platform/kernel/linux-rpi.git
17 months agoselftests/bpf: Fix s390 sock_field test failure
Yonghong Song [Tue, 16 May 2023 21:49:45 +0000 (14:49 -0700)]
selftests/bpf: Fix s390 sock_field test failure

llvm patch [1] enabled cross-function optimization for func arguments
(ArgumentPromotion) at -O2 level. And this caused s390 sock_fields
test failure ([2]). The failure is gone right now as patch [1] was
reverted in [3]. But it is possible that patch [3] will be reverted
again and then the test failure in [2] will show up again. So it is
desirable to fix the failure regardless.

The following is an analysis why sock_field test fails with
llvm patch [1].

The main problem is in
  static __noinline bool sk_dst_port__load_word(struct bpf_sock *sk)
  {
        __u32 *word = (__u32 *)&sk->dst_port;
        return word[0] == bpf_htons(0xcafe);
  }
  static __noinline bool sk_dst_port__load_half(struct bpf_sock *sk)
  {
        __u16 *half = (__u16 *)&sk->dst_port;
        return half[0] == bpf_htons(0xcafe);
  }
  ...
  int read_sk_dst_port(struct __sk_buff *skb)
  {
...
        sk = skb->sk;
...
        if (!sk_dst_port__load_word(sk))
                RET_LOG();
        if (!sk_dst_port__load_half(sk))
                RET_LOG();
...
  }

Through some cross-function optimization by ArgumentPromotion
optimization, the compiler does:
  static __noinline bool sk_dst_port__load_word(__u32 word_val)
  {
        return word_val == bpf_htons(0xcafe);
  }
  static __noinline bool sk_dst_port__load_half(__u16 half_val)
  {
        return half_val == bpf_htons(0xcafe);
  }
  ...
  int read_sk_dst_port(struct __sk_buff *skb)
  {
        ...
        sk = skb->sk;
        ...
        __u32 *word = (__u32 *)&sk->dst_port;
        __u32 word_val = word[0];
        ...
        if (!sk_dst_port__load_word(word_val))
                RET_LOG();

        __u16 half_val = word_val >> 16;
        if (!sk_dst_port__load_half(half_val))
                RET_LOG();
        ...
  }

In current uapi bpf.h, we have
  struct bpf_sock {
...
        __be16 dst_port;        /* network byte order */
        __u16 :16;              /* zero padding */
...
  };
But the old kernel (e.g., 5.6) we have
  struct bpf_sock {
...
__u32 dst_port;         /* network byte order */
...
  };

So for backward compatability reason, 4-byte load of
dst_port is converted to 2-byte load internally.
Specifically, 'word_val = word[0]' is replaced by 2-byte load
by the verifier and this caused the trouble for later
sk_dst_port__load_half() where half_val becomes 0.

Typical usr program won't have such a code pattern tiggering
the above bug, so let us fix the test failure with source
code change. Adding an empty asm volatile statement seems
enough to prevent undesired transformation.

  [1] https://reviews.llvm.org/D148269
  [2] https://lore.kernel.org/bpf/e7f2c5e8-a50c-198d-8f95-388165f1e4fd@meta.com/
  [3] https://reviews.llvm.org/rG141be5c062ecf22bd287afffd310e8ac4711444a

Tested-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20230516214945.1013578-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agoselftests/bpf: improve netcnt test robustness
Andrii Nakryiko [Mon, 15 May 2023 20:48:33 +0000 (13:48 -0700)]
selftests/bpf: improve netcnt test robustness

Change netcnt to demand at least 10K packets, as we frequently see some
stray packet arriving during the test in BPF CI. It seems more important
to make sure we haven't lost any packet than enforcing exact number of
packets.

Cc: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230515204833.2832000-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agoRevert "net: Remove low_thresh in ip defrag"
Jakub Kicinski [Wed, 17 May 2023 03:41:12 +0000 (20:41 -0700)]
Revert "net: Remove low_thresh in ip defrag"

This reverts commit b2cbac9b9b28730e9e53be20b6cdf979d3b9f27e.

We have multiple reports of obvious breakage from this patch.

Reported-by: Ido Schimmel <idosch@idosch.org>
Link: https://lore.kernel.org/all/ZGIRWjNcfqI8yY8W@shredder/
Link: https://lore.kernel.org/all/CADJHv_sDK=0RrMA2FTZQV5fw7UQ+qY=HG21Wu5qb0V9vvx5w6A@mail.gmail.com/
Reported-by: syzbot+a5e719ac7c268e414c95@syzkaller.appspotmail.com
Reported-by: syzbot+a03fd670838d927d9cd8@syzkaller.appspotmail.com
Fixes: b2cbac9b9b28 ("net: Remove low_thresh in ip defrag")
Link: https://lore.kernel.org/r/20230517034112.1261835-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf...
Jakub Kicinski [Wed, 17 May 2023 02:50:05 +0000 (19:50 -0700)]
Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2023-05-16

We've added 57 non-merge commits during the last 19 day(s) which contain
a total of 63 files changed, 3293 insertions(+), 690 deletions(-).

The main changes are:

1) Add precision propagation to verifier for subprogs and callbacks,
   from Andrii Nakryiko.

2) Improve BPF's {g,s}setsockopt() handling with wrong option lengths,
   from Stanislav Fomichev.

3) Utilize pahole v1.25 for the kernel's BTF generation to filter out
   inconsistent function prototypes, from Alan Maguire.

4) Various dyn-pointer verifier improvements to relax restrictions,
   from Daniel Rosenberg.

5) Add a new bpf_task_under_cgroup() kfunc for designated task,
   from Feng Zhou.

6) Unblock tests for arm64 BPF CI after ftrace supporting direct call,
   from Florent Revest.

7) Add XDP hint kfunc metadata for RX hash/timestamp for igc,
   from Jesper Dangaard Brouer.

8) Add several new dyn-pointer kfuncs to ease their usability,
   from Joanne Koong.

9) Add in-depth LRU internals description and dot function graph,
   from Joe Stringer.

10) Fix KCSAN report on bpf_lru_list when accessing node->ref,
    from Martin KaFai Lau.

11) Only dump unprivileged_bpf_disabled log warning upon write,
    from Kui-Feng Lee.

12) Extend test_progs to directly passing allow/denylist file,
    from Stephen Veiss.

13) Fix BPF trampoline memleak upon failure attaching to fentry,
    from Yafang Shao.

14) Fix emitting struct bpf_tcp_sock type in vmlinux BTF,
    from Yonghong Song.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (57 commits)
  bpf: Fix memleak due to fentry attach failure
  bpf: Remove bpf trampoline selector
  bpf, arm64: Support struct arguments in the BPF trampoline
  bpftool: JIT limited misreported as negative value on aarch64
  bpf: fix calculation of subseq_idx during precision backtracking
  bpf: Remove anonymous union in bpf_kfunc_call_arg_meta
  bpf: Document EFAULT changes for sockopt
  selftests/bpf: Correctly handle optlen > 4096
  selftests/bpf: Update EFAULT {g,s}etsockopt selftests
  bpf: Don't EFAULT for {g,s}setsockopt with wrong optlen
  libbpf: fix offsetof() and container_of() to work with CO-RE
  bpf: Address KCSAN report on bpf_lru_list
  bpf: Add --skip_encoding_btf_inconsistent_proto, --btf_gen_optimized to pahole flags for v1.25
  selftests/bpf: Accept mem from dynptr in helper funcs
  bpf: verifier: Accept dynptr mem as mem in helpers
  selftests/bpf: Check overflow in optional buffer
  selftests/bpf: Test allowing NULL buffer in dynptr slice
  bpf: Allow NULL buffers in bpf_dynptr_slice(_rw)
  selftests/bpf: Add testcase for bpf_task_under_cgroup
  bpf: Add bpf_task_under_cgroup() kfunc
  ...
====================

Link: https://lore.kernel.org/r/20230515225603.27027-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'spdx-conversion-for-bonding-8390-and-i825xx-drivers'
Paolo Abeni [Tue, 16 May 2023 13:39:09 +0000 (15:39 +0200)]
Merge branch 'spdx-conversion-for-bonding-8390-and-i825xx-drivers'

Bagas Sanjaya says:

====================
SPDX conversion for bonding, 8390, and i825xx drivers

This series is SPDX conversion for bonding, 8390, and i825xx driver
subsystems. It is splitted from v2 of my SPDX conversion series in
response to Didi's GPL full name fixes [1] to make it easily
digestible.

The conversion in this series is divided by each subsystem and by
license type.

[1]: https://lore.kernel.org/linux-spdx/20230512100620.36807-1-bagasdotme@gmail.com/
====================

Link: https://lore.kernel.org/r/20230515060714.621952-1-bagasdotme@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: ethernet: i825xx: sun3_8256: Add SPDX license identifier
Bagas Sanjaya [Mon, 15 May 2023 06:07:15 +0000 (13:07 +0700)]
net: ethernet: i825xx: sun3_8256: Add SPDX license identifier

The boilerplate reads that sun3_8256 driver is an extension to Linux
kernel core, hence add SPDX license identifier for GPL 2.0.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Michael Hipp <hippm@informatik.uni-tuebingen.de>
Cc: Sam Creasey <sammy@sammy.net>
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: ethernet: i825xx: Replace unversioned GPL (GPL 1.0) notice with SPDX identifier
Bagas Sanjaya [Mon, 15 May 2023 06:07:14 +0000 (13:07 +0700)]
net: ethernet: i825xx: Replace unversioned GPL (GPL 1.0) notice with SPDX identifier

Replace unversioned GPL boilerplate notice with corresponding SPDX
license identifier, which is GPL 1.0+.

Cc: Donald Becker <becker@scyld.com>
Cc: Richard Hirst <richard@sleepie.demon.co.uk>
Cc: Sam Creasey <sammy@sammy.net>
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: ethernet: 8390: Replace GPL 2.0 boilerplate with SPDX identifier
Bagas Sanjaya [Mon, 15 May 2023 06:07:13 +0000 (13:07 +0700)]
net: ethernet: 8390: Replace GPL 2.0 boilerplate with SPDX identifier

The boilerplate refers to COPYING in the top-level directory of kernel
tree. Replace it with corresponding SPDX license identifier.

Cc: Donald Becker <becker@scyld.com>
Cc: Peter De Schrijver <p2@mind.be>
Cc: Topi Kanerva <topi@susanna.oulu.fi>
Cc: Alain Malek <Alain.Malek@cryogen.com>
Cc: Bruce Abbott <bhabbott@inhb.co.nz>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Richard Fontana <rfontana@redhat.com>
Acked-by: Greg Ungerer <gerg@linux-m68k.org>
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: ethernet: 8390: Convert unversioned GPL notice to SPDX license identifier
Bagas Sanjaya [Mon, 15 May 2023 06:07:12 +0000 (13:07 +0700)]
net: ethernet: 8390: Convert unversioned GPL notice to SPDX license identifier

Replace boilerplate notice for unversioned GPL to SPDX tag for GPL 1.0+.
For ne2k-pci.c, only add SPDX tag and keep the boilerplate instead,
since the boilerplate notes that it must be preserved.

Cc: David A. Hinds <dahinds@users.sourceforge.net>
Cc: Donald Becker <becker@scyld.com>
Cc: Alan Cox <alan@linux.intel.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Richard Fontana <rfontana@redhat.com>
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: bonding: Add SPDX identifier to remaining files
Bagas Sanjaya [Mon, 15 May 2023 06:07:11 +0000 (13:07 +0700)]
net: bonding: Add SPDX identifier to remaining files

Previous batches of SPDX conversion missed bond_main.c and bonding_priv.h
because these files doesn't mention intended GPL version. Add SPDX identifier
to these files, assuming GPL 1.0+.

Cc: Thomas Davis <tadavis@lbl.gov>
Cc: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: skbuff: update comment about pfmemalloc propagating
Yunsheng Lin [Mon, 15 May 2023 05:01:07 +0000 (13:01 +0800)]
net: skbuff: update comment about pfmemalloc propagating

__skb_fill_page_desc_noacc() is not doing any pfmemalloc
propagating, and yet it has a comment about that, commit
84ce071e38a6 ("net: introduce __skb_fill_page_desc_noacc")
may have accidentally moved it to __skb_fill_page_desc_noacc(),
so move it back to __skb_fill_page_desc() which is supposed
to be doing pfmemalloc propagating.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
CC: Pavel Begunkov <asml.silence@gmail.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/20230515050107.46397-1-linyunsheng@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agobpf: Fix memleak due to fentry attach failure
Yafang Shao [Mon, 15 May 2023 13:08:47 +0000 (13:08 +0000)]
bpf: Fix memleak due to fentry attach failure

If it fails to attach fentry, the allocated bpf trampoline image will be
left in the system. That can be verified by checking /proc/kallsyms.

This meamleak can be verified by a simple bpf program as follows:

  SEC("fentry/trap_init")
  int fentry_run()
  {
      return 0;
  }

It will fail to attach trap_init because this function is freed after
kernel init, and then we can find the trampoline image is left in the
system by checking /proc/kallsyms.

  $ tail /proc/kallsyms
  ffffffffc0613000 t bpf_trampoline_6442453466_1  [bpf]
  ffffffffc06c3000 t bpf_trampoline_6442453466_1  [bpf]

  $ bpftool btf dump file /sys/kernel/btf/vmlinux | grep "FUNC 'trap_init'"
  [2522] FUNC 'trap_init' type_id=119 linkage=static

  $ echo $((6442453466 & 0x7fffffff))
  2522

Note that there are two left bpf trampoline images, that is because the
libbpf will fallback to raw tracepoint if -EINVAL is returned.

Fixes: e21aa341785c ("bpf: Fix fexit trampoline.")
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <song@kernel.org>
Cc: Jiri Olsa <olsajiri@gmail.com>
Link: https://lore.kernel.org/bpf/20230515130849.57502-2-laoar.shao@gmail.com
17 months agobpf: Remove bpf trampoline selector
Yafang Shao [Mon, 15 May 2023 13:08:48 +0000 (13:08 +0000)]
bpf: Remove bpf trampoline selector

After commit e21aa341785c ("bpf: Fix fexit trampoline."), the selector is only
used to indicate how many times the bpf trampoline image are updated and been
displayed in the trampoline ksym name. After the trampoline is freed, the
selector will start from 0 again. So the selector is a useless value to the
user. We can remove it.

If the user want to check whether the bpf trampoline image has been updated
or not, the user can compare the address. Each time the trampoline image is
updated, the address will change consequently. Jiri also pointed out another
issue that perf is still using the old name "bpf_trampoline_%lu", so this
change can fix the issue in perf.

Fixes: e21aa341785c ("bpf: Fix fexit trampoline.")
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <song@kernel.org>
Cc: Jiri Olsa <olsajiri@gmail.com>
Link: https://lore.kernel.org/bpf/ZFvOOlrmHiY9AgXE@krava
Link: https://lore.kernel.org/bpf/20230515130849.57502-3-laoar.shao@gmail.com
17 months agobpf, arm64: Support struct arguments in the BPF trampoline
Florent Revest [Thu, 11 May 2023 14:05:07 +0000 (16:05 +0200)]
bpf, arm64: Support struct arguments in the BPF trampoline

This extends the BPF trampoline JIT to support attachment to functions
that take small structures (up to 128bit) as argument. This is trivially
achieved by saving/restoring a number of "argument registers" rather
than a number of arguments.

The AAPCS64 section 6.8.2 describes the parameter passing ABI.
"Composite types" (like C structs) below 16 bytes (as enforced by the
BPF verifier) are provided as part of the 8 argument registers as
explained in the section C.12.

Signed-off-by: Florent Revest <revest@chromium.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Acked-by: Xu Kuohai <xukuohai@huawei.com>
Link: https://lore.kernel.org/bpf/20230511140507.514888-1-revest@chromium.org
17 months agobpftool: JIT limited misreported as negative value on aarch64
Alan Maguire [Fri, 12 May 2023 11:31:34 +0000 (12:31 +0100)]
bpftool: JIT limited misreported as negative value on aarch64

On aarch64, "bpftool feature" reports an incorrect BPF JIT limit:

$ sudo /sbin/bpftool feature
Scanning system configuration...
bpf() syscall restricted to privileged users
JIT compiler is enabled
JIT compiler hardening is disabled
JIT compiler kallsyms exports are enabled for root
skipping kernel config, can't open file: No such file or directory
Global memory limit for JIT compiler for unprivileged users is -201326592 bytes

This is because /proc/sys/net/core/bpf_jit_limit reports

$ sudo cat /proc/sys/net/core/bpf_jit_limit
68169519595520

...and an int is assumed in read_procfs().  Change read_procfs()
to return a long to avoid negative value reporting.

Fixes: 7a4522bbef0c ("tools: bpftool: add probes for /proc/ eBPF parameters")
Reported-by: Nicky Veitch <nicky.veitch@oracle.com>
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20230512113134.58996-1-alan.maguire@oracle.com
17 months agobpf: fix calculation of subseq_idx during precision backtracking
Andrii Nakryiko [Mon, 15 May 2023 18:07:10 +0000 (11:07 -0700)]
bpf: fix calculation of subseq_idx during precision backtracking

Subsequent instruction index (subseq_idx) is an index of an instruction
that was verified/executed by verifier after the currently processed
instruction. It is maintained during precision backtracking processing
and is used to detect various subprog calling conditions.

This patch fixes the bug with incorrectly resetting subseq_idx to -1
when going from child state to parent state during backtracking. If we
don't maintain correct subseq_idx we can misidentify subprog calls
leading to precision tracking bugs.

One such case was triggered by test_global_funcs/global_func9 test where
global subprog call happened to be the very last instruction in parent
state, leading to subseq_idx==-1, triggering WARN_ONCE:

  [   36.045754] verifier backtracking bug
  [   36.045764] WARNING: CPU: 13 PID: 2073 at kernel/bpf/verifier.c:3503 __mark_chain_precision+0xcc6/0xde0
  [   36.046819] Modules linked in: aesni_intel(E) crypto_simd(E) cryptd(E) kvm_intel(E) kvm(E) irqbypass(E) i2c_piix4(E) serio_raw(E) i2c_core(E) crc32c_intel)
  [   36.048040] CPU: 13 PID: 2073 Comm: test_progs Tainted: G        W  OE      6.3.0-07976-g4d585f48ee6b-dirty #972
  [   36.048783] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014
  [   36.049648] RIP: 0010:__mark_chain_precision+0xcc6/0xde0
  [   36.050038] Code: 3d 82 c6 05 bb 35 32 02 01 e8 66 21 ec ff 0f 0b b8 f2 ff ff ff e9 30 f5 ff ff 48 c7 c7 f3 61 3d 82 4c 89 0c 24 e8 4a 21 ec ff <0f> 0b 4c0

With the fix precision tracking across multiple states works correctly now:

mark_precise: frame0: last_idx 45 first_idx 38 subseq_idx -1
mark_precise: frame0: regs=r8 stack= before 44: (61) r7 = *(u32 *)(r10 -4)
mark_precise: frame0: regs=r8 stack= before 43: (85) call pc+41
mark_precise: frame0: regs=r8 stack= before 42: (07) r1 += -48
mark_precise: frame0: regs=r8 stack= before 41: (bf) r1 = r10
mark_precise: frame0: regs=r8 stack= before 40: (63) *(u32 *)(r10 -48) = r1
mark_precise: frame0: regs=r8 stack= before 39: (b4) w1 = 0
mark_precise: frame0: regs=r8 stack= before 38: (85) call pc+38
mark_precise: frame0: parent state regs=r8 stack=:  R0_w=scalar() R1_w=map_value(off=4,ks=4,vs=8,imm=0) R6=1 R7_w=scalar() R8_r=P0 R10=fpm
mark_precise: frame0: last_idx 36 first_idx 28 subseq_idx 38
mark_precise: frame0: regs=r8 stack= before 36: (18) r1 = 0xffff888104f2ed14
mark_precise: frame0: regs=r8 stack= before 35: (85) call pc+33
mark_precise: frame0: regs=r8 stack= before 33: (18) r1 = 0xffff888104f2ed10
mark_precise: frame0: regs=r8 stack= before 32: (85) call pc+36
mark_precise: frame0: regs=r8 stack= before 31: (07) r1 += -4
mark_precise: frame0: regs=r8 stack= before 30: (bf) r1 = r10
mark_precise: frame0: regs=r8 stack= before 29: (63) *(u32 *)(r10 -4) = r7
mark_precise: frame0: regs=r8 stack= before 28: (4c) w7 |= w0
mark_precise: frame0: parent state regs=r8 stack=:  R0_rw=scalar() R6=1 R7_rw=scalar() R8_rw=P0 R10=fp0 fp-48_r=mmmmmmmm
mark_precise: frame0: last_idx 27 first_idx 16 subseq_idx 28
mark_precise: frame0: regs=r8 stack= before 27: (85) call pc+31
mark_precise: frame0: regs=r8 stack= before 26: (b7) r1 = 0
mark_precise: frame0: regs=r8 stack= before 25: (b7) r8 = 0

Note how subseq_idx starts out as -1, then is preserved as 38 and then 28 as we
go up the parent state chain.

Reported-by: Alexei Starovoitov <ast@kernel.org>
Fixes: fde2a3882bd0 ("bpf: support precision propagation in the presence of subprogs")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230515180710.1535018-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agobpf: Remove anonymous union in bpf_kfunc_call_arg_meta
Dave Marchevsky [Wed, 10 May 2023 21:30:47 +0000 (14:30 -0700)]
bpf: Remove anonymous union in bpf_kfunc_call_arg_meta

For kfuncs like bpf_obj_drop and bpf_refcount_acquire - which take
user-defined types as input - the verifier needs to track the specific
type passed in when checking a particular kfunc call. This requires
tracking (btf, btf_id) tuple. In commit 7c50b1cb76ac
("bpf: Add bpf_refcount_acquire kfunc") I added an anonymous union with
inner structs named after the specific kfuncs tracking this information,
with the goal of making it more obvious which kfunc this data was being
tracked / expected to be tracked on behalf of.

In a recent series adding a new user of this tuple, Alexei mentioned
that he didn't like this union usage as it doesn't really help with
readability or bug-proofing ([0]). In an offline convo we agreed to
have the tuple be fields (arg_btf, arg_btf_id), with comments in
bpf_kfunc_call_arg_meta definition enumerating the uses of the fields by
kfunc-specific handling logic. Such a pattern is used by struct
bpf_reg_state without trouble.

Accordingly, this patch removes the anonymous union in favor of arg_btf
and arg_btf_id fields and comment enumerating their current uses. The
patch also removes struct btf_and_id, which was only being used by the
removed union's inner structs.

This is a mechanical change, existing linked_list and rbtree tests will
validate that correct (btf, btf_id) are being passed.

  [0]: https://lore.kernel.org/bpf/20230505021707.vlyiwy57vwxglbka@dhcp-172-26-102-232.dhcp.thefacebook.com

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230510213047.1633612-1-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agonfc: llcp: fix possible use of uninitialized variable in nfc_llcp_send_connect()
Krzysztof Kozlowski [Sat, 13 May 2023 11:52:04 +0000 (13:52 +0200)]
nfc: llcp: fix possible use of uninitialized variable in nfc_llcp_send_connect()

If sock->service_name is NULL, the local variable
service_name_tlv_length will not be assigned by nfc_llcp_build_tlv(),
later leading to using value frmo the stack.  Smatch warning:

  net/nfc/llcp_commands.c:442 nfc_llcp_send_connect() error: uninitialized symbol 'service_name_tlv_length'.

Fixes: de9e5aeb4f40 ("NFC: llcp: Fix usage of llcp_add_tlv()")
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: mcs: Remove unneeded semicolon
Yang Li [Mon, 15 May 2023 08:56:45 +0000 (16:56 +0800)]
octeontx2-pf: mcs: Remove unneeded semicolon

./drivers/net/ethernet/marvell/octeontx2/nic/cn10k_macsec.c:242:2-3: Unneeded semicolon
./drivers/net/ethernet/marvell/octeontx2/nic/cn10k_macsec.c:476:2-3: Unneeded semicolon

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=4947
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: ethernet: microchip: vcap: Remove extra semicolon
Anup Sharma [Sat, 13 May 2023 11:06:54 +0000 (16:36 +0530)]
net: ethernet: microchip: vcap: Remove extra semicolon

Remove the extra semicolon at end. Issue identified using
semicolon.cocci Coccinelle semantic patch.

drivers/net/ethernet/microchip/vcap/vcap_api.c:1124:3-4: Unneeded semicolon
drivers/net/ethernet/microchip/vcap/vcap_api.c:1165:3-4: Unneeded semicolon
drivers/net/ethernet/microchip/vcap/vcap_api.c:1239:3-4: Unneeded semicolon
drivers/net/ethernet/microchip/vcap/vcap_api.c:1287:3-4: Unneeded semicolon

Signed-off-by: Anup Sharma <anupnewsmail@gmail.com>
Changes:
V1 -> V2: Target tree included in the subject line.

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'octeontx2-pf-HTB'
David S. Miller [Mon, 15 May 2023 08:31:08 +0000 (09:31 +0100)]
Merge branch 'octeontx2-pf-HTB'

Hariprasad Kelam says:

====================
octeontx2-pf: HTB offload support

octeontx2 silicon and CN10K transmit interface consists of five
transmit levels starting from MDQ, TL4 to TL1. Once packets are
submitted to MDQ, hardware picks all active MDQs using strict
priority, and MDQs having the same priority level are chosen using
round robin. Each packet will traverse MDQ, TL4 to TL1 levels.
Each level contains an array of queues to support scheduling and
shaping.

As HTB supports classful queuing mechanism by supporting rate and
ceil and allow the user to control the absolute bandwidth to
particular classes of traffic the same can be achieved by
configuring shapers and schedulers on different transmit levels.

This series of patches adds support for HTB offload,

Patch1: Allow strict priority parameter in HTB offload mode.

Patch2: Rename existing total tx queues for better readability

Patch3: defines APIs such that the driver can dynamically initialize/
        deinitialize the send queues.

Patch4: Refactors transmit alloc/free calls as preparation for QOS
        offload code.

Patch5: moves rate limiting logic to common header which will be used
        by qos offload code.

Patch6: Adds actual HTB offload support.

Patch7: exposes qos send queue stats over ethtool.

Patch8: Add documentation about htb offload flow in driver
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agodocs: octeontx2: Add Documentation for QOS
Hariprasad Kelam [Sat, 13 May 2023 08:51:43 +0000 (14:21 +0530)]
docs: octeontx2: Add Documentation for QOS

Add QOS example configuration along with tc-htb commands

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: ethtool expose qos stats
Hariprasad Kelam [Sat, 13 May 2023 08:51:42 +0000 (14:21 +0530)]
octeontx2-pf: ethtool expose qos stats

This patch extends ethtool stats support for QoS send queues as well.
upon the number of transmit channels change request, Ensures the real
number of transmit queues are equal to active QoS send queues plus
configured transmit queues.

    ethtool -S eth0
    txq_qos0: bytes: 3021391800
    txq_qos0: frames: 1998275
    txq_qos1: bytes: 4619766312
    txq_qos1: frames: 3055401
    ...
    ...

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: Add support for HTB offload
Naveen Mamindlapalli [Sat, 13 May 2023 08:51:41 +0000 (14:21 +0530)]
octeontx2-pf: Add support for HTB offload

This patch registers callbacks to support HTB offload.

Below are features supported,

- supports traffic shaping on the given class by honoring rate and ceil
configuration.

- supports traffic scheduling,  which prioritizes different types of
traffic based on strict priority values.

- supports the creation of leaf to inner classes such that parent node
rate limits apply to all child nodes.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: Prepare for QOS offload
Hariprasad Kelam [Sat, 13 May 2023 08:51:40 +0000 (14:21 +0530)]
octeontx2-pf: Prepare for QOS offload

This patch moves rate limiting definitions to a common header file and
adds csr definitions required for QOS code.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: Refactor schedular queue alloc/free calls
Hariprasad Kelam [Sat, 13 May 2023 08:51:39 +0000 (14:21 +0530)]
octeontx2-pf: Refactor schedular queue alloc/free calls

1. Upon txschq free request, the transmit schedular config in hardware
is not getting reset. This patch adds necessary changes to do the same.

2. Current implementation calls txschq alloc during interface
initialization and in response handler updates the default txschq array.
This creates a problem for htb offload where txsch alloc will be called
for every tc class. This patch addresses the issue by reading txschq
response in mbox caller function instead in the response handler.

3. Current otx2_txschq_stop routine tries to free all txschq nodes
allocated to the interface. This creates a problem for htb offload.
This patch introduces the otx2_txschq_free_one to free txschq in a
given level.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: qos send queues management
Subbaraya Sundeep [Sat, 13 May 2023 08:51:38 +0000 (14:21 +0530)]
octeontx2-pf: qos send queues management

Current implementation is such that the number of Send queues (SQs)
are decided on the device probe which is equal to the number of online
cpus. These SQs are allocated and deallocated in interface open and c
lose calls respectively.

This patch defines new APIs for initializing and deinitializing Send
queues dynamically and allocates more number of transmit queues for
QOS feature.

Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: Rename tot_tx_queues to non_qos_queues
Hariprasad Kelam [Sat, 13 May 2023 08:51:37 +0000 (14:21 +0530)]
octeontx2-pf: Rename tot_tx_queues to non_qos_queues

current implementation is such that tot_tx_queues contains both
xdp queues and normal tx queues. which will be allocated in interface
open calls and deallocated on interface down calls respectively.

With addition of QOS, where send quees are allocated/deallacated upon
user request Qos send queues won't be part of tot_tx_queues. So this
patch renames tot_tx_queues to non_qos_queues.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosch_htb: Allow HTB priority parameter in offload mode
Naveen Mamindlapalli [Sat, 13 May 2023 08:51:36 +0000 (14:21 +0530)]
sch_htb: Allow HTB priority parameter in offload mode

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

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

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Co-developed-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: Remove low_thresh in ip defrag
Angus Chen [Fri, 12 May 2023 01:01:52 +0000 (09:01 +0800)]
net: Remove low_thresh in ip defrag

As low_thresh has no work in fragment reassembles,del it.
And Mark it deprecated in sysctl Document.

Signed-off-by: Angus Chen <angus.chen@jaguarmicro.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge tag 'wireless-next-2023-05-12' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Mon, 15 May 2023 07:37:17 +0000 (08:37 +0100)]
Merge tag 'wireless-next-2023-05-12' of git://git./linux/kernel/git/wireless/wireless-next

Kalle valo says:

====================
wireless-next patches for v6.5

The first pull request for v6.5 and only driver changes this time.
rtl8xxxu has been making lots of progress lately and now has AP mode
support.

Major changes:

rtl8xxxu

* AP mode support, initially only for rtl8188f

rtw89

* provide RSSI, EVN and SNR statistics via debugfs

* support U-NII-4 channels on 5 GHz band
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'bpf: Don't EFAULT for {g,s}setsockopt with wrong optlen'
Martin KaFai Lau [Sat, 13 May 2023 23:20:16 +0000 (16:20 -0700)]
Merge branch 'bpf: Don't EFAULT for {g,s}setsockopt with wrong optlen'

Stanislav Fomichev says:

====================
optval larger than PAGE_SIZE leads to EFAULT if the BPF program
isn't careful enough. This is often overlooked and might break
completely unrelated socket options. Instead of EFAULT,
let's ignore BPF program buffer changes. See the first patch for
more info.

In addition, clearly document this corner case and reset optlen
in our selftests (in case somebody copy-pastes from them).

v6:
- no changes; resending due to screwing up v5 series with the unrelated
  patch

v5:
- goto in the selftest (Martin)
- set IP_TOS to zero to avoid endianness complications (Martin)

v4:
- ignore retval as well when optlen > PAGE_SIZE (Martin)

v3:
- don't hard-code PAGE_SIZE (Martin)
- reset orig_optlen in getsockopt when kernel part succeeds (Martin)
====================

Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
17 months agobpf: Document EFAULT changes for sockopt
Stanislav Fomichev [Thu, 11 May 2023 17:04:56 +0000 (10:04 -0700)]
bpf: Document EFAULT changes for sockopt

And add examples for how to correctly handle large optlens.
This is less relevant now when we don't EFAULT anymore, but
that's still the correct thing to do.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230511170456.1759459-5-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
17 months agoselftests/bpf: Correctly handle optlen > 4096
Stanislav Fomichev [Thu, 11 May 2023 17:04:55 +0000 (10:04 -0700)]
selftests/bpf: Correctly handle optlen > 4096

Even though it's not relevant in selftests, the people
might still copy-paste from them. So let's take care
of optlen > 4096 cases explicitly.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230511170456.1759459-4-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
17 months agoselftests/bpf: Update EFAULT {g,s}etsockopt selftests
Stanislav Fomichev [Thu, 11 May 2023 17:04:54 +0000 (10:04 -0700)]
selftests/bpf: Update EFAULT {g,s}etsockopt selftests

Instead of assuming EFAULT, let's assume the BPF program's
output is ignored.

Remove "getsockopt: deny arbitrary ctx->retval" because it
was actually testing optlen. We have separate set of tests
for retval.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230511170456.1759459-3-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
17 months agobpf: Don't EFAULT for {g,s}setsockopt with wrong optlen
Stanislav Fomichev [Thu, 11 May 2023 17:04:53 +0000 (10:04 -0700)]
bpf: Don't EFAULT for {g,s}setsockopt with wrong optlen

With the way the hooks implemented right now, we have a special
condition: optval larger than PAGE_SIZE will expose only first 4k into
BPF; any modifications to the optval are ignored. If the BPF program
doesn't handle this condition by resetting optlen to 0,
the userspace will get EFAULT.

The intention of the EFAULT was to make it apparent to the
developers that the program is doing something wrong.
However, this inadvertently might affect production workloads
with the BPF programs that are not too careful (i.e., returning EFAULT
for perfectly valid setsockopt/getsockopt calls).

Let's try to minimize the chance of BPF program screwing up userspace
by ignoring the output of those BPF programs (instead of returning
EFAULT to the userspace). pr_info_once those cases to
the dmesg to help with figuring out what's going wrong.

Fixes: 0d01da6afc54 ("bpf: implement getsockopt and setsockopt hooks")
Suggested-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20230511170456.1759459-2-sdf@google.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
17 months agosfc: fix use-after-free in efx_tc_flower_record_encap_match()
Edward Cree [Fri, 12 May 2023 15:35:58 +0000 (16:35 +0100)]
sfc: fix use-after-free in efx_tc_flower_record_encap_match()

When writing error messages to extack for pseudo collisions, we can't
 use encap->type as encap has already been freed.  Fortunately the
 same value is stored in local variable em_type, so use that instead.

Fixes: 3c9561c0a5b9 ("sfc: support TC decap rules matching on enc_ip_tos")
Reported-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: phylink: constify fwnode arguments
Russell King (Oracle) [Fri, 12 May 2023 16:58:37 +0000 (17:58 +0100)]
net: phylink: constify fwnode arguments

Both phylink_create() and phylink_fwnode_phy_connect() do not modify
the fwnode argument that they are passed, so lets constify these.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: fec: using the standard return codes when xdp xmit errors
Shenwei Wang [Fri, 12 May 2023 13:20:10 +0000 (08:20 -0500)]
net: fec: using the standard return codes when xdp xmit errors

This patch standardizes the inconsistent return values for unsuccessful
XDP transmits by using standardized error codes (-EBUSY or -ENOMEM).

Signed-off-by: Shenwei Wang <shenwei.wang@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: macb: Shorten max_tx_len to 4KiB - 56 on mpfs
Daire McNamara [Fri, 12 May 2023 12:20:32 +0000 (13:20 +0100)]
net: macb: Shorten max_tx_len to 4KiB - 56 on mpfs

On mpfs, with SRAM configured for 4 queues, setting max_tx_len
to GEM_TX_MAX_LEN=0x3f0 results multiple AMBA errors.
Setting max_tx_len to (4KiB - 56) removes those errors.

The details are described in erratum 1686 by Cadence

The max jumbo frame size is also reduced for mpfs to (4KiB - 56).

Signed-off-by: Daire McNamara <daire.mcnamara@microchip.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoping: Convert hlist_nulls to plain hlist.
Kuniyuki Iwashima [Wed, 10 May 2023 21:54:43 +0000 (14:54 -0700)]
ping: Convert hlist_nulls to plain hlist.

Since introduced in commit c319b4d76b9e ("net: ipv4: add IPPROTO_ICMP
socket kind"), ping socket does not use SLAB_TYPESAFE_BY_RCU nor check
nulls marker in loops.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'skb_frag_fill_page_desc'
David S. Miller [Sat, 13 May 2023 18:47:56 +0000 (19:47 +0100)]
Merge branch 'skb_frag_fill_page_desc'

Yunsheng Lin says:

====================
net: introduce skb_frag_fill_page_desc()

Most users use __skb_frag_set_page()/skb_frag_off_set()/
skb_frag_size_set() to fill the page desc for a skb frag.
It does not make much sense to calling __skb_frag_set_page()
without calling skb_frag_off_set(), as the offset may depend
on whether the page is head page or tail page, so add
skb_frag_fill_page_desc() to fill the page desc for a skb
frag.

In the future, we can make sure the page in the frag is
head page of compound page or a base page, if not, we
may warn about that and convert the tail page to head
page and update the offset accordingly, if we see a warning
about that, we also fix the caller to fill the head page
in the frag. when the fixing is done, we may remove the
warning and converting.

In this way, we can remove the compound_head() or use
page_ref_*() like the below case:
https://elixir.bootlin.com/linux/latest/source/net/core/page_pool.c#L881
https://elixir.bootlin.com/linux/latest/source/include/linux/skbuff.h#L3383

It may also convert net stack to use the folio easier.

V1: repost with all the ack/review tags included.
RFC: remove a local variable as pointed out by Simon.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: remove __skb_frag_set_page()
Yunsheng Lin [Thu, 11 May 2023 01:12:13 +0000 (09:12 +0800)]
net: remove __skb_frag_set_page()

The remaining users calling __skb_frag_set_page() with
page being NULL seems to be doing defensive programming,
as shinfo->nr_frags is already decremented, so remove
them.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: introduce and use skb_frag_fill_page_desc()
Yunsheng Lin [Thu, 11 May 2023 01:12:12 +0000 (09:12 +0800)]
net: introduce and use skb_frag_fill_page_desc()

Most users use __skb_frag_set_page()/skb_frag_off_set()/
skb_frag_size_set() to fill the page desc for a skb frag.

Introduce skb_frag_fill_page_desc() to do that.

net/bpf/test_run.c does not call skb_frag_off_set() to
set the offset, "copy_from_user(page_address(page), ...)"
and 'shinfo' being part of the 'data' kzalloced in
bpf_test_init() suggest that it is assuming offset to be
initialized as zero, so call skb_frag_fill_page_desc()
with offset being zero for this case.

Also, skb_frag_set_page() is not used anymore, so remove
it.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoselftests: net: vxlan: Add tests for vxlan nolocalbypass option.
Vladimir Nikishkin [Fri, 12 May 2023 03:40:34 +0000 (11:40 +0800)]
selftests: net: vxlan: Add tests for vxlan nolocalbypass option.

Add test to make sure that the localbypass option is on by default.

Add test to change vxlan localbypass to nolocalbypass and check
that packets are delivered to userspace.

Signed-off-by: Vladimir Nikishkin <vladimir@nikishkin.pw>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: vxlan: Add nolocalbypass option to vxlan.
Vladimir Nikishkin [Fri, 12 May 2023 03:40:33 +0000 (11:40 +0800)]
net: vxlan: Add nolocalbypass option to vxlan.

If a packet needs to be encapsulated towards a local destination IP, the
packet will undergo a "local bypass" and be injected into the Rx path as
if it was received by the target VXLAN device without undergoing
encapsulation. If such a device does not exist, the packet will be
dropped.

There are scenarios where we do not want to perform such a bypass, but
instead want the packet to be encapsulated and locally received by a
user space program for post-processing.

To that end, add a new VXLAN device attribute that controls whether a
"local bypass" is performed or not. Default to performing a bypass to
maintain existing behavior.

Signed-off-by: Vladimir Nikishkin <vladimir@nikishkin.pw>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'broadcom-phy-wol'
David S. Miller [Sat, 13 May 2023 15:56:29 +0000 (16:56 +0100)]
Merge branch 'broadcom-phy-wol'

Florian Fainelli says:

====================
Support for Wake-on-LAN for Broadcom PHYs

This patch series adds support for Wake-on-LAN to the Broadcom PHY
driver. Specifically the BCM54210E/B50212E are capable of supporting
Wake-on-LAN using an external pin typically wired up to a system's GPIO.

These PHY operate a programmable Ethernet MAC destination address
comparator which will fire up an interrupt whenever a match is received.
Because of that, it was necessary to introduce patch #1 which allows the
PHY driver's ->suspend() routine to be called unconditionally. This is
necessary in our case because we need a hook point into the device
suspend/resume flow to enable the wake-up interrupt as late as possible.

Patch #2 adds support for the Broadcom PHY library and driver for
Wake-on-LAN proper with the WAKE_UCAST, WAKE_MCAST, WAKE_BCAST,
WAKE_MAGIC and WAKE_MAGICSECURE. Note that WAKE_FILTER is supportable,
however this will require further discussions and be submitted as a RFC
series later on.

Patch #3 updates the GENET driver to defer to the PHY for Wake-on-LAN if
the PHY supports it, thus allowing the MAC to be powered down to
conserve power.

Changes in v3:

- collected Reviewed-by tags
- explicitly use return 0 in bcm54xx_phy_probe() (Paolo)

Changes in v2:

- introduce PHY_ALWAYS_CALL_SUSPEND and only have the Broadcom PHY
  driver set this flag to minimize changes to the suspend flow to only
  drivers that need it

- corrected possibly uninitialized variable in bcm54xx_set_wakeup_irq
  (Simon)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: bcmgenet: Add support for PHY-based Wake-on-LAN
Florian Fainelli [Thu, 11 May 2023 17:21:10 +0000 (10:21 -0700)]
net: bcmgenet: Add support for PHY-based Wake-on-LAN

If available, interrogate the PHY to find out whether we can use it for
Wake-on-LAN. This can be a more power efficient way of implementing
that feature, especially when the MAC is powered off in low power
states.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: phy: broadcom: Add support for Wake-on-LAN
Florian Fainelli [Thu, 11 May 2023 17:21:09 +0000 (10:21 -0700)]
net: phy: broadcom: Add support for Wake-on-LAN

Add support for WAKE_UCAST, WAKE_MCAST, WAKE_BCAST, WAKE_MAGIC and
WAKE_MAGICSECURE. This is only supported with the BCM54210E and
compatible Ethernet PHYs. Using the in-band interrupt or an out of band
GPIO interrupts are supported.

Broadcom PHYs will generate a Wake-on-LAN level low interrupt on LED4 as
soon as one of the supported patterns is being matched. That includes
generating such an interrupt even if the PHY is operated during normal
modes. If WAKE_UCAST is selected, this could lead to the LED4 interrupt
firing up for every packet being received which is absolutely
undesirable from a performance point of view.

Because the Wake-on-LAN configuration can be set long before the system
is actually put to sleep, we cannot have an interrupt service routine to
clear on read the interrupt status register and ensure that new packet
matches will be detected.

It is desirable to enable the Wake-on-LAN interrupt as late as possible
during the system suspend process such that we limit the number of
interrupts to be handled by the system, but also conversely feed into
the Linux's system suspend way of dealing with interrupts in and around
the points of no return.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: phy: Allow drivers to always call into ->suspend()
Florian Fainelli [Thu, 11 May 2023 17:21:08 +0000 (10:21 -0700)]
net: phy: Allow drivers to always call into ->suspend()

A few PHY drivers are currently attempting to not suspend the PHY when
Wake-on-LAN is enabled, however that code is not currently executing at
all due to an early check in phy_suspend().

This prevents PHY drivers from making an appropriate decisions and put
the hardware into a low power state if desired.

In order to allow the PHY drivers to opt into getting their ->suspend
routine to be called, add a PHY_ALWAYS_CALL_SUSPEND bit which can be
set. A boolean that tracks whether the PHY or the attached MAC has
Wake-on-LAN enabled is also provided for convenience.

If phydev::wol_enabled then the PHY shall not prevent its own
Wake-on-LAN detection logic from working and shall not prevent the
Ethernet MAC from receiving packets for matching.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agolibbpf: fix offsetof() and container_of() to work with CO-RE
Andrii Nakryiko [Tue, 9 May 2023 06:55:02 +0000 (23:55 -0700)]
libbpf: fix offsetof() and container_of() to work with CO-RE

It seems like __builtin_offset() doesn't preserve CO-RE field
relocations properly. So if offsetof() macro is defined through
__builtin_offset(), CO-RE-enabled BPF code using container_of() will be
subtly and silently broken.

To avoid this problem, redefine offsetof() and container_of() in the
form that works with CO-RE relocations more reliably.

Fixes: 5fbc220862fc ("tools/libpf: Add offsetof/container_of macro in bpf_helpers.h")
Reported-by: Lennart Poettering <lennart@poettering.net>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20230509065502.2306180-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agobpf: Address KCSAN report on bpf_lru_list
Martin KaFai Lau [Thu, 11 May 2023 04:37:48 +0000 (21:37 -0700)]
bpf: Address KCSAN report on bpf_lru_list

KCSAN reported a data-race when accessing node->ref.
Although node->ref does not have to be accurate,
take this chance to use a more common READ_ONCE() and WRITE_ONCE()
pattern instead of data_race().

There is an existing bpf_lru_node_is_ref() and bpf_lru_node_set_ref().
This patch also adds bpf_lru_node_clear_ref() to do the
WRITE_ONCE(node->ref, 0) also.

==================================================================
BUG: KCSAN: data-race in __bpf_lru_list_rotate / __htab_lru_percpu_map_update_elem

write to 0xffff888137038deb of 1 bytes by task 11240 on cpu 1:
__bpf_lru_node_move kernel/bpf/bpf_lru_list.c:113 [inline]
__bpf_lru_list_rotate_active kernel/bpf/bpf_lru_list.c:149 [inline]
__bpf_lru_list_rotate+0x1bf/0x750 kernel/bpf/bpf_lru_list.c:240
bpf_lru_list_pop_free_to_local kernel/bpf/bpf_lru_list.c:329 [inline]
bpf_common_lru_pop_free kernel/bpf/bpf_lru_list.c:447 [inline]
bpf_lru_pop_free+0x638/0xe20 kernel/bpf/bpf_lru_list.c:499
prealloc_lru_pop kernel/bpf/hashtab.c:290 [inline]
__htab_lru_percpu_map_update_elem+0xe7/0x820 kernel/bpf/hashtab.c:1316
bpf_percpu_hash_update+0x5e/0x90 kernel/bpf/hashtab.c:2313
bpf_map_update_value+0x2a9/0x370 kernel/bpf/syscall.c:200
generic_map_update_batch+0x3ae/0x4f0 kernel/bpf/syscall.c:1687
bpf_map_do_batch+0x2d9/0x3d0 kernel/bpf/syscall.c:4534
__sys_bpf+0x338/0x810
__do_sys_bpf kernel/bpf/syscall.c:5096 [inline]
__se_sys_bpf kernel/bpf/syscall.c:5094 [inline]
__x64_sys_bpf+0x43/0x50 kernel/bpf/syscall.c:5094
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd

read to 0xffff888137038deb of 1 bytes by task 11241 on cpu 0:
bpf_lru_node_set_ref kernel/bpf/bpf_lru_list.h:70 [inline]
__htab_lru_percpu_map_update_elem+0x2f1/0x820 kernel/bpf/hashtab.c:1332
bpf_percpu_hash_update+0x5e/0x90 kernel/bpf/hashtab.c:2313
bpf_map_update_value+0x2a9/0x370 kernel/bpf/syscall.c:200
generic_map_update_batch+0x3ae/0x4f0 kernel/bpf/syscall.c:1687
bpf_map_do_batch+0x2d9/0x3d0 kernel/bpf/syscall.c:4534
__sys_bpf+0x338/0x810
__do_sys_bpf kernel/bpf/syscall.c:5096 [inline]
__se_sys_bpf kernel/bpf/syscall.c:5094 [inline]
__x64_sys_bpf+0x43/0x50 kernel/bpf/syscall.c:5094
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd

value changed: 0x01 -> 0x00

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 11241 Comm: syz-executor.3 Not tainted 6.3.0-rc7-syzkaller-00136-g6a66fdd29ea1 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/30/2023
==================================================================

Reported-by: syzbot+ebe648a84e8784763f82@syzkaller.appspotmail.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20230511043748.1384166-1-martin.lau@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agobpf: Add --skip_encoding_btf_inconsistent_proto, --btf_gen_optimized to pahole flags...
Alan Maguire [Wed, 10 May 2023 13:02:41 +0000 (14:02 +0100)]
bpf: Add --skip_encoding_btf_inconsistent_proto, --btf_gen_optimized to pahole flags for v1.25

v1.25 of pahole supports filtering out functions with multiple inconsistent
function prototypes or optimized-out parameters from the BTF representation.
These present problems because there is no additional info in BTF saying which
inconsistent prototype matches which function instance to help guide attachment,
and functions with optimized-out parameters can lead to incorrect assumptions
about register contents.

So for now, filter out such functions while adding BTF representations for
functions that have "."-suffixes (foo.isra.0) but not optimized-out parameters.
This patch assumes that below linked changes land in pahole for v1.25.

Issues with pahole filtering being too aggressive in removing functions
appear to be resolved now, but CI and further testing will confirm.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20230510130241.1696561-1-alan.maguire@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
17 months agoMerge branch 'sfc-decap'
David S. Miller [Fri, 12 May 2023 09:37:02 +0000 (10:37 +0100)]
Merge branch 'sfc-decap'

Edward Cree says:

====================
sfc: more flexible encap matches on TC decap rules

This series extends the TC offload support on EF100 to support optionally
 matching on the IP ToS and UDP source port of the outer header in rules
 performing tunnel decapsulation.  Both of these fields allow masked
 matches if the underlying hardware supports it (current EF100 hardware
 supports masking on ToS, but only exact-match on source port).
Given that the source port is typically populated from a hash of inner
 header entropy, it's not clear whether filtering on it is useful, but
 since we can support it we may as well expose the capability.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosfc: support TC decap rules matching on enc_src_port
Edward Cree [Thu, 11 May 2023 19:47:31 +0000 (20:47 +0100)]
sfc: support TC decap rules matching on enc_src_port

Allow efx_tc_encap_match entries to include a udp_sport and a
 udp_sport_mask.  As with enc_ip_tos, use pseudos to enforce that all
 encap matches within a given <src_ip,dst_ip,udp_dport> tuple have
 the same udp_sport_mask.
Note that since we use a single layer of pseudos for both fields, two
 matches that differ in (say) udp_sport value aren't permitted to have
 different ip_tos_mask, even though this would technically be safe.
Current userland TC does not support setting enc_src_port; this patch
 was tested with an iproute2 patched to support it.

Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosfc: support TC decap rules matching on enc_ip_tos
Edward Cree [Thu, 11 May 2023 19:47:30 +0000 (20:47 +0100)]
sfc: support TC decap rules matching on enc_ip_tos

Allow efx_tc_encap_match entries to include an ip_tos and ip_tos_mask.
To avoid partially-overlapping Outer Rules (which can lead to undefined
 behaviour in the hardware), store extra "pseudo" entries in our
 encap_match hashtable, which are used to enforce that all Outer Rule
 entries within a given <src_ip,dst_ip,udp_dport> tuple (or IPv6
 equivalent) have the same ip_tos_mask.
The "direct" encap_match entry takes a reference on the "pseudo",
 allowing it to be destroyed when all "direct" entries using it are
 removed.
efx_tc_em_pseudo_type is an enum rather than just a bool because in
 future an additional pseudo-type will be added to support Conntrack
 offload.

Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosfc: populate enc_ip_tos matches in MAE outer rules
Edward Cree [Thu, 11 May 2023 19:47:29 +0000 (20:47 +0100)]
sfc: populate enc_ip_tos matches in MAE outer rules

Currently tc.c will block them before they get here, but following
 patch will change that.
Use the extack message from efx_mae_check_encap_match_caps() instead
 of writing a new one, since there's now more being fed in than just
 an IP version.

Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosfc: release encap match in efx_tc_flow_free()
Edward Cree [Thu, 11 May 2023 19:47:28 +0000 (20:47 +0100)]
sfc: release encap match in efx_tc_flow_free()

When force-freeing leftover entries from our match_action_ht, call
 efx_tc_delete_rule(), which releases all the rule's resources, rather
 than open-coding it.  The open-coded version was missing a call to
 release the rule's encap match (if any).
It probably doesn't matter as everything's being torn down anyway, but
 it's cleaner this way and prevents further error messages potentially
 being logged by efx_tc_encap_match_free() later on.
Move efx_tc_flow_free() further down the file to avoid introducing a
 forward declaration of efx_tc_delete_rule().

Fixes: 17654d84b47c ("sfc: add offloading of 'foreign' TC (decap) rules")
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: liquidio: lio_main: Remove unnecessary (void*) conversions
wuych [Fri, 12 May 2023 02:44:03 +0000 (10:44 +0800)]
net: liquidio: lio_main: Remove unnecessary (void*) conversions

Pointer variables of void * type do not require type cast.

Signed-off-by: wuych <yunchuan@nfschina.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agosctp: add bpf_bypass_getsockopt proto callback
Alexander Mikhalitsyn [Thu, 11 May 2023 13:25:06 +0000 (15:25 +0200)]
sctp: add bpf_bypass_getsockopt proto callback

Implement ->bpf_bypass_getsockopt proto callback and filter out
SCTP_SOCKOPT_PEELOFF, SCTP_SOCKOPT_PEELOFF_FLAGS and SCTP_SOCKOPT_CONNECTX3
socket options from running eBPF hook on them.

SCTP_SOCKOPT_PEELOFF and SCTP_SOCKOPT_PEELOFF_FLAGS options do fd_install(),
and if BPF_CGROUP_RUN_PROG_GETSOCKOPT hook returns an error after success of
the original handler sctp_getsockopt(...), userspace will receive an error
from getsockopt syscall and will be not aware that fd was successfully
installed into a fdtable.

As pointed by Marcelo Ricardo Leitner it seems reasonable to skip
bpf getsockopt hook for SCTP_SOCKOPT_CONNECTX3 sockopt too.
Because internaly, it triggers connect() and if error is masked
then userspace will be confused.

This patch was born as a result of discussion around a new SCM_PIDFD interface:
https://lore.kernel.org/all/20230413133355.350571-3-aleksandr.mikhalitsyn@canonical.com/

Fixes: 0d01da6afc54 ("bpf: implement getsockopt and setsockopt hooks")
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Cc: Xin Long <lucien.xin@gmail.com>
Cc: linux-sctp@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: netdev@vger.kernel.org
Suggested-by: Stanislav Fomichev <sdf@google.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
Acked-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'selftests-fcnal'
David S. Miller [Fri, 12 May 2023 08:43:56 +0000 (09:43 +0100)]
Merge branch 'selftests-fcnal'

Guillaume Nault says:

====================
selftests: fcnal: Test SO_DONTROUTE socket option.

The objective is to cover kernel paths that use the RTO_ONLINK flag
in .flowi4_tos. This way we'll be able to safely remove this flag in
the future by properly setting .flowi4_scope instead. With these
selftests in place, we can make sure this won't introduce regressions.

For more context, the final objective is to convert .flowi4_tos to
dscp_t, to ensure that ECN bits don't influence route and fib-rule
lookups (see commit a410a0cf9885 ("ipv6: Define dscp_t and stop taking
ECN bits into account in fib6-rules")).

These selftests only cover IPv4, as SO_DONTROUTE has no effect on IPv6
sockets.

v2:
  - Use two different nettest options for setting SO_DONTROUTE either
    on the server or on the client socket.

  - Use the above feature to run a single 'nettest -B' instance per
    test (instead of having two nettest processes for server and
    client).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoselftests: fcnal: Test SO_DONTROUTE on raw and ping sockets.
Guillaume Nault [Thu, 11 May 2023 14:39:46 +0000 (16:39 +0200)]
selftests: fcnal: Test SO_DONTROUTE on raw and ping sockets.

Use ping -r to test the kernel behaviour with raw and ping sockets
having the SO_DONTROUTE option.

Since ipv4_ping_novrf() is called with different values of
net.ipv4.ping_group_range, then it tests both raw and ping sockets
(ping uses ping sockets if its user ID belongs to ping_group_range
and raw sockets otherwise).

With both socket types, sending packets to a neighbour (on link) host,
should work. When the host is behind a router, sending should fail.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoselftests: fcnal: Test SO_DONTROUTE on UDP sockets.
Guillaume Nault [Thu, 11 May 2023 14:39:39 +0000 (16:39 +0200)]
selftests: fcnal: Test SO_DONTROUTE on UDP sockets.

Use nettest --client-dontroute to test the kernel behaviour with UDP
sockets having the SO_DONTROUTE option. Sending packets to a neighbour
(on link) host, should work. When the host is behind a router, sending
should fail.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoselftests: fcnal: Test SO_DONTROUTE on TCP sockets.
Guillaume Nault [Thu, 11 May 2023 14:39:32 +0000 (16:39 +0200)]
selftests: fcnal: Test SO_DONTROUTE on TCP sockets.

Use nettest --{client,server}-dontroute to test the kernel behaviour
with TCP sockets having the SO_DONTROUTE option. Sending packets to a
neighbour (on link) host, should work. When the host is behind a
router, sending should fail.

Client and server sockets are tested independently, so that we can
cover different TCP kernel paths.

SO_DONTROUTE also affects the syncookies path. So ipv4_tcp_dontroute()
is made to work with or without syncookies, to cover both paths.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoselftests: Add SO_DONTROUTE option to nettest.
Guillaume Nault [Thu, 11 May 2023 14:39:25 +0000 (16:39 +0200)]
selftests: Add SO_DONTROUTE option to nettest.

Add --client-dontroute and --server-dontroute options to nettest. They
allow to set the SO_DONTROUTE option to the client and server sockets
respectively. This will be used by the following patches to test
the SO_DONTROUTE kernel behaviour with TCP and UDP.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agobonding: Always assign be16 value to vlan_proto
Simon Horman [Thu, 11 May 2023 15:07:12 +0000 (17:07 +0200)]
bonding: Always assign be16 value to vlan_proto

The type of the vlan_proto field is __be16.
And most users of the field use it as such.

In the case of setting or testing the field for the special VLAN_N_VID
value, host byte order is used. Which seems incorrect.

It also seems somewhat odd to store a VLAN ID value in a field that is
otherwise used to store Ether types.

Address this issue by defining BOND_VLAN_PROTO_NONE, a big endian value.
0xffff was chosen somewhat arbitrarily. What is important is that it
doesn't overlap with any valid VLAN Ether types.

I don't believe the problems described above are a bug because
VLAN_N_VID in both little-endian and big-endian byte order does not
conflict with any supported VLAN Ether types in big-endian byte order.

Reported by sparse as:

 .../bond_main.c:2857:26: warning: restricted __be16 degrades to integer
 .../bond_main.c:2863:20: warning: restricted __be16 degrades to integer
 .../bond_main.c:2939:40: warning: incorrect type in assignment (different base types)
 .../bond_main.c:2939:40:    expected restricted __be16 [usertype] vlan_proto
 .../bond_main.c:2939:40:    got int

No functional changes intended.
Compile tested only.

Signed-off-by: Simon Horman <horms@kernel.org>
Acked-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoMerge branch 'net-handshake-fixes'
David S. Miller [Fri, 12 May 2023 08:24:08 +0000 (09:24 +0100)]
Merge branch 'net-handshake-fixes'

Chuck Lever says:

====================
Bug fixes for net/handshake

Please consider these for merge via net-next.

Paolo observed that there is a possible leak of sock->file. I
haven't looked into that yet, but it seems to be separate from
the fixes in this series, so no need to hold these up.

Changes since v2:
- Address Paolo comment regarding handshake_dup()

Changes since v1:
- Rework "Fix handshake_dup() ref counting"
- Unpin sock->file when a handshake is cancelled
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: Enable the SNI extension to work properly
Chuck Lever [Thu, 11 May 2023 15:49:50 +0000 (11:49 -0400)]
net/handshake: Enable the SNI extension to work properly

Enable the upper layer protocol to specify the SNI peername. This
avoids the need for tlshd to use a DNS lookup, which can return a
hostname that doesn't match the incoming certificate's SubjectName.

Fixes: 2fd5532044a8 ("net/handshake: Add a kernel API for requesting a TLSv1.3 handshake")
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: Unpin sock->file if a handshake is cancelled
Chuck Lever [Thu, 11 May 2023 15:49:17 +0000 (11:49 -0400)]
net/handshake: Unpin sock->file if a handshake is cancelled

If user space never calls DONE, sock->file's reference count remains
elevated. Enable sock->file to be freed eventually in this case.

Reported-by: Jakub Kacinski <kuba@kernel.org>
Fixes: 3b3009ea8abb ("net/handshake: Create a NETLINK service for handling handshake requests")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: handshake_genl_notify() shouldn't ignore @flags
Chuck Lever [Thu, 11 May 2023 15:48:45 +0000 (11:48 -0400)]
net/handshake: handshake_genl_notify() shouldn't ignore @flags

Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Fixes: 3b3009ea8abb ("net/handshake: Create a NETLINK service for handling handshake requests")
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: Fix uninitialized local variable
Chuck Lever [Thu, 11 May 2023 15:48:13 +0000 (11:48 -0400)]
net/handshake: Fix uninitialized local variable

trace_handshake_cmd_done_err() simply records the pointer in @req,
so initializing it to NULL is sufficient and safe.

Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Fixes: 3b3009ea8abb ("net/handshake: Create a NETLINK service for handling handshake requests")
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: Fix handshake_dup() ref counting
Chuck Lever [Thu, 11 May 2023 15:47:40 +0000 (11:47 -0400)]
net/handshake: Fix handshake_dup() ref counting

If get_unused_fd_flags() fails, we ended up calling fput(sock->file)
twice.

Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Fixes: 3b3009ea8abb ("net/handshake: Create a NETLINK service for handling handshake requests")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet/handshake: Remove unneeded check from handshake_dup()
Chuck Lever [Thu, 11 May 2023 15:47:09 +0000 (11:47 -0400)]
net/handshake: Remove unneeded check from handshake_dup()

handshake_req_submit() now verifies that the socket has a file.

Fixes: 3b3009ea8abb ("net/handshake: Create a NETLINK service for handling handshake requests")
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoipvlan: Remove NULL check before dev_{put, hold}
Yang Li [Thu, 11 May 2023 07:21:19 +0000 (15:21 +0800)]
ipvlan: Remove NULL check before dev_{put, hold}

The call netdev_{put, hold} of dev_{put, hold} will check NULL,
so there is no need to check before using dev_{put, hold},
remove it to silence the warning:

./drivers/net/ipvlan/ipvlan_core.c:559:3-11: WARNING: NULL check before dev_{put, hold} functions is not needed.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=4930
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agoocteontx2-pf: mcs: Offload extended packet number(XPN) feature
Subbaraya Sundeep [Thu, 11 May 2023 06:17:12 +0000 (11:47 +0530)]
octeontx2-pf: mcs: Offload extended packet number(XPN) feature

The macsec hardware block supports XPN cipher suites also.
Hence added changes to offload XPN feature. Changes include
configuring SecY policy to XPN cipher suite, Salt and SSCI values.
64 bit packet number is passed instead of 32 bit packet number.

Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: samsung: sxgbe: Make sxgbe_drv_remove() return void
Uwe Kleine-König [Wed, 10 May 2023 20:02:47 +0000 (22:02 +0200)]
net: samsung: sxgbe: Make sxgbe_drv_remove() return void

sxgbe_drv_remove() returned zero unconditionally, so it can be converted
to return void without losing anything. The upside is that it becomes
more obvious in its callers that there is no error to handle.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 months agonet: enc28j60: Use threaded interrupt instead of workqueue
Philipp Rosenberger [Tue, 9 May 2023 04:28:56 +0000 (06:28 +0200)]
net: enc28j60: Use threaded interrupt instead of workqueue

The Microchip ENC28J60 SPI Ethernet driver schedules a work item from
the interrupt handler because accesses to the SPI bus may sleep.

On PREEMPT_RT (which forces interrupt handling into threads) this
old-fashioned approach unnecessarily increases latency because an
interrupt results in first waking the interrupt thread, then scheduling
the work item.  So, a double indirection to handle an interrupt.

Avoid by converting the driver to modern threaded interrupt handling.

Signed-off-by: Philipp Rosenberger <p.rosenberger@kunbus.com>
Signed-off-by: Zhi Han <hanzhi09@gmail.com>
[lukas: rewrite commit message, linewrap request_threaded_irq() call]
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Piotr Raczynski <piotr.raczynski@intel.com>
Link: https://lore.kernel.org/r/342380d989ce26bc49f0e5d45fbb0416a5f7809f.1683606193.git.lukas@wunner.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 11 May 2023 16:06:26 +0000 (09:06 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Cross-merge networking fixes. No conflicts.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge tag 'net-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 11 May 2023 13:42:47 +0000 (08:42 -0500)]
Merge tag 'net-6.4-rc2' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from netfilter.

  Current release - regressions:

   - mtk_eth_soc: fix NULL pointer dereference

  Previous releases - regressions:

   - core:
      - skb_partial_csum_set() fix against transport header magic value
      - fix load-tearing on sk->sk_stamp in sock_recv_cmsgs().
      - annotate sk->sk_err write from do_recvmmsg()
      - add vlan_get_protocol_and_depth() helper

   - netlink: annotate accesses to nlk->cb_running

   - netfilter: always release netdev hooks from notifier

  Previous releases - always broken:

   - core: deal with most data-races in sk_wait_event()

   - netfilter: fix possible bug_on with enable_hooks=1

   - eth: bonding: fix send_peer_notif overflow

   - eth: xpcs: fix incorrect number of interfaces

   - eth: ipvlan: fix out-of-bounds caused by unclear skb->cb

   - eth: stmmac: Initialize MAC_ONEUS_TIC_COUNTER register"

* tag 'net-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (31 commits)
  af_unix: Fix data races around sk->sk_shutdown.
  af_unix: Fix a data race of sk->sk_receive_queue->qlen.
  net: datagram: fix data-races in datagram_poll()
  net: mscc: ocelot: fix stat counter register values
  ipvlan:Fix out-of-bounds caused by unclear skb->cb
  docs: networking: fix x25-iface.rst heading & index order
  gve: Remove the code of clearing PBA bit
  tcp: add annotations around sk->sk_shutdown accesses
  net: add vlan_get_protocol_and_depth() helper
  net: pcs: xpcs: fix incorrect number of interfaces
  net: deal with most data-races in sk_wait_event()
  net: annotate sk->sk_err write from do_recvmmsg()
  netlink: annotate accesses to nlk->cb_running
  kselftest: bonding: add num_grat_arp test
  selftests: forwarding: lib: add netns support for tc rule handle stats get
  Documentation: bonding: fix the doc of peer_notif_delay
  bonding: fix send_peer_notif overflow
  net: ethernet: mtk_eth_soc: fix NULL pointer dereference
  selftests: nft_flowtable.sh: check ingress/egress chain too
  selftests: nft_flowtable.sh: monitor result file sizes
  ...

17 months agoMerge tag 'media/v6.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Thu, 11 May 2023 13:35:52 +0000 (08:35 -0500)]
Merge tag 'media/v6.4-2' of git://git./linux/kernel/git/mchehab/linux-media

Pull media fixes from Mauro Carvalho Chehab:

 - fix some unused-variable warning in mtk-mdp3

 - ignore unused suspend operations in nxp

 - some driver fixes in rcar-vin

* tag 'media/v6.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
  media: platform: mtk-mdp3: work around unused-variable warning
  media: nxp: ignore unused suspend operations
  media: rcar-vin: Select correct interrupt mode for V4L2_FIELD_ALTERNATE
  media: rcar-vin: Fix NV12 size alignment
  media: rcar-vin: Gen3 can not scale NV12

17 months agowifi: rtw89: suppress the log for specific SER called CMDPSR_FRZTO
Chin-Yen Lee [Mon, 8 May 2023 08:43:35 +0000 (16:43 +0800)]
wifi: rtw89: suppress the log for specific SER called CMDPSR_FRZTO

For 8852CE, there is abnormal state called CMDPSR_FRZTO,
which occasionally happens in some platforms, and could be
found by firmware and fixed in current SER flow, so we add
suppress function to avoid verbose message for this resolved case.

Signed-off-by: Chin-Yen Lee <timlee@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508084335.42953-4-pkshih@realtek.com
17 months agowifi: rtw89: ser: L1 add pre-M0 and post-M0 states
Zong-Zhe Yang [Mon, 8 May 2023 08:43:34 +0000 (16:43 +0800)]
wifi: rtw89: ser: L1 add pre-M0 and post-M0 states

Newer FW re-design SER (syetem error recovery) L1 (level 1) flow.
New L1 flow will expect two extra states before original L1 flow.

* Before:
fw --- M1 --> driver
fw <-- M2 --- driver
fw --- M3 --> driver
fw <-- M4 --- driver
fw --- M5 --> driver

* After:
fw --- pre-M0  --> driver
fw <-- post-M0 --- driver
fw --- M1 --> driver
fw <-- M2 --- driver
fw --- M3 --> driver
fw <-- M4 --- driver
fw --- M5 --> driver

Then before M1, FW gets one more interval to deal with things that FW
should have handled well. To consider backward/forward compatibility,
FW and driver won't change flow from M1 to M5. (only except that halt
trigger control will change a little bit.) So, there will be two differnt
starting points of SER L1.
* old FW: SER L1 starts from M1
* new FW: SER L1 starts from pre-M0
Then, driver adds the new SER L1 entry and also keep the original one
instead of changing it.

Signed-off-by: Zong-Zhe Yang <kevin_yang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508084335.42953-3-pkshih@realtek.com
17 months agowifi: rtw89: pci: fix interrupt enable mask for HALT C2H of RTL8851B
Zong-Zhe Yang [Mon, 8 May 2023 08:43:33 +0000 (16:43 +0800)]
wifi: rtw89: pci: fix interrupt enable mask for HALT C2H of RTL8851B

RTL8851B keeps almost the same interrupt flow as RTL8852A and RTL8852B.
But, it uses a different bitmask for interrupt indicator of FW HALT C2H.
So, we make a chip judgement in pci when configuring interrupt mask.

Signed-off-by: Zong-Zhe Yang <kevin_yang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508084335.42953-2-pkshih@realtek.com
17 months agowifi: rtw89: support U-NII-4 channels on 5GHz band
Zong-Zhe Yang [Mon, 8 May 2023 08:12:11 +0000 (16:12 +0800)]
wifi: rtw89: support U-NII-4 channels on 5GHz band

U-NII-4 band, i.e 5.9GHz channels, can be supported by chip 8852C, 8852B
and 8851B. But, it is not supported by chip 8852A. Flag support_unii4 is
added in chip info and defined by chip accordingly to indicate that.
We reference this flag of runtime chip to decide whether to register
5.9GHz channels.

After that, we consider if U-NII-4 band is allowed by our regulatory
rule of U-NII-4. If chip::support_unii4 but not regd::allow_unii4,
we stll do not register 5.9GHz channels.

Signed-off-by: Zong-Zhe Yang <kevin_yang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508081211.38760-4-pkshih@realtek.com
17 months agowifi: rtw89: regd: judge UNII-4 according to BIOS and chip
Zong-Zhe Yang [Mon, 8 May 2023 08:12:10 +0000 (16:12 +0800)]
wifi: rtw89: regd: judge UNII-4 according to BIOS and chip

For realtek regulatory, there are following two kinds of configurations
to determine whether to allow UNII-4 band, i.e. 5.9GHz channels.

1. default setting according to whether chip support it or not
2. evaluate realtek ACPI DSM with RTW89_ACPI_DSM_FUNC_59G_EN (func. 6)

If (1) is false, we won't try (2) and just disallow UNII-4. Otherwise,
if (2) is not supported or returns a non-specific value, we follow the
default setting either. Besides, this commit aims to add decision logic
in rtw89 regulatory. Actually, driver doesn't register UNII-4 yet. That
will be handled by another commit.

Signed-off-by: Zong-Zhe Yang <kevin_yang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508081211.38760-3-pkshih@realtek.com
17 months agowifi: rtw89: introduce realtek ACPI DSM method
Zong-Zhe Yang [Mon, 8 May 2023 08:12:09 +0000 (16:12 +0800)]
wifi: rtw89: introduce realtek ACPI DSM method

Introduce realtek ACPI DSM method to get required BIOS
configurations. It will be used in the following commits.

And, enum rtw89_acpi_dsm_func is added for listing the functions
which are currently recognized.

Signed-off-by: Zong-Zhe Yang <kevin_yang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230508081211.38760-2-pkshih@realtek.com
17 months agowifi: mwifiex: Fix the size of a memory allocation in mwifiex_ret_802_11_scan()
Christophe JAILLET [Sat, 6 May 2023 13:53:15 +0000 (15:53 +0200)]
wifi: mwifiex: Fix the size of a memory allocation in mwifiex_ret_802_11_scan()

The type of "mwifiex_adapter->nd_info" is "struct cfg80211_wowlan_nd_info",
not "struct cfg80211_wowlan_nd_match".

Use struct_size() to ease the computation of the needed size.

The current code over-allocates some memory, so is safe.
But it wastes 32 bytes.

Fixes: 7d7f07d8c5d3 ("mwifiex: add wowlan net-detect support")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/7a6074fb056d2181e058a3cc6048d8155c20aec7.1683371982.git.christophe.jaillet@wanadoo.fr
17 months agowifi: rtw88: unlock on error path in rtw_ops_add_interface()
Dan Carpenter [Wed, 3 May 2023 15:09:55 +0000 (18:09 +0300)]
wifi: rtw88: unlock on error path in rtw_ops_add_interface()

Call mutex_unlock(&rtwdev->mutex); before returning on this error path.

Fixes: f0e741e4ddbc ("wifi: rtw88: add bitmap for dynamic port settings")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/ddd10a74-5982-4f65-8c59-c1cca558d239@kili.mountain
17 months agowifi: wilc1000: Increase ASSOC response buffer
Amisha Patel [Tue, 9 May 2023 17:29:08 +0000 (17:29 +0000)]
wifi: wilc1000: Increase ASSOC response buffer

In recent access points, information element is longer as they include
additional data which exceeds 256 bytes. To accommodate longer
association response, increase the ASSOC response buffer.

Signed-off-by: Amisha Patel <amisha.patel@microchip.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230509172811.4953-1-amisha.patel@microchip.com
17 months agowifi: wilc1000: fix for absent RSN capabilities WFA testcase
Amisha Patel [Fri, 21 Apr 2023 18:10:20 +0000 (18:10 +0000)]
wifi: wilc1000: fix for absent RSN capabilities WFA testcase

Mandatory WFA testcase
CT_Security_WPA2Personal_STA_RSNEBoundsVerification-AbsentRSNCap,
performs bounds verfication on Beacon and/or Probe response frames. It
failed and observed the reason to be absence of cipher suite and AKM
suite in RSN information. To fix this, enable the RSN flag before extracting RSN
capabilities.

Fixes: cd21d99e595e ("wifi: wilc1000: validate pairwise and authentication suite offsets")
Signed-off-by: Amisha Patel <amisha.patel@microchip.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230421181005.4865-1-amisha.patel@microchip.com
17 months agoMerge branch 'net-mvneta-reduce-size-of-tso-header-allocation'
Paolo Abeni [Thu, 11 May 2023 11:05:19 +0000 (13:05 +0200)]
Merge branch 'net-mvneta-reduce-size-of-tso-header-allocation'

Russell King says:

====================
net: mvneta: reduce size of TSO header allocation

With reference to
https://forum.turris.cz/t/random-kernel-exceptions-on-hbl-tos-7-0/18865/
https://github.com/openwrt/openwrt/pull/12375#issuecomment-1528842334

It appears that mvneta attempts an order-6 allocation for the TSO
header memory. While this succeeds early on in the system's life time,
trying order-6 allocations later can result in failure due to memory
fragmentation.

Firstly, the reason it's so large is that we take the number of
transmit descriptors, and allocate a TSO header buffer for each, and
each TSO header is 256 bytes. The driver uses a simple mechanism to
determine the address - it uses the transmit descriptor index as an
index into the TSO header memory.

(The first obvious question is: do there need to be this
many? Won't each TSO header always have at least one bit
of data to go with it? In other words, wouldn't the maximum
number of TSO headers that a ring could accept be the number
of ring entries divided by 2?)

There is no real need for this memory to be an order-6 allocation,
since nothing in hardware requires this buffer to be contiguous.

Therefore, this series splits this order-6 allocation up into 32
order-1 allocations (8k pages on 4k page platforms), each giving
32 TSO headers per page.

In order to do this, these patches:

1) fix a horrible transmit path error-cleanup bug - the existing
   code unmaps from the first descriptor that was allocated at
   interface bringup, not the first descriptor that the packet
   is using, resulting in the wrong descriptors being unmapped.

2) since xdp support was added, we now have buf->type which indicates
   what this transmit buffer contains. Use this to mark TSO header
   buffers.

3) get rid of IS_TSO_HEADER(), instead using buf->type to determine
   whether this transmit buffer needs to be DMA-unmapped.

4) move tso_build_hdr() into mvneta_tso_put_hdr() to keep all the
   TSO header building code together.

5) split the TSO header allocation into chunks of order-1 pages.

This has now been tested by the Turris folk and has been found to fix
the allocation error.
====================

Link: https://lore.kernel.org/r/ZFtuhJOC03qpASt2@shell.armlinux.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: mvneta: allocate TSO header DMA memory in chunks
Russell King (Oracle) [Wed, 10 May 2023 10:16:03 +0000 (11:16 +0100)]
net: mvneta: allocate TSO header DMA memory in chunks

Now that we no longer need to check whether the DMA address is within
the TSO header DMA memory range for the queue, we can allocate the TSO
header DMA memory in chunks rather than one contiguous order-6 chunk,
which can stress the kernel's memory subsystems to allocate.

Instead, use order-1 (8k) allocations, which will result in 32 order-1
pages containing 32 TSO headers.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: mvneta: move tso_build_hdr() into mvneta_tso_put_hdr()
Russell King (Oracle) [Wed, 10 May 2023 10:15:58 +0000 (11:15 +0100)]
net: mvneta: move tso_build_hdr() into mvneta_tso_put_hdr()

Move tso_build_hdr() into mvneta_tso_put_hdr() so that all the TSO
header building code is in one place.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: mvneta: use buf->type to determine whether to dma-unmap
Russell King (Oracle) [Wed, 10 May 2023 10:15:53 +0000 (11:15 +0100)]
net: mvneta: use buf->type to determine whether to dma-unmap

Now that we use a different buffer type for TSO headers, we can use
buf->type to determine whether the original buffer was DMA-mapped or
not. The rules are:

MVNETA_TYPE_XDP_TX - from a DMA pool, no unmap is required
MVNETA_TYPE_XDP_NDO - dma_map_single()'d
MVNETA_TYPE_SKB - normal skbuff, dma_map_single()'d
MVNETA_TYPE_TSO - from the TSO buffer area

This means we only need to call dma_unmap_single() on the XDP_NDO and
SKB types of buffer, and we no longer need the private IS_TSO_HEADER()
which relies on the TSO region being contiguously allocated.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: mvneta: mark mapped and tso buffers separately
Russell King (Oracle) [Wed, 10 May 2023 10:15:48 +0000 (11:15 +0100)]
net: mvneta: mark mapped and tso buffers separately

Mark dma-mapped skbs and TSO buffers separately, so we can use
buf->type to identify their differences.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: mvneta: fix transmit path dma-unmapping on error
Russell King (Oracle) [Wed, 10 May 2023 10:15:42 +0000 (11:15 +0100)]
net: mvneta: fix transmit path dma-unmapping on error

The transmit code assumes that the transmit descriptors that are used
begin with the first descriptor in the ring, but this may not be the
case. Fix this by providing a new function that dma-unmaps a range of
numbered descriptor entries, and use that to do the unmapping.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agotcp: make the first N SYN RTO backoffs linear
David Morley [Tue, 9 May 2023 18:05:58 +0000 (18:05 +0000)]
tcp: make the first N SYN RTO backoffs linear

Currently the SYN RTO schedule follows an exponential backoff
scheme, which can be unnecessarily conservative in cases where
there are link failures. In such cases, it's better to
aggressively try to retransmit packets, so it takes routers
less time to find a repath with a working link.

We chose a default value for this sysctl of 4, to follow
the macOS and IOS backoff scheme of 1,1,1,1,1,2,4,8, ...
MacOS and IOS have used this backoff schedule for over
a decade, since before this 2009 IETF presentation
discussed the behavior:
https://www.ietf.org/proceedings/75/slides/tcpm-1.pdf

This commit makes the SYN RTO schedule start with a number of
linear backoffs given by the following sysctl:
* tcp_syn_linear_timeouts

This changes the SYN RTO scheme to be: init_rto_val for
tcp_syn_linear_timeouts, exp backoff starting at init_rto_val

For example if init_rto_val = 1 and tcp_syn_linear_timeouts = 2, our
backoff scheme would be: 1, 1, 1, 2, 4, 8, 16, ...

Signed-off-by: David Morley <morleyd@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Tested-by: David Morley <morleyd@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230509180558.2541885-1-morleyd.kernel@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: wwan: iosm: clean up unused struct members
M Chetan Kumar [Tue, 9 May 2023 16:36:35 +0000 (22:06 +0530)]
net: wwan: iosm: clean up unused struct members

Below members are unused.
- td_tag member defined in struct ipc_pipe.
- adb_finish_timer & params defined in struct iosm_mux.

Remove it to avoid unexpected usage.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/92ee483d79dfc871ed7408da8fec60b395ff3a9c.1683649868.git.m.chetan.kumar@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: wwan: iosm: remove unused enum definition
M Chetan Kumar [Tue, 9 May 2023 16:36:22 +0000 (22:06 +0530)]
net: wwan: iosm: remove unused enum definition

ipc_time_unit enum is defined but not used.
Remove it to avoid unexpected usage.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/8295a6138f13c686590ee4021384ee992f717408.1683649868.git.m.chetan.kumar@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: wwan: iosm: remove unused macro definition
M Chetan Kumar [Tue, 9 May 2023 16:35:55 +0000 (22:05 +0530)]
net: wwan: iosm: remove unused macro definition

IOSM_IF_ID_PAYLOAD is defined but not used.
Remove it to avoid unexpected usage.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/0697e811cb7f10b4fd8f99e66bda1329efdd3d1d.1683649868.git.m.chetan.kumar@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>