fix
[oweals/gnunet.git] / src / fs / gnunet-service-fs.c
index 41be75e988c1822da9c298c7a61e95b999ff06ca..ccdd76de235fb1add6c72bd6d857fd95dd9d53d7 100644 (file)
  * - TTL/priority calculations are absent!
  * TODO:
  * - have non-zero preference / priority for requests we initiate!
- * - track stats for hot-path routing
  * - implement hot-path routing decision procedure
- * - implement: bound_priority, test_load_too_high, validate_nblock
- * - add content migration support (store locally) [or create new service]
+ * - implement: bound_priority, test_load_too_high
  * - statistics
  */
 #include "platform.h"
 #include "gnunet_signatures.h"
 #include "gnunet_statistics_service.h"
 #include "gnunet_util_lib.h"
-#include "gnunet-service-fs_drq.h"
 #include "gnunet-service-fs_indexing.h"
 #include "fs.h"
 
-#define DEBUG_FS GNUNET_YES
+#define DEBUG_FS GNUNET_NO
 
 /**
  * Maximum number of outgoing messages we queue per peer.
@@ -89,6 +86,11 @@ static uint64_t max_pending_requests = (32 * 1024);
  */
 struct PendingMessage;
 
+/**
+ * Our connection to the datastore.
+ */
+static struct GNUNET_DATASTORE_Handle *dsh;
+
 
 /**
  * Function called upon completion of a transmission.
@@ -472,12 +474,13 @@ struct PendingRequest
   GNUNET_PEER_Id *used_pids;
   
   /**
-   * Our entry in the DRQ (non-NULL while we wait for our
+   * Our entry in the queue (non-NULL while we wait for our
    * turn to interact with the local database).
    */
-  struct DatastoreRequestQueue *drq;
+  struct GNUNET_DATASTORE_QueueEntry *qe;
 
   /**
+
    * Size of the 'bf' (in bytes).
    */
   size_t bf_size;
@@ -545,8 +548,62 @@ struct PendingRequest
   /**
    * Remove this request after transmission of the current response.
    */
-  int do_remove;
+  int16_t do_remove;
+
+  /**
+   * GNUNET_YES if we should not forward this request to other peers.
+   */
+  int16_t local_only;
+
+};
+
+
+/**
+ * Block that is ready for migration to other peers.  Actual data is at the end of the block.
+ */
+struct MigrationReadyBlock
+{
+
+  /**
+   * This is a doubly-linked list.
+   */
+  struct MigrationReadyBlock *next;
+
+  /**
+   * This is a doubly-linked list.
+   */
+  struct MigrationReadyBlock *prev;
+
+  /**
+   * Query for the block.
+   */
+  GNUNET_HashCode query;
+
+  /**
+   * When does this block expire? 
+   */
+  struct GNUNET_TIME_Absolute expiration;
+
+  /**
+   * Peers we would consider forwarding this
+   * block to.  Zero for empty entries.
+   */
+  GNUNET_PEER_Id target_list[MIGRATION_LIST_SIZE];
+
+  /**
+   * Size of the block.
+   */
+  size_t size;
+
+  /**
+   *  Number of targets already used.
+   */
+  unsigned int used_targets;
 
+  /**
+   * Type of the block.
+   */
+  enum GNUNET_BLOCK_Type type;
 };
 
 
@@ -601,10 +658,333 @@ static struct ClientList *client_list;
  */
 static struct GNUNET_CORE_Handle *core;
 
+/**
+ * Head of linked list of blocks that can be migrated.
+ */
+static struct MigrationReadyBlock *mig_head;
+
+/**
+ * Tail of linked list of blocks that can be migrated.
+ */
+static struct MigrationReadyBlock *mig_tail;
+
+/**
+ * Request to datastore for migration (or NULL).
+ */
+static struct GNUNET_DATASTORE_QueueEntry *mig_qe;
+
+/**
+ * ID of task that collects blocks for migration.
+ */
+static GNUNET_SCHEDULER_TaskIdentifier mig_task;
+
+/**
+ * What is the maximum frequency at which we are allowed to
+ * poll the datastore for migration content?
+ */
+static struct GNUNET_TIME_Relative min_migration_delay;
+
+/**
+ * Size of the doubly-linked list of migration blocks.
+ */
+static unsigned int mig_size;
+
+/**
+ * Are we allowed to migrate content to this peer.
+ */
+static int active_migration;
+
+
+/**
+ * Transmit messages by copying it to the target buffer
+ * "buf".  "buf" will be NULL and "size" zero if the socket was closed
+ * for writing in the meantime.  In that case, do nothing
+ * (the disconnect or shutdown handler will take care of the rest).
+ * If we were able to transmit messages and there are still more
+ * pending, ask core again for further calls to this function.
+ *
+ * @param cls closure, pointer to the 'struct ConnectedPeer*'
+ * @param size number of bytes available in buf
+ * @param buf where the callee should write the message
+ * @return number of bytes written to buf
+ */
+static size_t
+transmit_to_peer (void *cls,
+                 size_t size, void *buf);
+
 
 /* ******************* clean up functions ************************ */
 
 
+/**
+ * Delete the given migration block.
+ *
+ * @param mb block to delete
+ */
+static void
+delete_migration_block (struct MigrationReadyBlock *mb)
+{
+  GNUNET_CONTAINER_DLL_remove (mig_head,
+                              mig_tail,
+                              mb);
+  GNUNET_PEER_decrement_rcs (mb->target_list,
+                            MIGRATION_LIST_SIZE);
+  mig_size--;
+  GNUNET_free (mb);
+}
+
+
+/**
+ * Compare the distance of two peers to a key.
+ *
+ * @param key key
+ * @param p1 first peer
+ * @param p2 second peer
+ * @return GNUNET_YES if P1 is closer to key than P2
+ */
+static int
+is_closer (const GNUNET_HashCode *key,
+          const struct GNUNET_PeerIdentity *p1,
+          const struct GNUNET_PeerIdentity *p2)
+{
+  return GNUNET_CRYPTO_hash_xorcmp (&p1->hashPubKey,
+                                   &p2->hashPubKey,
+                                   key);
+}
+
+
+/**
+ * Consider migrating content to a given peer.
+ *
+ * @param cls 'struct MigrationReadyBlock*' to select
+ *            targets for (or NULL for none)
+ * @param key ID of the peer 
+ * @param value 'struct ConnectedPeer' of the peer
+ * @return GNUNET_YES (always continue iteration)
+ */
+static int
+consider_migration (void *cls,
+                   const GNUNET_HashCode *key,
+                   void *value)
+{
+  struct MigrationReadyBlock *mb = cls;
+  struct ConnectedPeer *cp = value;
+  struct MigrationReadyBlock *pos;
+  struct GNUNET_PeerIdentity cppid;
+  struct GNUNET_PeerIdentity otherpid;
+  struct GNUNET_PeerIdentity worstpid;
+  size_t msize;
+  unsigned int i;
+  unsigned int repl;
+  
+  /* consider 'cp' as a migration target for mb */
+  if (mb != NULL)
+    {
+      GNUNET_PEER_resolve (cp->pid,
+                          &cppid);
+      repl = MIGRATION_LIST_SIZE;
+      for (i=0;i<MIGRATION_LIST_SIZE;i++)
+       {
+         if (mb->target_list[i] == 0)
+           {
+             mb->target_list[i] = cp->pid;
+             GNUNET_PEER_change_rc (mb->target_list[i], 1);
+             repl = MIGRATION_LIST_SIZE;
+             break;
+           }
+         GNUNET_PEER_resolve (mb->target_list[i],
+                              &otherpid);
+         if ( (repl == MIGRATION_LIST_SIZE) &&
+              is_closer (&mb->query,
+                         &cppid,
+                         &otherpid)) 
+           {
+             repl = i;
+             worstpid = otherpid;
+           }
+         else if ( (repl != MIGRATION_LIST_SIZE) &&
+                   (is_closer (&mb->query,
+                               &worstpid,
+                               &otherpid) ) )
+           {
+             repl = i;
+             worstpid = otherpid;
+           }       
+       }
+      if (repl != MIGRATION_LIST_SIZE) 
+       {
+         GNUNET_PEER_change_rc (mb->target_list[repl], -1);
+         mb->target_list[repl] = cp->pid;
+         GNUNET_PEER_change_rc (mb->target_list[repl], 1);
+       }
+    }
+
+  /* consider scheduling transmission to cp for content migration */
+  if (cp->cth != NULL)
+    return GNUNET_YES; 
+  msize = 0;
+  pos = mig_head;
+  while (pos != NULL)
+    {
+      for (i=0;i<MIGRATION_LIST_SIZE;i++)
+       {
+         if (cp->pid == pos->target_list[i])
+           {
+             if (msize == 0)
+               msize = pos->size;
+             else
+               msize = GNUNET_MIN (msize,
+                                   pos->size);
+             break;
+           }
+       }
+      pos = pos->next;
+    }
+  if (msize == 0)
+    return GNUNET_YES; /* no content available */
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+             "Trying to migrate at least %u bytes to peer `%s'\n",
+             msize,
+             GNUNET_h2s (key));
+#endif
+  cp->cth 
+    = GNUNET_CORE_notify_transmit_ready (core,
+                                        0, GNUNET_TIME_UNIT_FOREVER_REL,
+                                        (const struct GNUNET_PeerIdentity*) key,
+                                        msize + sizeof (struct PutMessage),
+                                        &transmit_to_peer,
+                                        cp);
+  return GNUNET_YES;
+}
+
+
+/**
+ * Task that is run periodically to obtain blocks for content
+ * migration
+ * 
+ * @param cls unused
+ * @param tc scheduler context (also unused)
+ */
+static void
+gather_migration_blocks (void *cls,
+                        const struct GNUNET_SCHEDULER_TaskContext *tc);
+
+
+/**
+ * If the migration task is not currently running, consider
+ * (re)scheduling it with the appropriate delay.
+ */
+static void
+consider_migration_gathering ()
+{
+  struct GNUNET_TIME_Relative delay;
+
+  if (mig_qe != NULL)
+    return;
+  if (mig_task != GNUNET_SCHEDULER_NO_TASK)
+    return;
+  delay = GNUNET_TIME_relative_multiply (GNUNET_TIME_UNIT_SECONDS,
+                                        mig_size);
+  delay = GNUNET_TIME_relative_divide (delay,
+                                      MAX_MIGRATION_QUEUE);
+  delay = GNUNET_TIME_relative_max (delay,
+                                   min_migration_delay);
+  mig_task = GNUNET_SCHEDULER_add_delayed (sched,
+                                          delay,
+                                          &gather_migration_blocks,
+                                          NULL);
+}
+
+
+/**
+ * Process content offered for migration.
+ *
+ * @param cls closure
+ * @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_migration_content (void *cls,
+                          const GNUNET_HashCode * key,
+                          uint32_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 MigrationReadyBlock *mb;
+  
+  if (key == NULL)
+    {
+      mig_qe = NULL;
+      if (mig_size < MAX_MIGRATION_QUEUE)  
+       consider_migration_gathering ();
+      return;
+    }
+  if (type == GNUNET_BLOCK_TYPE_ONDEMAND)
+    {
+      if (GNUNET_OK !=
+         GNUNET_FS_handle_on_demand_block (key, size, data,
+                                           type, priority, anonymity,
+                                           expiration, uid, 
+                                           &process_migration_content,
+                                           NULL))
+       GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
+      return;
+    }
+#if DEBUG_FS
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+             "Retrieved block `%s' of type %u for migration\n",
+             GNUNET_h2s (key),
+             type);
+#endif
+  mb = GNUNET_malloc (sizeof (struct MigrationReadyBlock) + size);
+  mb->query = *key;
+  mb->expiration = expiration;
+  mb->size = size;
+  mb->type = type;
+  memcpy (&mb[1], data, size);
+  GNUNET_CONTAINER_DLL_insert_after (mig_head,
+                                    mig_tail,
+                                    mig_tail,
+                                    mb);
+  mig_size++;
+  GNUNET_CONTAINER_multihashmap_iterate (connected_peers,
+                                        &consider_migration,
+                                        mb);
+  GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
+}
+
+
+/**
+ * Task that is run periodically to obtain blocks for content
+ * migration
+ * 
+ * @param cls unused
+ * @param tc scheduler context (also unused)
+ */
+static void
+gather_migration_blocks (void *cls,
+                        const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+  mig_task = GNUNET_SCHEDULER_NO_TASK;
+  mig_qe = GNUNET_DATASTORE_get_random (dsh, 0, -1,
+                                       GNUNET_TIME_UNIT_FOREVER_REL,
+                                       &process_migration_content, NULL);
+  GNUNET_assert (mig_qe != NULL);
+}
+
+
 /**
  * We're done with a particular message list entry.
  * Free all associated resources.
@@ -687,10 +1067,10 @@ destroy_pending_request (struct PendingRequest *pr)
   (void) GNUNET_CONTAINER_multihashmap_remove (query_request_map,
                                               &pr->query,
                                               pr);
-  if (pr->drq != NULL)
-    {
-      GNUNET_FS_drq_get_cancel (pr->drq);
-      pr->drq = NULL;
+  if (pr->qe != NULL)
+     {
+      GNUNET_DATASTORE_cancel (pr->qe);
+      pr->qe = NULL;
     }
   if (pr->client_request_list != NULL)
     {
@@ -761,16 +1141,26 @@ peer_connect_handler (void *cls,
                      uint32_t distance)
 {
   struct ConnectedPeer *cp;
-
+  struct MigrationReadyBlock *pos;
+  
   cp = GNUNET_malloc (sizeof (struct ConnectedPeer));
   cp->pid = GNUNET_PEER_intern (peer);
-  GNUNET_CONTAINER_multihashmap_put (connected_peers,
-                                    &peer->hashPubKey,
-                                    cp,
-                                    GNUNET_CONTAINER_MULTIHASHMAPOPTION_UNIQUE_ONLY);
+  GNUNET_break (GNUNET_OK ==
+               GNUNET_CONTAINER_multihashmap_put (connected_peers,
+                                                  &peer->hashPubKey,
+                                                  cp,
+                                                  GNUNET_CONTAINER_MULTIHASHMAPOPTION_UNIQUE_ONLY));
+
+  pos = mig_head;
+  while (NULL != pos)
+    {
+      (void) consider_migration (pos, &peer->hashPubKey, cp);
+      pos = pos->next;
+    }
 }
 
 
+
 /**
  * Free (each) request made by the peer.
  *
@@ -787,9 +1177,10 @@ destroy_request (void *cls,
   const struct GNUNET_PeerIdentity * peer = cls;
   struct PendingRequest *pr = value;
   
-  GNUNET_CONTAINER_multihashmap_remove (peer_request_map,
-                                       &peer->hashPubKey,
-                                       pr);
+  GNUNET_break (GNUNET_YES ==
+               GNUNET_CONTAINER_multihashmap_remove (peer_request_map,
+                                                     &peer->hashPubKey,
+                                                     pr));
   destroy_pending_request (pr);
   return GNUNET_YES;
 }
@@ -809,6 +1200,8 @@ peer_disconnect_handler (void *cls,
   struct ConnectedPeer *cp;
   struct PendingMessage *pm;
   unsigned int i;
+  struct MigrationReadyBlock *pos;
+  struct MigrationReadyBlock *next;
 
   GNUNET_CONTAINER_multihashmap_get_multiple (peer_request_map,
                                              &peer->hashPubKey,
@@ -826,9 +1219,35 @@ peer_disconnect_handler (void *cls,
          cp->last_client_replies[i] = NULL;
        }
     }
-  GNUNET_CONTAINER_multihashmap_remove (connected_peers,
-                                       &peer->hashPubKey,
-                                       cp);
+  GNUNET_break (GNUNET_YES ==
+               GNUNET_CONTAINER_multihashmap_remove (connected_peers,
+                                                     &peer->hashPubKey,
+                                                     cp));
+  /* remove this peer from migration considerations; schedule
+     alternatives */
+  next = mig_head;
+  while (NULL != (pos = next))
+    {
+      next = pos->next;
+      for (i=0;i<MIGRATION_LIST_SIZE;i++)
+       {
+         if (pos->target_list[i] == cp->pid)
+           {
+             GNUNET_PEER_change_rc (pos->target_list[i], -1);
+             pos->target_list[i] = 0;
+            }
+         }
+      if (pos->used_targets >= GNUNET_CONTAINER_multihashmap_size (connected_peers))
+       {
+         delete_migration_block (pos);
+         consider_migration_gathering ();
+          continue;
+       }
+      GNUNET_CONTAINER_multihashmap_iterate (connected_peers,
+                                            &consider_migration,
+                                            pos);
+    }
+
   GNUNET_PEER_change_rc (cp->pid, -1);
   GNUNET_PEER_decrement_rcs (cp->last_p2p_replies, P2P_SUCCESS_LIST_SIZE);
   if (NULL != cp->cth)
@@ -958,6 +1377,16 @@ static void
 shutdown_task (void *cls,
               const struct GNUNET_SCHEDULER_TaskContext *tc)
 {
+  if (mig_qe != NULL)
+    {
+      GNUNET_DATASTORE_cancel (mig_qe);
+      mig_qe = NULL;
+    }
+  if (GNUNET_SCHEDULER_NO_TASK != mig_task)
+    {
+      GNUNET_SCHEDULER_cancel (sched, mig_task);
+      mig_task = GNUNET_SCHEDULER_NO_TASK;
+    }
   while (client_list != NULL)
     handle_client_disconnect (NULL,
                              client_list->client);
@@ -983,6 +1412,12 @@ shutdown_task (void *cls,
       GNUNET_STATISTICS_destroy (stats, GNUNET_NO);
       stats = NULL;
     }
+  GNUNET_DATASTORE_disconnect (dsh,
+                              GNUNET_NO);
+  while (mig_head != NULL)
+    delete_migration_block (mig_head);
+  GNUNET_assert (0 == mig_size);
+  dsh = NULL;
   sched = NULL;
   cfg = NULL;  
 }
@@ -992,7 +1427,7 @@ shutdown_task (void *cls,
 
 
 /**
- * Transmit the given message by copying it to the target buffer
+ * Transmit messages by copying it to the target buffer
  * "buf".  "buf" will be NULL and "size" zero if the socket was closed
  * for writing in the meantime.  In that case, do nothing
  * (the disconnect or shutdown handler will take care of the rest).
@@ -1012,7 +1447,11 @@ transmit_to_peer (void *cls,
   char *cbuf = buf;
   struct GNUNET_PeerIdentity pid;
   struct PendingMessage *pm;
+  struct MigrationReadyBlock *mb;
+  struct MigrationReadyBlock *next;
+  struct PutMessage migm;
   size_t msize;
+  unsigned int i;
  
   cp->cth = NULL;
   if (NULL == buf)
@@ -1044,9 +1483,64 @@ transmit_to_peer (void *cls,
                                                   &transmit_to_peer,
                                                   cp);
     }
-#if DEBUG_FS > 2
+  else
+    {      
+      next = mig_head;
+      while (NULL != (mb = next))
+       {
+         next = mb->next;
+         for (i=0;i<MIGRATION_LIST_SIZE;i++)
+           {
+             if ( (cp->pid == mb->target_list[i]) &&
+                  (mb->size + sizeof (migm) <= size) )
+               {
+                 GNUNET_PEER_change_rc (mb->target_list[i], -1);
+                 mb->target_list[i] = 0;
+                 mb->used_targets++;
+                 migm.header.size = htons (sizeof (migm) + mb->size);
+                 migm.header.type = htons (GNUNET_MESSAGE_TYPE_FS_PUT);
+                 migm.type = htonl (mb->type);
+                 migm.expiration = GNUNET_TIME_absolute_hton (mb->expiration);
+                 memcpy (&cbuf[msize], &migm, sizeof (migm));
+                 msize += sizeof (migm);
+                 size -= sizeof (migm);
+                 memcpy (&cbuf[msize], &mb[1], mb->size);
+                 msize += mb->size;
+                 size -= mb->size;
+#if DEBUG_FS
+                 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                             "Pushing migration block `%s' (%u bytes) to `%s'\n",
+                             GNUNET_h2s (&mb->query),
+                             mb->size,
+                             GNUNET_i2s (&pid));
+#endif   
+                 break;
+               }
+             else
+               {
+#if DEBUG_FS
+                 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                             "Migration block `%s' (%u bytes) is not on migration list for peer `%s'\n",
+                             GNUNET_h2s (&mb->query),
+                             mb->size,
+                             GNUNET_i2s (&pid));
+#endif   
+               }
+           }
+         if ( (mb->used_targets >= MIGRATION_TARGET_COUNT) ||
+              (mb->used_targets >= GNUNET_CONTAINER_multihashmap_size (connected_peers)) )
+           {
+             delete_migration_block (mb);
+             consider_migration_gathering ();
+           }
+       }
+      consider_migration (NULL, 
+                         &pid.hashPubKey,
+                         cp);
+    }
+#if DEBUG_FS
   GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
-             "Transmitting %u bytes to peer %u.\n",
+             "Transmitting %u bytes to peer %u\n",
              msize,
              cp->pid);
 #endif
@@ -1091,25 +1585,27 @@ add_to_pending_messages_for_peer (struct ConnectedPeer *cp,
   cp->pending_requests++;
   if (cp->pending_requests > MAX_QUEUE_PER_PEER)
     destroy_pending_message (cp->pending_messages_tail, 0);  
-  if (cp->cth == NULL)
-    {
-      /* need to schedule transmission */
-      GNUNET_PEER_resolve (cp->pid, &pid);
-      cp->cth = GNUNET_CORE_notify_transmit_ready (core,
-                                                  cp->pending_messages_head->priority,
-                                                  MAX_TRANSMIT_DELAY,
-                                                  &pid,
-                                                  cp->pending_messages_head->msize,
-                                                  &transmit_to_peer,
-                                                  cp);
-    }
+  GNUNET_PEER_resolve (cp->pid, &pid);
+  if (NULL != cp->cth)
+    GNUNET_CORE_notify_transmit_ready_cancel (cp->cth);
+  /* need to schedule transmission */
+  cp->cth = GNUNET_CORE_notify_transmit_ready (core,
+                                              cp->pending_messages_head->priority,
+                                              MAX_TRANSMIT_DELAY,
+                                              &pid,
+                                              cp->pending_messages_head->msize,
+                                              &transmit_to_peer,
+                                              cp);
   if (cp->cth == NULL)
     {
 #if DEBUG_FS
       GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
                  "Failed to schedule transmission with core!\n");
 #endif
-      /* FIXME: call stats (rare, bad case) */
+      GNUNET_STATISTICS_update (stats,
+                               gettext_noop ("# CORE transmission failures"),
+                               1,
+                               GNUNET_NO);
     }
 }
 
@@ -1600,6 +2096,8 @@ forward_request_task (void *cls,
 #endif
       return; /* already pending */
     }
+  if (GNUNET_YES == pr->local_only)
+    return; /* configured to not do P2P search */
   /* (1) select target */
   psc.pr = pr;
   psc.target_score = DBL_MIN;
@@ -1653,7 +2151,7 @@ forward_request_task (void *cls,
   pr->irc = GNUNET_CORE_peer_change_preference (sched, cfg,
                                                &psc.target,
                                                GNUNET_CONSTANTS_SERVICE_TIMEOUT, 
-                                               GNUNET_BANDWIDTH_value_init ((uint32_t) -1 /* no limit */), 
+                                               GNUNET_BANDWIDTH_value_init (UINT32_MAX),
                                                DBLOCK_SIZE * 2, 
                                                (uint64_t) cp->inc_preference,
                                                &target_reservation_cb,
@@ -1764,7 +2262,10 @@ struct ProcessReplyClosure
    */
   const void *data;
 
-  // FIXME: add 'struct ConnectedPeer' to track 'last_xxx_replies' here!
+  /**
+   * Who gave us this reply? NULL for local host.
+   */
+  struct ConnectedPeer *sender;
 
   /**
    * When the reply expires.
@@ -1791,6 +2292,11 @@ struct ProcessReplyClosure
    * How much was this reply worth to us?
    */
   uint32_t priority;
+
+  /**
+   * Did we finish processing the associated request?
+   */ 
+  int finished;
 };
 
 
@@ -1814,6 +2320,7 @@ process_reply (void *cls,
   struct ClientList *cl;
   struct PutMessage *pm;
   struct ConnectedPeer *cp;
+  struct GNUNET_TIME_Relative cur_delay;
   GNUNET_HashCode chash;
   GNUNET_HashCode mhash;
   size_t msize;
@@ -1828,6 +2335,39 @@ process_reply (void *cls,
                            gettext_noop ("# replies received and matched"),
                            1,
                            GNUNET_NO);
+  if (prq->sender != NULL)
+    {
+      /* FIXME: should we be more precise here and not use
+        "start_time" but a peer-specific time stamp? */
+      cur_delay = GNUNET_TIME_absolute_get_duration (pr->start_time);
+      prq->sender->avg_delay.value
+       = (prq->sender->avg_delay.value * 
+          (RUNAVG_DELAY_N - 1) + cur_delay.value) / RUNAVG_DELAY_N; 
+      prq->sender->avg_priority
+       = (prq->sender->avg_priority * 
+          (RUNAVG_DELAY_N - 1) + pr->priority) / (double) RUNAVG_DELAY_N;
+      if (pr->cp != NULL)
+       {
+         GNUNET_PEER_change_rc (prq->sender->last_p2p_replies
+                                [prq->sender->last_p2p_replies_woff % P2P_SUCCESS_LIST_SIZE], 
+                                -1);
+         GNUNET_PEER_change_rc (pr->cp->pid, 1);
+         prq->sender->last_p2p_replies
+           [(prq->sender->last_p2p_replies_woff++) % P2P_SUCCESS_LIST_SIZE]
+           = pr->cp->pid;
+       }
+      else
+       {
+         if (NULL != prq->sender->last_client_replies
+             [(prq->sender->last_client_replies_woff) % CS2P_SUCCESS_LIST_SIZE])
+           GNUNET_SERVER_client_drop (prq->sender->last_client_replies
+                                      [(prq->sender->last_client_replies_woff) % CS2P_SUCCESS_LIST_SIZE]);
+         prq->sender->last_client_replies
+           [(prq->sender->last_client_replies_woff++) % CS2P_SUCCESS_LIST_SIZE]
+           = pr->client_request_list->client_list->client;
+         GNUNET_SERVER_client_keep (pr->client_request_list->client_list->client);
+       }
+    }
   GNUNET_CRYPTO_hash (prq->data,
                      prq->size,
                      &chash);
@@ -1838,13 +2378,13 @@ process_reply (void *cls,
       /* only possible reply, stop requesting! */
       while (NULL != pr->pending_head)
        destroy_pending_message_list_entry (pr->pending_head);
-      if (pr->drq != NULL)
+      if (pr->qe != NULL)
        {
          if (pr->client_request_list != NULL)
            GNUNET_SERVER_receive_done (pr->client_request_list->client_list->client, 
                                        GNUNET_YES);
-         GNUNET_FS_drq_get_cancel (pr->drq);
-         pr->drq = NULL;
+         GNUNET_DATASTORE_cancel (pr->qe);
+         pr->qe = NULL;
        }
       pr->do_remove = GNUNET_YES;
       if (pr->task != GNUNET_SCHEDULER_NO_TASK)
@@ -1918,7 +2458,7 @@ process_reply (void *cls,
     }
   prq->priority += pr->remaining_priority;
   pr->remaining_priority = 0;
-  if (pr->client_request_list != NULL)
+  if (NULL != pr->client_request_list)
     {
       GNUNET_STATISTICS_update (stats,
                                gettext_noop ("# replies received for local clients"),
@@ -1954,7 +2494,10 @@ process_reply (void *cls,
        }
       GNUNET_break (cl->th != NULL);
       if (pr->do_remove)               
-       destroy_pending_request (pr);           
+       {
+         prq->finished = GNUNET_YES;
+         destroy_pending_request (pr);         
+       }
     }
   else
     {
@@ -1974,7 +2517,7 @@ process_reply (void *cls,
       reply->cont = &transmit_reply_continuation;
       reply->cont_cls = pr;
       reply->msize = msize;
-      reply->priority = (uint32_t) -1; /* send replies first! */
+      reply->priority = UINT32_MAX; /* send replies first! */
       pm = (struct PutMessage*) &reply[1];
       pm->header.type = htons (GNUNET_MESSAGE_TYPE_FS_PUT);
       pm->header.size = htons (msize);
@@ -1983,11 +2526,27 @@ process_reply (void *cls,
       memcpy (&pm[1], prq->data, prq->size);
       add_to_pending_messages_for_peer (cp, reply, pr);
     }
-  // FIXME: implement hot-path routing statistics keeping!
   return GNUNET_YES;
 }
 
 
+/**
+ * Continuation called to notify client about result of the
+ * operation.
+ *
+ * @param cls closure
+ * @param success GNUNET_SYSERR on failure
+ * @param msg NULL on success, otherwise an error message
+ */
+static void 
+put_migration_continuation (void *cls,
+                           int success,
+                           const char *msg)
+{
+  /* FIXME */
+}
+
+
 /**
  * Handle P2P "PUT" message.
  *
@@ -2058,17 +2617,35 @@ handle_p2p_put (void *cls,
                            GNUNET_NO);
   /* now, lookup 'query' */
   prq.data = (const void*) &put[1];
+  if (other != NULL)
+    prq.sender = GNUNET_CONTAINER_multihashmap_get (connected_peers,
+                                                   &other->hashPubKey);
   prq.size = dsize;
   prq.type = type;
   prq.expiration = expiration;
   prq.priority = 0;
+  prq.finished = GNUNET_NO;
   GNUNET_CONTAINER_multihashmap_get_multiple (query_request_map,
                                              &query,
                                              &process_reply,
                                              &prq);
-  // FIXME: if migration is on and load is low,
-  // queue to store data in datastore;
-  // use "prq.priority" for that!
+  if (GNUNET_YES == active_migration)
+    {
+#if DEBUG_FS
+      GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                 "Replicating result for query `%s' with priority %u\n",
+                 GNUNET_h2s (&query),
+                 prq.priority);
+#endif
+      GNUNET_DATASTORE_put (dsh,
+                           0, &query, dsize, &put[1],
+                           type, prq.priority, 1 /* anonymity */, 
+                           expiration, 
+                           1 + prq.priority, MAX_DATASTORE_QUEUE,
+                           GNUNET_CONSTANTS_SERVICE_TIMEOUT,
+                           &put_migration_continuation, 
+                           NULL);
+    }
   return GNUNET_OK;
 }
 
@@ -2169,7 +2746,7 @@ process_local_reply (void *cls,
       GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
                  "Done processing local replies, forwarding request to other peers.\n");
 #endif
-      pr->drq = NULL;
+      pr->qe = NULL;
       if (pr->client_request_list != NULL)
        {
          GNUNET_SERVER_receive_done (pr->client_request_list->client_list->client, 
@@ -2223,7 +2800,8 @@ process_local_reply (void *cls,
                                            anonymity, expiration, uid, 
                                            &process_local_reply,
                                            pr))
-       GNUNET_FS_drq_get_next (GNUNET_YES);
+      if (pr->qe != NULL)
+       GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
       return;
     }
   /* check for duplicates */
@@ -2244,7 +2822,8 @@ process_local_reply (void *cls,
                                gettext_noop ("# results filtered by query bloomfilter"),
                                1,
                                GNUNET_NO);
-      GNUNET_FS_drq_get_next (GNUNET_YES);
+      if (pr->qe != NULL)
+       GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
       return;
     }
 #if DEBUG_FS
@@ -2274,18 +2853,27 @@ process_local_reply (void *cls,
                                             &query))
     {
       GNUNET_break (0);
-      /* FIXME: consider removing the block? */
-      GNUNET_FS_drq_get_next (GNUNET_YES);
+      GNUNET_DATASTORE_remove (dsh,
+                              key,
+                              size, data,
+                              -1, -1, 
+                              GNUNET_TIME_UNIT_FOREVER_REL,
+                              NULL, NULL);
+      GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
       return;
     }
   prq.type = type;
   prq.priority = priority;  
+  prq.finished = GNUNET_NO;
   process_reply (&prq, key, pr);
-
+  if (prq.finished == GNUNET_YES)
+    return;
+  if (pr->qe == NULL)
+    return; /* done here */
   if ( (type == GNUNET_BLOCK_TYPE_DBLOCK) ||
        (type == GNUNET_BLOCK_TYPE_IBLOCK) ) 
     {
-      GNUNET_FS_drq_get_next (GNUNET_NO);
+      GNUNET_DATASTORE_get_next (dsh, GNUNET_NO);
       return;
     }
   if ( (pr->client_request_list == NULL) &&
@@ -2300,10 +2888,10 @@ process_local_reply (void *cls,
                                gettext_noop ("# processing result set cut short due to load"),
                                1,
                                GNUNET_NO);
-      GNUNET_FS_drq_get_next (GNUNET_NO);
+      GNUNET_DATASTORE_get_next (dsh, GNUNET_NO);
       return;
     }
-  GNUNET_FS_drq_get_next (GNUNET_YES);
+  GNUNET_DATASTORE_get_next (dsh, GNUNET_YES);
 }
 
 
@@ -2496,11 +3084,12 @@ handle_p2p_get (void *cls,
   pr = GNUNET_malloc (sizeof (struct PendingRequest) + 
                      (have_ns ? sizeof(GNUNET_HashCode) : 0));
   if (have_ns)
-    pr->namespace = (GNUNET_HashCode*) &pr[1];
+    {
+      pr->namespace = (GNUNET_HashCode*) &pr[1];
+      memcpy (&pr[1], &opt[bits++], sizeof (GNUNET_HashCode));
+    }
   pr->type = type;
   pr->mingle = ntohl (gm->filter_mutator);
-  if (0 != (bm & GET_MESSAGE_BIT_SKS_NAMESPACE))    
-    memcpy (&pr[1], &opt[bits++], sizeof (GNUNET_HashCode));
   if (0 != (bm & GET_MESSAGE_BIT_TRANSMIT_TO))
     pr->target_pid = GNUNET_PEER_intern ((const struct GNUNET_PeerIdentity*) &opt[bits++]);
 
@@ -2526,7 +3115,7 @@ handle_p2p_get (void *cls,
                                gettext_noop ("# requests dropped due TTL underflow"),
                                1,
                                GNUNET_NO);
-      /* integer underflow => drop (should be very rare)! */
+      /* integer underflow => drop (should be very rare)! */      
       GNUNET_free (pr);
       return GNUNET_OK;
     } 
@@ -2581,14 +3170,16 @@ handle_p2p_get (void *cls,
     }
 
   pr->cp = cp;
-  GNUNET_CONTAINER_multihashmap_put (query_request_map,
-                                    &gm->query,
-                                    pr,
-                                    GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE);
-  GNUNET_CONTAINER_multihashmap_put (peer_request_map,
-                                    &other->hashPubKey,
-                                    pr,
-                                    GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE);
+  GNUNET_break (GNUNET_OK ==
+               GNUNET_CONTAINER_multihashmap_put (query_request_map,
+                                                  &gm->query,
+                                                  pr,
+                                                  GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE));
+  GNUNET_break (GNUNET_OK ==
+               GNUNET_CONTAINER_multihashmap_put (peer_request_map,
+                                                  &other->hashPubKey,
+                                                  pr,
+                                                  GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE));
   
   pr->hnode = GNUNET_CONTAINER_heap_insert (requests_by_expiration_heap,
                                            pr,
@@ -2614,12 +3205,14 @@ handle_p2p_get (void *cls,
     type = GNUNET_BLOCK_TYPE_ANY; /* to get on-demand as well */
   timeout = GNUNET_TIME_relative_multiply (BASIC_DATASTORE_REQUEST_DELAY,
                                           (pr->priority + 1)); 
-  pr->drq = GNUNET_FS_drq_get (&gm->query,
-                              type,                           
-                              &process_local_reply,
-                              pr,
-                              timeout,
-                              GNUNET_NO);
+  pr->qe = GNUNET_DATASTORE_get (dsh,
+                                &gm->query,
+                                type,                         
+                                pr->priority + 1,
+                                MAX_DATASTORE_QUEUE,                            
+                                timeout,
+                                &process_local_reply,
+                                pr);
 
   /* Are multiple results possible?  If so, start processing remotely now! */
   switch (pr->type)
@@ -2783,6 +3376,10 @@ handle_start_search (void *cls,
   pr->anonymity_level = ntohl (sm->anonymity_level); 
   refresh_bloomfilter (pr);
   pr->query = sm->query;
+  if (0 == (1 & ntohl (sm->options)))
+    pr->local_only = GNUNET_NO;
+  else
+    pr->local_only = GNUNET_YES;
   switch (type)
     {
     case GNUNET_BLOCK_TYPE_DBLOCK:
@@ -2799,55 +3396,25 @@ handle_start_search (void *cls,
     default:
       break;
     }
-  GNUNET_CONTAINER_multihashmap_put (query_request_map,
-                                    &sm->query,
-                                    pr,
-                                    GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE);
+  GNUNET_break (GNUNET_OK ==
+               GNUNET_CONTAINER_multihashmap_put (query_request_map,
+                                                  &sm->query,
+                                                  pr,
+                                                  GNUNET_CONTAINER_MULTIHASHMAPOPTION_MULTIPLE));
   if (type == GNUNET_BLOCK_TYPE_DBLOCK)
     type = GNUNET_BLOCK_TYPE_ANY; /* get on-demand blocks too! */
-  pr->drq = GNUNET_FS_drq_get (&sm->query,
-                              type,                           
-                              &process_local_reply,
-                              pr,
-                              GNUNET_CONSTANTS_SERVICE_TIMEOUT,
-                              GNUNET_YES);
+  pr->qe = GNUNET_DATASTORE_get (dsh,
+                                &sm->query,
+                                type,
+                                -3, -1,
+                                GNUNET_CONSTANTS_SERVICE_TIMEOUT,                             
+                                &process_local_reply,
+                                pr);
 }
 
 
 /* **************************** Startup ************************ */
 
-
-/**
- * List of handlers for P2P messages
- * that we care about.
- */
-static struct GNUNET_CORE_MessageHandler p2p_handlers[] =
-  {
-    { &handle_p2p_get, 
-      GNUNET_MESSAGE_TYPE_FS_GET, 0 },
-    { &handle_p2p_put, 
-      GNUNET_MESSAGE_TYPE_FS_PUT, 0 },
-    { NULL, 0, 0 }
-  };
-
-
-/**
- * List of handlers for the messages understood by this
- * service.
- */
-static struct GNUNET_SERVER_MessageHandler handlers[] = {
-  {&GNUNET_FS_handle_index_start, NULL, 
-   GNUNET_MESSAGE_TYPE_FS_INDEX_START, 0},
-  {&GNUNET_FS_handle_index_list_get, NULL, 
-   GNUNET_MESSAGE_TYPE_FS_INDEX_LIST_GET, sizeof(struct GNUNET_MessageHeader) },
-  {&GNUNET_FS_handle_unindex, NULL, GNUNET_MESSAGE_TYPE_FS_UNINDEX, 
-   sizeof (struct UnindexMessage) },
-  {&handle_start_search, NULL, GNUNET_MESSAGE_TYPE_FS_START_SEARCH, 
-   0 },
-  {NULL, NULL, 0, 0}
-};
-
-
 /**
  * Process fs requests.
  *
@@ -2860,9 +3427,30 @@ main_init (struct GNUNET_SCHEDULER_Handle *s,
           struct GNUNET_SERVER_Handle *server,
           const struct GNUNET_CONFIGURATION_Handle *c)
 {
+  static const struct GNUNET_CORE_MessageHandler p2p_handlers[] =
+    {
+      { &handle_p2p_get, 
+       GNUNET_MESSAGE_TYPE_FS_GET, 0 },
+      { &handle_p2p_put, 
+       GNUNET_MESSAGE_TYPE_FS_PUT, 0 },
+      { NULL, 0, 0 }
+    };
+  static const struct GNUNET_SERVER_MessageHandler handlers[] = {
+    {&GNUNET_FS_handle_index_start, NULL, 
+     GNUNET_MESSAGE_TYPE_FS_INDEX_START, 0},
+    {&GNUNET_FS_handle_index_list_get, NULL, 
+     GNUNET_MESSAGE_TYPE_FS_INDEX_LIST_GET, sizeof(struct GNUNET_MessageHeader) },
+    {&GNUNET_FS_handle_unindex, NULL, GNUNET_MESSAGE_TYPE_FS_UNINDEX, 
+     sizeof (struct UnindexMessage) },
+    {&handle_start_search, NULL, GNUNET_MESSAGE_TYPE_FS_START_SEARCH, 
+     0 },
+    {NULL, NULL, 0, 0}
+  };
+
   sched = s;
   cfg = c;
   stats = GNUNET_STATISTICS_create (sched, "fs", cfg);
+  min_migration_delay = GNUNET_TIME_UNIT_SECONDS; // FIXME: get from config
   connected_peers = GNUNET_CONTAINER_multihashmap_create (128); // FIXME: get size from config
   query_request_map = GNUNET_CONTAINER_multihashmap_create (128); // FIXME: get size from config
   peer_request_map = GNUNET_CONTAINER_multihashmap_create (128); // FIXME: get size from config
@@ -2890,9 +3478,20 @@ main_init (struct GNUNET_SCHEDULER_Handle *s,
       requests_by_expiration_heap = NULL;
       GNUNET_CONTAINER_multihashmap_destroy (peer_request_map);
       peer_request_map = NULL;
-
+      if (dsh != NULL)
+       {
+         GNUNET_DATASTORE_disconnect (dsh, GNUNET_NO);
+         dsh = NULL;
+       }
       return GNUNET_SYSERR;
-    }  
+    }
+  /* FIXME: distinguish between sending and storing in options? */
+  if (active_migration) 
+    {
+      GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+                 _("Content migration is enabled, will start to gather data\n"));
+      consider_migration_gathering ();
+    }
   GNUNET_SERVER_disconnect_notify (server, 
                                   &handle_client_disconnect,
                                   NULL);
@@ -2919,11 +3518,22 @@ run (void *cls,
      struct GNUNET_SERVER_Handle *server,
      const struct GNUNET_CONFIGURATION_Handle *cfg)
 {
-  if ( (GNUNET_OK != GNUNET_FS_drq_init (sched, cfg)) ||
-       (GNUNET_OK != GNUNET_FS_indexing_init (sched, cfg)) ||
+  active_migration = GNUNET_CONFIGURATION_get_value_yesno (cfg,
+                                                          "FS",
+                                                          "ACTIVEMIGRATION");
+  dsh = GNUNET_DATASTORE_connect (cfg,
+                                 sched);
+  if (dsh == NULL)
+    {
+      GNUNET_SCHEDULER_shutdown (sched);
+      return;
+    }
+  if ( (GNUNET_OK != GNUNET_FS_indexing_init (sched, cfg, dsh)) ||
        (GNUNET_OK != main_init (sched, server, cfg)) )
     {    
       GNUNET_SCHEDULER_shutdown (sched);
+      GNUNET_DATASTORE_disconnect (dsh, GNUNET_NO);
+      dsh = NULL;
       return;   
     }
 }