vitalif
/
mtcp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
mtcp/mtcp/src/io_module.c

740 lines
20 KiB
C
Raw Blame History

/* for I/O module def'ns */
#include "io_module.h"
/* for num_devices decl */
#include "config.h"
/* std lib funcs */
#include <stdlib.h>
/* std io funcs */
#include <stdio.h>
/* strcmp func etc. */
#include <string.h>
/* for ifreq struct */
#include <net/if.h>
/* for ioctl */
#include <sys/ioctl.h>
#ifndef DISABLE_DPDK
#define RTE_ARGC_MAX (RTE_MAX_ETHPORTS << 1) + 9
/* for dpdk ethernet functions (get mac addresses) */
#include <rte_ethdev.h>
#include <dpdk_iface_common.h>
/* for ceil func */
#include <math.h>
/* for retrieving rte version(s) */
#include <rte_version.h>
#endif /* DISABLE_DPDK */
/* for TRACE_* */
#include "debug.h"
/* for inet_* */
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
/* for getopt() */
#include <unistd.h>
/* for getifaddrs */
#include <sys/types.h>
#include <ifaddrs.h>
#ifdef ENABLE_ONVM
/* for onvm */
#include "onvm_nflib.h"
#include "onvm_pkt_helper.h"
/* for dpdk/onvm big ints */
#include <gmp.h>
#endif
/* for file opening */
#include <sys/stat.h>
#include <fcntl.h>
/* for netmap macros */
#include "netmap_user.h"
/*----------------------------------------------------------------------------*/
io_module_func *current_iomodule_func = &dpdk_module_func;
#ifndef DISABLE_DPDK
enum rte_proc_type_t eal_proc_type_detect(void);
/**
* DPDK's RTE consumes some huge pages for internal bookkeeping.
* Therefore, it is not always safe to reserve the exact amount
* of pages for our stack (e.g. dividing requested mem, in MB, by
* (1<<20) would be insufficient). Hence, the following value.
*/
#define RTE_SOCKET_MEM_SHIFT ((1<<19)|(1<<18))
#endif
/*----------------------------------------------------------------------------*/
#define ALL_STRING "all"
#define MAX_PROCLINE_LEN 1024
#define MAX(a, b) ((a)>(b)?(a):(b))
#define MIN(a, b) ((a)<(b)?(a):(b))
/*----------------------------------------------------------------------------*/
/* onvm struct for port info lookup */
extern struct port_info *ports;
#ifndef DISABLE_PSIO
static int
GetNumQueues()
{
FILE *fp;
char buf[MAX_PROCLINE_LEN];
int queue_cnt;
fp = fopen("/proc/interrupts", "r");
if (!fp) {
TRACE_CONFIG("Failed to read data from /proc/interrupts!\n");
return -1;
}
/* count number of NIC queues from /proc/interrupts */
queue_cnt = 0;
while (!feof(fp)) {
if (fgets(buf, MAX_PROCLINE_LEN, fp) == NULL)
break;
/* "xge0-rx" is the keyword for counting queues */
if (strstr(buf, "xge0-rx")) {
queue_cnt++;
}
}
fclose(fp);
return queue_cnt;
}
#endif /* !PSIO */
/*----------------------------------------------------------------------------*/
#ifndef DISABLE_DPDK
/**
* returns max numa ID while probing for rte devices
*/
static int
probe_all_rte_devices(char **argv, int *argc, char *dev_name_list)
{
PciDevice pd;
int fd, numa_id = -1;
static char end[] = "";
static const char delim[] = " \t";
static char *dev_tokenizer;
char *dev_token, *saveptr;
dev_tokenizer = strdup(dev_name_list);
if (dev_tokenizer == NULL) {
TRACE_ERROR("Can't allocate memory for dev_tokenizer!\n");
exit(EXIT_FAILURE);
}
fd = open(DEV_PATH, O_RDONLY);
if (fd == -1) {
TRACE_ERROR("Error opening dpdk-face!\n");
exit(EXIT_FAILURE);
}
dev_token = strtok_r(dev_tokenizer, delim, &saveptr);
while (dev_token != NULL) {
strcpy(pd.ifname, dev_token);
if (ioctl(fd, FETCH_PCI_ADDRESS, &pd) == -1) {
TRACE_DBG("Could not find pci info on dpdk "
"device: %s. Is it a dpdk-attached "
"interface?\n", dev_token);
goto loop_over;
}
argv[*argc] = strdup("-w");
argv[*argc + 1] = calloc(PCI_LENGTH, 1);
if (argv[*argc] == NULL ||
argv[*argc + 1] == NULL) {
TRACE_ERROR("Memory allocation error!\n");
exit(EXIT_FAILURE);
}
sprintf(argv[*argc + 1], PCI_DOM":"PCI_BUS":"
PCI_DEVICE"."PCI_FUNC,
pd.pa.domain, pd.pa.bus, pd.pa.device,
pd.pa.function);
*argc += 2;
if (pd.numa_socket > numa_id) numa_id = pd.numa_socket;
loop_over:
dev_token = strtok_r(NULL, delim, &saveptr);
}
/* add the terminating "" sequence */
argv[*argc] = end;
close(fd);
free(dev_tokenizer);
return numa_id;
}
#endif /* !DISABLE_DPDK */
/*----------------------------------------------------------------------------*/
int
SetNetEnv(char *dev_name_list, char *port_stat_list)
{
int eidx = 0;
int i, j;
int set_all_inf = (strncmp(dev_name_list, ALL_STRING, sizeof(ALL_STRING))==0);
TRACE_CONFIG("Loading interface setting\n");
CONFIG.eths = (struct eth_table *)
calloc(MAX_DEVICES, sizeof(struct eth_table));
if (!CONFIG.eths) {
TRACE_ERROR("Can't allocate space for CONFIG.eths\n");
exit(EXIT_FAILURE);
}
if (current_iomodule_func == &ps_module_func) {
#ifndef DISABLE_PSIO
struct ifreq ifr;
/* calculate num_devices now! */
num_devices = ps_list_devices(devices);
if (num_devices == -1) {
perror("ps_list_devices");
exit(EXIT_FAILURE);
}
/* Create socket */
int sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP);
if (sock == -1) {
TRACE_ERROR("socket");
exit(EXIT_FAILURE);
}
/* To Do: Parse dev_name_list rather than use strstr */
for (i = 0; i < num_devices; i++) {
strcpy(ifr.ifr_name, devices[i].name);
/* getting interface information */
if (ioctl(sock, SIOCGIFFLAGS, &ifr) == 0) {
if (!set_all_inf && strstr(dev_name_list, ifr.ifr_name) == NULL)
continue;
/* Setting informations */
eidx = CONFIG.eths_num++;
strcpy(CONFIG.eths[eidx].dev_name, ifr.ifr_name);
CONFIG.eths[eidx].ifindex = devices[i].ifindex;
/* getting address */
if (ioctl(sock, SIOCGIFADDR, &ifr) == 0 ) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].ip_addr = *(uint32_t *)&sin;
}
if (ioctl(sock, SIOCGIFHWADDR, &ifr) == 0 ) {
for (j = 0; j < ETH_ALEN; j ++) {
CONFIG.eths[eidx].haddr[j] = ifr.ifr_addr.sa_data[j];
}
}
/* Net MASK */
if (ioctl(sock, SIOCGIFNETMASK, &ifr) == 0) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].netmask = *(uint32_t *)&sin;
}
/* add to attached devices */
for (j = 0; j < num_devices_attached; j++) {
if (devices_attached[j] == devices[i].ifindex) {
break;
}
}
devices_attached[num_devices_attached] = devices[i].ifindex;
num_devices_attached++;
} else {
perror("SIOCGIFFLAGS");
}
}
num_queues = GetNumQueues();
if (num_queues <= 0) {
TRACE_CONFIG("Failed to find NIC queues!\n");
close(sock);
return -1;
}
if (num_queues > num_cpus) {
TRACE_CONFIG("Too many NIC queues available.\n");
close(sock);
return -1;
}
close(sock);
#endif /* !PSIO_MODULE */
} else if (current_iomodule_func == &dpdk_module_func) {
#ifndef DISABLE_DPDK
int cpu = CONFIG.num_cores;
mpz_t _cpumask;
char cpumaskbuf[32] = "";
char mem_channels[8] = "";
char socket_mem_str[32] = "";
// int i;
int ret, socket_mem;
static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
/* STEP 1: first determine CPU mask */
mpz_init(_cpumask);
if (!mpz_cmp(_cpumask, CONFIG._cpumask)) {
/* get the cpu mask */
for (ret = 0; ret < cpu; ret++)
mpz_setbit(_cpumask, ret);
gmp_sprintf(cpumaskbuf, "%ZX", _cpumask);
} else
gmp_sprintf(cpumaskbuf, "%ZX", CONFIG._cpumask);
mpz_clear(_cpumask);
/* STEP 2: determine memory channels per socket */
/* get the mem channels per socket */
if (CONFIG.num_mem_ch == 0) {
TRACE_ERROR("DPDK module requires # of memory channels "
"per socket parameter!\n");
exit(EXIT_FAILURE);
}
sprintf(mem_channels, "%d", CONFIG.num_mem_ch);
/* STEP 3: determine socket memory */
/* get socket memory threshold (in MB) */
socket_mem =
RTE_ALIGN_CEIL((unsigned long)ceil((CONFIG.num_cores *
(CONFIG.rcvbuf_size +
CONFIG.sndbuf_size +
sizeof(struct tcp_stream) +
sizeof(struct tcp_recv_vars) +
sizeof(struct tcp_send_vars) +
sizeof(struct fragment_ctx)) *
CONFIG.max_concurrency)/RTE_SOCKET_MEM_SHIFT),
RTE_CACHE_LINE_SIZE);
/* initialize the rte env, what a waste of implementation effort! */
int argc = 6;//8;
char *argv[RTE_ARGC_MAX] = {"",
"-c",
cpumaskbuf,
"-n",
mem_channels,
#if 0
"--socket-mem",
socket_mem_str,
#endif
"--proc-type=auto"
};
ret = probe_all_rte_devices(argv, &argc, dev_name_list);
/* STEP 4: build up socket mem parameter */
sprintf(socket_mem_str, "%d", socket_mem);
#if 0
char *smsptr = socket_mem_str + strlen(socket_mem_str);
for (i = 1; i < ret + 1; i++) {
sprintf(smsptr, ",%d", socket_mem);
smsptr += strlen(smsptr);
}
TRACE_DBG("socket_mem: %s\n", socket_mem_str);
#endif
/*
* re-set getopt extern variable optind.
* this issue was a bitch to debug
* rte_eal_init() internally uses getopt() syscall
* mtcp applications that also use an `external' getopt
* will cause a violent crash if optind is not reset to zero
* prior to calling the func below...
* see man getopt(3) for more details
*/
optind = 0;
#ifdef DEBUG
/* print argv's */
for (i = 0; i < argc; i++)
TRACE_INFO("argv[%d]: %s\n", i, argv[i]);
#endif
/* initialize the dpdk eal env */
ret = rte_eal_init(argc, argv);
if (ret < 0) {
TRACE_ERROR("Invalid EAL args!\n");
exit(EXIT_FAILURE);
}
/* give me the count of 'detected' ethernet ports */
#if (RTE_VER_YEAR <= 18) && (RTE_VER_MONTH <= 02)
num_devices = rte_eth_dev_count();
#else
num_devices = rte_eth_dev_count_avail();
#endif
if (num_devices == 0) {
TRACE_ERROR("No Ethernet port!\n");
exit(EXIT_FAILURE);
}
/* get mac addr entries of 'detected' dpdk ports */
for (ret = 0; ret < num_devices; ret++)
rte_eth_macaddr_get(ret, &ports_eth_addr[ret]);
num_queues = MIN(CONFIG.num_cores, MAX_CPUS);
struct ifaddrs *ifap;
struct ifaddrs *iter_if;
char *seek;
if (getifaddrs(&ifap) != 0) {
perror("getifaddrs: ");
exit(EXIT_FAILURE);
}
iter_if = ifap;
do {
if (iter_if->ifa_addr->sa_family == AF_INET &&
!set_all_inf &&
(seek=strstr(dev_name_list, iter_if->ifa_name)) != NULL &&
/* check if the interface was not aliased */
*(seek + strlen(iter_if->ifa_name)) != ':') {
struct ifreq ifr;
/* Setting informations */
eidx = CONFIG.eths_num++;
strcpy(CONFIG.eths[eidx].dev_name, iter_if->ifa_name);
strcpy(ifr.ifr_name, iter_if->ifa_name);
/* Create socket */
int sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP);
if (sock == -1) {
perror("socket");
exit(EXIT_FAILURE);
}
/* getting address */
if (ioctl(sock, SIOCGIFADDR, &ifr) == 0 ) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].ip_addr = *(uint32_t *)&sin;
}
if (ioctl(sock, SIOCGIFHWADDR, &ifr) == 0 ) {
for (j = 0; j < ETH_ALEN; j ++) {
CONFIG.eths[eidx].haddr[j] = ifr.ifr_addr.sa_data[j];
}
}
/* Net MASK */
if (ioctl(sock, SIOCGIFNETMASK, &ifr) == 0) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].netmask = *(uint32_t *)&sin;
}
close(sock);
for (j = 0; j < num_devices; j++) {
if (!memcmp(&CONFIG.eths[eidx].haddr[0], &ports_eth_addr[j],
ETH_ALEN))
CONFIG.eths[eidx].ifindex = j;
}
/* add to attached devices */
for (j = 0; j < num_devices_attached; j++) {
if (devices_attached[j] == CONFIG.eths[eidx].ifindex) {
break;
}
}
devices_attached[num_devices_attached] = CONFIG.eths[eidx].ifindex;
num_devices_attached++;
fprintf(stderr, "Total number of attached devices: %d\n",
num_devices_attached);
fprintf(stderr, "Interface name: %s\n",
iter_if->ifa_name);
}
iter_if = iter_if->ifa_next;
} while (iter_if != NULL);
freeifaddrs(ifap);
#if 0
/*
* XXX: It seems that there is a bug in the RTE SDK.
* The dynamically allocated rte_argv params are left
* as dangling pointers. Freeing them causes program
* to crash.
*/
/* free up all resources */
for (; rte_argc >= 9; rte_argc--) {
if (rte_argv[rte_argc] != NULL) {
fprintf(stderr, "Cleaning up rte_argv[%d]: %s (%p)\n",
rte_argc, rte_argv[rte_argc], rte_argv[rte_argc]);
free(rte_argv[rte_argc]);
rte_argv[rte_argc] = NULL;
}
}
#endif
/* check if process is primary or secondary */
CONFIG.multi_process_is_master = (eal_proc_type_detect() == RTE_PROC_PRIMARY) ?
1 : 0;
#endif /* !DISABLE_DPDK */
} else if (current_iomodule_func == &netmap_module_func) {
#ifndef DISABLE_NETMAP
struct ifaddrs *ifap;
struct ifaddrs *iter_if;
char *seek;
num_queues = MIN(CONFIG.num_cores, MAX_CPUS);
if (getifaddrs(&ifap) != 0) {
perror("getifaddrs: ");
exit(EXIT_FAILURE);
}
iter_if = ifap;
do {
if (iter_if->ifa_addr->sa_family == AF_INET &&
!set_all_inf &&
(seek=strstr(dev_name_list, iter_if->ifa_name)) != NULL &&
/* check if the interface was not aliased */
*(seek + strlen(iter_if->ifa_name)) != ':') {
struct ifreq ifr;
/* Setting informations */
eidx = CONFIG.eths_num++;
strcpy(CONFIG.eths[eidx].dev_name, iter_if->ifa_name);
strcpy(ifr.ifr_name, iter_if->ifa_name);
/* Create socket */
int sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP);
if (sock == -1) {
perror("socket");
exit(EXIT_FAILURE);
}
/* getting address */
if (ioctl(sock, SIOCGIFADDR, &ifr) == 0 ) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].ip_addr = *(uint32_t *)&sin;
}
if (ioctl(sock, SIOCGIFHWADDR, &ifr) == 0 ) {
for (j = 0; j < ETH_ALEN; j ++) {
CONFIG.eths[eidx].haddr[j] = ifr.ifr_addr.sa_data[j];
}
}
/* Net MASK */
if (ioctl(sock, SIOCGIFNETMASK, &ifr) == 0) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].netmask = *(uint32_t *)&sin;
}
close(sock);
#if 0
for (j = 0; j < num_devices; j++) {
if (!memcmp(&CONFIG.eths[eidx].haddr[0], &ports_eth_addr[j],
ETH_ALEN))
CONFIG.eths[eidx].ifindex = ifr.ifr_ifindex;
#endif
CONFIG.eths[eidx].ifindex = eidx;
TRACE_INFO("Ifindex of interface %s is: %d\n",
ifr.ifr_name, CONFIG.eths[eidx].ifindex);
#if 0
}
#endif
/* add to attached devices */
for (j = 0; j < num_devices_attached; j++) {
if (devices_attached[j] == CONFIG.eths[eidx].ifindex) {
break;
}
}
devices_attached[num_devices_attached] = if_nametoindex(ifr.ifr_name);
num_devices_attached++;
fprintf(stderr, "Total number of attached devices: %d\n",
num_devices_attached);
fprintf(stderr, "Interface name: %s\n",
iter_if->ifa_name);
}
iter_if = iter_if->ifa_next;
} while (iter_if != NULL);
freeifaddrs(ifap);
#endif /* !DISABLE_NETMAP */
} else if (current_iomodule_func == &onvm_module_func) {
#ifdef ENABLE_ONVM
int cpu = CONFIG.num_cores;
mpz_t cpumask;
char cpumaskbuf[32];
char mem_channels[8];
char service[6];
char instance[6];
int ret;
mpz_init(cpumask);
/* get the cpu mask */
for (ret = 0; ret < cpu; ret++)
mpz_setbit(cpumask, ret);
gmp_sprintf(cpumaskbuf, "%ZX", cpumask);
mpz_clear(cpumask);
/* get the mem channels per socket */
if (CONFIG.num_mem_ch == 0) {
TRACE_ERROR("DPDK module requires # of memory channels "
"per socket parameter!\n");
exit(EXIT_FAILURE);
}
sprintf(mem_channels, "%d", CONFIG.num_mem_ch);
sprintf(service, "%d", CONFIG.onvm_serv);
sprintf(instance, "%d", CONFIG.onvm_inst);
/* initialize the rte env first, what a waste of implementation effort! */
char *argv[] = {"",
"-c",
cpumaskbuf,
"-n",
mem_channels,
"--proc-type=secondary",
"--",
"-r",
service,
instance,
""
};
const int argc = 10;
/*
* re-set getopt extern variable optind.
* this issue was a bitch to debug
* rte_eal_init() internally uses getopt() syscall
* mtcp applications that also use an `external' getopt
* will cause a violent crash if optind is not reset to zero
* prior to calling the func below...
* see man getopt(3) for more details
*/
optind = 0;
/* initialize the dpdk eal env */
ret = onvm_nflib_init(argc, argv, "mtcp_nf", &CONFIG.nf_info);
if (ret < 0) {
TRACE_ERROR("Invalid EAL args!\n");
exit(EXIT_FAILURE);
}
/* give me the count of 'detected' ethernet ports */
num_devices = ports->num_ports;
if (num_devices == 0) {
TRACE_ERROR("No Ethernet port!\n");
exit(EXIT_FAILURE);
}
num_queues = MIN(CONFIG.num_cores, MAX_CPUS);
struct ifaddrs *ifap;
struct ifaddrs *iter_if;
char *seek;
if (getifaddrs(&ifap) != 0) {
perror("getifaddrs: ");
exit(EXIT_FAILURE);
}
iter_if = ifap;
do {
if (iter_if->ifa_addr->sa_family == AF_INET &&
!set_all_inf &&
(seek=strstr(dev_name_list, iter_if->ifa_name)) != NULL &&
/* check if the interface was not aliased */
*(seek + strlen(iter_if->ifa_name)) != ':') {
struct ifreq ifr;
/* Setting informations */
eidx = CONFIG.eths_num++;
strcpy(CONFIG.eths[eidx].dev_name, iter_if->ifa_name);
strcpy(ifr.ifr_name, iter_if->ifa_name);
/* Create socket */
int sock = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP);
if (sock == -1) {
perror("socket");
}
/* getting address */
if (ioctl(sock, SIOCGIFADDR, &ifr) == 0 ) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].ip_addr = *(uint32_t *)&sin;
}
for (j = 0; j < ETH_ALEN; j ++) {
CONFIG.eths[eidx].haddr[j] = ports->mac[eidx].addr_bytes[j];
};
/* Net MASK */
if (ioctl(sock, SIOCGIFNETMASK, &ifr) == 0) {
struct in_addr sin = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr;
CONFIG.eths[eidx].netmask = *(uint32_t *)&sin;
}
close(sock);
CONFIG.eths[eidx].ifindex = ports->id[eidx];
devices_attached[num_devices_attached] = CONFIG.eths[eidx].ifindex;
num_devices_attached++;
fprintf(stderr, "Total number of attached devices: %d\n",
num_devices_attached);
fprintf(stderr, "Interface name: %s\n",
iter_if->ifa_name);
}
iter_if = iter_if->ifa_next;
} while (iter_if != NULL);
freeifaddrs(ifap);
#endif /* ENABLE_ONVM */
}
CONFIG.nif_to_eidx = (int*)calloc(MAX_DEVICES, sizeof(int));
if (!CONFIG.nif_to_eidx) {
exit(EXIT_FAILURE);
}
for (i = 0; i < MAX_DEVICES; ++i) {
CONFIG.nif_to_eidx[i] = -1;
}
for (i = 0; i < CONFIG.eths_num; ++i) {
j = CONFIG.eths[i].ifindex;
if (j >= MAX_DEVICES) {
TRACE_ERROR("ifindex of eths_%d exceed the limit: %d\n", i, j);
exit(EXIT_FAILURE);
}
/* the physic port index of the i-th port listed in the config file is j*/
CONFIG.nif_to_eidx[j] = i;
/* finally set the port stats option `on' */
if (strcmp(CONFIG.eths[i].dev_name, port_stat_list) == 0)
CONFIG.eths[i].stat_print = TRUE;
}
return 0;
}
/*----------------------------------------------------------------------------*/
int
FetchEndianType()
{
#ifndef DISABLE_DPDK
char *argv;
char **argp = &argv;
/* dpdk_module_func/onvm_module_func logic down below */
if (current_iomodule_func == &dpdk_module_func) {
(*current_iomodule_func).dev_ioctl(NULL, CONFIG.eths[0].ifindex, DRV_NAME, (void *)argp);
if (!strcmp(*argp, "net_i40e"))
return 1;
}
#endif
return 0;
}
/*----------------------------------------------------------------------------*/
int
CheckIOModuleAccessPermissions()
{
int fd;
/* check if netmap module can access I/O with sudo privileges */
if (current_iomodule_func == &netmap_module_func) {
fd = open(NETMAP_DEVICE_NAME, O_RDONLY);
if (fd != -1)
close(fd);
return fd;
}
/* sudo privileges are definitely needed otherwise */
if (geteuid())
return -1;
return 0;
}
/*----------------------------------------------------------------------------*/
Reference in New Issue View Git Blame Copy Permalink