open62541/plugins/ua_nodestore_hashmap.c

/* This work is licensed under a Creative Commons CCZero 1.0 Universal License.
 * See http://creativecommons.org/publicdomain/zero/1.0/ for more information. 
 *
 *    Copyright 2014-2019 (c) Fraunhofer IOSB (Author: Julius Pfrommer)
 *    Copyright 2017 (c) Julian Grothoff
 *    Copyright 2017 (c) Stefan Profanter, fortiss GmbH
 */

#include <open62541/plugin/nodestore_default.h>

#if UA_MULTITHREADING >= 100
#define BEGIN_CRITSECT(NODEMAP) UA_LOCK(NODEMAP->lock)
#define END_CRITSECT(NODEMAP) UA_UNLOCK(NODEMAP->lock)
#else
#define BEGIN_CRITSECT(NODEMAP) do {} while(0)
#define END_CRITSECT(NODEMAP) do {} while(0)
#endif

/* container_of */
#define container_of(ptr, type, member) \
    (type *)((uintptr_t)ptr - offsetof(type,member))

/* The default Nodestore is simply a hash-map from NodeIds to Nodes. To find an
 * entry, iterate over candidate positions according to the NodeId hash.
 *
 * - Tombstone or non-matching NodeId: continue searching
 * - Matching NodeId: Return the entry
 * - NULL: Abort the search
 *
 * The nodestore uses atomic operations to set entries of the hash-map. If
 * UA_ENABLE_IMMUTABLE_NODES is configured, the nodestore allows read-access
 * from an interrupt without seeing corrupted nodes. For true multi-threaded
 * access, a mutex is used.
 *
 * Multi-threading without a mutex could be realized with the Linux RCU mechanism.
 * But this is not done for this implementation of the nodestore. */

typedef struct UA_NodeMapEntry {
    UA_UInt32 nodeIdHash;
    struct UA_NodeMapEntry *orig; /* the version this is a copy from (or NULL) */
    UA_UInt16 refCount; /* How many consumers have a reference to the node? */
    UA_Boolean deleted; /* Node was marked as deleted and can be deleted when refCount == 0 */
    UA_Node node;
} UA_NodeMapEntry;

#define UA_NODEMAP_MINSIZE 64
#define UA_NODEMAP_TOMBSTONE ((UA_NodeMapEntry*)0x01)

typedef struct {
    UA_NodeMapEntry **entries;
    UA_UInt32 size;
    UA_UInt32 count;
    UA_UInt32 sizePrimeIndex;
#if UA_MULTITHREADING >= 100
    UA_LOCK_TYPE(lock) /* Protect access */
#endif
} UA_NodeMap;

/*********************/
/* HashMap Utilities */
/*********************/

/* The size of the hash-map is always a prime number. They are chosen to be
 * close to the next power of 2. So the size ca. doubles with each prime. */
static UA_UInt32 const primes[] = {
    7,         13,         31,         61,         127,         251,
    509,       1021,       2039,       4093,       8191,        16381,
    32749,     65521,      131071,     262139,     524287,      1048573,
    2097143,   4194301,    8388593,    16777213,   33554393,    67108859,
    134217689, 268435399,  536870909,  1073741789, 2147483647,  4294967291
};

static UA_UInt32 mod(UA_UInt32 h, UA_UInt32 size) { return h % size; }
static UA_UInt32 mod2(UA_UInt32 h, UA_UInt32 size) { return 1 + (h % (size - 2)); }

static UA_UInt16
higher_prime_index(UA_UInt32 n) {
    UA_UInt16 low  = 0;
    UA_UInt16 high = (UA_UInt16)(sizeof(primes) / sizeof(UA_UInt32));
    while(low != high) {
        UA_UInt16 mid = (UA_UInt16)(low + ((high - low) / 2));
        if(n > primes[mid])
            low = (UA_UInt16)(mid + 1);
        else
            high = mid;
    }
    return low;
}

/* Returns an empty slot or null if the nodeid exists or if no empty slot is found. */
static UA_NodeMapEntry **
findFreeSlot(const UA_NodeMap *ns, const UA_NodeId *nodeid) {
    UA_NodeMapEntry **retval = NULL;
    UA_UInt32 h = UA_NodeId_hash(nodeid);
    UA_UInt32 size = ns->size;
    UA_UInt64 idx = mod(h, size); /* Use 64bit container to avoid overflow  */
    UA_UInt32 startIdx = (UA_UInt32)idx;
    UA_UInt32 hash2 = mod2(h, size);
    UA_NodeMapEntry *entry = NULL;

    do {
        entry = ns->entries[(UA_UInt32)idx];
        if(entry > UA_NODEMAP_TOMBSTONE &&
           UA_NodeId_equal(&entry->node.nodeId, nodeid))
            return NULL;
        if(!retval && entry <= UA_NODEMAP_TOMBSTONE)
            retval = &ns->entries[(UA_UInt32)idx];
        idx += hash2;
        if(idx >= size)
            idx -= size;
    } while((UA_UInt32)idx != startIdx && entry);

    /* NULL is returned if there is no free slot (idx == startIdx).
     * Otherwise the first free slot is returned after we are sure,
     * that the node id cannot be found in the used hashmap (!entry). */
    return retval;
}

/* The occupancy of the table after the call will be about 50% */
static UA_StatusCode
expand(UA_NodeMap *ns) {
    UA_UInt32 osize = ns->size;
    UA_UInt32 count = ns->count;
    /* Resize only when table after removal of unused elements is either too
       full or too empty */
    if(count * 2 < osize && (count * 8 > osize || osize <= UA_NODEMAP_MINSIZE))
        return UA_STATUSCODE_GOOD;

    UA_NodeMapEntry **oentries = ns->entries;
    UA_UInt32 nindex = higher_prime_index(count * 2);
    UA_UInt32 nsize = primes[nindex];
    UA_NodeMapEntry **nentries = (UA_NodeMapEntry **)UA_calloc(nsize, sizeof(UA_NodeMapEntry*));
    if(!nentries)
        return UA_STATUSCODE_BADOUTOFMEMORY;

    ns->entries = nentries;
    ns->size = nsize;
    ns->sizePrimeIndex = nindex;

    /* recompute the position of every entry and insert the pointer */
    for(size_t i = 0, j = 0; i < osize && j < count; ++i) {
        if(oentries[i] <= UA_NODEMAP_TOMBSTONE)
            continue;
        UA_NodeMapEntry **e = findFreeSlot(ns, &oentries[i]->node.nodeId);
        UA_assert(e);
        *e = oentries[i];
        ++j;
    }

    UA_free(oentries);
    return UA_STATUSCODE_GOOD;
}

static UA_NodeMapEntry *
newEntry(UA_NodeClass nodeClass) {
    size_t size = sizeof(UA_NodeMapEntry) - sizeof(UA_Node);
    switch(nodeClass) {
    case UA_NODECLASS_OBJECT:
        size += sizeof(UA_ObjectNode);
        break;
    case UA_NODECLASS_VARIABLE:
        size += sizeof(UA_VariableNode);
        break;
    case UA_NODECLASS_METHOD:
        size += sizeof(UA_MethodNode);
        break;
    case UA_NODECLASS_OBJECTTYPE:
        size += sizeof(UA_ObjectTypeNode);
        break;
    case UA_NODECLASS_VARIABLETYPE:
        size += sizeof(UA_VariableTypeNode);
        break;
    case UA_NODECLASS_REFERENCETYPE:
        size += sizeof(UA_ReferenceTypeNode);
        break;
    case UA_NODECLASS_DATATYPE:
        size += sizeof(UA_DataTypeNode);
        break;
    case UA_NODECLASS_VIEW:
        size += sizeof(UA_ViewNode);
        break;
    default:
        return NULL;
    }
    UA_NodeMapEntry *entry = (UA_NodeMapEntry*)UA_calloc(1, size);
    if(!entry)
        return NULL;
    entry->node.nodeClass = nodeClass;
    return entry;
}

static void
deleteEntry(UA_NodeMapEntry *entry) {
    UA_Node_clear(&entry->node);
    UA_free(entry);
}

static void
cleanupEntry(UA_NodeMapEntry *entry) {
    if(entry->deleted && entry->refCount == 0)
        deleteEntry(entry);
}

static UA_StatusCode
clearSlot(UA_NodeMap *ns, UA_NodeMapEntry **slot) {
    UA_NodeMapEntry *entry = *slot;
    if(UA_atomic_cmpxchg((void**)slot, entry, UA_NODEMAP_TOMBSTONE) != entry)
        return UA_STATUSCODE_BADINTERNALERROR;
    entry->deleted = true;
    cleanupEntry(entry);
    --ns->count;
    /* Downsize the hashmap if it is very empty */
    if(ns->count * 8 < ns->size && ns->size > 32)
        expand(ns); /* Can fail. Just continue with the bigger hashmap. */
    return UA_STATUSCODE_GOOD;
}

static UA_NodeMapEntry **
findOccupiedSlot(const UA_NodeMap *ns, const UA_NodeId *nodeid) {
    UA_UInt32 h = UA_NodeId_hash(nodeid);
    UA_UInt32 size = ns->size;
    UA_UInt64 idx = mod(h, size); /* Use 64bit container to avoid overflow */
    UA_UInt32 hash2 = mod2(h, size);
    UA_UInt32 startIdx = (UA_UInt32)idx;
    UA_NodeMapEntry *entry = NULL;

    do {
        entry = ns->entries[(UA_UInt32)idx];
        if(entry > UA_NODEMAP_TOMBSTONE &&
           entry->nodeIdHash == h &&
           UA_NodeId_equal(&entry->node.nodeId, nodeid))
            return &ns->entries[(UA_UInt32)idx];
        idx += hash2;
        if(idx >= size)
            idx -= size;
    } while((UA_UInt32)idx != startIdx && entry);

    /* NULL is returned if there is no free slot (idx == startIdx)
     * and the node id is not found or if the end of the used slots (!entry)
     * is reached. */
    return NULL;
}

/***********************/
/* Interface functions */
/***********************/

static UA_Node *
UA_NodeMap_newNode(void *context, UA_NodeClass nodeClass) {
    UA_NodeMapEntry *entry = newEntry(nodeClass);
    if(!entry)
        return NULL;
    return &entry->node;
}

static void
UA_NodeMap_deleteNode(void *context, UA_Node *node) {
#if UA_MULTITHREADING >= 100
    UA_NodeMap *ns = (UA_NodeMap*)context;
#endif
    BEGIN_CRITSECT(ns);
    UA_NodeMapEntry *entry = container_of(node, UA_NodeMapEntry, node);
    UA_assert(&entry->node == node);
    deleteEntry(entry);
    END_CRITSECT(ns);
}

static const UA_Node *
UA_NodeMap_getNode(void *context, const UA_NodeId *nodeid) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    BEGIN_CRITSECT(ns);
    UA_NodeMapEntry **entry = findOccupiedSlot(ns, nodeid);
    if(!entry) {
        END_CRITSECT(ns);
        return NULL;
    }
    ++(*entry)->refCount;
    END_CRITSECT(ns);
    return (const UA_Node*)&(*entry)->node;
}

static void
UA_NodeMap_releaseNode(void *context, const UA_Node *node) {
    if (!node)
        return;
#if UA_MULTITHREADING >= 100
    UA_NodeMap *ns = (UA_NodeMap*)context;
#endif
    BEGIN_CRITSECT(ns);
    UA_NodeMapEntry *entry = container_of(node, UA_NodeMapEntry, node);
    UA_assert(&entry->node == node);
    UA_assert(entry->refCount > 0);
    --entry->refCount;
    cleanupEntry(entry);
    END_CRITSECT(ns);
}

static UA_StatusCode
UA_NodeMap_getNodeCopy(void *context, const UA_NodeId *nodeid,
                       UA_Node **outNode) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    BEGIN_CRITSECT(ns);
    UA_NodeMapEntry **slot = findOccupiedSlot(ns, nodeid);
    if(!slot) {
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADNODEIDUNKNOWN;
    }
    UA_NodeMapEntry *entry = *slot;
    UA_NodeMapEntry *newItem = newEntry(entry->node.nodeClass);
    if(!newItem) {
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADOUTOFMEMORY;
    }
    UA_StatusCode retval = UA_Node_copy(&entry->node, &newItem->node);
    if(retval == UA_STATUSCODE_GOOD) {
        newItem->orig = entry; /* Store the pointer to the original */
        *outNode = &newItem->node;
    } else {
        deleteEntry(newItem);
    }
    END_CRITSECT(ns);
    return retval;
}

static UA_StatusCode
UA_NodeMap_removeNode(void *context, const UA_NodeId *nodeid) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    BEGIN_CRITSECT(ns);
    UA_NodeMapEntry **slot = findOccupiedSlot(ns, nodeid);
    UA_StatusCode retval = UA_STATUSCODE_GOOD;
    if(slot)
        retval = clearSlot(ns, slot);
    else
        retval = UA_STATUSCODE_BADNODEIDUNKNOWN;
    END_CRITSECT(ns);
    return retval;
}

static UA_StatusCode
UA_NodeMap_insertNode(void *context, UA_Node *node,
                      UA_NodeId *addedNodeId) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    BEGIN_CRITSECT(ns);
    if(ns->size * 3 <= ns->count * 4) {
        if(expand(ns) != UA_STATUSCODE_GOOD) {
            END_CRITSECT(ns);
            return UA_STATUSCODE_BADINTERNALERROR;
        }
    }

    UA_NodeMapEntry **slot;
    if(node->nodeId.identifierType == UA_NODEIDTYPE_NUMERIC &&
            node->nodeId.identifier.numeric == 0) {
        /* Create a random nodeid: Start at least with 50,000 to make sure we
         * don not conflict with nodes from the spec. If we find a conflict, we
         * just try another identifier until we have tried all possible
         * identifiers. Since the size is prime and we don't change the increase
         * val, we will reach the starting id again. E.g. adding a nodeset will
         * create children while there are still other nodes which need to be
         * created. Thus the node ids may collide. */
        UA_UInt32 size = ns->size;
        UA_UInt64 identifier = mod(50000 + size+1, UA_UINT32_MAX); /* Use 64bit to
                                                                    * avoid overflow */
        UA_UInt32 increase = mod2(ns->count+1, size);
        UA_UInt32 startId = (UA_UInt32)identifier; /* mod ensures us that the id
                                                    * is a valid 32 bit integer */

        do {
            node->nodeId.identifier.numeric = (UA_UInt32)identifier;
            slot = findFreeSlot(ns, &node->nodeId);
            if(slot)
                break;
            identifier += increase;
            if(identifier >= size)
                identifier -= size;
        } while((UA_UInt32)identifier != startId);
    } else {
        slot = findFreeSlot(ns, &node->nodeId);
    }

    if(!slot) {
        deleteEntry(container_of(node, UA_NodeMapEntry, node));
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADNODEIDEXISTS;
    }

    /* Copy the NodeId */
    UA_StatusCode retval = UA_STATUSCODE_GOOD;
    if(addedNodeId) {
        retval = UA_NodeId_copy(&node->nodeId, addedNodeId);
        if(retval != UA_STATUSCODE_GOOD) {
            deleteEntry(container_of(node, UA_NodeMapEntry, node));
            END_CRITSECT(ns);
            return retval;
        }
    }

    /* Store the hash */
    UA_NodeMapEntry *newEntry = container_of(node, UA_NodeMapEntry, node);
    newEntry->nodeIdHash = UA_NodeId_hash(&node->nodeId);

    /* Insert the node */
    UA_NodeMapEntry *oldEntry = *slot;
    if(UA_atomic_cmpxchg((void**)slot, oldEntry, newEntry) != oldEntry) {
        deleteEntry(container_of(node, UA_NodeMapEntry, node));
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADNODEIDEXISTS;
    }
    ++ns->count;
    END_CRITSECT(ns);
    return retval;
}

static UA_StatusCode
UA_NodeMap_replaceNode(void *context, UA_Node *node) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    UA_NodeMapEntry *newEntryContainer = container_of(node, UA_NodeMapEntry, node);

    BEGIN_CRITSECT(ns);

    /* Find the node */
    UA_NodeMapEntry **slot = findOccupiedSlot(ns, &node->nodeId);
    if(!slot) {
        deleteEntry(newEntryContainer);
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADNODEIDUNKNOWN;
    }
    UA_NodeMapEntry *oldEntryContainer = *slot;

    /* The node was already updated since the copy was made? */
    if(oldEntryContainer != newEntryContainer->orig) {
        deleteEntry(newEntryContainer);
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADINTERNALERROR;
    }

    /* Store the hash */
    newEntryContainer->nodeIdHash = oldEntryContainer->nodeIdHash;

    /* Replace the entry with an atomic operation */
    if(UA_atomic_cmpxchg((void**)slot, oldEntryContainer,
                         newEntryContainer) != oldEntryContainer) {
        deleteEntry(newEntryContainer);
        END_CRITSECT(ns);
        return UA_STATUSCODE_BADINTERNALERROR;
    }

    oldEntryContainer->deleted = true;
    cleanupEntry(oldEntryContainer);
    END_CRITSECT(ns);
    return UA_STATUSCODE_GOOD;
}

static void
UA_NodeMap_iterate(void *context, UA_NodestoreVisitor visitor,
                   void *visitorContext) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
    BEGIN_CRITSECT(ns);
    for(UA_UInt32 i = 0; i < ns->size; ++i) {
        if(ns->entries[i] > UA_NODEMAP_TOMBSTONE) {
            END_CRITSECT(ns);
            UA_NodeMapEntry *entry = ns->entries[i];
            entry->refCount++;
            visitor(visitorContext, &entry->node);
            entry->refCount--;
            cleanupEntry(entry);
            BEGIN_CRITSECT(ns);
        }
    }
    END_CRITSECT(ns);
}

static void
UA_NodeMap_delete(void *context) {
    UA_NodeMap *ns = (UA_NodeMap*)context;
#if UA_MULTITHREADING >= 100
    UA_LOCK_DESTROY(ns->lock);
#endif
    UA_UInt32 size = ns->size;
    UA_NodeMapEntry **entries = ns->entries;
    for(UA_UInt32 i = 0; i < size; ++i) {
        if(entries[i] > UA_NODEMAP_TOMBSTONE) {
            /* On debugging builds, check that all nodes were release */
            UA_assert(entries[i]->refCount == 0);
            /* Delete the node */
            deleteEntry(entries[i]);
        }
    }
    UA_free(ns->entries);
    UA_free(ns);
}

UA_StatusCode
UA_Nodestore_HashMap(UA_Nodestore *ns) {
    /* Allocate and initialize the nodemap */
    UA_NodeMap *nodemap = (UA_NodeMap*)UA_malloc(sizeof(UA_NodeMap));
    if(!nodemap)
        return UA_STATUSCODE_BADOUTOFMEMORY;
    nodemap->sizePrimeIndex = higher_prime_index(UA_NODEMAP_MINSIZE);
    nodemap->size = primes[nodemap->sizePrimeIndex];
    nodemap->count = 0;
    nodemap->entries = (UA_NodeMapEntry**)
        UA_calloc(nodemap->size, sizeof(UA_NodeMapEntry*));
    if(!nodemap->entries) {
        UA_free(nodemap);
        return UA_STATUSCODE_BADOUTOFMEMORY;
    }

#if UA_MULTITHREADING >= 100
    UA_LOCK_INIT(nodemap->lock)
#endif

    /* Populate the nodestore */
    ns->context = nodemap;
    ns->clear = UA_NodeMap_delete;
    ns->newNode = UA_NodeMap_newNode;
    ns->deleteNode = UA_NodeMap_deleteNode;
    ns->getNode = UA_NodeMap_getNode;
    ns->releaseNode = UA_NodeMap_releaseNode;
    ns->getNodeCopy = UA_NodeMap_getNodeCopy;
    ns->insertNode = UA_NodeMap_insertNode;
    ns->replaceNode = UA_NodeMap_replaceNode;
    ns->removeNode = UA_NodeMap_removeNode;
    ns->iterate = UA_NodeMap_iterate;

    return UA_STATUSCODE_GOOD;
}