xref: /illumos-gate/usr/src/uts/common/sys/strsubr.h (revision a07094369b21309434206d9b3601d162693466fc)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
23 /*	  All Rights Reserved  	*/
24 
25 
26 /*
27  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
28  * Use is subject to license terms.
29  */
30 
31 #ifndef _SYS_STRSUBR_H
32 #define	_SYS_STRSUBR_H
33 
34 #pragma ident	"%Z%%M%	%I%	%E% SMI"	/* SVr4.0 1.17 */
35 
36 /*
37  * WARNING:
38  * Everything in this file is private, belonging to the
39  * STREAMS subsystem.  The only guarantee made about the
40  * contents of this file is that if you include it, your
41  * code will not port to the next release.
42  */
43 #include <sys/stream.h>
44 #include <sys/stropts.h>
45 #include <sys/kstat.h>
46 #include <sys/uio.h>
47 #include <sys/proc.h>
48 
49 #ifdef	__cplusplus
50 extern "C" {
51 #endif
52 
53 /*
54  * In general, the STREAMS locks are disjoint; they are only held
55  * locally, and not simultaneously by a thread.  However, module
56  * code, including at the stream head, requires some locks to be
57  * acquired in order for its safety.
58  *	1. Stream level claim.  This prevents the value of q_next
59  *		from changing while module code is executing.
60  *	2. Queue level claim.  This prevents the value of q_ptr
61  *		from changing while put or service code is executing.
62  *		In addition, it provides for queue single-threading
63  *		for QPAIR and PERQ MT-safe modules.
64  *	3. Stream head lock.  May be held by the stream head module
65  *		to implement a read/write/open/close monitor.
66  *	   Note: that the only types of twisted stream supported are
67  *	   the pipe and transports which have read and write service
68  *	   procedures on both sides of the twist.
69  *	4. Queue lock.  May be acquired by utility routines on
70  *		behalf of a module.
71  */
72 
73 /*
74  * In general, sd_lock protects the consistency of the stdata
75  * structure.  Additionally, it is used with sd_monitor
76  * to implement an open/close monitor.  In particular, it protects
77  * the following fields:
78  *	sd_iocblk
79  *	sd_flag
80  *	sd_copyflag
81  *	sd_iocid
82  *	sd_iocwait
83  *	sd_sidp
84  *	sd_pgidp
85  *	sd_wroff
86  *	sd_tail
87  *	sd_rerror
88  *	sd_werror
89  *	sd_pushcnt
90  *	sd_sigflags
91  *	sd_siglist
92  *	sd_pollist
93  *	sd_mark
94  *	sd_closetime
95  *	sd_wakeq
96  *	sd_uiordq
97  *	sd_uiowrq
98  *	sd_maxblk
99  *
100  * The following fields are modified only by the allocator, which
101  * has exclusive access to them at that time:
102  *	sd_wrq
103  *	sd_strtab
104  *
105  * The following field is protected by the overlying file system
106  * code, guaranteeing single-threading of opens:
107  *	sd_vnode
108  *
109  * Stream-level locks should be acquired before any queue-level locks
110  *	are acquired.
111  *
112  * The stream head write queue lock(sd_wrq) is used to protect the
113  * fields qn_maxpsz and qn_minpsz because freezestr() which is
114  * necessary for strqset() only gets the queue lock.
115  */
116 
117 /*
118  * Function types for the parameterized stream head.
119  * The msgfunc_t takes the parameters:
120  * 	msgfunc(vnode_t *vp, mblk_t *mp, strwakeup_t *wakeups,
121  *		strsigset_t *firstmsgsigs, strsigset_t *allmsgsigs,
122  *		strpollset_t *pollwakeups);
123  * It returns an optional message to be processed by the stream head.
124  *
125  * The parameters for errfunc_t are:
126  *	errfunc(vnode *vp, int ispeek, int *clearerr);
127  * It returns an errno and zero if there was no pending error.
128  */
129 typedef uint_t	strwakeup_t;
130 typedef uint_t	strsigset_t;
131 typedef short	strpollset_t;
132 typedef uintptr_t callbparams_id_t;
133 typedef	mblk_t	*(*msgfunc_t)(vnode_t *, mblk_t *, strwakeup_t *,
134 			strsigset_t *, strsigset_t *, strpollset_t *);
135 typedef int 	(*errfunc_t)(vnode_t *, int, int *);
136 
137 /*
138  * Per stream sd_lock in putnext may be replaced by per cpu stream_putlocks
139  * each living in a separate cache line. putnext/canputnext grabs only one of
140  * stream_putlocks while strlock() (called on behalf of insertq()/removeq())
141  * acquires all stream_putlocks. Normally stream_putlocks are only employed
142  * for highly contended streams that have SQ_CIPUT queues in the critical path
143  * (e.g. NFS/UDP stream).
144  *
145  * stream_putlocks are dynamically assigned to stdata structure through
146  * sd_ciputctrl pointer possibly when a stream is already in use. Since
147  * strlock() uses stream_putlocks only under sd_lock acquiring sd_lock when
148  * assigning stream_putlocks to the stream ensures synchronization with
149  * strlock().
150  *
151  * For lock ordering purposes stream_putlocks are treated as the extension of
152  * sd_lock and are always grabbed right after grabbing sd_lock and released
153  * right before releasing sd_lock except putnext/canputnext where only one of
154  * stream_putlocks locks is used and where it is the first lock to grab.
155  */
156 
157 typedef struct ciputctrl_str {
158 	union _ciput_un {
159 		uchar_t	pad[64];
160 		struct _ciput_str {
161 			kmutex_t	ciput_lck;
162 			ushort_t	ciput_cnt;
163 		} ciput_str;
164 	} ciput_un;
165 } ciputctrl_t;
166 
167 #define	ciputctrl_lock	ciput_un.ciput_str.ciput_lck
168 #define	ciputctrl_count	ciput_un.ciput_str.ciput_cnt
169 
170 /*
171  * Header for a stream: interface to rest of system.
172  */
173 typedef struct stdata {
174 	struct queue	*sd_wrq;	/* write queue */
175 	struct msgb	*sd_iocblk;	/* return block for ioctl */
176 	struct vnode	*sd_vnode;	/* pointer to associated vnode */
177 	struct streamtab *sd_strtab;	/* pointer to streamtab for stream */
178 	uint_t		sd_flag;	/* state/flags */
179 	uint_t		sd_iocid;	/* ioctl id */
180 	struct pid	*sd_sidp;	/* controlling session info */
181 	struct pid	*sd_pgidp;	/* controlling process group info */
182 	ushort_t	sd_unused;	/* UNUSED, retained for binary */
183 					/* compatibility */
184 	ushort_t	sd_wroff;	/* write offset */
185 	ushort_t	sd_tail;	/* reserved space in written mblks */
186 	int		sd_rerror;	/* error to return on read ops */
187 	int		sd_werror;	/* error to return on write ops */
188 	int		sd_pushcnt;	/* number of pushes done on stream */
189 	int		sd_sigflags;	/* logical OR of all siglist events */
190 	struct strsig	*sd_siglist;	/* pid linked list to rcv SIGPOLL sig */
191 	struct pollhead sd_pollist;	/* list of all pollers to wake up */
192 	struct msgb	*sd_mark;	/* "marked" message on read queue */
193 	clock_t		sd_closetime;	/* time to wait to drain q in close */
194 	kmutex_t	sd_lock;	/* protect head consistency */
195 	kcondvar_t	sd_monitor;	/* open/close/push/pop monitor */
196 	kcondvar_t	sd_iocmonitor;	/* ioctl single-threading */
197 	kcondvar_t	sd_refmonitor;	/* sd_refcnt monitor */
198 	ssize_t		sd_qn_minpsz;	/* These two fields are a performance */
199 	ssize_t		sd_qn_maxpsz;	/* enhancements, cache the values in */
200 					/* the stream head so we don't have */
201 					/* to ask the module below the stream */
202 					/* head to get this information. */
203 	struct stdata	*sd_mate;	/* pointer to twisted stream mate */
204 	kthread_id_t	sd_freezer;	/* thread that froze stream */
205 	kmutex_t	sd_reflock;	/* Protects sd_refcnt */
206 	int		sd_refcnt;	/* number of claimstr */
207 	uint_t		sd_wakeq;	/* strwakeq()'s copy of sd_flag */
208 	struct queue	*sd_struiordq;	/* sync barrier struio() read queue */
209 	struct queue	*sd_struiowrq;	/* sync barrier struio() write queue */
210 	char		sd_struiodnak;	/* defer NAK of M_IOCTL by rput() */
211 	struct msgb	*sd_struionak;	/* pointer M_IOCTL mblk(s) to NAK */
212 	caddr_t		sd_t_audit_data; /* For audit purposes only */
213 	ssize_t		sd_maxblk;	/* maximum message block size */
214 	uint_t		sd_rput_opt;	/* options/flags for strrput */
215 	uint_t		sd_wput_opt;	/* options/flags for write/putmsg */
216 	uint_t		sd_read_opt;	/* options/flags for strread */
217 	msgfunc_t	sd_rprotofunc;	/* rput M_*PROTO routine */
218 	msgfunc_t	sd_rputdatafunc; /* read M_DATA routine */
219 	msgfunc_t	sd_rmiscfunc;	/* rput routine (non-data/proto) */
220 	msgfunc_t	sd_wputdatafunc; /* wput M_DATA routine */
221 	errfunc_t	sd_rderrfunc;	/* read side error callback */
222 	errfunc_t	sd_wrerrfunc;	/* write side error callback */
223 	/*
224 	 * support for low contention concurrent putnext.
225 	 */
226 	ciputctrl_t	*sd_ciputctrl;
227 	uint_t		sd_nciputctrl;
228 
229 	int		sd_anchor;	/* position of anchor in stream */
230 	/*
231 	 * Service scheduling at the stream head.
232 	 */
233 	kmutex_t	sd_qlock;
234 	struct queue	*sd_qhead;	/* Head of queues to be serviced. */
235 	struct queue	*sd_qtail;	/* Tail of queues to be serviced. */
236 	void		*sd_servid;	/* Service ID for bckgrnd schedule */
237 	ushort_t	sd_svcflags;	/* Servicing flags */
238 	short		sd_nqueues;	/* Number of queues in the list */
239 	kcondvar_t	sd_qcv;		/* Waiters for qhead to become empty */
240 	kcondvar_t	sd_zcopy_wait;
241 	uint_t		sd_copyflag;	/* copy-related flags */
242 } stdata_t;
243 
244 /*
245  * stdata servicing flags.
246  */
247 #define	STRS_WILLSERVICE	0x01
248 #define	STRS_SCHEDULED		0x02
249 
250 #define	STREAM_NEEDSERVICE(stp)	((stp)->sd_qhead != NULL)
251 
252 /*
253  * stdata flag field defines
254  */
255 #define	IOCWAIT		0x00000001	/* Someone is doing an ioctl */
256 #define	RSLEEP		0x00000002	/* Someone wants to read/recv msg */
257 #define	WSLEEP		0x00000004	/* Someone wants to write */
258 #define	STRPRI		0x00000008	/* An M_PCPROTO is at stream head */
259 #define	STRHUP		0x00000010	/* Device has vanished */
260 #define	STWOPEN		0x00000020	/* waiting for 1st open */
261 #define	STPLEX		0x00000040	/* stream is being multiplexed */
262 #define	STRISTTY	0x00000080	/* stream is a terminal */
263 #define	STRGETINPROG	0x00000100	/* (k)strgetmsg is running */
264 #define	IOCWAITNE	0x00000200	/* STR_NOERROR ioctl running */
265 #define	STRDERR		0x00000400	/* fatal read error from M_ERROR */
266 #define	STWRERR		0x00000800	/* fatal write error from M_ERROR */
267 #define	STRDERRNONPERSIST 0x00001000	/* nonpersistent read errors */
268 #define	STWRERRNONPERSIST 0x00002000	/* nonpersistent write errors */
269 #define	STRCLOSE	0x00004000	/* wait for a close to complete */
270 #define	SNDMREAD	0x00008000	/* used for read notification */
271 #define	OLDNDELAY	0x00010000	/* use old TTY semantics for */
272 					/* NDELAY reads and writes */
273 	/*		0x00020000	   unused */
274 	/*		0x00040000	   unused */
275 #define	STRTOSTOP	0x00080000	/* block background writes */
276 	/*		0x00100000	   unused */
277 	/*		0x00200000	   unused */
278 #define	STRMOUNT	0x00400000	/* stream is mounted */
279 #define	STRNOTATMARK	0x00800000	/* Not at mark (when empty read q) */
280 #define	STRDELIM	0x01000000	/* generate delimited messages */
281 #define	STRATMARK	0x02000000	/* At mark (due to MSGMARKNEXT) */
282 #define	STZCNOTIFY	0x04000000	/* wait for zerocopy mblk to be acked */
283 #define	STRPLUMB	0x08000000	/* push/pop pending */
284 #define	STREOF		0x10000000	/* End-of-file indication */
285 #define	STREOPENFAIL	0x20000000	/* indicates if re-open has failed */
286 #define	STRMATE		0x40000000	/* this stream is a mate */
287 #define	STRHASLINKS	0x80000000	/* I_LINKs under this stream */
288 
289 /*
290  * Copy-related flags (sd_copyflag), set by SO_COPYOPT.
291  */
292 #define	STZCVMSAFE	0x00000001	/* safe to borrow file (segmapped) */
293 					/* pages instead of bcopy */
294 #define	STZCVMUNSAFE	0x00000002	/* unsafe to borrow file pages */
295 #define	STRCOPYCACHED	0x00000004	/* copy should NOT bypass cache */
296 
297 /*
298  * Options and flags for strrput (sd_rput_opt)
299  */
300 #define	SR_POLLIN	0x00000001	/* pollwakeup needed for band0 data */
301 #define	SR_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
302 #define	SR_CONSOL_DATA	0x00000004	/* Consolidate M_DATA onto q_last */
303 #define	SR_IGN_ZEROLEN	0x00000008	/* Ignore zero-length M_DATA */
304 
305 /*
306  * Options and flags for strwrite/strputmsg (sd_wput_opt)
307  */
308 #define	SW_SIGPIPE	0x00000001	/* Send SIGPIPE for write error */
309 #define	SW_RECHECK_ERR	0x00000002	/* Recheck errors in strwrite loop */
310 #define	SW_SNDZERO	0x00000004	/* send 0-length msg down pipe/FIFO */
311 
312 /*
313  * Options and flags for strread (sd_read_opt)
314  */
315 #define	RD_MSGDIS	0x00000001	/* read msg discard */
316 #define	RD_MSGNODIS	0x00000002	/* read msg no discard */
317 #define	RD_PROTDAT	0x00000004	/* read M_[PC]PROTO contents as data */
318 #define	RD_PROTDIS	0x00000008	/* discard M_[PC]PROTO blocks and */
319 					/* retain data blocks */
320 /*
321  * Flags parameter for strsetrputhooks() and strsetwputhooks().
322  * These flags define the interface for setting the above internal
323  * flags in sd_rput_opt and sd_wput_opt.
324  */
325 #define	SH_CONSOL_DATA	0x00000001	/* Consolidate M_DATA onto q_last */
326 #define	SH_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
327 #define	SH_IGN_ZEROLEN	0x00000004	/* Drop zero-length M_DATA */
328 
329 #define	SH_SIGPIPE	0x00000100	/* Send SIGPIPE for write error */
330 #define	SH_RECHECK_ERR	0x00000200	/* Recheck errors in strwrite loop */
331 
332 /*
333  * Each queue points to a sync queue (the inner perimeter) which keeps
334  * track of the number of threads that are inside a given queue (sq_count)
335  * and also is used to implement the asynchronous putnext
336  * (by queuing messages if the queue can not be entered.)
337  *
338  * Messages are queued on sq_head/sq_tail including deferred qwriter(INNER)
339  * messages. The sq_head/sq_tail list is a singly-linked list with
340  * b_queue recording the queue and b_prev recording the function to
341  * be called (either the put procedure or a qwriter callback function.)
342  *
343  * The sq_count counter tracks the number of threads that are
344  * executing inside the perimeter or (in the case of outer perimeters)
345  * have some work queued for them relating to the perimeter. The sq_rmqcount
346  * counter tracks the subset which are in removeq() (usually invoked from
347  * qprocsoff(9F)).
348  *
349  * In addition a module writer can declare that the module has an outer
350  * perimeter (by setting D_MTOUTPERIM) in which case all inner perimeter
351  * syncq's for the module point (through sq_outer) to an outer perimeter
352  * syncq. The outer perimeter consists of the doubly linked list (sq_onext and
353  * sq_oprev) linking all the inner perimeter syncq's with out outer perimeter
354  * syncq. This is used to implement qwriter(OUTER) (an asynchronous way of
355  * getting exclusive access at the outer perimeter) and outer_enter/exit
356  * which are used by the framework to acquire exclusive access to the outer
357  * perimeter during open and close of modules that have set D_MTOUTPERIM.
358  *
359  * In the inner perimeter case sq_save is available for use by machine
360  * dependent code. sq_head/sq_tail are used to queue deferred messages on
361  * the inner perimeter syncqs and to queue become_writer requests on the
362  * outer perimeter syncqs.
363  *
364  * Note: machine dependent optimized versions of putnext may depend
365  * on the order of sq_flags and sq_count (so that they can e.g.
366  * read these two fields in a single load instruction.)
367  *
368  * Per perimeter SQLOCK/sq_count in putnext/put may be replaced by per cpu
369  * sq_putlocks/sq_putcounts each living in a separate cache line. Obviously
370  * sq_putlock[x] protects sq_putcount[x]. putnext/put routine will grab only 1
371  * of sq_putlocks and update only 1 of sq_putcounts. strlock() and many
372  * other routines in strsubr.c and ddi.c will grab all sq_putlocks (as well as
373  * SQLOCK) and figure out the count value as the sum of sq_count and all of
374  * sq_putcounts. The idea is to make critical fast path -- putnext -- much
375  * faster at the expense of much less often used slower path like
376  * strlock(). One known case where entersq/strlock is executed pretty often is
377  * SpecWeb but since IP is SQ_CIOC and socket TCP/IP stream is nextless
378  * there's no need to grab multiple sq_putlocks and look at sq_putcounts. See
379  * strsubr.c for more comments.
380  *
381  * Note regular SQLOCK and sq_count are still used in many routines
382  * (e.g. entersq(), rwnext()) in the same way as before sq_putlocks were
383  * introduced.
384  *
385  * To understand when all sq_putlocks need to be held and all sq_putcounts
386  * need to be added up one needs to look closely at putnext code. Basically if
387  * a routine like e.g. wait_syncq() needs to be sure that perimeter is empty
388  * all sq_putlocks/sq_putcounts need to be held/added up. On the other hand
389  * there's no need to hold all sq_putlocks and count all sq_putcounts in
390  * routines like leavesq()/dropsq() and etc. since the are usually exit
391  * counterparts of entersq/outer_enter() and etc. which have already either
392  * prevented put entry poins from executing or did not care about put
393  * entrypoints. entersq() doesn't need to care about sq_putlocks/sq_putcounts
394  * if the entry point has a shared access since put has the highest degree of
395  * concurrency and such entersq() does not intend to block out put
396  * entrypoints.
397  *
398  * Before sq_putcounts were introduced the standard way to wait for perimeter
399  * to become empty was:
400  *
401  *	mutex_enter(SQLOCK(sq));
402  *	while (sq->sq_count > 0) {
403  *		sq->sq_flags |= SQ_WANTWAKEUP;
404  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
405  *	}
406  *	mutex_exit(SQLOCK(sq));
407  *
408  * The new way is:
409  *
410  * 	mutex_enter(SQLOCK(sq));
411  *	count = sq->sq_count;
412  *	SQ_PUTLOCKS_ENTER(sq);
413  *	SUM_SQ_PUTCOUNTS(sq, count);
414  *	while (count != 0) {
415  *		sq->sq_flags |= SQ_WANTWAKEUP;
416  *		SQ_PUTLOCKS_EXIT(sq);
417  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
418  *		count = sq->sq_count;
419  *		SQ_PUTLOCKS_ENTER(sq);
420  *		SUM_SQ_PUTCOUNTS(sq, count);
421  *	}
422  *	SQ_PUTLOCKS_EXIT(sq);
423  *	mutex_exit(SQLOCK(sq));
424  *
425  * Note that SQ_WANTWAKEUP is set before dropping SQ_PUTLOCKS. This makes sure
426  * putnext won't skip a wakeup.
427  *
428  * sq_putlocks are treated as the extension of SQLOCK for lock ordering
429  * purposes and are always grabbed right after grabbing SQLOCK and released
430  * right before releasing SQLOCK. This also allows dynamic creation of
431  * sq_putlocks while holding SQLOCK (by making sq_ciputctrl non null even when
432  * the stream is already in use). Only in putnext one of sq_putlocks
433  * is grabbed instead of SQLOCK. putnext return path remembers what counter it
434  * incremented and decrements the right counter on its way out.
435  */
436 
437 struct syncq {
438 	kmutex_t	sq_lock;	/* atomic access to syncq */
439 	uint16_t	sq_count;	/* # threads inside */
440 	uint16_t	sq_flags;	/* state and some type info */
441 	/*
442 	 * Distributed syncq scheduling
443 	 *  The list of queue's is handled by sq_head and
444 	 *  sq_tail fields.
445 	 *
446 	 *  The list of events is handled by the sq_evhead and sq_evtail
447 	 *  fields.
448 	 */
449 	queue_t		*sq_head;	/* queue of deferred messages */
450 	queue_t		*sq_tail;	/* queue of deferred messages */
451 	mblk_t		*sq_evhead;	/* Event message on the syncq */
452 	mblk_t		*sq_evtail;
453 	uint_t		sq_nqueues;	/* # of queues on this sq */
454 	/*
455 	 * Concurrency and condition variables
456 	 */
457 	uint16_t	sq_type;	/* type (concurrency) of syncq */
458 	uint16_t	sq_rmqcount;	/* # threads inside removeq() */
459 	kcondvar_t 	sq_wait;	/* block on this sync queue */
460 	kcondvar_t 	sq_exitwait;	/* waiting for thread to leave the */
461 					/* inner perimeter */
462 	/*
463 	 * Handling synchronous callbacks such as qtimeout and qbufcall
464 	 */
465 	ushort_t	sq_callbflags;	/* flags for callback synchronization */
466 	callbparams_id_t sq_cancelid;	/* id of callback being cancelled */
467 	struct callbparams *sq_callbpend;	/* Pending callbacks */
468 
469 	/*
470 	 * Links forming an outer perimeter from one outer syncq and
471 	 * a set of inner sync queues.
472 	 */
473 	struct syncq	*sq_outer;	/* Pointer to outer perimeter */
474 	struct syncq	*sq_onext;	/* Linked list of syncq's making */
475 	struct syncq	*sq_oprev;	/* up the outer perimeter. */
476 	/*
477 	 * support for low contention concurrent putnext.
478 	 */
479 	ciputctrl_t	*sq_ciputctrl;
480 	uint_t		sq_nciputctrl;
481 	/*
482 	 * Counter for the number of threads wanting to become exclusive.
483 	 */
484 	uint_t		sq_needexcl;
485 	/*
486 	 * These two fields are used for scheduling a syncq for
487 	 * background processing. The sq_svcflag is protected by
488 	 * SQLOCK lock.
489 	 */
490 	struct syncq	*sq_next;	/* for syncq scheduling */
491 	void *		sq_servid;
492 	uint_t		sq_servcount;	/* # pending background threads */
493 	uint_t		sq_svcflags;	/* Scheduling flags	*/
494 	clock_t		sq_tstamp;	/* Time when was enabled */
495 	/*
496 	 * Maximum priority of the queues on this syncq.
497 	 */
498 	pri_t		sq_pri;
499 };
500 typedef struct syncq syncq_t;
501 
502 /*
503  * sync queue scheduling flags (for sq_svcflags).
504  */
505 #define	SQ_SERVICE	0x1		/* being serviced */
506 #define	SQ_BGTHREAD	0x2		/* awaiting service by bg thread */
507 #define	SQ_DISABLED	0x4		/* don't put syncq in service list */
508 
509 /*
510  * FASTPUT bit in sd_count/putcount.
511  */
512 #define	SQ_FASTPUT	0x8000
513 #define	SQ_FASTMASK	0x7FFF
514 
515 /*
516  * sync queue state flags
517  */
518 #define	SQ_EXCL		0x0001		/* exclusive access to inner */
519 					/*	perimeter */
520 #define	SQ_BLOCKED	0x0002		/* qprocsoff */
521 #define	SQ_FROZEN	0x0004		/* freezestr */
522 #define	SQ_WRITER	0x0008		/* qwriter(OUTER) pending or running */
523 #define	SQ_MESSAGES	0x0010		/* messages on syncq */
524 #define	SQ_WANTWAKEUP	0x0020		/* do cv_broadcast on sq_wait */
525 #define	SQ_WANTEXWAKEUP	0x0040		/* do cv_broadcast on sq_exitwait */
526 #define	SQ_EVENTS	0x0080		/* Events pending */
527 #define	SQ_QUEUED	(SQ_MESSAGES | SQ_EVENTS)
528 #define	SQ_FLAGMASK	0x00FF
529 
530 /*
531  * Test a queue to see if inner perimeter is exclusive.
532  */
533 #define	PERIM_EXCL(q)	((q)->q_syncq->sq_flags & SQ_EXCL)
534 
535 /*
536  * If any of these flags are set it is not possible for a thread to
537  * enter a put or service procedure. Instead it must either block
538  * or put the message on the syncq.
539  */
540 #define	SQ_GOAWAY	(SQ_EXCL|SQ_BLOCKED|SQ_FROZEN|SQ_WRITER|\
541 			SQ_QUEUED)
542 /*
543  * If any of these flags are set it not possible to drain the syncq
544  */
545 #define	SQ_STAYAWAY	(SQ_BLOCKED|SQ_FROZEN|SQ_WRITER)
546 
547 /*
548  * Flags to trigger syncq tail processing.
549  */
550 #define	SQ_TAIL		(SQ_QUEUED|SQ_WANTWAKEUP|SQ_WANTEXWAKEUP)
551 
552 /*
553  * Syncq types (stored in sq_type)
554  * The SQ_TYPES_IN_FLAGS (ciput) are also stored in sq_flags
555  * for performance reasons. Thus these type values have to be in the low
556  * 16 bits and not conflict with the sq_flags values above.
557  *
558  * Notes:
559  *  - putnext() and put() assume that the put procedures have the highest
560  *    degree of concurrency. Thus if any of the SQ_CI* are set then SQ_CIPUT
561  *    has to be set. This restriction can be lifted by adding code to putnext
562  *    and put that check that sq_count == 0 like entersq does.
563  *  - putnext() and put() does currently not handle !SQ_COPUT
564  *  - In order to implement !SQ_COCB outer_enter has to be fixed so that
565  *    the callback can be cancelled while cv_waiting in outer_enter.
566  *  - If SQ_CISVC needs to be implemented, qprocsoff() needs to wait
567  *    for the currently running services to stop (wait for QINSERVICE
568  *    to go off). disable_svc called from qprcosoff disables only
569  *    services that will be run in future.
570  *
571  * All the SQ_CO flags are set when there is no outer perimeter.
572  */
573 #define	SQ_CIPUT	0x0100		/* Concurrent inner put proc */
574 #define	SQ_CISVC	0x0200		/* Concurrent inner svc proc */
575 #define	SQ_CIOC		0x0400		/* Concurrent inner open/close */
576 #define	SQ_CICB		0x0800		/* Concurrent inner callback */
577 #define	SQ_COPUT	0x1000		/* Concurrent outer put proc */
578 #define	SQ_COSVC	0x2000		/* Concurrent outer svc proc */
579 #define	SQ_COOC		0x4000		/* Concurrent outer open/close */
580 #define	SQ_COCB		0x8000		/* Concurrent outer callback */
581 
582 /* Types also kept in sq_flags for performance */
583 #define	SQ_TYPES_IN_FLAGS	(SQ_CIPUT)
584 
585 #define	SQ_CI		(SQ_CIPUT|SQ_CISVC|SQ_CIOC|SQ_CICB)
586 #define	SQ_CO		(SQ_COPUT|SQ_COSVC|SQ_COOC|SQ_COCB)
587 #define	SQ_TYPEMASK	(SQ_CI|SQ_CO)
588 
589 /*
590  * Flag combinations passed to entersq and leavesq to specify the type
591  * of entry point.
592  */
593 #define	SQ_PUT		(SQ_CIPUT|SQ_COPUT)
594 #define	SQ_SVC		(SQ_CISVC|SQ_COSVC)
595 #define	SQ_OPENCLOSE	(SQ_CIOC|SQ_COOC)
596 #define	SQ_CALLBACK	(SQ_CICB|SQ_COCB)
597 
598 /*
599  * Other syncq types which are not copied into flags.
600  */
601 #define	SQ_PERMOD	0x01		/* Syncq is PERMOD */
602 
603 /*
604  * Asynchronous callback qun*** flag.
605  * The mechanism these flags are used in is one where callbacks enter
606  * the perimeter thanks to framework support. To use this mechanism
607  * the q* and qun* flavors of the callback routines must be used.
608  * e.g. qtimeout and quntimeout. The synchronization provided by the flags
609  * avoids deadlocks between blocking qun* routines and the perimeter
610  * lock.
611  */
612 #define	SQ_CALLB_BYPASSED	0x01		/* bypassed callback fn */
613 
614 /*
615  * Cancel callback mask.
616  * The mask expands as the number of cancelable callback types grows
617  * Note - separate callback flag because different callbacks have
618  * overlapping id space.
619  */
620 #define	SQ_CALLB_CANCEL_MASK	(SQ_CANCEL_TOUT|SQ_CANCEL_BUFCALL)
621 
622 #define	SQ_CANCEL_TOUT		0x02		/* cancel timeout request */
623 #define	SQ_CANCEL_BUFCALL	0x04		/* cancel bufcall request */
624 
625 typedef struct callbparams {
626 	syncq_t		*cbp_sq;
627 	void		(*cbp_func)(void *);
628 	void		*cbp_arg;
629 	callbparams_id_t cbp_id;
630 	uint_t		cbp_flags;
631 	struct callbparams *cbp_next;
632 	size_t		cbp_size;
633 } callbparams_t;
634 
635 typedef struct strbufcall {
636 	void		(*bc_func)(void *);
637 	void		*bc_arg;
638 	size_t		bc_size;
639 	bufcall_id_t	bc_id;
640 	struct strbufcall *bc_next;
641 	kthread_id_t	bc_executor;
642 } strbufcall_t;
643 
644 /*
645  * Structure of list of processes to be sent SIGPOLL/SIGURG signal
646  * on request.  The valid S_* events are defined in stropts.h.
647  */
648 typedef struct strsig {
649 	struct pid	*ss_pidp;	/* pid/pgrp pointer */
650 	pid_t		ss_pid;		/* positive pid, negative pgrp */
651 	int		ss_events;	/* S_* events */
652 	struct strsig	*ss_next;
653 } strsig_t;
654 
655 /*
656  * bufcall list
657  */
658 struct bclist {
659 	strbufcall_t	*bc_head;
660 	strbufcall_t	*bc_tail;
661 };
662 
663 /*
664  * Structure used to track mux links and unlinks.
665  */
666 struct mux_node {
667 	major_t		 mn_imaj;	/* internal major device number */
668 	uint16_t	 mn_indegree;	/* number of incoming edges */
669 	struct mux_node *mn_originp;	/* where we came from during search */
670 	struct mux_edge *mn_startp;	/* where search left off in mn_outp */
671 	struct mux_edge *mn_outp;	/* list of outgoing edges */
672 	uint_t		 mn_flags;	/* see below */
673 };
674 
675 /*
676  * Flags for mux_nodes.
677  */
678 #define	VISITED	1
679 
680 /*
681  * Edge structure - a list of these is hung off the
682  * mux_node to represent the outgoing edges.
683  */
684 struct mux_edge {
685 	struct mux_node	*me_nodep;	/* edge leads to this node */
686 	struct mux_edge	*me_nextp;	/* next edge */
687 	int		 me_muxid;	/* id of link */
688 };
689 
690 /*
691  * Queue info
692  *
693  * The syncq is included here to reduce memory fragmentation
694  * for kernel memory allocators that only allocate in sizes that are
695  * powers of two. If the kernel memory allocator changes this should
696  * be revisited.
697  */
698 typedef struct queinfo {
699 	struct queue	qu_rqueue;	/* read queue - must be first */
700 	struct queue	qu_wqueue;	/* write queue - must be second */
701 	struct syncq	qu_syncq;	/* syncq - must be third */
702 } queinfo_t;
703 
704 /*
705  * Multiplexed streams info
706  */
707 typedef struct linkinfo {
708 	struct linkblk	li_lblk;	/* must be first */
709 	struct file	*li_fpdown;	/* file pointer for lower stream */
710 	struct linkinfo	*li_next;	/* next in list */
711 	struct linkinfo *li_prev;	/* previous in list */
712 } linkinfo_t;
713 
714 /*
715  * List of syncq's used by freeezestr/unfreezestr
716  */
717 typedef struct syncql {
718 	struct syncql	*sql_next;
719 	syncq_t		*sql_sq;
720 } syncql_t;
721 
722 typedef struct sqlist {
723 	syncql_t	*sqlist_head;
724 	size_t		sqlist_size;		/* structure size in bytes */
725 	size_t		sqlist_index;		/* next free entry in array */
726 	syncql_t	sqlist_array[4];	/* 4 or more entries */
727 } sqlist_t;
728 
729 typedef struct perdm {
730 	struct perdm		*dm_next;
731 	syncq_t			*dm_sq;
732 	struct streamtab	*dm_str;
733 	uint_t			dm_ref;
734 } perdm_t;
735 
736 #define	NEED_DM(dmp, qflag) \
737 	(dmp == NULL && (qflag & (QPERMOD | QMTOUTPERIM)))
738 
739 /*
740  * fmodsw_impl_t is used within the kernel. fmodsw is used by
741  * the modules/drivers. The information is copied from fmodsw
742  * defined in the module/driver into the fmodsw_impl_t structure
743  * during the module/driver initialization.
744  */
745 typedef struct fmodsw_impl	fmodsw_impl_t;
746 
747 struct fmodsw_impl {
748 	fmodsw_impl_t		*f_next;
749 	char			f_name[FMNAMESZ + 1];
750 	struct streamtab	*f_str;
751 	uint32_t		f_qflag;
752 	uint32_t		f_sqtype;
753 	perdm_t			*f_dmp;
754 	uint32_t		f_ref;
755 	uint32_t		f_hits;
756 };
757 
758 typedef enum {
759 	FMODSW_HOLD =	0x00000001,
760 	FMODSW_LOAD =	0x00000002
761 } fmodsw_flags_t;
762 
763 typedef struct cdevsw_impl {
764 	struct streamtab	*d_str;
765 	uint32_t		d_qflag;
766 	uint32_t		d_sqtype;
767 	perdm_t			*d_dmp;
768 } cdevsw_impl_t;
769 
770 /*
771  * Enumeration of the types of access that can be requested for a
772  * controlling terminal under job control.
773  */
774 enum jcaccess {
775 	JCREAD,			/* read data on a ctty */
776 	JCWRITE,		/* write data to a ctty */
777 	JCSETP,			/* set ctty parameters */
778 	JCGETP			/* get ctty parameters */
779 };
780 
781 /*
782  * Finding related queues
783  */
784 #define	STREAM(q)	((q)->q_stream)
785 #define	SQ(rq)		((syncq_t *)((rq) + 2))
786 
787 /*
788  * Locking macros
789  */
790 #define	QLOCK(q)	(&(q)->q_lock)
791 #define	SQLOCK(sq)	(&(sq)->sq_lock)
792 
793 #define	STREAM_PUTLOCKS_ENTER(stp) {					       \
794 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
795 		if ((stp)->sd_ciputctrl != NULL) {			       \
796 			int i;						       \
797 			int nlocks = (stp)->sd_nciputctrl;		       \
798 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
799 			for (i = 0; i <= nlocks; i++) {			       \
800 				mutex_enter(&cip[i].ciputctrl_lock);	       \
801 			}						       \
802 		}							       \
803 	}
804 
805 #define	STREAM_PUTLOCKS_EXIT(stp) {					       \
806 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
807 		if ((stp)->sd_ciputctrl != NULL) {			       \
808 			int i;						       \
809 			int nlocks = (stp)->sd_nciputctrl;		       \
810 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
811 			for (i = 0; i <= nlocks; i++) {			       \
812 				mutex_exit(&cip[i].ciputctrl_lock);	       \
813 			}						       \
814 		}							       \
815 	}
816 
817 #define	SQ_PUTLOCKS_ENTER(sq) {						       \
818 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
819 		if ((sq)->sq_ciputctrl != NULL) {			       \
820 			int i;						       \
821 			int nlocks = (sq)->sq_nciputctrl;		       \
822 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
823 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
824 			for (i = 0; i <= nlocks; i++) {			       \
825 				mutex_enter(&cip[i].ciputctrl_lock);	       \
826 			}						       \
827 		}							       \
828 	}
829 
830 #define	SQ_PUTLOCKS_EXIT(sq) {						       \
831 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
832 		if ((sq)->sq_ciputctrl != NULL) {			       \
833 			int i;						       \
834 			int nlocks = (sq)->sq_nciputctrl;		       \
835 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
836 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
837 			for (i = 0; i <= nlocks; i++) {			       \
838 				mutex_exit(&cip[i].ciputctrl_lock);	       \
839 			}						       \
840 		}							       \
841 	}
842 
843 #define	SQ_PUTCOUNT_SETFAST(sq) {					\
844 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				\
845 		if ((sq)->sq_ciputctrl != NULL) {			\
846 			int i;						\
847 			int nlocks = (sq)->sq_nciputctrl;		\
848 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		\
849 			ASSERT((sq)->sq_type & SQ_CIPUT);		\
850 			for (i = 0; i <= nlocks; i++) {			\
851 				mutex_enter(&cip[i].ciputctrl_lock);	\
852 				cip[i].ciputctrl_count |= SQ_FASTPUT;	\
853 				mutex_exit(&cip[i].ciputctrl_lock);	\
854 			}						\
855 		}							\
856 	}
857 
858 #define	SQ_PUTCOUNT_CLRFAST(sq) {					\
859 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				\
860 		if ((sq)->sq_ciputctrl != NULL) {			\
861 			int i;						\
862 			int nlocks = (sq)->sq_nciputctrl;		\
863 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		\
864 			ASSERT((sq)->sq_type & SQ_CIPUT);		\
865 			for (i = 0; i <= nlocks; i++) {			\
866 				mutex_enter(&cip[i].ciputctrl_lock);	\
867 				cip[i].ciputctrl_count &= ~SQ_FASTPUT;	\
868 				mutex_exit(&cip[i].ciputctrl_lock);	\
869 			}						\
870 		}							\
871 	}
872 
873 
874 #ifdef	DEBUG
875 
876 #define	SQ_PUTLOCKS_HELD(sq) {						       \
877 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
878 		if ((sq)->sq_ciputctrl != NULL) {			       \
879 			int i;						       \
880 			int nlocks = (sq)->sq_nciputctrl;		       \
881 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
882 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
883 			for (i = 0; i <= nlocks; i++) {			       \
884 				ASSERT(MUTEX_HELD(&cip[i].ciputctrl_lock));    \
885 			}						       \
886 		}							       \
887 	}
888 
889 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) {				       \
890 		if ((sq)->sq_ciputctrl != NULL) {			       \
891 			int i;						       \
892 			uint_t count = 0;				       \
893 			int ncounts = (sq)->sq_nciputctrl;		       \
894 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
895 			for (i = 0; i <= ncounts; i++) {		       \
896 				count +=				       \
897 				    (((sq)->sq_ciputctrl[i].ciputctrl_count) & \
898 				    SQ_FASTMASK);			       \
899 			}						       \
900 			ASSERT(count == (countcheck));			       \
901 		}							       \
902 	}
903 
904 #define	SUMCHECK_CIPUTCTRL_COUNTS(ciput, nciput, countcheck) {		       \
905 		int i;							       \
906 		uint_t count = 0;					       \
907 		ASSERT((ciput) != NULL);				       \
908 		for (i = 0; i <= (nciput); i++) {			       \
909 			count += (((ciput)[i].ciputctrl_count) &	       \
910 			    SQ_FASTMASK);				       \
911 		}							       \
912 		ASSERT(count == (countcheck));				       \
913 	}
914 
915 #else	/* DEBUG */
916 
917 #define	SQ_PUTLOCKS_HELD(sq)
918 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck)
919 #define	SUMCHECK_CIPUTCTRL_COUNTS(sq, nciput, countcheck)
920 
921 #endif	/* DEBUG */
922 
923 #define	SUM_SQ_PUTCOUNTS(sq, count) {					       \
924 		if ((sq)->sq_ciputctrl != NULL) {			       \
925 			int i;						       \
926 			int ncounts = (sq)->sq_nciputctrl;		       \
927 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
928 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
929 			for (i = 0; i <= ncounts; i++) {		       \
930 				(count) += ((cip[i].ciputctrl_count) &	       \
931 				    SQ_FASTMASK);			       \
932 			}						       \
933 		}							       \
934 	}
935 
936 #define	CLAIM_QNEXT_LOCK(stp)	mutex_enter(&(stp)->sd_lock)
937 #define	RELEASE_QNEXT_LOCK(stp)	mutex_exit(&(stp)->sd_lock)
938 
939 /*
940  * syncq message manipulation macros.
941  */
942 /*
943  * Put a message on the queue syncq.
944  * Assumes QLOCK held.
945  */
946 #define	SQPUT_MP(qp, mp)						\
947 	{								\
948 		qp->q_syncqmsgs++;					\
949 		if (qp->q_sqhead == NULL) {				\
950 			qp->q_sqhead = qp->q_sqtail = mp;		\
951 		} else {						\
952 			qp->q_sqtail->b_next = mp;			\
953 			qp->q_sqtail = mp;				\
954 		}							\
955 	}
956 
957 /*
958  * Miscellaneous parameters and flags.
959  */
960 
961 /*
962  * Default timeout in milliseconds for ioctls and close
963  */
964 #define	STRTIMOUT 15000
965 
966 /*
967  * Flag values for stream io
968  */
969 #define	WRITEWAIT	0x1	/* waiting for write event */
970 #define	READWAIT	0x2	/* waiting for read event */
971 #define	NOINTR		0x4	/* error is not to be set for signal */
972 #define	GETWAIT		0x8	/* waiting for getmsg event */
973 
974 /*
975  * These flags need to be unique for stream io name space
976  * and copy modes name space.  These flags allow strwaitq
977  * and strdoioctl to proceed as if signals or errors on the stream
978  * head have not occurred; i.e. they will be detected by some other
979  * means.
980  * STR_NOSIG does not allow signals to interrupt the call
981  * STR_NOERROR does not allow stream head read, write or hup errors to
982  * affect the call.  When used with strdoioctl(), if a previous ioctl
983  * is pending and times out, STR_NOERROR will cause strdoioctl() to not
984  * return ETIME. If, however, the requested ioctl times out, ETIME
985  * will be returned (use ic_timout instead)
986  * STR_PEEK is used to inform strwaitq that the reader is peeking at data
987  * and that a non-persistent error should not be cleared.
988  * STR_DELAYERR is used to inform strwaitq that it should not check errors
989  * after being awoken since, in addition to an error, there might also be
990  * data queued on the stream head read queue.
991  */
992 #define	STR_NOSIG	0x10	/* Ignore signals during strdoioctl/strwaitq */
993 #define	STR_NOERROR	0x20	/* Ignore errors during strdoioctl/strwaitq */
994 #define	STR_PEEK	0x40	/* Peeking behavior on non-persistent errors */
995 #define	STR_DELAYERR	0x80	/* Do not check errors on return */
996 
997 /*
998  * Copy modes for tty and I_STR ioctls
999  */
1000 #define	U_TO_K 	01			/* User to Kernel */
1001 #define	K_TO_K  02			/* Kernel to Kernel */
1002 
1003 /*
1004  * Mux defines.
1005  */
1006 #define	LINKNORMAL	0x01		/* normal mux link */
1007 #define	LINKPERSIST	0x02		/* persistent mux link */
1008 #define	LINKTYPEMASK	0x03		/* bitmask of all link types */
1009 #define	LINKCLOSE	0x04		/* unlink from strclose */
1010 
1011 /*
1012  * Definitions of Streams macros and function interfaces.
1013  */
1014 
1015 /*
1016  * Obsolete queue scheduling macros. They are not used anymore, but still kept
1017  * here for 3-d party modules and drivers who might still use them.
1018  */
1019 #define	setqsched()
1020 #define	qready()	1
1021 
1022 #ifdef _KERNEL
1023 #define	runqueues()
1024 #define	queuerun()
1025 #endif
1026 
1027 /* compatibility module for style 2 drivers with DR race condition */
1028 #define	DRMODNAME	"drcompat"
1029 
1030 /*
1031  * Macros dealing with mux_nodes.
1032  */
1033 #define	MUX_VISIT(X)	((X)->mn_flags |= VISITED)
1034 #define	MUX_CLEAR(X)	((X)->mn_flags &= (~VISITED)); \
1035 			((X)->mn_originp = NULL)
1036 #define	MUX_DIDVISIT(X)	((X)->mn_flags & VISITED)
1037 
1038 
1039 /*
1040  * Twisted stream macros
1041  */
1042 #define	STRMATED(X)	((X)->sd_flag & STRMATE)
1043 #define	STRLOCKMATES(X)	if (&((X)->sd_lock) > &(((X)->sd_mate)->sd_lock)) { \
1044 				mutex_enter(&((X)->sd_lock)); \
1045 				mutex_enter(&(((X)->sd_mate)->sd_lock));  \
1046 			} else {  \
1047 				mutex_enter(&(((X)->sd_mate)->sd_lock)); \
1048 				mutex_enter(&((X)->sd_lock)); \
1049 			}
1050 #define	STRUNLOCKMATES(X)	mutex_exit(&((X)->sd_lock)); \
1051 			mutex_exit(&(((X)->sd_mate)->sd_lock))
1052 
1053 #ifdef _KERNEL
1054 
1055 extern void strinit(void);
1056 extern int strdoioctl(struct stdata *, struct strioctl *, int, int,
1057     cred_t *, int *);
1058 extern void strsendsig(struct strsig *, int, uchar_t, int);
1059 extern void str_sendsig(vnode_t *, int, uchar_t, int);
1060 extern void strhup(struct stdata *);
1061 extern int qattach(queue_t *, dev_t *, int, cred_t *, fmodsw_impl_t *,
1062     boolean_t);
1063 extern int qreopen(queue_t *, dev_t *, int, cred_t *);
1064 extern void qdetach(queue_t *, int, int, cred_t *, boolean_t);
1065 extern void enterq(queue_t *);
1066 extern void leaveq(queue_t *);
1067 extern int putiocd(mblk_t *, caddr_t, int, cred_t *);
1068 extern int getiocd(mblk_t *, caddr_t, int);
1069 extern struct linkinfo *alloclink(queue_t *, queue_t *, struct file *);
1070 extern void lbfree(struct linkinfo *);
1071 extern int linkcycle(stdata_t *, stdata_t *);
1072 extern struct linkinfo *findlinks(stdata_t *, int, int);
1073 extern queue_t *getendq(queue_t *);
1074 extern int mlink(vnode_t *, int, int, cred_t *, int *, int);
1075 extern int mlink_file(vnode_t *, int, struct file *, cred_t *, int *, int);
1076 extern int munlink(struct stdata *, struct linkinfo *, int, cred_t *, int *);
1077 extern int munlinkall(struct stdata *, int, cred_t *, int *);
1078 extern void mux_addedge(stdata_t *, stdata_t *, int);
1079 extern void mux_rmvedge(stdata_t *, int);
1080 extern int devflg_to_qflag(struct streamtab *, uint32_t, uint32_t *,
1081     uint32_t *);
1082 extern void setq(queue_t *, struct qinit *, struct qinit *, perdm_t *,
1083     uint32_t, uint32_t, boolean_t);
1084 extern perdm_t *hold_dm(struct streamtab *, uint32_t, uint32_t);
1085 extern void rele_dm(perdm_t *);
1086 extern int strmakectl(struct strbuf *, int32_t, int32_t, mblk_t **);
1087 extern int strmakedata(ssize_t *, struct uio *, stdata_t *, int32_t, mblk_t **);
1088 extern int strmakemsg(struct strbuf *, ssize_t *, struct uio *,
1089     struct stdata *, int32_t, mblk_t **);
1090 extern int strgetmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t *,
1091     int *, int, rval_t *);
1092 extern int strputmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t,
1093     int flag, int fmode);
1094 extern int strstartplumb(struct stdata *, int, int);
1095 extern void strendplumb(struct stdata *);
1096 extern int stropen(struct vnode *, dev_t *, int, cred_t *);
1097 extern int strclose(struct vnode *, int, cred_t *);
1098 extern int strpoll(register struct stdata *, short, int, short *,
1099     struct pollhead **);
1100 extern void strclean(struct vnode *);
1101 extern void str_cn_clean();	/* XXX hook for consoles signal cleanup */
1102 extern int strwrite(struct vnode *, struct uio *, cred_t *);
1103 extern int strwrite_common(struct vnode *, struct uio *, cred_t *, int);
1104 extern int kstrwritemp(struct vnode *, mblk_t *, ushort_t);
1105 extern int strread(struct vnode *, struct uio *, cred_t *);
1106 extern int strioctl(struct vnode *, int, intptr_t, int, int, cred_t *, int *);
1107 extern int strrput(queue_t *, mblk_t *);
1108 extern int strrput_nondata(queue_t *, mblk_t *);
1109 extern mblk_t *strrput_proto(vnode_t *, mblk_t *,
1110     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1111 extern mblk_t *strrput_misc(vnode_t *, mblk_t *,
1112     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1113 extern int getiocseqno(void);
1114 extern int strwaitbuf(size_t, int);
1115 extern int strwaitq(stdata_t *, int, ssize_t, int, clock_t, int *);
1116 extern void stralloctty(struct stdata *);
1117 extern void strfreectty(struct stdata *);
1118 extern struct stdata *shalloc(queue_t *);
1119 extern void shfree(struct stdata *s);
1120 extern queue_t *allocq(void);
1121 extern void freeq(queue_t *);
1122 extern qband_t *allocband(void);
1123 extern void freeband(qband_t *);
1124 extern void freebs_enqueue(mblk_t *, dblk_t *);
1125 extern void setqback(queue_t *, unsigned char);
1126 extern int strcopyin(void *, void *, size_t, int);
1127 extern int strcopyout(void *, void *, size_t, int);
1128 extern void strsignal(struct stdata *, int, int32_t);
1129 extern clock_t str_cv_wait(kcondvar_t *, kmutex_t *, clock_t, int);
1130 extern void disable_svc(queue_t *);
1131 extern void remove_runlist(queue_t *);
1132 extern void wait_svc(queue_t *);
1133 extern void backenable(queue_t *, uchar_t);
1134 extern void set_qend(queue_t *);
1135 extern int strgeterr(stdata_t *, int32_t, int);
1136 extern void qenable_locked(queue_t *);
1137 extern mblk_t *getq_noenab(queue_t *);
1138 extern void rmvq_noenab(queue_t *, mblk_t *);
1139 extern void qbackenable(queue_t *, uchar_t);
1140 
1141 extern void strblock(queue_t *);
1142 extern void strunblock(queue_t *);
1143 extern int qclaimed(queue_t *);
1144 extern int straccess(struct stdata *, enum jcaccess);
1145 
1146 extern void entersq(syncq_t *, int);
1147 extern void leavesq(syncq_t *, int);
1148 extern void claimq(queue_t *);
1149 extern void releaseq(queue_t *);
1150 extern void claimstr(queue_t *);
1151 extern void releasestr(queue_t *);
1152 extern void removeq(queue_t *);
1153 extern void insertq(struct stdata *, queue_t *);
1154 extern void drain_syncq(syncq_t *);
1155 extern void qfill_syncq(syncq_t *, queue_t *, mblk_t *);
1156 extern void qdrain_syncq(syncq_t *, queue_t *);
1157 extern int flush_syncq(syncq_t *, queue_t *);
1158 extern void wait_sq_svc(syncq_t *);
1159 
1160 extern void outer_enter(syncq_t *, uint16_t);
1161 extern void outer_exit(syncq_t *);
1162 extern void qwriter_inner(queue_t *, mblk_t *, void (*)());
1163 extern void qwriter_outer(queue_t *, mblk_t *, void (*)());
1164 
1165 extern callbparams_t *callbparams_alloc(syncq_t *, void (*)(void *),
1166     void *, int);
1167 extern void callbparams_free(syncq_t *, callbparams_t *);
1168 extern void callbparams_free_id(syncq_t *, callbparams_id_t, int32_t);
1169 extern void qcallbwrapper(void *);
1170 
1171 extern mblk_t *esballoc_wait(unsigned char *, size_t, uint_t, frtn_t *);
1172 extern mblk_t *esballoca(unsigned char *, size_t, uint_t, frtn_t *);
1173 extern mblk_t *desballoca(unsigned char *, size_t, uint_t, frtn_t *);
1174 extern int do_sendfp(struct stdata *, struct file *, struct cred *);
1175 extern int frozenstr(queue_t *);
1176 extern size_t xmsgsize(mblk_t *);
1177 
1178 extern void putnext_tail(syncq_t *, queue_t *, uint32_t);
1179 extern void stream_willservice(stdata_t *);
1180 extern void stream_runservice(stdata_t *);
1181 
1182 extern void strmate(vnode_t *, vnode_t *);
1183 extern queue_t *strvp2wq(vnode_t *);
1184 extern vnode_t *strq2vp(queue_t *);
1185 extern mblk_t *allocb_wait(size_t, uint_t, uint_t, int *);
1186 extern mblk_t *allocb_cred(size_t, cred_t *);
1187 extern mblk_t *allocb_cred_wait(size_t, uint_t, int *, cred_t *);
1188 extern mblk_t *allocb_tmpl(size_t, const mblk_t *);
1189 extern mblk_t *allocb_tryhard(size_t);
1190 extern void mblk_setcred(mblk_t *, cred_t *);
1191 extern void strpollwakeup(vnode_t *, short);
1192 extern int putnextctl_wait(queue_t *, int);
1193 
1194 extern int kstrputmsg(struct vnode *, mblk_t *, struct uio *, ssize_t,
1195     unsigned char, int, int);
1196 extern int kstrgetmsg(struct vnode *, mblk_t **, struct uio *,
1197     unsigned char *, int *, clock_t, rval_t *);
1198 
1199 extern void strsetrerror(vnode_t *, int, int, errfunc_t);
1200 extern void strsetwerror(vnode_t *, int, int, errfunc_t);
1201 extern void strseteof(vnode_t *, int);
1202 extern void strflushrq(vnode_t *, int);
1203 extern void strsetrputhooks(vnode_t *, uint_t, msgfunc_t, msgfunc_t);
1204 extern void strsetwputhooks(vnode_t *, uint_t, clock_t);
1205 extern void strsetrwputdatahooks(vnode_t *, msgfunc_t, msgfunc_t);
1206 extern int strwaitmark(vnode_t *);
1207 extern void strsignal_nolock(stdata_t *, int, int32_t);
1208 
1209 struct multidata_s;
1210 struct pdesc_s;
1211 extern int hcksum_assoc(mblk_t *, struct multidata_s *, struct pdesc_s  *,
1212     uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, int);
1213 extern void hcksum_retrieve(mblk_t *, struct multidata_s *, struct pdesc_s *,
1214     uint32_t *, uint32_t *, uint32_t *, uint32_t *, uint32_t *);
1215 extern unsigned int bcksum(uchar_t *, int, unsigned int);
1216 extern boolean_t is_vmloaned_mblk(mblk_t *, struct multidata_s *,
1217     struct pdesc_s *);
1218 
1219 extern int fmodsw_register(const char *, struct streamtab *, int);
1220 extern int fmodsw_unregister(const char *);
1221 extern fmodsw_impl_t *fmodsw_find(const char *, fmodsw_flags_t);
1222 extern void fmodsw_rele(fmodsw_impl_t *);
1223 
1224 extern void freemsgchain(mblk_t *);
1225 extern mblk_t *copymsgchain(mblk_t *);
1226 
1227 extern mblk_t *mcopyinuio(struct stdata *, uio_t *, ssize_t, ssize_t, int *);
1228 
1229 /*
1230  * shared or externally configured data structures
1231  */
1232 extern ssize_t strmsgsz;		/* maximum stream message size */
1233 extern ssize_t strctlsz;		/* maximum size of ctl message */
1234 extern int nstrpush;			/* maximum number of pushes allowed */
1235 
1236 /*
1237  * Bufcalls related variables.
1238  */
1239 extern struct bclist strbcalls;		/* List of bufcalls */
1240 extern kmutex_t	strbcall_lock;		/* Protects the list of bufcalls */
1241 extern kcondvar_t strbcall_cv;		/* Signaling when a bufcall is added */
1242 extern kcondvar_t bcall_cv;	/* wait of executing bufcall completes */
1243 
1244 extern frtn_t frnop;
1245 
1246 extern struct kmem_cache *ciputctrl_cache;
1247 extern int n_ciputctrl;
1248 extern int max_n_ciputctrl;
1249 extern int min_n_ciputctrl;
1250 
1251 extern cdevsw_impl_t *devimpl;
1252 #endif	/* _KERNEL */
1253 
1254 /*
1255  * Note: Use of these macros are restricted to kernel/unix and
1256  * intended for the STREAMS framework.
1257  * All modules/drivers should include sys/ddi.h.
1258  *
1259  * Finding related queues
1260  */
1261 #define		_OTHERQ(q)	((q)->q_flag&QREADR? (q)+1: (q)-1)
1262 #define		_WR(q)		((q)->q_flag&QREADR? (q)+1: (q))
1263 #define		_RD(q)		((q)->q_flag&QREADR? (q): (q)-1)
1264 #define		_SAMESTR(q)	(!((q)->q_flag & QEND))
1265 
1266 /*
1267  * These are also declared here for modules/drivers that erroneously
1268  * include strsubr.h after ddi.h or fail to include ddi.h at all.
1269  */
1270 extern struct queue *OTHERQ(queue_t *); /* stream.h */
1271 extern struct queue *RD(queue_t *);
1272 extern struct queue *WR(queue_t *);
1273 extern int SAMESTR(queue_t *);
1274 
1275 /*
1276  * The following hardware checksum related macros are private
1277  * interfaces that are subject to change without notice.
1278  */
1279 #ifdef _KERNEL
1280 #define	DB_CKSUMSTART(mp)	((mp)->b_datap->db_cksumstart)
1281 #define	DB_CKSUMEND(mp)		((mp)->b_datap->db_cksumend)
1282 #define	DB_CKSUMSTUFF(mp)	((mp)->b_datap->db_cksumstuff)
1283 #define	DB_CKSUMFLAGS(mp)	((mp)->b_datap->db_struioun.cksum.flags)
1284 #define	DB_CKSUM16(mp)		((mp)->b_datap->db_cksum16)
1285 #define	DB_CKSUM32(mp)		((mp)->b_datap->db_cksum32)
1286 #endif	/* _KERNEL */
1287 
1288 #ifdef	__cplusplus
1289 }
1290 #endif
1291 
1292 
1293 #endif	/* _SYS_STRSUBR_H */
1294