TLS device offload allows to perform zerocopy sendfile transmissions.
FreeBSD provides this feature by default, and Linux 5.19 introduced it
as an opt-in. Zerocopy improves the TX rate significantly, but has a
side effect: if the underlying file is changed while being transmitted,
and a TCP retransmission happens, the receiver may get a TLS record
containing both new and old data, which leads to an authentication
failure and termination of connection. This effect is the reason Linux
makes a copy on sendfile by default.
This commit adds support for TLS zerocopy sendfile on Linux disabled by
default to avoid any unlikely backward compatibility issues on Linux,
although sacrificing consistency in OpenSSL's behavior on Linux and
FreeBSD. A new option called KTLSTxZerocopySendfile is added to enable
the new zerocopy behavior on Linux. This option should be used when the
the application guarantees that the file is not modified during
transmission, or it doesn't care about breaking the connection.
The related documentation is also added in this commit. The unit test
added doesn't test the actual functionality (it would require specific
hardware and a non-local peer), but solely checks that it's possible to
set the new option flag.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Boris Pismenny <borisp@nvidia.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18650)
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19346)
And so clean a few useless includes
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19721)
The latest kernel (including stable kernel) has fixed the issue
of decryption failure in CCM mode in TLS 1.3. It is necessary to
reenable CCM mode for KTLS.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17207)
The demux and record RX implemented lists internally. This changes them over
to using list.h.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/19377)
Instead of implementing a list internally.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/19377)
As opposed to implementing a linked list explicitly.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/19377)
This is instead of re-implementing a linked list itself.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/19377)
Properly handle the case where there is pending write data and we want
to send an alert.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19550)
Added SFRAME_LIST structure and QUIC_RSTREAM object to
manage received stream data.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19351)
We need to check whether the sent_messages has actually buffered any
messages in it. If not we won't free the old record layer later when we
clear out the old buffered messages and a memory leak will result.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
The SSL3 prefix no longer seems appropriate. We choose TLS_RL_RECORD instead
of TLS_RECORD because that type already exists elsewhere.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
The SSL3 prefix no longer seems appropriate.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
We move many of the declarations in record.h and record_local.h into
locations inside ssl/record/methods instead. Also many declarations were
no longer required and could be removed completely.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
This file is used by libssl record layer methods and therefore should now
be in the methods subdir
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
They are no longer used and can be removed.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
These fields are instead held in the new record layer code and are
therefore no longer needed.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
They are no longer needed. The new record layer handles this.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
Those fields are no longer used. Their previous function is now in the new
record layer.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19586)
Otherwise, this causes a warning on platforms where 'uint32_t' is
defined as 'unsigned long int' instead of 'unsigned int'.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19322)
For some reason djgpp uses '(unsigned) long int' for (u)int32_t. This
causes errors with -Werror=format, even though these types are in
practice identical.
Obvious solution: cast to the types indicated by the format string.
For asn1_time_test.c I changed the format string to %lli since time_t
may be 'long long' some platforms.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19322)
We had two different macros for calculating the potential growth due to
encryption. The macro we use for allocating the underlying buffer should be
the same one that we use for reserving bytes for encryption growth.
Also if we are adding the MAC independently of the cipher algorithm then
the encryption growth will not include that MAC so we should remove it
from the amount of bytes that we reserve for that growth. Otherwise we
might exceed our buffer size and the WPACKET_reserve operation will
fail.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19622)
We fix dtls_get_max_record_overhead() to give a better value for the max
record overhead. We can't realistically handle the compression case so we
just ignore that.
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19516)
Stitched ciphersuites can grow by more during encryption than the code
allowed for. We fix the calculation and add an assert to check we go it
right.
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19516)
|uclen| is created from three byte values, so this seems a bit
redundant, but if it makes coverity happy
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19519)
During read pipelining we must ensure that the buffer is sufficiently large
to read enough data to fill our pipelines. We also remove some code that
moved data to the start of the packet if we can. This was unnecessary
because of later code which would end up moving it anyway. The earlier move
was also incorrect in the case that |clearold| was 0. This would cause the
read pipelining code to fail with sufficiently large records.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19456)
The pipeline input/output buf arrays must remain accessible to the
EVP_CIPHER_CTX until EVP_Cipher is subsequently called. This fixes an
asan error discovered by the newly added pipeline test.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19456)
TLS pipelining provides the ability for libssl to read or write multiple
records in parallel. It requires special ciphers to do this, and there are
currently no built-in ciphers that provide this capability. However, the
dasync engine does have such a cipher, so we add a test for this capability
using that engine.
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19456)
Remove two function pointers from the OSSL_RECORD_METHOD. Those functions
were no-ops and were never called.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19472)
If rule_str ended in a "-", "l" was incremented one byte past the
end of the buffer. This resulted in an out-of-bounds read when "l"
is dereferenced at the end of the loop. It is safest to just return
early in this case since the condition occurs inside a nested loop.
CLA: trivial
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19166)
The re-occuring surprise is that in Win32, size_t is 32 bits...
Fixed by changing size_t to uint64_t in QUIC_CC
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19345)
This test was disabled during the record write record layer refactor.
We can now enable it again.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19470)
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19464)
This partially reverts commit 30eba7f359.
This is legitimate use of the stack functions and no error
should be reported apart from the NULL return value.
Fixes#19389
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19400)
