[Devel] [PATCH VZ10 v3 1/4] connector: annotate ve->cn with __rcu
Vasileios Almpanis
vasileios.almpanis at virtuozzo.com
Tue Aug 25 19:19:51 MSK 2026
The following patches will clear ve->cn on container stop and free the
per-VE connector state only after an RCU grace period, so that the
proc event delivery path can use it under rcu_read_lock() even when
the reported task no longer pins the VE.
Prepare for that: annotate ve->cn with __rcu and switch all accesses
to the RCU accessors so sparse can verify the protocol. All current
users run either in the context of a task alive in the VE or from
the VE start/stop hooks under ve->op_sem (or on ve0), so plain readers
use rcu_dereference_check(ve->cn, 1) with a comment and the start/stop
hooks use rcu_dereference_protected() with the proper lockdep
condition.
No functional change intended
https://virtuozzo.atlassian.net/browse/VSTOR-140421
Signed-off-by: Vasileios Almpanis <vasileios.almpanis at virtuozzo.com>
Feature: ve: ve generic structures
---
drivers/connector/cn_proc.c | 48 +++++++++++++++++++++++++++++++------------
drivers/connector/connector.c | 43 +++++++++++++++++++++++++-------------
include/linux/ve.h | 2 +-
3 files changed, 65 insertions(+), 28 deletions(-)
diff --git a/drivers/connector/cn_proc.c b/drivers/connector/cn_proc.c
index d41c8aca1866..d4ce1697dd0b 100644
--- a/drivers/connector/cn_proc.c
+++ b/drivers/connector/cn_proc.c
@@ -84,6 +84,8 @@ static int cn_filter(struct sock *dsk, struct sk_buff *skb, void *data)
static inline void send_msg_ve(struct ve_struct *ve, struct cn_msg *msg)
{
+ /* See the comment in proc_event_num_listeners() */
+ struct cn_private *cn = rcu_dereference_check(ve->cn, 1);
struct local_event *le_ptr;
__u32 filter_data[2];
@@ -93,9 +95,9 @@ static inline void send_msg_ve(struct ve_struct *ve, struct cn_msg *msg)
* so be of the safe side.
*/
BUILD_BUG_ON(offsetof(struct local_event, lock) != 0);
- local_lock(&ve->cn->local_event->lock);
+ local_lock(&cn->local_event->lock);
- le_ptr = this_cpu_ptr(ve->cn->local_event);
+ le_ptr = this_cpu_ptr(cn->local_event);
msg->seq = le_ptr->count++;
((struct proc_event *)msg->data)->cpu = smp_processor_id();
@@ -116,7 +118,7 @@ static inline void send_msg_ve(struct ve_struct *ve, struct cn_msg *msg)
cn_netlink_send_mult_ve(ve, msg, msg->len, 0, CN_IDX_PROC, GFP_NOWAIT,
cn_filter, (void *)filter_data);
- local_unlock(&ve->cn->local_event->lock);
+ local_unlock(&cn->local_event->lock);
}
static struct cn_msg *cn_msg_fill(__u8 *buffer, struct ve_struct *ve,
@@ -147,8 +149,15 @@ static struct cn_msg *cn_msg_fill(__u8 *buffer, struct ve_struct *ve,
static int proc_event_num_listeners(struct ve_struct *ve)
{
- if (ve->cn)
- return atomic_read(&ve->cn->proc_event_num_listeners);
+ /*
+ * Callers not under rcu_read_lock() are pinned by a live task
+ * of this VE (or run on ve0 whose connector state lives as long
+ * as the connector itself).
+ */
+ struct cn_private *cn = rcu_dereference_check(ve->cn, 1);
+
+ if (cn)
+ return atomic_read(&cn->proc_event_num_listeners);
return 0;
}
@@ -414,6 +423,8 @@ static void cn_proc_mcast_ctl(struct cn_msg *msg,
enum proc_cn_event ev_type = 0;
int err = 0, initial = 0;
struct sock *sk = NULL;
+ /* current is a live task of this VE, it cannot be stopped under us */
+ struct cn_private *cn = rcu_dereference_check(ve->cn, 1);
/*
* Events are reported with respect to the initial pid
@@ -467,11 +478,11 @@ static void cn_proc_mcast_ctl(struct cn_msg *msg,
switch (mc_op) {
case PROC_CN_MCAST_LISTEN:
if (initial || (prev_mc_op != PROC_CN_MCAST_LISTEN))
- atomic_inc(&ve->cn->proc_event_num_listeners);
+ atomic_inc(&cn->proc_event_num_listeners);
break;
case PROC_CN_MCAST_IGNORE:
if (!initial && (prev_mc_op != PROC_CN_MCAST_IGNORE))
- atomic_dec(&ve->cn->proc_event_num_listeners);
+ atomic_dec(&cn->proc_event_num_listeners);
((struct proc_input *)(sk->sk_user_data))->event_type =
PROC_EVENT_NONE;
break;
@@ -488,13 +499,18 @@ int cn_proc_init_ve(struct ve_struct *ve)
{
int err, cpu;
struct local_event *le_ptr;
+ struct cn_private *cn;
- ve->cn->local_event = alloc_percpu(struct local_event);
- if (!ve->cn->local_event)
+ cn = rcu_dereference_protected(ve->cn,
+ lockdep_is_held(&ve->op_sem) ||
+ ve_is_super(ve));
+
+ cn->local_event = alloc_percpu(struct local_event);
+ if (!cn->local_event)
return -ENOMEM;
for_each_possible_cpu(cpu) {
- le_ptr = per_cpu_ptr(ve->cn->local_event, cpu);
+ le_ptr = per_cpu_ptr(cn->local_event, cpu);
local_lock_init(&le_ptr->lock);
}
@@ -503,15 +519,21 @@ int cn_proc_init_ve(struct ve_struct *ve)
&cn_proc_mcast_ctl);
if (err) {
pr_warn("VE#%d: cn_proc failed to register\n", ve->veid);
- free_percpu(ve->cn->local_event);
+ free_percpu(cn->local_event);
return err;
}
- atomic_set(&ve->cn->proc_event_num_listeners, 0);
+ atomic_set(&cn->proc_event_num_listeners, 0);
return 0;
}
void cn_proc_fini_ve(struct ve_struct *ve)
{
+ struct cn_private *cn;
+
+ cn = rcu_dereference_protected(ve->cn,
+ lockdep_is_held(&ve->op_sem) ||
+ ve_is_super(ve));
+
cn_del_callback_ve(ve, &cn_proc_event_id);
- free_percpu(ve->cn->local_event);
+ free_percpu(cn->local_event);
}
diff --git a/drivers/connector/connector.c b/drivers/connector/connector.c
index 39a697803d0f..bf4a83f3f870 100644
--- a/drivers/connector/connector.c
+++ b/drivers/connector/connector.c
@@ -29,7 +29,13 @@ MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_CONNECTOR);
static struct cn_dev *get_cdev(struct ve_struct *ve)
{
- return &ve->cn->cdev;
+ /*
+ * Callers not under rcu_read_lock() are pinned by a live task
+ * of this VE or run from the VE start/stop hooks.
+ */
+ struct cn_private *cn = rcu_dereference_check(ve->cn, 1);
+
+ return cn ? &cn->cdev : NULL;
}
/*
@@ -230,12 +236,13 @@ int cn_add_callback_ve(struct ve_struct *ve,
void (*callback)(struct cn_msg *,
struct netlink_skb_parms *))
{
- struct cn_dev *dev = get_cdev(ve);
+ /* See the comment in get_cdev() */
+ struct cn_private *cn = rcu_dereference_check(ve->cn, 1);
- if (!ve->cn->cn_already_initialized)
+ if (!cn || !cn->cn_already_initialized)
return -EAGAIN;
- return cn_queue_add_callback(dev->cbdev, name, id, callback);
+ return cn_queue_add_callback(cn->cdev.cbdev, name, id, callback);
}
/*
@@ -297,6 +304,7 @@ static int __maybe_unused cn_proc_show(struct seq_file *m, void *v)
static int cn_init_ve(void *data)
{
struct ve_struct *ve = data;
+ struct cn_private *cn;
struct cn_dev *dev;
struct netlink_kernel_cfg cfg = {
.groups = CN_NETLINK_USERS + 0xf,
@@ -308,11 +316,12 @@ static int cn_init_ve(void *data)
struct net *net;
int err;
- ve->cn = kzalloc(sizeof(*ve->cn), GFP_KERNEL);
- if (!ve->cn)
+ cn = kzalloc(sizeof(*cn), GFP_KERNEL);
+ if (!cn)
return -ENOMEM;
+ rcu_assign_pointer(ve->cn, cn);
- dev = &ve->cn->cdev;
+ dev = &cn->cdev;
/*
* This is a hook, hooks are called under a single lock, so ve_nsproxy will
@@ -331,7 +340,7 @@ static int cn_init_ve(void *data)
goto netlink_release;
}
- ve->cn->cn_already_initialized = 1;
+ cn->cn_already_initialized = 1;
if (!proc_ve_create_single("connector", S_IRUGO, net->proc_net,
cn_proc_show)) {
@@ -353,18 +362,24 @@ static int cn_init_ve(void *data)
netlink_release:
netlink_kernel_release(dev->nls);
free_cn:
- kfree(ve->cn);
- ve->cn = NULL;
+ RCU_INIT_POINTER(ve->cn, NULL);
+ kfree(cn);
goto net_unlock;
}
static void cn_fini_ve(void *data)
{
struct ve_struct *ve = data;
- struct cn_dev *dev = get_cdev(ve);
+ struct cn_private *cn;
+ struct cn_dev *dev;
struct net *net;
- ve->cn->cn_already_initialized = 0;
+ cn = rcu_dereference_protected(ve->cn,
+ lockdep_is_held(&ve->op_sem) ||
+ ve_is_super(ve));
+ dev = &cn->cdev;
+
+ cn->cn_already_initialized = 0;
cn_proc_fini_ve(ve);
@@ -379,8 +394,8 @@ static void cn_fini_ve(void *data)
cn_queue_free_dev(dev->cbdev);
netlink_kernel_release(dev->nls);
- kfree(ve->cn);
- ve->cn = NULL;
+ RCU_INIT_POINTER(ve->cn, NULL);
+ kfree(cn);
}
#ifdef CONFIG_VE
diff --git a/include/linux/ve.h b/include/linux/ve.h
index 46e67d00cf78..2dc93e42784b 100644
--- a/include/linux/ve.h
+++ b/include/linux/ve.h
@@ -102,7 +102,7 @@ struct ve_struct {
char core_pattern[CORENAME_MAX_SIZE];
#endif
#ifdef CONFIG_CONNECTOR
- struct cn_private *cn;
+ struct cn_private __rcu *cn;
#endif
struct kthread_worker *kthreadd_worker;
--
2.43.0
More information about the Devel
mailing list