Re: Is there any plan for ZFS and timerfd updates on stable/14?
Date: Sun, 17 Sep 2023 00:39:50 UTC
On Sun, 3 Sep 2023 13:33:28 +0900 Tomoaki AOKI <junchoon@dec.sakura.ne.jp> wrote: > On Sat, 2 Sep 2023 22:47:53 -0500 > Jake Freeland <jake@technologyfriends.net> wrote: > > > On Sat, Sep 2, 2023 at 10:40 PM Warner Losh <imp@bsdimp.com> wrote: > > > > > > > > > > > On Sat, Sep 2, 2023, 9:36 PM Jake Freeland <jake@technologyfriends.net> > > > wrote: > > > > > >> On Sat, Sep 2, 2023 at 10:31 PM Tomoaki AOKI <junchoon@dec.sakura.ne.jp> > > >> wrote: > > >> > > >>> Hi. > > >>> > > >>> There are discussions about deadlocks issue of ZFS on freebsd-current > > >>> ML, starting from [1] last month. > > >>> IIRC, at least some fixes (candidates?) are merged to main, but not yet > > >>> to stable/14. > > >>> > > >>> Upcoming (aleready released? or still rc3?) OpenZFS 2.2-release seems > > >>> to have most of them. So my 1st question is "Is there any plan to > > >>> import vendor/openzfs/zfs-2.2-release into stable/14 BEFORE BRANCHING > > >>> releng/14? > > >>> > > >>> And one more. timerfd is added at last-minutes BEFORE stable/14 is > > >>> branched, and already have not-yet-MFC'ed fixes [2], [3], [4] and > > >>> Differential revision D41600 on Phablicator [5] related to memory leaks > > >>> and locks. > > >>> Additionally, splitting out lib32 part to proper place is proposed > > >>> as D41640 [6]. Both [5] and [6] are accepted but not yet landed. > > >>> Also, D41641 [7] proposes namespace pollution adjustments. This can be > > >>> optional? > > >>> > > >>> Memory leaks and improper locks can lead system to security issues or > > >>> deadlocks, so it would be benefical if landed and MFC'ed BEFORE > > >>> releng/14 branches. > > >>> > > >>> Is there any plan to do so? At least, existing deadlocks should be > > >>> considered as SHOW-STOPPER and resolved. > > >>> > > >> > > >> The plan is to get all of those patches in before releng/14.0, I believe. > > >> > > >> What are your thoughts, Warner? > > >> > > > > > > Sounds like the reviews are done or nearly so. I've not had time to look > > > closely to be sure... I'd planned on making time Tuesday morning. > > > > > > > Yes. All reviews are good to go. > > > > Jake Freeland > > Glad to know. Thanks! > Looking forward to see them landed / MFC'ed before releng/14 branches. > > Regards. > > > > > > > > > > > Warner > > > > > > > > >> Thanks, > > >> Jake Freeland > > >> > > >> > > >>> > > >>> I myself am bitten by several deadlocks on poudriere full builds after > > >>> upgrading base from stable/13 to stable/14, finally finished with > > >>> increasing kern.maxvnodes after powercycle on each deadlock and > > >>> continue. > > >>> > > >>> > > >>> Thanks in advance! > > >>> > > >>> [1] > > >>> > > >>> https://lists.freebsd.org/archives/freebsd-current/2023-August/004162.html > > >>> > > >>> [2] > > >>> > > >>> https://cgit.freebsd.org/src/commit/?id=02f534b57f84d6f4f97c337b05b383c8b3aaf18c > > >>> > > >>> [3] > > >>> > > >>> https://cgit.freebsd.org/src/commit/?id=5eab523053db79b4bd4f926c7d7ac04444d9c1da > > >>> > > >>> [4] > > >>> > > >>> https://cgit.freebsd.org/src/commit/?id=f4296cfb409a48de00bfa60e76f686c2b031876f > > >>> > > >>> [5] https://reviews.freebsd.org/D41600 > > >>> > > >>> [6] https://reviews.freebsd.org/D41640 > > >>> > > >>> [7] https://reviews.freebsd.org/D41641 > > >>> > > >>> -- > > >>> Tomoaki AOKI <junchoon@dec.sakura.ne.jp> > > -- > Tomoaki AOKI <junchoon@dec.sakura.ne.jp> Hi. Thanks for your hard work on it. I could confirm commits to main as below, but they are not yet MFC'ed/MFS'/ed. It would be nice these commits to be incorporated at worst on first RC. Any plans for MFC and following MFS? commit 02f534b57f84d6f4f97c337b05b383c8b3aaf18c timerfd: fix up a memory leak and missing locking commit 5eab523053db79b4bd4f926c7d7ac04444d9c1da timerfd: compute fflags before calling falloc commit f4296cfb409a48de00bfa60e76f686c2b031876f timerfd: convert timerfd_list_lock from sx to mtx commit a1f506156c4db885d3cc177c93e9c8a28d535d30 timerfd: Define a locking regime commit 918966a27479b4fb7c4c8999c4926d83c2c081e5 timerfd: Relocate 32-bit compat code commit fb5daae920bae84e3eec8175bf9e46304c3b2ae6 timerfd: Namespace pollution adjustments Note that first 3 commits are authord/committed by mjg@. These are all I could confirm landed with commit histories of sys/kern/sys_timerfd.c and sys/sys/timerfd.c, excluding already-in-stable/14 one. I've found a related commit to sys/compat/linux/linux_event.c but intentionally excluded it, as it's already MFC'ed and MFS'ed. And I found another request on dev-commits-src-main ML archive [1], without any reply. Attached is the hand-merged patch to cherry pick them to stable/14, created before last 3 commits landed on main. HTH. [1] https://lists.freebsd.org/archives/dev-commits-src-main/2023-September/018407.html Regards. -- Tomoaki AOKI <junchoon@dec.sakura.ne.jp>