Fix for ARC sysctls ignored at runtime

This change leverage module_param_call() to run arc_tuning_update()
immediately after the ARC tunable has been updated as suggested in
cffa8372 code review.

A simple test case is added to the ZFS Test Suite to prevent future
regressions in functionality.

Reviewed-by: Matt Macy <mmacy@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #9487  
Closes #9489
This commit is contained in:
loli10K 2019-10-27 00:22:19 +02:00 committed by Brian Behlendorf
parent 6963414d70
commit e35704647e
10 changed files with 178 additions and 42 deletions

View File

@ -76,26 +76,26 @@ enum scope_prefix_types {
/* /*
* Declare a module parameter / sysctl node * Declare a module parameter / sysctl node
* *
* scope_prefix the part of the the sysctl / sysfs tree the node resides under * "scope_prefix" the part of the the sysctl / sysfs tree the node resides under
* (currently a no-op on Linux) * (currently a no-op on Linux)
* name_prefix the part of the variable name that will be excluded from the * "name_prefix" the part of the variable name that will be excluded from the
* exported names on platforms with a hierarchical namespace * exported names on platforms with a hierarchical namespace
* name the part of the variable that will be exposed on platforms with a * "name" the part of the variable that will be exposed on platforms with a
* hierarchical namespace, or as name_prefix ## name on Linux * hierarchical namespace, or as name_prefix ## name on Linux
* type the variable type * "type" the variable type
* perm the permissions (read/write or read only) * "perm" the permissions (read/write or read only)
* desc a brief description of the option * "desc" a brief description of the option
* *
* Examples: * Examples:
* ZFS_MODULE_PARAM(zfs_vdev_mirror, zfs_vdev_mirror_, rotating_inc, UINT, * ZFS_MODULE_PARAM(zfs_vdev_mirror, zfs_vdev_mirror_, rotating_inc, UINT,
* ZMOD_RW, "Rotating media load increment for non-seeking I/O's"); * ZMOD_RW, "Rotating media load increment for non-seeking I/O's");
* on FreeBSD: * on FreeBSD:
* vfs.zfs.vdev.mirror.rotating_inc * vfs.zfs.vdev.mirror.rotating_inc
* on Linux: * on Linux:
* zfs_vdev_mirror_rotating_inc * zfs_vdev_mirror_rotating_inc
* *
* *ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, UINT, ZMOD_RW, * ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, UINT, ZMOD_RW,
* "Limit one prefetch call to this size"); * "Limit one prefetch call to this size");
* on FreeBSD: * on FreeBSD:
* vfs.zfs.dmu_prefetch_max * vfs.zfs.dmu_prefetch_max
* on Linux: * on Linux:
@ -108,5 +108,33 @@ enum scope_prefix_types {
MODULE_PARM_DESC(name_prefix ## name, desc) MODULE_PARM_DESC(name_prefix ## name, desc)
/* END CSTYLED */ /* END CSTYLED */
/*
* Declare a module parameter / sysctl node
*
* "scope_prefix" the part of the the sysctl / sysfs tree the node resides under
* (currently a no-op on Linux)
* "name_prefix" the part of the variable name that will be excluded from the
* exported names on platforms with a hierarchical namespace
* "name" the part of the variable that will be exposed on platforms with a
* hierarchical namespace, or as name_prefix ## name on Linux
* "setfunc" setter function
* "getfunc" getter function
* "perm" the permissions (read/write or read only)
* "desc" a brief description of the option
*
* Examples:
* ZFS_MODULE_PARAM_CALL(zfs_spa, spa_, slop_shift, param_set_slop_shift,
* param_get_int, ZMOD_RW, "Reserved free space in pool");
* on FreeBSD:
* vfs.zfs.spa_slop_shift
* on Linux:
* spa_slop_shift
*/
/* BEGIN CSTYLED */
#define ZFS_MODULE_PARAM_CALL(scope_prefix, name_prefix, name, setfunc, getfunc, perm, desc) \
CTASSERT_GLOBAL((sizeof (scope_prefix) == sizeof (enum scope_prefix_types))); \
module_param_call(name_prefix ## name, setfunc, getfunc, &name_prefix ## name, perm); \
MODULE_PARM_DESC(name_prefix ## name, desc)
/* END CSTYLED */
#endif /* _MOD_COMPAT_H */ #endif /* _MOD_COMPAT_H */

View File

@ -611,6 +611,10 @@ extern void arc_prune_async(int64_t);
extern int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg); extern int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg);
extern uint64_t arc_free_memory(void); extern uint64_t arc_free_memory(void);
extern int64_t arc_available_memory(void); extern int64_t arc_available_memory(void);
extern void arc_tuning_update(void);
extern int param_set_arc_long(const char *buf, zfs_kernel_param_t *kp);
extern int param_set_arc_int(const char *buf, zfs_kernel_param_t *kp);
#ifdef __cplusplus #ifdef __cplusplus
} }

View File

@ -198,7 +198,13 @@ extern int aok;
/* /*
* Tunables. * Tunables.
*/ */
typedef struct zfs_kernel_param {
const char *name; /* unused stub */
} zfs_kernel_param_t;
#define ZFS_MODULE_PARAM(scope_prefix, name_prefix, name, type, perm, desc) #define ZFS_MODULE_PARAM(scope_prefix, name_prefix, name, type, perm, desc)
#define ZFS_MODULE_PARAM_CALL(scope_prefix, name_prefix, name, setfunc, \
getfunc, perm, desc)
/* /*
* Exported symbols * Exported symbols

View File

@ -357,6 +357,34 @@ arc_lowmem_fini(void)
{ {
spl_unregister_shrinker(&arc_shrinker); spl_unregister_shrinker(&arc_shrinker);
} }
int
param_set_arc_long(const char *buf, zfs_kernel_param_t *kp)
{
int error;
error = param_set_long(buf, kp);
if (error < 0)
return (SET_ERROR(error));
arc_tuning_update();
return (0);
}
int
param_set_arc_int(const char *buf, zfs_kernel_param_t *kp)
{
int error;
error = param_set_int(buf, kp);
if (error < 0)
return (SET_ERROR(error));
arc_tuning_update();
return (0);
}
#else /* _KERNEL */ #else /* _KERNEL */
int64_t int64_t
arc_available_memory(void) arc_available_memory(void)

View File

@ -24,6 +24,7 @@
* Copyright (c) 2011, 2019 by Delphix. All rights reserved. * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
* Copyright (c) 2014 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 by Saso Kiselkov. All rights reserved.
* Copyright 2017 Nexenta Systems, Inc. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
*/ */
/* /*
@ -814,7 +815,6 @@ static void arc_hdr_alloc_abd(arc_buf_hdr_t *, boolean_t);
static void arc_access(arc_buf_hdr_t *, kmutex_t *); static void arc_access(arc_buf_hdr_t *, kmutex_t *);
static boolean_t arc_is_overflowing(void); static boolean_t arc_is_overflowing(void);
static void arc_buf_watch(arc_buf_t *); static void arc_buf_watch(arc_buf_t *);
static void arc_tuning_update(void);
static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *);
static uint32_t arc_bufc_to_flags(arc_buf_contents_t); static uint32_t arc_bufc_to_flags(arc_buf_contents_t);
@ -6873,10 +6873,12 @@ arc_state_multilist_index_func(multilist_t *ml, void *obj)
/* /*
* Called during module initialization and periodically thereafter to * Called during module initialization and periodically thereafter to
* apply reasonable changes to the exposed performance tunings. Non-zero * apply reasonable changes to the exposed performance tunings. Can also be
* zfs_* values which differ from the currently set values will be applied. * called explicitly by param_set_arc_*() functions when ARC tunables are
* updated manually. Non-zero zfs_* values which differ from the currently set
* values will be applied.
*/ */
static void void
arc_tuning_update(void) arc_tuning_update(void)
{ {
uint64_t allmem = arc_all_memory(); uint64_t allmem = arc_all_memory();
@ -8694,20 +8696,21 @@ EXPORT_SYMBOL(arc_add_prune_callback);
EXPORT_SYMBOL(arc_remove_prune_callback); EXPORT_SYMBOL(arc_remove_prune_callback);
/* BEGIN CSTYLED */ /* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, min, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min, param_set_arc_long,
"Min arc size"); param_get_long, ZMOD_RW, "Min arc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, max, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, max, param_set_arc_long,
"Max arc size"); param_get_long, ZMOD_RW, "Max arc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_limit, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit, param_set_arc_long,
"Metadata limit for arc size"); param_get_long, ZMOD_RW, "Metadata limit for arc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_limit_percent, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit_percent,
param_set_arc_long, param_get_long, ZMOD_RW,
"Percent of arc size for arc meta limit"); "Percent of arc size for arc meta limit");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_min, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_min, param_set_arc_long,
"Min arc metadata"); param_get_long, ZMOD_RW, "Min arc metadata");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_prune, INT, ZMOD_RW, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_prune, INT, ZMOD_RW,
"Meta objects to scan for prune"); "Meta objects to scan for prune");
@ -8718,20 +8721,20 @@ ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_adjust_restarts, INT, ZMOD_RW,
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_strategy, INT, ZMOD_RW, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_strategy, INT, ZMOD_RW,
"Meta reclaim strategy"); "Meta reclaim strategy");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, grow_retry, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, grow_retry, param_set_arc_int,
"Seconds before growing arc size"); param_get_int, ZMOD_RW, "Seconds before growing arc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, p_dampener_disable, INT, ZMOD_RW, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, p_dampener_disable, INT, ZMOD_RW,
"Disable arc_p adapt dampener"); "Disable arc_p adapt dampener");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, shrink_shift, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, shrink_shift, param_set_arc_int,
"log2(fraction of arc to reclaim)"); param_get_int, ZMOD_RW, "log2(fraction of arc to reclaim)");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, pc_percent, UINT, ZMOD_RW, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, pc_percent, UINT, ZMOD_RW,
"Percent of pagecache to reclaim arc to"); "Percent of pagecache to reclaim arc to");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, p_min_shift, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, p_min_shift, param_set_arc_int,
"arc_c shift to calc min/max arc_p"); param_get_int, ZMOD_RW, "arc_c shift to calc min/max arc_p");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, average_blocksize, INT, ZMOD_RD, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, average_blocksize, INT, ZMOD_RD,
"Target average block size"); "Target average block size");
@ -8739,10 +8742,11 @@ ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, average_blocksize, INT, ZMOD_RD,
ZFS_MODULE_PARAM(zfs, zfs_, compressed_arc_enabled, INT, ZMOD_RW, ZFS_MODULE_PARAM(zfs, zfs_, compressed_arc_enabled, INT, ZMOD_RW,
"Disable compressed arc buffers"); "Disable compressed arc buffers");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, min_prefetch_ms, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prefetch_ms, param_set_arc_int,
"Min life of prefetch block in ms"); param_get_int, ZMOD_RW, "Min life of prefetch block in ms");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, min_prescient_prefetch_ms, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prescient_prefetch_ms,
param_set_arc_int, param_get_int, ZMOD_RW,
"Min life of prescient prefetched block in ms"); "Min life of prescient prefetched block in ms");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_max, ULONG, ZMOD_RW, ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_max, ULONG, ZMOD_RW,
@ -8772,16 +8776,17 @@ ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_again, INT, ZMOD_RW,
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, norw, INT, ZMOD_RW, ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, norw, INT, ZMOD_RW,
"No reads during writes"); "No reads during writes");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, lotsfree_percent, INT, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, lotsfree_percent, param_set_arc_int,
"System free memory I/O throttle in bytes"); param_get_int, ZMOD_RW, "System free memory I/O throttle in bytes");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, sys_free, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, sys_free, param_set_arc_long,
"System free memory target size in bytes"); param_get_long, ZMOD_RW, "System free memory target size in bytes");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_limit, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit, param_set_arc_long,
"Minimum bytes of dnodes in arc"); param_get_long, ZMOD_RW, "Minimum bytes of dnodes in arc");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_limit_percent, ULONG, ZMOD_RW, ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit_percent,
param_set_arc_long, param_get_long, ZMOD_RW,
"Percent of ARC meta buffers for dnodes"); "Percent of ARC meta buffers for dnodes");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_reduce_percent, ULONG, ZMOD_RW, ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_reduce_percent, ULONG, ZMOD_RW,

View File

@ -26,6 +26,7 @@
* Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved.
* Copyright (c) 2017 Datto Inc. * Copyright (c) 2017 Datto Inc.
* Copyright (c) 2017, Intel Corporation. * Copyright (c) 2017, Intel Corporation.
* Copyright (c) 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
*/ */
#include <sys/zfs_context.h> #include <sys/zfs_context.h>
@ -2920,9 +2921,8 @@ module_param_call(zfs_deadman_ziotime_ms, param_set_deadman_ziotime,
MODULE_PARM_DESC(zfs_deadman_ziotime_ms, MODULE_PARM_DESC(zfs_deadman_ziotime_ms,
"IO expiration time in milliseconds"); "IO expiration time in milliseconds");
module_param_call(spa_slop_shift, param_set_slop_shift, param_get_int, ZFS_MODULE_PARAM_CALL(zfs_spa, spa_, slop_shift, param_set_slop_shift,
&spa_slop_shift, 0644); param_get_int, ZMOD_RW, "Reserved free space in pool");
MODULE_PARM_DESC(spa_slop_shift, "Reserved free space in pool");
module_param_call(zfs_deadman_failmode, param_set_deadman_failmode, module_param_call(zfs_deadman_failmode, param_set_deadman_failmode,
param_get_charp, &zfs_deadman_failmode, 0644); param_get_charp, &zfs_deadman_failmode, 0644);

View File

@ -25,7 +25,8 @@ tests = ['posix_001_pos', 'posix_002_pos', 'posix_003_pos']
tags = ['functional', 'acl', 'posix'] tags = ['functional', 'acl', 'posix']
[tests/functional/arc:Linux] [tests/functional/arc:Linux]
tests = ['dbufstats_001_pos', 'dbufstats_002_pos', 'dbufstats_003_pos'] tests = ['dbufstats_001_pos', 'dbufstats_002_pos', 'dbufstats_003_pos',
'arcstats_runtime_tuning']
tags = ['functional', 'arc'] tags = ['functional', 'arc']
[tests/functional/atime:Linux] [tests/functional/atime:Linux]

View File

@ -2,6 +2,7 @@ pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/tests/functional/arc
dist_pkgdata_SCRIPTS = \ dist_pkgdata_SCRIPTS = \
cleanup.ksh \ cleanup.ksh \
setup.ksh \ setup.ksh \
arcstats_runtime_tuning.ksh \
dbufstats_001_pos.ksh \ dbufstats_001_pos.ksh \
dbufstats_002_pos.ksh \ dbufstats_002_pos.ksh \
dbufstats_003_pos.ksh dbufstats_003_pos.ksh

View File

@ -0,0 +1,46 @@
#!/bin/ksh -p
#
# 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.
#
#
# Copyright 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
#
. $STF_SUITE/include/libtest.shlib
. $STF_SUITE/tests/perf/perf.shlib
function cleanup
{
# Set tunables to their recorded actual size and then to their original
# value: this works for previously unconfigured tunables.
log_must set_tunable64 zfs_arc_min "$MINSIZE"
log_must set_tunable64 zfs_arc_min "$ZFS_ARC_MIN"
log_must set_tunable64 zfs_arc_max "$MAXSIZE"
log_must set_tunable64 zfs_arc_max "$ZFS_ARC_MAX"
}
log_onexit cleanup
ZFS_ARC_MAX="$(get_tunable zfs_arc_max)"
ZFS_ARC_MIN="$(get_tunable zfs_arc_min)"
MINSIZE="$(get_min_arc_size)"
MAXSIZE="$(get_max_arc_size)"
log_assert "ARC tunables should be updated dynamically"
for size in $((MAXSIZE/4)) $((MAXSIZE/3)) $((MAXSIZE/2)) $MAXSIZE; do
log_must set_tunable64 zfs_arc_max "$size"
log_must test "$(get_max_arc_size)" == "$size"
log_must set_tunable64 zfs_arc_min "$size"
log_must test "$(get_min_arc_size)" == "$size"
done
log_pass "ARC tunables can be updated dynamically"

View File

@ -373,6 +373,23 @@ function get_directory
echo $directory echo $directory
} }
function get_min_arc_size
{
if is_linux; then
typeset -l min_arc_size=`awk '$1 == "c_min" { print $3 }' \
/proc/spl/kstat/zfs/arcstats`
else
typeset -l min_arc_size=$(dtrace -qn 'BEGIN {
printf("%u\n", `arc_stats.arcstat_c_min.value.ui64);
exit(0);
}')
fi
[[ $? -eq 0 ]] || log_fail "get_min_arc_size failed"
echo $min_arc_size
}
function get_max_arc_size function get_max_arc_size
{ {
if is_linux; then if is_linux; then