1TC(8) Linux TC(8)
2
6 tc - show / manipulate traffic control settings
7
9 tc [ OPTIONS ] qdisc [ add | change | replace | link | delete ] dev DEV
10 [ parent qdisc-id | root ] [ handle qdisc-id ] [ ingress_block
11 BLOCK_INDEX ] [ egress_block BLOCK_INDEX ] qdisc [ qdisc specific
12 parameters ]
13 tc [ OPTIONS ] class [ add | change | replace | delete ] dev DEV parent
15 qdisc-id [ classid class-id ] qdisc [ qdisc specific parameters ]
16 tc [ OPTIONS ] filter [ add | change | replace | delete | get ] dev DEV
18 [ parent qdisc-id | root ] [ handle filter-id ] protocol protocol prio
19 priority filtertype [ filtertype specific parameters ] flowid flow-id
20 tc [ OPTIONS ] filter [ add | change | replace | delete | get ] block
22 BLOCK_INDEX [ handle filter-id ] protocol protocol prio priority fil‐
23 tertype [ filtertype specific parameters ] flowid flow-id
24 tc [ OPTIONS ] chain [ add | delete | get ] dev DEV [ parent qdisc-id |
26 root ] filtertype [ filtertype specific parameters ]
27 tc [ OPTIONS ] chain [ add | delete | get ] block BLOCK_INDEX filter‐
29 type [ filtertype specific parameters ]
30 tc [ OPTIONS ] [ FORMAT ] qdisc show [ dev DEV ]
33 tc [ OPTIONS ] [ FORMAT ] class show dev DEV
35 tc [ OPTIONS ] filter show dev DEV
37 tc [ OPTIONS ] filter show block BLOCK_INDEX
39 tc [ OPTIONS ] chain show dev DEV
41 tc [ OPTIONS ] chain show block BLOCK_INDEX
43 tc [ OPTIONS ] monitor [ file FILENAME ]
46 OPTIONS := { [ -force ] -b[atch] [ filename ] | [ -n[etns] name ] | [
49 -N[umeric] ] | [ -nm | -nam[es] ] | [ { -cf | -c[onf] } [ filename ] ]
50 [ -t[imestamp] ] | [ -t[short] | [ -o[neline] ] }
51 FORMAT := { -s[tatistics] | -d[etails] | -r[aw] | -i[ec] | -g[raph] |
53 -j[json] | -p[retty] | -col[or] }
54
57 Tc is used to configure Traffic Control in the Linux kernel. Traffic
58 Control consists of the following:
59 SHAPING
62 When traffic is shaped, its rate of transmission is under con‐
63 trol. Shaping may be more than lowering the available bandwidth
64 - it is also used to smooth out bursts in traffic for better
65 network behaviour. Shaping occurs on egress.
66 SCHEDULING
69 By scheduling the transmission of packets it is possible to
70 improve interactivity for traffic that needs it while still
71 guaranteeing bandwidth to bulk transfers. Reordering is also
72 called prioritizing, and happens only on egress.
73 POLICING
76 Whereas shaping deals with transmission of traffic, policing
77 pertains to traffic arriving. Policing thus occurs on ingress.
78 DROPPING
81 Traffic exceeding a set bandwidth may also be dropped forthwith,
82 both on ingress and on egress.
83 Processing of traffic is controlled by three kinds of objects: qdiscs,
86 classes and filters.
87
90 qdisc is short for 'queueing discipline' and it is elementary to under‐
91 standing traffic control. Whenever the kernel needs to send a packet to
92 an interface, it is enqueued to the qdisc configured for that inter‐
93 face. Immediately afterwards, the kernel tries to get as many packets
94 as possible from the qdisc, for giving them to the network adaptor
95 driver.
96 A simple QDISC is the 'pfifo' one, which does no processing at all and
98 is a pure First In, First Out queue. It does however store traffic when
99 the network interface can't handle it momentarily.
100
103 Some qdiscs can contain classes, which contain further qdiscs - traffic
104 may then be enqueued in any of the inner qdiscs, which are within the
105 classes. When the kernel tries to dequeue a packet from such a class‐
106 ful qdisc it can come from any of the classes. A qdisc may for example
107 prioritize certain kinds of traffic by trying to dequeue from certain
108 classes before others.
109
112 A filter is used by a classful qdisc to determine in which class a
113 packet will be enqueued. Whenever traffic arrives at a class with sub‐
114 classes, it needs to be classified. Various methods may be employed to
115 do so, one of these are the filters. All filters attached to the class
116 are called, until one of them returns with a verdict. If no verdict was
117 made, other criteria may be available. This differs per qdisc.
118 It is important to notice that filters reside within qdiscs - they are
120 not masters of what happens.
121 The available filters are:
123 basic Filter packets based on an ematch expression. See tc-ematch(8)
125 for details.
126 bpf Filter packets using (e)BPF, see tc-bpf(8) for details.
128 cgroup Filter packets based on the control group of their process. See
130 tc-cgroup(8) for details.
131 flow, flower
133 Flow-based classifiers, filtering packets based on their flow
134 (identified by selectable keys). See tc-flow(8) and tc-flower(8)
135 for details.
136 fw Filter based on fwmark. Directly maps fwmark value to traffic
138 class. See tc-fw(8).
139 route Filter packets based on routing table. See tc-route(8) for
141 details.
142 rsvp Match Resource Reservation Protocol (RSVP) packets.
144 tcindex
146 Filter packets based on traffic control index. See tc-tcin‐
147 dex(8).
148 u32 Generic filtering on arbitrary packet data, assisted by syntax
150 to abstract common operations. See tc-u32(8) for details.
151 matchall
153 Traffic control filter that matches every packet. See tc-
154 matchall(8) for details.
155
158 The classless qdiscs are:
159 choke CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for
161 unresponsive flows) is a classless qdisc designed to both iden‐
162 tify and penalize flows that monopolize the queue. CHOKe is a
163 variation of RED, and the configuration is similar to RED.
164 codel CoDel (pronounced "coddle") is an adaptive "no-knobs" active
166 queue management algorithm (AQM) scheme that was developed to
167 address the shortcomings of RED and its variants.
168 [p|b]fifo
170 Simplest usable qdisc, pure First In, First Out behaviour. Lim‐
171 ited in packets or in bytes.
172 fq Fair Queue Scheduler realises TCP pacing and scales to millions
174 of concurrent flows per qdisc.
175 fq_codel
177 Fair Queuing Controlled Delay is queuing discipline that com‐
178 bines Fair Queuing with the CoDel AQM scheme. FQ_Codel uses a
179 stochastic model to classify incoming packets into different
180 flows and is used to provide a fair share of the bandwidth to
181 all the flows using the queue. Each such flow is managed by the
182 CoDel queuing discipline. Reordering within a flow is avoided
183 since Codel internally uses a FIFO queue.
184 gred Generalized Random Early Detection combines multiple RED queues
186 in order to achieve multiple drop priorities. This is required
187 to realize Assured Forwarding (RFC 2597).
188 hhf Heavy-Hitter Filter differentiates between small flows and the
190 opposite, heavy-hitters. The goal is to catch the heavy-hitters
191 and move them to a separate queue with less priority so that
192 bulk traffic does not affect the latency of critical traffic.
193 ingress
195 This is a special qdisc as it applies to incoming traffic on an
196 interface, allowing for it to be filtered and policed.
197 mqprio The Multiqueue Priority Qdisc is a simple queuing discipline
199 that allows mapping traffic flows to hardware queue ranges using
200 priorities and a configurable priority to traffic class mapping.
201 A traffic class in this context is a set of contiguous qdisc
202 classes which map 1:1 to a set of hardware exposed queues.
203 multiq Multiqueue is a qdisc optimized for devices with multiple Tx
205 queues. It has been added for hardware that wishes to avoid
206 head-of-line blocking. It will cycle though the bands and ver‐
207 ify that the hardware queue associated with the band is not
208 stopped prior to dequeuing a packet.
209 netem Network Emulator is an enhancement of the Linux traffic control
211 facilities that allow to add delay, packet loss, duplication and
212 more other characteristics to packets outgoing from a selected
213 network interface.
214 pfifo_fast
216 Standard qdisc for 'Advanced Router' enabled kernels. Consists
217 of a three-band queue which honors Type of Service flags, as
218 well as the priority that may be assigned to a packet.
219 pie Proportional Integral controller-Enhanced (PIE) is a control
221 theoretic active queue management scheme. It is based on the
222 proportional integral controller but aims to control delay.
223 red Random Early Detection simulates physical congestion by randomly
225 dropping packets when nearing configured bandwidth allocation.
226 Well suited to very large bandwidth applications.
227 rr Round-Robin qdisc with support for multiqueue network devices.
229 Removed from Linux since kernel version 2.6.27.
230 sfb Stochastic Fair Blue is a classless qdisc to manage congestion
232 based on packet loss and link utilization history while trying
233 to prevent non-responsive flows (i.e. flows that do not react to
234 congestion marking or dropped packets) from impacting perfor‐
235 mance of responsive flows. Unlike RED, where the marking proba‐
236 bility has to be configured, BLUE tries to determine the ideal
237 marking probability automatically.
238 sfq Stochastic Fairness Queueing reorders queued traffic so each
240 'session' gets to send a packet in turn.
241 tbf The Token Bucket Filter is suited for slowing traffic down to a
243 precisely configured rate. Scales well to large bandwidths.
244
246 In the absence of classful qdiscs, classless qdiscs can only be
247 attached at the root of a device. Full syntax:
248 tc qdisc add dev DEV root QDISC QDISC-PARAMETERS
250 To remove, issue
252 tc qdisc del dev DEV root
254 The pfifo_fast qdisc is the automatic default in the absence of a con‐
256 figured qdisc.
257
260 The classful qdiscs are:
261 ATM Map flows to virtual circuits of an underlying asynchronous
263 transfer mode device.
264 CBQ Class Based Queueing implements a rich linksharing hierarchy of
266 classes. It contains shaping elements as well as prioritizing
267 capabilities. Shaping is performed using link idle time calcula‐
268 tions based on average packet size and underlying link band‐
269 width. The latter may be ill-defined for some interfaces.
270 DRR The Deficit Round Robin Scheduler is a more flexible replacement
272 for Stochastic Fairness Queuing. Unlike SFQ, there are no built-
273 in queues -- you need to add classes and then set up filters to
274 classify packets accordingly. This can be useful e.g. for using
275 RED qdiscs with different settings for particular traffic. There
276 is no default class -- if a packet cannot be classified, it is
277 dropped.
278 DSMARK Classify packets based on TOS field, change TOS field of packets
280 based on classification.
281 HFSC Hierarchical Fair Service Curve guarantees precise bandwidth and
283 delay allocation for leaf classes and allocates excess bandwidth
284 fairly. Unlike HTB, it makes use of packet dropping to achieve
285 low delays which interactive sessions benefit from.
286 HTB The Hierarchy Token Bucket implements a rich linksharing hierar‐
288 chy of classes with an emphasis on conforming to existing prac‐
289 tices. HTB facilitates guaranteeing bandwidth to classes, while
290 also allowing specification of upper limits to inter-class shar‐
291 ing. It contains shaping elements, based on TBF and can priori‐
292 tize classes.
293 PRIO The PRIO qdisc is a non-shaping container for a configurable
295 number of classes which are dequeued in order. This allows for
296 easy prioritization of traffic, where lower classes are only
297 able to send if higher ones have no packets available. To facil‐
298 itate configuration, Type Of Service bits are honored by
299 default.
300 QFQ Quick Fair Queueing is an O(1) scheduler that provides near-
302 optimal guarantees, and is the first to achieve that goal with a
303 constant cost also with respect to the number of groups and the
304 packet length. The QFQ algorithm has no loops, and uses very
305 simple instructions and data structures that lend themselves
306 very well to a hardware implementation.
307
309 Classes form a tree, where each class has a single parent. A class may
310 have multiple children. Some qdiscs allow for runtime addition of
311 classes (CBQ, HTB) while others (PRIO) are created with a static number
312 of children.
313 Qdiscs which allow dynamic addition of classes can have zero or more
315 subclasses to which traffic may be enqueued.
316 Furthermore, each class contains a leaf qdisc which by default has
318 pfifo behaviour, although another qdisc can be attached in place. This
319 qdisc may again contain classes, but each class can have only one leaf
320 qdisc.
321 When a packet enters a classful qdisc it can be classified to one of
323 the classes within. Three criteria are available, although not all
324 qdiscs will use all three:
325 tc filters
327 If tc filters are attached to a class, they are consulted first
328 for relevant instructions. Filters can match on all fields of a
329 packet header, as well as on the firewall mark applied by
330 ipchains or iptables.
331 Type of Service
333 Some qdiscs have built in rules for classifying packets based on
334 the TOS field.
335 skb->priority
337 Userspace programs can encode a class-id in the 'skb->priority'
338 field using the SO_PRIORITY option.
339 Each node within the tree can have its own filters but higher level
341 filters may also point directly to lower classes.
342 If classification did not succeed, packets are enqueued to the leaf
344 qdisc attached to that class. Check qdisc specific manpages for
345 details, however.
346
349 All qdiscs, classes and filters have IDs, which can either be specified
350 or be automatically assigned.
351 IDs consist of a major number and a minor number, separated by a colon
353 - major:minor. Both major and minor are hexadecimal numbers and are
354 limited to 16 bits. There are two special values: root is signified by
355 major and minor of all ones, and unspecified is all zeros.
356 QDISCS A qdisc, which potentially can have children, gets assigned a
359 major number, called a 'handle', leaving the minor number names‐
360 pace available for classes. The handle is expressed as '10:'.
361 It is customary to explicitly assign a handle to qdiscs expected
362 to have children.
363 CLASSES
366 Classes residing under a qdisc share their qdisc major number,
367 but each have a separate minor number called a 'classid' that
368 has no relation to their parent classes, only to their parent
369 qdisc. The same naming custom as for qdiscs applies.
370 FILTERS
373 Filters have a three part ID, which is only needed when using a
374 hashed filter hierarchy.
375
378 The following parameters are widely used in TC. For other parameters,
379 see the man pages for individual qdiscs.
380 RATES Bandwidths or rates. These parameters accept a floating point
383 number, possibly followed by either a unit (both SI and IEC
384 units supported), or a float followed by a '%' character to
385 specify the rate as a percentage of the device's speed (e.g. 5%,
386 99.5%). Warning: specifying the rate as a percentage means a
387 fraction of the current speed; if the speed changes, the value
388 will not be recalculated.
389 bit or a bare number
391 Bits per second
392 kbit Kilobits per second
394 mbit Megabits per second
396 gbit Gigabits per second
398 tbit Terabits per second
400 bps Bytes per second
402 kbps Kilobytes per second
404 mbps Megabytes per second
406 gbps Gigabytes per second
408 tbps Terabytes per second
410 To specify in IEC units, replace the SI prefix (k-, m-, g-, t-)
413 with IEC prefix (ki-, mi-, gi- and ti-) respectively.
414 TC store rates as a 32-bit unsigned integer in bps internally,
417 so we can specify a max rate of 4294967295 bps.
418 TIMES Length of time. Can be specified as a floating point number fol‐
421 lowed by an optional unit:
422 s, sec or secs
424 Whole seconds
425 ms, msec or msecs
427 Milliseconds
428 us, usec, usecs or a bare number
430 Microseconds.
431 TC defined its own time unit (equal to microsecond) and stores
434 time values as 32-bit unsigned integer, thus we can specify a
435 max time value of 4294967295 usecs.
436 SIZES Amounts of data. Can be specified as a floating point number
439 followed by an optional unit:
440 b or a bare number
442 Bytes.
443 kbit Kilobits
445 kb or k
447 Kilobytes
448 mbit Megabits
450 mb or m
452 Megabytes
453 gbit Gigabits
455 gb or g
457 Gigabytes
458 TC stores sizes internally as 32-bit unsigned integer in byte,
461 so we can specify a max size of 4294967295 bytes.
462 VALUES Other values without a unit. These parameters are interpreted
465 as decimal by default, but you can indicate TC to interpret them
466 as octal and hexadecimal by adding a '0' or '0x' prefix respec‐
467 tively.
468
471 The following commands are available for qdiscs, classes and filter:
472 add Add a qdisc, class or filter to a node. For all entities, a par‐
474 ent must be passed, either by passing its ID or by attaching
475 directly to the root of a device. When creating a qdisc or a
476 filter, it can be named with the handle parameter. A class is
477 named with the classid parameter.
478 delete A qdisc can be deleted by specifying its handle, which may also
481 be 'root'. All subclasses and their leaf qdiscs are automati‐
482 cally deleted, as well as any filters attached to them.
483 change Some entities can be modified 'in place'. Shares the syntax of
486 'add', with the exception that the handle cannot be changed and
487 neither can the parent. In other words, change cannot move a
488 node.
489 replace
492 Performs a nearly atomic remove/add on an existing node id. If
493 the node does not exist yet it is created.
494 get Displays a single filter given the interface DEV, qdisc-id, pri‐
497 ority, protocol and filter-id.
498 show Displays all filters attached to the given interface. A valid
501 parent ID must be passed.
502 link Only available for qdiscs and performs a replace where the node
505 must exist already.
506
509 The tc utility can monitor events generated by the kernel such as
510 adding/deleting qdiscs, filters or actions, or modifying existing ones.
511 The following command is available for monitor :
513 file If the file option is given, the tc does not listen to kernel
515 events, but opens the given file and dumps its contents. The
516 file has to be in binary format and contain netlink messages.
517
520 -b, -b filename, -batch, -batch filename
521 read commands from provided file or standard input and invoke
522 them. First failure will cause termination of tc.
523 -force don't terminate tc on errors in batch mode. If there were any
526 errors during execution of the commands, the application return
527 code will be non zero.
528 -o, -oneline
531 output each record on a single line, replacing line feeds with
532 the '\' character. This is convenient when you want to count
533 records with wc(1) or to grep(1) the output.
534 -n, -net, -netns <NETNS>
537 switches tc to the specified network namespace NETNS. Actually
538 it just simplifies executing of:
539 ip netns exec NETNS tc [ OPTIONS ] OBJECT { COMMAND | help }
541 to
543 tc -n[etns] NETNS [ OPTIONS ] OBJECT { COMMAND | help }
545 -N, -Numeric
548 Print the number of protocol, scope, dsfield, etc directly
549 instead of converting it to human readable name.
550 -cf, -conf <FILENAME>
553 specifies path to the config file. This option is used in con‐
554 junction with other options (e.g. -nm).
555 -t, -timestamp
558 When tc monitor runs, print timestamp before the event message
559 in format:
560 Timestamp: <Day> <Month> <DD> <hh:mm:ss> <YYYY> <usecs> usec
561 -ts, -tshort
564 When tc monitor runs, prints short timestamp before the event
565 message in format:
566 [<YYYY>-<MM>-<DD>T<hh:mm:ss>.<ms>]
567
570 The show command has additional formatting options:
571 -s, -stats, -statistics
574 output more statistics about packet usage.
575 -d, -details
578 output more detailed information about rates and cell sizes.
579 -r, -raw
582 output raw hex values for handles.
583 -p, -pretty
586 for u32 filter, decode offset and mask values to equivalent fil‐
587 ter commands based on TCP/IP. In JSON output, add whitespace to
588 improve readability.
589 -iec print rates in IEC units (ie. 1K = 1024).
592 -g, -graph
595 shows classes as ASCII graph. Prints generic stats info under
596 each class if -s option was specified. Classes can be filtered
597 only by dev option.
598 -c[color][={always|auto|never}
601 Configure color output. If parameter is omitted or always, color
602 output is enabled regardless of stdout state. If parameter is
603 auto, stdout is checked to be a terminal before enabling color
604 output. If parameter is never, color output is disabled. If
605 specified multiple times, the last one takes precedence. This
606 flag is ignored if -json is also given.
607 -j, -json
610 Display results in JSON format.
611 -nm, -name
614 resolve class name from /etc/iproute2/tc_cls file or from file
615 specified by -cf option. This file is just a mapping of classid
616 to class name:
617 # Here is comment
619 1:40 voip # Here is another comment
620 1:50 web
621 1:60 ftp
622 1:2 home
623 tc will not fail if -nm was specified without -cf option but
625 /etc/iproute2/tc_cls file does not exist, which makes it possi‐
626 ble to pass -nm option for creating tc alias.
627
630 tc -g class show dev eth0
631 Shows classes as ASCII graph on eth0 interface.
632 tc -g -s class show dev eth0
634 Shows classes as ASCII graph with stats info under each class.
635
638 tc was written by Alexey N. Kuznetsov and added in Linux 2.2.
639
641 tc-basic(8), tc-bfifo(8), tc-bpf(8), tc-cake(8), tc-cbq(8), tc-
642 cgroup(8), tc-choke(8), tc-codel(8), tc-drr(8), tc-ematch(8), tc-
643 flow(8), tc-flower(8), tc-fq(8), tc-fq_codel(8), tc-fw(8), tc-hfsc(7),
644 tc-hfsc(8), tc-htb(8), tc-mqprio(8), tc-pfifo(8), tc-pfifo_fast(8), tc-
645 pie(8), tc-red(8), tc-route(8), tc-sfb(8), tc-sfq(8), tc-stab(8), tc-
646 tbf(8), tc-tcindex(8), tc-u32(8),
647 User documentation at http://lartc.org/, but please direct bugreports
648 and patches to: <netdev@vger.kernel.org>
649
652 Manpage maintained by bert hubert (ahu@ds9a.nl)
653iproute2 16 December 2001 TC(8)