push the KERNEL_LOCK deeper on read(2) and write(2)

unlocks read(2) and write(2) syscalls families, and push the KERNEL_LOCK
deeper in the code path. KERNEL_LOCK is managed per file type in fileops
handlers (fo_read, fo_write, and fo_close). read(2) and write(2) on
socket are KERNEL_LOCK-free.

initial work from mpi@ and ians@

ok mpi@ kettenis@ visa@ ians@

5 files changed, 52 insertions, 21 deletions

diff --git a/sys/kern/sys_generic.c b/sys/kern/sys_generic.c
index 5fb4cf04258..f72ca9fd3cb 100644
--- a/sys/kern/sys_generic.c
+++ b/sys/kern/sys_generic.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sys_generic.c,v 1.124 2019/06/21 09:39:48 visa Exp $	*/
+/*	$OpenBSD: sys_generic.c,v 1.125 2019/06/22 06:48:25 semarie Exp $	*/
 /*	$NetBSD: sys_generic.c,v 1.24 1996/03/29 00:25:32 cgd Exp $	*/
 
 /*
@@ -366,8 +366,11 @@ dofilewritev(struct proc *p, int fd, struct uio *uio, int flags,
 		if (uio->uio_resid != cnt && (error == ERESTART ||
 		    error == EINTR || error == EWOULDBLOCK))
 			error = 0;
-		if (error == EPIPE)
+		if (error == EPIPE) {
+			KERNEL_LOCK();
 			ptsignal(p, SIGPIPE, STHREAD);
+			KERNEL_UNLOCK();
+		}
 	}
 	cnt -= uio->uio_resid;
 
diff --git a/sys/kern/sys_pipe.c b/sys/kern/sys_pipe.c
index 71522053130..2d2c2210e5e 100644
--- a/sys/kern/sys_pipe.c
+++ b/sys/kern/sys_pipe.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sys_pipe.c,v 1.87 2018/11/13 13:02:20 visa Exp $	*/
+/*	$OpenBSD: sys_pipe.c,v 1.88 2019/06/22 06:48:25 semarie Exp $	*/
 
 /*
  * Copyright (c) 1996 John S. Dyson
@@ -314,9 +314,11 @@ pipe_read(struct file *fp, struct uio *uio, int fflags)
 	int error;
 	size_t size, nread = 0;
 
+	KERNEL_LOCK();
+
 	error = pipelock(rpipe);
 	if (error)
-		return (error);
+		goto done;
 
 	++rpipe->pipe_busy;
 
@@ -420,6 +422,8 @@ unlocked_error:
 	if ((rpipe->pipe_buffer.size - rpipe->pipe_buffer.cnt) >= PIPE_BUF)
 		pipeselwakeup(rpipe);
 
+done:
+	KERNEL_UNLOCK();
 	return (error);
 }
 
@@ -430,6 +434,8 @@ pipe_write(struct file *fp, struct uio *uio, int fflags)
 	size_t orig_resid;
 	struct pipe *wpipe, *rpipe;
 
+	KERNEL_LOCK();
+
 	rpipe = fp->f_data;
 	wpipe = rpipe->pipe_peer;
 
@@ -437,7 +443,8 @@ pipe_write(struct file *fp, struct uio *uio, int fflags)
 	 * detect loss of pipe read side, issue SIGPIPE if lost.
 	 */
 	if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
-		return (EPIPE);
+		error = EPIPE;
+		goto done;
 	}
 	++wpipe->pipe_busy;
 
@@ -471,7 +478,7 @@ pipe_write(struct file *fp, struct uio *uio, int fflags)
 			wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 			wakeup(wpipe);
 		}
-		return (error);
+		goto done;
 	}
 
 	orig_resid = uio->uio_resid;
@@ -642,6 +649,8 @@ retrywrite:
 	if (wpipe->pipe_buffer.cnt)
 		pipeselwakeup(wpipe);
 
+done:
+	KERNEL_UNLOCK();
 	return (error);
 }
 
diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master
index 8c6ea62e801..d7439af23cf 100644
--- a/sys/kern/syscalls.master
+++ b/sys/kern/syscalls.master
@@ -1,4 +1,4 @@
-;	$OpenBSD: syscalls.master,v 1.191 2019/05/13 22:55:27 beck Exp $
+;	$OpenBSD: syscalls.master,v 1.192 2019/06/22 06:49:14 semarie Exp $
 ;	$NetBSD: syscalls.master,v 1.32 1996/04/23 10:24:21 mycroft Exp $
 
 ;	@(#)syscalls.master	8.2 (Berkeley) 1/13/94
@@ -51,8 +51,8 @@
 0	INDIR		{ int sys_syscall(int number, ...); }
 1	STD		{ void sys_exit(int rval); }
 2	STD		{ int sys_fork(void); }
-3	STD		{ ssize_t sys_read(int fd, void *buf, size_t nbyte); }
-4	STD		{ ssize_t sys_write(int fd, const void *buf, \
+3	STD NOLOCK	{ ssize_t sys_read(int fd, void *buf, size_t nbyte); }
+4	STD NOLOCK	{ ssize_t sys_write(int fd, const void *buf, \
 			    size_t nbyte); }
 5	STD		{ int sys_open(const char *path, \
 			    int flags, ... mode_t mode); }
@@ -247,9 +247,9 @@
 118	STD		{ int sys_getsockopt(int s, int level, int name, \
 			    void *val, socklen_t *avalsize); }
 119	STD		{ int sys_thrkill(pid_t tid, int signum, void *tcb); }
-120	STD		{ ssize_t sys_readv(int fd, \
+120	STD NOLOCK	{ ssize_t sys_readv(int fd, \
 			    const struct iovec *iovp, int iovcnt); }
-121	STD		{ ssize_t sys_writev(int fd, \
+121	STD NOLOCK	{ ssize_t sys_writev(int fd, \
 			    const struct iovec *iovp, int iovcnt); }
 122	STD		{ int sys_kill(int pid, int signum); }
 123	STD		{ int sys_fchown(int fd, uid_t uid, gid_t gid); }
@@ -316,9 +316,9 @@
 170	OBSOL		msgsys10
 171	OBSOL		shmsys10
 172	UNIMPL
-173	STD		{ ssize_t sys_pread(int fd, void *buf, \
+173	STD NOLOCK	{ ssize_t sys_pread(int fd, void *buf, \
 			  size_t nbyte, int pad, off_t offset); }
-174	STD		{ ssize_t sys_pwrite(int fd, const void *buf, \
+174	STD NOLOCK	{ ssize_t sys_pwrite(int fd, const void *buf, \
 			  size_t nbyte, int pad, off_t offset); }
 175	UNIMPL		ntp_gettime
 176	UNIMPL		ntp_adjtime
@@ -452,10 +452,10 @@
 264	STD		{ int sys_fhopen(const fhandle_t *fhp, int flags); }
 265	UNIMPL
 266	UNIMPL
-267	STD		{ ssize_t sys_preadv(int fd, \
+267	STD NOLOCK	{ ssize_t sys_preadv(int fd, \
 			  const struct iovec *iovp, int iovcnt, \
 			  int pad, off_t offset); }
-268	STD		{ ssize_t sys_pwritev(int fd, \
+268	STD NOLOCK	{ ssize_t sys_pwritev(int fd, \
 			  const struct iovec *iovp, int iovcnt, \
 			  int pad, off_t offset); }
 269	STD		{ int sys_kqueue(void); }
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index c02f8f0cba1..520126602c1 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_vnops.c,v 1.98 2019/06/21 09:39:48 visa Exp $	*/
+/*	$OpenBSD: vfs_vnops.c,v 1.99 2019/06/22 06:48:25 semarie Exp $	*/
 /*	$NetBSD: vfs_vnops.c,v 1.20 1996/02/04 02:18:41 christos Exp $	*/
 
 /*
@@ -343,6 +343,8 @@ vn_read(struct file *fp, struct uio *uio, int fflags)
 	off_t offset;
 	int error;
 
+	KERNEL_LOCK();
+
 	/*
 	 * Check below can race.  We can block on the vnode lock
 	 * and resume with a different `fp->f_offset' value.
@@ -353,11 +355,15 @@ vn_read(struct file *fp, struct uio *uio, int fflags)
 		offset = uio->uio_offset;
 
 	/* no wrap around of offsets except on character devices */
-	if (vp->v_type != VCHR && count > LLONG_MAX - offset)
-		return (EINVAL);
+	if (vp->v_type != VCHR && count > LLONG_MAX - offset) {
+		error = EINVAL;
+		goto done;
+	}
 
-	if (vp->v_type == VDIR)
-		return (EISDIR);
+	if (vp->v_type == VDIR) {
+		error = EISDIR;
+		goto done;
+	}
 
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 	if ((fflags & FO_POSITION) == 0)
@@ -367,6 +373,8 @@ vn_read(struct file *fp, struct uio *uio, int fflags)
 	if ((fflags & FO_POSITION) == 0)
 		fp->f_offset += count - uio->uio_resid;
 	VOP_UNLOCK(vp);
+done:
+	KERNEL_UNLOCK();
 	return (error);
 }
 
@@ -381,6 +389,8 @@ vn_write(struct file *fp, struct uio *uio, int fflags)
 	int error, ioflag = IO_UNIT;
 	size_t count;
 
+	KERNEL_LOCK();
+
 	/* note: pwrite/pwritev are unaffected by O_APPEND */
 	if (vp->v_type == VREG && (fp->f_flag & O_APPEND) &&
 	    (fflags & FO_POSITION) == 0)
@@ -402,6 +412,8 @@ vn_write(struct file *fp, struct uio *uio, int fflags)
 			fp->f_offset += count - uio->uio_resid;
 	}
 	VOP_UNLOCK(vp);
+
+	KERNEL_UNLOCK();
 	return (error);
 }
 
diff --git a/sys/sys/file.h b/sys/sys/file.h
index b2a1f95e898..0054c6df76b 100644
--- a/sys/sys/file.h
+++ b/sys/sys/file.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: file.h,v 1.53 2018/08/20 16:00:22 mpi Exp $	*/
+/*	$OpenBSD: file.h,v 1.54 2019/06/22 06:48:25 semarie Exp $	*/
 /*	$NetBSD: file.h,v 1.11 1995/03/26 20:24:13 jtc Exp $	*/
 
 /*
@@ -46,6 +46,13 @@ struct stat;
 struct file;
 struct ucred;
 
+/**
+ * File operations.
+ * The following entries could be called without KERNEL_LOCK hold:
+ * - fo_read
+ * - fo_write
+ * - fo_close
+ */
 struct	fileops {
 	int	(*fo_read)(struct file *, struct uio *, int);
 	int	(*fo_write)(struct file *, struct uio *, int);