Since commit ac4e4af1e5 ("macvtap: Consistently use rcu functions"),
Thomas gets two different warnings :
BUG: using smp_processor_id() in preemptible [00000000] code: vhost-45891/45892
caller is macvtap_do_read+0x45c/0x600 [macvtap]
CPU: 1 PID: 45892 Comm: vhost-45891 Not tainted 3.11.0-bisecttest #13
Call Trace:
([<00000000001126ee>] show_trace+0x126/0x144)
[<00000000001127d2>] show_stack+0xc6/0xd4
[<000000000068bcec>] dump_stack+0x74/0xd8
[<0000000000481066>] debug_smp_processor_id+0xf6/0x114
[<000003ff802e9a18>] macvtap_do_read+0x45c/0x600 [macvtap]
[<000003ff802e9c1c>] macvtap_recvmsg+0x60/0x88 [macvtap]
[<000003ff80318c5e>] handle_rx+0x5b2/0x800 [vhost_net]
[<000003ff8028f77c>] vhost_worker+0x15c/0x1c4 [vhost]
[<000000000015f3ac>] kthread+0xd8/0xe4
[<00000000006934a6>] kernel_thread_starter+0x6/0xc
[<00000000006934a0>] kernel_thread_starter+0x0/0xc
And
BUG: using smp_processor_id() in preemptible [00000000] code: vhost-45897/45898
caller is macvlan_start_xmit+0x10a/0x1b4 [macvlan]
CPU: 1 PID: 45898 Comm: vhost-45897 Not tainted 3.11.0-bisecttest #16
Call Trace:
([<00000000001126ee>] show_trace+0x126/0x144)
[<00000000001127d2>] show_stack+0xc6/0xd4
[<000000000068bdb8>] dump_stack+0x74/0xd4
[<0000000000481132>] debug_smp_processor_id+0xf6/0x114
[<000003ff802b72ca>] macvlan_start_xmit+0x10a/0x1b4 [macvlan]
[<000003ff802ea69a>] macvtap_get_user+0x982/0xbc4 [macvtap]
[<000003ff802ea92a>] macvtap_sendmsg+0x4e/0x60 [macvtap]
[<000003ff8031947c>] handle_tx+0x494/0x5ec [vhost_net]
[<000003ff8028f77c>] vhost_worker+0x15c/0x1c4 [vhost]
[<000000000015f3ac>] kthread+0xd8/0xe4
[<000000000069356e>] kernel_thread_starter+0x6/0xc
[<0000000000693568>] kernel_thread_starter+0x0/0xc
2 locks held by vhost-45897/45898:
#0: (&vq->mutex){+.+.+.}, at: [<000003ff8031903c>] handle_tx+0x54/0x5ec [vhost_net]
#1: (rcu_read_lock){.+.+..}, at: [<000003ff802ea53c>] macvtap_get_user+0x824/0xbc4 [macvtap]
In the first case, macvtap_put_user() calls macvlan_count_rx()
in a preempt-able context, and this is not allowed.
In the second case, macvtap_get_user() calls
macvlan_start_xmit() with BH enabled, and this is not allowed.
Reported-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Bisected-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Cc: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
network.
Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than
MAX_SKB_FRAGS.
This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify
vhost.
We can do further optimization on top.
This bug were introduced from b92946e291
(macvtap: zerocopy: validate vectors before building skb).
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 441ac0fcaa
(macvtap: Convert to using rtnl lock) forget to return what
macvtap_ioctl_set_queue() returns to its caller. This may break multiqueue API
by always falling through to TUNGETFEATURES.
Cc: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Userspace may produce vectors greater than MAX_SKB_FRAGS. When we try to
linearize parts of the skb to let the rest of iov to be fit in
the frags, we need count copylen into linear when calling macvtap_alloc_skb()
instead of partly counting it into data_len. Since this breaks
zerocopy_sg_from_iovec() since its inner counter assumes nr_frags should
be zero at beginning. This cause nr_frags to be increased wrongly without
setting the correct frags.
This bug were introduced from b92946e291
(macvtap: zerocopy: validate vectors before building skb).
Cc: Michael S. Tsirkin <mst@redhat.com>
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>
Conflicts:
drivers/net/ethernet/freescale/fec_main.c
drivers/net/ethernet/renesas/sh_eth.c
net/ipv4/gre.c
The GRE conflict is between a bug fix (kfree_skb --> kfree_skb_list)
and the splitting of the gre.c code into seperate files.
The FEC conflict was two sets of changes adding ethtool support code
in an "!CONFIG_M5272" CPP protected block.
Finally the sh_eth.c conflict was between one commit add bits set
in the .eesr_err_check mask whilst another commit removed the
.tx_error_check member and assignments.
Signed-off-by: David S. Miller <davem@davemloft.net>
When macvtap forwards skb to its tap, it needs to check
if GSO needs to be performed. This is sometimes necessary
when the HW device performed GRO, but the guest reading
from the tap does not support it (ex: Windows 7).
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the user issues TUNSETOFFLOAD ioctl, macvtap does not do
anything other then to verify arguments. This patch adds
functionality to allow users to actually control offload features.
NETIF_F_GSO and NETIF_F_GRO are always on, but the rest of the
features can be controlled.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently macvtap uses rcu_bh functions in its
user facing fuction macvtap_get_user() and macvtap_put_user().
However, its packet handlers use normal rcu as the rcu_read_lock()
is taken in netif_receive_skb(). We can safely discontinue
the usage or rcu with bh disabled.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Macvtap uses a private lock to protect the relationship between
macvtap_queue and macvlan_dev. The private lock is not needed
since the relationship is managed by user via open(), release(),
and dellink() calls. dellink() already happens under rtnl, so
we can safely convert open() and release(), and use it in ioctl()
as well.
Suggested by Eric Dumazet.
Signed-off-by: Vlad Yasevich <vyasevic@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
get user pages might fail partially in macvtap zero copy
mode. To recover we need to put all pages that we got,
but code used a wrong index resulting in double-free
errors.
Reported-by: Brad Hubbard <bhubbard@redhat.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>
Return -EINVAL on illegal flag instead of uninitialized value. This fixes the
kbuild test warning.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds TUNSETQUEUE ioctl to let userspace can temporarily disable or
enable a queue of macvtap. This is used to be compatible at API layer of tuntap
to simplify the userspace to manage the queues. This is done through introducing
a linked list to track all taps while using vlan->taps array to only track
active taps.
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>
Linear search were used in both get_slot() and macvtap_get_queue(), this is
because:
- macvtap didn't reshuffle the array of taps when create or destroy a queue, so
when adding a new queue, macvtap must do linear search to find a location for
the new queue. This will also complicate the TUNSETQUEUE implementation for
multiqueue API.
- the queue itself didn't track the queue index, so the we must do a linear
search in the array to find the location of a existed queue.
The solution is straightforward: reshuffle the array and introduce a queue_index
to macvtap_queue.
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>
Factor out the device holding logic to a macvtap_get_vlan(), this will be also
used by multiqueue API.
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>
There's no need to add self to waitqueue if doing a nonblock read. This could
help to avoid the spinlock contention.
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Complier may generate codes that re-read the vlan->numvtaps during
macvtap_get_queue(). This may lead a race if vlan->numvtaps were changed in the
same time and which can lead unexpected result (e.g. very huge value).
We need prevent the compiler from generating such codes by adding an
ACCESS_ONCE() to make sure vlan->numvtaps were only read once.
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So far, only net_device * could be passed along with netdevice notifier
event. This patch provides a possibility to pass custom structure
able to provide info that event listener needs to know.
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
v2->v3: fix typo on simeth
shortened dev_getter
shortened notifier_info struct name
v1->v2: fix notifier_call parameter in call_netdevice_notifier()
Signed-off-by: David S. Miller <davem@davemloft.net>
Switch to use the new help skb_probe_transport_header() to do the l4 header
probing for untrusted sources. For packets with partial csum, the header should
already been set by skb_partial_csum_set().
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set the transport header for 1) some drivers (e.g ixgbe) needs l4 header 2)
precise packet length estimation (introduced in 1def9238) needs l4 header to
compute header length.
For the packets with partial checksum, the patch just set the transport header
to csum_start. Otherwise tries to use skb_flow_dissect() to get l4 offset, if it
fails, just pretend no l4 header.
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Patch cef401de7b (net: fix possible wrong checksum
generation) fixed wrong checksum calculation but it broke TSO by
defining new GSO type but not a netdev feature for that type.
net_gso_ok() would not allow hardware checksum/segmentation
offload of such packets without the feature.
Following patch fixes TSO and wrong checksum. This patch uses
same logic that Eric Dumazet used. Patch introduces new flag
SKBTX_SHARED_FRAG if at least one frag can be modified by
the user. but SKBTX_SHARED_FRAG flag is kept in skb shared
info tx_flags rather than gso_type.
tx_flags is better compared to gso_type since we can have skb with
shared frag without gso packet. It does not link SHARED_FRAG to
GSO, So there is no need to define netdev feature for this.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pravin Shelar mentioned that GSO could potentially generate
wrong TX checksum if skb has fragments that are overwritten
by the user between the checksum computation and transmit.
He suggested to linearize skbs but this extra copy can be
avoided for normal tcp skbs cooked by tcp_sendmsg().
This patch introduces a new SKB_GSO_SHARED_FRAG flag, set
in skb_shinfo(skb)->gso_type if at least one frag can be
modified by the user.
Typical sources of such possible overwrites are {vm}splice(),
sendfile(), and macvtap/tun/virtio_net drivers.
Tested:
$ netperf -H 7.7.8.84
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
7.7.8.84 () port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 16384 16384 10.00 3959.52
$ netperf -H 7.7.8.84 -t TCP_SENDFILE
TCP SENDFILE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.8.84 ()
port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 16384 16384 10.00 3216.80
Performance of the SENDFILE is impacted by the extra allocation and
copy, and because we use order-0 pages, while the TCP_STREAM uses
bigger pages.
Reported-by: Pravin Shelar <pshelar@nicira.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet
Jason Wang
David S. Miller
Vlad Yasevich
Michael S. Tsirkin
Jiri Pirko
Tejun Heo
Pravin B Shelar