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 #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 
  48 #ifdef  __cplusplus
  49 extern "C" {
  50 #endif
  51 
  52 /*
  53  * In general, the STREAMS locks are disjoint; they are only held
  54  * locally, and not simultaneously by a thread.  However, module
  55  * code, including at the stream head, requires some locks to be
  56  * acquired in order for its safety.
  57  *      1. Stream level claim.  This prevents the value of q_next
  58  *              from changing while module code is executing.
  59  *      2. Queue level claim.  This prevents the value of q_ptr
  60  *              from changing while put or service code is executing.
  61  *              In addition, it provides for queue single-threading
  62  *              for QPAIR and PERQ MT-safe modules.
  63  *      3. Stream head lock.  May be held by the stream head module
  64  *              to implement a read/write/open/close monitor.
  65  *         Note: that the only types of twisted stream supported are
  66  *         the pipe and transports which have read and write service
  67  *         procedures on both sides of the twist.
  68  *      4. Queue lock.  May be acquired by utility routines on
  69  *              behalf of a module.
  70  */
  71 
  72 /*
  73  * In general, sd_lock protects the consistency of the stdata
  74  * structure.  Additionally, it is used with sd_monitor
  75  * to implement an open/close monitor.  In particular, it protects
  76  * the following fields:
  77  *      sd_iocblk
  78  *      sd_flag
  79  *      sd_copyflag
  80  *      sd_iocid
  81  *      sd_iocwait
  82  *      sd_sidp
  83  *      sd_pgidp
  84  *      sd_wroff
  85  *      sd_tail
  86  *      sd_rerror
  87  *      sd_werror
  88  *      sd_pushcnt
  89  *      sd_sigflags
  90  *      sd_siglist
  91  *      sd_pollist
  92  *      sd_mark
  93  *      sd_closetime
  94  *      sd_wakeq
  95  *      sd_maxblk
  96  *
  97  * The following fields are modified only by the allocator, which
  98  * has exclusive access to them at that time:
  99  *      sd_wrq
 100  *      sd_strtab
 101  *
 102  * The following field is protected by the overlying file system
 103  * code, guaranteeing single-threading of opens:
 104  *      sd_vnode
 105  *
 106  * Stream-level locks should be acquired before any queue-level locks
 107  *      are acquired.
 108  *
 109  * The stream head write queue lock(sd_wrq) is used to protect the
 110  * fields qn_maxpsz and qn_minpsz because freezestr() which is
 111  * necessary for strqset() only gets the queue lock.
 112  */
 113 
 114 /*
 115  * Function types for the parameterized stream head.
 116  * The msgfunc_t takes the parameters:
 117  *      msgfunc(vnode_t *vp, mblk_t *mp, strwakeup_t *wakeups,
 118  *              strsigset_t *firstmsgsigs, strsigset_t *allmsgsigs,
 119  *              strpollset_t *pollwakeups);
 120  * It returns an optional message to be processed by the stream head.
 121  *
 122  * The parameters for errfunc_t are:
 123  *      errfunc(vnode *vp, int ispeek, int *clearerr);
 124  * It returns an errno and zero if there was no pending error.
 125  */
 126 typedef uint_t  strwakeup_t;
 127 typedef uint_t  strsigset_t;
 128 typedef short   strpollset_t;
 129 typedef uintptr_t callbparams_id_t;
 130 typedef mblk_t  *(*msgfunc_t)(vnode_t *, mblk_t *, strwakeup_t *,
 131                         strsigset_t *, strsigset_t *, strpollset_t *);
 132 typedef int     (*errfunc_t)(vnode_t *, int, int *);
 133 
 134 /*
 135  * Per stream sd_lock in putnext may be replaced by per cpu stream_putlocks
 136  * each living in a separate cache line. putnext/canputnext grabs only one of
 137  * stream_putlocks while strlock() (called on behalf of insertq()/removeq())
 138  * acquires all stream_putlocks. Normally stream_putlocks are only employed
 139  * for highly contended streams that have SQ_CIPUT queues in the critical path
 140  * (e.g. NFS/UDP stream).
 141  *
 142  * stream_putlocks are dynamically assigned to stdata structure through
 143  * sd_ciputctrl pointer possibly when a stream is already in use. Since
 144  * strlock() uses stream_putlocks only under sd_lock acquiring sd_lock when
 145  * assigning stream_putlocks to the stream ensures synchronization with
 146  * strlock().
 147  *
 148  * For lock ordering purposes stream_putlocks are treated as the extension of
 149  * sd_lock and are always grabbed right after grabbing sd_lock and released
 150  * right before releasing sd_lock except putnext/canputnext where only one of
 151  * stream_putlocks locks is used and where it is the first lock to grab.
 152  */
 153 
 154 typedef struct ciputctrl_str {
 155         union _ciput_un {
 156                 uchar_t pad[64];
 157                 struct _ciput_str {
 158                         kmutex_t        ciput_lck;
 159                         ushort_t        ciput_cnt;
 160                 } ciput_str;
 161         } ciput_un;
 162 } ciputctrl_t;
 163 
 164 #define ciputctrl_lock  ciput_un.ciput_str.ciput_lck
 165 #define ciputctrl_count ciput_un.ciput_str.ciput_cnt
 166 
 167 /*
 168  * Header for a stream: interface to rest of system.
 169  *
 170  * NOTE: While this is a consolidation-private structure, some unbundled and
 171  *       third-party products inappropriately make use of some of the fields.
 172  *       As such, please take care to not gratuitously change any offsets of
 173  *       existing members.
 174  */
 175 typedef struct stdata {
 176         struct queue    *sd_wrq;        /* write queue */
 177         struct msgb     *sd_iocblk;     /* return block for ioctl */
 178         struct vnode    *sd_vnode;      /* pointer to associated vnode */
 179         struct streamtab *sd_strtab;    /* pointer to streamtab for stream */
 180         uint_t          sd_flag;        /* state/flags */
 181         uint_t          sd_iocid;       /* ioctl id */
 182         struct pid      *sd_sidp;       /* controlling session info */
 183         struct pid      *sd_pgidp;      /* controlling process group info */
 184         ushort_t        sd_tail;        /* reserved space in written mblks */
 185         ushort_t        sd_wroff;       /* write offset */
 186         int             sd_rerror;      /* error to return on read ops */
 187         int             sd_werror;      /* error to return on write ops */
 188         int             sd_pushcnt;     /* number of pushes done on stream */
 189         int             sd_sigflags;    /* logical OR of all siglist events */
 190         struct strsig   *sd_siglist;    /* pid linked list to rcv SIGPOLL sig */
 191         struct pollhead sd_pollist;     /* list of all pollers to wake up */
 192         struct msgb     *sd_mark;       /* "marked" message on read queue */
 193         clock_t         sd_closetime;   /* time to wait to drain q in close */
 194         kmutex_t        sd_lock;        /* protect head consistency */
 195         kcondvar_t      sd_monitor;     /* open/close/push/pop monitor */
 196         kcondvar_t      sd_iocmonitor;  /* ioctl single-threading */
 197         kcondvar_t      sd_refmonitor;  /* sd_refcnt monitor */
 198         ssize_t         sd_qn_minpsz;   /* These two fields are a performance */
 199         ssize_t         sd_qn_maxpsz;   /* enhancements, cache the values in */
 200                                         /* the stream head so we don't have */
 201                                         /* to ask the module below the stream */
 202                                         /* head to get this information. */
 203         struct stdata   *sd_mate;       /* pointer to twisted stream mate */
 204         kthread_id_t    sd_freezer;     /* thread that froze stream */
 205         kmutex_t        sd_reflock;     /* Protects sd_refcnt */
 206         int             sd_refcnt;      /* number of claimstr */
 207         uint_t          sd_wakeq;       /* strwakeq()'s copy of sd_flag */
 208         struct queue    *sd_struiordq;  /* sync barrier struio() read queue */
 209         struct queue    *sd_struiowrq;  /* sync barrier struio() write queue */
 210         char            sd_struiodnak;  /* defer NAK of M_IOCTL by rput() */
 211         struct msgb     *sd_struionak;  /* pointer M_IOCTL mblk(s) to NAK */
 212         caddr_t         sd_t_audit_data; /* For audit purposes only */
 213         ssize_t         sd_maxblk;      /* maximum message block size */
 214         uint_t          sd_rput_opt;    /* options/flags for strrput */
 215         uint_t          sd_wput_opt;    /* options/flags for write/putmsg */
 216         uint_t          sd_read_opt;    /* options/flags for strread */
 217         msgfunc_t       sd_rprotofunc;  /* rput M_*PROTO routine */
 218         msgfunc_t       sd_rputdatafunc; /* read M_DATA routine */
 219         msgfunc_t       sd_rmiscfunc;   /* rput routine (non-data/proto) */
 220         msgfunc_t       sd_wputdatafunc; /* wput M_DATA routine */
 221         errfunc_t       sd_rderrfunc;   /* read side error callback */
 222         errfunc_t       sd_wrerrfunc;   /* write side error callback */
 223         /*
 224          * support for low contention concurrent putnext.
 225          */
 226         ciputctrl_t     *sd_ciputctrl;
 227         uint_t          sd_nciputctrl;
 228 
 229         int             sd_anchor;      /* position of anchor in stream */
 230         /*
 231          * Service scheduling at the stream head.
 232          */
 233         kmutex_t        sd_qlock;
 234         struct queue    *sd_qhead;      /* Head of queues to be serviced. */
 235         struct queue    *sd_qtail;      /* Tail of queues to be serviced. */
 236         void            *sd_servid;     /* Service ID for bckgrnd schedule */
 237         ushort_t        sd_svcflags;    /* Servicing flags */
 238         short           sd_nqueues;     /* Number of queues in the list */
 239         kcondvar_t      sd_qcv;         /* Waiters for qhead to become empty */
 240         kcondvar_t      sd_zcopy_wait;
 241         uint_t          sd_copyflag;    /* copy-related flags */
 242         zoneid_t        sd_anchorzone;  /* Allow removal from same zone only */
 243         struct msgb     *sd_cmdblk;     /* reply from _I_CMD */
 244 } stdata_t;
 245 
 246 /*
 247  * stdata servicing flags.
 248  */
 249 #define STRS_WILLSERVICE        0x01
 250 #define STRS_SCHEDULED          0x02
 251 
 252 #define STREAM_NEEDSERVICE(stp) ((stp)->sd_qhead != NULL)
 253 
 254 /*
 255  * stdata flag field defines
 256  */
 257 #define IOCWAIT         0x00000001      /* Someone is doing an ioctl */
 258 #define RSLEEP          0x00000002      /* Someone wants to read/recv msg */
 259 #define WSLEEP          0x00000004      /* Someone wants to write */
 260 #define STRPRI          0x00000008      /* An M_PCPROTO is at stream head */
 261 #define STRHUP          0x00000010      /* Device has vanished */
 262 #define STWOPEN         0x00000020      /* waiting for 1st open */
 263 #define STPLEX          0x00000040      /* stream is being multiplexed */
 264 #define STRISTTY        0x00000080      /* stream is a terminal */
 265 #define STRGETINPROG    0x00000100      /* (k)strgetmsg is running */
 266 #define IOCWAITNE       0x00000200      /* STR_NOERROR ioctl running */
 267 #define STRDERR         0x00000400      /* fatal read error from M_ERROR */
 268 #define STWRERR         0x00000800      /* fatal write error from M_ERROR */
 269 #define STRDERRNONPERSIST 0x00001000    /* nonpersistent read errors */
 270 #define STWRERRNONPERSIST 0x00002000    /* nonpersistent write errors */
 271 #define STRCLOSE        0x00004000      /* wait for a close to complete */
 272 #define SNDMREAD        0x00008000      /* used for read notification */
 273 #define OLDNDELAY       0x00010000      /* use old TTY semantics for */
 274                                         /* NDELAY reads and writes */
 275         /*              0x00020000         unused */
 276         /*              0x00040000         unused */
 277 #define STRTOSTOP       0x00080000      /* block background writes */
 278 #define STRCMDWAIT      0x00100000      /* someone is doing an _I_CMD */
 279         /*              0x00200000         unused */
 280 #define STRMOUNT        0x00400000      /* stream is mounted */
 281 #define STRNOTATMARK    0x00800000      /* Not at mark (when empty read q) */
 282 #define STRDELIM        0x01000000      /* generate delimited messages */
 283 #define STRATMARK       0x02000000      /* At mark (due to MSGMARKNEXT) */
 284 #define STZCNOTIFY      0x04000000      /* wait for zerocopy mblk to be acked */
 285 #define STRPLUMB        0x08000000      /* push/pop pending */
 286 #define STREOF          0x10000000      /* End-of-file indication */
 287 #define STREOPENFAIL    0x20000000      /* indicates if re-open has failed */
 288 #define STRMATE         0x40000000      /* this stream is a mate */
 289 #define STRHASLINKS     0x80000000      /* I_LINKs under this stream */
 290 
 291 /*
 292  * Copy-related flags (sd_copyflag), set by SO_COPYOPT.
 293  */
 294 #define STZCVMSAFE      0x00000001      /* safe to borrow file (segmapped) */
 295                                         /* pages instead of bcopy */
 296 #define STZCVMUNSAFE    0x00000002      /* unsafe to borrow file pages */
 297 #define STRCOPYCACHED   0x00000004      /* copy should NOT bypass cache */
 298 
 299 /*
 300  * Options and flags for strrput (sd_rput_opt)
 301  */
 302 #define SR_POLLIN       0x00000001      /* pollwakeup needed for band0 data */
 303 #define SR_SIGALLDATA   0x00000002      /* Send SIGPOLL for all M_DATA */
 304 #define SR_CONSOL_DATA  0x00000004      /* Consolidate M_DATA onto q_last */
 305 #define SR_IGN_ZEROLEN  0x00000008      /* Ignore zero-length M_DATA */
 306 
 307 /*
 308  * Options and flags for strwrite/strputmsg (sd_wput_opt)
 309  */
 310 #define SW_SIGPIPE      0x00000001      /* Send SIGPIPE for write error */
 311 #define SW_RECHECK_ERR  0x00000002      /* Recheck errors in strwrite loop */
 312 #define SW_SNDZERO      0x00000004      /* send 0-length msg down pipe/FIFO */
 313 
 314 /*
 315  * Options and flags for strread (sd_read_opt)
 316  */
 317 #define RD_MSGDIS       0x00000001      /* read msg discard */
 318 #define RD_MSGNODIS     0x00000002      /* read msg no discard */
 319 #define RD_PROTDAT      0x00000004      /* read M_[PC]PROTO contents as data */
 320 #define RD_PROTDIS      0x00000008      /* discard M_[PC]PROTO blocks and */
 321                                         /* retain data blocks */
 322 /*
 323  * Flags parameter for strsetrputhooks() and strsetwputhooks().
 324  * These flags define the interface for setting the above internal
 325  * flags in sd_rput_opt and sd_wput_opt.
 326  */
 327 #define SH_CONSOL_DATA  0x00000001      /* Consolidate M_DATA onto q_last */
 328 #define SH_SIGALLDATA   0x00000002      /* Send SIGPOLL for all M_DATA */
 329 #define SH_IGN_ZEROLEN  0x00000004      /* Drop zero-length M_DATA */
 330 
 331 #define SH_SIGPIPE      0x00000100      /* Send SIGPIPE for write error */
 332 #define SH_RECHECK_ERR  0x00000200      /* Recheck errors in strwrite loop */
 333 
 334 /*
 335  * Each queue points to a sync queue (the inner perimeter) which keeps
 336  * track of the number of threads that are inside a given queue (sq_count)
 337  * and also is used to implement the asynchronous putnext
 338  * (by queuing messages if the queue can not be entered.)
 339  *
 340  * Messages are queued on sq_head/sq_tail including deferred qwriter(INNER)
 341  * messages. The sq_head/sq_tail list is a singly-linked list with
 342  * b_queue recording the queue and b_prev recording the function to
 343  * be called (either the put procedure or a qwriter callback function.)
 344  *
 345  * The sq_count counter tracks the number of threads that are
 346  * executing inside the perimeter or (in the case of outer perimeters)
 347  * have some work queued for them relating to the perimeter. The sq_rmqcount
 348  * counter tracks the subset which are in removeq() (usually invoked from
 349  * qprocsoff(9F)).
 350  *
 351  * In addition a module writer can declare that the module has an outer
 352  * perimeter (by setting D_MTOUTPERIM) in which case all inner perimeter
 353  * syncq's for the module point (through sq_outer) to an outer perimeter
 354  * syncq. The outer perimeter consists of the doubly linked list (sq_onext and
 355  * sq_oprev) linking all the inner perimeter syncq's with out outer perimeter
 356  * syncq. This is used to implement qwriter(OUTER) (an asynchronous way of
 357  * getting exclusive access at the outer perimeter) and outer_enter/exit
 358  * which are used by the framework to acquire exclusive access to the outer
 359  * perimeter during open and close of modules that have set D_MTOUTPERIM.
 360  *
 361  * In the inner perimeter case sq_save is available for use by machine
 362  * dependent code. sq_head/sq_tail are used to queue deferred messages on
 363  * the inner perimeter syncqs and to queue become_writer requests on the
 364  * outer perimeter syncqs.
 365  *
 366  * Note: machine dependent optimized versions of putnext may depend
 367  * on the order of sq_flags and sq_count (so that they can e.g.
 368  * read these two fields in a single load instruction.)
 369  *
 370  * Per perimeter SQLOCK/sq_count in putnext/put may be replaced by per cpu
 371  * sq_putlocks/sq_putcounts each living in a separate cache line. Obviously
 372  * sq_putlock[x] protects sq_putcount[x]. putnext/put routine will grab only 1
 373  * of sq_putlocks and update only 1 of sq_putcounts. strlock() and many
 374  * other routines in strsubr.c and ddi.c will grab all sq_putlocks (as well as
 375  * SQLOCK) and figure out the count value as the sum of sq_count and all of
 376  * sq_putcounts. The idea is to make critical fast path -- putnext -- much
 377  * faster at the expense of much less often used slower path like
 378  * strlock(). One known case where entersq/strlock is executed pretty often is
 379  * SpecWeb but since IP is SQ_CIOC and socket TCP/IP stream is nextless
 380  * there's no need to grab multiple sq_putlocks and look at sq_putcounts. See
 381  * strsubr.c for more comments.
 382  *
 383  * Note regular SQLOCK and sq_count are still used in many routines
 384  * (e.g. entersq(), rwnext()) in the same way as before sq_putlocks were
 385  * introduced.
 386  *
 387  * To understand when all sq_putlocks need to be held and all sq_putcounts
 388  * need to be added up one needs to look closely at putnext code. Basically if
 389  * a routine like e.g. wait_syncq() needs to be sure that perimeter is empty
 390  * all sq_putlocks/sq_putcounts need to be held/added up. On the other hand
 391  * there's no need to hold all sq_putlocks and count all sq_putcounts in
 392  * routines like leavesq()/dropsq() and etc. since the are usually exit
 393  * counterparts of entersq/outer_enter() and etc. which have already either
 394  * prevented put entry poins from executing or did not care about put
 395  * entrypoints. entersq() doesn't need to care about sq_putlocks/sq_putcounts
 396  * if the entry point has a shared access since put has the highest degree of
 397  * concurrency and such entersq() does not intend to block out put
 398  * entrypoints.
 399  *
 400  * Before sq_putcounts were introduced the standard way to wait for perimeter
 401  * to become empty was:
 402  *
 403  *      mutex_enter(SQLOCK(sq));
 404  *      while (sq->sq_count > 0) {
 405  *              sq->sq_flags |= SQ_WANTWAKEUP;
 406  *              cv_wait(&sq->sq_wait, SQLOCK(sq));
 407  *      }
 408  *      mutex_exit(SQLOCK(sq));
 409  *
 410  * The new way is:
 411  *
 412  *      mutex_enter(SQLOCK(sq));
 413  *      count = sq->sq_count;
 414  *      SQ_PUTLOCKS_ENTER(sq);
 415  *      SUM_SQ_PUTCOUNTS(sq, count);
 416  *      while (count != 0) {
 417  *              sq->sq_flags |= SQ_WANTWAKEUP;
 418  *              SQ_PUTLOCKS_EXIT(sq);
 419  *              cv_wait(&sq->sq_wait, SQLOCK(sq));
 420  *              count = sq->sq_count;
 421  *              SQ_PUTLOCKS_ENTER(sq);
 422  *              SUM_SQ_PUTCOUNTS(sq, count);
 423  *      }
 424  *      SQ_PUTLOCKS_EXIT(sq);
 425  *      mutex_exit(SQLOCK(sq));
 426  *
 427  * Note that SQ_WANTWAKEUP is set before dropping SQ_PUTLOCKS. This makes sure
 428  * putnext won't skip a wakeup.
 429  *
 430  * sq_putlocks are treated as the extension of SQLOCK for lock ordering
 431  * purposes and are always grabbed right after grabbing SQLOCK and released
 432  * right before releasing SQLOCK. This also allows dynamic creation of
 433  * sq_putlocks while holding SQLOCK (by making sq_ciputctrl non null even when
 434  * the stream is already in use). Only in putnext one of sq_putlocks
 435  * is grabbed instead of SQLOCK. putnext return path remembers what counter it
 436  * incremented and decrements the right counter on its way out.
 437  */
 438 
 439 struct syncq {
 440         kmutex_t        sq_lock;        /* atomic access to syncq */
 441         uint16_t        sq_count;       /* # threads inside */
 442         uint16_t        sq_flags;       /* state and some type info */
 443         /*
 444          * Distributed syncq scheduling
 445          *  The list of queue's is handled by sq_head and
 446          *  sq_tail fields.
 447          *
 448          *  The list of events is handled by the sq_evhead and sq_evtail
 449          *  fields.
 450          */
 451         queue_t         *sq_head;       /* queue of deferred messages */
 452         queue_t         *sq_tail;       /* queue of deferred messages */
 453         mblk_t          *sq_evhead;     /* Event message on the syncq */
 454         mblk_t          *sq_evtail;
 455         uint_t          sq_nqueues;     /* # of queues on this sq */
 456         /*
 457          * Concurrency and condition variables
 458          */
 459         uint16_t        sq_type;        /* type (concurrency) of syncq */
 460         uint16_t        sq_rmqcount;    /* # threads inside removeq() */
 461         kcondvar_t      sq_wait;        /* block on this sync queue */
 462         kcondvar_t      sq_exitwait;    /* waiting for thread to leave the */
 463                                         /* inner perimeter */
 464         /*
 465          * Handling synchronous callbacks such as qtimeout and qbufcall
 466          */
 467         ushort_t        sq_callbflags;  /* flags for callback synchronization */
 468         callbparams_id_t sq_cancelid;   /* id of callback being cancelled */
 469         struct callbparams *sq_callbpend;       /* Pending callbacks */
 470 
 471         /*
 472          * Links forming an outer perimeter from one outer syncq and
 473          * a set of inner sync queues.
 474          */
 475         struct syncq    *sq_outer;      /* Pointer to outer perimeter */
 476         struct syncq    *sq_onext;      /* Linked list of syncq's making */
 477         struct syncq    *sq_oprev;      /* up the outer perimeter. */
 478         /*
 479          * support for low contention concurrent putnext.
 480          */
 481         ciputctrl_t     *sq_ciputctrl;
 482         uint_t          sq_nciputctrl;
 483         /*
 484          * Counter for the number of threads wanting to become exclusive.
 485          */
 486         uint_t          sq_needexcl;
 487         /*
 488          * These two fields are used for scheduling a syncq for
 489          * background processing. The sq_svcflag is protected by
 490          * SQLOCK lock.
 491          */
 492         struct syncq    *sq_next;       /* for syncq scheduling */
 493         void *          sq_servid;
 494         uint_t          sq_servcount;   /* # pending background threads */
 495         uint_t          sq_svcflags;    /* Scheduling flags     */
 496         clock_t         sq_tstamp;      /* Time when was enabled */
 497         /*
 498          * Maximum priority of the queues on this syncq.
 499          */
 500         pri_t           sq_pri;
 501 };
 502 typedef struct syncq syncq_t;
 503 
 504 /*
 505  * sync queue scheduling flags (for sq_svcflags).
 506  */
 507 #define SQ_SERVICE      0x1             /* being serviced */
 508 #define SQ_BGTHREAD     0x2             /* awaiting service by bg thread */
 509 #define SQ_DISABLED     0x4             /* don't put syncq in service list */
 510 
 511 /*
 512  * FASTPUT bit in sd_count/putcount.
 513  */
 514 #define SQ_FASTPUT      0x8000
 515 #define SQ_FASTMASK     0x7FFF
 516 
 517 /*
 518  * sync queue state flags
 519  */
 520 #define SQ_EXCL         0x0001          /* exclusive access to inner */
 521                                         /*      perimeter */
 522 #define SQ_BLOCKED      0x0002          /* qprocsoff */
 523 #define SQ_FROZEN       0x0004          /* freezestr */
 524 #define SQ_WRITER       0x0008          /* qwriter(OUTER) pending or running */
 525 #define SQ_MESSAGES     0x0010          /* messages on syncq */
 526 #define SQ_WANTWAKEUP   0x0020          /* do cv_broadcast on sq_wait */
 527 #define SQ_WANTEXWAKEUP 0x0040          /* do cv_broadcast on sq_exitwait */
 528 #define SQ_EVENTS       0x0080          /* Events pending */
 529 #define SQ_QUEUED       (SQ_MESSAGES | SQ_EVENTS)
 530 #define SQ_FLAGMASK     0x00FF
 531 
 532 /*
 533  * Test a queue to see if inner perimeter is exclusive.
 534  */
 535 #define PERIM_EXCL(q)   ((q)->q_syncq->sq_flags & SQ_EXCL)
 536 
 537 /*
 538  * If any of these flags are set it is not possible for a thread to
 539  * enter a put or service procedure. Instead it must either block
 540  * or put the message on the syncq.
 541  */
 542 #define SQ_GOAWAY       (SQ_EXCL|SQ_BLOCKED|SQ_FROZEN|SQ_WRITER|\
 543                         SQ_QUEUED)
 544 /*
 545  * If any of these flags are set it not possible to drain the syncq
 546  */
 547 #define SQ_STAYAWAY     (SQ_BLOCKED|SQ_FROZEN|SQ_WRITER)
 548 
 549 /*
 550  * Flags to trigger syncq tail processing.
 551  */
 552 #define SQ_TAIL         (SQ_QUEUED|SQ_WANTWAKEUP|SQ_WANTEXWAKEUP)
 553 
 554 /*
 555  * Syncq types (stored in sq_type)
 556  * The SQ_TYPES_IN_FLAGS (ciput) are also stored in sq_flags
 557  * for performance reasons. Thus these type values have to be in the low
 558  * 16 bits and not conflict with the sq_flags values above.
 559  *
 560  * Notes:
 561  *  - putnext() and put() assume that the put procedures have the highest
 562  *    degree of concurrency. Thus if any of the SQ_CI* are set then SQ_CIPUT
 563  *    has to be set. This restriction can be lifted by adding code to putnext
 564  *    and put that check that sq_count == 0 like entersq does.
 565  *  - putnext() and put() does currently not handle !SQ_COPUT
 566  *  - In order to implement !SQ_COCB outer_enter has to be fixed so that
 567  *    the callback can be cancelled while cv_waiting in outer_enter.
 568  *  - If SQ_CISVC needs to be implemented, qprocsoff() needs to wait
 569  *    for the currently running services to stop (wait for QINSERVICE
 570  *    to go off). disable_svc called from qprcosoff disables only
 571  *    services that will be run in future.
 572  *
 573  * All the SQ_CO flags are set when there is no outer perimeter.
 574  */
 575 #define SQ_CIPUT        0x0100          /* Concurrent inner put proc */
 576 #define SQ_CISVC        0x0200          /* Concurrent inner svc proc */
 577 #define SQ_CIOC         0x0400          /* Concurrent inner open/close */
 578 #define SQ_CICB         0x0800          /* Concurrent inner callback */
 579 #define SQ_COPUT        0x1000          /* Concurrent outer put proc */
 580 #define SQ_COSVC        0x2000          /* Concurrent outer svc proc */
 581 #define SQ_COOC         0x4000          /* Concurrent outer open/close */
 582 #define SQ_COCB         0x8000          /* Concurrent outer callback */
 583 
 584 /* Types also kept in sq_flags for performance */
 585 #define SQ_TYPES_IN_FLAGS       (SQ_CIPUT)
 586 
 587 #define SQ_CI           (SQ_CIPUT|SQ_CISVC|SQ_CIOC|SQ_CICB)
 588 #define SQ_CO           (SQ_COPUT|SQ_COSVC|SQ_COOC|SQ_COCB)
 589 #define SQ_TYPEMASK     (SQ_CI|SQ_CO)
 590 
 591 /*
 592  * Flag combinations passed to entersq and leavesq to specify the type
 593  * of entry point.
 594  */
 595 #define SQ_PUT          (SQ_CIPUT|SQ_COPUT)
 596 #define SQ_SVC          (SQ_CISVC|SQ_COSVC)
 597 #define SQ_OPENCLOSE    (SQ_CIOC|SQ_COOC)
 598 #define SQ_CALLBACK     (SQ_CICB|SQ_COCB)
 599 
 600 /*
 601  * Other syncq types which are not copied into flags.
 602  */
 603 #define SQ_PERMOD       0x01            /* Syncq is PERMOD */
 604 
 605 /*
 606  * Asynchronous callback qun*** flag.
 607  * The mechanism these flags are used in is one where callbacks enter
 608  * the perimeter thanks to framework support. To use this mechanism
 609  * the q* and qun* flavors of the callback routines must be used.
 610  * e.g. qtimeout and quntimeout. The synchronization provided by the flags
 611  * avoids deadlocks between blocking qun* routines and the perimeter
 612  * lock.
 613  */
 614 #define SQ_CALLB_BYPASSED       0x01            /* bypassed callback fn */
 615 
 616 /*
 617  * Cancel callback mask.
 618  * The mask expands as the number of cancelable callback types grows
 619  * Note - separate callback flag because different callbacks have
 620  * overlapping id space.
 621  */
 622 #define SQ_CALLB_CANCEL_MASK    (SQ_CANCEL_TOUT|SQ_CANCEL_BUFCALL)
 623 
 624 #define SQ_CANCEL_TOUT          0x02            /* cancel timeout request */
 625 #define SQ_CANCEL_BUFCALL       0x04            /* cancel bufcall request */
 626 
 627 typedef struct callbparams {
 628         syncq_t         *cbp_sq;
 629         void            (*cbp_func)(void *);
 630         void            *cbp_arg;
 631         callbparams_id_t cbp_id;
 632         uint_t          cbp_flags;
 633         struct callbparams *cbp_next;
 634         size_t          cbp_size;
 635 } callbparams_t;
 636 
 637 typedef struct strbufcall {
 638         void            (*bc_func)(void *);
 639         void            *bc_arg;
 640         size_t          bc_size;
 641         bufcall_id_t    bc_id;
 642         struct strbufcall *bc_next;
 643         kthread_id_t    bc_executor;
 644 } strbufcall_t;
 645 
 646 /*
 647  * Structure of list of processes to be sent SIGPOLL/SIGURG signal
 648  * on request.  The valid S_* events are defined in stropts.h.
 649  */
 650 typedef struct strsig {
 651         struct pid      *ss_pidp;       /* pid/pgrp pointer */
 652         pid_t           ss_pid;         /* positive pid, negative pgrp */
 653         int             ss_events;      /* S_* events */
 654         struct strsig   *ss_next;
 655 } strsig_t;
 656 
 657 /*
 658  * bufcall list
 659  */
 660 struct bclist {
 661         strbufcall_t    *bc_head;
 662         strbufcall_t    *bc_tail;
 663 };
 664 
 665 /*
 666  * Structure used to track mux links and unlinks.
 667  */
 668 struct mux_node {
 669         major_t          mn_imaj;       /* internal major device number */
 670         uint16_t         mn_indegree;   /* number of incoming edges */
 671         struct mux_node *mn_originp;    /* where we came from during search */
 672         struct mux_edge *mn_startp;     /* where search left off in mn_outp */
 673         struct mux_edge *mn_outp;       /* list of outgoing edges */
 674         uint_t           mn_flags;      /* see below */
 675 };
 676 
 677 /*
 678  * Flags for mux_nodes.
 679  */
 680 #define VISITED 1
 681 
 682 /*
 683  * Edge structure - a list of these is hung off the
 684  * mux_node to represent the outgoing edges.
 685  */
 686 struct mux_edge {
 687         struct mux_node *me_nodep;      /* edge leads to this node */
 688         struct mux_edge *me_nextp;      /* next edge */
 689         int              me_muxid;      /* id of link */
 690         dev_t            me_dev;        /* dev_t - used for kernel PUNLINK */
 691 };
 692 
 693 /*
 694  * Queue info
 695  *
 696  * The syncq is included here to reduce memory fragmentation
 697  * for kernel memory allocators that only allocate in sizes that are
 698  * powers of two. If the kernel memory allocator changes this should
 699  * be revisited.
 700  */
 701 typedef struct queinfo {
 702         struct queue    qu_rqueue;      /* read queue - must be first */
 703         struct queue    qu_wqueue;      /* write queue - must be second */
 704         struct syncq    qu_syncq;       /* syncq - must be third */
 705 } queinfo_t;
 706 
 707 /*
 708  * Multiplexed streams info
 709  */
 710 typedef struct linkinfo {
 711         struct linkblk  li_lblk;        /* must be first */
 712         struct file     *li_fpdown;     /* file pointer for lower stream */
 713         struct linkinfo *li_next;       /* next in list */
 714         struct linkinfo *li_prev;       /* previous in list */
 715 } linkinfo_t;
 716 
 717 /*
 718  * List of syncq's used by freeezestr/unfreezestr
 719  */
 720 typedef struct syncql {
 721         struct syncql   *sql_next;
 722         syncq_t         *sql_sq;
 723 } syncql_t;
 724 
 725 typedef struct sqlist {
 726         syncql_t        *sqlist_head;
 727         size_t          sqlist_size;            /* structure size in bytes */
 728         size_t          sqlist_index;           /* next free entry in array */
 729         syncql_t        sqlist_array[4];        /* 4 or more entries */
 730 } sqlist_t;
 731 
 732 typedef struct perdm {
 733         struct perdm            *dm_next;
 734         syncq_t                 *dm_sq;
 735         struct streamtab        *dm_str;
 736         uint_t                  dm_ref;
 737 } perdm_t;
 738 
 739 #define NEED_DM(dmp, qflag) \
 740         (dmp == NULL && (qflag & (QPERMOD | QMTOUTPERIM)))
 741 
 742 /*
 743  * fmodsw_impl_t is used within the kernel. fmodsw is used by
 744  * the modules/drivers. The information is copied from fmodsw
 745  * defined in the module/driver into the fmodsw_impl_t structure
 746  * during the module/driver initialization.
 747  */
 748 typedef struct fmodsw_impl      fmodsw_impl_t;
 749 
 750 struct fmodsw_impl {
 751         fmodsw_impl_t           *f_next;
 752         char                    f_name[FMNAMESZ + 1];
 753         struct streamtab        *f_str;
 754         uint32_t                f_qflag;
 755         uint32_t                f_sqtype;
 756         perdm_t                 *f_dmp;
 757         uint32_t                f_ref;
 758         uint32_t                f_hits;
 759 };
 760 
 761 typedef enum {
 762         FMODSW_HOLD =   0x00000001,
 763         FMODSW_LOAD =   0x00000002
 764 } fmodsw_flags_t;
 765 
 766 typedef struct cdevsw_impl {
 767         struct streamtab        *d_str;
 768         uint32_t                d_qflag;
 769         uint32_t                d_sqtype;
 770         perdm_t                 *d_dmp;
 771 } cdevsw_impl_t;
 772 
 773 /*
 774  * Enumeration of the types of access that can be requested for a
 775  * controlling terminal under job control.
 776  */
 777 enum jcaccess {
 778         JCREAD,                 /* read data on a ctty */
 779         JCWRITE,                /* write data to a ctty */
 780         JCSETP,                 /* set ctty parameters */
 781         JCGETP                  /* get ctty parameters */
 782 };
 783 
 784 struct str_stack {
 785         netstack_t      *ss_netstack;   /* Common netstack */
 786 
 787         kmutex_t        ss_sad_lock;    /* autopush lock */
 788         mod_hash_t      *ss_sad_hash;
 789         size_t          ss_sad_hash_nchains;
 790         struct saddev   *ss_saddev;     /* sad device array */
 791         int             ss_sadcnt;      /* number of sad devices */
 792 
 793         int             ss_devcnt;      /* number of mux_nodes */
 794         struct mux_node *ss_mux_nodes;  /* mux info for cycle checking */
 795 };
 796 typedef struct str_stack str_stack_t;
 797 
 798 /*
 799  * Finding related queues
 800  */
 801 #define STREAM(q)       ((q)->q_stream)
 802 #define SQ(rq)          ((syncq_t *)((rq) + 2))
 803 
 804 /*
 805  * Get the module/driver name for a queue.  Since some queues don't have
 806  * q_info structures (e.g., see log_makeq()), fall back to "?".
 807  */
 808 #define Q2NAME(q) \
 809         (((q)->q_qinfo != NULL && (q)->q_qinfo->qi_minfo->mi_idname != NULL) ? \
 810         (q)->q_qinfo->qi_minfo->mi_idname : "?")
 811 
 812 /*
 813  * Locking macros
 814  */
 815 #define QLOCK(q)        (&(q)->q_lock)
 816 #define SQLOCK(sq)      (&(sq)->sq_lock)
 817 
 818 #define STREAM_PUTLOCKS_ENTER(stp) {                                           \
 819                 ASSERT(MUTEX_HELD(&(stp)->sd_lock));                            \
 820                 if ((stp)->sd_ciputctrl != NULL) {                          \
 821                         int i;                                                 \
 822                         int nlocks = (stp)->sd_nciputctrl;                  \
 823                         ciputctrl_t *cip = (stp)->sd_ciputctrl;                     \
 824                         for (i = 0; i <= nlocks; i++) {                             \
 825                                 mutex_enter(&cip[i].ciputctrl_lock);               \
 826                         }                                                      \
 827                 }                                                              \
 828         }
 829 
 830 #define STREAM_PUTLOCKS_EXIT(stp) {                                            \
 831                 ASSERT(MUTEX_HELD(&(stp)->sd_lock));                            \
 832                 if ((stp)->sd_ciputctrl != NULL) {                          \
 833                         int i;                                                 \
 834                         int nlocks = (stp)->sd_nciputctrl;                  \
 835                         ciputctrl_t *cip = (stp)->sd_ciputctrl;                     \
 836                         for (i = 0; i <= nlocks; i++) {                             \
 837                                 mutex_exit(&cip[i].ciputctrl_lock);        \
 838                         }                                                      \
 839                 }                                                              \
 840         }
 841 
 842 #define SQ_PUTLOCKS_ENTER(sq) {                                                \
 843                 ASSERT(MUTEX_HELD(SQLOCK(sq)));                                \
 844                 if ((sq)->sq_ciputctrl != NULL) {                           \
 845                         int i;                                                 \
 846                         int nlocks = (sq)->sq_nciputctrl;                   \
 847                         ciputctrl_t *cip = (sq)->sq_ciputctrl;                      \
 848                         ASSERT((sq)->sq_type & SQ_CIPUT);                       \
 849                         for (i = 0; i <= nlocks; i++) {                             \
 850                                 mutex_enter(&cip[i].ciputctrl_lock);               \
 851                         }                                                      \
 852                 }                                                              \
 853         }
 854 
 855 #define SQ_PUTLOCKS_EXIT(sq) {                                                 \
 856                 ASSERT(MUTEX_HELD(SQLOCK(sq)));                                \
 857                 if ((sq)->sq_ciputctrl != NULL) {                           \
 858                         int i;                                                 \
 859                         int nlocks = (sq)->sq_nciputctrl;                   \
 860                         ciputctrl_t *cip = (sq)->sq_ciputctrl;                      \
 861                         ASSERT((sq)->sq_type & SQ_CIPUT);                       \
 862                         for (i = 0; i <= nlocks; i++) {                             \
 863                                 mutex_exit(&cip[i].ciputctrl_lock);        \
 864                         }                                                      \
 865                 }                                                              \
 866         }
 867 
 868 #define SQ_PUTCOUNT_SETFAST(sq) {                                       \
 869                 ASSERT(MUTEX_HELD(SQLOCK(sq)));                         \
 870                 if ((sq)->sq_ciputctrl != NULL) {                    \
 871                         int i;                                          \
 872                         int nlocks = (sq)->sq_nciputctrl;            \
 873                         ciputctrl_t *cip = (sq)->sq_ciputctrl;               \
 874                         ASSERT((sq)->sq_type & SQ_CIPUT);                \
 875                         for (i = 0; i <= nlocks; i++) {                      \
 876                                 mutex_enter(&cip[i].ciputctrl_lock);        \
 877                                 cip[i].ciputctrl_count |= SQ_FASTPUT;   \
 878                                 mutex_exit(&cip[i].ciputctrl_lock); \
 879                         }                                               \
 880                 }                                                       \
 881         }
 882 
 883 #define SQ_PUTCOUNT_CLRFAST(sq) {                                       \
 884                 ASSERT(MUTEX_HELD(SQLOCK(sq)));                         \
 885                 if ((sq)->sq_ciputctrl != NULL) {                    \
 886                         int i;                                          \
 887                         int nlocks = (sq)->sq_nciputctrl;            \
 888                         ciputctrl_t *cip = (sq)->sq_ciputctrl;               \
 889                         ASSERT((sq)->sq_type & SQ_CIPUT);                \
 890                         for (i = 0; i <= nlocks; i++) {                      \
 891                                 mutex_enter(&cip[i].ciputctrl_lock);        \
 892                                 cip[i].ciputctrl_count &= ~SQ_FASTPUT;      \
 893                                 mutex_exit(&cip[i].ciputctrl_lock); \
 894                         }                                               \
 895                 }                                                       \
 896         }
 897 
 898 
 899 #ifdef  DEBUG
 900 
 901 #define SQ_PUTLOCKS_HELD(sq) {                                                 \
 902                 ASSERT(MUTEX_HELD(SQLOCK(sq)));                                \
 903                 if ((sq)->sq_ciputctrl != NULL) {                           \
 904                         int i;                                                 \
 905                         int nlocks = (sq)->sq_nciputctrl;                   \
 906                         ciputctrl_t *cip = (sq)->sq_ciputctrl;                      \
 907                         ASSERT((sq)->sq_type & SQ_CIPUT);                       \
 908                         for (i = 0; i <= nlocks; i++) {                             \
 909                                 ASSERT(MUTEX_HELD(&cip[i].ciputctrl_lock));    \
 910                         }                                                      \
 911                 }                                                              \
 912         }
 913 
 914 #define SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) {                                \
 915                 if ((sq)->sq_ciputctrl != NULL) {                           \
 916                         int i;                                                 \
 917                         uint_t count = 0;                                      \
 918                         int ncounts = (sq)->sq_nciputctrl;                  \
 919                         ASSERT((sq)->sq_type & SQ_CIPUT);                       \
 920                         for (i = 0; i <= ncounts; i++) {                    \
 921                                 count +=                                       \
 922                                     (((sq)->sq_ciputctrl[i].ciputctrl_count) & \
 923                                     SQ_FASTMASK);                              \
 924                         }                                                      \
 925                         ASSERT(count == (countcheck));                         \
 926                 }                                                              \
 927         }
 928 
 929 #define SUMCHECK_CIPUTCTRL_COUNTS(ciput, nciput, countcheck) {                 \
 930                 int i;                                                         \
 931                 uint_t count = 0;                                              \
 932                 ASSERT((ciput) != NULL);                                       \
 933                 for (i = 0; i <= (nciput); i++) {                           \
 934                         count += (((ciput)[i].ciputctrl_count) &           \
 935                             SQ_FASTMASK);                                      \
 936                 }                                                              \
 937                 ASSERT(count == (countcheck));                                 \
 938         }
 939 
 940 #else   /* DEBUG */
 941 
 942 #define SQ_PUTLOCKS_HELD(sq)
 943 #define SUMCHECK_SQ_PUTCOUNTS(sq, countcheck)
 944 #define SUMCHECK_CIPUTCTRL_COUNTS(sq, nciput, countcheck)
 945 
 946 #endif  /* DEBUG */
 947 
 948 #define SUM_SQ_PUTCOUNTS(sq, count) {                                          \
 949                 if ((sq)->sq_ciputctrl != NULL) {                           \
 950                         int i;                                                 \
 951                         int ncounts = (sq)->sq_nciputctrl;                  \
 952                         ciputctrl_t *cip = (sq)->sq_ciputctrl;                      \
 953                         ASSERT((sq)->sq_type & SQ_CIPUT);                       \
 954                         for (i = 0; i <= ncounts; i++) {                    \
 955                                 (count) += ((cip[i].ciputctrl_count) &             \
 956                                     SQ_FASTMASK);                              \
 957                         }                                                      \
 958                 }                                                              \
 959         }
 960 
 961 #define CLAIM_QNEXT_LOCK(stp)   mutex_enter(&(stp)->sd_lock)
 962 #define RELEASE_QNEXT_LOCK(stp) mutex_exit(&(stp)->sd_lock)
 963 
 964 /*
 965  * syncq message manipulation macros.
 966  */
 967 /*
 968  * Put a message on the queue syncq.
 969  * Assumes QLOCK held.
 970  */
 971 #define SQPUT_MP(qp, mp)                                                \
 972         {                                                               \
 973                 qp->q_syncqmsgs++;                                   \
 974                 if (qp->q_sqhead == NULL) {                          \
 975                         qp->q_sqhead = qp->q_sqtail = mp;         \
 976                 } else {                                                \
 977                         qp->q_sqtail->b_next = mp;                        \
 978                         qp->q_sqtail = mp;                           \
 979                 }                                                       \
 980                 set_qfull(qp);                                          \
 981         }
 982 
 983 /*
 984  * Miscellaneous parameters and flags.
 985  */
 986 
 987 /*
 988  * Default timeout in milliseconds for ioctls and close
 989  */
 990 #define STRTIMOUT 15000
 991 
 992 /*
 993  * Flag values for stream io
 994  */
 995 #define WRITEWAIT       0x1     /* waiting for write event */
 996 #define READWAIT        0x2     /* waiting for read event */
 997 #define NOINTR          0x4     /* error is not to be set for signal */
 998 #define GETWAIT         0x8     /* waiting for getmsg event */
 999 
1000 /*
1001  * These flags need to be unique for stream io name space
1002  * and copy modes name space.  These flags allow strwaitq
1003  * and strdoioctl to proceed as if signals or errors on the stream
1004  * head have not occurred; i.e. they will be detected by some other
1005  * means.
1006  * STR_NOSIG does not allow signals to interrupt the call
1007  * STR_NOERROR does not allow stream head read, write or hup errors to
1008  * affect the call.  When used with strdoioctl(), if a previous ioctl
1009  * is pending and times out, STR_NOERROR will cause strdoioctl() to not
1010  * return ETIME. If, however, the requested ioctl times out, ETIME
1011  * will be returned (use ic_timout instead)
1012  * STR_PEEK is used to inform strwaitq that the reader is peeking at data
1013  * and that a non-persistent error should not be cleared.
1014  * STR_DELAYERR is used to inform strwaitq that it should not check errors
1015  * after being awoken since, in addition to an error, there might also be
1016  * data queued on the stream head read queue.
1017  */
1018 #define STR_NOSIG       0x10    /* Ignore signals during strdoioctl/strwaitq */
1019 #define STR_NOERROR     0x20    /* Ignore errors during strdoioctl/strwaitq */
1020 #define STR_PEEK        0x40    /* Peeking behavior on non-persistent errors */
1021 #define STR_DELAYERR    0x80    /* Do not check errors on return */
1022 
1023 /*
1024  * Copy modes for tty and I_STR ioctls
1025  */
1026 #define U_TO_K  01                      /* User to Kernel */
1027 #define K_TO_K  02                      /* Kernel to Kernel */
1028 
1029 /*
1030  * Mux defines.
1031  */
1032 #define LINKNORMAL      0x01            /* normal mux link */
1033 #define LINKPERSIST     0x02            /* persistent mux link */
1034 #define LINKTYPEMASK    0x03            /* bitmask of all link types */
1035 #define LINKCLOSE       0x04            /* unlink from strclose */
1036 
1037 /*
1038  * Definitions of Streams macros and function interfaces.
1039  */
1040 
1041 /*
1042  * Obsolete queue scheduling macros. They are not used anymore, but still kept
1043  * here for 3-d party modules and drivers who might still use them.
1044  */
1045 #define setqsched()
1046 #define qready()        1
1047 
1048 #ifdef _KERNEL
1049 #define runqueues()
1050 #define queuerun()
1051 #endif
1052 
1053 /* compatibility module for style 2 drivers with DR race condition */
1054 #define DRMODNAME       "drcompat"
1055 
1056 /*
1057  * Macros dealing with mux_nodes.
1058  */
1059 #define MUX_VISIT(X)    ((X)->mn_flags |= VISITED)
1060 #define MUX_CLEAR(X)    ((X)->mn_flags &= (~VISITED)); \
1061                         ((X)->mn_originp = NULL)
1062 #define MUX_DIDVISIT(X) ((X)->mn_flags & VISITED)
1063 
1064 
1065 /*
1066  * Twisted stream macros
1067  */
1068 #define STRMATED(X)     ((X)->sd_flag & STRMATE)
1069 #define STRLOCKMATES(X) if (&((X)->sd_lock) > &(((X)->sd_mate)->sd_lock)) { \
1070                                 mutex_enter(&((X)->sd_lock)); \
1071                                 mutex_enter(&(((X)->sd_mate)->sd_lock));  \
1072                         } else {  \
1073                                 mutex_enter(&(((X)->sd_mate)->sd_lock)); \
1074                                 mutex_enter(&((X)->sd_lock)); \
1075                         }
1076 #define STRUNLOCKMATES(X)       mutex_exit(&((X)->sd_lock)); \
1077                         mutex_exit(&(((X)->sd_mate)->sd_lock))
1078 
1079 #ifdef _KERNEL
1080 
1081 extern void strinit(void);
1082 extern int strdoioctl(struct stdata *, struct strioctl *, int, int,
1083     cred_t *, int *);
1084 extern void strsendsig(struct strsig *, int, uchar_t, int);
1085 extern void str_sendsig(vnode_t *, int, uchar_t, int);
1086 extern void strhup(struct stdata *);
1087 extern int qattach(queue_t *, dev_t *, int, cred_t *, fmodsw_impl_t *,
1088     boolean_t);
1089 extern int qreopen(queue_t *, dev_t *, int, cred_t *);
1090 extern void qdetach(queue_t *, int, int, cred_t *, boolean_t);
1091 extern void enterq(queue_t *);
1092 extern void leaveq(queue_t *);
1093 extern int putiocd(mblk_t *, caddr_t, int, cred_t *);
1094 extern int getiocd(mblk_t *, caddr_t, int);
1095 extern struct linkinfo *alloclink(queue_t *, queue_t *, struct file *);
1096 extern void lbfree(struct linkinfo *);
1097 extern int linkcycle(stdata_t *, stdata_t *, str_stack_t *);
1098 extern struct linkinfo *findlinks(stdata_t *, int, int, str_stack_t *);
1099 extern queue_t *getendq(queue_t *);
1100 extern int mlink(vnode_t *, int, int, cred_t *, int *, int);
1101 extern int mlink_file(vnode_t *, int, struct file *, cred_t *, int *, int);
1102 extern int munlink(struct stdata *, struct linkinfo *, int, cred_t *, int *,
1103     str_stack_t *);
1104 extern int munlinkall(struct stdata *, int, cred_t *, int *, str_stack_t *);
1105 extern void mux_addedge(stdata_t *, stdata_t *, int, str_stack_t *);
1106 extern void mux_rmvedge(stdata_t *, int, str_stack_t *);
1107 extern int devflg_to_qflag(struct streamtab *, uint32_t, uint32_t *,
1108     uint32_t *);
1109 extern void setq(queue_t *, struct qinit *, struct qinit *, perdm_t *,
1110     uint32_t, uint32_t, boolean_t);
1111 extern perdm_t *hold_dm(struct streamtab *, uint32_t, uint32_t);
1112 extern void rele_dm(perdm_t *);
1113 extern int strmakectl(struct strbuf *, int32_t, int32_t, mblk_t **);
1114 extern int strmakedata(ssize_t *, struct uio *, stdata_t *, int32_t, mblk_t **);
1115 extern int strmakemsg(struct strbuf *, ssize_t *, struct uio *,
1116     struct stdata *, int32_t, mblk_t **);
1117 extern int strgetmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t *,
1118     int *, int, rval_t *);
1119 extern int strputmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t,
1120     int flag, int fmode);
1121 extern int strstartplumb(struct stdata *, int, int);
1122 extern void strendplumb(struct stdata *);
1123 extern int stropen(struct vnode *, dev_t *, int, cred_t *);
1124 extern int strclose(struct vnode *, int, cred_t *);
1125 extern int strpoll(register struct stdata *, short, int, short *,
1126     struct pollhead **);
1127 extern void strclean(struct vnode *);
1128 extern void str_cn_clean();     /* XXX hook for consoles signal cleanup */
1129 extern int strwrite(struct vnode *, struct uio *, cred_t *);
1130 extern int strwrite_common(struct vnode *, struct uio *, cred_t *, int);
1131 extern int strread(struct vnode *, struct uio *, cred_t *);
1132 extern int strioctl(struct vnode *, int, intptr_t, int, int, cred_t *, int *);
1133 extern int strrput(queue_t *, mblk_t *);
1134 extern int strrput_nondata(queue_t *, mblk_t *);
1135 extern mblk_t *strrput_proto(vnode_t *, mblk_t *,
1136     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1137 extern mblk_t *strrput_misc(vnode_t *, mblk_t *,
1138     strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *);
1139 extern int getiocseqno(void);
1140 extern int strwaitbuf(size_t, int);
1141 extern int strwaitq(stdata_t *, int, ssize_t, int, clock_t, int *);
1142 extern struct stdata *shalloc(queue_t *);
1143 extern void shfree(struct stdata *s);
1144 extern queue_t *allocq(void);
1145 extern void freeq(queue_t *);
1146 extern qband_t *allocband(void);
1147 extern void freeband(qband_t *);
1148 extern void freebs_enqueue(mblk_t *, dblk_t *);
1149 extern void setqback(queue_t *, unsigned char);
1150 extern int strcopyin(void *, void *, size_t, int);
1151 extern int strcopyout(void *, void *, size_t, int);
1152 extern void strsignal(struct stdata *, int, int32_t);
1153 extern clock_t str_cv_wait(kcondvar_t *, kmutex_t *, clock_t, int);
1154 extern void disable_svc(queue_t *);
1155 extern void enable_svc(queue_t *);
1156 extern void remove_runlist(queue_t *);
1157 extern void wait_svc(queue_t *);
1158 extern void backenable(queue_t *, uchar_t);
1159 extern void set_qend(queue_t *);
1160 extern int strgeterr(stdata_t *, int32_t, int);
1161 extern void qenable_locked(queue_t *);
1162 extern mblk_t *getq_noenab(queue_t *, ssize_t);
1163 extern void rmvq_noenab(queue_t *, mblk_t *);
1164 extern void qbackenable(queue_t *, uchar_t);
1165 extern void set_qfull(queue_t *);
1166 
1167 extern void strblock(queue_t *);
1168 extern void strunblock(queue_t *);
1169 extern int qclaimed(queue_t *);
1170 extern int straccess(struct stdata *, enum jcaccess);
1171 
1172 extern void entersq(syncq_t *, int);
1173 extern void leavesq(syncq_t *, int);
1174 extern void claimq(queue_t *);
1175 extern void releaseq(queue_t *);
1176 extern void claimstr(queue_t *);
1177 extern void releasestr(queue_t *);
1178 extern void removeq(queue_t *);
1179 extern void insertq(struct stdata *, queue_t *);
1180 extern void drain_syncq(syncq_t *);
1181 extern void qfill_syncq(syncq_t *, queue_t *, mblk_t *);
1182 extern void qdrain_syncq(syncq_t *, queue_t *);
1183 extern int flush_syncq(syncq_t *, queue_t *);
1184 extern void wait_sq_svc(syncq_t *);
1185 
1186 extern void outer_enter(syncq_t *, uint16_t);
1187 extern void outer_exit(syncq_t *);
1188 extern void qwriter_inner(queue_t *, mblk_t *, void (*)());
1189 extern void qwriter_outer(queue_t *, mblk_t *, void (*)());
1190 
1191 extern callbparams_t *callbparams_alloc(syncq_t *, void (*)(void *),
1192     void *, int);
1193 extern void callbparams_free(syncq_t *, callbparams_t *);
1194 extern void callbparams_free_id(syncq_t *, callbparams_id_t, int32_t);
1195 extern void qcallbwrapper(void *);
1196 
1197 extern mblk_t *esballoc_wait(unsigned char *, size_t, uint_t, frtn_t *);
1198 extern mblk_t *esballoca(unsigned char *, size_t, uint_t, frtn_t *);
1199 extern mblk_t *desballoca(unsigned char *, size_t, uint_t, frtn_t *);
1200 extern int do_sendfp(struct stdata *, struct file *, struct cred *);
1201 extern int frozenstr(queue_t *);
1202 extern size_t xmsgsize(mblk_t *);
1203 
1204 extern void putnext_tail(syncq_t *, queue_t *, uint32_t);
1205 extern void stream_willservice(stdata_t *);
1206 extern void stream_runservice(stdata_t *);
1207 
1208 extern void strmate(vnode_t *, vnode_t *);
1209 extern queue_t *strvp2wq(vnode_t *);
1210 extern vnode_t *strq2vp(queue_t *);
1211 extern mblk_t *allocb_wait(size_t, uint_t, uint_t, int *);
1212 extern mblk_t *allocb_cred(size_t, cred_t *, pid_t);
1213 extern mblk_t *allocb_cred_wait(size_t, uint_t, int *, cred_t *, pid_t);
1214 extern mblk_t *allocb_tmpl(size_t, const mblk_t *);
1215 extern mblk_t *allocb_tryhard(size_t);
1216 extern void mblk_copycred(mblk_t *, const mblk_t *);
1217 extern void mblk_setcred(mblk_t *, cred_t *, pid_t);
1218 extern cred_t *msg_getcred(const mblk_t *, pid_t *);
1219 extern struct ts_label_s *msg_getlabel(const mblk_t *);
1220 extern cred_t *msg_extractcred(mblk_t *, pid_t *);
1221 extern void strpollwakeup(vnode_t *, short);
1222 extern int putnextctl_wait(queue_t *, int);
1223 
1224 extern int kstrputmsg(struct vnode *, mblk_t *, struct uio *, ssize_t,
1225     unsigned char, int, int);
1226 extern int kstrgetmsg(struct vnode *, mblk_t **, struct uio *,
1227     unsigned char *, int *, clock_t, rval_t *);
1228 
1229 extern void strsetrerror(vnode_t *, int, int, errfunc_t);
1230 extern void strsetwerror(vnode_t *, int, int, errfunc_t);
1231 extern void strseteof(vnode_t *, int);
1232 extern void strflushrq(vnode_t *, int);
1233 extern void strsetrputhooks(vnode_t *, uint_t, msgfunc_t, msgfunc_t);
1234 extern void strsetwputhooks(vnode_t *, uint_t, clock_t);
1235 extern void strsetrwputdatahooks(vnode_t *, msgfunc_t, msgfunc_t);
1236 extern int strwaitmark(vnode_t *);
1237 extern void strsignal_nolock(stdata_t *, int, uchar_t);
1238 
1239 struct multidata_s;
1240 struct pdesc_s;
1241 extern int hcksum_assoc(mblk_t *, struct multidata_s *, struct pdesc_s  *,
1242     uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, int);
1243 extern void hcksum_retrieve(mblk_t *, struct multidata_s *, struct pdesc_s *,
1244     uint32_t *, uint32_t *, uint32_t *, uint32_t *, uint32_t *);
1245 extern void lso_info_set(mblk_t *, uint32_t, uint32_t);
1246 extern void lso_info_cleanup(mblk_t *);
1247 extern unsigned int bcksum(uchar_t *, int, unsigned int);
1248 extern boolean_t is_vmloaned_mblk(mblk_t *, struct multidata_s *,
1249     struct pdesc_s *);
1250 
1251 extern int fmodsw_register(const char *, struct streamtab *, int);
1252 extern int fmodsw_unregister(const char *);
1253 extern fmodsw_impl_t *fmodsw_find(const char *, fmodsw_flags_t);
1254 extern void fmodsw_rele(fmodsw_impl_t *);
1255 
1256 extern void freemsgchain(mblk_t *);
1257 extern mblk_t *copymsgchain(mblk_t *);
1258 
1259 extern mblk_t *mcopyinuio(struct stdata *, uio_t *, ssize_t, ssize_t, int *);
1260 
1261 /*
1262  * shared or externally configured data structures
1263  */
1264 extern ssize_t strmsgsz;                /* maximum stream message size */
1265 extern ssize_t strctlsz;                /* maximum size of ctl message */
1266 extern int nstrpush;                    /* maximum number of pushes allowed */
1267 
1268 /*
1269  * Bufcalls related variables.
1270  */
1271 extern struct bclist strbcalls;         /* List of bufcalls */
1272 extern kmutex_t strbcall_lock;          /* Protects the list of bufcalls */
1273 extern kcondvar_t strbcall_cv;          /* Signaling when a bufcall is added */
1274 extern kcondvar_t bcall_cv;     /* wait of executing bufcall completes */
1275 
1276 extern frtn_t frnop;
1277 
1278 extern struct kmem_cache *ciputctrl_cache;
1279 extern int n_ciputctrl;
1280 extern int max_n_ciputctrl;
1281 extern int min_n_ciputctrl;
1282 
1283 extern cdevsw_impl_t *devimpl;
1284 
1285 /*
1286  * esballoc queue for throttling
1287  */
1288 typedef struct esb_queue {
1289         kmutex_t        eq_lock;
1290         uint_t          eq_len;         /* number of queued messages */
1291         mblk_t          *eq_head;       /* head of queue */
1292         mblk_t          *eq_tail;       /* tail of queue */
1293         uint_t          eq_flags;       /* esballoc queue flags */
1294 } esb_queue_t;
1295 
1296 /*
1297  * esballoc flags for queue processing.
1298  */
1299 #define ESBQ_PROCESSING 0x01    /* queue is being processed */
1300 #define ESBQ_TIMER      0x02    /* timer is active */
1301 
1302 extern void esballoc_queue_init(void);
1303 
1304 #endif  /* _KERNEL */
1305 
1306 /*
1307  * Note: Use of these macros are restricted to kernel/unix and
1308  * intended for the STREAMS framework.
1309  * All modules/drivers should include sys/ddi.h.
1310  *
1311  * Finding related queues
1312  */
1313 #define         _OTHERQ(q)      ((q)->q_flag&QREADR? (q)+1: (q)-1)
1314 #define         _WR(q)          ((q)->q_flag&QREADR? (q)+1: (q))
1315 #define         _RD(q)          ((q)->q_flag&QREADR? (q): (q)-1)
1316 #define         _SAMESTR(q)     (!((q)->q_flag & QEND))
1317 
1318 /*
1319  * These are also declared here for modules/drivers that erroneously
1320  * include strsubr.h after ddi.h or fail to include ddi.h at all.
1321  */
1322 extern struct queue *OTHERQ(queue_t *); /* stream.h */
1323 extern struct queue *RD(queue_t *);
1324 extern struct queue *WR(queue_t *);
1325 extern int SAMESTR(queue_t *);
1326 
1327 /*
1328  * The following hardware checksum related macros are private
1329  * interfaces that are subject to change without notice.
1330  */
1331 #ifdef _KERNEL
1332 #define DB_CKSUMSTART(mp)       ((mp)->b_datap->db_cksumstart)
1333 #define DB_CKSUMEND(mp)         ((mp)->b_datap->db_cksumend)
1334 #define DB_CKSUMSTUFF(mp)       ((mp)->b_datap->db_cksumstuff)
1335 #define DB_CKSUMFLAGS(mp)       ((mp)->b_datap->db_struioun.cksum.flags)
1336 #define DB_CKSUM16(mp)          ((mp)->b_datap->db_cksum16)
1337 #define DB_CKSUM32(mp)          ((mp)->b_datap->db_cksum32)
1338 #define DB_LSOFLAGS(mp)         ((mp)->b_datap->db_struioun.cksum.flags)
1339 #define DB_LSOMSS(mp)           ((mp)->b_datap->db_struioun.cksum.pad)
1340 #endif  /* _KERNEL */
1341 
1342 #ifdef  __cplusplus
1343 }
1344 #endif
1345 
1346 
1347 #endif  /* _SYS_STRSUBR_H */