NTP_CONF(5) | File | NTP_CONF(5) |
ntp.conf | [--option-name] [--option-name value] |
All arguments must be options.
The file format is similar to other UNIX configuration files. Comments begin with a ‘#
’ character and extend to the end of the line; blank lines are ignored. Configuration commands consist of an initial keyword followed by a list of arguments, some of which may be optional, separated by whitespace. Commands may not be continued over multiple lines. Arguments may be host names, host addresses written in numeric, dotted-quad form, integers, floating point numbers (when specifying times in seconds) and text strings.
The rest of this page describes the configuration and control options. The “Notes on Configuring NTP and Setting up an NTP Subnet” page (available as part of the HTML documentation provided in /usr/share/doc/ntp) contains an extended discussion of these options. In addition to the discussion of general Configuration Options, there are sections describing the following supported functionality and the options used to control it:
Following these is a section describing Miscellaneous Options. While there is a rich set of options available, the only required option is one or more pool, server, peer, broadcast or manycastclient commands.
If the Basic Socket Interface Extensions for IPv6 (RFC-2553) is detected, support for the IPv6 address family is generated in addition to the default support of the IPv4 address family. In a few cases, including the reslist billboard generated by ntpdc, IPv6 addresses are automatically generated. IPv6 addresses can be identified by the presence of colons “:” in the address field. IPv6 addresses can be used almost everywhere where IPv4 addresses can be used, with the exception of reference clock addresses, which are always IPv4.
Note that in contexts where a host name is expected, a -4 qualifier preceding the host name forces DNS resolution to the IPv4 namespace, while a -6 qualifier forces DNS resolution to the IPv6 namespace. See IPv6 references for the equivalent classes for that address family.
These five commands specify the time server name or address to be used and the mode in which to operate. The address can be either a DNS name or an IP address in dotted-quad notation. Additional information on association behavior can be found in the “Association Management” page (available as part of the HTML documentation provided in /usr/share/doc/ntp).
Options:
NTPv4 retains the NTPv3 scheme, properly described as symmetric key cryptography and, in addition, provides a new Autokey scheme based on public key cryptography. Public key cryptography is generally considered more secure than symmetric key cryptography, since the security is based on a private value which is generated by each server and never revealed. With Autokey all key distribution and management functions involve only public values, which considerably simplifies key distribution and storage. Public key management is based on X.509 certificates, which can be provided by commercial services or produced by utility programs in the OpenSSL software library or the NTPv4 distribution.
While the algorithms for symmetric key cryptography are included in the NTPv4 distribution, public key cryptography requires the OpenSSL software library to be installed before building the NTP distribution. Directions for doing that are on the Building and Installing the Distribution page.
Authentication is configured separately for each association using the key or autokey subcommand on the peer, server, broadcast and manycastclient configuration commands as described in Configuration Options page. The authentication options described below specify the locations of the key files, if other than default, which symmetric keys are trusted and the interval between various operations, if other than default.
Authentication is always enabled, although ineffective if not configured as described below. If a NTP packet arrives including a message authentication code (MAC), it is accepted only if it passes all cryptographic checks. The checks require correct key ID, key value and message digest. If the packet has been modified in any way or replayed by an intruder, it will fail one or more of these checks and be discarded. Furthermore, the Autokey scheme requires a preliminary protocol exchange to obtain the server certificate, verify its credentials and initialize the protocol
The auth flag controls whether new associations or remote configuration commands require cryptographic authentication. This flag can be set or reset by the enable and disable commands and also by remote configuration commands sent by a ntpdc(8) program running in another machine. If this flag is enabled, which is the default case, new broadcast client and symmetric passive associations and remote configuration commands must be cryptographically authenticated using either symmetric key or public key cryptography. If this flag is disabled, these operations are effective even if not cryptographic authenticated. It should be understood that operating with the auth flag disabled invites a significant vulnerability where a rogue hacker can masquerade as a falseticker and seriously disrupt system timekeeping. It is important to note that this flag has no purpose other than to allow or disallow a new association in response to new broadcast and symmetric active messages and remote configuration commands and, in particular, the flag has no effect on the authentication process itself.
An attractive alternative where multicast support is available is manycast mode, in which clients periodically troll for servers as described in the Automatic NTP Configuration Options page. Either symmetric key or public key cryptographic authentication can be used in this mode. The principle advantage of manycast mode is that potential servers need not be configured in advance, since the client finds them during regular operation, and the configuration files for all clients can be identical.
The security model and protocol schemes for both symmetric key and public key cryptography are summarized below; further details are in the briefings, papers and reports at the NTP project page linked from http://www.ntp.org/
.
When ntpd(8) is first started, it reads the key file specified in the keys configuration command and installs the keys in the key cache. However, individual keys must be activated with the trusted command before use. This allows, for instance, the installation of possibly several batches of keys and then activating or deactivating each batch remotely using ntpdc(8). This also provides a revocation capability that can be used if a key becomes compromised. The requestkey command selects the key used as the password for the ntpdc(8) utility, while the controlkey command selects the key used as the password for the ntpq(8) utility.
The Autokey protocol has several modes of operation corresponding to the various NTP modes supported. Most modes use a special cookie which can be computed independently by the client and server, but encrypted in transmission. All modes use in addition a variant of the S-KEY scheme, in which a pseudo-random key list is generated and used in reverse order. These schemes are described along with an executive summary, current status, briefing slides and reading list on the Autonomous Authentication page.
The specific cryptographic environment used by Autokey servers and clients is determined by a set of files and soft links generated by the ntp-keygen(1ntpkeygenmdoc) program. This includes a required host key file, required certificate file and optional sign key file, leapsecond file and identity scheme files. The digest/signature scheme is specified in the X.509 certificate along with the matching sign key. There are several schemes available in the OpenSSL software library, each identified by a specific string such as md5WithRSAEncryption, which stands for the MD5 message digest with RSA encryption scheme. The current NTP distribution supports all the schemes in the OpenSSL library, including those based on RSA and DSA digital signatures.
NTP secure groups can be used to define cryptographic compartments and security hierarchies. It is important that every host in the group be able to construct a certificate trail to one or more trusted hosts in the same group. Each group host runs the Autokey protocol to obtain the certificates for all hosts along the trail to one or more trusted hosts. This requires the configuration file in all hosts to be engineered so that, even under anticipated failure conditions, the NTP subnet will form such that every group host can find a trail to at least one trusted host.
By convention, the name of an Autokey host is the name returned by the Unix gethostname(2) system call or equivalent in other systems. By the system design model, there are no provisions to allow alternate names or aliases. However, this is not to say that DNS aliases, different names for each interface, etc., are constrained in any way.
It is also important to note that Autokey verifies authenticity using the host name, network address and public keys, all of which are bound together by the protocol specifically to deflect masquerade attacks. For this reason Autokey includes the source and destinatino IP addresses in message digest computations and so the same addresses must be available at both the server and client. For this reason operation with network address translation schemes is not possible. This reflects the intended robust security model where government and corporate NTP servers are operated outside firewall perimeters.
The cryptotype of an association is determined at the time of mobilization, either at configuration time or some time later when a message of appropriate cryptotype arrives. When mobilized by a server or peer configuration command and no key or autokey subcommands are present, the association is not authenticated; if the key subcommand is present, the association is authenticated using the symmetric key ID specified; if the autokey subcommand is present, the association is authenticated using Autokey.
When multiple identity schemes are supported in the Autokey protocol, the first message exchange determines which one is used. The client request message contains bits corresponding to which schemes it has available. The server response message contains bits corresponding to which schemes it has available. Both server and client match the received bits with their own and select a common scheme.
Following the principle that time is a public value, a server responds to any client packet that matches its cryptotype capabilities. Thus, a server receiving an unauthenticated packet will respond with an unauthenticated packet, while the same server receiving a packet of a cryptotype it supports will respond with packets of that cryptotype. However, unconfigured broadcast or manycast client associations or symmetric passive associations will not be mobilized unless the server supports a cryptotype compatible with the first packet received. By default, unauthenticated associations will not be mobilized unless overridden in a decidedly dangerous way.
Some examples may help to reduce confusion. Client Alice has no specific cryptotype selected. Server Bob has both a symmetric key file and minimal Autokey files. Alice's unauthenticated messages arrive at Bob, who replies with unauthenticated messages. Cathy has a copy of Bob's symmetric key file and has selected key ID 4 in messages to Bob. Bob verifies the message with his key ID 4. If it's the same key and the message is verified, Bob sends Cathy a reply authenticated with that key. If verification fails, Bob sends Cathy a thing called a crypto-NAK, which tells her something broke. She can see the evidence using the ntpq(8) program.
Denise has rolled her own host key and certificate. She also uses one of the identity schemes as Bob. She sends the first Autokey message to Bob and they both dance the protocol authentication and identity steps. If all comes out okay, Denise and Bob continue as described above.
It should be clear from the above that Bob can support all the girls at the same time, as long as he has compatible authentication and identity credentials. Now, Bob can act just like the girls in his own choice of servers; he can run multiple configured associations with multiple different servers (or the same server, although that might not be useful). But, wise security policy might preclude some cryptotype combinations; for instance, running an identity scheme with one server and no authentication with another might not be wise.
Certificates imported from OpenSSL or public certificate authorities have certian limitations. The certificate should be in ASN.1 syntax, X.509 Version 3 format and encoded in PEM, which is the same format used by OpenSSL. The overall length of the certificate encoded in ASN.1 must not exceed 1024 bytes. The subject distinguished name field (CN) is the fully qualified name of the host on which it is used; the remaining subject fields are ignored. The certificate extension fields must not contain either a subject key identifier or a issuer key identifier field; however, an extended key usage field for a trusted host must contain the value trustRoot;. Other extension fields are ignored.
49213 525.624 127.127.4.1 93 226 00:08:29.606 D
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The next field shows the clock address in dotted-quad notation. The final field shows the last timecode received from the clock in decoded ASCII format, where meaningful. In some clock drivers a good deal of additional information can be gathered and displayed as well. See information specific to each clock for further details.
49213 525.624 127.127.4.1 message
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The next field shows the peer address in dotted-quad notation, The final message field includes the message type and certain ancillary information. See the Authentication Options section for further information.
50935 75440.031 0.000006019 13.778190 0.000351733 0.0133806
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The next five fields show time offset (seconds), frequency offset (parts per million - PPM), RMS jitter (seconds), Allan deviation (PPM) and clock discipline time constant.
48773 10847.650 127.127.4.1 9714 -0.001605376 0.000000000 0.001424877 0.000958674
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The next two fields show the peer address in dotted-quad notation and status, respectively. The status field is encoded in hex in the format described in Appendix A of the NTP specification RFC 1305. The final four fields show the offset, delay, dispersion and RMS jitter, all in seconds.
50928 2132.543 128.4.1.1 128.4.1.20 3102453281.584327000 3102453281.58622800031 02453332.540806000 3102453332.541458000
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The next two fields show the remote peer or clock address followed by the local address in dotted-quad notation. The final four fields show the originate, receive, transmit and final NTP timestamps in order. The timestamp values are as received and before processing by the various data smoothing and mitigation algorithms.
50928 2132.543 36000 81965 0 9546 56 71793 512 540 10 147
The first two fields show the date (Modified Julian Day) and time (seconds and fraction past UTC midnight). The remaining ten fields show the statistics counter values accumulated since the last generated line.
Note that this command can be sent from the ntpdc(8) program running at a remote location.
/
’). This can be modified using the file argument to the filegen statement. No .. elements are allowed in this component to prevent filenames referring to parts outside the filesystem hierarchy denoted by prefix..
’ to concatenated prefix and filename strings, and appending the decimal representation of the process ID of the ntpd(8) server process..
’ and a day specification in the form YYYYMMdd. YYYY is a 4-digit year number (e.g., 1992). MM is a two digit month number. dd is a two digit day number. Thus, all information written at 10 December 1992 would end up in a file named prefix filename.19921210.The restriction facility was implemented in conformance with the access policies for the original NSFnet backbone time servers. Later the facility was expanded to deflect cryptographic and clogging attacks. While this facility may be useful for keeping unwanted or broken or malicious clients from congesting innocent servers, it should not be considered an alternative to the NTP authentication facilities. Source address based restrictions are easily circumvented by a determined cracker.
Clients can be denied service because they are explicitly included in the restrict list created by the restrict command or implicitly as the result of cryptographic or rate limit violations. Cryptographic violations include certificate or identity verification failure; rate limit violations generally result from defective NTP implementations that send packets at abusive rates. Some violations cause denied service only for the offending packet, others cause denied service for a timed period and others cause the denied service for an indefinate period. When a client or network is denied access for an indefinate period, the only way at present to remove the restrictions is by restarting the server.
A client receiving a KoD performs a set of sanity checks to minimize security exposure, then updates the stratum and reference identifier peer variables, sets the access denied (TEST4) bit in the peer flash variable and sends a message to the log. As long as the TEST4 bit is set, the client will send no further packets to the server. The only way at present to recover from this condition is to restart the protocol at both the client and server. This happens automatically at the client when the association times out. It will happen at the server only if the server operator cooperates.
Default restriction list entries with the flags ignore, interface, ntpport, for each of the local host's interface addresses are inserted into the table at startup to prevent the server from attempting to synchronize to its own time. A default entry is also always present, though if it is otherwise unconfigured; no flags are associated with the default entry (i.e., everything besides your own NTP server is unrestricted).
Note that the manycasting paradigm does not coincide with the anycast paradigm described in RFC-1546, which is designed to find a single server from a clique of servers providing the same service. The manycast paradigm is designed to find a plurality of redundant servers satisfying defined optimality criteria.
Manycasting can be used with either symmetric key or public key cryptography. The public key infrastructure (PKI) offers the best protection against compromised keys and is generally considered stronger, at least with relatively large key sizes. It is implemented using the Autokey protocol and the OpenSSL cryptographic library available from http://www.openssl.org/
. The library can also be used with other NTPv4 modes as well and is highly recommended, especially for broadcast modes.
A persistent manycast client association is configured using the manycastclient command, which is similar to the server command but with a multicast (IPv4 class D or IPv6 prefix FF) group address. The IANA has designated IPv4 address 224.1.1.1 and IPv6 address FF05::101 (site local) for NTP. When more servers are needed, it broadcasts manycast client messages to this address at the minimum feasible rate and minimum feasible time-to-live (TTL) hops, depending on how many servers have already been found. There can be as many manycast client associations as different group address, each one serving as a template for a future ephemeral unicast client/server association.
Manycast servers configured with the manycastserver command listen on the specified group address for manycast client messages. Note the distinction between manycast client, which actively broadcasts messages, and manycast server, which passively responds to them. If a manycast server is in scope of the current TTL and is itself synchronized to a valid source and operating at a stratum level equal to or lower than the manycast client, it replies to the manycast client message with an ordinary unicast server message.
The manycast client receiving this message mobilizes an ephemeral client/server association according to the matching manycast client template, but only if cryptographically authenticated and the server stratum is less than or equal to the client stratum. Authentication is explicitly required and either symmetric key or public key (Autokey) can be used. Then, the client polls the server at its unicast address in burst mode in order to reliably set the host clock and validate the source. This normally results in a volley of eight client/server at 2-s intervals during which both the synchronization and cryptographic protocols run concurrently. Following the volley, the client runs the NTP intersection and clustering algorithms, which act to discard all but the "best" associations according to stratum and synchronization distance. The surviving associations then continue in ordinary client/server mode.
The manycast client polling strategy is designed to reduce as much as possible the volume of manycast client messages and the effects of implosion due to near-simultaneous arrival of manycast server messages. The strategy is determined by the manycastclient, tos and ttl configuration commands. The manycast poll interval is normally eight times the system poll interval, which starts out at the minpoll value specified in the manycastclient, command and, under normal circumstances, increments to the maxpolll value specified in this command. Initially, the TTL is set at the minimum hops specified by the ttl command. At each retransmission the TTL is increased until reaching the maximum hops specified by this command or a sufficient number client associations have been found. Further retransmissions use the same TTL.
The quality and reliability of the suite of associations discovered by the manycast client is determined by the NTP mitigation algorithms and the minclock and minsane values specified in the tos configuration command. At least minsane candidate servers must be available and the mitigation algorithms produce at least minclock survivors in order to synchronize the clock. Byzantine agreement principles require at least four candidates in order to correctly discard a single falseticker. For legacy purposes, minsane defaults to 1 and minclock defaults to 3. For manycast service minsane should be explicitly set to 4, assuming at least that number of servers are available.
If at least minclock servers are found, the manycast poll interval is immediately set to eight times maxpoll. If less than minclock servers are found when the TTL has reached the maximum hops, the manycast poll interval is doubled. For each transmission after that, the poll interval is doubled again until reaching the maximum of eight times maxpoll. Further transmissions use the same poll interval and TTL values. Note that while all this is going on, each client/server association found is operating normally it the system poll interval.
Administratively scoped multicast boundaries are normally specified by the network router configuration and, in the case of IPv6, the link/site scope prefix. By default, the increment for TTL hops is 32 starting from 31; however, the ttl configuration command can be used to modify the values to match the scope rules.
It is often useful to narrow the range of acceptable servers which can be found by manycast client associations. Because manycast servers respond only when the client stratum is equal to or greater than the server stratum, primary (stratum 1) servers fill find only primary servers in TTL range, which is probably the most common objective. However, unless configured otherwise, all manycast clients in TTL range will eventually find all primary servers in TTL range, which is probably not the most common objective in large networks. The tos command can be used to modify this behavior. Servers with stratum below floor or above ceiling specified in the tos command are strongly discouraged during the selection process; however, these servers may be temporally accepted if the number of servers within TTL range is less than minclock.
The above actions occur for each manycast client message, which repeats at the designated poll interval. However, once the ephemeral client association is mobilized, subsequent manycast server replies are discarded, since that would result in a duplicate association. If during a poll interval the number of client associations falls below minclock, all manycast client prototype associations are reset to the initial poll interval and TTL hops and operation resumes from the beginning. It is important to avoid frequent manycast client messages, since each one requires all manycast servers in TTL range to respond. The result could well be an implosion, either minor or major, depending on the number of servers in range. The recommended value for maxpoll is 12 (4,096 s).
It is possible and frequently useful to configure a host as both manycast client and manycast server. A number of hosts configured this way and sharing a common group address will automatically organize themselves in an optimum configuration based on stratum and synchronization distance. For example, consider an NTP subnet of two primary servers and a hundred or more dependent clients. With two exceptions, all servers and clients have identical configuration files including both multicastclient and multicastserver commands using, for instance, multicast group address 239.1.1.1. The only exception is that each primary server configuration file must include commands for the primary reference source such as a GPS receiver.
The remaining configuration files for all secondary servers and clients have the same contents, except for the tos command, which is specific for each stratum level. For stratum 1 and stratum 2 servers, that command is not necessary. For stratum 3 and above servers the floor value is set to the intended stratum number. Thus, all stratum 3 configuration files are identical, all stratum 4 files are identical and so forth.
Once operations have stabilized in this scenario, the primary servers will find the primary reference source and each other, since they both operate at the same stratum (1), but not with any secondary server or client, since these operate at a higher stratum. The secondary servers will find the servers at the same stratum level. If one of the primary servers loses its GPS receiver, it will continue to operate as a client and other clients will time out the corresponding association and re-associate accordingly.
Some administrators prefer to avoid running ntpd(8) continuously and run either ntpdate(8) or ntpd(8) -q as a cron job. In either case the servers must be configured in advance and the program fails if none are available when the cron job runs. A really slick application of manycast is with ntpd(8) -q. The program wakes up, scans the local landscape looking for the usual suspects, selects the best from among the rascals, sets the clock and then departs. Servers do not have to be configured in advance and all clients throughout the network can have the same configuration file.
About once an hour or less often if the poll interval exceeds this, the client regenerates the Autokey key list. This is in general transparent in client/server mode. However, about once per day the server private value used to generate cookies is refreshed along with all manycast client associations. In this case all cryptographic values including certificates is refreshed. If a new certificate has been generated since the last refresh epoch, it will automatically revoke all prior certificates that happen to be in the certificate cache. At the same time, the manycast scheme starts all over from the beginning and the expanding ring shrinks to the minimum and increments from there while collecting all servers in scope.
A reference clock will generally (though not always) be a radio timecode receiver which is synchronized to a source of standard time such as the services offered by the NRC in Canada and NIST and USNO in the US. The interface between the computer and the timecode receiver is device dependent, but is usually a serial port. A device driver specific to each reference clock must be selected and compiled in the distribution; however, most common radio, satellite and modem clocks are included by default. Note that an attempt to configure a reference clock when the driver has not been compiled or the hardware port has not been appropriately configured results in a scalding remark to the system log file, but is otherwise non hazardous.
For the purposes of configuration, ntpd(8) treats reference clocks in a manner analogous to normal NTP peers as much as possible. Reference clocks are identified by a syntactically correct but invalid IP address, in order to distinguish them from normal NTP peers. Reference clock addresses are of the form 127.127.
t.u, where t is an integer denoting the clock type and u indicates the unit number in the range 0-3. While it may seem overkill, it is in fact sometimes useful to configure multiple reference clocks of the same type, in which case the unit numbers must be unique.
The server command is used to configure a reference clock, where the address argument in that command is the clock address. The key, version and ttl options are not used for reference clock support. The mode option is added for reference clock support, as described below. The prefer option can be useful to persuade the server to cherish a reference clock with somewhat more enthusiasm than other reference clocks or peers. Further information on this option can be found in the “Mitigation Rules and the prefer Keyword” (available as part of the HTML documentation provided in /usr/share/doc/ntp) page. The minpoll and maxpoll options have meaning only for selected clock drivers. See the individual clock driver document pages for additional information.
The fudge command is used to provide additional information for individual clock drivers and normally follows immediately after the server command. The address argument specifies the clock address. The refid and stratum options can be used to override the defaults for the device. There are two optional device-dependent time offsets and four flags that can be included in the fudge command as well.
The stratum number of a reference clock is by default zero. Since the ntpd(8) daemon adds one to the stratum of each peer, a primary server ordinarily displays an external stratum of one. In order to provide engineered backups, it is often useful to specify the reference clock stratum as greater than zero. The stratum option is used for this purpose. Also, in cases involving both a reference clock and a pulse-per-second (PPS) discipline signal, it is useful to specify the reference clock identifier as other than the default, depending on the driver. The refid option is used for this purpose. Except where noted, these options apply to all clock drivers.
127.127.
t.u [prefer] [mode int] [minpoll int] [maxpoll int]127.127.
t.u [time1 sec] [time2 sec] [stratum int] [refid string] [mode int] [flag1 0 | 1] [flag2 0 | 1] [flag3 0 | 1] [flag4 0 | 1]The file format consists of a single line containing a single floating point number, which records the frequency offset measured in parts-per-million (PPM). The file is updated by first writing the current drift value into a temporary file and then renaming this file to replace the old version. This implies that ntpd(8) must have write permission for the directory the drift file is located in, and that file system links, symbolic or otherwise, should be avoided.
=
’, ‘+
’ and ‘-
’, where ‘=
’ sets the syslog(3) priority mask, ‘+
’ adds and ‘-
’ removes messages. syslog(3) messages can be controlled in four classes (clock, peer, sys and sync). Within these classes four types of messages can be controlled: informational messages (info), event messages (events), statistics messages (statistics) and status messages (status).Configuration keywords are formed by concatenating the message class with the event class. The all prefix can be used instead of a message class. A message class may also be followed by the all keyword to enable/disable all messages of the respective message class.Thus, a minimal log configuration could look like this:
logconfig =syncstatus +sysevents
This would just list the synchronizations state of ntpd(8) and the major system events. For a simple reference server, the following minimum message configuration could be useful:
logconfig =syncall +clockall
This configuration will list all clock information and synchronization information. All other events and messages about peers, system events and so on is suppressed.
The variables operate as follows:
The trap receiver will generally log event messages and other information from the server in a log file. While such monitor programs may also request their own trap dynamically, configuring a trap receiver will ensure that no messages are lost when the server is started.
In addition to the manual pages provided, comprehensive documentation is available on the world wide web at http://www.ntp.org/
. A snapshot of this documentation is available in HTML format in /usr/share/doc/ntp.
David L. Mills, Network Time Protocol (Version 4), RFC5905.
The ntpkey_host files are really digital certificates. These should be obtained via secure directory services when they become universally available.
Please send bug reports to: http://bugs.ntp.org, bugs@ntp.org
This manual page was AutoGen-erated from the ntp.conf option definitions.
December 24 2013 | SunOS 5.10 |