/* * Ouroboros - Copyright (C) 2016 * * Random Deletion Ring Buffer for Data Units * * Dimitri Staessens <dimitri.staessens@intec.ugent.be> * Sander Vrijders <sander.vrijders@intec.ugent.be> * * 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/errno.h> #include <ouroboros/shm_rdrbuff.h> #include <ouroboros/time_utils.h> #include <pthread.h> #include <sys/mman.h> #include <fcntl.h> #include <unistd.h> #include <stdlib.h> #include <string.h> #include <signal.h> #include <sys/stat.h> #include <stdbool.h> #include <assert.h> #define OUROBOROS_PREFIX "shm_rdrbuff" #include <ouroboros/logs.h> #define SHM_BLOCKS_SIZE (SHM_BUFFER_SIZE * SHM_RDRB_BLOCK_SIZE) #define SHM_FILE_SIZE (SHM_BLOCKS_SIZE + 3 * sizeof (size_t) \ + sizeof(pthread_mutex_t) + 2 * sizeof(pthread_cond_t) \ + sizeof(pid_t)) #define get_head_ptr(rdrb) \ ((struct shm_du_buff *)(rdrb->shm_base + (*rdrb->ptr_head * \ SHM_RDRB_BLOCK_SIZE))) #define get_tail_ptr(rdrb) \ ((struct shm_du_buff *)(rdrb->shm_base + (*rdrb->ptr_tail * \ SHM_RDRB_BLOCK_SIZE))) #define idx_to_du_buff_ptr(rdrb, idx) \ ((struct shm_du_buff *)(rdrb->shm_base + (idx * SHM_RDRB_BLOCK_SIZE))) #define block_ptr_to_idx(rdrb, sdb) \ (((uint8_t *)sdb - rdrb->shm_base) / SHM_RDRB_BLOCK_SIZE) #define shm_rdrb_used(rdrb) \ ((*rdrb->ptr_head + SHM_BUFFER_SIZE - *rdrb->ptr_tail) \ & (SHM_BUFFER_SIZE - 1)) #define shm_rdrb_free(rdrb, i) \ (shm_rdrb_used(rdrb) + i < SHM_BUFFER_SIZE) #define shm_rdrb_empty(rdrb) \ (*rdrb->ptr_tail == *rdrb->ptr_head) struct shm_du_buff { size_t size; #ifdef SHM_RDRB_MULTI_BLOCK size_t blocks; #endif size_t du_head; size_t du_tail; pid_t dst_api; size_t idx; }; struct shm_rdrbuff { uint8_t * shm_base; /* start of blocks */ size_t * ptr_head; /* start of ringbuffer head */ size_t * ptr_tail; /* start of ringbuffer tail */ pthread_mutex_t * lock; /* lock all free space in shm */ size_t * choked; /* stale sdu detection */ pthread_cond_t * healthy; /* du map is healthy */ pthread_cond_t * full; /* run sanitizer when buffer full */ pid_t * api; /* api of the irmd owner */ enum qos_cube qos; /* qos id which this buffer serves */ int fd; }; static void garbage_collect(struct shm_rdrbuff * rdrb) { #ifdef SHM_RDRB_MULTI_BLOCK struct shm_du_buff * sdb; while (!shm_rdrb_empty(rdrb) && (sdb = get_tail_ptr(rdrb))->dst_api == -1) *rdrb->ptr_tail = (*rdrb->ptr_tail + sdb->blocks) & (SHM_BUFFER_SIZE - 1); #else while (!shm_rdrb_empty(rdrb) && get_tail_ptr(rdrb)->dst_api == -1) *rdrb->ptr_tail = (*rdrb->ptr_tail + 1) & (SHM_BUFFER_SIZE - 1); #endif } static void clean_sdus(struct shm_rdrbuff * rdrb, pid_t api) { size_t idx = *rdrb->ptr_tail; struct shm_du_buff * buf; while (idx != *rdrb->ptr_head) { buf = idx_to_du_buff_ptr(rdrb, idx); if (buf->dst_api == api) buf->dst_api = -1; #ifdef SHM_RDRB_MULTI_BLOCK idx = (idx + buf->blocks) & (SHM_BUFFER_SIZE - 1); #else idx = (idx + 1) & (SHM_BUFFER_SIZE - 1); #endif } garbage_collect(rdrb); *rdrb->choked = 0; } static char * rdrb_filename(enum qos_cube qos) { int chars = 0; char * str; int qm = QOS_MAX; do { qm /= 10; ++chars; } while (qm > 0); str = malloc(strlen(SHM_RDRB_PREFIX) + chars + 1); if (str == NULL) { LOG_ERR("Failed to create shm_rdrbuff: Out of Memory."); return NULL; } sprintf(str, "%s%d", SHM_RDRB_PREFIX, (int) qos); return str; } /* FIXME: create a ringbuffer for each qos cube in the system */ struct shm_rdrbuff * shm_rdrbuff_create() { struct shm_rdrbuff * rdrb; mode_t mask; int shm_fd; uint8_t * shm_base; pthread_mutexattr_t mattr; pthread_condattr_t cattr; enum qos_cube qos = QOS_CUBE_BE; char * shm_rdrb_fn = rdrb_filename(qos); if (shm_rdrb_fn == NULL) { LOG_ERR("Could not create rdrbuff. Out of Memory"); return NULL; } rdrb = malloc(sizeof *rdrb); if (rdrb == NULL) { LOG_DBGF("Could not allocate struct."); return NULL; } mask = umask(0); shm_fd = shm_open(shm_rdrb_fn, O_CREAT | O_EXCL | O_RDWR, 0666); if (shm_fd == -1) { LOG_DBGF("Failed creating shared memory map."); free(shm_rdrb_fn); free(rdrb); return NULL; } umask(mask); if (ftruncate(shm_fd, SHM_FILE_SIZE - 1) < 0) { LOG_DBGF("Failed to extend shared memory map."); free(shm_rdrb_fn); free(rdrb); return NULL; } #ifndef __APPLE if (write(shm_fd, "", 1) != 1) { LOG_DBGF("Failed to finalise extension of shared memory map."); free(shm_rdrb_fn); free(rdrb); return NULL; } #endif shm_base = mmap(NULL, SHM_FILE_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0); if (shm_base == MAP_FAILED) { LOG_DBGF("Failed to map shared memory."); if (shm_unlink(shm_rdrb_fn) == -1) LOG_DBGF("Failed to remove invalid shm."); free(shm_rdrb_fn); free(rdrb); return NULL; } rdrb->shm_base = shm_base; rdrb->ptr_head = (size_t *) ((uint8_t *) rdrb->shm_base + SHM_BLOCKS_SIZE); rdrb->ptr_tail = rdrb->ptr_head + 1; rdrb->lock = (pthread_mutex_t *) (rdrb->ptr_tail + 1); rdrb->choked = (size_t *) (rdrb->lock + 1); rdrb->healthy = (pthread_cond_t *) (rdrb->choked + 1); rdrb->full = rdrb->healthy + 1; rdrb->api = (pid_t *) (rdrb->full + 1); pthread_mutexattr_init(&mattr); pthread_mutexattr_setpshared(&mattr, PTHREAD_PROCESS_SHARED); #ifndef __APPLE__ pthread_mutexattr_setrobust(&mattr, PTHREAD_MUTEX_ROBUST); #endif pthread_mutex_init(rdrb->lock, &mattr); pthread_condattr_init(&cattr); pthread_condattr_setpshared(&cattr, PTHREAD_PROCESS_SHARED); pthread_cond_init(rdrb->full, &cattr); pthread_cond_init(rdrb->healthy, &cattr); *rdrb->ptr_head = 0; *rdrb->ptr_tail = 0; *rdrb->choked = 0; *rdrb->api = getpid(); rdrb->qos = qos; rdrb->fd = shm_fd; free(shm_rdrb_fn); return rdrb; } /* FIXME: open a ringbuffer for each qos cube in the system */ struct shm_rdrbuff * shm_rdrbuff_open() { struct shm_rdrbuff * rdrb; int shm_fd; uint8_t * shm_base; enum qos_cube qos = QOS_CUBE_BE; char * shm_rdrb_fn = rdrb_filename(qos); if (shm_rdrb_fn == NULL) { LOG_ERR("Could not create rdrbuff. Out of Memory"); return NULL; } rdrb = malloc(sizeof *rdrb); if (rdrb == NULL) { LOG_DBGF("Could not allocate struct."); return NULL; } shm_fd = shm_open(shm_rdrb_fn, O_RDWR, 0666); if (shm_fd < 0) { LOG_DBGF("Failed opening shared memory."); free(shm_rdrb_fn); free(rdrb); return NULL; } shm_base = mmap(NULL, SHM_FILE_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0); if (shm_base == MAP_FAILED) { LOG_DBGF("Failed to map shared memory."); if (close(shm_fd) == -1) LOG_DBG("Failed to close invalid shm."); if (shm_unlink(shm_rdrb_fn) == -1) LOG_DBG("Failed to unlink invalid shm."); free(shm_rdrb_fn); free(rdrb); return NULL; } rdrb->shm_base = shm_base; rdrb->ptr_head = (size_t *) ((uint8_t *) rdrb->shm_base + SHM_BLOCKS_SIZE); rdrb->ptr_tail = rdrb->ptr_head + 1; rdrb->lock = (pthread_mutex_t *) (rdrb->ptr_tail + 1); rdrb->choked = (size_t *) (rdrb->lock + 1); rdrb->healthy = (pthread_cond_t *) (rdrb->choked + 1); rdrb->full = rdrb->healthy + 1; rdrb->api = (pid_t *) (rdrb->full + 1); rdrb->qos = qos; rdrb->fd = shm_fd; free(shm_rdrb_fn); return rdrb; } void * shm_rdrbuff_sanitize(void * o) { struct shm_rdrbuff * rdrb = (struct shm_rdrbuff *) o; struct timespec intv = {SHM_DU_TIMEOUT_MICROS / MILLION, (SHM_DU_TIMEOUT_MICROS % MILLION) * 1000}; pid_t api; assert(o); #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_WARN("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif pthread_cleanup_push((void (*)(void *)) pthread_mutex_unlock, (void *) rdrb->lock); while (true) { int ret = 0; struct timespec now; struct timespec dl; #ifdef __APPLE__ pthread_cond_wait(rdrb->full, rdrb->lock); #else if (pthread_cond_wait(rdrb->full, rdrb->lock) == EOWNERDEAD) { LOG_WARN("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif *rdrb->choked = 1; garbage_collect(rdrb); if (shm_rdrb_empty(rdrb)) { pthread_cond_broadcast(rdrb->healthy); continue; } api = get_tail_ptr(rdrb)->dst_api; if (kill(api, 0)) { LOG_DBGF("Dead process %d left stale sdu.", api); clean_sdus(rdrb, api); pthread_cond_broadcast(rdrb->healthy); continue; } clock_gettime(CLOCK_REALTIME, &now); ts_add(&now, &intv, &dl); while (*rdrb->choked) { ret = pthread_cond_timedwait(rdrb->healthy, rdrb->lock, &dl); if (!ret) continue; #ifndef __APPLE__ if (ret == EOWNERDEAD) { LOG_WARN("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif if (ret == ETIMEDOUT) { LOG_DBGF("SDU timed out (dst: %d).", api); clean_sdus(rdrb, api); } } pthread_cond_broadcast(rdrb->healthy); } pthread_cleanup_pop(true); return (void *) 0; } void shm_rdrbuff_close(struct shm_rdrbuff * rdrb) { assert(rdrb); if (close(rdrb->fd) < 0) LOG_DBGF("Couldn't close shared memory."); if (munmap(rdrb->shm_base, SHM_FILE_SIZE) == -1) LOG_DBGF("Couldn't unmap shared memory."); free(rdrb); } void shm_rdrbuff_destroy(struct shm_rdrbuff * rdrb) { char * shm_rdrb_fn; assert(rdrb); if (getpid() != *rdrb->api && kill(*rdrb->api, 0) == 0) { LOG_DBG("Process %d tried to destroy active rdrb.", getpid()); return; } if (close(rdrb->fd) < 0) LOG_DBG("Couldn't close shared memory."); if (munmap(rdrb->shm_base, SHM_FILE_SIZE) == -1) LOG_DBG("Couldn't unmap shared memory."); shm_rdrb_fn = rdrb_filename(rdrb->qos); if (shm_rdrb_fn == NULL) { LOG_ERR("Could not create rdrbuff. Out of Memory"); return; } if (shm_unlink(shm_rdrb_fn) == -1) LOG_DBG("Failed to unlink shm."); free(rdrb); free(shm_rdrb_fn); } ssize_t shm_rdrbuff_write(struct shm_rdrbuff * rdrb, pid_t dst_api, size_t headspace, size_t tailspace, uint8_t * data, size_t len) { struct shm_du_buff * sdb; size_t size = headspace + len + tailspace; #ifdef SHM_RDRB_MULTI_BLOCK long blocks = 0; long padblocks = 0; #endif int sz = size + sizeof *sdb; uint8_t * write_pos; assert(rdrb); assert(data); #ifndef SHM_RDRB_MULTI_BLOCK if (sz > SHM_RDRB_BLOCK_SIZE) { LOG_DBGF("Multi-block SDUs disabled. Dropping."); return -1; } #endif #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_DBGF("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif #ifdef SHM_RDRB_MULTI_BLOCK while (sz > 0) { sz -= SHM_RDRB_BLOCK_SIZE; ++blocks; } if (blocks + *rdrb->ptr_head > SHM_BUFFER_SIZE) padblocks = SHM_BUFFER_SIZE - *rdrb->ptr_head; if (!shm_rdrb_free(rdrb, (blocks + padblocks))) { #else if (!shm_rdrb_free(rdrb, 1)) { #endif pthread_cond_signal(rdrb->full); pthread_mutex_unlock(rdrb->lock); return -1; } #ifdef SHM_RDRB_MULTI_BLOCK if (padblocks) { sdb = get_head_ptr(rdrb); sdb->size = 0; sdb->blocks = padblocks; sdb->dst_api = -1; sdb->du_head = 0; sdb->du_tail = 0; sdb->idx = *rdrb->ptr_head; *rdrb->ptr_head = 0; } #endif sdb = get_head_ptr(rdrb); sdb->size = size; sdb->dst_api = dst_api; sdb->du_head = headspace; sdb->du_tail = sdb->du_head + len; #ifdef SHM_RDRB_MULTI_BLOCK sdb->blocks = blocks; #endif write_pos = ((uint8_t *) (sdb + 1)) + headspace; memcpy(write_pos, data, len); sdb->idx = *rdrb->ptr_head; #ifdef SHM_RDRB_MULTI_BLOCK *rdrb->ptr_head = (*rdrb->ptr_head + blocks) & (SHM_BUFFER_SIZE - 1); #else *rdrb->ptr_head = (*rdrb->ptr_head + 1) & (SHM_BUFFER_SIZE - 1); #endif pthread_mutex_unlock(rdrb->lock); return sdb->idx; } ssize_t shm_rdrbuff_write_b(struct shm_rdrbuff * rdrb, pid_t dst_api, size_t headspace, size_t tailspace, uint8_t * data, size_t len) { struct shm_du_buff * sdb; size_t size = headspace + len + tailspace; #ifdef SHM_RDRB_MULTI_BLOCK long blocks = 0; long padblocks = 0; #endif int sz = size + sizeof *sdb; uint8_t * write_pos; assert(rdrb); assert(data); #ifndef SHM_RDRB_MULTI_BLOCK if (sz > SHM_RDRB_BLOCK_SIZE) { LOG_DBGF("Multi-block SDUs disabled. Dropping."); return -1; } #endif #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_DBGF("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock, (void *) rdrb->lock); #ifdef SHM_RDRB_MULTI_BLOCK while (sz > 0) { sz -= SHM_RDRB_BLOCK_SIZE; ++blocks; } if (blocks + *rdrb->ptr_head > SHM_BUFFER_SIZE) padblocks = SHM_BUFFER_SIZE - *rdrb->ptr_head; while (!shm_rdrb_free(rdrb, (blocks + padblocks))) { #else while (!shm_rdrb_free(rdrb, 1)) { #endif pthread_cond_signal(rdrb->full); pthread_cond_wait(rdrb->healthy, rdrb->lock); } #ifdef SHM_RDRB_MULTI_BLOCK if (padblocks) { sdb = get_head_ptr(rdrb); sdb->size = 0; sdb->blocks = padblocks; sdb->dst_api = -1; sdb->du_head = 0; sdb->du_tail = 0; sdb->idx = *rdrb->ptr_head; *rdrb->ptr_head = 0; } #endif sdb = get_head_ptr(rdrb); sdb->size = size; sdb->dst_api = dst_api; sdb->du_head = headspace; sdb->du_tail = sdb->du_head + len; #ifdef SHM_RDRB_MULTI_BLOCK sdb->blocks = blocks; #endif write_pos = ((uint8_t *) (sdb + 1)) + headspace; memcpy(write_pos, data, len); sdb->idx = *rdrb->ptr_head; #ifdef SHM_RDRB_MULTI_BLOCK *rdrb->ptr_head = (*rdrb->ptr_head + blocks) & (SHM_BUFFER_SIZE - 1); #else *rdrb->ptr_head = (*rdrb->ptr_head + 1) & (SHM_BUFFER_SIZE - 1); #endif pthread_cleanup_pop(true); return sdb->idx; } int shm_rdrbuff_read(uint8_t ** dst, struct shm_rdrbuff * rdrb, ssize_t idx) { size_t len = 0; struct shm_du_buff * sdb; assert(dst); assert(rdrb); if (idx > SHM_BUFFER_SIZE) return -1; #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_DBGF("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif if (shm_rdrb_empty(rdrb)) { pthread_mutex_unlock(rdrb->lock); return -1; } sdb = idx_to_du_buff_ptr(rdrb, idx); len = sdb->du_tail - sdb->du_head; *dst = ((uint8_t *) (sdb + 1)) + sdb->du_head; pthread_mutex_unlock(rdrb->lock); return len; } struct shm_du_buff * shm_rdrbuff_get(struct shm_rdrbuff * rdrb, ssize_t idx) { struct shm_du_buff * sdb; assert(rdrb); if (idx > SHM_BUFFER_SIZE) return NULL; #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_DBGF("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif if (shm_rdrb_empty(rdrb)) { pthread_mutex_unlock(rdrb->lock); return NULL; } sdb = idx_to_du_buff_ptr(rdrb, idx); pthread_mutex_unlock(rdrb->lock); return sdb; } int shm_rdrbuff_remove(struct shm_rdrbuff * rdrb, ssize_t idx) { assert(rdrb); if (idx > SHM_BUFFER_SIZE) return -1; #ifdef __APPLE__ pthread_mutex_lock(rdrb->lock); #else if (pthread_mutex_lock(rdrb->lock) == EOWNERDEAD) { LOG_DBGF("Recovering dead mutex."); pthread_mutex_consistent(rdrb->lock); } #endif if (shm_rdrb_empty(rdrb)) { pthread_mutex_unlock(rdrb->lock); return -1; } idx_to_du_buff_ptr(rdrb, idx)->dst_api = -1; if (idx != *rdrb->ptr_tail) { pthread_mutex_unlock(rdrb->lock); return 0; } garbage_collect(rdrb); *rdrb->choked = 0; pthread_cond_broadcast(rdrb->healthy); pthread_mutex_unlock(rdrb->lock); return 0; } size_t shm_du_buff_get_idx(struct shm_du_buff * sdb) { assert(sdb); return sdb->idx; } uint8_t * shm_du_buff_head(struct shm_du_buff * sdb) { assert(sdb); return (uint8_t *) (sdb + 1) + sdb->du_head; } uint8_t * shm_du_buff_tail(struct shm_du_buff * sdb) { assert(sdb); return (uint8_t *) (sdb + 1) + sdb->du_tail; } uint8_t * shm_du_buff_head_alloc(struct shm_du_buff * sdb, size_t size) { uint8_t * buf = NULL; assert(sdb); if ((long) (sdb->du_head - size) < 0) { LOG_ERR("Failed to allocate PCI headspace."); return NULL; } sdb->du_head -= size; buf = (uint8_t *) (sdb + 1) + sdb->du_head; return buf; } uint8_t * shm_du_buff_tail_alloc(struct shm_du_buff * sdb, size_t size) { uint8_t * buf = NULL; assert(sdb); if (sdb->du_tail + size >= sdb->size) { LOG_ERR("Failed to allocate PCI tailspace."); return NULL; } buf = (uint8_t *) (sdb + 1) + sdb->du_tail; sdb->du_tail += size; return buf; } int shm_du_buff_head_release(struct shm_du_buff * sdb, size_t size) { assert(sdb); if (size > sdb->du_tail - sdb->du_head) { LOG_DBGF("Tried to release beyond SDU boundary."); return -EOVERFLOW; } sdb->du_head += size; return 0; } int shm_du_buff_tail_release(struct shm_du_buff * sdb, size_t size) { assert(sdb); if (size > sdb->du_tail - sdb->du_head) { LOG_ERR("Tried to release beyond SDU boundary."); return -EOVERFLOW; } sdb->du_tail -= size; return 0; }