return 0;
}
+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;
+}
+
static int nh_notifier_grp_info_init(struct nh_notifier_info *info,
const struct nexthop *nh)
{
if (nhg->mpath)
return nh_notifier_mp_info_init(info, nhg);
+ else if (nhg->resilient)
+ return nh_notifier_res_table_info_init(info, nhg);
return -EINVAL;
}
if (nhg->mpath)
kfree(info->nh_grp);
+ else if (nhg->resilient)
+ vfree(info->nh_res_table);
}
static int nh_notifier_info_init(struct nh_notifier_info *info,
return notifier_to_errno(err);
}
+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);
+}
+
/* There are three users of RES_TABLE, and NHs etc. referenced from there:
*
* 1) a collection of callbacks for NH maintenance. This operates under
*/
#define nh_res_dereference(p) (rcu_dereference_raw(p))
+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);
+ err = nh_notifier_mp_info_init(&info, nhg);
+ 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);
+}
+
static int call_nexthop_notifier(struct notifier_block *nb, struct net *net,
enum nexthop_event_type event_type,
struct nexthop *nh,
}
static bool nh_res_bucket_migrate(struct nh_res_table *res_table,
- u16 bucket_index, bool force)
+ u16 bucket_index, bool notify, bool force)
{
struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index];
struct nh_grp_entry *new_nhge;
+ struct netlink_ext_ack extack;
+ int err;
new_nhge = list_first_entry_or_null(&res_table->uw_nh_entries,
struct nh_grp_entry,
*/
return false;
+ 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);
+ }
+ }
+
nh_res_bucket_set_nh(bucket, new_nhge);
nh_res_bucket_set_idle(res_table, bucket);
#define NH_RES_UPKEEP_DW_MINIMUM_INTERVAL (HZ / 2)
-static void nh_res_table_upkeep(struct nh_res_table *res_table)
+static void nh_res_table_upkeep(struct nh_res_table *res_table, bool notify)
{
unsigned long now = jiffies;
unsigned long deadline;
if (nh_res_bucket_should_migrate(res_table, bucket,
&deadline, &force)) {
- if (!nh_res_bucket_migrate(res_table, i, force)) {
+ if (!nh_res_bucket_migrate(res_table, i, notify,
+ force)) {
unsigned long idle_point;
/* A driver can override the migration
struct nh_res_table *res_table;
res_table = container_of(dw, struct nh_res_table, upkeep_dw);
- nh_res_table_upkeep(res_table);
+ nh_res_table_upkeep(res_table, true);
}
static void nh_res_table_cancel_upkeep(struct nh_res_table *res_table)
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;
- nh_res_table_upkeep(old_res_table);
+ nh_res_table_upkeep(old_res_table, true);
}
static void nh_mp_group_rebalance(struct nh_group *nhg)
list_del(&nhge->nh_list);
nexthop_put(nhge->nh);
- err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, &extack);
- if (err)
- pr_err("%s\n", extack._msg);
+ /* Removal of a NH from a resilient group is notified through
+ * bucket notifications.
+ */
+ if (newg->mpath) {
+ err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
+ &extack);
+ if (err)
+ pr_err("%s\n", extack._msg);
+ }
if (nlinfo)
nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
return -EINVAL;
}
+ /* 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;
+
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)
nhg->has_v4 = has_v4;
}
+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;
+
+ if (nhg->mpath) {
+ 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;
+}
+
static int replace_nexthop_single(struct net *net, struct nexthop *old,
struct nexthop *new,
struct netlink_ext_ack *extack)
list_for_each_entry(nhge, &old->grp_list, nh_list) {
struct nexthop *nhp = nhge->nh_parent;
- err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
- extack);
+ err = replace_nexthop_single_notify(net, nhp, old, oldi, newi,
+ extack);
if (err)
goto err_notify;
}
list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) {
struct nexthop *nhp = nhge->nh_parent;
- call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, extack);
+ replace_nexthop_single_notify(net, nhp, old, newi, oldi, NULL);
}
call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack);
return err;
}
nh_res_group_rebalance(nhg, res_table);
- nh_res_table_upkeep(res_table);
+
+ /* Do not send bucket notifications, we do full
+ * notification below.
+ */
+ nh_res_table_upkeep(res_table, false);
}
}
rb_link_node_rcu(&new_nh->rb_node, parent, pp);
rb_insert_color(&new_nh->rb_node, root);
+ /* The initial insertion is a full notification for mpath as well
+ * as resilient groups.
+ */
rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack);
if (rc)
rb_erase(&new_nh->rb_node, &net->nexthop.rb_root);
projects / users / dwmw2 / linux.git / commitdiff