[PATCH] [RFC] kernfs: Allow vm_ops->close() if VMA is never split

From: Richard Weinberger <richard@nod.at>
To: linux-kernel@vger.kernel.org
Cc: tj@kernel.org, gregkh@linuxfoundation.org, ebiederm@xmission.com,
	dan.j.williams@intel.com, Richard Weinberger <richard@nod.at>
Subject: [PATCH] [RFC] kernfs: Allow vm_ops->close() if VMA is never split
Date: Sat,  4 Jul 2020 22:56:19 +0200	[thread overview]
Message-ID: <20200704205619.11172-1-richard@nod.at> (raw)

10 years ago commit a6849fa1f7d7 ("sysfs: Fail bin file mmap if vma close is implemented.")
removed support for vm_ops->close() for mmap on sysfs.
As far I understand the reason is that due to the wrapping in kernfs
every VMA split operation needs to be tracked to call vm_ops->close()
for all fragments. This is not feasible with reasonable effort.

Since commit 31383c6865a5 ("mm, hugetlbfs: introduce ->split() to vm_operations_struct")
we can get notified as soon a VMA is split, this can help to relax the restriction.
So I propose to allow having a custom close under the condition that a
VMA cannot get split.

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/kernfs/file.c | 42 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 34 insertions(+), 8 deletions(-)

diff --git a/fs/kernfs/file.c b/fs/kernfs/file.c
index 06b342d8462b..82b09e72acff 100644
--- a/fs/kernfs/file.c
+++ b/fs/kernfs/file.c
@@ -347,6 +347,38 @@ static void kernfs_vma_open(struct vm_area_struct *vma)
 	kernfs_put_active(of->kn);
 }
 
+static void kernfs_vma_close(struct vm_area_struct *vma)
+{
+	struct file *file = vma->vm_file;
+	struct kernfs_open_file *of = kernfs_of(file);
+
+	if (!of->vm_ops)
+		return;
+
+	if (!kernfs_get_active(of->kn))
+		return;
+
+	if (of->vm_ops->close)
+		of->vm_ops->close(vma);
+
+	kernfs_put_active(of->kn);
+}
+
+static int kernfs_vma_split(struct vm_area_struct *vma, unsigned long addr)
+{
+	struct file *file = vma->vm_file;
+	struct kernfs_open_file *of = kernfs_of(file);
+
+	/*
+	 * We cannot keep track of split operations,
+	 * so refuse splitting a VMA if someone uses close.
+	 */
+	if (of->vm_ops && of->vm_ops->close)
+		return -EINVAL;
+
+	return 0;
+}
+
 static vm_fault_t kernfs_vma_fault(struct vm_fault *vmf)
 {
 	struct file *file = vmf->vma->vm_file;
@@ -457,6 +489,8 @@ static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
 
 static const struct vm_operations_struct kernfs_vm_ops = {
 	.open		= kernfs_vma_open,
+	.close		= kernfs_vma_close,
+	.split		= kernfs_vma_split,
 	.fault		= kernfs_vma_fault,
 	.page_mkwrite	= kernfs_vma_page_mkwrite,
 	.access		= kernfs_vma_access,
@@ -505,14 +539,6 @@ static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
 	if (of->mmapped && of->vm_ops != vma->vm_ops)
 		goto out_put;
 
-	/*
-	 * It is not possible to successfully wrap close.
-	 * So error if someone is trying to use close.
-	 */
-	rc = -EINVAL;
-	if (vma->vm_ops && vma->vm_ops->close)
-		goto out_put;
-
 	rc = 0;
 	of->mmapped = true;
 	of->vm_ops = vma->vm_ops;
-- 
2.26.2

