6174 /dev/zvol does not show pool directories

Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jason King <jason.brian.king@gmail.com>
Approved by: Dan McDonald <danmcd@omniti.com>

3 files changed, 91 insertions, 21 deletions

diff --git a/usr/src/uts/common/fs/dev/sdev_vfsops.c b/usr/src/uts/common/fs/dev/sdev_vfsops.c
index ea9cb6374a..00e981ce9c 100644
--- a/usr/src/uts/common/fs/dev/sdev_vfsops.c
+++ b/usr/src/uts/common/fs/dev/sdev_vfsops.c
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -56,6 +57,7 @@
 #include <sys/fs/dv_node.h>
 #include <sys/sunndi.h>
 #include <sys/mntent.h>
+#include <sys/disp.h>
 
 /*
  * /dev vfs operations.
@@ -66,6 +68,7 @@
  */
 struct sdev_data *sdev_origins; /* mount info for origins under /dev */
 kmutex_t sdev_lock; /* used for mount/unmount/rename synchronization */
+taskq_t *sdev_taskq = NULL;
 
 /*
  * static
@@ -254,6 +257,20 @@ sdev_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap,
 	mutex_enter(&sdev_lock);
 
 	/*
+	 * Check that the taskq has been created. We can't do this in our
+	 * _init or devinit because they run too early for ddi_taskq_create.
+	 */
+	if (sdev_taskq == NULL) {
+		sdev_taskq = taskq_create("sdev", 1, minclsyspri, 1, 1, 0);
+		if (sdev_taskq == NULL) {
+			error = ENOMEM;
+			mutex_exit(&sdev_lock);
+			VN_RELE(avp);
+			goto cleanup;
+		}
+	}
+
+	/*
 	 * handling installation
 	 */
 	if (uap->flags & MS_REMOUNT) {
diff --git a/usr/src/uts/common/fs/dev/sdev_zvolops.c b/usr/src/uts/common/fs/dev/sdev_zvolops.c
index e4c3acf787..54061b48cf 100644
--- a/usr/src/uts/common/fs/dev/sdev_zvolops.c
+++ b/usr/src/uts/common/fs/dev/sdev_zvolops.c
@@ -41,14 +41,18 @@
 #include <sys/vfs_opreg.h>
 
 struct vnodeops	*devzvol_vnodeops;
+static major_t devzvol_major;
+static taskq_ent_t devzvol_zclist_task;
+
+static kmutex_t devzvol_mtx;
+/* Below are protected by devzvol_mtx */
+static boolean_t devzvol_isopen;
+static boolean_t devzvol_zclist_task_running = B_FALSE;
 static uint64_t devzvol_gen = 0;
 static uint64_t devzvol_zclist;
 static size_t devzvol_zclist_size;
 static ldi_ident_t devzvol_li;
 static ldi_handle_t devzvol_lh;
-static kmutex_t devzvol_mtx;
-static boolean_t devzvol_isopen;
-static major_t devzvol_major;
 
 /*
  * we need to use ddi_mod* since fs/dev gets loaded early on in
@@ -310,19 +314,20 @@ devzvol_validate(struct sdev_node *dv)
 }
 
 /*
- * creates directories as needed in response to a readdir
+ * Taskq callback to update the devzvol_zclist.
+ *
+ * We need to defer this to the taskq to avoid it running with a user
+ * context that might be associated with some non-global zone, and thus
+ * not being able to list all of the pools on the entire system.
  */
-void
-devzvol_create_pool_dirs(struct vnode *dvp)
+/*ARGSUSED*/
+static void
+devzvol_update_zclist_cb(void *arg)
 {
 	zfs_cmd_t	*zc;
-	nvlist_t *nv = NULL;
-	nvpair_t *elem = NULL;
-	size_t size;
-	int pools = 0;
-	int rc;
+	int		rc;
+	size_t		size;
 
-	sdcmn_err13(("devzvol_create_pool_dirs"));
 	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
 	mutex_enter(&devzvol_mtx);
 	zc->zc_cookie = devzvol_gen;
@@ -337,22 +342,67 @@ devzvol_create_pool_dirs(struct vnode *dvp)
 				kmem_free((void *)(uintptr_t)devzvol_zclist,
 				    devzvol_zclist_size);
 			devzvol_zclist = zc->zc_nvlist_dst;
+			/* Keep the alloc'd size, not the nvlist size. */
 			devzvol_zclist_size = size;
 			break;
-		case EEXIST:
+		default:
 			/*
-			 * no change in the configuration; still need
-			 * to do lookups in case we did a lookup in
-			 * zvol/rdsk but not zvol/dsk (or vice versa)
+			 * Either there was no change in pool configuration
+			 * since we last asked (rc == EEXIST) or we got a
+			 * catastrophic error.
+			 *
+			 * Give up memory and exit.
 			 */
 			kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst,
 			    size);
 			break;
-		default:
-			kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst,
-			    size);
-			goto out;
 	}
+
+	VERIFY(devzvol_zclist_task_running == B_TRUE);
+	devzvol_zclist_task_running = B_FALSE;
+	mutex_exit(&devzvol_mtx);
+
+	kmem_free(zc, sizeof (zfs_cmd_t));
+}
+
+static void
+devzvol_update_zclist(void)
+{
+	mutex_enter(&devzvol_mtx);
+	if (devzvol_zclist_task_running == B_TRUE) {
+		mutex_exit(&devzvol_mtx);
+		goto wait;
+	}
+
+	devzvol_zclist_task_running = B_TRUE;
+
+	taskq_dispatch_ent(sdev_taskq, devzvol_update_zclist_cb, NULL, 0,
+	    &devzvol_zclist_task);
+
+	mutex_exit(&devzvol_mtx);
+
+wait:
+	taskq_wait(sdev_taskq);
+}
+
+/*
+ * Creates sub-directories for each zpool as needed in response to a
+ * readdir on one of the /dev/zvol/{dsk,rdsk} directories.
+ */
+void
+devzvol_create_pool_dirs(struct vnode *dvp)
+{
+	nvlist_t *nv = NULL;
+	nvpair_t *elem = NULL;
+	int pools = 0;
+	int rc;
+
+	sdcmn_err13(("devzvol_create_pool_dirs"));
+
+	devzvol_update_zclist();
+
+	mutex_enter(&devzvol_mtx);
+
 	rc = nvlist_unpack((char *)(uintptr_t)devzvol_zclist,
 	    devzvol_zclist_size, &nv, 0);
 	if (rc) {
@@ -385,7 +435,6 @@ devzvol_create_pool_dirs(struct vnode *dvp)
 	}
 out:
 	mutex_exit(&devzvol_mtx);
-	kmem_free(zc, sizeof (zfs_cmd_t));
 }
 
 /*ARGSUSED3*/
diff --git a/usr/src/uts/common/sys/fs/sdev_impl.h b/usr/src/uts/common/sys/fs/sdev_impl.h
index 19a147d88a..7ea6b88ec2 100644
--- a/usr/src/uts/common/sys/fs/sdev_impl.h
+++ b/usr/src/uts/common/sys/fs/sdev_impl.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015 Joyent, Inc.  All rights reserved.
  */
 
 #ifndef _SYS_SDEV_IMPL_H
@@ -35,6 +36,7 @@ extern "C" {
 #include <sys/vfs_opreg.h>
 #include <sys/list.h>
 #include <sys/nvpair.h>
+#include <sys/sunddi.h>
 
 /*
  * sdev_nodes are the file-system specific part of the
@@ -541,6 +543,8 @@ extern int		sdev_nc_disable;
 extern int		sdev_nc_disable_reset;
 extern int		sdev_nc_verbose;
 
+extern taskq_t		*sdev_taskq;
+
 /*
  * misc. defines
  */