[PATCH 1/4] fuse: remove FUSE_NOTIFY_CODE_MAX from <uapi/linux/fuse.h>

[PATCH 1/4] fuse: remove FUSE_NOTIFY_CODE_MAX from <uapi/linux/fuse.h>

[Miklos Szeredi]

Constants that change value from version to version have no place in an
interface definition.

Hopefully this won't break anything.

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
 include/uapi/linux/fuse.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index 6b9fb8b08768..30bf0846547f 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -680,7 +680,6 @@ enum fuse_notify_code {
 	FUSE_NOTIFY_DELETE = 6,
 	FUSE_NOTIFY_RESEND = 7,
 	FUSE_NOTIFY_INC_EPOCH = 8,
-	FUSE_NOTIFY_CODE_MAX,
 };
 
 /* The read buffer is required to be at least 8k, but may be much larger */
-- 
2.49.0

[PATCH 2/4] fuse: fix possibly missing fuse_copy_finish() call in fuse_notify()

[Miklos Szeredi]

In case of FUSE_NOTIFY_RESEND and FUSE_NOTIFY_INC_EPOCH fuse_copy_finish()
isn't called.

Fix by always calling fuse_copy_finish() after fuse_notify().  It's a no-op
if called a second time.

Fixes: 760eac73f9f6 ("fuse: Introduce a new notification type for resend pending requests")
Fixes: 2396356a945b ("fuse: add more control over cache invalidation behaviour")
Cc: <stable@vger.kernel.org> # v6.9
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
 fs/fuse/dev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index df793003eb0c..85d05a5e40e9 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -2178,7 +2178,7 @@ static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
 	 */
 	if (!oh.unique) {
 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
-		goto out;
+		goto copy_finish;
 	}
 
 	err = -EINVAL;
-- 
2.49.0

Re: [PATCH 2/4] fuse: fix possibly missing fuse_copy_finish() call in fuse_notify()

[Joanne Koong]

On Tue, Sep 2, 2025 at 7:44 AM Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> In case of FUSE_NOTIFY_RESEND and FUSE_NOTIFY_INC_EPOCH fuse_copy_finish()
> isn't called.
>
> Fix by always calling fuse_copy_finish() after fuse_notify().  It's a no-op
> if called a second time.
>
> Fixes: 760eac73f9f6 ("fuse: Introduce a new notification type for resend pending requests")
> Fixes: 2396356a945b ("fuse: add more control over cache invalidation behaviour")
> Cc: <stable@vger.kernel.org> # v6.9
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
> ---
>  fs/fuse/dev.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index df793003eb0c..85d05a5e40e9 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -2178,7 +2178,7 @@ static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
>          */
>         if (!oh.unique) {
>                 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
> -               goto out;
> +               goto copy_finish;
>         }
>
>         err = -EINVAL;
> --
> 2.49.0
>
>

[PATCH 3/4] fuse: remove redundant calls to fuse_copy_finish() in fuse_notify()

[Miklos Szeredi]

Remove tail calls of fuse_copy_finish(), since it's now done from
fuse_dev_do_write().

No functional change.

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
 fs/fuse/dev.c | 79 +++++++++++++++------------------------------------
 1 file changed, 23 insertions(+), 56 deletions(-)

diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index 85d05a5e40e9..1258acee9704 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -1622,35 +1622,31 @@ static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
 			    struct fuse_copy_state *cs)
 {
 	struct fuse_notify_poll_wakeup_out outarg;
-	int err = -EINVAL;
+	int err;
 
 	if (size != sizeof(outarg))
-		goto err;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto err;
+		return err;
 
 	fuse_copy_finish(cs);
 	return fuse_notify_poll_wakeup(fc, &outarg);
-
-err:
-	fuse_copy_finish(cs);
-	return err;
 }
 
 static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
 				   struct fuse_copy_state *cs)
 {
 	struct fuse_notify_inval_inode_out outarg;
-	int err = -EINVAL;
+	int err;
 
 	if (size != sizeof(outarg))
-		goto err;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto err;
+		return err;
 	fuse_copy_finish(cs);
 
 	down_read(&fc->killsb);
@@ -1658,10 +1654,6 @@ static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
 				       outarg.off, outarg.len);
 	up_read(&fc->killsb);
 	return err;
-
-err:
-	fuse_copy_finish(cs);
-	return err;
 }
 
 static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
@@ -1669,29 +1661,26 @@ static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
 {
 	struct fuse_notify_inval_entry_out outarg;
 	int err;
-	char *buf = NULL;
+	char *buf;
 	struct qstr name;
 
-	err = -EINVAL;
 	if (size < sizeof(outarg))
-		goto err;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto err;
+		return err;
 
-	err = -ENAMETOOLONG;
 	if (outarg.namelen > fc->name_max)
-		goto err;
+		return -ENAMETOOLONG;
 
 	err = -EINVAL;
 	if (size != sizeof(outarg) + outarg.namelen + 1)
-		goto err;
+		return -EINVAL;
 
-	err = -ENOMEM;
 	buf = kzalloc(outarg.namelen + 1, GFP_KERNEL);
 	if (!buf)
-		goto err;
+		return -ENOMEM;
 
 	name.name = buf;
 	name.len = outarg.namelen;
@@ -1704,12 +1693,8 @@ static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
 	down_read(&fc->killsb);
 	err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
 	up_read(&fc->killsb);
-	kfree(buf);
-	return err;
-
 err:
 	kfree(buf);
-	fuse_copy_finish(cs);
 	return err;
 }
 
@@ -1718,29 +1703,25 @@ static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
 {
 	struct fuse_notify_delete_out outarg;
 	int err;
-	char *buf = NULL;
+	char *buf;
 	struct qstr name;
 
-	err = -EINVAL;
 	if (size < sizeof(outarg))
-		goto err;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto err;
+		return err;
 
-	err = -ENAMETOOLONG;
 	if (outarg.namelen > fc->name_max)
-		goto err;
+		return -ENAMETOOLONG;
 
-	err = -EINVAL;
 	if (size != sizeof(outarg) + outarg.namelen + 1)
-		goto err;
+		return -EINVAL;
 
-	err = -ENOMEM;
 	buf = kzalloc(outarg.namelen + 1, GFP_KERNEL);
 	if (!buf)
-		goto err;
+		return -ENOMEM;
 
 	name.name = buf;
 	name.len = outarg.namelen;
@@ -1753,12 +1734,8 @@ static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
 	down_read(&fc->killsb);
 	err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
 	up_read(&fc->killsb);
-	kfree(buf);
-	return err;
-
 err:
 	kfree(buf);
-	fuse_copy_finish(cs);
 	return err;
 }
 
@@ -1776,17 +1753,15 @@ static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
 	loff_t file_size;
 	loff_t end;
 
-	err = -EINVAL;
 	if (size < sizeof(outarg))
-		goto out_finish;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto out_finish;
+		return err;
 
-	err = -EINVAL;
 	if (size - sizeof(outarg) != outarg.size)
-		goto out_finish;
+		return -EINVAL;
 
 	nodeid = outarg.nodeid;
 
@@ -1846,8 +1821,6 @@ static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
 	iput(inode);
 out_up_killsb:
 	up_read(&fc->killsb);
-out_finish:
-	fuse_copy_finish(cs);
 	return err;
 }
 
@@ -1962,13 +1935,12 @@ static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
 	u64 nodeid;
 	int err;
 
-	err = -EINVAL;
 	if (size != sizeof(outarg))
-		goto copy_finish;
+		return -EINVAL;
 
 	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
 	if (err)
-		goto copy_finish;
+		return err;
 
 	fuse_copy_finish(cs);
 
@@ -1984,10 +1956,6 @@ static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
 	up_read(&fc->killsb);
 
 	return err;
-
-copy_finish:
-	fuse_copy_finish(cs);
-	return err;
 }
 
 /*
@@ -2098,7 +2066,6 @@ static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
 		return fuse_notify_inc_epoch(fc);
 
 	default:
-		fuse_copy_finish(cs);
 		return -EINVAL;
 	}
 }
-- 
2.49.0

Re: [PATCH 3/4] fuse: remove redundant calls to fuse_copy_finish() in fuse_notify()

[Joanne Koong]

On Tue, Sep 2, 2025 at 7:44 AM Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> Remove tail calls of fuse_copy_finish(), since it's now done from
> fuse_dev_do_write().
>
> No functional change.
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Reviewed-by: Joanne Koong <joannelkoong@gmail.com>

> ---
>  fs/fuse/dev.c | 79 +++++++++++++++------------------------------------
>  1 file changed, 23 insertions(+), 56 deletions(-)
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 85d05a5e40e9..1258acee9704 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -1622,35 +1622,31 @@ static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
>                             struct fuse_copy_state *cs)
>  {
>         struct fuse_notify_poll_wakeup_out outarg;
> -       int err = -EINVAL;
> +       int err;
>
>         if (size != sizeof(outarg))
> -               goto err;
> +               return -EINVAL;
>
>         err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>         if (err)
> -               goto err;
> +               return err;
>
>         fuse_copy_finish(cs);

Maybe worth also removing fuse_copy_finish() here (and for the other
notify handlers) since it's going to get called anyways?

Thanks,
Joanne

>         return fuse_notify_poll_wakeup(fc, &outarg);
> -

Re: [PATCH 3/4] fuse: remove redundant calls to fuse_copy_finish() in fuse_notify()

[Luis Henriques]

On Tue, Sep 02 2025, Miklos Szeredi wrote:

> Remove tail calls of fuse_copy_finish(), since it's now done from
> fuse_dev_do_write().
>
> No functional change.
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>  fs/fuse/dev.c | 79 +++++++++++++++------------------------------------
>  1 file changed, 23 insertions(+), 56 deletions(-)
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 85d05a5e40e9..1258acee9704 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -1622,35 +1622,31 @@ static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
>  			    struct fuse_copy_state *cs)
>  {
>  	struct fuse_notify_poll_wakeup_out outarg;
> -	int err = -EINVAL;
> +	int err;
>  
>  	if (size != sizeof(outarg))
> -		goto err;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto err;
> +		return err;
>  
>  	fuse_copy_finish(cs);
>  	return fuse_notify_poll_wakeup(fc, &outarg);
> -
> -err:
> -	fuse_copy_finish(cs);
> -	return err;
>  }
>  
>  static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
>  				   struct fuse_copy_state *cs)
>  {
>  	struct fuse_notify_inval_inode_out outarg;
> -	int err = -EINVAL;
> +	int err;
>  
>  	if (size != sizeof(outarg))
> -		goto err;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto err;
> +		return err;
>  	fuse_copy_finish(cs);
 
I wonder if these extra fuse_copy_finish() calls should also be removed.
It doesn't seem to be a problem to call it twice, but maybe it's not
needed, or am I missing something?  This happens in a few places.

Other than that, and FWIW, the series look good to me.

Cheers,
-- 
Luís

> 
>  	down_read(&fc->killsb);
> @@ -1658,10 +1654,6 @@ static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
>  				       outarg.off, outarg.len);
>  	up_read(&fc->killsb);
>  	return err;
> -
> -err:
> -	fuse_copy_finish(cs);
> -	return err;
>  }
>  
>  static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
> @@ -1669,29 +1661,26 @@ static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
>  {
>  	struct fuse_notify_inval_entry_out outarg;
>  	int err;
> -	char *buf = NULL;
> +	char *buf;
>  	struct qstr name;
>  
> -	err = -EINVAL;
>  	if (size < sizeof(outarg))
> -		goto err;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto err;
> +		return err;
>  
> -	err = -ENAMETOOLONG;
>  	if (outarg.namelen > fc->name_max)
> -		goto err;
> +		return -ENAMETOOLONG;
>  
>  	err = -EINVAL;
>  	if (size != sizeof(outarg) + outarg.namelen + 1)
> -		goto err;
> +		return -EINVAL;
>  
> -	err = -ENOMEM;
>  	buf = kzalloc(outarg.namelen + 1, GFP_KERNEL);
>  	if (!buf)
> -		goto err;
> +		return -ENOMEM;
>  
>  	name.name = buf;
>  	name.len = outarg.namelen;
> @@ -1704,12 +1693,8 @@ static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
>  	down_read(&fc->killsb);
>  	err = fuse_reverse_inval_entry(fc, outarg.parent, 0, &name, outarg.flags);
>  	up_read(&fc->killsb);
> -	kfree(buf);
> -	return err;
> -
>  err:
>  	kfree(buf);
> -	fuse_copy_finish(cs);
>  	return err;
>  }
>  
> @@ -1718,29 +1703,25 @@ static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
>  {
>  	struct fuse_notify_delete_out outarg;
>  	int err;
> -	char *buf = NULL;
> +	char *buf;
>  	struct qstr name;
>  
> -	err = -EINVAL;
>  	if (size < sizeof(outarg))
> -		goto err;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto err;
> +		return err;
>  
> -	err = -ENAMETOOLONG;
>  	if (outarg.namelen > fc->name_max)
> -		goto err;
> +		return -ENAMETOOLONG;
>  
> -	err = -EINVAL;
>  	if (size != sizeof(outarg) + outarg.namelen + 1)
> -		goto err;
> +		return -EINVAL;
>  
> -	err = -ENOMEM;
>  	buf = kzalloc(outarg.namelen + 1, GFP_KERNEL);
>  	if (!buf)
> -		goto err;
> +		return -ENOMEM;
>  
>  	name.name = buf;
>  	name.len = outarg.namelen;
> @@ -1753,12 +1734,8 @@ static int fuse_notify_delete(struct fuse_conn *fc, unsigned int size,
>  	down_read(&fc->killsb);
>  	err = fuse_reverse_inval_entry(fc, outarg.parent, outarg.child, &name, 0);
>  	up_read(&fc->killsb);
> -	kfree(buf);
> -	return err;
> -
>  err:
>  	kfree(buf);
> -	fuse_copy_finish(cs);
>  	return err;
>  }
>  
> @@ -1776,17 +1753,15 @@ static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
>  	loff_t file_size;
>  	loff_t end;
>  
> -	err = -EINVAL;
>  	if (size < sizeof(outarg))
> -		goto out_finish;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto out_finish;
> +		return err;
>  
> -	err = -EINVAL;
>  	if (size - sizeof(outarg) != outarg.size)
> -		goto out_finish;
> +		return -EINVAL;
>  
>  	nodeid = outarg.nodeid;
>  
> @@ -1846,8 +1821,6 @@ static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
>  	iput(inode);
>  out_up_killsb:
>  	up_read(&fc->killsb);
> -out_finish:
> -	fuse_copy_finish(cs);
>  	return err;
>  }
>  
> @@ -1962,13 +1935,12 @@ static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
>  	u64 nodeid;
>  	int err;
>  
> -	err = -EINVAL;
>  	if (size != sizeof(outarg))
> -		goto copy_finish;
> +		return -EINVAL;
>  
>  	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
>  	if (err)
> -		goto copy_finish;
> +		return err;
>  
>  	fuse_copy_finish(cs);
>  
> @@ -1984,10 +1956,6 @@ static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
>  	up_read(&fc->killsb);
>  
>  	return err;
> -
> -copy_finish:
> -	fuse_copy_finish(cs);
> -	return err;
>  }
>  
>  /*
> @@ -2098,7 +2066,6 @@ static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
>  		return fuse_notify_inc_epoch(fc);
>  
>  	default:
> -		fuse_copy_finish(cs);
>  		return -EINVAL;
>  	}
>  }
> -- 
> 2.49.0
>

Re: [PATCH 3/4] fuse: remove redundant calls to fuse_copy_finish() in fuse_notify()

[Miklos Szeredi]

On Wed, 3 Sept 2025 at 12:26, Luis Henriques <luis@igalia.com> wrote:

> I wonder if these extra fuse_copy_finish() calls should also be removed.
> It doesn't seem to be a problem to call it twice, but maybe it's not
> needed, or am I missing something?  This happens in a few places.

It's not a clear no-op, since the put_page() for the userspace buffer
is being moved to after processing the notification.

I don't think this should cause problems, but it's something that I
wouldn't like to do without a bit of thought.

Thanks,
Miklos

[PATCH 4/4] fuse: add prune notification

[Miklos Szeredi]

Some fuse servers need to prune their caches, which can only be done if the
kernel's own dentry/inode caches are pruned first to avoid dangling
references.

Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
of.  Inodes with active references are skipped.

A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
the FUSE_EXPIRE_ONLY flag.  Differences in the interface are

FUSE_NOTIFY_INVAL_ENTRY:

  - can only prune one dentry

  - dentry is determined by parent ID and name

  - if inode has multiple aliases (cached hard links), then they would have
    to be invalidated individually to be able to get rid of the inode

FUSE_NOTIFY_PRUNE:

  - can prune multiple inodes

  - inodes determined by their node ID

  - aliases are taken care of automatically

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
 fs/fuse/dev.c             | 39 +++++++++++++++++++++++++++++++++++++++
 fs/fuse/fuse_i.h          |  6 ++++++
 fs/fuse/inode.c           | 11 +++++++++++
 include/uapi/linux/fuse.h |  8 ++++++++
 4 files changed, 64 insertions(+)

diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index 1258acee9704..4229b38546bb 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -2034,6 +2034,42 @@ static int fuse_notify_inc_epoch(struct fuse_conn *fc)
 	return 0;
 }
 
+static int fuse_notify_prune(struct fuse_conn *fc, unsigned int size,
+			     struct fuse_copy_state *cs)
+{
+	struct fuse_notify_prune_out outarg;
+	const unsigned int batch = 512;
+	u64 *nodeids __free(kfree) = kmalloc(sizeof(u64) * batch, GFP_KERNEL);
+	unsigned int num, i;
+	int err;
+
+	if (!nodeids)
+		return -ENOMEM;
+
+	if (size < sizeof(outarg))
+		return -EINVAL;
+
+	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
+	if (err)
+		return err;
+
+	if (size - sizeof(outarg) != outarg.count * sizeof(u64))
+		return -EINVAL;
+
+	for (; outarg.count; outarg.count -= num) {
+		num = min(batch, outarg.count);
+		err = fuse_copy_one(cs, nodeids, num * sizeof(u64));
+		if (err)
+			return err;
+
+		scoped_guard(rwsem_read, &fc->killsb) {
+			for (i = 0; i < num; i++)
+				fuse_try_prune_one_inode(fc, nodeids[i]);
+		}
+	}
+	return 0;
+}
+
 static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
 		       unsigned int size, struct fuse_copy_state *cs)
 {
@@ -2065,6 +2101,9 @@ static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
 	case FUSE_NOTIFY_INC_EPOCH:
 		return fuse_notify_inc_epoch(fc);
 
+	case FUSE_NOTIFY_PRUNE:
+		return fuse_notify_prune(fc, size, cs);
+
 	default:
 		return -EINVAL;
 	}
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 233c6111f768..fb6604120b53 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -1413,6 +1413,12 @@ int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
 			     u64 child_nodeid, struct qstr *name, u32 flags);
 
+/*
+ * Try to prune this inode.  If neither the inode itself nor dentries associated
+ * with this inode have any external reference, then the inode can be freed.
+ */
+void fuse_try_prune_one_inode(struct fuse_conn *fc, u64 nodeid);
+
 int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
 		 bool isdir);
 
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 5b7897bf7e45..a4d361b34d06 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -585,6 +585,17 @@ int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 	return 0;
 }
 
+void fuse_try_prune_one_inode(struct fuse_conn *fc, u64 nodeid)
+{
+	struct inode *inode;
+
+	inode = fuse_ilookup(fc, nodeid,  NULL);
+	if (!inode)
+		return;
+	d_prune_aliases(inode);
+	iput(inode);
+}
+
 bool fuse_lock_inode(struct inode *inode)
 {
 	bool locked = false;
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index 30bf0846547f..c13e1f9a2f12 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -239,6 +239,7 @@
  *  7.45
  *  - add FUSE_COPY_FILE_RANGE_64
  *  - add struct fuse_copy_file_range_out
+ *  - add FUSE_NOTIFY_PRUNE
  */
 
 #ifndef _LINUX_FUSE_H
@@ -680,6 +681,7 @@ enum fuse_notify_code {
 	FUSE_NOTIFY_DELETE = 6,
 	FUSE_NOTIFY_RESEND = 7,
 	FUSE_NOTIFY_INC_EPOCH = 8,
+	FUSE_NOTIFY_PRUNE = 9,
 };
 
 /* The read buffer is required to be at least 8k, but may be much larger */
@@ -1118,6 +1120,12 @@ struct fuse_notify_retrieve_in {
 	uint64_t	dummy4;
 };
 
+struct fuse_notify_prune_out {
+	uint32_t	count;
+	uint32_t	padding;
+	uint64_t	spare;
+};
+
 struct fuse_backing_map {
 	int32_t		fd;
 	uint32_t	flags;
-- 
2.49.0

Re: [PATCH 4/4] fuse: add prune notification

[Joanne Koong]

On Tue, Sep 2, 2025 at 7:44 AM Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> Some fuse servers need to prune their caches, which can only be done if the
> kernel's own dentry/inode caches are pruned first to avoid dangling
> references.
>
> Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
> of.  Inodes with active references are skipped.
>
> A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
> the FUSE_EXPIRE_ONLY flag.  Differences in the interface are
>
> FUSE_NOTIFY_INVAL_ENTRY:
>
>   - can only prune one dentry
>
>   - dentry is determined by parent ID and name
>
>   - if inode has multiple aliases (cached hard links), then they would have
>     to be invalidated individually to be able to get rid of the inode
>
> FUSE_NOTIFY_PRUNE:
>
>   - can prune multiple inodes
>
>   - inodes determined by their node ID
>
>   - aliases are taken care of automatically
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Reviewed-by: Joanne Koong <joannelkoong@gmail.com>

Re: [PATCH 4/4] fuse: add prune notification

[Jim Harris]

[-- Attachment #1: Type: text/plain, Size: 1188 bytes --]



> On Sep 2, 2025, at 7:41 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
> 
> External email: Use caution opening links or attachments
> 
> 
> Some fuse servers need to prune their caches, which can only be done if the
> kernel's own dentry/inode caches are pruned first to avoid dangling
> references.
> 
> Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
> of.  Inodes with active references are skipped.
> 
> A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
> the FUSE_EXPIRE_ONLY flag.  Differences in the interface are
> 
> FUSE_NOTIFY_INVAL_ENTRY:
> 
>  - can only prune one dentry
> 
>  - dentry is determined by parent ID and name
> 
>  - if inode has multiple aliases (cached hard links), then they would have
>    to be invalidated individually to be able to get rid of the inode
> 
> FUSE_NOTIFY_PRUNE:
> 
>  - can prune multiple inodes
> 
>  - inodes determined by their node ID
> 
>  - aliases are taken care of automatically
> 
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Thanks Miklos, this looks great. I’ll give this a spin in our virtio-fs FUSE device.

<snip>

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 4312 bytes --]

Re: [PATCH 4/4] fuse: add prune notification

[Jim Harris]

[-- Attachment #1: Type: text/plain, Size: 1536 bytes --]



> On Sep 4, 2025, at 2:07 PM, Jim Harris <jiharris@nvidia.com> wrote:
> 
> 
> 
>> On Sep 2, 2025, at 7:41 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>> 
>> External email: Use caution opening links or attachments
>> 
>> 
>> Some fuse servers need to prune their caches, which can only be done if the
>> kernel's own dentry/inode caches are pruned first to avoid dangling
>> references.
>> 
>> Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
>> of.  Inodes with active references are skipped.
>> 
>> A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
>> the FUSE_EXPIRE_ONLY flag.  Differences in the interface are
>> 
>> FUSE_NOTIFY_INVAL_ENTRY:
>> 
>> - can only prune one dentry
>> 
>> - dentry is determined by parent ID and name
>> 
>> - if inode has multiple aliases (cached hard links), then they would have
>>   to be invalidated individually to be able to get rid of the inode
>> 
>> FUSE_NOTIFY_PRUNE:
>> 
>> - can prune multiple inodes
>> 
>> - inodes determined by their node ID
>> 
>> - aliases are taken care of automatically
>> 
>> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> 
> Thanks Miklos, this looks great. I’ll give this a spin in our virtio-fs FUSE device.

Hi Miklos,

I was finally was able to give these patches a spin. They work great, a nice
simplification compared to FUSE_NOTIFY_INVAL_ENTRY. Feel free to add:

Tested-by: Jim Harris <jim.harris@nvidia.com>

Best regards,

-Jim


[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 4312 bytes --]

Re: [PATCH 4/4] fuse: add prune notification

[Jim Harris]

[-- Attachment #1: Type: text/plain, Size: 1536 bytes --]



> On Sep 4, 2025, at 2:07 PM, Jim Harris <jiharris@nvidia.com> wrote:
> 
> 
> 
>> On Sep 2, 2025, at 7:41 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>> 
>> External email: Use caution opening links or attachments
>> 
>> 
>> Some fuse servers need to prune their caches, which can only be done if the
>> kernel's own dentry/inode caches are pruned first to avoid dangling
>> references.
>> 
>> Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
>> of.  Inodes with active references are skipped.
>> 
>> A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
>> the FUSE_EXPIRE_ONLY flag.  Differences in the interface are
>> 
>> FUSE_NOTIFY_INVAL_ENTRY:
>> 
>> - can only prune one dentry
>> 
>> - dentry is determined by parent ID and name
>> 
>> - if inode has multiple aliases (cached hard links), then they would have
>>   to be invalidated individually to be able to get rid of the inode
>> 
>> FUSE_NOTIFY_PRUNE:
>> 
>> - can prune multiple inodes
>> 
>> - inodes determined by their node ID
>> 
>> - aliases are taken care of automatically
>> 
>> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> 
> Thanks Miklos, this looks great. I’ll give this a spin in our virtio-fs FUSE device.

Hi Miklos,

I was finally was able to give these patches a spin. They work great, a nice
simplification compared to FUSE_NOTIFY_INVAL_ENTRY. Feel free to add:

Tested-by: Jim Harris <jim.harris@nvidia.com>

Best regards,

-Jim


[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 4312 bytes --]

Re: [PATCH 4/4] fuse: add prune notification

[Bernd Schubert]

Hi Miklos,

I'm sorry for late review.

On 9/2/25 16:41, Miklos Szeredi wrote:
> Some fuse servers need to prune their caches, which can only be done if the
> kernel's own dentry/inode caches are pruned first to avoid dangling
> references.
> 
> Add FUSE_NOTIFY_PRUNE, which takes an array of node ID's to try and get rid
> of.  Inodes with active references are skipped.
> 
> A similar functionality is already provided by FUSE_NOTIFY_INVAL_ENTRY with
> the FUSE_EXPIRE_ONLY flag.  Differences in the interface are
> 
> FUSE_NOTIFY_INVAL_ENTRY:
> 
>   - can only prune one dentry
> 
>   - dentry is determined by parent ID and name
> 
>   - if inode has multiple aliases (cached hard links), then they would have
>     to be invalidated individually to be able to get rid of the inode
> 
> FUSE_NOTIFY_PRUNE:
> 
>   - can prune multiple inodes
> 
>   - inodes determined by their node ID
> 
>   - aliases are taken care of automatically


It actually solves another issue we had run into. FUSE_NOTIFY_INVAL_ENTRY takes 
the parent dir lock. Problem with the parent lock is our DLM and lock order.
I.e. kernel holds parent lock for some dir operation, calls into
fuse-server, which tries to acquires the DLM lock. DLM lock might
be taken already and is doing FUSE_NOTIFY_INVAL_ENTRY - deadlock through
the parent dir lock. We are running with an additional patch from Yong Ze,
which is still for single entries.


Thanks,
Bernd


> 
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>  fs/fuse/dev.c             | 39 +++++++++++++++++++++++++++++++++++++++
>  fs/fuse/fuse_i.h          |  6 ++++++
>  fs/fuse/inode.c           | 11 +++++++++++
>  include/uapi/linux/fuse.h |  8 ++++++++
>  4 files changed, 64 insertions(+)
> 
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 1258acee9704..4229b38546bb 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -2034,6 +2034,42 @@ static int fuse_notify_inc_epoch(struct fuse_conn *fc)
>  	return 0;
>  }
>  
> +static int fuse_notify_prune(struct fuse_conn *fc, unsigned int size,
> +			     struct fuse_copy_state *cs)
> +{
> +	struct fuse_notify_prune_out outarg;
> +	const unsigned int batch = 512;
> +	u64 *nodeids __free(kfree) = kmalloc(sizeof(u64) * batch, GFP_KERNEL);
> +	unsigned int num, i;
> +	int err;
> +
> +	if (!nodeids)
> +		return -ENOMEM;
> +
> +	if (size < sizeof(outarg))
> +		return -EINVAL;
> +
> +	err = fuse_copy_one(cs, &outarg, sizeof(outarg));
> +	if (err)
> +		return err;
> +
> +	if (size - sizeof(outarg) != outarg.count * sizeof(u64))
> +		return -EINVAL;
> +
> +	for (; outarg.count; outarg.count -= num) {
> +		num = min(batch, outarg.count);
> +		err = fuse_copy_one(cs, nodeids, num * sizeof(u64));
> +		if (err)
> +			return err;
> +
> +		scoped_guard(rwsem_read, &fc->killsb) {
> +			for (i = 0; i < num; i++)
> +				fuse_try_prune_one_inode(fc, nodeids[i]);
> +		}
> +	}
> +	return 0;
> +}
> +
>  static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
>  		       unsigned int size, struct fuse_copy_state *cs)
>  {
> @@ -2065,6 +2101,9 @@ static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
>  	case FUSE_NOTIFY_INC_EPOCH:
>  		return fuse_notify_inc_epoch(fc);
>  
> +	case FUSE_NOTIFY_PRUNE:
> +		return fuse_notify_prune(fc, size, cs);
> +
>  	default:
>  		return -EINVAL;
>  	}
> diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
> index 233c6111f768..fb6604120b53 100644
> --- a/fs/fuse/fuse_i.h
> +++ b/fs/fuse/fuse_i.h
> @@ -1413,6 +1413,12 @@ int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
>  int fuse_reverse_inval_entry(struct fuse_conn *fc, u64 parent_nodeid,
>  			     u64 child_nodeid, struct qstr *name, u32 flags);
>  
> +/*
> + * Try to prune this inode.  If neither the inode itself nor dentries associated
> + * with this inode have any external reference, then the inode can be freed.
> + */
> +void fuse_try_prune_one_inode(struct fuse_conn *fc, u64 nodeid);
> +
>  int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
>  		 bool isdir);
>  
> diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
> index 5b7897bf7e45..a4d361b34d06 100644
> --- a/fs/fuse/inode.c
> +++ b/fs/fuse/inode.c
> @@ -585,6 +585,17 @@ int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
>  	return 0;
>  }
>  
> +void fuse_try_prune_one_inode(struct fuse_conn *fc, u64 nodeid)
> +{
> +	struct inode *inode;
> +
> +	inode = fuse_ilookup(fc, nodeid,  NULL);
> +	if (!inode)
> +		return;
> +	d_prune_aliases(inode);
> +	iput(inode);

I also wonder if FUSE_NOTIFY_PRUNE shouldn't handle dir inodes and
call d_invalidate() to also prune child entries. 



Thanks,
Bernd

Re: [PATCH 4/4] fuse: add prune notification

[Miklos Szeredi]

On Wed, 1 Oct 2025 at 14:29, Bernd Schubert <bernd@bsbernd.com> wrote:

> I also wonder if FUSE_NOTIFY_PRUNE shouldn't handle dir inodes and
> call d_invalidate() to also prune child entries.

d_invalidate unconditionally unhashes the dentry, which this interface
should only do if the dentry is unused.

However, it could call shrink_dcache_parent() before trying to prune the dentry.

Thanks,
Miklos

Re: [PATCH 1/4] fuse: remove FUSE_NOTIFY_CODE_MAX from <uapi/linux/fuse.h>

[Joanne Koong]

On Tue, Sep 2, 2025 at 7:46 AM Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> Constants that change value from version to version have no place in an
> interface definition.
>
> Hopefully this won't break anything.
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
> ---
>  include/uapi/linux/fuse.h | 1 -
>  1 file changed, 1 deletion(-)
>
> diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
> index 6b9fb8b08768..30bf0846547f 100644
> --- a/include/uapi/linux/fuse.h
> +++ b/include/uapi/linux/fuse.h
> @@ -680,7 +680,6 @@ enum fuse_notify_code {
>         FUSE_NOTIFY_DELETE = 6,
>         FUSE_NOTIFY_RESEND = 7,
>         FUSE_NOTIFY_INC_EPOCH = 8,
> -       FUSE_NOTIFY_CODE_MAX,
>  };
>
>  /* The read buffer is required to be at least 8k, but may be much larger */
> --
> 2.49.0
>
>