-fixing #2139
[oweals/gnunet.git] / src / fs / gnunet-service-fs_pr.c
index 343ec3ada5f9c9f94375d0939425afaad8236d1f..c16f94464e5fd9f54d363de1ef005b366cd0a81d 100644 (file)
  * @author Christian Grothoff
  */
 #include "platform.h"
+#include "gnunet_load_lib.h"
+#include "gnunet-service-fs.h"
+#include "gnunet-service-fs_cp.h"
+#include "gnunet-service-fs_indexing.h"
+#include "gnunet-service-fs_pe.h"
 #include "gnunet-service-fs_pr.h"
 
+/**
+ * Maximum size of the datastore queue for P2P operations.  Needs to
+ * be large enough to queue MAX_QUEUE_PER_PEER operations for roughly
+ * the number of active (connected) peers.
+ */
+#define MAX_DATASTORE_QUEUE (16 * MAX_QUEUE_PER_PEER)
+
+/**
+ * Bandwidth value of a 0-priority content (must be fairly high
+ * compared to query since content is typically significantly larger
+ * -- and more valueable since it can take many queries to get one
+ * piece of content).
+ */
+#define CONTENT_BANDWIDTH_VALUE 800
+
+/**
+ * Hard limit on the number of results we may get from the datastore per query.
+ */
+#define MAX_RESULTS (100 * 1024)
+
+/**
+ * An active request.
+ */
+struct GSF_PendingRequest
+{
+  /**
+   * Public data for the request.
+   */
+  struct GSF_PendingRequestData public_data;
+
+  /**
+   * Function to call if we encounter a reply.
+   */
+  GSF_PendingRequestReplyHandler rh;
+
+  /**
+   * Closure for 'rh'
+   */
+  void *rh_cls;
+
+  /**
+   * Array of hash codes of replies we've already seen.
+   */
+  GNUNET_HashCode *replies_seen;
+
+  /**
+   * Bloomfilter masking replies we've already seen.
+   */
+  struct GNUNET_CONTAINER_BloomFilter *bf;
+
+  /**
+   * Entry for this pending request in the expiration heap, or NULL.
+   */
+  struct GNUNET_CONTAINER_HeapNode *hnode;
+
+  /**
+   * Datastore queue entry for this request (or NULL for none).
+   */
+  struct GNUNET_DATASTORE_QueueEntry *qe;
+
+  /**
+   * DHT request handle for this request (or NULL for none).
+   */
+  struct GNUNET_DHT_GetHandle *gh;
+
+  /**
+   * Function to call upon completion of the local get
+   * request, or NULL for none.
+   */
+  GSF_LocalLookupContinuation llc_cont;
+
+  /**
+   * Closure for llc_cont.
+   */
+  void *llc_cont_cls;
+
+  /**
+   * Last result from the local datastore lookup evaluation.
+   */
+  enum GNUNET_BLOCK_EvaluationResult local_result;
+
+  /**
+   * Identity of the peer that we should use for the 'sender'
+   * (recipient of the response) when forwarding (0 for none).
+   */
+  GNUNET_PEER_Id sender_pid;
+
+  /**
+   * Identity of the peer that we should never forward this query
+   * to since it originated this query (0 for none).
+   */
+  GNUNET_PEER_Id origin_pid;
+
+  /**
+   * Time we started the last datastore lookup.
+   */
+  struct GNUNET_TIME_Absolute qe_start;
+
+  /**
+   * Task that warns us if the local datastore lookup takes too long.
+   */
+  GNUNET_SCHEDULER_TaskIdentifier warn_task;
+
+  /**
+   * Current offset for querying our local datastore for results.
+   * Starts at a random value, incremented until we get the same
+   * UID again (detected using 'first_uid'), which is then used
+   * to termiante the iteration.
+   */
+  uint64_t local_result_offset;
+
+  /**
+   * Unique ID of the first result from the local datastore;
+   * used to detect wrap-around of the offset.
+   */
+  uint64_t first_uid;
+
+  /**
+   * Number of valid entries in the 'replies_seen' array.
+   */
+  unsigned int replies_seen_count;
+
+  /**
+   * Length of the 'replies_seen' array.
+   */
+  unsigned int replies_seen_size;
+
+  /**
+   * Mingle value we currently use for the bf.
+   */
+  uint32_t mingle;
+
+  /**
+   * Do we have a first UID yet?
+   */
+  unsigned int have_first_uid;
+
+};
+
+
+/**
+ * All pending requests, ordered by the query.  Entries
+ * are of type 'struct GSF_PendingRequest*'.
+ */
+static struct GNUNET_CONTAINER_MultiHashMap *pr_map;
+
+
+/**
+ * Datastore 'PUT' load tracking.
+ */
+static struct GNUNET_LOAD_Value *datastore_put_load;
+
+
+/**
+ * Are we allowed to migrate content to this peer.
+ */
+static int active_to_migration;
+
+
+/**
+ * Size of the datastore queue we assume for common requests.
+ * Determined based on the network quota.
+ */
+static unsigned int datastore_queue_size;
+
+/**
+ * Heap with the request that will expire next at the top.  Contains
+ * pointers of type "struct PendingRequest*"; these will *also* be
+ * aliased from the "requests_by_peer" data structures and the
+ * "requests_by_query" table.  Note that requests from our clients
+ * don't expire and are thus NOT in the "requests_by_expiration"
+ * (or the "requests_by_peer" tables).
+ */
+static struct GNUNET_CONTAINER_Heap *requests_by_expiration_heap;
+
 
 /**
- * Create a new pending request.  
+ * Maximum number of requests (from other peers, overall) that we're
+ * willing to have pending at any given point in time.  Can be changed
+ * via the configuration file (32k is just the default).
+ */
+static unsigned long long max_pending_requests = (32 * 1024);
+
+
+
+/**
+ * Recalculate our bloom filter for filtering replies.  This function
+ * will create a new bloom filter from scratch, so it should only be
+ * called if we have no bloomfilter at all (and hence can create a
+ * fresh one of minimal size without problems) OR if our peer is the
+ * initiator (in which case we may resize to larger than mimimum size).
+ *
+ * @param pr request for which the BF is to be recomputed
+ */
+static void
+refresh_bloomfilter (struct GSF_PendingRequest *pr)
+{
+  if (pr->bf != NULL)
+    GNUNET_CONTAINER_bloomfilter_free (pr->bf);
+  pr->mingle =
+      GNUNET_CRYPTO_random_u32 (GNUNET_CRYPTO_QUALITY_WEAK, UINT32_MAX);
+  pr->bf =
+      GNUNET_BLOCK_construct_bloomfilter (pr->mingle, pr->replies_seen,
+                                          pr->replies_seen_count);
+}
+
+
+/**
+ * Create a new pending request.
  *
  * @param options request options
  * @param type type of the block that is being requested
  * @param query key for the lookup
  * @param namespace namespace to lookup, NULL for no namespace
  * @param target preferred target for the request, NULL for none
- * @param bf bloom filter for known replies, can be NULL
+ * @param bf_data raw data for bloom filter for known replies, can be NULL
+ * @param bf_size number of bytes in bf_data
  * @param mingle mingle value for bf
  * @param anonymity_level desired anonymity level
  * @param priority maximum outgoing cummulative request priority to use
+ * @param ttl current time-to-live for the request
+ * @param sender_pid peer ID to use for the sender when forwarding, 0 for none
+ * @param origin_pid peer ID of origin of query (do not loop back)
  * @param replies_seen hash codes of known local replies
  * @param replies_seen_count size of the 'replies_seen' array
  * @param rh handle to call when we get a reply
  */
 struct GSF_PendingRequest *
 GSF_pending_request_create_ (enum GSF_PendingRequestOptions options,
-                            enum GNUNET_BLOCK_Type type,
-                            const GNUNET_HashCode *query,
-                            const GNUNET_HashCode *namespace,
-                            const struct GNUNET_PeerIdentity *target,
-                            struct GNUNET_CONTAINER_BloomFilter *bf,
-                            int32_t mingle,
-                            uint32_t anonymity_level,
-                            uint32_t priority,
-                            const GNUNET_HashCode *replies_seen,
-                            unsigned int replies_seen_count,
-                            GSF_PendingRequestReplyHandler rh,
-                            void *rh_cls)
+                             enum GNUNET_BLOCK_Type type,
+                             const GNUNET_HashCode * query,
+                             const GNUNET_HashCode * namespace,
+                             const struct GNUNET_PeerIdentity *target,
+                             const char *bf_data, size_t bf_size,
+                             uint32_t mingle, uint32_t anonymity_level,
+                             uint32_t priority, int32_t ttl,
+                             GNUNET_PEER_Id sender_pid,
+                             GNUNET_PEER_Id origin_pid,
+                             const GNUNET_HashCode * replies_seen,
+                             unsigned int replies_seen_count,
+                             GSF_PendingRequestReplyHandler rh, void *rh_cls)
 {
-  return NULL; // FIXME
-}
+  struct GSF_PendingRequest *pr;
+  struct GSF_PendingRequest *dpr;
 
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+              "Creating request handle for `%s' of type %d\n",
+              GNUNET_h2s (query), type);
+#endif
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Pending requests created"), 1,
+                            GNUNET_NO);
+  pr = GNUNET_malloc (sizeof (struct GSF_PendingRequest));
+  pr->local_result_offset =
+      GNUNET_CRYPTO_random_u64 (GNUNET_CRYPTO_QUALITY_WEAK, UINT64_MAX);
+  pr->public_data.query = *query;
+  if (GNUNET_BLOCK_TYPE_FS_SBLOCK == type)
+  {
+    GNUNET_assert (NULL != namespace);
+    pr->public_data.namespace = *namespace;
+  }
+  if (NULL != target)
+  {
+    pr->public_data.target = *target;
+    pr->public_data.has_target = GNUNET_YES;
+  }
+  pr->public_data.anonymity_level = anonymity_level;
+  pr->public_data.priority = priority;
+  pr->public_data.original_priority = priority;
+  pr->public_data.options = options;
+  pr->public_data.type = type;
+  pr->public_data.start_time = GNUNET_TIME_absolute_get ();
+  pr->sender_pid = sender_pid;
+  pr->origin_pid = origin_pid;
+  pr->rh = rh;
+  pr->rh_cls = rh_cls;
+  GNUNET_assert ((sender_pid != 0) || (0 == (options & GSF_PRO_FORWARD_ONLY)));
+  if (ttl >= 0)
+    pr->public_data.ttl =
+        GNUNET_TIME_relative_to_absolute (GNUNET_TIME_relative_multiply
+                                          (GNUNET_TIME_UNIT_SECONDS,
+                                           (uint32_t) ttl));
+  else
+    pr->public_data.ttl =
+        GNUNET_TIME_absolute_subtract (pr->public_data.start_time,
+                                       GNUNET_TIME_relative_multiply
+                                       (GNUNET_TIME_UNIT_SECONDS,
+                                        (uint32_t) (-ttl)));
+  if (replies_seen_count > 0)
+  {
+    pr->replies_seen_size = replies_seen_count;
+    pr->replies_seen =
+        GNUNET_malloc (sizeof (GNUNET_HashCode) * pr->replies_seen_size);
+    memcpy (pr->replies_seen, replies_seen,
+            replies_seen_count * sizeof (GNUNET_HashCode));
+    pr->replies_seen_count = replies_seen_count;
+  }
+  if (NULL != bf_data)
+  {
+    pr->bf =
+        GNUNET_CONTAINER_bloomfilter_init (bf_data, bf_size,
+                                           GNUNET_CONSTANTS_BLOOMFILTER_K);
+    pr->mingle = mingle;
+  }
+  else if ((replies_seen_count > 0) &&
+           (0 != (options & GSF_PRO_BLOOMFILTER_FULL_REFRESH)))
+  {
+    refresh_bloomfilter (pr);
+  }
+  GNUNET_CONTAINER_multihashmap_put (pr_map, query, pr,
+                                     GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE);
+  if (0 == (options & GSF_PRO_REQUEST_NEVER_EXPIRES))
+  {
+    pr->hnode =
+        GNUNET_CONTAINER_heap_insert (requests_by_expiration_heap, pr,
+                                      pr->public_data.ttl.abs_value);
+    /* make sure we don't track too many requests */
+    while (GNUNET_CONTAINER_heap_get_size (requests_by_expiration_heap) >
+           max_pending_requests)
+    {
+      dpr = GNUNET_CONTAINER_heap_peek (requests_by_expiration_heap);
+      GNUNET_assert (dpr != NULL);
+      if (pr == dpr)
+        break;                  /* let the request live briefly... */
+      if (NULL != dpr->rh)
+       dpr->rh (dpr->rh_cls, GNUNET_BLOCK_EVALUATION_REQUEST_VALID, dpr,
+                UINT32_MAX, GNUNET_TIME_UNIT_FOREVER_ABS, GNUNET_TIME_UNIT_FOREVER_ABS,
+                 GNUNET_BLOCK_TYPE_ANY, NULL, 0);
+      GSF_pending_request_cancel_ (dpr, GNUNET_YES);
+    }
+  }
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Pending requests active"), 1,
+                            GNUNET_NO);
+  return pr;
+}
 
 /**
- * Update a given pending request with additional replies
- * that have been seen.
+ * Obtain the public data associated with a pending request
  *
- * @param pr request to update
- * @param replies_seen hash codes of replies that we've seen
- * @param replies_seen_count size of the replies_seen array
+ * @param pr pending request
+ * @return associated public data
  */
-void
-GSF_pending_request_update_ (struct GSF_PendingRequest *pr,
-                            const GNUNET_HashCode *replies_seen,
-                            unsigned int replies_seen_count)
+struct GSF_PendingRequestData *
+GSF_pending_request_get_data_ (struct GSF_PendingRequest *pr)
 {
-  // FIXME
+  return &pr->public_data;
 }
 
 
-
 /**
- * Get the query for a given pending request.
+ * Test if two pending requests are compatible (would generate
+ * the same query modulo filters and should thus be processed
+ * jointly).
  *
- * @param pr the request
- * @return pointer to the query (only valid as long as pr is valid)
+ * @param pra a pending request
+ * @param prb another pending request
+ * @return GNUNET_OK if the requests are compatible
  */
-const GNUNET_HashCode *
-GSF_pending_request_get_query_ (const struct GSF_PendingRequest *pr)
+int
+GSF_pending_request_is_compatible_ (struct GSF_PendingRequest *pra,
+                                    struct GSF_PendingRequest *prb)
 {
-  return NULL; // FIXME
+  if ((pra->public_data.type != prb->public_data.type) ||
+      (0 !=
+       memcmp (&pra->public_data.query, &prb->public_data.query,
+               sizeof (GNUNET_HashCode))) ||
+      ((pra->public_data.type == GNUNET_BLOCK_TYPE_FS_SBLOCK) &&
+       (0 !=
+        memcmp (&pra->public_data.namespace, &prb->public_data.namespace,
+                sizeof (GNUNET_HashCode)))))
+    return GNUNET_NO;
+  return GNUNET_OK;
 }
 
 
+
 /**
- * Get the type of a given pending request.
+ * Update a given pending request with additional replies
+ * that have been seen.
  *
- * @param pr the request
- * @return query type
+ * @param pr request to update
+ * @param replies_seen hash codes of replies that we've seen
+ * @param replies_seen_count size of the replies_seen array
  */
-enum GNUNET_BLOCK_Type
-GSF_pending_request_get_type_ (const struct GSF_PendingRequest *pr)
+void
+GSF_pending_request_update_ (struct GSF_PendingRequest *pr,
+                             const GNUNET_HashCode * replies_seen,
+                             unsigned int replies_seen_count)
 {
-  return 0; // FIXME
+  unsigned int i;
+  GNUNET_HashCode mhash;
+
+  if (replies_seen_count + pr->replies_seen_count < pr->replies_seen_count)
+    return;                     /* integer overflow */
+  if (0 != (pr->public_data.options & GSF_PRO_BLOOMFILTER_FULL_REFRESH))
+  {
+    /* we're responsible for the BF, full refresh */
+    if (replies_seen_count + pr->replies_seen_count > pr->replies_seen_size)
+      GNUNET_array_grow (pr->replies_seen, pr->replies_seen_size,
+                         replies_seen_count + pr->replies_seen_count);
+    memcpy (&pr->replies_seen[pr->replies_seen_count], replies_seen,
+            sizeof (GNUNET_HashCode) * replies_seen_count);
+    pr->replies_seen_count += replies_seen_count;
+    refresh_bloomfilter (pr);
+  }
+  else
+  {
+    if (NULL == pr->bf)
+    {
+      /* we're not the initiator, but the initiator did not give us
+       * any bloom-filter, so we need to create one on-the-fly */
+      pr->mingle =
+          GNUNET_CRYPTO_random_u32 (GNUNET_CRYPTO_QUALITY_WEAK, UINT32_MAX);
+      pr->bf =
+          GNUNET_BLOCK_construct_bloomfilter (pr->mingle, replies_seen,
+                                              replies_seen_count);
+    }
+    else
+    {
+      for (i = 0; i < pr->replies_seen_count; i++)
+      {
+        GNUNET_BLOCK_mingle_hash (&replies_seen[i], pr->mingle, &mhash);
+        GNUNET_CONTAINER_bloomfilter_add (pr->bf, &mhash);
+      }
+    }
+  }
 }
 
 
@@ -119,10 +475,154 @@ GSF_pending_request_get_type_ (const struct GSF_PendingRequest *pr)
  */
 size_t
 GSF_pending_request_get_message_ (struct GSF_PendingRequest *pr,
-                                 size_t buf_size,
-                                 void *buf)
+                                  size_t buf_size, void *buf)
+{
+  char lbuf[GNUNET_SERVER_MAX_MESSAGE_SIZE];
+  struct GetMessage *gm;
+  GNUNET_HashCode *ext;
+  size_t msize;
+  unsigned int k;
+  uint32_t bm;
+  uint32_t prio;
+  size_t bf_size;
+  struct GNUNET_TIME_Absolute now;
+  int64_t ttl;
+  int do_route;
+
+#if DEBUG_FS
+  if (buf_size > 0)
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Building request message for `%s' of type %d\n",
+                GNUNET_h2s (&pr->public_data.query), pr->public_data.type);
+#endif
+  k = 0;
+  bm = 0;
+  do_route = (0 == (pr->public_data.options & GSF_PRO_FORWARD_ONLY));
+  if ((!do_route) && (pr->sender_pid == 0))
+  {
+    GNUNET_break (0);
+    do_route = GNUNET_YES;
+  }
+  if (!do_route)
+  {
+    bm |= GET_MESSAGE_BIT_RETURN_TO;
+    k++;
+  }
+  if (GNUNET_BLOCK_TYPE_FS_SBLOCK == pr->public_data.type)
+  {
+    bm |= GET_MESSAGE_BIT_SKS_NAMESPACE;
+    k++;
+  }
+  if (GNUNET_YES == pr->public_data.has_target)
+  {
+    bm |= GET_MESSAGE_BIT_TRANSMIT_TO;
+    k++;
+  }
+  bf_size = GNUNET_CONTAINER_bloomfilter_get_size (pr->bf);
+  msize = sizeof (struct GetMessage) + bf_size + k * sizeof (GNUNET_HashCode);
+  GNUNET_assert (msize < GNUNET_SERVER_MAX_MESSAGE_SIZE);
+  if (buf_size < msize)
+    return msize;
+  gm = (struct GetMessage *) lbuf;
+  gm->header.type = htons (GNUNET_MESSAGE_TYPE_FS_GET);
+  gm->header.size = htons (msize);
+  gm->type = htonl (pr->public_data.type);
+  if (do_route)
+    prio =
+        GNUNET_CRYPTO_random_u32 (GNUNET_CRYPTO_QUALITY_WEAK,
+                                  pr->public_data.priority + 1);
+  else
+    prio = 0;
+  pr->public_data.priority -= prio;
+  gm->priority = htonl (prio);
+  now = GNUNET_TIME_absolute_get ();
+  ttl = (int64_t) (pr->public_data.ttl.abs_value - now.abs_value);
+  gm->ttl = htonl (ttl / 1000);
+  gm->filter_mutator = htonl (pr->mingle);
+  gm->hash_bitmap = htonl (bm);
+  gm->query = pr->public_data.query;
+  ext = (GNUNET_HashCode *) & gm[1];
+  k = 0;
+  if (!do_route)
+    GNUNET_PEER_resolve (pr->sender_pid,
+                         (struct GNUNET_PeerIdentity *) &ext[k++]);
+  if (GNUNET_BLOCK_TYPE_FS_SBLOCK == pr->public_data.type)
+    memcpy (&ext[k++], &pr->public_data.namespace, sizeof (GNUNET_HashCode));
+  if (GNUNET_YES == pr->public_data.has_target)
+    ext[k++] = pr->public_data.target.hashPubKey;
+  if (pr->bf != NULL)
+    GNUNET_assert (GNUNET_SYSERR !=
+                   GNUNET_CONTAINER_bloomfilter_get_raw_data (pr->bf,
+                                                              (char *) &ext[k],
+                                                              bf_size));
+  memcpy (buf, gm, msize);
+  return msize;
+}
+
+
+/**
+ * Iterator to free pending requests.
+ *
+ * @param cls closure, unused
+ * @param key current key code
+ * @param value value in the hash map (pending request)
+ * @return GNUNET_YES (we should continue to iterate)
+ */
+static int
+clean_request (void *cls, const GNUNET_HashCode * key, void *value)
 {
-  return 0; // FIXME
+  struct GSF_PendingRequest *pr = value;
+  GSF_LocalLookupContinuation cont;
+
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+              "Cleaning up pending request for `%s'.\n", GNUNET_h2s (key));
+#endif
+  if (NULL != (cont = pr->llc_cont))
+  {
+    pr->llc_cont = NULL;
+    cont (pr->llc_cont_cls, pr, pr->local_result);
+  }
+  GSF_plan_notify_request_done_ (pr);
+  GNUNET_free_non_null (pr->replies_seen);
+  if (NULL != pr->bf)
+  {
+    GNUNET_CONTAINER_bloomfilter_free (pr->bf);
+    pr->bf = NULL;
+  }
+  GNUNET_PEER_change_rc (pr->sender_pid, -1);
+  pr->sender_pid = 0;
+  GNUNET_PEER_change_rc (pr->origin_pid, -1);
+  pr->origin_pid = 0;
+  if (NULL != pr->hnode)
+  {
+    GNUNET_CONTAINER_heap_remove_node (pr->hnode);
+    pr->hnode = NULL;
+  }
+  if (NULL != pr->qe)
+  {
+    GNUNET_DATASTORE_cancel (pr->qe);
+    pr->qe = NULL;
+  }
+  if (NULL != pr->gh)
+  {
+    GNUNET_DHT_get_stop (pr->gh);
+    pr->gh = NULL;
+  }
+  if (GNUNET_SCHEDULER_NO_TASK != pr->warn_task)
+  {
+    GNUNET_SCHEDULER_cancel (pr->warn_task);
+    pr->warn_task = GNUNET_SCHEDULER_NO_TASK;
+  }
+  GNUNET_assert (GNUNET_OK ==
+                 GNUNET_CONTAINER_multihashmap_remove (pr_map,
+                                                       &pr->public_data.query,
+                                                       pr));
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Pending requests active"), -1,
+                            GNUNET_NO);
+  GNUNET_free (pr);
+  return GNUNET_YES;
 }
 
 
@@ -130,10 +630,46 @@ GSF_pending_request_get_message_ (struct GSF_PendingRequest *pr,
  * Explicitly cancel a pending request.
  *
  * @param pr request to cancel
+ * @param full_cleanup fully purge the request
  */
 void
-GSF_pending_request_cancel_ (struct GSF_PendingRequest *pr)
+GSF_pending_request_cancel_ (struct GSF_PendingRequest *pr, int full_cleanup)
 {
+  GSF_LocalLookupContinuation cont;
+
+  if (NULL == pr_map)
+    return;                     /* already cleaned up! */
+  if (GNUNET_YES != full_cleanup)
+  {
+    /* make request inactive (we're no longer interested in more results),
+     * but do NOT remove from our data-structures, we still need it there
+     * to prevent the request from looping */
+    pr->rh = NULL;
+    if (NULL != (cont = pr->llc_cont))
+    {
+      pr->llc_cont = NULL;
+      cont (pr->llc_cont_cls, pr, pr->local_result);
+    }
+    GSF_plan_notify_request_done_ (pr);
+    if (NULL != pr->qe)
+    {
+      GNUNET_DATASTORE_cancel (pr->qe);
+      pr->qe = NULL;
+    }
+    if (NULL != pr->gh)
+    {
+      GNUNET_DHT_get_stop (pr->gh);
+      pr->gh = NULL;
+    }
+    if (GNUNET_SCHEDULER_NO_TASK != pr->warn_task)
+    {
+      GNUNET_SCHEDULER_cancel (pr->warn_task);
+      pr->warn_task = GNUNET_SCHEDULER_NO_TASK;
+    }
+    return;
+  }
+  GNUNET_assert (GNUNET_YES ==
+                 clean_request (NULL, &pr->public_data.query, pr));
 }
 
 
@@ -144,31 +680,938 @@ GSF_pending_request_cancel_ (struct GSF_PendingRequest *pr)
  * @param cls closure for it
  */
 void
-GSF_iterate_pending_requests_ (GSF_PendingRequestIterator it,
-                              void *cls)
+GSF_iterate_pending_requests_ (GSF_PendingRequestIterator it, void *cls)
 {
-  // FIXME
+  GNUNET_CONTAINER_multihashmap_iterate (pr_map,
+                                         (GNUNET_CONTAINER_HashMapIterator) it,
+                                         cls);
 }
 
 
+
+
+/**
+ * Closure for "process_reply" function.
+ */
+struct ProcessReplyClosure
+{
+  /**
+   * The data for the reply.
+   */
+  const void *data;
+
+  /**
+   * Who gave us this reply? NULL for local host (or DHT)
+   */
+  struct GSF_ConnectedPeer *sender;
+
+  /**
+   * When the reply expires.
+   */
+  struct GNUNET_TIME_Absolute expiration;
+
+  /**
+   * Size of data.
+   */
+  size_t size;
+
+  /**
+   * Type of the block.
+   */
+  enum GNUNET_BLOCK_Type type;
+
+  /**
+   * How much was this reply worth to us?
+   */
+  uint32_t priority;
+
+  /**
+   * Anonymity requirements for this reply.
+   */
+  uint32_t anonymity_level;
+
+  /**
+   * Evaluation result (returned).
+   */
+  enum GNUNET_BLOCK_EvaluationResult eval;
+
+  /**
+   * Did we find a matching request?
+   */
+  int request_found;
+};
+
+
+/**
+ * Update the performance data for the sender (if any) since
+ * the sender successfully answered one of our queries.
+ *
+ * @param prq information about the sender
+ * @param pr request that was satisfied
+ */
+static void
+update_request_performance_data (struct ProcessReplyClosure *prq,
+                                 struct GSF_PendingRequest *pr)
+{
+  if (prq->sender == NULL)
+    return;
+  GSF_peer_update_performance_ (prq->sender, pr->public_data.start_time,
+                                prq->priority);
+}
+
+
+/**
+ * We have received a reply; handle it!
+ *
+ * @param cls response (struct ProcessReplyClosure)
+ * @param key our query
+ * @param value value in the hash map (info about the query)
+ * @return GNUNET_YES (we should continue to iterate)
+ */
+static int
+process_reply (void *cls, const GNUNET_HashCode * key, void *value)
+{
+  struct ProcessReplyClosure *prq = cls;
+  struct GSF_PendingRequest *pr = value;
+  GNUNET_HashCode chash;
+  struct GNUNET_TIME_Absolute last_transmission;
+
+  if (NULL == pr->rh)
+    return GNUNET_YES;
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+              "Matched result (type %u) for query `%s' with pending request\n",
+              (unsigned int) prq->type, GNUNET_h2s (key));
+#endif
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# replies received and matched"), 1,
+                            GNUNET_NO);
+  prq->eval =
+      GNUNET_BLOCK_evaluate (GSF_block_ctx, prq->type, key, &pr->bf, pr->mingle,
+                             &pr->public_data.namespace,
+                             (prq->type ==
+                              GNUNET_BLOCK_TYPE_FS_SBLOCK) ?
+                             sizeof (GNUNET_HashCode) : 0, prq->data,
+                             prq->size);
+  switch (prq->eval)
+  {
+  case GNUNET_BLOCK_EVALUATION_OK_MORE:
+    update_request_performance_data (prq, pr);
+    break;
+  case GNUNET_BLOCK_EVALUATION_OK_LAST:
+    /* short cut: stop processing early, no BF-update, etc. */
+    update_request_performance_data (prq, pr);
+    GNUNET_LOAD_update (GSF_rt_entry_lifetime,
+                        GNUNET_TIME_absolute_get_duration (pr->
+                                                           public_data.start_time).rel_value);
+    if (!GSF_request_plan_reference_get_last_transmission_ (pr->public_data.rpr_head, prq->sender, &last_transmission))
+      last_transmission.abs_value = GNUNET_TIME_UNIT_FOREVER_ABS.abs_value;
+    /* pass on to other peers / local clients */
+    pr->rh (pr->rh_cls, prq->eval, pr, prq->anonymity_level, prq->expiration,
+            last_transmission, prq->type, prq->data, prq->size);
+    return GNUNET_YES;
+  case GNUNET_BLOCK_EVALUATION_OK_DUPLICATE:
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop
+                              ("# duplicate replies discarded (bloomfilter)"),
+                              1, GNUNET_NO);
+#if DEBUG_FS && 0
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Duplicate response `%s', discarding.\n", GNUNET_h2s (&mhash));
+#endif
+    return GNUNET_YES;          /* duplicate */
+  case GNUNET_BLOCK_EVALUATION_RESULT_INVALID:
+    return GNUNET_YES;          /* wrong namespace */
+  case GNUNET_BLOCK_EVALUATION_REQUEST_VALID:
+    GNUNET_break (0);
+    return GNUNET_YES;
+  case GNUNET_BLOCK_EVALUATION_REQUEST_INVALID:
+    GNUNET_break (0);
+    return GNUNET_YES;
+  case GNUNET_BLOCK_EVALUATION_TYPE_NOT_SUPPORTED:
+    GNUNET_log (GNUNET_ERROR_TYPE_ERROR, _("Unsupported block type %u\n"),
+                prq->type);
+    return GNUNET_NO;
+  }
+  /* update bloomfilter */
+  GNUNET_CRYPTO_hash (prq->data, prq->size, &chash);
+  GSF_pending_request_update_ (pr, &chash, 1);
+  if (NULL == prq->sender)
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Found result for query `%s' in local datastore\n",
+                GNUNET_h2s (key));
+#endif
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop ("# results found locally"), 1,
+                              GNUNET_NO);
+  }
+  else
+  {
+    GSF_dht_lookup_ (pr);
+  }
+  prq->priority += pr->public_data.original_priority;
+  pr->public_data.priority = 0;
+  pr->public_data.original_priority = 0;
+  pr->public_data.results_found++;
+  prq->request_found = GNUNET_YES;
+  /* finally, pass on to other peer / local client */
+  if (!GSF_request_plan_reference_get_last_transmission_ (pr->public_data.rpr_head, prq->sender, &last_transmission))
+    last_transmission.abs_value = GNUNET_TIME_UNIT_FOREVER_ABS.abs_value;
+  pr->rh (pr->rh_cls, prq->eval, pr, prq->anonymity_level, prq->expiration,
+          last_transmission, prq->type, prq->data, prq->size);
+  return GNUNET_YES;
+}
+
+
+/**
+ * Context for the 'put_migration_continuation'.
+ */
+struct PutMigrationContext
+{
+
+  /**
+   * Start time for the operation.
+   */
+  struct GNUNET_TIME_Absolute start;
+
+  /**
+   * Request origin.
+   */
+  struct GNUNET_PeerIdentity origin;
+
+  /**
+   * GNUNET_YES if we had a matching request for this block,
+   * GNUNET_NO if not.
+   */
+  int requested;
+};
+
+
+/**
+ * Continuation called to notify client about result of the
+ * operation.
+ *
+ * @param cls closure
+ * @param success GNUNET_SYSERR on failure
+ * @param min_expiration minimum expiration time required for content to be stored
+ * @param msg NULL on success, otherwise an error message
+ */
+static void
+put_migration_continuation (void *cls, int success, 
+                           struct GNUNET_TIME_Absolute min_expiration,
+                           const char *msg)
+{
+  struct PutMigrationContext *pmc = cls;
+  struct GSF_ConnectedPeer *cp;
+  struct GNUNET_TIME_Relative mig_pause;
+  struct GSF_PeerPerformanceData *ppd;
+
+  if (NULL != datastore_put_load)
+  {
+    if (GNUNET_SYSERR != success)
+    {
+      GNUNET_LOAD_update (datastore_put_load, 
+                         GNUNET_TIME_absolute_get_duration (pmc->start).rel_value);
+    }
+    else
+    {
+      /* on queue failure / timeout, increase the put load dramatically */
+      GNUNET_LOAD_update (datastore_put_load, 
+                         GNUNET_TIME_UNIT_MINUTES.rel_value);
+    }
+  }
+  cp = GSF_peer_get_ (&pmc->origin);
+  if (GNUNET_OK == success)
+  {
+    if (NULL != cp)
+    {
+      ppd = GSF_get_peer_performance_data_ (cp);
+      ppd->migration_delay.rel_value /= 2;
+    }
+    GNUNET_free (pmc);
+    return;
+  }
+  if ( (GNUNET_NO == success) && 
+       (GNUNET_NO == pmc->requested) && 
+       (NULL != cp) )
+  {
+    ppd = GSF_get_peer_performance_data_ (cp);
+    if (min_expiration.abs_value > 0)
+    {
+#if DEBUG_FS
+      GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, 
+                 "Asking to stop migration for %llu ms because datastore is full\n",
+                 (unsigned long long) GNUNET_TIME_absolute_get_remaining (min_expiration).rel_value);
+#endif
+      GSF_block_peer_migration_ (cp, min_expiration);      
+    }
+    else
+    {
+      ppd->migration_delay = GNUNET_TIME_relative_max (GNUNET_TIME_UNIT_SECONDS,
+                                                      ppd->migration_delay);
+      ppd->migration_delay = GNUNET_TIME_relative_min (GNUNET_TIME_UNIT_HOURS,
+                                                      ppd->migration_delay);
+      mig_pause.rel_value = GNUNET_CRYPTO_random_u64 (GNUNET_CRYPTO_QUALITY_WEAK,
+                                                     ppd->migration_delay.rel_value);
+      ppd->migration_delay = GNUNET_TIME_relative_multiply (ppd->migration_delay, 2);
+#if DEBUG_FS
+      GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, 
+                 "Replicated content already exists locally, asking to stop migration for %llu ms\n",
+                 (unsigned long long) mig_pause.rel_value);
+#endif
+      GSF_block_peer_migration_ (cp, GNUNET_TIME_relative_to_absolute (mig_pause));
+    }
+  }
+  GNUNET_free (pmc);
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Datastore `PUT' failures"), 1,
+                            GNUNET_NO);
+}
+
+
+/**
+ * Test if the DATABASE (PUT) load on this peer is too high
+ * to even consider processing the query at
+ * all.
+ *
+ * @return GNUNET_YES if the load is too high to do anything (load high)
+ *         GNUNET_NO to process normally (load normal or low)
+ */
+static int
+test_put_load_too_high (uint32_t priority)
+{
+  double ld;
+
+  if (NULL == datastore_put_load)
+    return GNUNET_NO;
+  if (GNUNET_LOAD_get_average (datastore_put_load) < 50)
+    return GNUNET_NO;           /* very fast */
+  ld = GNUNET_LOAD_get_load (datastore_put_load);
+  if (ld < 2.0 * (1 + priority))
+    return GNUNET_NO;
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop
+                            ("# storage requests dropped due to high load"), 1,
+                            GNUNET_NO);
+  return GNUNET_YES;
+}
+
+
+/**
+ * Iterator called on each result obtained for a DHT
+ * operation that expects a reply
+ *
+ * @param cls closure
+ * @param exp when will this value expire
+ * @param key key of the result
+ * @param get_path peers on reply path (or NULL if not recorded)
+ * @param get_path_length number of entries in get_path
+ * @param put_path peers on the PUT path (or NULL if not recorded)
+ * @param put_path_length number of entries in get_path
+ * @param type type of the result
+ * @param size number of bytes in data
+ * @param data pointer to the result data
+ */
+static void
+handle_dht_reply (void *cls, struct GNUNET_TIME_Absolute exp,
+                  const GNUNET_HashCode * key,
+                  const struct GNUNET_PeerIdentity *get_path,
+                  unsigned int get_path_length,
+                  const struct GNUNET_PeerIdentity *put_path,
+                  unsigned int put_path_length, enum GNUNET_BLOCK_Type type,
+                  size_t size, const void *data)
+{
+  struct GSF_PendingRequest *pr = cls;
+  struct ProcessReplyClosure prq;
+  struct PutMigrationContext *pmc;
+
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Replies received from DHT"), 1,
+                            GNUNET_NO);
+  memset (&prq, 0, sizeof (prq));
+  prq.data = data;
+  prq.expiration = exp;
+  /* do not allow migrated content to live longer than 1 year */
+  prq.expiration = GNUNET_TIME_absolute_min (GNUNET_TIME_relative_to_absolute (GNUNET_TIME_UNIT_YEARS),
+                                            prq.expiration);
+  prq.size = size;
+  prq.type = type;
+  process_reply (&prq, key, pr);
+  if ((GNUNET_YES == active_to_migration) &&
+      (GNUNET_NO == test_put_load_too_high (prq.priority)))
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Replicating result for query `%s' with priority %u\n",
+                GNUNET_h2s (key), prq.priority);
+#endif
+    pmc = GNUNET_malloc (sizeof (struct PutMigrationContext));
+    pmc->start = GNUNET_TIME_absolute_get ();
+    pmc->requested = GNUNET_YES;
+    if (NULL ==
+        GNUNET_DATASTORE_put (GSF_dsh, 0, key, size, data, type, prq.priority,
+                              1 /* anonymity */ ,
+                              0 /* replication */ ,
+                              exp, 1 + prq.priority, MAX_DATASTORE_QUEUE,
+                              GNUNET_CONSTANTS_SERVICE_TIMEOUT,
+                              &put_migration_continuation, pmc))
+    {
+      put_migration_continuation (pmc, GNUNET_SYSERR, GNUNET_TIME_UNIT_ZERO_ABS, NULL);
+    }
+  }
+}
+
+
+/**
+ * Consider looking up the data in the DHT (anonymity-level permitting).
+ *
+ * @param pr the pending request to process
+ */
+void
+GSF_dht_lookup_ (struct GSF_PendingRequest *pr)
+{
+  const void *xquery;
+  size_t xquery_size;
+  struct GNUNET_PeerIdentity pi;
+  char buf[sizeof (GNUNET_HashCode) * 2];
+
+  if (0 != pr->public_data.anonymity_level)
+    return;
+  if (NULL != pr->gh)
+  {
+    GNUNET_DHT_get_stop (pr->gh);
+    pr->gh = NULL;
+  }
+  xquery = NULL;
+  xquery_size = 0;
+  if (GNUNET_BLOCK_TYPE_FS_SBLOCK == pr->public_data.type)
+  {
+    xquery = buf;
+    memcpy (buf, &pr->public_data.namespace, sizeof (GNUNET_HashCode));
+    xquery_size = sizeof (GNUNET_HashCode);
+  }
+  if (0 != (pr->public_data.options & GSF_PRO_FORWARD_ONLY))
+  {
+    GNUNET_assert (0 != pr->sender_pid);
+    GNUNET_PEER_resolve (pr->sender_pid, &pi);
+    memcpy (&buf[xquery_size], &pi, sizeof (struct GNUNET_PeerIdentity));
+    xquery_size += sizeof (struct GNUNET_PeerIdentity);
+  }
+  pr->gh =
+      GNUNET_DHT_get_start (GSF_dht, GNUNET_TIME_UNIT_FOREVER_REL,
+                            pr->public_data.type, &pr->public_data.query,
+                            5 /* DEFAULT_GET_REPLICATION */ ,
+                            GNUNET_DHT_RO_DEMULTIPLEX_EVERYWHERE,
+                            /* FIXME: can no longer pass pr->bf/pr->mingle... */
+                            xquery, xquery_size, &handle_dht_reply, pr);
+}
+
+
+/**
+ * Task that issues a warning if the datastore lookup takes too long.
+ *
+ * @param cls the 'struct GSF_PendingRequest'
+ * @param tc task context
+ */
+static void
+warn_delay_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+  struct GSF_PendingRequest *pr = cls;
+
+  GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
+              _("Datastore lookup already took %llu ms!\n"),
+              (unsigned long long)
+              GNUNET_TIME_absolute_get_duration (pr->qe_start).rel_value);
+  pr->warn_task =
+      GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES, &warn_delay_task,
+                                    pr);
+}
+
+
+/**
+ * Task that issues a warning if the datastore lookup takes too long.
+ *
+ * @param cls the 'struct GSF_PendingRequest'
+ * @param tc task context
+ */
+static void
+odc_warn_delay_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+  struct GSF_PendingRequest *pr = cls;
+
+  GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
+              _("On-demand lookup already took %llu ms!\n"),
+              (unsigned long long)
+              GNUNET_TIME_absolute_get_duration (pr->qe_start).rel_value);
+  pr->warn_task =
+      GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES,
+                                    &odc_warn_delay_task, pr);
+}
+
+
+/**
+ * We're processing (local) results for a search request
+ * from another peer.  Pass applicable results to the
+ * peer and if we are done either clean up (operation
+ * complete) or forward to other peers (more results possible).
+ *
+ * @param cls our closure (struct PendingRequest)
+ * @param key key for the content
+ * @param size number of bytes in data
+ * @param data content stored
+ * @param type type of the content
+ * @param priority priority of the content
+ * @param anonymity anonymity-level for the content
+ * @param expiration expiration time for the content
+ * @param uid unique identifier for the datum;
+ *        maybe 0 if no unique identifier is available
+ */
+static void
+process_local_reply (void *cls, const GNUNET_HashCode * key, size_t size,
+                     const void *data, enum GNUNET_BLOCK_Type type,
+                     uint32_t priority, uint32_t anonymity,
+                     struct GNUNET_TIME_Absolute expiration, uint64_t uid)
+{
+  struct GSF_PendingRequest *pr = cls;
+  GSF_LocalLookupContinuation cont;
+  struct ProcessReplyClosure prq;
+  GNUNET_HashCode query;
+  unsigned int old_rf;
+
+  GNUNET_SCHEDULER_cancel (pr->warn_task);
+  pr->warn_task = GNUNET_SCHEDULER_NO_TASK;
+  if (NULL != pr->qe)
+  {
+    pr->qe = NULL;
+    if (NULL == key)
+    {
+      GNUNET_STATISTICS_update (GSF_stats,
+                                gettext_noop
+                                ("# Datastore lookups concluded (no results)"),
+                                1, GNUNET_NO);
+    }
+    if (GNUNET_NO == pr->have_first_uid)
+    {
+      pr->first_uid = uid;
+      pr->have_first_uid = 1;
+    }
+    else
+    {
+      if ((uid == pr->first_uid) && (key != NULL))
+      {
+        GNUNET_STATISTICS_update (GSF_stats,
+                                  gettext_noop
+                                  ("# Datastore lookups concluded (seen all)"),
+                                  1, GNUNET_NO);
+        key = NULL;             /* all replies seen! */
+      }
+      pr->have_first_uid++;
+      if ((pr->have_first_uid > MAX_RESULTS) && (key != NULL))
+      {
+        GNUNET_STATISTICS_update (GSF_stats,
+                                  gettext_noop
+                                  ("# Datastore lookups aborted (more than MAX_RESULTS)"),
+                                  1, GNUNET_NO);
+        key = NULL;             /* all replies seen! */
+      }
+    }
+  }
+  if (NULL == key)
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG | GNUNET_ERROR_TYPE_BULK,
+                "No further local responses available.\n");
+#endif
+    if ((pr->public_data.type == GNUNET_BLOCK_TYPE_FS_DBLOCK) ||
+        (pr->public_data.type == GNUNET_BLOCK_TYPE_FS_IBLOCK))
+      GNUNET_STATISTICS_update (GSF_stats,
+                                gettext_noop
+                                ("# requested DBLOCK or IBLOCK not found"), 1,
+                                GNUNET_NO);
+    goto check_error_and_continue;
+  }
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+              "Received reply for `%s' of type %d with UID %llu from datastore.\n",
+              GNUNET_h2s (key), type, (unsigned long long) uid);
+#endif
+  if (type == GNUNET_BLOCK_TYPE_FS_ONDEMAND)
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Found ONDEMAND block, performing on-demand encoding\n");
+#endif
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop
+                              ("# on-demand blocks matched requests"), 1,
+                              GNUNET_NO);
+    pr->qe_start = GNUNET_TIME_absolute_get ();
+    pr->warn_task =
+        GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES,
+                                      &odc_warn_delay_task, pr);
+    if (GNUNET_OK ==
+        GNUNET_FS_handle_on_demand_block (key, size, data, type, priority,
+                                          anonymity, expiration, uid,
+                                          &process_local_reply, pr))
+    {
+      GNUNET_STATISTICS_update (GSF_stats,
+                                gettext_noop
+                                ("# on-demand lookups performed successfully"),
+                                1, GNUNET_NO);
+      return;                   /* we're done */
+    }
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop ("# on-demand lookups failed"), 1,
+                              GNUNET_NO);
+    GNUNET_SCHEDULER_cancel (pr->warn_task);
+    pr->warn_task =
+        GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES,
+                                      &warn_delay_task, pr);
+    pr->qe =
+        GNUNET_DATASTORE_get_key (GSF_dsh, pr->local_result_offset - 1,
+                                  &pr->public_data.query,
+                                  pr->public_data.type ==
+                                  GNUNET_BLOCK_TYPE_FS_DBLOCK ?
+                                  GNUNET_BLOCK_TYPE_ANY : pr->public_data.type,
+                                  (0 !=
+                                   (GSF_PRO_PRIORITY_UNLIMITED &
+                                    pr->public_data.options)) ? UINT_MAX : 1
+                                  /* queue priority */ ,
+                                  (0 !=
+                                   (GSF_PRO_PRIORITY_UNLIMITED &
+                                    pr->public_data.options)) ? UINT_MAX :
+                                  datastore_queue_size
+                                  /* max queue size */ ,
+                                  GNUNET_TIME_UNIT_FOREVER_REL,
+                                  &process_local_reply, pr);
+    if (NULL != pr->qe)
+      return;                   /* we're done */
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop
+                              ("# Datastore lookups concluded (error queueing)"),
+                              1, GNUNET_NO);
+    goto check_error_and_continue;
+  }
+  old_rf = pr->public_data.results_found;
+  memset (&prq, 0, sizeof (prq));
+  prq.data = data;
+  prq.expiration = expiration;
+  prq.size = size;
+  if (GNUNET_OK !=
+      GNUNET_BLOCK_get_key (GSF_block_ctx, type, data, size, &query))
+  {
+    GNUNET_break (0);
+    GNUNET_DATASTORE_remove (GSF_dsh, key, size, data, -1, -1,
+                             GNUNET_TIME_UNIT_FOREVER_REL, NULL, NULL);
+    pr->qe_start = GNUNET_TIME_absolute_get ();
+    pr->warn_task =
+        GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES,
+                                      &warn_delay_task, pr);
+    pr->qe =
+        GNUNET_DATASTORE_get_key (GSF_dsh, pr->local_result_offset - 1,
+                                  &pr->public_data.query,
+                                  pr->public_data.type ==
+                                  GNUNET_BLOCK_TYPE_FS_DBLOCK ?
+                                  GNUNET_BLOCK_TYPE_ANY : pr->public_data.type,
+                                  (0 !=
+                                   (GSF_PRO_PRIORITY_UNLIMITED &
+                                    pr->public_data.options)) ? UINT_MAX : 1
+                                  /* queue priority */ ,
+                                  (0 !=
+                                   (GSF_PRO_PRIORITY_UNLIMITED &
+                                    pr->public_data.options)) ? UINT_MAX :
+                                  datastore_queue_size
+                                  /* max queue size */ ,
+                                  GNUNET_TIME_UNIT_FOREVER_REL,
+                                  &process_local_reply, pr);
+    if (pr->qe == NULL)
+    {
+      GNUNET_STATISTICS_update (GSF_stats,
+                                gettext_noop
+                                ("# Datastore lookups concluded (error queueing)"),
+                                1, GNUNET_NO);
+      goto check_error_and_continue;
+    }
+    return;
+  }
+  prq.type = type;
+  prq.priority = priority;
+  prq.request_found = GNUNET_NO;
+  prq.anonymity_level = anonymity;
+  if ((old_rf == 0) && (pr->public_data.results_found == 0))
+    GSF_update_datastore_delay_ (pr->public_data.start_time);
+  process_reply (&prq, key, pr);
+  pr->local_result = prq.eval;
+  if (prq.eval == GNUNET_BLOCK_EVALUATION_OK_LAST)
+  {
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop
+                              ("# Datastore lookups concluded (found last result)"),
+                              1, GNUNET_NO);
+    goto check_error_and_continue;
+  }
+  if ((0 == (GSF_PRO_PRIORITY_UNLIMITED & pr->public_data.options)) &&
+      ((GNUNET_YES == GSF_test_get_load_too_high_ (0)) ||
+       (pr->public_data.results_found > 5 + 2 * pr->public_data.priority)))
+  {
+#if DEBUG_FS > 2
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "Load too high, done with request\n");
+#endif
+    GNUNET_STATISTICS_update (GSF_stats,
+                              gettext_noop
+                              ("# Datastore lookups concluded (load too high)"),
+                              1, GNUNET_NO);
+    goto check_error_and_continue;
+  }
+  pr->qe_start = GNUNET_TIME_absolute_get ();
+  pr->warn_task =
+      GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES, &warn_delay_task,
+                                    pr);
+  pr->qe =
+      GNUNET_DATASTORE_get_key (GSF_dsh, pr->local_result_offset++,
+                                &pr->public_data.query,
+                                pr->public_data.type ==
+                                GNUNET_BLOCK_TYPE_FS_DBLOCK ?
+                                GNUNET_BLOCK_TYPE_ANY : pr->public_data.type,
+                                (0 !=
+                                 (GSF_PRO_PRIORITY_UNLIMITED & pr->
+                                  public_data.options)) ? UINT_MAX : 1
+                                /* queue priority */ ,
+                                (0 !=
+                                 (GSF_PRO_PRIORITY_UNLIMITED & pr->
+                                  public_data.options)) ? UINT_MAX :
+                                datastore_queue_size
+                                /* max queue size */ ,
+                                GNUNET_TIME_UNIT_FOREVER_REL,
+                                &process_local_reply, pr);
+  /* check if we successfully queued another datastore request;
+   * if so, return, otherwise call our continuation (if we have
+   * any) */
+check_error_and_continue:
+  if (NULL != pr->qe)
+    return;
+  if (GNUNET_SCHEDULER_NO_TASK != pr->warn_task)
+  {
+    GNUNET_SCHEDULER_cancel (pr->warn_task);
+    pr->warn_task = GNUNET_SCHEDULER_NO_TASK;
+  }
+  if (NULL == (cont = pr->llc_cont))
+    return;                     /* no continuation */
+  pr->llc_cont = NULL;
+  cont (pr->llc_cont_cls, pr, pr->local_result);
+}
+
+
+/**
+ * Is the given target a legitimate peer for forwarding the given request?
+ *
+ * @param pr request
+ * @param target
+ * @return GNUNET_YES if this request could be forwarded to the given peer
+ */
+int
+GSF_pending_request_test_target_ (struct GSF_PendingRequest *pr,
+                                  const struct GNUNET_PeerIdentity *target)
+{
+  struct GNUNET_PeerIdentity pi;
+
+  if (0 == pr->origin_pid)
+    return GNUNET_YES;
+  GNUNET_PEER_resolve (pr->origin_pid, &pi);
+  return (0 ==
+          memcmp (&pi, target,
+                  sizeof (struct GNUNET_PeerIdentity))) ? GNUNET_NO :
+      GNUNET_YES;
+}
+
+
+/**
+ * Look up the request in the local datastore.
+ *
+ * @param pr the pending request to process
+ * @param cont function to call at the end
+ * @param cont_cls closure for cont
+ */
+void
+GSF_local_lookup_ (struct GSF_PendingRequest *pr,
+                   GSF_LocalLookupContinuation cont, void *cont_cls)
+{
+  GNUNET_assert (NULL == pr->gh);
+  GNUNET_assert (NULL == pr->llc_cont);
+  pr->llc_cont = cont;
+  pr->llc_cont_cls = cont_cls;
+  pr->qe_start = GNUNET_TIME_absolute_get ();
+  pr->warn_task =
+      GNUNET_SCHEDULER_add_delayed (GNUNET_TIME_UNIT_MINUTES, &warn_delay_task,
+                                    pr);
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# Datastore lookups initiated"), 1,
+                            GNUNET_NO);
+  pr->qe =
+      GNUNET_DATASTORE_get_key (GSF_dsh, pr->local_result_offset++,
+                                &pr->public_data.query,
+                                pr->public_data.type ==
+                                GNUNET_BLOCK_TYPE_FS_DBLOCK ?
+                                GNUNET_BLOCK_TYPE_ANY : pr->public_data.type,
+                                (0 !=
+                                 (GSF_PRO_PRIORITY_UNLIMITED & pr->
+                                  public_data.options)) ? UINT_MAX : 1
+                                /* queue priority */ ,
+                                (0 !=
+                                 (GSF_PRO_PRIORITY_UNLIMITED & pr->
+                                  public_data.options)) ? UINT_MAX :
+                                datastore_queue_size
+                                /* max queue size */ ,
+                                GNUNET_TIME_UNIT_FOREVER_REL,
+                                &process_local_reply, pr);
+  if (NULL != pr->qe)
+    return;
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop
+                            ("# Datastore lookups concluded (error queueing)"),
+                            1, GNUNET_NO);
+  GNUNET_SCHEDULER_cancel (pr->warn_task);
+  pr->warn_task = GNUNET_SCHEDULER_NO_TASK;
+  pr->llc_cont = NULL;
+  if (NULL != cont)
+    cont (cont_cls, pr, pr->local_result);
+}
+
+
+
 /**
  * Handle P2P "CONTENT" message.  Checks that the message is
  * well-formed and then checks if there are any pending requests for
  * this content and possibly passes it on (to local clients or other
  * peers).  Does NOT perform migration (content caching at this peer).
  *
- * @param other the other peer involved (sender or receiver, NULL
+ * @param cp the other peer involved (sender or receiver, NULL
  *        for loopback messages where we are both sender and receiver)
  * @param message the actual message
- * @return how valueable was the content to us (0 for not at all),
+ * @return GNUNET_OK if the message was well-formed,
  *         GNUNET_SYSERR if the message was malformed (close connection,
  *         do not cache under any circumstances)
  */
 int
-GSF_handle_p2p_content_ (const struct GNUNET_PeerIdentity *other,
-                        const struct GNUNET_MessageHeader *message)
+GSF_handle_p2p_content_ (struct GSF_ConnectedPeer *cp,
+                         const struct GNUNET_MessageHeader *message)
 {
-  return GNUNET_SYSERR; // FIXME
+  const struct PutMessage *put;
+  uint16_t msize;
+  size_t dsize;
+  enum GNUNET_BLOCK_Type type;
+  struct GNUNET_TIME_Absolute expiration;
+  GNUNET_HashCode query;
+  struct ProcessReplyClosure prq;
+  struct GNUNET_TIME_Relative block_time;
+  double putl;
+  struct PutMigrationContext *pmc;
+
+  msize = ntohs (message->size);
+  if (msize < sizeof (struct PutMessage))
+  {
+    GNUNET_break_op (0);
+    return GNUNET_SYSERR;
+  }
+  put = (const struct PutMessage *) message;
+  dsize = msize - sizeof (struct PutMessage);
+  type = ntohl (put->type);
+  expiration = GNUNET_TIME_absolute_ntoh (put->expiration);
+  /* do not allow migrated content to live longer than 1 year */
+  expiration = GNUNET_TIME_absolute_min (GNUNET_TIME_relative_to_absolute (GNUNET_TIME_UNIT_YEARS),
+                                        expiration);
+  if (type == GNUNET_BLOCK_TYPE_FS_ONDEMAND)
+    return GNUNET_SYSERR;
+  if (GNUNET_OK !=
+      GNUNET_BLOCK_get_key (GSF_block_ctx, type, &put[1], dsize, &query))
+  {
+    GNUNET_break_op (0);
+    return GNUNET_SYSERR;
+  }
+  GNUNET_STATISTICS_update (GSF_stats,
+                            gettext_noop ("# GAP PUT messages received"), 1,
+                            GNUNET_NO);
+  /* now, lookup 'query' */
+  prq.data = (const void *) &put[1];
+  if (NULL != cp)
+    prq.sender = cp;
+  else
+    prq.sender = NULL;
+  prq.size = dsize;
+  prq.type = type;
+  prq.expiration = expiration;
+  prq.priority = 0;
+  prq.anonymity_level = UINT32_MAX;
+  prq.request_found = GNUNET_NO;
+  GNUNET_CONTAINER_multihashmap_get_multiple (pr_map, &query, &process_reply,
+                                              &prq);
+  if (NULL != cp)
+  {
+    GSF_connected_peer_change_preference_ (cp,
+                                           CONTENT_BANDWIDTH_VALUE +
+                                           1000 * prq.priority);
+    GSF_get_peer_performance_data_ (cp)->trust += prq.priority;
+  }
+  if ((GNUNET_YES == active_to_migration) &&
+      (GNUNET_NO == test_put_load_too_high (prq.priority)))
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Replicating result for query `%s' with priority %u\n",
+                GNUNET_h2s (&query), prq.priority);
+#endif
+    pmc = GNUNET_malloc (sizeof (struct PutMigrationContext));
+    pmc->start = GNUNET_TIME_absolute_get ();
+    pmc->requested = prq.request_found;
+    GNUNET_assert (0 != GSF_get_peer_performance_data_ (cp)->pid);
+    GNUNET_PEER_resolve (GSF_get_peer_performance_data_ (cp)->pid,
+                         &pmc->origin);
+    if (NULL ==
+        GNUNET_DATASTORE_put (GSF_dsh, 0, &query, dsize, &put[1], type,
+                              prq.priority, 1 /* anonymity */ ,
+                              0 /* replication */ ,
+                              expiration, 1 + prq.priority, MAX_DATASTORE_QUEUE,
+                              GNUNET_CONSTANTS_SERVICE_TIMEOUT,
+                              &put_migration_continuation, pmc))
+    {
+      put_migration_continuation (pmc, GNUNET_SYSERR, GNUNET_TIME_UNIT_ZERO_ABS, NULL);
+    }
+  }
+  else
+  {
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                "Choosing not to keep content `%s' (%d/%d)\n",
+                GNUNET_h2s (&query), active_to_migration,
+                test_put_load_too_high (prq.priority));
+#endif
+  }
+  putl = GNUNET_LOAD_get_load (datastore_put_load);
+  if ((NULL != (cp = prq.sender)) && (GNUNET_NO == prq.request_found) &&
+      ((GNUNET_YES != active_to_migration) ||
+       (putl > 2.5 * (1 + prq.priority))))
+  {
+    if (GNUNET_YES != active_to_migration)
+      putl = 1.0 + GNUNET_CRYPTO_random_u32 (GNUNET_CRYPTO_QUALITY_WEAK, 5);
+    block_time =
+        GNUNET_TIME_relative_multiply (GNUNET_TIME_UNIT_MILLISECONDS,
+                                       5000 +
+                                       GNUNET_CRYPTO_random_u32
+                                       (GNUNET_CRYPTO_QUALITY_WEAK,
+                                        (unsigned int) (60000 * putl * putl)));
+#if DEBUG_FS
+    GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, 
+               "Asking to stop migration for %llu ms because of load %f and events %d/%d\n",
+               (unsigned long long) block_time.rel_value,
+               putl,
+               active_to_migration,
+               (GNUNET_NO == prq.request_found));
+#endif
+    GSF_block_peer_migration_ (cp, GNUNET_TIME_relative_to_absolute (block_time));
+  }
+  return GNUNET_OK;
 }
 
 
@@ -178,7 +1621,39 @@ GSF_handle_p2p_content_ (const struct GNUNET_PeerIdentity *other,
 void
 GSF_pending_request_init_ ()
 {
-  // FIXME
+  unsigned long long bps;
+
+  if (GNUNET_OK !=
+      GNUNET_CONFIGURATION_get_value_number (GSF_cfg, "fs",
+                                             "MAX_PENDING_REQUESTS",
+                                             &max_pending_requests))
+  {
+    GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+                _
+                ("Configuration fails to specify `%s', assuming default value."),
+                "MAX_PENDING_REQUESTS");
+  }
+  if (GNUNET_OK !=
+      GNUNET_CONFIGURATION_get_value_size (GSF_cfg, "ats", "WAN_QUOTA_OUT",
+                                           &bps))
+  {
+    GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+                _
+                ("Configuration fails to specify `%s', assuming default value."),
+                "WAN_QUOTA_OUT");
+    bps = 65536;
+  }
+  /* queue size should be #queries we can have pending and satisfy within
+   * a carry interval: */
+  datastore_queue_size =
+      bps * GNUNET_CONSTANTS_MAX_BANDWIDTH_CARRY_S / DBLOCK_SIZE;
+
+  active_to_migration =
+      GNUNET_CONFIGURATION_get_value_yesno (GSF_cfg, "FS", "CONTENT_CACHING");
+  datastore_put_load = GNUNET_LOAD_value_init (DATASTORE_LOAD_AUTODECLINE);
+  pr_map = GNUNET_CONTAINER_multihashmap_create (32 * 1024);
+  requests_by_expiration_heap =
+      GNUNET_CONTAINER_heap_create (GNUNET_CONTAINER_HEAP_ORDER_MIN);
 }
 
 
@@ -188,7 +1663,13 @@ GSF_pending_request_init_ ()
 void
 GSF_pending_request_done_ ()
 {
-  // FIXME
+  GNUNET_CONTAINER_multihashmap_iterate (pr_map, &clean_request, NULL);
+  GNUNET_CONTAINER_multihashmap_destroy (pr_map);
+  pr_map = NULL;
+  GNUNET_CONTAINER_heap_destroy (requests_by_expiration_heap);
+  requests_by_expiration_heap = NULL;
+  GNUNET_LOAD_value_free (datastore_put_load);
+  datastore_put_load = NULL;
 }