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, 408 insertions, 0 deletions

diff --git a/src/lib/shm_flow_set.c b/src/lib/shm_flow_set.c
new file mode 100644
index 00000000..c960bd25
--- /dev/null
+++ b/src/lib/shm_flow_set.c
@@ -0,0 +1,408 @@
+/*
+ * Ouroboros - Copyright (C) 2016
+ *
+ * Management of flow_sets for fqueue
+ *
+ *    Dimitri Staessens <[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
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <ouroboros/config.h>
+#include <ouroboros/lockfile.h>
+#include <ouroboros/time_utils.h>
+#include <ouroboros/shm_flow_set.h>
+#include <ouroboros/fqueue.h>
+#include <ouroboros/errno.h>
+
+#define OUROBOROS_PREFIX "shm_flow_set"
+
+#include <ouroboros/logs.h>
+
+#include <pthread.h>
+#include <sys/mman.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <string.h>
+#include <assert.h>
+
+#define FN_MAX_CHARS 255
+
+#define FQUEUESIZE (SHM_BUFFER_SIZE * sizeof(int))
+
+#define SHM_FLOW_SET_FILE_SIZE (IRMD_MAX_FLOWS * sizeof(ssize_t)          \
+                                + AP_MAX_FQUEUES * sizeof(size_t)         \
+                                + AP_MAX_FQUEUES * sizeof(pthread_cond_t) \
+                                + AP_MAX_FQUEUES * FQUEUESIZE             \
+                                + sizeof(pthread_mutex_t))
+
+#define fqueue_ptr(fs, idx) (fs->fqueues + SHM_BUFFER_SIZE * idx)
+
+struct shm_flow_set {
+        ssize_t *         mtable;
+        size_t *          heads;
+        pthread_cond_t *  conds;
+        int *             fqueues;
+        pthread_mutex_t * lock;
+
+        pid_t             api;
+};
+
+struct shm_flow_set * shm_flow_set_create()
+{
+        struct shm_flow_set * set;
+        ssize_t *             shm_base;
+        pthread_mutexattr_t   mattr;
+        pthread_condattr_t    cattr;
+        char                  fn[FN_MAX_CHARS];
+        mode_t                mask;
+        int                   shm_fd;
+        int                   i;
+
+        sprintf(fn, SHM_FLOW_SET_PREFIX "%d", getpid());
+
+        set = malloc(sizeof(*set));
+        if (set == NULL) {
+                LOG_DBG("Could not allocate struct.");
+                return NULL;
+        }
+
+        mask = umask(0);
+
+        shm_fd = shm_open(fn, O_CREAT | O_EXCL | O_RDWR, 0666);
+        if (shm_fd == -1) {
+                LOG_DBG("Failed creating flag file.");
+                free(set);
+                return NULL;
+        }
+
+        umask(mask);
+
+        if (ftruncate(shm_fd, SHM_FLOW_SET_FILE_SIZE - 1) < 0) {
+                LOG_DBG("Failed to extend flag file.");
+                free(set);
+                close(shm_fd);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_FLOW_SET_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                LOG_DBG("Failed to map shared memory.");
+                if (shm_unlink(fn) == -1)
+                        LOG_DBG("Failed to remove invalid shm.");
+
+                free(set);
+                return NULL;
+        }
+
+        set->mtable  = shm_base;
+        set->heads   = (size_t *) (set->mtable + IRMD_MAX_FLOWS);
+        set->conds   = (pthread_cond_t *)(set->heads + AP_MAX_FQUEUES);
+        set->fqueues = (int *) (set->conds + AP_MAX_FQUEUES);
+        set->lock    = (pthread_mutex_t *)
+                (set->fqueues + AP_MAX_FQUEUES * SHM_BUFFER_SIZE);
+
+        pthread_mutexattr_init(&mattr);
+#ifndef __APPLE__
+        pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST);
+#endif
+        pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED);
+        pthread_mutex_init(set->lock, &mattr);
+
+        pthread_condattr_init(&cattr);
+        pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED);
+#ifndef __APPLE__
+        pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK);
+#endif
+        for (i = 0; i < AP_MAX_FQUEUES; ++i) {
+                set->heads[i] = 0;
+                pthread_cond_init(&set->conds[i], &cattr);
+        }
+
+        for (i = 0; i < IRMD_MAX_FLOWS; ++i)
+                set->mtable[i] = -1;
+
+        set->api = getpid();
+
+        return set;
+}
+
+struct shm_flow_set * shm_flow_set_open(pid_t api)
+{
+        struct shm_flow_set * set;
+        ssize_t *             shm_base;
+        char                  fn[FN_MAX_CHARS];
+        int                   shm_fd;
+
+        sprintf(fn, SHM_FLOW_SET_PREFIX "%d", api);
+
+        set = malloc(sizeof(*set));
+        if (set == NULL) {
+                LOG_DBG("Could not allocate struct.");
+                return NULL;
+        }
+
+        shm_fd = shm_open(fn, O_RDWR, 0666);
+        if (shm_fd == -1) {
+                LOG_DBG("%d failed opening shared memory %s.", getpid(), fn);
+                free(set);
+                return NULL;
+        }
+
+        shm_base = mmap(NULL,
+                        SHM_FLOW_SET_FILE_SIZE,
+                        PROT_READ | PROT_WRITE,
+                        MAP_SHARED,
+                        shm_fd,
+                        0);
+
+        close(shm_fd);
+
+        if (shm_base == MAP_FAILED) {
+                LOG_DBG("Failed to map shared memory.");
+                if (shm_unlink(fn) == -1)
+                        LOG_DBG("Failed to remove invalid shm.");
+                free(set);
+                return NULL;
+        }
+
+        set->mtable  = shm_base;
+        set->heads   = (size_t *) (set->mtable + IRMD_MAX_FLOWS);
+        set->conds   = (pthread_cond_t *)(set->heads + AP_MAX_FQUEUES);
+        set->fqueues = (int *) (set->conds + AP_MAX_FQUEUES);
+        set->lock    = (pthread_mutex_t *)
+                (set->fqueues + AP_MAX_FQUEUES * SHM_BUFFER_SIZE);
+
+        set->api = api;
+
+        return set;
+}
+
+void shm_flow_set_destroy(struct shm_flow_set * set)
+{
+        char fn[25];
+        struct lockfile * lf = NULL;
+
+        assert(set);
+
+        if (set->api != getpid()) {
+                lf = lockfile_open();
+                if (lf == NULL) {
+                        LOG_ERR("Failed to open lockfile.");
+                        return;
+                }
+
+                if (lockfile_owner(lf) == getpid()) {
+                        LOG_DBG("Flow set %d destroyed by IRMd %d.",
+                                set->api, getpid());
+                        lockfile_close(lf);
+                } else {
+                        LOG_ERR("AP-I %d tried to destroy flowset owned by %d.",
+                                getpid(), set->api);
+                        lockfile_close(lf);
+                        return;
+                }
+        }
+
+        sprintf(fn, SHM_FLOW_SET_PREFIX "%d", set->api);
+
+        if (munmap(set->mtable, SHM_FLOW_SET_FILE_SIZE) == -1)
+                LOG_DBG("Couldn't unmap shared memory.");
+
+        if (shm_unlink(fn) == -1)
+                LOG_DBG("Failed to unlink shm.");
+
+        free(set);
+}
+
+void shm_flow_set_close(struct shm_flow_set * set)
+{
+        assert(set);
+
+        if (munmap(set->mtable, SHM_FLOW_SET_FILE_SIZE) == -1)
+                LOG_DBG("Couldn't unmap shared memory.");
+
+        free(set);
+}
+
+void shm_flow_set_zero(struct shm_flow_set * shm_set,
+                       ssize_t               idx)
+{
+        ssize_t i = 0;
+
+        assert(!(idx < 0) && idx < AP_MAX_FQUEUES);
+
+        pthread_mutex_lock(shm_set->lock);
+
+        for (i = 0; i < IRMD_MAX_FLOWS; ++i)
+                if (shm_set->mtable[i] == idx)
+                        shm_set->mtable[i] = -1;
+
+        shm_set->heads[idx] = 0;
+
+        pthread_mutex_unlock(shm_set->lock);
+}
+
+
+int shm_flow_set_add(struct shm_flow_set * shm_set,
+                     ssize_t               idx,
+                     int                   port_id)
+{
+        assert(shm_set);
+        assert(!(port_id < 0) && port_id < IRMD_MAX_FLOWS);
+        assert(!(idx < 0) && idx < AP_MAX_FQUEUES);
+
+        pthread_mutex_lock(shm_set->lock);
+
+        if (shm_set->mtable[port_id] != -1) {
+                pthread_mutex_unlock(shm_set->lock);
+                return -EPERM;
+        }
+
+        shm_set->mtable[port_id] = idx;
+
+        pthread_mutex_unlock(shm_set->lock);
+
+        return 0;
+}
+
+void shm_flow_set_del(struct shm_flow_set * shm_set,
+                      ssize_t               idx,
+                      int                   port_id)
+{
+        assert(shm_set);
+        assert(!(port_id < 0) && port_id < IRMD_MAX_FLOWS);
+        assert(!(idx < 0) && idx < AP_MAX_FQUEUES);
+
+        pthread_mutex_lock(shm_set->lock);
+
+        if (shm_set->mtable[port_id] == idx)
+                shm_set->mtable[port_id] = -1;
+
+        pthread_mutex_unlock(shm_set->lock);
+}
+
+int shm_flow_set_has(struct shm_flow_set * shm_set,
+                     ssize_t               idx,
+                     int                   port_id)
+{
+        int ret = 0;
+
+        assert(shm_set);
+        assert(!(port_id < 0) && port_id < IRMD_MAX_FLOWS);
+        assert(!(idx < 0) && idx < AP_MAX_FQUEUES);
+
+
+        pthread_mutex_lock(shm_set->lock);
+
+        if (shm_set->mtable[port_id] == idx)
+                ret = 1;
+
+        pthread_mutex_unlock(shm_set->lock);
+
+        return ret;
+}
+
+void shm_flow_set_notify(struct shm_flow_set * shm_set, int port_id)
+{
+        assert(shm_set);
+        assert(!(port_id < 0) && port_id < IRMD_MAX_FLOWS);
+
+        pthread_mutex_lock(shm_set->lock);
+
+        if (shm_set->mtable[port_id] == -1) {
+                pthread_mutex_unlock(shm_set->lock);
+                return;
+        }
+
+        *(fqueue_ptr(shm_set, shm_set->mtable[port_id]) +
+                     (shm_set->heads[shm_set->mtable[port_id]])++) = port_id;
+
+        pthread_cond_signal(&shm_set->conds[shm_set->mtable[port_id]]);
+
+        pthread_mutex_unlock(shm_set->lock);
+}
+
+
+int shm_flow_set_wait(const struct shm_flow_set * shm_set,
+                      ssize_t                     idx,
+                      int *                       fqueue,
+                      const struct timespec *     timeout)
+{
+        int ret = 0;
+        struct timespec abstime;
+
+        assert(shm_set);
+        assert(!(idx < 0) && idx < AP_MAX_FQUEUES);
+
+#ifdef __APPLE__
+        pthread_mutex_lock(shm_set->lock);
+#else
+        if (pthread_mutex_lock(shm_set->lock) == EOWNERDEAD) {
+                LOG_DBG("Recovering dead mutex.");
+                pthread_mutex_consistent(shm_set->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 *) shm_set->lock);
+
+        while (shm_set->heads[idx] == 0 && ret != -ETIMEDOUT) {
+                if (timeout != NULL)
+                        ret = pthread_cond_timedwait(shm_set->conds + idx,
+                                                     shm_set->lock,
+                                                     &abstime);
+                else
+                        ret = pthread_cond_wait(shm_set->conds + idx,
+                                                shm_set->lock);
+#ifndef __APPLE__
+                if (ret == EOWNERDEAD) {
+                        LOG_DBG("Recovering dead mutex.");
+                        pthread_mutex_consistent(shm_set->lock);
+                }
+#endif
+                if (ret == ETIMEDOUT) {
+                        ret = -ETIMEDOUT;
+                        break;
+                }
+        }
+
+        if (ret != -ETIMEDOUT) {
+                memcpy(fqueue,
+                       fqueue_ptr(shm_set, idx),
+                       shm_set->heads[idx] * sizeof(int));
+                ret = shm_set->heads[idx];
+                shm_set->heads[idx] = 0;
+        }
+
+        pthread_cleanup_pop(true);
+
+        return ret;
+}