Improving DPDK capture interface and docs v7

doc/userguide/install.rst

@@ -57,6 +57,9 @@ Common configure options
 
     Enables GeoIP support for detection.
 
+.. option:: --enable-dpdk
+
+    Enables `DPDK <https://www.dpdk.org/>` packet capture method.
 
 Dependencies
 ^^^^^^^^^^^^

src/runmode-dpdk.c

@@ -930,7 +930,7 @@ static int DeviceValidateMTU(const DPDKIfaceConfig *iconf, const struct rte_eth_
         SCReturnInt(-ERANGE);
     }
 
-#if RTE_VER_YEAR < 21 || RTE_VER_YEAR == 21 && RTE_VER_MONTH < 11
+#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0)
     // check if jumbo frames are set and are available
     if (iconf->mtu > RTE_ETHER_MAX_LEN &&
             !(dev_info->rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME)) {
@@ -944,7 +944,7 @@ static int DeviceValidateMTU(const DPDKIfaceConfig *iconf, const struct rte_eth_
 
 static void DeviceSetMTU(struct rte_eth_conf *port_conf, uint16_t mtu)
 {
-#if RTE_VER_YEAR > 21 || RTE_VER_YEAR == 21 && RTE_VER_MONTH == 11
+#if RTE_VERSION >= RTE_VERSION_NUM(21, 11, 0, 0)
     port_conf->rxmode.mtu = mtu;
 #else
     port_conf->rxmode.max_rx_pkt_len = mtu;
@@ -954,6 +954,24 @@ static void DeviceSetMTU(struct rte_eth_conf *port_conf, uint16_t mtu)
 #endif
 }
 
+/**
+ * \param port_id - queried port
+ * \param socket_id - socket ID of the queried port
+ * \return positive number on success, negative on failure (errno)
+ */
+static int32_t DeviceSetSocketID(uint16_t port_id, int32_t *socket_id)
+{
+    rte_errno = 0;
+    int retval = rte_eth_dev_socket_id(port_id);
+    *socket_id = retval;
+
+#if RTE_VERSION >= RTE_VERSION_NUM(22, 11, 0, 0) // DPDK API changed since 22.11
+    retval = -rte_errno;
+#endif
+
+    return retval;
+}
+
 static void DeviceInitPortConf(const DPDKIfaceConfig *iconf,
         const struct rte_eth_dev_info *dev_info, struct rte_eth_conf *port_conf)
 {
@@ -1002,7 +1020,8 @@ static void DeviceInitPortConf(const DPDKIfaceConfig *iconf,
 
     if (iconf->checksum_mode == CHECKSUM_VALIDATION_DISABLE) {
         SCLogConfig("%s: checksum validation disabled", iconf->iface);
-    } else if (dev_info->rx_offload_capa & RTE_ETH_RX_OFFLOAD_CHECKSUM) {
+    } else if ((dev_info->rx_offload_capa & RTE_ETH_RX_OFFLOAD_CHECKSUM) ==
+               RTE_ETH_RX_OFFLOAD_CHECKSUM) { // multibit comparison to make sure all bits are set
         if (iconf->checksum_mode == CHECKSUM_VALIDATION_ENABLE &&
                 iconf->flags & DPDK_RX_CHECKSUM_OFFLOAD) {
             SCLogConfig("%s: IP, TCP and UDP checksum validation offloaded", iconf->iface);
@@ -1150,7 +1169,14 @@ static int DeviceConfigureIPS(DPDKIfaceConfig *iconf)
             SCReturnInt(retval);
         }
 
-        if (rte_eth_dev_socket_id(iconf->port_id) != rte_eth_dev_socket_id(iconf->out_port_id)) {
+        int32_t out_port_socket_id;
+        retval = DeviceSetSocketID(iconf->port_id, &out_port_socket_id);
+        if (retval < 0) {
+            SCLogError("%s: invalid socket id (err=%d)", iconf->out_iface, retval);
+            SCReturnInt(retval);
+        }
+
+        if (iconf->socket_id != out_port_socket_id) {
             SCLogWarning("%s: out iface %s is not on the same NUMA node", iconf->iface,
                     iconf->out_iface);
         }
@@ -1190,12 +1216,11 @@ static int DeviceConfigure(DPDKIfaceConfig *iconf)
         SCReturnInt(retval);
     }
 
-    retval = rte_eth_dev_socket_id(iconf->port_id);
+    retval = DeviceSetSocketID(iconf->port_id, &iconf->socket_id);
     if (retval < 0) {
         SCLogError("%s: invalid socket id (err=%d)", iconf->iface, retval);
         SCReturnInt(retval);
     }
-    iconf->socket_id = retval;
 
     retval = rte_eth_dev_info_get(iconf->port_id, &dev_info);
     if (retval != 0) {
@@ -1337,6 +1362,13 @@ static void *ParseDpdkConfigAndConfigureDevice(const char *iface)
     // This counter is increased by worker threads that individually pick queue IDs.
     SC_ATOMIC_RESET(iconf->queue_id);
     SC_ATOMIC_RESET(iconf->inconsitent_numa_cnt);
+
+    // initialize LiveDev DPDK values
+    LiveDevice *ldev_instance = LiveGetDevice(iface);
+    if (ldev_instance == NULL) {
+        FatalError("Device %s is not registered as a live device", iface);
+    }
+    ldev_instance->dpdk_vars.pkt_mp = iconf->pkt_mempool;
     return iconf;
 }
 

src/source-dpdk.c

@@ -121,6 +121,7 @@ typedef struct DPDKThreadVars_ {
     uint64_t dropped;
     uint16_t port_id;
     uint16_t queue_id;
+    int32_t port_socket_id;
     struct rte_mempool *pkt_mempool;
     struct rte_mbuf *received_mbufs[BURST_SIZE];
 } DPDKThreadVars;
@@ -267,17 +268,17 @@ void TmModuleDecodeDPDKRegister(void)
 
 static inline void DPDKDumpCounters(DPDKThreadVars *ptv)
 {
-    struct rte_eth_stats eth_stats;
-    int retval = rte_eth_stats_get(ptv->port_id, &eth_stats);
-    if (unlikely(retval != 0)) {
-        SCLogError("Failed to get stats for port id %d: %s", ptv->port_id, rte_strerror(-retval));
-        return;
-    }
-
     /* Some NICs (e.g. Intel) do not support queue statistics and the drops can be fetched only on
      * the port level. Therefore setting it to the first worker to have at least continuous update
      * on the dropped packets. */
     if (ptv->queue_id == 0) {
+        struct rte_eth_stats eth_stats;
+        int retval = rte_eth_stats_get(ptv->port_id, &eth_stats);
+        if (unlikely(retval != 0)) {
+            SCLogError("%s: failed to get stats: %s", ptv->livedev->dev, rte_strerror(-retval));
+            return;
+        }
+
         StatsSetUI64(ptv->tv, ptv->capture_dpdk_packets,
                 ptv->pkts + eth_stats.imissed + eth_stats.ierrors + eth_stats.rx_nombuf);
         SC_ATOMIC_SET(ptv->livedev->pkts,
@@ -344,6 +345,7 @@ static TmEcode ReceiveDPDKLoop(ThreadVars *tv, void *data, void *slot)
     uint16_t nb_rx;
     time_t last_dump = 0;
     time_t current_time;
+    bool segmented_mbufs_warned = 0;
 
     DPDKThreadVars *ptv = (DPDKThreadVars *)data;
     TmSlot *s = (TmSlot *)slot;
@@ -408,6 +410,23 @@ static TmEcode ReceiveDPDKLoop(ThreadVars *tv, void *data, void *slot)
                 }
             }
 
+            if (!rte_pktmbuf_is_contiguous(p->dpdk_v.mbuf) && !segmented_mbufs_warned) {
+                char warn_s[] = "Segmented mbufs detected! Redmine Ticket #6012 "
+                                "Check your configuration or report the issue";
+                enum rte_proc_type_t eal_t = rte_eal_process_type();
+                if (eal_t == RTE_PROC_SECONDARY) {
+                    SCLogWarning("%s. To avoid segmented mbufs, "
+                                 "try to increase mbuf size in your primary application",
+                            warn_s);
+                } else if (eal_t == RTE_PROC_PRIMARY) {
+                    SCLogWarning("%s. To avoid segmented mbufs, "
+                                 "try to increase MTU in your suricata.yaml",
+                            warn_s);
+                }
+
+                segmented_mbufs_warned = 1;
+            }
+
             PacketSetData(p, rte_pktmbuf_mtod(p->dpdk_v.mbuf, uint8_t *),
                     rte_pktmbuf_pkt_len(p->dpdk_v.mbuf));
             if (TmThreadsSlotProcessPkt(ptv->tv, ptv->slot, p) != TM_ECODE_OK) {
@@ -473,15 +492,16 @@ static TmEcode ReceiveDPDKThreadInit(ThreadVars *tv, const void *initdata, void
     ptv->threads = dpdk_config->threads;
     ptv->port_id = dpdk_config->port_id;
     ptv->out_port_id = dpdk_config->out_port_id;
+    ptv->port_socket_id = dpdk_config->socket_id;
     // pass the pointer to the mempool and then forget about it. Mempool is freed in thread deinit.
     ptv->pkt_mempool = dpdk_config->pkt_mempool;
     dpdk_config->pkt_mempool = NULL;
 
     thread_numa = GetNumaNode();
-    if (thread_numa >= 0 && thread_numa != rte_eth_dev_socket_id(ptv->port_id)) {
+    if (thread_numa >= 0 && thread_numa != ptv->port_socket_id) {
         SC_ATOMIC_ADD(dpdk_config->inconsitent_numa_cnt, 1);
         SCLogPerf("%s: NIC is on NUMA %d, thread on NUMA %d", dpdk_config->iface,
-                rte_eth_dev_socket_id(ptv->port_id), thread_numa);
+                ptv->port_socket_id, thread_numa);
     }
 
     uint16_t queue_id = SC_ATOMIC_ADD(dpdk_config->queue_id, 1);
@@ -644,10 +664,7 @@ static TmEcode ReceiveDPDKThreadDeinit(ThreadVars *tv, void *data)
         rte_eth_dev_stop(ptv->out_port_id);
     }
 
-    if (ptv->queue_id == 0 && ptv->pkt_mempool != NULL) {
-        rte_mempool_free(ptv->pkt_mempool);
-        ptv->pkt_mempool = NULL;
-    }
+    ptv->pkt_mempool = NULL; // MP is released when device is closed
 
     SCFree(ptv);
     SCReturnInt(TM_ECODE_OK);

src/source-dpdk.h

@@ -46,7 +46,7 @@ typedef struct DPDKIfaceConfig_ {
 #ifdef HAVE_DPDK
     char iface[RTE_ETH_NAME_MAX_LEN];
     uint16_t port_id;
-    uint16_t socket_id;
+    int32_t socket_id;
     /* number of threads - zero means all available */
     int threads;
     /* IPS mode */

src/util-device.h

@@ -18,6 +18,10 @@
 #ifndef __UTIL_DEVICE_H__
 #define __UTIL_DEVICE_H__
 
+#ifdef HAVE_DPDK
+#include <rte_mempool.h>
+#endif /* HAVE_DPDK */
+
 #include "queue.h"
 
 #define OFFLOAD_FLAG_SG     (1<<0)
@@ -35,6 +39,12 @@ int LiveGetOffload(void);
 
 #define MAX_DEVNAME 10
 
+#ifdef HAVE_DPDK
+typedef struct {
+    struct rte_mempool *pkt_mp;
+} DPDKDeviceResources;
+#endif /* HAVE_DPDK */
+
 /** storage for live device names */
 typedef struct LiveDevice_ {
     char *dev;  /**< the device (e.g. "eth0") */
@@ -51,6 +61,10 @@ typedef struct LiveDevice_ {
 
     uint32_t tenant_id;     /**< tenant id in multi-tenancy */
     uint32_t offload_orig;  /**< original offload settings to restore @exit */
+#ifdef HAVE_DPDK
+    // DPDK resources that needs to be cleaned after workers are stopped and devices closed
+    DPDKDeviceResources dpdk_vars;
+#endif
 } LiveDevice;
 
 typedef struct LiveDeviceName_ {

src/util-dpdk-i40e.c

@@ -38,7 +38,7 @@
 
 #define I40E_RSS_HKEY_LEN 52
 
-#if RTE_VER_YEAR <= 19
+#if RTE_VERSION < RTE_VERSION_NUM(20, 0, 0, 0)
 static int i40eDeviceEnableSymHash(
         int port_id, const char *port_name, uint32_t ftype, enum rte_eth_hash_function function)
 {
@@ -349,7 +349,7 @@ static int i40eDeviceSetRSSWithFlows(int port_id, const char *port_name, int nb_
     return 0;
 }
 
-#endif /* RTE_VER_YEAR < 19 */
+#endif /* RTE_VERSION < RTE_VERSION_NUM(20,0,0,0) */
 
 int i40eDeviceSetRSS(int port_id, int nb_rx_queues)
 {
@@ -364,25 +364,25 @@ int i40eDeviceSetRSS(int port_id, int nb_rx_queues)
         return retval;
     }
 
-#if RTE_VER_YEAR <= 19
-    i40eDeviceSetRSSWithFilter(port_id, port_name);
-#else
+#if RTE_VERSION >= RTE_VERSION_NUM(20, 0, 0, 0)
     i40eDeviceSetRSSWithFlows(port_id, port_name, nb_rx_queues);
+#else
+    i40eDeviceSetRSSWithFilter(port_id, port_name);
 #endif
     return 0;
 }
 
 void i40eDeviceSetRSSHashFunction(uint64_t *rss_hf)
 {
-    if (RTE_VER_YEAR <= 19)
-        *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_TCP |
-                  RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_NONFRAG_IPV4_SCTP |
-                  RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_FRAG_IPV6 |
-                  RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_NONFRAG_IPV6_UDP |
-                  RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_NONFRAG_IPV6_OTHER | RTE_ETH_RSS_SCTP;
-    else
-        *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_FRAG_IPV6 |
-                  RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
+#if RTE_VERSION >= RTE_VERSION_NUM(20, 0, 0, 0)
+    *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_FRAG_IPV6 |
+              RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
+#else
+    *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_NONFRAG_IPV4_UDP |
+              RTE_ETH_RSS_NONFRAG_IPV4_SCTP | RTE_ETH_RSS_NONFRAG_IPV4_OTHER |
+              RTE_ETH_RSS_FRAG_IPV6 | RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_NONFRAG_IPV6_UDP |
+              RTE_ETH_RSS_NONFRAG_IPV6_SCTP | RTE_ETH_RSS_NONFRAG_IPV6_OTHER | RTE_ETH_RSS_SCTP;
+#endif
 }
 
 #endif /* HAVE_DPDK */

src/util-dpdk-ice.c

@@ -37,12 +37,13 @@
 
 void iceDeviceSetRSSHashFunction(uint64_t *rss_hf)
 {
-    if (RTE_VER_YEAR <= 19)
-        *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_FRAG_IPV6 |
-                  RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
-    else
-        *rss_hf = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER |
-                  RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 | RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
+#if RTE_VERSION < RTE_VERSION_NUM(20, 0, 0, 0)
+    *rss_hf = RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER | RTE_ETH_RSS_FRAG_IPV6 |
+              RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
+#else
+    *rss_hf = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 | RTE_ETH_RSS_NONFRAG_IPV4_OTHER |
+              RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 | RTE_ETH_RSS_NONFRAG_IPV6_OTHER;
+#endif
 }
 
 #endif /* HAVE_DPDK */

src/util-dpdk.c

@@ -51,6 +51,9 @@ void DPDKCloseDevice(LiveDevice *ldev)
 
         SCLogInfo("%s: closing device", ldev->dev);
         rte_eth_dev_close(port_id);
+
+        SCLogInfo("%s: releasing packet mempool", ldev->dev);
+        rte_mempool_free(ldev->dpdk_vars.pkt_mp);
     }
 #endif
 }

src/util-dpdk.h

@@ -34,13 +34,13 @@
 #include <rte_mempool.h>
 #include <rte_mbuf.h>
 #include <rte_flow.h>
+#include <rte_version.h>
 
-#if RTE_VER_YEAR < 22
+#if RTE_VERSION < RTE_VERSION_NUM(22, 0, 0, 0)
 #define RTE_ETH_MQ_RX_RSS ETH_MQ_RX_RSS
-
 #endif
 
-#if RTE_VER_YEAR < 21 || RTE_VER_YEAR == 21 && RTE_VER_MONTH < 11
+#if RTE_VERSION < RTE_VERSION_NUM(21, 11, 0, 0)
 #define RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE DEV_TX_OFFLOAD_MBUF_FAST_FREE
 
 #define RTE_ETH_RX_OFFLOAD_CHECKSUM DEV_RX_OFFLOAD_CHECKSUM