netdev - Re: net-sched 04/05: sch_htb: move hash and sibling list removal to htb

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 02 Jul 2008 12:11:06 +0200
From:	Patrick McHardy <kaber@...sh.net>
To:	Jarek Poplawski <jarkao2@...il.com>
CC:	netdev@...r.kernel.org, devik@....cz
Subject: Re: net-sched 04/05: sch_htb: move hash and sibling list removal
 to htb_delete

Jarek Poplawski wrote:
> Patrick McHardy wrote, On 07/01/2008 04:34 PM:
> ...
>> diff --git a/net/sched/sch_htb.c b/net/sched/sch_htb.c
>> index 0284791..879f0b6 100644
>> --- a/net/sched/sch_htb.c
>> +++ b/net/sched/sch_htb.c
>> @@ -1238,14 +1238,6 @@ static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
>>  
>>  	tcf_destroy_chain(&cl->filter_list);
>>  
>> -	while (!list_empty(&cl->children))
>> -		htb_destroy_class(sch, list_entry(cl->children.next,
>> -						  struct htb_class, sibling));
>> -
>> -	/* note: this delete may happen twice (see htb_delete) */
>> -	hlist_del_init(&cl->hlist);
>> -	list_del(&cl->sibling);
>> -
>>  	if (cl->prio_activity)
>>  		htb_deactivate(q, cl);
> 
> I'll try to check this all more later, but this probably "ain't" good:
> during deactivation a class can use a parent class, so there would be
> a use after kfree if it's not "parents after children". IMHO, it's
> better to do a separate version of htb_destroy_class() for
> htb_destroy(), and skip there htb_deactivate(), tcf_destroy_chain()
> and htb_safe_rb_erase() which are not needed at the moment.

Good point.

Actually deactivation in htb_destroy_class is unnecessary, in
htb_delete() its done immediately anyway (as it should), in
htb_destroy() the entire qdisc is killed atomically and thus
there is no need for deactivation of single classes.

The tcf_destroy_chain() call is harmless since all filters are
already gone when going through qdisc_destroy(). Which leaves
htb_safe_rb_erase(), which looks like it should also be performed
in htb_delete() since otherwise the class will still be visible
in the rb tree in the period between dropping the lock in
htb_delete() and the final destruction in htb_put(). Similar
to deactivation, removal from the rb tree is unnecessary in
the qdisc_destroy() case.

Attached is a incremental patch and the full new patch that
makes these changes.

View attachment "x" of type "text/plain" (755 bytes)

View attachment "04.diff" of type "text/x-diff" (3043 bytes)