xref: /illumos-gate/usr/src/uts/common/sys/strsubr.h (revision 78a2e113edb6fe0a0382b403b55d92e8f0bba78f)
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 2010 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 /*
31  * Copyright 2019 OmniOS Community Edition (OmniOSce) Association.
32  */
33 
34 #ifndef _SYS_STRSUBR_H
35 #define	_SYS_STRSUBR_H
36 
37 /*
38  * WARNING:
39  * Everything in this file is private, belonging to the
40  * STREAMS subsystem.  The only guarantee made about the
41  * contents of this file is that if you include it, your
42  * code will not port to the next release.
43  */
44 #include <sys/stream.h>
45 #include <sys/stropts.h>
46 #include <sys/vnode.h>
47 #include <sys/kstat.h>
48 #include <sys/uio.h>
49 #include <sys/proc.h>
50 #include <sys/netstack.h>
51 #include <sys/modhash.h>
52 
53 #ifdef	__cplusplus
54 extern "C" {
55 #endif
56 
57 /*
58  * In general, the STREAMS locks are disjoint; they are only held
59  * locally, and not simultaneously by a thread.  However, module
60  * code, including at the stream head, requires some locks to be
61  * acquired in order for its safety.
62  *	1. Stream level claim.  This prevents the value of q_next
63  *		from changing while module code is executing.
64  *	2. Queue level claim.  This prevents the value of q_ptr
65  *		from changing while put or service code is executing.
66  *		In addition, it provides for queue single-threading
67  *		for QPAIR and PERQ MT-safe modules.
68  *	3. Stream head lock.  May be held by the stream head module
69  *		to implement a read/write/open/close monitor.
70  *	   Note: that the only types of twisted stream supported are
71  *	   the pipe and transports which have read and write service
72  *	   procedures on both sides of the twist.
73  *	4. Queue lock.  May be acquired by utility routines on
74  *		behalf of a module.
75  */
76 
77 /*
78  * In general, sd_lock protects the consistency of the stdata
79  * structure.  Additionally, it is used with sd_monitor
80  * to implement an open/close monitor.  In particular, it protects
81  * the following fields:
82  *	sd_iocblk
83  *	sd_flag
84  *	sd_copyflag
85  *	sd_iocid
86  *	sd_iocwait
87  *	sd_sidp
88  *	sd_pgidp
89  *	sd_wroff
90  *	sd_tail
91  *	sd_rerror
92  *	sd_werror
93  *	sd_pushcnt
94  *	sd_sigflags
95  *	sd_siglist
96  *	sd_pollist
97  *	sd_mark
98  *	sd_closetime
99  *	sd_wakeq
100  *	sd_maxblk
101  *
102  * The following fields are modified only by the allocator, which
103  * has exclusive access to them at that time:
104  *	sd_wrq
105  *	sd_strtab
106  *
107  * The following field is protected by the overlying file system
108  * code, guaranteeing single-threading of opens:
109  *	sd_vnode
110  *
111  * Stream-level locks should be acquired before any queue-level locks
112  *	are acquired.
113  *
114  * The stream head write queue lock(sd_wrq) is used to protect the
115  * fields qn_maxpsz and qn_minpsz because freezestr() which is
116  * necessary for strqset() only gets the queue lock.
117  */
118 
119 /*
120  * Function types for the parameterized stream head.
121  * The msgfunc_t takes the parameters:
122  *	msgfunc(vnode_t *vp, mblk_t *mp, strwakeup_t *wakeups,
123  *		strsigset_t *firstmsgsigs, strsigset_t *allmsgsigs,
124  *		strpollset_t *pollwakeups);
125  * It returns an optional message to be processed by the stream head.
126  *
127  * The parameters for errfunc_t are:
128  *	errfunc(vnode *vp, int ispeek, int *clearerr);
129  * It returns an errno and zero if there was no pending error.
130  */
131 typedef uint_t	strwakeup_t;
132 typedef uint_t	strsigset_t;
133 typedef short	strpollset_t;
134 typedef uintptr_t callbparams_id_t;
135 typedef	mblk_t	*(*msgfunc_t)(vnode_t *, mblk_t *, strwakeup_t *,
136 			strsigset_t *, strsigset_t *, strpollset_t *);
137 typedef int	(*errfunc_t)(vnode_t *, int, int *);
138 
139 /*
140  * Per stream sd_lock in putnext may be replaced by per cpu stream_putlocks
141  * each living in a separate cache line. putnext/canputnext grabs only one of
142  * stream_putlocks while strlock() (called on behalf of insertq()/removeq())
143  * acquires all stream_putlocks. Normally stream_putlocks are only employed
144  * for highly contended streams that have SQ_CIPUT queues in the critical path
145  * (e.g. NFS/UDP stream).
146  *
147  * stream_putlocks are dynamically assigned to stdata structure through
148  * sd_ciputctrl pointer possibly when a stream is already in use. Since
149  * strlock() uses stream_putlocks only under sd_lock acquiring sd_lock when
150  * assigning stream_putlocks to the stream ensures synchronization with
151  * strlock().
152  *
153  * For lock ordering purposes stream_putlocks are treated as the extension of
154  * sd_lock and are always grabbed right after grabbing sd_lock and released
155  * right before releasing sd_lock except putnext/canputnext where only one of
156  * stream_putlocks locks is used and where it is the first lock to grab.
157  */
158 
159 typedef struct ciputctrl_str {
160 	union _ciput_un {
161 		uchar_t	pad[64];
162 		struct _ciput_str {
163 			kmutex_t	ciput_lck;
164 			ushort_t	ciput_cnt;
165 		} ciput_str;
166 	} ciput_un;
167 } ciputctrl_t;
168 
169 #define	ciputctrl_lock	ciput_un.ciput_str.ciput_lck
170 #define	ciputctrl_count	ciput_un.ciput_str.ciput_cnt
171 
172 /*
173  * Header for a stream: interface to rest of system.
174  *
175  * NOTE: While this is a consolidation-private structure, some unbundled and
176  *       third-party products inappropriately make use of some of the fields.
177  *       As such, please take care to not gratuitously change any offsets of
178  *       existing members.
179  */
180 typedef struct stdata {
181 	struct queue	*sd_wrq;	/* write queue */
182 	struct msgb	*sd_iocblk;	/* return block for ioctl */
183 	struct vnode	*sd_vnode;	/* pointer to associated vnode */
184 	struct streamtab *sd_strtab;	/* pointer to streamtab for stream */
185 	uint_t		sd_flag;	/* state/flags */
186 	uint_t		sd_iocid;	/* ioctl id */
187 	struct pid	*sd_sidp;	/* controlling session info */
188 	struct pid	*sd_pgidp;	/* controlling process group info */
189 	ushort_t	sd_tail;	/* reserved space in written mblks */
190 	ushort_t	sd_wroff;	/* write offset */
191 	int		sd_rerror;	/* error to return on read ops */
192 	int		sd_werror;	/* error to return on write ops */
193 	int		sd_pushcnt;	/* number of pushes done on stream */
194 	int		sd_sigflags;	/* logical OR of all siglist events */
195 	struct strsig	*sd_siglist;	/* pid linked list to rcv SIGPOLL sig */
196 	struct pollhead sd_pollist;	/* list of all pollers to wake up */
197 	struct msgb	*sd_mark;	/* "marked" message on read queue */
198 	clock_t		sd_closetime;	/* time to wait to drain q in close */
199 	kmutex_t	sd_lock;	/* protect head consistency */
200 	kcondvar_t	sd_monitor;	/* open/close/push/pop monitor */
201 	kcondvar_t	sd_iocmonitor;	/* ioctl single-threading */
202 	kcondvar_t	sd_refmonitor;	/* sd_refcnt monitor */
203 	ssize_t		sd_qn_minpsz;	/* These two fields are a performance */
204 	ssize_t		sd_qn_maxpsz;	/* enhancements, cache the values in */
205 					/* the stream head so we don't have */
206 					/* to ask the module below the stream */
207 					/* head to get this information. */
208 	struct stdata	*sd_mate;	/* pointer to twisted stream mate */
209 	kthread_id_t	sd_freezer;	/* thread that froze stream */
210 	kmutex_t	sd_reflock;	/* Protects sd_refcnt */
211 	int		sd_refcnt;	/* number of claimstr */
212 	uint_t		sd_wakeq;	/* strwakeq()'s copy of sd_flag */
213 	struct queue	*sd_struiordq;	/* sync barrier struio() read queue */
214 	struct queue	*sd_struiowrq;	/* sync barrier struio() write queue */
215 	char		sd_struiodnak;	/* defer NAK of M_IOCTL by rput() */
216 	struct msgb	*sd_struionak;	/* pointer M_IOCTL mblk(s) to NAK */
217 	caddr_t		sd_t_audit_data; /* For audit purposes only */
218 	ssize_t		sd_maxblk;	/* maximum message block size */
219 	uint_t		sd_rput_opt;	/* options/flags for strrput */
220 	uint_t		sd_wput_opt;	/* options/flags for write/putmsg */
221 	uint_t		sd_read_opt;	/* options/flags for strread */
222 	msgfunc_t	sd_rprotofunc;	/* rput M_*PROTO routine */
223 	msgfunc_t	sd_rputdatafunc; /* read M_DATA routine */
224 	msgfunc_t	sd_rmiscfunc;	/* rput routine (non-data/proto) */
225 	msgfunc_t	sd_wputdatafunc; /* wput M_DATA routine */
226 	errfunc_t	sd_rderrfunc;	/* read side error callback */
227 	errfunc_t	sd_wrerrfunc;	/* write side error callback */
228 	/*
229 	 * support for low contention concurrent putnext.
230 	 */
231 	ciputctrl_t	*sd_ciputctrl;
232 	uint_t		sd_nciputctrl;
233 
234 	int		sd_anchor;	/* position of anchor in stream */
235 	/*
236 	 * Service scheduling at the stream head.
237 	 */
238 	kmutex_t	sd_qlock;
239 	struct queue	*sd_qhead;	/* Head of queues to be serviced. */
240 	struct queue	*sd_qtail;	/* Tail of queues to be serviced. */
241 	void		*sd_servid;	/* Service ID for bckgrnd schedule */
242 	ushort_t	sd_svcflags;	/* Servicing flags */
243 	short		sd_nqueues;	/* Number of queues in the list */
244 	kcondvar_t	sd_qcv;		/* Waiters for qhead to become empty */
245 	kcondvar_t	sd_zcopy_wait;
246 	uint_t		sd_copyflag;	/* copy-related flags */
247 	zoneid_t	sd_anchorzone;	/* Allow removal from same zone only */
248 	struct msgb	*sd_cmdblk;	/* reply from _I_CMD */
249 
250 	/*
251 	 * When a STREAMS device is cloned, the sd_vnode element of this
252 	 * structure is replaced by a pointer to a common vnode shared across
253 	 * all streams that are using the device. In this case, it is no longer
254 	 * possible to get from the stream head back to the original vnode via
255 	 * sd_vnode. Therefore, when such a device is cloned, the parent vnode -
256 	 * i.e. that which was created during the device clone in spec_clone()
257 	 * - is kept in sd_pvnode.
258 	 */
259 	struct vnode	*sd_pvnode;
260 } stdata_t;
261 
262 /*
263  * stdata servicing flags.
264  */
265 #define	STRS_WILLSERVICE	0x01
266 #define	STRS_SCHEDULED		0x02
267 
268 #define	STREAM_NEEDSERVICE(stp)	((stp)->sd_qhead != NULL)
269 
270 /*
271  * stdata flag field defines
272  */
273 #define	IOCWAIT		0x00000001	/* Someone is doing an ioctl */
274 #define	RSLEEP		0x00000002	/* Someone wants to read/recv msg */
275 #define	WSLEEP		0x00000004	/* Someone wants to write */
276 #define	STRPRI		0x00000008	/* An M_PCPROTO is at stream head */
277 #define	STRHUP		0x00000010	/* Device has vanished */
278 #define	STWOPEN		0x00000020	/* waiting for 1st open */
279 #define	STPLEX		0x00000040	/* stream is being multiplexed */
280 #define	STRISTTY	0x00000080	/* stream is a terminal */
281 #define	STRGETINPROG	0x00000100	/* (k)strgetmsg is running */
282 #define	IOCWAITNE	0x00000200	/* STR_NOERROR ioctl running */
283 #define	STRDERR		0x00000400	/* fatal read error from M_ERROR */
284 #define	STWRERR		0x00000800	/* fatal write error from M_ERROR */
285 #define	STRDERRNONPERSIST 0x00001000	/* nonpersistent read errors */
286 #define	STWRERRNONPERSIST 0x00002000	/* nonpersistent write errors */
287 #define	STRCLOSE	0x00004000	/* wait for a close to complete */
288 #define	SNDMREAD	0x00008000	/* used for read notification */
289 #define	OLDNDELAY	0x00010000	/* use old TTY semantics for */
290 					/* NDELAY reads and writes */
291 	/*		0x00020000	   unused */
292 	/*		0x00040000	   unused */
293 #define	STRTOSTOP	0x00080000	/* block background writes */
294 #define	STRCMDWAIT	0x00100000	/* someone is doing an _I_CMD */
295 	/*		0x00200000	   unused */
296 #define	STRMOUNT	0x00400000	/* stream is mounted */
297 #define	STRNOTATMARK	0x00800000	/* Not at mark (when empty read q) */
298 #define	STRDELIM	0x01000000	/* generate delimited messages */
299 #define	STRATMARK	0x02000000	/* At mark (due to MSGMARKNEXT) */
300 #define	STZCNOTIFY	0x04000000	/* wait for zerocopy mblk to be acked */
301 #define	STRPLUMB	0x08000000	/* push/pop pending */
302 #define	STREOF		0x10000000	/* End-of-file indication */
303 #define	STREOPENFAIL	0x20000000	/* indicates if re-open has failed */
304 #define	STRMATE		0x40000000	/* this stream is a mate */
305 #define	STRHASLINKS	0x80000000	/* I_LINKs under this stream */
306 
307 /*
308  * Copy-related flags (sd_copyflag), set by SO_COPYOPT.
309  */
310 #define	STZCVMSAFE	0x00000001	/* safe to borrow file (segmapped) */
311 					/* pages instead of bcopy */
312 #define	STZCVMUNSAFE	0x00000002	/* unsafe to borrow file pages */
313 #define	STRCOPYCACHED	0x00000004	/* copy should NOT bypass cache */
314 
315 /*
316  * Options and flags for strrput (sd_rput_opt)
317  */
318 #define	SR_POLLIN	0x00000001	/* pollwakeup needed for band0 data */
319 #define	SR_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
320 #define	SR_CONSOL_DATA	0x00000004	/* Consolidate M_DATA onto q_last */
321 #define	SR_IGN_ZEROLEN	0x00000008	/* Ignore zero-length M_DATA */
322 
323 /*
324  * Options and flags for strwrite/strputmsg (sd_wput_opt)
325  */
326 #define	SW_SIGPIPE	0x00000001	/* Send SIGPIPE for write error */
327 #define	SW_RECHECK_ERR	0x00000002	/* Recheck errors in strwrite loop */
328 #define	SW_SNDZERO	0x00000004	/* send 0-length msg down pipe/FIFO */
329 
330 /*
331  * Options and flags for strread (sd_read_opt)
332  */
333 #define	RD_MSGDIS	0x00000001	/* read msg discard */
334 #define	RD_MSGNODIS	0x00000002	/* read msg no discard */
335 #define	RD_PROTDAT	0x00000004	/* read M_[PC]PROTO contents as data */
336 #define	RD_PROTDIS	0x00000008	/* discard M_[PC]PROTO blocks and */
337 					/* retain data blocks */
338 /*
339  * Flags parameter for strsetrputhooks() and strsetwputhooks().
340  * These flags define the interface for setting the above internal
341  * flags in sd_rput_opt and sd_wput_opt.
342  */
343 #define	SH_CONSOL_DATA	0x00000001	/* Consolidate M_DATA onto q_last */
344 #define	SH_SIGALLDATA	0x00000002	/* Send SIGPOLL for all M_DATA */
345 #define	SH_IGN_ZEROLEN	0x00000004	/* Drop zero-length M_DATA */
346 
347 #define	SH_SIGPIPE	0x00000100	/* Send SIGPIPE for write error */
348 #define	SH_RECHECK_ERR	0x00000200	/* Recheck errors in strwrite loop */
349 
350 /*
351  * Each queue points to a sync queue (the inner perimeter) which keeps
352  * track of the number of threads that are inside a given queue (sq_count)
353  * and also is used to implement the asynchronous putnext
354  * (by queuing messages if the queue can not be entered.)
355  *
356  * Messages are queued on sq_head/sq_tail including deferred qwriter(INNER)
357  * messages. The sq_head/sq_tail list is a singly-linked list with
358  * b_queue recording the queue and b_prev recording the function to
359  * be called (either the put procedure or a qwriter callback function.)
360  *
361  * The sq_count counter tracks the number of threads that are
362  * executing inside the perimeter or (in the case of outer perimeters)
363  * have some work queued for them relating to the perimeter. The sq_rmqcount
364  * counter tracks the subset which are in removeq() (usually invoked from
365  * qprocsoff(9F)).
366  *
367  * In addition a module writer can declare that the module has an outer
368  * perimeter (by setting D_MTOUTPERIM) in which case all inner perimeter
369  * syncq's for the module point (through sq_outer) to an outer perimeter
370  * syncq. The outer perimeter consists of the doubly linked list (sq_onext and
371  * sq_oprev) linking all the inner perimeter syncq's with out outer perimeter
372  * syncq. This is used to implement qwriter(OUTER) (an asynchronous way of
373  * getting exclusive access at the outer perimeter) and outer_enter/exit
374  * which are used by the framework to acquire exclusive access to the outer
375  * perimeter during open and close of modules that have set D_MTOUTPERIM.
376  *
377  * In the inner perimeter case sq_save is available for use by machine
378  * dependent code. sq_head/sq_tail are used to queue deferred messages on
379  * the inner perimeter syncqs and to queue become_writer requests on the
380  * outer perimeter syncqs.
381  *
382  * Note: machine dependent optimized versions of putnext may depend
383  * on the order of sq_flags and sq_count (so that they can e.g.
384  * read these two fields in a single load instruction.)
385  *
386  * Per perimeter SQLOCK/sq_count in putnext/put may be replaced by per cpu
387  * sq_putlocks/sq_putcounts each living in a separate cache line. Obviously
388  * sq_putlock[x] protects sq_putcount[x]. putnext/put routine will grab only 1
389  * of sq_putlocks and update only 1 of sq_putcounts. strlock() and many
390  * other routines in strsubr.c and ddi.c will grab all sq_putlocks (as well as
391  * SQLOCK) and figure out the count value as the sum of sq_count and all of
392  * sq_putcounts. The idea is to make critical fast path -- putnext -- much
393  * faster at the expense of much less often used slower path like
394  * strlock(). One known case where entersq/strlock is executed pretty often is
395  * SpecWeb but since IP is SQ_CIOC and socket TCP/IP stream is nextless
396  * there's no need to grab multiple sq_putlocks and look at sq_putcounts. See
397  * strsubr.c for more comments.
398  *
399  * Note regular SQLOCK and sq_count are still used in many routines
400  * (e.g. entersq(), rwnext()) in the same way as before sq_putlocks were
401  * introduced.
402  *
403  * To understand when all sq_putlocks need to be held and all sq_putcounts
404  * need to be added up one needs to look closely at putnext code. Basically if
405  * a routine like e.g. wait_syncq() needs to be sure that perimeter is empty
406  * all sq_putlocks/sq_putcounts need to be held/added up. On the other hand
407  * there's no need to hold all sq_putlocks and count all sq_putcounts in
408  * routines like leavesq()/dropsq() and etc. since the are usually exit
409  * counterparts of entersq/outer_enter() and etc. which have already either
410  * prevented put entry poins from executing or did not care about put
411  * entrypoints. entersq() doesn't need to care about sq_putlocks/sq_putcounts
412  * if the entry point has a shared access since put has the highest degree of
413  * concurrency and such entersq() does not intend to block out put
414  * entrypoints.
415  *
416  * Before sq_putcounts were introduced the standard way to wait for perimeter
417  * to become empty was:
418  *
419  *	mutex_enter(SQLOCK(sq));
420  *	while (sq->sq_count > 0) {
421  *		sq->sq_flags |= SQ_WANTWAKEUP;
422  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
423  *	}
424  *	mutex_exit(SQLOCK(sq));
425  *
426  * The new way is:
427  *
428  *	mutex_enter(SQLOCK(sq));
429  *	count = sq->sq_count;
430  *	SQ_PUTLOCKS_ENTER(sq);
431  *	SUM_SQ_PUTCOUNTS(sq, count);
432  *	while (count != 0) {
433  *		sq->sq_flags |= SQ_WANTWAKEUP;
434  *		SQ_PUTLOCKS_EXIT(sq);
435  *		cv_wait(&sq->sq_wait, SQLOCK(sq));
436  *		count = sq->sq_count;
437  *		SQ_PUTLOCKS_ENTER(sq);
438  *		SUM_SQ_PUTCOUNTS(sq, count);
439  *	}
440  *	SQ_PUTLOCKS_EXIT(sq);
441  *	mutex_exit(SQLOCK(sq));
442  *
443  * Note that SQ_WANTWAKEUP is set before dropping SQ_PUTLOCKS. This makes sure
444  * putnext won't skip a wakeup.
445  *
446  * sq_putlocks are treated as the extension of SQLOCK for lock ordering
447  * purposes and are always grabbed right after grabbing SQLOCK and released
448  * right before releasing SQLOCK. This also allows dynamic creation of
449  * sq_putlocks while holding SQLOCK (by making sq_ciputctrl non null even when
450  * the stream is already in use). Only in putnext one of sq_putlocks
451  * is grabbed instead of SQLOCK. putnext return path remembers what counter it
452  * incremented and decrements the right counter on its way out.
453  */
454 
455 struct syncq {
456 	kmutex_t	sq_lock;	/* atomic access to syncq */
457 	uint16_t	sq_count;	/* # threads inside */
458 	uint16_t	sq_flags;	/* state and some type info */
459 	/*
460 	 * Distributed syncq scheduling
461 	 *  The list of queue's is handled by sq_head and
462 	 *  sq_tail fields.
463 	 *
464 	 *  The list of events is handled by the sq_evhead and sq_evtail
465 	 *  fields.
466 	 */
467 	queue_t		*sq_head;	/* queue of deferred messages */
468 	queue_t		*sq_tail;	/* queue of deferred messages */
469 	mblk_t		*sq_evhead;	/* Event message on the syncq */
470 	mblk_t		*sq_evtail;
471 	uint_t		sq_nqueues;	/* # of queues on this sq */
472 	/*
473 	 * Concurrency and condition variables
474 	 */
475 	uint16_t	sq_type;	/* type (concurrency) of syncq */
476 	uint16_t	sq_rmqcount;	/* # threads inside removeq() */
477 	kcondvar_t	sq_wait;	/* block on this sync queue */
478 	kcondvar_t	sq_exitwait;	/* waiting for thread to leave the */
479 					/* inner perimeter */
480 	/*
481 	 * Handling synchronous callbacks such as qtimeout and qbufcall
482 	 */
483 	ushort_t	sq_callbflags;	/* flags for callback synchronization */
484 	callbparams_id_t sq_cancelid;	/* id of callback being cancelled */
485 	struct callbparams *sq_callbpend;	/* Pending callbacks */
486 
487 	/*
488 	 * Links forming an outer perimeter from one outer syncq and
489 	 * a set of inner sync queues.
490 	 */
491 	struct syncq	*sq_outer;	/* Pointer to outer perimeter */
492 	struct syncq	*sq_onext;	/* Linked list of syncq's making */
493 	struct syncq	*sq_oprev;	/* up the outer perimeter. */
494 	/*
495 	 * support for low contention concurrent putnext.
496 	 */
497 	ciputctrl_t	*sq_ciputctrl;
498 	uint_t		sq_nciputctrl;
499 	/*
500 	 * Counter for the number of threads wanting to become exclusive.
501 	 */
502 	uint_t		sq_needexcl;
503 	/*
504 	 * These two fields are used for scheduling a syncq for
505 	 * background processing. The sq_svcflag is protected by
506 	 * SQLOCK lock.
507 	 */
508 	struct syncq	*sq_next;	/* for syncq scheduling */
509 	void *		sq_servid;
510 	uint_t		sq_servcount;	/* # pending background threads */
511 	uint_t		sq_svcflags;	/* Scheduling flags	*/
512 	clock_t		sq_tstamp;	/* Time when was enabled */
513 	/*
514 	 * Maximum priority of the queues on this syncq.
515 	 */
516 	pri_t		sq_pri;
517 };
518 typedef struct syncq syncq_t;
519 
520 /*
521  * sync queue scheduling flags (for sq_svcflags).
522  */
523 #define	SQ_SERVICE	0x1		/* being serviced */
524 #define	SQ_BGTHREAD	0x2		/* awaiting service by bg thread */
525 #define	SQ_DISABLED	0x4		/* don't put syncq in service list */
526 
527 /*
528  * FASTPUT bit in sd_count/putcount.
529  */
530 #define	SQ_FASTPUT	0x8000
531 #define	SQ_FASTMASK	0x7FFF
532 
533 /*
534  * sync queue state flags
535  */
536 #define	SQ_EXCL		0x0001		/* exclusive access to inner */
537 					/*	perimeter */
538 #define	SQ_BLOCKED	0x0002		/* qprocsoff */
539 #define	SQ_FROZEN	0x0004		/* freezestr */
540 #define	SQ_WRITER	0x0008		/* qwriter(OUTER) pending or running */
541 #define	SQ_MESSAGES	0x0010		/* messages on syncq */
542 #define	SQ_WANTWAKEUP	0x0020		/* do cv_broadcast on sq_wait */
543 #define	SQ_WANTEXWAKEUP	0x0040		/* do cv_broadcast on sq_exitwait */
544 #define	SQ_EVENTS	0x0080		/* Events pending */
545 #define	SQ_QUEUED	(SQ_MESSAGES | SQ_EVENTS)
546 #define	SQ_FLAGMASK	0x00FF
547 
548 /*
549  * Test a queue to see if inner perimeter is exclusive.
550  */
551 #define	PERIM_EXCL(q)	((q)->q_syncq->sq_flags & SQ_EXCL)
552 
553 /*
554  * If any of these flags are set it is not possible for a thread to
555  * enter a put or service procedure. Instead it must either block
556  * or put the message on the syncq.
557  */
558 #define	SQ_GOAWAY	(SQ_EXCL|SQ_BLOCKED|SQ_FROZEN|SQ_WRITER|\
559 			SQ_QUEUED)
560 /*
561  * If any of these flags are set it not possible to drain the syncq
562  */
563 #define	SQ_STAYAWAY	(SQ_BLOCKED|SQ_FROZEN|SQ_WRITER)
564 
565 /*
566  * Flags to trigger syncq tail processing.
567  */
568 #define	SQ_TAIL		(SQ_QUEUED|SQ_WANTWAKEUP|SQ_WANTEXWAKEUP)
569 
570 /*
571  * Syncq types (stored in sq_type)
572  * The SQ_TYPES_IN_FLAGS (ciput) are also stored in sq_flags
573  * for performance reasons. Thus these type values have to be in the low
574  * 16 bits and not conflict with the sq_flags values above.
575  *
576  * Notes:
577  *  - putnext() and put() assume that the put procedures have the highest
578  *    degree of concurrency. Thus if any of the SQ_CI* are set then SQ_CIPUT
579  *    has to be set. This restriction can be lifted by adding code to putnext
580  *    and put that check that sq_count == 0 like entersq does.
581  *  - putnext() and put() does currently not handle !SQ_COPUT
582  *  - In order to implement !SQ_COCB outer_enter has to be fixed so that
583  *    the callback can be cancelled while cv_waiting in outer_enter.
584  *  - If SQ_CISVC needs to be implemented, qprocsoff() needs to wait
585  *    for the currently running services to stop (wait for QINSERVICE
586  *    to go off). disable_svc called from qprcosoff disables only
587  *    services that will be run in future.
588  *
589  * All the SQ_CO flags are set when there is no outer perimeter.
590  */
591 #define	SQ_CIPUT	0x0100		/* Concurrent inner put proc */
592 #define	SQ_CISVC	0x0200		/* Concurrent inner svc proc */
593 #define	SQ_CIOC		0x0400		/* Concurrent inner open/close */
594 #define	SQ_CICB		0x0800		/* Concurrent inner callback */
595 #define	SQ_COPUT	0x1000		/* Concurrent outer put proc */
596 #define	SQ_COSVC	0x2000		/* Concurrent outer svc proc */
597 #define	SQ_COOC		0x4000		/* Concurrent outer open/close */
598 #define	SQ_COCB		0x8000		/* Concurrent outer callback */
599 
600 /* Types also kept in sq_flags for performance */
601 #define	SQ_TYPES_IN_FLAGS	(SQ_CIPUT)
602 
603 #define	SQ_CI		(SQ_CIPUT|SQ_CISVC|SQ_CIOC|SQ_CICB)
604 #define	SQ_CO		(SQ_COPUT|SQ_COSVC|SQ_COOC|SQ_COCB)
605 #define	SQ_TYPEMASK	(SQ_CI|SQ_CO)
606 
607 /*
608  * Flag combinations passed to entersq and leavesq to specify the type
609  * of entry point.
610  */
611 #define	SQ_PUT		(SQ_CIPUT|SQ_COPUT)
612 #define	SQ_SVC		(SQ_CISVC|SQ_COSVC)
613 #define	SQ_OPENCLOSE	(SQ_CIOC|SQ_COOC)
614 #define	SQ_CALLBACK	(SQ_CICB|SQ_COCB)
615 
616 /*
617  * Other syncq types which are not copied into flags.
618  */
619 #define	SQ_PERMOD	0x01		/* Syncq is PERMOD */
620 
621 /*
622  * Asynchronous callback qun*** flag.
623  * The mechanism these flags are used in is one where callbacks enter
624  * the perimeter thanks to framework support. To use this mechanism
625  * the q* and qun* flavors of the callback routines must be used.
626  * e.g. qtimeout and quntimeout. The synchronization provided by the flags
627  * avoids deadlocks between blocking qun* routines and the perimeter
628  * lock.
629  */
630 #define	SQ_CALLB_BYPASSED	0x01		/* bypassed callback fn */
631 
632 /*
633  * Cancel callback mask.
634  * The mask expands as the number of cancelable callback types grows
635  * Note - separate callback flag because different callbacks have
636  * overlapping id space.
637  */
638 #define	SQ_CALLB_CANCEL_MASK	(SQ_CANCEL_TOUT|SQ_CANCEL_BUFCALL)
639 
640 #define	SQ_CANCEL_TOUT		0x02		/* cancel timeout request */
641 #define	SQ_CANCEL_BUFCALL	0x04		/* cancel bufcall request */
642 
643 typedef struct callbparams {
644 	syncq_t		*cbp_sq;
645 	void		(*cbp_func)(void *);
646 	void		*cbp_arg;
647 	callbparams_id_t cbp_id;
648 	uint_t		cbp_flags;
649 	struct callbparams *cbp_next;
650 	size_t		cbp_size;
651 } callbparams_t;
652 
653 typedef struct strbufcall {
654 	void		(*bc_func)(void *);
655 	void		*bc_arg;
656 	size_t		bc_size;
657 	bufcall_id_t	bc_id;
658 	struct strbufcall *bc_next;
659 	kthread_id_t	bc_executor;
660 } strbufcall_t;
661 
662 /*
663  * Structure of list of processes to be sent SIGPOLL/SIGURG signal
664  * on request.  The valid S_* events are defined in stropts.h.
665  */
666 typedef struct strsig {
667 	struct pid	*ss_pidp;	/* pid/pgrp pointer */
668 	pid_t		ss_pid;		/* positive pid, negative pgrp */
669 	int		ss_events;	/* S_* events */
670 	struct strsig	*ss_next;
671 } strsig_t;
672 
673 /*
674  * bufcall list
675  */
676 struct bclist {
677 	strbufcall_t	*bc_head;
678 	strbufcall_t	*bc_tail;
679 };
680 
681 /*
682  * Structure used to track mux links and unlinks.
683  */
684 struct mux_node {
685 	major_t		 mn_imaj;	/* internal major device number */
686 	uint16_t	 mn_indegree;	/* number of incoming edges */
687 	struct mux_node *mn_originp;	/* where we came from during search */
688 	struct mux_edge *mn_startp;	/* where search left off in mn_outp */
689 	struct mux_edge *mn_outp;	/* list of outgoing edges */
690 	uint_t		 mn_flags;	/* see below */
691 };
692 
693 /*
694  * Flags for mux_nodes.
695  */
696 #define	VISITED	1
697 
698 /*
699  * Edge structure - a list of these is hung off the
700  * mux_node to represent the outgoing edges.
701  */
702 struct mux_edge {
703 	struct mux_node	*me_nodep;	/* edge leads to this node */
704 	struct mux_edge	*me_nextp;	/* next edge */
705 	int		 me_muxid;	/* id of link */
706 	dev_t		 me_dev;	/* dev_t - used for kernel PUNLINK */
707 };
708 
709 /*
710  * Queue info
711  *
712  * The syncq is included here to reduce memory fragmentation
713  * for kernel memory allocators that only allocate in sizes that are
714  * powers of two. If the kernel memory allocator changes this should
715  * be revisited.
716  */
717 typedef struct queinfo {
718 	struct queue	qu_rqueue;	/* read queue - must be first */
719 	struct queue	qu_wqueue;	/* write queue - must be second */
720 	struct syncq	qu_syncq;	/* syncq - must be third */
721 } queinfo_t;
722 
723 /*
724  * Multiplexed streams info
725  */
726 typedef struct linkinfo {
727 	struct linkblk	li_lblk;	/* must be first */
728 	struct file	*li_fpdown;	/* file pointer for lower stream */
729 	struct linkinfo	*li_next;	/* next in list */
730 	struct linkinfo *li_prev;	/* previous in list */
731 } linkinfo_t;
732 
733 /*
734  * List of syncq's used by freeezestr/unfreezestr
735  */
736 typedef struct syncql {
737 	struct syncql	*sql_next;
738 	syncq_t		*sql_sq;
739 } syncql_t;
740 
741 typedef struct sqlist {
742 	syncql_t	*sqlist_head;
743 	size_t		sqlist_size;		/* structure size in bytes */
744 	size_t		sqlist_index;		/* next free entry in array */
745 	syncql_t	sqlist_array[4];	/* 4 or more entries */
746 } sqlist_t;
747 
748 typedef struct perdm {
749 	struct perdm		*dm_next;
750 	syncq_t			*dm_sq;
751 	struct streamtab	*dm_str;
752 	uint_t			dm_ref;
753 } perdm_t;
754 
755 #define	NEED_DM(dmp, qflag) \
756 	(dmp == NULL && (qflag & (QPERMOD | QMTOUTPERIM)))
757 
758 /*
759  * fmodsw_impl_t is used within the kernel. fmodsw is used by
760  * the modules/drivers. The information is copied from fmodsw
761  * defined in the module/driver into the fmodsw_impl_t structure
762  * during the module/driver initialization.
763  */
764 typedef struct fmodsw_impl	fmodsw_impl_t;
765 
766 struct fmodsw_impl {
767 	fmodsw_impl_t		*f_next;
768 	char			f_name[FMNAMESZ + 1];
769 	struct streamtab	*f_str;
770 	uint32_t		f_qflag;
771 	uint32_t		f_sqtype;
772 	perdm_t			*f_dmp;
773 	uint32_t		f_ref;
774 	uint32_t		f_hits;
775 };
776 
777 typedef enum {
778 	FMODSW_HOLD =	0x00000001,
779 	FMODSW_LOAD =	0x00000002
780 } fmodsw_flags_t;
781 
782 typedef struct cdevsw_impl {
783 	struct streamtab	*d_str;
784 	uint32_t		d_qflag;
785 	uint32_t		d_sqtype;
786 	perdm_t			*d_dmp;
787 } cdevsw_impl_t;
788 
789 /*
790  * Enumeration of the types of access that can be requested for a
791  * controlling terminal under job control.
792  */
793 enum jcaccess {
794 	JCREAD,			/* read data on a ctty */
795 	JCWRITE,		/* write data to a ctty */
796 	JCSETP,			/* set ctty parameters */
797 	JCGETP			/* get ctty parameters */
798 };
799 
800 struct str_stack {
801 	netstack_t	*ss_netstack;	/* Common netstack */
802 
803 	kmutex_t	ss_sad_lock;	/* autopush lock */
804 	mod_hash_t	*ss_sad_hash;
805 	size_t		ss_sad_hash_nchains;
806 	struct saddev	*ss_saddev;	/* sad device array */
807 	int		ss_sadcnt;	/* number of sad devices */
808 
809 	int		ss_devcnt;	/* number of mux_nodes */
810 	struct mux_node	*ss_mux_nodes;	/* mux info for cycle checking */
811 };
812 typedef struct str_stack str_stack_t;
813 
814 /*
815  * Finding related queues
816  */
817 #define	STREAM(q)	((q)->q_stream)
818 #define	SQ(rq)		((syncq_t *)((rq) + 2))
819 
820 /*
821  * Get the module/driver name for a queue.  Since some queues don't have
822  * q_info structures (e.g., see log_makeq()), fall back to "?".
823  */
824 #define	Q2NAME(q) \
825 	(((q)->q_qinfo != NULL && (q)->q_qinfo->qi_minfo->mi_idname != NULL) ? \
826 	(q)->q_qinfo->qi_minfo->mi_idname : "?")
827 
828 /*
829  * Locking macros
830  */
831 #define	QLOCK(q)	(&(q)->q_lock)
832 #define	SQLOCK(sq)	(&(sq)->sq_lock)
833 
834 #define	STREAM_PUTLOCKS_ENTER(stp) {					       \
835 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
836 		if ((stp)->sd_ciputctrl != NULL) {			       \
837 			int i;						       \
838 			int nlocks = (stp)->sd_nciputctrl;		       \
839 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
840 			for (i = 0; i <= nlocks; i++) {			       \
841 				mutex_enter(&cip[i].ciputctrl_lock);	       \
842 			}						       \
843 		}							       \
844 	}
845 
846 #define	STREAM_PUTLOCKS_EXIT(stp) {					       \
847 		ASSERT(MUTEX_HELD(&(stp)->sd_lock));			       \
848 		if ((stp)->sd_ciputctrl != NULL) {			       \
849 			int i;						       \
850 			int nlocks = (stp)->sd_nciputctrl;		       \
851 			ciputctrl_t *cip = (stp)->sd_ciputctrl;		       \
852 			for (i = 0; i <= nlocks; i++) {			       \
853 				mutex_exit(&cip[i].ciputctrl_lock);	       \
854 			}						       \
855 		}							       \
856 	}
857 
858 #define	SQ_PUTLOCKS_ENTER(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 			}						       \
868 		}							       \
869 	}
870 
871 #define	SQ_PUTLOCKS_EXIT(sq) {						       \
872 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
873 		if ((sq)->sq_ciputctrl != NULL) {			       \
874 			int i;						       \
875 			int nlocks = (sq)->sq_nciputctrl;		       \
876 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
877 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
878 			for (i = 0; i <= nlocks; i++) {			       \
879 				mutex_exit(&cip[i].ciputctrl_lock);	       \
880 			}						       \
881 		}							       \
882 	}
883 
884 #define	SQ_PUTCOUNT_SETFAST(sq) {					\
885 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				\
886 		if ((sq)->sq_ciputctrl != NULL) {			\
887 			int i;						\
888 			int nlocks = (sq)->sq_nciputctrl;		\
889 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		\
890 			ASSERT((sq)->sq_type & SQ_CIPUT);		\
891 			for (i = 0; i <= nlocks; i++) {			\
892 				mutex_enter(&cip[i].ciputctrl_lock);	\
893 				cip[i].ciputctrl_count |= SQ_FASTPUT;	\
894 				mutex_exit(&cip[i].ciputctrl_lock);	\
895 			}						\
896 		}							\
897 	}
898 
899 #define	SQ_PUTCOUNT_CLRFAST(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 				mutex_enter(&cip[i].ciputctrl_lock);	\
908 				cip[i].ciputctrl_count &= ~SQ_FASTPUT;	\
909 				mutex_exit(&cip[i].ciputctrl_lock);	\
910 			}						\
911 		}							\
912 	}
913 
914 
915 #ifdef	DEBUG
916 
917 #define	SQ_PUTLOCKS_HELD(sq) {						       \
918 		ASSERT(MUTEX_HELD(SQLOCK(sq)));				       \
919 		if ((sq)->sq_ciputctrl != NULL) {			       \
920 			int i;						       \
921 			int nlocks = (sq)->sq_nciputctrl;		       \
922 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
923 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
924 			for (i = 0; i <= nlocks; i++) {			       \
925 				ASSERT(MUTEX_HELD(&cip[i].ciputctrl_lock));    \
926 			}						       \
927 		}							       \
928 	}
929 
930 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) {				       \
931 		if ((sq)->sq_ciputctrl != NULL) {			       \
932 			int i;						       \
933 			uint_t count = 0;				       \
934 			int ncounts = (sq)->sq_nciputctrl;		       \
935 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
936 			for (i = 0; i <= ncounts; i++) {		       \
937 				count +=				       \
938 				    (((sq)->sq_ciputctrl[i].ciputctrl_count) & \
939 				    SQ_FASTMASK);			       \
940 			}						       \
941 			ASSERT(count == (countcheck));			       \
942 		}							       \
943 	}
944 
945 #define	SUMCHECK_CIPUTCTRL_COUNTS(ciput, nciput, countcheck) {		       \
946 		int i;							       \
947 		uint_t count = 0;					       \
948 		ASSERT((ciput) != NULL);				       \
949 		for (i = 0; i <= (nciput); i++) {			       \
950 			count += (((ciput)[i].ciputctrl_count) &	       \
951 			    SQ_FASTMASK);				       \
952 		}							       \
953 		ASSERT(count == (countcheck));				       \
954 	}
955 
956 #else	/* DEBUG */
957 
958 #define	SQ_PUTLOCKS_HELD(sq)
959 #define	SUMCHECK_SQ_PUTCOUNTS(sq, countcheck)
960 #define	SUMCHECK_CIPUTCTRL_COUNTS(sq, nciput, countcheck)
961 
962 #endif	/* DEBUG */
963 
964 #define	SUM_SQ_PUTCOUNTS(sq, count) {					       \
965 		if ((sq)->sq_ciputctrl != NULL) {			       \
966 			int i;						       \
967 			int ncounts = (sq)->sq_nciputctrl;		       \
968 			ciputctrl_t *cip = (sq)->sq_ciputctrl;		       \
969 			ASSERT((sq)->sq_type & SQ_CIPUT);		       \
970 			for (i = 0; i <= ncounts; i++) {		       \
971 				(count) += ((cip[i].ciputctrl_count) &	       \
972 				    SQ_FASTMASK);			       \
973 			}						       \
974 		}							       \
975 	}
976 
977 #define	CLAIM_QNEXT_LOCK(stp)	mutex_enter(&(stp)->sd_lock)
978 #define	RELEASE_QNEXT_LOCK(stp)	mutex_exit(&(stp)->sd_lock)
979 
980 /*
981  * syncq message manipulation macros.
982  */
983 /*
984  * Put a message on the queue syncq.
985  * Assumes QLOCK held.
986  */
987 #define	SQPUT_MP(qp, mp)						\
988 	{								\
989 		qp->q_syncqmsgs++;					\
990 		if (qp->q_sqhead == NULL) {				\
991 			qp->q_sqhead = qp->q_sqtail = mp;		\
992 		} else {						\
993 			qp->q_sqtail->b_next = mp;			\
994 			qp->q_sqtail = mp;				\
995 		}							\
996 		set_qfull(qp);						\
997 	}
998 
999 /*
1000  * Miscellaneous parameters and flags.
1001  */
1002 
1003 /*
1004  * Default timeout in milliseconds for ioctls and close
1005  */
1006 #define	STRTIMOUT 15000
1007 
1008 /*
1009  * Flag values for stream io
1010  */
1011 #define	WRITEWAIT	0x1	/* waiting for write event */
1012 #define	READWAIT	0x2	/* waiting for read event */
1013 #define	NOINTR		0x4	/* error is not to be set for signal */
1014 #define	GETWAIT		0x8	/* waiting for getmsg event */
1015 
1016 /*
1017  * These flags need to be unique for stream io name space
1018  * and copy modes name space.  These flags allow strwaitq
1019  * and strdoioctl to proceed as if signals or errors on the stream
1020  * head have not occurred; i.e. they will be detected by some other
1021  * means.
1022  * STR_NOSIG does not allow signals to interrupt the call
1023  * STR_NOERROR does not allow stream head read, write or hup errors to
1024  * affect the call.  When used with strdoioctl(), if a previous ioctl
1025  * is pending and times out, STR_NOERROR will cause strdoioctl() to not
1026  * return ETIME. If, however, the requested ioctl times out, ETIME
1027  * will be returned (use ic_timout instead)
1028  * STR_PEEK is used to inform strwaitq that the reader is peeking at data
1029  * and that a non-persistent error should not be cleared.
1030  * STR_DELAYERR is used to inform strwaitq that it should not check errors
1031  * after being awoken since, in addition to an error, there might also be
1032  * data queued on the stream head read queue.
1033  */
1034 #define	STR_NOSIG	0x10	/* Ignore signals during strdoioctl/strwaitq */
1035 #define	STR_NOERROR	0x20	/* Ignore errors during strdoioctl/strwaitq */
1036 #define	STR_PEEK	0x40	/* Peeking behavior on non-persistent errors */
1037 #define	STR_DELAYERR	0x80	/* Do not check errors on return */
1038 
1039 /*
1040  * Copy modes for tty and I_STR ioctls
1041  */
1042 #define	U_TO_K	01			/* User to Kernel */
1043 #define	K_TO_K  02			/* Kernel to Kernel */
1044 
1045 /*
1046  * Mux defines.
1047  */
1048 #define	LINKNORMAL	0x01		/* normal mux link */
1049 #define	LINKPERSIST	0x02		/* persistent mux link */
1050 #define	LINKTYPEMASK	0x03		/* bitmask of all link types */
1051 #define	LINKCLOSE	0x04		/* unlink from strclose */
1052 
1053 /*
1054  * Definitions of Streams macros and function interfaces.
1055  */
1056 
1057 /*
1058  * Obsolete queue scheduling macros. They are not used anymore, but still kept
1059  * here for 3-d party modules and drivers who might still use them.
1060  */
1061 #define	setqsched()
1062 #define	qready()	1
1063 
1064 #ifdef _KERNEL
1065 #define	runqueues()
1066 #define	queuerun()
1067 #endif
1068 
1069 /* compatibility module for style 2 drivers with DR race condition */
1070 #define	DRMODNAME	"drcompat"
1071 
1072 /*
1073  * Macros dealing with mux_nodes.
1074  */
1075 #define	MUX_VISIT(X)	((X)->mn_flags |= VISITED)
1076 #define	MUX_CLEAR(X)	((X)->mn_flags &= (~VISITED)); \
1077 			((X)->mn_originp = NULL)
1078 #define	MUX_DIDVISIT(X)	((X)->mn_flags & VISITED)
1079 
1080 
1081 /*
1082  * Twisted stream macros
1083  */
1084 #define	STRMATED(X)	((X)->sd_flag & STRMATE)
1085 #define	STRLOCKMATES(X)	if (&((X)->sd_lock) > &(((X)->sd_mate)->sd_lock)) { \
1086 				mutex_enter(&((X)->sd_lock)); \
1087 				mutex_enter(&(((X)->sd_mate)->sd_lock));  \
1088 			} else {  \
1089 				mutex_enter(&(((X)->sd_mate)->sd_lock)); \
1090 				mutex_enter(&((X)->sd_lock)); \
1091 			}
1092 #define	STRUNLOCKMATES(X)	mutex_exit(&((X)->sd_lock)); \
1093 			mutex_exit(&(((X)->sd_mate)->sd_lock))
1094 
1095 #if defined(_KERNEL) || defined(_FAKE_KERNEL)
1096 
1097 extern void strinit(void);
1098 extern int strdoioctl(struct stdata *, struct strioctl *, int, int,
1099     cred_t *, int *);
1100 extern void strsendsig(struct strsig *, int, uchar_t, int);
1101 extern void str_sendsig(vnode_t *, int, uchar_t, int);
1102 extern void strhup(struct stdata *);
1103 extern int qattach(queue_t *, dev_t *, int, cred_t *, fmodsw_impl_t *,
1104     boolean_t);
1105 extern int qreopen(queue_t *, dev_t *, int, cred_t *);
1106 extern void qdetach(queue_t *, int, int, cred_t *, boolean_t);
1107 extern void enterq(queue_t *);
1108 extern void leaveq(queue_t *);
1109 extern int putiocd(mblk_t *, caddr_t, int, cred_t *);
1110 extern int getiocd(mblk_t *, caddr_t, int);
1111 extern struct linkinfo *alloclink(queue_t *, queue_t *, struct file *);
1112 extern void lbfree(struct linkinfo *);
1113 extern int linkcycle(stdata_t *, stdata_t *, str_stack_t *);
1114 extern struct linkinfo *findlinks(stdata_t *, int, int, str_stack_t *);
1115 extern queue_t *getendq(queue_t *);
1116 extern int mlink(vnode_t *, int, int, cred_t *, int *, int);
1117 extern int mlink_file(vnode_t *, int, struct file *, cred_t *, int *, int);
1118 extern int munlink(struct stdata *, struct linkinfo *, int, cred_t *, int *,
1119     str_stack_t *);
1120 extern int munlinkall(struct stdata *, int, cred_t *, int *, str_stack_t *);
1121 extern void mux_addedge(stdata_t *, stdata_t *, int, str_stack_t *);
1122 extern void mux_rmvedge(stdata_t *, int, str_stack_t *);
1123 extern int devflg_to_qflag(struct streamtab *, uint32_t, uint32_t *,
1124     uint32_t *);
1125 extern void setq(queue_t *, struct qinit *, struct qinit *, perdm_t *,
1126     uint32_t, uint32_t, boolean_t);
1127 extern perdm_t *hold_dm(struct streamtab *, uint32_t, uint32_t);
1128 extern void rele_dm(perdm_t *);
1129 extern int strmakectl(struct strbuf *, int32_t, int32_t, mblk_t **);
1130 extern int strmakedata(ssize_t *, struct uio *, stdata_t *, int32_t, mblk_t **);
1131 extern int strmakemsg(struct strbuf *, ssize_t *, struct uio *,
1132     struct stdata *, int32_t, mblk_t **);
1133 extern int strgetmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t *,
1134     int *, int, rval_t *);
1135 extern int strputmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t,
1136     int flag, int fmode);
1137 extern int strstartplumb(struct stdata *, int, int);
1138 extern void strendplumb(struct stdata *);
1139 extern int stropen(struct vnode *, dev_t *, int, cred_t *);
1140 extern int strclose(struct vnode *, int, cred_t *);
1141 extern int strpoll(register struct stdata *, short, int, short *,
1142     struct pollhead **);
1143 extern void strclean(struct vnode *);
1144 extern void str_cn_clean();	/* XXX hook for consoles signal cleanup */
1145 extern int strwrite(struct vnode *, struct uio *, cred_t *);
1146 extern int strwrite_common(struct vnode *, struct uio *, cred_t *, int);
1147 extern int strread(struct vnode *, struct uio *, cred_t *);
1148 extern int strioctl(struct vnode *, int, intptr_t, int, int, cred_t *, int *);
1149 extern int strrput(queue_t *, mblk_t *);
1150 extern int strrput_nondata(queue_t *, mblk_t *);
1151 extern mblk_t *strrput_proto(vnode_t *, mblk_t *,
1152     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1153 extern mblk_t *strrput_misc(vnode_t *, mblk_t *,
1154     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1155 extern int getiocseqno(void);
1156 extern int strwaitbuf(size_t, int);
1157 extern int strwaitq(stdata_t *, int, ssize_t, int, clock_t, int *);
1158 extern struct stdata *shalloc(queue_t *);
1159 extern void shfree(struct stdata *s);
1160 extern queue_t *allocq(void);
1161 extern void freeq(queue_t *);
1162 extern qband_t *allocband(void);
1163 extern void freeband(qband_t *);
1164 extern void freebs_enqueue(mblk_t *, dblk_t *);
1165 extern void setqback(queue_t *, unsigned char);
1166 extern int strcopyin(void *, void *, size_t, int);
1167 extern int strcopyout(void *, void *, size_t, int);
1168 extern void strsignal(struct stdata *, int, int32_t);
1169 extern clock_t str_cv_wait(kcondvar_t *, kmutex_t *, clock_t, int);
1170 extern void disable_svc(queue_t *);
1171 extern void enable_svc(queue_t *);
1172 extern void remove_runlist(queue_t *);
1173 extern void wait_svc(queue_t *);
1174 extern void backenable(queue_t *, uchar_t);
1175 extern void set_qend(queue_t *);
1176 extern int strgeterr(stdata_t *, int32_t, int);
1177 extern void qenable_locked(queue_t *);
1178 extern mblk_t *getq_noenab(queue_t *, ssize_t);
1179 extern void rmvq_noenab(queue_t *, mblk_t *);
1180 extern void qbackenable(queue_t *, uchar_t);
1181 extern void set_qfull(queue_t *);
1182 
1183 extern void strblock(queue_t *);
1184 extern void strunblock(queue_t *);
1185 extern int qclaimed(queue_t *);
1186 extern int straccess(struct stdata *, enum jcaccess);
1187 
1188 extern void entersq(syncq_t *, int);
1189 extern void leavesq(syncq_t *, int);
1190 extern void claimq(queue_t *);
1191 extern void releaseq(queue_t *);
1192 extern void claimstr(queue_t *);
1193 extern void releasestr(queue_t *);
1194 extern void removeq(queue_t *);
1195 extern void insertq(struct stdata *, queue_t *);
1196 extern void drain_syncq(syncq_t *);
1197 extern void qfill_syncq(syncq_t *, queue_t *, mblk_t *);
1198 extern void qdrain_syncq(syncq_t *, queue_t *);
1199 extern int flush_syncq(syncq_t *, queue_t *);
1200 extern void wait_sq_svc(syncq_t *);
1201 
1202 extern void outer_enter(syncq_t *, uint16_t);
1203 extern void outer_exit(syncq_t *);
1204 extern void qwriter_inner(queue_t *, mblk_t *, void (*)());
1205 extern void qwriter_outer(queue_t *, mblk_t *, void (*)());
1206 
1207 extern callbparams_t *callbparams_alloc(syncq_t *, void (*)(void *),
1208     void *, int);
1209 extern void callbparams_free(syncq_t *, callbparams_t *);
1210 extern void callbparams_free_id(syncq_t *, callbparams_id_t, int32_t);
1211 extern void qcallbwrapper(void *);
1212 
1213 extern mblk_t *esballoc_wait(unsigned char *, size_t, uint_t, frtn_t *);
1214 extern mblk_t *esballoca(unsigned char *, size_t, uint_t, frtn_t *);
1215 extern mblk_t *desballoca(unsigned char *, size_t, uint_t, frtn_t *);
1216 extern int do_sendfp(struct stdata *, struct file *, struct cred *);
1217 extern int frozenstr(queue_t *);
1218 extern size_t xmsgsize(mblk_t *);
1219 
1220 extern void putnext_tail(syncq_t *, queue_t *, uint32_t);
1221 extern void stream_willservice(stdata_t *);
1222 extern void stream_runservice(stdata_t *);
1223 
1224 extern void strmate(vnode_t *, vnode_t *);
1225 extern queue_t *strvp2wq(vnode_t *);
1226 extern vnode_t *strq2vp(queue_t *);
1227 extern mblk_t *allocb_wait(size_t, uint_t, uint_t, int *);
1228 extern mblk_t *allocb_cred(size_t, cred_t *, pid_t);
1229 extern mblk_t *allocb_cred_wait(size_t, uint_t, int *, cred_t *, pid_t);
1230 extern mblk_t *allocb_tmpl(size_t, const mblk_t *);
1231 extern mblk_t *allocb_tryhard(size_t);
1232 extern void mblk_copycred(mblk_t *, const mblk_t *);
1233 extern void mblk_setcred(mblk_t *, cred_t *, pid_t);
1234 extern cred_t *msg_getcred(const mblk_t *, pid_t *);
1235 extern struct ts_label_s *msg_getlabel(const mblk_t *);
1236 extern cred_t *msg_extractcred(mblk_t *, pid_t *);
1237 extern void strpollwakeup(vnode_t *, short);
1238 extern int putnextctl_wait(queue_t *, int);
1239 
1240 extern int kstrputmsg(struct vnode *, mblk_t *, struct uio *, ssize_t,
1241     unsigned char, int, int);
1242 extern int kstrgetmsg(struct vnode *, mblk_t **, struct uio *,
1243     unsigned char *, int *, clock_t, rval_t *);
1244 
1245 extern void strsetrerror(vnode_t *, int, int, errfunc_t);
1246 extern void strsetwerror(vnode_t *, int, int, errfunc_t);
1247 extern void strseteof(vnode_t *, int);
1248 extern void strflushrq(vnode_t *, int);
1249 extern void strsetrputhooks(vnode_t *, uint_t, msgfunc_t, msgfunc_t);
1250 extern void strsetwputhooks(vnode_t *, uint_t, clock_t);
1251 extern void strsetrwputdatahooks(vnode_t *, msgfunc_t, msgfunc_t);
1252 extern int strwaitmark(vnode_t *);
1253 extern void strsignal_nolock(stdata_t *, int, uchar_t);
1254 
1255 struct multidata_s;
1256 struct pdesc_s;
1257 extern int hcksum_assoc(mblk_t *, struct multidata_s *, struct pdesc_s  *,
1258     uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, int);
1259 extern void hcksum_retrieve(mblk_t *, struct multidata_s *, struct pdesc_s *,
1260     uint32_t *, uint32_t *, uint32_t *, uint32_t *, uint32_t *);
1261 extern void lso_info_set(mblk_t *, uint32_t, uint32_t);
1262 extern void lso_info_cleanup(mblk_t *);
1263 extern unsigned int bcksum(uchar_t *, int, unsigned int);
1264 extern boolean_t is_vmloaned_mblk(mblk_t *, struct multidata_s *,
1265     struct pdesc_s *);
1266 
1267 extern int fmodsw_register(const char *, struct streamtab *, int);
1268 extern int fmodsw_unregister(const char *);
1269 extern fmodsw_impl_t *fmodsw_find(const char *, fmodsw_flags_t);
1270 extern void fmodsw_rele(fmodsw_impl_t *);
1271 
1272 extern void freemsgchain(mblk_t *);
1273 extern mblk_t *copymsgchain(mblk_t *);
1274 
1275 extern mblk_t *mcopyinuio(struct stdata *, uio_t *, ssize_t, ssize_t, int *);
1276 
1277 /*
1278  * shared or externally configured data structures
1279  */
1280 extern ssize_t strmsgsz;		/* maximum stream message size */
1281 extern ssize_t strctlsz;		/* maximum size of ctl message */
1282 extern int nstrpush;			/* maximum number of pushes allowed */
1283 
1284 /*
1285  * Bufcalls related variables.
1286  */
1287 extern struct bclist strbcalls;		/* List of bufcalls */
1288 extern kmutex_t	strbcall_lock;		/* Protects the list of bufcalls */
1289 extern kcondvar_t strbcall_cv;		/* Signaling when a bufcall is added */
1290 extern kcondvar_t bcall_cv;	/* wait of executing bufcall completes */
1291 
1292 extern frtn_t frnop;
1293 
1294 extern struct kmem_cache *ciputctrl_cache;
1295 extern int n_ciputctrl;
1296 extern int max_n_ciputctrl;
1297 extern int min_n_ciputctrl;
1298 
1299 extern cdevsw_impl_t *devimpl;
1300 
1301 /*
1302  * esballoc queue for throttling
1303  */
1304 typedef struct esb_queue {
1305 	kmutex_t	eq_lock;
1306 	uint_t		eq_len;		/* number of queued messages */
1307 	mblk_t		*eq_head;	/* head of queue */
1308 	mblk_t		*eq_tail;	/* tail of queue */
1309 	uint_t		eq_flags;	/* esballoc queue flags */
1310 } esb_queue_t;
1311 
1312 /*
1313  * esballoc flags for queue processing.
1314  */
1315 #define	ESBQ_PROCESSING	0x01	/* queue is being processed */
1316 #define	ESBQ_TIMER	0x02	/* timer is active */
1317 
1318 extern void esballoc_queue_init(void);
1319 
1320 #endif	/* _KERNEL */
1321 
1322 /*
1323  * Note: Use of these macros are restricted to kernel/unix and
1324  * intended for the STREAMS framework.
1325  * All modules/drivers should include sys/ddi.h.
1326  *
1327  * Finding related queues
1328  */
1329 #define		_OTHERQ(q)	((q)->q_flag&QREADR? (q)+1: (q)-1)
1330 #define		_WR(q)		((q)->q_flag&QREADR? (q)+1: (q))
1331 #define		_RD(q)		((q)->q_flag&QREADR? (q): (q)-1)
1332 #define		_SAMESTR(q)	(!((q)->q_flag & QEND))
1333 
1334 /*
1335  * These are also declared here for modules/drivers that erroneously
1336  * include strsubr.h after ddi.h or fail to include ddi.h at all.
1337  */
1338 extern struct queue *OTHERQ(queue_t *); /* stream.h */
1339 extern struct queue *RD(queue_t *);
1340 extern struct queue *WR(queue_t *);
1341 extern int SAMESTR(queue_t *);
1342 
1343 /*
1344  * The following hardware checksum related macros are private
1345  * interfaces that are subject to change without notice.
1346  */
1347 #ifdef _KERNEL
1348 #define	DB_CKSUMSTART(mp)	((mp)->b_datap->db_cksumstart)
1349 #define	DB_CKSUMEND(mp)		((mp)->b_datap->db_cksumend)
1350 #define	DB_CKSUMSTUFF(mp)	((mp)->b_datap->db_cksumstuff)
1351 #define	DB_CKSUMFLAGS(mp)	((mp)->b_datap->db_struioun.cksum.flags)
1352 #define	DB_CKSUM16(mp)		((mp)->b_datap->db_cksum16)
1353 #define	DB_CKSUM32(mp)		((mp)->b_datap->db_cksum32)
1354 #define	DB_LSOFLAGS(mp)		((mp)->b_datap->db_struioun.cksum.flags)
1355 #define	DB_LSOMSS(mp)		((mp)->b_datap->db_struioun.cksum.pad)
1356 #endif	/* _KERNEL */
1357 
1358 #ifdef	__cplusplus
1359 }
1360 #endif
1361 
1362 
1363 #endif	/* _SYS_STRSUBR_H */
1364