The above function was running while holding the store lock with a read
lock. Unfortunately it actually modifies the store, so a write lock is
required instead.
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/13660)
The provider operation_bits array can see concurrent access by multiple
threads and can be reallocated at any time. Therefore we need to ensure
that it is appropriately locked.
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/13660)
This introduces a separate activation counter, and the function
ossl_provider_deactivate() for provider deactivation.
Something to be noted is that if the reference count goes down to
zero, we don't care if the activation count is non-zero (i.e. someone
forgot to call ossl_provider_deactivate()). Since there are no more
references to the provider, it doesn't matter.
The important thing is that deactivation doesn't remove the provider
as long as there are references to it, for example because there are
live methods associated with that provider, but still makes the
provider unavailable to create new methods from.
Fixes#13503Fixes#12157
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/13661)
This includes error reporting for libcrypto sub-libraries in surprising
places.
This was done using util/err-to-raise
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/13318)
This change makes the naming more consistent, because three different terms
were used for the same thing. (The term libctx was used by far most often.)
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12621)
Many of the new types introduced by OpenSSL 3.0 have an OSSL_ prefix,
e.g., OSSL_CALLBACK, OSSL_PARAM, OSSL_ALGORITHM, OSSL_SERIALIZER.
The OPENSSL_CTX type stands out a little by using a different prefix.
For consistency reasons, this type is renamed to OSSL_LIB_CTX.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12621)
A provider without `provider_query_operation()` is admittedly quite
useless, yet technically the base provider functions are not mandatory
according to our documentation.
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/13058)
This replaces the older 'file:' loader that is now an engine.
It's still possible to use the older 'file:' loader by explicitly
using the engine, and tests will remain for it as long as ENGINEs are
still supported (even through deprecated).
To support this storemgmt implementation, a few internal OSSL_DECODER
modifications are needed:
- An internal function that implements most of
OSSL_DECODER_CTX_new_by_EVP_PKEY(), but operates on an already
existing OSSL_DECODER_CTX instead of allocating a new one.
- Allow direct creation of a OSSL_DECODER from an OSSL_ALGORITHM.
It isn't attached to any provider, and is only used internally, to
simply catch any DER encoded object to be passed back to the
object callback with no further checking. This implementation
becomes the last resort decoder, when all "normal"
decodation attempts (i.e. those that are supposed to result
in an OpenSSL object of some sort) have failed.
Because file_store_attach() uses BIO_tell(), we must also support
BIO_ctrl() as a libcrypto upcall.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12587)
Fix OPENSSL_realloc failure case; `provider->operation_bits` memory
is lost when `OPENSSL_realloc()` returns NULL.
`operation_bits_sz` is never set to the length of the allocated array.
This means that operation_bits is always reallocated in
`ossl_provider_set_operation_bit()`, possibly shrinking the array.
In addition, it means that the `memset()` always zeros out the
whole reallocated array, not just the new part. Also, because
`operation_bits_sz` is always zero, the value of `*result` in
`ossl_provider_test_operation_bit()` will always be zero.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/12760)
If an attempt is made to load a provider and it fails, the fall-back mechanism
should be disabled to prevent the user getting some weird happening. E.g. a
failure to load the FIPS provider should not allow the default to load as a
fall-back.
The OSSL_PROVIDER_try_load() call has been added, to allow a provider to be
loaded without disabling the fall-back mechanism if it fails.
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/12625)
The default and legacy providers currently return 1 for status and self test checks.
Added test to show the 3 different stages the self test can be run (for installation, loading and on demand).
For the fips provider:
- If the on demand self test fails, then any subsequent fetches should also fail. To implement this the
cached algorithms are flushed on failure.
- getting the self test callback in the fips provider is a bit complicated since the callback hangs off the core
libctx (as it is set by the application) not the actual fips library context. Also the callback can be set at
any time not just during the OSSL_provider_init() so it is calculated each time before doing any self test.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11752)
As the ERR_raise() is setup at this point returng a range of negative values for errors is not required.
This will need to be revisited if the code ever moves to running from the DEP.
Added a -config option to the fips install so that it can test if a fips module is loadable from configuration.
(The -verify option only uses the generated config, whereas -config uses the normal way of including the generated data via another config file).
Added more failure tests for the raised errors.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12346)
It is not a failure to call OSSL_PROVIDER_get_capabilities() with a
provider loaded that has no capabilities.
Fixes#12286
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/12292)
The new naming scheme consistently usese the `OSSL_FUNC_` prefix for all
functions which are dispatched between the core and providers.
This change includes in particular all up- and downcalls, i.e., the
dispatched functions passed from core to provider and vice versa.
- OSSL_core_ -> OSSL_FUNC_core_
- OSSL_provider_ -> OSSL_FUNC_core_
For operations and their function dispatch tables, the following convention
is used:
Type | Name (evp_generic_fetch(3)) |
---------------------|-----------------------------------|
operation | OSSL_OP_FOO |
function id | OSSL_FUNC_FOO_FUNCTION_NAME |
function "name" | OSSL_FUNC_foo_function_name |
function typedef | OSSL_FUNC_foo_function_name_fn |
function ptr getter | OSSL_FUNC_foo_function_name |
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/12222)
Provide a function to applications to query the capabilities that a
provider can perform.
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11914)
Implement a function which enables us to get hold of the provider ctx
for a loaded provider.
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11834)
Calling 'OSSL_PROVIDER_available(NULL, "default")' would search for
the "default" provider, and in doing so, activate it if necessary,
thereby detecting that it's available... and then immediately free
it, which could deactivate that provider, even though it should stay
available.
We solve this by incrementing the refcount for activated fallbacks one
extra time, thereby simulating an explicit OSSL_PROVIDER_load(), and
compensate for it with an extra ossl_provider_free() when emptying the
provider store.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11926)
Renamed some values in core_names i.e Some DH specific names were changed to use DH instead of FFC.
Added some strings values related to RSA keys.
Moved set_params related docs out of EVP_PKEY_CTX_ctrl.pod into its own file.
Updated Keyexchange and signature code and docs.
Moved some common DSA/DH docs into a shared EVP_PKEY-FFC.pod.
Moved Ed25519.pod into EVP_SIGNATURE-ED25519.pod and reworked it.
Added some usage examples. As a result of the usage examples the following change was also made:
ec allows OSSL_PKEY_PARAM_USE_COFACTOR_ECDH as a settable gen parameter.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11610)
When a desired algorithm wasn't available, we didn't register anywhere
that an attempt had been made, with the result that next time the same
attempt was made, the whole process would be done again.
To avoid this churn, we register a bit for each operation that has
been queried in the libcrypto provider object, and test it before
trying the same query and method construction loop again.
If course, if the provider has told us not to cache, we don't register
this bit.
Fixes#11814
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11842)
A provider could be linked against a different version of libcrypto than
the version of libcrypto that loaded the provider. Different versions of
libcrypto could define opaque types differently. It must never occur that
a type created in one libcrypto is used directly by the other libcrypto.
This will cause crashes.
We can "cheat" for "built-in" providers that are part of libcrypto itself,
because we know that the two libcrypto versions are the same - but not for
other providers.
To ensure this does not occur we use different types names for the handful
of opaque types that are passed between the core and providers.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11758)
There are concerns that if |prov->provctx| is populated early,
sensitive information may leak from the provider. Therefore, we use a
temporary variable, and only assign it to |prov->provctx| when the
provider init function has returned successfully.
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11777)
This macro is used to determine if certain pieces of code should
become part of the FIPS module or not. The old name was confusing.
Fixes#11538
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11539)
Starting `cnt` from 1 would work if we weren't using cnt itself to
access elements of the array returned calling the provider callback.
As it is before this commit, we have 2 problems:
- first, in the unlikely case that the incoming array was "empty" (only
contains the terminator item) we would skip past it and potentially
end up with oob reads;
- otherwise, at the end of the while loop, `cnt` will be equal to the
number of items in the input array, not 1 more. We then add 1 more to
the zalloc call to account for the library name item, and we fill all
of it (relying on zalloc to have zeroed the terminator item).
The first read access that will read the list up to the terminator
will result in a OOB read as we did not allocate enough space to also
contain the terminator.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11427)
Add a -provider option to allow providers to be loaded. This option can be
specified multiple times.
Add a -provider_path option to allow the path to providers to be specified.
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11167)
This propagates ERR_set_mark(), and ERR_clear_last_mark() and
ERR_pop_to_mark() for provider use.
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/10803)
Added an API to optionally set a self test callback.
The callback has the following 2 purposes
(1) Output information about the KAT tests.
(2) Allow the ability to corrupt one of the KAT's
The fipsinstall program uses the API.
Some KATS are not included in this PR since the required functionality did not yet exist in the provider.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/10374)
Fixes#8322
The leak-checking (and backtrace option, on some platforms) provided
by crypto-mdebug and crypto-mdebug-backtrace have been mostly neutered;
only the "make malloc fail" capability remains. OpenSSL recommends using
the compiler's leak-detection instead.
The OPENSSL_DEBUG_MEMORY environment variable is no longer used.
CRYPTO_mem_ctrl(), CRYPTO_set_mem_debug(), CRYPTO_mem_leaks(),
CRYPTO_mem_leaks_fp() and CRYPTO_mem_leaks_cb() return a failure code.
CRYPTO_mem_debug_{malloc,realloc,free}() have been removed. All of the
above are now deprecated.
Merge (now really small) mem_dbg.c into mem.c
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/10572)
The BIO_vprintf() will allow the provider to print any text, given a
BIO supplied by libcrypto.
Additionally, we add a provider library with functions to collect all
the currently supplied BIO upcalls, as well as wrappers around those
upcalls.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/10394)
The fips self test lock is deallocated in platform specific ways that may
occur after we do mem leak checking. If we don't know how to free it for
a particular platform then we just leak it deliberately. So we
temporarily disable the mem leak checking while we allocate the lock.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9939)
Currently, there are two different directories which contain internal
header files of libcrypto which are meant to be shared internally:
While header files in 'include/internal' are intended to be shared
between libcrypto and libssl, the files in 'crypto/include/internal'
are intended to be shared inside libcrypto only.
To make things complicated, the include search path is set up in such
a way that the directive #include "internal/file.h" could refer to
a file in either of these two directoroes. This makes it necessary
in some cases to add a '_int.h' suffix to some files to resolve this
ambiguity:
#include "internal/file.h" # located in 'include/internal'
#include "internal/file_int.h" # located in 'crypto/include/internal'
This commit moves the private crypto headers from
'crypto/include/internal' to 'include/crypto'
As a result, the include directives become unambiguous
#include "internal/file.h" # located in 'include/internal'
#include "crypto/file.h" # located in 'include/crypto'
hence the superfluous '_int.h' suffixes can be stripped.
The files 'store_int.h' and 'store.h' need to be treated specially;
they are joined into a single file.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9333)
This will only be required until everything is moved to providers and a NULL
provider pointer won't be possible.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9662)
More PR's related to self test will be derived from this PR.
Note: the code removed in core_get_params() was causing a freeze since the
fips module was being loaded from a config file, which then called core_get_params()
which then tried to init the config fle again...
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9596)
Quite a few adaptations are needed, most prominently the added code
to allow provider based MACs.
As part of this, all the old information functions are gone, except
for EVP_MAC_name(). Some of them will reappear later, for example
EVP_MAC_do_all() in some form.
MACs by EVP_PKEY was particularly difficult to deal with, as they
need to allocate and deallocate EVP_MAC_CTXs "under the hood", and
thereby implicitly fetch the corresponding EVP_MAC. This means that
EVP_MACs can't be constant in a EVP_MAC_CTX, as their reference count
may need to be incremented and decremented as part of the allocation
or deallocation of the EVP_MAC_CTX. It may be that other provider
based EVP operation types may need to be handled in a similar manner.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/8877)
It was argued that names like SOMETHING_set_param_types were confusing,
and a rename has been proposed to SOMETHING_settable_params, and by
consequence, SOMETHING_get_param_types is renamed
SOMETHING_gettable_params.
This changes implements this change for the dispatched provider and
core functions.
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/9591)
We move an OPENSSL_init_crypto call slightly earlier in the process to
avoid calling it while holding the store lock. This can lead to deadlocks.
Fixes the no-engine build.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/9547)
Previously we only loaded the config file by default for libssl. Now we do
it for libcrypto too.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9492)
The core now supplies its own versions of ERR_new(), ERR_set_debug()
and ERR_vset_error(). This should suffice for a provider to have any
OpenSSL compatible functionlity it desires.
The main difference between the ERR functions and the core
counterparts is that the core counterparts take an OSSL_PROVIDER
parameter instead of the library number. That way, providers do not
need to know what number they have been assigned, that information
stays in the core.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/9452)
Public function OSSL_PROVIDER_available() takes a library context and
a provider name, and returns 1 if it's available for use, i.e. if it's
possible to fetch implementations from it, otherwise 0.
Internal function ossl_provider_activated() returns 1 if the given
OSSL_PROVIDER is activated, otherwise 0.
To make this possible, the activation of fallbacks got refactored out
to a separate function, which ended up simplifying the code.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9398)
The core provides a number of essential functions as "upcalls" to
providers. Some of those were just utility functions that wrap other
upcalls - which don't seem essential and bloat the interface. We should
remove them in order to simplify the interface.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/9432)
Also make OPENSSL_hexstr2buf available to providers.
EVP control functions need hexstring conversion, so move any
memory-allocating functions in o_str.c into new file mem_str.c
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/8886)
The FIPS module inner provider doesn't need to deal with error reason
strings or error library number, since it uses the outer provider's
error reporting upcalls. We therefore disable that code in
crypto/provider_core.c when building the FIPS module.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9174)
The idea is that providers should only have to report a reason code.
The library code is considered to be libcrypto internal, and are
allocated dynamically and automatically for providers on creation.
We reserve the upper 8 bits of the reason code for internal OpenSSL
use. This allows our own providers to report errors in form of a
packed number that includes library number, function number and
reason number.
With this, a provider can potentially use any reason number it wants
from 1 to 16777216, although the current error semantics really only
allow 1 to 4095 (because only the lower 12 bits are currently
considered an actual reason code by the ERR subsystem).
A provider can provide a reason string table in form of an array of
ERR_STRING_DATA, with each item containing just the reason code and
the associated string, with the dispatch function numbered
OSSL_FUNC_PROVIDER_GET_REASON_STRINGS matching the type
OSSL_provider_get_reason_strings_fn.
If available, libcrypto will call that function on provider
activation.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9174)
