Merged in dstaesse/ouroboros/be-rbuff-ll (pull request #450)

lib: Lockless implementation of the rbuff

3 files changed, 789 insertions, 404 deletions

diff --git a/src/lib/shm_rbuff.c b/src/lib/shm_rbuff.c
index b8db7c19..dc1d9f0a 100644
--- a/src/lib/shm_rbuff.c
+++ b/src/lib/shm_rbuff.c
@@ -1,406 +1,6 @@
-/*
- * Ouroboros - Copyright (C) 2016 - 2017
- *
- * Ring buffer for incoming SDUs
- *
- *    Dimitri Staessens <[email protected]>
- *    Sander Vrijders   <[email protected]>
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public License
- * version 2.1 as published by the Free Software Foundation.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
- * 02110-1301 USA
- */
-
-#include <ouroboros/config.h>
-#include <ouroboros/shm_rbuff.h>
-#include <ouroboros/lockfile.h>
-#include <ouroboros/time_utils.h>
-#include <ouroboros/errno.h>
-
-#include <pthread.h>
-#include <sys/mman.h>
-#include <fcntl.h>
-#include <stdlib.h>
-#include <string.h>
-#include <stdio.h>
-#include <stdint.h>
-#include <unistd.h>
-#include <signal.h>
-#include <sys/stat.h>
-#include <assert.h>
-#include <stdbool.h>
-
-#define FN_MAX_CHARS 255
-#define RB_OPEN 0
-#define RB_CLOSED 1
-
-#define SHM_RBUFF_FILE_SIZE ((SHM_BUFFER_SIZE) * sizeof(ssize_t)          \
-                             + 3 * sizeof(size_t)                         \
-                             + sizeof(pthread_mutex_t)                    \
-                             + 2 * sizeof (pthread_cond_t))
-
-#define shm_rbuff_used(rb) ((*rb->head + (SHM_BUFFER_SIZE) - *rb->tail)   \
-                            & ((SHM_BUFFER_SIZE) - 1))
-#define shm_rbuff_free(rb) (shm_rbuff_used(rb) + 1 < (SHM_BUFFER_SIZE))
-#define shm_rbuff_empty(rb) (*rb->head == *rb->tail)
-#define head_el_ptr(rb) (rb->shm_base + *rb->head)
-#define tail_el_ptr(rb) (rb->shm_base + *rb->tail)
-
-struct shm_rbuff {
-        ssize_t *         shm_base; /* start of entry                */
-        size_t *          head;     /* start of ringbuffer head      */
-        size_t *          tail;     /* start of ringbuffer tail      */
-        size_t *          acl;      /* access control                */
-        pthread_mutex_t * lock;     /* lock all free space in shm    */
-        pthread_cond_t *  add;      /* SDU arrived                   */
-        pthread_cond_t *  del;      /* SDU removed                   */
-        pid_t             api;      /* api of the owner              */
-        int               port_id;  /* port_id of the flow           */
-};
-
-struct shm_rbuff * shm_rbuff_create(pid_t api, int port_id)
-{
-        struct shm_rbuff *  rb;
-        int                 shm_fd;
-        ssize_t *           shm_base;
-        pthread_mutexattr_t mattr;
-        pthread_condattr_t  cattr;
-        char                fn[FN_MAX_CHARS];
-        mode_t              mask;
-
-        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
-
-        rb = malloc(sizeof(*rb));
-        if (rb == NULL)
-                return NULL;
-
-        mask = umask(0);
-
-        shm_fd = shm_open(fn, O_CREAT | O_EXCL | O_RDWR, 0666);
-        if (shm_fd == -1) {
-                free(rb);
-                return NULL;
-        }
-
-        umask(mask);
-
-        if (ftruncate(shm_fd, SHM_RBUFF_FILE_SIZE - 1) < 0) {
-                free(rb);
-                close(shm_fd);
-                return NULL;
-        }
-
-        shm_base = mmap(NULL,
-                        SHM_RBUFF_FILE_SIZE,
-                        PROT_READ | PROT_WRITE,
-                        MAP_SHARED,
-                        shm_fd,
-                        0);
-
-        close(shm_fd);
-
-        if (shm_base == MAP_FAILED) {
-                shm_unlink(fn);
-                free(rb);
-                return NULL;
-        }
-
-        rb->shm_base = shm_base;
-        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
-        rb->tail     = rb->head + 1;
-        rb->acl      = rb->tail + 1;
-        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
-        rb->add      = (pthread_cond_t *) (rb->lock + 1);
-        rb->del      = rb->add + 1;
-
-        pthread_mutexattr_init(&mattr);
-#ifndef __APPLE__
-        pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST);
-#endif
-        pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
-        pthread_mutex_init(rb->lock, &mattr);
-
-        pthread_condattr_init(&cattr);
-        pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED);
-#ifndef __APPLE__
-        pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK);
-#endif
-        pthread_cond_init(rb->add, &cattr);
-        pthread_cond_init(rb->del, &cattr);
-
-        *rb->acl = RB_OPEN;
-        *rb->head = 0;
-        *rb->tail = 0;
-
-        rb->api = api;
-        rb->port_id = port_id;
-
-        return rb;
-}
-
-struct shm_rbuff * shm_rbuff_open(pid_t api, int port_id)
-{
-        struct shm_rbuff * rb;
-        int                shm_fd;
-        ssize_t *          shm_base;
-        char               fn[FN_MAX_CHARS];
-
-        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
-
-        rb = malloc(sizeof(*rb));
-        if (rb == NULL)
-                return NULL;
-
-        shm_fd = shm_open(fn, O_RDWR, 0666);
-        if (shm_fd == -1) {
-                free(rb);
-                return NULL;
-        }
-
-        shm_base = mmap(NULL,
-                        SHM_RBUFF_FILE_SIZE,
-                        PROT_READ | PROT_WRITE,
-                        MAP_SHARED,
-                        shm_fd,
-                        0);
-
-        close(shm_fd);
-
-        if (shm_base == MAP_FAILED) {
-                shm_unlink(fn);
-                free(rb);
-                return NULL;
-        }
-
-        rb->shm_base = shm_base;
-        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
-        rb->tail     = rb->head + 1;
-        rb->acl      = rb->tail + 1;
-        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
-        rb->add      = (pthread_cond_t *) (rb->lock + 1);
-        rb->del      = rb->add + 1;
-
-        rb->api = api;
-        rb->port_id = port_id;
-
-        return rb;
-}
-
-void shm_rbuff_close(struct shm_rbuff * rb)
-{
-        assert(rb);
-
-        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
-
-        free(rb);
-}
-
-void shm_rbuff_destroy(struct shm_rbuff * rb)
-{
-        char fn[FN_MAX_CHARS];
-
-        assert(rb);
-
-#ifdef CONFIG_OUROBOROS_DEBUG
-        pthread_mutex_lock(rb->lock);
-
-        assert(shm_rbuff_empty(rb));
-
-        pthread_mutex_unlock(rb->lock);
-#endif
-        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", rb->api, rb->port_id);
-
-        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
-        shm_unlink(fn);
-
-        free(rb);
-}
-
-int shm_rbuff_write(struct shm_rbuff * rb, size_t idx)
-{
-        assert(rb);
-        assert(idx < SHM_BUFFER_SIZE);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        if (*rb->acl == RB_CLOSED) {
-                pthread_mutex_unlock(rb->lock);
-                return -ENOTALLOC;
-        }
-
-        if (!shm_rbuff_free(rb)) {
-                pthread_mutex_unlock(rb->lock);
-                return -EAGAIN;
-        }
-
-        if (shm_rbuff_empty(rb))
-                pthread_cond_broadcast(rb->add);
-
-        *head_el_ptr(rb) = (ssize_t) idx;
-        *rb->head = (*rb->head + 1) & ((SHM_BUFFER_SIZE) -1);
-
-        pthread_mutex_unlock(rb->lock);
-
-        return 0;
-}
-
-ssize_t shm_rbuff_read(struct shm_rbuff * rb)
-{
-        ssize_t ret = 0;
-
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        if (shm_rbuff_empty(rb)) {
-                pthread_mutex_unlock(rb->lock);
-                return -EAGAIN;
-        }
-
-        ret = *tail_el_ptr(rb);
-        *rb->tail = (*rb->tail + 1) & ((SHM_BUFFER_SIZE) - 1);
-        pthread_cond_broadcast(rb->del);
-
-        pthread_mutex_unlock(rb->lock);
-
-        return ret;
-}
-
-ssize_t shm_rbuff_read_b(struct shm_rbuff *      rb,
-                         const struct timespec * timeout)
-{
-        struct timespec abstime;
-        ssize_t idx = -1;
-
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        if (timeout != NULL) {
-                clock_gettime(PTHREAD_COND_CLOCK, &abstime);
-                ts_add(&abstime, timeout, &abstime);
-        }
-
-        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
-                             (void *) rb->lock);
-
-        while (shm_rbuff_empty(rb) && (idx != -ETIMEDOUT)) {
-                if (timeout != NULL)
-                        idx = -pthread_cond_timedwait(rb->add,
-                                                      rb->lock,
-                                                      &abstime);
-                else
-                        idx = -pthread_cond_wait(rb->add, rb->lock);
-#ifndef __APPLE__
-                if (idx == -EOWNERDEAD)
-                        pthread_mutex_consistent(rb->lock);
-#endif
-        }
-
-        if (idx != -ETIMEDOUT) {
-                idx = *tail_el_ptr(rb);
-                *rb->tail = (*rb->tail + 1) & ((SHM_BUFFER_SIZE) - 1);
-                pthread_cond_broadcast(rb->del);
-        }
-
-        pthread_cleanup_pop(true);
-
-        return idx;
-}
-
-void shm_rbuff_block(struct shm_rbuff * rb)
-{
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        *rb->acl = RB_CLOSED;
-
-        pthread_mutex_unlock(rb->lock);
-}
-
-void shm_rbuff_unblock(struct shm_rbuff * rb)
-{
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        *rb->acl = RB_OPEN;
-
-        pthread_mutex_unlock(rb->lock);
-}
-
-void shm_rbuff_fini(struct shm_rbuff * rb)
-{
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
-#else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
-#endif
-        assert(*rb->acl == RB_CLOSED);
-
-        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
-                             (void *) rb->lock);
-
-        while (!shm_rbuff_empty(rb))
-#ifdef __APPLE__
-                pthread_cond_wait(rb->del, rb->lock);
-#else
-                if (pthread_cond_wait(rb->del, rb->lock) == EOWNERDEAD)
-                        pthread_mutex_consistent(rb->lock);
-#endif
-        pthread_cleanup_pop(true);
-}
-
-size_t shm_rbuff_queued(struct shm_rbuff * rb)
-{
-        size_t ret;
-
-        assert(rb);
-
-#ifdef __APPLE__
-        pthread_mutex_lock(rb->lock);
+#if (defined (SHM_RBUFF_LOCKLESS) && \
+     (defined(__GNUC__) || defined (__clang__)))
+#include "shm_rbuff_ll.c"
 #else
-        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
-                pthread_mutex_consistent(rb->lock);
+#include "shm_rbuff_pthr.c"
 #endif
-
-        ret = shm_rbuff_used(rb);
-
-        pthread_mutex_unlock(rb->lock);
-
-        return ret;
-}
diff --git a/src/lib/shm_rbuff_ll.c b/src/lib/shm_rbuff_ll.c
new file mode 100644
index 00000000..b5b8bed2
--- /dev/null
+++ b/src/lib/shm_rbuff_ll.c
@@ -0,0 +1,379 @@
+/*
+ * Ouroboros - Copyright (C) 2016 - 2017
+ *
+ * Lockless ring buffer for incoming SDUs
+ *
+ *    Dimitri Staessens <[email protected]>
+ *    Sander Vrijders   <[email protected]>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * version 2.1 as published by the Free Software Foundation.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ */
+
+#include <ouroboros/config.h>
+#include <ouroboros/shm_rbuff.h>
+#include <ouroboros/lockfile.h>
+#include <ouroboros/time_utils.h>
+#include <ouroboros/errno.h>
+
+#include <pthread.h>
+#include <sys/mman.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <assert.h>
+#include <stdbool.h>
+
+#define FN_MAX_CHARS 255
+#define RB_OPEN 0
+#define RB_CLOSED 1
+
+#define SHM_RBUFF_FILE_SIZE ((SHM_BUFFER_SIZE) * sizeof(ssize_t)        \
+                             + 3 * sizeof(size_t)                       \
+                             + sizeof(pthread_mutex_t)                  \
+                             + 2 * sizeof (pthread_cond_t))
+
+#define RB_HEAD __sync_fetch_and_add(rb->head, 0)
+#define RB_TAIL __sync_fetch_and_add(rb->tail, 0)
+
+#define shm_rbuff_used(rb) ((RB_HEAD + (SHM_BUFFER_SIZE) - RB_TAIL)     \
+                            & ((SHM_BUFFER_SIZE) - 1))
+#define shm_rbuff_free(rb) (shm_rbuff_used(rb) + 1 < (SHM_BUFFER_SIZE))
+#define shm_rbuff_empty(rb) (RB_HEAD == RB_TAIL)
+#define head_el_ptr(rb) (rb->shm_base + RB_HEAD)
+#define tail_el_ptr(rb) (rb->shm_base + RB_TAIL)
+
+struct shm_rbuff {
+        ssize_t *         shm_base; /* start of entry                */
+        size_t *          head;     /* start of ringbuffer head      */
+        size_t *          tail;     /* start of ringbuffer tail      */
+        size_t *          acl;      /* access control                */
+        pthread_mutex_t * lock;     /* lock all free space in shm    */
+        pthread_cond_t *  add;      /* SDU arrived                   */
+        pthread_cond_t *  del;      /* SDU removed                   */
+        pid_t             api;      /* api of the owner              */
+        int               port_id;  /* port_id of the flow           */
+};
+
+struct shm_rbuff * shm_rbuff_create(pid_t api, int port_id)
+{
+        struct shm_rbuff *  rb;
+        int                 shm_fd;
+        ssize_t *           shm_base;
+        pthread_mutexattr_t mattr;
+        pthread_condattr_t  cattr;
+        char                fn[FN_MAX_CHARS];
+        mode_t              mask;
+
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
+
+        rb = malloc(sizeof(*rb));
+        if (rb == NULL)
+                return NULL;
+
+        mask = umask(0);
+
+        shm_fd = shm_open(fn, O_CREAT | O_EXCL | O_RDWR, 0666);
+        if (shm_fd == -1) {
+                free(rb);
+                return NULL;
+        }
+
+        umask(mask);
+
+        if (ftruncate(shm_fd, SHM_RBUFF_FILE_SIZE - 1) < 0) {
+                free(rb);
+                close(shm_fd);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_RBUFF_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                shm_unlink(fn);
+                free(rb);
+                return NULL;
+        }
+
+        rb->shm_base = shm_base;
+        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
+        rb->tail     = rb->head + 1;
+        rb->acl      = rb->tail + 1;
+        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
+        rb->add      = (pthread_cond_t *) (rb->lock + 1);
+        rb->del      = rb->add + 1;
+
+        pthread_mutexattr_init(&mattr);
+#ifndef __APPLE__
+        pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST);
+#endif
+        pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
+        pthread_mutex_init(rb->lock, &mattr);
+
+        pthread_condattr_init(&cattr);
+        pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED);
+#ifndef __APPLE__
+        pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK);
+#endif
+        pthread_cond_init(rb->add, &cattr);
+        pthread_cond_init(rb->del, &cattr);
+
+        *rb->acl = RB_OPEN;
+        *rb->head = 0;
+        *rb->tail = 0;
+
+        rb->api = api;
+        rb->port_id = port_id;
+
+        return rb;
+}
+
+struct shm_rbuff * shm_rbuff_open(pid_t api, int port_id)
+{
+        struct shm_rbuff * rb;
+        int                shm_fd;
+        ssize_t *          shm_base;
+        char               fn[FN_MAX_CHARS];
+
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
+
+        rb = malloc(sizeof(*rb));
+        if (rb == NULL)
+                return NULL;
+
+        shm_fd = shm_open(fn, O_RDWR, 0666);
+        if (shm_fd == -1) {
+                free(rb);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_RBUFF_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                shm_unlink(fn);
+                free(rb);
+                return NULL;
+        }
+
+        rb->shm_base = shm_base;
+        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
+        rb->tail     = rb->head + 1;
+        rb->acl      = rb->tail + 1;
+        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
+        rb->add      = (pthread_cond_t *) (rb->lock + 1);
+        rb->del      = rb->add + 1;
+
+        rb->api = api;
+        rb->port_id = port_id;
+
+        return rb;
+}
+
+void shm_rbuff_close(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
+
+        free(rb);
+}
+
+void shm_rbuff_destroy(struct shm_rbuff * rb)
+{
+        char fn[FN_MAX_CHARS];
+
+        assert(rb);
+
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", rb->api, rb->port_id);
+
+        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
+        shm_unlink(fn);
+
+        free(rb);
+}
+
+int shm_rbuff_write(struct shm_rbuff * rb,
+                    size_t             idx)
+{
+        size_t ohead;
+        size_t nhead;
+
+        assert(rb);
+        assert(idx < SHM_BUFFER_SIZE);
+
+        if (__sync_fetch_and_add(rb->acl, 0)) /* CLOSED */
+                return -ENOTALLOC;
+
+        if (!shm_rbuff_free(rb))
+                return -EAGAIN;
+
+        if (shm_rbuff_empty(rb))
+                pthread_cond_broadcast(rb->add);
+
+        ohead = RB_HEAD;
+
+        *head_el_ptr(rb) = (ssize_t) idx;
+
+        do {
+                ohead = nhead;
+                nhead = (ohead + 1) & ((SHM_BUFFER_SIZE) - 1);
+                nhead = __sync_val_compare_and_swap(rb->head, ohead, nhead);
+        } while (nhead != ohead);
+
+        return 0;
+}
+
+ssize_t shm_rbuff_read(struct shm_rbuff * rb)
+{
+        size_t otail;
+        size_t ntail;
+
+        assert(rb);
+
+        if (shm_rbuff_empty(rb))
+                return -EAGAIN;
+
+        otail = RB_TAIL;
+
+        do {
+                otail = ntail;
+                ntail = (otail + 1) & ((SHM_BUFFER_SIZE) - 1);
+                ntail = __sync_val_compare_and_swap(rb->tail, otail, ntail);
+        } while (ntail != otail);
+
+        pthread_cond_broadcast(rb->del);
+
+        return *tail_el_ptr(rb);
+}
+
+ssize_t shm_rbuff_read_b(struct shm_rbuff *      rb,
+                         const struct timespec * timeout)
+{
+        struct timespec abstime;
+        ssize_t idx = -1;
+
+        assert(rb);
+
+        /* try a non-blocking read first */
+        idx = shm_rbuff_read(rb);
+        if (idx != -EAGAIN)
+                return idx;
+
+        if (timeout != NULL) {
+                clock_gettime(PTHREAD_COND_CLOCK, &abstime);
+                ts_add(&abstime, timeout, &abstime);
+        }
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
+                             (void *) rb->lock);
+
+        while (shm_rbuff_empty(rb) && (idx != -ETIMEDOUT)) {
+                if (timeout != NULL)
+                        idx = -pthread_cond_timedwait(rb->add,
+                                                      rb->lock,
+                                                      &abstime);
+                else
+                        idx = -pthread_cond_wait(rb->add, rb->lock);
+#ifndef __APPLE__
+                if (idx == -EOWNERDEAD)
+                        pthread_mutex_consistent(rb->lock);
+#endif
+        }
+
+        if (idx != -ETIMEDOUT) {
+                /* do a nonblocking read */
+                idx = shm_rbuff_read(rb);
+
+                assert(idx >= 0);
+        }
+
+        pthread_cleanup_pop(true);
+
+        return idx;
+}
+
+void shm_rbuff_block(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        __sync_bool_compare_and_swap(rb->acl, RB_OPEN, RB_CLOSED);
+}
+
+void shm_rbuff_unblock(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        __sync_bool_compare_and_swap(rb->acl, RB_CLOSED, RB_OPEN);
+}
+
+void shm_rbuff_fini(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        assert(__sync_fetch_and_add(rb->acl, 0) == RB_CLOSED);
+
+        if (shm_rbuff_empty)
+                return;
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+
+        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
+                             (void *) rb->lock);
+
+        while (!shm_rbuff_empty(rb))
+#ifdef __APPLE__
+                pthread_cond_wait(rb->del, rb->lock);
+#else
+                if (pthread_cond_wait(rb->del, rb->lock) == EOWNERDEAD)
+                        pthread_mutex_consistent(rb->lock);
+#endif
+        pthread_cleanup_pop(true);
+}
+
+size_t shm_rbuff_queued(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        return shm_rbuff_used(rb);
+}
diff --git a/src/lib/shm_rbuff_pthr.c b/src/lib/shm_rbuff_pthr.c
new file mode 100644
index 00000000..d3c1e143
--- /dev/null
+++ b/src/lib/shm_rbuff_pthr.c
@@ -0,0 +1,406 @@
+/*
+ * Ouroboros - Copyright (C) 2016 - 2017
+ *
+ * Ring buffer for incoming SDUs
+ *
+ *    Dimitri Staessens <[email protected]>
+ *    Sander Vrijders   <[email protected]>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * version 2.1 as published by the Free Software Foundation.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ */
+
+#include <ouroboros/config.h>
+#include <ouroboros/shm_rbuff.h>
+#include <ouroboros/lockfile.h>
+#include <ouroboros/time_utils.h>
+#include <ouroboros/errno.h>
+
+#include <pthread.h>
+#include <sys/mman.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <assert.h>
+#include <stdbool.h>
+
+#define FN_MAX_CHARS 255
+#define RB_OPEN 0
+#define RB_CLOSED 1
+
+#define SHM_RBUFF_FILE_SIZE ((SHM_BUFFER_SIZE) * sizeof(ssize_t)          \
+                             + 3 * sizeof(size_t)                         \
+                             + sizeof(pthread_mutex_t)                    \
+                             + 2 * sizeof (pthread_cond_t))
+
+#define shm_rbuff_used(rb) ((*rb->head + (SHM_BUFFER_SIZE) - *rb->tail)   \
+                            & ((SHM_BUFFER_SIZE) - 1))
+#define shm_rbuff_free(rb) (shm_rbuff_used(rb) + 1 < (SHM_BUFFER_SIZE))
+#define shm_rbuff_empty(rb) (*rb->head == *rb->tail)
+#define head_el_ptr(rb) (rb->shm_base + *rb->head)
+#define tail_el_ptr(rb) (rb->shm_base + *rb->tail)
+
+struct shm_rbuff {
+        ssize_t *         shm_base; /* start of entry                */
+        size_t *          head;     /* start of ringbuffer head      */
+        size_t *          tail;     /* start of ringbuffer tail      */
+        size_t *          acl;      /* access control                */
+        pthread_mutex_t * lock;     /* lock all free space in shm    */
+        pthread_cond_t *  add;      /* SDU arrived                   */
+        pthread_cond_t *  del;      /* SDU removed                   */
+        pid_t             api;      /* api of the owner              */
+        int               port_id;  /* port_id of the flow           */
+};
+
+struct shm_rbuff * shm_rbuff_create(pid_t api, int port_id)
+{
+        struct shm_rbuff *  rb;
+        int                 shm_fd;
+        ssize_t *           shm_base;
+        pthread_mutexattr_t mattr;
+        pthread_condattr_t  cattr;
+        char                fn[FN_MAX_CHARS];
+        mode_t              mask;
+
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
+
+        rb = malloc(sizeof(*rb));
+        if (rb == NULL)
+                return NULL;
+
+        mask = umask(0);
+
+        shm_fd = shm_open(fn, O_CREAT | O_EXCL | O_RDWR, 0666);
+        if (shm_fd == -1) {
+                free(rb);
+                return NULL;
+        }
+
+        umask(mask);
+
+        if (ftruncate(shm_fd, SHM_RBUFF_FILE_SIZE - 1) < 0) {
+                free(rb);
+                close(shm_fd);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_RBUFF_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                shm_unlink(fn);
+                free(rb);
+                return NULL;
+        }
+
+        rb->shm_base = shm_base;
+        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
+        rb->tail     = rb->head + 1;
+        rb->acl      = rb->tail + 1;
+        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
+        rb->add      = (pthread_cond_t *) (rb->lock + 1);
+        rb->del      = rb->add + 1;
+
+        pthread_mutexattr_init(&mattr);
+#ifndef __APPLE__
+        pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST);
+#endif
+        pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
+        pthread_mutex_init(rb->lock, &mattr);
+
+        pthread_condattr_init(&cattr);
+        pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED);
+#ifndef __APPLE__
+        pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK);
+#endif
+        pthread_cond_init(rb->add, &cattr);
+        pthread_cond_init(rb->del, &cattr);
+
+        *rb->acl = RB_OPEN;
+        *rb->head = 0;
+        *rb->tail = 0;
+
+        rb->api = api;
+        rb->port_id = port_id;
+
+        return rb;
+}
+
+struct shm_rbuff * shm_rbuff_open(pid_t api, int port_id)
+{
+        struct shm_rbuff * rb;
+        int                shm_fd;
+        ssize_t *          shm_base;
+        char               fn[FN_MAX_CHARS];
+
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", api, port_id);
+
+        rb = malloc(sizeof(*rb));
+        if (rb == NULL)
+                return NULL;
+
+        shm_fd = shm_open(fn, O_RDWR, 0666);
+        if (shm_fd == -1) {
+                free(rb);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_RBUFF_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                shm_unlink(fn);
+                free(rb);
+                return NULL;
+        }
+
+        rb->shm_base = shm_base;
+        rb->head     = (size_t *) (rb->shm_base + (SHM_BUFFER_SIZE));
+        rb->tail     = rb->head + 1;
+        rb->acl      = rb->tail + 1;
+        rb->lock     = (pthread_mutex_t *) (rb->acl + 1);
+        rb->add      = (pthread_cond_t *) (rb->lock + 1);
+        rb->del      = rb->add + 1;
+
+        rb->api = api;
+        rb->port_id = port_id;
+
+        return rb;
+}
+
+void shm_rbuff_close(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
+
+        free(rb);
+}
+
+void shm_rbuff_destroy(struct shm_rbuff * rb)
+{
+        char fn[FN_MAX_CHARS];
+
+        assert(rb);
+
+#ifdef CONFIG_OUROBOROS_DEBUG
+        pthread_mutex_lock(rb->lock);
+
+        assert(shm_rbuff_empty(rb));
+
+        pthread_mutex_unlock(rb->lock);
+#endif
+        sprintf(fn, SHM_RBUFF_PREFIX "%d.%d", rb->api, rb->port_id);
+
+        munmap(rb->shm_base, SHM_RBUFF_FILE_SIZE);
+        shm_unlink(fn);
+
+        free(rb);
+}
+
+int shm_rbuff_write(struct shm_rbuff * rb, size_t idx)
+{
+        assert(rb);
+        assert(idx < SHM_BUFFER_SIZE);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        if (*rb->acl == RB_CLOSED) {
+                pthread_mutex_unlock(rb->lock);
+                return -ENOTALLOC;
+        }
+
+        if (!shm_rbuff_free(rb)) {
+                pthread_mutex_unlock(rb->lock);
+                return -EAGAIN;
+        }
+
+        if (shm_rbuff_empty(rb))
+                pthread_cond_broadcast(rb->add);
+
+        *head_el_ptr(rb) = (ssize_t) idx;
+        *rb->head = (*rb->head + 1) & ((SHM_BUFFER_SIZE) -1);
+
+        pthread_mutex_unlock(rb->lock);
+
+        return 0;
+}
+
+ssize_t shm_rbuff_read(struct shm_rbuff * rb)
+{
+        ssize_t ret = 0;
+
+        assert(rb);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        if (shm_rbuff_empty(rb)) {
+                pthread_mutex_unlock(rb->lock);
+                return -EAGAIN;
+        }
+
+        ret = *tail_el_ptr(rb);
+        *rb->tail = (*rb->tail + 1) & ((SHM_BUFFER_SIZE) - 1);
+        pthread_cond_broadcast(rb->del);
+
+        pthread_mutex_unlock(rb->lock);
+
+        return ret;
+}
+
+ssize_t shm_rbuff_read_b(struct shm_rbuff *      rb,
+                         const struct timespec * timeout)
+{
+        struct timespec abstime;
+        ssize_t idx = -1;
+
+        assert(rb);
+
+        if (timeout != NULL) {
+                clock_gettime(PTHREAD_COND_CLOCK, &abstime);
+                ts_add(&abstime, timeout, &abstime);
+        }
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
+                             (void *) rb->lock);
+
+        while (shm_rbuff_empty(rb) && (idx != -ETIMEDOUT)) {
+                if (timeout != NULL)
+                        idx = -pthread_cond_timedwait(rb->add,
+                                                      rb->lock,
+                                                      &abstime);
+                else
+                        idx = -pthread_cond_wait(rb->add, rb->lock);
+#ifndef __APPLE__
+                if (idx == -EOWNERDEAD)
+                        pthread_mutex_consistent(rb->lock);
+#endif
+        }
+
+        if (idx != -ETIMEDOUT) {
+                idx = *tail_el_ptr(rb);
+                *rb->tail = (*rb->tail + 1) & ((SHM_BUFFER_SIZE) - 1);
+                pthread_cond_broadcast(rb->del);
+        }
+
+        pthread_cleanup_pop(true);
+
+        return idx;
+}
+
+void shm_rbuff_block(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        *rb->acl = RB_CLOSED;
+
+        pthread_mutex_unlock(rb->lock);
+}
+
+void shm_rbuff_unblock(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        *rb->acl = RB_OPEN;
+
+        pthread_mutex_unlock(rb->lock);
+}
+
+void shm_rbuff_fini(struct shm_rbuff * rb)
+{
+        assert(rb);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+        assert(*rb->acl == RB_CLOSED);
+
+        pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock,
+                             (void *) rb->lock);
+
+        while (!shm_rbuff_empty(rb))
+#ifdef __APPLE__
+                pthread_cond_wait(rb->del, rb->lock);
+#else
+                if (pthread_cond_wait(rb->del, rb->lock) == EOWNERDEAD)
+                        pthread_mutex_consistent(rb->lock);
+#endif
+        pthread_cleanup_pop(true);
+}
+
+size_t shm_rbuff_queued(struct shm_rbuff * rb)
+{
+        size_t ret;
+
+        assert(rb);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(rb->lock);
+#else
+        if (pthread_mutex_lock(rb->lock) == EOWNERDEAD)
+                pthread_mutex_consistent(rb->lock);
+#endif
+
+        ret = shm_rbuff_used(rb);
+
+        pthread_mutex_unlock(rb->lock);
+
+        return ret;
+}