fuse: Fix O_DIRECT operations vs cached writeback misorder

The problem is:

1. write cached data to a file
2. read directly from the same file (via another fd)

The 2nd operation may read stale data, i.e. the one that was in a file
before the 1st op. Problem is in how fuse manages writeback.

When direct op occurs the core kernel code calls filemap_write_and_wait
to flush all the cached ops in flight. But fuse acks the writeback right
after the ->writepages callback exits w/o waiting for the real write to
happen. Thus the subsequent direct op proceeds while the real writeback
is still in flight. This is a problem for backends that reorder operation.

Fix this by making the fuse direct IO callback explicitly wait on the
in-flight writeback to finish.

Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>

3 files changed, 41 insertions, 9 deletions

diff --git a/fs/fuse/cuse.c b/fs/fuse/cuse.c
index b96a49b..23e363f 100644
--- a/fs/fuse/cuse.c
+++ b/fs/fuse/cuse.c
@@ -95,7 +95,7 @@ static ssize_t cuse_read(struct file *file, char __user *buf, size_t count,
 	struct iovec iov = { .iov_base = buf, .iov_len = count };
 	struct fuse_io_priv io = { .async = 0, .file = file };
 
-	return fuse_direct_io(&io, &iov, 1, count, &pos, 0);
+	return fuse_direct_io(&io, &iov, 1, count, &pos, FUSE_DIO_CUSE);
 }
 
 static ssize_t cuse_write(struct file *file, const char __user *buf,
@@ -109,7 +109,8 @@ static ssize_t cuse_write(struct file *file, const char __user *buf,
 	 * No locking or generic_write_checks(), the server is
 	 * responsible for locking and sanity checks.
 	 */
-	return fuse_direct_io(&io, &iov, 1, count, &pos, 1);
+	return fuse_direct_io(&io, &iov, 1, count, &pos,
+			      FUSE_DIO_WRITE | FUSE_DIO_CUSE);
 }
 
 static int cuse_open(struct inode *inode, struct file *file)
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index d93f2a1..2764330 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -358,12 +358,13 @@ u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
 }
 
 /*
- * Check if page is under writeback
+ * Check if any page in a range is under writeback
  *
  * This is currently done by walking the list of writepage requests
  * for the inode, which can be pretty inefficient.
  */
-static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
+static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
+				   pgoff_t idx_to)
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
@@ -376,8 +377,8 @@ static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
 
 		BUG_ON(req->inode != inode);
 		curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
-		if (curr_index <= index &&
-		    index < curr_index + req->num_pages) {
+		if (idx_from < curr_index + req->num_pages &&
+		    curr_index <= idx_to) {
 			found = true;
 			break;
 		}
@@ -387,6 +388,11 @@ static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
 	return found;
 }
 
+static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
+{
+	return fuse_range_is_writeback(inode, index, index);
+}
+
 /*
  * Wait for page writeback to be completed.
  *
@@ -1364,13 +1370,18 @@ static inline int fuse_iter_npages(const struct iov_iter *ii_p)
 
 ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov,
 		       unsigned long nr_segs, size_t count, loff_t *ppos,
-		       int write)
+		       int flags)
 {
+	int write = flags & FUSE_DIO_WRITE;
+	int cuse = flags & FUSE_DIO_CUSE;
 	struct file *file = io->file;
+	struct inode *inode = file->f_mapping->host;
 	struct fuse_file *ff = file->private_data;
 	struct fuse_conn *fc = ff->fc;
 	size_t nmax = write ? fc->max_write : fc->max_read;
 	loff_t pos = *ppos;
+	pgoff_t idx_from = pos >> PAGE_CACHE_SHIFT;
+	pgoff_t idx_to = (pos + count - 1) >> PAGE_CACHE_SHIFT;
 	ssize_t res = 0;
 	struct fuse_req *req;
 	struct iov_iter ii;
@@ -1384,6 +1395,14 @@ ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov,
 	if (IS_ERR(req))
 		return PTR_ERR(req);
 
+	if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
+		if (!write)
+			mutex_lock(&inode->i_mutex);
+		fuse_sync_writes(inode);
+		if (!write)
+			mutex_unlock(&inode->i_mutex);
+	}
+
 	while (count) {
 		size_t nres;
 		fl_owner_t owner = current->files;
@@ -1472,7 +1491,8 @@ static ssize_t __fuse_direct_write(struct fuse_io_priv *io,
 
 	res = generic_write_checks(file, ppos, &count, 0);
 	if (!res)
-		res = fuse_direct_io(io, iov, nr_segs, count, ppos, 1);
+		res = fuse_direct_io(io, iov, nr_segs, count, ppos,
+				     FUSE_DIO_WRITE);
 
 	fuse_invalidate_attr(inode);
 
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 1e6ad6d..a257ed8e 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -868,9 +868,20 @@ int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
 
 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 		 bool isdir);
+
+/**
+ * fuse_direct_io() flags
+ */
+
+/** If set, it is WRITE; otherwise - READ */
+#define FUSE_DIO_WRITE (1 << 0)
+
+/** CUSE pass fuse_direct_io() a file which f_mapping->host is not from FUSE */
+#define FUSE_DIO_CUSE  (1 << 1)
+
 ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov,
 		       unsigned long nr_segs, size_t count, loff_t *ppos,
-		       int write);
+		       int flags);
 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
 		   unsigned int flags);
 long fuse_ioctl_common(struct file *file, unsigned int cmd,