Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Fri, Oct 23, 2015 at 04:14:10AM +, Kosuke Tatsukawa wrote:
>> J. Bruce Fields wrote:
>> > On Fri, Oct 16, 2015 at 02:28:10AM +, Kosuke Tatsukawa wrote:
>> >> Tatsukawa Kosuke wrote:
>> >> > J. Bruce Fields wrote:
>> >> >> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
>> >> >>> Tatsukawa Kosuke wrote:
>> >> >>> > J. Bruce Fields wrote:
>> >> >>> >> Thanks for the detailed investigation.
>> >> >>> >> 
>> >> >>> >> I think it would be worth adding a comment if that might help 
>> >> >>> >> someone
>> >> >>> >> having to reinvestigate this again some day.
>> >> >>> > 
>> >> >>> > It would be nice, but I find it difficult to write a comment in the
>> >> >>> > sunrpc layer why a memory barrier isn't necessary, using the 
>> >> >>> > knowledge
>> >> >>> > of how nfsd uses it, and the current implementation of the network 
>> >> >>> > code.
>> >> >>> > 
>> >> >>> > Personally, I would prefer removing the call to waitqueue_active() 
>> >> >>> > which
>> >> >>> > would make the memory barrier totally unnecessary at the cost of a
>> >> >>> > spin_lock + spin_unlock by unconditionally calling
>> >> >>> > wake_up_interruptible.
>> >> >>> 
>> >> >>> On second thought, the callbacks will be called frequently from the 
>> >> >>> tcp
>> >> >>> code, so it wouldn't be a good idea.
>> >> >> 
>> >> >> So, I was even considering documenting it like this, if it's not
>> >> >> overkill.
>> >> >> 
>> >> >> Hmm... but if this is right, then we may as well ask why we're doing 
>> >> >> the
>> >> >> wakeups at all.  Might be educational to test the code with them
>> >> >> removed.
>> >> > 
>> >> > sk_write_space will be called in sock_wfree() with UDP/IP each time
>> >> > kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
>> >> > SOCK_NOSPACE has been set.
>> >> > 
>> >> > sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
>> >> > and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
>> >> > barrier with sk_data_ready called right after __skb_queue_tail().
>> >> > I think this hasn't caused any problems because sk_data_ready wasn't
>> >> > used.
>> >> 
>> >> Actually, svc_udp_data_ready() calls set_bit() which is an atomic
>> >> operation.  So there won't be a problem unless svsk is NULL.
>> > 
>> > So is it true that every caller of these socket callbacks has adequate
>> > memory barriers between the time the change is made visible and the time
>> > the callback is called?
>> > 
>> > If so, then there's nothing really specific about nfsd here.
>> > 
>> > In that case maybe it's the networking code that use some documentation,
>> > if it doesn't already?  (Or maybe common helper functions for this
>> > 
>> >if (waitqueue_active(wq))
>> >wake_up(wq)
>> > 
>> > pattern?)
>> 
>> Some of the other places defining these callback functions are using
>>   static inline bool wq_has_sleeper(struct socket_wq *wq)
>> defined in include/net/sock.h
>> 
>> The comment above the function explains that it was introduced for
>> exactly this purpose.
>> 
>> Even thought the argument variable uses the same name "wq", it has a
>> different type from the wq used in svcsock.c (struct socket_wq *
>> vs. wait_queue_head_t *).
> 
> OK, thanks.  So, I guess it still sounds like the code is OK as is, but
> maybe my comment wasn't.  Here's another attempt.

Thank you.  By now the patch looks completely different from my original
patch, so I don't think I deserve to be mentioned in the Author line.


> --b.
> 
> commit b805ca58a81a
> Author: Kosuke Tatsukawa 
> Date:   Fri Oct 9 01:44:07 2015 +
> 
> svcrpc: document lack of some memory barriers
> 
> We're missing memory barriers in net/sunrpc/svcsock.c in some spots we'd
> expect them.  But it doesn't appear they're necessary in our case, and
> this is likely a hot path--for now just document the odd behavior.
> 
> I found this issue when I was looking through the linux source code
> for places calling waitqueue_active() before wake_up*(), but without
> preceding memory barriers, after sending a patch to fix a similar
> issue in drivers/tty/n_tty.c  (Details about the original issue can be
> found here: https://lkml.org/lkml/2015/9/28/849).

I should have used the stable link format to refer to the disucssion in
LKML instead of the lkml.org URL.  The stable link is

http://lkml.kernel.org/r/17ec94b0a072c34b8dcf0d30ad16044a02871...@bpxm09gp.gisp.nec.co.jp


> 
> Signed-off-by: Kosuke Tatsukawa 
> [bfie...@redhat.com,nfbr...@novell.com: document instead of adding 
> barriers]
> Signed-off-by: J. Bruce Fields 
> 
> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> index 48923730722d..1f71eece04d3 100644
> --- a/net/sunrpc/svcsock.c
> +++ b/net/sunrpc/svcsock.c
> @@ -399,6 +399,31 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
>   return 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Fri, Oct 23, 2015 at 04:14:10AM +, Kosuke Tatsukawa wrote:
> J. Bruce Fields wrote:
> > On Fri, Oct 16, 2015 at 02:28:10AM +, Kosuke Tatsukawa wrote:
> >> Tatsukawa Kosuke wrote:
> >> > J. Bruce Fields wrote:
> >> >> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
> >> >>> Tatsukawa Kosuke wrote:
> >> >>> > J. Bruce Fields wrote:
> >> >>> >> Thanks for the detailed investigation.
> >> >>> >> 
> >> >>> >> I think it would be worth adding a comment if that might help 
> >> >>> >> someone
> >> >>> >> having to reinvestigate this again some day.
> >> >>> > 
> >> >>> > It would be nice, but I find it difficult to write a comment in the
> >> >>> > sunrpc layer why a memory barrier isn't necessary, using the 
> >> >>> > knowledge
> >> >>> > of how nfsd uses it, and the current implementation of the network 
> >> >>> > code.
> >> >>> > 
> >> >>> > Personally, I would prefer removing the call to waitqueue_active() 
> >> >>> > which
> >> >>> > would make the memory barrier totally unnecessary at the cost of a
> >> >>> > spin_lock + spin_unlock by unconditionally calling
> >> >>> > wake_up_interruptible.
> >> >>> 
> >> >>> On second thought, the callbacks will be called frequently from the tcp
> >> >>> code, so it wouldn't be a good idea.
> >> >> 
> >> >> So, I was even considering documenting it like this, if it's not
> >> >> overkill.
> >> >> 
> >> >> Hmm... but if this is right, then we may as well ask why we're doing the
> >> >> wakeups at all.  Might be educational to test the code with them
> >> >> removed.
> >> > 
> >> > sk_write_space will be called in sock_wfree() with UDP/IP each time
> >> > kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
> >> > SOCK_NOSPACE has been set.
> >> > 
> >> > sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
> >> > and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
> >> > barrier with sk_data_ready called right after __skb_queue_tail().
> >> > I think this hasn't caused any problems because sk_data_ready wasn't
> >> > used.
> >> 
> >> Actually, svc_udp_data_ready() calls set_bit() which is an atomic
> >> operation.  So there won't be a problem unless svsk is NULL.
> > 
> > So is it true that every caller of these socket callbacks has adequate
> > memory barriers between the time the change is made visible and the time
> > the callback is called?
> > 
> > If so, then there's nothing really specific about nfsd here.
> > 
> > In that case maybe it's the networking code that use some documentation,
> > if it doesn't already?  (Or maybe common helper functions for this
> > 
> > if (waitqueue_active(wq))
> > wake_up(wq)
> > 
> > pattern?)
> 
> Some of the other places defining these callback functions are using
>   static inline bool wq_has_sleeper(struct socket_wq *wq)
> defined in include/net/sock.h
> 
> The comment above the function explains that it was introduced for
> exactly this purpose.
> 
> Even thought the argument variable uses the same name "wq", it has a
> different type from the wq used in svcsock.c (struct socket_wq *
> vs. wait_queue_head_t *).

OK, thanks.  So, I guess it still sounds like the code is OK as is, but
maybe my comment wasn't.  Here's another attempt.

--b.

commit b805ca58a81a
Author: Kosuke Tatsukawa 
Date:   Fri Oct 9 01:44:07 2015 +

svcrpc: document lack of some memory barriers

We're missing memory barriers in net/sunrpc/svcsock.c in some spots we'd
expect them.  But it doesn't appear they're necessary in our case, and
this is likely a hot path--for now just document the odd behavior.

I found this issue when I was looking through the linux source code
for places calling waitqueue_active() before wake_up*(), but without
preceding memory barriers, after sending a patch to fix a similar
issue in drivers/tty/n_tty.c  (Details about the original issue can be
found here: https://lkml.org/lkml/2015/9/28/849).

Signed-off-by: Kosuke Tatsukawa 
[bfie...@redhat.com,nfbr...@novell.com: document instead of adding barriers]
Signed-off-by: J. Bruce Fields 

diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 48923730722d..1f71eece04d3 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -399,6 +399,31 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
return svc_port_is_privileged(svc_addr(rqstp));
 }
 
+static bool sunrpc_waitqueue_active(wait_queue_head_t *wq)
+{
+   if (!wq)
+   return false;
+   /*
+* Kosuke Tatsukawa points out there should normally be a memory
+* barrier here--see wq_has_sleeper().
+*
+* It appears that isn't currently necessary, though, basically
+* because callers all appear to have sufficient memory barriers
+* between the time the relevant change is made and the
+* time they call these callbacks.
+ 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Fri, Oct 16, 2015 at 02:28:10AM +, Kosuke Tatsukawa wrote:
>> Tatsukawa Kosuke wrote:
>> > J. Bruce Fields wrote:
>> >> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
>> >>> Tatsukawa Kosuke wrote:
>> >>> > J. Bruce Fields wrote:
>> >>> >> Thanks for the detailed investigation.
>> >>> >> 
>> >>> >> I think it would be worth adding a comment if that might help someone
>> >>> >> having to reinvestigate this again some day.
>> >>> > 
>> >>> > It would be nice, but I find it difficult to write a comment in the
>> >>> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
>> >>> > of how nfsd uses it, and the current implementation of the network 
>> >>> > code.
>> >>> > 
>> >>> > Personally, I would prefer removing the call to waitqueue_active() 
>> >>> > which
>> >>> > would make the memory barrier totally unnecessary at the cost of a
>> >>> > spin_lock + spin_unlock by unconditionally calling
>> >>> > wake_up_interruptible.
>> >>> 
>> >>> On second thought, the callbacks will be called frequently from the tcp
>> >>> code, so it wouldn't be a good idea.
>> >> 
>> >> So, I was even considering documenting it like this, if it's not
>> >> overkill.
>> >> 
>> >> Hmm... but if this is right, then we may as well ask why we're doing the
>> >> wakeups at all.  Might be educational to test the code with them
>> >> removed.
>> > 
>> > sk_write_space will be called in sock_wfree() with UDP/IP each time
>> > kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
>> > SOCK_NOSPACE has been set.
>> > 
>> > sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
>> > and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
>> > barrier with sk_data_ready called right after __skb_queue_tail().
>> > I think this hasn't caused any problems because sk_data_ready wasn't
>> > used.
>> 
>> Actually, svc_udp_data_ready() calls set_bit() which is an atomic
>> operation.  So there won't be a problem unless svsk is NULL.
> 
> So is it true that every caller of these socket callbacks has adequate
> memory barriers between the time the change is made visible and the time
> the callback is called?
> 
> If so, then there's nothing really specific about nfsd here.
> 
> In that case maybe it's the networking code that use some documentation,
> if it doesn't already?  (Or maybe common helper functions for this
> 
>   if (waitqueue_active(wq))
>   wake_up(wq)
> 
> pattern?)

Some of the other places defining these callback functions are using
  static inline bool wq_has_sleeper(struct socket_wq *wq)
defined in include/net/sock.h

The comment above the function explains that it was introduced for
exactly this purpose.

Even thought the argument variable uses the same name "wq", it has a
different type from the wq used in svcsock.c (struct socket_wq *
vs. wait_queue_head_t *).


> --b.
> 
>> 
>> 
>> >> --b.
>> >> 
>> >> commit 0882cfeb39e0
>> >> Author: J. Bruce Fields 
>> >> Date:   Thu Oct 15 16:53:41 2015 -0400
>> >> 
>> >> svcrpc: document lack of some memory barriers.
>> >> 
>> >> Kosuke Tatsukawa points out an odd lack of memory barriers in some 
>> >> sites
>> >> here.  I think the code's correct, but it's probably worth 
>> >> documenting.
>> >> 
>> >> Reported-by: Kosuke Tatsukawa 
>> >> 
>> >> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
>> >> index 856407fa085e..90480993ec4a 100644
>> >> --- a/net/sunrpc/svcsock.c
>> >> +++ b/net/sunrpc/svcsock.c
>> >> @@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst 
>> >> *rqstp)
>> >>   return svc_port_is_privileged(svc_addr(rqstp));
>> >>  }
>> >>  
>> >> +static void svc_no_smp_mb(void)
>> >> +{
>> >> + /*
>> >> +  * Kosuke Tatsukawa points out there should normally be an
>> >> +  * smp_mb() at the callsites of this function.  (Either that or
>> >> +  * we could just drop the waitqueue_active() checks.)
>> >> +  *
>> >> +  * It appears they aren't currently necessary, though, basically
>> >> +  * because nfsd does non-blocking reads from these sockets, so
>> >> +  * the only places we wait on this waitqueue is in sendpage and
>> >> +  * sendmsg, which won't be waiting for wakeups on newly arrived
>> >> +  * data.
>> >> +  *
>> >> +  * Maybe we should add the memory barriers anyway, but these are
>> >> +  * hot paths so we'd need to be convinced there's no sigificant
>> >> +  * penalty.
>> >> +  */
>> >> +}
>> >> +
>> >>  /*
>> >>   * INET callback when data has been received on the socket.
>> >>   */
>> >> @@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
>> >>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>> >>   svc_xprt_enqueue(>sk_xprt);
>> >>   }
>> >> - smp_mb();
>> >> + svc_no_smp_mb();
>> >>   if (wq && waitqueue_active(wq))
>> >>   wake_up_interruptible(wq);
>> >>  }
>> >> @@ -433,7 +452,7 @@ static void svc_write_space(struct sock 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Fri, Oct 16, 2015 at 02:28:10AM +, Kosuke Tatsukawa wrote:
> Tatsukawa Kosuke wrote:
> > J. Bruce Fields wrote:
> >> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
> >>> Tatsukawa Kosuke wrote:
> >>> > J. Bruce Fields wrote:
> >>> >> Thanks for the detailed investigation.
> >>> >> 
> >>> >> I think it would be worth adding a comment if that might help someone
> >>> >> having to reinvestigate this again some day.
> >>> > 
> >>> > It would be nice, but I find it difficult to write a comment in the
> >>> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
> >>> > of how nfsd uses it, and the current implementation of the network code.
> >>> > 
> >>> > Personally, I would prefer removing the call to waitqueue_active() which
> >>> > would make the memory barrier totally unnecessary at the cost of a
> >>> > spin_lock + spin_unlock by unconditionally calling
> >>> > wake_up_interruptible.
> >>> 
> >>> On second thought, the callbacks will be called frequently from the tcp
> >>> code, so it wouldn't be a good idea.
> >> 
> >> So, I was even considering documenting it like this, if it's not
> >> overkill.
> >> 
> >> Hmm... but if this is right, then we may as well ask why we're doing the
> >> wakeups at all.  Might be educational to test the code with them
> >> removed.
> > 
> > sk_write_space will be called in sock_wfree() with UDP/IP each time
> > kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
> > SOCK_NOSPACE has been set.
> > 
> > sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
> > and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
> > barrier with sk_data_ready called right after __skb_queue_tail().
> > I think this hasn't caused any problems because sk_data_ready wasn't
> > used.
> 
> Actually, svc_udp_data_ready() calls set_bit() which is an atomic
> operation.  So there won't be a problem unless svsk is NULL.

So is it true that every caller of these socket callbacks has adequate
memory barriers between the time the change is made visible and the time
the callback is called?

If so, then there's nothing really specific about nfsd here.

In that case maybe it's the networking code that use some documentation,
if it doesn't already?  (Or maybe common helper functions for this

if (waitqueue_active(wq))
wake_up(wq)

pattern?)

--b.

> 
> 
> >> --b.
> >> 
> >> commit 0882cfeb39e0
> >> Author: J. Bruce Fields 
> >> Date:   Thu Oct 15 16:53:41 2015 -0400
> >> 
> >> svcrpc: document lack of some memory barriers.
> >> 
> >> Kosuke Tatsukawa points out an odd lack of memory barriers in some 
> >> sites
> >> here.  I think the code's correct, but it's probably worth documenting.
> >> 
> >> Reported-by: Kosuke Tatsukawa 
> >> 
> >> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> >> index 856407fa085e..90480993ec4a 100644
> >> --- a/net/sunrpc/svcsock.c
> >> +++ b/net/sunrpc/svcsock.c
> >> @@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst 
> >> *rqstp)
> >>return svc_port_is_privileged(svc_addr(rqstp));
> >>  }
> >>  
> >> +static void svc_no_smp_mb(void)
> >> +{
> >> +  /*
> >> +   * Kosuke Tatsukawa points out there should normally be an
> >> +   * smp_mb() at the callsites of this function.  (Either that or
> >> +   * we could just drop the waitqueue_active() checks.)
> >> +   *
> >> +   * It appears they aren't currently necessary, though, basically
> >> +   * because nfsd does non-blocking reads from these sockets, so
> >> +   * the only places we wait on this waitqueue is in sendpage and
> >> +   * sendmsg, which won't be waiting for wakeups on newly arrived
> >> +   * data.
> >> +   *
> >> +   * Maybe we should add the memory barriers anyway, but these are
> >> +   * hot paths so we'd need to be convinced there's no sigificant
> >> +   * penalty.
> >> +   */
> >> +}
> >> +
> >>  /*
> >>   * INET callback when data has been received on the socket.
> >>   */
> >> @@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
> >>set_bit(XPT_DATA, >sk_xprt.xpt_flags);
> >>svc_xprt_enqueue(>sk_xprt);
> >>}
> >> -  smp_mb();
> >> +  svc_no_smp_mb();
> >>if (wq && waitqueue_active(wq))
> >>wake_up_interruptible(wq);
> >>  }
> >> @@ -433,7 +452,7 @@ static void svc_write_space(struct sock *sk)
> >>svc_xprt_enqueue(>sk_xprt);
> >>}
> >>  
> >> -  smp_mb();
> >> +  svc_no_smp_mb();
> >>if (wq && waitqueue_active(wq)) {
> >>dprintk("RPC svc_write_space: someone sleeping on %p\n",
> >>   svsk);
> >> @@ -789,7 +808,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
> >>}
> >>  
> >>wq = sk_sleep(sk);
> >> -  smp_mb();
> >> +  svc_no_smp_mb();
> >>if (wq && waitqueue_active(wq))
> >>wake_up_interruptible_all(wq);
> >>  }
> >> @@ -811,7 +830,7 @@ static void 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

Tatsukawa Kosuke wrote:
> J. Bruce Fields wrote:
>> On Wed, Oct 14, 2015 at 03:57:13AM +, Kosuke Tatsukawa wrote:
>>> J. Bruce Fields wrote:
>>> > On Mon, Oct 12, 2015 at 10:41:06AM +, Kosuke Tatsukawa wrote:
>>> >> J. Bruce Fields wrote:
>>> >> > On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
>>> >> >> Neil Brown wrote:
>>> >> >> > Kosuke Tatsukawa  writes:
>>> >> >> > 
>>> >> >> >> There are several places in net/sunrpc/svcsock.c which calls
>>> >> >> >> waitqueue_active() without calling a memory barrier.  Add a memory
>>> >> >> >> barrier just as in wq_has_sleeper().
>>> >> >> >>
>>> >> >> >> I found this issue when I was looking through the linux source code
>>> >> >> >> for places calling waitqueue_active() before wake_up*(), but 
>>> >> >> >> without
>>> >> >> >> preceding memory barriers, after sending a patch to fix a similar
>>> >> >> >> issue in drivers/tty/n_tty.c  (Details about the original issue 
>>> >> >> >> can be
>>> >> >> >> found here: https://lkml.org/lkml/2015/9/28/849).
>>> >> >> > 
>>> >> >> > hi,
>>> >> >> > this feels like the wrong approach to the problem.  It requires 
>>> >> >> > extra
>>> >> >> > 'smb_mb's to be spread around which are hard to understand as easy 
>>> >> >> > to
>>> >> >> > forget.
>>> >> >> > 
>>> >> >> > A quick look seems to suggest that (nearly) every waitqueue_active()
>>> >> >> > will need an smb_mb.  Could we just put the smb_mb() inside
>>> >> >> > waitqueue_active()??
>>> >> >> 
>>> >> >> 
>>> >> >> There are around 200 occurrences of waitqueue_active() in the kernel
>>> >> >> source, and most of the places which use it before wake_up are either
>>> >> >> protected by some spin lock, or already has a memory barrier or some
>>> >> >> kind of atomic operation before it.
>>> >> >> 
>>> >> >> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
>>> >> >> many cases and won't be a good idea.
>>> >> >> 
>>> >> >> Another way to solve this problem is to remove the waitqueue_active(),
>>> >> >> making the code look like this;
>>> >> >>   if (wq)
>>> >> >>   wake_up_interruptible(wq);
>>> >> >> This also fixes the problem because the spinlock in the wake_up*() 
>>> >> >> acts
>>> >> >> as a memory barrier and prevents the code from being reordered by the
>>> >> >> CPU (and it also makes the resulting code is much simpler).
>>> >> > 
>>> >> > I might not care which we did, except I don't have the means to test
>>> >> > this quickly, and I guess this is some of our most frequently called
>>> >> > code.
>>> >> > 
>>> >> > I suppose your patch is the most conservative approach, as the
>>> >> > alternative is a spinlock/unlock in wake_up_interruptible, which I
>>> >> > assume is necessarily more expensive than an smp_mb().
>>> >> > 
>>> >> > As far as I can tell it's been this way since forever.  (Well, since a
>>> >> > 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" 
>>> >> > which
>>> >> > removed some spinlocks from the data_ready routines.)
>>> >> > 
>>> >> > I don't understand what the actual race is yet (which code exactly is
>>> >> > missing the wakeup in this case?  nfsd threads seem to instead get
>>> >> > woken up by the wake_up_process() in svc_xprt_do_enqueue().)
>>> >> 
>>> >> Thank you for the reply.  I tried looking into this.
>>> >> 
>>> >> The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
>>> >> svc_udp_init(), which are both called from svc_setup_socket().
>>> >> svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
>>> >> callback port related code.
>>> >> 
>>> >> Maybe I'm wrong, but there might not be any kernel code that is using
>>> >> the socket's wait queue in this case.
>>> > 
>>> > As Trond points out there are probably waiters internal to the
>>> > networking code.
>>> 
>>> Trond and Bruce, thank you for the comment.  I was able to find the call
>>> to the wait function that was called from nfsd.
>>> 
>>> sk_stream_wait_connect() and sk_stream_wait_memory() were called from
>>> either do_tcp_sendpages() or tcp_sendmsg() called from within
>>> svc_send().  sk_stream_wait_connect() shouldn't be called at this point,
>>> because the socket has already been used to receive the rpc request.
>>> 
>>> On the wake_up side, sk_write_space() is called from the following
>>> locations.  The relevant ones seems to be preceded by atomic_sub or a
>>> memory barrier.
>>> + ksocknal_write_space 
>>> [drivers/staging/lustre/lnet/klnds/socklnd/socklnd_lib.c:633]
>>> + atm_pop_raw [net/atm/raw.c:40]
>>> + sock_setsockopt [net/core/sock.c:740]
>>> + sock_wfree [net/core/sock.c:1630]
>>>   Preceded by atomic_sub in sock_wfree()
>>> + ccid3_hc_tx_packet_recv [net/dccp/ccids/ccid3.c:442]
>>> + do_tcp_sendpages [net/ipv4/tcp.c:1008]
>>> + tcp_sendmsg [net/ipv4/tcp.c:1300]
>>> + do_tcp_setsockopt [net/ipv4/tcp.c:2597]
>>> + tcp_new_space [net/ipv4/tcp_input.c:4885]
>>>   Preceded by smp_mb__after_atomic in tcp_check_space()

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
> Tatsukawa Kosuke wrote:
> > J. Bruce Fields wrote:
> >> Thanks for the detailed investigation.
> >> 
> >> I think it would be worth adding a comment if that might help someone
> >> having to reinvestigate this again some day.
> > 
> > It would be nice, but I find it difficult to write a comment in the
> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
> > of how nfsd uses it, and the current implementation of the network code.
> > 
> > Personally, I would prefer removing the call to waitqueue_active() which
> > would make the memory barrier totally unnecessary at the cost of a
> > spin_lock + spin_unlock by unconditionally calling
> > wake_up_interruptible.
> 
> On second thought, the callbacks will be called frequently from the tcp
> code, so it wouldn't be a good idea.

So, I was even considering documenting it like this, if it's not
overkill.

Hmm... but if this is right, then we may as well ask why we're doing the
wakeups at all.  Might be educational to test the code with them
removed.

--b.

commit 0882cfeb39e0
Author: J. Bruce Fields 
Date:   Thu Oct 15 16:53:41 2015 -0400

svcrpc: document lack of some memory barriers.

Kosuke Tatsukawa points out an odd lack of memory barriers in some sites
here.  I think the code's correct, but it's probably worth documenting.

Reported-by: Kosuke Tatsukawa 

diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 856407fa085e..90480993ec4a 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
return svc_port_is_privileged(svc_addr(rqstp));
 }
 
+static void svc_no_smp_mb(void)
+{
+   /*
+* Kosuke Tatsukawa points out there should normally be an
+* smp_mb() at the callsites of this function.  (Either that or
+* we could just drop the waitqueue_active() checks.)
+*
+* It appears they aren't currently necessary, though, basically
+* because nfsd does non-blocking reads from these sockets, so
+* the only places we wait on this waitqueue is in sendpage and
+* sendmsg, which won't be waiting for wakeups on newly arrived
+* data.
+*
+* Maybe we should add the memory barriers anyway, but these are
+* hot paths so we'd need to be convinced there's no sigificant
+* penalty.
+*/
+}
+
 /*
  * INET callback when data has been received on the socket.
  */
@@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
set_bit(XPT_DATA, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
@@ -433,7 +452,7 @@ static void svc_write_space(struct sock *sk)
svc_xprt_enqueue(>sk_xprt);
}
 
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq)) {
dprintk("RPC svc_write_space: someone sleeping on %p\n",
   svsk);
@@ -789,7 +808,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
}
 
wq = sk_sleep(sk);
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible_all(wq);
 }
@@ -811,7 +830,7 @@ static void svc_tcp_state_change(struct sock *sk)
set_bit(XPT_CLOSE, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible_all(wq);
 }
@@ -827,7 +846,7 @@ static void svc_tcp_data_ready(struct sock *sk)
set_bit(XPT_DATA, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
@@ -1599,7 +1618,7 @@ static void svc_sock_detach(struct svc_xprt *xprt)
sk->sk_write_space = svsk->sk_owspace;
 
wq = sk_sleep(sk);
-   smp_mb();
+   svc_no_smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
>> Tatsukawa Kosuke wrote:
>> > J. Bruce Fields wrote:
>> >> Thanks for the detailed investigation.
>> >> 
>> >> I think it would be worth adding a comment if that might help someone
>> >> having to reinvestigate this again some day.
>> > 
>> > It would be nice, but I find it difficult to write a comment in the
>> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
>> > of how nfsd uses it, and the current implementation of the network code.
>> > 
>> > Personally, I would prefer removing the call to waitqueue_active() which
>> > would make the memory barrier totally unnecessary at the cost of a
>> > spin_lock + spin_unlock by unconditionally calling
>> > wake_up_interruptible.
>> 
>> On second thought, the callbacks will be called frequently from the tcp
>> code, so it wouldn't be a good idea.
> 
> So, I was even considering documenting it like this, if it's not
> overkill.
> 
> Hmm... but if this is right, then we may as well ask why we're doing the
> wakeups at all.  Might be educational to test the code with them
> removed.

sk_write_space will be called in sock_wfree() with UDP/IP each time
kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
SOCK_NOSPACE has been set.

sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
barrier with sk_data_ready called right after __skb_queue_tail().
I think this hasn't caused any problems because sk_data_ready wasn't
used.


> --b.
> 
> commit 0882cfeb39e0
> Author: J. Bruce Fields 
> Date:   Thu Oct 15 16:53:41 2015 -0400
> 
> svcrpc: document lack of some memory barriers.
> 
> Kosuke Tatsukawa points out an odd lack of memory barriers in some sites
> here.  I think the code's correct, but it's probably worth documenting.
> 
> Reported-by: Kosuke Tatsukawa 
> 
> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> index 856407fa085e..90480993ec4a 100644
> --- a/net/sunrpc/svcsock.c
> +++ b/net/sunrpc/svcsock.c
> @@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
>   return svc_port_is_privileged(svc_addr(rqstp));
>  }
>  
> +static void svc_no_smp_mb(void)
> +{
> + /*
> +  * Kosuke Tatsukawa points out there should normally be an
> +  * smp_mb() at the callsites of this function.  (Either that or
> +  * we could just drop the waitqueue_active() checks.)
> +  *
> +  * It appears they aren't currently necessary, though, basically
> +  * because nfsd does non-blocking reads from these sockets, so
> +  * the only places we wait on this waitqueue is in sendpage and
> +  * sendmsg, which won't be waiting for wakeups on newly arrived
> +  * data.
> +  *
> +  * Maybe we should add the memory barriers anyway, but these are
> +  * hot paths so we'd need to be convinced there's no sigificant
> +  * penalty.
> +  */
> +}
> +
>  /*
>   * INET callback when data has been received on the socket.
>   */
> @@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }
> @@ -433,7 +452,7 @@ static void svc_write_space(struct sock *sk)
>   svc_xprt_enqueue(>sk_xprt);
>   }
>  
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq)) {
>   dprintk("RPC svc_write_space: someone sleeping on %p\n",
>  svsk);
> @@ -789,7 +808,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
>   }
>  
>   wq = sk_sleep(sk);
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible_all(wq);
>  }
> @@ -811,7 +830,7 @@ static void svc_tcp_state_change(struct sock *sk)
>   set_bit(XPT_CLOSE, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible_all(wq);
>  }
> @@ -827,7 +846,7 @@ static void svc_tcp_data_ready(struct sock *sk)
>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }
> @@ -1599,7 +1618,7 @@ static void svc_sock_detach(struct svc_xprt *xprt)
>   sk->sk_write_space = svsk->sk_owspace;
>  
>   wq = sk_sleep(sk);
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }
---
Kosuke TATSUKAWA  | 3rd IT Platform Department
   

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Neil Brown]

"J. Bruce Fields"  writes:

> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
>> Tatsukawa Kosuke wrote:
>> > J. Bruce Fields wrote:
>> >> Thanks for the detailed investigation.
>> >> 
>> >> I think it would be worth adding a comment if that might help someone
>> >> having to reinvestigate this again some day.
>> > 
>> > It would be nice, but I find it difficult to write a comment in the
>> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
>> > of how nfsd uses it, and the current implementation of the network code.
>> > 
>> > Personally, I would prefer removing the call to waitqueue_active() which
>> > would make the memory barrier totally unnecessary at the cost of a
>> > spin_lock + spin_unlock by unconditionally calling
>> > wake_up_interruptible.
>> 
>> On second thought, the callbacks will be called frequently from the tcp
>> code, so it wouldn't be a good idea.
>
> So, I was even considering documenting it like this, if it's not
> overkill.
>
> Hmm... but if this is right, then we may as well ask why we're doing the
> wakeups at all.  Might be educational to test the code with them
> removed.
>
> --b.
>
> commit 0882cfeb39e0
> Author: J. Bruce Fields 
> Date:   Thu Oct 15 16:53:41 2015 -0400
>
> svcrpc: document lack of some memory barriers.
> 
> Kosuke Tatsukawa points out an odd lack of memory barriers in some sites
> here.  I think the code's correct, but it's probably worth documenting.
> 
> Reported-by: Kosuke Tatsukawa 
>
> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> index 856407fa085e..90480993ec4a 100644
> --- a/net/sunrpc/svcsock.c
> +++ b/net/sunrpc/svcsock.c
> @@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
>   return svc_port_is_privileged(svc_addr(rqstp));
>  }
>  
> +static void svc_no_smp_mb(void)
> +{
> + /*
> +  * Kosuke Tatsukawa points out there should normally be an
> +  * smp_mb() at the callsites of this function.  (Either that or
> +  * we could just drop the waitqueue_active() checks.)
> +  *
> +  * It appears they aren't currently necessary, though, basically
> +  * because nfsd does non-blocking reads from these sockets, so
> +  * the only places we wait on this waitqueue is in sendpage and
> +  * sendmsg, which won't be waiting for wakeups on newly arrived
> +  * data.
> +  *
> +  * Maybe we should add the memory barriers anyway, but these are
> +  * hot paths so we'd need to be convinced there's no sigificant
> +  * penalty.
> +  */
> +}
> +
>  /*
>   * INET callback when data has been received on the socket.
>   */
> @@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }
> @@ -433,7 +452,7 @@ static void svc_write_space(struct sock *sk)
>   svc_xprt_enqueue(>sk_xprt);
>   }
>  
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq)) {
>   dprintk("RPC svc_write_space: someone sleeping on %p\n",
>  svsk);
> @@ -789,7 +808,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
>   }
>  
>   wq = sk_sleep(sk);
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible_all(wq);
>  }
> @@ -811,7 +830,7 @@ static void svc_tcp_state_change(struct sock *sk)
>   set_bit(XPT_CLOSE, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible_all(wq);
>  }
> @@ -827,7 +846,7 @@ static void svc_tcp_data_ready(struct sock *sk)
>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }
> @@ -1599,7 +1618,7 @@ static void svc_sock_detach(struct svc_xprt *xprt)
>   sk->sk_write_space = svsk->sk_owspace;
>  
>   wq = sk_sleep(sk);
> - smp_mb();
> + svc_no_smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }

I would feel a lot more comfortable if you instead created:

static inline bool sunrpc_waitqueue_active(struct wait_queue_head *wq)
{
if (!wq)
return false;
/* long comment abot not needing a memory barrier */
return waitqueue_active(wq);
}

and then replace various "if (wq && waitqueue_active(wq))" calls with
if (sunrpc_waitqueue_active(wq))"

The comment seems readable and seems to make sense.

NeilBrown


signature.asc
Description: 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

Tatsukawa Kosuke wrote:
> J. Bruce Fields wrote:
>> On Thu, Oct 15, 2015 at 11:44:20AM +, Kosuke Tatsukawa wrote:
>>> Tatsukawa Kosuke wrote:
>>> > J. Bruce Fields wrote:
>>> >> Thanks for the detailed investigation.
>>> >> 
>>> >> I think it would be worth adding a comment if that might help someone
>>> >> having to reinvestigate this again some day.
>>> > 
>>> > It would be nice, but I find it difficult to write a comment in the
>>> > sunrpc layer why a memory barrier isn't necessary, using the knowledge
>>> > of how nfsd uses it, and the current implementation of the network code.
>>> > 
>>> > Personally, I would prefer removing the call to waitqueue_active() which
>>> > would make the memory barrier totally unnecessary at the cost of a
>>> > spin_lock + spin_unlock by unconditionally calling
>>> > wake_up_interruptible.
>>> 
>>> On second thought, the callbacks will be called frequently from the tcp
>>> code, so it wouldn't be a good idea.
>> 
>> So, I was even considering documenting it like this, if it's not
>> overkill.
>> 
>> Hmm... but if this is right, then we may as well ask why we're doing the
>> wakeups at all.  Might be educational to test the code with them
>> removed.
> 
> sk_write_space will be called in sock_wfree() with UDP/IP each time
> kfree_skb() is called.  With TCP/IP, sk_write_space is only called if
> SOCK_NOSPACE has been set.
> 
> sk_data_ready will be called in both tcp_rcv_established() for TCP/IP
> and in sock_queue_rcv_skb() for UDP/IP.  The latter lacks a memory
> barrier with sk_data_ready called right after __skb_queue_tail().
> I think this hasn't caused any problems because sk_data_ready wasn't
> used.

Actually, svc_udp_data_ready() calls set_bit() which is an atomic
operation.  So there won't be a problem unless svsk is NULL.


>> --b.
>> 
>> commit 0882cfeb39e0
>> Author: J. Bruce Fields 
>> Date:   Thu Oct 15 16:53:41 2015 -0400
>> 
>> svcrpc: document lack of some memory barriers.
>> 
>> Kosuke Tatsukawa points out an odd lack of memory barriers in some sites
>> here.  I think the code's correct, but it's probably worth documenting.
>> 
>> Reported-by: Kosuke Tatsukawa 
>> 
>> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
>> index 856407fa085e..90480993ec4a 100644
>> --- a/net/sunrpc/svcsock.c
>> +++ b/net/sunrpc/svcsock.c
>> @@ -399,6 +399,25 @@ static int svc_sock_secure_port(struct svc_rqst *rqstp)
>>  return svc_port_is_privileged(svc_addr(rqstp));
>>  }
>>  
>> +static void svc_no_smp_mb(void)
>> +{
>> +/*
>> + * Kosuke Tatsukawa points out there should normally be an
>> + * smp_mb() at the callsites of this function.  (Either that or
>> + * we could just drop the waitqueue_active() checks.)
>> + *
>> + * It appears they aren't currently necessary, though, basically
>> + * because nfsd does non-blocking reads from these sockets, so
>> + * the only places we wait on this waitqueue is in sendpage and
>> + * sendmsg, which won't be waiting for wakeups on newly arrived
>> + * data.
>> + *
>> + * Maybe we should add the memory barriers anyway, but these are
>> + * hot paths so we'd need to be convinced there's no sigificant
>> + * penalty.
>> + */
>> +}
>> +
>>  /*
>>   * INET callback when data has been received on the socket.
>>   */
>> @@ -414,7 +433,7 @@ static void svc_udp_data_ready(struct sock *sk)
>>  set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>>  svc_xprt_enqueue(>sk_xprt);
>>  }
>> -smp_mb();
>> +svc_no_smp_mb();
>>  if (wq && waitqueue_active(wq))
>>  wake_up_interruptible(wq);
>>  }
>> @@ -433,7 +452,7 @@ static void svc_write_space(struct sock *sk)
>>  svc_xprt_enqueue(>sk_xprt);
>>  }
>>  
>> -smp_mb();
>> +svc_no_smp_mb();
>>  if (wq && waitqueue_active(wq)) {
>>  dprintk("RPC svc_write_space: someone sleeping on %p\n",
>> svsk);
>> @@ -789,7 +808,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
>>  }
>>  
>>  wq = sk_sleep(sk);
>> -smp_mb();
>> +svc_no_smp_mb();
>>  if (wq && waitqueue_active(wq))
>>  wake_up_interruptible_all(wq);
>>  }
>> @@ -811,7 +830,7 @@ static void svc_tcp_state_change(struct sock *sk)
>>  set_bit(XPT_CLOSE, >sk_xprt.xpt_flags);
>>  svc_xprt_enqueue(>sk_xprt);
>>  }
>> -smp_mb();
>> +svc_no_smp_mb();
>>  if (wq && waitqueue_active(wq))
>>  wake_up_interruptible_all(wq);
>>  }
>> @@ -827,7 +846,7 @@ static void svc_tcp_data_ready(struct sock *sk)
>>  set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>>  svc_xprt_enqueue(>sk_xprt);
>>  }
>> -smp_mb();
>> +svc_no_smp_mb();
>>  if (wq && waitqueue_active(wq))
>>  wake_up_interruptible(wq);
>>  }
>> @@ -1599,7 +1618,7 @@ static void svc_sock_detach(struct svc_xprt *xprt)
>>  sk->sk_write_space 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Wed, Oct 14, 2015 at 03:57:13AM +, Kosuke Tatsukawa wrote:
> J. Bruce Fields wrote:
> > On Mon, Oct 12, 2015 at 10:41:06AM +, Kosuke Tatsukawa wrote:
> >> J. Bruce Fields wrote:
> >> > On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
> >> >> Neil Brown wrote:
> >> >> > Kosuke Tatsukawa  writes:
> >> >> > 
> >> >> >> There are several places in net/sunrpc/svcsock.c which calls
> >> >> >> waitqueue_active() without calling a memory barrier.  Add a memory
> >> >> >> barrier just as in wq_has_sleeper().
> >> >> >>
> >> >> >> I found this issue when I was looking through the linux source code
> >> >> >> for places calling waitqueue_active() before wake_up*(), but without
> >> >> >> preceding memory barriers, after sending a patch to fix a similar
> >> >> >> issue in drivers/tty/n_tty.c  (Details about the original issue can 
> >> >> >> be
> >> >> >> found here: https://lkml.org/lkml/2015/9/28/849).
> >> >> > 
> >> >> > hi,
> >> >> > this feels like the wrong approach to the problem.  It requires extra
> >> >> > 'smb_mb's to be spread around which are hard to understand as easy to
> >> >> > forget.
> >> >> > 
> >> >> > A quick look seems to suggest that (nearly) every waitqueue_active()
> >> >> > will need an smb_mb.  Could we just put the smb_mb() inside
> >> >> > waitqueue_active()??
> >> >> 
> >> >> 
> >> >> There are around 200 occurrences of waitqueue_active() in the kernel
> >> >> source, and most of the places which use it before wake_up are either
> >> >> protected by some spin lock, or already has a memory barrier or some
> >> >> kind of atomic operation before it.
> >> >> 
> >> >> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
> >> >> many cases and won't be a good idea.
> >> >> 
> >> >> Another way to solve this problem is to remove the waitqueue_active(),
> >> >> making the code look like this;
> >> >> if (wq)
> >> >> wake_up_interruptible(wq);
> >> >> This also fixes the problem because the spinlock in the wake_up*() acts
> >> >> as a memory barrier and prevents the code from being reordered by the
> >> >> CPU (and it also makes the resulting code is much simpler).
> >> > 
> >> > I might not care which we did, except I don't have the means to test
> >> > this quickly, and I guess this is some of our most frequently called
> >> > code.
> >> > 
> >> > I suppose your patch is the most conservative approach, as the
> >> > alternative is a spinlock/unlock in wake_up_interruptible, which I
> >> > assume is necessarily more expensive than an smp_mb().
> >> > 
> >> > As far as I can tell it's been this way since forever.  (Well, since a
> >> > 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
> >> > removed some spinlocks from the data_ready routines.)
> >> > 
> >> > I don't understand what the actual race is yet (which code exactly is
> >> > missing the wakeup in this case?  nfsd threads seem to instead get
> >> > woken up by the wake_up_process() in svc_xprt_do_enqueue().)
> >> 
> >> Thank you for the reply.  I tried looking into this.
> >> 
> >> The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
> >> svc_udp_init(), which are both called from svc_setup_socket().
> >> svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
> >> callback port related code.
> >> 
> >> Maybe I'm wrong, but there might not be any kernel code that is using
> >> the socket's wait queue in this case.
> > 
> > As Trond points out there are probably waiters internal to the
> > networking code.
> 
> Trond and Bruce, thank you for the comment.  I was able to find the call
> to the wait function that was called from nfsd.
> 
> sk_stream_wait_connect() and sk_stream_wait_memory() were called from
> either do_tcp_sendpages() or tcp_sendmsg() called from within
> svc_send().  sk_stream_wait_connect() shouldn't be called at this point,
> because the socket has already been used to receive the rpc request.
> 
> On the wake_up side, sk_write_space() is called from the following
> locations.  The relevant ones seems to be preceded by atomic_sub or a
> memory barrier.
> + ksocknal_write_space 
> [drivers/staging/lustre/lnet/klnds/socklnd/socklnd_lib.c:633]
> + atm_pop_raw [net/atm/raw.c:40]
> + sock_setsockopt [net/core/sock.c:740]
> + sock_wfree [net/core/sock.c:1630]
>   Preceded by atomic_sub in sock_wfree()
> + ccid3_hc_tx_packet_recv [net/dccp/ccids/ccid3.c:442]
> + do_tcp_sendpages [net/ipv4/tcp.c:1008]
> + tcp_sendmsg [net/ipv4/tcp.c:1300]
> + do_tcp_setsockopt [net/ipv4/tcp.c:2597]
> + tcp_new_space [net/ipv4/tcp_input.c:4885]
>   Preceded by smp_mb__after_atomic in tcp_check_space()
> + llc_conn_state_process [net/llc/llc_conn.c:148]
> + pipe_rcv_status [net/phonet/pep.c:312]
> + pipe_do_rcv [net/phonet/pep.c:440]
> + pipe_start_flow_control [net/phonet/pep.c:554]
> + svc_sock_setbufsize [net/sunrpc/svcsock.c:45]
> 
> sk_state_change() calls related to TCP/IP were called from the 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Wed, Oct 14, 2015 at 03:57:13AM +, Kosuke Tatsukawa wrote:
>> J. Bruce Fields wrote:
>> > On Mon, Oct 12, 2015 at 10:41:06AM +, Kosuke Tatsukawa wrote:
>> >> J. Bruce Fields wrote:
>> >> > On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
>> >> >> Neil Brown wrote:
>> >> >> > Kosuke Tatsukawa  writes:
>> >> >> > 
>> >> >> >> There are several places in net/sunrpc/svcsock.c which calls
>> >> >> >> waitqueue_active() without calling a memory barrier.  Add a memory
>> >> >> >> barrier just as in wq_has_sleeper().
>> >> >> >>
>> >> >> >> I found this issue when I was looking through the linux source code
>> >> >> >> for places calling waitqueue_active() before wake_up*(), but without
>> >> >> >> preceding memory barriers, after sending a patch to fix a similar
>> >> >> >> issue in drivers/tty/n_tty.c  (Details about the original issue can 
>> >> >> >> be
>> >> >> >> found here: https://lkml.org/lkml/2015/9/28/849).
>> >> >> > 
>> >> >> > hi,
>> >> >> > this feels like the wrong approach to the problem.  It requires extra
>> >> >> > 'smb_mb's to be spread around which are hard to understand as easy to
>> >> >> > forget.
>> >> >> > 
>> >> >> > A quick look seems to suggest that (nearly) every waitqueue_active()
>> >> >> > will need an smb_mb.  Could we just put the smb_mb() inside
>> >> >> > waitqueue_active()??
>> >> >> 
>> >> >> 
>> >> >> There are around 200 occurrences of waitqueue_active() in the kernel
>> >> >> source, and most of the places which use it before wake_up are either
>> >> >> protected by some spin lock, or already has a memory barrier or some
>> >> >> kind of atomic operation before it.
>> >> >> 
>> >> >> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
>> >> >> many cases and won't be a good idea.
>> >> >> 
>> >> >> Another way to solve this problem is to remove the waitqueue_active(),
>> >> >> making the code look like this;
>> >> >>if (wq)
>> >> >>wake_up_interruptible(wq);
>> >> >> This also fixes the problem because the spinlock in the wake_up*() acts
>> >> >> as a memory barrier and prevents the code from being reordered by the
>> >> >> CPU (and it also makes the resulting code is much simpler).
>> >> > 
>> >> > I might not care which we did, except I don't have the means to test
>> >> > this quickly, and I guess this is some of our most frequently called
>> >> > code.
>> >> > 
>> >> > I suppose your patch is the most conservative approach, as the
>> >> > alternative is a spinlock/unlock in wake_up_interruptible, which I
>> >> > assume is necessarily more expensive than an smp_mb().
>> >> > 
>> >> > As far as I can tell it's been this way since forever.  (Well, since a
>> >> > 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
>> >> > removed some spinlocks from the data_ready routines.)
>> >> > 
>> >> > I don't understand what the actual race is yet (which code exactly is
>> >> > missing the wakeup in this case?  nfsd threads seem to instead get
>> >> > woken up by the wake_up_process() in svc_xprt_do_enqueue().)
>> >> 
>> >> Thank you for the reply.  I tried looking into this.
>> >> 
>> >> The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
>> >> svc_udp_init(), which are both called from svc_setup_socket().
>> >> svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
>> >> callback port related code.
>> >> 
>> >> Maybe I'm wrong, but there might not be any kernel code that is using
>> >> the socket's wait queue in this case.
>> > 
>> > As Trond points out there are probably waiters internal to the
>> > networking code.
>> 
>> Trond and Bruce, thank you for the comment.  I was able to find the call
>> to the wait function that was called from nfsd.
>> 
>> sk_stream_wait_connect() and sk_stream_wait_memory() were called from
>> either do_tcp_sendpages() or tcp_sendmsg() called from within
>> svc_send().  sk_stream_wait_connect() shouldn't be called at this point,
>> because the socket has already been used to receive the rpc request.
>> 
>> On the wake_up side, sk_write_space() is called from the following
>> locations.  The relevant ones seems to be preceded by atomic_sub or a
>> memory barrier.
>> + ksocknal_write_space 
>> [drivers/staging/lustre/lnet/klnds/socklnd/socklnd_lib.c:633]
>> + atm_pop_raw [net/atm/raw.c:40]
>> + sock_setsockopt [net/core/sock.c:740]
>> + sock_wfree [net/core/sock.c:1630]
>>   Preceded by atomic_sub in sock_wfree()
>> + ccid3_hc_tx_packet_recv [net/dccp/ccids/ccid3.c:442]
>> + do_tcp_sendpages [net/ipv4/tcp.c:1008]
>> + tcp_sendmsg [net/ipv4/tcp.c:1300]
>> + do_tcp_setsockopt [net/ipv4/tcp.c:2597]
>> + tcp_new_space [net/ipv4/tcp_input.c:4885]
>>   Preceded by smp_mb__after_atomic in tcp_check_space()
>> + llc_conn_state_process [net/llc/llc_conn.c:148]
>> + pipe_rcv_status [net/phonet/pep.c:312]
>> + pipe_do_rcv [net/phonet/pep.c:440]
>> + pipe_start_flow_control 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Mon, Oct 12, 2015 at 10:41:06AM +, Kosuke Tatsukawa wrote:
>> J. Bruce Fields wrote:
>> > On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
>> >> Neil Brown wrote:
>> >> > Kosuke Tatsukawa  writes:
>> >> > 
>> >> >> There are several places in net/sunrpc/svcsock.c which calls
>> >> >> waitqueue_active() without calling a memory barrier.  Add a memory
>> >> >> barrier just as in wq_has_sleeper().
>> >> >>
>> >> >> I found this issue when I was looking through the linux source code
>> >> >> for places calling waitqueue_active() before wake_up*(), but without
>> >> >> preceding memory barriers, after sending a patch to fix a similar
>> >> >> issue in drivers/tty/n_tty.c  (Details about the original issue can be
>> >> >> found here: https://lkml.org/lkml/2015/9/28/849).
>> >> > 
>> >> > hi,
>> >> > this feels like the wrong approach to the problem.  It requires extra
>> >> > 'smb_mb's to be spread around which are hard to understand as easy to
>> >> > forget.
>> >> > 
>> >> > A quick look seems to suggest that (nearly) every waitqueue_active()
>> >> > will need an smb_mb.  Could we just put the smb_mb() inside
>> >> > waitqueue_active()??
>> >> 
>> >> 
>> >> There are around 200 occurrences of waitqueue_active() in the kernel
>> >> source, and most of the places which use it before wake_up are either
>> >> protected by some spin lock, or already has a memory barrier or some
>> >> kind of atomic operation before it.
>> >> 
>> >> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
>> >> many cases and won't be a good idea.
>> >> 
>> >> Another way to solve this problem is to remove the waitqueue_active(),
>> >> making the code look like this;
>> >>   if (wq)
>> >>   wake_up_interruptible(wq);
>> >> This also fixes the problem because the spinlock in the wake_up*() acts
>> >> as a memory barrier and prevents the code from being reordered by the
>> >> CPU (and it also makes the resulting code is much simpler).
>> > 
>> > I might not care which we did, except I don't have the means to test
>> > this quickly, and I guess this is some of our most frequently called
>> > code.
>> > 
>> > I suppose your patch is the most conservative approach, as the
>> > alternative is a spinlock/unlock in wake_up_interruptible, which I
>> > assume is necessarily more expensive than an smp_mb().
>> > 
>> > As far as I can tell it's been this way since forever.  (Well, since a
>> > 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
>> > removed some spinlocks from the data_ready routines.)
>> > 
>> > I don't understand what the actual race is yet (which code exactly is
>> > missing the wakeup in this case?  nfsd threads seem to instead get
>> > woken up by the wake_up_process() in svc_xprt_do_enqueue().)
>> 
>> Thank you for the reply.  I tried looking into this.
>> 
>> The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
>> svc_udp_init(), which are both called from svc_setup_socket().
>> svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
>> callback port related code.
>> 
>> Maybe I'm wrong, but there might not be any kernel code that is using
>> the socket's wait queue in this case.
> 
> As Trond points out there are probably waiters internal to the
> networking code.

Trond and Bruce, thank you for the comment.  I was able to find the call
to the wait function that was called from nfsd.

sk_stream_wait_connect() and sk_stream_wait_memory() were called from
either do_tcp_sendpages() or tcp_sendmsg() called from within
svc_send().  sk_stream_wait_connect() shouldn't be called at this point,
because the socket has already been used to receive the rpc request.

On the wake_up side, sk_write_space() is called from the following
locations.  The relevant ones seems to be preceded by atomic_sub or a
memory barrier.
+ ksocknal_write_space 
[drivers/staging/lustre/lnet/klnds/socklnd/socklnd_lib.c:633]
+ atm_pop_raw [net/atm/raw.c:40]
+ sock_setsockopt [net/core/sock.c:740]
+ sock_wfree [net/core/sock.c:1630]
  Preceded by atomic_sub in sock_wfree()
+ ccid3_hc_tx_packet_recv [net/dccp/ccids/ccid3.c:442]
+ do_tcp_sendpages [net/ipv4/tcp.c:1008]
+ tcp_sendmsg [net/ipv4/tcp.c:1300]
+ do_tcp_setsockopt [net/ipv4/tcp.c:2597]
+ tcp_new_space [net/ipv4/tcp_input.c:4885]
  Preceded by smp_mb__after_atomic in tcp_check_space()
+ llc_conn_state_process [net/llc/llc_conn.c:148]
+ pipe_rcv_status [net/phonet/pep.c:312]
+ pipe_do_rcv [net/phonet/pep.c:440]
+ pipe_start_flow_control [net/phonet/pep.c:554]
+ svc_sock_setbufsize [net/sunrpc/svcsock.c:45]

sk_state_change() calls related to TCP/IP were called from the following
places.
+ inet_shutdown [net/ipv4/af_inet.c:825]
  This shouldn't be called when waiting
+ tcp_done [net/ipv4/tcp.c:3078]
  spin_lock*/spin_unlock* is called in lock_timer_base
+ tcp_fin [net/ipv4/tcp_input.c:4031]
  atomic_long_sub is called from sk_memory_allocated_sub called 

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

J. Bruce Fields wrote:
> On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
>> Neil Brown wrote:
>> > Kosuke Tatsukawa  writes:
>> > 
>> >> There are several places in net/sunrpc/svcsock.c which calls
>> >> waitqueue_active() without calling a memory barrier.  Add a memory
>> >> barrier just as in wq_has_sleeper().
>> >>
>> >> I found this issue when I was looking through the linux source code
>> >> for places calling waitqueue_active() before wake_up*(), but without
>> >> preceding memory barriers, after sending a patch to fix a similar
>> >> issue in drivers/tty/n_tty.c  (Details about the original issue can be
>> >> found here: https://lkml.org/lkml/2015/9/28/849).
>> > 
>> > hi,
>> > this feels like the wrong approach to the problem.  It requires extra
>> > 'smb_mb's to be spread around which are hard to understand as easy to
>> > forget.
>> > 
>> > A quick look seems to suggest that (nearly) every waitqueue_active()
>> > will need an smb_mb.  Could we just put the smb_mb() inside
>> > waitqueue_active()??
>> 
>> 
>> There are around 200 occurrences of waitqueue_active() in the kernel
>> source, and most of the places which use it before wake_up are either
>> protected by some spin lock, or already has a memory barrier or some
>> kind of atomic operation before it.
>> 
>> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
>> many cases and won't be a good idea.
>> 
>> Another way to solve this problem is to remove the waitqueue_active(),
>> making the code look like this;
>>  if (wq)
>>  wake_up_interruptible(wq);
>> This also fixes the problem because the spinlock in the wake_up*() acts
>> as a memory barrier and prevents the code from being reordered by the
>> CPU (and it also makes the resulting code is much simpler).
> 
> I might not care which we did, except I don't have the means to test
> this quickly, and I guess this is some of our most frequently called
> code.
> 
> I suppose your patch is the most conservative approach, as the
> alternative is a spinlock/unlock in wake_up_interruptible, which I
> assume is necessarily more expensive than an smp_mb().
> 
> As far as I can tell it's been this way since forever.  (Well, since a
> 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
> removed some spinlocks from the data_ready routines.)
> 
> I don't understand what the actual race is yet (which code exactly is
> missing the wakeup in this case?  nfsd threads seem to instead get
> woken up by the wake_up_process() in svc_xprt_do_enqueue().)

Thank you for the reply.  I tried looking into this.

The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
svc_udp_init(), which are both called from svc_setup_socket().
svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
callback port related code.

Maybe I'm wrong, but there might not be any kernel code that is using
the socket's wait queue in this case.

Best regards.
---
Kosuke TATSUKAWA  | 3rd IT Platform Department
  | IT Platform Division, NEC Corporation
  | ta...@ab.jp.nec.com
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Mon, Oct 12, 2015 at 10:41:06AM +, Kosuke Tatsukawa wrote:
> J. Bruce Fields wrote:
> > On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
> >> Neil Brown wrote:
> >> > Kosuke Tatsukawa  writes:
> >> > 
> >> >> There are several places in net/sunrpc/svcsock.c which calls
> >> >> waitqueue_active() without calling a memory barrier.  Add a memory
> >> >> barrier just as in wq_has_sleeper().
> >> >>
> >> >> I found this issue when I was looking through the linux source code
> >> >> for places calling waitqueue_active() before wake_up*(), but without
> >> >> preceding memory barriers, after sending a patch to fix a similar
> >> >> issue in drivers/tty/n_tty.c  (Details about the original issue can be
> >> >> found here: https://lkml.org/lkml/2015/9/28/849).
> >> > 
> >> > hi,
> >> > this feels like the wrong approach to the problem.  It requires extra
> >> > 'smb_mb's to be spread around which are hard to understand as easy to
> >> > forget.
> >> > 
> >> > A quick look seems to suggest that (nearly) every waitqueue_active()
> >> > will need an smb_mb.  Could we just put the smb_mb() inside
> >> > waitqueue_active()??
> >> 
> >> 
> >> There are around 200 occurrences of waitqueue_active() in the kernel
> >> source, and most of the places which use it before wake_up are either
> >> protected by some spin lock, or already has a memory barrier or some
> >> kind of atomic operation before it.
> >> 
> >> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
> >> many cases and won't be a good idea.
> >> 
> >> Another way to solve this problem is to remove the waitqueue_active(),
> >> making the code look like this;
> >>if (wq)
> >>wake_up_interruptible(wq);
> >> This also fixes the problem because the spinlock in the wake_up*() acts
> >> as a memory barrier and prevents the code from being reordered by the
> >> CPU (and it also makes the resulting code is much simpler).
> > 
> > I might not care which we did, except I don't have the means to test
> > this quickly, and I guess this is some of our most frequently called
> > code.
> > 
> > I suppose your patch is the most conservative approach, as the
> > alternative is a spinlock/unlock in wake_up_interruptible, which I
> > assume is necessarily more expensive than an smp_mb().
> > 
> > As far as I can tell it's been this way since forever.  (Well, since a
> > 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
> > removed some spinlocks from the data_ready routines.)
> > 
> > I don't understand what the actual race is yet (which code exactly is
> > missing the wakeup in this case?  nfsd threads seem to instead get
> > woken up by the wake_up_process() in svc_xprt_do_enqueue().)
> 
> Thank you for the reply.  I tried looking into this.
> 
> The callbacks in net/sunrpc/svcsock.c are set up in svc_tcp_init() and
> svc_udp_init(), which are both called from svc_setup_socket().
> svc_setup_socket() is called (indirectly) from lockd, nfsd, and nfsv4
> callback port related code.
> 
> Maybe I'm wrong, but there might not be any kernel code that is using
> the socket's wait queue in this case.

As Trond points out there are probably waiters internal to the
networking code.

--b.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

Neil Brown wrote:
> Kosuke Tatsukawa  writes:
> 
>> There are several places in net/sunrpc/svcsock.c which calls
>> waitqueue_active() without calling a memory barrier.  Add a memory
>> barrier just as in wq_has_sleeper().
>>
>> I found this issue when I was looking through the linux source code
>> for places calling waitqueue_active() before wake_up*(), but without
>> preceding memory barriers, after sending a patch to fix a similar
>> issue in drivers/tty/n_tty.c  (Details about the original issue can be
>> found here: https://lkml.org/lkml/2015/9/28/849).
> 
> hi,
> this feels like the wrong approach to the problem.  It requires extra
> 'smb_mb's to be spread around which are hard to understand as easy to
> forget.
> 
> A quick look seems to suggest that (nearly) every waitqueue_active()
> will need an smb_mb.  Could we just put the smb_mb() inside
> waitqueue_active()??


There are around 200 occurrences of waitqueue_active() in the kernel
source, and most of the places which use it before wake_up are either
protected by some spin lock, or already has a memory barrier or some
kind of atomic operation before it.

Simply adding smp_mb() to waitqueue_active() would incur extra cost in
many cases and won't be a good idea.

Another way to solve this problem is to remove the waitqueue_active(),
making the code look like this;
if (wq)
wake_up_interruptible(wq);
This also fixes the problem because the spinlock in the wake_up*() acts
as a memory barrier and prevents the code from being reordered by the
CPU (and it also makes the resulting code is much simpler).
---
Kosuke TATSUKAWA  | 3rd IT Platform Department
  | IT Platform Division, NEC Corporation
  | ta...@ab.jp.nec.com
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Trond Myklebust]

On Fri, Oct 9, 2015 at 5:18 PM, J. Bruce Fields  wrote:
>
> On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
> > Neil Brown wrote:
> > > Kosuke Tatsukawa  writes:
> > >
> > >> There are several places in net/sunrpc/svcsock.c which calls
> > >> waitqueue_active() without calling a memory barrier.  Add a memory
> > >> barrier just as in wq_has_sleeper().
> > >>
> > >> I found this issue when I was looking through the linux source code
> > >> for places calling waitqueue_active() before wake_up*(), but without
> > >> preceding memory barriers, after sending a patch to fix a similar
> > >> issue in drivers/tty/n_tty.c  (Details about the original issue can be
> > >> found here: https://lkml.org/lkml/2015/9/28/849).
> > >
> > > hi,
> > > this feels like the wrong approach to the problem.  It requires extra
> > > 'smb_mb's to be spread around which are hard to understand as easy to
> > > forget.
> > >
> > > A quick look seems to suggest that (nearly) every waitqueue_active()
> > > will need an smb_mb.  Could we just put the smb_mb() inside
> > > waitqueue_active()??
> > 
> >
> > There are around 200 occurrences of waitqueue_active() in the kernel
> > source, and most of the places which use it before wake_up are either
> > protected by some spin lock, or already has a memory barrier or some
> > kind of atomic operation before it.
> >
> > Simply adding smp_mb() to waitqueue_active() would incur extra cost in
> > many cases and won't be a good idea.
> >
> > Another way to solve this problem is to remove the waitqueue_active(),
> > making the code look like this;
> >   if (wq)
> >   wake_up_interruptible(wq);
> > This also fixes the problem because the spinlock in the wake_up*() acts
> > as a memory barrier and prevents the code from being reordered by the
> > CPU (and it also makes the resulting code is much simpler).
>
> I might not care which we did, except I don't have the means to test
> this quickly, and I guess this is some of our most frequently called
> code.
>
> I suppose your patch is the most conservative approach, as the
> alternative is a spinlock/unlock in wake_up_interruptible, which I
> assume is necessarily more expensive than an smp_mb().
>
> As far as I can tell it's been this way since forever.  (Well, since a
> 2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
> removed some spinlocks from the data_ready routines.)
>
> I don't understand what the actual race is yet (which code exactly is
> missing the wakeup in this case?  nfsd threads seem to instead get
> woken up by the wake_up_process() in svc_xprt_do_enqueue().)
>

Those threads still use blocking calls for sendpage() and sendmsg(),
so presumably they may be affected.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[J. Bruce Fields]

On Fri, Oct 09, 2015 at 06:29:44AM +, Kosuke Tatsukawa wrote:
> Neil Brown wrote:
> > Kosuke Tatsukawa  writes:
> > 
> >> There are several places in net/sunrpc/svcsock.c which calls
> >> waitqueue_active() without calling a memory barrier.  Add a memory
> >> barrier just as in wq_has_sleeper().
> >>
> >> I found this issue when I was looking through the linux source code
> >> for places calling waitqueue_active() before wake_up*(), but without
> >> preceding memory barriers, after sending a patch to fix a similar
> >> issue in drivers/tty/n_tty.c  (Details about the original issue can be
> >> found here: https://lkml.org/lkml/2015/9/28/849).
> > 
> > hi,
> > this feels like the wrong approach to the problem.  It requires extra
> > 'smb_mb's to be spread around which are hard to understand as easy to
> > forget.
> > 
> > A quick look seems to suggest that (nearly) every waitqueue_active()
> > will need an smb_mb.  Could we just put the smb_mb() inside
> > waitqueue_active()??
> 
> 
> There are around 200 occurrences of waitqueue_active() in the kernel
> source, and most of the places which use it before wake_up are either
> protected by some spin lock, or already has a memory barrier or some
> kind of atomic operation before it.
> 
> Simply adding smp_mb() to waitqueue_active() would incur extra cost in
> many cases and won't be a good idea.
> 
> Another way to solve this problem is to remove the waitqueue_active(),
> making the code look like this;
>   if (wq)
>   wake_up_interruptible(wq);
> This also fixes the problem because the spinlock in the wake_up*() acts
> as a memory barrier and prevents the code from being reordered by the
> CPU (and it also makes the resulting code is much simpler).

I might not care which we did, except I don't have the means to test
this quickly, and I guess this is some of our most frequently called
code.

I suppose your patch is the most conservative approach, as the
alternative is a spinlock/unlock in wake_up_interruptible, which I
assume is necessarily more expensive than an smp_mb().

As far as I can tell it's been this way since forever.  (Well, since a
2002 patch "NFSD: TCP: rationalise locking in RPC server routines" which
removed some spinlocks from the data_ready routines.)

I don't understand what the actual race is yet (which code exactly is
missing the wakeup in this case?  nfsd threads seem to instead get
woken up by the wake_up_process() in svc_xprt_do_enqueue().)

--b.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Kosuke Tatsukawa]

There are several places in net/sunrpc/svcsock.c which calls
waitqueue_active() without calling a memory barrier.  Add a memory
barrier just as in wq_has_sleeper().

I found this issue when I was looking through the linux source code
for places calling waitqueue_active() before wake_up*(), but without
preceding memory barriers, after sending a patch to fix a similar
issue in drivers/tty/n_tty.c  (Details about the original issue can be
found here: https://lkml.org/lkml/2015/9/28/849).

Signed-off-by: Kosuke Tatsukawa 
---
v2:
  - Fixed compiler warnings caused by type mismatch
v1:
  - https://lkml.org/lkml/2015/10/8/993
---
 net/sunrpc/svcsock.c |6 ++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 0c81202..ec19444 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -414,6 +414,7 @@ static void svc_udp_data_ready(struct sock *sk)
set_bit(XPT_DATA, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
+   smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
@@ -432,6 +433,7 @@ static void svc_write_space(struct sock *sk)
svc_xprt_enqueue(>sk_xprt);
}
 
+   smp_mb();
if (wq && waitqueue_active(wq)) {
dprintk("RPC svc_write_space: someone sleeping on %p\n",
   svsk);
@@ -787,6 +789,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
}
 
wq = sk_sleep(sk);
+   smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible_all(wq);
 }
@@ -808,6 +811,7 @@ static void svc_tcp_state_change(struct sock *sk)
set_bit(XPT_CLOSE, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
+   smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible_all(wq);
 }
@@ -823,6 +827,7 @@ static void svc_tcp_data_ready(struct sock *sk)
set_bit(XPT_DATA, >sk_xprt.xpt_flags);
svc_xprt_enqueue(>sk_xprt);
}
+   smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
@@ -1594,6 +1599,7 @@ static void svc_sock_detach(struct svc_xprt *xprt)
sk->sk_write_space = svsk->sk_owspace;
 
wq = sk_sleep(sk);
+   smp_mb();
if (wq && waitqueue_active(wq))
wake_up_interruptible(wq);
 }
-- 
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] sunrpc: fix waitqueue_active without memory barrier in sunrpc

[Neil Brown]

Kosuke Tatsukawa  writes:

> There are several places in net/sunrpc/svcsock.c which calls
> waitqueue_active() without calling a memory barrier.  Add a memory
> barrier just as in wq_has_sleeper().
>
> I found this issue when I was looking through the linux source code
> for places calling waitqueue_active() before wake_up*(), but without
> preceding memory barriers, after sending a patch to fix a similar
> issue in drivers/tty/n_tty.c  (Details about the original issue can be
> found here: https://lkml.org/lkml/2015/9/28/849).

hi,
this feels like the wrong approach to the problem.  It requires extra
'smb_mb's to be spread around which are hard to understand as easy to
forget.

A quick look seems to suggest that (nearly) every waitqueue_active()
will need an smb_mb.  Could we just put the smb_mb() inside
waitqueue_active()??

Thanks,
NeilBrown


>
> Signed-off-by: Kosuke Tatsukawa 
> ---
> v2:
>   - Fixed compiler warnings caused by type mismatch
> v1:
>   - https://lkml.org/lkml/2015/10/8/993
> ---
>  net/sunrpc/svcsock.c |6 ++
>  1 files changed, 6 insertions(+), 0 deletions(-)
>
> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> index 0c81202..ec19444 100644
> --- a/net/sunrpc/svcsock.c
> +++ b/net/sunrpc/svcsock.c
> @@ -414,6 +414,7 @@ static void svc_udp_data_ready(struct sock *sk)
>   set_bit(XPT_DATA, >sk_xprt.xpt_flags);
>   svc_xprt_enqueue(>sk_xprt);
>   }
> + smp_mb();
>   if (wq && waitqueue_active(wq))
>   wake_up_interruptible(wq);
>  }


signature.asc
Description: PGP signature