Unverified Commit c4c162c1 authored by Alexander Motin's avatar Alexander Motin Committed by GitHub
Browse files

Use wmsum for arc, abd, dbuf and zfetch statistics. (#12172)



wmsum was designed exactly for cases like these with many updates
and rare reads.  It allows to completely avoid atomic operations on
congested global variables.
Reviewed-by: default avatarBrian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: default avatarMark Maybee <mark.maybee@delphix.com>
Signed-off-by: default avatarAlexander Motin <mav@FreeBSD.org>
Sponsored-By: iXsystems, Inc.
Closes #12172
parent 9ffcaa37
......@@ -27,6 +27,7 @@
#define _ABD_IMPL_H
#include <sys/abd.h>
#include <sys/wmsum.h>
#ifdef __cplusplus
extern "C" {
......@@ -82,9 +83,8 @@ void abd_iter_unmap(struct abd_iter *);
/*
* Helper macros
*/
#define ABDSTAT(stat) (abd_stats.stat.value.ui64)
#define ABDSTAT_INCR(stat, val) \
atomic_add_64(&abd_stats.stat.value.ui64, (val))
wmsum_add(&abd_sums.stat, (val))
#define ABDSTAT_BUMP(stat) ABDSTAT_INCR(stat, 1)
#define ABDSTAT_BUMPDOWN(stat) ABDSTAT_INCR(stat, -1)
......
......@@ -33,6 +33,7 @@
#include <sys/zio_crypt.h>
#include <sys/zthr.h>
#include <sys/aggsum.h>
#include <sys/wmsum.h>
#ifdef __cplusplus
extern "C" {
......@@ -563,7 +564,6 @@ typedef struct arc_stats {
kstat_named_t arcstat_c;
kstat_named_t arcstat_c_min;
kstat_named_t arcstat_c_max;
/* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_size;
/*
* Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
......@@ -592,14 +592,12 @@ typedef struct arc_stats {
* (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
* caches), and arc_buf_t structures (allocated via arc_buf_t
* cache).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_hdr_size;
/*
* Number of bytes consumed by ARC buffers of type equal to
* ARC_BUFC_DATA. This is generally consumed by buffers backing
* on disk user data (e.g. plain file contents).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_data_size;
/*
......@@ -607,22 +605,18 @@ typedef struct arc_stats {
* ARC_BUFC_METADATA. This is generally consumed by buffers
* backing on disk data that is used for internal ZFS
* structures (e.g. ZAP, dnode, indirect blocks, etc).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_metadata_size;
/*
* Number of bytes consumed by dmu_buf_impl_t objects.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_dbuf_size;
/*
* Number of bytes consumed by dnode_t objects.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_dnode_size;
/*
* Number of bytes consumed by bonus buffers.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_bonus_size;
#if defined(COMPAT_FREEBSD11)
......@@ -637,7 +631,6 @@ typedef struct arc_stats {
* arc_anon state. This includes *all* buffers in the arc_anon
* state; e.g. data, metadata, evictable, and unevictable buffers
* are all included in this value.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_size;
/*
......@@ -645,7 +638,6 @@ typedef struct arc_stats {
* following criteria: backing buffers of type ARC_BUFC_DATA,
* residing in the arc_anon state, and are eligible for eviction
* (e.g. have no outstanding holds on the buffer).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_evictable_data;
/*
......@@ -653,7 +645,6 @@ typedef struct arc_stats {
* following criteria: backing buffers of type ARC_BUFC_METADATA,
* residing in the arc_anon state, and are eligible for eviction
* (e.g. have no outstanding holds on the buffer).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_anon_evictable_metadata;
/*
......@@ -661,7 +652,6 @@ typedef struct arc_stats {
* arc_mru state. This includes *all* buffers in the arc_mru
* state; e.g. data, metadata, evictable, and unevictable buffers
* are all included in this value.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_size;
/*
......@@ -669,7 +659,6 @@ typedef struct arc_stats {
* following criteria: backing buffers of type ARC_BUFC_DATA,
* residing in the arc_mru state, and are eligible for eviction
* (e.g. have no outstanding holds on the buffer).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_evictable_data;
/*
......@@ -677,7 +666,6 @@ typedef struct arc_stats {
* following criteria: backing buffers of type ARC_BUFC_METADATA,
* residing in the arc_mru state, and are eligible for eviction
* (e.g. have no outstanding holds on the buffer).
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_evictable_metadata;
/*
......@@ -688,21 +676,18 @@ typedef struct arc_stats {
* don't actually have ARC buffers linked off of these headers.
* Thus, *if* the headers had associated ARC buffers, these
* buffers *would have* consumed this number of bytes.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_ghost_size;
/*
* Number of bytes that *would have been* consumed by ARC
* buffers that are eligible for eviction, of type
* ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_ghost_evictable_data;
/*
* Number of bytes that *would have been* consumed by ARC
* buffers that are eligible for eviction, of type
* ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mru_ghost_evictable_metadata;
/*
......@@ -710,42 +695,36 @@ typedef struct arc_stats {
* arc_mfu state. This includes *all* buffers in the arc_mfu
* state; e.g. data, metadata, evictable, and unevictable buffers
* are all included in this value.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_size;
/*
* Number of bytes consumed by ARC buffers that are eligible for
* eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
* state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_evictable_data;
/*
* Number of bytes consumed by ARC buffers that are eligible for
* eviction, of type ARC_BUFC_METADATA, and reside in the
* arc_mfu state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_evictable_metadata;
/*
* Total number of bytes that *would have been* consumed by ARC
* buffers in the arc_mfu_ghost state. See the comment above
* arcstat_mru_ghost_size for more details.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_ghost_size;
/*
* Number of bytes that *would have been* consumed by ARC
* buffers that are eligible for eviction, of type
* ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_ghost_evictable_data;
/*
* Number of bytes that *would have been* consumed by ARC
* buffers that are eligible for eviction, of type
* ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
* Not updated directly; only synced in arc_kstat_update.
*/
kstat_named_t arcstat_mfu_ghost_evictable_metadata;
kstat_named_t arcstat_l2_hits;
......@@ -779,7 +758,6 @@ typedef struct arc_stats {
kstat_named_t arcstat_l2_io_error;
kstat_named_t arcstat_l2_lsize;
kstat_named_t arcstat_l2_psize;
/* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_l2_hdr_size;
/*
* Number of L2ARC log blocks written. These are used for restoring the
......@@ -860,7 +838,6 @@ typedef struct arc_stats {
kstat_named_t arcstat_tempreserve;
kstat_named_t arcstat_loaned_bytes;
kstat_named_t arcstat_prune;
/* Not updated directly; only synced in arc_kstat_update. */
kstat_named_t arcstat_meta_used;
kstat_named_t arcstat_meta_limit;
kstat_named_t arcstat_dnode_limit;
......@@ -876,6 +853,96 @@ typedef struct arc_stats {
kstat_named_t arcstat_abd_chunk_waste_size;
} arc_stats_t;
typedef struct arc_sums {
wmsum_t arcstat_hits;
wmsum_t arcstat_misses;
wmsum_t arcstat_demand_data_hits;
wmsum_t arcstat_demand_data_misses;
wmsum_t arcstat_demand_metadata_hits;
wmsum_t arcstat_demand_metadata_misses;
wmsum_t arcstat_prefetch_data_hits;
wmsum_t arcstat_prefetch_data_misses;
wmsum_t arcstat_prefetch_metadata_hits;
wmsum_t arcstat_prefetch_metadata_misses;
wmsum_t arcstat_mru_hits;
wmsum_t arcstat_mru_ghost_hits;
wmsum_t arcstat_mfu_hits;
wmsum_t arcstat_mfu_ghost_hits;
wmsum_t arcstat_deleted;
wmsum_t arcstat_mutex_miss;
wmsum_t arcstat_access_skip;
wmsum_t arcstat_evict_skip;
wmsum_t arcstat_evict_not_enough;
wmsum_t arcstat_evict_l2_cached;
wmsum_t arcstat_evict_l2_eligible;
wmsum_t arcstat_evict_l2_eligible_mfu;
wmsum_t arcstat_evict_l2_eligible_mru;
wmsum_t arcstat_evict_l2_ineligible;
wmsum_t arcstat_evict_l2_skip;
wmsum_t arcstat_hash_collisions;
wmsum_t arcstat_hash_chains;
aggsum_t arcstat_size;
wmsum_t arcstat_compressed_size;
wmsum_t arcstat_uncompressed_size;
wmsum_t arcstat_overhead_size;
wmsum_t arcstat_hdr_size;
wmsum_t arcstat_data_size;
wmsum_t arcstat_metadata_size;
wmsum_t arcstat_dbuf_size;
aggsum_t arcstat_dnode_size;
wmsum_t arcstat_bonus_size;
wmsum_t arcstat_l2_hits;
wmsum_t arcstat_l2_misses;
wmsum_t arcstat_l2_prefetch_asize;
wmsum_t arcstat_l2_mru_asize;
wmsum_t arcstat_l2_mfu_asize;
wmsum_t arcstat_l2_bufc_data_asize;
wmsum_t arcstat_l2_bufc_metadata_asize;
wmsum_t arcstat_l2_feeds;
wmsum_t arcstat_l2_rw_clash;
wmsum_t arcstat_l2_read_bytes;
wmsum_t arcstat_l2_write_bytes;
wmsum_t arcstat_l2_writes_sent;
wmsum_t arcstat_l2_writes_done;
wmsum_t arcstat_l2_writes_error;
wmsum_t arcstat_l2_writes_lock_retry;
wmsum_t arcstat_l2_evict_lock_retry;
wmsum_t arcstat_l2_evict_reading;
wmsum_t arcstat_l2_evict_l1cached;
wmsum_t arcstat_l2_free_on_write;
wmsum_t arcstat_l2_abort_lowmem;
wmsum_t arcstat_l2_cksum_bad;
wmsum_t arcstat_l2_io_error;
wmsum_t arcstat_l2_lsize;
wmsum_t arcstat_l2_psize;
aggsum_t arcstat_l2_hdr_size;
wmsum_t arcstat_l2_log_blk_writes;
wmsum_t arcstat_l2_log_blk_asize;
wmsum_t arcstat_l2_log_blk_count;
wmsum_t arcstat_l2_rebuild_success;
wmsum_t arcstat_l2_rebuild_abort_unsupported;
wmsum_t arcstat_l2_rebuild_abort_io_errors;
wmsum_t arcstat_l2_rebuild_abort_dh_errors;
wmsum_t arcstat_l2_rebuild_abort_cksum_lb_errors;
wmsum_t arcstat_l2_rebuild_abort_lowmem;
wmsum_t arcstat_l2_rebuild_size;
wmsum_t arcstat_l2_rebuild_asize;
wmsum_t arcstat_l2_rebuild_bufs;
wmsum_t arcstat_l2_rebuild_bufs_precached;
wmsum_t arcstat_l2_rebuild_log_blks;
wmsum_t arcstat_memory_throttle_count;
wmsum_t arcstat_memory_direct_count;
wmsum_t arcstat_memory_indirect_count;
wmsum_t arcstat_prune;
aggsum_t arcstat_meta_used;
wmsum_t arcstat_async_upgrade_sync;
wmsum_t arcstat_demand_hit_predictive_prefetch;
wmsum_t arcstat_demand_hit_prescient_prefetch;
wmsum_t arcstat_raw_size;
wmsum_t arcstat_cached_only_in_progress;
wmsum_t arcstat_abd_chunk_waste_size;
} arc_sums_t;
typedef struct arc_evict_waiter {
list_node_t aew_node;
kcondvar_t aew_cv;
......@@ -885,7 +952,7 @@ typedef struct arc_evict_waiter {
#define ARCSTAT(stat) (arc_stats.stat.value.ui64)
#define ARCSTAT_INCR(stat, val) \
atomic_add_64(&arc_stats.stat.value.ui64, (val))
wmsum_add(&arc_sums.stat, (val))
#define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
#define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
......@@ -899,6 +966,7 @@ typedef struct arc_evict_waiter {
extern taskq_t *arc_prune_taskq;
extern arc_stats_t arc_stats;
extern arc_sums_t arc_sums;
extern hrtime_t arc_growtime;
extern boolean_t arc_warm;
extern int arc_grow_retry;
......@@ -906,7 +974,6 @@ extern int arc_no_grow_shift;
extern int arc_shrink_shift;
extern kmutex_t arc_prune_mtx;
extern list_t arc_prune_list;
extern aggsum_t arc_size;
extern arc_state_t *arc_mfu;
extern arc_state_t *arc_mru;
extern uint_t zfs_arc_pc_percent;
......
......@@ -69,6 +69,15 @@ static abd_stats_t abd_stats = {
{ "linear_data_size", KSTAT_DATA_UINT64 },
};
struct {
wmsum_t abdstat_struct_size;
wmsum_t abdstat_scatter_cnt;
wmsum_t abdstat_scatter_data_size;
wmsum_t abdstat_scatter_chunk_waste;
wmsum_t abdstat_linear_cnt;
wmsum_t abdstat_linear_data_size;
} abd_sums;
/*
* The size of the chunks ABD allocates. Because the sizes allocated from the
* kmem_cache can't change, this tunable can only be modified at boot. Changing
......@@ -271,16 +280,46 @@ abd_free_zero_scatter(void)
kmem_free(abd_zero_buf, zfs_abd_chunk_size);
}
static int
abd_kstats_update(kstat_t *ksp, int rw)
{
abd_stats_t *as = ksp->ks_data;
if (rw == KSTAT_WRITE)
return (EACCES);
as->abdstat_struct_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_struct_size);
as->abdstat_scatter_cnt.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_cnt);
as->abdstat_scatter_data_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_data_size);
as->abdstat_scatter_chunk_waste.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_chunk_waste);
as->abdstat_linear_cnt.value.ui64 =
wmsum_value(&abd_sums.abdstat_linear_cnt);
as->abdstat_linear_data_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_linear_data_size);
return (0);
}
void
abd_init(void)
{
abd_chunk_cache = kmem_cache_create("abd_chunk", zfs_abd_chunk_size, 0,
NULL, NULL, NULL, NULL, 0, KMC_NODEBUG);
wmsum_init(&abd_sums.abdstat_struct_size, 0);
wmsum_init(&abd_sums.abdstat_scatter_cnt, 0);
wmsum_init(&abd_sums.abdstat_scatter_data_size, 0);
wmsum_init(&abd_sums.abdstat_scatter_chunk_waste, 0);
wmsum_init(&abd_sums.abdstat_linear_cnt, 0);
wmsum_init(&abd_sums.abdstat_linear_data_size, 0);
abd_ksp = kstat_create("zfs", 0, "abdstats", "misc", KSTAT_TYPE_NAMED,
sizeof (abd_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
if (abd_ksp != NULL) {
abd_ksp->ks_data = &abd_stats;
abd_ksp->ks_update = abd_kstats_update;
kstat_install(abd_ksp);
}
......@@ -297,6 +336,13 @@ abd_fini(void)
abd_ksp = NULL;
}
wmsum_fini(&abd_sums.abdstat_struct_size);
wmsum_fini(&abd_sums.abdstat_scatter_cnt);
wmsum_fini(&abd_sums.abdstat_scatter_data_size);
wmsum_fini(&abd_sums.abdstat_scatter_chunk_waste);
wmsum_fini(&abd_sums.abdstat_linear_cnt);
wmsum_fini(&abd_sums.abdstat_linear_data_size);
kmem_cache_destroy(abd_chunk_cache);
abd_chunk_cache = NULL;
}
......
......@@ -132,6 +132,20 @@ static abd_stats_t abd_stats = {
{ "scatter_sg_table_retry", KSTAT_DATA_UINT64 },
};
struct {
wmsum_t abdstat_struct_size;
wmsum_t abdstat_linear_cnt;
wmsum_t abdstat_linear_data_size;
wmsum_t abdstat_scatter_cnt;
wmsum_t abdstat_scatter_data_size;
wmsum_t abdstat_scatter_chunk_waste;
wmsum_t abdstat_scatter_orders[MAX_ORDER];
wmsum_t abdstat_scatter_page_multi_chunk;
wmsum_t abdstat_scatter_page_multi_zone;
wmsum_t abdstat_scatter_page_alloc_retry;
wmsum_t abdstat_scatter_sg_table_retry;
} abd_sums;
#define abd_for_each_sg(abd, sg, n, i) \
for_each_sg(ABD_SCATTER(abd).abd_sgl, sg, n, i)
......@@ -687,6 +701,40 @@ abd_free_zero_scatter(void)
#endif /* _KERNEL */
}
static int
abd_kstats_update(kstat_t *ksp, int rw)
{
abd_stats_t *as = ksp->ks_data;
if (rw == KSTAT_WRITE)
return (EACCES);
as->abdstat_struct_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_struct_size);
as->abdstat_linear_cnt.value.ui64 =
wmsum_value(&abd_sums.abdstat_linear_cnt);
as->abdstat_linear_data_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_linear_data_size);
as->abdstat_scatter_cnt.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_cnt);
as->abdstat_scatter_data_size.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_data_size);
as->abdstat_scatter_chunk_waste.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_chunk_waste);
for (int i = 0; i < MAX_ORDER; i++) {
as->abdstat_scatter_orders[i].value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_orders[i]);
}
as->abdstat_scatter_page_multi_chunk.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_page_multi_chunk);
as->abdstat_scatter_page_multi_zone.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_page_multi_zone);
as->abdstat_scatter_page_alloc_retry.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_page_alloc_retry);
as->abdstat_scatter_sg_table_retry.value.ui64 =
wmsum_value(&abd_sums.abdstat_scatter_sg_table_retry);
return (0);
}
void
abd_init(void)
{
......@@ -695,6 +743,19 @@ abd_init(void)
abd_cache = kmem_cache_create("abd_t", sizeof (abd_t),
0, NULL, NULL, NULL, NULL, NULL, 0);
wmsum_init(&abd_sums.abdstat_struct_size, 0);
wmsum_init(&abd_sums.abdstat_linear_cnt, 0);
wmsum_init(&abd_sums.abdstat_linear_data_size, 0);
wmsum_init(&abd_sums.abdstat_scatter_cnt, 0);
wmsum_init(&abd_sums.abdstat_scatter_data_size, 0);
wmsum_init(&abd_sums.abdstat_scatter_chunk_waste, 0);
for (i = 0; i < MAX_ORDER; i++)
wmsum_init(&abd_sums.abdstat_scatter_orders[i], 0);
wmsum_init(&abd_sums.abdstat_scatter_page_multi_chunk, 0);
wmsum_init(&abd_sums.abdstat_scatter_page_multi_zone, 0);
wmsum_init(&abd_sums.abdstat_scatter_page_alloc_retry, 0);
wmsum_init(&abd_sums.abdstat_scatter_sg_table_retry, 0);
abd_ksp = kstat_create("zfs", 0, "abdstats", "misc", KSTAT_TYPE_NAMED,
sizeof (abd_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
if (abd_ksp != NULL) {
......@@ -705,6 +766,7 @@ abd_init(void)
KSTAT_DATA_UINT64;
}
abd_ksp->ks_data = &abd_stats;
abd_ksp->ks_update = abd_kstats_update;
kstat_install(abd_ksp);
}
......@@ -721,6 +783,19 @@ abd_fini(void)
abd_ksp = NULL;
}
wmsum_fini(&abd_sums.abdstat_struct_size);
wmsum_fini(&abd_sums.abdstat_linear_cnt);
wmsum_fini(&abd_sums.abdstat_linear_data_size);
wmsum_fini(&abd_sums.abdstat_scatter_cnt);
wmsum_fini(&abd_sums.abdstat_scatter_data_size);
wmsum_fini(&abd_sums.abdstat_scatter_chunk_waste);
for (int i = 0; i < MAX_ORDER; i++)
wmsum_fini(&abd_sums.abdstat_scatter_orders[i]);
wmsum_fini(&abd_sums.abdstat_scatter_page_multi_chunk);
wmsum_fini(&abd_sums.abdstat_scatter_page_multi_zone);
wmsum_fini(&abd_sums.abdstat_scatter_page_alloc_retry);
wmsum_fini(&abd_sums.abdstat_scatter_sg_table_retry);
if (abd_cache) {
kmem_cache_destroy(abd_cache);
abd_cache = NULL;
......
......@@ -135,7 +135,7 @@ arc_available_memory(void)
static uint64_t
arc_evictable_memory(void)
{
int64_t asize = aggsum_value(&arc_size);
int64_t asize = aggsum_value(&arc_sums.arcstat_size);
uint64_t arc_clean =
zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) +
zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) +
......
This diff is collapsed.
......@@ -52,6 +52,7 @@
#include <sys/vdev.h>
#include <cityhash.h>
#include <sys/spa_impl.h>
#include <sys/wmsum.h>
kstat_t *dbuf_ksp;
......@@ -135,8 +136,22 @@ dbuf_stats_t dbuf_stats = {
{ "metadata_cache_overflow", KSTAT_DATA_UINT64 }
};
struct {
wmsum_t cache_count;
wmsum_t cache_total_evicts;
wmsum_t cache_levels[DN_MAX_LEVELS];
wmsum_t cache_levels_bytes[DN_MAX_LEVELS];
wmsum_t hash_hits;
wmsum_t hash_misses;
wmsum_t hash_collisions;
wmsum_t hash_chains;
wmsum_t hash_insert_race;
wmsum_t metadata_cache_count;
wmsum_t metadata_cache_overflow;
} dbuf_sums;
#define DBUF_STAT_INCR(stat, val) \
atomic_add_64(&dbuf_stats.stat.value.ui64, (val));
wmsum_add(&dbuf_sums.stat, val);
#define DBUF_STAT_DECR(stat, val) \
DBUF_STAT_INCR(stat, -(val));
#define DBUF_STAT_BUMP(stat) \
......@@ -297,8 +312,6 @@ dbuf_dest(void *vdb, void *unused)
*/
static dbuf_hash_table_t dbuf_hash_table;
static uint64_t dbuf_hash_count;
/*
* We use Cityhash for this. It's fast, and has good hash properties without
* requiring any large static buffers.
......@@ -409,8 +422,8 @@ dbuf_hash_insert(dmu_buf_impl_t *db)
db->db_hash_next = h->hash_table[idx];
h->hash_table[idx] = db;
mutex_exit(DBUF_HASH_MUTEX(h, idx));
atomic_inc_64(&dbuf_hash_count);
DBUF_STAT_MAX(hash_elements_max, dbuf_hash_count);
uint64_t he = atomic_inc_64_nv(&dbuf_stats.hash_elements.value.ui64);
DBUF_STAT_MAX(hash_elements_max, he);
return (NULL);
}
......@@ -483,7 +496,7 @@ dbuf_hash_remove(dmu_buf_impl_t *db)
h->hash_table[idx]->db_hash_next == NULL)
DBUF_STAT_BUMPDOWN(hash_chains);
mutex_exit(DBUF_HASH_MUTEX(h, idx));
atomic_dec_64(&dbuf_hash_count);
atomic_dec_64(&dbuf_stats.hash_elements.value.ui64);
}
typedef enum {
......@@ -767,19 +780,40 @@ dbuf_kstat_update(kstat_t *ksp, int rw)
{
dbuf_stats_t *ds = ksp->ks_data;
if (rw == KSTAT_WRITE) {
if (rw == KSTAT_WRITE)
return (SET_ERROR(EACCES));
} else {
ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count(
&dbuf_caches[DB_DBUF_METADATA_CACHE].size);
ds->cache_size_bytes.value.ui64 =
zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size);
ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes();
ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes();
ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes();
ds->hash_elements.value.ui64 = dbuf_hash_count;
}
ds->cache_count.value.ui64 =
wmsum_value(&dbuf_sums.cache_count);
ds->cache_size_bytes.value.ui64 =
zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size);
ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes();
ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes();
ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes();
ds->cache_total_evicts.value.ui64 =
wmsum_value(&dbuf_sums.cache_total_evicts);
for (int i = 0; i < DN_MAX_LEVELS; i++) {
ds->cache_levels[i].value.ui64 =
wmsum_value(&dbuf_sums.cache_levels[i]);
ds->cache_levels_bytes[i].value.ui64 =
wmsum_value(&dbuf_sums.cache_levels_bytes[i]);
}
ds->hash_hits.value.ui64 =
wmsum_value(&dbuf_sums.hash_hits);
ds->hash_misses.value.ui64 =
wmsum_value(&dbuf_sums.hash_misses);
ds->hash_collisions.value.ui64 =
wmsum_value(&dbuf_sums.hash_collisions);
ds->hash_chains.value.ui64 =
wmsum_value(&dbuf_sums.hash_chains);
ds->hash_insert_race.value.ui64 =
wmsum_value(&dbuf_sums.hash_insert_race);
ds->metadata_cache_count.value.ui64 =
wmsum_value(&dbuf_sums.metadata_cache_count);
ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count(
&dbuf_caches[DB_DBUF_METADATA_CACHE].size);
ds->metadata_cache_overflow.value.ui64 =
wmsum_value(&dbuf_sums.metadata_cache_overflow);
return (0);
}
......@@ -846,6 +880,20 @@ dbuf_init(void)
dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread,
NULL, 0, &p0, TS_RUN, minclsyspri);