BTW, this is another occurance of the same failure condition: You did not build a full snapshot!
This diff is old enough there was lots of time to build a snapshot with it. Vitaliy Makkoveev <o...@bsdbox.dev> wrote: > Sorry about that. Please give me 5 min to fix it. > > > On 11 Aug 2024, at 23:46, Theo Buehler <t...@theobuehler.org> wrote: > > > > On Sun, Aug 11, 2024 at 09:10:53AM -0600, Vitaliy Makkoveev wrote: > >> CVSROOT: /cvs > >> Module name: src > >> Changes by: m...@cvs.openbsd.org 2024/08/11 09:10:53 > >> > >> Modified files: > >> sys/kern : kern_exit.c kern_fork.c kern_sysctl.c > >> sys/sys : proc.h > >> > >> Log message: > >> Make exit1() wait sysctl(2) `allprocess' loops. > >> > >> Regardless on wired userland memory, KERN_FILE_BYPID and KERN_FILE_BYUID > >> `allprocess' loops have netlock provided sleep points, so concurrent > >> process exit(1) could crash kernel. > >> > >> The main exit1() problem is that process teardown begins while process > >> is still linked to `allprocess' list, and current code doesn't allow to > >> unlink it first. Wait for concurrent sysctl(2) `allprocess' loops > >> between PS_EXITING bit setting and list unlinking. Both KERN_FILE_BYPID > >> and KERN_FILE_BYUID loops do PS_EXITING check and won't deal with dying > >> process. Concurrent exit1() thread will wait loops keeping process > >> linked to `allprocess' list. > >> > >> Tested with i386 dpb(1) run. > >> Stress tests and ok bluhm. > >> > > > > libkvm fails to build with this diff because userland doesn't know about > > struct refcount: > > > > cc -O2 -pipe -g -D_LIBKVM -MD -MP -c /usr/src/lib/libkvm/kvm_proc.c -o > > kvm_proc.o > > In file included from /usr/src/lib/libkvm/kvm_proc.c:76: > > /usr/include/sys/proc.h:145:16: error: field has incomplete type 'struct > > refcnt' > > struct refcnt ps_refcnt; > > ^ >
