cifs: support nested dfs links over reconnect

Mounting a dfs link that has nested links was already supported at
mount(2), so make it work over reconnect as well.

Make the following case work:

* mount //root/dfs/link /mnt -o ...
  - final share: /server/share

* in server settings
  - change target folder of /root/dfs/link3 to /server/share2
  - change target folder of /root/dfs/link2 to /root/dfs/link3
  - change target folder of /root/dfs/link to /root/dfs/link2

* mount -o remount,... /mnt
 - refresh all dfs referrals
 - mark current connection for failover
 - cifs_reconnect() reconnects to root server
 - tree_connect()
   * checks that /root/dfs/link2 is a link, then chase it
   * checks that root/dfs/link3 is a link, then chase it
   * finally tree connect to /server/share2

If the mounted share is no longer accessible and a reconnect had been
triggered, the client will retry it from both last referral
path (/root/dfs/link3) and original referral path (/root/dfs/link).

Any new referral paths found while chasing dfs links over reconnect,
it will be updated to TCP_Server_Info::leaf_fullpath, accordingly.

Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>

fs/cifs/cifs_dfs_ref.c

@@ -307,12 +307,8 @@ static struct vfsmount *cifs_dfs_do_mount(struct dentry *mntpt,
 static struct vfsmount *cifs_dfs_do_automount(struct dentry *mntpt)
 {
 	struct cifs_sb_info *cifs_sb;
-	struct cifs_ses *ses;
-	struct cifs_tcon *tcon;
 	void *page;
-	char *full_path, *root_path;
+	char *full_path;
-	unsigned int xid;
-	int rc;
 	struct vfsmount *mnt;
 
 	cifs_dbg(FYI, "in %s\n", __func__);
@@ -324,8 +320,6 @@ static struct vfsmount *cifs_dfs_do_automount(struct dentry *mntpt)
 	 * the double backslashes usually used in the UNC. This function
 	 * gives us the latter, so we must adjust the result.
 	 */
-	mnt = ERR_PTR(-ENOMEM);
-
 	cifs_sb = CIFS_SB(mntpt->d_sb);
 	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_DFS) {
 		mnt = ERR_PTR(-EREMOTE);
@@ -341,60 +335,11 @@ static struct vfsmount *cifs_dfs_do_automount(struct dentry *mntpt)
 	}
 
 	convert_delimiter(full_path, '\\');
-
 	cifs_dbg(FYI, "%s: full_path: %s\n", __func__, full_path);
 
-	if (!cifs_sb_master_tlink(cifs_sb)) {
-		cifs_dbg(FYI, "%s: master tlink is NULL\n", __func__);
-		goto free_full_path;
-	}
-
-	tcon = cifs_sb_master_tcon(cifs_sb);
-	if (!tcon) {
-		cifs_dbg(FYI, "%s: master tcon is NULL\n", __func__);
-		goto free_full_path;
-	}
-
-	root_path = kstrdup(tcon->treeName, GFP_KERNEL);
-	if (!root_path) {
-		mnt = ERR_PTR(-ENOMEM);
-		goto free_full_path;
-	}
-	cifs_dbg(FYI, "%s: root path: %s\n", __func__, root_path);
-
-	ses = tcon->ses;
-	xid = get_xid();
-
-	/*
-	 * If DFS root has been expired, then unconditionally fetch it again to
-	 * refresh DFS referral cache.
-	 */
-	rc = dfs_cache_find(xid, ses, cifs_sb->local_nls, cifs_remap(cifs_sb),
-			    root_path + 1, NULL, NULL);
-	if (!rc) {
-		rc = dfs_cache_find(xid, ses, cifs_sb->local_nls,
-				    cifs_remap(cifs_sb), full_path + 1,
-				    NULL, NULL);
-	}
-
-	free_xid(xid);
-
-	if (rc) {
-		mnt = ERR_PTR(rc);
-		goto free_root_path;
-	}
-	/*
-	 * OK - we were able to get and cache a referral for @full_path.
-	 *
-	 * Now, pass it down to cifs_mount() and it will retry every available
-	 * node server in case of failures - no need to do it here.
-	 */
 	mnt = cifs_dfs_do_mount(mntpt, cifs_sb, full_path);
-	cifs_dbg(FYI, "%s: cifs_dfs_do_mount:%s , mnt:%p\n", __func__,
+	cifs_dbg(FYI, "%s: cifs_dfs_do_mount:%s , mnt:%p\n", __func__, full_path + 1, mnt);
-		 full_path + 1, mnt);
 
-free_root_path:
-	kfree(root_path);
 free_full_path:
 	free_dentry_path(page);
 cdda_exit:

fs/cifs/cifs_fs_sb.h

@@ -61,11 +61,6 @@ struct cifs_sb_info {
 	/* only used when CIFS_MOUNT_USE_PREFIX_PATH is set */
 	char *prepath;
 
-	/*
-	 * Canonical DFS path initially provided by the mount call. We might connect to something
-	 * different via DFS but we want to keep it to do failover properly.
-	 */
-	char *origin_fullpath; /* \\HOST\SHARE\[OPTIONAL PATH] */
 	/* randomly generated 128-bit number for indexing dfs mount groups in referral cache */
 	uuid_t dfs_mount_id;
 	/*

fs/cifs/cifsglob.h

@@ -697,6 +697,19 @@ struct TCP_Server_Info {
 #endif
 #ifdef CONFIG_CIFS_DFS_UPCALL
 	bool is_dfs_conn; /* if a dfs connection */
+	struct mutex refpath_lock; /* protects leaf_fullpath */
+	/*
+	 * Canonical DFS full paths that were used to chase referrals in mount and reconnect.
+	 *
+	 * origin_fullpath: first or original referral path
+	 * leaf_fullpath: last referral path (might be changed due to nested links in reconnect)
+	 *
+	 * current_fullpath: pointer to either origin_fullpath or leaf_fullpath
+	 * NOTE: cannot be accessed outside cifs_reconnect() and smb2_reconnect()
+	 *
+	 * format: \\HOST\SHARE\[OPTIONAL PATH]
+	 */
+	char *origin_fullpath, *leaf_fullpath, *current_fullpath;
 #endif
 };
 
@@ -1097,7 +1110,6 @@ struct cifs_tcon {
 	struct cached_fid crfid; /* Cached root fid */
 	/* BB add field for back pointer to sb struct(s)? */
 #ifdef CONFIG_CIFS_DFS_UPCALL
-	char *dfs_path; /* canonical DFS path */
 	struct list_head ulist; /* cache update list */
 #endif
 };
@@ -1948,4 +1960,14 @@ static inline bool is_tcon_dfs(struct cifs_tcon *tcon)
 		tcon->share_flags & (SHI1005_FLAGS_DFS | SHI1005_FLAGS_DFS_ROOT);
 }
 
+static inline bool cifs_is_referral_server(struct cifs_tcon *tcon,
+					   const struct dfs_info3_param *ref)
+{
+	/*
+	 * Check if all targets are capable of handling DFS referrals as per
+	 * MS-DFSC 2.2.4 RESP_GET_DFS_REFERRAL.
+	 */
+	return is_tcon_dfs(tcon) || (ref && (ref->flags & DFSREF_REFERRAL_SERVER));
+}
+
 #endif	/* _CIFS_GLOB_H */

fs/cifs/cifsproto.h

@@ -607,7 +607,7 @@ int smb2_parse_query_directory(struct cifs_tcon *tcon, struct kvec *rsp_iov,
 
 struct super_block *cifs_get_tcp_super(struct TCP_Server_Info *server);
 void cifs_put_tcp_super(struct super_block *sb);
-int update_super_prepath(struct cifs_tcon *tcon, char *prefix);
+int cifs_update_super_prepath(struct cifs_sb_info *cifs_sb, char *prefix);
 char *extract_hostname(const char *unc);
 char *extract_sharename(const char *unc);
 
@@ -634,4 +634,7 @@ static inline int cifs_create_options(struct cifs_sb_info *cifs_sb, int options)
 		return options;
 }
 
+struct super_block *cifs_get_tcon_super(struct cifs_tcon *tcon);
+void cifs_put_tcon_super(struct super_block *sb);
+
 #endif			/* _CIFSPROTO_H */

fs/cifs/dfs_cache.c

@@ -1364,9 +1364,9 @@ static void mark_for_reconnect_if_needed(struct cifs_tcon *tcon, struct dfs_cach
 }
 
 /* Refresh dfs referral of tcon and mark it for reconnect if needed */
-static int refresh_tcon(struct cifs_ses **sessions, struct cifs_tcon *tcon, bool force_refresh)
+static int __refresh_tcon(const char *path, struct cifs_ses **sessions, struct cifs_tcon *tcon,
+			  bool force_refresh)
 {
-	const char *path = tcon->dfs_path + 1;
 	struct cifs_ses *ses;
 	struct cache_entry *ce;
 	struct dfs_info3_param *refs = NULL;
@@ -1422,6 +1422,20 @@ out:
 	return rc;
 }
 
+static int refresh_tcon(struct cifs_ses **sessions, struct cifs_tcon *tcon, bool force_refresh)
+{
+	struct TCP_Server_Info *server = tcon->ses->server;
+
+	mutex_lock(&server->refpath_lock);
+	if (strcasecmp(server->leaf_fullpath, server->origin_fullpath))
+		__refresh_tcon(server->leaf_fullpath + 1, sessions, tcon, force_refresh);
+	mutex_unlock(&server->refpath_lock);
+
+	__refresh_tcon(server->origin_fullpath + 1, sessions, tcon, force_refresh);
+
+	return 0;
+}
+
 /**
  * dfs_cache_remount_fs - remount a DFS share
  *
@@ -1435,6 +1449,7 @@ out:
 int dfs_cache_remount_fs(struct cifs_sb_info *cifs_sb)
 {
 	struct cifs_tcon *tcon;
+	struct TCP_Server_Info *server;
 	struct mount_group *mg;
 	struct cifs_ses *sessions[CACHE_MAX_ENTRIES + 1] = {NULL};
 	int rc;
@@ -1443,13 +1458,15 @@ int dfs_cache_remount_fs(struct cifs_sb_info *cifs_sb)
 		return -EINVAL;
 
 	tcon = cifs_sb_master_tcon(cifs_sb);
-	if (!tcon->dfs_path) {
+	server = tcon->ses->server;
-		cifs_dbg(FYI, "%s: not a dfs tcon\n", __func__);
+
+	if (!server->origin_fullpath) {
+		cifs_dbg(FYI, "%s: not a dfs mount\n", __func__);
 		return 0;
 	}
 
 	if (uuid_is_null(&cifs_sb->dfs_mount_id)) {
-		cifs_dbg(FYI, "%s: tcon has no dfs mount group id\n", __func__);
+		cifs_dbg(FYI, "%s: no dfs mount group id\n", __func__);
 		return -EINVAL;
 	}
 
@@ -1457,7 +1474,7 @@ int dfs_cache_remount_fs(struct cifs_sb_info *cifs_sb)
 	mg = find_mount_group_locked(&cifs_sb->dfs_mount_id);
 	if (IS_ERR(mg)) {
 		mutex_unlock(&mount_group_list_lock);
-		cifs_dbg(FYI, "%s: tcon has ipc session to refresh referral\n", __func__);
+		cifs_dbg(FYI, "%s: no ipc session for refreshing referral\n", __func__);
 		return PTR_ERR(mg);
 	}
 	kref_get(&mg->refcount);
@@ -1498,9 +1515,12 @@ static void refresh_mounts(struct cifs_ses **sessions)
 
 	spin_lock(&cifs_tcp_ses_lock);
 	list_for_each_entry(server, &cifs_tcp_ses_list, tcp_ses_list) {
+		if (!server->is_dfs_conn)
+			continue;
+
 		list_for_each_entry(ses, &server->smb_ses_list, smb_ses_list) {
 			list_for_each_entry(tcon, &ses->tcon_list, tcon_list) {
-				if (tcon->dfs_path) {
+				if (!tcon->ipc && !tcon->need_reconnect) {
 					tcon->tc_count++;
 					list_add_tail(&tcon->ulist, &tcons);
 				}
@@ -1510,8 +1530,16 @@ static void refresh_mounts(struct cifs_ses **sessions)
 	spin_unlock(&cifs_tcp_ses_lock);
 
 	list_for_each_entry_safe(tcon, ntcon, &tcons, ulist) {
+		struct TCP_Server_Info *server = tcon->ses->server;
+
 		list_del_init(&tcon->ulist);
-		refresh_tcon(sessions, tcon, false);
+
+		mutex_lock(&server->refpath_lock);
+		if (strcasecmp(server->leaf_fullpath, server->origin_fullpath))
+			__refresh_tcon(server->leaf_fullpath + 1, sessions, tcon, false);
+		mutex_unlock(&server->refpath_lock);
+
+		__refresh_tcon(server->origin_fullpath + 1, sessions, tcon, false);
 		cifs_put_tcon(tcon);
 	}
 }

fs/cifs/misc.c

@@ -139,9 +139,6 @@ tconInfoFree(struct cifs_tcon *buf_to_free)
 	kfree(buf_to_free->nativeFileSystem);
 	kfree_sensitive(buf_to_free->password);
 	kfree(buf_to_free->crfid.fid);
-#ifdef CONFIG_CIFS_DFS_UPCALL
-	kfree(buf_to_free->dfs_path);
-#endif
 	kfree(buf_to_free);
 }
 
@@ -1288,69 +1285,20 @@ out:
 	return rc;
 }
 
-static void tcon_super_cb(struct super_block *sb, void *arg)
+int cifs_update_super_prepath(struct cifs_sb_info *cifs_sb, char *prefix)
 {
-	struct super_cb_data *sd = arg;
-	struct cifs_tcon *tcon = sd->data;
-	struct cifs_sb_info *cifs_sb;
-
-	if (sd->sb)
-		return;
-
-	cifs_sb = CIFS_SB(sb);
-	if (tcon->dfs_path && cifs_sb->origin_fullpath &&
-	    !strcasecmp(tcon->dfs_path, cifs_sb->origin_fullpath))
-		sd->sb = sb;
-}
-
-static inline struct super_block *cifs_get_tcon_super(struct cifs_tcon *tcon)
-{
-	return __cifs_get_super(tcon_super_cb, tcon);
-}
-
-static inline void cifs_put_tcon_super(struct super_block *sb)
-{
-	__cifs_put_super(sb);
-}
-#else
-static inline struct super_block *cifs_get_tcon_super(struct cifs_tcon *tcon)
-{
-	return ERR_PTR(-EOPNOTSUPP);
-}
-
-static inline void cifs_put_tcon_super(struct super_block *sb)
-{
-}
-#endif
-
-int update_super_prepath(struct cifs_tcon *tcon, char *prefix)
-{
-	struct super_block *sb;
-	struct cifs_sb_info *cifs_sb;
-	int rc = 0;
-
-	sb = cifs_get_tcon_super(tcon);
-	if (IS_ERR(sb))
-		return PTR_ERR(sb);
-
-	cifs_sb = CIFS_SB(sb);
-
 	kfree(cifs_sb->prepath);
 
 	if (prefix && *prefix) {
 		cifs_sb->prepath = kstrdup(prefix, GFP_ATOMIC);
-		if (!cifs_sb->prepath) {
+		if (!cifs_sb->prepath)
-			rc = -ENOMEM;
+			return -ENOMEM;
-			goto out;
-		}
 
 		convert_delimiter(cifs_sb->prepath, CIFS_DIR_SEP(cifs_sb));
 	} else
 		cifs_sb->prepath = NULL;
 
 	cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_USE_PREFIX_PATH;
-
+	return 0;
-out:
-	cifs_put_tcon_super(sb);
-	return rc;
 }
+#endif

fs/cifs/smb2ops.c

@@ -2844,6 +2844,7 @@ smb2_get_dfs_refer(const unsigned int xid, struct cifs_ses *ses,
 	struct fsctl_get_dfs_referral_req *dfs_req = NULL;
 	struct get_dfs_referral_rsp *dfs_rsp = NULL;
 	u32 dfs_req_size = 0, dfs_rsp_size = 0;
+	int retry_count = 0;
 
 	cifs_dbg(FYI, "%s: path: %s\n", __func__, search_name);
 
@@ -2895,11 +2896,14 @@ smb2_get_dfs_refer(const unsigned int xid, struct cifs_ses *ses,
 				true /* is_fsctl */,
 				(char *)dfs_req, dfs_req_size, CIFSMaxBufSize,
 				(char **)&dfs_rsp, &dfs_rsp_size);
-	} while (rc == -EAGAIN);
+		if (!is_retryable_error(rc))
+			break;
+		usleep_range(512, 2048);
+	} while (++retry_count < 5);
 
 	if (rc) {
-		if ((rc != -ENOENT) && (rc != -EOPNOTSUPP))
+		if (!is_retryable_error(rc) && rc != -ENOENT && rc != -EOPNOTSUPP)
-			cifs_tcon_dbg(VFS, "ioctl error in %s rc=%d\n", __func__, rc);
+			cifs_tcon_dbg(VFS, "%s: ioctl error: rc=%d\n", __func__, rc);
 		goto out;
 	}
 

fs/cifs/smb2pdu.c

@@ -155,7 +155,11 @@ smb2_reconnect(__le16 smb2_command, struct cifs_tcon *tcon,
 	if (tcon == NULL)
 		return 0;
 
-	if (smb2_command == SMB2_TREE_CONNECT)
+	/*
+	 * Need to also skip SMB2_IOCTL because it is used for checking nested dfs links in
+	 * cifs_tree_connect().
+	 */
+	if (smb2_command == SMB2_TREE_CONNECT || smb2_command == SMB2_IOCTL)
 		return 0;
 
 	if (tcon->tidStatus == CifsExiting) {