*
* 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, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_CLASS
+#include <linux/module.h>
+#include <linux/list.h>
#include <libcfs/libcfs.h>
-
-#ifdef __KERNEL__
-# include <linux/module.h>
-#endif
-
-/* hash_long() */
-#include <libcfs/libcfs_hash.h>
+#include <libcfs/libcfs_hash.h> /* hash_long() */
#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>
+
+enum {
+ LU_CACHE_PERCENT_MAX = 50,
+ LU_CACHE_PERCENT_DEFAULT = 20
+};
+
+#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
+/** This is set to roughly (20 * OSS_NTHRS_MAX) to prevent thrashing */
+#define LU_CACHE_NR_ZFS_LIMIT 10240
+
+#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
+ * - avoid unbalanced LRU list
+ */
+#define LU_SITE_BKT_BITS 8
+
+
+static unsigned int lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
+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;
+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);
/**
* Decrease reference counter on object. If last reference is freed, return
struct lu_object_header *top;
struct lu_site *site;
struct lu_object *orig;
- cfs_hash_bd_t bd;
+ struct cfs_hash_bd bd;
const struct lu_fid *fid;
top = o->lo_header;
if (fid_is_zero(fid)) {
LASSERT(top->loh_hash.next == NULL
&& top->loh_hash.pprev == NULL);
- LASSERT(cfs_list_empty(&top->loh_lru));
- if (!cfs_atomic_dec_and_test(&top->loh_ref))
+ LASSERT(list_empty(&top->loh_lru));
+ if (!atomic_dec_and_test(&top->loh_ref))
return;
- cfs_list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
+ 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);
}
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.
*/
- cfs_list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
+ 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 (!lu_object_is_dying(top)) {
- LASSERT(cfs_list_empty(&top->loh_lru));
- cfs_list_add_tail(&top->loh_lru, &bkt->lsb_lru);
+ 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);
+ bkt->lsb_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);
cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
return;
}
/*
- * If object is dying (will not be cached), removed it
+ * 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
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);
- cfs_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);
+ bkt->lsb_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);
}
{
struct lu_object *scan;
struct lu_object *top;
- cfs_list_t *layers;
+ struct list_head *layers;
unsigned int init_mask = 0;
unsigned int init_flag;
int clean;
*/
clean = 1;
init_flag = 1;
- cfs_list_for_each_entry(scan, layers, lo_linkage) {
+ list_for_each_entry(scan, layers, lo_linkage) {
if (init_mask & init_flag)
goto next;
clean = 0;
}
} while (!clean);
- cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
+ list_for_each_entry_reverse(scan, layers, lo_linkage) {
if (scan->lo_ops->loo_object_start != NULL) {
result = scan->lo_ops->loo_object_start(env, scan);
if (result != 0) {
*/
static void lu_object_free(const struct lu_env *env, struct lu_object *o)
{
- struct lu_site_bkt_data *bkt;
- struct lu_site *site;
- struct lu_object *scan;
- cfs_list_t *layers;
- cfs_list_t splice;
+ struct lu_site_bkt_data *bkt;
+ 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;
/*
* First call ->loo_object_delete() method to release all resources.
*/
- cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
+ list_for_each_entry_reverse(scan, layers, lo_linkage) {
if (scan->lo_ops->loo_object_delete != NULL)
scan->lo_ops->loo_object_delete(env, scan);
}
* necessary, because lu_object_header is freed together with the
* top-level slice.
*/
- CFS_INIT_LIST_HEAD(&splice);
- cfs_list_splice_init(layers, &splice);
- while (!cfs_list_empty(&splice)) {
+ INIT_LIST_HEAD(&splice);
+ list_splice_init(layers, &splice);
+ while (!list_empty(&splice)) {
/*
* Free layers in bottom-to-top order, so that object header
* lives as long as possible and ->loo_object_free() methods
* can look at its contents.
*/
o = container_of0(splice.prev, struct lu_object, lo_linkage);
- cfs_list_del_init(&o->lo_linkage);
+ list_del_init(&o->lo_linkage);
LASSERT(o->lo_ops->loo_object_free != NULL);
o->lo_ops->loo_object_free(env, o);
}
/**
* 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;
- cfs_list_t dispose;
- int did_sth;
- int start;
+ struct cfs_hash_bd bd;
+ struct cfs_hash_bd bd2;
+ struct list_head dispose;
+ int did_sth;
+ unsigned int start = 0;
int count;
int bnr;
- int i;
+ unsigned int i;
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_NO_LRU))
RETURN(0);
- CFS_INIT_LIST_HEAD(&dispose);
+ INIT_LIST_HEAD(&dispose);
/*
* 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;
- bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1;
+ 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.
+ */
+ 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_lock(s->ls_obj_hash, &bd, 1);
bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
- cfs_list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
- LASSERT(cfs_atomic_read(&h->loh_ref) == 0);
+ list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
+ LASSERT(atomic_read(&h->loh_ref) == 0);
cfs_hash_bd_get(s->ls_obj_hash, &h->loh_fid, &bd2);
LASSERT(bd.bd_bucket == bd2.bd_bucket);
cfs_hash_bd_del_locked(s->ls_obj_hash,
&bd2, &h->loh_hash);
- cfs_list_move(&h->loh_lru, &dispose);
+ list_move(&h->loh_lru, &dispose);
+ bkt->lsb_lru_len--;
+ percpu_counter_dec(&s->ls_lru_len_counter);
if (did_sth == 0)
did_sth = 1;
* Free everything on the dispose list. This is safe against
* races due to the reasons described in lu_object_put().
*/
- while (!cfs_list_empty(&dispose)) {
- h = container_of0(dispose.next,
- struct lu_object_header, loh_lru);
- cfs_list_del_init(&h->loh_lru);
- lu_object_free(env, lu_object_top(h));
- lprocfs_counter_incr(s->ls_stats, LU_SS_LRU_PURGED);
- }
+ while (!list_empty(&dispose)) {
+ h = container_of0(dispose.next,
+ struct lu_object_header, loh_lru);
+ list_del_init(&h->loh_lru);
+ lu_object_free(env, lu_object_top(h));
+ lprocfs_counter_incr(s->ls_stats, LU_SS_LRU_PURGED);
+ }
if (nr == 0)
break;
}
+ mutex_unlock(&s->ls_purge_mutex);
if (nr != 0 && did_sth && start != 0) {
start = 0; /* restart from the first bucket */
/* 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,
vsnprintf(key->lck_area + used,
ARRAY_SIZE(key->lck_area) - used, format, args);
if (complete) {
- if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
- libcfs_debug_msg(msgdata, "%s", key->lck_area);
+ if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
+ libcfs_debug_msg(msgdata, "%s\n", key->lck_area);
key->lck_area[0] = 0;
}
va_end(args);
lu_printer_t printer,
const struct lu_object_header *hdr)
{
- (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]",
- hdr, hdr->loh_flags, cfs_atomic_read(&hdr->loh_ref),
- PFID(&hdr->loh_fid),
- cfs_hlist_unhashed(&hdr->loh_hash) ? "" : " hash",
- cfs_list_empty((cfs_list_t *)&hdr->loh_lru) ? \
- "" : " lru",
- hdr->loh_attr & LOHA_EXISTS ? " exist":"");
+ (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]",
+ hdr, hdr->loh_flags, atomic_read(&hdr->loh_ref),
+ PFID(&hdr->loh_fid),
+ hlist_unhashed(&hdr->loh_hash) ? "" : " hash",
+ list_empty((struct list_head *)&hdr->loh_lru) ? \
+ "" : " lru",
+ hdr->loh_attr & LOHA_EXISTS ? " exist" : "");
}
EXPORT_SYMBOL(lu_object_header_print);
lu_object_header_print(env, cookie, printer, top);
(*printer)(env, cookie, "{\n");
- cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
+ list_for_each_entry(o, &top->loh_layers, lo_linkage) {
/*
* print `.' \a depth times followed by type name and address
*/
struct lu_object_header *top;
top = o->lo_header;
- cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
+ list_for_each_entry(o, &top->loh_layers, lo_linkage) {
if (o->lo_ops->loo_object_invariant != NULL &&
!o->lo_ops->loo_object_invariant(o))
return 0;
}
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;
- cfs_hlist_node_t *hnode;
- __u64 ver = cfs_hash_bd_version_get(bd);
+ struct lu_site_bkt_data *bkt;
+ struct lu_object_header *h;
+ struct hlist_node *hnode;
+ __u64 ver = cfs_hash_bd_version_get(bd);
if (*version == ver)
return ERR_PTR(-ENOENT);
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);
- cfs_list_del_init(&h->loh_lru);
+ if (!list_empty(&h->loh_lru)) {
+ list_del_init(&h->loh_lru);
+ bkt->lsb_lru_len--;
+ percpu_counter_dec(&s->ls_lru_len_counter);
+ }
return lu_object_top(h);
}
* drained), and moreover, lookup has to wait until object is freed.
*/
- 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)
-{
- cfs_hlist_node_t *hnode;
- struct lu_object_header *h;
-
- /* 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);
- return ERR_PTR(-ENOENT);
+ if (likely(waiter != NULL)) {
+ init_waitqueue_entry(waiter, current);
+ add_wait_queue(&bkt->lsb_marche_funebre, waiter);
+ set_current_state(TASK_UNINTERRUPTIBLE);
+ lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE);
}
- 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);
- cfs_list_del_init(&h->loh_lru);
- return lu_object_top(h);
+ return ERR_PTR(-EAGAIN);
}
/**
}
EXPORT_SYMBOL(lu_object_find);
+/*
+ * Limit the lu_object cache to a maximum of lu_cache_nr objects. Because
+ * the calculation for the number of objects to reclaim is not covered by
+ * a lock the maximum number of objects is capped by LU_CACHE_MAX_ADJUST.
+ * This ensures that many concurrent threads will not accidentally purge
+ * the entire cache.
+ */
+static void lu_object_limit(const struct lu_env *env,
+ struct lu_device *dev)
+{
+ __u64 size, nr;
+
+ if (lu_cache_nr == LU_CACHE_NR_UNLIMITED)
+ return;
+
+ size = cfs_hash_size_get(dev->ld_site->ls_obj_hash);
+ nr = (__u64)lu_cache_nr;
+ if (size <= nr)
+ 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,
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;
+ struct cfs_hash *hs;
+ struct cfs_hash_bd bd;
o = lu_object_alloc(env, dev, f, conf);
if (unlikely(IS_ERR(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);
+
return o;
}
struct lu_object *o;
struct lu_object *shadow;
struct lu_site *s;
- cfs_hash_t *hs;
- cfs_hash_bd_t bd;
+ struct cfs_hash *hs;
+ struct cfs_hash_bd bd;
__u64 version = 0;
/*
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;
}
struct lu_object *obj;
wait_queue_t wait;
+ if (conf != NULL && conf->loc_flags & LOC_F_NOWAIT)
+ return lu_object_find_try(env, dev, f, conf, NULL);
+
while (1) {
obj = lu_object_find_try(env, dev, f, conf, &wait);
if (obj != ERR_PTR(-EAGAIN))
* lu_object_find_try() already added waiter into the
* wait queue.
*/
- waitq_wait(&wait, TASK_UNINTERRUPTIBLE);
+ schedule();
bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f);
remove_wait_queue(&bkt->lsb_marche_funebre, &wait);
}
EXPORT_SYMBOL(lu_object_find_at);
/**
- * 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;
-
- 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);
- }
-}
-EXPORT_SYMBOL(lu_object_purge);
-
-/**
* Find object with given fid, and return its slice belonging to given device.
*/
struct lu_object *lu_object_find_slice(const struct lu_env *env,
const struct lu_fid *f,
const struct lu_object_conf *conf)
{
- struct lu_object *top;
- struct lu_object *obj;
+ struct lu_object *top;
+ struct lu_object *obj;
- top = lu_object_find(env, dev, f, conf);
- if (!IS_ERR(top)) {
- obj = lu_object_locate(top->lo_header, dev->ld_type);
- if (obj == NULL)
- lu_object_put(env, top);
- } else
- obj = top;
- return obj;
+ top = lu_object_find(env, dev, f, conf);
+ if (IS_ERR(top))
+ return top;
+
+ obj = lu_object_locate(top->lo_header, dev->ld_type);
+ if (unlikely(obj == NULL)) {
+ lu_object_put(env, top);
+ obj = ERR_PTR(-ENOENT);
+ }
+
+ return obj;
}
EXPORT_SYMBOL(lu_object_find_slice);
/**
* Global list of all device types.
*/
-static CFS_LIST_HEAD(lu_device_types);
+static struct list_head lu_device_types;
int lu_device_type_init(struct lu_device_type *ldt)
{
int result = 0;
- CFS_INIT_LIST_HEAD(&ldt->ldt_linkage);
+ 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)
- cfs_list_add(&ldt->ldt_linkage, &lu_device_types);
+
+ 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)
{
- cfs_list_del_init(&ldt->ldt_linkage);
+ 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);
}
EXPORT_SYMBOL(lu_device_type_fini);
-void lu_types_stop(void)
-{
- struct lu_device_type *ldt;
-
- cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
- if (ldt->ldt_device_nr == 0 && ldt->ldt_ops->ldto_stop)
- ldt->ldt_ops->ldto_stop(ldt);
- }
-}
-EXPORT_SYMBOL(lu_types_stop);
-
/**
* Global list of all sites on this node
*/
-static CFS_LIST_HEAD(lu_sites);
-static DEFINE_MUTEX(lu_sites_guard);
+static struct list_head lu_sites;
+static struct rw_semaphore lu_sites_guard;
/**
* Global environment used by site shrinker.
};
static int
-lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd,
- cfs_hlist_node_t *hnode, void *data)
-{
- struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data;
- struct lu_object_header *h;
-
- h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
- if (!cfs_list_empty(&h->loh_layers)) {
- const struct lu_object *o;
-
- o = lu_object_top(h);
- lu_object_print(arg->lsp_env, arg->lsp_cookie,
- arg->lsp_printer, o);
- } else {
- lu_object_header_print(arg->lsp_env, arg->lsp_cookie,
- arg->lsp_printer, h);
- }
- return 0;
+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;
+ struct lu_object_header *h;
+
+ h = hlist_entry(hnode, struct lu_object_header, loh_hash);
+ if (!list_empty(&h->loh_layers)) {
+ const struct lu_object *o;
+
+ o = lu_object_top(h);
+ lu_object_print(arg->lsp_env, arg->lsp_cookie,
+ arg->lsp_printer, o);
+ } else {
+ lu_object_header_print(arg->lsp_env, arg->lsp_cookie,
+ arg->lsp_printer, h);
+ }
+ return 0;
}
/**
}
EXPORT_SYMBOL(lu_site_print);
-enum {
- LU_CACHE_PERCENT_MAX = 50,
- LU_CACHE_PERCENT_DEFAULT = 20
-};
-
-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");
-
/**
* Return desired hash table order.
*/
-static int lu_htable_order(void)
+static unsigned long lu_htable_order(struct lu_device *top)
{
- unsigned long cache_size;
- int bits;
+ 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
+ * allows the ZFS ARC maximum flexibility in determining what buffers
+ * to cache. If Lustre has objects or buffer which it wants to ensure
+ * always stay cached it must maintain a hold on them.
+ */
+ if (strcmp(top->ld_type->ldt_name, LUSTRE_OSD_ZFS_NAME) == 0) {
+ lu_cache_percent = 1;
+ lu_cache_nr = LU_CACHE_NR_ZFS_LIMIT;
+ 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
*
* Size of lu_object is (arbitrary) taken as 1K (together with inode).
*/
- cache_size = num_physpages;
+ cache_size = totalram_pages;
#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,
- const void *key, unsigned mask)
+static unsigned lu_obj_hop_hash(struct cfs_hash *hs,
+ const void *key, unsigned mask)
{
- struct lu_fid *fid = (struct lu_fid *)key;
- __u32 hash;
+ struct lu_fid *fid = (struct lu_fid *)key;
+ __u32 hash;
- hash = fid_flatten32(fid);
- hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
- hash = cfs_hash_long(hash, hs->hs_bkt_bits);
+ hash = fid_flatten32(fid);
+ hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
+ hash = hash_long(hash, hs->hs_bkt_bits);
- /* give me another random factor */
- hash -= cfs_hash_long((unsigned long)hs, fid_oid(fid) % 11 + 3);
+ /* give me another random factor */
+ hash -= hash_long((unsigned long)hs, fid_oid(fid) % 11 + 3);
- hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
- hash |= (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
+ hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
+ hash |= (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
- return hash & mask;
+ return hash & mask;
}
-static void *lu_obj_hop_object(cfs_hlist_node_t *hnode)
+static void *lu_obj_hop_object(struct hlist_node *hnode)
{
- return cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
+ return hlist_entry(hnode, struct lu_object_header, loh_hash);
}
-static void *lu_obj_hop_key(cfs_hlist_node_t *hnode)
+static void *lu_obj_hop_key(struct hlist_node *hnode)
{
- struct lu_object_header *h;
+ struct lu_object_header *h;
- h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
- return &h->loh_fid;
+ h = hlist_entry(hnode, struct lu_object_header, loh_hash);
+ return &h->loh_fid;
}
-static int lu_obj_hop_keycmp(const void *key, cfs_hlist_node_t *hnode)
+static int lu_obj_hop_keycmp(const void *key, struct hlist_node *hnode)
{
- struct lu_object_header *h;
+ struct lu_object_header *h;
- h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
- return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
+ h = hlist_entry(hnode, struct lu_object_header, loh_hash);
+ return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
}
-static void lu_obj_hop_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+static void lu_obj_hop_get(struct cfs_hash *hs, struct hlist_node *hnode)
{
- struct lu_object_header *h;
-
- h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
- if (cfs_atomic_add_return(1, &h->loh_ref) == 1) {
- struct lu_site_bkt_data *bkt;
- cfs_hash_bd_t bd;
+ struct lu_object_header *h;
- cfs_hash_bd_get(hs, &h->loh_fid, &bd);
- bkt = cfs_hash_bd_extra_get(hs, &bd);
- bkt->lsb_busy++;
- }
+ h = hlist_entry(hnode, struct lu_object_header, loh_hash);
+ atomic_inc(&h->loh_ref);
}
-static void lu_obj_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *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,
void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d)
{
spin_lock(&s->ls_ld_lock);
- if (cfs_list_empty(&d->ld_linkage))
- cfs_list_add(&d->ld_linkage, &s->ls_ld_linkage);
+ if (list_empty(&d->ld_linkage))
+ list_add(&d->ld_linkage, &s->ls_ld_linkage);
spin_unlock(&s->ls_ld_lock);
}
EXPORT_SYMBOL(lu_dev_add_linkage);
void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d)
{
spin_lock(&s->ls_ld_lock);
- cfs_list_del_init(&d->ld_linkage);
+ list_del_init(&d->ld_linkage);
spin_unlock(&s->ls_ld_lock);
}
EXPORT_SYMBOL(lu_dev_del_linkage);
/**
- * Initialize site \a s, with \a d as the top level device.
- */
-#define LU_SITE_BITS_MIN 12
-#define LU_SITE_BITS_MAX 24
-/**
- * total 256 buckets, we don't want too many buckets because:
- * - consume too much memory
- * - avoid unbalanced LRU list
- */
-#define LU_SITE_BKT_BITS 8
-
+ * Initialize site \a s, with \a d as the top level device.
+ */
int lu_site_init(struct lu_site *s, struct lu_device *top)
{
- struct lu_site_bkt_data *bkt;
- cfs_hash_bd_t bd;
- char name[16];
- int bits;
- int i;
- ENTRY;
-
- memset(s, 0, sizeof *s);
- bits = lu_htable_order();
- snprintf(name, 16, "lu_site_%s", top->ld_type->ldt_name);
- for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX);
- bits >= LU_SITE_BITS_MIN; bits--) {
- s->ls_obj_hash = cfs_hash_create(name, bits, bits,
- bits - LU_SITE_BKT_BITS,
- sizeof(*bkt), 0, 0,
- &lu_site_hash_ops,
- CFS_HASH_SPIN_BKTLOCK |
- CFS_HASH_NO_ITEMREF |
- CFS_HASH_DEPTH |
- CFS_HASH_ASSERT_EMPTY);
- if (s->ls_obj_hash != NULL)
- break;
- }
+ struct lu_site_bkt_data *bkt;
+ struct cfs_hash_bd bd;
+ char name[16];
+ unsigned long bits;
+ unsigned int i;
+ int rc;
+ ENTRY;
- if (s->ls_obj_hash == NULL) {
- CERROR("failed to create lu_site hash with bits: %d\n", bits);
- return -ENOMEM;
- }
+ 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--) {
+ s->ls_obj_hash = cfs_hash_create(name, bits, bits,
+ bits - LU_SITE_BKT_BITS,
+ sizeof(*bkt), 0, 0,
+ &lu_site_hash_ops,
+ CFS_HASH_SPIN_BKTLOCK |
+ CFS_HASH_NO_ITEMREF |
+ CFS_HASH_DEPTH |
+ CFS_HASH_ASSERT_EMPTY |
+ CFS_HASH_COUNTER);
+ if (s->ls_obj_hash != NULL)
+ break;
+ }
+
+ if (s->ls_obj_hash == NULL) {
+ CERROR("failed to create lu_site hash with bits: %lu\n", bits);
+ return -ENOMEM;
+ }
cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
- CFS_INIT_LIST_HEAD(&bkt->lsb_lru);
+ INIT_LIST_HEAD(&bkt->lsb_lru);
init_waitqueue_head(&bkt->lsb_marche_funebre);
}
lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED,
0, "lru_purged", "lru_purged");
- CFS_INIT_LIST_HEAD(&s->ls_linkage);
+ INIT_LIST_HEAD(&s->ls_linkage);
s->ls_top_dev = top;
top->ld_site = s;
lu_device_get(top);
lu_ref_add(&top->ld_reference, "site-top", s);
- CFS_INIT_LIST_HEAD(&s->ls_ld_linkage);
+ INIT_LIST_HEAD(&s->ls_ld_linkage);
spin_lock_init(&s->ls_ld_lock);
lu_dev_add_linkage(s, top);
*/
void lu_site_fini(struct lu_site *s)
{
- mutex_lock(&lu_sites_guard);
- cfs_list_del_init(&s->ls_linkage);
- mutex_unlock(&lu_sites_guard);
+ down_write(&lu_sites_guard);
+ list_del_init(&s->ls_linkage);
+ 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)
- cfs_list_add(&s->ls_linkage, &lu_sites);
- mutex_unlock(&lu_sites_guard);
+ list_add(&s->ls_linkage, &lu_sites);
+ up_write(&lu_sites_guard);
return result;
}
EXPORT_SYMBOL(lu_site_init_finish);
*/
void lu_device_get(struct lu_device *d)
{
- cfs_atomic_inc(&d->ld_ref);
+ atomic_inc(&d->ld_ref);
}
EXPORT_SYMBOL(lu_device_get);
*/
void lu_device_put(struct lu_device *d)
{
- LASSERT(cfs_atomic_read(&d->ld_ref) > 0);
- cfs_atomic_dec(&d->ld_ref);
+ LASSERT(atomic_read(&d->ld_ref) > 0);
+ atomic_dec(&d->ld_ref);
}
EXPORT_SYMBOL(lu_device_put);
*/
int lu_device_init(struct lu_device *d, struct lu_device_type *t)
{
- if (t->ldt_device_nr++ == 0 && t->ldt_ops->ldto_start != NULL)
- t->ldt_ops->ldto_start(t);
- memset(d, 0, sizeof *d);
- cfs_atomic_set(&d->ld_ref, 0);
- d->ld_type = t;
- lu_ref_init(&d->ld_reference);
- CFS_INIT_LIST_HEAD(&d->ld_linkage);
- return 0;
+ if (atomic_inc_return(&t->ldt_device_nr) == 1 &&
+ t->ldt_ops->ldto_start != NULL)
+ t->ldt_ops->ldto_start(t);
+
+ memset(d, 0, sizeof *d);
+ d->ld_type = t;
+ lu_ref_init(&d->ld_reference);
+ INIT_LIST_HEAD(&d->ld_linkage);
+
+ return 0;
}
EXPORT_SYMBOL(lu_device_init);
*/
void lu_device_fini(struct lu_device *d)
{
- struct lu_device_type *t;
+ struct lu_device_type *t = d->ld_type;
- t = d->ld_type;
- if (d->ld_obd != NULL) {
- d->ld_obd->obd_lu_dev = NULL;
- d->ld_obd = NULL;
- }
+ if (d->ld_obd != NULL) {
+ d->ld_obd->obd_lu_dev = NULL;
+ d->ld_obd = NULL;
+ }
- lu_ref_fini(&d->ld_reference);
- LASSERTF(cfs_atomic_read(&d->ld_ref) == 0,
- "Refcount is %u\n", cfs_atomic_read(&d->ld_ref));
- LASSERT(t->ldt_device_nr > 0);
- if (--t->ldt_device_nr == 0 && t->ldt_ops->ldto_stop != NULL)
- t->ldt_ops->ldto_stop(t);
+ lu_ref_fini(&d->ld_reference);
+ LASSERTF(atomic_read(&d->ld_ref) == 0,
+ "Refcount is %u\n", atomic_read(&d->ld_ref));
+ LASSERT(atomic_read(&t->ldt_device_nr) > 0);
+
+ if (atomic_dec_and_test(&t->ldt_device_nr) &&
+ t->ldt_ops->ldto_stop != NULL)
+ t->ldt_ops->ldto_stop(t);
}
EXPORT_SYMBOL(lu_device_fini);
o->lo_dev = d;
lu_device_get(d);
lu_ref_add_at(&d->ld_reference, &o->lo_dev_ref, "lu_object", o);
- CFS_INIT_LIST_HEAD(&o->lo_linkage);
+ INIT_LIST_HEAD(&o->lo_linkage);
return 0;
}
{
struct lu_device *dev = o->lo_dev;
- LASSERT(cfs_list_empty(&o->lo_linkage));
+ LASSERT(list_empty(&o->lo_linkage));
if (dev != NULL) {
lu_ref_del_at(&dev->ld_reference, &o->lo_dev_ref,
*/
void lu_object_add_top(struct lu_object_header *h, struct lu_object *o)
{
- cfs_list_move(&o->lo_linkage, &h->loh_layers);
+ list_move(&o->lo_linkage, &h->loh_layers);
}
EXPORT_SYMBOL(lu_object_add_top);
*/
void lu_object_add(struct lu_object *before, struct lu_object *o)
{
- cfs_list_move(&o->lo_linkage, &before->lo_linkage);
+ list_move(&o->lo_linkage, &before->lo_linkage);
}
EXPORT_SYMBOL(lu_object_add);
int lu_object_header_init(struct lu_object_header *h)
{
memset(h, 0, sizeof *h);
- cfs_atomic_set(&h->loh_ref, 1);
- CFS_INIT_HLIST_NODE(&h->loh_hash);
- CFS_INIT_LIST_HEAD(&h->loh_lru);
- CFS_INIT_LIST_HEAD(&h->loh_layers);
+ atomic_set(&h->loh_ref, 1);
+ INIT_HLIST_NODE(&h->loh_hash);
+ INIT_LIST_HEAD(&h->loh_lru);
+ INIT_LIST_HEAD(&h->loh_layers);
lu_ref_init(&h->loh_reference);
return 0;
}
*/
void lu_object_header_fini(struct lu_object_header *h)
{
- LASSERT(cfs_list_empty(&h->loh_layers));
- LASSERT(cfs_list_empty(&h->loh_lru));
- LASSERT(cfs_hlist_unhashed(&h->loh_hash));
+ LASSERT(list_empty(&h->loh_layers));
+ LASSERT(list_empty(&h->loh_lru));
+ LASSERT(hlist_unhashed(&h->loh_hash));
lu_ref_fini(&h->loh_reference);
}
EXPORT_SYMBOL(lu_object_header_fini);
struct lu_object *lu_object_locate(struct lu_object_header *h,
const struct lu_device_type *dtype)
{
- struct lu_object *o;
+ struct lu_object *o;
- cfs_list_for_each_entry(o, &h->loh_layers, lo_linkage) {
- if (o->lo_dev->ld_type == dtype)
- return o;
- }
- return NULL;
+ list_for_each_entry(o, &h->loh_layers, lo_linkage) {
+ if (o->lo_dev->ld_type == dtype)
+ return o;
+ }
+ return NULL;
}
EXPORT_SYMBOL(lu_object_locate);
-
-
/**
* Finalize and free devices in the device stack.
*
}
}
}
-EXPORT_SYMBOL(lu_stack_fini);
enum {
/**
static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
-static DEFINE_SPINLOCK(lu_keys_guard);
+DEFINE_RWLOCK(lu_keys_guard);
+static atomic_t lu_key_initing_cnt = ATOMIC_INIT(0);
/**
* Global counter incremented whenever key is registered, unregistered,
*/
int lu_context_key_register(struct lu_context_key *key)
{
- int result;
- int i;
+ int result;
+ unsigned int i;
LASSERT(key->lct_init != NULL);
LASSERT(key->lct_fini != NULL);
LASSERT(key->lct_owner != NULL);
result = -ENFILE;
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
if (lu_keys[i] == NULL) {
key->lct_index = i;
- cfs_atomic_set(&key->lct_used, 1);
+ atomic_set(&key->lct_used, 1);
lu_keys[i] = key;
lu_ref_init(&key->lct_reference);
result = 0;
break;
}
}
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
return result;
}
EXPORT_SYMBOL(lu_context_key_register);
key = lu_keys[index];
LASSERT(key != NULL);
LASSERT(key->lct_fini != NULL);
- LASSERT(cfs_atomic_read(&key->lct_used) > 1);
+ LASSERT(atomic_read(&key->lct_used) > 1);
key->lct_fini(ctx, key, ctx->lc_value[index]);
lu_ref_del(&key->lct_reference, "ctx", ctx);
- cfs_atomic_dec(&key->lct_used);
+ atomic_dec(&key->lct_used);
LASSERT(key->lct_owner != NULL);
if ((ctx->lc_tags & LCT_NOREF) == 0) {
*/
void lu_context_key_degister(struct lu_context_key *key)
{
- LASSERT(cfs_atomic_read(&key->lct_used) >= 1);
+ LASSERT(atomic_read(&key->lct_used) >= 1);
LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
lu_context_key_quiesce(key);
++key_set_version;
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
key_fini(&lu_shrink_env.le_ctx, key->lct_index);
+
+ /**
+ * Wait until all transient contexts referencing this key have
+ * run lu_context_key::lct_fini() method.
+ */
+ while (atomic_read(&key->lct_used) > 1) {
+ write_unlock(&lu_keys_guard);
+ CDEBUG(D_INFO, "lu_context_key_degister: \"%s\" %p, %d\n",
+ key->lct_owner ? key->lct_owner->name : "", key,
+ atomic_read(&key->lct_used));
+ schedule();
+ write_lock(&lu_keys_guard);
+ }
if (lu_keys[key->lct_index]) {
lu_keys[key->lct_index] = NULL;
lu_ref_fini(&key->lct_reference);
}
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
- LASSERTF(cfs_atomic_read(&key->lct_used) == 1,
+ LASSERTF(atomic_read(&key->lct_used) == 1,
"key has instances: %d\n",
- cfs_atomic_read(&key->lct_used));
+ atomic_read(&key->lct_used));
}
EXPORT_SYMBOL(lu_context_key_degister);
/**
* List of remembered contexts. XXX document me.
*/
-static CFS_LIST_HEAD(lu_context_remembered);
+static struct list_head lu_context_remembered;
/**
* 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;
+ struct lu_context *ctx;
- if (!(key->lct_tags & LCT_QUIESCENT)) {
- /*
- * XXX layering violation.
- */
- key->lct_tags |= LCT_QUIESCENT;
+ if (!(key->lct_tags & LCT_QUIESCENT)) {
/*
* XXX memory barrier has to go here.
*/
- spin_lock(&lu_keys_guard);
- cfs_list_for_each_entry(ctx, &lu_context_remembered,
- lc_remember)
+ write_lock(&lu_keys_guard);
+ key->lct_tags |= LCT_QUIESCENT;
+
+ /**
+ * Wait until all lu_context_key::lct_init() methods
+ * have completed.
+ */
+ while (atomic_read(&lu_key_initing_cnt) > 0) {
+ write_unlock(&lu_keys_guard);
+ CDEBUG(D_INFO, "lu_context_key_quiesce: \"%s\""
+ " %p, %d (%d)\n",
+ key->lct_owner ? key->lct_owner->name : "",
+ key, atomic_read(&key->lct_used),
+ atomic_read(&lu_key_initing_cnt));
+ schedule();
+ write_lock(&lu_keys_guard);
+ }
+
+ list_for_each_entry(ctx, &lu_context_remembered,
+ lc_remember)
key_fini(ctx, key->lct_index);
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
++key_set_version;
}
}
-EXPORT_SYMBOL(lu_context_key_quiesce);
void lu_context_key_revive(struct lu_context_key *key)
{
key->lct_tags &= ~LCT_QUIESCENT;
++key_set_version;
}
-EXPORT_SYMBOL(lu_context_key_revive);
static void keys_fini(struct lu_context *ctx)
{
- int i;
+ unsigned int i;
if (ctx->lc_value == NULL)
return;
static int keys_fill(struct lu_context *ctx)
{
- int i;
+ unsigned int i;
+
+ /*
+ * A serialisation with lu_context_key_quiesce() is needed, but some
+ * "key->lct_init()" are calling kernel memory allocation routine and
+ * can't be called while holding a spin_lock.
+ * "lu_keys_guard" is held while incrementing "lu_key_initing_cnt"
+ * to ensure the start of the serialisation.
+ * An atomic_t variable is still used, in order not to reacquire the
+ * lock when decrementing the counter.
+ */
+ read_lock(&lu_keys_guard);
+ atomic_inc(&lu_key_initing_cnt);
+ read_unlock(&lu_keys_guard);
LINVRNT(ctx->lc_value != NULL);
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
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);
-
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))) {
+ atomic_dec(&lu_key_initing_cnt);
+ return PTR_ERR(value);
+ }
+
lu_ref_add_atomic(&key->lct_reference, "ctx", ctx);
- cfs_atomic_inc(&key->lct_used);
+ 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
}
ctx->lc_version = key_set_version;
}
+ atomic_dec(&lu_key_initing_cnt);
return 0;
}
ctx->lc_state = LCS_INITIALIZED;
ctx->lc_tags = tags;
if (tags & LCT_REMEMBER) {
- spin_lock(&lu_keys_guard);
- cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
- spin_unlock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
+ list_add(&ctx->lc_remember, &lu_context_remembered);
+ write_unlock(&lu_keys_guard);
} else {
- CFS_INIT_LIST_HEAD(&ctx->lc_remember);
+ INIT_LIST_HEAD(&ctx->lc_remember);
}
rc = keys_init(ctx);
ctx->lc_state = LCS_FINALIZED;
if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
- LASSERT(cfs_list_empty(&ctx->lc_remember));
+ LASSERT(list_empty(&ctx->lc_remember));
keys_fini(ctx);
} else { /* could race with key degister */
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
keys_fini(ctx);
- cfs_list_del_init(&ctx->lc_remember);
- spin_unlock(&lu_keys_guard);
+ list_del_init(&ctx->lc_remember);
+ write_unlock(&lu_keys_guard);
}
}
EXPORT_SYMBOL(lu_context_fini);
*/
void lu_context_exit(struct lu_context *ctx)
{
- int i;
+ 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) {
+ /* 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) {
- 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]);
- }
+ 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]);
+ }
}
+
+ if (ctx->lc_tags & LCT_REMEMBER)
+ read_unlock(&lu_keys_guard);
}
}
EXPORT_SYMBOL(lu_context_exit);
{
return likely(ctx->lc_version == key_set_version) ? 0 : keys_fill(ctx);
}
-EXPORT_SYMBOL(lu_context_refill);
/**
* lu_ctx_tags/lu_ses_tags will be updated if there are new types of
void lu_context_tags_update(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
lu_context_tags_default |= tags;
key_set_version++;
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
}
EXPORT_SYMBOL(lu_context_tags_update);
void lu_context_tags_clear(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
lu_context_tags_default &= ~tags;
key_set_version++;
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
}
EXPORT_SYMBOL(lu_context_tags_clear);
void lu_session_tags_update(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
lu_session_tags_default |= tags;
key_set_version++;
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
}
EXPORT_SYMBOL(lu_session_tags_update);
void lu_session_tags_clear(__u32 tags)
{
- spin_lock(&lu_keys_guard);
+ write_lock(&lu_keys_guard);
lu_session_tags_default &= ~tags;
key_set_version++;
- spin_unlock(&lu_keys_guard);
+ write_unlock(&lu_keys_guard);
}
EXPORT_SYMBOL(lu_session_tags_clear);
unsigned lss_busy;
} lu_site_stats_t;
-static void lu_site_stats_get(cfs_hash_t *hs,
+static void lu_site_stats_get(struct cfs_hash *hs,
lu_site_stats_t *stats, int populated)
{
- cfs_hash_bd_t bd;
- int i;
+ struct cfs_hash_bd bd;
+ unsigned int i;
cfs_hash_for_each_bucket(hs, &bd, i) {
struct lu_site_bkt_data *bkt = cfs_hash_bd_extra_get(hs, &bd);
- cfs_hlist_head_t *hhead;
+ struct hlist_head *hhead;
cfs_hash_bd_lock(hs, &bd, 1);
- stats->lss_busy += bkt->lsb_busy;
+ stats->lss_busy +=
+ cfs_hash_bd_count_get(&bd) - bkt->lsb_lru_len;
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));
}
cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
- if (!cfs_hlist_empty(hhead))
+ if (!hlist_empty(hhead))
stats->lss_populated++;
}
cfs_hash_bd_unlock(hs, &bd, 1);
}
}
-#ifdef __KERNEL__
/*
+ * 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)
+{
+ struct lu_site *s;
+ struct lu_site *tmp;
+ unsigned long cached = 0;
+
+ if (!(sc->gfp_mask & __GFP_FS))
+ return 0;
+
+ 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",
+ cached, sysctl_vfs_cache_pressure);
+
+ return cached;
+}
+
+static unsigned long lu_cache_shrink_scan(struct shrinker *sk,
+ struct shrink_control *sc)
+{
+ struct lu_site *s;
+ struct lu_site *tmp;
+ unsigned long remain = sc->nr_to_scan;
+ LIST_HEAD(splice);
+
+ if (!(sc->gfp_mask & __GFP_FS))
+ /* We must not take the lu_sites_guard lock when
+ * __GFP_FS is *not* set because of the deadlock
+ * possibility detailed above. Additionally,
+ * since we cannot determine the number of
+ * objects in the cache without taking this
+ * lock, we're in a particularly tough spot. As
+ * a result, we'll just lie and say our cache is
+ * empty. This _should_ be ok, as we can't
+ * reclaim objects when __GFP_FS is *not* set
+ * anyways.
+ */
+ return SHRINK_STOP;
+
+ 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);
+ /*
+ * Move just shrunk site to the tail of site list to
+ * assure shrinking fairness.
+ */
+ list_move_tail(&s->ls_linkage, &splice);
+ }
+ list_splice(&splice, lu_sites.prev);
+ up_write(&lu_sites_guard);
+
+ return sc->nr_to_scan - remain;
+}
+
+#ifndef HAVE_SHRINKER_COUNT
+/*
* There exists a potential lock inversion deadlock scenario when using
* Lustre on top of ZFS. This occurs between one of ZFS's
* buf_hash_table.ht_lock's, and Lustre's lu_sites_guard lock. Essentially,
* 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))
{
- lu_site_stats_t stats;
- struct lu_site *s;
- struct lu_site *tmp;
int cached = 0;
- int remain = shrink_param(sc, nr_to_scan);
- CFS_LIST_HEAD(splice);
-
- if (!(shrink_param(sc, gfp_mask) & __GFP_FS)) {
- if (remain != 0)
- return -1;
- else
- /* We must not take the lu_sites_guard lock when
- * __GFP_FS is *not* set because of the deadlock
- * possibility detailed above. Additionally,
- * since we cannot determine the number of
- * objects in the cache without taking this
- * lock, we're in a particularly tough spot. As
- * a result, we'll just lie and say our cache is
- * empty. This _should_ be ok, as we can't
- * reclaim objects when __GFP_FS is *not* set
- * anyways.
- */
- return 0;
- }
+ struct shrink_control scv = {
+ .nr_to_scan = shrink_param(sc, nr_to_scan),
+ .gfp_mask = shrink_param(sc, gfp_mask)
+ };
+#if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL)
+ struct shrinker* shrinker = NULL;
+#endif
- CDEBUG(D_INODE, "Shrink %d objects\n", remain);
- mutex_lock(&lu_sites_guard);
- cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
- if (shrink_param(sc, nr_to_scan) != 0) {
- remain = lu_site_purge(&lu_shrink_env, s, remain);
- /*
- * Move just shrunk site to the tail of site list to
- * assure shrinking fairness.
- */
- cfs_list_move_tail(&s->ls_linkage, &splice);
- }
+ CDEBUG(D_INODE, "Shrink %lu objects\n", scv.nr_to_scan);
- memset(&stats, 0, sizeof(stats));
- lu_site_stats_get(s->ls_obj_hash, &stats, 0);
- cached += stats.lss_total - stats.lss_busy;
- if (shrink_param(sc, nr_to_scan) && remain <= 0)
- break;
- }
- cfs_list_splice(&splice, lu_sites.prev);
- mutex_unlock(&lu_sites_guard);
+ if (scv.nr_to_scan != 0)
+ lu_cache_shrink_scan(shrinker, &scv);
- cached = (cached / 100) * sysctl_vfs_cache_pressure;
- if (shrink_param(sc, nr_to_scan) == 0)
- CDEBUG(D_INODE, "%d objects cached\n", cached);
- return cached;
+ cached = lu_cache_shrink_count(shrinker, &scv);
+ return cached;
}
+#endif /* HAVE_SHRINKER_COUNT */
+
+
/*
* Debugging stuff.
*/
/**
* 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;
void lu_context_keys_dump(void)
{
- int i;
+ unsigned int i;
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
struct lu_context_key *key;
CERROR("[%d]: %p %x (%p,%p,%p) %d %d \"%s\"@%p\n",
i, key, key->lct_tags,
key->lct_init, key->lct_fini, key->lct_exit,
- key->lct_index, cfs_atomic_read(&key->lct_used),
+ key->lct_index, atomic_read(&key->lct_used),
key->lct_owner ? key->lct_owner->name : "",
key->lct_owner);
lu_ref_print(&key->lct_reference);
}
}
}
-EXPORT_SYMBOL(lu_context_keys_dump);
-#else /* !__KERNEL__ */
-static int lu_cache_shrink(int nr, unsigned int gfp_mask)
-{
- return 0;
-}
-#endif /* __KERNEL__ */
/**
* Initialization of global lu_* data.
int lu_global_init(void)
{
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);
+ init_rwsem(&lu_sites_guard);
+
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;
* inode, one for ea. Unfortunately setting this high value results in
* lu_object/inode cache consuming all the memory.
*/
- lu_site_shrinker = set_shrinker(DEFAULT_SEEKS, lu_cache_shrink);
+ lu_site_shrinker = set_shrinker(DEFAULT_SEEKS, &shvar);
if (lu_site_shrinker == NULL)
return -ENOMEM;
* 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();
}
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
}
* Output site statistical counters into a buffer. Suitable for
* lprocfs_rd_*()-style functions.
*/
-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));
+int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m)
+{
+ lu_site_stats_t stats;
+
+ 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\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_print);
+EXPORT_SYMBOL(lu_site_stats_seq_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;
+ 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;
- 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);
}
EXPORT_SYMBOL(lu_buf_free);
-void lu_buf_alloc(struct lu_buf *buf, int size)
+void lu_buf_alloc(struct lu_buf *buf, size_t size)
{
LASSERT(buf);
LASSERT(buf->lb_buf == NULL);
}
EXPORT_SYMBOL(lu_buf_alloc);
-void lu_buf_realloc(struct lu_buf *buf, int size)
+void lu_buf_realloc(struct lu_buf *buf, size_t size)
{
lu_buf_free(buf);
lu_buf_alloc(buf, size);
}
EXPORT_SYMBOL(lu_buf_realloc);
-struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len)
+struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len)
{
if (buf->lb_buf == NULL && buf->lb_len == 0)
lu_buf_alloc(buf, len);
* old buffer remains unchanged on error
* \retval 0 or -ENOMEM
*/
-int lu_buf_check_and_grow(struct lu_buf *buf, int len)
+int lu_buf_check_and_grow(struct lu_buf *buf, size_t len)
{
char *ptr;
buf->lb_len = len;
return 0;
}
-EXPORT_SYMBOL(lu_buf_check_and_grow);
-