Skip to content
  1. Mar 01, 2017
    • Eric Dumazet's avatar
      net: solve a NAPI race · 39e6c820
      Eric Dumazet authored 
      

While playing with mlx4 hardware timestamping of RX packets, I found
that some packets were received by TCP stack with a ~200 ms delay...

Since the timestamp was provided by the NIC, and my probe was added
in tcp_v4_rcv() while in BH handler, I was confident it was not
a sender issue, or a drop in the network.

This would happen with a very low probability, but hurting RPC
workloads.

A NAPI driver normally arms the IRQ after the napi_complete_done(),
after NAPI_STATE_SCHED is cleared, so that the hard irq handler can grab
it.

Problem is that if another point in the stack grabs NAPI_STATE_SCHED bit
while IRQ are not disabled, we might have later an IRQ firing and
finding this bit set, right before napi_complete_done() clears it.

This can happen with busy polling users, or if gro_flush_timeout is
used. But some other uses of napi_schedule() in drivers can cause this
as well.

thread 1                                 thread 2 (could be on same cpu, or not)

// busy polling or napi_watchdog()
napi_schedule();
...
napi->poll()

device polling:
read 2 packets from ring buffer
                                          Additional 3rd packet is
available.
                                          device hard irq

                                          // does nothing because
NAPI_STATE_SCHED bit is owned by thread 1
                                          napi_schedule();

napi_complete_done(napi, 2);
rearm_irq();

Note that rearm_irq() will not force the device to send an additional
IRQ for the packet it already signaled (3rd packet in my example)

This patch adds a new NAPI_STATE_MISSED bit, that napi_schedule_prep()
can set if it could not grab NAPI_STATE_SCHED

Then napi_complete_done() properly reschedules the napi to make sure
we do not miss something.

Since we manipulate multiple bits at once, use cmpxchg() like in
sk_busy_loop() to provide proper transactions.

In v2, I changed napi_watchdog() to use a relaxed variant of
napi_schedule_prep() : No need to set NAPI_STATE_MISSED from this point.

In v3, I added more details in the changelog and clears
NAPI_STATE_MISSED in busy_poll_stop()

In v4, I added the ideas given by Alexander Duyck in v3 review

Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Cc: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      39e6c820
    • Dan Carpenter's avatar
      net/mlx4: && vs & typo · b2d0fe35
      Dan Carpenter authored 
      

Bitwise & was obviously intended here.

Fixes: 745d8ae4 ("net/mlx4: Spoofcheck and zero MAC can't coexist")
Signed-off-by: default avatarDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: default avatarTariq Toukan <tariqt@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      b2d0fe35
    • David Howells's avatar
      rxrpc: Fix deadlock between call creation and sendmsg/recvmsg · 540b1c48
      David Howells authored 
      

All the routines by which rxrpc is accessed from the outside are serialised
by means of the socket lock (sendmsg, recvmsg, bind,
rxrpc_kernel_begin_call(), ...) and this presents a problem:

 (1) If a number of calls on the same socket are in the process of
     connection to the same peer, a maximum of four concurrent live calls
     are permitted before further calls need to wait for a slot.

 (2) If a call is waiting for a slot, it is deep inside sendmsg() or
     rxrpc_kernel_begin_call() and the entry function is holding the socket
     lock.

 (3) sendmsg() and recvmsg() or the in-kernel equivalents are prevented
     from servicing the other calls as they need to take the socket lock to
     do so.

 (4) The socket is stuck until a call is aborted and makes its slot
     available to the waiter.

Fix this by:

 (1) Provide each call with a mutex ('user_mutex') that arbitrates access
     by the users of rxrpc separately for each specific call.

 (2) Make rxrpc_sendmsg() and rxrpc_recvmsg() unlock the socket as soon as
     they've got a call and taken its mutex.

     Note that I'm returning EWOULDBLOCK from recvmsg() if MSG_DONTWAIT is
     set but someone else has the lock.  Should I instead only return
     EWOULDBLOCK if there's nothing currently to be done on a socket, and
     sleep in this particular instance because there is something to be
     done, but we appear to be blocked by the interrupt handler doing its
     ping?

 (3) Make rxrpc_new_client_call() unlock the socket after allocating a new
     call, locking its user mutex and adding it to the socket's call tree.
     The call is returned locked so that sendmsg() can add data to it
     immediately.

     From the moment the call is in the socket tree, it is subject to
     access by sendmsg() and recvmsg() - even if it isn't connected yet.

 (4) Lock new service calls in the UDP data_ready handler (in
     rxrpc_new_incoming_call()) because they may already be in the socket's
     tree and the data_ready handler makes them live immediately if a user
     ID has already been preassigned.

     Note that the new call is locked before any notifications are sent
     that it is live, so doing mutex_trylock() *ought* to always succeed.
     Userspace is prevented from doing sendmsg() on calls that are in a
     too-early state in rxrpc_do_sendmsg().

 (5) Make rxrpc_new_incoming_call() return the call with the user mutex
     held so that a ping can be scheduled immediately under it.

     Note that it might be worth moving the ping call into
     rxrpc_new_incoming_call() and then we can drop the mutex there.

 (6) Make rxrpc_accept_call() take the lock on the call it is accepting and
     release the socket after adding the call to the socket's tree.  This
     is slightly tricky as we've dequeued the call by that point and have
     to requeue it.

     Note that requeuing emits a trace event.

 (7) Make rxrpc_kernel_send_data() and rxrpc_kernel_recv_data() take the
     new mutex immediately and don't bother with the socket mutex at all.

This patch has the nice bonus that calls on the same socket are now to some
extent parallelisable.

Note that we might want to move rxrpc_service_prealloc() calls out from the
socket lock and give it its own lock, so that we don't hang progress in
other calls because we're waiting for the allocator.

We probably also want to avoid calling rxrpc_notify_socket() from within
the socket lock (rxrpc_accept_call()).

Signed-off-by: default avatarDavid Howells <dhowells@redhat.com>
Tested-by: default avatarMarc Dionne <marc.c.dionne@auristor.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      540b1c48
  3. Feb 28, 2017
  5. Feb 25, 2017