btrfs: fix the memory leak caused in lzo_compress_pages()

[BUG]
Fstests generic/027 is pretty easy to trigger a slow but steady memory
leak if run with "-o compress=lzo" mount option.

Normally one single run of generic/027 is enough to eat up at least 4G ram.

[CAUSE]
In commit d408880 ("btrfs: subpage: make lzo_compress_pages()
compatible") we changed how @page_in is released.

But that refactoring makes @page_in only released after all pages being
compressed.

This leaves error path not releasing @page_in. And by "error path"
things like incompressible data will also be treated as an error
(-E2BIG).

Thus it can cause a memory leak if even nothing wrong happened.

[FIX]
Add check under @out label to release @page_in when needed, so when we
hit any error, the input page is properly released.

Reported-by: Josef Bacik <josef@toxicpanda.com>
Fixes: d408880 ("btrfs: subpage: make lzo_compress_pages() compatible")
Reviewed-and-tested-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/lzo.c

@@ -276,6 +276,8 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
 	*total_out = cur_out;
 	*total_in = cur_in - start;
 out:
+	if (page_in)
+		put_page(page_in);
 	*out_pages = DIV_ROUND_UP(cur_out, PAGE_SIZE);
 	return ret;
 }