libceph: delay debugfs initialization until we learn global_id
authorSage Weil <sage@inktank.com>
Sun, 19 Aug 2012 19:29:16 +0000 (12:29 -0700)
committerSage Weil <sage@inktank.com>
Mon, 20 Aug 2012 17:03:15 +0000 (10:03 -0700)
The debugfs directory includes the cluster fsid and our unique global_id.
We need to delay the initialization of the debug entry until we have
learned both the fsid and our global_id from the monitor or else the
second client can't create its debugfs entry and will fail (and multiple
client instances aren't properly reflected in debugfs).

Reported by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
fs/ceph/debugfs.c
net/ceph/ceph_common.c
net/ceph/debugfs.c
net/ceph/mon_client.c

index fb962efdacee8cdbbdff2c3324516d0356d2e2e5..6d59006bfa27e688d37499b88f45e28a22df24db 100644 (file)
@@ -201,6 +201,7 @@ int ceph_fs_debugfs_init(struct ceph_fs_client *fsc)
        int err = -ENOMEM;
 
        dout("ceph_fs_debugfs_init\n");
+       BUG_ON(!fsc->client->debugfs_dir);
        fsc->debugfs_congestion_kb =
                debugfs_create_file("writeback_congestion_kb",
                                    0600,
index 69e38db28e5f3efcda5ce871ac5682408eb5aa9e..a8020293f34210620eff72f6193ee35e064f5d69 100644 (file)
@@ -84,7 +84,6 @@ int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
                        return -1;
                }
        } else {
-               pr_info("client%lld fsid %pU\n", ceph_client_id(client), fsid);
                memcpy(&client->fsid, fsid, sizeof(*fsid));
        }
        return 0;
index 54b531a01121aafdbc31a9ba33fd60027280d32e..38b5dc1823d44961e6bbf7d52d36991920f5ddc1 100644 (file)
@@ -189,6 +189,9 @@ int ceph_debugfs_client_init(struct ceph_client *client)
        snprintf(name, sizeof(name), "%pU.client%lld", &client->fsid,
                 client->monc.auth->global_id);
 
+       dout("ceph_debugfs_client_init %p %s\n", client, name);
+
+       BUG_ON(client->debugfs_dir);
        client->debugfs_dir = debugfs_create_dir(name, ceph_debugfs_dir);
        if (!client->debugfs_dir)
                goto out;
@@ -234,6 +237,7 @@ out:
 
 void ceph_debugfs_client_cleanup(struct ceph_client *client)
 {
+       dout("ceph_debugfs_client_cleanup %p\n", client);
        debugfs_remove(client->debugfs_osdmap);
        debugfs_remove(client->debugfs_monmap);
        debugfs_remove(client->osdc.debugfs_file);
index 105d533b55f3bcc3e8a2e7a82b80a6459cbb9363..900ea0f043fc0d5cfd7c68b7abc37e607d822165 100644 (file)
@@ -310,6 +310,17 @@ int ceph_monc_open_session(struct ceph_mon_client *monc)
 }
 EXPORT_SYMBOL(ceph_monc_open_session);
 
+/*
+ * We require the fsid and global_id in order to initialize our
+ * debugfs dir.
+ */
+static bool have_debugfs_info(struct ceph_mon_client *monc)
+{
+       dout("have_debugfs_info fsid %d globalid %lld\n",
+            (int)monc->client->have_fsid, monc->auth->global_id);
+       return monc->client->have_fsid && monc->auth->global_id > 0;
+}
+
 /*
  * The monitor responds with mount ack indicate mount success.  The
  * included client ticket allows the client to talk to MDSs and OSDs.
@@ -320,9 +331,12 @@ static void ceph_monc_handle_map(struct ceph_mon_client *monc,
        struct ceph_client *client = monc->client;
        struct ceph_monmap *monmap = NULL, *old = monc->monmap;
        void *p, *end;
+       int had_debugfs_info, init_debugfs = 0;
 
        mutex_lock(&monc->mutex);
 
+       had_debugfs_info = have_debugfs_info(monc);
+
        dout("handle_monmap\n");
        p = msg->front.iov_base;
        end = p + msg->front.iov_len;
@@ -344,12 +358,22 @@ static void ceph_monc_handle_map(struct ceph_mon_client *monc,
 
        if (!client->have_fsid) {
                client->have_fsid = true;
+               if (!had_debugfs_info && have_debugfs_info(monc)) {
+                       pr_info("client%lld fsid %pU\n",
+                               ceph_client_id(monc->client),
+                               &monc->client->fsid);
+                       init_debugfs = 1;
+               }
                mutex_unlock(&monc->mutex);
-               /*
-                * do debugfs initialization without mutex to avoid
-                * creating a locking dependency
-                */
-               ceph_debugfs_client_init(client);
+
+               if (init_debugfs) {
+                       /*
+                        * do debugfs initialization without mutex to avoid
+                        * creating a locking dependency
+                        */
+                       ceph_debugfs_client_init(monc->client);
+               }
+
                goto out_unlocked;
        }
 out:
@@ -865,8 +889,10 @@ static void handle_auth_reply(struct ceph_mon_client *monc,
 {
        int ret;
        int was_auth = 0;
+       int had_debugfs_info, init_debugfs = 0;
 
        mutex_lock(&monc->mutex);
+       had_debugfs_info = have_debugfs_info(monc);
        if (monc->auth->ops)
                was_auth = monc->auth->ops->is_authenticated(monc->auth);
        monc->pending_auth = 0;
@@ -889,7 +915,22 @@ static void handle_auth_reply(struct ceph_mon_client *monc,
                __send_subscribe(monc);
                __resend_generic_request(monc);
        }
+
+       if (!had_debugfs_info && have_debugfs_info(monc)) {
+               pr_info("client%lld fsid %pU\n",
+                       ceph_client_id(monc->client),
+                       &monc->client->fsid);
+               init_debugfs = 1;
+       }
        mutex_unlock(&monc->mutex);
+
+       if (init_debugfs) {
+               /*
+                * do debugfs initialization without mutex to avoid
+                * creating a locking dependency
+                */
+               ceph_debugfs_client_init(monc->client);
+       }
 }
 
 static int __validate_auth(struct ceph_mon_client *monc)