Re: ng_socket refcount bug?

On Fri, Jun 19, 2026 at 04:15:29PM -0400, Kristof Provost wrote:
K> I pretty consistently see a panic running if_pppoe tests.
K> These tests use `/usr/libexec/pppoed -l pppoe-pap epair0b` (or similar) to
K> be a PPPoE server in one vnet (so that I can test if_pppoe in another
K> vnet).
K> 
K> Unfortunately I can’t share the if_pppoe module required to run them.
K> 
K> They reliably panic like this:
K> 
K> 	panic: ng node  won't die
K> 	cpuid = 39
K> 	time = 1781899222
K> 	KDB: stack backtrace:
K> 	db_trace_self_wrapper() at db_trace_self_wrapper+0x36/frame
K> 0xfffffe004ac1cb40
K> 	vpanic() at vpanic+0x149/frame 0xfffffe004ac1cc70
K> 	panic() at panic+0x43/frame 0xfffffe004ac1ccd0
K> 	vnet_netgraph_uninit() at vnet_netgraph_uninit+0x171/frame
K> 0xfffffe004ac1cd00
K> 	vnet_destroy() at vnet_destroy+0x154/frame 0xfffffe004ac1cd30
K> 	prison_deref() at prison_deref+0xb15/frame 0xfffffe004ac1cdb0
K> 	sys_jail_remove() at sys_jail_remove+0x1a7/frame 0xfffffe004ac1cdf0
K> 	amd64_syscall() at amd64_syscall+0x179/frame 0xfffffe004ac1cf30
K> 	fast_syscall_common() at fast_syscall_common+0xf8/frame 0xfffffe004ac1cf30
K> 	--- syscall (508, FreeBSD ELF64, jail_remove), rip = 0xfe32964660a, rsp =
K> 0xfe324ddabb8, rbp = 0xfe324ddac40 ---
K> 	KDB: enter: panic
K> 	[ thread pid 17271 tid 102985 ]
K> 	Stopped at      kdb_enter+0x33: movq    $0,0x15abeb2(%rip)

Fix pushed in 4d3cbeb7b407.

-- 
Gleb Smirnoff