net/tap: fix mbuf double free when writev fails
authorYunjian Wang <wangyunjian@huawei.com>
Thu, 16 Apr 2020 03:04:07 +0000 (11:04 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 21 Apr 2020 11:57:08 +0000 (13:57 +0200)
When the tap_write_mbufs() function return with break, mbuf was freed
without increasing num_packets, which could cause applications to free
the mbuf again. And the pmd_tx_burst() function should returns the
number of original packets it actually sent excluding tso mbufs.

Fixes: 9396ad334672 ("net/tap: fix reported number of Tx packets")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
drivers/net/tap/rte_eth_tap.c

index 4b2168e..6d54c25 100644 (file)
@@ -520,7 +520,7 @@ tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
        }
 }
 
-static inline void
+static inline int
 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
                        struct rte_mbuf **pmbufs,
                        uint16_t *num_packets, unsigned long *num_tx_bytes)
@@ -587,7 +587,7 @@ tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
                        seg_len = rte_pktmbuf_data_len(mbuf);
                        l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
                        if (seg_len < l234_hlen)
-                               break;
+                               return -1;
 
                        /* To change checksums, work on a * copy of l2, l3
                         * headers + l4 pseudo header
@@ -633,10 +633,12 @@ tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
                /* copy the tx frame data */
                n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
                if (n <= 0)
-                       break;
+                       return -1;
+
                (*num_packets)++;
                (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
        }
+       return 0;
 }
 
 /* Callback to handle sending packets from the tap interface
@@ -662,8 +664,8 @@ pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                uint16_t num_mbufs = 0;
                uint16_t tso_segsz = 0;
                int ret;
+               int num_tso_mbufs;
                uint16_t hdrs_len;
-               int j;
                uint64_t tso;
 
                tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
@@ -683,43 +685,51 @@ pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                                break;
                        }
                        gso_ctx->gso_size = tso_segsz;
-                       ret = rte_gso_segment(mbuf_in, /* packet to segment */
+                       /* 'mbuf_in' packet to segment */
+                       num_tso_mbufs = rte_gso_segment(mbuf_in,
                                gso_ctx, /* gso control block */
                                (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
                                RTE_DIM(gso_mbufs)); /* max tso mbufs */
 
                        /* ret contains the number of new created mbufs */
-                       if (ret < 0)
+                       if (num_tso_mbufs < 0)
                                break;
 
                        mbuf = gso_mbufs;
-                       num_mbufs = ret;
+                       num_mbufs = num_tso_mbufs;
                } else {
                        /* stats.errs will be incremented */
                        if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
                                break;
 
                        /* ret 0 indicates no new mbufs were created */
-                       ret = 0;
+                       num_tso_mbufs = 0;
                        mbuf = &mbuf_in;
                        num_mbufs = 1;
                }
 
-               tap_write_mbufs(txq, num_mbufs, mbuf,
+               ret = tap_write_mbufs(txq, num_mbufs, mbuf,
                                &num_packets, &num_tx_bytes);
+               if (ret == -1) {
+                       txq->stats.errs++;
+                       /* free tso mbufs */
+                       if (num_tso_mbufs > 0)
+                               rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
+                       break;
+               }
                num_tx++;
                /* free original mbuf */
                rte_pktmbuf_free(mbuf_in);
                /* free tso mbufs */
-               for (j = 0; j < ret; j++)
-                       rte_pktmbuf_free(mbuf[j]);
+               if (num_tso_mbufs > 0)
+                       rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
        }
 
        txq->stats.opackets += num_packets;
        txq->stats.errs += nb_pkts - num_tx;
        txq->stats.obytes += num_tx_bytes;
 
-       return num_packets;
+       return num_tx;
 }
 
 static const char *