OSDN Git Service

uclinux-h8/linux.git
3 years agoRevert "net: avoid double accounting for pure zerocopy skbs"
Jakub Kicinski [Tue, 2 Nov 2021 05:26:08 +0000 (22:26 -0700)]
Revert "net: avoid double accounting for pure zerocopy skbs"

This reverts commit f1a456f8f3fc5828d8abcad941860380ae147b1d.

  WARNING: CPU: 1 PID: 6819 at net/core/skbuff.c:5429 skb_try_coalesce+0x78b/0x7e0
  CPU: 1 PID: 6819 Comm: xxxxxxx Kdump: loaded Tainted: G S                5.15.0-04194-gd852503f7711 #16
  RIP: 0010:skb_try_coalesce+0x78b/0x7e0
  Code: e8 2a bf 41 ff 44 8b b3 bc 00 00 00 48 8b 7c 24 30 e8 19 c0 41 ff 44 89 f0 48 03 83 c0 00 00 00 48 89 44 24 40 e9 47 fb ff ff <0f> 0b e9 ca fc ff ff 4c 8d 70 ff 48 83 c0 07 48 89 44 24 38 e9 61
  RSP: 0018:ffff88881f449688 EFLAGS: 00010282
  RAX: 00000000fffffe96 RBX: ffff8881566e4460 RCX: ffffffff82079f7e
  RDX: 0000000000000003 RSI: dffffc0000000000 RDI: ffff8881566e47b0
  RBP: ffff8881566e46e0 R08: ffffed102619235d R09: ffffed102619235d
  R10: ffff888130c91ae3 R11: ffffed102619235c R12: ffff88881f4498a0
  R13: 0000000000000056 R14: 0000000000000009 R15: ffff888130c91ac0
  FS:  00007fec2cbb9700(0000) GS:ffff88881f440000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007fec1b060d80 CR3: 00000003acf94005 CR4: 00000000003706e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   <IRQ>
   tcp_try_coalesce+0xeb/0x290
   ? tcp_parse_options+0x610/0x610
   ? mark_held_locks+0x79/0xa0
   tcp_queue_rcv+0x69/0x2f0
   tcp_rcv_established+0xa49/0xd40
   ? tcp_data_queue+0x18a0/0x18a0
   tcp_v6_do_rcv+0x1c9/0x880
   ? rt6_mtu_change_route+0x100/0x100
   tcp_v6_rcv+0x1624/0x1830

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Tue, 2 Nov 2021 03:05:14 +0000 (20:05 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Merge in the fixes we had queued in case there was another -rc.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Jakub Kicinski [Tue, 2 Nov 2021 02:59:45 +0000 (19:59 -0700)]
Merge https://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2021-11-01

We've added 181 non-merge commits during the last 28 day(s) which contain
a total of 280 files changed, 11791 insertions(+), 5879 deletions(-).

The main changes are:

1) Fix bpf verifier propagation of 64-bit bounds, from Alexei.

2) Parallelize bpf test_progs, from Yucong and Andrii.

3) Deprecate various libbpf apis including af_xdp, from Andrii, Hengqi, Magnus.

4) Improve bpf selftests on s390, from Ilya.

5) bloomfilter bpf map type, from Joanne.

6) Big improvements to JIT tests especially on Mips, from Johan.

7) Support kernel module function calls from bpf, from Kumar.

8) Support typeless and weak ksym in light skeleton, from Kumar.

9) Disallow unprivileged bpf by default, from Pawan.

10) BTF_KIND_DECL_TAG support, from Yonghong.

11) Various bpftool cleanups, from Quentin.

* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (181 commits)
  libbpf: Deprecate AF_XDP support
  kbuild: Unify options for BTF generation for vmlinux and modules
  selftests/bpf: Add a testcase for 64-bit bounds propagation issue.
  bpf: Fix propagation of signed bounds from 64-bit min/max into 32-bit.
  bpf: Fix propagation of bounds from 64-bit min/max into 32-bit and var_off.
  selftests/bpf: Fix also no-alu32 strobemeta selftest
  bpf: Add missing map_delete_elem method to bloom filter map
  selftests/bpf: Add bloom map success test for userspace calls
  bpf: Add alignment padding for "map_extra" + consolidate holes
  bpf: Bloom filter map naming fixups
  selftests/bpf: Add test cases for struct_ops prog
  bpf: Add dummy BPF STRUCT_OPS for test purpose
  bpf: Factor out helpers for ctx access checking
  bpf: Factor out a helper to prepare trampoline for struct_ops prog
  selftests, bpf: Fix broken riscv build
  riscv, libbpf: Add RISC-V (RV64) support to bpf_tracing.h
  tools, build: Add RISC-V to HOSTARCH parsing
  riscv, bpf: Increase the maximum number of iterations
  selftests, bpf: Add one test for sockmap with strparser
  selftests, bpf: Fix test_txmsg_ingress_parser error
  ...
====================

Link: https://lore.kernel.org/r/20211102013123.9005-1-alexei.starovoitov@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'make-neighbor-eviction-controllable-by-userspace'
Jakub Kicinski [Tue, 2 Nov 2021 02:57:16 +0000 (19:57 -0700)]
Merge branch 'make-neighbor-eviction-controllable-by-userspace'

James Prestwood says:

====================
Make neighbor eviction controllable by userspace
====================

Link: https://lore.kernel.org/r/20211101173630.300969-1-prestwoj@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests: net: add arp_ndisc_evict_nocarrier
James Prestwood [Mon, 1 Nov 2021 17:36:30 +0000 (10:36 -0700)]
selftests: net: add arp_ndisc_evict_nocarrier

This tests the sysctl options for ARP/ND:

/net/ipv4/conf/<iface>/arp_evict_nocarrier
/net/ipv4/conf/all/arp_evict_nocarrier
/net/ipv6/conf/<iface>/ndisc_evict_nocarrier
/net/ipv6/conf/all/ndisc_evict_nocarrier

Signed-off-by: James Prestwood <prestwoj@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ndisc: introduce ndisc_evict_nocarrier sysctl parameter
James Prestwood [Mon, 1 Nov 2021 17:36:29 +0000 (10:36 -0700)]
net: ndisc: introduce ndisc_evict_nocarrier sysctl parameter

In most situations the neighbor discovery cache should be cleared on a
NOCARRIER event which is currently done unconditionally. But for wireless
roams the neighbor discovery cache can and should remain intact since
the underlying network has not changed.

This patch introduces a sysctl option ndisc_evict_nocarrier which can
be disabled by a wireless supplicant during a roam. This allows packets
to be sent after a roam immediately without having to wait for
neighbor discovery.

A user reported roughly a 1 second delay after a roam before packets
could be sent out (note, on IPv4). This delay was due to the ARP
cache being cleared. During testing of this same scenario using IPv6
no delay was noticed, but regardless there is no reason to clear
the ndisc cache for wireless roams.

Signed-off-by: James Prestwood <prestwoj@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: arp: introduce arp_evict_nocarrier sysctl parameter
James Prestwood [Mon, 1 Nov 2021 17:36:28 +0000 (10:36 -0700)]
net: arp: introduce arp_evict_nocarrier sysctl parameter

This change introduces a new sysctl parameter, arp_evict_nocarrier.
When set (default) the ARP cache will be cleared on a NOCARRIER event.
This new option has been defaulted to '1' which maintains existing
behavior.

Clearing the ARP cache on NOCARRIER is relatively new, introduced by:

commit 859bd2ef1fc1110a8031b967ee656c53a6260a76
Author: David Ahern <dsahern@gmail.com>
Date:   Thu Oct 11 20:33:49 2018 -0700

    net: Evict neighbor entries on carrier down

The reason for this changes is to prevent the ARP cache from being
cleared when a wireless device roams. Specifically for wireless roams
the ARP cache should not be cleared because the underlying network has not
changed. Clearing the ARP cache in this case can introduce significant
delays sending out packets after a roam.

A user reported such a situation here:

https://lore.kernel.org/linux-wireless/CACsRnHWa47zpx3D1oDq9JYnZWniS8yBwW1h0WAVZ6vrbwL_S0w@mail.gmail.com/

After some investigation it was found that the kernel was holding onto
packets until ARP finished which resulted in this 1 second delay. It
was also found that the first ARP who-has was never responded to,
which is actually what caues the delay. This change is more or less
working around this behavior, but again, there is no reason to clear
the cache on a roam anyways.

As for the unanswered who-has, we know the packet made it OTA since
it was seen while monitoring. Why it never received a response is
unknown. In any case, since this is a problem on the AP side of things
all that can be done is to work around it until it is solved.

Some background on testing/reproducing the packet delay:

Hardware:
 - 2 access points configured for Fast BSS Transition (Though I don't
   see why regular reassociation wouldn't have the same behavior)
 - Wireless station running IWD as supplicant
 - A device on network able to respond to pings (I used one of the APs)

Procedure:
 - Connect to first AP
 - Ping once to establish an ARP entry
 - Start a tcpdump
 - Roam to second AP
 - Wait for operstate UP event, and note the timestamp
 - Start pinging

Results:

Below is the tcpdump after UP. It was recorded the interface went UP at
10:42:01.432875.

10:42:01.461871 ARP, Request who-has 192.168.254.1 tell 192.168.254.71, length 28
10:42:02.497976 ARP, Request who-has 192.168.254.1 tell 192.168.254.71, length 28
10:42:02.507162 ARP, Reply 192.168.254.1 is-at ac:86:74:55:b0:20, length 46
10:42:02.507185 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 1, length 64
10:42:02.507205 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 2, length 64
10:42:02.507212 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 3, length 64
10:42:02.507219 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 4, length 64
10:42:02.507225 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 5, length 64
10:42:02.507232 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 6, length 64
10:42:02.515373 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 1, length 64
10:42:02.521399 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 2, length 64
10:42:02.521612 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 3, length 64
10:42:02.521941 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 4, length 64
10:42:02.522419 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 5, length 64
10:42:02.523085 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 6, length 64

You can see the first ARP who-has went out very quickly after UP, but
was never responded to. Nearly a second later the kernel retries and
gets a response. Only then do the ping packets go out. If an ARP entry
is manually added prior to UP (after the cache is cleared) it is seen
that the first ping is never responded to, so its not only an issue with
ARP but with data packets in general.

As mentioned prior, the wireless interface was also monitored to verify
the ping/ARP packet made it OTA which was observed to be true.

Signed-off-by: James Prestwood <prestwoj@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agolibbpf: Deprecate AF_XDP support
Magnus Karlsson [Fri, 29 Oct 2021 09:01:11 +0000 (11:01 +0200)]
libbpf: Deprecate AF_XDP support

Deprecate AF_XDP support in libbpf ([0]). This has been moved to
libxdp as it is a better fit for that library. The AF_XDP support only
uses the public libbpf functions and can therefore just use libbpf as
a library from libxdp. The libxdp APIs are exactly the same so it
should just be linking with libxdp instead of libbpf for the AF_XDP
functionality. If not, please submit a bug report. Linking with both
libraries is supported but make sure you link in the correct order so
that the new functions in libxdp are used instead of the deprecated
ones in libbpf.

Libxdp can be found at https://github.com/xdp-project/xdp-tools.

  [0] Closes: https://github.com/libbpf/libbpf/issues/270

Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20211029090111.4733-1-magnus.karlsson@gmail.com
3 years agokbuild: Unify options for BTF generation for vmlinux and modules
Jiri Olsa [Fri, 29 Oct 2021 12:57:29 +0000 (14:57 +0200)]
kbuild: Unify options for BTF generation for vmlinux and modules

Using new PAHOLE_FLAGS variable to pass extra arguments to
pahole for both vmlinux and modules BTF data generation.

Adding new scripts/pahole-flags.sh script that detect and
prints pahole options.

[ fixed issues found by kernel test robot ]

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211029125729.70002-1-jolsa@kernel.org
3 years agoselftests/bpf: Add a testcase for 64-bit bounds propagation issue.
Alexei Starovoitov [Mon, 1 Nov 2021 22:21:53 +0000 (15:21 -0700)]
selftests/bpf: Add a testcase for 64-bit bounds propagation issue.

./test_progs-no_alu32 -vv -t twfw

Before the 64-bit_into_32-bit fix:
19: (25) if r1 > 0x3f goto pc+6
 R1_w=inv(id=0,umax_value=63,var_off=(0x0; 0xff),s32_max_value=255,u32_max_value=255)

and eventually:

invalid access to map value, value_size=8 off=7 size=8
R6 max value is outside of the allowed memory range
libbpf: failed to load object 'no_alu32/twfw.o'

After the fix:
19: (25) if r1 > 0x3f goto pc+6
 R1_w=inv(id=0,umax_value=63,var_off=(0x0; 0x3f))

verif_twfw:OK

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211101222153.78759-3-alexei.starovoitov@gmail.com
3 years agobpf: Fix propagation of signed bounds from 64-bit min/max into 32-bit.
Alexei Starovoitov [Mon, 1 Nov 2021 22:21:52 +0000 (15:21 -0700)]
bpf: Fix propagation of signed bounds from 64-bit min/max into 32-bit.

Similar to unsigned bounds propagation fix signed bounds.
The 'Fixes' tag is a hint. There is no security bug here.
The verifier was too conservative.

Fixes: 3f50f132d840 ("bpf: Verifier, do explicit ALU32 bounds tracking")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211101222153.78759-2-alexei.starovoitov@gmail.com
3 years agobpf: Fix propagation of bounds from 64-bit min/max into 32-bit and var_off.
Alexei Starovoitov [Mon, 1 Nov 2021 22:21:51 +0000 (15:21 -0700)]
bpf: Fix propagation of bounds from 64-bit min/max into 32-bit and var_off.

Before this fix:
166: (b5) if r2 <= 0x1 goto pc+22
from 166 to 189: R2=invP(id=1,umax_value=1,var_off=(0x0; 0xffffffff))

After this fix:
166: (b5) if r2 <= 0x1 goto pc+22
from 166 to 189: R2=invP(id=1,umax_value=1,var_off=(0x0; 0x1))

While processing BPF_JLE the reg_set_min_max() would set true_reg->umax_value = 1
and call __reg_combine_64_into_32(true_reg).

Without the fix it would not pass the condition:
if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value))

since umin_value == 0 at this point.
Before commit 10bf4e83167c the umin was incorrectly ingored.
The commit 10bf4e83167c fixed the correctness issue, but pessimized
propagation of 64-bit min max into 32-bit min max and corresponding var_off.

Fixes: 10bf4e83167c ("bpf: Fix propagation of 32 bit unsigned bounds from 64 bit bounds")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211101222153.78759-1-alexei.starovoitov@gmail.com
3 years agonet: vmxnet3: remove multiple false checks in vmxnet3_ethtool.c
Jean Sacren [Sun, 31 Oct 2021 01:27:28 +0000 (19:27 -0600)]
net: vmxnet3: remove multiple false checks in vmxnet3_ethtool.c

In one if branch, (ec->rx_coalesce_usecs != 0) is checked.  When it is
checked again in two more places, it is always false and has no effect
on the whole check expression.  We should remove it in both places.

In another if branch, (ec->use_adaptive_rx_coalesce != 0) is checked.
When it is checked again, it is always false.  We should remove the
entire branch with it.

In addition we might as well let C precedence dictate by getting rid of
two pairs of parentheses in the neighboring lines in order to keep
expressions on both sides of '||' in balance with checkpatch warning
silenced.

Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Link: https://lore.kernel.org/r/20211031012728.8325-1-sakiwit@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'accurate-memory-charging-for-msg_zerocopy'
Jakub Kicinski [Mon, 1 Nov 2021 23:33:29 +0000 (16:33 -0700)]
Merge branch 'accurate-memory-charging-for-msg_zerocopy'

Talal Ahmad says:

====================
Accurate Memory Charging For MSG_ZEROCOPY

This series improves the accuracy of msg_zerocopy memory accounting.
At present, when msg_zerocopy is used memory is charged twice for the
data - once when user space allocates it, and then again within
__zerocopy_sg_from_iter. The memory charging in the kernel is excessive
because data is held in user pages and is never actually copied to skb
fragments. This leads to incorrectly inflated memory statistics for
programs passing MSG_ZEROCOPY.

We reduce this inaccuracy by introducing the notion of "pure" zerocopy
SKBs - where all the frags in the SKB are backed by pinned userspace
pages, and none are backed by copied pages. For such SKBs, tracked via
the new SKBFL_PURE_ZEROCOPY flag, we elide sk_mem_charge/uncharge
calls, leading to more accurate accounting.

However, SKBs can also be coalesced by the stack at present,
potentially leading to "impure" SKBs. We restrict this coalescing so
it can only happen within the sendmsg() system call itself, for the
most recently allocated SKB. While this can lead to a small degree of
double-charging of memory, this case does not arise often in practice
for workloads that set MSG_ZEROCOPY.

Testing verified that memory usage in the kernel is lowered.
Instrumentation with counters also showed that accounting at time
charging and uncharging is balanced.
====================

Link: https://lore.kernel.org/r/20211030020542.3870542-1-mailtalalahmad@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: avoid double accounting for pure zerocopy skbs
Talal Ahmad [Sat, 30 Oct 2021 02:05:42 +0000 (22:05 -0400)]
net: avoid double accounting for pure zerocopy skbs

Track skbs with only zerocopy data and avoid charging them to kernel
memory to correctly account the memory utilization for msg_zerocopy.
All of the data in such skbs is held in user pages which are already
accounted to user. Before this change, they are charged again in
kernel in __zerocopy_sg_from_iter. The charging in kernel is
excessive because data is not being copied into skb frags. This
excessive charging can lead to kernel going into memory pressure
state which impacts all sockets in the system adversely. Mark pure
zerocopy skbs with a SKBFL_PURE_ZEROCOPY flag and remove
charge/uncharge for data in such skbs.

Initially, an skb is marked pure zerocopy when it is empty and in
zerocopy path. skb can then change from a pure zerocopy skb to mixed
data skb (zerocopy and copy data) if it is at tail of write queue and
there is room available in it and non-zerocopy data is being sent in
the next sendmsg call. At this time sk_mem_charge is done for the pure
zerocopied data and the pure zerocopy flag is unmarked. We found that
this happens very rarely on workloads that pass MSG_ZEROCOPY.

A pure zerocopy skb can later be coalesced into normal skb if they are
next to each other in queue but this patch prevents coalescing from
happening. This avoids complexity of charging when skb downgrades from
pure zerocopy to mixed. This is also rare.

In sk_wmem_free_skb, if it is a pure zerocopy skb, an sk_mem_uncharge
for SKB_TRUESIZE(MAX_TCP_HEADER) is done for sk_mem_charge in
tcp_skb_entail for an skb without data.

Testing with the msg_zerocopy.c benchmark between two hosts(100G nics)
with zerocopy showed that before this patch the 'sock' variable in
memory.stat for cgroup2 that tracks sum of sk_forward_alloc,
sk_rmem_alloc and sk_wmem_queued is around 1822720 and with this
change it is 0. This is due to no charge to sk_forward_alloc for
zerocopy data and shows memory utilization for kernel is lowered.

Signed-off-by: Talal Ahmad <talalahmad@google.com>
Acked-by: Arjun Roy <arjunroy@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agotcp: rename sk_wmem_free_skb
Talal Ahmad [Sat, 30 Oct 2021 02:05:41 +0000 (22:05 -0400)]
tcp: rename sk_wmem_free_skb

sk_wmem_free_skb() is only used by TCP.

Rename it to make this clear, and move its declaration to
include/net/tcp.h

Signed-off-by: Talal Ahmad <talalahmad@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonetdevsim: fix uninit value in nsim_drv_configure_vfs()
Jakub Kicinski [Mon, 1 Nov 2021 22:18:45 +0000 (15:18 -0700)]
netdevsim: fix uninit value in nsim_drv_configure_vfs()

Build bot points out that I missed initializing ret
after refactoring.

Reported-by: kernel test robot <lkp@intel.com>
Fixes: 1c401078bcf3 ("netdevsim: move details of vf config to dev")
Link: https://lore.kernel.org/r/20211101221845.3188490-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoselftests/bpf: Fix also no-alu32 strobemeta selftest
Andrii Nakryiko [Mon, 1 Nov 2021 23:01:18 +0000 (16:01 -0700)]
selftests/bpf: Fix also no-alu32 strobemeta selftest

Previous fix aded bpf_clamp_umax() helper use to re-validate boundaries.
While that works correctly, it introduces more branches, which blows up
past 1 million instructions in no-alu32 variant of strobemeta selftests.

Switching len variable from u32 to u64 also fixes the issue and reduces
the number of validated instructions, so use that instead. Fix this
patch and bpf_clamp_umax() removed, both alu32 and no-alu32 selftests
pass.

Fixes: 0133c20480b1 ("selftests/bpf: Fix strobemeta selftest regression")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211101230118.1273019-1-andrii@kernel.org
3 years agobpf: Add missing map_delete_elem method to bloom filter map
Eric Dumazet [Sun, 31 Oct 2021 17:13:53 +0000 (10:13 -0700)]
bpf: Add missing map_delete_elem method to bloom filter map

Without it, kernel crashes in map_delete_elem(), as reported
by syzbot.

BUG: kernel NULL pointer dereference, address: 0000000000000000
PGD 72c97067 P4D 72c97067 PUD 1e20c067 PMD 0
Oops: 0010 [#1] PREEMPT SMP KASAN
CPU: 0 PID: 6518 Comm: syz-executor196 Not tainted 5.15.0-rc3-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:0x0
Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.
RSP: 0018:ffffc90002bafcb8 EFLAGS: 00010246
RAX: dffffc0000000000 RBX: 1ffff92000575f9f RCX: 0000000000000000
RDX: 1ffffffff1327aba RSI: 0000000000000000 RDI: ffff888025a30c00
RBP: ffffc90002baff08 R08: 0000000000000000 R09: 0000000000000001
R10: ffffffff818525d8 R11: 0000000000000000 R12: ffffffff8993d560
R13: ffff888025a30c00 R14: ffff888024bc0000 R15: 0000000000000000
FS:  0000555557491300(0000) GS:ffff8880b9c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffffffffd6 CR3: 0000000070189000 CR4: 00000000003506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 map_delete_elem kernel/bpf/syscall.c:1220 [inline]
 __sys_bpf+0x34f1/0x5ee0 kernel/bpf/syscall.c:4606
 __do_sys_bpf kernel/bpf/syscall.c:4719 [inline]
 __se_sys_bpf kernel/bpf/syscall.c:4717 [inline]
 __x64_sys_bpf+0x75/0xb0 kernel/bpf/syscall.c:4717
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]

Fixes: 9330986c0300 ("bpf: Add bloom filter map implementation")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211031171353.4092388-1-eric.dumazet@gmail.com
3 years agoMerge branch '"map_extra" and bloom filter fixups'
Alexei Starovoitov [Mon, 1 Nov 2021 21:16:03 +0000 (14:16 -0700)]
Merge branch '"map_extra" and bloom filter fixups'

Joanne Koong says:

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

There are 3 patches in this patchset:

1/3 - Bloom filter naming fixups (kernel/bpf/bloom_filter.c)

2/3 - Add alignment padding for map_extra, rearrange fields in
bpf_map struct to consolidate holes

3/3 - Bloom filter tests (prog_tests/bloom_filter_map):
Add test for successful userspace calls, some refactoring to
use bpf_create_map instead of bpf_create_map_xattr

v1 -> v2:
    * In prog_tests/bloom_filter_map: remove unneeded line break,
also change the inner_map_test to use bpf_create_map instead
of bpf_create_map_xattr.
    * Add acked-bys to commit messages
====================

Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agoselftests/bpf: Add bloom map success test for userspace calls
Joanne Koong [Fri, 29 Oct 2021 22:49:09 +0000 (15:49 -0700)]
selftests/bpf: Add bloom map success test for userspace calls

This patch has two changes:
1) Adds a new function "test_success_cases" to test
successfully creating + adding + looking up a value
in a bloom filter map from the userspace side.

2) Use bpf_create_map instead of bpf_create_map_xattr in
the "test_fail_cases" and test_inner_map to make the
code look cleaner.

Signed-off-by: Joanne Koong <joannekoong@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211029224909.1721024-4-joannekoong@fb.com
3 years agobpf: Add alignment padding for "map_extra" + consolidate holes
Joanne Koong [Fri, 29 Oct 2021 22:49:08 +0000 (15:49 -0700)]
bpf: Add alignment padding for "map_extra" + consolidate holes

This patch makes 2 changes regarding alignment padding
for the "map_extra" field.

1) In the kernel header, "map_extra" and "btf_value_type_id"
are rearranged to consolidate the hole.

Before:
struct bpf_map {
...
        u32 max_entries; /*    36     4 */
        u32 map_flags; /*    40     4 */

        /* XXX 4 bytes hole, try to pack */

        u64 map_extra; /*    48     8 */
        int spin_lock_off; /*    56     4 */
        int timer_off; /*    60     4 */
        /* --- cacheline 1 boundary (64 bytes) --- */
        u32 id; /*    64     4 */
        int numa_node; /*    68     4 */
...
        bool frozen; /*   117     1 */

        /* XXX 10 bytes hole, try to pack */

        /* --- cacheline 2 boundary (128 bytes) --- */
...
        struct work_struct work; /*   144    72 */

        /* --- cacheline 3 boundary (192 bytes) was 24 bytes ago --- */
struct mutex freeze_mutex; /*   216   144  */

        /* --- cacheline 5 boundary (320 bytes) was 40 bytes ago --- */
        u64 writecnt;  /*   360     8 */

    /* size: 384, cachelines: 6, members: 26 */
    /* sum members: 354, holes: 2, sum holes: 14 */
    /* padding: 16 */
    /* forced alignments: 2, forced holes: 1, sum forced holes: 10 */

} __attribute__((__aligned__(64)));

After:
struct bpf_map {
...
        u32 max_entries; /*    36     4 */
        u64 map_extra; /*    40     8  */
        u32 map_flags; /*    48     4 */
        int spin_lock_off; /*    52     4 */
        int timer_off; /*    56     4 */
        u32 id; /*    60     4 */

        /* --- cacheline 1 boundary (64 bytes) --- */
        int numa_node; /*    64     4 */
...
bool frozen /*   113     1  */

        /* XXX 14 bytes hole, try to pack */

        /* --- cacheline 2 boundary (128 bytes) --- */
...
        struct work_struct work; /*   144    72 */

        /* --- cacheline 3 boundary (192 bytes) was 24 bytes ago --- */
        struct mutex freeze_mutex; /*   216   144 */

        /* --- cacheline 5 boundary (320 bytes) was 40 bytes ago --- */
        u64 writecnt;       /*   360     8 */

    /* size: 384, cachelines: 6, members: 26 */
    /* sum members: 354, holes: 1, sum holes: 14 */
    /* padding: 16 */
    /* forced alignments: 2, forced holes: 1, sum forced holes: 14 */

} __attribute__((__aligned__(64)));

2) Add alignment padding to the bpf_map_info struct
More details can be found in commit 36f9814a494a ("bpf: fix uapi hole
for 32 bit compat applications")

Signed-off-by: Joanne Koong <joannekoong@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211029224909.1721024-3-joannekoong@fb.com
3 years agobpf: Bloom filter map naming fixups
Joanne Koong [Fri, 29 Oct 2021 22:49:07 +0000 (15:49 -0700)]
bpf: Bloom filter map naming fixups

This patch has two changes in the kernel bloom filter map
implementation:

1) Change the names of map-ops functions to include the
"bloom_map" prefix.

As Martin pointed out on a previous patchset, having generic
map-ops names may be confusing in tracing and in perf-report.

2) Drop the "& 0xF" when getting nr_hash_funcs, since we
already ascertain that no other bits in map_extra beyond the
first 4 bits can be set.

Signed-off-by: Joanne Koong <joannekoong@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211029224909.1721024-2-joannekoong@fb.com
3 years agoMerge branch 'introduce dummy BPF STRUCT_OPS'
Alexei Starovoitov [Mon, 1 Nov 2021 21:10:00 +0000 (14:10 -0700)]
Merge branch 'introduce dummy BPF STRUCT_OPS'

Hou Tao says:

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

Hi,

Currently the test of BPF STRUCT_OPS depends on the specific bpf
implementation (e.g, tcp_congestion_ops), but it can not cover all
basic functionalities (e.g, return value handling), so introduce
a dummy BPF STRUCT_OPS for test purpose.

Instead of loading a userspace-implemeted bpf_dummy_ops map into
kernel and calling the specific function by writing to sysfs provided
by bpf_testmode.ko, only loading bpf_dummy_ops related prog into
kernel and calling these prog by bpf_prog_test_run(). The latter
is more flexible and has no dependency on extra kernel module.

Now the return value handling is supported by test_1(...) ops,
and passing multiple arguments is supported by test_2(...) ops.
If more is needed, test_x(...) ops can be added afterwards.

Comments are always welcome.
Regards,
Hou

Change Log:
v4:
 * add Acked-by tags in patch 1~4
 * patch 2: remove unncessary comments and update commit message
            accordingly
 * patch 4: remove unnecessary nr checking in dummy_ops_init_args()

v3: https://www.spinics.net/lists/bpf/msg48303.html
 * rebase on bpf-next
 * address comments for Martin, mainly include: merge patch 3 &
   patch 4 in v2, fix names of btf ctx access check helpers,
   handle CONFIG_NET, fix leak in dummy_ops_init_args(), and
   simplify bpf_dummy_init()
 * patch 4: use a loop to check args in test_dummy_multiple_args()

v2: https://www.spinics.net/lists/bpf/msg47948.html
 * rebase on bpf-next
 * add test_2(...) ops to test the passing of multiple arguments
 * a new patch (patch #2) is added to factor out ctx access helpers
 * address comments from Martin & Andrii

v1: https://www.spinics.net/lists/bpf/msg46787.html

RFC: https://www.spinics.net/lists/bpf/msg46117.html
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agoselftests/bpf: Add test cases for struct_ops prog
Hou Tao [Mon, 25 Oct 2021 06:40:25 +0000 (14:40 +0800)]
selftests/bpf: Add test cases for struct_ops prog

Running a BPF_PROG_TYPE_STRUCT_OPS prog for dummy_st_ops::test_N()
through bpf_prog_test_run(). Four test cases are added:
(1) attach dummy_st_ops should fail
(2) function return value of bpf_dummy_ops::test_1() is expected
(3) pointer argument of bpf_dummy_ops::test_1() works as expected
(4) multiple arguments passed to bpf_dummy_ops::test_2() are correct

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211025064025.2567443-5-houtao1@huawei.com
3 years agobpf: Add dummy BPF STRUCT_OPS for test purpose
Hou Tao [Mon, 25 Oct 2021 06:40:24 +0000 (14:40 +0800)]
bpf: Add dummy BPF STRUCT_OPS for test purpose

Currently the test of BPF STRUCT_OPS depends on the specific bpf
implementation of tcp_congestion_ops, but it can not cover all
basic functionalities (e.g, return value handling), so introduce
a dummy BPF STRUCT_OPS for test purpose.

Loading a bpf_dummy_ops implementation from userspace is prohibited,
and its only purpose is to run BPF_PROG_TYPE_STRUCT_OPS program
through bpf(BPF_PROG_TEST_RUN). Now programs for test_1() & test_2()
are supported. The following three cases are exercised in
bpf_dummy_struct_ops_test_run():

(1) test and check the value returned from state arg in test_1(state)
The content of state is copied from userspace pointer and copied back
after calling test_1(state). The user pointer is saved in an u64 array
and the array address is passed through ctx_in.

(2) test and check the return value of test_1(NULL)
Just simulate the case in which an invalid input argument is passed in.

(3) test multiple arguments passing in test_2(state, ...)
5 arguments are passed through ctx_in in form of u64 array. The first
element of array is userspace pointer of state and others 4 arguments
follow.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211025064025.2567443-4-houtao1@huawei.com
3 years agobpf: Factor out helpers for ctx access checking
Hou Tao [Mon, 25 Oct 2021 06:40:23 +0000 (14:40 +0800)]
bpf: Factor out helpers for ctx access checking

Factor out two helpers to check the read access of ctx for raw tp
and BTF function. bpf_tracing_ctx_access() is used to check
the read access to argument is valid, and bpf_tracing_btf_ctx_access()
checks whether the btf type of argument is valid besides the checking
of argument read. bpf_tracing_btf_ctx_access() will be used by the
following patch.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211025064025.2567443-3-houtao1@huawei.com
3 years agobpf: Factor out a helper to prepare trampoline for struct_ops prog
Hou Tao [Mon, 25 Oct 2021 06:40:22 +0000 (14:40 +0800)]
bpf: Factor out a helper to prepare trampoline for struct_ops prog

Factor out a helper bpf_struct_ops_prepare_trampoline() to prepare
trampoline for BPF_PROG_TYPE_STRUCT_OPS prog. It will be used by
.test_run callback in following patch.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211025064025.2567443-2-houtao1@huawei.com
3 years agoselftests, bpf: Fix broken riscv build
Björn Töpel [Thu, 28 Oct 2021 16:10:57 +0000 (18:10 +0200)]
selftests, bpf: Fix broken riscv build

This patch is closely related to commit 6016df8fe874 ("selftests/bpf:
Fix broken riscv build"). When clang includes the system include
directories, but targeting BPF program, __BITS_PER_LONG defaults to
32, unless explicitly set. Work around this problem, by explicitly
setting __BITS_PER_LONG to __riscv_xlen.

Signed-off-by: Björn Töpel <bjorn@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211028161057.520552-5-bjorn@kernel.org
3 years agoriscv, libbpf: Add RISC-V (RV64) support to bpf_tracing.h
Björn Töpel [Thu, 28 Oct 2021 16:10:56 +0000 (18:10 +0200)]
riscv, libbpf: Add RISC-V (RV64) support to bpf_tracing.h

Add macros for 64-bit RISC-V PT_REGS to bpf_tracing.h.

Signed-off-by: Björn Töpel <bjorn@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211028161057.520552-4-bjorn@kernel.org
3 years agotools, build: Add RISC-V to HOSTARCH parsing
Björn Töpel [Thu, 28 Oct 2021 16:10:55 +0000 (18:10 +0200)]
tools, build: Add RISC-V to HOSTARCH parsing

Add RISC-V to the HOSTARCH parsing, so that ARCH is "riscv", and not
"riscv32" or "riscv64".

This affects the perf and libbpf builds, so that arch specific
includes are correctly picked up for RISC-V.

Signed-off-by: Björn Töpel <bjorn@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211028161057.520552-3-bjorn@kernel.org
3 years agoriscv, bpf: Increase the maximum number of iterations
Björn Töpel [Thu, 28 Oct 2021 16:10:54 +0000 (18:10 +0200)]
riscv, bpf: Increase the maximum number of iterations

Now that BPF programs can be up to 1M instructions, it is not uncommon
that a program requires more than the current 16 iterations to
converge.

Bump it to 32, which is enough for selftests/bpf, and test_bpf.ko.

Signed-off-by: Björn Töpel <bjorn@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211028161057.520552-2-bjorn@kernel.org
3 years agoselftests, bpf: Add one test for sockmap with strparser
Liu Jian [Fri, 29 Oct 2021 14:12:16 +0000 (22:12 +0800)]
selftests, bpf: Add one test for sockmap with strparser

Add the test to check sockmap with strparser is working well.

Signed-off-by: Liu Jian <liujian56@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20211029141216.211899-3-liujian56@huawei.com
3 years agoselftests, bpf: Fix test_txmsg_ingress_parser error
Liu Jian [Fri, 29 Oct 2021 14:12:15 +0000 (22:12 +0800)]
selftests, bpf: Fix test_txmsg_ingress_parser error

After "skmsg: lose offset info in sk_psock_skb_ingress", the test case
with ktls failed. This because ktls parser(tls_read_size) return value
is 285 not 256.

The case like this:

tls_sk1 --> redir_sk --> tls_sk2

tls_sk1 sent out 512 bytes data, after tls related processing redir_sk
recved 570 btyes data, and redirect 512 (skb_use_parser) bytes data to
tls_sk2; but tls_sk2 needs 285 * 2 bytes data, receive timeout occurred.

Signed-off-by: Liu Jian <liujian56@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20211029141216.211899-2-liujian56@huawei.com
3 years agoskmsg: Lose offset info in sk_psock_skb_ingress
Liu Jian [Fri, 29 Oct 2021 14:12:14 +0000 (22:12 +0800)]
skmsg: Lose offset info in sk_psock_skb_ingress

If sockmap enable strparser, there are lose offset info in
sk_psock_skb_ingress(). If the length determined by parse_msg function is not
skb->len, the skb will be converted to sk_msg multiple times, and userspace
app will get the data multiple times.

Fix this by get the offset and length from strp_msg. And as Cong suggested,
add one bit in skb->_sk_redir to distinguish enable or disable strparser.

Fixes: 604326b41a6fb ("bpf, sockmap: convert to generic sk_msg interface")
Signed-off-by: Liu Jian <liujian56@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Cong Wang <cong.wang@bytedance.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20211029141216.211899-1-liujian56@huawei.com
3 years agoselftests/bpf: Fix strobemeta selftest regression
Andrii Nakryiko [Fri, 29 Oct 2021 18:29:07 +0000 (11:29 -0700)]
selftests/bpf: Fix strobemeta selftest regression

After most recent nightly Clang update strobemeta selftests started
failing with the following error (relevant portion of assembly included):

  1624: (85) call bpf_probe_read_user_str#114
  1625: (bf) r1 = r0
  1626: (18) r2 = 0xfffffffe
  1628: (5f) r1 &= r2
  1629: (55) if r1 != 0x0 goto pc+7
  1630: (07) r9 += 104
  1631: (6b) *(u16 *)(r9 +0) = r0
  1632: (67) r0 <<= 32
  1633: (77) r0 >>= 32
  1634: (79) r1 = *(u64 *)(r10 -456)
  1635: (0f) r1 += r0
  1636: (7b) *(u64 *)(r10 -456) = r1
  1637: (79) r1 = *(u64 *)(r10 -368)
  1638: (c5) if r1 s< 0x1 goto pc+778
  1639: (bf) r6 = r8
  1640: (0f) r6 += r7
  1641: (b4) w1 = 0
  1642: (6b) *(u16 *)(r6 +108) = r1
  1643: (79) r3 = *(u64 *)(r10 -352)
  1644: (79) r9 = *(u64 *)(r10 -456)
  1645: (bf) r1 = r9
  1646: (b4) w2 = 1
  1647: (85) call bpf_probe_read_user_str#114

  R1 unbounded memory access, make sure to bounds check any such access

In the above code r0 and r1 are implicitly related. Clang knows that,
but verifier isn't able to infer this relationship.

Yonghong Song narrowed down this "regression" in code generation to
a recent Clang optimization change ([0]), which for BPF target generates
code pattern that BPF verifier can't handle and loses track of register
boundaries.

This patch works around the issue by adding an BPF assembly-based helper
that helps to prove to the verifier that upper bound of the register is
a given constant by controlling the exact share of generated BPF
instruction sequence. This fixes the immediate issue for strobemeta
selftest.

  [0] https://github.com/llvm/llvm-project/commit/acabad9ff6bf13e00305d9d8621ee8eafc1f8b08

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/20211029182907.166910-1-andrii@kernel.org
3 years agobpf: Disallow unprivileged bpf by default
Pawan Gupta [Fri, 29 Oct 2021 19:43:54 +0000 (12:43 -0700)]
bpf: Disallow unprivileged bpf by default

Disabling unprivileged BPF would help prevent unprivileged users from
creating certain conditions required for potential speculative execution
side-channel attacks on unmitigated affected hardware.

A deep dive on such attacks and current mitigations is available here [0].

Sync with what many distros are currently applying already, and disable
unprivileged BPF by default. An admin can enable this at runtime, if
necessary, as described in 08389d888287 ("bpf: Add kconfig knob for
disabling unpriv bpf by default").

  [0] "BPF and Spectre: Mitigating transient execution attacks", Daniel Borkmann, eBPF Summit '21
      https://ebpf.io/summit-2021-slides/eBPF_Summit_2021-Keynote-Daniel_Borkmann-BPF_and_Spectre.pdf

Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Link: https://lore.kernel.org/bpf/0ace9ce3f97656d5f62d11093ad7ee81190c3c25.1635535215.git.pawan.kumar.gupta@linux.intel.com
3 years agoMerge branch 'SMC-tracepoints'
David S. Miller [Mon, 1 Nov 2021 13:39:14 +0000 (13:39 +0000)]
Merge branch 'SMC-tracepoints'

Tony Lu says:

====================
Tracepoints for SMC

This patch set introduces tracepoints for SMC, including the tracepoints
basic code. The tracepoitns would help us to track SMC's behaviors by
automatic tools, or other BPF tools, and zero overhead if not enabled.

Compared with kprobe and other dymatic tools, the tracepoints are
considered as stable API, and less overhead for tracing with easy-to-use
API.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smc: Introduce tracepoint for smcr link down
Tony Lu [Mon, 1 Nov 2021 07:39:16 +0000 (15:39 +0800)]
net/smc: Introduce tracepoint for smcr link down

SMC-R link down event is important to help us find links' issues, we
should track this event, especially in the single nic mode, which means
upper layer connection would be shut down. Then find out the direct
link-down reason in time, not only increased the counter, also the
location of the code who triggered this event.

Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smc: Introduce tracepoints for tx and rx msg
Tony Lu [Mon, 1 Nov 2021 07:39:14 +0000 (15:39 +0800)]
net/smc: Introduce tracepoints for tx and rx msg

This introduce two tracepoints for smc tx and rx msg to help us
diagnosis issues of data path. These two tracepoitns don't cover the
path of CORK or MSG_MORE in tx, just the top half of data path.

Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/smc: Introduce tracepoint for fallback
Tony Lu [Mon, 1 Nov 2021 07:39:12 +0000 (15:39 +0800)]
net/smc: Introduce tracepoint for fallback

This introduces tracepoint for smc fallback to TCP, so that we can track
which connection and why it fallbacks, and map the clcsocks' pointer with
/proc/net/tcp to find more details about TCP connections. Compared with
kprobe or other dynamic tracing, tracepoints are stable and easy to use.

Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'amt-driver'
David S. Miller [Mon, 1 Nov 2021 13:36:09 +0000 (13:36 +0000)]
Merge branch 'amt-driver'

Taehee Yoo says:

====================
amt: add initial driver for Automatic Multicast Tunneling (AMT)

This is an implementation of AMT(Automatic Multicast Tunneling), RFC 7450.
https://datatracker.ietf.org/doc/html/rfc7450

This implementation supports IGMPv2, IGMPv3, MLDv1, MLDv2, and IPv4
underlay.

 Summary of RFC 7450
The purpose of this protocol is to provide multicast tunneling.
The main use-case of this protocol is to provide delivery multicast
traffic from a multicast-enabled network to sites that lack multicast
connectivity to the source network.
There are two roles in AMT protocol, Gateway, and Relay.
The main purpose of Gateway mode is to forward multicast listening
information(IGMP, MLD) to the source.
The main purpose of Relay mode is to forward multicast data to listeners.
These multicast traffics(IGMP, MLD, multicast data packets) are tunneled.

Listeners are located behind Gateway endpoint.
But gateway itself can be a listener too.
Senders are located behind Relay endpoint.

    ___________       _________       _______       ________
   |           |     |         |     |       |     |        |
   | Listeners <-----> Gateway <-----> Relay <-----> Source |
   |___________|     |_________|     |_______|     |________|
      IGMP/MLD---------(encap)----------->
         <-------------(decap)--------(encap)------Multicast Data

 Usage of AMT interface
1. Create gateway interface
ip link add amtg type amt mode gateway local 10.0.0.1 discovery 10.0.0.2 \
dev gw1_rt gateway_port 2268 relay_port 2268

2. Create Relay interface
ip link add amtr type amt mode relay local 10.0.0.2 dev relay_rt \
relay_port 2268 max_tunnels 4

v1 -> v2:
 - Eliminate sparse warnings.
   - Use bool type instead of __be16 for identifying v4/v6 protocol.

v2 -> v3:
 - Fix compile warning due to unsed variable.
 - Add missing spinlock comment.
 - Update help message of amt in Kconfig.

v3 -> v4:
 - Split patch.
 - Use CHECKSUM_NONE instead of CHECKSUM_UNNECESSARY.
 - Fix compile error.

v4 -> v5:
 - Remove unnecessary rcu_read_lock().
 - Remove unnecessary amt_change_mtu().
 - Change netlink error message.
 - Add validation for IFLA_AMT_LOCAL_IP and IFLA_AMT_DISCOVERY_IP.
 - Add comments in amt.h.
 - Add missing dev_put() in error path of amt_newlink().
 - Fix typo.
 - Add BUILD_BUG_ON() in amt_smb_cb().
 - Use macro instead of magic values.
 - Use kzalloc() instead of kmalloc().
 - Add selftest script.

v5 -> v6:
 - Reset remote_ip in amt_dev_stop().

v6 -> v7:
 - Fix compile error.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: add amt interface selftest script
Taehee Yoo [Sun, 31 Oct 2021 16:00:06 +0000 (16:00 +0000)]
selftests: add amt interface selftest script

This is selftest script for amt interface.
This script includes basic forwarding scenarion and torture scenario.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoamt: add mld report message handler
Taehee Yoo [Sun, 31 Oct 2021 16:00:05 +0000 (16:00 +0000)]
amt: add mld report message handler

In the previous patch, igmp report handler was added.
That handler can be used for mld too.
So, it uses that common code to parse mld report message.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoamt: add multicast(IGMP) report message handler
Taehee Yoo [Sun, 31 Oct 2021 16:00:04 +0000 (16:00 +0000)]
amt: add multicast(IGMP) report message handler

amt 'Relay' interface manages multicast groups(igmp/mld) and sources.
In order to manage, it should have the function to parse igmp/mld
report messages. So, this adds the logic for parsing igmp report messages
and saves them on their own data structure.

   struct amt_group_node means one group(igmp/mld).
   struct amt_source_node means one source.

The same source can't exist in the same group.
The same group can exist in the same tunnel because it manages
the host address too.

The group information is used when forwarding multicast data.
If there are no groups in the specific tunnel, Relay doesn't forward it.

Although Relay manages sources, it doesn't support the source filtering
feature. Because the reason to manage sources is just that in order
to manage group more correctly.

In the next patch, MLD part will be added.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoamt: add data plane of amt interface
Taehee Yoo [Sun, 31 Oct 2021 16:00:03 +0000 (16:00 +0000)]
amt: add data plane of amt interface

Before forwarding multicast traffic, the amt interface establishes between
gateway and relay. In order to establish, amt defined some message type
and those message flow looks like the below.

                      Gateway                  Relay
                      -------                  -----
                         :        Request        :
                     [1] |           N           |
                         |---------------------->|
                         |    Membership Query   | [2]
                         |    N,MAC,gADDR,gPORT  |
                         |<======================|
                     [3] |   Membership Update   |
                         |   ({G:INCLUDE({S})})  |
                         |======================>|
                         |                       |
    ---------------------:-----------------------:---------------------
   |                     |                       |                     |
   |                     |    *Multicast Data    |  *IP Packet(S,G)    |
   |                     |      gADDR,gPORT      |<-----------------() |
   |    *IP Packet(S,G)  |<======================|                     |
   | ()<-----------------|                       |                     |
   |                     |                       |                     |
    ---------------------:-----------------------:---------------------
                         ~                       ~
                         ~        Request        ~
                     [4] |           N'          |
                         |---------------------->|
                         |   Membership Query    | [5]
                         | N',MAC',gADDR',gPORT' |
                         |<======================|
                     [6] |                       |
                         |       Teardown        |
                         |   N,MAC,gADDR,gPORT   |
                         |---------------------->|
                         |                       | [7]
                         |   Membership Update   |
                         |  ({G:INCLUDE({S})})   |
                         |======================>|
                         |                       |
    ---------------------:-----------------------:---------------------
   |                     |                       |                     |
   |                     |    *Multicast Data    |  *IP Packet(S,G)    |
   |                     |     gADDR',gPORT'     |<-----------------() |
   |    *IP Packet (S,G) |<======================|                     |
   | ()<-----------------|                       |                     |
   |                     |                       |                     |
    ---------------------:-----------------------:---------------------
                         |                       |
                         :                       :

1. Discovery
 - Sent by Gateway to Relay
 - To find Relay unique ip address
2. Advertisement
 - Sent by Relay to Gateway
 - Contains the unique IP address
3. Request
 - Sent by Gateway to Relay
 - Solicit to receive 'Query' message.
4. Query
 - Sent by Relay to Gateway
 - Contains General Query message.
5. Update
 - Sent by  Gateway to Relay
 - Contains report message.
6. Multicast Data
 - Sent by Relay to Gateway
 - encapsulated multicast traffic.
7. Teardown
 - Not supported at this time.

Except for the Teardown message, it supports all messages.

In the next patch, IGMP/MLD logic will be added.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoamt: add control plane of amt interface
Taehee Yoo [Sun, 31 Oct 2021 16:00:02 +0000 (16:00 +0000)]
amt: add control plane of amt interface

It adds definitions and control plane code for AMT.
this is very similar to udp tunneling interfaces such as gtp, vxlan, etc.
In the next patch, data plane code will be added.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'netdevsim-device-and-bus'
David S. Miller [Mon, 1 Nov 2021 13:29:41 +0000 (13:29 +0000)]
Merge branch 'netdevsim-device-and-bus'

Jakub Kicinski says:

====================
netdevsim: improve separation between device and bus

VF config falls strangely in between device and bus
responsibilities today. Because of this bus.c sticks fingers
directly into struct nsim_dev and we look at nsim_bus_dev
in many more places than necessary.

Make bus.c contain pure interface code, and move
the particulars of the logic (which touch on eswitch,
devlink reloads etc) to dev.c. Rename the functions
at the boundary of the interface to make the separation
clearer.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: rename 'driver' entry points
Jakub Kicinski [Sat, 30 Oct 2021 23:15:05 +0000 (16:15 -0700)]
netdevsim: rename 'driver' entry points

Rename functions serving as driver entry points
from nsim_dev_... to nsim_drv_... this makes the
API boundary between bus and dev clearer.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: move max vf config to dev
Jakub Kicinski [Sat, 30 Oct 2021 23:15:04 +0000 (16:15 -0700)]
netdevsim: move max vf config to dev

max_vfs is a strange little beast because the file
hangs off of nsim's debugfs, but it configures a field
in the bus device. Move it to dev.c, let's look at it
as if the device driver was imposing VF limit based
on FW info (like pci_sriov_set_totalvfs()).

Again, when moving refactor the function not to hold
the vfs lock pointlessly while parsing the input.
Wrap the access from the read side in READ_ONCE()
to appease concurrency checkers. Do not check if
return value from snprintf() is negative...

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: move details of vf config to dev
Jakub Kicinski [Sat, 30 Oct 2021 23:15:03 +0000 (16:15 -0700)]
netdevsim: move details of vf config to dev

Since "eswitch" configuration was added bus.c contains
a lot of device details which really belong to dev.c.

Restructure the code while moving it.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: move vfconfig to nsim_dev
Jakub Kicinski [Sat, 30 Oct 2021 23:15:02 +0000 (16:15 -0700)]
netdevsim: move vfconfig to nsim_dev

When netdevsim got split into the faux bus vfconfig ended
up in the bus device (think pci_dev) which is strange because
it contains very networky not to say netdevy information.
Move it to nsim_dev, which is the driver "priv" structure
for the device.

To make sure we don't race with probe/remove take
the device lock (much like PCI).

While at it remove the NULL-checking of vfconfigs.
It appears to be pointless.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: take rtnl_lock when assigning num_vfs
Jakub Kicinski [Sat, 30 Oct 2021 23:15:01 +0000 (16:15 -0700)]
netdevsim: take rtnl_lock when assigning num_vfs

Legacy VF NDOs look at num_vfs and then based on that
index into vfconfig. If we don't rtnl_lock() num_vfs
may get set to 0 and vfconfig freed/replaced while
the NDO is running.

We don't need to protect replacing vfconfig since it's
only done when num_vfs is 0.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'devlink-locking'
David S. Miller [Mon, 1 Nov 2021 13:26:07 +0000 (13:26 +0000)]
Merge branch 'devlink-locking'

Jakub Kicinski says:

====================
improve ethtool/rtnl vs devlink locking

During ethtool netlink development we decided to move some of
the commmands to devlink. Since we don't want drivers to implement
both devlink and ethtool version of the commands ethtool ioctl
falls back to calling devlink. Unfortunately devlink locks must
be taken before rtnl_lock. This results in a questionable
dev_hold() / rtnl_unlock() / devlink / rtnl_lock() / dev_put()
pattern.

This method "works" but it working depends on drivers in question
not doing much in ethtool_ops->begin / complete, and on the netdev
not having needs_free_netdev set.

Since commit 437ebfd90a25 ("devlink: Count struct devlink consumers")
we can hold a reference on a devlink instance and prevent it from
going away (sort of like netdev with dev_hold()). We can use this
to create a more natural reference nesting where we get a ref on
the devlink instance and make the devlink call entirely outside
of the rtnl_lock section.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: don't drop the rtnl_lock half way thru the ioctl
Jakub Kicinski [Sat, 30 Oct 2021 17:18:51 +0000 (10:18 -0700)]
ethtool: don't drop the rtnl_lock half way thru the ioctl

devlink compat code needs to drop rtnl_lock to take
devlink->lock to ensure correct lock ordering.

This is problematic because we're not strictly guaranteed
that the netdev will not disappear after we re-lock.
It may open a possibility of nested ->begin / ->complete
calls.

Instead of calling into devlink under rtnl_lock take
a ref on the devlink instance and make the call after
we've dropped rtnl_lock.

We (continue to) assume that netdevs have an implicit
reference on the devlink returned from ndo_get_devlink_port

Note that ndo_get_devlink_port will now get called
under rtnl_lock. That should be fine since none of
the drivers seem to be taking serious locks inside
ndo_get_devlink_port.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: expose get/put functions
Jakub Kicinski [Sat, 30 Oct 2021 17:18:50 +0000 (10:18 -0700)]
devlink: expose get/put functions

Allow those who hold implicit reference on a devlink instance
to try to take a full ref on it. This will be used from netdev
code which has an implicit ref because of driver call ordering.

Note that after recent changes devlink_unregister() may happen
before netdev unregister, but devlink_free() should still happen
after, so we are safe to try, but we can't just refcount_inc()
and assume it's not zero.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: handle info/flash data copying outside rtnl_lock
Jakub Kicinski [Sat, 30 Oct 2021 17:18:49 +0000 (10:18 -0700)]
ethtool: handle info/flash data copying outside rtnl_lock

We need to increase the lifetime of the data for .get_info
and .flash_update beyond their handlers inside rtnl_lock.

Allocate a union on the heap and use it instead.

Note that we now copy the ethcmd before we lookup dev,
hopefully there is no crazy user space depending on error
codes.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: push the rtnl_lock into dev_ethtool()
Jakub Kicinski [Sat, 30 Oct 2021 17:18:48 +0000 (10:18 -0700)]
ethtool: push the rtnl_lock into dev_ethtool()

Don't take the lock in net/core/dev_ioctl.c,
we'll have things to do outside rtnl_lock soon.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'mana-misc'
David S. Miller [Mon, 1 Nov 2021 13:21:49 +0000 (13:21 +0000)]
Merge branch 'mana-misc'

Dexuan Cui says:

====================
net: mana: some misc patches

Patch 1 is a small fix.

Patch 2 reports OS info to the PF driver.
Before the patch, the req fields were all zeros.

Patch 3 fixes and cleans up the error handling of HWC creation failure.

Patch 4 adds the callbacks for hibernation/kexec. It's based on patch 3.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mana: Support hibernation and kexec
Dexuan Cui [Sat, 30 Oct 2021 00:54:08 +0000 (17:54 -0700)]
net: mana: Support hibernation and kexec

Implement the suspend/resume/shutdown callbacks for hibernation/kexec.

Add mana_gd_setup() and mana_gd_cleanup() for some common code, and
use them in the mand_gd_* callbacks.

Reuse mana_probe/remove() for the hibernation path.

Signed-off-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mana: Improve the HWC error handling
Dexuan Cui [Sat, 30 Oct 2021 00:54:07 +0000 (17:54 -0700)]
net: mana: Improve the HWC error handling

Currently when the HWC creation fails, the error handling is flawed,
e.g. if mana_hwc_create_channel() -> mana_hwc_establish_channel() fails,
the resources acquired in mana_hwc_init_queues() is not released.

Enhance mana_hwc_destroy_channel() to do the proper cleanup work and
call it accordingly.

Signed-off-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mana: Report OS info to the PF driver
Dexuan Cui [Sat, 30 Oct 2021 00:54:06 +0000 (17:54 -0700)]
net: mana: Report OS info to the PF driver

The PF driver might use the OS info for statistical purposes.

Signed-off-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mana: Fix the netdev_err()'s vPort argument in mana_init_port()
Dexuan Cui [Sat, 30 Oct 2021 00:54:05 +0000 (17:54 -0700)]
net: mana: Fix the netdev_err()'s vPort argument in mana_init_port()

Use the correct port index rather than 0.

Signed-off-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'mptcp-selftests'
David S. Miller [Mon, 1 Nov 2021 13:19:49 +0000 (13:19 +0000)]
Merge branch 'mptcp-selftests'

Mat Martineau says:

====================
mptcp: Some selftest improvements

Here are a couple of selftest changes for MPTCP.

Patch 1 fixes a mistake where the wrong protocol (TCP vs MPTCP) could be
requested on the listening socket in some link failure tests.

Patch 2 refactors the simulataneous flow tests to improve timing
accuracy and give more consistent results.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: mptcp: more stable simult_flows tests
Paolo Abeni [Fri, 29 Oct 2021 23:55:59 +0000 (16:55 -0700)]
selftests: mptcp: more stable simult_flows tests

Currently the simult_flows.sh self-tests are not very stable,
especially when running on slow VMs.

The tests measure runtime for transfers on multiple subflows
and check that the time is near the theoretical maximum.

The current test infra introduces a bit of jitter in test
runtime, due to multiple explicit delays. Additionally the
runtime is measured by the shell script wrapper. On a slow
VM, the script overhead is measurable and subject to relevant
jitter.

One solution to make the test more stable would be adding more
slack to the expected time; that could possibly hide real
regressions. Instead move the measurement inside the command
doing the transfer, and drop most unneeded sleeps.

Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: mptcp: fix proto type in link_failure tests
Geliang Tang [Fri, 29 Oct 2021 23:55:58 +0000 (16:55 -0700)]
selftests: mptcp: fix proto type in link_failure tests

In listener_ns, we should pass srv_proto argument to mptcp_connect command,
not cl_proto.

Fixes: 7d1e6f1639044 ("selftests: mptcp: add testcase for active-back")
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: delay complete()
Sukadev Bhattiprolu [Fri, 29 Oct 2021 22:03:16 +0000 (15:03 -0700)]
ibmvnic: delay complete()

If we get CRQ_INIT, we set errno to -EIO and first call complete() to
notify the waiter. Then we try to schedule a FAILOVER reset. If this
occurs while adapter is in PROBING state, ibmvnic_reset() changes the
error code to EAGAIN and returns without scheduling the FAILOVER. The
purpose of setting error code to EAGAIN is to ask the waiter to retry.

But due to the earlier complete() call, the waiter may already have seen
the -EIO response and decided not to retry. This can cause intermittent
failures when bringing up ibmvnic adapters during boot, specially in
in kexec/kdump kernels.

Defer the complete() call until after scheduling the reset.

Also streamline the error code to EAGAIN. Don't see why we need EIO
sometimes. All 3 callers of ibmvnic_reset_init() can handle EAGAIN.

Fixes: 17c8705838a5 ("ibmvnic: Return error code if init interrupted by transport event")
Reported-by: Vaishnavi Bhat <vaish123@in.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: Process crqs after enabling interrupts
Sukadev Bhattiprolu [Fri, 29 Oct 2021 22:03:15 +0000 (15:03 -0700)]
ibmvnic: Process crqs after enabling interrupts

Soon after registering a CRQ it is possible that we get a fail over or
maybe a CRQ_INIT from the VIOS while interrupts were disabled.

Look for any such CRQs after enabling interrupts.

Otherwise we can intermittently fail to bring up ibmvnic adapters during
boot, specially in kexec/kdump kernels.

Fixes: 032c5e82847a ("Driver for IBM System i/p VNIC protocol")
Reported-by: Vaishnavi Bhat <vaish123@in.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: don't stop queue in xmit
Sukadev Bhattiprolu [Fri, 29 Oct 2021 22:03:14 +0000 (15:03 -0700)]
ibmvnic: don't stop queue in xmit

If adapter's resetting bit is on, discard the packet but don't stop the
transmit queue - instead leave that to the reset code. With this change,
it is possible that we may get several calls to ibmvnic_xmit() that simply
discard packets and return.

But if we stop the queue here, we might end up doing so just after
__ibmvnic_open() started the queues (during a hard/soft reset) and before
the ->resetting bit was cleared. If that happens, there will be no one to
restart queue and transmissions will be blocked indefinitely.

This can cause a TIMEOUT reset and with auto priority failover enabled,
an unnecessary FAILOVER reset to less favored backing device and then a
FAILOVER back to the most favored backing device. If we hit the window
repeatedly, we can get stuck in a loop of TIMEOUT, FAILOVER, FAILOVER
resets leaving the adapter unusable for extended periods of time.

Fixes: 7f5b030830fe ("ibmvnic: Free skb's in cases of failure in transmit")
Reported-by: Abdul Haleem <abdhalee@in.ibm.com>
Reported-by: Vaishnavi Bhat <vaish123@in.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'SO_MARK-routing'
David S. Miller [Mon, 1 Nov 2021 13:12:48 +0000 (13:12 +0000)]
Merge branch 'SO_MARK-routing'

Jakub Kicinski says:

====================
udp6: allow SO_MARK ctrl msg to affect routing

Looks like SO_MARK from cmsg does not affect routing policy.
This seems accidental.

I opted for net because of the discrepancy between IPv4
and IPv6, but it never worked and doesn't cause crashes..
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: udp: test for passing SO_MARK as cmsg
Jakub Kicinski [Fri, 29 Oct 2021 15:51:35 +0000 (08:51 -0700)]
selftests: udp: test for passing SO_MARK as cmsg

Before fix:
|  Case IPv6 rejection returned 0, expected 1
|FAIL - 1/4 cases failed

With the fix:
| OK

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoudp6: allow SO_MARK ctrl msg to affect routing
Jakub Kicinski [Fri, 29 Oct 2021 15:51:34 +0000 (08:51 -0700)]
udp6: allow SO_MARK ctrl msg to affect routing

Commit c6af0c227a22 ("ip: support SO_MARK cmsg")
added propagation of SO_MARK from cmsg to skb->mark.
For IPv4 and raw sockets the mark also affects route
lookup, but in case of IPv6 the flow info is
initialized before cmsg is parsed.

Fixes: c6af0c227a22 ("ip: support SO_MARK cmsg")
Reported-and-tested-by: Xintong Hu <huxintong@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonfp: flower: Allow ipv6gretap interface for offloading
Yu Xiao [Fri, 29 Oct 2021 15:04:29 +0000 (17:04 +0200)]
nfp: flower: Allow ipv6gretap interface for offloading

The tunnel_type check only allows for "netif_is_gretap", but for
OVS the port is actually "netif_is_ip6gretap" when setting up GRE
for ipv6, which means offloading request was rejected before.

Therefore, adding "netif_is_ip6gretap" allow ipv6gretap interface
for offloading.

Signed-off-by: Yu Xiao <yu.xiao@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: populate supported_interfaces member
Marek Behún [Thu, 28 Oct 2021 17:00:14 +0000 (18:00 +0100)]
net: dsa: populate supported_interfaces member

Add a new DSA switch operation, phylink_get_interfaces, which should
fill in which PHY_INTERFACE_MODE_* are supported by given port.

Use this before phylink_create() to fill phylinks supported_interfaces
member, allowing phylink to determine which PHY_INTERFACE_MODEs are
supported.

Signed-off-by: Marek Behún <kabel@kernel.org>
[tweaked patch and description to add more complete support -- rmk]
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
David S. Miller [Mon, 1 Nov 2021 13:05:20 +0000 (13:05 +0000)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
100GbE Intel Wired LAN Driver Updates 2021-10-29

This series contains updates to ice and iavf drivers and virtchnl header
file.

Brett removes vlan_promisc argument from a function call for ice driver.
In the virtchnl header file he removes an unused, reserved define and
converts raw value defines to instead use the BIT macro.

Marcin adds syncing of MAC addresses when creating switchdev VFs to
remove error messages on link up and stops showing buffer information
for port representors to remove duplicated entries being displayed for
ice driver.

Karen introduces a helper to go from pci_dev to iavf_adapter in the
iavf driver.

Przemyslaw fixes an issue where iavf was attempting to free IRQs before
calling disable.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Mon, 1 Nov 2021 13:01:44 +0000 (13:01 +0000)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
pull request (net-next): ipsec-next 2021-10-30

Just two minor changes this time:

1) Remove some superfluous header files from xfrm4_tunnel.c
   From Mianhan Liu.

2) Simplify some error checks in xfrm_input().
   From luo penghao.

Please pull or let me know if there are problems.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Mon, 1 Nov 2021 12:59:58 +0000 (12:59 +0000)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next:

1) Use array_size() in ebtables, from Gustavo A. R. Silva.

2) Attach IPS_ASSURED to internal UDP stream state, reported by
   Maciej Zenczykowski.

3) Add NFT_META_IFTYPE to match on the interface type either
   from ingress or egress.

4) Generalize pktinfo->tprot_set to flags field.

5) Allow to match on inner headers / payload data.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'mlx5-updates-2021-10-29' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Mon, 1 Nov 2021 12:53:24 +0000 (12:53 +0000)]
Merge tag 'mlx5-updates-2021-10-29' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2021-10-29

1) Minor trivial refactoring and improvements
2) Check for unsupported parameters fields in SW steering
3) Support TC offload for OVS internal port, from Ariel, see below.

Ariel Levkovich says:

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

Support HW offload of TC rules involving OVS internal port
device type as the filter device or the destination
device.

The support is for flows which explicitly use the internal
port as source or destination device as well as indirect offload
for flows performing tunnel set or unset via a tunnel device
and the internal port is the tunnel overlay device.

Since flows with internal port as source port are added
as egress rules while redirecting to internal port is done
as an ingress redirect, the series introduces the necessary
changes in mlx5_core driver to support the new types of flows
and actions.

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

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

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetfilter: nft_payload: support for inner header matching / mangling
Pablo Neira Ayuso [Thu, 28 Oct 2021 20:15:00 +0000 (22:15 +0200)]
netfilter: nft_payload: support for inner header matching / mangling

Allow to match and mangle on inner headers / payload data after the
transport header. There is a new field in the pktinfo structure that
stores the inner header offset which is calculated only when requested.
Only TCP and UDP supported at this stage.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nf_tables: convert pktinfo->tprot_set to flags field
Pablo Neira Ayuso [Thu, 28 Oct 2021 19:47:55 +0000 (21:47 +0200)]
netfilter: nf_tables: convert pktinfo->tprot_set to flags field

Generalize boolean field to store more flags on the pktinfo structure.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nft_meta: add NFT_META_IFTYPE
Pablo Neira Ayuso [Mon, 25 Oct 2021 13:43:29 +0000 (15:43 +0200)]
netfilter: nft_meta: add NFT_META_IFTYPE

Generalize NFT_META_IIFTYPE to NFT_META_IFTYPE which allows you to match
on the interface type of the skb->dev field. This field is used by the
netdev family to add an implicit dependency to skip non-ethernet packets
when matching on layer 3 and 4 TCP/IP header fields.

For backward compatibility, add the NFT_META_IIFTYPE alias to
NFT_META_IFTYPE.

Add __NFT_META_IIFTYPE, to be used by userspace in the future to match
specifically on the iiftype.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: set on IPS_ASSURED if flows enters internal stream state
Pablo Neira Ayuso [Mon, 25 Oct 2021 09:26:49 +0000 (11:26 +0200)]
netfilter: conntrack: set on IPS_ASSURED if flows enters internal stream state

The internal stream state sets the timeout to 120 seconds 2 seconds
after the creation of the flow, attach this internal stream state to the
IPS_ASSURED flag for consistent event reporting.

Before this patch:

      [NEW] udp      17 30 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 [UNREPLIED] src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282
   [UPDATE] udp      17 30 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282
   [UPDATE] udp      17 30 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282 [ASSURED]
  [DESTROY] udp      17 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282 [ASSURED]

Note IPS_ASSURED for the flow not yet in the internal stream state.

after this update:

      [NEW] udp      17 30 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 [UNREPLIED] src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282
   [UPDATE] udp      17 30 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282
   [UPDATE] udp      17 120 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282 [ASSURED]
  [DESTROY] udp      17 src=10.246.11.13 dst=216.239.35.0 sport=37282 dport=123 src=216.239.35.0 dst=10.246.11.13 sport=123 dport=37282 [ASSURED]

Before this patch, short-lived UDP flows never entered IPS_ASSURED, so
they were already candidate flow to be deleted by early_drop under
stress.

Before this patch, IPS_ASSURED is set on regardless the internal stream
state, attach this internal stream state to IPS_ASSURED.

packet #1 (original direction) enters NEW state
packet #2 (reply direction) enters ESTABLISHED state, sets on IPS_SEEN_REPLY
paclet #3 (any direction) sets on IPS_ASSURED (if 2 seconds since the
          creation has passed by).

Reported-by: Maciej Żenczykowski <zenczykowski@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonet: bridge: switchdev: fix shim definition for br_switchdev_mdb_notify
Vladimir Oltean [Fri, 29 Oct 2021 22:36:06 +0000 (01:36 +0300)]
net: bridge: switchdev: fix shim definition for br_switchdev_mdb_notify

br_switchdev_mdb_notify() is conditionally compiled only when
CONFIG_NET_SWITCHDEV=y and CONFIG_BRIDGE_IGMP_SNOOPING=y. It is called
from br_mdb.c, which is conditionally compiled only when
CONFIG_BRIDGE_IGMP_SNOOPING=y.

The shim definition of br_switchdev_mdb_notify() is therefore needed for
the case where CONFIG_NET_SWITCHDEV=n, however we mistakenly put it
there for the case where CONFIG_BRIDGE_IGMP_SNOOPING=n. This results in
build failures when CONFIG_BRIDGE_IGMP_SNOOPING=y and
CONFIG_NET_SWITCHDEV=n.

To fix this, put the shim definition right next to
br_switchdev_fdb_notify(), which is properly guarded by NET_SWITCHDEV=n.
Since this is called only from br_mdb.c, we need not take any extra
safety precautions, when NET_SWITCHDEV=n and BRIDGE_IGMP_SNOOPING=n,
this shim definition will be absent but nobody will be needing it.

Fixes: 9776457c784f ("net: bridge: mdb: move all switchdev logic to br_switchdev.c")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20211029223606.3450523-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue
Jakub Kicinski [Sat, 30 Oct 2021 04:23:18 +0000 (21:23 -0700)]
Merge branch '1GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
1GbE Intel Wired LAN Driver Updates 2021-10-29

This series contains updates to igc driver only.

Sasha removes an unnecessary media type check, adds a new device ID, and
changes a device reset to a port reset command.
====================

Link: https://lore.kernel.org/r/20211029174101.2970935-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agobnxt_en: Remove not used other ULP define
Leon Romanovsky [Fri, 29 Oct 2021 17:18:15 +0000 (20:18 +0300)]
bnxt_en: Remove not used other ULP define

There is only one bnxt ULP in the upstream kernel and definition
for other ULP can be safely removed.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://lore.kernel.org/r/3a8ea720b28ec4574648012d2a00208f1144eff5.1635527693.git.leonro@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
Jakub Kicinski [Sat, 30 Oct 2021 04:19:29 +0000 (21:19 -0700)]
Merge branch '40GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
40GbE Intel Wired LAN Driver Updates 2021-10-29

This series contains updates to i40e, ice, igb, and ixgbevf drivers.

Yang Li simplifies return statements of bool values for i40e and ice.

Jan Kundrát corrects problems with I2C bit-banging for igb.

Colin Ian King removes unneeded variable initialization for ixgbevf.
====================

Link: https://lore.kernel.org/r/20211029164641.2714265-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonetdevsim: remove max_vfs dentry
Jakub Kicinski [Thu, 28 Oct 2021 21:17:53 +0000 (14:17 -0700)]
netdevsim: remove max_vfs dentry

Commit d395381909a3 ("netdevsim: Add max_vfs to bus_dev")
added this file and saved the dentry for no apparent reason.

Link: https://lore.kernel.org/r/20211028211753.22612-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet/mlx5: Support internal port as decap route device
Ariel Levkovich [Fri, 8 Jan 2021 23:38:04 +0000 (01:38 +0200)]
net/mlx5: Support internal port as decap route device

When performing route device lookup for decap action, support
the case of ovs internal port as the lookup result.

In such case, an internal port struct is mapped and attached
to the flow attributes so that the source port matching of the
rule will match on the internal port's metadata value.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Term table handling of internal port rules
Ariel Levkovich [Tue, 7 Sep 2021 20:30:59 +0000 (23:30 +0300)]
net/mlx5e: Term table handling of internal port rules

Adjust termination table logic to handle rules which
involve internal port as filter or forwarding device.

For cases where the rule forwards from internal port
to uplink, always choose to go via termination table.
This is because it is not known from where the packet
originally arrived to the internal port and it is possible
that it came from the uplink itself, in which case
a term table is required to perform hairpin.
If the packet arrived from a vport, going via term
table has no effect.

For cases where the rule forwards to an internal port
from uplink the rep pointer will point to the uplink rep,
avoid going via termination table as it is not required.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Add indirect tc offload of ovs internal port
Ariel Levkovich [Thu, 29 Apr 2021 07:03:29 +0000 (10:03 +0300)]
net/mlx5e: Add indirect tc offload of ovs internal port

Register callbacks for tc blocks of ovs internal port devices.

This allows an indirect offloading rules that apply on
such devices as the filter device.

In case a rule is added to a tc block of an internal port,
the mlx5 driver will implicitly add a matching on the internal
port's unique vport metadata value to the rule's matching list.
Therefore, only packets that previously hit a rule that redirects
to an internal port and got the vport metadata overwritten to the
internal port's unique metadata, can match on such indirect rule.

Offloading of both ingress and egress tc blocks of internal ports
is supported as opposed to other devices where only ingress block
offloading is supported.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Offload internal port as encap route device
Ariel Levkovich [Fri, 8 Jan 2021 20:03:48 +0000 (22:03 +0200)]
net/mlx5e: Offload internal port as encap route device

When pefroming encap action, a route lookup is performed
to find the routing device the packet should be forwarded
to after the encapsulation. This is the device that has the
local tunnel ip address.

This change adds support to offload an encap rule where the
route device ends up being an ovs internal port.
In such case, the driver will add a HW rule that will encapsulate
the packet with the tunnel header and will overwrite the vport
metadata in reg_c0 to the internal port metadata value.
Finally, the packet will be forwarded to the root table to be
processed again with the indication that it came from an internal
port.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Offload tc rules that redirect to ovs internal port
Ariel Levkovich [Fri, 8 Jan 2021 19:42:39 +0000 (21:42 +0200)]
net/mlx5e: Offload tc rules that redirect to ovs internal port

Allow offloading rules that redirect to ovs internal port
ingress and egress.

To support redirect to ingress device, offloading of REDIRECT_INGRESS
action is added.

When a tc rule redirects to ovs internal port, the hw rule will
overwrite the input vport value in reg_c0 with a new vport metadata
value that is mapped for this internal port using the internal
port mapping api that is introduce in previous patches.
After that the hw rule will redirect the packet to the root table
to continue processing with the new vport metadata value.

The new vport metadata value indicates that this packet is now
arriving through an internal port and therefore should be processed
using rules that apply on the same internal port as the filter device.
Therefore, following rules that apply on this internal port will have
to match on the same vport metadata value as part of their matching
keys to make sure the packet belongs to the internal port.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Accept action skbedit in the tc actions list
Ariel Levkovich [Mon, 11 Jan 2021 19:42:59 +0000 (21:42 +0200)]
net/mlx5e: Accept action skbedit in the tc actions list

Setting the skb packet type field to host is usually
done when performing forwarding to ingress device.

This is required since the receive handling that is used
by the redirect to ingress action checks whether the packet
doesn't belong to this host and drops the packet in such case.

In order to be able to offload action redirect ingress, tc offload
code needs to accept the skbedit ptype action as well.

There's no special handling in HW for such action since it will
be followed by a redirect action and therefore, this code
only allows us to accept such action in the actions list but
not performing anything specific in HW for it.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5: E-Switch, Add ovs internal port mapping to metadata support
Ariel Levkovich [Thu, 29 Apr 2021 06:53:41 +0000 (09:53 +0300)]
net/mlx5: E-Switch, Add ovs internal port mapping to metadata support

Adding infrastructure to map ovs internal port device to vport
match metadata to support offload of rules with internal port as
the filter device or as the destination device.

The infrastructure allows adding and removing internal port device
to an eswitch database and getting a unique vport metadata value to
be placed and match on in reg_c0 when offloading rules that are coming
from or going to an internal port.

The new int port metadata can be written to the source port register
in HW to indicate that current source port of the packet is the
internal port and not one of the actual HW vports (uplink or VF).
Using this method, it is possible to offload TC rules with an OVS
internal port as their destination port (overwriting the src vport
register) or as the filter port (matching on the value of the src
vport register and making sure it matches to the internal port's
value).

There is also a need to handle a miss case where the packet's
src port value was changed in HW to an internal port but a following
rule which matches on this new src port value wasn't found in HW.

In such case, the packet will be forwarded to the driver with
metadata which allows driver to restore the info of the internal
port's netdevice. Once this info is restored, the uplink driver
can forward the packet to the relevant netdevice in SW.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Use generic name for the forwarding dev pointer
Ariel Levkovich [Tue, 26 Oct 2021 17:55:05 +0000 (20:55 +0300)]
net/mlx5e: Use generic name for the forwarding dev pointer

Rename tun_dev to fwd_dev within mlx5e_tc_update_priv struct
since future implementation may introduce other device types
which the handler is forwarding to.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: Refactor rx handler of represetor device
Ariel Levkovich [Tue, 26 Oct 2021 13:01:03 +0000 (16:01 +0300)]
net/mlx5e: Refactor rx handler of represetor device

Move the ownership of skb forwarding to network stack to the
tc update_skb handler as different cases will require different
handling of the skb.

While the tc handler will take care of the various cases and
properly handle the handover of the skb to the network stack
and freeing the skb, the main rx handler will be kept clean
from branches and usage of flags.

Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5: DR, Add check for unsupported fields in match param
Muhammad Sammar [Mon, 5 Jul 2021 12:39:38 +0000 (15:39 +0300)]
net/mlx5: DR, Add check for unsupported fields in match param

When a matcher is being built, we "consume" (clear) mask fields one by one,
and to verify that we do support all the required fields we check if the
whole mask was consumed, else the matching request includes unsupported
fields.

Signed-off-by: Muhammad Sammar <muhammads@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
3 years agonet/mlx5: Allow skipping counter refresh on creation
Paul Blakey [Sun, 11 Jul 2021 13:56:54 +0000 (16:56 +0300)]
net/mlx5: Allow skipping counter refresh on creation

CT creates a counter for each CT rule, and for each such counter,
fs_counters tries to queue mlx5_fc_stats_work() work again via
mod_delayed_work(0) call to refresh all counters. This call has a
large performance impact when reaching high insertion rate and
accounts for ~8% of the insertion time when using software steering.

Allow skipping the refresh of all counters during counter creation.
Change CT to use this refresh skipping for it's counters.

Signed-off-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5e: IPsec: Refactor checksum code in tx data path
Raed Salem [Tue, 26 Oct 2021 07:10:42 +0000 (10:10 +0300)]
net/mlx5e: IPsec: Refactor checksum code in tx data path

Part of code that is related solely to IPsec is always compiled in the
driver code regardless if the IPsec functionality is enabled or disabled
in the driver code, this will add unnecessary branch in case IPsec is
disabled at Tx data path.

Move IPsec related code to IPsec related file such that in case of IPsec
is disabled and because of unlikely macro the compiler should be able to
optimize and omit the checksum IPsec code all together from Tx data path

Signed-off-by: Raed Salem <raeds@nvidia.com>
Reviewed-by: Emeel Hakim <ehakim@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
3 years agonet/mlx5: CT: Remove warning of ignore_flow_level support for VFs
Paul Blakey [Tue, 31 Aug 2021 08:57:52 +0000 (11:57 +0300)]
net/mlx5: CT: Remove warning of ignore_flow_level support for VFs

ignore_flow_level isn't supported for VFs, and so it causes
post_act and ct to warn about it.

Instead of disabling CT for VFs, and a driver update will be need
to enable CT again once firmware support this, remove this warning
specifically for VFs. This way, it could be automatically enabled on
future firmwares where VFs support ignore_flow_level capability.

Signed-off-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Maor Dickman <maord@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>