Use a different mutex to protect the kqueue klist in logsoftc.

knote_locked() will call wakeup() and with it the SCHED_LOCK and by that
makes log_mtx no longer a leaf lock. By using an own lock for the klist
we can keep log_mtx a leaf lock and with that printf(9) can be used in
most contexts again.
OK mvs@

diff --git a/sys/kern/subr_log.c b/sys/kern/subr_log.c
index f1e504d..e64b85d 100644 (file)
--- a/sys/kern/subr_log.c
+++ b/sys/kern/subr_log.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: subr_log.c,v 1.78 2023/09/22 20:03:05 mvs Exp $       */
+/*     $OpenBSD: subr_log.c,v 1.79 2024/07/24 13:37:05 claudio Exp $   */
 /*     $NetBSD: subr_log.c,v 1.11 1996/03/30 22:24:44 christos Exp $   */
 
 /*
@@ -73,10 +73,11 @@
 /*
  * Locking:
  *     L       log_mtx
+ *     Q       log_kq_mtx
  */
 struct logsoftc {
        int     sc_state;               /* [L] see above for possibilities */
-       struct  klist sc_klist;         /* process waiting on kevent call */
+       struct  klist sc_klist;         /* [Q] process waiting on kevent call */
        struct  sigio_ref sc_sigio;     /* async I/O registration */
        int     sc_need_wakeup;         /* if set, wake up waiters */
        struct timeout sc_tick;         /* wakeup poll timeout */
@@ -97,6 +98,8 @@ struct        rwlock syslogf_rwlock = RWLOCK_INITIALIZER("syslogf");
  */
 struct mutex log_mtx =
     MUTEX_INITIALIZER_FLAGS(IPL_HIGH, "logmtx", MTX_NOWITNESS);
+struct mutex log_kq_mtx =
+    MUTEX_INITIALIZER_FLAGS(IPL_HIGH, "logkqmtx", MTX_NOWITNESS);
 
 void filt_logrdetach(struct knote *kn);
 int filt_logread(struct knote *kn, long hint);
@@ -208,7 +211,7 @@ logopen(dev_t dev, int flags, int mode, struct proc *p)
        if (log_open)
                return (EBUSY);
        log_open = 1;
-       klist_init_mutex(&logsoftc.sc_klist, &log_mtx);
+       klist_init_mutex(&logsoftc.sc_klist, &log_kq_mtx);
        sigio_init(&logsoftc.sc_sigio);
        timeout_set(&logsoftc.sc_tick, logtick, NULL);
        timeout_add_msec(&logsoftc.sc_tick, LOG_TICK);
@@ -336,7 +339,9 @@ filt_logread(struct knote *kn, long hint)
 {
        struct msgbuf *mbp = kn->kn_hook;
 
+       mtx_enter(&log_mtx);
        kn->kn_data = msgbuf_getlen(mbp);
+       mtx_leave(&log_mtx);
        return (kn->kn_data != 0);
 }
 
@@ -345,9 +350,9 @@ filt_logmodify(struct kevent *kev, struct knote *kn)
 {
        int active;
 
-       mtx_enter(&log_mtx);
+       mtx_enter(&log_kq_mtx);
        active = knote_modify(kev, kn);
-       mtx_leave(&log_mtx);
+       mtx_leave(&log_kq_mtx);
 
        return (active);
 }
@@ -357,9 +362,9 @@ filt_logprocess(struct knote *kn, struct kevent *kev)
 {
        int active;
 
-       mtx_enter(&log_mtx);
+       mtx_enter(&log_kq_mtx);
        active = knote_process(kn, kev);
-       mtx_leave(&log_mtx);
+       mtx_leave(&log_kq_mtx);
 
        return (active);
 }
@@ -404,9 +409,10 @@ logtick(void *arg)
        state = logsoftc.sc_state;
        if (logsoftc.sc_state & LOG_RDWAIT)
                logsoftc.sc_state &= ~LOG_RDWAIT;
-       knote_locked(&logsoftc.sc_klist, 0);
        mtx_leave(&log_mtx);
 
+       knote(&logsoftc.sc_klist, 0);
+
        if (state & LOG_ASYNC)
                pgsigio(&logsoftc.sc_sigio, SIGIO, 0);
        if (state & LOG_RDWAIT)