smb2: The Pure-Rust SMB Client That Outruns macOS's Own

Copy a few gigabytes off a network drive and you quietly accept a speed limit — whatever your operating system's own file client can manage feels like a law of nature.

On a Gigabit NAS, smb2 breaks that law. The pure-Rust SMB client, in benchmarks its author published, beats the native macOS SMB client on every operation — by as much as 5× on downloads — and it has only been on crates.io since April.

That alone makes it worth a look. What makes it worth a verdict is everything around the speed: a test suite most funded libraries would envy, a refreshingly honest list of limitations, and a provenance question every engineering leader now has to weigh.

TL;DR — Trial. Put smb2 on a real but non-critical path today. The thing keeping it from a clean "Adopt" isn't the code; it's the bus factor.

(This is a Rust Crate Radar Deep Dive — every verdict is checked against the crate's live maintenance and activity, not vibes.)

What it actually is

smb2 is an async, runtime-agnostic SMB 2.x/3.x client written straight from Microsoft's MS-SMB2 specification. There is no C library underneath it and no SMB1 — a deliberate, security-minded omission. Talking to a Windows share or a Samba server from Rust used to mean binding to a C client; smb2 makes it a pure-Cargo dependency.

The API is the boring kind, which is a compliment:

let mut client = smb2::connect(&addr, &user, &pass).await?;
let mut share = client.connect_share("Documents").await?;
let data = client.read_file(&mut share, "report.pdf").await?;

A runnable version lives in the series' examples repo, on the Digest #1 branch.

The benchmark that earns attention

On a Gigabit NAS, smb2 beats the native macOS SMB client on every operation — and runs 3–8× faster than the existing smb crate.

The speed isn't an accident of micro-optimization; it's architectural. smb2 pipelines its I/O — it batches compound requests and rides a sliding window with adaptive chunking, so it isn't idling between round-trips the way a request-response client does. On a high-latency or lossy link, that design is the difference, and it's exactly where a from-the-spec implementation can out-engineer a decades-old default.

What actually earns the verdict: the testing

Speed is easy to stage in a benchmark; durability is not. The reason smb2 reaches Trial rather than Assess is its test discipline, which is unusual for a crate this young: roughly 970 tests, property tests, 12 fuzz targets, and — the part worth pausing on — integration tests against 14 Docker-based Samba containers in CI, covering encryption-required servers, high latency, dropped connections, and tiny read sizes. A testing feature lets your application spin up the same containers, so you can prove the integration against your own access patterns before you commit.

That is more rigor than many funded, decade-old libraries carry. It is the single strongest signal that this isn't a weekend project.

The catch

Two things hold smb2 back from a clean "Adopt," and the author is upfront about both.

It is a single maintainer — David Veszelovszki — which is real bus-factor risk for anything you'd build a product on. And it was written with heavy AI assistance, stated plainly in the project's own notes. The test suite above is precisely what makes that provenance defensible: the code is held to spec by machinery, not vibes. But "one person, AI-assisted" is a profile to weigh deliberately, not wave through.

It is also not yet feature-complete against its older rival. The more established smb crate has multi-channel and QUIC/RDMA support that smb2 omits. If you need those, this isn't your client yet.

Evaluation

Run smb2 through the Crate Radar evaluation tool and the maintenance picture is healthy, which is what tips the trade-offs toward Trial rather than Assess:

Problem fit — Strong. Pure-Rust SMB with no C dependency is a real, under-served need.
Maturity — Early. 0.11.3, pre-1.0, first published April 2026; expect breaking changes.
Activity — Active. Last commit six days ago; eleven releases in the past year.
Stewardship — Watch. Single maintainer, AI-assisted; offset by an exceptional test suite.
Ecosystem fit — Strong. Async and runtime-agnostic; MSRV 1.85; dual MIT/Apache.
Exit cost — Low. A client behind a thin call surface; swapping it back out is cheap.

The verdict

Trial. If you touch SMB from Rust, put smb2 on a real but non-critical path now — a sync job, an internal tool, an importer — and lean on its container test harness to prove it against your servers. The benchmarks are real and the testing is better than its age suggests.

The one thing to keep your eye on isn't a line of code. It's whether a second maintainer ever shows up. The day this stops being one person's project is the day it graduates to Adopt.

Next on the Radar: a Category Showdown of async runtimes. Got a crate you want put under the tool? Get in touch.

Sources: smb2 on GitHub · crates.io · author's benchmarks & write-up