Push kernel lock deep down to sys_sysctl(). At least network subset of

sysctl(8) MIBs relies on netlock or another locks and doesn't require
kernel lock, so unlock it. The protocols layer *_sysctl()s are left
under kernel lock and will be sequentially unlocked later.

ok bluhm@

diff --git a/sys/kern/kern_sysctl.c b/sys/kern/kern_sysctl.c
index 226518b..1c5adf4 100644 (file)
--- a/sys/kern/kern_sysctl.c
+++ b/sys/kern/kern_sysctl.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: kern_sysctl.c,v 1.411 2023/01/22 12:05:44 mvs Exp $   */
+/*     $OpenBSD: kern_sysctl.c,v 1.412 2023/05/04 09:40:36 mvs Exp $   */
 /*     $NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $     */
 
 /*-
@@ -168,7 +168,7 @@ sys_sysctl(struct proc *p, void *v, register_t *retval)
                syscallarg(void *) new;
                syscallarg(size_t) newlen;
        } */ *uap = v;
-       int error, dolock = 1;
+       int error, dokernellock = 1, dolock = 1;
        size_t savelen = 0, oldlen = 0;
        sysctlfn *fn;
        int name[CTL_MAXNAME];
@@ -203,6 +203,7 @@ sys_sysctl(struct proc *p, void *v, register_t *retval)
                break;
        case CTL_NET:
                fn = net_sysctl;
+               dokernellock = 0;
                break;
        case CTL_FS:
                fn = fs_sysctl;
@@ -230,19 +231,22 @@ sys_sysctl(struct proc *p, void *v, register_t *retval)
        if (SCARG(uap, oldlenp) &&
            (error = copyin(SCARG(uap, oldlenp), &oldlen, sizeof(oldlen))))
                return (error);
+       if (dokernellock)
+               KERNEL_LOCK();
        if (SCARG(uap, old) != NULL) {
                if ((error = rw_enter(&sysctl_lock, RW_WRITE|RW_INTR)) != 0)
-                       return (error);
+                       goto unlock;
                if (dolock) {
                        if (atop(oldlen) > uvmexp.wiredmax - uvmexp.wired) {
                                rw_exit_write(&sysctl_lock);
-                               return (ENOMEM);
+                               error = ENOMEM;
+                               goto unlock;
                        }
                        error = uvm_vslock(p, SCARG(uap, old), oldlen,
                            PROT_READ | PROT_WRITE);
                        if (error) {
                                rw_exit_write(&sysctl_lock);
-                               return (error);
+                               goto unlock;
                        }
                }
                savelen = oldlen;
@@ -254,6 +258,9 @@ sys_sysctl(struct proc *p, void *v, register_t *retval)
                        uvm_vsunlock(p, SCARG(uap, old), savelen);
                rw_exit_write(&sysctl_lock);
        }
+unlock:
+       if (dokernellock)
+               KERNEL_UNLOCK();
        if (error)
                return (error);
        if (SCARG(uap, oldlenp))

diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master
index 7f43762..3faf5dd 100644 (file)
--- a/sys/kern/syscalls.master
+++ b/sys/kern/syscalls.master
@@ -1,4 +1,4 @@
-;      $OpenBSD: syscalls.master,v 1.246 2023/02/25 09:55:46 mvs Exp $
+;      $OpenBSD: syscalls.master,v 1.247 2023/05/04 09:40:36 mvs Exp $
 ;      $NetBSD: syscalls.master,v 1.32 1996/04/23 10:24:21 mycroft Exp $
 
 ;      @(#)syscalls.master     8.2 (Berkeley) 1/13/94
@@ -361,7 +361,7 @@
 199    OBSOL           pad_lseek
 200    OBSOL           pad_truncate
 201    OBSOL           pad_ftruncate
-202    STD             { int sys_sysctl(const int *name, u_int namelen, \
+202    STD NOLOCK      { int sys_sysctl(const int *name, u_int namelen, \
                            void *old, size_t *oldlenp, void *new, \
                            size_t newlen); }
 203    STD             { int sys_mlock(const void *addr, size_t len); }

diff --git a/sys/kern/uipc_domain.c b/sys/kern/uipc_domain.c
index 1534c0d..3b5c0bb 100644 (file)
--- a/sys/kern/uipc_domain.c
+++ b/sys/kern/uipc_domain.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: uipc_domain.c,v 1.60 2022/08/14 01:58:28 jsg Exp $    */
+/*     $OpenBSD: uipc_domain.c,v 1.61 2023/05/04 09:40:36 mvs Exp $    */
 /*     $NetBSD: uipc_domain.c,v 1.14 1996/02/09 19:00:44 christos Exp $        */
 
 /*
@@ -213,9 +213,13 @@ net_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
                    newp, newlen));
 #endif
 #if NPFLOW > 0
-       if (family == PF_PFLOW)
-               return (pflow_sysctl(name + 1, namelen - 1, oldp, oldlenp,
-                   newp, newlen));
+       if (family == PF_PFLOW) {
+               KERNEL_LOCK();
+               error = pflow_sysctl(name + 1, namelen - 1, oldp, oldlenp,
+                   newp, newlen);
+               KERNEL_UNLOCK();
+               return (error);
+       }
 #endif
 #ifdef PIPEX
        if (family == PF_PIPEX)
@@ -223,9 +227,13 @@ net_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
                    newp, newlen));
 #endif
 #ifdef MPLS
-       if (family == PF_MPLS)
-               return (mpls_sysctl(name + 1, namelen - 1, oldp, oldlenp,
-                   newp, newlen));
+       if (family == PF_MPLS) {
+               KERNEL_LOCK();
+               error = mpls_sysctl(name + 1, namelen - 1, oldp, oldlenp,
+                   newp, newlen);
+               KERNEL_UNLOCK();
+               return (error);
+       }
 #endif
        dp = pffinddomain(family);
        if (dp == NULL)
@@ -236,8 +244,10 @@ net_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
        protocol = name[1];
        for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
                if (pr->pr_protocol == protocol && pr->pr_sysctl) {
+                       KERNEL_LOCK();
                        error = (*pr->pr_sysctl)(name + 2, namelen - 2,
                            oldp, oldlenp, newp, newlen);
+                       KERNEL_UNLOCK();
                        return (error);
                }
        return (ENOPROTOOPT);