The commit is pushed to "branch-rh10-6.12.0-211.16.1.12.x.vz10-ovz" and will
appear at [email protected]:openvz/vzkernel.git
after rh10-6.12.0-211.16.1.12.1.vz10
------>
commit 0ed374bd0965f8bd1afc3fc199398eb836aeff8c
Author: Vladimir Riabchun <[email protected]>
Date: Mon Jun 15 19:26:25 2026 +0200
ve/sunrpc: Implement a per-ve sunrpc killer
We currently have /proc/<pid>/net/rpc/kill-tasks feature for
aborting pending RPC tasks. To fast-stop container, we need
to abort all RPC tasks, so we must iterate over all VE PIDs.
There are several problems with this approach:
1. In real life with cgroups-v2 we just process container
init PID, all other network namespaces are ignored.
2. There is a risk of breaking unrelated NFS connections if a PID,
that was a container process, is reused while we are
stopping container.
3. This may be slow - container might have a lot of processes
and a few network namespaces.
To fix this all, create a per-ve interface for aborting
RPC requests in all VE network namespaces.
https://virtuozzo.atlassian.net/browse/VSTOR-126316
Feature: nfs/nfsd: containerization
Signed-off-by: Vladimir Riabchun <[email protected]>
Reviewed-by: Pavel Tikhomirov <[email protected]>
---
include/linux/sunrpc/clnt.h | 1 +
include/linux/ve.h | 4 ++++
kernel/ve/ve.c | 46 +++++++++++++++++++++++++++++++++++++++++++++
net/sunrpc/clnt.c | 3 ++-
net/sunrpc/sunrpc_syms.c | 3 +++
5 files changed, 56 insertions(+), 1 deletion(-)
diff --git a/include/linux/sunrpc/clnt.h b/include/linux/sunrpc/clnt.h
index 7c6e9413db96c..0c4c582a1f1c3 100644
--- a/include/linux/sunrpc/clnt.h
+++ b/include/linux/sunrpc/clnt.h
@@ -257,6 +257,7 @@ void rpc_clnt_probe_trunked_xprts(struct
rpc_clnt *,
const char *rpc_proc_name(const struct rpc_task *task);
+void rpc_set_kill_net(struct net *net, bool new_value);
int rpc_task_kill_proc_init(struct net *net);
void rpc_task_kill_proc_fini(struct net *net);
diff --git a/include/linux/ve.h b/include/linux/ve.h
index b46899a447d15..7c5515548ad23 100644
--- a/include/linux/ve.h
+++ b/include/linux/ve.h
@@ -251,6 +251,8 @@ extern bool is_ve_init_net(const struct net *net);
void ve_setup_task(struct task_struct *p, struct ve_struct *ve);
+void ve_set_rpc_kill_fn(void (*fn)(struct net *, bool));
+
static inline struct vdso_image *ve_get_vdso_32(struct ve_struct *ve)
{
return ve->vdso_32;
@@ -357,6 +359,8 @@ static inline int vz_security_protocol_check(struct net
*net, int protocol) { re
static inline void ve_setup_task(struct task_struct *p, struct ve_struct *ve)
{ }
+static inline void ve_set_rpc_kill_fn(void (*fn)(struct net *, bool)) { }
+
static inline struct vdso_image *ve_get_vdso_32(struct ve_struct *ve)
{
return (struct vdso_image *)&vdso_image_32;
diff --git a/kernel/ve/ve.c b/kernel/ve/ve.c
index 231c0300e929d..7485ee4ad009a 100644
--- a/kernel/ve/ve.c
+++ b/kernel/ve/ve.c
@@ -96,6 +96,8 @@ EXPORT_SYMBOL(nr_ve);
static DEFINE_IDR(ve_idr);
+static void (__rcu *rpc_kill_net_fn)(struct net *, bool);
+
struct ve_struct *get_ve(struct ve_struct *ve)
{
if (ve)
@@ -1474,6 +1476,45 @@ static ssize_t ve_write_ctty(struct kernfs_open_file
*of, char *buf,
return ret;
}
+void ve_set_rpc_kill_fn(void (*fn)(struct net *, bool))
+{
+ rcu_assign_pointer(rpc_kill_net_fn, fn);
+ /*
+ * ve_rpc_kill_write might be using old function.
+ * rpc_set_kill_net -> NULL is dangerous, block sunrpc exit while we use
+ * functions from this module.
+ */
+ if (!fn)
+ synchronize_rcu();
+}
+EXPORT_SYMBOL(ve_set_rpc_kill_fn);
+
+static int ve_rpc_kill_write(struct cgroup_subsys_state *css,
+ struct cftype *cft, u64 val)
+{
+ struct net *net;
+ struct ve_struct *ve = css_to_ve(css);
+ void (*fn)(struct net *net, bool);
+
+ guard(rwsem_read)(&net_rwsem); /* for_each_net protection */
+ guard(rcu)(); /* Begin rpc_kill_net_fn usage section
*/
+
+ fn = rcu_dereference(rpc_kill_net_fn);
+ if (!fn) {
+ pr_info_ratelimited("SUNRPC module is not loaded.\n");
+ return 0;
+ }
+
+ for_each_net(net) {
+ if (net->owner_ve != ve)
+ continue;
+ /* rpc_set_kill_net is atomic. */
+ fn(net, val);
+ }
+
+ return 0;
+}
+
static struct cftype ve_cftypes[] = {
{
@@ -1569,6 +1610,11 @@ static struct cftype ve_cftypes[] = {
.flags = CFTYPE_ONLY_ON_ROOT,
.write = ve_write_ctty,
},
+ {
+ .name = "rpc_kill",
+ .flags = CFTYPE_NOT_ON_ROOT,
+ .write_u64 = ve_rpc_kill_write,
+ },
{ }
};
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index d794a4af01bcd..4d52ea3eb17c6 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -3433,8 +3433,9 @@ rpc_clnt_swap_deactivate(struct rpc_clnt *clnt)
EXPORT_SYMBOL_GPL(rpc_clnt_swap_deactivate);
#endif /* CONFIG_SUNRPC_SWAP */
-static void rpc_set_kill_net(struct net *net, bool new_value)
+void rpc_set_kill_net(struct net *net, bool new_value)
{
+ /* Note: function must be atomic, used under RCU read-lock. */
struct rpc_clnt *clnt;
struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
if (sn->kill_tasks == new_value)
diff --git a/net/sunrpc/sunrpc_syms.c b/net/sunrpc/sunrpc_syms.c
index bab6cab294052..7e957b482fde7 100644
--- a/net/sunrpc/sunrpc_syms.c
+++ b/net/sunrpc/sunrpc_syms.c
@@ -13,6 +13,7 @@
#include <linux/uio.h>
#include <linux/unistd.h>
#include <linux/init.h>
+#include <linux/ve.h>
#include <linux/sunrpc/sched.h>
#include <linux/sunrpc/clnt.h>
@@ -114,6 +115,7 @@ init_sunrpc(void)
#endif
svc_init_xprt_sock(); /* svc sock transport */
init_socket_xprt(); /* clnt sock transport */
+ ve_set_rpc_kill_fn(rpc_set_kill_net);
return 0;
out5:
@@ -131,6 +133,7 @@ init_sunrpc(void)
static void __exit
cleanup_sunrpc(void)
{
+ ve_set_rpc_kill_fn(NULL);
rpc_sysfs_exit();
rpc_cleanup_clids();
xprt_cleanup_ids();
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel
