int do_item_link(item *it) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
assert(it->nbytes < (1024 * 1024)); /* 1MB max size */
it->it_flags |= ITEM_LINKED;
it->time = current_time;
assoc_insert(it);
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
#ifdef USE_REPLICATION
/* Allocate a new CAS ID on link. */
if(!(it->it_flags & ITEM_REPDATA))
it->cas_id = get_cas_id();
#else
/* Allocate a new CAS ID on link. */
it->cas_id = get_cas_id();
#endif /* USE_REPLICATION */
item_link_q(it);
return 1;
}
//将item插入到哈希表和LRU队列中，hv为哈希值
int do_item_link(item *it, const uint32_t hv) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
//确保这个item已经从slab分配出去并且还没插入到LRU队列中
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
mutex_lock(&cache_lock);
//加入link标记
it->it_flags |= ITEM_LINKED;
it->time = current_time;
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
//插入到hash表中
assoc_insert(it, hv);
//item插入到链表中
item_link_q(it);
//引用计数加1
refcount_incr(&it->refcount);
mutex_unlock(&cache_lock);
return 1;
}
int item_free(item *it) {
size_t ntotal = 0;
if (NULL == it)
return 0;
/* ntotal may be wrong, if 'it' is not a full item. */
ntotal = ITEM_ntotal(it);
if (ntotal > settings.item_buf_size){
if (settings.verbose > 1) {
fprintf(stderr, "ntotal: %d, use free() directly.\n", ntotal);
}
free(it);
}else{
if (0 != item_add_to_freelist(it)) {
if (settings.verbose > 1) {
fprintf(stderr, "ntotal: %d, add a item buffer to freelist fail, use free() directly.\n", ntotal);
}
free(it);
}else{
if (settings.verbose > 1) {
fprintf(stderr, "ntotal: %d, add a item buffer to freelist.\n", ntotal);
}
}
}
return 0;
}
/*@null@*/
void do_item_stats_sizes(ADD_STAT add_stats, void *c) {
/* max 1MB object, divided into 32 bytes size buckets */
const int num_buckets = 32768;
unsigned int *histogram = calloc(num_buckets, sizeof(int));
if (histogram != NULL) {
int i;
/* build the histogram */
for (i = 0; i < LARGEST_ID; i++) {
item *iter = heads[i];
while (iter) {
int ntotal = ITEM_ntotal(iter);
int bucket = ntotal / 32;
if ((ntotal % 32) != 0) bucket++;
if (bucket < num_buckets) histogram[bucket]++;
iter = iter->next;
}
}
/* write the buffer */
for (i = 0; i < num_buckets; i++) {
if (histogram[i] != 0) {
char key[8];
int klen = 0;
klen = snprintf(key, sizeof(key), "%d", i * 32);
assert(klen < sizeof(key));
APPEND_STAT(key, "%u", histogram[i]);
}
}
free(histogram);
}
add_stats(NULL, 0, NULL, 0, c);
}
/* 形成了一个完成的 item 后, 就要把它放入两个数据结构中, 一是 memcached 的哈希表,
memcached 运行过程中只有一个哈希表, 二是 item 所在的 slabclass 的 LRU 队列. */
int do_item_link(item *it, const uint32_t hv) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
mutex_lock(&cache_lock);
it->it_flags |= ITEM_LINKED;
it->time = current_time;
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
/* 把 item 放入哈希表 */
assoc_insert(it, hv);
/* 把 item 放入 LRU 队列*/
item_link_q(it);
refcount_incr(&it->refcount);
mutex_unlock(&cache_lock);
return 1;
}
int do_item_link(item *it) {
MEMCACHED_ITEM_LINK(ITEM_key(it), it->nkey, it->nbytes);
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
assert(it->nbytes < (1024 * 1024)); /* 1MB max size */
it->it_flags |= ITEM_LINKED;
it->time = current_time;
assoc_insert(it);
#ifdef MOXI_ITEM_MALLOC
it->refcount++;
#endif
STATS_LOCK();
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
STATS_UNLOCK();
/* Allocate a new CAS ID on link. */
ITEM_set_cas(it, (settings.use_cas) ? get_cas_id() : 0);
item_link_q(it);
return 1;
}
/* I think there's no way for this to be accurate without using the CAS value.
* Since items getting their time value bumped will pass this validation.
*/
void item_stats_sizes_remove(item *it) {
if (stats_sizes_hist == NULL || stats_sizes_cas_min > ITEM_get_cas(it))
return;
int ntotal = ITEM_ntotal(it);
int bucket = ntotal / 32;
if ((ntotal % 32) != 0) bucket++;
if (bucket < stats_sizes_buckets) stats_sizes_hist[bucket]--;
}
void item_unlink(item *it) {
if (it->it_flags & ITEM_LINKED) {
it->it_flags &= ~ITEM_LINKED;
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
assoc_delete(ITEM_key(it));
item_unlink_q(it);
}
if (it->refcount == 0) item_free(it);
}
void do_item_unlink(item *it) {
// MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nbytes);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items--;
assoc_delete(ITEM_key(it), it->nkey);
item_unlink_q(it);
if (it->refcount == 0) item_free(it);
}
}
void do_item_unlink(item *it) {
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey);
item_unlink_q(it);
if (it->refcount == 0) item_free(it);
}
}
void item_free(item *it) {
unsigned int ntotal = ITEM_ntotal(it);
assert((it->it_flags & ITEM_LINKED) == 0);
assert(it != heads[it->slabs_clsid]);
assert(it != tails[it->slabs_clsid]);
assert(it->refcount == 0);
/* so slab size changer can tell later if item is already free or not */
it->slabs_clsid = 0;
it->it_flags |= ITEM_SLABBED;
slabs_free(it, ntotal);
}
/*@null@*/
char *do_item_stats_sizes(uint32_t (*add_stats)(char *buf,
const char *key, const uint16_t klen, const char *val,
const uint32_t vlen, void *cookie), void *c, int *bytes) {
const int num_buckets = 32768; /* max 1MB object, divided into 32 bytes size buckets */
unsigned int *histogram = (unsigned int *)malloc((size_t)num_buckets * sizeof(int));
char *buf = (char *)malloc(2 * 1024 * 1024); /* 2MB max response size */
char *ptr = buf;
uint32_t nbytes, linelen = 0;
int i;
if (histogram == 0 || buf == 0) {
if (histogram) free(histogram);
if (buf) free(buf);
*bytes = -1;
return NULL;
}
/* build the histogram */
memset(histogram, 0, (size_t)num_buckets * sizeof(int));
for (i = 0; i < LARGEST_ID; i++) {
item *iter = heads[i];
while (iter) {
int ntotal = ITEM_ntotal(iter);
int bucket = ntotal / 32;
if ((ntotal % 32) != 0) bucket++;
if (bucket < num_buckets) histogram[bucket]++;
iter = iter->next;
}
}
/* write the buffer */
*bytes = 0;
char key[128];
char val[128];
for (i = 0; i < num_buckets; i++) {
if (histogram[i] != 0) {
sprintf(key, "%d", i * 32);
sprintf(val, "%u", histogram[i]);
nbytes = add_stats(ptr, key, strlen(key), val, strlen(val), c);
linelen += nbytes;
ptr += nbytes;
}
}
nbytes = add_stats(ptr, NULL, 0, NULL, 0, c);
*bytes = linelen + nbytes;
free(histogram);
return buf;
}
/* FIXME: Is it necessary to keep this copy/pasted code? */
void do_item_unlink_nolock(item *it, const uint32_t hv) {
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey, hv);
item_unlink_q(it);
do_item_remove(it);
}
}
/* slawek */
void do_item_unlink_nolock_nostat(item *it, const uint32_t hv, uint64_t *items_removed, uint64_t *bytes_removed) {
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
*items_removed += 1;
*bytes_removed += ITEM_ntotal(it);
assoc_delete(ITEM_key(it), it->nkey, hv);
item_unlink_q(it);
do_item_remove(it);
}
}
void item_free(item *it) {
size_t ntotal = ITEM_ntotal(it);
unsigned int clsid;
assert((it->it_flags & ITEM_LINKED) == 0);
assert(it != heads[it->slabs_clsid]);
assert(it != tails[it->slabs_clsid]);
assert(it->refcount == 0);
/* so slab size changer can tell later if item is already free or not */
clsid = ITEM_clsid(it);
DEBUG_REFCNT(it, 'F');
slabs_free(it, ntotal, clsid);
}
//释放item
void item_free(item *it) {
size_t ntotal = ITEM_ntotal(it);//获得item的大小
unsigned int clsid;
assert((it->it_flags & ITEM_LINKED) == 0);//判断item的状态是否正确
assert(it != heads[it->slabs_clsid]);//item不能为LRU的头指针
assert(it != tails[it->slabs_clsid]);//item不能为LRU的尾指针
assert(it->refcount == 0);//释放时，需保证引用次数为0
/* so slab size changer can tell later if item is already free or not */
clsid = it->slabs_clsid;
it->slabs_clsid = 0;//断开slabclass的链接
DEBUG_REFCNT(it, 'F');
slabs_free(it, ntotal, clsid);//slabclass结构执行释放
}
/* FIXME: Is it necessary to keep this copy/pasted code? */
void do_item_unlink_nolock(item *it, const uint32_t hv) {
syslog(LOG_INFO, "[%s:%s:%d]", __FILE__, __func__, __LINE__);
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey, hv);
item_unlink_q(it);
do_item_remove(it);
}
}
void item_free(item *it) {
size_t ntotal = ITEM_ntotal(it);
unsigned int clsid;
assert((it->it_flags & ITEM_LINKED) == 0);
assert(it->refcount == 0);
/* so slab size changer can tell later if item is already free or not */
clsid = it->slabs_clsid;
it->slabs_clsid = 0;
// it->it_flags |= ITEM_SLABBED;
DEBUG_REFCNT(it, 'F');
//[TODO] allocate management
zfree(it);
}
int item_link(item *it) {
assert((it->it_flags & (ITEM_LINKED|ITEM_SLABBED)) == 0);
assert(it->nbytes < 1048576);
it->it_flags |= ITEM_LINKED;
it->time = current_time;
assoc_insert(ITEM_key(it), it);
stats.curr_bytes += ITEM_ntotal(it);
stats.curr_items += 1;
stats.total_items += 1;
item_link_q(it);
return 1;
}
static void item_free(struct default_engine *engine, hash_item *it) {
size_t ntotal = ITEM_ntotal(engine, it);
unsigned int clsid;
assert((it->iflag & ITEM_LINKED) == 0);
assert(it != engine->items.heads[it->slabs_clsid]);
assert(it != engine->items.tails[it->slabs_clsid]);
assert(it->refcount == 0);
/* so slab size changer can tell later if item is already free or not */
clsid = it->slabs_clsid;
it->slabs_clsid = 0;
it->iflag |= ITEM_SLABBED;
DEBUG_REFCNT(it, 'F');
slabs_free(engine, it, ntotal, clsid);
}
//释放item
void item_free(item *it) {
syslog(LOG_INFO, "[%s:%s:%d]", __FILE__, __func__, __LINE__);
size_t ntotal = ITEM_ntotal(it);
unsigned int clsid;
assert((it->it_flags & ITEM_LINKED) == 0);//没有在hash表和LUR链中
assert(it != heads[it->slabs_clsid]);
assert(it != tails[it->slabs_clsid]);
assert(it->refcount == 0);
/* so slab size changer can tell later if item is already free or not */
clsid = it->slabs_clsid;
it->slabs_clsid = 0;
DEBUG_REFCNT(it, 'F');
slabs_free(it, ntotal, clsid);
}
/* 0 for Success
-1 for SERVER_ERROR
*/
int bdb_set(char *key, item *it){
pthread_rwlock_rdlock(&qlist_ht_lock);
queue_t *q = (queue_t *)hashtable_search(qlist_htp, (void *)key);
DB_TXN *txnp = NULL;
int ret;
if (NULL == q) {
pthread_rwlock_unlock(&qlist_ht_lock);
ret = bdb_create_queue(key);
if (0 != ret){
return -1;
}
/* search again */
pthread_rwlock_rdlock(&qlist_ht_lock);
q = (queue_t *)hashtable_search(qlist_htp, (void *)key);
}
if (NULL != q) {
db_recno_t recno;
DBT dbkey, dbdata;
BDB_CLEANUP_DBT();
dbkey.data = &recno;
dbkey.ulen = sizeof(recno);
dbkey.flags = DB_DBT_USERMEM;
dbdata.data = it;
dbdata.size = ITEM_ntotal(it);
ret = envp->txn_begin(envp, NULL, &txnp, 0);
CHECK_DB_RET(ret);
ret = q->dbp->put(q->dbp, txnp, &dbkey, &dbdata, DB_APPEND);
CHECK_DB_RET(ret);
ret = txnp->commit(txnp, 0);
CHECK_DB_RET(ret);
pthread_mutex_lock(&(q->lock));
(q->set_hits)++;
pthread_mutex_unlock(&(q->lock));
}
pthread_rwlock_unlock(&qlist_ht_lock);
return 0;
dberr:
if (txnp != NULL){
txnp->abort(txnp);
}
if (settings.verbose > 1) {
fprintf(stderr, "bdb_set: %s\n", db_strerror(ret));
}
pthread_rwlock_unlock(&qlist_ht_lock);
return -1;
}
void do_item_unlink(item *it, const uint32_t hv) {
syslog(LOG_INFO, "[%s:%s:%d]", __FILE__, __func__, __LINE__);
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
mutex_lock(&cache_lock);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;//设置为非linked
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey, hv); //从hash表中删除
item_unlink_q(it); //从LRU链中删除
do_item_remove(it);
}
mutex_unlock(&cache_lock);
}
//将item从hashtable和LRU链中移除,而且还释放掉 item 所占的内存 (其实只是把 item 放到空闲链表中),是do_item_link的逆操作
void do_item_unlink(item *it, const uint32_t hv)
{
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
mutex_lock(&cache_lock);//执行同步
if ((it->it_flags & ITEM_LINKED) != 0) {//判断状态值，保证item还在LRU队列中
it->it_flags &= ~ITEM_LINKED;//修改状态值
STATS_LOCK();//更新统计信息
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey, hv);//从Hash表中删除
item_unlink_q(it);//将item从slabclass对应的LRU队列摘除
do_item_remove(it);//释放 item 所占的内存
}
mutex_unlock(&cache_lock);
}
void do_item_unlink(item *it) {
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
assoc_delete(ITEM_key(it), it->nkey);
item_unlink_q(it);
#ifndef MOXI_ITEM_MALLOC
if (it->refcount == 0) item_free(it);
#endif
}
}
void do_item_unlink(struct default_engine *engine, hash_item *it) {
MEMCACHED_ITEM_UNLINK(item_get_key(it), it->nkey, it->nbytes);
if ((it->iflag & ITEM_LINKED) != 0) {
it->iflag &= ~ITEM_LINKED;
pthread_mutex_lock(&engine->stats.lock);
engine->stats.curr_bytes -= ITEM_ntotal(engine, it);
engine->stats.curr_items -= 1;
pthread_mutex_unlock(&engine->stats.lock);
assoc_delete(engine, engine->server.core->hash(item_get_key(it),
it->nkey, 0),
item_get_key(it), it->nkey);
item_unlink_q(engine, it);
if (it->refcount == 0) {
item_free(engine, it);
}
}
}
static void do_item_link_q(item *it) { /* item is the new head */
item **head, **tail;
assert((it->it_flags & ITEM_SLABBED) == 0);
head = &heads[it->slabs_clsid];
tail = &tails[it->slabs_clsid];
assert(it != *head);
assert((*head && *tail) || (*head == 0 && *tail == 0));
it->prev = 0;
it->next = *head;
if (it->next) it->next->prev = it;
*head = it;
if (*tail == 0) *tail = it;
sizes[it->slabs_clsid]++;
sizes_bytes[it->slabs_clsid] += ITEM_ntotal(it);
return;
}
//从哈希表和LRU中删除
void do_item_unlink(item *it, const uint32_t hv) {
MEMCACHED_ITEM_UNLINK(ITEM_key(it), it->nkey, it->nbytes);
mutex_lock(&cache_lock);
if ((it->it_flags & ITEM_LINKED) != 0) {
it->it_flags &= ~ITEM_LINKED;
STATS_LOCK();
stats.curr_bytes -= ITEM_ntotal(it);
stats.curr_items -= 1;
STATS_UNLOCK();
//从哈希表中删除
assoc_delete(ITEM_key(it), it->nkey, hv);
//从链表中删除
item_unlink_q(it);
//向slab归还这个item
do_item_remove(it);
}
mutex_unlock(&cache_lock);
}
/* 0 for Success
-1 for SERVER_ERROR
*/
int item_put(char *key, size_t nkey, item *it){
int ret;
DBT dbkey, dbdata;
BDB_CLEANUP_DBT();
dbkey.data = key;
dbkey.size = nkey;
dbdata.data = it;
dbdata.size = ITEM_ntotal(it);
ret = dbp->put(dbp, NULL, &dbkey, &dbdata, 0);
if (ret == 0) {
return 0;
} else {
if (settings.verbose > 1) {
fprintf(stderr, "dbp->put: %s\n", db_strerror(ret));
}
return -1;
}
}
/*@null@*/
static void do_item_stats_sizes(struct default_engine *engine,
ADD_STAT add_stats, const void *c) {
/* max 1MB object, divided into 32 bytes size buckets */
const int num_buckets = 32768;
unsigned int *histogram = calloc(num_buckets, sizeof(unsigned int));
if (histogram != NULL) {
int i;
/* build the histogram */
for (i = 0; i < POWER_LARGEST; i++) {
hash_item *iter = engine->items.heads[i];
while (iter) {
size_t ntotal = ITEM_ntotal(engine, iter);
size_t bucket = ntotal / 32;
if ((ntotal % 32) != 0) {
bucket++;
}
if (bucket < num_buckets) {
histogram[bucket]++;
}
iter = iter->next;
}
}
/* write the buffer */
for (i = 0; i < num_buckets; i++) {
if (histogram[i] != 0) {
char key[8], val[32];
int klen, vlen;
klen = snprintf(key, sizeof(key), "%d", i * 32);
vlen = snprintf(val, sizeof(val), "%u", histogram[i]);
if (klen > 0 && klen < sizeof(key) && vlen > 0 &&
vlen < sizeof(val)) {
add_stats(key, klen, val, vlen, c);
}
}
}
free(histogram);
}
}