Support memcached

Makefile

@@ -31,6 +31,18 @@ REDIS_ZIP_NAME := 7.0-rc2
 REDIS_PATH := downloads/$(REDIS_NAME)
 REDIS := redis
 
+MEMCACHED_SOURCE := http://www.memcached.org/files/memcached-1.6.15.tar.gz
+MEMCACHED_NAME := memcached-1.6.15
+MEMCACHED_ZIP_NAME := memcached-1.6.15
+MEMCACHED_PATH := downloads/$(MEMCACHED_NAME)
+MEMCACHED := memcached
+
+LIBEVENT_SOURCE := https://github.com/libevent/libevent/archive/refs/tags/release-2.1.12-stable.tar.gz
+LIBEVENT_NAME := libevent-release-2.1.12-stable
+LIBEVENT_ZIP_NAME := release-2.1.12-stable
+LIBEVENT_PATH := downloads/$(LIBEVENT_NAME)
+LIBEVENT := libevent
+
 LIBDUMMY_PATH := $(shell find $(shell pwd) -type f -name "libdummy.so") | sed 's_/_\\/_g'
 PWD := $(shell pwd)
 OUT := downloads
@@ -89,6 +101,29 @@ $(REDIS):
 	cd $(OUT) && patch -p1 < ../patches/redis_network.patch && patch -p1 < ../patches/redis_server.patch
 	cd $(REDIS_PATH) && make -j$(nproc)
 
+$(LIBEVENT):
+	@echo "download libevent..."
+	wget $(LIBEVENT_SOURCE) --no-check-certificate
+	mkdir -p $(LIBEVENT_PATH)
+	tar -zxvf $(LIBEVENT_ZIP_NAME).tar.gz -C $(OUT)
+	rm $(LIBEVENT_ZIP_NAME).tar.gz
+	cd $(LIBEVENT_PATH) && mkdir -p local && ./autogen.sh && ./configure --prefix=$(PWD)/$(LIBEVENT_PATH)/local
+	cd $(OUT) && patch -p1 < ../patches/libevent.patch
+
+
+$(MEMCACHED): $(LIBEVENT)
+	@echo "download memcached..."
+	wget $(MEMCACHED_SOURCE)
+	mkdir -p $(MEMCACHED_PATH)
+	tar -zxvf $(MEMCACHED_ZIP_NAME).tar.gz -C $(OUT)
+	rm $(MEMCACHED_ZIP_NAME).tar.gz
+	cd $(MEMCACHED_PATH) && mkdir -p local && ./configure --prefix=$(PWD)/$(MEMCACHED_PATH)/local
+	cd $(OUT) && patch -p1 < ../patches/memcached.patch
+	scripts/libevent.sh $(LIBEVENT_PATH)
+	cd $(LIBEVENT_PATH) && make -j$(nproc) && sudo make install
+	scripts/memcached.sh $(MEMCACHED_PATH)
+	cd $(MEMCACHED_PATH) && make -j$(nproc)
+
 test-nginx-perf:
 	./$(NGX_PATH)/objs/nginx
 
@@ -119,6 +154,8 @@ else ifeq ($(strip $(TARGET)),redis)
 TARGET = redis
 else ifeq ($(strip $(TARGET)),echo)
 TARGET = echo
+else ifeq ($(strip $(TARGET)),mcached)
+TARGET = mcached
 endif
 
 # set configuration flag of Nginx

module/include/aarch64_syscall.h

@@ -2,4 +2,5 @@
 #define __ESCA_shutdown 210
 #define __ESCA_close 57
 #define __ESCA_write 64
-#define __ESCA_sendto 206
+#define __ESCA_sendto 206
+#define __ESCA_sendmsg 211

module/include/esca.h

@@ -8,7 +8,7 @@
 
 #include <stdatomic.h>
 
-#define DEFAULT_CONFIG_PATH ""
+#define DEFAULT_CONFIG_PATH "/home/eecheng/SAIO/esca.conf"
 #define CONFIG_ARG_MAX_BYTES 128
 
 /* Limit */

module/include/x86_syscall.h

@@ -2,4 +2,5 @@
 #define __ESCA_shutdown 48
 #define __ESCA_close 3
 #define __ESCA_write 1
-#define __ESCA_sendto 44
+#define __ESCA_sendto 44
+#define __ESCA_sendmsg 46

patches/libevent.patch

@@ -0,0 +1,19 @@
+--- old/libevent-release-2.1.12-stable/event.c	2022-06-09 01:52:06.746406656 +0000
++++ new/libevent-release-2.1.12-stable/event.c	2022-06-09 01:51:04.594620903 +0000
+@@ -1639,7 +1639,7 @@
+ 	int count = 0;
+
+ 	EVUTIL_ASSERT(activeq != NULL);
+-
++batch_start();
+ 	for (evcb = TAILQ_FIRST(activeq); evcb; evcb = TAILQ_FIRST(activeq)) {
+ 		struct event *ev=NULL;
+ 		if (evcb->evcb_flags & EVLIST_INIT) {
+@@ -1747,6 +1747,7 @@
+ 		if (base->event_continue)
+ 			break;
+ 	}
++batch_flush();
+ 	return count;
+ }
+

patches/memcached.patch

@@ -0,0 +1,20 @@
+--- old/memcached-1.6.15/thread.c	2022-06-09 01:59:43.605398261 +0000
++++ new/memcached-1.6.15/thread.c	2022-06-09 02:01:16.581275340 +0000
+@@ -497,6 +497,8 @@
+ /*
+  * Worker thread: main event loop
+  */
++extern void init_worker(int);
++int worker_index = 0;
+ static void *worker_libevent(void *arg) {
+     LIBEVENT_THREAD *me = arg;
+
+@@ -514,7 +516,7 @@
+     }
+
+     register_thread_initialized();
+-
++    init_worker(worker_index++);
+     event_base_loop(me->base, 0);
+
+     // same mechanism used to watch for all threads exiting.

scripts/libevent.sh

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+
+lib_saio_dir=$(find $(pwd) -type d -name "wrapper" | sed 's_/_\\/_g')
+libpath=$1
+
+# modify libevent
+sed -i "s/LDFLAGS =.*/& -Wl,-rpath -Wl,${lib_saio_dir}/" ${libpath}/Makefile
+sed -i "s/LIBS =.*/& -L${lib_saio_dir} -ldummy/" ${libpath}/Makefile

scripts/memcached.sh

@@ -0,0 +1,10 @@
+#!/usr/bin/env bash
+
+lib_saio_dir=$(find $(pwd) -type d -name "wrapper" | sed 's_/_\\/_g')
+libevent_dir=$(find $(pwd) -type d -path "*/local/lib" | sed 's_/_\\/_g')
+mempath=$1
+
+# modify memcached
+sed -i "s/LDFLAGS =.*/& -Wl,-rpath -Wl,${lib_saio_dir}/" ${mempath}/Makefile
+sed -i "s/LDFLAGS =.*/& -Wl,-rpath -Wl,${libevent_dir}/" ${mempath}/Makefile
+sed -i "s/LIBS =.*/& -L${lib_saio_dir} -ldummy/" ${mempath}/Makefile

wrapper/mcached.c

@@ -0,0 +1,65 @@
+ssize_t sendmsg(int sockfd, const struct msghdr* msg, int flags)
+{
+    if (!in_segment) {
+        return real_sendmsg(sockfd, msg, flags);
+    }
+
+    int idx = this_worker_id * RATIO + (sockfd % RATIO);
+    batch_num++;
+
+    int i = table[idx]->tail_table;
+    int j = table[idx]->tail_entry;
+
+    /* check out of bound */
+    msg_offset = msg_offset & MSG_MASK;
+    iov_offset = iov_offset & IOV_MASK;
+
+    int res = 0;
+    struct iovec* iov = msg->msg_iov;
+    int iovlen = msg->msg_iovlen, start_iov_offset = iov_offset;
+
+    for (int k = 0; k < iovlen; k++) {
+        int ll = iov[k].iov_len; // number of bytes
+
+        /* handle string */
+        if (pool_offset + (ull)(ll) > MAX_POOL_SIZE)
+            pool_offset = 0;
+
+        memcpy((void*)((unsigned char*)mpool + pool_offset), iov[k].iov_base, ll);
+
+        iov_offset = iov_offset & IOV_MASK;
+        iovpool[iov_offset].iov_base = (void*)((unsigned char*)mpool + pool_offset);
+        iovpool[iov_offset].iov_len = ll;
+
+        /* update index of iovec array */
+        iov_offset++;
+
+        /* update offset of character pool */
+        pool_offset += (size_t)(ll);
+
+        res += ll;
+    }
+
+    /* copy msghdr */
+    msgpool[msg_offset].msg_name = NULL; // FIXME: should be more general
+    msgpool[msg_offset].msg_namelen = 0;
+    msgpool[msg_offset].msg_iov = (struct iovec*)(&iovpool[start_iov_offset]);
+    msgpool[msg_offset].msg_iovlen = iovlen;
+    msgpool[msg_offset].msg_control = NULL;
+    msgpool[msg_offset].msg_controllen = 0;
+    msgpool[msg_offset].msg_flags = msg->msg_flags;
+
+    table[idx]->user_tables[i][j].sysnum = __ESCA_sendmsg;
+    table[idx]->user_tables[i][j].nargs = 3;
+    table[idx]->user_tables[i][j].args[0] = sockfd;
+    table[idx]->user_tables[i][j].args[1] = (long)(&msgpool[msg_offset]);
+    table[idx]->user_tables[i][j].args[2] = flags;
+
+    msg_offset++;
+    update_index(idx);
+
+    esca_smp_store_release(&table[idx]->user_tables[i][j].rstatus, BENTRY_BUSY);
+
+    /* assume success */
+    return res;
+}

wrapper/preload.c

@@ -14,9 +14,11 @@ int syscall_num; /* number of syscall triggered currently */
 size_t pgsize;
 
 void* mpool; /* memory pool */
-ull pool_offset;
 struct iovec* iovpool; /* pool for iovector */
+struct msghdr* msgpool; /* pool for msgpool */
+ull pool_offset;
 ull iov_offset;
+ull msg_offset;
 
 /* Global configurable variable */
 int ESCA_LOCALIZE;
@@ -78,6 +80,8 @@ void init_worker(int idx)
     pool_offset = 0;
     iovpool = (struct iovec*)malloc(sizeof(struct iovec) * MAX_POOL_IOV_SIZE);
     iov_offset = 0;
+    msgpool = (struct msghdr*)malloc(sizeof(struct msghdr) * MAX_POOL_MSG_SIZE);
+    msg_offset = 0;
 
 init_worker_exit:
     return;
@@ -170,6 +174,7 @@ __attribute__((constructor)) static void setup(void)
     real_writev = real_writev ? real_writev : dlsym(RTLD_NEXT, "writev");
     real_shutdown = real_shutdown ? real_shutdown : dlsym(RTLD_NEXT, "shutdown");
     real_sendfile = real_sendfile ? real_sendfile : dlsym(RTLD_NEXT, "sendfile");
+    real_sendmsg = real_sendmsg ? real_sendmsg : dlsym(RTLD_NEXT, "sendmsg");
 
     /* configuration */
     config = malloc(sizeof(esca_config_t));

wrapper/preload.h

@@ -3,7 +3,12 @@
 #define DEPLOY_TAGET 1
 #define MAX_TABLE_SIZE 64
 #define MAX_POOL_SIZE 900000000
-#define MAX_POOL_IOV_SIZE 1000
+
+/* optimize: order two */
+#define MAX_POOL_IOV_SIZE 1024
+#define MAX_POOL_MSG_SIZE 1024
+#define IOV_MASK (MAX_POOL_IOV_SIZE - 1)
+#define MSG_MASK (MAX_POOL_MSG_SIZE - 1)
 #define POOL_UNIT 8
 
 #include "../module/include/esca.h"
@@ -15,7 +20,9 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/socket.h>
 #include <sys/stat.h>
+#include <sys/types.h>
 #include <sys/uio.h>
 #include <time.h>
 
@@ -35,3 +42,5 @@ typedef long (*shutdown_t)(int fd, int how);
 shutdown_t real_shutdown;
 typedef long (*sendfile_t)(int out_fd, int in_fd, off_t* offset, size_t count);
 sendfile_t real_sendfile;
+typedef long (*sendmsg_t)(int sockfd, const struct msghdr* msg, int flags);
+sendmsg_t real_sendmsg;