Count number of threads that enter and leave dynamically registered

syscalls. On the dynamic syscall deregistration, wait until all
threads leave the syscall code. This somewhat increases the safety
of the loadable modules unloading.

Reviewed by:	jhb
Tested by:	pho
MFC after:	1 month

6 files changed, 101 insertions, 29 deletions

diff --git a/sys/compat/freebsd32/syscalls.master b/sys/compat/freebsd32/syscalls.master
index 1afe586..4f1fc28 100644
--- a/sys/compat/freebsd32/syscalls.master
+++ b/sys/compat/freebsd32/syscalls.master
@@ -386,16 +386,16 @@
 ;
 ; The following are reserved for loadable syscalls
 ;
-210	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-211	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-212	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-213	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-214	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-215	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-216	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-217	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-218	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-219	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
+210	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+211	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+212	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+213	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+214	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+215	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+216	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+217	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+218	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+219	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
 
 ;
 ; The following were introduced with NetBSD/4.4Lite-2
diff --git a/sys/kern/kern_syscalls.c b/sys/kern/kern_syscalls.c
index 9cb7b68..d1a5c0d 100644
--- a/sys/kern/kern_syscalls.c
+++ b/sys/kern/kern_syscalls.c
@@ -28,12 +28,15 @@
 __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
+#include <sys/kernel.h>
 #include <sys/lock.h>
 #include <sys/module.h>
 #include <sys/sx.h>
 #include <sys/syscall.h>
 #include <sys/sysent.h>
 #include <sys/sysproto.h>
+#include <sys/systm.h>
+#include <machine/atomic.h>
 
 /*
  * Acts like "nosys" but can be identified in sysent for dynamic call
@@ -55,6 +58,51 @@ lkmressys(struct thread *td, struct nosys_args *args)
 	return (nosys(td, args));
 }
 
+static void
+syscall_thread_drain(struct sysent *se)
+{
+	u_int32_t cnt, oldcnt;
+
+	do {
+		oldcnt = se->sy_thrcnt;
+		KASSERT((oldcnt & SY_THR_STATIC) == 0,
+		    ("drain on static syscall"));
+		cnt = oldcnt | SY_THR_DRAINING;
+	} while (atomic_cmpset_acq_32(&se->sy_thrcnt, oldcnt, cnt) == 0);
+	while (atomic_cmpset_32(&se->sy_thrcnt, SY_THR_DRAINING,
+	    SY_THR_ABSENT) == 0)
+		pause("scdrn", hz/2);
+}
+
+int
+syscall_thread_enter(struct thread *td, struct sysent *se)
+{
+	u_int32_t cnt, oldcnt;
+
+	do {
+		oldcnt = se->sy_thrcnt;
+		if ((oldcnt & SY_THR_STATIC) != 0)
+			return (0);
+		if ((oldcnt & (SY_THR_DRAINING | SY_THR_ABSENT)) != 0)
+			return (ENOSYS);
+		cnt = oldcnt + SY_THR_INCR;
+	} while (atomic_cmpset_acq_32(&se->sy_thrcnt, oldcnt, cnt) == 0);
+	return (0);
+}
+
+void
+syscall_thread_exit(struct thread *td, struct sysent *se)
+{
+	u_int32_t cnt, oldcnt;
+
+	do {
+		oldcnt = se->sy_thrcnt;
+		if ((oldcnt & SY_THR_STATIC) != 0)
+			return;
+		cnt = oldcnt - SY_THR_INCR;
+	} while (atomic_cmpset_rel_32(&se->sy_thrcnt, oldcnt, cnt) == 0);
+}
+
 int
 syscall_register(int *offset, struct sysent *new_sysent,
     struct sysent *old_sysent)
@@ -74,8 +122,12 @@ syscall_register(int *offset, struct sysent *new_sysent,
 	    sysent[*offset].sy_call != (sy_call_t *)lkmressys)
 		return (EEXIST);
 
+	KASSERT(sysent[*offset].sy_thrcnt == SY_THR_ABSENT,
+	    ("dynamic syscall is not protected"));
 	*old_sysent = sysent[*offset];
+	new_sysent->sy_thrcnt = SY_THR_ABSENT;
 	sysent[*offset] = *new_sysent;
+	atomic_store_rel_32(&sysent[*offset].sy_thrcnt, 0);
 	return (0);
 }
 
@@ -83,8 +135,10 @@ int
 syscall_deregister(int *offset, struct sysent *old_sysent)
 {
 
-	if (*offset)
+	if (*offset) {
+		syscall_thread_drain(&sysent[*offset]);
 		sysent[*offset] = *old_sysent;
+	}
 	return (0);
 }
 
diff --git a/sys/kern/makesyscalls.sh b/sys/kern/makesyscalls.sh
index 4e55f96..46e04fc 100644
--- a/sys/kern/makesyscalls.sh
+++ b/sys/kern/makesyscalls.sh
@@ -253,6 +253,10 @@ s/\$//g
 		f=4			# toss number, type, audit event
 		argc= 0;
 		argssize = "0"
+		thr_flag = "SY_THR_STATIC"
+		if (flag("NOTSTATIC")) {
+			thr_flag = "SY_THR_ABSENT"
+		}
 		if ($NF != "}") {
 			funcalias=$(NF-2)
 			argalias=$(NF-1)
@@ -401,10 +405,10 @@ s/\$//g
 		printf("\t{ %s, (sy_call_t *)", argssize) > sysent
 		column = 8 + 2 + length(argssize) + 15
 		if (flag("NOSTD")) {
-			printf("%s },", "lkmressys, AUE_NULL, NULL, 0, 0, 0") > sysent
+			printf("%s },", "lkmressys, AUE_NULL, NULL, 0, 0, 0, SY_THR_ABSENT") > sysent
 			column = column + length("lkmressys") + length("AUE_NULL") + 3
 		} else {
-			printf("%s, %s, NULL, 0, 0, %s },", funcname, auditev, flags) > sysent
+			printf("%s, %s, NULL, 0, 0, %s, %s },", funcname, auditev, flags, thr_flag) > sysent
 			column = column + length(funcname) + length(auditev) + length(flags) + 3
 		} 
 		align_sysent_comment(column)
@@ -472,13 +476,13 @@ s/\$//g
 			    prefix, funcname, auditev) > sysaue
 		}
 		if (flag("NOSTD")) {
-			printf("\t{ %s, (sy_call_t *)%s, %s, NULL, 0, 0, 0 },",
+			printf("\t{ %s, (sy_call_t *)%s, %s, NULL, 0, 0, 0, SY_THR_ABSENT },",
 			    "0", "lkmressys", "AUE_NULL") > sysent
 			align_sysent_comment(8 + 2 + length("0") + 15 + \
 			    length("lkmressys") + length("AUE_NULL") + 3)
 		} else {
-			printf("\t{ %s(%s,%s), %s, NULL, 0, 0, %s },",
-			    wrap, argssize, funcname, auditev, flags) > sysent
+			printf("\t{ %s(%s,%s), %s, NULL, 0, 0, %s, %s },",
+			    wrap, argssize, funcname, auditev, flags, thr_flag) > sysent
 			align_sysent_comment(8 + 9 + length(argssize) + 1 + \
 			    length(funcname) + length(auditev) + \
 			    length(flags) + 4)
@@ -498,7 +502,7 @@ s/\$//g
 		next
 	}
 	type("OBSOL") {
-		printf("\t{ 0, (sy_call_t *)nosys, AUE_NULL, NULL, 0, 0, 0 },") > sysent
+		printf("\t{ 0, (sy_call_t *)nosys, AUE_NULL, NULL, 0, 0, 0, SY_THR_ABSENT },") > sysent
 		align_sysent_comment(34)
 		printf("/* %d = obsolete %s */\n", syscall, comment) > sysent
 		printf("\t\"obs_%s\",\t\t\t/* %d = obsolete %s */\n",
@@ -509,7 +513,7 @@ s/\$//g
 		next
 	}
 	type("UNIMPL") {
-		printf("\t{ 0, (sy_call_t *)nosys, AUE_NULL, NULL, 0, 0, 0 },\t\t\t/* %d = %s */\n",
+		printf("\t{ 0, (sy_call_t *)nosys, AUE_NULL, NULL, 0, 0, 0, SY_THR_ABSENT },\t\t\t/* %d = %s */\n",
 		    syscall, comment) > sysent
 		printf("\t\"#%d\",\t\t\t/* %d = %s */\n",
 		    syscall, syscall, comment) > sysnames
diff --git a/sys/kern/subr_trap.c b/sys/kern/subr_trap.c
index 19bd5d4..4298429 100644
--- a/sys/kern/subr_trap.c
+++ b/sys/kern/subr_trap.c
@@ -298,6 +298,9 @@ syscallenter(struct thread *td, struct syscall_args *sa)
 			if (error != 0)
 				goto retval;
 		}
+		error = syscall_thread_enter(td, sa->callp);
+		if (error != 0)
+			goto retval;
 
 #ifdef KDTRACE_HOOKS
 		/*
@@ -327,6 +330,7 @@ syscallenter(struct thread *td, struct syscall_args *sa)
 			(*systrace_probe_func)(sa->callp->sy_return, sa->code,
 			    sa->callp, sa->args);
 #endif
+		syscall_thread_exit(td, sa->callp);
 		CTR4(KTR_SYSC, "syscall: p=%p error=%d return %#lx %#lx",
 		    p, error, td->td_retval[0], td->td_retval[1]);
 	}
diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master
index 6a0ea65..8e11134 100644
--- a/sys/kern/syscalls.master
+++ b/sys/kern/syscalls.master
@@ -40,7 +40,7 @@
 ;	NOPROTO	same as STD except do not create structure or
 ;		function prototype in sys/sysproto.h.  Does add a
 ;		definition to syscall.h besides adding a sysent.
-
+;	NONSTATIC syscall is loadable
 ;
 ; Please copy any additions and changes to the following compatability tables:
 ; sys/compat/freebsd32/syscalls.master
@@ -403,16 +403,16 @@
 ;
 ; The following are reserved for loadable syscalls
 ;
-210	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-211	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-212	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-213	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-214	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-215	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-216	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-217	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-218	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
-219	AUE_NULL	NODEF	lkmnosys lkmnosys nosys_args int
+210	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+211	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+212	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+213	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+214	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+215	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+216	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+217	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+218	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
+219	AUE_NULL	NODEF|NOTSTATIC	lkmnosys lkmnosys nosys_args int
 
 ;
 ; The following were introduced with NetBSD/4.4Lite-2
diff --git a/sys/sys/sysent.h b/sys/sys/sysent.h
index 99e44a0..641bdae 100644
--- a/sys/sys/sysent.h
+++ b/sys/sys/sysent.h
@@ -61,8 +61,15 @@ struct sysent {			/* system call table */
 	u_int32_t sy_entry;	/* DTrace entry ID for systrace. */
 	u_int32_t sy_return;	/* DTrace return ID for systrace. */
 	u_int32_t sy_flags;	/* General flags for system calls. */
+	u_int32_t sy_thrcnt;
 };
 
+#define	SY_THR_FLAGMASK	0x7
+#define	SY_THR_STATIC	0x1
+#define	SY_THR_DRAINING	0x2
+#define	SY_THR_ABSENT	0x4
+#define	SY_THR_INCR	0x8
+
 struct image_params;
 struct __sigset;
 struct syscall_args;
@@ -211,6 +218,9 @@ struct nosys_args;
 int	lkmnosys(struct thread *, struct nosys_args *);
 int	lkmressys(struct thread *, struct nosys_args *);
 
+int	syscall_thread_enter(struct thread *td, struct sysent *se);
+void	syscall_thread_exit(struct thread *td, struct sysent *se);
+
 #endif /* _KERNEL */
 
 #endif /* !_SYS_SYSENT_H_ */