Re: PROBLEM: DFS Caching feature causing problems traversing multi-tier DFS setups

[Paulo Alcantara <pc@cjr.nz>]

Hi Matthew,

Thanks for the report.

Matthew Ruffell <matthew.ruffell@canonical.com> writes:

> We have come across a problem where kernels 5.0-rc1 and onwards cannot mount
> a multi tier DFS setup, while kernels 4.20 and below can mount the share fine.
>
> The DFS tiering structure looks like this:
>
> Domain virtual DFS (i.e. \\company.com\folders\share)
> |-- Domain controller DFS (i.e. \\regional-dc.company.com\folders\share)
>     |-- Regional DFS Server (i.e. \\regional-dfs.company.com\folders\share)
>         |-- Actual file server (i.e. \\regional-svr.company.com\share)
>
> On the 5.x series kernels, after getting the DFS referrals list through to the
> Regional DFS Server, which responds with the correct server/share, instead of
> going to the Actual file server, the kernel backtracks from the Regional DFS
> Server back to the Domain controller and requests the share there. Of course,
> this share does not exist on the Domain controller, as it only exists on the
> Actual file server, and the connection dies.

I've got some DFS cache patches[1] and haven't sent them yet due to lack
of time and testing. Those contain a lot of important fixes but none of
them seem to fix the issue you're having -- thus I won't ask you to
apply them on top.

Instead, could you please try below changes?

Thanks,
Paulo

[1] https://git.cjr.nz/linux.git/log/?h=cifs-dfscache

diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 3991d6c8f255..9158d5d14ac9 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -4777,6 +4777,15 @@ static int is_path_remote(struct cifs_sb_info *cifs_sb, struct smb_vol *vol,
 }
 
 #ifdef CONFIG_CIFS_DFS_UPCALL
+static inline void set_root_tcon(struct cifs_tcon *tcon,
+				 struct cifs_tcon **root)
+{
+	spin_lock(&cifs_tcp_ses_lock);
+	tcon->tc_count++;
+	spin_unlock(&cifs_tcp_ses_lock);
+	*root = tcon;
+}
+
 int cifs_mount(struct cifs_sb_info *cifs_sb, struct smb_vol *vol)
 {
 	int rc = 0;
@@ -4878,18 +4887,10 @@ int cifs_mount(struct cifs_sb_info *cifs_sb, struct smb_vol *vol)
 	/* Cache out resolved root server */
 	(void)dfs_cache_find(xid, ses, cifs_sb->local_nls, cifs_remap(cifs_sb),
 			     root_path + 1, NULL, NULL);
-	/*
-	 * Save root tcon for additional DFS requests to update or create a new
-	 * DFS cache entry, or even perform DFS failover.
-	 */
-	spin_lock(&cifs_tcp_ses_lock);
-	tcon->tc_count++;
-	tcon->dfs_path = root_path;
+	kfree(root_path);
 	root_path = NULL;
-	tcon->remap = cifs_remap(cifs_sb);
-	spin_unlock(&cifs_tcp_ses_lock);
 
-	root_tcon = tcon;
+	set_root_tcon(tcon, &root_tcon);
 
 	for (count = 1; ;) {
 		if (!rc && tcon) {
@@ -4926,6 +4927,15 @@ int cifs_mount(struct cifs_sb_info *cifs_sb, struct smb_vol *vol)
 			mount_put_conns(cifs_sb, xid, server, ses, tcon);
 			rc = mount_get_conns(vol, cifs_sb, &xid, &server, &ses,
 					     &tcon);
+			/*
+			 * Ensure that DFS referrals go through new root server.
+			 */
+			if (!rc && tcon &&
+			    (tcon->share_flags & (SHI1005_FLAGS_DFS |
+						  SHI1005_FLAGS_DFS_ROOT))) {
+				cifs_put_tcon(root_tcon);
+				set_root_tcon(tcon, &root_tcon);
+			}
 		}
 		if (rc) {
 			if (rc == -EACCES || rc == -EOPNOTSUPP)