Re: [take21 1/4] kevent: Core files.

[Eric Dumazet <dada1@cosmosbay.com>]

Evgeniy Polyakov a e'crit :
> On Sat, Oct 28, 2006 at 03:23:40PM +0200, Eric Dumazet (dada1@cosmosbay.com) wrote:
>>> diff --git a/kernel/kevent/kevent_user.c b/kernel/kevent/kevent_user.c
>>> index 711a8a8..ecee668 100644
>>> --- a/kernel/kevent/kevent_user.c
>>> +++ b/kernel/kevent/kevent_user.c
>>> @@ -235,6 +235,36 @@ static void kevent_free_rcu(struct rcu_h
>>> }
>>>
>>> /*
>>> + * Must be called under u->ready_lock.
>>> + * This function removes kevent from ready queue and 
>>> + * tries to add new kevent into ring buffer.
>>> + */
>>> +static void kevent_remove_ready(struct kevent *k)
>>> +{
>>> +	struct kevent_user *u = k->user;
>>> +
>>> +	list_del(&k->ready_entry);
>> Arg... no
>>
>> You cannot call list_del() , then check overflow_kevent.
>>
>> I you call list_del on what happens to be the kevent pointed by 
>> overflow_kevent, you loose...
> 
> This function is always called from appropriate context, where it is
> guaranteed that it is safe to call list_del:
> 1. when kevent is removed. It is called after check, that given kevent 
> is in the ready queue.
> 2. when dequeued from ready queue, which means that it can be removed
> from that queue.
> 

Could you please check the list_del() function ?

file include/linux/list.h

static inline void list_del(struct list_head *entry)
{
   __list_del(entry->prev, entry->next);
   entry->next = LIST_POISON1;
   entry->prev = LIST_POISON2;
}

So, after calling list_del(&k->read_entry);
next and prev are basically destroyed.

So when you write later :

+        if (!err || u->overflow_kevent == k) {
+            if (u->overflow_kevent->ready_entry.next == &u->ready_list)
+                u->overflow_kevent = NULL;
+            else
+                u->overflow_kevent = + 
list_entry(u->overflow_kevent->ready_entry.next, + 
struct kevent, ready_entry);
+        }


then you have a problem, since

list_entry(k->ready_entry.next, struct kevent, ready_entry);

will give you garbage.

Eric