WASD Configuration

5.Conditional Configuration

Request processing (WASD_CONFIG_MAP) and authorization (WASD_CONFIG_AUTH) rules may be conditionally applied depending on request, server or other charactersistics. These include

• server host name, port
• client IP address and host name
• browser-accepted content-types, character sets, languages, encodings
• browser identification string
• scheme ("http:" or "https:", i.e. is it a secure request?)
• HTTP method (GET, POST, etc.)
• request path, query string, cookie data, refering page
• virtual host:port specified in request header
• system information (hardware, Alpha/IA64/X86, node name, VMS version, etc.)
• local time
• random number generation

5.1Service Conditionals

As described in 2.3.1 [[virtual-server]] a [[host:port]] rule applies subsequent configuration depending on whether the request service matches the specified service. This makes it a fundamental element of conditional configuration.

Note that service conditionals impose a boundary on the scope of if..endif constructs. That is, an if..endif may not span a virtual service conditional. A conditional flow syntax error is reported if an if..endif construct is not properly closed before encountering a subsequent [[host:port]] rule.

5.2If..endif Conditionals

These may be nested up to a maximum depth of eight, are not case sensitive and generally match via string comparison, although some tests are performed as boolean operations, by converting the conditional parameter to a number before comparison, and IP address parameters will accept a network mask as well as a string pattern.

String Matching

The basis of much conditional decision making is string pattern matching. Both wildcard and regular expression based pattern matching is available (4. String Matching). Wildcard matching in conditional tests is greedy. Regular expression matching, in common with usage throughout WASD, is differentiated from wildcard patterns using a leading "^" character.

Conditional Syntax

Conditional expressions and processing flow structures may be used in the following formats. Conditional and rule text may be indented for clarifying structure.

Logical operators are also supported, in conjunction with precedence ordering parentheses, allowing moderately complex compound expressions to be applied in conditionals.

There are two more conditional structures that allow previous decisions to be reused. These are unif and the ifif. The first unconditionally includes rules regardless of the current state of execution. The second resumes execution only if the previous if or elif expression was true. The else statement may also be used after an unif to continue only if the previous expression was false. The purpose of these constructs are to allow a single decision statement to include both conditional and unconditional rules.

5.3Conditional Keywords

The following keywords provide a match between the corresponding request or other value and a string immediately following the delimiting colon. White space or other reserved characters may not be included unless preceded by a backslash. The actual value being used in the conditional matching may be observed using the mapping item of the WATCH facility.

Conditional Keywords

5.3.1Notepad: Keyword

The request notepad is a string storage area that can be used to store and retrieve ad hoc information during path mapping and subsequent authorization processing. The notepad contents can be changed using the SET notepad=<string> or appended to using SET notepad=+<string> (10.5.5 SET Rule). These contents then can be subsequently detected using the notepad: conditional keyword (or the obsolescent 'NO' mapping conditional) and used to control subsequent mapping or authorization processing.

Notepad information persists across internal redirection processing (10.5.2 REDIRECT Rule) and so may be used when the regenerated request is mapped and authorized. To prevent such information from unexpectedly interfering with internally redirected requests a notepad="" can be used to empty the storage area.

The dictionary facility provides similar and arguably superior functionailtiy. See 5.5.4 WATCH Dictionary. In fact notepad is now implemented as a dictionary entry.

5.3.2Rand: Keyword

At the commencement of each pass a new pseudo-random number is generated (and therefore remains constant during that pass). The rand: conditional is intended to allow some sort of distribution to be built into a set of rules, where each pass (request) generates a different one. The random conditional accepts two parameters, a modulas number, which is used to modulas the base number, and a comparison number, which is compared to the modulas result.

Hence the following conditional rules

5.3.3Request: Keyword

Looks through each of the lines of the request header for the specified request field and/or value. This may be used to detect the presence of specific or unknown (to the server) request fields. When detecting a specified just field the name can be provided

Note that all request fields known to the server have a specific associated conditional keyword (i.e. "user-agent:" for the above example). To determine whether any request fields unknown to the server have been supplied use the request: keyword as in the following example.

5.3.4Instance: and Robin: Keywords

Both of these conditionals are designed to allow the redistribution of requests between clustered WASD services. They are WASD-aware and so allow a slightly more tailored distribution than perhaps an IP package round-robin implementation might. Each tests for the current operation of WASD on a particular node (using the DLM) before allowing the selection of that node as a target. This can allow some systems to be shutting down or starting up, or have WASD shutdown for any reason, without requiring any extraordinary procedures to allow for the change in processing environment.

Instance:

The instance: directive allows testing for a particular cluster member having a WASD instance currently running. This can allow requests to be redirected or reverse-proxied to a particular system with the knowlege that it should be processed (of course there is a small window of uncertainty as events such as system shutdown and startup occur asynchronously). The behaviour of the conditional block is entirely determinate based on which node names have a WASD instance and the order of evaluation. Compare this to a similar construct using the robin: directive, as described below.

This conditional is deployed in two phases. In the first, it contains a comma-separated list of node names (that are expected to have instances of WASD instantiated). In the second, containing a single node name, allowing the selected node to be tested. For example.

If none of the node names specified in the first phase is currently running a WASD instance the rule returns false, otherwise true. If true the above example has conditional block processed with each of the node names successively tested. If NODE1 has a WASD instance executing it returns true and the associated redirect is performed. The same for NODE2 and NODE3. At least one of these would be expected to test true otherwise the outer conditional established during phase one would have been expected to return false.

Robin:

The robin: conditional allows rules to be applied sequentially against specified members of a cluster that currently have instances of WASD running. This is obviously intended to allow a form of load sharing and/or with redundancy (not balancing, as no evaluation of the selected target's current workload is performed, see below). As with the instance: directive above, there is, of course, a small window of potential uncertainty as events such as system shutdown and startup occur asynchronously and may impact availability between the phase one test and ultimate request distribution.

This conditional is again used in two phases. The first, containing a comma-separated list of node names (that are expected to have instances of WASD instantiated). The second, containing a single node name, allowing the selected node (from phase one) to have a rule applied. For example.

In this case round-robining will be made through four node names. Of course these do not have to represent all the systems in the cluster currently available or having WASD instantiated. The first time the 'robin:' rule containing multiple names is called X861 will be selected. The second time ALPHA1, the third ALPHA2, and the fourth IA64A. With the fifth call X861 is returned to, the sixth ALPHA1, etc. In addition, the selected nodename is verified to have a instance of WASD currently running (using the DLM and WASD's instance awareness). If it does not, round-robining is applied again until one is found (if none is available the phase one conditional returns false). This is most significant as it ensures that the selected node should be able to respond to a redirected or (reverse-)proxied requested. This is the selection set-up phase.

Then there is the selection application phase. Inside the set-up conditional other conditionals apply the selection made in the first phase (through simple nodename string comparison). The rule, in the above example a redirect, is applied if that was the node selected.

During selection set-up unequal weighting can be applied to the round-robin algorithm by including particular node names more than once.

In the above example, the node ALPHA will be selected twice as often as either of X861 and X862 (and because of the ordering interleaved with the X86 selections).

5.3.5Time: Keyword

The time: conditional allows server behaviour to change according to the time of day, week, or even year. It compares the supplied parameter to the current system time in one of three ways.

1. The supplied parameter is in the form "1200-1759", which should be read as "twelve noon to five fifty-nine PM" (i.e. as a time range in minutes, generalized as hhmm-hhmm), where the first is the start time and the second the end time. If the current time is within that range (inclusive) the conditional returns true, otherwise false. If the range doesn't look correct false is always returned.
   if (time:0000-0000) it's midnight elif (time:0001-1159) it's AM elif (time:1200-1200) it's noon else it's PM endif
2. If the supplied parameter is a single digit it is compared to the VMS day of the week (1-Monday, 2-Tuesday … 7-Sunday).
   if (time:6 || time:7) it's the weekend else it's the working week endif
3. If the supplied string is not in either of the formats described above it is treated as a string match with a VMS comparision time (i.e. yyyy-mm-dd hh-mm-ss.hh).
   if (time:%%%%-05-*) it's the month of May endif

5.3.6Trnlnm: Keyword

The trnlnm: conditional dynamically translates a logical name and uses the value. One mandatory and up to two optional parameters may be supplied.

The logical-name must be supplied; without it false is always returned. If just the logical-name is supplied the conditional returns true if the name exists or false if it does not. The default name-table is LNM$FILE_DEV. When the optional name-table is supplied the lookup is confined to that table. If the optional string-to-match is supplied it is matched against the value of the logical and the result returned.

5.3.7Host Addresses

Host names or addresses can be an alpha-numeric string (if DNS lookup is enabled) or dotted-decimal network address, a slash, then a dotted-decimal mask. For example "131.185.250.0/255.255.255.192". This has a 6 bit subnet. It operates by bitwise-ANDing the client host address with the mask, bitwise-ANDing the network address supplied with the mask, then comparing the two results for equality. Using the above example the host 131.185.250.250 would be accepted, but 131.185.250.50 would be rejected. Equivalent notation for this rule would be "131.185.250.0/26".

5.4Examples

The following provides a collection of examples of conditional mapping and authorization rules illustrating the use of wildcard matching, network mask matching and the various formats in which the rules may be blocked.

1. This first example shows an EXEC mapping rule being applied to a path if the request query string contains the string "example".
   if (query-string:*example*) exec /* /cgi-bin/example/*
2. In this example a block of mapping statements is processed if the virtual service of the request matches that in the conditional, otherwise the block is skipped. Note the indentation to help clarify the structure.
   if (service:the.host.name:80) pass /web/* /dka0/the_host_name_web/* pass /graphics/* /dka100/graphics/* pass * "404 Resource not found." endif
3. This example a series of tests allow a form of case processing where the first to match will be processed and terminate the matching process. In this case if a match does not occur rule processing continues after the endif.
   if (service:the.host.name:80) pass /web/* /dka0/the_host_name_web/* elif (service:next.host.name:80) pass /web/* /dka0/next_host_name_web/* elif (service:another.host.name:80) pass /web/* /dka0/another_host_name_web/* endif pass /graphics/* /dka100/graphics/* pass * "404 Resource not found."
4. In this (somewhat contrived) example a nested test is used to check (virtual) server name and that the request is being handled via Secure Sockets Layer (SSL) for security. If it is not an informative message is supplied. The else and the quotes are not really required but included here for illustration.
   if (server-name:the.host.name) if (scheme:"https") pass /secure/* /dka0/the_host_name_web/secure/* else pass * /dka0/the_host_name_web/secure/only-via-SSL.html endif endif
5. This would be another way to accomplish a similar objective to example 4. This uses a negation operator to exclude access to successive mappings if not requesting via SSL.
   if (server-name:the.host.name) if (!SSL:) pass * /web/secure/only-via-SSL.html endif pass /secure/* /web/secure/* pass /other/* /web/other/* pass /web/* /web/web/* pass * "404 Resource not found." endif
6. This example shows the use of a compound conditional using the AND and OR operators. It also illustrates the use of a network mask. It will exclude all access to the specified path unless the request is originating from within a specified network (perhaps an intranet) or via SSL.
   if (path:/sensitive/* && !(remote-addr:131.185.250.0/24 || SSL:)) pass * 404 "Access denied (SSL only)." endif
7. This example illustrates restricting authentication to SSL.
   [[*]] ["Your VMS password"=VMS] if (!request-scheme:https) * r+w,#0 endif
8. Logical name translation may be used to dynamically alter the flow of rule interpretation.
   if (trnlnm:HTTPD_EXAMPLE) pass /* /example/* else pass /* /* endif
9. Using a site administrator's /DO=NOTE= entry to modify rule processing. In this example the contingency of a broken back-end processor has been prepared for and a document advising clients of the temporary problem is redirected to once the administrator enters
   $ HTTPD /DO=NOTE=PROBLEM /ALL
   at the command-line (or via the online equivalent). Note that in this example external clients are provided with the problem advice document while internal clients may still access the back-end for troubleshooting purposes.
   if (note:PROBLEM && !remote-addr:131.185.0.0/16) pass /* /problem_with_backend.html else pass /* /backend/* endif

   Of course there are a multitude of possibilities based on this idea!

5.5Dictionary

The per-request dictionary stores key-value string pairs related to request processing. Some entries are generated and used internally by the server and others may be inserted, value changed, removed and tested by the server admin for conditional processing purposes.

The dictionary was initially introduced as an abstraction layer between the significantly different HTTP/2 and HTTP/1.n header semantics and server internal processing. Its utility was then extended into configuration. It is implemented as a standard hash table with collision lists. The small cost in terms of processing is completely offset by its effectiveness.

5.5.1Configuration Entries

Dictionary entries may be configured using the SET dict=key=value mapping rule or the DICT key=value meta keyword. These are known as configuration entries. Keys must begin with an alpha-numeric character but otherwise keys and values may contain any printable character, with some needing to be escaped in the text of configuration files. These are some examples of each.

If an existing key is (re-)inserted it overwrites the old value.

An entry can have an empty value.

An entry may be removed from the dictionary by prefixing the key name with an exclamation point.

All configuration entries may be removed by using the exclamation point with an empty key.

5.5.2Other Entries

As mentioned, the server generates and uses dictionary entries during request processing. There are multiple types of entry, generally insulated from each other for good reason. These entries are also available for conditional testing.

Dictionary Entries

The "if (dict:expression)" contruct first checks for a configuration entry, then for an request header field entry, then finally for an internal entry (response entries are only available for testing after response processing begins and so not in the search list). It is also possible to test for a key of a specific type by prefixing the key name with the type character. This example shows a request header field being conditionally processed.

It is also possible to set an entry of a specific type by prefixing the key with the type character. For example the following will set a response header field that will be included in the header when returned to the client.

Setting any non-configuration entry should only be undertaken by the literati or the brave.

5.5.3Entry Substitution

The value of a dictionary entry can be derived in whole or part from the value of another entry or entries. This uses a somewhat familiar substitution syntax. A contrived example shows an entry being set that transfers back the request user-agent header field as a response header field.

5.5.4WATCH Dictionary

The content of a request's dictionary at significant stages of request processing can be viewed using the [x]Internal item of a WATCH report. See WATCH Facility of WASD Features and Facilities.

A request dictionary WATCH point is similar to the following (end of request processing) example. Note that all of the entry types described above are present in the example, including two configured entries. Note also that two of the internal entries contain embedded line-breaks and empty lines. This is an HTTP/2 request and the expanded (HTTP/1.n style) request_header and response_header entries are due to WATCH items Request [x]Header and Response [x]Header also being checked. They were not required for request processing.

The first three digit number is simply the entry count in order of insertion. The second, either square bracketed or period delimited, is the hash table entry. The square brackets indicate the head of the hash table, the periods down the collision list. The single punctuation character is use to indicate and differentiate the entry type. Then are the key and equate-separated value. The brace enclosed numbers are the length of the key and value respectively.