On Sun, Feb 11, 2024 at 09:16:41PM +0300, Vitaliy Makkoveev wrote:
> > > @@ -919,7 +929,7 @@ sbappendaddr(struct socket *so, struct s
> > >  	struct mbuf *m, *n, *nlast;
> > >  	int space = asa->sa_len;
> > >  
> > > -	soassertlocked(so);
> > > +	sbmtxassertlocked(so, sb);
> > 
> > What about the other sbappend...() functions?  If you replace
> > soassertlocked() also there, we see if you missed something.
> 
> Will replace it later, when corresponding sbappend*() will be protected
> with `sb_mtx'. This time this conversion is useless, but grows the diff.

My point is that you place the assertion in the functions where you
added the mutex protection.  This way assert will not find any bug.

The interesting thing is, whether the socket buffer fields are
modified elsewhere without protection.  sbmtxassertlocked() knows
which kind of socket needs the mutex.  So it can be placed in the
other sbappend*() functions now.  Then unexpecteded calls from
sockets with SB_MTXLOCK flag are reported.  If you know that
everything is correct, you don't need the assert at all.

Of course not in this diff, that is already waiting to be commited
for a week.

bluhm