lib: Demultiplex the fast path

The fast path will now use an incoming ring buffer per flow per
process. This necessitated the development of a new method for the
asynchronous io call, which is now based on an event queue system for
scalability (fqueue). The ipcpd's and tools have been updated to this
API.

1 files changed, 329 insertions, 171 deletions

diff --git a/src/lib/dev.c b/src/lib/dev.c
index 77c2d06a..f735e72b 100644
--- a/src/lib/dev.c
+++ b/src/lib/dev.c
@@ -3,7 +3,8 @@
  *
  * API for applications
  *
- *    Sander Vrijders <[email protected]>
+ *    Dimitri Staessens <[email protected]>
+ *    Sander Vrijders   <[email protected]>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -26,20 +27,24 @@
 #include <ouroboros/sockets.h>
 #include <ouroboros/fcntl.h>
 #include <ouroboros/bitmap.h>
+#include <ouroboros/shm_flow_set.h>
 #include <ouroboros/shm_rdrbuff.h>
-#include <ouroboros/shm_ap_rbuff.h>
+#include <ouroboros/shm_rbuff.h>
 #include <ouroboros/utils.h>
-#include <ouroboros/select.h>
+#include <ouroboros/fqueue.h>
 
 #include <stdlib.h>
 #include <string.h>
 #include <stdio.h>
 
 struct flow_set {
-        bool dirty;
-        bool b[IRMD_MAX_FLOWS]; /* working copy */
-        bool s[IRMD_MAX_FLOWS]; /* safe copy */
-        pthread_rwlock_t lock;
+        size_t idx;
+};
+
+struct fqueue {
+        int    fqueue[SHM_BUFFER_SIZE]; /* safe copy from shm */
+        size_t fqsize;
+        size_t next;
 };
 
 enum port_state {
@@ -124,7 +129,9 @@ enum port_state port_wait_assign(struct port * p)
 }
 
 struct flow {
-        struct shm_ap_rbuff * rb;
+        struct shm_rbuff *    rx_rb;
+        struct shm_rbuff *    tx_rb;
+        struct shm_flow_set * set;
         int                   port_id;
         int                   oflags;
 
@@ -139,10 +146,11 @@ struct {
         pid_t                 api;
 
         struct shm_rdrbuff *  rdrb;
-        struct shm_ap_rbuff * rb;
+        struct shm_flow_set * fqset;
         pthread_rwlock_t      data_lock;
 
         struct bmp *          fds;
+        struct bmp *          fqueues;
         struct flow *         flows;
         struct port *         ports;
 
@@ -194,40 +202,52 @@ int ap_init(char * ap_name)
         if (ai.fds == NULL)
                 return -ENOMEM;
 
-        ai.rdrb = shm_rdrbuff_open();
-        if (ai.rdrb == NULL) {
+        ai.fqueues = bmp_create(AP_MAX_FQUEUES, 0);
+        if (ai.fqueues == NULL) {
+                bmp_destroy(ai.fds);
+                return -ENOMEM;
+        }
+
+        ai.fqset = shm_flow_set_create();
+        if (ai.fqset == NULL) {
+                bmp_destroy(ai.fqueues);
                 bmp_destroy(ai.fds);
                 return -1;
         }
 
-        ai.rb = shm_ap_rbuff_create();
-        if (ai.rb == NULL) {
-                shm_rdrbuff_close(ai.rdrb);
+        ai.rdrb = shm_rdrbuff_open();
+        if (ai.rdrb == NULL) {
+                shm_flow_set_destroy(ai.fqset);
+                bmp_destroy(ai.fqueues);
                 bmp_destroy(ai.fds);
                 return -1;
         }
 
         ai.flows = malloc(sizeof(*ai.flows) * AP_MAX_FLOWS);
         if (ai.flows == NULL) {
-                shm_ap_rbuff_destroy(ai.rb);
                 shm_rdrbuff_close(ai.rdrb);
+                shm_flow_set_destroy(ai.fqset);
+                bmp_destroy(ai.fqueues);
                 bmp_destroy(ai.fds);
                 return -1;
         }
 
         for (i = 0; i < AP_MAX_FLOWS; ++i) {
-                ai.flows[i].rb = NULL;
+                ai.flows[i].rx_rb   = NULL;
+                ai.flows[i].tx_rb   = NULL;
+                ai.flows[i].set     = NULL;
                 ai.flows[i].port_id = -1;
-                ai.flows[i].oflags = 0;
-                ai.flows[i].api = -1;
+                ai.flows[i].oflags  = 0;
+                ai.flows[i].api     = -1;
                 ai.flows[i].timeout = NULL;
         }
 
         ai.ports = malloc(sizeof(*ai.ports) * IRMD_MAX_FLOWS);
-        if (ai.flows == NULL) {
+        if (ai.ports == NULL) {
                 free(ai.flows);
-                shm_ap_rbuff_destroy(ai.rb);
                 shm_rdrbuff_close(ai.rdrb);
+                shm_flow_set_destroy(ai.fqset);
+                bmp_destroy(ai.fqueues);
                 bmp_destroy(ai.fds);
                 return -1;
         }
@@ -253,16 +273,10 @@ void ap_fini()
 
         pthread_rwlock_wrlock(&ai.data_lock);
 
-        /* remove all remaining sdus */
-        while ((i = shm_ap_rbuff_pop_idx(ai.rb)) >= 0)
-                shm_rdrbuff_remove(ai.rdrb, i);
-
-        if (ai.fds != NULL)
-                bmp_destroy(ai.fds);
-        if (ai.rb != NULL)
-                shm_ap_rbuff_destroy(ai.rb);
-        if (ai.rdrb != NULL)
-                shm_rdrbuff_close(ai.rdrb);
+        bmp_destroy(ai.fds);
+        bmp_destroy(ai.fqueues);
+        shm_flow_set_destroy(ai.fqset);
+        shm_rdrbuff_close(ai.rdrb);
 
         if (ai.daf_name != NULL)
                 free(ai.daf_name);
@@ -270,8 +284,15 @@ void ap_fini()
         pthread_rwlock_rdlock(&ai.flows_lock);
 
         for (i = 0; i < AP_MAX_FLOWS; ++i) {
-                if (ai.flows[i].rb != NULL)
-                        shm_ap_rbuff_close(ai.flows[i].rb);
+                if (ai.flows[i].tx_rb != NULL) {
+                        int idx;
+                        while ((idx = shm_rbuff_read(ai.flows[i].rx_rb)) >= 0)
+                                shm_rdrbuff_remove(ai.rdrb, idx);
+                        shm_rbuff_destroy(ai.flows[i].rx_rb);
+                        shm_rbuff_close(ai.flows[i].tx_rb);
+                        shm_flow_set_close(ai.flows[i].set);
+                }
+
                 if (ai.flows[i].timeout != NULL)
                         free(ai.flows[i].timeout);
         }
@@ -328,8 +349,8 @@ int flow_accept(char ** ae_name)
                 return -1;
         }
 
-        ai.flows[fd].rb = shm_ap_rbuff_open(recv_msg->api);
-        if (ai.flows[fd].rb == NULL) {
+        ai.flows[fd].rx_rb = shm_rbuff_create(recv_msg->port_id);
+        if (ai.flows[fd].rx_rb == NULL) {
                 bmp_release(ai.fds, fd);
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
@@ -337,10 +358,24 @@ int flow_accept(char ** ae_name)
                 return -1;
         }
 
+        ai.flows[fd].set = shm_flow_set_open(recv_msg->api);
+        if (ai.flows[fd].set == NULL) {
+                bmp_release(ai.fds, fd);
+                shm_rbuff_destroy(ai.flows[fd].rx_rb);
+                shm_rbuff_close(ai.flows[fd].tx_rb);
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                irm_msg__free_unpacked(recv_msg, NULL);
+                return -1;
+        }
+
+
         if (ae_name != NULL) {
                 *ae_name = strdup(recv_msg->ae_name);
                 if (*ae_name == NULL) {
-                        shm_ap_rbuff_close(ai.flows[fd].rb);
+                        shm_rbuff_destroy(ai.flows[fd].tx_rb);
+                        shm_rbuff_close(ai.flows[fd].tx_rb);
+                        shm_flow_set_close(ai.flows[fd].set);
                         bmp_release(ai.fds, fd);
                         pthread_rwlock_unlock(&ai.flows_lock);
                         pthread_rwlock_unlock(&ai.data_lock);
@@ -356,8 +391,6 @@ int flow_accept(char ** ae_name)
         ai.ports[recv_msg->port_id].fd    = fd;
         ai.ports[recv_msg->port_id].state = PORT_ID_ASSIGNED;
 
-        shm_ap_rbuff_open_port(ai.rb, recv_msg->port_id);
-
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
@@ -410,6 +443,17 @@ int flow_alloc_resp(int fd, int response)
 
         ret = recv_msg->result;
 
+        pthread_rwlock_wrlock(&ai.flows_lock);
+
+        ai.flows[fd].tx_rb = shm_rbuff_open(ai.flows[fd].api,
+                                            ai.flows[fd].port_id);
+        if (ai.flows[fd].tx_rb == NULL) {
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -1;
+        }
+
+        pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
         irm_msg__free_unpacked(recv_msg, NULL);
@@ -461,8 +505,11 @@ int flow_alloc(char * dst_name, char * src_ae_name, struct qos_spec * qos)
                 return -1;
         }
 
-        ai.flows[fd].rb = shm_ap_rbuff_open(recv_msg->api);
-        if (ai.flows[fd].rb == NULL) {
+        ai.flows[fd].port_id = recv_msg->port_id;
+        ai.flows[fd].oflags  = FLOW_O_DEFAULT;
+        ai.flows[fd].api     = recv_msg->api;
+        ai.flows[fd].rx_rb   = shm_rbuff_create(recv_msg->port_id);
+        if (ai.flows[fd].rx_rb == NULL) {
                 bmp_release(ai.fds, fd);
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
@@ -470,9 +517,26 @@ int flow_alloc(char * dst_name, char * src_ae_name, struct qos_spec * qos)
                 return -1;
         }
 
-        ai.flows[fd].port_id = recv_msg->port_id;
-        ai.flows[fd].oflags  = FLOW_O_DEFAULT;
-        ai.flows[fd].api     = recv_msg->api;
+        ai.flows[fd].tx_rb   = shm_rbuff_open(recv_msg->api, recv_msg->port_id);
+        if (ai.flows[fd].tx_rb == NULL) {
+                shm_rbuff_destroy(ai.flows[fd].rx_rb);
+                bmp_release(ai.fds, fd);
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                irm_msg__free_unpacked(recv_msg, NULL);
+                return -1;
+        }
+
+        ai.flows[fd].set = shm_flow_set_open(recv_msg->api);
+        if (ai.flows[fd].set == NULL) {
+                shm_rbuff_close(ai.flows[fd].tx_rb);
+                shm_rbuff_destroy(ai.flows[fd].rx_rb);
+                bmp_release(ai.fds, fd);
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                irm_msg__free_unpacked(recv_msg, NULL);
+                return -1;
+        }
 
         ai.ports[recv_msg->port_id].fd    = fd;
         ai.ports[recv_msg->port_id].state = PORT_ID_ASSIGNED;
@@ -480,8 +544,6 @@ int flow_alloc(char * dst_name, char * src_ae_name, struct qos_spec * qos)
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
-        shm_ap_rbuff_open_port(ai.rb, recv_msg->port_id);
-
         irm_msg__free_unpacked(recv_msg, NULL);
 
         return fd;
@@ -548,7 +610,7 @@ int flow_dealloc(int fd)
                 return -ENOTALLOC;
         }
 
-        if (shm_ap_rbuff_close_port(ai.rb, ai.flows[fd].port_id) == -EBUSY) {
+        if (shm_rbuff_block(ai.flows[fd].rx_rb) == -EBUSY) {
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
                 return -EBUSY;
@@ -559,8 +621,10 @@ int flow_dealloc(int fd)
         port_destroy(&ai.ports[msg.port_id]);
 
         ai.flows[fd].port_id = -1;
-        shm_ap_rbuff_close(ai.flows[fd].rb);
-        ai.flows[fd].rb = NULL;
+        shm_rbuff_destroy(ai.flows[fd].rx_rb);
+        ai.flows[fd].rx_rb = NULL;
+        shm_rbuff_close(ai.flows[fd].tx_rb);
+        ai.flows[fd].tx_rb = NULL;
         ai.flows[fd].oflags = 0;
         ai.flows[fd].api = -1;
         if (ai.flows[fd].timeout != NULL) {
@@ -604,9 +668,9 @@ int flow_cntl(int fd, int cmd, int oflags)
         case FLOW_F_SETFL: /* SET FLOW FLAGS */
                 ai.flows[fd].oflags = oflags;
                 if (oflags & FLOW_O_WRONLY)
-                        shm_ap_rbuff_close_port(ai.rb, ai.flows[fd].port_id);
+                        shm_rbuff_block(ai.flows[fd].rx_rb);
                 if (oflags & FLOW_O_RDWR)
-                        shm_ap_rbuff_open_port(ai.rb, ai.flows[fd].port_id);
+                        shm_rbuff_unblock(ai.flows[fd].rx_rb);
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
                 return old;
@@ -620,7 +684,6 @@ int flow_cntl(int fd, int cmd, int oflags)
 ssize_t flow_write(int fd, void * buf, size_t count)
 {
         ssize_t idx;
-        struct rb_entry e;
 
         if (buf == NULL)
                 return 0;
@@ -653,13 +716,10 @@ ssize_t flow_write(int fd, void * buf, size_t count)
                 if (idx < 0) {
                         pthread_rwlock_unlock(&ai.flows_lock);
                         pthread_rwlock_unlock(&ai.data_lock);
-                        return -idx;
+                        return idx;
                 }
 
-                e.index   = idx;
-                e.port_id = ai.flows[fd].port_id;
-
-                if (shm_ap_rbuff_write(ai.flows[fd].rb, &e) < 0) {
+                if (shm_rbuff_write(ai.flows[fd].tx_rb, idx) < 0) {
                         shm_rdrbuff_remove(ai.rdrb, idx);
                         pthread_rwlock_unlock(&ai.flows_lock);
                         pthread_rwlock_unlock(&ai.data_lock);
@@ -667,7 +727,7 @@ ssize_t flow_write(int fd, void * buf, size_t count)
                 }
         } else { /* blocking */
                 struct shm_rdrbuff * rdrb = ai.rdrb;
-                pid_t                api = ai.flows[fd].api;
+                pid_t                api  = ai.flows[fd].api;
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
 
@@ -681,17 +741,16 @@ ssize_t flow_write(int fd, void * buf, size_t count)
                 pthread_rwlock_rdlock(&ai.data_lock);
                 pthread_rwlock_rdlock(&ai.flows_lock);
 
-                e.index   = idx;
-                e.port_id = ai.flows[fd].port_id;
-
-                if (shm_ap_rbuff_write(ai.flows[fd].rb, &e) < 0) {
-                        shm_rdrbuff_remove(ai.rdrb, e.index);
+                if (shm_rbuff_write(ai.flows[fd].tx_rb, idx) < 0) {
+                        shm_rdrbuff_remove(ai.rdrb, idx);
                         pthread_rwlock_unlock(&ai.flows_lock);
                         pthread_rwlock_unlock(&ai.data_lock);
                         return -ENOTALLOC;
                 }
         }
 
+        shm_flow_set_notify(ai.flows[fd].set, ai.flows[fd].port_id);
+
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
@@ -717,15 +776,14 @@ ssize_t flow_read(int fd, void * buf, size_t count)
         }
 
         if (ai.flows[fd].oflags & FLOW_O_NONBLOCK) {
-                idx = shm_ap_rbuff_read_port(ai.rb, ai.flows[fd].port_id);
+                idx = shm_rbuff_read(ai.flows[fd].rx_rb);
                 pthread_rwlock_unlock(&ai.flows_lock);
         } else {
-                struct shm_ap_rbuff * rb      = ai.rb;
-                int                   port_id = ai.flows[fd].port_id;
-                struct timespec *     timeout = ai.flows[fd].timeout;
+                struct shm_rbuff * rb     = ai.flows[fd].rx_rb;
+                struct timespec * timeout = ai.flows[fd].timeout;
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
-                idx = shm_ap_rbuff_read_port_b(rb, port_id, timeout);
+                idx = shm_rbuff_read_b(rb, timeout);
                 pthread_rwlock_rdlock(&ai.data_lock);
         }
 
@@ -757,79 +815,163 @@ struct flow_set * flow_set_create()
         if (set == NULL)
                 return NULL;
 
-        if (pthread_rwlock_init(&set->lock, NULL)) {
+        assert(ai.fqueues);
+
+        set->idx = bmp_allocate(ai.fqueues);
+        if (!bmp_is_id_valid(ai.fqueues, set->idx)) {
                 free(set);
                 return NULL;
         }
 
-        memset(set->b, 0, IRMD_MAX_FLOWS);
-        memset(set->s, 0, IRMD_MAX_FLOWS);
+        return set;
+}
 
-        set->dirty = true;
+void flow_set_destroy(struct flow_set * set)
+{
+        if (set == NULL)
+                return;
 
-        return set;
+        flow_set_zero(set);
+        bmp_release(ai.fqueues, set->idx);
+        free(set);
 }
 
-void flow_set_zero(struct flow_set * set)
+struct fqueue * fqueue_create()
 {
-        pthread_rwlock_wrlock(&set->lock);
-        memset(set->b, 0, IRMD_MAX_FLOWS);
-        set->dirty = true;
-        pthread_rwlock_unlock(&set->lock);
+        struct fqueue * fq = malloc(sizeof(*fq));
+        if (fq == NULL)
+                return NULL;
+
+        memset(fq->fqueue, -1, SHM_BUFFER_SIZE);
+        fq->fqsize = 0;
+        fq->next   = 0;
+
+        return fq;
 }
 
-void flow_set_add(struct flow_set * set, int fd)
+void fqueue_destroy(struct fqueue * fq)
 {
-        pthread_rwlock_wrlock(&set->lock);
-        set->b[ai.flows[fd].port_id] = true;
-        set->dirty = true;
-        pthread_rwlock_unlock(&set->lock);
+        if (fq == NULL)
+                return
+        free(fq);
 }
 
-void flow_set_del(struct flow_set * set, int fd)
+void flow_set_zero(struct flow_set * set)
 {
-        pthread_rwlock_wrlock(&set->lock);
-        set->b[ai.flows[fd].port_id] = false;
-        set->dirty = true;
-        pthread_rwlock_unlock(&set->lock);
+        if (set == NULL)
+                return;
+
+        pthread_rwlock_rdlock(&ai.data_lock);
+
+        shm_flow_set_zero(ai.fqset, set->idx);
+
+        pthread_rwlock_unlock(&ai.data_lock);
 }
 
-bool flow_set_has(struct flow_set * set, int fd)
+int flow_set_add(struct flow_set * set, int fd)
 {
-        bool ret;
-        pthread_rwlock_rdlock(&set->lock);
-        ret = set->b[ai.flows[fd].port_id];
-        pthread_rwlock_unlock(&set->lock);
+        int ret;
+
+        if (set == NULL)
+                return -EINVAL;
+
+        pthread_rwlock_rdlock(&ai.data_lock);
+        pthread_rwlock_rdlock(&ai.flows_lock);
+
+        ret = shm_flow_set_add(ai.fqset, set->idx, ai.flows[fd].port_id);
+
+        pthread_rwlock_unlock(&ai.flows_lock);
+        pthread_rwlock_unlock(&ai.data_lock);
+
         return ret;
 }
 
-void flow_set_destroy(struct flow_set * set)
+void flow_set_del(struct flow_set * set, int fd)
 {
-        pthread_rwlock_destroy(&set->lock);
-        free(set);
+        if (set == NULL)
+                return;
+
+        pthread_rwlock_rdlock(&ai.data_lock);
+        pthread_rwlock_rdlock(&ai.flows_lock);
+
+        if (ai.flows[fd].port_id >= 0)
+                shm_flow_set_del(ai.fqset, set->idx, ai.flows[fd].port_id);
+
+        pthread_rwlock_unlock(&ai.flows_lock);
+        pthread_rwlock_unlock(&ai.data_lock);
 }
 
-static void flow_set_cpy(struct flow_set * set)
+bool flow_set_has(struct flow_set * set, int fd)
 {
-        pthread_rwlock_rdlock(&set->lock);
-        if (set->dirty)
-                memcpy(set->s, set->b, IRMD_MAX_FLOWS);
-        set->dirty = false;
-        pthread_rwlock_unlock(&set->lock);
+        bool ret = false;
+
+        if (set == NULL || fd < 0)
+                return false;
+
+        pthread_rwlock_rdlock(&ai.data_lock);
+        pthread_rwlock_rdlock(&ai.flows_lock);
+
+        if (ai.flows[fd].port_id < 0) {
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return false;
+        }
+
+        ret = (shm_flow_set_has(ai.fqset, set->idx, ai.flows[fd].port_id) == 1);
+
+        pthread_rwlock_unlock(&ai.flows_lock);
+        pthread_rwlock_unlock(&ai.data_lock);
+
+        return ret;
 }
 
-int flow_select(struct flow_set * set, const struct timespec * timeout)
+int fqueue_next(struct fqueue * fq)
 {
-        int port_id;
-        if (set == NULL) {
-                port_id = shm_ap_rbuff_peek_b(ai.rb, NULL, timeout);
-        } else {
-                flow_set_cpy(set);
-                port_id = shm_ap_rbuff_peek_b(ai.rb, (bool *) set->s, timeout);
+        int fd;
+
+        if (fq == NULL)
+                return -EINVAL;
+
+        if (fq->next == fq->fqsize) {
+                fq->fqsize = 0;
+                fq->next = 0;
+                return -EPERM;
         }
-        if (port_id < 0)
-                return port_id;
-        return ai.ports[port_id].fd;
+
+        pthread_rwlock_rdlock(&ai.data_lock);
+        pthread_rwlock_rdlock(&ai.flows_lock);
+
+        fd = ai.ports[fq->fqueue[fq->next++]].fd;
+
+        pthread_rwlock_unlock(&ai.flows_lock);
+        pthread_rwlock_unlock(&ai.data_lock);
+
+        return fd;
+}
+
+int flow_event_wait(struct flow_set *       set,
+                    struct fqueue *         fq,
+                    const struct timespec * timeout)
+{
+        int ret;
+
+        if (set == NULL)
+                return -EINVAL;
+
+        if (fq->fqsize > 0)
+                return 0;
+
+        ret = shm_flow_set_wait(ai.fqset, set->idx, fq->fqueue, timeout);
+        if (ret == -ETIMEDOUT)
+                return -ETIMEDOUT;
+
+        if (ret < 0)
+                return ret;
+
+        fq->fqsize = ret;
+        fq->next   = 0;
+
+        return 0;
 }
 
 /* ipcp-dev functions */
@@ -848,8 +990,8 @@ int np1_flow_alloc(pid_t n_api, int port_id)
                 return -1;
         }
 
-        ai.flows[fd].rb = shm_ap_rbuff_open(n_api);
-        if (ai.flows[fd].rb == NULL) {
+        ai.flows[fd].rx_rb = shm_rbuff_create(port_id);
+        if (ai.flows[fd].rx_rb == NULL) {
                 bmp_release(ai.fds, fd);
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
@@ -863,8 +1005,6 @@ int np1_flow_alloc(pid_t n_api, int port_id)
         ai.ports[port_id].fd = fd;
         port_set_state(&ai.ports[port_id], PORT_ID_ASSIGNED);
 
-        shm_ap_rbuff_open_port(ai.rb, port_id);
-
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
@@ -890,7 +1030,6 @@ int np1_flow_dealloc(int port_id)
 int np1_flow_resp(pid_t n_api, int port_id)
 {
         int fd;
-        struct shm_ap_rbuff * rb;
 
         port_wait_assign(&ai.ports[port_id]);
 
@@ -904,18 +1043,26 @@ int np1_flow_resp(pid_t n_api, int port_id)
                 return fd;
         }
 
-        rb = shm_ap_rbuff_open(n_api);
-        if (rb == NULL) {
+        ai.flows[fd].tx_rb = shm_rbuff_open(n_api, port_id);
+        if (ai.flows[fd].tx_rb == NULL) {
                 ai.flows[fd].port_id = -1;
+                shm_rbuff_destroy(ai.flows[fd].rx_rb);
                 port_destroy(&ai.ports[port_id]);
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
                 return -1;
         }
 
-        ai.flows[fd].rb = rb;
-
-        shm_ap_rbuff_open_port(ai.rb, port_id);
+        ai.flows[fd].set = shm_flow_set_open(n_api);
+        if (ai.flows[fd].set == NULL) {
+                shm_rbuff_close(ai.flows[fd].tx_rb);
+                ai.flows[fd].port_id = -1;
+                shm_rbuff_destroy(ai.flows[fd].rx_rb);
+                port_destroy(&ai.ports[port_id]);
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -1;
+        }
 
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
@@ -929,9 +1076,9 @@ int ipcp_create_r(pid_t api)
         irm_msg_t * recv_msg = NULL;
         int ret = -1;
 
-        msg.code         = IRM_MSG_CODE__IPCP_CREATE_R;
-        msg.has_api      = true;
-        msg.api          = api;
+        msg.code    = IRM_MSG_CODE__IPCP_CREATE_R;
+        msg.has_api = true;
+        msg.api     = api;
 
         recv_msg = send_recv_irm_msg(&msg);
         if (recv_msg == NULL)
@@ -958,11 +1105,11 @@ int ipcp_flow_req_arr(pid_t  api, char * dst_name, char * src_ae_name)
         if (dst_name == NULL || src_ae_name == NULL)
                 return -EINVAL;
 
-        msg.code          = IRM_MSG_CODE__IPCP_FLOW_REQ_ARR;
-        msg.has_api       = true;
-        msg.api           = api;
-        msg.dst_name      = dst_name;
-        msg.ae_name       = src_ae_name;
+        msg.code     = IRM_MSG_CODE__IPCP_FLOW_REQ_ARR;
+        msg.has_api  = true;
+        msg.api      = api;
+        msg.dst_name = dst_name;
+        msg.ae_name  = src_ae_name;
 
         pthread_rwlock_rdlock(&ai.data_lock);
         pthread_rwlock_wrlock(&ai.flows_lock);
@@ -974,7 +1121,7 @@ int ipcp_flow_req_arr(pid_t  api, char * dst_name, char * src_ae_name)
                 return -1; /* -ENOMOREFDS */
         }
 
-        ai.flows[fd].rb    = NULL;
+        ai.flows[fd].tx_rb    = NULL;
 
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
@@ -996,8 +1143,16 @@ int ipcp_flow_req_arr(pid_t  api, char * dst_name, char * src_ae_name)
         pthread_rwlock_rdlock(&ai.data_lock);
         pthread_rwlock_wrlock(&ai.flows_lock);
 
+        ai.flows[fd].rx_rb = shm_rbuff_create(port_id);
+        if (ai.flows[fd].rx_rb == NULL) {
+                ai.flows[fd].port_id = -1;
+                port_destroy(&ai.ports[port_id]);
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -1;
+        }
+
         ai.flows[fd].port_id = port_id;
-        ai.flows[fd].rb      = NULL;
 
         ai.ports[port_id].fd = fd;
         port_set_state(&(ai.ports[port_id]), PORT_ID_ASSIGNED);
@@ -1019,16 +1174,13 @@ int ipcp_flow_alloc_reply(int fd, int response)
 
         pthread_rwlock_rdlock(&ai.data_lock);
         pthread_rwlock_rdlock(&ai.flows_lock);
-        msg.port_id      = ai.flows[fd].port_id;
+        msg.port_id = ai.flows[fd].port_id;
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
         msg.has_response = true;
         msg.response     = response;
 
-        if (response)
-                shm_ap_rbuff_open_port(ai.rb, msg.port_id);
-
         recv_msg = send_recv_irm_msg(&msg);
         if (recv_msg == NULL)
                 return -1;
@@ -1039,6 +1191,26 @@ int ipcp_flow_alloc_reply(int fd, int response)
         }
 
         ret = recv_msg->result;
+
+        pthread_rwlock_wrlock(&ai.flows_lock);
+
+        ai.flows[fd].tx_rb = shm_rbuff_open(ai.flows[fd].api,
+                                            ai.flows[fd].port_id);
+        if (ai.flows[fd].tx_rb == NULL) {
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -1;
+        }
+
+        ai.flows[fd].set = shm_flow_set_open(ai.flows[fd].api);
+        if (ai.flows[fd].set == NULL) {
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -1;
+        }
+
+        pthread_rwlock_unlock(&ai.flows_lock);
+
         irm_msg__free_unpacked(recv_msg, NULL);
 
         return ret;
@@ -1061,7 +1233,7 @@ int ipcp_flow_read(int fd, struct shm_du_buff ** sdb)
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
-        idx = shm_ap_rbuff_read_port(ai.rb, port_id);
+        idx = shm_rbuff_read(ai.flows[fd].rx_rb);
         if (idx < 0) {
                 pthread_rwlock_rdlock(&ai.data_lock);
                 pthread_rwlock_rdlock(&ai.flows_lock);
@@ -1081,7 +1253,7 @@ int ipcp_flow_read(int fd, struct shm_du_buff ** sdb)
 
 int ipcp_flow_write(int fd, struct shm_du_buff * sdb)
 {
-        struct rb_entry e;
+        ssize_t idx;
 
         if (sdb == NULL)
                 return -EINVAL;
@@ -1095,16 +1267,16 @@ int ipcp_flow_write(int fd, struct shm_du_buff * sdb)
                 return -EPERM;
         }
 
-        if (ai.flows[fd].rb == NULL) {
+        if (ai.flows[fd].tx_rb == NULL) {
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
                 return -EPERM;
         }
 
-        e.index = shm_du_buff_get_idx(sdb);
-        e.port_id = ai.flows[fd].port_id;
+        idx = shm_du_buff_get_idx(sdb);
 
-        shm_ap_rbuff_write(ai.flows[fd].rb, &e);
+        shm_rbuff_write(ai.flows[fd].tx_rb, idx);
+        shm_flow_set_notify(ai.flows[fd].set, ai.flows[fd].port_id);
 
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
@@ -1112,46 +1284,28 @@ int ipcp_flow_write(int fd, struct shm_du_buff * sdb)
         return 0;
 }
 
-struct rb_entry * local_flow_read(int fd)
+ssize_t local_flow_read(int fd)
 {
-        int port_id;
-        struct rb_entry * e = NULL;
-
-        pthread_rwlock_rdlock(&ai.data_lock);
-        pthread_rwlock_rdlock(&ai.flows_lock);
-
-        port_id = ai.flows[fd].port_id;
-
-        pthread_rwlock_unlock(&ai.flows_lock);
-        pthread_rwlock_unlock(&ai.data_lock);
-
-        if (port_id != -1) {
-                e = malloc(sizeof(*e));
-                if (e == NULL)
-                        return NULL;
-                e->index = shm_ap_rbuff_read_port(ai.rb, port_id);
-        }
-
-        return e;
+        return shm_rbuff_read(ai.flows[fd].rx_rb);
 }
 
-int local_flow_write(int fd, struct rb_entry * e)
+int local_flow_write(int fd, ssize_t idx)
 {
-        if (e == NULL || fd < 0)
+        if (fd < 0)
                 return -EINVAL;
 
         pthread_rwlock_rdlock(&ai.data_lock);
         pthread_rwlock_rdlock(&ai.flows_lock);
 
-        if (ai.flows[fd].rb == NULL) {
+        if (ai.flows[fd].tx_rb == NULL) {
                 pthread_rwlock_unlock(&ai.flows_lock);
                 pthread_rwlock_unlock(&ai.data_lock);
                 return -EPERM;
         }
 
-        e->port_id = ai.flows[fd].port_id;
+        shm_rbuff_write(ai.flows[fd].tx_rb, idx);
 
-        shm_ap_rbuff_write(ai.flows[fd].rb, e);
+        shm_flow_set_notify(ai.flows[fd].set, ai.flows[fd].port_id);
 
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
@@ -1159,22 +1313,26 @@ int local_flow_write(int fd, struct rb_entry * e)
         return 0;
 }
 
-int ipcp_read_shim(struct shm_du_buff ** sdb)
+int ipcp_read_shim(int fd, struct shm_du_buff ** sdb)
 {
-        int fd;
-        struct rb_entry * e = shm_ap_rbuff_read(ai.rb);
+        ssize_t idx;
 
         pthread_rwlock_rdlock(&ai.data_lock);
         pthread_rwlock_rdlock(&ai.flows_lock);
 
-        fd = ai.ports[e->port_id].fd;
+        if (ai.flows[fd].rx_rb == NULL) {
+                pthread_rwlock_unlock(&ai.flows_lock);
+                pthread_rwlock_unlock(&ai.data_lock);
+                return -EPERM;
+        }
 
-        *sdb = shm_rdrbuff_get(ai.rdrb, e->index);
+        idx = shm_rbuff_read(ai.flows[fd].rx_rb);
+        *sdb = shm_rdrbuff_get(ai.rdrb, idx);
 
         pthread_rwlock_unlock(&ai.flows_lock);
         pthread_rwlock_unlock(&ai.data_lock);
 
-        return fd;
+        return 0;
 }
 
 void ipcp_flow_del(struct shm_du_buff * sdb)