1 /proc/sys/net/ipv4/* Variables:
7 Forward Packets between interfaces.
9 This variable is special, its change resets all configuration
10 parameters to their default state (RFC1122 for hosts, RFC1812
13 ip_default_ttl - INTEGER
16 ip_no_pmtu_disc - BOOLEAN
17 Disable Path MTU Discovery.
22 ipfrag_high_thresh - INTEGER
23 Maximum memory used to reassemble IP fragments. When
24 ipfrag_high_thresh bytes of memory is allocated for this purpose,
25 the fragment handler will toss packets until ipfrag_low_thresh
28 ipfrag_low_thresh - INTEGER
29 See ipfrag_high_thresh
32 Time in seconds to keep an IP fragment in memory.
36 inet_peer_threshold - INTEGER
37 The approximate size of the storage. Starting from this threshold
38 entries will be thrown aggressively. This threshold also determines
39 entries' time-to-live and time intervals between garbage collection
40 passes. More entries, less time-to-live, less GC interval.
42 inet_peer_minttl - INTEGER
43 Minimum time-to-live of entries. Should be enough to cover fragment
44 time-to-live on the reassembling side. This minimum time-to-live is
45 guaranteed if the pool size is less than inet_peer_threshold.
46 Measured in jiffies(1).
48 inet_peer_maxttl - INTEGER
49 Maximum time-to-live of entries. Unused entries will expire after
50 this period of time if there is no memory pressure on the pool (i.e.
51 when the number of entries in the pool is very small).
52 Measured in jiffies(1).
54 inet_peer_gc_mintime - INTEGER
55 Minimum interval between garbage collection passes. This interval is
56 in effect under high memory pressure on the pool.
57 Measured in jiffies(1).
59 inet_peer_gc_maxtime - INTEGER
60 Minimum interval between garbage collection passes. This interval is
61 in effect under low (or absent) memory pressure on the pool.
62 Measured in jiffies(1).
66 tcp_syn_retries - INTEGER
67 Number of times initial SYNs for an active TCP connection attempt
68 will be retransmitted. Should not be higher than 255. Default value
69 is 5, which corresponds to ~180seconds.
71 tcp_synack_retries - INTEGER
72 Number of times SYNACKs for a passive TCP connection attempt will
73 be retransmitted. Should not be higher than 255. Default value
74 is 5, which corresponds to ~180seconds.
76 tcp_keepalive_time - INTEGER
77 How often TCP sends out keepalive messages when keepalive is enabled.
80 tcp_keepalive_probes - INTEGER
81 How many keepalive probes TCP sends out, until it decides that the
82 connection is broken. Default value: 9.
84 tcp_keepalive_intvl - INTEGER
85 How frequently the probes are send out. Multiplied by
86 tcp_keepalive_probes it is time to kill not responding connection,
87 after probes started. Default value: 75sec i.e. connection
88 will be aborted after ~11 minutes of retries.
90 tcp_retries1 - INTEGER
91 How many times to retry before deciding that something is wrong
92 and it is necessary to report this suspection to network layer.
93 Minimal RFC value is 3, it is default, which corresponds
94 to ~3sec-8min depending on RTO.
96 tcp_retries2 - INTEGER
97 How may times to retry before killing alive TCP connection.
98 RFC1122 says that the limit should be longer than 100 sec.
99 It is too small number. Default value 15 corresponds to ~13-30min
102 tcp_orphan_retries - INTEGER
103 How may times to retry before killing TCP connection, closed
104 by our side. Default value 7 corresponds to ~50sec-16min
105 depending on RTO. If you machine is loaded WEB server,
106 you should think about lowering this value, such sockets
107 may consume significant resources. Cf. tcp_max_orphans.
109 tcp_fin_timeout - INTEGER
110 Time to hold socket in state FIN-WAIT-2, if it was closed
111 by our side. Peer can be broken and never close its side,
112 or even died unexpectedly. Default value is 60sec.
113 Usual value used in 2.2 was 180 seconds, you may restore
114 it, but remember that if your machine is even underloaded WEB server,
115 you risk to overflow memory with kilotons of dead sockets,
116 FIN-WAIT-2 sockets are less dangerous than FIN-WAIT-1,
117 because they eat maximum 1.5K of memory, but they tend
118 to live longer. Cf. tcp_max_orphans.
120 tcp_max_tw_buckets - INTEGER
121 Maximal number of timewait sockets held by system simultaneously.
122 If this number is exceeded time-wait socket is immediately destroyed
123 and warning is printed. This limit exists only to prevent
124 simple DoS attacks, you _must_ not lower the limit artificially,
125 but rather increase it (probably, after increasing installed memory),
126 if network conditions require more than default value.
128 tcp_tw_recycle - BOOLEAN
129 Enable fast recycling TIME-WAIT sockets. Default value is 0.
130 It should not be changed without advice/request of technical
133 tcp_tw_reuse - BOOLEAN
134 Allow to reuse TIME-WAIT sockets for new connections when it is
135 safe from protocol viewpoint. Default value is 0.
136 It should not be changed without advice/request of technical
139 tcp_max_orphans - INTEGER
140 Maximal number of TCP sockets not attached to any user file handle,
141 held by system. If this number is exceeded orphaned connections are
142 reset immediately and warning is printed. This limit exists
143 only to prevent simple DoS attacks, you _must_ not rely on this
144 or lower the limit artificially, but rather increase it
145 (probably, after increasing installed memory),
146 if network conditions require more than default value,
147 and tune network services to linger and kill such states
148 more aggressively. Let me to remind again: each orphan eats
149 up to ~64K of unswappable memory.
151 tcp_abort_on_overflow - BOOLEAN
152 If listening service is too slow to accept new connections,
153 reset them. Default state is FALSE. It means that if overflow
154 occurred due to a burst, connection will recover. Enable this
155 option _only_ if you are really sure that listening daemon
156 cannot be tuned to accept connections faster. Enabling this
157 option can harm clients of your server.
159 tcp_syncookies - BOOLEAN
160 Only valid when the kernel was compiled with CONFIG_SYNCOOKIES
161 Send out syncookies when the syn backlog queue of a socket
162 overflows. This is to prevent against the common 'syn flood attack'
165 Note, that syncookies is fallback facility.
166 It MUST NOT be used to help highly loaded servers to stand
167 against legal connection rate. If you see synflood warnings
168 in your logs, but investigation shows that they occur
169 because of overload with legal connections, you should tune
170 another parameters until this warning disappear.
171 See: tcp_max_syn_backlog, tcp_synack_retries, tcp_abort_on_overflow.
173 syncookies seriously violate TCP protocol, do not allow
174 to use TCP extensions, can result in serious degradation
175 of some services (f.e. SMTP relaying), visible not by you,
176 but your clients and relays, contacting you. While you see
177 synflood warnings in logs not being really flooded, your server
178 is seriously misconfigured.
181 Use the Host requirements interpretation of the TCP urg pointer field.
182 Most hosts use the older BSD interpretation, so if you turn this on
183 Linux might not communicate correctly with them.
186 tcp_max_syn_backlog - INTEGER
187 Maximal number of remembered connection requests, which are
188 still did not receive an acknowledgement from connecting client.
189 Default value is 1024 for systems with more than 128Mb of memory,
190 and 128 for low memory machines. If server suffers of overload,
191 try to increase this number.
193 tcp_window_scaling - BOOLEAN
194 Enable window scaling as defined in RFC1323.
196 tcp_timestamps - BOOLEAN
197 Enable timestamps as defined in RFC1323.
200 Enable select acknowledgments (SACKS).
203 Enable FACK congestion avoidance and fast restransmission.
204 The value is not used, if tcp_sack is not enabled.
207 Allows TCP to send "duplicate" SACKs.
210 Enable Explicit Congestion Notification in TCP.
212 tcp_reordering - INTEGER
213 Maximal reordering of packets in a TCP stream.
216 tcp_retrans_collapse - BOOLEAN
217 Bug-to-bug compatibility with some broken printers.
218 On retransmit try to send bigger packets to work around bugs in
221 tcp_wmem - vector of 3 INTEGERs: min, default, max
222 min: Amount of memory reserved for send buffers for TCP socket.
223 Each TCP socket has rights to use it due to fact of its birth.
226 default: Amount of memory allowed for send buffers for TCP socket
227 by default. This value overrides net.core.wmem_default used
228 by other protocols, it is usually lower than net.core.wmem_default.
231 max: Maximal amount of memory allowed for automatically selected
232 send buffers for TCP socket. This value does not override
233 net.core.wmem_max, "static" selection via SO_SNDBUF does not use this.
236 tcp_rmem - vector of 3 INTEGERs: min, default, max
237 min: Minimal size of receive buffer used by TCP sockets.
238 It is guaranteed to each TCP socket, even under moderate memory
242 default: default size of receive buffer used by TCP sockets.
243 This value overrides net.core.rmem_default used by other protocols.
244 Default: 87380 bytes. This value results in window of 65535 with
245 default setting of tcp_adv_win_scale and tcp_app_win:0 and a bit
246 less for default tcp_app_win. See below about these variables.
248 max: maximal size of receive buffer allowed for automatically
249 selected receiver buffers for TCP socket. This value does not override
250 net.core.rmem_max, "static" selection via SO_RCVBUF does not use this.
251 Default: 87380*2 bytes.
253 tcp_mem - vector of 3 INTEGERs: min, pressure, max
254 low: below this number of pages TCP is not bothered about its
257 pressure: when amount of memory allocated by TCP exceeds this number
258 of pages, TCP moderates its memory consumption and enters memory
259 pressure mode, which is exited when memory consumtion falls
262 high: number of pages allowed for queueing by all TCP sockets.
264 Defaults are calculated at boot time from amount of available
267 tcp_app_win - INTEGER
268 Reserve max(window/2^tcp_app_win, mss) of window for application
269 buffer. Value 0 is special, it means that nothing is reserved.
272 tcp_adv_win_scale - INTEGER
273 Count buffering overhead as bytes/2^tcp_adv_win_scale
274 (if tcp_adv_win_scale > 0) or bytes-bytes/2^(-tcp_adv_win_scale),
278 tcp_rfc1337 - BOOLEAN
279 If set, the TCP stack behaves conforming to RFC1337. If unset,
280 we are not conforming to RFC, but prevent TCP TIME_WAIT
284 tcp_low_latency - BOOLEAN
285 If set, the TCP stack makes decisions that prefer lower
286 latency as opposed to higher throughput. By default, this
287 option is not set meaning that higher throughput is preferred.
288 An example of an application where this default should be
289 changed would be a Beowulf compute cluster.
292 tcp_vegas_cong_avoid - BOOLEAN
293 Enable TCP Vegas congestion avoidance algorithm.
294 TCP Vegas is a sender-side only change to TCP that anticipates
295 the onset of congestion by estimating the bandwidth. TCP Vegas
296 adjusts the sending rate by modifying the congestion
297 window. TCP Vegas should provide less packet loss, but it is
298 not as aggressive as TCP Reno.
302 Enable BIC TCP congestion control algorithm.
303 BIC-TCP is a sender-side only change that ensures a linear RTT
304 fairness under large windows while offering both scalability and
305 bounded TCP-friendliness. The protocol combines two schemes
306 called additive increase and binary search increase. When the
307 congestion window is large, additive increase with a large
308 increment ensures linear RTT fairness as well as good
309 scalability. Under small congestion windows, binary search
310 increase provides TCP friendliness.
313 tcp_bic_low_window - INTEGER
314 Sets the threshold window (in packets) where BIC TCP starts to
315 adjust the congestion window. Below this threshold BIC TCP behaves
316 the same as the default TCP Reno.
319 tcp_bic_fast_convergence - BOOLEAN
320 Forces BIC TCP to more quickly respond to changes in congestion
321 window. Allows two flows sharing the same connection to converge
325 tcp_default_win_scale - INTEGER
326 Sets the minimum window scale TCP will negotiate for on all
330 ip_local_port_range - 2 INTEGERS
331 Defines the local port range that is used by TCP and UDP to
332 choose the local port. The first number is the first, the
333 second the last local port number. Default value depends on
334 amount of memory available on the system:
336 < 128Mb 1024-4999 or even less.
337 This number defines number of active connections, which this
338 system can issue simultaneously to systems not supporting
339 TCP extensions (timestamps). With tcp_tw_recycle enabled
340 (i.e. by default) range 1024-4999 is enough to issue up to
341 2000 connections per second to systems supporting timestamps.
343 ip_nonlocal_bind - BOOLEAN
344 If set, allows processes to bind() to non-local IP adresses,
345 which can be quite useful - but may break some applications.
349 If set non-zero, enables support for dynamic addresses.
350 If set to a non-zero value larger than 1, a kernel log
351 message will be printed when dynamic address rewriting
355 icmp_echo_ignore_all - BOOLEAN
356 icmp_echo_ignore_broadcasts - BOOLEAN
357 If either is set to true, then the kernel will ignore either all
358 ICMP ECHO requests sent to it or just those to broadcast/multicast
359 addresses, respectively.
361 icmp_ratelimit - INTEGER
362 Limit the maximal rates for sending ICMP packets whose type matches
363 icmp_ratemask (see below) to specific targets.
364 0 to disable any limiting, otherwise the maximal rate in jiffies(1)
367 icmp_ratemask - INTEGER
368 Mask made of ICMP types for which rates are being limited.
369 Significant bits: IHGFEDCBA9876543210
370 Default mask: 0000001100000011000 (6168)
372 Bit definitions (see include/linux/icmp.h):
374 3 Destination Unreachable *
379 C Parameter Problem *
384 H Address Mask Request
387 * These are rate limited by default (see default mask above)
389 icmp_ignore_bogus_error_responses - BOOLEAN
390 Some routers violate RFC1122 by sending bogus responses to broadcast
391 frames. Such violations are normally logged via a kernel warning.
392 If this is set to TRUE, the kernel will not give such warnings, which
393 will avoid log file clutter.
396 igmp_max_memberships - INTEGER
397 Change the maximum number of multicast groups we can subscribe to.
400 conf/interface/* changes special settings per interface (where "interface" is
401 the name of your network interface)
402 conf/all/* is special, changes the settings for all interfaces
405 log_martians - BOOLEAN
406 Log packets with impossible addresses to kernel log.
407 log_martians for the interface will be enabled if at least one of
408 conf/{all,interface}/log_martians is set to TRUE,
409 it will be disabled otherwise
411 accept_redirects - BOOLEAN
412 Accept ICMP redirect messages.
413 accept_redirects for the interface will be enabled if:
414 - both conf/{all,interface}/accept_redirects are TRUE in the case forwarding
415 for the interface is enabled
417 - at least one of conf/{all,interface}/accept_redirects is TRUE in the case
418 forwarding for the interface is disabled
419 accept_redirects for the interface will be disabled otherwise
424 Enable IP forwarding on this interface.
426 mc_forwarding - BOOLEAN
427 Do multicast routing. The kernel needs to be compiled with CONFIG_MROUTE
428 and a multicast routing daemon is required.
429 conf/all/mc_forwarding must also be set to TRUE to enable multicast routing
433 Integer value used to differentiate the devices by the medium they
434 are attached to. Two devices can have different id values when
435 the broadcast packets are received only on one of them.
436 The default value 0 means that the device is the only interface
437 to its medium, value of -1 means that medium is not known.
439 Currently, it is used to change the proxy_arp behavior:
440 the proxy_arp feature is enabled for packets forwarded between
441 two devices attached to different media.
445 proxy_arp for the interface will be enabled if at least one of
446 conf/{all,interface}/proxy_arp is set to TRUE,
447 it will be disabled otherwise
449 shared_media - BOOLEAN
450 Send(router) or accept(host) RFC1620 shared media redirects.
451 Overrides ip_secure_redirects.
452 shared_media for the interface will be enabled if at least one of
453 conf/{all,interface}/shared_media is set to TRUE,
454 it will be disabled otherwise
457 secure_redirects - BOOLEAN
458 Accept ICMP redirect messages only for gateways,
459 listed in default gateway list.
460 secure_redirects for the interface will be enabled if at least one of
461 conf/{all,interface}/secure_redirects is set to TRUE,
462 it will be disabled otherwise
465 send_redirects - BOOLEAN
466 Send redirects, if router.
467 send_redirects for the interface will be enabled if at least one of
468 conf/{all,interface}/send_redirects is set to TRUE,
469 it will be disabled otherwise
472 bootp_relay - BOOLEAN
473 Accept packets with source address 0.b.c.d destined
474 not to this host as local ones. It is supposed, that
475 BOOTP relay daemon will catch and forward such packets.
476 conf/all/bootp_relay must also be set to TRUE to enable BOOTP relay
481 accept_source_route - BOOLEAN
482 Accept packets with SRR option.
483 conf/all/accept_source_route must also be set to TRUE to accept packets
484 with SRR option on the interface
485 default TRUE (router)
489 1 - do source validation by reversed path, as specified in RFC1812
490 Recommended option for single homed hosts and stub network
491 routers. Could cause troubles for complicated (not loop free)
492 networks running a slow unreliable protocol (sort of RIP),
493 or using static routes.
495 0 - No source validation.
497 conf/all/rp_filter must also be set to TRUE to do source validation
500 Default value is 0. Note that some distributions enable it
504 1 - Allows you to have multiple network interfaces on the same
505 subnet, and have the ARPs for each interface be answered
506 based on whether or not the kernel would route a packet from
507 the ARP'd IP out that interface (therefore you must use source
508 based routing for this to work). In other words it allows control
509 of which cards (usually 1) will respond to an arp request.
511 0 - (default) The kernel can respond to arp requests with addresses
512 from other interfaces. This may seem wrong but it usually makes
513 sense, because it increases the chance of successful communication.
514 IP addresses are owned by the complete host on Linux, not by
515 particular interfaces. Only for more complex setups like load-
516 balancing, does this behaviour cause problems.
518 arp_filter for the interface will be enabled if at least one of
519 conf/{all,interface}/arp_filter is set to TRUE,
520 it will be disabled otherwise
522 arp_announce - INTEGER
523 Define different restriction levels for announcing the local
524 source IP address from IP packets in ARP requests sent on
526 0 - (default) Use any local address, configured on any interface
527 1 - Try to avoid local addresses that are not in the target's
528 subnet for this interface. This mode is useful when target
529 hosts reachable via this interface require the source IP
530 address in ARP requests to be part of their logical network
531 configured on the receiving interface. When we generate the
532 request we will check all our subnets that include the
533 target IP and will preserve the source address if it is from
534 such subnet. If there is no such subnet we select source
535 address according to the rules for level 2.
536 2 - Always use the best local address for this target.
537 In this mode we ignore the source address in the IP packet
538 and try to select local address that we prefer for talks with
539 the target host. Such local address is selected by looking
540 for primary IP addresses on all our subnets on the outgoing
541 interface that include the target IP address. If no suitable
542 local address is found we select the first local address
543 we have on the outgoing interface or on all other interfaces,
544 with the hope we will receive reply for our request and
545 even sometimes no matter the source IP address we announce.
547 The max value from conf/{all,interface}/arp_announce is used.
549 Increasing the restriction level gives more chance for
550 receiving answer from the resolved target while decreasing
551 the level announces more valid sender's information.
554 Define different modes for sending replies in response to
555 received ARP requests that resolve local target IP addresses:
556 0 - (default): reply for any local target IP address, configured
558 1 - reply only if the target IP address is local address
559 configured on the incoming interface
560 2 - reply only if the target IP address is local address
561 configured on the incoming interface and both with the
562 sender's IP address are part from same subnet on this interface
563 3 - do not reply for local addresses configured with scope host,
564 only resolutions for global and link addresses are replied
566 8 - do not reply for all local addresses
568 The max value from conf/{all,interface}/arp_ignore is used
569 when ARP request is received on the {interface}
572 Allows you to write a number, which can be used as required.
575 (1) Jiffie: internal timeunit for the kernel. On the i386 1/100s, on the
576 Alpha 1/1024s. See the HZ define in /usr/include/asm/param.h for the exact
577 value on your system.
586 delon.nicolas@wanadoo.fr
591 /proc/sys/net/ipv6/* Variables:
593 IPv6 has no global variables such as tcp_*. tcp_* settings under ipv4/ also
594 apply to IPv6 [XXX?].
597 Default value for IPV6_V6ONLY socket option,
598 which restricts use of the IPv6 socket to IPv6 communication
600 TRUE: disable IPv4-mapped address feature
601 FALSE: enable IPv4-mapped address feature
603 Default: FALSE (as specified in RFC2553bis)
606 Change the interface-specific default settings.
610 Change all the interface-specific settings.
612 [XXX: Other special features than forwarding?]
614 conf/all/forwarding - BOOLEAN
615 Enable global IPv6 forwarding between all interfaces.
617 IPv4 and IPv6 work differently here; e.g. netfilter must be used
618 to control which interfaces may forward packets and which not.
620 This also sets all interfaces' Host/Router setting
621 'forwarding' to the specified value. See below for details.
623 This referred to as global forwarding.
626 Change special settings per interface.
628 The functional behaviour for certain settings is different
629 depending on whether local forwarding is enabled or not.
632 Accept Router Advertisements; autoconfigure using them.
634 Functional default: enabled if local forwarding is disabled.
635 disabled if local forwarding is enabled.
637 accept_redirects - BOOLEAN
640 Functional default: enabled if local forwarding is disabled.
641 disabled if local forwarding is enabled.
644 Autoconfigure addresses using Prefix Information in Router
647 Functional default: enabled if accept_ra is enabled.
648 disabled if accept_ra is disabled.
650 dad_transmits - INTEGER
651 The amount of Duplicate Address Detection probes to send.
655 Configure interface-specific Host/Router behaviour.
657 Note: It is recommended to have the same setting on all
658 interfaces; mixed router/host scenarios are rather uncommon.
662 By default, Host behaviour is assumed. This means:
664 1. IsRouter flag is not set in Neighbour Advertisements.
665 2. Router Solicitations are being sent when necessary.
666 3. If accept_ra is TRUE (default), accept Router
667 Advertisements (and do autoconfiguration).
668 4. If accept_redirects is TRUE (default), accept Redirects.
672 If local forwarding is enabled, Router behaviour is assumed.
673 This means exactly the reverse from the above:
675 1. IsRouter flag is set in Neighbour Advertisements.
676 2. Router Solicitations are not sent.
677 3. Router Advertisements are ignored.
678 4. Redirects are ignored.
680 Default: FALSE if global forwarding is disabled (default),
684 Default Hop Limit to set.
688 Default Maximum Transfer Unit
689 Default: 1280 (IPv6 required minimum)
691 router_solicitation_delay - INTEGER
692 Number of seconds to wait after interface is brought up
693 before sending Router Solicitations.
696 router_solicitation_interval - INTEGER
697 Number of seconds to wait between Router Solicitations.
700 router_solicitations - INTEGER
701 Number of Router Solicitations to send until assuming no
707 Limit the maximal rates for sending ICMPv6 packets.
708 0 to disable any limiting, otherwise the maximal rate in jiffies(1)
712 Pekka Savola <pekkas@netcore.fi>
713 YOSHIFUJI Hideaki / USAGI Project <yoshfuji@linux-ipv6.org>
715 $Id: ip-sysctl.txt,v 1.19.2.1 2001/12/13 08:59:27 davem Exp $