OSDN Git Service

uclinux-h8/linux.git
2 years agobpftool: Migrate from bpf_prog_test_run_xattr
Delyan Kratunov [Wed, 2 Feb 2022 23:54:22 +0000 (15:54 -0800)]
bpftool: Migrate from bpf_prog_test_run_xattr

bpf_prog_test_run is being deprecated in favor of the OPTS-based
bpf_prog_test_run_opts.

Signed-off-by: Delyan Kratunov <delyank@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220202235423.1097270-4-delyank@fb.com
2 years agoselftests/bpf: Migrate from bpf_prog_test_run_xattr
Delyan Kratunov [Wed, 2 Feb 2022 23:54:21 +0000 (15:54 -0800)]
selftests/bpf: Migrate from bpf_prog_test_run_xattr

bpf_prog_test_run_xattr is being deprecated in favor of the OPTS-based
bpf_prog_test_run_opts.
We end up unable to use CHECK_ATTR so replace usages with ASSERT_* calls.
Also, prog_run_xattr is now prog_run_opts.

Signed-off-by: Delyan Kratunov <delyank@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220202235423.1097270-3-delyank@fb.com
2 years agoselftests/bpf: Migrate from bpf_prog_test_run
Delyan Kratunov [Wed, 2 Feb 2022 23:54:20 +0000 (15:54 -0800)]
selftests/bpf: Migrate from bpf_prog_test_run

bpf_prog_test_run is being deprecated in favor of the OPTS-based
bpf_prog_test_run_opts.
We end up unable to use CHECK in most cases, so replace usages with
ASSERT_* calls.

Signed-off-by: Delyan Kratunov <delyank@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220202235423.1097270-2-delyank@fb.com
2 years agoMerge branch 'bpf-btf-dwarf5'
Daniel Borkmann [Wed, 2 Feb 2022 10:19:34 +0000 (11:19 +0100)]
Merge branch 'bpf-btf-dwarf5'

Nathan Chancellor says:

====================
This series allows CONFIG_DEBUG_INFO_DWARF5 to be selected with
CONFIG_DEBUG_INFO_BTF=y by checking the pahole version.

The first four patches add CONFIG_PAHOLE_VERSION and
scripts/pahole-version.sh to clean up all the places that pahole's
version is transformed into a 3-digit form.

The fourth patch adds a PAHOLE_VERSION dependency to DEBUG_INFO_DWARF5
so that there are no build errors when it is selected with
DEBUG_INFO_BTF.

I build tested Fedora's aarch64 and x86_64 config with ToT clang 14.0.0
and GCC 11 with CONFIG_DEBUG_INFO_DWARF5 enabled with both pahole 1.21
and 1.23.
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2 years agolib/Kconfig.debug: Allow BTF + DWARF5 with pahole 1.21+
Nathan Chancellor [Tue, 1 Feb 2022 20:56:24 +0000 (13:56 -0700)]
lib/Kconfig.debug: Allow BTF + DWARF5 with pahole 1.21+

Commit 98cd6f521f10 ("Kconfig: allow explicit opt in to DWARF v5")
prevented CONFIG_DEBUG_INFO_DWARF5 from being selected when
CONFIG_DEBUG_INFO_BTF is enabled because pahole had issues with clang's
DWARF5 info. This was resolved by [1], which is in pahole v1.21.

Allow DEBUG_INFO_DWARF5 to be selected with DEBUG_INFO_BTF when using
pahole v1.21 or newer.

[1]: https://git.kernel.org/pub/scm/devel/pahole/pahole.git/commit/?id=7d8e829f636f47aba2e1b6eda57e74d8e31f733c

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220201205624.652313-6-nathan@kernel.org
2 years agolib/Kconfig.debug: Use CONFIG_PAHOLE_VERSION
Nathan Chancellor [Tue, 1 Feb 2022 20:56:23 +0000 (13:56 -0700)]
lib/Kconfig.debug: Use CONFIG_PAHOLE_VERSION

Now that CONFIG_PAHOLE_VERSION exists, use it in the definition of
CONFIG_PAHOLE_HAS_SPLIT_BTF and CONFIG_PAHOLE_HAS_BTF_TAG to reduce the
amount of duplication across the tree.

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220201205624.652313-5-nathan@kernel.org
2 years agoscripts/pahole-flags.sh: Use pahole-version.sh
Nathan Chancellor [Tue, 1 Feb 2022 20:56:22 +0000 (13:56 -0700)]
scripts/pahole-flags.sh: Use pahole-version.sh

Use pahole-version.sh to get pahole's version code to reduce the amount
of duplication across the tree.

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220201205624.652313-4-nathan@kernel.org
2 years agokbuild: Add CONFIG_PAHOLE_VERSION
Nathan Chancellor [Tue, 1 Feb 2022 20:56:21 +0000 (13:56 -0700)]
kbuild: Add CONFIG_PAHOLE_VERSION

There are a few different places where pahole's version is turned into a
three digit form with the exact same command. Move this command into
scripts/pahole-version.sh to reduce the amount of duplication across the
tree.

Create CONFIG_PAHOLE_VERSION so the version code can be used in Kconfig
to enable and disable configuration options based on the pahole version,
which is already done in a couple of places.

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220201205624.652313-3-nathan@kernel.org
2 years agoMAINTAINERS: Add scripts/pahole-flags.sh to BPF section
Nathan Chancellor [Tue, 1 Feb 2022 20:56:20 +0000 (13:56 -0700)]
MAINTAINERS: Add scripts/pahole-flags.sh to BPF section

Currently, scripts/pahole-flags.sh has no formal maintainer. Add it to
the BPF section so that patches to it can be properly reviewed and
picked up.

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220201205624.652313-2-nathan@kernel.org
2 years agoMerge branch 'bpf-drop-libbpf-from-preload'
Daniel Borkmann [Tue, 1 Feb 2022 22:56:19 +0000 (23:56 +0100)]
Merge branch 'bpf-drop-libbpf-from-preload'

Alexei Starovoitov says:

====================
CO-RE in the kernel support allows bpf preload to switch to light
skeleton and remove libbpf dependency.

This reduces the size of bpf_preload_umd from 300kbyte to 19kbyte
and eventually will make "kernel skeleton" possible.
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2 years agobpf: Drop libbpf, libelf, libz dependency from bpf preload.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:28 +0000 (14:05 -0800)]
bpf: Drop libbpf, libelf, libz dependency from bpf preload.

Drop libbpf, libelf, libz dependency from bpf preload.
This reduces bpf_preload_umd binary size
from 1.7M to 30k unstripped with debug info
and from 300k to 19k stripped.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-8-alexei.starovoitov@gmail.com
2 years agobpf: Open code obj_get_info_by_fd in bpf preload.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:27 +0000 (14:05 -0800)]
bpf: Open code obj_get_info_by_fd in bpf preload.

Open code obj_get_info_by_fd in bpf preload.
It's the last part of libbpf that preload/iterators were using.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-7-alexei.starovoitov@gmail.com
2 years agobpf: Convert bpf preload to light skeleton.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:26 +0000 (14:05 -0800)]
bpf: Convert bpf preload to light skeleton.

Convert bpffs preload iterators to light skeleton.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-6-alexei.starovoitov@gmail.com
2 years agobpf: Remove unnecessary setrlimit from bpf preload.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:25 +0000 (14:05 -0800)]
bpf: Remove unnecessary setrlimit from bpf preload.

BPF programs and maps are memcg accounted. setrlimit is obsolete.
Remove its use from bpf preload.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-5-alexei.starovoitov@gmail.com
2 years agolibbpf: Open code raw_tp_open and link_create commands.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:24 +0000 (14:05 -0800)]
libbpf: Open code raw_tp_open and link_create commands.

Open code raw_tracepoint_open and link_create used by light skeleton
to be able to avoid full libbpf eventually.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-4-alexei.starovoitov@gmail.com
2 years agolibbpf: Open code low level bpf commands.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:23 +0000 (14:05 -0800)]
libbpf: Open code low level bpf commands.

Open code low level bpf commands used by light skeleton to
be able to avoid full libbpf eventually.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-3-alexei.starovoitov@gmail.com
2 years agolibbpf: Add support for bpf iter in light skeleton.
Alexei Starovoitov [Mon, 31 Jan 2022 22:05:22 +0000 (14:05 -0800)]
libbpf: Add support for bpf iter in light skeleton.

bpf iterator programs should use bpf_link_create to attach instead of
bpf_raw_tracepoint_open like other tracing programs.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220131220528.98088-2-alexei.starovoitov@gmail.com
2 years agoMerge branch 'libbpf: deprecate xdp_cpumap, xdp_devmap and classifier sec definitions'
Andrii Nakryiko [Tue, 1 Feb 2022 17:51:32 +0000 (09:51 -0800)]
Merge branch 'libbpf: deprecate xdp_cpumap, xdp_devmap and classifier sec definitions'

Lorenzo Bianconi says:

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

Deprecate xdp_cpumap, xdp_devmap and classifier sec definitions.
Update cpumap/devmap samples and kselftests.

Changes since v2:
- update warning log
- split libbpf and samples/kselftests changes
- deprecate classifier sec definition

Changes since v1:
- refer to Libbpf-1.0-migration-guide in the warning rised by libbpf
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agosamples/bpf: Update cpumap/devmap sec_name
Lorenzo Bianconi [Tue, 1 Feb 2022 14:58:10 +0000 (15:58 +0100)]
samples/bpf: Update cpumap/devmap sec_name

Substitute deprecated xdp_cpumap and xdp_devmap sec_name with
xdp/cpumap and xdp/devmap respectively.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/509201497c6c4926bc941f1cba24173cf500e760.1643727185.git.lorenzo@kernel.org
2 years agoselftests/bpf: Update cpumap/devmap sec_name
Lorenzo Bianconi [Tue, 1 Feb 2022 14:58:09 +0000 (15:58 +0100)]
selftests/bpf: Update cpumap/devmap sec_name

Substitute deprecated xdp_cpumap and xdp_devmap sec_name with
xdp/cpumap and xdp/devmap respectively in bpf kselftests.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/9a4286cd36781e2c31ba3773bfdcf45cf1bbaa9e.1643727185.git.lorenzo@kernel.org
2 years agolibbpf: Deprecate xdp_cpumap, xdp_devmap and classifier sec definitions
Lorenzo Bianconi [Tue, 1 Feb 2022 14:58:08 +0000 (15:58 +0100)]
libbpf: Deprecate xdp_cpumap, xdp_devmap and classifier sec definitions

Deprecate xdp_cpumap, xdp_devmap and classifier sec definitions.
Introduce xdp/devmap and xdp/cpumap definitions according to the
standard for SEC("") in libbpf:
- prog_type.prog_flags/attach_place

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/5c7bd9426b3ce6a31d9a4b1f97eb299e1467fc52.1643727185.git.lorenzo@kernel.org
2 years agolibbpf: Deprecate btf_ext rec_size APIs
Dave Marchevsky [Tue, 1 Feb 2022 01:46:10 +0000 (17:46 -0800)]
libbpf: Deprecate btf_ext rec_size APIs

btf_ext__{func,line}_info_rec_size functions are used in conjunction
with already-deprecated btf_ext__reloc_{func,line}_info functions. Since
struct btf_ext is opaque to the user it was necessary to expose rec_size
getters in the past.

btf_ext__reloc_{func,line}_info were deprecated in commit 8505e8709b5ee
("libbpf: Implement generalized .BTF.ext func/line info adjustment")
as they're not compatible with support for multiple programs per
section. It was decided[0] that users of these APIs should implement their
own .btf.ext parsing to access this data, in which case the rec_size
getters are unnecessary. So deprecate them from libbpf 0.7.0 onwards.

  [0] Closes: https://github.com/libbpf/libbpf/issues/277

Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220201014610.3522985-1-davemarchevsky@fb.com
2 years agobpf: make bpf_copy_from_user_task() gpl only
Kenta Tada [Fri, 28 Jan 2022 17:09:06 +0000 (02:09 +0900)]
bpf: make bpf_copy_from_user_task() gpl only

access_process_vm() is exported by EXPORT_SYMBOL_GPL().

Signed-off-by: Kenta Tada <Kenta.Tada@sony.com>
Link: https://lore.kernel.org/r/20220128170906.21154-1-Kenta.Tada@sony.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge branch 'Split bpf_sock dst_port field'
Alexei Starovoitov [Mon, 31 Jan 2022 20:39:12 +0000 (12:39 -0800)]
Merge branch 'Split bpf_sock dst_port field'

Jakub Sitnicki says:

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

This is a follow-up to discussion around the idea of making dst_port in struct
bpf_sock a 16-bit field that happened in [1].

v2:
- use an anonymous field for zero padding (Alexei)

v1:
- keep dst_field offset unchanged to prevent existing BPF program breakage
  (Martin)
- allow 8-bit loads from dst_port[0] and [1]
- add test coverage for the verifier and the context access converter

[1] https://lore.kernel.org/bpf/87sftbobys.fsf@cloudflare.com/
====================

Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Extend verifier and bpf_sock tests for dst_port loads
Jakub Sitnicki [Sun, 30 Jan 2022 11:55:18 +0000 (12:55 +0100)]
selftests/bpf: Extend verifier and bpf_sock tests for dst_port loads

Add coverage to the verifier tests and tests for reading bpf_sock fields to
ensure that 32-bit, 16-bit, and 8-bit loads from dst_port field are allowed
only at intended offsets and produce expected values.

While 16-bit and 8-bit access to dst_port field is straight-forward, 32-bit
wide loads need be allowed and produce a zero-padded 16-bit value for
backward compatibility.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/r/20220130115518.213259-3-jakub@cloudflare.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Make dst_port field in struct bpf_sock 16-bit wide
Jakub Sitnicki [Sun, 30 Jan 2022 11:55:17 +0000 (12:55 +0100)]
bpf: Make dst_port field in struct bpf_sock 16-bit wide

Menglong Dong reports that the documentation for the dst_port field in
struct bpf_sock is inaccurate and confusing. From the BPF program PoV, the
field is a zero-padded 16-bit integer in network byte order. The value
appears to the BPF user as if laid out in memory as so:

  offsetof(struct bpf_sock, dst_port) + 0  <port MSB>
                                      + 8  <port LSB>
                                      +16  0x00
                                      +24  0x00

32-, 16-, and 8-bit wide loads from the field are all allowed, but only if
the offset into the field is 0.

32-bit wide loads from dst_port are especially confusing. The loaded value,
after converting to host byte order with bpf_ntohl(dst_port), contains the
port number in the upper 16-bits.

Remove the confusion by splitting the field into two 16-bit fields. For
backward compatibility, allow 32-bit wide loads from offsetof(struct
bpf_sock, dst_port).

While at it, allow loads 8-bit loads at offset [0] and [1] from dst_port.

Reported-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/r/20220130115518.213259-2-jakub@cloudflare.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge branch 'selftests/bpf: use temp netns for testing'
Alexei Starovoitov [Fri, 28 Jan 2022 03:21:28 +0000 (19:21 -0800)]
Merge branch 'selftests/bpf: use temp netns for testing'

Hangbin Liu says:

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

There are some bpf tests using hard code netns name like ns0, ns1, etc.
This kind of ns name is easily used by other tests or system. If there
is already a such netns, all the related tests will failed. So let's
use temp netns name for testing.

The first patch not only change to temp netns. But also fixed an interface
index issue. So I add fixes tag. For the later patches, I think that
should be an update instead of fixes, so the fixes tag is not added.
====================

Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_xdp_redirect: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:17 +0000 (16:17 +0800)]
selftests/bpf/test_xdp_redirect: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-8-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_xdp_meta: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:16 +0000 (16:17 +0800)]
selftests/bpf/test_xdp_meta: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-7-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_tcp_check_syncookie: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:15 +0000 (16:17 +0800)]
selftests/bpf/test_tcp_check_syncookie: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: Lorenz Bauer <lmb@cloudflare.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-6-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_lwt_seg6local: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:14 +0000 (16:17 +0800)]
selftests/bpf/test_lwt_seg6local: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-5-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_xdp_vlan: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:13 +0000 (16:17 +0800)]
selftests/bpf/test_xdp_vlan: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-4-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_xdp_veth: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:12 +0000 (16:17 +0800)]
selftests/bpf/test_xdp_veth: use temp netns for testing

Use temp netns instead of hard code name for testing in case the
netns already exists.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-3-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf/test_xdp_redirect_multi: use temp netns for testing
Hangbin Liu [Tue, 25 Jan 2022 08:17:11 +0000 (16:17 +0800)]
selftests/bpf/test_xdp_redirect_multi: use temp netns for testing

Use temp netns instead of hard code name for testing in case the netns
already exists.

Remove the hard code interface index when creating the veth interfaces.
Because when the system loads some virtual interface modules, e.g. tunnels.
the ifindex of 2 will be used and the cmd will fail.

As the netns has not created if checking environment failed. Trap the
clean up function after checking env.

Fixes: 8955c1a32987 ("selftests/bpf/xdp_redirect_multi: Limit the tests in netns")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Link: https://lore.kernel.org/r/20220125081717.1260849-2-liuhangbin@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf, x86: Remove unnecessary handling of BPF_SUB atomic op
Hou Tao [Thu, 27 Jan 2022 08:32:40 +0000 (16:32 +0800)]
bpf, x86: Remove unnecessary handling of BPF_SUB atomic op

According to the LLVM commit (https://reviews.llvm.org/D72184),
sync_fetch_and_sub() is implemented as a negation followed by
sync_fetch_and_add(), so there will be no BPF_SUB op, thus just
remove it. BPF_SUB is also rejected by the verifier anyway.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Brendan Jackman <jackmanb@google.com>
Link: https://lore.kernel.org/bpf/20220127083240.1425481-1-houtao1@huawei.com
2 years agoMerge branch 'bpf: add __user tagging support in vmlinux BTF'
Alexei Starovoitov [Thu, 27 Jan 2022 20:03:47 +0000 (12:03 -0800)]
Merge branch 'bpf: add __user tagging support in vmlinux BTF'

Yonghong Song says:

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

The __user attribute is currently mainly used by sparse for type checking.
The attribute indicates whether a memory access is in user memory address
space or not. Such information is important during tracing kernel
internal functions or data structures as accessing user memory often
has different mechanisms compared to accessing kernel memory. For example,
the perf-probe needs explicit command line specification to indicate a
particular argument or string in user-space memory ([1], [2], [3]).
Currently, vmlinux BTF is available in kernel with many distributions.
If __user attribute information is available in vmlinux BTF, the explicit
user memory access information from users will not be necessary as
the kernel can figure it out by itself with vmlinux BTF.

Besides the above possible use for perf/probe, another use case is
for bpf verifier. Currently, for bpf BPF_PROG_TYPE_TRACING type of bpf
programs, users can write direct code like
  p->m1->m2
and "p" could be a function parameter. Without __user information in BTF,
the verifier will assume p->m1 accessing kernel memory and will generate
normal loads. Let us say "p" actually tagged with __user in the source
code.  In such cases, p->m1 is actually accessing user memory and direct
load is not right and may produce incorrect result. For such cases,
bpf_probe_read_user() will be the correct way to read p->m1.

To support encoding __user information in BTF, a new attribute
  __attribute__((btf_type_tag("<arbitrary_string>")))
is implemented in clang ([4]). For example, if we have
  #define __user __attribute__((btf_type_tag("user")))
during kernel compilation, the attribute "user" information will
be preserved in dwarf. After pahole converting dwarf to BTF, __user
information will be available in vmlinux BTF and such information
can be used by bpf verifier, perf/probe or other use cases.

Currently btf_type_tag is only supported in clang (>= clang14) and
pahole (>= 1.23). gcc support is also proposed and under development ([5]).

In the rest of patch set, Patch 1 added support of __user btf_type_tag
during compilation. Patch 2 added bpf verifier support to utilize __user
tag information to reject bpf programs not using proper helper to access
user memories. Patches 3-5 are for bpf selftests which demonstrate verifier
can reject direct user memory accesses.

  [1] http://lkml.kernel.org/r/155789874562.26965.10836126971405890891.stgit@devnote2
  [2] http://lkml.kernel.org/r/155789872187.26965.4468456816590888687.stgit@devnote2
  [3] http://lkml.kernel.org/r/155789871009.26965.14167558859557329331.stgit@devnote2
  [4] https://reviews.llvm.org/D111199
  [5] https://lore.kernel.org/bpf/0cbeb2fb-1a18-f690-e360-24b1c90c2a91@fb.com/

Changelog:
  v2 -> v3:
    - remove FLAG_DONTCARE enumerator and just use 0 as dontcare flag.
    - explain how btf type_tag is encoded in btf type chain.
  v1 -> v2:
    - use MEM_USER flag for PTR_TO_BTF_ID reg type instead of a separate
      field to encode __user tag.
    - add a test with kernel function __sys_getsockname which has __user tagged
      argument.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agodocs/bpf: clarify how btf_type_tag gets encoded in the type chain
Yonghong Song [Thu, 27 Jan 2022 15:46:27 +0000 (07:46 -0800)]
docs/bpf: clarify how btf_type_tag gets encoded in the type chain

Clarify where the BTF_KIND_TYPE_TAG gets encoded in the type chain,
so applications and kernel can properly parse them.

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154627.665163-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: specify pahole version requirement for btf_tag test
Yonghong Song [Thu, 27 Jan 2022 15:46:22 +0000 (07:46 -0800)]
selftests/bpf: specify pahole version requirement for btf_tag test

Specify pahole version requirement (1.23) for btf_tag subtests
btf_type_tag_user_{mod1, mod2, vmlinux}.

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154622.663337-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: add a selftest with __user tag
Yonghong Song [Thu, 27 Jan 2022 15:46:16 +0000 (07:46 -0800)]
selftests/bpf: add a selftest with __user tag

Added a selftest with three__user usages: a __user pointer-type argument
in bpf_testmod, a __user pointer-type struct member in bpf_testmod,
and a __user pointer-type struct member in vmlinux. In all cases,
directly accessing the user memory will result verification failure.

  $ ./test_progs -v -n 22/3
  ...
  libbpf: prog 'test_user1': BPF program load failed: Permission denied
  libbpf: prog 'test_user1': -- BEGIN PROG LOAD LOG --
  R1 type=ctx expected=fp
  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  ; int BPF_PROG(test_user1, struct bpf_testmod_btf_type_tag_1 *arg)
  0: (79) r1 = *(u64 *)(r1 +0)
  func 'bpf_testmod_test_btf_type_tag_user_1' arg0 has btf_id 136561 type STRUCT 'bpf_testmod_btf_type_tag_1'
  1: R1_w=user_ptr_bpf_testmod_btf_type_tag_1(id=0,off=0,imm=0)
  ; g = arg->a;
  1: (61) r1 = *(u32 *)(r1 +0)
  R1 invalid mem access 'user_ptr_'
  ...
  #22/3 btf_tag/btf_type_tag_user_mod1:OK

  $ ./test_progs -v -n 22/4
  ...
  libbpf: prog 'test_user2': BPF program load failed: Permission denied
  libbpf: prog 'test_user2': -- BEGIN PROG LOAD LOG --
  R1 type=ctx expected=fp
  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  ; int BPF_PROG(test_user2, struct bpf_testmod_btf_type_tag_2 *arg)
  0: (79) r1 = *(u64 *)(r1 +0)
  func 'bpf_testmod_test_btf_type_tag_user_2' arg0 has btf_id 136563 type STRUCT 'bpf_testmod_btf_type_tag_2'
  1: R1_w=ptr_bpf_testmod_btf_type_tag_2(id=0,off=0,imm=0)
  ; g = arg->p->a;
  1: (79) r1 = *(u64 *)(r1 +0)          ; R1_w=user_ptr_bpf_testmod_btf_type_tag_1(id=0,off=0,imm=0)
  ; g = arg->p->a;
  2: (61) r1 = *(u32 *)(r1 +0)
  R1 invalid mem access 'user_ptr_'
  ...
  #22/4 btf_tag/btf_type_tag_user_mod2:OK

  $ ./test_progs -v -n 22/5
  ...
  libbpf: prog 'test_sys_getsockname': BPF program load failed: Permission denied
  libbpf: prog 'test_sys_getsockname': -- BEGIN PROG LOAD LOG --
  R1 type=ctx expected=fp
  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  ; int BPF_PROG(test_sys_getsockname, int fd, struct sockaddr *usockaddr,
  0: (79) r1 = *(u64 *)(r1 +8)
  func '__sys_getsockname' arg1 has btf_id 2319 type STRUCT 'sockaddr'
  1: R1_w=user_ptr_sockaddr(id=0,off=0,imm=0)
  ; g = usockaddr->sa_family;
  1: (69) r1 = *(u16 *)(r1 +0)
  R1 invalid mem access 'user_ptr_'
  ...
  #22/5 btf_tag/btf_type_tag_user_vmlinux:OK

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154616.659314-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: rename btf_decl_tag.c to test_btf_decl_tag.c
Yonghong Song [Thu, 27 Jan 2022 15:46:11 +0000 (07:46 -0800)]
selftests/bpf: rename btf_decl_tag.c to test_btf_decl_tag.c

The uapi btf.h contains the following declaration:
  struct btf_decl_tag {
       __s32   component_idx;
  };

The skeleton will also generate a struct with name
"btf_decl_tag" for bpf program btf_decl_tag.c.

Rename btf_decl_tag.c to test_btf_decl_tag.c so
the corresponding skeleton struct name becomes
"test_btf_decl_tag". This way, we could include
uapi btf.h in prog_tests/btf_tag.c.
There is no functionality change for this patch.

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154611.656699-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: reject program if a __user tagged memory accessed in kernel way
Yonghong Song [Thu, 27 Jan 2022 15:46:06 +0000 (07:46 -0800)]
bpf: reject program if a __user tagged memory accessed in kernel way

BPF verifier supports direct memory access for BPF_PROG_TYPE_TRACING type
of bpf programs, e.g., a->b. If "a" is a pointer
pointing to kernel memory, bpf verifier will allow user to write
code in C like a->b and the verifier will translate it to a kernel
load properly. If "a" is a pointer to user memory, it is expected
that bpf developer should be bpf_probe_read_user() helper to
get the value a->b. Without utilizing BTF __user tagging information,
current verifier will assume that a->b is a kernel memory access
and this may generate incorrect result.

Now BTF contains __user information, it can check whether the
pointer points to a user memory or not. If it is, the verifier
can reject the program and force users to use bpf_probe_read_user()
helper explicitly.

In the future, we can easily extend btf_add_space for other
address space tagging, for example, rcu/percpu etc.

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154606.654961-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agocompiler_types: define __user as __attribute__((btf_type_tag("user")))
Yonghong Song [Thu, 27 Jan 2022 15:46:00 +0000 (07:46 -0800)]
compiler_types: define __user as __attribute__((btf_type_tag("user")))

The __user attribute is currently mainly used by sparse for type checking.
The attribute indicates whether a memory access is in user memory address
space or not. Such information is important during tracing kernel
internal functions or data structures as accessing user memory often
has different mechanisms compared to accessing kernel memory. For example,
the perf-probe needs explicit command line specification to indicate a
particular argument or string in user-space memory ([1], [2], [3]).
Currently, vmlinux BTF is available in kernel with many distributions.
If __user attribute information is available in vmlinux BTF, the explicit
user memory access information from users will not be necessary as
the kernel can figure it out by itself with vmlinux BTF.

Besides the above possible use for perf/probe, another use case is
for bpf verifier. Currently, for bpf BPF_PROG_TYPE_TRACING type of bpf
programs, users can write direct code like
  p->m1->m2
and "p" could be a function parameter. Without __user information in BTF,
the verifier will assume p->m1 accessing kernel memory and will generate
normal loads. Let us say "p" actually tagged with __user in the source
code.  In such cases, p->m1 is actually accessing user memory and direct
load is not right and may produce incorrect result. For such cases,
bpf_probe_read_user() will be the correct way to read p->m1.

To support encoding __user information in BTF, a new attribute
  __attribute__((btf_type_tag("<arbitrary_string>")))
is implemented in clang ([4]). For example, if we have
  #define __user __attribute__((btf_type_tag("user")))
during kernel compilation, the attribute "user" information will
be preserved in dwarf. After pahole converting dwarf to BTF, __user
information will be available in vmlinux BTF.

The following is an example with latest upstream clang (clang14) and
pahole 1.23:

  [$ ~] cat test.c
  #define __user __attribute__((btf_type_tag("user")))
  int foo(int __user *arg) {
          return *arg;
  }
  [$ ~] clang -O2 -g -c test.c
  [$ ~] pahole -JV test.o
  ...
  [1] INT int size=4 nr_bits=32 encoding=SIGNED
  [2] TYPE_TAG user type_id=1
  [3] PTR (anon) type_id=2
  [4] FUNC_PROTO (anon) return=1 args=(3 arg)
  [5] FUNC foo type_id=4
  [$ ~]

You can see for the function argument "int __user *arg", its type is
described as
  PTR -> TYPE_TAG(user) -> INT
The kernel can use this information for bpf verification or other
use cases.

Current btf_type_tag is only supported in clang (>= clang14) and
pahole (>= 1.23). gcc support is also proposed and under development ([5]).

  [1] http://lkml.kernel.org/r/155789874562.26965.10836126971405890891.stgit@devnote2
  [2] http://lkml.kernel.org/r/155789872187.26965.4468456816590888687.stgit@devnote2
  [3] http://lkml.kernel.org/r/155789871009.26965.14167558859557329331.stgit@devnote2
  [4] https://reviews.llvm.org/D111199
  [5] https://lore.kernel.org/bpf/0cbeb2fb-1a18-f690-e360-24b1c90c2a91@fb.com/

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127154600.652613-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agocgroup/bpf: fast path skb BPF filtering
Pavel Begunkov [Thu, 27 Jan 2022 14:09:13 +0000 (14:09 +0000)]
cgroup/bpf: fast path skb BPF filtering

Even though there is a static key protecting from overhead from
cgroup-bpf skb filtering when there is nothing attached, in many cases
it's not enough as registering a filter for one type will ruin the fast
path for all others. It's observed in production servers I've looked
at but also in laptops, where registration is done during init by
systemd or something else.

Add a per-socket fast path check guarding from such overhead. This
affects both receive and transmit paths of TCP, UDP and other
protocols. It showed ~1% tx/s improvement in small payload UDP
send benchmarks using a real NIC and in a server environment and the
number jumps to 2-3% for preemtible kernels.

Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/d8c58857113185a764927a46f4b5a058d36d3ec3.1643292455.git.asml.silence@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: fix a clang compilation error
Yonghong Song [Thu, 27 Jan 2022 16:37:26 +0000 (08:37 -0800)]
selftests/bpf: fix a clang compilation error

When building selftests/bpf with clang
  make -j LLVM=1
  make -C tools/testing/selftests/bpf -j LLVM=1
I hit the following compilation error:

  trace_helpers.c:152:9: error: variable 'found' is used uninitialized whenever 'while' loop exits because its condition is false [-Werror,-Wsometimes-uninitialized]
          while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) {
                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  trace_helpers.c:161:7: note: uninitialized use occurs here
          if (!found)
               ^~~~~
  trace_helpers.c:152:9: note: remove the condition if it is always true
          while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) {
                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                 1
  trace_helpers.c:145:12: note: initialize the variable 'found' to silence this warning
          bool found;
                    ^
                     = false

It is possible that for sane /proc/self/maps we may never hit the above issue
in practice. But let us initialize variable 'found' properly to silence the
compilation error.

Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20220127163726.1442032-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests, xsk: Fix bpf_res cleanup test
Magnus Karlsson [Tue, 25 Jan 2022 08:29:45 +0000 (09:29 +0100)]
selftests, xsk: Fix bpf_res cleanup test

After commit 710ad98c363a ("veth: Do not record rx queue hint in veth_xmit"),
veth no longer receives traffic on the same queue as it was sent on. This
breaks the bpf_res test for the AF_XDP selftests as the socket tied to
queue 1 will not receive traffic anymore.

Modify the test so that two sockets are tied to queue id 0 using a shared
umem instead. When killing the first socket enter the second socket into
the xskmap so that traffic will flow to it. This will still test that the
resources are not cleaned up until after the second socket dies, without
having to rely on veth supporting rx_queue hints.

Reported-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20220125082945.26179-1-magnus.karlsson@gmail.com
2 years agoMerge branch 'xsk-batching'
Daniel Borkmann [Thu, 27 Jan 2022 16:25:33 +0000 (17:25 +0100)]
Merge branch 'xsk-batching'

Maciej Fijalkowski says:

====================
Unfortunately, similar scalability issues that were addressed for XDP
processing in ice, exist for XDP in the zero-copy driver used by AF_XDP.
Let's resolve them in mostly the same way as we did in [0] and utilize
the Tx batching API from XSK buffer pool.

Move the array of Tx descriptors that is used with batching approach to
the XSK buffer pool. This means that future users of this API will not
have to carry the array on their own side, they can simple refer to
pool's tx_desc array.

We also improve the Rx side where we extend ice_alloc_rx_buf_zc() to
handle the ring wrap and bump Rx tail more frequently. By doing so,
Rx side is adjusted to Tx and it was needed for l2fwd scenario.

Here are the improvements of performance numbers that this set brings
measured with xdpsock app in busy poll mode for 1 and 2 core modes.
Both Tx and Rx rings were sized to 1k length and busy poll budget was
256.

----------------------------------------------------------------
     |      txonly:      |      l2fwd      |      rxdrop
----------------------------------------------------------------
1C   |       149%        |       14%       |        3%
----------------------------------------------------------------
2C   |       134%        |       20%       |        5%
----------------------------------------------------------------

Next step will be to introduce batching onto Rx side.

v5:
* collect acks
* fix typos
* correct comments showing cache line boundaries in ice_tx_ring struct
v4 - address Alexandr's review:
* new patch (2) for making sure ring size is pow(2) when attaching
  xsk socket
* don't open code ALIGN_DOWN (patch 3)
* resign from storing tx_thresh in ice_tx_ring (patch 4)
* scope variables in a better way for Tx batching (patch 7)
v3:
* drop likely() that was wrapping napi_complete_done (patch 1)
* introduce configurable Tx threshold (patch 2)
* handle ring wrap on Rx side when allocating buffers (patch 3)
* respect NAPI budget when cleaning Tx descriptors in ZC (patch 6)
v2:
* introduce new patch that resets @next_dd and @next_rs fields
* use batching API for AF_XDP Tx on ice side

  [0]: https://lore.kernel.org/bpf/20211015162908.145341-8-anthony.l.nguyen@intel.com/
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2 years agoice: xsk: Borrow xdp_tx_active logic from i40e
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:46 +0000 (17:04 +0100)]
ice: xsk: Borrow xdp_tx_active logic from i40e

One of the things that commit 5574ff7b7b3d ("i40e: optimize AF_XDP Tx
completion path") introduced was the @xdp_tx_active field. Its usage
from i40e can be adjusted to ice driver and give us positive performance
results.

If the descriptor that @next_dd points to has been sent by HW (its DD
bit is set), then we are sure that at least quarter of the ring is ready
to be cleaned. If @xdp_tx_active is 0 which means that related xdp_ring
is not used for XDP_{TX, REDIRECT} workloads, then we know how many XSK
entries should placed to completion queue, IOW walking through the ring
can be skipped.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-9-maciej.fijalkowski@intel.com
2 years agoice: xsk: Improve AF_XDP ZC Tx and use batching API
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:45 +0000 (17:04 +0100)]
ice: xsk: Improve AF_XDP ZC Tx and use batching API

Apply the logic that was done for regular XDP from commit 9610bd988df9
("ice: optimize XDP_TX workloads") to the ZC side of the driver. On top
of that, introduce batching to Tx that is inspired by i40e's
implementation with adjustments to the cleaning logic - take into the
account NAPI budget in ice_clean_xdp_irq_zc().

Separating the stats structs onto separate cache lines seemed to improve
the performance.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-8-maciej.fijalkowski@intel.com
2 years agoice: xsk: Avoid potential dead AF_XDP Tx processing
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:44 +0000 (17:04 +0100)]
ice: xsk: Avoid potential dead AF_XDP Tx processing

Commit 9610bd988df9 ("ice: optimize XDP_TX workloads") introduced
@next_dd and @next_rs to ice_tx_ring struct. Currently, their state is
not restored in ice_clean_tx_ring(), which was not causing any troubles
as the XDP rings are gone after we're done with XDP prog on interface.

For upcoming usage of mentioned fields in AF_XDP, this might expose us
to a potential dead Tx side. Scenario would look like following (based
on xdpsock):

- two xdpsock instances are spawned in Tx mode
- one of them is killed
- XDP prog is kept on interface due to the other xdpsock still running
  * this means that XDP rings stayed in place
- xdpsock is launched again on same queue id that was terminated on
- @next_dd and @next_rs setting is bogus, therefore transmit side is
  broken

To protect us from the above, restore the initial @next_rs and @next_dd
values when cleaning the Tx ring.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-7-maciej.fijalkowski@intel.com
2 years agoi40e: xsk: Move tmp desc array from driver to pool
Magnus Karlsson [Tue, 25 Jan 2022 16:04:43 +0000 (17:04 +0100)]
i40e: xsk: Move tmp desc array from driver to pool

Move desc_array from the driver to the pool. The reason behind this is
that we can then reuse this array as a temporary storage for descriptors
in all zero-copy drivers that use the batched interface. This will make
it easier to add batching to more drivers.

i40e is the only driver that has a batched Tx zero-copy
implementation, so no need to touch any other driver.

Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-6-maciej.fijalkowski@intel.com
2 years agoice: Make Tx threshold dependent on ring length
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:42 +0000 (17:04 +0100)]
ice: Make Tx threshold dependent on ring length

XDP_TX workloads use a concept of Tx threshold that indicates the
interval of setting RS bit on descriptors which in turn tells the HW to
generate an interrupt to signal the completion of Tx on HW side. It is
currently based on a constant value of 32 which might not work out well
for various sizes of ring combined with for example batch size that can
be set via SO_BUSY_POLL_BUDGET.

Internal tests based on AF_XDP showed that most convenient setup of
mentioned threshold is when it is equal to quarter of a ring length.

Make use of recently introduced ICE_RING_QUARTER macro and use this
value as a substitute for ICE_TX_THRESH.

Align also ethtool -G callback so that next_dd/next_rs fields are up to
date in terms of the ring size.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-5-maciej.fijalkowski@intel.com
2 years agoice: xsk: Handle SW XDP ring wrap and bump tail more often
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:41 +0000 (17:04 +0100)]
ice: xsk: Handle SW XDP ring wrap and bump tail more often

Currently, if ice_clean_rx_irq_zc() processed the whole ring and
next_to_use != 0, then ice_alloc_rx_buf_zc() would not refill the whole
ring even if the XSK buffer pool would have enough free entries (either
from fill ring or the internal recycle mechanism) - it is because ring
wrap is not handled.

Improve the logic in ice_alloc_rx_buf_zc() to address the problem above.
Do not clamp the count of buffers that is passed to
xsk_buff_alloc_batch() in case when next_to_use + buffer count >=
rx_ring->count,  but rather split it and have two calls to the mentioned
function - one for the part up until the wrap and one for the part after
the wrap.

Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-4-maciej.fijalkowski@intel.com
2 years agoice: xsk: Force rings to be sized to power of 2
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:40 +0000 (17:04 +0100)]
ice: xsk: Force rings to be sized to power of 2

With the upcoming introduction of batching to XSK data path,
performance wise it will be the best to have the ring descriptor count
to be aligned to power of 2.

Check if ring sizes that user is going to attach the XSK socket fulfill
the condition above. For Tx side, although check is being done against
the Tx queue and in the end the socket will be attached to the XDP
queue, it is fine since XDP queues get the ring->count setting from Tx
queues.

Suggested-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-3-maciej.fijalkowski@intel.com
2 years agoice: Remove likely for napi_complete_done
Maciej Fijalkowski [Tue, 25 Jan 2022 16:04:39 +0000 (17:04 +0100)]
ice: Remove likely for napi_complete_done

Remove the likely before napi_complete_done as this is the unlikely case
when busy-poll is used. Removing this has a positive performance impact
for busy-poll and no negative impact to the regular case.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20220125160446.78976-2-maciej.fijalkowski@intel.com
2 years agobpf: remove unused static inlines
Jakub Kicinski [Wed, 26 Jan 2022 18:54:12 +0000 (10:54 -0800)]
bpf: remove unused static inlines

Remove two dead stubs, sk_msg_clear_meta() was never
used, use of xskq_cons_is_full() got replaced by
xsk_tx_writeable() in v5.10.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20220126185412.2776254-1-kuba@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: fix uprobe offset calculation in selftests
Andrii Nakryiko [Wed, 26 Jan 2022 19:30:58 +0000 (11:30 -0800)]
selftests/bpf: fix uprobe offset calculation in selftests

Fix how selftests determine relative offset of a function that is
uprobed. Previously, there was an assumption that uprobed function is
always in the first executable region, which is not always the case
(libbpf CI hits this case now). So get_base_addr() approach in isolation
doesn't work anymore. So teach get_uprobe_offset() to determine correct
memory mapping and calculate uprobe offset correctly.

While at it, I merged together two implementations of
get_uprobe_offset() helper, moving powerpc64-specific logic inside (had
to add extra {} block to avoid unused variable error for insn).

Also ensured that uprobed functions are never inlined, but are still
static (and thus local to each selftest), by using a no-op asm volatile
block internally. I didn't want to keep them global __weak, because some
tests use uprobe's ref counter offset (to test USDT-like logic) which is
not compatible with non-refcounted uprobe. So it's nicer to have each
test uprobe target local to the file and guaranteed to not be inlined or
skipped by the compiler (which can happen with static functions,
especially if compiling selftests with -O2).

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220126193058.3390292-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Fix a clang compilation error
Yonghong Song [Wed, 26 Jan 2022 18:19:40 +0000 (10:19 -0800)]
selftests/bpf: Fix a clang compilation error

Compiling kernel and selftests/bpf with latest llvm like blow:
  make -j LLVM=1
  make -C tools/testing/selftests/bpf -j LLVM=1
I hit the following compilation error:
  /.../prog_tests/log_buf.c:215:6: error: variable 'log_buf' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]
          if (!ASSERT_OK_PTR(raw_btf_data, "raw_btf_data_good"))
              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  /.../prog_tests/log_buf.c:264:7: note: uninitialized use occurs here
          free(log_buf);
               ^~~~~~~
  /.../prog_tests/log_buf.c:215:2: note: remove the 'if' if its condition is always false
          if (!ASSERT_OK_PTR(raw_btf_data, "raw_btf_data_good"))
          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  /.../prog_tests/log_buf.c:205:15: note: initialize the variable 'log_buf' to silence this warning
          char *log_buf;
                       ^
                        = NULL
  1 error generated.

Compiler rightfully detected that log_buf is uninitialized in one of failure path as indicated
in the above.

Proper initialization of 'log_buf' variable fixed the issue.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220126181940.4105997-1-yhs@fb.com
2 years agobpf: fix register_btf_kfunc_id_set for !CONFIG_DEBUG_INFO_BTF
Stanislav Fomichev [Wed, 26 Jan 2022 00:13:40 +0000 (16:13 -0800)]
bpf: fix register_btf_kfunc_id_set for !CONFIG_DEBUG_INFO_BTF

Commit dee872e124e8 ("bpf: Populate kfunc BTF ID sets in struct btf")
breaks loading of some modules when CONFIG_DEBUG_INFO_BTF is not set.
register_btf_kfunc_id_set returns -ENOENT to the callers when
there is no module btf. Let's return 0 (success) instead to let
those modules work in !CONFIG_DEBUG_INFO_BTF cases.

Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Fixes: dee872e124e8 ("bpf: Populate kfunc BTF ID sets in struct btf")
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/20220126001340.1573649-1-sdf@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests: bpf: Less strict size check in sockopt_sk
Felix Maurer [Tue, 25 Jan 2022 16:58:23 +0000 (17:58 +0100)]
selftests: bpf: Less strict size check in sockopt_sk

Originally, the kernel strictly checked the size of the optval in
getsockopt(TCP_ZEROCOPY_RECEIVE) to be equal to sizeof(struct
tcp_zerocopy_receive). With c8856c0514549, this was changed to allow
optvals of different sizes.

The bpf code in the sockopt_sk test was still performing the strict size
check. This fix adapts the kernel behavior from c8856c0514549 in the
selftest, i.e., just check if the required fields are there.

Fixes: 9cacf81f81611 ("bpf: Remove extra lock_sock for TCP_ZEROCOPY_RECEIVE")
Signed-off-by: Felix Maurer <fmaurer@redhat.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/r/6f569cca2e45473f9a724d54d03fdfb45f29e35f.1643129402.git.fmaurer@redhat.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge branch 'libbpf: deprecate some setter and getter APIs'
Alexei Starovoitov [Wed, 26 Jan 2022 01:59:07 +0000 (17:59 -0800)]
Merge branch 'libbpf: deprecate some setter and getter APIs'

Andrii Nakryiko says:

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

Another batch of simple deprecations. One of the last few, hopefully, as we
are getting close to deprecating all the planned APIs/features. See
individual patches for details.

v1->v2:
  - rebased on latest bpf-next, fixed Closes: reference.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoperf: use generic bpf_program__set_type() to set BPF prog type
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:54 +0000 (11:42 -0800)]
perf: use generic bpf_program__set_type() to set BPF prog type

bpf_program__set_<type>() APIs are deprecated, use generic
bpf_program__set_type() APIs instead.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-8-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agosamples/bpf: use preferred getters/setters instead of deprecated ones
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:53 +0000 (11:42 -0800)]
samples/bpf: use preferred getters/setters instead of deprecated ones

Use preferred setter and getter APIs instead of deprecated ones.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-7-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: use preferred setter/getter APIs instead of deprecated ones
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:52 +0000 (11:42 -0800)]
selftests/bpf: use preferred setter/getter APIs instead of deprecated ones

Switch to using preferred setters and getters instead of deprecated ones.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-6-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpftool: use preferred setters/getters instead of deprecated ones
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:51 +0000 (11:42 -0800)]
bpftool: use preferred setters/getters instead of deprecated ones

Use bpf_program__type() instead of discouraged bpf_program__get_type().
Also switch to bpf_map__set_max_entries() instead of bpf_map__resize().

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-5-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agolibbpf: deprecate bpf_program__is_<type>() and bpf_program__set_<type>() APIs
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:50 +0000 (11:42 -0800)]
libbpf: deprecate bpf_program__is_<type>() and bpf_program__set_<type>() APIs

Not sure why these APIs were added in the first place instead of
a completely generic (and not requiring constantly adding new APIs with
each new BPF program type) bpf_program__type() and
bpf_program__set_type() APIs. But as it is right now, there are 13 such
specialized is_type/set_type APIs, while latest kernel is already at 30+
BPF program types.

Instead of completing the set of APIs and keep chasing kernel's
bpf_prog_type enum, deprecate existing subset and recommend generic
bpf_program__type() and bpf_program__set_type() APIs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-4-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agolibbpf: deprecate bpf_map__resize()
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:49 +0000 (11:42 -0800)]
libbpf: deprecate bpf_map__resize()

Deprecated bpf_map__resize() in favor of bpf_map__set_max_entries()
setter. In addition to having a surprising name (users often don't
realize that they need to use bpf_map__resize()), the name also implies
some magic way of resizing BPF map after it is created, which is clearly
not the case.

Another minor annoyance is that bpf_map__resize() disallows 0 value for
max_entries, which in some cases is totally acceptable (e.g., like for
BPF perf buf case to let libbpf auto-create one buffer per each
available CPU core).

  [0] Closes: https://github.com/libbpf/libbpf/issues/304

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-3-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agolibbpf: hide and discourage inconsistently named getters
Andrii Nakryiko [Mon, 24 Jan 2022 19:42:48 +0000 (11:42 -0800)]
libbpf: hide and discourage inconsistently named getters

Move a bunch of "getters" into libbpf_legacy.h to keep them there in
libbpf 1.0. See [0] for discussion of "Discouraged APIs". These getters
don't add any maintenance burden and are simple alias, but they are
inconsistent in naming. So keep them in libbpf_legacy.h instead of
libbpf.h to "hide" them in favor of preferred getters ([1]). Also add two
missing getters: bpf_program__type() and bpf_program__expected_attach_type().

  [0] https://github.com/libbpf/libbpf/wiki/Libbpf:-the-road-to-v1.0#handling-deprecation-of-apis-and-functionality
  [1] Closes: https://github.com/libbpf/libbpf/issues/307

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124194254.2051434-2-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge branch 'Fix the incorrect register read for syscalls on x86_64'
Andrii Nakryiko [Tue, 25 Jan 2022 04:55:27 +0000 (20:55 -0800)]
Merge branch 'Fix the incorrect register read for syscalls on x86_64'

Kenta Tada says:

====================
Currently, rcx is read as the fourth parameter of syscall on x86_64.
But x86_64 Linux System Call convention uses r10 actually.
This commit adds the wrapper for users who want to access to
syscall params to analyze the user space.

Changelog:
----------
v1 -> v2:
- Rebase to current bpf-next
https://lore.kernel.org/bpf/20211222213924.1869758-1-andrii@kernel.org/

v2 -> v3:
- Modify the definition of SYSCALL macros for only targeted archs.
- Define __BPF_TARGET_MISSING variants for completeness.
- Remove CORE variants. These macros will not be used.
- Add a selftest.

v3 -> v4:
- Modify a selftest not to use serial tests.
- Modify a selftest to use ASSERT_EQ().
- Extract syscall wrapper for all the other tests.
- Add CORE variants.

v4 -> v5:
- Modify the CORE variant macro not to read memory directly.
- Remove the unnecessary comment.
- Add a selftest for the CORE variant.
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoselftests/bpf: Add a test to confirm PT_REGS_PARM4_SYSCALL
Kenta Tada [Mon, 24 Jan 2022 14:16:22 +0000 (23:16 +0900)]
selftests/bpf: Add a test to confirm PT_REGS_PARM4_SYSCALL

Add a selftest to verify the behavior of PT_REGS_xxx
and the CORE variant.

Signed-off-by: Kenta Tada <Kenta.Tada@sony.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220124141622.4378-4-Kenta.Tada@sony.com
2 years agolibbpf: Fix the incorrect register read for syscalls on x86_64
Kenta Tada [Mon, 24 Jan 2022 14:16:21 +0000 (23:16 +0900)]
libbpf: Fix the incorrect register read for syscalls on x86_64

Currently, rcx is read as the fourth parameter of syscall on x86_64.
But x86_64 Linux System Call convention uses r10 actually.
This commit adds the wrapper for users who want to access to
syscall params to analyze the user space.

Signed-off-by: Kenta Tada <Kenta.Tada@sony.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220124141622.4378-3-Kenta.Tada@sony.com
2 years agoselftests/bpf: Extract syscall wrapper
Kenta Tada [Mon, 24 Jan 2022 14:16:20 +0000 (23:16 +0900)]
selftests/bpf: Extract syscall wrapper

Extract the helper to set up SYS_PREFIX for fentry and kprobe selftests
that use __x86_sys_* attach functions.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Kenta Tada <Kenta.Tada@sony.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220124141622.4378-2-Kenta.Tada@sony.com
2 years agolibbpf: Mark bpf_object__open_xattr() deprecated
Christy Lee [Tue, 25 Jan 2022 01:09:17 +0000 (17:09 -0800)]
libbpf: Mark bpf_object__open_xattr() deprecated

Mark bpf_object__open_xattr() as deprecated, use
bpf_object__open_file() instead.

  [0] Closes: https://github.com/libbpf/libbpf/issues/287

Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220125010917.679975-1-christylee@fb.com
2 years agoMerge branch 'deprecate bpf_object__open_buffer() API'
Andrii Nakryiko [Tue, 25 Jan 2022 04:37:34 +0000 (20:37 -0800)]
Merge branch 'deprecate bpf_object__open_buffer() API'

Christy Lee says:

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

Deprecate bpf_object__open_buffer() API, replace all usage
with bpf_object__open_mem().

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

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoperf: Stop using bpf_object__open_buffer() API
Christy Lee [Tue, 25 Jan 2022 00:59:23 +0000 (16:59 -0800)]
perf: Stop using bpf_object__open_buffer() API

bpf_object__open_buffer() API is deprecated, use the unified opts
bpf_object__open_mem() API in perf instead. This requires at least
libbpf 0.0.6.

Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220125005923.418339-3-christylee@fb.com
2 years agolibbpf: Mark bpf_object__open_buffer() API deprecated
Christy Lee [Tue, 25 Jan 2022 00:59:22 +0000 (16:59 -0800)]
libbpf: Mark bpf_object__open_buffer() API deprecated

Deprecate bpf_object__open_buffer() API in favor of the unified
opts-based bpf_object__open_mem() API.

Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220125005923.418339-2-christylee@fb.com
2 years agoMerge branch 'Add bpf_copy_from_user_task helper and sleepable bpf iterator programs'
Alexei Starovoitov [Tue, 25 Jan 2022 03:55:41 +0000 (19:55 -0800)]
Merge branch 'Add bpf_copy_from_user_task helper and sleepable bpf iterator programs'

Kenny Yu says:

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

This patch series makes the following changes:
* Adds a new bpf helper `bpf_copy_from_user_task` to read user space
  memory from a different task.
* Adds the ability to create sleepable bpf iterator programs.

As an example of how this will be used, at Meta we are using bpf task
iterator programs and this new bpf helper to read C++ async stack traces of
a running process for debugging C++ binaries in production.

Changes since v6:
* Split first patch into two patches: first patch to add support
  for bpf iterators to use sleepable helpers, and the second to add
  the new bpf helper.
* Simplify implementation of `bpf_copy_from_user_task` based on feedback.
* Add to docs that the destination buffer will be zero-ed on error.

Changes since v5:
* Rename `bpf_access_process_vm` to `bpf_copy_from_user_task`.
* Change return value to be all-or-nothing. If we get a partial read,
  memset all bytes to 0 and return -EFAULT.
* Add to docs that the helper can only be used by sleepable BPF programs.
* Fix nits in selftests.

Changes since v4:
* Make `flags` into u64.
* Use `user_ptr` arg name to be consistent with `bpf_copy_from_user`.
* Add an extra check in selftests to verify access_process_vm calls
  succeeded.

Changes since v3:
* Check if `flags` is 0 and return -EINVAL if not.
* Rebase on latest bpf-next branch and fix merge conflicts.

Changes since v2:
* Reorder arguments in `bpf_access_process_vm` to match existing related
  bpf helpers (e.g. `bpf_probe_read_kernel`, `bpf_probe_read_user`,
  `bpf_copy_from_user`).
* `flags` argument is provided for future extensibility and is not
  currently used, and we always invoke `access_process_vm` with no flags.
* Merge bpf helper patch and `bpf_iter_run_prog` patch together for better
  bisectability in case of failures.
* Clean up formatting and comments in selftests.

Changes since v1:
* Fixed "Invalid wait context" issue in `bpf_iter_run_prog` by using
  `rcu_read_lock_trace()` for sleepable bpf iterator programs.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Add test for sleepable bpf iterator programs
Kenny Yu [Mon, 24 Jan 2022 18:54:03 +0000 (10:54 -0800)]
selftests/bpf: Add test for sleepable bpf iterator programs

This adds a test for bpf iterator programs to make use of sleepable
bpf helpers.

Signed-off-by: Kenny Yu <kennyyu@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124185403.468466-5-kennyyu@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agolibbpf: Add "iter.s" section for sleepable bpf iterator programs
Kenny Yu [Mon, 24 Jan 2022 18:54:02 +0000 (10:54 -0800)]
libbpf: Add "iter.s" section for sleepable bpf iterator programs

This adds a new bpf section "iter.s" to allow bpf iterator programs to
be sleepable.

Signed-off-by: Kenny Yu <kennyyu@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124185403.468466-4-kennyyu@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Add bpf_copy_from_user_task() helper
Kenny Yu [Mon, 24 Jan 2022 18:54:01 +0000 (10:54 -0800)]
bpf: Add bpf_copy_from_user_task() helper

This adds a helper for bpf programs to read the memory of other
tasks.

As an example use case at Meta, we are using a bpf task iterator program
and this new helper to print C++ async stack traces for all threads of
a given process.

Signed-off-by: Kenny Yu <kennyyu@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124185403.468466-3-kennyyu@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Add support for bpf iterator programs to use sleepable helpers
Kenny Yu [Mon, 24 Jan 2022 18:54:00 +0000 (10:54 -0800)]
bpf: Add support for bpf iterator programs to use sleepable helpers

This patch allows bpf iterator programs to use sleepable helpers by
changing `bpf_iter_run_prog` to use the appropriate synchronization.
With sleepable bpf iterator programs, we can no longer use
`rcu_read_lock()` and must use `rcu_read_lock_trace()` instead
to protect the bpf program.

Signed-off-by: Kenny Yu <kennyyu@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20220124185403.468466-2-kennyyu@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Jakub Kicinski [Mon, 24 Jan 2022 23:42:28 +0000 (15:42 -0800)]
Merge https://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2022-01-24

We've added 80 non-merge commits during the last 14 day(s) which contain
a total of 128 files changed, 4990 insertions(+), 895 deletions(-).

The main changes are:

1) Add XDP multi-buffer support and implement it for the mvneta driver,
   from Lorenzo Bianconi, Eelco Chaudron and Toke Høiland-Jørgensen.

2) Add unstable conntrack lookup helpers for BPF by using the BPF kfunc
   infra, from Kumar Kartikeya Dwivedi.

3) Extend BPF cgroup programs to export custom ret value to userspace via
   two helpers bpf_get_retval() and bpf_set_retval(), from YiFei Zhu.

4) Add support for AF_UNIX iterator batching, from Kuniyuki Iwashima.

5) Complete missing UAPI BPF helper description and change bpf_doc.py script
   to enforce consistent & complete helper documentation, from Usama Arif.

6) Deprecate libbpf's legacy BPF map definitions and streamline XDP APIs to
   follow tc-based APIs, from Andrii Nakryiko.

7) Support BPF_PROG_QUERY for BPF programs attached to sockmap, from Di Zhu.

8) Deprecate libbpf's bpf_map__def() API and replace users with proper getters
   and setters, from Christy Lee.

9) Extend libbpf's btf__add_btf() with an additional hashmap for strings to
   reduce overhead, from Kui-Feng Lee.

10) Fix bpftool and libbpf error handling related to libbpf's hashmap__new()
    utility function, from Mauricio Vásquez.

11) Add support to BTF program names in bpftool's program dump, from Raman Shukhau.

12) Fix resolve_btfids build to pick up host flags, from Connor O'Brien.

* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (80 commits)
  selftests, bpf: Do not yet switch to new libbpf XDP APIs
  selftests, xsk: Fix rx_full stats test
  bpf: Fix flexible_array.cocci warnings
  xdp: disable XDP_REDIRECT for xdp frags
  bpf: selftests: add CPUMAP/DEVMAP selftests for xdp frags
  bpf: selftests: introduce bpf_xdp_{load,store}_bytes selftest
  net: xdp: introduce bpf_xdp_pointer utility routine
  bpf: generalise tail call map compatibility check
  libbpf: Add SEC name for xdp frags programs
  bpf: selftests: update xdp_adjust_tail selftest to include xdp frags
  bpf: test_run: add xdp_shared_info pointer in bpf_test_finish signature
  bpf: introduce frags support to bpf_prog_test_run_xdp()
  bpf: move user_size out of bpf_test_init
  bpf: add frags support to xdp copy helpers
  bpf: add frags support to the bpf_xdp_adjust_tail() API
  bpf: introduce bpf_xdp_get_buff_len helper
  net: mvneta: enable jumbo frames if the loaded XDP program support frags
  bpf: introduce BPF_F_XDP_HAS_FRAGS flag in prog_flags loading the ebpf program
  net: mvneta: add frags support to XDP_TX
  xdp: add frags support to xdp_return_{buff/frame}
  ...
====================

Link: https://lore.kernel.org/r/20220124221235.18993-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoselftests, bpf: Do not yet switch to new libbpf XDP APIs
Daniel Borkmann [Mon, 24 Jan 2022 22:01:28 +0000 (23:01 +0100)]
selftests, bpf: Do not yet switch to new libbpf XDP APIs

Revert commit 544356524dd6 ("selftests/bpf: switch to new libbpf XDP APIs")
for now given this will heavily conflict with 4b27480dcaa7 ("bpf/selftests:
convert xdp_link test to ASSERT_* macros") upon merge. Andrii agreed to redo
the conversion cleanly after trees merged.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoMerge tag 'linux-can-fixes-for-5.17-20220124' of git://git.kernel.org/pub/scm/linux...
Jakub Kicinski [Mon, 24 Jan 2022 20:17:58 +0000 (12:17 -0800)]
Merge tag 'linux-can-fixes-for-5.17-20220124' of git://git./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2022-01-24

The first patch updates the email address of Brian Silverman from his
former employer to his private address.

The next patch fixes DT bindings information for the tcan4x5x SPI CAN
driver.

The following patch targets the m_can driver and fixes the
introduction of FIFO bulk read support.

Another patch for the tcan4x5x driver, which fixes the max register
value for the regmap config.

The last patch for the flexcan driver marks the RX mailbox support for
the MCF5441X as support.

* tag 'linux-can-fixes-for-5.17-20220124' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can:
  can: flexcan: mark RX via mailboxes as supported on MCF5441X
  can: tcan4x5x: regmap: fix max register value
  can: m_can: m_can_fifo_{read,write}: don't read or write from/to FIFO if length is 0
  dt-bindings: can: tcan4x5x: fix mram-cfg RX FIFO config
  mailmap: update email address of Brian Silverman
====================

Link: https://lore.kernel.org/r/20220124175955.3464134-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agocan: flexcan: mark RX via mailboxes as supported on MCF5441X
Marc Kleine-Budde [Fri, 21 Jan 2022 08:22:34 +0000 (09:22 +0100)]
can: flexcan: mark RX via mailboxes as supported on MCF5441X

Most flexcan IP cores support 2 RX modes:
- FIFO
- mailbox

The flexcan IP core on the MCF5441X cannot receive CAN RTR messages
via mailboxes. However the mailbox mode is more performant. The commit

1c45f5778a3b ("can: flexcan: add ethtool support to change rx-rtr setting during runtime")

added support to switch from FIFO to mailbox mode on these cores.

After testing the mailbox mode on the MCF5441X by Angelo Dureghello,
this patch marks it (without RTR capability) as supported. Further the
IP core overview table is updated, that RTR reception via mailboxes is
not supported.

Link: https://lore.kernel.org/all/20220121084425.3141218-1-mkl@pengutronix.de
Tested-by: Angelo Dureghello <angelo@kernel-space.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: tcan4x5x: regmap: fix max register value
Marc Kleine-Budde [Fri, 14 Jan 2022 17:50:54 +0000 (18:50 +0100)]
can: tcan4x5x: regmap: fix max register value

The MRAM of the tcan4x5x has a size of 2K and starts at 0x8000. There
are no further registers in the tcan4x5x making 0x87fc the biggest
addressable register.

This patch fixes the max register value of the regmap config from
0x8ffc to 0x87fc.

Fixes: 6e1caaf8ed22 ("can: tcan4x5x: fix max register value")
Link: https://lore.kernel.org/all/20220119064011.2943292-1-mkl@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: m_can: m_can_fifo_{read,write}: don't read or write from/to FIFO if length is 0
Marc Kleine-Budde [Fri, 14 Jan 2022 14:35:01 +0000 (15:35 +0100)]
can: m_can: m_can_fifo_{read,write}: don't read or write from/to FIFO if length is 0

In order to optimize FIFO access, especially on m_can cores attached
to slow busses like SPI, in patch

e39381770ec9 ("can: m_can: Disable IRQs on FIFO bus errors")

bulk read/write support has been added to the m_can_fifo_{read,write}
functions.

That change leads to the tcan driver to call
regmap_bulk_{read,write}() with a length of 0 (for CAN frames with 0
data length). regmap treats this as an error:

| tcan4x5x spi1.0 tcan4x5x0: FIFO write returned -22

This patch fixes the problem by not calling the
cdev->ops->{read,write)_fifo() in case of a 0 length read/write.

Fixes: e39381770ec9 ("can: m_can: Disable IRQs on FIFO bus errors")
Link: https://lore.kernel.org/all/20220114155751.2651888-1-mkl@pengutronix.de
Cc: stable@vger.kernel.org
Cc: Matt Kline <matt@bitbashing.io>
Cc: Chandrasekar Ramakrishnan <rcsekar@samsung.com>
Reported-by: Michael Anochin <anochin@photo-meter.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agodt-bindings: can: tcan4x5x: fix mram-cfg RX FIFO config
Marc Kleine-Budde [Fri, 14 Jan 2022 17:47:41 +0000 (18:47 +0100)]
dt-bindings: can: tcan4x5x: fix mram-cfg RX FIFO config

This tcan4x5x only comes with 2K of MRAM, a RX FIFO with a dept of 32
doesn't fit into the MRAM. Use a depth of 16 instead.

Fixes: 4edd396a1911 ("dt-bindings: can: tcan4x5x: Add DT bindings for TCAN4x5X driver")
Link: https://lore.kernel.org/all/20220119062951.2939851-1-mkl@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agomailmap: update email address of Brian Silverman
Marc Kleine-Budde [Mon, 10 Jan 2022 08:20:19 +0000 (09:20 +0100)]
mailmap: update email address of Brian Silverman

Brian Silverman's address at bluerivertech.com is not valid anymore,
use Brian's private email address instead.

Link: https://lore.kernel.org/all/20220110082359.2019735-1-mkl@pengutronix.de
Cc: Brian Silverman <bsilver16384@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agoselftests, xsk: Fix rx_full stats test
Magnus Karlsson [Fri, 21 Jan 2022 12:35:08 +0000 (13:35 +0100)]
selftests, xsk: Fix rx_full stats test

Fix the rx_full stats test so that it correctly reports pass even when
the fill ring is not full of buffers.

Fixes: 872a1184dbf2 ("selftests: xsk: Put the same buffer only once in the fill ring")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Link: https://lore.kernel.org/bpf/20220121123508.12759-1-magnus.karlsson@gmail.com
2 years agobpf: Fix flexible_array.cocci warnings
kernel test robot [Sat, 22 Jan 2022 11:09:44 +0000 (12:09 +0100)]
bpf: Fix flexible_array.cocci warnings

Zero-length and one-element arrays are deprecated, see:
Documentation/process/deprecated.rst

Flexible-array members should be used instead.

Generated by: scripts/coccinelle/misc/flexible_array.cocci

Fixes: c1ff181ffabc ("selftests/bpf: Extend kfunc selftests")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: kernel test robot <lkp@intel.com>
Signed-off-by: Julia Lawall <julia.lawall@inria.fr>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/bpf/alpine.DEB.2.22.394.2201221206320.12220@hadrien
2 years agonet: stmmac: remove unused members in struct stmmac_priv
Jisheng Zhang [Sun, 23 Jan 2022 12:27:58 +0000 (20:27 +0800)]
net: stmmac: remove unused members in struct stmmac_priv

The tx_coalesce and mii_irq are not used at all now, so remove them.

Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: atlantic: Use the bitmap API instead of hand-writing it
Christophe JAILLET [Sun, 23 Jan 2022 06:53:46 +0000 (07:53 +0100)]
net: atlantic: Use the bitmap API instead of hand-writing it

Simplify code by using bitmap_weight() and bitmap_zero() instead of
hand-writing these functions.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoping: fix the sk_bound_dev_if match in ping_lookup
Xin Long [Sat, 22 Jan 2022 11:40:56 +0000 (06:40 -0500)]
ping: fix the sk_bound_dev_if match in ping_lookup

When 'ping' changes to use PING socket instead of RAW socket by:

   # sysctl -w net.ipv4.ping_group_range="0 100"

the selftests 'router_broadcast.sh' will fail, as such command

  # ip vrf exec vrf-h1 ping -I veth0 198.51.100.255 -b

can't receive the response skb by the PING socket. It's caused by mismatch
of sk_bound_dev_if and dif in ping_rcv() when looking up the PING socket,
as dif is vrf-h1 if dif's master was set to vrf-h1.

This patch is to fix this regression by also checking the sk_bound_dev_if
against sdif so that the packets can stil be received even if the socket
is not bound to the vrf device but to the real iif.

Fixes: c319b4d76b9e ("net: ipv4: add IPPROTO_ICMP socket kind")
Reported-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet/smc: Transitional solution for clcsock race issue
Wen Gu [Sat, 22 Jan 2022 09:43:09 +0000 (17:43 +0800)]
net/smc: Transitional solution for clcsock race issue

We encountered a crash in smc_setsockopt() and it is caused by
accessing smc->clcsock after clcsock was released.

 BUG: kernel NULL pointer dereference, address: 0000000000000020
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 0 P4D 0
 Oops: 0000 [#1] PREEMPT SMP PTI
 CPU: 1 PID: 50309 Comm: nginx Kdump: loaded Tainted: G E     5.16.0-rc4+ #53
 RIP: 0010:smc_setsockopt+0x59/0x280 [smc]
 Call Trace:
  <TASK>
  __sys_setsockopt+0xfc/0x190
  __x64_sys_setsockopt+0x20/0x30
  do_syscall_64+0x34/0x90
  entry_SYSCALL_64_after_hwframe+0x44/0xae
 RIP: 0033:0x7f16ba83918e
  </TASK>

This patch tries to fix it by holding clcsock_release_lock and
checking whether clcsock has already been released before access.

In case that a crash of the same reason happens in smc_getsockopt()
or smc_switch_to_fallback(), this patch also checkes smc->clcsock
in them too. And the caller of smc_switch_to_fallback() will identify
whether fallback succeeds according to the return value.

Fixes: fd57770dd198 ("net/smc: wait for pending work before clcsock release_sock")
Link: https://lore.kernel.org/lkml/5dd7ffd1-28e2-24cc-9442-1defec27375e@linux.ibm.com/T/
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoibmvnic: remove unused ->wait_capability
Sukadev Bhattiprolu [Sat, 22 Jan 2022 02:59:21 +0000 (18:59 -0800)]
ibmvnic: remove unused ->wait_capability

With previous bug fix, ->wait_capability flag is no longer needed and can
be removed.

Fixes: 249168ad07cd ("ibmvnic: Make CRQ interrupt tasklet wait for all capabilities crqs")
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoibmvnic: don't spin in tasklet
Sukadev Bhattiprolu [Sat, 22 Jan 2022 02:59:20 +0000 (18:59 -0800)]
ibmvnic: don't spin in tasklet

ibmvnic_tasklet() continuously spins waiting for responses to all
capability requests. It does this to avoid encountering an error
during initialization of the vnic. However if there is a bug in the
VIOS and we do not receive a response to one or more queries the
tasklet ends up spinning continuously leading to hard lock ups.

If we fail to receive a message from the VIOS it is reasonable to
timeout the login attempt rather than spin indefinitely in the tasklet.

Fixes: 249168ad07cd ("ibmvnic: Make CRQ interrupt tasklet wait for all capabilities crqs")
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoibmvnic: init ->running_cap_crqs early
Sukadev Bhattiprolu [Sat, 22 Jan 2022 02:59:19 +0000 (18:59 -0800)]
ibmvnic: init ->running_cap_crqs early

We use ->running_cap_crqs to determine when the ibmvnic_tasklet() should
send out the next protocol message type. i.e when we get back responses
to all our QUERY_CAPABILITY CRQs we send out REQUEST_CAPABILITY crqs.
Similiary, when we get responses to all the REQUEST_CAPABILITY crqs, we
send out the QUERY_IP_OFFLOAD CRQ.

We currently increment ->running_cap_crqs as we send out each CRQ and
have the ibmvnic_tasklet() send out the next message type, when this
running_cap_crqs count drops to 0.

This assumes that all the CRQs of the current type were sent out before
the count drops to 0. However it is possible that we send out say 6 CRQs,
get preempted and receive all the 6 responses before we send out the
remaining CRQs. This can result in ->running_cap_crqs count dropping to
zero before all messages of the current type were sent and we end up
sending the next protocol message too early.

Instead initialize the ->running_cap_crqs upfront so the tasklet will
only send the next protocol message after all responses are received.

Use the cap_reqs local variable to also detect any discrepancy (either
now or in future) in the number of capability requests we actually send.

Currently only send_query_cap() is affected by this behavior (of sending
next message early) since it is called from the worker thread (during
reset) and from application thread (during ->ndo_open()) and they can be
preempted. send_request_cap() is only called from the tasklet  which
processes CRQ responses sequentially, is not be affected.  But to
maintain the existing symmtery with send_query_capability() we update
send_request_capability() also.

Fixes: 249168ad07cd ("ibmvnic: Make CRQ interrupt tasklet wait for all capabilities crqs")
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoibmvnic: Allow extra failures before disabling
Sukadev Bhattiprolu [Sat, 22 Jan 2022 02:59:18 +0000 (18:59 -0800)]
ibmvnic: Allow extra failures before disabling

If auto-priority-failover (APF) is enabled and there are at least two
backing devices of different priorities, some resets like fail-over,
change-param etc can cause at least two back to back failovers. (Failover
from high priority backing device to lower priority one and then back
to the higher priority one if that is still functional).

Depending on the timimg of the two failovers it is possible to trigger
a "hard" reset and for the hard reset to fail due to failovers. When this
occurs, the driver assumes that the network is unstable and disables the
VNIC for a 60-second "settling time". This in turn can cause the ethtool
command to fail with "No such device" while the vnic automatically recovers
a little while later.

Given that it's possible to have two back to back failures, allow for extra
failures before disabling the vnic for the settling time.

Fixes: f15fde9d47b8 ("ibmvnic: delay next reset if hard reset fails")
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoipv4: fix ip option filtering for locally generated fragments
Jakub Kicinski [Sat, 22 Jan 2022 00:57:31 +0000 (16:57 -0800)]
ipv4: fix ip option filtering for locally generated fragments

During IP fragmentation we sanitize IP options. This means overwriting
options which should not be copied with NOPs. Only the first fragment
has the original, full options.

ip_fraglist_prepare() copies the IP header and options from previous
fragment to the next one. Commit 19c3401a917b ("net: ipv4: place control
buffer handling away from fragmentation iterators") moved sanitizing
options before ip_fraglist_prepare() which means options are sanitized
and then overwritten again with the old values.

Fixing this is not enough, however, nor did the sanitization work
prior to aforementioned commit.

ip_options_fragment() (which does the sanitization) uses ipcb->opt.optlen
for the length of the options. ipcb->opt of fragments is not populated
(it's 0), only the head skb has the state properly built. So even when
called at the right time ip_options_fragment() does nothing. This seems
to date back all the way to v2.5.44 when the fast path for pre-fragmented
skbs had been introduced. Prior to that ip_options_build() would have been
called for every fragment (in fact ever since v2.5.44 the fragmentation
handing in ip_options_build() has been dead code, I'll clean it up in
-next).

In the original patch (see Link) caixf mentions fixing the handling
for fragments other than the second one, but I'm not sure how _any_
fragment could have had their options sanitized with the code
as it stood.

Tested with python (MTU on lo lowered to 1000 to force fragmentation):

  import socket
  s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
  s.setsockopt(socket.IPPROTO_IP, socket.IP_OPTIONS,
               bytearray([7,4,5,192, 20|0x80,4,1,0]))
  s.sendto(b'1'*2000, ('127.0.0.1', 1234))

Before:

IP (tos 0x0, ttl 64, id 1053, offset 0, flags [+], proto UDP (17), length 996, options (RR [bad length 4] [bad ptr 5] 192.148.4.1,,RA value 256))
    localhost.36500 > localhost.search-agent: UDP, length 2000
IP (tos 0x0, ttl 64, id 1053, offset 968, flags [+], proto UDP (17), length 996, options (RR [bad length 4] [bad ptr 5] 192.148.4.1,,RA value 256))
    localhost > localhost: udp
IP (tos 0x0, ttl 64, id 1053, offset 1936, flags [none], proto UDP (17), length 100, options (RR [bad length 4] [bad ptr 5] 192.148.4.1,,RA value 256))
    localhost > localhost: udp

After:

IP (tos 0x0, ttl 96, id 42549, offset 0, flags [+], proto UDP (17), length 996, options (RR [bad length 4] [bad ptr 5] 192.148.4.1,,RA value 256))
    localhost.51607 > localhost.search-agent: UDP, bad length 2000 > 960
IP (tos 0x0, ttl 96, id 42549, offset 968, flags [+], proto UDP (17), length 996, options (NOP,NOP,NOP,NOP,RA value 256))
    localhost > localhost: udp
IP (tos 0x0, ttl 96, id 42549, offset 1936, flags [none], proto UDP (17), length 100, options (NOP,NOP,NOP,NOP,RA value 256))
    localhost > localhost: udp

RA (20 | 0x80) is now copied as expected, RR (7) is "NOPed out".

Link: https://lore.kernel.org/netdev/20220107080559.122713-1-ooppublic@163.com/
Fixes: 19c3401a917b ("net: ipv4: place control buffer handling away from fragmentation iterators")
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: caixf <ooppublic@163.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet-procfs: show net devices bound packet types
Jianguo Wu [Fri, 21 Jan 2022 09:15:31 +0000 (17:15 +0800)]
net-procfs: show net devices bound packet types

After commit:7866a621043f ("dev: add per net_device packet type chains"),
we can not get packet types that are bound to a specified net device by
/proc/net/ptype, this patch fix the regression.

Run "tcpdump -i ens192 udp -nns0" Before and after apply this patch:

Before:
  [root@localhost ~]# cat /proc/net/ptype
  Type Device      Function
  0800          ip_rcv
  0806          arp_rcv
  86dd          ipv6_rcv

After:
  [root@localhost ~]# cat /proc/net/ptype
  Type Device      Function
  ALL  ens192   tpacket_rcv
  0800          ip_rcv
  0806          arp_rcv
  86dd          ipv6_rcv

v1 -> v2:
  - fix the regression rather than adding new /proc API as
    suggested by Stephen Hemminger.

Fixes: 7866a621043f ("dev: add per net_device packet type chains")
Signed-off-by: Jianguo Wu <wujianguo@chinatelecom.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>