c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-12-26 19:19:32 +03:00
Code Issues Packages Projects Releases Wiki Activity

Illumos #4756 Fix metaslab_group_preload deadlock

Browse Source 
4756 metaslab_group_preload() could deadlock
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Approved by: Garrett D'Amore <garrett@damore.org>

The metaslab_group_preload() function grabs the mg_lock and then later
tries to grab the metaslab lock. This lock ordering may lead to a
deadlock since other consumers of the mg_lock will grab the metaslab
lock first.

References:
  https://www.illumos.org/issues/4756
  https://github.com/illumos/illumos-gate/commit/30beaff

Ported-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2488
This commit is contained in:
George Wilson 2014-06-12 15:29:11 -08:00 committed by Brian Behlendorf
parent 3c51c5cb1f
commit 080b310015
2 changed files with 25 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
cmd/ztest/ztest.c
View File

@ -20,7 +20,7 @@
*/ */
/* /*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved. * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved. * Copyright (c) 2013 Steven Hartland. All rights reserved.
*/ */
@ -187,6 +187,7 @@ static const ztest_shared_opts_t ztest_opts_defaults = {
extern uint64_t metaslab_gang_bang; extern uint64_t metaslab_gang_bang;
extern uint64_t metaslab_df_alloc_threshold; extern uint64_t metaslab_df_alloc_threshold;
extern int metaslab_preload_limit;
static ztest_shared_opts_t *ztest_shared_opts; static ztest_shared_opts_t *ztest_shared_opts;
static ztest_shared_opts_t ztest_opts; static ztest_shared_opts_t ztest_opts;
@ -5734,6 +5735,7 @@ ztest_run(ztest_shared_t *zs)
kernel_init(FREAD | FWRITE); kernel_init(FREAD | FWRITE);
VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG)); VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
spa->spa_debug = B_TRUE; spa->spa_debug = B_TRUE;
metaslab_preload_limit = ztest_random(20) + 1;
ztest_spa = spa; ztest_spa = spa;
VERIFY0(dmu_objset_own(ztest_opts.zo_pool, VERIFY0(dmu_objset_own(ztest_opts.zo_pool,

25
module/zfs/metaslab.c
View File

@ -1242,6 +1242,8 @@ metaslab_preload(void *arg)
metaslab_t *msp = arg; metaslab_t *msp = arg;
spa_t *spa = msp->ms_group->mg_vd->vdev_spa; spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock));
mutex_enter(&msp->ms_lock); mutex_enter(&msp->ms_lock);
metaslab_load_wait(msp); metaslab_load_wait(msp);
if (!msp->ms_loaded) if (!msp->ms_loaded)
@ -1266,19 +1268,36 @@ metaslab_group_preload(metaslab_group_t *mg)
taskq_wait(mg->mg_taskq); taskq_wait(mg->mg_taskq);
return; return;
} }
mutex_enter(&mg->mg_lock);
mutex_enter(&mg->mg_lock);
/* /*
* Prefetch the next potential metaslabs * Load the next potential metaslabs
*/ */
for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) { msp = avl_first(t);
while (msp != NULL) {
metaslab_t *msp_next = AVL_NEXT(t, msp);
/* If we have reached our preload limit then we're done */ /* If we have reached our preload limit then we're done */
if (++m > metaslab_preload_limit) if (++m > metaslab_preload_limit)
break; break;
/*
* We must drop the metaslab group lock here to preserve
* lock ordering with the ms_lock (when grabbing both
* the mg_lock and the ms_lock, the ms_lock must be taken
* first). As a result, it is possible that the ordering
* of the metaslabs within the avl tree may change before
* we reacquire the lock. The metaslab cannot be removed from
* the tree while we're in syncing context so it is safe to
* drop the mg_lock here. If the metaslabs are reordered
* nothing will break -- we just may end up loading a
* less than optimal one.
*/
mutex_exit(&mg->mg_lock);
VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload, VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload,
msp, TQ_PUSHPAGE) != 0); msp, TQ_PUSHPAGE) != 0);
mutex_enter(&mg->mg_lock);
msp = msp_next;
} }
mutex_exit(&mg->mg_lock); mutex_exit(&mg->mg_lock);
} }