@@ -224,7 +224,6 @@ struct net_device_core_stats {
#include <linux/static_key.h>
extern struct static_key_false rps_needed;
extern struct static_key_false rfs_needed;
-extern struct cpumask rps_default_mask;
#endif
struct neighbour;
@@ -6,6 +6,7 @@
struct ctl_table_header;
struct prot_inuse;
+struct cpumask;
struct netns_core {
/* core sysctls */
@@ -17,6 +18,10 @@ struct netns_core {
#ifdef CONFIG_PROC_FS
struct prot_inuse __percpu *prot_inuse;
#endif
+
+#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
+ struct cpumask *rps_default_mask;
+#endif
};
#endif
@@ -1060,6 +1060,18 @@ static const struct kobj_type rx_queue_ktype = {
.get_ownership = rx_queue_get_ownership,
};
+static int rx_queue_default_mask(struct net_device *dev,
+ struct netdev_rx_queue *queue)
+{
+#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
+ struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
+
+ if (rps_default_mask && !cpumask_empty(rps_default_mask))
+ return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
+#endif
+ return 0;
+}
+
static int rx_queue_add_kobject(struct net_device *dev, int index)
{
struct netdev_rx_queue *queue = dev->_rx + index;
@@ -1083,13 +1095,10 @@ static int rx_queue_add_kobject(struct net_device *dev, int index)
goto err;
}
-#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
- if (!cpumask_empty(&rps_default_mask)) {
- error = netdev_rx_queue_set_rps_mask(queue, &rps_default_mask);
- if (error)
- goto err;
- }
-#endif
+ error = rx_queue_default_mask(dev, queue);
+ if (error)
+ goto err;
+
kobject_uevent(kobj, KOBJ_ADD);
return error;
@@ -74,24 +74,46 @@ static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
#endif
#ifdef CONFIG_RPS
-struct cpumask rps_default_mask;
+
+static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
+{
+ struct cpumask *rps_default_mask;
+
+ if (net->core.rps_default_mask)
+ return net->core.rps_default_mask;
+
+ if (!zalloc_cpumask_var(&rps_default_mask, GFP_KERNEL))
+ return NULL;
+
+ /* pairs with READ_ONCE in rx_queue_default_mask() */
+ WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
+ return rps_default_mask;
+}
static int rps_default_mask_sysctl(struct ctl_table *table, int write,
void *buffer, size_t *lenp, loff_t *ppos)
{
+ struct net *net = (struct net *)table->data;
int err = 0;
rtnl_lock();
if (write) {
- err = cpumask_parse(buffer, &rps_default_mask);
+ struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
+
+ err = -ENOMEM;
+ if (!rps_default_mask)
+ goto done;
+
+ err = cpumask_parse(buffer, rps_default_mask);
if (err)
goto done;
- err = rps_cpumask_housekeeping(&rps_default_mask);
+ err = rps_cpumask_housekeeping(rps_default_mask);
if (err)
goto done;
} else {
- dump_cpumask(buffer, lenp, ppos, &rps_default_mask);
+ dump_cpumask(buffer, lenp, ppos,
+ net->core.rps_default_mask ? : cpu_none_mask);
}
done:
@@ -508,11 +530,6 @@ static struct ctl_table net_core_table[] = {
.mode = 0644,
.proc_handler = rps_sock_flow_sysctl
},
- {
- .procname = "rps_default_mask",
- .mode = 0644,
- .proc_handler = rps_default_mask_sysctl
- },
#endif
#ifdef CONFIG_NET_FLOW_LIMIT
{
@@ -639,6 +656,14 @@ static struct ctl_table net_core_table[] = {
};
static struct ctl_table netns_core_table[] = {
+#if IS_ENABLED(CONFIG_RPS)
+ {
+ .procname = "rps_default_mask",
+ .data = &init_net,
+ .mode = 0644,
+ .proc_handler = rps_default_mask_sysctl
+ },
+#endif
{
.procname = "somaxconn",
.data = &init_net.core.sysctl_somaxconn,
@@ -706,6 +731,9 @@ static __net_exit void sysctl_core_net_exit(struct net *net)
tbl = net->core.sysctl_hdr->ctl_table_arg;
unregister_net_sysctl_table(net->core.sysctl_hdr);
BUG_ON(tbl == netns_core_table);
+#if IS_ENABLED(CONFIG_RPS)
+ kfree(net->core.rps_default_mask);
+#endif
kfree(tbl);
}
@@ -716,10 +744,6 @@ static __net_initdata struct pernet_operations sysctl_core_ops = {
static __init int sysctl_core_init(void)
{
-#if IS_ENABLED(CONFIG_RPS)
- cpumask_copy(&rps_default_mask, cpu_none_mask);
-#endif
-
register_net_sysctl(&init_net, "net/core", net_core_table);
return register_pernet_subsys(&sysctl_core_ops);
}
That really was meant to be a per netns attribute from the beginning. The idea is that once proper isolation is in place in the main namespace, additional demux in the child namespaces will be redundant. Let's make child netns default rps mask empty by default. To avoid bloating the netns with a possibly large cpumask, allocate it on-demand during the first write operation. Signed-off-by: Paolo Abeni <pabeni@redhat.com> --- include/linux/netdevice.h | 1 - include/net/netns/core.h | 5 ++++ net/core/net-sysfs.c | 23 ++++++++++++------ net/core/sysctl_net_core.c | 50 ++++++++++++++++++++++++++++---------- 4 files changed, 58 insertions(+), 21 deletions(-)