This patch split out XDP logic into a single function. This make it to
be reused by XDP batching path in the following patch.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we're sure not to go native XDP, there's no need for several things
like bh and rcu stuffs. So this patch introduces a helper to build skb
and hold page refcnt. When we found we will go through skb path, build
skb directly.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There's no need to duplicate page get logic in each action. So this
patch tries to get page and calculate the offset before processing XDP
actions (except for XDP_DROP), and undo them when meet errors (we
don't care the performance on errors). This will be used for factoring
out XDP logic.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch move the bh enabling a little bit earlier, this will be
used for factoring out the core XDP logic of tuntap.
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch introduces a new sock flag - SOCK_XDP. This will be used
for notifying the upper layer that XDP program is attached on the
lower socket, and requires for extra headroom.

TUN will be the first user.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

0x3ff in tun_hashfn is mask of TUN_NUM_FLOW_ENTRIES, instead
of hardcode, define a macro to setup the relationship with
TUN_NUM_FLOW_ENTRIES
Signed-off-by: NLi RongQing <lirongqing@baidu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When f_setown is called a pid and a pid type are stored.  Replace the use
of PIDTYPE_PID with PIDTYPE_TGID as PIDTYPE_TGID goes to the entire thread
group.  Replace the use of PIDTYPE_MAX with PIDTYPE_PID as PIDTYPE_PID now
is only for a thread.

Update the users of __f_setown to use PIDTYPE_TGID instead of
PIDTYPE_PID.

For now the code continues to capture task_pid (when task_tgid would
really be appropriate), and iterate on PIDTYPE_PID (even when type ==
PIDTYPE_TGID) out of an abundance of caution to preserve existing
behavior.

Oleg Nesterov suggested using the test to ensure we use PIDTYPE_PID
for tgid lookup also be used to avoid taking the tasklist lock.
Suggested-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>

On XDP_TX we need to free up the frame only when tun_xdp_tx() returns a
negative value. A positive value indicates that the packet is
successfully enqueued to the ptr_ring, so freeing the page causes
use-after-free.

Fixes: 735fc405 ("xdp: change ndo_xdp_xmit API to support bulking")
Signed-off-by: NToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Acked-by: NJason Wang <jasowang@redhat.com>
Acked-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

prog_attached of struct netdev_bpf should have been superseded
by simply setting prog_id long time ago, but we kept it around
to allow offloading drivers to communicate attachment mode (drv
vs hw).  Subsequently drivers were also allowed to report back
attachment flags (prog_flags), and since nowadays only programs
attached will XDP_FLAGS_HW_MODE can get offloaded, we can tell
the attachment mode from the flags driver reports.  Remove
prog_attached member.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NQuentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>

This patch makes it so that instead of passing a void pointer as the
accel_priv we instead pass a net_device pointer as sb_dev. Making this
change allows us to pass the subordinate device through to the fallback
function eventually so that we can keep the actual code in the
ndo_select_queue call as focused on possible on the exception cases.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Tun, tap, virtio, packet and uml vector all use struct virtio_net_hdr
to communicate packet metadata to userspace.

For skbuffs with vlan, the first two return the packet as it may have
existed on the wire, inserting the VLAN tag in the user buffer.  Then
virtio_net_hdr.csum_start needs to be adjusted by VLAN_HLEN bytes.

Commit f09e2249 ("macvtap: restore vlan header on user read")
added this feature to macvtap. Commit 3ce9b20f ("macvtap: Fix
csum_start when VLAN tags are present") then fixed up csum_start.

Virtio, packet and uml do not insert the vlan header in the user
buffer.

When introducing virtio_net_hdr_from_skb to deduplicate filling in
the virtio_net_hdr, the variant from macvtap which adds VLAN_HLEN was
applied uniformly, breaking csum offset for packets with vlan on
virtio and packet.

Make insertion of VLAN_HLEN optional. Convert the callers to pass it
when needed.

Fixes: e858fae2 ("virtio_net: use common code for virtio_net_hdr and skb GSO conversion")
Fixes: 1276f24e ("packet: use common code for virtio_net_hdr and skb GSO conversion")
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Remove the ndo_xdp_flush call implementation tun_xdp_flush
as no callers of ndo_xdp_flush are left.

The tun drivers XDP_TX implementation also used tun_xdp_flush (and
tun_xdp_xmit).  This is easily solved by passing the XDP_XMIT_FLUSH
flag to tun_xdp_xmit in tun_xdp_tx.
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>

Let user space set whatever it would like to advertise for the
tun interface.  Preserve the existing defaults.
Signed-off-by: NChas Williams <3chas3@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When passed the XDP_XMIT_FLUSH flag tun_xdp_xmit now performs the same
kind of socket wake up as in tun_xdp_flush(). The wake up code from
tun_xdp_flush is generalized and shared with tun_xdp_xmit.
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Acked-by: NSong Liu <songliubraving@fb.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>

This patch only change the API and reject any use of flags. This is an
intermediate step that allows us to implement the flush flag operation
later, for each individual driver in a separate patch.

The plan is to implement flush operation via XDP_XMIT_FLUSH flag
and then remove XDP_XMIT_FLAGS_NONE when done.
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Acked-by: NSong Liu <songliubraving@fb.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>

Calling XDP redirection requires bh disabled. Softirq can call another
XDP function and redirection functions, then the percpu static variable
ri->map can be overwritten to NULL.

This is a generic XDP case called from tun.

[ 3535.736058] BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
[ 3535.743974] PGD 0 P4D 0
[ 3535.746530] Oops: 0000 [#1] SMP PTI
[ 3535.750049] Modules linked in: vhost_net vhost tap tun bridge stp llc ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter sunrpc vfat fat ext4 mbcache jbd2 intel_rapl skx_edac nfit libnvdimm x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm ipmi_ssif irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc ses aesni_intel crypto_simd cryptd enclosure hpwdt hpilo glue_helper ipmi_si pcspkr wmi mei_me ioatdma mei ipmi_devintf shpchp dca ipmi_msghandler lpc_ich acpi_power_meter sch_fq_codel ip_tables xfs libcrc32c sd_mod mgag200 i2c_algo_bit drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops ttm drm smartpqi i40e crc32c_intel scsi_transport_sas tg3 i2c_core ptp pps_core
[ 3535.813456] CPU: 5 PID: 1630 Comm: vhost-1614 Not tainted 4.17.0-rc4 #2
[ 3535.820127] Hardware name: HPE ProLiant DL360 Gen10/ProLiant DL360 Gen10, BIOS U32 11/14/2017
[ 3535.828732] RIP: 0010:__xdp_map_lookup_elem+0x5/0x30
[ 3535.833740] RSP: 0018:ffffb4bc47bf7c58 EFLAGS: 00010246
[ 3535.839009] RAX: ffff9fdfcfea1c40 RBX: 0000000000000000 RCX: ffff9fdf27fe3100
[ 3535.846205] RDX: ffff9fdfca769200 RSI: 0000000000000000 RDI: 0000000000000000
[ 3535.853402] RBP: ffffb4bc491d9000 R08: 00000000000045ad R09: 0000000000000ec0
[ 3535.860597] R10: 0000000000000001 R11: ffff9fdf26c3ce4e R12: ffff9fdf9e72c000
[ 3535.867794] R13: 0000000000000000 R14: fffffffffffffff2 R15: ffff9fdfc82cdd00
[ 3535.874990] FS:  0000000000000000(0000) GS:ffff9fdfcfe80000(0000) knlGS:0000000000000000
[ 3535.883152] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 3535.888948] CR2: 0000000000000018 CR3: 0000000bde724004 CR4: 00000000007626e0
[ 3535.896145] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 3535.903342] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 3535.910538] PKRU: 55555554
[ 3535.913267] Call Trace:
[ 3535.915736]  xdp_do_generic_redirect+0x7a/0x310
[ 3535.920310]  do_xdp_generic.part.117+0x285/0x370
[ 3535.924970]  tun_get_user+0x5b9/0x1260 [tun]
[ 3535.929279]  tun_sendmsg+0x52/0x70 [tun]
[ 3535.933237]  handle_tx+0x2ad/0x5f0 [vhost_net]
[ 3535.937721]  vhost_worker+0xa5/0x100 [vhost]
[ 3535.942030]  kthread+0xf5/0x130
[ 3535.945198]  ? vhost_dev_ioctl+0x3b0/0x3b0 [vhost]
[ 3535.950031]  ? kthread_bind+0x10/0x10
[ 3535.953727]  ret_from_fork+0x35/0x40
[ 3535.957334] Code: 0e 74 15 83 f8 10 75 05 e9 49 aa b3 ff f3 c3 0f 1f 80 00 00 00 00 f3 c3 e9 29 9d b3 ff 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 <8b> 47 18 83 f8 0e 74 0d 83 f8 10 75 05 e9 49 a9 b3 ff 31 c0 c3
[ 3535.976387] RIP: __xdp_map_lookup_elem+0x5/0x30 RSP: ffffb4bc47bf7c58
[ 3535.982883] CR2: 0000000000000018
[ 3535.987096] ---[ end trace 383b299dd1430240 ]---
[ 3536.131325] Kernel panic - not syncing: Fatal exception
[ 3536.137484] Kernel Offset: 0x26a00000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
[ 3536.281406] ---[ end Kernel panic - not syncing: Fatal exception ]---

And a kernel with generic case fixed still panics in tun driver XDP
redirect, because it disabled only preemption, but not bh.

[ 2055.128746] BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
[ 2055.136662] PGD 0 P4D 0
[ 2055.139219] Oops: 0000 [#1] SMP PTI
[ 2055.142736] Modules linked in: vhost_net vhost tap tun bridge stp llc ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter sunrpc vfat fat ext4 mbcache jbd2 intel_rapl skx_edac nfit libnvdimm x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc ses aesni_intel ipmi_ssif crypto_simd enclosure cryptd hpwdt glue_helper ioatdma hpilo wmi dca pcspkr ipmi_si acpi_power_meter ipmi_devintf shpchp mei_me ipmi_msghandler mei lpc_ich sch_fq_codel ip_tables xfs libcrc32c sd_mod mgag200 i2c_algo_bit drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops ttm drm i40e smartpqi tg3 scsi_transport_sas crc32c_intel i2c_core ptp pps_core
[ 2055.206142] CPU: 6 PID: 1693 Comm: vhost-1683 Tainted: G        W         4.17.0-rc5-fix-tun+ #1
[ 2055.215011] Hardware name: HPE ProLiant DL360 Gen10/ProLiant DL360 Gen10, BIOS U32 11/14/2017
[ 2055.223617] RIP: 0010:__xdp_map_lookup_elem+0x5/0x30
[ 2055.228624] RSP: 0018:ffff998b07607cc0 EFLAGS: 00010246
[ 2055.233892] RAX: ffff8dbd8e235700 RBX: ffff8dbd8ff21c40 RCX: 0000000000000004
[ 2055.241089] RDX: ffff998b097a9000 RSI: 0000000000000000 RDI: 0000000000000000
[ 2055.248286] RBP: 0000000000000000 R08: 00000000000065a8 R09: 0000000000005d80
[ 2055.255483] R10: 0000000000000040 R11: ffff8dbcf0100000 R12: ffff998b097a9000
[ 2055.262681] R13: ffff8dbd8c98c000 R14: 0000000000000000 R15: ffff998b07607d78
[ 2055.269879] FS:  0000000000000000(0000) GS:ffff8dbd8ff00000(0000) knlGS:0000000000000000
[ 2055.278039] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2055.283834] CR2: 0000000000000018 CR3: 0000000c0c8cc005 CR4: 00000000007626e0
[ 2055.291030] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 2055.298227] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 2055.305424] PKRU: 55555554
[ 2055.308153] Call Trace:
[ 2055.310624]  xdp_do_redirect+0x7b/0x380
[ 2055.314499]  tun_get_user+0x10fe/0x12a0 [tun]
[ 2055.318895]  tun_sendmsg+0x52/0x70 [tun]
[ 2055.322852]  handle_tx+0x2ad/0x5f0 [vhost_net]
[ 2055.327337]  vhost_worker+0xa5/0x100 [vhost]
[ 2055.331646]  kthread+0xf5/0x130
[ 2055.334813]  ? vhost_dev_ioctl+0x3b0/0x3b0 [vhost]
[ 2055.339646]  ? kthread_bind+0x10/0x10
[ 2055.343343]  ret_from_fork+0x35/0x40
[ 2055.346950] Code: 0e 74 15 83 f8 10 75 05 e9 e9 aa b3 ff f3 c3 0f 1f 80 00 00 00 00 f3 c3 e9 c9 9d b3 ff 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 <8b> 47 18 83 f8 0e 74 0d 83 f8 10 75 05 e9 e9 a9 b3 ff 31 c0 c3
[ 2055.366004] RIP: __xdp_map_lookup_elem+0x5/0x30 RSP: ffff998b07607cc0
[ 2055.372500] CR2: 0000000000000018
[ 2055.375856] ---[ end trace 2a2dcc5e9e174268 ]---
[ 2055.523626] Kernel panic - not syncing: Fatal exception
[ 2055.529796] Kernel Offset: 0x2e000000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
[ 2055.677539] ---[ end Kernel panic - not syncing: Fatal exception ]---

v2:
 - Removed preempt_disable/enable since local_bh_disable will prevent
   preemption as well, feedback from Jason Wang.

Fixes: 761876c8 ("tap: XDP support")
Signed-off-by: NToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Acked-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch change the API for ndo_xdp_xmit to support bulking
xdp_frames.

When kernel is compiled with CONFIG_RETPOLINE, XDP sees a huge slowdown.
Most of the slowdown is caused by DMA API indirect function calls, but
also the net_device->ndo_xdp_xmit() call.

Benchmarked patch with CONFIG_RETPOLINE, using xdp_redirect_map with
single flow/core test (CPU E5-1650 v4 @ 3.60GHz), showed
performance improved:
 for driver ixgbe: 6,042,682 pps -> 6,853,768 pps = +811,086 pps
 for driver i40e : 6,187,169 pps -> 6,724,519 pps = +537,350 pps

With frames avail as a bulk inside the driver ndo_xdp_xmit call,
further optimizations are possible, like bulk DMA-mapping for TX.

Testing without CONFIG_RETPOLINE show the same performance for
physical NIC drivers.

The virtual NIC driver tun sees a huge performance boost, as it can
avoid doing per frame producer locking, but instead amortize the
locking cost over the bulk.

V2: Fix compile errors reported by kbuild test robot <lkp@intel.com>
V4: Isolated ndo, driver changes and callers.
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>

When link is down, writes to the device might fail with
-EIO. Userspace needs an indication when the status is resolved.  As a
fix, tun_net_open() attempts to wake up writers - but that is only
effective if SOCKWQ_ASYNC_NOSPACE has been set in the past. This is
not the case of vhost_net which only poll for EPOLLOUT after it meets
errors during sendmsg().

This patch fixes this by making sure SOCKWQ_ASYNC_NOSPACE is set when
socket is not writable or device is down to guarantee EPOLLOUT will be
raised in either tun_chr_poll() or tun_sock_write_space() after device
is up.

Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Eric Dumazet <edumazet@google.com>
Fixes: 1bd4978a ("tun: honor IFF_UP in tun_get_user()")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After commit b196d88a ("tun: fix use after free for ptr_ring") we
need clean up tx ring during release(). But unfortunately, it tries to
do the cleanup blindly after socket were destroyed which will lead
another use-after-free. Fix this by doing the cleanup before dropping
the last reference of the socket in __tun_detach().
Reported-by: NAndrei Vagin <avagin@virtuozzo.com>
Acked-by: NAndrei Vagin <avagin@virtuozzo.com>
Fixes: b196d88a ("tun: fix use after free for ptr_ring")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We used to initialize ptr_ring during TUNSETIFF, this is because its
size depends on the tx_queue_len of netdevice. And we try to clean it
up when socket were detached from netdevice. A race were spotted when
trying to do uninit during a read which will lead a use after free for
pointer ring. Solving this by always initialize a zero size ptr_ring
in open() and do resizing during TUNSETIFF, and then we can safely do
cleanup during close(). With this, there's no need for the workaround
that was introduced by commit 4df0bfc7 ("tun: fix a memory leak
for tfile->tx_array").

Reported-by: syzbot+e8b902c3c3fadf0a9dba@syzkaller.appspotmail.com
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Fixes: 1576d986 ("tun: switch to use skb array for tx")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since net ns of tun device is assigned on the device creation,
and it never changes, we do not need to use any lock to get it
from alive tun.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently, the tun driver, in absence of an eBPF steering program,
always compute the rxhash in its rx path, even when such value
is later unused due to additional checks (

This changeset moves the all the related checks just before the
__skb_get_hash_symmetric(), so that the latter is no more computed
when unneeded.

Also replace an unneeded RCU section with rcu_access_pointer().
Signed-off-by: NPaolo Abeni <pabeni@redhat.com>
Acked-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

w/ bpf_xdp_adjust_tail helper xdp's data_end pointer could be changed as
well (only "decrease" of pointer's location is going to be supported).
changing of this pointer will change packet's size.
for tun driver we need to adjust XDP_PASS handling by recalculating
length of the packet if it was passed to the TCP/IP stack
(in case if after xdp's prog run data_end pointer was adjusted)
Reviewed-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NNikita V. Shirokov <tehnerd@tehnerd.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>

Bogus trimming in tun_net_xmit() causes truncated vlan packets.

skb->len is correct whether or not skb_vlan_tag_present() is true. There
is no more reason to adjust the skb length on xmit in this driver than
any other driver. tun_put_user() adds 4 bytes to the total for tagged
packets because it transmits the tag inline to userspace.  This is
similar to a nic transmitting the tag inline on the wire.

Reproducing the bug by sending any tagged packet through back-to-back
connected tap interfaces:

 socat TUN,tun-type=tap,iff-up,tun-name=in TUN,tun-type=tap,iff-up,tun-name=out &
 ip link add link in name in.20 type vlan id 20
 ip addr add 10.9.9.9/24 dev in.20
 ip link set in.20 up
 tshark -nxxi in -f arp -c1 2>/dev/null &
 tshark -nxxi out -f arp -c1 2>/dev/null &
 ping -c 1 10.9.9.5 >/dev/null 2>&1

The output from the 'in' and 'out' interfaces are different when the
bug is present:

 Capturing on 'in'
 0000  ff ff ff ff ff ff 76 cf 76 37 d5 0a 81 00 00 14   ......v.v7......
 0010  08 06 00 01 08 00 06 04 00 01 76 cf 76 37 d5 0a   ..........v.v7..
 0020  0a 09 09 09 00 00 00 00 00 00 0a 09 09 05         ..............

 Capturing on 'out'
 0000  ff ff ff ff ff ff 76 cf 76 37 d5 0a 81 00 00 14   ......v.v7......
 0010  08 06 00 01 08 00 06 04 00 01 76 cf 76 37 d5 0a   ..........v.v7..
 0020  0a 09 09 09 00 00 00 00 00 00                     ..........

Fixes: aff3d70a ("tun: allow to attach ebpf socket filter")
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: NBjørn Mork <bjorn@mork.no>
Acked-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Changing API ndo_xdp_xmit to take a struct xdp_frame instead of struct
xdp_buff.  This brings xdp_return_frame and ndp_xdp_xmit in sync.

This builds towards changing the API further to become a bulk API,
because xdp_buff is not a queue-able object while xdp_frame is.

V4: Adjust for commit 59655a5b ("tuntap: XDP_TX can use native XDP")
V7: Adjust for commit d9314c47 ("i40e: add support for XDP_REDIRECT")
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Changing API xdp_return_frame() to take struct xdp_frame as argument,
seems like a natural choice. But there are some subtle performance
details here that needs extra care, which is a deliberate choice.

When de-referencing xdp_frame on a remote CPU during DMA-TX
completion, result in the cache-line is change to "Shared"
state. Later when the page is reused for RX, then this xdp_frame
cache-line is written, which change the state to "Modified".

This situation already happens (naturally) for, virtio_net, tun and
cpumap as the xdp_frame pointer is the queued object.  In tun and
cpumap, the ptr_ring is used for efficiently transferring cache-lines
(with pointers) between CPUs. Thus, the only option is to
de-referencing xdp_frame.

It is only the ixgbe driver that had an optimization, in which it can
avoid doing the de-reference of xdp_frame.  The driver already have
TX-ring queue, which (in case of remote DMA-TX completion) have to be
transferred between CPUs anyhow.  In this data area, we stored a
struct xdp_mem_info and a data pointer, which allowed us to avoid
de-referencing xdp_frame.

To compensate for this, a prefetchw is used for telling the cache
coherency protocol about our access pattern.  My benchmarks show that
this prefetchw is enough to compensate the ixgbe driver.

V7: Adjust for commit d9314c47 ("i40e: add support for XDP_REDIRECT")
V8: Adjust for commit bd658dda ("net/mlx5e: Separate dma base address
and offset in dma_sync call")
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use the IDA infrastructure for getting a cyclic increasing ID number,
that is used for keeping track of each registered allocator per
RX-queue xdp_rxq_info.  Instead of using the IDR infrastructure, which
uses a radix tree, use a dynamic rhashtable, for creating ID to
pointer lookup table, because this is faster.

The problem that is being solved here is that, the xdp_rxq_info
pointer (stored in xdp_buff) cannot be used directly, as the
guaranteed lifetime is too short.  The info is needed on a
(potentially) remote CPU during DMA-TX completion time . In an
xdp_frame the xdp_mem_info is stored, when it got converted from an
xdp_buff, which is sufficient for the simple page refcnt based recycle
schemes.

For more advanced allocators there is a need to store a pointer to the
registered allocator.  Thus, there is a need to guard the lifetime or
validity of the allocator pointer, which is done through this
rhashtable ID map to pointer. The removal and validity of of the
allocator and helper struct xdp_mem_allocator is guarded by RCU.  The
allocator will be created by the driver, and registered with
xdp_rxq_info_reg_mem_model().

It is up-to debate who is responsible for freeing the allocator
pointer or invoking the allocator destructor function.  In any case,
this must happen via RCU freeing.

Use the IDA infrastructure for getting a cyclic increasing ID number,
that is used for keeping track of each registered allocator per
RX-queue xdp_rxq_info.

V4: Per req of Jason Wang
- Use xdp_rxq_info_reg_mem_model() in all drivers implementing
  XDP_REDIRECT, even-though it's not strictly necessary when
  allocator==NULL for type MEM_TYPE_PAGE_SHARED (given it's zero).

V6: Per req of Alex Duyck
- Introduce rhashtable_lookup() call in later patch

V8: Address sparse should be static warnings (from kbuild test robot)
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The tuntap driver invented it's own driver specific way of queuing
XDP packets, by storing the xdp_buff information in the top of
the XDP frame data.

Convert it over to use the more generic xdp_frame structure.  The
main problem with the in-driver method is that the xdp_rxq_info pointer
cannot be trused/used when dequeueing the frame.

V3: Remove check based on feedback from Jason
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

I added dumping of link information about tun devices over netlink in
commit 1ec010e7 ("tun: export flags, uid, gid, queue information
over netlink"), but didn't add the missing netlink notifications when
the device's exported properties change.

This patch adds notifications when owner/group or flags are modified,
when queues are attached/detached, and when a tun fd is closed.
Reported-by: NThomas Haller <thaller@redhat.com>
Fixes: 1ec010e7 ("tun: export flags, uid, gid, queue information over netlink")
Signed-off-by: NSabrina Dubroca <sd@queasysnail.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Otherwise, register_netdevice advertises the creation of the device with
the default flags, instead of what the user requested.
Reported-by: NThomas Haller <thaller@redhat.com>
Fixes: 1ec010e7 ("tun: export flags, uid, gid, queue information over netlink")
Signed-off-by: NSabrina Dubroca <sd@queasysnail.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now we have ndo_xdp_xmit, switch to use it instead of the slow generic
XDP TX routine. XDP_TX on TAP gets ~20% improvements from ~1.5Mpps to
~1.8Mpps on 2.60GHz Core(TM) i7-5600U.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After commit fc72d1d5 ("tuntap: XDP transmission"), we can
actually queueing XDP pointers in the pointer ring, so we should
examine the pointer type before freeing the pointer.

Fixes: fc72d1d5 ("tuntap: XDP transmission")
Reported-by: NMichael S. Tsirkin <mst@redhat.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We don't flush batched XDP packets through xdp_do_flush_map(), this
will cause packets stall at TX queue. Consider we don't do XDP on NAPI
poll(), the only possible fix is to call xdp_do_flush_map()
immediately after xdp_do_redirect().

Note, this in fact won't try to batch packets through devmap, we could
address in the future.
Reported-by: NChristoffer Dall <christoffer.dall@linaro.org>
Fixes: 761876c8 ("tap: XDP support")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Except for tuntap, all other drivers' XDP was implemented at NAPI
poll() routine in a bh. This guarantees all XDP operation were done at
the same CPU which is required by e.g BFP_MAP_TYPE_PERCPU_ARRAY. But
for tuntap, we do it in process context and we try to protect XDP
processing by RCU reader lock. This is insufficient since
CONFIG_PREEMPT_RCU can preempt the RCU reader critical section which
breaks the assumption that all XDP were processed in the same CPU.

Fixing this by simply disabling preemption during XDP processing.

Fixes: 761876c8 ("tap: XDP support")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit 762c330d. The
reason is we try to batch packets for devmap which causes calling
xdp_do_flush() in the process context. Simply disabling preemption
may not work since process may move among processors which lead
xdp_do_flush() to miss some flushes on some processors.

So simply revert the patch, a follow-up patch will add the xdp flush
correctly.
Reported-by: NChristoffer Dall <christoffer.dall@linaro.org>
Fixes: 762c330d ("tuntap: add missing xdp flush")
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NSabrina Dubroca <sd@queasysnail.net>
Reviewed-by: NStefano Brivio <sbrivio@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

<Mark Rutland reported>
    While fuzzing arm64 v4.16-rc1 with Syzkaller, I've been hitting a
    misaligned atomic in __skb_clone:

        atomic_inc(&(skb_shinfo(skb)->dataref));

   where dataref doesn't have the required natural alignment, and the
   atomic operation faults. e.g. i often see it aligned to a single
   byte boundary rather than a four byte boundary.

   AFAICT, the skb_shared_info is misaligned at the instant it's
   allocated in __napi_alloc_skb()  __napi_alloc_skb()
</end of report>

Problem is caused by tun_napi_alloc_frags() using
napi_alloc_frag() with user provided seg sizes,
leading to other users of this API getting unaligned
page fragments.

Since we would like to not necessarily add paddings or alignments to
the frags that tun_napi_alloc_frags() attaches to the skb, switch to
another page frag allocator.

As a bonus skb_page_frag_refill() can use GFP_KERNEL allocations,
meaning that we can not deplete memory reserves as easily.

Fixes: 90e33d45 ("tun: enable napi_gro_frags() for TUN/TAP driver")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NMark Rutland <mark.rutland@arm.com>
Tested-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds possibility to get tun device's net namespace fd
in the same way we allow to do that for sockets.

Socket ioctl numbers do not intersect with tun-specific, and there
is already SIOCSIFHWADDR used in tun code. So, SIOCGSKNS number
is choosen instead of custom-made for this functionality.

Note, that open_related_ns() uses plain get_net_ns() and it's safe
(net can't be already dead at this moment):

  tun socket is allocated via sk_alloc() with zero last arg (kern = 0).
  So, each alive socket increments net::count, and the socket is definitely
  alive during ioctl syscall.

Also, common variable net is introduced, so small cleanup in TUNSETIFF
is made.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is the mindless scripted replacement of kernel use of POLL*
variables as described by Al, done by this script:

    for V in IN OUT PRI ERR RDNORM RDBAND WRNORM WRBAND HUP RDHUP NVAL MSG; do
        L=`git grep -l -w POLL$V | grep -v '^t' | grep -v /um/ | grep -v '^sa' | grep -v '/poll.h$'|grep -v '^D'`
        for f in $L; do sed -i "-es/^\([^\"]*\)\(\<POLL$V\>\)/\\1E\\2/" $f; done
    done

with de-mangling cleanups yet to come.

NOTE! On almost all architectures, the EPOLL* constants have the same
values as the POLL* constants do.  But they keyword here is "almost".
For various bad reasons they aren't the same, and epoll() doesn't
actually work quite correctly in some cases due to this on Sparc et al.

The next patch from Al will sort out the final differences, and we
should be all done.
Scripted-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>