On Wed, 2003-04-16 at 13:27, Linux RedHat wrote: > > On Wed, 2003-04-16 at 12:13, Linux RedHat wrote: > >> Hi all, > >> > >> I am wondering if I can use traffic control to balance the traffic coming into > >> my computer in the following way... > >> > >> +----------+ +----------+ +----------+ +----------+ > >> | Internet | | Internet | | Internet | | Internet | > >> | Server 1 | | Server 2 | | Server 3 | ... | Server n | > >> +----------+ +----------+ +----------+ +----------+ > >> Port ????? Port ????? Port ????? Port ????? > >> | | | | > >> +--------------+------+-------+-----------------+ > >> | > >> | eth1 > >> +----------+ > >> | Linux | > >> | Router | > >> +----------+ > >> | eth0 > >> | > >> +----------+ > >> | PC on | > >> | Lan | > >> +----------+ > >> > >> Multiple servers sending data to one pc on the lan using multiple ports. I want > >> to load balance the traffic leaving eth0 towards the PC based on the number of > >> connections. > >> > >> I am wondering if I will be able to set up classes/filters for a set number of > >> connections ( for arguments sake - say 10 connections ) regardless of what port > >> they are on and thus configure my tc setup to load balance those 10 connections > >> - allowing them to borrow from each other as usual. > >> > >> All I am looking for here is a short yes/no answer and possible a hint as to > >> what part of the howto to start looking at. I want to figure out how to do it > >> myself, but don't want to waste time trying if it is not possible. > >> > >> Thanks for your time spent reading/answering. > >> > >> Leigh > > > > hi leigh, > > > > checkout the linux virtual server page as well -- this is the exact schema > > discussed in the howto. (http://www.linuxvirtualserver.org/) > > > > cheers > > > > christopher cuse > > Thanks for that Christopher... > > Unfortunately that's not quite what I am after : > > The servers referred to in my diagram are not "known" servers. They could be any > servers e.g. public ftps, web sites etc. The virtualserver pages seem to be designed > for balancing the load on "known" servers while I want to balance the traffic *to* > the client which has connected to several previously undefined servers and is > downloading from each of them. > > The purpose here is to prevent one server from sending so fast that it "hogs" the > bandwidth available on my internet connection. > > An example would be that I am downloading via ftp from two servers where the first > server can send data at 200KB/s and the second server can only send at 50KB/s. My > internet connection can only handle 120KB/s so I want each connection to be > guaranteed 60KB/s but allow borrowing of bandwidth where available. (In reality I > will be implementing traffic control to allow for prioritising interactive traffic > and also balancing the load between multiple client PC's on the lan but for the > purposes of this discussion I am ignoring that part of the structure) > > This is possibly a similar problem to sharing the bandwidth available to multiple > client pcs but on top of that sharing the bandwidth available to each pc among that > pc's active connections. > > Hope that clarifies my project a bit... I have just realised that I used the term > "load balancing" in my original post and may have misled you. I meant "bandwidth > sharing". > > Leigh hi leigh, well, i would probably start with a script (see attached cbq-init) which should help you get your feet wet and serve as a knowledge foundation upon which you will build your final tc hierarchy. cheers christopher cuse
#!/bin/bash # # cbq.init v0.7.2 # Copyright (C) 1999 Pavel Golubev <pg@xxxxxxxxxxxxx> # # chkconfig: 2345 11 89 # description: sets up CBQ-based traffic control # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation; either version 2 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA # # To get the latest version, check on Freshmeat for actual location: # # http://freshmeat.net/projects/cbq.init # # # VERSION HISTORY # --------------- # v0.7.2-Paulo Sedrez # - fix time2abs to allow hours with leading zero in TIME rules # - Svetlin Simeonov <zvero@xxxxxxxxx> # - fix cbq_device_list to allow VLAN interfaces # - Mark Davis <mark.davis@xxxxxx> # - ignore *~ backup files when looking for classes # - Mike Boyer <boyer@xxxxxxxxxxxxxxxxxx> # - fix to allow arguments to be passed to "restart" command # v0.7.1- Lubomir Bulej <pallas@xxxxxxxx> # - default value for PERTURB # - fixed small bug in RULE parser to correctly parse rules with # identical source and destination fields # - faster initial scanning of DEVICE fields # v0.7 - Lubomir Bulej <pallas@xxxxxxxx> # - lots of various cleanups and reorganizations; the parsing is now # some 40% faster, but the class ID must be in range 0x0002-0xffff # (again). Because of the number of internal changes and the above # class ID restriction, I bumped the version to 0.7 to indicate # something might have got broken :) # - changed PRIO_{U32,FW,ROUTE} to PRIO_{RULE,MARK,REALM} # for consistency with filter keywords # - exposed "compile" command # - Catalin Petrescu <taz@xxxxxxxx> # - support for port masks in RULE (u32) filter # - Jordan Vrtanoski <obeliks@xxxxxxxxx> # - support for week days in TIME rules # v0.6.4- Lubomir Bulej <pallas@xxxxxxxx> # - added PRIO_* variables to allow easy control of filter priorities # - added caching to speed up CBQ start, the cache is invalidated # whenever any of the configuration files changes # - updated the readme section + some cosmetic fixes # v0.6.3- Lubomir Bulej <pallas@xxxxxxxx> # - removed setup of (unnecessary) class 1:1 - all classes # now use qdisc's default class 1:0 as their parent # - minor fix in the timecheck branch - classes # without leaf qdisc were not updated # - minor fix to avoid timecheck failure when run # at time with minutes equal to 08 or 09 # - respect CBQ_PATH setting in environment # - made PRIO=5 default, rendering it optional in configs # - added support for route filter, see notes about REALM keyword # - added support for fw filter, see notes about MARK keyword # - added filter display to "list" and "stats" commands # - readme section update + various cosmetic fixes # v0.6.2- Catalin Petrescu <taz@xxxxxxxx> # - added tunnels interface handling # v0.6.1- Pavel Golubev <pg@xxxxxxxxxxxxx> # - added sch_prio module loading # (thanks johan@xxxxxxxxxxxxxxxxxx for reminding) # - resolved errors resulting from stricter syntax checking in bash2 # - Lubomir Bulej <pallas@xxxxxxxx> # - various cosmetic fixes # v0.6 - Lubomir Bulej <pallas@xxxxxxxx> # - attempt to limit number of spawned processes by utilizing # more of sed power (use sed instead of grep+cut) # - simplified TIME parser, using bash builtins # - added initial support for SFQ as leaf qdisc # - reworked the documentation part a little # - incorporated pending patches and ideas submitted by # following people for versions 0.3 into version 0.6 # - Miguel Freitas <miguel@xxxxxxxxxxxxxxxx> # - in case of overlapping TIME parameters, the last match is taken # - Juanjo Ciarlante <jjo@xxxxxxxxxxxxxx> # - chkconfig tags, list + stats startup parameters # - optional tc & ip command logging (into /var/run/cbq-*) # - Rafal Maszkowski <rzm@xxxxxxxxxx> # - PEAK parameter for setting TBF's burst peak rate # - fix for many config files (use find instead of ls) # v0.5.1- Lubomir Bulej <pallas@xxxxxxxx> # - fixed little but serious bug in RULE parser # v0.5 - Lubomir Bulej <pallas@xxxxxxxx> # - added options PARENT, LEAF, ISOLATED and BOUNDED. This allows # (with some attention to config file ordering) for creating # hierarchical structures of shapers with classes able (or unable) # to borrow bandwidth from their parents. # - class ID check allows hexadecimal numbers # - rewritten & simplified RULE parser # - cosmetic changes to improve readability # - reorganization to avoid duplicate code (timecheck etc.) # - timecheck doesn't check classes without TIME fields anymore # v0.4 - Lubomir Bulej <pallas@xxxxxxxx> # - small bugfix in RULE parsing code # - simplified configuration parsing code # - several small cosmetic changes # - TIME parameter can be now specified more than once allowing you to # differentiate RATE throughout the whole day. Time overlapping is # not checked, first match is taken. Midnight wrap (eg. 20:00-6:00) # is allowed and taken care of. # v0.3a4- fixed small bug in IF operator. Thanks to # Rafal Maszkowski <rzm@xxxxxxxxxx> # v0.3a3- fixed grep bug when using more than 10 eth devices. Thanks to David # Trcka <trcka@xxxxxxx>. # v0.3a2- fixed bug in "if" operator. Thanks kad@xxxxxxxxxxxxxxxx # v0.3a - added TIME parameter. Example: TIME=00:00-19:00;64Kbit/6Kbit # So, between 00:00 and 19:00 the RATE will be 64Kbit. # Just start "cbq.init timecheck" periodically from cron # (every 10 minutes for example). DON'T FORGET though, to run # "cbq.init start" for CBQ to initialize. # v0.2 - Some cosmetic changes. Now it is more compatible with old bash # version. Thanks to Stanislav V. Voronyi <stas@xxxxxxxxxxxxxxxxxxxxx>. # v0.1 - First public release # # # README # ------ # # First of all - this is just a SIMPLE EXAMPLE of CBQ power. # Don't ask me "why" and "how" :) # # This script is meant to simplify setup and management of relatively simple # CBQ-based traffic control on Linux. Access to advanced networking features # of Linux kernel is provided by "ip" and "tc" utilities from A. Kuznetsov's # iproute2 package, available at ftp://ftp.inr.ac.ru/ip-routing. Because the # utilities serve primarily to translate user wishes to RTNETLINK commands, # their interface is rather spartan, intolerant and requires quite a lot of # typing. And typing is what this script attempts to reduce :) # # The advanced networking stuff in Linux is pretty flexible and this script # aims to bring some of its features to the not-so-hard-core Linux users. Of # course, there is a tradeoff between simplicity and flexibility and you may # realize that the flexibility suffered too much for your needs -- time to # face "ip" and "tc" interface. # # To speed up the "start" command, simple caching was introduced in version # 0.6.4. The caching works so that the sequence of "tc" commands for given # configuration is stored in a file (/var/cache/cbq.init by default) which # is used next time the "start" command is run to avoid repeated parsing of # configuration files. This cache is invalidated whenever any of the CBQ # configuration files changes. If you want to run "cbq.init start" without # caching, run it as "cbq.init start nocache". If you want to force cache # invalidation, run it as "cbq.init start invalidate". Caching is disabled # if you have logging enabled (ie. CBQ_DEBUG is not empty). # # If you only want cqb.init to translate your configuration to "tc" commands, # use "compile" command which will output "tc" commands required to build # your configuration. Bear in mind that "compile" does not check if the "tc" # commands were successful - this is done (in certain places) only when the # "start nocache" command is used, which is also useful when creating the # configuration to check whether it is completely valid. # # All CBQ parameters are valid for Ethernet interfaces only, The script was # tested on various Linux kernel versions from series 2.1 to 2.4 and several # distributions with KSI Linux (Nostromo version) as the premier one. # # # HOW DOES IT WORK? # ----------------- # # Every traffic class must be described by a file in the $CBQ_PATH directory # (/etc/sysconfig/cbq by default) - one file per class. # # The config file names must obey mandatory format: cbq-<clsid>.<name> where # <clsid> is two-byte hexadecimal number in range <0002-FFFF> (which in fact # is a CBQ class ID) and <name> is the name of the class -- anything to help # you distinguish the configuration files. For small amount of classes it is # often possible (and convenient) to let <clsid> resemble bandwidth of the # class. # # Example of valid config name: # cbq-1280.My_first_shaper # # # The configuration file may contain the following parameters: # ### Device parameters # # DEVICE=<ifname>,<bandwidth>[,<weight>] mandatory # DEVICE=eth0,10Mbit,1Mbit # # <ifname> is the name of the interface you want to control # traffic on, e.g. eth0 # <bandwidth> is the physical bandwidth of the device, e.g. for # ethernet 10Mbit or 100Mbit, for arcnet 2Mbit # <weight> is tuning parameter that should be proportional to # <bandwidth>. As a rule of thumb: <weight> = <bandwidth> / 10 # # When you have more classes on one interface, it is enough to specify # <bandwidth> [and <weight>] only once, therefore in other files you only # need to set DEVICE=<ifname>. # ### Class parameters # # RATE=<speed> mandatory # RATE=5Mbit # # Bandwidth allocated to the class. Traffic going through the class is # shaped to conform to specified rate. You can use Kbit, Mbit or bps, # Kbps and Mbps as suffices. If you don't specify any unit, bits/sec # are used. Also note that "bps" means "bytes per second", not bits. # # WEIGHT=<speed> mandatory # WEIGHT=500Kbit # # Tuning parameter that should be proportional to RATE. As a rule # of thumb, use WEIGHT ~= RATE / 10. # # PRIO=<1-8> optional, default 5 # PRIO=5 # # Priority of class traffic. The higher the number, the lesser # the priority. Priority of 5 is just fine. # # PARENT=<clsid> optional, default not set # PARENT=1280 # # Specifies ID of the parent class to which you want this class be # attached. You might want to use LEAF=none for the parent class as # mentioned below. By using this parameter and carefully ordering the # configuration files, it is possible to create simple hierarchical # structures of CBQ classes. The ordering is important so that parent # classes are constructed prior to their children. # # LEAF=none|tbf|sfq optional, default "tbf" # # Tells the script to attach specified leaf queueing discipline to CBQ # class. By default, TBF is used. Note that attaching TBF to CBQ class # shapes the traffic to conform to TBF parameters and prevents the class # from borrowing bandwidth from its parent even if you have BOUNDED set # to "no". To allow the class to borrow bandwith (provided it is not # bounded), you must set LEAF to "none" or "sfq". # # If you want to ensure (approximately) fair sharing of bandwidth among # several hosts in the same class, you might want to specify LEAF=sfq to # attach SFQ as leaf queueing discipline to that class. # # BOUNDED=yes|no optional, default "yes" # # If set to "yes", the class is not allowed to borrow bandwidth from # its parent class in overlimit situation. If set to "no", the class # will be allowed to borrow bandwidth from its parent. # # Note: Don't forget to set LEAF to "none" or "sfq", otherwise the class will # have TBF attached to itself and will not be able to borrow unused # bandwith from its parent. # # ISOLATED=yes|no optional, default "no" # # If set to "yes", the class will not lend unused bandwidth to # its children. # ### TBF qdisc parameters # # BUFFER=<bytes>[/<bytes>] optional, default "10Kb/8" # # This parameter controls the depth of the token bucket. In other # words it represents the maximal burst size the class can send. # The optional part of parameter is used to determine the length # of intervals in packet sizes, for which the transmission times # are kept. # # LIMIT=<bytes> optional, default "15Kb" # # This parameter determines the maximal length of backlog. If # the queue contains more data than specified by LIMIT, the # newly arriving packets are dropped. The length of backlog # determines queue latency in case of congestion. # # PEAK=<speed> optional, default not set # # Maximal peak rate for short-term burst traffic. This allows you # to control the absolute peak rate the class can send at, because # single TBF that allows 256Kbit/s would of course allow rate of # 512Kbit for half a second or 1Mbit for a quarter of second. # # MTU=<bytes> optional, default "1500" # # Maximum number of bytes that can be sent at once over the # physical medium. This parameter is required when you specify # PEAK parameter. It defaults to MTU of ethernet - for other # media types you might want to change it. # # Note: Setting TBF as leaf qdisc will effectively prevent the class from # borrowing bandwidth from the ancestor class, because even if the # class allows more traffic to pass through, it is then shaped to # conform to TBF. # ### SFQ qdisc parameters # # The SFQ queueing discipline is a cheap way for sharing class bandwidth # among several hosts. As it is stochastic, the fairness is approximate but # it will do the job in most cases. If you want real fairness, you should # probably use WRR (weighted round robin) or WFQ queueing disciplines. Note # that SFQ does not do any traffic shaping - the shaping is done by the CBQ # class the SFQ is attached to. # # QUANTUM=<bytes> optional, default not set # # This parameter should not be set lower than link MTU, for ethernet # it is 1500b, or (with MAC header) 1514b which is the value used # in Alexey Kuznetsov's examples. # # PERTURB=<seconds> optional, default "10" # # Period of hash function perturbation. If unset, hash reconfiguration # will never take place which is what you probably don't want. The # default value of 10 seconds is probably a good one. # ### Filter parameters # # RULE=[[saddr[/prefix]][:port[/mask]],][daddr[/prefix]][:port[/mask]] # # These parameters make up "u32" filter rules that select traffic for # each of the classes. You can use multiple RULE fields per config. # # The optional port mask should only be used by advanced users who # understand how the u32 filter works. # # Some examples: # # RULE=10.1.1.0/24:80 # selects traffic going to port 80 in network 10.1.1.0 # # RULE=10.2.2.5 # selects traffic going to any port on single host 10.2.2.5 # # RULE=10.2.2.5:20/0xfffe # selects traffic going to ports 20 and 21 on host 10.2.2.5 # # RULE=:25,10.2.2.128/26:5000 # selects traffic going from anywhere on port 50 to # port 5000 in network 10.2.2.128 # # RULE=10.5.5.5:80, # selects traffic going from port 80 of single host 10.5.5.5 # # # # REALM=[srealm,][drealm] # # These parameters make up "route" filter rules that classify traffic # according to packet source/destination realms. For information about # realms, see Alexey Kuznetsov's IP Command Reference. This script # does not define any realms, it justs builds "tc filter" commands # for you if you need to classify traffic this way. # # Realm is either a decimal number or a string referencing entry in # /etc/iproute2/rt_realms (usually). # # Some examples: # # REALM=russia,internet # selects traffic going from realm "russia" to realm "internet" # # REALM=freenet, # selects traffic going from realm "freenet" # # REALM=10 # selects traffic going to realm 10 # # # # MARK=<mark> # # These parameters make up "fw" filter rules that select traffic for # each of the classes accoring to firewall "mark". Mark is a decimal # number packets are tagged with if firewall rules say so. You can # use multiple MARK fields per config. # # # Note: Rules for different filter types can be combined. Attention must be # paid to the priority of filter rules, which can be set below through # the PRIO_{RULE,MARK,REALM} variables. # ### Time ranging parameters # # TIME=[<dow>,<dow>, ...,<dow>/]<from>-<till>;<rate>/<weight>[/<peak>] # TIME=0,1,2,5/18:00-06:00;256Kbit/25Kbit # TIME=60123/18:00-06:00;256Kbit/25Kbit # TIME=18:00-06:00;256Kbit/25Kbit # # This parameter allows you to differentiate the class bandwidth # throughout the day. You can specify multiple TIME parameters, if # the times overlap, last match is taken. The fields <rate>, <weight> # and <peak> correspond to parameters RATE, WEIGHT and PEAK (which # is optional and applies to TBF leaf qdisc only). # # You can also specify days of week when the TIME rule applies. <dow> # is numeric, 0 corresponds to sunday, 1 corresponds to monday, etc. # ### # # Sample configuration file: cbq-1280.My_first_shaper # # -------------------------------------------------------------------------- # DEVICE=eth0,10Mbit,1Mbit # RATE=128Kbit # WEIGHT=10Kbit # PRIO=5 # RULE=192.128.1.0/24 # -------------------------------------------------------------------------- # # The configuration says that we will control traffic on 10Mbit ethernet # device eth0 and the traffic going to network 192.168.1.0 will be # processed with priority 5 and shaped to rate of 128Kbit. # # Note that you can control outgoing traffic only. If you want to control # traffic in both directions, you must set up CBQ for both interfaces. # # Consider the following example: # # +---------+ 192.168.1.1 # BACKBONE -----eth0-| linux |-eth1------*-[client] # +---------+ # # Imagine you want to shape traffic from backbone to the client to 28Kbit # and traffic in the opposite direction to 128Kbit. You need to setup CBQ # on both eth0 and eth1 interfaces, thus you need two config files: # # cbq-028.backbone-client # -------------------------------------------------------------------------- # DEVICE=eth1,10Mbit,1Mbit # RATE=28Kbit # WEIGHT=2Kbit # PRIO=5 # RULE=192.168.1.1 # -------------------------------------------------------------------------- # # cbq-128.client-backbone # -------------------------------------------------------------------------- # DEVICE=eth0,10Mbit,1Mbit # RATE=128Kbit # WEIGHT=10Kbit # PRIO=5 # RULE=192.168.1.1, # -------------------------------------------------------------------------- # # Pay attention to comma "," in the RULE field - it denotes source address! # # Enjoy. # ############################################################################# export LANG="C" ### Default filter priorities (must be different) PRIO_RULE=${PRIO_RULE:-100} PRIO_MARK=${PRIO_MARK:-200} PRIO_REALM=${PRIO_REALM:-300} ### Default CBQ_PATH & CBQ_CACHE settings CBQ_PATH=${CBQ_PATH:-/etc/sysconfig/cbq} CBQ_CACHE=${CBQ_CACHE:-/var/cache/cbq.init} ### Uncomment to enable logfile for debugging #CBQ_DEBUG="/var/run/cbq-$1" ### Modules to probe for. Uncomment the last CBQ_PROBE ### line if you have QoS support compiled into kernel CBQ_PROBE="sch_cbq sch_tbf sch_sfq sch_prio" CBQ_PROBE="$CBQ_PROBE cls_fw cls_u32 cls_route" #CBQ_PROBE="" ### Keywords required for qdisc & class configuration CBQ_WORDS="DEVICE|RATE|WEIGHT|PRIO|PARENT|LEAF|BOUNDED|ISOLATED" CBQ_WORDS="$CBQ_WORDS|BUFFER|LIMIT|PEAK|MTU|QUANTUM|PERTURB" ############################################################################# ############################# SUPPORT FUNCTIONS ############################# ############################################################################# ### Get list of network devices cbq_device_list () { ip link show| sed -n "/^[0-9]/ \ { s/^[0-9]\+: \([a-z0-9.]\+\)[:@].*/\1/; p; }" } # cbq_device_list ### Remove root class from device $1 cbq_device_off () { tc qdisc del dev $1 root 2> /dev/null } # cbq_device_off ### Remove CBQ from all devices cbq_off () { for dev in `cbq_device_list`; do cbq_device_off $dev done } # cbq_off ### Prefixed message cbq_message () { echo -e "**CBQ: $@" } # cbq_message ### Failure message cbq_failure () { cbq_message "$@" exit 1 } # cbq_failure ### Failure w/ cbq-off cbq_fail_off () { cbq_message "$@" cbq_off exit 1 } # cbq_fail_off ### Convert time to absolute value cbq_time2abs () { local min=${1##*:}; min=${min##0} local hrs=${1%%:*}; hrs=${hrs##0} echo $[hrs*60 + min] } # cbq_time2abs ### Display CBQ setup cbq_show () { for dev in `cbq_device_list`; do [ `tc qdisc show dev $dev| wc -l` -eq 0 ] && continue echo -e "### $dev: queueing disciplines\n" tc $1 qdisc show dev $dev; echo [ `tc class show dev $dev| wc -l` -eq 0 ] && continue echo -e "### $dev: traffic classes\n" tc $1 class show dev $dev; echo [ `tc filter show dev $dev| wc -l` -eq 0 ] && continue echo -e "### $dev: filtering rules\n" tc $1 filter show dev $dev; echo done } # cbq_show ### Check configuration and load DEVICES, DEVFIELDS and CLASSLIST from $1 cbq_init () { ### Get a list of configured classes CLASSLIST=`find $1 -type f -name 'cbq-*' -not -name '*~' \ -maxdepth 1 -printf "%f\n"| sort` [ -z "$CLASSLIST" ] && cbq_failure "no configuration files found in $1!" ### Gather all DEVICE fields from $1/cbq-* DEVFIELDS=`find $1 -type f -name 'cbq-*' -not -name '*~' \ -maxdepth 1| xargs sed -n 's/#.*//; s/[[:space:]]//g; \ /^DEVICE=[^,]*,[^,]*\(,[^,]*\)\?/ { s/.*=//; p; }'|\ sort -u` [ -z "$DEVFIELDS" ] && cbq_failure "no DEVICE field found in $1/cbq-*!" ### Check for different DEVICE fields for the same device DEVICES=`echo "$DEVFIELDS"| sed 's/,.*//'| sort -u` [ `echo "$DEVICES"| wc -l` -ne `echo "$DEVFIELDS"| wc -l` ] && cbq_failure "different DEVICE fields for single device!\n$DEVFIELDS" } # cbq_init ### Load class configuration from $1/$2 cbq_load_class () { CLASS=`echo $2| sed 's/^cbq-0*//; s/^\([0-9a-fA-F]\+\).*/\1/'` CFILE=`sed -n 's/#.*//; s/[[:space:]]//g; /^[[:alnum:]]\+=[[:alnum:].,:;/*-]\+$/ p' $1/$2` ### Check class number IDVAL=`/usr/bin/printf "%d" 0x$CLASS 2> /dev/null` [ $? -ne 0 -o $IDVAL -lt 2 -o $IDVAL -gt 65535 ] && cbq_fail_off "class ID of $2 must be in range <0002-FFFF>!" ### Set defaults & load class RATE=""; WEIGHT=""; PARENT=""; PRIO=5 LEAF=tbf; BOUNDED=yes; ISOLATED=no BUFFER=10Kb/8; LIMIT=15Kb; MTU=1500 PEAK=""; PERTURB=10; QUANTUM="" eval `echo "$CFILE"| grep -E "^($CBQ_WORDS)="` ### Require RATE/WEIGHT [ -z "$RATE" -o -z "$WEIGHT" ] && cbq_fail_off "missing RATE or WEIGHT in $2!" ### Class device DEVICE=${DEVICE%%,*} [ -z "$DEVICE" ] && cbq_fail_off "missing DEVICE field in $2!" BANDWIDTH=`echo "$DEVFIELDS"| sed -n "/^$DEVICE,/ \ { s/[^,]*,\([^,]*\).*/\1/; p; q; }"` ### Convert to "tc" options PEAK=${PEAK:+peakrate $PEAK} PERTURB=${PERTURB:+perturb $PERTURB} QUANTUM=${QUANTUM:+quantum $QUANTUM} [ "$BOUNDED" = "no" ] && BOUNDED="" || BOUNDED="bounded" [ "$ISOLATED" = "yes" ] && ISOLATED="isolated" || ISOLATED="" } # cbq_load_class ############################################################################# #################################### INIT ################################### ############################################################################# ### Check for presence of ip-route2 in usual place [ -x /sbin/tc -a -x /sbin/ip ] || cbq_failure "ip-route2 utilities not installed or executable!" ### Various tweaks if [ "$1" = "compile" ]; then ### no module probing CBQ_PROBE="" ### echo-only version of "tc" command tc () { echo "tc $@" } # tc elif [ -n "$CBQ_DEBUG" ]; then echo -e "# `date`" > $CBQ_DEBUG ### Logging version of "ip" command ip () { echo -e "\n# ip $@" >> $CBQ_DEBUG /sbin/ip "$@" 2>&1 | tee -a $CBQ_DEBUG } # ip ### Logging version of "tc" command tc () { echo -e "\n# tc $@" >> $CBQ_DEBUG /sbin/tc "$@" 2>&1 | tee -a $CBQ_DEBUG } # tc fi # command logging case "$1" in ############################################################################# ############################### START/COMPILE ############################### ############################################################################# start|compile) ### Probe QoS modules (start only) for module in $CBQ_PROBE; do modprobe $module || cbq_failure "failed to load module $module" done ### If we are in compile/nocache/logging mode, don't bother with cache if [ "$1" != "compile" -a "$2" != "nocache" -a -z "$CBQ_DEBUG" ]; then VALID=1 ### validate the cache [ "$2" = "invalidate" -o ! -f $CBQ_CACHE ] && VALID=0 if [ $VALID -eq 1 ]; then [ `find $CBQ_PATH -maxdepth 1 -newer $CBQ_CACHE| \ wc -l` -gt 0 ] && VALID=0 fi ### compile the config if the cache is invalid if [ $VALID -ne 1 ]; then $0 compile > $CBQ_CACHE || cbq_fail_off "failed to compile CBQ configuration!" fi ### run the cached commands exec /bin/sh $CBQ_CACHE 2> /dev/null fi ### Load DEVICES, DEVFIELDS and CLASSLIST cbq_init $CBQ_PATH ### Setup root qdisc on all configured devices for dev in $DEVICES; do ### Retrieve device bandwidth and, optionally, weight DEVTEMP=`echo "$DEVFIELDS"| sed -n "/^$dev,/ { s/$dev,//; p; q; }"` DEVBWDT=${DEVTEMP%%,*}; DEVWGHT=${DEVTEMP##*,} [ "$DEVBWDT" = "$DEVWGHT" ] && DEVWGHT="" ### Device bandwidth is required if [ -z "$DEVBWDT" ]; then cbq_message "could not determine bandwidth for device $dev!" cbq_failure "please set up the DEVICE fields properly!" fi ### Check if the device is there ip link show $dev &> /dev/null || cbq_fail_off "device $dev not found!" ### Remove old root qdisc from device cbq_device_off $dev ### Setup root qdisc + class for device tc qdisc add dev $dev root handle 1 cbq \ bandwidth $DEVBWDT avpkt 1000 cell 8 ### Set weight of the root class if set [ -n "$DEVWGHT" ] && tc class change dev $dev root cbq weight $DEVWGHT allot 1514 [ "$1" = "compile" ] && echo done # dev ### Setup traffic classes for classfile in $CLASSLIST; do cbq_load_class $CBQ_PATH $classfile ### Create the class tc class add dev $DEVICE parent 1:$PARENT classid 1:$CLASS cbq \ bandwidth $BANDWIDTH rate $RATE weight $WEIGHT prio $PRIO \ allot 1514 cell 8 maxburst 20 avpkt 1000 $BOUNDED $ISOLATED || cbq_fail_off "failed to add class $CLASS with parent $PARENT on $DEVICE!" ### Create leaf qdisc if set if [ "$LEAF" = "tbf" ]; then tc qdisc add dev $DEVICE parent 1:$CLASS handle $CLASS tbf \ rate $RATE buffer $BUFFER limit $LIMIT mtu $MTU $PEAK elif [ "$LEAF" = "sfq" ]; then tc qdisc add dev $DEVICE parent 1:$CLASS handle $CLASS sfq \ $PERTURB $QUANTUM fi ### Create fw filter for MARK fields for mark in `echo "$CFILE"| sed -n '/^MARK/ { s/.*=//; p; }'`; do ### Attach fw filter to root class tc filter add dev $DEVICE parent 1:0 protocol ip \ prio $PRIO_MARK handle $mark fw classid 1:$CLASS done ### mark ### Create route filter for REALM fields for realm in `echo "$CFILE"| sed -n '/^REALM/ { s/.*=//; p; }'`; do ### Split realm into source & destination realms SREALM=${realm%%,*}; DREALM=${realm##*,} [ "$SREALM" = "$DREALM" ] && SREALM="" ### Convert asterisks to empty strings SREALM=${SREALM#\*}; DREALM=${DREALM#\*} ### Attach route filter to the root class tc filter add dev $DEVICE parent 1:0 protocol ip \ prio $PRIO_REALM route ${SREALM:+from $SREALM} \ ${DREALM:+to $DREALM} classid 1:$CLASS done ### realm ### Create u32 filter for RULE fields for rule in `echo "$CFILE"| sed -n '/^RULE/ { s/.*=//; p; }'`; do ### Split rule into source & destination SRC=${rule%%,*}; DST=${rule##*,} [ "$SRC" = "$rule" ] && SRC="" ### Split destination into address, port & mask fields DADDR=${DST%%:*}; DTEMP=${DST##*:} [ "$DADDR" = "$DST" ] && DTEMP="" DPORT=${DTEMP%%/*}; DMASK=${DTEMP##*/} [ "$DPORT" = "$DTEMP" ] && DMASK="0xffff" ### Split up source (if specified) SADDR=""; SPORT="" if [ -n "$SRC" ]; then SADDR=${SRC%%:*}; STEMP=${SRC##*:} [ "$SADDR" = "$SRC" ] && STEMP="" SPORT=${STEMP%%/*}; SMASK=${STEMP##*/} [ "$SPORT" = "$STEMP" ] && SMASK="0xffff" fi ### Convert asterisks to empty strings SADDR=${SADDR#\*}; DADDR=${DADDR#\*} ### Compose u32 filter rules u32_s="${SPORT:+match ip sport $SPORT $SMASK}" u32_s="${SADDR:+match ip src $SADDR} $u32_s" u32_d="${DPORT:+match ip dport $DPORT $DMASK}" u32_d="${DADDR:+match ip dst $DADDR} $u32_d" ### Uncomment the following if you want to see parsed rules #echo "$rule: $u32_s $u32_d" ### Attach u32 filter to the appropriate class tc filter add dev $DEVICE parent 1:0 protocol ip \ prio $PRIO_RULE u32 $u32_s $u32_d classid 1:$CLASS done ### rule [ "$1" = "compile" ] && echo done ### classfile ;; ############################################################################# ################################# TIME CHECK ################################ ############################################################################# timecheck) ### Get time + weekday TIME_TMP=`date +%w/%k:%M` TIME_DOW=${TIME_TMP%%/*} TIME_NOW=${TIME_TMP##*/} TIME_ABS=`cbq_time2abs $TIME_NOW` ### Load DEVICES, DEVFIELDS and CLASSLIST cbq_init $CBQ_PATH ### Run through all classes for classfile in $CLASSLIST; do ### Gather all TIME rules from class config TIMESET=`sed -n 's/#.*//; s/[[:space:]]//g; /^TIME/ { s/.*=//; p; }' \ $CBQ_PATH/$classfile` [ -z "$TIMESET" ] && continue MATCH=0; CHANGE=0 for timerule in $TIMESET; do ### Split TIME rule to pieces TIMESPEC=${timerule%%;*}; PARAMS=${timerule##*;} WEEKDAYS=${TIMESPEC%%/*}; INTERVAL=${TIMESPEC##*/} BEG_TIME=${INTERVAL%%-*}; END_TIME=${INTERVAL##*-} ### Check the day-of-week (if present) [ "$WEEKDAYS" != "$INTERVAL" -a \ -n "${WEEKDAYS##*$TIME_DOW*}" ] && continue ### Compute interval boundaries BEG_ABS=`cbq_time2abs $BEG_TIME` END_ABS=`cbq_time2abs $END_TIME` ### Midnight wrap fixup if [ $BEG_ABS -gt $END_ABS ]; then [ $TIME_ABS -le $END_ABS ] && TIME_ABS=$[TIME_ABS + 24*60] END_ABS=$[END_ABS + 24*60] fi ### If the time matches, remember params and set MATCH flag if [ $TIME_ABS -ge $BEG_ABS -a $TIME_ABS -lt $END_ABS ]; then TMP_RATE=${PARAMS%%/*}; PARAMS=${PARAMS#*/} TMP_WGHT=${PARAMS%%/*}; TMP_PEAK=${PARAMS##*/} [ "$TMP_PEAK" = "$TMP_WGHT" ] && TMP_PEAK="" TMP_PEAK=${TMP_PEAK:+peakrate $TMP_PEAK} MATCH=1 fi done ### timerule cbq_load_class $CBQ_PATH $classfile ### Get current RATE of CBQ class RATE_NOW=`tc class show dev $DEVICE| sed -n \ "/cbq 1:$CLASS / { s/.*rate //; s/ .*//; p; q; }"` [ -z "$RATE_NOW" ] && continue ### Time interval matched if [ $MATCH -ne 0 ]; then ### Check if there is any change in class RATE if [ "$RATE_NOW" != "$TMP_RATE" ]; then NEW_RATE="$TMP_RATE" NEW_WGHT="$TMP_WGHT" NEW_PEAK="$TMP_PEAK" CHANGE=1 fi ### Match not found, reset to default RATE if necessary elif [ "$RATE_NOW" != "$RATE" ]; then NEW_WGHT="$WEIGHT" NEW_RATE="$RATE" NEW_PEAK="$PEAK" CHANGE=1 fi ### If there are no changes, go for next class [ $CHANGE -eq 0 ] && continue ### Replace CBQ class tc class replace dev $DEVICE classid 1:$CLASS cbq \ bandwidth $BANDWIDTH rate $NEW_RATE weight $NEW_WGHT prio $PRIO \ allot 1514 cell 8 maxburst 20 avpkt 1000 $BOUNDED $ISOLATED ### Replace leaf qdisc (if any) if [ "$LEAF" = "tbf" ]; then tc qdisc replace dev $DEVICE handle $CLASS tbf \ rate $NEW_RATE buffer $BUFFER limit $LIMIT mtu $MTU $NEW_PEAK fi cbq_message "$TIME_NOW: class $CLASS on $DEVICE changed rate ($RATE_NOW -> $NEW_RATE)" done ### class file ;; ############################################################################# ################################## THE REST ################################# ############################################################################# stop) cbq_off ;; list) cbq_show ;; stats) cbq_show -s ;; restart) shift $0 stop $0 start "$@" ;; *) echo "Usage: `basename $0` {start|compile|stop|restart|timecheck|list|stats}" esac
