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
14 Default value of TTL field (Time To Live) for outgoing (but not
15 forwarded) IP packets. Should be between 1 and 255 inclusive.
16 Default: 64 (as recommended by RFC1700)
18 ip_no_pmtu_disc - INTEGER
19 Disable Path MTU Discovery. If enabled in mode 1 and a
20 fragmentation-required ICMP is received, the PMTU to this
21 destination will be set to min_pmtu (see below). You will need
22 to raise min_pmtu to the smallest interface MTU on your system
23 manually if you want to avoid locally generated fragments.
25 In mode 2 incoming Path MTU Discovery messages will be
26 discarded. Outgoing frames are handled the same as in mode 1,
27 implicitly setting IP_PMTUDISC_DONT on every created socket.
29 Mode 3 is a hardend pmtu discover mode. The kernel will only
30 accept fragmentation-needed errors if the underlying protocol
31 can verify them besides a plain socket lookup. Current
32 protocols for which pmtu events will be honored are TCP, SCTP
33 and DCCP as they verify e.g. the sequence number or the
34 association. This mode should not be enabled globally but is
35 only intended to secure e.g. name servers in namespaces where
36 TCP path mtu must still work but path MTU information of other
37 protocols should be discarded. If enabled globally this mode
38 could break other protocols.
44 default 552 - minimum discovered Path MTU
46 ip_forward_use_pmtu - BOOLEAN
47 By default we don't trust protocol path MTUs while forwarding
48 because they could be easily forged and can lead to unwanted
49 fragmentation by the router.
50 You only need to enable this if you have user-space software
51 which tries to discover path mtus by itself and depends on the
52 kernel honoring this information. This is normally not the
59 fwmark_reflect - BOOLEAN
60 Controls the fwmark of kernel-generated IPv4 reply packets that are not
61 associated with a socket for example, TCP RSTs or ICMP echo replies).
62 If unset, these packets have a fwmark of zero. If set, they have the
63 fwmark of the packet they are replying to.
66 fib_multipath_use_neigh - BOOLEAN
67 Use status of existing neighbor entry when determining nexthop for
68 multipath routes. If disabled, neighbor information is not used and
69 packets could be directed to a failed nexthop. Only valid for kernels
70 built with CONFIG_IP_ROUTE_MULTIPATH enabled.
76 fib_multipath_hash_policy - INTEGER
77 Controls which hash policy to use for multipath routes. Only valid
78 for kernels built with CONFIG_IP_ROUTE_MULTIPATH enabled.
84 route/max_size - INTEGER
85 Maximum number of routes allowed in the kernel. Increase
86 this when using large numbers of interfaces and/or routes.
87 From linux kernel 3.6 onwards, this is deprecated for ipv4
88 as route cache is no longer used.
90 neigh/default/gc_thresh1 - INTEGER
91 Minimum number of entries to keep. Garbage collector will not
92 purge entries if there are fewer than this number.
95 neigh/default/gc_thresh2 - INTEGER
96 Threshold when garbage collector becomes more aggressive about
97 purging entries. Entries older than 5 seconds will be cleared
98 when over this number.
101 neigh/default/gc_thresh3 - INTEGER
102 Maximum number of neighbor entries allowed. Increase this
103 when using large numbers of interfaces and when communicating
104 with large numbers of directly-connected peers.
107 neigh/default/unres_qlen_bytes - INTEGER
108 The maximum number of bytes which may be used by packets
109 queued for each unresolved address by other network layers.
111 Setting negative value is meaningless and will return error.
112 Default: 65536 Bytes(64KB)
114 neigh/default/unres_qlen - INTEGER
115 The maximum number of packets which may be queued for each
116 unresolved address by other network layers.
117 (deprecated in linux 3.3) : use unres_qlen_bytes instead.
118 Prior to linux 3.3, the default value is 3 which may cause
119 unexpected packet loss. The current default value is calculated
120 according to default value of unres_qlen_bytes and true size of
124 mtu_expires - INTEGER
125 Time, in seconds, that cached PMTU information is kept.
127 min_adv_mss - INTEGER
128 The advertised MSS depends on the first hop route MTU, but will
129 never be lower than this setting.
133 ipfrag_high_thresh - INTEGER
134 Maximum memory used to reassemble IP fragments. When
135 ipfrag_high_thresh bytes of memory is allocated for this purpose,
136 the fragment handler will toss packets until ipfrag_low_thresh
137 is reached. This also serves as a maximum limit to namespaces
138 different from the initial one.
140 ipfrag_low_thresh - INTEGER
141 Maximum memory used to reassemble IP fragments before the kernel
142 begins to remove incomplete fragment queues to free up resources.
143 The kernel still accepts new fragments for defragmentation.
145 ipfrag_time - INTEGER
146 Time in seconds to keep an IP fragment in memory.
148 ipfrag_max_dist - INTEGER
149 ipfrag_max_dist is a non-negative integer value which defines the
150 maximum "disorder" which is allowed among fragments which share a
151 common IP source address. Note that reordering of packets is
152 not unusual, but if a large number of fragments arrive from a source
153 IP address while a particular fragment queue remains incomplete, it
154 probably indicates that one or more fragments belonging to that queue
155 have been lost. When ipfrag_max_dist is positive, an additional check
156 is done on fragments before they are added to a reassembly queue - if
157 ipfrag_max_dist (or more) fragments have arrived from a particular IP
158 address between additions to any IP fragment queue using that source
159 address, it's presumed that one or more fragments in the queue are
160 lost. The existing fragment queue will be dropped, and a new one
161 started. An ipfrag_max_dist value of zero disables this check.
163 Using a very small value, e.g. 1 or 2, for ipfrag_max_dist can
164 result in unnecessarily dropping fragment queues when normal
165 reordering of packets occurs, which could lead to poor application
166 performance. Using a very large value, e.g. 50000, increases the
167 likelihood of incorrectly reassembling IP fragments that originate
168 from different IP datagrams, which could result in data corruption.
173 inet_peer_threshold - INTEGER
174 The approximate size of the storage. Starting from this threshold
175 entries will be thrown aggressively. This threshold also determines
176 entries' time-to-live and time intervals between garbage collection
177 passes. More entries, less time-to-live, less GC interval.
179 inet_peer_minttl - INTEGER
180 Minimum time-to-live of entries. Should be enough to cover fragment
181 time-to-live on the reassembling side. This minimum time-to-live is
182 guaranteed if the pool size is less than inet_peer_threshold.
185 inet_peer_maxttl - INTEGER
186 Maximum time-to-live of entries. Unused entries will expire after
187 this period of time if there is no memory pressure on the pool (i.e.
188 when the number of entries in the pool is very small).
194 Limit of socket listen() backlog, known in userspace as SOMAXCONN.
195 Defaults to 128. See also tcp_max_syn_backlog for additional tuning
198 tcp_abort_on_overflow - BOOLEAN
199 If listening service is too slow to accept new connections,
200 reset them. Default state is FALSE. It means that if overflow
201 occurred due to a burst, connection will recover. Enable this
202 option _only_ if you are really sure that listening daemon
203 cannot be tuned to accept connections faster. Enabling this
204 option can harm clients of your server.
206 tcp_adv_win_scale - INTEGER
207 Count buffering overhead as bytes/2^tcp_adv_win_scale
208 (if tcp_adv_win_scale > 0) or bytes-bytes/2^(-tcp_adv_win_scale),
210 Possible values are [-31, 31], inclusive.
213 tcp_allowed_congestion_control - STRING
214 Show/set the congestion control choices available to non-privileged
215 processes. The list is a subset of those listed in
216 tcp_available_congestion_control.
217 Default is "reno" and the default setting (tcp_congestion_control).
219 tcp_app_win - INTEGER
220 Reserve max(window/2^tcp_app_win, mss) of window for application
221 buffer. Value 0 is special, it means that nothing is reserved.
224 tcp_autocorking - BOOLEAN
225 Enable TCP auto corking :
226 When applications do consecutive small write()/sendmsg() system calls,
227 we try to coalesce these small writes as much as possible, to lower
228 total amount of sent packets. This is done if at least one prior
229 packet for the flow is waiting in Qdisc queues or device transmit
230 queue. Applications can still use TCP_CORK for optimal behavior
231 when they know how/when to uncork their sockets.
234 tcp_available_congestion_control - STRING
235 Shows the available congestion control choices that are registered.
236 More congestion control algorithms may be available as modules,
239 tcp_base_mss - INTEGER
240 The initial value of search_low to be used by the packetization layer
241 Path MTU discovery (MTU probing). If MTU probing is enabled,
242 this is the initial MSS used by the connection.
244 tcp_congestion_control - STRING
245 Set the congestion control algorithm to be used for new
246 connections. The algorithm "reno" is always available, but
247 additional choices may be available based on kernel configuration.
248 Default is set as part of kernel configuration.
249 For passive connections, the listener congestion control choice
251 [see setsockopt(listenfd, SOL_TCP, TCP_CONGESTION, "name" ...) ]
254 Allows TCP to send "duplicate" SACKs.
256 tcp_early_retrans - INTEGER
257 Tail loss probe (TLP) converts RTOs occurring due to tail
258 losses into fast recovery (draft-ietf-tcpm-rack). Note that
259 TLP requires RACK to function properly (see tcp_recovery below)
266 Control use of Explicit Congestion Notification (ECN) by TCP.
267 ECN is used only when both ends of the TCP connection indicate
268 support for it. This feature is useful in avoiding losses due
269 to congestion by allowing supporting routers to signal
270 congestion before having to drop packets.
272 0 Disable ECN. Neither initiate nor accept ECN.
273 1 Enable ECN when requested by incoming connections and
274 also request ECN on outgoing connection attempts.
275 2 Enable ECN when requested by incoming connections
276 but do not request ECN on outgoing connections.
279 tcp_ecn_fallback - BOOLEAN
280 If the kernel detects that ECN connection misbehaves, enable fall
281 back to non-ECN. Currently, this knob implements the fallback
282 from RFC3168, section 6.1.1.1., but we reserve that in future,
283 additional detection mechanisms could be implemented under this
284 knob. The value is not used, if tcp_ecn or per route (or congestion
285 control) ECN settings are disabled.
286 Default: 1 (fallback enabled)
289 Enable FACK congestion avoidance and fast retransmission.
290 The value is not used, if tcp_sack is not enabled.
292 tcp_fin_timeout - INTEGER
293 The length of time an orphaned (no longer referenced by any
294 application) connection will remain in the FIN_WAIT_2 state
295 before it is aborted at the local end. While a perfectly
296 valid "receive only" state for an un-orphaned connection, an
297 orphaned connection in FIN_WAIT_2 state could otherwise wait
298 forever for the remote to close its end of the connection.
303 Enables Forward RTO-Recovery (F-RTO) defined in RFC5682.
304 F-RTO is an enhanced recovery algorithm for TCP retransmission
305 timeouts. It is particularly beneficial in networks where the
306 RTT fluctuates (e.g., wireless). F-RTO is sender-side only
307 modification. It does not require any support from the peer.
309 By default it's enabled with a non-zero value. 0 disables F-RTO.
311 tcp_invalid_ratelimit - INTEGER
312 Limit the maximal rate for sending duplicate acknowledgments
313 in response to incoming TCP packets that are for an existing
314 connection but that are invalid due to any of these reasons:
316 (a) out-of-window sequence number,
317 (b) out-of-window acknowledgment number, or
318 (c) PAWS (Protection Against Wrapped Sequence numbers) check failure
320 This can help mitigate simple "ack loop" DoS attacks, wherein
321 a buggy or malicious middlebox or man-in-the-middle can
322 rewrite TCP header fields in manner that causes each endpoint
323 to think that the other is sending invalid TCP segments, thus
324 causing each side to send an unterminating stream of duplicate
325 acknowledgments for invalid segments.
327 Using 0 disables rate-limiting of dupacks in response to
328 invalid segments; otherwise this value specifies the minimal
329 space between sending such dupacks, in milliseconds.
331 Default: 500 (milliseconds).
333 tcp_keepalive_time - INTEGER
334 How often TCP sends out keepalive messages when keepalive is enabled.
337 tcp_keepalive_probes - INTEGER
338 How many keepalive probes TCP sends out, until it decides that the
339 connection is broken. Default value: 9.
341 tcp_keepalive_intvl - INTEGER
342 How frequently the probes are send out. Multiplied by
343 tcp_keepalive_probes it is time to kill not responding connection,
344 after probes started. Default value: 75sec i.e. connection
345 will be aborted after ~11 minutes of retries.
347 tcp_l3mdev_accept - BOOLEAN
348 Enables child sockets to inherit the L3 master device index.
349 Enabling this option allows a "global" listen socket to work
350 across L3 master domains (e.g., VRFs) with connected sockets
351 derived from the listen socket to be bound to the L3 domain in
352 which the packets originated. Only valid when the kernel was
353 compiled with CONFIG_NET_L3_MASTER_DEV.
355 tcp_low_latency - BOOLEAN
356 If set, the TCP stack makes decisions that prefer lower
357 latency as opposed to higher throughput. By default, this
358 option is not set meaning that higher throughput is preferred.
359 An example of an application where this default should be
360 changed would be a Beowulf compute cluster.
363 tcp_max_orphans - INTEGER
364 Maximal number of TCP sockets not attached to any user file handle,
365 held by system. If this number is exceeded orphaned connections are
366 reset immediately and warning is printed. This limit exists
367 only to prevent simple DoS attacks, you _must_ not rely on this
368 or lower the limit artificially, but rather increase it
369 (probably, after increasing installed memory),
370 if network conditions require more than default value,
371 and tune network services to linger and kill such states
372 more aggressively. Let me to remind again: each orphan eats
373 up to ~64K of unswappable memory.
375 tcp_max_syn_backlog - INTEGER
376 Maximal number of remembered connection requests, which have not
377 received an acknowledgment from connecting client.
378 The minimal value is 128 for low memory machines, and it will
379 increase in proportion to the memory of machine.
380 If server suffers from overload, try increasing this number.
382 tcp_max_tw_buckets - INTEGER
383 Maximal number of timewait sockets held by system simultaneously.
384 If this number is exceeded time-wait socket is immediately destroyed
385 and warning is printed. This limit exists only to prevent
386 simple DoS attacks, you _must_ not lower the limit artificially,
387 but rather increase it (probably, after increasing installed memory),
388 if network conditions require more than default value.
390 tcp_mem - vector of 3 INTEGERs: min, pressure, max
391 min: below this number of pages TCP is not bothered about its
394 pressure: when amount of memory allocated by TCP exceeds this number
395 of pages, TCP moderates its memory consumption and enters memory
396 pressure mode, which is exited when memory consumption falls
399 max: number of pages allowed for queueing by all TCP sockets.
401 Defaults are calculated at boot time from amount of available
404 tcp_min_rtt_wlen - INTEGER
405 The window length of the windowed min filter to track the minimum RTT.
406 A shorter window lets a flow more quickly pick up new (higher)
407 minimum RTT when it is moved to a longer path (e.g., due to traffic
408 engineering). A longer window makes the filter more resistant to RTT
409 inflations such as transient congestion. The unit is seconds.
412 tcp_moderate_rcvbuf - BOOLEAN
413 If set, TCP performs receive buffer auto-tuning, attempting to
414 automatically size the buffer (no greater than tcp_rmem[2]) to
415 match the size required by the path for full throughput. Enabled by
418 tcp_mtu_probing - INTEGER
419 Controls TCP Packetization-Layer Path MTU Discovery. Takes three
422 1 - Disabled by default, enabled when an ICMP black hole detected
423 2 - Always enabled, use initial MSS of tcp_base_mss.
425 tcp_probe_interval - INTEGER
426 Controls how often to start TCP Packetization-Layer Path MTU
427 Discovery reprobe. The default is reprobing every 10 minutes as
430 tcp_probe_threshold - INTEGER
431 Controls when TCP Packetization-Layer Path MTU Discovery probing
432 will stop in respect to the width of search range in bytes. Default
435 tcp_no_metrics_save - BOOLEAN
436 By default, TCP saves various connection metrics in the route cache
437 when the connection closes, so that connections established in the
438 near future can use these to set initial conditions. Usually, this
439 increases overall performance, but may sometimes cause performance
440 degradation. If set, TCP will not cache metrics on closing
443 tcp_orphan_retries - INTEGER
444 This value influences the timeout of a locally closed TCP connection,
445 when RTO retransmissions remain unacknowledged.
446 See tcp_retries2 for more details.
448 The default value is 8.
449 If your machine is a loaded WEB server,
450 you should think about lowering this value, such sockets
451 may consume significant resources. Cf. tcp_max_orphans.
453 tcp_recovery - INTEGER
454 This value is a bitmap to enable various experimental loss recovery
457 RACK: 0x1 enables the RACK loss detection for fast detection of lost
458 retransmissions and tail drops.
462 tcp_reordering - INTEGER
463 Initial reordering level of packets in a TCP stream.
464 TCP stack can then dynamically adjust flow reordering level
465 between this initial value and tcp_max_reordering
468 tcp_max_reordering - INTEGER
469 Maximal reordering level of packets in a TCP stream.
470 300 is a fairly conservative value, but you might increase it
471 if paths are using per packet load balancing (like bonding rr mode)
474 tcp_retrans_collapse - BOOLEAN
475 Bug-to-bug compatibility with some broken printers.
476 On retransmit try to send bigger packets to work around bugs in
479 tcp_retries1 - INTEGER
480 This value influences the time, after which TCP decides, that
481 something is wrong due to unacknowledged RTO retransmissions,
482 and reports this suspicion to the network layer.
483 See tcp_retries2 for more details.
485 RFC 1122 recommends at least 3 retransmissions, which is the
488 tcp_retries2 - INTEGER
489 This value influences the timeout of an alive TCP connection,
490 when RTO retransmissions remain unacknowledged.
491 Given a value of N, a hypothetical TCP connection following
492 exponential backoff with an initial RTO of TCP_RTO_MIN would
493 retransmit N times before killing the connection at the (N+1)th RTO.
495 The default value of 15 yields a hypothetical timeout of 924.6
496 seconds and is a lower bound for the effective timeout.
497 TCP will effectively time out at the first RTO which exceeds the
498 hypothetical timeout.
500 RFC 1122 recommends at least 100 seconds for the timeout,
501 which corresponds to a value of at least 8.
503 tcp_rfc1337 - BOOLEAN
504 If set, the TCP stack behaves conforming to RFC1337. If unset,
505 we are not conforming to RFC, but prevent TCP TIME_WAIT
509 tcp_rmem - vector of 3 INTEGERs: min, default, max
510 min: Minimal size of receive buffer used by TCP sockets.
511 It is guaranteed to each TCP socket, even under moderate memory
515 default: initial size of receive buffer used by TCP sockets.
516 This value overrides net.core.rmem_default used by other protocols.
517 Default: 87380 bytes. This value results in window of 65535 with
518 default setting of tcp_adv_win_scale and tcp_app_win:0 and a bit
519 less for default tcp_app_win. See below about these variables.
521 max: maximal size of receive buffer allowed for automatically
522 selected receiver buffers for TCP socket. This value does not override
523 net.core.rmem_max. Calling setsockopt() with SO_RCVBUF disables
524 automatic tuning of that socket's receive buffer size, in which
525 case this value is ignored.
526 Default: between 87380B and 6MB, depending on RAM size.
529 Enable select acknowledgments (SACKS).
531 tcp_slow_start_after_idle - BOOLEAN
532 If set, provide RFC2861 behavior and time out the congestion
533 window after an idle period. An idle period is defined at
534 the current RTO. If unset, the congestion window will not
535 be timed out after an idle period.
539 Use the Host requirements interpretation of the TCP urgent pointer field.
540 Most hosts use the older BSD interpretation, so if you turn this on
541 Linux might not communicate correctly with them.
544 tcp_synack_retries - INTEGER
545 Number of times SYNACKs for a passive TCP connection attempt will
546 be retransmitted. Should not be higher than 255. Default value
547 is 5, which corresponds to 31seconds till the last retransmission
548 with the current initial RTO of 1second. With this the final timeout
549 for a passive TCP connection will happen after 63seconds.
551 tcp_syncookies - BOOLEAN
552 Only valid when the kernel was compiled with CONFIG_SYN_COOKIES
553 Send out syncookies when the syn backlog queue of a socket
554 overflows. This is to prevent against the common 'SYN flood attack'
557 Note, that syncookies is fallback facility.
558 It MUST NOT be used to help highly loaded servers to stand
559 against legal connection rate. If you see SYN flood warnings
560 in your logs, but investigation shows that they occur
561 because of overload with legal connections, you should tune
562 another parameters until this warning disappear.
563 See: tcp_max_syn_backlog, tcp_synack_retries, tcp_abort_on_overflow.
565 syncookies seriously violate TCP protocol, do not allow
566 to use TCP extensions, can result in serious degradation
567 of some services (f.e. SMTP relaying), visible not by you,
568 but your clients and relays, contacting you. While you see
569 SYN flood warnings in logs not being really flooded, your server
570 is seriously misconfigured.
572 If you want to test which effects syncookies have to your
573 network connections you can set this knob to 2 to enable
574 unconditionally generation of syncookies.
576 tcp_fastopen - INTEGER
577 Enable TCP Fast Open (RFC7413) to send and accept data in the opening
580 The client support is enabled by flag 0x1 (on by default). The client
581 then must use sendmsg() or sendto() with the MSG_FASTOPEN flag,
582 rather than connect() to send data in SYN.
584 The server support is enabled by flag 0x2 (off by default). Then
585 either enable for all listeners with another flag (0x400) or
586 enable individual listeners via TCP_FASTOPEN socket option with
587 the option value being the length of the syn-data backlog.
589 The values (bitmap) are
590 0x1: (client) enables sending data in the opening SYN on the client.
591 0x2: (server) enables the server support, i.e., allowing data in
592 a SYN packet to be accepted and passed to the
593 application before 3-way handshake finishes.
594 0x4: (client) send data in the opening SYN regardless of cookie
595 availability and without a cookie option.
596 0x200: (server) accept data-in-SYN w/o any cookie option present.
597 0x400: (server) enable all listeners to support Fast Open by
598 default without explicit TCP_FASTOPEN socket option.
602 Note that that additional client or server features are only
603 effective if the basic support (0x1 and 0x2) are enabled respectively.
605 tcp_syn_retries - INTEGER
606 Number of times initial SYNs for an active TCP connection attempt
607 will be retransmitted. Should not be higher than 127. Default value
608 is 6, which corresponds to 63seconds till the last retransmission
609 with the current initial RTO of 1second. With this the final timeout
610 for an active TCP connection attempt will happen after 127seconds.
612 tcp_timestamps - INTEGER
613 Enable timestamps as defined in RFC1323.
615 1: Enable timestamps as defined in RFC1323 and use random offset for
616 each connection rather than only using the current time.
617 2: Like 1, but without random offsets.
620 tcp_min_tso_segs - INTEGER
621 Minimal number of segments per TSO frame.
622 Since linux-3.12, TCP does an automatic sizing of TSO frames,
623 depending on flow rate, instead of filling 64Kbytes packets.
624 For specific usages, it's possible to force TCP to build big
625 TSO frames. Note that TCP stack might split too big TSO packets
626 if available window is too small.
629 tcp_pacing_ss_ratio - INTEGER
630 sk->sk_pacing_rate is set by TCP stack using a ratio applied
631 to current rate. (current_rate = cwnd * mss / srtt)
632 If TCP is in slow start, tcp_pacing_ss_ratio is applied
633 to let TCP probe for bigger speeds, assuming cwnd can be
634 doubled every other RTT.
637 tcp_pacing_ca_ratio - INTEGER
638 sk->sk_pacing_rate is set by TCP stack using a ratio applied
639 to current rate. (current_rate = cwnd * mss / srtt)
640 If TCP is in congestion avoidance phase, tcp_pacing_ca_ratio
641 is applied to conservatively probe for bigger throughput.
644 tcp_tso_win_divisor - INTEGER
645 This allows control over what percentage of the congestion window
646 can be consumed by a single TSO frame.
647 The setting of this parameter is a choice between burstiness and
648 building larger TSO frames.
651 tcp_tw_reuse - BOOLEAN
652 Allow to reuse TIME-WAIT sockets for new connections when it is
653 safe from protocol viewpoint. Default value is 0.
654 It should not be changed without advice/request of technical
657 tcp_window_scaling - BOOLEAN
658 Enable window scaling as defined in RFC1323.
660 tcp_wmem - vector of 3 INTEGERs: min, default, max
661 min: Amount of memory reserved for send buffers for TCP sockets.
662 Each TCP socket has rights to use it due to fact of its birth.
665 default: initial size of send buffer used by TCP sockets. This
666 value overrides net.core.wmem_default used by other protocols.
667 It is usually lower than net.core.wmem_default.
670 max: Maximal amount of memory allowed for automatically tuned
671 send buffers for TCP sockets. This value does not override
672 net.core.wmem_max. Calling setsockopt() with SO_SNDBUF disables
673 automatic tuning of that socket's send buffer size, in which case
674 this value is ignored.
675 Default: between 64K and 4MB, depending on RAM size.
677 tcp_notsent_lowat - UNSIGNED INTEGER
678 A TCP socket can control the amount of unsent bytes in its write queue,
679 thanks to TCP_NOTSENT_LOWAT socket option. poll()/select()/epoll()
680 reports POLLOUT events if the amount of unsent bytes is below a per
681 socket value, and if the write queue is not full. sendmsg() will
682 also not add new buffers if the limit is hit.
684 This global variable controls the amount of unsent data for
685 sockets not using TCP_NOTSENT_LOWAT. For these sockets, a change
686 to the global variable has immediate effect.
688 Default: UINT_MAX (0xFFFFFFFF)
690 tcp_workaround_signed_windows - BOOLEAN
691 If set, assume no receipt of a window scaling option means the
692 remote TCP is broken and treats the window as a signed quantity.
693 If unset, assume the remote TCP is not broken even if we do
694 not receive a window scaling option from them.
697 tcp_thin_linear_timeouts - BOOLEAN
698 Enable dynamic triggering of linear timeouts for thin streams.
699 If set, a check is performed upon retransmission by timeout to
700 determine if the stream is thin (less than 4 packets in flight).
701 As long as the stream is found to be thin, up to 6 linear
702 timeouts may be performed before exponential backoff mode is
703 initiated. This improves retransmission latency for
704 non-aggressive thin streams, often found to be time-dependent.
705 For more information on thin streams, see
706 Documentation/networking/tcp-thin.txt
709 tcp_limit_output_bytes - INTEGER
710 Controls TCP Small Queue limit per tcp socket.
711 TCP bulk sender tends to increase packets in flight until it
712 gets losses notifications. With SNDBUF autotuning, this can
713 result in a large amount of packets queued in qdisc/device
714 on the local machine, hurting latency of other flows, for
715 typical pfifo_fast qdiscs.
716 tcp_limit_output_bytes limits the number of bytes on qdisc
717 or device to reduce artificial RTT/cwnd and reduce bufferbloat.
720 tcp_challenge_ack_limit - INTEGER
721 Limits number of Challenge ACK sent per second, as recommended
722 in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
727 udp_l3mdev_accept - BOOLEAN
728 Enabling this option allows a "global" bound socket to work
729 across L3 master domains (e.g., VRFs) with packets capable of
730 being received regardless of the L3 domain in which they
731 originated. Only valid when the kernel was compiled with
732 CONFIG_NET_L3_MASTER_DEV.
734 udp_mem - vector of 3 INTEGERs: min, pressure, max
735 Number of pages allowed for queueing by all UDP sockets.
737 min: Below this number of pages UDP is not bothered about its
738 memory appetite. When amount of memory allocated by UDP exceeds
739 this number, UDP starts to moderate memory usage.
741 pressure: This value was introduced to follow format of tcp_mem.
743 max: Number of pages allowed for queueing by all UDP sockets.
745 Default is calculated at boot time from amount of available memory.
747 udp_rmem_min - INTEGER
748 Minimal size of receive buffer used by UDP sockets in moderation.
749 Each UDP socket is able to use the size for receiving data, even if
750 total pages of UDP sockets exceed udp_mem pressure. The unit is byte.
753 udp_wmem_min - INTEGER
754 Minimal size of send buffer used by UDP sockets in moderation.
755 Each UDP socket is able to use the size for sending data, even if
756 total pages of UDP sockets exceed udp_mem pressure. The unit is byte.
761 cipso_cache_enable - BOOLEAN
762 If set, enable additions to and lookups from the CIPSO label mapping
763 cache. If unset, additions are ignored and lookups always result in a
764 miss. However, regardless of the setting the cache is still
765 invalidated when required when means you can safely toggle this on and
766 off and the cache will always be "safe".
769 cipso_cache_bucket_size - INTEGER
770 The CIPSO label cache consists of a fixed size hash table with each
771 hash bucket containing a number of cache entries. This variable limits
772 the number of entries in each hash bucket; the larger the value the
773 more CIPSO label mappings that can be cached. When the number of
774 entries in a given hash bucket reaches this limit adding new entries
775 causes the oldest entry in the bucket to be removed to make room.
778 cipso_rbm_optfmt - BOOLEAN
779 Enable the "Optimized Tag 1 Format" as defined in section 3.4.2.6 of
780 the CIPSO draft specification (see Documentation/netlabel for details).
781 This means that when set the CIPSO tag will be padded with empty
782 categories in order to make the packet data 32-bit aligned.
785 cipso_rbm_structvalid - BOOLEAN
786 If set, do a very strict check of the CIPSO option when
787 ip_options_compile() is called. If unset, relax the checks done during
788 ip_options_compile(). Either way is "safe" as errors are caught else
789 where in the CIPSO processing code but setting this to 0 (False) should
790 result in less work (i.e. it should be faster) but could cause problems
791 with other implementations that require strict checking.
796 ip_local_port_range - 2 INTEGERS
797 Defines the local port range that is used by TCP and UDP to
798 choose the local port. The first number is the first, the
799 second the last local port number.
800 If possible, it is better these numbers have different parity.
801 (one even and one odd values)
802 The default values are 32768 and 60999 respectively.
804 ip_local_reserved_ports - list of comma separated ranges
805 Specify the ports which are reserved for known third-party
806 applications. These ports will not be used by automatic port
807 assignments (e.g. when calling connect() or bind() with port
808 number 0). Explicit port allocation behavior is unchanged.
810 The format used for both input and output is a comma separated
811 list of ranges (e.g. "1,2-4,10-10" for ports 1, 2, 3, 4 and
812 10). Writing to the file will clear all previously reserved
813 ports and update the current list with the one given in the
816 Note that ip_local_port_range and ip_local_reserved_ports
817 settings are independent and both are considered by the kernel
818 when determining which ports are available for automatic port
821 You can reserve ports which are not in the current
822 ip_local_port_range, e.g.:
824 $ cat /proc/sys/net/ipv4/ip_local_port_range
826 $ cat /proc/sys/net/ipv4/ip_local_reserved_ports
829 although this is redundant. However such a setting is useful
830 if later the port range is changed to a value that will
831 include the reserved ports.
835 ip_unprivileged_port_start - INTEGER
836 This is a per-namespace sysctl. It defines the first
837 unprivileged port in the network namespace. Privileged ports
838 require root or CAP_NET_BIND_SERVICE in order to bind to them.
839 To disable all privileged ports, set this to 0. It may not
840 overlap with the ip_local_reserved_ports range.
844 ip_nonlocal_bind - BOOLEAN
845 If set, allows processes to bind() to non-local IP addresses,
846 which can be quite useful - but may break some applications.
850 If set non-zero, enables support for dynamic addresses.
851 If set to a non-zero value larger than 1, a kernel log
852 message will be printed when dynamic address rewriting
856 ip_early_demux - BOOLEAN
857 Optimize input packet processing down to one demux for
858 certain kinds of local sockets. Currently we only do this
859 for established TCP sockets.
861 It may add an additional cost for pure routing workloads that
862 reduces overall throughput, in such case you should disable it.
865 icmp_echo_ignore_all - BOOLEAN
866 If set non-zero, then the kernel will ignore all ICMP ECHO
870 icmp_echo_ignore_broadcasts - BOOLEAN
871 If set non-zero, then the kernel will ignore all ICMP ECHO and
872 TIMESTAMP requests sent to it via broadcast/multicast.
875 icmp_ratelimit - INTEGER
876 Limit the maximal rates for sending ICMP packets whose type matches
877 icmp_ratemask (see below) to specific targets.
878 0 to disable any limiting,
879 otherwise the minimal space between responses in milliseconds.
880 Note that another sysctl, icmp_msgs_per_sec limits the number
881 of ICMP packets sent on all targets.
884 icmp_msgs_per_sec - INTEGER
885 Limit maximal number of ICMP packets sent per second from this host.
886 Only messages whose type matches icmp_ratemask (see below) are
887 controlled by this limit.
890 icmp_msgs_burst - INTEGER
891 icmp_msgs_per_sec controls number of ICMP packets sent per second,
892 while icmp_msgs_burst controls the burst size of these packets.
895 icmp_ratemask - INTEGER
896 Mask made of ICMP types for which rates are being limited.
897 Significant bits: IHGFEDCBA9876543210
898 Default mask: 0000001100000011000 (6168)
900 Bit definitions (see include/linux/icmp.h):
902 3 Destination Unreachable *
907 C Parameter Problem *
912 H Address Mask Request
915 * These are rate limited by default (see default mask above)
917 icmp_ignore_bogus_error_responses - BOOLEAN
918 Some routers violate RFC1122 by sending bogus responses to broadcast
919 frames. Such violations are normally logged via a kernel warning.
920 If this is set to TRUE, the kernel will not give such warnings, which
921 will avoid log file clutter.
924 icmp_errors_use_inbound_ifaddr - BOOLEAN
926 If zero, icmp error messages are sent with the primary address of
927 the exiting interface.
929 If non-zero, the message will be sent with the primary address of
930 the interface that received the packet that caused the icmp error.
931 This is the behaviour network many administrators will expect from
932 a router. And it can make debugging complicated network layouts
935 Note that if no primary address exists for the interface selected,
936 then the primary address of the first non-loopback interface that
937 has one will be used regardless of this setting.
941 igmp_max_memberships - INTEGER
942 Change the maximum number of multicast groups we can subscribe to.
945 Theoretical maximum value is bounded by having to send a membership
946 report in a single datagram (i.e. the report can't span multiple
947 datagrams, or risk confusing the switch and leaving groups you don't
950 The number of supported groups 'M' is bounded by the number of group
951 report entries you can fit into a single datagram of 65535 bytes.
953 M = 65536-sizeof (ip header)/(sizeof(Group record))
955 Group records are variable length, with a minimum of 12 bytes.
956 So net.ipv4.igmp_max_memberships should not be set higher than:
958 (65536-24) / 12 = 5459
960 The value 5459 assumes no IP header options, so in practice
961 this number may be lower.
963 igmp_max_msf - INTEGER
964 Maximum number of addresses allowed in the source filter list for a
969 Controls the IGMP query robustness variable (see RFC2236 8.1).
970 Default: 2 (as specified by RFC2236 8.1)
971 Minimum: 1 (as specified by RFC6636 4.5)
973 force_igmp_version - INTEGER
974 0 - (default) No enforcement of a IGMP version, IGMPv1/v2 fallback
975 allowed. Will back to IGMPv3 mode again if all IGMPv1/v2 Querier
976 Present timer expires.
977 1 - Enforce to use IGMP version 1. Will also reply IGMPv1 report if
978 receive IGMPv2/v3 query.
979 2 - Enforce to use IGMP version 2. Will fallback to IGMPv1 if receive
980 IGMPv1 query message. Will reply report if receive IGMPv3 query.
981 3 - Enforce to use IGMP version 3. The same react with default 0.
983 Note: this is not the same with force_mld_version because IGMPv3 RFC3376
984 Security Considerations does not have clear description that we could
985 ignore other version messages completely as MLDv2 RFC3810. So make
986 this value as default 0 is recommended.
988 conf/interface/* changes special settings per interface (where
989 "interface" is the name of your network interface)
991 conf/all/* is special, changes the settings for all interfaces
993 log_martians - BOOLEAN
994 Log packets with impossible addresses to kernel log.
995 log_martians for the interface will be enabled if at least one of
996 conf/{all,interface}/log_martians is set to TRUE,
997 it will be disabled otherwise
999 accept_redirects - BOOLEAN
1000 Accept ICMP redirect messages.
1001 accept_redirects for the interface will be enabled if:
1002 - both conf/{all,interface}/accept_redirects are TRUE in the case
1003 forwarding for the interface is enabled
1005 - at least one of conf/{all,interface}/accept_redirects is TRUE in the
1006 case forwarding for the interface is disabled
1007 accept_redirects for the interface will be disabled otherwise
1011 forwarding - BOOLEAN
1012 Enable IP forwarding on this interface. This controls whether packets
1013 received _on_ this interface can be forwarded.
1015 mc_forwarding - BOOLEAN
1016 Do multicast routing. The kernel needs to be compiled with CONFIG_MROUTE
1017 and a multicast routing daemon is required.
1018 conf/all/mc_forwarding must also be set to TRUE to enable multicast
1019 routing for the interface
1022 Integer value used to differentiate the devices by the medium they
1023 are attached to. Two devices can have different id values when
1024 the broadcast packets are received only on one of them.
1025 The default value 0 means that the device is the only interface
1026 to its medium, value of -1 means that medium is not known.
1028 Currently, it is used to change the proxy_arp behavior:
1029 the proxy_arp feature is enabled for packets forwarded between
1030 two devices attached to different media.
1034 proxy_arp for the interface will be enabled if at least one of
1035 conf/{all,interface}/proxy_arp is set to TRUE,
1036 it will be disabled otherwise
1038 proxy_arp_pvlan - BOOLEAN
1039 Private VLAN proxy arp.
1040 Basically allow proxy arp replies back to the same interface
1041 (from which the ARP request/solicitation was received).
1043 This is done to support (ethernet) switch features, like RFC
1044 3069, where the individual ports are NOT allowed to
1045 communicate with each other, but they are allowed to talk to
1046 the upstream router. As described in RFC 3069, it is possible
1047 to allow these hosts to communicate through the upstream
1048 router by proxy_arp'ing. Don't need to be used together with
1051 This technology is known by different names:
1052 In RFC 3069 it is called VLAN Aggregation.
1053 Cisco and Allied Telesyn call it Private VLAN.
1054 Hewlett-Packard call it Source-Port filtering or port-isolation.
1055 Ericsson call it MAC-Forced Forwarding (RFC Draft).
1057 shared_media - BOOLEAN
1058 Send(router) or accept(host) RFC1620 shared media redirects.
1059 Overrides secure_redirects.
1060 shared_media for the interface will be enabled if at least one of
1061 conf/{all,interface}/shared_media is set to TRUE,
1062 it will be disabled otherwise
1065 secure_redirects - BOOLEAN
1066 Accept ICMP redirect messages only to gateways listed in the
1067 interface's current gateway list. Even if disabled, RFC1122 redirect
1069 Overridden by shared_media.
1070 secure_redirects for the interface will be enabled if at least one of
1071 conf/{all,interface}/secure_redirects is set to TRUE,
1072 it will be disabled otherwise
1075 send_redirects - BOOLEAN
1076 Send redirects, if router.
1077 send_redirects for the interface will be enabled if at least one of
1078 conf/{all,interface}/send_redirects is set to TRUE,
1079 it will be disabled otherwise
1082 bootp_relay - BOOLEAN
1083 Accept packets with source address 0.b.c.d destined
1084 not to this host as local ones. It is supposed, that
1085 BOOTP relay daemon will catch and forward such packets.
1086 conf/all/bootp_relay must also be set to TRUE to enable BOOTP relay
1089 Not Implemented Yet.
1091 accept_source_route - BOOLEAN
1092 Accept packets with SRR option.
1093 conf/all/accept_source_route must also be set to TRUE to accept packets
1094 with SRR option on the interface
1095 default TRUE (router)
1098 accept_local - BOOLEAN
1099 Accept packets with local source addresses. In combination with
1100 suitable routing, this can be used to direct packets between two
1101 local interfaces over the wire and have them accepted properly.
1104 route_localnet - BOOLEAN
1105 Do not consider loopback addresses as martian source or destination
1106 while routing. This enables the use of 127/8 for local routing purposes.
1110 0 - No source validation.
1111 1 - Strict mode as defined in RFC3704 Strict Reverse Path
1112 Each incoming packet is tested against the FIB and if the interface
1113 is not the best reverse path the packet check will fail.
1114 By default failed packets are discarded.
1115 2 - Loose mode as defined in RFC3704 Loose Reverse Path
1116 Each incoming packet's source address is also tested against the FIB
1117 and if the source address is not reachable via any interface
1118 the packet check will fail.
1120 Current recommended practice in RFC3704 is to enable strict mode
1121 to prevent IP spoofing from DDos attacks. If using asymmetric routing
1122 or other complicated routing, then loose mode is recommended.
1124 The max value from conf/{all,interface}/rp_filter is used
1125 when doing source validation on the {interface}.
1127 Default value is 0. Note that some distributions enable it
1130 arp_filter - BOOLEAN
1131 1 - Allows you to have multiple network interfaces on the same
1132 subnet, and have the ARPs for each interface be answered
1133 based on whether or not the kernel would route a packet from
1134 the ARP'd IP out that interface (therefore you must use source
1135 based routing for this to work). In other words it allows control
1136 of which cards (usually 1) will respond to an arp request.
1138 0 - (default) The kernel can respond to arp requests with addresses
1139 from other interfaces. This may seem wrong but it usually makes
1140 sense, because it increases the chance of successful communication.
1141 IP addresses are owned by the complete host on Linux, not by
1142 particular interfaces. Only for more complex setups like load-
1143 balancing, does this behaviour cause problems.
1145 arp_filter for the interface will be enabled if at least one of
1146 conf/{all,interface}/arp_filter is set to TRUE,
1147 it will be disabled otherwise
1149 arp_announce - INTEGER
1150 Define different restriction levels for announcing the local
1151 source IP address from IP packets in ARP requests sent on
1153 0 - (default) Use any local address, configured on any interface
1154 1 - Try to avoid local addresses that are not in the target's
1155 subnet for this interface. This mode is useful when target
1156 hosts reachable via this interface require the source IP
1157 address in ARP requests to be part of their logical network
1158 configured on the receiving interface. When we generate the
1159 request we will check all our subnets that include the
1160 target IP and will preserve the source address if it is from
1161 such subnet. If there is no such subnet we select source
1162 address according to the rules for level 2.
1163 2 - Always use the best local address for this target.
1164 In this mode we ignore the source address in the IP packet
1165 and try to select local address that we prefer for talks with
1166 the target host. Such local address is selected by looking
1167 for primary IP addresses on all our subnets on the outgoing
1168 interface that include the target IP address. If no suitable
1169 local address is found we select the first local address
1170 we have on the outgoing interface or on all other interfaces,
1171 with the hope we will receive reply for our request and
1172 even sometimes no matter the source IP address we announce.
1174 The max value from conf/{all,interface}/arp_announce is used.
1176 Increasing the restriction level gives more chance for
1177 receiving answer from the resolved target while decreasing
1178 the level announces more valid sender's information.
1180 arp_ignore - INTEGER
1181 Define different modes for sending replies in response to
1182 received ARP requests that resolve local target IP addresses:
1183 0 - (default): reply for any local target IP address, configured
1185 1 - reply only if the target IP address is local address
1186 configured on the incoming interface
1187 2 - reply only if the target IP address is local address
1188 configured on the incoming interface and both with the
1189 sender's IP address are part from same subnet on this interface
1190 3 - do not reply for local addresses configured with scope host,
1191 only resolutions for global and link addresses are replied
1193 8 - do not reply for all local addresses
1195 The max value from conf/{all,interface}/arp_ignore is used
1196 when ARP request is received on the {interface}
1198 arp_notify - BOOLEAN
1199 Define mode for notification of address and device changes.
1200 0 - (default): do nothing
1201 1 - Generate gratuitous arp requests when device is brought up
1202 or hardware address changes.
1204 arp_accept - BOOLEAN
1205 Define behavior for gratuitous ARP frames who's IP is not
1206 already present in the ARP table:
1207 0 - don't create new entries in the ARP table
1208 1 - create new entries in the ARP table
1210 Both replies and requests type gratuitous arp will trigger the
1211 ARP table to be updated, if this setting is on.
1213 If the ARP table already contains the IP address of the
1214 gratuitous arp frame, the arp table will be updated regardless
1215 if this setting is on or off.
1217 mcast_solicit - INTEGER
1218 The maximum number of multicast probes in INCOMPLETE state,
1219 when the associated hardware address is unknown. Defaults
1222 ucast_solicit - INTEGER
1223 The maximum number of unicast probes in PROBE state, when
1224 the hardware address is being reconfirmed. Defaults to 3.
1226 app_solicit - INTEGER
1227 The maximum number of probes to send to the user space ARP daemon
1228 via netlink before dropping back to multicast probes (see
1229 mcast_resolicit). Defaults to 0.
1231 mcast_resolicit - INTEGER
1232 The maximum number of multicast probes after unicast and
1233 app probes in PROBE state. Defaults to 0.
1235 disable_policy - BOOLEAN
1236 Disable IPSEC policy (SPD) for this interface
1238 disable_xfrm - BOOLEAN
1239 Disable IPSEC encryption on this interface, whatever the policy
1241 igmpv2_unsolicited_report_interval - INTEGER
1242 The interval in milliseconds in which the next unsolicited
1243 IGMPv1 or IGMPv2 report retransmit will take place.
1244 Default: 10000 (10 seconds)
1246 igmpv3_unsolicited_report_interval - INTEGER
1247 The interval in milliseconds in which the next unsolicited
1248 IGMPv3 report retransmit will take place.
1249 Default: 1000 (1 seconds)
1251 promote_secondaries - BOOLEAN
1252 When a primary IP address is removed from this interface
1253 promote a corresponding secondary IP address instead of
1254 removing all the corresponding secondary IP addresses.
1256 drop_unicast_in_l2_multicast - BOOLEAN
1257 Drop any unicast IP packets that are received in link-layer
1258 multicast (or broadcast) frames.
1259 This behavior (for multicast) is actually a SHOULD in RFC
1260 1122, but is disabled by default for compatibility reasons.
1263 drop_gratuitous_arp - BOOLEAN
1264 Drop all gratuitous ARP frames, for example if there's a known
1265 good ARP proxy on the network and such frames need not be used
1266 (or in the case of 802.11, must not be used to prevent attacks.)
1271 Allows you to write a number, which can be used as required.
1274 xfrm4_gc_thresh - INTEGER
1275 The threshold at which we will start garbage collecting for IPv4
1276 destination cache entries. At twice this value the system will
1277 refuse new allocations. The value must be set below the flowcache
1278 limit (4096 * number of online cpus) to take effect.
1280 igmp_link_local_mcast_reports - BOOLEAN
1281 Enable IGMP reports for link local multicast groups in the
1286 kuznet@ms2.inr.ac.ru
1292 delon.nicolas@wanadoo.fr
1297 /proc/sys/net/ipv6/* Variables:
1299 IPv6 has no global variables such as tcp_*. tcp_* settings under ipv4/ also
1300 apply to IPv6 [XXX?].
1302 bindv6only - BOOLEAN
1303 Default value for IPV6_V6ONLY socket option,
1304 which restricts use of the IPv6 socket to IPv6 communication
1306 TRUE: disable IPv4-mapped address feature
1307 FALSE: enable IPv4-mapped address feature
1309 Default: FALSE (as specified in RFC3493)
1311 flowlabel_consistency - BOOLEAN
1312 Protect the consistency (and unicity) of flow label.
1313 You have to disable it to use IPV6_FL_F_REFLECT flag on the
1319 auto_flowlabels - INTEGER
1320 Automatically generate flow labels based on a flow hash of the
1321 packet. This allows intermediate devices, such as routers, to
1322 identify packet flows for mechanisms like Equal Cost Multipath
1323 Routing (see RFC 6438).
1324 0: automatic flow labels are completely disabled
1325 1: automatic flow labels are enabled by default, they can be
1326 disabled on a per socket basis using the IPV6_AUTOFLOWLABEL
1328 2: automatic flow labels are allowed, they may be enabled on a
1329 per socket basis using the IPV6_AUTOFLOWLABEL socket option
1330 3: automatic flow labels are enabled and enforced, they cannot
1331 be disabled by the socket option
1334 flowlabel_state_ranges - BOOLEAN
1335 Split the flow label number space into two ranges. 0-0x7FFFF is
1336 reserved for the IPv6 flow manager facility, 0x80000-0xFFFFF
1337 is reserved for stateless flow labels as described in RFC6437.
1342 anycast_src_echo_reply - BOOLEAN
1343 Controls the use of anycast addresses as source addresses for ICMPv6
1349 idgen_delay - INTEGER
1350 Controls the delay in seconds after which time to retry
1351 privacy stable address generation if a DAD conflict is
1353 Default: 1 (as specified in RFC7217)
1355 idgen_retries - INTEGER
1356 Controls the number of retries to generate a stable privacy
1357 address if a DAD conflict is detected.
1358 Default: 3 (as specified in RFC7217)
1361 Controls the MLD query robustness variable (see RFC3810 9.1).
1362 Default: 2 (as specified by RFC3810 9.1)
1363 Minimum: 1 (as specified by RFC6636 4.5)
1367 ip6frag_high_thresh - INTEGER
1368 Maximum memory used to reassemble IPv6 fragments. When
1369 ip6frag_high_thresh bytes of memory is allocated for this purpose,
1370 the fragment handler will toss packets until ip6frag_low_thresh
1373 ip6frag_low_thresh - INTEGER
1374 See ip6frag_high_thresh
1376 ip6frag_time - INTEGER
1377 Time in seconds to keep an IPv6 fragment in memory.
1380 Change the interface-specific default settings.
1384 Change all the interface-specific settings.
1386 [XXX: Other special features than forwarding?]
1388 conf/all/forwarding - BOOLEAN
1389 Enable global IPv6 forwarding between all interfaces.
1391 IPv4 and IPv6 work differently here; e.g. netfilter must be used
1392 to control which interfaces may forward packets and which not.
1394 This also sets all interfaces' Host/Router setting
1395 'forwarding' to the specified value. See below for details.
1397 This referred to as global forwarding.
1402 fwmark_reflect - BOOLEAN
1403 Controls the fwmark of kernel-generated IPv6 reply packets that are not
1404 associated with a socket for example, TCP RSTs or ICMPv6 echo replies).
1405 If unset, these packets have a fwmark of zero. If set, they have the
1406 fwmark of the packet they are replying to.
1410 Change special settings per interface.
1412 The functional behaviour for certain settings is different
1413 depending on whether local forwarding is enabled or not.
1416 Accept Router Advertisements; autoconfigure using them.
1418 It also determines whether or not to transmit Router
1419 Solicitations. If and only if the functional setting is to
1420 accept Router Advertisements, Router Solicitations will be
1423 Possible values are:
1424 0 Do not accept Router Advertisements.
1425 1 Accept Router Advertisements if forwarding is disabled.
1426 2 Overrule forwarding behaviour. Accept Router Advertisements
1427 even if forwarding is enabled.
1429 Functional default: enabled if local forwarding is disabled.
1430 disabled if local forwarding is enabled.
1432 accept_ra_defrtr - BOOLEAN
1433 Learn default router in Router Advertisement.
1435 Functional default: enabled if accept_ra is enabled.
1436 disabled if accept_ra is disabled.
1438 accept_ra_from_local - BOOLEAN
1439 Accept RA with source-address that is found on local machine
1440 if the RA is otherwise proper and able to be accepted.
1441 Default is to NOT accept these as it may be an un-intended
1445 enabled if accept_ra_from_local is enabled
1446 on a specific interface.
1447 disabled if accept_ra_from_local is disabled
1448 on a specific interface.
1450 accept_ra_min_hop_limit - INTEGER
1451 Minimum hop limit Information in Router Advertisement.
1453 Hop limit Information in Router Advertisement less than this
1454 variable shall be ignored.
1458 accept_ra_pinfo - BOOLEAN
1459 Learn Prefix Information in Router Advertisement.
1461 Functional default: enabled if accept_ra is enabled.
1462 disabled if accept_ra is disabled.
1464 accept_ra_rt_info_max_plen - INTEGER
1465 Maximum prefix length of Route Information in RA.
1467 Route Information w/ prefix larger than or equal to this
1468 variable shall be ignored.
1470 Functional default: 0 if accept_ra_rtr_pref is enabled.
1471 -1 if accept_ra_rtr_pref is disabled.
1473 accept_ra_rtr_pref - BOOLEAN
1474 Accept Router Preference in RA.
1476 Functional default: enabled if accept_ra is enabled.
1477 disabled if accept_ra is disabled.
1479 accept_ra_mtu - BOOLEAN
1480 Apply the MTU value specified in RA option 5 (RFC4861). If
1481 disabled, the MTU specified in the RA will be ignored.
1483 Functional default: enabled if accept_ra is enabled.
1484 disabled if accept_ra is disabled.
1486 accept_redirects - BOOLEAN
1489 Functional default: enabled if local forwarding is disabled.
1490 disabled if local forwarding is enabled.
1492 accept_source_route - INTEGER
1493 Accept source routing (routing extension header).
1495 >= 0: Accept only routing header type 2.
1496 < 0: Do not accept routing header.
1501 Autoconfigure addresses using Prefix Information in Router
1504 Functional default: enabled if accept_ra_pinfo is enabled.
1505 disabled if accept_ra_pinfo is disabled.
1507 dad_transmits - INTEGER
1508 The amount of Duplicate Address Detection probes to send.
1511 forwarding - INTEGER
1512 Configure interface-specific Host/Router behaviour.
1514 Note: It is recommended to have the same setting on all
1515 interfaces; mixed router/host scenarios are rather uncommon.
1517 Possible values are:
1518 0 Forwarding disabled
1519 1 Forwarding enabled
1523 By default, Host behaviour is assumed. This means:
1525 1. IsRouter flag is not set in Neighbour Advertisements.
1526 2. If accept_ra is TRUE (default), transmit Router
1528 3. If accept_ra is TRUE (default), accept Router
1529 Advertisements (and do autoconfiguration).
1530 4. If accept_redirects is TRUE (default), accept Redirects.
1534 If local forwarding is enabled, Router behaviour is assumed.
1535 This means exactly the reverse from the above:
1537 1. IsRouter flag is set in Neighbour Advertisements.
1538 2. Router Solicitations are not sent unless accept_ra is 2.
1539 3. Router Advertisements are ignored unless accept_ra is 2.
1540 4. Redirects are ignored.
1542 Default: 0 (disabled) if global forwarding is disabled (default),
1543 otherwise 1 (enabled).
1546 Default Hop Limit to set.
1550 Default Maximum Transfer Unit
1551 Default: 1280 (IPv6 required minimum)
1553 ip_nonlocal_bind - BOOLEAN
1554 If set, allows processes to bind() to non-local IPv6 addresses,
1555 which can be quite useful - but may break some applications.
1558 router_probe_interval - INTEGER
1559 Minimum interval (in seconds) between Router Probing described
1564 router_solicitation_delay - INTEGER
1565 Number of seconds to wait after interface is brought up
1566 before sending Router Solicitations.
1569 router_solicitation_interval - INTEGER
1570 Number of seconds to wait between Router Solicitations.
1573 router_solicitations - INTEGER
1574 Number of Router Solicitations to send until assuming no
1575 routers are present.
1578 use_oif_addrs_only - BOOLEAN
1579 When enabled, the candidate source addresses for destinations
1580 routed via this interface are restricted to the set of addresses
1581 configured on this interface (vis. RFC 6724, section 4).
1585 use_tempaddr - INTEGER
1586 Preference for Privacy Extensions (RFC3041).
1587 <= 0 : disable Privacy Extensions
1588 == 1 : enable Privacy Extensions, but prefer public
1589 addresses over temporary addresses.
1590 > 1 : enable Privacy Extensions and prefer temporary
1591 addresses over public addresses.
1592 Default: 0 (for most devices)
1593 -1 (for point-to-point devices and loopback devices)
1595 temp_valid_lft - INTEGER
1596 valid lifetime (in seconds) for temporary addresses.
1597 Default: 604800 (7 days)
1599 temp_prefered_lft - INTEGER
1600 Preferred lifetime (in seconds) for temporary addresses.
1601 Default: 86400 (1 day)
1603 keep_addr_on_down - INTEGER
1604 Keep all IPv6 addresses on an interface down event. If set static
1605 global addresses with no expiration time are not flushed.
1610 Default: 0 (addresses are removed)
1612 max_desync_factor - INTEGER
1613 Maximum value for DESYNC_FACTOR, which is a random value
1614 that ensures that clients don't synchronize with each
1615 other and generate new addresses at exactly the same time.
1616 value is in seconds.
1619 regen_max_retry - INTEGER
1620 Number of attempts before give up attempting to generate
1621 valid temporary addresses.
1624 max_addresses - INTEGER
1625 Maximum number of autoconfigured addresses per interface. Setting
1626 to zero disables the limitation. It is not recommended to set this
1627 value too large (or to zero) because it would be an easy way to
1628 crash the kernel by allowing too many addresses to be created.
1631 disable_ipv6 - BOOLEAN
1632 Disable IPv6 operation. If accept_dad is set to 2, this value
1633 will be dynamically set to TRUE if DAD fails for the link-local
1635 Default: FALSE (enable IPv6 operation)
1637 When this value is changed from 1 to 0 (IPv6 is being enabled),
1638 it will dynamically create a link-local address on the given
1639 interface and start Duplicate Address Detection, if necessary.
1641 When this value is changed from 0 to 1 (IPv6 is being disabled),
1642 it will dynamically delete all address on the given interface.
1644 accept_dad - INTEGER
1645 Whether to accept DAD (Duplicate Address Detection).
1647 1: Enable DAD (default)
1648 2: Enable DAD, and disable IPv6 operation if MAC-based duplicate
1649 link-local address has been found.
1651 force_tllao - BOOLEAN
1652 Enable sending the target link-layer address option even when
1653 responding to a unicast neighbor solicitation.
1656 Quoting from RFC 2461, section 4.4, Target link-layer address:
1658 "The option MUST be included for multicast solicitations in order to
1659 avoid infinite Neighbor Solicitation "recursion" when the peer node
1660 does not have a cache entry to return a Neighbor Advertisements
1661 message. When responding to unicast solicitations, the option can be
1662 omitted since the sender of the solicitation has the correct link-
1663 layer address; otherwise it would not have be able to send the unicast
1664 solicitation in the first place. However, including the link-layer
1665 address in this case adds little overhead and eliminates a potential
1666 race condition where the sender deletes the cached link-layer address
1667 prior to receiving a response to a previous solicitation."
1669 ndisc_notify - BOOLEAN
1670 Define mode for notification of address and device changes.
1671 0 - (default): do nothing
1672 1 - Generate unsolicited neighbour advertisements when device is brought
1673 up or hardware address changes.
1675 mldv1_unsolicited_report_interval - INTEGER
1676 The interval in milliseconds in which the next unsolicited
1677 MLDv1 report retransmit will take place.
1678 Default: 10000 (10 seconds)
1680 mldv2_unsolicited_report_interval - INTEGER
1681 The interval in milliseconds in which the next unsolicited
1682 MLDv2 report retransmit will take place.
1683 Default: 1000 (1 second)
1685 force_mld_version - INTEGER
1686 0 - (default) No enforcement of a MLD version, MLDv1 fallback allowed
1687 1 - Enforce to use MLD version 1
1688 2 - Enforce to use MLD version 2
1690 suppress_frag_ndisc - INTEGER
1691 Control RFC 6980 (Security Implications of IPv6 Fragmentation
1692 with IPv6 Neighbor Discovery) behavior:
1693 1 - (default) discard fragmented neighbor discovery packets
1694 0 - allow fragmented neighbor discovery packets
1696 optimistic_dad - BOOLEAN
1697 Whether to perform Optimistic Duplicate Address Detection (RFC 4429).
1698 0: disabled (default)
1701 use_optimistic - BOOLEAN
1702 If enabled, do not classify optimistic addresses as deprecated during
1703 source address selection. Preferred addresses will still be chosen
1704 before optimistic addresses, subject to other ranking in the source
1705 address selection algorithm.
1706 0: disabled (default)
1709 stable_secret - IPv6 address
1710 This IPv6 address will be used as a secret to generate IPv6
1711 addresses for link-local addresses and autoconfigured
1712 ones. All addresses generated after setting this secret will
1713 be stable privacy ones by default. This can be changed via the
1714 addrgenmode ip-link. conf/default/stable_secret is used as the
1715 secret for the namespace, the interface specific ones can
1716 overwrite that. Writes to conf/all/stable_secret are refused.
1718 It is recommended to generate this secret during installation
1719 of a system and keep it stable after that.
1721 By default the stable secret is unset.
1723 drop_unicast_in_l2_multicast - BOOLEAN
1724 Drop any unicast IPv6 packets that are received in link-layer
1725 multicast (or broadcast) frames.
1727 By default this is turned off.
1729 drop_unsolicited_na - BOOLEAN
1730 Drop all unsolicited neighbor advertisements, for example if there's
1731 a known good NA proxy on the network and such frames need not be used
1732 (or in the case of 802.11, must not be used to prevent attacks.)
1734 By default this is turned off.
1736 enhanced_dad - BOOLEAN
1737 Include a nonce option in the IPv6 neighbor solicitation messages used for
1738 duplicate address detection per RFC7527. A received DAD NS will only signal
1739 a duplicate address if the nonce is different. This avoids any false
1740 detection of duplicates due to loopback of the NS messages that we send.
1741 The nonce option will be sent on an interface unless both of
1742 conf/{all,interface}/enhanced_dad are set to FALSE.
1747 Limit the maximal rates for sending ICMPv6 packets.
1748 0 to disable any limiting,
1749 otherwise the minimal space between responses in milliseconds.
1752 xfrm6_gc_thresh - INTEGER
1753 The threshold at which we will start garbage collecting for IPv6
1754 destination cache entries. At twice this value the system will
1755 refuse new allocations. The value must be set below the flowcache
1756 limit (4096 * number of online cpus) to take effect.
1760 Pekka Savola <pekkas@netcore.fi>
1761 YOSHIFUJI Hideaki / USAGI Project <yoshfuji@linux-ipv6.org>
1764 /proc/sys/net/bridge/* Variables:
1766 bridge-nf-call-arptables - BOOLEAN
1767 1 : pass bridged ARP traffic to arptables' FORWARD chain.
1771 bridge-nf-call-iptables - BOOLEAN
1772 1 : pass bridged IPv4 traffic to iptables' chains.
1776 bridge-nf-call-ip6tables - BOOLEAN
1777 1 : pass bridged IPv6 traffic to ip6tables' chains.
1781 bridge-nf-filter-vlan-tagged - BOOLEAN
1782 1 : pass bridged vlan-tagged ARP/IP/IPv6 traffic to {arp,ip,ip6}tables.
1786 bridge-nf-filter-pppoe-tagged - BOOLEAN
1787 1 : pass bridged pppoe-tagged IP/IPv6 traffic to {ip,ip6}tables.
1791 bridge-nf-pass-vlan-input-dev - BOOLEAN
1792 1: if bridge-nf-filter-vlan-tagged is enabled, try to find a vlan
1793 interface on the bridge and set the netfilter input device to the vlan.
1794 This allows use of e.g. "iptables -i br0.1" and makes the REDIRECT
1795 target work with vlan-on-top-of-bridge interfaces. When no matching
1796 vlan interface is found, or this switch is off, the input device is
1797 set to the bridge interface.
1798 0: disable bridge netfilter vlan interface lookup.
1801 proc/sys/net/sctp/* Variables:
1803 addip_enable - BOOLEAN
1804 Enable or disable extension of Dynamic Address Reconfiguration
1805 (ADD-IP) functionality specified in RFC5061. This extension provides
1806 the ability to dynamically add and remove new addresses for the SCTP
1809 1: Enable extension.
1811 0: Disable extension.
1816 Enable or disable pf (pf is short for potentially failed) state. A value
1817 of pf_retrans > path_max_retrans also disables pf state. That is, one of
1818 both pf_enable and pf_retrans > path_max_retrans can disable pf state.
1819 Since pf_retrans and path_max_retrans can be changed by userspace
1820 application, sometimes user expects to disable pf state by the value of
1821 pf_retrans > path_max_retrans, but occasionally the value of pf_retrans
1822 or path_max_retrans is changed by the user application, this pf state is
1823 enabled. As such, it is necessary to add this to dynamically enable
1824 and disable pf state. See:
1825 https://datatracker.ietf.org/doc/draft-ietf-tsvwg-sctp-failover for
1834 addip_noauth_enable - BOOLEAN
1835 Dynamic Address Reconfiguration (ADD-IP) requires the use of
1836 authentication to protect the operations of adding or removing new
1837 addresses. This requirement is mandated so that unauthorized hosts
1838 would not be able to hijack associations. However, older
1839 implementations may not have implemented this requirement while
1840 allowing the ADD-IP extension. For reasons of interoperability,
1841 we provide this variable to control the enforcement of the
1842 authentication requirement.
1844 1: Allow ADD-IP extension to be used without authentication. This
1845 should only be set in a closed environment for interoperability
1846 with older implementations.
1848 0: Enforce the authentication requirement
1852 auth_enable - BOOLEAN
1853 Enable or disable Authenticated Chunks extension. This extension
1854 provides the ability to send and receive authenticated chunks and is
1855 required for secure operation of Dynamic Address Reconfiguration
1858 1: Enable this extension.
1859 0: Disable this extension.
1863 prsctp_enable - BOOLEAN
1864 Enable or disable the Partial Reliability extension (RFC3758) which
1865 is used to notify peers that a given DATA should no longer be expected.
1873 The limit of the number of new packets that can be initially sent. It
1874 controls how bursty the generated traffic can be.
1878 association_max_retrans - INTEGER
1879 Set the maximum number for retransmissions that an association can
1880 attempt deciding that the remote end is unreachable. If this value
1881 is exceeded, the association is terminated.
1885 max_init_retransmits - INTEGER
1886 The maximum number of retransmissions of INIT and COOKIE-ECHO chunks
1887 that an association will attempt before declaring the destination
1888 unreachable and terminating.
1892 path_max_retrans - INTEGER
1893 The maximum number of retransmissions that will be attempted on a given
1894 path. Once this threshold is exceeded, the path is considered
1895 unreachable, and new traffic will use a different path when the
1896 association is multihomed.
1900 pf_retrans - INTEGER
1901 The number of retransmissions that will be attempted on a given path
1902 before traffic is redirected to an alternate transport (should one
1903 exist). Note this is distinct from path_max_retrans, as a path that
1904 passes the pf_retrans threshold can still be used. Its only
1905 deprioritized when a transmission path is selected by the stack. This
1906 setting is primarily used to enable fast failover mechanisms without
1907 having to reduce path_max_retrans to a very low value. See:
1908 http://www.ietf.org/id/draft-nishida-tsvwg-sctp-failover-05.txt
1909 for details. Note also that a value of pf_retrans > path_max_retrans
1910 disables this feature. Since both pf_retrans and path_max_retrans can
1911 be changed by userspace application, a variable pf_enable is used to
1916 rto_initial - INTEGER
1917 The initial round trip timeout value in milliseconds that will be used
1918 in calculating round trip times. This is the initial time interval
1919 for retransmissions.
1924 The maximum value (in milliseconds) of the round trip timeout. This
1925 is the largest time interval that can elapse between retransmissions.
1930 The minimum value (in milliseconds) of the round trip timeout. This
1931 is the smallest time interval the can elapse between retransmissions.
1935 hb_interval - INTEGER
1936 The interval (in milliseconds) between HEARTBEAT chunks. These chunks
1937 are sent at the specified interval on idle paths to probe the state of
1938 a given path between 2 associations.
1942 sack_timeout - INTEGER
1943 The amount of time (in milliseconds) that the implementation will wait
1948 valid_cookie_life - INTEGER
1949 The default lifetime of the SCTP cookie (in milliseconds). The cookie
1950 is used during association establishment.
1954 cookie_preserve_enable - BOOLEAN
1955 Enable or disable the ability to extend the lifetime of the SCTP cookie
1956 that is used during the establishment phase of SCTP association
1958 1: Enable cookie lifetime extension.
1963 cookie_hmac_alg - STRING
1964 Select the hmac algorithm used when generating the cookie value sent by
1965 a listening sctp socket to a connecting client in the INIT-ACK chunk.
1970 Ability to assign md5 or sha1 as the selected alg is predicated on the
1971 configuration of those algorithms at build time (CONFIG_CRYPTO_MD5 and
1972 CONFIG_CRYPTO_SHA1).
1974 Default: Dependent on configuration. MD5 if available, else SHA1 if
1975 available, else none.
1977 rcvbuf_policy - INTEGER
1978 Determines if the receive buffer is attributed to the socket or to
1979 association. SCTP supports the capability to create multiple
1980 associations on a single socket. When using this capability, it is
1981 possible that a single stalled association that's buffering a lot
1982 of data may block other associations from delivering their data by
1983 consuming all of the receive buffer space. To work around this,
1984 the rcvbuf_policy could be set to attribute the receiver buffer space
1985 to each association instead of the socket. This prevents the described
1988 1: rcvbuf space is per association
1989 0: rcvbuf space is per socket
1993 sndbuf_policy - INTEGER
1994 Similar to rcvbuf_policy above, this applies to send buffer space.
1996 1: Send buffer is tracked per association
1997 0: Send buffer is tracked per socket.
2001 sctp_mem - vector of 3 INTEGERs: min, pressure, max
2002 Number of pages allowed for queueing by all SCTP sockets.
2004 min: Below this number of pages SCTP is not bothered about its
2005 memory appetite. When amount of memory allocated by SCTP exceeds
2006 this number, SCTP starts to moderate memory usage.
2008 pressure: This value was introduced to follow format of tcp_mem.
2010 max: Number of pages allowed for queueing by all SCTP sockets.
2012 Default is calculated at boot time from amount of available memory.
2014 sctp_rmem - vector of 3 INTEGERs: min, default, max
2015 Only the first value ("min") is used, "default" and "max" are
2018 min: Minimal size of receive buffer used by SCTP socket.
2019 It is guaranteed to each SCTP socket (but not association) even
2020 under moderate memory pressure.
2024 sctp_wmem - vector of 3 INTEGERs: min, default, max
2025 Currently this tunable has no effect.
2027 addr_scope_policy - INTEGER
2028 Control IPv4 address scoping - draft-stewart-tsvwg-sctp-ipv4-00
2030 0 - Disable IPv4 address scoping
2031 1 - Enable IPv4 address scoping
2032 2 - Follow draft but allow IPv4 private addresses
2033 3 - Follow draft but allow IPv4 link local addresses
2038 /proc/sys/net/core/*
2039 Please see: Documentation/sysctl/net.txt for descriptions of these entries.
2042 /proc/sys/net/unix/*
2043 max_dgram_qlen - INTEGER
2044 The maximum length of dgram socket receive queue
2051 /proc/sys/net/irda/*
2052 fast_poll_increase FIXME
2053 warn_noreply_time FIXME
2054 discovery_slots FIXME
2057 discovery_timeout FIXME
2058 lap_keepalive_time FIXME
2059 max_noreply_time FIXME
2060 max_tx_data_size FIXME
2062 min_tx_turn_time FIXME