Beginning with kernel 2.6.37, it has been changed the meaning of dropped packet count. Before, dropped packets was most likely due to an error. Now, the rx_dropped counter shows statistics for dropped frames because of:
If any frames meet those conditions, they are dropped before the protocol stack and the rx_dropped counter is incremented.
# netstat -i Kernel Interface table Iface MTU RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg eth0 1500 653090942 0 1481 0 84429784 0 0 0 BMRU lo 65536 1280117 0 0 0 1280117 0 0 0 LRU #
# ip -s link show 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN mode DEFAULT group default qlen 1000 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 RX: bytes packets errors dropped overrun mcast 119101765 1280127 0 0 0 0 TX: bytes packets errors dropped carrier collsns 119101765 1280127 0 0 0 0 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP mode DEFAULT group default qlen 1000 link/ether 02:01:76:1b:xx:xx brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped overrun mcast 61367859241 653095320 0 1481 0 0 TX: bytes packets errors dropped carrier collsns 269612612688 84429834 0 0 0 0
# ifconfig eth0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 inet xxx.xxx.xxx.xxx netmask 255.255.254.0 broadcast xxx.xxx.xxx.xxx inet6 fe80::1:xxxx:xxxx:xxxx prefixlen 64 scopeid 0x20<link> ether 02:01:76:xx:xx:xx txqueuelen 1000 (Ethernet) RX packets 653131438 bytes 61371164765 (61.3 GB) RX errors 0 dropped 1481 overruns 0 frame 0 TX packets 84435136 bytes 269624405533 (269.6 GB) TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 lo: flags=73<UP,LOOPBACK,RUNNING> mtu 65536 inet 127.0.0.1 netmask 255.0.0.0 inet6 ::1 prefixlen 128 scopeid 0x10<host> loop txqueuelen 1000 (Local Loopback) RX packets 1280223 bytes 119110409 (119.1 MB) RX errors 0 dropped 0 overruns 0 frame 0 TX packets 1280223 bytes 119110409 (119.1 MB) TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
# ethtool -S eth0 NIC statistics: rx_queue_0_packets: 653279140 rx_queue_0_bytes: 61384496586 rx_queue_0_drops: 0 rx_queue_0_xdp_packets: 0 rx_queue_0_xdp_tx: 0 rx_queue_0_xdp_redirects: 0 rx_queue_0_xdp_drops: 0 rx_queue_0_kicks: 10165 tx_queue_0_packets: 84452496 tx_queue_0_bytes: 269671658132 tx_queue_0_xdp_tx: 0 tx_queue_0_xdp_tx_drops: 0 tx_queue_0_kicks: 47507630
# ethtool -g eth0 Ring parameters for eth0: Pre-set maximums: RX: 4094 RX Mini: 0 RX Jumbo: 0 TX: 4094 Current hardware settings: RX: 256 RX Mini: 0 RX Jumbo: 0 TX: 256
ethtool -G eth0 rx 512
auto eth0 iface eth0 inet static address 192.168.1.2 netmask 255.255.255.0 gateway 192.168.1.1 dns-nameservers 192.168.1.1 post-up ethtool -G eth0 rx 512
RHEL 7 drop statistics cover more reasons for discards of unwanted packets than previous version of RHEL. For details, see this article.
Packets can be dropped at other levels in the network stack. These will not be counted as discards/drops/etc shown by ethtool -S ethX.
ネットワークスタックのその他のレベルで、パケットが破棄される場合があります。そのようなパケットは、ethtool -S ethX で示される discard、drop などには含まれません。
ip -s link and ifconfig report the rx_drop* counter increasing but no drops are seen with ethtool -S ethX
netstat -i reports RX-ERR and RX-DRP
This is not a bug. Previous kernels were not accounting for these dropped packets (unknown protocol for example)
Seeing high rx_drop_counters in RHEL 7 from ifconfig|ip -s link is not a definitive conclusion of a network problem. To gain more traction of a network problem “ethtool -S” can be used for Network Card Statistics or “netstat -s” for Network Application statistics.
ip -s linkを使うな。
RHEL 7 differentiates packet loss from older o/s (RHEL 6 and RHEL 5), as rx drop counter can increment from the following reasons:
Note: While tcpdump is running, the above frame does not increase rx drop count.
Care should be taken to confirm that frames are not being legitimately dropped. A quick way to test this (WARNING: this test does not work for bonding interfaces) is to force the NIC into promiscuous mode:
host:~# ifconfig <interface> promisc
And then watching the rx_dropped counter. If it stops incrementing while the NIC is in promiscuous mode; then it is more than likely showing drops because of the reasons listed earlier.
If additional frames continue to be shown as dropped, further investigation might be warranted. However, since IP communication and/or the applications that use it have methods in place for recovering missing packets, a small number of dropped packets is not expected to have any significant impact, and is not worth investigating. Additionally, some applications, such as video streaming, don't consider a few lost packets to be something that requires any recovery attempts, and they proceed without any delay.
The “other drops” you are referring to are probably non-IPv4 packets being discarded by the Ethernet driver, does the incrementing of RX-DRP stop for as long as you are actively running a tcpdump? See the following from my book:
RX-DRP Culprit 1: Unknown or Undesired Protocol Type In every Ethernet frame is a header field called “EtherType”. This field specifies the OSI Layer 3 protocol that the Ethernet frame is carrying. A very common value for this header field is 0x0800, which indicates that the frame is carrying an Internet Protocol version 4 (IPv4) packet. Look at this excerpt from Stage 6 of “A Millisecond in the life of a frame”: Stage 6: At a later time the CPU begins SoftIRQ processing and looks in the ring buffer. If a descriptor is present, the CPU retrieves the frame from the associated receive socket buffer, clears the descriptor referencing the frame in the ring buffer, and sends the frame to all “registered receivers” which will be the SecureXL acceleration driver. If a tcpdump capture is currently running, libpcap will also be a “registered receiver” in that case and get a copy of the frame as well. The SoftIRQ processing continues until all ring buffers are completely emptied, or various packet count or time limits have been reached. During hardware interrupt processing, the NIC driver will examine the EtherType field and verify there is a “registered receiver” present for the protocol specified in the frame header. If there is not, the frame is discarded and RX-DRP is incremented. Example: an Ethernet frame arrives with an EtherType of 0x86dd indicating the presence of IPv6 in the Ethernet frame. If IPv6 has not been enabled on the firewall (it is off by default), the frame will be discarded by the NIC driver and RX-DRP incremented. What other protocols are known to cause this effect in the real world? Let’s take a look at a brief sampling of other possible rogue EtherTypes you may see, that is by no means complete: - Appletalk (0x809b) - IPX (0x8137 or 0x8138) - Ethernet Flow Control (0x8808) if NIC flow control is disabled - Jumbo Frames (0x8870) if the firewall is not configured to process jumbo frames The dropping of these protocols for which there is no “registered receiver” does cause a very small amount of overhead on the firewall during hardware interrupt processing, but unless the number of frames discarded in this way exceeds 0.1% of all inbound packets, you probably shouldn’t worry too much about it. An easy way to confirm that the lack of a registered receiver is the cause of RX-DRPs is to perform the following test: 1. In a SSH or terminal window, run watch -d netstat -ni and confirm the constant incrementing of RX-DRP on (interface). 2. In a second SSH session, run tcpdump -ni (interface) host 220.127.116.11 Does the near constant incrementing of RX-DRP on that interface suddenly stop as long as the tcpdump is still running, and resume when the tcpdump is stopped? If so, the lack of a registered receiver is indeed the cause of the RX-DRPs. The specified filter expression (host 18.104.22.168 in our example) does not actually matter, since libpcap will register to receive all protocols on behalf of the running tcpdump, and then filter the packets based on the provided tcpdump expression. So as long as the tcpdump is running, there is essentially a registered received for everything.