This happens usually if an template object is created
and there is an out of memory error before the ASN1_OP_NEW_POST
method is called, but asn1_item_embed_free calls now the
ASN1_OP_FREE_POST which may crash because the object is not
properly initialized. Apparently that is only an issue with
the ASN1_OP_FREE_POST handling of crypot/x509/x_crl.c, which
ought to be tolerant to incomplete initialized objects.
The error can be reproduced with the reproducible error injection patch:
$ ERROR_INJECT=1652890550 ../util/shlib_wrap.sh ./asn1-test ./corpora/asn1/0ff17293911f54d1538b9896563a4048d67d9ee4
#0 0x7faae9dbeeba in __sanitizer_print_stack_trace ../../../../gcc-trunk/libsanitizer/asan/asan_stack.cpp:87
#1 0x408dc4 in my_malloc fuzz/test-corpus.c:114
#2 0x7faae99f2430 in CRYPTO_zalloc crypto/mem.c:230
#3 0x7faae97f09e5 in ASN1_STRING_type_new crypto/asn1/asn1_lib.c:341
#4 0x7faae98118f7 in asn1_primitive_new crypto/asn1/tasn_new.c:318
#5 0x7faae9812401 in asn1_item_embed_new crypto/asn1/tasn_new.c:78
#6 0x7faae9812401 in asn1_template_new crypto/asn1/tasn_new.c:240
#7 0x7faae9812315 in asn1_item_embed_new crypto/asn1/tasn_new.c:137
#8 0x7faae9812315 in asn1_template_new crypto/asn1/tasn_new.c:240
#9 0x7faae9812a54 in asn1_item_embed_new crypto/asn1/tasn_new.c:137
#10 0x7faae9812a54 in ASN1_item_ex_new crypto/asn1/tasn_new.c:39
#11 0x7faae980be51 in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:325
#12 0x7faae980c813 in asn1_template_noexp_d2i crypto/asn1/tasn_dec.c:611
#13 0x7faae980d288 in asn1_template_ex_d2i crypto/asn1/tasn_dec.c:518
#14 0x7faae980b9ce in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:382
#15 0x7faae980caf5 in asn1_template_noexp_d2i crypto/asn1/tasn_dec.c:643
#16 0x7faae980d7d3 in asn1_template_ex_d2i crypto/asn1/tasn_dec.c:494
#17 0x7faae980b9ce in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:382
#18 0x7faae980dd1f in ASN1_item_ex_d2i crypto/asn1/tasn_dec.c:124
#19 0x7faae980de35 in ASN1_item_d2i crypto/asn1/tasn_dec.c:114
#20 0x40712c in FuzzerTestOneInput fuzz/asn1.c:301
#21 0x40893b in testfile fuzz/test-corpus.c:182
#22 0x406b86 in main fuzz/test-corpus.c:226
#23 0x7faae8eb1f44 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x21f44)
AddressSanitizer:DEADLYSIGNAL
=================================================================
==1194==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000010 (pc 0x7faae9b0625f bp 0x7fffffe41a00 sp 0x7fffffe41920 T0)
==1194==The signal is caused by a READ memory access.
==1194==Hint: address points to the zero page.
#0 0x7faae9b0625f in crl_cb crypto/x509/x_crl.c:258
#1 0x7faae9811255 in asn1_item_embed_free crypto/asn1/tasn_fre.c:113
#2 0x7faae9812a65 in asn1_item_embed_new crypto/asn1/tasn_new.c:150
#3 0x7faae9812a65 in ASN1_item_ex_new crypto/asn1/tasn_new.c:39
#4 0x7faae980be51 in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:325
#5 0x7faae980c813 in asn1_template_noexp_d2i crypto/asn1/tasn_dec.c:611
#6 0x7faae980d288 in asn1_template_ex_d2i crypto/asn1/tasn_dec.c:518
#7 0x7faae980b9ce in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:382
#8 0x7faae980caf5 in asn1_template_noexp_d2i crypto/asn1/tasn_dec.c:643
#9 0x7faae980d7d3 in asn1_template_ex_d2i crypto/asn1/tasn_dec.c:494
#10 0x7faae980b9ce in asn1_item_embed_d2i crypto/asn1/tasn_dec.c:382
#11 0x7faae980dd1f in ASN1_item_ex_d2i crypto/asn1/tasn_dec.c:124
#12 0x7faae980de35 in ASN1_item_d2i crypto/asn1/tasn_dec.c:114
#13 0x40712c in FuzzerTestOneInput fuzz/asn1.c:301
#14 0x40893b in testfile fuzz/test-corpus.c:182
#15 0x406b86 in main fuzz/test-corpus.c:226
#16 0x7faae8eb1f44 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x21f44)
AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV crypto/x509/x_crl.c:258 in crl_cb
==1194==ABORTING
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18360)
The loop unrolling and use of EOR3 can improve N2 performance
by up to 32%
Signed-off-by: XiaokangQian <xiaokang.qian@arm.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18350)
Fixes#18306
Signed-off-by: Randall S. Becker <rsbecker@nexbridge.com>
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/18351)
To avoid the issue of overflowing the buffer start while
building up the help string prefix this rewrite of the
string building logic does multiple smaller writes to
opt_printf_stderr. While this is slower it completely
avoids the buffer overflow issue and does not place
any (unchecked) length constraints on the name of passed
options. Instead such long options are gracefully
wrapped onto the next line.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12265)
This can't currently happen due to sizeof(start) being way larger than MAX_OPT_HELP_WIDTH,
but wasn't checked for previously. With this patch there still remains one (static) OOB,
when the length of the option name and the valtype2param string for that argument overflow
the buffer in opt_print. This is kinda unlikely, unless someone intentionally crafts a
long option name, in which case this would become some trivial stack buffer overrun with
possibility to overwrite pointer to the OPTIONS structure (a long o->name is critical here).
I sincerely hope we trust our built-in documentation to not exploit ourselves.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12265)
Rather than relying on the locale code working, instead implement these
functions directly.
Fixes#18322
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/18344)
This improves the performance of this function and the ones that rely on it
(ossl_lh_strcasehash primarily).
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/18344)
The FIPS module checksum needs to know that configuration.h is
generated from configuration.h.in, so that information is conserved.
To make this possible, it's now possible to have attributes with the
GENERATE keyword, and the attribute "skip" is added to make a keyword
a no-op, which makes it informative only.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
(Merged from https://github.com/openssl/openssl/pull/16378)
The goal is to avoid having too much of the OpenSSL source rebuilt
because include/openssl/configuration.h, or even because it was a
Makefile target that was called upon (some make implementations
consider the use of a target as an update of that target, even if it
wasn't really updated).
To resolve this, we move the production of include/openssl/configuration.h
to configdata.pm, and only update it if there were any actual changes.
Fixes#16377
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
(Merged from https://github.com/openssl/openssl/pull/16378)
The goal is to avoid having too much of the OpenSSL source rebuilt
because configdata.pm got a new time stamp. The conditions for
updating configdata.pm are now:
1. its time stamp is older than Configure's, or...
2. its contents has changed.
Fixes#16377
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
(Merged from https://github.com/openssl/openssl/pull/16378)
This improvement seems to roughly halve the time it takes to run the
ossl_lh_strcasehash function.
It should have no impact on the strings we hash and search for often (algorithm
names, property strings).
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18354)
Common markdown styles usually show 4-column indents to separate the
list marker and the list item text. That's a common template for
writing new markdown files.
On the other hand, we do have some files (such as CHANGES.md) where we
use a different style.
From a markdown perspective, both are perfectly OK, and there's no
reason to enforce either.
Therefore, the best thing is to exclude this particular rule.
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18297)
These routines make use of the rev8 instruction in the Zbb extension
to accelerate byte-swapping when OpenSSL is built specifically for
a machine that supports Zbb.
Reviewed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Henry Brausen <henry.brausen@vrull.eu>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17640)
ghash-riscv64.pl implements 128-bit galois field multiplication for
use in the GCM mode using RISC-V carryless multiplication primitives.
The clmul-accelerated routine can be selected by setting the Zbb and
Zbc bits of the OPENSSL_riscvcap environment variable at runtime.
Reviewed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Henry Brausen <henry.brausen@vrull.eu>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17640)
RISC-V cpuid implementation allows bitmanip extensions Zb[abcs] to
be enabled at runtime using OPENSSL_riscvcap environment variable.
For example, to specify 64-bit RISC-V with the G,C,Zba,Zbb,Zbc
extensions, one could write: OPENSSL_riscvcap="rv64gc_zba_zbb_zbc"
Architecture string parsing is still very primitive, but can be
expanded in the future. Currently, only bitmanip extensions Zba, Zbb,
Zbc and Zbs are supported.
Includes implementation of constant-time CRYPTO_memcmp in riscv64 asm,
as well as OPENSSL_cleanse. Assembly implementations are written using
perlasm.
Reviewed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Henry Brausen <henry.brausen@vrull.eu>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17640)
This implementation is based on the four-table approach, along the same
lines as the non-constant-time implementation in aes_core.c The
implementation is in perlasm.
Utility functions are defined to automatically stack/unstack registers
as needed for prologues and epilogues. See riscv-elf-psabi-doc at
https://github.com/riscv-non-isa/riscv-elf-psabi-doc/ for ABI details.
Reviewed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Henry Brausen <henry.brausen@vrull.eu>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17640)
This commit removes platform defines the interfere with loading and resolution
of platform and memory model variants of integer types and includes the
appropriate files, stdint.h and sys/types.h where the types are defined.
Fixes#17669
Signed-off-by: Randall S. Becker <rsbecker@nexbridge.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18325)
Just extending the comment so coveriety knows as well.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18326)
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16237)
As there are no limitations for HMACs used in PBKDF2 inside PBES2,
as more specifically the SHA-3 hashes are drop-in replacements for
SHA-2 hashes, we can easily add support for SHA-3 here.
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16237)
When a socket connection is done using BIO_new_connect,
the ktls_enable is done too early, and fails with ENOTCONN.
Therefore the KLTS ioctl will fail later with ENOPROTOOPT.
Fix that by doing the ktls_enable after the connection
succeeded, not when the socket is created as that will
always fail.
One example where this happens is doit_localhost in
test/ssl_old_test.c, and therefore, contrary to the expectation
the -client_ktls option did never enable the client KTLS
connection, but this was not noticed, because there was no
diagnostic output, and it was only visible with strace output.
Also enhanced the ssl_old_test -client_ktls/-server_ktls
options together with -v option to print a summary line
if and how KTLS was negotiated in server and client.
While I am already there adjusted the usage info of
the -s_cert, -s_key commands, and allow -time to print the
timings of ktls connections.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18318)
It seems the return value of EVP_PKEY_copy_parameters() in
ssl_set_cert_and_key(), and could lead to null pointer dereference in
EVP_PKEY_eq() function.
However those functions are complicated and this fix is suggested by
a static analyzer, so please advise.
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18071)
Otherwise the implementation is unnecessarily duplicated in legacy.so.
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18282)
It also allows for passing -DOPENSSL_NO_LOCALE as a workaround
to ./Configure command.
Fixes#18233
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18282)
Otherwise optional properties can be incorrectly ignored.
Fixes#18262
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18269)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18269)
Otherwise, it seems nmake doesn't invoke perl properly.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18215)
Reviewed-by: Philipp Tomsich <philipp.tomsich@vrull.eu>
Signed-off-by: Henry Brausen <henry.brausen@vrull.eu>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18275)
As of clang-14 the strict aliasing is causing code to magically disappear.
By explicitly inlining the code, the aliasing problem evaporates.
Fixes#18225
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18258)
Even though the function is not part of the public api, it is not
entirely removed, in order to minimize the chance of breakage,
because it is exported from libcrypto. Instead, we keep a dummy
implementation.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17974)
This commit removes the entire initialization and cleanup of the
error string hash table (`int_error_hash`) if `no-err` is configured.
The only operative function remaining is `ERR_get_next_error_library()`.
That is the reason why the `err_string_lock` and hence the
`do_err_strings_init()` function can't be removed entirely.
Fixes#17971
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17974)
rotated_mac is a 64-byte aligned buffer of size 64 and rotate_offset is secret.
Consider a weaker leakage model(CL) where only cacheline base address is leaked,
i.e address/32 for 32-byte cacheline(CL32).
Previous code used to perform two loads
1. rotated_mac[rotate_offset ^ 32] and
2. rotated_mac[rotate_offset++]
which would leak 2q + 1, 2q for 0 <= rotate_offset < 32
and 2q, 2q + 1 for 32 <= rotate_offset < 64
The proposed fix performs load operations which will always leak 2q, 2q + 1 and
selects the appropriate value in constant-time.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18033)
When reusing an algctx (it was always freed on reinitialization,
prior to #18105), assert that the associated digest is provided.
We implicitly rely on this for algctx reuse to be safe (since
an implicit fetch could potentially change the digest object used,
including provider, which accordingly could change the layout of the
algctx object.
From code inspection, this is currently always the case -- the only
way to set an algctx requires the provider to be set, and the only
ways to change or remove a provider without destroying the entier
EVP_MD_CTX will also free the algctx. Adding an assertion will help
ensure that this remains true as the code evolves.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/18224)
