nflog: only increment packets_nobufs when recv() returns an error

Errno should only be valid when recv() returns a `-1`, indicating an
error.

I believe the intended behavior here is for packets_nobufs to
be a counter that reports back how many times recv() returns
an ENOBUFS during a packet capture. Because of the existing logic
however, packets_nobufs begins incrementing for every recv() call
once the first ENOBUFS error is seen, since errno is not reset when
there are no errors returned from recv().

Before (counter deviates from strace):

    # tcpdump output
    38069 packets captured
    38069 packets received by filter
    38061 packets dropped by kernel

    # strace output
    % time     seconds  usecs/call     calls    errors syscall
    ------ ----------- ----------- --------- --------- ----------------
    26.47    0.282728           7     38067         3 recvfrom

After (counter matches strace):

    # tcpdump output
    38095 packets captured
    38095 packets received by filter
    7 packets dropped by kernel

    # strace output
    % time     seconds  usecs/call     calls    errors syscall
    ------ ----------- ----------- --------- --------- ----------------
    27.11    0.258596           6     38096         7 recvfrom

diff --git a/pcap-netfilter-linux.c b/pcap-netfilter-linux.c
index 33204a54e045bed86b69928f586cfb9c1ca66b10..dad5add490352f116dcb0ed5be7fcf9400f39ea6 100644 (file)
--- a/pcap-netfilter-linux.c
+++ b/pcap-netfilter-linux.c
@@ -123,7 +123,7 @@ netfilter_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_c
                                handle->break_loop = 0;
                                return PCAP_ERROR_BREAK;
                        }
-                       if (errno == ENOBUFS)
+                       if (len == -1 && errno == ENOBUFS)
                                handlep->packets_nobufs++;
                } while ((len == -1) && (errno == EINTR || errno == ENOBUFS));