Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse

Pull fuse update from Miklos Szeredi:
 "This series adds cached writeback support to fuse, improving write
  throughput"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  fuse: fix "uninitialized variable" warning
  fuse: Turn writeback cache on
  fuse: Fix O_DIRECT operations vs cached writeback misorder
  fuse: fuse_flush() should wait on writeback
  fuse: Implement write_begin/write_end callbacks
  fuse: restructure fuse_readpage()
  fuse: Flush files on wb close
  fuse: Trust kernel i_mtime only
  fuse: Trust kernel i_size only
  fuse: Connection bit for enabling writeback
  fuse: Prepare to handle short reads
  fuse: Linking file to inode helper

6 files changed, 384 insertions, 84 deletions

diff --git a/fs/fuse/cuse.c b/fs/fuse/cuse.c
index b96a49b37d6..23e363f3830 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/dir.c b/fs/fuse/dir.c
index 1d1292c581c..5b4e035b364 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -839,6 +839,14 @@ static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
 			  struct kstat *stat)
 {
 	unsigned int blkbits;
+	struct fuse_conn *fc = get_fuse_conn(inode);
+
+	/* see the comment in fuse_change_attributes() */
+	if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
+		attr->size = i_size_read(inode);
+		attr->mtime = inode->i_mtime.tv_sec;
+		attr->mtimensec = inode->i_mtime.tv_nsec;
+	}
 
 	stat->dev = inode->i_sb->s_dev;
 	stat->ino = attr->ino;
@@ -1477,12 +1485,16 @@ static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
 				 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
 }
 
-static bool update_mtime(unsigned ivalid)
+static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
 {
 	/* Always update if mtime is explicitly set  */
 	if (ivalid & ATTR_MTIME_SET)
 		return true;
 
+	/* Or if kernel i_mtime is the official one */
+	if (trust_local_mtime)
+		return true;
+
 	/* If it's an open(O_TRUNC) or an ftruncate(), don't update */
 	if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
 		return false;
@@ -1491,7 +1503,8 @@ static bool update_mtime(unsigned ivalid)
 	return true;
 }
 
-static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
+static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg,
+			   bool trust_local_mtime)
 {
 	unsigned ivalid = iattr->ia_valid;
 
@@ -1510,11 +1523,11 @@ static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
 		if (!(ivalid & ATTR_ATIME_SET))
 			arg->valid |= FATTR_ATIME_NOW;
 	}
-	if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
+	if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_mtime)) {
 		arg->valid |= FATTR_MTIME;
 		arg->mtime = iattr->ia_mtime.tv_sec;
 		arg->mtimensec = iattr->ia_mtime.tv_nsec;
-		if (!(ivalid & ATTR_MTIME_SET))
+		if (!(ivalid & ATTR_MTIME_SET) && !trust_local_mtime)
 			arg->valid |= FATTR_MTIME_NOW;
 	}
 }
@@ -1563,6 +1576,63 @@ void fuse_release_nowrite(struct inode *inode)
 	spin_unlock(&fc->lock);
 }
 
+static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_req *req,
+			      struct inode *inode,
+			      struct fuse_setattr_in *inarg_p,
+			      struct fuse_attr_out *outarg_p)
+{
+	req->in.h.opcode = FUSE_SETATTR;
+	req->in.h.nodeid = get_node_id(inode);
+	req->in.numargs = 1;
+	req->in.args[0].size = sizeof(*inarg_p);
+	req->in.args[0].value = inarg_p;
+	req->out.numargs = 1;
+	if (fc->minor < 9)
+		req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
+	else
+		req->out.args[0].size = sizeof(*outarg_p);
+	req->out.args[0].value = outarg_p;
+}
+
+/*
+ * Flush inode->i_mtime to the server
+ */
+int fuse_flush_mtime(struct file *file, bool nofail)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct fuse_inode *fi = get_fuse_inode(inode);
+	struct fuse_conn *fc = get_fuse_conn(inode);
+	struct fuse_req *req = NULL;
+	struct fuse_setattr_in inarg;
+	struct fuse_attr_out outarg;
+	int err;
+
+	if (nofail) {
+		req = fuse_get_req_nofail_nopages(fc, file);
+	} else {
+		req = fuse_get_req_nopages(fc);
+		if (IS_ERR(req))
+			return PTR_ERR(req);
+	}
+
+	memset(&inarg, 0, sizeof(inarg));
+	memset(&outarg, 0, sizeof(outarg));
+
+	inarg.valid |= FATTR_MTIME;
+	inarg.mtime = inode->i_mtime.tv_sec;
+	inarg.mtimensec = inode->i_mtime.tv_nsec;
+
+	fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
+	fuse_request_send(fc, req);
+	err = req->out.h.error;
+	fuse_put_request(fc, req);
+
+	if (!err)
+		clear_bit(FUSE_I_MTIME_DIRTY, &fi->state);
+
+	return err;
+}
+
 /*
  * Set attributes, and at the same time refresh them.
  *
@@ -1580,8 +1650,10 @@ int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 	struct fuse_setattr_in inarg;
 	struct fuse_attr_out outarg;
 	bool is_truncate = false;
+	bool is_wb = fc->writeback_cache;
 	loff_t oldsize;
 	int err;
+	bool trust_local_mtime = is_wb && S_ISREG(inode->i_mode);
 
 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 		attr->ia_valid |= ATTR_FORCE;
@@ -1610,7 +1682,7 @@ int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 
 	memset(&inarg, 0, sizeof(inarg));
 	memset(&outarg, 0, sizeof(outarg));
-	iattr_to_fattr(attr, &inarg);
+	iattr_to_fattr(attr, &inarg, trust_local_mtime);
 	if (file) {
 		struct fuse_file *ff = file->private_data;
 		inarg.valid |= FATTR_FH;
@@ -1621,17 +1693,7 @@ int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 		inarg.valid |= FATTR_LOCKOWNER;
 		inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
 	}
-	req->in.h.opcode = FUSE_SETATTR;
-	req->in.h.nodeid = get_node_id(inode);
-	req->in.numargs = 1;
-	req->in.args[0].size = sizeof(inarg);
-	req->in.args[0].value = &inarg;
-	req->out.numargs = 1;
-	if (fc->minor < 9)
-		req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
-	else
-		req->out.args[0].size = sizeof(outarg);
-	req->out.args[0].value = &outarg;
+	fuse_setattr_fill(fc, req, inode, &inarg, &outarg);
 	fuse_request_send(fc, req);
 	err = req->out.h.error;
 	fuse_put_request(fc, req);
@@ -1648,10 +1710,18 @@ int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 	}
 
 	spin_lock(&fc->lock);
+	/* the kernel maintains i_mtime locally */
+	if (trust_local_mtime && (attr->ia_valid & ATTR_MTIME)) {
+		inode->i_mtime = attr->ia_mtime;
+		clear_bit(FUSE_I_MTIME_DIRTY, &fi->state);
+	}
+
 	fuse_change_attributes_common(inode, &outarg.attr,
 				      attr_timeout(&outarg));
 	oldsize = inode->i_size;
-	i_size_write(inode, outarg.attr.size);
+	/* see the comment in fuse_change_attributes() */
+	if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
+		i_size_write(inode, outarg.attr.size);
 
 	if (is_truncate) {
 		/* NOTE: this may release/reacquire fc->lock */
@@ -1663,7 +1733,8 @@ int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 	 * Only call invalidate_inode_pages2() after removing
 	 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
 	 */
-	if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
+	if ((is_truncate || !is_wb) &&
+	    S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
 		truncate_pagecache(inode, outarg.attr.size);
 		invalidate_inode_pages2(inode->i_mapping);
 	}
@@ -1875,6 +1946,17 @@ static int fuse_removexattr(struct dentry *entry, const char *name)
 	return err;
 }
 
+static int fuse_update_time(struct inode *inode, struct timespec *now,
+			    int flags)
+{
+	if (flags & S_MTIME) {
+		inode->i_mtime = *now;
+		set_bit(FUSE_I_MTIME_DIRTY, &get_fuse_inode(inode)->state);
+		BUG_ON(!S_ISREG(inode->i_mode));
+	}
+	return 0;
+}
+
 static const struct inode_operations fuse_dir_inode_operations = {
 	.lookup		= fuse_lookup,
 	.mkdir		= fuse_mkdir,
@@ -1914,6 +1996,7 @@ static const struct inode_operations fuse_common_inode_operations = {
 	.getxattr	= fuse_getxattr,
 	.listxattr	= fuse_listxattr,
 	.removexattr	= fuse_removexattr,
+	.update_time	= fuse_update_time,
 };
 
 static const struct inode_operations fuse_symlink_inode_operations = {
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 77bcc303c3a..65df7d8be4f 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -188,6 +188,22 @@ int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 }
 EXPORT_SYMBOL_GPL(fuse_do_open);
 
+static void fuse_link_write_file(struct file *file)
+{
+	struct inode *inode = file_inode(file);
+	struct fuse_conn *fc = get_fuse_conn(inode);
+	struct fuse_inode *fi = get_fuse_inode(inode);
+	struct fuse_file *ff = file->private_data;
+	/*
+	 * file may be written through mmap, so chain it onto the
+	 * inodes's write_file list
+	 */
+	spin_lock(&fc->lock);
+	if (list_empty(&ff->write_entry))
+		list_add(&ff->write_entry, &fi->write_files);
+	spin_unlock(&fc->lock);
+}
+
 void fuse_finish_open(struct inode *inode, struct file *file)
 {
 	struct fuse_file *ff = file->private_data;
@@ -208,6 +224,8 @@ void fuse_finish_open(struct inode *inode, struct file *file)
 		spin_unlock(&fc->lock);
 		fuse_invalidate_attr(inode);
 	}
+	if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
+		fuse_link_write_file(file);
 }
 
 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
@@ -292,6 +310,15 @@ static int fuse_open(struct inode *inode, struct file *file)
 
 static int fuse_release(struct inode *inode, struct file *file)
 {
+	struct fuse_conn *fc = get_fuse_conn(inode);
+
+	/* see fuse_vma_close() for !writeback_cache case */
+	if (fc->writeback_cache)
+		filemap_write_and_wait(file->f_mapping);
+
+	if (test_bit(FUSE_I_MTIME_DIRTY, &get_fuse_inode(inode)->state))
+		fuse_flush_mtime(file, true);
+
 	fuse_release_common(file, FUSE_RELEASE);
 
 	/* return value is ignored by VFS */
@@ -333,12 +360,13 @@ 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);
@@ -351,8 +379,8 @@ 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;
 		}
@@ -362,6 +390,11 @@ 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.
  *
@@ -376,6 +409,21 @@ static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
 	return 0;
 }
 
+/*
+ * Wait for all pending writepages on the inode to finish.
+ *
+ * This is currently done by blocking further writes with FUSE_NOWRITE
+ * and waiting for all sent writes to complete.
+ *
+ * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
+ * could conflict with truncation.
+ */
+static void fuse_sync_writes(struct inode *inode)
+{
+	fuse_set_nowrite(inode);
+	fuse_release_nowrite(inode);
+}
+
 static int fuse_flush(struct file *file, fl_owner_t id)
 {
 	struct inode *inode = file_inode(file);
@@ -391,6 +439,14 @@ static int fuse_flush(struct file *file, fl_owner_t id)
 	if (fc->no_flush)
 		return 0;
 
+	err = filemap_write_and_wait(file->f_mapping);
+	if (err)
+		return err;
+
+	mutex_lock(&inode->i_mutex);
+	fuse_sync_writes(inode);
+	mutex_unlock(&inode->i_mutex);
+
 	req = fuse_get_req_nofail_nopages(fc, file);
 	memset(&inarg, 0, sizeof(inarg));
 	inarg.fh = ff->fh;
@@ -411,21 +467,6 @@ static int fuse_flush(struct file *file, fl_owner_t id)
 	return err;
 }
 
-/*
- * Wait for all pending writepages on the inode to finish.
- *
- * This is currently done by blocking further writes with FUSE_NOWRITE
- * and waiting for all sent writes to complete.
- *
- * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
- * could conflict with truncation.
- */
-static void fuse_sync_writes(struct inode *inode)
-{
-	fuse_set_nowrite(inode);
-	fuse_release_nowrite(inode);
-}
-
 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
 		      int datasync, int isdir)
 {
@@ -459,6 +500,12 @@ int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
 
 	fuse_sync_writes(inode);
 
+	if (test_bit(FUSE_I_MTIME_DIRTY, &get_fuse_inode(inode)->state)) {
+		int err = fuse_flush_mtime(file, false);
+		if (err)
+			goto out;
+	}
+
 	req = fuse_get_req_nopages(fc);
 	if (IS_ERR(req)) {
 		err = PTR_ERR(req);
@@ -655,7 +702,33 @@ static void fuse_read_update_size(struct inode *inode, loff_t size,
 	spin_unlock(&fc->lock);
 }
 
-static int fuse_readpage(struct file *file, struct page *page)
+static void fuse_short_read(struct fuse_req *req, struct inode *inode,
+			    u64 attr_ver)
+{
+	size_t num_read = req->out.args[0].size;
+	struct fuse_conn *fc = get_fuse_conn(inode);
+
+	if (fc->writeback_cache) {
+		/*
+		 * A hole in a file. Some data after the hole are in page cache,
+		 * but have not reached the client fs yet. So, the hole is not
+		 * present there.
+		 */
+		int i;
+		int start_idx = num_read >> PAGE_CACHE_SHIFT;
+		size_t off = num_read & (PAGE_CACHE_SIZE - 1);
+
+		for (i = start_idx; i < req->num_pages; i++) {
+			zero_user_segment(req->pages[i], off, PAGE_CACHE_SIZE);
+			off = 0;
+		}
+	} else {
+		loff_t pos = page_offset(req->pages[0]) + num_read;
+		fuse_read_update_size(inode, pos, attr_ver);
+	}
+}
+
+static int fuse_do_readpage(struct file *file, struct page *page)
 {
 	struct fuse_io_priv io = { .async = 0, .file = file };
 	struct inode *inode = page->mapping->host;
@@ -667,10 +740,6 @@ static int fuse_readpage(struct file *file, struct page *page)
 	u64 attr_ver;
 	int err;
 
-	err = -EIO;
-	if (is_bad_inode(inode))
-		goto out;
-
 	/*
 	 * Page writeback can extend beyond the lifetime of the
 	 * page-cache page, so make sure we read a properly synced
@@ -679,9 +748,8 @@ static int fuse_readpage(struct file *file, struct page *page)
 	fuse_wait_on_page_writeback(inode, page->index);
 
 	req = fuse_get_req(fc, 1);
-	err = PTR_ERR(req);
 	if (IS_ERR(req))
-		goto out;
+		return PTR_ERR(req);
 
 	attr_ver = fuse_get_attr_version(fc);
 
@@ -692,18 +760,32 @@ static int fuse_readpage(struct file *file, struct page *page)
 	req->page_descs[0].length = count;
 	num_read = fuse_send_read(req, &io, pos, count, NULL);
 	err = req->out.h.error;
-	fuse_put_request(fc, req);
 
 	if (!err) {
 		/*
 		 * Short read means EOF.  If file size is larger, truncate it
 		 */
 		if (num_read < count)
-			fuse_read_update_size(inode, pos + num_read, attr_ver);
+			fuse_short_read(req, inode, attr_ver);
 
 		SetPageUptodate(page);
 	}
 
+	fuse_put_request(fc, req);
+
+	return err;
+}
+
+static int fuse_readpage(struct file *file, struct page *page)
+{
+	struct inode *inode = page->mapping->host;
+	int err;
+
+	err = -EIO;
+	if (is_bad_inode(inode))
+		goto out;
+
+	err = fuse_do_readpage(file, page);
 	fuse_invalidate_atime(inode);
  out:
 	unlock_page(page);
@@ -726,13 +808,9 @@ static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
 		/*
 		 * Short read means EOF. If file size is larger, truncate it
 		 */
-		if (!req->out.h.error && num_read < count) {
-			loff_t pos;
+		if (!req->out.h.error && num_read < count)
+			fuse_short_read(req, inode, req->misc.read.attr_ver);
 
-			pos = page_offset(req->pages[0]) + num_read;
-			fuse_read_update_size(inode, pos,
-					      req->misc.read.attr_ver);
-		}
 		fuse_invalidate_atime(inode);
 	}
 
@@ -922,16 +1000,21 @@ static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
 	return req->misc.write.out.size;
 }
 
-void fuse_write_update_size(struct inode *inode, loff_t pos)
+bool fuse_write_update_size(struct inode *inode, loff_t pos)
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
+	bool ret = false;
 
 	spin_lock(&fc->lock);
 	fi->attr_version = ++fc->attr_version;
-	if (pos > inode->i_size)
+	if (pos > inode->i_size) {
 		i_size_write(inode, pos);
+		ret = true;
+	}
 	spin_unlock(&fc->lock);
+
+	return ret;
 }
 
 static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
@@ -1116,6 +1199,15 @@ static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 	struct iov_iter i;
 	loff_t endbyte = 0;
 
+	if (get_fuse_conn(inode)->writeback_cache) {
+		/* Update size (EOF optimization) and mode (SUID clearing) */
+		err = fuse_update_attributes(mapping->host, NULL, file, NULL);
+		if (err)
+			return err;
+
+		return generic_file_aio_write(iocb, iov, nr_segs, pos);
+	}
+
 	WARN_ON(iocb->ki_pos != pos);
 
 	ocount = 0;
@@ -1289,13 +1381,18 @@ 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;
@@ -1309,6 +1406,14 @@ 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;
@@ -1397,7 +1502,8 @@ 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);
 
@@ -1885,6 +1991,77 @@ out:
 	return err;
 }
 
+/*
+ * It's worthy to make sure that space is reserved on disk for the write,
+ * but how to implement it without killing performance need more thinking.
+ */
+static int fuse_write_begin(struct file *file, struct address_space *mapping,
+		loff_t pos, unsigned len, unsigned flags,
+		struct page **pagep, void **fsdata)
+{
+	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
+	struct fuse_conn *fc = get_fuse_conn(file->f_dentry->d_inode);
+	struct page *page;
+	loff_t fsize;
+	int err = -ENOMEM;
+
+	WARN_ON(!fc->writeback_cache);
+
+	page = grab_cache_page_write_begin(mapping, index, flags);
+	if (!page)
+		goto error;
+
+	fuse_wait_on_page_writeback(mapping->host, page->index);
+
+	if (PageUptodate(page) || len == PAGE_CACHE_SIZE)
+		goto success;
+	/*
+	 * Check if the start this page comes after the end of file, in which
+	 * case the readpage can be optimized away.
+	 */
+	fsize = i_size_read(mapping->host);
+	if (fsize <= (pos & PAGE_CACHE_MASK)) {
+		size_t off = pos & ~PAGE_CACHE_MASK;
+		if (off)
+			zero_user_segment(page, 0, off);
+		goto success;
+	}
+	err = fuse_do_readpage(file, page);
+	if (err)
+		goto cleanup;
+success:
+	*pagep = page;
+	return 0;
+
+cleanup:
+	unlock_page(page);
+	page_cache_release(page);
+error:
+	return err;
+}
+
+static int fuse_write_end(struct file *file, struct address_space *mapping,
+		loff_t pos, unsigned len, unsigned copied,
+		struct page *page, void *fsdata)
+{
+	struct inode *inode = page->mapping->host;
+
+	if (!PageUptodate(page)) {
+		/* Zero any unwritten bytes at the end of the page */
+		size_t endoff = (pos + copied) & ~PAGE_CACHE_MASK;
+		if (endoff)
+			zero_user_segment(page, endoff, PAGE_CACHE_SIZE);
+		SetPageUptodate(page);
+	}
+
+	fuse_write_update_size(inode, pos + copied);
+	set_page_dirty(page);
+	unlock_page(page);
+	page_cache_release(page);
+
+	return copied;
+}
+
 static int fuse_launder_page(struct page *page)
 {
 	int err = 0;
@@ -1946,20 +2123,9 @@ static const struct vm_operations_struct fuse_file_vm_ops = {
 
 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
 {
-	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
-		struct inode *inode = file_inode(file);
-		struct fuse_conn *fc = get_fuse_conn(inode);
-		struct fuse_inode *fi = get_fuse_inode(inode);
-		struct fuse_file *ff = file->private_data;
-		/*
-		 * file may be written through mmap, so chain it onto the
-		 * inodes's write_file list
-		 */
-		spin_lock(&fc->lock);
-		if (list_empty(&ff->write_entry))
-			list_add(&ff->write_entry, &fi->write_files);
-		spin_unlock(&fc->lock);
-	}
+	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
+		fuse_link_write_file(file);
+
 	file_accessed(file);
 	vma->vm_ops = &fuse_file_vm_ops;
 	return 0;
@@ -2606,7 +2772,7 @@ static void fuse_register_polled_file(struct fuse_conn *fc,
 {
 	spin_lock(&fc->lock);
 	if (RB_EMPTY_NODE(&ff->polled_node)) {
-		struct rb_node **link, *parent;
+		struct rb_node **link, *uninitialized_var(parent);
 
 		link = fuse_find_polled_node(fc, ff->kh, &parent);
 		BUG_ON(*link);
@@ -2850,8 +3016,16 @@ static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
 		goto out;
 
 	/* we could have extended the file */
-	if (!(mode & FALLOC_FL_KEEP_SIZE))
-		fuse_write_update_size(inode, offset + length);
+	if (!(mode & FALLOC_FL_KEEP_SIZE)) {
+		bool changed = fuse_write_update_size(inode, offset + length);
+
+		if (changed && fc->writeback_cache) {
+			struct fuse_inode *fi = get_fuse_inode(inode);
+
+			inode->i_mtime = current_fs_time(inode->i_sb);
+			set_bit(FUSE_I_MTIME_DIRTY, &fi->state);
+		}
+	}
 
 	if (mode & FALLOC_FL_PUNCH_HOLE)
 		truncate_pagecache_range(inode, offset, offset + length - 1);
@@ -2915,6 +3089,8 @@ static const struct address_space_operations fuse_file_aops  = {
 	.set_page_dirty	= __set_page_dirty_nobuffers,
 	.bmap		= fuse_bmap,
 	.direct_IO	= fuse_direct_IO,
+	.write_begin	= fuse_write_begin,
+	.write_end	= fuse_write_end,
 };
 
 void fuse_init_file_inode(struct inode *inode)
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 2da5db2c8bd..a257ed8ebee 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -119,6 +119,8 @@ enum {
 	FUSE_I_INIT_RDPLUS,
 	/** An operation changing file size is in progress  */
 	FUSE_I_SIZE_UNSTABLE,
+	/** i_mtime has been updated locally; a flush to userspace needed */
+	FUSE_I_MTIME_DIRTY,
 };
 
 struct fuse_conn;
@@ -480,6 +482,9 @@ struct fuse_conn {
 	/** Set if bdi is valid */
 	unsigned bdi_initialized:1;
 
+	/** write-back cache policy (default is write-through) */
+	unsigned writeback_cache:1;
+
 	/*
 	 * The following bitfields are only for optimization purposes
 	 * and hence races in setting them will not cause malfunction
@@ -863,9 +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,
@@ -873,7 +889,9 @@ long fuse_ioctl_common(struct file *file, unsigned int cmd,
 unsigned fuse_file_poll(struct file *file, poll_table *wait);
 int fuse_dev_release(struct inode *inode, struct file *file);
 
-void fuse_write_update_size(struct inode *inode, loff_t pos);
+bool fuse_write_update_size(struct inode *inode, loff_t pos);
+
+int fuse_flush_mtime(struct file *file, bool nofail);
 
 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
 		    struct file *file);
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 9c761b611c5..b4bff1b1502 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -170,8 +170,11 @@ void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 	inode->i_blocks  = attr->blocks;
 	inode->i_atime.tv_sec   = attr->atime;
 	inode->i_atime.tv_nsec  = attr->atimensec;
-	inode->i_mtime.tv_sec   = attr->mtime;
-	inode->i_mtime.tv_nsec  = attr->mtimensec;
+	/* mtime from server may be stale due to local buffered write */
+	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
+		inode->i_mtime.tv_sec   = attr->mtime;
+		inode->i_mtime.tv_nsec  = attr->mtimensec;
+	}
 	inode->i_ctime.tv_sec   = attr->ctime;
 	inode->i_ctime.tv_nsec  = attr->ctimensec;
 
@@ -197,6 +200,7 @@ void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
+	bool is_wb = fc->writeback_cache;
 	loff_t oldsize;
 	struct timespec old_mtime;
 
@@ -211,10 +215,16 @@ void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 	fuse_change_attributes_common(inode, attr, attr_valid);
 
 	oldsize = inode->i_size;
-	i_size_write(inode, attr->size);
+	/*
+	 * In case of writeback_cache enabled, the cached writes beyond EOF
+	 * extend local i_size without keeping userspace server in sync. So,
+	 * attr->size coming from server can be stale. We cannot trust it.
+	 */
+	if (!is_wb || !S_ISREG(inode->i_mode))
+		i_size_write(inode, attr->size);
 	spin_unlock(&fc->lock);
 
-	if (S_ISREG(inode->i_mode)) {
+	if (!is_wb && S_ISREG(inode->i_mode)) {
 		bool inval = false;
 
 		if (oldsize != attr->size) {
@@ -243,6 +253,8 @@ static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 {
 	inode->i_mode = attr->mode & S_IFMT;
 	inode->i_size = attr->size;
+	inode->i_mtime.tv_sec  = attr->mtime;
+	inode->i_mtime.tv_nsec = attr->mtimensec;
 	if (S_ISREG(inode->i_mode)) {
 		fuse_init_common(inode);
 		fuse_init_file_inode(inode);
@@ -289,7 +301,9 @@ struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 		return NULL;
 
 	if ((inode->i_state & I_NEW)) {
-		inode->i_flags |= S_NOATIME|S_NOCMTIME;
+		inode->i_flags |= S_NOATIME;
+		if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
+			inode->i_flags |= S_NOCMTIME;
 		inode->i_generation = generation;
 		inode->i_data.backing_dev_info = &fc->bdi;
 		fuse_init_inode(inode, attr);
@@ -873,6 +887,8 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 			}
 			if (arg->flags & FUSE_ASYNC_DIO)
 				fc->async_dio = 1;
+			if (arg->flags & FUSE_WRITEBACK_CACHE)
+				fc->writeback_cache = 1;
 		} else {
 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 			fc->no_lock = 1;
@@ -900,7 +916,8 @@ static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
-		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO;
+		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
+		FUSE_WRITEBACK_CACHE;
 	req->in.h.opcode = FUSE_INIT;
 	req->in.numargs = 1;
 	req->in.args[0].size = sizeof(*arg);
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index 60bb2f9f7b7..cf4750e1bb4 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -93,6 +93,9 @@
  *
  * 7.22
  *  - add FUSE_ASYNC_DIO
+ *
+ * 7.23
+ *  - add FUSE_WRITEBACK_CACHE
  */
 
 #ifndef _LINUX_FUSE_H
@@ -128,7 +131,7 @@
 #define FUSE_KERNEL_VERSION 7
 
 /** Minor version number of this interface */
-#define FUSE_KERNEL_MINOR_VERSION 22
+#define FUSE_KERNEL_MINOR_VERSION 23
 
 /** The node ID of the root inode */
 #define FUSE_ROOT_ID 1
@@ -219,6 +222,7 @@ struct fuse_file_lock {
  * FUSE_DO_READDIRPLUS: do READDIRPLUS (READDIR+LOOKUP in one)
  * FUSE_READDIRPLUS_AUTO: adaptive readdirplus
  * FUSE_ASYNC_DIO: asynchronous direct I/O submission
+ * FUSE_WRITEBACK_CACHE: use writeback cache for buffered writes
  */
 #define FUSE_ASYNC_READ		(1 << 0)
 #define FUSE_POSIX_LOCKS	(1 << 1)
@@ -236,6 +240,7 @@ struct fuse_file_lock {
 #define FUSE_DO_READDIRPLUS	(1 << 13)
 #define FUSE_READDIRPLUS_AUTO	(1 << 14)
 #define FUSE_ASYNC_DIO		(1 << 15)
+#define FUSE_WRITEBACK_CACHE	(1 << 16)
 
 /**
  * CUSE INIT request/reply flags