moonchen opened a new pull request, #13307: URL: https://github.com/apache/trafficserver/pull/13307
## Summary `dnsBufAllocator` is declared `ClassAllocator<HostEnt, false>` (i.e. `Destruct_on_free = false`), so `HostEnt::free()` returns the block to the free pool without running `~HostEnt`. `HostEnt` embeds an `SRVHosts` whose `std::vector<SRV> hosts` owns heap storage. `ClassAllocator::alloc()` always placement-constructs (`::new (ptr) C(...)`), so the next allocation of a recycled block default-constructs a fresh empty vector over it, overwriting and orphaning the previous buffer. Result: every recycle of a `HostEnt` that held one or more SRV records (populated via `srv_hosts.hosts.push_back` in `dns_process`) leaks one vector buffer, up to `hostdb_round_robin_max_count` × `sizeof(SRV)` bytes. The leak is conditional on `T_SRV` responses; A/AAAA/PTR responses leave the vector empty and are unaffected. This is a long-standing latent leak: it was harmless while `SRVHosts::hosts` was a fixed array and became a leak once it changed to `std::vector`. ## Fix Drop the `false` and use the default `Destruct_on_free = true` so `~HostEnt` (hence `~vector<SRV>`) runs on free. `alloc()` already placement-constructs, so construction and destruction stay balanced. `HostEnt::free()` (reached only when a `Ptr<HostEnt>` refcount hits zero) is the sole destruction path, no `HostEnt` field is read after free, and the type is never treated as POD, so running the destructor is safe. The hot A/AAAA path only destructs an empty vector. ## Testing Verified under AddressSanitizer/LeakSanitizer with a standalone reproduction faithful to the `ClassAllocator` alloc/free semantics: 1000 SRV recycle cycles leak ~16 MB with `Destruct_on_free = false` and zero with `true`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
