async zvol minor node creation interferes with receive

When we finish a zfs receive, dmu_recv_end_sync() calls
zvol_create_minors(async=TRUE).  This kicks off some other threads that
create the minor device nodes (in /dev/zvol/poolname/...).  These async
threads call zvol_prefetch_minors_impl() and zvol_create_minor(), which
both call dmu_objset_own(), which puts a "long hold" on the dataset.
Since the zvol minor node creation is asynchronous, this can happen
after the `ZFS_IOC_RECV[_NEW]` ioctl and `zfs receive` process have
completed.

After the first receive ioctl has completed, userland may attempt to do
another receive into the same dataset (e.g. the next incremental
stream).  This second receive and the asynchronous minor node creation
can interfere with one another in several different ways, because they
both require exclusive access to the dataset:

1. When the second receive is finishing up, dmu_recv_end_check() does
dsl_dataset_handoff_check(), which can fail with EBUSY if the async
minor node creation already has a "long hold" on this dataset.  This
causes the 2nd receive to fail.

2. The async udev rule can fail if zvol_id and/or systemd-udevd try to
open the device while the the second receive's async attempt at minor
node creation owns the dataset (via zvol_prefetch_minors_impl).  This
causes the minor node (/dev/zd*) to exist, but the udev-generated
/dev/zvol/... to not exist.

3. The async minor node creation can silently fail with EBUSY if the
first receive's zvol_create_minor() trys to own the dataset while the
second receive's zvol_prefetch_minors_impl already owns the dataset.

To address these problems, this change synchronously creates the minor
node.  To avoid the lock ordering problems that the asynchrony was
introduced to fix (see #3681), we create the minor nodes from open
context, with no locks held, rather than from syncing contex as was
originally done.

Implementation notes:

We generally do not need to traverse children or prefetch anything (e.g.
when running the recv, snapshot, create, or clone subcommands of zfs).
We only need recursion when importing/opening a pool and when loading
encryption keys.  The existing recursive, asynchronous, prefetching code
is preserved for use in these cases.

Channel programs may need to create zvol minor nodes, when creating a
snapshot of a zvol with the snapdev property set.  We figure out what
snapshots are created when running the LUA program in syncing context.
In this case we need to remember what snapshots were created, and then
try to create their minor nodes from open context, after the LUA code
has completed.

There are additional zvol use cases that asynchronously own the dataset,
which can cause similar problems.  E.g. changing the volmode or snapdev
properties.  These are less problematic because they are not recursive
and don't touch datasets that are not involved in the operation, there
is still potential for interference with subsequent operations.  In the
future, these cases should be similarly converted to create the zvol
minor node synchronously from open context.

The async tasks of removing and renaming minors do not own the objset,
so they do not have this problem.  However, it may make sense to also
convert these operations to happen synchronously from open context, in
the future.

Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
External-issue: DLPX-65948
Closes #7863
Closes #9885

include/sys/zcp.h

@@ -131,6 +131,14 @@ typedef struct zcp_run_info {
 	 */
 	nvlist_t	*zri_outnvl;
 
+	/*
+	 * The keys of this nvlist are datasets which may be zvols and may need
+	 * to have device minor nodes created.  This information is passed from
+	 * syncing context (where the zvol is created) to open context (where we
+	 * create the minor nodes).
+	 */
+	nvlist_t	*zri_new_zvols;
+
 	/*
 	 * The errno number returned to caller of zcp_eval().
 	 */

include/sys/zvol.h

@@ -35,20 +35,20 @@
 #define	SPEC_MAXOFFSET_T	((1LL << ((NBBY * sizeof (daddr32_t)) + \
 				DEV_BSHIFT - 1)) - 1)
 
-extern void zvol_create_minors(spa_t *spa, const char *name, boolean_t async);
-extern void zvol_remove_minors(spa_t *spa, const char *name, boolean_t async);
-extern void zvol_rename_minors(spa_t *spa, const char *oldname,
-    const char *newname, boolean_t async);
+extern void zvol_create_minor(const char *);
+extern void zvol_create_minors_recursive(const char *);
+extern void zvol_remove_minors(spa_t *, const char *, boolean_t);
+extern void zvol_rename_minors(spa_t *, const char *, const char *, boolean_t);
 
 #ifdef _KERNEL
 struct zvol_state;
 typedef struct zvol_state zvol_state_handle_t;
 
-extern int zvol_check_volsize(uint64_t volsize, uint64_t blocksize);
-extern int zvol_check_volblocksize(const char *name, uint64_t volblocksize);
-extern int zvol_get_stats(objset_t *os, nvlist_t *nv);
+extern int zvol_check_volsize(uint64_t, uint64_t);
+extern int zvol_check_volblocksize(const char *, uint64_t);
+extern int zvol_get_stats(objset_t *, nvlist_t *);
 extern boolean_t zvol_is_zvol(const char *);
-extern void zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
+extern void zvol_create_cb(objset_t *, void *, cred_t *, dmu_tx_t *);
 extern int zvol_set_volsize(const char *, uint64_t);
 extern int zvol_set_volblocksize(const char *, uint64_t);
 extern int zvol_set_snapdev(const char *, zprop_source_t, uint64_t);

lib/libzpool/kernel.c

@@ -1013,7 +1013,12 @@ kmem_cache_reap_active(void)
 void *zvol_tag = "zvol_tag";
 
 void
-zvol_create_minors(spa_t *spa, const char *name, boolean_t async)
+zvol_create_minor(spa_t *spa, const char *name, boolean_t async)
+{
+}
+
+void
+zvol_create_minors_recursive(spa_t *spa, const char *name, boolean_t async)
 {
 }
 

module/os/linux/zfs/zvol_os.c

@@ -842,7 +842,7 @@ zvol_free(zvol_state_t *zv)
  * device is live and ready for use.
  */
 static int
-zvol_create_minor(const char *name)
+zvol_os_create_minor(const char *name)
 {
 	zvol_state_t *zv;
 	objset_t *os;
@@ -967,7 +967,7 @@ out_doi:
 		ida_simple_remove(&zvol_ida, idx);
 	}
 
-	return (SET_ERROR(error));
+	return (error);
 }
 
 static void
@@ -1014,7 +1014,7 @@ zvol_set_capacity_impl(zvol_state_t *zv, uint64_t capacity)
 const static zvol_platform_ops_t zvol_linux_ops = {
 	.zv_free = zvol_free,
 	.zv_rename_minor = zvol_rename_minor,
-	.zv_create_minor = zvol_create_minor,
+	.zv_create_minor = zvol_os_create_minor,
 	.zv_update_volsize = zvol_update_volsize,
 	.zv_clear_private = zvol_clear_private,
 	.zv_is_zvol = zvol_is_zvol_impl,

module/zfs/dmu_objset.c

@@ -1263,7 +1263,6 @@ dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
 	}
 
 	spa_history_log_internal_ds(ds, "create", tx, " ");
-	zvol_create_minors(spa, doca->doca_name, B_TRUE);
 
 	dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
 	dsl_dir_rele(pdd, FTAG);
@@ -1293,9 +1292,13 @@ dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
 	 */
 	doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp;
 
-	return (dsl_sync_task(name,
+	int rv = dsl_sync_task(name,
 	    dmu_objset_create_check, dmu_objset_create_sync, &doca,
-	    6, ZFS_SPACE_CHECK_NORMAL));
+	    6, ZFS_SPACE_CHECK_NORMAL);
+
+	if (rv == 0)
+		zvol_create_minor(name);
+	return (rv);
 }
 
 typedef struct dmu_objset_clone_arg {
@@ -1376,7 +1379,6 @@ dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
 	dsl_dataset_name(origin, namebuf);
 	spa_history_log_internal_ds(ds, "clone", tx,
 	    "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object);
-	zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE);
 	dsl_dataset_rele(ds, FTAG);
 	dsl_dataset_rele(origin, FTAG);
 	dsl_dir_rele(pdd, FTAG);
@@ -1391,9 +1393,14 @@ dmu_objset_clone(const char *clone, const char *origin)
 	doca.doca_origin = origin;
 	doca.doca_cred = CRED();
 
-	return (dsl_sync_task(clone,
+	int rv = dsl_sync_task(clone,
 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
-	    6, ZFS_SPACE_CHECK_NORMAL));
+	    6, ZFS_SPACE_CHECK_NORMAL);
+
+	if (rv == 0)
+		zvol_create_minor(clone);
+
+	return (rv);
 }
 
 int

module/zfs/dmu_recv.c

@@ -2859,6 +2859,12 @@ out:
 	if (drc->drc_next_rrd != NULL)
 		kmem_free(drc->drc_next_rrd, sizeof (*drc->drc_next_rrd));
 
+	/*
+	 * The objset will be invalidated by dmu_recv_end() when we do
+	 * dsl_dataset_clone_swap_sync_impl().
+	 */
+	drc->drc_os = NULL;
+
 	kmem_free(rwa, sizeof (*rwa));
 	nvlist_free(drc->drc_begin_nvl);
 	if ((drc->drc_featureflags & DMU_BACKUP_FEATURE_DEDUP) &&
@@ -3085,8 +3091,6 @@ dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
 		    &drc->drc_ivset_guid, tx));
 	}
 
-	zvol_create_minors(dp->dp_spa, drc->drc_tofs, B_TRUE);
-
 	/*
 	 * Release the hold from dmu_recv_begin.  This must be done before
 	 * we return to open context, so that when we free the dataset's dnode
@@ -3195,9 +3199,20 @@ dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
 	if (error != 0) {
 		dmu_recv_cleanup_ds(drc);
 		nvlist_free(drc->drc_keynvl);
-	} else if (drc->drc_guid_to_ds_map != NULL) {
-		(void) add_ds_to_guidmap(drc->drc_tofs, drc->drc_guid_to_ds_map,
-		    drc->drc_newsnapobj, drc->drc_raw);
+	} else {
+		if (drc->drc_newfs) {
+			zvol_create_minor(drc->drc_tofs);
+		}
+		char *snapname = kmem_asprintf("%s@%s",
+		    drc->drc_tofs, drc->drc_tosnap);
+		zvol_create_minor(snapname);
+		kmem_strfree(snapname);
+
+		if (drc->drc_guid_to_ds_map != NULL) {
+			(void) add_ds_to_guidmap(drc->drc_tofs,
+			    drc->drc_guid_to_ds_map,
+			    drc->drc_newsnapobj, drc->drc_raw);
+		}
 	}
 	return (error);
 }

module/zfs/dsl_crypt.c

@@ -854,7 +854,7 @@ spa_keystore_load_wkey(const char *dsname, dsl_crypto_params_t *dcp,
 	dsl_pool_rele(dp, FTAG);
 
 	/* create any zvols under this ds */
-	zvol_create_minors(dp->dp_spa, dsname, B_TRUE);
+	zvol_create_minors_recursive(dsname);
 
 	return (0);
 

module/zfs/dsl_dataset.c

@@ -1869,7 +1869,6 @@ dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
 			dsl_props_set_sync_impl(ds->ds_prev,
 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
 		}
-		zvol_create_minors(dp->dp_spa, nvpair_name(pair), B_TRUE);
 		dsl_dataset_rele(ds, FTAG);
 	}
 }
@@ -1944,6 +1943,13 @@ dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
 		fnvlist_free(suspended);
 	}
 
+	if (error == 0) {
+		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
+		    pair = nvlist_next_nvpair(snaps, pair)) {
+			zvol_create_minor(nvpair_name(pair));
+		}
+	}
+
 	return (error);
 }
 

module/zfs/spa.c

@@ -5114,7 +5114,7 @@ spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
 	}
 
 	if (firstopen)
-		zvol_create_minors(spa, spa_name(spa), B_TRUE);
+		zvol_create_minors_recursive(spa_name(spa));
 
 	*spapp = spa;
 
@@ -6083,7 +6083,7 @@ spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
 
 	mutex_exit(&spa_namespace_lock);
 
-	zvol_create_minors(spa, pool, B_TRUE);
+	zvol_create_minors_recursive(pool);
 
 	return (0);
 }

module/zfs/zcp.c

@@ -100,6 +100,7 @@
 #include <sys/zcp_iter.h>
 #include <sys/zcp_prop.h>
 #include <sys/zcp_global.h>
+#include <sys/zvol.h>
 
 #ifndef KM_NORMALPRI
 #define	KM_NORMALPRI	0
@@ -1155,6 +1156,7 @@ zcp_eval(const char *poolname, const char *program, boolean_t sync,
 	runinfo.zri_space_used = 0;
 	runinfo.zri_curinstrs = 0;
 	runinfo.zri_maxinstrs = instrlimit;
+	runinfo.zri_new_zvols = fnvlist_alloc();
 
 	if (sync) {
 		err = dsl_sync_task_sig(poolname, NULL, zcp_eval_sync,
@@ -1166,6 +1168,16 @@ zcp_eval(const char *poolname, const char *program, boolean_t sync,
 	}
 	lua_close(state);
 
+	/*
+	 * Create device minor nodes for any new zvols.
+	 */
+	for (nvpair_t *pair = nvlist_next_nvpair(runinfo.zri_new_zvols, NULL);
+	    pair != NULL;
+	    pair = nvlist_next_nvpair(runinfo.zri_new_zvols, pair)) {
+		zvol_create_minor(nvpair_name(pair));
+	}
+	fnvlist_free(runinfo.zri_new_zvols);
+
 	return (runinfo.zri_result);
 }
 

module/zfs/zcp_synctask.c

@@ -276,6 +276,16 @@ zcp_synctask_snapshot(lua_State *state, boolean_t sync, nvlist_t *err_details)
 	err = zcp_sync_task(state, dsl_dataset_snapshot_check,
 	    dsl_dataset_snapshot_sync, &ddsa, sync, dsname);
 
+	if (err == 0) {
+		/*
+		 * We may need to create a new device minor node for this
+		 * dataset (if it is a zvol and the "snapdev" property is set).
+		 * Save it in the nvlist so that it can be processed in open
+		 * context.
+		 */
+		fnvlist_add_boolean(ri->zri_new_zvols, dsname);
+	}
+
 	zcp_deregister_cleanup(state, zch);
 	fnvlist_free(ddsa.ddsa_snaps);
 

module/zfs/zfs_ioctl.c

@@ -3312,8 +3312,9 @@ zfs_ioc_create(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl)
 
 			/*
 			 * Volumes will return EBUSY and cannot be destroyed
-			 * until all asynchronous minor handling has completed.
-			 * Wait for the spa_zvol_taskq to drain then retry.
+			 * until all asynchronous minor handling (e.g. from
+			 * setting the volmode property) has completed. Wait for
+			 * the spa_zvol_taskq to drain then retry.
 			 */
 			error2 = dsl_destroy_head(fsname);
 			while ((error2 == EBUSY) && (type == DMU_OST_ZVOL)) {

module/zfs/zvol.c

@@ -97,7 +97,6 @@ krwlock_t zvol_state_lock;
 const zvol_platform_ops_t *ops;
 
 typedef enum {
-	ZVOL_ASYNC_CREATE_MINORS,
 	ZVOL_ASYNC_REMOVE_MINORS,
 	ZVOL_ASYNC_RENAME_MINORS,
 	ZVOL_ASYNC_SET_SNAPDEV,
@@ -1098,17 +1097,14 @@ zvol_create_minors_cb(const char *dsname, void *arg)
  * 'visible' (which also verifies that the parent is a zvol), and if so,
  * a minor node for that snapshot is created.
  */
-static int
-zvol_create_minors_impl(const char *name)
+void
+zvol_create_minors_recursive(const char *name)
 {
-	int error = 0;
-	fstrans_cookie_t cookie;
-	char *atp, *parent;
 	list_t minors_list;
 	minors_job_t *job;
 
 	if (zvol_inhibit_dev)
-		return (0);
+		return;
 
 	/*
 	 * This is the list for prefetch jobs. Whenever we found a match
@@ -1122,26 +1118,22 @@ zvol_create_minors_impl(const char *name)
 	list_create(&minors_list, sizeof (minors_job_t),
 	    offsetof(minors_job_t, link));
 
-	parent = kmem_alloc(MAXPATHLEN, KM_SLEEP);
-	(void) strlcpy(parent, name, MAXPATHLEN);
 
-	if ((atp = strrchr(parent, '@')) != NULL) {
+	if (strchr(name, '@') != NULL) {
 		uint64_t snapdev;
 
-		*atp = '\0';
-		error = dsl_prop_get_integer(parent, "snapdev",
+		int error = dsl_prop_get_integer(name, "snapdev",
 		    &snapdev, NULL);
 
 		if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
-			error = ops->zv_create_minor(name);
+			(void) ops->zv_create_minor(name);
 	} else {
-		cookie = spl_fstrans_mark();
-		error = dmu_objset_find(parent, zvol_create_minors_cb,
+		fstrans_cookie_t cookie = spl_fstrans_mark();
+		(void) dmu_objset_find(name, zvol_create_minors_cb,
 		    &minors_list, DS_FIND_CHILDREN);
 		spl_fstrans_unmark(cookie);
 	}
 
-	kmem_free(parent, MAXPATHLEN);
 	taskq_wait_outstanding(system_taskq, 0);
 
 	/*
@@ -1151,14 +1143,40 @@ zvol_create_minors_impl(const char *name)
 	while ((job = list_head(&minors_list)) != NULL) {
 		list_remove(&minors_list, job);
 		if (!job->error)
-			ops->zv_create_minor(job->name);
+			(void) ops->zv_create_minor(job->name);
 		kmem_strfree(job->name);
 		kmem_free(job, sizeof (minors_job_t));
 	}
 
 	list_destroy(&minors_list);
+}
 
-	return (SET_ERROR(error));
+void
+zvol_create_minor(const char *name)
+{
+	/*
+	 * Note: the dsl_pool_config_lock must not be held.
+	 * Minor node creation needs to obtain the zvol_state_lock.
+	 * zvol_open() obtains the zvol_state_lock and then the dsl pool
+	 * config lock.  Therefore, we can't have the config lock now if
+	 * we are going to wait for the zvol_state_lock, because it
+	 * would be a lock order inversion which could lead to deadlock.
+	 */
+
+	if (zvol_inhibit_dev)
+		return;
+
+	if (strchr(name, '@') != NULL) {
+		uint64_t snapdev;
+
+		int error = dsl_prop_get_integer(name,
+		    "snapdev", &snapdev, NULL);
+
+		if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
+			(void) ops->zv_create_minor(name);
+	} else {
+		(void) ops->zv_create_minor(name);
+	}
 }
 
 /*
@@ -1366,7 +1384,7 @@ zvol_set_volmode_impl(char *name, uint64_t volmode)
 	/*
 	 * It's unfortunate we need to remove minors before we create new ones:
 	 * this is necessary because our backing gendisk (zvol_state->zv_disk)
-	 * coule be different when we set, for instance, volmode from "geom"
+	 * could be different when we set, for instance, volmode from "geom"
 	 * to "dev" (or vice versa).
 	 * A possible optimization is to modify our consumers so we don't get
 	 * called when "volmode" does not change.
@@ -1426,14 +1444,11 @@ zvol_task_free(zvol_task_t *task)
  * The worker thread function performed asynchronously.
  */
 static void
-zvol_task_cb(void *param)
+zvol_task_cb(void *arg)
 {
-	zvol_task_t *task = (zvol_task_t *)param;
+	zvol_task_t *task = arg;
 
 	switch (task->op) {
-	case ZVOL_ASYNC_CREATE_MINORS:
-		(void) zvol_create_minors_impl(task->name1);
-		break;
 	case ZVOL_ASYNC_REMOVE_MINORS:
 		zvol_remove_minors_impl(task->name1);
 		break;
@@ -1634,21 +1649,6 @@ zvol_set_volmode(const char *ddname, zprop_source_t source, uint64_t volmode)
 	    zvol_set_volmode_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
 }
 
-void
-zvol_create_minors(spa_t *spa, const char *name, boolean_t async)
-{
-	zvol_task_t *task;
-	taskqid_t id;
-
-	task = zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS, name, NULL, ~0ULL);
-	if (task == NULL)
-		return;
-
-	id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
-	if ((async == B_FALSE) && (id != TASKQID_INVALID))
-		taskq_wait_id(spa->spa_zvol_taskq, id);
-}
-
 void
 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
 {