Precedence of DLPI Primitives

Overview

For each precedence table, the rows (labeled "PRIM X") correspond to primitives that are on the given queue and the columns (labeled "PRIM Y") correspond to primitives that are about to be placed on that queue. Each pair of primitives (PRIM X, PRIM Y) may be manipulated resulting in:

• Change of order, where the order of a pair of primitives is reversed if, and only if, the second primitive in the pair (PRIM Y) is of a type defined to be able to advance ahead of the first primitive in the pair (PRIM X).

• Deletion, where a primitive (PRIM X) may be deleted if, and only if, the primitive that follows it (PRIM Y) is defined to be destructive with respect to that primitive. Destructive primitives may always be added to the queue. Some primitives may cause both primitives in the pair to be destroyed.

The precedence rules define the allowed manipulations of a pair of DLPI primitives. Whether these actions are performed is the choice of the DLS provider for user-originated primitives and the choice of the DLS user for provider-originated primitives.

Write Queue Precedence

For connection establishment primitives, this table represents the possible pairs of DLPI primitives when connect indications/responses are single-threaded. For the multi-threading scenario, the following rules apply:

• ADL_CONNECT_RES primitive has no precedence over either a DL_CONNECT_RES or a DL_DISCONNECT_REQ primitive that is associated with another connection correlation number (dl_correlation), and should therefore be placed on the queue behind such primitives.

• Similarly, a DL_DISCONNECT_REQ primitive has no precedence over either a DL_CONNECT_RES or a DL_DISCONNECT_REQ primitive that is associated with another connection correlation number, and should therefore be placed on the queue behind such primitives. Notice, however, that a DL_DISCONNECT_REQ does have precedence over a DL_CONNECT_RES primitive that is associated with the same correlation number (this is indicated in the table below).

Table: Write Queue Precedence

Read Queue Precedence

For connection establishment primitives, this table represents the possible pairs of DLPI primitives when connect indications/responses are single-threaded. For the multi-threading scenario, the following rules apply:

1. A DL_CONNECT_IND primitive has no precedence over either a DL_CONNECT_IND or a DL_DISCONNECT_IND primitive that is associated with another connection correlation number (dl_correlation), and should therefore be placed on the queue behind such primitives.

2. Similarly, a DL_DISCONNECT_IND primitive has no precedence over either a DL_CONNECT_IND or a DL_DISCONNECT_IND primitive that is associated with another connection correlation number, and should therefore be placed on the queue behind such primitives.

3. A DL_DISCONNECT_IND does have precedence over a DL_CONNECT_IND primitive that is associated with the same correlation number (this is indicated in the table below). If a DL_DISCONNECT_IND is about to be placed on the DLS user's read queue, the user should scan the read queue for a possible DL_CONNECT_IND primitive with a matching correlation number. If a match is found, both the DL_DISCONNECT_IND and matching DL_CONNECT_IND should be removed.

If the DLS user is a user-level process, it's read queue is the stream head read queue. Because a user process has no control over the placement of DLS primitives on the stream head read queue, a DLS user cannot straightforwardly initiate the actions specified in the following precedence table. Except for the connection establishment scenario, the DLS user can ignore the precedence rules defined in the table below. This is equivalent to saying the DLS user's read queue contains at most one primitive.

The only exception to this rule is the processing of connect indication/response primitives. A problem arises if a user issues a DL_CONNECT_RES primitive when a DL_DISCONNECT_IND is on the stream head read queue. The DLS provider will not be expecting the connect response because it has forwarded the disconnect indication to the DLS user and is in the DL_IDLE state. It will therefore generate an error upon seeing the DL_CONNECT_RES.To avoid this error, the DLS user should not respond to a DL_CONNECT_IND primitive if the stream head read queue is not empty. The assumption here is a non-empty queue may be holding a disconnect indication that is associated with the connect indication that is being processed.

When connect indications/responses are single-threaded, a non-empty read queue can only contain a DL_DISCONNECT_IND, which must be associated with the outstanding DL_CONNECT_IND. This DL_DISCONNECT_IND primitive indicates to the DLS user that the DL_CONNECT_IND is to be removed. The DLS user should not issue a response to the DL_CONNECT_IND if a DL_DISCONNECT_IND is received.

The multi-threaded scenario is slightly more complex, because multiple DL_CONNECT_IND and DL_DISCONNECT_IND primitives may be interspersed on the stream head read queue. In this scenario, the DLS user should retrieve all indications on the queue before responding to a given connect indication.

If a queued primitive is a DL_CONNECT_IND, it should be stored by the user process for eventual response. If a queued primitive is a DL_DISCONNECT_IND, it should be matched (using the correlation number) against any stored connect indications. The matched connect indication should then be removed, just as is done in the single-threaded scenario.

Table: Read Queue Precedence