NFS: Count the bytes of skipped subrequests in nfs_lock_and_join_requests()
If we skip a subrequest due to a zero refcount, we should still count
the byte range that it covered so that we accurately reconstruct the
original request size.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 36d34a4..f68083d 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -504,8 +504,12 @@ nfs_lock_and_join_requests(struct page *page)
for (subreq = head->wb_this_page; subreq != head;
subreq = subreq->wb_this_page) {
- if (!kref_get_unless_zero(&subreq->wb_kref))
+ if (!kref_get_unless_zero(&subreq->wb_kref)) {
+ if (subreq->wb_offset == head->wb_offset + total_bytes)
+ total_bytes += subreq->wb_bytes;
continue;
+ }
+
while (!nfs_lock_request(subreq)) {
/*
* Unlock page to allow nfs_page_group_sync_on_bit()