The Candid library causes a Denial of Service while parsing a specially crafted payload with 'empty' data type. For example, if the payload is `record { * ; empty }` and the canister interface expects `record { * }` then the Rust candid decoder treats empty as an extra field required by the type. The problem with the type empty is that the candid Rust library wrongly categorizes empty as a recoverable error when skipping the field and thus causing an infinite decoding loop. Canisters using affected versions of candid are exposed to denial of service by causing the decoding to run indefinitely until the canister traps due to reaching maximum instruction limit per execution round. Repeated exposure to the payload will result in degraded performance of the canister. Note: Canisters written in Motoko are unaffected.

Motoko's incremental garbage collector is impacted by an uninitialized memory access bug, caused by incorrect use of write barriers in a few locations. This vulnerability could potentially allow unauthorized read or write access to a Canister's memory. However, exploiting this bug requires the Canister to enable the incremental garbage collector or enhanced orthogonal persistence, which are non-default features in Motoko.

Impact: The library offers a function to generate an ed25519 key pair via Ed25519KeyIdentity.generate with an optional param to provide a 32 byte seed value, which will then be used as the secret key. When no seed value is provided, it is expected that the library generates the secret key using secure randomness. However, a recent change broke this guarantee and uses an insecure seed for key pair generation. Since the private key of this identity (535yc-uxytb-gfk7h-tny7p-vjkoe-i4krp-3qmcl-uqfgr-cpgej-yqtjq-rqe) is compromised, one could lose funds associated with the principal on ledgers or lose access to a canister where this principal is the controller.

When storing unbounded types in a BTreeMap, a node is represented as a linked list of "memory chunks". It was discovered recently that when we deallocate a node, in some cases only the first memory chunk is deallocated, and the rest of the memory chunks remain (incorrectly) allocated, causing a memory leak. In the worst case, depending on how a canister uses the BTreeMap, an adversary could interact with the canister through its API and trigger interactions with the map that keep consuming memory due to the memory leak. This could potentially lead to using an excessive amount of memory, or even running out of memory. This issue has been fixed in #212 https://github.com/dfinity/stable-structures/pull/212  by changing the logic for deallocating nodes to ensure that all of a node's memory chunks are deallocated and users are asked to upgrade to version 0.6.4.. Tests have been added to prevent regressions of this nature moving forward. Note: Users of stable-structure < 0.6.0 are not affected. Users who are not storing unbounded types in BTreeMap are not affected and do not need to upgrade. Otherwise, an upgrade to version 0.6.4 is necessary.