Re: tail -f is broken in recent CURRENT ( afd5bc630930 Sunday Jul 6 )

On Mon, Jul 7, 2025 at 4:31 PM Konstantin Belousov <kostikbel@gmail.com> wrote:
>
> On Mon, Jul 07, 2025 at 04:14:22PM +0300, Oleg Nauman wrote:
> > On Mon, Jul 7, 2025 at 3:53 PM Konstantin Belousov <kostikbel@gmail.com> wrote:
> > >
> > > On Mon, Jul 07, 2025 at 03:36:12PM +0300, Oleg Nauman wrote:
> > > >  I noticed that
> > > > tail -f /var/log/messages never receive updates if running with recent
> > > > CURRENT ( afd5bc630930 Sunday Jul 6 )
> > > >
> > > > 60092: read(3,"talled\nJul  6 16:59:54 mysystem pkg-static[86368]:
> > > > gstreamer1-plugins-mpg123-1.26.2 deinstalled\nJul  6 16:59:57 mysytem
> > > > pkg"...,32768) = 2666 (0xa6a)
> > > > 60092: munmap(0x801403000,297578)                = 0 (0x0)
> > > > 60092: read(3,0x80182c600,32768)                 = 0 (0x0)
> > > > 60092: kqueue()                                  = 5 (0x5)
> > > > 60092: fstatfs(3,{
> > > > fstypename=ufs,mntonname=/,mntfromname=/dev/gpt/FreeBSD-UFS,fsid=133a996124e0f956
> > > > }) = 0 (0x0)
> > > > 60092: kevent(5,{ 3,EVFILT_READ,EV_ADD|EV_ENABLE|EV_CLEAR,0,0,0x0
> > > > },1,0x0,0,{ 0.000000000 }) = 0 (0x0)
> > > > 60092: kevent(5,0x0,0,0x801814000,1,0x0)         ERR#4 'Interrupted
> > > > system call' <--- killed at this point by me
> > > > 60095: select(7,{ 6 },0x0,0x0,0x0)               ERR#4 'Interrupted system call'
> > > > 60092: SIGNAL 2 (SIGINT) code=SI_KERNEL
> > > > 60095: SIGNAL 2 (SIGINT) code=SI_KERNEL
> > > > 60095: process killed, signal = 2
> > > > 60092: process killed, signal = 2
> > > >
> > > >  Likely something is wrong with kqueue and/or kevent
> > >
> > > Try this
> > >
> > > diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h
> > > index 3ed469bdce6d..2c6947103c94 100644
> > > --- a/sys/sys/vnode.h
> > > +++ b/sys/sys/vnode.h
> > > @@ -1032,7 +1032,7 @@ void      vop_rename_fail(struct vop_rename_args *ap);
> > >  #define VOP_WRITE_POST(ap, ret)                                                \
> > >         noffset = (ap)->a_uio->uio_offset;                              \
> > >         if (noffset > ooffset) {                                        \
> > > -               if (VN_KNLIST_EMPTY((ap)->a_vp)) {                      \
> > > +               if (!VN_KNLIST_EMPTY((ap)->a_vp)) {                     \
> > >                         VFS_KNOTE_LOCKED((ap)->a_vp, NOTE_WRITE |       \
> > >                             (noffset > osize ? NOTE_EXTEND : 0));       \
> > >                 }                                                       \
> > >
> >
> >  Buildworld is running,  please give me two hours or so
>
> World is not relevant, just rebuild the kernel.
>

 Well, I can confirm that patched kernel restore tail -f behavior

Thank you