When I've been reworking vznetstat mem management I've added destroying
of VE's statistics upon containers stop. This is incorrect, we need to
keep it while module is alive.

Still one can destroy statistics with VZCTL_TC_DESTROY_STAT
or VZCTL_TC_DESTROY_ALL_STAT (of course it clears it for
non running VEs only).

https://jira.sw.ru/browse/PSBM-35178

Signed-off-by: Cyrill Gorcunov <[email protected]>
CC: Andrey Vagin <[email protected]>
CC: Vladimir Davydov <[email protected]>
CC: Konstantin Khorenko <[email protected]>
CC: Pavel Emelyanov <[email protected]>
CC: Igor Sukhih <[email protected]>
---
 kernel/ve/vznetstat/vznetstat.c |   15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

Index: linux-pcs7.git/kernel/ve/vznetstat/vznetstat.c
===================================================================
--- linux-pcs7.git.orig/kernel/ve/vznetstat/vznetstat.c
+++ linux-pcs7.git/kernel/ve/vznetstat/vznetstat.c
@@ -1096,13 +1096,7 @@ static void __net_exit net_exit_acct(str
 {
        struct ve_struct *ve = net->owner_ve;
 
-       if (ve->stat) {
-               venet_acct_put_stat(ve->stat);
-               if (atomic_read(&ve->stat->users) == 0) {
-                       venet_acct_destroy_stat(ve->veid);
-                       ve->stat = NULL;
-               }
-       }
+       venet_acct_put_stat(ve->stat);
 }
 
 static struct pernet_operations __net_initdata net_acct_ops = {
@@ -1142,10 +1136,17 @@ int __init venetstat_init(void)
 
 void __exit venetstat_exit(void)
 {
+       struct ve_struct *ve;
+
        unregister_pernet_subsys(&net_acct_ops);
        vzioctl_unregister(&tc_ioctl_info);
        venet_acct_destroy_all_stat();
 
+       mutex_lock(&ve_list_lock);
+       for_each_ve(ve)
+               ve->stat = NULL;
+       mutex_unlock(&ve_list_lock);
+
 #if CONFIG_PROC_FS
        remove_proc_entry("venetstat_v6", proc_vz_dir);
        remove_proc_entry("venetstat", proc_vz_dir);
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to