Fix to the hashtable for misisng mod of hashsize

Addition of multiple reader/single writer locking mechanism
This commit is contained in:
Mark Riddoch 2013-06-24 13:09:11 +02:00
parent 69de408d8a
commit aeda47c69d
5 changed files with 156 additions and 8 deletions

View File

@ -395,7 +395,8 @@ config.o: config.c /usr/include/stdio.h /usr/include/features.h \
../include/service.h ../include/spinlock.h ../include/thread.h \
/usr/include/pthread.h /usr/include/sched.h /usr/include/bits/sched.h \
/usr/include/bits/setjmp.h ../include/dcb.h ../include/buffer.h \
../include/server.h ../include/users.h ../include/hashtable.h
../include/server.h ../include/users.h ../include/hashtable.h \
../include/atomic.h
users.o: users.c /usr/include/stdlib.h /usr/include/features.h \
/usr/include/sys/cdefs.h /usr/include/bits/wordsize.h \
/usr/include/gnu/stubs.h /usr/include/gnu/stubs-64.h \
@ -408,7 +409,9 @@ users.o: users.c /usr/include/stdlib.h /usr/include/features.h \
/usr/include/bits/sigset.h /usr/include/bits/time.h \
/usr/include/sys/sysmacros.h /usr/include/bits/pthreadtypes.h \
/usr/include/alloca.h /usr/include/string.h /usr/include/xlocale.h \
../include/users.h ../include/hashtable.h ../include/atomic.h
../include/users.h ../include/hashtable.h ../include/spinlock.h \
../include/thread.h /usr/include/pthread.h /usr/include/sched.h \
/usr/include/bits/sched.h /usr/include/bits/setjmp.h ../include/atomic.h
hashtable.o: hashtable.c /usr/include/stdlib.h /usr/include/features.h \
/usr/include/sys/cdefs.h /usr/include/bits/wordsize.h \
/usr/include/gnu/stubs.h /usr/include/gnu/stubs-64.h \
@ -421,4 +424,6 @@ hashtable.o: hashtable.c /usr/include/stdlib.h /usr/include/features.h \
/usr/include/bits/sigset.h /usr/include/bits/time.h \
/usr/include/sys/sysmacros.h /usr/include/bits/pthreadtypes.h \
/usr/include/alloca.h /usr/include/string.h /usr/include/xlocale.h \
../include/hashtable.h
../include/hashtable.h ../include/spinlock.h ../include/thread.h \
/usr/include/pthread.h /usr/include/sched.h /usr/include/bits/sched.h \
/usr/include/bits/setjmp.h ../include/atomic.h

View File

@ -22,6 +22,29 @@
/**
* @file hashtable.c General purpose hashtable routines
*
* The hashtable can be create with a custom number of hash buckets,
* a hash function and optional functions to call make copies of the key
* and value and to free them.
*
* The hashtable is arrange as a set of linked lists, the number of linked
* lists beign the hashsize as requested by the user. Entries are hashed by
* calling the hash function that is passed in by the user, this is used as
* an index into the array of linked lists, usign modulo hashsize.
*
* The linked lists are searched using the key comparison function that is
* passed into the hash table creation routine.
*
* By default the hash table keeps the original pointers that are passed in
* for the keys and values, however two functions can be supplied to copy these
* a copy function and a free function. Please note the same function is used for
* the key and the value, if the actions required are different the called functions
* must understand how to differenate the key and value.
*
* The hash table implements a single write, multiple reader locking policy by
* using a pair of counters and a spinlock. The spinlock is used to protect the
* number of readers and writers counters when taking out locks. Releasing of
* locks uses pure atomic actions and thus does not require spinlock protection.
*
* @verbatim
* Revision History
*
@ -31,9 +54,14 @@
* @endverbatim
*/
static void hashtable_read_lock(HASHTABLE *table);
static void hashtable_read_unlock(HASHTABLE *table);
static void hashtable_write_lock(HASHTABLE *table);
static void hashtable_write_unlock(HASHTABLE *table);
/**
* Special null function used as default memory allfunctions in the hashtable
* implementation. This avoids havign to special case the code that manipulates
* implementation. This avoids having to special case the code that manipulates
* the keys and values
*
* @param data The data pointer
@ -65,6 +93,9 @@ HASHTABLE *rval;
rval->cmpfn = cmpfn;
rval->copyfn = nullfn;
rval->freefn = nullfn;
rval->n_readers = 0;
rval->writelock = 0;
spinlock_init(&rval->spin);
if ((rval->entries = calloc(size, sizeof(HASHENTRIES))) == NULL)
{
free(rval);
@ -86,6 +117,7 @@ hashtable_free(HASHTABLE *table)
int i;
HASHENTRIES *entry, *ptr;
hashtable_write_lock(table);
for (i = 0; i < table->hashsize; i++)
{
entry = table->entries[i];
@ -135,9 +167,10 @@ hashtable_memory_fns(HASHTABLE *table, HASHMEMORYFN copyfn, HASHMEMORYFN freefn)
int
hashtable_add(HASHTABLE *table, void *key, void *value)
{
int hashkey = table->hashfn(key);
int hashkey = table->hashfn(key) % table->hashsize;
HASHENTRIES *entry;
hashtable_write_lock(table);
entry = table->entries[hashkey % table->hashsize];
while (entry->next && entry->key && table->cmpfn(key, entry->key) != 0)
{
@ -152,18 +185,23 @@ HASHENTRIES *entry;
else if (table->cmpfn(key, entry->key) == 0)
{
/* Duplicate key value */
hashtable_write_unlock(table);
return 0;
}
else
{
HASHENTRIES *ptr = (HASHENTRIES *)malloc(sizeof(HASHENTRIES));
if (ptr == NULL)
{
hashtable_write_unlock(table);
return 0;
}
ptr->key = table->copyfn(key);
ptr->value = table->copyfn(value);
ptr->next = NULL;
entry->next = ptr;
}
hashtable_write_unlock(table);
return 1;
}
@ -177,9 +215,10 @@ HASHENTRIES *entry;
int
hashtable_delete(HASHTABLE *table, void *key)
{
int hashkey = table->hashfn(key);
int hashkey = table->hashfn(key) % table->hashsize;
HASHENTRIES *entry, *ptr;
hashtable_write_lock(table);
entry = table->entries[hashkey % table->hashsize];
while (entry && entry->key && table->cmpfn(key, entry->key) != 0)
{
@ -188,6 +227,7 @@ HASHENTRIES *entry, *ptr;
if (entry == NULL || entry->key == NULL)
{
/* Not found */
hashtable_write_unlock(table);
return 0;
}
@ -218,12 +258,16 @@ HASHENTRIES *entry, *ptr;
while (ptr && ptr->next != entry)
ptr = ptr->next;
if (ptr == NULL)
{
hashtable_write_unlock(table);
return 0; /* This should never happen */
}
ptr->next = entry->next;
table->freefn(entry->key);
table->freefn(entry->value);
free(entry);
}
hashtable_write_unlock(table);
return 1;
}
@ -237,9 +281,10 @@ HASHENTRIES *entry, *ptr;
void *
hashtable_fetch(HASHTABLE *table, void *key)
{
int hashkey = table->hashfn(key);
int hashkey = table->hashfn(key) % table->hashsize;
HASHENTRIES *entry;
hashtable_read_lock(table);
entry = table->entries[hashkey % table->hashsize];
while (entry && entry->key && table->cmpfn(key, entry->key) != 0)
{
@ -247,11 +292,102 @@ HASHENTRIES *entry;
}
if (entry == NULL || entry->key == NULL)
{
hashtable_read_unlock(table);
return NULL;
}
else
{
hashtable_read_unlock(table);
return entry->value;
}
}
/**
* Take a read lock on the hashtable.
*
* The hashtable support multiple readers and a single writer,
* we have a spinlock to protect the two counts, n_readers and
* writelock.
*
* We take the hashtable spinlock and then check that writelock
* is set to zero. If not we release the spinlock and do dirty
* reads of writelock until it goes to 0. Once it is zero we
* acquire the spinlock again and test that writelock is still
* 0.
*
* With writelock set to zero we increment n_readers with the
* spinlock still held.
*
* @param table The hashtable to lock.
*/
static void
hashtable_read_lock(HASHTABLE *table)
{
spinlock_acquire(&table->spin);
while (table->writelock)
{
spinlock_release(&table->spin);
while (table->writelock)
;
spinlock_acquire(&table->spin);
}
table->n_readers++;
spinlock_release(&table->spin);
}
/**
* Release a previously obtained readlock.
*
* Simply decrement the n_readers value for the hash table
*
* @param table The hash table to unlock
*/
static void
hashtable_read_unlock(HASHTABLE *table)
{
atomic_add(&table->n_readers, -1);
}
/**
* Obtain an exclusive write lock for the hash table.
*
* We acquire the hashtable spinlock, check for the number of
* readers beign zero. If it is not we hold the spinlock and
* loop waiting for the n_readers to reach zero. This will prevent
* any new readers beign granted access but will not prevent current
* readers releasing the read lock.
*
* Once we have no readers we increment writelock and test if we are
* the only writelock holder, if not we repeat the process. We hold
* the spinlock throughout the process since both read and write
* locks do not require the spinlock to be acquired.
*
* @param table The table to lock for updates
*/
static void
hashtable_write_lock(HASHTABLE *table)
{
int available;
spinlock_acquire(&table->spin);
do {
while (table->n_readers)
;
available = atomic_add(&table->writelock, 1);
if (available != 0)
atomic_add(&table->writelock, -1);
} while (available != 0);
spinlock_release(&table->spin);
}
/**
* Release the write lock on the hash table.
*
* @param table The hash table to unlock
*/
static void
hashtable_write_unlock(HASHTABLE *table)
{
atomic_add(&table->writelock, -1);
}

View File

@ -30,6 +30,8 @@
*
* @endverbatim
*/
#include <spinlock.h>
#include <atomic.h>
/**
* The entries within a hashtable.
@ -58,6 +60,9 @@ typedef struct hashtable {
int (*cmpfn)(void *, void *); /**< The key comparison function */
HASHMEMORYFN copyfn; /**< Optional copy function */
HASHMEMORYFN freefn; /**< Optional free function */
SPINLOCK spin; /**< Internal spinlock for the hashtable */
int n_readers; /**< Number of clients reading the table */
int writelock; /**< The table is locked by a writer */
} HASHTABLE;
extern HASHTABLE *hashtable_alloc(int, int (*hashfn)(), int (*cmpfn)());

View File

@ -31,7 +31,7 @@
#include <thread.h>
typedef struct spinlock {
int lock;
volatile int lock;
#if DEBUG
int spins;
int acquired;

View File

@ -18,6 +18,7 @@
* Copyright SkySQL Ab 2013
*/
#include <hashtable.h>
/**
* @file users.h The functions to manipulate the table of users maintained
* for each service
@ -40,6 +41,7 @@ typedef struct {
int n_deletes; /**< The number of deletes */
int n_fetches; /**< The number of fetchs */
} USERS_STATS;
/**
* The user table, this contains the username and authentication data required
* for the authentication implementation within the gateway.