Revert "nfs: don't reuse partially completed requests in nfs_lock_and_join_requests"

[ Upstream commit 66f9dac ]

This reverts commit b571cfc.

This patch appears to assume that if one request is complete, then the
others will complete too before unlocking. That is not a valid
assumption, since other requests could hit a non-fatal error or a short
write that would cause them not to complete.

Reported-by: Igor Raits <igor@gooddata.com>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=219508
Fixes: b571cfc ("nfs: don't reuse partially completed requests in nfs_lock_and_join_requests")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

Author: Trond Myklebust

fs/nfs/write.c

@@ -144,6 +144,31 @@ static void nfs_io_completion_put(struct nfs_io_completion *ioc)
 		kref_put(&ioc->refcount, nfs_io_completion_release);
 }
 
+static void
+nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode)
+{
+	if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) {
+		kref_get(&req->wb_kref);
+		atomic_long_inc(&NFS_I(inode)->nrequests);
+	}
+}
+
+static int
+nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode)
+{
+	int ret;
+
+	if (!test_bit(PG_REMOVE, &req->wb_flags))
+		return 0;
+	ret = nfs_page_group_lock(req);
+	if (ret)
+		return ret;
+	if (test_and_clear_bit(PG_REMOVE, &req->wb_flags))
+		nfs_page_set_inode_ref(req, inode);
+	nfs_page_group_unlock(req);
+	return 0;
+}
+
 /**
  * nfs_folio_find_head_request - find head request associated with a folio
  * @folio: pointer to folio
@@ -540,7 +565,6 @@ static struct nfs_page *nfs_lock_and_join_requests(struct folio *folio)
 	struct inode *inode = folio->mapping->host;
 	struct nfs_page *head, *subreq;
 	struct nfs_commit_info cinfo;
-	bool removed;
 	int ret;
 
 	/*
@@ -565,40 +589,25 @@ static struct nfs_page *nfs_lock_and_join_requests(struct folio *folio)
 		goto retry;
 	}
 
-	ret = nfs_page_group_lock(head);
+	ret = nfs_cancel_remove_inode(head, inode);
 	if (ret < 0)
 		goto out_unlock;
 
-	removed = test_bit(PG_REMOVE, &head->wb_flags);
+	ret = nfs_page_group_lock(head);
+	if (ret < 0)
+		goto out_unlock;
 
 	/* lock each request in the page group */
 	for (subreq = head->wb_this_page;
 	     subreq != head;
 	     subreq = subreq->wb_this_page) {
-		if (test_bit(PG_REMOVE, &subreq->wb_flags))
-			removed = true;
 		ret = nfs_page_group_lock_subreq(head, subreq);
 		if (ret < 0)
 			goto out_unlock;
 	}
 
 	nfs_page_group_unlock(head);
 
-	/*
-	 * If PG_REMOVE is set on any request, I/O on that request has
-	 * completed, but some requests were still under I/O at the time
-	 * we locked the head request.
-	 *
-	 * In that case the above wait for all requests means that all I/O
-	 * has now finished, and we can restart from a clean slate.  Let the
-	 * old requests go away and start from scratch instead.
-	 */
-	if (removed) {
-		nfs_unroll_locks(head, head);
-		nfs_unlock_and_release_request(head);
-		goto retry;
-	}
-
 	nfs_init_cinfo_from_inode(&cinfo, inode);
 	nfs_join_page_group(head, &cinfo, inode);
 	return head;