Dave observed number of machines hitting OOM on the UDP send
path. The workload seems to be sending large UDP packets over
loopback. Since loopback has MTU of 64k kernel will try to
allocate an skb with up to 64k of head space. This has a good
chance of failing under memory pressure. What's worse if
the message length is <32k the allocation may trigger an
OOM killer.

This is entirely avoidable, we can use an skb with page frags.

af_unix solves a similar problem by limiting the head
length to SKB_MAX_ALLOC. This seems like a good and simple
approach. It means that UDP messages > 16kB will now
use fragments if underlying device supports SG, if extra
allocator pressure causes regressions in real workloads
we can switch to trying the large allocation first and
falling back.

v4: pre-calculate all the additions to alloclen so
    we can be sure it won't go over order-2
Reported-by: Dave Jones <dsj@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure that SO_NETNS_COOKIE returns a non-zero value, and
that sockets from different namespaces have a distinct cookie
value.
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

It's getting more common to run nested container environments for
testing cloud software. One of such examples is Kind [1] which runs a
Kubernetes cluster in Docker containers on a single host. Each container
acts as a Kubernetes node, and thus can run any Pod (aka container)
inside the former. This approach simplifies testing a lot, as it
eliminates complicated VM setups.

Unfortunately, such a setup breaks some functionality when cgroupv2 BPF
programs are used for load-balancing. The load-balancer BPF program
needs to detect whether a request originates from the host netns or a
container netns in order to allow some access, e.g. to a service via a
loopback IP address. Typically, the programs detect this by comparing
netns cookies with the one of the init ns via a call to
bpf_get_netns_cookie(NULL). However, in nested environments the latter
cannot be used given the Kubernetes node's netns is outside the init ns.
To fix this, we need to pass the Kubernetes node netns cookie to the
program in a different way: by extending getsockopt() with a
SO_NETNS_COOKIE option, the orchestrator which runs in the Kubernetes
node netns can retrieve the cookie and pass it to the program instead.

Thus, this is following up on Eric's commit 3d368ab8 ("net:
initialize net->net_cookie at netns setup") to allow retrieval via
SO_NETNS_COOKIE.  This is also in line in how we retrieve socket cookie
via SO_COOKIE.

  [1] https://kind.sigs.k8s.io/Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Martynas Pumputis <m@lambda.lt>
Cc: Eric Dumazet <edumazet@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Dmytro Linkin says:

====================
Fixes for devlink rate objects API

Patch #1 fixes not decreased refcount of parent node for destroyed leaf
object.

Patch #2 fixes incorect eswitch mode check.

Patch #3 protects list traversing with a lock.

====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Devlink eswitch set command doesn't hold devlink->lock, which makes
possible race condition between rate list traversing and others devlink
rate KAPI calls, like devlink_rate_nodes_destroy().
Hold devlink lock while traversing the list.

Fixes: a8ecb93e ("devlink: Introduce rate nodes")
Signed-off-by: Dmytro Linkin <dlinkin@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When eswitch is disabled, querying its current mode results in error.
Due to this when trying to set the eswitch mode for mlx5 devices, it
fails to set the eswitch switchdev mode.
Hence remove such check.

Fixes: a8ecb93e ("devlink: Introduce rate nodes")
Signed-off-by: Dmytro Linkin <dlinkin@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Port functions, like SFs, can be deleted by the user when its leaf rate
object has parent node. In such case node refcnt won't be decreased
which blocks the node from deletion later.
Do simple refcnt decrease, since driver in cleanup stage. This:
1) assumes that driver took proper internal parent unset action;
2) allows to avoid nested callbacks call and deadlock.

Fixes: d7555984 ("devlink: Allow setting parent node of rate objects")
Signed-off-by: Dmytro Linkin <dlinkin@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

virtio_find_vqs_ctx() is defined but never be called currently,
it is the right place to use it.
Signed-off-by: Xianting Tian <xianting.tian@linux.alibaba.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The commit d26b698d ("net/tls: add skeleton of MIB statistics")
introduced __TLS_DEC_STATS(), but it is not used and __SNMP_DEC_STATS() is
not defined also. Let's remove it.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

This commit adds two stats for the socket migration feature to evaluate the
effectiveness: LINUX_MIB_TCPMIGRATEREQ(SUCCESS|FAILURE).

If the migration fails because of the own_req race in receiving ACK and
sending SYN+ACK paths, we do not increment the failure stat. Then another
CPU is responsible for the req.

Link: https://lore.kernel.org/bpf/CAK6E8=cgFKuGecTzSCSQ8z3YJ_163C0uwO9yRvfDSE7vOe9mJA@mail.gmail.com/Suggested-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

TCP checksums on received packets may be set to NULL by the sender if CSO
is enabled. The hypervisor flags these packets as check-sum-ok and the
skb is then flagged CHECKSUM_UNNECESSARY. If these packets are then
forwarded the sender will not request CSO due to the CHECKSUM_UNNECESSARY
flag. The result is a TCP packet sent with a bad checksum. This change
sets up CHECKSUM_PARTIAL on these packets causing the sender to correctly
request CSUM offload.
Signed-off-by: David Wilder <dwilder@us.ibm.com>
Reviewed-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
Tested-by: Cristobal Forno <cforno12@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Saeed Mahameed says:

====================
mlx5-net-next-2021-06-22

1) Various minor cleanups and fixes from net-next branch
2) Optimize mlx5 feature check on tx and
   a fix to allow Vxlan with Ipsec offloads
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next:

1) Skip non-SCTP packets in the new SCTP chunk support for nft_exthdr,
   from Phil Sutter.

2) Simplify TCP option sanity check for TCP packets, also from Phil.

3) Add a new expression to store when the rule has been used last time.

4) Pass the hook state object to log function, from Florian Westphal.

5) Document the new sysctl knobs to tune the flowtable timeouts,
   from Oz Shlomo.

6) Fix snprintf error check in the new nfnetlink_hook infrastructure,
   from Dan Carpenter.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

According to a comment in commit 99513cfa ("selftest: Fixes for
icmp_redirect test") the test "IPv6: mtu exception plus redirect" is
expected to fail, because of a bug in the IPv6 logic that hasn't been
fixed yet apparently.

We should probably consider this failure as an "expected failure",
therefore change the script to return XFAIL for that particular test and
also report the total amount of expected failures at the end of the run.
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Yunsheng Lin says:

====================
Some optimization for lockless qdisc

Patch 1: remove unnecessary seqcount operation.
Patch 2: implement TCQ_F_CAN_BYPASS.
Patch 3: remove qdisc->empty.

Performance data for pktgen in queue_xmit mode + dummy netdev
with pfifo_fast:

 threads    unpatched           patched             delta
    1       2.60Mpps            3.21Mpps             +23%
    2       3.84Mpps            5.56Mpps             +44%
    4       5.52Mpps            5.58Mpps             +1%
    8       2.77Mpps            2.76Mpps             -0.3%
   16       2.24Mpps            2.23Mpps             -0.4%

Performance for IP forward testing: 1.05Mpps increases to
1.16Mpps, about 10% improvement.

V3: Add 'Acked-by' from Jakub and 'Tested-by' from Vladimir,
    and resend based on latest net-next.
V2: Adjust the comment and commit log according to discussion
    in V1.
V1: Drop RFC tag, add nolock_qdisc_is_empty() and do the qdisc
    empty checking without the protection of qdisc->seqlock to
    aviod doing unnecessary spin_trylock() for contention case.
RFC v4: Use STATE_MISSED and STATE_DRAINING to indicate non-empty
        qdisc, and add patch 1 and 3.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As MISSED and DRAINING state are used to indicate a non-empty
qdisc, qdisc->empty is not longer needed, so remove it.
Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # flexcan
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently pfifo_fast has both TCQ_F_CAN_BYPASS and TCQ_F_NOLOCK
flag set, but queue discipline by-pass does not work for lockless
qdisc because skb is always enqueued to qdisc even when the qdisc
is empty, see __dev_xmit_skb().

This patch calls sch_direct_xmit() to transmit the skb directly
to the driver for empty lockless qdisc, which aviod enqueuing
and dequeuing operation.

As qdisc->empty is not reliable to indicate a empty qdisc because
there is a time window between enqueuing and setting qdisc->empty.
So we use the MISSED state added in commit a90c57f2 ("net:
sched: fix packet stuck problem for lockless qdisc"), which
indicate there is lock contention, suggesting that it is better
not to do the qdisc bypass in order to avoid packet out of order
problem.

In order to make MISSED state reliable to indicate a empty qdisc,
we need to ensure that testing and clearing of MISSED state is
within the protection of qdisc->seqlock, only setting MISSED state
can be done without the protection of qdisc->seqlock. A MISSED
state testing is added without the protection of qdisc->seqlock to
aviod doing unnecessary spin_trylock() for contention case.

As the enqueuing is not within the protection of qdisc->seqlock,
there is still a potential data race as mentioned by Jakub [1]:

      thread1               thread2             thread3
qdisc_run_begin() # true
                        qdisc_run_begin(q)
                             set(MISSED)
pfifo_fast_dequeue
  clear(MISSED)
  # recheck the queue
qdisc_run_end()
                            enqueue skb1
                                             qdisc empty # true
                                          qdisc_run_begin() # true
                                          sch_direct_xmit() # skb2
                         qdisc_run_begin()
                            set(MISSED)

When above happens, skb1 enqueued by thread2 is transmited after
skb2 is transmited by thread3 because MISSED state setting and
enqueuing is not under the qdisc->seqlock. If qdisc bypass is
disabled, skb1 has better chance to be transmited quicker than
skb2.

This patch does not take care of the above data race, because we
view this as similar as below:
Even at the same time CPU1 and CPU2 write the skb to two socket
which both heading to the same qdisc, there is no guarantee that
which skb will hit the qdisc first, because there is a lot of
factor like interrupt/softirq/cache miss/scheduling afffecting
that.

There are below cases that need special handling:
1. When MISSED state is cleared before another round of dequeuing
   in pfifo_fast_dequeue(), and __qdisc_run() might not be able to
   dequeue all skb in one round and call __netif_schedule(), which
   might result in a non-empty qdisc without MISSED set. In order
   to avoid this, the MISSED state is set for lockless qdisc and
   __netif_schedule() will be called at the end of qdisc_run_end.

2. The MISSED state also need to be set for lockless qdisc instead
   of calling __netif_schedule() directly when requeuing a skb for
   a similar reason.

3. For netdev queue stopped case, the MISSED case need clearing
   while the netdev queue is stopped, otherwise there may be
   unnecessary __netif_schedule() calling. So a new DRAINING state
   is added to indicate this case, which also indicate a non-empty
   qdisc.

4. As there is already netif_xmit_frozen_or_stopped() checking in
   dequeue_skb() and sch_direct_xmit(), which are both within the
   protection of qdisc->seqlock, but the same checking in
   __dev_xmit_skb() is without the protection, which might cause
   empty indication of a lockless qdisc to be not reliable. So
   remove the checking in __dev_xmit_skb(), and the checking in
   the protection of qdisc->seqlock seems enough to avoid the cpu
   consumption problem for netdev queue stopped case.

1. https://lkml.org/lkml/2021/5/29/215Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # flexcan
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qdisc->running seqcount operation is mainly used to do heuristic
locking on q->busylock for locked qdisc, see qdisc_is_running()
and __dev_xmit_skb().

So avoid doing seqcount operation for qdisc with TCQ_F_NOLOCK
flag.
Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # flexcan
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The packet is VXLAN packet over IPsec transport mode tunnel
which has the following format: [IP1 | ESP | UDP | VXLAN | IP2 | TCP]
NVIDIA ConnectX card cannot do checksum offload for two L4 headers.
The solution is using the checksum partial offload similar to
VXLAN | TCP packet. Hardware calculates IP1, IP2 and TCP checksums and
software calculates UDP checksum. However, unlike VXLAN | TCP case,
IPsec's mlx5 driver cannot access the inner plaintext IP protocol type.
Therefore, inner_ipproto is added in the sec_path structure
to provide this information. Also, utilize the skb's csum_start to
program L4 inner checksum offset.

While at it, remove the call to mlx5e_set_eseg_swp and setup software parser
fields directly in mlx5e_ipsec_set_swp. mlx5e_set_eseg_swp is not
needed as the two features (GENEVE and IPsec) are different and adding
this sharing layer creates unnecessary complexity and affect
performance.

For the case VXLAN packet over IPsec tunnel mode tunnel, checksum offload
is disabled because the hardware does not support checksum offload for
three L3 (IP) headers.
Signed-off-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Huy Nguyen <huyn@nvidia.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

The inner_ipproto saves the inner IP protocol of the plain
text packet. This allows vendor's IPsec feature making offload
decision at skb's features_check and configuring hardware at
ndo_start_xmit.

For example, ConnectX6-DX IPsec device needs the plaintext's
IP protocol to support partial checksum offload on
VXLAN/GENEVE packet over IPsec transport mode tunnel.
Signed-off-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Huy Nguyen <huyn@nvidia.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Acked-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

mlx5e_ipsec_feature_check belongs to mlx5e_tunnel_features_check.
Also, IPsec is not the default configuration so it should be
checked at the end instead of the beginning of mlx5e_features_check.
Signed-off-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Huy Nguyen <huyn@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

remove "default n" and "No" is default
Signed-off-by: caihuoqing <caihuoqing@baidu.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

There is a spelling mistake in a mlx5_core_err error message. Fix it.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

When CONFIG_CPUMASK_OFFSTACK is unset, cpumask_var_t is not a pointer
but a single element array, meaning its address in a structure cannot be
NULL as long as it is not the first element, which it is not. This
results in a clang warning:

drivers/net/ethernet/mellanox/mlx5/core/eq.c:715:14: warning: address of
array 'param->affinity' will always evaluate to 'true'
[-Wpointer-bool-conversion]
        if (!param->affinity)
            ~~~~~~~~^~~~~~~~
1 warning generated.

The helper cpumask_available was added in commit f7e30f01 ("cpumask:
Add helper cpumask_available()") to handle situations like this so use
it to keep the meaning of the code the same while resolving the warning.

Fixes: e4e3f24b ("net/mlx5: Provide cpumask at EQ creation phase")
Link: https://github.com/ClangBuiltLinux/linux/issues/1400Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

The error code is missing in this code scenario, add the error code
'-ENOMEM' to the return value 'err'.

Eliminate the follow smatch warning:

drivers/net/ethernet/mellanox/mlx5/core/fs_core.c:2973 mlx5_init_fs()
warn: missing error code 'err'.
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Fixes: 4a98544d ("net/mlx5: Move chains ft pool to be used by all firmware steering").
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Mat Martineau says:

====================
mptcp: Connection-time 'C' flag and two fixes

Here are six more patches from the MPTCP tree.

Most of them add support for the 'C' flag in the MPTCP connection-time
option headers. This flag affects how the initial address and port are
treated by each peer. Normally one peer may send MP_JOIN requests to the
remote address and port that were used when initiating the MPTCP
connection. The 'C' bit indicates that MP_JOINs should only be sent to
remote addresses that have been advertised with ADD_ADDR.

The other two patches are unrelated improvements.

Patches 1-4: Add the 'C' flag feature, a sysctl to optionally enable it,
and a selftest.

Patch 5: Adjust rp_filter settings in a selftest.

Patch 6: Improve rbuf cleanup for MPTCP sockets.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The current cleanup rbuf tries a bit too hard to avoid acquiring
the subflow socket lock. We may end-up delaying the needed ack,
or skip acking a blocked subflow.

Address the above extending the conditions used to trigger the cleanup
to reflect more closely what TCP does and invoking tcp_cleanup_rbuf()
on all the active subflows.

Note that we can't replicate the exact tests implemented in
tcp_cleanup_rbuf(), as MPTCP lacks some of the required info - e.g.
ping-pong mode.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

To turn rp_filter off we should:

  echo 0 > /proc/sys/net/ipv4/conf/default/rp_filter

and

  echo 0 > /proc/sys/net/ipv4/conf/all/rp_filter

before NIC created.
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Yonglong Li <liyonglong@chinatelecom.cn>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch added a new argument '-d' for mptcp_join.sh script, to invoke
the testcases for the MP_CAPABLE 'C' flag.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch added a new flag named deny_join_id0 in struct
mptcp_options_received. Set it when MP_CAPABLE with the flag
MPTCP_CAP_DENYJOIN_ID0 is received.

Also add a new flag remote_deny_join_id0 in struct mptcp_pm_data. When the
flag deny_join_id0 is set, set this remote_deny_join_id0 flag.

In mptcp_pm_create_subflow_or_signal_addr, if the remote_deny_join_id0 flag
is set, and the remote address id is zero, stop this connection.
Suggested-by: Florian Westphal <fw@strlen.de>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch defined a new flag MPTCP_CAP_DENY_JOIN_ID0 for the third bit,
labeled "C" of the MP_CAPABLE option.

Add a new flag allow_join_id0 in struct mptcp_out_options. If this flag is
set, send out the MP_CAPABLE option with the flag MPTCP_CAP_DENY_JOIN_ID0.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch added a new sysctl, named allow_join_initial_addr_port, to
control whether allow peers to send join requests to the IP address and
port number used by the initial subflow.
Suggested-by: Florian Westphal <fw@strlen.de>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Xin Long says:

====================
sctp: implement RFC8899: Packetization Layer Path MTU Discovery for SCTP transport

Overview(From RFC8899):

  In contrast to PMTUD, Packetization Layer Path MTU Discovery
  (PLPMTUD) [RFC4821] introduces a method that does not rely upon
  reception and validation of PTB messages.  It is therefore more
  robust than Classical PMTUD.  This has become the recommended
  approach for implementing discovery of the PMTU [BCP145].

  It uses a general strategy in which the PL sends probe packets to
  search for the largest size of unfragmented datagram that can be sent
  over a network path.  Probe packets are sent to explore using a
  larger packet size.  If a probe packet is successfully delivered (as
  determined by the PL), then the PLPMTU is raised to the size of the
  successful probe.  If a black hole is detected (e.g., where packets
  of size PLPMTU are consistently not received), the method reduces the
  PLPMTU.

SCTP Probe Packets:

  As the RFC suggested, the probe packets consist of an SCTP common header
  followed by a HEARTBEAT chunk and a PAD chunk. The PAD chunk is used to
  control the length of the probe packet.  The HEARTBEAT chunk is used to
  trigger the sending of a HEARTBEAT ACK chunk to confirm this probe on
  the HEARTBEAT sender.

  The HEARTBEAT chunk also carries a Heartbeat Information parameter that
  includes the probe size to help an implementation associate a HEARTBEAT
  ACK with the size of probe that was sent. The sender use the nonce and
  the probe size to verify the information returned.

Detailed Implementation on SCTP:

                       +------+
              +------->| Base |-----------------+ Connectivity
              |        +------+                 | or BASE_PLPMTU
              |           |                     | confirmation failed
              |           |                     v
              |           | Connectivity    +-------+
              |           | and BASE_PLPMTU | Error |
              |           | confirmed       +-------+
              |           |                     | Consistent
              |           v                     | connectivity
   Black Hole |       +--------+                | and BASE_PLPMTU
    detected  |       | Search |<---------------+ confirmed
              |       +--------+
              |          ^  |
              |          |  |
              |    Raise |  | Search
              |    timer |  | algorithm
              |  expired |  | completed
              |          |  |
              |          |  v
              |   +-----------------+
              +---| Search Complete |
                  +-----------------+

  When PLPMTUD is enabled, it's in Base state, and starts to probe with
  BASE_PLPMTU (1200). If this probe succeeds, it goes to Search state;
  If this probe fails, it goes to Error state under which pl.pmtu goes
  down to MIN_PLPMTU (512) and keeps probing with BASE_PLPMTU until it
  succeeds and goes to Search state.

  During the Search state, the probe size is growing by a Big step (32)
  every time when the last probe succeeds at the beginning. Once a probe
  (such as 1420) fails after trying MAX_PROBES (3) times, the probe_size
  goes back to the last one (1420 - 32 = 1388), meanwhile 'probe_high'
  is set to 1420 and the growing step becomes a Small one (4). Then the
  probe is continuing with a Small step grown each round. Until it gets
  the optimal size (such as 1400) when probe with its next probe size
  (1404) fails, it sync this size to pathmtu and goes to Complete state.

  In Complete state, it will only does a probe check for the pathmtu just
  set, if it fails, which means a Black Hole is detected and it goes back
  to Base state. If it succeeds, it goes back to Search state again, and
  probe is continuing with growing a Small step (1400 + 4). If this probe
  fails, probe_high is set and goes back to 1388 and then Complete state,
  which is kind of a loop normally. However if the env's pathmtu changes
  to a big size somehow, this probe will succeed and then probe continues
  with growing a Big step (1400 + 32) each round until another probe fails.

PTB Messages Process:

  PLPMTUD doesn't rely on these package to find the pmtu, and shouldn't
  trust it either. When processing them, it only changes the probe_size
  to PL_PTB_SIZE(info - hlen) if 'pl.pmtu < PL_PTB_SIZE < the current
  probe_size' druing Search state. As this could help probe_size to get
  to the optimal size faster, for exmaple:

  pl.pmtu = 1388, probe_size = 1420, while the env's pathmtu = 1400.
  When probe_size is 1420, a Toobig packet with 1400 comes back. If probe
  size changes to use 1400, it will save quite a few rounds to get there.
  But of course after having this value, PLPMTUD will still verify it on
  its own before using it.

Patches:

  - Patch 1-6: introduce some new constants/variables from the RFC, systcl
    and members in transport, APIs for the following patches, chunks and
    a timer for the probe sending and some codes for the probe receiving.

  - Patch 7-9: implement the state transition on the tx path, rx path and
    toobig ICMP packet processing. This is the main algorithm part.

  - Patch 10: activate this feature

  - Patch 11-14: improve the process for ICMP packets for SCTP over UDP,
    so that it can also be covered by this feature.

Tests:

  - do sysctl and setsockopt tests for this feature's enabling and disabling.

  - get these pr_debug points for this feature by
      # cat /sys/kernel/debug/dynamic_debug/control | grep PLP
    and enable them on kernel dynamic debug, then play with the pathmtu and
    check if the state transition and plpmtu change match the RFC.

  - do the above tests for SCTP over IPv4/IPv6 and SCTP over UDP.

v1->v2:
  - See Patch 06/14.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Previously, sctp over udp was using udp tunnel's icmp err process, which
only does sk lookup on sctp side. However for sctp's icmp error process,
there are more things to do, like syncing assoc pmtu/retransmit packets
for toobig type err, and starting proto_unreach_timer for unreach type
err etc.

Now after adding PLPMTUD, which also requires to process toobig type err
on sctp side. This patch is to process icmp err on sctp side by parsing
the type/code/info in .encap_err_lookup and call sctp's icmp processing
functions. Note as the 'redirect' err process needs to know the outer
ip(v6) header's, we have to leave it to udp(v6)_err to handle it.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch is to extract sctp_v4_err_handle() from sctp_v4_err() to
only handle the icmp err after the sock lookup, and it also makes
the code clearer.

sctp_v4_err_handle() will be used in sctp over udp's err handling
in the following patch.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch is to extract sctp_v6_err_handle() from sctp_v6_err() to
only handle the icmp err after the sock lookup, and it also makes
the code clearer.

sctp_v6_err_handle() will be used in sctp over udp's err handling
in the following patch.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Same as in tcp_v6_err() and __udp6_lib_err(), there's no need to
hold idev in sctp_v6_err(), so just call __in6_dev_get() instead.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sctp_transport_pl_reset() is called whenever any of these 3 members in
transport is changed:

  - probe_interval
  - param_flags & SPP_PMTUD_ENABLE
  - state == ACTIVE

If all are true, start the PLPMTUD when it's not yet started. If any of
these is false, stop the PLPMTUD when it's already running.

sctp_transport_pl_update() is called when the transport dst has changed.
It will restart the PLPMTUD probe. Again, the pathmtu won't change but
use the dst's mtu until the Search phase is done.

Note that after using PLPMTUD, the pathmtu is only initialized with the
dst mtu when the transport dst changes. At other time it is updated by
pl.pmtu. So sctp_transport_pmtu_check() will be called only when PLPMTUD
is disabled in sctp_packet_config().

After this patch, the PLPMTUD feature from RFC8899 will be activated
and can be used by users.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

PLPMTUD will short-circuit the old process for icmp TOOBIG packets.
This part is described in rfc8899#section-4.6.2 (PL_PTB_SIZE =
PTB_SIZE - other_headers_len). Note that from rfc8899#section-5.2
State Machine, each case below is for some specific states only:

  a) PL_PTB_SIZE < MIN_PLPMTU || PL_PTB_SIZE >= PROBED_SIZE,
     discard it, for any state

  b) MIN_PLPMTU < PL_PTB_SIZE < BASE_PLPMTU,
     Base -> Error, for Base state

  c) BASE_PLPMTU <= PL_PTB_SIZE < PLPMTU,
     Search -> Base or Complete -> Base, for Search and Complete states.

  d) PLPMTU < PL_PTB_SIZE < PROBED_SIZE,
     set pl.probe_size to PL_PTB_SIZE then verify it, for Search state.

The most important one is case d), which will help find the optimal
fast during searching. Like when pathmtu = 1392 for SCTP over IPv4,
the search will be (20 is iphdr_len):

  1. probe with 1200 - 20
  2. probe with 1232 - 20
  3. probe with 1264 - 20
  ...
  7. probe with 1388 - 20
  8. probe with 1420 - 20

When sending the probe with 1420 - 20, TOOBIG may come with PL_PTB_SIZE =
1392 - 20. Then it matches case d), and saves some rounds to try with the
1392 - 20 probe. But of course, PLPMTUD doesn't trust TOOBIG packets, and
it will go back to the common searching once the probe with the new size
can't be verified.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

As described in rfc8899#section-5.2, when a probe succeeds, there might
be the following state transitions:

  - Base -> Search, occurs when probe succeeds with BASE_PLPMTU,
    pl.pmtu is not changing,
    pl.probe_size increases by SCTP_PL_BIG_STEP,

  - Error -> Search, occurs when probe succeeds with BASE_PLPMTU,
    pl.pmtu is changed from SCTP_MIN_PLPMTU to SCTP_BASE_PLPMTU,
    pl.probe_size increases by SCTP_PL_BIG_STEP.

  - Search -> Search Complete, occurs when probe succeeds with the probe
    size SCTP_MAX_PLPMTU less than pl.probe_high,
    pl.pmtu is not changing, but update *pathmtu* with it,
    pl.probe_size is set back to pl.pmtu to double check it.

  - Search Complete -> Search, occurs when probe succeeds with the probe
    size equal to pl.pmtu,
    pl.pmtu is not changing,
    pl.probe_size increases by SCTP_PL_MIN_STEP.

So search process can be described as:

 1. When it just enters 'Search' state, *pathmtu* is not updated with
    pl.pmtu, and probe_size increases by a big step (SCTP_PL_BIG_STEP)
    each round.

 2. Until pl.probe_high is set when a probe fails, and probe_size
    decreases back to pl.pmtu, as described in the last patch.

 3. When the probe with the new size succeeds, probe_size changes to
    increase by a small step (SCTP_PL_MIN_STEP) due to pl.probe_high
    is set.

 4. Until probe_size is next to pl.probe_high, the searching finishes and
    it goes to 'Complete' state and updates *pathmtu* with pl.pmtu, and
    then probe_size is set to pl.pmtu to confirm by once more probe.

 5. This probe occurs after "30 * probe_inteval", a much longer time than
    that in Search state. Once it is done it goes to 'Search' state again
    with probe_size increased by SCTP_PL_MIN_STEP.

As we can see above, during the searching, pl.pmtu changes while *pathmtu*
doesn't. *pathmtu* is only updated when the search finishes by which it
gets an optimal value for it. A big step is used at the beginning until
it gets close to the optimal value, then it changes to a small step until
it has this optimal value.

The small step is also used in 'Complete' until it goes to 'Search' state
again and the probe with 'pmtu + the small step' succeeds, which means a
higher size could be used. Then probe_size changes to increase by a big
step again until it gets close to the next optimal value.

Note that anytime when black hole is detected, it goes directly to 'Base'
state with pl.pmtu set to SCTP_BASE_PLPMTU, as described in the last patch.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>