/*
This file is part of GNUnet
- Copyright (C) 2013 Christian Grothoff (and other contributing authors)
+ Copyright (C) 2013-2016 GNUnet e.V.
GNUnet is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published
*/
/**
* @file set/gnunet-service-set_union.c
+
* @brief two-peer set operations
* @author Florian Dold
*/
#include "platform.h"
#include "gnunet_util_lib.h"
+#include "gnunet_statistics_service.h"
#include "gnunet-service-set.h"
#include "ibf.h"
#include "gnunet-service-set_union_strata_estimator.h"
#include <gcrypt.h>
+#define LOG(kind,...) GNUNET_log_from (kind, "set-union",__VA_ARGS__)
+
+
/**
* Number of IBFs in a strata estimator.
*/
#define SE_STRATA_COUNT 32
+
/**
* Size of the IBFs in the strata estimator.
*/
#define SE_IBF_SIZE 80
+
/**
- * hash num parameter for the difference digests and strata estimators
+ * The hash num parameter for the difference digests and strata estimators.
*/
#define SE_IBF_HASH_NUM 4
* Choose this value so that computing the IBF is still cheaper
* than transmitting all values.
*/
-#define MAX_IBF_ORDER (16)
+#define MAX_IBF_ORDER (20)
/**
* Number of buckets used in the ibf per estimated
enum UnionOperationPhase
{
/**
- * We sent the request message, and expect a strata estimator
+ * We sent the request message, and expect a strata estimator.
*/
PHASE_EXPECT_SE,
* We sent the strata estimator, and expect an IBF. This phase is entered once
* upon initialization and later via #PHASE_EXPECT_ELEMENTS_AND_REQUESTS.
*
+ * XXX: could use better wording.
+ *
* After receiving the complete IBF, we enter #PHASE_EXPECT_ELEMENTS
*/
PHASE_EXPECT_IBF,
PHASE_EXPECT_IBF_CONT,
/**
- * We are sending request and elements,
- * and thus only expect elements from the other peer.
- *
- * We are currently decoding an IBF until it can no longer be decoded,
- * we currently send requests and expect elements
- * The remote peer is in #PHASE_EXPECT_ELEMENTS_AND_REQUESTS
+ * We are decoding an IBF.
*/
- PHASE_EXPECT_ELEMENTS,
+ PHASE_INVENTORY_ACTIVE,
/**
- * We are expecting elements and requests, and send
- * requested elements back to the other peer.
- *
- * We are in this phase if we have SENT an IBF for the remote peer to decode.
- * We expect requests, send elements or could receive an new IBF, which takes
- * us via #PHASE_EXPECT_IBF to phase #PHASE_EXPECT_ELEMENTS
- *
- * The remote peer is thus in:
- * #PHASE_EXPECT_ELEMENTS
+ * The other peer is decoding the IBF we just sent.
*/
- PHASE_EXPECT_ELEMENTS_AND_REQUESTS,
+ PHASE_INVENTORY_PASSIVE,
/**
- * The protocol is over.
- * Results may still have to be sent to the client.
+ * The protocol is almost finished, but we still have to flush our message
+ * queue and/or expect some elements.
*/
- PHASE_FINISHED
+ PHASE_FINISH_CLOSING,
+
+ /**
+ * In the penultimate phase,
+ * we wait until all our demands
+ * are satisfied. Then we send a done
+ * message, and wait for another done message.*/
+ PHASE_FINISH_WAITING,
+
+ /**
+ * In the ultimate phase, we wait until
+ * our demands are satisfied and then
+ * quit (sending another DONE message). */
+ PHASE_DONE
};
*/
struct OperationState
{
-
/**
* Copy of the set's strata estimator at the time of
- * creation of this operation
+ * creation of this operation.
*/
struct StrataEstimator *se;
/**
- * The ibf we currently receive
+ * The IBF we currently receive.
*/
struct InvertibleBloomFilter *remote_ibf;
/**
- * IBF of the set's element.
+ * The IBF with the local set's element.
*/
struct InvertibleBloomFilter *local_ibf;
*/
struct GNUNET_CONTAINER_MultiHashMap32 *key_to_element;
- /**
- * Iterator for sending elements on the key to element mapping to the client.
- */
- struct GNUNET_CONTAINER_MultiHashMap32Iterator *full_result_iter;
-
/**
* Current state of the operation.
*/
int client_done_sent;
/**
- * Number of ibf buckets received
+ * Number of ibf buckets already received into the @a remote_ibf.
*/
unsigned int ibf_buckets_received;
+ /**
+ * Hashes for elements that we have demanded from the other peer.
+ */
+ struct GNUNET_CONTAINER_MultiHashMap *demanded_hashes;
+
+ /**
+ * Salt that we're using for sending IBFs
+ */
+ uint32_t salt_send;
+
+ /**
+ * Salt for the IBF we've received and that we're currently decoding.
+ */
+ uint32_t salt_receive;
};
/**
* The actual element associated with the key.
+ *
+ * Only owned by the union operation if element->operation
+ * is #GNUNET_YES.
*/
struct ElementEntry *element;
-
- /**
- * Element that collides with this element
- * on the ibf key. All colliding entries must have the same ibf key.
- */
- struct KeyEntry *next_colliding;
};
{
struct KeyEntry *k = value;
- while (NULL != k)
+ GNUNET_assert (NULL != k);
+ if (GNUNET_YES == k->element->remote)
{
- struct KeyEntry *k_tmp = k;
-
- k = k->next_colliding;
- if (GNUNET_YES == k_tmp->element->remote)
- {
- GNUNET_free (k_tmp->element);
- k_tmp->element = NULL;
- }
- GNUNET_free (k_tmp);
+ GNUNET_free (k->element);
+ k->element = NULL;
}
+ GNUNET_free (k);
return GNUNET_YES;
}
static void
union_op_cancel (struct Operation *op)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "destroying union op\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "destroying union op\n");
/* check if the op was canceled twice */
GNUNET_assert (NULL != op->state);
if (NULL != op->state->remote_ibf)
ibf_destroy (op->state->remote_ibf);
op->state->remote_ibf = NULL;
}
+ if (NULL != op->state->demanded_hashes)
+ {
+ GNUNET_CONTAINER_multihashmap_destroy (op->state->demanded_hashes);
+ op->state->demanded_hashes = NULL;
+ }
if (NULL != op->state->local_ibf)
{
ibf_destroy (op->state->local_ibf);
}
GNUNET_free (op->state);
op->state = NULL;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "destroying union op done\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "destroying union op done\n");
}
struct GNUNET_MQ_Envelope *ev;
struct GNUNET_SET_ResultMessage *msg;
- GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
- "union operation failed\n");
+ LOG (GNUNET_ERROR_TYPE_ERROR,
+ "union operation failed\n");
ev = GNUNET_MQ_msg (msg, GNUNET_MESSAGE_TYPE_SET_RESULT);
msg->result_status = htons (GNUNET_SET_STATUS_FAILURE);
msg->request_id = htonl (op->spec->client_request_id);
* a salt.
*
* @param src the hash code
- * @param salt salt to use
* @return the derived IBF key
*/
static struct IBF_Key
-get_ibf_key (const struct GNUNET_HashCode *src,
- uint16_t salt)
+get_ibf_key (const struct GNUNET_HashCode *src)
{
struct IBF_Key key;
+ uint16_t salt = 0;
- GNUNET_CRYPTO_hkdf (&key, sizeof (key),
- GCRY_MD_SHA512, GCRY_MD_SHA256,
- src, sizeof *src,
- &salt, sizeof (salt),
- NULL, 0);
+ GNUNET_CRYPTO_kdf (&key, sizeof (key),
+ src, sizeof *src,
+ &salt, sizeof (salt),
+ NULL, 0);
return key;
}
/**
- * Iterator to create the mapping between ibf keys
- * and element entries.
+ * Iterator over the mapping from IBF keys to element entries. Checks if we
+ * have an element with a given GNUNET_HashCode.
*
* @param cls closure
* @param key current key code
* @param value value in the hash map
- * @return #GNUNET_YES if we should continue to iterate,
- * #GNUNET_NO if not.
- */
-static int
-op_register_element_iterator (void *cls,
- uint32_t key,
- void *value)
-{
- struct KeyEntry *const new_k = cls;
- struct KeyEntry *old_k = value;
-
- GNUNET_assert (NULL != old_k);
- /* check if our ibf key collides with the ibf key in the existing entry */
- if (old_k->ibf_key.key_val == new_k->ibf_key.key_val)
- {
- /* insert the the new key in the collision chain */
- new_k->next_colliding = old_k->next_colliding;
- old_k->next_colliding = new_k;
- /* signal to the caller that we were able to insert into a colliding bucket */
- return GNUNET_NO;
- }
- return GNUNET_YES;
-}
-
-
-/**
- * Iterator to create the mapping between ibf keys
- * and element entries.
- *
- * @param cls closure
- * @param key current key code
- * @param value value in the hash map
- * @return #GNUNET_YES (we should continue to iterate)
+ * @return #GNUNET_YES if we should search further,
+ * #GNUNET_NO if we've found the element.
*/
static int
op_has_element_iterator (void *cls,
struct KeyEntry *k = value;
GNUNET_assert (NULL != k);
- while (NULL != k)
- {
- if (0 == GNUNET_CRYPTO_hash_cmp (&k->element->element_hash,
- element_hash))
- return GNUNET_NO;
- k = k->next_colliding;
- }
+ if (0 == GNUNET_CRYPTO_hash_cmp (&k->element->element_hash,
+ element_hash))
+ return GNUNET_NO;
return GNUNET_YES;
}
int ret;
struct IBF_Key ibf_key;
- ibf_key = get_ibf_key (element_hash, op->spec->salt);
+ ibf_key = get_ibf_key (element_hash);
ret = GNUNET_CONTAINER_multihashmap32_get_multiple (op->state->key_to_element,
(uint32_t) ibf_key.key_val,
op_has_element_iterator,
* This is done to speed up re-tried operations, if some elements
* were transmitted, and then the IBF fails to decode.
*
+ * XXX: clarify ownership, doesn't sound right.
+ *
* @param op the union operation
* @param ee the element entry
*/
op_register_element (struct Operation *op,
struct ElementEntry *ee)
{
- int ret;
struct IBF_Key ibf_key;
struct KeyEntry *k;
- ibf_key = get_ibf_key (&ee->element_hash, op->spec->salt);
+ ibf_key = get_ibf_key (&ee->element_hash);
k = GNUNET_new (struct KeyEntry);
k->element = ee;
k->ibf_key = ibf_key;
- ret = GNUNET_CONTAINER_multihashmap32_get_multiple (op->state->key_to_element,
+ GNUNET_assert (GNUNET_OK ==
+ GNUNET_CONTAINER_multihashmap32_put (op->state->key_to_element,
(uint32_t) ibf_key.key_val,
- op_register_element_iterator,
- k);
+ k,
+ GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE));
+}
- /* was the element inserted into a colliding bucket? */
- if (GNUNET_SYSERR == ret)
- return;
- GNUNET_CONTAINER_multihashmap32_put (op->state->key_to_element,
- (uint32_t) ibf_key.key_val,
- k,
- GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE);
+
+static void
+salt_key (const struct IBF_Key *k_in,
+ uint32_t salt,
+ struct IBF_Key *k_out)
+{
+ int s = salt % 64;
+ uint64_t x = k_in->key_val;
+ /* rotate ibf key */
+ x = (x >> s) | (x << (64 - s));
+ k_out->key_val = x;
+}
+
+
+static void
+unsalt_key (const struct IBF_Key *k_in,
+ uint32_t salt,
+ struct IBF_Key *k_out)
+{
+ int s = salt % 64;
+ uint64_t x = k_in->key_val;
+ x = (x << s) | (x >> (64 - s));
+ k_out->key_val = x;
}
uint32_t key,
void *value)
{
- struct InvertibleBloomFilter *ibf = cls;
+ struct Operation *op = cls;
struct KeyEntry *ke = value;
-
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "inserting %x into ibf\n",
- ke->ibf_key.key_val);
- ibf_insert (ibf, ke->ibf_key);
+ struct IBF_Key salted_key;
+
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "[OP %x] inserting %lx (hash %s) into ibf\n",
+ (void *) op,
+ (unsigned long) ke->ibf_key.key_val,
+ GNUNET_h2s (&ke->element->element_hash));
+ salt_key (&ke->ibf_key, op->state->salt_send, &salted_key);
+ ibf_insert (op->state->local_ibf, salted_key);
return GNUNET_YES;
}
*
* @param op the union operation
* @param size size of the ibf to create
+ * @return #GNUNET_OK on success, #GNUNET_SYSERR on failure
*/
-static void
+static int
prepare_ibf (struct Operation *op,
- uint16_t size)
+ uint32_t size)
{
if (NULL == op->state->key_to_element)
{
if (NULL != op->state->local_ibf)
ibf_destroy (op->state->local_ibf);
op->state->local_ibf = ibf_create (size, SE_IBF_HASH_NUM);
+ if (NULL == op->state->local_ibf)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to allocate local IBF\n");
+ return GNUNET_SYSERR;
+ }
GNUNET_CONTAINER_multihashmap32_iterate (op->state->key_to_element,
&prepare_ibf_iterator,
- op->state->local_ibf);
+ op);
+ return GNUNET_OK;
}
/**
* Send an ibf of appropriate size.
*
+ * Fragments the IBF into multiple messages if necessary.
+ *
* @param op the union operation
* @param ibf_order order of the ibf to send, size=2^order
+ * @return #GNUNET_OK on success, #GNUNET_SYSERR on failure
*/
-static void
+static int
send_ibf (struct Operation *op,
uint16_t ibf_order)
{
unsigned int buckets_sent = 0;
struct InvertibleBloomFilter *ibf;
- prepare_ibf (op, 1<<ibf_order);
+ if (GNUNET_OK !=
+ prepare_ibf (op, 1<<ibf_order))
+ {
+ /* allocation failed */
+ return GNUNET_SYSERR;
+ }
+
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sending ibf of size %u\n",
+ 1<<ibf_order);
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending ibf of size %u\n",
- 1<<ibf_order);
+ {
+ char name[64] = { 0 };
+ snprintf (name, sizeof (name), "# sent IBF (order %u)", ibf_order);
+ GNUNET_STATISTICS_update (_GSS_statistics, name, 1, GNUNET_NO);
+ }
ibf = op->state->local_ibf;
ev = GNUNET_MQ_msg_extra (msg,
buckets_in_message * IBF_BUCKET_SIZE,
GNUNET_MESSAGE_TYPE_SET_UNION_P2P_IBF);
- msg->reserved = 0;
+ msg->reserved1 = 0;
+ msg->reserved2 = 0;
msg->order = ibf_order;
- msg->offset = htons (buckets_sent);
+ msg->offset = htonl (buckets_sent);
+ msg->salt = htonl (op->state->salt_send);
ibf_write_slice (ibf, buckets_sent,
buckets_in_message, &msg[1]);
buckets_sent += buckets_in_message;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "ibf chunk size %u, %u/%u sent\n",
- buckets_in_message,
- buckets_sent,
- 1<<ibf_order);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "ibf chunk size %u, %u/%u sent\n",
+ buckets_in_message,
+ buckets_sent,
+ 1<<ibf_order);
GNUNET_MQ_send (op->mq, ev);
}
- op->state->phase = PHASE_EXPECT_ELEMENTS_AND_REQUESTS;
+ /* The other peer must decode the IBF, so
+ * we're passive. */
+ op->state->phase = PHASE_INVENTORY_PASSIVE;
+ return GNUNET_OK;
}
static void
send_strata_estimator (struct Operation *op)
{
+ const struct StrataEstimator *se = op->state->se;
struct GNUNET_MQ_Envelope *ev;
struct GNUNET_MessageHeader *strata_msg;
-
+ char *buf;
+ size_t len;
+ uint16_t type;
+
+ buf = GNUNET_malloc (se->strata_count * IBF_BUCKET_SIZE * se->ibf_size);
+ len = strata_estimator_write (op->state->se,
+ buf);
+ if (len < se->strata_count * IBF_BUCKET_SIZE * se->ibf_size)
+ type = GNUNET_MESSAGE_TYPE_SET_UNION_P2P_SEC;
+ else
+ type = GNUNET_MESSAGE_TYPE_SET_UNION_P2P_SE;
ev = GNUNET_MQ_msg_header_extra (strata_msg,
- SE_STRATA_COUNT * IBF_BUCKET_SIZE * SE_IBF_SIZE,
- GNUNET_MESSAGE_TYPE_SET_UNION_P2P_SE);
- strata_estimator_write (op->state->se, &strata_msg[1]);
+ len,
+ type);
+ GNUNET_memcpy (&strata_msg[1],
+ buf,
+ len);
+ GNUNET_free (buf);
GNUNET_MQ_send (op->mq,
ev);
op->state->phase = PHASE_EXPECT_IBF;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sent SE, expecting IBF\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sent SE, expecting IBF\n");
}
*
* @param cls the union operation
* @param mh the message
+ * @param is_compressed #GNUNET_YES if the estimator is compressed
* @return #GNUNET_SYSERR if the tunnel should be disconnected,
* #GNUNET_OK otherwise
*/
static int
handle_p2p_strata_estimator (void *cls,
- const struct GNUNET_MessageHeader *mh)
+ const struct GNUNET_MessageHeader *mh,
+ int is_compressed)
{
struct Operation *op = cls;
struct StrataEstimator *remote_se;
int diff;
+ size_t len;
+
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# bytes of SE received",
+ ntohs (mh->size),
+ GNUNET_NO);
if (op->state->phase != PHASE_EXPECT_SE)
{
GNUNET_break (0);
return GNUNET_SYSERR;
}
- if (ntohs (mh->size) !=
- SE_STRATA_COUNT * SE_IBF_SIZE * IBF_BUCKET_SIZE +
- sizeof (struct GNUNET_MessageHeader))
+ len = ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader);
+ if ( (GNUNET_NO == is_compressed) &&
+ (len != SE_STRATA_COUNT * SE_IBF_SIZE * IBF_BUCKET_SIZE) )
{
fail_union_operation (op);
GNUNET_break (0);
remote_se = strata_estimator_create (SE_STRATA_COUNT,
SE_IBF_SIZE,
SE_IBF_HASH_NUM);
- strata_estimator_read (&mh[1], remote_se);
+ if (NULL == remote_se)
+ {
+ /* insufficient resources, fail */
+ fail_union_operation (op);
+ return GNUNET_SYSERR;
+ }
+ if (GNUNET_OK !=
+ strata_estimator_read (&mh[1],
+ len,
+ is_compressed,
+ remote_se))
+ {
+ /* decompression failed */
+ fail_union_operation (op);
+ strata_estimator_destroy (remote_se);
+ return GNUNET_SYSERR;
+ }
GNUNET_assert (NULL != op->state->se);
diff = strata_estimator_difference (remote_se,
op->state->se);
strata_estimator_destroy (remote_se);
strata_estimator_destroy (op->state->se);
op->state->se = NULL;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "got se diff=%d, using ibf size %d\n",
- diff,
- 1<<get_order_from_difference (diff));
- send_ibf (op,
- get_order_from_difference (diff));
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "got se diff=%d, using ibf size %d\n",
+ diff,
+ 1<<get_order_from_difference (diff));
+ if (GNUNET_OK !=
+ send_ibf (op,
+ get_order_from_difference (diff)))
+ {
+ /* Internal error, best we can do is shut the connection */
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to send IBF, closing connection\n");
+ fail_union_operation (op);
+ return GNUNET_SYSERR;
+ }
return GNUNET_OK;
}
* @param value the key entry
*/
static int
-send_element_iterator (void *cls,
- uint32_t key,
- void *value)
+send_offers_iterator (void *cls,
+ uint32_t key,
+ void *value)
{
struct SendElementClosure *sec = cls;
- struct IBF_Key ibf_key = sec->ibf_key;
struct Operation *op = sec->op;
struct KeyEntry *ke = value;
+ struct GNUNET_MQ_Envelope *ev;
+ struct GNUNET_MessageHeader *mh;
- if (ke->ibf_key.key_val != ibf_key.key_val)
+ /* Detect 32-bit key collision for the 64-bit IBF keys. */
+ if (ke->ibf_key.key_val != sec->ibf_key.key_val)
return GNUNET_YES;
- while (NULL != ke)
- {
- const struct GNUNET_SET_Element *const element = &ke->element->element;
- struct GNUNET_MQ_Envelope *ev;
- struct GNUNET_MessageHeader *mh;
- GNUNET_assert (ke->ibf_key.key_val == ibf_key.key_val);
- ev = GNUNET_MQ_msg_header_extra (mh,
- element->size,
- GNUNET_MESSAGE_TYPE_SET_P2P_ELEMENTS);
- if (NULL == ev)
- {
- /* element too large */
- GNUNET_break (0);
- continue;
- }
- memcpy (&mh[1],
- element->data,
- element->size);
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending element (%s) to peer\n",
- GNUNET_h2s (&ke->element->element_hash));
- GNUNET_MQ_send (op->mq, ev);
- ke = ke->next_colliding;
- }
- return GNUNET_NO;
+ ev = GNUNET_MQ_msg_header_extra (mh,
+ sizeof (struct GNUNET_HashCode),
+ GNUNET_MESSAGE_TYPE_SET_UNION_P2P_OFFER);
+
+ GNUNET_assert (NULL != ev);
+ *(struct GNUNET_HashCode *) &mh[1] = ke->element->element_hash;
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "[OP %x] sending element offer (%s) to peer\n",
+ (void *) op,
+ GNUNET_h2s (&ke->element->element_hash));
+ GNUNET_MQ_send (op->mq, ev);
+ return GNUNET_YES;
}
/**
- * Send all elements that have the specified IBF key
- * to the remote peer of the union operation
+ * Send offers (in the form of GNUNET_Hash-es) to the remote peer for the given IBF key.
*
* @param op union operation
* @param ibf_key IBF key of interest
*/
static void
-send_elements_for_key (struct Operation *op,
- struct IBF_Key ibf_key)
+send_offers_for_key (struct Operation *op,
+ struct IBF_Key ibf_key)
{
struct SendElementClosure send_cls;
send_cls.op = op;
(void) GNUNET_CONTAINER_multihashmap32_get_multiple (op->state->key_to_element,
(uint32_t) ibf_key.key_val,
- &send_element_iterator,
+ &send_offers_iterator,
&send_cls);
}
/**
* Decode which elements are missing on each side, and
- * send the appropriate elemens and requests
+ * send the appropriate offers and inquiries.
*
* @param op union operation
+ * @return #GNUNET_OK on success, #GNUNET_SYSERR on failure
*/
-static void
+static int
decode_and_send (struct Operation *op)
{
struct IBF_Key key;
unsigned int num_decoded;
struct InvertibleBloomFilter *diff_ibf;
- GNUNET_assert (PHASE_EXPECT_ELEMENTS == op->state->phase);
+ GNUNET_assert (PHASE_INVENTORY_ACTIVE == op->state->phase);
- prepare_ibf (op, op->state->remote_ibf->size);
+ if (GNUNET_OK !=
+ prepare_ibf (op, op->state->remote_ibf->size))
+ {
+ GNUNET_break (0);
+ /* allocation failed */
+ return GNUNET_SYSERR;
+ }
diff_ibf = ibf_dup (op->state->local_ibf);
ibf_subtract (diff_ibf, op->state->remote_ibf);
ibf_destroy (op->state->remote_ibf);
op->state->remote_ibf = NULL;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "decoding IBF (size=%u)\n",
- diff_ibf->size);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "decoding IBF (size=%u)\n",
+ diff_ibf->size);
num_decoded = 0;
- last_key.key_val = 0;
+ key.key_val = 0; /* just to avoid compiler thinking we use undef'ed variable */
while (1)
{
res = ibf_decode (diff_ibf, &side, &key);
if (res == GNUNET_OK)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "decoded ibf key %lx\n",
- key.key_val);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "decoded ibf key %lx\n",
+ (unsigned long) key.key_val);
num_decoded += 1;
if ( (num_decoded > diff_ibf->size) ||
- (num_decoded > 1 && last_key.key_val == key.key_val) )
+ ( (num_decoded > 1) &&
+ (last_key.key_val == key.key_val) ) )
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "detected cyclic ibf (decoded %u/%u)\n",
- num_decoded,
- diff_ibf->size);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "detected cyclic ibf (decoded %u/%u)\n",
+ num_decoded,
+ diff_ibf->size);
cycle_detected = GNUNET_YES;
}
}
next_order++;
if (next_order <= MAX_IBF_ORDER)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "decoding failed, sending larger ibf (size %u)\n",
- 1<<next_order);
- send_ibf (op, next_order);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "decoding failed, sending larger ibf (size %u)\n",
+ 1<<next_order);
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of IBF retries",
+ 1,
+ GNUNET_NO);
+ op->state->salt_send++;
+ if (GNUNET_OK !=
+ send_ibf (op, next_order))
+ {
+ /* Internal error, best we can do is shut the connection */
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to send IBF, closing connection\n");
+ fail_union_operation (op);
+ ibf_destroy (diff_ibf);
+ return GNUNET_SYSERR;
+ }
}
else
{
- GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
- "set union failed: reached ibf limit\n");
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of failed union operations (too large)",
+ 1,
+ GNUNET_NO);
+ // XXX: Send the whole set, element-by-element
+ LOG (GNUNET_ERROR_TYPE_ERROR,
+ "set union failed: reached ibf limit\n");
+ fail_union_operation (op);
+ ibf_destroy (diff_ibf);
+ return GNUNET_SYSERR;
}
break;
}
{
struct GNUNET_MQ_Envelope *ev;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "transmitted all values, sending DONE\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "transmitted all values, sending DONE\n");
ev = GNUNET_MQ_msg_header (GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DONE);
GNUNET_MQ_send (op->mq, ev);
+ /* We now wait until we get a DONE message back
+ * and then wait for our MQ to be flushed and all our
+ * demands be delivered. */
break;
}
if (1 == side)
{
- send_elements_for_key (op, key);
+ struct IBF_Key unsalted_key;
+ unsalt_key (&key, op->state->salt_receive, &unsalted_key);
+ send_offers_for_key (op, unsalted_key);
}
else if (-1 == side)
{
struct GNUNET_MQ_Envelope *ev;
- struct GNUNET_MessageHeader *msg;
+ struct InquiryMessage *msg;
- /* It may be nice to merge multiple requests, but with cadet's corking it is not worth
+ /* It may be nice to merge multiple requests, but with CADET's corking it is not worth
* the effort additional complexity. */
- ev = GNUNET_MQ_msg_header_extra (msg,
- sizeof (struct IBF_Key),
- GNUNET_MESSAGE_TYPE_SET_P2P_ELEMENT_REQUESTS);
-
- memcpy (&msg[1],
+ ev = GNUNET_MQ_msg_extra (msg,
+ sizeof (struct IBF_Key),
+ GNUNET_MESSAGE_TYPE_SET_UNION_P2P_INQUIRY);
+ msg->salt = htonl (op->state->salt_receive);
+ GNUNET_memcpy (&msg[1],
&key,
sizeof (struct IBF_Key));
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending element request\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sending element inquiry for IBF key %lx\n",
+ (unsigned long) key.key_val);
GNUNET_MQ_send (op->mq, ev);
}
else
}
}
ibf_destroy (diff_ibf);
+ return GNUNET_OK;
}
/**
* Handle an IBF message from a remote peer.
*
+ * Reassemble the IBF from multiple pieces, and
+ * process the whole IBF once possible.
+ *
* @param cls the union operation
* @param mh the header of the message
* @return #GNUNET_SYSERR if the tunnel should be disconnected,
return GNUNET_SYSERR;
}
msg = (const struct IBFMessage *) mh;
- if ( (op->state->phase == PHASE_EXPECT_ELEMENTS_AND_REQUESTS) ||
+ if ( (op->state->phase == PHASE_INVENTORY_PASSIVE) ||
(op->state->phase == PHASE_EXPECT_IBF) )
{
op->state->phase = PHASE_EXPECT_IBF_CONT;
GNUNET_assert (NULL == op->state->remote_ibf);
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Creating new ibf of size %u\n",
- 1 << msg->order);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "Creating new ibf of size %u\n",
+ 1 << msg->order);
op->state->remote_ibf = ibf_create (1<<msg->order, SE_IBF_HASH_NUM);
+ op->state->salt_receive = ntohl (msg->salt);
+ LOG (GNUNET_ERROR_TYPE_DEBUG, "Receiving new IBF with salt %u\n", op->state->salt_receive);
+ if (NULL == op->state->remote_ibf)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to parse remote IBF, closing connection\n");
+ fail_union_operation (op);
+ return GNUNET_SYSERR;
+ }
op->state->ibf_buckets_received = 0;
- if (0 != ntohs (msg->offset))
+ if (0 != ntohl (msg->offset))
{
GNUNET_break_op (0);
fail_union_operation (op);
}
else if (op->state->phase == PHASE_EXPECT_IBF_CONT)
{
- if ( (ntohs (msg->offset) != op->state->ibf_buckets_received) ||
- (1<<msg->order != op->state->remote_ibf->size) )
+ if (ntohl (msg->offset) != op->state->ibf_buckets_received)
+ {
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return GNUNET_SYSERR;
+ }
+ if (1<<msg->order != op->state->remote_ibf->size)
+ {
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return GNUNET_SYSERR;
+ }
+ if (ntohl (msg->salt) != op->state->salt_receive)
{
GNUNET_break_op (0);
fail_union_operation (op);
return GNUNET_SYSERR;
}
}
+ else
+ {
+ GNUNET_assert (0);
+ }
buckets_in_message = (ntohs (msg->header.size) - sizeof *msg) / IBF_BUCKET_SIZE;
return GNUNET_SYSERR;
}
+ GNUNET_assert (NULL != op->state->remote_ibf);
+
ibf_read_slice (&msg[1],
op->state->ibf_buckets_received,
buckets_in_message,
if (op->state->ibf_buckets_received == op->state->remote_ibf->size)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "received full ibf\n");
- op->state->phase = PHASE_EXPECT_ELEMENTS;
- decode_and_send (op);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "received full ibf\n");
+ op->state->phase = PHASE_INVENTORY_ACTIVE;
+ if (GNUNET_OK !=
+ decode_and_send (op))
+ {
+ /* Internal error, best we can do is shut down */
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to decode IBF, closing connection\n");
+ return GNUNET_SYSERR;
+ }
}
return GNUNET_OK;
}
*
* @param op union operation
* @param element element to send
+ * @param status status to send with the new element
*/
static void
send_client_element (struct Operation *op,
- struct GNUNET_SET_Element *element)
+ struct GNUNET_SET_Element *element,
+ int status)
{
struct GNUNET_MQ_Envelope *ev;
struct GNUNET_SET_ResultMessage *rm;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending element (size %u) to client\n",
- element->size);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sending element (size %u) to client\n",
+ element->size);
GNUNET_assert (0 != op->spec->client_request_id);
ev = GNUNET_MQ_msg_extra (rm, element->size, GNUNET_MESSAGE_TYPE_SET_RESULT);
if (NULL == ev)
GNUNET_break (0);
return;
}
- rm->result_status = htons (GNUNET_SET_STATUS_OK);
+ rm->result_status = htons (status);
rm->request_id = htonl (op->spec->client_request_id);
rm->element_type = element->element_type;
- memcpy (&rm[1], element->data, element->size);
+ GNUNET_memcpy (&rm[1], element->data, element->size);
GNUNET_MQ_send (op->spec->set->client_mq, ev);
}
rm->result_status = htons (GNUNET_SET_STATUS_DONE);
rm->element_type = htons (0);
GNUNET_MQ_send (op->spec->set->client_mq, ev);
+ /* Will also call the union-specific cancel function. */
_GSS_operation_destroy (op, GNUNET_YES);
- op->keep--;
- if (0 == op->keep)
- GNUNET_free (op);
}
-/**
- * Send all remaining elements in the full result iterator.
- *
- * @param cls operation
- */
static void
-send_remaining_elements (void *cls)
+maybe_finish (struct Operation *op)
{
- struct Operation *op = cls;
- struct KeyEntry *ke;
- int res;
+ unsigned int num_demanded;
- res = GNUNET_CONTAINER_multihashmap32_iterator_next (op->state->full_result_iter,
- NULL,
- (const void **) &ke);
- if (GNUNET_NO == res)
- {
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending done and destroy because iterator ran out\n");
- send_done_and_destroy (op);
- return;
- }
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending elements from key entry\n");
- while (1)
+ num_demanded = GNUNET_CONTAINER_multihashmap_size (op->state->demanded_hashes);
+
+ if (PHASE_FINISH_WAITING == op->state->phase)
{
- struct GNUNET_MQ_Envelope *ev;
- struct GNUNET_SET_ResultMessage *rm;
- struct GNUNET_SET_Element *element;
-
- element = &ke->element->element;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending element (size %u) to client (full set)\n",
- element->size);
- GNUNET_assert (0 != op->spec->client_request_id);
- ev = GNUNET_MQ_msg_extra (rm,
- element->size,
- GNUNET_MESSAGE_TYPE_SET_RESULT);
- if (NULL == ev)
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "In PHASE_FINISH_WAITING, pending %u demands\n",
+ num_demanded);
+ if (0 == num_demanded)
{
- GNUNET_MQ_discard (ev);
- GNUNET_break (0);
- continue;
- }
- rm->result_status = htons (GNUNET_SET_STATUS_OK);
- rm->request_id = htonl (op->spec->client_request_id);
- rm->element_type = element->element_type;
- memcpy (&rm[1], element->data, element->size);
- if (NULL == ke->next_colliding)
- {
- GNUNET_MQ_notify_sent (ev, send_remaining_elements, op);
- GNUNET_MQ_send (op->spec->set->client_mq, ev);
- break;
- }
- GNUNET_MQ_send (op->spec->set->client_mq, ev);
- ke = ke->next_colliding;
- }
-}
+ struct GNUNET_MQ_Envelope *ev;
+ op->state->phase = PHASE_DONE;
+ ev = GNUNET_MQ_msg_header (GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DONE);
+ GNUNET_MQ_send (op->mq, ev);
-/**
- * Send a result message to the client indicating
- * that the operation is over.
- * After the result done message has been sent to the client,
- * destroy the evaluate operation.
- *
- * @param op union operation
- */
-static void
-finish_and_destroy (struct Operation *op)
-{
- GNUNET_assert (GNUNET_NO == op->state->client_done_sent);
- op->keep++;
- if (GNUNET_SET_RESULT_FULL == op->spec->result_mode)
+ /* We now wait until the other peer closes the channel
+ * after it got all elements from us. */
+ }
+ }
+ if (PHASE_FINISH_CLOSING == op->state->phase)
{
- /* prevent that the op is free'd by the tunnel end handler */
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sending full result set\n");
- GNUNET_assert (NULL == op->state->full_result_iter);
- op->state->full_result_iter =
- GNUNET_CONTAINER_multihashmap32_iterator_create (op->state->key_to_element);
- send_remaining_elements (op);
- return;
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "In PHASE_FINISH_CLOSING, pending %u demands\n",
+ num_demanded);
+ if (0 == num_demanded)
+ {
+ op->state->phase = PHASE_DONE;
+ send_done_and_destroy (op);
+ }
}
- send_done_and_destroy (op);
}
{
struct Operation *op = cls;
struct ElementEntry *ee;
+ const struct GNUNET_SET_ElementMessage *emsg;
uint16_t element_size;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Got element from peer\n");
- if ( (op->state->phase != PHASE_EXPECT_ELEMENTS) &&
- (op->state->phase != PHASE_EXPECT_ELEMENTS_AND_REQUESTS) )
+ if (0 == GNUNET_CONTAINER_multihashmap_size (op->state->demanded_hashes))
{
+ GNUNET_break_op (0);
fail_union_operation (op);
+ return;
+ }
+ if (ntohs (mh->size) < sizeof (struct GNUNET_SET_ElementMessage))
+ {
GNUNET_break_op (0);
+ fail_union_operation (op);
return;
}
- element_size = ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader);
+
+ emsg = (const struct GNUNET_SET_ElementMessage *) mh;
+
+ element_size = ntohs (mh->size) - sizeof (struct GNUNET_SET_ElementMessage);
ee = GNUNET_malloc (sizeof (struct ElementEntry) + element_size);
- memcpy (&ee[1], &mh[1], element_size);
+ GNUNET_memcpy (&ee[1], &emsg[1], element_size);
ee->element.size = element_size;
ee->element.data = &ee[1];
+ ee->element.element_type = ntohs (emsg->element_type);
ee->remote = GNUNET_YES;
- GNUNET_CRYPTO_hash (ee->element.data,
- ee->element.size,
- &ee->element_hash);
+ GNUNET_SET_element_hash (&ee->element, &ee->element_hash);
- if (GNUNET_YES == op_has_element (op, &ee->element_hash))
+ if (GNUNET_NO ==
+ GNUNET_CONTAINER_multihashmap_remove (op->state->demanded_hashes,
+ &ee->element_hash,
+ NULL))
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "got existing element from peer\n");
+ /* We got something we didn't demand, since it's not in our map. */
+ GNUNET_break_op (0);
GNUNET_free (ee);
+ fail_union_operation (op);
return;
}
- op_register_element (op, ee);
- /* only send results immediately if the client wants it */
- if (GNUNET_SET_RESULT_ADDED == op->spec->result_mode)
- send_client_element (op, &ee->element);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "Got element (size %u, hash %s) from peer\n",
+ (unsigned int) element_size,
+ GNUNET_h2s (&ee->element_hash));
+
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# received elements",
+ 1,
+ GNUNET_NO);
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# exchanged elements",
+ 1,
+ GNUNET_NO);
+
+ if (GNUNET_YES == op_has_element (op, &ee->element_hash))
+ {
+ /* Got repeated element. Should not happen since
+ * we track demands. */
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# repeated elements",
+ 1,
+ GNUNET_NO);
+ GNUNET_free (ee);
+ }
+ else
+ {
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "Registering new element from remote peer\n");
+ op_register_element (op, ee);
+ /* only send results immediately if the client wants it */
+ switch (op->spec->result_mode)
+ {
+ case GNUNET_SET_RESULT_ADDED:
+ send_client_element (op, &ee->element, GNUNET_SET_STATUS_OK);
+ break;
+ case GNUNET_SET_RESULT_SYMMETRIC:
+ send_client_element (op, &ee->element, GNUNET_SET_STATUS_ADD_LOCAL);
+ break;
+ default:
+ /* Result mode not supported, should have been caught earlier. */
+ GNUNET_break (0);
+ break;
+ }
+ }
+
+ maybe_finish (op);
}
/**
- * Handle an element request from a remote peer.
+ * Send offers (for GNUNET_Hash-es) in response
+ * to inquiries (for IBF_Key-s).
*
* @param cls the union operation
* @param mh the message
*/
static void
-handle_p2p_element_requests (void *cls,
- const struct GNUNET_MessageHeader *mh)
+handle_p2p_inquiry (void *cls,
+ const struct GNUNET_MessageHeader *mh)
{
struct Operation *op = cls;
const struct IBF_Key *ibf_key;
unsigned int num_keys;
+ struct InquiryMessage *msg;
/* look up elements and send them */
- if (op->state->phase != PHASE_EXPECT_ELEMENTS_AND_REQUESTS)
+ if (op->state->phase != PHASE_INVENTORY_PASSIVE)
{
GNUNET_break_op (0);
fail_union_operation (op);
return;
}
- num_keys = (ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
- / sizeof (struct IBF_Key);
- if ((ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
+ num_keys = (ntohs (mh->size) - sizeof (struct InquiryMessage))
+ / sizeof (struct IBF_Key);
+ if ((ntohs (mh->size) - sizeof (struct InquiryMessage))
!= num_keys * sizeof (struct IBF_Key))
{
GNUNET_break_op (0);
return;
}
- ibf_key = (const struct IBF_Key *) &mh[1];
+ msg = (struct InquiryMessage *) mh;
+
+ ibf_key = (const struct IBF_Key *) &msg[1];
while (0 != num_keys--)
{
- send_elements_for_key (op, *ibf_key);
+ struct IBF_Key unsalted_key;
+ unsalt_key (ibf_key, ntohl (msg->salt), &unsalted_key);
+ send_offers_for_key (op, unsalted_key);
ibf_key++;
}
}
+/**
+ * FIXME
+ */
+static void
+handle_p2p_demand (void *cls,
+ const struct GNUNET_MessageHeader *mh)
+{
+ struct Operation *op = cls;
+ struct ElementEntry *ee;
+ struct GNUNET_SET_ElementMessage *emsg;
+ const struct GNUNET_HashCode *hash;
+ unsigned int num_hashes;
+ struct GNUNET_MQ_Envelope *ev;
+
+ num_hashes = (ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
+ / sizeof (struct GNUNET_HashCode);
+ if ((ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
+ != num_hashes * sizeof (struct GNUNET_HashCode))
+ {
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return;
+ }
+
+ for (hash = (const struct GNUNET_HashCode *) &mh[1];
+ num_hashes > 0;
+ hash++, num_hashes--)
+ {
+ ee = GNUNET_CONTAINER_multihashmap_get (op->spec->set->content->elements, hash);
+ if (NULL == ee)
+ {
+ /* Demand for non-existing element. */
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return;
+ }
+ if (GNUNET_NO == _GSS_is_element_of_operation (ee, op))
+ {
+ /* Probably confused lazily copied sets. */
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return;
+ }
+ ev = GNUNET_MQ_msg_extra (emsg, ee->element.size, GNUNET_MESSAGE_TYPE_SET_P2P_ELEMENTS);
+ GNUNET_memcpy (&emsg[1], ee->element.data, ee->element.size);
+ emsg->reserved = htons (0);
+ emsg->element_type = htons (ee->element.element_type);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "[OP %x] Sending demanded element (size %u, hash %s) to peer\n",
+ (void *) op,
+ (unsigned int) ee->element.size,
+ GNUNET_h2s (&ee->element_hash));
+ GNUNET_MQ_send (op->mq, ev);
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# exchanged elements",
+ 1,
+ GNUNET_NO);
+
+ switch (op->spec->result_mode)
+ {
+ case GNUNET_SET_RESULT_ADDED:
+ /* Nothing to do. */
+ break;
+ case GNUNET_SET_RESULT_SYMMETRIC:
+ send_client_element (op, &ee->element, GNUNET_SET_STATUS_ADD_REMOTE);
+ break;
+ default:
+ /* Result mode not supported, should have been caught earlier. */
+ GNUNET_break (0);
+ break;
+ }
+ }
+}
+
+
+/**
+ * Handle offers (of GNUNET_HashCode-s) and
+ * respond with demands (of GNUNET_HashCode-s).
+ *
+ * @param cls the union operation
+ * @param mh the message
+ */
+static void
+handle_p2p_offer (void *cls,
+ const struct GNUNET_MessageHeader *mh)
+{
+ struct Operation *op = cls;
+ const struct GNUNET_HashCode *hash;
+ unsigned int num_hashes;
+
+ /* look up elements and send them */
+ if ( (op->state->phase != PHASE_INVENTORY_PASSIVE) &&
+ (op->state->phase != PHASE_INVENTORY_ACTIVE))
+ {
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return;
+ }
+ num_hashes = (ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
+ / sizeof (struct GNUNET_HashCode);
+ if ((ntohs (mh->size) - sizeof (struct GNUNET_MessageHeader))
+ != num_hashes * sizeof (struct GNUNET_HashCode))
+ {
+ GNUNET_break_op (0);
+ fail_union_operation (op);
+ return;
+ }
+
+ for (hash = (const struct GNUNET_HashCode *) &mh[1];
+ num_hashes > 0;
+ hash++, num_hashes--)
+ {
+ struct ElementEntry *ee;
+ struct GNUNET_MessageHeader *demands;
+ struct GNUNET_MQ_Envelope *ev;
+
+ ee = GNUNET_CONTAINER_multihashmap_get (op->spec->set->content->elements,
+ hash);
+ if (NULL != ee)
+ if (GNUNET_YES == _GSS_is_element_of_operation (ee, op))
+ continue;
+
+ if (GNUNET_YES ==
+ GNUNET_CONTAINER_multihashmap_contains (op->state->demanded_hashes,
+ hash))
+ {
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "Skipped sending duplicate demand\n");
+ continue;
+ }
+
+ GNUNET_assert (GNUNET_OK ==
+ GNUNET_CONTAINER_multihashmap_put (op->state->demanded_hashes,
+ hash,
+ NULL,
+ GNUNET_CONTAINER_MULTIHASHMAPOPTION_UNIQUE_FAST));
+
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "[OP %x] Requesting element (hash %s)\n",
+ (void *) op, GNUNET_h2s (hash));
+ ev = GNUNET_MQ_msg_header_extra (demands,
+ sizeof (struct GNUNET_HashCode),
+ GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DEMAND);
+ *(struct GNUNET_HashCode *) &demands[1] = *hash;
+ GNUNET_MQ_send (op->mq, ev);
+ }
+}
+
+
/**
* Handle a done message from a remote peer
*
const struct GNUNET_MessageHeader *mh)
{
struct Operation *op = cls;
- struct GNUNET_MQ_Envelope *ev;
- if (op->state->phase == PHASE_EXPECT_ELEMENTS_AND_REQUESTS)
+ if (op->state->phase == PHASE_INVENTORY_PASSIVE)
{
- /* we got all requests, but still have to send our elements as response */
-
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "got DONE, sending final DONE after elements\n");
- op->state->phase = PHASE_FINISHED;
- ev = GNUNET_MQ_msg_header (GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DONE);
- GNUNET_MQ_send (op->mq, ev);
+ /* We got all requests, but still have to send our elements in response. */
+
+ op->state->phase = PHASE_FINISH_WAITING;
+
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "got DONE (as passive partner), waiting for our demands to be satisfied\n");
+ /* The active peer is done sending offers
+ * and inquiries. This means that all
+ * our responses to that (demands and offers)
+ * must be in flight (queued or in mesh).
+ *
+ * We should notify the active peer once
+ * all our demands are satisfied, so that the active
+ * peer can quit if we gave him everything.
+ */
+ maybe_finish (op);
return;
}
- if (op->state->phase == PHASE_EXPECT_ELEMENTS)
+ if (op->state->phase == PHASE_INVENTORY_ACTIVE)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "got final DONE\n");
- op->state->phase = PHASE_FINISHED;
- finish_and_destroy (op);
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "got DONE (as active partner), waiting to finish\n");
+ /* All demands of the other peer are satisfied,
+ * and we processed all offers, thus we know
+ * exactly what our demands must be.
+ *
+ * We'll close the channel
+ * to the other peer once our demands are met.
+ */
+ op->state->phase = PHASE_FINISH_CLOSING;
+ maybe_finish (op);
return;
}
GNUNET_break_op (0);
struct GNUNET_MQ_Envelope *ev;
struct OperationRequestMessage *msg;
+ GNUNET_assert (NULL == op->state);
op->state = GNUNET_new (struct OperationState);
+ op->state->demanded_hashes = GNUNET_CONTAINER_multihashmap_create (32, GNUNET_NO);
/* copy the current generation's strata estimator for this operation */
op->state->se = strata_estimator_dup (op->spec->set->state->se);
/* we started the operation, thus we have to send the operation request */
op->state->phase = PHASE_EXPECT_SE;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Initiating union operation evaluation\n");
+ op->state->salt_receive = op->state->salt_send = 42;
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "Initiating union operation evaluation\n");
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of total union operations",
+ 1,
+ GNUNET_NO);
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of initiated union operations",
+ 1,
+ GNUNET_NO);
ev = GNUNET_MQ_msg_nested_mh (msg,
GNUNET_MESSAGE_TYPE_SET_P2P_OPERATION_REQUEST,
opaque_context);
return;
}
msg->operation = htonl (GNUNET_SET_OPERATION_UNION);
- msg->app_id = op->spec->app_id;
GNUNET_MQ_send (op->mq,
ev);
if (NULL != opaque_context)
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sent op request with context message\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sent op request with context message\n");
else
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "sent op request without context message\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "sent op request without context message\n");
}
static void
union_accept (struct Operation *op)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "accepting set union operation\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "accepting set union operation\n");
+ GNUNET_assert (NULL == op->state);
+
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of accepted union operations",
+ 1,
+ GNUNET_NO);
+ GNUNET_STATISTICS_update (_GSS_statistics,
+ "# of total union operations",
+ 1,
+ GNUNET_NO);
+
op->state = GNUNET_new (struct OperationState);
op->state->se = strata_estimator_dup (op->spec->set->state->se);
+ op->state->demanded_hashes = GNUNET_CONTAINER_multihashmap_create (32, GNUNET_NO);
+ op->state->salt_receive = op->state->salt_send = 42;
/* kick off the operation */
send_strata_estimator (op);
}
* We maintain one strata estimator per set and then manipulate it over the
* lifetime of the set, as recreating a strata estimator would be expensive.
*
- * @return the newly created set
+ * @return the newly created set, NULL on error
*/
static struct SetState *
union_set_create (void)
{
struct SetState *set_state;
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "union set created\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "union set created\n");
set_state = GNUNET_new (struct SetState);
set_state->se = strata_estimator_create (SE_STRATA_COUNT,
SE_IBF_SIZE, SE_IBF_HASH_NUM);
+ if (NULL == set_state->se)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+ "Failed to allocate strata estimator\n");
+ GNUNET_free (set_state);
+ return NULL;
+ }
return set_state;
}
union_add (struct SetState *set_state, struct ElementEntry *ee)
{
strata_estimator_insert (set_state->se,
- get_ibf_key (&ee->element_hash, 0));
+ get_ibf_key (&ee->element_hash));
}
union_remove (struct SetState *set_state, struct ElementEntry *ee)
{
strata_estimator_remove (set_state->se,
- get_ibf_key (&ee->element_hash, 0));
+ get_ibf_key (&ee->element_hash));
}
union_handle_p2p_message (struct Operation *op,
const struct GNUNET_MessageHeader *mh)
{
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "received p2p message (t: %u, s: %u)\n",
- ntohs (mh->type),
- ntohs (mh->size));
+ //LOG (GNUNET_ERROR_TYPE_DEBUG,
+ // "received p2p message (t: %u, s: %u)\n",
+ // ntohs (mh->type),
+ // ntohs (mh->size));
switch (ntohs (mh->type))
{
case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_IBF:
return handle_p2p_ibf (op, mh);
case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_SE:
- return handle_p2p_strata_estimator (op, mh);
+ return handle_p2p_strata_estimator (op, mh, GNUNET_NO);
+ case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_SEC:
+ return handle_p2p_strata_estimator (op, mh, GNUNET_YES);
case GNUNET_MESSAGE_TYPE_SET_P2P_ELEMENTS:
handle_p2p_elements (op, mh);
break;
- case GNUNET_MESSAGE_TYPE_SET_P2P_ELEMENT_REQUESTS:
- handle_p2p_element_requests (op, mh);
+ case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_INQUIRY:
+ handle_p2p_inquiry (op, mh);
break;
case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DONE:
handle_p2p_done (op, mh);
break;
+ case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_OFFER:
+ handle_p2p_offer (op, mh);
+ break;
+ case GNUNET_MESSAGE_TYPE_SET_UNION_P2P_DEMAND:
+ handle_p2p_demand (op, mh);
+ break;
default:
- /* something wrong with cadet's message handlers? */
+ /* Something wrong with cadet's message handlers? */
GNUNET_assert (0);
}
return GNUNET_OK;
}
+
/**
- * handler for peer-disconnects, notifies the client
- * about the aborted operation in case the op was not concluded
+ * Handler for peer-disconnects, notifies the client
+ * about the aborted operation in case the op was not concluded.
*
* @param op the destroyed operation
*/
static void
union_peer_disconnect (struct Operation *op)
{
- if (PHASE_FINISHED != op->state->phase)
+ if (PHASE_DONE != op->state->phase)
{
struct GNUNET_MQ_Envelope *ev;
struct GNUNET_SET_ResultMessage *msg;
msg->element_type = htons (0);
GNUNET_MQ_send (op->spec->set->client_mq,
ev);
- GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
- "other peer disconnected prematurely\n");
+ LOG (GNUNET_ERROR_TYPE_WARNING,
+ "other peer disconnected prematurely, phase %u\n",
+ op->state->phase);
_GSS_operation_destroy (op,
GNUNET_YES);
return;
}
// else: the session has already been concluded
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "other peer disconnected (finished)\n");
+ LOG (GNUNET_ERROR_TYPE_DEBUG,
+ "other peer disconnected (finished)\n");
if (GNUNET_NO == op->state->client_done_sent)
- finish_and_destroy (op);
+ send_done_and_destroy (op);
+}
+
+
+/**
+ * Copy union-specific set state.
+ *
+ * @param set source set for copying the union state
+ * @return a copy of the union-specific set state
+ */
+static struct SetState *
+union_copy_state (struct Set *set)
+{
+ struct SetState *new_state;
+
+ new_state = GNUNET_new (struct SetState);
+ GNUNET_assert ( (NULL != set->state) && (NULL != set->state->se) );
+ new_state->se = strata_estimator_dup (set->state->se);
+
+ return new_state;
}
.accept = &union_accept,
.peer_disconnect = &union_peer_disconnect,
.cancel = &union_op_cancel,
+ .copy_state = &union_copy_state,
};
return &union_vt;