*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2015, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_CLASS
-#include <libcfs/libcfs.h>
#include <linux/module.h>
+#include <linux/list.h>
+#include <libcfs/libcfs.h>
#include <libcfs/libcfs_hash.h> /* hash_long() */
#include <obd_class.h>
#include <obd_support.h>
#include <lustre_fid.h>
#include <lu_object.h>
#include <lu_ref.h>
-#include <libcfs/list.h>
enum {
LU_CACHE_PERCENT_MAX = 50,
LU_CACHE_PERCENT_DEFAULT = 20
};
-#define LU_CACHE_NR_MAX_ADJUST 128
+#define LU_CACHE_NR_MAX_ADJUST 512
#define LU_CACHE_NR_UNLIMITED -1
#define LU_CACHE_NR_DEFAULT LU_CACHE_NR_UNLIMITED
#define LU_CACHE_NR_LDISKFS_LIMIT LU_CACHE_NR_UNLIMITED
LASSERT(list_empty(&top->loh_lru));
list_add_tail(&top->loh_lru, &bkt->lsb_lru);
bkt->lsb_lru_len++;
- lprocfs_counter_incr(site->ls_stats, LU_SS_LRU_LEN);
+ percpu_counter_inc(&site->ls_lru_len_counter);
CDEBUG(D_INODE, "Add %p to site lru. hash: %p, bkt: %p, "
"lru_len: %ld\n",
o, site->ls_obj_hash, bkt, bkt->lsb_lru_len);
list_del_init(&top->loh_lru);
bkt = cfs_hash_bd_extra_get(obj_hash, &bd);
bkt->lsb_lru_len--;
- lprocfs_counter_decr(site->ls_stats, LU_SS_LRU_LEN);
+ percpu_counter_dec(&site->ls_lru_len_counter);
}
cfs_hash_bd_del_locked(obj_hash, &bd, &top->loh_hash);
cfs_hash_bd_unlock(obj_hash, &bd, 1);
/**
* Free \a nr objects from the cold end of the site LRU list.
+ * if canblock is 0, then don't block awaiting for another
+ * instance of lu_site_purge() to complete
*/
-int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
+int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s,
+ int nr, int canblock)
{
struct lu_object_header *h;
struct lu_object_header *temp;
* It doesn't make any sense to make purge threads parallel, that can
* only bring troubles to us. See LU-5331.
*/
- mutex_lock(&s->ls_purge_mutex);
+ if (canblock != 0)
+ mutex_lock(&s->ls_purge_mutex);
+ else if (mutex_trylock(&s->ls_purge_mutex) == 0)
+ goto out;
+
did_sth = 0;
cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
if (i < start)
&bd2, &h->loh_hash);
list_move(&h->loh_lru, &dispose);
bkt->lsb_lru_len--;
- lprocfs_counter_decr(s->ls_stats, LU_SS_LRU_LEN);
+ percpu_counter_dec(&s->ls_lru_len_counter);
if (did_sth == 0)
did_sth = 1;
/* race on s->ls_purge_start, but nobody cares */
s->ls_purge_start = i % CFS_HASH_NBKT(s->ls_obj_hash);
+out:
return nr;
}
-EXPORT_SYMBOL(lu_site_purge);
+EXPORT_SYMBOL(lu_site_purge_objects);
/*
* Object printing.
if (!list_empty(&h->loh_lru)) {
list_del_init(&h->loh_lru);
bkt->lsb_lru_len--;
- lprocfs_counter_decr(s->ls_stats, LU_SS_LRU_LEN);
+ percpu_counter_dec(&s->ls_lru_len_counter);
}
return lu_object_top(h);
}
size = cfs_hash_size_get(dev->ld_site->ls_obj_hash);
nr = (__u64)lu_cache_nr;
- if (size > nr)
- lu_site_purge(env, dev->ld_site,
- MIN(size - nr, LU_CACHE_NR_MAX_ADJUST));
+ if (size <= nr)
+ return;
- return;
+ lu_site_purge_objects(env, dev->ld_site,
+ MIN(size - nr, LU_CACHE_NR_MAX_ADJUST), 0);
}
static struct lu_object *lu_object_new(const struct lu_env *env,
}
EXPORT_SYMBOL(lu_object_find_slice);
-/**
- * Global list of all device types.
- */
-static struct list_head lu_device_types;
-
int lu_device_type_init(struct lu_device_type *ldt)
{
int result = 0;
atomic_set(&ldt->ldt_device_nr, 0);
- INIT_LIST_HEAD(&ldt->ldt_linkage);
if (ldt->ldt_ops->ldto_init)
result = ldt->ldt_ops->ldto_init(ldt);
- if (result == 0) {
- spin_lock(&obd_types_lock);
- list_add(&ldt->ldt_linkage, &lu_device_types);
- spin_unlock(&obd_types_lock);
- }
-
return result;
}
EXPORT_SYMBOL(lu_device_type_init);
void lu_device_type_fini(struct lu_device_type *ldt)
{
- spin_lock(&obd_types_lock);
- list_del_init(&ldt->ldt_linkage);
- spin_unlock(&obd_types_lock);
if (ldt->ldt_ops->ldto_fini)
ldt->ldt_ops->ldto_fini(ldt);
}
/**
* Global list of all sites on this node
*/
-static struct list_head lu_sites;
-static DEFINE_MUTEX(lu_sites_guard);
+static LIST_HEAD(lu_sites);
+static DECLARE_RWSEM(lu_sites_guard);
/**
* Global environment used by site shrinker.
#if BITS_PER_LONG == 32
/* limit hashtable size for lowmem systems to low RAM */
- if (cache_size > 1 << (30 - PAGE_CACHE_SHIFT))
- cache_size = 1 << (30 - PAGE_CACHE_SHIFT) * 3 / 4;
+ if (cache_size > 1 << (30 - PAGE_SHIFT))
+ cache_size = 1 << (30 - PAGE_SHIFT) * 3 / 4;
#endif
/* clear off unreasonable cache setting. */
lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
}
cache_size = cache_size / 100 * lu_cache_percent *
- (PAGE_CACHE_SIZE / 1024);
+ (PAGE_SIZE / 1024);
for (bits = 1; (1 << bits) < cache_size; ++bits) {
;
char name[16];
unsigned long bits;
unsigned int i;
+ int rc;
ENTRY;
memset(s, 0, sizeof *s);
mutex_init(&s->ls_purge_mutex);
+
+#ifdef HAVE_PERCPU_COUNTER_INIT_GFP_FLAG
+ rc = percpu_counter_init(&s->ls_lru_len_counter, 0, GFP_NOFS);
+#else
+ rc = percpu_counter_init(&s->ls_lru_len_counter, 0);
+#endif
+ if (rc)
+ return -ENOMEM;
+
snprintf(name, sizeof(name), "lu_site_%s", top->ld_type->ldt_name);
for (bits = lu_htable_order(top);
bits >= LU_SITE_BITS_MIN; bits--) {
0, "cache_death_race", "cache_death_race");
lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED,
0, "lru_purged", "lru_purged");
- /*
- * Unlike other counters, lru_len can be decremented so
- * need lc_sum instead of just lc_count
- */
- lprocfs_counter_init(s->ls_stats, LU_SS_LRU_LEN,
- LPROCFS_CNTR_AVGMINMAX, "lru_len", "lru_len");
INIT_LIST_HEAD(&s->ls_linkage);
s->ls_top_dev = top;
*/
void lu_site_fini(struct lu_site *s)
{
- mutex_lock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
list_del_init(&s->ls_linkage);
- mutex_unlock(&lu_sites_guard);
+ up_write(&lu_sites_guard);
+
+ percpu_counter_destroy(&s->ls_lru_len_counter);
if (s->ls_obj_hash != NULL) {
cfs_hash_putref(s->ls_obj_hash);
int lu_site_init_finish(struct lu_site *s)
{
int result;
- mutex_lock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
result = lu_context_refill(&lu_shrink_env.le_ctx);
if (result == 0)
list_add(&s->ls_linkage, &lu_sites);
- mutex_unlock(&lu_sites_guard);
+ up_write(&lu_sites_guard);
return result;
}
EXPORT_SYMBOL(lu_site_init_finish);
/**
* List of remembered contexts. XXX document me.
*/
-static struct list_head lu_context_remembered;
+static LIST_HEAD(lu_context_remembered);
/**
* Destroy \a key in all remembered contexts. This is used to destroy key
list_for_each_entry(ctx, &lu_context_remembered,
lc_remember)
key_fini(ctx, key->lct_index);
- write_unlock(&lu_keys_guard);
+
++key_set_version;
+ write_unlock(&lu_keys_guard);
}
}
void lu_context_key_revive(struct lu_context_key *key)
{
- key->lct_tags &= ~LCT_QUIESCENT;
- ++key_set_version;
+ write_lock(&lu_keys_guard);
+ key->lct_tags &= ~LCT_QUIESCENT;
+ ++key_set_version;
+ write_unlock(&lu_keys_guard);
}
static void keys_fini(struct lu_context *ctx)
static int keys_fill(struct lu_context *ctx)
{
unsigned int i;
+ unsigned pre_version;
/*
* A serialisation with lu_context_key_quiesce() is needed, but some
*/
read_lock(&lu_keys_guard);
atomic_inc(&lu_key_initing_cnt);
+ pre_version = key_set_version;
read_unlock(&lu_keys_guard);
- LINVRNT(ctx->lc_value != NULL);
- for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
- struct lu_context_key *key;
+refill:
+ LINVRNT(ctx->lc_value != NULL);
+ for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
+ struct lu_context_key *key;
- key = lu_keys[i];
- if (ctx->lc_value[i] == NULL && key != NULL &&
- (key->lct_tags & ctx->lc_tags) &&
- /*
- * Don't create values for a LCT_QUIESCENT key, as this
- * will pin module owning a key.
- */
- !(key->lct_tags & LCT_QUIESCENT)) {
- void *value;
-
- LINVRNT(key->lct_init != NULL);
- LINVRNT(key->lct_index == i);
+ key = lu_keys[i];
+ if (ctx->lc_value[i] == NULL && key != NULL &&
+ (key->lct_tags & ctx->lc_tags) &&
+ /*
+ * Don't create values for a LCT_QUIESCENT key, as this
+ * will pin module owning a key.
+ */
+ !(key->lct_tags & LCT_QUIESCENT)) {
+ void *value;
+
+ LINVRNT(key->lct_init != NULL);
+ LINVRNT(key->lct_index == i);
LASSERT(key->lct_owner != NULL);
if (!(ctx->lc_tags & LCT_NOREF) &&
lu_ref_add_atomic(&key->lct_reference, "ctx", ctx);
atomic_inc(&key->lct_used);
- /*
- * This is the only place in the code, where an
- * element of ctx->lc_value[] array is set to non-NULL
- * value.
- */
- ctx->lc_value[i] = value;
- if (key->lct_exit != NULL)
- ctx->lc_tags |= LCT_HAS_EXIT;
- }
- ctx->lc_version = key_set_version;
- }
+ /*
+ * This is the only place in the code, where an
+ * element of ctx->lc_value[] array is set to non-NULL
+ * value.
+ */
+ ctx->lc_value[i] = value;
+ if (key->lct_exit != NULL)
+ ctx->lc_tags |= LCT_HAS_EXIT;
+ }
+ }
+
+ read_lock(&lu_keys_guard);
+ if (pre_version != key_set_version) {
+ pre_version = key_set_version;
+ read_unlock(&lu_keys_guard);
+ goto refill;
+ }
+
atomic_dec(&lu_key_initing_cnt);
- return 0;
+ read_unlock(&lu_keys_guard);
+ return 0;
}
static int keys_init(struct lu_context *ctx)
LINVRNT(ctx->lc_state == LCS_ENTERED);
ctx->lc_state = LCS_LEFT;
if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) {
+ /* could race with key quiescency */
+ if (ctx->lc_tags & LCT_REMEMBER)
+ read_lock(&lu_keys_guard);
+
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
- /* could race with key quiescency */
- if (ctx->lc_tags & LCT_REMEMBER)
- read_lock(&lu_keys_guard);
if (ctx->lc_value[i] != NULL) {
struct lu_context_key *key;
key->lct_exit(ctx,
key, ctx->lc_value[i]);
}
- if (ctx->lc_tags & LCT_REMEMBER)
- read_unlock(&lu_keys_guard);
}
+
+ if (ctx->lc_tags & LCT_REMEMBER)
+ read_unlock(&lu_keys_guard);
}
}
EXPORT_SYMBOL(lu_context_exit);
/*
- * lu_cache_shrink_count returns the number of cached objects that are
- * candidates to be freed by shrink_slab(). A counter, which tracks
- * the number of items in the site's lru, is maintained in the per cpu
- * stats of each site. The counter is incremented when an object is added
- * to a site's lru and decremented when one is removed. The number of
- * free-able objects is the sum of all per cpu counters for all sites.
+ * lu_cache_shrink_count() returns an approximate number of cached objects
+ * that can be freed by shrink_slab(). A counter, which tracks the
+ * number of items in the site's lru, is maintained in a percpu_counter
+ * for each site. The percpu values are incremented and decremented as
+ * objects are added or removed from the lru. The percpu values are summed
+ * and saved whenever a percpu value exceeds a threshold. Thus the saved,
+ * summed value at any given time may not accurately reflect the current
+ * lru length. But this value is sufficiently accurate for the needs of
+ * a shrinker.
*
* Using a per cpu counter is a compromise solution to concurrent access:
* lu_object_put() can update the counter without locking the site and
if (!(sc->gfp_mask & __GFP_FS))
return 0;
- mutex_lock(&lu_sites_guard);
- list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
- cached += ls_stats_read(s->ls_stats, LU_SS_LRU_LEN);
- }
- mutex_unlock(&lu_sites_guard);
+ down_read(&lu_sites_guard);
+ list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage)
+ cached += percpu_counter_read_positive(&s->ls_lru_len_counter);
+ up_read(&lu_sites_guard);
cached = (cached / 100) * sysctl_vfs_cache_pressure;
CDEBUG(D_INODE, "%ld objects cached, cache pressure %d\n",
*/
return SHRINK_STOP;
- mutex_lock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
remain = lu_site_purge(&lu_shrink_env, s, remain);
/*
list_move_tail(&s->ls_linkage, &splice);
}
list_splice(&splice, lu_sites.prev);
- mutex_unlock(&lu_sites_guard);
+ up_write(&lu_sites_guard);
return sc->nr_to_scan - remain;
}
CDEBUG(D_INFO, "Lustre LU module (%p).\n", &lu_keys);
- INIT_LIST_HEAD(&lu_device_types);
- INIT_LIST_HEAD(&lu_context_remembered);
- INIT_LIST_HEAD(&lu_sites);
-
result = lu_ref_global_init();
if (result != 0)
return result;
* conservatively. This should not be too bad, because this
* environment is global.
*/
- mutex_lock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
- mutex_unlock(&lu_sites_guard);
+ up_write(&lu_sites_guard);
if (result != 0)
return result;
* Tear shrinker environment down _after_ de-registering
* lu_global_key, because the latter has a value in the former.
*/
- mutex_lock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
lu_env_fini(&lu_shrink_env);
- mutex_unlock(&lu_sites_guard);
+ up_write(&lu_sites_guard);
lu_ref_global_fini();
}
struct lprocfs_counter ret;
lprocfs_stats_collect(stats, idx, &ret);
- if (idx == LU_SS_LRU_LEN)
- /*
- * protect against counter on cpu A being decremented
- * before counter is incremented on cpu B; unlikely
- */
- return (__u32)((ret.lc_sum > 0) ? ret.lc_sum : 0);
- else
- return (__u32)ret.lc_count;
+ return (__u32)ret.lc_count;
#else
return 0;
#endif
memset(&stats, 0, sizeof(stats));
lu_site_stats_get(s->ls_obj_hash, &stats, 1);
- seq_printf(m, "%d/%d %d/%d %d %d %d %d %d %d %d %d\n",
+ seq_printf(m, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
stats.lss_busy,
stats.lss_total,
stats.lss_populated,
ls_stats_read(s->ls_stats, LU_SS_CACHE_MISS),
ls_stats_read(s->ls_stats, LU_SS_CACHE_RACE),
ls_stats_read(s->ls_stats, LU_SS_CACHE_DEATH_RACE),
- ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED),
- ls_stats_read(s->ls_stats, LU_SS_LRU_LEN));
+ ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED));
return 0;
}
EXPORT_SYMBOL(lu_site_stats_seq_print);
{
struct lu_site *s = o->lo_dev->ld_site;
struct lu_fid *old = &o->lo_header->loh_fid;
- struct lu_object *shadow;
- wait_queue_t waiter;
struct cfs_hash *hs;
struct cfs_hash_bd bd;
- __u64 version = 0;
LASSERT(fid_is_zero(old));
+ /* supposed to be unique */
hs = s->ls_obj_hash;
cfs_hash_bd_get_and_lock(hs, (void *)fid, &bd, 1);
- shadow = htable_lookup(s, &bd, fid, &waiter, &version);
- /* supposed to be unique */
- LASSERT(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT);
+#ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
+ {
+ __u64 version = 0;
+ wait_queue_t waiter;
+ struct lu_object *shadow;
+ shadow = htable_lookup(s, &bd, fid, &waiter, &version);
+ /* supposed to be unique */
+ LASSERT(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT);
+ }
+#endif
*old = *fid;
cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
cfs_hash_bd_unlock(hs, &bd, 1);
buf->lb_len = len;
return 0;
}
+EXPORT_SYMBOL(lu_buf_check_and_grow);