vhost_zerocopy_callback accesses VQ right after it drops a ubuf
reference. In theory, this could race with device removal which waits
on the ubuf kref, and crash on use after free.
Do all accesses within rcu read side critical section, and synchronize
on release.
Since callbacks are always invoked from bh, synchronize_rcu_bh seems
enough and will help release complete a bit faster.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
vhost checked the counter within the refcnt before decrementing. It
really wanted to know that it is the one that has the last reference, as
a way to batch freeing resources a bit more efficiently.
Note: we only let refcount go to 0 on device release.
This works well but we now access the ref counter twice so there's a
race: all users might see a high count and decide to defer freeing
resources.
In the end no one initiates freeing resources until the last reference
is gone (which is on VM shotdown so might happen after a looooong time).
Let's do what we probably should have done straight away:
switch from kref to plain atomic, documenting the
semantics, return the refcount value atomically after decrement,
then use that to avoid the deadlock.
Reported-by: Qin Chuanyu <qinchuanyu@huawei.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since vhost_dev_init() forever return 0, some branches are never run,
therefore need to be removed.
Signed-off-by: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As Michael point out, We used to limit the max pending DMAs to get better cache
utilization. But it was not done correctly since it was one done when there's no
new buffers submitted from guest. Guest can easily exceeds the limitation by
keeping sending packets.
So this patch moves the check into main loop. Tests shows about 5%-10%
improvement on per cpu throughput for guest tx.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We used to poll vhost queue before making DMA is done, this is racy if vhost
thread were waked up before marking DMA is done which can result the signal to
be missed. Fix this by always polling the vhost thread before DMA is done.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, even if the packet length is smaller than VHOST_GOODCOPY_LEN, if
upend_idx != done_idx we still set zcopy_used to true and rollback this choice
later. This could be avoided by determining zerocopy once by checking all
conditions at one time before.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We tend to batch the used adding and signaling in vhost_zerocopy_callback()
which may result more than 100 used buffers to be updated in
vhost_zerocopy_signal_used() in some cases. So switch to use
vhost_add_used_and_signal_n() to avoid multiple calls to
vhost_add_used_and_signal(). Which means much less times of used index
updating and memory barriers.
2% performance improvement were seen on netperf TCP_RR test.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
None of its caller use its return value, so let it return void.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now, vq->private_data is always accessed under vq mutex. No need to play
the vhost rcu trick.
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
$ make C=1 M=drivers/vhost
drivers/vhost/net.c:168:5: warning: symbol 'vhost_net_set_ubuf_info' was not declared. Should it be static?
drivers/vhost/net.c:194:6: warning: symbol 'vhost_net_vq_reset' was not declared. Should it be static?
drivers/vhost/scsi.c:219:6: warning: symbol 'tcm_vhost_done_inflight' was not declared. Should it be static?
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost_net_ubuf_put_and_wait has a confusing name:
it will actually also free it's argument.
Thus since commit 1280c27f8e
"vhost-net: flush outstanding DMAs on memory change"
vhost_net_flush tries to use the argument after passing it
to vhost_net_ubuf_put_and_wait, this results
in use after free.
To fix, don't free the argument in vhost_net_ubuf_put_and_wait,
add an new API for callers that want to free ubufs.
Acked-by: Asias He <asias@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost_net_clear_ubuf_info didn't clear ubuf_info
after kfree, this could trigger double free.
Fix this and simplify this code to make it more robust: make sure
ubuf info is always freed through vhost_net_clear_ubuf_info.
Reported-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If device has an owner, we shouldn't touch ubuf_info
since it might be in use.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we decide not use zero-copy, msg.control should be set to NULL otherwise
macvtap/tap may set zerocopy callbacks which may decrease the kref of ubufs
wrongly.
Bug were introduced by commit cedb9bdce0
(vhost-net: skip head management if no outstanding).
This solves the following warnings:
WARNING: at include/linux/kref.h:47 handle_tx+0x477/0x4b0 [vhost_net]()
Modules linked in: vhost_net macvtap macvlan tun nfsd exportfs bridge stp llc openvswitch kvm_amd kvm bnx2 megaraid_sas [last unloaded: tun]
CPU: 5 PID: 8670 Comm: vhost-8668 Not tainted 3.10.0-rc2+ #1566
Hardware name: Dell Inc. PowerEdge R715/00XHKG, BIOS 1.5.2 04/19/2011
ffffffffa0198323 ffff88007c9ebd08 ffffffff81796b73 ffff88007c9ebd48
ffffffff8103d66b 000000007b773e20 ffff8800779f0000 ffff8800779f43f0
ffff8800779f8418 000000000000015c 0000000000000062 ffff88007c9ebd58
Call Trace:
[<ffffffff81796b73>] dump_stack+0x19/0x1e
[<ffffffff8103d66b>] warn_slowpath_common+0x6b/0xa0
[<ffffffff8103d6b5>] warn_slowpath_null+0x15/0x20
[<ffffffffa0197627>] handle_tx+0x477/0x4b0 [vhost_net]
[<ffffffffa0197690>] handle_tx_kick+0x10/0x20 [vhost_net]
[<ffffffffa019541e>] vhost_worker+0xfe/0x1a0 [vhost_net]
[<ffffffffa0195320>] ? vhost_attach_cgroups_work+0x30/0x30 [vhost_net]
[<ffffffffa0195320>] ? vhost_attach_cgroups_work+0x30/0x30 [vhost_net]
[<ffffffff81061f46>] kthread+0xc6/0xd0
[<ffffffff81061e80>] ? kthread_freezable_should_stop+0x70/0x70
[<ffffffff817a1aec>] ret_from_fork+0x7c/0xb0
[<ffffffff81061e80>] ? kthread_freezable_should_stop+0x70/0x70
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
- Rename vhost_ubuf to vhost_net_ubuf
- Rename vhost_zcopy_mask to vhost_net_zcopy_mask
- Make funcs static
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost.h should not depend on device specific marcos like
VHOST_NET_F_VIRTIO_NET_HDR and VIRTIO_NET_F_MRG_RXBUF.
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
RESET_OWNER ioctl would leave the fd in a bad state if
memory allocation failed: device is stopped
but owner is not reset. Make state changes
after allocating memory, such that a failed
ioctl has no effect.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
On top of 'vhost: Allow device specific fields per vq', we can move device
specific fields to device virt queue from vhost virt queue.
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This is useful for any device who wants device specific fields per vq.
For example, tcm_vhost wants a per vq field to track requests which are
in flight on the vq. Also, on top of this we can add patches to move
things like ubufs from vhost.h out to net.c.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
After commit 2b8b328b61 (vhost_net: handle polling
errors when setting backend), we in fact track the polling state through
poll->wqh, so there's no need to duplicate the work with an extra
vhost_net_polling_state. So this patch removes this and make the code simpler.
This patch also removes the all tx starting/stopping code in tx path according
to Michael's suggestion.
Netperf test shows almost the same result in stream test, but gets improvements
on TCP_RR tests (both zerocopy or copy) especially on low load cases.
Tested between multiqueue kvm guest and external host with two direct
connected 82599s.
zerocopy disabled:
sessions|transaction rates|normalize|
before/after/+improvements
1 | 9510.24/11727.29/+23.3% | 693.54/887.68/+28.0% |
25| 192931.50/241729.87/+25.3% | 2376.80/2771.70/+16.6% |
50| 277634.64/291905.76/+5% | 3118.36/3230.11/+3.6% |
zerocopy enabled:
sessions|transaction rates|normalize|
before/after/+improvements
1 | 7318.33/11929.76/+63.0% | 521.86/843.30/+61.6% |
25| 167264.88/242422.15/+44.9% | 2181.60/2788.16/+27.8% |
50| 272181.02/294347.04/+8.1% | 3071.56/3257.85/+6.1% |
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ubuf info allocator uses guest controlled head as an index,
so a malicious guest could put the same head entry in the ring twice,
and we will get two callbacks on the same value.
To fix use upend_idx which is guaranteed to be unique.
Reported-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Cc: stable@kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the polling errors were ignored, which can lead following issues:
- vhost remove itself unconditionally from waitqueue when stopping the poll,
this may crash the kernel since the previous attempt of starting may fail to
add itself to the waitqueue
- userspace may think the backend were successfully set even when the polling
failed.
Solve this by:
- check poll->wqh before trying to remove from waitqueue
- report polling errors in vhost_poll_start(), tx_poll_start(), the return value
will be checked and returned when userspace want to set the backend
After this fix, there still could be a polling failure after backend is set, it
will addressed by the next patch.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, when vhost_init_used() fails the sock refcnt and ubufs were
leaked. Correct this by calling vhost_init_used() before assign ubufs and
restore the oldsock when it fails.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Zero copy TX has been around for a while now.
We seem to be down to eliminating theoretical bugs
and performance tuning at this point:
it's probably time to enable it by default so that
most users get the benefit.
Keep the flag around meanwhile so users can experiment
with disabling this if they experience regressions.
I expect that we will remove it in the future.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
For short packets zerocopy mode adds overhead
of managing heads which isn't necessary: we
could simly update used ring directly
same as with zerocopy disabled.
Things seem to run a bit faster if we detect
and bypass head management when zcopy isn't used.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When memory map changes, we need to flush outstanding
DMAs as they might in theory reference old memory addresses.
To do this simply stop initiating new DMAs
and wait for ubufs ref count to drop to 0.
Afterwards reset the count back to 1.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vring changes already do a flush internally where appropriate, so we do
not need a second flush.
It's currently not very expensive but a follow-up patch makes flush more
heavy-weight, so remove the extra flush here to avoid regressing
performance if call or kick fds are changed on data path.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
These packet counters are used to drive the zercopy
selection heuristic so nothing too bad happens if they are off a bit -
and they are also reset once in a while.
But it's cleaner to clear them when backend is set so that
we start in a known state.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
Minor conflict between the BCM_CNIC define removal in net-next
and a bug fix added to net. Based upon a conflict resolution
patch posted by Stephen Rothwell.
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems that to avoid deadlocks it is enough to poll vq before
we are going to use the last buffer. This is faster than
c70aa540c7.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even when vhost-net is in zero-copy transmit mode,
net core might still decide to copy the skb later
which is somewhat slower than a copy in user
context: data copy overhead is added to the cost of
page pin/unpin. The result is that enabling tx zero copy
option leads to higher CPU utilization for guest to guest
and guest to host traffic.
To fix this, suppress zero copy tx after a given number of
packets triggered late data copy. Re-enable periodically
to detect workload changes.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Zerocopy handling code is vhost-net specific.
Move it from vhost.c/vhost.h out to net.c
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Better document macros for DMA tracking. Add an
explicit one for DMA in progress instead of
relying on user supplying len != 1.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We copy head count to a 16 bit field, this works by chance on LE but on
BE guest gets 0. Fix it up.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Alexander Graf <agraf@suse.de>
Cc: stable@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
In order for other vhost devices to use the VHOST_FEATURES bits the
vhost-net specific bits need to be moved to their own VHOST_NET_FEATURES
constant.
(Asias: Update drivers/vhost/test.c to use VHOST_NET_FEATURES)
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Cc: Zhi Yong Wu <wuzhy@cn.ibm.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Asias He <asias@redhat.com>
Signed-off-by: Nicholas A. Bellinger <nab@risingtidesystems.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Take vlan header length into account, when vlan id is stored as
vlan_tci. Otherwise tagged packets coming from macvtap will be
truncated.
Signed-off-by: Basil Gor <basil.gor@gmail.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a packet were fully copied in zerocopy, we don't wait for the DMA done to
mark the done flag, so after the packet were passed to lower device, we need to
add used and signal guest immediately.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Currently, we restart tx polling unconditionally when sendmsg()
fails. This would cause unnecessary wakeups of vhost wokers and waste
cpu utlization when evil userspace(guest driver) is able to hit EFAULT or
EINVAL.
The polling is only needed when the socket send buffer were exceeded or not
enough memory. So fix this by restarting polling only when sendmsg() returns
EAGAIN/ENOBUFS.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When we want to disable vhost_net backend while there's a tx work, a possible
NULL pointer defernece may happen we we try to deference the vq->bufs after
vhost_net_set_backend() assign a NULL to it.
As suggested by Michael, fix this by checking the vq->bufs instead of
vhost_sock_zcopy().
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The skb struct ubuf_info callback gets passed struct ubuf_info
itself, not the arg value as the field name and the function signature
seem to imply. Rename the arg field to ctx to match usage,
add documentation and change the callback argument type
to make usage clear and to have compiler check correctness.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We shouldn't hold any locks on release path. Pass a flag to
vhost_dev_cleanup to use the lockdep info correctly.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Sasha Levin <levinsasha928@gmail.com>
By adding some module aliases, programs (or users) won't have to explicitly
call modprobe. Vhost-net will always be available if built into the kernel.
It does require assigning a permanent minor number for depmod to work.
Also:
- use C99 style initialization.
- add missing entry in documentation for loop-control
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-By: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The meth for calculating the # of outstanding buffers gives
incorrect results when vq->upend_idx wraps around zero.
Fix that.
Signed-off-by: Shirley Ma <xma@us.ibm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
On backend change, we flushed out outstanding skbs
but forgot to update the used ring, so that
done entries were left in the ubuf_info ring.
As a result we lose heads or complete incorrect ones,
crashing the guest or leaking memory.
Fix by updating the used ring.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Move the used ring initialization after backend was set. This
makes it possible to disable the backend and tweak the used ring,
then restart. This will also make it possible to log the used ring
write correctly.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
>From: Shirley Ma <mashirle@us.ibm.com>
This adds experimental zero copy support in vhost-net,
disabled by default. To enable, set
experimental_zcopytx module option to 1.
This patch maintains the outstanding userspace buffers in the
sequence it is delivered to vhost. The outstanding userspace buffers
will be marked as done once the lower device buffers DMA has finished.
This is monitored through last reference of kfree_skb callback. Two
buffer indices are used for this purpose.
The vhost-net device passes the userspace buffers info to lower device
skb through message control. DMA done status check and guest
notification are handled by handle_tx: in the worst case is all buffers
in the vq are in pending/done status, so we need to notify guest to
release DMA done buffers first before we get any new buffers from the
vq.
One known problem is that if the guest stops submitting
buffers, buffers might never get used until some
further action, e.g. device reset. This does not
seem to affect linux guests.
Signed-off-by: Shirley <xma@us.ibm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Michael S. Tsirkin
Zhi Yong Wu
Jason Wang
Asias He
David S. Miller
Stefan Hajnoczi
Basil Gor
stephen hemminger
Shirley Ma