platform/kernel/linux-rpi.git
3 years agosamples/bpf: Add missing option to xdp_sample_pkts usage
Wang Hai [Tue, 15 Jun 2021 13:57:24 +0000 (21:57 +0800)]
samples/bpf: Add missing option to xdp_sample_pkts usage

xdp_sample_pkts usage() is missing the introduction of the
"-S" option, this patch adds it.

Fixes: d50ecc46d18f ("samples/bpf: Attach XDP programs in driver mode by default")
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/20210615135724.29528-1-wanghai38@huawei.com
3 years agosamples/bpf: Add missing option to xdp_fwd usage
Wang Hai [Tue, 15 Jun 2021 13:55:54 +0000 (21:55 +0800)]
samples/bpf: Add missing option to xdp_fwd usage

xdp_fwd usage() is missing the introduction of the "-S"
and "-F" options, this patch adds it.

Fixes: d50ecc46d18f ("samples/bpf: Attach XDP programs in driver mode by default")
Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/20210615135554.29158-1-wanghai38@huawei.com
3 years agobpf: Fix typo in kernel/bpf/bpf_lsm.c
Shuyi Cheng [Wed, 16 Jun 2021 02:04:36 +0000 (10:04 +0800)]
bpf: Fix typo in kernel/bpf/bpf_lsm.c

Fix s/sleeable/sleepable/ typo in a comment.

Signed-off-by: Shuyi Cheng <chengshuyi@linux.alibaba.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1623809076-97907-1-git-send-email-chengshuyi@linux.alibaba.com
3 years agoselftests/bpf: Whitelist test_progs.h from .gitignore
Daniel Xu [Wed, 16 Jun 2021 21:52:11 +0000 (14:52 -0700)]
selftests/bpf: Whitelist test_progs.h from .gitignore

Somehow test_progs.h was being included by the existing rule:

    /test_progs*

This is bad because:

    1) test_progs.h is a checked in file
    2) grep-like tools like ripgrep[0] respect gitignore and
       test_progs.h was being hidden from searches

[0]: https://github.com/BurntSushi/ripgrep

Fixes: 74b5a5968fe8 ("selftests/bpf: Replace test_progs and test_maps w/ general rule")
Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/a46f64944bf678bc652410ca6028d3450f4f7f4b.1623880296.git.dxu@dxuuu.xyz
3 years agoMerge branch 'bpf-sock-migration'
Daniel Borkmann [Tue, 15 Jun 2021 16:01:06 +0000 (18:01 +0200)]
Merge branch 'bpf-sock-migration'

Kuniyuki Iwashima says:

====================
The SO_REUSEPORT option allows sockets to listen on the same port and to
accept connections evenly. However, there is a defect in the current
implementation [1]. When a SYN packet is received, the connection is tied
to a listening socket. Accordingly, when the listener is closed, in-flight
requests during the three-way handshake and child sockets in the accept
queue are dropped even if other listeners on the same port could accept
such connections.

This situation can happen when various server management tools restart
server (such as nginx) processes. For instance, when we change nginx
configurations and restart it, it spins up new workers that respect the
new configuration and closes all listeners on the old workers, resulting
in the in-flight ACK of 3WHS is responded by RST.

To avoid such a situation, users have to know deeply how the kernel handles
SYN packets and implement connection draining by eBPF [2]:

  1. Stop routing SYN packets to the listener by eBPF.
  2. Wait for all timers to expire to complete requests
  3. Accept connections until EAGAIN, then close the listener.

  or

  1. Start counting SYN packets and accept syscalls using the eBPF map.
  2. Stop routing SYN packets.
  3. Accept connections up to the count, then close the listener.

In either way, we cannot close a listener immediately. However, ideally,
the application need not drain the not yet accepted sockets because 3WHS
and tying a connection to a listener are just the kernel behaviour. The
root cause is within the kernel, so the issue should be addressed in kernel
space and should not be visible to user space. This patchset fixes it so
that users need not take care of kernel implementation and connection
draining. With this patchset, the kernel redistributes requests and
connections from a listener to the others in the same reuseport group
at/after close or shutdown syscalls.

Although some software does connection draining, there are still merits in
migration. For some security reasons, such as replacing TLS certificates,
we may want to apply new settings as soon as possible and/or we may not be
able to wait for connection draining. The sockets in the accept queue have
not started application sessions yet. So, if we do not drain such sockets,
they can be handled by the newer listeners and could have a longer
lifetime. It is difficult to drain all connections in every case, but we
can decrease such aborted connections by migration. In that sense,
migration is always better than draining.

Moreover, auto-migration simplifies user space logic and also works well in
a case where we cannot modify and build a server program to implement the
workaround.

Note that the source and destination listeners MUST have the same settings
at the socket API level; otherwise, applications may face inconsistency and
cause errors. In such a case, we have to use the eBPF program to select a
specific listener or to cancel migration.

Special thanks to Martin KaFai Lau for bouncing ideas and exchanging code
snippets along the way.

Link:
 [1] The SO_REUSEPORT socket option
 https://lwn.net/Articles/542629/

 [2] Re: [PATCH 1/1] net: Add SO_REUSEPORT_LISTEN_OFF socket option as drain mode
 https://lore.kernel.org/netdev/1458828813.10868.65.camel@edumazet-glaptop3.roam.corp.google.com/

Changelog:
 v8:
  * Make reuse const in reuseport_sock_index()
  * Don't use __reuseport_add_sock() in reuseport_alloc()
  * Change the arg of the second memcpy() in reuseport_grow()
  * Fix coding style to use goto in reuseport_alloc()
  * Keep sk_refcnt uninitialized in inet_reqsk_clone()
  * Initialize ireq_opt and ipv6_opt separately in reqsk_migrate_reset()

  [ This series does not include a stats patch suggested by Yuchung Cheng
    not to drop Acked-by/Reviewed-by tags and save reviewer's time. I will
    post the patch as a follow up after this series is merged. ]

 v7:
 https://lore.kernel.org/bpf/20210521182104.18273-1-kuniyu@amazon.co.jp/
  * Prevent attaching/detaching a bpf prog via shutdowned socket
  * Fix typo in commit messages
  * Split selftest into subtests

 v6:
 https://lore.kernel.org/bpf/20210517002258.75019-1-kuniyu@amazon.co.jp/
  * Change description in ip-sysctl.rst
  * Test IPPROTO_TCP before reading tfo_listener
  * Move reqsk_clone() to inet_connection_sock.c and rename to
    inet_reqsk_clone()
  * Pass req->rsk_listener to inet_csk_reqsk_queue_drop() and
    reqsk_queue_removed() in the migration path of receiving ACK
  * s/ARG_PTR_TO_SOCKET/PTR_TO_SOCKET/ in sk_reuseport_is_valid_access()
  * In selftest, use atomic ops to increment global vars, drop ACK by XDP,
    enable force fastopen, use "skel->bss" instead of "skel->data"

 v5:
 https://lore.kernel.org/bpf/20210510034433.52818-1-kuniyu@amazon.co.jp/
  * Move initializtion of sk_node from 6th to 5th patch
  * Initialize sk_refcnt in reqsk_clone()
  * Modify some definitions in reqsk_timer_handler()
  * Validate in which path/state migration happens in selftest

 v4:
 https://lore.kernel.org/bpf/20210427034623.46528-1-kuniyu@amazon.co.jp/
  * Make some functions and variables 'static' in selftest
  * Remove 'scalability' from the cover letter

 v3:
 https://lore.kernel.org/bpf/20210420154140.80034-1-kuniyu@amazon.co.jp/
  * Add sysctl back for reuseport_grow()
  * Add helper functions to manage socks[]
  * Separate migration related logic into functions: reuseport_resurrect(),
    reuseport_stop_listen_sock(), reuseport_migrate_sock()
  * Clone request_sock to be migrated
  * Migrate request one by one
  * Pass child socket to eBPF prog

 v2:
 https://lore.kernel.org/netdev/20201207132456.65472-1-kuniyu@amazon.co.jp/
  * Do not save closed sockets in socks[]
  * Revert 607904c357c61adf20b8fd18af765e501d61a385
  * Extract inet_csk_reqsk_queue_migrate() into a single patch
  * Change the spin_lock order to avoid lockdep warning
  * Add static to __reuseport_select_sock
  * Use refcount_inc_not_zero() in reuseport_select_migrated_sock()
  * Set the default attach type in bpf_prog_load_check_attach()
  * Define new proto of BPF_FUNC_get_socket_cookie
  * Fix test to be compiled successfully
  * Update commit messages

 v1:
 https://lore.kernel.org/netdev/20201201144418.35045-1-kuniyu@amazon.co.jp/
  * Remove the sysctl option
  * Enable migration if eBPF progam is not attached
  * Add expected_attach_type to check if eBPF program can migrate sockets
  * Add a field to tell migration type to eBPF program
  * Support BPF_FUNC_get_socket_cookie to get the cookie of sk
  * Allocate an empty skb if skb is NULL
  * Pass req_to_sk(req)->sk_hash because listener's hash is zero
  * Update commit messages and coverletter

 RFC:
 https://lore.kernel.org/netdev/20201117094023.3685-1-kuniyu@amazon.co.jp/
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
3 years agobpf: Test BPF_SK_REUSEPORT_SELECT_OR_MIGRATE.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:24 +0000 (21:32 +0900)]
bpf: Test BPF_SK_REUSEPORT_SELECT_OR_MIGRATE.

This patch adds a test for BPF_SK_REUSEPORT_SELECT_OR_MIGRATE and
removes 'static' from settimeo() in network_helpers.c.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-12-kuniyu@amazon.co.jp
3 years agolibbpf: Set expected_attach_type for BPF_PROG_TYPE_SK_REUSEPORT.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:23 +0000 (21:32 +0900)]
libbpf: Set expected_attach_type for BPF_PROG_TYPE_SK_REUSEPORT.

This commit introduces a new section (sk_reuseport/migrate) and sets
expected_attach_type to two each section in BPF_PROG_TYPE_SK_REUSEPORT
program.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-11-kuniyu@amazon.co.jp
3 years agobpf: Support socket migration by eBPF.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:22 +0000 (21:32 +0900)]
bpf: Support socket migration by eBPF.

This patch introduces a new bpf_attach_type for BPF_PROG_TYPE_SK_REUSEPORT
to check if the attached eBPF program is capable of migrating sockets. When
the eBPF program is attached, we run it for socket migration if the
expected_attach_type is BPF_SK_REUSEPORT_SELECT_OR_MIGRATE or
net.ipv4.tcp_migrate_req is enabled.

Currently, the expected_attach_type is not enforced for the
BPF_PROG_TYPE_SK_REUSEPORT type of program. Thus, this commit follows the
earlier idea in the commit aac3fc320d94 ("bpf: Post-hooks for sys_bind") to
fix up the zero expected_attach_type in bpf_prog_load_fixup_attach_type().

Moreover, this patch adds a new field (migrating_sk) to sk_reuseport_md to
select a new listener based on the child socket. migrating_sk varies
depending on if it is migrating a request in the accept queue or during
3WHS.

  - accept_queue : sock (ESTABLISHED/SYN_RECV)
  - 3WHS         : request_sock (NEW_SYN_RECV)

In the eBPF program, we can select a new listener by
BPF_FUNC_sk_select_reuseport(). Also, we can cancel migration by returning
SK_DROP. This feature is useful when listeners have different settings at
the socket API level or when we want to free resources as soon as possible.

  - SK_PASS with selected_sk, select it as a new listener
  - SK_PASS with selected_sk NULL, fallbacks to the random selection
  - SK_DROP, cancel the migration.

There is a noteworthy point. We select a listening socket in three places,
but we do not have struct skb at closing a listener or retransmitting a
SYN+ACK. On the other hand, some helper functions do not expect skb is NULL
(e.g. skb_header_pointer() in BPF_FUNC_skb_load_bytes(), skb_tail_pointer()
in BPF_FUNC_skb_load_bytes_relative()). So we allocate an empty skb
temporarily before running the eBPF program.

Suggested-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/netdev/20201123003828.xjpjdtk4ygl6tg6h@kafai-mbp.dhcp.thefacebook.com/
Link: https://lore.kernel.org/netdev/20201203042402.6cskdlit5f3mw4ru@kafai-mbp.dhcp.thefacebook.com/
Link: https://lore.kernel.org/netdev/20201209030903.hhow5r53l6fmozjn@kafai-mbp.dhcp.thefacebook.com/
Link: https://lore.kernel.org/bpf/20210612123224.12525-10-kuniyu@amazon.co.jp
3 years agobpf: Support BPF_FUNC_get_socket_cookie() for BPF_PROG_TYPE_SK_REUSEPORT.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:21 +0000 (21:32 +0900)]
bpf: Support BPF_FUNC_get_socket_cookie() for BPF_PROG_TYPE_SK_REUSEPORT.

We will call sock_reuseport.prog for socket migration in the next commit,
so the eBPF program has to know which listener is closing to select a new
listener.

We can currently get a unique ID of each listener in the userspace by
calling bpf_map_lookup_elem() for BPF_MAP_TYPE_REUSEPORT_SOCKARRAY map.

This patch makes the pointer of sk available in sk_reuseport_md so that we
can get the ID by BPF_FUNC_get_socket_cookie() in the eBPF program.

Suggested-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/netdev/20201119001154.kapwihc2plp4f7zc@kafai-mbp.dhcp.thefacebook.com/
Link: https://lore.kernel.org/bpf/20210612123224.12525-9-kuniyu@amazon.co.jp
3 years agotcp: Migrate TCP_NEW_SYN_RECV requests at receiving the final ACK.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:20 +0000 (21:32 +0900)]
tcp: Migrate TCP_NEW_SYN_RECV requests at receiving the final ACK.

This patch also changes the code to call reuseport_migrate_sock() and
inet_reqsk_clone(), but unlike the other cases, we do not call
inet_reqsk_clone() right after reuseport_migrate_sock().

Currently, in the receive path for TCP_NEW_SYN_RECV sockets, its listener
has three kinds of refcnt:

  (A) for listener itself
  (B) carried by reuqest_sock
  (C) sock_hold() in tcp_v[46]_rcv()

While processing the req, (A) may disappear by close(listener). Also, (B)
can disappear by accept(listener) once we put the req into the accept
queue. So, we have to hold another refcnt (C) for the listener to prevent
use-after-free.

For socket migration, we call reuseport_migrate_sock() to select a listener
with (A) and to increment the new listener's refcnt in tcp_v[46]_rcv().
This refcnt corresponds to (C) and is cleaned up later in tcp_v[46]_rcv().
Thus we have to take another refcnt (B) for the newly cloned request_sock.

In inet_csk_complete_hashdance(), we hold the count (B), clone the req, and
try to put the new req into the accept queue. By migrating req after
winning the "own_req" race, we can avoid such a worst situation:

  CPU 1 looks up req1
  CPU 2 looks up req1, unhashes it, then CPU 1 loses the race
  CPU 3 looks up req2, unhashes it, then CPU 2 loses the race
  ...

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-8-kuniyu@amazon.co.jp
3 years agotcp: Migrate TCP_NEW_SYN_RECV requests at retransmitting SYN+ACKs.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:19 +0000 (21:32 +0900)]
tcp: Migrate TCP_NEW_SYN_RECV requests at retransmitting SYN+ACKs.

As with the preceding patch, this patch changes reqsk_timer_handler() to
call reuseport_migrate_sock() and inet_reqsk_clone() to migrate in-flight
requests at retransmitting SYN+ACKs. If we can select a new listener and
clone the request, we resume setting the SYN+ACK timer for the new req. If
we can set the timer, we call inet_ehash_insert() to unhash the old req and
put the new req into ehash.

The noteworthy point here is that by unhashing the old req, another CPU
processing it may lose the "own_req" race in tcp_v[46]_syn_recv_sock() and
drop the final ACK packet. However, the new timer will recover this
situation.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-7-kuniyu@amazon.co.jp
3 years agotcp: Migrate TCP_ESTABLISHED/TCP_SYN_RECV sockets in accept queues.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:18 +0000 (21:32 +0900)]
tcp: Migrate TCP_ESTABLISHED/TCP_SYN_RECV sockets in accept queues.

When we call close() or shutdown() for listening sockets, each child socket
in the accept queue are freed at inet_csk_listen_stop(). If we can get a
new listener by reuseport_migrate_sock() and clone the request by
inet_reqsk_clone(), we try to add it into the new listener's accept queue
by inet_csk_reqsk_queue_add(). If it fails, we have to call __reqsk_free()
to call sock_put() for its listener and free the cloned request.

After putting the full socket into ehash, tcp_v[46]_syn_recv_sock() sets
NULL to ireq_opt/pktopts in struct inet_request_sock, but ipv6_opt can be
non-NULL. So, we have to set NULL to ipv6_opt of the old request to avoid
double free.

Note that we do not update req->rsk_listener and instead clone the req to
migrate because another path may reference the original request. If we
protected it by RCU, we would need to add rcu_read_lock() in many places.

Suggested-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/netdev/20201209030903.hhow5r53l6fmozjn@kafai-mbp.dhcp.thefacebook.com/
Link: https://lore.kernel.org/bpf/20210612123224.12525-6-kuniyu@amazon.co.jp
3 years agotcp: Add reuseport_migrate_sock() to select a new listener.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:17 +0000 (21:32 +0900)]
tcp: Add reuseport_migrate_sock() to select a new listener.

reuseport_migrate_sock() does the same check done in
reuseport_listen_stop_sock(). If the reuseport group is capable of
migration, reuseport_migrate_sock() selects a new listener by the child
socket hash and increments the listener's sk_refcnt beforehand. Thus, if we
fail in the migration, we have to decrement it later.

We will support migration by eBPF in the later commits.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-5-kuniyu@amazon.co.jp
3 years agotcp: Keep TCP_CLOSE sockets in the reuseport group.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:16 +0000 (21:32 +0900)]
tcp: Keep TCP_CLOSE sockets in the reuseport group.

When we close a listening socket, to migrate its connections to another
listener in the same reuseport group, we have to handle two kinds of child
sockets. One is that a listening socket has a reference to, and the other
is not.

The former is the TCP_ESTABLISHED/TCP_SYN_RECV sockets, and they are in the
accept queue of their listening socket. So we can pop them out and push
them into another listener's queue at close() or shutdown() syscalls. On
the other hand, the latter, the TCP_NEW_SYN_RECV socket is during the
three-way handshake and not in the accept queue. Thus, we cannot access
such sockets at close() or shutdown() syscalls. Accordingly, we have to
migrate immature sockets after their listening socket has been closed.

Currently, if their listening socket has been closed, TCP_NEW_SYN_RECV
sockets are freed at receiving the final ACK or retransmitting SYN+ACKs. At
that time, if we could select a new listener from the same reuseport group,
no connection would be aborted. However, we cannot do that because
reuseport_detach_sock() sets NULL to sk_reuseport_cb and forbids access to
the reuseport group from closed sockets.

This patch allows TCP_CLOSE sockets to remain in the reuseport group and
access it while any child socket references them. The point is that
reuseport_detach_sock() was called twice from inet_unhash() and
sk_destruct(). This patch replaces the first reuseport_detach_sock() with
reuseport_stop_listen_sock(), which checks if the reuseport group is
capable of migration. If capable, it decrements num_socks, moves the socket
backwards in socks[] and increments num_closed_socks. When all connections
are migrated, sk_destruct() calls reuseport_detach_sock() to remove the
socket from socks[], decrement num_closed_socks, and set NULL to
sk_reuseport_cb.

By this change, closed or shutdowned sockets can keep sk_reuseport_cb.
Consequently, calling listen() after shutdown() can cause EADDRINUSE or
EBUSY in inet_csk_bind_conflict() or reuseport_add_sock() which expects
such sockets not to have the reuseport group. Therefore, this patch also
loosens such validation rules so that a socket can listen again if it has a
reuseport group with num_closed_socks more than 0.

When such sockets listen again, we handle them in reuseport_resurrect(). If
there is an existing reuseport group (reuseport_add_sock() path), we move
the socket from the old group to the new one and free the old one if
necessary. If there is no existing group (reuseport_alloc() path), we
allocate a new reuseport group, detach sk from the old one, and free it if
necessary, not to break the current shutdown behaviour:

  - we cannot carry over the eBPF prog of shutdowned sockets
  - we cannot attach/detach an eBPF prog to/from listening sockets via
    shutdowned sockets

Note that when the number of sockets gets over U16_MAX, we try to detach a
closed socket randomly to make room for the new listening socket in
reuseport_grow().

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-4-kuniyu@amazon.co.jp
3 years agotcp: Add num_closed_socks to struct sock_reuseport.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:15 +0000 (21:32 +0900)]
tcp: Add num_closed_socks to struct sock_reuseport.

As noted in the following commit, a closed listener has to hold the
reference to the reuseport group for socket migration. This patch adds a
field (num_closed_socks) to struct sock_reuseport to manage closed sockets
within the same reuseport group. Moreover, this and the following commits
introduce some helper functions to split socks[] into two sections and keep
TCP_LISTEN and TCP_CLOSE sockets in each section. Like a double-ended
queue, we will place TCP_LISTEN sockets from the front and TCP_CLOSE
sockets from the end.

  TCP_LISTEN---------->       <-------TCP_CLOSE
  +---+---+  ---  +---+  ---  +---+  ---  +---+
  | 0 | 1 |  ...  | i |  ...  | j |  ...  | k |
  +---+---+  ---  +---+  ---  +---+  ---  +---+

  i = num_socks - 1
  j = max_socks - num_closed_socks
  k = max_socks - 1

This patch also extends reuseport_add_sock() and reuseport_grow() to
support num_closed_socks.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-3-kuniyu@amazon.co.jp
3 years agonet: Introduce net.ipv4.tcp_migrate_req.
Kuniyuki Iwashima [Sat, 12 Jun 2021 12:32:14 +0000 (21:32 +0900)]
net: Introduce net.ipv4.tcp_migrate_req.

This commit adds a new sysctl option: net.ipv4.tcp_migrate_req. If this
option is enabled or eBPF program is attached, we will be able to migrate
child sockets from a listener to another in the same reuseport group after
close() or shutdown() syscalls.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Benjamin Herrenschmidt <benh@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20210612123224.12525-2-kuniyu@amazon.co.jp
3 years agolibbpf: Set NLM_F_EXCL when creating qdisc
Kumar Kartikeya Dwivedi [Sat, 12 Jun 2021 02:35:01 +0000 (08:05 +0530)]
libbpf: Set NLM_F_EXCL when creating qdisc

This got lost during the refactoring across versions. We always use
NLM_F_EXCL when creating some TC object, so reflect what the function
says and set the flag.

Fixes: 715c5ce454a6 ("libbpf: Add low level TC-BPF management API")
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210612023502.1283837-3-memxor@gmail.com
3 years agolibbpf: Remove unneeded check for flags during tc detach
Kumar Kartikeya Dwivedi [Sat, 12 Jun 2021 02:35:00 +0000 (08:05 +0530)]
libbpf: Remove unneeded check for flags during tc detach

Coverity complained about this being unreachable code. It is right
because we already enforce flags to be unset, so a check validating
the flag value is redundant.

Fixes: 715c5ce454a6 ("libbpf: Add low level TC-BPF management API")
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210612023502.1283837-2-memxor@gmail.com
3 years agotools/bpftool: Fix error return code in do_batch()
Zhihao Cheng [Wed, 9 Jun 2021 11:59:16 +0000 (19:59 +0800)]
tools/bpftool: Fix error return code in do_batch()

Fix to return a negative error code from the error handling
case instead of 0, as done elsewhere in this function.

Fixes: 668da745af3c2 ("tools: bpftool: add support for quotations ...")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20210609115916.2186872-1-chengzhihao1@huawei.com
3 years agolibbpf: Simplify the return expression of bpf_object__init_maps function
Wang Hai [Wed, 9 Jun 2021 11:56:51 +0000 (19:56 +0800)]
libbpf: Simplify the return expression of bpf_object__init_maps function

There is no need for special treatment of the 'ret == 0' case.
This patch simplifies the return expression.

Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20210609115651.3392580-1-wanghai38@huawei.com
3 years agoselftests, bpf: Make docs tests fail more reliably
Joe Stringer [Tue, 8 Jun 2021 01:57:56 +0000 (18:57 -0700)]
selftests, bpf: Make docs tests fail more reliably

Previously, if rst2man caught errors, then these would be ignored and
the output file would be written anyway. This would allow developers to
introduce regressions in the docs comments in the BPF headers.

Additionally, even if you instruct rst2man to fail out, it will still
write out to the destination target file, so if you ran the tests twice
in a row it would always pass. Use a temporary file for the initial run
to ensure that if rst2man fails out under "--strict" mode, subsequent
runs will not automatically pass.

Tested via ./tools/testing/selftests/bpf/test_doc_build.sh

Signed-off-by: Joe Stringer <joe@cilium.io>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20210608015756.340385-1-joe@cilium.io
3 years agolibbpf: Fix pr_warn type warnings on 32bit
Michal Suchanek [Fri, 4 Jun 2021 11:24:48 +0000 (13:24 +0200)]
libbpf: Fix pr_warn type warnings on 32bit

The printed value is ptrdiff_t and is formatted wiht %ld. This works on
64bit but produces a warning on 32bit. Fix the format specifier to %td.

Fixes: 67234743736a ("libbpf: Generate loader program out of BPF ELF file.")
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20210604112448.32297-1-msuchanek@suse.de
3 years agotools/bpftool: Fix cross-build
Jean-Philippe Brucker [Thu, 3 Jun 2021 17:05:16 +0000 (19:05 +0200)]
tools/bpftool: Fix cross-build

When the bootstrap and final bpftool have different architectures, we
need to build two distinct disasm.o objects. Add a recipe for the
bootstrap disasm.o.

After commit d510296d331a ("bpftool: Use syscall/loader program in
"prog load" and "gen skeleton" command.") cross-building bpftool didn't
work anymore, because the bootstrap bpftool was linked using objects
from different architectures:

  $ make O=/tmp/bpftool ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- -C tools/bpf/bpftool/ V=1
  [...]
  aarch64-linux-gnu-gcc ... -c -MMD -o /tmp/bpftool/disasm.o /home/z/src/linux/kernel/bpf/disasm.c
  gcc ... -c -MMD -o /tmp/bpftool//bootstrap/main.o main.c
  gcc ... -o /tmp/bpftool//bootstrap/bpftool /tmp/bpftool//bootstrap/main.o ... /tmp/bpftool/disasm.o
  /usr/bin/ld: /tmp/bpftool/disasm.o: Relocations in generic ELF (EM: 183)
  /usr/bin/ld: /tmp/bpftool/disasm.o: Relocations in generic ELF (EM: 183)
  /usr/bin/ld: /tmp/bpftool/disasm.o: Relocations in generic ELF (EM: 183)
  /usr/bin/ld: /tmp/bpftool/disasm.o: error adding symbols: file in wrong format
  collect2: error: ld returned 1 exit status
  [...]

The final bpftool was built for e.g. arm64, while the bootstrap bpftool,
executed on the host, was built for x86. The problem here was that disasm.o
linked into the bootstrap bpftool was arm64 rather than x86. With the fix
we build two disasm.o, one for the target bpftool in arm64, and one for
the bootstrap bpftool in x86.

Fixes: d510296d331a ("bpftool: Use syscall/loader program in "prog load" and "gen skeleton" command.")
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210603170515.1854642-1-jean-philippe@linaro.org
3 years agoselftests/bpf: Add xdp_redirect_multi into .gitignore
Andrii Nakryiko [Thu, 3 Jun 2021 00:40:26 +0000 (17:40 -0700)]
selftests/bpf: Add xdp_redirect_multi into .gitignore

When xdp_redirect_multi test binary was added recently, it wasn't added to
.gitignore. Fix that.

Fixes: d23292476297 ("selftests/bpf: Add xdp_redirect_multi test")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210603004026.2698513-5-andrii@kernel.org
3 years agolibbpf: Install skel_internal.h header used from light skeletons
Andrii Nakryiko [Thu, 3 Jun 2021 00:40:25 +0000 (17:40 -0700)]
libbpf: Install skel_internal.h header used from light skeletons

Light skeleton code assumes skel_internal.h header to be installed system-wide
by libbpf package. Make sure it is actually installed.

Fixes: 67234743736a ("libbpf: Generate loader program out of BPF ELF file.")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210603004026.2698513-4-andrii@kernel.org
3 years agolibbpf: Refactor header installation portions of Makefile
Andrii Nakryiko [Thu, 3 Jun 2021 00:40:24 +0000 (17:40 -0700)]
libbpf: Refactor header installation portions of Makefile

As we gradually get more headers that have to be installed, it's quite
annoying to copy/paste long $(call) commands. So extract that logic and do
a simple $(foreach) over the list of headers.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210603004026.2698513-3-andrii@kernel.org
3 years agolibbpf: Move few APIs from 0.4 to 0.5 version
Andrii Nakryiko [Thu, 3 Jun 2021 00:40:23 +0000 (17:40 -0700)]
libbpf: Move few APIs from 0.4 to 0.5 version

Official libbpf 0.4 release doesn't include three APIs that were tentatively
put into 0.4 section. Fix libbpf.map and move these three APIs:

  - bpf_map__initial_value;
  - bpf_map_lookup_and_delete_elem_flags;
  - bpf_object__gen_loader.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210603004026.2698513-2-andrii@kernel.org
3 years agobpf, tnums: Provably sound, faster, and more precise algorithm for tnum_mul
Harishankar Vishwanathan [Mon, 31 May 2021 02:01:57 +0000 (22:01 -0400)]
bpf, tnums: Provably sound, faster, and more precise algorithm for tnum_mul

This patch introduces a new algorithm for multiplication of tristate
numbers (tnums) that is provably sound. It is faster and more precise when
compared to the existing method.

Like the existing method, this new algorithm follows the long
multiplication algorithm. The idea is to generate partial products by
multiplying each bit in the multiplier (tnum a) with the multiplicand
(tnum b), and adding the partial products after appropriately bit-shifting
them. The new algorithm, however, uses just a single loop over the bits of
the multiplier (tnum a) and accumulates only the uncertain components of
the multiplicand (tnum b) into a mask-only tnum. The following paper
explains the algorithm in more detail: https://arxiv.org/abs/2105.05398.

A natural way to construct the tnum product is by performing a tnum
addition on all the partial products. This algorithm presents another
method of doing this: decompose each partial product into two tnums,
consisting of the values and the masks separately. The mask-sum is
accumulated within the loop in acc_m. The value-sum tnum is generated
using a.value * b.value. The tnum constructed by tnum addition of the
value-sum and the mask-sum contains all possible summations of concrete
values drawn from the partial product tnums pairwise. We prove this result
in the paper.

Our evaluations show that the new algorithm is overall more precise
(producing tnums with less uncertain components) than the existing method.
As an illustrative example, consider the input tnums A and B. The numbers
in the parenthesis correspond to (value;mask).

  A                = 000000x1 (1;2)
  B                = 0010011x (38;1)
  A * B (existing) = xxxxxxxx (0;255)
  A * B (new)      = 0x1xxxxx (32;95)

Importantly, we present a proof of soundness of the new algorithm in the
aforementioned paper. Additionally, we show that this new algorithm is
empirically faster than the existing method.

Co-developed-by: Matan Shachnai <m.shachnai@rutgers.edu>
Co-developed-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Co-developed-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Matan Shachnai <m.shachnai@rutgers.edu>
Signed-off-by: Srinivas Narayana <srinivas.narayana@rutgers.edu>
Signed-off-by: Santosh Nagarakatte <santosh.nagarakatte@rutgers.edu>
Signed-off-by: Harishankar Vishwanathan <harishankar.vishwanathan@rutgers.edu>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Edward Cree <ecree.xilinx@gmail.com>
Link: https://arxiv.org/abs/2105.05398
Link: https://lore.kernel.org/bpf/20210531020157.7386-1-harishankar.vishwanathan@rutgers.edu
3 years agobpf, devmap: Remove drops variable from bq_xmit_all()
Hangbin Liu [Fri, 28 May 2021 02:43:56 +0000 (22:43 -0400)]
bpf, devmap: Remove drops variable from bq_xmit_all()

As Colin pointed out, the first drops assignment after declaration will
be overwritten by the second drops assignment before using, which makes
it useless.

Since the drops variable will be used only once. Just remove it and
use "cnt - sent" in trace_xdp_devmap_xmit().

Fixes: cb261b594b41 ("bpf: Run devmap xdp_prog on flush instead of bulk enqueue")
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20210528024356.24333-1-liuhangbin@gmail.com
3 years agobpf, docs: Add llvm_reloc.rst to explain llvm bpf relocations
Yonghong Song [Wed, 26 May 2021 15:24:57 +0000 (08:24 -0700)]
bpf, docs: Add llvm_reloc.rst to explain llvm bpf relocations

LLVM upstream commit https://reviews.llvm.org/D102712 made some changes
to bpf relocations to make them llvm linker lld friendly. The scope of
existing relocations R_BPF_64_{64,32} is narrowed and new relocations
R_BPF_64_{ABS32,ABS64,NODYLD32} are introduced.

Let us add some documentation about llvm bpf relocations so people can
understand how to resolve them properly in their respective tools.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20210526152457.335210-1-yhs@fb.com
3 years agolibbpf: Move BPF_SEQ_PRINTF and BPF_SNPRINTF to bpf_helpers.h
Florent Revest [Wed, 26 May 2021 16:46:43 +0000 (18:46 +0200)]
libbpf: Move BPF_SEQ_PRINTF and BPF_SNPRINTF to bpf_helpers.h

These macros are convenient wrappers around the bpf_seq_printf and
bpf_snprintf helpers. They are currently provided by bpf_tracing.h which
targets low level tracing primitives. bpf_helpers.h is a better fit.

The __bpf_narg and __bpf_apply are needed in both files and provided
twice. __bpf_empty isn't used anywhere and is removed from bpf_tracing.h

Reported-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Florent Revest <revest@chromium.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210526164643.2881368-1-revest@chromium.org
3 years agoMerge branch 'bpf-xdp-bcast'
Daniel Borkmann [Wed, 26 May 2021 07:46:17 +0000 (09:46 +0200)]
Merge branch 'bpf-xdp-bcast'

Hangbin Liu says:

====================
This patchset is a new implementation for XDP multicast support based
on my previous 2 maps implementation[1]. The reason is that Daniel thinks
the exclude map implementation is missing proper bond support in XDP
context. And there is a plan to add native XDP bonding support. Adding
a exclude map in the helper also increases the complexity of verifier and
has drawbacks on performance.

The new implementation just add two new flags BPF_F_BROADCAST and
BPF_F_EXCLUDE_INGRESS to extend xdp_redirect_map for broadcast support.

With BPF_F_BROADCAST the packet will be broadcasted to all the interfaces
in the map. with BPF_F_EXCLUDE_INGRESS the ingress interface will be
excluded when do broadcasting.

The patchv11 link is here [2].

  [1] https://lore.kernel.org/bpf/20210223125809.1376577-1-liuhangbin@gmail.com
  [2] https://lore.kernel.org/bpf/20210513070447.1878448-1-liuhangbin@gmail.com

v12: As Daniel pointed out:
  a) defined as const u64 for flag_mask and action_mask in
     __bpf_xdp_redirect_map()
  b) remove BPF_F_ACTION_MASK in uapi header
  c) remove EXPORT_SYMBOL_GPL for xdpf_clone()

v11:
  a) Use unlikely() when checking if this is for broadcast redirecting.
  b) Fix a tracepoint NULL pointer issue Jesper found
  c) Remove BPF_F_REDIR_MASK and just use OR flags to make the reader more
     clear about what's flags we are using
  d) Add the performace number with multi veth interfaces in patch 01
     description.
  e) remove some sleeps to reduce the testing time in patch04. Re-struct the
     test and make clear what flags we are testing.

v10: use READ/WRITE_ONCE when read/write map instead of xchg()
v9: Update patch 01 commit description
v8: use hlist_for_each_entry_rcu() when looping the devmap hash ojbs
v7: No need to free xdpf in dev_map_enqueue_clone() if xdpf_clone failed.
v6: Fix a skb leak in the error path for generic XDP
v5: Just walk the map directly to get interfaces as get_next_key() of devmap
    hash may restart looping from the first key if the device get removed.
    After update the performace has improved 10% compired with v4.
v4: Fix flags never cleared issue in patch 02. Update selftest to cover this.
v3: Rebase the code based on latest bpf-next
v2: fix flag renaming issue in patch 02
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
3 years agoselftests/bpf: Add xdp_redirect_multi test
Hangbin Liu [Wed, 19 May 2021 09:07:47 +0000 (17:07 +0800)]
selftests/bpf: Add xdp_redirect_multi test

Add a bpf selftest for new helper xdp_redirect_map_multi(). In this
test there are 3 forward groups and 1 exclude group. The test will
redirect each interface's packets to all the interfaces in the forward
group, and exclude the interface in exclude map.

Two maps (DEVMAP, DEVMAP_HASH) and two xdp modes (generic, drive) will
be tested. XDP egress program will also be tested by setting pkt src MAC
to egress interface's MAC address.

For more test details, you can find it in the test script. Here is
the test result.
]# time ./test_xdp_redirect_multi.sh
Pass: xdpgeneric arp(F_BROADCAST) ns1-1
Pass: xdpgeneric arp(F_BROADCAST) ns1-2
Pass: xdpgeneric arp(F_BROADCAST) ns1-3
Pass: xdpgeneric IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-1
Pass: xdpgeneric IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-2
Pass: xdpgeneric IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-3
Pass: xdpgeneric IPv6 (no flags) ns1-1
Pass: xdpgeneric IPv6 (no flags) ns1-2
Pass: xdpdrv arp(F_BROADCAST) ns1-1
Pass: xdpdrv arp(F_BROADCAST) ns1-2
Pass: xdpdrv arp(F_BROADCAST) ns1-3
Pass: xdpdrv IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-1
Pass: xdpdrv IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-2
Pass: xdpdrv IPv4 (F_BROADCAST|F_EXCLUDE_INGRESS) ns1-3
Pass: xdpdrv IPv6 (no flags) ns1-1
Pass: xdpdrv IPv6 (no flags) ns1-2
Pass: xdpegress mac ns1-2
Pass: xdpegress mac ns1-3
Summary: PASS 18, FAIL 0

real    1m18.321s
user    0m0.123s
sys     0m0.350s

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210519090747.1655268-5-liuhangbin@gmail.com
3 years agosample/bpf: Add xdp_redirect_map_multi for redirect_map broadcast test
Hangbin Liu [Wed, 19 May 2021 09:07:46 +0000 (17:07 +0800)]
sample/bpf: Add xdp_redirect_map_multi for redirect_map broadcast test

This is a sample for xdp redirect broadcast. In the sample we could forward
all packets between given interfaces. There is also an option -X that could
enable 2nd xdp_prog on egress interface.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210519090747.1655268-4-liuhangbin@gmail.com
3 years agoxdp: Extend xdp_redirect_map with broadcast support
Hangbin Liu [Wed, 19 May 2021 09:07:45 +0000 (17:07 +0800)]
xdp: Extend xdp_redirect_map with broadcast support

This patch adds two flags BPF_F_BROADCAST and BPF_F_EXCLUDE_INGRESS to
extend xdp_redirect_map for broadcast support.

With BPF_F_BROADCAST the packet will be broadcasted to all the interfaces
in the map. with BPF_F_EXCLUDE_INGRESS the ingress interface will be
excluded when do broadcasting.

When getting the devices in dev hash map via dev_map_hash_get_next_key(),
there is a possibility that we fall back to the first key when a device
was removed. This will duplicate packets on some interfaces. So just walk
the whole buckets to avoid this issue. For dev array map, we also walk the
whole map to find valid interfaces.

Function bpf_clear_redirect_map() was removed in
commit ee75aef23afe ("bpf, xdp: Restructure redirect actions").
Add it back as we need to use ri->map again.

With test topology:
  +-------------------+             +-------------------+
  | Host A (i40e 10G) |  ---------- | eno1(i40e 10G)    |
  +-------------------+             |                   |
                                    |   Host B          |
  +-------------------+             |                   |
  | Host C (i40e 10G) |  ---------- | eno2(i40e 10G)    |
  +-------------------+             |                   |
                                    |          +------+ |
                                    | veth0 -- | Peer | |
                                    | veth1 -- |      | |
                                    | veth2 -- |  NS  | |
                                    |          +------+ |
                                    +-------------------+

On Host A:
 # pktgen/pktgen_sample03_burst_single_flow.sh -i eno1 -d $dst_ip -m $dst_mac -s 64

On Host B(Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz, 128G Memory):
Use xdp_redirect_map and xdp_redirect_map_multi in samples/bpf for testing.
All the veth peers in the NS have a XDP_DROP program loaded. The
forward_map max_entries in xdp_redirect_map_multi is modify to 4.

Testing the performance impact on the regular xdp_redirect path with and
without patch (to check impact of additional check for broadcast mode):

5.12 rc4         | redirect_map        i40e->i40e      |    2.0M |  9.7M
5.12 rc4         | redirect_map        i40e->veth      |    1.7M | 11.8M
5.12 rc4 + patch | redirect_map        i40e->i40e      |    2.0M |  9.6M
5.12 rc4 + patch | redirect_map        i40e->veth      |    1.7M | 11.7M

Testing the performance when cloning packets with the redirect_map_multi
test, using a redirect map size of 4, filled with 1-3 devices:

5.12 rc4 + patch | redirect_map multi  i40e->veth (x1) |    1.7M | 11.4M
5.12 rc4 + patch | redirect_map multi  i40e->veth (x2) |    1.1M |  4.3M
5.12 rc4 + patch | redirect_map multi  i40e->veth (x3) |    0.8M |  2.6M

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/20210519090747.1655268-3-liuhangbin@gmail.com
3 years agobpf: Run devmap xdp_prog on flush instead of bulk enqueue
Jesper Dangaard Brouer [Wed, 19 May 2021 09:07:44 +0000 (17:07 +0800)]
bpf: Run devmap xdp_prog on flush instead of bulk enqueue

This changes the devmap XDP program support to run the program when the
bulk queue is flushed instead of before the frame is enqueued. This has
a couple of benefits:

- It "sorts" the packets by destination devmap entry, and then runs the
  same BPF program on all the packets in sequence. This ensures that we
  keep the XDP program and destination device properties hot in I-cache.

- It makes the multicast implementation simpler because it can just
  enqueue packets using bq_enqueue() without having to deal with the
  devmap program at all.

The drawback is that if the devmap program drops the packet, the enqueue
step is redundant. However, arguably this is mostly visible in a
micro-benchmark, and with more mixed traffic the I-cache benefit should
win out. The performance impact of just this patch is as follows:

Using 2 10Gb i40e NIC, redirecting one to another, or into a veth interface,
which do XDP_DROP on veth peer. With xdp_redirect_map in sample/bpf, send
pkts via pktgen cmd:
./pktgen_sample03_burst_single_flow.sh -i eno1 -d $dst_ip -m $dst_mac -t 10 -s 64

There are about +/- 0.1M deviation for native testing, the performance
improved for the base-case, but some drop back with xdp devmap prog attached.

Version          | Test                           | Generic | Native | Native + 2nd xdp_prog
5.12 rc4         | xdp_redirect_map   i40e->i40e  |    1.9M |   9.6M |  8.4M
5.12 rc4         | xdp_redirect_map   i40e->veth  |    1.7M |  11.7M |  9.8M
5.12 rc4 + patch | xdp_redirect_map   i40e->i40e  |    1.9M |   9.8M |  8.0M
5.12 rc4 + patch | xdp_redirect_map   i40e->veth  |    1.7M |  12.0M |  9.4M

When bq_xmit_all() is called from bq_enqueue(), another packet will
always be enqueued immediately after, so clearing dev_rx, xdp_prog and
flush_node in bq_xmit_all() is redundant. Move the clear to __dev_flush(),
and only check them once in bq_enqueue() since they are all modified
together.

This change also has the side effect of extending the lifetime of the
RCU-protected xdp_prog that lives inside the devmap entries: Instead of
just living for the duration of the XDP program invocation, the
reference now lives all the way until the bq is flushed. This is safe
because the bq flush happens at the end of the NAPI poll loop, so
everything happens between a local_bh_disable()/local_bh_enable() pair.
However, this is by no means obvious from looking at the call sites; in
particular, some drivers have an additional rcu_read_lock() around only
the XDP program invocation, which only confuses matters further.
Cleaning this up will be done in a separate patch series.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20210519090747.1655268-2-liuhangbin@gmail.com
3 years agoMerge branch 'libbpf: error reporting changes for v1.0'
Alexei Starovoitov [Wed, 26 May 2021 00:32:35 +0000 (17:32 -0700)]
Merge branch 'libbpf: error reporting changes for v1.0'

Andrii Nakryiko says:

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

Implement error reporting changes discussed in "Libbpf: the road to v1.0"
([0]) document.

Libbpf gets a new API, libbpf_set_strict_mode() which accepts a set of flags
that turn on a set of libbpf 1.0 changes, that might be potentially breaking.
It's possible to opt-in into all current and future 1.0 features by specifying
LIBBPF_STRICT_ALL flag.

When some of the 1.0 "features" are requested, libbpf APIs might behave
differently. In this patch set a first set of changes are implemented, all
related to the way libbpf returns errors. See individual patches for details.

Patch #1 adds a no-op libbpf_set_strict_mode() functionality to enable
updating selftests.

Patch #2 gets rid of all the bad code patterns that will break in libbpf 1.0
(exact -1 comparison for low-level APIs, direct IS_ERR() macro usage to check
pointer-returning APIs for error, etc). These changes make selftest work in
both legacy and 1.0 libbpf modes. Selftests also opt-in into 100% libbpf 1.0
mode to automatically gain all the subsequent changes, which will come in
follow up patches.

Patch #3 streamlines error reporting for low-level APIs wrapping bpf() syscall.

Patch #4 streamlines errors for all the rest APIs.

Patch #5 ensures that BPF skeletons propagate errors properly as well, as
currently on error some APIs will return NULL with no way of checking exact
error code.

  [0] https://docs.google.com/document/d/1UyjTZuPFWiPFyKk1tV5an11_iaRuec6U-ZESZ54nNTY

v1->v2:
  - move libbpf_set_strict_mode() implementation to patch #1, where it belongs
    (Alexei);
  - add acks, slight rewording of commit messages.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agobpftool: Set errno on skeleton failures and propagate errors
Andrii Nakryiko [Tue, 25 May 2021 03:59:35 +0000 (20:59 -0700)]
bpftool: Set errno on skeleton failures and propagate errors

Follow libbpf's error handling conventions and pass through errors and errno
properly. Skeleton code always returned NULL on errors (not ERR_PTR(err)), so
there are no backwards compatibility concerns. But now we also set errno
properly, so it's possible to distinguish different reasons for failure, if
necessary.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210525035935.1461796-6-andrii@kernel.org
3 years agolibbpf: Streamline error reporting for high-level APIs
Andrii Nakryiko [Tue, 25 May 2021 03:59:34 +0000 (20:59 -0700)]
libbpf: Streamline error reporting for high-level APIs

Implement changes to error reporting for high-level libbpf APIs to make them
less surprising and less error-prone to users:
  - in all the cases when error happens, errno is set to an appropriate error
    value;
  - in libbpf 1.0 mode, all pointer-returning APIs return NULL on error and
    error code is communicated through errno; this applies both to APIs that
    already returned NULL before (so now they communicate more detailed error
    codes), as well as for many APIs that used ERR_PTR() macro and encoded
    error numbers as fake pointers.
  - in legacy (default) mode, those APIs that were returning ERR_PTR(err),
    continue doing so, but still set errno.

With these changes, errno can be always used to extract actual error,
regardless of legacy or libbpf 1.0 modes. This is utilized internally in
libbpf in places where libbpf uses it's own high-level APIs.
libbpf_get_error() is adapted to handle both cases completely transparently to
end-users (and is used by libbpf consistently as well).

More context, justification, and discussion can be found in "Libbpf: the road
to v1.0" document ([0]).

  [0] https://docs.google.com/document/d/1UyjTZuPFWiPFyKk1tV5an11_iaRuec6U-ZESZ54nNTY

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210525035935.1461796-5-andrii@kernel.org
3 years agolibbpf: Streamline error reporting for low-level APIs
Andrii Nakryiko [Tue, 25 May 2021 03:59:33 +0000 (20:59 -0700)]
libbpf: Streamline error reporting for low-level APIs

Ensure that low-level APIs behave uniformly across the libbpf as follows:
  - in case of an error, errno is always set to the correct error code;
  - when libbpf 1.0 mode is enabled with LIBBPF_STRICT_DIRECT_ERRS option to
    libbpf_set_strict_mode(), return -Exxx error value directly, instead of -1;
  - by default, until libbpf 1.0 is released, keep returning -1 directly.

More context, justification, and discussion can be found in "Libbpf: the road
to v1.0" document ([0]).

  [0] https://docs.google.com/document/d/1UyjTZuPFWiPFyKk1tV5an11_iaRuec6U-ZESZ54nNTY

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210525035935.1461796-4-andrii@kernel.org
3 years agoselftests/bpf: Turn on libbpf 1.0 mode and fix all IS_ERR checks
Andrii Nakryiko [Tue, 25 May 2021 03:59:32 +0000 (20:59 -0700)]
selftests/bpf: Turn on libbpf 1.0 mode and fix all IS_ERR checks

Turn ony libbpf 1.0 mode. Fix all the explicit IS_ERR checks that now will be
broken because libbpf returns NULL on error (and sets errno). Fix
ASSERT_OK_PTR and ASSERT_ERR_PTR to work for both old mode and new modes and
use them throughout selftests. This is trivial to do by using
libbpf_get_error() API that all libbpf users are supposed to use, instead of
IS_ERR checks.

A bunch of checks also did explicit -1 comparison for various fd-returning
APIs. Such checks are replaced with >= 0 or < 0 cases.

There were also few misuses of bpf_object__find_map_by_name() in test_maps.
Those are fixed in this patch as well.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210525035935.1461796-3-andrii@kernel.org
3 years agolibbpf: Add libbpf_set_strict_mode() API to turn on libbpf 1.0 behaviors
Andrii Nakryiko [Tue, 25 May 2021 03:59:31 +0000 (20:59 -0700)]
libbpf: Add libbpf_set_strict_mode() API to turn on libbpf 1.0 behaviors

Add libbpf_set_strict_mode() API that allows application to simulate libbpf
1.0 breaking changes before libbpf 1.0 is released. This will help users
migrate gradually and with confidence.

For now only ALL or NONE options are available, subsequent patches will add
more flags. This patch is preliminary for selftests/bpf changes.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210525035935.1461796-2-andrii@kernel.org
3 years agoxsk: Use kvcalloc to support large umems
Magnus Karlsson [Fri, 21 May 2021 08:33:01 +0000 (10:33 +0200)]
xsk: Use kvcalloc to support large umems

Use kvcalloc() instead of kcalloc() to support large umems with, on my
server, one million pages or more in the umem.

Reported-by: Dan Siemon <dan@coverfire.com>
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Björn Töpel <bjorn@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20210521083301.26921-1-magnus.karlsson@gmail.com
3 years agobpf: Fix spelling mistakes
Zhen Lei [Tue, 25 May 2021 02:56:59 +0000 (10:56 +0800)]
bpf: Fix spelling mistakes

Fix some spelling mistakes in comments:
aother ==> another
Netiher ==> Neither
desribe ==> describe
intializing ==> initializing
funciton ==> function
wont ==> won't and move the word 'the' at the end to the next line
accross ==> across
pathes ==> paths
triggerred ==> triggered
excute ==> execute
ether ==> either
conervative ==> conservative
convetion ==> convention
markes ==> marks
interpeter ==> interpreter

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210525025659.8898-2-thunder.leizhen@huawei.com
3 years agosamples: bpf: Ix kernel-doc syntax in file header
Aditya Srivastava [Sun, 23 May 2021 15:14:08 +0000 (20:44 +0530)]
samples: bpf: Ix kernel-doc syntax in file header

The opening comment mark '/**' is used for highlighting the beginning of
kernel-doc comments.
The header for samples/bpf/ibumad_kern.c follows this syntax, but
the content inside does not comply with kernel-doc.

This line was probably not meant for kernel-doc parsing, but is parsed
due to the presence of kernel-doc like comment syntax(i.e, '/**'), which
causes unexpected warnings from kernel-doc:
warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
 * ibumad BPF sample kernel side

Provide a simple fix by replacing this occurrence with general comment
format, i.e. '/*', to prevent kernel-doc from parsing it.

Signed-off-by: Aditya Srivastava <yashsri421@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://lore.kernel.org/bpf/20210523151408.22280-1-yashsri421@gmail.com
3 years agolibbpf: Add support for new llvm bpf relocations
Yonghong Song [Sat, 22 May 2021 16:23:41 +0000 (09:23 -0700)]
libbpf: Add support for new llvm bpf relocations

LLVM patch https://reviews.llvm.org/D102712
narrowed the scope of existing R_BPF_64_64
and R_BPF_64_32 relocations, and added three
new relocations, R_BPF_64_ABS64, R_BPF_64_ABS32
and R_BPF_64_NODYLD32. The main motivation is
to make relocations linker friendly.

This change, unfortunately, breaks libbpf build,
and we will see errors like below:
  libbpf: ELF relo #0 in section #6 has unexpected type 2 in
     /home/yhs/work/bpf-next/tools/testing/selftests/bpf/bpf_tcp_nogpl.o
  Error: failed to link
     '/home/yhs/work/bpf-next/tools/testing/selftests/bpf/bpf_tcp_nogpl.o':
     Unknown error -22 (-22)
The new relocation R_BPF_64_ABS64 is generated
and libbpf linker sanity check doesn't understand it.
Relocation section '.rel.struct_ops' at offset 0x1410 contains 1 entries:
    Offset             Info             Type               Symbol's Value  Symbol's Name
0000000000000018  0000000700000002 R_BPF_64_ABS64         0000000000000000 nogpltcp_init

Look at the selftests/bpf/bpf_tcp_nogpl.c,
  void BPF_STRUCT_OPS(nogpltcp_init, struct sock *sk)
  {
  }

  SEC(".struct_ops")
  struct tcp_congestion_ops bpf_nogpltcp = {
          .init           = (void *)nogpltcp_init,
          .name           = "bpf_nogpltcp",
  };
The new llvm relocation scheme categorizes 'nogpltcp_init' reference
as R_BPF_64_ABS64 instead of R_BPF_64_64 which is used to specify
ld_imm64 relocation in the new scheme.

Let us fix the linker sanity checking by including
R_BPF_64_ABS64 and R_BPF_64_ABS32. There is no need to
check R_BPF_64_NODYLD32 which is used for .BTF and .BTF.ext.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20210522162341.3687617-1-yhs@fb.com
3 years agoMerge branch 'Add lookup_and_delete_elem support to BPF hash map types'
Andrii Nakryiko [Mon, 24 May 2021 20:30:53 +0000 (13:30 -0700)]
Merge branch 'Add lookup_and_delete_elem support to BPF hash map types'

Denis Salopek says:

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

This patch series extends the existing bpf_map_lookup_and_delete_elem()
functionality with 4 more map types:
 - BPF_MAP_TYPE_HASH,
 - BPF_MAP_TYPE_PERCPU_HASH,
 - BPF_MAP_TYPE_LRU_HASH and
 - BPF_MAP_TYPE_LRU_PERCPU_HASH.

Patch 1 adds most of its functionality and logic as well as
documentation.

As it was previously limited to only stacks and queues which do not
support the BPF_F_LOCK flag, patch 2 enables its usage by adding a new
libbpf API bpf_map_lookup_and_delete_elem_flags() based on the existing
bpf_map_lookup_elem_flags().

Patch 3 adds selftests for lookup_and_delete_elem().

Changes in patch 1:
v7: Minor formating nits, add Acked-by.
v6: Remove unneeded flag check, minor code/format fixes.
v5: Split patch to 3 patches. Extend BPF_MAP_LOOKUP_AND_DELETE_ELEM
documentation with this changes.
v4: Fix the return value for unsupported map types.
v3: Add bpf_map_lookup_and_delete_elem_flags() and enable BPF_F_LOCK
flag, change CHECKs to ASSERT_OKs, initialize variables to 0.
v2: Add functionality for LRU/per-CPU, add test_progs tests.

Changes in patch 2:
v7: No change.
v6: Add Acked-by.
v5: Move to the newest libbpf version (0.4.0).

Changes in patch 3:
v7: Remove ASSERT_GE macro which is already added in some other commit,
change ASSERT_OK to ASSERT_OK_PTR, add Acked-by.
v6: Remove PERCPU macros, add ASSERT_GE macro to test_progs.h, remove
leftover code.
v5: Use more appropriate macros. Better check for changed value.
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
3 years agoselftests/bpf: Add bpf_lookup_and_delete_elem tests
Denis Salopek [Tue, 11 May 2021 21:00:06 +0000 (23:00 +0200)]
selftests/bpf: Add bpf_lookup_and_delete_elem tests

Add bpf selftests and extend existing ones for a new function
bpf_lookup_and_delete_elem() for (percpu) hash and (percpu) LRU hash map
types.
In test_lru_map and test_maps we add an element, lookup_and_delete it,
then check whether it's deleted.
The newly added lookup_and_delete prog tests practically do the same
thing but additionally use a BPF program to change the value of the
element for LRU maps.

Signed-off-by: Denis Salopek <denis.salopek@sartura.hr>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/d30d3e0060c1f750e133579623cf1c60ff58f3d9.1620763117.git.denis.salopek@sartura.hr
3 years agobpf: Extend libbpf with bpf_map_lookup_and_delete_elem_flags
Denis Salopek [Tue, 11 May 2021 21:00:05 +0000 (23:00 +0200)]
bpf: Extend libbpf with bpf_map_lookup_and_delete_elem_flags

Add bpf_map_lookup_and_delete_elem_flags() libbpf API in order to use
the BPF_F_LOCK flag with the map_lookup_and_delete_elem() function.

Signed-off-by: Denis Salopek <denis.salopek@sartura.hr>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/15b05dafe46c7e0750d110f233977372029d1f62.1620763117.git.denis.salopek@sartura.hr
3 years agobpf: Add lookup_and_delete_elem support to hashtab
Denis Salopek [Tue, 11 May 2021 21:00:04 +0000 (23:00 +0200)]
bpf: Add lookup_and_delete_elem support to hashtab

Extend the existing bpf_map_lookup_and_delete_elem() functionality to
hashtab map types, in addition to stacks and queues.
Create a new hashtab bpf_map_ops function that does lookup and deletion
of the element under the same bucket lock and add the created map_ops to
bpf.h.

Signed-off-by: Denis Salopek <denis.salopek@sartura.hr>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/4d18480a3e990ffbf14751ddef0325eed3be2966.1620763117.git.denis.salopek@sartura.hr
3 years agolibbpf: Skip bpf_object__probe_loading for light skeleton
Stanislav Fomichev [Fri, 21 May 2021 03:06:53 +0000 (20:06 -0700)]
libbpf: Skip bpf_object__probe_loading for light skeleton

I'm getting the following error when running 'gen skeleton -L' as
regular user:

libbpf: Error in bpf_object__probe_loading():Operation not permitted(1).
Couldn't load trivial BPF program. Make sure your kernel supports BPF
(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is set to big enough
value.

Fixes: 67234743736a ("libbpf: Generate loader program out of BPF ELF file.")
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210521030653.2626513-1-sdf@google.com
3 years agoethernet: ucc_geth: Use kmemdup() rather than kmalloc+memcpy
YueHaibing [Mon, 24 May 2021 01:07:01 +0000 (09:07 +0800)]
ethernet: ucc_geth: Use kmemdup() rather than kmalloc+memcpy

Issue identified with Coccinelle.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: r6040: Allow restarting auto-negotiation
Florian Fainelli [Sun, 23 May 2021 15:58:42 +0000 (08:58 -0700)]
net: r6040: Allow restarting auto-negotiation

Use phy_ethtool_nway_reset() since the driver makes use of the PHY
library.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'r6040-cleanups'
David S. Miller [Mon, 24 May 2021 00:20:53 +0000 (17:20 -0700)]
Merge branch 'r6040-cleanups'

Florian Fainelli says:

====================
net: r6040: Non-functional changes

These two patches clean up the r6040 driver a little bit in preparation
for adding additional features such as dumping MAC counters and properly
dealing with DMA-API mapping.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: r6040: Use ETH_FCS_LEN
Florian Fainelli [Sun, 23 May 2021 15:54:11 +0000 (08:54 -0700)]
net: r6040: Use ETH_FCS_LEN

Instead of the open coded constant 4.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: r6040: Use logical or for MDIO operations
Florian Fainelli [Sun, 23 May 2021 15:54:10 +0000 (08:54 -0700)]
net: r6040: Use logical or for MDIO operations

This is not a functional change, but we should be using a logical or to
assign the bits we will be writing to the MDIO read and write registers.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoehea: Use DEVICE_ATTR_*() macro
YueHaibing [Sun, 23 May 2021 06:02:23 +0000 (14:02 +0800)]
ehea: Use DEVICE_ATTR_*() macro

Use DEVICE_ATTR_*() helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agosfc: falcon: use DEVICE_ATTR_*() macro
YueHaibing [Sun, 23 May 2021 03:24:09 +0000 (11:24 +0800)]
sfc: falcon: use DEVICE_ATTR_*() macro

Use DEVICE_ATTR_*() helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agosfc: use DEVICE_ATTR_*() macro
YueHaibing [Sun, 23 May 2021 03:20:30 +0000 (11:20 +0800)]
sfc: use DEVICE_ATTR_*() macro

Use DEVICE_ATTR_*() helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ftgmac100: add missing error return code in ftgmac100_probe()
Yang Yingliang [Sat, 22 May 2021 12:02:46 +0000 (20:02 +0800)]
net: ftgmac100: add missing error return code in ftgmac100_probe()

The variables will be free on path err_phy_connect, it should
return error code, or it will cause double free when calling
ftgmac100_remove().

Fixes: bd466c3fb5a4 ("net/faraday: Support NCSI mode")
Fixes: 39bfab8844a0 ("net: ftgmac100: Add support for DT phy-handle property")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: don't print error from dpaa2_mac_connect if that's EPROBE_DEFER
Vladimir Oltean [Fri, 21 May 2021 14:12:20 +0000 (17:12 +0300)]
dpaa2-eth: don't print error from dpaa2_mac_connect if that's EPROBE_DEFER

When booting a board with DPAA2 interfaces defined statically via DPL
(as opposed to creating them dynamically using restool), the driver will
print an unspecific error message.

This change adds the error code to the message, and avoids printing
altogether if the error code is EPROBE_DEFER, because that is not a
cause of alarm.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'dpaa2-eth-of_node'
David S. Miller [Fri, 21 May 2021 21:05:04 +0000 (14:05 -0700)]
Merge branch 'dpaa2-eth-of_node'

Ioana Ciornei says:

====================
dpaa2-eth: setup the of_node

This patch set allows DSA to work with a DPAA2 master device by setting
up the of_node to point to the corresponding MAC DTS node, so that
of_find_net_device_by_node() can find it.
As an added benefit, udev rules can now be used to create a naming
scheme based on the physical MAC.

The second patch renames the debugfs directory to use the DPNI name
instead of the netdev name, since the latter can be changed after probe
time.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: name the debugfs directory after the DPNI object
Ioana Ciornei [Fri, 21 May 2021 13:25:30 +0000 (16:25 +0300)]
dpaa2-eth: name the debugfs directory after the DPNI object

Name the debugfs directory after the DPNI object instead of the netdev
name since this can be changed after probe by udev rules.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: setup the of_node field of the device
Ioana Ciornei [Fri, 21 May 2021 13:25:29 +0000 (16:25 +0300)]
dpaa2-eth: setup the of_node field of the device

When the DPNI object is connected to a DPMAC, setup the of_node to point
to the DTS device node of that specific MAC. This enables other drivers,
for example the DSA subsystem, to find the net_device by its device
node.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'sja1105-stats'
David S. Miller [Fri, 21 May 2021 21:01:41 +0000 (14:01 -0700)]
Merge branch 'sja1105-stats'

Vladimir Oltean says:

====================
Ethtool statistics counters cleanup for SJA1105 DSA driver

This series removes some reported data from ethtool -S which were not
counters at all, and reorganizes the code such that counters can be read
individually and not just all at once.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: don't use burst SPI reads for port statistics
Vladimir Oltean [Fri, 21 May 2021 13:16:08 +0000 (16:16 +0300)]
net: dsa: sja1105: don't use burst SPI reads for port statistics

The current internal sja1105 driver API is optimized for retrieving many
statistics counters at once. But the switch does not do atomic snapshotting
for them anyway.

In case we start reporting the hardware port counters through
ndo_get_stats64 as well, not just ethtool, it would be good to be able
to read individual port counters and not all of them.

Additionally, since Arnd Bergmann's commit ae1804de93f6 ("dsa: sja1105:
dynamically allocate stats structure"), sja1105_get_ethtool_stats
allocates memory dynamically, since struct sja1105_port_status was
deemed to consume too much stack memory. That is not ideal.
The large structure is only needed because of the burst read.
If we read statistics one by one, we can consume less memory, and
we can avoid dynamic allocation.

Additionally, latency-sensitive interfaces such as PTP operations (for
phc2sys) might suffer if the SPI mutex is being held for too long, which
happens in the case of SPI burst reads. By reading counters one by one,
we give a chance for higher priority processes to preempt and take the
SPI bus mutex for accessing the PTP clock.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: stop reporting the queue levels in ethtool port counters
Vladimir Oltean [Fri, 21 May 2021 13:16:07 +0000 (16:16 +0300)]
net: dsa: sja1105: stop reporting the queue levels in ethtool port counters

The queue levels are not counters, but instead they represent the
occupancy of the MAC TX queues. Having these in ethtool port counters is
not helpful, so remove them.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: hns3: Fix return of uninitialized variable ret
Colin Ian King [Fri, 21 May 2021 10:01:46 +0000 (11:01 +0100)]
net: hns3: Fix return of uninitialized variable ret

In the unlikely event that rule_cnt is zero the variable ret is
not assigned a value and function hclge_dbg_dump_fd_tcam can end
up returning an unitialized value in ret. Fix this by explicitly
setting ret to zero before the for-loop.

Addresses-Coverity: ("Uninitialized scalar variable")
Fixes: b5a0b70d77b9 ("net: hns3: refactor dump fd tcam of debugfs")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoatm: Fix typo
zuoqilin [Fri, 21 May 2021 09:45:22 +0000 (17:45 +0800)]
atm: Fix typo

Change 'contol' to 'control'.

Signed-off-by: zuoqilin <zuoqilin@yulong.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: Fix inconsistent indenting
Jiapeng Chong [Fri, 21 May 2021 09:40:14 +0000 (17:40 +0800)]
net: phy: Fix inconsistent indenting

Eliminate the follow smatch warning:

drivers/net/phy/phy_device.c:2886 phy_probe() warn: inconsistent
indenting.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agosfc: farch: fix compile warning in efx_farch_dimension_resources()
Yang Yingliang [Fri, 21 May 2021 03:57:21 +0000 (11:57 +0800)]
sfc: farch: fix compile warning in efx_farch_dimension_resources()

Fix the following kernel build warning when CONFIG_SFC_SRIOV is disabled:

  drivers/net/ethernet/sfc/farch.c: In function ‘efx_farch_dimension_resources’:
  drivers/net/ethernet/sfc/farch.c:1671:21: warning: variable ‘buftbl_min’ set but not used [-Wunused-but-set-variable]
    unsigned vi_count, buftbl_min, total_tx_channels;

Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bonding: bond_alb: Fix some typos in bond_alb.c
Wang Hai [Fri, 21 May 2021 03:31:35 +0000 (11:31 +0800)]
net: bonding: bond_alb: Fix some typos in bond_alb.c

s/becase/because/
s/reqeusts/requests/
s/funcions/functions/
s/addreses/addresses/

Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocaif_virtio: Fix some typos in caif_virtio.c
Wang Hai [Fri, 21 May 2021 03:24:55 +0000 (11:24 +0800)]
caif_virtio: Fix some typos in caif_virtio.c

s/patckets/packets/
s/avilable/available/
s/tbe/the/

Signed-off-by: Wang Hai <wanghai38@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'wan-cleanups'
David S. Miller [Fri, 21 May 2021 20:26:41 +0000 (13:26 -0700)]
Merge branch 'wan-cleanups'

Guangbin Huang says:

====================
net: wan: clean up some code style issues

This patchset clean up some code style issues.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: add necessary () to macro argument
Peng Li [Fri, 21 May 2021 01:08:17 +0000 (09:08 +0800)]
net: wan: add necessary () to macro argument

Macro argument 'card' and 'port' may be better as
'(card)' and '(port)' to avoid precedence issues.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: add braces {} to all arms of the statement
Peng Li [Fri, 21 May 2021 01:08:16 +0000 (09:08 +0800)]
net: wan: add braces {} to all arms of the statement

Braces {} should be used on all arms of this statement.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: remove redundant blank lines
Peng Li [Fri, 21 May 2021 01:08:15 +0000 (09:08 +0800)]
net: wan: remove redundant blank lines

This patch removes some redundant blank lines.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: fix the code style issue about trailing statements
Peng Li [Fri, 21 May 2021 01:08:14 +0000 (09:08 +0800)]
net: wan: fix the code style issue about trailing statements

Trailing statements should be on next line.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: add some required spaces
Peng Li [Fri, 21 May 2021 01:08:13 +0000 (09:08 +0800)]
net: wan: add some required spaces

Add space required after that close brace '}'.
Add space required before the open parenthesis '('.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wan: fix an code style issue about "foo* bar"
Peng Li [Fri, 21 May 2021 01:08:12 +0000 (09:08 +0800)]
net: wan: fix an code style issue about "foo* bar"

Fix the checkpatch error as "foo* bar" should be "foo *bar".

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'sja1105-spi'
David S. Miller [Fri, 21 May 2021 20:23:29 +0000 (13:23 -0700)]
Merge branch 'sja1105-spi'

Vladimir Oltean says:

====================
Adapt the sja1105 DSA driver to the SPI controller's transfer limits

This series changes the SPI transfer procedure in sja1105 to take into
consideration the buffer size limitations that the SPI controller driver
might have.

Changes in v2:
Remove the driver's use of cs_change and send multiple, smaller SPI
messages instead of a single large one.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: adapt to a SPI controller with a limited max transfer size
Vladimir Oltean [Thu, 20 May 2021 21:16:57 +0000 (00:16 +0300)]
net: dsa: sja1105: adapt to a SPI controller with a limited max transfer size

The static config of the sja1105 switch is a long stream of bytes which
is programmed to the hardware in chunks (portions with the chip select
continuously asserted) of max 256 bytes each. Each chunk is a
spi_message composed of 2 spi_transfers: the buffer with the data and a
preceding buffer with the SPI access header.

Only that certain SPI controllers, such as the spi-sc18is602 I2C-to-SPI
bridge, cannot keep the chip select asserted for that long.
The spi_max_transfer_size() and spi_max_message_size() functions are how
the controller can impose its hardware limitations upon the SPI
peripheral driver.

For the sja1105 driver to work with these controllers, both buffers must
be smaller than the transfer limit, and their sum must be smaller than
the message limit.

Regression-tested on a switch connected to a controller with no
limitations (spi-fsl-dspi) as well as with one with caps for both
max_transfer_size and max_message_size (spi-sc18is602).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: send multiple spi_messages instead of using cs_change
Vladimir Oltean [Thu, 20 May 2021 21:16:56 +0000 (00:16 +0300)]
net: dsa: sja1105: send multiple spi_messages instead of using cs_change

The sja1105 driver has been described by Mark Brown as "not using the
[ SPI ] API at all idiomatically" due to the use of cs_change:
https://patchwork.kernel.org/project/netdevbpf/patch/20210520135031.2969183-1-olteanv@gmail.com/

According to include/linux/spi/spi.h, the chip select is supposed to be
asserted for the entire length of a SPI message, as long as cs_change is
false for all member transfers. The cs_change flag changes the following:

(i) When a non-final SPI transfer has cs_change = true, the chip select
    should temporarily deassert and then reassert starting with the next
    transfer.
(ii) When a final SPI transfer has cs_change = true, the chip select
     should remain asserted until the following SPI message.

The sja1105 driver only uses cs_change for its first property, to form a
single SPI message whose layout can be seen below:

                                             this is an entire, single spi_message
           _______________________________________________________________________________________________
          /                                                                                               \
          +-------------+---------------+-------------+---------------+ ... +-------------+---------------+
          | hdr_xfer[0] | chunk_xfer[0] | hdr_xfer[1] | chunk_xfer[1] |     | hdr_xfer[n] | chunk_xfer[n] |
          +-------------+---------------+-------------+---------------+ ... +-------------+---------------+
cs_change      false          true           false           true                false          false

           ____________________________  _____________________________       _____________________________
CS line __/                            \/                             \ ... /                             \__

The fact of the matter is that spi_max_message_size() has an ambiguous
meaning if any non-final transfer has cs_change = true.

If the SPI master has a limitation in that it cannot keep the chip
select asserted for more than, say, 200 bytes (like the spi-sc18is602),
the normal thing for it to do is to implement .max_transfer_size and
.max_message_size, and limit both to 200: in the "worst case" where
cs_change is always false, then the controller can, indeed, not send
messages larger than 200 bytes.

But the fact that the SPI controller's max_message_size does not
necessarily mean that we cannot send messages larger than that.
Notably, if the SPI master special-cases the transfers with cs_change
and treats every chip select toggling as an entirely new transaction,
then a SPI message can easily exceed that limit. So there is a
temptation to ignore the controller's reported max_message_size when
using cs_change = true in non-final transfers.

But that can lead to false conclusions. As Mark points out, the SPI
controller might have a different kind of limitation with the max
message size, that has nothing at all to do with how long it can keep
the chip select asserted.
For example, that might be the case if the device is able to offload the
chip select changes to the hardware as part of the data stream, and it
packs the entire stream of commands+data (corresponding to a SPI
message) into a single DMA transfer that is itself limited in size.

So the only thing we can do is avoid ambiguity by not using cs_change at
all. Instead of sending a single spi_message, we now send multiple SPI
messages as follows:

                  spi_message 0                 spi_message 1                       spi_message n
           ____________________________   ___________________________        _____________________________
          /                            \ /                           \      /                             \
          +-------------+---------------+-------------+---------------+ ... +-------------+---------------+
          | hdr_xfer[0] | chunk_xfer[0] | hdr_xfer[1] | chunk_xfer[1] |     | hdr_xfer[n] | chunk_xfer[n] |
          +-------------+---------------+-------------+---------------+ ... +-------------+---------------+
cs_change      false          true           false           true                false          false

           ____________________________  _____________________________       _____________________________
CS line __/                            \/                             \ ... /                             \__

which is clearer because the max_message_size limit is now easier to
enforce. What is transmitted on the wire stays, of course, the same.

Additionally, because we send no more than 2 transfers at a time, we now
avoid dynamic memory allocation too, which might be seen as an
improvement by some.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: add driver for Motorcomm yt8511 phy
Peter Geis [Thu, 20 May 2021 16:32:30 +0000 (12:32 -0400)]
net: phy: add driver for Motorcomm yt8511 phy

Add a driver for the Motorcomm yt8511 phy that will be used in the
production Pine64 rk3566-quartz64 development board.
It supports gigabit transfer speeds, rgmii, and 125mhz clk output.

Signed-off-by: Peter Geis <pgwipeout@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: xilinx_emaclite: Do not print real IOMEM pointer
YueHaibing [Wed, 19 May 2021 02:47:04 +0000 (10:47 +0800)]
net: xilinx_emaclite: Do not print real IOMEM pointer

Printing kernel pointers is discouraged because they might leak kernel
memory layout.  This fixes smatch warning:

drivers/net/ethernet/xilinx/xilinx_emaclite.c:1191 xemaclite_of_probe() warn:
 argument 4 to %08lX specifier is cast from pointer

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: cdc_ncm: use DEVICE_ATTR_RW macro
YueHaibing [Thu, 20 May 2021 13:46:19 +0000 (21:46 +0800)]
net: cdc_ncm: use DEVICE_ATTR_RW macro

Use DEVICE_ATTR_RW helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: usb: hso: use DEVICE_ATTR_RO macro
YueHaibing [Thu, 20 May 2021 13:41:16 +0000 (21:41 +0800)]
net: usb: hso: use DEVICE_ATTR_RO macro

Use DEVICE_ATTR_RO helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: atm: use DEVICE_ATTR_RO macro
YueHaibing [Thu, 20 May 2021 13:36:45 +0000 (21:36 +0800)]
net: atm: use DEVICE_ATTR_RO macro

Use DEVICE_ATTR_RO helper instead of plain DEVICE_ATTR,
which makes the code a bit shorter and easier to read.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: net: devlink_port_split.py: skip the test if no devlink device
Po-Hsu Lin [Thu, 20 May 2021 10:49:54 +0000 (18:49 +0800)]
selftests: net: devlink_port_split.py: skip the test if no devlink device

When there is no devlink device, the following command will return:
  $ devlink -j dev show
  {dev:{}}

This will cause IndexError when trying to access the first element
in dev of this json dataset. Use the kselftest framework skip code
to skip this test in this case.

Example output with this change:
  # selftests: net: devlink_port_split.py
  # no devlink device was found, test skipped
  ok 7 selftests: net: devlink_port_split.py # SKIP

Link: https://bugs.launchpad.net/bugs/1928889
Signed-off-by: Po-Hsu Lin <po-hsu.lin@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: remove default label from to_string switch
Michal Suchanek [Thu, 20 May 2021 06:50:34 +0000 (08:50 +0200)]
ibmvnic: remove default label from to_string switch

This way the compiler warns when a new value is added to the enum but
not to the string translation like:

drivers/net/ethernet/ibm/ibmvnic.c: In function 'adapter_state_to_string':
drivers/net/ethernet/ibm/ibmvnic.c:832:2: warning: enumeration value 'VNIC_FOOBAR' not handled in switch [-Wswitch]
  switch (state) {
  ^~~~~~
drivers/net/ethernet/ibm/ibmvnic.c: In function 'reset_reason_to_string':
drivers/net/ethernet/ibm/ibmvnic.c:1935:2: warning: enumeration value 'VNIC_RESET_FOOBAR' not handled in switch [-Wswitch]
  switch (reason) {
  ^~~~~~

Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Acked-by: Lijun Pan <lijunp213@gmail.com>
Link: https://lore.kernel.org/netdev/CAOhMmr701LecfuNM+EozqbiTxFvDiXjFdY2aYeKJYaXq9kqVDg@mail.gmail.com/
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoNFC: st21nfca: remove unnecessary variable and labels
wengjianfeng [Thu, 20 May 2021 01:05:50 +0000 (09:05 +0800)]
NFC: st21nfca: remove unnecessary variable and labels

assign vlue (EIO/EPROTO) to variable r, and goto exit label,
but just return r follow exit label, so we delete exit label,
and just replace with return sentence.

Signed-off-by: wengjianfeng <wengjianfeng@yulong.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bond-cleanups'
David S. Miller [Thu, 20 May 2021 22:43:25 +0000 (15:43 -0700)]
Merge branch 'bond-cleanups'

Guangbin Huang says:

====================
net: bonding: clean up some code style issues

This patchset cleans up some code style issues.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bonding: use tabs instead of space for code indent
Yufeng Mo [Thu, 20 May 2021 06:18:35 +0000 (14:18 +0800)]
net: bonding: use tabs instead of space for code indent

Code indent should use tabs where possible, so
use tabs instead of space for code indent.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bonding: remove unnecessary braces
Yufeng Mo [Thu, 20 May 2021 06:18:34 +0000 (14:18 +0800)]
net: bonding: remove unnecessary braces

Braces {} are not necessary for single statement blocks,
so remove these braces {}.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bonding: fix code indent for conditional statements
Yufeng Mo [Thu, 20 May 2021 06:18:33 +0000 (14:18 +0800)]
net: bonding: fix code indent for conditional statements

Fix incorrect code indent for conditional statements.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bonding: add some required blank lines
Yufeng Mo [Thu, 20 May 2021 06:18:32 +0000 (14:18 +0800)]
net: bonding: add some required blank lines

Add some blank lines after declarations as required.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue
David S. Miller [Thu, 20 May 2021 22:13:28 +0000 (15:13 -0700)]
Merge branch '1GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
1GbE Intel Wired LAN Driver Updates 2021-05-20

This series contains updates to igc driver only.

Andre Guedes says:

This series adds AF_XDP zero-copy feature to igc driver.

The initial patches do some code refactoring, preparing the code base to
land the AF_XDP zero-copy feature, avoiding code duplications. The last
patches of the series are the ones implementing the feature.

The last patch which indeed implements AF_XDP zero-copy support was
originally way too lengthy so, for the sake of code review, I broke it
up into two patches: one adding support for the RX functionality and the
other one adding TX support.
---
v2:
Patch 8/9 - "igc: Enable RX via AF_XDP zero-copy"
 * In XDP_PASS flow, copy metadata too into the skb.
 * When HW timestamp is added by the NIC, after copying it into
   a local variable, update xdp_buff->data_meta so that
   metadata length when XDP program is called 0.
 * In igc_xdp_enable_pool(), call xsk_pool_dma_unmap() on
   failure.

Known issues:
 When an XDP application is running in Tx-Only mode with Zero-Copy
 enabled, it is not expected to add the frames to the fill-queue. I have
 noticed the following two issues in this scenario:
 - If XDP_USE_NEED_WAKEUP flag is not set by application, igc_poll()
   will go into infinite loop because the buffer allocation resulting
   in igc_clean_rx_irq_zc() indicating that all work is not done and NAPI
   should keep polling. This does not occur if XDP_USE_NEED_WAKEUP
   flag is set.
 - Since there are no buffers allocated by userspace for the fill
   queue, there is no memory allocated for the NIC to copy the data
   to. If the packet received is destined to the hardware queue where
   XDP application is running, no packets are received even on other
   queues.
 Both these issues can be mitigated by adding a few frames to the
 fill queue. The second issue can also be mitigated by making sure no
 packets are being received on the hardware queue where Rx is running.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'net-leading-spaces'
David S. Miller [Thu, 20 May 2021 22:10:57 +0000 (15:10 -0700)]
Merge branch 'net-leading-spaces'

Hui Tang says:

====================
net: remove leading spaces before tabs

There are a few leading spaces before tabs and remove it by running the
following commard:

        $ find . -name '*.[ch]' | xargs sed -r -i 's/^[ ]+\t/\t/'
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomii: remove leading spaces before tabs
Hui Tang [Thu, 20 May 2021 03:47:54 +0000 (11:47 +0800)]
mii: remove leading spaces before tabs

There are a few leading spaces before tabs and remove it by running
the following commard:

    $ find . -name '*.[ch]' | xargs sed -r -i 's/^[ ]+\t/\t/'

Signed-off-by: Hui Tang <tanghui20@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoifb: remove leading spaces before tabs
Hui Tang [Thu, 20 May 2021 03:47:53 +0000 (11:47 +0800)]
ifb: remove leading spaces before tabs

There are a few leading spaces before tabs and remove it by running
the following commard:

    $ find . -name '*.[ch]' | xargs sed -r -i 's/^[ ]+\t/\t/'

Signed-off-by: Hui Tang <tanghui20@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>