[PATCH] ceph: don't rely on error_string to validate blocklisted session.

[PATCH] ceph: don't rely on error_string to validate blocklisted session.

[khiremat]

From: Kotresh HR <khiremat@redhat.com>

The "error_string" in the metadata of MClientSession is being
parsed by kclient to validate whether the session is blocklisted.
The "error_string" is for humans and shouldn't be relied on it.
Hence added the flag to MClientsession to indicate the session
is blocklisted.

URL: https://tracker.ceph.com/issues/47450
Signed-off-by: Kotresh HR <khiremat@redhat.com>
---
 fs/ceph/mds_client.c         | 13 +++++++++++++
 include/linux/ceph/ceph_fs.h |  2 ++
 2 files changed, 15 insertions(+)

diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 44bc780b2b0e..f3c023c17963 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -3407,6 +3407,19 @@ static void handle_session(struct ceph_mds_session *session,
 		}
 	}
 
+	if (msg_version >= 5) {
+		u32 len;
+		u32 flags;
+		/* version >= 4, metric_spec (struct_v, struct_cv, len, metric_flag) */
+	        ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 2, bad);
+		/* version >= 5, flags   */
+                ceph_decode_32_safe(&p, end, flags, bad);
+		if (flags & CEPH_SESSION_BLOCKLISTED) {
+		        pr_info("mds%d session blocklisted\n", session->s_mds);
+			blocklisted = true;
+		}
+	}
+
 	mutex_lock(&mdsc->mutex);
 	if (op == CEPH_SESSION_CLOSE) {
 		ceph_get_mds_session(session);
diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
index bc2699feddbe..7ad6c3d0db7d 100644
--- a/include/linux/ceph/ceph_fs.h
+++ b/include/linux/ceph/ceph_fs.h
@@ -302,6 +302,8 @@ enum {
 	CEPH_SESSION_REQUEST_FLUSH_MDLOG,
 };
 
+#define CEPH_SESSION_BLOCKLISTED	(1 << 0)  /* session blocklisted */
+
 extern const char *ceph_session_op_name(int op);
 
 struct ceph_mds_session_head {
-- 
2.25.1

Re: [PATCH] ceph: don't rely on error_string to validate blocklisted session.

[Jeff Layton]

On Thu, 2021-09-23 at 18:56 +0530, khiremat@redhat.com wrote:
> From: Kotresh HR <khiremat@redhat.com>
> 
> The "error_string" in the metadata of MClientSession is being
> parsed by kclient to validate whether the session is blocklisted.
> The "error_string" is for humans and shouldn't be relied on it.
> Hence added the flag to MClientsession to indicate the session
> is blocklisted.
> 
> URL: https://tracker.ceph.com/issues/47450
> Signed-off-by: Kotresh HR <khiremat@redhat.com>
> ---
>  fs/ceph/mds_client.c         | 13 +++++++++++++
>  include/linux/ceph/ceph_fs.h |  2 ++
>  2 files changed, 15 insertions(+)
> 
> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> index 44bc780b2b0e..f3c023c17963 100644
> --- a/fs/ceph/mds_client.c
> +++ b/fs/ceph/mds_client.c
> @@ -3407,6 +3407,19 @@ static void handle_session(struct ceph_mds_session *session,
>  		}
>  	}
>  
> +	if (msg_version >= 5) {
> +		u32 len;
> +		u32 flags;
> +		/* version >= 4, metric_spec (struct_v, struct_cv, len, metric_flag) */
> +	        ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 2, bad);
> +		/* version >= 5, flags   */
> +                ceph_decode_32_safe(&p, end, flags, bad);
> +		if (flags & CEPH_SESSION_BLOCKLISTED) {
> +		        pr_info("mds%d session blocklisted\n", session->s_mds);

Things are likely to not work well if this flag is set. Maybe this
should be a pr_warn?

> +			blocklisted = true;
> +		}
> +	}
> +
>  	mutex_lock(&mdsc->mutex);
>  	if (op == CEPH_SESSION_CLOSE) {
>  		ceph_get_mds_session(session);
> diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
> index bc2699feddbe..7ad6c3d0db7d 100644
> --- a/include/linux/ceph/ceph_fs.h
> +++ b/include/linux/ceph/ceph_fs.h
> @@ -302,6 +302,8 @@ enum {
>  	CEPH_SESSION_REQUEST_FLUSH_MDLOG,
>  };
>  
> +#define CEPH_SESSION_BLOCKLISTED	(1 << 0)  /* session blocklisted */
> +
>  extern const char *ceph_session_op_name(int op);
>  
>  struct ceph_mds_session_head {

This looks reasonable, but incomplete.

If the msg_version >= 5, then we probably also don't need to scan for
"blacklisted" in the strings, right? In fact, it looks like we can avoid
doing most of __decode_session_metadata in that case and can just skip
over the text blob section.
-- 
Jeff Layton <jlayton@kernel.org>