Part II Kernel Interface

Chapter 7 STREAMS Framework – Kernel Level

Because the STREAMS subsystem of UNIX® provides a framework on which communications services can be built, it is often called the STREAMS framework. This framework consists of the stream head and a series of utilities (put, putnext), kernel structures (mblk, dblk), and linkages (queues) that facilitate the interconnections between modules, drivers, and basic system calls. This chapter describes the STREAMS components from the kernel developer's perspective.

• Overview of Streams in Kernel Space

• Stream Head

• Kernel–Level Messages

• Message Queues

• Entry Points

• Flow Control in Service Procedures

Overview of Streams in Kernel Space

Chapter 1, Overview of STREAMS describes a stream as a full-duplex processing and data transfer path between a STREAMS driver in kernel space and a process in user space. In the kernel, a stream consists of a stream head, a driver, and zero or more modules between the driver and the stream head.

The stream head is the end of the stream nearest the user process. All system calls made by user-level applications on a stream are processed by the stream head.

Messages are the containers in which data and control information is passed between the stream head, modules, and drivers. The stream head is responsible for translating the appropriate messages between the user application and the kernel. Messages are simply pointers to structures (mblk, dblk) that describe the data contained in them. Messages are categorized by type according to the purpose and priority of the message.

Queues are the basic elements by which the stream head, modules, and drivers are connected. Queues identify the open, close, put, and service entry points. Additionally, queues specify parameters and private data for use by modules and drivers, and are the repository for messages destined for deferred processing.

In the rest of this chapter, the word “modules” refers to modules, drivers, or multiplexers, except where noted.

Stream Head

The stream head interacts between applications in the user space and the rest of the stream in kernel space. The stream head is responsible for configuring the plumbing of the stream through open, close, push, pop, link, and unlink operations. It also translates user data into messages to be passed down the stream, and translates messages that arrive at the stream head into user data. Any characteristics of the stream that can be modified by the user application or the underlying stream are controlled by the stream head, which also informs users of data arrival and events such as error conditions.

If an application makes a system call with a STREAMS file descriptor, the stream head routines are invoked, resulting in data copying, message generation, or control operations. Only the stream head can copy data between the user space and kernel space. All other parts of the stream pass data by way of messages and are thus isolated from direct interaction with users of the stream.

Kernel–Level Messages

Chapter 3, STREAMS Application-Level Mechanisms discusses messages from the application perspective. The following sections discuss message types, message structure and linkage; how messages are sent and received; and message queues and priority from the kernel perspective.

Message Types

Several STREAMS messages differ in their purpose and queueing priority. The message types are briefly described and classified, according to their queueing priority, in Table 7–1 and Table 7–2. A detailed discussion of message types is in Chapter 8, STREAMS Kernel-Level Mechanisms.

Some message types are defined as high-priority types. Ordinary or normal messages can have a normal priority of 0, or a priority (also called a band) from 1 to 255.

Message Structure

A STREAMS message in its simplest form contains three elements—a message block, a data block, and a data buffer. The data buffer is the location in memory where the actual data of the message is stored. The data block (datab(9S) describes the data buffer—where it starts, where it ends, the message types, and how many message blocks reference it. The message block (msgb(9S)) describes the data block and how the data is used.

The data block has a typedef of dblk_t and has the following public elements:

struct datab {
	unsigned char       *db_base;          /* first byte of buffer */
	unsigned char       *db_lim;           /* last byte+1 of buffer */
	unsigned char        db_ref;           /* msg count ptg to this blk */
	unsigned char        db_type;          /* msg type */
};

typedef struct datab dblk_t;

The datab structure specifies the data buffers' fixed limits (db_base and db_lim), a reference count field (db_ref), and the message type field (db_type). db_base points to the address where the data buffer starts, db_lim points one byte beyond where the data buffer ends, and db_ref maintains a count of the number of message blocks sharing the data buffer.

db_base, db_lim, and db_ref should not be modified directly. db_type is modified under carefully monitored conditions, such as changing the message type to reuse the message block.

In a simple message, the message block references the data block, identifying for each message the address where the message data begins and ends. Each simple message block refers to the data block to identify these addresses, which must be within the confines of the buffer such that db_base ≥ b_rptr ≥≥ b_wptr ≥ db_lim. For ordinary messages, a priority band can be indicated, and this band is used if the message is queued.

Figure 7–1 shows the linkages between msgb, datab, and the data buffer in a simple message.

Figure 7–1 Simple Message Referencing the Data Block

The message block (see msgb(9S)) has a typedef of mblk_t and has the following public elements:

struct msgb {
	struct msgb            *b_next;      /*next msg in queue*/
	struct msgb            *b_prev;      /*previous msg in queue*/
	struct msgb            *b_cont;      /*next msg block of message*/
	unsigned char          *b_rptr;      /*1st unread byte in bufr*/
	unsigned char          *b_wptr;      /*1st unwritten byte in bufr*/
	struct datab           *b_datap;     /*data block*/
	unsigned char           b_band;      /*message priority*/
	unsigned short          b_flag;      /*message flags*/
};

The STREAMS framework uses the b_next and b_prev fields to link messages into queues. b_rptr and b_wptr specify the current read and write pointers respectively, in the data buffer pointed to by b_datap. The fields b_rptr and b_wptr are maintained by drivers and modules.

The field b_band specifies a priority band where the message is placed when it is queued using the STREAMS utility routines. This field has no meaning for high-priority messages and is set to zero for these messages. When a message is allocated using allocb(9F), the b_band field is initially set to zero. Modules and drivers can set this field to a value from 0 to 255 depending on the number of priority bands needed. Lower numbers represent lower priority. The kernel incurs overhead in maintaining bands if nonzero numbers are used.

Message block data elements must not modify b_next, b_prev, or b_datap. The first two fields are modified by utility routines such as putq(9F) and getq(9F). Message block data elements can modify b_cont, b_rptr, b_wptr, b_band (for ordinary messages types), and b_flag.

The SunOS environment places b_band in the msgb structure. Some other STREAMS implementations place b_band in the datab structure. The SunOS implementation is more flexible because each message is independent. For shared data blocks, the b_band can differ in the SunOS implementation, but not in other implementations.

Message Linkage

A complex message can consist of several linked message blocks. If buffer size is limited or if processing expands the message, multiple message blocks are formed in the message, as shown in Figure 7–2. When a message is composed of multiple message blocks, the type associated with the first message block determines the overall message type, regardless of the types of the attached message blocks.

Figure 7–2 Linked Message Blocks

Queued Messages

A put procedure processes single messages immediately and can pass the message to the next module's put procedure using put or putnext. Alternatively, the message is linked on the message queue for later processing, to be processed by a module's service procedure (putq(9F)). Note that only the first module of a set of linked modules is linked to the next message in the queue.

Think of linked message blocks as a concatenation of messages. Queued messages are a linked list of individual messages that can also be linked message blocks.

Figure 7–3 Queued Messages

In Figure 7–3 messages are queued: Message 1 being the first message on the queue, followed by Message 2 and Message 3. Notice that Message 1 is a linked message consisting of more than one mblk.

Modules or drivers must not modify b_next and b_prev. These fields are modified by utility routines such as putq(9F) and getq(9F).

Shared Data

In Figure 7–4, two message blocks are shown pointing to one data block. db_ref indicates that there are two references (mblks) to the data block. db_base and db_lim point to an address range in the data buffer. The b_rptr and b_wptr of both message blocks must fall within the assigned range specified by the data block.

Figure 7–4 Shared Data Block

Data blocks are shared using utility routines (see dupmsg(9F) or dupb(9F)). STREAMS maintains a count of the message blocks sharing a data block in the db_ref field.

These two mblks share the same data and datablock. If a module changes the contents of the data or message type, it is visible to the owner of the message block.

When modifying data contained in the dblk or data buffer, if the reference count of the message is greater than one, the module should copy the message using copymsg(9F), free the duplicated message, and then change the appropriate data.

Hardening Information. At Sun, it is assumed that a message with a db_ref > 1 is a "read-only" message and can be read but not modified. If the module wishes to modify the data, it should first copy the message, and free the original:

if ( dbp->db_ref > 1 ) {
	dblk_t	*newdbp;

	/* Get a copy of the message */
	newdbp = copymsg(dbp);

	/* Free the original */
	freemsg(dbp);

	/* make sure that we are now using the new dbp */
	dbp = newdbp;
}

STREAMS provides utility routines and macros (identified in Appendix B, Kernel Utility Interface Summary) to assist in managing messages and message queues, and in other areas of module and driver development. Always use utility routines to operate on a message queue or to free or allocate messages. If messages are manipulated in the queue without using the STREAMS utilities, the message ordering can become confused and cause inconsistent results.

Not adhering to the DDI/DKI can result in panics and system crashes.

Sending and Receiving Messages

Among the message types, the most commonly used messages are M_DATA, M_PROTO, and M_PCPROTO. These messages can be passed between a process and the topmost module in a stream, with the same message boundary alignment maintained between user and kernel space. This allows a user process to function, to some degree, as a module above the stream and maintain a service interface. M_PROTO and M_PCPROTO messages carry service interface information among modules, drivers, and user processes.

Modules and drivers do not interact directly with any interfaces except open(2) and close(2). The stream head translates and passes all messages between user processes and the uppermost STREAMS module. Message transfers between a process and the stream head can occur in different forms. For example, M_DATA and M_PROTO messages can be transferred in their direct form by getmsg(2) and putmsg(2). Alternatively, write(2) creates one or more M_DATA messages from the data buffer supplied in the call. M_DATA messages received at the stream head are consumed by read(2) and copied into the user buffer.

Any module or driver can send any message in either direction on a stream. However, based on their intended use in STREAMS and their treatment by the stream head, certain messages can be categorized as upstream, downstream, or bidirectional. For example, M_DATA, M_PROTO, or M_PCPROTO messages can be sent in both directions. Other message types such as M_IOACK are sent upstream to be processed only by the stream head. Messages to be sent downstream are silently discarded if received by the stream head. Table 7–1 and Table 7–2 indicate the intended direction of message types.

STREAMS enables modules to create messages and pass them to neighboring modules. read(2) and write(2) are not enough to enable a user process to generate and receive all messages. In the first place, read(2) and write(2) are byte-stream oriented with no concept of message boundaries. The message boundary of each service primitive must be preserved so that the start and end of each primitive can be located in order to support service interfaces. Furthermore, read(2) and write(2) offer only one buffer to the user for transmitting and receiving STREAMS messages. If control information and data is placed in a single buffer, the user has to parse the contents of the buffer to separate the data from the control information. Furthermore, read(2) and write(2) offer only one buffer to the user for transmitting and receiving STREAMS messages. If control information and data is placed in a single buffer, the user has to parse the contents of the buffer to separate the data from the control information.

getmsg(2) and putmsg(2) enable a user process and the stream to pass data and control information between one another while maintaining distinct message boundaries.

Data Alignment

Hardening Information. There is no guarantee in STREAMS that a b_rptr or b_wptr will fall on a proper bit alignment. Most modules that pass data structures with pointers try to retain the desired bit alignment. If the module is in a stream where this is reasonably guaranteed, it does not need to check data alignment. However, for the purpose of hardening, modules that are concerned about data alignment should verify that pointers are properly aligned, or copy data in mblks to local structures that are properly aligned (see bcopy(3C)).

Hardening Information. Ensure that the changing of pointers is uniform (b_rptr <= b_rptr). Keep pointers inside db_base and db_lim. It is easier to recover from an error if b_rptr and b_wptr are inside db_base and db_lim.

When a module changes the b_rptr and/or the b_wptr, it should verify the following relationship:

db_base <= b_rptr <= b_wptr <= db_lim

and

db_base < db_lim

Message Queues and Message Priority

Message queues grow when the STREAMS scheduler is delayed from calling a service procedure by system activity, or when the procedure is blocked by flow control. When called by the scheduler, a module's service procedure processes queued messages in a FIFO manner (getq(9F)). However, some messages associated with certain conditions, such as M_ERROR, must reach their stream destination as rapidly as possible. This is accomplished by associating priorities with the messages. These priorities imply a certain ordering of messages in the queue, as shown in Figure 7–5.

Each message has a priority band associated with it. Ordinary messages have a priority band of zero. The priority band of high-priority messages is ignored since they are high priority and thus not affected by flow control. putq(9F) places high-priority messages at the head of the message queue, followed by priority band messages (expedited data) and ordinary messages.

Figure 7–5 Message Ordering in a Queue

When a message is queued, it is placed after the messages of the same priority already in the queue (in other words, FIFO within their order of queueing). This affects the flow-control parameters associated with the band of the same priority. Message priorities range from 0 (normal) to 255 (highest). This provides up to 256 bands of message flow within a stream. An example of how to implement expedited data would be with one extra band of data flow (priority band 1), is shown in the following figure. Queues are explained in detail in the next section.

Figure 7–6 Message Ordering with One Priority Band

High-priority messages are not subject to flow control. When they are queued by putq(9F), the associated queue is always scheduled, even if the queue has been disabled (noenable(9F)). When the service procedure is called by the stream's scheduler, the procedure uses getq(9F) to retrieve the first message on queue, which is a high-priority message. Service procedures must be implemented to act on high-priority messages immediately. The mechanisms just mentioned—priority message queueing, absence of flow control, and immediate processing by a procedure—result in rapid transport of high-priority messages between the originating and destination components in the stream.

In general, high-priority messages should be processed immediately by the module's put procedure and not placed on the service queue.

A service procedure must never queue a high-priority message on its own queue because an infinite loop results. The enqueuing triggers the queue to be immediately scheduled again.

Message Queues

The queue is the fundamental component of a stream. It is the interface between a STREAMS module and the rest of the stream, and is the repository for deferred message processing. For each instance of an open driver or pushed module or stream head, a pair of queues is allocated, one for the read side of the stream and one for the write side.

The RD(9F), WR(9F), and OTHERQ(9F) routines allow reference of one queue from the other. Given a queue, RD(9F) returns a pointer to the read queue, WR(9F) returns a pointer to the write queue and OTHERQ(9F) returns a pointer to the opposite queue of the pair (see QUEUE(9S)).

By convention, queue pairs are depicted graphically as side- by-side blocks, with the write queue on the left and the read queue on the right (see following figure).

Figure 7–7 Queue Pair Allocation

queue() Structure

As previously discussed, messages are ordered in message queues. Message queues, message priority, service procedures, and basic flow control all combine in STREAMS. A service procedure processes the messages in its queue. If there is no service procedure for a queue, putq(9F) does not schedule the queue to be run. The module developer must ensure that the messages in the queue are processed. Message priority and flow control are associated with message queues.

The queue structure is defined in stream.h as a typedef queue_t, and has the following public elements:

struct  qinit   *q_qinfo;	/* procs and limits for queue */
struct  msgb    *q_first;	/* first data block */
struct  msgb    *q_last;	/* last data block */
struct  queue   *q_next;	/* Q of next stream */
struct  queue   *q_link;	/* to next Q for scheduling */
void            *q_ptr;		/* to private data structure */
size_t          q_count;	/* number of bytes on Q */
uint            q_flag;		/* queue state */
ssize_t         q_minpsz;	/* min packet size accepted by this module */
ssize_t         q_maxpsz;	/* max packet size accepted by this module */
size_t          q_hiwat;	/* queue high–water mark */
size_t          q_lowat;	/* queue low–water mark */

q_first points to the first message on the queue, and q_last points to the last message on the queue. q_count is used in flow control and contains the total number of bytes contained in normal and high-priority messages in band 0 of this queue. Each band is flow controlled individually and has its own count. For more details, see qband Structure. qsize(9F) can be used to determine the total number of messages on the queue. q_flag indicates the state of the queue. See Table 7–3 for the definitions of these flags.

q_minpsz contains the minimum packet size accepted by the queue, and q_maxpsz contains the maximum packet size accepted by the queue. These are suggested limits, and some implementations of STREAMS may not enforce them. The SunOS stream head enforces these values but they are voluntary at the module level. You should design modules to handle messages of any size.

q_hiwat indicates the limiting maximum number of bytes that can be put on a queue before flow control occurs. q_lowat indicates the lower limit where STREAMS flow control is released.

q_ptr is the element of the queue structure where modules can put values or pointers to data structures that are private to the module. This data can include any information required by the module for processing messages passed through the module, such as state information, module IDs, routing tables, and so on. Effectively, this element can be used any way the module or driver writer chooses. q_ptr can be accessed or changed directly by the driver, and is typically initialized in the open(9E) routine.

When a queue pair is allocated, streamtab initializes q_qinfo, and module_info initializes q_minpsz, q_maxpsz, q_hiwat, and q_lowat. Copying values from the module_info structure enables them to be changed in the queue without modifying the streamtab and module_info values.

The following table lists the queue(9S) flags.

Using Queue Information

The q_first, q_last, q_count, and q_flags components must not be modified by the module, and should be accessed using strqget(9F). The values of q_minpsz, q_maxpsz, q_hiwat, and q_lowat are accessed through strqget(9F), and are modified by strqset(9F). q_ptr can be accessed and modified by the module and contains data private to the module.

All other accesses to fields in the queue(9S) structure should be made through STREAMS utility routines (see Appendix B, “STREAMS Utilities”). Modules and drivers should not change any fields not explicitly listed previously.

strqget(9F) enables modules and drivers to get information about a queue or particular band of the queue. This insulates the STREAMS data structures from the modules and drivers. The prototype for the strqget(9F) routine is:

int
strqget(queue_t *q, qfields_t what, unsigned char pri, void *valp)

q specifies from which queue the information is to be retrieved; what defines the queue_t field value to obtain (see the following structure fields). pri identifies a specific priority band. The value of the field is returned in valp. The fields that can be obtained are defined in <sys/stream.h> and shown here as:

QHIWAT              /* high–water mark */
QLOWAT              /* low–water mark */
QMAXPSZ             /* largest packet accepted */
QMINPSZ             /* smallest packet accepted */
QCOUNT              /* approx. size (in bytes) of data */
QFIRST              /* first message */
QLAST               /* last message */
QFLAG               /* status */

strqset(9F) enables modules and drivers to change information about a queue or a band of the queue. This also insulates the STREAMS data structures from the modules and drivers. Its prototype is:

int 
strqset(queue_t *q. qfields_t what, unsigned char pri, intptr_t val)

The q, what, and pri fields are the same as in strqget(9F), but the information to be updated is provided in val instead of through a pointer. If the field is read-only, EPERM is returned and the field is left unchanged. The following fields are read-only: QCOUNT, QFIRST, QLAST, and QFLAG.

Hardening Information. Access queue structure information, through strqget() and strqset() only. Do not access the queue structure directly.

Entry Points

The q_qinfo component points to a qinit structure. This structure defines the module's entry point procedures for each queue, which include the following:

int        (*qi_putp)();       /* put procedure */
int        (*qi_srvp)();       /* service procedure */
int        (*qi_qopen)();      /* open procedure */
int        (*qi_qclose)();     /* close procedure */
struct module_info *qi_minfo;  /* module information structure */

There is generally a unique q_init structure associated with the read queue and the write queue. qi_putp identifies the put procedure for the module. qi_srvp identifies the optional service procedure for the module.

The open and close entry points are required for the read-side queue. The put procedure is generally required on both queues and the service procedure is optional.

Hardening Information. If the put procedure is not defined and a subsequent put is done to the module, a panic occurs. As a precaution, putnext should be declared as the module's put procedure.

If a module only requires a service procedure, putq(9F) can be used as the module's put procedure. If the service procedure is not defined, the module's put procedure must not queue data (putq).

The qi_qopen member of the read-side qinit structure identifies the open(9E) entry point of the module. The qi_qclose member of the read-side qinit structure identifies the close(9E) entry point of the module.

The qi_minfo member points to the module_info(9S) structure.

struct module_info {
		ushort   mi_idnum;          /* module id number */
		char     *mi_idname;        /* module name */
		ssize_t  mi_minpsz;         /* min packet size accepted */
		ssize_t  mi_maxpsz;         /* max packet size accepted */
		size_t   mi_hiwat;          /* hi-water mark */
		size_t   mi_lowat;          /* lo-water mark */
};

mi_idnum is the module's unique identifier defined by the developer and used in strlog(9F). mi_idname is an ASCII string containing the name of the module. mi_minpsz is the initial minimum packet size of the queue. mi_maxpsz is the initial maximum packet size of the queue. mi_hiwat is the initial high–water mark of the queue. mi_lowat is the initial low–water mark of the queue.

open Routine

The open(9E) routine of a device is called once for the initial open of the device, then is called again on subsequent reopens of the stream. Module open routines are called once for the initial push onto the stream and again on subsequent reopens of the stream.

Figure 7–8 Order of a Module's open Procedure

The stream is analogous to a stack. Initially the driver is opened and as modules are pushed onto the stream, their open routines are invoked. Once the stream is built, this order reverses if a reopen of the stream occurs. For example, while building the stream shown in Figure 7–8, device A's open routine is called, followed by B's and C's when they are pushed onto the stream. If the stream is reopened, Module C's open routine is called first, followed by B's, and finally by A's.

Usually the module or driver does not check this, but the issue is raised so that dependencies on the order of open routines are not introduced by the programmer. Note that although an open can happen more than once, close is only called once. See the next section on the close routine for more details. If a file is duplicated (dup(2)) the stream is not reopened.

The prototype for the open entry point is:

int prefix_open(queue_t *q, dev_t *devp, int oflag, int sflag, 
						cred_t *cred_p)

q

Pointer to the read queue of this module.

devp

Pointer to a device number that is always associated with the device at the end of the stream. Modules cannot modify this value, but drivers can, as described in Chapter 9, STREAMS Drivers.

oflag

For devices, oflag can contain the following bit mask values: FEXCL, FNDELAY, FREAD, and FWRITE. See Chapter 9, STREAMS Drivers for more information on drivers.

sflag

When the open is associated with a driver, sflag is set to 0 or CLONEOPEN, see Chapter 9, STREAMS Drivers, Cloning STREAMS Drivers for more details. If the open is associated with a module, sflag contains the value MODOPEN.

cred_p

Pointer to the user credentials structure.

The open routines to devices are serialized . If more than one process attempts to open the device, only one proceeds and the others wait until the first finishes. Interrupts are not blocked during an open. The driver's interrupt routine must continue to handle interrupts when multiple processes are opening the same device. See Chapter 9, STREAMS Drivers for more information.

The open routines for both drivers and modules have user context. For example, they can do blocking operations, but the blocking routine should return in the event of a signal. In other words, q_wait_sig is allowed, but q_wait is not.

If the module or driver is to allocate a controlling terminal, it should send an M_SETOPTS message with SO_ISTTY set to the stream head.

The open routine usually initializes the q_ptr member of the queue. q_ptr is generally initialized to some private data structure that contains various state information private to the module or driver. The module's close routine is responsible for freeing resources allocated by the module including q_ptr. The following example shows a simple open routine.

Example 7–1 A Simple open Routine

/* example of a module open */
int xx_open(queue_t *q, dev_t *devp, int oflag, int sflag, cred_t *crp)
{
	struct xxstr *xx_ptr;

	xx_ptr = kmemzalloc(sizeof(struct xxstr), KM_SLEEP);
	xx_ptr->xx_foo = 1;
	q->q_ptr = WR(q)->q_ptr = xx_ptr;
	qprocson(q);
	return (0);
}

In a multithreaded environment, data can flow up the stream during the open. A module receiving this data before its open routine finishes initialization can panic. To eliminate this problem, modules and drivers are not linked into the stream until qprocson(9F) is called (messages flow around the module). The following figure illustrates this process. See Chapter 12, Multithreaded STREAMS for more information on the multithreaded environment and the use of perimeters.

Figure 7–9 Messages Flowing Around the Module Before qprocson

The module or driver instance is guaranteed to be single-threaded before qprocson(9F) is called, except for interrupts or callbacks that must be handled separately.

Hardening Information. qprocson must be called before calling qbufcall(9F), qtimeout(9F), qwait(9F), or qwait_sig(9F).

Before a module calls qprocson, it must be ready to accept data via the module's put procedure so all data structures must be fully initialized (see put Procedure). The most common method for calling qprocson is to call this function just before returning from a successful open (see Example 7–1).

For a multithreaded environment, verify you are using the correct perimeter before accessing data. See Chapter 12, Multithreaded STREAMS for more information on the multithreaded environment and the use of perimeters.

close Routine

The close routine of devices is called only during the last close of the device. Module close routines are called during the last close or if the module is popped.

The prototype for the close entry point is:

int prefix_close (queue *q, int flag, cred_t * cred_p)

q

is a pointer to the read queue of the module.

flag

is analogous to the oflag parameter of the open entry point. If FNBLOCK or FNDELAY is set, then the module should attempt to avoid blocking during the close.

cred_p

is a pointer to the user credential structure.

Like open, the close entry point has user context and can block. Likewise, the blocking routines should return in the event of a signal. Device drivers must take into consideration that interrupts are not blocked during close. The close routine must cancel all pending and qbufcall callbacks, and process any remaining messages on its service queue.

The open and close procedures are only used on the read side of the queue and can be set to NULL in the write-side qinit structure initialization. Example 7–2 shows an example of a module close routine.

Example 7–2 Example of a Module Close

/* example of a module close */
static int
xx_close(queue_t *, *rq, int flag, cred_t *credp)
{
		struct xxstr   *xxp;
  
   /*
    * Disable xxput() and xxsrv() procedures on this queue.
    */
		qprocsoff(rq);
		xxp = (struct xxstr *) rq->q_ptr;

   /*
    * Cancel any pending timeout.
    * This example assumes that the timeout was issued
    * against the write queue.
    */

		if (xxp->xx_timeoutid != 0) {
			(void) quntimeout(WR(rq), xxp->xx_timeoutid);
			xxp->xx_timeoutid=0;
    }
   /*
    * Cancel any pending bufcalls.
    * This example assumes that the bufcall was issued
    * against the write queue.
    */
		if (xxp->xx_bufcallid !=0) {
			(void) qunbufcall(WR(rq), xxp->xx_bufcallid);
 		xxp->xx_bufcallid = 0;
		}
		rq->q_ptr = WR(rq)->q_ptr = NULL;

   /*
    * Free resources allocated during open
    */
		kmem_free (xxp, sizeof (struct xxstr));
		return (0);
}

qprocsoff does the inverse operation shown in Figure 7–9. This supports the need for cancelling callbacks before a qprocsoff.

qprocsoff is typically called at the begining of the close routine. The module can no longer receive messages from adjoining modules. The queue, however, still has pointers to it's adjoining modules and can putnext. However, as the queue is no longer inserted into the stream, these messages may be out of order from other messages in the stream, so it is best to process these messages before qprocsoff.

qwait is used because a module needs to get some response from another module or driver in the STREAM (i.e. a DLPI disconnect message sent downstream). qwait and qwait_sig must also be called before qprocsoff because once the queue is removed from the stream, there will be no way for the reply message to reach the queue.

put Procedure

The put procedure is the mechanism that other modules use to pass messages into this module. This procedure is called via the putor putnext routines on behalf of other modules. The queue's put procedure is invoked by the preceding module to process a message immediately (see put(9F) and putnext(9F)). Most modules will have a put routine. The common exception is on the read-side of drivers because there will not typically be a module downstream to the driver.

Hardening Information. putnext is used by adjoining modules to ensure that the next module's queue is intact. Use of put cannot guarantee that the queue being called is currently valid and inserted into a stream; you must ensure that the queue is valid when using put.

A driver's put procedure must do one of the following:

• Process and free the message.

• Process and route the message back upstream.

• Queue the message to be processed by the driver's service procedure.

All M_IOCTL type messages must be acknowledged through M_IOACK or rejected through M_IOCNACK. M_IOCTL messages should not be freed. Drivers must free any unrecognized message.

A module's put procedure must do one of the following as shown in Figure 7–10:

• Process and free the message.

• Process the message and pass it to the next module or driver.

• Queue the message to be processed later by the module's service procedure.

Unrecognized messages are passed to the next module or driver. The stream operates more efficiently when messages are processed in the put procedure. Processing a message with the service procedure imposes some latency on the message.

Figure 7–10 Flow of put Procedure

If the next module is flow controlled (see canput(9F) and canputnext(9F)), the put procedure can queue the message for processing by the next service procedure (see putq(9F)). The put routine is always called before the component's corresponding srv(9E) service routine, so always use put for immediate message processing.

Hardening Information. canput and canputnext operate similar to put and putnext; that is the next functions verify the integrity of the next queue. Not using the next functions can cause panics as the queue being referenced might have already been closed.

The preferred naming convention for a put procedure reflects the direction of the message flow. The read put procedure is suffixed by r (rput), and the write procedure by w (wput). For example, the read-side put procedure for module xx is declared as int xxrput (queue_t *q, mblk_t *mp). The write-side put procedure is declared as int xxwput(queue_t *q, mblk_t *mp), where q points to the corresponding read or write queue and mp points to the message to be processed.

Although high-priority messages can be placed on the service queue, processing them immediately in the put procedure is better. (See the stub code in Example 7–3.) Place ordinary or priority-band messages on the service queue (putq(9F)) if:

• The stream has been flow controlled; that is, canput fails.

• There are already messages on the service queue, that is, q_first is not NULL.

• Deferred processing is desired.

If other messages already exist on the queue and the put procedure does not queue new messages (provided they are not high-priority), messages are reordered. If the next module is flow controlled (see canput(9F) and canputnext(9F)), the put procedure can queue the message for processing by the service procedure (see putq(9F)).

Example 7–3 Example of a Module put Procedure

/*example of a module put procedure */
int
xxrput(queue_t *,mblk_t, *mp)
{
   /*
    * If the message is a high-priority message or
    * the next module is not flow controlled and we have not
    * already deferred processing, then:
    */

    if (mp->b_datap->db_type >= QPCTL ||
             (canputnext(q) && q->q_first == NULL)) {
        /*
         * Process message
         */

         .
         .
         .
         putnext(q,mp);
    } else {
         /*
          * put message on service queue
          */
          putq(q,mp);
     }
     return (0);
}

Put procedures must never call putq, putbq, or qenable if the module does not have a service procedure.

Hardening Information. Once a message is passed using a putq, put, putnext, as well as the perimeter function qwriter, it cannot be accessed again because the use of this message has been given to the new routine. If a reference needs to be retained by the module, it should copy it by using copyb, copymsg, dupb, or dupmsg.

A module need not process the message immediately, and can queue it for later processing by the service procedure (see putq(9F)).

The SunOS STREAMS framework is multithreaded. Unsafe (nonmultithreaded) modules are not supported. To make multithreading of modules easier, the SunOS STREAMS framework uses perimeters (see MT STREAMS Perimeters for more information).

Mutex locks must not be held across a call to put(9F), putnext(9F), or qreply(9F).

Because of the asynchronous nature of STREAMS, do not assume that a module's put procedure has been called just because put(9F), putnext(9F), or qreply(9F) has returned.

Queue service Procedure

A queue's service procedure is invoked to process messages on the queue. It removes successive messages from the queue, processes them, and calls the put procedure of the next module in the stream to give the processed message to the next queue.

The service procedure is optional. A module or driver can use a service procedure for the following reasons:

• Streams flow control is implemented by service procedures. If the next component on the stream has been flow controlled, the put procedure can queue the message. (See Flow Control in Service Procedures in Chapter 7, STREAMS Framework – Kernel Level for more on flow control.)

• Resource allocation recovery. If a put or service procedure cannot allocate a resource, such as memory, the message is usually queued to process later.

• A device driver can queue a message and get out of interrupt context.

• To combine multiple messages into larger messages.

The service procedure is invoked by the STREAMS scheduler. A STREAMS service procedure is scheduled to run if:

• The queue is not disabled (noenable(9F)) and the message being queued is either the first message on the queue, or a priority band message.

• The message being queued (putq(9F) or putbq(9F)) is a high-priority message,

• The queue has been back-enabled because flow control has been relieved,

• The queue has been explicitly enabled (qenable(9F)).

A service procedure usually processes all messages on its queue (getq(9F)) or takes appropriate action to ensure it is re-enabled (qenable(9F)) at a later time. Figure 7–11 shows the flow of a service procedure.

High-priority messages (db_type and MSG_HIPRI)must never be placed back on a service queue (putbq(9F)); this can cause an infinite loop.

Put procedures must never call putq, putbq, or qenable if the module does not have a service procedure.

Figure 7–11 Flow of service Procedure

The following example shows the stub code for a module service procedure.

Example 7–4 Module service Procedure

/*example of a module service procedure */
int
xxrsrv(queue_t *q)
{
		mblk_t *mp;
   /*
    * While there are still messages on our service queue
    */
		while ((mp = getq(q) != NULL) {
			/*
       * We check for high priority messages, but
       * none is ever seen since the put procedure
       * never queues them.
       * If the next module is not flow controlled, then
       */
      if (mp->b_datap->db_type >= QPCTL || (canputnext (q)) {
				/*
				 * process message
				 */
				.
				.
				.
				putnext (q, mp);
			} else {
				/*
          * put message back on service queue
          */
				putbq(q,mp);
				break;
			}
		}
		return (0);
}

qband Structure

The queue flow information for each band, other than band 0, is contained in a qband(9S) structure. This structure is not visible to other modules. For accessible information see strqget(9F) and strqset(9F). qband is defined as follows:

typedef struct qband {
		struct  qband    *qb_next;   /* next band's info */
		size_t  qb_count;            /* number of bytes in band */
		struct  msgb     *qb_first;  /* beginning of band's data */
		struct  msgb     *qb_last;   /* end of band's data */
		size_t  qb_hiwat;            /* high-water mark for band */
		size_t  qb_lowat;            /* low–water mark for band */
		uint    qb_flag;             /* see below */
} qband_t;

The structure contains pointers to the linked list of messages in the queue. These pointers, qb_first and qb_last, denote the beginning and end of messages for the particular band. The qb_count field is analogous to the queue's q_count field. However, qb_count only applies to the messages in the queue in the band of data flow represented by the corresponding qband structure. In contrast, q_count only contains information regarding normal and high-priority messages.

Each band has a separate high and low watermark, qb_hiwat and qb_lowat. These are initially set to the queue's q_hiwat and q_lowat respectively. Modules and drivers can change these values through the strqset(9F) function. The QB_FULL flag for qb_flag denotes that the particular band is full.

The qband(9S) structures are not preallocated per queue. Rather, they are allocated when a message with a priority greater than zero is placed in the queue using putq(9F), putbq(9F), or insq(9F). Since band allocation can fail, these routines return 0 on failure and 1 on success. Once a qband(9S) structure is allocated, it remains associated with the queue until the queue is freed. strqset(9F) and strqget(9F) cause qband(9S) allocation. Sending a message to a band causes all bands up to and including that one to be created.

Using qband Information

The STREAMS utility routines should be used when manipulating the fields in the queue and qband(9S) structures. strqget(9F) and strqset(9F) are used to access band information.

Drivers and modules can change the qb_hiwat and qb_lowat fields of the qband structure. Drivers and modules can only read the qb_count, qb_first, qb_last, and qb_flag fields of the qband structure. Only the fields listed previously can be referenced.

There are fields in the qband structure that are reserved and are not documented. These fields are subject to undocumented, unnotified change at any time.

The following figure shows a queue with two extra bands of flow.

Figure 7–12 Data Structure Linkage

Several routines are provided to aid you in controlling each priority band of data flow. These routines are

• flushband(9F)

• bcanputnext(9F)

• strqget(9F)

• strqset(9F)

flushband(9F) is discussed in Flushing Priority Band. bcanputnext(9F) is discussed in Flow Control in Service Procedures, and the other two routines are described in the following section. Appendix B, Kernel Utility Interface Summary also has a description of these routines.

Message Processing Procedures

Typically, put procedures are required in pushable modules, but service procedures are optional. If the put routine queues messages, a corresponding service routine must be present to handle the queued messages. If the put routine does not queue messages, the service routine is not required.

Figure 7–10 shows typical processing flow for a put procedure which works as follows:

• A message is received by the put procedure associated with the queue, where some processing can be performed on the message.

• The put procedure determines if the message can be sent to the next module by the use of canput(9F) or canputnext(9F).

• If the next module is flow controlled, the put procedure queues the message using putq(9F).

• putq(9F) places the message in the queue based on its priority.

• Then, putq(9F) makes the queue ready for execution by the STREAMS scheduler, following all other queues currently scheduled.

• If the next module is not flow controlled, the put procedure does any processing needed on the message and sends it to the next module using putnext(9F). Note that if the module does not have a service procedure it cannot queue the message, and must process and send the message to the next module.

Figure 7–11 shows typical processing flow for a service procedure that works as follows:

• When the system goes from kernel mode to user mode, the STREAMS scheduler calls the service procedure.

• The service procedure gets the first message (q_first) from the message queue using the getq(9F) utility.

• The put procedure determines if the message can be sent to the next module using canput(9F) or canputnext(9F).

• If the next module is flow controlled, the put procedure requeues the message with putbq(9F), and then returns.

• If the next module is not flow controlled, the service procedure processes the message and passes it to the put procedure of the next queue with putnext(9F).

• The service procedure gets the next message and processes it. This processing continues until the queue is empty or flow control blocks further processing. The service procedure returns to the caller.

A service or put procedure must never block since it has no user context. It must always return to its caller.

If no processing is required in the put procedure, the procedure does not have to be explicitly declared. Rather, putq(9F) can be placed in the qinit(9S) structure declaration for the appropriate queue side to queue the message for the service procedure. For example:

static struct qinit winit = { putq, modwsrv, ...... };

More typically, put procedures process high-priority messages to avoid queueing them.

Device drivers associated with hardware are examples of STREAMS devices that might not have a put procedure. Since there are no queues below the hardware level, another module does not call the module's put procedure. Data comes into the stream from an interrupt routine, and is either processed or queued for the service procedure.

A STREAMS filter is an example of a module without a service procedure—messages passed to it are either passed or filtered. Flow control is described in Flow Control in Service Procedures.

The key attribute of a service procedure in the STREAMS architecture is delayed processing. When a service procedure is used in a module, the module developer is implying that there are other, more time-sensitive activities to be performed elsewhere in this stream, in other streams, or in the system in general.

The presence of a service procedure is mandatory if the flow control mechanism is to be utilized by the queue. If you do not implement flow control, queues can overflow and hang the system.

Flow Control in Service Procedures

The STREAMS flow control mechanism is voluntary and operates between the two nearest queues in a stream containing service procedures (see Figure 7–13). Messages are held on a queue only if a service procedure is present in the associated queue.

Messages accumulate on a queue when the queue's service procedure processing does not keep pace with the message arrival rate, or when the procedure is blocked from placing its messages on the following STREAMS component by the flow control mechanism. Pushable modules can contain independent upstream and downstream limits. The stream head contains a preset upstream limit (which can be modified by a special message sent from downstream) and a driver can contain a downstream limit. See M_SETOPTS for more information.

Flow control operates as follows:

• Each time a STREAMS message-handling routine (for example, putq(9F)) adds or removes a message from a message queue, the limits are checked. STREAMS calculates the total size of all message blocks (bp->b_wptr - bp->b_rptr) on the message queue.

  ---

  Note –

  bp is a pointer to the buffer header structure allocated by bp_mapin(), b_wptr is the first unwritten byte in the buffer, and b_rptr is the first unread byte in the buffer. See msgb(9S) STREAMS message block structure.

  ---

• The total is compared to the queue high and low watermark values. If the total exceeds the high watermark value, an internal full indicator is set for the queue. The operation of the service procedure in this queue is not affected if the indicator is set, and the service procedure continues to be scheduled.

• The next part of flow control processing occurs in the nearest preceding queue that contains a service procedure. In the following figure, if D is full and C has no service procedure, then B is the nearest preceding queue.

Figure 7–13 Flow Control Mechanism

• The service procedure in B uses canputnext(9F) to check if a queue ahead is marked full. If messages cannot be sent, the scheduler blocks the service procedure in B from further execution. B remains blocked until the low watermark of the full queue, D, is reached.

• While B is blocked, any messages except high-priority messages arriving at B accumulate on its message queue. High-priority messages are not subject to flow control. Eventually, B can reach a full state and the full condition propagates back to the preceding module in the stream.

• When the service procedure processing on D causes the message block total to fall below the low watermark, the full indicator is turned off. STREAMS then schedules the nearest preceding blocked queue (B in this case). This automatic scheduling is called back-enabling a queue.

Modules and drivers need to observe the message priority. High-priority messages, determined by the type of the first block in the message,

mp->b_datap->db_type >= QPCTL

are not subject to flow control. They should be processed immediately and forwarded, as appropriate.

For ordinary messages, flow control must be tested before any processing is performed. canputnext(9F) determines if the forward path from the queue is blocked by flow control.

This is the general flow control processing of ordinary messages:

• Retrieve the message at the head of the queue with getq(9F).

• Determine if the message type is high priority and not to be processed here.

• If so, pass the message to the put procedure of the following queue with putnext(9F).

• Use canputnext(9F) to determine if messages can be sent onward.

• If messages cannot be forwarded, put the message back in the queue with putbq(9F) and return from the procedure.

High-priority messages must be processed and not placed back on the queue.

• Otherwise, process the message.

The canonical representation of this processing within a service procedure is:

while (getq() != NULL)
	if (high priority message || no flow control) {
		process message
		putnext()
	} else {
		putbq()
		return
	}

Expedited data has its own flow control with the same processing method as that of ordinary messages. bcanputnext(9F) provides modules and drivers with a test of flow control in a priority band. It returns 1 if a message of the given priority can be placed in the queue. It returns 0 if the priority band is flow controlled. If the band does not exist in the queue in question, the routine returns 1.

If the band is flow controlled, the higher bands are not affected. However, lower bands are also stopped from sending messages. Without this, lower priority messages can be passed along ahead of the flow-controlled higher priority messages.

The call bcanputnext(q, 0); is equivalent to the call canputnext(q);.

A service procedure must process all messages in its queue unless flow control prevents this.

A service procedure must continue processing messages from its queue until getq(9F) returns NULL. When an ordinary message is queued by putq(9F), the service procedure is scheduled only if the queue was previously empty, and a previous getq(9F) call returns NULL (that is, the QWANTR flag is set). If there are messages in the queue, putq(9F) presumes the service procedure is blocked by flow control and the procedure is automatically rescheduled by STREAMS when the block is removed. If the service procedure cannot complete processing as a result of conditions other than flow control (for example, no buffers), it must ensure a later return (for example, by bufcall(9F)) or discard all messages in the queue. If this is not done, STREAMS never schedules the service procedure to be run unless the queue's put procedure queues a priority message with putq(9F).

High-priority messages are discarded only if there is already a high-priority message on the stream head read queue. That is, there can be only one high-priority message (PC_PROTO) present on the stream head read queue at any time.

putbq(9F) replaces a message at the beginning of the appropriate section of the message queue according to its priority. This might not be the same position at which the message was retrieved by the preceding getq(9F). A subsequent getq(9F) might return a different message.

putq(9F) checks only the priority band in the first message. If a high-priority message is passed to putq with a nonzero b_band value, b_band is reset to 0 before placing the message in the queue. If the message is passed to putq(9F) with a b_band value that is greater than the number of qband(9S)structures associated with the queue, putq(9F) tries to allocate a new qband(9S) structure for each band, up to and including the band of the message.

rmvq and insq work similarly. If you try to insert a message out of order in a queue with insq(9F), the message is not inserted and the routine fails.

putq(9F) does not schedule a queue if noenable(9F) was previously called for the queue. noenable(9F) forces putq(9F) to queue the message when called by this queue, but not to schedule the service procedure. noenable(9F) does not prevent the queue from being scheduled by a flow control back-enable. The inverse of noenable(9F) is enableok(9F).

The service procedure is written using the following algorithm:

while ((bp = getq(q)) != NULL) {
	if (queclass (bp) == QPCTL) {
		/* Process the message */
		putnext(q, bp);
	 } else if (bcanputnext(q, bp->b_band)) {
		/* Process the message */
		putnext(q, bp);
	 } else {
		putbq(q, bp);
		return;
	 }
 }

If the module or driver ignores priority bands, the algorithm is the same as described in the previous paragraphs, except that canputnext(q) is substituted for bcanputnex(q, bp->b_band).

qenable(9F), another flow-control utility, enables a module or driver to cause one of its queues, or another module's queues, to be scheduled. qenable(9F) can also be used to delay message processing. An example of this is a buffer module that gathers messages in its message queue and forwards them as a single, larger message. This module uses noenable(9F) to inhibit its service procedure and queues messages with its put procedure until a certain byte count or “in queue” time has been reached. When either of these conditions is met, the module calls qenable(9F) to cause its service procedure to run.

Another example is a communication line discipline module that implements end-to-end (for example, to a remote system) flow control. Outbound data is held on the write side message queue until the read side receives a transmit window from the remote end of the network.

STREAMS routines are called at different priority levels. Interrupt routines are called at the interrupt priority of the interrupting device. Service routines are called with interrupts enabled (so that service routines for STREAMS drivers can be interrupted by their own interrupt routines).

Chapter 8 STREAMS Kernel-Level Mechanisms

This chapter describes the STREAMS kernel-level mechanisms:

• ioctl Processing

• Message Allocation and Freeing

• Extended STREAMS Buffers

• General ioctl Processing

• M_FLUSH Message Handling

• Driver and Module Service Interfaces

• Common ioctl Interfaces

• signal Message

ioctl Processing

STREAMS is a special type of character device driver that is different from the historical character input/output (I/O) mechanism in several ways.

In the classical device driver, all ioctl(2) calls are processed by the single device driver, which is responsible for their resolution. The classical device driver has user context, that is, all data can be copied directly to and from user space.

By contrast, the stream head itself can process some ioctl(2) calls (defined in streamio(7I)). Generally, STREAMS ioctl(2) calls operate independently of any particular module or driver on the stream. This means the valid ioctl(2) calls that are processed on a stream change over time, as modules are pushed and popped on the stream. The stream modules have no user context and must rely on the stream head to perform copyin and copyout requests.

There is no user context in a module or driver when the information associated with the ioctl(2) call is received. This prevents use of ddi_copyin(9F) or ddi_copyout(9F) by the module. No user context also prevents the module and driver from associating any kernel data with the currently running process. In any case, by the time the module or driver receives the ioctl(2) call, the process generating can have exited.

STREAMS enables user processes to control functions on specific modules and drivers in a stream using ioctl(2) calls. In fact, many streamio(7I) ioctl(2) commands go no further than the stream head. They are fully processed there and no related messages are sent downstream. For an I_STR ioctl(2) or an unrecognized ioctl(2) command, the stream head creates an M_IOCTL message, which includes the ioctl(2) argument. This is then sent downstream to be processed by the pertinent module or driver. STREAMS ensures that there is only one user-driven M_IOCTL operating on a stream at a time. The M_IOCTL message is the precursor message type carrying ioctl(2) information to modules. Other message types are used to complete the ioctl processing in the stream. Each module has its own set of M_IOCTL messages it must recognize.

Hardening Information. Modules and drivers should never assume that user data is correct. Users might be able to pass offsets that exceed the buffers supplied, or data that might be in kernel space. Values should always be checked against the range of data that is requested or supplied. Otherwise, panics or data corruption may occur.

Message Allocation and Freeing

The allocb(9F) utility routine allocates a message and the space to hold the data for the message. allocb(9F) returns a pointer to a message block containing a data buffer of at least the size requested, providing there is enough memory available. The routine returns NULL on failure. allocb(9F) always returns a message of type M_DATA. The type can then be changed if required. b_rptr and b_wptr are set to db_base (see msgb(9S) and datab(9S)), which is the start of the memory location for the message buffer.

STREAMS often provides buffers that are bit aligned, but there is no guarantee that db_base or db_lim reside on bit-aligned boundaries. If bit or page alignment is required on module-supplied buffers use esballoc For more information about esballoc see Extended STREAMS Buffers.

allocb(9F) can return a buffer larger than the size requested. If allocb(9F) indicates that buffers are not available (allocb(9F) fails), the put or service procedure cannot block to wait for a buffer to become available. Instead, bufcall(9F) defers processing in the module or the driver until a buffer becomes available.

If message space allocation is done by the put procedure and allocb(9F) fails, the message is usually discarded. If the allocation fails in the service routine, the message is returned to the queue. bufcall(9F) is called to set a call to the service routine when a message buffer becomes available, and the service routine returns.

freeb(9F) releases the message block descriptor and the corresponding data block, if the reference count (see datab(9S)) is equal to 1. If the reference count exceeds 1, the data block is not released.

freemsg(9F) releases all message blocks in a message. It uses freeb(9F) to free all message blocks and corresponding data blocks.

In Example 8–1, allocb(9F) is used by the bappend subroutine that appends a character to a message block.

Example 8–1 Use of allocb

/*
 * Append a character to a message block.
 * If (*bpp) is null, it will allocate a new block
 * Returns 0 when the message block is full, 1 otherwise
 */
#define MODBLKSZ		128			/* size of message blocks */

static int bappend(mblk_t **bpp, int ch)
{
 	mblk_t *bp;

 	if ((bp = *bpp) != NULL) {
 			if (bp->b_wptr >= bp->b_datap->db_lim)
 				return (0);
 	} else {
 			if ((*bpp = bp = allocb(MODBLKSZ, BPRI_MED)) == NULL)
 				return (1);
 	}
 	*bp->b_wptr++ = ch;
 	return 1;
}

bappend receives a pointer to a message block and a character as arguments. If a message block is supplied (*bpp != NULL), bappend checks if there is room for more data in the block. If not, it fails. If there is no message block, a block of at least MODBLKSZ is allocated through allocb(9F).

If allocb(9F) fails, bappend returns success and discards the character. If the original message block is not full or the allocb(9F) is successful, bappend stores the character in the block.

Example 8–2 shows the processing of all the message blocks in any downstream data (type M_DATA) messages. freemsg(9F) frees messages.

Example 8–2 Subroutine modwput

/* Write side put procedure */
static int modwput(queue_t *q, mblk_t *mp)
{
 	switch (mp->b_datap->db_type) {
 	default:
 			putnext(q, mp);			/* Don't do these, pass along */
 			break;

	case M_DATA: {
 			mblk_t *bp;
			struct mblk_t *nmp = NULL, *nbp = NULL;

			for (bp = mp; bp != NULL; bp = bp->b_cont) {
 				while (bp->b_rptr < bp->b_wptr) {
 						if (*bp->b_rptr == '\n')
 								if (!bappend(&nbp, '\r'))
 									goto newblk;
 						if (!bappend(&nbp, *bp->b_rptr))
 								goto newblk;

						bp->b_rptr++;
 						continue;

				newblk:
 						if (nmp == NULL)
 								nmp = nbp;
 						else { /* link msg blk to tail of nmp */
 								linkb(nmp, nbp);
 								nbp = NULL;
 						}
 				}
 			}
			if (nmp == NULL)
	 			nmp = nbp;
 			else
	 			linkb(nmp, nbp);
	 		freemsg(mp); /* de-allocate message */
 			if (nmp)
 				putnext(q, nmp);
 			break;
 	 	}
 	}
}

Data messages are scanned and filtered. modwput copies the original message into new blocks, modifying as it copies. nbp points to the current new message block. nmp points to the new message being formed as multiple M_DATA message blocks. The outer for loop goes through each message block of the original message. The inner while loop goes through each byte. bappend is used to add characters to the current or new block. If bappend fails, the current new block is full. If nmp is NULL, nmp is pointed at the new block. If nmp is not NULL, the new block is linked to the end of nmp by use of linkb(9F).

At the end of the loops, the final new block is linked to nmp. The original message (all message blocks) is returned to the pool by freemsg(9F). If a new message exists, it is sent downstream.

Recovering From No Buffers

bufcall(9F) can be used to recover from an allocb(9F) failure. The call syntax is as follows:

bufcall_id_t bufcall(int size, int pri, void(*func)(), long arg);

qbufcall(9F) and qunbufcall(9F) must be used with perimeters.

bufcall(9F) calls (*func)(arg) when a buffer of size bytes is available. When func is called, it has no user context and must return without blocking. Also, there is no guarantee that when func is called, a buffer will still be available.

On success, bufcall returns a nonzero identifier that can be used as a parameter to unbufcall(9F) to cancel the request later. On failure, 0 is returned and the requested function is never called.

Care must be taken to avoid deadlock when holding resources while waiting for bufcall to call (*func(arg). bufcall should be used sparingly.

Read Device Interrupt Handler

Example 8–3 is an example of a read device interrupt handler.

Example 8–3 Device Interrupt Handler

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
buffcall_id_t id;						/* hold id val for unbufcall */

dev_rintr(dev)
{
 	/* process incoming message ... */
 	/* allocate new buffer for device */
 	dev_re_load(dev);
}

/*
 * Reload device with a new receive buffer
 */
dev_re_load(dev)
{
 	mblk_t *bp;
 	id = 0;						/* begin with no waiting for buffers */
 	if ((bp = allocb(DEVBLKSZ, BPRI_MED)) == NULL) {
 			cmn_err(CE_WARN,"dev:allocbfailure(size%d)\n",
 				 DEVBLKSZ);
 			/*
 			 * Allocation failed. Use bufcall to
 			 * schedule a call to ourselves.
 			 */
 			id = bufcall(DEVBLKSZ,BPRI_MED,dev_re_load,dev);
 			return;
 	}

 	/* pass buffer to device ... */
}

See Chapter 12, Multithreaded STREAMS for more information on the uses of unbufcall(9F). These references to unbufcall are protected by MT locks.

Because bufcall(9F) can fail, there is still a chance that the device will hang. A better strategy if bufcall(9F) fails is to discard the current input message and resubmit that buffer to the device. Losing input data is preferable to the device hanging.

Write Service Procedure

Example 8–4 is an example of a write service procedure.

Example 8–4 Write Service Procedure

static int mod_wsrv(queue_t *q)
{
 	extern int qenable();
 	mblk_t *mp, *bp;
		while (mp = getq(q)) {
			/* check for priority messages and canput ... */

			/* Allocate a header to prepend to the message.
 		 * If the allocb fails, use bufcall to reschedule.
 		 */
 		if ((bp = allocb(HDRSZ, BPRI_MED)) == NULL) {
 			if (!(id=bufcall(HDRSZ,BPRI_MED,qenable, q))) {
  				timeout(qenable, (caddr_t)q,
					drv_usectohz());
 				/*
 				 * Put the msg back and exit, we will be
 				 * re-enabled later
					 */
 				putbq(q, mp);
 				return;
 			}
 			/* process message .... */
 		}
		}
	}

mod_wsrv prefixes each output message with a header.

In this example, mod_wsrv illustrates a potential deadlock case. If allocb(9F) fails, mod_wsrv tends to recover without loss of data and calls bufcall(9F). In this case, the routine passed to bufcall(9F) is qenable(9F). When a buffer is available, the service procedure is automatically re-enabled. Before exiting, the current message is put back in the queue. Example 8–4 deals with bufcall(9F) failure by calling timeout(9F).

timeout(9F) schedules the given function to be run with the given argument in the given number of clock cycles. In this example, if bufcall(9F) fails, the system runs qenable(9F) after two seconds have passed.

Releasing Callback Requests

When allocb(9F) fails and bufcall(9F) is called, a callback is pending until a buffer is actually returned. Because this callback is asynchronous, it must be released before all processing is complete. To release this queued event, use unbufcall(9F).

Pass the id returned by bufcall(9F) to unbufcall(9F). Then close the driver in the normal way. If this sequence of unbufcall(9F) and xxclose is not followed, the callback can occur when the driver is already closed. This is one of the most difficult types of problems to find and debug.

All bufcall(9F) and timeouts must be canceled in the close routine.

Extended STREAMS Buffers

Some hardware using the STREAMS mechanism supports memory-mapped I/O (see mmap(2)) which allows the sharing of buffers between users, the kernel, and the I/O card. Modules and drivers that need bit-aligned or page-aligned buffers should use extended STREAMS buffers by calling esballoc (also see ddi_umem_alloc).

If the hardware supports memory-mapped I/O, data received from the hardware is placed in the DARAM (dual-access RAM) section of the I/O card. Since DARAM is memory that is shared between the kernel and the I/O card, coordinated data transfer between the kernel and the I/O card is eliminated. Once in kernel space, the data buffer is manipulated as if it were a kernel resident buffer. Similarly, data sent downstream is placed in the DARAM and forwarded to the network.

In a typical network arrangement, data is received from the network by the I/O card. The controller reads the block of data into the card's internal buffer. It interrupts the host computer to notify that data have arrived. The STREAMS driver gives the controller the kernel address where the data block is to go and the number of bytes to transfer. After the controller has read the data into its buffer and verified the checksum, it copies the data into main memory to the address specified by the DMA (direct memory access) memory address. Once in the kernel space, the data is packaged into message blocks and processed in the usual manner.

When data is transmitted from a user process to the network, it is copied from the user space to the kernel space, packaged as a message block, and sent to the downstream driver. The driver interrupts the I/O card, signaling that data is ready to be transmitted to the network. The controller copies the data from the kernel space to the internal buffer on the I/O card, and from there it is placed on the network.

The STREAMS buffer allocation mechanism enables the allocation of message and data blocks to point directly to a client-supplied (non-STREAMS) buffer. Message and data blocks allocated this way are indistinguishable from the normal data blocks. The client-supplied buffers are processed as if they were normal STREAMS data buffers.

Drivers can attach non-STREAMS data buffers and also free them. This is done as follows:

• Allocation - If the drivers use DARAM without using STREAMS resources and without depending on upstream modules or need to use privately allocated buffers, a data and message block can be allocated without an allocated data buffer. Use esballoc(9F). This returns a message block and data block without an associated STREAMS buffer. The buffer used is the one supplied by the caller in the calling sequence.

• Freeing - Each driver using non-STREAMS resources in a STREAMS environment must manage those resources completely, including freeing them. To make this as transparent as possible, a driver-dependent routine is executed if freeb(9F) is called to free a message and data block with an attached non-STREAMS buffer.

freeb(9F) detects when a buffer is a client supplied, non-STREAMS buffer. If it is, freeb(9F) finds the free_rtn(9S) structure associated with the buffer. After calling the driver-dependent routine (defined in free_rtn(9S)) to free the buffer, freeb(9F) frees the message and data block.

The free routine should not reference any dynamically allocated data structures that are freed when the driver is closed, as messages can exist in a stream after the driver is closed. For example, when a stream is closed, the driver close routine is called and its private data structure can be deallocated. If the driver sends a message created by esballoc upstream, that message can still be on the stream head read queue. When the stream head read queue is flushed, the message is freed and a call is made to the driver's free routine after the driver has been closed.

The format of the free_rtn(9S) structure is as follows:

void (*free_func)();   /*driver dependent free routine*/
char *free_arg;        /* argument for free_rtn */

The structure has two fields: a pointer to a function and a location for any argument passed to the function. Instead of defining a specific number of arguments, free_arg is defined as a char *. This way, drivers can pass pointers to structures if more than one argument is needed.

The method by which free_func is called is implementation-specific. Do not assume that free_func is called directly from STREAMS utility routines like freeb(9F). The free_func function must not call another module's put procedure nor try to acquire a private module lock that can be held by another thread across a call to a STREAMS utility routine that could free a message block. Otherwise, lock recursion and deadlock could occur.

esballoc(9F), provides a common interface for allocating and initializing data blocks. It makes the allocation as transparent to the driver as possible and provides a way to modify the fields of the data block, since modification should only be performed by STREAMS. The driver calls this routine to attach its own data buffer to a newly allocated message and data block. If the routine successfully completes the allocation and assigns the buffer, it returns a pointer to the message block. The driver is responsible for supplying the arguments to esballoc(9F), a pointer to its data buffer, the size of the buffer, the priority of the data block, and a pointer to the free_rtn structure. All arguments should be non-NULL. See Appendix B, Kernel Utility Interface Summary, for a description of esballoc(9F).

esballoc(9F) Example

Example 8–5 (which will not compile) shows how extended buffers are managed in the multithreaded environment. The driver maintains a pool of special memory that is allocated by esballoc(9F). The allocator free routine uses the queue struct assigned to the driver or other queue private data, so the allocator and the close routine need to coordinate to ensure that no outstanding esballoc(9F) memory blocks remain after the close. The special memory blocks are of type ebm_t, the counter is ebm, and the mutex mp and the condition variable cvp are used to implement the coordination.

Example 8–5 esballoc Example

ebm_t *
special_new()
{
		mutex_enter(&mp);
		/*
		 * allocate some special memory
		 */
		esballoc();
		/*
		 * increment counter
		 */
		ebm++;
		mutex_exit(&mp);
}

void
special_free()
{
		mutex_enter(&mp);
		/*
		 * de-allocate some special memory
		 */
		freeb();
	
		/*
		 * decrement counter
		 */
		ebm--;
		if (ebm == 0)
			cv_broadcast(&cvp);
		mutex_exit(&mp);
}

open_close(q, .....)
	....
{
		/*
		 * do some stuff
		 */
		/*
		 * Time to decommission the special allocator.  Are there
		 * any outstanding allocations from it?
		 */
		mutex_enter(&mp);
		while (ebm > 0)
			cv_wait(&cvp, &mp);
	
		mutex_exit(&mp);
}

The close routine must wait for all esballoc(9F) memory to be freed.

General ioctl Processing

Please see the ioctl() section in the Writing Device Drivers for information on the 64–bit data structure macros.

When the stream head is called to process an ioctl(2) that it does not recognize, it creates an M_IOCTL message and sends it down the stream. An M_IOCTL message is a single M_IOCTL message block followed by zero or more M_DATA blocks. The M_IOCTL message block has the form of an iocblk(9S) structure. This structure contains the following elements.

int        ioc_cmd;              /* ioctls command type */
cred_t     *ioc_cr;              /* full credentials */
uint       ioc_id;               /* ioctl id */
uint       ioc_count;            /* byte cnt in data field */
int        ioc_error;            /* error code */
int        ioc_rval;             /* return value */

For an I_STR ioctl(2), ioc_cmd contains the command supplied by the user in the ic_cmd member of the strioctl structure defined in streamio(7I). For others, ioc_cmd contains the value of the cmd argument in the call to ioctl(2). The ioc_cr field contains the credentials of the user process.

The ioc_id field is a unique identifier used by the stream head to identify the ioctl and its response messages.

The ioc_count field indicates the number of bytes of data associated with this ioctl request. If the value is greater than zero, there will be one or more M_DATA mblks linked to the M_IOCTL mblkb_cont field. If the value of the ioc_count field is zero, there will be no M_DATA mblk associated with the M_IOCTL mblk. If the value of ioc_count is equal to the special value TRANSPARENT, then there is one M_DATA mblk linked to this mblk and its contents will be the value of the argument passed to ioctl(2). This can be a user address or numeric value. (see Transparent ioctl Processing).

An M_IOCTL message is processed by the first module or driver that recognizes it. If a module does not recognize the command, it should pass it down. If a driver does not recognize the command, it should send a negative acknowledgement or M_IOCNAK message upstream. In all circumstances, a module or driver processing an M_IOCTL message must acknowledge it.

Modules must always pass unrecognized messages on. Drivers should negatively acknowledge unrecognized ioctl(2) messages and free any other unrecognized message.

If a module or driver finds an error in an M_IOCTL message for any reason, it must produce a negative acknowledgement message. To do this, set the message type to M_IOCNAK and send the message upstream. No data or return value can be sent. If ioc_error is set to 0, the stream head causes the ioctl(2) to fail with EINVAL. Optionally, the module can set ioc_error to an alternate error number.

ioc_error can be set to a nonzero value in both M_IOCACK and M_IOCNAK. This causes the value to be returned as an error number to the process that sent the ioctl(2).

If a module checks what the ioctl(2) of other modules below it are doing, the module should not just search for a specific M_IOCTL on the write side, but also look for M_IOCACK or M_IOCNAK on the read side. For example, suppose the module's write side sees TCSETA (see termio(7I)) and records what is being set. The read-side processing knows that the module is waiting for an answer for the ioctl(2). When the read-side processing sees an ack or nak, it checks for the same ioctl(2) by checking the command (here TCSETA) and the ioc_id. If these match, the module can use the information previously saved.

If you have the module check, for example, the TCSETA/TCGETA group of ioctl(2) calls as they pass up or down a stream, you must never assume that because TCSETA comes down it actually has a data buffer attached to it. The user can form TCSETA as an I_STR call and accidentally give a NULL data buffer pointer. Always check b_cont to see if it is NULL before using it as an index to the data block that goes with M_IOCTL messages.

The TCGETA call, if formed as an I_STR call with a data buffer pointer set to a value by the user, always has a data buffer attached to b_cont from the main message block. Do not assume that the data block is missing and allocate a new buffer, then assign b_cont to point to it, because the original buffer will be lost.

STREAMS ioctl Issues

Regular device drivers have user context in the ioctl(9E) call. However, in a STREAMS driver or module, the only guarantee of user context is in the open(9E) and close(9E) routines. Some indication of the calling context where data is used is therefore necessary.

The notion of data models as well as new macros for handling data structure access are discussed in Writing Device Drivers. A STREAMS driver or module writer should use these flags and macros when dealing with structures that change size between data models.

A flag value that represents the data model of the entity invoking the operation has been added to the ioc_flag field of the iocblk(9S) structure, the cq_flag of the copyreq(9S) structure, and the cp_flag of the copyresp(9S) structure.

The data model flag is one of these possibilities:

IOC_ILP32




IOC_LP64

In addition, IOC_NATIVE is conditionally defined to match the data model of the kernel implementation.

By looking at the data model flag field of the relevant iocblk(9S), copyreq(9S), or copyresp(9S) structures, the STREAMS module can determine the best method of handling the data.

The layout of the iocblk, copyreq, and copyresp structures is different between the 32-bit and 64-bit kernels. Be cautious of any data structure overloading in the cp_private, cq_private, or cq_filler fields because alignment has changed.

I_STR ioctl Processing

The transparent and nontransparent methods implement ioctl(2) in the STREAMS driver or module itself, rather than in the stream head. I_STR ioctl(2) (also referred to as nontransparent ioctl(2)) is created when a user requests an I_STR ioctl(2) and specifies a pointer to a strioctl structure as the argument. For example, assuming that fd is an open lp STREAMS device and LP_CRLF is a valid option, the user could make a request by issuing the struct in the following example:

Example 8–6 Struct for Nontransparent ioctl

struct strioctl *str;
short lp_opt = LP_CRLF;

str.ic_cmd = SET_OPTIONS;
str.ic_timout = -1;
str.ic_dp = (char *)&lp_opt;
str.ic_len = sizeof (lp_opt)

ioctl(fd, I_STR, &str);

On receipt of the I_STR ioctl(2) request, the stream head creates an M_IOCTL message. ioc_cmd is set to SET_OPTIONS, ioc_count is set to the value contained in ic_len (in this example sizeof (short)). An M_DATA mblk is linked to the M_IOCTL mblk and the data pointed to by ic_dp is copied into it (in this case LP_CRLF).

Example 8–7 illustrates processing associated with an I_STR ioctl(2). lpdoioctl illustrates driver M_IOCTL processing, which also applies to modules. lpdoioctl is called by lp write-side put or service procedure to process M_IOCTL messages. This example is for a driver.

Example 8–7 I_STR ioctl(2) Driver

static void
lpdoioctl (queue_t *q, mblk_t	 *mp)
{
		struct iocblk *iocp;
		struct lp *lp;

		lp = (struct lp *)q->q_ptr;

		/* 1st block contains iocblk structure */
		iocp = (struct iocblk *)mp->b_rptr;

		switch (iocp->ioc_cmd) {
		  case SET_OPTIONS:
			  /* Count should be exactly one short's worth
			   * (for this example) */
			  if (iocp->ioc_count != sizeof(short))
				  goto iocnak;
			  if (mp->b_cont == NULL)
				  goto lognak; /* not shown in this example */
			  /* Actual data is in 2nd message block */
			  iocp->ioc_error = lpsetopt (lp, *(short *)mp->b_cont->b_rptr)

			  /* ACK the ioctl */
			  mp->b_datap->db_type = M_IOCACK;
			  iocp->ioc_count = 0;
			  qreply(q, mp);
			  break;
		  default:
			  iocnak:
			  /* NAK the ioctl */
			  mp->b_datap->db_type = M_IOCNAK;
			  qreply(q, mp);
		}
	}

This example recogizes only one command, SET_OPTIONS. The ioc_count contains the number of user-supplied data bytes. ioc_count must equal the size of a short.

    switch (iocp->ioc_cmd) {
			case SET_OPTIONS:
				/* Count should be exactly one short's worth
				 * (for this example) */
				if (iocp->ioc_count != sizeof(short))
					goto iocnak;
				if (mp->b_cont == NULL)
					goto lognak; /* not shown in this example */

Once the command has been verified, lpsetopt is called to process the request. lpsetopt returns 0 if the request is satisfied, otherwise an error number is returned. If ioc_error is nonzero, on receipt of the acknowledgement the stream head returns -1 to the application's ioctl(2) request and sets errno to the value of ioc_error.

        /* Actual data is in 2nd message block */
			iocp->ioc_error = lpsetopt (lp, *(short *)mp->b_cont->b_rptr)

The ioctl(2) is acknowledged. This includes changing the M_IOCTL message type to M_IOCACK and setting the ioc_count field to zero to indicate that no data is to be returned to the user. Finally, the message is sent upstream using qreply(9F).

If ioc_count was left nonzero, the stream head would copy that many bytes from the second through the nth message blocks into the user buffer. You must set ioc_count if you want to pass any data back to the user.

           /* ACK the ioctl */
				mp->b_datap->db_type = M_IOCACK;
				iocp->ioc_count = 0;
				qreply(q, mp);
				break;

In the default case for unrecognized commands or malformed requests, a nak is generated. This is done by changing the message type to an M_IOCNAK and sending it back upstream.

          default:
				iocnak:
				/* NAK the ioctl */
				mp->b_datap->db_type = M_IOCNAK;
				qreply(q, mp);

A module does not acknowledge (nak) an unrecognized command, but passes the message on. A module does not acknowledge (nak) a malformed request.

Transparent ioctl

Transparent ioctls are used from within a module to tell the stream head to perform a copyin or copyout on behalf of the module. The stream head must have knowledge of the data model of the caller in order to process the copyin and copyout properly. The user should use the ioctl macros as described in Writing Device Drivers when coding a STREAMS module that uses Transparent ioctls.

Transparent ioctl Messages

The transparent STREAMS ioctl(2) mechanism is needed because user context does not exist in modules and drivers when an ioctl(2) is processed. This prevents them from using the kernel ddi_copyin/ddi_copyout functions.

Transparent ioctl(2) enable you to write an application using conventional ioctl(2) semantics instead of the I_STR ioctl(2) and an strioctl structure. The difference between transparent and nontransparent ioctl(2) processing in a STREAMS driver and module is the way data is transferred from user to kernel space.

The transparent ioctl(2) mechanism allows backward compatibility for older programs. This transparency only works for modules and drivers that support transparent ioctl(2). Trying to use transparent ioctl(2) on a stream that does not support them makes the driver send an error message upstream, causing the ioctl to fail.

The following example illustrates the semantic difference between a nontransparent and transparent ioctl(2). A module that translates arbitrary character is pushed on the stream The ioctl(2) specifies the translation to do (in this case all uppercase vowels are changed to lowercase). A transparent ioctl(2) uses XCASE instead of I_STR to inform the module directly.

Assume that fd points to a STREAMS device and that the conversion module has been pushed onto it. The semantics of a nontransparent I_STR command to inform the module to change the case of AEIOU are:

strioctl.ic_cmd = XCASE;
strioctl.ic_timout = 0;
strioctl.ic_dp = "AEIOU"
strioctl.ic_len = strlen(strioctl.ic_dp);
ioctl(fd,I_STR, &strioctl);

When the stream head receives the I_STR ioctl(2) it creates an M_IOCTL message with the ioc_cmd set to XCASE and the data specified by ic_dp. AEIOU is copied into the first mblk following the M_IOCTL mblk.

The same ioctl(2) specified as a transparent ioctl(2) is called as follows:

ioctl(fd, XCASE, "AEIOU");

The stream head creates an M_IOCTL message with the ioc_cmd set to XCASE, but the data is not copied in. Instead, ioc_count is set to TRANSPARENT and the address of the user data is placed in the first mblk following the M_IOCTL mblk. The module then requests the stream head to copy in the data ("AEIOU") from user space.

Unlike the nontransparent ioctl(2), which can specify a timeout parameter, transparent ioctl(2)s block until processing is complete.

Incorrectly written drivers can cause applications using transparent ioctl(2) to block indefinitely.

Even though this process is simpler in the application, transparent ioctl adds considerable complexity to modules and drivers, and additional overhead to the time required to process the request.

The form of the M_IOCTL message generated by the stream head for a transparent ioctl(2) is a single M_IOCTL message block followed by one M_DATA block. The form of the iocblk(9S) structure in the M_IOCTL block is the same as described under general ioctl(2) processing. However, ioc_cmd is set to the value of the command argument in ioctl(2) and ioc_count is set to the special value of TRANSPARENT. The value TRANSPARENT distinguishes when an I_STR ioctl(2) can specify a value of ioc_cmd that is equivalent to the command argument of a transparent ioctl(2). The b_cont block of the message contains the value of the arg parameter in the call.

If a module processes a specific ioc_cmd and does not validate the ioc_count field of the M_IOCTL message, the module breaks when transparent ioctl(2) is performed with the same command.

Write modules and drivers to support both transparent and I_STR ioctl(2).

All M_IOCTL message types (M_COPYIN, M_COPYOUT, M_IOCDATA,M_IOCACK and M_IOCNACK) have some similar data structures and sizes. You can reuse these structures instead of reallocating them. Note the similarities in the command type, credentials, and id.

The iocblk(9S) structure is contained in M_IOCTL, M_IOCACK and M_IOCNAK message types. For the transparent case, M_IOCTL has one M_DATA message linked to it. This message contains a copy of the argument passed to ioctl(2). Transparent processing of M_IOCACK and M_IONAK does not allow any messages to be linked to them.

The copyreq(9S) structure is contained in M_COPYIN and M_COPYOUT message types. The M_COPYIN message type must not have any other message linked to it (that is, b_cont == NULL). The M_COPYOUT message type must have one or more M_DATA messages linked to it. These messages contain the data to be copied into user space.

The copyresp(9S) structure is contained in M_IOCDATA response message types. These messages are generated by the stream head in response to an M_COPYIN or M_COPYOUT request. If the message is in response to an M_COPYOUT request, the message has no messages attached to it (b_cont is NULL). If the response is to an M_COPYIN, then zero or more M_DATA message types are attached to the M_IOCDATA message. These attached messages contain a copy of the user data requested by the M_COPYIN message.

The iocblk(9S), copyreq(9S), and copyresp(9S) structures contain a field indicating the type of ioctl(2) command, a pointer to the user's credentials, and a unique identifier for this ioctl(2). These fields must be preserved.

The structure member cq_private is reserved for use by the module. M_COPYIN and M_COPYOUT request messages contain a cq_private field that can be set to contain state information for ioctl(2) processing (which identifies what the subsequent M_IOCDATA response message contains). This state is returned in cp_private in the M_IOCDATA message. This state information determines the next step in processing the message. Keeping the state in the message makes the message self-describing and simplifies the ioctl(2) processing.

For each piece of data that the module copies from user space, an M_COPYIN message is sent to the stream head. The M_COPYIN message specifies the user address (cq_addr) and number of bytes (cq_size) to copy from user space. The stream head responds to the M_COPYIN request with a M_IOCDATA message. The b_cont field of the M_IOCDATA mblk contains the contents pointed to by the M_COPYIN request. Likewise, for each piece of data that the module copies to user space, an M_COPYOUT message is sent to the stream head. Specify the user address (cq_addr) and number of bytes to copy (cq_size). The data to be copied is linked to the M_COPYOUT message as one or more M_DATA messages. The stream head responds to M_COPYOUT requests with an M_IOCDATA message, but b_cont is null.

After the module has finished processing the ioctl (that is, all M_COPYIN and M_COPYOUT requests have been processed), the ioctl(2) must be acknowledged with an M_IOCACK to indicate successful completion of the command or an M_IOCNAK to indicate failure.

If an error occurs when attempting to copy data to or from user address space, the stream head will set cp_rval in the M_IOCDATA message to the error number. In the event of such an error, the M_IOCDATA message should be freed by the module or driver. No acknowledgement of the ioctl(2) is sent in this case.

Transparent ioctl Examples

Following are three examples of transparent ioctl(2) processing. Example 8–8 and Example 8–9 illustrate how to use M_COPYIN to copy data from user space. Example 8–10 illustrates how to use M_COPYOUT to copy data to user space. Example 8–11 is a more complex example showing state transitions that combine M_COPYIN and M_COPYOUT.

In these examples the message blocks are reused to avoid the overhead of allocating, copying, and releasing messages. This is standard practice.

The stream head guarantees that the size of the message block containing an iocblk(9S) structure is large enough to also hold the copyreq(9S) and copyresp(9S) structures.

M_COPYIN Example

Please see the copyin section in Writing Device Drivers for information on the 64–bit data structure macros.

Example 8–8 illustrates the processing of a transparent ioctl(2) request only (nontransparent request processing is not shown). In this example, the contents of a user buffer are to be transferred into the kernel as part of an ioctl call of the form

ioctl(fd, SET_ADDR, (caddr_t) &bufadd);

where bufadd is a struct address whose elements are:

struct address {	
     int				ad_len;;		/* buffer length in bytes */
     caddr_t		ad_addr;		/* buffer address */
};

This requires two pairs of messages (request and response) following receipt of the M_IOCTL message: the first copyin(9F), shown in Example 8–8, copies the structure (address) , and the second copyin(9F), shown in Example 8–9, copies the buffer (address.ad.addr). Two states are maintained and processed in this example: GETSTRUCT is for copying the address structure and GETADDR for copying the ad_addr of the structure.

The transparent part of the SET_ADDR M_IOCTL message processing requires that the address structure be copied from user address space. To accomplish this, the M_IOCTL message processing issues an M_COPYIN request to the stream head.

Example 8–8 M_COPYIN: Copy the address Structure

	struct address {		/* same members as in user space */
		int		ad_len;	/* length in bytes */
		caddr_t	ad_addr;	/* buffer address */
	};

	/* state values (overloaded in private field) */
	#define GETSTRUCT		0		/* address structure */
	#define GETADDR		1		/* byte string from ad_addr */

	static void xxioc(queue_t *q, mblk_t *mp);

	static int
	xxwput(q, mp)
		queue_t *q;		/* write queue */
		mblk_t *mp;
	{
		struct iocblk *iocbp;
		struct copyreq *cqp;

		switch (mp->b_datap->db_type) {
			.
			.
			.
			case M_IOCTL:
				/* Process ioctl commands */
				iocbp = (struct iocblk *)mp->b_rptr;
				switch (iocbp->ioc_cmd) {
					case SET_ADDR;
						if (iocbp->ioc_count != TRANSPARENT) {
						  /* do non-transparent processing here
						   *       (not shown here) */
						} else {
						 /* ioctl command is transparent 
						  * Reuse M_IOCTL block for first M_COPYIN 
						  * request of address structure */
						 cqp = (struct copyreq *)mp->b_rptr;
						 /* Get user space structure address from linked 
						  * M_DATA block */
						 cqp->cq_addr = *(caddr_t *) mp->b_cont->b_rptr;
						 cqp->cq_size = sizeof(struct address);
						 /* MUST free linked blks */
						 freemsg(mp->b_cont);
						 mp->b_cont = NULL;

						 /* identify response */
						 cqp->cq_private = (mblk_t *)GETSTRUCT;

						 /* Finish describing M_COPYIN message */
						 cqp->cq_flag = 0;
						 mp->b_datap->db_type = M_COPYIN;
						 mp->b_wptr = mp->b_rptr + sizeof(struct copyreq);
						 qreply(q, mp);
					  break;
					default: /* M_IOCTL not for us */
					  /* if module, pass on */
					  /* if driver, nak ioctl */
					  break;
				} /* switch (iocbp->ioc_cmd) */
				break;
			case M_IOCDATA:
				/* all M_IOCDATA processing done here */
				xxioc(q, mp);
				break;
		}
		return (0);
	}

xxwput() verifies that the SET_ADDR is TRANSPARENT to avoid confusion with an I_STR ioctl(2), which uses a value of ioc_cmd equivalent to the command argument of a transparent ioctl(2).

The if else statement checks whether the size count is equal to TRANSPARENT. If it is equal, the message was not generated from an I_STR ioctl(2) and the else clause of the if else executes.

	if (iocbp->ioc_count != TRANSPARENT) {
			/* do non-transparent processing here (not shown here) */
	} else {

The mblk is reused and mapped into a copyreq(9S) structure. The user space address of bufadd is contained in the b_cont of the M_IOCTL mblk. This address and its size are copied into the copyreq(9S) message. The b_cont of the copy request mblk is not needed, so it is freed and then filled with NULL.

	cqp = (struct copyreq *)mp->b_rptr;
			/* Get user space structure address from linked M_DATA block */
			cqp->cq_addr = *(caddr_t *) mp->b_cont->b_rptr;
			cqp->cq_size = sizeof(struct address);
			/* MUST free linked blks */
			freemsg(mp->b_cont);
			mp->b_cont = NULL;

The layout of the iocblk, copyreq, and copyresp structures is different between 32–bit and 64–bit kernels. Be careful not to overload any data structure in the cp_private or the cq_filler fields because alignment has changed.

Example 8–9 M_COPYIN: Copy the Buffer Address

	xxioc(queue_t *q, mblk_t *mp)			/* M_IOCDATA processing */
	{
		struct iocblk *iocbp;
		struct copyreq *cqp;
		struct copyresp *csp;
		struct address *ap;

		csp = (struct copyresp *)mp->b_rptr;
		iocbp = (struct iocblk *)mp->b_rptr;

		/* validate this M_IOCDATA is for this module */
		switch (csp->cp_cmd) {
			case SET_ADDR:
				if (csp->cp_rval){ /* GETSTRUCT or GETADDR fail */
					freemsg(mp);
					return;
				}
				switch ((int)csp->cp_private){ /* determine state */
					case GETSTRUCT:		/* user structure has arrived */
					  /* reuse M_IOCDATA block */
					  mp->b_datap->db_type = M_COPYIN;
					  mp->b_wptr = mp->b_rptr + sizeof (struct copyreq);
					  cqp = (struct copyreq *)mp->b_rptr;
					  /* user structure */
					  ap = (struct address *)mp->b_cont->b_rptr;
					  /* buffer length */
					  cqp->cq_size = ap->ad_len;
					  /* user space buffer address */
					  cqp->cq_addr = ap->ad_addr;
					  freemsg(mp->b_cont);
					  mp->b_cont = NULL;
					  cqp->cq_flag = 0;
					  cqp->cp_private=(mblk_t *)GETADDR;  /*nxt st*/
					  qreply(q, mp);
					  break;

					case GETADDR:				/* user address is here */
					  /* hypothetical routine */
					  if (xx_set_addr(mp->b_cont) == FAILURE) {
						  mp->b_datap->db_type = M_IOCNAK;
						  iocbp->ioc_error = EIO;
					  } else {
						  mp->b_datap->db_type=M_IOCACK;/*success*/
						  /* can have been overwritten */
						  iocbp->ioc_error = 0;
						  iocbp->ioc_count = 0;
						  iocbp->ioc_rval = 0;
					  }
					  mp->b_wptr=mp->b_rptr + sizeof (struct ioclk);
					  freemsg(mp->b_cont);
					  mp->b_cont = NULL;
					  qreply(q, mp);
					  break;

					default: /* invalid state: can't happen */
					  freemsg(mp->b_cont);
					  mp->b_cont = NULL;
					  mp->b_datap->db_type = M_IOCNAK;
					  mp->b_wptr = mp->rptr + sizeof(struct iocblk);
					  /* can have been overwritten */
					  iocbp->ioc_error = EINVAL;
					  qreply(q, mp);
					  break;
				}
				break;						/* switch (cp_private) */

			default: /* M_IOCDATA not for us */
				/* if module, pass message on */
				/* if driver, free message */
				break;

On receipt of the M_IOCDATA message for the SET_ADDR command, xxioc() checks cp_rval. If an error occurred during the copyin operation, cp_rval is set. The mblk is freed and, if necessary, xxioc() cleans up from previous M_IOCTL requests, freeing memory, resetting state variables, and so on. The stream head returns the appropriate error to the user.

              if (csp->cp_rval){ /* GETSTRUCT or GETADDR fail */
					freemsg(mp);
					return;

If no error occurred during the copyin operation, the switch statement determines whether to process the user structure, GETSTRUCT, or user address, GETADDR.

             switch ((int)csp->cp_private){  /*determine state*/

The cp_private field set to GETSTRUCT indicates that the linked b_cont mblk contains a copy of the user's address structure. The example then copies the actual address specified in address.ad_addr. The program issues another M_COPYIN request to the stream head, but this time cq_private contains GETADDR to indicate that the M_IOCDATA response will contain a copy of address.ad_addr. The stream head copies the information at the requested user address and sends it downstream in another, final M_IOCDATA message.

				  case GETSTRUCT:		/* user structure has arrived */
					  /* reuse M_IOCDATA block */
					  mp->b_datap->db_type = M_COPYIN;
					  mp->b_wptr = mp->b_rptr + sizeof (struct copyreq);
					  cqp = (struct copyreq *)mp->b_rptr;
					  /* user structure */
					  ap = (struct address *)mp->b_cont->b_rptr;
					  /* buffer length */
					  cqp->cq_size = ap->ad_len;
					  /* user space buffer address */
					  cqp->cq_addr = ap->ad_addr;
					  freemsg(mp->b_cont);
					  mp->b_cont = NULL;
					  cqp->cq_flag = 0;
					  cqp->cp_private=(mblk_t *)GETADDR;  /*nxt st*/
					  qreply(q, mp);
					  break;

The final M_IOCDATA message arrives from the stream head. cp_private contains GETADDR. The ad_addr data is contained in the b_cont link of the mblk. If the address is successfully processed by xx_set_addr() (not shown here), the message is acknowledged with an M_IOCACK message. If xx_set_addr() fails, the message is rejected with an M_IOCNAK message]. xx_set_addr() processes the user address from the ioctl(2).

After the final M_IOCDATA message is processed, the module acknowledges the ioctl(2) to let the stream head know that processing is complete. This is done by sending an M_IOCACK message upstream if the request was successfully processed. Always set ioc_error to zero, otherwise an error code could be passed to the user application. Set ioc_rval and ioc_count to zero to reflect that a return value of 0 and no data is to be passed upstream. If the request cannot be processed, either an M_IOCNAK or M_IOCACK can be sent upstream with an appropriate error number. When sending an M_IOCNAK or M_IOCACK, freeing the linked M_DATA block is not mandatory. It is more efficient to use the stream head handle to free the linked M_DATA block.

If ioc_error is set in an M_IOCNAK or M_IOCNACK message, this error code will be returned to the user. If no error code is set in an M_IOCNAK message, EINVAL will be returned to the user.

                case GETADDR:			/* user address is here */
						/* hypothetical routine */
						if (xx_set_addr(mp->b_cont) == FAILURE) {
							mp->b_datap->db_type = M_IOCNAK;
							iocbp->ioc_error = EIO;
						} else {
							mp->b_datap->db_type=M_IOCACK;/*success*/
							/* can have been overwritten */
							iocbp->ioc_error = 0;
							iocbp->ioc_count = 0;
							iocbp->ioc_rval = 0;
						}
						mp->b_wptr=mp->b_rptr + sizeof (struct ioclk);
						freemsg(mp->b_cont);
						mp->b_cont = NULL;
						qreply(q, mp);
						break;

M_COPYOUT Example

Please see the copyout section in Writing Device Drivers for information on the 64–bit data structure macros.

The following code excerpts return option values for the STREAMS device by placing them in the user's options structure. This is done by a transparent ioctl(2) call of the form

struct options optadd;

ioctl(fd, GET_OPTIONS,(caddr_t) &optadd) 

or by a nontransparent I_STR call

	struct strioctl opts_strioctl;
	structure options optadd;

	opts_strioctl.ic_cmd = GET_OPTIONS;
	opts_strioctl.ic_timeout = -1
	opts_strioctl.ic_len = sizeof (struct options);
	opts_strioctl.ic_dp = (char *)&optadd;
	ioctl(fd, I_STR, (caddr_t) &opts_strioctl) 

In the nontransparent I_STR case, opts_strioctl.ic_dp points to the options structure optadd.

Example 8–7 illustrates support of both the I_STR and transparent forms of ioctl(2). The transparent form requires a single M_COPYOUT message following receipt of the M_IOCTL to copy out the contents of the structure. xxwput() is the write-side put procedure of module or driver xx.

Example 8–10 M_COPYOUT

	struct options {			/* same members as in user space */
		int			op_one;
		int			op_two;
		short			op_three;
		long			op_four;
	};

	static int
	xxwput (queue_t *q, mblk_t *mp)
	{
		struct iocblk *iocbp;
		struct copyreq *cqp;
		struct copyresp *csp;
		int transparent = 0;

		switch (mp->b_datap->db_type) {
			.
			.
			.
			case M_IOCTL:
				iocbp = (struct iocblk *)mp->b_rptr;
				switch (iocbp->ioc_cmd) {
					case GET_OPTIONS:
						if (iocbp->ioc_count == TRANSPARENT) {
						  transparent = 1;
						  cqp = (struct copyreq *)mp->b_rptr;
						  cqp->cq_size = sizeof(struct options);
						  /* Get struct address from
								linked M_DATA block */
						  cqp->cq_addr = (caddr_t) 
								*(caddr_t *)mp->b_cont->b_rptr;
						  cqp->cq_flag = 0;
						  /* No state necessary - we will only ever 
							* get one M_IOCDATA from the Stream head 
							* indicating success or failure for 
							* the copyout */
						}
						if (mp->b_cont)
							freemsg(mp->b_cont);
						if ((mp->b_cont = 
									allocb(sizeof(struct options), 
												BPRI_MED)) == NULL) {
						  mp->b_datap->db_type = M_IOCNAK;
						  iocbp->ioc_error = EAGAIN;
						  qreply(q, mp);
						  break;
						}
						/* hypothetical routine */
						xx_get_options(mp->b_cont);
						if (transparent) {
						  mp->b_datap->db_type = M_COPYOUT;
						  mp->b_wptr = mp->b_rptr + sizeof(struct copyreq);
						} else {
						  mp->b_datap->db_type = M_IOCACK;
						  iocbp->ioc_count = sizeof(struct options);
						}
						qreply(q, mp);
						break;

					default: /* M_IOCTL not for us */
						/*if module, pass on;if driver, nak ioctl*/
						break;
				} /* switch (iocbp->ioc_cmd) */
				break;

			case M_IOCDATA:
				csp = (struct copyresp *)mp->b_rptr;
				/* M_IOCDATA not for us */
				if (csp->cmd != GET_OPTIONS) {
					/*if module/pass on, if driver/free message*/
					break;
				}
				if ( csp->cp_rval ) {
					freemsg(mp);	/* failure */
					return (0);
				}
				/* Data successfully copied out, ack */

				/* reuse M_IOCDATA for ack */
				mp->b_datap->db_type = M_IOCACK;
				mp->b_wptr = mp->b_rptr + sizeof(struct iocblk);
				/* can have been overwritten */
				iocbp->ioc_error = 0;
				iocbp->ioc_count = 0;
				iocbp->ioc_rval = 0;
				qreply(q, mp);
				break;
				.
				.
				.
			} /* switch (mp->b_datap->db_type) */
			return (0);

xxwput() first checks whether the ioctl(2) command is transparent. If it is, the message is reused as an M_COPYOUT copy request message. The pointer to the receiving buffer is in the linked message and is copied into cq_addr. Because only a single copy out is being done, no state information needs to be stored in cq_private. The original linked message is freed, in case it isn't big enough to hold the request.

if (iocbp->ioc_count == TRANSPARENT) {
		transparent = 1;
		cqp = (struct copyreq *)mp->b_rptr;
		cqp->cq_size = sizeof(struct options);
		/* Get struct address from linked M_DATA block */
		cqp->cq_addr = (caddr_t) 
									*(caddr_t *)mp->b_cont->b_rptr;
		cqp->cq_flag = 0;
		/* No state necessary - we will only ever get one 
		 * M_IOCDATA from the Stream head indicating 
		 * success or failure for the copyout */
		}
		if (mp->b_cont)
			freemsg(mp->b_cont);

As an optimization, the following code checks the size of the message for reuse:

mp->b_cont->b_datap->db_lim 
- mp->b_cont->b_datap->db_base >= sizeof (struct options)

Hardening Information. After message reuse, make sure to retain the relation:

db_base <= b_rptr <= b_wptr <= db_lim

A new linked message is allocated to hold the option request. When using the transparent ioctl(2) M_COPYOUT command, data contained in the linked message is passed to the stream head. The stream head will copy the data to the user's address space and issue an M_IOCDATA in response to the M_COPYOUT message, which the module must acknowledge in an M_IOCACK message.

            /* hypothetical routine */
				xx_get_options(mp->b_cont);
				if (transparent) {
					mp->b_datap->db_type = M_COPYOUT;
					mp->b_wptr = mp->b_rptr + sizeof(struct copyreq);
				} else {
					mp->b_datap->db_type = M_IOCACK;
					iocbp->ioc_count = sizeof(struct options);
				}

If the message is not transparent (is issued through an I_STR ioctl(2)), the data is sent with the M_IOCACK acknowledgement message and copied into the buffer specified by the strioctl data structure. ioc_error, ioc_count, and ioc_rval are cleared to prevent any stale data from being passed back to the stream head.

           /* reuse M_IOCDATA for ack */
				mp->b_datap->db_type = M_IOCACK;
				mp->b_wptr = mp->b_rptr + sizeof(struct iocblk);
				/* can have been overwritten */
				iocbp->ioc_error = 0;
				iocbp->ioc_count = 0;
				iocbp->ioc_rval = 0;
				qreply(q, mp);
				break;

Bidirectional Data Transfer Example

Example 8–11 illustrates bidirectional data transfer between the kernel and application during transparent ioctl(2) processing. It also shows how to use more complex state information.

The user wants to send and receive data from user buffers as part of a transparent ioctl(2) call of the form:

	ioctl(fd, XX_IOCTL, (caddr_t) &addr_xxdata) 

Example 8–11 Bidirectional Data Transfer

struct xxdata {             /* same members in user space */
   int         x_inlen;     /* number of bytes copied in */
   caddr_t     x_inaddr;    /* buf addr of data copied in */
   int         x_outlen;    /* number of bytes copied out */
   caddr_t     x_outaddr;   /* buf addr of data copied out */
};
/* State information for ioctl processing */
struct state {
		int         st_state;    /* see below */
		struct xxdata		st_data;		/* see above */
};
/* state values */

#define GETSTRUC     0   /* get xxdata structure */
#define GETINDATA    1   /* get data from x_inaddr */
#define PUTOUTDATA   2   /* get response from M_COPYOUT */

static void xxioc(queue_t *q, mblk_t *mp);

static int
xxwput (queue_t *q, 	mblk_t *mp) {
		struct iocblk *iocbp;
		struct copyreq *cqp;
		struct state *stp;
		mblk_t *tmp;

		switch (mp->b_datap->db_type) {
			.
			.
			.
			case M_IOCTL:
				iocbp = (struct iocblk *)mp->b_rptr;
				switch (iocbp->ioc_cmd) {
				case XX_IOCTL:
				/* do non-transparent processing. (See I_STR ioctl
				 * processing discussed in previous section.)
				 */
				/*Reuse M_IOCTL block for M_COPYIN request*/

				cqp = (struct copyreq *)mp->b_rptr;

				/* Get structure's user address from
				 * linked M_DATA block */

				cqp->cq_addr = (caddr_t)
				 *(long *)mp->b_cont->b_rptr;
				freemsg(mp->b_cont);
				mp->b_cont = NULL;

				/* Allocate state buffer */

				if ((tmp = allocb(sizeof(struct state),
				 BPRI_MED)) == NULL) {
						mp->b_datap->db_type = M_IOCNAK;
						iocbp->ioc_error = EAGAIN;
						qreply(q, mp);
						break;
				}
				tmp->b_wptr += sizeof(struct state);
				stp = (struct state *)tmp->b_rptr;
				stp->st_state = GETSTRUCT;
				cqp->cq_private = tmp;

				/* Finish describing M_COPYIN message */

				cqp->cq_size = sizeof(struct xxdata);
				cqp->cq_flag = 0;
				mp->b_datap->db_type = M_COPYIN;
				mp->b_wptr=mp->b_rptr+sizeof(struct copyreq);
				qreply(q, mp);
				break;

			default: /* M_IOCTL not for us */
				/* if module, pass on */
				/* if driver, nak ioctl */
				break;

			} /* switch (iocbp->ioc_cmd) */
			break;

	case M_IOCDATA:
			xxioc(q, mp);   /*all M_IOCDATA processing here*/
			break;
			.
			.
			.
	} /* switch (mp->b_datap->db_type) */
}

Three pairs of messages are required following the M_IOCTL message:

1. case GETSTRUCT copies the structure into the message buffer.

2. case GETINDATA copies the user buffer into the message buffer.

3. case PUTOUTDATA copies the second message buffer into the user buffer.

xxwput() is the write-side put procedure for module or driver xx. xxwput allocates a message block to contain the state structure and reuses the M_IOCTL to create an M_COPYIN message to read in the xxdata structure.

M_IOCDATA processing is done in xxioc() as shown in the following example:

Example 8–12 M_IOCDATA Processing

xxioc(					/* M_IOCDATA processing */
	queue_t *q,
	mblk_t *mp)
{
	struct iocblk *iocbp;
	struct copyreq *cqp;
	struct copyresp *csp;
	struct state *stp;
	mblk_t *xx_indata();

	csp = (struct copyresp *)mp->b_rptr;
	iocbp = (struct iocblk *)mp->b_rptr;
	switch (csp->cp_cmd) {

	case XX_IOCTL:
			if (csp->cp_rval) { /* failure */
				if (csp->cp_private) /* state structure */
						freemsg(csp->cp_private);
				freemsg(mp);
				return;
			 }
			stp = (struct state *)csp->cp_private->b_rptr;
			switch (stp->st_state) {

			case GETSTRUCT:			/* xxdata structure copied in */
										/* save structure */

				stp->st_data =
				 *(struct xxdata *)mp->b_cont->b_rptr;
				freemsg(mp->b_cont);
				mp->b_cont = NULL;
				/* Reuse M_IOCDATA to copyin data */
				mp->b_datap->db_type = M_COPYIN;
				cqp = (struct copyreq *)mp->b_rptr;
				cqp->cq_size = stp->st_data.x_inlen;
				cqp->cq_addr = stp->st_data.x_inaddr;
				cqp->cq_flag = 0;
				stp->st_state = GETINDATA; /* next state */
				qreply(q, mp);
				break;

			case GETINDATA: /* data successfully copied in */
				/* Process input, return output */
				if ((mp->b_cont = xx_indata(mp->b_cont))
				 == NULL) { /* hypothetical */
							/* fail xx_indata */
							mp->b_datap->db_type = M_IOCNAK;
							mp->b_wptr = mp->b_rptr +
								sizeof(struct iocblk);
						iocbp->ioc_error = EIO;
						qreply(q, mp);
						break;
				}
				mp->b_datap->db_type = M_COPYOUT;
				cqp = (struct copyreq *)mp->b_rptr;
				cqp->cq_size = min(msgdsize(mp->b_cont),
				 stp->st_data.x_outlen);
				cqp->cq_addr = stp->st_data.x_outaddr;
				cqp->cq_flag = 0;
				stp->st_state = PUTOUTDATA; /* next state */
				qreply(q, mp);
				break;

			case PUTOUTDATA: /* data copied out, ack ioctl */
				freemsg(csp->cp_private); /*state structure*/
				mp->b_datap->db_type = M_IOCACK;
				mp->b_wtpr = mp->b_rptr + sizeof (struct iocblk);
            /* can have been overwritten */
				iocbp->ioc_error = 0;
				iocbp->ioc_count = 0;
				iocbp->ioc_rval = 0;
				qreply(q, mp);
				break;

			default: /* invalid state: can't happen */
				freemsg(mp->b_cont);
				mp->b_cont = NULL;
				mp->b_datap->db_type = M_IOCNAK;
				mp->b_wptr=mp->b_rptr + sizeof (struct iocblk);
				iocbp->ioc_error = EINVAL;
				qreply(q, mp);
				break;
			} /* switch (stp->st_state) */
			break;
	default: /* M_IOCDATA not for us */
			/* if module, pass message on */
			/* if driver, free message */
			break;
	} /* switch (csp->cp_cmd) */
}

At case GETSTRUCT, the user xxdata structure is copied into the module's state structure (pointed to by cp_private in the message) and the M_IOCDATA message is reused to create a second M_COPYIN message to read the user data.

At case GETINDATA, the input user data is processed by xx_indata (not supplied in the example), which frees the linked M_DATA block and returns the output data message block. The M_IOCDATA message is reused to create an M_COPYOUT message to write the user data.

At case PUTOUTDATA, the message block containing the state structure is freed and an acknowledgement is sent upstream.

Care must be taken at the “can't happen” default case since the message block containing the state structure (cp_private) is not returned to the pool because it might not be valid. This might result in a lost block. The ASSERT helps find errors in the module if a “can't happen” condition occurs.

I_LIST ioctl(2)Example

The I_LIST ioctl(2) lists the drivers and module in a stream.

Example 8–13 List a Stream's Drivers and Modules

#include <stdio.h>
#include <string.h>
#include <stropts.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/socket.h>

main(int argc, const char **argv)
{
		int               s, i;
		int               mods;
		struct str_list   mod_list;
		struct str_mlist *mlist;

		/* Get a socket... */
		if((s = socket(AF_INET, SOCK_STREAM, 0)) <= 0) {
			perror("socket: ");
			exit(1);
		}

		/* Determine the number of modules in the stream. */
		if((mods = ioctl(s, I_LIST, 0)) < 0){
			perror("I_LIST ioctl");
		}
		if(mods == 0) {
			printf("No modules\n");
			exit(1);
		} else {
			printf("%d modules\n", mods);
		}
		/* Allocate memory for all of the module names... */
		mlist = (struct str_mlist *) calloc(mods, sizeof(struct str_mlist));
		if(mlist == 0){
			perror("malloc failure");
			exit(1);
		}
		mod_list.sl_modlist = mlist;
		mod_list.sl_nmods = mods;

		/* Do the ioctl and get the module names. */
		if(ioctl(s, I_LIST, &mod_list) < 0){
			perror("I_LIST ioctl fetch");
			exit(1);
		}

		/* Print out the name of the modules */
		for(i = 0; i < mods; i++) {
			printf("s: %s\n", mod_list.sl_modlist[i].l_name);
		}

		free(mlist);

		exit(0);
}

M_FLUSH Message Handling

All modules and drivers are expected to handle M_FLUSH messages. An M_FLUSH message can originate at the stream head or from a module or a driver. The user can cause data to be flushed from queued messages of a stream by submiting an I_FLUSH ioctl(2). Data can be flushed from the read side, write side, or both sides of a stream.

ioctl(fd,I_FLUSH, arg);

The first byte of the M_FLUSH message is an option flag. The following table describes the possible values for this flag.

Flushing According to Priority Bands

In addition to being able to flush all the data from a queue, a specific band can be flushed using the I_FLUSHBAND ioctl(2).

ioctl(fd, I_FLUSHBAND, bandp); 

The ioctl(2) is passed a pointer to a bandinfo structure. The bi_pri field indicates the band priority to be flushed (from 0 to 255). The bi_flag field indicates the type of flushing to be done. The legal values for bi_flag are defined in Table 8–1. bandinfo has the following format:

struct bandinfo {
		unsigned char       bi_pri;
		in                  bi_flag;
};

See M_FLUSH for details on how modules and drivers should handle flush band requests.

Figure 8–1 and Figure 8–2 further demonstrate flushing the entire stream due to a line break. Figure 8–1 shows the flushing of the write side of a stream, and Figure 8–2 shows the flushing of the read side of a stream.

Figure 8–1 Flushing the Write Side of a Stream

The following discussion describes the sequence of events shown in Figure 8–1 (dotted lines mean flushed queues):

1. A break is detected by a driver.

2. The driver generates an M_BREAK message and sends it upstream.

3. The module translates the M_BREAK into an M_FLUSH message with FLUSHW set, then sends it upstream.

4. The stream head does not flush the write queue (no messages are ever queued there).

5. The stream head turns the message around (sends it down the write side).

6. The module flushes its write queue.

7. The message is passed downstream.

8. The driver flushes its write queue and frees the message.

Figure 8–2 shows flushing the read side of a stream.

Figure 8–2 Flushing the Read Side of a Stream

The following discussion describes the sequence of events.

1. After generating the first M_FLUSH message, the module generates an M_FLUSH with FLUSHR set and sends it downstream.

2. The driver flushes its read queue.

3. The driver turns the message around (sends it up the read side).

4. The module flushes its read queue.

5. The message is passed upstream.

6. The stream head flushes the read queue and frees the message.

The following code shows line discipline module for flush handling.

Example 8–14 Line Discipline Module for Flush Handling

static int
ld_put(
 	queue_t *q,					/* pointer to read/write queue */
 	mblk_t *mp)					/* pointer to message being passed */
{
 	switch (mp->b_datap->db_type) {
 		default:
 			putq(q, mp); /* queue everything */
			return (0);					 /* except flush */

 		case M_FLUSH:
 			if (*mp->b_rptr & FLUSHW)				  /* flush write q */
 					flushq(WR(q), FLUSHDATA);

 			if (*mp->b_rptr & FLUSHR)				  /* flush read q */
 					flushq(RD(q), FLUSHDATA);

 			putnext(q, mp);							  /* pass it on */
 			return(0);
 	}
}

The above example uses FLUSHDATA as the argument to flushq. This code will flush data type messages (M_DATA, M_DELAY, M_PROTO and M_PCPROTO) only. Flushing non-data messages, such as M_IOCTL, M_IOCACK, and M_IOCNAK, might flush messages that contain critical state in the stream. If an M_IOCACK message is flushed, a thread waiting at the stream head for a non-data message response will never receive it. Care should be taken when using flushq.

The stream head turns around the M_FLUSH message if FLUSHW is set (FLUSHR is cleared). A driver turns around M_FLUSH if FLUSHR is set (should mask off FLUSHW).

Flushing Priority Band

The bi_flag field is one of FLUSHR, FLUSHW, or FLUSHRW.

The following example shows flushing according to the priority band.

Example 8–15 Priority Band Data Flush Handling

queue_t *rdq;								/* read queue */
queue_t *wrq;								/* write queue */

	case M_FLUSH:
		if (*bp->b_rptr & FLUSHBAND) {
			if (*bp->b_rptr & FLUSHW)
				flushband(wrq, FLUSHDATA, *(bp->b_rptr + 1));
			if (*bp->b_rptr & FLUSHR)
				flushband(rdq, FLUSHDATA, *(bp->b_rptr + 1));
		} else {
			if (*bp->b_rptr & FLUSHW)
				flushq(wrq, FLUSHDATA);
			if (*bp->b_rptr & FLUSHR)
				flushq(rdq, FLUSHDATA);
		}
		/*
		 * modules pass the message on;
		 * drivers shut off FLUSHW and loop the message
		 * up the read-side if FLUSHR is set; otherwise,
		 * drivers free the message.
		 */
		break;

Note that modules and drivers are not required to treat messages as flowing in separate bands. Modules and drivers can view the queue as having only two bands of flow, normal and high priority. However, the latter alternative flushes the entire queue whenever an M_FLUSH message is received.

The field b_flag of the msgb structure provides a way for the stream head to stop M_FLUSH messages from being reflected forever when the stream is used as a pipe. When the stream head receives an M_FLUSH message, it sets the MSGNOLOOP flag in the b_flag field before reflecting the message down the write side of the stream. If the stream head receives an M_FLUSH message with this flag set, the message is freed rather than reflected.

Figure 8–3 Interfaces Affecting Drivers

The set of STREAMS utilities available to drivers are listed in Appendix B, Kernel Utility Interface Summary. No system-defined macros that manipulate global kernel data or introduce structure-size dependencies are permitted in these utilities. So, some utilities that have been implemented as macros in the prior Solaris operating environment releases are implemented as functions in the SunOS 5 system. This does not preclude the existence of both macro and function versions of these utilities. Driver source code should include a header file that picks up function declarations while the core operating system source should include a header file that defines the macros. With the DKI interface, the following STREAMS utilities are implemented as C programming language functions: datamsg(9F), OTHERQ(9F), putnext(9F), RD(9F), and WR(9F).

Replacing macros such as RD with function equivalents in the driver source code allows driver objects to be insulated from changes in the data structures and their size, increasing the useful lifetime of driver source code and objects. Multithreaded drivers are also protected against changes in implementation-specific STREAMS synchronization.

The DKI defines an interface suitable for drivers and there is no need for drivers to access global kernel data structures directly. The kernel function drv_getparm(9F) fetches information from these structures. This restriction has an important consequence. Because drivers are not permitted to access global kernel data structures directly, changes in the contents/offsets of information within these structures will not break objects.

Driver and Module Service Interfaces

STREAMS provides the means to implement a service interface between any two components in a stream, and between a user process and the topmost module in the stream. A service interface is a set of primitives defined at the boundary between a service user and a service provider (see Figure 8–5). Rules define a service and the allowable state transitions that result as these primitives are passed between the user and the provider. These rules are typically represented by a state machine. In STREAMS, the service user and provider are implemented in a module, driver, or user process. The primitives are carried bidirectionally between a service user and provider in M_PROTO and M_PCPROTO messages.

PROTO messages (M_PROTO and M_PCPROTO) can be multiblock. The second through last blocks are of type M_DATA. The first block in a PROTO message contains the control part of the primitive in a form agreed upon by the user and provider. The block is not intended to carry protocol headers. (Upstream PROTO messages can have multiple PROTO blocks at the start of the message, although its use is not recommended. getmsg(2) compacts the blocks into a single control part when sending to a user process.) The M_DATA block contains any data part associated with the primitive. The data part can be processed in a module that receives it, or it can be sent to the next stream component, along with any data generated by the module. The contents of PROTO messages and their allowable sequences are determined by the service interface specification.

PROTO messages can be sent bidirectionally (upstream and downstream) on a stream and between a stream and a user process. putmsg(2) and getmsg(2) system calls are analogous to write(2) and read(2) except that the former allow both data and control parts to be (separately) passed, and they retain the message boundaries across the user-stream interface. putmsg(2) and getmsg(2) separately copy the control part (M_PROTO or M_PCPROTO block) and data part (M_DATA blocks) between the stream and user process.

An M_PCPROTO message normally is used to acknowledge primitives composed of other messages. M_PCPROTO ensures that the acknowledgement reaches the service user before any other message. If the service user is a user process, the stream head will only store a single M_PCPROTO message, and discard subsequent M_PCPROTO messages until the first one is read with getmsg(2).

Figure 8–4 Protocol Substitution

By defining a service interface through which applications interact with a transport protocol, you can substitute a different protocol below the service interface that is completely transparent to the application. In Figure 8–4, the same application can run over the Transmission Control Protocol (TCP) and the ISO transport protocol. Of course, the service interface must define a set of services common to both protocols.

The three components of any service interface are the service user, the service provider, and the service interface itself, as seen in Figure 8–5.

Figure 8–5 Service Interface

Typically, an application makes requests of a service provider using some well-defined service primitive. Responses and event indications are also passed from the provider to the user using service primitives.

Each service interface primitive is a distinct STREAMS message that has two parts, a control part and a data part. The control part contains information that identifies the primitive and includes all necessary parameters. The data part contains user data associated with that primitive.

An example of a service interface primitive is a transport protocol connect request. This primitive requests the transport protocol service provider to establish a connection with another transport user. The parameters associated with this primitive can include a destination protocol address and specific protocol options to be associated with that connection. Some transport protocols also allow a user to send data with the connect request. A STREAMS message is used to define this primitive. The control part identifies the primitive as a connect request and includes the protocol address and options. The data part contains the associated user data.

Service Interface Library Example

The service interface library example presented here includes four functions that enable a user do the following:

• Establish a stream to the service provider and bind a protocol address to the stream

• Send data to a remote user

• Receive data from a remote user

• Close the stream connected to the provider

First, the structure and constant definitions required by the library are shown in the following code. These typically reside in a header file associated with the service interface.

The defined structures describe the contents of the control part of each service interface message passed between the service user and service provider. The first field of each control part defines the type of primitive being passed.

Example 8–16 Service Interface Library Header File

/*
 * Primitives initiated by the service user.
 */
 #define BIND_REQ                   1   /* bind request */
 #define UNITDATA_REQ               2   /* unitdata request */

/*
 * Primitives initiated by the service provider.
 */
 #define OK_ACK                     3   /* bind acknowledgement */
 #define ERROR_ACK                  4   /* error acknowledgement */
 #define UNITDATA_IND               5   /* unitdata indication */

/*
 * The following structure definitions define the format
 * of the control part of the service interface message
 * of the above primitives.
 */
struct bind_req {                      /* bind request */
 	t_scalar_t    PRIM_type;            /* always BIND_REQ */
 	t_uscalar_t   BIND_addr;            /* addr to bind */
};
 struct unitdata_req {                 /* unitdata request */
 	t_scalar_t    PRIM_type;            /* always UNITDATA_REQ */
 	t_scalar_t    DEST_addr;            /* destination addr */
};

struct ok_ack {                        /* positiv acknowledgement*/
 	t_scalar_t    PRIM_type;            /* always OK_ACK */
};

struct error_ack {                     /* error acknowledgement */
 	t_scalar_t    PRIM_type;            /* always ERROR_ACK */
 	t_scalar_t    UNIX_error;           /* UNIX systemerror code */
};

struct unitdata_ind {                  /* unitdata indication */
 	t_scalar_t    PRIM_type;            /* always UNITDATA_IND */
 	t_scalar_t    SRC_addr;             /* source addr */
};

/* union of all primitives */
union primitives {
 	t_scalar_t             type;
 	struct bind_req        bind_req;
 	struct unitdata_req    unitdata_req;
 	struct ok_ack          ok_ack;
 	struct error_ack       error_ack;
 	struct unitdata_ind    unitdata_ind;
};

/* header files needed by library */
#include <stropts.h>
#include <stdio.h>
#include <errno.h>

Five primitives are defined. The first two represent requests from the service user to the service provider.

BIND_REQ

Asks the provider to bind a specified protocol address. It requires an acknowledgement from the provider to verify that the contents of the request were syntactically correct.

UNITDATA_REQ

Asks the provider to send data to the specified destination address. It does not require an acknowledgement from the provider.

The three other primitives represent acknowledgements of requests, or indications of incoming events, and are passed from the service provider to the service user.

OK_ACK

Informs the user that a previous bind request was received successfully by the service provider.

ERROR_ACK

Informs the user that a nonfatal error was found in the previous bind request. It indicates that no action was taken with the primitive that caused the error.

UNITDATA_IND

Indicates that data destined for the user has arrived.

Module Service Interface Example

The following code is part of a module that illustrates the concept of a service interface. The module implements a simple service interface and mirrors the service interface library example. The following rules pertain to service interfaces.

• Modules and drivers that support a service interface must act upon all PROTO messages and not pass them through.

• Modules can be inserted between a service user and a service provider to manipulate the data part as it passes between them. However, these modules cannot alter the contents of the control part (PROTO block, first message block) nor alter the boundaries of the control or data parts. That is, the message blocks comprising the data part can be changed, but the message cannot be split into separate messages nor combined with other messages.

  In addition, modules and drivers must observe the rule that high-priority messages are not subject to flow control and forward them accordingly.

Service Primitive Declarations

The service interface primitives are defined in the declarations shown in the following example:

Example 8–17 Service Primitive Declarations

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/errno.h>

/* Primitives initiated by the service user */

#define BIND_REQ				1		/* bind request */
#define UNITDATA_REQ			2		/* unitdata request */

 /* Primitives initiated by the service provider */

#define OK_ACK				3		/* bind acknowledgement */
#define ERROR_ACK				4		/* error acknowledgement */
#define UNITDATA_IND			5		/* unitdata indication */
/*
 * The following structures define the format of the
 * stream message block of the above primitives.

 */
struct bind_req {                       /* bind request */
   t_scalar_t    PRIM_type;             /* always BIND_REQ */
   t_uscalar_t   BIND_addr;             /* addr to bind	*/
};
struct unitdata_req {                   /* unitdata request */
   t_scalar_t    PRIM_type;             /* always UNITDATA_REQ */
   t_scalar_t    DEST_addr;             /* dest addr */
};
struct ok_ack {                         /* ok acknowledgement */
   t_scalar_t    PRIM_type;             /* always OK_ACK */
};
struct error_ack {                      /* error acknowledgement */
   t_scalar_t    PRIM_type;             /* always ERROR_ACK */
   t_scalar_t    UNIX_error;            /* UNIX system error code*/
};
struct unitdata_ind {                   /* unitdata indication */
   t_scalar_t    PRIM_type;             /* always UNITDATA_IND */
   t_scalar_t    SRC_addr;              /* source addr */
};

union primitives {							/* union of all primitives */
   long                      type;
   struct bind_req           bind_req;
   struct unitdata_req       unitdata_req;
   struct ok_ack             ok_ack;
   struct error_ack          error_ack;
   struct unitdata_ind       unitdata_ind;
};
struct dgproto {                        /* structure minor device */
   short state;                         /* current provider state */
   long addr;                           /* net address */
};

/* Provider states */
#define IDLE 0
#define BOUND 1

In general, the M_PROTO or M_PCPROTO block is described by a data structure containing the service interface information in this example, union primitives.

The module recognizes two commands:

BIND_REQ

Give this stream a protocol address (for example, give it a name on the network). After a BIND_REQ is completed, data from other senders will find their way through the network to this particular stream.

UNITDATA_REQ

Send data to the specified address.

The module generates three messages:

OK_ACK

A positive acknowledgement (ack) of BIND_REQ.

ERROR_ACK

A negative acknowledgement (nak) of BIND_REQ.

UNITDATA_IND

Data from the network has been received.

The acknowledgement of a BIND_REQ informs the user that the request was syntactically correct (or incorrect if ERROR_ACK). The receipt of a BIND_REQ is acknowledged with an M_PCPROTO to ensure that the acknowledgement reaches the user before any other message. For example, if a UNITDATA_IND comes through before the bind is completed, the application cannot send data to the proper address

The driver uses a per-minor device data structure, dgproto, which contains the following:

state

Current state of the service provider IDLE or BOUND.

addr

Network address that has been bound to this stream.

The module open procedure is assumed to have set the write queue q_ptr to point at the appropriate private data structure, although this is not shown explicitly.

Service Interface Procedure

The write put procedure, protowput(), is shown in the following example:

Example 8–18 Service Interface protoput Procedure

static int protowput(queue_t *q, mblk_t *mp)
{
 	union primitives *proto;
 	struct dgproto *dgproto;
 	int err;
 	dgproto = (struct dgproto *) q->q_ptr;  /* priv data struct */
 	switch (mp->b_datap->db_type) {
 	default:
 			/* don't understand it */
 			mp->b_datap->db_type = M_ERROR;
 			mp->b_rptr = mp->b_wptr = mp->b_datap->db_base;
 			*mp->b_wptr++ = EPROTO;
 			qreply(q, mp);
 			break;
 	case M_FLUSH: /* standard flush handling goes here ... */
 			break;
 	case M_PROTO:
 			/* Protocol message -> user request */
 			proto = (union primitives *) mp->b_rptr;
 			switch (proto->type) {
 			default:
 				mp->b_datap->db_type = M_ERROR;
 				mp->b_rptr = mp->b_wptr = mp->b_datap->db_base;
 				*mp->b_wptr++ = EPROTO;
 				qreply(q, mp);
 				return;
 			case BIND_REQ:
 				if (dgproto->state != IDLE) {
 						err = EINVAL;
 						goto error_ack;
 				}
 				if (mp->b_wptr - mp->b_rptr !=
 				 sizeof(struct bind_req)) {
 						err = EINVAL;
 						goto error_ack;
 				}
 				if (err = chkaddr(proto->bind_req.BIND_addr))
 						goto error_ack;
 				dgproto->state = BOUND;
 				dgproto->addr = proto->bind_req.BIND_addr;
 				mp->b_datap->db_type = M_PCPROTO;
 				proto->type = OK_ACK;
 				mp->b_wptr=mp->b_rptr+sizeof(structok_ack);
 				qreply(q, mp);
 				break;
			error_ack:
 				mp->b_datap->db_type = M_PCPROTO;
 				proto->type = ERROR_ACK;
 				proto->error_ack.UNIX_error = err;
 				mp->b_wptr = mp->b_rptr+sizeof(structerror_ack);
 				qreply(q, mp);
 				break;
 			case UNITDATA_REQ:
 				if (dgproto->state != BOUND)
 						goto bad;
 				if (mp->b_wptr - mp->b_rptr !=
 					 sizeof(struct unitdata_req))
 						goto bad;
 				if(err=chkaddr(proto->unitdata_req.DEST_addr))
 						goto bad;
 				putq(q, mp);
 				/* start device or mux output ... */
 				break;
 			bad:
 				freemsg(mp);
 				break;
 			}
	 }
return(0);
}

The write put procedure, protowput(), switches on the message type. The only types accepted are M_FLUSH and M_PROTO. For M_FLUSH messages, the driver performs the canonical flush handling (not shown). For M_PROTO messages, the driver assumes that the message block contains a union primitive and switches on the type field. Two types are understood: BIND_REQ and UNITDATA_REQ.

For a BIND_REQ, the current state is checked; it must be IDLE. Next, the message size is checked. If it is the correct size, the passed-in address is verified for legality by calling chkaddr. If everything checks, the incoming message is converted into an OK_ACK and sent upstream. If there was any error, the incoming message is converted into an ERROR_ACK and sent upstream.

For UNITDATA_REQ, the state is also checked; it must be BOUND. As above, the message size and destination address are checked. If there is any error, the message is discarded. If all is well, the message is put in the queue, and the lower half of the driver is started.

If the write put procedure receives a message type that it does not understand, either a bad b_datap->db_type or bad proto->type, the message is converted into an M_ERROR message and is then sent upstream.

The generation of UNITDATA_IND messages (not shown in the example) would normally occur in the device interrupt if this is a hardware driver or in the lower read put procedure if this is a multiplexer. The algorithm is simple: the data part of the message is prefixed by an M_PROTO message block that contains a unitdata_ind structure and is sent upstream.

Message Type Change Rules

To change a message type, use the following rules:

• You can only change the M_IOCTL family of message types to other M_IOCTL message types.

• M_DATA, M_PROTO, and M_PCPROTO are dependent on the modules, driver and service provider interfaces defined.

• A message type should not be changed if the reference count > 1.

• The data of a message should not be modified if the reference count > 1.

• All other message types are interchangeable as long as sufficient space has been allocated in the data buffer of the message.

Common ioctl Interfaces

Many ioctl operations are common to a class of STREAMS drivers or STREAMS modules. Modules that deal with terminals usually implement a subset of the termio(7I) ioctls. Similarly, drivers that deal with audio devices usually implement a subset of the audio(7I) interfaces.

Because no data structures have changed size as a result of the LP64 data model for either termio(7I) or audio(7I), you do no need to use any of the structure macros to decode any of these ioctls.

FIORDCHK

The FIORDCHK ioctl returns a count (in bytes) of the number of bytes to be read as the return value. Although FIORDCHK should be able to return more than MAXINT bytes, it is constrained to returning an int by the type of the ioctl(2) function.

FIONREAD

The FIONREAD ioctl returns the number of data bytes (in all data messages queued) in the location pointed to by the arg parameter. The ioctl returns a 32-bit quantity for both 32-bit and 64-bit application., Therefore, code that passes the address of a long variable needs to be changed to pass an int variable for 64–bit applications.

I_NREAD

The I_NREAD ioctl (streamio(7I)) is an informational ioctl that counts the data bytes as well as the number of messages in the stream head read queue. The number of data bytes is returned in the location pointed to by the arg parameter of the ioctl. The number of messages in the stream head read queue is returned as the return value of the ioctl.

Like FIONREAD, the arg parameter to the I_NREAD ioctl should be a pointer to an int, not a long. And, like FIORDCHK, the return value is constrained to be less than or equal to MAXINT bytes, even if more data is available.

signal Message

STREAMS modules and drivers send signals to application processes through a special signal message. If the signal specified by the module or driver is not SIGPOLL (see signal(3C)), the signal is delivered to the process group associated with the stream. If the signal is SIGPOLL, the signal is only sent to processes that have registered for the signal by using the I_SETSIG ioctl(2).

Modules or drivers use an M_SIG message to insert an explicit in-band signal into a message stream. For example, a message can be sent to the application process immediately before a particular service interface message. When the M_SIG message reaches the head of the stream read queue, a signal is generated and the M_SIG message is removed. The service interface message is the next message to be processed by the user. (The M_SIG message is usually defined as part of the service interface of the driver or module.)

Chapter 9 STREAMS Drivers

This chapter describes the operation of STREAMS drivers and some of the processing typically required in the drivers.

• STREAMS Device Drivers

• STREAMS Driver Entry Points

• STREAMS Driver Code Samples

STREAMS Device Drivers

STREAMS drivers can be considered a subset of device drivers in general and character device drivers in particular. While there are some differences between STREAMS drivers and non-STREAMS drivers, much of the information contained in Writing Device Drivers also applies to STREAMS drivers.

The word module is used differently when talking about drivers. A device driver is a kernel-loadable module that provides the interface between a device and the Device Driver Interface, and is linked to the kernel when it is first invoked.

STREAMS drivers share a basic programming model with STREAMS modules. Information common to both drivers and modules is discussed in Chapter 10, STREAMS Modules. After summarizing some basic device driver concepts, this chapter discusses several topics specific to STREAMS device drivers (and not covered elsewhere) and then presents code samples illustrating basic STREAMS driver processing.

Basic Driver

A device driver is a loadable kernel module that translates between an I/O device and the kernel to operate the device.

Device drivers can also be software-only, implementing a pseudo-device such as RAM disk or a pseudo-terminal that only exists in software.

In the Solaris operating environment, the interface between the kernel and device drivers is called the Device Driver Interface (DDI/DKI). This interface is specified in the Section 9E manual pages that specify the driver entry points. Section 9 also details the kernel data structures (9S) and utility functions (9F) available to drivers.

The DDI protects the kernel from device specifics. Application programs and the rest of the kernel need little (if any) device-specific code to use the device. The DDI makes the system more portable and easier to maintain.

There are three basic types of device drivers corresponding to the three basic types of devices. Character devices handle data serially and transfer data to and from the processor one character at a time, the same as keyboards and low performance printers. Serial block devices and drivers also handle data serially, but transfer data to and from memory without processor intervention, the same as tape drives. Direct access block devices and drivers also transfer data without processor intervention and blocks of storage on the device can be addressed directly, the same as disk drives.

There are two types of character device drivers: standard character device drivers and STREAMS device drivers. STREAMS is a separate programming model for writing a character driver. Devices that receive data asynchronously (such as terminal and network devices) are suited to a STREAMS implementation.

STREAMS drivers share some kinds of processing with STREAMS modules. Important differences between drivers and modules include how the application manipulates drivers and modules and how interrupts are handled. In STREAMS, drivers are opened and modules are pushed. A device driver has an interrupt routine to process hardware interrupts.

STREAMS Driver Entry Points

STREAMS drivers have five different points of contact with the kernel:

STREAMS Configuration Entry Points

As with other SunOS 5 drivers, STREAMS drivers are dynamically linked and loaded when referred to for the first time. For example, when the system is initially booted, the STREAMS pseudo-tty slave pseudo-driver (pts(7D)) is loaded automatically into the kernel when it is first opened.

In STREAMS, the header declarations differ between drivers and modules. The word “module” is used in two different ways when talking about drivers. There are STREAMS modules, which are pushable nondriver entities, and there are kernel-loadable modules, which are components of the kernel. See the appropriate chapters in Writing Device Drivers.

The kernel configuration mechanism must distinguish between STREAMS devices and traditional character devices because system calls to STREAMS drivers are processed by STREAMS routines, not by the system driver routines. The streamtab pointer in the cb_ops(9S) structure provides this distinction. If it is NULL, there are no STREAMS routines to execute; otherwise, STREAMS drivers initialize streamtab with a pointer to a streamtab(9S) structure containing the driver's STREAMS queue processing entry points.

STREAMS Initialization Entry Points

The initialization entry points of STREAMS drivers must perform the same tasks as those of non-STREAMS drivers. See Writing Device Drivers for more information.

STREAMS Table-Driven Entry Points

In non-STREAMS drivers, most of the driver's work is accomplished through the entry points in the cb_ops(9S) structure. For STREAMS drivers, most of the work is accomplished through the message-based STREAMS queue processing entry points.

Figure 9–1 shows multiple streams (corresponding to minor devices) connecting to a common driver. There are two distinct streams opened from the same major device. Consequently, they have the same streamtab and the same driver procedures.

Figure 9–1 Device Driver Streams

Multiple instances (minor devices) of the same driver are handled during the initial open for each device. Typically, a driver stores the queue address in a driver-private structure that is uniquely identified by the minor device number. (The DDI/DKI provides a mechanism for uniform handling of driver-private structures; see ddi_soft_state(9F)). The q_ptr of the queue points to the private data structure entry. When the messages are received by the queue, the calls to the driver put and service procedures pass the address of the queue, enabling the procedures to determine the associated device through the q_ptr field.

STREAMS guarantees that only one open or close can be active at a time per major/minor device pair.

STREAMS Queue Processing Entry Points

STREAMS device drivers have processing routines that are registered with the framework through the streamtab structure. The put procedure is a driver's entry point, but it is a message (not system) interface. STREAMS drivers and STREAMS modules implement these entry points similarly, as described in Entry Points.

The stream head translates write(2) and ioctl(2) calls into messages and sends them downstream to be processed by the driver's write queue put(9E) procedure. read is seen directly only by the stream head, which contains the functions required to process system calls. A STREAMS driver does not check system interfaces other than open and close, but it can detect the absence of a read indirectly if flow control propagates from the stream head to the driver and affects the driver's ability to send messages upstream.

For read-side processing, when the driver is ready to send data or other information to a user process, it prepares a message and sends it upstream to the read queue of the appropriate (minor device) stream. The driver's open routine generally stores the queue address corresponding to this stream.

For write-side (or output) processing, the driver receives messages in place of a write call. If the message cannot be sent immediately to the hardware, it may be stored on the driver's write message queue. Subsequent output interrupts can remove messages from this queue.

A driver is at the end of a stream. As a result, drivers must include standard processing for certain message types that a module might be able to pass to the next component. For example, a driver must process all M_IOCTL messages; otherwise, the stream head blocks for an M_IOCNAK, M_IOCACK, or until the timeout (potentially infinite) expires. If a driver does not understand an ioctl(2), an M_IOCNAK message is sent upstream.

Messages that are not understood by the drivers should be freed.

The stream head locks up the stream when it receives an M_ERROR message, so driver developers should be careful when using the M_ERROR message.

STREAMS Interrupt Handlers

Most hardware drivers have an interrupt handler routine. You must supply an interrupt routine for the device's driver. The interrupt handling for STREAMS drivers is not fundamentally different from that for other device drivers. Drivers usually register interrupt handlers in their attach(9E)entry point, using ddi_add_intr(9F). Drivers unregister the interrupt handler at detach time using ddi_remove_intr(9F).

The system also supports software interrupts. The routines ddi_add_softintr(9F) and ddi_remove_softintr(9F) register and unregister (respectively) soft-interrupt handlers. A software interrupt is generated by calling ddi_trigger_softintr(9F).

See Writing Device Drivers for more information.

Driver Unloading

STREAMS drivers can prevent unloading through the standard driver detach(9E) entry point.

STREAMS Driver Code Samples

The following discussion describes characteristics of a STREAMS driver:

• Basic hardware/pseudo drivers

  This type of driver communicates with a specific piece of hardware (or simulated hardware). The lp example simulates a simple printer driver.

• Clonable drivers

  The STREAMS framework supports a CLONEOPEN facility, which allows multiple streams to be opened from a single special file. If a STREAMS device driver chooses to support CLONEOPEN, it can be referred to as a clonable device. The attach(9E) routines from two Solaris drivers, ptm(7D) and log(7D), illustrate two approaches to cloning.

• Multiple instances in drivers

  A multiplexer driver is a regular STREAMS driver that can handle multiple streams connected to it instead of just one stream. Multiple connections occur when more than one minor device of the same driver is in use. See Cloning STREAMS Drivers for more information.

Printer Driver Example

Example 9–1 is a sample print driver for an interrupt-per-character line printer. The driver is unidirectional—it has no read-side processing. It demonstrates some differences between module and driver programming, including the following:

• Declarations for driver configuration

• Open handling

• A driver is passed a device number

• Flush handling

• A driver must loop M_FLUSH messages back upstream

• Interrupt routine

• A driver registers interrupt handler and processes interrupts

Most of the STREAMS processing in the driver is independent of the actual printer hardware; in this example, actual interaction with the printer is limited to the lpoutchar function, which prints one character at a time. For purposes of demonstration, the “printer hardware” is actually the system console, accessed through cmn_err(9F). Since there's no actual hardware to generate a genuine hardware interrupt, lpoutchar simulates interrupts using ddi_trigger_softintr(9F). For a real printer, the lpoutchar function is rewritten to send a character to the printer, which should generate a hardware interrupt.

The driver declarations follow. After specifying header files (include <sys/ddi.h> and <sys/sunddi.h> as the last two header files), the driver declares a per-printer structure, struct lp. This structure contains members that enable the driver to keep track of each instance of the driver, such as flags (what the driver is doing), msg (the current STREAMS print message), qptr (pointer to the stream's write queue), dip (the instance's device information handle), iblock_cookie (for registering an interrupt handler), siid (the handle of the soft interrupt), and lp_lock (a mutex to protect the data structure from multithreaded race conditions). The driver next defines the bits for the flags member of struct lp; the driver defines only one flag, BUSY.

Following function prototypes, the driver provides some standard STREAMS declarations: a module_info(9S) structure (minfo), a qinit(9S) structure for the read side (rinit) that is initialized by the driver's open and close entry points, a qinit(9S) structure for the write side (winit) that is initialized by the write put procedure, and a streamtab(9S) that points to rinit and winit. The values in the module name and ID fields in the module_info(9S) structure must be unique in the system. Because the driver is unidirectional, there is no read side put or service procedure. The flow control limits for use on the write side are 50 bytes for the low-watermark and 150 bytes for the high-watermark.

The driver next declares lp_state. This is an anchor on which the various “soft-state” functions provided by the DDK operate. The ddi_soft_state(9F) manual page describes how to maintain multiple instances of a driver.

The driver next declares acb_ops(9S) structure, which is required in all device drivers. In non-STREAMS device drivers, cb_ops(9S) contains vectors to the table-driven entry points. For STREAMS drivers, however, cb_ops(9S) contains mostly nodev entries. The cb_stream field, however, is initialized with a pointer to the driver's streamtab(9S) structure. This indicates to the kernel that this driver is a STREAMS driver.

Next, the driver declares a dev_ops(9S) structure, which points to the various initialization entry points as well as to the cb_ops(9S) structure. Finally, the driver declares a struct moldrv and a struct modlinkage for use by the kernel linker when the driver is dynamically loaded. struct moldrv contains a pointer to mod_driverops (a significant difference between a STREAMS driver and a STREAMS module—a STREAMS module would contain a pointer to mod_strops instead).

Example 9–1 Simple Line Printer Driver

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/signal.h>
#include <sys/errno.h>
#include <sys/cred.h>
#include <sys/stat.h>
#include <sys/modctl.h>
#include <sys/conf.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

/* This is a private data structure, one per minor device number */

struct lp {
	short flags; /* flags -- see below */
	mblk_t *msg; /* current message being output */
	queue_t *qptr; /* back pointer to write queue */
	dev_info_t *dip; /* devinfo handle */
	ddi_iblock_cookie_t iblock_cookie;
	ddi_softintr_t siid;
	kmutex_t lp_lock; /* sync lock */
};

/* flags bits */

#define BUSY 1 /* dev is running, int is forthcoming */

/*
 * Function prototypes.
 */
static int lpattach(dev_info_t *, ddi_attach_cmd_t);
static int lpdetach(dev_info_t *, ddi_detach_cmd_t);
static int lpgetinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
static int lpidentify(dev_info_t *);
static uint lpintr(caddr_t lp);
static void lpout(struct lp *lp);
static void lpoutchar(struct lp *lp, char c);
static int lpopen(queue_t*, dev_t*, int, int, cred_t*);
static int lpclose(queue_t*, int, cred_t*);
static int lpwput(queue_t*, mblk_t*);

/* Standard Streams declarations */

static struct module_info minfo = {
	0xaabb,
	"lp",
	0,
	INFPSZ,
	150,
	50
};

static struct qinit rinit = {
	(int (*)()) NULL,
	(int (*)()) NULL,
	lpopen,
	lpclose,
	(int (*)()) NULL,
	&minfo,
	NULL
};

static struct qinit winit = {
	lpwput,
	(int (*)()) NULL,
	(int (*)()) NULL,
	(int (*)()) NULL,
	(int (*)()) NULL,
	&minfo,
	NULL
};

static struct streamtab lpstrinfo = { &rinit, &winit, NULL, NULL };

/*
 * An opaque handle where our lp lives
 */
static void *lp_state;

/* Module Loading/Unloading and Autoconfiguration declarations */

static struct cb_ops lp_cb_ops = {
	nodev, /* cb_open */
	nodev, /* cb_close */
	nodev, /* cb_strategy */
	nodev, /* cb_print */
	nodev, /* cb_dump */
	nodev, /* cb_read */
	nodev, /* cb_write */
	nodev, /* cb_ioctl */
	nodev, /* cb_devmap */
	nodev, /* cb_mmap */
	nodev, /* cb_segmap */
	nochpoll, /* cb_chpoll */
	ddi_prop_op, /* cb_prop_op */
	&lpstrinfo, /* cb_stream */
	D_MP | D_NEW, /* cb_flag */
};

static struct dev_ops lp_ops = {
	DEVO_REV, /* devo_rev */
	0, /* devo_refcnt */
	lpgetinfo, /* devo_getinfo */
	lpidentify, /* devo_identify */
	nulldev, /* devo_probe */
	lpattach, /* devo_attach */
	lpdetach, /* devo_detach */
	nodev, /* devo_reset */
	&lp_cb_ops, /* devo_cb_ops */
	(struct bus_ops *)NULL /* devo_bus_ops */
};

/*
 * Module linkage information for the kernel.
 */
static struct modldrv modldrv = {
	&mod_driverops,
	"Simple Sample Printer Streams Driver", /* Description */
	&lp_ops, /* driver ops */
};

static struct modlinkage modlinkage = {
	MODREV_1, &modldrv, NULL
};

Example 9–2 shows the required driver configuration entry points _init(9E), _fini(9E), and _info(9E). In addition to installing the driver using mod_install(9F), the _init entry point also initializes the per-instance driver structure using ddi_soft_state_init(9F). _fini(9E) performs the complementary calls to mod_remove(9F) and ddi_soft_state_fini(9F) to unload the driver and release the resources used by the soft-state routines.

Example 9–2 Configuration Entry Points

int
_init(void)
{
	int e;

	if ((e = ddi_soft_state_init(&lp_state,
	 sizeof (struct lp), 1)) != 0) {
		return (e);
	}

	if ((e = mod_install(&modlinkage)) != 0) {
		ddi_soft_state_fini(&lp_state);
	}

	return (e);
}

int
_fini(void)
{
	int e;

	if ((e = mod_remove(&modlinkage)) != 0) {
		return (e);
	}
	ddi_soft_state_fini(&lp_state);
	return (e);
}

int
_info(struct modinfo *modinfop)
{
	return (mod_info(&modlinkage, modinfop));
}

Example 9–3 shows the lp driver's implementation of the initialization entry points. In lpidentify, the driver ensures that the name of the device being attached is “lp”.

lpattach first uses ddi_soft_state_zalloc(9F) to allocate a per-instance structure for the printer being attached. Next it creates a node in the device tree for the printer using ddi_create_minor_node(9F); user programs use the node to access the device. lpattach then registers the driver interrupt handler because the sample is driver pseudo-hardware, the driver uses soft interrupts. A driver for a real printer would use ddi_add_intr(9F) instead of ddi_add_softintr(9F). A driver for a real printer would also need to perform any other required hardware initialization in lpattach. Finally, lpattach initializes the per-instance mutex.

In lpdetach, the driver undoes everything it did in lpattach.

lpgetinfo uses the soft-state structures to obtain the required information.

Example 9–3 Initialization Entry Points

static int
lpidentify(dev_info_t *dip)
{
	if (strcmp(ddi_get_name(dip), "lp") == 0) {
		return (DDI_IDENTIFIED);
	} else
		return (DDI_NOT_IDENTIFIED);
}

static int
lpattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
{
	int instance;
	struct lp *lpp;

	switch (cmd) {

	case DDI_ATTACH:

		instance = ddi_get_instance(dip);

		if (ddi_soft_state_zalloc(lp_state, instance) != DDI_SUCCESS) {
			cmn_err(CE_CONT, "%s%d: can't allocate state\n",
			 ddi_get_name(dip), instance);
			return (DDI_FAILURE);
		} else
			lpp = ddi_get_soft_state(lp_state, instance);

		if (ddi_create_minor_node(dip, "strlp", S_IFCHR,
		 instance, NULL, 0) == DDI_FAILURE) {
			ddi_remove_minor_node(dip, NULL);
			goto attach_failed;

		}

		lpp->dip = dip;
		ddi_set_driver_private(dip, (caddr_t)lpp);

		/* add (soft) interrupt */


		if (ddi_add_softintr(dip, DDI_SOFTINT_LOW, &lpp->siid,
		 &lpp->iblock_cookie, 0, lpintr, (caddr_t)lpp)
		 != DDI_SUCCESS) {
			ddi_remove_minor_node(dip, NULL);
			goto attach_failed;
		}

		mutex_init(&lpp->lp_lock, "lp lock", MUTEX_DRIVER,
		 (void *)lpp->iblock_cookie);


		ddi_report_dev(dip);
		return (DDI_SUCCESS);

	default:
		return (DDI_FAILURE);
	}

attach_failed:
	/*
	 * Use our own detach routine to toss
	 * away any stuff we allocated above.
	 */
	(void) lpdetach(dip, DDI_DETACH);
	return (DDI_FAILURE);
}

static int
lpdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
{
	int instance;
	struct lp *lpp;

	switch (cmd) {

	case DDI_DETACH:
		/*
		 * Undo what we did in lpattach, freeing resources
		 * and removing things we installed. The system
		 * framework guarantees we are not active with this devinfo
		 * node in any other entry points at this time.
		 */
		ddi_prop_remove_all(dip);
		instance = ddi_get_instance(dip);
		lpp = ddi_get_soft_state(lp_state, instance);
		ddi_remove_minor_node(dip, NULL);
		ddi_remove_softintr(lpp->siid);
		ddi_soft_state_free(lp_state, instance);
		return (DDI_SUCCESS);

	default:
		return (DDI_FAILURE);
	}
}

/*ARGSUSED*/
static int
lpgetinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
          void **result)
{
	struct lp *lpp;
	int error = DDI_FAILURE;

	switch (infocmd) {
	case DDI_INFO_DEVT2DEVINFO:
		if ((lpp = ddi_get_soft_state(lp_state,
		 getminor((dev_t)arg))) != NULL) {
			*result = lpp->dip;
			error = DDI_SUCCESS;
		} else
			*result = NULL;
		break;

	case DDI_INFO_DEVT2INSTANCE:
		*result = (void *)getminor((dev_t)arg);
		error = DDI_SUCCESS;
		break;

	default:
		break;
	}

	return (error);
}

The STREAMS mechanism allows only one stream per minor device. The driver open routine is called whenever a STREAMS device is opened. open matches the correct private data structure with the stream using ddi_get_soft_state(9F). The driver open, lpopen in Example 9–4, has the same interface as the module open.

The stream flag, sflag, must have the value 0, indicating a normal driver open. devp pointers to the major/minor device number for the port. After checking sflag, lpopen uses devp to find the correct soft-state structure.

The next check, if (q->q_ptr)..., determines if the printer is already open. q_ptr is a driver or module private data pointer. It can be used by the driver for any purpose and is initialized to zero by STREAMS before the first open. In this example, the driver sets the value of q_ptr, in both the read and write queue structures, to point to the device's per-instance data structure. If the pointer is non-NULL, it means the printer is already open, so lpopen returns EBUSY to avoid merging printouts from multiple users.

The driver close routine is called by the stream head. Any messages left in the queue are automatically removed by STREAMS. The stream is dismantled and data structures are released.

Example 9–4 Queue Processing Entry Points

/*ARGSUSED*/
static int
lpopen(
	queue_t *q,	/* read queue */
	dev_t *devp,
	int flag,
	int sflag,
	cred_t *credp)

{

	struct lp *lp;

	if (sflag)	/* driver refuses to do module or clone open */
			return (ENXIO);

	if ((lp = ddi_get_soft_state(lp_state, getminor(*devp))) == NULL)
			return (ENXIO);

	/* Check if open already. Can't have multiple opens */

	if (q->q_ptr) {
			return (EBUSY);
	}

	lp->qptr = WR(q);
	q->q_ptr = (char *) lp;
	WR(q)->q_ptr = (char *) lp;
	qprocson(q);
	return (0);
}

/*ARGSUSED*/
static int
lpclose(
	queue_t *q,		/* read queue */
	int flag,
	cred_t *credp)
{

	struct lp *lp;

	qprocsoff(q);
	lp = (struct lp *) q->q_ptr;

	/*
	 * Free message, queue is automatically
	 * flushed by STREAMS
	 */
	mutex_enter(&lp->lp_lock);

	if (lp->msg) {
			freemsg(lp->msg);
			lp->msg = NULL;
	}

	lp->flags = 0;
	mutex_exit(&lp->lp_lock);

	return (0);
}

There are no physical pointers between the read and write queue of a pair. WR(9F) is a queue pointer function. WR(9F) generates the write pointer from the read pointer. RD(9F) and OTHERQ(9F) are additional queue pointer functions. RD(9F) generates the read pointer from the write pointer, and OTHERQ(9F) generates the mate pointer from either.

Driver Flush Handling

The write put procedure in Example 9–5, lpwput, illustrates driver M_FLUSH handling. Note that all drivers are expected to incorporate flush handling.

If FLUSHW is set, the write message queue is flushed, and (in this example) the leading message (lp->msg) is also flushed. lp_lock protects the driver's per-instance data structure.

In most drivers, if FLUSHR is set, the read queue is flushed. However, in this example, no messages are ever placed on the read queue, so flushing it is not necessary. The FLUSHW bit is cleared and the message is sent upstream using qreply(9F). If FLUSHR is not set, the message is discarded.

The stream head always performs the following actions on flush requests received on the read side from downstream. If FLUSHR is set, messages waiting to be sent to user space are flushed. If FLUSHW is set, the stream head clears the FLUSHR bit and sends the M_FLUSH message downstream. In this manner, a single M_FLUSH message sent from the driver can reach all queues in a stream. A module must send two M_FLUSH messages to have the same effect.

lpwput queues M_DATA and M_IOCTL messages and if the device is not busy, starts output by calling lpout. Message types that are not recognized are discarded (in the default case of the switch).

Example 9–5 Driver Flush Handling

static int lpwput(
	queue_t *q,	/* write queue */
	mblk_t *mp)	/* message pointer */
{
	struct lp *lp;

	lp = (struct lp *)q->q_ptr;

	switch (mp->b_datap->db_type) {
	default:
		freemsg(mp);
		break;

	case M_FLUSH: /* Canonical flush handling */
		if (*mp->b_rptr & FLUSHW) {
			flushq(q, FLUSHDATA);
			mutex_enter(&lp->lp_lock); /* lock any access to lp */

			if (lp->msg) {
				freemsg(lp->msg);
				lp->msg = NULL;
			}

			mutex_exit(&lp->lp_lock);

		}

		if (*mp->b_rptr & FLUSHR) {
			*mp->b_rptr &= ~FLUSHW;
			qreply(q, mp);
		} else
			freemsg(mp);

		break;

	case M_IOCTL:
	case M_DATA:
		(void) putq(q, mp);
		mutex_enter(&lp->lp_lock);

		if (!(lp->flags & BUSY))
			lpout(lp);

		mutex_exit(&lp->lp_lock);

	}
	return (0);
}

Print Driver Interrupt

Example 9–6 shows the interrupt handling for the printer driver.

lpintr is the driver-interrupt handler registered by the attach routine.

lpout takes a single character from the queue and sends it to the printer. For convenience, the message currently being output is stored in lp->msg in the per-instance structure. This assumes that the message is called with the mutex held.

lpoutchar sends a single character to the printer (in this case the system console using cmn_err(9F)) and interrupts when complete. Of course, hardware would generate a hard interrupt, so the call to ddi_trigger_softintr(9F) would be unnecessary.

Example 9–6 Driver Interrupt Handling

/* Device interrupt routine */static uint
lpintr(caddr_t lp)	 /* minor device number of lp */
{

	struct lp *lpp = (struct lp *)lp;
	
	mutex_enter(&lpp->lp_lock);

	if (!(lpp->flags & BUSY)) {
			mutex_exit(&lpp->lp_lock);
			return (DDI_INTR_UNCLAIMED);
	}

	lpp->flags &= ~BUSY;
	lpout(lpp);
	mutex_exit(&lpp->lp_lock);

	return (DDI_INTR_CLAIMED);
}

/* Start output to device - used by put procedure and driver */

static void
lpout(
	struct lp *lp)
{

	mblk_t *bp;
	queue_t *q;

	q = lp->qptr;

 loop:
	if ((bp = lp->msg) == NULL) { /*no current message*/
			if ((bp = getq(q)) == NULL) {
				lp->flags &= ~BUSY;	
				return;
			}
			if (bp->b_datap->db_type == M_IOCTL) {
				/* lpdoioctl(lp, bp); */
				goto loop;
			}

			lp->msg = bp; /* new message */

		}

	if (bp->b_rptr >= bp->b_wptr) { /* validate message */

			bp = lp->msg->b_cont;
			lp->msg->b_cont = NULL;
			freeb(lp->msg);
			lp->msg = bp;
			goto loop;
	}

	lpoutchar(lp, *bp->b_rptr++); /*output one character*/
	lp->flags |= BUSY;
}

static void
lpoutchar(
	struct lp *lp,
	char c)
{
	cmn_err(CE_CONT, “%c”, c);
	ddi_trigger_softintr(lp->siid);
}

Driver Flow Control

The same utilities (described in Chapter 10, STREAMS Modules) and mechanisms used for module flow control are used by drivers.

When the message is queued, putq(9F) increments the value of q_count by the size of the message and compares the result to the driver's write high-watermark (q_hiwat) value. If the count reaches q_hiwat, putq(9F) sets the internal FULL indicator for the driver write queue. This causes messages from upstream to be halted (canputnext(9F) returns FALSE) until the write queue count drops below q_lowat. The driver messages waiting to be output through lpout are dequeued by the driver output interrupt routine with getq(9F), which decrements the count. If the resulting count is below q_lowat, getq(9F) back-enables any upstream queue that had been blocked.

For priority band data, qb_count, qb_hiwat, and qb_lowat are used.

STREAMS with flow control can be used on the driver read side to handle temporary upstream blocks.

To some extent, a driver or a module can control when its upstream transmission becomes blocked. Control is available through the M_SETOPTS message (see Appendix A, Message Types) to modify the stream head read-side flow control limits.

Cloning STREAMS Drivers

To eliminate polling, STREAMS drivers can be made clonable. If a STREAMS driver is implemented as a clonable device, a single node in the file system can be opened to access any unused device that the driver controls. This special node guarantees that each user is allocated a separate stream to the driver for each open call. Each stream is associated with an unused minor device, so the total number of streams that may be connected to a particular clonable driver is limited only by the number of minor devices configured for that driver.

In previous examples, each user process connected a stream to a driver by explicitly opening a particular minor device of the driver. Each minor device had its own node in the device tree file system. Often, there is a need for a user process to connect a new stream to a driver regardless of which minor device is used to access the driver. In the past, this forced the user process to poll the various minor device nodes of the driver for an available minor device.

The clone model is useful, for example, in a networking environment where a protocol pseudo-device driver requires each user to open a separate stream over which it establishes communication. (The decision to implement a STREAMS driver as a clonable device is made by the designers of the device driver. Knowledge of the clone driver implementation is not required to use it.)

There are two ways to open as a clone device. The first is to use the STREAMS framework-provided clone device, which arranges to open the device with the CLONEOPEN flag passed in. This method is demonstrated in Example 9–7, which shows the attach and open routines for the pseudo-terminal master ptm(7D) driver. The second way is to have the driver open itself as a clone device, without intervention from the system clone device. This method is demonstrated in the attach and open routines for the log(7D) device in Example 9–8.

The ptm(7D) device, which uses the system-provided clone device, sets up two nodes in the device file system. One has a major number of 23 (ptm's assigned major number) and a minor number of 0. The other node has a major number of 11 (the clone device's assigned major number) and a minor number of 23 (ptm's assigned major number). The driver's attach routine (see Example 9–7) calls to ddi_create_minor_node(9F) twice. First, to set up the “normal” node (major number 23); second, to specify CLONE_DEV as the last parameter, making the system create the node with major 11.

crw-rw-rw-   1 sys       11, 23 Mar  6 02:05 clone:ptmx
crw-------   1 sys       23,  0 Mar  6 02:05 ptm:ptmajor

When the special file /devices/pseudo/clone@0:ptmx is opened, the clone driver code in the kernel (accessed by major 11) passes the CLONEOPEN flag in the sflag parameter to the ptm(7D) open routine. ptm's open routine checks sflag to make sure it is being called by the clone driver. The open routine next attempts to find an unused minor device for the open by searching its table of minor devices. (PT_ENTER_WRITE and PT_EXIT_WRITE are driver-defined macros for entering and exiting the driver's mutex.) If it succeeds (and following other open processing), the open routine constructs a new dev_t with the new minor number, which it passes back to its caller in the devp parameter. (The new minor number is available to the user program that opened the clonable device through an fstat(2) call.)

Example 9–7 Opening a System Clone Device

static int
ptm_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
{
	  if (cmd != DDI_ATTACH)
			return (DDI_FAILURE);

	  if (ddi_create_minor_node(devi, "ptmajor", S_IFCHR, 0, NULL, 0) 
				== DDI_FAILURE) {
			ddi_remove_minor_node(devi, NULL);
			return (DDI_FAILURE);
	  }
	  if (ddi_create_minor_node(devi, "ptmx", S_IFCHR, 0, NULL, CLONE_DEV) 
				== DDI_FAILURE) {
			ddi_remove_minor_node(devi, NULL);
			return (DDI_FAILURE);
	  }
	  ptm_dip = devi;
	  return (DDI_SUCCESS);
}

static int
ptmopen(
		queue_t	*rqp,		/* pointer to the read side queue */
		dev_t	*devp,		/* pointer to stream tail's dev */
		int		oflag,		/* the user open(2) supplied flags */
		int		sflag,		/* open state flag */
		cred_t	*credp)	/* credentials */
{
		struct pt_ttys	*ptmp;
		mblk_t	*mop;		/* ptr to a setopts message block */
		minor_t	dev;

		if (sflag != CLONEOPEN) {
			return (EINVAL);
		}

		for (dev = 0; dev < pt_cnt; dev++) {
			ptmp = &ptms_tty[dev];
			PT_ENTER_WRITE(ptmp);
			if (ptmp->pt_state & (PTMOPEN | PTSOPEN | PTLOCK)) {
				PT_EXIT_WRITE(ptmp);
			} else
				break;
		}

		if (dev >= pt_cnt) {
			return (ENODEV);
		}

		... <other open processing> ...

		/*
		 * The input, devp, is a major device number, the output is put into
		 * into the same parm as a major,minor pair.
		 */
		*devp = makedevice(getmajor(*devp), dev);
		return (0);
}

The log(7D) driver uses the second method; it clones itself without intervention from the system clone device. The log(7D) driver's attach routine (in Example 9–8) is similar to the one in ptm(7D). It creates two nodes using ddi_create_minor_node(9F), but neither specifies CLONE_DEV as the last parameter. Instead, one of the devices has minor 0, the other minor CLONEMIN. These two devices provide log(7D) two interfaces: the first write-only, the second read-write (see the man page log(7D) for more information). Users open one node or the other. If they open the CONSWMIN (clonable, read-write) node, the open routine checks its table of minor devices for an unused device. If it is successful, it (like the ptm(7D) open routine) returns the new dev_t to its caller in devp.

Example 9–8 Opening the log Driver

static int
log_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
{
		if (ddi_create_minor_node(devi, "conslog", S_IFCHR, 0, NULL, NULL)
				 == DDI_FAILURE ||
				 ddi_create_minor_node(devi, "log", S_IFCHR, 5, NULL, NULL) 
				 == DDI_FAILURE) {
			ddi_remove_minor_node(devi, NULL);
			return (-1);
		}
		log_dip = devi;
		return (DDI_SUCCESS);
}

static int
logopen(
		queue_t *q,
		dev_t *devp,
		int flag,
		int sflag,
		cred_t *cr)
{
		int i;
		struct log *lp;

		/*
		 * A MODOPEN is invalid and so is a CLONEOPEN.
		 * This is because a clone open comes in as a CLONEMIN device open!!
		 */
		if (sflag)
			return (ENXIO);

		mutex_enter(&log_lock);
		switch (getminor(*devp)) {

		case CONSWMIN:
			if (flag & FREAD) { /* you can only write to this minor */
				mutex_exit(&log_lock);
				return (EINVAL);
			}
			if (q->q_ptr) { /* already open */
				mutex_exit(&log_lock);
				return (0);
			}
			lp = &log_log[CONSWMIN];
			break;

		case CLONEMIN:
			/*
			 * Find an unused minor > CLONEMIN.
			 */
				i = CLONEMIN + 1;
			for (lp = &log_log[i]; i < log_cnt; i++, lp++) {
				if (!(lp->log_state & LOGOPEN))
					break;
			}
			if (i >= log_cnt) {
				mutex_exit(&log_lock);
				return (ENXIO);
			}
			*devp = makedevice(getmajor(*devp), i); /* clone it */
			break;

		default:
			mutex_exit(&log_lock);
			return (ENXIO);
		}

		/*
		 * Finish device initialization.
		 */
		lp->log_state = LOGOPEN;
		lp->log_rdq = q;
		q->q_ptr = (void *)lp;
		WR(q)->q_ptr = (void *)lp;
		mutex_exit(&log_lock);
		qprocson(q);
		return (0);
}

Loop-Around Driver

The loop-around driver is a pseudo-driver that loops data from one open stream to another open stream. The associated files are almost like a full-duplex pipe to user processes. The streams are not physically linked. The driver is a simple multiplexer that passes messages from one stream's write queue to the other stream's read queue.

To create a connection, a process opens two streams, obtains the minor device number associated with one of the returned file descriptors, and sends the device number in an ioctl(2) to the other stream. For each open, the driver open places the passed queue pointer in a driver interconnection table, indexed by the device number. When the driver later receives an M_IOCTL message, it uses the device number to locate the other stream's interconnection table entry, and stores the appropriate queue pointers in both of the streams' interconnection table entries.

Subsequently, when messages other than M_IOCTL or M_FLUSH are received by the driver on either stream's write side, the messages are switched to the read queue following the driver on the other stream's read side. The resultant logical connection is shown in Figure 9–2. Flow control between the two streams must be handled explicitly, since STREAMS do not automatically propagate flow control information between two streams that are not physically connected.

Figure 9–2 Loop-Around Streams

Example 9–9 shows the loop-around driver code. The loop structure contains the interconnection information for a pair of streams. loop_loop is indexed by the minor device number. When a stream is opened to the driver, the driver places the address of the corresponding loop_loop element in the q_ptr (private data structure pointer) of the read-side and write-side queues. Since STREAMS clears q_ptr when the queue is allocated, a NULL value of q_ptr indicates an initial open. loop_loop verifies that this stream is connected to another open stream.

The code presented here for the loop-around driver represents a single-threaded, uniprocessor implementation. Chapter 12, Multithreaded STREAMS presents multiprocessor and multithreading issues such as locking to prevent race conditions and data corruption.

Example 9–9 contains the declarations for the driver.

Example 9–9 Declarations for the Loop-Around Driver

/* Loop-around driver */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/signal.h>
#include <sys/errno.h>
#include <sys/cred.h>
#include <sys/stat.h>
#include <sys/modctl.h>
#include <sys/conf.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

static int loop_identify(dev_info_t *);
static int loop_attach(dev_info_t *, ddi_attach_cmd_t);
static int loop_detach(dev_info_t *, ddi_detach_cmd_t);
static int loop_devinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
static int loopopen (queue_t*, dev_t*, int, int, cred_t*);
static int loopclose (queue_t*, int, cred_t*);
static int loopwput (queue_t*, mblk_t*);
static int loopwsrv (queue_t*);
static int looprsrv (queue_t*);

static dev_info_t *loop_dip;	/* private devinfo pointer */

static struct module_info minfo = {
		0xee12,
		“loop”,
		0,
		INFPSZ,
		512,
		128
};

static struct qinit rinit = {
		(int (*)()) NULL,
		looprsrv,
		loopopen,
		loopclose,
		(int (*)()) NULL,
		&minfo,
		NULL
};

static struct qinit winit = {
		loopwput,
		loopwsrv,
		(int (*)()) NULL,
		(int (*)()) NULL,
		(int (*)()) NULL,
		&minfo,
		NULL
};

static struct streamtab loopinfo= {
		&rinit,
		&winit,
		NULL,
		NULL
};

struct 	loop {
		queue_t *qptr;			/* back pointer to write queue */
		queue_t *oqptr;		/* pointer to connected read queue */
};

#define LOOP_CONF_FLAG (D_NEW | D_MP)

static struct cb_ops cb_loop_ops = {
   nulldev,               /* cb_open */
   nulldev,               /* cb_close */
   nodev,                 /* cb_strategy */
   nodev,                 /* cb_print */
   nodev,                 /* cb_dump */
   nodev,                 /* cb_read */
   nodev,                 /* cb_write */
   nodev,                 /* cb_ioctl */
   nodev,                 /* cb_devmap */
   nodev,                 /* cb_mmap */
   nodev,                 /* cb_segmap */
   nochpoll,              /* cb_chpoll */
   ddi_prop_op,           /* cb_prop_op */
   ( &loopinfo),          /* cb_stream */
   (int)(LOOP_CONF_FLAG)  /* cb_flag */
};

static struct dev_ops loop_ops = {
   DEVO_REV,                /* devo_rev */
   0,                       /* devo_refcnt */
   (loop_devinfo),          /* devo_getinfo */
   (loop_identify),         /* devo_identify */
   (nulldev),               /* devo_probe */
   (loop_attach),           /* devo_attach */
   (loop_detach),           /* devo_detach */
   (nodev),                 /* devo_reset */
   &(cb_loop_ops),          /* devo_cb_ops */
   (struct bus_ops *)NULL,  /* devo_bus_ops */
   (int (*)()) NULL         /* devo_power */
};

#define LOOP_SET ((`l'<<8)|1) /* in a .h file */
#define NLOOP	64
static struct loop loop_loop[NLOOP];
static int loop_cnt = NLOOP;

/*
 * Module linkage information for the kernel.
 */
extern struct mod_ops mod_strmodops;

static struct modldrv modldrv = {
		&mod_driverops, "STREAMS loop driver", &loop_ops
};

static struct modlinkage modlinkage = {
		MODREV_1, &modldrv, NULL
};

_init()
{
		return (mod_install(&modlinkage));
}

_info(modinfop)
		struct modinfo *modinfop;
{
		return (mod_info(&modlinkage, modinfop));
}

_fini(void)
{
		return (mod_remove(&modlinkage));
}

Example 9–10 contains the initialization routines.

Example 9–10 Initialization Routines for the Loop-around Driver

static int
loop_identify(dev_info_t *devi)
{
	  if (strcmp(ddi_get_name(devi), "loop") == 0)
			return (DDI_IDENTIFIED);
	  else
			return (DDI_NOT_IDENTIFIED);
}

static int
loop_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
{
	  if (cmd != DDI_ATTACH)
			return (DDI_FAILURE);

	  if (ddi_create_minor_node(devi, "loopmajor", S_IFCHR, 0, NULL, 0) 
				== DDI_FAILURE) {
			ddi_remove_minor_node(devi, NULL);
			return (DDI_FAILURE);
	  }
	  if (ddi_create_minor_node(devi, "loopx", S_IFCHR, 0, NULL, CLONE_DEV)
				 == DDI_FAILURE) {
			ddi_remove_minor_node(devi, NULL);
			return (DDI_FAILURE);
	  }

	  loop_dip = devi;

	  return (DDI_SUCCESS);
}

static int
loop_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
{
	  if (cmd != DDI_DETACH)
			return (DDI_FAILURE);

	  ddi_remove_minor_node(devi, NULL);
	  return (DDI_SUCCESS);
}

/*ARGSUSED*/
static int
loop_devinfo(
	  dev_info_t *dip, 
	  ddi_info_cmd_t infocmd, 
	  void *arg, 
	  void **result)
{
	  int error;

	  switch (infocmd) {
	  case DDI_INFO_DEVT2DEVINFO:
			if (loop_dip == NULL) {
				error = DDI_FAILURE;
			} else {
				*result = (void *) loop_dip;
				error = DDI_SUCCESS;
			}
			break;
	  case DDI_INFO_DEVT2INSTANCE:
			*result = (void *)0;
			error = DDI_SUCCESS;
			break;
	  default:
			error = DDI_FAILURE;
	  }
	  return (error);

}

The open procedure (in Example 9–11) includes canonical clone processing that enables a single file system node to yield a new minor device/vnode each time the driver is opened. In loopopen, sflag can be CLONEOPEN, indicating that the driver picks an unused minor device. In this case, the driver scans its private loop_loop data structure to find an unused minor device number. If sflag is not set to CLONEOPEN, the passed-in minor device specified by getminor(*devp) is used.

Example 9–11 Opening the Loop-Around Driver

/*ARGSUSED*/
static int loopopen(
		queue_t *q,
		dev_t *devp,
		int flag,
		int sflag,
		cred_t *credp)
{
		struct loop *loop;
		minor_t newminor;

		if (q->q_ptr) /* already open */
			return(0);

		/*
		 * If CLONEOPEN, pick a minor device number to use.
		 * Otherwise, check the minor device range.
		 */

		if (sflag == CLONEOPEN) {
			for (newminor = 0; newminor < loop_cnt; newminor++ ) {
				if (loop_loop[newminor].qptr == NULL)
					break;
			}
		} else
			newminor = getminor(*devp);

		if (newminor >= loop_cnt)
			return(ENXIO);

		/*
		 * construct new device number and reset devp
		 * getmajor gets the major number
		 */

		*devp = makedevice(getmajor(*devp), newminor);
		loop = &loop_loop[newminor];
		WR(q)->q_ptr = (char *) loop;
		q->q_ptr = (char *) loop;
		loop->qptr = WR(q);
		loop->oqptr = NULL;

		qprocson(q);

		return(0);
}

Because the messages are switched to the read queue following the other stream's read side, the driver needs a put procedure only on its write side. loopwput (in Example 9–12) shows another use of an ioctl(2). The driver supports the ioc_cmd value LOOP_SET in the iocblk(9S) of the M_IOCTL message. LOOP_SET makes the driver connect the current open stream to the stream indicated in the message. The second block of the M_IOCTL message holds an integer that specifies the minor device number of the stream to which to connect.

The LOOP_SET ioctl(2) processing involves several checks:

• Does the second block have the proper amount of data?

• Is the “to” device in range?

• Is the “to” device open?

• Is the current stream disconnected?

• Is the “to” stream disconnected?

If these checks pass, the read queue pointers for the two streams are stored in the respective oqptr fields. This cross-connects the two streams indirectly, through loop_loop.

The put procedure incorporates canonical flush handling.

loopwput queues all other messages (for example, M_DATA or M_PROTO) for processing by its service procedure. A check is made that the stream is connected. If not, M_ERROR is sent to the stream head. Certain message types can be sent upstream by drivers and modules to the stream head where they are translated into actions detectable by user processes. These messages may also modify the state of the stream head:

M_ERROR

Causes the stream head to lock up. Message transmission between stream and user processes is terminated. All subsequent system calls except close(2) and poll(2) fail. Also causes M_FLUSH, clearing all message queues, to be sent downstream by the stream head.

M_HANGUP

Terminates input from a user process to the stream. All subsequent system calls that would send messages downstream fail. Once the stream head read message queue is empty, EOF is returned on reads. This can also result in SIGHUP being sent to the process group's session leader.

M_SIG/M_PCSIG

Causes a specified signal to be sent to the process group associated with the stream.

putnextctl(9F) and putnextctl1(9F) allocate a nondata (that is, not M_DATA, M_DELAY, M_PROTO, or M_PCPROTO) type message, place one byte in the message (for putnextctl1(9F)), and call the put(9E) procedure of the specified queue.

Example 9–12 Use of ioctl to Copy Data From User Space to Kernel Space

static int loopwput(queue_t *q, mblk_t *mp)
{
		struct loop *loop;
		int to;

		loop = (struct loop *)q->q_ptr;

		switch (mp->b_datap->db_type) {
			case M_IOCTL: {
				struct iocblk	*iocp;
				int		error=0;

				iocp = (struct iocblk *)mp->b_rptr;

				switch (iocp->ioc_cmd) {

	 				case LOOP_SET: {
						/*
						 * if this is a transparent ioctl return an error; 
						 * the complete solution is to convert the message 
						 * into an M_COPYIN message so that the data is
						 * ultimately copied from user space 
						 * to kernel space.
						 */

						if (iocp->ioc_count == TRANSPARENT) {
							error = EINVAL;
							goto iocnak;
						}

						/* fetch other minor device number */

						to = *(int *)mp->b_cont->b_rptr;

						/*
						 * Sanity check. ioc_count contains the amount
						 * of user supplied data which must equal the
						 * size of an int.
						 */

						if (iocp->ioc_count != sizeof(int)) {
							error = EINVAL;
							goto iocnak;
						}

						/* Is the minor device number in range? */

						if (to >= loop_cnt || to < 0) {
							error = ENXIO;
							goto iocnak;
						}

						/* Is the other device open? */

						if (!loop_loop[to].qptr) {
							error = ENXIO;
							goto iocnak;
						}

						/* Check if either dev is currently connected */

						if (loop->oqptr || loop_loop[to].oqptr) {
							error = EBUSY;
							goto iocnak;
						}

						/* Cross connect the streams through 
						 * the loopstruct 
						 */

						loop->oqptr = RD(loop_loop[to].qptr);
						loop_loop[to].oqptr = RD(q);

						/*
						 * Return successful ioctl. Set ioc_count
						 * to zero, since no data is returned.
						 */

						mp->b_datap->db_type = M_IOCACK;
						iocp->ioc_count = 0;
						qreply(q, mp);
						break;
					}

					default:
						error = EINVAL;
						iocnak:

						/*
						 * Bad ioctl. Setting ioc_error causes 
						 * the ioctl call to return that particular errno.
						 * By default, ioctl returns EINVAL on failure.
						 */

						mp->b_datap->db_type = M_IOCNAK;
						iocp->ioc_error = error;
						qreply(q, mp);
						break;
					}

					break;
				}

				case M_FLUSH: {

					if (*mp->b_rptr & FLUSHW) {
						flushq(q, FLUSHALL);	/* write */
						if (loop->oqptr)
							flushq(loop->oqptr, FLUSHALL);
						/* read on other side equals write on this side */
					}

					if (*mp->b_rptr & FLUSHR) {
						flushq(RD(q), FLUSHALL);
						if (loop->oqptr != NULL)
							flushq(WR(loop->oqptr), FLUSHALL);
					}

					switch(*mp->b_rptr) {

					case FLUSHW:
						*mp->b_rptr = FLUSHR;
						break;

					case FLUSHR:
						*mp->b_rptr = FLUSHW;
						break;

					}

					if (loop->oqptr != NULL)
						(void) putnext(loop->oqptr, mp);
					break;
				}

				default: /* If this Stream isn't connected, send 
							 * M_ERROR upstream.
							 */
					if (loop->oqptr == NULL) {
						freemsg(mp);
						(void) putnextctl1(RD(q), M_ERROR, ENXIO);
						break;
					}
					(void) putq(q, mp);

				}

		return (0);
}

Service procedures are required in this example on both the write side and read side for flow control (see Example 9–13). The write service procedure, loopwsrv, takes on the canonical form. The queue being written to is not downstream, but upstream (found through oqptr) on the other stream.

In this case, there is no read-side put procedure so the read service procedure, looprsrv, is not scheduled by an associated put procedure, as has been done previously. looprsrv is scheduled only by being back-enabled when its upstream flow control blockage is released. The purpose of the procedure is to re-enable the writer (loopwsrv) by using oqptr to find the related queue. loopwsrv cannot be directly back-enabled by STREAMS because there is no direct queue linkage between the two streams. Note that no message is queued to the read service procedure. Messages are kept on the write side so that flow control can propagate up to the stream head. qenable(9F) schedules the write-side service procedure of the other stream.

Example 9–13 Loop-Around Driver Flow Control

staticintloopwsrv (queue_t*q)
{
		mblk_t *mp;
		structloop *loop;
		loop = (structloop*) q->q_ptr;

		while ((mp = getq (q)) != NULL){
		  /* Check if we can put the message up 
		   * the other Stream read queue
		   */
		  if (mp->b_datap->db_type <= QPCTL && !canputnext (loop->oqptr)) {
				(void) putbq (q,mp);				/*read-side is blocked*/
				break;
		  }

		  /* send message to queue following other Stream read queue */

		  (void) putnext (loop->oqptr, mp);
		}
		return (0);
}

staticintlooprsrv (queue_t*q)
{
		/* Enter only when "backenabled" by flow control */
		structloop *loop;
		loop = (structloop*) q->q_ptr;
		if (loop->oqptr == NULL)
			return (0);

		/*manually enable write service procedure*/
		qenable (WR (loop->oqptr));
		return (0);
}

loopclose breaks the connection between the streams, as shown in Example 9–14. loopclose sends an M_HANGUP message up the connected stream to the stream head.

Example 9–14 Breaking Stream Connections for Loop-Around Device

/*ARGSUSED*/
static int loopclose (
		queue_t *q,
		int flag,
		cred_t *credp)
{
		struct loop *loop;

		loop = (struct loop *)q->q_ptr;
		loop->qptr = NULL;

		/*
		 * If we are connected to another stream, break the linkage, and 
		 * send a hangup message. The hangup message causes the stream head 
		 * to reject writes, allow the queued data to be read completely,
		 * and then return EOF on subsequent reads.
		 */

		if (loop->oqptr) {
			(void) putnextctl(loop->oqptr, M_HANGUP);
			((struct loop *)loop->oqptr->q_ptr)->oqptr = NULL;
			loop->oqptr = NULL;
		}

		qprocsoff(q);
		return (0);
}

An application using this driver would first open the clone device node created in the attach routine (/devices/pseudo/clone@0:loopx) two times to obtain two streams. The application can determine the minor numbers of the devices by using fstat(2). Next, it joins the two streams by using the STREAMS I_STR ioctl(2) (see streamio(7I)) to pass the LOOP_SET ioctl(2) with one of the stream's minor numbers as an argument to the other stream. Once this is completed, the data sent to one stream using write(2) or putmsg(2) can be retrieved from the other stream with read(2) or getmsg(2). The application also can interpose STREAMS modules between the stream heads and the driver using the I_PUSH ioctl(2).

Summarizing STREAMS Device Drivers

STREAMS device drivers are in many ways similar to non-STREAMS device drivers. The following points summarize the differences between STREAMS drivers and other drivers:

• Drivers must have attach(9E), probe(9E), and identify(9E) entry points to initialize the driver. The attach routine initializes the driver. Software drivers usually have little to initialize, because there is no hardware involved.

• Drivers have open(9E) and close(9E) routines.

• Most drivers have an interrupt handler routine. The driver developer is responsible for supplying an interrupt routine for the device's driver. In addition to hardware interrupts, the system also supports software interrupts. A software interrupt is generated by calling ddi_trigger_softintr(9F).

• All minor nodes are generated by ddi_create_minor_node(9F).

STREAMS device drivers also are similar to STREAMS modules. The following points summarize some of the differences between STREAMS modules and drivers.

• Messages that are not understood by the drivers should be freed.

• A driver must process all M_IOCTL messages. Otherwise, the stream head blocks for an M_IOCNAK, M_IOCACK, or until the timeout (potentially infinite) expires.

• If a driver does not understand an ioctl(2), an M_IOCNAK message must be sent upstream.

• The stream head locks up the stream when it receives an M_ERROR message, so driver developers should be careful when using the M_ERROR message.

• A hardware driver must provide an interrupt routine.

• Multithreaded drivers must protect their own data structures.

For more information on global driver issues and non-STREAMS drivers, see Writing Device Drivers.

Chapter 10 STREAMS Modules

This chapter provides specific examples of how modules work, including code samples.

• Module Overview

• Data Flow Control

• Design Guidelines

Module Overview

STREAMS modules process messages as they flow through the stream between an application and a character device driver. A STREAMS module is a pair of initialized queue structures and the specified kernel-level procedures that process data, status, and control information for the two queues. A stream can contain zero or more modules. Application processes push (stack) modules on a stream using the I_PUSH ioctl(2) and pop (unstack) them using the I_POP ioctl(2).

STREAMS Module Configuration

Like device drivers, STREAMS modules are dynamically linked and can be loaded into and unloaded from the running kernel.

The word module is used differently when talking about drivers. A device driver is a kernel-loadable module that provides the interface between a device and the Device Driver Interface, and is linked to the kernel when it is first invoked.

A loadable module must provide linkage information to the kernel in an initialized modlstrmod(9S) and three entry points: _init(9E), _info(9E), and _fini(9E).

STREAMS modules can be unloaded from the kernel when not pushed onto a stream. A STREAMS module can prevent itself from being unloaded by returning an error (selected from errno.h) from its _fini(9E) routine (EBUSY is a good choice).

Module Procedures

STREAMS module procedures (open, close, put, service) have already been described in the previous chapters. This section shows some examples and further describes attributes common to module put and service procedures.

A module's put procedure is called by the preceding module, driver, or stream head, and always before that queue's service procedure. The put procedure does any immediate processing (for example, high-priority messages), while the corresponding service procedure performs deferred processing.

The service procedure is used primarily for performing deferred processing, with a secondary task to implement flow control. Once the service procedure is enabled, it can start but not finish before running user-level code. The put and service procedures must not block because there is no thread synchronization being done.

Example 10–1 shows a STREAMS module read-side put procedure.

Example 10–1 Read-side put Procedure

static int
modrput (queue_t *q, mblk_t *mp)
{
		struct mod_prv *modptr;

		modptr = (struct mod_prv *) q->q_ptr;  /*state info*/

		if (mp->b_datap->db_type >= QPCTL){	/*proc pri msg*/
			putnext(q, mp); 						/* and pass it on */
			return (0);
		}

		switch(mp->b_datap->db_type) {
			case M_DATA:				/* can process message data */
				putq(q, mp);	/* queue msg for service procedure */
			return (0);

			case M_PROTO:			/* handle protocol control message */
					.
					.
					.

			default:
					putnext(q, mp);		
					return (0);
		}
}

The preceding code does the following:

• A pointer to a queue defining an instance of the module and a pointer to a message are passed to the put procedure.

• The put procedure performs a switch on the type of the message. For each message type, the put procedure either enqueues the message for further processing by the module service procedure, or passes the message to the next module in the stream.

• High-priority messages are typically processed immediately. Immediate processing is not required by the put procedure and the message is passed to the next module.

• Ordinary (or normal) messages are either queued or passed along the stream.

Example 10–2 shows a module write-side put procedure.

Example 10–2 Write-side put Procedure

static int
modwput (queue_t *q, mblk_t *mp)
{
 	struct mod_prv *modptr;

 modptr = (struct mod_prv *) q->q_ptr;		/*state info*/

 if (mp->b_datap->db_type >= QPCTL){	/* proc pri msg and pass it on */
			putnext(q, mp);
			return (0);
		}

 switch(mp->b_datap->db_type) {
		case M_DATA:			/* can process message data queue msg */
			putq(q, mp);		/* for service procedure or pass message */
								/* along with putnext(q,mp) */
			return (0);

		case M_PROTO:
				.
				.
				.

		case M_IOCTL:		/* if cmd in msg is recognized */
							/* process message and send reply back */
							/* else pass message downstream */

		default:
			putnext(q, mp);
			return (0);
	 }
}

The write-side put procedure, unlike the read side, can be passed M_IOCTL messages. The module must recognize and process the ioctl(2) command, or pass the message downstream if it does not recognize the command.

Example 10–3 shows a general scenario employed by the module's service procedure.

Example 10–3 STREAMS Module Service Procedure

static int
modrsrv (queue_t *q)
{
		mblk_t *mp;

		while ((mp = getq(q)) != NULL) {
			/* flow control check */
			if (!(mp->b_datap->db_type >= QPCTL) && !canputnext(q)) {	
				putbq(q, mp);						/* return message */
				return (0);
			}
			/* process the message */
				.
				.
				.
			putnext(q, mp); /* pass the result */
		}
		return (0);
}

The steps are:

1. Retrieve the first message from the queue using getq(9F).

2. If the message is high priority, process it immediately and pass it along the stream.

   Otherwise, the service procedure should use canputnext(9F) to determine if the next module or driver that enqueues messages is within acceptable flow-control limits. canputnext(9F) searches the stream for the next module, driver, or the stream head with a service procedure. When it finds one, it looks at the total message space currently allocated to the queue for messages. If the amount of space currently used at that queue reaches the high-water mark, canputnext(9F) returns false (zero). If the next queue with a service procedure is within acceptable flow-control limits, canputnext(9F) returns true (nonzero).

3. If canputnext(9F) returns false, the service procedure returns the message to its own queue with putbq(9F). The service procedure can do no further processing at this time, and it returns to the caller.

   If canputnext(9F) returns true, the service procedure completes any processing of the message. This can involve retrieving more messages from the queue, allocating and deallocating header and trailer information, and performing control functions for the module.

4. When the service procedure is finished processing the message, it calls putnext(9F) to pass the resulting message to the next queue.

These steps are repeated until getq(9F) returns NULL (the queue is empty) or canputnext(9F) returns false.

Filter Module Example

The module shown next, crmod in Example 10–4, is an asymmetric filter. On the write side, a newline is changed to a carriage return followed by a newline. On the read side, no conversion is done.

Example 10–4 Filter Module

/* Simple filter
 * converts newline -> carriage return, newline
 */
#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

static struct module_info minfo =
	{ 0x09, "crmod", 0, INFPSZ, 512, 128 };

static int modopen (queue_t*, dev_t*, int, int, cred_t*);
static int modrput (queue_t*, mblk_t*);
static int modwput (queue_t*, mblk_t*);
static int modwsrv (queue_t*);
static int modclose (queue_t*, int, cred_t*);

static struct qinit rinit = {
	modrput, NULL, modopen, modclose, NULL, &minfo, NULL};

static struct qinit winit = {
	modwput, modwsrv, NULL, NULL, NULL, &minfo, NULL};

struct streamtab crmdinfo={ &rinit, &winit, NULL, NULL};

stropts.h includes definitions of flush message options common to user applications. modrput is like modput from the null module.

In contrast to the null module example, a single module_info structure is shared by the read side and write side. The module_info includes the flow control high-water and low-water marks (512 and 128) for the write queue. (Though the same module_info is used on the read queue side, the read side has no service procedure so flow control is not used.) The qinit contains the service procedure pointer.

The write-side put procedure, the beginning of the service procedure, and an example of flushing a queue are shown in Example 10–5.

Example 10–5 Flushing a Queue

static int
modwput(queue_t *q, mblk_t *mp)
{
	if (mp->b_datap->db_type >= QPCTL && mp->b_datap->db_type != M_FLUSH)
			putnext(q, mp);
	else
			putq(q, mp);				 /* Put it on the queue */
	return (0);
}
static int 
modwsrv(queue_t *q)
{
	mblk_t *mp;

	while ((mp = getq(q)) != NULL) {
			switch (mp->b_datap->db_type) {
				default:
					if (canputnext(q)) {
							putnext(q, mp);
							break;
			 		} else {
							putbq(q, mp);
							return (0);
					 }

				case M_FLUSH:
				    if (*mp->b_rptr & FLUSHW)
						    flushq(q, FLUSHDATA);
				    putnext(q, mp);
				    break;

				case M_DATA: {
					mblk_t *nbp = NULL;
					mblk_t *next;
					if (!canputnext(q)) {
						putbq(q, mp);
						return (0);
					}
			/* Filter data, appending to queue */
			for (; mp != NULL; mp = next) {
					while (mp->b_rptr < mp->b_wptr) {
							if (*mp->b_rptr == '\n')
								if (!bappend(&nbp, '\r'))
										goto push;
							if (!bappend(&nbp, *mp->b_rptr))
								goto push;
							mp->b_rptr++;
							continue;
					push:
							if (nbp)
								putnext(q, nbp);
							nbp = NULL;
							if (!canputnext(q)) {
								if (mp->b_rptr>=mp->b_wptr){
										next = mp->b_cont;
										freeb(mp);
										mp=next;
								}
								if (mp)
										putbq(q, mp);
								return (0);
							}
					} /* while */
					next = mp->b_cont;
					freeb(mp);
					if (nbp)
						putnext(q, nbp);
				} /* for */
			}
		} /* switch */
	}
	return (0);
}					

modwsrv() is the write service procedure. It takes a single argument, which is a pointer to the write queue. modwsrv() switches on the message type, M_FLUSH or M_DATA. modwsrv() processes only one high-priority message, M_FLUSH. No other high-priority messages should reach modwsrv. High-priority messages other than type M_FLUSH use putnext(9F) to avoid scheduling. The others are queued for the service procedure. An M_FLUSH message is a request to remove messages on one or both queues. It can be processed in the put or service procedure.

For an M_FLUSH message, modwsrv() checks the first data byte. If FLUSHW is set, the write queue is flushed by flushq(9F), which takes two arguments, the queue pointer and a flag. The flag indicates what should be flushed, data messages (FLUSHDATA) or everything (FLUSHALL). Data includes M_DATA, M_DELAY, M_PROTO, and M_PCPROTO messages. The choice of what types of messages to flush is specific to each module.

Ordinary messages are returned to the queue if canputnext(9F) returns false, indicating the downstream path is blocked.

The differences in M_DATA processing between this and the example in Message Allocation and Freeing relate to the manner in which the new messages are forwarded and flow controlled. For the purpose of demonstrating alternative means of processing messages, this version creates individual new messages rather than a single message containing multiple message blocks. When a new message block is full, it is immediately forwarded with putnext(9F) rather than being linked into a single large message. This alternative is not desirable because message boundaries are altered, and because of the additional overhead of handling and scheduling multiple messages.

When the filter processing is performed (following push), flow control is checked (with canputnext(9F)) after each new message is forwarded. This is necessary because there is no provision to hold the new message until the queue becomes unblocked. If the downstream path is blocked, the remaining part of the original message is returned to the queue. Otherwise, processing continues.

Data Flow Control

To support the STREAMS flow control mechanism, modules that use service procedures must invoke canputnext(9F) before calling putnext(9F), and use appropriate values for the high-water and low-water marks. If your module has a service procedure, you manage the flow control. If you don't have a service procedure, then there is no need to do anything.

The queue hiwat and lowat values limit the amount of data that can be placed on a queue. These limits prevent depletion of buffers in the buffer pool. Flow control is advisory in nature and can be bypassed. It is managed by high-water and low-water marks and regulated by the utility routines getq(9F), putq(9F), putbq(9F), insq(9F), rmvq(9F), and canputnext(9F).

The following scenario takes place normally in flow control:

A driver sends data to a module using putnext(9F), and the module's put procedure queues data using putq(9F). Calling putq(9F) enables the service procedure and executes it at some indeterminate time in the future. When the service procedure runs, it retrieves the data by calling getq(9F).

If the module cannot process data at the rate at which the driver is sending the data, the following happens:

When the message is queued, putq(9F) increments the value of q_count by the size of the message and compares the result to the module's high-water limit (q_hiwat) value for the queue. If the count reaches q_hiwat, putq(9F) sets the internal FULL indicator for the queue. This causes messages from upstream in the case of a write-side queue or downstream in the case of a read-side queue to be halted (canputnext(9F) returns FALSE) until the queue count drops below q_lowat. getq(9F) decrements the queue count. If the resulting count is below q_lowat, getq(9F) back-enables and causes the service procedure to be called for any blocked queue. (Flow control does not prevent reaching q_hiwat on a queue. Flow control can exceed its maximum value before canputnext(9F) detects QFULL and flow is stopped.)

The next example show a line discipline module's flow control. Example 10–6 shows a read-side line discipline module and a write-side line discipline module. Note that the read side is the same as the write side but without the M_IOCTL processing.

Example 10–6 Read-side Line Discipline Module

/* read side line discipline module flow control */
static mblk_t *read_canon(mblk_t *);

static int
ld_read_srv(
	queue_t *q)							/* pointer to read queue */
{
	mblk_t *mp;							/* original message */
	mblk_t *bp;							/* canonicalized message */

	while ((mp = getq(q)) != NULL) {
			switch (mp->b_datap->db_type) { /* type of msg */
			case M_DATA:	 /* data message */
				if (canputnext(q)) {
						bp = read_canon(mp);
						putnext(q, bp);
				} else {
						putbq(q, mp); /* put messagebackinqueue */
						return (0);
				}
				break;

			default:
				if (mp->b_datap->db_type >= QPCTL)
						putnext(q, mp); 		/* high-priority message */
				else { /* ordinary message */
						if (canputnext(q))
								 putnext(q, mp);
						else {
								 putbq(q, mp);
								 return (0);
						}
				}
				break;
			}
	}
return (0);
}

/* write side line discipline module flow control */
static int
ld_write_srv(
	queue_t *q)								/* pointer to write queue */
{
	mblk_t *mp;								/* original message */
	mblk_t *bp;								/* canonicalized message */

	while ((mp = getq(q)) != NULL) {
			switch (mp->b_datap->db_type) { /* type of msg */
			case M_DATA:			 /* data message */
				if (canputnext(q)) {
						bp = write_canon(mp);
						putnext(q, bp);
				} else {
						putbq(q, mp);
						return (0);
				}
				break;

			case M_IOCTL:
				ld_ioctl(q, mp);
				break;

			default:
				if (mp->b_datap->db_type >= QPCTL)
						putnext(q, mp);		/* high priority message */
				else { 						/* ordinary message */
						if (canputnext(q))		
								putnext(q, mp);
						else {
								putbq(q, mp);
								return (0);
						}
				}
				break;
			}
	}
return (0);
}

Design Guidelines

Module developers should follow these guidelines:

• If a module cannot process the message types, the message types must be passed to the next module.

• The module that acts on an M_IOCTL message should send an M_IOCACK or M_IOCNAK message in response to the ioctl(2). If the module cannot process the ioctl(2), it should pass the M_IOCTL message to the next module.

• Modules should not pertain to any particular driver but should be compatible with all drivers.

• In general, modules should not require the data in an M_DATA message to follow a particular format, such as a specific alignment. This means modules can be arbitrarily pushed on top of each other in a sensible fashion. Not following this rule can limit module usability.

• Filter modules pushed between a service user and a service provider should not alter the contents of the M_PROTO or M_PCPROTO block in messages. The contents of the data blocks can be changed, but the message boundaries must be preserved.

htonl(3N) and ntohl(3N)

The htonl(3N) and ntohl(3N) conversion routines follow the XNS5 publications. The functions continue to convert 32-bit quantities between network byte order and host byte order.

Chapter 11 Configuring STREAMS Drivers and Modules

This chapter contains information about configuring STREAMS drivers and modules into the Solaris operating environment. It describes how to configure a driver and a module for the STREAMS framework only. For more in-depth information on the general configuration mechanism, see Writing Device Drivers.

This chapter has the following organization:

• Kernel Data Structures

• STREAMS Driver Entry Points

• Tunable Parameters

• STREAMS Administrative Driver

Kernel Data Structures

The following sections contain descriptions of the pointer relationships maintained by the kernel and the various data structures used in STREAMS drivers. When the kernel accesses a driver, it uses a sequence of pointers in various data structures. It looks first at the data structure relationship, and then the entry point interface for loading the driver into the kernel and accessing the driver from the application level.

The order of data structures the kernel uses to get to a driver is as follows:

modlinkage(9S)

Contains the revision number and a list of drivers to dynamically load. It is used by mod_install in the _init routine to load the module into the kernel. Points to a modldrv(9S) or modlstrmod(9S).

modldrv(9S)

Contains information about the driver being loaded and points to the devops structure.

modlstrmod(9S)

Points to an fmodsw(9S) structure (which points to a streamtab(9S)). Only used by STREAMS modules.

dev_ops(9S)

Contains list of entry points for a driver, such as identify, attach, and info. Also points to a cb_ops(9S) structure.

cb_ops(9S)

Points to list of threadable entry points to driver, like open, close, read, write, ioctl. Also points to the streamtab.

streamtab(9S)

Points to the read and write queue init structures.

qinit(9S)

Points to the entry points of the STREAMS portion of the driver, such as put, srv, open, close, as well as the mod_info structure. These entry points only process messages.

Each STREAMS driver or module contains the linkage connections for the various data structures which is a list of pointers to dev_ops(9S) structures. In each dev_ops(9S) structure is a pointer to the cb_ops(9S) structure. In the cb_ops(9S) structure contains a pointer to the streamtab struct. If the driver is not a STREAMS driver, streamtab is NULL. If the driver is a STREAMS driver, streamtab contains initialization routines for the driver.

modlinkage

The definition of modlinkage(9S) is:

struct modlinkage {
   int      ml_rev;           /* rev of loadable modules system */
   void     *ml_linkage[4];   /* NULL terminated list of linkage 
                               * structures */
};

modldrv

The definition of modldrv(9S) is:

struct modldrv {
		struct mod_ops   *drv_modops;
		char             *drv_linkinfo;
		struct dev_ops   *drv_dev_ops;
};

modlstrmod

The definition of modlstrmod(9S) is below. It does not point to dev_ops(9S) structures because modules can only be pushed onto an existing stream.

struct modlstrmod {
		struct mod_ops      *strmod_modops;
		char                *strmod_linkinfo;
		struct fmodsw       *strmod_fmodsw;
};

dev_ops

The dev_ops(9S) structure represents a specific class or type of device. Each dev_ops(9S) structure represents a unique device to the operating system. Each device has its own dev_ops(9S) structure, which in turn contains a cb_ops(9S) structure.

struct dev_ops  {
  int       devo_rev;                 /* Driver build version	*/
  int       devo_refcnt;              /* device reference count	*/
  int       (*devo_getinfo)(dev_info_t *dip, ddi_info_cmd_t infocmd, 
									 void *arg, void **result);
  int       (*devo_identify)(dev_info_t *dip);
  int       (*devo_probe)(dev_info_t *dip);
  int       (*devo_attach)(dev_info_t *dip, ddi_attach_cmd_t cmd);
  int       (*devo_detach)(dev_info_t *dip, ddi_detach_cmd_t cmd);
  int       (*devo_reset)(dev_info_t *dip, ddi_reset_cmd_t cmd);
  struct cb_ops      *devo_cb_ops;    /* cb_ops ptr for leaf driver*/
  struct bus_ops     *devo_bus_ops;   /* ptr for nexus drivers */
};

cb_ops

The cb_ops(9S) structure is the SunOS 5 version of the cdevsw and bdevsw tables of previous versions of UNIX System V. It contains character and block device information and the driver entry points for non-STREAMS drivers.

struct cb_ops  {
		int		*cb_open)(dev_t *devp, int flag, int otyp, 
						cred_t *credp);
		int		(*cb_close)(dev_t dev, int flag, int otyp, 
						cred_t *credp);
		int		(*cb_strategy)(struct buf *bp);
		int		(*cb_print)(dev_t dev, char *str);
		int		(*cb_dump)(dev_t dev, caddr_t addr,daddr_t blkno, 
						int nblk);
		int		(*cb_read)(dev_t dev, struct uio *uiop, cred_t *credp);
		int		(*cb_write)(dev_t dev, struct uio *uiop, cred_t *credp);
		int		(*cb_ioctl)(dev_t dev, int cmd, int arg, int mode,
						cred_t *credp, int *rvalp);
		int		(*cb_devmap)(dev_t dev, dev_info_t *dip, 
						ddi_devmap_data_t *dvdp, ddi_devmap_cmd_t cmd, 
						off_t offset, unsigned int len, unsigned int prot, 
						cred_t *credp);
		int		(*cb_mmap)(dev_t dev, off_t off, int prot);
		int		(*cb_segmap)(dev_t dev, off_t off, struct as *asp, 
						caddr_t *addrp, off_t len, unsigned int prot, 
						unsigned int maxprot, unsigned int flags, 
						cred_t *credp);
		int		(*cb_chpoll)(dev_t dev, short events, int anyyet, 
						short *reventsp, struct pollhead **phpp);
		int		(*cb_prop_op)(dev_t dev, dev_info_t *dip, 
						ddi_prop_op_t prop_op, int mod_flags, char *name, 
						caddr_t valuep, int *length);

     struct streamtab *cb_str;		/* streams information */

		/*
		 * The cb_flag fields are here to tell the system a bit about 
		 * the device. The bit definitions are in <sys/conf.h>.
		 */
     int		cb_flag;					/* driver compatibility flag */
};

streamtab

The streamtab(9S) structure contains pointers to the structures that hold the routines that actually initialize the reading and writing for a module.

If streamtab is NULL, there are no STREAMS routines and the entire driver is treated as though it was a non-STREAMS driver. The streamtab(9S) indirectly identifies the appropriate open, close, put, service, and administration routines. These driver and module routines should generally be declared static.

struct streamtab {
		struct qinit     *st_rdinit;      /* defines read queue */
		struct qinit     *st_wrinit;      /* defines write queue */
		struct qinit     *st_muxrinit;    /* for multiplexing */
		struct qinit     *st_muxwinit;    /* drivers only */
};

qinit

The qinit(9S) structure (also shown in Appendix A) contains pointers to the STREAMS entry points. These routines are called by the module-loading code in the kernel.

struct qinit {
 	int         (*qi_putp)();            /* put procedure */
 	int         (*qi_srvp)();            /* service procedure */
 	int         (*qi_qopen)();           /* called on each open or push*/
 	int         (*qi_qclose)();          /* called on last close or pop*/
 	int         (*qi_qadmin)();          /* reserved for future use */
 	struct module_info     *qi_minfo;    /* info struct */
 	struct module_stat     *qi_mstat;    /* stats struct (opt)*/
};

STREAMS Driver Entry Points

As described in Chapter 9, STREAMS Drivers and as seen in the previous data structures, there are four STREAMS driver entry points:

load kernel module

_init(9E), _fini(9E), _info(9E)

dev_ops

identify(9E), attach(9E), getinfo(9E)

cb_ops

open(9E), close(9E), read(9E), write(9E), ioctl(9E)

streamtab

put(9E), srv(9E)

pts Example

The following real example was taken from the Solaris operating environment. The driver pts(7D) is the pseudo terminal slave driver.

Example 11–1 Stream Pseudo Terminal Module

/*
 * Slave Stream Pseudo Terminal Module
 */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/stat.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include <sys/cmn_err.h>
#include <sys/modctl.h>
#include <sys/conf.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

static int ptsopen (queue_t*, dev_t*, int, int, cred_t);
static int ptsclose (queue_t*, int, cred_t*);
static int ptswput (queue_t*, mblk_t*);
static int ptsrsrv (queue_t*);
static int ptswsrv (queue_t*);

static int pts_devinfo(dev_info_t *dip, ddi_info_cmd_t infocmd,
      void *arg,void **result);

static struct module_info pts_info = {
		0xface,
		"pts",
		0,
		512,
		512,
		128
};

static struct qinit ptsrint = {
		NULL,
		ptsrsrv,
		ptsopen,
		ptsclose,
		NULL,
		&pts_info,
		NULL
};

static struct qinit ptswint = {
		ptswput,
		ptswsrv,
		NULL,
		NULL,
		NULL,
		&pts_info,
		NULL
};

static struct streamtab ptsinfo = {
		&ptsrint,
		&ptswint,
		NULL,
		NULL
};

static int pts_identify(dev_info_t *devi);
static int pts_attach(dev_info_t *devi, ddi_attach_cmd_t cmd);
static int pts_detach(dev_info_t *devi, ddi_detach_cmd_t cmd);
static dev_info_t *pts_dip;				/* private copy of devinfo ptr */

extern kmutex_t pt_lock;
extern pt_cnt;
static struct cb_ops cb_pts_ops = {
   nulldev,       /* cb_open */ 
   nulldev,       /* cb_close */ 
   nodev,         /* cb_strategy */
   nodev,         /* cb_print */
   nodev,         /* cb_dump */
   nodev,         /* cb_read */
   nodev,         /* cb_write */
   nodev,         /* cb_ioctl */
   nodev,         /* cb_devmap */
   nodev,         /* cb_mmap */
   nodev,         /* cb_segmap */
   nochpoll,      /* cb_chpoll */
   ddi_prop_op,   /* cb_prop_op */
   &ptsinfo,      /* cb_stream */
   D_MP           /* cb_flag */
};

static struct dev_ops pts_ops = {
   DEVO_REV,      /* devo_rev */
   0,             /* devo_refcnt */   
   pts_devinfo,   /* devo_getinfo */ 
   pts_identify,  /* devo_identify */
   nulldev,       /* devo_probe */
   pts_attach,    /* devo_attach */
   pts_detach,    /* devo_detach */
   nodev,         /* devo_reset */
   &cb_pts_ops,   /* devo_cb_ops */
   (struct bus_ops*) NULL   /* devo_bus_ops */
};

/*
 * Module linkage information for the kernel.
 */

static struct modldrv modldrv = {
			&mod_driverops, 		/* Type of module: a pseudo driver */
			"Slave Stream Pseudo Terminal driver'pts'",
			&pts_ops,				/* driver ops */
};

static struct modlinkage modlinkage = {
		MODREV_1,
		(void *)&modldrv,
		NULL
};

int
_init(void)
{
		return (mod_install(&modlinkage));
}

int
_fini(void)
{
		return (mod_remove(&modlinkage));
}

int
_info(struct modinfo *modinfop)
{
		return (mod_info(&modlinkage, modinfop));
}

static int
pts_identify(dev_info_t *devi)
{
		if (strcmp(ddi_get_name(devi), "pts") == 0)
			return (DDI_IDENTIFIED);
		else
			return (DDI_NOT_IDENTIFIED);
}

static int
pts_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
{
		int i;
		char name[5];

		if (cmd != DDI_ATTACH)
			return (DDI_FAILURE);
	
		for (i = 0; i < pt_cnt; i++) {
			(void) sprintf(name, "%d", i);
			if (ddi_create_minor_node(devi, name, S_IFCHR, i, NULL, 0) 
						== DDI_FAILURE) {
				ddi_remove_minor_node(devi, NULL);
				return (DDI_FAILURE);
			}
		}
		return (DDI_SUCCESS);
}

static int
pts_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
{
		ddi_remove_minor_node(devi, NULL);
		return (DDI_SUCCESS);
}

static int
pts_devinfo (dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
					 void **result)
{
		int error;

		switch (infocmd)   {
			case DDI_INFO_DEVT2DEVINFO:
				if (pts_dip == NULL) {
					error = DDI_FAILURE;
				} else {
					*result = (void *) pts_dip;
					error = DDI_SUCCESS;
				}
				break;
			case DDI_INFO_DEVT2INSTANCE:
				*result = (void *) 0;
				error = DDI_SUCCESS;
				break;
			default:
				error = DDI_FAILURE;
		}
		return (error);
}

/* the open, close, wput, rsrv, and wsrv routines are presented
 * here solely for the sake of showing how they interact with the
 * configuration data structures and routines. Therefore, the 
 * bulk of their code is not included.
 */
static int
ptsopen(rqp, devp, oflag, sflag, credp)
		queue_t *rqp; 	/* pointer to the read side queue */
		dev_t   *devp;			/* pointer to stream tail's dev */
		int	 oflag; 			/* the user open(2) supplied flags */
		int 	sflag; 			/* open state flag */
		cred_t  *credp;		/* credentials */
{
		qprocson(rqp);
		return (0);
}

static int
ptsclose(rqp, flag, credp)
		queue_t		*rqp;
		int			flag;
		cred_t		*credp;
{
		qprocsoff(rqp);
		return (0);
}

static int
ptswput(qp, mp)
		queue_t		*qp;
		mblk_t		*mp;
{
		return (0);
}

static int
ptsrsrv(qp)
		queue_t *qp;
{
		return (0);
}

static int
ptswsrv(qp)
		queue_t 	*qp;
{
		return (0);
}

STREAMS Module Configuration

The following example shows the structures you need if you are working with a module instead of a driver. Notice that a modlstrmod(9S) is used in modlinkage(9S), and fmodsw(9S) points to streamtab(9S) instead of going through dev_ops(9S).

Example 11–2 Module Structures

extern struct streamtab pteminfo;

static struct fmodsw fsw = {
		"ptem",
		&pteminfo,
		D_NEW | D_MP
};

/*
 * Module linkage information for the kernel.
 */
extern struct mod_ops mod_strmodops;

static struct modlstrmod modlstrmod = {
		&mod_strmodops,
		"pty hardware emulator",
		&fsw
};

static struct modlinkage modlinkage = {
		MODREV_1, 
		(void *)&modlstrmod,
		NULL
};

Compilation

Below are some compile, assemble, and link lines for an example driver with two C source files (example_one.c and example_two.c) and an assembly language source file (example_asm.s).

cc -D_KERNEL -c example_one.c
cc -D_KERNEL -c example_two.c
as -P -D_ASM -D_KERNEL -I. -o example_asm.o example_asm.s
ld -r -o example example_one.o example_two.o example_asm.o

Kernel Loading

See Writing Device Drivers for more information on the sequence of installing and loading device drivers. The basic procedure is to copy your driver to /kernel/drv and your module to /kernel/strmod. For drivers run add_drv(1M).

The autoload facility looks for modules to reside in /kernel/strmod. If the object resides elsewhere the module will not be loaded.

Checking the Module Type

Below is sample code that enables a driver to determine if it is running as a regular driver, a module, or a cloneable driver. The open routine returns sflag, which is checked.

	if (sflag == MODOPEN)
 			/* then the module is being pushed */
		else if (sflag == CLONEOPEN)
 			/* then its being opened as a clonable driver */
 	else
 			/* its being opened as a regular driver */

Tunable Parameters

Certain system parameters referred to by STREAMS are configurable when building a new operating system (see the file /etc/system and the SunOS User's Guide to System Administration for further details). These parameters are:

nstrpush

Maximum number (should be at least 8) of modules that can be pushed onto a single stream.

strmsgsz

Maximum number of bytes of information that a single system call can pass to a stream to be placed into the data part of a message (in M_DATA blocks). Any write(2) exceeding this size is broken into multiple messages. A putmsg(2) with a data part exceeding this size fails with ERANGE. If STRMSGSZ is set to 0, the number of bytes passed to a stream is infinite.

strctlsz

Maximum number of bytes of information that a single system call can pass to a stream to be placed into the control part of a message (in an M_PROTO or M_PCPROTO block). A putmsg(2) with a control part that exceeds this size fails with ERANGE.

STREAMS Administrative Driver

The autopush(1M) facility configures the list of modules for a STREAMS device. It automatically pushes a prespecified list (/etc/iu.ap) of modules onto the stream when the STREAMS device is opened and the device is not already open.

The STREAMS Administrative Driver (SAD) (see the sad(7D) man page) provides an interface to the autopush mechanism. System administrators can open the SAD driver and set or get autopush information on other drivers. The SAD driver caches the list of modules to push for each driver. When the driver is opened the stream head checks the SAD's cache to determine if the device is configured to have modules pushed automatically. If an entry is found, the modules are pushed. If the device has been opened but not closed, another open does not cause the list of the prespecified modules to be pushed again.

Three options configure the module list:

• Configure for each minor device (that is, a specific major and minor device number)

• Configure for a range of minor devices within a major device

• Configure for all minor devices within a major device

In addition, when configuring the module list, an optional anchor can be placed within the module list. See STREAMS Anchors for more information.

When the module list is cleared, a range of minor devices has to be cleared as a range and not in parts.

Application Interface

The SAD driver is accessed through the /dev/sad/admin or /dev/sad/user node. After the device is initialized, a program can perform any autopush configuration. The program should open the SAD driver, read a configuration file to find out what modules need to be configured for which devices, format the information into strapush structures, and make the SAD_SAP ioctl(2) calls. See the sad(7D) man page for more information.

All autopush operations are performed through SAD_SAP ioctl(2) commands to set or get autopush information. Only the root user can set autopush information, but any user can get the autopush information for a device.

The SAD_SAP ioctl is a form of ioctl(fd, cmd, arg), where fd is the file descriptor of the SAD driver, cmd is either SAD_SAP (set autopush information) or SAD_GAP (get autopush information), and arg is a pointer to the structure strapush.

The strapush structure is shown in the following example:

Example 11–3 strapush Structure

/*
 * maximum number of modules that can be pushed on a
 * stream using the autopush feature should be no greater
 * than nstrpush
 */
#define MAXAPUSH 8

/* autopush information common to user and kernel */

struct apcommon {
   uint     apc_cmd;          /* command - see below */
   major_t  apc_major;        /* major device number */
   minor_t  apc_minor;        /* minor device number */
   minor_t  apc_lastminor;    /* last minor dev # for range */
   uint     apc_npush;        /* number of modules to push */
};

/* ap_cmd - various options of autopush */
#define SAP_CLEAR       0 /* remove configuration list */
#define SAP_ONE         1 /* configure one minor device */
#define SAP_RANGE       2 /* config range of minor devices */
#define SAP_ALL         3 /* configure all minor devices */

/* format of autopush ioctls */
struct strapush {
		struct apcommon sap_common;
		char sap_list[MAXAPUSH] [FMNAMESZ + 1]; /* module list */
};

#define sap_cmd           sap_common.apc_cmd
#define sap_major         sap_common.apc_major
#define sap_minor         sap_common.apc_minor
#define sap_lastminor     sap_common.apc_lastminor
#define sap_npush         sap_common.apc_npush

A device is identified by its major device number, sap_major. The SAD_SAP ioctl(2) has the following options:

SAP_ONE

Configures a single minor device, sap_minor, of a driver

SAP_RANGE

Configures a range of minor devices from sap_minor to sap_lastminor, inclusive

SAP_ALL

Configures all minor devices of a device

SAP_CLEAR

Clears the previous settings by removing the entry with the matching sap_major and sap_minor fields

The list of modules is specified as a list of module names in sap_list. MAXAPUSH defines the maximum number of modules to push automatically.

A user can query the current configuration status of a given major/minor device by issuing the SAD_GAP ioctl(2) with sap_major and sap_minor values of the device set. On successful return from this system call, the strapush structure is filled in with the corresponding information for the device. The maximum number of entries that the SAD driver can cache is determined by the tunable parameter NAUTOPUSH which is found in the SAD driver's master file.

The following is an example of an autopush configuration file in /etc/iu.ap:

#	major      minor         lastminor      modules

	wc         0             0              ldterm ttcompat
	zs         0             1              ldterm ttcompat
	ptsl       0             15             ldterm ttcompat

The first line configures a single minor device whose major name is wc. Minor numbers start and end at 0, creating only one minor number. The modules automatically pushed are ldterm and ttcompat. The second line configures the zs driver whose minor device numbers are 0 and 1, and automatically pushes the same modules. The last line configures the ptsl driver whose minor device numbers are from 0 to 15, and automatically pushes the same modules.

STREAMS Anchors

An anchor is a lock that prevents the removal of a STREAMS module with an I_POP call. You place an anchor in a stream on the module you want to lock. All modules at or below the anchor are locked, and can only be popped by a privileged process.

An anchor is a lock that prevents the removal of a STREAMS module with an I_POP call. You place an anchor in a stream on the module you want to lock. All modules at or below the anchor are locked, and can only be popped by a privileged process.

Anchors and Data Flow

Hardening Information. Anchors do not affect the flow of data in the stream or any other properties of the stream other than to lock down its plumbing. Any process can place an anchor on a stream, but once placed, it can only be removed by a privileged process.

An anchor is a per-stream entity; that is, there is exactly one per stream, and the anchor is moved upstream or downstream as needed. When a stream is created, the anchor is conceptually at the driver and therefore has no effect on the stream. By issuing the I_ANCHOR ioctl on a stream, a process places the anchor at the STREAMS module directly below the stream head. This means that a process can move an existing anchor upstream by pushing additional STREAMS modules and calling I_ANCHOR again.

Although anchors conceptually exist at a specific location in the stream, they are not a data processing element and therefore do not physically exist in the stream (for example, you will not find them parsing q_next pointers.) This means that anchors will not appear in ioctls such as I_LOOK, and they are not included in the module count on the stream.

To remove an anchor, a process pops the module at which the anchor was placed. The anchor will only allow a privileged process to pop modules at or below it, which provides security. Once an anchor has been removed, the anchor is not reset to its previous location in the stream, but rather positioned at the STREAMS driver again. When an unprivileged process attempts to pop an anchored module, the ioctl returns with EPERM.

The I_ANCHOR ioctl is processed completely in the stream head, and is never sent downstream. If a module or driver sends an I_ANCHOR to the stream head, the anchor is silently discarded.

Using Anchors

An anchor can be placed on a STREAMS module by adding an [anchor] flag to an autopush configuration file or by directly calling the I_ANCHOR ioctl.

For example, this configuration file specifies that autopush should place an anchor between foo and babar in the bb stream:

#	major      minor       lastminor     modules

	aa         0           0             foo babar
	bb         0           1             foo [anchor] babar
	bb         131072      131073        foo [anchor] babar

The following two examples illustrate the use of anchors in a client/server setting in which file descriptors are being passed. They call the I_ANCHOR ioctl directly.

In this example, the server program, fd_server.c, opens a stream, pushes modules on to it, and places an anchor on rlmod. The client program, fd_client.c attempts to pop modules, but can only pop rlmod or any modules below it if the client is run as root. That is, if the client is run as non-root, the I_POP fails.

This example also shows that once the module with the anchor on it is popped by the privileged root process, the anchor is destroyed (technically, it is moved back to the driver, where it has no effect). Subsequent attempts by the client to pop modules will succeed, even if the client is run as non-root.

Finally, this example also illustrates the effect of passing file descriptors, rather than copying modules or the stream as a whole. Specifically, because the stream is not duplicated, all instances of the client operate on the same stream. In this case, running the client repeatedly causes it to work down the list of modules, popping each one off in turn, until all modules have been removed from the stream.

Example 11–4 STREAMS Anchors fd_server.c

#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <fcntl.h>
#include <stdio.h>
#include <stropts.h>
#include <sys/conf.h>
#include <sys/stat.h>
#include <unistd.h>
#include <stdlib.h>

#define	SPIPE_PATH	"/tmp/mypipe"

int
main(void)
{
	int 			pipefd[2];
	struct strrecvfd	strrecvfd;
	int			streamfd;

	/*
	 * Open a stream to hand back to the client.  Since this
	 * is just an example, we don't really care what we open;
	 * make a rlmod<->udp<->ip stream.  Stick an anchor above
	 * rlmod so the client cannot I_POP rlmod unless it's root.
	 */
	streamfd = open("/dev/udp", O_RDWR);
	if (streamfd == -1) {
		perror("open");
		return (EXIT_FAILURE);
	}

	if (ioctl(streamfd, I_PUSH, "rlmod") == -1) {
		perror("ioctl (I_PUSH) rlmod");
		return (EXIT_FAILURE);
	}

	if (ioctl(streamfd, I_ANCHOR, 0) == -1) {
		perror("ioctl (I_ANCHOR)");
		return (EXIT_FAILURE);
	}

	/*
	 * Open ourselves for business by making a mounted stream.
	 */
	if (pipe(pipefd) == -1) {
		perror("pipe");
		return (EXIT_FAILURE);
	}

	if (ioctl(pipefd[1], I_PUSH, "connld") == -1) {
		perror("ioctl (I_PUSH) connld");
		return (EXIT_FAILURE);
	}

	(void) umask(0);
	(void) close(creat(SPIPE_PATH, 0666));

	if (fattach(pipefd[1], SPIPE_PATH) == -1) {
		perror("fattach");
		return (EXIT_FAILURE);
	}

	/*
	 * Accept clients (iterative server)
	 */
	for (;;) {

		if (ioctl(pipefd[0], I_RECVFD, &strrecvfd) == -1) {
			perror("ioctl (I_RECVFD)");
			return (EXIT_FAILURE);
		}

		/*
		 * Send the STREAMS descriptor back to the client.
		 */
		if (ioctl(strrecvfd.fd, I_SENDFD, streamfd) == -1) {
			perror("ioctl (I_SENDFD)");
			return (EXIT_FAILURE);
		}
	}
}

Example 11–5 STREAMS Anchors fd_client.c

#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <fcntl.h>
#include <stdio.h>
#include <stropts.h>
#include <sys/conf.h>
#include <unistd.h>
#include <stdlib.h>

#define	SPIPE_PATH	"/tmp/mypipe"

int
main(void)
{
	int 			serverfd;
	struct strrecvfd	strrecvfd;

	/*
	 * Open a connection to the server.
	 */
	serverfd = open(SPIPE_PATH, O_RDWR);
	if (serverfd == -1) {
		perror("open");
		return (EXIT_FAILURE);
	}

	/*
	 * Receive the STREAMS descriptor from the server.
	 */
	if (ioctl(serverfd, I_RECVFD, &strrecvfd) == -1) {
		perror("ioctl (I_RECVFD)");
		return (EXIT_FAILURE);
	}

	(void) printf("received the STREAMS descriptor; attempting to pop "
	    "the top module\n");

	/*
	 * Try to remove the top module from the stream.
	 */
	if (ioctl(strrecvfd.fd, I_POP, 0) == -1)
		perror("ioctl (I_POP)");

	(void) printf("modules on the stream: ");
	(void) fflush(stdout);

	/*
	 * Print out what the stream currently looks like.
	 */
	(void) dup2(strrecvfd.fd, 0);
	(void) system("strconf | paste -s -d' ' -");

	return (EXIT_SUCCESS);
}

Chapter 12 Multithreaded STREAMS

This chapter describes how to write a multithreaded STREAMS driver or module. It covers the necessary conversion topics so that new and existing STREAMS modules and drivers run in the multithreaded kernel. It describes STREAMS-specific multithreading issues and techniques. For general information, see Writing Device Drivers.

This chapter contains the following information:

• Multithreaded (MT) STREAMS Overview

• MT STREAMS Framework

• MT STREAMS Perimeters

• MT SAFE Modules and Drivers

• Routines Used Inside a Perimeter

• MT SAFE Modules Using Explicit Locks

• Preparing to Port

• Sample Multithreaded Device Driver Using a Per Module Inner Perimeter

• Sample Multithreaded Module With Outer Perimeter

Multithreaded (MT) STREAMS Overview

The SunOS 5 operating system is fully multithreaded, which means that it can make effective use of the available parallelism of a symmetric shared-memory multiprocessor computer. All kernel subsystems are multithreaded: scheduler, virtual memory, file systems, block/character/STREAMS I/O, networking protocols, and device drivers.

MT STREAMS requires you to use some different concepts and terminology. These concepts apply not only to STREAMS drivers, but to all device drivers in the Solaris operating environment. For a more complete description of these terms, see Writing Device Drivers. Additionally, see Chapter 1, Overview of STREAMS of this guide for definitions and Chapter 8, STREAMS Kernel-Level Mechanisms for elements of MT drivers.

Some of the multithreaded terms and ideas are.

Thread

Sequence of instructions executed within context of a process.

Lock

Mechanism to restrict access to data structures.

Single Threaded

Restricting access to a single thread.

Multithreaded

Allowing two or more threads access to a data element.

Multiprocessing

Two or more CPUs concurrently executing the Operating System.

Concurrency

Simultaneous execution.

Preemption

Suspending execution for the next thread to run.

Monitor

Portion of code that is single-threaded.

Mutual Exclusion

Exclusive access to a data element by a single thread at one time.

Condition Variables

Kernel event synchronization primitives.

Counting Semaphores

Memory-based synchronization mechanism.

Readers/Writer Locks

Data lock allowing one writer or many readers at one time.

Callback

On occurrence of a specific event, call a module function.

Synchronous Access

Only one thread is allowed in the perimeter. Upon return from a call the action is complete; when the thread is done, the job is done.

Asynchronous Access

Multiple threads are allowed in the perimeter. Upon return form a call there is no guarantee that the job is complete.

Perimeter Claim

A thread has synchronous access in the perimeter. The claim prevents subsequent synchronous access until the claim is released.

Exclusive Access

Calling a synchronous entry point in the perimeter.

Writer

A thread that has exclusive access to a perimeter.

MT STREAMS Framework

The STREAMS framework consists of the stream head, documented STREAMS data structures (such as queue_t, mblk_t) and STREAMS utility routines including STREAMS facilities documented in the Device Driver Interface (DDI). The STREAMS framework enables multiple kernel threads to concurrently enter and execute code defined by each module, including the open, close, put, and service procedures of each queue within the system.

The first goal of the SunOS 5 system is to preserve the interface and flavor of STREAMS and to shield module code as much as possible from the impact of migrating to the multithreaded kernel. Most of the locking is hidden from the programmer and performed by the STREAMS kernel framework. As long as module code uses the standard, documented programmatic interfaces to shared kernel data structures (such as queue_t, mblk_t, and dblk_t), and adheres to the DDI/DKI, the user does not have to explicitly lock these framework data structures.

The second goal is to make writing MT SAFE modules simple. One of the ways that the framework accomplishes this is by using the MT STREAMS perimeter mechanisms for controlling and restricting concurrent access to a STREAMS module. STREAMS perimeters allow the module writer to select the level of concurrency that a module can tolerate.

STREAMS Framework Integrity

The STREAMS framework ensures the integrity of the STREAMS data structures, such as queue_t, mblk_t, and dblk_t as long as the module conforms to the DDI/DKI, and does not directly access global operating system data structures or facilities not described in the DDI/DKI.

The q_next fields of the queue_t structure are not modified by the framework while a thread is actively executing within a synchronous entry point. However the q_next field might change while a thread is executing within an asynchronous entry point.

The q_ptr field is considered private to the module and the framework will not manipulate its value. When making a module MT Safe, the integrity of the module-private data structures must be ensured by the module itself. This integrity can be guaranteed by creating private locks, or by the control of concurrency within the module by the use of STREAMS perimeters. Knowing what the framework supports is critical in deciding what the module writer must provide.

Hardening Information. As in previous Solaris operating environment releases, a module must not call another module's put or service procedures directly. The DDI/DKI routines putnext(9F), put(9F), and other routines in Section 9F must be used to pass a message to another queue. Calling another module's routines directly circumvents the design of the MT STREAMS framework and can yield unknown results.

Hardening Information. Once a message is passed using a putq, put, putnext, as well as the perimeter function qwriter, it cannot be accessed again because the use of this message has been given to the new routine. If a reference needs to be retained by the module, it should copy it by using copyb, copymsg, dupb, or dupmsg.

Message Ordering

The STREAMS framework guarantees the ordering of messages along a stream if all the modules in the stream preserve message ordering internally. This ordering guarantee only applies to messages that are sent along the same stream and produced by the same source.

The STREAMS framework does not guarantee that a message has been seen by the next put procedure when the call to putnext(9F) or qreply(9F) returns.

MT STREAMS Perimeters

Solaris STREAMS uses a facility known as perimeters for handling thread concurrency in STREAMS modules. Perimeters allow the module writer to select conditions that will result in exclusive access to a queue, a pair of queues, or all the queues in a module. This makes multithreading issues easier to work with in STREAMS.

Perimeters work somewhat like reader/writer locks, where there can be many readers, but only one writer. A synchronous access to the perimeter is similar to holding the writer lock, in that only one thread can be in the perimeter at a time. Synchronous entry will hold a perimeter exclusively until the thread eventually unwinds out of the perimeter (usually when it returns from a put/putnext or a qwriter call that initially invoked the synchronous behavior). While a thread has synchronous access to the perimeter, all other access (synchronous or asynchronous) will be deferred.

An asynchronous access is similar to the reader lock, where many threads can be in the perimeter at a time, including the possible recursive entry of a thread previously entering the perimeter. The asynchronous "claim" is not released until the thread winds out of the put/putnext. Because asynchronous access is similar to the reader lock, any synchronous access will be deferred or blocked until all asynchronous claims are released.

Hardening Information. The perimeter data is private to the STREAMS framework, and should not be modified by the module itself.

STREAMS enables the definition of two synchronous perimeters. One is an inner perimeter, and is used to define synchronous entry points on a queue or a queue pair (the read and write queue's for a specific module instance). It also identifies an outer perimeter, which is made up of all the inner perimeters for all the queues for a specific module.

There is also a special inner perimeter, PERMOD, that is similar to the outer perimeter, but does not have the overhead of the outer perimeter. PERMOD identifies a single synchronous entry point for all queues for this module. Because PERMOD is like a hybrid of an inner perimeter and outer perimeter, the PERMOD perimeter cannot have an outer perimeter.

Inner Perimeters

For the most part, the module writer does not need to specify an inner perimeter, as the STREAMS framework automatically creates it for the module. What needs to be specified is the type of perimeter, and the concurrency of the perimeter.

Inner perimeters come in two types:

D_PERQ

Enables synchronous entry to be different between the read queue and the write queue. Therefore, if a synchronous putnext is occurring on the read queue, a synchronous or asynchronous putnext can occur on the write queue (or other (a)synchronous access on the write queue).

D_QPAIR

Protects both the read queue and the write queue, so synchronous access to one queue will prevent synchronous or asynchronous access to the other queue.

Another perimeter, D_PERMOD is slightly different, and is discussed in PERMOD Perimeter.

An inner perimeter becomes exclusive (writer) whenever an inner synchronous entry point is encountered. By default all the entry points are considered to be synchronous until enabled as "shared" entry points. As previously stated, synchronous entry points remain exclusive until the thread returns to the caller of the synchronous entry point. If the synchronous function calls putnext, the perimeter remains exclusive across the putnext, up till the synchronous function can return to its caller, and subsequent entries into the perimeter will be deferred.

Inner perimeters can specify additional concurrency on the STREAMS entry points for open/close, put/putnext, service, and callbacks as shown in Table 12–4.

Outer Perimeters

The module writer can also specify an outer perimeter. An outer perimeter is the linked list of all inner perimeters for all queues associated with the specified module. Entering the outer perimeter is equivalent to entering each of the inner perimeters. As this can also be an expensive operation, the outer perimeter is only entered synchronously, and upon successful completion of a qwriter(PERIM_OUTER) makes the outer perimeter exclusive. This also has the effect of making each of the inner perimeters exclusive.

Use of outer perimeters is reserved for module data that has an effect on all queue instances of the module, such as module state that might allow messages to pass between other instances of the module, information that allows a driver to configure shared hardware, or at open/close time when information is needed for all open instances for a module.

Outer perimeters, at this time, have only one concurrency modifier. This is D_MTOCEXCL, and instructs the framework to enter the outer perimeter on each open and close of queues for the module.

PERMOD Perimeter

The PERMOD perimeter is a hybrid of the inner and outer perimeter. It is implemented primarily for modules that might have a large number of queue instances, and cannot afford the latency for entering the outer perimeter. Because it is a hybrid, PERMOD perimeters cannot have an outer perimeter, and modules that have D_MTPERMOD and D_MTOUTPERIM defined will fail at open. As PERMOD perimeters are implemented as inner perimeters, they share all the concurrency states as the inner perimeter, see Table 12–2.

Hot Perimeters

All STREAMS modules and drivers in the Solaris operating environment must be D_MTSAFE, and must account for multithreading. Specifying an inner and/or outer perimeter will handle concurrency issues that the module writer may encounter while developing the module or driver. Experienced STREAMS programmers might decide that the perimeter should not have any synchronous entry points, and should run fully hot. To define a fully-hot perimeter, the module writer need only specify the D_MTSAFE flag without an inner perimeter type (D_MTPERQ, D_MTQPAIR, D_MTPERMOD) and without an outer perimeter (D_MTOUTPERIM).

Hardening Information. All STREAMS entry points run concurrently, and in a multiprocessor environment, there can be a put procedure running simultaneously with a service procedure or even a close procedure. So the writer must take precautions against kernel panics by making sure that other concurrent threads will not reference data the current thread is trying to change or remove.

Defining Perimeter Types

To configure a module with perimeter types and concurrency types, use the f_flag field in fmodsw(9s) with D_MTSAFE or'd with the appropriate perimeter type flags. See Table 12–1.

The easiest method is to initially implement your module and configure it to be per-module single threaded, and increase the level of concurrency as needed. Sample Multithreaded Device Driver Using a Per Module Inner Perimeter provides a complete example of using a per-module perimeter, and Sample Multithreaded Module With Outer Perimeter provides a complete example with a higher level of concurrency.

To configure a driver with perimeter and concurrency types, put MT_SAFE and the appropriate perimeter flags in the cb_flag field of the cb_ops structure for the driver.

Choosing a Perimeter Type

Table 12–1 summarizes examples of when to use an inner perimeter, or both an inner and outer perimeter for a STREAMS module.

Several flags specify the inner and outer perimeters (see Table 12–2 and Table 12–3). These flags fall into three categories:

• Define the presence and scope of the inner perimeter

• Define the presence of the outer perimeter (which can have only one scope)

• Modify the default concurrency for the different entry points

You configure the inner perimeter by choosing one of the mutually exclusive flags shown in Table 12–2.

Configure the outer perimeter using the flag shown in Table 12–3.

By default all synchronous entry points enter the inner perimeter exclusively and enter the outer perimeter shared. To modify this behavior use the flags shown in Table 12–4.

Hardening Information. Concurrency flags designated with a preceding underbar "_" are experimental, and their behavior might change in the future and should not be relied upon.

MT SAFE Modules and Drivers

A module or a driver can be either MT SAFE or MT UNSAFE. A module or driver is MT SAFE when its data values are correct regardless of the order that multiple threads access and modify the data. For MT SAFE mode, use MT STREAMS perimeters to restrict the concurrency in a module or driver to:

• Per-module single threading

• Per queue-pair single threading

• Per queue single threading

• Per queue or per queue-pair single threading of the put and service procedures with per module single threading of the open and close routines

• Unrestricted concurrency in the put and service procedures with the ability to restrict the concurrency when handling messages that modify data

• Completely unrestricted concurrency

MT SAFE Module

To configure a module as being MT SAFE, use the f_flag field in fmodsw(9S).

The easiest method is to initially implement your module and configure it to be per-module single threaded, and increase the level of concurrency as needed. Sample Multithreaded Device Driver Using a Per Module Inner Perimeter provides a complete example of using a per-module perimeter, and Sample Multithreaded Module With Outer Perimeter provides a complete example with a higher level of concurrency.

MT SAFE modules can use different MT STREAMS perimeters to restrict the concurrency in the module to a concurrency that is natural given the data structures that the module contains, thereby removing the need for module private locks (see MT STREAMS Perimeters for information on perimeters). A module that requires unrestricted concurrency can be configured to have no perimeters. Such modules have to use explicit locking primitives to protect their data structures. While such modules can exploit the maximum level of concurrency allowed by the underlying hardware platform, they are more complex to develop and support. See MT SAFE Modules Using Explicit Locks.

Independent of the perimeters, there will be at most one thread allowed within any given queue's service procedure.

Your MT SAFE modules should use perimeters and avoid using module private locks (mutex, condition variables, readers/writer, or semaphore). Should you opt to use module private locks, you need to read MT SAFE Modules Using Explicit Locks along with this section.

MT UNSAFE mode for STREAMS modules was temporarily supported as an aid in porting SVR4 modules; however, MT UNSAFE is not supported after SVR4. Beginning with the release of the Solaris 7 operating environment, no MT UNSAFE module or driver has been supported.

Upper and lower multiplexors share the same perimeter type and concurrency level.

MT SAFE Driver

To configure a driver as being MT SAFE, initialize the cb_ops(9S) and dev_ops(9S) data structures. This code must be in the header section of your module. For more information, see Example 12–1, and dev_ops(9S).

The driver is configured to be MT SAFE by setting the cb_flag to D_MP. It also specifies any MT STREAMS perimeters by setting flags in the cb_flag field. (See mt-streams(9F).)

Routines Used Inside a Perimeter

This section describes the routines and data fields used after you enter a perimeter.

qprocson/qprocsoff

The routines qprocson(9F) and qprocsoff(9F) respectively enable and disable the put and service procedures of the queue pair. Before calling qprocson(9F) and after calling qprocsoff(9F), the module's put and service procedures are disabled; messages flow around the module as if it were not present in the stream.

Call qprocson(9F) in the first open of a module, but only after allocating and initializing any module resources on which the put and service procedures depend. Call the qprocsoff routine in the close routine of the module before deallocating any resources on which the put and service procedures depend.

To avoid deadlocks, modules must not hold private locks across the calls to qprocson(9F) or qprocsoff(9F).

qtimeout/qunbufcall

The timeout(9F) and bufcall(9F) callbacks are asynchronous. For a module using MT STREAMS perimeters, the timeout(9F) and bufcall(9F) callback functions execute outside the scope of the perimeters. This makes synchronization of callbacks with the rest of the module complex.

To make timeout(9F) and bufcall(9F) functionality easier to use for modules with perimeters, there are additional interfaces that use synchronous callbacks. These routines are qtimeout(9F), quntimeout(9F), qbufcall(9F), and qunbufcall(9F). When using these routines, the callback functions are executed inside the perimeters, and hence have the same concurrency restrictions as the put and service procedures.

qwriter

Modules can use the qwriter(9F) function to upgrade from shared to exclusive access at a perimeter. For example, a module with an outer perimeter can use qwriter(9F) in the put procedure to upgrade to exclusive access at the outer perimeter. A module where the put procedure runs with shared access at the inner perimeter (D_MTPUTSHARED) can use qwriter(9F) in the put procedure to upgrade to exclusive access at the inner perimeter.

Returning from a qwriter call does not mean that the callback function has executed. If the framework can become exclusive in the qwriter call, it will enter the perimeter synchronously, and execute the callback. If it cannot, the callback will be deferred. It is a good idea for any caller of qwriter to immediately return to its caller as there is little that can be accomplished in this thread of execution.

Hardening Information. Do not call qwriter with another queue, as qwriter assumes that the caller has already made a claim to the perimeter that the queue is associated with (asynchronous entry), and calling another perimeter will cause problems.

qwriter(9F) cannot be used in the open or close procedures. If a module needs exclusive access at the outer perimeter in the open and/or close procedures, it has to specify that the outer perimeter should always be entered exclusively for open and close (using D_MTOCEXCL).

The STREAMS framework guarantees that all deferred qwriter(9F) callbacks associated with a queue have executed before the module's close routine is called for that queue.

For an example of a driver using qwriter(9F) see Example 12–2.

qwait

A module that uses perimeters and must wait in its open or close procedure for a message from another STREAMS module has to wait outside the perimeters; otherwise, the message would never be allowed to enter its put and service procedures. This is accomplished by using the qwait(9F) interface. See qwriter(9F) man page for an example.

Asynchronous Callback Functions

Interrupt handlers and other asynchronous callback functions require special care by the module writer, because they can execute asynchronously to threads executing within the module open, close, put, and service procedures.

For modules using perimeters, use qtimeout(9F) and qbufcall(9F) instead of timeout(9F) and bufcall(9F). The qtimeout and qbufcall callbacks are synchronous and consequently introduce no special synchronization requirements.

Because a thread can enter the module at any time, you must ensure that the asynchronous callback function acquires the proper private locks before accessing private module data structures, and releases these locks before returning. You must cancel any outstanding registered callback routines before the data structures on which the callback routines depend are deallocated and the module closed.

• For hardware device interrupts, this involves disabling the device interrupts.

• Outstanding callbacks from timeout(9F) and bufcall(9F) must be canceled by calling untimeout(9F) and unbufcall(9F).

  The module cannot hold certain private locks across calls to untimeout(9F) or unbufcall(9F). These locks are those that the module's timeout(9F) or bufcall(9F) callback functions acquire. See MT SAFE Modules Using Explicit Locks.

• If outstanding callbacks from esballoc(9F) are associated with a particular stream, they must be allowed to complete before the module close routine deallocates the private data structures on which they depend.

close() Race Conditions

Because the callback functions are by nature asynchronous, they can be executing or about to execute at the time the module close routine is called. You must cancel all outstanding callback and interrupt conditions before deallocating those data structures or returning from the close routine.

The callback functions scheduled with timeout(9F) and bufcall(9F) are guaranteed to have been canceled by the time untimeout(9F) and unbufcall(9F) return. The same is true for qtimeout(9F) and qbufcall(9F) by the time quntimeout(9F) and qunbufcall(9F) return. You must also take responsibility for other asynchronous routines, including esballoc(9F) callbacks and hardware, as well as software interrupts.

Unloading a Module that Uses esballoc

The STREAMS framework prevents a module or driver text from being unloaded while there are open instances of the module or driver. If a module does not cancel all callbacks in the last close routine, it should not be allowed to be unloaded.

This is an issue mainly for modules and drivers using esballoc because esballoc callbacks cannot be canceled. Thus, modules and drivers using esballoc have to be prepared to handle calls to the esballoc callback free function after the last instance of the module or driver has been closed.

Modules and drivers can maintain a semaphore count of outstanding callbacks. They can deny an unload by making the _fini(9E) routine return EBUSY if there are outstanding callbacks.

Use of the q_next Field

The q_next field in the queue_t structure can be referenced in open, close, put, and service procedures as well as the synchronous callback procedures (scheduled with qtimeout(9F), qbufcall(9F), and qwriter(9F)). However, the value in the q_next field should not be trusted. It is relevant to the STREAMS framework, but may not be relevant to a specific module.

All other module code, such as interrupt routines, timeout(9F) and esballoc(9F) callback routines, cannot dereference q_next. Those routines have to use the “next” version of all functions. For instance, use canputnext(9F) instead of dereferencing q_next and using canput(9F).

MT SAFE Modules Using Explicit Locks

Although the result is not reliable, you can use explicit locks either instead of perimeters or to augment the concurrency restrictions provided by the perimeters.

Explicit locks cannot be used to preserve message ordering in a module because of the risk of re-entering the module. Use MT STREAMS perimeters to preserve message ordering.

All four types of kernel synchronization primitives are available to the module writer: mutexes, readers/writer locks, semaphores, and condition variables. Because cv_wait implies a context switch, it can only be called from the module's open and close procedures, which are executed with valid process context. You must use the synchronization primitives to protect accesses and ensure the integrity of private module data structures.

Constraints When Using Locks

When adding locks in a module, observe these constraints:

• Avoid holding module private locks across calls to putnext(9F). The module might be re-entered by the same thread that called putnext(9F), causing the module to try to acquire a lock that it already holds. This can cause kernel panic.

• Do not hold module private locks, acquired in put or service procedures, across the calls to qprocson(9F) or qprocsoff(9F). Doing this causes deadlock, since qprocson(9F) and qprocsoff(9F) wait until all threads leave the inner perimeter.

• Similarly, do not hold locks, acquired in the timeout(9F) and bufcall(9F) callback procedures, across the calls to untimeout(9F) or unbufcall(9F). Doing this causes deadlock, because untimeout(9F)and unbufcall(9F) wait until an already executing callback has completed.

The first restriction deters using module private locks to preserve message ordering. The preferred mechanism is to use MT STREAMS perimeters to preserve message ordering.

Preserving Message Ordering

Module private locks cannot be used to preserve message ordering because they cannot be held across calls to putnext(9F) and the other messages that pass routines to other modules. The alternatives for preserving message ordering are:

• Use MT STREAMS perimeters.

• Pass all messages through the service procedures. The service procedure can drop the locks before calling putnext(9F) or qreply(9F), without reordering messages, because the framework guarantees that at most, one thread will execute in the service procedure for a given queue.

Use perimeters to avoid the performance penalty for using service procedures.

Preparing to Port

When modifying a STREAMS driver to take advantage of the multithreaded kernel, a level of MT safety is selected according to:

• The desired degree of concurrency

• The natural concurrency of the underlying module

• The amount of effort or complexity required

Much of the effort in conversion is simply determining the appropriate degree of data sharing and the corresponding granularity of locking (see Table 12–1). The actual time spent configuring perimeters and/or installing locks should be much smaller than the time spent in analysis.

To port your module, you must understand the data structures used within your module, as well as the accesses to those data structures. You must fully understand the relationship between all portions of the module and private data within that module, and to use the MT STREAMS perimeters (or the synchronization primitives available) to maintain the integrity of these private data structures.

You must explicitly restrict access to private module data structures as appropriate to ensure the integrity of these data structures. You must use the MT STREAMS perimeters to restrict the concurrency in the module so that the parts of the module that modify private data are single-threaded with respect to the parts of the module that read the same data. (For more information about perimeters, see MT STREAMS Perimeters.) Besides perimeters, you can use the synchronization primitives available (mutex, condition variables, readers/writer, semaphore) to explicitly restrict access to module private data appropriate for the operations within the module on that data.

The first step in multithreading a module or driver is to analyze the module, breaking the entire module up into a list of individual operations and the private data structures referenced in each operation. Part of this first step is deciding upon a level of concurrency for the module. Ask yourself which of these operations can be multithreaded and which must be single-threaded. Try to find a level of concurrency that is “natural” for the module and matches one of the available perimeters (or, alternatively, requires the minimal number of locks) , and has a simple and straightforward implementation. Avoid additional unnecessary complexity.

Typical questions to ask are:

• What data structures are maintained within the module?

• What types of accesses are made to each field of these data structures?

• When is each data structure accessed destructively (written) and when is it accessed non-destructively (read)?

• Which operations within the module should be allowed to execute concurrently?

• Is per module single-threading appropriate for the module?

• Is per queue-pair or per queue single-threading appropriate?

• What are the message ordering requirements?

Porting to the SunOS 5 System

When porting a STREAMS module or driver from the SunOS 4 system to the SunOS 5 system, the module should be examined with respect to the following areas:

• The SunOS 5 Device Driver Interface (DDI/DKI)

• The SunOS 5 MT design

For portability and correct operation, each module must adhere to the SunOS DDI/DKI. Several facilities available in previous releases of the SunOS system have changed and can take different arguments, or produce different side effects, or no longer exist in the SunOS 5 system. The module writer should carefully review the module with respect to the DDI/DKI.

Each module that accesses underlying Sun-specific features included in the SunOS 5 system should conform to the Device Driver Interface. The SunOS 5 DDI defines the interface used by the device driver to register device hardware interrupts, access device node properties, map device slave memory, and establish and synchronize memory mappings for DVMA (Direct Virtual Memory Access). These areas are primarily applicable to hardware device drivers. Refer to the Device Driver Interface Specification within the Writing Device Drivers for details on the SunOS 5 DDI and DVMA.

The kernel networking subsystem in the SunOS 5 system is based on STREAMS. Datalink drivers that used the ifnet interface in the SunOS 4 system must be converted to use DLPI for the SunOS 5 system. Refer to the Data Link Provider Interface, Revision 2 specification.

After reviewing the module for conformance to the SunOS 5 DKI and DDI specifications, you should be able to consider the impact of multithreading on the module.

Sample Multithreaded Device Driver Using a Per Module Inner Perimeter

Example 12–1 is a sample multithreaded, loadable, STREAMS pseudo-driver. The driver MT design is the simplest possible based on using a per module inner perimeter. Thus, only one thread can execute in the driver at any time. In addition, a quntimeout(9F) synchronous callback routine is used. The driver cancels an outstanding qtimeout(9F) by calling quntimeout(9F) in the close routine. See close() Race Conditions.

Example 12–1 Multithreaded, Loadable, STREAMS Pseudo-Driver

/*
 * Example SunOS 5 multithreaded STREAMS pseudo device driver.
 * Using a D_MTPERMOD inner perimeter.
 */

#include		<sys/types.h>
#include		<sys/errno.h>
#include		<sys/stropts.h>
#include		<sys/stream.h>
#include		<sys/strlog.h>
#include		<sys/cmn_err.h>
#include		<sys/modctl.h>
#include		<sys/kmem.h>
#include		<sys/conf.h>
#include		<sys/ksynch.h>
#include		<sys/stat.h>
#include		<sys/ddi.h>
#include		<sys/sunddi.h>

/*
 * Function prototypes.
 */
static			int xxidentify(dev_info_t *);
static			int xxattach(dev_info_t *, ddi_attach_cmd_t);
static			int xxdetach(dev_info_t *, ddi_detach_cmd_t);
static			int xxgetinfo(dev_info_t *,ddi_info_cmd_t,void *,void**);
static			int xxopen(queue_t *, dev_t *, int, int, cred_t *);
static			int xxclose(queue_t *, int, cred_t *);
static			int xxwput(queue_t *, mblk_t *);
static			int xxwsrv(queue_t *);
static			void xxtick(caddr_t);

/*
 * Streams Declarations
 */
static struct module_info xxm_info = {
   99,            /* mi_idnum */
   "xx",          /* mi_idname */
   0,             /* mi_minpsz */
   INFPSZ,        /* mi_maxpsz */
   0,             /* mi_hiwat */
   0              /* mi_lowat */
};

static struct qinit xxrinit = {
		NULL,           /* qi_putp */
		NULL,           /* qi_srvp */
		xxopen,         /* qi_qopen */
		xxclose,        /* qi_qclose */
		NULL,           /* qi_qadmin */
		&xxm_info,      /* qi_minfo */
		NULL            /* qi_mstat */
};

static struct qinit xxwinit = {
		xxwput,         /* qi_putp */
		xxwsrv,         /* qi_srvp */
		NULL,           /* qi_qopen */
		NULL,           /* qi_qclose */
		NULL,           /* qi_qadmin */
		&xxm_info,      /* qi_minfo */
		NULL            /* qi_mstat */
};

static struct streamtab xxstrtab = {
		&xxrinit,       /* st_rdinit */
		&xxwinit,       /* st_wrinit */
		NULL,           /* st_muxrinit */
		NULL            /* st_muxwrinit */
};

/*
 * define the xx_ops structure.
 */

static 				struct cb_ops cb_xx_ops = {
		nodev,            /* cb_open */
		nodev,            /* cb_close */
		nodev,            /* cb_strategy */
		nodev,            /* cb_print */
		nodev,            /* cb_dump */
		nodev,            /* cb_read */
		nodev,            /* cb_write */
		nodev,            /* cb_ioctl */
		nodev,            /* cb_devmap */
		nodev,            /* cb_mmap */
		nodev,            /* cb_segmap */
		nochpoll,         /* cb_chpoll */
		ddi_prop_op,      /* cb_prop_op */
		&xxstrtab,        /* cb_stream */
		(D_NEW|D_MP|D_MTPERMOD) /* cb_flag */
};

static struct dev_ops xx_ops = {
		DEVO_REV,         /* devo_rev */
		0,                /* devo_refcnt */
		xxgetinfo,        /* devo_getinfo */
		xxidentify,       /* devo_identify */
		nodev,            /* devo_probe */
		xxattach,         /* devo_attach */
		xxdetach,         /* devo_detach */
		nodev,            /* devo_reset */
		&cb_xx_ops,       /* devo_cb_ops */
		(struct bus_ops *)NULL /* devo_bus_ops */
};


/*
 * Module linkage information for the kernel.
 */
static struct modldrv modldrv = {
		&mod_driverops,   /* Type of module. This one is a driver */
		"xx",             /* Driver name */
		&xx_ops,          /* driver ops */
};

static struct modlinkage modlinkage = {
		MODREV_1,
		&modldrv,
		NULL
};

/*
 * Driver private data structure. One is allocated per Stream.
 */
struct xxstr {
		struct		xxstr *xx_next;	/* pointer to next in list */
		queue_t		*xx_rq;				/* read side queue pointer */
		minor_t		xx_minor;			/* minor device # (for clone) */
		int			xx_timeoutid;		/* id returned from timeout() */
};

/*
 * Linked list of opened Stream xxstr structures.
 * No need for locks protecting it since the whole module is
 * single threaded using the D_MTPERMOD perimeter.
 */
static struct xxstr						*xxup = NULL;


/*
 * Module Config entry points
 */

_init(void)
{
	  return (mod_install(&modlinkage));
}

_fini(void)
{
	  return (mod_remove(&modlinkage));
}

_info(struct modinfo *modinfop)
{
	  return (mod_info(&modlinkage, modinfop));
}

/*
 * Auto Configuration entry points
 */

/* Identify device. */
static int
xxidentify(dev_info_t *dip)
{
	  if (strcmp(ddi_get_name(dip), "xx") == 0)
			return (DDI_IDENTIFIED);
	  else
			return (DDI_NOT_IDENTIFIED);
}

/* Attach device. */
static int
xxattach(dev_info_t *dip, ddi_attach_cmd_t cmd)
{
	  /* This creates the device node. */
	  if (ddi_create_minor_node(dip, "xx", S_IFCHR, ddi_get_instance(dip), 
				DDI_PSEUDO, CLONE_DEV) == DDI_FAILURE) {
			return (DDI_FAILURE);
	  }
	  ddi_report_dev(dip);
	  return (DDI_SUCCESS);
}

/* Detach device. */
static int
xxdetach(dev_info_t *dip, ddi_detach_cmd_t cmd)
{
	  ddi_remove_minor_node(dip, NULL);
	  return (DDI_SUCCESS);
}

/* ARGSUSED */
static int
xxgetinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,	 
				void **resultp)
{
	  dev_t dev = (dev_t) arg;
	  int instance, ret = DDI_FAILURE;

	  devstate_t *sp;
	  state *statep;
	  instance = getminor(dev);

	  switch (infocmd) {
			case DDI_INFO_DEVT2DEVINFO:
				if ((sp = ddi_get_soft_state(statep, 
						getminor((dev_t) arg))) != NULL) {
					*resultp = sp->devi;
					ret = DDI_SUCCESS;
				} else
					*result = NULL;
				break;

			case DDI_INFO_DEVT2INSTANCE:
				*resultp = (void *)instance;
				ret = DDI_SUCCESS;
				break;

			default:
				break;
	  }
	  return (ret);
}

static
xxopen(rq, devp, flag, sflag, credp)
	  queue_t			*rq;
	  dev_t				*devp;
	  int				flag;
	  int				sflag;
	  cred_t			*credp;
{
	  struct xxstr *xxp;
	  struct xxstr **prevxxp;
	  minor_t 			minordev;

	  /* If this stream already open - we're done. */
	  if (rq->q_ptr)
			return (0);

    /* Determine minor device number. */
	  prevxxp = & xxup;
	  if (sflag == CLONEOPEN) {
			minordev = 0;
			while ((xxp = *prevxxp) != NULL) {
				if (minordev < xxp->xx_minor)
					break;
				minordev++;
				prevxxp = &xxp->xx_next;
			}
			*devp = makedevice(getmajor(*devp), minordev)
	  } else
			minordev = getminor(*devp);

	  /* Allocate our private per-Stream data structure. */
	  if ((xxp = kmem_alloc(sizeof (struct xxstr), KM_SLEEP)) == NULL)
			return (ENOMEM);

	  /* Point q_ptr at it. */
	  rq->q_ptr = WR(rq)->q_ptr = (char *) xxp;

	  /* Initialize it. */
	  xxp->xx_minor = minordev;
	  xxp->xx_timeoutid = 0;
	  xxp->xx_rq = rq;

	  /* Link new entry into the list of active entries. */
	  xxp->xx_next = *prevxxp;
	  *prevxxp = xxp;

	  /* Enable xxput() and xxsrv() procedures on this queue. */
	  qprocson(rq);

	  return (0);
}

static
xxclose(rq, flag, credp)
	  queue_t			*rq;
	  int				flag;
	  cred_t			*credp;

{
	  struct		xxstr		*xxp;
	  struct		xxstr		**prevxxp;

	  /* Disable xxput() and xxsrv() procedures on this queue. */
	  qprocsoff(rq);
	  /* Cancel any pending timeout. */
		 xxp = (struct xxstr *) rq->q_ptr;
		 if (xxp->xx_timeoutid != 0) {
	 		 (void) quntimeout(rq, xxp->xx_timeoutid);
	 		 xxp->xx_timeoutid = 0;
		 }
	  /* Unlink per-stream entry from the active list and free it. */
	  for (prevxxp = &xxup; (xxp = *prevxxp) != NULL; 
				prevxxp = &xxp->xx_next)
			if (xxp == (struct xxstr *) rq->q_ptr)
				break;
	  *prevxxp = xxp->xx_next;
	  kmem_free (xxp, sizeof (struct xxstr));

	  rq->q_ptr = WR(rq)->q_ptr = NULL;

	  return (0);
}

static
xxwput(wq, mp)
	  queue_t		*wq;
	  mblk_t		*mp;
{
	  struct xxstr	*xxp = (struct xxstr *)wq->q_ptr;

		/* write your code here */
     /* *** Sacha's Comments *** broken */
 		freemsg(mp);
		mp = NULL;

		if (mp != NULL)
			putnext(wq, mp);
}

static
xxwsrv(wq)
		queue_t		*wq;
{
		mblk_t		*mp;
		struct xxstr	*xxp;

		xxp = (struct xxstr *) wq->q_ptr;

		while (mp = getq(wq)) {
			/* write your code here */
			freemsg(mp);

			/* for example, start a timeout */
			if (xxp->xx_timeoutid != 0) {
				/* cancel running timeout */
				(void) quntimeout(wq, xxp->xx_timeoutid);
			}
			xxp->xx_timeoutid = qtimeout(wq, xxtick, (char *)xxp, 10);
		}
}

static void
xxtick(arg)
		caddr_t arg;
{
		struct xxstr *xxp = (struct xxstr *)arg;

		xxp->xx_timeoutid = 0;      /* timeout has run */
		/* write your code here */

}

Sample Multithreaded Module With Outer Perimeter

Example 12–2 is a sample multithreaded, loadable STREAMS module. The module MT design is a relatively simple one, based on a per queue-pair inner perimeter plus an outer perimeter. The inner perimeter protects per-instance data structure (accessed through the q_ptr field) and the module global data is protected by the outer perimeter. The outer perimeter is configured so that the open and close routines have exclusive access to the outer perimeter. This is necessary because they both modify the global-linked list of instances. Other routines that modify global data are run as qwriter(9F) callbacks, giving them exclusive access to the whole module.

Example 12–2 Multithread Module with Outer Perimeter

/*
 * Example SunOS 5 multi-threaded STREAMS module.
 * Using a per-queue-pair inner perimeter plus an outer perimeter.
 */

#include		<sys/types.h>
#include		<sys/errno.h>
#include		<sys/stropts.h>
#include		<sys/stream.h>
#include		<sys/strlog.h>
#include		<sys/cmn_err.h>
#include		<sys/kmem.h>
#include		<sys/conf.h>
#include		<sys/ksynch.h>
#include		<sys/modctl.h>
#include		<sys/stat.h>
#include		<sys/ddi.h>
#include		<sys/sunddi.h>

/*
 * Function prototypes.
 */
static			int xxopen(queue_t *, dev_t *, int, int, cred_t *);
static			int xxclose(queue_t *, int, cred_t *);
static			int xxwput(queue_t *, mblk_t *);
static			int xxwsrv(queue_t *);
static			void xxwput_ioctl(queue_t *, mblk_t *);
static			int xxrput(queue_t *, mblk_t *);
static			void xxtick(caddr_t);

/*
 * Streams Declarations
 */
static struct module_info xxm_info = {
   99,					/* mi_idnum */
   “xx”,			/* mi_idname */
   0,					/* mi_minpsz */
   INFPSZ,			/* mi_maxpsz */
   0,					/* mi_hiwat */
   0					/* mi_lowat */
};
/*
 * Define the read-side qinit structure
 */
static struct qinit xxrinit = {
		xxrput,         /* qi_putp */
		NULL,           /* qi_srvp */
		xxopen,         /* qi_qopen */
		xxclose,        /* qi_qclose */
		NULL,           /* qi_qadmin */
		&xxm_info,      /* qi_minfo */
		NULL            /* qi_mstat */
};
/*
 * Define the write-side qinit structure
 */
static struct qinit xxwinit = {
		xxwput,         /* qi_putp */
		xxwsr,          /* qi_srvp */
		NULL,           /* qi_qopen */
		NULL,           /* qi_qclose */
		NULL,           /* qi_qadmin */
		&xxm_info,      /* qi_minfo */
		NULL            /* qi_mstat */
};

static struct streamtab xxstrtab = {
		&xxrini,        /* st_rdinit */
		&xxwini,        /* st_wrinit */
		NULL,           /* st_muxrinit */
		NULL            /* st_muxwrinit */
};

/*
 * define the fmodsw structure.
 */

static struct fmodsw xx_fsw = {
		“xx”,         /* f_name */
		&xxstrtab,      /* f_str */
		(D_NEW|D_MP|D_MTQPAIR|D_MTOUTPERIM|D_MTOCEXCL) /* f_flag */
};

/*
 * Module linkage information for the kernel.
 */
static struct modlstrmod modlstrmod = {
		&mod_strmodops,	/* Type of module; a STREAMS module */
		“xx module”,		/* Module name */
		&xx_fsw,				/* fmodsw */
};

static struct modlinkage modlinkage = {
		MODREV_1,
		&modlstrmod,
		NULL
};

/*
 * Module private data structure. One is allocated per stream.
 */
struct xxstr {
		struct		xxstr *xx_next;	/* pointer to next in list */
		queue_t		*xx_rq;				/* read side queue pointer */
		int			xx_timeoutid;		/* id returned from timeout() */
};

/*
 * Linked list of opened stream xxstr structures and other module
 * global data. Protected by the outer perimeter.
 */
static struct xxstr						*xxup = NULL;
static int some_module_global_data;


/*
 * Module Config entry points
 */
int
_init(void)
{
		return (mod_install(&modlinkage));
}
int
_fini(void)
{
		return (mod_remove(&modlinkage));
}
int
_info(struct modinfo *modinfop)
{
		return (mod_info(&modlinkage, modinfop));
}


static int
xxopen(queue_t *rq,dev_t *devp,int flag,int sflag, cred_t *credp)
{
		struct xxstr *xxp;
		/* If this stream already open - we're done. */
		if (rq->q_ptr)
			return (0);
		/* We must be a module */
		if (sflag != MODOPEN)
			return (EINVAL);

		/*
		 * The perimeter flag D_MTOCEXCL implies that the open and
		 * close routines have exclusive access to the module global
		 * data structures.
		 *
		 * Allocate our private per-stream data structure.
		 */
	 	xxp = kmem_alloc(sizeof (struct xxstr),KM_SLEEP);

		/* Point q_ptr at it. */
		rq->q_ptr = WR(rq)->q_ptr = (char *) xxp;

		/* Initialize it. */
		xxp->xx_rq = rq;
		xxp->xx_timeoutid = 0;

		/* Link new entry into the list of active entries. */
		xxp->xx_next = xxup;
		xxup = xxp;

		/* Enable xxput() and xxsrv() procedures on this queue. */
		qprocson(rq);
		/* Return success */
		return (0);
}

static int
xxclose(queue_t,*rq, int flag,cred_t *credp)
{
		struct			xxstr				*xxp;
		struct			xxstr				**prevxxp;

		/* Disable xxput() and xxsrv() procedures on this queue. */
		qprocsoff(rq);
		/* Cancel any pending timeout. */
	 	xxp = (struct xxstr *) rq->q_ptr;
	 	if (xxp->xx_timeoutid != 0) {
	 		(void) quntimeout(WR(rq), xxp->xx_timeoutid);
	 	 	xxp->xx_timeoutid = 0;
	 	}
		/*
		 * D_MTOCEXCL implies that the open and close routines have
		 * exclusive access to the module global data structures.
		 *
		 * Unlink per-stream entry from the active list and free it.
		 */
		for (prevxxp = &xxup; (xxp = *prevxxp) != NULL; 
				prevxxp = &xxp->xx_next) {
			if (xxp == (struct xxstr *) rq->q_ptr)
				break;
		}
		*prevxxp = xxp->xx_next;
		kmem_free (xxp, sizeof (struct xxstr));
		rq->q_ptr = WR(rq)->q_ptr = NULL;
		return (0);
}

static int
xxrput(queue_t, *wq, mblk_t *mp)
{
		struct xxstr	*xxp = (struct xxstr *)wq->q_ptr;
	
		/*
		 * Write your code here. Can read “some_module_global_data”
		 * since we have shared access at the outer perimeter.
		 */
		putnext(wq, mp);
}

/* qwriter callback function for handling M_IOCTL messages */
static void
xxwput_ioctl(queue_t, *wq, mblk_t *mp)
{
		struct xxstr				*xxp = (struct xxstr *)wq->q_ptr;

		/*
		 * Write your code here. Can modify “some_module_global_data”
		 * since we have exclusive access at the outer perimeter.
		 */
		mp->b_datap->db_type = M_IOCNAK;
		qreply(wq, mp);
}

static
xxwput(queue_t *wq, mblk_t *mp)
{
		struct xxstr				*xxp = (struct xxstr *)wq->q_ptr;

		if (mp->b_datap->db_type == M_IOCTL) {
			/* M_IOCTL will modify the module global data */
			qwriter(wq, mp, xxwput_ioctl, PERIM_OUTER);
			return;
		}
		/*
		 * Write your code here. Can read “some_module_global_data”
		 * since we have exclusive access at the outer perimeter.
		 */
		putnext(wq, mp);
}

static
xxwsrv(queue_t wq)
{
		mblk_t			*mp;
		struct xxstr	*xxp= (struct xxstr *) wq->q_ptr;

		while (mp = getq(wq)) {
		/*
		 * Write your code here. Can read “some_module_global_data”
		 * since we have exclusive access at the outer perimeter.
		 */
			freemsg(mp);

			/* for example, start a timeout */
			if (xxp->xx_timeoutid != 0) {
				/* cancel running timeout */
				(void) quntimeout(wq, xxp->xx_timeoutid);
			}
			xxp->xx_timeoutid = qtimeout(wq, xxtick, (char *)xxp, 10);
		}
}

static void
xxtick(arg)
		caddr_t arg;
{
		struct xxstr *xxp = (struct xxstr *)arg;

		xxp->xx_timeoutid = 0;      /* timeout has run */
		/*
		 * Write your code here. Can read “some_module_global_data”
		 * since we have shared access at the outer perimeter.
		 */
}

Chapter 13 STREAMS Multiplex Drivers

This chapter describes how STREAMS multiplexing configurations are created and also discusses multiplexing drivers. A STREAMS multiplexer is a driver with multiple streams connected to it. The primary function of the multiplexing driver is to switch messages among the connected streams. Multiplexer configurations are created from the user level by system calls.

This chapter contains the following information:

• STREAMS Multiplexers

• Connecting And Disconnecting Lower Streams

• Multiplexer Construction Example

• Multiplexing Driver Example

• Persistent Links

• Design Guidelines

STREAMS Multiplexers

STREAMS-related system calls are used to set up the “plumbing,” or stream interconnections, for multiplexing drivers. The subset of these calls that allows a user to connect (and disconnect) streams below a driver is referred to as multiplexing. This type of connection is referred to as a one-to-M, or lower, multiplexer configuration. This configuration must always contain a multiplexing driver, which is recognized by STREAMS as having special characteristics.

Multiple streams can be connected above a driver by open(2) calls. This accommodates the loop-around driver and the driver that handled multiple minor devices in Chapter 9, STREAMS Drivers. There is no difference between the connections to these drivers. Only the functions performed by the driver are different. In the multiplexing case, the driver routes data between multiple streams. In the device driver case, the driver routes data between user processes and associated physical ports. Multiplexing with streams connected above is referred to as an N-to-1, or upper, multiplexer. STREAMS does not provide any facilities beyond open and close to connect or disconnect upper streams for multiplexing.

From the driver's perspective, upper and lower configurations differ only in the way they are initially connected to the driver. The implementation requirements are the same: route the data and handle flow control. All multiplexer drivers require special developer-provided software to perform the multiplexing data routing and to handle flow control. STREAMS does not directly support flow control among multiplexed streams. M-to-N multiplexing configurations are implemented by using both of these mechanisms in a driver.

As discussed in Chapter 9, STREAMS Drivers, the multiple streams that represent minor devices are actually distinct streams in which the driver keeps track of each stream attached to it. The STREAMS subsystem does not recognize any relationship between the streams. The same is true for STREAMS multiplexers of any configuration. The multiplexed streams are distinct and the driver must be implemented to do most of the work.

In addition to upper and lower multiplexers, more complex configurations can be created by connecting streams containing multiplexers to other multiplexer drivers. With such a diversity of needs for multiplexers, providing general-purpose multiplexer drivers is not possible. Rather, STREAMS provides a general purpose multiplexing facility. The facility enables you to set up the intermodule or driver plumbing to create multiplexer configurations of generally unlimited interconnection.

Building a Multiplexer

The example in this section builds a protocol multiplexer with the multiplexing configuration shown in Figure 13–1. To free users from the need to know about the underlying protocol structure, a user-level daemon process is built to maintain the multiplexing configuration. Users can then access the transport protocol directly by opening the transport protocol (TP) driver device node.

An internetworking protocol driver (IP) routes data from a single upper stream to one of two lower streams. This driver supports two STREAMS connections beneath it. These connections are to two distinct networks; one for the IEEE 802.3 standard through the 802.3 driver, and another to the IEEE 802.4 standard through the 802.4 driver. The TP driver multiplexes upper streams over a single stream to the IP driver.

Figure 13–1 Protocol Multiplexer

Example 13–1 shows how this daemon process sets up the protocol multiplexer. The necessary declarations and initialization for the daemon program follow.

Example 13–1 Protocol Daemon

#include <fcntl.h>
#include <stropts.h>
void
main()
{
		int	fd_802_4,
				fd_802_3,
				fd_ip,
				fd_tp;
		/* daemon-ize this process */

		switch (fork()) {
			case 0:
				break;
			case -1:
				perror("fork failed");
				exit(2);
			default:
				exit(0);
	}
	(void)setsid();

This multilevel multiplexed stream configuration is built from the bottom up. The example begins by first constructing the IP multiplexer. This multiplexing device driver is treated like any other software driver. It owns a node in the Solaris file system and is opened just like any other STREAMS device driver.

The first step is to open the multiplexing driver and the 802.4 driver, thus creating separate streams above each driver as shown in Figure 13–2. The stream to the 802.4 driver may now be connected below the multiplexing IP driver using the I_LINK ioctl(2).

Figure 13–2 Streams Before Link

The sequence of instructions to this point is:

	if ((fd_802_4 = open("/dev/802_4", O_RDWR)) < 0) {
			perror("open of /dev/802_4 failed");
			exit(1);
		}
		if ((fd_ip = open("/dev/ip", O_RDWR)) < 0) {
			perror("open of /dev/ip failed");
			exit(2);
		}
		/* now link 802.4 to underside of IP */
		if (ioctl(fd_ip, I_LINK, fd_802_4) < 0) {
			perror("I_LINK ioctl failed");
			exit(3);
		}

I_LINK takes two file descriptors as arguments. The first file descriptor, fd_ip, is the stream connected to the multiplexing driver, and the second file descriptor, fd_802_4, is the stream to be connected below the multiplexer. The complete stream to the 802.4 driver is connected below the IP driver. The stream head's queues of the 802.4 driver are used by the IP driver to manage the lower half of the multiplexer.

I_LINK returns an integer value, muxid, which is used by the multiplexing driver to identify the stream just connected below it. muxid is ignored in the example, but it is useful for dismantling a multiplexer or routing data through the multiplexer. Its significance is discussed in Dismantling a Multiplexer.

The following sequence of system calls continues building the Internetworking Protocol multiplexer (IP):

	if ((fd_802_3 = open("/dev/802_3", O_RDWR)) < 0) {
			perror("open of /dev/802_3 failed");
			exit(4);
		}
		if (ioctl(fd_ip, I_LINK, fd_802_3) < 0) {
			perror("I_LINK ioctl failed");
			exit(5);
		}

The stream above the multiplexing driver used to establish the lower connections is the controlling stream and has special significance when dismantling the multiplexing configuration. This is illustrated in Dismantling a Multiplexer. The stream referenced by fd_ip is the controlling stream for the IP multiplexer.

The order in which the streams in the multiplexing configuration are opened is unimportant. If intermediate modules in the stream are necessary between the IP driver and media drivers, these modules must be added to the streams associated with the media drivers (using I_PUSH) before the media drivers are attached below the multiplexer.

The number of streams that can be linked to a multiplexer is restricted by the design of the particular multiplexer. The manual page describing each driver describes such restrictions (see SunOS Reference Manual, Intro(7)) . However, only one I_LINK operation is allowed for each lower stream; a single stream cannot be linked below two multiplexers simultaneously.

Continuing with the example, the IP driver is now linked below the transport protocol (TP) multiplexing driver. As seen in Figure 13–1, only one link is supported below the transport driver. This link is formed by the following sequence of system calls:

	if ((fd_tp = open("/dev/tp", O_RDWR)) < 0) {
			perror("open of /dev/tp failed");
			exit(6);
		}
		if (ioctl(fd_tp, I_LINK, fd_ip) < 0) {
			perror("I_LINK ioctl failed");
			exit(7);
		}

Because the controlling stream of the IP multiplexer has been linked below the TP multiplexer, the controlling stream for the new multilevel multiplexer configuration is the stream above the TP multiplexer.

At this point, the file descriptors associated with the lower drivers can be closed without affecting the operation of the multiplexer. If these file descriptors are not closed, all subsequent read(2), write(2), ioctl(2), poll(2), getmsg(2), and putmsg(2) calls issued to them fail. That is because I_LINK associates the stream head of each linked stream with the multiplexer, so the user may not access that stream directly for the duration of the link.

The following sequence of system calls completes the daemon example:

	close(fd_802_4);
		close(fd_802_3);
		close(fd_ip);
		/* Hold multiplexer open forever or at least til this process
      is terminated by an external UNIX signal */
		pause();
	}

The transport driver supports several simultaneous streams. These streams are multiplexed over the single stream connected to the IP multiplexer. The mechanism for establishing multiple streams above the transport multiplexer is actually a by-product of the way in which streams are created between a user process and a driver. By opening different minor devices of a STREAMS driver, separate streams will be connected to that driver. The driver must be designed with the intelligence to route data from the single lower stream to the appropriate upper stream.

The daemon process maintains the multiplexed stream configuration through an open stream (the controlling stream) to the transport driver. Meanwhile, other users can access the services of the transport protocol by opening new streams to the transport driver; they are freed from the need for any unnecessary knowledge of the underlying protocol configurations and subnetworks that support the transport service.

Multilevel multiplexing configurations should be assembled from the bottom up. That is because the passing of ioctl(2) through the multiplexer is determined by the nature of the multiplexing driver and cannot generally be relied on.

Dismantling a Multiplexer

Streams connected to a multiplexing driver from above with open(2), can be dismantled by closing each stream with close(2). The mechanism for dismantling streams that have been linked below a multiplexing driver is less obvious, and is described in Disconnecting Lower Streams.

I_UNLINK ioctl(2) disconnects each multiplexer link below a multiplexing driver individually. This command has the form:

ioctl(fd, I_UNLINK, muxid);

where fd is a file descriptor associated with a stream connected to the multiplexing driver from above, and muxid is the identifier that was returned by I_LINK when a driver was linked below the multiplexer. Each lower driver may be disconnected individually in this way, or a special muxid value of MUXID_ALL can be used to disconnect all drivers from the multiplexer simultaneously.

In the multiplexing daemon program, the multiplexer is never explicitly dismantled. That is because all links associated with a multiplexing driver are automatically dismantled when the controlling stream associated with that multiplexer is closed. Because the controlling stream is open to a driver, only the final call of close for that stream will close it. In this case, the daemon is the only process that has opened the controlling stream, so the multiplexing configuration will be dismantled when the daemon exits.

For the automatic dismantling mechanism to work in the multilevel, multiplexed stream configuration, the controlling stream for each multiplexer at each level must be linked under the next higher-level multiplexer. In the example, the controlling stream for the IP driver was linked under the TP driver. This resulted in a single controlling stream for the full, multilevel configuration. Because the multiplexing program relied on closing the controlling stream to dismantle the multiplexed stream configuration instead of using explicit I_UNLINK calls, the muxid values returned by I_LINK could be ignored.

An important side effect of automatic dismantling on the close is that a process cannot build a multiplexing configuration with I_LINK and then exit. exit(2) closes all files associated with the process, including the controlling stream. To keep the configuration intact, the process must exist for the life of that multiplexer. That is the motivation for implementing the multiplexer as a daemon processs, see Multiplexing Driver Example.

If the process uses persistent links through I_PLINK ioctl(2), the multiplexer configuration remains intact after the process exits. These links are described in Persistent Links.

Routing Data Through a Multiplexer

As demonstrated, STREAMS provides a mechanism for building multiplexed stream configurations. However, the criteria by which a multiplexer routes data are driver dependent. For example, the protocol multiplexer might use address information found in a protocol header to determine the subnetwork over which data should be routed. You must define its routing criteria.

One routing option available to the multiplexer is to use the muxid value to determine the stream to which data is routed (remember that each multiplexer link has a muxid). I_LINK passes the muxid value to the driver and returns this value to the user. The driver can therefore specify that the muxid value accompany data routed through it. For example, if a multiplexer routed data from a single upper stream to one of several lower streams (as did the IP driver), the multiplexer can require the user to insert the muxid of the desired lower stream into the first four bytes of each message passed to it. The driver can then match the muxid in each message with the muxid of each lower stream, and route the data accordingly.

Connecting And Disconnecting Lower Streams

Multiple streams are created above a driver/multiplexer by use of the open system call on either different minor device, or on a cloneable device file. Note that any driver that handles more than one minor device is considered an upper multiplexer.

To connect streams below a multiplexer requires additional software in the multiplexer. The main difference between STREAMS lower multiplexers and STREAMS device drivers is that multiplexers are pseudo-devices and multiplexers have two additional qinit structures, pointed to by fields in streamtab(9S): the lower half read-side qinit(9S) and the lower half write-side qinit(9S).

The multiplexer is divided into two parts: the lower half and the upper half. The multiplexer queue structures allocated when the multiplexer was opened use the usual qinit entries from the multiplexer's streamtab(9S). This is the same as any open of the STREAMS device. When a lower stream is linked beneath the multiplexer, the qinit structures at the stream head are substituted by the lower half qinit(9S) structures identified in the streamstab for the multiplexers. Once the linkage is made, the multiplexer switches messages between upper and lower streams. When messages reach the top of the lower stream, they are handled by put and service routines specified in the bottom half of the multiplexer.

Connecting Lower Streams

A lower multiplexer is connected as follows: the initial open to a multiplexing driver creates a stream, as in any other driver. open uses the st_rdinit and st_wrinit elements of the streamtab structure to initialize the driver queues.. At this point, the only distinguishing characteristics of this stream are non-NULL entries in the streamtab(9S) st_muxrinit and st_muxwinit fields.

These fields are ignored by open. Any other stream subsequently opened to this driver will have the same streamtab and thereby the same mux fields.

Next, another file is opened to create a (soon-to-be) lower stream. The driver for the lower stream is typically a device driver This stream has no distinguishing characteristics. It can include any driver compatible with the multiplexer. Any modules required on the lower stream must be pushed onto it now.

Next, this lower stream is connected below the multiplexing driver with an I_LINK ioctl(2) (see streamio(7I)). The stream head points to the stream head routines as its procedures (through its queue). An I_LINK to the upper stream, referencing the lower stream, causes STREAMS to modify the contents of the stream head's queues in the lower stream. The pointers to the stream head routines, and other values, in the stream head's queues are replaced with those contained in the mux fields of the multiplexing driver's streamtab. Changing the stream head routines on the lower stream means that all subsequent messages sent upstream by the lower stream's driver are passed to the put procedure designated in st_muxrinit, the multiplexing driver. The I_LINK also establishes this upper stream as the control stream for this lower stream. STREAMS remembers the relationship between these two streams until the upper stream is closed or the lower stream is unlinked.

Finally, the stream head sends an M_IOCTL message with ioc_cmd set to I_LINK to the multiplexing driver. The M_DATA part of the M_IOCTL contains a linkblk(9S) structure. The multiplexing driver stores information from the linkblk(9S) structure in private storage and returns an M_IOCACK acknowledgement. l_index is returned to the process requesting the I_LINK. This value is used later by the process to disconnect the stream.

An I_LINK is required for each lower stream connected to the driver. Additional upper streams can be connected to the multiplexing driver by open calls. Any message type can be sent from a lower stream to user processes along any of the upper streams. The upper streams provide the only interface between the user processes and the multiplexer.

No direct data structure linkage is established for the linked streams. The read queue's q_next is NULL and the write queue's q_next points to the first entity on the lower stream. Messages flowing upstream from a lower driver (a device driver or another multiplexer) will enter the multiplexing driver put procedure with the queue represented in l_qbot as the queue_t for the put procedure. The multiplexing driver has to route the messages to the appropriate upper (or lower) stream. Similarly, a message coming downstream from user space on any upper stream has to be processed and routed, if required, by the driver.

It is the responsibility of the driver to handle routing of messages between the upper and lower streams, or between any lateral stream that is part of the multiplexer. This operation is not handled by the STREAMS framework.

In general, multiplexing drivers should be implemented so that new streams can be dynamically connected to (and existing streams disconnected from) the driver without interfering with its ongoing operation. The number of streams that can be connected to a multiplexer is implementation dependent.

Disconnecting Lower Streams

Dismantling a lower multiplexer is accomplished by disconnecting (unlinking) the lower streams. Unlinking can be initiated in three ways:

• An I_UNLINK ioctl(2) referencing a specific stream

• An I_UNLINK indicating all lower streams

• The last close of the control stream

As in the link, an unlink sends a linkblk(9S) structure to the driver in an M_IOCTL message. The I_UNLINK call, which unlinks a single stream, uses the l_index value returned in the I_LINK to specify the lower stream to be unlinked. The latter two calls must designate a file corresponding to a control stream, which causes all the lower streams that were previously linked by this control stream to be unlinked. However, the driver sees a series of individual unlinks.

If no open references exist for a lower stream, a subsequent unlink will automatically close the stream. Otherwise, the lower stream must be closed by close(2) following the unlink. STREAMS automatically dismantles all cascaded multiplexers (below other multiplexing streams) if their controlling stream is closed. An I_UNLINK leaves lower, cascaded multiplexing streams intact unless the stream file descriptor was previously closed.

Multiplexer Construction Example

This section describes an example of multiplexer construction and usage. Multiple upper and lower streams interface to the multiplexer driver.

The Ethernet, LAPB, and IEEE 802.2 device drivers terminate links to other nodes. The multiplexer driver is an Internet Protocol (IP) multiplexer that switches data among the various nodes or sends data upstream to users in the system. The net modules typically provide a convergence function that matches the multiplexer driver and device driver interface.

Streams A, B, and C are opened by the process, and modules are pushed as needed. Two upper streams are opened to the IP multiplexer. The rightmost stream represents multiple streams, each connected to a process using the network. The stream second from the right provides a direct path to the multiplexer for supervisory functions. The control stream, leading to a process, sets up and supervises this configuration. It is always directly connected to the IP driver. Although not shown, modules can be pushed on the control stream.

After the streams are opened, the supervisory process typically transfers routing information to the IP drivers (and any other multiplexers above the IP), and initializes the links. As each link becomes operational, its stream is connected below the IP driver. If a more complex multiplexing configuration is required, the IP multiplexer stream with all its connected links can be connected below another multiplexer driver.

Multiplexing Driver Example

This section contains an example of a multiplexing driver that implements an N-to-1 configuration. This configuration might be used for terminal windows, where each transmission to or from the terminal identifies the window. This resembles a typical device driver, with two differences: the device-handling functions are performed by a separate driver, connected as a lower stream, and the device information (that is, relevant user process) is contained in the input data rather than in an interrupt call.

Each upper stream is created by open(2). A single lower stream is opened and then it is linked by use of the multiplexing facility. This lower stream might connect to the TTY driver. The implementation of this example is a foundation for an M-to-N multiplexer.

As in the loop-around driver (Chapter 9, STREAMS Drivers), flow control requires the use of standard and special code because physical connectivity among the streams is broken at the driver. Different approaches are used for flow control on the lower stream, for messages coming upstream from the device driver, and on the upper streams, for messages coming downstream from the user processes.

The code presented here for the multiplexing driver represents a single-threaded, uniprocessor implementation. See Chapter 12, Multithreaded STREAMS for details on multiprocessor and multithreading issues such as locking for data corruption and to prevent race conditions.

Example 13–2 is of multiplexer declarations:

Example 13–2 Multiplexer Declarations

#include <sys/types.h>
#include <sys/param.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/errno.h>
#include <sys/cred.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

static int muxopen (queue_t*, dev_t*, int, int, cred_t*);
static int muxclose (queue_t*, int, cred_t*);
static int muxuwput (queue_t*, mblk_t*);
static int muxlwsrv (queue_t*);
static int muxlrput (queue_t*, mblk_t*);
static int muxuwsrv (queue_t*);

static struct module_info info = {
	0xaabb, "mux", 0, INFPSZ, 512, 128 };

static struct qinit urinit = {							/* upper read */
	NULL, NULL, muxopen, muxclose, NULL, &info, NULL };

static struct qinit uwinit = {							/* upper write */
	muxuwput, muxuwsrv, NULL, NULL, NULL, &info, NULL };

static struct qinit lrinit = { /* lower read */
	muxlrput, NULL, NULL, NULL, NULL, &info, NULL };

static struct qinit lwinit = { /* lower write */
	NULL, muxlwsrv, NULL, NULL, NULL, &info, NULL };

struct streamtab muxinfo = {
	&urinit, &uwinit, &lrinit, &lwinit };

struct mux {
	queue_t *qptr;					/* back pointer to read queue */
	int		bufcid;					/* bufcall return value */
};
extern struct mux mux_mux[];
extern int mux_cnt;     /* max number of muxes */

static queue_t *muxbot; 			/* linked lower queue */
static int muxerr;					/* set if error of hangup on
lower strm */

The four streamtab entries correspond to the upper read, upper write, lower read, and lower write qinit structures. The multiplexing qinit structures replace those in each lower stream head (in this case there is only one) after the I_LINK has concluded successfully. In a multiplexing configuration, the processing performed by the multiplexing driver can be partitioned between the upper and lower queues. There must be an upper-stream write put procedure and lower-stream read put procedure. If the queue procedures of the opposite upper/lower queue are not needed, the queue can be skipped, and the message put to the following queue.

In the example, the upper read-side procedures are not used. The lower-stream read queue put procedure transfers the message directly to the read queue upstream from the multiplexer. There is no lower write put procedure because the upper write put procedure directly feeds the lower write queue downstream from the multiplexer.

The driver uses a private data structure, mux. mux_mux[dev] points back to the opened upper read queue. This is used to route messages coming upstream from the driver to the appropriate upper queue. It is also used to find a free major or minor device for a CLONEOPEN driver open case.

Example 13–3, the upper queue open, contains the canonical driver open code.

Example 13–3 Upper Queue Open

static int
muxopen(queue_t *q, dev_t *devp, int flag,
				 int sflag, cred_t *credp)
{
	struct mux *mux;
	minor_t device;

	if (q->q_ptr)
			return(EBUSY);

	if (sflag == CLONEOPEN) {
			for (device = 0; device < mux_cnt; device++)
				if (mux_mux[device].qptr == 0)
						break;

			*devp=makedevice(getmajor(*devp), device);
	}
	else {
			device = getminor(*devp);
			if (device >= mux_cnt)
				return ENXIO;
	}

	mux = &mux_mux[device];
	mux->qptr = q;
	q->q_ptr = (char *) mux;
	WR(q)->q_ptr = (char *) mux;
	qprocson(q);
	return (0);
}

muxopen checks for a clone or ordinary open call. It initializes q_ptr to point at the mux_mux[] structure.

The core multiplexer processing is as follows: downstream data written to an upper stream is queued on the corresponding upper write message queue if the lower stream is flow controlled. This allows flow control to propagate toward the stream head for each upper stream. A lower write service procedure, rather than a write put procedure, is used so that flow control, coming up from the driver below, may be handled.

On the lower read side, data coming up the lower stream are passed to the lower read put procedure. The procedure routes the data to an upper stream based on the first byte of the message. This byte holds the minor device number of an upper stream. The put procedure handles flow control by testing the upper stream at the first upper read queue beyond the driver.

Upper Write put Procedure Sample

muxuwput, the upper-queue write put procedure, traps ioctl calls, in particular I_LINK and I_UNLINK:

Example 13–4 bufcall Callback Routine

static int
/*
* This is our callback routine used by bufcall() to inform us
* when buffers become available
*/
static void mux_qenable(long ql)
{
		queue_t *q = (queue_t *ql);
		struct mux *mux;

		mux = (struct mux *)(q->q_ptr);
		mux->bufcid = 0;
		qenable(q);
}
muxuwput(queue_t *q, mblk_t *mp)
{
	struct mux *mux;

	mux = (struct mux *)q->q_ptr;
	switch (mp->b_datap->db_type) {
	case M_IOCTL: {
			struct iocblk *iocp;
			struct linkblk *linkp;
			/*
			 * ioctl. Only channel 0 can do ioctls. Two
			 * calls are recognized: LINK, and UNLINK
			 */
			if (mux != mux_mux)
				goto iocnak;

			iocp = (struct iocblk *) mp->b_rptr;
			switch (iocp->ioc_cmd) {
			case I_LINK:
				/*
				 *Link. The data contains a linkblk structure
				 *Remember the bottom queue in muxbot.
				 */
				if (muxbot != NULL)
						goto iocnak;

				linkp=(struct linkblk *) mp->b_cont->b_rptr;
				muxbot = linkp->l_qbot;
				muxerr = 0;

				mp->b_datap->db_type = M_IOCACK;
				iocp->ioc_count = 0;
				qreply(q, mp);
				break;
			case I_UNLINK:
				/*
				 * Unlink. The data contains a linkblk struct.
				 * Should not fail an unlink. Null out muxbot.
				 */
				linkp=(struct linkblk *) mp->b_cont->b_rptr;
				muxbot = NULL;
				mp->b_datap->db_type = M_IOCACK;
				iocp->ioc_count = 0;
				qreply(q, mp);
				break;
			default:
			iocnak:
				/* fail ioctl */
				mp->b_datap->db_type = M_IOCNAK;
				qreply(q, mp);
			}
			break;
	}
	case M_FLUSH:
			if (*mp->b_rptr & FLUSHW)
				flushq(q, FLUSHDATA);
			if (*mp->b_rptr & FLUSHR) {
				*mp->b_rptr &= ~FLUSHW;
				qreply(q, mp);
			} else
				freemsg(mp);
			break;





	case M_DATA:{
			 */
			 * Data. If we have no lower queue --> fail
			 * Otherwise, queue the data and invoke the lower
			 * service procedure.
		mblk_t *bp;
			if (muxerr || muxbot == NULL)
				goto bad;
		if ((bp = allocb(1, BPRI_MED)) == NULL) {
				putbq(q, mp);
				mux->bufcid = bufcall(1, BPRI_MED,
					mux_qenable, (long)q);
				break;
			}
			*bp->b_wptr++ = (struct mux *)q->ptr - mux_mux;
			bp->b_cont = mp;
			putq(q, bp);
			break;
}
	default:
	bad:
			/*
			 * Send an error message upstream.
			 */
			mp->b_datap->db_type = M_ERROR;
			mp->b_rptr = mp->b_wptr = mp->b_datap->db_base;
			*mp->b_wptr++ = EINVAL;
			qreply(q, mp);
 	}
}

First, there is a check to enforce that the stream associated with minor device 0 will be the single, controlling stream. The ioctls are only accepted on this stream. As described previously, a controlling stream is the one that issues the I_LINK. There should be only a single control stream. I_LINK and I_UNLINK include a linkblk structure containing the following fields:

l_qtop is the upper write queue from which the ioctl(2) comes. It always equals q for an I_LINK, and NULL for I_PLINK.

l_qbot is the new lower write queue. It is the former stream head write queue and is where the multiplexer gets and puts its data.

l_index is a unique (system-wide) identifier for the link. It can be used for routing or during selective unlinks. Since the example only supports a single link, l_index is not used.

For I_LINK, l_qbot is saved in muxbot and a positive acknowledgement is generated. From this point on, until an I_UNLINK occurs, data from upper queues will be routed through muxbot. Note that when an I_LINK, is received, the lower stream has already been connected. This enables the driver to send messages downstream to perform any initialization functions. Returning an M_IOCNAK message (negative acknowledgement) in response to an I_LINK causes the lower stream to be disconnected.

The I_UNLINK handling code nulls out muxbot and generates a positive acknowledgement. A negative acknowledgement should not be returned to an I_UNLINK. The stream head ensures that the lower stream is connected to a multiplexer before sending an I_UNLINK M_IOCTL.

Drivers can handle the persistent link requests I_PLINK and I_PUNLINK ioctl(2) in the same manner, except that l_qtop in the linkblk structure passed to the put routine is NULL instead of identifying the controlling stream.

muxuwput handles M_FLUSH messages as a normal driver does, except that there are no messages queued on the upper read queue, so there is no need to call flushq if FLUSHR is set.

M_DATA messages are not placed on the lower write message queue. They are queued on the upper write message queue. When flow control subsides on the lower stream, the lower service procedure, muxlwsrv, is scheduled to start output. This is similar to starting output on a device driver.

Upper Write service Procedure Sample

The following example shows the code for the upper multiplexer write service procedure:

Example 13–5 Upper Multiplexer Write Service Procedure

static int muxuwsrv(queue_t *q)
{
	mblk_t *mp;
	struct mux *muxp;
	muxp = (struct mux *)q->q_ptr;

	if (!muxbot) {
			flushq(q, FLUSHALL);
			return (0);
	}
	if (muxerr) {
			flushq(q, FLUSHALL);
			return (0);
	}
	while (mp = getq(q)) {
			if (canputnext(muxbot))
				putnext(muxbot, mp);
			else {
				putbq(q, mp);
				return(0);
			}
	}
	return (0);
}

As long as there is a stream still linked under the multiplexer and there are no errors, the service procedure will take a message off the queue and send it downstream, if flow control allows.

Lower Write service Procedure

muxlwsrv, the lower (linked) queue write service procedure is scheduled as a result of flow control subsiding downstream (it is back-enabled).

static int muxlwsrv(queue_t *q)
{
	int i;

	for (i = 0; i < mux_cnt; i++)
			if (mux_mux[i].qptr && mux_mux[i].qptr->q_first)
				qenable(mux_mux[i].qptr);
	return (0);
}

muxlwsrv steps through all possible upper queues. If a queue is active and there are messages on the queue, then its upper write service procedure is enabled through qenable.

Lower Read put Procedure

The lower (linked) queue read put procedure is shown in the following example:

Example 13–6 Lower Read put Procedure

static int
muxlrput(queue_t *q, mblk_t *mp)
{
	queue_t *uq;
	int device;

	if(muxerr) {
			freemsg(mp);
			return (0);
	}

	switch(mp->b_datap->db_type) {
	case M_FLUSH:
			/*
			 * Flush queues. NOTE: sense of tests is reversed
			 * since we are acting like a "stream head"
			 */
			if (*mp->b_rptr & FLUSHW) {
				*mp->b_rptr &= ~FLUSHR;
				qreply(q, mp);
			} else
				freemsg(mp);
			break;
	case M_ERROR:
	case M_HANGUP:
			muxerr = 1;
			freemsg(mp);
			break;
	case M_DATA:
			/*
			 * Route message. First byte indicates
			 * device to send to. No flow control.
			 *
			 * Extract and delete device number. If the
			 * leading block is now empty and more blocks
			 * follow, strip the leading block.
			 */
			device = *mp->b_rptr++;

			/* Sanity check. Device must be in range */
			if (device < 0 || device >= mux_cnt) {
				freemsg(mp);
				break;
			}
			/*
			 * If upper stream is open and not backed up,
			 * send the message there, otherwise discard it.
			 */
			uq = mux_mux[device].qptr;
			if (uq != NULL && canputnext(uq))
				putnext(uq, mp);
			else
				freemsg(mp);
			break;
	default:
			freemsg(mp);
	}
	return (0);
}

muxlrput receives messages from the linked stream. In this case, it is acting as a stream head and handles M_FLUSH messages. The code is the reverse of a driver, handling M_FLUSH messages from upstream. There is no need to flush the read queue because no data is ever placed in it.

muxlrput also handles M_ERROR and M_HANGUP messages. If one is received, it locks up the upper streams by setting muxerr.

M_DATA messages are routed by checking the first data byte of the message. This byte contains the minor device of the upper stream. Several checks examine whether:

• The device is in range

• The upper stream is open

• The upper stream is full

This multiplexer does not support flow control on the read side; it is merely a router. If the message passes all checks, it is put to the proper upper queue. Otherwise, the message is discarded.

The upper stream close routine clears the mux entry so this queue will no longer be found. Outstanding bufcalls are not cleared.

/*
* Upper queue close
*/
static int
muxclose(queue_t *q, int flag, cred_t *credp)
{
	struct mux *mux;

	mux = (struct mux *) q->q_ptr;
	qprocsoff(q);
	if (mux->bufcid != 0)
		unbufcall(mux->bufcid);
	mux->bufcid = 0;
	mux->ptr = NULL;
	q->q_ptr = NULL;
	WR(q)->q_ptr = NULL;
	return(0);
}

Persistent Links

Keeping a process running merely to hold the multiplexer configuration together is not always desirable, so, “free standing” links below a multiplexer are needed. A persistent link is such a link. It is similar to a STREAMS multiplexer link except that a process is not needed to hold the links together. After the multiplexer has been set up, the process may close all file descriptors and exit, and the multiplexer remains intact.

With I_LINK and I_UNLINK ioctl(2) the file descriptor associated with the stream above the multiplexer used to set up the lower multiplexer connections must remain open for the duration of the configuration. Closing the file descriptor associated with the controlling stream dismantles the whole multiplexing configuration.

Two ioctl(2)s, I_PLINK and I_PUNLINK, are used to create and remove persistent links that are associated with the stream above the multiplexer. close(2) and I_UNLINK are not able to disconnect the persistent links (see strconf(1) and strchg(1)).

The format of I_PLINK is:

ioctl(fd0, I_PLINK, fd1)

The first file descriptor, fd0, must reference the stream connected to the multiplexing driver and the second file descriptor, fd1, must reference the stream to be connected below the multiplexer. The persistent link can be created as follows:

upper_stream_fd = open("/dev/mux", O_RDWR);
lower_stream_fd = open("/dev/driver", O_RDWR);
muxid = ioctl(upper_stream_fd, I_PLINK, lower_stream_fd);
/*
 * save muxid in a file
 */
exit(0);

The persistent link can still exist even if the file descriptor associated with the upper stream to the multiplexing driver is closed. The I_PLINK ioctl(2) returns an integer value, muxid, that can be used for dismantling the multiplexing configuration. If the process that created the persistent link still exists, it may pass the muxid value to some other process to dismantle the link, if the dismantling is desired, or it can leave the muxid value in a file so that other processes may find it later.

Several users can open the MUX driver and send data to Driver1 since the persistent link to Driver1 remains intact.

The I_PUNLINK ioctl(2) is used to dismantle the persistent link. Its format is:

ioctl(fd0, I_PUNLINK, muxid)

where fd0 is the file descriptor associated with stream connected to the multiplexing driver from above. The muxid is returned by the I_PLINK ioctl(2) for the stream that was connected below the multiplexer. I_PUNLINK removes the persistent link between the multiplexer referenced by fd0 and the stream to the driver designated by the muxid. Each of the bottom persistent links can be disconnected individually. An I_PUNLINK ioctl(2) with the muxid value of MUXID_ALL will remove all persistent links below the multiplexing driver referenced by fd0.

The following code example shows how to dismantle the previously given configuration:

fd = open("/dev/mux", O_RDWR);
/*
 * retrieve muxid from the file
 */
ioctl(fd, I_PUNLINK, muxid);
exit(0);

Do not use the I_PLINK and I_PUNLINK ioctls with I_LINK and I_UNLINK. Any attempt to unlink a regular link with I_PUNLINK or to unlink a persistent link with the I_UNLINK ioctl(2) causes the errno value of EINVAL to be returned.

Because multilevel multiplexing configurations are allowed in STREAMS, persistent links could exist below a multiplexer whose stream is connected to the above multiplexer by regular links. Closing the file descriptor associated with the controlling stream will remove the regular link but not the persistent links below it. On the other hand, regular links are allowed to exist below a multiplexer whose stream is connected to the above multiplexer with persistent links. In this case, the regular links will be removed if the persistent link above is removed and no other references to the lower streams exist.

The construction of cycles is not allowed when creating links. A cycle could be constructed by:

1. Creating a persistent link of multiplexer 2 below multiplexer 1

2. Closing the controlling file descriptor associated with the multiplexer 2

3. Reopening the file descriptor again

4. Linking the multiplexer 1 below the multiplexer 2

Design Guidelines

The following are general multiplexer design guidelines:

• The upper half of the multiplexer acts like the end of the upper stream. The lower half of the multiplexer acts like the head of the lower stream. Service procedures are used for flow control.

• Message routing is based on multiplexer-specific criteria.

• When one stream is being fed by many streams, flow control may have to take place. Then all feeding streams on the other end of the multiplexer have to be enabled when the flow control is relieved.

• When one stream is feeding many streams, flow control may also have to take place. Be careful not to starve other streams when one becomes flow controlled.

• Upper and lower multiplexers share the same perimeter type and concurrency level. (See MT STREAMS Perimeters for information about perimeters.)