zed: detect and offline physically removed devices

This commit adds a new test case to the ZFS Test Suite to verify ZED
can detect when a device is physically removed from a running system:
the device will be offlined if a spare is not available in the pool.

We implement this by using the existing libudev functionality and
without relying solely on the FM kernel module capabilities which have
been observed to be unreliable with some kernels.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #1537
Closes #7926
This commit is contained in:
loli10K 2018-09-18 23:45:52 +02:00 committed by Brian Behlendorf
parent 13c59bb76b
commit d48091de81
11 changed files with 341 additions and 43 deletions

View File

@ -12,6 +12,7 @@
/*
* Copyright (c) 2016, Intel Corporation.
* Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>
*/
#include <libnvpair.h>
@ -53,13 +54,25 @@ pthread_t g_agents_tid;
libzfs_handle_t *g_zfs_hdl;
/* guid search data */
typedef enum device_type {
DEVICE_TYPE_L2ARC, /* l2arc device */
DEVICE_TYPE_SPARE, /* spare device */
DEVICE_TYPE_PRIMARY /* any primary pool storage device */
} device_type_t;
typedef struct guid_search {
uint64_t gs_pool_guid;
uint64_t gs_vdev_guid;
char *gs_devid;
device_type_t gs_vdev_type;
uint64_t gs_vdev_expandtime; /* vdev expansion time */
} guid_search_t;
static void
/*
* Walks the vdev tree recursively looking for a matching devid.
* Returns B_TRUE as soon as a matching device is found, B_FALSE otherwise.
*/
static boolean_t
zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
{
guid_search_t *gsp = arg;
@ -72,19 +85,47 @@ zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
*/
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN,
&child, &children) == 0) {
for (c = 0; c < children; c++)
zfs_agent_iter_vdev(zhp, child[c], gsp);
return;
for (c = 0; c < children; c++) {
if (zfs_agent_iter_vdev(zhp, child[c], gsp)) {
gsp->gs_vdev_type = DEVICE_TYPE_PRIMARY;
return (B_TRUE);
}
}
}
/*
* On a devid match, grab the vdev guid
* Iterate over any spares and cache devices
*/
if ((gsp->gs_vdev_guid == 0) &&
(nvlist_lookup_string(nvl, ZPOOL_CONFIG_DEVID, &path) == 0) &&
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_SPARES,
&child, &children) == 0) {
for (c = 0; c < children; c++) {
if (zfs_agent_iter_vdev(zhp, child[c], gsp)) {
gsp->gs_vdev_type = DEVICE_TYPE_L2ARC;
return (B_TRUE);
}
}
}
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_L2CACHE,
&child, &children) == 0) {
for (c = 0; c < children; c++) {
if (zfs_agent_iter_vdev(zhp, child[c], gsp)) {
gsp->gs_vdev_type = DEVICE_TYPE_SPARE;
return (B_TRUE);
}
}
}
/*
* On a devid match, grab the vdev guid and expansion time, if any.
*/
if ((nvlist_lookup_string(nvl, ZPOOL_CONFIG_DEVID, &path) == 0) &&
(strcmp(gsp->gs_devid, path) == 0)) {
(void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID,
&gsp->gs_vdev_guid);
(void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_EXPANSION_TIME,
&gsp->gs_vdev_expandtime);
return (B_TRUE);
}
return (B_FALSE);
}
static int
@ -99,7 +140,7 @@ zfs_agent_iter_pool(zpool_handle_t *zhp, void *arg)
if ((config = zpool_get_config(zhp, NULL)) != NULL) {
if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
&nvl) == 0) {
zfs_agent_iter_vdev(zhp, nvl, gsp);
(void) zfs_agent_iter_vdev(zhp, nvl, gsp);
}
}
/*
@ -148,6 +189,8 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
struct timeval tv;
int64_t tod[2];
uint64_t pool_guid = 0, vdev_guid = 0;
guid_search_t search = { 0 };
device_type_t devtype = DEVICE_TYPE_PRIMARY;
class = "resource.fs.zfs.removed";
subclass = "";
@ -156,30 +199,55 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
(void) nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, &pool_guid);
(void) nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &vdev_guid);
(void) gettimeofday(&tv, NULL);
tod[0] = tv.tv_sec;
tod[1] = tv.tv_usec;
(void) nvlist_add_int64_array(payload, FM_EREPORT_TIME, tod, 2);
/*
* For multipath, ZFS_EV_VDEV_GUID is missing so find it.
* For multipath, spare and l2arc devices ZFS_EV_VDEV_GUID or
* ZFS_EV_POOL_GUID may be missing so find them.
*/
if (vdev_guid == 0) {
guid_search_t search = { 0 };
(void) nvlist_lookup_string(nvl, DEV_IDENTIFIER,
&search.gs_devid);
(void) zpool_iter(g_zfs_hdl, zfs_agent_iter_pool, &search);
pool_guid = search.gs_pool_guid;
vdev_guid = search.gs_vdev_guid;
devtype = search.gs_vdev_type;
(void) nvlist_lookup_string(nvl, DEV_IDENTIFIER,
&search.gs_devid);
(void) zpool_iter(g_zfs_hdl, zfs_agent_iter_pool,
&search);
pool_guid = search.gs_pool_guid;
vdev_guid = search.gs_vdev_guid;
/*
* We want to avoid reporting "remove" events coming from
* libudev for VDEVs which were expanded recently (10s) and
* avoid activating spares in response to partitions being
* deleted and created in rapid succession.
*/
if (search.gs_vdev_expandtime != 0 &&
search.gs_vdev_expandtime + 10 > tv.tv_sec) {
zed_log_msg(LOG_INFO, "agent post event: ignoring '%s' "
"for recently expanded device '%s'", EC_DEV_REMOVE,
search.gs_devid);
goto out;
}
(void) nvlist_add_uint64(payload,
FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, pool_guid);
(void) nvlist_add_uint64(payload,
FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vdev_guid);
(void) gettimeofday(&tv, NULL);
tod[0] = tv.tv_sec;
tod[1] = tv.tv_usec;
(void) nvlist_add_int64_array(payload, FM_EREPORT_TIME, tod, 2);
switch (devtype) {
case DEVICE_TYPE_L2ARC:
(void) nvlist_add_string(payload,
FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
VDEV_TYPE_L2CACHE);
break;
case DEVICE_TYPE_SPARE:
(void) nvlist_add_string(payload,
FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, VDEV_TYPE_SPARE);
break;
case DEVICE_TYPE_PRIMARY:
(void) nvlist_add_string(payload,
FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, VDEV_TYPE_DISK);
break;
}
zed_log_msg(LOG_INFO, "agent post event: mapping '%s' to '%s'",
EC_DEV_REMOVE, class);
@ -193,6 +261,7 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
list_insert_tail(&agent_events, event);
(void) pthread_mutex_unlock(&agent_lock);
out:
(void) pthread_cond_signal(&agent_cond);
}

View File

@ -476,7 +476,20 @@ zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data)
&child, &children) == 0) {
for (c = 0; c < children; c++)
zfs_iter_vdev(zhp, child[c], data);
return;
}
/*
* Iterate over any spares and cache devices
*/
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_SPARES,
&child, &children) == 0) {
for (c = 0; c < children; c++)
zfs_iter_vdev(zhp, child[c], data);
}
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_L2CACHE,
&child, &children) == 0) {
for (c = 0; c < children; c++)
zfs_iter_vdev(zhp, child[c], data);
}
/* once a vdev was matched and processed there is nothing left to do */

View File

@ -22,6 +22,7 @@
* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
*
* Copyright (c) 2016, Intel Corporation.
* Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>
*/
/*
@ -126,6 +127,15 @@ find_vdev(libzfs_handle_t *zhdl, nvlist_t *nv, uint64_t search_guid)
return (ret);
}
if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
&child, &children) != 0)
return (NULL);
for (c = 0; c < children; c++) {
if ((ret = find_vdev(zhdl, child[c], search_guid)) != NULL)
return (ret);
}
return (NULL);
}
@ -167,9 +177,10 @@ find_by_guid(libzfs_handle_t *zhdl, uint64_t pool_guid, uint64_t vdev_guid,
/*
* Given a vdev, attempt to replace it with every known spare until one
* succeeds.
* succeeds or we run out of devices to try.
* Return whether we were successful or not in replacing the device.
*/
static void
static boolean_t
replace_with_spare(fmd_hdl_t *hdl, zpool_handle_t *zhp, nvlist_t *vdev)
{
nvlist_t *config, *nvroot, *replacement;
@ -182,14 +193,14 @@ replace_with_spare(fmd_hdl_t *hdl, zpool_handle_t *zhp, nvlist_t *vdev)
config = zpool_get_config(zhp, NULL);
if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
&nvroot) != 0)
return;
return (B_FALSE);
/*
* Find out if there are any hot spares available in the pool.
*/
if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
&spares, &nspares) != 0)
return;
return (B_FALSE);
/*
* lookup "ashift" pool property, we may need it for the replacement
@ -226,12 +237,17 @@ replace_with_spare(fmd_hdl_t *hdl, zpool_handle_t *zhp, nvlist_t *vdev)
dev_name, basename(spare_name));
if (zpool_vdev_attach(zhp, dev_name, spare_name,
replacement, B_TRUE) == 0)
break;
replacement, B_TRUE) == 0) {
free(dev_name);
nvlist_free(replacement);
return (B_TRUE);
}
}
free(dev_name);
nvlist_free(replacement);
return (B_FALSE);
}
/*
@ -304,10 +320,14 @@ zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
fmd_hdl_debug(hdl, "zfs_retire_recv: '%s'", class);
/*
* If this is a resource notifying us of device removal, then simply
* check for an available spare and continue.
* If this is a resource notifying us of device removal then simply
* check for an available spare and continue unless the device is a
* l2arc vdev, in which case we just offline it.
*/
if (strcmp(class, "resource.fs.zfs.removed") == 0) {
char *devtype;
char *devname;
if (nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
&pool_guid) != 0 ||
nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
@ -318,8 +338,21 @@ zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
&vdev)) == NULL)
return;
if (fmd_prop_get_int32(hdl, "spare_on_remove"))
replace_with_spare(hdl, zhp, vdev);
devname = zpool_vdev_name(NULL, zhp, vdev, B_FALSE);
/* Can't replace l2arc with a spare: offline the device */
if (nvlist_lookup_string(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
&devtype) == 0 && strcmp(devtype, VDEV_TYPE_L2CACHE) == 0) {
fmd_hdl_debug(hdl, "zpool_vdev_offline '%s'", devname);
zpool_vdev_offline(zhp, devname, B_TRUE);
} else if (!fmd_prop_get_int32(hdl, "spare_on_remove") ||
replace_with_spare(hdl, zhp, vdev) == B_FALSE) {
/* Could not handle with spare: offline the device */
fmd_hdl_debug(hdl, "zpool_vdev_offline '%s'", devname);
zpool_vdev_offline(zhp, devname, B_TRUE);
}
free(devname);
zpool_close(zhp);
return;
}
@ -463,7 +496,7 @@ zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
/*
* Attempt to substitute a hot spare.
*/
replace_with_spare(hdl, zhp, vdev);
(void) replace_with_spare(hdl, zhp, vdev);
zpool_close(zhp);
}

View File

@ -720,6 +720,7 @@ typedef struct zpool_load_policy {
#define ZPOOL_CONFIG_MMP_HOSTNAME "mmp_hostname" /* not stored on disk */
#define ZPOOL_CONFIG_MMP_HOSTID "mmp_hostid" /* not stored on disk */
#define ZPOOL_CONFIG_ALLOCATION_BIAS "alloc_bias" /* not stored on disk */
#define ZPOOL_CONFIG_EXPANSION_TIME "expansion_time" /* not stored */
/*
* The persistent vdev state is stored as separate values rather than a single

View File

@ -344,6 +344,7 @@ struct vdev {
uint64_t vdev_leaf_zap;
hrtime_t vdev_mmp_pending; /* 0 if write finished */
uint64_t vdev_mmp_kstat_id; /* to find kstat entry */
uint64_t vdev_expansion_time; /* vdev's last expansion time */
/*
* For DTrace to work in userland (libzpool) context, these fields must

View File

@ -3424,6 +3424,7 @@ vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
spa->spa_autoexpand);
vd->vdev_expansion_time = gethrestime_sec();
}
vdev_reopen(tvd);

View File

@ -515,6 +515,10 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
if (vd->vdev_crtxg)
fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
if (vd->vdev_expansion_time)
fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
vd->vdev_expansion_time);
if (flags & VDEV_CONFIG_MOS) {
if (vd->vdev_leaf_zap != 0) {
ASSERT(vd->vdev_ops->vdev_op_leaf);

View File

@ -543,10 +543,10 @@ tests = ['exec_001_pos', 'exec_002_neg']
tags = ['functional', 'exec']
[tests/functional/fault]
tests = ['auto_online_001_pos', 'auto_replace_001_pos', 'auto_spare_001_pos',
'auto_spare_002_pos', 'auto_spare_ashift', 'auto_spare_multiple',
'auto_spare_shared', 'scrub_after_resilver', 'decrypt_fault',
'decompress_fault','zpool_status_-s']
tests = ['auto_offline_001_pos', 'auto_online_001_pos', 'auto_replace_001_pos',
'auto_spare_001_pos', 'auto_spare_002_pos', 'auto_spare_ashift',
'auto_spare_multiple', 'auto_spare_shared', 'scrub_after_resilver',
'decrypt_fault', 'decompress_fault', 'zpool_status_-s']
tags = ['functional', 'fault']
[tests/functional/features/async_destroy]

View File

@ -96,8 +96,7 @@ for type in " " mirror raidz raidz2; do
fi
typeset prev_size=$(get_pool_prop size $TESTPOOL1)
typeset zfs_prev_size=$(zfs get -p avail $TESTPOOL1 | tail -1 | \
awk '{print $3}')
typeset zfs_prev_size=$(get_prop avail $TESTPOOL1)
# Expand each device as appropriate being careful to add an artificial
# delay to ensure we get a single history entry for each. This makes
@ -117,8 +116,7 @@ for type in " " mirror raidz raidz2; do
log_must zpool online -e $TESTPOOL1 $FILE_RAW
typeset expand_size=$(get_pool_prop size $TESTPOOL1)
typeset zfs_expand_size=$(zfs get -p avail $TESTPOOL1 | tail -1 | \
awk '{print $3}')
typeset zfs_expand_size=$(get_prop avail $TESTPOOL1)
log_note "$TESTPOOL1 $type has previous size: $prev_size and " \
"expanded size: $expand_size"

View File

@ -2,6 +2,7 @@ pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/tests/functional/fault
dist_pkgdata_SCRIPTS = \
setup.ksh \
cleanup.ksh \
auto_offline_001_pos.ksh \
auto_online_001_pos.ksh \
auto_replace_001_pos.ksh \
auto_spare_001_pos.ksh \

View File

@ -0,0 +1,177 @@
#!/bin/ksh -p
#
# CDDL HEADER START
#
# This file and its contents are supplied under the terms of the
# Common Development and Distribution License ("CDDL"), version 1.0.
# You may only use this file in accordance with the terms of version
# 1.0 of the CDDL.
#
# A full copy of the text of the CDDL should have accompanied this
# source. A copy of the CDDL is also available via the Internet at
# http://www.illumos.org/license/CDDL.
#
# CDDL HEADER END
#
#
# Copyright 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
#
. $STF_SUITE/include/libtest.shlib
. $STF_SUITE/tests/functional/events/events_common.kshlib
. $STF_SUITE/tests/functional/fault/fault.cfg
#
# DESCRIPTION:
# Testing Fault Management Agent ZED Logic - Physically removed device is
# offlined and onlined when reattached
#
# STRATEGY:
# 1. Create a pool
# 2. Simulate physical removal of one device
# 3. Verify the device is offlined
# 4. Reattach the device
# 5. Verify the device is onlined
# 6. Repeat the same tests with a spare device: zed will use the spare to handle
# the removed data device
# 7. Repeat the same tests again with a faulted spare device: zed should offline
# the removed data device if no spare is available
#
# NOTE: the use of 'block_device_wait' throughout the test helps avoid race
# conditions caused by mixing creation/removal events from partitioning the
# disk (zpool create) and events from physically removing it (remove_disk).
#
verify_runnable "both"
if is_linux; then
# Add one 512b scsi_debug device (4Kn would generate IO errors)
# NOTE: must be larger than other "file" vdevs and minimum SPA devsize:
# add 32m of fudge
load_scsi_debug $(($SPA_MINDEVSIZE/1024/1024+32)) 1 1 1 '512b'
else
log_unsupported "scsi debug module unsupported"
fi
function cleanup
{
destroy_pool $TESTPOOL
rm -f $filedev1
rm -f $filedev2
rm -f $filedev3
rm -f $sparedev
unload_scsi_debug
}
log_assert "ZED detects physically removed devices"
log_onexit cleanup
filedev1="$TEST_BASE_DIR/file-vdev-1"
filedev2="$TEST_BASE_DIR/file-vdev-2"
filedev3="$TEST_BASE_DIR/file-vdev-3"
sparedev="$TEST_BASE_DIR/file-vdev-spare"
removedev=$(get_debug_device)
typeset poolconfs=("mirror $filedev1 $removedev"
"raidz $filedev1 $removedev"
"raidz2 $filedev1 $filedev2 $removedev"
"raidz3 $filedev1 $filedev2 $filedev3 $removedev"
"$filedev1 cache $removedev"
"mirror $filedev1 $filedev2 cache $removedev"
"raidz $filedev1 $filedev2 $filedev3 cache $removedev"
)
log_must truncate -s $SPA_MINDEVSIZE $filedev1
log_must truncate -s $SPA_MINDEVSIZE $filedev2
log_must truncate -s $SPA_MINDEVSIZE $filedev3
log_must truncate -s $SPA_MINDEVSIZE $sparedev
for conf in "${poolconfs[@]}"
do
# 1. Create a pool
log_must zpool create -f $TESTPOOL $conf
block_device_wait
# 2. Simulate physical removal of one device
remove_disk $removedev
# 3. Verify the device is offlined
log_must wait_vdev_state $TESTPOOL $removedev "OFFLINE"
# 4. Reattach the device
insert_disk $removedev
# 5. Verify the device is onlined
log_must wait_vdev_state $TESTPOOL $removedev "ONLINE"
# cleanup
destroy_pool $TESTPOOL
log_must parted "/dev/${removedev}" -s -- mklabel msdos
block_device_wait
done
# 6. Repeat the same tests with a spare device: zed will use the spare to handle
# the removed data device
for conf in "${poolconfs[@]}"
do
# 1. Create a pool with a spare
log_must zpool create -f $TESTPOOL $conf
block_device_wait
log_must zpool add $TESTPOOL spare $sparedev
# 3. Simulate physical removal of one device
remove_disk $removedev
# 4. Verify the device is handled by the spare unless is a l2arc disk
# which can only be offlined
if [[ $(echo "$conf" | grep -c 'cache') -eq 0 ]]; then
log_must wait_hotspare_state $TESTPOOL $sparedev "INUSE"
else
log_must wait_vdev_state $TESTPOOL $removedev "OFFLINE"
fi
# 5. Reattach the device
insert_disk $removedev
# 6. Verify the device is onlined
log_must wait_vdev_state $TESTPOOL $removedev "ONLINE"
# cleanup
destroy_pool $TESTPOOL
log_must parted "/dev/${removedev}" -s -- mklabel msdos
block_device_wait
done
# 7. Repeat the same tests again with a faulted spare device: zed should offline
# the removed data device if no spare is available
for conf in "${poolconfs[@]}"
do
# 1. Create a pool with a spare
log_must zpool create -f $TESTPOOL $conf
block_device_wait
log_must zpool add $TESTPOOL spare $sparedev
# 2. Fault the spare device making it unavailable
log_must zpool offline -f $TESTPOOL $sparedev
log_must wait_hotspare_state $TESTPOOL $sparedev "FAULTED"
# 3. Simulate physical removal of one device
remove_disk $removedev
# 4. Verify the device is offlined
log_must wait_vdev_state $TESTPOOL $removedev "OFFLINE"
# 5. Reattach the device
insert_disk $removedev
# 6. Verify the device is onlined
log_must wait_vdev_state $TESTPOOL $removedev "ONLINE"
# cleanup
destroy_pool $TESTPOOL
log_must parted "/dev/${removedev}" -s -- mklabel msdos
block_device_wait
done
log_pass "ZED detects physically removed devices"