Bug #19873: Thread#native_thread_id is incorrectly cached across fork on Linux

• Status: Closed
• Assignee: -
• Target version: -
• Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN

[kjtsanaktsidis (KJ Tsanaktsidis)]

On Linux, when a process forks, the native thread ID of the (only) thread in the forked child should change.

>> libc = Fiddle.dlopen(nil)
=> #<Fiddle::Handle:0x0000ffffb0d8e900>
>> gettid = Fiddle::Function.new(libc['gettid'], [], Fiddle::TYPE_INT)
=>
#<Fiddle::Function:0x0000ffffb15d1130
...
>> gettid.call
=> 33
>> fork { puts gettid.call }; nil;
=> nil
36

However, the value of Thread#native_thread_id is (incorrectly) still set to the old value.

>> Thread.current.native_thread_id
=> 33
>> fork { puts Thread.current.native_thread_id }; nil;
=> nil
33

I think #native_thread_id should be re-fetched from the operating system via gettid(2) after fork.

This patch should fix it I think? https://github.com/ruby/ruby/pull/8418

[kjtsanaktsidis (KJ Tsanaktsidis)]

• Description updated (diff)

• Status changed from Open to Closed

Applied in changeset git|0117a6d389c5b607eaec1b7f917132efa356f665.

---

Fix Thread#native_thread_id being cached across fork (#8418)

The native thread ID can and does change on some operating systems (e.g.
Linux) after forking, so it needs to be re-queried.

[Bug #19873]