Skip to contentSkip to conversation

cifs: all initializations for tcon should happen in tcon_info_alloc

PATCH In review [email protected] 3 participants 5 messages Jun 30, 2025 – Jul 1, 2025

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cifs: all initializations for tcon should happen in tcon_info_alloc

From: Shyam Prasad N <[email protected]>

Today, a few work structs inside tcon are initialized inside cifs_get_tcon and not in tcon_info_alloc. As a result, if a tcon is obtained from tcon_info_alloc, but not called as a part of cifs_get_tcon, we may trip over.

---
 fs/smb/client/cifsproto.h | 1 +
 fs/smb/client/connect.c   | 8 +-------
 fs/smb/client/misc.c      | 6 ++++++
 3 files changed, 8 insertions(+), 7 deletions(-)
fs/smb/client/cifsproto.h +1 -0
index 66093fa78aed..045227ed4efc 100644
@@ -136,6 +136,7 @@ extern int SendReceiveBlockingLock(const unsigned int xid,
 			struct smb_hdr *out_buf,
 			int *bytes_returned);
 
+void smb2_query_server_interfaces(struct work_struct *work);
 void
 cifs_signal_cifsd_for_reconnect(struct TCP_Server_Info *server,
 				      bool all_channels);
fs/smb/client/connect.c +1 -7
index c48869c29e15..16c4f7fa1f34 100644
@@ -97,7 +97,7 @@ static int reconn_set_ipaddr_from_hostname(struct TCP_Server_Info *server)
 	return rc;
 }
 
-static void smb2_query_server_interfaces(struct work_struct *work)
+void smb2_query_server_interfaces(struct work_struct *work)
 {
 	int rc;
 	int xid;
@@ -2866,20 +2866,14 @@ cifs_get_tcon(struct cifs_ses *ses, struct smb3_fs_context *ctx)
 	tcon->max_cached_dirs = ctx->max_cached_dirs;
 	tcon->nodelete = ctx->nodelete;
 	tcon->local_lease = ctx->local_lease;
-	INIT_LIST_HEAD(&tcon->pending_opens);
 	tcon->status = TID_GOOD;
 
-	INIT_DELAYED_WORK(&tcon->query_interfaces,
-			  smb2_query_server_interfaces);
 	if (ses->server->dialect >= SMB30_PROT_ID &&
 	    (ses->server->capabilities & SMB2_GLOBAL_CAP_MULTI_CHANNEL)) {
 		/* schedule query interfaces poll */
 		queue_delayed_work(cifsiod_wq, &tcon->query_interfaces,
 				   (SMB_INTERFACE_POLL_INTERVAL * HZ));
 	}
-#ifdef CONFIG_CIFS_DFS_UPCALL
-	INIT_DELAYED_WORK(&tcon->dfs_cache_work, dfs_cache_refresh);
-#endif
 	spin_lock(&cifs_tcp_ses_lock);
 	list_add(&tcon->tcon_list, &ses->tcon_list);
 	spin_unlock(&cifs_tcp_ses_lock);
fs/smb/client/misc.c +6 -0
index e77017f47084..da23cc12a52c 100644
@@ -151,6 +151,12 @@ tcon_info_alloc(bool dir_leases_enabled, enum smb3_tcon_ref_trace trace)
 #ifdef CONFIG_CIFS_DFS_UPCALL
 	INIT_LIST_HEAD(&ret_buf->dfs_ses_list);
 #endif
+	INIT_LIST_HEAD(&ret_buf->pending_opens);
+	INIT_DELAYED_WORK(&ret_buf->query_interfaces,
+			  smb2_query_server_interfaces);
+#ifdef CONFIG_CIFS_DFS_UPCALL
+	INIT_DELAYED_WORK(&ret_buf->dfs_cache_work, dfs_cache_refresh);
+#endif
 
 	return ret_buf;
 }
Signature · 1 line
2.43.0
#1 lore raw

tentatively merged into cifs-2.6.git for-next pending testing and more review

Quoted 82 lines · On Mon, Jun 30, 2025 at 12:40 PM <[email protected]> wrote:

From: Shyam Prasad N <[email protected]>

Today, a few work structs inside tcon are initialized inside cifs_get_tcon and not in tcon_info_alloc. As a result, if a tcon is obtained from tcon_info_alloc, but not called as a part of cifs_get_tcon, we may trip over.

---
 fs/smb/client/cifsproto.h | 1 +
 fs/smb/client/connect.c   | 8 +-------
 fs/smb/client/misc.c      | 6 ++++++
 3 files changed, 8 insertions(+), 7 deletions(-)
fs/smb/client/cifsproto.h +1 -0
index 66093fa78aed..045227ed4efc 100644
@@ -136,6 +136,7 @@ extern int SendReceiveBlockingLock(const unsigned int xid,
                        struct smb_hdr *out_buf,
                        int *bytes_returned);
 
+void smb2_query_server_interfaces(struct work_struct *work);
 void
 cifs_signal_cifsd_for_reconnect(struct TCP_Server_Info *server,
                                      bool all_channels);
fs/smb/client/connect.c +1 -7
index c48869c29e15..16c4f7fa1f34 100644
@@ -97,7 +97,7 @@ static int reconn_set_ipaddr_from_hostname(struct TCP_Server_Info *server)
        return rc;
 }
 
-static void smb2_query_server_interfaces(struct work_struct *work)
+void smb2_query_server_interfaces(struct work_struct *work)
 {
        int rc;
        int xid;
@@ -2866,20 +2866,14 @@ cifs_get_tcon(struct cifs_ses *ses, struct smb3_fs_context *ctx)
        tcon->max_cached_dirs = ctx->max_cached_dirs;
        tcon->nodelete = ctx->nodelete;
        tcon->local_lease = ctx->local_lease;
-       INIT_LIST_HEAD(&tcon->pending_opens);
        tcon->status = TID_GOOD;
 
-       INIT_DELAYED_WORK(&tcon->query_interfaces,
-                         smb2_query_server_interfaces);
        if (ses->server->dialect >= SMB30_PROT_ID &&
            (ses->server->capabilities & SMB2_GLOBAL_CAP_MULTI_CHANNEL)) {
                /* schedule query interfaces poll */
                queue_delayed_work(cifsiod_wq, &tcon->query_interfaces,
                                   (SMB_INTERFACE_POLL_INTERVAL * HZ));
        }
-#ifdef CONFIG_CIFS_DFS_UPCALL
-       INIT_DELAYED_WORK(&tcon->dfs_cache_work, dfs_cache_refresh);
-#endif
        spin_lock(&cifs_tcp_ses_lock);
        list_add(&tcon->tcon_list, &ses->tcon_list);
        spin_unlock(&cifs_tcp_ses_lock);
fs/smb/client/misc.c +6 -1
index e77017f47084..da23cc12a52c 100644
@@ -151,6 +151,12 @@ tcon_info_alloc(bool dir_leases_enabled, enum smb3_tcon_ref_trace trace)
 #ifdef CONFIG_CIFS_DFS_UPCALL
        INIT_LIST_HEAD(&ret_buf->dfs_ses_list);
 #endif
+       INIT_LIST_HEAD(&ret_buf->pending_opens);
+       INIT_DELAYED_WORK(&ret_buf->query_interfaces,
+                         smb2_query_server_interfaces);
+#ifdef CONFIG_CIFS_DFS_UPCALL
+       INIT_DELAYED_WORK(&ret_buf->dfs_cache_work, dfs_cache_refresh);
+#endif
 
        return ret_buf;
 }
--

2.43.0

Signature · 3 lines
Thanks,

Steve
maintainer Reviewed-by on v1 1/1
#2 lore raw
In reply to [email protected]
Quoted 8 lines · [email protected] writes:

From: Shyam Prasad N <[email protected]>

Today, a few work structs inside tcon are initialized inside cifs_get_tcon and not in tcon_info_alloc. As a result, if a tcon is obtained from tcon_info_alloc, but not called as a part of cifs_get_tcon, we may trip over.

stable? Makes no sense.

Quoted 6 lines
---
 fs/smb/client/cifsproto.h | 1 +
 fs/smb/client/connect.c   | 8 +-------
 fs/smb/client/misc.c      | 6 ++++++
 3 files changed, 8 insertions(+), 7 deletions(-)

Otherwise, looks good:

#3 lore raw
Quoted 5 lines · Paulo Alcantara
Quoted 2 lines · [email protected] writes:

From: Shyam Prasad N <[email protected]>

Quoted 8 lines
Quoted 6 lines

Today, a few work structs inside tcon are initialized inside cifs_get_tcon and not in tcon_info_alloc. As a result, if a tcon is obtained from tcon_info_alloc, but not called as a part of cifs_get_tcon, we may trip over.

stable? Makes no sense.

I feel this is a serious one. If some code were to use tcon_info_alloc, they'd expect that it's fully initialized, but they'd end up with the problem that you and David saw.
I feel that this is the correct fix to that problem (although that addresses the problem of unnecessary scheduling of work).

Quoted 11 lines
Quoted 7 lines
---
 fs/smb/client/cifsproto.h | 1 +
 fs/smb/client/connect.c   | 8 +-------
 fs/smb/client/misc.c      | 6 ++++++
 3 files changed, 8 insertions(+), 7 deletions(-)

Otherwise, looks good:

Thanks.

Signature · 2 lines
Regards,
Shyam
maintainer
#4 lore raw
Quoted 11 lines · Shyam Prasad N
Quoted 10 lines · Paulo Alcantara
Quoted 7 lines · [email protected] writes:

From: Shyam Prasad N <[email protected]>

Today, a few work structs inside tcon are initialized inside cifs_get_tcon and not in tcon_info_alloc. As a result, if a tcon is obtained from tcon_info_alloc, but not called as a part of cifs_get_tcon, we may trip over.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

stable? Makes no sense.

I feel this is a serious one. If some code were to use tcon_info_alloc, they'd expect that it's fully initialized, but they'd end up with the problem that you and David saw.

Yes, I understand you want to be safe. But you're not fixing any existing problem with this patch, hence Cc stable didn't make sense to me.

Quoted 2 lines

I feel that this is the correct fix to that problem (although that addresses the problem of unnecessary scheduling of work).

You'd just mask the real problem with this. Without the WARN_ON() on the uninitialized delayed worker we wouldn't have found the actual bug, though.