5 This document explains the meaning of SNMP counters.
9 All layer 4 packets and ICMP packets will change these counters, but
10 these counters won't be changed by layer 2 packets (such as STP) or
15 Defined in `RFC1213 ipInReceives`_
17 .. _RFC1213 ipInReceives: https://tools.ietf.org/html/rfc1213#page-26
19 The number of packets received by the IP layer. It gets increasing at the
20 beginning of ip_rcv function, always be updated together with
21 IpExtInOctets. It will be increased even if the packet is dropped
22 later (e.g. due to the IP header is invalid or the checksum is wrong
23 and so on). It indicates the number of aggregated segments after
28 Defined in `RFC1213 ipInDelivers`_
30 .. _RFC1213 ipInDelivers: https://tools.ietf.org/html/rfc1213#page-28
32 The number of packets delivers to the upper layer protocols. E.g. TCP, UDP,
33 ICMP and so on. If no one listens on a raw socket, only kernel
34 supported protocols will be delivered, if someone listens on the raw
35 socket, all valid IP packets will be delivered.
39 Defined in `RFC1213 ipOutRequests`_
41 .. _RFC1213 ipOutRequests: https://tools.ietf.org/html/rfc1213#page-28
43 The number of packets sent via IP layer, for both single cast and
44 multicast packets, and would always be updated together with
47 * IpExtInOctets and IpExtOutOctets
49 They are Linux kernel extensions, no RFC definitions. Please note,
50 RFC1213 indeed defines ifInOctets and ifOutOctets, but they
51 are different things. The ifInOctets and ifOutOctets include the MAC
52 layer header size but IpExtInOctets and IpExtOutOctets don't, they
53 only include the IP layer header and the IP layer data.
55 * IpExtInNoECTPkts, IpExtInECT1Pkts, IpExtInECT0Pkts, IpExtInCEPkts
57 They indicate the number of four kinds of ECN IP packets, please refer
58 `Explicit Congestion Notification`_ for more details.
60 .. _Explicit Congestion Notification: https://tools.ietf.org/html/rfc3168#page-6
62 These 4 counters calculate how many packets received per ECN
63 status. They count the real frame number regardless the LRO/GRO. So
64 for the same packet, you might find that IpInReceives count 1, but
65 IpExtInNoECTPkts counts 2 or more.
69 Defined in `RFC1213 ipInHdrErrors`_. It indicates the packet is
70 dropped due to the IP header error. It might happen in both IP input
73 .. _RFC1213 ipInHdrErrors: https://tools.ietf.org/html/rfc1213#page-27
77 Defined in `RFC1213 ipInAddrErrors`_. It will be increased in two
78 scenarios: (1) The IP address is invalid. (2) The destination IP
79 address is not a local address and IP forwarding is not enabled
81 .. _RFC1213 ipInAddrErrors: https://tools.ietf.org/html/rfc1213#page-27
85 This counter means the packet is dropped when the IP stack receives a
86 packet and can't find a route for it from the route table. It might
87 happen when IP forwarding is enabled and the destination IP address is
88 not a local address and there is no route for the destination IP
93 Defined in `RFC1213 ipInUnknownProtos`_. It will be increased if the
94 layer 4 protocol is unsupported by kernel. If an application is using
95 raw socket, kernel will always deliver the packet to the raw socket
96 and this counter won't be increased.
98 .. _RFC1213 ipInUnknownProtos: https://tools.ietf.org/html/rfc1213#page-27
100 * IpExtInTruncatedPkts
102 For IPv4 packet, it means the actual data size is smaller than the
103 "Total Length" field in the IPv4 header.
107 Defined in `RFC1213 ipInDiscards`_. It indicates the packet is dropped
108 in the IP receiving path and due to kernel internal reasons (e.g. no
111 .. _RFC1213 ipInDiscards: https://tools.ietf.org/html/rfc1213#page-28
115 Defined in `RFC1213 ipOutDiscards`_. It indicates the packet is
116 dropped in the IP sending path and due to kernel internal reasons.
118 .. _RFC1213 ipOutDiscards: https://tools.ietf.org/html/rfc1213#page-28
122 Defined in `RFC1213 ipOutNoRoutes`_. It indicates the packet is
123 dropped in the IP sending path and no route is found for it.
125 .. _RFC1213 ipOutNoRoutes: https://tools.ietf.org/html/rfc1213#page-29
129 * IcmpInMsgs and IcmpOutMsgs
131 Defined by `RFC1213 icmpInMsgs`_ and `RFC1213 icmpOutMsgs`_
133 .. _RFC1213 icmpInMsgs: https://tools.ietf.org/html/rfc1213#page-41
134 .. _RFC1213 icmpOutMsgs: https://tools.ietf.org/html/rfc1213#page-43
136 As mentioned in the RFC1213, these two counters include errors, they
137 would be increased even if the ICMP packet has an invalid type. The
138 ICMP output path will check the header of a raw socket, so the
139 IcmpOutMsgs would still be updated if the IP header is constructed by
144 | These counters include most of common ICMP types, they are:
145 | IcmpInDestUnreachs: `RFC1213 icmpInDestUnreachs`_
146 | IcmpInTimeExcds: `RFC1213 icmpInTimeExcds`_
147 | IcmpInParmProbs: `RFC1213 icmpInParmProbs`_
148 | IcmpInSrcQuenchs: `RFC1213 icmpInSrcQuenchs`_
149 | IcmpInRedirects: `RFC1213 icmpInRedirects`_
150 | IcmpInEchos: `RFC1213 icmpInEchos`_
151 | IcmpInEchoReps: `RFC1213 icmpInEchoReps`_
152 | IcmpInTimestamps: `RFC1213 icmpInTimestamps`_
153 | IcmpInTimestampReps: `RFC1213 icmpInTimestampReps`_
154 | IcmpInAddrMasks: `RFC1213 icmpInAddrMasks`_
155 | IcmpInAddrMaskReps: `RFC1213 icmpInAddrMaskReps`_
156 | IcmpOutDestUnreachs: `RFC1213 icmpOutDestUnreachs`_
157 | IcmpOutTimeExcds: `RFC1213 icmpOutTimeExcds`_
158 | IcmpOutParmProbs: `RFC1213 icmpOutParmProbs`_
159 | IcmpOutSrcQuenchs: `RFC1213 icmpOutSrcQuenchs`_
160 | IcmpOutRedirects: `RFC1213 icmpOutRedirects`_
161 | IcmpOutEchos: `RFC1213 icmpOutEchos`_
162 | IcmpOutEchoReps: `RFC1213 icmpOutEchoReps`_
163 | IcmpOutTimestamps: `RFC1213 icmpOutTimestamps`_
164 | IcmpOutTimestampReps: `RFC1213 icmpOutTimestampReps`_
165 | IcmpOutAddrMasks: `RFC1213 icmpOutAddrMasks`_
166 | IcmpOutAddrMaskReps: `RFC1213 icmpOutAddrMaskReps`_
168 .. _RFC1213 icmpInDestUnreachs: https://tools.ietf.org/html/rfc1213#page-41
169 .. _RFC1213 icmpInTimeExcds: https://tools.ietf.org/html/rfc1213#page-41
170 .. _RFC1213 icmpInParmProbs: https://tools.ietf.org/html/rfc1213#page-42
171 .. _RFC1213 icmpInSrcQuenchs: https://tools.ietf.org/html/rfc1213#page-42
172 .. _RFC1213 icmpInRedirects: https://tools.ietf.org/html/rfc1213#page-42
173 .. _RFC1213 icmpInEchos: https://tools.ietf.org/html/rfc1213#page-42
174 .. _RFC1213 icmpInEchoReps: https://tools.ietf.org/html/rfc1213#page-42
175 .. _RFC1213 icmpInTimestamps: https://tools.ietf.org/html/rfc1213#page-42
176 .. _RFC1213 icmpInTimestampReps: https://tools.ietf.org/html/rfc1213#page-43
177 .. _RFC1213 icmpInAddrMasks: https://tools.ietf.org/html/rfc1213#page-43
178 .. _RFC1213 icmpInAddrMaskReps: https://tools.ietf.org/html/rfc1213#page-43
180 .. _RFC1213 icmpOutDestUnreachs: https://tools.ietf.org/html/rfc1213#page-44
181 .. _RFC1213 icmpOutTimeExcds: https://tools.ietf.org/html/rfc1213#page-44
182 .. _RFC1213 icmpOutParmProbs: https://tools.ietf.org/html/rfc1213#page-44
183 .. _RFC1213 icmpOutSrcQuenchs: https://tools.ietf.org/html/rfc1213#page-44
184 .. _RFC1213 icmpOutRedirects: https://tools.ietf.org/html/rfc1213#page-44
185 .. _RFC1213 icmpOutEchos: https://tools.ietf.org/html/rfc1213#page-45
186 .. _RFC1213 icmpOutEchoReps: https://tools.ietf.org/html/rfc1213#page-45
187 .. _RFC1213 icmpOutTimestamps: https://tools.ietf.org/html/rfc1213#page-45
188 .. _RFC1213 icmpOutTimestampReps: https://tools.ietf.org/html/rfc1213#page-45
189 .. _RFC1213 icmpOutAddrMasks: https://tools.ietf.org/html/rfc1213#page-45
190 .. _RFC1213 icmpOutAddrMaskReps: https://tools.ietf.org/html/rfc1213#page-46
192 Every ICMP type has two counters: 'In' and 'Out'. E.g., for the ICMP
193 Echo packet, they are IcmpInEchos and IcmpOutEchos. Their meanings are
194 straightforward. The 'In' counter means kernel receives such a packet
195 and the 'Out' counter means kernel sends such a packet.
199 They are IcmpMsgInType[N] and IcmpMsgOutType[N], the [N] indicates the
200 ICMP type number. These counters track all kinds of ICMP packets. The
201 ICMP type number definition could be found in the `ICMP parameters`_
204 .. _ICMP parameters: https://www.iana.org/assignments/icmp-parameters/icmp-parameters.xhtml
206 For example, if the Linux kernel sends an ICMP Echo packet, the
207 IcmpMsgOutType8 would increase 1. And if kernel gets an ICMP Echo Reply
208 packet, IcmpMsgInType0 would increase 1.
212 This counter indicates the checksum of the ICMP packet is
213 wrong. Kernel verifies the checksum after updating the IcmpInMsgs and
214 before updating IcmpMsgInType[N]. If a packet has bad checksum, the
215 IcmpInMsgs would be updated but none of IcmpMsgInType[N] would be updated.
217 * IcmpInErrors and IcmpOutErrors
219 Defined by `RFC1213 icmpInErrors`_ and `RFC1213 icmpOutErrors`_
221 .. _RFC1213 icmpInErrors: https://tools.ietf.org/html/rfc1213#page-41
222 .. _RFC1213 icmpOutErrors: https://tools.ietf.org/html/rfc1213#page-43
224 When an error occurs in the ICMP packet handler path, these two
225 counters would be updated. The receiving packet path use IcmpInErrors
226 and the sending packet path use IcmpOutErrors. When IcmpInCsumErrors
227 is increased, IcmpInErrors would always be increased too.
229 relationship of the ICMP counters
230 ---------------------------------
231 The sum of IcmpMsgOutType[N] is always equal to IcmpOutMsgs, as they
232 are updated at the same time. The sum of IcmpMsgInType[N] plus
233 IcmpInErrors should be equal or larger than IcmpInMsgs. When kernel
234 receives an ICMP packet, kernel follows below logic:
236 1. increase IcmpInMsgs
237 2. if has any error, update IcmpInErrors and finish the process
238 3. update IcmpMsgOutType[N]
239 4. handle the packet depending on the type, if has any error, update
240 IcmpInErrors and finish the process
242 So if all errors occur in step (2), IcmpInMsgs should be equal to the
243 sum of IcmpMsgOutType[N] plus IcmpInErrors. If all errors occur in
244 step (4), IcmpInMsgs should be equal to the sum of
245 IcmpMsgOutType[N]. If the errors occur in both step (2) and step (4),
246 IcmpInMsgs should be less than the sum of IcmpMsgOutType[N] plus
253 Defined in `RFC1213 tcpInSegs`_
255 .. _RFC1213 tcpInSegs: https://tools.ietf.org/html/rfc1213#page-48
257 The number of packets received by the TCP layer. As mentioned in
258 RFC1213, it includes the packets received in error, such as checksum
259 error, invalid TCP header and so on. Only one error won't be included:
260 if the layer 2 destination address is not the NIC's layer 2
261 address. It might happen if the packet is a multicast or broadcast
262 packet, or the NIC is in promiscuous mode. In these situations, the
263 packets would be delivered to the TCP layer, but the TCP layer will discard
264 these packets before increasing TcpInSegs. The TcpInSegs counter
265 isn't aware of GRO. So if two packets are merged by GRO, the TcpInSegs
266 counter would only increase 1.
270 Defined in `RFC1213 tcpOutSegs`_
272 .. _RFC1213 tcpOutSegs: https://tools.ietf.org/html/rfc1213#page-48
274 The number of packets sent by the TCP layer. As mentioned in RFC1213,
275 it excludes the retransmitted packets. But it includes the SYN, ACK
276 and RST packets. Doesn't like TcpInSegs, the TcpOutSegs is aware of
277 GSO, so if a packet would be split to 2 by GSO, TcpOutSegs will
282 Defined in `RFC1213 tcpActiveOpens`_
284 .. _RFC1213 tcpActiveOpens: https://tools.ietf.org/html/rfc1213#page-47
286 It means the TCP layer sends a SYN, and come into the SYN-SENT
287 state. Every time TcpActiveOpens increases 1, TcpOutSegs should always
292 Defined in `RFC1213 tcpPassiveOpens`_
294 .. _RFC1213 tcpPassiveOpens: https://tools.ietf.org/html/rfc1213#page-47
296 It means the TCP layer receives a SYN, replies a SYN+ACK, come into
299 * TcpExtTCPRcvCoalesce
301 When packets are received by the TCP layer and are not be read by the
302 application, the TCP layer will try to merge them. This counter
303 indicate how many packets are merged in such situation. If GRO is
304 enabled, lots of packets would be merged by GRO, these packets
305 wouldn't be counted to TcpExtTCPRcvCoalesce.
307 * TcpExtTCPAutoCorking
309 When sending packets, the TCP layer will try to merge small packets to
310 a bigger one. This counter increase 1 for every packet merged in such
311 situation. Please refer to the LWN article for more details:
312 https://lwn.net/Articles/576263/
314 * TcpExtTCPOrigDataSent
316 This counter is explained by `kernel commit f19c29e3e391`_, I pasted the
319 TCPOrigDataSent: number of outgoing packets with original data (excluding
320 retransmission but including data-in-SYN). This counter is different from
321 TcpOutSegs because TcpOutSegs also tracks pure ACKs. TCPOrigDataSent is
322 more useful to track the TCP retransmission rate.
326 This counter is explained by `kernel commit f19c29e3e391`_, I pasted the
329 TCPSynRetrans: number of SYN and SYN/ACK retransmits to break down
330 retransmissions into SYN, fast-retransmits, timeout retransmits, etc.
332 * TCPFastOpenActiveFail
334 This counter is explained by `kernel commit f19c29e3e391`_, I pasted the
337 TCPFastOpenActiveFail: Fast Open attempts (SYN/data) failed because
338 the remote does not accept it or the attempts timed out.
340 .. _kernel commit f19c29e3e391: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f19c29e3e391a66a273e9afebaf01917245148cd
342 * TcpExtListenOverflows and TcpExtListenDrops
344 When kernel receives a SYN from a client, and if the TCP accept queue
345 is full, kernel will drop the SYN and add 1 to TcpExtListenOverflows.
346 At the same time kernel will also add 1 to TcpExtListenDrops. When a
347 TCP socket is in LISTEN state, and kernel need to drop a packet,
348 kernel would always add 1 to TcpExtListenDrops. So increase
349 TcpExtListenOverflows would let TcpExtListenDrops increasing at the
350 same time, but TcpExtListenDrops would also increase without
351 TcpExtListenOverflows increasing, e.g. a memory allocation fail would
352 also let TcpExtListenDrops increase.
354 Note: The above explanation is based on kernel 4.10 or above version, on
355 an old kernel, the TCP stack has different behavior when TCP accept
356 queue is full. On the old kernel, TCP stack won't drop the SYN, it
357 would complete the 3-way handshake. As the accept queue is full, TCP
358 stack will keep the socket in the TCP half-open queue. As it is in the
359 half open queue, TCP stack will send SYN+ACK on an exponential backoff
360 timer, after client replies ACK, TCP stack checks whether the accept
361 queue is still full, if it is not full, moves the socket to the accept
362 queue, if it is full, keeps the socket in the half-open queue, at next
363 time client replies ACK, this socket will get another chance to move
371 Defined in `RFC1213 tcpEstabResets`_.
373 .. _RFC1213 tcpEstabResets: https://tools.ietf.org/html/rfc1213#page-48
377 Defined in `RFC1213 tcpAttemptFails`_.
379 .. _RFC1213 tcpAttemptFails: https://tools.ietf.org/html/rfc1213#page-48
383 Defined in `RFC1213 tcpOutRsts`_. The RFC says this counter indicates
384 the 'segments sent containing the RST flag', but in linux kernel, this
385 couner indicates the segments kerenl tried to send. The sending
386 process might be failed due to some errors (e.g. memory alloc failed).
388 .. _RFC1213 tcpOutRsts: https://tools.ietf.org/html/rfc1213#page-52
390 * TcpExtTCPSpuriousRtxHostQueues
392 When the TCP stack wants to retransmit a packet, and finds that packet
393 is not lost in the network, but the packet is not sent yet, the TCP
394 stack would give up the retransmission and update this counter. It
395 might happen if a packet stays too long time in a qdisc or driver
400 The socket receives a RST packet in Establish or CloseWait state.
404 This counter indicates many keepalive packets were sent. The keepalive
405 won't be enabled by default. A userspace program could enable it by
406 setting the SO_KEEPALIVE socket option.
408 * TcpExtTCPSpuriousRTOs
410 The spurious retransmission timeout detected by the `F-RTO`_
413 .. _F-RTO: https://tools.ietf.org/html/rfc5682
417 When kernel receives a TCP packet, it has two paths to handler the
418 packet, one is fast path, another is slow path. The comment in kernel
419 code provides a good explanation of them, I pasted them below::
421 It is split into a fast path and a slow path. The fast path is
424 - A zero window was announced from us
425 - zero window probing
426 is only handled properly on the slow path.
427 - Out of order segments arrived.
428 - Urgent data is expected.
429 - There is no buffer space left
430 - Unexpected TCP flags/window values/header lengths are received
431 (detected by checking the TCP header against pred_flags)
432 - Data is sent in both directions. The fast path only supports pure senders
433 or pure receivers (this means either the sequence number or the ack
434 value must stay constant)
435 - Unexpected TCP option.
437 Kernel will try to use fast path unless any of the above conditions
438 are satisfied. If the packets are out of order, kernel will handle
439 them in slow path, which means the performance might be not very
440 good. Kernel would also come into slow path if the "Delayed ack" is
441 used, because when using "Delayed ack", the data is sent in both
442 directions. When the TCP window scale option is not used, kernel will
443 try to enable fast path immediately when the connection comes into the
444 established state, but if the TCP window scale option is used, kernel
445 will disable the fast path at first, and try to enable it after kernel
448 * TcpExtTCPPureAcks and TcpExtTCPHPAcks
450 If a packet set ACK flag and has no data, it is a pure ACK packet, if
451 kernel handles it in the fast path, TcpExtTCPHPAcks will increase 1,
452 if kernel handles it in the slow path, TcpExtTCPPureAcks will
457 If a TCP packet has data (which means it is not a pure ACK packet),
458 and this packet is handled in the fast path, TcpExtTCPHPHits will
464 * TcpExtTCPAbortOnData
466 It means TCP layer has data in flight, but need to close the
467 connection. So TCP layer sends a RST to the other side, indicate the
468 connection is not closed very graceful. An easy way to increase this
469 counter is using the SO_LINGER option. Please refer to the SO_LINGER
470 section of the `socket man page`_:
472 .. _socket man page: http://man7.org/linux/man-pages/man7/socket.7.html
474 By default, when an application closes a connection, the close function
475 will return immediately and kernel will try to send the in-flight data
476 async. If you use the SO_LINGER option, set l_onoff to 1, and l_linger
477 to a positive number, the close function won't return immediately, but
478 wait for the in-flight data are acked by the other side, the max wait
479 time is l_linger seconds. If set l_onoff to 1 and set l_linger to 0,
480 when the application closes a connection, kernel will send a RST
481 immediately and increase the TcpExtTCPAbortOnData counter.
483 * TcpExtTCPAbortOnClose
485 This counter means the application has unread data in the TCP layer when
486 the application wants to close the TCP connection. In such a situation,
487 kernel will send a RST to the other side of the TCP connection.
489 * TcpExtTCPAbortOnMemory
491 When an application closes a TCP connection, kernel still need to track
492 the connection, let it complete the TCP disconnect process. E.g. an
493 app calls the close method of a socket, kernel sends fin to the other
494 side of the connection, then the app has no relationship with the
495 socket any more, but kernel need to keep the socket, this socket
496 becomes an orphan socket, kernel waits for the reply of the other side,
497 and would come to the TIME_WAIT state finally. When kernel has no
498 enough memory to keep the orphan socket, kernel would send an RST to
499 the other side, and delete the socket, in such situation, kernel will
500 increase 1 to the TcpExtTCPAbortOnMemory. Two conditions would trigger
501 TcpExtTCPAbortOnMemory:
503 1. the memory used by the TCP protocol is higher than the third value of
504 the tcp_mem. Please refer the tcp_mem section in the `TCP man page`_:
506 .. _TCP man page: http://man7.org/linux/man-pages/man7/tcp.7.html
508 2. the orphan socket count is higher than net.ipv4.tcp_max_orphans
511 * TcpExtTCPAbortOnTimeout
513 This counter will increase when any of the TCP timers expire. In such
514 situation, kernel won't send RST, just give up the connection.
516 * TcpExtTCPAbortOnLinger
518 When a TCP connection comes into FIN_WAIT_2 state, instead of waiting
519 for the fin packet from the other side, kernel could send a RST and
520 delete the socket immediately. This is not the default behavior of
521 Linux kernel TCP stack. By configuring the TCP_LINGER2 socket option,
522 you could let kernel follow this behavior.
524 * TcpExtTCPAbortFailed
526 The kernel TCP layer will send RST if the `RFC2525 2.17 section`_ is
527 satisfied. If an internal error occurs during this process,
528 TcpExtTCPAbortFailed will be increased.
530 .. _RFC2525 2.17 section: https://tools.ietf.org/html/rfc2525#page-50
532 TCP Hybrid Slow Start
533 =====================
534 The Hybrid Slow Start algorithm is an enhancement of the traditional
535 TCP congestion window Slow Start algorithm. It uses two pieces of
536 information to detect whether the max bandwidth of the TCP path is
537 approached. The two pieces of information are ACK train length and
538 increase in packet delay. For detail information, please refer the
539 `Hybrid Slow Start paper`_. Either ACK train length or packet delay
540 hits a specific threshold, the congestion control algorithm will come
541 into the Congestion Avoidance state. Until v4.20, two congestion
542 control algorithms are using Hybrid Slow Start, they are cubic (the
543 default congestion control algorithm) and cdg. Four snmp counters
544 relate with the Hybrid Slow Start algorithm.
546 .. _Hybrid Slow Start paper: https://pdfs.semanticscholar.org/25e9/ef3f03315782c7f1cbcd31b587857adae7d1.pdf
548 * TcpExtTCPHystartTrainDetect
550 How many times the ACK train length threshold is detected
552 * TcpExtTCPHystartTrainCwnd
554 The sum of CWND detected by ACK train length. Dividing this value by
555 TcpExtTCPHystartTrainDetect is the average CWND which detected by the
558 * TcpExtTCPHystartDelayDetect
560 How many times the packet delay threshold is detected.
562 * TcpExtTCPHystartDelayCwnd
564 The sum of CWND detected by packet delay. Dividing this value by
565 TcpExtTCPHystartDelayDetect is the average CWND which detected by the
568 TCP retransmission and congestion control
569 =========================================
570 The TCP protocol has two retransmission mechanisms: SACK and fast
571 recovery. They are exclusive with each other. When SACK is enabled,
572 the kernel TCP stack would use SACK, or kernel would use fast
573 recovery. The SACK is a TCP option, which is defined in `RFC2018`_,
574 the fast recovery is defined in `RFC6582`_, which is also called
577 The TCP congestion control is a big and complex topic. To understand
578 the related snmp counter, we need to know the states of the congestion
579 control state machine. There are 5 states: Open, Disorder, CWR,
580 Recovery and Loss. For details about these states, please refer page 5
581 and page 6 of this document:
582 https://pdfs.semanticscholar.org/0e9c/968d09ab2e53e24c4dca5b2d67c7f7140f8e.pdf
584 .. _RFC2018: https://tools.ietf.org/html/rfc2018
585 .. _RFC6582: https://tools.ietf.org/html/rfc6582
587 * TcpExtTCPRenoRecovery and TcpExtTCPSackRecovery
589 When the congestion control comes into Recovery state, if sack is
590 used, TcpExtTCPSackRecovery increases 1, if sack is not used,
591 TcpExtTCPRenoRecovery increases 1. These two counters mean the TCP
592 stack begins to retransmit the lost packets.
594 * TcpExtTCPSACKReneging
596 A packet was acknowledged by SACK, but the receiver has dropped this
597 packet, so the sender needs to retransmit this packet. In this
598 situation, the sender adds 1 to TcpExtTCPSACKReneging. A receiver
599 could drop a packet which has been acknowledged by SACK, although it is
600 unusual, it is allowed by the TCP protocol. The sender doesn't really
601 know what happened on the receiver side. The sender just waits until
602 the RTO expires for this packet, then the sender assumes this packet
603 has been dropped by the receiver.
605 * TcpExtTCPRenoReorder
607 The reorder packet is detected by fast recovery. It would only be used
608 if SACK is disabled. The fast recovery algorithm detects recorder by
609 the duplicate ACK number. E.g., if retransmission is triggered, and
610 the original retransmitted packet is not lost, it is just out of
611 order, the receiver would acknowledge multiple times, one for the
612 retransmitted packet, another for the arriving of the original out of
613 order packet. Thus the sender would find more ACks than its
614 expectation, and the sender knows out of order occurs.
618 The reorder packet is detected when a hole is filled. E.g., assume the
619 sender sends packet 1,2,3,4,5, and the receiving order is
620 1,2,4,5,3. When the sender receives the ACK of packet 3 (which will
621 fill the hole), two conditions will let TcpExtTCPTSReorder increase
622 1: (1) if the packet 3 is not re-retransmitted yet. (2) if the packet
623 3 is retransmitted but the timestamp of the packet 3's ACK is earlier
624 than the retransmission timestamp.
626 * TcpExtTCPSACKReorder
628 The reorder packet detected by SACK. The SACK has two methods to
629 detect reorder: (1) DSACK is received by the sender. It means the
630 sender sends the same packet more than one times. And the only reason
631 is the sender believes an out of order packet is lost so it sends the
632 packet again. (2) Assume packet 1,2,3,4,5 are sent by the sender, and
633 the sender has received SACKs for packet 2 and 5, now the sender
634 receives SACK for packet 4 and the sender doesn't retransmit the
635 packet yet, the sender would know packet 4 is out of order. The TCP
636 stack of kernel will increase TcpExtTCPSACKReorder for both of the
639 * TcpExtTCPSlowStartRetrans
641 The TCP stack wants to retransmit a packet and the congestion control
644 * TcpExtTCPFastRetrans
646 The TCP stack wants to retransmit a packet and the congestion control
649 * TcpExtTCPLostRetransmit
651 A SACK points out that a retransmission packet is lost again.
653 * TcpExtTCPRetransFail
655 The TCP stack tries to deliver a retransmission packet to lower layers
656 but the lower layers return an error.
658 * TcpExtTCPSynRetrans
660 The TCP stack retransmits a SYN packet.
664 The DSACK is defined in `RFC2883`_. The receiver uses DSACK to report
665 duplicate packets to the sender. There are two kinds of
666 duplications: (1) a packet which has been acknowledged is
667 duplicate. (2) an out of order packet is duplicate. The TCP stack
668 counts these two kinds of duplications on both receiver side and
671 .. _RFC2883 : https://tools.ietf.org/html/rfc2883
673 * TcpExtTCPDSACKOldSent
675 The TCP stack receives a duplicate packet which has been acked, so it
676 sends a DSACK to the sender.
678 * TcpExtTCPDSACKOfoSent
680 The TCP stack receives an out of order duplicate packet, so it sends a
685 The TCP stack receives a DSACK, which indicates an acknowledged
686 duplicate packet is received.
688 * TcpExtTCPDSACKOfoRecv
690 The TCP stack receives a DSACK, which indicate an out of order
691 duplicate packet is received.
693 invalid SACK and DSACK
694 ======================
695 When a SACK (or DSACK) block is invalid, a corresponding counter would
696 be updated. The validation method is base on the start/end sequence
697 number of the SACK block. For more details, please refer the comment
698 of the function tcp_is_sackblock_valid in the kernel source code. A
699 SACK option could have up to 4 blocks, they are checked
700 individually. E.g., if 3 blocks of a SACk is invalid, the
701 corresponding counter would be updated 3 times. The comment of the
702 `Add counters for discarded SACK blocks`_ patch has additional
705 .. _Add counters for discarded SACK blocks: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=18f02545a9a16c9a89778b91a162ad16d510bb32
707 * TcpExtTCPSACKDiscard
709 This counter indicates how many SACK blocks are invalid. If the invalid
710 SACK block is caused by ACK recording, the TCP stack will only ignore
711 it and won't update this counter.
713 * TcpExtTCPDSACKIgnoredOld and TcpExtTCPDSACKIgnoredNoUndo
715 When a DSACK block is invalid, one of these two counters would be
716 updated. Which counter will be updated depends on the undo_marker flag
717 of the TCP socket. If the undo_marker is not set, the TCP stack isn't
718 likely to re-transmit any packets, and we still receive an invalid
719 DSACK block, the reason might be that the packet is duplicated in the
720 middle of the network. In such scenario, TcpExtTCPDSACKIgnoredNoUndo
721 will be updated. If the undo_marker is set, TcpExtTCPDSACKIgnoredOld
722 will be updated. As implied in its name, it might be an old packet.
726 The linux networking stack stores data in sk_buff struct (skb for
727 short). If a SACK block acrosses multiple skb, the TCP stack will try
728 to re-arrange data in these skb. E.g. if a SACK block acknowledges seq
729 10 to 15, skb1 has seq 10 to 13, skb2 has seq 14 to 20. The seq 14 and
730 15 in skb2 would be moved to skb1. This operation is 'shift'. If a
731 SACK block acknowledges seq 10 to 20, skb1 has seq 10 to 13, skb2 has
732 seq 14 to 20. All data in skb2 will be moved to skb1, and skb2 will be
733 discard, this operation is 'merge'.
735 * TcpExtTCPSackShifted
739 * TcpExtTCPSackMerged
743 * TcpExtTCPSackShiftFallback
745 A skb should be shifted or merged, but the TCP stack doesn't do it for
752 The TCP layer receives an out of order packet and has enough memory
757 The TCP layer receives an out of order packet but doesn't have enough
758 memory, so drops it. Such packets won't be counted into
763 The received out of order packet has an overlay with the previous
764 packet. the overlay part will be dropped. All of TcpExtTCPOFOMerge
765 packets will also be counted into TcpExtTCPOFOQueue.
769 PAWS (Protection Against Wrapped Sequence numbers) is an algorithm
770 which is used to drop old packets. It depends on the TCP
771 timestamps. For detail information, please refer the `timestamp wiki`_
772 and the `RFC of PAWS`_.
774 .. _RFC of PAWS: https://tools.ietf.org/html/rfc1323#page-17
775 .. _timestamp wiki: https://en.wikipedia.org/wiki/Transmission_Control_Protocol#TCP_timestamps
779 Packets are dropped by PAWS in Syn-Sent status.
783 Packets are dropped by PAWS in any status other than Syn-Sent.
787 In some scenarios, kernel would avoid sending duplicate ACKs too
788 frequently. Please find more details in the tcp_invalid_ratelimit
789 section of the `sysctl document`_. When kernel decides to skip an ACK
790 due to tcp_invalid_ratelimit, kernel would update one of below
791 counters to indicate the ACK is skipped in which scenario. The ACK
792 would only be skipped if the received packet is either a SYN packet or
795 .. _sysctl document: https://www.kernel.org/doc/Documentation/networking/ip-sysctl.txt
797 * TcpExtTCPACKSkippedSynRecv
799 The ACK is skipped in Syn-Recv status. The Syn-Recv status means the
800 TCP stack receives a SYN and replies SYN+ACK. Now the TCP stack is
801 waiting for an ACK. Generally, the TCP stack doesn't need to send ACK
802 in the Syn-Recv status. But in several scenarios, the TCP stack need
803 to send an ACK. E.g., the TCP stack receives the same SYN packet
804 repeately, the received packet does not pass the PAWS check, or the
805 received packet sequence number is out of window. In these scenarios,
806 the TCP stack needs to send ACK. If the ACk sending frequency is higher than
807 tcp_invalid_ratelimit allows, the TCP stack will skip sending ACK and
808 increase TcpExtTCPACKSkippedSynRecv.
811 * TcpExtTCPACKSkippedPAWS
813 The ACK is skipped due to PAWS (Protect Against Wrapped Sequence
814 numbers) check fails. If the PAWS check fails in Syn-Recv, Fin-Wait-2
815 or Time-Wait statuses, the skipped ACK would be counted to
816 TcpExtTCPACKSkippedSynRecv, TcpExtTCPACKSkippedFinWait2 or
817 TcpExtTCPACKSkippedTimeWait. In all other statuses, the skipped ACK
818 would be counted to TcpExtTCPACKSkippedPAWS.
820 * TcpExtTCPACKSkippedSeq
822 The sequence number is out of window and the timestamp passes the PAWS
823 check and the TCP status is not Syn-Recv, Fin-Wait-2, and Time-Wait.
825 * TcpExtTCPACKSkippedFinWait2
827 The ACK is skipped in Fin-Wait-2 status, the reason would be either
828 PAWS check fails or the received sequence number is out of window.
830 * TcpExtTCPACKSkippedTimeWait
832 Tha ACK is skipped in Time-Wait status, the reason would be either
833 PAWS check failed or the received sequence number is out of window.
835 * TcpExtTCPACKSkippedChallenge
837 The ACK is skipped if the ACK is a challenge ACK. The RFC 5961 defines
838 3 kind of challenge ACK, please refer `RFC 5961 section 3.2`_,
839 `RFC 5961 section 4.2`_ and `RFC 5961 section 5.2`_. Besides these
840 three scenarios, In some TCP status, the linux TCP stack would also
841 send challenge ACKs if the ACK number is before the first
842 unacknowledged number (more strict than `RFC 5961 section 5.2`_).
844 .. _RFC 5961 section 3.2: https://tools.ietf.org/html/rfc5961#page-7
845 .. _RFC 5961 section 4.2: https://tools.ietf.org/html/rfc5961#page-9
846 .. _RFC 5961 section 5.2: https://tools.ietf.org/html/rfc5961#page-11
850 * TcpExtTCPWantZeroWindowAdv
852 Depending on current memory usage, the TCP stack tries to set receive
853 window to zero. But the receive window might still be a no-zero
854 value. For example, if the previous window size is 10, and the TCP
855 stack receives 3 bytes, the current window size would be 7 even if the
856 window size calculated by the memory usage is zero.
858 * TcpExtTCPToZeroWindowAdv
860 The TCP receive window is set to zero from a no-zero value.
862 * TcpExtTCPFromZeroWindowAdv
864 The TCP receive window is set to no-zero value from zero.
869 The TCP Delayed ACK is a technique which is used for reducing the
870 packet count in the network. For more details, please refer the
873 .. _Delayed ACK wiki: https://en.wikipedia.org/wiki/TCP_delayed_acknowledgment
877 A delayed ACK timer expires. The TCP stack will send a pure ACK packet
878 and exit the delayed ACK mode.
880 * TcpExtDelayedACKLocked
882 A delayed ACK timer expires, but the TCP stack can't send an ACK
883 immediately due to the socket is locked by a userspace program. The
884 TCP stack will send a pure ACK later (after the userspace program
885 unlock the socket). When the TCP stack sends the pure ACK later, the
886 TCP stack will also update TcpExtDelayedACKs and exit the delayed ACK
889 * TcpExtDelayedACKLost
891 It will be updated when the TCP stack receives a packet which has been
892 ACKed. A Delayed ACK loss might cause this issue, but it would also be
893 triggered by other reasons, such as a packet is duplicated in the
896 Tail Loss Probe (TLP)
897 =====================
898 TLP is an algorithm which is used to detect TCP packet loss. For more
899 details, please refer the `TLP paper`_.
901 .. _TLP paper: https://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01
903 * TcpExtTCPLossProbes
905 A TLP probe packet is sent.
907 * TcpExtTCPLossProbeRecovery
909 A packet loss is detected and recovered by TLP.
913 TCP Fast Open is a technology which allows data transfer before the
914 3-way handshake complete. Please refer the `TCP Fast Open wiki`_ for a
917 .. _TCP Fast Open wiki: https://en.wikipedia.org/wiki/TCP_Fast_Open
919 * TcpExtTCPFastOpenActive
921 When the TCP stack receives an ACK packet in the SYN-SENT status, and
922 the ACK packet acknowledges the data in the SYN packet, the TCP stack
923 understand the TFO cookie is accepted by the other side, then it
924 updates this counter.
926 * TcpExtTCPFastOpenActiveFail
928 This counter indicates that the TCP stack initiated a TCP Fast Open,
929 but it failed. This counter would be updated in three scenarios: (1)
930 the other side doesn't acknowledge the data in the SYN packet. (2) The
931 SYN packet which has the TFO cookie is timeout at least once. (3)
932 after the 3-way handshake, the retransmission timeout happens
933 net.ipv4.tcp_retries1 times, because some middle-boxes may black-hole
934 fast open after the handshake.
936 * TcpExtTCPFastOpenPassive
938 This counter indicates how many times the TCP stack accepts the fast
941 * TcpExtTCPFastOpenPassiveFail
943 This counter indicates how many times the TCP stack rejects the fast
944 open request. It is caused by either the TFO cookie is invalid or the
945 TCP stack finds an error during the socket creating process.
947 * TcpExtTCPFastOpenListenOverflow
949 When the pending fast open request number is larger than
950 fastopenq->max_qlen, the TCP stack will reject the fast open request
951 and update this counter. When this counter is updated, the TCP stack
952 won't update TcpExtTCPFastOpenPassive or
953 TcpExtTCPFastOpenPassiveFail. The fastopenq->max_qlen is set by the
954 TCP_FASTOPEN socket operation and it could not be larger than
955 net.core.somaxconn. For example:
957 setsockopt(sfd, SOL_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen));
959 * TcpExtTCPFastOpenCookieReqd
961 This counter indicates how many times a client wants to request a TFO
966 SYN cookies are used to mitigate SYN flood, for details, please refer
967 the `SYN cookies wiki`_.
969 .. _SYN cookies wiki: https://en.wikipedia.org/wiki/SYN_cookies
971 * TcpExtSyncookiesSent
973 It indicates how many SYN cookies are sent.
975 * TcpExtSyncookiesRecv
977 How many reply packets of the SYN cookies the TCP stack receives.
979 * TcpExtSyncookiesFailed
981 The MSS decoded from the SYN cookie is invalid. When this counter is
982 updated, the received packet won't be treated as a SYN cookie and the
983 TcpExtSyncookiesRecv counter wont be updated.
987 For details of challenge ACK, please refer the explaination of
988 TcpExtTCPACKSkippedChallenge.
990 * TcpExtTCPChallengeACK
992 The number of challenge acks sent.
994 * TcpExtTCPSYNChallenge
996 The number of challenge acks sent in response to SYN packets. After
997 updates this counter, the TCP stack might send a challenge ACK and
998 update the TcpExtTCPChallengeACK counter, or it might also skip to
999 send the challenge and update the TcpExtTCPACKSkippedChallenge.
1003 When a socket is under memory pressure, the TCP stack will try to
1004 reclaim memory from the receiving queue and out of order queue. One of
1005 the reclaiming method is 'collapse', which means allocate a big sbk,
1006 copy the contiguous skbs to the single big skb, and free these
1011 The TCP stack tries to reclaim memory for a socket. After updates this
1012 counter, the TCP stack will try to collapse the out of order queue and
1013 the receiving queue. If the memory is still not enough, the TCP stack
1014 will try to discard packets from the out of order queue (and update the
1015 TcpExtOfoPruned counter)
1019 The TCP stack tries to discard packet on the out of order queue.
1023 After 'collapse' and discard packets from the out of order queue, if
1024 the actually used memory is still larger than the max allowed memory,
1025 this counter will be updated. It means the 'prune' fails.
1027 * TcpExtTCPRcvCollapsed
1029 This counter indicates how many skbs are freed during 'collapse'.
1036 Run the ping command against the public dns server 8.8.8.8::
1038 nstatuser@nstat-a:~$ ping 8.8.8.8 -c 1
1039 PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data.
1040 64 bytes from 8.8.8.8: icmp_seq=1 ttl=119 time=17.8 ms
1042 --- 8.8.8.8 ping statistics ---
1043 1 packets transmitted, 1 received, 0% packet loss, time 0ms
1044 rtt min/avg/max/mdev = 17.875/17.875/17.875/0.000 ms
1048 nstatuser@nstat-a:~$ nstat
1054 IcmpInEchoReps 1 0.0
1057 IcmpMsgInType0 1 0.0
1058 IcmpMsgOutType8 1 0.0
1059 IpExtInOctets 84 0.0
1060 IpExtOutOctets 84 0.0
1061 IpExtInNoECTPkts 1 0.0
1063 The Linux server sent an ICMP Echo packet, so IpOutRequests,
1064 IcmpOutMsgs, IcmpOutEchos and IcmpMsgOutType8 were increased 1. The
1065 server got ICMP Echo Reply from 8.8.8.8, so IpInReceives, IcmpInMsgs,
1066 IcmpInEchoReps and IcmpMsgInType0 were increased 1. The ICMP Echo Reply
1067 was passed to the ICMP layer via IP layer, so IpInDelivers was
1068 increased 1. The default ping data size is 48, so an ICMP Echo packet
1069 and its corresponding Echo Reply packet are constructed by:
1071 * 14 bytes MAC header
1072 * 20 bytes IP header
1073 * 16 bytes ICMP header
1074 * 48 bytes data (default value of the ping command)
1076 So the IpExtInOctets and IpExtOutOctets are 20+16+48=84.
1080 On server side, we run::
1082 nstatuser@nstat-b:~$ nc -lknv 0.0.0.0 9000
1083 Listening on [0.0.0.0] (family 0, port 9000)
1085 On client side, we run::
1087 nstatuser@nstat-a:~$ nc -nv 192.168.122.251 9000
1088 Connection to 192.168.122.251 9000 port [tcp/*] succeeded!
1090 The server listened on tcp 9000 port, the client connected to it, they
1091 completed the 3-way handshake.
1093 On server side, we can find below nstat output::
1095 nstatuser@nstat-b:~$ nstat | grep -i tcp
1096 TcpPassiveOpens 1 0.0
1099 TcpExtTCPPureAcks 1 0.0
1101 On client side, we can find below nstat output::
1103 nstatuser@nstat-a:~$ nstat | grep -i tcp
1104 TcpActiveOpens 1 0.0
1108 When the server received the first SYN, it replied a SYN+ACK, and came into
1109 SYN-RCVD state, so TcpPassiveOpens increased 1. The server received
1110 SYN, sent SYN+ACK, received ACK, so server sent 1 packet, received 2
1111 packets, TcpInSegs increased 2, TcpOutSegs increased 1. The last ACK
1112 of the 3-way handshake is a pure ACK without data, so
1113 TcpExtTCPPureAcks increased 1.
1115 When the client sent SYN, the client came into the SYN-SENT state, so
1116 TcpActiveOpens increased 1, the client sent SYN, received SYN+ACK, sent
1117 ACK, so client sent 2 packets, received 1 packet, TcpInSegs increased
1118 1, TcpOutSegs increased 2.
1124 nstatuser@nstat-b:~$ nc -lkv 0.0.0.0 9000
1125 Listening on [0.0.0.0] (family 0, port 9000)
1129 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1130 Connection to nstat-b 9000 port [tcp/*] succeeded!
1132 Input a string in the nc client ('hello' in our example)::
1134 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1135 Connection to nstat-b 9000 port [tcp/*] succeeded!
1138 The client side nstat output::
1140 nstatuser@nstat-a:~$ nstat
1147 TcpExtTCPPureAcks 1 0.0
1148 TcpExtTCPOrigDataSent 1 0.0
1149 IpExtInOctets 52 0.0
1150 IpExtOutOctets 58 0.0
1151 IpExtInNoECTPkts 1 0.0
1153 The server side nstat output::
1155 nstatuser@nstat-b:~$ nstat
1162 IpExtInOctets 58 0.0
1163 IpExtOutOctets 52 0.0
1164 IpExtInNoECTPkts 1 0.0
1166 Input a string in nc client side again ('world' in our exmaple)::
1168 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1169 Connection to nstat-b 9000 port [tcp/*] succeeded!
1173 Client side nstat output::
1175 nstatuser@nstat-a:~$ nstat
1182 TcpExtTCPHPAcks 1 0.0
1183 TcpExtTCPOrigDataSent 1 0.0
1184 IpExtInOctets 52 0.0
1185 IpExtOutOctets 58 0.0
1186 IpExtInNoECTPkts 1 0.0
1189 Server side nstat output::
1191 nstatuser@nstat-b:~$ nstat
1198 TcpExtTCPHPHits 1 0.0
1199 IpExtInOctets 58 0.0
1200 IpExtOutOctets 52 0.0
1201 IpExtInNoECTPkts 1 0.0
1203 Compare the first client-side nstat and the second client-side nstat,
1204 we could find one difference: the first one had a 'TcpExtTCPPureAcks',
1205 but the second one had a 'TcpExtTCPHPAcks'. The first server-side
1206 nstat and the second server-side nstat had a difference too: the
1207 second server-side nstat had a TcpExtTCPHPHits, but the first
1208 server-side nstat didn't have it. The network traffic patterns were
1209 exactly the same: the client sent a packet to the server, the server
1210 replied an ACK. But kernel handled them in different ways. When the
1211 TCP window scale option is not used, kernel will try to enable fast
1212 path immediately when the connection comes into the established state,
1213 but if the TCP window scale option is used, kernel will disable the
1214 fast path at first, and try to enable it after kerenl receives
1215 packets. We could use the 'ss' command to verify whether the window
1216 scale option is used. e.g. run below command on either server or
1219 nstatuser@nstat-a:~$ ss -o state established -i '( dport = :9000 or sport = :9000 )
1220 Netid Recv-Q Send-Q Local Address:Port Peer Address:Port
1221 tcp 0 0 192.168.122.250:40654 192.168.122.251:9000
1222 ts sack cubic wscale:7,7 rto:204 rtt:0.98/0.49 mss:1448 pmtu:1500 rcvmss:536 advmss:1448 cwnd:10 bytes_acked:1 segs_out:2 segs_in:1 send 118.2Mbps lastsnd:46572 lastrcv:46572 lastack:46572 pacing_rate 236.4Mbps rcv_space:29200 rcv_ssthresh:29200 minrtt:0.98
1224 The 'wscale:7,7' means both server and client set the window scale
1225 option to 7. Now we could explain the nstat output in our test:
1227 In the first nstat output of client side, the client sent a packet, server
1228 reply an ACK, when kernel handled this ACK, the fast path was not
1229 enabled, so the ACK was counted into 'TcpExtTCPPureAcks'.
1231 In the second nstat output of client side, the client sent a packet again,
1232 and received another ACK from the server, in this time, the fast path is
1233 enabled, and the ACK was qualified for fast path, so it was handled by
1234 the fast path, so this ACK was counted into TcpExtTCPHPAcks.
1236 In the first nstat output of server side, fast path was not enabled,
1237 so there was no 'TcpExtTCPHPHits'.
1239 In the second nstat output of server side, the fast path was enabled,
1240 and the packet received from client qualified for fast path, so it
1241 was counted into 'TcpExtTCPHPHits'.
1243 TcpExtTCPAbortOnClose
1244 ---------------------
1245 On the server side, we run below python script::
1252 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1253 s.bind(('0.0.0.0', port))
1255 sock, addr = s.accept()
1259 This python script listen on 9000 port, but doesn't read anything from
1262 On the client side, we send the string "hello" by nc::
1264 nstatuser@nstat-a:~$ echo "hello" | nc nstat-b 9000
1266 Then, we come back to the server side, the server has received the "hello"
1267 packet, and the TCP layer has acked this packet, but the application didn't
1268 read it yet. We type Ctrl-C to terminate the server script. Then we
1269 could find TcpExtTCPAbortOnClose increased 1 on the server side::
1271 nstatuser@nstat-b:~$ nstat | grep -i abort
1272 TcpExtTCPAbortOnClose 1 0.0
1274 If we run tcpdump on the server side, we could find the server sent a
1275 RST after we type Ctrl-C.
1277 TcpExtTCPAbortOnMemory and TcpExtTCPAbortOnTimeout
1278 ---------------------------------------------------
1279 Below is an example which let the orphan socket count be higher than
1280 net.ipv4.tcp_max_orphans.
1281 Change tcp_max_orphans to a smaller value on client::
1283 sudo bash -c "echo 10 > /proc/sys/net/ipv4/tcp_max_orphans"
1285 Client code (create 64 connection to server)::
1287 nstatuser@nstat-a:~$ cat client_orphan.py
1291 server = 'nstat-b' # server address
1296 connection_list = []
1299 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1300 s.connect((server, port))
1301 connection_list.append(s)
1302 print("connection_count: %d" % len(connection_list))
1307 Server code (accept 64 connection from client)::
1309 nstatuser@nstat-b:~$ cat server_orphan.py
1316 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1317 s.bind(('0.0.0.0', port))
1319 connection_list = []
1321 sock, addr = s.accept()
1322 connection_list.append((sock, addr))
1323 print("connection_count: %d" % len(connection_list))
1325 Run the python scripts on server and client.
1329 python3 server_orphan.py
1333 python3 client_orphan.py
1335 Run iptables on server::
1337 sudo iptables -A INPUT -i ens3 -p tcp --destination-port 9000 -j DROP
1339 Type Ctrl-C on client, stop client_orphan.py.
1341 Check TcpExtTCPAbortOnMemory on client::
1343 nstatuser@nstat-a:~$ nstat | grep -i abort
1344 TcpExtTCPAbortOnMemory 54 0.0
1346 Check orphane socket count on client::
1348 nstatuser@nstat-a:~$ ss -s
1349 Total: 131 (kernel 0)
1350 TCP: 14 (estab 1, closed 0, orphaned 10, synrecv 0, timewait 0/0), ports 0
1352 Transport Total IP IPv6
1360 The explanation of the test: after run server_orphan.py and
1361 client_orphan.py, we set up 64 connections between server and
1362 client. Run the iptables command, the server will drop all packets from
1363 the client, type Ctrl-C on client_orphan.py, the system of the client
1364 would try to close these connections, and before they are closed
1365 gracefully, these connections became orphan sockets. As the iptables
1366 of the server blocked packets from the client, the server won't receive fin
1367 from the client, so all connection on clients would be stuck on FIN_WAIT_1
1368 stage, so they will keep as orphan sockets until timeout. We have echo
1369 10 to /proc/sys/net/ipv4/tcp_max_orphans, so the client system would
1370 only keep 10 orphan sockets, for all other orphan sockets, the client
1371 system sent RST for them and delete them. We have 64 connections, so
1372 the 'ss -s' command shows the system has 10 orphan sockets, and the
1373 value of TcpExtTCPAbortOnMemory was 54.
1375 An additional explanation about orphan socket count: You could find the
1376 exactly orphan socket count by the 'ss -s' command, but when kernel
1377 decide whither increases TcpExtTCPAbortOnMemory and sends RST, kernel
1378 doesn't always check the exactly orphan socket count. For increasing
1379 performance, kernel checks an approximate count firstly, if the
1380 approximate count is more than tcp_max_orphans, kernel checks the
1381 exact count again. So if the approximate count is less than
1382 tcp_max_orphans, but exactly count is more than tcp_max_orphans, you
1383 would find TcpExtTCPAbortOnMemory is not increased at all. If
1384 tcp_max_orphans is large enough, it won't occur, but if you decrease
1385 tcp_max_orphans to a small value like our test, you might find this
1386 issue. So in our test, the client set up 64 connections although the
1387 tcp_max_orphans is 10. If the client only set up 11 connections, we
1388 can't find the change of TcpExtTCPAbortOnMemory.
1390 Continue the previous test, we wait for several minutes. Because of the
1391 iptables on the server blocked the traffic, the server wouldn't receive
1392 fin, and all the client's orphan sockets would timeout on the
1393 FIN_WAIT_1 state finally. So we wait for a few minutes, we could find
1394 10 timeout on the client::
1396 nstatuser@nstat-a:~$ nstat | grep -i abort
1397 TcpExtTCPAbortOnTimeout 10 0.0
1399 TcpExtTCPAbortOnLinger
1400 ----------------------
1401 The server side code::
1403 nstatuser@nstat-b:~$ cat server_linger.py
1409 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1410 s.bind(('0.0.0.0', port))
1412 sock, addr = s.accept()
1416 The client side code::
1418 nstatuser@nstat-a:~$ cat client_linger.py
1422 server = 'nstat-b' # server address
1425 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1426 s.setsockopt(socket.SOL_SOCKET, socket.SO_LINGER, struct.pack('ii', 1, 10))
1427 s.setsockopt(socket.SOL_TCP, socket.TCP_LINGER2, struct.pack('i', -1))
1428 s.connect((server, port))
1431 Run server_linger.py on server::
1433 nstatuser@nstat-b:~$ python3 server_linger.py
1435 Run client_linger.py on client::
1437 nstatuser@nstat-a:~$ python3 client_linger.py
1439 After run client_linger.py, check the output of nstat::
1441 nstatuser@nstat-a:~$ nstat | grep -i abort
1442 TcpExtTCPAbortOnLinger 1 0.0
1444 TcpExtTCPRcvCoalesce
1445 --------------------
1446 On the server, we run a program which listen on TCP port 9000, but
1447 doesn't read any data::
1452 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1453 s.bind(('0.0.0.0', port))
1455 sock, addr = s.accept()
1459 Save the above code as server_coalesce.py, and run::
1461 python3 server_coalesce.py
1463 On the client, save below code as client_coalesce.py::
1468 s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
1469 s.connect((server, port))
1473 nstatuser@nstat-a:~$ python3 -i client_coalesce.py
1475 We use '-i' to come into the interactive mode, then a packet::
1480 Send a packet again::
1485 On the server, run nstat::
1487 ubuntu@nstat-b:~$ nstat
1494 TcpExtTCPRcvCoalesce 1 0.0
1495 IpExtInOctets 110 0.0
1496 IpExtOutOctets 104 0.0
1497 IpExtInNoECTPkts 2 0.0
1499 The client sent two packets, server didn't read any data. When
1500 the second packet arrived at server, the first packet was still in
1501 the receiving queue. So the TCP layer merged the two packets, and we
1502 could find the TcpExtTCPRcvCoalesce increased 1.
1504 TcpExtListenOverflows and TcpExtListenDrops
1505 -------------------------------------------
1506 On server, run the nc command, listen on port 9000::
1508 nstatuser@nstat-b:~$ nc -lkv 0.0.0.0 9000
1509 Listening on [0.0.0.0] (family 0, port 9000)
1511 On client, run 3 nc commands in different terminals::
1513 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1514 Connection to nstat-b 9000 port [tcp/*] succeeded!
1516 The nc command only accepts 1 connection, and the accept queue length
1517 is 1. On current linux implementation, set queue length to n means the
1518 actual queue length is n+1. Now we create 3 connections, 1 is accepted
1519 by nc, 2 in accepted queue, so the accept queue is full.
1521 Before running the 4th nc, we clean the nstat history on the server::
1523 nstatuser@nstat-b:~$ nstat -n
1525 Run the 4th nc on the client::
1527 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1529 If the nc server is running on kernel 4.10 or higher version, you
1530 won't see the "Connection to ... succeeded!" string, because kernel
1531 will drop the SYN if the accept queue is full. If the nc client is running
1532 on an old kernel, you would see that the connection is succeeded,
1533 because kernel would complete the 3 way handshake and keep the socket
1534 on half open queue. I did the test on kernel 4.15. Below is the nstat
1537 nstatuser@nstat-b:~$ nstat
1542 TcpExtListenOverflows 4 0.0
1543 TcpExtListenDrops 4 0.0
1544 IpExtInOctets 240 0.0
1545 IpExtInNoECTPkts 4 0.0
1547 Both TcpExtListenOverflows and TcpExtListenDrops were 4. If the time
1548 between the 4th nc and the nstat was longer, the value of
1549 TcpExtListenOverflows and TcpExtListenDrops would be larger, because
1550 the SYN of the 4th nc was dropped, the client was retrying.
1552 IpInAddrErrors, IpExtInNoRoutes and IpOutNoRoutes
1553 -------------------------------------------------
1554 server A IP address: 192.168.122.250
1555 server B IP address: 192.168.122.251
1556 Prepare on server A, add a route to server B::
1558 $ sudo ip route add 8.8.8.8/32 via 192.168.122.251
1560 Prepare on server B, disable send_redirects for all interfaces::
1562 $ sudo sysctl -w net.ipv4.conf.all.send_redirects=0
1563 $ sudo sysctl -w net.ipv4.conf.ens3.send_redirects=0
1564 $ sudo sysctl -w net.ipv4.conf.lo.send_redirects=0
1565 $ sudo sysctl -w net.ipv4.conf.default.send_redirects=0
1567 We want to let sever A send a packet to 8.8.8.8, and route the packet
1568 to server B. When server B receives such packet, it might send a ICMP
1569 Redirect message to server A, set send_redirects to 0 will disable
1572 First, generate InAddrErrors. On server B, we disable IP forwarding::
1574 $ sudo sysctl -w net.ipv4.conf.all.forwarding=0
1576 On server A, we send packets to 8.8.8.8::
1580 On server B, we check the output of nstat::
1585 IpInAddrErrors 3 0.0
1586 IpExtInOctets 180 0.0
1587 IpExtInNoECTPkts 3 0.0
1589 As we have let server A route 8.8.8.8 to server B, and we disabled IP
1590 forwarding on server B, Server A sent packets to server B, then server B
1591 dropped packets and increased IpInAddrErrors. As the nc command would
1592 re-send the SYN packet if it didn't receive a SYN+ACK, we could find
1593 multiple IpInAddrErrors.
1595 Second, generate IpExtInNoRoutes. On server B, we enable IP
1598 $ sudo sysctl -w net.ipv4.conf.all.forwarding=1
1600 Check the route table of server B and remove the default route::
1603 default via 192.168.122.1 dev ens3 proto static
1604 192.168.122.0/24 dev ens3 proto kernel scope link src 192.168.122.251
1605 $ sudo ip route delete default via 192.168.122.1 dev ens3 proto static
1607 On server A, we contact 8.8.8.8 again::
1610 nc: connect to 8.8.8.8 port 53 (tcp) failed: Network is unreachable
1612 On server B, run nstat::
1619 IcmpOutDestUnreachs 1 0.0
1620 IcmpMsgOutType3 1 0.0
1621 IpExtInNoRoutes 1 0.0
1622 IpExtInOctets 60 0.0
1623 IpExtOutOctets 88 0.0
1624 IpExtInNoECTPkts 1 0.0
1626 We enabled IP forwarding on server B, when server B received a packet
1627 which destination IP address is 8.8.8.8, server B will try to forward
1628 this packet. We have deleted the default route, there was no route for
1629 8.8.8.8, so server B increase IpExtInNoRoutes and sent the "ICMP
1630 Destination Unreachable" message to server A.
1632 Third, generate IpOutNoRoutes. Run ping command on server B::
1635 connect: Network is unreachable
1637 Run nstat on server B::
1643 We have deleted the default route on server B. Server B couldn't find
1644 a route for the 8.8.8.8 IP address, so server B increased
1647 TcpExtTCPACKSkippedSynRecv
1648 --------------------------
1649 In this test, we send 3 same SYN packets from client to server. The
1650 first SYN will let server create a socket, set it to Syn-Recv status,
1651 and reply a SYN/ACK. The second SYN will let server reply the SYN/ACK
1652 again, and record the reply time (the duplicate ACK reply time). The
1653 third SYN will let server check the previous duplicate ACK reply time,
1654 and decide to skip the duplicate ACK, then increase the
1655 TcpExtTCPACKSkippedSynRecv counter.
1657 Run tcpdump to capture a SYN packet::
1659 nstatuser@nstat-a:~$ sudo tcpdump -c 1 -w /tmp/syn.pcap port 9000
1660 tcpdump: listening on ens3, link-type EN10MB (Ethernet), capture size 262144 bytes
1662 Open another terminal, run nc command::
1664 nstatuser@nstat-a:~$ nc nstat-b 9000
1666 As the nstat-b didn't listen on port 9000, it should reply a RST, and
1667 the nc command exited immediately. It was enough for the tcpdump
1668 command to capture a SYN packet. A linux server might use hardware
1669 offload for the TCP checksum, so the checksum in the /tmp/syn.pcap
1670 might be not correct. We call tcprewrite to fix it::
1672 nstatuser@nstat-a:~$ tcprewrite --infile=/tmp/syn.pcap --outfile=/tmp/syn_fixcsum.pcap --fixcsum
1674 On nstat-b, we run nc to listen on port 9000::
1676 nstatuser@nstat-b:~$ nc -lkv 9000
1677 Listening on [0.0.0.0] (family 0, port 9000)
1679 On nstat-a, we blocked the packet from port 9000, or nstat-a would send
1682 nstatuser@nstat-a:~$ sudo iptables -A INPUT -p tcp --sport 9000 -j DROP
1684 Send 3 SYN repeatly to nstat-b::
1686 nstatuser@nstat-a:~$ for i in {1..3}; do sudo tcpreplay -i ens3 /tmp/syn_fixcsum.pcap; done
1688 Check snmp cunter on nstat-b::
1690 nstatuser@nstat-b:~$ nstat | grep -i skip
1691 TcpExtTCPACKSkippedSynRecv 1 0.0
1693 As we expected, TcpExtTCPACKSkippedSynRecv is 1.
1695 TcpExtTCPACKSkippedPAWS
1696 -----------------------
1697 To trigger PAWS, we could send an old SYN.
1699 On nstat-b, let nc listen on port 9000::
1701 nstatuser@nstat-b:~$ nc -lkv 9000
1702 Listening on [0.0.0.0] (family 0, port 9000)
1704 On nstat-a, run tcpdump to capture a SYN::
1706 nstatuser@nstat-a:~$ sudo tcpdump -w /tmp/paws_pre.pcap -c 1 port 9000
1707 tcpdump: listening on ens3, link-type EN10MB (Ethernet), capture size 262144 bytes
1709 On nstat-a, run nc as a client to connect nstat-b::
1711 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1712 Connection to nstat-b 9000 port [tcp/*] succeeded!
1714 Now the tcpdump has captured the SYN and exit. We should fix the
1717 nstatuser@nstat-a:~$ tcprewrite --infile /tmp/paws_pre.pcap --outfile /tmp/paws.pcap --fixcsum
1719 Send the SYN packet twice::
1721 nstatuser@nstat-a:~$ for i in {1..2}; do sudo tcpreplay -i ens3 /tmp/paws.pcap; done
1723 On nstat-b, check the snmp counter::
1725 nstatuser@nstat-b:~$ nstat | grep -i skip
1726 TcpExtTCPACKSkippedPAWS 1 0.0
1728 We sent two SYN via tcpreplay, both of them would let PAWS check
1729 failed, the nstat-b replied an ACK for the first SYN, skipped the ACK
1730 for the second SYN, and updated TcpExtTCPACKSkippedPAWS.
1732 TcpExtTCPACKSkippedSeq
1733 ----------------------
1734 To trigger TcpExtTCPACKSkippedSeq, we send packets which have valid
1735 timestamp (to pass PAWS check) but the sequence number is out of
1736 window. The linux TCP stack would avoid to skip if the packet has
1737 data, so we need a pure ACK packet. To generate such a packet, we
1738 could create two sockets: one on port 9000, another on port 9001. Then
1739 we capture an ACK on port 9001, change the source/destination port
1740 numbers to match the port 9000 socket. Then we could trigger
1741 TcpExtTCPACKSkippedSeq via this packet.
1743 On nstat-b, open two terminals, run two nc commands to listen on both
1744 port 9000 and port 9001::
1746 nstatuser@nstat-b:~$ nc -lkv 9000
1747 Listening on [0.0.0.0] (family 0, port 9000)
1749 nstatuser@nstat-b:~$ nc -lkv 9001
1750 Listening on [0.0.0.0] (family 0, port 9001)
1752 On nstat-a, run two nc clients::
1754 nstatuser@nstat-a:~$ nc -v nstat-b 9000
1755 Connection to nstat-b 9000 port [tcp/*] succeeded!
1757 nstatuser@nstat-a:~$ nc -v nstat-b 9001
1758 Connection to nstat-b 9001 port [tcp/*] succeeded!
1760 On nstat-a, run tcpdump to capture an ACK::
1762 nstatuser@nstat-a:~$ sudo tcpdump -w /tmp/seq_pre.pcap -c 1 dst port 9001
1763 tcpdump: listening on ens3, link-type EN10MB (Ethernet), capture size 262144 bytes
1765 On nstat-b, send a packet via the port 9001 socket. E.g. we sent a
1766 string 'foo' in our example::
1768 nstatuser@nstat-b:~$ nc -lkv 9001
1769 Listening on [0.0.0.0] (family 0, port 9001)
1770 Connection from nstat-a 42132 received!
1773 On nstat-a, the tcpdump should have caputred the ACK. We should check
1774 the source port numbers of the two nc clients::
1776 nstatuser@nstat-a:~$ ss -ta '( dport = :9000 || dport = :9001 )' | tee
1777 State Recv-Q Send-Q Local Address:Port Peer Address:Port
1778 ESTAB 0 0 192.168.122.250:50208 192.168.122.251:9000
1779 ESTAB 0 0 192.168.122.250:42132 192.168.122.251:9001
1781 Run tcprewrite, change port 9001 to port 9000, chagne port 42132 to
1784 nstatuser@nstat-a:~$ tcprewrite --infile /tmp/seq_pre.pcap --outfile /tmp/seq.pcap -r 9001:9000 -r 42132:50208 --fixcsum
1786 Now the /tmp/seq.pcap is the packet we need. Send it to nstat-b::
1788 nstatuser@nstat-a:~$ for i in {1..2}; do sudo tcpreplay -i ens3 /tmp/seq.pcap; done
1790 Check TcpExtTCPACKSkippedSeq on nstat-b::
1792 nstatuser@nstat-b:~$ nstat | grep -i skip
1793 TcpExtTCPACKSkippedSeq 1 0.0