sfrench/cifs-2.6.git
2 months agoionic: rearrange ionic_qcq
Shannon Nelson [Wed, 6 Mar 2024 23:29:55 +0000 (15:29 -0800)]
ionic: rearrange ionic_qcq

Rearange a few fields for better cache use and to put the
flags field up into the first cacheline rather than the last.

    struct ionic_qcq
Before: /* size: 2176, cachelines: 34, members: 23 */
After:  /* size: 2112, cachelines: 33, members: 23 */

Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: carry idev in ionic_cq struct
Shannon Nelson [Wed, 6 Mar 2024 23:29:54 +0000 (15:29 -0800)]
ionic: carry idev in ionic_cq struct

Remove the idev field from ionic_queue, which saves us a
bit of space, and add it into ionic_cq where there's room
within some cacheline padding.  Use this pointer rather
than doing a multi level reference from lif->ionic.

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: refactor skb building
Shannon Nelson [Wed, 6 Mar 2024 23:29:53 +0000 (15:29 -0800)]
ionic: refactor skb building

The existing ionic_rx_frags() code is a bit of a mess and can
be cleaned up by unrolling the first frag/header setup from
the loop, then reworking the do-while-loop into a for-loop.  We
rename the function to a more descriptive ionic_rx_build_skb().
We also change a couple of related variable names for readability.

Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: fold adminq clean into service routine
Shannon Nelson [Wed, 6 Mar 2024 23:29:52 +0000 (15:29 -0800)]
ionic: fold adminq clean into service routine

Since the AdminQ clean is a simple action called from only
one place, fold it back into the service routine.

Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: use specialized desc info structs
Shannon Nelson [Wed, 6 Mar 2024 23:29:51 +0000 (15:29 -0800)]
ionic: use specialized desc info structs

Make desc_info structure specific to the queue type, which
allows us to cut down the Rx and AdminQ descriptor sizes by
not including all the fields needed for the Tx desriptors.

Before:
    struct ionic_desc_info {
/* size: 464, cachelines: 8, members: 6 */

After:
    struct ionic_tx_desc_info {
/* size: 464, cachelines: 8, members: 6 */
    struct ionic_rx_desc_info {
/* size: 224, cachelines: 4, members: 2 */
    struct ionic_admin_desc_info {
/* size: 8, cachelines: 1, members: 1 */

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: remove the cq_info to save more memory
Shannon Nelson [Wed, 6 Mar 2024 23:29:50 +0000 (15:29 -0800)]
ionic: remove the cq_info to save more memory

With a little simple math we don't need another struct array to
find the completion structs, so we can remove the ionic_cq_info
altogether.  This doesn't really save anything in the ionic_cq
since it gets padded out to the cacheline, but it does remove
the parallel array allocation of 8 * num_descriptors, or about
8 Kbytes per queue in a default configuration.

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: remove callback pointer from desc_info
Shannon Nelson [Wed, 6 Mar 2024 23:29:49 +0000 (15:29 -0800)]
ionic: remove callback pointer from desc_info

By reworking the queue service routines to have their own
servicing loops we can remove the cb pointer from desc_info
to save another 8 bytes per descriptor,

This simplifies some of the queue handling indirection and makes
the code a little easier to follow, and keeps service code in
one place rather than jumping between code files.

   struct ionic_desc_info
Before:  /* size: 472, cachelines: 8, members: 7 */
After:   /* size: 464, cachelines: 8, members: 6 */

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: move adminq-notifyq handling to main file
Shannon Nelson [Wed, 6 Mar 2024 23:29:48 +0000 (15:29 -0800)]
ionic: move adminq-notifyq handling to main file

Move the AdminQ and NotifyQ queue handling to ionic_main.c with
the rest of the adminq code.

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: drop q mapping
Shannon Nelson [Wed, 6 Mar 2024 23:29:47 +0000 (15:29 -0800)]
ionic: drop q mapping

Now that we're not using desc_info pointers mapped in every q
we can simplify and drop the unnecessary utility functions.

Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoionic: remove desc, sg_desc and cmb_desc from desc_info
Shannon Nelson [Wed, 6 Mar 2024 23:29:46 +0000 (15:29 -0800)]
ionic: remove desc, sg_desc and cmb_desc from desc_info

Remove the struct pointers from desc_info to use less space.
Instead of pointers in every desc_info to its descriptor,
we can use the queue descriptor index to find the individual
desc, desc_info, and sgl structs in their parallel arrays.

   struct ionic_desc_info
Before:  /* size: 496, cachelines: 8, members: 10 */
After:   /* size: 472, cachelines: 8, members: 7 */

Suggested-by: Neel Patel <npatel2@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
David S. Miller [Fri, 8 Mar 2024 11:43:21 +0000 (11:43 +0000)]
Merge branch '40GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2024-03-06 (iavf, i40e, ixgbe)

This series contains updates to iavf, i40e, and ixgbe drivers.

Alexey Kodanev removes duplicate calls related to cloud filters on iavf
and unnecessary null checks on i40e.

Maciej adds helper functions for common code relating to updating
statistics for ixgbe.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoAdd Jeff Kirsher to .get_maintainer.ignore
Jakub Kicinski [Wed, 6 Mar 2024 15:47:03 +0000 (07:47 -0800)]
Add Jeff Kirsher to .get_maintainer.ignore

Jeff was retired as the Intel driver maintainer in
commit 6667df916fce ("MAINTAINERS: Update MAINTAINERS for
Intel ethernet drivers"), and his address bounces.
But he has signed-off a lot of patches over the years
so get_maintainer insists on CCing him.

We haven't heard from him since he left Intel, so remapping
the address via mailmap is also pointless. Add to ignored
addresses.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoMerge branch 'ipv6-lockless-dump-addrs'
David S. Miller [Fri, 8 Mar 2024 11:15:36 +0000 (11:15 +0000)]
Merge branch 'ipv6-lockless-dump-addrs'

Eric Dumazet says:

====================
ipv6: lockless inet6_dump_addr()

This series removes RTNL locking to dump ipv6 addresses.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoipv6: remove RTNL protection from inet6_dump_addr()
Eric Dumazet [Wed, 6 Mar 2024 15:51:44 +0000 (15:51 +0000)]
ipv6: remove RTNL protection from inet6_dump_addr()

We can now remove RTNL acquisition while running
inet6_dump_addr(), inet6_dump_ifmcaddr()
and inet6_dump_ifacaddr().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoipv6: use xa_array iterator to implement inet6_dump_addr()
Eric Dumazet [Wed, 6 Mar 2024 15:51:43 +0000 (15:51 +0000)]
ipv6: use xa_array iterator to implement inet6_dump_addr()

inet6_dump_addr() can use the new xa_array iterator
for better scalability.

Make it ready for RCU-only protection.
RTNL use is removed in the following patch.

Also properly return 0 at the end of a dump to avoid
and extra recvmsg() to get NLMSG_DONE.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoipv6: make in6_dump_addrs() lockless
Eric Dumazet [Wed, 6 Mar 2024 15:51:42 +0000 (15:51 +0000)]
ipv6: make in6_dump_addrs() lockless

in6_dump_addrs() is called with RCU protection.

There is no need holding idev->lock to iterate through unicast addresses.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoipv6: make inet6_fill_ifaddr() lockless
Eric Dumazet [Wed, 6 Mar 2024 15:51:41 +0000 (15:51 +0000)]
ipv6: make inet6_fill_ifaddr() lockless

Make inet6_fill_ifaddr() lockless, and add approriate annotations
on ifa->tstamp, ifa->valid_lft, ifa->preferred_lft, ifa->ifa_proto
and ifa->rt_priority.

Also constify 2nd argument of inet6_fill_ifaddr(), inet6_fill_ifmcaddr()
and inet6_fill_ifacaddr().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoMerge tag 'ipsec-next-2024-03-06' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 8 Mar 2024 10:56:05 +0000 (10:56 +0000)]
Merge tag 'ipsec-next-2024-03-06' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
1) Introduce forwarding of ICMP Error messages. That is specified
   in RFC 4301 but was never implemented. From Antony Antony.

2) Use KMEM_CACHE instead of kmem_cache_create in xfrm6_tunnel_init()
   and xfrm_policy_init(). From Kunwu Chan.

3) Do not allocate stats in the xfrm interface driver, this can be done
   on net core now. From Breno Leitao.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoMerge branch 'nexthop-group-stats'
David S. Miller [Fri, 8 Mar 2024 10:35:48 +0000 (10:35 +0000)]
Merge branch 'nexthop-group-stats'

Petr Machata says:

====================
Support for nexthop group statistics

ECMP is a fundamental component in L3 designs. However, it's fragile. Many
factors influence whether an ECMP group will operate as intended: hash
policy (i.e. the set of fields that contribute to ECMP hash calculation),
neighbor validity, hash seed (which might lead to polarization) or the type
of ECMP group used (hash-threshold or resilient).

At the same time, collection of statistics that would help an operator
determine that the group performs as desired, is difficult.

A solution that we present in this patchset is to add counters to next hop
group entries. For SW-datapath deployments, this will on its own allow
collection and evaluation of relevant statistics. For HW-datapath
deployments, we further add a way to request that HW counters be installed
for a given group, in-kernel interfaces to collect the HW statistics, and
netlink interfaces to query them.

For example:

    # ip nexthop replace id 4000 group 4001/4002 hw_stats on

    # ip -s -d nexthop show id 4000
    id 4000 group 4001/4002 scope global proto unspec offload hw_stats on used on
      stats:
        id 4001 packets 5002 packets_hw 5000
        id 4002 packets 4999 packets_hw 4999

The point of the patchset is visibility of ECMP balance, and that is
influenced by packet headers, not their payload. Correspondingly, we only
include packet counters in the statistics, not byte counters.

We also decided to model HW statistics as a nexthop group attribute, not an
arbitrary nexthop one. The latter would count any traffic going through a
given nexthop, regardless of which ECMP group it is in, or any at all. The
reason is again hat the point of the patchset is ECMP balance visibility,
not arbitrary inspection of how busy a particular nexthop is.
Implementation of individual-nexthop statistics is certainly possible, and
could well follow the general approach we are taking in this patchset.
For resilient groups, per-bucket statistics could be done in a similar
manner as well.

This patchset contains the core code. mlxsw support will be sent in a
follow-up patch set.

This patchset progresses as follows:

- Patches #1 and #2 add support for a new next-hop object attribute,
  NHA_OP_FLAGS. That is meant to carry various op-specific signaling, in
  particular whether SW- and HW-collected nexthop stats should be part of
  the get or dump response. The idea is to avoid wasting message space, and
  time for collection of HW statistics, when the values are not needed.

- Patches #3 and #4 add SW-datapath stats and corresponding UAPI.

- Patches #5, #6 and #7 add support fro HW-datapath stats and UAPI.
  Individual drivers still need to contribute the appropriate HW-specific
  support code.

v4:
- Patch #2:
    - s/nla_get_bitfield32/nla_get_u32/ in __nh_valid_dump_req().

v3:
- Patch #3:
    - Convert to u64_stats_t
- Patch #4:
    - Give a symbolic name to the set of all valid dump flags
      for the NHA_OP_FLAGS attribute.
    - Convert to u64_stats_t
- Patch #6:
    - Use a named constant for the NHA_HW_STATS_ENABLE policy.

v2:
- Patch #2:
    - Change OP_FLAGS to u32, enforce through NLA_POLICY_MASK
- Patch #3:
    - Set err on nexthop_create_group() error path
- Patch #4:
    - Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS
    - Rename jump target in nla_put_nh_group_stats() to avoid
      having to rename further in the patchset.
- Patch #7:
    - Use uint to encode NHA_GROUP_STATS_ENTRY_PACKETS_HW
    - Do not cancel outside of nesting in nla_put_nh_group_stats()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Expose nexthop group HW stats to user space
Ido Schimmel [Wed, 6 Mar 2024 12:49:21 +0000 (13:49 +0100)]
net: nexthop: Expose nexthop group HW stats to user space

Add netlink support for reading NH group hardware stats.

Stats collection is done through a new notifier,
NEXTHOP_EVENT_HW_STATS_REPORT_DELTA. Drivers that implement HW counters for
a given NH group are thereby asked to collect the stats and report back to
core by calling nh_grp_hw_stats_report_delta(). This is similar to what
netdevice L3 stats do.

Besides exposing number of packets that passed in the HW datapath, also
include information on whether any driver actually realizes the counters.
The core can tell based on whether it got any _report_delta() reports from
the drivers. This allows enabling the statistics at the group at any time,
with drivers opting into supporting them. This is also in line with what
netdevice L3 stats are doing.

So as not to waste time and space, tie the collection and reporting of HW
stats with a new op flag, NHA_OP_FLAG_DUMP_HW_STATS.

Co-developed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Kees Cook <keescook@chromium.org> # For the __counted_by bits
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Add ability to enable / disable hardware statistics
Ido Schimmel [Wed, 6 Mar 2024 12:49:20 +0000 (13:49 +0100)]
net: nexthop: Add ability to enable / disable hardware statistics

Add netlink support for enabling collection of HW statistics on nexthop
groups.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Add hardware statistics notifications
Ido Schimmel [Wed, 6 Mar 2024 12:49:19 +0000 (13:49 +0100)]
net: nexthop: Add hardware statistics notifications

Add hw_stats field to several notifier structures to communicate to the
drivers that HW statistics should be configured for nexthops within a given
group.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Expose nexthop group stats to user space
Ido Schimmel [Wed, 6 Mar 2024 12:49:18 +0000 (13:49 +0100)]
net: nexthop: Expose nexthop group stats to user space

Add netlink support for reading NH group stats.

This data is only for statistics of the traffic in the SW datapath. HW
nexthop group statistics will be added in the following patches.

Emission of the stats is keyed to a new op_stats flag to avoid cluttering
the netlink message with stats if the user doesn't need them:
NHA_OP_FLAG_DUMP_STATS.

Co-developed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Add nexthop group entry stats
Ido Schimmel [Wed, 6 Mar 2024 12:49:17 +0000 (13:49 +0100)]
net: nexthop: Add nexthop group entry stats

Add nexthop group entry stats to count the number of packets forwarded
via each nexthop in the group. The stats will be exposed to user space
for better data path observability in the next patch.

The per-CPU stats pointer is placed at the beginning of 'struct
nh_grp_entry', so that all the fields accessed for the data path reside
on the same cache line:

struct nh_grp_entry {
        struct nexthop *           nh;                   /*     0     8 */
        struct nh_grp_entry_stats * stats;               /*     8     8 */
        u8                         weight;               /*    16     1 */

        /* XXX 7 bytes hole, try to pack */

        union {
                struct {
                        atomic_t   upper_bound;          /*    24     4 */
                } hthr;                                  /*    24     4 */
                struct {
                        struct list_head uw_nh_entry;    /*    24    16 */
                        u16        count_buckets;        /*    40     2 */
                        u16        wants_buckets;        /*    42     2 */
                } res;                                   /*    24    24 */
        };                                               /*    24    24 */
        struct list_head           nh_list;              /*    48    16 */
        /* --- cacheline 1 boundary (64 bytes) --- */
        struct nexthop *           nh_parent;            /*    64     8 */

        /* size: 72, cachelines: 2, members: 6 */
        /* sum members: 65, holes: 1, sum holes: 7 */
        /* last cacheline: 8 bytes */
};

Co-developed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Add NHA_OP_FLAGS
Petr Machata [Wed, 6 Mar 2024 12:49:16 +0000 (13:49 +0100)]
net: nexthop: Add NHA_OP_FLAGS

In order to add per-nexthop statistics, but still not increase netlink
message size for consumers that do not care about them, there needs to be a
toggle through which the user indicates their desire to get the statistics.
To that end, add a new attribute, NHA_OP_FLAGS. The idea is to be able to
use the attribute for carrying of arbitrary operation-specific flags, i.e.
not make it specific for get / dump.

Add the new attribute to get and dump policies, but do not actually allow
any flags yet -- those will come later as the flags themselves are defined.
Add the necessary parsing code.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: nexthop: Adjust netlink policy parsing for a new attribute
Petr Machata [Wed, 6 Mar 2024 12:49:15 +0000 (13:49 +0100)]
net: nexthop: Adjust netlink policy parsing for a new attribute

A following patch will introduce a new attribute, op-specific flags to
adjust the behavior of an operation. Different operations will recognize
different flags.

- To make the differentiation possible, stop sharing the policies for get
  and del operations.

- To allow querying for presence of the attribute, have all the attribute
  arrays sized to NHA_MAX, regardless of what is permitted by policy, and
  pass the corresponding value to nlmsg_parse() as well.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoocteontx2-pf: Add TC flower offload support for TCP flags
Sai Krishna [Tue, 5 Mar 2024 18:16:06 +0000 (23:46 +0530)]
octeontx2-pf: Add TC flower offload support for TCP flags

This patch adds TC offload support for matching TCP flags
from TCP header.

Example usage:
tc qdisc add dev eth0 ingress

TC rule to drop the TCP SYN packets:
tc filter add dev eth0 ingress protocol ip flower ip_proto tcp tcp_flags
0x02/0x3f skip_sw action drop

Signed-off-by: Sai Krishna <saikrishnag@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: Add skb addr and sock addr to arguments of tracepoint tcp_probe.
fuyuanli [Tue, 5 Mar 2024 03:04:17 +0000 (11:04 +0800)]
tcp: Add skb addr and sock addr to arguments of tracepoint tcp_probe.

It is useful to expose skb addr and sock addr to user in tracepoint
tcp_probe, so that we can get more information while monitoring
receiving of tcp data, by ebpf or other ways.

For example, we need to identify a packet by seq and end_seq when
calculate transmit latency between layer 2 and layer 4 by ebpf, but which is
not available in tcp_probe, so we can only use kprobe hooking
tcp_rcv_established to get them. But we can use tcp_probe directly if skb
addr and sock addr are available, which is more efficient.

Signed-off-by: fuyuanli <fuyuanli@didiglobal.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: dqs: add NIC stall detector based on BQL
Jakub Kicinski [Mon, 4 Mar 2024 14:08:47 +0000 (06:08 -0800)]
net: dqs: add NIC stall detector based on BQL

softnet_data->time_squeeze is sometimes used as a proxy for
host overload or indication of scheduling problems. In practice
this statistic is very noisy and has hard to grasp units -
e.g. is 10 squeezes a second to be expected, or high?

Delaying network (NAPI) processing leads to drops on NIC queues
but also RTT bloat, impacting pacing and CA decisions.
Stalls are a little hard to detect on the Rx side, because
there may simply have not been any packets received in given
period of time. Packet timestamps help a little bit, but
again we don't know if packets are stale because we're
not keeping up or because someone (*cough* cgroups)
disabled IRQs for a long time.

We can, however, use Tx as a proxy for Rx stalls. Most drivers
use combined Rx+Tx NAPIs so if Tx gets starved so will Rx.
On the Tx side we know exactly when packets get queued,
and completed, so there is no uncertainty.

This patch adds stall checks to BQL. Why BQL? Because
it's a convenient place to add such checks, already
called by most drivers, and it has copious free space
in its structures (this patch adds no extra cache
references or dirtying to the fast path).

The algorithm takes one parameter - max delay AKA stall
threshold and increments a counter whenever NAPI got delayed
for at least that amount of time. It also records the length
of the longest stall.

To be precise every time NAPI has not polled for at least
stall thrs we check if there were any Tx packets queued
between last NAPI run and now - stall_thrs/2.

Unlike the classic Tx watchdog this mechanism does not
ignore stalls caused by Tx being disabled, or loss of link.
I don't think the check is worth the complexity, and
stall is a stall, whether due to host overload, flow
control, link down... doesn't matter much to the application.

We have been running this detector in production at Meta
for 2 years, with the threshold of 8ms. It's the lowest
value where false positives become rare. There's still
a constant stream of reported stalls (especially without
the ksoftirqd deferral patches reverted), those who like
their stall metrics to be 0 may prefer higher value.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agonet: chelsio: remove unused function calc_tx_descs
Colin Ian King [Thu, 7 Mar 2024 11:22:37 +0000 (11:22 +0000)]
net: chelsio: remove unused function calc_tx_descs

The inlined helper function calc_tx_descs is not used and is redundant.
Remove it.

Cleans up clang scan build warning:
drivers/net/ethernet/chelsio/cxgb4/sge.c:814:28: warning: unused
function 'calc_tx_descs' [-Wunused-function]

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agoMerge branch 'netdev-add-per-queue-statistics'
Jakub Kicinski [Fri, 8 Mar 2024 05:13:28 +0000 (21:13 -0800)]
Merge branch 'netdev-add-per-queue-statistics'

Jakub Kicinski says:

====================
netdev: add per-queue statistics

Per queue stats keep coming up, so it's about time someone laid
the foundation. This series adds the uAPI, a handful of stats
and a sample support for bnxt. It's not very comprehensive in
terms of stat types or driver support. The expectation is that
the support will grow organically. If we have the basic pieces
in place it will be easy for reviewers to request new stats,
or use of the API in place of ethtool -S.

See patch 3 for sample output.

v2: https://lore.kernel.org/all/20240229010221.2408413-1-kuba@kernel.org/
v1: https://lore.kernel.org/all/20240226211015.1244807-1-kuba@kernel.org/
rfc: https://lore.kernel.org/all/20240222223629.158254-1-kuba@kernel.org/
====================

Link: https://lore.kernel.org/r/20240306195509.1502746-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoeth: bnxt: support per-queue statistics
Jakub Kicinski [Wed, 6 Mar 2024 19:55:09 +0000 (11:55 -0800)]
eth: bnxt: support per-queue statistics

Support per-queue statistics API in bnxt.

$ ethtool -S eth0
NIC statistics:
     [0]: rx_ucast_packets: 1418
     [0]: rx_mcast_packets: 178
     [0]: rx_bcast_packets: 0
     [0]: rx_discards: 0
     [0]: rx_errors: 0
     [0]: rx_ucast_bytes: 1141815
     [0]: rx_mcast_bytes: 16766
     [0]: rx_bcast_bytes: 0
     [0]: tx_ucast_packets: 1734
...

$ ./cli.py --spec netlink/specs/netdev.yaml \
   --dump qstats-get --json '{"scope": "queue"}'
[{'ifindex': 2,
  'queue-id': 0,
  'queue-type': 'rx',
  'rx-alloc-fail': 0,
  'rx-bytes': 1164931,
  'rx-packets': 1641},
...
 {'ifindex': 2,
  'queue-id': 0,
  'queue-type': 'tx',
  'tx-bytes': 631494,
  'tx-packets': 1771},
...

Reset the per queue counters:
$ ethtool -L eth0 combined 4

Inspect again:

$ ./cli.py --spec netlink/specs/netdev.yaml \
   --dump qstats-get --json '{"scope": "queue"}'
[{'ifindex': 2,
  'queue-id': 0,
  'queue-type': 'rx',
  'rx-alloc-fail': 0,
  'rx-bytes': 32397,
  'rx-packets': 145},
...
 {'ifindex': 2,
  'queue-id': 0,
  'queue-type': 'tx',
  'tx-bytes': 37481,
  'tx-packets': 196},
...

$ ethtool -S eth0 | head
NIC statistics:
     [0]: rx_ucast_packets: 174
     [0]: rx_mcast_packets: 3
     [0]: rx_bcast_packets: 0
     [0]: rx_discards: 0
     [0]: rx_errors: 0
     [0]: rx_ucast_bytes: 37151
     [0]: rx_mcast_bytes: 267
     [0]: rx_bcast_bytes: 0
     [0]: tx_ucast_packets: 267
...

Totals are still correct:

$ ./cli.py --spec netlink/specs/netdev.yaml --dump qstats-get
[{'ifindex': 2,
  'rx-alloc-fail': 0,
  'rx-bytes': 281949995,
  'rx-packets': 216524,
  'tx-bytes': 52694905,
  'tx-packets': 75546}]
$ ip -s link show dev eth0
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP mode DEFAULT group default qlen 1000
    link/ether 14:23:f2:61:05:40 brd ff:ff:ff:ff:ff:ff
    RX:  bytes packets errors dropped  missed   mcast
     282519546  218100      0       0       0     516
    TX:  bytes packets errors dropped carrier collsns
      53323054   77674      0       0       0       0

Acked-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Amritha Nambiar <amritha.nambiar@intel.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240306195509.1502746-4-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonetdev: add queue stat for alloc failures
Jakub Kicinski [Wed, 6 Mar 2024 19:55:08 +0000 (11:55 -0800)]
netdev: add queue stat for alloc failures

Rx alloc failures are commonly counted by drivers.
Support reporting those via netdev-genl queue stats.

Acked-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Amritha Nambiar <amritha.nambiar@intel.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240306195509.1502746-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonetdev: add per-queue statistics
Jakub Kicinski [Wed, 6 Mar 2024 19:55:07 +0000 (11:55 -0800)]
netdev: add per-queue statistics

The ethtool-nl family does a good job exposing various protocol
related and IEEE/IETF statistics which used to get dumped under
ethtool -S, with creative names. Queue stats don't have a netlink
API, yet, and remain a lion's share of ethtool -S output for new
drivers. Not only is that bad because the names differ driver to
driver but it's also bug-prone. Intuitively drivers try to report
only the stats for active queues, but querying ethtool stats
involves multiple system calls, and the number of stats is
read separately from the stats themselves. Worse still when user
space asks for values of the stats, it doesn't inform the kernel
how big the buffer is. If number of stats increases in the meantime
kernel will overflow user buffer.

Add a netlink API for dumping queue stats. Queue information is
exposed via the netdev-genl family, so add the stats there.
Support per-queue and sum-for-device dumps. Latter will be useful
when subsequent patches add more interesting common stats than
just bytes and packets.

The API does not currently distinguish between HW and SW stats.
The expectation is that the source of the stats will either not
matter much (good packets) or be obvious (skb alloc errors).

Acked-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Amritha Nambiar <amritha.nambiar@intel.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240306195509.1502746-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge branch 'net-group-together-hot-data'
Jakub Kicinski [Fri, 8 Mar 2024 05:12:45 +0000 (21:12 -0800)]
Merge branch 'net-group-together-hot-data'

Eric Dumazet says:

====================
net: group together hot data

While our recent structure reorganizations were focused
on increasing max throughput, there is still an
area where improvements are much needed.

In many cases, a cpu handles one packet at a time,
instead of a nice batch.

Hardware interrupt.
 -> Software interrupt.
   -> Network/Protocol stacks.

If the cpu was idle or busy in other layers,
it has to pull many cache lines.

This series adds a new net_hotdata structure, where
some critical (and read-mostly) data used in
rx and tx path is packed in a small number of cache lines.

Synthetic benchmarks will not see much difference,
but latency of single packet should improve.

net_hodata current size on 64bit is 416 bytes,
but might grow in the future.

Also move RPS definitions to a new include file.
====================

Link: https://lore.kernel.org/r/20240306160031.874438-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move rps_sock_flow_table to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:31 +0000 (16:00 +0000)]
net: move rps_sock_flow_table to net_hotdata

rps_sock_flow_table and rps_cpu_mask are used in fast path.

Move them to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-19-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: introduce include/net/rps.h
Eric Dumazet [Wed, 6 Mar 2024 16:00:30 +0000 (16:00 +0000)]
net: introduce include/net/rps.h

Move RPS related structures and helpers from include/linux/netdevice.h
and include/net/sock.h to a new include file.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-18-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoipv6: move tcp_ipv6_hash_secret and udp_ipv6_hash_secret to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:29 +0000 (16:00 +0000)]
ipv6: move tcp_ipv6_hash_secret and udp_ipv6_hash_secret to net_hotdata

Use a 32bit hole in "struct net_offload" to store
the remaining 32bit secrets used by TCPv6 and UDPv6.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-17-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoipv6: move inet6_ehash_secret and udp6_ehash_secret into net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:28 +0000 (16:00 +0000)]
ipv6: move inet6_ehash_secret and udp6_ehash_secret into net_hotdata

"struct inet6_protocol" has a 32bit hole in 32bit arches.

Use it to store the 32bit secret used by UDP and TCP,
to increase cache locality in rx path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-16-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoinet: move inet_ehash_secret and udp_ehash_secret into net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:27 +0000 (16:00 +0000)]
inet: move inet_ehash_secret and udp_ehash_secret into net_hotdata

"struct net_protocol" has a 32bit hole in 32bit arches.

Use it to store the 32bit secret used by UDP and TCP,
to increase cache locality in rx path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-15-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoinet: move tcp_protocol and udp_protocol to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:26 +0000 (16:00 +0000)]
inet: move tcp_protocol and udp_protocol to net_hotdata

These structures are read in rx path, move them to net_hotdata
for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-14-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoipv6: move tcpv6_protocol and udpv6_protocol to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:25 +0000 (16:00 +0000)]
ipv6: move tcpv6_protocol and udpv6_protocol to net_hotdata

These structures are read in rx path, move them to net_hotdata
for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-13-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoudp: move udpv4_offload and udpv6_offload to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:24 +0000 (16:00 +0000)]
udp: move udpv4_offload and udpv6_offload to net_hotdata

These structures are used in GRO and GSO paths.
Move them to net_hodata for better cache locality.

v2: udpv6_offload definition depends on CONFIG_INET=y

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-12-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move skbuff_cache(s) to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:23 +0000 (16:00 +0000)]
net: move skbuff_cache(s) to net_hotdata

skbuff_cache, skbuff_fclone_cache and skb_small_head_cache
are used in rx/tx fast paths.

Move them to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-11-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move dev_rx_weight to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:22 +0000 (16:00 +0000)]
net: move dev_rx_weight to net_hotdata

dev_rx_weight is read from process_backlog().

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-10-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move dev_tx_weight to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:21 +0000 (16:00 +0000)]
net: move dev_tx_weight to net_hotdata

dev_tx_weight is used in tx fast path.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-9-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move tcpv4_offload and tcpv6_offload to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:20 +0000 (16:00 +0000)]
net: move tcpv4_offload and tcpv6_offload to net_hotdata

These are used in TCP fast paths.

Move them into net_hotdata for better cache locality.

v2: tcpv6_offload definition depends on CONFIG_INET

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-8-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move ip_packet_offload and ipv6_packet_offload to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:19 +0000 (16:00 +0000)]
net: move ip_packet_offload and ipv6_packet_offload to net_hotdata

These structures are used in GRO and GSO paths.

v2: ipv6_packet_offload definition depends on CONFIG_INET

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-7-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move netdev_max_backlog to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:18 +0000 (16:00 +0000)]
net: move netdev_max_backlog to net_hotdata

netdev_max_backlog is used in rx fat path.

Move it to net_hodata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-6-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move ptype_all into net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:17 +0000 (16:00 +0000)]
net: move ptype_all into net_hotdata

ptype_all is used in rx/tx fast paths.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move netdev_tstamp_prequeue into net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:16 +0000 (16:00 +0000)]
net: move netdev_tstamp_prequeue into net_hotdata

netdev_tstamp_prequeue is used in rx path.

Move it to net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: move netdev_budget and netdev_budget to net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:15 +0000 (16:00 +0000)]
net: move netdev_budget and netdev_budget to net_hotdata

netdev_budget and netdev_budget are used in rx path (net_rx_action())

Move them into net_hotdata for better cache locality.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: introduce struct net_hotdata
Eric Dumazet [Wed, 6 Mar 2024 16:00:14 +0000 (16:00 +0000)]
net: introduce struct net_hotdata

Instead of spreading networking critical fields
all over the places, add a custom net_hotdata
structure so that we can precisely control its layout.

In this first patch, move :

- gro_normal_batch used in rx (GRO stack)
- offload_base used in rx and tx (GRO and TSO stacks)

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306160031.874438-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge branch 'selftests-mptcp-share-code-and-fix-shellcheck-warnings'
Jakub Kicinski [Fri, 8 Mar 2024 05:06:58 +0000 (21:06 -0800)]
Merge branch 'selftests-mptcp-share-code-and-fix-shellcheck-warnings'

Matthieu Baerts says:

====================
selftests: mptcp: share code and fix shellcheck warnings

This series cleans MPTCP selftests code.

Patch 1 stops using 'iptables-legacy' if available, but uses 'iptables',
which is likely 'iptables-nft' behind.

Patches 2, 4 and 6 move duplicated code to mptcp_lib.sh. Patch 3 is a
preparation for patch 4, and patch 5 adds generic actions at the
creation and deletion of netns.

Patches 7 to 11 disable a few shellcheck warnings, and fix the rest, so
it is easy to spot real issues later. MPTCP CI is checking that now.

Patch 12 avoids redoing some actions at init time twice, e.g. restarting
the pm events tool.

v1: https://lore.kernel.org/r/20240305-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v1-0-66618ea5504e@kernel.org
====================

Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-0-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: userspace pm: avoid relaunching pm events
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:43:01 +0000 (10:43 +0100)]
selftests: userspace pm: avoid relaunching pm events

'make_connection' is launched twice: once for IPv4, once for IPv6.

But then, the "pm_nl_ctl events" was launched a first time, killed, then
relaunched after for no particular reason.

We can then move this code, and the generation of the temp file to
exchange, to the init part, and remove extra conditions that no longer
needed.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-12-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: simult flows: fix shellcheck warnings
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:43:00 +0000 (10:43 +0100)]
selftests: mptcp: simult flows: fix shellcheck warnings

shellcheck recently helped to prevent issues. It is then good to fix the
other harmless issues in order to spot "real" ones later.

Here, two categories of warnings are now ignored:

- SC2317: Command appears to be unreachable. The cleanup() function is
  invoked indirectly via the EXIT trap.

- SC2086: Double quote to prevent globbing and word splitting. This is
  recommended, but the current usage is correct and there is no need to
  do all these modifications to be compliant with this rule.

For the modifications:

  - SC2034: ksft_skip appears unused.
  - SC2004: $/${} is unnecessary on arithmetic variables.

Now this script is shellcheck (0.9.0) compliant. We can easily spot new
issues.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-11-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: pm netlink: fix shellcheck warnings
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:42:59 +0000 (10:42 +0100)]
selftests: mptcp: pm netlink: fix shellcheck warnings

shellcheck recently helped to prevent issues. It is then good to fix the
other harmless issues in order to spot "real" ones later.

Here, two categories of warnings are now ignored:

- SC2317: Command appears to be unreachable. The cleanup() function is
  invoked indirectly via the EXIT trap.

- SC2086: Double quote to prevent globbing and word splitting. This is
  recommended, but the current usage is correct and there is no need to
  do all these modifications to be compliant with this rule.

For the modifications:

  - SC2034: ksft_skip appears unused.
  - SC2154: optstring is referenced but not assigned.
  - SC2006: Use $(...) notation instead of legacy backticks `...`.

Now this script is shellcheck (0.9.0) compliant. We can easily spot new
issues.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-10-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: sockopt: fix shellcheck warnings
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:42:58 +0000 (10:42 +0100)]
selftests: mptcp: sockopt: fix shellcheck warnings

shellcheck recently helped to prevent issues. It is then good to fix the
other harmless issues in order to spot "real" ones later.

Here, two categories of warnings are now ignored:

- SC2317: Command appears to be unreachable. The cleanup() function is
  invoked indirectly via the EXIT trap.

- SC2086: Double quote to prevent globbing and word splitting. This is
  recommended, but the current usage is correct and there is no need to
  do all these modifications to be compliant with this rule.

For the modifications:

  - SC2034: ksft_skip appears unused.
  - SC2006: Use $(...) notation instead of legacy backticks `...`.
  - SC2145: Argument mixes string and array. Use * or separate argument.

Now this script is shellcheck (0.9.0) compliant. We can easily spot new
issues.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-9-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: connect: fix shellcheck warnings
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:42:57 +0000 (10:42 +0100)]
selftests: mptcp: connect: fix shellcheck warnings

shellcheck recently helped to prevent issues. It is then good to fix the
other harmless issues in order to spot "real" ones later.

Here, two categories of warnings are now ignored:

- SC2317: Command appears to be unreachable. The cleanup() function is
  invoked indirectly via the EXIT trap.

- SC2086: Double quote to prevent globbing and word splitting. This is
  recommended, but the current usage is correct and there is no need to
  do all these modifications to be compliant with this rule.

For the modifications:

  - SC2034: ksft_skip appears unused.
  - SC2181: Check exit code directly with e.g. 'if mycmd;', not
            indirectly with $?.
  - SC2004: $/${} is unnecessary on arithmetic variables.
  - SC2155: Declare and assign separately to avoid masking return
            values.
  - SC2166: Prefer [ p ] && [ q ] as [ p -a q ] is not well defined.
  - SC2059: Don't use variables in the printf format string. Use printf
            '..%s..' "$foo".

Now this script is shellcheck (0.9.0) compliant. We can easily spot new
issues.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-8-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: diag: fix shellcheck warnings
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:42:56 +0000 (10:42 +0100)]
selftests: mptcp: diag: fix shellcheck warnings

shellcheck recently helped to prevent issues. It is then good to fix the
other harmless issues in order to spot "real" ones later.

Here, two categories of warnings are now ignored:

- SC2317: Command appears to be unreachable. The cleanup() function is
  invoked indirectly via the EXIT trap.

- SC2086: Double quote to prevent globbing and word splitting. This is
  recommended, but the current usage is correct and there is no need to
  do all these modifications to be compliant with this rule.

For the modifications:

  - SC2034: ksft_skip appears unused.
  - SC2046: Quote '$(get_msk_inuse)' to prevent word splitting.
  - SC2006: Use $(...) notation instead of legacy backticks `...`.

Now this script is shellcheck (0.9.0) compliant. We can easily spot new
issues.

Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-7-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: add mptcp_lib_events helper
Geliang Tang [Wed, 6 Mar 2024 09:42:55 +0000 (10:42 +0100)]
selftests: mptcp: add mptcp_lib_events helper

To avoid duplicated code in different MPTCP selftests, we can add and
use helpers defined in mptcp_lib.sh.

This patch unifies "pm_nl_ctl events" related code in userspace_pm.sh
and mptcp_join.sh into a helper mptcp_lib_events(). Define it in
mptcp_lib.sh and use it in both scripts.

Note that mptcp_lib_kill_wait is now call before starting 'events' for
mptcp_join.sh as well, but that's fine: each test is started from a new
netns, so there will not be any existing pid there, and nothing is done
when mptcp_lib_kill_wait is called with 0.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-6-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: more operations in ns_init/exit
Geliang Tang [Wed, 6 Mar 2024 09:42:54 +0000 (10:42 +0100)]
selftests: mptcp: more operations in ns_init/exit

Set more the default sysctl values in mptcp_lib_ns_init(). It is fine to
do that everywhere, because they could be overridden latter if needed.

mptcp_lib_ns_exit() now also try to remove temp netns files used for the
stats even for selftests not using them. That's fine to do that because
these files have a unique name.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-5-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: add mptcp_lib_ns_init/exit helpers
Geliang Tang [Wed, 6 Mar 2024 09:42:53 +0000 (10:42 +0100)]
selftests: mptcp: add mptcp_lib_ns_init/exit helpers

Add helpers mptcp_lib_ns_init() and mptcp_lib_ns_exit() in mptcp_lib.sh
to initialize and delete the given namespaces. Then every test script
can invoke these helpers and use all namespaces.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-4-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: add local variables rndh
Geliang Tang [Wed, 6 Mar 2024 09:42:52 +0000 (10:42 +0100)]
selftests: mptcp: add local variables rndh

This patch adds local variables rndh in do_transfer() functions both in
mptcp_connect.sh and simult_flows.sh, setting it with ${ns1:4}, not the
global variable rndh. The global one is hidden in the next commit.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-3-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: add mptcp_lib_check_tools helper
Geliang Tang [Wed, 6 Mar 2024 09:42:51 +0000 (10:42 +0100)]
selftests: mptcp: add mptcp_lib_check_tools helper

This patch exports check_tools() helper from mptcp_join.sh into
mptcp_lib.sh as a public one mptcp_lib_check_tools(). The arguments
"ip", "ss", "iptables" and "ip6tables" are passed into this helper
to indicate whether to check ip tool, ss tool, iptables and ip6tables
tools.

This helper can be used in every scripts.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-2-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests: mptcp: stop forcing iptables-legacy
Matthieu Baerts (NGI0) [Wed, 6 Mar 2024 09:42:50 +0000 (10:42 +0100)]
selftests: mptcp: stop forcing iptables-legacy

Commit 0c4cd3f86a40 ("selftests: mptcp: join: use 'iptables-legacy' if
available") and commit a5a5990c099d ("selftests: mptcp: sockopt: use
'iptables-legacy' if available") forced using iptables-legacy if
available.

This was needed because of some issues that were visible when testing
the kselftests on a v5.15.x with iptables-nft as default backend. It
looks like these errors are no longer present. As mentioned by Pablo [1],
the errors were maybe due to missing kernel config. We can then use
iptables-nft if it is the default one, instead of using a legacy tool.

We can then check the variables iptables and ip6tables are valid. We can
keep the variables to easily change it later or add options.

Link: https://lore.kernel.org/netdev/ZbFiixyMFpQnxzCH@calendula/
Suggested-by: Pablo Neira Ayuso <pablo@netfilter.org>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240306-upstream-net-next-20240304-selftests-mptcp-shared-code-shellcheck-v2-1-bc79e6e5e6a0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge tag 'rxrpc-iothread-20240305' of git://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Fri, 8 Mar 2024 04:59:57 +0000 (20:59 -0800)]
Merge tag 'rxrpc-iothread-20240305' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
Here are some changes to AF_RXRPC:

 (1) Cache the transmission serial number of ACK and DATA packets in the
     rxrpc_txbuf struct and log this in the retransmit tracepoint.

 (2) Don't use atomics on rxrpc_txbuf::flags[*] and cache the intended wire
     header flags there too to avoid duplication.

 (3) Cache the wire checksum in rxrpc_txbuf to make it easier to create
     jumbo packets in future (which will require altering the wire header
     to a jumbo header and restoring it back again for retransmission).

 (4) Fix the protocol names in the wire ACK trailer struct.

 (5) Strip all the barriers and atomics out of the call timer tracking[*].

 (6) Remove atomic handling from call->tx_transmitted and
     call->acks_prev_seq[*].

 (7) Don't bother resetting the DF flag after UDP packet transmission.  To
     change it, we now call directly into UDP code, so it's quick just to
     set it every time.

 (8) Merge together the DF/non-DF branches of the DATA transmission to
     reduce duplication in the code.

 (9) Add a kvec array into rxrpc_txbuf and start moving things over to it.
     This paves the way for using page frags.

(10) Split (sub)packet preparation and timestamping out of the DATA
     transmission function.  This helps pave the way for future jumbo
     packet generation.

(11) In rxkad, don't pick values out of the wire header stored in
     rxrpc_txbuf, buf rather find them elsewhere so we can remove the wire
     header from there.

(12) Move rxrpc_send_ACK() to output.c so that it can be merged with
     rxrpc_send_ack_packet().

(13) Use rxrpc_txbuf::kvec[0] to access the wire header for the packet
     rather than directly accessing the copy in rxrpc_txbuf.  This will
     allow that to be removed to a page frag.

(14) Switch from keeping the transmission buffers in rxrpc_txbuf allocated
     in the slab to allocating them using page fragment allocators.  There
     are separate allocators for DATA packets (which persist for a while)
     and control packets (which are discarded immediately).

     We can then turn on MSG_SPLICE_PAGES when transmitting DATA and ACK
     packets.

     We can also get rid of the RCU cleanup on rxrpc_txbufs, preferring
     instead to release the page frags as soon as possible.

(15) Parse received packets before handling timeouts as the former may
     reset the latter.

(16) Make sure we don't retransmit DATA packets after all the packets have
     been ACK'd.

(17) Differentiate traces for PING ACK transmission.

(18) Switch to keeping timeouts as ktime_t rather than a number of jiffies
     as the latter is too coarse a granularity.  Only set the call timer at
     the end of the call event function from the aggregate of all the
     timeouts, thereby reducing the number of timer calls made.  In future,
     it might be possible to reduce the number of timers from one per call
     to one per I/O thread and to use a high-precision timer.

(19) Record RTT probes after successful transmission rather than recording
     it before and then cancelling it after if unsuccessful[*].  This
     allows a number of calls to get the current time to be removed.

(20) Clean up the resend algorithm as there's now no need to walk the
     transmission buffer under lock[*].  DATA packets can be retransmitted
     as soon as they're found rather than being queued up and transmitted
     when the locked is dropped.

(21) When initially parsing a received ACK packet, extract some of the
     fields from the ack info to the skbuff private data.  This makes it
     easier to do path MTU discovery in the future when the call to which a
     PING RESPONSE ACK refers has been deallocated.

[*] Possible with the move of almost all code from softirq context to the
    I/O thread.

Link: https://lore.kernel.org/r/20240301163807.385573-1-dhowells@redhat.com/
Link: https://lore.kernel.org/r/20240304084322.705539-1-dhowells@redhat.com/
* tag 'rxrpc-iothread-20240305' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs: (21 commits)
  rxrpc: Extract useful fields from a received ACK to skb priv data
  rxrpc: Clean up the resend algorithm
  rxrpc: Record probes after transmission and reduce number of time-gets
  rxrpc: Use ktimes for call timeout tracking and set the timer lazily
  rxrpc: Differentiate PING ACK transmission traces.
  rxrpc: Don't permit resending after all Tx packets acked
  rxrpc: Parse received packets before dealing with timeouts
  rxrpc: Do zerocopy using MSG_SPLICE_PAGES and page frags
  rxrpc: Use rxrpc_txbuf::kvec[0] instead of rxrpc_txbuf::wire
  rxrpc: Move rxrpc_send_ACK() to output.c with rxrpc_send_ack_packet()
  rxrpc: Don't pick values out of the wire header when setting up security
  rxrpc: Split up the DATA packet transmission function
  rxrpc: Add a kvec[] to the rxrpc_txbuf struct
  rxrpc: Merge together DF/non-DF branches of data Tx function
  rxrpc: Do lazy DF flag resetting
  rxrpc: Remove atomic handling on some fields only used in I/O thread
  rxrpc: Strip barriers and atomics off of timer tracking
  rxrpc: Fix the names of the fields in the ACK trailer struct
  rxrpc: Note cksum in txbuf
  rxrpc: Convert rxrpc_txbuf::flags into a mask and don't use atomics
  ...
====================

Link: https://lore.kernel.org/r/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: usbnet: Remove generic .ndo_get_stats64
Breno Leitao [Wed, 6 Mar 2024 14:26:42 +0000 (06:26 -0800)]
net: usbnet: Remove generic .ndo_get_stats64

Commit 3e2f544dd8a33 ("net: get stats64 if device if driver is
configured") moved the callback to dev_get_tstats64() to net core, so,
unless the driver is doing some custom stats collection, it does not
need to set .ndo_get_stats64.

Since this driver is now relying in NETDEV_PCPU_STAT_TSTATS, then, it
doesn't need to set the dev_get_tstats64() generic .ndo_get_stats64
function pointer.

Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Oliver Neukum <oneukum@suse.com>
Link: https://lore.kernel.org/r/20240306142643.2429409-2-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: usbnet: Leverage core stats allocator
Breno Leitao [Wed, 6 Mar 2024 14:26:41 +0000 (06:26 -0800)]
net: usbnet: Leverage core stats allocator

With commit 34d21de99cea9 ("net: Move {l,t,d}stats allocation to core and
convert veth & vrf"), stats allocation could be done on net core
instead of in this driver.

With this new approach, the driver doesn't have to bother with error
handling (allocation failure checking, making sure free happens in the
right spot, etc). This is core responsibility now.

Remove the allocation in the usbnet driver and leverage the network
core allocation instead.

Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Oliver Neukum <oneukum@suse.com>
Link: https://lore.kernel.org/r/20240306142643.2429409-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agodpll: spec: use proper enum for pin capabilities attribute
Jiri Pirko [Wed, 6 Mar 2024 12:07:39 +0000 (13:07 +0100)]
dpll: spec: use proper enum for pin capabilities attribute

The enum is defined, however the pin capabilities attribute does
refer to it. Add this missing enum field.

This fixes ynl cli output:

Example current output:
$ sudo ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/dpll.yaml --do pin-get --json '{"id": 0}'
{'capabilities': 4,
 ...
Example new output:
$ sudo ./tools/net/ynl/cli.py --spec Documentation/netlink/specs/dpll.yaml --do pin-get --json '{"id": 0}'
{'capabilities': {'state-can-change'},
 ...

Fixes: 3badff3a25d8 ("dpll: spec: Add Netlink spec in YAML")
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20240306120739.1447621-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: dsa: mv88e6xxx: update 88e6185 PCS driver to use neg_mode
Russell King (Oracle) [Wed, 6 Mar 2024 10:51:46 +0000 (10:51 +0000)]
net: dsa: mv88e6xxx: update 88e6185 PCS driver to use neg_mode

Update the Marvell 88e6185 PCS driver to use neg_mode rather than the
mode argument to match the other updated PCS drivers.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://lore.kernel.org/r/E1rhosE-003yuc-FM@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: pcs: rzn1-miic: update PCS driver to use neg_mode
Russell King (Oracle) [Wed, 6 Mar 2024 10:51:41 +0000 (10:51 +0000)]
net: pcs: rzn1-miic: update PCS driver to use neg_mode

Update the RZN1-MIIC PCS driver to use neg_mode rather than the mode
argument to match the other updated PCS drivers.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/E1rhos9-003yuW-Az@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: phy: marvell: add comment about m88e1111_config_init_1000basex()
Russell King (Oracle) [Wed, 6 Mar 2024 10:51:36 +0000 (10:51 +0000)]
net: phy: marvell: add comment about m88e1111_config_init_1000basex()

The comment in m88e1111_config_init_1000basex() is wrong - it claims
that Autoneg will be enabled, but this doesn't actually happen.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/E1rhos4-003yuQ-5p@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonetlink: let core handle error cases in dump operations
Eric Dumazet [Wed, 6 Mar 2024 10:24:26 +0000 (10:24 +0000)]
netlink: let core handle error cases in dump operations

After commit b5a899154aa9 ("netlink: handle EMSGSIZE errors
in the core"), we can remove some code that was not 100 % correct
anyway.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240306102426.245689-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoselftests/net: fix waiting time for ipv6_gc test in fib_tests.sh.
Kui-Feng Lee [Tue, 5 Mar 2024 18:39:49 +0000 (10:39 -0800)]
selftests/net: fix waiting time for ipv6_gc test in fib_tests.sh.

ipv6_gc fails occasionally. According to the study, fib6_run_gc() using
jiffies_round() to round the GC interval could increase the waiting time up
to 750ms (3/4 seconds). The timer has a granularity of 512ms at the range
4s to 32s. That means a route with an expiration time E seconds can wait
for more than E * 2 + 1 seconds if the GC interval is also E seconds.

E * 2 + 2 seconds should be enough for waiting for removing routes.

Also remove a check immediately after replacing 5 routes since it is very
likely to remove some of routes before completing the last route with a
slow environment.

Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
Link: https://lore.kernel.org/r/20240305183949.258473-1-thinker.li@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agompls: Do not orphan the skb
Christoph Paasch [Wed, 6 Mar 2024 18:11:17 +0000 (10:11 -0800)]
mpls: Do not orphan the skb

We observed that TCP-pacing was falling back to the TCP-layer pacing
instead of utilizing sch_fq for the pacing. This causes significant
CPU-usage due to the hrtimer running on a per-TCP-connection basis.

The issue is that mpls_xmit() calls skb_orphan() and thus sets
skb->sk to NULL. Which implies that many of the goodies of TCP won't
work. Pacing falls back to TCP-layer pacing. TCP Small Queues does not
work, ...

It is safe to remove this call to skb_orphan() in mpls_xmit() as there
really is not reason for it to be there. It appears that this call to
skb_orphan comes from the very initial implementation of MPLS.

Cc: Roopa Prabhu <roopa@nvidia.com>
Reported-by: Craig Taylor <cmtaylor@apple.com>
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Link: https://lore.kernel.org/r/20240306181117.77419-1-cpaasch@apple.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: dsa: Leverage core stats allocator
Florian Fainelli [Wed, 6 Mar 2024 20:04:09 +0000 (12:04 -0800)]
net: dsa: Leverage core stats allocator

With commit 34d21de99cea9 ("net: Move {l,t,d}stats allocation to core
and convert veth & vrf"), stats allocation could be done on net core
instead of in this driver.

With this new approach, the driver doesn't have to bother with error
handling (allocation failure checking, making sure free happens in the
right spot, etc). This is core responsibility now.

Remove the allocation in the DSA user network device code and leverage
the network core allocation instead.

Signed-off-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Breno Leitao <leitao@debian.org>
Link: https://lore.kernel.org/r/20240306200416.2973179-1-florian.fainelli@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoatm: fore200e: Convert to platform remove callback returning void
Uwe Kleine-König [Wed, 6 Mar 2024 21:23:44 +0000 (22:23 +0100)]
atm: fore200e: Convert to platform remove callback returning void

The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart
from emitting a warning) and this typically results in resource leaks.

To improve here there is a quest to make the remove callback return
void. In the first step of this quest all drivers are converted to
.remove_new(), which already returns void. Eventually after all drivers
are converted, .remove_new() will be renamed to .remove().

Trivially convert this driver from always returning zero in the remove
callback to the void returning variant.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240306212344.97985-2-u.kleine-koenig@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge branch 'tools-net-ynl-add-support-for-nlctrl-netlink-family'
Jakub Kicinski [Fri, 8 Mar 2024 04:28:41 +0000 (20:28 -0800)]
Merge branch 'tools-net-ynl-add-support-for-nlctrl-netlink-family'

Donald Hunter says:

====================
tools/net/ynl: Add support for nlctrl netlink family

This series adds a new YNL spec for the nlctrl family, plus some fixes
and enhancements for ynl.

Patch 1 fixes an extack decoding bug
Patch 2 gives cleaner netlink error reporting
Patch 3 fixes an array-nest codegen bug
Patch 4 adds nest-type-value support to ynl
Patch 5 fixes the ynl schemas to allow empty enum-name attrs
Patch 6 contains the nlctrl spec
====================

Link: https://lore.kernel.org/r/20240306231046.97158-1-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agodoc/netlink/specs: Add spec for nlctrl netlink family
Donald Hunter [Wed, 6 Mar 2024 23:10:46 +0000 (23:10 +0000)]
doc/netlink/specs: Add spec for nlctrl netlink family

Add a spec for the nlctrl family.

Example usage:

./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/nlctrl.yaml \
    --do getfamily --json '{"family-name": "nlctrl"}'

./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/nlctrl.yaml \
    --dump getpolicy --json '{"family-name": "nlctrl"}'

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-7-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agodoc/netlink: Allow empty enum-name in ynl specs
Donald Hunter [Wed, 6 Mar 2024 23:10:45 +0000 (23:10 +0000)]
doc/netlink: Allow empty enum-name in ynl specs

Update the ynl schemas to allow the specification of empty enum names
for all enum code generation.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-6-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agotools/net/ynl: Add nest-type-value decoding
Donald Hunter [Wed, 6 Mar 2024 23:10:44 +0000 (23:10 +0000)]
tools/net/ynl: Add nest-type-value decoding

The nlctrl genetlink-legacy family uses nest-type-value encoding as
described in Documentation/userspace-api/netlink/genetlink-legacy.rst

Add nest-type-value decoding to ynl.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-5-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agotools/net/ynl: Fix c codegen for array-nest
Donald Hunter [Wed, 6 Mar 2024 23:10:43 +0000 (23:10 +0000)]
tools/net/ynl: Fix c codegen for array-nest

ynl-gen-c generates e.g. 'calloc(mcast_groups, sizeof(*dst->mcast_groups))'
for array-nest attrs when it should be 'n_mcast_groups'.

Add a 'n_' prefix in the generated code for array-nests.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-4-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agotools/net/ynl: Report netlink errors without stacktrace
Donald Hunter [Wed, 6 Mar 2024 23:10:42 +0000 (23:10 +0000)]
tools/net/ynl: Report netlink errors without stacktrace

ynl does not handle NlError exceptions so they get reported like program
failures. Handle the NlError exceptions and report the netlink errors
more cleanly.

Example now:

Netlink error: No such file or directory
nl_len = 44 (28) nl_flags = 0x300 nl_type = 2
error: -2 extack: {'bad-attr': '.op'}

Example before:

Traceback (most recent call last):
  File "/home/donaldh/net-next/./tools/net/ynl/cli.py", line 81, in <module>
    main()
  File "/home/donaldh/net-next/./tools/net/ynl/cli.py", line 69, in main
    reply = ynl.dump(args.dump, attrs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/donaldh/net-next/tools/net/ynl/lib/ynl.py", line 906, in dump
    return self._op(method, vals, [], dump=True)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/donaldh/net-next/tools/net/ynl/lib/ynl.py", line 872, in _op
    raise NlError(nl_msg)
lib.ynl.NlError: Netlink error: No such file or directory
nl_len = 44 (28) nl_flags = 0x300 nl_type = 2
error: -2 extack: {'bad-attr': '.op'}

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-3-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agotools/net/ynl: Fix extack decoding for netlink-raw
Donald Hunter [Wed, 6 Mar 2024 23:10:41 +0000 (23:10 +0000)]
tools/net/ynl: Fix extack decoding for netlink-raw

Extack decoding was using a hard-coded msg header size of 20 but
netlink-raw has a header size of 16.

Use a protocol specific msghdr_size() when decoding the attr offssets.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240306231046.97158-2-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge branch 'isdn-constify-struct-class-usage'
Jakub Kicinski [Fri, 8 Mar 2024 04:26:25 +0000 (20:26 -0800)]
Merge branch 'isdn-constify-struct-class-usage'

Ricardo B. Marliere says:

====================
isdn: constify struct class usage

This is a simple and straight forward cleanup series that aims to make the
class structures in isdn constant. This has been possible since 2023 [1].

[1]: https://lore.kernel.org/all/2023040248-customary-release-4aec@gregkh/
====================

Link: https://lore.kernel.org/r/20240305-class_cleanup-isdn-v1-0-6f0edca75b61@marliere.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoisdn: capi: make capi_class constant
Ricardo B. Marliere [Tue, 5 Mar 2024 20:04:48 +0000 (17:04 -0300)]
isdn: capi: make capi_class constant

Since commit 43a7206b0963 ("driver core: class: make class_register() take
a const *"), the driver core allows for struct class to be in read-only
memory, so move the capi_class structure to be declared at build time
placing it into read-only memory, instead of having to be dynamically
allocated at boot time.

Suggested-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ricardo B. Marliere <ricardo@marliere.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240305-class_cleanup-isdn-v1-2-6f0edca75b61@marliere.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoisdn: mISDN: make elements_class constant
Ricardo B. Marliere [Tue, 5 Mar 2024 20:04:47 +0000 (17:04 -0300)]
isdn: mISDN: make elements_class constant

Since commit 43a7206b0963 ("driver core: class: make class_register() take
a const *"), the driver core allows for struct class to be in read-only
memory, so move the elements_class structure to be declared at build time
placing it into read-only memory, instead of having to be dynamically
allocated at boot time.

Suggested-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ricardo B. Marliere <ricardo@marliere.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240305-class_cleanup-isdn-v1-1-6f0edca75b61@marliere.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agodt-bindings: net: dp83822: change ti,rmii-mode description
Jérémie Dautheribes [Tue, 5 Mar 2024 14:13:09 +0000 (15:13 +0100)]
dt-bindings: net: dp83822: change ti,rmii-mode description

Drop reference to the 25MHz clock as it has nothing to do with connecting
the PHY and the MAC.
Add info about the reference clock direction between the PHY and the MAC
as it depends on the selected rmii mode.

Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jérémie Dautheribes <jeremie.dautheribes@bootlin.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20240305141309.127669-1-jeremie.dautheribes@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agonet: x25: remove dead links from Kconfig
Justin Swartz [Wed, 6 Mar 2024 11:26:59 +0000 (13:26 +0200)]
net: x25: remove dead links from Kconfig

Remove the "You can read more about X.25 at" links provided in
Kconfig as they have not pointed at any relevant pages for quite
a while.

An old copy of https://www.sangoma.com/tutorials/x25/ can be
retrieved via https://archive.org/web/ but nothing useful seems
to have been preserved for http://docwiki.cisco.com/wiki/X.25

For the sake of necromancy and those who really did want to
read more about X.25, a previous incarnation of Kconfig included
a link to:
http://www.cisco.com/univercd/cc/td/doc/product/software/ios11/cbook/cx25.htm

Which can still be read at:
https://web.archive.org/web/20071013101232/http://cisco.com/en/US/docs/ios/11_0/router/configuration/guide/cx25.html

Signed-off-by: Justin Swartz <justin.swartz@risingedge.co.za>
Acked-by: Martin Schiller <ms@dev.tdt.de>
Link: https://lore.kernel.org/r/20240306112659.25375-1-justin.swartz@risingedge.co.za
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agotools: ynl: check for overflow of constructed messages
Jakub Kicinski [Tue, 5 Mar 2024 18:50:00 +0000 (10:50 -0800)]
tools: ynl: check for overflow of constructed messages

Donald points out that we don't check for overflows.
Stash the length of the message on nlmsg_pid (nlmsg_seq would
do as well). This allows the attribute helpers to remain
self-contained (no extra arguments). Also let the put
helpers continue to return nothing. The error is checked
only in (newly introduced) ynl_msg_end().

Reviewed-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20240305185000.964773-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 7 Mar 2024 18:19:05 +0000 (10:19 -0800)]
Merge git://git./linux/kernel/git/netdev/net

Cross-merge networking fixes after downstream PR.

No conflicts.

Adjacent changes:

net/core/page_pool_user.c
  0b11b1c5c320 ("netdev: let netlink core handle -EMSGSIZE errors")
  429679dcf7d9 ("page_pool: fix netlink dump stop/resume")

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 months agoMerge tag 'net-6.8-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 7 Mar 2024 17:23:33 +0000 (09:23 -0800)]
Merge tag 'net-6.8-rc8' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from bpf, ipsec and netfilter.

  No solution yet for the stmmac issue mentioned in the last PR, but it
  proved to be a lockdep false positive, not a blocker.

  Current release - regressions:

   - dpll: move all dpll<>netdev helpers to dpll code, fix build
     regression with old compilers

  Current release - new code bugs:

   - page_pool: fix netlink dump stop/resume

  Previous releases - regressions:

   - bpf: fix verifier to check bpf_func_state->callback_depth when
     pruning states as otherwise unsafe programs could get accepted

   - ipv6: avoid possible UAF in ip6_route_mpath_notify()

   - ice: reconfig host after changing MSI-X on VF

   - mlx5:
       - e-switch, change flow rule destination checking
       - add a memory barrier to prevent a possible null-ptr-deref
       - switch to using _bh variant of of spinlock where needed

  Previous releases - always broken:

   - netfilter: nf_conntrack_h323: add protection for bmp length out of
     range

   - bpf: fix to zero-initialise xdp_rxq_info struct before running XDP
     program in CPU map which led to random xdp_md fields

   - xfrm: fix UDP encapsulation in TX packet offload

   - netrom: fix data-races around sysctls

   - ice:
       - fix potential NULL pointer dereference in ice_bridge_setlink()
       - fix uninitialized dplls mutex usage

   - igc: avoid returning frame twice in XDP_REDIRECT

   - i40e: disable NAPI right after disabling irqs when handling
     xsk_pool

   - geneve: make sure to pull inner header in geneve_rx()

   - sparx5: fix use after free inside sparx5_del_mact_entry

   - dsa: microchip: fix register write order in ksz8_ind_write8()

  Misc:

   - selftests: mptcp: fixes for diag.sh"

* tag 'net-6.8-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (63 commits)
  net: pds_core: Fix possible double free in error handling path
  netrom: Fix data-races around sysctl_net_busy_read
  netrom: Fix a data-race around sysctl_netrom_link_fails_count
  netrom: Fix a data-race around sysctl_netrom_routing_control
  netrom: Fix a data-race around sysctl_netrom_transport_no_activity_timeout
  netrom: Fix a data-race around sysctl_netrom_transport_requested_window_size
  netrom: Fix a data-race around sysctl_netrom_transport_busy_delay
  netrom: Fix a data-race around sysctl_netrom_transport_acknowledge_delay
  netrom: Fix a data-race around sysctl_netrom_transport_maximum_tries
  netrom: Fix a data-race around sysctl_netrom_transport_timeout
  netrom: Fix data-races around sysctl_netrom_network_ttl_initialiser
  netrom: Fix a data-race around sysctl_netrom_obsolescence_count_initialiser
  netrom: Fix a data-race around sysctl_netrom_default_path_quality
  netfilter: nf_conntrack_h323: Add protection for bmp length out of range
  netfilter: nf_tables: mark set as dead when unbinding anonymous set with timeout
  netfilter: nft_ct: fix l3num expectations with inet pseudo family
  netfilter: nf_tables: reject constant set with timeout
  netfilter: nf_tables: disallow anonymous set with timeout flag
  net/rds: fix WARNING in rds_conn_connect_if_down
  net: dsa: microchip: fix register write order in ksz8_ind_write8()
  ...

2 months agoMerge branch 'tcp-add-two-missing-addresses-when-using-trace'
Paolo Abeni [Thu, 7 Mar 2024 14:29:51 +0000 (15:29 +0100)]
Merge branch 'tcp-add-two-missing-addresses-when-using-trace'

Jason Xing says:

====================
tcp: add two missing addresses when using trace

When I reviewed other people's patch [1], I noticed that similar things
also happen in tcp_event_skb class and tcp_event_sk_skb class. They
don't print those two addrs of skb/sk which already exist.

In this patch, I just do as other trace functions do, like
trace_net_dev_start_xmit(), to know the exact flow or skb we would like
to know in case some systems doesn't support BPF programs well or we
have to use /sys/kernel/debug/tracing only for some reasons.

[1]
Link: https://lore.kernel.org/netdev/CAL+tcoAhvFhXdr1WQU8mv_6ZX5nOoNpbOLAB6=C+DB-qXQ11Ew@mail.gmail.com/
v2
Link: https://lore.kernel.org/netdev/CANn89iJcScraKAUk1GzZFoOO20RtC9iXpiJ4LSOWT5RUAC_QQA@mail.gmail.com/
1. change the description.
====================

Link: https://lore.kernel.org/r/20240304092934.76698-1-kerneljasonxing@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agotcp: add tracing of skbaddr in tcp_event_skb class
Jason Xing [Mon, 4 Mar 2024 09:29:34 +0000 (17:29 +0800)]
tcp: add tracing of skbaddr in tcp_event_skb class

Use the existing parameter and print the address of skbaddr
as other trace functions do.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agotcp: add tracing of skb/skaddr in tcp_event_sk_skb class
Jason Xing [Mon, 4 Mar 2024 09:29:33 +0000 (17:29 +0800)]
tcp: add tracing of skb/skaddr in tcp_event_sk_skb class

Printing the addresses can help us identify the exact skb/sk
for those system in which it's not that easy to run BPF program.
As we can see, it already fetches those, then use it directly
and it will print like below:

...tcp_retransmit_skb: skbaddr=XXX skaddr=XXX family=AF_INET...

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agoMerge branch 'doc-sfp-phylink-update-the-porting-guide'
Paolo Abeni [Thu, 7 Mar 2024 14:27:07 +0000 (15:27 +0100)]
Merge branch 'doc-sfp-phylink-update-the-porting-guide'

Maxime Chevallier says:

====================
doc: sfp-phylink: update the porting guide

Here's a V3 for an update on the phylink porting guide. The only
difference with V2 is a whitespace fix along with a line-wrap.

The main point of the update is the description of a basic process to
follow to expose one or more PCS to phylink. Let me know if you spot any
inaccuracies in the guide.

The second patch is a simple fixup on some in-code doc that was spotted
while updating the guide.

Link to V2: https://lore.kernel.org/netdev/20240228095755.1499577-1-maxime.chevallier@bootlin.com/
Link to V1: https://lore.kernel.org/netdev/20240220160406.3363002-1-maxime.chevallier@bootlin.com/
====================

Link: https://lore.kernel.org/r/20240301164309.3643849-1-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phylink: clean the pcs_get_state documentation
Maxime Chevallier [Fri, 1 Mar 2024 16:43:08 +0000 (17:43 +0100)]
net: phylink: clean the pcs_get_state documentation

commit 4d72c3bb60dd ("net: phylink: strip out pre-March 2020 legacy code")
dropped the mac_pcs_get_state ops in phylink_mac_ops in favor of
dedicated PCS operation pcs_get_state. However, the documentation for
the pcs_get_state ops was incorrectly converted and now self-references.

Drop the extra comment.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agodoc: sfp-phylink: update the porting guide with PCS handling
Maxime Chevallier [Fri, 1 Mar 2024 16:43:07 +0000 (17:43 +0100)]
doc: sfp-phylink: update the porting guide with PCS handling

Now that phylink has a comprehensive PCS support, update the porting
guide to explain the process of supporting the PCS configuration. This
also removed outdated references to phylink_config fields that no longer
exists.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: pds_core: Fix possible double free in error handling path
Yongzhi Liu [Wed, 6 Mar 2024 10:57:14 +0000 (18:57 +0800)]
net: pds_core: Fix possible double free in error handling path

When auxiliary_device_add() returns error and then calls
auxiliary_device_uninit(), Callback function pdsc_auxbus_dev_release
calls kfree(padev) to free memory. We shouldn't call kfree(padev)
again in the error handling path.

Fix this by cleaning up the redundant kfree() and putting
the error handling back to where the errors happened.

Fixes: 4569cce43bc6 ("pds_core: add auxiliary_bus devices")
Signed-off-by: Yongzhi Liu <hyperlyzcs@gmail.com>
Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240306105714.20597-1-hyperlyzcs@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>