platform/kernel/linux-rpi.git
20 months agoselftests/bpf: add pre bpf_prog_test_run_opts() callback for test_loader
Eduard Zingerman [Thu, 20 Apr 2023 23:23:16 +0000 (02:23 +0300)]
selftests/bpf: add pre bpf_prog_test_run_opts() callback for test_loader

When a test case is annotated with __retval tag the test_loader engine
would use libbpf's bpf_prog_test_run_opts() to do a test run of the
program and compare retvals.

This commit allows to perform arbitrary actions on bpf object right
before test loader invokes bpf_prog_test_run_opts(). This could be
used to setup some state for program execution, e.g. fill some maps.

Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20230420232317.2181776-4-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: fix __retval() being always ignored
Eduard Zingerman [Thu, 20 Apr 2023 23:23:15 +0000 (02:23 +0300)]
selftests/bpf: fix __retval() being always ignored

Florian Westphal found a bug in and suggested a fix for test_loader.c
processing of __retval tag. Because of this bug the function
test_loader.c:do_prog_test_run() never executed and all __retval test
tags were ignored.

If this bug is fixed a number of test cases from
progs/verifier_array_access.c fail with retval not matching the
expected value. This test was recently converted to use test_loader.c
and inline assembly in [1]. When doing the conversion I missed the
important detail of test_verifier.c operation: when it creates
fixup_map_array_ro, fixup_map_array_wo and fixup_map_array_small it
populates these maps with a dummy record.

Disabling the __retval checks for the affected verifier_array_access
in this commit to avoid false-postivies in any potential bisects.
The issue is addressed in the next patch.

I verified that the __retval tags are now respected by changing
expected return values for all tests annotated with __retval, and
checking that these tests started to fail.

[1] https://lore.kernel.org/bpf/20230325025524.144043-1-eddyz87@gmail.com/

Fixes: 19a8e06f5f91 ("selftests/bpf: Tests execution support for test_loader.c")
Reported-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/bpf/f4c4aee644425842ee6aa8edf1da68f0a8260e7c.camel@gmail.com/T/
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20230420232317.2181776-3-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: disable program test run for progs/refcounted_kptr.c
Eduard Zingerman [Thu, 20 Apr 2023 23:23:14 +0000 (02:23 +0300)]
selftests/bpf: disable program test run for progs/refcounted_kptr.c

Florian Westphal found a bug in test_loader.c processing of __retval
tag. Because of this bug the function test_loader.c:do_prog_test_run()
never executed and all __retval test tags were ignored. This hid an
issue with progs/refcounted_kptr.c tests.

When __retval tag bug is fixed and refcounted_kptr.c tests are run
kernel reports various issues and eventually hangs. Shortest reproducer
is the following command run a few times:

  $ for i in $(seq 1 4); do (./test_progs --allow=refcounted_kptr &); done

Commenting out __retval tags for these tests until this issue is resolved.

Reported-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/bpf/f4c4aee644425842ee6aa8edf1da68f0a8260e7c.camel@gmail.com/T/
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20230420232317.2181776-2-eddyz87@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpftool: Replace "__fallthrough" by a comment to address merge conflict
Quentin Monnet [Thu, 20 Apr 2023 00:33:33 +0000 (01:33 +0100)]
bpftool: Replace "__fallthrough" by a comment to address merge conflict

The recent support for inline annotations in control flow graphs
generated by bpftool introduced the usage of the "__fallthrough" macro
in a switch/case block in btf_dumper.c. This change went through the
bpf-next tree, but resulted in a merge conflict in linux-next, because
this macro has been renamed "fallthrough" (no underscores) in the
meantime.

To address the conflict, we temporarily switch to a simple comment
instead of a macro.

Related: commit f7a858bffcdd ("tools: Rename __fallthrough to fallthrough")

Fixes: 9fd496848b1c ("bpftool: Support inline annotations when dumping the CFG of a program")
Reported-by: Sven Schnelle <svens@linux.ibm.com>
Reported-by: Thomas Richter <tmricht@linux.ibm.com>
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/all/yt9dttxlwal7.fsf@linux.ibm.com/
Link: https://lore.kernel.org/bpf/20230412123636.2358949-1-tmricht@linux.ibm.com/
Link: https://lore.kernel.org/bpf/20230420003333.90901-1-quentin@isovalent.com
20 months agoMerge branch 'Access variable length array relaxed for integer type'
Alexei Starovoitov [Thu, 20 Apr 2023 04:29:39 +0000 (21:29 -0700)]
Merge branch 'Access variable length array relaxed for integer type'

Feng zhou says:

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

From: Feng Zhou <zhoufeng.zf@bytedance.com>

Add support for integer type of accessing variable length array.
Add a selftest to check it.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Add test to access integer type of variable array
Feng Zhou [Thu, 20 Apr 2023 03:27:35 +0000 (11:27 +0800)]
selftests/bpf: Add test to access integer type of variable array

Add prog test for accessing integer type of variable array in tracing
program.
In addition, hook load_balance function to access sd->span[0], only
to confirm whether the load is successful. Because there is no direct
way to trigger load_balance call.

Co-developed-by: Chengming Zhou <zhouchengming@bytedance.com>
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com>
Link: https://lore.kernel.org/r/20230420032735.27760-3-zhoufeng.zf@bytedance.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: support access variable length array of integer type
Feng Zhou [Thu, 20 Apr 2023 03:27:34 +0000 (11:27 +0800)]
bpf: support access variable length array of integer type

After this commit:
bpf: Support variable length array in tracing programs (9c5f8a1008a1)
Trace programs can access variable length array, but for structure
type. This patch adds support for integer type.

Example:
Hook load_balance
struct sched_domain {
...
unsigned long span[];
}

The access: sd->span[0].

Co-developed-by: Chengming Zhou <zhouchengming@bytedance.com>
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com>
Link: https://lore.kernel.org/r/20230420032735.27760-2-zhoufeng.zf@bytedance.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/xsk: Fix munmap for hugepage allocated umem
Magnus Karlsson [Tue, 18 Apr 2023 14:36:17 +0000 (16:36 +0200)]
selftests/xsk: Fix munmap for hugepage allocated umem

Fix the unmapping of hugepage allocated umems so that they are
properly unmapped. The new test referred to in the fixes label,
introduced a test that allocated a umem that is not a multiple of a 2M
hugepage size. This is fine for mmap() that rounds the size up the
nearest multiple of 2M. But munmap() requires the size to be a
multiple of the hugepage size in order for it to unmap the region. The
current behaviour of not properly unmapping the umem, was discovered
when further additions of tests that require hugepages (unaligned mode
tests only) started failing as the system was running out of
hugepages.

Fixes: c0801598e543 ("selftests: xsk: Add test UNALIGNED_INV_DESC_4K1_FRAME_SIZE")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230418143617.27762-1-magnus.karlsson@gmail.com
20 months agoMerge branch 'Provide bpf_for() and bpf_for_each() by libbpf'
Alexei Starovoitov [Tue, 18 Apr 2023 19:45:11 +0000 (12:45 -0700)]
Merge branch 'Provide bpf_for() and bpf_for_each() by libbpf'

Andrii Nakryiko says:

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

This patch set moves bpf_for(), bpf_for_each(), and bpf_repeat() macros from
selftests-internal bpf_misc.h header to libbpf-provided bpf_helpers.h header.
To do this in a way to allow users to feature-detect and guard such
bpf_for()/bpf_for_each() uses on old kernels we also extend libbpf to improve
unresolved kfunc calls handling and reporting. This lets us mark
bpf_iter_num_{new,next,destroy}() declarations as __weak, and thus not fail
program loading outright if such kfuncs are missing on the host kernel.

Patches #1 and #2 do some simple clean ups and logging improvements. Patch #3
adds kfunc call poisoning and log fixup logic and is the hear of this patch
set, effectively. Patch #4 adds selftest for this logic. Patches #4 and #5
move bpf_for()/bpf_for_each()/bpf_repeat() into bpf_helpers.h header and mark
kfuncs as __weak to allow users to feature-detect and guard their uses.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agolibbpf: mark bpf_iter_num_{new,next,destroy} as __weak
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:48 +0000 (17:21 -0700)]
libbpf: mark bpf_iter_num_{new,next,destroy} as __weak

Mark bpf_iter_num_{new,next,destroy}() kfuncs declared for
bpf_for()/bpf_repeat() macros as __weak to allow users to feature-detect
their presence and guard bpf_for()/bpf_repeat() loops accordingly for
backwards compatibility with old kernels.

Now that libbpf supports kfunc calls poisoning and better reporting of
unresolved (but called) kfuncs, declaring number iterator kfuncs in
bpf_helpers.h won't degrade user experience and won't cause unnecessary
kernel feature dependencies.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-7-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agolibbpf: move bpf_for(), bpf_for_each(), and bpf_repeat() into bpf_helpers.h
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:47 +0000 (17:21 -0700)]
libbpf: move bpf_for(), bpf_for_each(), and bpf_repeat() into bpf_helpers.h

To make it easier for bleeding-edge BPF applications, such as sched_ext,
to utilize open-coded iterators, move bpf_for(), bpf_for_each(), and
bpf_repeat() macros from selftests/bpf-internal bpf_misc.h helper, to
libbpf-provided bpf_helpers.h header.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-6-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: add missing __weak kfunc log fixup test
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:46 +0000 (17:21 -0700)]
selftests/bpf: add missing __weak kfunc log fixup test

Add test validating that libbpf correctly poisons and reports __weak
unresolved kfuncs in post-processed verifier log.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-5-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agolibbpf: improve handling of unresolved kfuncs
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:45 +0000 (17:21 -0700)]
libbpf: improve handling of unresolved kfuncs

Currently, libbpf leaves `call #0` instruction for __weak unresolved
kfuncs, which might lead to a confusing verifier log situations, where
invalid `call #0` will be treated as successfully validated.

We can do better. Libbpf already has an established mechanism of
poisoning instructions that failed some form of resolution (e.g., CO-RE
relocation and BPF map set to not be auto-created). Libbpf doesn't fail
them outright to allow users to guard them through other means, and as
long as BPF verifier can prove that such poisoned instructions cannot be
ever reached, this doesn't consistute an invalid BPF program. If user
didn't guard such code, libbpf will extract few pieces of information to
tie such poisoned instructions back to additional information about what
entitity wasn't resolved (e.g., BPF map name, or CO-RE relocation
information).

__weak unresolved kfuncs fit this model well, so this patch extends
libbpf with poisioning and log fixup logic for kfunc calls.

Note, this poisoning is done only for kfunc *calls*, not kfunc address
resolution (ldimm64 instructions). The former cannot be ever valid, if
reached, so it's safe to poison them. The latter is a valid mechanism to
check if __weak kfunc ksym was resolved, and do necessary guarding and
work arounds based on this result, supported in most recent kernels. As
such, libbpf keeps such ldimm64 instructions as loading zero, never
poisoning them.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agolibbpf: report vmlinux vs module name when dealing with ksyms
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:44 +0000 (17:21 -0700)]
libbpf: report vmlinux vs module name when dealing with ksyms

Currently libbpf always reports "kernel" as a source of ksym BTF type,
which is ambiguous given ksym's BTF can come from either vmlinux or
kernel module BTFs. Make this explicit and log module name, if used BTF
is from kernel module.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-3-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agolibbpf: misc internal libbpf clean ups around log fixup
Andrii Nakryiko [Tue, 18 Apr 2023 00:21:43 +0000 (17:21 -0700)]
libbpf: misc internal libbpf clean ups around log fixup

Normalize internal constants, field names, and comments related to log
fixup. Also add explicit `ext_idx` alias for relocation where relocation
is pointing to extern description for additional information.

No functional changes, just a clean up before subsequent additions.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230418002148.3255690-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Add a selftest for checking subreg equality
Yonghong Song [Mon, 17 Apr 2023 22:21:39 +0000 (15:21 -0700)]
selftests/bpf: Add a selftest for checking subreg equality

Add a selftest to ensure subreg equality if source register
upper 32bit is 0. Without previous patch, the test will
fail verification.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20230417222139.360607-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Improve verifier u32 scalar equality checking
Yonghong Song [Mon, 17 Apr 2023 22:21:34 +0000 (15:21 -0700)]
bpf: Improve verifier u32 scalar equality checking

In [1], I tried to remove bpf-specific codes to prevent certain
llvm optimizations, and add llvm TTI (target transform info) hooks
to prevent those optimizations. During this process, I found
if I enable llvm SimplifyCFG:shouldFoldTwoEntryPHINode
transformation, I will hit the following verification failure with selftests:

  ...
  8: (18) r1 = 0xffffc900001b2230       ; R1_w=map_value(off=560,ks=4,vs=564,imm=0)
  10: (61) r1 = *(u32 *)(r1 +0)         ; R1_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
  ; if (skb->tstamp == EGRESS_ENDHOST_MAGIC)
  11: (79) r2 = *(u64 *)(r6 +152)       ; R2_w=scalar() R6=ctx(off=0,imm=0)
  ; if (skb->tstamp == EGRESS_ENDHOST_MAGIC)
  12: (55) if r2 != 0xb9fbeef goto pc+10        ; R2_w=195018479
  13: (bc) w2 = w1                      ; R1_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
  ; if (test < __NR_TESTS)
  14: (a6) if w1 < 0x9 goto pc+1 16: R0=2 R1_w=scalar(umax=8,var_off=(0x0; 0xf)) R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R6=ctx(off=0,imm=0) R10=fp0
  ;
  16: (27) r2 *= 28                     ; R2_w=scalar(umax=120259084260,var_off=(0x0; 0x1ffffffffc),s32_max=2147483644,u32_max=-4)
  17: (18) r3 = 0xffffc900001b2118      ; R3_w=map_value(off=280,ks=4,vs=564,imm=0)
  19: (0f) r3 += r2                     ; R2_w=scalar(umax=120259084260,var_off=(0x0; 0x1ffffffffc),s32_max=2147483644,u32_max=-4) R3_w=map_value(off=280,ks=4,vs=564,umax=120259084260,var_off=(0x0; 0x1ffffffffc),s32_max=2147483644,u32_max=-4)
  20: (61) r2 = *(u32 *)(r3 +0)
  R3 unbounded memory access, make sure to bounds check any such access
  processed 97 insns (limit 1000000) max_states_per_insn 1 total_states 10 peak_states 10 mark_read 6
  -- END PROG LOAD LOG --
  libbpf: prog 'ingress_fwdns_prio100': failed to load: -13
  libbpf: failed to load object 'test_tc_dtime'
  libbpf: failed to load BPF skeleton 'test_tc_dtime': -13
  ...

At insn 14, with condition 'w1 < 9', register r1 is changed from an arbitrary
u32 value to `scalar(umax=8,var_off=(0x0; 0xf))`. Register r2, however, remains
as an arbitrary u32 value. Current verifier won't claim r1/r2 equality if
the previous mov is alu32 ('w2 = w1').

If r1 upper 32bit value is not 0, we indeed cannot clamin r1/r2 equality
after 'w2 = w1'. But in this particular case, we know r1 upper 32bit value
is 0, so it is safe to claim r1/r2 equality. This patch exactly did this.
For a 32bit subreg mov, if the src register upper 32bit is 0,
it is okay to claim equality between src and dst registers.

With this patch, the above verification sequence becomes

  ...
  8: (18) r1 = 0xffffc9000048e230       ; R1_w=map_value(off=560,ks=4,vs=564,imm=0)
  10: (61) r1 = *(u32 *)(r1 +0)         ; R1_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
  ; if (skb->tstamp == EGRESS_ENDHOST_MAGIC)
  11: (79) r2 = *(u64 *)(r6 +152)       ; R2_w=scalar() R6=ctx(off=0,imm=0)
  ; if (skb->tstamp == EGRESS_ENDHOST_MAGIC)
  12: (55) if r2 != 0xb9fbeef goto pc+10        ; R2_w=195018479
  13: (bc) w2 = w1                      ; R1_w=scalar(id=6,umax=4294967295,var_off=(0x0; 0xffffffff)) R2_w=scalar(id=6,umax=4294967295,var_off=(0x0; 0xffffffff))
  ; if (test < __NR_TESTS)
  14: (a6) if w1 < 0x9 goto pc+1        ; R1_w=scalar(id=6,umin=9,umax=4294967295,var_off=(0x0; 0xffffffff))
  ...
  from 14 to 16: R0=2 R1_w=scalar(id=6,umax=8,var_off=(0x0; 0xf)) R2_w=scalar(id=6,umax=8,var_off=(0x0; 0xf)) R6=ctx(off=0,imm=0) R10=fp0
  16: (27) r2 *= 28                     ; R2_w=scalar(umax=224,var_off=(0x0; 0xfc))
  17: (18) r3 = 0xffffc9000048e118      ; R3_w=map_value(off=280,ks=4,vs=564,imm=0)
  19: (0f) r3 += r2
  20: (61) r2 = *(u32 *)(r3 +0)         ; R2_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R3_w=map_value(off=280,ks=4,vs=564,umax=224,var_off=(0x0; 0xfc),s32_max=252,u32_max=252)
  ...

and eventually the bpf program can be verified successfully.

  [1] https://reviews.llvm.org/D147968

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20230417222134.359714-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: lirc program type should not require SYS_CAP_ADMIN
Sean Young [Mon, 17 Apr 2023 08:17:48 +0000 (09:17 +0100)]
bpf: lirc program type should not require SYS_CAP_ADMIN

Make it possible to load lirc program type with just CAP_BPF. There is
nothing exceptional about lirc programs that means they require
SYS_CAP_ADMIN.

In order to attach or detach a lirc program type you need permission to
open /dev/lirc0; if you have permission to do that, you can alter all
sorts of lirc receiving options. Changing the IR protocol decoder is no
different.

Right now on a typical distribution /dev/lirc devices are only
read/write by root. Ideally we would make them group read/write like
other devices so that local users can use them without becoming root.

Signed-off-by: Sean Young <sean@mess.org>
Link: https://lore.kernel.org/r/ZD0ArKpwnDBJZsrE@gofer.mess.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Set skb redirect and from_ingress info in __bpf_tx_skb
Daniel Borkmann [Mon, 17 Apr 2023 13:49:15 +0000 (15:49 +0200)]
bpf: Set skb redirect and from_ingress info in __bpf_tx_skb

There are some use-cases where it is desirable to use bpf_redirect()
in combination with ifb device, which currently is not supported, for
example, around filtering inbound traffic with BPF to then push it to
ifb which holds the qdisc for shaping in contrast to doing that on the
egress device.

Toke mentions the following case related to OpenWrt:

   Because there's not always a single egress on the other side. These are
   mainly home routers, which tend to have one or more WiFi devices bridged
   to one or more ethernet ports on the LAN side, and a single upstream WAN
   port. And the objective is to control the total amount of traffic going
   over the WAN link (in both directions), to deal with bufferbloat in the
   ISP network (which is sadly still all too prevalent).

   In this setup, the traffic can be split arbitrarily between the links
   on the LAN side, and the only "single bottleneck" is the WAN link. So we
   install both egress and ingress shapers on this, configured to something
   like 95-98% of the true link bandwidth, thus moving the queues into the
   qdisc layer in the router. It's usually necessary to set the ingress
   bandwidth shaper a bit lower than the egress due to being "downstream"
   of the bottleneck link, but it does work surprisingly well.

   We usually use something like a matchall filter to put all ingress
   traffic on the ifb, so doing the redirect from BPF has not been an
   immediate requirement thus far. However, it does seem a bit odd that
   this is not possible, and we do have a BPF-based filter that layers on
   top of this kind of setup, which currently uses u32 as the ingress
   filter and so it could presumably be improved to use BPF instead if
   that was available.

Reported-by: Toke Høiland-Jørgensen <toke@redhat.com>
Reported-by: Yafang Shao <laoar.shao@gmail.com>
Reported-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://git.openwrt.org/?p=project/qosify.git;a=blob;f=README
Link: https://lore.kernel.org/bpf/875y9yzbuy.fsf@toke.dk
Link: https://lore.kernel.org/r/8cebc8b2b6e967e10cbafe2ffd6795050e74accd.1681739137.git.daniel@iogearbox.net
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoMerge branch 'Remove KF_KPTR_GET kfunc flag'
Alexei Starovoitov [Sun, 16 Apr 2023 15:51:24 +0000 (08:51 -0700)]
Merge branch 'Remove KF_KPTR_GET kfunc flag'

David Vernet says:

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

We've managed to improve the UX for kptrs significantly over the last 9
months. All of the existing use cases which previously had KF_KPTR_GET
kfuncs (struct bpf_cpumask *, struct task_struct *, and struct cgroup *)
have all been updated to be synchronized using RCU. In other words,
their KF_KPTR_GET kfuncs have been removed in favor of KF_RCU |
KF_ACQUIRE kfuncs, with the pointers themselves also being readable from
maps in an RCU read region thanks to the types being RCU safe.

While KF_KPTR_GET was a logical starting point for kptrs, it's become
clear that they're not the correct abstraction. KF_KPTR_GET is a flag
that essentially does nothing other than enforcing that the argument to
a function is a pointer to a referenced kptr map value. At first glance,
that's a useful thing to guarantee to a kfunc. It gives kfuncs the
ability to try and acquire a reference on that kptr without requiring
the BPF prog to do something like this:

struct kptr_type *in_map, *new = NULL;

in_map = bpf_kptr_xchg(&map->value, NULL);
if (in_map) {
new = bpf_kptr_type_acquire(in_map);
in_map = bpf_kptr_xchg(&map->value, in_map);
if (in_map)
bpf_kptr_type_release(in_map);
}

That's clearly a pretty ugly (and racy) UX, and if using KF_KPTR_GET is
the only alternative, it's better than nothing. However, the problem
with any KF_KPTR_GET kfunc lies in the fact that it always requires some
kind of synchronization in order to safely do an opportunistic acquire
of the kptr in the map. This is because a BPF program running on another
CPU could do a bpf_kptr_xchg() on that map value, and free the kptr
after it's been read by the KF_KPTR_GET kfunc. For example, the
now-removed bpf_task_kptr_get() kfunc did the following:

struct task_struct *bpf_task_kptr_get(struct task_struct **pp)
{
    struct task_struct *p;

rcu_read_lock();
p = READ_ONCE(*pp);
/* If p is non-NULL, it could still be freed by another CPU,
   * so we have to do an opportunistic refcount_inc_not_zero()
 * and return NULL if the task will be freed after the
 * current RCU read region.
 */
|f (p && !refcount_inc_not_zero(&p->rcu_users))
p = NULL;
rcu_read_unlock();

return p;
}

In other words, the kfunc uses RCU to ensure that the task remains valid
after it's been peeked from the map. However, this is completely
redundant with just defining a KF_RCU kfunc that itself does a
refcount_inc_not_zero(), which is exactly what bpf_task_acquire() now
does.

So, the question of whether KF_KPTR_GET is useful is actually, "Are
there any synchronization mechanisms / safety flags that are required by
certain kptrs, but which are not provided by the verifier to kfuncs?"
The answer to that question today is "No", because every kptr we
currently care about is RCU protected.

Even if the answer ever became "yes", the proper way to support that
referenced kptr type would be to add support for whatever
synchronization mechanism it requires in the verifier, rather than
giving kfuncs a flag that says, "Here's a pointer to a referenced kptr
in a map, do whatever you need to do."

With all that said -- so as to allow us to consolidate the kfunc API,
and simplify the verifier, this patchset removes the KF_KPTR_GET kfunc
flag.
---

This is v2 of this patchset

v1: https://lore.kernel.org/all/20230415103231.236063-1-void@manifault.com/

Changelog:
----------

v1 -> v2:
- Fix KF_RU -> KF_RCU typo in commit summary for patch 2/3, and in cover
  letter (Alexei)
- In order to reduce churn, don't shift all KF_* flags down by 1. We'll
  just fill the now-empty slot the next time we add a flag (Alexei)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf,docs: Remove KF_KPTR_GET from documentation
David Vernet [Sun, 16 Apr 2023 08:49:28 +0000 (03:49 -0500)]
bpf,docs: Remove KF_KPTR_GET from documentation

A prior patch removed KF_KPTR_GET from the kernel. Now that it's no
longer accessible to kfunc authors, this patch removes it from the BPF
kfunc documentation.

Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230416084928.326135-4-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Remove KF_KPTR_GET kfunc flag
David Vernet [Sun, 16 Apr 2023 08:49:27 +0000 (03:49 -0500)]
bpf: Remove KF_KPTR_GET kfunc flag

We've managed to improve the UX for kptrs significantly over the last 9
months. All of the existing use cases which previously had KF_KPTR_GET
kfuncs (struct bpf_cpumask *, struct task_struct *, and struct cgroup *)
have all been updated to be synchronized using RCU. In other words,
their KF_KPTR_GET kfuncs have been removed in favor of KF_RCU |
KF_ACQUIRE kfuncs, with the pointers themselves also being readable from
maps in an RCU read region thanks to the types being RCU safe.

While KF_KPTR_GET was a logical starting point for kptrs, it's become
clear that they're not the correct abstraction. KF_KPTR_GET is a flag
that essentially does nothing other than enforcing that the argument to
a function is a pointer to a referenced kptr map value. At first glance,
that's a useful thing to guarantee to a kfunc. It gives kfuncs the
ability to try and acquire a reference on that kptr without requiring
the BPF prog to do something like this:

struct kptr_type *in_map, *new = NULL;

in_map = bpf_kptr_xchg(&map->value, NULL);
if (in_map) {
        new = bpf_kptr_type_acquire(in_map);
        in_map = bpf_kptr_xchg(&map->value, in_map);
        if (in_map)
                bpf_kptr_type_release(in_map);
}

That's clearly a pretty ugly (and racy) UX, and if using KF_KPTR_GET is
the only alternative, it's better than nothing. However, the problem
with any KF_KPTR_GET kfunc lies in the fact that it always requires some
kind of synchronization in order to safely do an opportunistic acquire
of the kptr in the map. This is because a BPF program running on another
CPU could do a bpf_kptr_xchg() on that map value, and free the kptr
after it's been read by the KF_KPTR_GET kfunc. For example, the
now-removed bpf_task_kptr_get() kfunc did the following:

struct task_struct *bpf_task_kptr_get(struct task_struct **pp)
{
            struct task_struct *p;

        rcu_read_lock();
        p = READ_ONCE(*pp);
        /* If p is non-NULL, it could still be freed by another CPU,
         * so we have to do an opportunistic refcount_inc_not_zero()
         * and return NULL if the task will be freed after the
         * current RCU read region.
         */
        |f (p && !refcount_inc_not_zero(&p->rcu_users))
                p = NULL;
        rcu_read_unlock();

        return p;
}

In other words, the kfunc uses RCU to ensure that the task remains valid
after it's been peeked from the map. However, this is completely
redundant with just defining a KF_RCU kfunc that itself does a
refcount_inc_not_zero(), which is exactly what bpf_task_acquire() now
does.

So, the question of whether KF_KPTR_GET is useful is actually, "Are
there any synchronization mechanisms / safety flags that are required by
certain kptrs, but which are not provided by the verifier to kfuncs?"
The answer to that question today is "No", because every kptr we
currently care about is RCU protected.

Even if the answer ever became "yes", the proper way to support that
referenced kptr type would be to add support for whatever
synchronization mechanism it requires in the verifier, rather than
giving kfuncs a flag that says, "Here's a pointer to a referenced kptr
in a map, do whatever you need to do."

With all that said -- so as to allow us to consolidate the kfunc API,
and simplify the verifier a bit, this patch removes KF_KPTR_GET, and all
relevant logic from the verifier.

Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230416084928.326135-3-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Remove bpf_kfunc_call_test_kptr_get() test kfunc
David Vernet [Sun, 16 Apr 2023 08:49:26 +0000 (03:49 -0500)]
bpf: Remove bpf_kfunc_call_test_kptr_get() test kfunc

We've managed to improve the UX for kptrs significantly over the last 9
months. All of the prior main use cases, struct bpf_cpumask *, struct
task_struct *, and struct cgroup *, have all been updated to be
synchronized mainly using RCU. In other words, their KF_ACQUIRE kfunc
calls are all KF_RCU, and the pointers themselves are MEM_RCU and can be
accessed in an RCU read region in BPF.

In a follow-on change, we'll be removing the KF_KPTR_GET kfunc flag.
This patch prepares for that by removing the
bpf_kfunc_call_test_kptr_get() kfunc, and all associated selftests.

Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230416084928.326135-2-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoMerge branch 'Shared ownership for local kptrs'
Alexei Starovoitov [Sun, 16 Apr 2023 00:36:50 +0000 (17:36 -0700)]
Merge branch 'Shared ownership for local kptrs'

Dave Marchevsky says:

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

This series adds support for refcounted local kptrs to the verifier. A local
kptr is 'refcounted' if its type contains a struct bpf_refcount field:

  struct refcounted_node {
    long data;
    struct bpf_list_node ll;
    struct bpf_refcount ref;
  };

bpf_refcount is used to implement shared ownership for local kptrs.

Motivating usecase
==================

If a struct has two collection node fields, e.g.:

  struct node {
    long key;
    long val;
    struct bpf_rb_node rb;
    struct bpf_list_node ll;
  };

It's not currently possible to add a node to both the list and rbtree:

  long bpf_prog(void *ctx)
  {
    struct node *n = bpf_obj_new(typeof(*n));
    if (!n) { /* ... */ }

    bpf_spin_lock(&lock);

    bpf_list_push_back(&head, &n->ll);
    bpf_rbtree_add(&root, &n->rb, less); /* Assume a resonable less() */
    bpf_spin_unlock(&lock);
  }

The above program will fail verification due to current owning / non-owning ref
logic: after bpf_list_push_back, n is a non-owning reference and thus cannot be
passed to bpf_rbtree_add. The only way to get an owning reference for the node
that was added is to bpf_list_pop_{front,back} it.

More generally, verifier ownership semantics expect that a node has one
owner (program, collection, or stashed in map) with exclusive ownership
of the node's lifetime. The owner free's the node's underlying memory when it
itself goes away.

Without a shared ownership concept it's impossible to express many real-world
usecases such that they pass verification.

Semantic Changes
================

Before this series, the verifier could make this statement: "whoever has the
owning reference has exclusive ownership of the referent's lifetime". As
demonstrated in the previous section, this implies that a BPF program can't
have an owning reference to some node if that node is in a collection. If
such a state were possible, the node would have multiple owners, each thinking
they have exclusive ownership. In order to support shared ownership it's
necessary to modify the exclusive ownership semantic.

After this series' changes, an owning reference has ownership of the referent's
lifetime, but it's not necessarily exclusive. The referent's underlying memory
is guaranteed to be valid (i.e. not free'd) until the reference is dropped or
used for collection insert.

This change doesn't affect UX of owning or non-owning references much:

  * insert kfuncs (bpf_rbtree_add, bpf_list_push_{front,back}) still require
    an owning reference arg, as ownership still must be passed to the
    collection in a shared-ownership world.

  * non-owning references still refer to valid memory without claiming
    any ownership.

One important conclusion that followed from "exclusive ownership" statement
is no longer valid, though. In exclusive-ownership world, if a BPF prog has
an owning reference to a node, the verifier can conclude that no collection has
ownership of it. This conclusion was used to avoid runtime checking in the
implementations of insert and remove operations (""has the node already been
{inserted, removed}?").

In a shared-ownership world the aforementioned conclusion is no longer valid,
which necessitates doing runtime checking in insert and remove operation
kfuncs, and those functions possibly failing to insert or remove anything.

Luckily the verifier changes necessary to go from exclusive to shared ownership
were fairly minimal. Patches in this series which do change verifier semantics
generally have some summary dedicated to explaining why certain usecases
Just Work for shared ownership without verifier changes.

Implementation
==============

The changes in this series can be categorized as follows:

  * struct bpf_refcount opaque field + plumbing
  * support for refcounted kptrs in bpf_obj_new and bpf_obj_drop
  * bpf_refcount_acquire kfunc
    * enables shared ownershp by bumping refcount + acquiring owning ref
  * support for possibly-failing collection insertion and removal
    * insertion changes are more complex

If a patch's changes have some nuance to their effect - or lack of effect - on
verifier behavior, the patch summary talks about it at length.

Patch contents:
  * Patch 1 removes btf_field_offs struct
  * Patch 2 adds struct bpf_refcount and associated plumbing
  * Patch 3 modifies semantics of bpf_obj_drop and bpf_obj_new to handle
    refcounted kptrs
  * Patch 4 adds bpf_refcount_acquire
  * Patches 5-7 add support for possibly-failing collection insert and remove
  * Patch 8 centralizes constructor-like functionality for local kptr types
  * Patch 9 adds tests for new functionality

base-commit: 4a1e885c6d143ff1b557ec7f3fc6ddf39c51502f

Changelog:

v1 -> v2: lore.kernel.org/bpf/20230410190753.2012798-1-davemarchevsky@fb.com

Patch #s used below refer to the patch's position in v1 unless otherwise
specified.

  * General
    * Rebase onto latest bpf-next (base-commit updated above)

  * Patch 4 - "bpf: Add bpf_refcount_acquire kfunc"
    * Fix typo in summary (Alexei)
  * Patch 7 - "Migrate bpf_rbtree_remove to possibly fail"
    * Modify a paragraph in patch summary to more clearly state that only
      bpf_rbtree_remove's non-owning ref clobbering behavior is changed by the
      patch (Alexei)
    * refcount_off == -1 -> refcount_off < 0  in "node type w/ both list
      and rb_node fields" check, since any negative value means "no
      bpf_refcount field found", and furthermore refcount_off is never
      explicitly set to -1, but rather -EINVAL. (Alexei)
    * Instead of just changing "btf: list_node and rb_node in same struct" test
      expectation to pass instead of fail, do some refactoring to test both
      "list_node, rb_node, and bpf_refcount" (success) and "list_node, rb_node,
      _no_ bpf_refcount" (failure) cases. This ensures that logic change in
      previous bullet point is correct.
      * v1's "btf: list_node and rb_node in same struct" test changes didn't
        add bpf_refcount, so the fact that btf load succeeded w/ list and
        rb_nodes but no bpf_refcount field is further proof that this logic
        was incorrect in v1.
  * Patch 8 - "bpf: Centralize btf_field-specific initialization logic"
    * Instead of doing __init_field_infer_size in kfuncs when taking
      bpf_list_head type input which might've been 0-initialized in map, go
      back to simple oneliner initialization. Add short comment explaining why
      this is necessary. (Alexei)
  * Patch 9 - "selftests/bpf: Add refcounted_kptr tests"
    * Don't __always_inline helper fns in progs/refcounted_kptr.c (Alexei)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Add refcounted_kptr tests
Dave Marchevsky [Sat, 15 Apr 2023 20:18:11 +0000 (13:18 -0700)]
selftests/bpf: Add refcounted_kptr tests

Test refcounted local kptr functionality added in previous patches in
the series.

Usecases which pass verification:

* Add refcounted local kptr to both tree and list. Then, read and -
  possibly, depending on test variant - delete from tree, then list.
  * Also test doing read-and-maybe-delete in opposite order
* Stash a refcounted local kptr in a map_value, then add it to a
  rbtree. Read from both, possibly deleting after tree read.
* Add refcounted local kptr to both tree and list. Then, try reading and
  deleting twice from one of the collections.
* bpf_refcount_acquire of just-added non-owning ref should work, as
  should bpf_refcount_acquire of owning ref just out of bpf_obj_new

Usecases which fail verification:

* The simple successful bpf_refcount_acquire cases from above should
  both fail to verify if the newly-acquired owning ref is not dropped

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-10-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Centralize btf_field-specific initialization logic
Dave Marchevsky [Sat, 15 Apr 2023 20:18:10 +0000 (13:18 -0700)]
bpf: Centralize btf_field-specific initialization logic

All btf_fields in an object are 0-initialized by memset in
bpf_obj_init. This might not be a valid initial state for some field
types, in which case kfuncs that use the type will properly initialize
their input if it's been 0-initialized. Some BPF graph collection types
and kfuncs do this: bpf_list_{head,node} and bpf_rb_node.

An earlier patch in this series added the bpf_refcount field, for which
the 0 state indicates that the refcounted object should be free'd.
bpf_obj_init treats this field specially, setting refcount to 1 instead
of relying on scattered "refcount is 0? Must have just been initialized,
let's set to 1" logic in kfuncs.

This patch extends this treatment to list and rbtree field types,
allowing most scattered initialization logic in kfuncs to be removed.

Note that bpf_{list_head,rb_root} may be inside a BPF map, in which case
they'll be 0-initialized without passing through the newly-added logic,
so scattered initialization logic must remain for these collection root
types.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-9-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Migrate bpf_rbtree_remove to possibly fail
Dave Marchevsky [Sat, 15 Apr 2023 20:18:09 +0000 (13:18 -0700)]
bpf: Migrate bpf_rbtree_remove to possibly fail

This patch modifies bpf_rbtree_remove to account for possible failure
due to the input rb_node already not being in any collection.
The function can now return NULL, and does when the aforementioned
scenario occurs. As before, on successful removal an owning reference to
the removed node is returned.

Adding KF_RET_NULL to bpf_rbtree_remove's kfunc flags - now KF_RET_NULL |
KF_ACQUIRE - provides the desired verifier semantics:

  * retval must be checked for NULL before use
  * if NULL, retval's ref_obj_id is released
  * retval is a "maybe acquired" owning ref, not a non-owning ref,
    so it will live past end of critical section (bpf_spin_unlock), and
    thus can be checked for NULL after the end of the CS

BPF programs must add checks
============================

This does change bpf_rbtree_remove's verifier behavior. BPF program
writers will need to add NULL checks to their programs, but the
resulting UX looks natural:

  bpf_spin_lock(&glock);

  n = bpf_rbtree_first(&ghead);
  if (!n) { /* ... */}
  res = bpf_rbtree_remove(&ghead, &n->node);

  bpf_spin_unlock(&glock);

  if (!res)  /* Newly-added check after this patch */
    return 1;

  n = container_of(res, /* ... */);
  /* Do something else with n */
  bpf_obj_drop(n);
  return 0;

The "if (!res)" check above is the only addition necessary for the above
program to pass verification after this patch.

bpf_rbtree_remove no longer clobbers non-owning refs
====================================================

An issue arises when bpf_rbtree_remove fails, though. Consider this
example:

  struct node_data {
    long key;
    struct bpf_list_node l;
    struct bpf_rb_node r;
    struct bpf_refcount ref;
  };

  long failed_sum;

  void bpf_prog()
  {
    struct node_data *n = bpf_obj_new(/* ... */);
    struct bpf_rb_node *res;
    n->key = 10;

    bpf_spin_lock(&glock);

    bpf_list_push_back(&some_list, &n->l); /* n is now a non-owning ref */
    res = bpf_rbtree_remove(&some_tree, &n->r, /* ... */);
    if (!res)
      failed_sum += n->key;  /* not possible */

    bpf_spin_unlock(&glock);
    /* if (res) { do something useful and drop } ... */
  }

The bpf_rbtree_remove in this example will always fail. Similarly to
bpf_spin_unlock, bpf_rbtree_remove is a non-owning reference
invalidation point. The verifier clobbers all non-owning refs after a
bpf_rbtree_remove call, so the "failed_sum += n->key" line will fail
verification, and in fact there's no good way to get information about
the node which failed to add after the invalidation. This patch removes
non-owning reference invalidation from bpf_rbtree_remove to allow the
above usecase to pass verification. The logic for why this is now
possible is as follows:

Before this series, bpf_rbtree_add couldn't fail and thus assumed that
its input, a non-owning reference, was in the tree. But it's easy to
construct an example where two non-owning references pointing to the same
underlying memory are acquired and passed to rbtree_remove one after
another (see rbtree_api_release_aliasing in
selftests/bpf/progs/rbtree_fail.c).

So it was necessary to clobber non-owning refs to prevent this
case and, more generally, to enforce "non-owning ref is definitely
in some collection" invariant. This series removes that invariant and
the failure / runtime checking added in this patch provide a clean way
to deal with the aliasing issue - just fail to remove.

Because the aliasing issue prevented by clobbering non-owning refs is no
longer an issue, this patch removes the invalidate_non_owning_refs
call from verifier handling of bpf_rbtree_remove. Note that
bpf_spin_unlock - the other caller of invalidate_non_owning_refs -
clobbers non-owning refs for a different reason, so its clobbering
behavior remains unchanged.

No BPF program changes are necessary for programs to remain valid as a
result of this clobbering change. A valid program before this patch
passed verification with its non-owning refs having shorter (or equal)
lifetimes due to more aggressive clobbering.

Also, update existing tests to check bpf_rbtree_remove retval for NULL
where necessary, and move rbtree_api_release_aliasing from
progs/rbtree_fail.c to progs/rbtree.c since it's now expected to pass
verification.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-8-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Modify linked_list tests to work with macro-ified inserts
Dave Marchevsky [Sat, 15 Apr 2023 20:18:08 +0000 (13:18 -0700)]
selftests/bpf: Modify linked_list tests to work with macro-ified inserts

The linked_list tests use macros and function pointers to reduce code
duplication. Earlier in the series, bpf_list_push_{front,back} were
modified to be macros, expanding to invoke actual kfuncs
bpf_list_push_{front,back}_impl. Due to this change, a code snippet
like:

  void (*p)(void *, void *) = (void *)&bpf_list_##op;
  p(hexpr, nexpr);

meant to do bpf_list_push_{front,back}(hexpr, nexpr), will no longer
work as it's no longer valid to do &bpf_list_push_{front,back} since
they're no longer functions.

This patch fixes issues of this type, along with two other minor changes
- one improvement and one fix - both related to the node argument to
list_push_{front,back}.

  * The fix: migration of list_push tests away from (void *, void *)
    func ptr uncovered that some tests were incorrectly passing pointer
    to node, not pointer to struct bpf_list_node within the node. This
    patch fixes such issues (CHECK(..., f) -> CHECK(..., &f->node))

  * The improvement: In linked_list tests, the struct foo type has two
    list_node fields: node and node2, at byte offsets 0 and 40 within
    the struct, respectively. Currently node is used in ~all tests
    involving struct foo and lists. The verifier needs to do some work
    to account for the offset of bpf_list_node within the node type, so
    using node2 instead of node exercises that logic more in the tests.
    This patch migrates linked_list tests to use node2 instead of node.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-7-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Migrate bpf_rbtree_add and bpf_list_push_{front,back} to possibly fail
Dave Marchevsky [Sat, 15 Apr 2023 20:18:07 +0000 (13:18 -0700)]
bpf: Migrate bpf_rbtree_add and bpf_list_push_{front,back} to possibly fail

Consider this code snippet:

  struct node {
    long key;
    bpf_list_node l;
    bpf_rb_node r;
    bpf_refcount ref;
  }

  int some_bpf_prog(void *ctx)
  {
    struct node *n = bpf_obj_new(/*...*/), *m;

    bpf_spin_lock(&glock);

    bpf_rbtree_add(&some_tree, &n->r, /* ... */);
    m = bpf_refcount_acquire(n);
    bpf_rbtree_add(&other_tree, &m->r, /* ... */);

    bpf_spin_unlock(&glock);

    /* ... */
  }

After bpf_refcount_acquire, n and m point to the same underlying memory,
and that node's bpf_rb_node field is being used by the some_tree insert,
so overwriting it as a result of the second insert is an error. In order
to properly support refcounted nodes, the rbtree and list insert
functions must be allowed to fail. This patch adds such support.

The kfuncs bpf_rbtree_add, bpf_list_push_{front,back} are modified to
return an int indicating success/failure, with 0 -> success, nonzero ->
failure.

bpf_obj_drop on failure
=======================

Currently the only reason an insert can fail is the example above: the
bpf_{list,rb}_node is already in use. When such a failure occurs, the
insert kfuncs will bpf_obj_drop the input node. This allows the insert
operations to logically fail without changing their verifier owning ref
behavior, namely the unconditional release_reference of the input
owning ref.

With insert that always succeeds, ownership of the node is always passed
to the collection, since the node always ends up in the collection.

With a possibly-failed insert w/ bpf_obj_drop, ownership of the node
is always passed either to the collection (success), or to bpf_obj_drop
(failure). Regardless, it's correct to continue unconditionally
releasing the input owning ref, as something is always taking ownership
from the calling program on insert.

Keeping owning ref behavior unchanged results in a nice default UX for
insert functions that can fail. If the program's reaction to a failed
insert is "fine, just get rid of this owning ref for me and let me go
on with my business", then there's no reason to check for failure since
that's default behavior. e.g.:

  long important_failures = 0;

  int some_bpf_prog(void *ctx)
  {
    struct node *n, *m, *o; /* all bpf_obj_new'd */

    bpf_spin_lock(&glock);
    bpf_rbtree_add(&some_tree, &n->node, /* ... */);
    bpf_rbtree_add(&some_tree, &m->node, /* ... */);
    if (bpf_rbtree_add(&some_tree, &o->node, /* ... */)) {
      important_failures++;
    }
    bpf_spin_unlock(&glock);
  }

If we instead chose to pass ownership back to the program on failed
insert - by returning NULL on success or an owning ref on failure -
programs would always have to do something with the returned ref on
failure. The most likely action is probably "I'll just get rid of this
owning ref and go about my business", which ideally would look like:

  if (n = bpf_rbtree_add(&some_tree, &n->node, /* ... */))
    bpf_obj_drop(n);

But bpf_obj_drop isn't allowed in a critical section and inserts must
occur within one, so in reality error handling would become a
hard-to-parse mess.

For refcounted nodes, we can replicate the "pass ownership back to
program on failure" logic with this patch's semantics, albeit in an ugly
way:

  struct node *n = bpf_obj_new(/* ... */), *m;

  bpf_spin_lock(&glock);

  m = bpf_refcount_acquire(n);
  if (bpf_rbtree_add(&some_tree, &n->node, /* ... */)) {
    /* Do something with m */
  }

  bpf_spin_unlock(&glock);
  bpf_obj_drop(m);

bpf_refcount_acquire is used to simulate "return owning ref on failure".
This should be an uncommon occurrence, though.

Addition of two verifier-fixup'd args to collection inserts
===========================================================

The actual bpf_obj_drop kfunc is
bpf_obj_drop_impl(void *, struct btf_struct_meta *), with bpf_obj_drop
macro populating the second arg with 0 and the verifier later filling in
the arg during insn fixup.

Because bpf_rbtree_add and bpf_list_push_{front,back} now might do
bpf_obj_drop, these kfuncs need a btf_struct_meta parameter that can be
passed to bpf_obj_drop_impl.

Similarly, because the 'node' param to those insert functions is the
bpf_{list,rb}_node within the node type, and bpf_obj_drop expects a
pointer to the beginning of the node, the insert functions need to be
able to find the beginning of the node struct. A second
verifier-populated param is necessary: the offset of {list,rb}_node within the
node type.

These two new params allow the insert kfuncs to correctly call
__bpf_obj_drop_impl:

  beginning_of_node = bpf_rb_node_ptr - offset
  if (already_inserted)
    __bpf_obj_drop_impl(beginning_of_node, btf_struct_meta->record);

Similarly to other kfuncs with "hidden" verifier-populated params, the
insert functions are renamed with _impl prefix and a macro is provided
for common usage. For example, bpf_rbtree_add kfunc is now
bpf_rbtree_add_impl and bpf_rbtree_add is now a macro which sets
"hidden" args to 0.

Due to the two new args BPF progs will need to be recompiled to work
with the new _impl kfuncs.

This patch also rewrites the "hidden argument" explanation to more
directly say why the BPF program writer doesn't need to populate the
arguments with anything meaningful.

How does this new logic affect non-owning references?
=====================================================

Currently, non-owning refs are valid until the end of the critical
section in which they're created. We can make this guarantee because, if
a non-owning ref exists, the referent was added to some collection. The
collection will drop() its nodes when it goes away, but it can't go away
while our program is accessing it, so that's not a problem. If the
referent is removed from the collection in the same CS that it was added
in, it can't be bpf_obj_drop'd until after CS end. Those are the only
two ways to free the referent's memory and neither can happen until
after the non-owning ref's lifetime ends.

On first glance, having these collection insert functions potentially
bpf_obj_drop their input seems like it breaks the "can't be
bpf_obj_drop'd until after CS end" line of reasoning. But we care about
the memory not being _freed_ until end of CS end, and a previous patch
in the series modified bpf_obj_drop such that it doesn't free refcounted
nodes until refcount == 0. So the statement can be more accurately
rewritten as "can't be free'd until after CS end".

We can prove that this rewritten statement holds for any non-owning
reference produced by collection insert functions:

* If the input to the insert function is _not_ refcounted
  * We have an owning reference to the input, and can conclude it isn't
    in any collection
    * Inserting a node in a collection turns owning refs into
      non-owning, and since our input type isn't refcounted, there's no
      way to obtain additional owning refs to the same underlying
      memory
  * Because our node isn't in any collection, the insert operation
    cannot fail, so bpf_obj_drop will not execute
  * If bpf_obj_drop is guaranteed not to execute, there's no risk of
    memory being free'd

* Otherwise, the input to the insert function is refcounted
  * If the insert operation fails due to the node's list_head or rb_root
    already being in some collection, there was some previous successful
    insert which passed refcount to the collection
  * We have an owning reference to the input, it must have been
    acquired via bpf_refcount_acquire, which bumped the refcount
  * refcount must be >= 2 since there's a valid owning reference and the
    node is already in a collection
  * Insert triggering bpf_obj_drop will decr refcount to >= 1, never
    resulting in a free

So although we may do bpf_obj_drop during the critical section, this
will never result in memory being free'd, and no changes to non-owning
ref logic are needed in this patch.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-6-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Add bpf_refcount_acquire kfunc
Dave Marchevsky [Sat, 15 Apr 2023 20:18:06 +0000 (13:18 -0700)]
bpf: Add bpf_refcount_acquire kfunc

Currently, BPF programs can interact with the lifetime of refcounted
local kptrs in the following ways:

  bpf_obj_new  - Initialize refcount to 1 as part of new object creation
  bpf_obj_drop - Decrement refcount and free object if it's 0
  collection add - Pass ownership to the collection. No change to
                   refcount but collection is responsible for
   bpf_obj_dropping it

In order to be able to add a refcounted local kptr to multiple
collections we need to be able to increment the refcount and acquire a
new owning reference. This patch adds a kfunc, bpf_refcount_acquire,
implementing such an operation.

bpf_refcount_acquire takes a refcounted local kptr and returns a new
owning reference to the same underlying memory as the input. The input
can be either owning or non-owning. To reinforce why this is safe,
consider the following code snippets:

  struct node *n = bpf_obj_new(typeof(*n)); // A
  struct node *m = bpf_refcount_acquire(n); // B

In the above snippet, n will be alive with refcount=1 after (A), and
since nothing changes that state before (B), it's obviously safe. If
n is instead added to some rbtree, we can still safely refcount_acquire
it:

  struct node *n = bpf_obj_new(typeof(*n));
  struct node *m;

  bpf_spin_lock(&glock);
  bpf_rbtree_add(&groot, &n->node, less);   // A
  m = bpf_refcount_acquire(n);              // B
  bpf_spin_unlock(&glock);

In the above snippet, after (A) n is a non-owning reference, and after
(B) m is an owning reference pointing to the same memory as n. Although
n has no ownership of that memory's lifetime, it's guaranteed to be
alive until the end of the critical section, and n would be clobbered if
we were past the end of the critical section, so it's safe to bump
refcount.

Implementation details:

* From verifier's perspective, bpf_refcount_acquire handling is similar
  to bpf_obj_new and bpf_obj_drop. Like the former, it returns a new
  owning reference matching input type, although like the latter, type
  can be inferred from concrete kptr input. Verifier changes in
  {check,fixup}_kfunc_call and check_kfunc_args are largely copied from
  aforementioned functions' verifier changes.

* An exception to the above is the new KF_ARG_PTR_TO_REFCOUNTED_KPTR
  arg, indicated by new "__refcounted_kptr" kfunc arg suffix. This is
  necessary in order to handle both owning and non-owning input without
  adding special-casing to "__alloc" arg handling. Also a convenient
  place to confirm that input type has bpf_refcount field.

* The implemented kfunc is actually bpf_refcount_acquire_impl, with
  'hidden' second arg that the verifier sets to the type's struct_meta
  in fixup_kfunc_call.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-5-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Support refcounted local kptrs in existing semantics
Dave Marchevsky [Sat, 15 Apr 2023 20:18:05 +0000 (13:18 -0700)]
bpf: Support refcounted local kptrs in existing semantics

A local kptr is considered 'refcounted' when it is of a type that has a
bpf_refcount field. When such a kptr is created, its refcount should be
initialized to 1; when destroyed, the object should be free'd only if a
refcount decr results in 0 refcount.

Existing logic always frees the underlying memory when destroying a
local kptr, and 0-initializes all btf_record fields. This patch adds
checks for "is local kptr refcounted?" and new logic for that case in
the appropriate places.

This patch focuses on changing existing semantics and thus conspicuously
does _not_ provide a way for BPF programs in increment refcount. That
follows later in the series.

__bpf_obj_drop_impl is modified to do the right thing when it sees a
refcounted type. Container types for graph nodes (list, tree, stashed in
map) are migrated to use __bpf_obj_drop_impl as a destructor for their
nodes instead of each having custom destruction code in their _free
paths. Now that "drop" isn't a synonym for "free" when the type is
refcounted it makes sense to centralize this logic.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-4-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Introduce opaque bpf_refcount struct and add btf_record plumbing
Dave Marchevsky [Sat, 15 Apr 2023 20:18:04 +0000 (13:18 -0700)]
bpf: Introduce opaque bpf_refcount struct and add btf_record plumbing

A 'struct bpf_refcount' is added to the set of opaque uapi/bpf.h types
meant for use in BPF programs. Similarly to other opaque types like
bpf_spin_lock and bpf_rbtree_node, the verifier needs to know where in
user-defined struct types a bpf_refcount can be located, so necessary
btf_record plumbing is added to enable this. bpf_refcount is sized to
hold a refcount_t.

Similarly to bpf_spin_lock, the offset of a bpf_refcount is cached in
btf_record as refcount_off in addition to being in the field array.
Caching refcount_off makes sense for this field because further patches
in the series will modify functions that take local kptrs (e.g.
bpf_obj_drop) to change their behavior if the type they're operating on
is refcounted. So enabling fast "is this type refcounted?" checks is
desirable.

No such verifier behavior changes are introduced in this patch, just
logic to recognize 'struct bpf_refcount' in btf_record.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-3-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Remove btf_field_offs, use btf_record's fields instead
Dave Marchevsky [Sat, 15 Apr 2023 20:18:03 +0000 (13:18 -0700)]
bpf: Remove btf_field_offs, use btf_record's fields instead

The btf_field_offs struct contains (offset, size) for btf_record fields,
sorted by offset. btf_field_offs is always used in conjunction with
btf_record, which has btf_field 'fields' array with (offset, type), the
latter of which btf_field_offs' size is derived from via
btf_field_type_size.

This patch adds a size field to struct btf_field and sorts btf_record's
fields by offset, making it possible to get rid of btf_field_offs. Less
data duplication and less code complexity results.

Since btf_field_offs' lifetime closely followed the btf_record used to
populate it, most complexity wins are from removal of initialization
code like:

  if (btf_record_successfully_initialized) {
    foffs = btf_parse_field_offs(rec);
    if (IS_ERR_OR_NULL(foffs))
      // free the btf_record and return err
  }

Other changes in this patch are pretty mechanical:

  * foffs->field_off[i] -> rec->fields[i].offset
  * foffs->field_sz[i] -> rec->fields[i].size
  * Sort rec->fields in btf_parse_fields before returning
    * It's possible that this is necessary independently of other
      changes in this patch. btf_record_find in syscall.c expects
      btf_record's fields to be sorted by offset, yet there's no
      explicit sorting of them before this patch, record's fields are
      populated in the order they're read from BTF struct definition.
      BTF docs don't say anything about the sortedness of struct fields.
  * All functions taking struct btf_field_offs * input now instead take
    struct btf_record *. All callsites of these functions already have
    access to the correct btf_record.

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230415201811.343116-2-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agosamples/bpf: sampleip: Replace PAGE_OFFSET with _text address
Rong Tao [Wed, 12 Apr 2023 08:16:24 +0000 (16:16 +0800)]
samples/bpf: sampleip: Replace PAGE_OFFSET with _text address

Macro PAGE_OFFSET(0xffff880000000000) in sampleip_user.c is inaccurate,
for example, in aarch64 architecture, this value depends on the
CONFIG_ARM64_VA_BITS compilation configuration, this value defaults to 48,
the corresponding PAGE_OFFSET is 0xffff800000000000, if we use the value
defined in sampleip_user.c, then all KSYMs obtained by sampleip are (user)

Symbol error due to PAGE_OFFSET error:

    $ sudo ./sampleip 1
    Sampling at 99 Hertz for 1 seconds. Ctrl-C also ends.
    ADDR                KSYM                             COUNT
    0xffff80000810ceb8  (user)                           1
    0xffffb28ec880      (user)                           1
    0xffff8000080c82b8  (user)                           1
    0xffffb23fed24      (user)                           1
    0xffffb28944fc      (user)                           1
    0xffff8000084628bc  (user)                           1
    0xffffb2a935c0      (user)                           1
    0xffff80000844677c  (user)                           1
    0xffff80000857a3a4  (user)                           1
    ...

A few examples of addresses in the CONFIG_ARM64_VA_BITS=48 environment in
the aarch64 environment:

    $ sudo head /proc/kallsyms
    ffff8000080a0000 T _text
    ffff8000080b0000 t gic_handle_irq
    ffff8000080b0000 T _stext
    ffff8000080b0000 T __irqentry_text_start
    ffff8000080b00b0 t gic_handle_irq
    ffff8000080b0230 t gic_handle_irq
    ffff8000080b03b4 T __irqentry_text_end
    ffff8000080b03b8 T __softirqentry_text_start
    ffff8000080b03c0 T __do_softirq
    ffff8000080b0718 T __entry_text_start

We just need to replace the PAGE_OFFSET with the address _text in
/proc/kallsyms to solve this problem:

    $ sudo ./sampleip 1
    Sampling at 99 Hertz for 1 seconds. Ctrl-C also ends.
    ADDR                KSYM                             COUNT
    0xffffb2892ab0      (user)                           1
    0xffffb2b1edfc      (user)                           1
    0xffff800008462834  __arm64_sys_ppoll                1
    0xffff8000084b87f4  eventfd_read                     1
    0xffffb28e6788      (user)                           1
    0xffff8000081e96d8  rcu_all_qs                       1
    0xffffb2ada878      (user)                           1
    ...

Signed-off-by: Rong Tao <rongtao@cestc.cn>
Link: https://lore.kernel.org/r/tencent_A0E82E0BEE925285F8156D540731DF805F05@qq.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Support 64-bit pointers to kfuncs
Ilya Leoshkevich [Wed, 12 Apr 2023 23:06:32 +0000 (01:06 +0200)]
bpf: Support 64-bit pointers to kfuncs

test_ksyms_module fails to emit a kfunc call targeting a module on
s390x, because the verifier stores the difference between kfunc
address and __bpf_call_base in bpf_insn.imm, which is s32, and modules
are roughly (1 << 42) bytes away from the kernel on s390x.

Fix by keeping BTF id in bpf_insn.imm for BPF_PSEUDO_KFUNC_CALLs,
and storing the absolute address in bpf_kfunc_desc.

Introduce bpf_jit_supports_far_kfunc_call() in order to limit this new
behavior to the s390x JIT. Otherwise other JITs need to be modified,
which is not desired.

Introduce bpf_get_kfunc_addr() instead of exposing both
find_kfunc_desc() and struct bpf_kfunc_desc.

In addition to sorting kfuncs by imm, also sort them by offset, in
order to handle conflicting imms from different modules. Do this on
all architectures in order to simplify code.

Factor out resolving specialized kfuncs (XPD and dynptr) from
fixup_kfunc_call(). This was required in the first place, because
fixup_kfunc_call() uses find_kfunc_desc(), which returns a const
pointer, so it's not possible to modify kfunc addr without stripping
const, which is not nice. It also removes repetition of code like:

if (bpf_jit_supports_far_kfunc_call())
desc->addr = func;
else
insn->imm = BPF_CALL_IMM(func);

and separates kfunc_desc_tab fixups from kfunc_call fixups.

Suggested-by: Jiri Olsa <olsajiri@gmail.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20230412230632.885985-1-iii@linux.ibm.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agobpf: Add preempt_count_{sub,add} into btf id deny list
Yafang [Thu, 13 Apr 2023 02:52:48 +0000 (02:52 +0000)]
bpf: Add preempt_count_{sub,add} into btf id deny list

The recursion check in __bpf_prog_enter* and __bpf_prog_exit*
leave preempt_count_{sub,add} unprotected. When attaching trampoline to
them we get panic as follows,

[  867.843050] BUG: TASK stack guard page was hit at 0000000009d325cf (stack is 0000000046a46a15..00000000537e7b28)
[  867.843064] stack guard page: 0000 [#1] PREEMPT SMP NOPTI
[  867.843067] CPU: 8 PID: 11009 Comm: trace Kdump: loaded Not tainted 6.2.0+ #4
[  867.843100] Call Trace:
[  867.843101]  <TASK>
[  867.843104]  asm_exc_int3+0x3a/0x40
[  867.843108] RIP: 0010:preempt_count_sub+0x1/0xa0
[  867.843135]  __bpf_prog_enter_recur+0x17/0x90
[  867.843148]  bpf_trampoline_6442468108_0+0x2e/0x1000
[  867.843154]  ? preempt_count_sub+0x1/0xa0
[  867.843157]  preempt_count_sub+0x5/0xa0
[  867.843159]  ? migrate_enable+0xac/0xf0
[  867.843164]  __bpf_prog_exit_recur+0x2d/0x40
[  867.843168]  bpf_trampoline_6442468108_0+0x55/0x1000
...
[  867.843788]  preempt_count_sub+0x5/0xa0
[  867.843793]  ? migrate_enable+0xac/0xf0
[  867.843829]  __bpf_prog_exit_recur+0x2d/0x40
[  867.843837] BUG: IRQ stack guard page was hit at 0000000099bd8228 (stack is 00000000b23e2bc4..000000006d95af35)
[  867.843841] BUG: IRQ stack guard page was hit at 000000005ae07924 (stack is 00000000ffd69623..0000000014eb594c)
[  867.843843] BUG: IRQ stack guard page was hit at 00000000028320f0 (stack is 00000000034b6438..0000000078d1bcec)
[  867.843842]  bpf_trampoline_6442468108_0+0x55/0x1000
...

That is because in __bpf_prog_exit_recur, the preempt_count_{sub,add} are
called after prog->active is decreased.

Fixing this by adding these two functions into btf ids deny list.

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Yafang <laoar.shao@gmail.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jiri Olsa <olsajiri@gmail.com>
Acked-by: Hao Luo <haoluo@google.com>
Link: https://lore.kernel.org/r/20230413025248.79764-1-laoar.shao@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Workaround for older vm_sockets.h.
Alexei Starovoitov [Fri, 14 Apr 2023 02:54:17 +0000 (19:54 -0700)]
selftests/bpf: Workaround for older vm_sockets.h.

Some distros ship with older vm_sockets.h that doesn't have VMADDR_CID_LOCAL
which causes selftests build to fail:
/tmp/work/bpf/bpf/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c:261:18: error: ‘VMADDR_CID_LOCAL’ undeclared (first use in this function); did you mean ‘VMADDR_CID_HOST’?
    261 |  addr->svm_cid = VMADDR_CID_LOCAL;
        |                  ^~~~~~~~~~~~~~~~
        |                  VMADDR_CID_HOST

Workaround this issue by defining it on demand.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Fix merge conflict due to SYS() macro change.
Alexei Starovoitov [Fri, 14 Apr 2023 00:22:48 +0000 (17:22 -0700)]
selftests/bpf: Fix merge conflict due to SYS() macro change.

Fix merge conflict between bpf/bpf-next trees due to change of arguments in SYS() macro.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoDaniel Borkmann says:
Jakub Kicinski [Thu, 13 Apr 2023 23:11:22 +0000 (16:11 -0700)]
Daniel Borkmann says:

====================
pull-request: bpf-next 2023-04-13

We've added 260 non-merge commits during the last 36 day(s) which contain
a total of 356 files changed, 21786 insertions(+), 11275 deletions(-).

The main changes are:

1) Rework BPF verifier log behavior and implement it as a rotating log
   by default with the option to retain old-style fixed log behavior,
   from Andrii Nakryiko.

2) Adds support for using {FOU,GUE} encap with an ipip device operating
   in collect_md mode and add a set of BPF kfuncs for controlling encap
   params, from Christian Ehrig.

3) Allow BPF programs to detect at load time whether a particular kfunc
   exists or not, and also add support for this in light skeleton,
   from Alexei Starovoitov.

4) Optimize hashmap lookups when key size is multiple of 4,
   from Anton Protopopov.

5) Enable RCU semantics for task BPF kptrs and allow referenced kptr
   tasks to be stored in BPF maps, from David Vernet.

6) Add support for stashing local BPF kptr into a map value via
   bpf_kptr_xchg(). This is useful e.g. for rbtree node creation
   for new cgroups, from Dave Marchevsky.

7) Fix BTF handling of is_int_ptr to skip modifiers to work around
   tracing issues where a program cannot be attached, from Feng Zhou.

8) Migrate a big portion of test_verifier unit tests over to
   test_progs -a verifier_* via inline asm to ease {read,debug}ability,
   from Eduard Zingerman.

9) Several updates to the instruction-set.rst documentation
   which is subject to future IETF standardization
   (https://lwn.net/Articles/926882/), from Dave Thaler.

10) Fix BPF verifier in the __reg_bound_offset's 64->32 tnum sub-register
    known bits information propagation, from Daniel Borkmann.

11) Add skb bitfield compaction work related to BPF with the overall goal
    to make more of the sk_buff bits optional, from Jakub Kicinski.

12) BPF selftest cleanups for build id extraction which stand on its own
    from the upcoming integration work of build id into struct file object,
    from Jiri Olsa.

13) Add fixes and optimizations for xsk descriptor validation and several
    selftest improvements for xsk sockets, from Kal Conley.

14) Add BPF links for struct_ops and enable switching implementations
    of BPF TCP cong-ctls under a given name by replacing backing
    struct_ops map, from Kui-Feng Lee.

15) Remove a misleading BPF verifier env->bypass_spec_v1 check on variable
    offset stack read as earlier Spectre checks cover this,
    from Luis Gerhorst.

16) Fix issues in copy_from_user_nofault() for BPF and other tracers
    to resemble copy_from_user_nmi() from safety PoV, from Florian Lehner
    and Alexei Starovoitov.

17) Add --json-summary option to test_progs in order for CI tooling to
    ease parsing of test results, from Manu Bretelle.

18) Batch of improvements and refactoring to prep for upcoming
    bpf_local_storage conversion to bpf_mem_cache_{alloc,free} allocator,
    from Martin KaFai Lau.

19) Improve bpftool's visual program dump which produces the control
    flow graph in a DOT format by adding C source inline annotations,
    from Quentin Monnet.

20) Fix attaching fentry/fexit/fmod_ret/lsm to modules by extracting
    the module name from BTF of the target and searching kallsyms of
    the correct module, from Viktor Malik.

21) Improve BPF verifier handling of '<const> <cond> <non_const>'
    to better detect whether in particular jmp32 branches are taken,
    from Yonghong Song.

22) Allow BPF TCP cong-ctls to write app_limited of struct tcp_sock.
    A built-in cc or one from a kernel module is already able to write
    to app_limited, from Yixin Shen.

Conflicts:

Documentation/bpf/bpf_devel_QA.rst
  b7abcd9c656b ("bpf, doc: Link to submitting-patches.rst for general patch submission info")
  0f10f647f455 ("bpf, docs: Use internal linking for link to netdev subsystem doc")
https://lore.kernel.org/all/20230307095812.236eb1be@canb.auug.org.au/

include/net/ip_tunnels.h
  bc9d003dc48c3 ("ip_tunnel: Preserve pointer const in ip_tunnel_info_opts")
  ac931d4cdec3d ("ipip,ip_tunnel,sit: Add FOU support for externally controlled ipip devices")
https://lore.kernel.org/all/20230413161235.4093777-1-broonie@kernel.org/

net/bpf/test_run.c
  e5995bc7e2ba ("bpf, test_run: fix crashes due to XDP frame overwriting/corruption")
  294635a8165a ("bpf, test_run: fix &xdp_frame misplacement for LIVE_FRAMES")
https://lore.kernel.org/all/20230320102619.05b80a98@canb.auug.org.au/
====================

Link: https://lore.kernel.org/r/20230413191525.7295-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 13 Apr 2023 23:04:28 +0000 (16:04 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Conflicts:

tools/testing/selftests/net/config
  62199e3f1658 ("selftests: net: Add VXLAN MDB test")
  3a0385be133e ("selftests: add the missing CONFIG_IP_SCTP in net config")

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoMerge tag 'net-6.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 13 Apr 2023 22:33:04 +0000 (15:33 -0700)]
Merge tag 'net-6.3-rc7' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Jakub Kicinski:
 "Including fixes from bpf, and bluetooth.

  Not all that quiet given spring celebrations, but "current" fixes are
  thinning out, which is encouraging. One outstanding regression in the
  mlx5 driver when using old FW, not blocking but we're pushing for a
  fix.

  Current release - new code bugs:

   - eth: enetc: workaround for unresponsive pMAC after receiving
     express traffic

  Previous releases - regressions:

   - rtnetlink: restore RTM_NEW/DELLINK notification behavior, keep the
     pid/seq fields 0 for backward compatibility

  Previous releases - always broken:

   - sctp: fix a potential overflow in sctp_ifwdtsn_skip

   - mptcp:
      - use mptcp_schedule_work instead of open-coding it and make the
        worker check stricter, to avoid scheduling work on closed
        sockets
      - fix NULL pointer dereference on fastopen early fallback

   - skbuff: fix memory corruption due to a race between skb coalescing
     and releasing clones confusing page_pool reference counting

   - bonding: fix neighbor solicitation validation on backup slaves

   - bpf: tcp: use sock_gen_put instead of sock_put in bpf_iter_tcp

   - bpf: arm64: fixed a BTI error on returning to patched function

   - openvswitch: fix race on port output leading to inf loop

   - sfp: initialize sfp->i2c_block_size at sfp allocation to avoid
     returning a different errno than expected

   - phy: nxp-c45-tja11xx: unregister PTP, purge queues on remove

   - Bluetooth: fix printing errors if LE Connection times out

   - Bluetooth: assorted UaF, deadlock and data race fixes

   - eth: macb: fix memory corruption in extended buffer descriptor mode

  Misc:

   - adjust the XDP Rx flow hash API to also include the protocol layers
     over which the hash was computed"

* tag 'net-6.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (50 commits)
  selftests/bpf: Adjust bpf_xdp_metadata_rx_hash for new arg
  mlx4: bpf_xdp_metadata_rx_hash add xdp rss hash type
  veth: bpf_xdp_metadata_rx_hash add xdp rss hash type
  mlx5: bpf_xdp_metadata_rx_hash add xdp rss hash type
  xdp: rss hash types representation
  selftests/bpf: xdp_hw_metadata remove bpf_printk and add counters
  skbuff: Fix a race between coalescing and releasing SKBs
  net: macb: fix a memory corruption in extended buffer descriptor mode
  selftests: add the missing CONFIG_IP_SCTP in net config
  udp6: fix potential access to stale information
  selftests: openvswitch: adjust datapath NL message declaration
  selftests: mptcp: userspace pm: uniform verify events
  mptcp: fix NULL pointer dereference on fastopen early fallback
  mptcp: stricter state check in mptcp_worker
  mptcp: use mptcp_schedule_work instead of open-coding it
  net: enetc: workaround for unresponsive pMAC after receiving express traffic
  sctp: fix a potential overflow in sctp_ifwdtsn_skip
  net: qrtr: Fix an uninit variable access bug in qrtr_tx_resume()
  rtnetlink: Restore RTM_NEW/DELLINK notification behavior
  net: ti/cpsw: Add explicit platform_device.h and of_platform.h includes
  ...

20 months agoMerge tag 'devicetree-fixes-for-6.2-3' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 13 Apr 2023 22:21:56 +0000 (15:21 -0700)]
Merge tag 'devicetree-fixes-for-6.2-3' of git://git./linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:

 - Fix interaction between fw_devlink and DT overlays causing devices to
   not be probed

 - Fix the compatible string for loongson,cpu-interrupt-controller

* tag 'devicetree-fixes-for-6.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  treewide: Fix probing of devices in DT overlays
  dt-bindings: interrupt-controller: loongarch: Fix mismatched compatible

20 months agoMerge tag 'pinctrl-v6.3-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Thu, 13 Apr 2023 22:17:59 +0000 (15:17 -0700)]
Merge tag 'pinctrl-v6.3-3' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fix from Linus Walleij:
 "This is just a revert of the AMD fix, because the fix broke some
  laptops. We are working on a proper solution"

* tag 'pinctrl-v6.3-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  Revert "pinctrl: amd: Disable and mask interrupts on resume"

20 months agoMerge tag 'drm-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Thu, 13 Apr 2023 21:58:55 +0000 (14:58 -0700)]
Merge tag 'drm-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Daniel Vetter:

 - two fbcon regressions

 - amdgpu: dp mst, smu13

 - i915: dual link dsi for tgl+

 - armada, nouveau, drm/sched, fbmem

* tag 'drm-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm:
  fbcon: set_con2fb_map needs to set con2fb_map!
  fbcon: Fix error paths in set_con2fb_map
  drm/amd/pm: correct the pcie link state check for SMU13
  drm/amd/pm: correct SMU13.0.7 max shader clock reporting
  drm/amd/pm: correct SMU13.0.7 pstate profiling clock settings
  drm/amd/display: Pass the right info to drm_dp_remove_payload
  drm/armada: Fix a potential double free in an error handling path
  fbmem: Reject FB_ACTIVATE_KD_TEXT from userspace
  drm/nouveau/fb: add missing sysmen flush callbacks
  drm/i915/dsi: fix DSS CTL register offsets for TGL+
  drm/scheduler: Fix UAF race in drm_sched_entity_push_job()

20 months agoMerge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
Jakub Kicinski [Thu, 13 Apr 2023 20:04:44 +0000 (13:04 -0700)]
Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2023-04-13

We've added 6 non-merge commits during the last 1 day(s) which contain
a total of 14 files changed, 205 insertions(+), 38 deletions(-).

The main changes are:

1) One late straggler fix on the XDP hints side which fixes
   bpf_xdp_metadata_rx_hash kfunc API before the release goes out
   in order to provide information on the RSS hash type,
   from Jesper Dangaard Brouer.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
  selftests/bpf: Adjust bpf_xdp_metadata_rx_hash for new arg
  mlx4: bpf_xdp_metadata_rx_hash add xdp rss hash type
  veth: bpf_xdp_metadata_rx_hash add xdp rss hash type
  mlx5: bpf_xdp_metadata_rx_hash add xdp rss hash type
  xdp: rss hash types representation
  selftests/bpf: xdp_hw_metadata remove bpf_printk and add counters
====================

Link: https://lore.kernel.org/r/20230413192939.10202-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoMerge tag 'drm-misc-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm-misc...
Daniel Vetter [Thu, 13 Apr 2023 18:47:58 +0000 (20:47 +0200)]
Merge tag 'drm-misc-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

Short summary of fixes pull:

 * armada: Fix double free
 * fb: Clear FB_ACTIVATE_KD_TEXT in ioctl
 * nouveau: Add missing callbacks
 * scheduler: Fix use-after-free error

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20230413184233.GA8148@linux-uq9g
20 months agobpf, sockmap: Revert buggy deadlock fix in the sockhash and sockmap
Daniel Borkmann [Thu, 13 Apr 2023 18:28:42 +0000 (20:28 +0200)]
bpf, sockmap: Revert buggy deadlock fix in the sockhash and sockmap

syzbot reported a splat and bisected it to recent commit ed17aa92dc56 ("bpf,
sockmap: fix deadlocks in the sockhash and sockmap"):

  [...]
  WARNING: CPU: 1 PID: 9280 at kernel/softirq.c:376 __local_bh_enable_ip+0xbe/0x130 kernel/softirq.c:376
  Modules linked in:
  CPU: 1 PID: 9280 Comm: syz-executor.1 Not tainted 6.2.0-syzkaller-13249-gd319f344561d #0
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/30/2023
  RIP: 0010:__local_bh_enable_ip+0xbe/0x130 kernel/softirq.c:376
  [...]
  Call Trace:
  <TASK>
  spin_unlock_bh include/linux/spinlock.h:395 [inline]
  sock_map_del_link+0x2ea/0x510 net/core/sock_map.c:165
  sock_map_unref+0xb0/0x1d0 net/core/sock_map.c:184
  sock_hash_delete_elem+0x1ec/0x2a0 net/core/sock_map.c:945
  map_delete_elem kernel/bpf/syscall.c:1536 [inline]
  __sys_bpf+0x2edc/0x53e0 kernel/bpf/syscall.c:5053
  __do_sys_bpf kernel/bpf/syscall.c:5166 [inline]
  __se_sys_bpf kernel/bpf/syscall.c:5164 [inline]
  __x64_sys_bpf+0x79/0xc0 kernel/bpf/syscall.c:5164
  do_syscall_x64 arch/x86/entry/common.c:50 [inline]
  do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
  entry_SYSCALL_64_after_hwframe+0x63/0xcd
  RIP: 0033:0x7fe8f7c8c169
  </TASK>
  [...]

Revert for now until we have a proper solution.

Fixes: ed17aa92dc56 ("bpf, sockmap: fix deadlocks in the sockhash and sockmap")
Reported-by: syzbot+49f6cef45247ff249498@syzkaller.appspotmail.com
Cc: Hsin-Wei Hung <hsinweih@uci.edu>
Cc: Xin Liu <liuxin350@huawei.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/000000000000f1db9605f939720e@google.com/
20 months agoMerge branch 'XDP-hints: change RX-hash kfunc bpf_xdp_metadata_rx_hash'
Alexei Starovoitov [Thu, 13 Apr 2023 18:05:49 +0000 (11:05 -0700)]
Merge branch 'XDP-hints: change RX-hash kfunc bpf_xdp_metadata_rx_hash'

Jesper Dangaard Brouer says:

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

Current API for bpf_xdp_metadata_rx_hash() returns the raw RSS hash value,
but doesn't provide information on the RSS hash type (part of 6.3-rc).

This patchset proposal is to change the function call signature via adding
a pointer value argument for providing the RSS hash type.

Patchset also removes all bpf_printk's from xdp_hw_metadata program
that we expect driver developers to use. Instead counters are introduced
for relaying e.g. skip and fail info.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: Adjust bpf_xdp_metadata_rx_hash for new arg
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:49:00 +0000 (21:49 +0200)]
selftests/bpf: Adjust bpf_xdp_metadata_rx_hash for new arg

Update BPF selftests to use the new RSS type argument for kfunc
bpf_xdp_metadata_rx_hash.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132894068.340624.8914711185697163690.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agomlx4: bpf_xdp_metadata_rx_hash add xdp rss hash type
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:48:55 +0000 (21:48 +0200)]
mlx4: bpf_xdp_metadata_rx_hash add xdp rss hash type

Update API for bpf_xdp_metadata_rx_hash() with arg for xdp rss hash type
via matching individual Completion Queue Entry (CQE) status bits.

Fixes: ab46182d0dcb ("net/mlx4_en: Support RX XDP metadata")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132893562.340624.12779118462402031248.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoveth: bpf_xdp_metadata_rx_hash add xdp rss hash type
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:48:50 +0000 (21:48 +0200)]
veth: bpf_xdp_metadata_rx_hash add xdp rss hash type

Update API for bpf_xdp_metadata_rx_hash() with arg for xdp rss hash type.

The veth driver currently only support XDP-hints based on SKB code path.
The SKB have lost information about the RSS hash type, by compressing
the information down to a single bitfield skb->l4_hash, that only knows
if this was a L4 hash value.

In preparation for veth, the xdp_rss_hash_type have an L4 indication
bit that allow us to return a meaningful L4 indication when working
with SKB based packets.

Fixes: 306531f0249f ("veth: Support RX XDP metadata")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132893055.340624.16209448340644513469.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agomlx5: bpf_xdp_metadata_rx_hash add xdp rss hash type
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:48:45 +0000 (21:48 +0200)]
mlx5: bpf_xdp_metadata_rx_hash add xdp rss hash type

Update API for bpf_xdp_metadata_rx_hash() with arg for xdp rss hash type
via mapping table.

The mlx5 hardware can also identify and RSS hash IPSEC.  This indicate
hash includes SPI (Security Parameters Index) as part of IPSEC hash.

Extend xdp core enum xdp_rss_hash_type with IPSEC hash type.

Fixes: bc8d405b1ba9 ("net/mlx5e: Support RX XDP metadata")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132892548.340624.11185734579430124869.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoxdp: rss hash types representation
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:48:40 +0000 (21:48 +0200)]
xdp: rss hash types representation

The RSS hash type specifies what portion of packet data NIC hardware used
when calculating RSS hash value. The RSS types are focused on Internet
traffic protocols at OSI layers L3 and L4. L2 (e.g. ARP) often get hash
value zero and no RSS type. For L3 focused on IPv4 vs. IPv6, and L4
primarily TCP vs UDP, but some hardware supports SCTP.

Hardware RSS types are differently encoded for each hardware NIC. Most
hardware represent RSS hash type as a number. Determining L3 vs L4 often
requires a mapping table as there often isn't a pattern or sorting
according to ISO layer.

The patch introduce a XDP RSS hash type (enum xdp_rss_hash_type) that
contains both BITs for the L3/L4 types, and combinations to be used by
drivers for their mapping tables. The enum xdp_rss_type_bits get exposed
to BPF via BTF, and it is up to the BPF-programmer to match using these
defines.

This proposal change the kfunc API bpf_xdp_metadata_rx_hash() adding
a pointer value argument for provide the RSS hash type.
Change signature for all xmo_rx_hash calls in drivers to make it compile.

The RSS type implementations for each driver comes as separate patches.

Fixes: 3d76a4d3d4e5 ("bpf: XDP metadata RX kfuncs")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132892042.340624.582563003880565460.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agoselftests/bpf: xdp_hw_metadata remove bpf_printk and add counters
Jesper Dangaard Brouer [Wed, 12 Apr 2023 19:48:35 +0000 (21:48 +0200)]
selftests/bpf: xdp_hw_metadata remove bpf_printk and add counters

The tool xdp_hw_metadata can be used by driver developers
implementing XDP-hints metadata kfuncs.

Remove all bpf_printk calls, as the tool already transfers all the
XDP-hints related information via metadata area to AF_XDP
userspace process.

Add counters for providing remaining information about failure and
skipped packet events.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/168132891533.340624.7313781245316405141.stgit@firesoul
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
20 months agofbcon: set_con2fb_map needs to set con2fb_map!
Daniel Vetter [Wed, 12 Apr 2023 15:31:46 +0000 (17:31 +0200)]
fbcon: set_con2fb_map needs to set con2fb_map!

I got really badly confused in d443d9386472 ("fbcon: move more common
code into fb_open()") because we set the con2fb_map before the failure
points, which didn't look good.

But in trying to fix that I moved the assignment into the wrong path -
we need to do it for _all_ vc we take over, not just the first one
(which additionally requires the call to con2fb_acquire_newinfo).

I've figured this out because of a KASAN bug report, where the
fbcon_registered_fb and fbcon_display arrays went out of sync in
fbcon_mode_deleted() because the con2fb_map pointed at the old
fb_info, but the modes and everything was updated for the new one.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Helge Deller <deller@gmx.de>
Tested-by: Xingyuan Mo <hdthky0@gmail.com>
Fixes: d443d9386472 ("fbcon: move more common code into fb_open()")
Reported-by: Xingyuan Mo <hdthky0@gmail.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Xingyuan Mo <hdthky0@gmail.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Helge Deller <deller@gmx.de>
Cc: <stable@vger.kernel.org> # v5.19+
20 months agofbcon: Fix error paths in set_con2fb_map
Daniel Vetter [Wed, 12 Apr 2023 15:23:49 +0000 (17:23 +0200)]
fbcon: Fix error paths in set_con2fb_map

This is a regressoin introduced in b07db3958485 ("fbcon: Ditch error
handling for con2fb_release_oldinfo"). I failed to realize what the if
(!err) checks. The mentioned commit was dropping the
con2fb_release_oldinfo() return value but the if (!err) was also
checking whether the con2fb_acquire_newinfo() function call above
failed or not.

Fix this with an early return statement.

Note that there's still a difference compared to the orginal state of
the code, the below lines are now also skipped on error:

if (!search_fb_in_map(info_idx))
info_idx = newidx;

These are only needed when we've actually thrown out an old fb_info
from the console mappings, which only happens later on.

Also move the fbcon_add_cursor_work() call into the same if block,
it's all protected by console_lock so doesn't matter when we set up
the blinking cursor delayed work anyway. This further simplifies the
control flow and allows us to ditch the found local variable.

v2: Clarify commit message (Javier)

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Helge Deller <deller@gmx.de>
Tested-by: Xingyuan Mo <hdthky0@gmail.com>
Fixes: b07db3958485 ("fbcon: Ditch error handling for con2fb_release_oldinfo")
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Xingyuan Mo <hdthky0@gmail.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Helge Deller <deller@gmx.de>
Cc: <stable@vger.kernel.org> # v5.19+
20 months agoskbuff: Fix a race between coalescing and releasing SKBs
Liang Chen [Thu, 13 Apr 2023 09:03:53 +0000 (17:03 +0800)]
skbuff: Fix a race between coalescing and releasing SKBs

Commit 1effe8ca4e34 ("skbuff: fix coalescing for page_pool fragment
recycling") allowed coalescing to proceed with non page pool page and page
pool page when @from is cloned, i.e.

to->pp_recycle    --> false
from->pp_recycle  --> true
skb_cloned(from)  --> true

However, it actually requires skb_cloned(@from) to hold true until
coalescing finishes in this situation. If the other cloned SKB is
released while the merging is in process, from_shinfo->nr_frags will be
set to 0 toward the end of the function, causing the increment of frag
page _refcount to be unexpectedly skipped resulting in inconsistent
reference counts. Later when SKB(@to) is released, it frees the page
directly even though the page pool page is still in use, leading to
use-after-free or double-free errors. So it should be prohibited.

The double-free error message below prompted us to investigate:
BUG: Bad page state in process swapper/1  pfn:0e0d1
page:00000000c6548b28 refcount:-1 mapcount:0 mapping:0000000000000000
index:0x2 pfn:0xe0d1
flags: 0xfffffc0000000(node=0|zone=1|lastcpupid=0x1fffff)
raw: 000fffffc0000000 0000000000000000 ffffffff00000101 0000000000000000
raw: 0000000000000002 0000000000000000 ffffffffffffffff 0000000000000000
page dumped because: nonzero _refcount

CPU: 1 PID: 0 Comm: swapper/1 Tainted: G            E      6.2.0+
Call Trace:
 <IRQ>
dump_stack_lvl+0x32/0x50
bad_page+0x69/0xf0
free_pcp_prepare+0x260/0x2f0
free_unref_page+0x20/0x1c0
skb_release_data+0x10b/0x1a0
napi_consume_skb+0x56/0x150
net_rx_action+0xf0/0x350
? __napi_schedule+0x79/0x90
__do_softirq+0xc8/0x2b1
__irq_exit_rcu+0xb9/0xf0
common_interrupt+0x82/0xa0
</IRQ>
<TASK>
asm_common_interrupt+0x22/0x40
RIP: 0010:default_idle+0xb/0x20

Fixes: 53e0961da1c7 ("page_pool: add frag page recycling support in page pool")
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230413090353.14448-1-liangchen.linux@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonet: macb: fix a memory corruption in extended buffer descriptor mode
Roman Gushchin [Wed, 12 Apr 2023 23:21:44 +0000 (16:21 -0700)]
net: macb: fix a memory corruption in extended buffer descriptor mode

For quite some time we were chasing a bug which looked like a sudden
permanent failure of networking and mmc on some of our devices.
The bug was very sensitive to any software changes and even more to
any kernel debug options.

Finally we got a setup where the problem was reproducible with
CONFIG_DMA_API_DEBUG=y and it revealed the issue with the rx dma:

[   16.992082] ------------[ cut here ]------------
[   16.996779] DMA-API: macb ff0b0000.ethernet: device driver tries to free DMA memory it has not allocated [device address=0x0000000875e3e244] [size=1536 bytes]
[   17.011049] WARNING: CPU: 0 PID: 85 at kernel/dma/debug.c:1011 check_unmap+0x6a0/0x900
[   17.018977] Modules linked in: xxxxx
[   17.038823] CPU: 0 PID: 85 Comm: irq/55-8000f000 Not tainted 5.4.0 #28
[   17.045345] Hardware name: xxxxx
[   17.049528] pstate: 60000005 (nZCv daif -PAN -UAO)
[   17.054322] pc : check_unmap+0x6a0/0x900
[   17.058243] lr : check_unmap+0x6a0/0x900
[   17.062163] sp : ffffffc010003c40
[   17.065470] x29: ffffffc010003c40 x28: 000000004000c03c
[   17.070783] x27: ffffffc010da7048 x26: ffffff8878e38800
[   17.076095] x25: ffffff8879d22810 x24: ffffffc010003cc8
[   17.081407] x23: 0000000000000000 x22: ffffffc010a08750
[   17.086719] x21: ffffff8878e3c7c0 x20: ffffffc010acb000
[   17.092032] x19: 0000000875e3e244 x18: 0000000000000010
[   17.097343] x17: 0000000000000000 x16: 0000000000000000
[   17.102647] x15: ffffff8879e4a988 x14: 0720072007200720
[   17.107959] x13: 0720072007200720 x12: 0720072007200720
[   17.113261] x11: 0720072007200720 x10: 0720072007200720
[   17.118565] x9 : 0720072007200720 x8 : 000000000000022d
[   17.123869] x7 : 0000000000000015 x6 : 0000000000000098
[   17.129173] x5 : 0000000000000000 x4 : 0000000000000000
[   17.134475] x3 : 00000000ffffffff x2 : ffffffc010a1d370
[   17.139778] x1 : b420c9d75d27bb00 x0 : 0000000000000000
[   17.145082] Call trace:
[   17.147524]  check_unmap+0x6a0/0x900
[   17.151091]  debug_dma_unmap_page+0x88/0x90
[   17.155266]  gem_rx+0x114/0x2f0
[   17.158396]  macb_poll+0x58/0x100
[   17.161705]  net_rx_action+0x118/0x400
[   17.165445]  __do_softirq+0x138/0x36c
[   17.169100]  irq_exit+0x98/0xc0
[   17.172234]  __handle_domain_irq+0x64/0xc0
[   17.176320]  gic_handle_irq+0x5c/0xc0
[   17.179974]  el1_irq+0xb8/0x140
[   17.183109]  xiic_process+0x5c/0xe30
[   17.186677]  irq_thread_fn+0x28/0x90
[   17.190244]  irq_thread+0x208/0x2a0
[   17.193724]  kthread+0x130/0x140
[   17.196945]  ret_from_fork+0x10/0x20
[   17.200510] ---[ end trace 7240980785f81d6f ]---

[  237.021490] ------------[ cut here ]------------
[  237.026129] DMA-API: exceeded 7 overlapping mappings of cacheline 0x0000000021d79e7b
[  237.033886] WARNING: CPU: 0 PID: 0 at kernel/dma/debug.c:499 add_dma_entry+0x214/0x240
[  237.041802] Modules linked in: xxxxx
[  237.061637] CPU: 0 PID: 0 Comm: swapper/0 Tainted: G        W         5.4.0 #28
[  237.068941] Hardware name: xxxxx
[  237.073116] pstate: 80000085 (Nzcv daIf -PAN -UAO)
[  237.077900] pc : add_dma_entry+0x214/0x240
[  237.081986] lr : add_dma_entry+0x214/0x240
[  237.086072] sp : ffffffc010003c30
[  237.089379] x29: ffffffc010003c30 x28: ffffff8878a0be00
[  237.094683] x27: 0000000000000180 x26: ffffff8878e387c0
[  237.099987] x25: 0000000000000002 x24: 0000000000000000
[  237.105290] x23: 000000000000003b x22: ffffffc010a0fa00
[  237.110594] x21: 0000000021d79e7b x20: ffffffc010abe600
[  237.115897] x19: 00000000ffffffef x18: 0000000000000010
[  237.121201] x17: 0000000000000000 x16: 0000000000000000
[  237.126504] x15: ffffffc010a0fdc8 x14: 0720072007200720
[  237.131807] x13: 0720072007200720 x12: 0720072007200720
[  237.137111] x11: 0720072007200720 x10: 0720072007200720
[  237.142415] x9 : 0720072007200720 x8 : 0000000000000259
[  237.147718] x7 : 0000000000000001 x6 : 0000000000000000
[  237.153022] x5 : ffffffc010003a20 x4 : 0000000000000001
[  237.158325] x3 : 0000000000000006 x2 : 0000000000000007
[  237.163628] x1 : 8ac721b3a7dc1c00 x0 : 0000000000000000
[  237.168932] Call trace:
[  237.171373]  add_dma_entry+0x214/0x240
[  237.175115]  debug_dma_map_page+0xf8/0x120
[  237.179203]  gem_rx_refill+0x190/0x280
[  237.182942]  gem_rx+0x224/0x2f0
[  237.186075]  macb_poll+0x58/0x100
[  237.189384]  net_rx_action+0x118/0x400
[  237.193125]  __do_softirq+0x138/0x36c
[  237.196780]  irq_exit+0x98/0xc0
[  237.199914]  __handle_domain_irq+0x64/0xc0
[  237.204000]  gic_handle_irq+0x5c/0xc0
[  237.207654]  el1_irq+0xb8/0x140
[  237.210789]  arch_cpu_idle+0x40/0x200
[  237.214444]  default_idle_call+0x18/0x30
[  237.218359]  do_idle+0x200/0x280
[  237.221578]  cpu_startup_entry+0x20/0x30
[  237.225493]  rest_init+0xe4/0xf0
[  237.228713]  arch_call_rest_init+0xc/0x14
[  237.232714]  start_kernel+0x47c/0x4a8
[  237.236367] ---[ end trace 7240980785f81d70 ]---

Lars was fast to find an explanation: according to the datasheet
bit 2 of the rx buffer descriptor entry has a different meaning in the
extended mode:
  Address [2] of beginning of buffer, or
  in extended buffer descriptor mode (DMA configuration register [28] = 1),
  indicates a valid timestamp in the buffer descriptor entry.

The macb driver didn't mask this bit while getting an address and it
eventually caused a memory corruption and a dma failure.

The problem is resolved by explicitly clearing the problematic bit
if hw timestamping is used.

Fixes: 7b4296148066 ("net: macb: Add support for PTP timestamps in DMA descriptors")
Signed-off-by: Roman Gushchin <roman.gushchin@linux.dev>
Co-developed-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230412232144.770336-1-roman.gushchin@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoselftests: add the missing CONFIG_IP_SCTP in net config
Xin Long [Wed, 12 Apr 2023 15:13:06 +0000 (11:13 -0400)]
selftests: add the missing CONFIG_IP_SCTP in net config

The selftest sctp_vrf needs CONFIG_IP_SCTP set in config
when building the kernel, so add it.

Fixes: a61bd7b9fef3 ("selftests: add a selftest for sctp vrf")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Link: https://lore.kernel.org/r/61dddebc4d2dd98fe7fb145e24d4b2430e42b572.1681312386.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoudp6: fix potential access to stale information
Eric Dumazet [Wed, 12 Apr 2023 13:03:08 +0000 (13:03 +0000)]
udp6: fix potential access to stale information

lena wang reported an issue caused by udpv6_sendmsg()
mangling msg->msg_name and msg->msg_namelen, which
are later read from ____sys_sendmsg() :

/*
 * If this is sendmmsg() and sending to current destination address was
 * successful, remember it.
 */
if (used_address && err >= 0) {
used_address->name_len = msg_sys->msg_namelen;
if (msg_sys->msg_name)
memcpy(&used_address->name, msg_sys->msg_name,
       used_address->name_len);
}

udpv6_sendmsg() wants to pretend the remote address family
is AF_INET in order to call udp_sendmsg().

A fix would be to modify the address in-place, instead
of using a local variable, but this could have other side effects.

Instead, restore initial values before we return from udpv6_sendmsg().

Fixes: c71d8ebe7a44 ("net: Fix security_socket_sendmsg() bypass problem.")
Reported-by: lena wang <lena.wang@mediatek.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20230412130308.1202254-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoselftests: openvswitch: adjust datapath NL message declaration
Aaron Conole [Wed, 12 Apr 2023 11:58:28 +0000 (07:58 -0400)]
selftests: openvswitch: adjust datapath NL message declaration

The netlink message for creating a new datapath takes an array
of ports for the PID creation.  This shouldn't cause much issue
but correct it for future cases where we need to do decode of
datapath information that could include the per-cpu PID map.

Fixes: 25f16c873fb1 ("selftests: add openvswitch selftest suite")
Signed-off-by: Aaron Conole <aconole@redhat.com>
Link: https://lore.kernel.org/r/20230412115828.3991806-1-aconole@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoMerge branch 'mptcp-more-fixes-for-6-3'
Jakub Kicinski [Thu, 13 Apr 2023 16:59:00 +0000 (09:59 -0700)]
Merge branch 'mptcp-more-fixes-for-6-3'

Matthieu Baerts says:

====================
mptcp: more fixes for 6.3

Patch 1 avoids scheduling the MPTCP worker on a closed socket on some
edge cases. It fixes issues that can be visible from v5.11.

Patch 2 makes sure the MPTCP worker doesn't try to manipulate
disconnected sockets. This is also a fix for an issue that can be
visible from v5.11.

Patch 3 fixes a NULL pointer dereference when MPTCP FastOpen is used
and an early fallback is done. A fix for v6.2.

Patch 4 improves the stability of the userspace PM selftest for a
subtest added in v6.2.
====================

Link: https://lore.kernel.org/r/20230411-upstream-net-20230411-mptcp-fixes-v1-0-ca540f3ef986@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoselftests: mptcp: userspace pm: uniform verify events
Matthieu Baerts [Tue, 11 Apr 2023 20:42:12 +0000 (22:42 +0200)]
selftests: mptcp: userspace pm: uniform verify events

Simply adding a "sleep" before checking something is usually not a good
idea because the time that has been picked can not be enough or too
much. The best is to wait for events with a timeout.

In this selftest, 'sleep 0.5' is used more than 40 times. It is always
used before calling a 'verify_*' function except for this
verify_listener_events which has been added later.

At the end, using all these 'sleep 0.5' seems to work: the slow CIs
don't complain so far. Also because it doesn't take too much time, we
can just add two more 'sleep 0.5' to uniform what is done before calling
a 'verify_*' function. For the same reasons, we can also delay a bigger
refactoring to replace all these 'sleep 0.5' by functions waiting for
events instead of waiting for a fix time and hope for the best.

Fixes: 6c73008aa301 ("selftests: mptcp: listener test for userspace PM")
Cc: stable@vger.kernel.org
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agomptcp: fix NULL pointer dereference on fastopen early fallback
Paolo Abeni [Tue, 11 Apr 2023 20:42:11 +0000 (22:42 +0200)]
mptcp: fix NULL pointer dereference on fastopen early fallback

In case of early fallback to TCP, subflow_syn_recv_sock() deletes
the subflow context before returning the newly allocated sock to
the caller.

The fastopen path does not cope with the above unconditionally
dereferencing the subflow context.

Fixes: 36b122baf6a8 ("mptcp: add subflow_v(4,6)_send_synack()")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agomptcp: stricter state check in mptcp_worker
Paolo Abeni [Tue, 11 Apr 2023 20:42:10 +0000 (22:42 +0200)]
mptcp: stricter state check in mptcp_worker

As reported by Christoph, the mptcp protocol can run the
worker when the relevant msk socket is in an unexpected state:

connect()
// incoming reset + fastclose
// the mptcp worker is scheduled
mptcp_disconnect()
// msk is now CLOSED
listen()
mptcp_worker()

Leading to the following splat:

divide error: 0000 [#1] PREEMPT SMP
CPU: 1 PID: 21 Comm: kworker/1:0 Not tainted 6.3.0-rc1-gde5e8fd0123c #11
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-2.el7 04/01/2014
Workqueue: events mptcp_worker
RIP: 0010:__tcp_select_window+0x22c/0x4b0 net/ipv4/tcp_output.c:3018
RSP: 0018:ffffc900000b3c98 EFLAGS: 00010293
RAX: 000000000000ffd7 RBX: 000000000000ffd7 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffffffff8214ce97 RDI: 0000000000000004
RBP: 000000000000ffd7 R08: 0000000000000004 R09: 0000000000010000
R10: 000000000000ffd7 R11: ffff888005afa148 R12: 000000000000ffd7
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffff88803ed00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000405270 CR3: 000000003011e006 CR4: 0000000000370ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 tcp_select_window net/ipv4/tcp_output.c:262 [inline]
 __tcp_transmit_skb+0x356/0x1280 net/ipv4/tcp_output.c:1345
 tcp_transmit_skb net/ipv4/tcp_output.c:1417 [inline]
 tcp_send_active_reset+0x13e/0x320 net/ipv4/tcp_output.c:3459
 mptcp_check_fastclose net/mptcp/protocol.c:2530 [inline]
 mptcp_worker+0x6c7/0x800 net/mptcp/protocol.c:2705
 process_one_work+0x3bd/0x950 kernel/workqueue.c:2390
 worker_thread+0x5b/0x610 kernel/workqueue.c:2537
 kthread+0x138/0x170 kernel/kthread.c:376
 ret_from_fork+0x2c/0x50 arch/x86/entry/entry_64.S:308
 </TASK>

This change addresses the issue explicitly checking for bad states
before running the mptcp worker.

Fixes: e16163b6e2b7 ("mptcp: refactor shutdown and close")
Cc: stable@vger.kernel.org
Reported-by: Christoph Paasch <cpaasch@apple.com>
Link: https://github.com/multipath-tcp/mptcp_net-next/issues/374
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agomptcp: use mptcp_schedule_work instead of open-coding it
Paolo Abeni [Tue, 11 Apr 2023 20:42:09 +0000 (22:42 +0200)]
mptcp: use mptcp_schedule_work instead of open-coding it

Beyond reducing code duplication this also avoids scheduling
the mptcp_worker on a closed socket on some edge scenarios.

The addressed issue is actually older than the blamed commit
below, but this fix needs it as a pre-requisite.

Fixes: ba8f48f7a4d7 ("mptcp: introduce mptcp_schedule_work")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agosamples/bpf: Fix fout leak in hbm's run_bpf_prog
Hao Zeng [Tue, 11 Apr 2023 08:43:49 +0000 (16:43 +0800)]
samples/bpf: Fix fout leak in hbm's run_bpf_prog

Fix fout being fopen'ed but then not subsequently fclose'd. In the affected
branch, fout is otherwise going out of scope.

Signed-off-by: Hao Zeng <zenghao@kylinos.cn>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230411084349.1999628-1-zenghao@kylinos.cn
20 months agoxsk: Elide base_addr comparison in xp_unaligned_validate_desc
Kal Conley [Tue, 11 Apr 2023 13:00:25 +0000 (15:00 +0200)]
xsk: Elide base_addr comparison in xp_unaligned_validate_desc

Remove redundant (base_addr >= pool->addrs_cnt) comparison from the
conditional.

In particular, addr is computed as:

    addr = base_addr + offset

... where base_addr and offset are stored as 48-bit and 16-bit unsigned
integers, respectively. The above sum cannot overflow u64 since base_addr
has a maximum value of 0x0000ffffffffffff and offset has a maximum value
of 0xffff (implying a maximum sum of 0x000100000000fffe). Since overflow
is impossible, it follows that addr >= base_addr.

Now if (base_addr >= pool->addrs_cnt), then clearly:

    addr >= base_addr
         >= pool->addrs_cnt

Thus, (base_addr >= pool->addrs_cnt) implies (addr >= pool->addrs_cnt).
Subsequently, the former comparison is unnecessary in the conditional
since for any boolean expressions A and B, (A || B) && (A -> B) is
equivalent to B.

Signed-off-by: Kal Conley <kal.conley@dectris.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20230411130025.19704-1-kal.conley@dectris.com
20 months agoselftests/bpf: Fix compiler warnings in bpf_testmod for kfuncs
Andrii Nakryiko [Wed, 12 Apr 2023 03:46:47 +0000 (20:46 -0700)]
selftests/bpf: Fix compiler warnings in bpf_testmod for kfuncs

Add -Wmissing-prototypes ignore in bpf_testmod.c, similarly to what we
do in kernel code proper.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/oe-kbuild-all/202304080951.l14IDv3n-lkp@intel.com/
Link: https://lore.kernel.org/bpf/20230412034647.3968143-1-andrii@kernel.org
20 months agoxsk: Simplify xp_aligned_validate_desc implementation
Kal Conley [Mon, 10 Apr 2023 12:18:41 +0000 (14:18 +0200)]
xsk: Simplify xp_aligned_validate_desc implementation

Perform the chunk boundary check like the page boundary check in
xp_desc_crosses_non_contig_pg(). This simplifies the implementation and
reduces the number of branches.

Signed-off-by: Kal Conley <kal.conley@dectris.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20230410121841.643254-1-kal.conley@dectris.com
20 months agoselftests/bpf: Remove stand-along test_verifier_log test binary
Andrii Nakryiko [Wed, 12 Apr 2023 17:06:55 +0000 (10:06 -0700)]
selftests/bpf: Remove stand-along test_verifier_log test binary

test_prog's prog_tests/verifier_log.c is superseding test_verifier_log
stand-alone test. It cover same checks and adds more, and is also
integrated into test_progs test runner.

Just remove test_verifier_log.c.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230412170655.1866831-1-andrii@kernel.org
20 months agoselftests/bpf: Keep the loop in bpf_testmod_loop_test
Song Liu [Wed, 12 Apr 2023 21:04:23 +0000 (14:04 -0700)]
selftests/bpf: Keep the loop in bpf_testmod_loop_test

Some compilers (for example clang-15) optimize bpf_testmod_loop_test and
remove the loop:

gcc version
(gdb) disassemble bpf_testmod_loop_test
Dump of assembler code for function bpf_testmod_loop_test:
   0x0000000000000570 <+0>:     callq  0x575 <bpf_testmod_loop_test+5>
   0x0000000000000575 <+5>:     xor    %eax,%eax
   0x0000000000000577 <+7>:     test   %edi,%edi
   0x0000000000000579 <+9>:     jle    0x587 <bpf_testmod_loop_test+23>
   0x000000000000057b <+11>:    xor    %edx,%edx
   0x000000000000057d <+13>:    add    %edx,%eax
   0x000000000000057f <+15>:    add    $0x1,%edx
   0x0000000000000582 <+18>:    cmp    %edx,%edi
   0x0000000000000584 <+20>:    jne    0x57d <bpf_testmod_loop_test+13>
   0x0000000000000586 <+22>:    retq
   0x0000000000000587 <+23>:    retq

clang-15 version
(gdb) disassemble bpf_testmod_loop_test
Dump of assembler code for function bpf_testmod_loop_test:
   0x0000000000000450 <+0>:     nopl   0x0(%rax,%rax,1)
   0x0000000000000455 <+5>:     test   %edi,%edi
   0x0000000000000457 <+7>:     jle    0x46b <bpf_testmod_loop_test+27>
   0x0000000000000459 <+9>:     lea    -0x1(%rdi),%eax
   0x000000000000045c <+12>:    lea    -0x2(%rdi),%ecx
   0x000000000000045f <+15>:    imul   %rax,%rcx
   0x0000000000000463 <+19>:    shr    %rcx
   0x0000000000000466 <+22>:    lea    -0x1(%rdi,%rcx,1),%eax
   0x000000000000046a <+26>:    retq
   0x000000000000046b <+27>:    xor    %eax,%eax
   0x000000000000046d <+29>:    retq

Note: The jne instruction is removed in clang-15 version.

Force the compile to keep the loop by making sum volatile.

Signed-off-by: Song Liu <song@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230412210423.900851-4-song@kernel.org
20 months agoselftests/bpf: Fix leaked bpf_link in get_stackid_cannot_attach
Song Liu [Wed, 12 Apr 2023 21:04:22 +0000 (14:04 -0700)]
selftests/bpf: Fix leaked bpf_link in get_stackid_cannot_attach

skel->links.oncpu is leaked in one case. This causes test perf_branches
fails when it runs after get_stackid_cannot_attach:

./test_progs -t get_stackid_cannot_attach,perf_branches
84      get_stackid_cannot_attach:OK
test_perf_branches_common:PASS:test_perf_branches_load 0 nsec
test_perf_branches_common:PASS:attach_perf_event 0 nsec
test_perf_branches_common:PASS:set_affinity 0 nsec
check_good_sample:FAIL:output not valid no valid sample from prog
146/1   perf_branches/perf_branches_hw:FAIL
146/2   perf_branches/perf_branches_no_hw:OK
146     perf_branches:FAIL

All error logs:
test_perf_branches_common:PASS:test_perf_branches_load 0 nsec
test_perf_branches_common:PASS:attach_perf_event 0 nsec
test_perf_branches_common:PASS:set_affinity 0 nsec
check_good_sample:FAIL:output not valid no valid sample from prog
146/1   perf_branches/perf_branches_hw:FAIL
146     perf_branches:FAIL
Summary: 1/1 PASSED, 0 SKIPPED, 1 FAILED

Fix this by adding the missing bpf_link__destroy().

Fixes: 346938e9380c ("selftests/bpf: Add get_stackid_cannot_attach")
Signed-off-by: Song Liu <song@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230412210423.900851-3-song@kernel.org
20 months agoselftests/bpf: Use read_perf_max_sample_freq() in perf_event_stackmap
Song Liu [Wed, 12 Apr 2023 21:04:21 +0000 (14:04 -0700)]
selftests/bpf: Use read_perf_max_sample_freq() in perf_event_stackmap

Currently, perf_event sample period in perf_event_stackmap is set too low
that the test fails randomly. Fix this by using the max sample frequency,
from read_perf_max_sample_freq().

Move read_perf_max_sample_freq() to testing_helpers.c. Replace the CHECK()
with if-printf, as CHECK is not available in testing_helpers.c.

Fixes: 1da4864c2b20 ("selftests/bpf: Add callchain_stackid")
Signed-off-by: Song Liu <song@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230412210423.900851-2-song@kernel.org
20 months agoMerge tag 'drm-intel-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm-intel...
Daniel Vetter [Thu, 13 Apr 2023 12:24:44 +0000 (14:24 +0200)]
Merge tag 'drm-intel-fixes-2023-04-13' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v6.3-rc7:
- Fix dual link DSI for TGL+

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/877cugckzu.fsf@intel.com
20 months agoselftests/bpf: Fix use of uninitialized op_name in log tests
Lorenz Bauer [Thu, 13 Apr 2023 09:47:40 +0000 (10:47 +0100)]
selftests/bpf: Fix use of uninitialized op_name in log tests

One of the test assertions uses an uninitialized op_name, which leads
to some headscratching if it fails. Use a string constant instead.

Fixes: b1a7a480a112 ("selftests/bpf: Add fixed vs rotating verifier log tests")
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230413094740.18041-1-lmb@isovalent.com
20 months agoMerge branch 'net-use-read_once-write_once-for-ring-index-accesses'
Paolo Abeni [Thu, 13 Apr 2023 11:30:22 +0000 (13:30 +0200)]
Merge branch 'net-use-read_once-write_once-for-ring-index-accesses'

Jakub Kicinski says:

====================
net: use READ_ONCE/WRITE_ONCE for ring index accesses

Small follow up to the lockless ring stop/start macros.
Update the doc and the drivers suggested by Eric:
https://lore.kernel.org/all/CANn89iJrBGSybMX1FqrhCEMWT3Nnz2=2+aStsbbwpWzKHjk51g@mail.gmail.com/

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

Link: https://lore.kernel.org/r/20230412015038.674023-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agomlx4: use READ_ONCE/WRITE_ONCE for ring indexes
Jakub Kicinski [Wed, 12 Apr 2023 01:50:38 +0000 (18:50 -0700)]
mlx4: use READ_ONCE/WRITE_ONCE for ring indexes

Eric points out that we should make sure that ring index updates
are wrapped in the appropriate READ_ONCE/WRITE_ONCE macros.

Suggested-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agobnxt: use READ_ONCE/WRITE_ONCE for ring indexes
Jakub Kicinski [Wed, 12 Apr 2023 01:50:37 +0000 (18:50 -0700)]
bnxt: use READ_ONCE/WRITE_ONCE for ring indexes

Eric points out that we should make sure that ring index updates
are wrapped in the appropriate READ_ONCE/WRITE_ONCE macros.

Suggested-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: docs: update the sample code in driver.rst
Jakub Kicinski [Wed, 12 Apr 2023 01:50:36 +0000 (18:50 -0700)]
net: docs: update the sample code in driver.rst

The sample code talks about single-queue devices and uses locks.
Update it to something resembling more modern code.
Make sure we mention use of READ_ONCE() / WRITE_ONCE().

Change the comment which talked about consumer on the xmit side.
AFAIU xmit is the producer and completions are a consumer.

Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agoMerge branch 'add-emac3-support-for-sa8540p-ride'
Paolo Abeni [Thu, 13 Apr 2023 10:50:47 +0000 (12:50 +0200)]
Merge branch 'add-emac3-support-for-sa8540p-ride'

Andrew Halaney says:

====================
Add EMAC3 support for sa8540p-ride

This is a forward port / upstream refactor of code delivered
downstream by Qualcomm over at [0] to enable the DWMAC5 based
implementation called EMAC3 on the sa8540p-ride dev board.

From what I can tell with the board schematic in hand,
as well as the code delivered, the main changes needed are:

    1. A new address space layout for dwmac5/EMAC3 MTL/DMA regs
    2. A new programming sequence required for the EMAC3 based platforms

This series makes the changes above as well as other housekeeping items
such as converting dt-bindings to yaml, etc.

As requested[1], it has been split up by compilation deps / maintainer tree.
I will post a link to the associated devicetree changes that together
with this series get the hardware functioning.

Patches 1-3 are clean ups of the currently supported dt-bindings and
IMO could be picked up as is independent of the rest of the series to
improve the current codebase. They've all been reviewed in prior
versions of the series.

Patches 5-7 are also clean ups of the driver and are worth picking up
independently as well. They don't all have explicit reviews but should
be good to go (trivial changes on non-reviewed bits).

The rest of the patches have new changes, lack review, or are specificly
being made to support the new hardware, so they should wait until the
series as a whole is deemed ready to go by the community.

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

Link: https://lore.kernel.org/r/20230411200409.455355-1-ahalaney@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: dwmac-qcom-ethqos: Add EMAC3 support
Andrew Halaney [Tue, 11 Apr 2023 20:04:09 +0000 (15:04 -0500)]
net: stmmac: dwmac-qcom-ethqos: Add EMAC3 support

Add the new programming sequence needed for EMAC3 based platforms such
as the sc8280xp family.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: dwmac-qcom-ethqos: Use loopback_en for all speeds
Andrew Halaney [Tue, 11 Apr 2023 20:04:08 +0000 (15:04 -0500)]
net: stmmac: dwmac-qcom-ethqos: Use loopback_en for all speeds

It seems that this variable should be used for all speeds, not just
1000/100.

While at it refactor it slightly to be more readable, including fixing
the typo in the variable name.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: dwmac-qcom-ethqos: Respect phy-mode and TX delay
Andrew Halaney [Tue, 11 Apr 2023 20:04:07 +0000 (15:04 -0500)]
net: stmmac: dwmac-qcom-ethqos: Respect phy-mode and TX delay

The driver currently sets a MAC TX delay of 2 ns no matter what the
phy-mode is. If the phy-mode indicates the phy is in charge of the
TX delay (rgmii-txid, rgmii-id), don't do it in the MAC.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: dwmac4: Allow platforms to specify some DMA/MTL offsets
Andrew Halaney [Tue, 11 Apr 2023 20:04:06 +0000 (15:04 -0500)]
net: stmmac: dwmac4: Allow platforms to specify some DMA/MTL offsets

Some platforms have dwmac4 implementations that have a different
address space layout than the default, resulting in the need to define
their own DMA/MTL offsets.

Extend the functions to allow a platform driver to indicate what its
addresses are, overriding the defaults.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: Pass stmmac_priv in some callbacks
Andrew Halaney [Tue, 11 Apr 2023 20:04:05 +0000 (15:04 -0500)]
net: stmmac: Pass stmmac_priv in some callbacks

Passing stmmac_priv to some of the callbacks allows hwif implementations
to grab some data that platforms can customize. Adjust the callbacks
accordingly in preparation of such a platform customization.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: Remove some unnecessary void pointers
Andrew Halaney [Tue, 11 Apr 2023 20:04:04 +0000 (15:04 -0500)]
net: stmmac: Remove some unnecessary void pointers

There's a few spots in the hardware interface where a void pointer is
used, but what's passed in and later cast out is always the same type.

Just use the proper type directly.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: Fix DMA typo
Andrew Halaney [Tue, 11 Apr 2023 20:04:03 +0000 (15:04 -0500)]
net: stmmac: Fix DMA typo

DAM is supposed to be DMA. Fix it to improve readability.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: stmmac: Remove unnecessary if statement brackets
Andrew Halaney [Tue, 11 Apr 2023 20:04:02 +0000 (15:04 -0500)]
net: stmmac: Remove unnecessary if statement brackets

The brackets are unnecessary, remove them to match the coding style
used in the kernel.

Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agodt-bindings: net: qcom,ethqos: Add Qualcomm sc8280xp compatibles
Andrew Halaney [Tue, 11 Apr 2023 20:04:01 +0000 (15:04 -0500)]
dt-bindings: net: qcom,ethqos: Add Qualcomm sc8280xp compatibles

The sc8280xp has a new version of the ETHQOS hardware in it, EMAC v3.
Add a compatible for this.

Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agodt-bindings: net: qcom,ethqos: Convert bindings to yaml
Bhupesh Sharma [Tue, 11 Apr 2023 20:04:00 +0000 (15:04 -0500)]
dt-bindings: net: qcom,ethqos: Convert bindings to yaml

Convert Qualcomm ETHQOS Ethernet devicetree binding to YAML.
In doing so add a new property for iommus since newer platforms support
using one, and without such make dtbs_check fails on them.

While at it, also update the MAINTAINERS file to point to the yaml
version of the bindings.

Signed-off-by: Bhupesh Sharma <bhupesh.sharma@linaro.org>
[halaney: Remove duplicated properties, add MAINTAINERS and iommus]
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agodt-bindings: net: snps,dwmac: Add Qualcomm Ethernet ETHQOS compatibles
Bhupesh Sharma [Tue, 11 Apr 2023 20:03:59 +0000 (15:03 -0500)]
dt-bindings: net: snps,dwmac: Add Qualcomm Ethernet ETHQOS compatibles

Add Qualcomm Ethernet ETHQOS compatible checks
in snps,dwmac YAML binding document.

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Bhupesh Sharma <bhupesh.sharma@linaro.org>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agodt-bindings: net: snps,dwmac: Update interrupt-names
Bhupesh Sharma [Tue, 11 Apr 2023 20:03:58 +0000 (15:03 -0500)]
dt-bindings: net: snps,dwmac: Update interrupt-names

As commit fc191af1bb0d ("net: stmmac: platform: Fix misleading
interrupt error msg") noted, not every stmmac based platform
makes use of the 'eth_wake_irq' or 'eth_lpi' interrupts.

So, update the 'interrupt-names' inside 'snps,dwmac' YAML
bindings to reflect the same.

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Bhupesh Sharma <bhupesh.sharma@linaro.org>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Tested-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: enetc: workaround for unresponsive pMAC after receiving express traffic
Vladimir Oltean [Tue, 11 Apr 2023 19:26:45 +0000 (22:26 +0300)]
net: enetc: workaround for unresponsive pMAC after receiving express traffic

I have observed an issue where the RX direction of the LS1028A ENETC pMAC
seems unresponsive. The minimal procedure to reproduce the issue is:

1. Connect ENETC port 0 with a loopback RJ45 cable to one of the Felix
   switch ports (0).

2. Bring the ports up (MAC Merge layer is not enabled on either end).

3. Send a large quantity of unidirectional (express) traffic from Felix
   to ENETC. I tried altering frame size and frame count, and it doesn't
   appear to be specific to either of them, but rather, to the quantity
   of octets received. Lowering the frame count, the minimum quantity of
   packets to reproduce relatively consistently seems to be around 37000
   frames at 1514 octets (w/o FCS) each.

4. Using ethtool --set-mm, enable the pMAC in the Felix and in the ENETC
   ports, in both RX and TX directions, and with verification on both
   ends.

5. Wait for verification to complete on both sides.

6. Configure a traffic class as preemptible on both ends.

7. Send some packets again.

The issue is at step 5, where the verification process of ENETC ends
(meaning that Felix responds with an SMD-R and ENETC sees the response),
but the verification process of Felix never ends (it remains VERIFYING).

If step 3 is skipped or if ENETC receives less traffic than
approximately that threshold, the test runs all the way through
(verification succeeds on both ends, preemptible traffic passes fine).

If, between step 4 and 5, the step below is also introduced:

4.1. Disable and re-enable PM0_COMMAND_CONFIG bit RX_EN

then again, the sequence of steps runs all the way through, and
verification succeeds, even if there was the previous RX traffic
injected into ENETC.

Traffic sent *by* the ENETC port prior to enabling the MAC Merge layer
does not seem to influence the verification result, only received
traffic does.

The LS1028A manual does not mention any relationship between
PM0_COMMAND_CONFIG and MMCSR, and the hardware people don't seem to
know for now either.

The bit that is toggled to work around the issue is also toggled
by enetc_mac_enable(), called from phylink's mac_link_down() and
mac_link_up() methods - which is how the workaround was found:
verification would work after a link down/up.

Fixes: c7b9e8086902 ("net: enetc: add support for MAC Merge layer")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230411192645.1896048-1-vladimir.oltean@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agobnxt_en: Allow to set switchdev mode without existing VFs
Ivan Vecera [Tue, 11 Apr 2023 12:04:42 +0000 (14:04 +0200)]
bnxt_en: Allow to set switchdev mode without existing VFs

Remove an inability of bnxt_en driver to set eswitch to switchdev
mode without existing VFs by:

1. Allow to set switchdev mode in bnxt_dl_eswitch_mode_set() so
   representors are created only when num_vfs > 0 otherwise just
   set bp->eswitch_mode
2. Do not automatically change bp->eswitch_mode during
   bnxt_vf_reps_create() and bnxt_vf_reps_destroy() calls so
   the eswitch mode is managed only by an user by devlink.
   Just set temporarily bp->eswitch_mode to legacy to avoid
   re-opening of representors during destroy.
3. Create representors in bnxt_sriov_enable() if current eswitch
   mode is switchdev one

Tested by this sequence:
1. Set PF interface up
2. Set PF's eswitch mode to switchdev
3. Created N VFs
4. Checked that N representors were created
5. Set eswitch mode to legacy
6. Checked that representors were deleted
7. Set eswitch mode back to switchdev
8. Checked that representors exist again for VFs
9. Deleted all VFs
10. Checked that all representors were deleted as well
11. Checked that current eswitch mode is still switchdev

Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Link: https://lore.kernel.org/r/20230411120443.126055-1-ivecera@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agosctp: fix a potential overflow in sctp_ifwdtsn_skip
Xin Long [Mon, 10 Apr 2023 19:43:30 +0000 (15:43 -0400)]
sctp: fix a potential overflow in sctp_ifwdtsn_skip

Currently, when traversing ifwdtsn skips with _sctp_walk_ifwdtsn, it only
checks the pos against the end of the chunk. However, the data left for
the last pos may be < sizeof(struct sctp_ifwdtsn_skip), and dereference
it as struct sctp_ifwdtsn_skip may cause coverflow.

This patch fixes it by checking the pos against "the end of the chunk -
sizeof(struct sctp_ifwdtsn_skip)" in sctp_ifwdtsn_skip, similar to
sctp_fwdtsn_skip.

Fixes: 0fc2ea922c8a ("sctp: implement validate_ftsn for sctp_stream_interleave")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Link: https://lore.kernel.org/r/2a71bffcd80b4f2c61fac6d344bb2f11c8fd74f7.1681155810.git.lucien.xin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agoMerge tag 'amd-drm-fixes-6.3-2023-04-12' of https://gitlab.freedesktop.org/agd5f...
Daniel Vetter [Thu, 13 Apr 2023 07:57:19 +0000 (09:57 +0200)]
Merge tag 'amd-drm-fixes-6.3-2023-04-12' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amd-drm-fixes-6.3-2023-04-12:

amdgpu:
- SMU13 fixes
- DP MST fix

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230412215637.7881-1-alexander.deucher@amd.com
20 months agonet: qrtr: Fix an uninit variable access bug in qrtr_tx_resume()
Ziyang Xuan [Mon, 10 Apr 2023 01:23:52 +0000 (09:23 +0800)]
net: qrtr: Fix an uninit variable access bug in qrtr_tx_resume()

Syzbot reported a bug as following:

=====================================================
BUG: KMSAN: uninit-value in qrtr_tx_resume+0x185/0x1f0 net/qrtr/af_qrtr.c:230
 qrtr_tx_resume+0x185/0x1f0 net/qrtr/af_qrtr.c:230
 qrtr_endpoint_post+0xf85/0x11b0 net/qrtr/af_qrtr.c:519
 qrtr_tun_write_iter+0x270/0x400 net/qrtr/tun.c:108
 call_write_iter include/linux/fs.h:2189 [inline]
 aio_write+0x63a/0x950 fs/aio.c:1600
 io_submit_one+0x1d1c/0x3bf0 fs/aio.c:2019
 __do_sys_io_submit fs/aio.c:2078 [inline]
 __se_sys_io_submit+0x293/0x770 fs/aio.c:2048
 __x64_sys_io_submit+0x92/0xd0 fs/aio.c:2048
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

Uninit was created at:
 slab_post_alloc_hook mm/slab.h:766 [inline]
 slab_alloc_node mm/slub.c:3452 [inline]
 __kmem_cache_alloc_node+0x71f/0xce0 mm/slub.c:3491
 __do_kmalloc_node mm/slab_common.c:967 [inline]
 __kmalloc_node_track_caller+0x114/0x3b0 mm/slab_common.c:988
 kmalloc_reserve net/core/skbuff.c:492 [inline]
 __alloc_skb+0x3af/0x8f0 net/core/skbuff.c:565
 __netdev_alloc_skb+0x120/0x7d0 net/core/skbuff.c:630
 qrtr_endpoint_post+0xbd/0x11b0 net/qrtr/af_qrtr.c:446
 qrtr_tun_write_iter+0x270/0x400 net/qrtr/tun.c:108
 call_write_iter include/linux/fs.h:2189 [inline]
 aio_write+0x63a/0x950 fs/aio.c:1600
 io_submit_one+0x1d1c/0x3bf0 fs/aio.c:2019
 __do_sys_io_submit fs/aio.c:2078 [inline]
 __se_sys_io_submit+0x293/0x770 fs/aio.c:2048
 __x64_sys_io_submit+0x92/0xd0 fs/aio.c:2048
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

It is because that skb->len requires at least sizeof(struct qrtr_ctrl_pkt)
in qrtr_tx_resume(). And skb->len equals to size in qrtr_endpoint_post().
But size is less than sizeof(struct qrtr_ctrl_pkt) when qrtr_cb->type
equals to QRTR_TYPE_RESUME_TX in qrtr_endpoint_post() under the syzbot
scenario. This triggers the uninit variable access bug.

Add size check when qrtr_cb->type equals to QRTR_TYPE_RESUME_TX in
qrtr_endpoint_post() to fix the bug.

Fixes: 5fdeb0d372ab ("net: qrtr: Implement outgoing flow control")
Reported-by: syzbot+4436c9630a45820fda76@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=c14607f0963d27d5a3d5f4c8639b500909e43540
Suggested-by: Manivannan Sadhasivam <mani@kernel.org>
Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230410012352.3997823-1-william.xuanziyang@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agoMerge branch 'net-thunderbolt-fix-for-sparse-warnings-and-typos'
Jakub Kicinski [Thu, 13 Apr 2023 04:09:14 +0000 (21:09 -0700)]
Merge branch 'net-thunderbolt-fix-for-sparse-warnings-and-typos'

Mika Westerberg says:

====================
net: thunderbolt: Fix for sparse warnings and typos

This series tries to fix the rest of the sparse warnings generated
against the driver. While there fix the two typos in comments as well.

The previous version of the series can be found here:
  https://lore.kernel.org/netdev/20230404053636.51597-1-mika.westerberg@linux.intel.com/
====================

Link: https://lore.kernel.org/r/20230411091049.12998-1-mika.westerberg@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonet: thunderbolt: Fix typos in comments
Mika Westerberg [Tue, 11 Apr 2023 09:10:49 +0000 (12:10 +0300)]
net: thunderbolt: Fix typos in comments

Fix two typos in comments:

  blongs -> belongs
  UPD -> UDP

No functional changes.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>