Illumos 5056 - ZFS deadlock on db_mtx and dn_holds

5056 ZFS deadlock on db_mtx and dn_holds
Author: Justin Gibbs <justing@spectralogic.com>
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>

References:
  https://www.illumos.org/issues/5056
  https://github.com/illumos/illumos-gate/commit/bc9014e

Porting Notes:

sa_handle_get_from_db():
  - the original patch includes an otherwise unmentioned fix for a
    possible usage of an uninitialised variable

dmu_objset_open_impl():
  - Under Illumos list_link_init() is the same as filling a list_node_t
    with NULLs, so they don't notice if they miss doing list_link_init()
    on a zero'd containing structure (e.g. allocated with kmem_zalloc as
    here). Under Linux, not so much: an uninitialised list_node_t goes
    "Boom!" some time later when it's used or destroyed.

dmu_objset_evict_dbufs():
  - reduce stack usage using kmem_alloc()

Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
This commit is contained in:
Justin T. Gibbs
2015-04-02 14:44:32 +11:00
committed by Brian Behlendorf
parent d683ddbb72
commit 0c66c32d1d
35 changed files with 645 additions and 316 deletions
+19 -1
View File
@@ -23,6 +23,7 @@
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
*/
/*
@@ -1095,6 +1096,8 @@ spa_activate(spa_t *spa, int mode)
list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
offsetof(vdev_t, vdev_config_dirty_node));
list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
offsetof(objset_t, os_evicting_node));
list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
offsetof(vdev_t, vdev_state_dirty_node));
@@ -1123,9 +1126,12 @@ spa_deactivate(spa_t *spa)
ASSERT(spa->spa_async_zio_root == NULL);
ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
spa_evicting_os_wait(spa);
txg_list_destroy(&spa->spa_vdev_txg_list);
list_destroy(&spa->spa_config_dirty_list);
list_destroy(&spa->spa_evicting_os_list);
list_destroy(&spa->spa_state_dirty_list);
taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
@@ -2136,6 +2142,11 @@ spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
mosconfig, &ereport);
}
/*
* Don't count references from objsets that are already closed
* and are making their way through the eviction process.
*/
spa_evicting_os_wait(spa);
spa->spa_minref = refcount_count(&spa->spa_refcount);
if (error) {
if (error != EEXIST) {
@@ -3775,6 +3786,11 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
spa_history_log_version(spa, "create");
/*
* Don't count references from objsets that are already closed
* and are making their way through the eviction process.
*/
spa_evicting_os_wait(spa);
spa->spa_minref = refcount_count(&spa->spa_refcount);
mutex_exit(&spa_namespace_lock);
@@ -4314,8 +4330,10 @@ spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
* modify its state. Objsets may be open only because they're dirty,
* so we have to force it to sync before checking spa_refcnt.
*/
if (spa->spa_sync_on)
if (spa->spa_sync_on) {
txg_wait_synced(spa->spa_dsl_pool, 0);
spa_evicting_os_wait(spa);
}
/*
* A pool cannot be exported or destroyed if there are active