New attr-based bridge_setlink can recurse lower devs and recover on err, so
remove old wrapper (including ndo_dflt_switchdev_port_bridge_setlink).
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add new switchdev_port_bridge_setlink that can be used by drivers
implementing .ndo_bridge_setlink to set switchdev bridge attributes.
Basically turn the raw rtnl_bridge_setlink netlink into switchdev attr
sets.  Proper netlink attr policy checking is done on the protinfo part of
the netlink msg.

Currently, for protinfo, only bridge port attrs BR_LEARNING and
BR_LEARNING_SYNC are parsed and passed to port driver.

For afspec, VLAN objs are passed so switchdev driver can set VLANs assigned
to SELF.  To illustrate with iproute2 cmd, we have:

	bridge vlan add vid 10 dev sw1p1 self master

To add VLAN 10 to port sw1p1 for both the bridge (master) and the device
(self).
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Like switchdev attr get/set, add new switchdev obj add/del.  switchdev objs
will be things like VLANs or FIB entries, so add/del fits better for
objects than get/set used for attributes.

Use same two-phase prepare-commit transaction model as in attr set.
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NSridhar Samudrala <sridhar.samudrala@intel.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

STP update is just a settable port attribute, so convert
switchdev_port_stp_update to an attr set.

For DSA, the prepare phase is skipped and STP updates are only done in the
commit phase.  This is because currently the DSA drivers don't need to
allocate any memory for STP updates and the STP update will not fail to HW
(unless something horrible goes wrong on the MDIO bus, in which case the
prepare phase wouldn't have been able to predict anyway).
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Switch ID is just a gettable port attribute.  Convert switchdev op
switchdev_parent_id_get to a switchdev attr.

Note: for sysfs and netlink interfaces, SWITCHDEV_ATTR_PORT_PARENT_ID is
called with SWITCHDEV_F_NO_RECUSE to limit switch ID user-visiblity to only
port netdevs.  So when a port is stacked under bond/bridge, the user can
only query switch id via the switch ports, but not via the upper devices
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add two new swdev ops for get/set switch port attributes.  Most swdev
interactions on a port are gets or sets on port attributes, so rather than
adding ops for each attribute, let's define clean get/set ops for all
attributes, and then we can have clear, consistent rules on how attributes
propagate on stacked devs.

Add the basic algorithms for get/set attr ops.  Use the same recusive algo
to walk lower devs we've used for STP updates, for example.  For get,
compare attr value for each lower dev and only return success if attr
values match across all lower devs.  For sets, set the same attr value for
all lower devs.  We'll use a two-phase prepare-commit transaction model for
sets.  In the first phase, the driver(s) are asked if attr set is OK.  If
all OK, the commit attr set in second phase.  A driver would NACK the
prepare phase if it can't set the attr due to lack of resources or support,
within it's control.  RTNL lock must be held across both phases because
we'll recurse all lower devs first in prepare phase, and then recurse all
lower devs again in commit phase.  If any lower dev fails the prepare
phase, we need to abort the transaction for all lower devs.

If lower dev recusion isn't desired, allow a flag SWITCHDEV_F_NO_RECURSE to
indicate get/set only work on port (lowest) device.
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Turned out that "switchdev" sticks. So just unify all related terms to use
this prefix.
Signed-off-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NRoopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: NAndy Gospodarek <gospo@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Turned out that "switchdev" sticks. So just unify all related terms to use
this prefix.
Signed-off-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Acked-by: NRoopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: NAndy Gospodarek <gospo@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In a GRED qdisc, if the default "virtual queue" (VQ) does not have drop
parameters configured, then packets for the default VQ are not subjected
to RED and are only dropped if the queue is larger than the net_device's
tx_queue_len. This behavior is useful for WRED mode, since these packets
will still influence the calculated average queue length and (therefore)
the drop probability for all of the other VQs. However, for some drivers
tx_queue_len is zero. In other cases the user may wish to make the limit
the same for all VQs (including the default VQ with no drop parameters).

This change adds a TCA_GRED_LIMIT attribute to set the GRED queue limit,
in bytes, during qdisc setup. (This limit is in bytes to be consistent
with the drop parameters.) The default limit is the same as for a bfifo
queue (tx_queue_len * psched_mtu). If the drop parameters of any VQ are
configured with a smaller limit than the GRED queue limit, that VQ will
still observe the smaller limit instead.
Signed-off-by: NDavid Ward <david.ward@ll.mit.edu>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change adds a function called skb_free_frag which is meant to
compliment the function netdev_alloc_frag.  The general idea is to enable a
more lightweight version of page freeing since we don't actually need all
the overhead of a put_page, and we don't quite fit the model of __free_pages.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change moves the __alloc_page_frag functionality out of the networking
stack and into the page allocation portion of mm.  The idea it so help make
this maintainable by placing it with other page allocation functions.

Since we are moving it from skbuff.c to page_alloc.c I have also renamed
the basic defines and structure from netdev_alloc_cache to page_frag_cache
to reflect that this is now part of a different kernel subsystem.

I have also added a simple __free_page_frag function which can handle
freeing the frags based on the skb->head pointer.  The model for this is
based off of __free_pages since we don't actually need to deal with all of
the cases that put_page handles.  I incorporated the virt_to_head_page call
and compound_order into the function as it actually allows for a signficant
size reduction by reducing code duplication.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change makes it so that we store the virtual address of the page
in the netdev_alloc_cache instead of the page pointer.  The idea behind
this is to avoid multiple calls to page_address since the virtual address
is required for every access, but the page pointer is only needed at
allocation or reset of the page.

While I was at it I also reordered the netdev_alloc_cache structure a bit
so that the size is always 16 bytes by dropping size in the case where
PAGE_SIZE is greater than or equal to 32KB.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While testing I found that the testing for pfmemalloc in build_skb was
rather expensive.  I found the issue to be two-fold.  First we have to get
from the virtual address to the head page and that comes at the cost of
something like 11 cycles.  Then there is the cost for reading pfmemalloc out
of the head page which can be cache cold due to the fact that
put_page_testzero is likely invalidating the cache-line on one or more
CPUs as the fragments can be shared.

To avoid this extra expense I have added a pfmemalloc member to the
netdev_alloc_cache.  I then pushed pieces of __alloc_rx_skb into
__napi_alloc_skb and __netdev_alloc_skb so that I could rewrite them to
make use of the cached pfmemalloc value.  The result is that my perf traces
show a reduction from 9.28% overhead to 3.7% for the code covered by
build_skb, __alloc_rx_skb, and __napi_alloc_skb when performing a test with
the packet being dropped instead of being handed to napi_gro_receive.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Only left enqueue_root() user is netem, and it looks not necessary :

qdisc_skb_cb(skb)->pkt_len is preserved after one skb_clone()
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Ingress qdisc has no other purpose than calling into tc_classify()
that executes attached classifier(s) and action(s).

It has a 1:1 relationship to dev->ingress_queue. After having commit
087c1a60 ("net: sched: run ingress qdisc without locks") removed
the central ingress lock, one major contention point is gone.

The extra indirection layers however, are not necessary for calling
into ingress qdisc. pktgen calling locally into netif_receive_skb()
with a dummy u32, single CPU result on a Supermicro X10SLM-F, Xeon
E3-1240: before ~21,1 Mpps, after patch ~22,9 Mpps.

We can redirect the private classifier list to the netdev directly,
without changing any classifier API bits (!) and execute on that from
handle_ing() side. The __QDISC_STATE_DEACTIVATE test can be removed,
ingress qdisc doesn't have a queue and thus dev_deactivate_queue()
is also not applicable, ingress_cl_list provides similar behaviour.
In other words, ingress qdisc acts like TCQ_F_BUILTIN qdisc.

One next possible step is the removal of the dev's ingress (dummy)
netdev_queue, and to only have the list member in the netdevice
itself.

Note, the filter chain is RCU protected and individual filter elements
are being kfree'd by sched subsystem after RCU grace period. RCU read
lock is being held by __netif_receive_skb_core().

Joint work with Alexei Starovoitov.
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Signed-off-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Given quite some code has been removed from ing_filter(), we can just
consolidate that function into handle_ing() and get rid of a few
instructions at the same time.
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These functions are no longer needed and no longer used kill them.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Utilize the new functionality of sk_alloc so that nothing needs to be
done to suprress the reference counting on kernel sockets.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that sk_alloc knows when a kernel socket is being allocated modify
it to not reference count the network namespace of kernel sockets.

Keep track of if a socket needs reference counting by adding a flag to
struct sock called sk_net_refcnt.

Update all of the callers of sock_create_kern to stop using
sk_change_net and sk_release_kernel as those hacks are no longer
needed, to avoid reference counting a kernel socket.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In preparation for changing how struct net is refcounted
on kernel sockets pass the knowledge that we are creating
a kernel socket from sock_create_kern through to sk_alloc.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is long overdue, and is part of cleaning up how we allocate kernel
sockets that don't reference count struct net.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There is no need for tun to do the weird network namespace refcounting.
The existing network namespace refcounting in tfile has almost exactly
the same lifetime.  So rewrite the code to use the struct sock network
namespace refcounting and remove the unnecessary hand rolled network
namespace refcounting and the unncesary tfile->net.

This change allows the tun code to directly call sock_put bypassing
sock_release and making SOCK_EXTERNALLY_ALLOCATED unnecessary.

Remove the now unncessary tun_release so that if anything tries to use
the sock_release code path the kernel will oops, and let us know about
the bug.

The macvtap code already uses it's internal socket this way.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For DCTCP or similar ECN based deployments on fabrics with shallow
buffers, hosts are responsible for a good part of the buffering.

This patch adds an optional ce_threshold to codel & fq_codel qdiscs,
so that DCTCP can have feedback from queuing in the host.

A DCTCP enabled egress port simply have a queue occupancy threshold
above which ECT packets get CE mark.

In codel language this translates to a sojourn time, so that one doesn't
have to worry about bytes or bandwidth but delays.

This makes the host an active participant in the health of the whole
network.

This also helps experimenting DCTCP in a setup without DCTCP compliant
fabric.

On following example, ce_threshold is set to 1ms, and we can see from
'ldelay xxx us' that TCP is not trying to go around the 5ms codel
target.

Queue has more capacity to absorb inelastic bursts (say from UDP
traffic), as queues are maintained to an optimal level.

lpaa23:~# ./tc -s -d qd sh dev eth1
qdisc mq 1: dev eth1 root
 Sent 87910654696 bytes 58065331 pkt (dropped 0, overlimits 0 requeues 42961)
 backlog 3108242b 364p requeues 42961
qdisc codel 8063: dev eth1 parent 1:1 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms
 Sent 7363778701 bytes 4863809 pkt (dropped 0, overlimits 0 requeues 5503)
 rate 2348Mbit 193919pps backlog 255866b 46p requeues 5503
  count 0 lastcount 0 ldelay 1.0ms drop_next 0us
  maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 72384
qdisc codel 8064: dev eth1 parent 1:2 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms
 Sent 7636486190 bytes 5043942 pkt (dropped 0, overlimits 0 requeues 5186)
 rate 2319Mbit 191538pps backlog 207418b 64p requeues 5186
  count 0 lastcount 0 ldelay 694us drop_next 0us
  maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 69873
qdisc codel 8065: dev eth1 parent 1:3 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms
 Sent 11569360142 bytes 7641602 pkt (dropped 0, overlimits 0 requeues 5554)
 rate 3041Mbit 251096pps backlog 210446b 59p requeues 5554
  count 0 lastcount 0 ldelay 889us drop_next 0us
  maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 37780
...
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Florian Westphal <fw@strlen.de>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Glenn Judd <glenn.judd@morganstanley.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Introduce xmit_mode 'netif_receive' for pktgen which generates the
packets using familiar pktgen commands, but feeds them into
netif_receive_skb() instead of ndo_start_xmit().

Default mode is called 'start_xmit'.

It is designed to test netif_receive_skb and ingress qdisc
performace only. Make sure to understand how it works before
using it for other rx benchmarking.

Sample script 'pktgen.sh':
\#!/bin/bash
function pgset() {
  local result

  echo $1 > $PGDEV

  result=`cat $PGDEV | fgrep "Result: OK:"`
  if [ "$result" = "" ]; then
    cat $PGDEV | fgrep Result:
  fi
}

[ -z "$1" ] && echo "Usage: $0 DEV" && exit 1
ETH=$1

PGDEV=/proc/net/pktgen/kpktgend_0
pgset "rem_device_all"
pgset "add_device $ETH"

PGDEV=/proc/net/pktgen/$ETH
pgset "xmit_mode netif_receive"
pgset "pkt_size 60"
pgset "dst 198.18.0.1"
pgset "dst_mac 90:e2:ba:ff:ff:ff"
pgset "count 10000000"
pgset "burst 32"

PGDEV=/proc/net/pktgen/pgctrl
echo "Running... ctrl^C to stop"
pgset "start"
echo "Done"
cat /proc/net/pktgen/$ETH

Usage:
$ sudo ./pktgen.sh eth2
...
Result: OK: 232376(c232372+d3) usec, 10000000 (60byte,0frags)
  43033682pps 20656Mb/sec (20656167360bps) errors: 10000000

Raw netif_receive_skb speed should be ~43 million packet
per second on 3.7Ghz x86 and 'perf report' should look like:
  37.69%  kpktgend_0   [kernel.vmlinux]  [k] __netif_receive_skb_core
  25.81%  kpktgend_0   [kernel.vmlinux]  [k] kfree_skb
   7.22%  kpktgend_0   [kernel.vmlinux]  [k] ip_rcv
   5.68%  kpktgend_0   [pktgen]          [k] pktgen_thread_worker

If fib_table_lookup is seen on top, it means skb was processed
by the stack. To benchmark netif_receive_skb only make sure
that 'dst_mac' of your pktgen script is different from
receiving device mac and it will be dropped by ip_rcv
Signed-off-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Allow flag NO_TIMESTAMP to turn timestamping on again, like other flags,
with a negation of the flag like !NO_TIMESTAMP.

Also document the option flag NO_TIMESTAMP.

Fixes: afb84b62 ("pktgen: add flag NO_TIMESTAMP to disable timestamping")
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

More accurately, listen all netns that have a nsid assigned into the netns
where the netlink socket is opened.
For this purpose, a netlink socket option is added:
NETLINK_LISTEN_ALL_NSID. When this option is set on a netlink socket, this
socket will receive netlink notifications from all netns that have a nsid
assigned into the netns where the socket has been opened. The nsid is sent
to userland via an anscillary data.

With this patch, a daemon needs only one socket to listen many netns. This
is useful when the number of netns is high.

Because 0 is a valid value for a nsid, the field nsid_is_set indicates if
the field nsid is valid or not. skb->cb is initialized to 0 on skb
allocation, thus we are sure that we will never send a nsid 0 by error to
the userland.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These flags and states have the same prefix (NETLINK_) that netlink socket
options. To avoid confusion and to be able to name a flag like a socket
option, let's use an other prefix: NETLINK_[S|F]_.

Note: a comment has been fixed, it was talking about
NETLINK_RECV_NO_ENOBUFS socket option instead of NETLINK_NO_ENOBUFS.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Before this patch, nsid were protected by the rtnl lock. The goal of this
patch is to be able to find a nsid without needing to hold the rtnl lock.

The next patch will introduce a netlink socket option to listen to all
netns that have a nsid assigned into the netns where the socket is opened.
Thus, it's important to call rtnl_net_notifyid() outside the spinlock, to
avoid a recursive lock (nsid are notified via rtnl). This was the main
reason of the previous patch.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There is no functional change with this patch. It will ease the refactoring
of the locking system that protects nsids and the support of the netlink
socket option NETLINK_LISTEN_ALL_NSID.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In a following commit, a new function will be introduced to only lookup for
a nsid (no allocation if the nsid doesn't exist). To avoid confusion, the
existing function is renamed.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The goal of this commit is to prepare the rework of the locking of nsnid
protection.
After this patch, rtnl_net_notifyid() will not call anymore __peernet2id(),
ie no idr_* operation into this function.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All callers of this function expect a nsid, not an error.
Thus, returns NETNSA_NSID_NOT_ASSIGNED in case of error so that callers
don't have to convert the error to NETNSA_NSID_NOT_ASSIGNED.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Under tcp memory pressure, calling epoll_wait() in edge triggered
mode after -EAGAIN, can result in an indefinite hang in epoll_wait(),
even when there is sufficient memory available to continue making
progress. The problem is that when __sk_mem_schedule() returns 0
under memory pressure, we do not set the SOCK_NOSPACE flag in the
tcp write paths (tcp_sendmsg() or do_tcp_sendpages()). Then, since
SOCK_NOSPACE is used to trigger wakeups when incoming acks create
sufficient new space in the write queue, all outstanding packets
are acked, but we never wake up with the the EPOLLOUT that we are
expecting from epoll_wait().

This issue is currently limited to epoll() when used in edge trigger
mode, since 'tcp_poll()', does in fact currently set SOCK_NOSPACE.
This is sufficient for poll()/select() and epoll() in level trigger
mode. However, in edge trigger mode, epoll() is relying on the write
path to set SOCK_NOSPACE. EPOLL(7) says that in edge-trigger mode we
can only call epoll_wait() after read/write return -EAGAIN. Thus, in
the case of the socket write, we are relying on the fact that
tcp_sendmsg()/network write paths are going to issue a wakeup for
us at some point in the future when we get -EAGAIN.

Normally, epoll() edge trigger works fine when we've exceeded the
sk->sndbuf because in that case we do set SOCK_NOSPACE. However, when
we return -EAGAIN from the write path b/c we are over the tcp memory
limits and not b/c we are over the sndbuf, we are never going to get
another wakeup.

I can reproduce this issue, using SO_SNDBUF, since __sk_mem_schedule()
will return 0, or failure more readily with SO_SNDBUF:

1) create socket and set SO_SNDBUF to N
2) add socket as edge trigger
3) write to socket and block in epoll on -EAGAIN
4) cause tcp mem pressure via: echo "<small val>" > net.ipv4.tcp_mem

The fix here is simply to set SOCK_NOSPACE in sk_stream_wait_memory()
when the socket is non-blocking. Note that SOCK_NOSPACE, in addition
to waking up outstanding waiters is also used to expand the size of
the sk->sndbuf. However, we will not expand it by setting it in this
case because tcp_should_expand_sndbuf(), ensures that no expansion
occurs when we are under tcp memory pressure.

Note that we could still hang if sk->sk_wmem_queue is 0, when we get
the -EAGAIN. In this case the SOCK_NOSPACE bit will not help, since we
are waiting for and event that will never happen. I believe
that this case is harder to hit (and did not hit in my testing),
in that over the tcp 'soft' memory limits, we continue to guarantee a
minimum write buffer size. Perhaps, we could return -ENOSPC in this
case, or maybe we simply issue a wakeup in this case, such that we
keep retrying the write. Note that this case is not specific to
epoll() ET, but rather would affect blocking sockets as well. So I
view this patch as bringing epoll() edge-trigger into sync with the
current poll()/select()/epoll() level trigger and blocking sockets
behavior.
Signed-off-by: NJason Baron <jbaron@akamai.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Seccomp has always been a special candidate when it comes to preparation
of its filters in seccomp_prepare_filter(). Due to the extra checks and
filter rewrite it partially duplicates code and has BPF internals exposed.

This patch adds a generic API inside the BPF code code that seccomp can use
and thus keep it's filter preparation code minimal and better maintainable.
The other side-effect is that now classic JITs can add seccomp support as
well by only providing a BPF_LDX | BPF_W | BPF_ABS translation.

Tested with seccomp and BPF test suites.
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Cc: Nicolas Schichan <nschichan@freebox.fr>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Kees Cook <keescook@chromium.org>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When seccomp BPF was added, it was discussed to add __GFP_NOWARN
flag for their configuration path as f.e. up to 32K allocations are
more prone to fail under stress. As we're going to reuse BPF API,
add __GFP_NOWARN flags where larger kmalloc() and friends allocations
could fail.

It doesn't make much sense to pass around __GFP_NOWARN everywhere as
an extra argument only for seccomp while we just as well could run
into similar issues for socket filters, where it's not desired to
have a user application throw a WARN() due to allocation failure.
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Cc: Nicolas Schichan <nschichan@freebox.fr>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Kees Cook <keescook@chromium.org>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Remove the calls to bpf_check_classic(), bpf_convert_filter() and
bpf_migrate_runtime() and let bpf_prepare_filter() take care of that
instead.

seccomp_check_filter() is passed to bpf_prepare_filter() so that it
gets called from there, after bpf_check_classic().

We can now remove exposure of two internal classic BPF functions
previously used by seccomp. The export of bpf_check_classic() symbol,
previously known as sk_chk_filter(), was there since pre git times,
and no in-tree module was using it, therefore remove it.

Joint work with Daniel Borkmann.
Signed-off-by: NNicolas Schichan <nschichan@freebox.fr>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Kees Cook <keescook@chromium.org>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is in preparation for use by the seccomp code, the rationale is
not to duplicate additional code within the seccomp layer, but instead,
have it abstracted and hidden within the classic BPF API.

As an interim step, this now also makes bpf_prepare_filter() visible
(not as exported symbol though), so that seccomp can reuse that code
path instead of reimplementing it.

Joint work with Daniel Borkmann.
Signed-off-by: NNicolas Schichan <nschichan@freebox.fr>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Kees Cook <keescook@chromium.org>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Diagnosing problems related to Window Probes has been hard because
we lack a counter.

TCPWinProbe counts the number of ACK packets a sender has to send
at regular intervals to make sure a reverse ACK packet opening back
a window had not been lost.

TCPKeepAlive counts the number of ACK packets sent to keep TCP
flows alive (SO_KEEPALIVE)
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Acked-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With the advent of small rto timers in datacenter TCP,
(ip route ... rto_min x), the following can happen :

1) Qdisc is full, transmit fails.

   TCP sets a timer based on icsk_rto to retry the transmit, without
   exponential backoff.
   With low icsk_rto, and lot of sockets, all cpus are servicing timer
   interrupts like crazy.
   Intent of the code was to retry with a timer between 200 (TCP_RTO_MIN)
   and 500ms (TCP_RESOURCE_PROBE_INTERVAL)

2) Receivers can send zero windows if they don't drain their receive queue.

   TCP sends zero window probes, based on icsk_rto current value, with
   exponential backoff.
   With /proc/sys/net/ipv4/tcp_retries2 being 15 (or even smaller in
   some cases), sender can abort in less than one or two minutes !
   If receiver stops the sender, it obviously doesn't care of very tight
   rto. Probability of dropping the ACK reopening the window is not
   worth the risk.

Lets change the base timer to be at least 200ms (TCP_RTO_MIN) for these
events (but not normal RTO based retransmits)

A followup patch adds a new SNMP counter, as it would have helped a lot
diagnosing this issue.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The legacy netlink API treated EPERM (permission denied) as
"operation not supported".
Reported-by: NTomi Ollila <tomi.ollila@iki.fi>
Signed-off-by: NRichard Alpe <richard.alpe@ericsson.com>
Reviewed-by: NErik Hugne <erik.hugne@ericsson.com>
Reviewed-by: NYing Xue <ying.xue@windriver.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>