fuse: Introduce fi->lock to protect write related fields

To minimize contention of fc->lock, this patch introduces a new spinlock
for protection fuse_inode metadata:

fuse_inode:
	writectr
	writepages
	write_files
	queued_writes
	attr_version

inode:
	i_size
	i_nlink
	i_mtime
	i_ctime

Also, it protects the fields changed in fuse_change_attributes_common()
(too many to list).
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fs/fuse/dir.c

@@ -658,7 +658,7 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
 		struct inode *inode = d_inode(entry);
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
+		spin_lock(&fi->lock);
 		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		/*
 		 * If i_nlink == 0 then unlink doesn't make sense, yet this can
@@ -668,7 +668,7 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
 		 */
 		if (inode->i_nlink > 0)
 			drop_nlink(inode);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		fuse_dir_changed(dir);
 		fuse_invalidate_entry_cache(entry);
@@ -812,10 +812,10 @@ static int fuse_link(struct dentry *entry, struct inode *newdir,
 	if (!err) {
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
+		spin_lock(&fi->lock);
 		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		inc_nlink(inode);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		fuse_update_ctime(inode);
 	} else if (err == -EINTR) {
@@ -1343,15 +1343,14 @@ static void iattr_to_fattr(struct fuse_conn *fc, struct iattr *iattr,
  */
 void fuse_set_nowrite(struct inode *inode)
 {
-	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	BUG_ON(!inode_is_locked(inode));
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	BUG_ON(fi->writectr < 0);
 	fi->writectr += FUSE_NOWRITE;
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 	wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
 }
 
@@ -1372,11 +1371,11 @@ static void __fuse_release_nowrite(struct inode *inode)
 
 void fuse_release_nowrite(struct inode *inode)
 {
-	struct fuse_conn *fc = get_fuse_conn(inode);
+	struct fuse_inode *fi = get_fuse_inode(inode);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	__fuse_release_nowrite(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
@@ -1511,7 +1510,7 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 		goto error;
 	}
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	/* the kernel maintains i_mtime locally */
 	if (trust_local_cmtime) {
 		if (attr->ia_valid & ATTR_MTIME)
@@ -1529,10 +1528,10 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 		i_size_write(inode, outarg.attr.size);
 
 	if (is_truncate) {
-		/* NOTE: this may release/reacquire fc->lock */
+		/* NOTE: this may release/reacquire fi->lock */
 		__fuse_release_nowrite(inode);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	/*
 	 * Only call invalidate_inode_pages2() after removing

fs/fuse/file.c

@@ -159,17 +159,16 @@ 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);
+	spin_lock(&fi->lock);
 	if (list_empty(&ff->write_entry))
 		list_add(&ff->write_entry, &fi->write_files);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 void fuse_finish_open(struct inode *inode, struct file *file)
@@ -186,10 +185,10 @@ void fuse_finish_open(struct inode *inode, struct file *file)
 	if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
+		spin_lock(&fi->lock);
 		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		i_size_write(inode, 0);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		if (fc->writeback_cache)
 			file_update_time(file);
@@ -231,8 +230,13 @@ static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
 	struct fuse_req *req = ff->reserved_req;
 	struct fuse_release_in *inarg = &req->misc.release.in;
 
+	/* Inode is NULL on error path of fuse_create_open() */
+	if (likely(fi)) {
+		spin_lock(&fi->lock);
+		list_del(&ff->write_entry);
+		spin_unlock(&fi->lock);
+	}
 	spin_lock(&fc->lock);
-	list_del(&ff->write_entry);
 	if (!RB_EMPTY_NODE(&ff->polled_node))
 		rb_erase(&ff->polled_node, &fc->polled_files);
 	spin_unlock(&fc->lock);
@@ -358,13 +362,12 @@ static struct fuse_req *fuse_find_writeback(struct fuse_inode *fi,
 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);
 	bool found;
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	found = fuse_find_writeback(fi, idx_from, idx_to);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	return found;
 }
@@ -607,9 +610,9 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
 			struct fuse_conn *fc = get_fuse_conn(inode);
 			struct fuse_inode *fi = get_fuse_inode(inode);
 
-			spin_lock(&fc->lock);
+			spin_lock(&fi->lock);
 			fi->attr_version = atomic64_inc_return(&fc->attr_version);
-			spin_unlock(&fc->lock);
+			spin_unlock(&fi->lock);
 		}
 
 		io->iocb->ki_complete(io->iocb, res, 0);
@@ -684,13 +687,13 @@ static void fuse_read_update_size(struct inode *inode, loff_t size,
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if (attr_ver == fi->attr_version && size < inode->i_size &&
 	    !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		i_size_write(inode, size);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 static void fuse_short_read(struct fuse_req *req, struct inode *inode,
@@ -1005,13 +1008,13 @@ bool fuse_write_update_size(struct inode *inode, loff_t pos)
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	bool ret = false;
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
 	if (pos > inode->i_size) {
 		i_size_write(inode, pos);
 		ret = true;
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	return ret;
 }
@@ -1490,11 +1493,11 @@ static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
 	wake_up(&fi->page_waitq);
 }
 
-/* Called under fc->lock, may release and reacquire it */
+/* Called under fi->lock, may release and reacquire it */
 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
 				loff_t size)
-__releases(fc->lock)
-__acquires(fc->lock)
+__releases(fi->lock)
+__acquires(fi->lock)
 {
 	struct fuse_req *aux, *next;
 	struct fuse_inode *fi = get_fuse_inode(req->inode);
@@ -1502,9 +1505,6 @@ __acquires(fc->lock)
 	__u64 data_size = req->num_pages * PAGE_SIZE;
 	bool queued;
 
-	if (!fc->connected)
-		goto out_free;
-
 	if (inarg->offset + data_size <= size) {
 		inarg->size = data_size;
 	} else if (inarg->offset < size) {
@@ -1515,14 +1515,17 @@ __acquires(fc->lock)
 	}
 
 	req->in.args[1].size = inarg->size;
-	fi->writectr++;
 	queued = fuse_request_queue_background(fc, req);
-	WARN_ON(!queued);
+	/* Fails on broken connection only */
+	if (unlikely(!queued))
+		goto out_free;
+
+	fi->writectr++;
 	return;
 
  out_free:
 	fuse_writepage_finish(fc, req);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	/* After fuse_writepage_finish() aux request list is private */
 	for (aux = req->misc.write.next; aux; aux = next) {
@@ -1534,18 +1537,18 @@ __acquires(fc->lock)
 
 	fuse_writepage_free(fc, req);
 	fuse_put_request(fc, req);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 }
 
 /*
  * If fi->writectr is positive (no truncate or fsync going on) send
  * all queued writepage requests.
  *
- * Called with fc->lock
+ * Called with fi->lock
  */
 void fuse_flush_writepages(struct inode *inode)
-__releases(fc->lock)
-__acquires(fc->lock)
+__releases(fi->lock)
+__acquires(fi->lock)
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
@@ -1565,7 +1568,7 @@ static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	mapping_set_error(inode->i_mapping, req->out.h.error);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	while (req->misc.write.next) {
 		struct fuse_conn *fc = get_fuse_conn(inode);
 		struct fuse_write_in *inarg = &req->misc.write.in;
@@ -1602,7 +1605,7 @@ static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
 	}
 	fi->writectr--;
 	fuse_writepage_finish(fc, req);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 	fuse_writepage_free(fc, req);
 }
 
@@ -1611,13 +1614,13 @@ static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
 {
 	struct fuse_file *ff = NULL;
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if (!list_empty(&fi->write_files)) {
 		ff = list_entry(fi->write_files.next, struct fuse_file,
 				write_entry);
 		fuse_file_get(ff);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	return ff;
 }
@@ -1688,11 +1691,11 @@ static int fuse_writepage_locked(struct page *page)
 	inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
 	inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	list_add(&req->writepages_entry, &fi->writepages);
 	list_add_tail(&req->list, &fi->queued_writes);
 	fuse_flush_writepages(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	end_page_writeback(page);
 
@@ -1741,16 +1744,15 @@ static void fuse_writepages_send(struct fuse_fill_wb_data *data)
 {
 	struct fuse_req *req = data->req;
 	struct inode *inode = data->inode;
-	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	int num_pages = req->num_pages;
 	int i;
 
 	req->ff = fuse_file_get(data->ff);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	list_add_tail(&req->list, &fi->queued_writes);
 	fuse_flush_writepages(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	for (i = 0; i < num_pages; i++)
 		end_page_writeback(data->orig_pages[i]);
@@ -1773,12 +1775,12 @@ static bool fuse_writepage_in_flight(struct fuse_req *new_req,
 
 	WARN_ON(new_req->num_pages != 0);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	list_del(&new_req->writepages_entry);
 	old_req = fuse_find_writeback(fi, page->index, page->index);
 	if (!old_req) {
 		list_add(&new_req->writepages_entry, &fi->writepages);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		return false;
 	}
 
@@ -1801,7 +1803,7 @@ static bool fuse_writepage_in_flight(struct fuse_req *new_req,
 		old_req->misc.write.next = new_req;
 	}
 
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	if (tmp) {
 		struct backing_dev_info *bdi = inode_to_bdi(new_req->inode);
@@ -1822,6 +1824,7 @@ static int fuse_writepages_fill(struct page *page,
 	struct fuse_fill_wb_data *data = _data;
 	struct fuse_req *req = data->req;
 	struct inode *inode = data->inode;
+	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct page *tmp_page;
 	bool is_writeback;
@@ -1892,9 +1895,9 @@ static int fuse_writepages_fill(struct page *page,
 		req->end = fuse_writepage_end;
 		req->inode = inode;
 
-		spin_lock(&fc->lock);
+		spin_lock(&fi->lock);
 		list_add(&req->writepages_entry, &fi->writepages);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 
 		data->req = req;
 	}
@@ -1917,12 +1920,12 @@ static int fuse_writepages_fill(struct page *page,
 	data->orig_pages[req->num_pages] = page;
 
 	/*
-	 * Protected by fc->lock against concurrent access by
+	 * Protected by fi->lock against concurrent access by
 	 * fuse_page_is_writeback().
 	 */
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	req->num_pages++;
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 out_unlock:
 	unlock_page(page);

fs/fuse/fuse_i.h

@@ -96,7 +96,7 @@ struct fuse_inode {
 	union {
 		/* Write related fields (regular file only) */
 		struct {
-			/* Files usable in writepage.  Protected by fc->lock */
+			/* Files usable in writepage.  Protected by fi->lock */
 			struct list_head write_files;
 
 			/* Writepages pending on truncate or fsync */
@@ -144,6 +144,9 @@ struct fuse_inode {
 
 	/** Lock for serializing lookup and readdir for back compatibility*/
 	struct mutex mutex;
+
+	/** Lock to protect write related fields */
+	spinlock_t lock;
 };
 
 /** FUSE inode state bits */

fs/fuse/inode.c

@@ -97,6 +97,7 @@ static struct inode *fuse_alloc_inode(struct super_block *sb)
 	fi->orig_ino = 0;
 	fi->state = 0;
 	mutex_init(&fi->mutex);
+	spin_lock_init(&fi->lock);
 	fi->forget = fuse_alloc_forget();
 	if (!fi->forget) {
 		kmem_cache_free(fuse_inode_cachep, inode);
@@ -163,6 +164,8 @@ void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
+	lockdep_assert_held(&fi->lock);
+
 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
 	fi->i_time = attr_valid;
 	WRITE_ONCE(fi->inval_mask, 0);
@@ -209,10 +212,10 @@ void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 	loff_t oldsize;
 	struct timespec64 old_mtime;
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		return;
 	}
 
@@ -227,7 +230,7 @@ void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 	 */
 	if (!is_wb || !S_ISREG(inode->i_mode))
 		i_size_write(inode, attr->size);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	if (!is_wb && S_ISREG(inode->i_mode)) {
 		bool inval = false;