linux-2.6-block.git
16 months agobpf: use bpf_map_kvcalloc in bpf_local_storage
Yafang Shao [Fri, 10 Feb 2023 15:47:32 +0000 (15:47 +0000)]
bpf: use bpf_map_kvcalloc in bpf_local_storage

Introduce new helper bpf_map_kvcalloc() for the memory allocation in
bpf_local_storage(). Then the allocation will charge the memory from the
map instead of from current, though currently they are the same thing as
it is only used in map creation path now. By charging map's memory into
the memcg from the map, it will be more clear.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Link: https://lore.kernel.org/r/20230210154734.4416-3-laoar.shao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
16 months agomm: memcontrol: add new kernel parameter cgroup.memory=nobpf
Yafang Shao [Fri, 10 Feb 2023 15:47:31 +0000 (15:47 +0000)]
mm: memcontrol: add new kernel parameter cgroup.memory=nobpf

Add new kernel parameter cgroup.memory=nobpf to allow user disable bpf
memory accounting. This is a preparation for the followup patch.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Link: https://lore.kernel.org/r/20230210154734.4416-2-laoar.shao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
16 months agodocs, bpf: Ensure IETF's BPF mailing list gets copied for ISA doc changes
Daniel Borkmann [Fri, 10 Feb 2023 22:40:31 +0000 (23:40 +0100)]
docs, bpf: Ensure IETF's BPF mailing list gets copied for ISA doc changes

Given BPF is increasingly being used beyond just the Linux kernel, with
implementations in NICs and other hardware, Windows, etc, there is an
ongoing effort to document and standardize parts of the existing BPF
infrastructure such as its ISA. As "source of truth" we decided some
time ago to rely on the in-tree documentation, in particular, starting
out with the Documentation/bpf/instruction-set.rst as a base for later
RFC drafts on the ISA. Therefore, we want to ensure that changes to that
document have bpf@ietf.org in Cc, so add a MAINTAINERS file entry with
a section on documents related to standardization efforts. For now, this
only relates to instruction-set.rst, and later additional files will be
added.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Dave Thaler <dthaler@microsoft.com>
Cc: bpf@ietf.org
Link: https://datatracker.ietf.org/doc/bofreq-thaler-bpf-ebpf/
Link: https://lore.kernel.org/r/57619c0dd8e354d82bf38745f99405e3babdc970.1676068387.git.daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
16 months agoDaniel Borkmann says:
Jakub Kicinski [Sat, 11 Feb 2023 01:51:27 +0000 (17:51 -0800)]
Daniel Borkmann says:

====================
pull-request: bpf-next 2023-02-11

We've added 96 non-merge commits during the last 14 day(s) which contain
a total of 152 files changed, 4884 insertions(+), 962 deletions(-).

There is a minor conflict in drivers/net/ethernet/intel/ice/ice_main.c
between commit 5b246e533d01 ("ice: split probe into smaller functions")
from the net-next tree and commit 66c0e13ad236 ("drivers: net: turn on
XDP features") from the bpf-next tree. Remove the hunk given ice_cfg_netdev()
is otherwise there a 2nd time, and add XDP features to the existing
ice_cfg_netdev() one:

        [...]
        ice_set_netdev_features(netdev);
        netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
                               NETDEV_XDP_ACT_XSK_ZEROCOPY;
        ice_set_ops(netdev);
        [...]

Stephen's merge conflict mail:
https://lore.kernel.org/bpf/20230207101951.21a114fa@canb.auug.org.au/

The main changes are:

1) Add support for BPF trampoline on s390x which finally allows to remove many
   test cases from the BPF CI's DENYLIST.s390x, from Ilya Leoshkevich.

2) Add multi-buffer XDP support to ice driver, from Maciej Fijalkowski.

3) Add capability to export the XDP features supported by the NIC.
   Along with that, add a XDP compliance test tool,
   from Lorenzo Bianconi & Marek Majtyka.

4) Add __bpf_kfunc tag for marking kernel functions as kfuncs,
   from David Vernet.

5) Add a deep dive documentation about the verifier's register
   liveness tracking algorithm, from Eduard Zingerman.

6) Fix and follow-up cleanups for resolve_btfids to be compiled
   as a host program to avoid cross compile issues,
   from Jiri Olsa & Ian Rogers.

7) Batch of fixes to the BPF selftest for xdp_hw_metadata which resulted
   when testing on different NICs, from Jesper Dangaard Brouer.

8) Fix libbpf to better detect kernel version code on Debian, from Hao Xiang.

9) Extend libbpf to add an option for when the perf buffer should
   wake up, from Jon Doron.

10) Follow-up fix on xdp_metadata selftest to just consume on TX
    completion, from Stanislav Fomichev.

11) Extend the kfuncs.rst document with description on kfunc
    lifecycle & stability expectations, from David Vernet.

12) Fix bpftool prog profile to skip attaching to offline CPUs,
    from Tonghao Zhang.

====================

Link: https://lore.kernel.org/r/20230211002037.8489-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoDocumentation: isdn: correct spelling
Randy Dunlap [Thu, 9 Feb 2023 07:13:44 +0000 (23:13 -0800)]
Documentation: isdn: correct spelling

Correct spelling problems for Documentation/isdn/ as reported
by codespell.

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/20230209071400.31476-9-rdunlap@infradead.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge branch 'net-move-more-duplicate-code-of-ovs-and-tc-conntrack-into-nf_conntrack_ovs'
Jakub Kicinski [Sat, 11 Feb 2023 00:23:05 +0000 (16:23 -0800)]
Merge branch 'net-move-more-duplicate-code-of-ovs-and-tc-conntrack-into-nf_conntrack_ovs'

Xin Long says:

====================
net: move more duplicate code of ovs and tc conntrack into nf_conntrack_ovs

We've moved some duplicate code into nf_nat_ovs in:

  "net: eliminate the duplicate code in the ct nat functions of ovs and tc"

This patchset addresses more code duplication in the conntrack of ovs
and tc then creates nf_conntrack_ovs for them, and four functions will
be extracted and moved into it:

  nf_ct_handle_fragments()
  nf_ct_skb_network_trim()
  nf_ct_helper()
  nf_ct_add_helper()
====================

Link: https://lore.kernel.org/r/cover.1675810210.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: extract nf_ct_handle_fragments to nf_conntrack_ovs
Xin Long [Tue, 7 Feb 2023 22:52:10 +0000 (17:52 -0500)]
net: extract nf_ct_handle_fragments to nf_conntrack_ovs

Now handle_fragments() in OVS and TC have the similar code, and
this patch removes the duplicate code by moving the function
to nf_conntrack_ovs.

Note that skb_clear_hash(skb) or skb->ignore_df = 1 should be
done only when defrag returns 0, as it does in other places
in kernel.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: sched: move frag check and tc_skb_cb update out of handle_fragments
Xin Long [Tue, 7 Feb 2023 22:52:09 +0000 (17:52 -0500)]
net: sched: move frag check and tc_skb_cb update out of handle_fragments

This patch has no functional changes and just moves frag check and
tc_skb_cb update out of handle_fragments, to make it easier to move
the duplicate code from handle_fragments() into nf_conntrack_ovs later.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoopenvswitch: move key and ovs_cb update out of handle_fragments
Xin Long [Tue, 7 Feb 2023 22:52:08 +0000 (17:52 -0500)]
openvswitch: move key and ovs_cb update out of handle_fragments

This patch has no functional changes and just moves key and ovs_cb update
out of handle_fragments, and skb_clear_hash() and skb->ignore_df change
into handle_fragments(), to make it easier to move the duplicate code
from handle_fragments() into nf_conntrack_ovs later.

Note that it changes to pass info->family to handle_fragments() instead
of key for the packet type check, as info->family is set according to
key->eth.type in ovs_ct_copy_action() when creating the action.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: extract nf_ct_skb_network_trim function to nf_conntrack_ovs
Xin Long [Tue, 7 Feb 2023 22:52:07 +0000 (17:52 -0500)]
net: extract nf_ct_skb_network_trim function to nf_conntrack_ovs

There are almost the same code in ovs_skb_network_trim() and
tcf_ct_skb_network_trim(), this patch extracts them into a function
nf_ct_skb_network_trim() and moves the function to nf_conntrack_ovs.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: create nf_conntrack_ovs for ovs and tc use
Xin Long [Tue, 7 Feb 2023 22:52:06 +0000 (17:52 -0500)]
net: create nf_conntrack_ovs for ovs and tc use

Similar to nf_nat_ovs created by Commit ebddb1404900 ("net: move the
nat function to nf_nat_ovs for ovs and tc"), this patch is to create
nf_conntrack_ovs to get these functions shared by OVS and TC only.

There are nf_ct_helper() and nf_ct_add_helper() from nf_conntrak_helper
in this patch, and will be more in the following patches.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agolibbpf: Fix alen calculation in libbpf_nla_dump_errormsg()
Ilya Leoshkevich [Fri, 10 Feb 2023 00:12:01 +0000 (01:12 +0100)]
libbpf: Fix alen calculation in libbpf_nla_dump_errormsg()

The code assumes that everything that comes after nlmsgerr are nlattrs.
When calculating their size, it does not account for the initial
nlmsghdr. This may lead to accessing uninitialized memory.

Fixes: bbf48c18ee0c ("libbpf: add error reporting in XDP")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-8-iii@linux.ibm.com
16 months agoselftests/bpf: Attach to fopen()/fclose() in attach_probe
Ilya Leoshkevich [Fri, 10 Feb 2023 00:12:00 +0000 (01:12 +0100)]
selftests/bpf: Attach to fopen()/fclose() in attach_probe

malloc() and free() may be completely replaced by sanitizers, use
fopen() and fclose() instead.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-7-iii@linux.ibm.com
16 months agoselftests/bpf: Attach to fopen()/fclose() in uprobe_autoattach
Ilya Leoshkevich [Fri, 10 Feb 2023 00:11:59 +0000 (01:11 +0100)]
selftests/bpf: Attach to fopen()/fclose() in uprobe_autoattach

malloc() and free() may be completely replaced by sanitizers, use
fopen() and fclose() instead.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-6-iii@linux.ibm.com
16 months agoselftests/bpf: Forward SAN_CFLAGS and SAN_LDFLAGS to runqslower and libbpf
Ilya Leoshkevich [Fri, 10 Feb 2023 00:11:58 +0000 (01:11 +0100)]
selftests/bpf: Forward SAN_CFLAGS and SAN_LDFLAGS to runqslower and libbpf

To get useful results from the Memory Sanitizer, all code running in a
process needs to be instrumented. When building tests with other
sanitizers, it's not strictly necessary, but is also helpful.
So make sure runqslower and libbpf are compiled with SAN_CFLAGS and
linked with SAN_LDFLAGS.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-5-iii@linux.ibm.com
16 months agoselftests/bpf: Split SAN_CFLAGS and SAN_LDFLAGS
Ilya Leoshkevich [Fri, 10 Feb 2023 00:11:57 +0000 (01:11 +0100)]
selftests/bpf: Split SAN_CFLAGS and SAN_LDFLAGS

Memory Sanitizer requires passing different options to CFLAGS and
LDFLAGS: besides the mandatory -fsanitize=memory, one needs to pass
header and library paths, and passing -L to a compilation step
triggers -Wunused-command-line-argument. So introduce a separate
variable for linker flags. Use $(SAN_CFLAGS) as a default in order to
avoid complicating the ASan usage.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-4-iii@linux.ibm.com
16 months agotools: runqslower: Add EXTRA_CFLAGS and EXTRA_LDFLAGS support
Ilya Leoshkevich [Fri, 10 Feb 2023 00:11:56 +0000 (01:11 +0100)]
tools: runqslower: Add EXTRA_CFLAGS and EXTRA_LDFLAGS support

This makes it possible to add sanitizer flags.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-3-iii@linux.ibm.com
16 months agoselftests/bpf: Quote host tools
Ilya Leoshkevich [Fri, 10 Feb 2023 00:11:55 +0000 (01:11 +0100)]
selftests/bpf: Quote host tools

Using HOSTCC="ccache clang" breaks building the tests, since, when it's
forwarded to e.g. bpftool, the child make sees HOSTCC=ccache and
"clang" is considered a target. Fix by quoting it, and also HOSTLD and
HOSTAR for consistency.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230210001210.395194-2-iii@linux.ibm.com
16 months agonet: skbuff: drop the word head from skb cache
Jakub Kicinski [Thu, 9 Feb 2023 06:06:42 +0000 (22:06 -0800)]
net: skbuff: drop the word head from skb cache

skbuff_head_cache is misnamed (perhaps for historical reasons?)
because it does not hold heads. Head is the buffer which skb->data
points to, and also where shinfo lives. struct sk_buff is a metadata
structure, not the head.

Eric recently added skb_small_head_cache (which allocates actual
head buffers), let that serve as an excuse to finally clean this up :)

Leave the user-space visible name intact, it could possibly be uAPI.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoMerge branch 'net-ipa-GSI'
David S. Miller [Fri, 10 Feb 2023 08:06:33 +0000 (08:06 +0000)]
Merge branch 'net-ipa-GSI'

Alex Elder says:

====================
net: ipa: prepare for GSI register updtaes

An upcoming series (or two) will convert the definitions of GSI
registers used by IPA so they use the "IPA reg" mechanism to specify
register offsets and their fields.  This will simplify implementing
the fairly large number of changes required in GSI registers to
support more than 32 GSI channels (introduced in IPA v5.0).

A few minor problems and inconsistencies were found, and they're
fixed here.  The last three patches in this series change the
"ipa_reg" code to separate the IPA-specific part (the base virtual
address, basically) from the generic register part, and the now-
generic code is renamed to use just "reg_" or "REG_" as a prefix
rather than "ipa_reg" or "IPA_REG_".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: generalize register field functions
Alex Elder [Wed, 8 Feb 2023 20:56:53 +0000 (14:56 -0600)]
net: ipa: generalize register field functions

Rename functions related to register fields so they don't appear to
be IPA-specific, and move their definitions into "reg.h":
    ipa_reg_fmask() -> reg_fmask()
    ipa_reg_bit() -> reg_bit()
    ipa_reg_field_max() -> reg_field_max()
    ipa_reg_encode() -> reg_encode()
    ipa_reg_decode() -> reg_decode()

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: generalize register offset functions
Alex Elder [Wed, 8 Feb 2023 20:56:52 +0000 (14:56 -0600)]
net: ipa: generalize register offset functions

Rename ipa_reg_offset() to be reg_offset() and move its definition
to "reg.h".  Rename ipa_reg_n_offset() to be reg_n_offset() also.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: start generalizing "ipa_reg"
Alex Elder [Wed, 8 Feb 2023 20:56:51 +0000 (14:56 -0600)]
net: ipa: start generalizing "ipa_reg"

IPA register definitions have evolved with each new version.  The
changes required to support more than 32 endpoints in IPA v5.0 made
it best to define a unified mechanism for defining registers and
their fields.

GSI register definitions, meanwhile, have remained fairly stable.
And even as the total number of IPA endpoints goes beyond 32, the
number of GSI channels on a given EE that underly endpoints still
remains 32 or less.

Despite that, GSI v3.0 (which is used with IPA v5.0) extends the
number of channels (and events) it supports to be about 256, and as
a result, many GSI register definitions must change significantly.
To address this, we'll use the same "ipa_reg" mechanism to define
the GSI registers.

As a first step in generalizing the "ipa_reg" to also support GSI
registers, isolate the definitions of the "ipa_reg" and "ipa_regs"
structure types (and some supporting macros) into a new header file,
and remove the "ipa_" and "IPA_" from symbol names.

Separate the IPA register ID validity checking from the generic
check that a register ID is in range.  Aside from that, this is
intended to have no functional effect on the code.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: GSI register cleanup
Alex Elder [Wed, 8 Feb 2023 20:56:50 +0000 (14:56 -0600)]
net: ipa: GSI register cleanup

Move some static inline function definitions out of "gsi_reg.h" and
into "gsi.c", which is the only place they're used.  Rename them so
their names identify the register they're associated with.

Move the gsi_channel_type enumerated type definition below the
offset and field definitions for the CH_C_CNTXT_0 register where
it's used.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: use bitmasks for GSI IRQ values
Alex Elder [Wed, 8 Feb 2023 20:56:49 +0000 (14:56 -0600)]
net: ipa: use bitmasks for GSI IRQ values

There are seven GSI interrupt types that can be signaled by a single
GSI IRQ.  These are represented in a bitmask, and the gsi_irq_type_id
enumerated type defines what each bit position represents.

Similarly, the global and general GSI interrupt types each has a set
of conditions it signals, and both types have an enumerated type
that defines which bit that represents each condition.

When used, these enumerated values are passed as an argument to BIT()
in *all* cases.  So clean up the code a little bit by defining the
enumerated type values as one-bit masks rather than bit positions.

Rename gsi_general_id to be gsi_general_irq_id for consistency.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: tighten up IPA register validity checking
Alex Elder [Wed, 8 Feb 2023 20:56:48 +0000 (14:56 -0600)]
net: ipa: tighten up IPA register validity checking

When checking the validity of an IPA register ID, compare it against
all possible ipa_reg_id values.

Rename the function ipa_reg_id_valid() to be specific about what's
being checked.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: add some new IPA versions
Alex Elder [Wed, 8 Feb 2023 20:56:47 +0000 (14:56 -0600)]
net: ipa: add some new IPA versions

Soon IPA v5.0+ will be supported, and when that happens we will be
able to enable support for the SDX65 (IPA v5.0), SM8450 (IPA v5.1),
and SM8550 (IPA v5.5).

Fix the comment about the GSI version used for IPA v3.1.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: get rid of ipa->reg_addr
Alex Elder [Wed, 8 Feb 2023 20:56:46 +0000 (14:56 -0600)]
net: ipa: get rid of ipa->reg_addr

The reg_addr field in the IPA structure is set but never used.
Get rid of it.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ipa: generic command param fix
Alex Elder [Wed, 8 Feb 2023 20:56:45 +0000 (14:56 -0600)]
net: ipa: generic command param fix

Starting at IPA v4.11, the GSI_GENERIC_COMMAND GSI register got a
new PARAMS field.  The code that encodes a value into that field
sets it unconditionally, which is wrong.

We currently only provide 0 as the field's value, so this error has
no real effect.  Still, it's a bug, so let's fix it.

Fix an (unrelated) incorrect comment as well.  Fields in the
ERROR_LOG GSI register actually *are* defined for IPA versions
prior to v3.5.1.

Fixes: fe68c43ce388 ("net: ipa: support enhanced channel flow control")
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: microchip: vcap: Add tc flower keys for lan966x
Horatiu Vultur [Wed, 8 Feb 2023 13:08:39 +0000 (14:08 +0100)]
net: microchip: vcap: Add tc flower keys for lan966x

Add the following TC flower filter keys to lan966x for IS2:
- ipv4_addr (sip and dip)
- ipv6_addr (sip and dip)
- control (IPv4 fragments)
- portnum (tcp and udp port numbers)
- basic (L3 and L4 protocol)
- vlan (outer vlan tag info)
- tcp (tcp flags)
- ip (tos field)

As the parsing of these keys is similar between lan966x and sparx5, move
the code in a separate file to be shared by these 2 chips. And put the
specific parsing outside of the common functions.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoMerge tag 'rxrpc-next-20230208' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 10 Feb 2023 08:00:05 +0000 (08:00 +0000)]
Merge tag 'rxrpc-next-20230208' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc development

Here are some miscellaneous changes for rxrpc:

 (1) Use consume_skb() rather than kfree_skb_reason().

 (2) Fix unnecessary waking when poking and already-poked call.

 (3) Add ack.rwind to the rxrpc_tx_ack tracepoint as this indicates how
     many incoming DATA packets we're telling the peer that we are
     currently willing to accept on this call.

 (4) Reduce duplicate ACK transmission.  We send ACKs to let the peer know
     that we're increasing the receive window (ack.rwind) as we consume
     packets locally.  Normal ACK transmission is triggered in three places
     and that leads to duplicates being sent.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: pcs: rzn1-miic: remove unused struct members and use miic variable
Clément Léger [Wed, 8 Feb 2023 16:12:49 +0000 (17:12 +0100)]
net: pcs: rzn1-miic: remove unused struct members and use miic variable

Remove unused bulk clocks struct from the miic state and use an already
existing miic variable in miic_config().

Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Link: https://lore.kernel.org/r/20230208161249.329631-1-clement.leger@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoopenvswitch: Use string_is_terminated() helper
Andy Shevchenko [Wed, 8 Feb 2023 13:31:53 +0000 (15:31 +0200)]
openvswitch: Use string_is_terminated() helper

Use string_is_terminated() helper instead of cpecific memchr() call.
This shows better the intention of the call.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230208133153.22528-3-andriy.shevchenko@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agogenetlink: Use string_is_terminated() helper
Andy Shevchenko [Wed, 8 Feb 2023 13:31:52 +0000 (15:31 +0200)]
genetlink: Use string_is_terminated() helper

Use string_is_terminated() helper instead of cpecific memchr() call.
This shows better the intention of the call.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230208133153.22528-2-andriy.shevchenko@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agostring_helpers: Move string_is_valid() to the header
Andy Shevchenko [Wed, 8 Feb 2023 13:31:51 +0000 (15:31 +0200)]
string_helpers: Move string_is_valid() to the header

Move string_is_valid() to the header for wider use.

While at it, rename to string_is_terminated() to be
precise about its semantics.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230208133153.22528-1-andriy.shevchenko@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: micrel: Cable Diagnostics feature for lan8841 PHY
Horatiu Vultur [Wed, 8 Feb 2023 11:44:06 +0000 (12:44 +0100)]
net: micrel: Cable Diagnostics feature for lan8841 PHY

Add support for cable diagnostics in lan8841 PHY. It has the same
registers layout as lan8814 PHY, therefore reuse the functionality.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20230208114406.1666671-1-horatiu.vultur@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonfp: support IPsec offloading for NFP3800
Huanhuan Wang [Wed, 8 Feb 2023 09:10:00 +0000 (10:10 +0100)]
nfp: support IPsec offloading for NFP3800

Add IPsec offloading support for NFP3800. Include data
plane and control plane.

Data plane: add IPsec packet process flow in NFP3800
datapath (NFDk).

Control plane: add an algorithm support distinction flow
in xfrm hook function xdo_dev_state_add(), as NFP3800 has
a different set of IPsec algorithm support.

This matches existing support for the NFP6000/NFP4000 and
their NFD3 datapath.

In addition, fixup the md_bytes calculation for NFD3 datapath
to make sure the two datapahts are keept in sync.

Signed-off-by: Huanhuan Wang <huanhuan.wang@corigine.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230208091000.4139974-1-simon.horman@corigine.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge branch 'net-introduce-rps_default_mask'
Jakub Kicinski [Fri, 10 Feb 2023 01:45:57 +0000 (17:45 -0800)]
Merge branch 'net-introduce-rps_default_mask'

Paolo Abeni says:

====================
net: introduce rps_default_mask

Real-time setups try hard to ensure proper isolation between time
critical applications and e.g. network processing performed by the
network stack in softirq and RPS is used to move the softirq
activity away from the isolated core.

If the network configuration is dynamic, with netns and devices
routinely created at run-time, enforcing the correct RPS setting
on each newly created device allowing to transient bad configuration
became complex.

Additionally, when multi-queue devices are involved, configuring rps
in user-space on each queue easily becomes very expensive, e.g.
some setups use veths with per cpu queues.

These series try to address the above, introducing a new
sysctl knob: rps_default_mask. The new sysctl entry allows
configuring a netns-wide RPS mask, to be enforced since receive
queue creation time without any fourther per device configuration
required.

Additionally, a simple self-test is introduced to check the
rps_default_mask behavior.
====================

Link: https://lore.kernel.org/r/cover.1675789134.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoself-tests: introduce self-tests for RPS default mask
Paolo Abeni [Tue, 7 Feb 2023 18:44:58 +0000 (19:44 +0100)]
self-tests: introduce self-tests for RPS default mask

Ensure that RPS default mask changes take place on
all newly created netns/devices and don't affect
existing ones.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: introduce default_rps_mask netns attribute
Paolo Abeni [Tue, 7 Feb 2023 18:44:57 +0000 (19:44 +0100)]
net: introduce default_rps_mask netns attribute

If RPS is enabled, this allows configuring a default rps
mask, which is effective since receive queue creation time.

A default RPS mask allows the system admin to ensure proper
isolation, avoiding races at network namespace or device
creation time.

The default RPS mask is initially empty, and can be
modified via a newly added sysctl entry.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet-sysctl: factor-out rpm mask manipulation helpers
Paolo Abeni [Tue, 7 Feb 2023 18:44:56 +0000 (19:44 +0100)]
net-sysctl: factor-out rpm mask manipulation helpers

Will simplify the following patch. No functional change
intended.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet-sysctl: factor out cpumask parsing helper
Paolo Abeni [Tue, 7 Feb 2023 18:44:55 +0000 (19:44 +0100)]
net-sysctl: factor out cpumask parsing helper

Will be used by the following patch to avoid code
duplication. No functional changes intended.

The only difference is that now flow_limit_cpu_sysctl() will
always compute the flow limit mask on each read operation,
even when read() will not return any byte to user-space.

Note that the new helper is placed under a new #ifdef at
the file start to better fit the usage in the later patch

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 9 Feb 2023 20:05:25 +0000 (12:05 -0800)]
Merge git://git./linux/kernel/git/netdev/net

net/devlink/leftover.c / net/core/devlink.c:
  565b4824c39f ("devlink: change port event netdev notifier from per-net to global")
  f05bd8ebeb69 ("devlink: move code to a dedicated directory")
  687125b5799c ("devlink: split out core code")
https://lore.kernel.org/all/20230208094657.379f2b1a@canb.auug.org.au/

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agotools/resolve_btfids: Pass HOSTCFLAGS as EXTRA_CFLAGS to prepare targets
Jiri Olsa [Thu, 9 Feb 2023 14:37:35 +0000 (15:37 +0100)]
tools/resolve_btfids: Pass HOSTCFLAGS as EXTRA_CFLAGS to prepare targets

Thorsten reported build issue with command line that defined extra
HOSTCFLAGS that were not passed into 'prepare' targets, but were
used to build resolve_btfids objects.

This results in build fail when these objects are linked together:

  /usr/bin/ld: /build.../tools/bpf/resolve_btfids//libbpf/libbpf.a(libbpf-in.o):
  relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a PIE \
  object; recompile with -fPIE

Fixing this by passing HOSTCFLAGS in EXTRA_CFLAGS as part of
HOST_OVERRIDES variable for prepare targets.

[1] https://lore.kernel.org/bpf/f7922132-6645-6316-5675-0ece4197bfff@leemhuis.info/

Fixes: 56a2df7615fa ("tools/resolve_btfids: Compile resolve_btfids as host program")
Reported-by: Thorsten Leemhuis <linux@leemhuis.info>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Tested-by: Thorsten Leemhuis <linux@leemhuis.info>
Acked-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/bpf/20230209143735.4112845-1-jolsa@kernel.org
16 months agonet: enable usercopy for skb_small_head_cache
Eric Dumazet [Wed, 8 Feb 2023 14:25:08 +0000 (14:25 +0000)]
net: enable usercopy for skb_small_head_cache

syzbot and other bots reported that we have to enable
user copy to/from skb->head. [1]

We can prevent access to skb_shared_info, which is a nice
improvement over standard kmem_cache.

Layout of these kmem_cache objects is:

< SKB_SMALL_HEAD_HEADROOM >< struct skb_shared_info >

usercopy: Kernel memory overwrite attempt detected to SLUB object 'skbuff_small_head' (offset 32, size 20)!
------------[ cut here ]------------
kernel BUG at mm/usercopy.c:102 !
invalid opcode: 0000 [#1] PREEMPT SMP KASAN
CPU: 1 PID: 1 Comm: swapper/0 Not tainted 6.2.0-rc6-syzkaller-01425-gcb6b2e11a42d #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/12/2023
RIP: 0010:usercopy_abort+0xbd/0xbf mm/usercopy.c:102
Code: e8 ee ad ba f7 49 89 d9 4d 89 e8 4c 89 e1 41 56 48 89 ee 48 c7 c7 20 2b 5b 8a ff 74 24 08 41 57 48 8b 54 24 20 e8 7a 17 fe ff <0f> 0b e8 c2 ad ba f7 e8 7d fb 08 f8 48 8b 0c 24 49 89 d8 44 89 ea
RSP: 0000:ffffc90000067a48 EFLAGS: 00010286
RAX: 000000000000006b RBX: ffffffff8b5b6ea0 RCX: 0000000000000000
RDX: ffff8881401c0000 RSI: ffffffff8166195c RDI: fffff5200000cf3b
RBP: ffffffff8a5b2a60 R08: 000000000000006b R09: 0000000000000000
R10: 0000000080000000 R11: 0000000000000000 R12: ffffffff8bf2a925
R13: ffffffff8a5b29a0 R14: 0000000000000014 R15: ffffffff8a5b2960
FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000000c48e000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
__check_heap_object+0xdd/0x110 mm/slub.c:4761
check_heap_object mm/usercopy.c:196 [inline]
__check_object_size mm/usercopy.c:251 [inline]
__check_object_size+0x1da/0x5a0 mm/usercopy.c:213
check_object_size include/linux/thread_info.h:199 [inline]
check_copy_size include/linux/thread_info.h:235 [inline]
copy_from_iter include/linux/uio.h:186 [inline]
copy_from_iter_full include/linux/uio.h:194 [inline]
memcpy_from_msg include/linux/skbuff.h:3977 [inline]
qrtr_sendmsg+0x65f/0x970 net/qrtr/af_qrtr.c:965
sock_sendmsg_nosec net/socket.c:722 [inline]
sock_sendmsg+0xde/0x190 net/socket.c:745
say_hello+0xf6/0x170 net/qrtr/ns.c:325
qrtr_ns_init+0x220/0x2b0 net/qrtr/ns.c:804
qrtr_proto_init+0x59/0x95 net/qrtr/af_qrtr.c:1296
do_one_initcall+0x141/0x790 init/main.c:1306
do_initcall_level init/main.c:1379 [inline]
do_initcalls init/main.c:1395 [inline]
do_basic_setup init/main.c:1414 [inline]
kernel_init_freeable+0x6f9/0x782 init/main.c:1634
kernel_init+0x1e/0x1d0 init/main.c:1522
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308
</TASK>

Fixes: bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Tested-by: Ido Schimmel <idosch@nvidia.com>
Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Link: https://lore.kernel.org/linux-next/CA+G9fYs-i-c2KTSA7Ai4ES_ZESY1ZnM=Zuo8P1jN00oed6KHMA@mail.gmail.com
Link: https://lore.kernel.org/r/20230208142508.3278406-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge tag 'net-6.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 9 Feb 2023 17:17:38 +0000 (09:17 -0800)]
Merge tag 'net-6.2-rc8' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from can and ipsec subtrees.

  Current release - regressions:

   - sched: fix off by one in htb_activate_prios()

   - eth: mana: fix accessing freed irq affinity_hint

   - eth: ice: fix out-of-bounds KASAN warning in virtchnl

  Current release - new code bugs:

   - eth: mtk_eth_soc: enable special tag when any MAC uses DSA

  Previous releases - always broken:

   - core: fix sk->sk_txrehash default

   - neigh: make sure used and confirmed times are valid

   - mptcp: be careful on subflow status propagation on errors

   - xfrm: prevent potential spectre v1 gadget in xfrm_xlate32_attr()

   - phylink: move phy_device_free() to correctly release phy device

   - eth: mlx5:
      - fix crash unsetting rx-vlan-filter in switchdev mode
      - fix hang on firmware reset
      - serialize module cleanup with reload and remove"

* tag 'net-6.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (57 commits)
  selftests: forwarding: lib: quote the sysctl values
  net: mscc: ocelot: fix all IPv6 getting trapped to CPU when PTP timestamping is used
  rds: rds_rm_zerocopy_callback() use list_first_entry()
  net: txgbe: Update support email address
  selftests: Fix failing VXLAN VNI filtering test
  selftests: mptcp: stop tests earlier
  selftests: mptcp: allow more slack for slow test-case
  mptcp: be careful on subflow status propagation on errors
  mptcp: fix locking for in-kernel listener creation
  mptcp: fix locking for setsockopt corner-case
  mptcp: do not wait for bare sockets' timeout
  net: ethernet: mtk_eth_soc: fix DSA TX tag hwaccel for switch port 0
  nfp: ethtool: fix the bug of setting unsupported port speed
  txhash: fix sk->sk_txrehash default
  net: ethernet: mtk_eth_soc: fix wrong parameters order in __xdp_rxq_info_reg()
  net: ethernet: mtk_eth_soc: enable special tag when any MAC uses DSA
  net: sched: sch: Fix off by one in htb_activate_prios()
  igc: Add ndo_tx_timeout support
  net: mana: Fix accessing freed irq affinity_hint
  hv_netvsc: Allocate memory in netvsc_dma_map() with GFP_ATOMIC
  ...

16 months agoMerge tag 'for-linus-2023020901' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 9 Feb 2023 17:09:13 +0000 (09:09 -0800)]
Merge tag 'for-linus-2023020901' of git://git./linux/kernel/git/hid/hid

Pull HID fixes from Benjamin Tissoires:

 - fix potential infinite loop with a badly crafted HID device (Xin
   Zhao)

 - fix regression from 6.1 in USB logitech devices potentially making
   their mouse wheel not working (Bastien Nocera)

 - clean up in AMD sensors, which fixes a long time resume bug (Mario
   Limonciello)

 - few device small fixes and quirks

* tag 'for-linus-2023020901' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid:
  HID: Ignore battery for ELAN touchscreen 29DF on HP
  HID: amd_sfh: if no sensors are enabled, clean up
  HID: logitech: Disable hi-res scrolling on USB
  HID: core: Fix deadloop in hid_apply_multiplier.
  HID: Ignore battery for Elan touchscreen on Asus TP420IA
  HID: elecom: add support for TrackBall 056E:011C

16 months agoMerge tag '6.2-rc8-smb3-client-fix' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Thu, 9 Feb 2023 17:00:26 +0000 (09:00 -0800)]
Merge tag '6.2-rc8-smb3-client-fix' of git://git.samba.org/sfrench/cifs-2.6

Pull cifx fix from Steve French:
 "Small fix for use after free"

* tag '6.2-rc8-smb3-client-fix' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Fix use-after-free in rdata->read_into_pages()

16 months agoselftests: forwarding: lib: quote the sysctl values
Hangbin Liu [Wed, 8 Feb 2023 03:21:10 +0000 (11:21 +0800)]
selftests: forwarding: lib: quote the sysctl values

When set/restore sysctl value, we should quote the value as some keys
may have multi values, e.g. net.ipv4.ping_group_range

Fixes: f5ae57784ba8 ("selftests: forwarding: lib: Add sysctl_set(), sysctl_restore()")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Link: https://lore.kernel.org/r/20230208032110.879205-1-liuhangbin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
16 months agonet: mscc: ocelot: fix all IPv6 getting trapped to CPU when PTP timestamping is used
Vladimir Oltean [Tue, 7 Feb 2023 18:31:17 +0000 (20:31 +0200)]
net: mscc: ocelot: fix all IPv6 getting trapped to CPU when PTP timestamping is used

While running this selftest which usually passes:

~/selftests/drivers/net/dsa# ./local_termination.sh eno0 swp0
TEST: swp0: Unicast IPv4 to primary MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to macvlan MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, promisc            [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, allmulti           [ OK ]
TEST: swp0: Multicast IPv4 to joined group                          [ OK ]
TEST: swp0: Multicast IPv4 to unknown group                         [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, allmulti               [ OK ]
TEST: swp0: Multicast IPv6 to joined group                          [ OK ]
TEST: swp0: Multicast IPv6 to unknown group                         [ OK ]
TEST: swp0: Multicast IPv6 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv6 to unknown group, allmulti               [ OK ]

if I start PTP timestamping then run it again (debug prints added by me),
the unknown IPv6 MC traffic is seen by the CPU port even when it should
have been dropped:

~/selftests/drivers/net/dsa# ptp4l -i swp0 -2 -P -m
ptp4l[225.410]: selected /dev/ptp1 as PTP clock
[  225.445746] mscc_felix 0000:00:00.5: ocelot_l2_ptp_trap_add: port 0 adding L2 PTP trap
[  225.453815] mscc_felix 0000:00:00.5: ocelot_ipv4_ptp_trap_add: port 0 adding IPv4 PTP event trap
[  225.462703] mscc_felix 0000:00:00.5: ocelot_ipv4_ptp_trap_add: port 0 adding IPv4 PTP general trap
[  225.471768] mscc_felix 0000:00:00.5: ocelot_ipv6_ptp_trap_add: port 0 adding IPv6 PTP event trap
[  225.480651] mscc_felix 0000:00:00.5: ocelot_ipv6_ptp_trap_add: port 0 adding IPv6 PTP general trap
ptp4l[225.488]: port 1: INITIALIZING to LISTENING on INIT_COMPLETE
ptp4l[225.488]: port 0: INITIALIZING to LISTENING on INIT_COMPLETE
^C
~/selftests/drivers/net/dsa# ./local_termination.sh eno0 swp0
TEST: swp0: Unicast IPv4 to primary MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to macvlan MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, promisc            [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, allmulti           [ OK ]
TEST: swp0: Multicast IPv4 to joined group                          [ OK ]
TEST: swp0: Multicast IPv4 to unknown group                         [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, allmulti               [ OK ]
TEST: swp0: Multicast IPv6 to joined group                          [ OK ]
TEST: swp0: Multicast IPv6 to unknown group                         [FAIL]
        reception succeeded, but should have failed
TEST: swp0: Multicast IPv6 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv6 to unknown group, allmulti               [ OK ]

The PGID_MCIPV6 is configured correctly to not flood to the CPU,
I checked that.

Furthermore, when I disable back PTP RX timestamping (ptp4l doesn't do
that when it exists), packets are RX filtered again as they should be:

~/selftests/drivers/net/dsa# hwstamp_ctl -i swp0 -r 0
[  218.202854] mscc_felix 0000:00:00.5: ocelot_l2_ptp_trap_del: port 0 removing L2 PTP trap
[  218.212656] mscc_felix 0000:00:00.5: ocelot_ipv4_ptp_trap_del: port 0 removing IPv4 PTP event trap
[  218.222975] mscc_felix 0000:00:00.5: ocelot_ipv4_ptp_trap_del: port 0 removing IPv4 PTP general trap
[  218.233133] mscc_felix 0000:00:00.5: ocelot_ipv6_ptp_trap_del: port 0 removing IPv6 PTP event trap
[  218.242251] mscc_felix 0000:00:00.5: ocelot_ipv6_ptp_trap_del: port 0 removing IPv6 PTP general trap
current settings:
tx_type 1
rx_filter 12
new settings:
tx_type 1
rx_filter 0
~/selftests/drivers/net/dsa# ./local_termination.sh eno0 swp0
TEST: swp0: Unicast IPv4 to primary MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to macvlan MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address                     [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, promisc            [ OK ]
TEST: swp0: Unicast IPv4 to unknown MAC address, allmulti           [ OK ]
TEST: swp0: Multicast IPv4 to joined group                          [ OK ]
TEST: swp0: Multicast IPv4 to unknown group                         [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv4 to unknown group, allmulti               [ OK ]
TEST: swp0: Multicast IPv6 to joined group                          [ OK ]
TEST: swp0: Multicast IPv6 to unknown group                         [ OK ]
TEST: swp0: Multicast IPv6 to unknown group, promisc                [ OK ]
TEST: swp0: Multicast IPv6 to unknown group, allmulti               [ OK ]

So it's clear that something in the PTP RX trapping logic went wrong.

Looking a bit at the code, I can see that there are 4 typos, which
populate "ipv4" VCAP IS2 key filter fields for IPv6 keys.

VCAP IS2 keys of type OCELOT_VCAP_KEY_IPV4 and OCELOT_VCAP_KEY_IPV6 are
handled by is2_entry_set(). OCELOT_VCAP_KEY_IPV4 looks at
&filter->key.ipv4, and OCELOT_VCAP_KEY_IPV6 at &filter->key.ipv6.
Simply put, when we populate the wrong key field, &filter->key.ipv6
fields "proto.mask" and "proto.value" remain all zeroes (or "don't care").
So is2_entry_set() will enter the "else" of this "if" condition:

if (msk == 0xff && (val == IPPROTO_TCP || val == IPPROTO_UDP))

and proceed to ignore the "proto" field. The resulting rule will match
on all IPv6 traffic, trapping it to the CPU.

This is the reason why the local_termination.sh selftest sees it,
because control traps are stronger than the PGID_MCIPV6 used for
flooding (from the forwarding data path).

But the problem is in fact much deeper. We trap all IPv6 traffic to the
CPU, but if we're bridged, we set skb->offload_fwd_mark = 1, so software
forwarding will not take place and IPv6 traffic will never reach its
destination.

The fix is simple - correct the typos.

I was intentionally inaccurate in the commit message about the breakage
occurring when any PTP timestamping is enabled. In fact it only happens
when L4 timestamping is requested (HWTSTAMP_FILTER_PTP_V2_EVENT or
HWTSTAMP_FILTER_PTP_V2_L4_EVENT). But ptp4l requests a larger RX
timestamping filter than it needs for "-2": HWTSTAMP_FILTER_PTP_V2_EVENT.
I wanted people skimming through git logs to not think that the bug
doesn't affect them because they only use ptp4l in L2 mode.

Fixes: 96ca08c05838 ("net: mscc: ocelot: set up traps for PTP packets")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230207183117.1745754-1-vladimir.oltean@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
16 months agords: rds_rm_zerocopy_callback() use list_first_entry()
Pietro Borrello [Tue, 7 Feb 2023 18:26:34 +0000 (18:26 +0000)]
rds: rds_rm_zerocopy_callback() use list_first_entry()

rds_rm_zerocopy_callback() uses list_entry() on the head of a list
causing a type confusion.
Use list_first_entry() to actually access the first element of the
rs_zcookie_queue list.

Fixes: 9426bbc6de99 ("rds: use list structure to track information for zerocopy completion notification")
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Pietro Borrello <borrello@diag.uniroma1.it>
Link: https://lore.kernel.org/r/20230202-rds-zerocopy-v3-1-83b0df974f9a@diag.uniroma1.it
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
16 months agoMerge tag 'ipsec-2023-02-08' of git://git.kernel.org/pub/scm/linux/kernel/git/klasser...
Jakub Kicinski [Thu, 9 Feb 2023 05:35:38 +0000 (21:35 -0800)]
Merge tag 'ipsec-2023-02-08' of git://git./linux/kernel/git/klassert/ipsec

Steffen Klassert says:

====================
ipsec 2023-02-08

1) Fix policy checks for nested IPsec tunnels when using
   xfrm interfaces. From Benedict Wong.

2) Fix netlink message expression on 32=>64-bit
   messages translators. From Anastasia Belova.

3) Prevent potential spectre v1 gadget in xfrm_xlate32_attr.
   From Eric Dumazet.

4) Always consistently use time64_t in xfrm_timer_handler.
   From Eric Dumazet.

5) Fix KCSAN reported bug: Multiple cpus can update use_time
   at the same time. From Eric Dumazet.

6) Fix SCP copy from IPv4 to IPv6 on interfamily tunnel.
   From Christian Hopps.

* tag 'ipsec-2023-02-08' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec:
  xfrm: fix bug with DSCP copy to v6 from v4 tunnel
  xfrm: annotate data-race around use_time
  xfrm: consistently use time64_t in xfrm_timer_handler()
  xfrm/compat: prevent potential spectre v1 gadget in xfrm_xlate32_attr()
  xfrm: compat: change expression for switch in xfrm_xlate64
  Fix XFRM-I support for nested ESP tunnels
====================

Link: https://lore.kernel.org/r/20230208114322.266510-1-steffen.klassert@secunet.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge tag 'linux-can-next-for-6.3-20230208' of git://git.kernel.org/pub/scm/linux...
Jakub Kicinski [Thu, 9 Feb 2023 05:32:19 +0000 (21:32 -0800)]
Merge tag 'linux-can-next-for-6.3-20230208' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
can-next 2023-02-08

The 1st patch is by Oliver Hartkopp and cleans up the CAN_RAW's
raw_setsockopt() for CAN_RAW_FD_FRAMES.

The 2nd patch is by me and fixes the compilation if
CONFIG_CAN_CALC_BITTIMING is disabled. (Problem introduced in last
pull request to next-next.)

* tag 'linux-can-next-for-6.3-20230208' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next:
  can: bittiming: can_calc_bittiming(): add missing parameter to no-op function
  can: raw: use temp variable instead of rolling back config
====================

Link: https://lore.kernel.org/r/20230208210014.3169347-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge tag 'mlx5-next-netdev-deadlock' of git://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Thu, 9 Feb 2023 05:00:54 +0000 (21:00 -0800)]
Merge tag 'mlx5-next-netdev-deadlock' of git://git./linux/kernel/git/mellanox/linux

Saeed Mahameed says:

====================
mlx5-next-netdev-deadlock

This series from Jiri solves a deadlock when removing a network namespace
with mlx5 devlink instance being in it.
The deadlock is between:
1) mlx5_ib->unregister_netdevice_notifier()
AND
2) mlx5_core->devlink_reload->cleanup_net()

To slove this introduced mlx5 netdev added/removed events to track uplink
netdev to be used for register_netdevice_notifier_dev_net() purposes.

* tag 'mlx5-next-netdev-deadlock' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
  RDMA/mlx5: Track netdev to avoid deadlock during netdev notifier unregister
  net/mlx5e: Propagate an internal event in case uplink netdev changes
  net/mlx5e: Fix trap event handling
  net/mlx5: Introduce CQE error syndrome
====================

Link: https://lore.kernel.org/r/20230208005626.72930-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: libwx: Remove unneeded semicolon
Yang Li [Wed, 8 Feb 2023 00:49:59 +0000 (08:49 +0800)]
net: libwx: Remove unneeded semicolon

./drivers/net/ethernet/wangxun/libwx/wx_lib.c:683:2-3: Unneeded semicolon

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3976
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230208004959.47553-1-yang.lee@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: libwx: clean up one inconsistent indenting
Yang Li [Wed, 8 Feb 2023 01:32:27 +0000 (09:32 +0800)]
net: libwx: clean up one inconsistent indenting

drivers/net/ethernet/wangxun/libwx/wx_lib.c:1835 wx_setup_all_rx_resources() warn: inconsistent indenting

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3981
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230208013227.111605-1-yang.lee@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agonet: txgbe: Update support email address
Jiawen Wu [Wed, 8 Feb 2023 02:30:35 +0000 (10:30 +0800)]
net: txgbe: Update support email address

Update new email address for Wangxun 10Gb NIC support team.

Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Link: https://lore.kernel.org/r/20230208023035.3371250-1-jiawenwu@trustnetic.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoRDMA/mlx5: Track netdev to avoid deadlock during netdev notifier unregister
Jiri Pirko [Tue, 1 Nov 2022 14:36:01 +0000 (15:36 +0100)]
RDMA/mlx5: Track netdev to avoid deadlock during netdev notifier unregister

When removing a network namespace with mlx5 devlink instance being in
it, following callchain is performed:

cleanup_net (takes down_read(&pernet_ops_rwsem)
devlink_pernet_pre_exit()
devlink_reload()
mlx5_devlink_reload_down()
mlx5_unload_one_devl_locked()
mlx5_detach_device()
del_adev()
mlx5r_remove()
__mlx5_ib_remove()
mlx5_ib_roce_cleanup()
mlx5_remove_netdev_notifier()
unregister_netdevice_notifier (takes down_write(&pernet_ops_rwsem)

This deadlocks.

Resolve this by converting to register_netdevice_notifier_dev_net()
which does not take pernet_ops_rwsem and moves the notifier block around
according to netdev it takes as arg.

Use previously introduced netdev added/removed events to track uplink
netdev to be used for register_netdevice_notifier_dev_net() purposes.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
16 months agonet/mlx5e: Propagate an internal event in case uplink netdev changes
Jiri Pirko [Tue, 1 Nov 2022 14:27:43 +0000 (15:27 +0100)]
net/mlx5e: Propagate an internal event in case uplink netdev changes

Whenever uplink netdev is set/cleared, propagate newly introduced event
to inform notifier blocks netdev was added/removed.

Move the set() helper to core.c from header, introduce clear() and
netdev_added_event_replay() helpers. The last one is going to be called
from rdma driver, so export it.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
16 months agonet/mlx5e: Fix trap event handling
Jiri Pirko [Thu, 24 Nov 2022 12:05:53 +0000 (13:05 +0100)]
net/mlx5e: Fix trap event handling

Current code does not return correct return value from event handler.
Fix it by returning NOTIFY_* and propagate err over newly introduce ctx
structure.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
16 months agoMerge tag 'mlx5-fixes-2023-02-07' of git://git.kernel.org/pub/scm/linux/kernel/git...
Jakub Kicinski [Thu, 9 Feb 2023 03:23:44 +0000 (19:23 -0800)]
Merge tag 'mlx5-fixes-2023-02-07' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5 fixes 2023-02-07

This series provides bug fixes to mlx5 driver.

* tag 'mlx5-fixes-2023-02-07' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
  net/mlx5: Serialize module cleanup with reload and remove
  net/mlx5: fw_tracer, Zero consumer index when reloading the tracer
  net/mlx5: fw_tracer, Clear load bit when freeing string DBs buffers
  net/mlx5: Expose SF firmware pages counter
  net/mlx5: Store page counters in a single array
  net/mlx5e: IPoIB, Show unknown speed instead of error
  net/mlx5e: Fix crash unsetting rx-vlan-filter in switchdev mode
  net/mlx5: Bridge, fix ageing of peer FDB entries
  net/mlx5: DR, Fix potential race in dr_rule_create_rule_nic
  net/mlx5e: Update rx ring hw mtu upon each rx-fcs flag change
====================

Link: https://lore.kernel.org/r/20230208030302.95378-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoMerge tag 'mlx5-updates-2023-02-07' of git://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Thu, 9 Feb 2023 03:05:58 +0000 (19:05 -0800)]
Merge tag 'mlx5-updates-2023-02-07' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2023-02-07

1) Minor and trivial code Cleanups

2) Minor fixes for net-next

3) From Shay: dynamic FW trace strings update.

* tag 'mlx5-updates-2023-02-07' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
  net/mlx5: fw_tracer, Add support for unrecognized string
  net/mlx5: fw_tracer, Add support for strings DB update event
  net/mlx5: fw_tracer, allow 0 size string DBs
  net/mlx5: fw_tracer: Fix debug print
  net/mlx5: fs, Remove redundant assignment of size
  net/mlx5: fs_core, Remove redundant variable err
  net/mlx5: Fix memory leak in error flow of port set buffer
  net/mlx5e: Remove incorrect debugfs_create_dir NULL check in TLS
  net/mlx5e: Remove incorrect debugfs_create_dir NULL check in hairpin
  net/mlx5: fs, Remove redundant vport_number assignment
  net/mlx5e: Remove redundant code for handling vlan actions
  net/mlx5e: Don't listen to remove flows event
  net/mlx5: fw reset: Skip device ID check if PCI link up failed
  net/mlx5: Remove redundant health work lock
  mlx5: reduce stack usage in mlx5_setup_tc
====================

Link: https://lore.kernel.org/r/20230208003712.68386-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agoselftests: Fix failing VXLAN VNI filtering test
Ido Schimmel [Tue, 7 Feb 2023 14:18:19 +0000 (16:18 +0200)]
selftests: Fix failing VXLAN VNI filtering test

iproute2 does not recognize the "group6" and "remote6" keywords. Fix by
using "group" and "remote" instead.

Before:

 # ./test_vxlan_vnifiltering.sh
 [...]
 Tests passed:  25
 Tests failed:   2

After:

 # ./test_vxlan_vnifiltering.sh
 [...]
 Tests passed:  27
 Tests failed:   0

Fixes: 3edf5f66c12a ("selftests: add new tests for vxlan vnifiltering")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230207141819.256689-1-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
16 months agosamples/bpf: Add openat2() enter/exit tracepoint to syscall_tp sample
Rong Tao [Wed, 8 Feb 2023 01:04:41 +0000 (09:04 +0800)]
samples/bpf: Add openat2() enter/exit tracepoint to syscall_tp sample

Commit fe3300897cbf("samples: bpf: fix syscall_tp due to unused syscall")
added openat() syscall tracepoints. This patch adds support for
openat2() as well.

Signed-off-by: Rong Tao <rongtao@cestc.cn>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/tencent_9381CB1A158ED7ADD12C4406034E21A3AC07@qq.com
16 months agolibbpf: Add sample_period to creation options
Jon Doron [Tue, 7 Feb 2023 08:19:16 +0000 (10:19 +0200)]
libbpf: Add sample_period to creation options

Add option to set when the perf buffer should wake up, by default the
perf buffer becomes signaled for every event that is being pushed to it.

In case of a high throughput of events it will be more efficient to wake
up only once you have X events ready to be read.

So your application can wakeup once and drain the entire perf buffer.

Signed-off-by: Jon Doron <jond@wiz.io>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230207081916.3398417-1-arilou@gmail.com
16 months agocan: bittiming: can_calc_bittiming(): add missing parameter to no-op function
Marc Kleine-Budde [Tue, 7 Feb 2023 20:15:01 +0000 (21:15 +0100)]
can: bittiming: can_calc_bittiming(): add missing parameter to no-op function

In commit 286c0e09e8e0 ("can: bittiming: can_changelink() pass extack
down callstack") a new parameter was added to can_calc_bittiming(),
however the static inline no-op (which is used if
CONFIG_CAN_CALC_BITTIMING is disabled) wasn't converted.

Add the new parameter to the static inline no-op of
can_calc_bittiming().

Fixes: 286c0e09e8e0 ("can: bittiming: can_changelink() pass extack down callstack")
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/20230207201734.2905618-1-mkl@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
16 months agocan: raw: use temp variable instead of rolling back config
Oliver Hartkopp [Fri, 3 Feb 2023 09:08:07 +0000 (10:08 +0100)]
can: raw: use temp variable instead of rolling back config

Introduce a temporary variable to check for an invalid configuration
attempt from user space. Before this patch the value was copied to
the real config variable and rolled back in the case of an error.

Suggested-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Link: https://lore.kernel.org/all/20230203090807.97100-1-socketcan@hartkopp.net
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
16 months agosfc: move xdp_features configuration in efx_pci_probe_post_io()
Lorenzo Bianconi [Wed, 8 Feb 2023 16:58:40 +0000 (17:58 +0100)]
sfc: move xdp_features configuration in efx_pci_probe_post_io()

Move xdp_features configuration from efx_pci_probe() to
efx_pci_probe_post_io() since it is where all the other basic netdev
features are initialised.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/9bd31c9a29bcf406ab90a249a28fc328e5578fd1.1675875404.git.lorenzo@kernel.org
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
16 months agobpf, docs: Add note about type convention
Dave Thaler [Fri, 27 Jan 2023 01:47:06 +0000 (01:47 +0000)]
bpf, docs: Add note about type convention

Add explanation about use of "u64", "u32", etc. as
the type convention used in BPF documentation.

Signed-off-by: Dave Thaler <dthaler@microsoft.com>
Acked-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230127014706.1005-1-dthaler1968@googlemail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
16 months agobpf/docs: Update design QA to be consistent with kfunc lifecycle docs
Toke Høiland-Jørgensen [Wed, 8 Feb 2023 16:41:43 +0000 (17:41 +0100)]
bpf/docs: Update design QA to be consistent with kfunc lifecycle docs

Cong pointed out that there are some inconsistencies between the BPF design
QA and the lifecycle expectations documentation we added for kfuncs. Let's
update the QA file to be consistent with the kfunc docs, and add references
where it makes sense. Also document that modules may export kfuncs now.

v3:
- Grammar nit + ack from David

v2:
- Fix repeated word (s/defined defined/defined/)

Reported-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: David Vernet <void@manifault.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/r/20230208164143.286392-1-toke@redhat.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
16 months agoMerge branch 'taprio-auto-qmaxsdu-new-tx'
David S. Miller [Wed, 8 Feb 2023 09:48:53 +0000 (09:48 +0000)]
Merge branch 'taprio-auto-qmaxsdu-new-tx'

Vladimir Oltean says:

====================
taprio automatic queueMaxSDU and new TXQ selection procedure

This patch set addresses 2 design limitations in the taprio software scheduler:

1. Software scheduling fundamentally prioritizes traffic incorrectly,
   in a way which was inspired from Intel igb/igc drivers and does not
   follow the inputs user space gives (traffic classes and TC to TXQ
   mapping). Patch 05/15 handles this, 01/15 - 04/15 are preparations
   for this work.

2. Software scheduling assumes that the gate for a traffic class closes
   as soon as the next interval begins. But this isn't true.
   If consecutive schedule entries have that traffic class gate open,
   there is no "gate close" event and taprio should keep dequeuing from
   that TC without interruptions. Patches 06/15 - 15/15 handle this.
   Patch 10/15 is a generic Qdisc change required for this to work.

Future development directions which depend on this patch set are:

- Propagating the automatic queueMaxSDU calculation down to offloading
  device drivers, instead of letting them calculate this, as
  vsc9959_tas_guard_bands_update() does today.

- A software data path for tc-taprio with preemptible traffic and
  Hold/Release events.

v1 at:
https://patchwork.kernel.org/project/netdevbpf/cover/20230128010719.2182346-1-vladimir.oltean@nxp.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: don't segment unnecessarily
Vladimir Oltean [Tue, 7 Feb 2023 13:54:40 +0000 (15:54 +0200)]
net/sched: taprio: don't segment unnecessarily

Improve commit 497cc00224cf ("taprio: Handle short intervals and large
packets") to only perform segmentation when skb->len exceeds what
taprio_dequeue() expects.

In practice, this will make the biggest difference when a traffic class
gate is always open in the schedule. This is because the max_frm_len
will be U32_MAX, and such large skb->len values as Kurt reported will be
sent just fine unsegmented.

What I don't seem to know how to handle is how to make sure that the
segmented skbs themselves are smaller than the maximum frame size given
by the current queueMaxSDU[tc]. Nonetheless, we still need to drop
those, otherwise the Qdisc will hang.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: split segmentation logic from qdisc_enqueue()
Vladimir Oltean [Tue, 7 Feb 2023 13:54:39 +0000 (15:54 +0200)]
net/sched: taprio: split segmentation logic from qdisc_enqueue()

The majority of the taprio_enqueue()'s function is spent doing TCP
segmentation, which doesn't look right to me. Compilers shouldn't have a
problem in inlining code no matter how we write it, so move the
segmentation logic to a separate function.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: automatically calculate queueMaxSDU based on TC gate durations
Vladimir Oltean [Tue, 7 Feb 2023 13:54:38 +0000 (15:54 +0200)]
net/sched: taprio: automatically calculate queueMaxSDU based on TC gate durations

taprio today has a huge problem with small TC gate durations, because it
might accept packets in taprio_enqueue() which will never be sent by
taprio_dequeue().

Since not much infrastructure was available, a kludge was added in
commit 497cc00224cf ("taprio: Handle short intervals and large
packets"), which segmented large TCP segments, but the fact of the
matter is that the issue isn't specific to large TCP segments (and even
worse, the performance penalty in segmenting those is absolutely huge).

In commit a54fc09e4cba ("net/sched: taprio: allow user input of per-tc
max SDU"), taprio gained support for queueMaxSDU, which is precisely the
mechanism through which packets should be dropped at qdisc_enqueue() if
they cannot be sent.

After that patch, it was necessary for the user to manually limit the
maximum MTU per TC. This change adds the necessary logic for taprio to
further limit the values specified (or not specified) by the user to
some minimum values which never allow oversized packets to be sent.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: keep the max_frm_len information inside struct sched_gate_list
Vladimir Oltean [Tue, 7 Feb 2023 13:54:37 +0000 (15:54 +0200)]
net/sched: keep the max_frm_len information inside struct sched_gate_list

I have one practical reason for doing this and one concerning correctness.

The practical reason has to do with a follow-up patch, which aims to mix
2 sources of max_sdu (one coming from the user and the other automatically
calculated based on TC gate durations @current link speed). Among those
2 sources of input, we must always select the smaller max_sdu value, but
this can change at various link speeds. So the max_sdu coming from the
user must be kept separated from the value that is operationally used
(the minimum of the 2), because otherwise we overwrite it and forget
what the user asked us to do.

To solve that, this patch proposes that struct sched_gate_list contains
the operationally active max_frm_len, and q->max_sdu contains just what
was requested by the user.

The reason having to do with correctness is based on the following
observation: the admin sched_gate_list becomes operational at a given
base_time in the future. Until then, it is inactive and applies no
shaping, all gates are open, etc. So the queueMaxSDU dropping shouldn't
apply either (this is a mechanism to ensure that packets smaller than
the largest gate duration for that TC don't hang the port; clearly it
makes little sense if the gates are always open).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: warn about missing size table
Vladimir Oltean [Tue, 7 Feb 2023 13:54:36 +0000 (15:54 +0200)]
net/sched: taprio: warn about missing size table

Vinicius intended taprio to take the L1 overhead into account when
estimating packet transmission time through user input, specifically
through the qdisc size table (man tc-stab).

Something like this:

tc qdisc replace dev $eth root stab overhead 24 taprio \
num_tc 8 \
map 0 1 2 3 4 5 6 7 \
queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
base-time 0 \
sched-entry S 0x7e 9000000 \
sched-entry S 0x82 1000000 \
max-sdu 0 0 0 0 0 0 0 200 \
flags 0x0 clockid CLOCK_TAI

Without the overhead being specified, transmission times will be
underestimated and will cause late transmissions. For an offloading
driver, it might even cause TX hangs if there is no open gate large
enough to send the maximum sized packets for that TC (including L1
overhead). Properly knowing the L1 overhead will ensure that we are able
to auto-calculate the queueMaxSDU per traffic class just right, and
avoid these hangs due to head-of-line blocking.

We can't make the stab mandatory due to existing setups, but we can warn
the user that it's important with a warning netlink extack.

Link: https://patchwork.kernel.org/project/netdevbpf/patch/20220505160357.298794-1-vladimir.oltean@nxp.com/
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: make stab available before ops->init() call
Vladimir Oltean [Tue, 7 Feb 2023 13:54:35 +0000 (15:54 +0200)]
net/sched: make stab available before ops->init() call

Some qdiscs like taprio turn out to be actually pretty reliant on a well
configured stab, to not underestimate the skb transmission time (by
properly accounting for L1 overhead).

In a future change, taprio will need the stab, if configured by the
user, to be available at ops->init() time. It will become even more
important in upcoming work, when the overhead will be used for the
queueMaxSDU calculation that is passed to an offloading driver.

However, rcu_assign_pointer(sch->stab, stab) is called right after
ops->init(), making it unavailable, and I don't really see a good reason
for that.

Move it earlier, which nicely seems to simplify the error handling path
as well.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: calculate guard band against actual TC gate close time
Vladimir Oltean [Tue, 7 Feb 2023 13:54:34 +0000 (15:54 +0200)]
net/sched: taprio: calculate guard band against actual TC gate close time

taprio_dequeue_from_txq() looks at the entry->end_time to determine
whether the skb will overrun its traffic class gate, as if at the end of
the schedule entry there surely is a "gate close" event for it. Hint:
maybe there isn't.

For each schedule entry, introduce an array of kernel times which
actually tracks when in the future will there be an *actual* gate close
event for that traffic class, and use that in the guard band overrun
calculation.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: calculate budgets per traffic class
Vladimir Oltean [Tue, 7 Feb 2023 13:54:33 +0000 (15:54 +0200)]
net/sched: taprio: calculate budgets per traffic class

Currently taprio assumes that the budget for a traffic class expires at
the end of the current interval as if the next interval contains a "gate
close" event for this traffic class.

This is, however, an unfounded assumption. Allow schedule entry
intervals to be fused together for a particular traffic class by
calculating the budget until the gate *actually* closes.

This means we need to keep budgets per traffic class, and we also need
to update the budget consumption procedure.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: rename close_time to end_time
Vladimir Oltean [Tue, 7 Feb 2023 13:54:32 +0000 (15:54 +0200)]
net/sched: taprio: rename close_time to end_time

There is a confusion in terms in taprio which makes what is called
"close_time" to be actually used for 2 things:

1. determining when an entry "closes" such that transmitted skbs are
   never allowed to overrun that time (?!)
2. an aid for determining when to advance and/or restart the schedule
   using the hrtimer

It makes more sense to call this so-called "close_time" "end_time",
because it's not clear at all to me what "closes". Future patches will
hopefully make better use of the term "to close".

This is an absolutely mechanical change.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: calculate tc gate durations
Vladimir Oltean [Tue, 7 Feb 2023 13:54:31 +0000 (15:54 +0200)]
net/sched: taprio: calculate tc gate durations

Current taprio code operates on a very simplistic (and incorrect)
assumption: that egress scheduling for a traffic class can only take
place for the duration of the current interval, or i.o.w., it assumes
that at the end of each schedule entry, there is a "gate close" event
for all traffic classes.

As an example, traffic sent with the schedule below will be jumpy, even
though all 8 TC gates are open, so there is absolutely no "gate close"
event (effectively a transition from BIT(tc)==1 to BIT(tc)==0 in
consecutive schedule entries):

tc qdisc replace dev veth0 parent root taprio \
num_tc 2 \
map 0 1 \
queues 1@0 1@1 \
base-time 0 \
sched-entry S 0xff 4000000000 \
clockid CLOCK_TAI \
flags 0x0

This qdisc simply does not have what it takes in terms of logic to
*actually* compute the durations of traffic classes. Also, it does not
recognize the need to use this information on a per-traffic-class basis:
it always looks at entry->interval and entry->close_time.

This change proposes that each schedule entry has an array called
tc_gate_duration[tc]. This holds the information: "for how long will
this traffic class gate remain open, starting from *this* schedule
entry". If the traffic class gate is always open, that value is equal to
the cycle time of the schedule.

We'll also need to keep track, for the purpose of queueMaxSDU[tc]
calculation, what is the maximum time duration for a traffic class
having an open gate. This gives us directly what is the maximum sized
packet that this traffic class will have to accept. For everything else
it has to qdisc_drop() it in qdisc_enqueue().

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: give higher priority to higher TCs in software dequeue mode
Vladimir Oltean [Tue, 7 Feb 2023 13:54:30 +0000 (15:54 +0200)]
net/sched: taprio: give higher priority to higher TCs in software dequeue mode

Current taprio software implementation is haunted by the shadow of the
igb/igc hardware model. It iterates over child qdiscs in increasing
order of TXQ index, therefore giving higher xmit priority to TXQ 0 and
lower to TXQ N. According to discussions with Vinicius, that is the
default (perhaps even unchangeable) prioritization scheme used for the
NICs that taprio was first written for (igb, igc), and we have a case of
two bugs canceling out, resulting in a functional setup on igb/igc, but
a less sane one on other NICs.

To the best of my understanding, taprio should prioritize based on the
traffic class, so it should really dequeue starting with the highest
traffic class and going down from there. We get to the TXQ using the
tc_to_txq[] netdev property.

TXQs within the same TC have the same (strict) priority, so we should
pick from them as fairly as we can. We can achieve that by implementing
something very similar to q->curband from multiq_dequeue().

Since igb/igc really do have TXQ 0 of higher hardware priority than
TXQ 1 etc, we need to preserve the behavior for them as well. We really
have no choice, because in txtime-assist mode, taprio is essentially a
software scheduler towards offloaded child tc-etf qdiscs, so the TXQ
selection really does matter (not all igb TXQs support ETF/SO_TXTIME,
says Kurt Kanzenbach).

To preserve the behavior, we need a capability bit so that taprio can
determine if it's running on igb/igc, or on something else. Because igb
doesn't offload taprio at all, we can't piggyback on the
qdisc_offload_query_caps() call from taprio_enable_offload(), but
instead we need a separate call which is also made for software
scheduling.

Introduce two static keys to minimize the performance penalty on systems
which only have igb/igc NICs, and on systems which only have other NICs.
For mixed systems, taprio will have to dynamically check whether to
dequeue using one prioritization algorithm or using the other.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: avoid calling child->ops->dequeue(child) twice
Vladimir Oltean [Tue, 7 Feb 2023 13:54:29 +0000 (15:54 +0200)]
net/sched: taprio: avoid calling child->ops->dequeue(child) twice

Simplify taprio_dequeue_from_txq() by noticing that we can goto one call
earlier than the previous skb_found label. This is possible because
we've unified the treatment of the child->ops->dequeue(child) return
call, we always try other TXQs now, instead of abandoning the root
dequeue completely if we failed in the peek() case.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: refactor one skb dequeue from TXQ to separate function
Vladimir Oltean [Tue, 7 Feb 2023 13:54:28 +0000 (15:54 +0200)]
net/sched: taprio: refactor one skb dequeue from TXQ to separate function

Future changes will refactor the TXQ selection procedure, and a lot of
stuff will become messy, the indentation of the bulk of the dequeue
procedure would increase, etc.

Break out the bulk of the function into a new one, which knows the TXQ
(child qdisc) we should perform a dequeue from.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: continue with other TXQs if one dequeue() failed
Vladimir Oltean [Tue, 7 Feb 2023 13:54:27 +0000 (15:54 +0200)]
net/sched: taprio: continue with other TXQs if one dequeue() failed

This changes the handling of an unlikely condition to not stop dequeuing
if taprio failed to dequeue the peeked skb in taprio_dequeue().

I've no idea when this can happen, but the only side effect seems to be
that the atomic_sub_return() call right above will have consumed some
budget. This isn't a big deal, since either that made us remain without
any budget (and therefore, we'd exit on the next peeked skb anyway), or
we could send some packets from other TXQs.

I'm making this change because in a future patch I'll be refactoring the
dequeue procedure to simplify it, and this corner case will have to go
away.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet/sched: taprio: delete peek() implementation
Vladimir Oltean [Tue, 7 Feb 2023 13:54:26 +0000 (15:54 +0200)]
net/sched: taprio: delete peek() implementation

There isn't any code in the network stack which calls taprio_peek().
We only see qdisc->ops->peek() being called on child qdiscs of other
classful qdiscs, never from the generic qdisc code. Whereas taprio is
never a child qdisc, it is always root.

This snippet of a comment from qdisc_peek_dequeued() seems to confirm:

/* we can reuse ->gso_skb because peek isn't called for root qdiscs */

Since I've been known to be wrong many times though, I'm not completely
removing it, but leaving a stub function in place which emits a warning.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoMerge branch 'mptcp-fixes'
David S. Miller [Wed, 8 Feb 2023 09:39:34 +0000 (09:39 +0000)]
Merge branch 'mptcp-fixes'

Matthieu Baerts says:

====================
mptcp: fixes for v6.2

Patch 1 clears resources earlier if there is no more reasons to keep
MPTCP sockets alive.

Patches 2 and 3 fix some locking issues visible in some rare corner
cases: the linked issues should be quite hard to reproduce.

Patch 4 makes sure subflows are correctly cleaned after the end of a
connection.

Patch 5 and 6 improve the selftests stability when running in a slow
environment by transfering data for a longer period on one hand and by
stopping the tests when all expected events have been observed on the
other hand.

All these patches fix issues introduced before v6.2.
====================

Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoselftests: mptcp: stop tests earlier
Matthieu Baerts [Tue, 7 Feb 2023 13:04:18 +0000 (14:04 +0100)]
selftests: mptcp: stop tests earlier

These 'endpoint' tests from 'mptcp_join.sh' selftest start a transfer in
the background and check the status during this transfer.

Once the expected events have been recorded, there is no reason to wait
for the data transfer to finish. It can be stopped earlier to reduce the
execution time by more than half.

For these tests, the exchanged data were not verified. Errors, if any,
were ignored but that's fine, plenty of other tests are looking at that.
It is then OK to mute stderr now that we are sure errors will be printed
(and still ignored) because the transfer is stopped before the end.

Fixes: e274f7154008 ("selftests: mptcp: add subflow limits test-cases")
Cc: stable@vger.kernel.org
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoselftests: mptcp: allow more slack for slow test-case
Paolo Abeni [Tue, 7 Feb 2023 13:04:17 +0000 (14:04 +0100)]
selftests: mptcp: allow more slack for slow test-case

A test-case is frequently failing on some extremely slow VMs.
The mptcp transfer completes before the script is able to do
all the required PM manipulation.

Address the issue in the simplest possible way, making the
transfer even more slow.

Additionally dump more info in case of failures, to help debugging
similar problems in the future and init dump_stats var.

Fixes: e274f7154008 ("selftests: mptcp: add subflow limits test-cases")
Cc: stable@vger.kernel.org
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/323
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agomptcp: be careful on subflow status propagation on errors
Paolo Abeni [Tue, 7 Feb 2023 13:04:16 +0000 (14:04 +0100)]
mptcp: be careful on subflow status propagation on errors

Currently the subflow error report callback unconditionally
propagates the fallback subflow status to the owning msk.

If the msk is already orphaned, the above prevents the code
from correctly tracking the msk moving to the TCP_CLOSE state
and doing the appropriate cleanup.

All the above causes increasing memory usage over time and
sporadic self-tests failures.

There is a great deal of infrastructure trying to propagate
correctly the fallback subflow status to the owning mptcp socket,
e.g. via mptcp_subflow_eof() and subflow_sched_work_if_closed():
in the error propagation path we need only to cope with unorphaned
sockets.

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/339
Fixes: 15cc10453398 ("mptcp: deliver ssk errors to msk")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agomptcp: fix locking for in-kernel listener creation
Paolo Abeni [Tue, 7 Feb 2023 13:04:15 +0000 (14:04 +0100)]
mptcp: fix locking for in-kernel listener creation

For consistency, in mptcp_pm_nl_create_listen_socket(), we need to
call the __mptcp_nmpc_socket() under the msk socket lock.

Note that as a side effect, mptcp_subflow_create_socket() needs a
'nested' lockdep annotation, as it will acquire the subflow (kernel)
socket lock under the in-kernel listener msk socket lock.

The current lack of locking is almost harmless, because the relevant
socket is not exposed to the user space, but in future we will add
more complexity to the mentioned helper, let's play safe.

Fixes: 1729cf186d8a ("mptcp: create the listening socket for new port")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agomptcp: fix locking for setsockopt corner-case
Paolo Abeni [Tue, 7 Feb 2023 13:04:14 +0000 (14:04 +0100)]
mptcp: fix locking for setsockopt corner-case

We need to call the __mptcp_nmpc_socket(), and later subflow socket
access under the msk socket lock, or e.g. a racing connect() could
change the socket status under the hood, with unexpected results.

Fixes: 54635bd04701 ("mptcp: add TCP_FASTOPEN_CONNECT socket option")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agomptcp: do not wait for bare sockets' timeout
Paolo Abeni [Tue, 7 Feb 2023 13:04:13 +0000 (14:04 +0100)]
mptcp: do not wait for bare sockets' timeout

If the peer closes all the existing subflows for a given
mptcp socket and later the application closes it, the current
implementation let it survive until the timewait timeout expires.

While the above is allowed by the protocol specification it
consumes resources for almost no reason and additionally
causes sporadic self-tests failures.

Let's move the mptcp socket to the TCP_CLOSE state when there are
no alive subflows at close time, so that the allocated resources
will be freed immediately.

Fixes: e16163b6e2b7 ("mptcp: refactor shutdown and close")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agoMerge branch 'micrel-lan8841-support'
David S. Miller [Wed, 8 Feb 2023 09:16:07 +0000 (09:16 +0000)]
Merge branch 'micrel-lan8841-support'

Horatiu Vultur says:

====================
net: micrel: Add support for lan8841 PHY

Add support for lan8841 PHY.

The first patch add the support for lan8841 PHY which can run at
10/100/1000Mbit. It also has support for other features, but they are not
added in this series.

The second patch updates the documentation for the dt-bindings which is
similar to the ksz9131.

v3->v4:
- add space between defines and function names
- inside lan8841_config_init use only ret variable

v2->v3:
- reuse ksz9131_config_init
- allow only open-drain configuration
- change from single patch to a patch series

v1->v2:
- Remove hardcoded values
- Fix typo in commit message
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agodt-bindings: net: micrel-ksz90x1.txt: Update for lan8841
Horatiu Vultur [Tue, 7 Feb 2023 10:52:12 +0000 (11:52 +0100)]
dt-bindings: net: micrel-ksz90x1.txt: Update for lan8841

The lan8841 has the same bindings as ksz9131, so just reuse the entire
section of ksz9131.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: micrel: Add support for lan8841 PHY
Horatiu Vultur [Tue, 7 Feb 2023 10:52:11 +0000 (11:52 +0100)]
net: micrel: Add support for lan8841 PHY

The LAN8841 is completely integrated triple-speed (10BASE-T/ 100BASE-TX/
1000BASE-T) Ethernet physical layer transceivers for transmission and
reception of data on standard CAT-5, as well as CAT-5e and CAT-6,
unshielded twisted pair (UTP) cables.
The LAN8841 offers the industry-standard GMII/MII as well as the RGMII.
Some of the features of the PHY are:
- Wake on LAN
- Auto-MDIX
- IEEE 1588-2008 (V2)
- LinkMD Capable diagnosis

Currently the patch offers support only for link configuration.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: lan966x: Add support for TC flower filter statistics
Horatiu Vultur [Tue, 7 Feb 2023 10:35:49 +0000 (11:35 +0100)]
net: lan966x: Add support for TC flower filter statistics

Add flower filter packet statistics. This will just read the TCAM
counter of the rule, which mention how many packages were hit by this
rule.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonet: ethernet: mtk_eth_soc: fix DSA TX tag hwaccel for switch port 0
Vladimir Oltean [Tue, 7 Feb 2023 10:30:27 +0000 (12:30 +0200)]
net: ethernet: mtk_eth_soc: fix DSA TX tag hwaccel for switch port 0

Arınç reports that on his MT7621AT Unielec U7621-06 board and MT7623NI
Bananapi BPI-R2, packets received by the CPU over mt7530 switch port 0
(of which this driver acts as the DSA master) are not processed
correctly by software. More precisely, they arrive without a DSA tag
(in packet or in the hwaccel area - skb_metadata_dst()), so DSA cannot
demux them towards the switch's interface for port 0. Traffic from other
ports receives a skb_metadata_dst() with the correct port and is demuxed
properly.

Looking at mtk_poll_rx(), it becomes apparent that this driver uses the
skb vlan hwaccel area:

union {
u32 vlan_all;
struct {
__be16 vlan_proto;
__u16 vlan_tci;
};
};

as a temporary storage for the VLAN hwaccel tag, or the DSA hwaccel tag.
If this is a DSA master it's a DSA hwaccel tag, and finally clears up
the skb VLAN hwaccel header.

I'm guessing that the problem is the (mis)use of API.
skb_vlan_tag_present() looks like this:

 #define skb_vlan_tag_present(__skb) (!!(__skb)->vlan_all)

So if both vlan_proto and vlan_tci are zeroes, skb_vlan_tag_present()
returns precisely false. I don't know for sure what is the format of the
DSA hwaccel tag, but I surely know that lowermost 3 bits of vlan_proto
are 0 when receiving from port 0:

unsigned int port = vlan_proto & GENMASK(2, 0);

If the RX descriptor has no other bits set to non-zero values in
RX_DMA_VTAG, then the call to __vlan_hwaccel_put_tag() will not, in
fact, make the subsequent skb_vlan_tag_present() return true, because
it's implemented like this:

static inline void __vlan_hwaccel_put_tag(struct sk_buff *skb,
  __be16 vlan_proto, u16 vlan_tci)
{
skb->vlan_proto = vlan_proto;
skb->vlan_tci = vlan_tci;
}

What we need to do to fix this problem (assuming this is the problem) is
to stop using skb->vlan_all as temporary storage for driver affairs, and
just create some local variables that serve the same purpose, but
hopefully better. Instead of calling skb_vlan_tag_present(), let's look
at a boolean has_hwaccel_tag which we set to true when the RX DMA
descriptors have something. Disambiguate based on netdev_uses_dsa()
whether this is a VLAN or DSA hwaccel tag, and only call
__vlan_hwaccel_put_tag() if we're certain it's a VLAN tag.

Arınç confirms that the treatment works, so this validates the
assumption.

Link: https://lore.kernel.org/netdev/704f3a72-fc9e-714a-db54-272e17612637@arinc9.com/
Fixes: 2d7605a72906 ("net: ethernet: mtk_eth_soc: enable hardware DSA untagging")
Reported-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Tested-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agonfp: ethtool: fix the bug of setting unsupported port speed
Yu Xiao [Tue, 7 Feb 2023 10:16:50 +0000 (11:16 +0100)]
nfp: ethtool: fix the bug of setting unsupported port speed

Unsupported port speed can be set and cause error. Now fixing it
and return an error if setting unsupported speed.

This fix depends on the following, which was included in v6.2-rc1:
commit a61474c41e8c ("nfp: ethtool: support reporting link modes").

Fixes: 7c698737270f ("nfp: add support for .set_link_ksettings()")
Signed-off-by: Yu Xiao <yu.xiao@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 months agotxhash: fix sk->sk_txrehash default
Kevin Yang [Tue, 7 Feb 2023 02:08:20 +0000 (02:08 +0000)]
txhash: fix sk->sk_txrehash default

This code fix a bug that sk->sk_txrehash gets its default enable
value from sysctl_txrehash only when the socket is a TCP listener.

We should have sysctl_txrehash to set the default sk->sk_txrehash,
no matter TCP, nor listerner/connector.

Tested by following packetdrill:
  0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
  +0 socket(..., SOCK_DGRAM, IPPROTO_UDP) = 4
  // SO_TXREHASH == 74, default to sysctl_txrehash == 1
  +0 getsockopt(3, SOL_SOCKET, 74, [1], [4]) = 0
  +0 getsockopt(4, SOL_SOCKET, 74, [1], [4]) = 0

Fixes: 26859240e4ee ("txhash: Add socket option to control TX hash rethink behavior")
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>