net/tap: fix flow and port commands
authorOphir Munk <ophirmu@mellanox.com>
Sat, 16 Sep 2017 22:32:38 +0000 (22:32 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 6 Oct 2017 00:49:48 +0000 (02:49 +0200)
This commit fixes two bugs related to tap devices. The first bug occurs
when executing in testpmd the following flow rule assuming tap device has
4 rx and tx pair queues
"flow create 0 ingress pattern eth / end actions queue index 5 / end"
This command will report on success and will print ""Flow rule #0 created"
although it should have failed as queue index number 5 does not exist

The second bug occurs when executing in testpmd "port start all" following
a port configuration. Assuming 1 pair of rx and tx queues an error is
reported: "Fail to start port 0"

Before this commit a fixed max number (16) of rx and tx queue pairs were
created on startup where the file descriptors (fds) of rx and tx pairs were
identical.  As a result in the first bug queue index 5 existed because the
tap device was created with 16 rx and tx queue pairs regardless of the
configured number of queues. In the second bug when tap device was started
tx fd was closed before opening it and executing ioctl() on it. However
closing the sole fd of the device caused ioctl to fail with "No such
device".

This commit creates the configured number of rx and tx queue pairs (up to
max 16) and assigns a unique fd to each queue. It was written to solve the
first bug and was found as the right fix for the second bug as well.

Fixes: 02f96a0a82d1 ("net/tap: add TUN/TAP device PMD")
Fixes: bf7b7f437b49 ("net/tap: create netdevice during probing")
Fixes: de96fe68ae95 ("net/tap: add basic flow API patterns and actions")
Cc: stable@dpdk.org
Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Pascal Mazon <pascal.mazon@6wind.com>
drivers/net/tap/rte_eth_tap.c
drivers/net/tap/rte_eth_tap.h
drivers/net/tap/tap_flow.c

index 9acea83..d926d4b 100644 (file)
@@ -603,8 +603,31 @@ tap_dev_stop(struct rte_eth_dev *dev)
 }
 
 static int
-tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
+tap_dev_configure(struct rte_eth_dev *dev)
 {
+       if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
+               RTE_LOG(ERR, PMD,
+                       "%s: number of rx queues %d exceeds max num of queues %d\n",
+                       dev->device->name,
+                       dev->data->nb_rx_queues,
+                       RTE_PMD_TAP_MAX_QUEUES);
+               return -1;
+       }
+       if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
+               RTE_LOG(ERR, PMD,
+                       "%s: number of tx queues %d exceeds max num of queues %d\n",
+                       dev->device->name,
+                       dev->data->nb_tx_queues,
+                       RTE_PMD_TAP_MAX_QUEUES);
+               return -1;
+       }
+
+       RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
+            dev->device->name, (void *)dev, dev->data->nb_tx_queues);
+
+       RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
+            dev->device->name, (void *)dev, dev->data->nb_rx_queues);
+
        return 0;
 }
 
@@ -650,8 +673,8 @@ tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        dev_info->if_index = internals->if_index;
        dev_info->max_mac_addrs = 1;
        dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
-       dev_info->max_rx_queues = internals->nb_queues;
-       dev_info->max_tx_queues = internals->nb_queues;
+       dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
+       dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
        dev_info->min_rx_bufsize = 0;
        dev_info->pci_dev = NULL;
        dev_info->speed_capa = tap_dev_speed_capa();
@@ -673,9 +696,9 @@ tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
        unsigned long rx_nombuf = 0, ierrors = 0;
        const struct pmd_internals *pmd = dev->data->dev_private;
 
-       imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
-               pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
-
+       /* rx queue statistics */
+       imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
+               dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
        for (i = 0; i < imax; i++) {
                tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
                tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
@@ -683,7 +706,13 @@ tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
                rx_bytes_total += tap_stats->q_ibytes[i];
                rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
                ierrors += pmd->rxq[i].stats.ierrors;
+       }
 
+       /* tx queue statistics */
+       imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
+               dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
+
+       for (i = 0; i < imax; i++) {
                tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
                tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
                tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
@@ -707,7 +736,7 @@ tap_stats_reset(struct rte_eth_dev *dev)
        int i;
        struct pmd_internals *pmd = dev->data->dev_private;
 
-       for (i = 0; i < pmd->nb_queues; i++) {
+       for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
                pmd->rxq[i].stats.ipackets = 0;
                pmd->rxq[i].stats.ibytes = 0;
                pmd->rxq[i].stats.ierrors = 0;
@@ -729,11 +758,15 @@ tap_dev_close(struct rte_eth_dev *dev)
        tap_flow_flush(dev, NULL);
        tap_flow_implicit_flush(internals, NULL);
 
-       for (i = 0; i < internals->nb_queues; i++) {
-               if (internals->rxq[i].fd != -1)
+       for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
+               if (internals->rxq[i].fd != -1) {
                        close(internals->rxq[i].fd);
-               internals->rxq[i].fd = -1;
-               internals->txq[i].fd = -1;
+                       internals->rxq[i].fd = -1;
+               }
+               if (internals->txq[i].fd != -1) {
+                       close(internals->txq[i].fd);
+                       internals->txq[i].fd = -1;
+               }
        }
 
        if (internals->remote_if_index) {
@@ -887,30 +920,57 @@ tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 static int
 tap_setup_queue(struct rte_eth_dev *dev,
                struct pmd_internals *internals,
-               uint16_t qid)
+               uint16_t qid,
+               int is_rx)
 {
+       int *fd;
+       int *other_fd;
+       const char *dir;
        struct pmd_internals *pmd = dev->data->dev_private;
        struct rx_queue *rx = &internals->rxq[qid];
        struct tx_queue *tx = &internals->txq[qid];
-       int fd = rx->fd == -1 ? tx->fd : rx->fd;
 
-       if (fd == -1) {
-               RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
-                       pmd->name, qid);
-               fd = tun_alloc(pmd);
-               if (fd < 0) {
+       if (is_rx) {
+               fd = &rx->fd;
+               other_fd = &tx->fd;
+               dir = "rx";
+       } else {
+               fd = &tx->fd;
+               other_fd = &rx->fd;
+               dir = "tx";
+       }
+       if (*fd != -1) {
+               /* fd for this queue already exists */
+               RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
+                       pmd->name, *fd, dir, qid);
+       } else if (*other_fd != -1) {
+               /* Only other_fd exists. dup it */
+               *fd = dup(*other_fd);
+               if (*fd < 0) {
+                       *fd = -1;
+                       RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
+                               pmd->name);
+                       return -1;
+               }
+               RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
+                       pmd->name, *other_fd, dir, qid, *fd);
+       } else {
+               /* Both RX and TX fds do not exist (equal -1). Create fd */
+               *fd = tun_alloc(pmd);
+               if (*fd < 0) {
+                       *fd = -1; /* restore original value */
                        RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
                                pmd->name);
                        return -1;
                }
+               RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
+                       pmd->name, dir, qid, *fd);
        }
 
-       rx->fd = fd;
-       tx->fd = fd;
        tx->mtu = &dev->data->mtu;
        rx->rxmode = &dev->data->dev_conf.rxmode;
 
-       return fd;
+       return *fd;
 }
 
 static int
@@ -932,10 +992,10 @@ tap_rx_queue_setup(struct rte_eth_dev *dev,
        int fd;
        int i;
 
-       if ((rx_queue_id >= internals->nb_queues) || !mp) {
+       if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
                RTE_LOG(WARNING, PMD,
-                       "nb_queues %d too small or mempool NULL\n",
-                       internals->nb_queues);
+                       "nb_rx_queues %d too small or mempool NULL\n",
+                       dev->data->nb_rx_queues);
                return -1;
        }
 
@@ -954,7 +1014,7 @@ tap_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->iovecs = iovecs;
 
        dev->data->rx_queues[rx_queue_id] = rxq;
-       fd = tap_setup_queue(dev, internals, rx_queue_id);
+       fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
        if (fd == -1) {
                ret = fd;
                goto error;
@@ -1002,11 +1062,11 @@ tap_tx_queue_setup(struct rte_eth_dev *dev,
        struct pmd_internals *internals = dev->data->dev_private;
        int ret;
 
-       if (tx_queue_id >= internals->nb_queues)
+       if (tx_queue_id >= dev->data->nb_tx_queues)
                return -1;
 
        dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
-       ret = tap_setup_queue(dev, internals, tx_queue_id);
+       ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
        if (ret == -1)
                return -1;
 
@@ -1166,7 +1226,6 @@ static const struct eth_dev_ops ops = {
        .filter_ctrl            = tap_dev_filter_ctrl,
 };
 
-
 static int
 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
                   char *remote_iface, int fixed_mac_type)
@@ -1193,8 +1252,8 @@ eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
        }
 
        pmd = dev->data->dev_private;
+       pmd->dev = dev;
        snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
-       pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 
        pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
        if (pmd->ioctl_sock == -1) {
@@ -1212,8 +1271,9 @@ eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
 
        data->dev_link = pmd_link;
        data->mac_addrs = &pmd->eth_addr;
-       data->nb_rx_queues = pmd->nb_queues;
-       data->nb_tx_queues = pmd->nb_queues;
+       /* Set the number of RX and TX queues */
+       data->nb_rx_queues = 0;
+       data->nb_tx_queues = 0;
 
        dev->data = data;
        dev->dev_ops = &ops;
@@ -1241,7 +1301,11 @@ eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
        }
 
        /* Immediately create the netdevice (this will create the 1st queue). */
-       if (tap_setup_queue(dev, pmd, 0) == -1)
+       /* rx queue */
+       if (tap_setup_queue(dev, pmd, 0, 1) == -1)
+               goto error_exit;
+       /* tx queue */
+       if (tap_setup_queue(dev, pmd, 0, 0) == -1)
                goto error_exit;
 
        ifr.ifr_mtu = dev->data->mtu;
@@ -1515,9 +1579,16 @@ rte_pmd_tap_remove(struct rte_vdev_device *dev)
                tap_flow_implicit_flush(internals, NULL);
                nl_final(internals->nlsk_fd);
        }
-       for (i = 0; i < internals->nb_queues; i++)
-               if (internals->rxq[i].fd != -1)
+       for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
+               if (internals->rxq[i].fd != -1) {
                        close(internals->rxq[i].fd);
+                       internals->rxq[i].fd = -1;
+               }
+               if (internals->txq[i].fd != -1) {
+                       close(internals->txq[i].fd);
+                       internals->txq[i].fd = -1;
+               }
+       }
 
        close(internals->ioctl_sock);
        rte_free(eth_dev->data->dev_private);
index 928a045..829f32f 100644 (file)
@@ -80,9 +80,9 @@ struct tx_queue {
 };
 
 struct pmd_internals {
+       struct rte_eth_dev *dev;          /* Ethernet device. */
        char remote_iface[RTE_ETH_NAME_MAX_LEN]; /* Remote netdevice name */
        char name[RTE_ETH_NAME_MAX_LEN];  /* Internal Tap device name */
-       uint16_t nb_queues;               /* Number of queues supported */
        struct ether_addr eth_addr;       /* Mac address of the device port */
        struct ifreq remote_initial_flags;   /* Remote netdevice flags on init */
        int remote_if_index;              /* remote netdevice IF_INDEX */
index 41f7345..aa33960 100644 (file)
@@ -1092,7 +1092,8 @@ priv_flow_process(struct pmd_internals *pmd,
                        if (action)
                                goto exit_action_not_supported;
                        action = 1;
-                       if (!queue || (queue->index >= pmd->nb_queues))
+                       if (!queue ||
+                           (queue->index > pmd->dev->data->nb_rx_queues - 1))
                                goto exit_action_not_supported;
                        if (flow)
                                err = add_action_skbedit(flow, queue->index);