Introduce the function csum_unfolded() that computes the unfolded
32-bit checksum of a data buffer, and call it from csum() that returns
the folded value.
Introduce csum_iov() that computes the checksum using csum_folded() on
all vectors of the iovec array and returns the folded result.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-ID: <20240303135114.1023026-4-lvivier@redhat.com>
[dwg: Fixed trivial cppcheck & clang-tidy regressions]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
If buffer is not aligned use sum_16b() only on the not aligned
part, and then use csum_avx2() on the remaining part
Remove unneeded now function csum_unaligned().
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-ID: <20240303135114.1023026-3-lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
In most places where we need to get ICMP definitions, we get them from
<netinet/ip_icmp.h>. However in checksum.c we instead include
<linux/icmp.h>. Change it to use <netinet/ip_icmp.h> for consistency.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
A number of checksum and hash functions require workarounds for the odd
behaviour of Type-Baased Alias Analysis. We have a detailed comment about
this on siphash_8b() and other functions reference that.
Move the main comment to csume_16b() instead, because we're going to
reorganise things in siphash.c.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
In practical terms, passt doesn't benefit from the additional
protection offered by the AGPL over the GPL, because it's not
suitable to be executed over a computer network.
Further, restricting the distribution under the version 3 of the GPL
wouldn't provide any practical advantage either, as long as the passt
codebase is concerned, and might cause unnecessary compatibility
dilemmas.
Change licensing terms to the GNU General Public License Version 2,
or any later version, with written permission from all current and
past contributors, namely: myself, David Gibson, Laine Stump, Andrea
Bolognani, Paul Holzinger, Richard W.M. Jones, Chris Kuhn, Florian
Weimer, Giuseppe Scrivano, Stefan Hajnoczi, and Vasiliy Ulyanov.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Recently, commit 4ddbcb9c0c ("tcp: Disable optimisations
for tcp_hash()") worked around yet another issue we hit with gcc 12
and '-flto -O2': some stores affecting the input data to siphash_20b()
were omitted altogether, and tcp_hash() wouldn't get the correct hash
for incoming connections.
Digging further into this revealed that, at least according to gcc's
interpretation of C99 aliasing rules, passing pointers to functions
with different types compared to the effective type of the object
(for example, a uint8_t pointer to an anonymous struct, as it happens
in tcp_hash()), doesn't guarantee that stores are not reordered
across the function call.
This means that, in general, our checksum and hash functions might
not see parts of input data that was intended to be provided by
callers.
Not even switching from uint8_t to character types, which should be
appropriate here, according to C99 (ISO/IEC 9899, TC3, draft N1256),
section 6.5, "Expressions", paragraph 7:
An object shall have its stored value accessed only by an lvalue
expression that has one of the following types:
[...]
— a character type.
does the trick. I guess this is also subject to interpretation:
casting passed pointers to character types, and then using those as
different types, might still violate (dubious) aliasing rules.
Disable gcc strict aliasing rules for potentially affected functions,
which, in turn, disables gcc's Type-Based Alias Analysis (TBAA)
optimisations based on those function arguments.
Drop the existing workarounds. Also the (seemingly?) bogus
'maybe-uninitialized' warning on the tcp_tap_handler() > tcp_hash() >
siphash_20b() path goes away with -fno-strict-aliasing on
siphash_20b().
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
We recently corrected some errors handling the endianness of IPv4
addresses. These are very easy errors to make since although we mostly
store them in network endianness, we sometimes need to manipulate them in
host endianness.
To reduce the chances of making such mistakes again, change to always using
a (struct in_addr) instead of a bare in_addr_t or uint32_t to store network
endian addresses. This makes it harder to accidentally do arithmetic or
comparisons on such addresses as if they were host endian.
We introduce a number of IN4_IS_ADDR_*() helpers to make it easier to
directly work with struct in_addr values. This has the additional benefit
of making the IPv4 and IPv6 paths more visually similar.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
We need to zero out the checksum field before calculating the
checksum, of course. I have no idea how this passed the "icmp" test
set, looking into it.
Reported-by: Paul Holzinger <pholzing@redhat.com>
Fixes: 67ab617172 ("Add csum_icmp4() helper for calculating ICMP checksums")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
tap_ip_send() is never used for TCP packets, we're unlikely to use it for
that in future, and the handling of TCP packets makes other cleanups
unnecessarily awkward. Remove it.
This is the only user of csum_tcp4(), so we can remove that as well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
We calculate IPv4 header checksums in at least two places, in dhcp() and
in tap_ip_send. Add a helper to handle this calculation in both places.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
At least two places in passt fill in UDP over IPv4 checksums, although
since UDP checksums are optional with IPv4 that just amounts to storing
a 0 (in tap_ip_send()) or leaving a 0 from an earlier initialization (in
dhcp()). For consistency, add a helper for this "calculation".
Just for the heck of it, add the option (compile time disabled for now) to
calculate real UDP checksums.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Add a helper for calculating UDP checksums when used over IPv6
For future flexibility, the new helper takes parameters for the fields in
the IPv6 pseudo-header, so an IPv6 header or pseudo-header doesn't need to
be explicitly constructed. It also allows the UDP header and payload to
be in separate buffers, although we don't use this yet.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Although tap_ip_send() is currently the only place calculating ICMP
checksums, create a helper function for symmetry with ICMPv6. For
future flexibility it allows the ICMPv6 header and payload to be in
separate buffers.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
At least two places in passt calculate ICMPv6 checksums, ndp() and
tap_ip_send(). Add a helper to handle this calculation in both places.
For future flexibility, the new helper takes parameters for the fields in
the IPv6 pseudo-header, so an IPv6 header or pseudo-header doesn't need to
be explicitly constructed. It also allows the ICMPv6 header and payload to
be in separate buffers, although we don't use this yet.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
With gcc 11 and 12, passing -flto, or -flto=auto, and -O2,
intra-procedural optimisation gets rid of a fundamental bit in ndp():
the store of hop_limit in the IPv6 header, before the checksum is
calculated, which on x86_64 looks like this:
ip6hr->hop_limit = IPPROTO_ICMPV6;
b8c0: c6 44 24 35 3a movb $0x3a,0x35(%rsp)
Here, hop_limit is temporarily set to the protocol number, to
conveniently get the IPv6 pseudo-header for ICMPv6 checksum
calculation in memory.
With LTO, the assignment just disappears from the binary.
This is rather visible as NDP messages get a wrong checksum, namely
the expected checksum plus 58, and they're ignored by the guest or
in the namespace, meaning we can't get any IPv6 routes, as reported
by Wenli Quan.
The issue affects a significant number of distribution builds,
including the ones for CentOS Stream 9, EPEL 9, Fedora >= 35,
Mageia Cauldron, and openSUSE Tumbleweed.
As a quick workaround, declare csum_unaligned() as "noipa" for gcc
11 and 12, with -flto and -O2. This disables inlining and cloning,
which causes the assignment to be compiled again.
Leave a TODO item: we should figure out if a gcc issue has already
been reported, and report one otherwise. There's no apparent
justification as to why the store could go away.
Reported-by: Wenli Quan <wquan@redhat.com>
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2129713
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
clang-tidy from LLVM 13.0.1 reports some new warnings from these
checkers:
- altera-unroll-loops, altera-id-dependent-backward-branch: ignore
for the moment being, add a TODO item
- bugprone-easily-swappable-parameters: ignore, nothing to do about
those
- readability-function-cognitive-complexity: ignore for the moment
being, add a TODO item
- altera-struct-pack-align: ignore, alignment is forced in protocol
headers
- concurrency-mt-unsafe: ignore for the moment being, add a TODO
item
Fix bugprone-implicit-widening-of-multiplication-result warnings,
though, that's doable and they seem to make sense.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
...mostly false positives, but a number of very relevant ones too,
in tcp_get_sndbuf(), tcp_conn_from_tap(), and siphash PREAMBLE().
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Unions and structs, you all have names now.
Take the chance to enable bugprone-reserved-identifier,
cert-dcl37-c, and cert-dcl51-cpp checkers in clang-tidy.
Provide a ffsl() weak declaration using gcc built-in.
Start reordering includes, but that's not enough for the
llvm-include-order checker yet.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
SPDX tags don't replace license files. Some notices were missing and
some tags were not according to the SPDX specification, too.
Now reuse --lint from the REUSE tool (https://reuse.software/) passes.
Reported-by: Martin Hauke <mardnh@gmx.de>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
...and further interleave register usage. This brings the csum()
overhead reported by perf(1) for 30 seconds of 64KiB TCP IPv4
frames, host to guest, from 7.2% to 5.8%.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
I left a TODO and never checked -- this actually seems to slightly
improve CPIs on AMD Naples (two 128-bit FMA units glued together).
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Provide an AVX2-based function using compiler intrinsics for
TCP/IP-style checksums. The load/unpack/add idea and implementation
is largely based on code from BESS (the Berkeley Extensible Software
Switch) licensed as 3-Clause BSD, with a number of modifications to
further decrease pipeline stalls and to minimise cache pollution.
This speeds up considerably data paths from sockets to tap
interfaces, decreasing overhead for checksum computation, with
16-64KiB packet buffers, from approximately 11% to 7%. The rest is
just syscalls at this point.
While at it, provide convenience targets in the Makefile for avx2,
avx2_debug, and debug targets -- these simply add target-specific
CFLAGS to the build.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
