-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2011 Whamcloud, Inc.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*/
#define DEBUG_SUBSYSTEM S_CLASS
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#include <libcfs/libcfs.h>
EXPORT_SYMBOL(lu_object_put);
/**
+ * Put object and don't keep in cache. This is temporary solution for
+ * multi-site objects when its layering is not constant.
+ */
+void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o)
+{
+ cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
+ &o->lo_header->loh_flags);
+ return lu_object_put(env, o);
+}
+EXPORT_SYMBOL(lu_object_put_nocache);
+
+/**
* Allocate new object.
*
* This follows object creation protocol, described in the comment within
* lu_global_init().
*/
struct lu_context_key lu_global_key = {
- .lct_tags = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD,
+ .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD |
+ LCT_MG_THREAD | LCT_CL_THREAD,
.lct_init = lu_global_key_init,
.lct_fini = lu_global_key_fini
};
int lu_cdebug_printer(const struct lu_env *env,
void *cookie, const char *format, ...)
{
- struct lu_cdebug_print_info *info = cookie;
- struct lu_cdebug_data *key;
+ struct libcfs_debug_msg_data *msgdata = cookie;
+ struct lu_cdebug_data *key;
int used;
int complete;
va_list args;
vsnprintf(key->lck_area + used,
ARRAY_SIZE(key->lck_area) - used, format, args);
if (complete) {
- if (cfs_cdebug_show(info->lpi_mask, info->lpi_subsys))
- libcfs_debug_msg(NULL, info->lpi_subsys, info->lpi_mask,
- (char *)info->lpi_file, info->lpi_fn,
- info->lpi_line, "%s", key->lck_area);
+ if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
+ libcfs_debug_msg(msgdata, "%s", key->lck_area);
key->lck_area[0] = 0;
}
va_end(args);
h = container_of0(hnode, struct lu_object_header, loh_hash);
if (likely(!lu_object_is_dying(h))) {
lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
+ cfs_list_del_init(&h->loh_lru);
return lu_object_top(h);
}
hs = s->ls_obj_hash;
cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
o = htable_lookup(s, &bd, f, waiter, &version);
- if (o != NULL && !cfs_list_empty(&o->lo_header->loh_lru))
- cfs_list_del_init(&o->lo_header->loh_lru);
cfs_hash_bd_unlock(hs, &bd, 1);
if (o != NULL)
return o;
* Global list of all sites on this node
*/
static CFS_LIST_HEAD(lu_sites);
-static CFS_DECLARE_MUTEX(lu_sites_guard);
+static CFS_DEFINE_MUTEX(lu_sites_guard);
/**
* Global environment used by site shrinker.
const void *key, unsigned mask)
{
struct lu_fid *fid = (struct lu_fid *)key;
- unsigned hash;
+ __u32 hash;
+
+ hash = fid_flatten32(fid);
+ hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
+ hash = cfs_hash_long(hash, hs->hs_bkt_bits);
+
+ /* give me another random factor */
+ hash -= cfs_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 = (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
- hash += fid_hash(fid, hs->hs_bkt_bits) << hs->hs_bkt_bits;
return hash & mask;
}
.hs_put_locked = lu_obj_hop_put_locked,
};
+void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d)
+{
+ cfs_spin_lock(&s->ls_ld_lock);
+ if (cfs_list_empty(&d->ld_linkage))
+ cfs_list_add(&d->ld_linkage, &s->ls_ld_linkage);
+ cfs_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)
+{
+ cfs_spin_lock(&s->ls_ld_lock);
+ cfs_list_del_init(&d->ld_linkage);
+ cfs_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 23
+#define LU_SITE_BITS_MAX 24
/**
- * total 128 buckets, we don't want too many buckets because:
+ * total 256 buckets, we don't want too many buckets because:
* - consume too much memory
* - avoid unbalanced LRU list
*/
-#define LU_SITE_BKT_BITS 7
+#define LU_SITE_BKT_BITS 8
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("lu_site", bits, bits,
+ s->ls_obj_hash = cfs_hash_create(name, bits, bits,
bits - LU_SITE_BKT_BITS,
sizeof(*bkt), 0, 0,
&lu_site_hash_ops,
lu_device_get(top);
lu_ref_add(&top->ld_reference, "site-top", s);
+ CFS_INIT_LIST_HEAD(&s->ls_ld_linkage);
+ cfs_spin_lock_init(&s->ls_ld_lock);
+
+ lu_dev_add_linkage(s, top);
+
RETURN(0);
}
EXPORT_SYMBOL(lu_site_init);
*/
void lu_site_fini(struct lu_site *s)
{
- cfs_down(&lu_sites_guard);
+ cfs_mutex_lock(&lu_sites_guard);
cfs_list_del_init(&s->ls_linkage);
- cfs_up(&lu_sites_guard);
+ cfs_mutex_unlock(&lu_sites_guard);
if (s->ls_obj_hash != NULL) {
cfs_hash_putref(s->ls_obj_hash);
int lu_site_init_finish(struct lu_site *s)
{
int result;
- cfs_down(&lu_sites_guard);
+ cfs_mutex_lock(&lu_sites_guard);
result = lu_context_refill(&lu_shrink_env.le_ctx);
if (result == 0)
cfs_list_add(&s->ls_linkage, &lu_sites);
- cfs_up(&lu_sites_guard);
+ cfs_mutex_unlock(&lu_sites_guard);
return result;
}
EXPORT_SYMBOL(lu_site_init_finish);
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;
}
EXPORT_SYMBOL(lu_device_init);
/* purge again. */
lu_site_purge(env, site, ~0);
- if (!cfs_hash_is_empty(site->ls_obj_hash)) {
- /*
- * Uh-oh, objects still exist.
- */
- static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR);
-
- lu_site_print(env, site, &cookie, lu_cdebug_printer);
- }
-
for (scan = top; scan != NULL; scan = next) {
const struct lu_device_type *ldt = scan->ld_type;
struct obd_type *type;
static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
-static cfs_spinlock_t lu_keys_guard = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(lu_keys_guard);
/**
* Global counter incremented whenever key is registered, unregistered,
key->lct_fini(ctx, key, ctx->lc_value[index]);
lu_ref_del(&key->lct_reference, "ctx", ctx);
cfs_atomic_dec(&key->lct_used);
- LASSERT(key->lct_owner != NULL);
- if (!(ctx->lc_tags & LCT_NOREF)) {
- LASSERT(cfs_module_refcount(key->lct_owner) > 0);
- cfs_module_put(key->lct_owner);
- }
- ctx->lc_value[index] = NULL;
- }
+
+ LASSERT(key->lct_owner != NULL);
+ if ((ctx->lc_tags & LCT_NOREF) == 0) {
+ LINVRNT(cfs_module_refcount(key->lct_owner) > 0);
+ cfs_module_put(key->lct_owner);
+ }
+ ctx->lc_value[index] = NULL;
+ }
}
/**
static void keys_fini(struct lu_context *ctx)
{
- int i;
+ int i;
- cfs_spin_lock(&lu_keys_guard);
- if (ctx->lc_value != NULL) {
- for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
- key_fini(ctx, i);
- OBD_FREE(ctx->lc_value,
- ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
- ctx->lc_value = NULL;
- }
- cfs_spin_unlock(&lu_keys_guard);
+ if (ctx->lc_value == NULL)
+ return;
+
+ for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
+ key_fini(ctx, i);
+
+ OBD_FREE(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
+ ctx->lc_value = NULL;
}
static int keys_fill(struct lu_context *ctx)
{
int i;
+ LINVRNT(ctx->lc_value != NULL);
for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
struct lu_context_key *key;
static int keys_init(struct lu_context *ctx)
{
- int result;
-
- OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
- if (likely(ctx->lc_value != NULL))
- result = keys_fill(ctx);
- else
- result = -ENOMEM;
+ OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
+ if (likely(ctx->lc_value != NULL))
+ return keys_fill(ctx);
- if (result != 0)
- keys_fini(ctx);
- return result;
+ return -ENOMEM;
}
/**
*/
int lu_context_init(struct lu_context *ctx, __u32 tags)
{
+ int rc;
+
memset(ctx, 0, sizeof *ctx);
ctx->lc_state = LCS_INITIALIZED;
ctx->lc_tags = tags;
cfs_spin_lock(&lu_keys_guard);
cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
cfs_spin_unlock(&lu_keys_guard);
- } else
- CFS_INIT_LIST_HEAD(&ctx->lc_remember);
- return keys_init(ctx);
+ } else {
+ CFS_INIT_LIST_HEAD(&ctx->lc_remember);
+ }
+
+ rc = keys_init(ctx);
+ if (rc != 0)
+ lu_context_fini(ctx);
+
+ return rc;
}
EXPORT_SYMBOL(lu_context_init);
*/
void lu_context_fini(struct lu_context *ctx)
{
- LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
- ctx->lc_state = LCS_FINALIZED;
- keys_fini(ctx);
- cfs_spin_lock(&lu_keys_guard);
- cfs_list_del_init(&ctx->lc_remember);
- cfs_spin_unlock(&lu_keys_guard);
+ LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
+ ctx->lc_state = LCS_FINALIZED;
+
+ if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
+ LASSERT(cfs_list_empty(&ctx->lc_remember));
+ keys_fini(ctx);
+
+ } else { /* could race with key degister */
+ cfs_spin_lock(&lu_keys_guard);
+ keys_fini(ctx);
+ cfs_list_del_init(&ctx->lc_remember);
+ cfs_spin_unlock(&lu_keys_guard);
+ }
}
EXPORT_SYMBOL(lu_context_fini);
/**
* Allocate for context all missing keys that were registered after context
- * creation.
+ * creation. key_set_version is only changed in rare cases when modules
+ * are loaded and removed.
*/
int lu_context_refill(struct lu_context *ctx)
{
- LINVRNT(ctx->lc_value != NULL);
- return ctx->lc_version == key_set_version ? 0 : keys_fill(ctx);
+ 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
+ * obd being added. Currently, this is only used on client side, specifically
+ * for echo device client, for other stack (like ptlrpc threads), context are
+ * 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;
+
+void lu_context_tags_update(__u32 tags)
+{
+ cfs_spin_lock(&lu_keys_guard);
+ lu_context_tags_default |= tags;
+ key_set_version ++;
+ cfs_spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_context_tags_update);
+
+void lu_context_tags_clear(__u32 tags)
+{
+ cfs_spin_lock(&lu_keys_guard);
+ lu_context_tags_default &= ~tags;
+ key_set_version ++;
+ cfs_spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_context_tags_clear);
+
+void lu_session_tags_update(__u32 tags)
+{
+ cfs_spin_lock(&lu_keys_guard);
+ lu_session_tags_default |= tags;
+ key_set_version ++;
+ cfs_spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_session_tags_update);
+
+void lu_session_tags_clear(__u32 tags)
+{
+ cfs_spin_lock(&lu_keys_guard);
+ lu_session_tags_default &= ~tags;
+ key_set_version ++;
+ cfs_spin_unlock(&lu_keys_guard);
+}
+EXPORT_SYMBOL(lu_session_tags_clear);
+
int lu_env_init(struct lu_env *env, __u32 tags)
{
int result;
}
EXPORT_SYMBOL(lu_env_refill);
+/**
+ * Currently, this API will only be used by echo client.
+ * Because echo client and normal lustre client will share
+ * same cl_env cache. So echo client needs to refresh
+ * the env context after it get one from the cache, especially
+ * when normal client and echo client co-exist in the same client.
+ */
+int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags,
+ __u32 stags)
+{
+ int result;
+
+ if ((env->le_ctx.lc_tags & ctags) != ctags) {
+ env->le_ctx.lc_version = 0;
+ env->le_ctx.lc_tags |= ctags;
+ }
+
+ if (env->le_ses && (env->le_ses->lc_tags & stags) != stags) {
+ env->le_ses->lc_version = 0;
+ env->le_ses->lc_tags |= stags;
+ }
+
+ result = lu_env_refill(env);
+
+ return result;
+}
+EXPORT_SYMBOL(lu_env_refill_by_tags);
+
static struct cfs_shrinker *lu_site_shrinker = NULL;
typedef struct lu_site_stats{
CDEBUG(D_INODE, "Shrink %d objects\n", remain);
}
- cfs_down(&lu_sites_guard);
+ cfs_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);
break;
}
cfs_list_splice(&splice, lu_sites.prev);
- cfs_up(&lu_sites_guard);
+ cfs_mutex_unlock(&lu_sites_guard);
cached = (cached / 100) * sysctl_vfs_cache_pressure;
if (shrink_param(sc, nr_to_scan) == 0)
* conservatively. This should not be too bad, because this
* environment is global.
*/
- cfs_down(&lu_sites_guard);
+ cfs_mutex_lock(&lu_sites_guard);
result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
- cfs_up(&lu_sites_guard);
+ cfs_mutex_unlock(&lu_sites_guard);
if (result != 0)
return result;
* Tear shrinker environment down _after_ de-registering
* lu_global_key, because the latter has a value in the former.
*/
- cfs_down(&lu_sites_guard);
+ cfs_mutex_lock(&lu_sites_guard);
lu_env_fini(&lu_shrink_env);
- cfs_up(&lu_sites_guard);
+ cfs_mutex_unlock(&lu_sites_guard);
lu_ref_global_fini();
}
}
}
EXPORT_SYMBOL(lu_kmem_fini);
+
+/**
+ * Temporary solution to be able to assign fid in ->do_create()
+ * till we have fully-functional OST fids
+ */
+void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
+ const struct lu_fid *fid)
+{
+ 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;
+ cfs_waitlink_t waiter;
+ cfs_hash_t *hs;
+ cfs_hash_bd_t bd;
+ __u64 version = 0;
+
+ LASSERT(fid_is_zero(old));
+
+ 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(shadow == NULL);
+ *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);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff