Linux 4.14 compat: vfs_read & vfs_write

The kernel_read & kernel_write functions have always wrapped the
vfs_read & vfs_write functions respectively.  However, they could
not be used by vn_rdwr() since the offset wasn't passed as a
pointer.  This prevented us from being able to properly update
the file offset.

Linux 4.14 unexported vfs_read & vfs_write but also changed the
signature of kernel_read & kernel_write to provide the needed
functionality.  Use these updated functions when available.

Reviewed-by: Pritam Baral <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #656 
Closes #667 

3 files changed, 105 insertions, 17 deletions

diff --git a/config/spl-build.m4 b/config/spl-build.m4
index 8e9dc99ff..7b66f2c86 100644
--- a/config/spl-build.m4
+++ b/config/spl-build.m4
@@ -52,6 +52,8 @@ AC_DEFUN([SPL_AC_CONFIG_KERNEL], [
 	SPL_AC_KMEM_CACHE_CREATE_USERCOPY
 	SPL_AC_WAIT_QUEUE_ENTRY_T
 	SPL_AC_WAIT_QUEUE_HEAD_ENTRY
+	SPL_AC_KERNEL_WRITE
+	SPL_AC_KERNEL_READ
 ])
 
 AC_DEFUN([SPL_AC_MODULE_SYMVERS], [
@@ -1594,3 +1596,61 @@ AC_DEFUN([SPL_AC_WAIT_QUEUE_HEAD_ENTRY], [
 		AC_MSG_RESULT(no)
 	])
 ])
+
+dnl #
+dnl # 4.14 API change
+dnl # kernel_write() which was introduced in 3.9 was updated to take
+dnl # the offset as a pointer which is needed by vn_rdwr().
+dnl #
+AC_DEFUN([SPL_AC_KERNEL_WRITE], [
+	AC_MSG_CHECKING([whether kernel_write() takes loff_t pointer])
+	tmp_flags="$EXTRA_KCFLAGS"
+	EXTRA_KCFLAGS="-Werror"
+	SPL_LINUX_TRY_COMPILE([
+		#include <linux/fs.h>
+	],[
+		struct file *file = NULL;
+		const void *buf = NULL;
+		size_t count = 0;
+		loff_t *pos = NULL;
+		ssize_t ret;
+
+		ret = kernel_write(file, buf, count, pos);
+	],[
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_KERNEL_WRITE_PPOS, 1,
+		    [kernel_write() take loff_t pointer])
+	],[
+		AC_MSG_RESULT(no)
+	])
+	EXTRA_KCFLAGS="$tmp_flags"
+])
+
+dnl #
+dnl # 4.14 API change
+dnl # kernel_read() which has existed for forever was updated to take
+dnl # the offset as a pointer which is needed by vn_rdwr().
+dnl #
+AC_DEFUN([SPL_AC_KERNEL_READ], [
+	AC_MSG_CHECKING([whether kernel_read() takes loff_t pointer])
+	tmp_flags="$EXTRA_KCFLAGS"
+	EXTRA_KCFLAGS="-Werror"
+	SPL_LINUX_TRY_COMPILE([
+		#include <linux/fs.h>
+	],[
+		struct file *file = NULL;
+		void *buf = NULL;
+		size_t count = 0;
+		loff_t *pos = NULL;
+		ssize_t ret;
+
+		ret = kernel_read(file, buf, count, pos);
+	],[
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_KERNEL_READ_PPOS, 1,
+		    [kernel_read() take loff_t pointer])
+	],[
+		AC_MSG_RESULT(no)
+	])
+	EXTRA_KCFLAGS="$tmp_flags"
+])
diff --git a/include/linux/file_compat.h b/include/linux/file_compat.h
index 7d61ba5f9..55ba2cc76 100644
--- a/include/linux/file_compat.h
+++ b/include/linux/file_compat.h
@@ -26,6 +26,7 @@
 #define	_SPL_FILE_COMPAT_H
 
 #include <linux/fs.h>
+#include <linux/uaccess.h>
 #ifdef HAVE_FDTABLE_HEADER
 #include <linux/fdtable.h>
 #endif
@@ -70,6 +71,46 @@ spl_filp_fallocate(struct file *fp, int mode, loff_t offset, loff_t len)
 	return (error);
 }
 
+static inline ssize_t
+spl_kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
+{
+#if defined(HAVE_KERNEL_WRITE_PPOS)
+	return (kernel_write(file, buf, count, pos));
+#else
+	mm_segment_t saved_fs;
+	ssize_t ret;
+
+	saved_fs = get_fs();
+	set_fs(get_ds());
+
+	ret = vfs_write(file, (__force const char __user *)buf, count, pos);
+
+	set_fs(saved_fs);
+
+	return (ret);
+#endif
+}
+
+static inline ssize_t
+spl_kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
+{
+#if defined(HAVE_KERNEL_READ_PPOS)
+	return (kernel_read(file, buf, count, pos));
+#else
+	mm_segment_t saved_fs;
+	ssize_t ret;
+
+	saved_fs = get_fs();
+	set_fs(get_ds());
+
+	ret = vfs_read(file, (void __user *)buf, count, pos);
+
+	set_fs(saved_fs);
+
+	return (ret);
+#endif
+}
+
 #ifdef HAVE_2ARGS_VFS_FSYNC
 #define	spl_filp_fsync(fp, sync)	vfs_fsync(fp, sync)
 #else
diff --git a/module/spl/spl-vnode.c b/module/spl/spl-vnode.c
index 0e4c386a3..19b3b76cd 100644
--- a/module/spl/spl-vnode.c
+++ b/module/spl/spl-vnode.c
@@ -211,35 +211,22 @@ int
 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 	uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 {
-	loff_t offset;
-	mm_segment_t saved_fs;
-	struct file *fp;
+	struct file *fp = vp->v_file;
+	loff_t offset = off;
 	int rc;
 
 	ASSERT(uio == UIO_WRITE || uio == UIO_READ);
-	ASSERT(vp);
-	ASSERT(vp->v_file);
 	ASSERT(seg == UIO_SYSSPACE);
 	ASSERT((ioflag & ~FAPPEND) == 0);
 
-	fp = vp->v_file;
-
-	offset = off;
 	if (ioflag & FAPPEND)
 		offset = fp->f_pos;
 
-	/* Writable user data segment must be briefly increased for this
-	 * process so we can use the user space read call paths to write
-	 * in to memory allocated by the kernel. */
-	saved_fs = get_fs();
-        set_fs(get_ds());
-
 	if (uio & UIO_WRITE)
-		rc = vfs_write(fp, addr, len, &offset);
+		rc = spl_kernel_write(fp, addr, len, &offset);
 	else
-		rc = vfs_read(fp, addr, len, &offset);
+		rc = spl_kernel_read(fp, addr, len, &offset);
 
-	set_fs(saved_fs);
 	fp->f_pos = offset;
 
 	if (rc < 0)