[PATCH net-next 1/8] rtnetlink: Revert "rtnetlink: add reference counting to prevent module unload while dump is in progress"

[Florian Westphal]

This reverts commit 019a316992ee0d9832b1c480c899d6bdf2a0a77e.

Peter Zijlstra reported:
  1) it not in fact a refcount, so using refcount_t is silly
  2) there is a distinct lack of memory barriers, so we can easily
     observe the decrement while the msg_handler is still in progress.
  3) waiting with a schedule()/yield() loop is complete crap and subject
     life-locks, imagine doing that rtnl_unregister_all() from a RT task.
----------

Moreover, the commit doesn't even work.
Problem is that netlink keeps the dump function address in the control
block to resume dumps, i.e. when netlink_dump_start() returns then
we may not be done with the dump, it can be restarted on next recv.

First, revert the broken commit to start from scratch.
Next patch will allow to store a pointer the owning module so we can
let netlink infrastructure grab a refcount to module while dump function
address is in use.

Signed-off-by: Florian Westphal <f...@strlen.de>
---
 net/core/rtnetlink.c | 13 -------------
 1 file changed, 13 deletions(-)

diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index dc5ad84ac096..c70f62137dd8 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -128,7 +128,6 @@ EXPORT_SYMBOL(lockdep_rtnl_is_held);
 #endif /* #ifdef CONFIG_PROVE_LOCKING */
 
 static struct rtnl_link __rcu *rtnl_msg_handlers[RTNL_FAMILY_MAX + 1];
-static refcount_t rtnl_msg_handlers_ref[RTNL_FAMILY_MAX + 1];
 
 static inline int rtm_msgindex(int msgtype)
 {
@@ -262,8 +261,6 @@ void rtnl_unregister_all(int protocol)
 
        synchronize_net();
 
-       while (refcount_read(&rtnl_msg_handlers_ref[protocol]) > 1)
-               schedule();
        kfree(handlers);
 }
 EXPORT_SYMBOL_GPL(rtnl_unregister_all);
@@ -4365,8 +4362,6 @@ static int rtnetlink_rcv_msg(struct sk_buff *skb, struct 
nlmsghdr *nlh,
                                goto err_unlock;
                }
 
-               refcount_inc(&rtnl_msg_handlers_ref[family]);
-
                if (type == RTM_GETLINK - RTM_BASE)
                        min_dump_alloc = rtnl_calcit(skb, nlh);
 
@@ -4380,7 +4375,6 @@ static int rtnetlink_rcv_msg(struct sk_buff *skb, struct 
nlmsghdr *nlh,
                        };
                        err = netlink_dump_start(rtnl, skb, nlh, &c);
                }
-               refcount_dec(&rtnl_msg_handlers_ref[family]);
                return err;
        }
 
@@ -4392,12 +4386,10 @@ static int rtnetlink_rcv_msg(struct sk_buff *skb, 
struct nlmsghdr *nlh,
 
        flags = READ_ONCE(handlers[type].flags);
        if (flags & RTNL_FLAG_DOIT_UNLOCKED) {
-               refcount_inc(&rtnl_msg_handlers_ref[family]);
                doit = READ_ONCE(handlers[type].doit);
                rcu_read_unlock();
                if (doit)
                        err = doit(skb, nlh, extack);
-               refcount_dec(&rtnl_msg_handlers_ref[family]);
                return err;
        }
 
@@ -4498,11 +4490,6 @@ static struct pernet_operations rtnetlink_net_ops = {
 
 void __init rtnetlink_init(void)
 {
-       int i;
-
-       for (i = 0; i < ARRAY_SIZE(rtnl_msg_handlers_ref); i++)
-               refcount_set(&rtnl_msg_handlers_ref[i], 1);
-
        if (register_pernet_subsys(&rtnetlink_net_ops))
                panic("rtnetlink_init: cannot initialize rtnetlink\n");
 
-- 
2.13.6