> On 14 Aug 2022, at 15:51, Kyle Evans <kev...@freebsd.org> wrote:
>
> On Sun, Aug 14, 2022 at 2:22 AM Alexander V. Chernikov
> <melif...@freebsd.org> wrote:
>>
>> The branch main has been updated by melifaro:
>>
>> URL:
>> https://cgit.FreeBSD.org/src/commit/?id=9b967bd65de6647aed68a141dc34f9b223a2593c
>>
>> commit 9b967bd65de6647aed68a141dc34f9b223a2593c
>> Author: Alexander V. Chernikov <melif...@freebsd.org>
>> AuthorDate: 2022-08-12 13:36:53 +0000
>> Commit: Alexander V. Chernikov <melif...@freebsd.org>
>> CommitDate: 2022-08-14 09:22:33 +0000
>>
>> domains: allow domains to be unloaded
>>
>> Add domain_remove() SYSUNINT callback that removes the domain
>> from the domain list if it has DOMF_UNLOADABLE flag set.
>> This change is required to support netlink ( D36002 ).
>>
>> Reviewed by: glebius
>> Differential Revision: https://reviews.freebsd.org/D36173
>> ---
>> sys/kern/uipc_domain.c | 23 +++++++++++++++++++++++
>> sys/sys/domain.h | 4 ++++
>> 2 files changed, 27 insertions(+)
>>
>> diff --git a/sys/kern/uipc_domain.c b/sys/kern/uipc_domain.c
>> index c70b3150cbf0..2cae08be089a 100644
>> --- a/sys/kern/uipc_domain.c
>> +++ b/sys/kern/uipc_domain.c
>> @@ -239,6 +239,29 @@ domain_add(void *data)
>> mtx_unlock(&dom_mtx);
>> }
>>
>> +void
>> +domain_remove(void *data)
>> +{
>> + struct domain *dp = (struct domain *)data;
>> +
>> + if ((dp->dom_flags & DOMF_UNLOADABLE) == 0)
>> + return;
>> +
>> + mtx_lock(&dom_mtx);
>> + if (domains == dp) {
>> + domains = dp->dom_next;
>> + } else {
>> + struct domain *curr;
>> + for (curr = domains; curr != NULL; curr = curr->dom_next) {
>> + if (curr->dom_next == dp) {
>> + curr->dom_next = dp->dom_next;
>> + break;
>> + }
>> + }
>> + }
>> + mtx_unlock(&dom_mtx);
>> +}
>> +
>> /* ARGSUSED*/
>> static void
>> domaininit(void *dummy)
>
> While it's not important for the one domain you care about, this
> should likely also check if any contained protocols have
> pr_fasttimo/pr_slowtimo and purge them from the appropriate list, just
> to be technically correct. (Or at least assert under INVARIANTS that
> they do not)
That’s a good point!
I should have provided more context in the commit message.
There is some ongoing work glebius@ is doing in the domain space.
Specifically, there are plans to remove this callbacks in the near future -
https://reviews.freebsd.org/D36163 stack provides more reasoning and
implementation details.
Additionally, the KPI may be a bit fluid and may change in the near/mid-term
future - that’s why the minimal working version was committed.
