The 'sys/net/if_tun_test:transient' testcase is now failing in CI: ===> sys/net/if_tun_test:transient Result: failed: Test case body timed out Standard output: Executing command [ jexec one ifconfig tun0 ] I can reproduce this error locally using Bricoler with 100 runs for good measure: $ bricoler run freebsd-src-regression-suite --param freebsd-src:url=/usr/src --param freebsd-src:branch= --param freebsd-src-regression-suite:hypervisor=bhyve --param freebsd-src-regression-suite:memory=4096 --param freebsd-src-regression-suite:ncpus=2 --param freebsd-src-regression-suite:parallelism=1 --param freebsd-src-regression-suite:count=100 --param freebsd-src-regression-suite:tests='sys/net/if_tun_test:transient' @kib, I have bisected the failure to your commit https://cgit.freebsd.org/src/commit/?id=4dbe6628179d8e6bf400bfdb4bfa869bdc102a56 (devfs: make destroy_dev() a release barrier for cdevpriv destructors runs). Please triage as necessary. @kevans, cc'd you since you added the tests in commit https://cgit.freebsd.org/src/commit/?id=a1174b3b1174754b1f69406bff4456d002e8f583 (tun(4)/tap(4): allow devices to be configured as transient).
Yeah, I could see that... I'll take this one.
(The problem is that transient tunnels will destroy the cloner and thus, the cdev, from the dtor; thus, we can't make any progress)
I have a change prepared that moves removal of the tun device into a taskqueue, I'm just working through the different bits we need to consider. The test is probably technically flakey after this if left untouched because we no longer 'guarantee' (it was never a written one) that the interface has been removed by the time close(2) returns. I'm not sure how much I care in practice: if you want it to be transient anyways, you might as well take advantage of /dev/tun cloning to avoid relying on an arbitrary unit number.
https://reviews.freebsd.org/D53438
A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/src/commit/?id=96c1d8db39dfeea78ea3f27d67649252a39bbf2e commit 96c1d8db39dfeea78ea3f27d67649252a39bbf2e Author: Kyle Evans <kevans@FreeBSD.org> AuthorDate: 2025-11-05 00:28:02 +0000 Commit: Kyle Evans <kevans@FreeBSD.org> CommitDate: 2025-11-05 00:28:02 +0000 if_tuntap: defer transient destroy_dev() to a taskqueue We're in the dtor, so we can't destroy it now without deadlocking after recent changes to make destroy_dev() provide a barrier. However, we know there isn't any other dtor to run, so we can go ahead and clean up our state and just prevent a use-after-free if someone races to open the device while we're trying to destroy it. tunopen() now uses the net epoch to protect against softc release by a concurrent tun_destroy(). While we're here, allow a destroy operation to proceed if we caught a signal in cv_wait_sig() but tun_busy dropped to 0 while we were waiting to acquire the lock. This was more of an inherent design flaw, rather than a bug in the below-refed commit. PR: 290575 Fixes: 4dbe6628179d ("devfs: make destroy_dev() a release [...]") Reviewed by: kib, markj Differential Revision: https://reviews.freebsd.org/D53438 sys/net/if_tuntap.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 57 insertions(+), 6 deletions(-)
The bug hasn't been MFC'd and the fix has landed in main; thanks for the report-