Use a write lock during ossl_provider_find() A "find" operation on a stack can end up sorting the underlying stack. In this case it is necessary to use a "write" lock to synchronise access to the stack across multiple threads. Reviewed-by: Tomas Mraz <tomas@openssl.org> (Merged from https://github.com/openssl/openssl/pull/16980)
diff --git a/crypto/provider_core.c b/crypto/provider_core.c index c9dba6d..a46a96c 100644 --- a/crypto/provider_core.c +++ b/crypto/provider_core.c
@@ -423,7 +423,11 @@ #endif tmpl.name = (char *)name; - if (!CRYPTO_THREAD_read_lock(store->lock)) + /* + * A "find" operation can sort the stack, and therefore a write lock is + * required. + */ + if (!CRYPTO_THREAD_write_lock(store->lock)) return NULL; if ((i = sk_OSSL_PROVIDER_find(store->providers, &tmpl)) != -1) prov = sk_OSSL_PROVIDER_value(store->providers, i);
diff --git a/doc/man3/DEFINE_STACK_OF.pod b/doc/man3/DEFINE_STACK_OF.pod index d715246..ec9eda8 100644 --- a/doc/man3/DEFINE_STACK_OF.pod +++ b/doc/man3/DEFINE_STACK_OF.pod
@@ -178,7 +178,10 @@ B<sk_I<TYPE>_find>() returns the index of a matching element or B<-1> if there is no match. Note that, in this case the comparison function will usually compare the values pointed to rather than the pointers themselves and -the order of elements in I<sk> can change. +the order of elements in I<sk> can change. Note that because the stack may be +sorted as the result of a B<sk_I<TYPE>_find>() call, if a lock is being used to +synchronise access to the stack across multiple threads, then that lock must be +a "write" lock. B<sk_I<TYPE>_find_ex>() operates like B<sk_I<TYPE>_find>() except when a comparison function has been specified and no matching element is found.