Introduce an RCU lock implementation as an alternative locking mechanism
to openssl. The api is documented in the ossl_rcu.pod
file
Read side implementaiton is comparable to that of RWLOCKS:
ossl_rcu_read_lock(lock);
<
critical section in which data can be accessed via
ossl_derefrence
>
ossl_rcu_read_unlock(lock);
Write side implementation is:
ossl_rcu_write_lock(lock);
<
critical section in which data can be updated via
ossl_assign_pointer
and stale data can optionally be scheduled for removal
via ossl_rcu_call
>
ossl_rcu_write_unlock(lock);
...
ossl_synchronize_rcu(lock);
ossl_rcu_call fixup
Reviewed-by: Hugo Landau <hlandau@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/22729)
The parameter list for CRYPTO_DOWN_REF for the icc on windows build was
incorrect.
This issue was introduced by 99fd5b2b10Fixes#23414
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Hugo Landau <hlandau@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23415)
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23307)
ubsan on clang17 has started warning about the following undefined
behavior:
crypto/lhash/lhash.c:299:12: runtime error: call to function err_string_data_hash through pointer to incorrect function type 'unsigned long (*)(const void *)'
[...]/crypto/err/err.c:184: note: err_string_data_hash defined here
#0 0x7fa569e3a434 in getrn [...]/crypto/lhash/lhash.c:299:12
#1 0x7fa569e39a46 in OPENSSL_LH_insert [...]/crypto/lhash/lhash.c:119:10
#2 0x7fa569d866ee in err_load_strings [...]/crypto/err/err.c:280:15
[...]
The issue occurs because, the generic hash functions (OPENSSL_LH_*) will
occasionaly call back to the type specific registered functions for hash
generation/comparison/free/etc, using functions of the (example)
prototype:
[return value] <hash|cmp|free> (void *, [void *], ...)
While the functions implementing hash|cmp|free|etc are defined as
[return value] <fnname> (TYPE *, [TYPE *], ...)
The compiler, not knowing the type signature of the function pointed to
by the implementation, performs no type conversion on the function
arguments
While the C language specification allows for pointers to data of one
type to be converted to pointers of another type, it does not
allow for pointers to functions with one signature to be called
while pointing to functions of another signature. Compilers often allow
this behavior, but strictly speaking it results in undefined behavior
As such, ubsan warns us about this issue
This is an potential fix for the issue, implemented using, in effect,
thunking macros. For each hash type, an additional set of wrapper
funtions is created (currently for compare and hash, but more will be
added for free/doall/etc). The corresponding thunking macros for each
type cases the actuall corresponding callback to a function pointer of
the proper type, and then calls that with the parameters appropriately
cast, avoiding the ubsan warning
This approach is adventageous as it maintains a level of type safety,
but comes at the cost of having to implement several additional
functions per hash table type.
Related to #22896
Reviewed-by: Sasa Nedvedicky <sashan@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23192)
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: David von Oheimb <david.von.oheimb@siemens.com>
(Merged from https://github.com/openssl/openssl/pull/19948)
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: David von Oheimb <david.von.oheimb@siemens.com>
(Merged from https://github.com/openssl/openssl/pull/19948)
During creation of the EVP_PKEY_CTX we were trying to discover what legacy
alg it corresponds to every time which was slow. Instead we move this into
the construction of the EVP_KEYMGMT.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23265)
X509_STORE_get0_objects returns a pointer to the X509_STORE's storage,
but this function is a bit deceptive. It is practically unusable in a
multi-threaded program. See, for example, RUSTSEC-2023-0072, a security
vulnerability caused by this OpenSSL API.
One might think that, if no other threads are mutating the X509_STORE,
it is safe to read the resulting list. However, the documention does not
mention that other logically-const operations on the X509_STORE, notably
certifcate verifications when a hash_dir is installed, will, under a
lock, write to the X509_STORE. The X509_STORE also internally re-sorts
the list on the first query.
If the caller knows to call X509_STORE_lock and X509_STORE_unlock, it
can work around this. But this is not obvious, and the documentation
does not discuss how X509_STORE_lock is very rarely safe to use. E.g.
one cannot call any APIs like X509_STORE_add_cert or
X509_STORE_CTX_get1_issuer while holding the lock because those
functions internally expect to take the lock. (X509_STORE_lock is
another such API which is not safe to export as public API.)
Rather than leave all this to the caller to figure out, the API should
have returned a shallow copy of the list, refcounting the values. Then
it could be internally locked and the caller can freely inspect the
result without synchronization with the X509_STORE.
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23224)
Current PowerPC-related defines omit Darwin ppc64 case.
Use __POWERPC__ in place of __ppc__ + __ppc64__
Fixes#23220
CLA: trivial
Reviewed-by: Tom Cosgrove <tom.cosgrove@arm.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/23245)
Fixes#22818
Reviewed-by: Neil Horman <nhorman@openssl.org>
Reviewed-by: Todd Short <todd.short@me.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/22860)
It would be helpful to be able to generate RSA's dmp1/dmq1/iqmp values
when not provided in the param list to EVP_PKEY_fromdata. Augment the
provider in ossl_rsa_fromdata to preform this generation iff:
a) At least p q n e and e are provided
b) the new parameter OSSL_PARAM_RSA_DERIVE_PQ is set to 1
Fixes#21826
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/21875)
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: David von Oheimb <david.von.oheimb@siemens.com>
(Merged from https://github.com/openssl/openssl/pull/20727)
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: David von Oheimb <david.von.oheimb@siemens.com>
(Merged from https://github.com/openssl/openssl/pull/20727)
If a malformed config file is provided such as the following:
openssl_conf = openssl_init
[openssl_init]
providers = provider_sect
[provider_sect]
= provider_sect
The config parsing library will crash overflowing the stack, as it
recursively parses the same provider_sect ad nauseum.
Prevent this by maintaing a list of visited nodes as we recurse through
referenced sections, and erroring out in the event we visit any given
section node more than once.
Note, adding the test for this revealed that our diagnostic code
inadvertently pops recorded errors off the error stack because
provider_conf_load returns success even in the event that a
configuration parse failed. The call path to provider_conf_load has been
updated in this commit to address that shortcoming, allowing recorded
errors to be visibile to calling applications.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/22898)
