platform/kernel/linux-starfive.git
15 months agoMerge branch 'Remove unused fields in cpumap & devmap'
Martin KaFai Lau [Tue, 1 Aug 2023 01:26:08 +0000 (18:26 -0700)]
Merge branch 'Remove unused fields in cpumap & devmap'

Hou Tao says:

====================
Patchset "Simplify xdp_do_redirect_map()/xdp_do_flush_map() and XDP
maps" [0] changed per-map flush list to global per-cpu flush list
for cpumap, devmap and xskmap, but it forgot to remove these unused
fields from cpumap and devmap. So just remove these unused fields.

Comments and suggestions are always welcome.

[0]: https://lore.kernel.org/bpf/20191219061006.21980-1-bjorn.topel@gmail.com
====================

Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agobpf, devmap: Remove unused dtab field from bpf_dtab_netdev
Hou Tao [Fri, 28 Jul 2023 01:49:42 +0000 (09:49 +0800)]
bpf, devmap: Remove unused dtab field from bpf_dtab_netdev

Commit 96360004b862 ("xdp: Make devmap flush_list common for all map
instances") removes the use of bpf_dtab_netdev::dtab in bq_enqueue(),
so just remove dtab from bpf_dtab_netdev.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Link: https://lore.kernel.org/r/20230728014942.892272-3-houtao@huaweicloud.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agobpf, cpumap: Remove unused cmap field from bpf_cpu_map_entry
Hou Tao [Fri, 28 Jul 2023 01:49:41 +0000 (09:49 +0800)]
bpf, cpumap: Remove unused cmap field from bpf_cpu_map_entry

Since commit cdfafe98cabe ("xdp: Make cpumap flush_list common for all
map instances"), cmap is no longer used, so just remove it.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20230728014942.892272-2-houtao@huaweicloud.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agonetfilter: bpf: Only define get_proto_defrag_hook() if necessary
Daniel Xu [Mon, 31 Jul 2023 21:55:00 +0000 (15:55 -0600)]
netfilter: bpf: Only define get_proto_defrag_hook() if necessary

Before, we were getting this warning:

  net/netfilter/nf_bpf_link.c:32:1: warning: 'get_proto_defrag_hook' defined but not used [-Wunused-function]

Guard the definition with CONFIG_NF_DEFRAG_IPV[4|6].

Fixes: 91721c2d02d3 ("netfilter: bpf: Support BPF_F_NETFILTER_IP_DEFRAG in netfilter link")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202307291213.fZ0zDmoG-lkp@intel.com/
Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Link: https://lore.kernel.org/r/b128b6489f0066db32c4772ae4aaee1480495929.1690840454.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Fix an array-index-out-of-bounds issue in disasm.c
Yonghong Song [Mon, 31 Jul 2023 20:45:34 +0000 (13:45 -0700)]
bpf: Fix an array-index-out-of-bounds issue in disasm.c

syzbot reported an array-index-out-of-bounds when printing out bpf
insns. Further investigation shows the insn is illegal but
is printed out due to log level 1 or 2 before actual insn verification
in do_check().

This particular illegal insn is a MOVSX insn with offset value 2.
The legal offset value for MOVSX should be 8, 16 and 32.
The disasm sign-extension-size array index is calculated as
 (insn->off / 8) - 1
and offset value 2 gives an out-of-bound index -1.

Tighten the checking for MOVSX insn in disasm.c to avoid
array-index-out-of-bounds issue.

Reported-by: syzbot+3758842a6c01012aa73b@syzkaller.appspotmail.com
Fixes: f835bb622299 ("bpf: Add kernel/bpftool asm support for new instructions")
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20230731204534.1975311-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agonet: remove duplicate INDIRECT_CALLABLE_DECLARE of udp[6]_ehashfn
Lorenz Bauer [Mon, 31 Jul 2023 09:42:53 +0000 (11:42 +0200)]
net: remove duplicate INDIRECT_CALLABLE_DECLARE of udp[6]_ehashfn

There are already INDIRECT_CALLABLE_DECLARE in the hashtable
headers, no need to declare them again.

Fixes: 0f495f761722 ("net: remove duplicate reuseport_lookup functions")
Suggested-by: Martin Lau <martin.lau@linux.dev>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230731-indir-call-v1-1-4cd0aeaee64f@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agodocs/bpf: Fix malformed documentation
Yonghong Song [Sun, 30 Jul 2023 00:42:51 +0000 (17:42 -0700)]
docs/bpf: Fix malformed documentation

Two issues are fixed:
  1. Malformed table due to newly-introduced BPF_MOVSX
  2. Missing reference link for ``Sign-extension load operations``

Fixes: 245d4c40c09b ("docs/bpf: Add documentation for new instructions")
Cc: bpf@ietf.org
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202307291840.Cqhj7uox-lkp@intel.com/
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230730004251.381307-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoMerge branch 'support-defragmenting-ipv-4-6-packets-in-bpf'
Alexei Starovoitov [Fri, 28 Jul 2023 23:52:08 +0000 (16:52 -0700)]
Merge branch 'support-defragmenting-ipv-4-6-packets-in-bpf'

Daniel Xu says:

====================
Support defragmenting IPv(4|6) packets in BPF

=== Context ===

In the context of a middlebox, fragmented packets are tricky to handle.
The full 5-tuple of a packet is often only available in the first
fragment which makes enforcing consistent policy difficult. There are
really only two stateless options, neither of which are very nice:

1. Enforce policy on first fragment and accept all subsequent fragments.
   This works but may let in certain attacks or allow data exfiltration.

2. Enforce policy on first fragment and drop all subsequent fragments.
   This does not really work b/c some protocols may rely on
   fragmentation. For example, DNS may rely on oversized UDP packets for
   large responses.

So stateful tracking is the only sane option. RFC 8900 [0] calls this
out as well in section 6.3:

    Middleboxes [...] should process IP fragments in a manner that is
    consistent with [RFC0791] and [RFC8200]. In many cases, middleboxes
    must maintain state in order to achieve this goal.

=== BPF related bits ===

Policy has traditionally been enforced from XDP/TC hooks. Both hooks
run before kernel reassembly facilities. However, with the new
BPF_PROG_TYPE_NETFILTER, we can rather easily hook into existing
netfilter reassembly infra.

The basic idea is we bump a refcnt on the netfilter defrag module and
then run the bpf prog after the defrag module runs. This allows bpf
progs to transparently see full, reassembled packets. The nice thing
about this is that progs don't have to carry around logic to detect
fragments.

=== Changelog ===

Changes from v5:

* Fix defrag disable codepaths

Changes from v4:

* Refactor module handling code to not sleep in rcu_read_lock()
* Also unify the v4 and v6 hook structs so they can share codepaths
* Fixed some checkpatch.pl formatting warnings

Changes from v3:

* Correctly initialize `addrlen` stack var for recvmsg()

Changes from v2:

* module_put() if ->enable() fails
* Fix CI build errors

Changes from v1:

* Drop bpf_program__attach_netfilter() patches
* static -> static const where appropriate
* Fix callback assignment order during registration
* Only request_module() if callbacks are missing
* Fix retval when modprobe fails in userspace
* Fix v6 defrag module name (nf_defrag_ipv6_hooks -> nf_defrag_ipv6)
* Simplify priority checking code
* Add warning if module doesn't assign callbacks in the future
* Take refcnt on module while defrag link is active

[0]: https://datatracker.ietf.org/doc/html/rfc8900
====================

Link: https://lore.kernel.org/r/cover.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: selftests: Add defrag selftests
Daniel Xu [Fri, 21 Jul 2023 20:22:49 +0000 (14:22 -0600)]
bpf: selftests: Add defrag selftests

These selftests tests 2 major scenarios: the BPF based defragmentation
can successfully be done and that packet pointers are invalidated after
calls to the kfunc. The logic is similar for both ipv4 and ipv6.

In the first scenario, we create a UDP client and UDP echo server. The
the server side is fairly straightforward: we attach the prog and simply
echo back the message.

The on the client side, we send fragmented packets to and expect the
reassembled message back from the server.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Link: https://lore.kernel.org/r/33e40fdfddf43be93f2cb259303f132f46750953.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: selftests: Support custom type and proto for client sockets
Daniel Xu [Fri, 21 Jul 2023 20:22:48 +0000 (14:22 -0600)]
bpf: selftests: Support custom type and proto for client sockets

Extend connect_to_fd_opts() to take optional type and protocol
parameters for the client socket. These parameters are useful when
opening a raw socket to send IP fragments.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Link: https://lore.kernel.org/r/9067db539efdfd608aa86a2b143c521337c111fc.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: selftests: Support not connecting client socket
Daniel Xu [Fri, 21 Jul 2023 20:22:47 +0000 (14:22 -0600)]
bpf: selftests: Support not connecting client socket

For connectionless protocols or raw sockets we do not want to actually
connect() to the server.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Link: https://lore.kernel.org/r/525c13d66dac2d640a1db922546842c051c6f2e6.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agonetfilter: bpf: Support BPF_F_NETFILTER_IP_DEFRAG in netfilter link
Daniel Xu [Fri, 21 Jul 2023 20:22:46 +0000 (14:22 -0600)]
netfilter: bpf: Support BPF_F_NETFILTER_IP_DEFRAG in netfilter link

This commit adds support for enabling IP defrag using pre-existing
netfilter defrag support. Basically all the flag does is bump a refcnt
while the link the active. Checks are also added to ensure the prog
requesting defrag support is run _after_ netfilter defrag hooks.

We also take care to avoid any issues w.r.t. module unloading -- while
defrag is active on a link, the module is prevented from unloading.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Reviewed-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/5cff26f97e55161b7d56b09ddcf5f8888a5add1d.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agonetfilter: defrag: Add glue hooks for enabling/disabling defrag
Daniel Xu [Fri, 21 Jul 2023 20:22:45 +0000 (14:22 -0600)]
netfilter: defrag: Add glue hooks for enabling/disabling defrag

We want to be able to enable/disable IP packet defrag from core
bpf/netfilter code. In other words, execute code from core that could
possibly be built as a module.

To help avoid symbol resolution errors, use glue hooks that the modules
will register callbacks with during module init.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Reviewed-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/f6a8824052441b72afe5285acedbd634bd3384c1.1689970773.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agodocs/bpf: Improve documentation for cpu=v4 instructions
Yonghong Song [Fri, 28 Jul 2023 22:51:05 +0000 (15:51 -0700)]
docs/bpf: Improve documentation for cpu=v4 instructions

Improve documentation for cpu=v4 instructions based on
David's suggestions.

Cc: bpf@ietf.org
Suggested-by: David Vernet <void@manifault.com>
Acked-by: David Vernet <void@manifault.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728225105.919595-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Non-atomically allocate freelist during prefill
YiFei Zhu [Fri, 28 Jul 2023 04:33:59 +0000 (04:33 +0000)]
bpf: Non-atomically allocate freelist during prefill

In internal testing of test_maps, we sometimes observed failures like:
  test_maps: test_maps.c:173: void test_hashmap_percpu(unsigned int, void *):
    Assertion `bpf_map_update_elem(fd, &key, value, BPF_ANY) == 0' failed.
where the errno is ENOMEM. After some troubleshooting and enabling
the warnings, we saw:
  [   91.304708] percpu: allocation failed, size=8 align=8 atomic=1, atomic alloc failed, no space left
  [   91.304716] CPU: 51 PID: 24145 Comm: test_maps Kdump: loaded Tainted: G                 N 6.1.38-smp-DEV #7
  [   91.304719] Hardware name: Google Astoria/astoria, BIOS 0.20230627.0-0 06/27/2023
  [   91.304721] Call Trace:
  [   91.304724]  <TASK>
  [   91.304730]  [<ffffffffa7ef83b9>] dump_stack_lvl+0x59/0x88
  [   91.304737]  [<ffffffffa7ef83f8>] dump_stack+0x10/0x18
  [   91.304738]  [<ffffffffa75caa0c>] pcpu_alloc+0x6fc/0x870
  [   91.304741]  [<ffffffffa75ca302>] __alloc_percpu_gfp+0x12/0x20
  [   91.304743]  [<ffffffffa756785e>] alloc_bulk+0xde/0x1e0
  [   91.304746]  [<ffffffffa7566c02>] bpf_mem_alloc_init+0xd2/0x2f0
  [   91.304747]  [<ffffffffa7547c69>] htab_map_alloc+0x479/0x650
  [   91.304750]  [<ffffffffa751d6e0>] map_create+0x140/0x2e0
  [   91.304752]  [<ffffffffa751d413>] __sys_bpf+0x5a3/0x6c0
  [   91.304753]  [<ffffffffa751c3ec>] __x64_sys_bpf+0x1c/0x30
  [   91.304754]  [<ffffffffa7ef847a>] do_syscall_64+0x5a/0x80
  [   91.304756]  [<ffffffffa800009b>] entry_SYSCALL_64_after_hwframe+0x63/0xcd

This makes sense, because in atomic context, percpu allocation would
not create new chunks; it would only create in non-atomic contexts.
And if during prefill all precpu chunks are full, -ENOMEM would
happen immediately upon next unit_alloc.

Prefill phase does not actually run in atomic context, so we can
use this fact to allocate non-atomically with GFP_KERNEL instead
of GFP_NOWAIT. This avoids the immediate -ENOMEM.

GFP_NOWAIT has to be used in unit_alloc when bpf program runs
in atomic context. Even if bpf program runs in non-atomic context,
in most cases, rcu read lock is enabled for the program so
GFP_NOWAIT is still needed. This is often also the case for
BPF_MAP_UPDATE_ELEM syscalls.

Signed-off-by: YiFei Zhu <zhuyifei@google.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20230728043359.3324347-1-zhuyifei@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Enable test test_progs-cpuv4 for gcc build kernel
Yonghong Song [Fri, 28 Jul 2023 05:57:45 +0000 (22:57 -0700)]
selftests/bpf: Enable test test_progs-cpuv4 for gcc build kernel

Currently, test_progs-cpuv4 is generated with clang build kernel
when bpf cpu=v4 is supported by the clang compiler.
Let us enable test_progs-cpuv4 for gcc build kernel as well.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728055745.2285202-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Fix compilation warning with -Wparentheses
Yonghong Song [Fri, 28 Jul 2023 05:57:40 +0000 (22:57 -0700)]
bpf: Fix compilation warning with -Wparentheses

The kernel test robot reported compilation warnings when -Wparentheses is
added to KBUILD_CFLAGS with gcc compiler. The following is the error message:

  .../bpf-next/kernel/bpf/verifier.c: In function ‘coerce_reg_to_size_sx’:
  .../bpf-next/kernel/bpf/verifier.c:5901:14:
    error: suggest parentheses around comparison in operand of ‘==’ [-Werror=parentheses]
    if (s64_max >= 0 == s64_min >= 0) {
        ~~~~~~~~^~~~
  .../bpf-next/kernel/bpf/verifier.c: In function ‘coerce_subreg_to_size_sx’:
  .../bpf-next/kernel/bpf/verifier.c:5965:14:
    error: suggest parentheses around comparison in operand of ‘==’ [-Werror=parentheses]
    if (s32_min >= 0 == s32_max >= 0) {
        ~~~~~~~~^~~~

To fix the issue, add proper parentheses for the above '>=' condition
to silence the warning/error.

I tried a few clang compilers like clang16 and clang18 and they do not emit
such warnings with -Wparentheses.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202307281133.wi0c4SqG-lkp@intel.com/
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20230728055740.2284534-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoMerge branch 'bpf-support-new-insns-from-cpu-v4'
Alexei Starovoitov [Fri, 28 Jul 2023 01:52:34 +0000 (18:52 -0700)]
Merge branch 'bpf-support-new-insns-from-cpu-v4'

Yonghong Song says:

====================
bpf: Support new insns from cpu v4

In previous discussion ([1]), it is agreed that we should introduce
cpu version 4 (llvm flag -mcpu=v4) which contains some instructions
which can simplify code, make code easier to understand, fix the
existing problem, or simply for feature completeness. More specifically,
the following new insns are proposed:
  . sign extended load
  . sign extended mov
  . bswap
  . signed div/mod
  . ja with 32-bit offset

This patch set added kernel support for insns proposed in [1] except
BPF_ST which already has full kernel support. Beside the above proposed
insns, LLVM will generate BPF_ST insn as well under -mcpu=v4.
The llvm patch ([2]) has been merged into llvm-project 'main' branch.

The patchset implements interpreter, jit and verifier support for these new
insns.

For this patch set, I tested cpu v2/v3/v4 and the selftests are all passed.
I also tested selftests introduced in this patch set with additional changes
beside normal jit testing (bpf_jit_enable = 1 and bpf_jit_harden = 0)
  - bpf_jit_enable = 0
  - bpf_jit_enable = 1 and bpf_jit_harden = 1
and both testing passed.

  [1] https://lore.kernel.org/bpf/4bfe98be-5333-1c7e-2f6d-42486c8ec039@meta.com/
  [2] https://reviews.llvm.org/D144829

Changelogs:
  v4 -> v5:
   . for v4, patch 8/17 missed in mailing list and patchwork, so resend.
   . rebase on top of master
  v3 -> v4:
   . some minor asm syntax adjustment based on llvm change.
   . add clang version and target arch guard for new tests
     so they can still compile with old llvm compilers.
   . some changes to the bpf doc.
  v2 -> v3:
   . add missed disasm change from v2.
   . handle signed load of ctx fields properly.
   . fix some interpreter sdiv/smod error when bpf_jit_enable = 0.
   . fix some verifier range bounding errors.
   . add more C tests.
  RFCv1 -> v2:
   . add more verifier supports for signed extend load and mov insns.
   . rename some insn names to be more consistent with intel practice.
   . add cpuv4 test runner for test progs.
   . add more unit and C tests.
   . add documentation.
====================

Link: https://lore.kernel.org/r/20230728011143.3710005-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agodocs/bpf: Add documentation for new instructions
Yonghong Song [Fri, 28 Jul 2023 01:13:42 +0000 (18:13 -0700)]
docs/bpf: Add documentation for new instructions

Add documentation in instruction-set.rst for new instruction encoding
and their corresponding operations. Also removed the question
related to 'no BPF_SDIV' in bpf_design_QA.rst since we have
BPF_SDIV insn now.

Cc: bpf@ietf.org
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011342.3724411-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Test ldsx with more complex cases
Yonghong Song [Fri, 28 Jul 2023 01:13:36 +0000 (18:13 -0700)]
selftests/bpf: Test ldsx with more complex cases

The following ldsx cases are tested:
  - signed readonly map value
  - read/write map value
  - probed memory
  - not-narrowed ctx field access
  - narrowed ctx field access.

Without previous proper verifier/git handling, the test will fail.

If cpuv4 is not supported either by compiler or by jit,
the test will be skipped.

  # ./test_progs -t ldsx_insn
  #113/1   ldsx_insn/map_val and probed_memory:SKIP
  #113/2   ldsx_insn/ctx_member_sign_ext:SKIP
  #113/3   ldsx_insn/ctx_member_narrow_sign_ext:SKIP
  #113     ldsx_insn:SKIP
  Summary: 1/0 PASSED, 3 SKIPPED, 0 FAILED

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011336.3723434-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add unit tests for new gotol insn
Yonghong Song [Fri, 28 Jul 2023 01:13:29 +0000 (18:13 -0700)]
selftests/bpf: Add unit tests for new gotol insn

Add unit tests for gotol insn.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011329.3721881-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add unit tests for new sdiv/smod insns
Yonghong Song [Fri, 28 Jul 2023 01:13:21 +0000 (18:13 -0700)]
selftests/bpf: Add unit tests for new sdiv/smod insns

Add unit tests for sdiv/smod insns.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011321.3720500-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add unit tests for new bswap insns
Yonghong Song [Fri, 28 Jul 2023 01:13:14 +0000 (18:13 -0700)]
selftests/bpf: Add unit tests for new bswap insns

Add unit tests for bswap insns.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011314.3720109-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add unit tests for new sign-extension mov insns
Yonghong Song [Fri, 28 Jul 2023 01:13:09 +0000 (18:13 -0700)]
selftests/bpf: Add unit tests for new sign-extension mov insns

Add unit tests for movsx insns.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011309.3719295-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add unit tests for new sign-extension load insns
Yonghong Song [Fri, 28 Jul 2023 01:13:04 +0000 (18:13 -0700)]
selftests/bpf: Add unit tests for new sign-extension load insns

Add unit tests for new ldsx insns. The test includes sign-extension
with a single value or with a value range.

If cpuv4 is not supported due to
  (1) older compiler, e.g., less than clang version 18, or
  (2) test runner test_progs and test_progs-no_alu32 which tests
      cpu v2 and v3, or
  (3) non-x86_64 arch not supporting new insns in jit yet,
a dummy program is added with below output:
  #318/1   verifier_ldsx/cpuv4 is not supported by compiler or jit, use a dummy test:OK
  #318     verifier_ldsx:OK
to indicate the test passed with a dummy test instead of actually
testing cpuv4. I am using a dummy prog to avoid changing the
verifier testing infrastructure. Once clang 18 is widely available
and other architectures support cpuv4, at least for CI run,
the dummy program can be removed.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011304.3719139-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Add a cpuv4 test runner for cpu=v4 testing
Yonghong Song [Fri, 28 Jul 2023 01:12:50 +0000 (18:12 -0700)]
selftests/bpf: Add a cpuv4 test runner for cpu=v4 testing

Similar to no-alu32 runner, if clang compiler supports -mcpu=v4,
a cpuv4 runner is created to test bpf programs compiled with
-mcpu=v4.

The following are some num-of-insn statistics for each newer
instructions based on existing selftests, excluding subsequent
cpuv4 insn specific tests.

   insn pattern                # of instructions
   reg = (s8)reg               4
   reg = (s16)reg              4
   reg = (s32)reg              144
   reg = *(s8 *)(reg + off)    13
   reg = *(s16 *)(reg + off)   14
   reg = *(s32 *)(reg + off)   15215
   reg = bswap16 reg           142
   reg = bswap32 reg           38
   reg = bswap64 reg           14
   reg s/= reg                 0
   reg s%= reg                 0
   gotol <offset>              58

Note that in llvm -mcpu=v4 implementation, the compiler is a little
bit conservative about generating 'gotol' insn (32-bit branch offset)
as it didn't precise count the number of insns (e.g., some insns are
debug insns, etc.). Compared to old 'goto' insn, newer 'gotol' insn
should have comparable verification states to 'goto' insn.

With current patch set, all selftests passed with -mcpu=v4
when running test_progs-cpuv4 binary. The -mcpu=v3 and -mcpu=v2 run
are also successful.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011250.3718252-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/bpf: Fix a test_verifier failure
Yonghong Song [Fri, 28 Jul 2023 01:12:44 +0000 (18:12 -0700)]
selftests/bpf: Fix a test_verifier failure

The following test_verifier subtest failed due to
new encoding for BSWAP.

  $ ./test_verifier
  ...
  #99/u invalid 64-bit BPF_END FAIL
  Unexpected success to load!
  verification time 215 usec
  stack depth 0
  processed 3 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
  #99/p invalid 64-bit BPF_END FAIL
  Unexpected success to load!
  verification time 198 usec
  stack depth 0
  processed 3 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0

Tighten the test so it still reports a failure.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011244.3717464-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Add kernel/bpftool asm support for new instructions
Yonghong Song [Wed, 28 Jun 2023 22:29:51 +0000 (15:29 -0700)]
bpf: Add kernel/bpftool asm support for new instructions

Add asm support for new instructions so kernel verifier and bpftool
xlated insn dumps can have proper asm syntax for new instructions.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Support new 32bit offset jmp instruction
Yonghong Song [Fri, 28 Jul 2023 01:12:31 +0000 (18:12 -0700)]
bpf: Support new 32bit offset jmp instruction

Add interpreter/jit/verifier support for 32bit offset jmp instruction.
If a conditional jmp instruction needs more than 16bit offset,
it can be simulated with a conditional jmp + a 32bit jmp insn.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011231.3716103-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Fix jit blinding with new sdiv/smov insns
Yonghong Song [Fri, 28 Jul 2023 01:12:25 +0000 (18:12 -0700)]
bpf: Fix jit blinding with new sdiv/smov insns

Handle new insns properly in bpf_jit_blind_insn() function.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011225.3715812-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Support new signed div/mod instructions.
Yonghong Song [Fri, 28 Jul 2023 01:12:19 +0000 (18:12 -0700)]
bpf: Support new signed div/mod instructions.

Add interpreter/jit support for new signed div/mod insns.
The new signed div/mod instructions are encoded with
unsigned div/mod instructions plus insn->off == 1.
Also add basic verifier support to ensure new insns get
accepted.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011219.3714605-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Support new unconditional bswap instruction
Yonghong Song [Fri, 28 Jul 2023 01:12:13 +0000 (18:12 -0700)]
bpf: Support new unconditional bswap instruction

The existing 'be' and 'le' insns will do conditional bswap
depends on host endianness. This patch implements
unconditional bswap insns.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011213.3712808-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Handle sign-extenstin ctx member accesses
Yonghong Song [Fri, 28 Jul 2023 01:12:07 +0000 (18:12 -0700)]
bpf: Handle sign-extenstin ctx member accesses

Currently, if user accesses a ctx member with signed types,
the compiler will generate an unsigned load followed by
necessary left and right shifts.

With the introduction of sign-extension load, compiler may
just emit a ldsx insn instead. Let us do a final movsx sign
extension to the final unsigned ctx load result to
satisfy original sign extension requirement.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011207.3712528-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Support new sign-extension mov insns
Yonghong Song [Fri, 28 Jul 2023 01:12:02 +0000 (18:12 -0700)]
bpf: Support new sign-extension mov insns

Add interpreter/jit support for new sign-extension mov insns.
The original 'MOV' insn is extended to support reg-to-reg
signed version for both ALU and ALU64 operations. For ALU mode,
the insn->off value of 8 or 16 indicates sign-extension
from 8- or 16-bit value to 32-bit value. For ALU64 mode,
the insn->off value of 8/16/32 indicates sign-extension
from 8-, 16- or 32-bit value to 64-bit value.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011202.3712300-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: Support new sign-extension load insns
Yonghong Song [Fri, 28 Jul 2023 01:11:56 +0000 (18:11 -0700)]
bpf: Support new sign-extension load insns

Add interpreter/jit support for new sign-extension load insns
which adds a new mode (BPF_MEMSX).
Also add verifier support to recognize these insns and to
do proper verification with new insns. In verifier, besides
to deduce proper bounds for the dst_reg, probed memory access
is also properly handled.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230728011156.3711870-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf, docs: fix BPF_NEG entry in instruction-set.rst
Jose E. Marchesi [Wed, 26 Jul 2023 09:25:43 +0000 (11:25 +0200)]
bpf, docs: fix BPF_NEG entry in instruction-set.rst

This patch fixes the documentation of the BPF_NEG instruction to
denote that it does not use the source register operand.

Signed-off-by: Jose E. Marchesi <jose.marchesi@oracle.com>
Acked-by: Dave Thaler <dthaler@microsoft.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230726092543.6362-1-jose.marchesi@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agobpf: work around -Wuninitialized warning
Arnd Bergmann [Tue, 25 Jul 2023 20:26:40 +0000 (22:26 +0200)]
bpf: work around -Wuninitialized warning

Splitting these out into separate helper functions means that we
actually pass an uninitialized variable into another function call
if dec_active() happens to not be inlined, and CONFIG_PREEMPT_RT
is disabled:

kernel/bpf/memalloc.c: In function 'add_obj_to_free_list':
kernel/bpf/memalloc.c:200:9: error: 'flags' is used uninitialized [-Werror=uninitialized]
  200 |         dec_active(c, flags);

Avoid this by passing the flags by reference, so they either get
initialized and dereferenced through a pointer, or the pointer never
gets accessed at all.

Fixes: 18e027b1c7c6d ("bpf: Factor out inc/dec of active flag into helpers.")
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20230725202653.2905259-1-arnd@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
15 months agoselftests/xsk: Fix spelling mistake "querrying" -> "querying"
Colin Ian King [Thu, 20 Jul 2023 10:48:15 +0000 (11:48 +0100)]
selftests/xsk: Fix spelling mistake "querrying" -> "querying"

There is a spelling mistake in an error message. Fix it.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Acked-by: Björn Töpel <bjorn@kernel.org>
Link: https://lore.kernel.org/r/20230720104815.123146-1-colin.i.king@gmail.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agoMerge branch 'Add SO_REUSEPORT support for TC bpf_sk_assign'
Martin KaFai Lau [Mon, 24 Jul 2023 20:28:41 +0000 (13:28 -0700)]
Merge branch 'Add SO_REUSEPORT support for TC bpf_sk_assign'

Lorenz Bauer says:

====================
We want to replace iptables TPROXY with a BPF program at TC ingress.
To make this work in all cases we need to assign a SO_REUSEPORT socket
to an skb, which is currently prohibited. This series adds support for
such sockets to bpf_sk_assing.

I did some refactoring to cut down on the amount of duplicate code. The
key to this is to use INDIRECT_CALL in the reuseport helpers. To show
that this approach is not just beneficial to TC sk_assign I removed
duplicate code for bpf_sk_lookup as well.

Joint work with Daniel Borkmann.

Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
---
Changes in v6:
- Reject unhashed UDP sockets in bpf_sk_assign to avoid ref leak
- Link to v5: https://lore.kernel.org/r/20230613-so-reuseport-v5-0-f6686a0dbce0@isovalent.com

Changes in v5:
- Drop reuse_sk == sk check in inet[6]_steal_stock (Kuniyuki)
- Link to v4: https://lore.kernel.org/r/20230613-so-reuseport-v4-0-4ece76708bba@isovalent.com

Changes in v4:
- WARN_ON_ONCE if reuseport socket is refcounted (Kuniyuki)
- Use inet[6]_ehashfn_t to shorten function declarations (Kuniyuki)
- Shuffle documentation patch around (Kuniyuki)
- Update commit message to explain why IPv6 needs EXPORT_SYMBOL
- Link to v3: https://lore.kernel.org/r/20230613-so-reuseport-v3-0-907b4cbb7b99@isovalent.com

Changes in v3:
- Fix warning re udp_ehashfn and udp6_ehashfn (Simon)
- Return higher scoring connected UDP reuseport sockets (Kuniyuki)
- Fix ipv6 module builds
- Link to v2: https://lore.kernel.org/r/20230613-so-reuseport-v2-0-b7c69a342613@isovalent.com

Changes in v2:
- Correct commit abbrev length (Kuniyuki)
- Reduce duplication (Kuniyuki)
- Add checks on sk_state (Martin)
- Split exporting inet[6]_lookup_reuseport into separate patch (Eric)

---
Daniel Borkmann (1):
      selftests/bpf: Test that SO_REUSEPORT can be used with sk_assign helper
====================

Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agoselftests/bpf: Test that SO_REUSEPORT can be used with sk_assign helper
Daniel Borkmann [Thu, 20 Jul 2023 15:30:12 +0000 (17:30 +0200)]
selftests/bpf: Test that SO_REUSEPORT can be used with sk_assign helper

We use two programs to check that the new reuseport logic is executed
appropriately.

The first is a TC clsact program which bpf_sk_assigns
the skb to a UDP or TCP socket created by user space. Since the test
communicates via lo we see both directions of packets in the eBPF.
Traffic ingressing to the reuseport socket is identified by looking
at the destination port. For TCP, we additionally need to make sure
that we only assign the initial SYN packets towards our listening
socket. The network stack then creates a request socket which
transitions to ESTABLISHED after the 3WHS.

The second is a reuseport program which shares the fact that
it has been executed with user space. This tells us that the delayed
lookup mechanism is working.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Co-developed-by: Lorenz Bauer <lmb@isovalent.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Cc: Joe Stringer <joe@cilium.io>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-8-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agobpf, net: Support SO_REUSEPORT sockets with bpf_sk_assign
Lorenz Bauer [Thu, 20 Jul 2023 15:30:11 +0000 (17:30 +0200)]
bpf, net: Support SO_REUSEPORT sockets with bpf_sk_assign

Currently the bpf_sk_assign helper in tc BPF context refuses SO_REUSEPORT
sockets. This means we can't use the helper to steer traffic to Envoy,
which configures SO_REUSEPORT on its sockets. In turn, we're blocked
from removing TPROXY from our setup.

The reason that bpf_sk_assign refuses such sockets is that the
bpf_sk_lookup helpers don't execute SK_REUSEPORT programs. Instead,
one of the reuseport sockets is selected by hash. This could cause
dispatch to the "wrong" socket:

    sk = bpf_sk_lookup_tcp(...) // select SO_REUSEPORT by hash
    bpf_sk_assign(skb, sk) // SK_REUSEPORT wasn't executed

Fixing this isn't as simple as invoking SK_REUSEPORT from the lookup
helpers unfortunately. In the tc context, L2 headers are at the start
of the skb, while SK_REUSEPORT expects L3 headers instead.

Instead, we execute the SK_REUSEPORT program when the assigned socket
is pulled out of the skb, further up the stack. This creates some
trickiness with regards to refcounting as bpf_sk_assign will put both
refcounted and RCU freed sockets in skb->sk. reuseport sockets are RCU
freed. We can infer that the sk_assigned socket is RCU freed if the
reuseport lookup succeeds, but convincing yourself of this fact isn't
straight forward. Therefore we defensively check refcounting on the
sk_assign sock even though it's probably not required in practice.

Fixes: 8e368dc72e86 ("bpf: Fix use of sk->sk_reuseport from sk_assign")
Fixes: cf7fbe660f2d ("bpf: Add socket assign support")
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Joe Stringer <joe@cilium.io>
Link: https://lore.kernel.org/bpf/CACAyw98+qycmpQzKupquhkxbvWK4OFyDuuLMBNROnfWMZxUWeA@mail.gmail.com/
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-7-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agonet: remove duplicate sk_lookup helpers
Lorenz Bauer [Thu, 20 Jul 2023 15:30:10 +0000 (17:30 +0200)]
net: remove duplicate sk_lookup helpers

Now that inet[6]_lookup_reuseport are parameterised on the ehashfn
we can remove two sk_lookup helpers.

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-6-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agonet: document inet[6]_lookup_reuseport sk_state requirements
Lorenz Bauer [Thu, 20 Jul 2023 15:30:09 +0000 (17:30 +0200)]
net: document inet[6]_lookup_reuseport sk_state requirements

The current implementation was extracted from inet[6]_lhash2_lookup
in commit 80b373f74f9e ("inet: Extract helper for selecting socket
from reuseport group") and commit 5df6531292b5 ("inet6: Extract helper
for selecting socket from reuseport group"). In the original context,
sk is always in TCP_LISTEN state and so did not have a separate check.

Add documentation that specifies which sk_state are valid to pass to
the function.

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-5-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agonet: remove duplicate reuseport_lookup functions
Lorenz Bauer [Thu, 20 Jul 2023 15:30:08 +0000 (17:30 +0200)]
net: remove duplicate reuseport_lookup functions

There are currently four copies of reuseport_lookup: one each for
(TCP, UDP)x(IPv4, IPv6). This forces us to duplicate all callers of
those functions as well. This is already the case for sk_lookup
helpers (inet,inet6,udp4,udp6)_lookup_run_bpf.

There are two differences between the reuseport_lookup helpers:

1. They call different hash functions depending on protocol
2. UDP reuseport_lookup checks that sk_state != TCP_ESTABLISHED

Move the check for sk_state into the caller and use the INDIRECT_CALL
infrastructure to cut down the helpers to one per IP version.

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-4-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agonet: export inet_lookup_reuseport and inet6_lookup_reuseport
Lorenz Bauer [Thu, 20 Jul 2023 15:30:07 +0000 (17:30 +0200)]
net: export inet_lookup_reuseport and inet6_lookup_reuseport

Rename the existing reuseport helpers for IPv4 and IPv6 so that they
can be invoked in the follow up commit. Export them so that building
DCCP and IPv6 as a module works.

No change in functionality.

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-3-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agobpf: reject unhashed sockets in bpf_sk_assign
Lorenz Bauer [Thu, 20 Jul 2023 15:30:06 +0000 (17:30 +0200)]
bpf: reject unhashed sockets in bpf_sk_assign

The semantics for bpf_sk_assign are as follows:

    sk = some_lookup_func()
    bpf_sk_assign(skb, sk)
    bpf_sk_release(sk)

That is, the sk is not consumed by bpf_sk_assign. The function
therefore needs to make sure that sk lives long enough to be
consumed from __inet_lookup_skb. The path through the stack for a
TCPv4 packet is roughly:

  netif_receive_skb_core: takes RCU read lock
    __netif_receive_skb_core:
      sch_handle_ingress:
        tcf_classify:
          bpf_sk_assign()
      deliver_ptype_list_skb:
        deliver_skb:
          ip_packet_type->func == ip_rcv:
            ip_rcv_core:
            ip_rcv_finish_core:
              dst_input:
                ip_local_deliver:
                  ip_local_deliver_finish:
                    ip_protocol_deliver_rcu:
                      tcp_v4_rcv:
                        __inet_lookup_skb:
                          skb_steal_sock

The existing helper takes advantage of the fact that everything
happens in the same RCU critical section: for sockets with
SOCK_RCU_FREE set bpf_sk_assign never takes a reference.
skb_steal_sock then checks SOCK_RCU_FREE again and does sock_put
if necessary.

This approach assumes that SOCK_RCU_FREE is never set on a sk
between bpf_sk_assign and skb_steal_sock, but this invariant is
violated by unhashed UDP sockets. A new UDP socket is created
in TCP_CLOSE state but without SOCK_RCU_FREE set. That flag is only
added in udp_lib_get_port() which happens when a socket is bound.

When bpf_sk_assign was added it wasn't possible to access unhashed
UDP sockets from BPF, so this wasn't a problem. This changed
in commit 0c48eefae712 ("sock_map: Lift socket state restriction
for datagram sockets"), but the helper wasn't adjusted accordingly.
The following sequence of events will therefore lead to a refcount
leak:

1. Add socket(AF_INET, SOCK_DGRAM) to a sockmap.
2. Pull socket out of sockmap and bpf_sk_assign it. Since
   SOCK_RCU_FREE is not set we increment the refcount.
3. bind() or connect() the socket, setting SOCK_RCU_FREE.
4. skb_steal_sock will now set refcounted = false due to
   SOCK_RCU_FREE.
5. tcp_v4_rcv() skips sock_put().

Fix the problem by rejecting unhashed sockets in bpf_sk_assign().
This matches the behaviour of __inet_lookup_skb which is ultimately
the goal of bpf_sk_assign().

Fixes: cf7fbe660f2d ("bpf: Add socket assign support")
Cc: Joe Stringer <joe@cilium.io>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-2-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agoudp: re-score reuseport groups when connected sockets are present
Lorenz Bauer [Thu, 20 Jul 2023 15:30:05 +0000 (17:30 +0200)]
udp: re-score reuseport groups when connected sockets are present

Contrary to TCP, UDP reuseport groups can contain TCP_ESTABLISHED
sockets. To support these properly we remember whether a group has
a connected socket and skip the fast reuseport early-return. In
effect we continue scoring all reuseport sockets and then choose the
one with the highest score.

The current code fails to re-calculate the score for the result of
lookup_reuseport. According to Kuniyuki Iwashima:

    1) SO_INCOMING_CPU is set
       -> selected sk might have +1 score

    2) BPF prog returns ESTABLISHED and/or SO_INCOMING_CPU sk
       -> selected sk will have more than 8

  Using the old score could trigger more lookups depending on the
  order that sockets are created.

    sk -> sk (SO_INCOMING_CPU) -> sk (ESTABLISHED)
    |     |
    `-> select the next SO_INCOMING_CPU sk
          |
          `-> select itself (We should save this lookup)

Fixes: efc6b6f6c311 ("udp: Improve load balancing for SO_REUSEPORT.")
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230720-so-reuseport-v6-1-7021b683cdae@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agoMAINTAINERS: Replace my email address
Yonghong Song [Tue, 25 Jul 2023 05:41:00 +0000 (22:41 -0700)]
MAINTAINERS: Replace my email address

Switch from corporate email address to linux.dev address.

Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20230725054100.1013421-1-yonghong.song@linux.dev
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
15 months agotcx: Fix splat in ingress_destroy upon tcx_entry_free
Daniel Borkmann [Fri, 21 Jul 2023 23:33:30 +0000 (01:33 +0200)]
tcx: Fix splat in ingress_destroy upon tcx_entry_free

On qdisc destruction, the ingress_destroy() needs to update the correct
entry, that is, tcx_entry_update must NULL the dev->tcx_ingress pointer.
Therefore, fix the typo.

Fixes: e420bed02507 ("bpf: Add fd-based tcx multi-prog infra with link support")
Reported-by: syzbot+bdcf141f362ef83335cf@syzkaller.appspotmail.com
Reported-by: syzbot+b202b7208664142954fa@syzkaller.appspotmail.com
Reported-by: syzbot+14736e249bce46091c18@syzkaller.appspotmail.com
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: syzbot+bdcf141f362ef83335cf@syzkaller.appspotmail.com
Tested-by: syzbot+b202b7208664142954fa@syzkaller.appspotmail.com
Tested-by: syzbot+14736e249bce46091c18@syzkaller.appspotmail.com
Tested-by: Petr Machata <petrm@nvidia.com>
Link: https://lore.kernel.org/r/20230721233330.5678-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoMerge branch 'ionic-FLR-support'
David S. Miller [Mon, 24 Jul 2023 09:42:34 +0000 (10:42 +0100)]
Merge branch 'ionic-FLR-support'

Shannon Nelson says:

====================
ionic: add FLR support

Add support for handing and recovering from a PCI FLR event.
This patchset first moves some code around to make it usable
from multiple paths, then adds the PCI error handler callbacks
for reset_prepare and reset_done.

Example test:
    echo 1 > /sys/bus/pci/devices/0000:2a:00.0/reset

v4:
 - don't remove ionic_dev_teardown() in ionic_probe() in patch 2/4
 - remove clear_bit() change from patch 3/4

v3:
Link: https://lore.kernel.org/netdev/20230717170001.30539-1-shannon.nelson@amd.com/
 - removed first patch, it is already merged into net

v2:
Link: https://lore.kernel.org/netdev/20230713192936.45152-1-shannon.nelson@amd.com/
 - removed redundant pci_save/restore_state() calls
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoionic: add FLR recovery support
Shannon Nelson [Thu, 20 Jul 2023 19:08:16 +0000 (12:08 -0700)]
ionic: add FLR recovery support

Add support for the PCI reset handlers in order to manage an FLR event.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoionic: pull out common bits from fw_up
Shannon Nelson [Thu, 20 Jul 2023 19:08:15 +0000 (12:08 -0700)]
ionic: pull out common bits from fw_up

Pull out some code from ionic_lif_handle_fw_up() that can be
used in the coming FLR recovery patch.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoionic: extract common bits from ionic_probe
Shannon Nelson [Thu, 20 Jul 2023 19:08:14 +0000 (12:08 -0700)]
ionic: extract common bits from ionic_probe

Pull out some chunks of code from ionic_probe() that will
be common in rebuild paths.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoionic: extract common bits from ionic_remove
Shannon Nelson [Thu, 20 Jul 2023 19:08:13 +0000 (12:08 -0700)]
ionic: extract common bits from ionic_remove

Pull out a chunk of code from ionic_remove() that will
be common in teardown paths.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoMerge branch 'phy-motorcomm-driver-strength'
David S. Miller [Mon, 24 Jul 2023 09:36:45 +0000 (10:36 +0100)]
Merge branch 'phy-motorcomm-driver-strength'

Samin Guo says:

====================
Add motorcomm phy pad-driver-strength-cfg support

The motorcomm phy (YT8531) supports the ability to adjust the drive
strength of the rx_clk/rx_data, and the default strength may not be
suitable for all boards. So add configurable options to better match
the boards.(e.g. StarFive VisionFive 2)

The first patch adds a description of dt-bingding, and the second patch adds
YT8531's parsing and settings for pad-driver-strength-cfg.

Changes since v4:
Patch 1:
- Removed register-related DS(3b) values and added vol descriptions (by Andrew Lunn)
- Dropped the type and added '-microamp' suffix. (by Rob Herring)
Patch 2:
- Return -EINVAL if the value in DT but it is invalid (by Andrew Lunn)

Changes since v3:
Patch 1:
- Used current values instead of register values
- Added units and numerical descriptions of driver-strength
Patch 2:
- Added a lookup table to listing the valid values in the schema (by Andrew Lunn)

Changes since v2:
Patch 2:
- Readjusted the order of YT8531_RGMII_xxx to below YTPHY_PAD_DRIVE_STRENGTH_REG (by Frank Sae)
- Reversed Christmas tree, sort these longest first, shortest last (by Andrew Lunn)
- Rebased on tag v6.4

Changes since v1:
Patch 1:
- Renamed "rx-xxx-driver-strength" to "motorcomm,rx-xxx-driver-strength" (by Frank Sae)
Patch 2:
- Added default values for rxc/rxd driver strength (by Frank Sea/Andrew Lunn)
- Added range checking when val is in DT (by Frank Sea/Andrew Lunn)

Previous versions:
v1 - https://patchwork.kernel.org/project/netdevbpf/cover/20230426063541.15378-1-samin.guo@starfivetech.com
v2 - https://patchwork.kernel.org/project/netdevbpf/cover/20230505090558.2355-1-samin.guo@starfivetech.com
v3 - https://patchwork.kernel.org/project/netdevbpf/cover/20230526090502.29835-1-samin.guo@starfivetech.com
v4 - https://patchwork.kernel.org/project/netdevbpf/cover/20230714101406.17686-1-samin.guo@starfivetech.com
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: phy: motorcomm: Add pad drive strength cfg support
Samin Guo [Thu, 20 Jul 2023 11:15:09 +0000 (19:15 +0800)]
net: phy: motorcomm: Add pad drive strength cfg support

The motorcomm phy (YT8531) supports the ability to adjust the drive
strength of the rx_clk/rx_data, and the default strength may not be
suitable for all boards. So add configurable options to better match
the boards.(e.g. StarFive VisionFive 2)

When we configure the drive strength, we need to read the current
LDO voltage value to ensure that it is a legal value at that LDO
voltage.

Reviewed-by: Hal Feng <hal.feng@starfivetech.com>
Signed-off-by: Samin Guo <samin.guo@starfivetech.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agodt-bindings: net: motorcomm: Add pad driver strength cfg
Samin Guo [Thu, 20 Jul 2023 11:15:08 +0000 (19:15 +0800)]
dt-bindings: net: motorcomm: Add pad driver strength cfg

The motorcomm phy (YT8531) supports the ability to adjust the drive
strength of the rx_clk/rx_data.

The YT8531 RGMII LDO voltage supports 1.8V/3.3V, and the
LDO voltage can be configured with hardware pull-up resistors to match
the SOC voltage (usually 1.8V). The software can read the registers
0xA001 obtain the current LDO voltage value.

Reviewed-by: Hal Feng <hal.feng@starfivetech.com>
Signed-off-by: Samin Guo <samin.guo@starfivetech.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoipv6: remove hard coded limitation on ipv6_pinfo
Eric Dumazet [Thu, 20 Jul 2023 11:09:01 +0000 (11:09 +0000)]
ipv6: remove hard coded limitation on ipv6_pinfo

IPv6 inet sockets are supposed to have a "struct ipv6_pinfo"
field at the end of their definition, so that inet6_sk_generic()
can derive from socket size the offset of the "struct ipv6_pinfo".

This is very fragile, and prevents adding bigger alignment
in sockets, because inet6_sk_generic() does not work
if the compiler adds padding after the ipv6_pinfo component.

We are currently working on a patch series to reorganize
TCP structures for better data locality and found issues
similar to the one fixed in commit f5d547676ca0
("tcp: fix tcp_inet6_sk() for 32bit kernels")

Alternative would be to force an alignment on "struct ipv6_pinfo",
greater or equal to __alignof__(any ipv6 sock) to ensure there is
no padding. This does not look great.

v2: fix typo in mptcp_proto_v6_init() (Paolo)

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Chao Wu <wwchao@google.com>
Cc: Wei Wang <weiwan@google.com>
Cc: Coco Li <lixiaoyan@google.com>
Cc: YiFei Zhu <zhuyifei@google.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: add sysctl accept_ra_min_rtr_lft
Patrick Rohr [Wed, 19 Jul 2023 14:52:13 +0000 (07:52 -0700)]
net: add sysctl accept_ra_min_rtr_lft

This change adds a new sysctl accept_ra_min_rtr_lft to specify the
minimum acceptable router lifetime in an RA. If the received RA router
lifetime is less than the configured value (and not 0), the RA is
ignored.
This is useful for mobile devices, whose battery life can be impacted
by networks that configure RAs with a short lifetime. On such networks,
the device should never gain IPv6 provisioning and should attempt to
drop RAs via hardware offload, if available.

Signed-off-by: Patrick Rohr <prohr@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: dsa: remove deprecated strncpy
justinstitt@google.com [Tue, 18 Jul 2023 22:56:38 +0000 (22:56 +0000)]
net: dsa: remove deprecated strncpy

`strncpy` is deprecated for use on NUL-terminated destination strings [1].

Even call sites utilizing length-bounded destination buffers should
switch over to using `strtomem` or `strtomem_pad`. In this case,
however, the compiler is unable to determine the size of the `data`
buffer which renders `strtomem` unusable. Due to this, `strscpy`
should be used.

It should be noted that most call sites already zero-initialize the
destination buffer. However, I've opted to use `strscpy_pad` to maintain
the same exact behavior that `strncpy` produced (zero-padded tail up to
`len`).

Also see [3].

[1]: www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings
[2]: elixir.bootlin.com/linux/v6.3/source/net/ethtool/ioctl.c#L1944
[3]: manpages.debian.org/testing/linux-manual-4.8/strscpy.9.en.html

Link: https://github.com/KSPP/linux/issues/90
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Justin Stitt <justinstitt@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoMerge branch 'process-connector-bug-fixes-and-enhancements'
David S. Miller [Sun, 23 Jul 2023 10:34:22 +0000 (11:34 +0100)]
Merge branch 'process-connector-bug-fixes-and-enhancements'

Anjali Kulkarni says:

====================
Process connector bug fixes & enhancements

Oracle DB is trying to solve a performance overhead problem it has been
facing for the past 10 years and using this patch series, we can fix this
issue.

Oracle DB runs on a large scale with 100000s of short lived processes,
starting up and exiting quickly. A process monitoring DB daemon which
tracks and cleans up after processes that have died without a proper exit
needs notifications only when a process died with a non-zero exit code
(which should be rare).

Due to the pmon architecture, which is distributed, each process is
independent and has minimal interaction with pmon. Hence fd based
solutions to track a process's spawning and exit cannot be used. Pmon
needs to detect the abnormal death of a process so it can cleanup after.
Currently it resorts to checking /proc every few seconds. Other methods
we tried like using system call to reduce the above overhead were not
accepted upstream.

With this change, we add event based filtering to proc connector module
so that DB can only listen to the events it is interested in. A new
event type PROC_EVENT_NONZERO_EXIT is added, which is only sent by kernel
to a listening application when any process exiting has a non-zero exit
status.

This change will give Oracle DB substantial performance savings - it takes
50ms to scan about 8K PIDs in /proc, about 500ms for 100K PIDs. DB does
this check every 3 secs, so over an hour we save 10secs for 100K PIDs.

With this, a client can register to listen for only exit or fork or a mix or
all of the events. This greatly enhances performance - currently, we
need to listen to all events, and there are 9 different types of events.
For eg. handling 3 types of events - 8K-forks + 8K-exits + 8K-execs takes
200ms, whereas handling 2 types - 8K-forks + 8K-exits takes about 150ms,
and handling just one type - 8K exits takes about 70ms.

Measuring the time using pidfds for monitoring 8K process exits took 4
times longer - 200ms, as compared to 70ms using only exit notifications
of proc connector. Hence, we cannot use pidfd for our use case.

This kind of a new event could also be useful to other applications like
Google's lmkd daemon, which needs a killed process's exit notification.

This patch series is organized as follows -

Patch 1 : Needed for patch 3 to work.
Patch 2 : Needed for patch 3 to work.
Patch 3 : Fixes some bugs in proc connector, details in the patch.
Patch 4 : Adds event based filtering for performance enhancements.
Patch 5 : Allow non-root users access to proc connector events.
Patch 6 : Selftest code for proc connector.

v9->v10 changes:
- Rebased to net-next, re-compiled and re-tested.

v8->v9 changes:
- Added sha1 ("title") of reversed patch as suggested by Eric Dumazet.

v7->v8 changes:
- Fixed an issue pointed by Liam Howlett in v7.

v6->v7 changes:
- Incorporated Liam Howlett's comments on v6
- Incorporated Kalesh Anakkur Purayil's comments

v5->v6 changes:
- Incorporated Liam Howlett's comments
- Removed FILTER define from proc_filter.c and added a "-f" run-time
  option to run new filter code.
- Made proc_filter.c a selftest in tools/testing/selftests/connector

v4->v5 changes:
- Change the cover letter
- Fix a small issue in proc_filter.c

v3->v4 changes:
- Fix comments by Jakub Kicinski to incorporate root access changes
  within bind call of connector

v2->v3 changes:
- Fix comments by Jakub Kicinski to separate netlink (patch 2) (after
  layering) from connector fixes (patch 3).
- Minor fixes suggested by Jakub.
- Add new multicast group level permissions check at netlink layer.
  Split this into netlink & connector layers (patches 6 & 7)

v1->v2 changes:
- Fix comments by Jakub Kicinski to keep layering within netlink and
  update kdocs.
- Move non-root users access patch last in series so remaining patches
  can go in first.

v->v1 changes:
- Changed commit log in patch 4 as suggested by Christian Brauner
- Changed patch 4 to make more fine grained access to non-root users
- Fixed warning in cn_proc.c,
Reported-by: kernel test robot <lkp@intel.com>
- Fixed some existing warnings in cn_proc.c
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoconnector/cn_proc: Selftest for proc connector
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:21 +0000 (13:18 -0700)]
connector/cn_proc: Selftest for proc connector

Run as ./proc_filter -f to run new filter code. Run without "-f" to run
usual proc connector code without the new filtering code.

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoconnector/cn_proc: Allow non-root users access
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:20 +0000 (13:18 -0700)]
connector/cn_proc: Allow non-root users access

There were a couple of reasons for not allowing non-root users access
initially  - one is there was some point no proper receive buffer
management in place for netlink multicast. But that should be long
fixed. See link below for more context.

Second is that some of the messages may contain data that is root only. But
this should be handled with a finer granularity, which is being done at the
protocol layer.  The only problematic protocols are nf_queue and the
firewall netlink. Hence, this restriction for non-root access was relaxed
for NETLINK_ROUTE initially:
https://lore.kernel.org/all/20020612013101.A22399@wotan.suse.de/

This restriction has also been removed for following protocols:
NETLINK_KOBJECT_UEVENT, NETLINK_AUDIT, NETLINK_SOCK_DIAG,
NETLINK_GENERIC, NETLINK_SELINUX.

Since process connector messages are not sensitive (process fork, exit
notifications etc.), and anyone can read /proc data, we can allow non-root
access here. However, since process event notification is not the only
consumer of NETLINK_CONNECTOR, we can make this change even more
fine grained than the protocol level, by checking for multicast group
within the protocol.

Allow non-root access for NETLINK_CONNECTOR via NL_CFG_F_NONROOT_RECV
but add new bind function cn_bind(), which allows non-root access only
for CN_IDX_PROC multicast group.

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoconnector/cn_proc: Performance improvements
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:19 +0000 (13:18 -0700)]
connector/cn_proc: Performance improvements

This patch adds the capability to filter messages sent by the proc
connector on the event type supplied in the message from the client
to the connector. The client can register to listen for an event type
given in struct proc_input.

This event based filteting will greatly enhance performance - handling
8K exits takes about 70ms, whereas 8K-forks + 8K-exits takes about 150ms
& handling 8K-forks + 8K-exits + 8K-execs takes 200ms. There are currently
9 different types of events, and we need to listen to all of them. Also,
measuring the time using pidfds for monitoring 8K process exits took
much longer - 200ms, as compared to 70ms using only exit notifications of
proc connector.

We also add a new event type - PROC_EVENT_NONZERO_EXIT, which is
only sent by kernel to a listening application when any process exiting,
has a non-zero exit status. This will help the clients like Oracle DB,
where a monitoring process wants notfications for non-zero process exits
so it can cleanup after them.

This kind of a new event could also be useful to other applications like
Google's lmkd daemon, which needs a killed process's exit notification.

The patch takes care that existing clients using old mechanism of not
sending the event type work without any changes.

cn_filter function checks to see if the event type being notified via
proc connector matches the event type requested by client, before
sending(matches) or dropping(does not match) a packet.

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoconnector/cn_proc: Add filtering to fix some bugs
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:18 +0000 (13:18 -0700)]
connector/cn_proc: Add filtering to fix some bugs

The current proc connector code has the foll. bugs - if there are more
than one listeners for the proc connector messages, and one of them
deregisters for listening using PROC_CN_MCAST_IGNORE, they will still get
all proc connector messages, as long as there is another listener.

Another issue is if one client calls PROC_CN_MCAST_LISTEN, and another one
calls PROC_CN_MCAST_IGNORE, then both will end up not getting any messages.

This patch adds filtering and drops packet if client has sent
PROC_CN_MCAST_IGNORE. This data is stored in the client socket's
sk_user_data. In addition, we only increment or decrement
proc_event_num_listeners once per client. This fixes the above issues.

cn_release is the release function added for NETLINK_CONNECTOR. It uses
the newly added netlink_release function added to netlink_sock. It will
free sk_user_data.

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonetlink: Add new netlink_release function
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:17 +0000 (13:18 -0700)]
netlink: Add new netlink_release function

A new function netlink_release is added in netlink_sock to store the
protocol's release function. This is called when the socket is deleted.
This can be supplied by the protocol via the release function in
netlink_kernel_cfg. This is being added for the NETLINK_CONNECTOR
protocol, so it can free it's data when socket is deleted.

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonetlink: Reverse the patch which removed filtering
Anjali Kulkarni [Wed, 19 Jul 2023 20:18:16 +0000 (13:18 -0700)]
netlink: Reverse the patch which removed filtering

To use filtering at the connector & cn_proc layers, we need to enable
filtering in the netlink layer. This reverses the patch which removed
netlink filtering - commit ID for that patch:
549017aa1bb7 (netlink: remove netlink_broadcast_filtered).

Signed-off-by: Anjali Kulkarni <anjali.k.kulkarni@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoMerge branch 'net-page_pool-remove-page_pool_release_page'
Jakub Kicinski [Sat, 22 Jul 2023 01:50:39 +0000 (18:50 -0700)]
Merge branch 'net-page_pool-remove-page_pool_release_page'

Jakub Kicinski says:

====================
net: page_pool: remove page_pool_release_page()

page_pool_return_page() is a historic artefact from before
recycling of pages attached to skbs was supported. Theoretical
uses for it may be thought up but in practice all existing
users can be converted to use skb_mark_for_recycle() instead.

This code was previously posted as part of the memory provider RFC.
https://lore.kernel.org/all/20230707183935.997267-1-kuba@kernel.org/
====================

Link: https://lore.kernel.org/r/20230720010409.1967072-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agonet: page_pool: merge page_pool_release_page() with page_pool_return_page()
Jakub Kicinski [Thu, 20 Jul 2023 01:04:09 +0000 (18:04 -0700)]
net: page_pool: merge page_pool_release_page() with page_pool_return_page()

Now that page_pool_release_page() is not exported we can
merge it with page_pool_return_page(). I believe that
the "Do not replace this with page_pool_return_page()"
comment was there in case page_pool_return_page() was
not inlined, to avoid two function calls.

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Reviewed-by: Yunsheng Lin <linyunsheng@huawei.com>
Link: https://lore.kernel.org/r/20230720010409.1967072-5-kuba@kernel.org
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agonet: page_pool: hide page_pool_release_page()
Jakub Kicinski [Thu, 20 Jul 2023 01:04:08 +0000 (18:04 -0700)]
net: page_pool: hide page_pool_release_page()

There seems to be no user calling page_pool_release_page()
for legit reasons, all the users simply haven't been converted
to skb-based recycling, yet. Previous changes converted them.
Update the docs, and unexport the function.

Link: https://lore.kernel.org/r/20230720010409.1967072-4-kuba@kernel.org
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoeth: stmmac: let page recycling happen with skbs
Jakub Kicinski [Thu, 20 Jul 2023 01:04:07 +0000 (18:04 -0700)]
eth: stmmac: let page recycling happen with skbs

stmmac removes pages from the page pool after attaching them
to skbs. Use page recycling instead.

skb heads are always copied, and pages are always from page
pool in this driver. We could as well mark all allocated skbs
for recycling.

Reviewed-by: Yunsheng Lin <linyunsheng@huawei.com>
Link: https://lore.kernel.org/r/20230720010409.1967072-3-kuba@kernel.org
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoeth: tsnep: let page recycling happen with skbs
Jakub Kicinski [Thu, 20 Jul 2023 01:04:06 +0000 (18:04 -0700)]
eth: tsnep: let page recycling happen with skbs

tsnep builds an skb with napi_build_skb() and then calls
page_pool_release_page() for the page in which that skb's
head sits. Use recycling instead, recycling of heads works
just fine.

Reviewed-by: Yunsheng Lin <linyunsheng@huawei.com>
Link: https://lore.kernel.org/r/20230720010409.1967072-2-kuba@kernel.org
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agogenetlink: add explicit ordering break check for split ops
Jiri Pirko [Thu, 20 Jul 2023 11:13:54 +0000 (13:13 +0200)]
genetlink: add explicit ordering break check for split ops

Currently, if cmd in the split ops array is of lower value than the
previous one, genl_validate_ops() continues to do the checks as if
the values are equal. This may result in non-obvious WARN_ON() hit in
these check.

Instead, check the incorrect ordering explicitly and put a WARN_ON()
in case it is broken.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230720111354.562242-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoMAINTAINERS: net: fix sort order
Marc Kleine-Budde [Thu, 20 Jul 2023 15:11:07 +0000 (17:11 +0200)]
MAINTAINERS: net: fix sort order

Linus seems to like the MAINTAINERS file sorted, see
c192ac735768 ("MAINTAINERS 2: Electric Boogaloo").

Since this is currently not the case, restore the sort order.

Fixes: 3abf3d15ffff ("MAINTAINERS: ASP 2.0 Ethernet driver maintainers")
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Acked-by: Justin Chen <justin.chen@broadcom.com>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://lore.kernel.org/r/20230720151107.679668-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoMerge branch 'octeontx2-pf-round-robin-sched'
David S. Miller [Fri, 21 Jul 2023 08:55:54 +0000 (09:55 +0100)]
Merge branch 'octeontx2-pf-round-robin-sched'

Hariprasad Kelam says:

====================
octeontx2-pf: support Round Robin scheduling

octeontx2 and CN10K silicons support Round Robin scheduling. When multiple
traffic flows reach transmit level with the same priority, with Round Robin
scheduling traffic flow with the highest quantum value is picked. With this
support, the user can add multiple classes with the same priority and
different quantum in htb offload.

This series of patches adds support for the same.

Patch1: implement transmit schedular allocation algorithm as preparation
        for support round robin scheduling.

Patch2: Allow quantum parameter in HTB offload mode.

Patch3: extends octeontx2 htb offload support for Round Robin scheduling

Patch4: extend QOS documentation for Round Robin scheduling

Hariprasad Kelam (1):
  docs: octeontx2: extend documentation for Round Robin scheduling

Naveen Mamindlapalli (3):
  octeontx2-pf: implement transmit schedular allocation algorithm
  sch_htb: Allow HTB quantum parameter in offload mode
  octeontx2-pf: htb offload support for Round Robin scheduling
---
v4 * update classid values in documentation.

v3 * 1. update QOS documentation for round robin scheduling
     2. added out of bound checks for quantum parameter

v2 * change data type of otx2_index_used to reduce size of structure
     otx2_qos_cfg
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agodocs: octeontx2: extend documentation for Round Robin scheduling
Hariprasad Kelam [Wed, 19 Jul 2023 11:04:43 +0000 (16:34 +0530)]
docs: octeontx2: extend documentation for Round Robin scheduling

Add example tc-htb commands for Round robin scheduling

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoocteontx2-pf: htb offload support for Round Robin scheduling
Naveen Mamindlapalli [Wed, 19 Jul 2023 11:04:42 +0000 (16:34 +0530)]
octeontx2-pf: htb offload support for Round Robin scheduling

When multiple traffic flows reach Transmit level with the same
priority, with Round robin scheduling traffic flow with the highest
quantum value is picked. With this support, the user can add multiple
classes with the same priority and different quantum. This patch
does necessary changes to support the same.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agosch_htb: Allow HTB quantum parameter in offload mode
Naveen Mamindlapalli [Wed, 19 Jul 2023 11:04:41 +0000 (16:34 +0530)]
sch_htb: Allow HTB quantum parameter in offload mode

The current implementation of HTB offload returns the EINVAL error for
quantum parameter. This patch removes the error returning checks for
'quantum' parameter and populates its value to tc_htb_qopt_offload
structure such that driver can use the same.

Add quantum parameter check in mlx5 driver, as mlx5 devices are not capable
of supporting the quantum parameter when htb offload is used. Report error
if quantum parameter is set to a non-default value.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoocteontx2-pf: implement transmit schedular allocation algorithm
Naveen Mamindlapalli [Wed, 19 Jul 2023 11:04:40 +0000 (16:34 +0530)]
octeontx2-pf: implement transmit schedular allocation algorithm

unlike strict priority, where number of classes are limited to max
8, there is no restriction on the number of dwrr child nodes unless
the count increases the max number of child nodes supported.

Hardware expects strict priority transmit schedular indexes mapped
to their priority. This patch adds defines transmit schedular allocation
algorithm such that the above requirement is honored.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoMerge branch 'mlxsw-enslavement'
David S. Miller [Fri, 21 Jul 2023 07:54:06 +0000 (08:54 +0100)]
Merge branch 'mlxsw-enslavement'

Petr Machata says:

====================
mlxsw: Permit enslavement to netdevices with uppers

The mlxsw driver currently makes the assumption that the user applies
configuration in a bottom-up manner. Thus netdevices need to be added to
the bridge before IP addresses are configured on that bridge or SVI added
on top of it. Enslaving a netdevice to another netdevice that already has
uppers is in fact forbidden by mlxsw for this reason. Despite this safety,
it is rather easy to get into situations where the offloaded configuration
is just plain wrong.

As an example, take a front panel port, configure an IP address: it gets a
RIF. Now enslave the port to the bridge, and the RIF is gone. Remove the
port from the bridge again, but the RIF never comes back. There is a number
of similar situations, where changing the configuration there and back
utterly breaks the offload.

Similarly, detaching a front panel port from a configured topology means
unoffloading of this whole topology -- VLAN uppers, next hops, etc.
Attaching the port back is then not permitted at all. If it were, it would
not result in a working configuration, because much of mlxsw is written to
react to changes in immediate configuration. There is nothing that would go
visit netdevices in the attached-to topology and offload existing routes
and VLAN memberships, for example.

In this patchset, introduce a number of replays to be invoked so that this
sort of post-hoc offload is supported. Then remove the vetoes that
disallowed enslavement of front panel ports to other netdevices with
uppers.

The patchset progresses as follows:

- In patch #1, fix an issue in the bridge driver. To my knowledge, the
  issue could not have resulted in a buggy behavior previously, and thus is
  packaged with this patchset instead of being sent separately to net.

- In patch #2, add a new helper to the switchdev code.

- In patch #3, drop mlxsw selftests that will not be relevant after this
  patchset anymore.

- Patches #4, #5, #6, #7 and #8 prepare the codebase for smoother
  introduction of the rest of the code.

- Patches #9, #10, #11, #12, #13 and #14 replay various aspects of upper
  configuration when a front panel port is introduced into a topology.
  Individual patches take care of bridge and LAG RIF memberships, switchdev
  replay, nexthop and neighbors replay, and MACVLAN offload.

- Patches #15 and #16 introduce RIFs for newly-relevant netdevices when a
  front panel port is enslaved (in which case all uppers are newly
  relevant), or, respectively, deslaved (in which case the newly-relevant
  netdevice is the one being deslaved).

- Up until this point, the introduced scaffolding was not really used,
  because mlxsw still forbids enslavement of mlxsw netdevices to uppers
  with uppers. In patch #17, this condition is finally relaxed.

A sizable selftest suite is available to test all this new code. That will
be sent in a separate patchset.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum: Permit enslavement to netdevices with uppers
Petr Machata [Wed, 19 Jul 2023 11:01:32 +0000 (13:01 +0200)]
mlxsw: spectrum: Permit enslavement to netdevices with uppers

Enslaving of front panel ports (and their uppers) to netdevices that
already have uppers is currently forbidden. In the previous patches, a
number of replays have been added. Those ensure that various bits of state,
such as next hops or switchdev objects, are offloaded when they become
relevant due to a mlxsw lower being introduced into the topology.

However the act of actually, for example, enslaving a front-panel port to
a bridge with uppers, has been vetoed so far. In this patch, remove the
vetoes and permit the operation.

mlxsw currently validates creation of "interesting" uppers. Thus creating
VLAN netdevices on top of 802.1ad bridges is forbidden if the bridge has an
mlxsw lower, but permitted in general. This validation code never gets run
when a port is introduced as a lower of an existing netdevice structure.

Thus when enslaving an mlxsw netdevice to netdevices with uppers, invoke
the PRECHANGEUPPER event handler for each netdevice above the one that the
front panel port is being enslaved to. This way the tower of netdevices
above the attachment point is validated.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Replay IP NETDEV_UP on device deslavement
Petr Machata [Wed, 19 Jul 2023 11:01:31 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Replay IP NETDEV_UP on device deslavement

When a netdevice is removed from a bridge or a LAG, and it has an IP
address, it should join the router and gain a RIF. Do that by replaying
address addition event on the netdevice.

When handling deslavement of LAG or its upper from a bridge device, the
replay should be done after all the lowers of the LAG have left the bridge.
Thus these scenarios are handled by passing replay_deslavement of false,
and by invoking, after the lowers have been processed, a new helper,
mlxsw_sp_netdevice_post_lag_event(), which does the per-LAG / -upper
handling, and in particular invokes the replay.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Replay IP NETDEV_UP on device enslavement
Petr Machata [Wed, 19 Jul 2023 11:01:30 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Replay IP NETDEV_UP on device enslavement

Enslaving of front panel ports (and their uppers) to netdevices that
already have uppers is currently forbidden. When this is permitted, any
uppers with IP addresses need to have the NETDEV_UP inetaddr event
replayed, so that any RIFs are created.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Replay neighbours when RIF is made
Petr Machata [Wed, 19 Jul 2023 11:01:29 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Replay neighbours when RIF is made

As neighbours are created, mlxsw is involved through the netevent
notifications. When at the time there is no RIF for a given neighbour, the
notification is not acted upon. When the RIF is later created, these
outstanding neighbours are left unoffloaded and cause traffic to go through
the SW datapath.

In order to fix this issue, as a RIF is created, walk the ARP and ND tables
and find neighbours for the netdevice that represents the RIF. Then
schedule neighbour work for them, allowing them to be offloaded.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Replay MACVLANs when RIF is made
Petr Machata [Wed, 19 Jul 2023 11:01:28 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Replay MACVLANs when RIF is made

If IP address is added to a MACVLAN netdevice, the effect is of configuring
VRRP on the RIF for the netdevice linked to the MACVLAN. Because the
MACVLAN offload is tied to existence of a RIF at the linked netdevice,
adding a MACVLAN is currently not allowed until a RIF is present.

If this requirement stays, it will never be possible to attach a first port
into a topology that involves a MACVLAN. Thus topologies would need to be
built in a certain order, which is impractical.

Additionally, IP address removal, which leads to disappearance of the RIF
that the MACVLAN depends on, cannot be vetoed. Thus even as things stand
now it is possible to get to a state where a MACVLAN netdevice exists
without a RIF, despite having mlxsw lowers. And once the MACVLAN is
un-offloaded due to RIF getting destroyed, recreating the RIF does not
bring it back.

In this patch, accept that MACVLAN can be created out of order and support
that use case.

One option would seem to be to simply recognize MACVLAN netdevices as
"interesting", and let the existing replay mechanisms take care of the
offload. However, that does not address the necessity to reoffload MACVLAN
once a RIF is created.

Thus add a new replay hook, symmetrical to mlxsw_sp_rif_macvlan_flush(),
called mlxsw_sp_rif_macvlan_replay(), which instead of unwinding the
existing offloads, applies the configuration as if the netdevice were
created just now.

Additionally, remove all vetoes and warning messages that checked for
presence of a RIF at the linked device.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Offload ethernet nexthops when RIF is made
Petr Machata [Wed, 19 Jul 2023 11:01:27 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Offload ethernet nexthops when RIF is made

As RIF is created, refresh each netxhop group tracked at the CRIF for which
the RIF was created.

Note that nothing needs to be done for IPIP nexthops. The RIF for these is
either available from the get-go, or will never be available, so no after
the fact offloading needs to be done.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Join RIFs of LAG upper VLANs
Petr Machata [Wed, 19 Jul 2023 11:01:26 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Join RIFs of LAG upper VLANs

In the following patches, the requirement that ports be only enslaved to
masters without uppers, is going to be relaxed. It will therefore be
necessary to join not only RIF for the immediate LAG, as is currently the
case, but also RIFs for VLAN netdevices upper to the LAG.

In this patch, extend mlxsw_sp_netdevice_router_join_lag() to walk the
uppers of a LAG being joined, and also join any VLAN ones.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_switchdev: Replay switchdev objects on port join
Petr Machata [Wed, 19 Jul 2023 11:01:25 +0000 (13:01 +0200)]
mlxsw: spectrum_switchdev: Replay switchdev objects on port join

Currently it never happens that a netdevice that is already a bridge slave
would suddenly become mlxsw upper. The only case where this might be
possible as far as mlxsw is concerned, is with LAG netdevices. But if a LAG
has any upper (e.g. is enslaved), enlaving mlxsw port to that LAG is
forbidden. Thus the only way to install a LAG between a bridge and a mlxsw
port is by first enslaving the port to the LAG, and then enslaving that LAG
to a bridge. At that point there are no bridge objects (such as port VLANs)
to replay. Those are added afterwards, and notified as they are created.
This holds even for the PVID.

However in the following patches, the requirement that ports be only
enslaved to masters without uppers, is going to be relaxed. It will
therefore be necessary to replay the existing bridge objects. Without this
replay, e.g. the mlxsw bridge_port_vlan objects are not instantiated, which
causes issues later, as a lot of code relies on their presence.

To that end, add a new notifier block whose sole role is to filter out
events related to the one relevant upper, and forward those to the existing
switchdev notifier block. Pass the new notifier block to
switchdev_bridge_port_offload() when the bridge port is created.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum: On port enslavement to a LAG, join upper's bridges
Petr Machata [Wed, 19 Jul 2023 11:01:24 +0000 (13:01 +0200)]
mlxsw: spectrum: On port enslavement to a LAG, join upper's bridges

Currently it never happens that a netdevice that is already a bridge slave
would suddenly become mlxsw upper. The only case where this might be
possible as far as mlxsw is concerned, is with LAG netdevices. But if a LAG
already has an upper, enslaving mlxsw port to that LAG is forbidden. Thus
the only way to install a LAG between a bridge and a mlxsw port is by first
enslaving the port to the LAG, and then enslaving that LAG to a bridge.

However in the following patches, the requirement that ports be only
enslaved to masters without uppers, is going to be relaxed. It will
therefore be necessary to join bridges of LAG uppers. Without this replay,
the mlxsw bridge_port objects are not instantiated, which causes issues
later, as a lot of code relies on their presence.

Therefore in this patch, when the first mlxsw physical netdevice is
enslaved to a LAG, consider bridges upper to the LAG (both the direct
master, if any, and any bridge masters of VLAN uppers), and have the
relevant netdevices join their bridges.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum: Add a replay_deslavement argument to event handlers
Petr Machata [Wed, 19 Jul 2023 11:01:23 +0000 (13:01 +0200)]
mlxsw: spectrum: Add a replay_deslavement argument to event handlers

When handling deslavement of LAG or its upper from a bridge device, when
the deslaved netdevice has an IP address, it should join the router. This
should be done after all the lowers of the LAG have left the bridge. The
replay intended to cause the device to join the router therefore cannot be
invoked unconditionally in the event handlers themselves. It can be done
right away if the handler is invoked for a sole device, but when it is
invoked repeated for each LAG lower, the replay needs to be postponed
until after this processing is done.

To that end, add a boolean parameter, replay_deslavement, to
mlxsw_sp_netdevice_port_upper_event(), mlxsw_sp_netdevice_port_vlan_event()
and one helper on the call path. Have the invocations that are done for
sole netdevices pass true, and those done for LAG lowers pass false.

Nothing depends on this flag at this point, but it removes some noise from
the patch that introduces the replay itself.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum: Allow event handlers to check unowned bridges
Petr Machata [Wed, 19 Jul 2023 11:01:22 +0000 (13:01 +0200)]
mlxsw: spectrum: Allow event handlers to check unowned bridges

Currently the bridge-related handlers bail out when the event is related to
a netdevice that is not an upper of one of the front-panel ports. In order
to allow enslavement of front-panel ports to bridges that already have
uppers, it will be necessary to replay CHANGEUPPER events to validate that
the configuration is offloadable. In order for the replay to be effective,
it must be possible to ignore unsupported configuration in the context of
an actual notifier event, but to still "veto" these configurations when the
validation is performed.

To that end, introduce two parameters to a number of handlers: mlxsw_sp,
because it will not be possible to deduce that from the netdevice lowers;
and process_foreign to indicate whether netdevices that are not front panel
uppers should be validated.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum: Split a helper out of mlxsw_sp_netdevice_event()
Petr Machata [Wed, 19 Jul 2023 11:01:21 +0000 (13:01 +0200)]
mlxsw: spectrum: Split a helper out of mlxsw_sp_netdevice_event()

Move the meat of mlxsw_sp_netdevice_event() to a separate function that
does just the validation. This separate helper will be possible to call
later for recursive ascent when validating attachment of a front panel port
to a bridge with uppers.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Extract a helper to schedule neighbour work
Petr Machata [Wed, 19 Jul 2023 11:01:20 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Extract a helper to schedule neighbour work

This will come in handy for neighbour replay.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agomlxsw: spectrum_router: Allow address handlers to run on bridge ports
Petr Machata [Wed, 19 Jul 2023 11:01:19 +0000 (13:01 +0200)]
mlxsw: spectrum_router: Allow address handlers to run on bridge ports

Currently the IP address event handlers bail out when the event is related
to a netdevice that is a bridge port or a member of a LAG. In order to
create a RIF when a bridged or LAG'd port is unenslaved, these event
handlers will be replayed. However, at the point in time when the
NETDEV_CHANGEUPPER event is delivered, informing of the loss of
enslavement, the port is still formally enslaved.

In order for the operation to have any effect, these handlers need an extra
parameter to indicate that the check for bridge or LAG membership should
not be done. In this patch, add an argument "nomaster" to several event
handlers.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoselftests: mlxsw: rtnetlink: Drop obsolete tests
Petr Machata [Wed, 19 Jul 2023 11:01:18 +0000 (13:01 +0200)]
selftests: mlxsw: rtnetlink: Drop obsolete tests

Support for enslaving ports to LAGs with uppers will be added in the
following patches. Selftests to make sure it actually does the right thing
are ready and will be sent as a follow-up.

Similarly, ordering of MACVLAN creation and RIF creation will be relaxed
and it will be permitted to create a MACVLAN first.

Thus these two tests are obsolete. Drop them.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: switchdev: Add a helper to replay objects on a bridge port
Petr Machata [Wed, 19 Jul 2023 11:01:17 +0000 (13:01 +0200)]
net: switchdev: Add a helper to replay objects on a bridge port

When a front panel joins a bridge via another netdevice (typically a LAG),
the driver needs to learn about the objects configured on the bridge port.
When the bridge port is offloaded by the driver for the first time, this
can be achieved by passing a notifier to switchdev_bridge_port_offload().
The notifier is then invoked for the individual objects (such as VLANs)
configured on the bridge, and can look for the interesting ones.

Calling switchdev_bridge_port_offload() when the second port joins the
bridge lower is unnecessary, but the replay is still needed. To that end,
add a new function, switchdev_bridge_port_replay(), which does only the
replay part of the _offload() function in exactly the same way as that
function.

Cc: Jiri Pirko <jiri@resnulli.us>
Cc: Ivan Vecera <ivecera@redhat.com>
Cc: Roopa Prabhu <roopa@nvidia.com>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: bridge@lists.linux-foundation.org
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: bridge: br_switchdev: Tolerate -EOPNOTSUPP when replaying MDB
Petr Machata [Wed, 19 Jul 2023 11:01:16 +0000 (13:01 +0200)]
net: bridge: br_switchdev: Tolerate -EOPNOTSUPP when replaying MDB

There are two kinds of MDB entries to be replayed: port MDB entries, and
host MDB entries. They are both replayed by br_switchdev_mdb_replay(). If
the driver supports one kind, but lacks the other, the first -EOPNOTSUPP
returned terminates the whole replay, including any further still-supported
objects in the list.

For this to cause issues, there must be MDB entries for both the host and
the port being replayed. In that case, if the driver bails out from
handling the host entry, the port entries are never replayed. However, the
replay is currently only done when a switchdev port joins a bridge. There
would be no port memberships at that point. Thus despite being erroneous,
the code does not cause observable bugs.

This is not an issue with other object kinds either, because there, each
function replays one object kind. If a driver does not support that kind,
it makes sense to bail out early. -EOPNOTSUPP is then ignored in
nbp_switchdev_sync_objs().

For MDB, suppress the -EOPNOTSUPP error code in br_switchdev_mdb_replay()
already, so that the whole list gets replayed.

The reason we need this patch is that a future patch will introduce a
replay that should be used when a front-panel port netdevice is enslaved to
a bridge lower, in particular a LAG. The LAG netdevice can already have
both host and port MDB entries. The port entries need to be replayed so
that they are offloaded on the port that joins the LAG.

Cc: Jiri Pirko <jiri@resnulli.us>
Cc: Ivan Vecera <ivecera@redhat.com>
Cc: Roopa Prabhu <roopa@nvidia.com>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: bridge@lists.linux-foundation.org
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agonet: ethernet: mtk_ppe: add MTK_FOE_ENTRY_V{1,2}_SIZE macros
Lorenzo Bianconi [Wed, 19 Jul 2023 10:29:49 +0000 (12:29 +0200)]
net: ethernet: mtk_ppe: add MTK_FOE_ENTRY_V{1,2}_SIZE macros

Introduce MTK_FOE_ENTRY_V{1,2}_SIZE macros in order to make more
explicit foe_entry size for different chipset revisions.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 months agoMerge branch 'nexthop-refactor-and-fix-nexthop-selection-for-multipath-routes'
Jakub Kicinski [Fri, 21 Jul 2023 03:23:23 +0000 (20:23 -0700)]
Merge branch 'nexthop-refactor-and-fix-nexthop-selection-for-multipath-routes'

Benjamin Poirier says:

====================
nexthop: Refactor and fix nexthop selection for multipath routes

In order to select a nexthop for multipath routes, fib_select_multipath()
is used with legacy nexthops and nexthop_select_path_hthr() is used with
nexthop objects. Those two functions perform a validity test on the
neighbor related to each nexthop but their logic is structured differently.
This causes a divergence in behavior and nexthop_select_path_hthr() may
return a nexthop that failed the neighbor validity test even if there was
one that passed.

Refactor nexthop_select_path_hthr() to make it more similar to
fib_select_multipath() and fix the problem mentioned above.

v1: https://lore.kernel.org/netdev/20230529201914.69828-1-bpoirier@nvidia.com/
====================

Link: https://lore.kernel.org/r/20230719-nh_select-v2-0-04383e89f868@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
15 months agoselftests: net: Add test cases for nexthop groups with invalid neighbors
Benjamin Poirier [Wed, 19 Jul 2023 13:57:27 +0000 (13:57 +0000)]
selftests: net: Add test cases for nexthop groups with invalid neighbors

Add test cases for hash threshold (multipath) nexthop groups with invalid
neighbors. Check that a nexthop with invalid neighbor is not selected when
there is another nexthop with a valid neighbor. Check that there is no
crash when there is no nexthop with a valid neighbor.

The first test fails before the previous commit in this series.

Signed-off-by: Benjamin Poirier <bpoirier@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230719-nh_select-v2-4-04383e89f868@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>