
|
|
newg->hash_threshold = nhg->hash_threshold;
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
newg->resilient = nhg->resilient;
|
2020-06-09 02:54:43 +00:00
|
|
|
newg->fdb_nh = nhg->fdb_nh;
|
2020-05-26 18:56:15 +00:00
|
|
|
newg->num_nh = nhg->num_nh;
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
/* copy old entries to new except the one getting removed */
|
|
|
|
nhges = nhg->nh_entries;
|
|
|
|
new_nhges = newg->nh_entries;
|
|
|
|
for (i = 0, j = 0; i < nhg->num_nh; ++i) {
|
2020-08-26 16:48:54 +00:00
|
|
|
struct nh_info *nhi;
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
/* current nexthop getting removed */
|
|
|
|
if (nhg->nh_entries[i].nh == nh) {
|
|
|
|
newg->num_nh--;
|
|
|
|
continue;
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-08-26 16:48:54 +00:00
|
|
|
nhi = rtnl_dereference(nhges[i].nh->nh_info);
|
|
|
|
if (nhi->family == AF_INET)
|
|
|
|
newg->has_v4 = true;
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
list_del(&nhges[i].nh_list);
|
|
|
|
new_nhges[j].nh_parent = nhges[i].nh_parent;
|
|
|
|
new_nhges[j].nh = nhges[i].nh;
|
|
|
|
new_nhges[j].weight = nhges[i].weight;
|
|
|
|
list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
|
|
|
|
j++;
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (newg->hash_threshold)
|
|
|
|
nh_hthr_group_rebalance(newg);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
else if (newg->resilient)
|
|
|
|
replace_nexthop_grp_res(nhg, newg);
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
rcu_assign_pointer(nhp->nh_grp, newg);
|
|
|
|
|
|
|
|
list_del(&nhge->nh_list);
|
|
|
|
nexthop_put(nhge->nh);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
/* Removal of a NH from a resilient group is notified through
|
|
|
|
* bucket notifications.
|
|
|
|
*/
|
2021-03-26 13:20:22 +00:00
|
|
|
if (newg->hash_threshold) {
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
|
|
|
|
&extack);
|
|
|
|
if (err)
|
|
|
|
pr_err("%s\n", extack._msg);
|
|
|
|
}
|
2020-11-04 13:30:33 +00:00
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nlinfo)
|
2020-05-26 18:56:14 +00:00
|
|
|
nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
|
|
|
|
struct nl_info *nlinfo)
|
|
|
|
{
|
|
|
|
struct nh_grp_entry *nhge, *tmp;
|
|
|
|
|
2020-05-26 18:56:14 +00:00
|
|
|
list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
|
|
|
|
remove_nh_grp_entry(net, nhge, nlinfo);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
/* make sure all see the newly published array before releasing rtnl */
|
2020-10-16 17:29:14 +00:00
|
|
|
synchronize_net();
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
struct nh_res_table *res_table;
|
2019-05-24 21:43:08 +00:00
|
|
|
int i, num_nh = nhg->num_nh;
|
|
|
|
|
|
|
|
for (i = 0; i < num_nh; ++i) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
|
|
|
|
if (WARN_ON(!nhge->nh))
|
|
|
|
continue;
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
list_del_init(&nhge->nh_list);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
|
|
|
|
if (nhg->resilient) {
|
|
|
|
res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
nh_res_table_cancel_upkeep(res_table);
|
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
/* not called for nexthop replace */
|
2019-06-04 03:19:51 +00:00
|
|
|
static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
|
|
|
|
{
|
2019-06-04 03:19:52 +00:00
|
|
|
struct fib6_info *f6i, *tmp;
|
2019-06-04 03:19:51 +00:00
|
|
|
bool do_flush = false;
|
|
|
|
struct fib_info *fi;
|
|
|
|
|
|
|
|
list_for_each_entry(fi, &nh->fi_list, nh_list) {
|
|
|
|
fi->fib_flags |= RTNH_F_DEAD;
|
|
|
|
do_flush = true;
|
|
|
|
}
|
|
|
|
if (do_flush)
|
|
|
|
fib_flush(net);
|
2019-06-04 03:19:52 +00:00
|
|
|
|
|
|
|
/* ip6_del_rt removes the entry from this list hence the _safe */
|
|
|
|
list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
|
|
|
|
/* __ip6_del_rt does a release, so do a hold here */
|
|
|
|
fib6_info_hold(f6i);
|
2020-04-27 20:56:46 +00:00
|
|
|
ipv6_stub->ip6_del_rt(net, f6i,
|
2022-07-12 00:15:33 +00:00
|
|
|
!READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode));
|
2019-06-04 03:19:52 +00:00
|
|
|
}
|
2019-06-04 03:19:51 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static void __remove_nexthop(struct net *net, struct nexthop *nh,
|
|
|
|
struct nl_info *nlinfo)
|
|
|
|
{
|
2019-06-04 03:19:51 +00:00
|
|
|
__remove_nexthop_fib(net, nh);
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nh->is_group) {
|
|
|
|
remove_nexthop_group(nh, nlinfo);
|
|
|
|
} else {
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
if (nhi->fib_nhc.nhc_dev)
|
|
|
|
hlist_del(&nhi->dev_hash);
|
|
|
|
|
|
|
|
remove_nexthop_from_groups(net, nh, nlinfo);
|
|
|
|
}
|
2019-05-24 21:43:05 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static void remove_nexthop(struct net *net, struct nexthop *nh,
|
2019-05-24 21:43:08 +00:00
|
|
|
struct nl_info *nlinfo)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
2020-11-04 13:30:24 +00:00
|
|
|
call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh, NULL);
|
2020-09-15 11:41:02 +00:00
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
/* remove from the tree */
|
|
|
|
rb_erase(&nh->rb_node, &net->nexthop.rb_root);
|
|
|
|
|
|
|
|
if (nlinfo)
|
|
|
|
nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
__remove_nexthop(net, nh, nlinfo);
|
2019-05-24 21:43:04 +00:00
|
|
|
nh_base_seq_inc(net);
|
|
|
|
|
|
|
|
nexthop_put(nh);
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
/* if any FIB entries reference this nexthop, any dst entries
|
|
|
|
* need to be regenerated
|
|
|
|
*/
|
net: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group
When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.
[1]
$ ip nexthop list
id 200 via 2002:db8::2 dev bridge.10 scope link onlink
id 201 via 2002:db8::3 dev bridge scope link onlink
id 203 group 201/200
$ ip -6 route
2001:db8::10 nhid 203 metric 1024 pref medium
nexthop via 2002:db8::3 dev bridge weight 1 onlink
nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink
Create rt6_info through one of the multipath legs, e.g.:
$ taskset -a -c 1 ./pkt_inj 24 bridge.10 2001:db8::10
(pkt_inj is just a custom packet generator, nothing special)
Then remove that leg from the group by replace (let's assume it is id
200 in this case):
$ ip nexthop replace id 203 group 201
Now remove the IPv6 route:
$ ip -6 route del 2001:db8::10/128
The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
(deleted) IPv6 route holds 1 reference over nhid 203
nh 203 holds 1 ref over id 201
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
$ ip nexthop replace id 203 group 201/200
And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.
To trigger the bug just delete the group (nhid 203):
$ ip nexthop del id 203
It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.
At this point the dependencies are:
(deleted, only unlinked) IPv6 route holds reference over group nh 203
(deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.
If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
$ ip nexthop del id 200
$ ip nexthop
$
Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.
Message from syslogd@dev at Nov 19 14:04:10 ...
kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Message from syslogd@dev at Nov 19 14:04:20 ...
kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-11-22 15:15:13 +00:00
|
|
|
static void nh_rt_cache_flush(struct net *net, struct nexthop *nh,
|
|
|
|
struct nexthop *replaced_nh)
|
2019-06-08 21:53:35 +00:00
|
|
|
{
|
|
|
|
struct fib6_info *f6i;
|
net: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group
When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.
[1]
$ ip nexthop list
id 200 via 2002:db8::2 dev bridge.10 scope link onlink
id 201 via 2002:db8::3 dev bridge scope link onlink
id 203 group 201/200
$ ip -6 route
2001:db8::10 nhid 203 metric 1024 pref medium
nexthop via 2002:db8::3 dev bridge weight 1 onlink
nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink
Create rt6_info through one of the multipath legs, e.g.:
$ taskset -a -c 1 ./pkt_inj 24 bridge.10 2001:db8::10
(pkt_inj is just a custom packet generator, nothing special)
Then remove that leg from the group by replace (let's assume it is id
200 in this case):
$ ip nexthop replace id 203 group 201
Now remove the IPv6 route:
$ ip -6 route del 2001:db8::10/128
The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
(deleted) IPv6 route holds 1 reference over nhid 203
nh 203 holds 1 ref over id 201
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
$ ip nexthop replace id 203 group 201/200
And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.
To trigger the bug just delete the group (nhid 203):
$ ip nexthop del id 203
It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.
At this point the dependencies are:
(deleted, only unlinked) IPv6 route holds reference over group nh 203
(deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.
If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
$ ip nexthop del id 200
$ ip nexthop
$
Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.
Message from syslogd@dev at Nov 19 14:04:10 ...
kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Message from syslogd@dev at Nov 19 14:04:20 ...
kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-11-22 15:15:13 +00:00
|
|
|
struct nh_group *nhg;
|
|
|
|
int i;
|
2019-06-08 21:53:35 +00:00
|
|
|
|
|
|
|
if (!list_empty(&nh->fi_list))
|
|
|
|
rt_cache_flush(net);
|
|
|
|
|
|
|
|
list_for_each_entry(f6i, &nh->f6i_list, nh_list)
|
|
|
|
ipv6_stub->fib6_update_sernum(net, f6i);
|
net: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group
When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.
[1]
$ ip nexthop list
id 200 via 2002:db8::2 dev bridge.10 scope link onlink
id 201 via 2002:db8::3 dev bridge scope link onlink
id 203 group 201/200
$ ip -6 route
2001:db8::10 nhid 203 metric 1024 pref medium
nexthop via 2002:db8::3 dev bridge weight 1 onlink
nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink
Create rt6_info through one of the multipath legs, e.g.:
$ taskset -a -c 1 ./pkt_inj 24 bridge.10 2001:db8::10
(pkt_inj is just a custom packet generator, nothing special)
Then remove that leg from the group by replace (let's assume it is id
200 in this case):
$ ip nexthop replace id 203 group 201
Now remove the IPv6 route:
$ ip -6 route del 2001:db8::10/128
The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
(deleted) IPv6 route holds 1 reference over nhid 203
nh 203 holds 1 ref over id 201
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
$ ip nexthop replace id 203 group 201/200
And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.
To trigger the bug just delete the group (nhid 203):
$ ip nexthop del id 203
It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.
At this point the dependencies are:
(deleted, only unlinked) IPv6 route holds reference over group nh 203
(deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.
If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
$ ip nexthop del id 200
$ ip nexthop
$
Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.
Message from syslogd@dev at Nov 19 14:04:10 ...
kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Message from syslogd@dev at Nov 19 14:04:20 ...
kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-11-22 15:15:13 +00:00
|
|
|
|
|
|
|
/* if an IPv6 group was replaced, we have to release all old
|
|
|
|
* dsts to make sure all refcounts are released
|
|
|
|
*/
|
|
|
|
if (!replaced_nh->is_group)
|
|
|
|
return;
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(replaced_nh->nh_grp);
|
|
|
|
for (i = 0; i < nhg->num_nh; i++) {
|
|
|
|
struct nh_grp_entry *nhge = &nhg->nh_entries[i];
|
|
|
|
struct nh_info *nhi = rtnl_dereference(nhge->nh->nh_info);
|
|
|
|
|
|
|
|
if (nhi->family == AF_INET6)
|
|
|
|
ipv6_stub->fib6_nh_release_dsts(&nhi->fib6_nh);
|
|
|
|
}
|
2019-06-08 21:53:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int replace_nexthop_grp(struct net *net, struct nexthop *old,
|
2021-03-11 18:03:12 +00:00
|
|
|
struct nexthop *new, const struct nh_config *cfg,
|
2019-06-08 21:53:35 +00:00
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
struct nh_res_table *tmp_table = NULL;
|
|
|
|
struct nh_res_table *new_res_table;
|
|
|
|
struct nh_res_table *old_res_table;
|
2019-06-08 21:53:35 +00:00
|
|
|
struct nh_group *oldg, *newg;
|
2020-11-04 13:30:30 +00:00
|
|
|
int i, err;
|
2019-06-08 21:53:35 +00:00
|
|
|
|
|
|
|
if (!new->is_group) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
oldg = rtnl_dereference(old->nh_grp);
|
|
|
|
newg = rtnl_dereference(new->nh_grp);
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (newg->hash_threshold != oldg->hash_threshold) {
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with one of a different type.");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (newg->hash_threshold) {
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new,
|
|
|
|
extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
} else if (newg->resilient) {
|
|
|
|
new_res_table = rtnl_dereference(newg->res_table);
|
|
|
|
old_res_table = rtnl_dereference(oldg->res_table);
|
|
|
|
|
|
|
|
/* Accept if num_nh_buckets was not given, but if it was
|
|
|
|
* given, demand that the value be correct.
|
|
|
|
*/
|
|
|
|
if (cfg->nh_grp_res_has_num_buckets &&
|
|
|
|
cfg->nh_grp_res_num_buckets !=
|
|
|
|
old_res_table->num_nh_buckets) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Can not change number of buckets of a resilient nexthop group.");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
/* Emit a pre-replace notification so that listeners could veto
|
|
|
|
* a potentially unsupported configuration. Otherwise,
|
|
|
|
* individual bucket replacement notifications would need to be
|
|
|
|
* vetoed, which is something that should only happen if the
|
|
|
|
* bucket is currently active.
|
|
|
|
*/
|
|
|
|
err = call_nexthop_res_table_notifiers(net, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (cfg->nh_grp_res_has_idle_timer)
|
|
|
|
old_res_table->idle_timer = cfg->nh_grp_res_idle_timer;
|
|
|
|
if (cfg->nh_grp_res_has_unbalanced_timer)
|
|
|
|
old_res_table->unbalanced_timer =
|
|
|
|
cfg->nh_grp_res_unbalanced_timer;
|
|
|
|
|
|
|
|
replace_nexthop_grp_res(oldg, newg);
|
|
|
|
|
|
|
|
tmp_table = new_res_table;
|
|
|
|
rcu_assign_pointer(newg->res_table, old_res_table);
|
|
|
|
rcu_assign_pointer(newg->spare->res_table, old_res_table);
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
/* update parents - used by nexthop code for cleanup */
|
|
|
|
for (i = 0; i < newg->num_nh; i++)
|
|
|
|
newg->nh_entries[i].nh_parent = old;
|
|
|
|
|
|
|
|
rcu_assign_pointer(old->nh_grp, newg);
|
|
|
|
|
2021-11-29 12:09:24 +00:00
|
|
|
/* Make sure concurrent readers are not using 'oldg' anymore. */
|
|
|
|
synchronize_net();
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (newg->resilient) {
|
|
|
|
rcu_assign_pointer(oldg->res_table, tmp_table);
|
|
|
|
rcu_assign_pointer(oldg->spare->res_table, tmp_table);
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
for (i = 0; i < oldg->num_nh; i++)
|
|
|
|
oldg->nh_entries[i].nh_parent = new;
|
|
|
|
|
|
|
|
rcu_assign_pointer(new->nh_grp, oldg);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-08-26 16:48:56 +00:00
|
|
|
static void nh_group_v4_update(struct nh_group *nhg)
|
|
|
|
{
|
|
|
|
struct nh_grp_entry *nhges;
|
|
|
|
bool has_v4 = false;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
nhges = nhg->nh_entries;
|
|
|
|
for (i = 0; i < nhg->num_nh; i++) {
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
nhi = rtnl_dereference(nhges[i].nh->nh_info);
|
|
|
|
if (nhi->family == AF_INET)
|
|
|
|
has_v4 = true;
|
|
|
|
}
|
|
|
|
nhg->has_v4 = has_v4;
|
|
|
|
}
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
static int replace_nexthop_single_notify_res(struct net *net,
|
|
|
|
struct nh_res_table *res_table,
|
|
|
|
struct nexthop *old,
|
|
|
|
struct nh_info *oldi,
|
|
|
|
struct nh_info *newi,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
u32 nhg_id = res_table->nhg_id;
|
|
|
|
int err;
|
|
|
|
u16 i;
|
|
|
|
|
|
|
|
for (i = 0; i < res_table->num_nh_buckets; i++) {
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
nhge = rtnl_dereference(bucket->nh_entry);
|
|
|
|
if (nhge->nh == old) {
|
|
|
|
err = __call_nexthop_res_bucket_notifiers(net, nhg_id,
|
|
|
|
i, true,
|
|
|
|
oldi, newi,
|
|
|
|
extack);
|
|
|
|
if (err)
|
|
|
|
goto err_notify;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err_notify:
|
|
|
|
while (i-- > 0) {
|
|
|
|
struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
nhge = rtnl_dereference(bucket->nh_entry);
|
|
|
|
if (nhge->nh == old)
|
|
|
|
__call_nexthop_res_bucket_notifiers(net, nhg_id, i,
|
|
|
|
true, newi, oldi,
|
|
|
|
extack);
|
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int replace_nexthop_single_notify(struct net *net,
|
|
|
|
struct nexthop *group_nh,
|
|
|
|
struct nexthop *old,
|
|
|
|
struct nh_info *oldi,
|
|
|
|
struct nh_info *newi,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg = rtnl_dereference(group_nh->nh_grp);
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold) {
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
return call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE,
|
|
|
|
group_nh, extack);
|
|
|
|
} else if (nhg->resilient) {
|
|
|
|
res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
return replace_nexthop_single_notify_res(net, res_table,
|
|
|
|
old, oldi, newi,
|
|
|
|
extack);
|
|
|
|
}
|
|
|
|
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
static int replace_nexthop_single(struct net *net, struct nexthop *old,
|
|
|
|
struct nexthop *new,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
2020-11-04 13:30:32 +00:00
|
|
|
u8 old_protocol, old_nh_flags;
|
2019-06-08 21:53:35 +00:00
|
|
|
struct nh_info *oldi, *newi;
|
2020-11-04 13:30:32 +00:00
|
|
|
struct nh_grp_entry *nhge;
|
2020-11-04 13:30:31 +00:00
|
|
|
int err;
|
2019-06-08 21:53:35 +00:00
|
|
|
|
|
|
|
if (new->is_group) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2020-11-04 13:30:31 +00:00
|
|
|
err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
/* Hardware flags were set on 'old' as 'new' is not in the red-black
|
|
|
|
* tree. Therefore, inherit the flags from 'old' to 'new'.
|
|
|
|
*/
|
|
|
|
new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
oldi = rtnl_dereference(old->nh_info);
|
|
|
|
newi = rtnl_dereference(new->nh_info);
|
|
|
|
|
|
|
|
newi->nh_parent = old;
|
|
|
|
oldi->nh_parent = new;
|
|
|
|
|
2020-11-04 13:30:32 +00:00
|
|
|
old_protocol = old->protocol;
|
|
|
|
old_nh_flags = old->nh_flags;
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
old->protocol = new->protocol;
|
|
|
|
old->nh_flags = new->nh_flags;
|
|
|
|
|
|
|
|
rcu_assign_pointer(old->nh_info, newi);
|
|
|
|
rcu_assign_pointer(new->nh_info, oldi);
|
|
|
|
|
2020-11-04 13:30:32 +00:00
|
|
|
/* Send a replace notification for all the groups using the nexthop. */
|
|
|
|
list_for_each_entry(nhge, &old->grp_list, nh_list) {
|
|
|
|
struct nexthop *nhp = nhge->nh_parent;
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
err = replace_nexthop_single_notify(net, nhp, old, oldi, newi,
|
|
|
|
extack);
|
2020-11-04 13:30:32 +00:00
|
|
|
if (err)
|
|
|
|
goto err_notify;
|
|
|
|
}
|
|
|
|
|
2020-08-26 16:48:56 +00:00
|
|
|
/* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially
|
|
|
|
* update IPv4 indication in all the groups using the nexthop.
|
|
|
|
*/
|
|
|
|
if (oldi->family == AF_INET && newi->family == AF_INET6) {
|
|
|
|
list_for_each_entry(nhge, &old->grp_list, nh_list) {
|
|
|
|
struct nexthop *nhp = nhge->nh_parent;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nhp->nh_grp);
|
|
|
|
nh_group_v4_update(nhg);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-06-08 21:53:35 +00:00
|
|
|
return 0;
|
2020-11-04 13:30:32 +00:00
|
|
|
|
|
|
|
err_notify:
|
|
|
|
rcu_assign_pointer(new->nh_info, newi);
|
|
|
|
rcu_assign_pointer(old->nh_info, oldi);
|
|
|
|
old->nh_flags = old_nh_flags;
|
|
|
|
old->protocol = old_protocol;
|
|
|
|
oldi->nh_parent = old;
|
|
|
|
newi->nh_parent = new;
|
|
|
|
list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) {
|
|
|
|
struct nexthop *nhp = nhge->nh_parent;
|
|
|
|
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
replace_nexthop_single_notify(net, nhp, old, newi, oldi, NULL);
|
2020-11-04 13:30:32 +00:00
|
|
|
}
|
|
|
|
call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack);
|
|
|
|
return err;
|
2019-06-08 21:53:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
|
|
|
|
struct nl_info *info)
|
|
|
|
{
|
|
|
|
struct fib6_info *f6i;
|
|
|
|
|
|
|
|
if (!list_empty(&nh->fi_list)) {
|
|
|
|
struct fib_info *fi;
|
|
|
|
|
|
|
|
/* expectation is a few fib_info per nexthop and then
|
|
|
|
* a lot of routes per fib_info. So mark the fib_info
|
|
|
|
* and then walk the fib tables once
|
|
|
|
*/
|
|
|
|
list_for_each_entry(fi, &nh->fi_list, nh_list)
|
|
|
|
fi->nh_updated = true;
|
|
|
|
|
|
|
|
fib_info_notify_update(net, info);
|
|
|
|
|
|
|
|
list_for_each_entry(fi, &nh->fi_list, nh_list)
|
|
|
|
fi->nh_updated = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_for_each_entry(f6i, &nh->f6i_list, nh_list)
|
|
|
|
ipv6_stub->fib6_rt_update(net, f6i, info);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
|
|
|
|
* linked to this nexthop and for all groups that the nexthop
|
|
|
|
* is a member of
|
|
|
|
*/
|
|
|
|
static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
|
|
|
|
struct nl_info *info)
|
|
|
|
{
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
__nexthop_replace_notify(net, nh, info);
|
|
|
|
|
|
|
|
list_for_each_entry(nhge, &nh->grp_list, nh_list)
|
|
|
|
__nexthop_replace_notify(net, nhge->nh_parent, info);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static int replace_nexthop(struct net *net, struct nexthop *old,
|
2021-03-11 18:03:12 +00:00
|
|
|
struct nexthop *new, const struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
2019-06-08 21:53:35 +00:00
|
|
|
bool new_is_reject = false;
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
/* check that existing FIB entries are ok with the
|
|
|
|
* new nexthop definition
|
|
|
|
*/
|
|
|
|
err = fib_check_nh_list(old, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = fib6_check_nh_list(old, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (!new->is_group) {
|
|
|
|
struct nh_info *nhi = rtnl_dereference(new->nh_info);
|
|
|
|
|
|
|
|
new_is_reject = nhi->reject_nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_for_each_entry(nhge, &old->grp_list, nh_list) {
|
|
|
|
/* if new nexthop is a blackhole, any groups using this
|
|
|
|
* nexthop cannot have more than 1 path
|
|
|
|
*/
|
|
|
|
if (new_is_reject &&
|
|
|
|
nexthop_num_path(nhge->nh_parent) > 1) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = fib_check_nh_list(nhge->nh_parent, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = fib6_check_nh_list(nhge->nh_parent, new, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (old->is_group)
|
2021-03-11 18:03:12 +00:00
|
|
|
err = replace_nexthop_grp(net, old, new, cfg, extack);
|
2019-06-08 21:53:35 +00:00
|
|
|
else
|
|
|
|
err = replace_nexthop_single(net, old, new, extack);
|
|
|
|
|
|
|
|
if (!err) {
|
net: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group
When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.
[1]
$ ip nexthop list
id 200 via 2002:db8::2 dev bridge.10 scope link onlink
id 201 via 2002:db8::3 dev bridge scope link onlink
id 203 group 201/200
$ ip -6 route
2001:db8::10 nhid 203 metric 1024 pref medium
nexthop via 2002:db8::3 dev bridge weight 1 onlink
nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink
Create rt6_info through one of the multipath legs, e.g.:
$ taskset -a -c 1 ./pkt_inj 24 bridge.10 2001:db8::10
(pkt_inj is just a custom packet generator, nothing special)
Then remove that leg from the group by replace (let's assume it is id
200 in this case):
$ ip nexthop replace id 203 group 201
Now remove the IPv6 route:
$ ip -6 route del 2001:db8::10/128
The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
(deleted) IPv6 route holds 1 reference over nhid 203
nh 203 holds 1 ref over id 201
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
$ ip nexthop replace id 203 group 201/200
And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.
To trigger the bug just delete the group (nhid 203):
$ ip nexthop del id 203
It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.
At this point the dependencies are:
(deleted, only unlinked) IPv6 route holds reference over group nh 203
(deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
nh 200 holds 1 ref over the net device and the route due to the stale
rt6_info
This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.
If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
$ ip nexthop del id 200
$ ip nexthop
$
Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.
Message from syslogd@dev at Nov 19 14:04:10 ...
kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Message from syslogd@dev at Nov 19 14:04:20 ...
kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-11-22 15:15:13 +00:00
|
|
|
nh_rt_cache_flush(net, old, new);
|
2019-06-08 21:53:35 +00:00
|
|
|
|
|
|
|
__remove_nexthop(net, new, NULL);
|
|
|
|
nexthop_put(new);
|
|
|
|
}
|
|
|
|
|
|
|
|
return err;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* called with rtnl_lock held */
|
|
|
|
static int insert_nexthop(struct net *net, struct nexthop *new_nh,
|
|
|
|
struct nh_config *cfg, struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct rb_node **pp, *parent = NULL, *next;
|
|
|
|
struct rb_root *root = &net->nexthop.rb_root;
|
|
|
|
bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
|
|
|
|
bool create = !!(cfg->nlflags & NLM_F_CREATE);
|
|
|
|
u32 new_id = new_nh->id;
|
2019-06-08 21:53:35 +00:00
|
|
|
int replace_notify = 0;
|
2019-05-24 21:43:04 +00:00
|
|
|
int rc = -EEXIST;
|
|
|
|
|
|
|
|
pp = &root->rb_node;
|
|
|
|
while (1) {
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
2020-08-26 16:48:53 +00:00
|
|
|
next = *pp;
|
2019-05-24 21:43:04 +00:00
|
|
|
if (!next)
|
|
|
|
break;
|
|
|
|
|
|
|
|
parent = next;
|
|
|
|
|
|
|
|
nh = rb_entry(parent, struct nexthop, rb_node);
|
|
|
|
if (new_id < nh->id) {
|
|
|
|
pp = &next->rb_left;
|
|
|
|
} else if (new_id > nh->id) {
|
|
|
|
pp = &next->rb_right;
|
|
|
|
} else if (replace) {
|
2021-03-11 18:03:12 +00:00
|
|
|
rc = replace_nexthop(net, nh, new_nh, cfg, extack);
|
2019-06-08 21:53:35 +00:00
|
|
|
if (!rc) {
|
2019-05-24 21:43:04 +00:00
|
|
|
new_nh = nh; /* send notification with old nh */
|
2019-06-08 21:53:35 +00:00
|
|
|
replace_notify = 1;
|
|
|
|
}
|
2019-05-24 21:43:04 +00:00
|
|
|
goto out;
|
|
|
|
} else {
|
|
|
|
/* id already exists and not a replace */
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (replace && !create) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
|
|
|
|
rc = -ENOENT;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (new_nh->is_group) {
|
|
|
|
struct nh_group *nhg = rtnl_dereference(new_nh->nh_grp);
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
|
|
|
|
if (nhg->resilient) {
|
|
|
|
res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
|
|
|
|
/* Not passing the number of buckets is OK when
|
|
|
|
* replacing, but not when creating a new group.
|
|
|
|
*/
|
|
|
|
if (!cfg->nh_grp_res_has_num_buckets) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Number of buckets not specified for nexthop group insertion");
|
|
|
|
rc = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
nh_res_group_rebalance(nhg, res_table);
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
|
|
|
|
/* Do not send bucket notifications, we do full
|
|
|
|
* notification below.
|
|
|
|
*/
|
2021-03-11 18:03:24 +00:00
|
|
|
nh_res_table_upkeep(res_table, false, false);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
rb_link_node_rcu(&new_nh->rb_node, parent, pp);
|
|
|
|
rb_insert_color(&new_nh->rb_node, root);
|
2020-11-04 13:30:29 +00:00
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
/* The initial insertion is a full notification for hash-threshold as
|
|
|
|
* well as resilient groups.
|
nexthop: Implement notifiers for resilient nexthop groups
Implement the following notifications towards drivers:
- NEXTHOP_EVENT_REPLACE, when a resilient nexthop group is created.
- NEXTHOP_EVENT_BUCKET_REPLACE any time there is a change in assignment of
next hops to hash table buckets. That includes replacements, deletions,
and delayed upkeep cycles. Some bucket notifications can be vetoed by the
driver, to make it possible to propagate bucket busy-ness flags from the
HW back to the algorithm. Some are however forced, e.g. if a next hop is
deleted, all buckets that use this next hop simply must be migrated,
whether the HW wishes so or not.
- NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, before a resilient nexthop group is
replaced. Usually the driver will get the bucket notifications as well,
and could veto those. But in some cases, a bucket may not be migrated
immediately, but during delayed upkeep, and that is too late to roll the
transaction back. This notification allows the driver to take a look and
veto the new proposed group up front, before anything is committed.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:18 +00:00
|
|
|
*/
|
2020-11-04 13:30:29 +00:00
|
|
|
rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack);
|
|
|
|
if (rc)
|
|
|
|
rb_erase(&new_nh->rb_node, &net->nexthop.rb_root);
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
out:
|
|
|
|
if (!rc) {
|
|
|
|
nh_base_seq_inc(net);
|
|
|
|
nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
|
2022-07-12 00:15:33 +00:00
|
|
|
if (replace_notify &&
|
|
|
|
READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode))
|
2019-06-08 21:53:35 +00:00
|
|
|
nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
/* rtnl */
|
|
|
|
/* remove all nexthops tied to a device being deleted */
|
2021-03-04 08:57:53 +00:00
|
|
|
static void nexthop_flush_dev(struct net_device *dev, unsigned long event)
|
2019-05-24 21:43:05 +00:00
|
|
|
{
|
|
|
|
unsigned int hash = nh_dev_hashfn(dev->ifindex);
|
|
|
|
struct net *net = dev_net(dev);
|
|
|
|
struct hlist_head *head = &net->nexthop.devhash[hash];
|
|
|
|
struct hlist_node *n;
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
|
|
|
|
if (nhi->fib_nhc.nhc_dev != dev)
|
|
|
|
continue;
|
|
|
|
|
2021-03-04 08:57:53 +00:00
|
|
|
if (nhi->reject_nh &&
|
|
|
|
(event == NETDEV_DOWN || event == NETDEV_CHANGE))
|
|
|
|
continue;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
remove_nexthop(net, nhi->nh_parent, NULL);
|
2019-05-24 21:43:05 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
/* rtnl; called when net namespace is deleted */
|
|
|
|
static void flush_all_nexthops(struct net *net)
|
|
|
|
{
|
|
|
|
struct rb_root *root = &net->nexthop.rb_root;
|
|
|
|
struct rb_node *node;
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
while ((node = rb_first(root))) {
|
|
|
|
nh = rb_entry(node, struct nexthop, rb_node);
|
2019-05-24 21:43:08 +00:00
|
|
|
remove_nexthop(net, nh, NULL);
|
2019-05-24 21:43:04 +00:00
|
|
|
cond_resched();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
static struct nexthop *nexthop_create_group(struct net *net,
|
|
|
|
struct nh_config *cfg)
|
|
|
|
{
|
|
|
|
struct nlattr *grps_attr = cfg->nh_grp;
|
|
|
|
struct nexthop_grp *entry = nla_data(grps_attr);
|
2020-05-26 18:56:15 +00:00
|
|
|
u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
|
2019-05-24 21:43:08 +00:00
|
|
|
struct nh_group *nhg;
|
|
|
|
struct nexthop *nh;
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
int err;
|
2019-05-24 21:43:08 +00:00
|
|
|
int i;
|
|
|
|
|
2020-08-22 12:06:36 +00:00
|
|
|
if (WARN_ON(!num_nh))
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
nh = nexthop_alloc();
|
|
|
|
if (!nh)
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
nh->is_group = 1;
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
nhg = nexthop_grp_alloc(num_nh);
|
2019-05-24 21:43:08 +00:00
|
|
|
if (!nhg) {
|
|
|
|
kfree(nh);
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
}
|
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
/* spare group used for removals */
|
|
|
|
nhg->spare = nexthop_grp_alloc(num_nh);
|
2020-06-01 11:12:01 +00:00
|
|
|
if (!nhg->spare) {
|
2020-05-26 18:56:15 +00:00
|
|
|
kfree(nhg);
|
|
|
|
kfree(nh);
|
2020-06-01 11:12:01 +00:00
|
|
|
return ERR_PTR(-ENOMEM);
|
2020-05-26 18:56:15 +00:00
|
|
|
}
|
|
|
|
nhg->spare->spare = nhg;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
for (i = 0; i < nhg->num_nh; ++i) {
|
|
|
|
struct nexthop *nhe;
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
nhe = nexthop_find_by_id(net, entry[i].id);
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
if (!nexthop_get(nhe)) {
|
|
|
|
err = -ENOENT;
|
2019-05-24 21:43:08 +00:00
|
|
|
goto out_no_nh;
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
nhi = rtnl_dereference(nhe->nh_info);
|
|
|
|
if (nhi->family == AF_INET)
|
|
|
|
nhg->has_v4 = true;
|
|
|
|
|
|
|
|
nhg->nh_entries[i].nh = nhe;
|
|
|
|
nhg->nh_entries[i].weight = entry[i].weight + 1;
|
|
|
|
list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
|
|
|
|
nhg->nh_entries[i].nh_parent = nh;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:14 +00:00
|
|
|
if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
|
2021-03-26 13:20:22 +00:00
|
|
|
nhg->hash_threshold = 1;
|
2021-03-11 18:03:14 +00:00
|
|
|
nhg->is_multipath = true;
|
2021-03-11 18:03:15 +00:00
|
|
|
} else if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES) {
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
struct nh_res_table *res_table;
|
|
|
|
|
|
|
|
res_table = nexthop_res_table_alloc(net, cfg->nh_id, cfg);
|
|
|
|
if (!res_table) {
|
|
|
|
err = -ENOMEM;
|
|
|
|
goto out_no_nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
rcu_assign_pointer(nhg->spare->res_table, res_table);
|
|
|
|
rcu_assign_pointer(nhg->res_table, res_table);
|
|
|
|
nhg->resilient = true;
|
|
|
|
nhg->is_multipath = true;
|
2021-03-11 18:03:14 +00:00
|
|
|
}
|
2021-01-28 12:49:16 +00:00
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
WARN_ON_ONCE(nhg->hash_threshold + nhg->resilient != 1);
|
2021-01-28 12:49:16 +00:00
|
|
|
|
2021-03-26 13:20:22 +00:00
|
|
|
if (nhg->hash_threshold)
|
|
|
|
nh_hthr_group_rebalance(nhg);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (cfg->nh_fdb)
|
2020-06-09 02:54:43 +00:00
|
|
|
nhg->fdb_nh = 1;
|
2020-05-22 05:26:13 +00:00
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
rcu_assign_pointer(nh->nh_grp, nhg);
|
|
|
|
|
|
|
|
return nh;
|
|
|
|
|
|
|
|
out_no_nh:
|
2021-01-07 14:48:22 +00:00
|
|
|
for (i--; i >= 0; --i) {
|
|
|
|
list_del(&nhg->nh_entries[i].nh_list);
|
2019-05-24 21:43:08 +00:00
|
|
|
nexthop_put(nhg->nh_entries[i].nh);
|
2021-01-07 14:48:22 +00:00
|
|
|
}
|
2019-05-24 21:43:08 +00:00
|
|
|
|
2020-05-26 18:56:15 +00:00
|
|
|
kfree(nhg->spare);
|
2019-05-24 21:43:08 +00:00
|
|
|
kfree(nhg);
|
|
|
|
kfree(nh);
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 18:03:16 +00:00
|
|
|
return ERR_PTR(err);
|
2019-05-24 21:43:08 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
static int nh_create_ipv4(struct net *net, struct nexthop *nh,
|
|
|
|
struct nh_info *nhi, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct fib_nh *fib_nh = &nhi->fib_nh;
|
|
|
|
struct fib_config fib_cfg = {
|
|
|
|
.fc_oif = cfg->nh_ifindex,
|
|
|
|
.fc_gw4 = cfg->gw.ipv4,
|
|
|
|
.fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
|
|
|
|
.fc_flags = cfg->nh_flags,
|
2021-09-02 05:20:14 +00:00
|
|
|
.fc_nlinfo = cfg->nlinfo,
|
2019-05-24 21:43:07 +00:00
|
|
|
.fc_encap = cfg->nh_encap,
|
|
|
|
.fc_encap_type = cfg->nh_encap_type,
|
2019-05-24 21:43:05 +00:00
|
|
|
};
|
2020-05-22 05:26:13 +00:00
|
|
|
u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN);
|
2019-08-22 12:53:40 +00:00
|
|
|
int err;
|
2019-05-24 21:43:05 +00:00
|
|
|
|
|
|
|
err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
|
|
|
|
if (err) {
|
|
|
|
fib_nh_release(net, fib_nh);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2020-06-09 02:54:43 +00:00
|
|
|
if (nhi->fdb_nh)
|
2020-05-22 05:26:13 +00:00
|
|
|
goto out;
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
/* sets nh_dev if successful */
|
|
|
|
err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
|
|
|
|
if (!err) {
|
|
|
|
nh->nh_flags = fib_nh->fib_nh_flags;
|
2019-06-04 03:19:50 +00:00
|
|
|
fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
|
2022-10-20 10:09:52 +00:00
|
|
|
!fib_nh->fib_nh_scope ? 0 : fib_nh->fib_nh_scope - 1);
|
2019-05-24 21:43:05 +00:00
|
|
|
} else {
|
|
|
|
fib_nh_release(net, fib_nh);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:06 +00:00
|
|
|
static int nh_create_ipv6(struct net *net, struct nexthop *nh,
|
|
|
|
struct nh_info *nhi, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct fib6_nh *fib6_nh = &nhi->fib6_nh;
|
|
|
|
struct fib6_config fib6_cfg = {
|
|
|
|
.fc_table = l3mdev_fib_table(cfg->dev),
|
|
|
|
.fc_ifindex = cfg->nh_ifindex,
|
|
|
|
.fc_gateway = cfg->gw.ipv6,
|
|
|
|
.fc_flags = cfg->nh_flags,
|
2021-09-02 05:20:14 +00:00
|
|
|
.fc_nlinfo = cfg->nlinfo,
|
2019-05-24 21:43:07 +00:00
|
|
|
.fc_encap = cfg->nh_encap,
|
|
|
|
.fc_encap_type = cfg->nh_encap_type,
|
2020-05-22 05:26:13 +00:00
|
|
|
.fc_is_fdb = cfg->nh_fdb,
|
2019-05-24 21:43:06 +00:00
|
|
|
};
|
2019-05-30 15:57:54 +00:00
|
|
|
int err;
|
2019-05-24 21:43:06 +00:00
|
|
|
|
|
|
|
if (!ipv6_addr_any(&cfg->gw.ipv6))
|
|
|
|
fib6_cfg.fc_flags |= RTF_GATEWAY;
|
|
|
|
|
|
|
|
/* sets nh_dev if successful */
|
|
|
|
err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
|
|
|
|
extack);
|
2021-11-23 10:27:19 +00:00
|
|
|
if (err) {
|
|
|
|
/* IPv6 is not enabled, don't call fib6_nh_release */
|
|
|
|
if (err == -EAFNOSUPPORT)
|
|
|
|
goto out;
|
2019-05-24 21:43:06 +00:00
|
|
|
ipv6_stub->fib6_nh_release(fib6_nh);
|
2021-11-23 10:27:19 +00:00
|
|
|
} else {
|
2019-05-24 21:43:06 +00:00
|
|
|
nh->nh_flags = fib6_nh->fib_nh_flags;
|
2021-11-23 10:27:19 +00:00
|
|
|
}
|
|
|
|
out:
|
2019-05-24 21:43:06 +00:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_info *nhi;
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
nh = nexthop_alloc();
|
|
|
|
if (!nh)
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
|
|
|
|
if (!nhi) {
|
|
|
|
kfree(nh);
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
}
|
|
|
|
|
|
|
|
nh->nh_flags = cfg->nh_flags;
|
|
|
|
nh->net = net;
|
|
|
|
|
|
|
|
nhi->nh_parent = nh;
|
|
|
|
nhi->family = cfg->nh_family;
|
|
|
|
nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (cfg->nh_fdb)
|
2020-06-09 02:54:43 +00:00
|
|
|
nhi->fdb_nh = 1;
|
2020-05-22 05:26:13 +00:00
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
if (cfg->nh_blackhole) {
|
|
|
|
nhi->reject_nh = 1;
|
|
|
|
cfg->nh_ifindex = net->loopback_dev->ifindex;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
switch (cfg->nh_family) {
|
|
|
|
case AF_INET:
|
|
|
|
err = nh_create_ipv4(net, nh, nhi, cfg, extack);
|
|
|
|
break;
|
2019-05-24 21:43:06 +00:00
|
|
|
case AF_INET6:
|
|
|
|
err = nh_create_ipv6(net, nh, nhi, cfg, extack);
|
|
|
|
break;
|
2019-05-24 21:43:05 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
if (err) {
|
|
|
|
kfree(nhi);
|
|
|
|
kfree(nh);
|
|
|
|
return ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
/* add the entry to the device based hash */
|
2020-06-09 02:54:43 +00:00
|
|
|
if (!nhi->fdb_nh)
|
2020-05-22 05:26:13 +00:00
|
|
|
nexthop_devhash_add(net, nhi);
|
2019-05-24 21:43:05 +00:00
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
rcu_assign_pointer(nh->nh_info, nhi);
|
|
|
|
|
|
|
|
return nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* called with rtnl lock held */
|
|
|
|
static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!cfg->nh_id) {
|
|
|
|
cfg->nh_id = nh_find_unused_id(net);
|
|
|
|
if (!cfg->nh_id) {
|
|
|
|
NL_SET_ERR_MSG(extack, "No unused id");
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (cfg->nh_grp)
|
|
|
|
nh = nexthop_create_group(net, cfg);
|
|
|
|
else
|
|
|
|
nh = nexthop_create(net, cfg, extack);
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
if (IS_ERR(nh))
|
|
|
|
return nh;
|
|
|
|
|
|
|
|
refcount_set(&nh->refcnt, 1);
|
|
|
|
nh->id = cfg->nh_id;
|
|
|
|
nh->protocol = cfg->nh_protocol;
|
|
|
|
nh->net = net;
|
|
|
|
|
|
|
|
err = insert_nexthop(net, nh, cfg, extack);
|
|
|
|
if (err) {
|
2019-05-24 21:43:08 +00:00
|
|
|
__remove_nexthop(net, nh, NULL);
|
2019-05-24 21:43:04 +00:00
|
|
|
nexthop_put(nh);
|
|
|
|
nh = ERR_PTR(err);
|
|
|
|
}
|
|
|
|
|
|
|
|
return nh;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:21 +00:00
|
|
|
static int rtm_nh_get_timer(struct nlattr *attr, unsigned long fallback,
|
|
|
|
unsigned long *timer_p, bool *has_p,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
unsigned long timer;
|
|
|
|
u32 value;
|
|
|
|
|
|
|
|
if (!attr) {
|
|
|
|
*timer_p = fallback;
|
|
|
|
*has_p = false;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
value = nla_get_u32(attr);
|
|
|
|
timer = clock_t_to_jiffies(value);
|
|
|
|
if (timer == ~0UL) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Timer value too large");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
*timer_p = timer;
|
|
|
|
*has_p = true;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rtm_to_nh_config_grp_res(struct nlattr *res, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_policy_new)] = {};
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (res) {
|
|
|
|
err = nla_parse_nested(tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_res_policy_new) - 1,
|
|
|
|
res, rtm_nh_res_policy_new, extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (tb[NHA_RES_GROUP_BUCKETS]) {
|
|
|
|
cfg->nh_grp_res_num_buckets =
|
|
|
|
nla_get_u16(tb[NHA_RES_GROUP_BUCKETS]);
|
|
|
|
cfg->nh_grp_res_has_num_buckets = true;
|
|
|
|
if (!cfg->nh_grp_res_num_buckets) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Number of buckets needs to be non-0");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
err = rtm_nh_get_timer(tb[NHA_RES_GROUP_IDLE_TIMER],
|
|
|
|
NH_RES_DEFAULT_IDLE_TIMER,
|
|
|
|
&cfg->nh_grp_res_idle_timer,
|
|
|
|
&cfg->nh_grp_res_has_idle_timer,
|
|
|
|
extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return rtm_nh_get_timer(tb[NHA_RES_GROUP_UNBALANCED_TIMER],
|
|
|
|
NH_RES_DEFAULT_UNBALANCED_TIMER,
|
|
|
|
&cfg->nh_grp_res_unbalanced_timer,
|
|
|
|
&cfg->nh_grp_res_has_unbalanced_timer,
|
|
|
|
extack);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
|
|
|
|
struct nlmsghdr *nlh, struct nh_config *cfg,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nhmsg *nhm = nlmsg_data(nlh);
|
2021-01-20 15:44:12 +00:00
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_new)];
|
2019-05-24 21:43:04 +00:00
|
|
|
int err;
|
|
|
|
|
2021-01-20 15:44:12 +00:00
|
|
|
err = nlmsg_parse(nlh, sizeof(*nhm), tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_policy_new) - 1,
|
|
|
|
rtm_nh_policy_new, extack);
|
2019-05-24 21:43:04 +00:00
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = -EINVAL;
|
|
|
|
if (nhm->resvd || nhm->nh_scope) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (nhm->nh_family) {
|
2019-05-24 21:43:05 +00:00
|
|
|
case AF_INET:
|
2019-05-24 21:43:06 +00:00
|
|
|
case AF_INET6:
|
2019-05-24 21:43:05 +00:00
|
|
|
break;
|
2019-05-24 21:43:08 +00:00
|
|
|
case AF_UNSPEC:
|
|
|
|
if (tb[NHA_GROUP])
|
|
|
|
break;
|
2020-03-12 22:50:22 +00:00
|
|
|
fallthrough;
|
2019-05-24 21:43:04 +00:00
|
|
|
default:
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid address family");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
memset(cfg, 0, sizeof(*cfg));
|
|
|
|
cfg->nlflags = nlh->nlmsg_flags;
|
|
|
|
cfg->nlinfo.portid = NETLINK_CB(skb).portid;
|
|
|
|
cfg->nlinfo.nlh = nlh;
|
|
|
|
cfg->nlinfo.nl_net = net;
|
|
|
|
|
|
|
|
cfg->nh_family = nhm->nh_family;
|
|
|
|
cfg->nh_protocol = nhm->nh_protocol;
|
|
|
|
cfg->nh_flags = nhm->nh_flags;
|
|
|
|
|
|
|
|
if (tb[NHA_ID])
|
|
|
|
cfg->nh_id = nla_get_u32(tb[NHA_ID]);
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (tb[NHA_FDB]) {
|
|
|
|
if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] ||
|
|
|
|
tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (nhm->nh_flags) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]);
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (tb[NHA_GROUP]) {
|
|
|
|
if (nhm->nh_family != AF_UNSPEC) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid family for group");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
cfg->nh_grp = tb[NHA_GROUP];
|
|
|
|
|
|
|
|
cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
|
|
|
|
if (tb[NHA_GROUP_TYPE])
|
|
|
|
cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
|
|
|
|
|
|
|
|
if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid group type");
|
|
|
|
goto out;
|
|
|
|
}
|
2021-03-11 18:03:21 +00:00
|
|
|
err = nh_check_attr_group(net, tb, ARRAY_SIZE(tb),
|
|
|
|
cfg->nh_grp_type, extack);
|
|
|
|
if (err)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES)
|
|
|
|
err = rtm_to_nh_config_grp_res(tb[NHA_RES_GROUP],
|
|
|
|
cfg, extack);
|
2019-05-24 21:43:08 +00:00
|
|
|
|
|
|
|
/* no other attributes should be set */
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
if (tb[NHA_BLACKHOLE]) {
|
2019-05-24 21:43:07 +00:00
|
|
|
if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
|
2020-05-22 05:26:13 +00:00
|
|
|
tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb");
|
2019-05-24 21:43:04 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
cfg->nh_blackhole = 1;
|
|
|
|
err = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (!cfg->nh_fdb && !tb[NHA_OIF]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops");
|
2019-05-24 21:43:04 +00:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (!cfg->nh_fdb && tb[NHA_OIF]) {
|
|
|
|
cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
|
|
|
|
if (cfg->nh_ifindex)
|
|
|
|
cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
|
2019-05-24 21:43:04 +00:00
|
|
|
|
2020-05-22 05:26:13 +00:00
|
|
|
if (!cfg->dev) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid device index");
|
|
|
|
goto out;
|
|
|
|
} else if (!(cfg->dev->flags & IFF_UP)) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Nexthop device is not up");
|
|
|
|
err = -ENETDOWN;
|
|
|
|
goto out;
|
|
|
|
} else if (!netif_carrier_ok(cfg->dev)) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
|
|
|
|
err = -ENETDOWN;
|
|
|
|
goto out;
|
|
|
|
}
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
err = -EINVAL;
|
|
|
|
if (tb[NHA_GATEWAY]) {
|
|
|
|
struct nlattr *gwa = tb[NHA_GATEWAY];
|
|
|
|
|
|
|
|
switch (cfg->nh_family) {
|
|
|
|
case AF_INET:
|
|
|
|
if (nla_len(gwa) != sizeof(u32)) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid gateway");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
cfg->gw.ipv4 = nla_get_be32(gwa);
|
|
|
|
break;
|
2019-05-24 21:43:06 +00:00
|
|
|
case AF_INET6:
|
|
|
|
if (nla_len(gwa) != sizeof(struct in6_addr)) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid gateway");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
cfg->gw.ipv6 = nla_get_in6_addr(gwa);
|
|
|
|
break;
|
2019-05-24 21:43:05 +00:00
|
|
|
default:
|
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"Unknown address family for gateway");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* device only nexthop (no gateway) */
|
|
|
|
if (cfg->nh_flags & RTNH_F_ONLINK) {
|
|
|
|
NL_SET_ERR_MSG(extack,
|
|
|
|
"ONLINK flag can not be set for nexthop without a gateway");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:07 +00:00
|
|
|
if (tb[NHA_ENCAP]) {
|
|
|
|
cfg->nh_encap = tb[NHA_ENCAP];
|
|
|
|
|
|
|
|
if (!tb[NHA_ENCAP_TYPE]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
|
|
|
|
err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
|
|
|
|
if (err < 0)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
} else if (tb[NHA_ENCAP_TYPE]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
err = 0;
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct net *net = sock_net(skb->sk);
|
|
|
|
struct nh_config cfg;
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
|
|
|
|
if (!err) {
|
|
|
|
nh = nexthop_add(net, &cfg, extack);
|
|
|
|
if (IS_ERR(nh))
|
|
|
|
err = PTR_ERR(nh);
|
|
|
|
}
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:24 +00:00
|
|
|
static int __nh_valid_get_del_req(const struct nlmsghdr *nlh,
|
|
|
|
struct nlattr **tb, u32 *id,
|
|
|
|
struct netlink_ext_ack *extack)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
|
|
|
struct nhmsg *nhm = nlmsg_data(nlh);
|
|
|
|
|
|
|
|
if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid values in header");
|
2021-01-28 12:49:24 +00:00
|
|
|
return -EINVAL;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!tb[NHA_ID]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Nexthop id is missing");
|
2021-01-28 12:49:24 +00:00
|
|
|
return -EINVAL;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
*id = nla_get_u32(tb[NHA_ID]);
|
2021-01-28 12:49:24 +00:00
|
|
|
if (!(*id)) {
|
2019-05-24 21:43:04 +00:00
|
|
|
NL_SET_ERR_MSG(extack, "Invalid nexthop id");
|
2021-01-28 12:49:24 +00:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_valid_get_del_req(const struct nlmsghdr *nlh, u32 *id,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get)];
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_policy_get) - 1,
|
|
|
|
rtm_nh_policy_get, extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return __nh_valid_get_del_req(nlh, tb, id, extack);
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct net *net = sock_net(skb->sk);
|
|
|
|
struct nl_info nlinfo = {
|
|
|
|
.nlh = nlh,
|
|
|
|
.nl_net = net,
|
|
|
|
.portid = NETLINK_CB(skb).portid,
|
|
|
|
};
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err;
|
|
|
|
u32 id;
|
|
|
|
|
|
|
|
err = nh_valid_get_del_req(nlh, &id, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
nh = nexthop_find_by_id(net, id);
|
|
|
|
if (!nh)
|
|
|
|
return -ENOENT;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
remove_nexthop(net, nh, &nlinfo);
|
2019-05-24 21:43:04 +00:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct net *net = sock_net(in_skb->sk);
|
|
|
|
struct sk_buff *skb = NULL;
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err;
|
|
|
|
u32 id;
|
|
|
|
|
|
|
|
err = nh_valid_get_del_req(nlh, &id, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = -ENOBUFS;
|
|
|
|
skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
|
|
|
if (!skb)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
err = -ENOENT;
|
|
|
|
nh = nexthop_find_by_id(net, id);
|
|
|
|
if (!nh)
|
|
|
|
goto errout_free;
|
|
|
|
|
|
|
|
err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
|
|
|
|
nlh->nlmsg_seq, 0);
|
|
|
|
if (err < 0) {
|
|
|
|
WARN_ON(err == -EMSGSIZE);
|
|
|
|
goto errout_free;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
errout_free:
|
|
|
|
kfree_skb(skb);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:19 +00:00
|
|
|
struct nh_dump_filter {
|
2021-03-11 18:03:22 +00:00
|
|
|
u32 nh_id;
|
2021-01-28 12:49:19 +00:00
|
|
|
int dev_idx;
|
|
|
|
int master_idx;
|
|
|
|
bool group_filter;
|
|
|
|
bool fdb_filter;
|
2021-03-11 18:03:22 +00:00
|
|
|
u32 res_bucket_nh_id;
|
2021-01-28 12:49:19 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
static bool nh_dump_filtered(struct nexthop *nh,
|
|
|
|
struct nh_dump_filter *filter, u8 family)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
|
|
|
const struct net_device *dev;
|
|
|
|
const struct nh_info *nhi;
|
|
|
|
|
2021-01-28 12:49:19 +00:00
|
|
|
if (filter->group_filter && !nh->is_group)
|
2019-05-24 21:43:08 +00:00
|
|
|
return true;
|
|
|
|
|
2021-01-28 12:49:19 +00:00
|
|
|
if (!filter->dev_idx && !filter->master_idx && !family)
|
2019-05-24 21:43:04 +00:00
|
|
|
return false;
|
|
|
|
|
2019-05-24 21:43:08 +00:00
|
|
|
if (nh->is_group)
|
|
|
|
return true;
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
nhi = rtnl_dereference(nh->nh_info);
|
|
|
|
if (family && nhi->family != family)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
dev = nhi->fib_nhc.nhc_dev;
|
2021-01-28 12:49:19 +00:00
|
|
|
if (filter->dev_idx && (!dev || dev->ifindex != filter->dev_idx))
|
2019-05-24 21:43:04 +00:00
|
|
|
return true;
|
|
|
|
|
2021-01-28 12:49:19 +00:00
|
|
|
if (filter->master_idx) {
|
2019-05-24 21:43:04 +00:00
|
|
|
struct net_device *master;
|
|
|
|
|
|
|
|
if (!dev)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
master = netdev_master_upper_dev_get((struct net_device *)dev);
|
2021-01-28 12:49:19 +00:00
|
|
|
if (!master || master->ifindex != filter->master_idx)
|
2019-05-24 21:43:04 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:20 +00:00
|
|
|
static int __nh_valid_dump_req(const struct nlmsghdr *nlh, struct nlattr **tb,
|
|
|
|
struct nh_dump_filter *filter,
|
|
|
|
struct netlink_ext_ack *extack)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
|
|
|
struct nhmsg *nhm;
|
|
|
|
u32 idx;
|
|
|
|
|
2021-01-20 15:44:11 +00:00
|
|
|
if (tb[NHA_OIF]) {
|
|
|
|
idx = nla_get_u32(tb[NHA_OIF]);
|
|
|
|
if (idx > INT_MAX) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid device index");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2021-01-28 12:49:19 +00:00
|
|
|
filter->dev_idx = idx;
|
2021-01-20 15:44:11 +00:00
|
|
|
}
|
|
|
|
if (tb[NHA_MASTER]) {
|
|
|
|
idx = nla_get_u32(tb[NHA_MASTER]);
|
|
|
|
if (idx > INT_MAX) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid master device index");
|
2019-05-24 21:43:04 +00:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
2021-01-28 12:49:19 +00:00
|
|
|
filter->master_idx = idx;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
2021-01-28 12:49:19 +00:00
|
|
|
filter->group_filter = nla_get_flag(tb[NHA_GROUPS]);
|
|
|
|
filter->fdb_filter = nla_get_flag(tb[NHA_FDB]);
|
2019-05-24 21:43:04 +00:00
|
|
|
|
|
|
|
nhm = nlmsg_data(nlh);
|
|
|
|
if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:20 +00:00
|
|
|
static int nh_valid_dump_req(const struct nlmsghdr *nlh,
|
|
|
|
struct nh_dump_filter *filter,
|
|
|
|
struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump)];
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_policy_dump) - 1,
|
|
|
|
rtm_nh_policy_dump, cb->extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return __nh_valid_dump_req(nlh, tb, filter, cb->extack);
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:21 +00:00
|
|
|
struct rtm_dump_nh_ctx {
|
|
|
|
u32 idx;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct rtm_dump_nh_ctx *
|
|
|
|
rtm_dump_nh_ctx(struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct rtm_dump_nh_ctx *ctx = (void *)cb->ctx;
|
|
|
|
|
|
|
|
BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
|
|
|
|
return ctx;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:22 +00:00
|
|
|
static int rtm_dump_walk_nexthops(struct sk_buff *skb,
|
|
|
|
struct netlink_callback *cb,
|
|
|
|
struct rb_root *root,
|
|
|
|
struct rtm_dump_nh_ctx *ctx,
|
2021-01-28 12:49:23 +00:00
|
|
|
int (*nh_cb)(struct sk_buff *skb,
|
|
|
|
struct netlink_callback *cb,
|
|
|
|
struct nexthop *nh, void *data),
|
|
|
|
void *data)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
|
|
|
struct rb_node *node;
|
2021-04-16 15:55:34 +00:00
|
|
|
int s_idx;
|
2019-05-24 21:43:04 +00:00
|
|
|
int err;
|
|
|
|
|
2021-01-28 12:49:21 +00:00
|
|
|
s_idx = ctx->idx;
|
2019-05-24 21:43:04 +00:00
|
|
|
for (node = rb_first(root); node; node = rb_next(node)) {
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
nh = rb_entry(node, struct nexthop, rb_node);
|
2021-04-16 15:55:34 +00:00
|
|
|
if (nh->id < s_idx)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
ctx->idx = nh->id;
|
2021-01-28 12:49:23 +00:00
|
|
|
err = nh_cb(skb, cb, nh, data);
|
|
|
|
if (err)
|
2021-01-28 12:49:22 +00:00
|
|
|
return err;
|
2019-05-24 21:43:04 +00:00
|
|
|
}
|
|
|
|
|
2021-04-16 15:55:34 +00:00
|
|
|
ctx->idx++;
|
2021-01-28 12:49:22 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:23 +00:00
|
|
|
static int rtm_dump_nexthop_cb(struct sk_buff *skb, struct netlink_callback *cb,
|
|
|
|
struct nexthop *nh, void *data)
|
|
|
|
{
|
|
|
|
struct nhmsg *nhm = nlmsg_data(cb->nlh);
|
|
|
|
struct nh_dump_filter *filter = data;
|
|
|
|
|
|
|
|
if (nh_dump_filtered(nh, filter, nhm->nh_family))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
return nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
|
|
|
|
NETLINK_CB(cb->skb).portid,
|
|
|
|
cb->nlh->nlmsg_seq, NLM_F_MULTI);
|
|
|
|
}
|
|
|
|
|
2021-01-28 12:49:22 +00:00
|
|
|
/* rtnl */
|
|
|
|
static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct rtm_dump_nh_ctx *ctx = rtm_dump_nh_ctx(cb);
|
|
|
|
struct net *net = sock_net(skb->sk);
|
|
|
|
struct rb_root *root = &net->nexthop.rb_root;
|
|
|
|
struct nh_dump_filter filter = {};
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nh_valid_dump_req(cb->nlh, &filter, cb);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
2021-01-28 12:49:23 +00:00
|
|
|
err = rtm_dump_walk_nexthops(skb, cb, root, ctx,
|
|
|
|
&rtm_dump_nexthop_cb, &filter);
|
2021-01-28 12:49:22 +00:00
|
|
|
if (err < 0) {
|
|
|
|
if (likely(skb->len))
|
|
|
|
goto out;
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
out:
|
|
|
|
err = skb->len;
|
|
|
|
out_err:
|
|
|
|
cb->seq = net->nexthop.seq;
|
|
|
|
nl_dump_check_consistent(cb, nlmsg_hdr(skb));
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:22 +00:00
|
|
|
static struct nexthop *
|
|
|
|
nexthop_find_group_resilient(struct net *net, u32 id,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nh_group *nhg;
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
nh = nexthop_find_by_id(net, id);
|
|
|
|
if (!nh)
|
|
|
|
return ERR_PTR(-ENOENT);
|
|
|
|
|
|
|
|
if (!nh->is_group) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Not a nexthop group");
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
if (!nhg->resilient) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Nexthop group not of type resilient");
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
|
|
|
return nh;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_valid_dump_nhid(struct nlattr *attr, u32 *nh_id_p,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
u32 idx;
|
|
|
|
|
|
|
|
if (attr) {
|
|
|
|
idx = nla_get_u32(attr);
|
|
|
|
if (!idx) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Invalid nexthop id");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
*nh_id_p = idx;
|
|
|
|
} else {
|
|
|
|
*nh_id_p = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_valid_dump_bucket_req(const struct nlmsghdr *nlh,
|
|
|
|
struct nh_dump_filter *filter,
|
|
|
|
struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct nlattr *res_tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_dump)];
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump_bucket)];
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_policy_dump_bucket) - 1,
|
|
|
|
rtm_nh_policy_dump_bucket, NULL);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = nh_valid_dump_nhid(tb[NHA_ID], &filter->nh_id, cb->extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (tb[NHA_RES_BUCKET]) {
|
|
|
|
size_t max = ARRAY_SIZE(rtm_nh_res_bucket_policy_dump) - 1;
|
|
|
|
|
|
|
|
err = nla_parse_nested(res_tb, max,
|
|
|
|
tb[NHA_RES_BUCKET],
|
|
|
|
rtm_nh_res_bucket_policy_dump,
|
|
|
|
cb->extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = nh_valid_dump_nhid(res_tb[NHA_RES_BUCKET_NH_ID],
|
|
|
|
&filter->res_bucket_nh_id,
|
|
|
|
cb->extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
return __nh_valid_dump_req(nlh, tb, filter, cb->extack);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct rtm_dump_res_bucket_ctx {
|
|
|
|
struct rtm_dump_nh_ctx nh;
|
|
|
|
u16 bucket_index;
|
|
|
|
u32 done_nh_idx; /* 1 + the index of the last fully processed NH. */
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct rtm_dump_res_bucket_ctx *
|
|
|
|
rtm_dump_res_bucket_ctx(struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct rtm_dump_res_bucket_ctx *ctx = (void *)cb->ctx;
|
|
|
|
|
|
|
|
BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
|
|
|
|
return ctx;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct rtm_dump_nexthop_bucket_data {
|
|
|
|
struct rtm_dump_res_bucket_ctx *ctx;
|
|
|
|
struct nh_dump_filter filter;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int rtm_dump_nexthop_bucket_nh(struct sk_buff *skb,
|
|
|
|
struct netlink_callback *cb,
|
|
|
|
struct nexthop *nh,
|
|
|
|
struct rtm_dump_nexthop_bucket_data *dd)
|
|
|
|
{
|
|
|
|
u32 portid = NETLINK_CB(cb->skb).portid;
|
|
|
|
struct nhmsg *nhm = nlmsg_data(cb->nlh);
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
u16 bucket_index;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (dd->ctx->nh.idx < dd->ctx->done_nh_idx)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
for (bucket_index = dd->ctx->bucket_index;
|
|
|
|
bucket_index < res_table->num_nh_buckets;
|
|
|
|
bucket_index++) {
|
|
|
|
struct nh_res_bucket *bucket;
|
|
|
|
struct nh_grp_entry *nhge;
|
|
|
|
|
|
|
|
bucket = &res_table->nh_buckets[bucket_index];
|
|
|
|
nhge = rtnl_dereference(bucket->nh_entry);
|
|
|
|
if (nh_dump_filtered(nhge->nh, &dd->filter, nhm->nh_family))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (dd->filter.res_bucket_nh_id &&
|
|
|
|
dd->filter.res_bucket_nh_id != nhge->nh->id)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
err = nh_fill_res_bucket(skb, nh, bucket, bucket_index,
|
|
|
|
RTM_NEWNEXTHOPBUCKET, portid,
|
|
|
|
cb->nlh->nlmsg_seq, NLM_F_MULTI,
|
|
|
|
cb->extack);
|
|
|
|
if (err < 0) {
|
|
|
|
if (likely(skb->len))
|
|
|
|
goto out;
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
dd->ctx->done_nh_idx = dd->ctx->nh.idx + 1;
|
|
|
|
bucket_index = 0;
|
|
|
|
|
|
|
|
out:
|
|
|
|
err = skb->len;
|
|
|
|
out_err:
|
|
|
|
dd->ctx->bucket_index = bucket_index;
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rtm_dump_nexthop_bucket_cb(struct sk_buff *skb,
|
|
|
|
struct netlink_callback *cb,
|
|
|
|
struct nexthop *nh, void *data)
|
|
|
|
{
|
|
|
|
struct rtm_dump_nexthop_bucket_data *dd = data;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
|
|
|
if (!nh->is_group)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
if (!nhg->resilient)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
return rtm_dump_nexthop_bucket_nh(skb, cb, nh, dd);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int rtm_dump_nexthop_bucket(struct sk_buff *skb,
|
|
|
|
struct netlink_callback *cb)
|
|
|
|
{
|
|
|
|
struct rtm_dump_res_bucket_ctx *ctx = rtm_dump_res_bucket_ctx(cb);
|
|
|
|
struct rtm_dump_nexthop_bucket_data dd = { .ctx = ctx };
|
|
|
|
struct net *net = sock_net(skb->sk);
|
|
|
|
struct nexthop *nh;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nh_valid_dump_bucket_req(cb->nlh, &dd.filter, cb);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (dd.filter.nh_id) {
|
|
|
|
nh = nexthop_find_group_resilient(net, dd.filter.nh_id,
|
|
|
|
cb->extack);
|
|
|
|
if (IS_ERR(nh))
|
|
|
|
return PTR_ERR(nh);
|
|
|
|
err = rtm_dump_nexthop_bucket_nh(skb, cb, nh, &dd);
|
|
|
|
} else {
|
|
|
|
struct rb_root *root = &net->nexthop.rb_root;
|
|
|
|
|
|
|
|
err = rtm_dump_walk_nexthops(skb, cb, root, &ctx->nh,
|
|
|
|
&rtm_dump_nexthop_bucket_cb, &dd);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (err < 0) {
|
|
|
|
if (likely(skb->len))
|
|
|
|
goto out;
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
err = skb->len;
|
|
|
|
out_err:
|
|
|
|
cb->seq = net->nexthop.seq;
|
|
|
|
nl_dump_check_consistent(cb, nlmsg_hdr(skb));
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2021-03-11 18:03:23 +00:00
|
|
|
static int nh_valid_get_bucket_req_res_bucket(struct nlattr *res,
|
|
|
|
u16 *bucket_index,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_get)];
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nla_parse_nested(tb, ARRAY_SIZE(rtm_nh_res_bucket_policy_get) - 1,
|
|
|
|
res, rtm_nh_res_bucket_policy_get, extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (!tb[NHA_RES_BUCKET_INDEX]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Bucket index is missing");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
*bucket_index = nla_get_u16(tb[NHA_RES_BUCKET_INDEX]);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh,
|
|
|
|
u32 *id, u16 *bucket_index,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get_bucket)];
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
|
|
|
|
ARRAY_SIZE(rtm_nh_policy_get_bucket) - 1,
|
|
|
|
rtm_nh_policy_get_bucket, extack);
|
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
err = __nh_valid_get_del_req(nlh, tb, id, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (!tb[NHA_RES_BUCKET]) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Bucket information is missing");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = nh_valid_get_bucket_req_res_bucket(tb[NHA_RES_BUCKET],
|
|
|
|
bucket_index, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int rtm_get_nexthop_bucket(struct sk_buff *in_skb, struct nlmsghdr *nlh,
|
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct net *net = sock_net(in_skb->sk);
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
struct sk_buff *skb = NULL;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
struct nexthop *nh;
|
|
|
|
u16 bucket_index;
|
|
|
|
int err;
|
|
|
|
u32 id;
|
|
|
|
|
|
|
|
err = nh_valid_get_bucket_req(nlh, &id, &bucket_index, extack);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
nh = nexthop_find_group_resilient(net, id, extack);
|
|
|
|
if (IS_ERR(nh))
|
|
|
|
return PTR_ERR(nh);
|
|
|
|
|
|
|
|
nhg = rtnl_dereference(nh->nh_grp);
|
|
|
|
res_table = rtnl_dereference(nhg->res_table);
|
|
|
|
if (bucket_index >= res_table->num_nh_buckets) {
|
|
|
|
NL_SET_ERR_MSG(extack, "Bucket index out of bounds");
|
|
|
|
return -ENOENT;
|
|
|
|
}
|
|
|
|
|
|
|
|
skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
|
|
|
if (!skb)
|
|
|
|
return -ENOBUFS;
|
|
|
|
|
|
|
|
err = nh_fill_res_bucket(skb, nh, &res_table->nh_buckets[bucket_index],
|
|
|
|
bucket_index, RTM_NEWNEXTHOPBUCKET,
|
|
|
|
NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
|
|
|
|
0, extack);
|
|
|
|
if (err < 0) {
|
|
|
|
WARN_ON(err == -EMSGSIZE);
|
|
|
|
goto errout_free;
|
|
|
|
}
|
|
|
|
|
|
|
|
return rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
|
|
|
|
|
|
|
|
errout_free:
|
|
|
|
kfree_skb(skb);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
|
|
|
|
{
|
|
|
|
unsigned int hash = nh_dev_hashfn(dev->ifindex);
|
|
|
|
struct net *net = dev_net(dev);
|
|
|
|
struct hlist_head *head = &net->nexthop.devhash[hash];
|
|
|
|
struct hlist_node *n;
|
|
|
|
struct nh_info *nhi;
|
|
|
|
|
|
|
|
hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
|
|
|
|
if (nhi->fib_nhc.nhc_dev == dev) {
|
|
|
|
if (nhi->family == AF_INET)
|
|
|
|
fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
|
|
|
|
orig_mtu);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rtnl */
|
|
|
|
static int nh_netdev_event(struct notifier_block *this,
|
|
|
|
unsigned long event, void *ptr)
|
|
|
|
{
|
|
|
|
struct net_device *dev = netdev_notifier_info_to_dev(ptr);
|
|
|
|
struct netdev_notifier_info_ext *info_ext;
|
|
|
|
|
|
|
|
switch (event) {
|
|
|
|
case NETDEV_DOWN:
|
|
|
|
case NETDEV_UNREGISTER:
|
2021-03-04 08:57:53 +00:00
|
|
|
nexthop_flush_dev(dev, event);
|
2019-05-24 21:43:05 +00:00
|
|
|
break;
|
|
|
|
case NETDEV_CHANGE:
|
|
|
|
if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
|
2021-03-04 08:57:53 +00:00
|
|
|
nexthop_flush_dev(dev, event);
|
2019-05-24 21:43:05 +00:00
|
|
|
break;
|
|
|
|
case NETDEV_CHANGEMTU:
|
|
|
|
info_ext = ptr;
|
|
|
|
nexthop_sync_mtu(dev, info_ext->ext.mtu);
|
|
|
|
rt_cache_flush(dev_net(dev));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return NOTIFY_DONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct notifier_block nh_netdev_notifier = {
|
|
|
|
.notifier_call = nh_netdev_event,
|
|
|
|
};
|
|
|
|
|
2020-11-04 13:30:35 +00:00
|
|
|
static int nexthops_dump(struct net *net, struct notifier_block *nb,
|
nexthop: Fix memory leaks in nexthop notification chain listeners
syzkaller discovered memory leaks [1] that can be reduced to the
following commands:
# ip nexthop add id 1 blackhole
# devlink dev reload pci/0000:06:00.0
As part of the reload flow, mlxsw will unregister its netdevs and then
unregister from the nexthop notification chain. Before unregistering
from the notification chain, mlxsw will receive delete notifications for
nexthop objects using netdevs registered by mlxsw or their uppers. mlxsw
will not receive notifications for nexthops using netdevs that are not
dismantled as part of the reload flow. For example, the blackhole
nexthop above that internally uses the loopback netdev as its nexthop
device.
One way to fix this problem is to have listeners flush their nexthop
tables after unregistering from the notification chain. This is
error-prone as evident by this patch and also not symmetric with the
registration path where a listener receives a dump of all the existing
nexthops.
Therefore, fix this problem by replaying delete notifications for the
listener being unregistered. This is symmetric to the registration path
and also consistent with the netdev notification chain.
The above means that unregister_nexthop_notifier(), like
register_nexthop_notifier(), will have to take RTNL in order to iterate
over the existing nexthops and that any callers of the function cannot
hold RTNL. This is true for mlxsw and netdevsim, but not for the VXLAN
driver. To avoid a deadlock, change the latter to unregister its nexthop
listener without holding RTNL, making it symmetric to the registration
path.
[1]
unreferenced object 0xffff88806173d600 (size 512):
comm "syz-executor.0", pid 1290, jiffies 4295583142 (age 143.507s)
hex dump (first 32 bytes):
41 9d 1e 60 80 88 ff ff 08 d6 73 61 80 88 ff ff A..`......sa....
08 d6 73 61 80 88 ff ff 01 00 00 00 00 00 00 00 ..sa............
backtrace:
[<ffffffff81a6b576>] kmemleak_alloc_recursive include/linux/kmemleak.h:43 [inline]
[<ffffffff81a6b576>] slab_post_alloc_hook+0x96/0x490 mm/slab.h:522
[<ffffffff81a716d3>] slab_alloc_node mm/slub.c:3206 [inline]
[<ffffffff81a716d3>] slab_alloc mm/slub.c:3214 [inline]
[<ffffffff81a716d3>] kmem_cache_alloc_trace+0x163/0x370 mm/slub.c:3231
[<ffffffff82e8681a>] kmalloc include/linux/slab.h:591 [inline]
[<ffffffff82e8681a>] kzalloc include/linux/slab.h:721 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_group_create drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:4918 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_new drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5054 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_event+0x59a/0x2910 drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5239
[<ffffffff813ef67d>] notifier_call_chain+0xbd/0x210 kernel/notifier.c:83
[<ffffffff813f0662>] blocking_notifier_call_chain kernel/notifier.c:318 [inline]
[<ffffffff813f0662>] blocking_notifier_call_chain+0x72/0xa0 kernel/notifier.c:306
[<ffffffff8384b9c6>] call_nexthop_notifiers+0x156/0x310 net/ipv4/nexthop.c:244
[<ffffffff83852bd8>] insert_nexthop net/ipv4/nexthop.c:2336 [inline]
[<ffffffff83852bd8>] nexthop_add net/ipv4/nexthop.c:2644 [inline]
[<ffffffff83852bd8>] rtm_new_nexthop+0x14e8/0x4d10 net/ipv4/nexthop.c:2913
[<ffffffff833e9a78>] rtnetlink_rcv_msg+0x448/0xbf0 net/core/rtnetlink.c:5572
[<ffffffff83608703>] netlink_rcv_skb+0x173/0x480 net/netlink/af_netlink.c:2504
[<ffffffff833de032>] rtnetlink_rcv+0x22/0x30 net/core/rtnetlink.c:5590
[<ffffffff836069de>] netlink_unicast_kernel net/netlink/af_netlink.c:1314 [inline]
[<ffffffff836069de>] netlink_unicast+0x5ae/0x7f0 net/netlink/af_netlink.c:1340
[<ffffffff83607501>] netlink_sendmsg+0x8e1/0xe30 net/netlink/af_netlink.c:1929
[<ffffffff832fde84>] sock_sendmsg_nosec net/socket.c:704 [inline]
[<ffffffff832fde84>] sock_sendmsg net/socket.c:724 [inline]
[<ffffffff832fde84>] ____sys_sendmsg+0x874/0x9f0 net/socket.c:2409
[<ffffffff83304a44>] ___sys_sendmsg+0x104/0x170 net/socket.c:2463
[<ffffffff83304c01>] __sys_sendmsg+0x111/0x1f0 net/socket.c:2492
[<ffffffff83304d5d>] __do_sys_sendmsg net/socket.c:2501 [inline]
[<ffffffff83304d5d>] __se_sys_sendmsg net/socket.c:2499 [inline]
[<ffffffff83304d5d>] __x64_sys_sendmsg+0x7d/0xc0 net/socket.c:2499
Fixes: 2a014b200bbd ("mlxsw: spectrum_router: Add support for nexthop objects")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-09-22 10:25:40 +00:00
|
|
|
enum nexthop_event_type event_type,
|
2020-11-04 13:30:35 +00:00
|
|
|
struct netlink_ext_ack *extack)
|
|
|
|
{
|
|
|
|
struct rb_root *root = &net->nexthop.rb_root;
|
|
|
|
struct rb_node *node;
|
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
for (node = rb_first(root); node; node = rb_next(node)) {
|
|
|
|
struct nexthop *nh;
|
|
|
|
|
|
|
|
nh = rb_entry(node, struct nexthop, rb_node);
|
nexthop: Fix memory leaks in nexthop notification chain listeners
syzkaller discovered memory leaks [1] that can be reduced to the
following commands:
# ip nexthop add id 1 blackhole
# devlink dev reload pci/0000:06:00.0
As part of the reload flow, mlxsw will unregister its netdevs and then
unregister from the nexthop notification chain. Before unregistering
from the notification chain, mlxsw will receive delete notifications for
nexthop objects using netdevs registered by mlxsw or their uppers. mlxsw
will not receive notifications for nexthops using netdevs that are not
dismantled as part of the reload flow. For example, the blackhole
nexthop above that internally uses the loopback netdev as its nexthop
device.
One way to fix this problem is to have listeners flush their nexthop
tables after unregistering from the notification chain. This is
error-prone as evident by this patch and also not symmetric with the
registration path where a listener receives a dump of all the existing
nexthops.
Therefore, fix this problem by replaying delete notifications for the
listener being unregistered. This is symmetric to the registration path
and also consistent with the netdev notification chain.
The above means that unregister_nexthop_notifier(), like
register_nexthop_notifier(), will have to take RTNL in order to iterate
over the existing nexthops and that any callers of the function cannot
hold RTNL. This is true for mlxsw and netdevsim, but not for the VXLAN
driver. To avoid a deadlock, change the latter to unregister its nexthop
listener without holding RTNL, making it symmetric to the registration
path.
[1]
unreferenced object 0xffff88806173d600 (size 512):
comm "syz-executor.0", pid 1290, jiffies 4295583142 (age 143.507s)
hex dump (first 32 bytes):
41 9d 1e 60 80 88 ff ff 08 d6 73 61 80 88 ff ff A..`......sa....
08 d6 73 61 80 88 ff ff 01 00 00 00 00 00 00 00 ..sa............
backtrace:
[<ffffffff81a6b576>] kmemleak_alloc_recursive include/linux/kmemleak.h:43 [inline]
[<ffffffff81a6b576>] slab_post_alloc_hook+0x96/0x490 mm/slab.h:522
[<ffffffff81a716d3>] slab_alloc_node mm/slub.c:3206 [inline]
[<ffffffff81a716d3>] slab_alloc mm/slub.c:3214 [inline]
[<ffffffff81a716d3>] kmem_cache_alloc_trace+0x163/0x370 mm/slub.c:3231
[<ffffffff82e8681a>] kmalloc include/linux/slab.h:591 [inline]
[<ffffffff82e8681a>] kzalloc include/linux/slab.h:721 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_group_create drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:4918 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_new drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5054 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_event+0x59a/0x2910 drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5239
[<ffffffff813ef67d>] notifier_call_chain+0xbd/0x210 kernel/notifier.c:83
[<ffffffff813f0662>] blocking_notifier_call_chain kernel/notifier.c:318 [inline]
[<ffffffff813f0662>] blocking_notifier_call_chain+0x72/0xa0 kernel/notifier.c:306
[<ffffffff8384b9c6>] call_nexthop_notifiers+0x156/0x310 net/ipv4/nexthop.c:244
[<ffffffff83852bd8>] insert_nexthop net/ipv4/nexthop.c:2336 [inline]
[<ffffffff83852bd8>] nexthop_add net/ipv4/nexthop.c:2644 [inline]
[<ffffffff83852bd8>] rtm_new_nexthop+0x14e8/0x4d10 net/ipv4/nexthop.c:2913
[<ffffffff833e9a78>] rtnetlink_rcv_msg+0x448/0xbf0 net/core/rtnetlink.c:5572
[<ffffffff83608703>] netlink_rcv_skb+0x173/0x480 net/netlink/af_netlink.c:2504
[<ffffffff833de032>] rtnetlink_rcv+0x22/0x30 net/core/rtnetlink.c:5590
[<ffffffff836069de>] netlink_unicast_kernel net/netlink/af_netlink.c:1314 [inline]
[<ffffffff836069de>] netlink_unicast+0x5ae/0x7f0 net/netlink/af_netlink.c:1340
[<ffffffff83607501>] netlink_sendmsg+0x8e1/0xe30 net/netlink/af_netlink.c:1929
[<ffffffff832fde84>] sock_sendmsg_nosec net/socket.c:704 [inline]
[<ffffffff832fde84>] sock_sendmsg net/socket.c:724 [inline]
[<ffffffff832fde84>] ____sys_sendmsg+0x874/0x9f0 net/socket.c:2409
[<ffffffff83304a44>] ___sys_sendmsg+0x104/0x170 net/socket.c:2463
[<ffffffff83304c01>] __sys_sendmsg+0x111/0x1f0 net/socket.c:2492
[<ffffffff83304d5d>] __do_sys_sendmsg net/socket.c:2501 [inline]
[<ffffffff83304d5d>] __se_sys_sendmsg net/socket.c:2499 [inline]
[<ffffffff83304d5d>] __x64_sys_sendmsg+0x7d/0xc0 net/socket.c:2499
Fixes: 2a014b200bbd ("mlxsw: spectrum_router: Add support for nexthop objects")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-09-22 10:25:40 +00:00
|
|
|
err = call_nexthop_notifier(nb, net, event_type, nh, extack);
|
2020-11-04 13:30:35 +00:00
|
|
|
if (err)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2020-11-04 13:30:34 +00:00
|
|
|
int register_nexthop_notifier(struct net *net, struct notifier_block *nb,
|
|
|
|
struct netlink_ext_ack *extack)
|
2020-05-22 05:26:15 +00:00
|
|
|
{
|
2020-11-04 13:30:35 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
rtnl_lock();
|
nexthop: Fix memory leaks in nexthop notification chain listeners
syzkaller discovered memory leaks [1] that can be reduced to the
following commands:
# ip nexthop add id 1 blackhole
# devlink dev reload pci/0000:06:00.0
As part of the reload flow, mlxsw will unregister its netdevs and then
unregister from the nexthop notification chain. Before unregistering
from the notification chain, mlxsw will receive delete notifications for
nexthop objects using netdevs registered by mlxsw or their uppers. mlxsw
will not receive notifications for nexthops using netdevs that are not
dismantled as part of the reload flow. For example, the blackhole
nexthop above that internally uses the loopback netdev as its nexthop
device.
One way to fix this problem is to have listeners flush their nexthop
tables after unregistering from the notification chain. This is
error-prone as evident by this patch and also not symmetric with the
registration path where a listener receives a dump of all the existing
nexthops.
Therefore, fix this problem by replaying delete notifications for the
listener being unregistered. This is symmetric to the registration path
and also consistent with the netdev notification chain.
The above means that unregister_nexthop_notifier(), like
register_nexthop_notifier(), will have to take RTNL in order to iterate
over the existing nexthops and that any callers of the function cannot
hold RTNL. This is true for mlxsw and netdevsim, but not for the VXLAN
driver. To avoid a deadlock, change the latter to unregister its nexthop
listener without holding RTNL, making it symmetric to the registration
path.
[1]
unreferenced object 0xffff88806173d600 (size 512):
comm "syz-executor.0", pid 1290, jiffies 4295583142 (age 143.507s)
hex dump (first 32 bytes):
41 9d 1e 60 80 88 ff ff 08 d6 73 61 80 88 ff ff A..`......sa....
08 d6 73 61 80 88 ff ff 01 00 00 00 00 00 00 00 ..sa............
backtrace:
[<ffffffff81a6b576>] kmemleak_alloc_recursive include/linux/kmemleak.h:43 [inline]
[<ffffffff81a6b576>] slab_post_alloc_hook+0x96/0x490 mm/slab.h:522
[<ffffffff81a716d3>] slab_alloc_node mm/slub.c:3206 [inline]
[<ffffffff81a716d3>] slab_alloc mm/slub.c:3214 [inline]
[<ffffffff81a716d3>] kmem_cache_alloc_trace+0x163/0x370 mm/slub.c:3231
[<ffffffff82e8681a>] kmalloc include/linux/slab.h:591 [inline]
[<ffffffff82e8681a>] kzalloc include/linux/slab.h:721 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_group_create drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:4918 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_new drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5054 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_event+0x59a/0x2910 drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5239
[<ffffffff813ef67d>] notifier_call_chain+0xbd/0x210 kernel/notifier.c:83
[<ffffffff813f0662>] blocking_notifier_call_chain kernel/notifier.c:318 [inline]
[<ffffffff813f0662>] blocking_notifier_call_chain+0x72/0xa0 kernel/notifier.c:306
[<ffffffff8384b9c6>] call_nexthop_notifiers+0x156/0x310 net/ipv4/nexthop.c:244
[<ffffffff83852bd8>] insert_nexthop net/ipv4/nexthop.c:2336 [inline]
[<ffffffff83852bd8>] nexthop_add net/ipv4/nexthop.c:2644 [inline]
[<ffffffff83852bd8>] rtm_new_nexthop+0x14e8/0x4d10 net/ipv4/nexthop.c:2913
[<ffffffff833e9a78>] rtnetlink_rcv_msg+0x448/0xbf0 net/core/rtnetlink.c:5572
[<ffffffff83608703>] netlink_rcv_skb+0x173/0x480 net/netlink/af_netlink.c:2504
[<ffffffff833de032>] rtnetlink_rcv+0x22/0x30 net/core/rtnetlink.c:5590
[<ffffffff836069de>] netlink_unicast_kernel net/netlink/af_netlink.c:1314 [inline]
[<ffffffff836069de>] netlink_unicast+0x5ae/0x7f0 net/netlink/af_netlink.c:1340
[<ffffffff83607501>] netlink_sendmsg+0x8e1/0xe30 net/netlink/af_netlink.c:1929
[<ffffffff832fde84>] sock_sendmsg_nosec net/socket.c:704 [inline]
[<ffffffff832fde84>] sock_sendmsg net/socket.c:724 [inline]
[<ffffffff832fde84>] ____sys_sendmsg+0x874/0x9f0 net/socket.c:2409
[<ffffffff83304a44>] ___sys_sendmsg+0x104/0x170 net/socket.c:2463
[<ffffffff83304c01>] __sys_sendmsg+0x111/0x1f0 net/socket.c:2492
[<ffffffff83304d5d>] __do_sys_sendmsg net/socket.c:2501 [inline]
[<ffffffff83304d5d>] __se_sys_sendmsg net/socket.c:2499 [inline]
[<ffffffff83304d5d>] __x64_sys_sendmsg+0x7d/0xc0 net/socket.c:2499
Fixes: 2a014b200bbd ("mlxsw: spectrum_router: Add support for nexthop objects")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-09-22 10:25:40 +00:00
|
|
|
err = nexthops_dump(net, nb, NEXTHOP_EVENT_REPLACE, extack);
|
2020-11-04 13:30:35 +00:00
|
|
|
if (err)
|
|
|
|
goto unlock;
|
|
|
|
err = blocking_notifier_chain_register(&net->nexthop.notifier_chain,
|
|
|
|
nb);
|
|
|
|
unlock:
|
|
|
|
rtnl_unlock();
|
|
|
|
return err;
|
2020-05-22 05:26:15 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(register_nexthop_notifier);
|
|
|
|
|
|
|
|
int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb)
|
|
|
|
{
|
nexthop: Fix memory leaks in nexthop notification chain listeners
syzkaller discovered memory leaks [1] that can be reduced to the
following commands:
# ip nexthop add id 1 blackhole
# devlink dev reload pci/0000:06:00.0
As part of the reload flow, mlxsw will unregister its netdevs and then
unregister from the nexthop notification chain. Before unregistering
from the notification chain, mlxsw will receive delete notifications for
nexthop objects using netdevs registered by mlxsw or their uppers. mlxsw
will not receive notifications for nexthops using netdevs that are not
dismantled as part of the reload flow. For example, the blackhole
nexthop above that internally uses the loopback netdev as its nexthop
device.
One way to fix this problem is to have listeners flush their nexthop
tables after unregistering from the notification chain. This is
error-prone as evident by this patch and also not symmetric with the
registration path where a listener receives a dump of all the existing
nexthops.
Therefore, fix this problem by replaying delete notifications for the
listener being unregistered. This is symmetric to the registration path
and also consistent with the netdev notification chain.
The above means that unregister_nexthop_notifier(), like
register_nexthop_notifier(), will have to take RTNL in order to iterate
over the existing nexthops and that any callers of the function cannot
hold RTNL. This is true for mlxsw and netdevsim, but not for the VXLAN
driver. To avoid a deadlock, change the latter to unregister its nexthop
listener without holding RTNL, making it symmetric to the registration
path.
[1]
unreferenced object 0xffff88806173d600 (size 512):
comm "syz-executor.0", pid 1290, jiffies 4295583142 (age 143.507s)
hex dump (first 32 bytes):
41 9d 1e 60 80 88 ff ff 08 d6 73 61 80 88 ff ff A..`......sa....
08 d6 73 61 80 88 ff ff 01 00 00 00 00 00 00 00 ..sa............
backtrace:
[<ffffffff81a6b576>] kmemleak_alloc_recursive include/linux/kmemleak.h:43 [inline]
[<ffffffff81a6b576>] slab_post_alloc_hook+0x96/0x490 mm/slab.h:522
[<ffffffff81a716d3>] slab_alloc_node mm/slub.c:3206 [inline]
[<ffffffff81a716d3>] slab_alloc mm/slub.c:3214 [inline]
[<ffffffff81a716d3>] kmem_cache_alloc_trace+0x163/0x370 mm/slub.c:3231
[<ffffffff82e8681a>] kmalloc include/linux/slab.h:591 [inline]
[<ffffffff82e8681a>] kzalloc include/linux/slab.h:721 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_group_create drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:4918 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_new drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5054 [inline]
[<ffffffff82e8681a>] mlxsw_sp_nexthop_obj_event+0x59a/0x2910 drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:5239
[<ffffffff813ef67d>] notifier_call_chain+0xbd/0x210 kernel/notifier.c:83
[<ffffffff813f0662>] blocking_notifier_call_chain kernel/notifier.c:318 [inline]
[<ffffffff813f0662>] blocking_notifier_call_chain+0x72/0xa0 kernel/notifier.c:306
[<ffffffff8384b9c6>] call_nexthop_notifiers+0x156/0x310 net/ipv4/nexthop.c:244
[<ffffffff83852bd8>] insert_nexthop net/ipv4/nexthop.c:2336 [inline]
[<ffffffff83852bd8>] nexthop_add net/ipv4/nexthop.c:2644 [inline]
[<ffffffff83852bd8>] rtm_new_nexthop+0x14e8/0x4d10 net/ipv4/nexthop.c:2913
[<ffffffff833e9a78>] rtnetlink_rcv_msg+0x448/0xbf0 net/core/rtnetlink.c:5572
[<ffffffff83608703>] netlink_rcv_skb+0x173/0x480 net/netlink/af_netlink.c:2504
[<ffffffff833de032>] rtnetlink_rcv+0x22/0x30 net/core/rtnetlink.c:5590
[<ffffffff836069de>] netlink_unicast_kernel net/netlink/af_netlink.c:1314 [inline]
[<ffffffff836069de>] netlink_unicast+0x5ae/0x7f0 net/netlink/af_netlink.c:1340
[<ffffffff83607501>] netlink_sendmsg+0x8e1/0xe30 net/netlink/af_netlink.c:1929
[<ffffffff832fde84>] sock_sendmsg_nosec net/socket.c:704 [inline]
[<ffffffff832fde84>] sock_sendmsg net/socket.c:724 [inline]
[<ffffffff832fde84>] ____sys_sendmsg+0x874/0x9f0 net/socket.c:2409
[<ffffffff83304a44>] ___sys_sendmsg+0x104/0x170 net/socket.c:2463
[<ffffffff83304c01>] __sys_sendmsg+0x111/0x1f0 net/socket.c:2492
[<ffffffff83304d5d>] __do_sys_sendmsg net/socket.c:2501 [inline]
[<ffffffff83304d5d>] __se_sys_sendmsg net/socket.c:2499 [inline]
[<ffffffff83304d5d>] __x64_sys_sendmsg+0x7d/0xc0 net/socket.c:2499
Fixes: 2a014b200bbd ("mlxsw: spectrum_router: Add support for nexthop objects")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-09-22 10:25:40 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
rtnl_lock();
|
|
|
|
err = blocking_notifier_chain_unregister(&net->nexthop.notifier_chain,
|
|
|
|
nb);
|
|
|
|
if (err)
|
|
|
|
goto unlock;
|
|
|
|
nexthops_dump(net, nb, NEXTHOP_EVENT_DEL, NULL);
|
|
|
|
unlock:
|
|
|
|
rtnl_unlock();
|
|
|
|
return err;
|
2020-05-22 05:26:15 +00:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(unregister_nexthop_notifier);
|
|
|
|
|
2020-11-04 13:30:28 +00:00
|
|
|
void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap)
|
|
|
|
{
|
|
|
|
struct nexthop *nexthop;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
nexthop = nexthop_find_by_id(net, id);
|
|
|
|
if (!nexthop)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
nexthop->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
|
|
|
|
if (offload)
|
|
|
|
nexthop->nh_flags |= RTNH_F_OFFLOAD;
|
|
|
|
if (trap)
|
|
|
|
nexthop->nh_flags |= RTNH_F_TRAP;
|
|
|
|
|
|
|
|
out:
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(nexthop_set_hw_flags);
|
|
|
|
|
2021-03-11 18:03:19 +00:00
|
|
|
void nexthop_bucket_set_hw_flags(struct net *net, u32 id, u16 bucket_index,
|
|
|
|
bool offload, bool trap)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
struct nh_res_bucket *bucket;
|
|
|
|
struct nexthop *nexthop;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
nexthop = nexthop_find_by_id(net, id);
|
|
|
|
if (!nexthop || !nexthop->is_group)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
nhg = rcu_dereference(nexthop->nh_grp);
|
|
|
|
if (!nhg->resilient)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (bucket_index >= nhg->res_table->num_nh_buckets)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
res_table = rcu_dereference(nhg->res_table);
|
|
|
|
bucket = &res_table->nh_buckets[bucket_index];
|
|
|
|
bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
|
|
|
|
if (offload)
|
|
|
|
bucket->nh_flags |= RTNH_F_OFFLOAD;
|
|
|
|
if (trap)
|
|
|
|
bucket->nh_flags |= RTNH_F_TRAP;
|
|
|
|
|
|
|
|
out:
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(nexthop_bucket_set_hw_flags);
|
|
|
|
|
2021-03-11 18:03:20 +00:00
|
|
|
void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets,
|
|
|
|
unsigned long *activity)
|
|
|
|
{
|
|
|
|
struct nh_res_table *res_table;
|
|
|
|
struct nexthop *nexthop;
|
|
|
|
struct nh_group *nhg;
|
|
|
|
u16 i;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
nexthop = nexthop_find_by_id(net, id);
|
|
|
|
if (!nexthop || !nexthop->is_group)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
nhg = rcu_dereference(nexthop->nh_grp);
|
|
|
|
if (!nhg->resilient)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Instead of silently ignoring some buckets, demand that the sizes
|
|
|
|
* be the same.
|
|
|
|
*/
|
|
|
|
res_table = rcu_dereference(nhg->res_table);
|
|
|
|
if (num_buckets != res_table->num_nh_buckets)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
for (i = 0; i < num_buckets; i++) {
|
|
|
|
if (test_bit(i, activity))
|
|
|
|
nh_res_bucket_set_busy(&res_table->nh_buckets[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(nexthop_res_grp_activity_update);
|
|
|
|
|
2022-02-08 04:50:31 +00:00
|
|
|
static void __net_exit nexthop_net_exit_batch(struct list_head *net_list)
|
2019-05-24 21:43:04 +00:00
|
|
|
{
|
2022-02-08 04:50:31 +00:00
|
|
|
struct net *net;
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
rtnl_lock();
|
2022-02-08 04:50:31 +00:00
|
|
|
list_for_each_entry(net, net_list, exit_list) {
|
|
|
|
flush_all_nexthops(net);
|
|
|
|
kfree(net->nexthop.devhash);
|
|
|
|
}
|
2019-05-24 21:43:04 +00:00
|
|
|
rtnl_unlock();
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __net_init nexthop_net_init(struct net *net)
|
|
|
|
{
|
2019-05-24 21:43:05 +00:00
|
|
|
size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
net->nexthop.rb_root = RB_ROOT;
|
2019-05-24 21:43:05 +00:00
|
|
|
net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
|
|
|
|
if (!net->nexthop.devhash)
|
|
|
|
return -ENOMEM;
|
2020-09-15 11:41:01 +00:00
|
|
|
BLOCKING_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain);
|
2019-05-24 21:43:04 +00:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct pernet_operations nexthop_net_ops = {
|
|
|
|
.init = nexthop_net_init,
|
2022-02-08 04:50:31 +00:00
|
|
|
.exit_batch = nexthop_net_exit_batch,
|
2019-05-24 21:43:04 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
static int __init nexthop_init(void)
|
|
|
|
{
|
|
|
|
register_pernet_subsys(&nexthop_net_ops);
|
|
|
|
|
2019-05-24 21:43:05 +00:00
|
|
|
register_netdevice_notifier(&nh_netdev_notifier);
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
|
|
|
|
rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
|
|
|
|
rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
|
|
|
|
rtm_dump_nexthop, 0);
|
|
|
|
|
|
|
|
rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
|
|
|
|
rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
|
|
|
|
|
|
|
|
rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
|
|
|
|
rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
|
|
|
|
|
2021-03-11 18:03:23 +00:00
|
|
|
rtnl_register(PF_UNSPEC, RTM_GETNEXTHOPBUCKET, rtm_get_nexthop_bucket,
|
2021-03-11 18:03:22 +00:00
|
|
|
rtm_dump_nexthop_bucket, 0);
|
|
|
|
|
2019-05-24 21:43:04 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
subsys_initcall(nexthop_init);
|