FreeBSD support by michael-yuji · Pull Request #3398 · apple/swift-nio

michael-yuji · 2025-10-06T01:16:27Z

This PR add FreeBSD support, currently build is passing against main swift branch, tests are not fully building but does not seems related to swift-nio itself

Apart from using kqueue, here's a list of non-trivial changes:

File copying

For file copying, we use copy_file_range instead of sendfile as it is available on all supported FreeBSD option, so now, copyItem's implementation is like this:

Platform	syscall
Darwin	copyfile(2)
FreeBSD	copy_file_range(2)
Linux	sendfile(2)

extended attributes

Unlike Linux and Darwin, the extended attribute apis are different in freebsd, e.g. extattr_get_fd instead of getxattr

pktinfo

FreeBSD does not have IP_PKTINFO for IPv4, instead we combine IP_RECVIF and IP_ORIGDSTADDR to get the destination address and interface index.

|swift-nio/Tests/NIOFSFoundationCompatTests/FileSystemFoundationCompatTests.swift:75:34: error: no exact matches in call to initializer
73 |         }
74 |
75 |         let contents = try await Data(contentsOf: path, maximumSizeAllowed: .bytes(1024))
   |                                  |- error: no exact matches in call to initializer
   |                                  |- note: found candidate with type '(NIOFilePath) -> Data'
   |                                  `- note: candidate has partially matching parameter list (contentsOf: __shared URL, options: Data.ReadingOptions)

Lukasa · 2025-10-06T07:31:23Z

@michael-yuji Has that issue been reported on Foundation? Not having the tests build is definitely going to be hiding bugs.

Lukasa · 2025-10-06T07:31:51Z

In the meantime, as Foundation is unlikely to fix the issue in 6.2, can I suggest we skip the test until 6.3 on FreeBSD?

Lukasa · 2025-10-06T07:32:49Z

Sources/_NIOFileSystem/Internal/System Calls/FileDescriptor+Syscalls.swift

 }

-#if canImport(Glibc) || canImport(Musl) || canImport(Bionic)
+#if !os(FreeBSD) && (canImport(Glibc) || canImport(Musl) || canImport(Bionic))


Under what circumstance would this check have previously compiled? Does FreeBSD report its libc as Glibc?

I see from the rest of the code that it does. Is there any way we can change that before formal support? It's going to make life very difficult, as well as being factually untrue.

Unfortunately FreeBSD does report its libc as Glibc, I do have a PR open (swiftlang/swift#79261) but we have decided that that's no bandwidth to make the overlay change into 6.2.x
cc @etcwilde

Sources/_NIOFileSystem/Internal/System Calls/Syscall.swift

Lukasa · 2025-10-06T07:38:00Z

Sources/_NIOFileSystem/Internal/SystemFileHandle.swift

                    // Okay, return empty value.
                    return .success([])
                #endif
+                #if !os(FreeBSD)


Is this here because noData is impossible or because it's unreachable?

it's because .nodata is not available on this platform

Lukasa · 2025-10-06T07:38:30Z

Sources/_NIOFileSystem/Internal/SystemFileHandle.swift

+                #endif
                default:
+                    #if os(FreeBSD)
+                    let error = FileSystemError.extattr_get_fd(


Does this behave meaningfully differently than fgetxattr? If not, can this #if os be moved into that function so it doesn't end up here?

fgetxattr is not available, the extended attribute apis are called differently

If it only has a different name, but otherwise behaves the same, please merge them into the same function.

Lukasa · 2025-10-06T07:39:42Z

Sources/_NIOFileSystem/FileSystem.swift

        }
-
-        #elseif canImport(Glibc) || canImport(Musl) || canImport(Bionic)
+        #elseif os(FreeBSD) || canImport(Glibc) || canImport(Musl) || canImport(Bionic)


canImport(Glibc) covers os(FreeBSD) currently.

Lukasa · 2025-10-06T07:48:28Z

Sources/NIOPosix/SelectorKqueue.swift

+        #if os(FreeBSD)
+        self.fflags = 0
+        self.data = 0
+        #else


It's not clear to me that this is right: as noted above, EVFILT_EXCEPT does appear to be available on FreeBSD, and I think we actually do want it.

Lukasa · 2025-10-06T07:49:40Z

Sources/NIOPosix/System.swift

    @inline(never)
    public static func opendir(pathname: String) throws -> OpaquePointer {
+#if os(FreeBSD)
+        sysOpendir(pathname)!


Please wrap this in the syscall helper.

Tests/NIOFSTests/Internal/SyscallTests.swift

Lukasa · 2025-10-06T07:50:55Z

Tests/NIOFSTests/FileHandleTests.swift

+        var nonThrowingErrnos: [Errno] = []
        var knownErrnos: [Errno: FileSystemError.Code] = [.notSupported: .unsupported]
-        #if canImport(Darwin)
+        #if canImport(Darwin) || os(FreeBSD)


This represents a behavioural change on macOS, which no longer has noData in its nonThrowingErrnos.

Lukasa · 2025-10-06T07:52:28Z

Package.swift

 // These platforms require a dependency on `NIOPosix` from `NIOHTTP1` to maintain backward
 // compatibility with previous NIO versions.
-let historicalNIOPosixDependencyRequired: [Platform] = [.macOS, .iOS, .tvOS, .watchOS, .linux, .android]
+let historicalNIOPosixDependencyRequired: [Platform] = [.macOS, .iOS, .tvOS, .watchOS, .linux, .android, .custom("freebsd"), .custom("FreeBSD")]


These platforms do not require a historical NIOPosix dependency, as they did not compile before.

michael-yuji · 2025-10-06T08:32:10Z

@michael-yuji Has that issue been reported on Foundation? Not having the tests build is definitely going to be hiding bugs.

I don't think it's a bug in Foundation, the initializer seems to come from Sources/_NIOFileSystemFoundationCompat/Data+FileSystem.switf. I suspect it's a bug in the current toolchain at main

skipping the test now sounds good to me

Lukasa · 2025-10-06T16:18:45Z

Let's also not merge this until swiftlang/swift#79261 lands, because the compiler guards will all change at that point.

michael-yuji · 2025-10-06T16:23:58Z

Let's also not merge this until swiftlang/swift#79261 lands, because the compiler guards will all change at that point.

I'm not sure that's a good idea, since we have announced support for FreeBSD and without swift-nio, anything swift on server would pretty much become useless on the platform which is less than ideal

Lukasa · 2025-10-06T16:40:41Z

Let me be more precise: if we land the change as-is, you must not change the name of the libc in FreeBSD, as doing so will be source-breaking. So, to put not too fine a point on it: if the above PR doesn't land now, it doesn't land until a source break is acceptable.

3405691582 · 2025-10-07T12:39:15Z

Sources/NIOFS/Internal/System Calls/Syscalls.swift

+}
+#endif
+
+#if !os(FreeBSD) && (canImport(Glibc) || canImport(Musl) || canImport(Android))


Would it be stylistically better to have

#if !os(FreeBSD) #if canImport(Glibc) ... #endif #endif

or retain the rather complex boolean expression here?

I ran into this when considering OpenBSD support; the stacked conditionals are easier to read but I don't know if there's a stylistic preference.

3405691582 · 2025-10-07T12:46:45Z

Sources/CNIOFreeBSD/shim.c

I erred on calling this just CNIOBSD in #3394. Would it be better to do the same here, so that we don't need to have one CNIO.* module per operating system? Or should I go and rename CNIOBSD to CNIOOpenBSD there?

Do the two have similar-enough libcs?

There are quite similar, especially due to pedigree, but there is definitely divergence, such as functionality in FreeBSD that OpenBSD does not have.

Let's keep them different then, at least for now.

weissi · 2025-10-08T21:30:08Z

Sources/CNIOFreeBSD/include/CNIOFreeBSD.h

+//
+// This source file is part of the SwiftNIO open source project
+//
+// Copyright (c) 2017-2018 Apple Inc. and the SwiftNIO project authors


Suggested change

// Copyright (c) 2017-2018 Apple Inc. and the SwiftNIO project authors

// Copyright (c) 2025 Apple Inc. and the SwiftNIO project authors

weissi · 2025-10-08T21:30:26Z

Sources/CNIOFreeBSD/shim.c

+//
+// This source file is part of the SwiftNIO open source project
+//
+// Copyright (c) 2017-2018 Apple Inc. and the SwiftNIO project authors


Suggested change

// Copyright (c) 2017-2018 Apple Inc. and the SwiftNIO project authors

// Copyright (c) 2025 Apple Inc. and the SwiftNIO project authors

elsewhere too

Lukasa

Ok @michael-yuji, I'm happy enough to go ahead with the Glibc module name for now. There are several outstanding pieces of review feedback, do you mind picking them up?

michael-yuji · 2025-10-19T00:53:47Z

Thanks @Lukasa, I'll picking those up hopefully sometime around next week.

michael-yuji added 8 commits October 5, 2025 16:37

FreeBSD lock/NSLock fixes

93729df

use kqueue selector for FreeBSD

c6c6e4f

preliminary FreeBSD support

e7270f0

remove clangd.core

485e39e

adjust pthread_cond_t compilation condition

b7a3bb3

fix linux build

795c611

fix linux build cont.

3d1d358

FreeBSD: use IP_RECVIF + IP_ORIGDSTADDR to work around IP_PKTINFO

8266590

Lukasa added the 🆕 semver/minor Adds new public API. label Oct 6, 2025

Lukasa reviewed Oct 6, 2025

View reviewed changes

3405691582 reviewed Oct 7, 2025

View reviewed changes

clean up

86e0ded

weissi reviewed Oct 8, 2025

View reviewed changes

Merge branch 'main' into freebsd-wip-main

37dec2f

Lukasa reviewed Oct 17, 2025

View reviewed changes

Lukasa mentioned this pull request Oct 22, 2025

Partial OpenBSD support. #3394

Merged

kkebo mentioned this pull request Jan 26, 2026

Add partial support for FreeBSD #3494

Open

	// Copyright (c) 2017-2018 Apple Inc. and the SwiftNIO project authors
	// Copyright (c) 2025 Apple Inc. and the SwiftNIO project authors

Conversation

michael-yuji commented Oct 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

File copying

extended attributes

pktinfo

Uh oh!

Lukasa commented Oct 6, 2025

Uh oh!

Lukasa commented Oct 6, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michael-yuji commented Oct 6, 2025

Uh oh!

Lukasa commented Oct 6, 2025

Uh oh!

michael-yuji commented Oct 6, 2025

Uh oh!

Lukasa commented Oct 6, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Lukasa left a comment

Choose a reason for hiding this comment

Uh oh!

michael-yuji commented Oct 19, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

michael-yuji commented Oct 6, 2025 •

edited

Loading