Manual browser: setkey(8)

SETKEY(8) System Manager's Manual SETKEY(8)


setkeymanually manipulate the IPsec SA/SP database


setkey [-knrv] file ...

setkey [-knrv] -c

setkey [-krv] -f filename

setkey [-aklPrv] -D

setkey [-Pvp] -F

setkey [-H] -x

setkey [-?V]


setkey adds, updates, dumps, or flushes Security Association Database (SAD) entries as well as Security Policy Database (SPD) entries in the kernel.

setkey takes a series of operations from standard input (if invoked with -c) or the file named filename (if invoked with -f filename).

(no flag)
Dump the SAD entries or SPD entries contained in the specified file.
Print short help.
setkey usually does not display dead SAD entries with -D. If -a is also specified, the dead SAD entries will be displayed as well. A dead SAD entry is one that has expired but remains in the system because it is referenced by some SPD entries.
Dump the SAD entries. If -P is also specified, the SPD entries are dumped. If -p is specified, the ports are displayed.
Flush the SAD entries. If -P is also specified, the SPD entries are flushed.
Add hexadecimal dump in -x mode.
On NetBSD, synonym for -H. On other systems, synonym for -?.
Use semantics used in kernel. Available only in Linux. See also -r.
Loop forever with short output on -D.
No action. The program will check validity of the input, but no changes to the SPD will be made.
Use semantics described in IPsec RFCs. This mode is default. For details see section RFC vs Linux kernel semantics. Available only in Linux. See also -k.
Loop forever and dump all the messages transmitted to the PF_KEY socket. -xx prints the unformatted timestamps.
Print version string.
Be verbose. The program will dump messages exchanged on the PF_KEY socket, including messages sent from other processes to the kernel.

Configuration syntax

With -c or -f on the command line, setkey accepts the following configuration syntax. Lines starting with hash signs (‘#’) are treated as comment lines.
add [-46n] src dst protocol spi [extensions] algorithm ...;
Add an SAD entry. add can fail for multiple reasons, including when the key length does not match the specified algorithm.
get [-46n] src dst protocol spi;
Show an SAD entry.
delete [-46n] src dst protocol spi;
Remove an SAD entry.
deleteall [-46n] src dst protocol;
Remove all SAD entries that match the specification.
flush [protocol];
Clear all SAD entries matched by the options. -F on the command line achieves the same functionality.
dump [protocol];
Dumps all SAD entries matched by the options. -D on the command line achieves the same functionality.
spdadd [-46n] src_range dst_range upperspec label policy;
Add an SPD entry.
spdadd tagged tag policy;
Add an SPD entry based on a PF tag. tag must be a string surrounded by double quotes.
spdupdate [-46n] src_range dst_range upperspec label policy;
Updates an SPD entry.
spdupdate tagged tag policy;
Update an SPD entry based on a PF tag. tag must be a string surrounded by double quotes.
spddelete [-46n] src_range dst_range upperspec -P direction;
Delete an SPD entry.
Clear all SPD entries. -FP on the command line achieves the same functionality.
Dumps all SPD entries. -DP on the command line achieves the same functionality.

Meta-arguments are as follows:

Source/destination of the secure communication is specified as an IPv4/v6 address, and an optional port number between square brackets. setkey can resolve a FQDN into numeric addresses. If the FQDN resolves into multiple addresses, setkey will install multiple SAD/SPD entries into the kernel by trying all possible combinations. -4, -6, and -n restrict the address resolution of FQDN in certain ways. -4 and -6 restrict results into IPv4/v6 addresses only, respectively. -n avoids FQDN resolution and requires addresses to be numeric addresses.

protocol is one of following:
ESP based on rfc2406
ESP based on rfc1827
UDP encapsulated ESP for NAT traversal (rfc3948)
AH based on rfc2402
AH based on rfc1826
TCP-MD5 based on rfc2385

Security Parameter Index (SPI) for the SAD and the SPD. spi must be a decimal number, or a hexadecimal number with a “0x” prefix. SPI values between 0 and 255 are reserved for future use by IANA and cannot be used. TCP-MD5 associations must use 0x1000 and therefore only have per-host granularity at this time.

take some of the following:
-m mode
Specify a security protocol mode for use. mode is one of following: transport, tunnel, or any. The default value is any.
-r size
Specify window size of bytes for replay prevention. size must be decimal number in 32-bit word. If size is zero or not specified, replay checks don't take place.
-u id
Specify the identifier of the policy entry in the SPD. See policy.
-f pad_option
defines the content of the ESP padding. pad_option is one of following:
All the paddings are zero.
A series of randomized values are used.
A series of sequential increasing numbers started from 1 are used.
-f nocyclic-seq
Don't allow cyclic sequence numbers.
-lh time
-ls time
Specify hard/soft life time duration of the SA measured in seconds.
-bh bytes
-bs bytes
Specify hard/soft life time duration of the SA measured in bytes transported.
-esp_frag bytes
Specify esp fragment size for NAT-T (only valid for NAT-T SAs).
-ctx doi algorithm context-name
Specify an access control label. The access control label is interpreted by the LSM (e.g., SELinux). Ultimately, it enables MAC on network communications.
The domain of interpretation, which is used by the IKE daemon to identify the domain in which negotiation takes place.
Indicates the LSM for which the label is generated (e.g., SELinux).
The string representation of the label that is interpreted by the LSM.

-E ealgo key
Specify an encryption algorithm ealgo for ESP.
-E ealgo key -A aalgo key
Specify an encryption algorithm ealgo, as well as a payload authentication algorithm aalgo, for ESP.
-A aalgo key
Specify an authentication algorithm for AH.
-C calgo [-R]
Specify a compression algorithm for IPComp. If -R is specified, the spi field value will be used as the IPComp CPI (compression parameter index) on wire as-is. If -R is not specified, the kernel will use well-known CPI on wire, and spi field will be used only as an index for kernel internal usage.

key must be a double-quoted character string, or a series of hexadecimal digits preceded by “0x”.

Possible values for ealgo, aalgo, and calgo are specified in the Algorithms sections.

These select the communications that should be secured by IPsec. They can be an IPv4/v6 address or an IPv4/v6 address range, and may be accompanied by a TCP/UDP port specification. This takes the following form:


prefixlen and port must be decimal numbers. The square brackets around port are really necessary, they are not man page meta-characters. For FQDN resolution, the rules applicable to src and dst apply here as well.

Upper-layer protocol to be used. You can use one of the words in /etc/protocols as upperspec, or icmp6, ip4, gre, or any. any stands for “any protocol”. You can also use the protocol number. Additional specification can be placed after the protocol name for some protocols. You can specify a type and/or a code of ICMP or ICMPv6. The type is separated from a code by single comma and the code must always be specified. GRE key can be specified in dotted-quad format or as plain number. When a zero is specified, the kernel deals with it as a wildcard. Note that the kernel can not distinguish a wildcard from an ICPMv6 type of zero.

For example, the following means that the policy doesn't require IPsec for any inbound Neighbor Solicitation.

spdadd ::/0 ::/0 icmp6 135,0 -P in none;

A second example of requiring transport mode encryption of specific GRE tunnel:

spdadd gre 1234 ipsec esp/transport//require;

Note: upperspec does not work against forwarding case at this moment, as it requires extra reassembly at the forwarding node (not implemented at this moment). There are many protocols in /etc/protocols, but all protocols except of TCP, UDP, GRE, and ICMP may not be suitable to use with IPsec. You have to consider carefully what to use.

label is the access control label for the policy. This label is interpreted by the LSM (e.g., SELinux). Ultimately, it enables MAC on network communications. When a policy contains an access control label, SAs negotiated with this policy will contain the label. Its format:
-ctx doi algorithm context-name
The domain of interpretation, which is used by the IKE daemon to identify the domain in which negotiation takes place.
Indicates the LSM for which the label is generated (e.g., SELinux).
The string representation of the label that is interpreted by the LSM.

policy is in one of the following three formats:
  • -P direction [priority specification] discard
  • -P direction [priority specification] none
  • -P direction [priority specification] ipsec protocol/mode/src-dst/level [...]

You must specify the direction of its policy as direction. Either out, in, or fwd can be used.

priority specification is used to control the placement of the policy within the SPD. Policy position is determined by a signed integer where higher priorities indicate the policy is placed closer to the beginning of the list and lower priorities indicate the policy is placed closer to the end of the list. Policies with equal priorities are added at the end of groups of such policies.

Priority can only be specified when setkey has been compiled against kernel headers that support policy priorities (Linux >= 2.6.6). If the kernel does not support priorities, a warning message will be printed the first time a priority specification is used. Policy priority takes one of the following formats:

{priority,prio} offset
offset is an integer in the range from -2147483647 to 214783648.
{priority,prio} base {+,-} offset
base is either low (-1073741824), def (0), or high (1073741824)

offset is an unsigned integer. It can be up to 1073741824 for positive offsets, and up to 1073741823 for negative offsets.

discard means the packet matching indexes will be discarded. none means that IPsec operation will not take place onto the packet. ipsec means that IPsec operation will take place onto the packet.

The protocol/mode/src-dst/level part specifies the rule how to process the packet. Either ah, esp, or ipcomp must be used as protocol. mode is either transport or tunnel. If mode is tunnel, you must specify the end-point addresses of the SA as src and dst with ‘-’ between these addresses, which is used to specify the SA to use. If mode is transport, both src and dst can be omitted. level is to be one of the following: default, use, require, or unique. If the SA is not available in every level, the kernel will ask the key exchange daemon to establish a suitable SA. default means the kernel consults the system wide default for the protocol you specified, e.g. the esp_trans_deflev sysctl variable, when the kernel processes the packet. use means that the kernel uses an SA if it's available, otherwise the kernel keeps normal operation. require means SA is required whenever the kernel sends a packet matched with the policy. unique is the same as require; in addition, it allows the policy to match the unique out-bound SA. You just specify the policy level unique, racoon(8) will configure the SA for the policy. If you configure the SA by manual keying for that policy, you can put a decimal number as the policy identifier after unique separated by a colon ‘:’ like: unique:number in order to bind this policy to the SA. number must be between 1 and 32767. It corresponds to extensions -u of the manual SA configuration. When you want to use SA bundle, you can define multiple rules. For example, if an IP header was followed by an AH header followed by an ESP header followed by an upper layer protocol header, the rule would be:

esp/transport//require ah/transport//require;
The rule order is very important.

When NAT-T is enabled in the kernel, policy matching for ESP over UDP packets may be done on endpoint addresses and port (this depends on the system. System that do not perform the port check cannot support multiple endpoints behind the same NAT). When using ESP over UDP, you can specify port numbers in the endpoint addresses to get the correct matching. Here is an example:

spdadd[any][any] any -P out ipsec 
    esp/tunnel/[4500]-[30000]/require ; 
These ports must be left unspecified (which defaults to 0) for anything other than ESP over UDP. They can be displayed in SPD dump using setkey -DPp.

Note that “discard” and “none” are not in the syntax described in ipsec_set_policy(3). There are a few differences in the syntax. See ipsec_set_policy(3) for detail.


The following list shows the supported algorithms. protocol and algorithm are almost orthogonal. These authentication algorithms can be used as aalgo in -A aalgo of the protocol parameter:

algorithm	keylen (bits) 
hmac-md5	128		ah: rfc2403 
		128		ah-old: rfc2085 
hmac-sha1	160		ah: rfc2404 
		160		ah-old: 128bit ICV (no document) 
keyed-md5	128		ah: 96bit ICV (no document) 
		128		ah-old: rfc1828 
keyed-sha1	160		ah: 96bit ICV (no document) 
		160		ah-old: 128bit ICV (no document) 
null		0 to 2048	for debugging 
hmac-sha256	256		ah: 128bit ICV (RFC4868) 
		256		ah-old: 128bit ICV (no document) 
hmac-sha384	384		ah: 192bit ICV (RFC4868) 
		384		ah-old: 128bit ICV (no document) 
hmac-sha512	512		ah: 256bit ICV (RFC4868) 
		512		ah-old: 128bit ICV (no document) 
hmac-ripemd160	160		ah: 96bit ICV (RFC2857) 
				ah-old: 128bit ICV (no document) 
aes-xcbc-mac	128		ah: 96bit ICV (RFC3566) 
		128		ah-old: 128bit ICV (no document) 
tcp-md5		8 to 640	tcp: rfc2385

These encryption algorithms can be used as ealgo in -E ealgo of the protocol parameter:

algorithm	keylen (bits) 
des-cbc		64		esp-old: rfc1829, esp: rfc2405 
3des-cbc	192		rfc2451 
null		0 to 2048	rfc2410 
blowfish-cbc	40 to 448	rfc2451 
cast128-cbc	40 to 128	rfc2451 
des-deriv	64		ipsec-ciph-des-derived-01 
3des-deriv	192		no document 
rijndael-cbc	128/192/256	rfc3602 
twofish-cbc	0 to 256	draft-ietf-ipsec-ciph-aes-cbc-01 
aes-ctr		160/224/288	rfc3686 
camellia-cbc	128/192/256	rfc4312 
aes-gcm-16	160/224/288	rfc4106 
aes-gmac	160/224/288	rfc4543

Note that the first 128/192/256 bits of a key for aes-ctr, aes-gcm-16 or aes-gmac will be used as AES key, and the remaining 32 bits will be used as nonce. Also note that aes-gmac does not encrypt the payload, it only provides authentication.

These compression algorithms can be used as calgo in -C calgo of the protocol parameter:

deflate		rfc2394

RFC vs Linux kernel semantics

The Linux kernel uses the fwd policy instead of the in policy for packets what are forwarded through that particular box.

In kernel mode, setkey manages and shows policies and SAs exactly as they are stored in the kernel.

In RFC mode, setkey

  • creates fwd policies for every in policy inserted
  • (not implemented yet) filters out all fwd policies


The command exits with 0 on success, and non-zero on errors.


add 3ffe:501:4819::1 3ffe:501:481d::1 esp 123457 
	-E des-cbc 0x3ffe05014819ffff ; 
add -6 ah 123456 
	-A hmac-sha1 "AH SA configuration!" ; 
add esp 0x10001 
	-E des-cbc 0x3ffe05014819ffff 
	-A hmac-md5 "authentication!!" ; 
get 3ffe:501:4819::1 3ffe:501:481d::1 ah 123456 ; 
flush ; 
dump esp ; 
spdadd[21][any] any 
	-P out ipsec esp/tunnel/ ; 
add tcp 0x1000 -A tcp-md5 "TCP-MD5 BGP secret" ; 
add esp 0x10001 
	-ctx 1 1 "system_u:system_r:unconfined_t:SystemLow-SystemHigh" 
	-E des-cbc 0x3ffe05014819ffff; 
spdadd any 
	-ctx 1 1 "system_u:system_r:unconfined_t:SystemLow-SystemHigh" 
	-P out ipsec esp/transport//require ;


ipsec_set_policy(3), racoon(8), sysctl(8)

Changed manual key configuration for IPsec,, October 1999.


The setkey command first appeared in the WIDE Hydrangea IPv6 protocol stack kit. The command was completely re-designed in June 1998.


setkey should report and handle syntax errors better.

For IPsec gateway configuration, src_range and dst_range with TCP/UDP port numbers does not work, as the gateway does not reassemble packets (it cannot inspect upper-layer headers).

February 18, 2012 NetBSD 7.0