8846533744 added a “power-on” test for the TLS KDF, but omitted to add
it to the documented list of these tests.
Change-Id: I13dbad4b9359e7dae0938d02ac53e5e011f50824
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44505
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
This imports d741debb320bf54e8575d35603a44d4eb40fa1f9 from upstream.
We've been managing the shared libraries already because our
arm-xlate.pl automatically adds .hidden to .extern lines, but nice to
reduce the diff. (This does result in some duplicate .hidden lines in
the generated output, but we still want the arm-xlate.pl patch to
automatically hide .globl.)
Removing .comm lines does change the generated output, but having each
asm file define its own copy of OPENSSL_armcap_P as a common symbol
always seemed odd. I recall some weird issue where the armv4.pl files
subtly rely on it for iOS's strange .indirect_symbol machinery. (Not
actually because iOS wants a common symbol but because arm-xlate.pl
repurposes .comm to trigger .indirect_symbol.) Fortunately, aarch64 is
much better about PC-relative addressing, so it should be a no-op.
The .comm lines have also previously caused weird issues
(https://boringssl-review.googlesource.com/c/boringssl/+/32324), so
it's generally nice to get rid of them.
Update-Note: If aarch64 builds get some weird error about relocations,
it's this CL's fault.
Change-Id: I763ffa6cda750d99694ded8a5b68d7b27b09cfc9
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44464
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
This imports 1ecc76f6746cefd502c7e9000bdfa4e5d7911386 and
41d62636fd996c031c0c7cef746476278583dc9e from upstream. These would have
rejected the mistake in OpenSSL's EDIPartyName sturcture.
Change-Id: I4eb218f9372bea0f7ff302321b9dc1992ef0c13a
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44424
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This change tweaks our ACVP config to better match what BoringCrypto
has previously tested with CAVP.
Change-Id: I7d7ce5153a3eb7355ae1516f06ff591ee2c9d902
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44385
Reviewed-by: David Benjamin <davidben@google.com>
See also CVE-2020-1971, f960d81215ebf3f65e03d4d5d857fb9b666d6920, and
aa0ad2011d3e7ad8a611da274ef7d9c7706e289b from upstream OpenSSL.
Unlike upstream's version, this CL opts for a simpler edipartyname_cmp.
GENERAL_NAME_cmp is already unsuitable for ordering, just equality,
which means there's no need to preserve return values from
ASN1_STRING_cmp. Additionally, the ASN.1 structure implies most fields
cannot be NULL.
(The change from other to x400Address is a no-op. They're the same type.
Just x400Address is a little clearer. Historical quirks of the
GENERAL_NAME structure.)
Change-Id: I4b0ffe8e931c8ef916794a486e6a0d6d684c0cc1
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44404
Reviewed-by: Adam Langley <agl@google.com>
It's insufficient to signal an error when the PWCT fails. We
additionally need to ensure that the invalid key material is not
returned.
Change-Id: Ic5ff719a688985a61c52540ce6d1ed279a493d27
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44306
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
The newest Rust Nightly is getting stricter about `forbid(warnings)`
which breaks the build.
Use "deny" instead of "forbid". And only deny when running Clippy in
CI/CD, so that when hacking on *ring* we don't have to deal with
warnings right away; we now only have to deal with them when we're ready
to submit a change to be merged.
If I perturb kOrder in the malleability check, our and Wycheproof's
tests don't easily notice. This adds some tests with s above and below
the order. EdDSA hashes the public key with the message, which
frustrates constructing actual boundary cases. Instead, these inputs
were found by generating many signatures.
This isn't ideal, but it is sensitive to the most significant 32 bits.
Change-Id: I7fc03758ab97650d0e94478f355ea7085ae0559a
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44346
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Change-Id: Iba527924a79733b28b12b65d8e1f613d7819eb34
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44345
Commit-Queue: Adam Langley <alangley@gmail.com>
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
<openssl/base.h> checks for a supported platform, but we don't check
endianness of ARM and MIPS, which are bi-endian. See
https://crbug.com/1153312#c7.
Switch this around. Documentation on which define is "official" is hard
to come by, so I mostly mimicked Chromium. Chromium detects
little-endian ARM and MIPS with __ARMEL__ and __MIPSEL__ respectively,
without looking at __arm__ or __mips__. It uses __aarch64__
instead of __AARCH64EL__, but I think that's an oversight. I can get
Clang to output for aarch64_be and that defines __aarch64__ with
__AARCH64EB__.
<openssl/arm_arch.h> (which we should simplify and align with base.h
once this CL sticks) also normalizes to __ARMEL__ over __BYTE_ORDER__
and friends. Although, interestingly, arm_arch.h defines its own
__ARMEL__ on GNUC aarch64, even though Clang does *not* define __ARMEL__
on aarch64. (I'm guessing this aligned for the benefit of the "armx"
bi-arch asm files.) This value is based on __BYTE_ORDER__, not
__ARMEL__, but it assumes GNUC arm always defines __ARMEL__, so I think
it's reasonable to assume GNUC aarch64 always defines __AARCH64EL__.
Given all this, probably the simplest thing that's most likely to work
is to use __ARMEL__, __MIPSEL__, and __AARCH64EL__. Note this does not
change the _M_* checks. _M_* are Windows's definitions, which I think we
can reasonably assume come with an endianness opinion. (Windows' ARM and
ARM64 ABIs mandate little-endian.) This aligns with Chromium.
Update-Note: CPU processor defines are a mess. If a little-endian ARM or
MIPS build breaks, some of the assumptions above may be wrong. In that
case, the output $CC -dM -E - < /dev/null on the offending toolchain
will be useful to fix it. If a big-endian ARM or MIPS build breaks, this
is working as intended. Any resulting binaries weren't producing the
right outputs.
Change-Id: I2a9e662d09df119a71226e91716d84e7ac3792aa
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44324
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
ARM Cortex-A57 and Cortex-A72 cores running in 32-bit mode are affected
by silicon errata #1742098 [0] and #1655431 [1], respectively, where the
second instruction of a AES instruction pair may execute twice if an
interrupt is taken right after the first instruction consumes an input
register of which a single 32-bit lane has been updated the last time it
was modified.
Shuffle the counter assignments around a bit so that the most recent
updates when the AES instruction pair executes are 128-bit wide.
[0] ARM-EPM-049219 v23 Cortex-A57 MPCore Software Developers Errata Notice
[1] ARM-EPM-012079 v11.0 Cortex-A72 MPCore Software Developers Errata Notice
(This is imported from upstream's
409c59e8f44ae56f2587cdd8a7ce611d0e3d91d9.)
The change is applied to both 32-bit and 64-bit for simplicity, but there
was no measurable performance difference, so leaving them aligned is
easiest.
Change-Id: Ic8e5f656f59ae8c2ecb2762a066c2c9064bb34c5
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44284
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
No need to use |sk_new|, which allocates a buffer that will immediately
be realloced.
Change-Id: If0a787beac19933d93c5f9a3a8b560edd027c16c
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44205
Reviewed-by: Adam Langley <agl@google.com>
Clarify that there are no truncation issues on targets where the range
of |unsigned| is smaller than the range of |size_t|.
Ensure that |poly1305_state| is (still) large enough. This is a good
idea independently of this change, but is especially important because
switching the fields to |size_t| might have enlarged the structures.
Change-Id: I16e408229c28fcba6c3592603ddb9431cf1f142d
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44244
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
The docs at os/signal.Notify warn about this signal delivery loss bug at
https://golang.org/pkg/os/signal/#Notify, which says:
Package signal will not block sending to c: the caller must ensure
that c has sufficient buffer space to keep up with the expected signal
rate. For a channel used for notification of just one signal value,
a buffer of size 1 is sufficient.
Discovered by one of Orijtech, Inc's internal static
analyzers that will eventually be donated to the Go project, and will
then be included when one runs:
go test
Change-Id: I5713f7087a195ac706240d32b53d2e4855d93a1c
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44264
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
This covers the use of EVP_sha256() added in 8846533744.
Change-Id: I8cd4c8e271de6a0b9a926e7186c7b24ffe849d67
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/44224
Commit-Queue: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
