A manual, empirical study of 850 uses of ‘unsafe’ in Rust and of 170 bugs (70 memory safety, 100 concurrency) based on studies of Servo, Tock, Parity Ethereum, TikV, Redox and five libraries and with most (145) of the bugs being in code after Rust 1.0 (2016). [I think that the choice has a bias towards system code so this may be a biased sample?] (Contrast with the much larger, automated studies in astrauskas:oopsla:2020 and evans:icse:2020.) Also describes a use-after-free checker and a double-lock checker for checking unsafe code.
They use the term ‘interior unsafe’ for a function that contains an unsafe block but that is not annotated as unsafe.
In their analysis of bugs, they distinguish four categories based on the ‘error propagation chain’ and whether it starts in safe or unsafe code and whether it ends in safe or unsafe code. e.g., if a safe function breaks an invariant which causes a later unsafe code block to perform an out-of-bounds access, that is a “safe → unsafe” bug.
Most interior unsafe functions ensure safety through type invariants instead of by performing runtime checks. That is, by restricting how their arguments can be constructed so that they can ensure that their arguments satisfy the invariant.
Uninitialized data can be a problem because assigning to that data will cause the previous contents to be dropped.
Use after free (“unsafe → safe”) can happen when safe code drops an object (because its lifetime ends) but unsafe code still accesses the object.
misunderstanding of lifetime [is] the main reason for most use-after-free and many other types of memory-safety bugs.
Deadlock (called “blocking bugs”) are all caused by interior unsafe code. They are broken down by type of lock (mutex, condvar, etc.) and involve failing to acquire locks, inconsistent lock ordering problems, forgetting to unlock.
What makes locking issues especially hard is that it is hard to understand the boundaries of critical sections because ‘unlock()’ is implicitly called at the end of variable lifetimes. They suggest tools could help highlight the range of lifetimes and critical sections and suggest that programmers explicitly use
mem::drop()both to fix problems and to be explicit about the end of critical sections.
Race conditions (called “non-blocking bugs”) are mostly caused by failure to protect shared resources or (rarely) message passing.
The most common issue is passing raw pointers (since they lack lifetime information and ownership checks); followed by sharing results from OS syscalls or hardware resources; shared mutable global variables; and incorrectly providing adding Sync on a struct.
They suggest careful review of code that implements Sync.