freebsd-skq/usr.sbin/nscd/cachelib.c

1219 lines
33 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2005 Michael Bushkov <bushman@rsu.ru>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/time.h>
#include <assert.h>
#include <stdlib.h>
#include <string.h>
#include "cachelib.h"
#include "debug.h"
#define INITIAL_ENTRIES_CAPACITY 32
#define ENTRIES_CAPACITY_STEP 32
#define STRING_SIMPLE_HASH_BODY(in_var, var, a, M) \
for ((var) = 0; *(in_var) != '\0'; ++(in_var)) \
(var) = ((a)*(var) + *(in_var)) % (M)
#define STRING_SIMPLE_MP2_HASH_BODY(in_var, var, a, M) \
for ((var) = 0; *(in_var) != 0; ++(in_var)) \
(var) = ((a)*(var) + *(in_var)) & (M - 1)
static int cache_elemsize_common_continue_func(struct cache_common_entry_ *,
struct cache_policy_item_ *);
static int cache_lifetime_common_continue_func(struct cache_common_entry_ *,
struct cache_policy_item_ *);
static void clear_cache_entry(struct cache_entry_ *);
static void destroy_cache_entry(struct cache_entry_ *);
static void destroy_cache_mp_read_session(struct cache_mp_read_session_ *);
static void destroy_cache_mp_write_session(struct cache_mp_write_session_ *);
static int entries_bsearch_cmp_func(const void *, const void *);
static int entries_qsort_cmp_func(const void *, const void *);
static struct cache_entry_ ** find_cache_entry_p(struct cache_ *,
const char *);
static void flush_cache_entry(struct cache_entry_ *);
static void flush_cache_policy(struct cache_common_entry_ *,
struct cache_policy_ *, struct cache_policy_ *,
int (*)(struct cache_common_entry_ *,
struct cache_policy_item_ *));
static int ht_items_cmp_func(const void *, const void *);
static int ht_items_fixed_size_left_cmp_func(const void *, const void *);
static hashtable_index_t ht_item_hash_func(const void *, size_t);
/*
* Hashing and comparing routines, that are used with the hash tables
*/
static int
ht_items_cmp_func(const void *p1, const void *p2)
{
struct cache_ht_item_data_ *hp1, *hp2;
size_t min_size;
int result;
hp1 = (struct cache_ht_item_data_ *)p1;
hp2 = (struct cache_ht_item_data_ *)p2;
assert(hp1->key != NULL);
assert(hp2->key != NULL);
if (hp1->key_size != hp2->key_size) {
min_size = (hp1->key_size < hp2->key_size) ? hp1->key_size :
hp2->key_size;
result = memcmp(hp1->key, hp2->key, min_size);
if (result == 0)
return ((hp1->key_size < hp2->key_size) ? -1 : 1);
else
return (result);
} else
return (memcmp(hp1->key, hp2->key, hp1->key_size));
}
static int
ht_items_fixed_size_left_cmp_func(const void *p1, const void *p2)
{
struct cache_ht_item_data_ *hp1, *hp2;
size_t min_size;
int result;
hp1 = (struct cache_ht_item_data_ *)p1;
hp2 = (struct cache_ht_item_data_ *)p2;
assert(hp1->key != NULL);
assert(hp2->key != NULL);
if (hp1->key_size != hp2->key_size) {
min_size = (hp1->key_size < hp2->key_size) ? hp1->key_size :
hp2->key_size;
result = memcmp(hp1->key, hp2->key, min_size);
if (result == 0)
if (min_size == hp1->key_size)
return (0);
else
return ((hp1->key_size < hp2->key_size) ? -1 : 1);
else
return (result);
} else
return (memcmp(hp1->key, hp2->key, hp1->key_size));
}
static hashtable_index_t
ht_item_hash_func(const void *p, size_t cache_entries_size)
{
struct cache_ht_item_data_ *hp;
size_t i;
hashtable_index_t retval;
hp = (struct cache_ht_item_data_ *)p;
assert(hp->key != NULL);
retval = 0;
for (i = 0; i < hp->key_size; ++i)
retval = (127 * retval + (unsigned char)hp->key[i]) %
cache_entries_size;
return retval;
}
HASHTABLE_GENERATE(cache_ht_, cache_ht_item_, struct cache_ht_item_data_, data,
ht_item_hash_func, ht_items_cmp_func);
/*
* Routines to sort and search the entries by name
*/
static int
entries_bsearch_cmp_func(const void *key, const void *ent)
{
assert(key != NULL);
assert(ent != NULL);
return (strcmp((char const *)key,
(*(struct cache_entry_ const **)ent)->name));
}
static int
entries_qsort_cmp_func(const void *e1, const void *e2)
{
assert(e1 != NULL);
assert(e2 != NULL);
return (strcmp((*(struct cache_entry_ const **)e1)->name,
(*(struct cache_entry_ const **)e2)->name));
}
static struct cache_entry_ **
find_cache_entry_p(struct cache_ *the_cache, const char *entry_name)
{
return ((struct cache_entry_ **)(bsearch(entry_name, the_cache->entries,
the_cache->entries_size, sizeof(struct cache_entry_ *),
entries_bsearch_cmp_func)));
}
static void
destroy_cache_mp_write_session(struct cache_mp_write_session_ *ws)
{
struct cache_mp_data_item_ *data_item;
TRACE_IN(destroy_cache_mp_write_session);
assert(ws != NULL);
while (!TAILQ_EMPTY(&ws->items)) {
data_item = TAILQ_FIRST(&ws->items);
TAILQ_REMOVE(&ws->items, data_item, entries);
free(data_item->value);
free(data_item);
}
free(ws);
TRACE_OUT(destroy_cache_mp_write_session);
}
static void
destroy_cache_mp_read_session(struct cache_mp_read_session_ *rs)
{
TRACE_IN(destroy_cache_mp_read_session);
assert(rs != NULL);
free(rs);
TRACE_OUT(destroy_cache_mp_read_session);
}
static void
destroy_cache_entry(struct cache_entry_ *entry)
{
struct cache_common_entry_ *common_entry;
struct cache_mp_entry_ *mp_entry;
struct cache_mp_read_session_ *rs;
struct cache_mp_write_session_ *ws;
struct cache_ht_item_ *ht_item;
struct cache_ht_item_data_ *ht_item_data;
TRACE_IN(destroy_cache_entry);
assert(entry != NULL);
if (entry->params->entry_type == CET_COMMON) {
common_entry = (struct cache_common_entry_ *)entry;
HASHTABLE_FOREACH(&(common_entry->items), ht_item) {
HASHTABLE_ENTRY_FOREACH(ht_item, data, ht_item_data)
{
free(ht_item_data->key);
free(ht_item_data->value);
}
HASHTABLE_ENTRY_CLEAR(ht_item, data);
}
HASHTABLE_DESTROY(&(common_entry->items), data);
/* FIFO policy is always first */
destroy_cache_fifo_policy(common_entry->policies[0]);
switch (common_entry->common_params.policy) {
case CPT_LRU:
destroy_cache_lru_policy(common_entry->policies[1]);
break;
case CPT_LFU:
destroy_cache_lfu_policy(common_entry->policies[1]);
break;
default:
break;
}
free(common_entry->policies);
} else {
mp_entry = (struct cache_mp_entry_ *)entry;
while (!TAILQ_EMPTY(&mp_entry->ws_head)) {
ws = TAILQ_FIRST(&mp_entry->ws_head);
TAILQ_REMOVE(&mp_entry->ws_head, ws, entries);
destroy_cache_mp_write_session(ws);
}
while (!TAILQ_EMPTY(&mp_entry->rs_head)) {
rs = TAILQ_FIRST(&mp_entry->rs_head);
TAILQ_REMOVE(&mp_entry->rs_head, rs, entries);
destroy_cache_mp_read_session(rs);
}
if (mp_entry->completed_write_session != NULL)
destroy_cache_mp_write_session(
mp_entry->completed_write_session);
if (mp_entry->pending_write_session != NULL)
destroy_cache_mp_write_session(
mp_entry->pending_write_session);
}
free(entry->name);
free(entry);
TRACE_OUT(destroy_cache_entry);
}
static void
clear_cache_entry(struct cache_entry_ *entry)
{
struct cache_mp_entry_ *mp_entry;
struct cache_common_entry_ *common_entry;
struct cache_ht_item_ *ht_item;
struct cache_ht_item_data_ *ht_item_data;
struct cache_policy_ *policy;
struct cache_policy_item_ *item, *next_item;
size_t entry_size;
int i;
if (entry->params->entry_type == CET_COMMON) {
common_entry = (struct cache_common_entry_ *)entry;
entry_size = 0;
HASHTABLE_FOREACH(&(common_entry->items), ht_item) {
HASHTABLE_ENTRY_FOREACH(ht_item, data, ht_item_data)
{
free(ht_item_data->key);
free(ht_item_data->value);
}
entry_size += HASHTABLE_ENTRY_SIZE(ht_item, data);
HASHTABLE_ENTRY_CLEAR(ht_item, data);
}
common_entry->items_size -= entry_size;
for (i = 0; i < common_entry->policies_size; ++i) {
policy = common_entry->policies[i];
next_item = NULL;
item = policy->get_first_item_func(policy);
while (item != NULL) {
next_item = policy->get_next_item_func(policy,
item);
policy->remove_item_func(policy, item);
policy->destroy_item_func(item);
item = next_item;
}
}
} else {
mp_entry = (struct cache_mp_entry_ *)entry;
if (mp_entry->rs_size == 0) {
if (mp_entry->completed_write_session != NULL) {
destroy_cache_mp_write_session(
mp_entry->completed_write_session);
mp_entry->completed_write_session = NULL;
}
memset(&mp_entry->creation_time, 0,
sizeof(struct timeval));
memset(&mp_entry->last_request_time, 0,
sizeof(struct timeval));
}
}
}
/*
* When passed to the flush_cache_policy, ensures that all old elements are
* deleted.
*/
static int
cache_lifetime_common_continue_func(struct cache_common_entry_ *entry,
struct cache_policy_item_ *item)
{
return ((item->last_request_time.tv_sec - item->creation_time.tv_sec >
entry->common_params.max_lifetime.tv_sec) ? 1: 0);
}
/*
* When passed to the flush_cache_policy, ensures that all elements, that
* exceed the size limit, are deleted.
*/
static int
cache_elemsize_common_continue_func(struct cache_common_entry_ *entry,
struct cache_policy_item_ *item)
{
return ((entry->items_size > entry->common_params.satisf_elemsize) ? 1
: 0);
}
/*
* Removes the elements from the cache entry, while the continue_func returns 1.
*/
static void
flush_cache_policy(struct cache_common_entry_ *entry,
struct cache_policy_ *policy,
struct cache_policy_ *connected_policy,
int (*continue_func)(struct cache_common_entry_ *,
struct cache_policy_item_ *))
{
struct cache_policy_item_ *item, *next_item, *connected_item;
struct cache_ht_item_ *ht_item;
struct cache_ht_item_data_ *ht_item_data, ht_key;
hashtable_index_t hash;
assert(policy != NULL);
next_item = NULL;
item = policy->get_first_item_func(policy);
while ((item != NULL) && (continue_func(entry, item) == 1)) {
next_item = policy->get_next_item_func(policy, item);
connected_item = item->connected_item;
policy->remove_item_func(policy, item);
memset(&ht_key, 0, sizeof(struct cache_ht_item_data_));
ht_key.key = item->key;
ht_key.key_size = item->key_size;
hash = HASHTABLE_CALCULATE_HASH(cache_ht_, &entry->items,
&ht_key);
assert(hash >= 0);
assert(hash < HASHTABLE_ENTRIES_COUNT(&entry->items));
ht_item = HASHTABLE_GET_ENTRY(&(entry->items), hash);
ht_item_data = HASHTABLE_ENTRY_FIND(cache_ht_, ht_item,
&ht_key);
assert(ht_item_data != NULL);
free(ht_item_data->key);
free(ht_item_data->value);
HASHTABLE_ENTRY_REMOVE(cache_ht_, ht_item, ht_item_data);
--entry->items_size;
policy->destroy_item_func(item);
if (connected_item != NULL) {
connected_policy->remove_item_func(connected_policy,
connected_item);
connected_policy->destroy_item_func(connected_item);
}
item = next_item;
}
}
static void
flush_cache_entry(struct cache_entry_ *entry)
{
struct cache_mp_entry_ *mp_entry;
struct cache_common_entry_ *common_entry;
struct cache_policy_ *policy, *connected_policy;
connected_policy = NULL;
if (entry->params->entry_type == CET_COMMON) {
common_entry = (struct cache_common_entry_ *)entry;
if ((common_entry->common_params.max_lifetime.tv_sec != 0) ||
(common_entry->common_params.max_lifetime.tv_usec != 0)) {
policy = common_entry->policies[0];
if (common_entry->policies_size > 1)
connected_policy = common_entry->policies[1];
flush_cache_policy(common_entry, policy,
connected_policy,
cache_lifetime_common_continue_func);
}
if ((common_entry->common_params.max_elemsize != 0) &&
common_entry->items_size >
common_entry->common_params.max_elemsize) {
if (common_entry->policies_size > 1) {
policy = common_entry->policies[1];
connected_policy = common_entry->policies[0];
} else {
policy = common_entry->policies[0];
connected_policy = NULL;
}
flush_cache_policy(common_entry, policy,
connected_policy,
cache_elemsize_common_continue_func);
}
} else {
mp_entry = (struct cache_mp_entry_ *)entry;
if ((mp_entry->mp_params.max_lifetime.tv_sec != 0)
|| (mp_entry->mp_params.max_lifetime.tv_usec != 0)) {
if (mp_entry->last_request_time.tv_sec -
mp_entry->last_request_time.tv_sec >
mp_entry->mp_params.max_lifetime.tv_sec)
clear_cache_entry(entry);
}
}
}
struct cache_ *
init_cache(struct cache_params const *params)
{
struct cache_ *retval;
TRACE_IN(init_cache);
assert(params != NULL);
retval = (struct cache_ *)calloc(1, sizeof(struct cache_));
assert(retval != NULL);
assert(params != NULL);
memcpy(&retval->params, params, sizeof(struct cache_params));
retval->entries = (struct cache_entry_ **)calloc(1,
sizeof(struct cache_entry_ *) * INITIAL_ENTRIES_CAPACITY);
assert(retval->entries != NULL);
retval->entries_capacity = INITIAL_ENTRIES_CAPACITY;
retval->entries_size = 0;
TRACE_OUT(init_cache);
return (retval);
}
void
destroy_cache(struct cache_ *the_cache)
{
TRACE_IN(destroy_cache);
assert(the_cache != NULL);
if (the_cache->entries != NULL) {
size_t i;
for (i = 0; i < the_cache->entries_size; ++i)
destroy_cache_entry(the_cache->entries[i]);
free(the_cache->entries);
}
free(the_cache);
TRACE_OUT(destroy_cache);
}
int
register_cache_entry(struct cache_ *the_cache,
struct cache_entry_params const *params)
{
int policies_size;
size_t entry_name_size;
struct cache_common_entry_ *new_common_entry;
struct cache_mp_entry_ *new_mp_entry;
TRACE_IN(register_cache_entry);
assert(the_cache != NULL);
if (find_cache_entry(the_cache, params->entry_name) != NULL) {
TRACE_OUT(register_cache_entry);
return (-1);
}
if (the_cache->entries_size == the_cache->entries_capacity) {
struct cache_entry_ **new_entries;
size_t new_capacity;
new_capacity = the_cache->entries_capacity +
ENTRIES_CAPACITY_STEP;
new_entries = (struct cache_entry_ **)calloc(1,
sizeof(struct cache_entry_ *) * new_capacity);
assert(new_entries != NULL);
memcpy(new_entries, the_cache->entries,
sizeof(struct cache_entry_ *)
* the_cache->entries_size);
free(the_cache->entries);
the_cache->entries = new_entries;
}
entry_name_size = strlen(params->entry_name);
switch (params->entry_type)
{
case CET_COMMON:
new_common_entry = (struct cache_common_entry_ *)calloc(1,
sizeof(struct cache_common_entry_));
assert(new_common_entry != NULL);
memcpy(&new_common_entry->common_params, params,
sizeof(struct common_cache_entry_params));
new_common_entry->params =
(struct cache_entry_params *)&new_common_entry->common_params;
new_common_entry->common_params.entry_name = (char *)calloc(1,
entry_name_size+1);
assert(new_common_entry->common_params.entry_name != NULL);
strncpy(new_common_entry->common_params.entry_name,
params->entry_name, entry_name_size);
new_common_entry->name =
new_common_entry->common_params.entry_name;
HASHTABLE_INIT(&(new_common_entry->items),
struct cache_ht_item_data_, data,
new_common_entry->common_params.cache_entries_size);
if (new_common_entry->common_params.policy == CPT_FIFO)
policies_size = 1;
else
policies_size = 2;
new_common_entry->policies = (struct cache_policy_ **)calloc(1,
sizeof(struct cache_policy_ *) * policies_size);
assert(new_common_entry->policies != NULL);
new_common_entry->policies_size = policies_size;
new_common_entry->policies[0] = init_cache_fifo_policy();
if (policies_size > 1) {
switch (new_common_entry->common_params.policy) {
case CPT_LRU:
new_common_entry->policies[1] =
init_cache_lru_policy();
break;
case CPT_LFU:
new_common_entry->policies[1] =
init_cache_lfu_policy();
break;
default:
break;
}
}
new_common_entry->get_time_func =
the_cache->params.get_time_func;
the_cache->entries[the_cache->entries_size++] =
(struct cache_entry_ *)new_common_entry;
break;
case CET_MULTIPART:
new_mp_entry = (struct cache_mp_entry_ *)calloc(1,
sizeof(struct cache_mp_entry_));
assert(new_mp_entry != NULL);
memcpy(&new_mp_entry->mp_params, params,
sizeof(struct mp_cache_entry_params));
new_mp_entry->params =
(struct cache_entry_params *)&new_mp_entry->mp_params;
new_mp_entry->mp_params.entry_name = (char *)calloc(1,
entry_name_size+1);
assert(new_mp_entry->mp_params.entry_name != NULL);
strncpy(new_mp_entry->mp_params.entry_name, params->entry_name,
entry_name_size);
new_mp_entry->name = new_mp_entry->mp_params.entry_name;
TAILQ_INIT(&new_mp_entry->ws_head);
TAILQ_INIT(&new_mp_entry->rs_head);
new_mp_entry->get_time_func = the_cache->params.get_time_func;
the_cache->entries[the_cache->entries_size++] =
(struct cache_entry_ *)new_mp_entry;
break;
}
qsort(the_cache->entries, the_cache->entries_size,
sizeof(struct cache_entry_ *), entries_qsort_cmp_func);
TRACE_OUT(register_cache_entry);
return (0);
}
int
unregister_cache_entry(struct cache_ *the_cache, const char *entry_name)
{
struct cache_entry_ **del_ent;
TRACE_IN(unregister_cache_entry);
assert(the_cache != NULL);
del_ent = find_cache_entry_p(the_cache, entry_name);
if (del_ent != NULL) {
destroy_cache_entry(*del_ent);
--the_cache->entries_size;
memmove(del_ent, del_ent + 1,
(&(the_cache->entries[--the_cache->entries_size]) -
del_ent) * sizeof(struct cache_entry_ *));
TRACE_OUT(unregister_cache_entry);
return (0);
} else {
TRACE_OUT(unregister_cache_entry);
return (-1);
}
}
struct cache_entry_ *
find_cache_entry(struct cache_ *the_cache, const char *entry_name)
{
struct cache_entry_ **result;
TRACE_IN(find_cache_entry);
result = find_cache_entry_p(the_cache, entry_name);
if (result == NULL) {
TRACE_OUT(find_cache_entry);
return (NULL);
} else {
TRACE_OUT(find_cache_entry);
return (*result);
}
}
/*
* Tries to read the element with the specified key from the cache. If the
* value_size is too small, it will be filled with the proper number, and
* the user will need to call cache_read again with the value buffer, that
* is large enough.
* Function returns 0 on success, -1 on error, and -2 if the value_size is too
* small.
*/
int
cache_read(struct cache_entry_ *entry, const char *key, size_t key_size,
char *value, size_t *value_size)
{
struct cache_common_entry_ *common_entry;
struct cache_ht_item_data_ item_data, *find_res;
struct cache_ht_item_ *item;
hashtable_index_t hash;
struct cache_policy_item_ *connected_item;
TRACE_IN(cache_read);
assert(entry != NULL);
assert(key != NULL);
assert(value_size != NULL);
assert(entry->params->entry_type == CET_COMMON);
common_entry = (struct cache_common_entry_ *)entry;
memset(&item_data, 0, sizeof(struct cache_ht_item_data_));
/* can't avoid the cast here */
item_data.key = (char *)key;
item_data.key_size = key_size;
hash = HASHTABLE_CALCULATE_HASH(cache_ht_, &common_entry->items,
&item_data);
assert(hash >= 0);
assert(hash < HASHTABLE_ENTRIES_COUNT(&common_entry->items));
item = HASHTABLE_GET_ENTRY(&(common_entry->items), hash);
find_res = HASHTABLE_ENTRY_FIND(cache_ht_, item, &item_data);
if (find_res == NULL) {
TRACE_OUT(cache_read);
return (-1);
}
if ((common_entry->common_params.max_lifetime.tv_sec != 0) ||
(common_entry->common_params.max_lifetime.tv_usec != 0)) {
if (find_res->fifo_policy_item->last_request_time.tv_sec -
find_res->fifo_policy_item->creation_time.tv_sec >
common_entry->common_params.max_lifetime.tv_sec) {
free(find_res->key);
free(find_res->value);
connected_item =
find_res->fifo_policy_item->connected_item;
if (connected_item != NULL) {
common_entry->policies[1]->remove_item_func(
common_entry->policies[1],
connected_item);
common_entry->policies[1]->destroy_item_func(
connected_item);
}
common_entry->policies[0]->remove_item_func(
common_entry->policies[0],
find_res->fifo_policy_item);
common_entry->policies[0]->destroy_item_func(
find_res->fifo_policy_item);
HASHTABLE_ENTRY_REMOVE(cache_ht_, item, find_res);
--common_entry->items_size;
}
}
if ((*value_size < find_res->value_size) || (value == NULL)) {
*value_size = find_res->value_size;
TRACE_OUT(cache_read);
return (-2);
}
*value_size = find_res->value_size;
memcpy(value, find_res->value, find_res->value_size);
++find_res->fifo_policy_item->request_count;
common_entry->get_time_func(
&find_res->fifo_policy_item->last_request_time);
common_entry->policies[0]->update_item_func(common_entry->policies[0],
find_res->fifo_policy_item);
if (find_res->fifo_policy_item->connected_item != NULL) {
connected_item = find_res->fifo_policy_item->connected_item;
memcpy(&connected_item->last_request_time,
&find_res->fifo_policy_item->last_request_time,
sizeof(struct timeval));
connected_item->request_count =
find_res->fifo_policy_item->request_count;
common_entry->policies[1]->update_item_func(
common_entry->policies[1], connected_item);
}
TRACE_OUT(cache_read);
return (0);
}
/*
* Writes the value with the specified key into the cache entry.
* Functions returns 0 on success, and -1 on error.
*/
int
cache_write(struct cache_entry_ *entry, const char *key, size_t key_size,
char const *value, size_t value_size)
{
struct cache_common_entry_ *common_entry;
struct cache_ht_item_data_ item_data, *find_res;
struct cache_ht_item_ *item;
hashtable_index_t hash;
struct cache_policy_ *policy, *connected_policy;
struct cache_policy_item_ *policy_item;
struct cache_policy_item_ *connected_policy_item;
TRACE_IN(cache_write);
assert(entry != NULL);
assert(key != NULL);
assert(value != NULL);
assert(entry->params->entry_type == CET_COMMON);
common_entry = (struct cache_common_entry_ *)entry;
memset(&item_data, 0, sizeof(struct cache_ht_item_data_));
/* can't avoid the cast here */
item_data.key = (char *)key;
item_data.key_size = key_size;
hash = HASHTABLE_CALCULATE_HASH(cache_ht_, &common_entry->items,
&item_data);
assert(hash >= 0);
assert(hash < HASHTABLE_ENTRIES_COUNT(&common_entry->items));
item = HASHTABLE_GET_ENTRY(&(common_entry->items), hash);
find_res = HASHTABLE_ENTRY_FIND(cache_ht_, item, &item_data);
if (find_res != NULL) {
TRACE_OUT(cache_write);
return (-1);
}
item_data.key = (char *)malloc(key_size);
memcpy(item_data.key, key, key_size);
item_data.value = (char *)malloc(value_size);
assert(item_data.value != NULL);
memcpy(item_data.value, value, value_size);
item_data.value_size = value_size;
policy_item = common_entry->policies[0]->create_item_func();
policy_item->key = item_data.key;
policy_item->key_size = item_data.key_size;
common_entry->get_time_func(&policy_item->creation_time);
if (common_entry->policies_size > 1) {
connected_policy_item =
common_entry->policies[1]->create_item_func();
memcpy(&connected_policy_item->creation_time,
&policy_item->creation_time,
sizeof(struct timeval));
connected_policy_item->key = policy_item->key;
connected_policy_item->key_size = policy_item->key_size;
connected_policy_item->connected_item = policy_item;
policy_item->connected_item = connected_policy_item;
}
item_data.fifo_policy_item = policy_item;
common_entry->policies[0]->add_item_func(common_entry->policies[0],
policy_item);
if (common_entry->policies_size > 1)
common_entry->policies[1]->add_item_func(
common_entry->policies[1], connected_policy_item);
HASHTABLE_ENTRY_STORE(cache_ht_, item, &item_data);
++common_entry->items_size;
if ((common_entry->common_params.max_elemsize != 0) &&
(common_entry->items_size >
common_entry->common_params.max_elemsize)) {
if (common_entry->policies_size > 1) {
policy = common_entry->policies[1];
connected_policy = common_entry->policies[0];
} else {
policy = common_entry->policies[0];
connected_policy = NULL;
}
flush_cache_policy(common_entry, policy, connected_policy,
cache_elemsize_common_continue_func);
}
TRACE_OUT(cache_write);
return (0);
}
/*
* Initializes the write session for the specified multipart entry. This
* session then should be filled with data either committed or abandoned by
* using close_cache_mp_write_session or abandon_cache_mp_write_session
* respectively.
* Returns NULL on errors (when there are too many opened write sessions for
* the entry).
*/
struct cache_mp_write_session_ *
open_cache_mp_write_session(struct cache_entry_ *entry)
{
struct cache_mp_entry_ *mp_entry;
struct cache_mp_write_session_ *retval;
TRACE_IN(open_cache_mp_write_session);
assert(entry != NULL);
assert(entry->params->entry_type == CET_MULTIPART);
mp_entry = (struct cache_mp_entry_ *)entry;
if ((mp_entry->mp_params.max_sessions > 0) &&
(mp_entry->ws_size == mp_entry->mp_params.max_sessions)) {
TRACE_OUT(open_cache_mp_write_session);
return (NULL);
}
retval = (struct cache_mp_write_session_ *)calloc(1,
sizeof(struct cache_mp_write_session_));
assert(retval != NULL);
TAILQ_INIT(&retval->items);
retval->parent_entry = mp_entry;
TAILQ_INSERT_HEAD(&mp_entry->ws_head, retval, entries);
++mp_entry->ws_size;
TRACE_OUT(open_cache_mp_write_session);
return (retval);
}
/*
* Writes data to the specified session. Return 0 on success and -1 on errors
* (when write session size limit is exceeded).
*/
int
cache_mp_write(struct cache_mp_write_session_ *ws, char *data,
size_t data_size)
{
struct cache_mp_data_item_ *new_item;
TRACE_IN(cache_mp_write);
assert(ws != NULL);
assert(ws->parent_entry != NULL);
assert(ws->parent_entry->params->entry_type == CET_MULTIPART);
if ((ws->parent_entry->mp_params.max_elemsize > 0) &&
(ws->parent_entry->mp_params.max_elemsize == ws->items_size)) {
TRACE_OUT(cache_mp_write);
return (-1);
}
new_item = (struct cache_mp_data_item_ *)calloc(1,
sizeof(struct cache_mp_data_item_));
assert(new_item != NULL);
new_item->value = (char *)malloc(data_size);
assert(new_item->value != NULL);
memcpy(new_item->value, data, data_size);
new_item->value_size = data_size;
TAILQ_INSERT_TAIL(&ws->items, new_item, entries);
++ws->items_size;
TRACE_OUT(cache_mp_write);
return (0);
}
/*
* Abandons the write session and frees all the connected resources.
*/
void
abandon_cache_mp_write_session(struct cache_mp_write_session_ *ws)
{
TRACE_IN(abandon_cache_mp_write_session);
assert(ws != NULL);
assert(ws->parent_entry != NULL);
assert(ws->parent_entry->params->entry_type == CET_MULTIPART);
TAILQ_REMOVE(&ws->parent_entry->ws_head, ws, entries);
--ws->parent_entry->ws_size;
destroy_cache_mp_write_session(ws);
TRACE_OUT(abandon_cache_mp_write_session);
}
/*
* Commits the session to the entry, for which it was created.
*/
void
close_cache_mp_write_session(struct cache_mp_write_session_ *ws)
{
TRACE_IN(close_cache_mp_write_session);
assert(ws != NULL);
assert(ws->parent_entry != NULL);
assert(ws->parent_entry->params->entry_type == CET_MULTIPART);
TAILQ_REMOVE(&ws->parent_entry->ws_head, ws, entries);
--ws->parent_entry->ws_size;
if (ws->parent_entry->completed_write_session == NULL) {
/*
* If there is no completed session yet, this will be the one
*/
ws->parent_entry->get_time_func(
&ws->parent_entry->creation_time);
ws->parent_entry->completed_write_session = ws;
} else {
/*
* If there is a completed session, then we'll save our session
* as a pending session. If there is already a pending session,
* it would be destroyed.
*/
if (ws->parent_entry->pending_write_session != NULL)
destroy_cache_mp_write_session(
ws->parent_entry->pending_write_session);
ws->parent_entry->pending_write_session = ws;
}
TRACE_OUT(close_cache_mp_write_session);
}
/*
* Opens read session for the specified entry. Returns NULL on errors (when
* there are no data in the entry, or the data are obsolete).
*/
struct cache_mp_read_session_ *
open_cache_mp_read_session(struct cache_entry_ *entry)
{
struct cache_mp_entry_ *mp_entry;
struct cache_mp_read_session_ *retval;
TRACE_IN(open_cache_mp_read_session);
assert(entry != NULL);
assert(entry->params->entry_type == CET_MULTIPART);
mp_entry = (struct cache_mp_entry_ *)entry;
if (mp_entry->completed_write_session == NULL) {
TRACE_OUT(open_cache_mp_read_session);
return (NULL);
}
if ((mp_entry->mp_params.max_lifetime.tv_sec != 0)
|| (mp_entry->mp_params.max_lifetime.tv_usec != 0)) {
if (mp_entry->last_request_time.tv_sec -
mp_entry->last_request_time.tv_sec >
mp_entry->mp_params.max_lifetime.tv_sec) {
flush_cache_entry(entry);
TRACE_OUT(open_cache_mp_read_session);
return (NULL);
}
}
retval = (struct cache_mp_read_session_ *)calloc(1,
sizeof(struct cache_mp_read_session_));
assert(retval != NULL);
retval->parent_entry = mp_entry;
retval->current_item = TAILQ_FIRST(
&mp_entry->completed_write_session->items);
TAILQ_INSERT_HEAD(&mp_entry->rs_head, retval, entries);
++mp_entry->rs_size;
mp_entry->get_time_func(&mp_entry->last_request_time);
TRACE_OUT(open_cache_mp_read_session);
return (retval);
}
/*
* Reads the data from the read session - step by step.
* Returns 0 on success, -1 on error (when there are no more data), and -2 if
* the data_size is too small. In the last case, data_size would be filled
* the proper value.
*/
int
cache_mp_read(struct cache_mp_read_session_ *rs, char *data, size_t *data_size)
{
TRACE_IN(cache_mp_read);
assert(rs != NULL);
if (rs->current_item == NULL) {
TRACE_OUT(cache_mp_read);
return (-1);
}
if (rs->current_item->value_size > *data_size) {
*data_size = rs->current_item->value_size;
if (data == NULL) {
TRACE_OUT(cache_mp_read);
return (0);
}
TRACE_OUT(cache_mp_read);
return (-2);
}
*data_size = rs->current_item->value_size;
memcpy(data, rs->current_item->value, rs->current_item->value_size);
rs->current_item = TAILQ_NEXT(rs->current_item, entries);
TRACE_OUT(cache_mp_read);
return (0);
}
/*
* Closes the read session. If there are no more read sessions and there is
* a pending write session, it will be committed and old
* completed_write_session will be destroyed.
*/
void
close_cache_mp_read_session(struct cache_mp_read_session_ *rs)
{
TRACE_IN(close_cache_mp_read_session);
assert(rs != NULL);
assert(rs->parent_entry != NULL);
TAILQ_REMOVE(&rs->parent_entry->rs_head, rs, entries);
--rs->parent_entry->rs_size;
if ((rs->parent_entry->rs_size == 0) &&
(rs->parent_entry->pending_write_session != NULL)) {
destroy_cache_mp_write_session(
rs->parent_entry->completed_write_session);
rs->parent_entry->completed_write_session =
rs->parent_entry->pending_write_session;
rs->parent_entry->pending_write_session = NULL;
}
destroy_cache_mp_read_session(rs);
TRACE_OUT(close_cache_mp_read_session);
}
int
transform_cache_entry(struct cache_entry_ *entry,
enum cache_transformation_t transformation)
{
TRACE_IN(transform_cache_entry);
switch (transformation) {
case CTT_CLEAR:
clear_cache_entry(entry);
TRACE_OUT(transform_cache_entry);
return (0);
case CTT_FLUSH:
flush_cache_entry(entry);
TRACE_OUT(transform_cache_entry);
return (0);
default:
TRACE_OUT(transform_cache_entry);
return (-1);
}
}
int
transform_cache_entry_part(struct cache_entry_ *entry,
enum cache_transformation_t transformation, const char *key_part,
size_t key_part_size, enum part_position_t part_position)
{
struct cache_common_entry_ *common_entry;
struct cache_ht_item_ *ht_item;
struct cache_ht_item_data_ *ht_item_data, ht_key;
struct cache_policy_item_ *item, *connected_item;
TRACE_IN(transform_cache_entry_part);
if (entry->params->entry_type != CET_COMMON) {
TRACE_OUT(transform_cache_entry_part);
return (-1);
}
if (transformation != CTT_CLEAR) {
TRACE_OUT(transform_cache_entry_part);
return (-1);
}
memset(&ht_key, 0, sizeof(struct cache_ht_item_data_));
ht_key.key = (char *)key_part; /* can't avoid casting here */
ht_key.key_size = key_part_size;
common_entry = (struct cache_common_entry_ *)entry;
HASHTABLE_FOREACH(&(common_entry->items), ht_item) {
do {
ht_item_data = HASHTABLE_ENTRY_FIND_SPECIAL(cache_ht_,
ht_item, &ht_key,
ht_items_fixed_size_left_cmp_func);
if (ht_item_data != NULL) {
item = ht_item_data->fifo_policy_item;
connected_item = item->connected_item;
common_entry->policies[0]->remove_item_func(
common_entry->policies[0],
item);
free(ht_item_data->key);
free(ht_item_data->value);
HASHTABLE_ENTRY_REMOVE(cache_ht_, ht_item,
ht_item_data);
--common_entry->items_size;
common_entry->policies[0]->destroy_item_func(
item);
if (common_entry->policies_size == 2) {
common_entry->policies[1]->remove_item_func(
common_entry->policies[1],
connected_item);
common_entry->policies[1]->destroy_item_func(
connected_item);
}
}
} while (ht_item_data != NULL);
}
TRACE_OUT(transform_cache_entry_part);
return (0);
}