[PATCH v3] fuse: invalidate page cache after sync and async direct writes

[PATCH v3] fuse: invalidate page cache after sync and async direct writes

[Cheng Ding via B4 Relay]

From: Cheng Ding <cding@ddn.com>

Fixes xfstests generic/451, similar to how commit b359af8275a9 ("fuse:
Invalidate the page cache after FOPEN_DIRECT_IO write") fixes xfstests
generic/209.

Signed-off-by: Cheng Ding <cding@ddn.com>
---
Changes in v3:
- Address review comments: fix typo
- Address review comments: move sb_init_dio_done_wq() to fuse_direct_IO()
  Note: We could skip sb_init_dio_done_wq() when io->blocking is true, but
  I opted to keep the change simpler.
- Link to v2: https://lore.kernel.org/r/20260306-xfstests-generic-451-v2-1-93b2d540304b@ddn.com

Changes in v2:
- Address review comments: move invalidation from fuse_direct_io() to
  fuse_direct_write_iter()
- Link to v1: https://lore.kernel.org/r/20260303-async-dio-aio-cache-invalidation-v1-1-fba0fd0426c3@ddn.com
---
 fs/fuse/file.c   | 59 +++++++++++++++++++++++++++++++++++++++++++++-----------
 fs/fuse/fuse_i.h |  1 +
 2 files changed, 49 insertions(+), 11 deletions(-)

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index b1bb7153cb78..c43fe74cdd46 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -23,6 +23,8 @@
 #include <linux/task_io_accounting_ops.h>
 #include <linux/iomap.h>
 
+int sb_init_dio_done_wq(struct super_block *sb);
+
 static int fuse_send_open(struct fuse_mount *fm, u64 nodeid,
 			  unsigned int open_flags, int opcode,
 			  struct fuse_open_out *outargp)
@@ -629,6 +631,19 @@ static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
 	return io->bytes < 0 ? io->size : io->bytes;
 }
 
+static void fuse_aio_invalidate_worker(struct work_struct *work)
+{
+	struct fuse_io_priv *io = container_of(work, struct fuse_io_priv, work);
+	struct address_space *mapping = io->iocb->ki_filp->f_mapping;
+	ssize_t res = fuse_get_res_by_io(io);
+	pgoff_t start = io->offset >> PAGE_SHIFT;
+	pgoff_t end = (io->offset + res - 1) >> PAGE_SHIFT;
+
+	invalidate_inode_pages2_range(mapping, start, end);
+	io->iocb->ki_complete(io->iocb, res);
+	kref_put(&io->refcnt, fuse_io_release);
+}
+
 /*
  * In case of short read, the caller sets 'pos' to the position of
  * actual end of fuse request in IO request. Otherwise, if bytes_requested
@@ -661,10 +676,11 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
 	spin_unlock(&io->lock);
 
 	if (!left && !io->blocking) {
+		struct inode *inode = file_inode(io->iocb->ki_filp);
+		struct address_space *mapping = io->iocb->ki_filp->f_mapping;
 		ssize_t res = fuse_get_res_by_io(io);
 
 		if (res >= 0) {
-			struct inode *inode = file_inode(io->iocb->ki_filp);
 			struct fuse_conn *fc = get_fuse_conn(inode);
 			struct fuse_inode *fi = get_fuse_inode(inode);
 
@@ -673,6 +689,17 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
 			spin_unlock(&fi->lock);
 		}
 
+		if (io->write && res > 0 && mapping->nrpages) {
+			/*
+			 * As in generic_file_direct_write(), invalidate after the
+			 * write, to invalidate read-ahead cache that may have competed
+			 * with the write.
+			 */
+			INIT_WORK(&io->work, fuse_aio_invalidate_worker);
+			queue_work(inode->i_sb->s_dio_done_wq, &io->work);
+			return;
+		}
+
 		io->iocb->ki_complete(io->iocb, res);
 	}
 
@@ -1738,15 +1765,6 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
 	if (res > 0)
 		*ppos = pos;
 
-	if (res > 0 && write && fopen_direct_io) {
-		/*
-		 * As in generic_file_direct_write(), invalidate after the
-		 * write, to invalidate read-ahead cache that may have competed
-		 * with the write.
-		 */
-		invalidate_inode_pages2_range(mapping, idx_from, idx_to);
-	}
-
 	return res > 0 ? res : err;
 }
 EXPORT_SYMBOL_GPL(fuse_direct_io);
@@ -1785,6 +1803,8 @@ static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
 static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
 {
 	struct inode *inode = file_inode(iocb->ki_filp);
+	struct address_space *mapping = inode->i_mapping;
+	loff_t pos = iocb->ki_pos;
 	ssize_t res;
 	bool exclusive;
 
@@ -1801,6 +1821,16 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
 					     FUSE_DIO_WRITE);
 			fuse_write_update_attr(inode, iocb->ki_pos, res);
 		}
+		if (res > 0 && mapping->nrpages) {
+			/*
+			 * As in generic_file_direct_write(), invalidate after
+			 * write, to invalidate read-ahead cache that may have
+			 * with the write.
+			 */
+			invalidate_inode_pages2_range(mapping,
+				pos >> PAGE_SHIFT,
+				(pos + res - 1) >> PAGE_SHIFT);
+		}
 	}
 	fuse_dio_unlock(iocb, exclusive);
 
@@ -2826,6 +2856,7 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 	size_t count = iov_iter_count(iter), shortened = 0;
 	loff_t offset = iocb->ki_pos;
 	struct fuse_io_priv *io;
+	bool async = ff->fm->fc->async_dio;
 
 	pos = offset;
 	inode = file->f_mapping->host;
@@ -2834,6 +2865,12 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 	if ((iov_iter_rw(iter) == READ) && (offset >= i_size))
 		return 0;
 
+	if ((iov_iter_rw(iter) == WRITE) && async && !inode->i_sb->s_dio_done_wq) {
+		ret = sb_init_dio_done_wq(inode->i_sb);
+		if (ret < 0)
+			return ret;
+	}
+
 	io = kmalloc_obj(struct fuse_io_priv);
 	if (!io)
 		return -ENOMEM;
@@ -2849,7 +2886,7 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 	 * By default, we want to optimize all I/Os with async request
 	 * submission to the client filesystem if supported.
 	 */
-	io->async = ff->fm->fc->async_dio;
+	io->async = async;
 	io->iocb = iocb;
 	io->blocking = is_sync_kiocb(iocb);
 
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 7f16049387d1..6e8c8cf6b2c8 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -377,6 +377,7 @@ union fuse_file_args {
 /** The request IO state (for asynchronous processing) */
 struct fuse_io_priv {
 	struct kref refcnt;
+	struct work_struct work;
 	int async;
 	spinlock_t lock;
 	unsigned reqs;

---
base-commit: 3c9332f821aa11552f19c331c5aa5299c78c7c94
change-id: 20260306-xfstests-generic-451-fed0f9d5a095

Best regards,
-- 
Cheng Ding <cding@ddn.com>

Re: [PATCH v3] fuse: invalidate page cache after sync and async direct writes

[Jingbo Xu]

On 3/9/26 4:35 PM, Cheng Ding via B4 Relay wrote:
> From: Cheng Ding <cding@ddn.com>
> 
> Fixes xfstests generic/451, similar to how commit b359af8275a9 ("fuse:
> Invalidate the page cache after FOPEN_DIRECT_IO write") fixes xfstests
> generic/209.
> 
> Signed-off-by: Cheng Ding <cding@ddn.com>

LGTM.

Reviewed-by: Jingbo Xu <jefflexu@linux.alibaba.com>

> ---
> Changes in v3:
> - Address review comments: fix typo
> - Address review comments: move sb_init_dio_done_wq() to fuse_direct_IO()
>   Note: We could skip sb_init_dio_done_wq() when io->blocking is true, but
>   I opted to keep the change simpler.
> - Link to v2: https://lore.kernel.org/r/20260306-xfstests-generic-451-v2-1-93b2d540304b@ddn.com
> 
> Changes in v2:
> - Address review comments: move invalidation from fuse_direct_io() to
>   fuse_direct_write_iter()
> - Link to v1: https://lore.kernel.org/r/20260303-async-dio-aio-cache-invalidation-v1-1-fba0fd0426c3@ddn.com
> ---
>  fs/fuse/file.c   | 59 +++++++++++++++++++++++++++++++++++++++++++++-----------
>  fs/fuse/fuse_i.h |  1 +
>  2 files changed, 49 insertions(+), 11 deletions(-)
> 
> diff --git a/fs/fuse/file.c b/fs/fuse/file.c
> index b1bb7153cb78..c43fe74cdd46 100644
> --- a/fs/fuse/file.c
> +++ b/fs/fuse/file.c
> @@ -23,6 +23,8 @@
>  #include <linux/task_io_accounting_ops.h>
>  #include <linux/iomap.h>
>  
> +int sb_init_dio_done_wq(struct super_block *sb);
> +
>  static int fuse_send_open(struct fuse_mount *fm, u64 nodeid,
>  			  unsigned int open_flags, int opcode,
>  			  struct fuse_open_out *outargp)
> @@ -629,6 +631,19 @@ static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
>  	return io->bytes < 0 ? io->size : io->bytes;
>  }
>  
> +static void fuse_aio_invalidate_worker(struct work_struct *work)
> +{
> +	struct fuse_io_priv *io = container_of(work, struct fuse_io_priv, work);
> +	struct address_space *mapping = io->iocb->ki_filp->f_mapping;
> +	ssize_t res = fuse_get_res_by_io(io);
> +	pgoff_t start = io->offset >> PAGE_SHIFT;
> +	pgoff_t end = (io->offset + res - 1) >> PAGE_SHIFT;
> +
> +	invalidate_inode_pages2_range(mapping, start, end);
> +	io->iocb->ki_complete(io->iocb, res);
> +	kref_put(&io->refcnt, fuse_io_release);
> +}
> +
>  /*
>   * In case of short read, the caller sets 'pos' to the position of
>   * actual end of fuse request in IO request. Otherwise, if bytes_requested
> @@ -661,10 +676,11 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
>  	spin_unlock(&io->lock);
>  
>  	if (!left && !io->blocking) {
> +		struct inode *inode = file_inode(io->iocb->ki_filp);
> +		struct address_space *mapping = io->iocb->ki_filp->f_mapping;
>  		ssize_t res = fuse_get_res_by_io(io);
>  
>  		if (res >= 0) {
> -			struct inode *inode = file_inode(io->iocb->ki_filp);
>  			struct fuse_conn *fc = get_fuse_conn(inode);
>  			struct fuse_inode *fi = get_fuse_inode(inode);
>  
> @@ -673,6 +689,17 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
>  			spin_unlock(&fi->lock);
>  		}
>  
> +		if (io->write && res > 0 && mapping->nrpages) {
> +			/*
> +			 * As in generic_file_direct_write(), invalidate after the
> +			 * write, to invalidate read-ahead cache that may have competed
> +			 * with the write.
> +			 */
> +			INIT_WORK(&io->work, fuse_aio_invalidate_worker);
> +			queue_work(inode->i_sb->s_dio_done_wq, &io->work);
> +			return;
> +		}
> +
>  		io->iocb->ki_complete(io->iocb, res);
>  	}
>  
> @@ -1738,15 +1765,6 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
>  	if (res > 0)
>  		*ppos = pos;
>  
> -	if (res > 0 && write && fopen_direct_io) {
> -		/*
> -		 * As in generic_file_direct_write(), invalidate after the
> -		 * write, to invalidate read-ahead cache that may have competed
> -		 * with the write.
> -		 */
> -		invalidate_inode_pages2_range(mapping, idx_from, idx_to);
> -	}
> -
>  	return res > 0 ? res : err;
>  }
>  EXPORT_SYMBOL_GPL(fuse_direct_io);
> @@ -1785,6 +1803,8 @@ static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
>  static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
>  {
>  	struct inode *inode = file_inode(iocb->ki_filp);
> +	struct address_space *mapping = inode->i_mapping;
> +	loff_t pos = iocb->ki_pos;
>  	ssize_t res;
>  	bool exclusive;
>  
> @@ -1801,6 +1821,16 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
>  					     FUSE_DIO_WRITE);
>  			fuse_write_update_attr(inode, iocb->ki_pos, res);
>  		}
> +		if (res > 0 && mapping->nrpages) {
> +			/*
> +			 * As in generic_file_direct_write(), invalidate after
> +			 * write, to invalidate read-ahead cache that may have
> +			 * with the write.
> +			 */
> +			invalidate_inode_pages2_range(mapping,
> +				pos >> PAGE_SHIFT,
> +				(pos + res - 1) >> PAGE_SHIFT);
> +		}
>  	}
>  	fuse_dio_unlock(iocb, exclusive);
>  
> @@ -2826,6 +2856,7 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
>  	size_t count = iov_iter_count(iter), shortened = 0;
>  	loff_t offset = iocb->ki_pos;
>  	struct fuse_io_priv *io;
> +	bool async = ff->fm->fc->async_dio;
>  
>  	pos = offset;
>  	inode = file->f_mapping->host;
> @@ -2834,6 +2865,12 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
>  	if ((iov_iter_rw(iter) == READ) && (offset >= i_size))
>  		return 0;
>  
> +	if ((iov_iter_rw(iter) == WRITE) && async && !inode->i_sb->s_dio_done_wq) {
> +		ret = sb_init_dio_done_wq(inode->i_sb);
> +		if (ret < 0)
> +			return ret;
> +	}
> +
>  	io = kmalloc_obj(struct fuse_io_priv);
>  	if (!io)
>  		return -ENOMEM;
> @@ -2849,7 +2886,7 @@ fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
>  	 * By default, we want to optimize all I/Os with async request
>  	 * submission to the client filesystem if supported.
>  	 */
> -	io->async = ff->fm->fc->async_dio;
> +	io->async = async;
>  	io->iocb = iocb;
>  	io->blocking = is_sync_kiocb(iocb);
>  
> diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
> index 7f16049387d1..6e8c8cf6b2c8 100644
> --- a/fs/fuse/fuse_i.h
> +++ b/fs/fuse/fuse_i.h
> @@ -377,6 +377,7 @@ union fuse_file_args {
>  /** The request IO state (for asynchronous processing) */
>  struct fuse_io_priv {
>  	struct kref refcnt;
> +	struct work_struct work;
>  	int async;
>  	spinlock_t lock;
>  	unsigned reqs;
> 
> ---
> base-commit: 3c9332f821aa11552f19c331c5aa5299c78c7c94
> change-id: 20260306-xfstests-generic-451-fed0f9d5a095
> 
> Best regards,

-- 
Thanks,
Jingbo