*
* 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, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2017, 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>
+#ifdef HAVE_PROCESSOR_H
+#include <linux/processor.h>
+#else
+#include <libcfs/linux/processor.h>
+#endif
+
+#include <libcfs/libcfs.h>
#include <libcfs/libcfs_hash.h> /* hash_long() */
+#include <libcfs/linux/linux-mem.h>
#include <obd_class.h>
#include <obd_support.h>
#include <lustre_disk.h>
#include <lustre_fid.h>
#include <lu_object.h>
#include <lu_ref.h>
-#include <libcfs/list.h>
+
+struct lu_site_bkt_data {
+ /**
+ * LRU list, updated on each access to object. Protected by
+ * bucket lock of lu_site::ls_obj_hash.
+ *
+ * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
+ * moved to the lu_site::ls_lru.prev (this is due to the non-existence
+ * of list_for_each_entry_safe_reverse()).
+ */
+ struct list_head lsb_lru;
+ /**
+ * Wait-queue signaled when an object in this site is ultimately
+ * destroyed (lu_object_free()). It is used by lu_object_find() to
+ * wait before re-trying when object in the process of destruction is
+ * found in the hash table.
+ *
+ * \see htable_lookup().
+ */
+ wait_queue_head_t lsb_marche_funebre;
+};
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
#define LU_SITE_BITS_MIN 12
#define LU_SITE_BITS_MAX 24
+#define LU_SITE_BITS_MAX_CL 19
/**
* total 256 buckets, we don't want too many buckets because:
* - consume too much memory
static unsigned int lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
-CFS_MODULE_PARM(lu_cache_percent, "i", int, 0644,
- "Percentage of memory to be used as lu_object cache");
+module_param(lu_cache_percent, int, 0644);
+MODULE_PARM_DESC(lu_cache_percent, "Percentage of memory to be used as lu_object cache");
static long lu_cache_nr = LU_CACHE_NR_DEFAULT;
-CFS_MODULE_PARM(lu_cache_nr, "l", long, 0644,
- "Maximum number of objects in lu_object cache");
+module_param(lu_cache_nr, long, 0644);
+MODULE_PARM_DESC(lu_cache_nr, "Maximum number of objects in lu_object cache");
static void lu_object_free(const struct lu_env *env, struct lu_object *o);
+static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx);
+
+wait_queue_head_t *
+lu_site_wq_from_fid(struct lu_site *site, struct lu_fid *fid)
+{
+ struct cfs_hash_bd bd;
+ struct lu_site_bkt_data *bkt;
+
+ cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
+ bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
+ return &bkt->lsb_marche_funebre;
+}
+EXPORT_SYMBOL(lu_site_wq_from_fid);
/**
* Decrease reference counter on object. If last reference is freed, return
*/
void lu_object_put(const struct lu_env *env, struct lu_object *o)
{
- struct lu_site_bkt_data *bkt;
- struct lu_object_header *top;
- struct lu_site *site;
- struct lu_object *orig;
- cfs_hash_bd_t bd;
- const struct lu_fid *fid;
-
- top = o->lo_header;
- site = o->lo_dev->ld_site;
- orig = o;
+ struct lu_site_bkt_data *bkt;
+ struct lu_object_header *top = o->lo_header;
+ struct lu_site *site = o->lo_dev->ld_site;
+ struct lu_object *orig = o;
+ struct cfs_hash_bd bd;
+ const struct lu_fid *fid = lu_object_fid(o);
+ bool is_dying;
/*
* till we have full fids-on-OST implemented anonymous objects
* are possible in OSP. such an object isn't listed in the site
* so we should not remove it from the site.
*/
- fid = lu_object_fid(o);
if (fid_is_zero(fid)) {
LASSERT(top->loh_hash.next == NULL
&& top->loh_hash.pprev == NULL);
return;
}
- cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
- bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
+ cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
+ bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
+ is_dying = lu_object_is_dying(top);
if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) {
- if (lu_object_is_dying(top)) {
-
+ /* at this point the object reference is dropped and lock is
+ * not taken, so lu_object should not be touched because it
+ * can be freed by concurrent thread. Use local variable for
+ * check.
+ */
+ if (is_dying) {
/*
* somebody may be waiting for this, currently only
* used for cl_object, see cl_object_put_last().
return;
}
- LASSERT(bkt->lsb_busy > 0);
- bkt->lsb_busy--;
- /*
- * When last reference is released, iterate over object
- * layers, and notify them that object is no longer busy.
- */
+ /*
+ * When last reference is released, iterate over object
+ * layers, and notify them that object is no longer busy.
+ */
list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
- if (o->lo_ops->loo_object_release != NULL)
- o->lo_ops->loo_object_release(env, o);
- }
+ if (o->lo_ops->loo_object_release != NULL)
+ o->lo_ops->loo_object_release(env, o);
+ }
- if (!lu_object_is_dying(top)) {
+ /* don't use local 'is_dying' here because if was taken without lock
+ * but here we need the latest actual value of it so check lu_object
+ * directly here.
+ */
+ if (!lu_object_is_dying(top) &&
+ (lu_object_exists(orig) || lu_object_is_cl(orig))) {
LASSERT(list_empty(&top->loh_lru));
list_add_tail(&top->loh_lru, &bkt->lsb_lru);
- cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
- return;
- }
+ percpu_counter_inc(&site->ls_lru_len_counter);
+ CDEBUG(D_INODE, "Add %p/%p to site lru. hash: %p, bkt: %p\n",
+ orig, top, site->ls_obj_hash, bkt);
+ cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
+ return;
+ }
- /*
- * If object is dying (will not be cached), removed it
- * from hash table and LRU.
- *
- * This is done with hash table and LRU lists locked. As the only
- * way to acquire first reference to previously unreferenced
- * object is through hash-table lookup (lu_object_find()),
- * or LRU scanning (lu_site_purge()), that are done under hash-table
- * and LRU lock, no race with concurrent object lookup is possible
- * and we can safely destroy object below.
- */
+ /*
+ * If object is dying (will not be cached) then remove it
+ * from hash table and LRU.
+ *
+ * This is done with hash table and LRU lists locked. As the only
+ * way to acquire first reference to previously unreferenced
+ * object is through hash-table lookup (lu_object_find()),
+ * or LRU scanning (lu_site_purge()), that are done under hash-table
+ * and LRU lock, no race with concurrent object lookup is possible
+ * and we can safely destroy object below.
+ */
if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags))
cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
- cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
- /*
- * Object was already removed from hash and lru above, can
- * kill it.
- */
- lu_object_free(env, orig);
+ cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
+ /*
+ * Object was already removed from hash and lru above, can
+ * kill it.
+ */
+ lu_object_free(env, orig);
}
EXPORT_SYMBOL(lu_object_put);
top = o->lo_header;
set_bit(LU_OBJECT_HEARD_BANSHEE, &top->loh_flags);
if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags)) {
- cfs_hash_t *obj_hash = o->lo_dev->ld_site->ls_obj_hash;
- cfs_hash_bd_t bd;
+ struct lu_site *site = o->lo_dev->ld_site;
+ struct cfs_hash *obj_hash = site->ls_obj_hash;
+ struct cfs_hash_bd bd;
cfs_hash_bd_get_and_lock(obj_hash, &top->loh_fid, &bd, 1);
- list_del_init(&top->loh_lru);
+ if (!list_empty(&top->loh_lru)) {
+ struct lu_site_bkt_data *bkt;
+
+ list_del_init(&top->loh_lru);
+ bkt = cfs_hash_bd_extra_get(obj_hash, &bd);
+ 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);
}
*/
static void lu_object_free(const struct lu_env *env, struct lu_object *o)
{
- struct lu_site_bkt_data *bkt;
+ wait_queue_head_t *wq;
struct lu_site *site;
struct lu_object *scan;
struct list_head *layers;
struct list_head splice;
- site = o->lo_dev->ld_site;
- layers = &o->lo_header->loh_layers;
- bkt = lu_site_bkt_from_fid(site, &o->lo_header->loh_fid);
+ site = o->lo_dev->ld_site;
+ layers = &o->lo_header->loh_layers;
+ wq = lu_site_wq_from_fid(site, &o->lo_header->loh_fid);
/*
* First call ->loo_object_delete() method to release all resources.
*/
o->lo_ops->loo_object_free(env, o);
}
- if (waitqueue_active(&bkt->lsb_marche_funebre))
- wake_up_all(&bkt->lsb_marche_funebre);
+ if (waitqueue_active(wq))
+ wake_up_all(wq);
}
/**
* 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;
struct lu_site_bkt_data *bkt;
- cfs_hash_bd_t bd;
- cfs_hash_bd_t bd2;
+ struct cfs_hash_bd bd;
+ struct cfs_hash_bd bd2;
struct list_head dispose;
int did_sth;
- unsigned int start;
+ unsigned int start = 0;
int count;
int bnr;
unsigned int i;
* Under LRU list lock, scan LRU list and move unreferenced objects to
* the dispose list, removing them from LRU and hash table.
*/
- start = s->ls_purge_start;
+ if (nr != ~0)
+ start = s->ls_purge_start;
bnr = (nr == ~0) ? -1 : nr / (int)CFS_HASH_NBKT(s->ls_obj_hash) + 1;
again:
/*
* 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)
cfs_hash_bd_del_locked(s->ls_obj_hash,
&bd2, &h->loh_hash);
list_move(&h->loh_lru, &dispose);
+ 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.
* Key, holding temporary buffer. This key is registered very early by
* lu_global_init().
*/
-struct lu_context_key lu_global_key = {
+static struct lu_context_key lu_global_key = {
.lct_tags = LCT_MD_THREAD | LCT_DT_THREAD |
LCT_MG_THREAD | LCT_CL_THREAD | LCT_LOCAL,
.lct_init = lu_global_key_init,
}
return 1;
}
-EXPORT_SYMBOL(lu_object_invariant);
static struct lu_object *htable_lookup(struct lu_site *s,
- cfs_hash_bd_t *bd,
+ struct cfs_hash_bd *bd,
const struct lu_fid *f,
- wait_queue_t *waiter,
__u64 *version)
{
struct lu_site_bkt_data *bkt;
struct lu_object_header *h;
- struct hlist_node *hnode;
- __u64 ver = cfs_hash_bd_version_get(bd);
+ struct hlist_node *hnode;
+ __u64 ver = cfs_hash_bd_version_get(bd);
- if (*version == ver)
- return ERR_PTR(-ENOENT);
-
- *version = ver;
- bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
- /* cfs_hash_bd_peek_locked is a somehow "internal" function
- * of cfs_hash, it doesn't add refcount on object. */
- hnode = cfs_hash_bd_peek_locked(s->ls_obj_hash, bd, (void *)f);
- if (hnode == NULL) {
- lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
+ if (*version == ver)
return ERR_PTR(-ENOENT);
- }
-
- h = container_of0(hnode, struct lu_object_header, loh_hash);
- if (likely(!lu_object_is_dying(h))) {
- cfs_hash_get(s->ls_obj_hash, hnode);
- lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
- list_del_init(&h->loh_lru);
- return lu_object_top(h);
- }
-
- /*
- * Lookup found an object being destroyed this object cannot be
- * returned (to assure that references to dying objects are eventually
- * drained), and moreover, lookup has to wait until object is freed.
- */
-
- if (likely(waiter != NULL)) {
- init_waitqueue_entry_current(waiter);
- add_wait_queue(&bkt->lsb_marche_funebre, waiter);
- set_current_state(TASK_UNINTERRUPTIBLE);
- lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE);
- }
-
- return ERR_PTR(-EAGAIN);
-}
-
-static struct lu_object *htable_lookup_nowait(struct lu_site *s,
- cfs_hash_bd_t *bd,
- const struct lu_fid *f)
-{
- struct hlist_node *hnode;
- struct lu_object_header *h;
+ *version = ver;
+ bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
/* cfs_hash_bd_peek_locked is a somehow "internal" function
* of cfs_hash, it doesn't add refcount on object. */
hnode = cfs_hash_bd_peek_locked(s->ls_obj_hash, bd, (void *)f);
- if (hnode == NULL) {
+ if (!hnode) {
lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
return ERR_PTR(-ENOENT);
}
h = container_of0(hnode, struct lu_object_header, loh_hash);
- if (unlikely(lu_object_is_dying(h)))
- return ERR_PTR(-ENOENT);
-
cfs_hash_get(s->ls_obj_hash, hnode);
lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
- list_del_init(&h->loh_lru);
+ if (!list_empty(&h->loh_lru)) {
+ list_del_init(&h->loh_lru);
+ 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));
-
- return;
-}
-
-static struct lu_object *lu_object_new(const struct lu_env *env,
- struct lu_device *dev,
- const struct lu_fid *f,
- const struct lu_object_conf *conf)
-{
- struct lu_object *o;
- cfs_hash_t *hs;
- cfs_hash_bd_t bd;
- struct lu_site_bkt_data *bkt;
-
- o = lu_object_alloc(env, dev, f, conf);
- if (unlikely(IS_ERR(o)))
- return o;
-
- hs = dev->ld_site->ls_obj_hash;
- cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
- bkt = cfs_hash_bd_extra_get(hs, &bd);
- cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
- bkt->lsb_busy++;
- cfs_hash_bd_unlock(hs, &bd, 1);
-
- lu_object_limit(env, dev);
+ if (size <= nr)
+ return;
- return o;
+ lu_site_purge_objects(env, dev->ld_site,
+ MIN(size - nr, LU_CACHE_NR_MAX_ADJUST), 0);
}
/**
* Core logic of lu_object_find*() functions.
- */
-static struct lu_object *lu_object_find_try(const struct lu_env *env,
- struct lu_device *dev,
- const struct lu_fid *f,
- const struct lu_object_conf *conf,
- wait_queue_t *waiter)
-{
- struct lu_object *o;
- struct lu_object *shadow;
- struct lu_site *s;
- cfs_hash_t *hs;
- cfs_hash_bd_t bd;
- __u64 version = 0;
-
- /*
- * This uses standard index maintenance protocol:
- *
- * - search index under lock, and return object if found;
- * - otherwise, unlock index, allocate new object;
- * - lock index and search again;
- * - if nothing is found (usual case), insert newly created
- * object into index;
- * - otherwise (race: other thread inserted object), free
- * object just allocated.
- * - unlock index;
- * - return object.
- *
- * For "LOC_F_NEW" case, we are sure the object is new established.
- * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
- * just alloc and insert directly.
- *
- * If dying object is found during index search, add @waiter to the
- * site wait-queue and return ERR_PTR(-EAGAIN).
- */
- if (conf != NULL && conf->loc_flags & LOC_F_NEW)
- return lu_object_new(env, dev, f, conf);
-
- s = dev->ld_site;
- hs = s->ls_obj_hash;
- cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
- o = htable_lookup(s, &bd, f, waiter, &version);
- cfs_hash_bd_unlock(hs, &bd, 1);
- if (!IS_ERR(o) || PTR_ERR(o) != -ENOENT)
- return o;
-
- /*
- * Allocate new object. This may result in rather complicated
- * operations, including fld queries, inode loading, etc.
- */
- o = lu_object_alloc(env, dev, f, conf);
- if (unlikely(IS_ERR(o)))
- return o;
-
- LASSERT(lu_fid_eq(lu_object_fid(o), f));
-
- cfs_hash_bd_lock(hs, &bd, 1);
-
- shadow = htable_lookup(s, &bd, f, waiter, &version);
- if (likely(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT)) {
- struct lu_site_bkt_data *bkt;
-
- bkt = cfs_hash_bd_extra_get(hs, &bd);
- cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
- bkt->lsb_busy++;
- cfs_hash_bd_unlock(hs, &bd, 1);
-
- lu_object_limit(env, dev);
-
- return o;
- }
-
- lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE);
- cfs_hash_bd_unlock(hs, &bd, 1);
- lu_object_free(env, o);
- return shadow;
-}
-
-/**
+ *
* Much like lu_object_find(), but top level device of object is specifically
* \a dev rather than top level device of the site. This interface allows
* objects of different "stacking" to be created within the same site.
const struct lu_fid *f,
const struct lu_object_conf *conf)
{
- struct lu_site_bkt_data *bkt;
- struct lu_object *obj;
- wait_queue_t wait;
-
- while (1) {
- if (conf != NULL && conf->loc_flags & LOC_F_NOWAIT) {
- obj = lu_object_find_try(env, dev, f, conf, NULL);
+ struct lu_object *o;
+ struct lu_object *shadow;
+ struct lu_site *s;
+ struct cfs_hash *hs;
+ struct cfs_hash_bd bd;
+ __u64 version = 0;
- return obj;
- }
+ /*
+ * This uses standard index maintenance protocol:
+ *
+ * - search index under lock, and return object if found;
+ * - otherwise, unlock index, allocate new object;
+ * - lock index and search again;
+ * - if nothing is found (usual case), insert newly created
+ * object into index;
+ * - otherwise (race: other thread inserted object), free
+ * object just allocated.
+ * - unlock index;
+ * - return object.
+ *
+ * For "LOC_F_NEW" case, we are sure the object is new established.
+ * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
+ * just alloc and insert directly.
+ *
+ */
+ s = dev->ld_site;
+ hs = s->ls_obj_hash;
+ cfs_hash_bd_get(hs, f, &bd);
+ if (!(conf && conf->loc_flags & LOC_F_NEW)) {
+ cfs_hash_bd_lock(hs, &bd, 1);
+ o = htable_lookup(s, &bd, f, &version);
+ cfs_hash_bd_unlock(hs, &bd, 1);
- obj = lu_object_find_try(env, dev, f, conf, &wait);
- if (obj != ERR_PTR(-EAGAIN))
- return obj;
- /*
- * lu_object_find_try() already added waiter into the
- * wait queue.
- */
- waitq_wait(&wait, TASK_UNINTERRUPTIBLE);
- bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f);
- remove_wait_queue(&bkt->lsb_marche_funebre, &wait);
+ if (!IS_ERR(o) || PTR_ERR(o) != -ENOENT)
+ return o;
}
-}
-EXPORT_SYMBOL(lu_object_find_at);
+ /*
+ * Allocate new object. This may result in rather complicated
+ * operations, including fld queries, inode loading, etc.
+ */
+ o = lu_object_alloc(env, dev, f, conf);
+ if (IS_ERR(o))
+ return o;
-/**
- * Try to find the object in cache without waiting for the dead object
- * to be released nor allocating object if no cached one was found.
- *
- * The found object will be set as LU_OBJECT_HEARD_BANSHEE for purging.
- */
-void lu_object_purge(const struct lu_env *env, struct lu_device *dev,
- const struct lu_fid *f)
-{
- struct lu_site *s = dev->ld_site;
- cfs_hash_t *hs = s->ls_obj_hash;
- cfs_hash_bd_t bd;
- struct lu_object *o;
+ LASSERT(lu_fid_eq(lu_object_fid(o), f));
- cfs_hash_bd_get_and_lock(hs, f, &bd, 1);
- o = htable_lookup_nowait(s, &bd, f);
- cfs_hash_bd_unlock(hs, &bd, 1);
- if (!IS_ERR(o)) {
- set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
- lu_object_put(env, o);
+ cfs_hash_bd_lock(hs, &bd, 1);
+
+ if (conf && conf->loc_flags & LOC_F_NEW)
+ shadow = ERR_PTR(-ENOENT);
+ else
+ shadow = htable_lookup(s, &bd, f, &version);
+ if (likely(PTR_ERR(shadow) == -ENOENT)) {
+ cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
+ cfs_hash_bd_unlock(hs, &bd, 1);
+
+ lu_object_limit(env, dev);
+
+ return o;
}
+
+ lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE);
+ cfs_hash_bd_unlock(hs, &bd, 1);
+ lu_object_free(env, o);
+ return shadow;
}
-EXPORT_SYMBOL(lu_object_purge);
+EXPORT_SYMBOL(lu_object_find_at);
/**
* Find object with given fid, and return its slice belonging to given device.
}
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.
};
static int
-lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd,
+lu_site_obj_print(struct cfs_hash *hs, struct cfs_hash_bd *bd,
struct hlist_node *hnode, void *data)
{
struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data;
{
unsigned long cache_size;
unsigned long bits;
+ unsigned long bits_max = LU_SITE_BITS_MAX;
/*
* For ZFS based OSDs the cache should be disabled by default. This
return LU_SITE_BITS_MIN;
}
+ if (strcmp(top->ld_type->ldt_name, LUSTRE_VVP_NAME) == 0)
+ bits_max = LU_SITE_BITS_MAX_CL;
+
/*
* Calculate hash table size, assuming that we want reasonable
* performance when 20% of total memory is occupied by cache of
#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) {
;
}
- return bits;
+
+ return clamp_t(typeof(bits), bits, LU_SITE_BITS_MIN, bits_max);
}
-static unsigned lu_obj_hop_hash(cfs_hash_t *hs,
+static unsigned lu_obj_hop_hash(struct cfs_hash *hs,
const void *key, unsigned mask)
{
struct lu_fid *fid = (struct lu_fid *)key;
return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
}
-static void lu_obj_hop_get(cfs_hash_t *hs, struct hlist_node *hnode)
+static void lu_obj_hop_get(struct cfs_hash *hs, struct hlist_node *hnode)
{
struct lu_object_header *h;
h = hlist_entry(hnode, struct lu_object_header, loh_hash);
- if (atomic_add_return(1, &h->loh_ref) == 1) {
- struct lu_site_bkt_data *bkt;
- cfs_hash_bd_t bd;
-
- cfs_hash_bd_get(hs, &h->loh_fid, &bd);
- bkt = cfs_hash_bd_extra_get(hs, &bd);
- bkt->lsb_busy++;
- }
+ atomic_inc(&h->loh_ref);
}
-static void lu_obj_hop_put_locked(cfs_hash_t *hs, struct hlist_node *hnode)
+static void lu_obj_hop_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
{
LBUG(); /* we should never called it */
}
-cfs_hash_ops_t lu_site_hash_ops = {
+static struct cfs_hash_ops lu_site_hash_ops = {
.hs_hash = lu_obj_hop_hash,
.hs_key = lu_obj_hop_key,
.hs_keycmp = lu_obj_hop_keycmp,
int lu_site_init(struct lu_site *s, struct lu_device *top)
{
struct lu_site_bkt_data *bkt;
- cfs_hash_bd_t bd;
+ struct cfs_hash_bd bd;
char name[16];
unsigned long bits;
unsigned int i;
+ int rc;
ENTRY;
memset(s, 0, sizeof *s);
mutex_init(&s->ls_purge_mutex);
- bits = lu_htable_order(top);
+
+#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 = clamp_t(typeof(bits), bits,
- LU_SITE_BITS_MIN, LU_SITE_BITS_MAX);
+ for (bits = lu_htable_order(top);
bits >= LU_SITE_BITS_MIN; bits--) {
s->ls_obj_hash = cfs_hash_create(name, bits, bits,
bits - LU_SITE_BKT_BITS,
*/
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);
}
}
}
-EXPORT_SYMBOL(lu_stack_fini);
enum {
/**
static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
-static DEFINE_SPINLOCK(lu_keys_guard);
+static DECLARE_RWSEM(lu_key_initing);
/**
* Global counter incremented whenever key is registered, unregistered,
* lu_context_refill(). No locking is provided, as initialization and shutdown
* are supposed to be externally serialized.
*/
-static unsigned key_set_version = 0;
+static atomic_t key_set_version = ATOMIC_INIT(0);
/**
* Register new key.
LASSERT(key->lct_owner != NULL);
result = -ENFILE;
- spin_lock(&lu_keys_guard);
+ atomic_set(&key->lct_used, 1);
+ lu_ref_init(&key->lct_reference);
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
- if (lu_keys[i] == NULL) {
- key->lct_index = i;
- atomic_set(&key->lct_used, 1);
- lu_keys[i] = key;
- lu_ref_init(&key->lct_reference);
- result = 0;
- ++key_set_version;
- break;
- }
+ if (lu_keys[i])
+ continue;
+ key->lct_index = i;
+ if (cmpxchg(&lu_keys[i], NULL, key) != NULL)
+ continue;
+
+ result = 0;
+ atomic_inc(&key_set_version);
+ break;
}
- spin_unlock(&lu_keys_guard);
+ if (result) {
+ lu_ref_fini(&key->lct_reference);
+ atomic_set(&key->lct_used, 0);
+ }
return result;
}
EXPORT_SYMBOL(lu_context_key_register);
key = lu_keys[index];
LASSERT(key != NULL);
LASSERT(key->lct_fini != NULL);
- LASSERT(atomic_read(&key->lct_used) > 1);
+ LASSERT(atomic_read(&key->lct_used) > 0);
key->lct_fini(ctx, key, ctx->lc_value[index]);
lu_ref_del(&key->lct_reference, "ctx", ctx);
- atomic_dec(&key->lct_used);
+ if (atomic_dec_and_test(&key->lct_used))
+ wake_up_var(&key->lct_used);
LASSERT(key->lct_owner != NULL);
if ((ctx->lc_tags & LCT_NOREF) == 0) {
lu_context_key_quiesce(key);
- ++key_set_version;
- spin_lock(&lu_keys_guard);
key_fini(&lu_shrink_env.le_ctx, key->lct_index);
- if (lu_keys[key->lct_index]) {
- lu_keys[key->lct_index] = NULL;
+
+ /**
+ * Wait until all transient contexts referencing this key have
+ * run lu_context_key::lct_fini() method.
+ */
+ atomic_dec(&key->lct_used);
+ wait_var_event(&key->lct_used, atomic_read(&key->lct_used) == 0);
+
+ if (!WARN_ON(lu_keys[key->lct_index] == NULL))
lu_ref_fini(&key->lct_reference);
- }
- spin_unlock(&lu_keys_guard);
- LASSERTF(atomic_read(&key->lct_used) == 1,
- "key has instances: %d\n",
- atomic_read(&key->lct_used));
+ smp_store_release(&lu_keys[key->lct_index], NULL);
}
EXPORT_SYMBOL(lu_context_key_degister);
/**
* List of remembered contexts. XXX document me.
*/
-static struct list_head lu_context_remembered;
+static LIST_HEAD(lu_context_remembered);
+static DEFINE_SPINLOCK(lu_context_remembered_guard);
/**
* Destroy \a key in all remembered contexts. This is used to destroy key
*/
void lu_context_key_quiesce(struct lu_context_key *key)
{
- struct lu_context *ctx;
- extern unsigned cl_env_cache_purge(unsigned nr);
+ struct lu_context *ctx;
- if (!(key->lct_tags & LCT_QUIESCENT)) {
- /*
- * XXX layering violation.
- */
- cl_env_cache_purge(~0);
- key->lct_tags |= LCT_QUIESCENT;
+ if (!(key->lct_tags & LCT_QUIESCENT)) {
/*
- * XXX memory barrier has to go here.
+ * The write-lock on lu_key_initing will ensure that any
+ * keys_fill() which didn't see LCT_QUIESCENT will have
+ * finished before we call key_fini().
*/
- spin_lock(&lu_keys_guard);
- list_for_each_entry(ctx, &lu_context_remembered,
- lc_remember)
+ down_write(&lu_key_initing);
+ key->lct_tags |= LCT_QUIESCENT;
+ up_write(&lu_key_initing);
+
+ spin_lock(&lu_context_remembered_guard);
+ list_for_each_entry(ctx, &lu_context_remembered, lc_remember) {
+ spin_until_cond(READ_ONCE(ctx->lc_state) != LCS_LEAVING);
key_fini(ctx, key->lct_index);
- spin_unlock(&lu_keys_guard);
- ++key_set_version;
+ }
+
+ spin_unlock(&lu_context_remembered_guard);
}
}
-EXPORT_SYMBOL(lu_context_key_quiesce);
void lu_context_key_revive(struct lu_context_key *key)
{
- key->lct_tags &= ~LCT_QUIESCENT;
- ++key_set_version;
+ key->lct_tags &= ~LCT_QUIESCENT;
+ atomic_inc(&key_set_version);
}
-EXPORT_SYMBOL(lu_context_key_revive);
static void keys_fini(struct lu_context *ctx)
{
static int keys_fill(struct lu_context *ctx)
{
unsigned int i;
+ int rc = 0;
- 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);
-
- value = key->lct_init(ctx, key);
- if (unlikely(IS_ERR(value)))
- return PTR_ERR(value);
+ /*
+ * A serialisation with lu_context_key_quiesce() is needed, to
+ * ensure we see LCT_QUIESCENT and don't allocate a new value
+ * after it freed one. The rwsem provides this. As down_read()
+ * does optimistic spinning while the writer is active, this is
+ * unlikely to ever sleep.
+ */
+ down_read(&lu_key_initing);
+ ctx->lc_version = atomic_read(&key_set_version);
+
+ LINVRNT(ctx->lc_value);
+ for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
+ struct lu_context_key *key;
+
+ key = lu_keys[i];
+ if (!ctx->lc_value[i] && key &&
+ (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))
- try_module_get(key->lct_owner);
+ if (!(ctx->lc_tags & LCT_NOREF) &&
+ try_module_get(key->lct_owner) == 0) {
+ /* module is unloading, skip this key */
+ continue;
+ }
+
+ value = key->lct_init(ctx, key);
+ if (unlikely(IS_ERR(value))) {
+ rc = PTR_ERR(value);
+ break;
+ }
+
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;
- }
- return 0;
+ /*
+ * 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;
+ }
+ }
+
+ up_read(&lu_key_initing);
+ return rc;
}
static int keys_init(struct lu_context *ctx)
ctx->lc_state = LCS_INITIALIZED;
ctx->lc_tags = tags;
if (tags & LCT_REMEMBER) {
- spin_lock(&lu_keys_guard);
+ spin_lock(&lu_context_remembered_guard);
list_add(&ctx->lc_remember, &lu_context_remembered);
- spin_unlock(&lu_keys_guard);
+ spin_unlock(&lu_context_remembered_guard);
} else {
INIT_LIST_HEAD(&ctx->lc_remember);
}
if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
LASSERT(list_empty(&ctx->lc_remember));
- keys_fini(ctx);
-
- } else { /* could race with key degister */
- spin_lock(&lu_keys_guard);
- keys_fini(ctx);
+ } else {
+ /* could race with key degister */
+ spin_lock(&lu_context_remembered_guard);
list_del_init(&ctx->lc_remember);
- spin_unlock(&lu_keys_guard);
+ spin_unlock(&lu_context_remembered_guard);
}
+ keys_fini(ctx);
}
EXPORT_SYMBOL(lu_context_fini);
{
unsigned int i;
- LINVRNT(ctx->lc_state == LCS_ENTERED);
- ctx->lc_state = LCS_LEFT;
- if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) {
+ LINVRNT(ctx->lc_state == LCS_ENTERED);
+ /*
+ * Disable preempt to ensure we get a warning if
+ * any lct_exit ever tries to sleep. That would hurt
+ * lu_context_key_quiesce() which spins waiting for us.
+ * This also ensure we aren't preempted while the state
+ * is LCS_LEAVING, as that too would cause problems for
+ * lu_context_key_quiesce().
+ */
+ preempt_disable();
+ /*
+ * Ensure lu_context_key_quiesce() sees LCS_LEAVING
+ * or we see LCT_QUIESCENT
+ */
+ smp_store_mb(ctx->lc_state, LCS_LEAVING);
+ if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value) {
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
- if (ctx->lc_value[i] != NULL) {
- struct lu_context_key *key;
-
- key = lu_keys[i];
- LASSERT(key != NULL);
- if (key->lct_exit != NULL)
- key->lct_exit(ctx,
- key, ctx->lc_value[i]);
- }
- }
+ struct lu_context_key *key;
+
+ key = lu_keys[i];
+ if (ctx->lc_value[i] &&
+ !(key->lct_tags & LCT_QUIESCENT) &&
+ key->lct_exit)
+ key->lct_exit(ctx, key, ctx->lc_value[i]);
+ }
}
+
+ smp_store_release(&ctx->lc_state, LCS_LEFT);
+ preempt_enable();
}
EXPORT_SYMBOL(lu_context_exit);
*/
int lu_context_refill(struct lu_context *ctx)
{
- return likely(ctx->lc_version == key_set_version) ? 0 : keys_fill(ctx);
+ if (likely(ctx->lc_version == atomic_read(&key_set_version)))
+ return 0;
+
+ return keys_fill(ctx);
}
-EXPORT_SYMBOL(lu_context_refill);
/**
* lu_ctx_tags/lu_ses_tags will be updated if there are new types of
* predefined when the lu_device type are registered, during the module probe
* phase.
*/
-__u32 lu_context_tags_default = 0;
-__u32 lu_session_tags_default = 0;
+u32 lu_context_tags_default;
+u32 lu_session_tags_default;
+#ifdef HAVE_SERVER_SUPPORT
void lu_context_tags_update(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ spin_lock(&lu_context_remembered_guard);
lu_context_tags_default |= tags;
- key_set_version++;
- spin_unlock(&lu_keys_guard);
+ atomic_inc(&key_set_version);
+ spin_unlock(&lu_context_remembered_guard);
}
EXPORT_SYMBOL(lu_context_tags_update);
void lu_context_tags_clear(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ spin_lock(&lu_context_remembered_guard);
lu_context_tags_default &= ~tags;
- key_set_version++;
- spin_unlock(&lu_keys_guard);
+ atomic_inc(&key_set_version);
+ spin_unlock(&lu_context_remembered_guard);
}
EXPORT_SYMBOL(lu_context_tags_clear);
void lu_session_tags_update(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ spin_lock(&lu_context_remembered_guard);
lu_session_tags_default |= tags;
- key_set_version++;
- spin_unlock(&lu_keys_guard);
+ atomic_inc(&key_set_version);
+ spin_unlock(&lu_context_remembered_guard);
}
EXPORT_SYMBOL(lu_session_tags_update);
void lu_session_tags_clear(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ spin_lock(&lu_context_remembered_guard);
lu_session_tags_default &= ~tags;
- key_set_version++;
- spin_unlock(&lu_keys_guard);
+ atomic_inc(&key_set_version);
+ spin_unlock(&lu_context_remembered_guard);
}
EXPORT_SYMBOL(lu_session_tags_clear);
+#endif /* HAVE_SERVER_SUPPORT */
int lu_env_init(struct lu_env *env, __u32 tags)
{
}
EXPORT_SYMBOL(lu_env_refill_by_tags);
+
+struct lu_env_item {
+ struct task_struct *lei_task; /* rhashtable key */
+ struct rhash_head lei_linkage;
+ struct lu_env *lei_env;
+};
+
+static const struct rhashtable_params lu_env_rhash_params = {
+ .key_len = sizeof(struct task_struct *),
+ .key_offset = offsetof(struct lu_env_item, lei_task),
+ .head_offset = offsetof(struct lu_env_item, lei_linkage),
+ };
+
+struct rhashtable lu_env_rhash;
+
+struct lu_env_percpu {
+ struct task_struct *lep_task;
+ struct lu_env *lep_env ____cacheline_aligned_in_smp;
+};
+
+static struct lu_env_percpu lu_env_percpu[NR_CPUS];
+
+int lu_env_add(struct lu_env *env)
+{
+ struct lu_env_item *lei, *old;
+
+ LASSERT(env);
+
+ OBD_ALLOC_PTR(lei);
+ if (!lei)
+ return -ENOMEM;
+
+ lei->lei_task = current;
+ lei->lei_env = env;
+
+ old = rhashtable_lookup_get_insert_fast(&lu_env_rhash,
+ &lei->lei_linkage,
+ lu_env_rhash_params);
+ LASSERT(!old);
+
+ return 0;
+}
+EXPORT_SYMBOL(lu_env_add);
+
+void lu_env_remove(struct lu_env *env)
+{
+ struct lu_env_item *lei;
+ const void *task = current;
+ int i;
+
+ for_each_possible_cpu(i) {
+ if (lu_env_percpu[i].lep_env == env) {
+ LASSERT(lu_env_percpu[i].lep_task == task);
+ lu_env_percpu[i].lep_task = NULL;
+ lu_env_percpu[i].lep_env = NULL;
+ }
+ }
+
+ rcu_read_lock();
+ lei = rhashtable_lookup_fast(&lu_env_rhash, &task,
+ lu_env_rhash_params);
+ if (lei && rhashtable_remove_fast(&lu_env_rhash, &lei->lei_linkage,
+ lu_env_rhash_params) == 0)
+ OBD_FREE_PTR(lei);
+ rcu_read_unlock();
+}
+EXPORT_SYMBOL(lu_env_remove);
+
+struct lu_env *lu_env_find(void)
+{
+ struct lu_env *env = NULL;
+ struct lu_env_item *lei;
+ const void *task = current;
+ int i = get_cpu();
+
+ if (lu_env_percpu[i].lep_task == current) {
+ env = lu_env_percpu[i].lep_env;
+ put_cpu();
+ LASSERT(env);
+ return env;
+ }
+
+ lei = rhashtable_lookup_fast(&lu_env_rhash, &task,
+ lu_env_rhash_params);
+ if (lei) {
+ env = lei->lei_env;
+ lu_env_percpu[i].lep_task = current;
+ lu_env_percpu[i].lep_env = env;
+ }
+ put_cpu();
+
+ return env;
+}
+EXPORT_SYMBOL(lu_env_find);
+
static struct shrinker *lu_site_shrinker;
typedef struct lu_site_stats{
unsigned lss_busy;
} lu_site_stats_t;
-static void lu_site_stats_get(cfs_hash_t *hs,
+static void lu_site_stats_get(const struct lu_site *s,
lu_site_stats_t *stats, int populated)
{
- cfs_hash_bd_t bd;
- unsigned int i;
+ struct cfs_hash *hs = s->ls_obj_hash;
+ struct cfs_hash_bd bd;
+ unsigned int i;
+ /*
+ * percpu_counter_sum_positive() won't accept a const pointer
+ * as it does modify the struct by taking a spinlock
+ */
+ struct lu_site *s2 = (struct lu_site *)s;
+ stats->lss_busy += cfs_hash_size_get(hs) -
+ percpu_counter_sum_positive(&s2->ls_lru_len_counter);
cfs_hash_for_each_bucket(hs, &bd, i) {
- struct lu_site_bkt_data *bkt = cfs_hash_bd_extra_get(hs, &bd);
- struct hlist_head *hhead;
+ struct hlist_head *hhead;
cfs_hash_bd_lock(hs, &bd, 1);
- stats->lss_busy += bkt->lsb_busy;
stats->lss_total += cfs_hash_bd_count_get(&bd);
stats->lss_max_search = max((int)stats->lss_max_search,
cfs_hash_bd_depmax_get(&bd));
}
+/*
+ * 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
+ * lu_cache_shrink_count can sum the counters without locking each
+ * ls_obj_hash bucket.
+ */
static unsigned long lu_cache_shrink_count(struct shrinker *sk,
struct shrink_control *sc)
{
- lu_site_stats_t stats;
struct lu_site *s;
struct lu_site *tmp;
unsigned long cached = 0;
if (!(sc->gfp_mask & __GFP_FS))
return 0;
- mutex_lock(&lu_sites_guard);
- list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
- memset(&stats, 0, sizeof(stats));
- lu_site_stats_get(s->ls_obj_hash, &stats, 0);
- cached += stats.lss_total - stats.lss_busy;
- }
- 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\n", cached);
+ CDEBUG(D_INODE, "%ld objects cached, cache pressure %d\n",
+ cached, sysctl_vfs_cache_pressure);
+
return cached;
}
*/
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;
}
* is safe to take the lu_sites_guard lock.
*
* Ideally we should accurately return the remaining number of cached
- * objects without taking the lu_sites_guard lock, but this is not
+ * objects without taking the lu_sites_guard lock, but this is not
* possible in the current implementation.
*/
static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
CDEBUG(D_INODE, "Shrink %lu objects\n", scv.nr_to_scan);
- lu_cache_shrink_scan(shrinker, &scv);
+ if (scv.nr_to_scan != 0)
+ lu_cache_shrink_scan(shrinker, &scv);
cached = lu_cache_shrink_count(shrinker, &scv);
- if (scv.nr_to_scan == 0)
- CDEBUG(D_INODE, "%d objects cached\n", cached);
return cached;
}
/**
* Environment to be used in debugger, contains all tags.
*/
-struct lu_env lu_debugging_env;
+static struct lu_env lu_debugging_env;
/**
* Debugging printer function using printk().
*/
int lu_printk_printer(const struct lu_env *env,
- void *unused, const char *format, ...)
+ void *unused, const char *format, ...)
{
va_list args;
}
}
}
-EXPORT_SYMBOL(lu_context_keys_dump);
/**
* Initialization of global lu_* data.
*/
int lu_global_init(void)
{
- int result;
+ int result;
DEF_SHRINKER_VAR(shvar, lu_cache_shrink,
lu_cache_shrink_count, lu_cache_shrink_scan);
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;
if (lu_site_shrinker == NULL)
return -ENOMEM;
+ result = rhashtable_init(&lu_env_rhash, &lu_env_rhash_params);
+
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);
+
+ rhashtable_destroy(&lu_env_rhash);
lu_ref_global_fini();
}
static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx)
{
-#ifdef LPROCFS
- struct lprocfs_counter ret;
+#ifdef CONFIG_PROC_FS
+ struct lprocfs_counter ret;
- lprocfs_stats_collect(stats, idx, &ret);
- return (__u32)ret.lc_count;
+ lprocfs_stats_collect(stats, idx, &ret);
+ return (__u32)ret.lc_count;
#else
- return 0;
+ return 0;
#endif
}
lu_site_stats_t stats;
memset(&stats, 0, sizeof(stats));
- lu_site_stats_get(s->ls_obj_hash, &stats, 1);
-
- return seq_printf(m, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
- stats.lss_busy,
- stats.lss_total,
- stats.lss_populated,
- CFS_HASH_NHLIST(s->ls_obj_hash),
- stats.lss_max_search,
- ls_stats_read(s->ls_stats, LU_SS_CREATED),
- ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
- 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));
+ lu_site_stats_get(s, &stats, 1);
+
+ seq_printf(m, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
+ stats.lss_busy,
+ stats.lss_total,
+ stats.lss_populated,
+ CFS_HASH_NHLIST(s->ls_obj_hash),
+ stats.lss_max_search,
+ ls_stats_read(s->ls_stats, LU_SS_CREATED),
+ ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
+ 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));
+ return 0;
}
EXPORT_SYMBOL(lu_site_stats_seq_print);
-int lu_site_stats_print(const struct lu_site *s, char *page, int count)
-{
- lu_site_stats_t stats;
-
- memset(&stats, 0, sizeof(stats));
- lu_site_stats_get(s->ls_obj_hash, &stats, 1);
-
- return snprintf(page, count, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
- stats.lss_busy,
- stats.lss_total,
- stats.lss_populated,
- CFS_HASH_NHLIST(s->ls_obj_hash),
- stats.lss_max_search,
- ls_stats_read(s->ls_stats, LU_SS_CREATED),
- ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
- 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));
-}
-EXPORT_SYMBOL(lu_site_stats_print);
-
/**
* Helper function to initialize a number of kmem slab caches at once.
*/
{
struct lu_site *s = o->lo_dev->ld_site;
struct lu_fid *old = &o->lo_header->loh_fid;
- struct lu_site_bkt_data *bkt;
- struct lu_object *shadow;
- wait_queue_t waiter;
- cfs_hash_t *hs;
- cfs_hash_bd_t bd;
- __u64 version = 0;
+ struct cfs_hash *hs;
+ struct cfs_hash_bd bd;
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;
+ struct lu_object *shadow;
+
+ shadow = htable_lookup(s, &bd, fid, &version);
+ /* supposed to be unique */
+ LASSERT(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT);
+ }
+#endif
*old = *fid;
- bkt = cfs_hash_bd_extra_get(hs, &bd);
cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
- bkt->lsb_busy++;
cfs_hash_bd_unlock(hs, &bd, 1);
}
EXPORT_SYMBOL(lu_object_assign_fid);
return 0;
}
EXPORT_SYMBOL(lu_buf_check_and_grow);
-