xref: /illumos-gate/usr/src/uts/common/sys/strsubr.h (revision b424305435881ac456a9343be2898f1f86440f31)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
22 /*	  All Rights Reserved  	*/
23 
24 
25 /*
26  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #ifndef _SYS_STRSUBR_H
31 #define	_SYS_STRSUBR_H
32 
33 /*
34  * WARNING:
35  * Everything in this file is private, belonging to the
36  * STREAMS subsystem.  The only guarantee made about the
37  * contents of this file is that if you include it, your
38  * code will not port to the next release.
39  */
40 #include <sys/stream.h>
41 #include <sys/stropts.h>
42 #include <sys/kstat.h>
43 #include <sys/uio.h>
44 #include <sys/proc.h>
45 #include <sys/netstack.h>
46 #include <sys/modhash.h>
47 #include <sys/sodirect.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_maxblk
97  *	sd_sodirect
98  *
99  * The following fields are modified only by the allocator, which
100  * has exclusive access to them at that time:
101  *	sd_wrq
102  *	sd_strtab
103  *
104  * The following field is protected by the overlying file system
105  * code, guaranteeing single-threading of opens:
106  *	sd_vnode
107  *
108  * Stream-level locks should be acquired before any queue-level locks
109  *	are acquired.
110  *
111  * The stream head write queue lock(sd_wrq) is used to protect the
112  * fields qn_maxpsz and qn_minpsz because freezestr() which is
113  * necessary for strqset() only gets the queue lock.
114  */
115 
116 /*
117  * Function types for the parameterized stream head.
118  * The msgfunc_t takes the parameters:
119  * 	msgfunc(vnode_t *vp, mblk_t *mp, strwakeup_t *wakeups,
120  *		strsigset_t *firstmsgsigs, strsigset_t *allmsgsigs,
121  *		strpollset_t *pollwakeups);
122  * It returns an optional message to be processed by the stream head.
123  *
124  * The parameters for errfunc_t are:
125  *	errfunc(vnode *vp, int ispeek, int *clearerr);
126  * It returns an errno and zero if there was no pending error.
127  */
128 typedef uint_t	strwakeup_t;
129 typedef uint_t	strsigset_t;
130 typedef short	strpollset_t;
131 typedef uintptr_t callbparams_id_t;
132 typedef	mblk_t	*(*msgfunc_t)(vnode_t *, mblk_t *, strwakeup_t *,
133 			strsigset_t *, strsigset_t *, strpollset_t *);
134 typedef int 	(*errfunc_t)(vnode_t *, int, int *);
135 
136 /*
137  * Per stream sd_lock in putnext may be replaced by per cpu stream_putlocks
138  * each living in a separate cache line. putnext/canputnext grabs only one of
139  * stream_putlocks while strlock() (called on behalf of insertq()/removeq())
140  * acquires all stream_putlocks. Normally stream_putlocks are only employed
141  * for highly contended streams that have SQ_CIPUT queues in the critical path
142  * (e.g. NFS/UDP stream).
143  *
144  * stream_putlocks are dynamically assigned to stdata structure through
145  * sd_ciputctrl pointer possibly when a stream is already in use. Since
146  * strlock() uses stream_putlocks only under sd_lock acquiring sd_lock when
147  * assigning stream_putlocks to the stream ensures synchronization with
148  * strlock().
149  *
150  * For lock ordering purposes stream_putlocks are treated as the extension of
151  * sd_lock and are always grabbed right after grabbing sd_lock and released
152  * right before releasing sd_lock except putnext/canputnext where only one of
153  * stream_putlocks locks is used and where it is the first lock to grab.
154  */
155 
156 typedef struct ciputctrl_str {
157 	union _ciput_un {
158 		uchar_t	pad[64];
159 		struct _ciput_str {
160 			kmutex_t	ciput_lck;
161 			ushort_t	ciput_cnt;
162 		} ciput_str;
163 	} ciput_un;
164 } ciputctrl_t;
165 
166 #define	ciputctrl_lock	ciput_un.ciput_str.ciput_lck
167 #define	ciputctrl_count	ciput_un.ciput_str.ciput_cnt
168 
169 /*
170  * Header for a stream: interface to rest of system.
171  *
172  * NOTE: While this is a consolidation-private structure, some unbundled and
173  *       third-party products inappropriately make use of some of the fields.
174  *       As such, please take care to not gratuitously change any offsets of
175  *       existing members.
176  */
177 typedef struct stdata {
178 	struct queue	*sd_wrq;	/* write queue */
179 	struct msgb	*sd_iocblk;	/* return block for ioctl */
180 	struct vnode	*sd_vnode;	/* pointer to associated vnode */
181 	struct streamtab *sd_strtab;	/* pointer to streamtab for stream */
182 	uint_t		sd_flag;	/* state/flags */
183 	uint_t		sd_iocid;	/* ioctl id */
184 	struct pid	*sd_sidp;	/* controlling session info */
185 	struct pid	*sd_pgidp;	/* controlling process group info */
186 	ushort_t	sd_tail;	/* reserved space in written mblks */
187 	ushort_t	sd_wroff;	/* write offset */
188 	int		sd_rerror;	/* error to return on read ops */
189 	int		sd_werror;	/* error to return on write ops */
190 	int		sd_pushcnt;	/* number of pushes done on stream */
191 	int		sd_sigflags;	/* logical OR of all siglist events */
192 	struct strsig	*sd_siglist;	/* pid linked list to rcv SIGPOLL sig */
193 	struct pollhead sd_pollist;	/* list of all pollers to wake up */
194 	struct msgb	*sd_mark;	/* "marked" message on read queue */
195 	clock_t		sd_closetime;	/* time to wait to drain q in close */
196 	kmutex_t	sd_lock;	/* protect head consistency */
197 	kcondvar_t	sd_monitor;	/* open/close/push/pop monitor */
198 	kcondvar_t	sd_iocmonitor;	/* ioctl single-threading */
199 	kcondvar_t	sd_refmonitor;	/* sd_refcnt monitor */
200 	ssize_t		sd_qn_minpsz;	/* These two fields are a performance */
201 	ssize_t		sd_qn_maxpsz;	/* enhancements, cache the values in */
202 					/* the stream head so we don't have */
203 					/* to ask the module below the stream */
204 					/* head to get this information. */
205 	struct stdata	*sd_mate;	/* pointer to twisted stream mate */
206 	kthread_id_t	sd_freezer;	/* thread that froze stream */
207 	kmutex_t	sd_reflock;	/* Protects sd_refcnt */
208 	int		sd_refcnt;	/* number of claimstr */
209 	uint_t		sd_wakeq;	/* strwakeq()'s copy of sd_flag */
210 	struct queue	*sd_struiordq;	/* sync barrier struio() read queue */
211 	struct queue	*sd_struiowrq;	/* sync barrier struio() write queue */
212 	char		sd_struiodnak;	/* defer NAK of M_IOCTL by rput() */
213 	struct msgb	*sd_struionak;	/* pointer M_IOCTL mblk(s) to NAK */
214 	caddr_t		sd_t_audit_data; /* For audit purposes only */
215 	ssize_t		sd_maxblk;	/* maximum message block size */
216 	uint_t		sd_rput_opt;	/* options/flags for strrput */
217 	uint_t		sd_wput_opt;	/* options/flags for write/putmsg */
218 	uint_t		sd_read_opt;	/* options/flags for strread */
219 	msgfunc_t	sd_rprotofunc;	/* rput M_*PROTO routine */
220 	msgfunc_t	sd_rputdatafunc; /* read M_DATA routine */
221 	msgfunc_t	sd_rmiscfunc;	/* rput routine (non-data/proto) */
222 	msgfunc_t	sd_wputdatafunc; /* wput M_DATA routine */
223 	errfunc_t	sd_rderrfunc;	/* read side error callback */
224 	errfunc_t	sd_wrerrfunc;	/* write side error callback */
225 	/*
226 	 * support for low contention concurrent putnext.
227 	 */
228 	ciputctrl_t	*sd_ciputctrl;
229 	uint_t		sd_nciputctrl;
230 
231 	int		sd_anchor;	/* position of anchor in stream */
232 	/*
233 	 * Service scheduling at the stream head.
234 	 */
235 	kmutex_t	sd_qlock;
236 	struct queue	*sd_qhead;	/* Head of queues to be serviced. */
237 	struct queue	*sd_qtail;	/* Tail of queues to be serviced. */
238 	void		*sd_servid;	/* Service ID for bckgrnd schedule */
239 	ushort_t	sd_svcflags;	/* Servicing flags */
240 	short		sd_nqueues;	/* Number of queues in the list */
241 	kcondvar_t	sd_qcv;		/* Waiters for qhead to become empty */
242 	kcondvar_t	sd_zcopy_wait;
243 	uint_t		sd_copyflag;	/* copy-related flags */
244 	zoneid_t	sd_anchorzone;	/* Allow removal from same zone only */
245 	struct msgb	*sd_cmdblk;	/* reply from _I_CMD */
246 	/*
247 	 * Support for socket direct.
248 	 */
249 	sodirect_t	*sd_sodirect;	/* pointer to shared sodirect_t */
250 } stdata_t;
251 
252 /*
253  * stdata servicing flags.
254  */
255 #define	STRS_WILLSERVICE	0x01
256 #define	STRS_SCHEDULED		0x02
257 
258 #define	STREAM_NEEDSERVICE(stp)	((stp)->sd_qhead != NULL)
259 
260 /*
261  * stdata flag field defines
262  */
263 #define	IOCWAIT		0x00000001	/* Someone is doing an ioctl */
264 #define	RSLEEP		0x00000002	/* Someone wants to read/recv msg */
265 #define	WSLEEP		0x00000004	/* Someone wants to write */
266 #define	STRPRI		0x00000008	/* An M_PCPROTO is at stream head */
267 #define	STRHUP		0x00000010	/* Device has vanished */
268 #define	STWOPEN		0x00000020	/* waiting for 1st open */
269 #define	STPLEX		0x00000040	/* stream is being multiplexed */
270 #define	STRISTTY	0x00000080	/* stream is a terminal */
271 #define	STRGETINPROG	0x00000100	/* (k)strgetmsg is running */
272 #define	IOCWAITNE	0x00000200	/* STR_NOERROR ioctl running */
273 #define	STRDERR		0x00000400	/* fatal read error from M_ERROR */
274 #define	STWRERR		0x00000800	/* fatal write error from M_ERROR */
275 #define	STRDERRNONPERSIST 0x00001000	/* nonpersistent read errors */
276 #define	STWRERRNONPERSIST 0x00002000	/* nonpersistent write errors */
277 #define	STRCLOSE	0x00004000	/* wait for a close to complete */
278 #define	SNDMREAD	0x00008000	/* used for read notification */
279 #define	OLDNDELAY	0x00010000	/* use old TTY semantics for */
280 					/* NDELAY reads and writes */
281 	/*		0x00020000	   unused */
282 	/*		0x00040000	   unused */
283 #define	STRTOSTOP	0x00080000	/* block background writes */
284 #define	STRCMDWAIT	0x00100000 	/* someone is doing an _I_CMD */
285 	/*		0x00200000	   unused */
286 #define	STRMOUNT	0x00400000	/* stream is mounted */
287 #define	STRNOTATMARK	0x00800000	/* Not at mark (when empty read q) */
288 #define	STRDELIM	0x01000000	/* generate delimited messages */
289 #define	STRATMARK	0x02000000	/* At mark (due to MSGMARKNEXT) */
290 #define	STZCNOTIFY	0x04000000	/* wait for zerocopy mblk to be acked */
291 #define	STRPLUMB	0x08000000	/* push/pop pending */
292 #define	STREOF		0x10000000	/* End-of-file indication */
293 #define	STREOPENFAIL	0x20000000	/* indicates if re-open has failed */
294 #define	STRMATE		0x40000000	/* this stream is a mate */
295 #define	STRHASLINKS	0x80000000	/* I_LINKs under this stream */
296 
297 /*
298  * Copy-related flags (sd_copyflag), set by SO_COPYOPT.
299  */
300 #define	STZCVMSAFE	0x00000001	/* safe to borrow file (segmapped) */
301 					/* pages instead of bcopy */
302 #define	STZCVMUNSAFE	0x00000002	/* unsafe to borrow file pages */
303 #define	STRCOPYCACHED	0x00000004	/* copy should NOT bypass cache */
304 
305 /*
306  * Options and flags for strrput (sd_rput_opt)
307  */
308 #define	SR_POLLIN	0x00000001	/* pollwakeup needed for band0 data */
309 #define	SR_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
310 #define	SR_CONSOL_DATA	0x00000004	/* Consolidate M_DATA onto q_last */
311 #define	SR_IGN_ZEROLEN	0x00000008	/* Ignore zero-length M_DATA */
312 
313 /*
314  * Options and flags for strwrite/strputmsg (sd_wput_opt)
315  */
316 #define	SW_SIGPIPE	0x00000001	/* Send SIGPIPE for write error */
317 #define	SW_RECHECK_ERR	0x00000002	/* Recheck errors in strwrite loop */
318 #define	SW_SNDZERO	0x00000004	/* send 0-length msg down pipe/FIFO */
319 
320 /*
321  * Options and flags for strread (sd_read_opt)
322  */
323 #define	RD_MSGDIS	0x00000001	/* read msg discard */
324 #define	RD_MSGNODIS	0x00000002	/* read msg no discard */
325 #define	RD_PROTDAT	0x00000004	/* read M_[PC]PROTO contents as data */
326 #define	RD_PROTDIS	0x00000008	/* discard M_[PC]PROTO blocks and */
327 					/* retain data blocks */
328 /*
329  * Flags parameter for strsetrputhooks() and strsetwputhooks().
330  * These flags define the interface for setting the above internal
331  * flags in sd_rput_opt and sd_wput_opt.
332  */
333 #define	SH_CONSOL_DATA	0x00000001	/* Consolidate M_DATA onto q_last */
334 #define	SH_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
335 #define	SH_IGN_ZEROLEN	0x00000004	/* Drop zero-length M_DATA */
336 
337 #define	SH_SIGPIPE	0x00000100	/* Send SIGPIPE for write error */
338 #define	SH_RECHECK_ERR	0x00000200	/* Recheck errors in strwrite loop */
339 
340 /*
341  * Each queue points to a sync queue (the inner perimeter) which keeps
342  * track of the number of threads that are inside a given queue (sq_count)
343  * and also is used to implement the asynchronous putnext
344  * (by queuing messages if the queue can not be entered.)
345  *
346  * Messages are queued on sq_head/sq_tail including deferred qwriter(INNER)
347  * messages. The sq_head/sq_tail list is a singly-linked list with
348  * b_queue recording the queue and b_prev recording the function to
349  * be called (either the put procedure or a qwriter callback function.)
350  *
351  * The sq_count counter tracks the number of threads that are
352  * executing inside the perimeter or (in the case of outer perimeters)
353  * have some work queued for them relating to the perimeter. The sq_rmqcount
354  * counter tracks the subset which are in removeq() (usually invoked from
355  * qprocsoff(9F)).
356  *
357  * In addition a module writer can declare that the module has an outer
358  * perimeter (by setting D_MTOUTPERIM) in which case all inner perimeter
359  * syncq's for the module point (through sq_outer) to an outer perimeter
360  * syncq. The outer perimeter consists of the doubly linked list (sq_onext and
361  * sq_oprev) linking all the inner perimeter syncq's with out outer perimeter
362  * syncq. This is used to implement qwriter(OUTER) (an asynchronous way of
363  * getting exclusive access at the outer perimeter) and outer_enter/exit
364  * which are used by the framework to acquire exclusive access to the outer
365  * perimeter during open and close of modules that have set D_MTOUTPERIM.
366  *
367  * In the inner perimeter case sq_save is available for use by machine
368  * dependent code. sq_head/sq_tail are used to queue deferred messages on
369  * the inner perimeter syncqs and to queue become_writer requests on the
370  * outer perimeter syncqs.
371  *
372  * Note: machine dependent optimized versions of putnext may depend
373  * on the order of sq_flags and sq_count (so that they can e.g.
374  * read these two fields in a single load instruction.)
375  *
376  * Per perimeter SQLOCK/sq_count in putnext/put may be replaced by per cpu
377  * sq_putlocks/sq_putcounts each living in a separate cache line. Obviously
378  * sq_putlock[x] protects sq_putcount[x]. putnext/put routine will grab only 1
379  * of sq_putlocks and update only 1 of sq_putcounts. strlock() and many
380  * other routines in strsubr.c and ddi.c will grab all sq_putlocks (as well as
381  * SQLOCK) and figure out the count value as the sum of sq_count and all of
382  * sq_putcounts. The idea is to make critical fast path -- putnext -- much
383  * faster at the expense of much less often used slower path like
384  * strlock(). One known case where entersq/strlock is executed pretty often is
385  * SpecWeb but since IP is SQ_CIOC and socket TCP/IP stream is nextless
386  * there's no need to grab multiple sq_putlocks and look at sq_putcounts. See
387  * strsubr.c for more comments.
388  *
389  * Note regular SQLOCK and sq_count are still used in many routines
390  * (e.g. entersq(), rwnext()) in the same way as before sq_putlocks were
391  * introduced.
392  *
393  * To understand when all sq_putlocks need to be held and all sq_putcounts
394  * need to be added up one needs to look closely at putnext code. Basically if
395  * a routine like e.g. wait_syncq() needs to be sure that perimeter is empty
396  * all sq_putlocks/sq_putcounts need to be held/added up. On the other hand
397  * there's no need to hold all sq_putlocks and count all sq_putcounts in
398  * routines like leavesq()/dropsq() and etc. since the are usually exit
399  * counterparts of entersq/outer_enter() and etc. which have already either
400  * prevented put entry poins from executing or did not care about put
401  * entrypoints. entersq() doesn't need to care about sq_putlocks/sq_putcounts
402  * if the entry point has a shared access since put has the highest degree of
403  * concurrency and such entersq() does not intend to block out put
404  * entrypoints.
405  *
406  * Before sq_putcounts were introduced the standard way to wait for perimeter
407  * to become empty was:
408  *
409  *	mutex_enter(SQLOCK(sq));
410  *	while (sq->sq_count > 0) {
411  *		sq->sq_flags |= SQ_WANTWAKEUP;
412  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
413  *	}
414  *	mutex_exit(SQLOCK(sq));
415  *
416  * The new way is:
417  *
418  * 	mutex_enter(SQLOCK(sq));
419  *	count = sq->sq_count;
420  *	SQ_PUTLOCKS_ENTER(sq);
421  *	SUM_SQ_PUTCOUNTS(sq, count);
422  *	while (count != 0) {
423  *		sq->sq_flags |= SQ_WANTWAKEUP;
424  *		SQ_PUTLOCKS_EXIT(sq);
425  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
426  *		count = sq->sq_count;
427  *		SQ_PUTLOCKS_ENTER(sq);
428  *		SUM_SQ_PUTCOUNTS(sq, count);
429  *	}
430  *	SQ_PUTLOCKS_EXIT(sq);
431  *	mutex_exit(SQLOCK(sq));
432  *
433  * Note that SQ_WANTWAKEUP is set before dropping SQ_PUTLOCKS. This makes sure
434  * putnext won't skip a wakeup.
435  *
436  * sq_putlocks are treated as the extension of SQLOCK for lock ordering
437  * purposes and are always grabbed right after grabbing SQLOCK and released
438  * right before releasing SQLOCK. This also allows dynamic creation of
439  * sq_putlocks while holding SQLOCK (by making sq_ciputctrl non null even when
440  * the stream is already in use). Only in putnext one of sq_putlocks
441  * is grabbed instead of SQLOCK. putnext return path remembers what counter it
442  * incremented and decrements the right counter on its way out.
443  */
444 
445 struct syncq {
446 	kmutex_t	sq_lock;	/* atomic access to syncq */
447 	uint16_t	sq_count;	/* # threads inside */
448 	uint16_t	sq_flags;	/* state and some type info */
449 	/*
450 	 * Distributed syncq scheduling
451 	 *  The list of queue's is handled by sq_head and
452 	 *  sq_tail fields.
453 	 *
454 	 *  The list of events is handled by the sq_evhead and sq_evtail
455 	 *  fields.
456 	 */
457 	queue_t		*sq_head;	/* queue of deferred messages */
458 	queue_t		*sq_tail;	/* queue of deferred messages */
459 	mblk_t		*sq_evhead;	/* Event message on the syncq */
460 	mblk_t		*sq_evtail;
461 	uint_t		sq_nqueues;	/* # of queues on this sq */
462 	/*
463 	 * Concurrency and condition variables
464 	 */
465 	uint16_t	sq_type;	/* type (concurrency) of syncq */
466 	uint16_t	sq_rmqcount;	/* # threads inside removeq() */
467 	kcondvar_t 	sq_wait;	/* block on this sync queue */
468 	kcondvar_t 	sq_exitwait;	/* waiting for thread to leave the */
469 					/* inner perimeter */
470 	/*
471 	 * Handling synchronous callbacks such as qtimeout and qbufcall
472 	 */
473 	ushort_t	sq_callbflags;	/* flags for callback synchronization */
474 	callbparams_id_t sq_cancelid;	/* id of callback being cancelled */
475 	struct callbparams *sq_callbpend;	/* Pending callbacks */
476 
477 	/*
478 	 * Links forming an outer perimeter from one outer syncq and
479 	 * a set of inner sync queues.
480 	 */
481 	struct syncq	*sq_outer;	/* Pointer to outer perimeter */
482 	struct syncq	*sq_onext;	/* Linked list of syncq's making */
483 	struct syncq	*sq_oprev;	/* up the outer perimeter. */
484 	/*
485 	 * support for low contention concurrent putnext.
486 	 */
487 	ciputctrl_t	*sq_ciputctrl;
488 	uint_t		sq_nciputctrl;
489 	/*
490 	 * Counter for the number of threads wanting to become exclusive.
491 	 */
492 	uint_t		sq_needexcl;
493 	/*
494 	 * These two fields are used for scheduling a syncq for
495 	 * background processing. The sq_svcflag is protected by
496 	 * SQLOCK lock.
497 	 */
498 	struct syncq	*sq_next;	/* for syncq scheduling */
499 	void *		sq_servid;
500 	uint_t		sq_servcount;	/* # pending background threads */
501 	uint_t		sq_svcflags;	/* Scheduling flags	*/
502 	clock_t		sq_tstamp;	/* Time when was enabled */
503 	/*
504 	 * Maximum priority of the queues on this syncq.
505 	 */
506 	pri_t		sq_pri;
507 };
508 typedef struct syncq syncq_t;
509 
510 /*
511  * sync queue scheduling flags (for sq_svcflags).
512  */
513 #define	SQ_SERVICE	0x1		/* being serviced */
514 #define	SQ_BGTHREAD	0x2		/* awaiting service by bg thread */
515 #define	SQ_DISABLED	0x4		/* don't put syncq in service list */
516 
517 /*
518  * FASTPUT bit in sd_count/putcount.
519  */
520 #define	SQ_FASTPUT	0x8000
521 #define	SQ_FASTMASK	0x7FFF
522 
523 /*
524  * sync queue state flags
525  */
526 #define	SQ_EXCL		0x0001		/* exclusive access to inner */
527 					/*	perimeter */
528 #define	SQ_BLOCKED	0x0002		/* qprocsoff */
529 #define	SQ_FROZEN	0x0004		/* freezestr */
530 #define	SQ_WRITER	0x0008		/* qwriter(OUTER) pending or running */
531 #define	SQ_MESSAGES	0x0010		/* messages on syncq */
532 #define	SQ_WANTWAKEUP	0x0020		/* do cv_broadcast on sq_wait */
533 #define	SQ_WANTEXWAKEUP	0x0040		/* do cv_broadcast on sq_exitwait */
534 #define	SQ_EVENTS	0x0080		/* Events pending */
535 #define	SQ_QUEUED	(SQ_MESSAGES | SQ_EVENTS)
536 #define	SQ_FLAGMASK	0x00FF
537 
538 /*
539  * Test a queue to see if inner perimeter is exclusive.
540  */
541 #define	PERIM_EXCL(q)	((q)->q_syncq->sq_flags & SQ_EXCL)
542 
543 /*
544  * If any of these flags are set it is not possible for a thread to
545  * enter a put or service procedure. Instead it must either block
546  * or put the message on the syncq.
547  */
548 #define	SQ_GOAWAY	(SQ_EXCL|SQ_BLOCKED|SQ_FROZEN|SQ_WRITER|\
549 			SQ_QUEUED)
550 /*
551  * If any of these flags are set it not possible to drain the syncq
552  */
553 #define	SQ_STAYAWAY	(SQ_BLOCKED|SQ_FROZEN|SQ_WRITER)
554 
555 /*
556  * Flags to trigger syncq tail processing.
557  */
558 #define	SQ_TAIL		(SQ_QUEUED|SQ_WANTWAKEUP|SQ_WANTEXWAKEUP)
559 
560 /*
561  * Syncq types (stored in sq_type)
562  * The SQ_TYPES_IN_FLAGS (ciput) are also stored in sq_flags
563  * for performance reasons. Thus these type values have to be in the low
564  * 16 bits and not conflict with the sq_flags values above.
565  *
566  * Notes:
567  *  - putnext() and put() assume that the put procedures have the highest
568  *    degree of concurrency. Thus if any of the SQ_CI* are set then SQ_CIPUT
569  *    has to be set. This restriction can be lifted by adding code to putnext
570  *    and put that check that sq_count == 0 like entersq does.
571  *  - putnext() and put() does currently not handle !SQ_COPUT
572  *  - In order to implement !SQ_COCB outer_enter has to be fixed so that
573  *    the callback can be cancelled while cv_waiting in outer_enter.
574  *  - If SQ_CISVC needs to be implemented, qprocsoff() needs to wait
575  *    for the currently running services to stop (wait for QINSERVICE
576  *    to go off). disable_svc called from qprcosoff disables only
577  *    services that will be run in future.
578  *
579  * All the SQ_CO flags are set when there is no outer perimeter.
580  */
581 #define	SQ_CIPUT	0x0100		/* Concurrent inner put proc */
582 #define	SQ_CISVC	0x0200		/* Concurrent inner svc proc */
583 #define	SQ_CIOC		0x0400		/* Concurrent inner open/close */
584 #define	SQ_CICB		0x0800		/* Concurrent inner callback */
585 #define	SQ_COPUT	0x1000		/* Concurrent outer put proc */
586 #define	SQ_COSVC	0x2000		/* Concurrent outer svc proc */
587 #define	SQ_COOC		0x4000		/* Concurrent outer open/close */
588 #define	SQ_COCB		0x8000		/* Concurrent outer callback */
589 
590 /* Types also kept in sq_flags for performance */
591 #define	SQ_TYPES_IN_FLAGS	(SQ_CIPUT)
592 
593 #define	SQ_CI		(SQ_CIPUT|SQ_CISVC|SQ_CIOC|SQ_CICB)
594 #define	SQ_CO		(SQ_COPUT|SQ_COSVC|SQ_COOC|SQ_COCB)
595 #define	SQ_TYPEMASK	(SQ_CI|SQ_CO)
596 
597 /*
598  * Flag combinations passed to entersq and leavesq to specify the type
599  * of entry point.
600  */
601 #define	SQ_PUT		(SQ_CIPUT|SQ_COPUT)
602 #define	SQ_SVC		(SQ_CISVC|SQ_COSVC)
603 #define	SQ_OPENCLOSE	(SQ_CIOC|SQ_COOC)
604 #define	SQ_CALLBACK	(SQ_CICB|SQ_COCB)
605 
606 /*
607  * Other syncq types which are not copied into flags.
608  */
609 #define	SQ_PERMOD	0x01		/* Syncq is PERMOD */
610 
611 /*
612  * Asynchronous callback qun*** flag.
613  * The mechanism these flags are used in is one where callbacks enter
614  * the perimeter thanks to framework support. To use this mechanism
615  * the q* and qun* flavors of the callback routines must be used.
616  * e.g. qtimeout and quntimeout. The synchronization provided by the flags
617  * avoids deadlocks between blocking qun* routines and the perimeter
618  * lock.
619  */
620 #define	SQ_CALLB_BYPASSED	0x01		/* bypassed callback fn */
621 
622 /*
623  * Cancel callback mask.
624  * The mask expands as the number of cancelable callback types grows
625  * Note - separate callback flag because different callbacks have
626  * overlapping id space.
627  */
628 #define	SQ_CALLB_CANCEL_MASK	(SQ_CANCEL_TOUT|SQ_CANCEL_BUFCALL)
629 
630 #define	SQ_CANCEL_TOUT		0x02		/* cancel timeout request */
631 #define	SQ_CANCEL_BUFCALL	0x04		/* cancel bufcall request */
632 
633 typedef struct callbparams {
634 	syncq_t		*cbp_sq;
635 	void		(*cbp_func)(void *);
636 	void		*cbp_arg;
637 	callbparams_id_t cbp_id;
638 	uint_t		cbp_flags;
639 	struct callbparams *cbp_next;
640 	size_t		cbp_size;
641 } callbparams_t;
642 
643 typedef struct strbufcall {
644 	void		(*bc_func)(void *);
645 	void		*bc_arg;
646 	size_t		bc_size;
647 	bufcall_id_t	bc_id;
648 	struct strbufcall *bc_next;
649 	kthread_id_t	bc_executor;
650 } strbufcall_t;
651 
652 /*
653  * Structure of list of processes to be sent SIGPOLL/SIGURG signal
654  * on request.  The valid S_* events are defined in stropts.h.
655  */
656 typedef struct strsig {
657 	struct pid	*ss_pidp;	/* pid/pgrp pointer */
658 	pid_t		ss_pid;		/* positive pid, negative pgrp */
659 	int		ss_events;	/* S_* events */
660 	struct strsig	*ss_next;
661 } strsig_t;
662 
663 /*
664  * bufcall list
665  */
666 struct bclist {
667 	strbufcall_t	*bc_head;
668 	strbufcall_t	*bc_tail;
669 };
670 
671 /*
672  * Structure used to track mux links and unlinks.
673  */
674 struct mux_node {
675 	major_t		 mn_imaj;	/* internal major device number */
676 	uint16_t	 mn_indegree;	/* number of incoming edges */
677 	struct mux_node *mn_originp;	/* where we came from during search */
678 	struct mux_edge *mn_startp;	/* where search left off in mn_outp */
679 	struct mux_edge *mn_outp;	/* list of outgoing edges */
680 	uint_t		 mn_flags;	/* see below */
681 };
682 
683 /*
684  * Flags for mux_nodes.
685  */
686 #define	VISITED	1
687 
688 /*
689  * Edge structure - a list of these is hung off the
690  * mux_node to represent the outgoing edges.
691  */
692 struct mux_edge {
693 	struct mux_node	*me_nodep;	/* edge leads to this node */
694 	struct mux_edge	*me_nextp;	/* next edge */
695 	int		 me_muxid;	/* id of link */
696 	dev_t		 me_dev;	/* dev_t - used for kernel PUNLINK */
697 };
698 
699 /*
700  * Queue info
701  *
702  * The syncq is included here to reduce memory fragmentation
703  * for kernel memory allocators that only allocate in sizes that are
704  * powers of two. If the kernel memory allocator changes this should
705  * be revisited.
706  */
707 typedef struct queinfo {
708 	struct queue	qu_rqueue;	/* read queue - must be first */
709 	struct queue	qu_wqueue;	/* write queue - must be second */
710 	struct syncq	qu_syncq;	/* syncq - must be third */
711 } queinfo_t;
712 
713 /*
714  * Multiplexed streams info
715  */
716 typedef struct linkinfo {
717 	struct linkblk	li_lblk;	/* must be first */
718 	struct file	*li_fpdown;	/* file pointer for lower stream */
719 	struct linkinfo	*li_next;	/* next in list */
720 	struct linkinfo *li_prev;	/* previous in list */
721 } linkinfo_t;
722 
723 /*
724  * List of syncq's used by freeezestr/unfreezestr
725  */
726 typedef struct syncql {
727 	struct syncql	*sql_next;
728 	syncq_t		*sql_sq;
729 } syncql_t;
730 
731 typedef struct sqlist {
732 	syncql_t	*sqlist_head;
733 	size_t		sqlist_size;		/* structure size in bytes */
734 	size_t		sqlist_index;		/* next free entry in array */
735 	syncql_t	sqlist_array[4];	/* 4 or more entries */
736 } sqlist_t;
737 
738 typedef struct perdm {
739 	struct perdm		*dm_next;
740 	syncq_t			*dm_sq;
741 	struct streamtab	*dm_str;
742 	uint_t			dm_ref;
743 } perdm_t;
744 
745 #define	NEED_DM(dmp, qflag) \
746 	(dmp == NULL && (qflag & (QPERMOD | QMTOUTPERIM)))
747 
748 /*
749  * fmodsw_impl_t is used within the kernel. fmodsw is used by
750  * the modules/drivers. The information is copied from fmodsw
751  * defined in the module/driver into the fmodsw_impl_t structure
752  * during the module/driver initialization.
753  */
754 typedef struct fmodsw_impl	fmodsw_impl_t;
755 
756 struct fmodsw_impl {
757 	fmodsw_impl_t		*f_next;
758 	char			f_name[FMNAMESZ + 1];
759 	struct streamtab	*f_str;
760 	uint32_t		f_qflag;
761 	uint32_t		f_sqtype;
762 	perdm_t			*f_dmp;
763 	uint32_t		f_ref;
764 	uint32_t		f_hits;
765 };
766 
767 typedef enum {
768 	FMODSW_HOLD =	0x00000001,
769 	FMODSW_LOAD =	0x00000002
770 } fmodsw_flags_t;
771 
772 typedef struct cdevsw_impl {
773 	struct streamtab	*d_str;
774 	uint32_t		d_qflag;
775 	uint32_t		d_sqtype;
776 	perdm_t			*d_dmp;
777 } cdevsw_impl_t;
778 
779 /*
780  * Enumeration of the types of access that can be requested for a
781  * controlling terminal under job control.
782  */
783 enum jcaccess {
784 	JCREAD,			/* read data on a ctty */
785 	JCWRITE,		/* write data to a ctty */
786 	JCSETP,			/* set ctty parameters */
787 	JCGETP			/* get ctty parameters */
788 };
789 
790 struct str_stack {
791 	netstack_t	*ss_netstack;	/* Common netstack */
792 
793 	kmutex_t	ss_sad_lock;	/* autopush lock */
794 	mod_hash_t	*ss_sad_hash;
795 	size_t		ss_sad_hash_nchains;
796 	struct saddev	*ss_saddev;	/* sad device array */
797 	int		ss_sadcnt;	/* number of sad devices */
798 
799 	int		ss_devcnt;	/* number of mux_nodes */
800 	struct mux_node	*ss_mux_nodes;	/* mux info for cycle checking */
801 };
802 typedef struct str_stack str_stack_t;
803 
804 /*
805  * Finding related queues
806  */
807 #define	STREAM(q)	((q)->q_stream)
808 #define	SQ(rq)		((syncq_t *)((rq) + 2))
809 
810 /*
811  * Locking macros
812  */
813 #define	QLOCK(q)	(&(q)->q_lock)
814 #define	SQLOCK(sq)	(&(sq)->sq_lock)
815 
816 #define	STREAM_PUTLOCKS_ENTER(stp) {					       \
817 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
818 		if ((stp)->sd_ciputctrl != NULL) {			       \
819 			int i;						       \
820 			int nlocks = (stp)->sd_nciputctrl;		       \
821 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
822 			for (i = 0; i <= nlocks; i++) {			       \
823 				mutex_enter(&cip[i].ciputctrl_lock);	       \
824 			}						       \
825 		}							       \
826 	}
827 
828 #define	STREAM_PUTLOCKS_EXIT(stp) {					       \
829 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
830 		if ((stp)->sd_ciputctrl != NULL) {			       \
831 			int i;						       \
832 			int nlocks = (stp)->sd_nciputctrl;		       \
833 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
834 			for (i = 0; i <= nlocks; i++) {			       \
835 				mutex_exit(&cip[i].ciputctrl_lock);	       \
836 			}						       \
837 		}							       \
838 	}
839 
840 #define	SQ_PUTLOCKS_ENTER(sq) {						       \
841 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
842 		if ((sq)->sq_ciputctrl != NULL) {			       \
843 			int i;						       \
844 			int nlocks = (sq)->sq_nciputctrl;		       \
845 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
846 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
847 			for (i = 0; i <= nlocks; i++) {			       \
848 				mutex_enter(&cip[i].ciputctrl_lock);	       \
849 			}						       \
850 		}							       \
851 	}
852 
853 #define	SQ_PUTLOCKS_EXIT(sq) {						       \
854 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
855 		if ((sq)->sq_ciputctrl != NULL) {			       \
856 			int i;						       \
857 			int nlocks = (sq)->sq_nciputctrl;		       \
858 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
859 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
860 			for (i = 0; i <= nlocks; i++) {			       \
861 				mutex_exit(&cip[i].ciputctrl_lock);	       \
862 			}						       \
863 		}							       \
864 	}
865 
866 #define	SQ_PUTCOUNT_SETFAST(sq) {					\
867 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				\
868 		if ((sq)->sq_ciputctrl != NULL) {			\
869 			int i;						\
870 			int nlocks = (sq)->sq_nciputctrl;		\
871 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		\
872 			ASSERT((sq)->sq_type & SQ_CIPUT);		\
873 			for (i = 0; i <= nlocks; i++) {			\
874 				mutex_enter(&cip[i].ciputctrl_lock);	\
875 				cip[i].ciputctrl_count |= SQ_FASTPUT;	\
876 				mutex_exit(&cip[i].ciputctrl_lock);	\
877 			}						\
878 		}							\
879 	}
880 
881 #define	SQ_PUTCOUNT_CLRFAST(sq) {					\
882 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				\
883 		if ((sq)->sq_ciputctrl != NULL) {			\
884 			int i;						\
885 			int nlocks = (sq)->sq_nciputctrl;		\
886 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		\
887 			ASSERT((sq)->sq_type & SQ_CIPUT);		\
888 			for (i = 0; i <= nlocks; i++) {			\
889 				mutex_enter(&cip[i].ciputctrl_lock);	\
890 				cip[i].ciputctrl_count &= ~SQ_FASTPUT;	\
891 				mutex_exit(&cip[i].ciputctrl_lock);	\
892 			}						\
893 		}							\
894 	}
895 
896 
897 #ifdef	DEBUG
898 
899 #define	SQ_PUTLOCKS_HELD(sq) {						       \
900 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
901 		if ((sq)->sq_ciputctrl != NULL) {			       \
902 			int i;						       \
903 			int nlocks = (sq)->sq_nciputctrl;		       \
904 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
905 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
906 			for (i = 0; i <= nlocks; i++) {			       \
907 				ASSERT(MUTEX_HELD(&cip[i].ciputctrl_lock));    \
908 			}						       \
909 		}							       \
910 	}
911 
912 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) {				       \
913 		if ((sq)->sq_ciputctrl != NULL) {			       \
914 			int i;						       \
915 			uint_t count = 0;				       \
916 			int ncounts = (sq)->sq_nciputctrl;		       \
917 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
918 			for (i = 0; i <= ncounts; i++) {		       \
919 				count +=				       \
920 				    (((sq)->sq_ciputctrl[i].ciputctrl_count) & \
921 				    SQ_FASTMASK);			       \
922 			}						       \
923 			ASSERT(count == (countcheck));			       \
924 		}							       \
925 	}
926 
927 #define	SUMCHECK_CIPUTCTRL_COUNTS(ciput, nciput, countcheck) {		       \
928 		int i;							       \
929 		uint_t count = 0;					       \
930 		ASSERT((ciput) != NULL);				       \
931 		for (i = 0; i <= (nciput); i++) {			       \
932 			count += (((ciput)[i].ciputctrl_count) &	       \
933 			    SQ_FASTMASK);				       \
934 		}							       \
935 		ASSERT(count == (countcheck));				       \
936 	}
937 
938 #else	/* DEBUG */
939 
940 #define	SQ_PUTLOCKS_HELD(sq)
941 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck)
942 #define	SUMCHECK_CIPUTCTRL_COUNTS(sq, nciput, countcheck)
943 
944 #endif	/* DEBUG */
945 
946 #define	SUM_SQ_PUTCOUNTS(sq, count) {					       \
947 		if ((sq)->sq_ciputctrl != NULL) {			       \
948 			int i;						       \
949 			int ncounts = (sq)->sq_nciputctrl;		       \
950 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
951 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
952 			for (i = 0; i <= ncounts; i++) {		       \
953 				(count) += ((cip[i].ciputctrl_count) &	       \
954 				    SQ_FASTMASK);			       \
955 			}						       \
956 		}							       \
957 	}
958 
959 #define	CLAIM_QNEXT_LOCK(stp)	mutex_enter(&(stp)->sd_lock)
960 #define	RELEASE_QNEXT_LOCK(stp)	mutex_exit(&(stp)->sd_lock)
961 
962 /*
963  * syncq message manipulation macros.
964  */
965 /*
966  * Put a message on the queue syncq.
967  * Assumes QLOCK held.
968  */
969 #define	SQPUT_MP(qp, mp)						\
970 	{								\
971 		qp->q_syncqmsgs++;					\
972 		if (qp->q_sqhead == NULL) {				\
973 			qp->q_sqhead = qp->q_sqtail = mp;		\
974 		} else {						\
975 			qp->q_sqtail->b_next = mp;			\
976 			qp->q_sqtail = mp;				\
977 		}							\
978 		set_qfull(qp);						\
979 	}
980 
981 /*
982  * Miscellaneous parameters and flags.
983  */
984 
985 /*
986  * Default timeout in milliseconds for ioctls and close
987  */
988 #define	STRTIMOUT 15000
989 
990 /*
991  * Flag values for stream io
992  */
993 #define	WRITEWAIT	0x1	/* waiting for write event */
994 #define	READWAIT	0x2	/* waiting for read event */
995 #define	NOINTR		0x4	/* error is not to be set for signal */
996 #define	GETWAIT		0x8	/* waiting for getmsg event */
997 
998 /*
999  * These flags need to be unique for stream io name space
1000  * and copy modes name space.  These flags allow strwaitq
1001  * and strdoioctl to proceed as if signals or errors on the stream
1002  * head have not occurred; i.e. they will be detected by some other
1003  * means.
1004  * STR_NOSIG does not allow signals to interrupt the call
1005  * STR_NOERROR does not allow stream head read, write or hup errors to
1006  * affect the call.  When used with strdoioctl(), if a previous ioctl
1007  * is pending and times out, STR_NOERROR will cause strdoioctl() to not
1008  * return ETIME. If, however, the requested ioctl times out, ETIME
1009  * will be returned (use ic_timout instead)
1010  * STR_PEEK is used to inform strwaitq that the reader is peeking at data
1011  * and that a non-persistent error should not be cleared.
1012  * STR_DELAYERR is used to inform strwaitq that it should not check errors
1013  * after being awoken since, in addition to an error, there might also be
1014  * data queued on the stream head read queue.
1015  */
1016 #define	STR_NOSIG	0x10	/* Ignore signals during strdoioctl/strwaitq */
1017 #define	STR_NOERROR	0x20	/* Ignore errors during strdoioctl/strwaitq */
1018 #define	STR_PEEK	0x40	/* Peeking behavior on non-persistent errors */
1019 #define	STR_DELAYERR	0x80	/* Do not check errors on return */
1020 
1021 /*
1022  * Copy modes for tty and I_STR ioctls
1023  */
1024 #define	U_TO_K 	01			/* User to Kernel */
1025 #define	K_TO_K  02			/* Kernel to Kernel */
1026 
1027 /*
1028  * Mux defines.
1029  */
1030 #define	LINKNORMAL	0x01		/* normal mux link */
1031 #define	LINKPERSIST	0x02		/* persistent mux link */
1032 #define	LINKTYPEMASK	0x03		/* bitmask of all link types */
1033 #define	LINKCLOSE	0x04		/* unlink from strclose */
1034 
1035 /*
1036  * Definitions of Streams macros and function interfaces.
1037  */
1038 
1039 /*
1040  * Obsolete queue scheduling macros. They are not used anymore, but still kept
1041  * here for 3-d party modules and drivers who might still use them.
1042  */
1043 #define	setqsched()
1044 #define	qready()	1
1045 
1046 #ifdef _KERNEL
1047 #define	runqueues()
1048 #define	queuerun()
1049 #endif
1050 
1051 /* compatibility module for style 2 drivers with DR race condition */
1052 #define	DRMODNAME	"drcompat"
1053 
1054 /*
1055  * Macros dealing with mux_nodes.
1056  */
1057 #define	MUX_VISIT(X)	((X)->mn_flags |= VISITED)
1058 #define	MUX_CLEAR(X)	((X)->mn_flags &= (~VISITED)); \
1059 			((X)->mn_originp = NULL)
1060 #define	MUX_DIDVISIT(X)	((X)->mn_flags & VISITED)
1061 
1062 
1063 /*
1064  * Twisted stream macros
1065  */
1066 #define	STRMATED(X)	((X)->sd_flag & STRMATE)
1067 #define	STRLOCKMATES(X)	if (&((X)->sd_lock) > &(((X)->sd_mate)->sd_lock)) { \
1068 				mutex_enter(&((X)->sd_lock)); \
1069 				mutex_enter(&(((X)->sd_mate)->sd_lock));  \
1070 			} else {  \
1071 				mutex_enter(&(((X)->sd_mate)->sd_lock)); \
1072 				mutex_enter(&((X)->sd_lock)); \
1073 			}
1074 #define	STRUNLOCKMATES(X)	mutex_exit(&((X)->sd_lock)); \
1075 			mutex_exit(&(((X)->sd_mate)->sd_lock))
1076 
1077 #ifdef _KERNEL
1078 
1079 extern void strinit(void);
1080 extern int strdoioctl(struct stdata *, struct strioctl *, int, int,
1081     cred_t *, int *);
1082 extern void strsendsig(struct strsig *, int, uchar_t, int);
1083 extern void str_sendsig(vnode_t *, int, uchar_t, int);
1084 extern void strhup(struct stdata *);
1085 extern int qattach(queue_t *, dev_t *, int, cred_t *, fmodsw_impl_t *,
1086     boolean_t);
1087 extern int qreopen(queue_t *, dev_t *, int, cred_t *);
1088 extern void qdetach(queue_t *, int, int, cred_t *, boolean_t);
1089 extern void enterq(queue_t *);
1090 extern void leaveq(queue_t *);
1091 extern int putiocd(mblk_t *, caddr_t, int, cred_t *);
1092 extern int getiocd(mblk_t *, caddr_t, int);
1093 extern struct linkinfo *alloclink(queue_t *, queue_t *, struct file *);
1094 extern void lbfree(struct linkinfo *);
1095 extern int linkcycle(stdata_t *, stdata_t *, str_stack_t *);
1096 extern struct linkinfo *findlinks(stdata_t *, int, int, str_stack_t *);
1097 extern queue_t *getendq(queue_t *);
1098 extern int mlink(vnode_t *, int, int, cred_t *, int *, int);
1099 extern int mlink_file(vnode_t *, int, struct file *, cred_t *, int *, int);
1100 extern int munlink(struct stdata *, struct linkinfo *, int, cred_t *, int *,
1101     str_stack_t *);
1102 extern int munlinkall(struct stdata *, int, cred_t *, int *, str_stack_t *);
1103 extern void mux_addedge(stdata_t *, stdata_t *, int, str_stack_t *);
1104 extern void mux_rmvedge(stdata_t *, int, str_stack_t *);
1105 extern int devflg_to_qflag(struct streamtab *, uint32_t, uint32_t *,
1106     uint32_t *);
1107 extern void setq(queue_t *, struct qinit *, struct qinit *, perdm_t *,
1108     uint32_t, uint32_t, boolean_t);
1109 extern perdm_t *hold_dm(struct streamtab *, uint32_t, uint32_t);
1110 extern void rele_dm(perdm_t *);
1111 extern int strmakectl(struct strbuf *, int32_t, int32_t, mblk_t **);
1112 extern int strmakedata(ssize_t *, struct uio *, stdata_t *, int32_t, mblk_t **);
1113 extern int strmakemsg(struct strbuf *, ssize_t *, struct uio *,
1114     struct stdata *, int32_t, mblk_t **);
1115 extern int strgetmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t *,
1116     int *, int, rval_t *);
1117 extern int strputmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t,
1118     int flag, int fmode);
1119 extern int strstartplumb(struct stdata *, int, int);
1120 extern void strendplumb(struct stdata *);
1121 extern int stropen(struct vnode *, dev_t *, int, cred_t *);
1122 extern int strclose(struct vnode *, int, cred_t *);
1123 extern int strpoll(register struct stdata *, short, int, short *,
1124     struct pollhead **);
1125 extern void strclean(struct vnode *);
1126 extern void str_cn_clean();	/* XXX hook for consoles signal cleanup */
1127 extern int strwrite(struct vnode *, struct uio *, cred_t *);
1128 extern int strwrite_common(struct vnode *, struct uio *, cred_t *, int);
1129 extern int kstrwritemp(struct vnode *, mblk_t *, ushort_t);
1130 extern int strread(struct vnode *, struct uio *, cred_t *);
1131 extern int strioctl(struct vnode *, int, intptr_t, int, int, cred_t *, int *);
1132 extern int strrput(queue_t *, mblk_t *);
1133 extern int strrput_nondata(queue_t *, mblk_t *);
1134 extern mblk_t *strrput_proto(vnode_t *, mblk_t *,
1135     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1136 extern mblk_t *strrput_misc(vnode_t *, mblk_t *,
1137     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1138 extern int getiocseqno(void);
1139 extern int strwaitbuf(size_t, int);
1140 extern int strwaitq(stdata_t *, int, ssize_t, int, clock_t, int *);
1141 extern struct stdata *shalloc(queue_t *);
1142 extern void shfree(struct stdata *s);
1143 extern queue_t *allocq(void);
1144 extern void freeq(queue_t *);
1145 extern qband_t *allocband(void);
1146 extern void freeband(qband_t *);
1147 extern void freebs_enqueue(mblk_t *, dblk_t *);
1148 extern void setqback(queue_t *, unsigned char);
1149 extern int strcopyin(void *, void *, size_t, int);
1150 extern int strcopyout(void *, void *, size_t, int);
1151 extern void strsignal(struct stdata *, int, int32_t);
1152 extern clock_t str_cv_wait(kcondvar_t *, kmutex_t *, clock_t, int);
1153 extern void disable_svc(queue_t *);
1154 extern void remove_runlist(queue_t *);
1155 extern void wait_svc(queue_t *);
1156 extern void backenable(queue_t *, uchar_t);
1157 extern void set_qend(queue_t *);
1158 extern int strgeterr(stdata_t *, int32_t, int);
1159 extern void qenable_locked(queue_t *);
1160 extern mblk_t *getq_noenab(queue_t *, ssize_t);
1161 extern void rmvq_noenab(queue_t *, mblk_t *);
1162 extern void qbackenable(queue_t *, uchar_t);
1163 extern void set_qfull(queue_t *);
1164 
1165 extern void strblock(queue_t *);
1166 extern void strunblock(queue_t *);
1167 extern int qclaimed(queue_t *);
1168 extern int straccess(struct stdata *, enum jcaccess);
1169 
1170 extern void entersq(syncq_t *, int);
1171 extern void leavesq(syncq_t *, int);
1172 extern void claimq(queue_t *);
1173 extern void releaseq(queue_t *);
1174 extern void claimstr(queue_t *);
1175 extern void releasestr(queue_t *);
1176 extern void removeq(queue_t *);
1177 extern void insertq(struct stdata *, queue_t *);
1178 extern void drain_syncq(syncq_t *);
1179 extern void qfill_syncq(syncq_t *, queue_t *, mblk_t *);
1180 extern void qdrain_syncq(syncq_t *, queue_t *);
1181 extern int flush_syncq(syncq_t *, queue_t *);
1182 extern void wait_sq_svc(syncq_t *);
1183 
1184 extern void outer_enter(syncq_t *, uint16_t);
1185 extern void outer_exit(syncq_t *);
1186 extern void qwriter_inner(queue_t *, mblk_t *, void (*)());
1187 extern void qwriter_outer(queue_t *, mblk_t *, void (*)());
1188 
1189 extern callbparams_t *callbparams_alloc(syncq_t *, void (*)(void *),
1190     void *, int);
1191 extern void callbparams_free(syncq_t *, callbparams_t *);
1192 extern void callbparams_free_id(syncq_t *, callbparams_id_t, int32_t);
1193 extern void qcallbwrapper(void *);
1194 
1195 extern mblk_t *esballoc_wait(unsigned char *, size_t, uint_t, frtn_t *);
1196 extern mblk_t *esballoca(unsigned char *, size_t, uint_t, frtn_t *);
1197 extern mblk_t *desballoca(unsigned char *, size_t, uint_t, frtn_t *);
1198 extern int do_sendfp(struct stdata *, struct file *, struct cred *);
1199 extern int frozenstr(queue_t *);
1200 extern size_t xmsgsize(mblk_t *);
1201 
1202 extern void putnext_tail(syncq_t *, queue_t *, uint32_t);
1203 extern void stream_willservice(stdata_t *);
1204 extern void stream_runservice(stdata_t *);
1205 
1206 extern void strmate(vnode_t *, vnode_t *);
1207 extern queue_t *strvp2wq(vnode_t *);
1208 extern vnode_t *strq2vp(queue_t *);
1209 extern mblk_t *allocb_wait(size_t, uint_t, uint_t, int *);
1210 extern mblk_t *allocb_cred(size_t, cred_t *);
1211 extern mblk_t *allocb_cred_wait(size_t, uint_t, int *, cred_t *);
1212 extern mblk_t *allocb_tmpl(size_t, const mblk_t *);
1213 extern mblk_t *allocb_tryhard(size_t);
1214 extern void mblk_setcred(mblk_t *, cred_t *);
1215 extern void strpollwakeup(vnode_t *, short);
1216 extern int putnextctl_wait(queue_t *, int);
1217 
1218 extern int kstrputmsg(struct vnode *, mblk_t *, struct uio *, ssize_t,
1219     unsigned char, int, int);
1220 extern int kstrgetmsg(struct vnode *, mblk_t **, struct uio *,
1221     unsigned char *, int *, clock_t, rval_t *);
1222 
1223 extern void strsetrerror(vnode_t *, int, int, errfunc_t);
1224 extern void strsetwerror(vnode_t *, int, int, errfunc_t);
1225 extern void strseteof(vnode_t *, int);
1226 extern void strflushrq(vnode_t *, int);
1227 extern void strsetrputhooks(vnode_t *, uint_t, msgfunc_t, msgfunc_t);
1228 extern void strsetwputhooks(vnode_t *, uint_t, clock_t);
1229 extern void strsetrwputdatahooks(vnode_t *, msgfunc_t, msgfunc_t);
1230 extern int strwaitmark(vnode_t *);
1231 extern void strsignal_nolock(stdata_t *, int, int32_t);
1232 
1233 struct multidata_s;
1234 struct pdesc_s;
1235 extern int hcksum_assoc(mblk_t *, struct multidata_s *, struct pdesc_s  *,
1236     uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, int);
1237 extern void hcksum_retrieve(mblk_t *, struct multidata_s *, struct pdesc_s *,
1238     uint32_t *, uint32_t *, uint32_t *, uint32_t *, uint32_t *);
1239 extern void lso_info_set(mblk_t *, uint32_t, uint32_t);
1240 extern void lso_info_get(mblk_t *, uint32_t *, uint32_t *);
1241 extern unsigned int bcksum(uchar_t *, int, unsigned int);
1242 extern boolean_t is_vmloaned_mblk(mblk_t *, struct multidata_s *,
1243     struct pdesc_s *);
1244 
1245 extern int fmodsw_register(const char *, struct streamtab *, int);
1246 extern int fmodsw_unregister(const char *);
1247 extern fmodsw_impl_t *fmodsw_find(const char *, fmodsw_flags_t);
1248 extern void fmodsw_rele(fmodsw_impl_t *);
1249 
1250 extern void freemsgchain(mblk_t *);
1251 extern mblk_t *copymsgchain(mblk_t *);
1252 
1253 extern mblk_t *mcopyinuio(struct stdata *, uio_t *, ssize_t, ssize_t, int *);
1254 
1255 /*
1256  * shared or externally configured data structures
1257  */
1258 extern ssize_t strmsgsz;		/* maximum stream message size */
1259 extern ssize_t strctlsz;		/* maximum size of ctl message */
1260 extern int nstrpush;			/* maximum number of pushes allowed */
1261 
1262 /*
1263  * Bufcalls related variables.
1264  */
1265 extern struct bclist strbcalls;		/* List of bufcalls */
1266 extern kmutex_t	strbcall_lock;		/* Protects the list of bufcalls */
1267 extern kcondvar_t strbcall_cv;		/* Signaling when a bufcall is added */
1268 extern kcondvar_t bcall_cv;	/* wait of executing bufcall completes */
1269 
1270 extern frtn_t frnop;
1271 
1272 extern struct kmem_cache *ciputctrl_cache;
1273 extern int n_ciputctrl;
1274 extern int max_n_ciputctrl;
1275 extern int min_n_ciputctrl;
1276 
1277 extern cdevsw_impl_t *devimpl;
1278 
1279 /*
1280  * esballoc queue for throttling
1281  */
1282 typedef struct esb_queue {
1283 	kmutex_t	eq_lock;
1284 	uint_t		eq_len;		/* number of queued messages */
1285 	mblk_t		*eq_head;	/* head of queue */
1286 	mblk_t		*eq_tail;	/* tail of queue */
1287 	uint_t		eq_flags;	/* esballoc queue flags */
1288 } esb_queue_t;
1289 
1290 /*
1291  * esballoc flags for queue processing.
1292  */
1293 #define	ESBQ_PROCESSING	0x01	/* queue is being processed */
1294 #define	ESBQ_TIMER	0x02	/* timer is active */
1295 
1296 extern void esballoc_queue_init(void);
1297 
1298 #endif	/* _KERNEL */
1299 
1300 /*
1301  * Note: Use of these macros are restricted to kernel/unix and
1302  * intended for the STREAMS framework.
1303  * All modules/drivers should include sys/ddi.h.
1304  *
1305  * Finding related queues
1306  */
1307 #define		_OTHERQ(q)	((q)->q_flag&QREADR? (q)+1: (q)-1)
1308 #define		_WR(q)		((q)->q_flag&QREADR? (q)+1: (q))
1309 #define		_RD(q)		((q)->q_flag&QREADR? (q): (q)-1)
1310 #define		_SAMESTR(q)	(!((q)->q_flag & QEND))
1311 
1312 /*
1313  * These are also declared here for modules/drivers that erroneously
1314  * include strsubr.h after ddi.h or fail to include ddi.h at all.
1315  */
1316 extern struct queue *OTHERQ(queue_t *); /* stream.h */
1317 extern struct queue *RD(queue_t *);
1318 extern struct queue *WR(queue_t *);
1319 extern int SAMESTR(queue_t *);
1320 
1321 /*
1322  * The following hardware checksum related macros are private
1323  * interfaces that are subject to change without notice.
1324  */
1325 #ifdef _KERNEL
1326 #define	DB_CKSUMSTART(mp)	((mp)->b_datap->db_cksumstart)
1327 #define	DB_CKSUMEND(mp)		((mp)->b_datap->db_cksumend)
1328 #define	DB_CKSUMSTUFF(mp)	((mp)->b_datap->db_cksumstuff)
1329 #define	DB_CKSUMFLAGS(mp)	((mp)->b_datap->db_struioun.cksum.flags)
1330 #define	DB_CKSUM16(mp)		((mp)->b_datap->db_cksum16)
1331 #define	DB_CKSUM32(mp)		((mp)->b_datap->db_cksum32)
1332 #define	DB_LSOFLAGS(mp)		((mp)->b_datap->db_struioun.cksum.flags)
1333 #define	DB_LSOMSS(mp)		((mp)->b_datap->db_struioun.cksum.pad)
1334 #endif	/* _KERNEL */
1335 
1336 #ifdef	__cplusplus
1337 }
1338 #endif
1339 
1340 
1341 #endif	/* _SYS_STRSUBR_H */
1342