From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
To: Eric Dumazet <dada1@cosmosbay.com>
Cc: Corey Minyard <minyard@acm.org>,
David Miller <davem@davemloft.net>,
shemminger@vyatta.com, benny+usenet@amorsen.dk,
netdev@vger.kernel.org,
Christoph Lameter <cl@linux-foundation.org>,
a.p.zijlstra@chello.nl, johnpol@2ka.mipt.ru,
Christian Bell <christian@myri.com>
Subject: Re: [PATCH 2/2] udp: RCU handling for Unicast packets.
Date: Wed, 29 Oct 2008 13:17:59 -0700 [thread overview]
Message-ID: <20081029201759.GF6732@linux.vnet.ibm.com> (raw)
In-Reply-To: <4908C0CD.5050406@cosmosbay.com>
On Wed, Oct 29, 2008 at 09:00:13PM +0100, Eric Dumazet wrote:
> Paul E. McKenney a écrit :
>> On Wed, Oct 29, 2008 at 01:28:15PM -0500, Corey Minyard wrote:
>>> Eric Dumazet wrote:
>>>> Corey Minyard a écrit :
>>>>> Paul E. McKenney wrote:
>>>>>> On Wed, Oct 29, 2008 at 05:09:53PM +0100, Eric Dumazet wrote:
>>>>>>
>>>>>>> Corey Minyard a écrit :
>>>>>>>
>>>>>>>> Eric Dumazet wrote:
>>>>>>>>
>>>>>>>>> Corey Minyard found a race added in commit
>>>>>>>>> 271b72c7fa82c2c7a795bc16896149933110672d
>>>>>>>>> (udp: RCU handling for Unicast packets.)
>>>>>>>>>
>>>>>>>>> "If the socket is moved from one list to another list in-between
>>>>>>>>> the time the hash is calculated and the next field is accessed,
>>>>>>>>> and the socket has moved to the end of the new list, the traversal
>>>>>>>>> will not complete properly on the list it should have, since the
>>>>>>>>> socket will be on the end of the new list and there's not a way to
>>>>>>>>> tell it's on a new list and restart the list traversal. I think
>>>>>>>>> that this can be solved by pre-fetching the "next" field (with
>>>>>>>>> proper barriers) before checking the hash."
>>>>>>>>>
>>>>>>>>> This patch corrects this problem, introducing a new
>>>>>>>>> sk_for_each_rcu_safenext()
>>>>>>>>> macro.
>>>>>>>>>
>>>>>>>> You also need the appropriate smp_wmb() in udp_lib_get_port() after
>>>>>>>> sk_hash is set, I think, so the next field is guaranteed to be
>>>>>>>> changed after the hash value is changed.
>>>>>>>>
>>>>>>> Not sure about this one Corey.
>>>>>>>
>>>>>>> If a reader catches previous value of item->sk_hash, two cases are to
>>>>>>> be taken into :
>>>>>>>
>>>>>>> 1) its udp_hashfn(net, sk->sk_hash) is != hash -> goto begin :
>>>>>>> Reader will redo its scan
>>>>>>>
>>>>>>> 2) its udp_hashfn(net, sk->sk_hash) is == hash
>>>>>>> -> next pointer is good enough : it points to next item in same hash
>>>>>>> chain.
>>>>>>> No need to rescan the chain at this point.
>>>>>>> Yes we could miss the fact that a new port was bound and this UDP
>>>>>>> message could be lost.
>>>>>>>
>>>>>> 3) its udp_hashfn(net, sk-sk_hash) is == hash, but only because it was
>>>>>> removed, freed, reallocated, and then readded with the same hash
>>>>>> value,
>>>>>> possibly carrying the reader to a new position in the same list.
>>>>>>
>>>>> If I understand this, without the smp_wmb(), it is possible that the
>>>>> next field can be written to main memory before the hash value is
>>>>> written. If that happens, the following can occur:
>>>>>
>>>>> CPU1 CPU2
>>>>> next is set to NULL (end of new list)
>>>> Well, if this item is injected to the same chain, next wont be set to
>>>> NULL.
>>>>
>>>> That would mean previous writers deleted all items from the chain.
>>> I put my comment in the wrong place, I wasn't talking about being
>>> injected into the same chain.
>>>
>>>> In this case, readers can see NULL, it is not a problem at all.
>>>> List is/was empty.
>>>> An application cannot complain a packet is not
>>>> handled if its bind() syscall is not yet completed :)
>>>>
>>>> If item is injected on another chain, we will detect hash mismatch and
>>>> redo full scan.
>>> If the item is injected onto the end of another chain, the next field
>>> will be NULL and you won't detect a hash mismatch. It's basically the
>>> same issue as the previous race, though a lot more subtle and unlikely.
>>> If you get (from the previous socket) an old value of "sk_hash" and (from
>>> the new socket) a new value of "next" that is NULL, you will terminate
>>> the loop when you should have restarted it. I'm pretty sure that can
>>> occur without the write barrier.
>> One way of dealing with this is to keep a tail pointer. Then, if the
>> element containing the NULL pointer doesn't match the tail pointer seen
>> at the start of the search, or if the tail pointer has changed,
>> restart the search. Memory barriers will be needed. ;-)
>
> Hum... Another way of handling all those cases and avoid memory barriers
> would be to have different "NULL" pointers.
>
> Each hash chain should have a unique "NULL" pointer (in the case of UDP, it
> can be the 128 values : [ (void*)0 .. (void *)127 ]
>
> Then, when performing a lookup, a reader should check the "NULL" pointer
> it get at the end of its lookup has is the "hash" value of its chain.
>
> If not -> restart the loop, aka "goto begin;" :)
>
> We could avoid memory barriers then.
>
> In the two cases Corey mentioned, this trick could let us avoid memory
> barriers.
> (existing one in sk_add_node_rcu(sk, &hslot->head); should be enough)
>
> What do you think ?
Kinky!!! ;-)
Then the rcu_dereference() would be supplying the needed memory barriers.
Hmmm... I guess that the only confusion would be if the element got
removed and then added to the same list. But then if its pointer was
pseudo-NULL, then that would mean that all subsequent elements had been
removed, and all preceding ones added after the scan started.
Which might well be harmless, but I must defer to you on this one at
the moment.
If you need a larger hash table, another approach would be to set the
pointer's low-order bit, allowing the upper bits to be a full-sized
index -- or even a pointer to the list header. Just make very sure
to clear the pointer when freeing, or an element on the freelist
could end up looking like a legitimate end of list... Which again
might well be safe, but why inflict this on oneself?
Thanx, Paul
next prev parent reply other threads:[~2008-10-29 20:18 UTC|newest]
Thread overview: 134+ messages / expand[flat|nested] mbox.gz Atom feed top
2008-10-06 18:50 [PATCH 3/3] Convert the UDP hash lock to RCU Corey Minyard
2008-10-06 21:22 ` Eric Dumazet
2008-10-06 21:40 ` David Miller
2008-10-06 23:08 ` Corey Minyard
2008-10-07 8:37 ` Evgeniy Polyakov
2008-10-07 14:16 ` Christoph Lameter
2008-10-07 14:29 ` Evgeniy Polyakov
2008-10-07 14:38 ` Christoph Lameter
2008-10-07 14:33 ` Paul E. McKenney
2008-10-07 14:45 ` Christoph Lameter
2008-10-07 15:07 ` Eric Dumazet
2008-10-07 15:07 ` Paul E. McKenney
2008-10-07 5:24 ` Eric Dumazet
2008-10-07 8:54 ` Benny Amorsen
2008-10-07 12:59 ` Eric Dumazet
2008-10-07 14:07 ` Stephen Hemminger
2008-10-07 20:55 ` David Miller
2008-10-07 21:20 ` Stephen Hemminger
2008-10-08 13:55 ` Eric Dumazet
2008-10-08 18:45 ` David Miller
2008-10-28 20:37 ` [PATCH 1/2] udp: introduce struct udp_table and multiple rwlocks Eric Dumazet
2008-10-28 21:23 ` Christian Bell
2008-10-28 21:31 ` Evgeniy Polyakov
2008-10-28 21:48 ` Eric Dumazet
2008-10-28 21:28 ` Evgeniy Polyakov
2008-10-28 20:42 ` [PATCH 2/2] udp: RCU handling for Unicast packets Eric Dumazet
2008-10-28 22:45 ` Eric Dumazet
2008-10-29 5:05 ` David Miller
2008-10-29 8:23 ` Eric Dumazet
2008-10-29 8:56 ` David Miller
2008-10-29 10:19 ` Eric Dumazet
2008-10-29 18:19 ` David Miller
2008-10-29 9:04 ` Eric Dumazet
2008-10-29 9:17 ` David Miller
2008-10-29 13:17 ` Corey Minyard
2008-10-29 14:36 ` Eric Dumazet
2008-10-29 15:34 ` Corey Minyard
2008-10-29 16:09 ` Eric Dumazet
2008-10-29 16:37 ` Paul E. McKenney
2008-10-29 17:22 ` Corey Minyard
2008-10-29 17:45 ` Eric Dumazet
2008-10-29 18:28 ` Corey Minyard
2008-10-29 18:52 ` Paul E. McKenney
2008-10-29 20:00 ` Eric Dumazet
2008-10-29 20:17 ` Paul E. McKenney [this message]
2008-10-29 21:29 ` Corey Minyard
2008-10-29 21:57 ` Eric Dumazet
2008-10-29 21:58 ` Paul E. McKenney
2008-10-29 22:08 ` Eric Dumazet
2008-10-30 3:22 ` Corey Minyard
2008-10-30 5:50 ` Eric Dumazet
2008-11-02 4:19 ` David Miller
2008-10-30 5:40 ` David Miller
2008-10-30 5:51 ` Eric Dumazet
2008-10-30 7:04 ` Eric Dumazet
2008-10-30 7:05 ` David Miller
2008-10-30 15:40 ` [PATCH] udp: Introduce special NULL pointers for hlist termination Eric Dumazet
2008-10-30 15:51 ` Stephen Hemminger
2008-10-30 16:28 ` Corey Minyard
2008-10-31 14:37 ` Eric Dumazet
2008-10-31 14:55 ` Pavel Emelyanov
2008-11-02 4:22 ` David Miller
2008-10-30 17:12 ` Eric Dumazet
2008-10-31 7:51 ` David Miller
2008-10-30 16:01 ` Peter Zijlstra
2008-10-31 0:14 ` Keith Owens
2008-11-13 13:13 ` [PATCH 0/3] net: RCU lookups for UDP, DCCP and TCP protocol Eric Dumazet
2008-11-13 17:20 ` Andi Kleen
2008-11-17 3:41 ` David Miller
2008-11-19 19:52 ` Christoph Lameter
2008-11-13 13:14 ` [PATCH 1/3] rcu: Introduce hlist_nulls variant of hlist Eric Dumazet
2008-11-13 13:29 ` Peter Zijlstra
2008-11-13 13:44 ` Eric Dumazet
2008-11-13 16:02 ` [PATCH 4/3] rcu: documents rculist_nulls Eric Dumazet
2008-11-14 15:16 ` Peter Zijlstra
2008-11-17 3:36 ` David Miller
2008-11-19 17:07 ` Paul E. McKenney
2008-11-14 15:16 ` [PATCH 1/3] rcu: Introduce hlist_nulls variant of hlist Peter Zijlstra
2008-11-19 17:01 ` Paul E. McKenney
2008-11-19 17:53 ` Eric Dumazet
2008-11-19 18:46 ` Paul E. McKenney
2008-11-19 18:53 ` Arnaldo Carvalho de Melo
2008-11-19 21:17 ` Paul E. McKenney
2008-11-19 20:39 ` Eric Dumazet
2008-11-19 21:21 ` Paul E. McKenney
2008-11-13 13:15 ` [PATCH 2/3] udp: Use hlist_nulls in UDP RCU code Eric Dumazet
2008-11-19 17:29 ` Paul E. McKenney
2008-11-19 17:53 ` Eric Dumazet
2008-11-13 13:15 ` [PATCH 3/3] net: Convert TCP & DCCP hash tables to use RCU / hlist_nulls Eric Dumazet
2008-11-13 13:34 ` Peter Zijlstra
2008-11-13 13:51 ` Eric Dumazet
2008-11-13 14:08 ` Christoph Lameter
2008-11-13 14:22 ` Peter Zijlstra
2008-11-13 14:27 ` Christoph Lameter
2008-11-19 17:53 ` Paul E. McKenney
2008-11-23 9:33 ` [PATCH] net: Convert TCP/DCCP listening hash tables to use RCU Eric Dumazet
2008-11-23 15:59 ` Paul E. McKenney
2008-11-23 18:42 ` Eric Dumazet
2008-11-23 19:17 ` Paul E. McKenney
2008-11-23 20:18 ` Eric Dumazet
2008-11-23 22:33 ` Paul E. McKenney
2008-11-24 1:23 ` David Miller
2008-10-30 11:04 ` [PATCH 2/2] udp: RCU handling for Unicast packets Peter Zijlstra
2008-10-30 11:30 ` Eric Dumazet
2008-10-30 18:25 ` Paul E. McKenney
2008-10-31 16:40 ` Eric Dumazet
2008-11-01 3:10 ` Paul E. McKenney
2008-10-29 17:32 ` Eric Dumazet
2008-10-29 18:11 ` Paul E. McKenney
2008-10-29 18:29 ` David Miller
2008-10-29 18:38 ` Paul E. McKenney
2008-10-29 18:36 ` Eric Dumazet
2008-10-29 18:20 ` David Miller
2008-10-30 11:12 ` Peter Zijlstra
2008-10-30 11:29 ` Eric Dumazet
2008-10-28 20:37 ` [PATCH 0/2] udp: Convert the UDP hash lock to RCU Eric Dumazet
2008-10-28 21:28 ` Stephen Hemminger
2008-10-28 21:50 ` Eric Dumazet
2008-10-07 16:43 ` [PATCH 3/3] " Corey Minyard
2008-10-07 18:26 ` David Miller
2008-10-08 8:35 ` Eric Dumazet
2008-10-08 16:38 ` David Miller
2008-10-07 8:31 ` Peter Zijlstra
2008-10-07 14:36 ` Paul E. McKenney
2008-10-07 18:29 ` David Miller
2008-10-06 22:07 ` Corey Minyard
2008-10-07 8:17 ` Peter Zijlstra
2008-10-07 9:24 ` Eric Dumazet
2008-10-07 14:15 ` Christoph Lameter
2008-10-07 14:38 ` Paul E. McKenney
2008-10-07 14:50 ` Eric Dumazet
2008-10-07 15:05 ` Paul E. McKenney
2008-10-07 15:09 ` Peter Zijlstra
2008-10-07 15:23 ` Christoph Lameter
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20081029201759.GF6732@linux.vnet.ibm.com \
--to=paulmck@linux.vnet.ibm.com \
--cc=a.p.zijlstra@chello.nl \
--cc=benny+usenet@amorsen.dk \
--cc=christian@myri.com \
--cc=cl@linux-foundation.org \
--cc=dada1@cosmosbay.com \
--cc=davem@davemloft.net \
--cc=johnpol@2ka.mipt.ru \
--cc=minyard@acm.org \
--cc=netdev@vger.kernel.org \
--cc=shemminger@vyatta.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).