1 .. SPDX-License-Identifier: GPL-2.0
7 /proc/sys/net/ipv4/* Variables
8 ==============================
11 - 0 - disabled (default)
14 Forward Packets between interfaces.
16 This variable is special, its change resets all configuration
17 parameters to their default state (RFC1122 for hosts, RFC1812
20 ip_default_ttl - INTEGER
21 Default value of TTL field (Time To Live) for outgoing (but not
22 forwarded) IP packets. Should be between 1 and 255 inclusive.
23 Default: 64 (as recommended by RFC1700)
25 ip_no_pmtu_disc - INTEGER
26 Disable Path MTU Discovery. If enabled in mode 1 and a
27 fragmentation-required ICMP is received, the PMTU to this
28 destination will be set to the smallest of the old MTU to
29 this destination and min_pmtu (see below). You will need
30 to raise min_pmtu to the smallest interface MTU on your system
31 manually if you want to avoid locally generated fragments.
33 In mode 2 incoming Path MTU Discovery messages will be
34 discarded. Outgoing frames are handled the same as in mode 1,
35 implicitly setting IP_PMTUDISC_DONT on every created socket.
37 Mode 3 is a hardened pmtu discover mode. The kernel will only
38 accept fragmentation-needed errors if the underlying protocol
39 can verify them besides a plain socket lookup. Current
40 protocols for which pmtu events will be honored are TCP, SCTP
41 and DCCP as they verify e.g. the sequence number or the
42 association. This mode should not be enabled globally but is
43 only intended to secure e.g. name servers in namespaces where
44 TCP path mtu must still work but path MTU information of other
45 protocols should be discarded. If enabled globally this mode
46 could break other protocols.
53 default 552 - minimum Path MTU. Unless this is changed manually,
54 each cached pmtu will never be lower than this setting.
56 ip_forward_use_pmtu - BOOLEAN
57 By default we don't trust protocol path MTUs while forwarding
58 because they could be easily forged and can lead to unwanted
59 fragmentation by the router.
60 You only need to enable this if you have user-space software
61 which tries to discover path mtus by itself and depends on the
62 kernel honoring this information. This is normally not the
72 fwmark_reflect - BOOLEAN
73 Controls the fwmark of kernel-generated IPv4 reply packets that are not
74 associated with a socket for example, TCP RSTs or ICMP echo replies).
75 If unset, these packets have a fwmark of zero. If set, they have the
76 fwmark of the packet they are replying to.
80 fib_multipath_use_neigh - BOOLEAN
81 Use status of existing neighbor entry when determining nexthop for
82 multipath routes. If disabled, neighbor information is not used and
83 packets could be directed to a failed nexthop. Only valid for kernels
84 built with CONFIG_IP_ROUTE_MULTIPATH enabled.
93 fib_multipath_hash_policy - INTEGER
94 Controls which hash policy to use for multipath routes. Only valid
95 for kernels built with CONFIG_IP_ROUTE_MULTIPATH enabled.
103 - 2 - Layer 3 or inner Layer 3 if present
104 - 3 - Custom multipath hash. Fields used for multipath hash calculation
105 are determined by fib_multipath_hash_fields sysctl
107 fib_multipath_hash_fields - UNSIGNED INTEGER
108 When fib_multipath_hash_policy is set to 3 (custom multipath hash), the
109 fields used for multipath hash calculation are determined by this
112 This value is a bitmask which enables various fields for multipath hash
117 ====== ============================
118 0x0001 Source IP address
119 0x0002 Destination IP address
121 0x0008 Unused (Flow Label)
123 0x0020 Destination port
124 0x0040 Inner source IP address
125 0x0080 Inner destination IP address
126 0x0100 Inner IP protocol
127 0x0200 Inner Flow Label
128 0x0400 Inner source port
129 0x0800 Inner destination port
130 ====== ============================
132 Default: 0x0007 (source IP, destination IP and IP protocol)
134 fib_sync_mem - UNSIGNED INTEGER
135 Amount of dirty memory from fib entries that can be backlogged before
136 synchronize_rcu is forced.
138 Default: 512kB Minimum: 64kB Maximum: 64MB
140 ip_forward_update_priority - INTEGER
141 Whether to update SKB priority from "TOS" field in IPv4 header after it
142 is forwarded. The new SKB priority is mapped from TOS field value
143 according to an rt_tos2priority table (see e.g. man tc-prio).
145 Default: 1 (Update priority.)
149 - 0 - Do not update priority.
150 - 1 - Update priority.
152 route/max_size - INTEGER
153 Maximum number of routes allowed in the kernel. Increase
154 this when using large numbers of interfaces and/or routes.
156 From linux kernel 3.6 onwards, this is deprecated for ipv4
157 as route cache is no longer used.
159 From linux kernel 6.3 onwards, this is deprecated for ipv6
160 as garbage collection manages cached route entries.
162 neigh/default/gc_thresh1 - INTEGER
163 Minimum number of entries to keep. Garbage collector will not
164 purge entries if there are fewer than this number.
168 neigh/default/gc_thresh2 - INTEGER
169 Threshold when garbage collector becomes more aggressive about
170 purging entries. Entries older than 5 seconds will be cleared
171 when over this number.
175 neigh/default/gc_thresh3 - INTEGER
176 Maximum number of non-PERMANENT neighbor entries allowed. Increase
177 this when using large numbers of interfaces and when communicating
178 with large numbers of directly-connected peers.
182 neigh/default/unres_qlen_bytes - INTEGER
183 The maximum number of bytes which may be used by packets
184 queued for each unresolved address by other network layers.
187 Setting negative value is meaningless and will return error.
189 Default: SK_WMEM_MAX, (same as net.core.wmem_default).
191 Exact value depends on architecture and kernel options,
192 but should be enough to allow queuing 256 packets
195 neigh/default/unres_qlen - INTEGER
196 The maximum number of packets which may be queued for each
197 unresolved address by other network layers.
199 (deprecated in linux 3.3) : use unres_qlen_bytes instead.
201 Prior to linux 3.3, the default value is 3 which may cause
202 unexpected packet loss. The current default value is calculated
203 according to default value of unres_qlen_bytes and true size of
208 neigh/default/interval_probe_time_ms - INTEGER
209 The probe interval for neighbor entries with NTF_MANAGED flag,
214 mtu_expires - INTEGER
215 Time, in seconds, that cached PMTU information is kept.
217 min_adv_mss - INTEGER
218 The advertised MSS depends on the first hop route MTU, but will
219 never be lower than this setting.
221 fib_notify_on_flag_change - INTEGER
222 Whether to emit RTM_NEWROUTE notifications whenever RTM_F_OFFLOAD/
223 RTM_F_TRAP/RTM_F_OFFLOAD_FAILED flags are changed.
225 After installing a route to the kernel, user space receives an
226 acknowledgment, which means the route was installed in the kernel,
227 but not necessarily in hardware.
228 It is also possible for a route already installed in hardware to change
229 its action and therefore its flags. For example, a host route that is
230 trapping packets can be "promoted" to perform decapsulation following
231 the installation of an IPinIP/VXLAN tunnel.
232 The notifications will indicate to user-space the state of the route.
234 Default: 0 (Do not emit notifications.)
238 - 0 - Do not emit notifications.
239 - 1 - Emit notifications.
240 - 2 - Emit notifications only for RTM_F_OFFLOAD_FAILED flag change.
244 ipfrag_high_thresh - LONG INTEGER
245 Maximum memory used to reassemble IP fragments.
247 ipfrag_low_thresh - LONG INTEGER
248 (Obsolete since linux-4.17)
249 Maximum memory used to reassemble IP fragments before the kernel
250 begins to remove incomplete fragment queues to free up resources.
251 The kernel still accepts new fragments for defragmentation.
253 ipfrag_time - INTEGER
254 Time in seconds to keep an IP fragment in memory.
256 ipfrag_max_dist - INTEGER
257 ipfrag_max_dist is a non-negative integer value which defines the
258 maximum "disorder" which is allowed among fragments which share a
259 common IP source address. Note that reordering of packets is
260 not unusual, but if a large number of fragments arrive from a source
261 IP address while a particular fragment queue remains incomplete, it
262 probably indicates that one or more fragments belonging to that queue
263 have been lost. When ipfrag_max_dist is positive, an additional check
264 is done on fragments before they are added to a reassembly queue - if
265 ipfrag_max_dist (or more) fragments have arrived from a particular IP
266 address between additions to any IP fragment queue using that source
267 address, it's presumed that one or more fragments in the queue are
268 lost. The existing fragment queue will be dropped, and a new one
269 started. An ipfrag_max_dist value of zero disables this check.
271 Using a very small value, e.g. 1 or 2, for ipfrag_max_dist can
272 result in unnecessarily dropping fragment queues when normal
273 reordering of packets occurs, which could lead to poor application
274 performance. Using a very large value, e.g. 50000, increases the
275 likelihood of incorrectly reassembling IP fragments that originate
276 from different IP datagrams, which could result in data corruption.
279 bc_forwarding - INTEGER
280 bc_forwarding enables the feature described in rfc1812#section-5.3.5.2
281 and rfc2644. It allows the router to forward directed broadcast.
282 To enable this feature, the 'all' entry and the input interface entry
289 inet_peer_threshold - INTEGER
290 The approximate size of the storage. Starting from this threshold
291 entries will be thrown aggressively. This threshold also determines
292 entries' time-to-live and time intervals between garbage collection
293 passes. More entries, less time-to-live, less GC interval.
295 inet_peer_minttl - INTEGER
296 Minimum time-to-live of entries. Should be enough to cover fragment
297 time-to-live on the reassembling side. This minimum time-to-live is
298 guaranteed if the pool size is less than inet_peer_threshold.
301 inet_peer_maxttl - INTEGER
302 Maximum time-to-live of entries. Unused entries will expire after
303 this period of time if there is no memory pressure on the pool (i.e.
304 when the number of entries in the pool is very small).
311 Limit of socket listen() backlog, known in userspace as SOMAXCONN.
312 Defaults to 4096. (Was 128 before linux-5.4)
313 See also tcp_max_syn_backlog for additional tuning for TCP sockets.
315 tcp_abort_on_overflow - BOOLEAN
316 If listening service is too slow to accept new connections,
317 reset them. Default state is FALSE. It means that if overflow
318 occurred due to a burst, connection will recover. Enable this
319 option _only_ if you are really sure that listening daemon
320 cannot be tuned to accept connections faster. Enabling this
321 option can harm clients of your server.
323 tcp_adv_win_scale - INTEGER
324 Obsolete since linux-6.6
325 Count buffering overhead as bytes/2^tcp_adv_win_scale
326 (if tcp_adv_win_scale > 0) or bytes-bytes/2^(-tcp_adv_win_scale),
329 Possible values are [-31, 31], inclusive.
333 tcp_allowed_congestion_control - STRING
334 Show/set the congestion control choices available to non-privileged
335 processes. The list is a subset of those listed in
336 tcp_available_congestion_control.
338 Default is "reno" and the default setting (tcp_congestion_control).
340 tcp_app_win - INTEGER
341 Reserve max(window/2^tcp_app_win, mss) of window for application
342 buffer. Value 0 is special, it means that nothing is reserved.
344 Possible values are [0, 31], inclusive.
348 tcp_autocorking - BOOLEAN
349 Enable TCP auto corking :
350 When applications do consecutive small write()/sendmsg() system calls,
351 we try to coalesce these small writes as much as possible, to lower
352 total amount of sent packets. This is done if at least one prior
353 packet for the flow is waiting in Qdisc queues or device transmit
354 queue. Applications can still use TCP_CORK for optimal behavior
355 when they know how/when to uncork their sockets.
359 tcp_available_congestion_control - STRING
360 Shows the available congestion control choices that are registered.
361 More congestion control algorithms may be available as modules,
364 tcp_base_mss - INTEGER
365 The initial value of search_low to be used by the packetization layer
366 Path MTU discovery (MTU probing). If MTU probing is enabled,
367 this is the initial MSS used by the connection.
369 tcp_mtu_probe_floor - INTEGER
370 If MTU probing is enabled this caps the minimum MSS used for search_low
375 tcp_min_snd_mss - INTEGER
376 TCP SYN and SYNACK messages usually advertise an ADVMSS option,
377 as described in RFC 1122 and RFC 6691.
379 If this ADVMSS option is smaller than tcp_min_snd_mss,
380 it is silently capped to tcp_min_snd_mss.
382 Default : 48 (at least 8 bytes of payload per segment)
384 tcp_congestion_control - STRING
385 Set the congestion control algorithm to be used for new
386 connections. The algorithm "reno" is always available, but
387 additional choices may be available based on kernel configuration.
388 Default is set as part of kernel configuration.
389 For passive connections, the listener congestion control choice
392 [see setsockopt(listenfd, SOL_TCP, TCP_CONGESTION, "name" ...) ]
395 Allows TCP to send "duplicate" SACKs.
397 tcp_early_retrans - INTEGER
398 Tail loss probe (TLP) converts RTOs occurring due to tail
399 losses into fast recovery (draft-ietf-tcpm-rack). Note that
400 TLP requires RACK to function properly (see tcp_recovery below)
410 Control use of Explicit Congestion Notification (ECN) by TCP.
411 ECN is used only when both ends of the TCP connection indicate
412 support for it. This feature is useful in avoiding losses due
413 to congestion by allowing supporting routers to signal
414 congestion before having to drop packets.
418 = =====================================================
419 0 Disable ECN. Neither initiate nor accept ECN.
420 1 Enable ECN when requested by incoming connections and
421 also request ECN on outgoing connection attempts.
422 2 Enable ECN when requested by incoming connections
423 but do not request ECN on outgoing connections.
424 = =====================================================
428 tcp_ecn_fallback - BOOLEAN
429 If the kernel detects that ECN connection misbehaves, enable fall
430 back to non-ECN. Currently, this knob implements the fallback
431 from RFC3168, section 6.1.1.1., but we reserve that in future,
432 additional detection mechanisms could be implemented under this
433 knob. The value is not used, if tcp_ecn or per route (or congestion
434 control) ECN settings are disabled.
436 Default: 1 (fallback enabled)
439 This is a legacy option, it has no effect anymore.
441 tcp_fin_timeout - INTEGER
442 The length of time an orphaned (no longer referenced by any
443 application) connection will remain in the FIN_WAIT_2 state
444 before it is aborted at the local end. While a perfectly
445 valid "receive only" state for an un-orphaned connection, an
446 orphaned connection in FIN_WAIT_2 state could otherwise wait
447 forever for the remote to close its end of the connection.
454 Enables Forward RTO-Recovery (F-RTO) defined in RFC5682.
455 F-RTO is an enhanced recovery algorithm for TCP retransmission
456 timeouts. It is particularly beneficial in networks where the
457 RTT fluctuates (e.g., wireless). F-RTO is sender-side only
458 modification. It does not require any support from the peer.
460 By default it's enabled with a non-zero value. 0 disables F-RTO.
462 tcp_fwmark_accept - BOOLEAN
463 If set, incoming connections to listening sockets that do not have a
464 socket mark will set the mark of the accepting socket to the fwmark of
465 the incoming SYN packet. This will cause all packets on that connection
466 (starting from the first SYNACK) to be sent with that fwmark. The
467 listening socket's mark is unchanged. Listening sockets that already
468 have a fwmark set via setsockopt(SOL_SOCKET, SO_MARK, ...) are
473 tcp_invalid_ratelimit - INTEGER
474 Limit the maximal rate for sending duplicate acknowledgments
475 in response to incoming TCP packets that are for an existing
476 connection but that are invalid due to any of these reasons:
478 (a) out-of-window sequence number,
479 (b) out-of-window acknowledgment number, or
480 (c) PAWS (Protection Against Wrapped Sequence numbers) check failure
482 This can help mitigate simple "ack loop" DoS attacks, wherein
483 a buggy or malicious middlebox or man-in-the-middle can
484 rewrite TCP header fields in manner that causes each endpoint
485 to think that the other is sending invalid TCP segments, thus
486 causing each side to send an unterminating stream of duplicate
487 acknowledgments for invalid segments.
489 Using 0 disables rate-limiting of dupacks in response to
490 invalid segments; otherwise this value specifies the minimal
491 space between sending such dupacks, in milliseconds.
493 Default: 500 (milliseconds).
495 tcp_keepalive_time - INTEGER
496 How often TCP sends out keepalive messages when keepalive is enabled.
499 tcp_keepalive_probes - INTEGER
500 How many keepalive probes TCP sends out, until it decides that the
501 connection is broken. Default value: 9.
503 tcp_keepalive_intvl - INTEGER
504 How frequently the probes are send out. Multiplied by
505 tcp_keepalive_probes it is time to kill not responding connection,
506 after probes started. Default value: 75sec i.e. connection
507 will be aborted after ~11 minutes of retries.
509 tcp_l3mdev_accept - BOOLEAN
510 Enables child sockets to inherit the L3 master device index.
511 Enabling this option allows a "global" listen socket to work
512 across L3 master domains (e.g., VRFs) with connected sockets
513 derived from the listen socket to be bound to the L3 domain in
514 which the packets originated. Only valid when the kernel was
515 compiled with CONFIG_NET_L3_MASTER_DEV.
517 Default: 0 (disabled)
519 tcp_low_latency - BOOLEAN
520 This is a legacy option, it has no effect anymore.
522 tcp_max_orphans - INTEGER
523 Maximal number of TCP sockets not attached to any user file handle,
524 held by system. If this number is exceeded orphaned connections are
525 reset immediately and warning is printed. This limit exists
526 only to prevent simple DoS attacks, you _must_ not rely on this
527 or lower the limit artificially, but rather increase it
528 (probably, after increasing installed memory),
529 if network conditions require more than default value,
530 and tune network services to linger and kill such states
531 more aggressively. Let me to remind again: each orphan eats
532 up to ~64K of unswappable memory.
534 tcp_max_syn_backlog - INTEGER
535 Maximal number of remembered connection requests (SYN_RECV),
536 which have not received an acknowledgment from connecting client.
538 This is a per-listener limit.
540 The minimal value is 128 for low memory machines, and it will
541 increase in proportion to the memory of machine.
543 If server suffers from overload, try increasing this number.
545 Remember to also check /proc/sys/net/core/somaxconn
546 A SYN_RECV request socket consumes about 304 bytes of memory.
548 tcp_max_tw_buckets - INTEGER
549 Maximal number of timewait sockets held by system simultaneously.
550 If this number is exceeded time-wait socket is immediately destroyed
551 and warning is printed. This limit exists only to prevent
552 simple DoS attacks, you _must_ not lower the limit artificially,
553 but rather increase it (probably, after increasing installed memory),
554 if network conditions require more than default value.
556 tcp_mem - vector of 3 INTEGERs: min, pressure, max
557 min: below this number of pages TCP is not bothered about its
560 pressure: when amount of memory allocated by TCP exceeds this number
561 of pages, TCP moderates its memory consumption and enters memory
562 pressure mode, which is exited when memory consumption falls
565 max: number of pages allowed for queueing by all TCP sockets.
567 Defaults are calculated at boot time from amount of available
570 tcp_min_rtt_wlen - INTEGER
571 The window length of the windowed min filter to track the minimum RTT.
572 A shorter window lets a flow more quickly pick up new (higher)
573 minimum RTT when it is moved to a longer path (e.g., due to traffic
574 engineering). A longer window makes the filter more resistant to RTT
575 inflations such as transient congestion. The unit is seconds.
577 Possible values: 0 - 86400 (1 day)
581 tcp_moderate_rcvbuf - BOOLEAN
582 If set, TCP performs receive buffer auto-tuning, attempting to
583 automatically size the buffer (no greater than tcp_rmem[2]) to
584 match the size required by the path for full throughput. Enabled by
587 tcp_mtu_probing - INTEGER
588 Controls TCP Packetization-Layer Path MTU Discovery. Takes three
592 - 1 - Disabled by default, enabled when an ICMP black hole detected
593 - 2 - Always enabled, use initial MSS of tcp_base_mss.
595 tcp_probe_interval - UNSIGNED INTEGER
596 Controls how often to start TCP Packetization-Layer Path MTU
597 Discovery reprobe. The default is reprobing every 10 minutes as
600 tcp_probe_threshold - INTEGER
601 Controls when TCP Packetization-Layer Path MTU Discovery probing
602 will stop in respect to the width of search range in bytes. Default
605 tcp_no_metrics_save - BOOLEAN
606 By default, TCP saves various connection metrics in the route cache
607 when the connection closes, so that connections established in the
608 near future can use these to set initial conditions. Usually, this
609 increases overall performance, but may sometimes cause performance
610 degradation. If set, TCP will not cache metrics on closing
613 tcp_no_ssthresh_metrics_save - BOOLEAN
614 Controls whether TCP saves ssthresh metrics in the route cache.
616 Default is 1, which disables ssthresh metrics.
618 tcp_orphan_retries - INTEGER
619 This value influences the timeout of a locally closed TCP connection,
620 when RTO retransmissions remain unacknowledged.
621 See tcp_retries2 for more details.
623 The default value is 8.
625 If your machine is a loaded WEB server,
626 you should think about lowering this value, such sockets
627 may consume significant resources. Cf. tcp_max_orphans.
629 tcp_recovery - INTEGER
630 This value is a bitmap to enable various experimental loss recovery
633 ========= =============================================================
634 RACK: 0x1 enables the RACK loss detection for fast detection of lost
635 retransmissions and tail drops. It also subsumes and disables
636 RFC6675 recovery for SACK connections.
638 RACK: 0x2 makes RACK's reordering window static (min_rtt/4).
640 RACK: 0x4 disables RACK's DUPACK threshold heuristic
641 ========= =============================================================
645 tcp_reflect_tos - BOOLEAN
646 For listening sockets, reuse the DSCP value of the initial SYN message
647 for outgoing packets. This allows to have both directions of a TCP
648 stream to use the same DSCP value, assuming DSCP remains unchanged for
649 the lifetime of the connection.
651 This options affects both IPv4 and IPv6.
653 Default: 0 (disabled)
655 tcp_reordering - INTEGER
656 Initial reordering level of packets in a TCP stream.
657 TCP stack can then dynamically adjust flow reordering level
658 between this initial value and tcp_max_reordering
662 tcp_max_reordering - INTEGER
663 Maximal reordering level of packets in a TCP stream.
664 300 is a fairly conservative value, but you might increase it
665 if paths are using per packet load balancing (like bonding rr mode)
669 tcp_retrans_collapse - BOOLEAN
670 Bug-to-bug compatibility with some broken printers.
671 On retransmit try to send bigger packets to work around bugs in
674 tcp_retries1 - INTEGER
675 This value influences the time, after which TCP decides, that
676 something is wrong due to unacknowledged RTO retransmissions,
677 and reports this suspicion to the network layer.
678 See tcp_retries2 for more details.
680 RFC 1122 recommends at least 3 retransmissions, which is the
683 tcp_retries2 - INTEGER
684 This value influences the timeout of an alive TCP connection,
685 when RTO retransmissions remain unacknowledged.
686 Given a value of N, a hypothetical TCP connection following
687 exponential backoff with an initial RTO of TCP_RTO_MIN would
688 retransmit N times before killing the connection at the (N+1)th RTO.
690 The default value of 15 yields a hypothetical timeout of 924.6
691 seconds and is a lower bound for the effective timeout.
692 TCP will effectively time out at the first RTO which exceeds the
693 hypothetical timeout.
695 RFC 1122 recommends at least 100 seconds for the timeout,
696 which corresponds to a value of at least 8.
698 tcp_rfc1337 - BOOLEAN
699 If set, the TCP stack behaves conforming to RFC1337. If unset,
700 we are not conforming to RFC, but prevent TCP TIME_WAIT
705 tcp_rmem - vector of 3 INTEGERs: min, default, max
706 min: Minimal size of receive buffer used by TCP sockets.
707 It is guaranteed to each TCP socket, even under moderate memory
712 default: initial size of receive buffer used by TCP sockets.
713 This value overrides net.core.rmem_default used by other protocols.
714 Default: 131072 bytes.
715 This value results in initial window of 65535.
717 max: maximal size of receive buffer allowed for automatically
718 selected receiver buffers for TCP socket. This value does not override
719 net.core.rmem_max. Calling setsockopt() with SO_RCVBUF disables
720 automatic tuning of that socket's receive buffer size, in which
721 case this value is ignored.
722 Default: between 131072 and 6MB, depending on RAM size.
725 Enable select acknowledgments (SACKS).
727 tcp_comp_sack_delay_ns - LONG INTEGER
728 TCP tries to reduce number of SACK sent, using a timer
729 based on 5% of SRTT, capped by this sysctl, in nano seconds.
730 The default is 1ms, based on TSO autosizing period.
732 Default : 1,000,000 ns (1 ms)
734 tcp_comp_sack_slack_ns - LONG INTEGER
735 This sysctl control the slack used when arming the
736 timer used by SACK compression. This gives extra time
737 for small RTT flows, and reduces system overhead by allowing
738 opportunistic reduction of timer interrupts.
740 Default : 100,000 ns (100 us)
742 tcp_comp_sack_nr - INTEGER
743 Max number of SACK that can be compressed.
744 Using 0 disables SACK compression.
748 tcp_backlog_ack_defer - BOOLEAN
749 If set, user thread processing socket backlog tries sending
750 one ACK for the whole queue. This helps to avoid potential
751 long latencies at end of a TCP socket syscall.
755 tcp_slow_start_after_idle - BOOLEAN
756 If set, provide RFC2861 behavior and time out the congestion
757 window after an idle period. An idle period is defined at
758 the current RTO. If unset, the congestion window will not
759 be timed out after an idle period.
764 Use the Host requirements interpretation of the TCP urgent pointer field.
765 Most hosts use the older BSD interpretation, so if you turn this on
766 Linux might not communicate correctly with them.
770 tcp_synack_retries - INTEGER
771 Number of times SYNACKs for a passive TCP connection attempt will
772 be retransmitted. Should not be higher than 255. Default value
773 is 5, which corresponds to 31seconds till the last retransmission
774 with the current initial RTO of 1second. With this the final timeout
775 for a passive TCP connection will happen after 63seconds.
777 tcp_syncookies - INTEGER
778 Only valid when the kernel was compiled with CONFIG_SYN_COOKIES
779 Send out syncookies when the syn backlog queue of a socket
780 overflows. This is to prevent against the common 'SYN flood attack'
783 Note, that syncookies is fallback facility.
784 It MUST NOT be used to help highly loaded servers to stand
785 against legal connection rate. If you see SYN flood warnings
786 in your logs, but investigation shows that they occur
787 because of overload with legal connections, you should tune
788 another parameters until this warning disappear.
789 See: tcp_max_syn_backlog, tcp_synack_retries, tcp_abort_on_overflow.
791 syncookies seriously violate TCP protocol, do not allow
792 to use TCP extensions, can result in serious degradation
793 of some services (f.e. SMTP relaying), visible not by you,
794 but your clients and relays, contacting you. While you see
795 SYN flood warnings in logs not being really flooded, your server
796 is seriously misconfigured.
798 If you want to test which effects syncookies have to your
799 network connections you can set this knob to 2 to enable
800 unconditionally generation of syncookies.
802 tcp_migrate_req - BOOLEAN
803 The incoming connection is tied to a specific listening socket when
804 the initial SYN packet is received during the three-way handshake.
805 When a listener is closed, in-flight request sockets during the
806 handshake and established sockets in the accept queue are aborted.
808 If the listener has SO_REUSEPORT enabled, other listeners on the
809 same port should have been able to accept such connections. This
810 option makes it possible to migrate such child sockets to another
811 listener after close() or shutdown().
813 The BPF_SK_REUSEPORT_SELECT_OR_MIGRATE type of eBPF program should
814 usually be used to define the policy to pick an alive listener.
815 Otherwise, the kernel will randomly pick an alive listener only if
816 this option is enabled.
818 Note that migration between listeners with different settings may
819 crash applications. Let's say migration happens from listener A to
820 B, and only B has TCP_SAVE_SYN enabled. B cannot read SYN data from
821 the requests migrated from A. To avoid such a situation, cancel
822 migration by returning SK_DROP in the type of eBPF program, or
827 tcp_fastopen - INTEGER
828 Enable TCP Fast Open (RFC7413) to send and accept data in the opening
831 The client support is enabled by flag 0x1 (on by default). The client
832 then must use sendmsg() or sendto() with the MSG_FASTOPEN flag,
833 rather than connect() to send data in SYN.
835 The server support is enabled by flag 0x2 (off by default). Then
836 either enable for all listeners with another flag (0x400) or
837 enable individual listeners via TCP_FASTOPEN socket option with
838 the option value being the length of the syn-data backlog.
840 The values (bitmap) are
842 ===== ======== ======================================================
843 0x1 (client) enables sending data in the opening SYN on the client.
844 0x2 (server) enables the server support, i.e., allowing data in
845 a SYN packet to be accepted and passed to the
846 application before 3-way handshake finishes.
847 0x4 (client) send data in the opening SYN regardless of cookie
848 availability and without a cookie option.
849 0x200 (server) accept data-in-SYN w/o any cookie option present.
850 0x400 (server) enable all listeners to support Fast Open by
851 default without explicit TCP_FASTOPEN socket option.
852 ===== ======== ======================================================
856 Note that additional client or server features are only
857 effective if the basic support (0x1 and 0x2) are enabled respectively.
859 tcp_fastopen_blackhole_timeout_sec - INTEGER
860 Initial time period in second to disable Fastopen on active TCP sockets
861 when a TFO firewall blackhole issue happens.
862 This time period will grow exponentially when more blackhole issues
863 get detected right after Fastopen is re-enabled and will reset to
864 initial value when the blackhole issue goes away.
865 0 to disable the blackhole detection.
867 By default, it is set to 0 (feature is disabled).
869 tcp_fastopen_key - list of comma separated 32-digit hexadecimal INTEGERs
870 The list consists of a primary key and an optional backup key. The
871 primary key is used for both creating and validating cookies, while the
872 optional backup key is only used for validating cookies. The purpose of
873 the backup key is to maximize TFO validation when keys are rotated.
875 A randomly chosen primary key may be configured by the kernel if
876 the tcp_fastopen sysctl is set to 0x400 (see above), or if the
877 TCP_FASTOPEN setsockopt() optname is set and a key has not been
878 previously configured via sysctl. If keys are configured via
879 setsockopt() by using the TCP_FASTOPEN_KEY optname, then those
880 per-socket keys will be used instead of any keys that are specified via
883 A key is specified as 4 8-digit hexadecimal integers which are separated
884 by a '-' as: xxxxxxxx-xxxxxxxx-xxxxxxxx-xxxxxxxx. Leading zeros may be
885 omitted. A primary and a backup key may be specified by separating them
886 by a comma. If only one key is specified, it becomes the primary key and
887 any previously configured backup keys are removed.
889 tcp_syn_retries - INTEGER
890 Number of times initial SYNs for an active TCP connection attempt
891 will be retransmitted. Should not be higher than 127. Default value
892 is 6, which corresponds to 67seconds (with tcp_syn_linear_timeouts = 4)
893 till the last retransmission with the current initial RTO of 1second.
894 With this the final timeout for an active TCP connection attempt
895 will happen after 131seconds.
897 tcp_timestamps - INTEGER
898 Enable timestamps as defined in RFC1323.
901 - 1: Enable timestamps as defined in RFC1323 and use random offset for
902 each connection rather than only using the current time.
903 - 2: Like 1, but without random offsets.
907 tcp_min_tso_segs - INTEGER
908 Minimal number of segments per TSO frame.
910 Since linux-3.12, TCP does an automatic sizing of TSO frames,
911 depending on flow rate, instead of filling 64Kbytes packets.
912 For specific usages, it's possible to force TCP to build big
913 TSO frames. Note that TCP stack might split too big TSO packets
914 if available window is too small.
918 tcp_tso_rtt_log - INTEGER
919 Adjustment of TSO packet sizes based on min_rtt
921 Starting from linux-5.18, TCP autosizing can be tweaked
922 for flows having small RTT.
924 Old autosizing was splitting the pacing budget to send 1024 TSO
927 tso_packet_size = sk->sk_pacing_rate / 1024;
929 With the new mechanism, we increase this TSO sizing using:
931 distance = min_rtt_usec / (2^tcp_tso_rtt_log)
932 tso_packet_size += gso_max_size >> distance;
934 This means that flows between very close hosts can use bigger
935 TSO packets, reducing their cpu costs.
937 If you want to use the old autosizing, set this sysctl to 0.
939 Default: 9 (2^9 = 512 usec)
941 tcp_pacing_ss_ratio - INTEGER
942 sk->sk_pacing_rate is set by TCP stack using a ratio applied
943 to current rate. (current_rate = cwnd * mss / srtt)
944 If TCP is in slow start, tcp_pacing_ss_ratio is applied
945 to let TCP probe for bigger speeds, assuming cwnd can be
946 doubled every other RTT.
950 tcp_pacing_ca_ratio - INTEGER
951 sk->sk_pacing_rate is set by TCP stack using a ratio applied
952 to current rate. (current_rate = cwnd * mss / srtt)
953 If TCP is in congestion avoidance phase, tcp_pacing_ca_ratio
954 is applied to conservatively probe for bigger throughput.
958 tcp_syn_linear_timeouts - INTEGER
959 The number of times for an active TCP connection to retransmit SYNs with
960 a linear backoff timeout before defaulting to an exponential backoff
961 timeout. This has no effect on SYNACK at the passive TCP side.
963 With an initial RTO of 1 and tcp_syn_linear_timeouts = 4 we would
964 expect SYN RTOs to be: 1, 1, 1, 1, 1, 2, 4, ... (4 linear timeouts,
965 and the first exponential backoff using 2^0 * initial_RTO).
968 tcp_tso_win_divisor - INTEGER
969 This allows control over what percentage of the congestion window
970 can be consumed by a single TSO frame.
971 The setting of this parameter is a choice between burstiness and
972 building larger TSO frames.
976 tcp_tw_reuse - INTEGER
977 Enable reuse of TIME-WAIT sockets for new connections when it is
978 safe from protocol viewpoint.
982 - 2 - enable for loopback traffic only
984 It should not be changed without advice/request of technical
989 tcp_window_scaling - BOOLEAN
990 Enable window scaling as defined in RFC1323.
992 tcp_shrink_window - BOOLEAN
993 This changes how the TCP receive window is calculated.
995 RFC 7323, section 2.4, says there are instances when a retracted
996 window can be offered, and that TCP implementations MUST ensure
997 that they handle a shrinking window, as specified in RFC 1122.
999 - 0 - Disabled. The window is never shrunk.
1000 - 1 - Enabled. The window is shrunk when necessary to remain within
1001 the memory limit set by autotuning (sk_rcvbuf).
1002 This only occurs if a non-zero receive window
1003 scaling factor is also in effect.
1007 tcp_wmem - vector of 3 INTEGERs: min, default, max
1008 min: Amount of memory reserved for send buffers for TCP sockets.
1009 Each TCP socket has rights to use it due to fact of its birth.
1013 default: initial size of send buffer used by TCP sockets. This
1014 value overrides net.core.wmem_default used by other protocols.
1016 It is usually lower than net.core.wmem_default.
1020 max: Maximal amount of memory allowed for automatically tuned
1021 send buffers for TCP sockets. This value does not override
1022 net.core.wmem_max. Calling setsockopt() with SO_SNDBUF disables
1023 automatic tuning of that socket's send buffer size, in which case
1024 this value is ignored.
1026 Default: between 64K and 4MB, depending on RAM size.
1028 tcp_notsent_lowat - UNSIGNED INTEGER
1029 A TCP socket can control the amount of unsent bytes in its write queue,
1030 thanks to TCP_NOTSENT_LOWAT socket option. poll()/select()/epoll()
1031 reports POLLOUT events if the amount of unsent bytes is below a per
1032 socket value, and if the write queue is not full. sendmsg() will
1033 also not add new buffers if the limit is hit.
1035 This global variable controls the amount of unsent data for
1036 sockets not using TCP_NOTSENT_LOWAT. For these sockets, a change
1037 to the global variable has immediate effect.
1039 Default: UINT_MAX (0xFFFFFFFF)
1041 tcp_workaround_signed_windows - BOOLEAN
1042 If set, assume no receipt of a window scaling option means the
1043 remote TCP is broken and treats the window as a signed quantity.
1044 If unset, assume the remote TCP is not broken even if we do
1045 not receive a window scaling option from them.
1049 tcp_thin_linear_timeouts - BOOLEAN
1050 Enable dynamic triggering of linear timeouts for thin streams.
1051 If set, a check is performed upon retransmission by timeout to
1052 determine if the stream is thin (less than 4 packets in flight).
1053 As long as the stream is found to be thin, up to 6 linear
1054 timeouts may be performed before exponential backoff mode is
1055 initiated. This improves retransmission latency for
1056 non-aggressive thin streams, often found to be time-dependent.
1057 For more information on thin streams, see
1058 Documentation/networking/tcp-thin.rst
1062 tcp_limit_output_bytes - INTEGER
1063 Controls TCP Small Queue limit per tcp socket.
1064 TCP bulk sender tends to increase packets in flight until it
1065 gets losses notifications. With SNDBUF autotuning, this can
1066 result in a large amount of packets queued on the local machine
1067 (e.g.: qdiscs, CPU backlog, or device) hurting latency of other
1068 flows, for typical pfifo_fast qdiscs. tcp_limit_output_bytes
1069 limits the number of bytes on qdisc or device to reduce artificial
1070 RTT/cwnd and reduce bufferbloat.
1072 Default: 1048576 (16 * 65536)
1074 tcp_challenge_ack_limit - INTEGER
1075 Limits number of Challenge ACK sent per second, as recommended
1076 in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
1077 Note that this per netns rate limit can allow some side channel
1078 attacks and probably should not be enabled.
1079 TCP stack implements per TCP socket limits anyway.
1080 Default: INT_MAX (unlimited)
1082 tcp_ehash_entries - INTEGER
1083 Show the number of hash buckets for TCP sockets in the current
1084 networking namespace.
1086 A negative value means the networking namespace does not own its
1087 hash buckets and shares the initial networking namespace's one.
1089 tcp_child_ehash_entries - INTEGER
1090 Control the number of hash buckets for TCP sockets in the child
1091 networking namespace, which must be set before clone() or unshare().
1093 If the value is not 0, the kernel uses a value rounded up to 2^n
1094 as the actual hash bucket size. 0 is a special value, meaning
1095 the child networking namespace will share the initial networking
1096 namespace's hash buckets.
1098 Note that the child will use the global one in case the kernel
1099 fails to allocate enough memory. In addition, the global hash
1100 buckets are spread over available NUMA nodes, but the allocation
1101 of the child hash table depends on the current process's NUMA
1102 policy, which could result in performance differences.
1104 Note also that the default value of tcp_max_tw_buckets and
1105 tcp_max_syn_backlog depend on the hash bucket size.
1107 Possible values: 0, 2^n (n: 0 - 24 (16Mi))
1111 tcp_plb_enabled - BOOLEAN
1112 If set and the underlying congestion control (e.g. DCTCP) supports
1113 and enables PLB feature, TCP PLB (Protective Load Balancing) is
1114 enabled. PLB is described in the following paper:
1115 https://doi.org/10.1145/3544216.3544226. Based on PLB parameters,
1116 upon sensing sustained congestion, TCP triggers a change in
1117 flow label field for outgoing IPv6 packets. A change in flow label
1118 field potentially changes the path of outgoing packets for switches
1119 that use ECMP/WCMP for routing.
1121 PLB changes socket txhash which results in a change in IPv6 Flow Label
1122 field, and currently no-op for IPv4 headers. It is possible
1123 to apply PLB for IPv4 with other network header fields (e.g. TCP
1124 or IPv4 options) or using encapsulation where outer header is used
1125 by switches to determine next hop. In either case, further host
1126 and switch side changes will be needed.
1128 When set, PLB assumes that congestion signal (e.g. ECN) is made
1129 available and used by congestion control module to estimate a
1130 congestion measure (e.g. ce_ratio). PLB needs a congestion measure to
1131 make repathing decisions.
1135 tcp_plb_idle_rehash_rounds - INTEGER
1136 Number of consecutive congested rounds (RTT) seen after which
1137 a rehash can be performed, given there are no packets in flight.
1138 This is referred to as M in PLB paper:
1139 https://doi.org/10.1145/3544216.3544226.
1141 Possible Values: 0 - 31
1145 tcp_plb_rehash_rounds - INTEGER
1146 Number of consecutive congested rounds (RTT) seen after which
1147 a forced rehash can be performed. Be careful when setting this
1148 parameter, as a small value increases the risk of retransmissions.
1149 This is referred to as N in PLB paper:
1150 https://doi.org/10.1145/3544216.3544226.
1152 Possible Values: 0 - 31
1156 tcp_plb_suspend_rto_sec - INTEGER
1157 Time, in seconds, to suspend PLB in event of an RTO. In order to avoid
1158 having PLB repath onto a connectivity "black hole", after an RTO a TCP
1159 connection suspends PLB repathing for a random duration between 1x and
1160 2x of this parameter. Randomness is added to avoid concurrent rehashing
1161 of multiple TCP connections. This should be set corresponding to the
1162 amount of time it takes to repair a failed link.
1164 Possible Values: 0 - 255
1168 tcp_plb_cong_thresh - INTEGER
1169 Fraction of packets marked with congestion over a round (RTT) to
1170 tag that round as congested. This is referred to as K in the PLB paper:
1171 https://doi.org/10.1145/3544216.3544226.
1173 The 0-1 fraction range is mapped to 0-256 range to avoid floating
1174 point operations. For example, 128 means that if at least 50% of
1175 the packets in a round were marked as congested then the round
1176 will be tagged as congested.
1178 Setting threshold to 0 means that PLB repaths every RTT regardless
1179 of congestion. This is not intended behavior for PLB and should be
1180 used only for experimentation purpose.
1182 Possible Values: 0 - 256
1186 tcp_pingpong_thresh - INTEGER
1187 The number of estimated data replies sent for estimated incoming data
1188 requests that must happen before TCP considers that a connection is a
1189 "ping-pong" (request-response) connection for which delayed
1190 acknowledgments can provide benefits.
1192 This threshold is 1 by default, but some applications may need a higher
1193 threshold for optimal performance.
1195 Possible Values: 1 - 255
1202 udp_l3mdev_accept - BOOLEAN
1203 Enabling this option allows a "global" bound socket to work
1204 across L3 master domains (e.g., VRFs) with packets capable of
1205 being received regardless of the L3 domain in which they
1206 originated. Only valid when the kernel was compiled with
1207 CONFIG_NET_L3_MASTER_DEV.
1209 Default: 0 (disabled)
1211 udp_mem - vector of 3 INTEGERs: min, pressure, max
1212 Number of pages allowed for queueing by all UDP sockets.
1214 min: Number of pages allowed for queueing by all UDP sockets.
1216 pressure: This value was introduced to follow format of tcp_mem.
1218 max: This value was introduced to follow format of tcp_mem.
1220 Default is calculated at boot time from amount of available memory.
1222 udp_rmem_min - INTEGER
1223 Minimal size of receive buffer used by UDP sockets in moderation.
1224 Each UDP socket is able to use the size for receiving data, even if
1225 total pages of UDP sockets exceed udp_mem pressure. The unit is byte.
1229 udp_wmem_min - INTEGER
1230 UDP does not have tx memory accounting and this tunable has no effect.
1232 udp_hash_entries - INTEGER
1233 Show the number of hash buckets for UDP sockets in the current
1234 networking namespace.
1236 A negative value means the networking namespace does not own its
1237 hash buckets and shares the initial networking namespace's one.
1239 udp_child_ehash_entries - INTEGER
1240 Control the number of hash buckets for UDP sockets in the child
1241 networking namespace, which must be set before clone() or unshare().
1243 If the value is not 0, the kernel uses a value rounded up to 2^n
1244 as the actual hash bucket size. 0 is a special value, meaning
1245 the child networking namespace will share the initial networking
1246 namespace's hash buckets.
1248 Note that the child will use the global one in case the kernel
1249 fails to allocate enough memory. In addition, the global hash
1250 buckets are spread over available NUMA nodes, but the allocation
1251 of the child hash table depends on the current process's NUMA
1252 policy, which could result in performance differences.
1254 Possible values: 0, 2^n (n: 7 (128) - 16 (64K))
1262 raw_l3mdev_accept - BOOLEAN
1263 Enabling this option allows a "global" bound socket to work
1264 across L3 master domains (e.g., VRFs) with packets capable of
1265 being received regardless of the L3 domain in which they
1266 originated. Only valid when the kernel was compiled with
1267 CONFIG_NET_L3_MASTER_DEV.
1269 Default: 1 (enabled)
1274 cipso_cache_enable - BOOLEAN
1275 If set, enable additions to and lookups from the CIPSO label mapping
1276 cache. If unset, additions are ignored and lookups always result in a
1277 miss. However, regardless of the setting the cache is still
1278 invalidated when required when means you can safely toggle this on and
1279 off and the cache will always be "safe".
1283 cipso_cache_bucket_size - INTEGER
1284 The CIPSO label cache consists of a fixed size hash table with each
1285 hash bucket containing a number of cache entries. This variable limits
1286 the number of entries in each hash bucket; the larger the value is, the
1287 more CIPSO label mappings that can be cached. When the number of
1288 entries in a given hash bucket reaches this limit adding new entries
1289 causes the oldest entry in the bucket to be removed to make room.
1293 cipso_rbm_optfmt - BOOLEAN
1294 Enable the "Optimized Tag 1 Format" as defined in section 3.4.2.6 of
1295 the CIPSO draft specification (see Documentation/netlabel for details).
1296 This means that when set the CIPSO tag will be padded with empty
1297 categories in order to make the packet data 32-bit aligned.
1301 cipso_rbm_structvalid - BOOLEAN
1302 If set, do a very strict check of the CIPSO option when
1303 ip_options_compile() is called. If unset, relax the checks done during
1304 ip_options_compile(). Either way is "safe" as errors are caught else
1305 where in the CIPSO processing code but setting this to 0 (False) should
1306 result in less work (i.e. it should be faster) but could cause problems
1307 with other implementations that require strict checking.
1314 ip_local_port_range - 2 INTEGERS
1315 Defines the local port range that is used by TCP and UDP to
1316 choose the local port. The first number is the first, the
1317 second the last local port number.
1318 If possible, it is better these numbers have different parity
1319 (one even and one odd value).
1320 Must be greater than or equal to ip_unprivileged_port_start.
1321 The default values are 32768 and 60999 respectively.
1323 ip_local_reserved_ports - list of comma separated ranges
1324 Specify the ports which are reserved for known third-party
1325 applications. These ports will not be used by automatic port
1326 assignments (e.g. when calling connect() or bind() with port
1327 number 0). Explicit port allocation behavior is unchanged.
1329 The format used for both input and output is a comma separated
1330 list of ranges (e.g. "1,2-4,10-10" for ports 1, 2, 3, 4 and
1331 10). Writing to the file will clear all previously reserved
1332 ports and update the current list with the one given in the
1335 Note that ip_local_port_range and ip_local_reserved_ports
1336 settings are independent and both are considered by the kernel
1337 when determining which ports are available for automatic port
1340 You can reserve ports which are not in the current
1341 ip_local_port_range, e.g.::
1343 $ cat /proc/sys/net/ipv4/ip_local_port_range
1345 $ cat /proc/sys/net/ipv4/ip_local_reserved_ports
1348 although this is redundant. However such a setting is useful
1349 if later the port range is changed to a value that will
1350 include the reserved ports. Also keep in mind, that overlapping
1351 of these ranges may affect probability of selecting ephemeral
1352 ports which are right after block of reserved ports.
1356 ip_unprivileged_port_start - INTEGER
1357 This is a per-namespace sysctl. It defines the first
1358 unprivileged port in the network namespace. Privileged ports
1359 require root or CAP_NET_BIND_SERVICE in order to bind to them.
1360 To disable all privileged ports, set this to 0. They must not
1361 overlap with the ip_local_port_range.
1365 ip_nonlocal_bind - BOOLEAN
1366 If set, allows processes to bind() to non-local IP addresses,
1367 which can be quite useful - but may break some applications.
1371 ip_autobind_reuse - BOOLEAN
1372 By default, bind() does not select the ports automatically even if
1373 the new socket and all sockets bound to the port have SO_REUSEADDR.
1374 ip_autobind_reuse allows bind() to reuse the port and this is useful
1375 when you use bind()+connect(), but may break some applications.
1376 The preferred solution is to use IP_BIND_ADDRESS_NO_PORT and this
1377 option should only be set by experts.
1380 ip_dynaddr - INTEGER
1381 If set non-zero, enables support for dynamic addresses.
1382 If set to a non-zero value larger than 1, a kernel log
1383 message will be printed when dynamic address rewriting
1388 ip_early_demux - BOOLEAN
1389 Optimize input packet processing down to one demux for
1390 certain kinds of local sockets. Currently we only do this
1391 for established TCP and connected UDP sockets.
1393 It may add an additional cost for pure routing workloads that
1394 reduces overall throughput, in such case you should disable it.
1398 ping_group_range - 2 INTEGERS
1399 Restrict ICMP_PROTO datagram sockets to users in the group range.
1400 The default is "1 0", meaning, that nobody (not even root) may
1401 create ping sockets. Setting it to "100 100" would grant permissions
1402 to the single group. "0 4294967294" would enable it for the world, "100
1403 4294967294" would enable it for the users, but not daemons.
1405 tcp_early_demux - BOOLEAN
1406 Enable early demux for established TCP sockets.
1410 udp_early_demux - BOOLEAN
1411 Enable early demux for connected UDP sockets. Disable this if
1412 your system could experience more unconnected load.
1416 icmp_echo_ignore_all - BOOLEAN
1417 If set non-zero, then the kernel will ignore all ICMP ECHO
1418 requests sent to it.
1422 icmp_echo_enable_probe - BOOLEAN
1423 If set to one, then the kernel will respond to RFC 8335 PROBE
1424 requests sent to it.
1428 icmp_echo_ignore_broadcasts - BOOLEAN
1429 If set non-zero, then the kernel will ignore all ICMP ECHO and
1430 TIMESTAMP requests sent to it via broadcast/multicast.
1434 icmp_ratelimit - INTEGER
1435 Limit the maximal rates for sending ICMP packets whose type matches
1436 icmp_ratemask (see below) to specific targets.
1437 0 to disable any limiting,
1438 otherwise the minimal space between responses in milliseconds.
1439 Note that another sysctl, icmp_msgs_per_sec limits the number
1440 of ICMP packets sent on all targets.
1444 icmp_msgs_per_sec - INTEGER
1445 Limit maximal number of ICMP packets sent per second from this host.
1446 Only messages whose type matches icmp_ratemask (see below) are
1447 controlled by this limit. For security reasons, the precise count
1448 of messages per second is randomized.
1452 icmp_msgs_burst - INTEGER
1453 icmp_msgs_per_sec controls number of ICMP packets sent per second,
1454 while icmp_msgs_burst controls the burst size of these packets.
1455 For security reasons, the precise burst size is randomized.
1459 icmp_ratemask - INTEGER
1460 Mask made of ICMP types for which rates are being limited.
1462 Significant bits: IHGFEDCBA9876543210
1464 Default mask: 0000001100000011000 (6168)
1466 Bit definitions (see include/linux/icmp.h):
1468 = =========================
1470 3 Destination Unreachable [1]_
1471 4 Source Quench [1]_
1474 B Time Exceeded [1]_
1475 C Parameter Problem [1]_
1480 H Address Mask Request
1481 I Address Mask Reply
1482 = =========================
1484 .. [1] These are rate limited by default (see default mask above)
1486 icmp_ignore_bogus_error_responses - BOOLEAN
1487 Some routers violate RFC1122 by sending bogus responses to broadcast
1488 frames. Such violations are normally logged via a kernel warning.
1489 If this is set to TRUE, the kernel will not give such warnings, which
1490 will avoid log file clutter.
1494 icmp_errors_use_inbound_ifaddr - BOOLEAN
1496 If zero, icmp error messages are sent with the primary address of
1497 the exiting interface.
1499 If non-zero, the message will be sent with the primary address of
1500 the interface that received the packet that caused the icmp error.
1501 This is the behaviour many network administrators will expect from
1502 a router. And it can make debugging complicated network layouts
1505 Note that if no primary address exists for the interface selected,
1506 then the primary address of the first non-loopback interface that
1507 has one will be used regardless of this setting.
1511 igmp_max_memberships - INTEGER
1512 Change the maximum number of multicast groups we can subscribe to.
1515 Theoretical maximum value is bounded by having to send a membership
1516 report in a single datagram (i.e. the report can't span multiple
1517 datagrams, or risk confusing the switch and leaving groups you don't
1520 The number of supported groups 'M' is bounded by the number of group
1521 report entries you can fit into a single datagram of 65535 bytes.
1523 M = 65536-sizeof (ip header)/(sizeof(Group record))
1525 Group records are variable length, with a minimum of 12 bytes.
1526 So net.ipv4.igmp_max_memberships should not be set higher than:
1528 (65536-24) / 12 = 5459
1530 The value 5459 assumes no IP header options, so in practice
1531 this number may be lower.
1533 igmp_max_msf - INTEGER
1534 Maximum number of addresses allowed in the source filter list for a
1540 Controls the IGMP query robustness variable (see RFC2236 8.1).
1542 Default: 2 (as specified by RFC2236 8.1)
1544 Minimum: 1 (as specified by RFC6636 4.5)
1546 force_igmp_version - INTEGER
1547 - 0 - (default) No enforcement of a IGMP version, IGMPv1/v2 fallback
1548 allowed. Will back to IGMPv3 mode again if all IGMPv1/v2 Querier
1549 Present timer expires.
1550 - 1 - Enforce to use IGMP version 1. Will also reply IGMPv1 report if
1551 receive IGMPv2/v3 query.
1552 - 2 - Enforce to use IGMP version 2. Will fallback to IGMPv1 if receive
1553 IGMPv1 query message. Will reply report if receive IGMPv3 query.
1554 - 3 - Enforce to use IGMP version 3. The same react with default 0.
1558 this is not the same with force_mld_version because IGMPv3 RFC3376
1559 Security Considerations does not have clear description that we could
1560 ignore other version messages completely as MLDv2 RFC3810. So make
1561 this value as default 0 is recommended.
1563 ``conf/interface/*``
1564 changes special settings per interface (where
1565 interface" is the name of your network interface)
1568 is special, changes the settings for all interfaces
1570 log_martians - BOOLEAN
1571 Log packets with impossible addresses to kernel log.
1572 log_martians for the interface will be enabled if at least one of
1573 conf/{all,interface}/log_martians is set to TRUE,
1574 it will be disabled otherwise
1576 accept_redirects - BOOLEAN
1577 Accept ICMP redirect messages.
1578 accept_redirects for the interface will be enabled if:
1580 - both conf/{all,interface}/accept_redirects are TRUE in the case
1581 forwarding for the interface is enabled
1585 - at least one of conf/{all,interface}/accept_redirects is TRUE in the
1586 case forwarding for the interface is disabled
1588 accept_redirects for the interface will be disabled otherwise
1595 forwarding - BOOLEAN
1596 Enable IP forwarding on this interface. This controls whether packets
1597 received _on_ this interface can be forwarded.
1599 mc_forwarding - BOOLEAN
1600 Do multicast routing. The kernel needs to be compiled with CONFIG_MROUTE
1601 and a multicast routing daemon is required.
1602 conf/all/mc_forwarding must also be set to TRUE to enable multicast
1603 routing for the interface
1606 Integer value used to differentiate the devices by the medium they
1607 are attached to. Two devices can have different id values when
1608 the broadcast packets are received only on one of them.
1609 The default value 0 means that the device is the only interface
1610 to its medium, value of -1 means that medium is not known.
1612 Currently, it is used to change the proxy_arp behavior:
1613 the proxy_arp feature is enabled for packets forwarded between
1614 two devices attached to different media.
1619 proxy_arp for the interface will be enabled if at least one of
1620 conf/{all,interface}/proxy_arp is set to TRUE,
1621 it will be disabled otherwise
1623 proxy_arp_pvlan - BOOLEAN
1624 Private VLAN proxy arp.
1626 Basically allow proxy arp replies back to the same interface
1627 (from which the ARP request/solicitation was received).
1629 This is done to support (ethernet) switch features, like RFC
1630 3069, where the individual ports are NOT allowed to
1631 communicate with each other, but they are allowed to talk to
1632 the upstream router. As described in RFC 3069, it is possible
1633 to allow these hosts to communicate through the upstream
1634 router by proxy_arp'ing. Don't need to be used together with
1637 This technology is known by different names:
1639 In RFC 3069 it is called VLAN Aggregation.
1640 Cisco and Allied Telesyn call it Private VLAN.
1641 Hewlett-Packard call it Source-Port filtering or port-isolation.
1642 Ericsson call it MAC-Forced Forwarding (RFC Draft).
1644 proxy_delay - INTEGER
1645 Delay proxy response.
1647 Delay response to a neighbor solicitation when proxy_arp
1648 or proxy_ndp is enabled. A random value between [0, proxy_delay)
1649 will be chosen, setting to zero means reply with no delay.
1650 Value in jiffies. Defaults to 80.
1652 shared_media - BOOLEAN
1653 Send(router) or accept(host) RFC1620 shared media redirects.
1654 Overrides secure_redirects.
1656 shared_media for the interface will be enabled if at least one of
1657 conf/{all,interface}/shared_media is set to TRUE,
1658 it will be disabled otherwise
1662 secure_redirects - BOOLEAN
1663 Accept ICMP redirect messages only to gateways listed in the
1664 interface's current gateway list. Even if disabled, RFC1122 redirect
1667 Overridden by shared_media.
1669 secure_redirects for the interface will be enabled if at least one of
1670 conf/{all,interface}/secure_redirects is set to TRUE,
1671 it will be disabled otherwise
1675 send_redirects - BOOLEAN
1676 Send redirects, if router.
1678 send_redirects for the interface will be enabled if at least one of
1679 conf/{all,interface}/send_redirects is set to TRUE,
1680 it will be disabled otherwise
1684 bootp_relay - BOOLEAN
1685 Accept packets with source address 0.b.c.d destined
1686 not to this host as local ones. It is supposed, that
1687 BOOTP relay daemon will catch and forward such packets.
1688 conf/all/bootp_relay must also be set to TRUE to enable BOOTP relay
1693 Not Implemented Yet.
1695 accept_source_route - BOOLEAN
1696 Accept packets with SRR option.
1697 conf/all/accept_source_route must also be set to TRUE to accept packets
1698 with SRR option on the interface
1705 accept_local - BOOLEAN
1706 Accept packets with local source addresses. In combination with
1707 suitable routing, this can be used to direct packets between two
1708 local interfaces over the wire and have them accepted properly.
1711 route_localnet - BOOLEAN
1712 Do not consider loopback addresses as martian source or destination
1713 while routing. This enables the use of 127/8 for local routing purposes.
1718 - 0 - No source validation.
1719 - 1 - Strict mode as defined in RFC3704 Strict Reverse Path
1720 Each incoming packet is tested against the FIB and if the interface
1721 is not the best reverse path the packet check will fail.
1722 By default failed packets are discarded.
1723 - 2 - Loose mode as defined in RFC3704 Loose Reverse Path
1724 Each incoming packet's source address is also tested against the FIB
1725 and if the source address is not reachable via any interface
1726 the packet check will fail.
1728 Current recommended practice in RFC3704 is to enable strict mode
1729 to prevent IP spoofing from DDos attacks. If using asymmetric routing
1730 or other complicated routing, then loose mode is recommended.
1732 The max value from conf/{all,interface}/rp_filter is used
1733 when doing source validation on the {interface}.
1735 Default value is 0. Note that some distributions enable it
1738 src_valid_mark - BOOLEAN
1739 - 0 - The fwmark of the packet is not included in reverse path
1740 route lookup. This allows for asymmetric routing configurations
1741 utilizing the fwmark in only one direction, e.g., transparent
1744 - 1 - The fwmark of the packet is included in reverse path route
1745 lookup. This permits rp_filter to function when the fwmark is
1746 used for routing traffic in both directions.
1748 This setting also affects the utilization of fmwark when
1749 performing source address selection for ICMP replies, or
1750 determining addresses stored for the IPOPT_TS_TSANDADDR and
1751 IPOPT_RR IP options.
1753 The max value from conf/{all,interface}/src_valid_mark is used.
1757 arp_filter - BOOLEAN
1758 - 1 - Allows you to have multiple network interfaces on the same
1759 subnet, and have the ARPs for each interface be answered
1760 based on whether or not the kernel would route a packet from
1761 the ARP'd IP out that interface (therefore you must use source
1762 based routing for this to work). In other words it allows control
1763 of which cards (usually 1) will respond to an arp request.
1765 - 0 - (default) The kernel can respond to arp requests with addresses
1766 from other interfaces. This may seem wrong but it usually makes
1767 sense, because it increases the chance of successful communication.
1768 IP addresses are owned by the complete host on Linux, not by
1769 particular interfaces. Only for more complex setups like load-
1770 balancing, does this behaviour cause problems.
1772 arp_filter for the interface will be enabled if at least one of
1773 conf/{all,interface}/arp_filter is set to TRUE,
1774 it will be disabled otherwise
1776 arp_announce - INTEGER
1777 Define different restriction levels for announcing the local
1778 source IP address from IP packets in ARP requests sent on
1781 - 0 - (default) Use any local address, configured on any interface
1782 - 1 - Try to avoid local addresses that are not in the target's
1783 subnet for this interface. This mode is useful when target
1784 hosts reachable via this interface require the source IP
1785 address in ARP requests to be part of their logical network
1786 configured on the receiving interface. When we generate the
1787 request we will check all our subnets that include the
1788 target IP and will preserve the source address if it is from
1789 such subnet. If there is no such subnet we select source
1790 address according to the rules for level 2.
1791 - 2 - Always use the best local address for this target.
1792 In this mode we ignore the source address in the IP packet
1793 and try to select local address that we prefer for talks with
1794 the target host. Such local address is selected by looking
1795 for primary IP addresses on all our subnets on the outgoing
1796 interface that include the target IP address. If no suitable
1797 local address is found we select the first local address
1798 we have on the outgoing interface or on all other interfaces,
1799 with the hope we will receive reply for our request and
1800 even sometimes no matter the source IP address we announce.
1802 The max value from conf/{all,interface}/arp_announce is used.
1804 Increasing the restriction level gives more chance for
1805 receiving answer from the resolved target while decreasing
1806 the level announces more valid sender's information.
1808 arp_ignore - INTEGER
1809 Define different modes for sending replies in response to
1810 received ARP requests that resolve local target IP addresses:
1812 - 0 - (default): reply for any local target IP address, configured
1814 - 1 - reply only if the target IP address is local address
1815 configured on the incoming interface
1816 - 2 - reply only if the target IP address is local address
1817 configured on the incoming interface and both with the
1818 sender's IP address are part from same subnet on this interface
1819 - 3 - do not reply for local addresses configured with scope host,
1820 only resolutions for global and link addresses are replied
1822 - 8 - do not reply for all local addresses
1824 The max value from conf/{all,interface}/arp_ignore is used
1825 when ARP request is received on the {interface}
1827 arp_notify - BOOLEAN
1828 Define mode for notification of address and device changes.
1830 == ==========================================================
1831 0 (default): do nothing
1832 1 Generate gratuitous arp requests when device is brought up
1833 or hardware address changes.
1834 == ==========================================================
1836 arp_accept - INTEGER
1837 Define behavior for accepting gratuitous ARP (garp) frames from devices
1838 that are not already present in the ARP table:
1840 - 0 - don't create new entries in the ARP table
1841 - 1 - create new entries in the ARP table
1842 - 2 - create new entries only if the source IP address is in the same
1843 subnet as an address configured on the interface that received the
1846 Both replies and requests type gratuitous arp will trigger the
1847 ARP table to be updated, if this setting is on.
1849 If the ARP table already contains the IP address of the
1850 gratuitous arp frame, the arp table will be updated regardless
1851 if this setting is on or off.
1853 arp_evict_nocarrier - BOOLEAN
1854 Clears the ARP cache on NOCARRIER events. This option is important for
1855 wireless devices where the ARP cache should not be cleared when roaming
1856 between access points on the same network. In most cases this should
1857 remain as the default (1).
1859 - 1 - (default): Clear the ARP cache on NOCARRIER events
1860 - 0 - Do not clear ARP cache on NOCARRIER events
1862 mcast_solicit - INTEGER
1863 The maximum number of multicast probes in INCOMPLETE state,
1864 when the associated hardware address is unknown. Defaults
1867 ucast_solicit - INTEGER
1868 The maximum number of unicast probes in PROBE state, when
1869 the hardware address is being reconfirmed. Defaults to 3.
1871 app_solicit - INTEGER
1872 The maximum number of probes to send to the user space ARP daemon
1873 via netlink before dropping back to multicast probes (see
1874 mcast_resolicit). Defaults to 0.
1876 mcast_resolicit - INTEGER
1877 The maximum number of multicast probes after unicast and
1878 app probes in PROBE state. Defaults to 0.
1880 disable_policy - BOOLEAN
1881 Disable IPSEC policy (SPD) for this interface
1883 disable_xfrm - BOOLEAN
1884 Disable IPSEC encryption on this interface, whatever the policy
1886 igmpv2_unsolicited_report_interval - INTEGER
1887 The interval in milliseconds in which the next unsolicited
1888 IGMPv1 or IGMPv2 report retransmit will take place.
1890 Default: 10000 (10 seconds)
1892 igmpv3_unsolicited_report_interval - INTEGER
1893 The interval in milliseconds in which the next unsolicited
1894 IGMPv3 report retransmit will take place.
1896 Default: 1000 (1 seconds)
1898 ignore_routes_with_linkdown - BOOLEAN
1899 Ignore routes whose link is down when performing a FIB lookup.
1901 promote_secondaries - BOOLEAN
1902 When a primary IP address is removed from this interface
1903 promote a corresponding secondary IP address instead of
1904 removing all the corresponding secondary IP addresses.
1906 drop_unicast_in_l2_multicast - BOOLEAN
1907 Drop any unicast IP packets that are received in link-layer
1908 multicast (or broadcast) frames.
1910 This behavior (for multicast) is actually a SHOULD in RFC
1911 1122, but is disabled by default for compatibility reasons.
1915 drop_gratuitous_arp - BOOLEAN
1916 Drop all gratuitous ARP frames, for example if there's a known
1917 good ARP proxy on the network and such frames need not be used
1918 (or in the case of 802.11, must not be used to prevent attacks.)
1924 Allows you to write a number, which can be used as required.
1928 xfrm4_gc_thresh - INTEGER
1929 (Obsolete since linux-4.14)
1930 The threshold at which we will start garbage collecting for IPv4
1931 destination cache entries. At twice this value the system will
1932 refuse new allocations.
1934 igmp_link_local_mcast_reports - BOOLEAN
1935 Enable IGMP reports for link local multicast groups in the
1941 kuznet@ms2.inr.ac.ru
1948 delon.nicolas@wanadoo.fr
1953 /proc/sys/net/ipv6/* Variables
1954 ==============================
1956 IPv6 has no global variables such as tcp_*. tcp_* settings under ipv4/ also
1957 apply to IPv6 [XXX?].
1959 bindv6only - BOOLEAN
1960 Default value for IPV6_V6ONLY socket option,
1961 which restricts use of the IPv6 socket to IPv6 communication
1964 - TRUE: disable IPv4-mapped address feature
1965 - FALSE: enable IPv4-mapped address feature
1967 Default: FALSE (as specified in RFC3493)
1969 flowlabel_consistency - BOOLEAN
1970 Protect the consistency (and unicity) of flow label.
1971 You have to disable it to use IPV6_FL_F_REFLECT flag on the
1979 auto_flowlabels - INTEGER
1980 Automatically generate flow labels based on a flow hash of the
1981 packet. This allows intermediate devices, such as routers, to
1982 identify packet flows for mechanisms like Equal Cost Multipath
1983 Routing (see RFC 6438).
1985 = ===========================================================
1986 0 automatic flow labels are completely disabled
1987 1 automatic flow labels are enabled by default, they can be
1988 disabled on a per socket basis using the IPV6_AUTOFLOWLABEL
1990 2 automatic flow labels are allowed, they may be enabled on a
1991 per socket basis using the IPV6_AUTOFLOWLABEL socket option
1992 3 automatic flow labels are enabled and enforced, they cannot
1993 be disabled by the socket option
1994 = ===========================================================
1998 flowlabel_state_ranges - BOOLEAN
1999 Split the flow label number space into two ranges. 0-0x7FFFF is
2000 reserved for the IPv6 flow manager facility, 0x80000-0xFFFFF
2001 is reserved for stateless flow labels as described in RFC6437.
2008 flowlabel_reflect - INTEGER
2009 Control flow label reflection. Needed for Path MTU
2010 Discovery to work with Equal Cost Multipath Routing in anycast
2011 environments. See RFC 7690 and:
2012 https://tools.ietf.org/html/draft-wang-6man-flow-label-reflection-01
2016 - 1: enabled for established flows
2018 Note that this prevents automatic flowlabel changes, as done
2019 in "tcp: change IPv6 flow-label upon receiving spurious retransmission"
2020 and "tcp: Change txhash on every SYN and RTO retransmit"
2022 - 2: enabled for TCP RESET packets (no active listener)
2023 If set, a RST packet sent in response to a SYN packet on a closed
2024 port will reflect the incoming flow label.
2026 - 4: enabled for ICMPv6 echo reply messages.
2030 fib_multipath_hash_policy - INTEGER
2031 Controls which hash policy to use for multipath routes.
2033 Default: 0 (Layer 3)
2037 - 0 - Layer 3 (source and destination addresses plus flow label)
2038 - 1 - Layer 4 (standard 5-tuple)
2039 - 2 - Layer 3 or inner Layer 3 if present
2040 - 3 - Custom multipath hash. Fields used for multipath hash calculation
2041 are determined by fib_multipath_hash_fields sysctl
2043 fib_multipath_hash_fields - UNSIGNED INTEGER
2044 When fib_multipath_hash_policy is set to 3 (custom multipath hash), the
2045 fields used for multipath hash calculation are determined by this
2048 This value is a bitmask which enables various fields for multipath hash
2051 Possible fields are:
2053 ====== ============================
2054 0x0001 Source IP address
2055 0x0002 Destination IP address
2059 0x0020 Destination port
2060 0x0040 Inner source IP address
2061 0x0080 Inner destination IP address
2062 0x0100 Inner IP protocol
2063 0x0200 Inner Flow Label
2064 0x0400 Inner source port
2065 0x0800 Inner destination port
2066 ====== ============================
2068 Default: 0x0007 (source IP, destination IP and IP protocol)
2070 anycast_src_echo_reply - BOOLEAN
2071 Controls the use of anycast addresses as source addresses for ICMPv6
2079 idgen_delay - INTEGER
2080 Controls the delay in seconds after which time to retry
2081 privacy stable address generation if a DAD conflict is
2084 Default: 1 (as specified in RFC7217)
2086 idgen_retries - INTEGER
2087 Controls the number of retries to generate a stable privacy
2088 address if a DAD conflict is detected.
2090 Default: 3 (as specified in RFC7217)
2093 Controls the MLD query robustness variable (see RFC3810 9.1).
2095 Default: 2 (as specified by RFC3810 9.1)
2097 Minimum: 1 (as specified by RFC6636 4.5)
2099 max_dst_opts_number - INTEGER
2100 Maximum number of non-padding TLVs allowed in a Destination
2101 options extension header. If this value is less than zero
2102 then unknown options are disallowed and the number of known
2103 TLVs allowed is the absolute value of this number.
2107 max_hbh_opts_number - INTEGER
2108 Maximum number of non-padding TLVs allowed in a Hop-by-Hop
2109 options extension header. If this value is less than zero
2110 then unknown options are disallowed and the number of known
2111 TLVs allowed is the absolute value of this number.
2115 max_dst_opts_length - INTEGER
2116 Maximum length allowed for a Destination options extension
2119 Default: INT_MAX (unlimited)
2121 max_hbh_length - INTEGER
2122 Maximum length allowed for a Hop-by-Hop options extension
2125 Default: INT_MAX (unlimited)
2127 skip_notify_on_dev_down - BOOLEAN
2128 Controls whether an RTM_DELROUTE message is generated for routes
2129 removed when a device is taken down or deleted. IPv4 does not
2130 generate this message; IPv6 does by default. Setting this sysctl
2131 to true skips the message, making IPv4 and IPv6 on par in relying
2132 on userspace caches to track link events and evict routes.
2134 Default: false (generate message)
2136 nexthop_compat_mode - BOOLEAN
2137 New nexthop API provides a means for managing nexthops independent of
2138 prefixes. Backwards compatibility with old route format is enabled by
2139 default which means route dumps and notifications contain the new
2140 nexthop attribute but also the full, expanded nexthop definition.
2141 Further, updates or deletes of a nexthop configuration generate route
2142 notifications for each fib entry using the nexthop. Once a system
2143 understands the new API, this sysctl can be disabled to achieve full
2144 performance benefits of the new API by disabling the nexthop expansion
2145 and extraneous notifications.
2146 Default: true (backward compat mode)
2148 fib_notify_on_flag_change - INTEGER
2149 Whether to emit RTM_NEWROUTE notifications whenever RTM_F_OFFLOAD/
2150 RTM_F_TRAP/RTM_F_OFFLOAD_FAILED flags are changed.
2152 After installing a route to the kernel, user space receives an
2153 acknowledgment, which means the route was installed in the kernel,
2154 but not necessarily in hardware.
2155 It is also possible for a route already installed in hardware to change
2156 its action and therefore its flags. For example, a host route that is
2157 trapping packets can be "promoted" to perform decapsulation following
2158 the installation of an IPinIP/VXLAN tunnel.
2159 The notifications will indicate to user-space the state of the route.
2161 Default: 0 (Do not emit notifications.)
2165 - 0 - Do not emit notifications.
2166 - 1 - Emit notifications.
2167 - 2 - Emit notifications only for RTM_F_OFFLOAD_FAILED flag change.
2170 Define the IOAM id of this node. Uses only 24 bits out of 32 in total.
2177 ioam6_id_wide - LONG INTEGER
2178 Define the wide IOAM id of this node. Uses only 56 bits out of 64 in
2179 total. Can be different from ioam6_id.
2182 Max: 0xFFFFFFFFFFFFFF
2184 Default: 0xFFFFFFFFFFFFFF
2188 ip6frag_high_thresh - INTEGER
2189 Maximum memory used to reassemble IPv6 fragments. When
2190 ip6frag_high_thresh bytes of memory is allocated for this purpose,
2191 the fragment handler will toss packets until ip6frag_low_thresh
2194 ip6frag_low_thresh - INTEGER
2195 See ip6frag_high_thresh
2197 ip6frag_time - INTEGER
2198 Time in seconds to keep an IPv6 fragment in memory.
2201 Change the interface-specific default settings.
2203 These settings would be used during creating new interfaces.
2207 Change all the interface-specific settings.
2209 [XXX: Other special features than forwarding?]
2211 conf/all/disable_ipv6 - BOOLEAN
2212 Changing this value is same as changing ``conf/default/disable_ipv6``
2213 setting and also all per-interface ``disable_ipv6`` settings to the same
2216 Reading this value does not have any particular meaning. It does not say
2217 whether IPv6 support is enabled or disabled. Returned value can be 1
2218 also in the case when some interface has ``disable_ipv6`` set to 0 and
2219 has configured IPv6 addresses.
2221 conf/all/forwarding - BOOLEAN
2222 Enable global IPv6 forwarding between all interfaces.
2224 IPv4 and IPv6 work differently here; e.g. netfilter must be used
2225 to control which interfaces may forward packets and which not.
2227 This also sets all interfaces' Host/Router setting
2228 'forwarding' to the specified value. See below for details.
2230 This referred to as global forwarding.
2235 fwmark_reflect - BOOLEAN
2236 Controls the fwmark of kernel-generated IPv6 reply packets that are not
2237 associated with a socket for example, TCP RSTs or ICMPv6 echo replies).
2238 If unset, these packets have a fwmark of zero. If set, they have the
2239 fwmark of the packet they are replying to.
2243 ``conf/interface/*``:
2244 Change special settings per interface.
2246 The functional behaviour for certain settings is different
2247 depending on whether local forwarding is enabled or not.
2250 Accept Router Advertisements; autoconfigure using them.
2252 It also determines whether or not to transmit Router
2253 Solicitations. If and only if the functional setting is to
2254 accept Router Advertisements, Router Solicitations will be
2257 Possible values are:
2259 == ===========================================================
2260 0 Do not accept Router Advertisements.
2261 1 Accept Router Advertisements if forwarding is disabled.
2262 2 Overrule forwarding behaviour. Accept Router Advertisements
2263 even if forwarding is enabled.
2264 == ===========================================================
2268 - enabled if local forwarding is disabled.
2269 - disabled if local forwarding is enabled.
2271 accept_ra_defrtr - BOOLEAN
2272 Learn default router in Router Advertisement.
2276 - enabled if accept_ra is enabled.
2277 - disabled if accept_ra is disabled.
2279 ra_defrtr_metric - UNSIGNED INTEGER
2280 Route metric for default route learned in Router Advertisement. This value
2281 will be assigned as metric for the default route learned via IPv6 Router
2282 Advertisement. Takes affect only if accept_ra_defrtr is enabled.
2287 Default: IP6_RT_PRIO_USER i.e. 1024.
2289 accept_ra_from_local - BOOLEAN
2290 Accept RA with source-address that is found on local machine
2291 if the RA is otherwise proper and able to be accepted.
2293 Default is to NOT accept these as it may be an un-intended
2298 - enabled if accept_ra_from_local is enabled
2299 on a specific interface.
2300 - disabled if accept_ra_from_local is disabled
2301 on a specific interface.
2303 accept_ra_min_hop_limit - INTEGER
2304 Minimum hop limit Information in Router Advertisement.
2306 Hop limit Information in Router Advertisement less than this
2307 variable shall be ignored.
2311 accept_ra_min_lft - INTEGER
2312 Minimum acceptable lifetime value in Router Advertisement.
2314 RA sections with a lifetime less than this value shall be
2315 ignored. Zero lifetimes stay unaffected.
2319 accept_ra_pinfo - BOOLEAN
2320 Learn Prefix Information in Router Advertisement.
2324 - enabled if accept_ra is enabled.
2325 - disabled if accept_ra is disabled.
2327 ra_honor_pio_life - BOOLEAN
2328 Whether to use RFC4862 Section 5.5.3e to determine the valid
2329 lifetime of an address matching a prefix sent in a Router
2330 Advertisement Prefix Information Option.
2332 - If enabled, the PIO valid lifetime will always be honored.
2333 - If disabled, RFC4862 section 5.5.3e is used to determine
2334 the valid lifetime of the address.
2336 Default: 0 (disabled)
2338 accept_ra_rt_info_min_plen - INTEGER
2339 Minimum prefix length of Route Information in RA.
2341 Route Information w/ prefix smaller than this variable shall
2346 * 0 if accept_ra_rtr_pref is enabled.
2347 * -1 if accept_ra_rtr_pref is disabled.
2349 accept_ra_rt_info_max_plen - INTEGER
2350 Maximum prefix length of Route Information in RA.
2352 Route Information w/ prefix larger than this variable shall
2357 * 0 if accept_ra_rtr_pref is enabled.
2358 * -1 if accept_ra_rtr_pref is disabled.
2360 accept_ra_rtr_pref - BOOLEAN
2361 Accept Router Preference in RA.
2365 - enabled if accept_ra is enabled.
2366 - disabled if accept_ra is disabled.
2368 accept_ra_mtu - BOOLEAN
2369 Apply the MTU value specified in RA option 5 (RFC4861). If
2370 disabled, the MTU specified in the RA will be ignored.
2374 - enabled if accept_ra is enabled.
2375 - disabled if accept_ra is disabled.
2377 accept_redirects - BOOLEAN
2382 - enabled if local forwarding is disabled.
2383 - disabled if local forwarding is enabled.
2385 accept_source_route - INTEGER
2386 Accept source routing (routing extension header).
2388 - >= 0: Accept only routing header type 2.
2389 - < 0: Do not accept routing header.
2394 Autoconfigure addresses using Prefix Information in Router
2399 - enabled if accept_ra_pinfo is enabled.
2400 - disabled if accept_ra_pinfo is disabled.
2402 dad_transmits - INTEGER
2403 The amount of Duplicate Address Detection probes to send.
2407 forwarding - INTEGER
2408 Configure interface-specific Host/Router behaviour.
2412 It is recommended to have the same setting on all
2413 interfaces; mixed router/host scenarios are rather uncommon.
2415 Possible values are:
2417 - 0 Forwarding disabled
2418 - 1 Forwarding enabled
2422 By default, Host behaviour is assumed. This means:
2424 1. IsRouter flag is not set in Neighbour Advertisements.
2425 2. If accept_ra is TRUE (default), transmit Router
2427 3. If accept_ra is TRUE (default), accept Router
2428 Advertisements (and do autoconfiguration).
2429 4. If accept_redirects is TRUE (default), accept Redirects.
2433 If local forwarding is enabled, Router behaviour is assumed.
2434 This means exactly the reverse from the above:
2436 1. IsRouter flag is set in Neighbour Advertisements.
2437 2. Router Solicitations are not sent unless accept_ra is 2.
2438 3. Router Advertisements are ignored unless accept_ra is 2.
2439 4. Redirects are ignored.
2441 Default: 0 (disabled) if global forwarding is disabled (default),
2442 otherwise 1 (enabled).
2445 Default Hop Limit to set.
2450 Default Maximum Transfer Unit
2452 Default: 1280 (IPv6 required minimum)
2454 ip_nonlocal_bind - BOOLEAN
2455 If set, allows processes to bind() to non-local IPv6 addresses,
2456 which can be quite useful - but may break some applications.
2460 router_probe_interval - INTEGER
2461 Minimum interval (in seconds) between Router Probing described
2466 router_solicitation_delay - INTEGER
2467 Number of seconds to wait after interface is brought up
2468 before sending Router Solicitations.
2472 router_solicitation_interval - INTEGER
2473 Number of seconds to wait between Router Solicitations.
2477 router_solicitations - INTEGER
2478 Number of Router Solicitations to send until assuming no
2479 routers are present.
2483 use_oif_addrs_only - BOOLEAN
2484 When enabled, the candidate source addresses for destinations
2485 routed via this interface are restricted to the set of addresses
2486 configured on this interface (vis. RFC 6724, section 4).
2490 use_tempaddr - INTEGER
2491 Preference for Privacy Extensions (RFC3041).
2493 * <= 0 : disable Privacy Extensions
2494 * == 1 : enable Privacy Extensions, but prefer public
2495 addresses over temporary addresses.
2496 * > 1 : enable Privacy Extensions and prefer temporary
2497 addresses over public addresses.
2501 * 0 (for most devices)
2502 * -1 (for point-to-point devices and loopback devices)
2504 temp_valid_lft - INTEGER
2505 valid lifetime (in seconds) for temporary addresses. If less than the
2506 minimum required lifetime (typically 5 seconds), temporary addresses
2507 will not be created.
2509 Default: 172800 (2 days)
2511 temp_prefered_lft - INTEGER
2512 Preferred lifetime (in seconds) for temporary addresses. If
2513 temp_prefered_lft is less than the minimum required lifetime (typically
2514 5 seconds), the preferred lifetime is the minimum required. If
2515 temp_prefered_lft is greater than temp_valid_lft, the preferred lifetime
2518 Default: 86400 (1 day)
2520 keep_addr_on_down - INTEGER
2521 Keep all IPv6 addresses on an interface down event. If set static
2522 global addresses with no expiration time are not flushed.
2525 * 0 : system default
2528 Default: 0 (addresses are removed)
2530 max_desync_factor - INTEGER
2531 Maximum value for DESYNC_FACTOR, which is a random value
2532 that ensures that clients don't synchronize with each
2533 other and generate new addresses at exactly the same time.
2534 value is in seconds.
2538 regen_max_retry - INTEGER
2539 Number of attempts before give up attempting to generate
2540 valid temporary addresses.
2544 max_addresses - INTEGER
2545 Maximum number of autoconfigured addresses per interface. Setting
2546 to zero disables the limitation. It is not recommended to set this
2547 value too large (or to zero) because it would be an easy way to
2548 crash the kernel by allowing too many addresses to be created.
2552 disable_ipv6 - BOOLEAN
2553 Disable IPv6 operation. If accept_dad is set to 2, this value
2554 will be dynamically set to TRUE if DAD fails for the link-local
2557 Default: FALSE (enable IPv6 operation)
2559 When this value is changed from 1 to 0 (IPv6 is being enabled),
2560 it will dynamically create a link-local address on the given
2561 interface and start Duplicate Address Detection, if necessary.
2563 When this value is changed from 0 to 1 (IPv6 is being disabled),
2564 it will dynamically delete all addresses and routes on the given
2565 interface. From now on it will not possible to add addresses/routes
2566 to the selected interface.
2568 accept_dad - INTEGER
2569 Whether to accept DAD (Duplicate Address Detection).
2571 == ==============================================================
2573 1 Enable DAD (default)
2574 2 Enable DAD, and disable IPv6 operation if MAC-based duplicate
2575 link-local address has been found.
2576 == ==============================================================
2578 DAD operation and mode on a given interface will be selected according
2579 to the maximum value of conf/{all,interface}/accept_dad.
2581 force_tllao - BOOLEAN
2582 Enable sending the target link-layer address option even when
2583 responding to a unicast neighbor solicitation.
2587 Quoting from RFC 2461, section 4.4, Target link-layer address:
2589 "The option MUST be included for multicast solicitations in order to
2590 avoid infinite Neighbor Solicitation "recursion" when the peer node
2591 does not have a cache entry to return a Neighbor Advertisements
2592 message. When responding to unicast solicitations, the option can be
2593 omitted since the sender of the solicitation has the correct link-
2594 layer address; otherwise it would not have be able to send the unicast
2595 solicitation in the first place. However, including the link-layer
2596 address in this case adds little overhead and eliminates a potential
2597 race condition where the sender deletes the cached link-layer address
2598 prior to receiving a response to a previous solicitation."
2600 ndisc_notify - BOOLEAN
2601 Define mode for notification of address and device changes.
2603 * 0 - (default): do nothing
2604 * 1 - Generate unsolicited neighbour advertisements when device is brought
2605 up or hardware address changes.
2607 ndisc_tclass - INTEGER
2608 The IPv6 Traffic Class to use by default when sending IPv6 Neighbor
2609 Discovery (Router Solicitation, Router Advertisement, Neighbor
2610 Solicitation, Neighbor Advertisement, Redirect) messages.
2611 These 8 bits can be interpreted as 6 high order bits holding the DSCP
2612 value and 2 low order bits representing ECN (which you probably want
2617 ndisc_evict_nocarrier - BOOLEAN
2618 Clears the neighbor discovery table on NOCARRIER events. This option is
2619 important for wireless devices where the neighbor discovery cache should
2620 not be cleared when roaming between access points on the same network.
2621 In most cases this should remain as the default (1).
2623 - 1 - (default): Clear neighbor discover cache on NOCARRIER events.
2624 - 0 - Do not clear neighbor discovery cache on NOCARRIER events.
2626 mldv1_unsolicited_report_interval - INTEGER
2627 The interval in milliseconds in which the next unsolicited
2628 MLDv1 report retransmit will take place.
2630 Default: 10000 (10 seconds)
2632 mldv2_unsolicited_report_interval - INTEGER
2633 The interval in milliseconds in which the next unsolicited
2634 MLDv2 report retransmit will take place.
2636 Default: 1000 (1 second)
2638 force_mld_version - INTEGER
2639 * 0 - (default) No enforcement of a MLD version, MLDv1 fallback allowed
2640 * 1 - Enforce to use MLD version 1
2641 * 2 - Enforce to use MLD version 2
2643 suppress_frag_ndisc - INTEGER
2644 Control RFC 6980 (Security Implications of IPv6 Fragmentation
2645 with IPv6 Neighbor Discovery) behavior:
2647 * 1 - (default) discard fragmented neighbor discovery packets
2648 * 0 - allow fragmented neighbor discovery packets
2650 optimistic_dad - BOOLEAN
2651 Whether to perform Optimistic Duplicate Address Detection (RFC 4429).
2653 * 0: disabled (default)
2656 Optimistic Duplicate Address Detection for the interface will be enabled
2657 if at least one of conf/{all,interface}/optimistic_dad is set to 1,
2658 it will be disabled otherwise.
2660 use_optimistic - BOOLEAN
2661 If enabled, do not classify optimistic addresses as deprecated during
2662 source address selection. Preferred addresses will still be chosen
2663 before optimistic addresses, subject to other ranking in the source
2664 address selection algorithm.
2666 * 0: disabled (default)
2669 This will be enabled if at least one of
2670 conf/{all,interface}/use_optimistic is set to 1, disabled otherwise.
2672 stable_secret - IPv6 address
2673 This IPv6 address will be used as a secret to generate IPv6
2674 addresses for link-local addresses and autoconfigured
2675 ones. All addresses generated after setting this secret will
2676 be stable privacy ones by default. This can be changed via the
2677 addrgenmode ip-link. conf/default/stable_secret is used as the
2678 secret for the namespace, the interface specific ones can
2679 overwrite that. Writes to conf/all/stable_secret are refused.
2681 It is recommended to generate this secret during installation
2682 of a system and keep it stable after that.
2684 By default the stable secret is unset.
2686 addr_gen_mode - INTEGER
2687 Defines how link-local and autoconf addresses are generated.
2689 = =================================================================
2690 0 generate address based on EUI64 (default)
2691 1 do no generate a link-local address, use EUI64 for addresses
2692 generated from autoconf
2693 2 generate stable privacy addresses, using the secret from
2694 stable_secret (RFC7217)
2695 3 generate stable privacy addresses, using a random secret if unset
2696 = =================================================================
2698 drop_unicast_in_l2_multicast - BOOLEAN
2699 Drop any unicast IPv6 packets that are received in link-layer
2700 multicast (or broadcast) frames.
2702 By default this is turned off.
2704 drop_unsolicited_na - BOOLEAN
2705 Drop all unsolicited neighbor advertisements, for example if there's
2706 a known good NA proxy on the network and such frames need not be used
2707 (or in the case of 802.11, must not be used to prevent attacks.)
2709 By default this is turned off.
2711 accept_untracked_na - INTEGER
2712 Define behavior for accepting neighbor advertisements from devices that
2713 are absent in the neighbor cache:
2715 - 0 - (default) Do not accept unsolicited and untracked neighbor
2718 - 1 - Add a new neighbor cache entry in STALE state for routers on
2719 receiving a neighbor advertisement (either solicited or unsolicited)
2720 with target link-layer address option specified if no neighbor entry
2721 is already present for the advertised IPv6 address. Without this knob,
2722 NAs received for untracked addresses (absent in neighbor cache) are
2725 This is as per router-side behavior documented in RFC9131.
2727 This has lower precedence than drop_unsolicited_na.
2729 This will optimize the return path for the initial off-link
2730 communication that is initiated by a directly connected host, by
2731 ensuring that the first-hop router which turns on this setting doesn't
2732 have to buffer the initial return packets to do neighbor-solicitation.
2733 The prerequisite is that the host is configured to send unsolicited
2734 neighbor advertisements on interface bringup. This setting should be
2735 used in conjunction with the ndisc_notify setting on the host to
2736 satisfy this prerequisite.
2738 - 2 - Extend option (1) to add a new neighbor cache entry only if the
2739 source IP address is in the same subnet as an address configured on
2740 the interface that received the neighbor advertisement.
2742 enhanced_dad - BOOLEAN
2743 Include a nonce option in the IPv6 neighbor solicitation messages used for
2744 duplicate address detection per RFC7527. A received DAD NS will only signal
2745 a duplicate address if the nonce is different. This avoids any false
2746 detection of duplicates due to loopback of the NS messages that we send.
2747 The nonce option will be sent on an interface unless both of
2748 conf/{all,interface}/enhanced_dad are set to FALSE.
2756 Limit the maximal rates for sending ICMPv6 messages.
2758 0 to disable any limiting,
2759 otherwise the minimal space between responses in milliseconds.
2763 ratemask - list of comma separated ranges
2764 For ICMPv6 message types matching the ranges in the ratemask, limit
2765 the sending of the message according to ratelimit parameter.
2767 The format used for both input and output is a comma separated
2768 list of ranges (e.g. "0-127,129" for ICMPv6 message type 0 to 127 and
2769 129). Writing to the file will clear all previous ranges of ICMPv6
2770 message types and update the current list with the input.
2772 Refer to: https://www.iana.org/assignments/icmpv6-parameters/icmpv6-parameters.xhtml
2773 for numerical values of ICMPv6 message types, e.g. echo request is 128
2774 and echo reply is 129.
2776 Default: 0-1,3-127 (rate limit ICMPv6 errors except Packet Too Big)
2778 echo_ignore_all - BOOLEAN
2779 If set non-zero, then the kernel will ignore all ICMP ECHO
2780 requests sent to it over the IPv6 protocol.
2784 echo_ignore_multicast - BOOLEAN
2785 If set non-zero, then the kernel will ignore all ICMP ECHO
2786 requests sent to it over the IPv6 protocol via multicast.
2790 echo_ignore_anycast - BOOLEAN
2791 If set non-zero, then the kernel will ignore all ICMP ECHO
2792 requests sent to it over the IPv6 protocol destined to anycast address.
2796 error_anycast_as_unicast - BOOLEAN
2797 If set to 1, then the kernel will respond with ICMP Errors
2798 resulting from requests sent to it over the IPv6 protocol destined
2799 to anycast address essentially treating anycast as unicast.
2803 xfrm6_gc_thresh - INTEGER
2804 (Obsolete since linux-4.14)
2805 The threshold at which we will start garbage collecting for IPv6
2806 destination cache entries. At twice this value the system will
2807 refuse new allocations.
2811 Pekka Savola <pekkas@netcore.fi>
2812 YOSHIFUJI Hideaki / USAGI Project <yoshfuji@linux-ipv6.org>
2815 /proc/sys/net/bridge/* Variables:
2816 =================================
2818 bridge-nf-call-arptables - BOOLEAN
2819 - 1 : pass bridged ARP traffic to arptables' FORWARD chain.
2824 bridge-nf-call-iptables - BOOLEAN
2825 - 1 : pass bridged IPv4 traffic to iptables' chains.
2830 bridge-nf-call-ip6tables - BOOLEAN
2831 - 1 : pass bridged IPv6 traffic to ip6tables' chains.
2836 bridge-nf-filter-vlan-tagged - BOOLEAN
2837 - 1 : pass bridged vlan-tagged ARP/IP/IPv6 traffic to {arp,ip,ip6}tables.
2842 bridge-nf-filter-pppoe-tagged - BOOLEAN
2843 - 1 : pass bridged pppoe-tagged IP/IPv6 traffic to {ip,ip6}tables.
2848 bridge-nf-pass-vlan-input-dev - BOOLEAN
2849 - 1: if bridge-nf-filter-vlan-tagged is enabled, try to find a vlan
2850 interface on the bridge and set the netfilter input device to the
2851 vlan. This allows use of e.g. "iptables -i br0.1" and makes the
2852 REDIRECT target work with vlan-on-top-of-bridge interfaces. When no
2853 matching vlan interface is found, or this switch is off, the input
2854 device is set to the bridge interface.
2856 - 0: disable bridge netfilter vlan interface lookup.
2860 ``proc/sys/net/sctp/*`` Variables:
2861 ==================================
2863 addip_enable - BOOLEAN
2864 Enable or disable extension of Dynamic Address Reconfiguration
2865 (ADD-IP) functionality specified in RFC5061. This extension provides
2866 the ability to dynamically add and remove new addresses for the SCTP
2869 1: Enable extension.
2871 0: Disable extension.
2876 Enable or disable pf (pf is short for potentially failed) state. A value
2877 of pf_retrans > path_max_retrans also disables pf state. That is, one of
2878 both pf_enable and pf_retrans > path_max_retrans can disable pf state.
2879 Since pf_retrans and path_max_retrans can be changed by userspace
2880 application, sometimes user expects to disable pf state by the value of
2881 pf_retrans > path_max_retrans, but occasionally the value of pf_retrans
2882 or path_max_retrans is changed by the user application, this pf state is
2883 enabled. As such, it is necessary to add this to dynamically enable
2884 and disable pf state. See:
2885 https://datatracker.ietf.org/doc/draft-ietf-tsvwg-sctp-failover for
2895 Unset or enable/disable pf (pf is short for potentially failed) state
2896 exposure. Applications can control the exposure of the PF path state
2897 in the SCTP_PEER_ADDR_CHANGE event and the SCTP_GET_PEER_ADDR_INFO
2898 sockopt. When it's unset, no SCTP_PEER_ADDR_CHANGE event with
2899 SCTP_ADDR_PF state will be sent and a SCTP_PF-state transport info
2900 can be got via SCTP_GET_PEER_ADDR_INFO sockopt; When it's enabled,
2901 a SCTP_PEER_ADDR_CHANGE event will be sent for a transport becoming
2902 SCTP_PF state and a SCTP_PF-state transport info can be got via
2903 SCTP_GET_PEER_ADDR_INFO sockopt; When it's disabled, no
2904 SCTP_PEER_ADDR_CHANGE event will be sent and it returns -EACCES when
2905 trying to get a SCTP_PF-state transport info via SCTP_GET_PEER_ADDR_INFO
2908 0: Unset pf state exposure, Compatible with old applications.
2910 1: Disable pf state exposure.
2912 2: Enable pf state exposure.
2916 addip_noauth_enable - BOOLEAN
2917 Dynamic Address Reconfiguration (ADD-IP) requires the use of
2918 authentication to protect the operations of adding or removing new
2919 addresses. This requirement is mandated so that unauthorized hosts
2920 would not be able to hijack associations. However, older
2921 implementations may not have implemented this requirement while
2922 allowing the ADD-IP extension. For reasons of interoperability,
2923 we provide this variable to control the enforcement of the
2924 authentication requirement.
2926 == ===============================================================
2927 1 Allow ADD-IP extension to be used without authentication. This
2928 should only be set in a closed environment for interoperability
2929 with older implementations.
2931 0 Enforce the authentication requirement
2932 == ===============================================================
2936 auth_enable - BOOLEAN
2937 Enable or disable Authenticated Chunks extension. This extension
2938 provides the ability to send and receive authenticated chunks and is
2939 required for secure operation of Dynamic Address Reconfiguration
2942 - 1: Enable this extension.
2943 - 0: Disable this extension.
2947 prsctp_enable - BOOLEAN
2948 Enable or disable the Partial Reliability extension (RFC3758) which
2949 is used to notify peers that a given DATA should no longer be expected.
2951 - 1: Enable extension
2957 The limit of the number of new packets that can be initially sent. It
2958 controls how bursty the generated traffic can be.
2962 association_max_retrans - INTEGER
2963 Set the maximum number for retransmissions that an association can
2964 attempt deciding that the remote end is unreachable. If this value
2965 is exceeded, the association is terminated.
2969 max_init_retransmits - INTEGER
2970 The maximum number of retransmissions of INIT and COOKIE-ECHO chunks
2971 that an association will attempt before declaring the destination
2972 unreachable and terminating.
2976 path_max_retrans - INTEGER
2977 The maximum number of retransmissions that will be attempted on a given
2978 path. Once this threshold is exceeded, the path is considered
2979 unreachable, and new traffic will use a different path when the
2980 association is multihomed.
2984 pf_retrans - INTEGER
2985 The number of retransmissions that will be attempted on a given path
2986 before traffic is redirected to an alternate transport (should one
2987 exist). Note this is distinct from path_max_retrans, as a path that
2988 passes the pf_retrans threshold can still be used. Its only
2989 deprioritized when a transmission path is selected by the stack. This
2990 setting is primarily used to enable fast failover mechanisms without
2991 having to reduce path_max_retrans to a very low value. See:
2992 http://www.ietf.org/id/draft-nishida-tsvwg-sctp-failover-05.txt
2993 for details. Note also that a value of pf_retrans > path_max_retrans
2994 disables this feature. Since both pf_retrans and path_max_retrans can
2995 be changed by userspace application, a variable pf_enable is used to
3000 ps_retrans - INTEGER
3001 Primary.Switchover.Max.Retrans (PSMR), it's a tunable parameter coming
3002 from section-5 "Primary Path Switchover" in rfc7829. The primary path
3003 will be changed to another active path when the path error counter on
3004 the old primary path exceeds PSMR, so that "the SCTP sender is allowed
3005 to continue data transmission on a new working path even when the old
3006 primary destination address becomes active again". Note this feature
3007 is disabled by initializing 'ps_retrans' per netns as 0xffff by default,
3008 and its value can't be less than 'pf_retrans' when changing by sysctl.
3012 rto_initial - INTEGER
3013 The initial round trip timeout value in milliseconds that will be used
3014 in calculating round trip times. This is the initial time interval
3015 for retransmissions.
3020 The maximum value (in milliseconds) of the round trip timeout. This
3021 is the largest time interval that can elapse between retransmissions.
3026 The minimum value (in milliseconds) of the round trip timeout. This
3027 is the smallest time interval the can elapse between retransmissions.
3031 hb_interval - INTEGER
3032 The interval (in milliseconds) between HEARTBEAT chunks. These chunks
3033 are sent at the specified interval on idle paths to probe the state of
3034 a given path between 2 associations.
3038 sack_timeout - INTEGER
3039 The amount of time (in milliseconds) that the implementation will wait
3044 valid_cookie_life - INTEGER
3045 The default lifetime of the SCTP cookie (in milliseconds). The cookie
3046 is used during association establishment.
3050 cookie_preserve_enable - BOOLEAN
3051 Enable or disable the ability to extend the lifetime of the SCTP cookie
3052 that is used during the establishment phase of SCTP association
3054 - 1: Enable cookie lifetime extension.
3059 cookie_hmac_alg - STRING
3060 Select the hmac algorithm used when generating the cookie value sent by
3061 a listening sctp socket to a connecting client in the INIT-ACK chunk.
3068 Ability to assign md5 or sha1 as the selected alg is predicated on the
3069 configuration of those algorithms at build time (CONFIG_CRYPTO_MD5 and
3070 CONFIG_CRYPTO_SHA1).
3072 Default: Dependent on configuration. MD5 if available, else SHA1 if
3073 available, else none.
3075 rcvbuf_policy - INTEGER
3076 Determines if the receive buffer is attributed to the socket or to
3077 association. SCTP supports the capability to create multiple
3078 associations on a single socket. When using this capability, it is
3079 possible that a single stalled association that's buffering a lot
3080 of data may block other associations from delivering their data by
3081 consuming all of the receive buffer space. To work around this,
3082 the rcvbuf_policy could be set to attribute the receiver buffer space
3083 to each association instead of the socket. This prevents the described
3086 - 1: rcvbuf space is per association
3087 - 0: rcvbuf space is per socket
3091 sndbuf_policy - INTEGER
3092 Similar to rcvbuf_policy above, this applies to send buffer space.
3094 - 1: Send buffer is tracked per association
3095 - 0: Send buffer is tracked per socket.
3099 sctp_mem - vector of 3 INTEGERs: min, pressure, max
3100 Number of pages allowed for queueing by all SCTP sockets.
3102 min: Below this number of pages SCTP is not bothered about its
3103 memory appetite. When amount of memory allocated by SCTP exceeds
3104 this number, SCTP starts to moderate memory usage.
3106 pressure: This value was introduced to follow format of tcp_mem.
3108 max: Number of pages allowed for queueing by all SCTP sockets.
3110 Default is calculated at boot time from amount of available memory.
3112 sctp_rmem - vector of 3 INTEGERs: min, default, max
3113 Only the first value ("min") is used, "default" and "max" are
3116 min: Minimal size of receive buffer used by SCTP socket.
3117 It is guaranteed to each SCTP socket (but not association) even
3118 under moderate memory pressure.
3122 sctp_wmem - vector of 3 INTEGERs: min, default, max
3123 Only the first value ("min") is used, "default" and "max" are
3126 min: Minimum size of send buffer that can be used by SCTP sockets.
3127 It is guaranteed to each SCTP socket (but not association) even
3128 under moderate memory pressure.
3132 addr_scope_policy - INTEGER
3133 Control IPv4 address scoping - draft-stewart-tsvwg-sctp-ipv4-00
3135 - 0 - Disable IPv4 address scoping
3136 - 1 - Enable IPv4 address scoping
3137 - 2 - Follow draft but allow IPv4 private addresses
3138 - 3 - Follow draft but allow IPv4 link local addresses
3143 The listening port for the local UDP tunneling sock. Normally it's
3144 using the IANA-assigned UDP port number 9899 (sctp-tunneling).
3146 This UDP sock is used for processing the incoming UDP-encapsulated
3147 SCTP packets (from RFC6951), and shared by all applications in the
3148 same net namespace. This UDP sock will be closed when the value is
3151 The value will also be used to set the src port of the UDP header
3152 for the outgoing UDP-encapsulated SCTP packets. For the dest port,
3153 please refer to 'encap_port' below.
3157 encap_port - INTEGER
3158 The default remote UDP encapsulation port.
3160 This value is used to set the dest port of the UDP header for the
3161 outgoing UDP-encapsulated SCTP packets by default. Users can also
3162 change the value for each sock/asoc/transport by using setsockopt.
3163 For further information, please refer to RFC6951.
3165 Note that when connecting to a remote server, the client should set
3166 this to the port that the UDP tunneling sock on the peer server is
3167 listening to and the local UDP tunneling sock on the client also
3168 must be started. On the server, it would get the encap_port from
3169 the incoming packet's source port.
3173 plpmtud_probe_interval - INTEGER
3174 The time interval (in milliseconds) for the PLPMTUD probe timer,
3175 which is configured to expire after this period to receive an
3176 acknowledgment to a probe packet. This is also the time interval
3177 between the probes for the current pmtu when the probe search
3180 PLPMTUD will be disabled when 0 is set, and other values for it
3185 reconf_enable - BOOLEAN
3186 Enable or disable extension of Stream Reconfiguration functionality
3187 specified in RFC6525. This extension provides the ability to "reset"
3188 a stream, and it includes the Parameters of "Outgoing/Incoming SSN
3189 Reset", "SSN/TSN Reset" and "Add Outgoing/Incoming Streams".
3191 - 1: Enable extension.
3192 - 0: Disable extension.
3196 intl_enable - BOOLEAN
3197 Enable or disable extension of User Message Interleaving functionality
3198 specified in RFC8260. This extension allows the interleaving of user
3199 messages sent on different streams. With this feature enabled, I-DATA
3200 chunk will replace DATA chunk to carry user messages if also supported
3201 by the peer. Note that to use this feature, one needs to set this option
3202 to 1 and also needs to set socket options SCTP_FRAGMENT_INTERLEAVE to 2
3203 and SCTP_INTERLEAVING_SUPPORTED to 1.
3205 - 1: Enable extension.
3206 - 0: Disable extension.
3210 ecn_enable - BOOLEAN
3211 Control use of Explicit Congestion Notification (ECN) by SCTP.
3212 Like in TCP, ECN is used only when both ends of the SCTP connection
3213 indicate support for it. This feature is useful in avoiding losses
3214 due to congestion by allowing supporting routers to signal congestion
3215 before having to drop packets.
3222 l3mdev_accept - BOOLEAN
3223 Enabling this option allows a "global" bound socket to work
3224 across L3 master domains (e.g., VRFs) with packets capable of
3225 being received regardless of the L3 domain in which they
3226 originated. Only valid when the kernel was compiled with
3227 CONFIG_NET_L3_MASTER_DEV.
3229 Default: 1 (enabled)
3232 ``/proc/sys/net/core/*``
3233 ========================
3235 Please see: Documentation/admin-guide/sysctl/net.rst for descriptions of these entries.
3238 ``/proc/sys/net/unix/*``
3239 ========================
3241 max_dgram_qlen - INTEGER
3242 The maximum length of dgram socket receive queue