On 06/07/17(Thu) 21:18, Juan Francisco Cantero Hurtado wrote:
> On Thu, Jul 06, 2017 at 08:29:39AM -0600, Aaron Bieber wrote:
> > Hola,
> >
> > For roughly a week I have been able to reliably produce panics on
> > amd64 (virtual machine and physical machine) using disk IO heavy
> > applications (gitea, syncthing).
> >
> > Unfortunately on the physical machine - the one I can reliably panic
> > within < 1 minute - I don't have a serial console, so here are links
> > to photos:
> >
> > https://goo.gl/photos/Z8g225UXXpCXQpGcA
> > mirror/google averse: https://deftly.net/panic-2017-07-06/
> >
> > Also with machdep.forceukbd=1, once I ran `mach ddbcpu 0`, every
> > subsequent key press puked out an "splassert" line, I attempted to run
> > a "trace" again, but things seemed to be completely locked up after that.
>
> I'm seeing the same panic but I can't run ddb. Also with syncthing.
Here's a fix. The problem certainly comes from a contented NET_LOCK()
which means that we now have a sleeping point where we're not allowed
to sleep.
Diff below revert the solock()/sounlock() dances in the kqueue filters,
let me know if it works for you.
Index: kern/uipc_socket.c
===================================================================
RCS file: /cvs/src/sys/kern/uipc_socket.c,v
retrieving revision 1.192
diff -u -p -r1.192 uipc_socket.c
--- kern/uipc_socket.c 4 Jul 2017 12:58:32 -0000 1.192
+++ kern/uipc_socket.c 7 Jul 2017 10:46:40 -0000
@@ -1998,10 +1998,8 @@ int
filt_sowrite(struct knote *kn, long hint)
{
struct socket *so = kn->kn_fp->f_data;
- int s, rv;
+ int rv;
- if (!(hint & NOTE_SUBMIT))
- s = solock(so);
kn->kn_data = sbspace(so, &so->so_snd);
if (so->so_state & SS_CANTSENDMORE) {
kn->kn_flags |= EV_EOF;
@@ -2017,8 +2015,6 @@ filt_sowrite(struct knote *kn, long hint
} else {
rv = (kn->kn_data >= so->so_snd.sb_lowat);
}
- if (!(hint & NOTE_SUBMIT))
- sounlock(s);
return (rv);
}
@@ -2027,13 +2023,8 @@ int
filt_solisten(struct knote *kn, long hint)
{
struct socket *so = kn->kn_fp->f_data;
- int s;
- if (!(hint & NOTE_SUBMIT))
- s = solock(so);
kn->kn_data = so->so_qlen;
- if (!(hint & NOTE_SUBMIT))
- sounlock(s);
return (kn->kn_data != 0);
}
Index: miscfs/fifofs/fifo_vnops.c
===================================================================
RCS file: /cvs/src/sys/miscfs/fifofs/fifo_vnops.c,v
retrieving revision 1.56
diff -u -p -r1.56 fifo_vnops.c
--- miscfs/fifofs/fifo_vnops.c 3 Jul 2017 08:31:39 -0000 1.56
+++ miscfs/fifofs/fifo_vnops.c 7 Jul 2017 10:46:57 -0000
@@ -531,10 +531,8 @@ int
filt_fiforead(struct knote *kn, long hint)
{
struct socket *so = (struct socket *)kn->kn_hook;
- int s, rv;
+ int rv;
- if (!(hint & NOTE_SUBMIT))
- s = solock(so);
kn->kn_data = so->so_rcv.sb_cc;
if (so->so_state & SS_CANTRCVMORE) {
kn->kn_flags |= EV_EOF;
@@ -543,8 +541,6 @@ filt_fiforead(struct knote *kn, long hin
kn->kn_flags &= ~EV_EOF;
rv = (kn->kn_data > 0);
}
- if (!(hint & NOTE_SUBMIT))
- sounlock(s);
return (rv);
}
@@ -563,10 +559,8 @@ int
filt_fifowrite(struct knote *kn, long hint)
{
struct socket *so = (struct socket *)kn->kn_hook;
- int s, rv;
+ int rv;
- if (!(hint & NOTE_SUBMIT))
- s = solock(so);
kn->kn_data = sbspace(so, &so->so_snd);
if (so->so_state & SS_CANTSENDMORE) {
kn->kn_flags |= EV_EOF;
@@ -575,8 +569,6 @@ filt_fifowrite(struct knote *kn, long hi
kn->kn_flags &= ~EV_EOF;
rv = (kn->kn_data >= so->so_snd.sb_lowat);
}
- if (!(hint & NOTE_SUBMIT))
- sounlock(s);
return (rv);
}
Index: sys/socketvar.h
===================================================================
RCS file: /cvs/src/sys/sys/socketvar.h,v
retrieving revision 1.72
diff -u -p -r1.72 socketvar.h
--- sys/socketvar.h 4 Jul 2017 12:58:32 -0000 1.72
+++ sys/socketvar.h 7 Jul 2017 10:46:32 -0000
@@ -186,7 +186,10 @@ static inline long
sbspace(struct socket *so, struct sockbuf *sb)
{
KASSERT(sb == &so->so_rcv || sb == &so->so_snd);
+#if 0
+ /* XXXSMP kqueue_scan() calling filt_sowrite() cannot sleep. */
soassertlocked(so);
+#endif
return lmin(sb->sb_hiwat - sb->sb_cc, sb->sb_mbmax - sb->sb_mbcnt);
}
