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 (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
  27  * Copyright 2017 Joyent, Inc.
  28  * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
  29  */
  30 
  31 #include <sys/types.h>
  32 #include <sys/sysmacros.h>
  33 #include <sys/param.h>
  34 #include <sys/errno.h>
  35 #include <sys/signal.h>
  36 #include <sys/stat.h>
  37 #include <sys/proc.h>
  38 #include <sys/cred.h>
  39 #include <sys/user.h>
  40 #include <sys/vnode.h>
  41 #include <sys/file.h>
  42 #include <sys/stream.h>
  43 #include <sys/strsubr.h>
  44 #include <sys/stropts.h>
  45 #include <sys/tihdr.h>
  46 #include <sys/var.h>
  47 #include <sys/poll.h>
  48 #include <sys/termio.h>
  49 #include <sys/ttold.h>
  50 #include <sys/systm.h>
  51 #include <sys/uio.h>
  52 #include <sys/cmn_err.h>
  53 #include <sys/sad.h>
  54 #include <sys/netstack.h>
  55 #include <sys/priocntl.h>
  56 #include <sys/jioctl.h>
  57 #include <sys/procset.h>
  58 #include <sys/session.h>
  59 #include <sys/kmem.h>
  60 #include <sys/filio.h>
  61 #include <sys/vtrace.h>
  62 #include <sys/debug.h>
  63 #include <sys/strredir.h>
  64 #include <sys/fs/fifonode.h>
  65 #include <sys/fs/snode.h>
  66 #include <sys/strlog.h>
  67 #include <sys/strsun.h>
  68 #include <sys/project.h>
  69 #include <sys/kbio.h>
  70 #include <sys/msio.h>
  71 #include <sys/tty.h>
  72 #include <sys/ptyvar.h>
  73 #include <sys/vuid_event.h>
  74 #include <sys/modctl.h>
  75 #include <sys/sunddi.h>
  76 #include <sys/sunldi_impl.h>
  77 #include <sys/autoconf.h>
  78 #include <sys/policy.h>
  79 #include <sys/dld.h>
  80 #include <sys/zone.h>
  81 #include <c2/audit.h>
  82 
  83 /*
  84  * This define helps improve the readability of streams code while
  85  * still maintaining a very old streams performance enhancement.  The
  86  * performance enhancement basically involved having all callers
  87  * of straccess() perform the first check that straccess() will do
  88  * locally before actually calling straccess().  (There by reducing
  89  * the number of unnecessary calls to straccess().)
  90  */
  91 #define i_straccess(x, y)       ((stp->sd_sidp == NULL) ? 0 : \
  92                                     (stp->sd_vnode->v_type == VFIFO) ? 0 : \
  93                                     straccess((x), (y)))
  94 
  95 /*
  96  * what is mblk_pull_len?
  97  *
  98  * If a streams message consists of many short messages,
  99  * a performance degradation occurs from copyout overhead.
 100  * To decrease the per mblk overhead, messages that are
 101  * likely to consist of many small mblks are pulled up into
 102  * one continuous chunk of memory.
 103  *
 104  * To avoid the processing overhead of examining every
 105  * mblk, a quick heuristic is used. If the first mblk in
 106  * the message is shorter than mblk_pull_len, it is likely
 107  * that the rest of the mblk will be short.
 108  *
 109  * This heuristic was decided upon after performance tests
 110  * indicated that anything more complex slowed down the main
 111  * code path.
 112  */
 113 #define MBLK_PULL_LEN 64
 114 uint32_t mblk_pull_len = MBLK_PULL_LEN;
 115 
 116 /*
 117  * The sgttyb_handling flag controls the handling of the old BSD
 118  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
 119  *
 120  * 0 - Emit no warnings at all and retain old, broken behavior.
 121  * 1 - Emit no warnings and silently handle new semantics.
 122  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
 123  *     (once per system invocation).  Handle with new semantics.
 124  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
 125  *     made (so that offenders drop core and are easy to debug).
 126  *
 127  * The "new semantics" are that TIOCGETP returns B38400 for
 128  * sg_[io]speed if the corresponding value is over B38400, and that
 129  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
 130  * bit rate."
 131  */
 132 int sgttyb_handling = 1;
 133 static boolean_t sgttyb_complaint;
 134 
 135 /* don't push drcompat module by default on Style-2 streams */
 136 static int push_drcompat = 0;
 137 
 138 /*
 139  * id value used to distinguish between different ioctl messages
 140  */
 141 static uint32_t ioc_id;
 142 
 143 static void putback(struct stdata *, queue_t *, mblk_t *, int);
 144 static void strcleanall(struct vnode *);
 145 static int strwsrv(queue_t *);
 146 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
 147 
 148 /*
 149  * qinit and module_info structures for stream head read and write queues
 150  */
 151 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
 152 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
 153 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
 154 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
 155 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
 156     FIFOLOWAT };
 157 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
 158 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
 159 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
 160 
 161 extern kmutex_t strresources;   /* protects global resources */
 162 extern kmutex_t muxifier;       /* single-threads multiplexor creation */
 163 
 164 static boolean_t msghasdata(mblk_t *bp);
 165 #define msgnodata(bp) (!msghasdata(bp))
 166 
 167 /*
 168  * Stream head locking notes:
 169  *      There are four monitors associated with the stream head:
 170  *      1. v_stream monitor: in stropen() and strclose() v_lock
 171  *              is held while the association of vnode and stream
 172  *              head is established or tested for.
 173  *      2. open/close/push/pop monitor: sd_lock is held while each
 174  *              thread bids for exclusive access to this monitor
 175  *              for opening or closing a stream.  In addition, this
 176  *              monitor is entered during pushes and pops.  This
 177  *              guarantees that during plumbing operations there
 178  *              is only one thread trying to change the plumbing.
 179  *              Any other threads present in the stream are only
 180  *              using the plumbing.
 181  *      3. read/write monitor: in the case of read, a thread holds
 182  *              sd_lock while trying to get data from the stream
 183  *              head queue.  if there is none to fulfill a read
 184  *              request, it sets RSLEEP and calls cv_wait_sig() down
 185  *              in strwaitq() to await the arrival of new data.
 186  *              when new data arrives in strrput(), sd_lock is acquired
 187  *              before testing for RSLEEP and calling cv_broadcast().
 188  *              the behavior of strwrite(), strwsrv(), and WSLEEP
 189  *              mirror this.
 190  *      4. ioctl monitor: sd_lock is gotten to ensure that only one
 191  *              thread is doing an ioctl at a time.
 192  */
 193 
 194 static int
 195 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
 196     int anchor, cred_t *crp, uint_t anchor_zoneid)
 197 {
 198         int error;
 199         fmodsw_impl_t *fp;
 200 
 201         if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
 202                 error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
 203                 return (error);
 204         }
 205         if (stp->sd_pushcnt >= nstrpush) {
 206                 return (EINVAL);
 207         }
 208 
 209         if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
 210                 stp->sd_flag |= STREOPENFAIL;
 211                 return (EINVAL);
 212         }
 213 
 214         /*
 215          * push new module and call its open routine via qattach
 216          */
 217         if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
 218                 return (error);
 219 
 220         /*
 221          * Check to see if caller wants a STREAMS anchor
 222          * put at this place in the stream, and add if so.
 223          */
 224         mutex_enter(&stp->sd_lock);
 225         if (anchor == stp->sd_pushcnt) {
 226                 stp->sd_anchor = stp->sd_pushcnt;
 227                 stp->sd_anchorzone = anchor_zoneid;
 228         }
 229         mutex_exit(&stp->sd_lock);
 230 
 231         return (0);
 232 }
 233 
 234 /*
 235  * Open a stream device.
 236  */
 237 int
 238 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
 239 {
 240         struct stdata *stp;
 241         queue_t *qp;
 242         int s;
 243         dev_t dummydev, savedev;
 244         struct autopush *ap;
 245         struct dlautopush dlap;
 246         int error = 0;
 247         ssize_t rmin, rmax;
 248         int cloneopen;
 249         queue_t *brq;
 250         major_t major;
 251         str_stack_t *ss;
 252         zoneid_t zoneid;
 253         uint_t anchor;
 254 
 255         /*
 256          * If the stream already exists, wait for any open in progress
 257          * to complete, then call the open function of each module and
 258          * driver in the stream.  Otherwise create the stream.
 259          */
 260         TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
 261 retry:
 262         mutex_enter(&vp->v_lock);
 263         if ((stp = vp->v_stream) != NULL) {
 264 
 265                 /*
 266                  * Waiting for stream to be created to device
 267                  * due to another open.
 268                  */
 269                 mutex_exit(&vp->v_lock);
 270 
 271                 if (STRMATED(stp)) {
 272                         struct stdata *strmatep = stp->sd_mate;
 273 
 274                         STRLOCKMATES(stp);
 275                         if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
 276                                 if (flag & (FNDELAY|FNONBLOCK)) {
 277                                         error = EAGAIN;
 278                                         mutex_exit(&strmatep->sd_lock);
 279                                         goto ckreturn;
 280                                 }
 281                                 mutex_exit(&stp->sd_lock);
 282                                 if (!cv_wait_sig(&strmatep->sd_monitor,
 283                                     &strmatep->sd_lock)) {
 284                                         error = EINTR;
 285                                         mutex_exit(&strmatep->sd_lock);
 286                                         mutex_enter(&stp->sd_lock);
 287                                         goto ckreturn;
 288                                 }
 289                                 mutex_exit(&strmatep->sd_lock);
 290                                 goto retry;
 291                         }
 292                         if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
 293                                 if (flag & (FNDELAY|FNONBLOCK)) {
 294                                         error = EAGAIN;
 295                                         mutex_exit(&strmatep->sd_lock);
 296                                         goto ckreturn;
 297                                 }
 298                                 mutex_exit(&strmatep->sd_lock);
 299                                 if (!cv_wait_sig(&stp->sd_monitor,
 300                                     &stp->sd_lock)) {
 301                                         error = EINTR;
 302                                         goto ckreturn;
 303                                 }
 304                                 mutex_exit(&stp->sd_lock);
 305                                 goto retry;
 306                         }
 307 
 308                         if (stp->sd_flag & (STRDERR|STWRERR)) {
 309                                 error = EIO;
 310                                 mutex_exit(&strmatep->sd_lock);
 311                                 goto ckreturn;
 312                         }
 313 
 314                         stp->sd_flag |= STWOPEN;
 315                         STRUNLOCKMATES(stp);
 316                 } else {
 317                         mutex_enter(&stp->sd_lock);
 318                         if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
 319                                 if (flag & (FNDELAY|FNONBLOCK)) {
 320                                         error = EAGAIN;
 321                                         goto ckreturn;
 322                                 }
 323                                 if (!cv_wait_sig(&stp->sd_monitor,
 324                                     &stp->sd_lock)) {
 325                                         error = EINTR;
 326                                         goto ckreturn;
 327                                 }
 328                                 mutex_exit(&stp->sd_lock);
 329                                 goto retry;  /* could be clone! */
 330                         }
 331 
 332                         if (stp->sd_flag & (STRDERR|STWRERR)) {
 333                                 error = EIO;
 334                                 goto ckreturn;
 335                         }
 336 
 337                         stp->sd_flag |= STWOPEN;
 338                         mutex_exit(&stp->sd_lock);
 339                 }
 340 
 341                 /*
 342                  * Open all modules and devices down stream to notify
 343                  * that another user is streaming.  For modules, set the
 344                  * last argument to MODOPEN and do not pass any open flags.
 345                  * Ignore dummydev since this is not the first open.
 346                  */
 347                 claimstr(stp->sd_wrq);
 348                 qp = stp->sd_wrq;
 349                 while (_SAMESTR(qp)) {
 350                         qp = qp->q_next;
 351                         if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
 352                                 break;
 353                 }
 354                 releasestr(stp->sd_wrq);
 355                 mutex_enter(&stp->sd_lock);
 356                 stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
 357                 stp->sd_rerror = 0;
 358                 stp->sd_werror = 0;
 359 ckreturn:
 360                 cv_broadcast(&stp->sd_monitor);
 361                 mutex_exit(&stp->sd_lock);
 362                 return (error);
 363         }
 364 
 365         /*
 366          * This vnode isn't streaming.  SPECFS already
 367          * checked for multiple vnodes pointing to the
 368          * same stream, so create a stream to the driver.
 369          */
 370         qp = allocq();
 371         stp = shalloc(qp);
 372 
 373         /*
 374          * Initialize stream head.  shalloc() has given us
 375          * exclusive access, and we have the vnode locked;
 376          * we can do whatever we want with stp.
 377          */
 378         stp->sd_flag = STWOPEN;
 379         stp->sd_siglist = NULL;
 380         stp->sd_pollist.ph_list = NULL;
 381         stp->sd_sigflags = 0;
 382         stp->sd_mark = NULL;
 383         stp->sd_closetime = STRTIMOUT;
 384         stp->sd_sidp = NULL;
 385         stp->sd_pgidp = NULL;
 386         stp->sd_vnode = vp;
 387         stp->sd_rerror = 0;
 388         stp->sd_werror = 0;
 389         stp->sd_wroff = 0;
 390         stp->sd_tail = 0;
 391         stp->sd_iocblk = NULL;
 392         stp->sd_cmdblk = NULL;
 393         stp->sd_pushcnt = 0;
 394         stp->sd_qn_minpsz = 0;
 395         stp->sd_qn_maxpsz = INFPSZ - 1;      /* used to check for initialization */
 396         stp->sd_maxblk = INFPSZ;
 397         qp->q_ptr = _WR(qp)->q_ptr = stp;
 398         STREAM(qp) = STREAM(_WR(qp)) = stp;
 399         vp->v_stream = stp;
 400         mutex_exit(&vp->v_lock);
 401         if (vp->v_type == VFIFO) {
 402                 stp->sd_flag |= OLDNDELAY;
 403                 /*
 404                  * This means, both for pipes and fifos
 405                  * strwrite will send SIGPIPE if the other
 406                  * end is closed. For putmsg it depends
 407                  * on whether it is a XPG4_2 application
 408                  * or not
 409                  */
 410                 stp->sd_wput_opt = SW_SIGPIPE;
 411 
 412                 /* setq might sleep in kmem_alloc - avoid holding locks. */
 413                 setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
 414                     SQ_CI|SQ_CO, B_FALSE);
 415 
 416                 set_qend(qp);
 417                 stp->sd_strtab = fifo_getinfo();
 418                 _WR(qp)->q_nfsrv = _WR(qp);
 419                 qp->q_nfsrv = qp;
 420                 /*
 421                  * Wake up others that are waiting for stream to be created.
 422                  */
 423                 mutex_enter(&stp->sd_lock);
 424                 /*
 425                  * nothing is be pushed on stream yet, so
 426                  * optimized stream head packetsizes are just that
 427                  * of the read queue
 428                  */
 429                 stp->sd_qn_minpsz = qp->q_minpsz;
 430                 stp->sd_qn_maxpsz = qp->q_maxpsz;
 431                 stp->sd_flag &= ~STWOPEN;
 432                 goto fifo_opendone;
 433         }
 434         /* setq might sleep in kmem_alloc - avoid holding locks. */
 435         setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
 436 
 437         set_qend(qp);
 438 
 439         /*
 440          * Open driver and create stream to it (via qattach).
 441          */
 442         savedev = *devp;
 443         cloneopen = (getmajor(*devp) == clone_major);
 444         if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
 445                 mutex_enter(&vp->v_lock);
 446                 vp->v_stream = NULL;
 447                 mutex_exit(&vp->v_lock);
 448                 mutex_enter(&stp->sd_lock);
 449                 cv_broadcast(&stp->sd_monitor);
 450                 mutex_exit(&stp->sd_lock);
 451                 freeq(_RD(qp));
 452                 shfree(stp);
 453                 return (error);
 454         }
 455         /*
 456          * Set sd_strtab after open in order to handle clonable drivers
 457          */
 458         stp->sd_strtab = STREAMSTAB(getmajor(*devp));
 459 
 460         /*
 461          * Historical note: dummydev used to be be prior to the initial
 462          * open (via qattach above), which made the value seen
 463          * inconsistent between an I_PUSH and an autopush of a module.
 464          */
 465         dummydev = *devp;
 466 
 467         /*
 468          * For clone open of old style (Q not associated) network driver,
 469          * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
 470          */
 471         brq = _RD(_WR(qp)->q_next);
 472         major = getmajor(*devp);
 473         if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
 474             ((brq->q_flag & _QASSOCIATED) == 0)) {
 475                 if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
 476                         cmn_err(CE_WARN, "cannot push " DRMODNAME
 477                             " streams module");
 478         }
 479 
 480         if (!NETWORK_DRV(major)) {
 481                 savedev = *devp;
 482         } else {
 483                 /*
 484                  * For network devices, process differently based on the
 485                  * return value from dld_autopush():
 486                  *
 487                  *   0: the passed-in device points to a GLDv3 datalink with
 488                  *   per-link autopush configuration; use that configuration
 489                  *   and ignore any per-driver autopush configuration.
 490                  *
 491                  *   1: the passed-in device points to a physical GLDv3
 492                  *   datalink without per-link autopush configuration.  The
 493                  *   passed in device was changed to refer to the actual
 494                  *   physical device (if it's not already); we use that new
 495                  *   device to look up any per-driver autopush configuration.
 496                  *
 497                  *   -1: neither of the above cases applied; use the initial
 498                  *   device to look up any per-driver autopush configuration.
 499                  */
 500                 switch (dld_autopush(&savedev, &dlap)) {
 501                 case 0:
 502                         zoneid = crgetzoneid(crp);
 503                         for (s = 0; s < dlap.dap_npush; s++) {
 504                                 error = push_mod(qp, &dummydev, stp,
 505                                     dlap.dap_aplist[s], dlap.dap_anchor, crp,
 506                                     zoneid);
 507                                 if (error != 0)
 508                                         break;
 509                         }
 510                         goto opendone;
 511                 case 1:
 512                         break;
 513                 case -1:
 514                         savedev = *devp;
 515                         break;
 516                 }
 517         }
 518         /*
 519          * Find the autopush configuration based on "savedev". Start with the
 520          * global zone. If not found check in the local zone.
 521          */
 522         zoneid = GLOBAL_ZONEID;
 523 retryap:
 524         ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
 525             netstack_str;
 526         if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
 527                 netstack_rele(ss->ss_netstack);
 528                 if (zoneid == GLOBAL_ZONEID) {
 529                         /*
 530                          * None found. Also look in the zone's autopush table.
 531                          */
 532                         zoneid = crgetzoneid(crp);
 533                         if (zoneid != GLOBAL_ZONEID)
 534                                 goto retryap;
 535                 }
 536                 goto opendone;
 537         }
 538         anchor = ap->ap_anchor;
 539         zoneid = crgetzoneid(crp);
 540         for (s = 0; s < ap->ap_npush; s++) {
 541                 error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
 542                     anchor, crp, zoneid);
 543                 if (error != 0)
 544                         break;
 545         }
 546         sad_ap_rele(ap, ss);
 547         netstack_rele(ss->ss_netstack);
 548 
 549 opendone:
 550 
 551         /*
 552          * let specfs know that open failed part way through
 553          */
 554         if (error) {
 555                 mutex_enter(&stp->sd_lock);
 556                 stp->sd_flag |= STREOPENFAIL;
 557                 mutex_exit(&stp->sd_lock);
 558         }
 559 
 560         /*
 561          * Wake up others that are waiting for stream to be created.
 562          */
 563         mutex_enter(&stp->sd_lock);
 564         stp->sd_flag &= ~STWOPEN;
 565 
 566         /*
 567          * As a performance concern we are caching the values of
 568          * q_minpsz and q_maxpsz of the module below the stream
 569          * head in the stream head.
 570          */
 571         mutex_enter(QLOCK(stp->sd_wrq->q_next));
 572         rmin = stp->sd_wrq->q_next->q_minpsz;
 573         rmax = stp->sd_wrq->q_next->q_maxpsz;
 574         mutex_exit(QLOCK(stp->sd_wrq->q_next));
 575 
 576         /* do this processing here as a performance concern */
 577         if (strmsgsz != 0) {
 578                 if (rmax == INFPSZ)
 579                         rmax = strmsgsz;
 580                 else
 581                         rmax = MIN(strmsgsz, rmax);
 582         }
 583 
 584         mutex_enter(QLOCK(stp->sd_wrq));
 585         stp->sd_qn_minpsz = rmin;
 586         stp->sd_qn_maxpsz = rmax;
 587         mutex_exit(QLOCK(stp->sd_wrq));
 588 
 589 fifo_opendone:
 590         cv_broadcast(&stp->sd_monitor);
 591         mutex_exit(&stp->sd_lock);
 592         return (error);
 593 }
 594 
 595 static int strsink(queue_t *, mblk_t *);
 596 static struct qinit deadrend = {
 597         strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
 598 };
 599 static struct qinit deadwend = {
 600         NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
 601 };
 602 
 603 /*
 604  * Close a stream.
 605  * This is called from closef() on the last close of an open stream.
 606  * Strclean() will already have removed the siglist and pollist
 607  * information, so all that remains is to remove all multiplexor links
 608  * for the stream, pop all the modules (and the driver), and free the
 609  * stream structure.
 610  */
 611 
 612 int
 613 strclose(struct vnode *vp, int flag, cred_t *crp)
 614 {
 615         struct stdata *stp;
 616         queue_t *qp;
 617         int rval;
 618         int freestp = 1;
 619         queue_t *rmq;
 620 
 621         TRACE_1(TR_FAC_STREAMS_FR,
 622             TR_STRCLOSE, "strclose:%p", vp);
 623         ASSERT(vp->v_stream);
 624 
 625         stp = vp->v_stream;
 626         ASSERT(!(stp->sd_flag & STPLEX));
 627         qp = stp->sd_wrq;
 628 
 629         /*
 630          * Needed so that strpoll will return non-zero for this fd.
 631          * Note that with POLLNOERR STRHUP does still cause POLLHUP.
 632          */
 633         mutex_enter(&stp->sd_lock);
 634         stp->sd_flag |= STRHUP;
 635         mutex_exit(&stp->sd_lock);
 636 
 637         /*
 638          * If the registered process or process group did not have an
 639          * open instance of this stream then strclean would not be
 640          * called. Thus at the time of closing all remaining siglist entries
 641          * are removed.
 642          */
 643         if (stp->sd_siglist != NULL)
 644                 strcleanall(vp);
 645 
 646         ASSERT(stp->sd_siglist == NULL);
 647         ASSERT(stp->sd_sigflags == 0);
 648 
 649         if (STRMATED(stp)) {
 650                 struct stdata *strmatep = stp->sd_mate;
 651                 int waited = 1;
 652 
 653                 STRLOCKMATES(stp);
 654                 while (waited) {
 655                         waited = 0;
 656                         while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
 657                                 mutex_exit(&strmatep->sd_lock);
 658                                 cv_wait(&stp->sd_monitor, &stp->sd_lock);
 659                                 mutex_exit(&stp->sd_lock);
 660                                 STRLOCKMATES(stp);
 661                                 waited = 1;
 662                         }
 663                         while (strmatep->sd_flag &
 664                             (STWOPEN|STRCLOSE|STRPLUMB)) {
 665                                 mutex_exit(&stp->sd_lock);
 666                                 cv_wait(&strmatep->sd_monitor,
 667                                     &strmatep->sd_lock);
 668                                 mutex_exit(&strmatep->sd_lock);
 669                                 STRLOCKMATES(stp);
 670                                 waited = 1;
 671                         }
 672                 }
 673                 stp->sd_flag |= STRCLOSE;
 674                 STRUNLOCKMATES(stp);
 675         } else {
 676                 mutex_enter(&stp->sd_lock);
 677                 stp->sd_flag |= STRCLOSE;
 678                 mutex_exit(&stp->sd_lock);
 679         }
 680 
 681         ASSERT(qp->q_first == NULL); /* No more delayed write */
 682 
 683         /* Check if an I_LINK was ever done on this stream */
 684         if (stp->sd_flag & STRHASLINKS) {
 685                 netstack_t *ns;
 686                 str_stack_t *ss;
 687 
 688                 ns = netstack_find_by_cred(crp);
 689                 ASSERT(ns != NULL);
 690                 ss = ns->netstack_str;
 691                 ASSERT(ss != NULL);
 692 
 693                 (void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
 694                 netstack_rele(ss->ss_netstack);
 695         }
 696 
 697         while (_SAMESTR(qp)) {
 698                 /*
 699                  * Holding sd_lock prevents q_next from changing in
 700                  * this stream.
 701                  */
 702                 mutex_enter(&stp->sd_lock);
 703                 if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
 704 
 705                         /*
 706                          * sleep until awakened by strwsrv() or timeout
 707                          */
 708                         for (;;) {
 709                                 mutex_enter(QLOCK(qp->q_next));
 710                                 if (!(qp->q_next->q_mblkcnt)) {
 711                                         mutex_exit(QLOCK(qp->q_next));
 712                                         break;
 713                                 }
 714                                 stp->sd_flag |= WSLEEP;
 715 
 716                                 /* ensure strwsrv gets enabled */
 717                                 qp->q_next->q_flag |= QWANTW;
 718                                 mutex_exit(QLOCK(qp->q_next));
 719                                 /* get out if we timed out or recv'd a signal */
 720                                 if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
 721                                     stp->sd_closetime, 0) <= 0) {
 722                                         break;
 723                                 }
 724                         }
 725                         stp->sd_flag &= ~WSLEEP;
 726                 }
 727                 mutex_exit(&stp->sd_lock);
 728 
 729                 rmq = qp->q_next;
 730                 if (rmq->q_flag & QISDRV) {
 731                         ASSERT(!_SAMESTR(rmq));
 732                         wait_sq_svc(_RD(qp)->q_syncq);
 733                 }
 734 
 735                 qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
 736         }
 737 
 738         /*
 739          * Since we call pollwakeup in close() now, the poll list should
 740          * be empty in most cases. The only exception is the layered devices
 741          * (e.g. the console drivers with redirection modules pushed on top
 742          * of it).  We have to do this after calling qdetach() because
 743          * the redirection module won't have torn down the console
 744          * redirection until after qdetach() has been invoked.
 745          */
 746         if (stp->sd_pollist.ph_list != NULL) {
 747                 pollwakeup(&stp->sd_pollist, POLLERR);
 748                 pollhead_clean(&stp->sd_pollist);
 749         }
 750         ASSERT(stp->sd_pollist.ph_list == NULL);
 751         ASSERT(stp->sd_sidp == NULL);
 752         ASSERT(stp->sd_pgidp == NULL);
 753 
 754         /* Prevent qenable from re-enabling the stream head queue */
 755         disable_svc(_RD(qp));
 756 
 757         /*
 758          * Wait until service procedure of each queue is
 759          * run, if QINSERVICE is set.
 760          */
 761         wait_svc(_RD(qp));
 762 
 763         /*
 764          * Now, flush both queues.
 765          */
 766         flushq(_RD(qp), FLUSHALL);
 767         flushq(qp, FLUSHALL);
 768 
 769         /*
 770          * If the write queue of the stream head is pointing to a
 771          * read queue, we have a twisted stream.  If the read queue
 772          * is alive, convert the stream head queues into a dead end.
 773          * If the read queue is dead, free the dead pair.
 774          */
 775         if (qp->q_next && !_SAMESTR(qp)) {
 776                 if (qp->q_next->q_qinfo == &deadrend) {       /* half-closed pipe */
 777                         flushq(qp->q_next, FLUSHALL); /* ensure no message */
 778                         shfree(qp->q_next->q_stream);
 779                         freeq(qp->q_next);
 780                         freeq(_RD(qp));
 781                 } else if (qp->q_next == _RD(qp)) {  /* fifo */
 782                         freeq(_RD(qp));
 783                 } else {                                /* pipe */
 784                         freestp = 0;
 785                         /*
 786                          * The q_info pointers are never accessed when
 787                          * SQLOCK is held.
 788                          */
 789                         ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
 790                         mutex_enter(SQLOCK(qp->q_syncq));
 791                         qp->q_qinfo = &deadwend;
 792                         _RD(qp)->q_qinfo = &deadrend;
 793                         mutex_exit(SQLOCK(qp->q_syncq));
 794                 }
 795         } else {
 796                 freeq(_RD(qp)); /* free stream head queue pair */
 797         }
 798 
 799         mutex_enter(&vp->v_lock);
 800         if (stp->sd_iocblk) {
 801                 if (stp->sd_iocblk != (mblk_t *)-1) {
 802                         freemsg(stp->sd_iocblk);
 803                 }
 804                 stp->sd_iocblk = NULL;
 805         }
 806         stp->sd_vnode = NULL;
 807         vp->v_stream = NULL;
 808         mutex_exit(&vp->v_lock);
 809         mutex_enter(&stp->sd_lock);
 810         freemsg(stp->sd_cmdblk);
 811         stp->sd_cmdblk = NULL;
 812         stp->sd_flag &= ~STRCLOSE;
 813         cv_broadcast(&stp->sd_monitor);
 814         mutex_exit(&stp->sd_lock);
 815 
 816         if (freestp)
 817                 shfree(stp);
 818         return (0);
 819 }
 820 
 821 static int
 822 strsink(queue_t *q, mblk_t *bp)
 823 {
 824         struct copyresp *resp;
 825 
 826         switch (bp->b_datap->db_type) {
 827         case M_FLUSH:
 828                 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
 829                         *bp->b_rptr &= ~FLUSHR;
 830                         bp->b_flag |= MSGNOLOOP;
 831                         /*
 832                          * Protect against the driver passing up
 833                          * messages after it has done a qprocsoff.
 834                          */
 835                         if (_OTHERQ(q)->q_next == NULL)
 836                                 freemsg(bp);
 837                         else
 838                                 qreply(q, bp);
 839                 } else {
 840                         freemsg(bp);
 841                 }
 842                 break;
 843 
 844         case M_COPYIN:
 845         case M_COPYOUT:
 846                 if (bp->b_cont) {
 847                         freemsg(bp->b_cont);
 848                         bp->b_cont = NULL;
 849                 }
 850                 bp->b_datap->db_type = M_IOCDATA;
 851                 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
 852                 resp = (struct copyresp *)bp->b_rptr;
 853                 resp->cp_rval = (caddr_t)1;  /* failure */
 854                 /*
 855                  * Protect against the driver passing up
 856                  * messages after it has done a qprocsoff.
 857                  */
 858                 if (_OTHERQ(q)->q_next == NULL)
 859                         freemsg(bp);
 860                 else
 861                         qreply(q, bp);
 862                 break;
 863 
 864         case M_IOCTL:
 865                 if (bp->b_cont) {
 866                         freemsg(bp->b_cont);
 867                         bp->b_cont = NULL;
 868                 }
 869                 bp->b_datap->db_type = M_IOCNAK;
 870                 /*
 871                  * Protect against the driver passing up
 872                  * messages after it has done a qprocsoff.
 873                  */
 874                 if (_OTHERQ(q)->q_next == NULL)
 875                         freemsg(bp);
 876                 else
 877                         qreply(q, bp);
 878                 break;
 879 
 880         default:
 881                 freemsg(bp);
 882                 break;
 883         }
 884 
 885         return (0);
 886 }
 887 
 888 /*
 889  * Clean up after a process when it closes a stream.  This is called
 890  * from closef for all closes, whereas strclose is called only for the
 891  * last close on a stream.  The siglist is scanned for entries for the
 892  * current process, and these are removed.
 893  */
 894 void
 895 strclean(struct vnode *vp)
 896 {
 897         strsig_t *ssp, *pssp, *tssp;
 898         stdata_t *stp;
 899         int update = 0;
 900 
 901         TRACE_1(TR_FAC_STREAMS_FR,
 902             TR_STRCLEAN, "strclean:%p", vp);
 903         stp = vp->v_stream;
 904         pssp = NULL;
 905         mutex_enter(&stp->sd_lock);
 906         ssp = stp->sd_siglist;
 907         while (ssp) {
 908                 if (ssp->ss_pidp == curproc->p_pidp) {
 909                         tssp = ssp->ss_next;
 910                         if (pssp)
 911                                 pssp->ss_next = tssp;
 912                         else
 913                                 stp->sd_siglist = tssp;
 914                         mutex_enter(&pidlock);
 915                         PID_RELE(ssp->ss_pidp);
 916                         mutex_exit(&pidlock);
 917                         kmem_free(ssp, sizeof (strsig_t));
 918                         update = 1;
 919                         ssp = tssp;
 920                 } else {
 921                         pssp = ssp;
 922                         ssp = ssp->ss_next;
 923                 }
 924         }
 925         if (update) {
 926                 stp->sd_sigflags = 0;
 927                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
 928                         stp->sd_sigflags |= ssp->ss_events;
 929         }
 930         mutex_exit(&stp->sd_lock);
 931 }
 932 
 933 /*
 934  * Used on the last close to remove any remaining items on the siglist.
 935  * These could be present on the siglist due to I_ESETSIG calls that
 936  * use process groups or processed that do not have an open file descriptor
 937  * for this stream (Such entries would not be removed by strclean).
 938  */
 939 static void
 940 strcleanall(struct vnode *vp)
 941 {
 942         strsig_t *ssp, *nssp;
 943         stdata_t *stp;
 944 
 945         stp = vp->v_stream;
 946         mutex_enter(&stp->sd_lock);
 947         ssp = stp->sd_siglist;
 948         stp->sd_siglist = NULL;
 949         while (ssp) {
 950                 nssp = ssp->ss_next;
 951                 mutex_enter(&pidlock);
 952                 PID_RELE(ssp->ss_pidp);
 953                 mutex_exit(&pidlock);
 954                 kmem_free(ssp, sizeof (strsig_t));
 955                 ssp = nssp;
 956         }
 957         stp->sd_sigflags = 0;
 958         mutex_exit(&stp->sd_lock);
 959 }
 960 
 961 /*
 962  * Retrieve the next message from the logical stream head read queue
 963  * using either rwnext (if sync stream) or getq_noenab.
 964  * It is the callers responsibility to call qbackenable after
 965  * it is finished with the message. The caller should not call
 966  * qbackenable until after any putback calls to avoid spurious backenabling.
 967  */
 968 mblk_t *
 969 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
 970     int *errorp)
 971 {
 972         mblk_t *bp;
 973         int error;
 974         ssize_t rbytes = 0;
 975 
 976         /* Holding sd_lock prevents the read queue from changing  */
 977         ASSERT(MUTEX_HELD(&stp->sd_lock));
 978 
 979         if (uiop != NULL && stp->sd_struiordq != NULL &&
 980             q->q_first == NULL &&
 981             (!first || (stp->sd_wakeq & RSLEEP))) {
 982                 /*
 983                  * Stream supports rwnext() for the read side.
 984                  * If this is the first time we're called by e.g. strread
 985                  * only do the downcall if there is a deferred wakeup
 986                  * (registered in sd_wakeq).
 987                  */
 988                 struiod_t uiod;
 989 
 990                 if (first)
 991                         stp->sd_wakeq &= ~RSLEEP;
 992 
 993                 (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
 994                     sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
 995                 uiod.d_mp = 0;
 996                 /*
 997                  * Mark that a thread is in rwnext on the read side
 998                  * to prevent strrput from nacking ioctls immediately.
 999                  * When the last concurrent rwnext returns
1000                  * the ioctls are nack'ed.
1001                  */
1002                 ASSERT(MUTEX_HELD(&stp->sd_lock));
1003                 stp->sd_struiodnak++;
1004                 /*
1005                  * Note: rwnext will drop sd_lock.
1006                  */
1007                 error = rwnext(q, &uiod);
1008                 ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1009                 mutex_enter(&stp->sd_lock);
1010                 stp->sd_struiodnak--;
1011                 while (stp->sd_struiodnak == 0 &&
1012                     ((bp = stp->sd_struionak) != NULL)) {
1013                         stp->sd_struionak = bp->b_next;
1014                         bp->b_next = NULL;
1015                         bp->b_datap->db_type = M_IOCNAK;
1016                         /*
1017                          * Protect against the driver passing up
1018                          * messages after it has done a qprocsoff.
1019                          */
1020                         if (_OTHERQ(q)->q_next == NULL)
1021                                 freemsg(bp);
1022                         else {
1023                                 mutex_exit(&stp->sd_lock);
1024                                 qreply(q, bp);
1025                                 mutex_enter(&stp->sd_lock);
1026                         }
1027                 }
1028                 ASSERT(MUTEX_HELD(&stp->sd_lock));
1029                 if (error == 0 || error == EWOULDBLOCK) {
1030                         if ((bp = uiod.d_mp) != NULL) {
1031                                 *errorp = 0;
1032                                 ASSERT(MUTEX_HELD(&stp->sd_lock));
1033                                 return (bp);
1034                         }
1035                         error = 0;
1036                 } else if (error == EINVAL) {
1037                         /*
1038                          * The stream plumbing must have
1039                          * changed while we were away, so
1040                          * just turn off rwnext()s.
1041                          */
1042                         error = 0;
1043                 } else if (error == EBUSY) {
1044                         /*
1045                          * The module might have data in transit using putnext
1046                          * Fall back on waiting + getq.
1047                          */
1048                         error = 0;
1049                 } else {
1050                         *errorp = error;
1051                         ASSERT(MUTEX_HELD(&stp->sd_lock));
1052                         return (NULL);
1053                 }
1054                 /*
1055                  * Try a getq in case a rwnext() generated mblk
1056                  * has bubbled up via strrput().
1057                  */
1058         }
1059         *errorp = 0;
1060         ASSERT(MUTEX_HELD(&stp->sd_lock));
1061 
1062         /*
1063          * If we have a valid uio, try and use this as a guide for how
1064          * many bytes to retrieve from the queue via getq_noenab().
1065          * Doing this can avoid unneccesary counting of overlong
1066          * messages in putback(). We currently only do this for sockets
1067          * and only if there is no sd_rputdatafunc hook.
1068          *
1069          * The sd_rputdatafunc hook transforms the entire message
1070          * before any bytes in it can be given to a client. So, rbytes
1071          * must be 0 if there is a hook.
1072          */
1073         if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK) &&
1074             (stp->sd_rputdatafunc == NULL))
1075                 rbytes = uiop->uio_resid;
1076 
1077         return (getq_noenab(q, rbytes));
1078 }
1079 
1080 /*
1081  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1082  * If the message does not fit in the uio the remainder of it is returned;
1083  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1084  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1085  * the error code, the message is consumed, and NULL is returned.
1086  */
1087 static mblk_t *
1088 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1089 {
1090         int error;
1091         ptrdiff_t n;
1092         mblk_t *nbp;
1093 
1094         ASSERT(bp->b_wptr >= bp->b_rptr);
1095 
1096         do {
1097                 if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1098                         ASSERT(n > 0);
1099 
1100                         error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1101                         if (error != 0) {
1102                                 freemsg(bp);
1103                                 *errorp = error;
1104                                 return (NULL);
1105                         }
1106                 }
1107 
1108                 bp->b_rptr += n;
1109                 while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1110                         nbp = bp;
1111                         bp = bp->b_cont;
1112                         freeb(nbp);
1113                 }
1114         } while (bp != NULL && uiop->uio_resid > 0);
1115 
1116         *errorp = 0;
1117         return (bp);
1118 }
1119 
1120 /*
1121  * Read a stream according to the mode flags in sd_flag:
1122  *
1123  * (default mode)               - Byte stream, msg boundaries are ignored
1124  * RD_MSGDIS (msg discard)      - Read on msg boundaries and throw away
1125  *                              any data remaining in msg
1126  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1127  *                              any remaining data on head of read queue
1128  *
1129  * Consume readable messages on the front of the queue until
1130  * ttolwp(curthread)->lwp_count
1131  * is satisfied, the readable messages are exhausted, or a message
1132  * boundary is reached in a message mode.  If no data was read and
1133  * the stream was not opened with the NDELAY flag, block until data arrives.
1134  * Otherwise return the data read and update the count.
1135  *
1136  * In default mode a 0 length message signifies end-of-file and terminates
1137  * a read in progress.  The 0 length message is removed from the queue
1138  * only if it is the only message read (no data is read).
1139  *
1140  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1141  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1142  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1143  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1144  * are unlinked from and M_DATA blocks in the message, the protos are
1145  * thrown away, and the data is read.
1146  */
1147 /* ARGSUSED */
1148 int
1149 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1150 {
1151         struct stdata *stp;
1152         mblk_t *bp, *nbp;
1153         queue_t *q;
1154         int error = 0;
1155         uint_t old_sd_flag;
1156         int first;
1157         char rflg;
1158         uint_t mark;            /* Contains MSG*MARK and _LASTMARK */
1159 #define _LASTMARK       0x8000  /* Distinct from MSG*MARK */
1160         short delim;
1161         unsigned char pri = 0;
1162         char waitflag;
1163         unsigned char type;
1164 
1165         TRACE_1(TR_FAC_STREAMS_FR,
1166             TR_STRREAD_ENTER, "strread:%p", vp);
1167         ASSERT(vp->v_stream);
1168         stp = vp->v_stream;
1169 
1170         mutex_enter(&stp->sd_lock);
1171 
1172         if ((error = i_straccess(stp, JCREAD)) != 0) {
1173                 mutex_exit(&stp->sd_lock);
1174                 return (error);
1175         }
1176 
1177         if (stp->sd_flag & (STRDERR|STPLEX)) {
1178                 error = strgeterr(stp, STRDERR|STPLEX, 0);
1179                 if (error != 0) {
1180                         mutex_exit(&stp->sd_lock);
1181                         return (error);
1182                 }
1183         }
1184 
1185         /*
1186          * Loop terminates when uiop->uio_resid == 0.
1187          */
1188         rflg = 0;
1189         waitflag = READWAIT;
1190         q = _RD(stp->sd_wrq);
1191         for (;;) {
1192                 ASSERT(MUTEX_HELD(&stp->sd_lock));
1193                 old_sd_flag = stp->sd_flag;
1194                 mark = 0;
1195                 delim = 0;
1196                 first = 1;
1197                 while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1198                         int done = 0;
1199 
1200                         ASSERT(MUTEX_HELD(&stp->sd_lock));
1201 
1202                         if (error != 0)
1203                                 goto oops;
1204 
1205                         if (stp->sd_flag & (STRHUP|STREOF)) {
1206                                 goto oops;
1207                         }
1208                         if (rflg && !(stp->sd_flag & STRDELIM)) {
1209                                 goto oops;
1210                         }
1211                         /*
1212                          * If a read(fd,buf,0) has been done, there is no
1213                          * need to sleep. We always have zero bytes to
1214                          * return.
1215                          */
1216                         if (uiop->uio_resid == 0) {
1217                                 goto oops;
1218                         }
1219 
1220                         qbackenable(q, 0);
1221 
1222                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1223                             "strread calls strwaitq:%p, %p, %p",
1224                             vp, uiop, crp);
1225                         if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1226                             uiop->uio_fmode, -1, &done)) != 0 || done) {
1227                                 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1228                                     "strread error or done:%p, %p, %p",
1229                                     vp, uiop, crp);
1230                                 if ((uiop->uio_fmode & FNDELAY) &&
1231                                     (stp->sd_flag & OLDNDELAY) &&
1232                                     (error == EAGAIN))
1233                                         error = 0;
1234                                 goto oops;
1235                         }
1236                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1237                             "strread awakes:%p, %p, %p", vp, uiop, crp);
1238                         if ((error = i_straccess(stp, JCREAD)) != 0) {
1239                                 goto oops;
1240                         }
1241                         first = 0;
1242                 }
1243 
1244                 ASSERT(MUTEX_HELD(&stp->sd_lock));
1245                 ASSERT(bp);
1246                 pri = bp->b_band;
1247                 /*
1248                  * Extract any mark information. If the message is not
1249                  * completely consumed this information will be put in the mblk
1250                  * that is putback.
1251                  * If MSGMARKNEXT is set and the message is completely consumed
1252                  * the STRATMARK flag will be set below. Likewise, if
1253                  * MSGNOTMARKNEXT is set and the message is
1254                  * completely consumed STRNOTATMARK will be set.
1255                  *
1256                  * For some unknown reason strread only breaks the read at the
1257                  * last mark.
1258                  */
1259                 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1260                 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1261                     (MSGMARKNEXT|MSGNOTMARKNEXT));
1262                 if (mark != 0 && bp == stp->sd_mark) {
1263                         if (rflg) {
1264                                 putback(stp, q, bp, pri);
1265                                 goto oops;
1266                         }
1267                         mark |= _LASTMARK;
1268                         stp->sd_mark = NULL;
1269                 }
1270                 if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1271                         delim = 1;
1272                 mutex_exit(&stp->sd_lock);
1273 
1274                 if (STREAM_NEEDSERVICE(stp))
1275                         stream_runservice(stp);
1276 
1277                 type = bp->b_datap->db_type;
1278 
1279                 switch (type) {
1280 
1281                 case M_DATA:
1282 ismdata:
1283                         if (msgnodata(bp)) {
1284                                 if (mark || delim) {
1285                                         freemsg(bp);
1286                                 } else if (rflg) {
1287 
1288                                         /*
1289                                          * If already read data put zero
1290                                          * length message back on queue else
1291                                          * free msg and return 0.
1292                                          */
1293                                         bp->b_band = pri;
1294                                         mutex_enter(&stp->sd_lock);
1295                                         putback(stp, q, bp, pri);
1296                                         mutex_exit(&stp->sd_lock);
1297                                 } else {
1298                                         freemsg(bp);
1299                                 }
1300                                 error =  0;
1301                                 goto oops1;
1302                         }
1303 
1304                         rflg = 1;
1305                         waitflag |= NOINTR;
1306                         bp = struiocopyout(bp, uiop, &error);
1307                         if (error != 0)
1308                                 goto oops1;
1309 
1310                         mutex_enter(&stp->sd_lock);
1311                         if (bp) {
1312                                 /*
1313                                  * Have remaining data in message.
1314                                  * Free msg if in discard mode.
1315                                  */
1316                                 if (stp->sd_read_opt & RD_MSGDIS) {
1317                                         freemsg(bp);
1318                                 } else {
1319                                         bp->b_band = pri;
1320                                         if ((mark & _LASTMARK) &&
1321                                             (stp->sd_mark == NULL))
1322                                                 stp->sd_mark = bp;
1323                                         bp->b_flag |= mark & ~_LASTMARK;
1324                                         if (delim)
1325                                                 bp->b_flag |= MSGDELIM;
1326                                         if (msgnodata(bp))
1327                                                 freemsg(bp);
1328                                         else
1329                                                 putback(stp, q, bp, pri);
1330                                 }
1331                         } else {
1332                                 /*
1333                                  * Consumed the complete message.
1334                                  * Move the MSG*MARKNEXT information
1335                                  * to the stream head just in case
1336                                  * the read queue becomes empty.
1337                                  *
1338                                  * If the stream head was at the mark
1339                                  * (STRATMARK) before we dropped sd_lock above
1340                                  * and some data was consumed then we have
1341                                  * moved past the mark thus STRATMARK is
1342                                  * cleared. However, if a message arrived in
1343                                  * strrput during the copyout above causing
1344                                  * STRATMARK to be set we can not clear that
1345                                  * flag.
1346                                  */
1347                                 if (mark &
1348                                     (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1349                                         if (mark & MSGMARKNEXT) {
1350                                                 stp->sd_flag &= ~STRNOTATMARK;
1351                                                 stp->sd_flag |= STRATMARK;
1352                                         } else if (mark & MSGNOTMARKNEXT) {
1353                                                 stp->sd_flag &= ~STRATMARK;
1354                                                 stp->sd_flag |= STRNOTATMARK;
1355                                         } else {
1356                                                 stp->sd_flag &=
1357                                                     ~(STRATMARK|STRNOTATMARK);
1358                                         }
1359                                 } else if (rflg && (old_sd_flag & STRATMARK)) {
1360                                         stp->sd_flag &= ~STRATMARK;
1361                                 }
1362                         }
1363 
1364                         /*
1365                          * Check for signal messages at the front of the read
1366                          * queue and generate the signal(s) if appropriate.
1367                          * The only signal that can be on queue is M_SIG at
1368                          * this point.
1369                          */
1370                         while ((((bp = q->q_first)) != NULL) &&
1371                             (bp->b_datap->db_type == M_SIG)) {
1372                                 bp = getq_noenab(q, 0);
1373                                 /*
1374                                  * sd_lock is held so the content of the
1375                                  * read queue can not change.
1376                                  */
1377                                 ASSERT(bp != NULL && DB_TYPE(bp) == M_SIG);
1378                                 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
1379                                 mutex_exit(&stp->sd_lock);
1380                                 freemsg(bp);
1381                                 if (STREAM_NEEDSERVICE(stp))
1382                                         stream_runservice(stp);
1383                                 mutex_enter(&stp->sd_lock);
1384                         }
1385 
1386                         if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1387                             delim ||
1388                             (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1389                                 goto oops;
1390                         }
1391                         continue;
1392 
1393                 case M_SIG:
1394                         strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1395                         freemsg(bp);
1396                         mutex_enter(&stp->sd_lock);
1397                         continue;
1398 
1399                 case M_PROTO:
1400                 case M_PCPROTO:
1401                         /*
1402                          * Only data messages are readable.
1403                          * Any others generate an error, unless
1404                          * RD_PROTDIS or RD_PROTDAT is set.
1405                          */
1406                         if (stp->sd_read_opt & RD_PROTDAT) {
1407                                 for (nbp = bp; nbp; nbp = nbp->b_next) {
1408                                         if ((nbp->b_datap->db_type ==
1409                                             M_PROTO) ||
1410                                             (nbp->b_datap->db_type ==
1411                                             M_PCPROTO)) {
1412                                                 nbp->b_datap->db_type = M_DATA;
1413                                         } else {
1414                                                 break;
1415                                         }
1416                                 }
1417                                 /*
1418                                  * clear stream head hi pri flag based on
1419                                  * first message
1420                                  */
1421                                 if (type == M_PCPROTO) {
1422                                         mutex_enter(&stp->sd_lock);
1423                                         stp->sd_flag &= ~STRPRI;
1424                                         mutex_exit(&stp->sd_lock);
1425                                 }
1426                                 goto ismdata;
1427                         } else if (stp->sd_read_opt & RD_PROTDIS) {
1428                                 /*
1429                                  * discard non-data messages
1430                                  */
1431                                 while (bp &&
1432                                     ((bp->b_datap->db_type == M_PROTO) ||
1433                                     (bp->b_datap->db_type == M_PCPROTO))) {
1434                                         nbp = unlinkb(bp);
1435                                         freeb(bp);
1436                                         bp = nbp;
1437                                 }
1438                                 /*
1439                                  * clear stream head hi pri flag based on
1440                                  * first message
1441                                  */
1442                                 if (type == M_PCPROTO) {
1443                                         mutex_enter(&stp->sd_lock);
1444                                         stp->sd_flag &= ~STRPRI;
1445                                         mutex_exit(&stp->sd_lock);
1446                                 }
1447                                 if (bp) {
1448                                         bp->b_band = pri;
1449                                         goto ismdata;
1450                                 } else {
1451                                         break;
1452                                 }
1453                         }
1454                         /* FALLTHRU */
1455                 case M_PASSFP:
1456                         if ((bp->b_datap->db_type == M_PASSFP) &&
1457                             (stp->sd_read_opt & RD_PROTDIS)) {
1458                                 freemsg(bp);
1459                                 break;
1460                         }
1461                         mutex_enter(&stp->sd_lock);
1462                         putback(stp, q, bp, pri);
1463                         mutex_exit(&stp->sd_lock);
1464                         if (rflg == 0)
1465                                 error = EBADMSG;
1466                         goto oops1;
1467 
1468                 default:
1469                         /*
1470                          * Garbage on stream head read queue.
1471                          */
1472                         cmn_err(CE_WARN, "bad %x found at stream head\n",
1473                             bp->b_datap->db_type);
1474                         freemsg(bp);
1475                         goto oops1;
1476                 }
1477                 mutex_enter(&stp->sd_lock);
1478         }
1479 oops:
1480         mutex_exit(&stp->sd_lock);
1481 oops1:
1482         qbackenable(q, pri);
1483         return (error);
1484 #undef  _LASTMARK
1485 }
1486 
1487 /*
1488  * Default processing of M_PROTO/M_PCPROTO messages.
1489  * Determine which wakeups and signals are needed.
1490  * This can be replaced by a user-specified procedure for kernel users
1491  * of STREAMS.
1492  */
1493 /* ARGSUSED */
1494 mblk_t *
1495 strrput_proto(vnode_t *vp, mblk_t *mp,
1496     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1497     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1498 {
1499         *wakeups = RSLEEP;
1500         *allmsgsigs = 0;
1501 
1502         switch (mp->b_datap->db_type) {
1503         case M_PROTO:
1504                 if (mp->b_band == 0) {
1505                         *firstmsgsigs = S_INPUT | S_RDNORM;
1506                         *pollwakeups = POLLIN | POLLRDNORM;
1507                 } else {
1508                         *firstmsgsigs = S_INPUT | S_RDBAND;
1509                         *pollwakeups = POLLIN | POLLRDBAND;
1510                 }
1511                 break;
1512         case M_PCPROTO:
1513                 *firstmsgsigs = S_HIPRI;
1514                 *pollwakeups = POLLPRI;
1515                 break;
1516         }
1517         return (mp);
1518 }
1519 
1520 /*
1521  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1522  * M_PASSFP messages.
1523  * Determine which wakeups and signals are needed.
1524  * This can be replaced by a user-specified procedure for kernel users
1525  * of STREAMS.
1526  */
1527 /* ARGSUSED */
1528 mblk_t *
1529 strrput_misc(vnode_t *vp, mblk_t *mp,
1530     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1531     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1532 {
1533         *wakeups = 0;
1534         *firstmsgsigs = 0;
1535         *allmsgsigs = 0;
1536         *pollwakeups = 0;
1537         return (mp);
1538 }
1539 
1540 /*
1541  * Stream read put procedure.  Called from downstream driver/module
1542  * with messages for the stream head.  Data, protocol, and in-stream
1543  * signal messages are placed on the queue, others are handled directly.
1544  */
1545 int
1546 strrput(queue_t *q, mblk_t *bp)
1547 {
1548         struct stdata   *stp;
1549         ulong_t         rput_opt;
1550         strwakeup_t     wakeups;
1551         strsigset_t     firstmsgsigs;   /* Signals if first message on queue */
1552         strsigset_t     allmsgsigs;     /* Signals for all messages */
1553         strsigset_t     signals;        /* Signals events to generate */
1554         strpollset_t    pollwakeups;
1555         mblk_t          *nextbp;
1556         uchar_t         band = 0;
1557         int             hipri_sig;
1558 
1559         stp = (struct stdata *)q->q_ptr;
1560         /*
1561          * Use rput_opt for optimized access to the SR_ flags except
1562          * SR_POLLIN. That flag has to be checked under sd_lock since it
1563          * is modified by strpoll().
1564          */
1565         rput_opt = stp->sd_rput_opt;
1566 
1567         ASSERT(qclaimed(q));
1568         TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1569             "strrput called with message type:q %p bp %p", q, bp);
1570 
1571         /*
1572          * Perform initial processing and pass to the parameterized functions.
1573          */
1574         ASSERT(bp->b_next == NULL);
1575 
1576         switch (bp->b_datap->db_type) {
1577         case M_DATA:
1578                 /*
1579                  * sockfs is the only consumer of STREOF and when it is set,
1580                  * it implies that the receiver is not interested in receiving
1581                  * any more data, hence the mblk is freed to prevent unnecessary
1582                  * message queueing at the stream head.
1583                  */
1584                 if (stp->sd_flag == STREOF) {
1585                         freemsg(bp);
1586                         return (0);
1587                 }
1588                 if ((rput_opt & SR_IGN_ZEROLEN) &&
1589                     bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1590                         /*
1591                          * Ignore zero-length M_DATA messages. These might be
1592                          * generated by some transports.
1593                          * The zero-length M_DATA messages, even if they
1594                          * are ignored, should effect the atmark tracking and
1595                          * should wake up a thread sleeping in strwaitmark.
1596                          */
1597                         mutex_enter(&stp->sd_lock);
1598                         if (bp->b_flag & MSGMARKNEXT) {
1599                                 /*
1600                                  * Record the position of the mark either
1601                                  * in q_last or in STRATMARK.
1602                                  */
1603                                 if (q->q_last != NULL) {
1604                                         q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1605                                         q->q_last->b_flag |= MSGMARKNEXT;
1606                                 } else {
1607                                         stp->sd_flag &= ~STRNOTATMARK;
1608                                         stp->sd_flag |= STRATMARK;
1609                                 }
1610                         } else if (bp->b_flag & MSGNOTMARKNEXT) {
1611                                 /*
1612                                  * Record that this is not the position of
1613                                  * the mark either in q_last or in
1614                                  * STRNOTATMARK.
1615                                  */
1616                                 if (q->q_last != NULL) {
1617                                         q->q_last->b_flag &= ~MSGMARKNEXT;
1618                                         q->q_last->b_flag |= MSGNOTMARKNEXT;
1619                                 } else {
1620                                         stp->sd_flag &= ~STRATMARK;
1621                                         stp->sd_flag |= STRNOTATMARK;
1622                                 }
1623                         }
1624                         if (stp->sd_flag & RSLEEP) {
1625                                 stp->sd_flag &= ~RSLEEP;
1626                                 cv_broadcast(&q->q_wait);
1627                         }
1628                         mutex_exit(&stp->sd_lock);
1629                         freemsg(bp);
1630                         return (0);
1631                 }
1632                 wakeups = RSLEEP;
1633                 if (bp->b_band == 0) {
1634                         firstmsgsigs = S_INPUT | S_RDNORM;
1635                         pollwakeups = POLLIN | POLLRDNORM;
1636                 } else {
1637                         firstmsgsigs = S_INPUT | S_RDBAND;
1638                         pollwakeups = POLLIN | POLLRDBAND;
1639                 }
1640                 if (rput_opt & SR_SIGALLDATA)
1641                         allmsgsigs = firstmsgsigs;
1642                 else
1643                         allmsgsigs = 0;
1644 
1645                 mutex_enter(&stp->sd_lock);
1646                 if ((rput_opt & SR_CONSOL_DATA) &&
1647                     (q->q_last != NULL) &&
1648                     (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1649                         /*
1650                          * Consolidate an M_DATA message onto an M_DATA,
1651                          * M_PROTO, or M_PCPROTO by merging it with q_last.
1652                          * The consolidation does not take place if
1653                          * the old message is marked with either of the
1654                          * marks or the delim flag or if the new
1655                          * message is marked with MSGMARK. The MSGMARK
1656                          * check is needed to handle the odd semantics of
1657                          * MSGMARK where essentially the whole message
1658                          * is to be treated as marked.
1659                          * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1660                          * new message to the front of the b_cont chain.
1661                          */
1662                         mblk_t *lbp = q->q_last;
1663                         unsigned char db_type = lbp->b_datap->db_type;
1664 
1665                         if ((db_type == M_DATA || db_type == M_PROTO ||
1666                             db_type == M_PCPROTO) &&
1667                             !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1668                                 rmvq_noenab(q, lbp);
1669                                 /*
1670                                  * The first message in the b_cont list
1671                                  * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1672                                  * We need to handle the case where we
1673                                  * are appending:
1674                                  *
1675                                  * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1676                                  * 2) a MSGMARKNEXT to a plain message.
1677                                  * 3) a MSGNOTMARKNEXT to a plain message
1678                                  * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1679                                  *    message.
1680                                  *
1681                                  * Thus we never append a MSGMARKNEXT or
1682                                  * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1683                                  */
1684                                 if (bp->b_flag & MSGMARKNEXT) {
1685                                         lbp->b_flag |= MSGMARKNEXT;
1686                                         lbp->b_flag &= ~MSGNOTMARKNEXT;
1687                                         bp->b_flag &= ~MSGMARKNEXT;
1688                                 } else if (bp->b_flag & MSGNOTMARKNEXT) {
1689                                         lbp->b_flag |= MSGNOTMARKNEXT;
1690                                         bp->b_flag &= ~MSGNOTMARKNEXT;
1691                                 }
1692 
1693                                 linkb(lbp, bp);
1694                                 bp = lbp;
1695                                 /*
1696                                  * The new message logically isn't the first
1697                                  * even though the q_first check below thinks
1698                                  * it is. Clear the firstmsgsigs to make it
1699                                  * not appear to be first.
1700                                  */
1701                                 firstmsgsigs = 0;
1702                         }
1703                 }
1704                 break;
1705 
1706         case M_PASSFP:
1707                 wakeups = RSLEEP;
1708                 allmsgsigs = 0;
1709                 if (bp->b_band == 0) {
1710                         firstmsgsigs = S_INPUT | S_RDNORM;
1711                         pollwakeups = POLLIN | POLLRDNORM;
1712                 } else {
1713                         firstmsgsigs = S_INPUT | S_RDBAND;
1714                         pollwakeups = POLLIN | POLLRDBAND;
1715                 }
1716                 mutex_enter(&stp->sd_lock);
1717                 break;
1718 
1719         case M_PROTO:
1720         case M_PCPROTO:
1721                 ASSERT(stp->sd_rprotofunc != NULL);
1722                 bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1723                     &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1724 #define ALLSIG  (S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1725                 S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1726 #define ALLPOLL (POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1727                 POLLWRBAND)
1728 
1729                 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1730                 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1731                 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1732                 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1733 
1734                 mutex_enter(&stp->sd_lock);
1735                 break;
1736 
1737         default:
1738                 ASSERT(stp->sd_rmiscfunc != NULL);
1739                 bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1740                     &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1741                 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1742                 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1743                 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1744                 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1745 #undef  ALLSIG
1746 #undef  ALLPOLL
1747                 mutex_enter(&stp->sd_lock);
1748                 break;
1749         }
1750         ASSERT(MUTEX_HELD(&stp->sd_lock));
1751 
1752         /* By default generate superset of signals */
1753         signals = (firstmsgsigs | allmsgsigs);
1754 
1755         /*
1756          * The  proto and misc functions can return multiple messages
1757          * as a b_next chain. Such messages are processed separately.
1758          */
1759 one_more:
1760         hipri_sig = 0;
1761         if (bp == NULL) {
1762                 nextbp = NULL;
1763         } else {
1764                 nextbp = bp->b_next;
1765                 bp->b_next = NULL;
1766 
1767                 switch (bp->b_datap->db_type) {
1768                 case M_PCPROTO:
1769                         /*
1770                          * Only one priority protocol message is allowed at the
1771                          * stream head at a time.
1772                          */
1773                         if (stp->sd_flag & STRPRI) {
1774                                 TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1775                                     "M_PCPROTO already at head");
1776                                 freemsg(bp);
1777                                 mutex_exit(&stp->sd_lock);
1778                                 goto done;
1779                         }
1780                         stp->sd_flag |= STRPRI;
1781                         hipri_sig = 1;
1782                         /* FALLTHRU */
1783                 case M_DATA:
1784                 case M_PROTO:
1785                 case M_PASSFP:
1786                         band = bp->b_band;
1787                         /*
1788                          * Marking doesn't work well when messages
1789                          * are marked in more than one band.  We only
1790                          * remember the last message received, even if
1791                          * it is placed on the queue ahead of other
1792                          * marked messages.
1793                          */
1794                         if (bp->b_flag & MSGMARK)
1795                                 stp->sd_mark = bp;
1796                         (void) putq(q, bp);
1797 
1798                         /*
1799                          * If message is a PCPROTO message, always use
1800                          * firstmsgsigs to determine if a signal should be
1801                          * sent as strrput is the only place to send
1802                          * signals for PCPROTO. Other messages are based on
1803                          * the STRGETINPROG flag. The flag determines if
1804                          * strrput or (k)strgetmsg will be responsible for
1805                          * sending the signals, in the firstmsgsigs case.
1806                          */
1807                         if ((hipri_sig == 1) ||
1808                             (((stp->sd_flag & STRGETINPROG) == 0) &&
1809                             (q->q_first == bp)))
1810                                 signals = (firstmsgsigs | allmsgsigs);
1811                         else
1812                                 signals = allmsgsigs;
1813                         break;
1814 
1815                 default:
1816                         mutex_exit(&stp->sd_lock);
1817                         (void) strrput_nondata(q, bp);
1818                         mutex_enter(&stp->sd_lock);
1819                         break;
1820                 }
1821         }
1822         ASSERT(MUTEX_HELD(&stp->sd_lock));
1823         /*
1824          * Wake sleeping read/getmsg and cancel deferred wakeup
1825          */
1826         if (wakeups & RSLEEP)
1827                 stp->sd_wakeq &= ~RSLEEP;
1828 
1829         wakeups &= stp->sd_flag;
1830         if (wakeups & RSLEEP) {
1831                 stp->sd_flag &= ~RSLEEP;
1832                 cv_broadcast(&q->q_wait);
1833         }
1834         if (wakeups & WSLEEP) {
1835                 stp->sd_flag &= ~WSLEEP;
1836                 cv_broadcast(&_WR(q)->q_wait);
1837         }
1838 
1839         if (pollwakeups != 0) {
1840                 if (pollwakeups == (POLLIN | POLLRDNORM)) {
1841                         /*
1842                          * Can't use rput_opt since it was not
1843                          * read when sd_lock was held and SR_POLLIN is changed
1844                          * by strpoll() under sd_lock.
1845                          */
1846                         if (!(stp->sd_rput_opt & SR_POLLIN))
1847                                 goto no_pollwake;
1848                         stp->sd_rput_opt &= ~SR_POLLIN;
1849                 }
1850                 mutex_exit(&stp->sd_lock);
1851                 pollwakeup(&stp->sd_pollist, pollwakeups);
1852                 mutex_enter(&stp->sd_lock);
1853         }
1854 no_pollwake:
1855 
1856         /*
1857          * strsendsig can handle multiple signals with a
1858          * single call.
1859          */
1860         if (stp->sd_sigflags & signals)
1861                 strsendsig(stp->sd_siglist, signals, band, 0);
1862         mutex_exit(&stp->sd_lock);
1863 
1864 
1865 done:
1866         if (nextbp == NULL)
1867                 return (0);
1868 
1869         /*
1870          * Any signals were handled the first time.
1871          * Wakeups and pollwakeups are redone to avoid any race
1872          * conditions - all the messages are not queued until the
1873          * last message has been processed by strrput.
1874          */
1875         bp = nextbp;
1876         signals = firstmsgsigs = allmsgsigs = 0;
1877         mutex_enter(&stp->sd_lock);
1878         goto one_more;
1879 }
1880 
1881 static void
1882 log_dupioc(queue_t *rq, mblk_t *bp)
1883 {
1884         queue_t *wq, *qp;
1885         char *modnames, *mnp, *dname;
1886         size_t maxmodstr;
1887         boolean_t islast;
1888 
1889         /*
1890          * Allocate a buffer large enough to hold the names of nstrpush modules
1891          * and one driver, with spaces between and NUL terminator.  If we can't
1892          * get memory, then we'll just log the driver name.
1893          */
1894         maxmodstr = nstrpush * (FMNAMESZ + 1);
1895         mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1896 
1897         /* march down write side to print log message down to the driver */
1898         wq = WR(rq);
1899 
1900         /* make sure q_next doesn't shift around while we're grabbing data */
1901         claimstr(wq);
1902         qp = wq->q_next;
1903         do {
1904                 dname = Q2NAME(qp);
1905                 islast = !SAMESTR(qp) || qp->q_next == NULL;
1906                 if (modnames == NULL) {
1907                         /*
1908                          * If we don't have memory, then get the driver name in
1909                          * the log where we can see it.  Note that memory
1910                          * pressure is a possible cause of these sorts of bugs.
1911                          */
1912                         if (islast) {
1913                                 modnames = dname;
1914                                 maxmodstr = 0;
1915                         }
1916                 } else {
1917                         mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1918                         if (!islast)
1919                                 *mnp++ = ' ';
1920                 }
1921                 qp = qp->q_next;
1922         } while (!islast);
1923         releasestr(wq);
1924         /* Cannot happen unless stream head is corrupt. */
1925         ASSERT(modnames != NULL);
1926         (void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1927             SL_CONSOLE|SL_TRACE|SL_ERROR,
1928             "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1929             rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1930             (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1931         if (maxmodstr != 0)
1932                 kmem_free(modnames, maxmodstr);
1933 }
1934 
1935 int
1936 strrput_nondata(queue_t *q, mblk_t *bp)
1937 {
1938         struct stdata *stp;
1939         struct iocblk *iocbp;
1940         struct stroptions *sop;
1941         struct copyreq *reqp;
1942         struct copyresp *resp;
1943         unsigned char bpri;
1944         unsigned char  flushed_already = 0;
1945 
1946         stp = (struct stdata *)q->q_ptr;
1947 
1948         ASSERT(!(stp->sd_flag & STPLEX));
1949         ASSERT(qclaimed(q));
1950 
1951         switch (bp->b_datap->db_type) {
1952         case M_ERROR:
1953                 /*
1954                  * An error has occurred downstream, the errno is in the first
1955                  * bytes of the message.
1956                  */
1957                 if ((bp->b_wptr - bp->b_rptr) == 2) {     /* New flavor */
1958                         unsigned char rw = 0;
1959 
1960                         mutex_enter(&stp->sd_lock);
1961                         if (*bp->b_rptr != NOERROR) {        /* read error */
1962                                 if (*bp->b_rptr != 0) {
1963                                         if (stp->sd_flag & STRDERR)
1964                                                 flushed_already |= FLUSHR;
1965                                         stp->sd_flag |= STRDERR;
1966                                         rw |= FLUSHR;
1967                                 } else {
1968                                         stp->sd_flag &= ~STRDERR;
1969                                 }
1970                                 stp->sd_rerror = *bp->b_rptr;
1971                         }
1972                         bp->b_rptr++;
1973                         if (*bp->b_rptr != NOERROR) {        /* write error */
1974                                 if (*bp->b_rptr != 0) {
1975                                         if (stp->sd_flag & STWRERR)
1976                                                 flushed_already |= FLUSHW;
1977                                         stp->sd_flag |= STWRERR;
1978                                         rw |= FLUSHW;
1979                                 } else {
1980                                         stp->sd_flag &= ~STWRERR;
1981                                 }
1982                                 stp->sd_werror = *bp->b_rptr;
1983                         }
1984                         if (rw) {
1985                                 TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1986                                     "strrput cv_broadcast:q %p, bp %p",
1987                                     q, bp);
1988                                 cv_broadcast(&q->q_wait); /* readers */
1989                                 cv_broadcast(&_WR(q)->q_wait); /* writers */
1990                                 cv_broadcast(&stp->sd_monitor); /* ioctllers */
1991 
1992                                 mutex_exit(&stp->sd_lock);
1993                                 pollwakeup(&stp->sd_pollist, POLLERR);
1994                                 mutex_enter(&stp->sd_lock);
1995 
1996                                 if (stp->sd_sigflags & S_ERROR)
1997                                         strsendsig(stp->sd_siglist, S_ERROR, 0,
1998                                             ((rw & FLUSHR) ? stp->sd_rerror :
1999                                             stp->sd_werror));
2000                                 mutex_exit(&stp->sd_lock);
2001                                 /*
2002                                  * Send the M_FLUSH only
2003                                  * for the first M_ERROR
2004                                  * message on the stream
2005                                  */
2006                                 if (flushed_already == rw) {
2007                                         freemsg(bp);
2008                                         return (0);
2009                                 }
2010 
2011                                 bp->b_datap->db_type = M_FLUSH;
2012                                 *bp->b_rptr = rw;
2013                                 bp->b_wptr = bp->b_rptr + 1;
2014                                 /*
2015                                  * Protect against the driver
2016                                  * passing up messages after
2017                                  * it has done a qprocsoff
2018                                  */
2019                                 if (_OTHERQ(q)->q_next == NULL)
2020                                         freemsg(bp);
2021                                 else
2022                                         qreply(q, bp);
2023                                 return (0);
2024                         } else
2025                                 mutex_exit(&stp->sd_lock);
2026                 } else if (*bp->b_rptr != 0) {               /* Old flavor */
2027                                 if (stp->sd_flag & (STRDERR|STWRERR))
2028                                         flushed_already = FLUSHRW;
2029                                 mutex_enter(&stp->sd_lock);
2030                                 stp->sd_flag |= (STRDERR|STWRERR);
2031                                 stp->sd_rerror = *bp->b_rptr;
2032                                 stp->sd_werror = *bp->b_rptr;
2033                                 TRACE_2(TR_FAC_STREAMS_FR,
2034                                     TR_STRRPUT_WAKE2,
2035                                     "strrput wakeup #2:q %p, bp %p", q, bp);
2036                                 cv_broadcast(&q->q_wait); /* the readers */
2037                                 cv_broadcast(&_WR(q)->q_wait); /* the writers */
2038                                 cv_broadcast(&stp->sd_monitor); /* ioctllers */
2039 
2040                                 mutex_exit(&stp->sd_lock);
2041                                 pollwakeup(&stp->sd_pollist, POLLERR);
2042                                 mutex_enter(&stp->sd_lock);
2043 
2044                                 if (stp->sd_sigflags & S_ERROR)
2045                                         strsendsig(stp->sd_siglist, S_ERROR, 0,
2046                                             (stp->sd_werror ? stp->sd_werror :
2047                                             stp->sd_rerror));
2048                                 mutex_exit(&stp->sd_lock);
2049 
2050                                 /*
2051                                  * Send the M_FLUSH only
2052                                  * for the first M_ERROR
2053                                  * message on the stream
2054                                  */
2055                                 if (flushed_already != FLUSHRW) {
2056                                         bp->b_datap->db_type = M_FLUSH;
2057                                         *bp->b_rptr = FLUSHRW;
2058                                         /*
2059                                          * Protect against the driver passing up
2060                                          * messages after it has done a
2061                                          * qprocsoff.
2062                                          */
2063                                 if (_OTHERQ(q)->q_next == NULL)
2064                                         freemsg(bp);
2065                                 else
2066                                         qreply(q, bp);
2067                                 return (0);
2068                                 }
2069                 }
2070                 freemsg(bp);
2071                 return (0);
2072 
2073         case M_HANGUP:
2074 
2075                 freemsg(bp);
2076                 mutex_enter(&stp->sd_lock);
2077                 stp->sd_werror = ENXIO;
2078                 stp->sd_flag |= STRHUP;
2079                 stp->sd_flag &= ~(WSLEEP|RSLEEP);
2080 
2081                 /*
2082                  * send signal if controlling tty
2083                  */
2084 
2085                 if (stp->sd_sidp) {
2086                         prsignal(stp->sd_sidp, SIGHUP);
2087                         if (stp->sd_sidp != stp->sd_pgidp)
2088                                 pgsignal(stp->sd_pgidp, SIGTSTP);
2089                 }
2090 
2091                 /*
2092                  * wake up read, write, and exception pollers and
2093                  * reset wakeup mechanism.
2094                  */
2095                 cv_broadcast(&q->q_wait);        /* the readers */
2096                 cv_broadcast(&_WR(q)->q_wait);   /* the writers */
2097                 cv_broadcast(&stp->sd_monitor);  /* the ioctllers */
2098                 strhup(stp);
2099                 mutex_exit(&stp->sd_lock);
2100                 return (0);
2101 
2102         case M_UNHANGUP:
2103                 freemsg(bp);
2104                 mutex_enter(&stp->sd_lock);
2105                 stp->sd_werror = 0;
2106                 stp->sd_flag &= ~STRHUP;
2107                 mutex_exit(&stp->sd_lock);
2108                 return (0);
2109 
2110         case M_SIG:
2111                 /*
2112                  * Someone downstream wants to post a signal.  The
2113                  * signal to post is contained in the first byte of the
2114                  * message.  If the message would go on the front of
2115                  * the queue, send a signal to the process group
2116                  * (if not SIGPOLL) or to the siglist processes
2117                  * (SIGPOLL).  If something is already on the queue,
2118                  * OR if we are delivering a delayed suspend (*sigh*
2119                  * another "tty" hack) and there's no one sleeping already,
2120                  * just enqueue the message.
2121                  */
2122                 mutex_enter(&stp->sd_lock);
2123                 if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2124                     !(stp->sd_flag & RSLEEP))) {
2125                         (void) putq(q, bp);
2126                         mutex_exit(&stp->sd_lock);
2127                         return (0);
2128                 }
2129                 mutex_exit(&stp->sd_lock);
2130                 /* FALLTHRU */
2131 
2132         case M_PCSIG:
2133                 /*
2134                  * Don't enqueue, just post the signal.
2135                  */
2136                 strsignal(stp, *bp->b_rptr, 0L);
2137                 freemsg(bp);
2138                 return (0);
2139 
2140         case M_CMD:
2141                 if (MBLKL(bp) != sizeof (cmdblk_t)) {
2142                         freemsg(bp);
2143                         return (0);
2144                 }
2145 
2146                 mutex_enter(&stp->sd_lock);
2147                 if (stp->sd_flag & STRCMDWAIT) {
2148                         ASSERT(stp->sd_cmdblk == NULL);
2149                         stp->sd_cmdblk = bp;
2150                         cv_broadcast(&stp->sd_monitor);
2151                         mutex_exit(&stp->sd_lock);
2152                 } else {
2153                         mutex_exit(&stp->sd_lock);
2154                         freemsg(bp);
2155                 }
2156                 return (0);
2157 
2158         case M_FLUSH:
2159                 /*
2160                  * Flush queues.  The indication of which queues to flush
2161                  * is in the first byte of the message.  If the read queue
2162                  * is specified, then flush it.  If FLUSHBAND is set, just
2163                  * flush the band specified by the second byte of the message.
2164                  *
2165                  * If a module has issued a M_SETOPT to not flush hi
2166                  * priority messages off of the stream head, then pass this
2167                  * flag into the flushq code to preserve such messages.
2168                  */
2169 
2170                 if (*bp->b_rptr & FLUSHR) {
2171                         mutex_enter(&stp->sd_lock);
2172                         if (*bp->b_rptr & FLUSHBAND) {
2173                                 ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2174                                 flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2175                         } else
2176                                 flushq_common(q, FLUSHALL,
2177                                     stp->sd_read_opt & RFLUSHPCPROT);
2178                         if ((q->q_first == NULL) ||
2179                             (q->q_first->b_datap->db_type < QPCTL))
2180                                 stp->sd_flag &= ~STRPRI;
2181                         else {
2182                                 ASSERT(stp->sd_flag & STRPRI);
2183                         }
2184                         mutex_exit(&stp->sd_lock);
2185                 }
2186                 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2187                         *bp->b_rptr &= ~FLUSHR;
2188                         bp->b_flag |= MSGNOLOOP;
2189                         /*
2190                          * Protect against the driver passing up
2191                          * messages after it has done a qprocsoff.
2192                          */
2193                         if (_OTHERQ(q)->q_next == NULL)
2194                                 freemsg(bp);
2195                         else
2196                                 qreply(q, bp);
2197                         return (0);
2198                 }
2199                 freemsg(bp);
2200                 return (0);
2201 
2202         case M_IOCACK:
2203         case M_IOCNAK:
2204                 iocbp = (struct iocblk *)bp->b_rptr;
2205                 /*
2206                  * If not waiting for ACK or NAK then just free msg.
2207                  * If incorrect id sequence number then just free msg.
2208                  * If already have ACK or NAK for user then this is a
2209                  *    duplicate, display a warning and free the msg.
2210                  */
2211                 mutex_enter(&stp->sd_lock);
2212                 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2213                     (stp->sd_iocid != iocbp->ioc_id)) {
2214                         /*
2215                          * If the ACK/NAK is a dup, display a message
2216                          * Dup is when sd_iocid == ioc_id, and
2217                          * sd_iocblk == <valid ptr> or -1 (the former
2218                          * is when an ioctl has been put on the stream
2219                          * head, but has not yet been consumed, the
2220                          * later is when it has been consumed).
2221                          */
2222                         if ((stp->sd_iocid == iocbp->ioc_id) &&
2223                             (stp->sd_iocblk != NULL)) {
2224                                 log_dupioc(q, bp);
2225                         }
2226                         freemsg(bp);
2227                         mutex_exit(&stp->sd_lock);
2228                         return (0);
2229                 }
2230 
2231                 /*
2232                  * Assign ACK or NAK to user and wake up.
2233                  */
2234                 stp->sd_iocblk = bp;
2235                 cv_broadcast(&stp->sd_monitor);
2236                 mutex_exit(&stp->sd_lock);
2237                 return (0);
2238 
2239         case M_COPYIN:
2240         case M_COPYOUT:
2241                 reqp = (struct copyreq *)bp->b_rptr;
2242 
2243                 /*
2244                  * If not waiting for ACK or NAK then just fail request.
2245                  * If already have ACK, NAK, or copy request, then just
2246                  * fail request.
2247                  * If incorrect id sequence number then just fail request.
2248                  */
2249                 mutex_enter(&stp->sd_lock);
2250                 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2251                     (stp->sd_iocid != reqp->cq_id)) {
2252                         if (bp->b_cont) {
2253                                 freemsg(bp->b_cont);
2254                                 bp->b_cont = NULL;
2255                         }
2256                         bp->b_datap->db_type = M_IOCDATA;
2257                         bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2258                         resp = (struct copyresp *)bp->b_rptr;
2259                         resp->cp_rval = (caddr_t)1;  /* failure */
2260                         mutex_exit(&stp->sd_lock);
2261                         putnext(stp->sd_wrq, bp);
2262                         return (0);
2263                 }
2264 
2265                 /*
2266                  * Assign copy request to user and wake up.
2267                  */
2268                 stp->sd_iocblk = bp;
2269                 cv_broadcast(&stp->sd_monitor);
2270                 mutex_exit(&stp->sd_lock);
2271                 return (0);
2272 
2273         case M_SETOPTS:
2274                 /*
2275                  * Set stream head options (read option, write offset,
2276                  * min/max packet size, and/or high/low water marks for
2277                  * the read side only).
2278                  */
2279 
2280                 bpri = 0;
2281                 sop = (struct stroptions *)bp->b_rptr;
2282                 mutex_enter(&stp->sd_lock);
2283                 if (sop->so_flags & SO_READOPT) {
2284                         switch (sop->so_readopt & RMODEMASK) {
2285                         case RNORM:
2286                                 stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2287                                 break;
2288 
2289                         case RMSGD:
2290                                 stp->sd_read_opt =
2291                                     ((stp->sd_read_opt & ~RD_MSGNODIS) |
2292                                     RD_MSGDIS);
2293                                 break;
2294 
2295                         case RMSGN:
2296                                 stp->sd_read_opt =
2297                                     ((stp->sd_read_opt & ~RD_MSGDIS) |
2298                                     RD_MSGNODIS);
2299                                 break;
2300                         }
2301                         switch (sop->so_readopt & RPROTMASK) {
2302                         case RPROTNORM:
2303                                 stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2304                                 break;
2305 
2306                         case RPROTDAT:
2307                                 stp->sd_read_opt =
2308                                     ((stp->sd_read_opt & ~RD_PROTDIS) |
2309                                     RD_PROTDAT);
2310                                 break;
2311 
2312                         case RPROTDIS:
2313                                 stp->sd_read_opt =
2314                                     ((stp->sd_read_opt & ~RD_PROTDAT) |
2315                                     RD_PROTDIS);
2316                                 break;
2317                         }
2318                         switch (sop->so_readopt & RFLUSHMASK) {
2319                         case RFLUSHPCPROT:
2320                                 /*
2321                                  * This sets the stream head to NOT flush
2322                                  * M_PCPROTO messages.
2323                                  */
2324                                 stp->sd_read_opt |= RFLUSHPCPROT;
2325                                 break;
2326                         }
2327                 }
2328                 if (sop->so_flags & SO_ERROPT) {
2329                         switch (sop->so_erropt & RERRMASK) {
2330                         case RERRNORM:
2331                                 stp->sd_flag &= ~STRDERRNONPERSIST;
2332                                 break;
2333                         case RERRNONPERSIST:
2334                                 stp->sd_flag |= STRDERRNONPERSIST;
2335                                 break;
2336                         }
2337                         switch (sop->so_erropt & WERRMASK) {
2338                         case WERRNORM:
2339                                 stp->sd_flag &= ~STWRERRNONPERSIST;
2340                                 break;
2341                         case WERRNONPERSIST:
2342                                 stp->sd_flag |= STWRERRNONPERSIST;
2343                                 break;
2344                         }
2345                 }
2346                 if (sop->so_flags & SO_COPYOPT) {
2347                         if (sop->so_copyopt & ZCVMSAFE) {
2348                                 stp->sd_copyflag |= STZCVMSAFE;
2349                                 stp->sd_copyflag &= ~STZCVMUNSAFE;
2350                         } else if (sop->so_copyopt & ZCVMUNSAFE) {
2351                                 stp->sd_copyflag |= STZCVMUNSAFE;
2352                                 stp->sd_copyflag &= ~STZCVMSAFE;
2353                         }
2354 
2355                         if (sop->so_copyopt & COPYCACHED) {
2356                                 stp->sd_copyflag |= STRCOPYCACHED;
2357                         }
2358                 }
2359                 if (sop->so_flags & SO_WROFF)
2360                         stp->sd_wroff = sop->so_wroff;
2361                 if (sop->so_flags & SO_TAIL)
2362                         stp->sd_tail = sop->so_tail;
2363                 if (sop->so_flags & SO_MINPSZ)
2364                         q->q_minpsz = sop->so_minpsz;
2365                 if (sop->so_flags & SO_MAXPSZ)
2366                         q->q_maxpsz = sop->so_maxpsz;
2367                 if (sop->so_flags & SO_MAXBLK)
2368                         stp->sd_maxblk = sop->so_maxblk;
2369                 if (sop->so_flags & SO_HIWAT) {
2370                         if (sop->so_flags & SO_BAND) {
2371                                 if (strqset(q, QHIWAT,
2372                                     sop->so_band, sop->so_hiwat)) {
2373                                         cmn_err(CE_WARN, "strrput: could not "
2374                                             "allocate qband\n");
2375                                 } else {
2376                                         bpri = sop->so_band;
2377                                 }
2378                         } else {
2379                                 q->q_hiwat = sop->so_hiwat;
2380                         }
2381                 }
2382                 if (sop->so_flags & SO_LOWAT) {
2383                         if (sop->so_flags & SO_BAND) {
2384                                 if (strqset(q, QLOWAT,
2385                                     sop->so_band, sop->so_lowat)) {
2386                                         cmn_err(CE_WARN, "strrput: could not "
2387                                             "allocate qband\n");
2388                                 } else {
2389                                         bpri = sop->so_band;
2390                                 }
2391                         } else {
2392                                 q->q_lowat = sop->so_lowat;
2393                         }
2394                 }
2395                 if (sop->so_flags & SO_MREADON)
2396                         stp->sd_flag |= SNDMREAD;
2397                 if (sop->so_flags & SO_MREADOFF)
2398                         stp->sd_flag &= ~SNDMREAD;
2399                 if (sop->so_flags & SO_NDELON)
2400                         stp->sd_flag |= OLDNDELAY;
2401                 if (sop->so_flags & SO_NDELOFF)
2402                         stp->sd_flag &= ~OLDNDELAY;
2403                 if (sop->so_flags & SO_ISTTY)
2404                         stp->sd_flag |= STRISTTY;
2405                 if (sop->so_flags & SO_ISNTTY)
2406                         stp->sd_flag &= ~STRISTTY;
2407                 if (sop->so_flags & SO_TOSTOP)
2408                         stp->sd_flag |= STRTOSTOP;
2409                 if (sop->so_flags & SO_TONSTOP)
2410                         stp->sd_flag &= ~STRTOSTOP;
2411                 if (sop->so_flags & SO_DELIM)
2412                         stp->sd_flag |= STRDELIM;
2413                 if (sop->so_flags & SO_NODELIM)
2414                         stp->sd_flag &= ~STRDELIM;
2415 
2416                 mutex_exit(&stp->sd_lock);
2417                 freemsg(bp);
2418 
2419                 /* Check backenable in case the water marks changed */
2420                 qbackenable(q, bpri);
2421                 return (0);
2422 
2423         /*
2424          * The following set of cases deal with situations where two stream
2425          * heads are connected to each other (twisted streams).  These messages
2426          * have no meaning at the stream head.
2427          */
2428         case M_BREAK:
2429         case M_CTL:
2430         case M_DELAY:
2431         case M_START:
2432         case M_STOP:
2433         case M_IOCDATA:
2434         case M_STARTI:
2435         case M_STOPI:
2436                 freemsg(bp);
2437                 return (0);
2438 
2439         case M_IOCTL:
2440                 /*
2441                  * Always NAK this condition
2442                  * (makes no sense)
2443                  * If there is one or more threads in the read side
2444                  * rwnext we have to defer the nacking until that thread
2445                  * returns (in strget).
2446                  */
2447                 mutex_enter(&stp->sd_lock);
2448                 if (stp->sd_struiodnak != 0) {
2449                         /*
2450                          * Defer NAK to the streamhead. Queue at the end
2451                          * the list.
2452                          */
2453                         mblk_t *mp = stp->sd_struionak;
2454 
2455                         while (mp && mp->b_next)
2456                                 mp = mp->b_next;
2457                         if (mp)
2458                                 mp->b_next = bp;
2459                         else
2460                                 stp->sd_struionak = bp;
2461                         bp->b_next = NULL;
2462                         mutex_exit(&stp->sd_lock);
2463                         return (0);
2464                 }
2465                 mutex_exit(&stp->sd_lock);
2466 
2467                 bp->b_datap->db_type = M_IOCNAK;
2468                 /*
2469                  * Protect against the driver passing up
2470                  * messages after it has done a qprocsoff.
2471                  */
2472                 if (_OTHERQ(q)->q_next == NULL)
2473                         freemsg(bp);
2474                 else
2475                         qreply(q, bp);
2476                 return (0);
2477 
2478         default:
2479 #ifdef DEBUG
2480                 cmn_err(CE_WARN,
2481                     "bad message type %x received at stream head\n",
2482                     bp->b_datap->db_type);
2483 #endif
2484                 freemsg(bp);
2485                 return (0);
2486         }
2487 
2488         /* NOTREACHED */
2489 }
2490 
2491 /*
2492  * Check if the stream pointed to by `stp' can be written to, and return an
2493  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2494  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2495  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2496  */
2497 static int
2498 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2499 {
2500         int error;
2501 
2502         ASSERT(MUTEX_HELD(&stp->sd_lock));
2503 
2504         /*
2505          * For modem support, POSIX states that on writes, EIO should
2506          * be returned if the stream has been hung up.
2507          */
2508         if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2509                 error = EIO;
2510         else
2511                 error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2512 
2513         if (error != 0) {
2514                 if (!(stp->sd_flag & STPLEX) &&
2515                     (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2516                         tsignal(curthread, SIGPIPE);
2517                         error = EPIPE;
2518                 }
2519         }
2520 
2521         return (error);
2522 }
2523 
2524 /*
2525  * Copyin and send data down a stream.
2526  * The caller will allocate and copyin any control part that precedes the
2527  * message and pass that in as mctl.
2528  *
2529  * Caller should *not* hold sd_lock.
2530  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2531  * under sd_lock in order to avoid missing a backenabling wakeup.
2532  *
2533  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2534  *
2535  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2536  * For sync streams we can only ignore flow control by reverting to using
2537  * putnext.
2538  *
2539  * If sd_maxblk is less than *iosize this routine might return without
2540  * transferring all of *iosize. In all cases, on return *iosize will contain
2541  * the amount of data that was transferred.
2542  */
2543 static int
2544 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2545     int b_flag, int pri, int flags)
2546 {
2547         struiod_t uiod;
2548         mblk_t *mp;
2549         queue_t *wqp = stp->sd_wrq;
2550         int error = 0;
2551         ssize_t count = *iosize;
2552 
2553         ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2554 
2555         if (uiop != NULL && count >= 0)
2556                 flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2557 
2558         if (!(flags & STRUIO_POSTPONE)) {
2559                 /*
2560                  * Use regular canputnext, strmakedata, putnext sequence.
2561                  */
2562                 if (pri == 0) {
2563                         if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2564                                 freemsg(mctl);
2565                                 return (EWOULDBLOCK);
2566                         }
2567                 } else {
2568                         if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2569                                 freemsg(mctl);
2570                                 return (EWOULDBLOCK);
2571                         }
2572                 }
2573 
2574                 if ((error = strmakedata(iosize, uiop, stp, flags,
2575                     &mp)) != 0) {
2576                         freemsg(mctl);
2577                         /*
2578                          * need to change return code to ENOMEM
2579                          * so that this is not confused with
2580                          * flow control, EAGAIN.
2581                          */
2582 
2583                         if (error == EAGAIN)
2584                                 return (ENOMEM);
2585                         else
2586                                 return (error);
2587                 }
2588                 if (mctl != NULL) {
2589                         if (mctl->b_cont == NULL)
2590                                 mctl->b_cont = mp;
2591                         else if (mp != NULL)
2592                                 linkb(mctl, mp);
2593                         mp = mctl;
2594                 } else if (mp == NULL)
2595                         return (0);
2596 
2597                 mp->b_flag |= b_flag;
2598                 mp->b_band = (uchar_t)pri;
2599 
2600                 if (flags & MSG_IGNFLOW) {
2601                         /*
2602                          * XXX Hack: Don't get stuck running service
2603                          * procedures. This is needed for sockfs when
2604                          * sending the unbind message out of the rput
2605                          * procedure - we don't want a put procedure
2606                          * to run service procedures.
2607                          */
2608                         putnext(wqp, mp);
2609                 } else {
2610                         stream_willservice(stp);
2611                         putnext(wqp, mp);
2612                         stream_runservice(stp);
2613                 }
2614                 return (0);
2615         }
2616         /*
2617          * Stream supports rwnext() for the write side.
2618          */
2619         if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2620                 freemsg(mctl);
2621                 /*
2622                  * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2623                  */
2624                 return (error == EAGAIN ? ENOMEM : error);
2625         }
2626         if (mctl != NULL) {
2627                 if (mctl->b_cont == NULL)
2628                         mctl->b_cont = mp;
2629                 else if (mp != NULL)
2630                         linkb(mctl, mp);
2631                 mp = mctl;
2632         } else if (mp == NULL) {
2633                 return (0);
2634         }
2635 
2636         mp->b_flag |= b_flag;
2637         mp->b_band = (uchar_t)pri;
2638 
2639         (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2640             sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2641         uiod.d_uio.uio_offset = 0;
2642         uiod.d_mp = mp;
2643         error = rwnext(wqp, &uiod);
2644         if (! uiod.d_mp) {
2645                 uioskip(uiop, *iosize);
2646                 return (error);
2647         }
2648         ASSERT(mp == uiod.d_mp);
2649         if (error == EINVAL) {
2650                 /*
2651                  * The stream plumbing must have changed while
2652                  * we were away, so just turn off rwnext()s.
2653                  */
2654                 error = 0;
2655         } else if (error == EBUSY || error == EWOULDBLOCK) {
2656                 /*
2657                  * Couldn't enter a perimeter or took a page fault,
2658                  * so fall-back to putnext().
2659                  */
2660                 error = 0;
2661         } else {
2662                 freemsg(mp);
2663                 return (error);
2664         }
2665         /* Have to check canput before consuming data from the uio */
2666         if (pri == 0) {
2667                 if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2668                         freemsg(mp);
2669                         return (EWOULDBLOCK);
2670                 }
2671         } else {
2672                 if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2673                         freemsg(mp);
2674                         return (EWOULDBLOCK);
2675                 }
2676         }
2677         ASSERT(mp == uiod.d_mp);
2678         /* Copyin data from the uio */
2679         if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2680                 freemsg(mp);
2681                 return (error);
2682         }
2683         uioskip(uiop, *iosize);
2684         if (flags & MSG_IGNFLOW) {
2685                 /*
2686                  * XXX Hack: Don't get stuck running service procedures.
2687                  * This is needed for sockfs when sending the unbind message
2688                  * out of the rput procedure - we don't want a put procedure
2689                  * to run service procedures.
2690                  */
2691                 putnext(wqp, mp);
2692         } else {
2693                 stream_willservice(stp);
2694                 putnext(wqp, mp);
2695                 stream_runservice(stp);
2696         }
2697         return (0);
2698 }
2699 
2700 /*
2701  * Write attempts to break the write request into messages conforming
2702  * with the minimum and maximum packet sizes set downstream.
2703  *
2704  * Write will not block if downstream queue is full and
2705  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2706  *
2707  * A write of zero bytes gets packaged into a zero length message and sent
2708  * downstream like any other message.
2709  *
2710  * If buffers of the requested sizes are not available, the write will
2711  * sleep until the buffers become available.
2712  *
2713  * Write (if specified) will supply a write offset in a message if it
2714  * makes sense. This can be specified by downstream modules as part of
2715  * a M_SETOPTS message.  Write will not supply the write offset if it
2716  * cannot supply any data in a buffer.  In other words, write will never
2717  * send down an empty packet due to a write offset.
2718  */
2719 /* ARGSUSED2 */
2720 int
2721 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2722 {
2723         return (strwrite_common(vp, uiop, crp, 0));
2724 }
2725 
2726 /* ARGSUSED2 */
2727 int
2728 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2729 {
2730         struct stdata *stp;
2731         struct queue *wqp;
2732         ssize_t rmin, rmax;
2733         ssize_t iosize;
2734         int waitflag;
2735         int tempmode;
2736         int error = 0;
2737         int b_flag;
2738 
2739         ASSERT(vp->v_stream);
2740         stp = vp->v_stream;
2741 
2742         mutex_enter(&stp->sd_lock);
2743 
2744         if ((error = i_straccess(stp, JCWRITE)) != 0) {
2745                 mutex_exit(&stp->sd_lock);
2746                 return (error);
2747         }
2748 
2749         if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2750                 error = strwriteable(stp, B_TRUE, B_TRUE);
2751                 if (error != 0) {
2752                         mutex_exit(&stp->sd_lock);
2753                         return (error);
2754                 }
2755         }
2756 
2757         mutex_exit(&stp->sd_lock);
2758 
2759         wqp = stp->sd_wrq;
2760 
2761         /* get these values from them cached in the stream head */
2762         rmin = stp->sd_qn_minpsz;
2763         rmax = stp->sd_qn_maxpsz;
2764 
2765         /*
2766          * Check the min/max packet size constraints.  If min packet size
2767          * is non-zero, the write cannot be split into multiple messages
2768          * and still guarantee the size constraints.
2769          */
2770         TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2771 
2772         ASSERT((rmax >= 0) || (rmax == INFPSZ));
2773         if (rmax == 0) {
2774                 return (0);
2775         }
2776         if (rmin > 0) {
2777                 if (uiop->uio_resid < rmin) {
2778                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2779                             "strwrite out:q %p out %d error %d",
2780                             wqp, 0, ERANGE);
2781                         return (ERANGE);
2782                 }
2783                 if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2784                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2785                             "strwrite out:q %p out %d error %d",
2786                             wqp, 1, ERANGE);
2787                         return (ERANGE);
2788                 }
2789         }
2790 
2791         /*
2792          * Do until count satisfied or error.
2793          */
2794         waitflag = WRITEWAIT | wflag;
2795         if (stp->sd_flag & OLDNDELAY)
2796                 tempmode = uiop->uio_fmode & ~FNDELAY;
2797         else
2798                 tempmode = uiop->uio_fmode;
2799 
2800         if (rmax == INFPSZ)
2801                 rmax = uiop->uio_resid;
2802 
2803         /*
2804          * Note that tempmode does not get used in strput/strmakedata
2805          * but only in strwaitq. The other routines use uio_fmode
2806          * unmodified.
2807          */
2808 
2809         /* LINTED: constant in conditional context */
2810         while (1) {     /* breaks when uio_resid reaches zero */
2811                 /*
2812                  * Determine the size of the next message to be
2813                  * packaged.  May have to break write into several
2814                  * messages based on max packet size.
2815                  */
2816                 iosize = MIN(uiop->uio_resid, rmax);
2817 
2818                 /*
2819                  * Put block downstream when flow control allows it.
2820                  */
2821                 if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2822                         b_flag = MSGDELIM;
2823                 else
2824                         b_flag = 0;
2825 
2826                 for (;;) {
2827                         int done = 0;
2828 
2829                         error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2830                         if (error == 0)
2831                                 break;
2832                         if (error != EWOULDBLOCK)
2833                                 goto out;
2834 
2835                         mutex_enter(&stp->sd_lock);
2836                         /*
2837                          * Check for a missed wakeup.
2838                          * Needed since strput did not hold sd_lock across
2839                          * the canputnext.
2840                          */
2841                         if (canputnext(wqp)) {
2842                                 /* Try again */
2843                                 mutex_exit(&stp->sd_lock);
2844                                 continue;
2845                         }
2846                         TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2847                             "strwrite wait:q %p wait", wqp);
2848                         if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2849                             tempmode, -1, &done)) != 0 || done) {
2850                                 mutex_exit(&stp->sd_lock);
2851                                 if ((vp->v_type == VFIFO) &&
2852                                     (uiop->uio_fmode & FNDELAY) &&
2853                                     (error == EAGAIN))
2854                                         error = 0;
2855                                 goto out;
2856                         }
2857                         TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2858                             "strwrite wake:q %p awakes", wqp);
2859                         if ((error = i_straccess(stp, JCWRITE)) != 0) {
2860                                 mutex_exit(&stp->sd_lock);
2861                                 goto out;
2862                         }
2863                         mutex_exit(&stp->sd_lock);
2864                 }
2865                 waitflag |= NOINTR;
2866                 TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2867                     "strwrite resid:q %p uiop %p", wqp, uiop);
2868                 if (uiop->uio_resid) {
2869                         /* Recheck for errors - needed for sockets */
2870                         if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2871                             (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2872                                 mutex_enter(&stp->sd_lock);
2873                                 error = strwriteable(stp, B_FALSE, B_TRUE);
2874                                 mutex_exit(&stp->sd_lock);
2875                                 if (error != 0)
2876                                         return (error);
2877                         }
2878                         continue;
2879                 }
2880                 break;
2881         }
2882 out:
2883         /*
2884          * For historical reasons, applications expect EAGAIN when a data
2885          * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2886          */
2887         if (error == ENOMEM)
2888                 error = EAGAIN;
2889         TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2890             "strwrite out:q %p out %d error %d", wqp, 2, error);
2891         return (error);
2892 }
2893 
2894 /*
2895  * Stream head write service routine.
2896  * Its job is to wake up any sleeping writers when a queue
2897  * downstream needs data (part of the flow control in putq and getq).
2898  * It also must wake anyone sleeping on a poll().
2899  * For stream head right below mux module, it must also invoke put procedure
2900  * of next downstream module.
2901  */
2902 int
2903 strwsrv(queue_t *q)
2904 {
2905         struct stdata *stp;
2906         queue_t *tq;
2907         qband_t *qbp;
2908         int i;
2909         qband_t *myqbp;
2910         int isevent;
2911         unsigned char   qbf[NBAND];     /* band flushing backenable flags */
2912 
2913         TRACE_1(TR_FAC_STREAMS_FR,
2914             TR_STRWSRV, "strwsrv:q %p", q);
2915         stp = (struct stdata *)q->q_ptr;
2916         ASSERT(qclaimed(q));
2917         mutex_enter(&stp->sd_lock);
2918         ASSERT(!(stp->sd_flag & STPLEX));
2919 
2920         if (stp->sd_flag & WSLEEP) {
2921                 stp->sd_flag &= ~WSLEEP;
2922                 cv_broadcast(&q->q_wait);
2923         }
2924         mutex_exit(&stp->sd_lock);
2925 
2926         /* The other end of a stream pipe went away. */
2927         if ((tq = q->q_next) == NULL) {
2928                 return (0);
2929         }
2930 
2931         /* Find the next module forward that has a service procedure */
2932         claimstr(q);
2933         tq = q->q_nfsrv;
2934         ASSERT(tq != NULL);
2935 
2936         if ((q->q_flag & QBACK)) {
2937                 if ((tq->q_flag & QFULL)) {
2938                         mutex_enter(QLOCK(tq));
2939                         if (!(tq->q_flag & QFULL)) {
2940                                 mutex_exit(QLOCK(tq));
2941                                 goto wakeup;
2942                         }
2943                         /*
2944                          * The queue must have become full again. Set QWANTW
2945                          * again so strwsrv will be back enabled when
2946                          * the queue becomes non-full next time.
2947                          */
2948                         tq->q_flag |= QWANTW;
2949                         mutex_exit(QLOCK(tq));
2950                 } else {
2951                 wakeup:
2952                         pollwakeup(&stp->sd_pollist, POLLWRNORM);
2953                         mutex_enter(&stp->sd_lock);
2954                         if (stp->sd_sigflags & S_WRNORM)
2955                                 strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2956                         mutex_exit(&stp->sd_lock);
2957                 }
2958         }
2959 
2960         isevent = 0;
2961         i = 1;
2962         bzero((caddr_t)qbf, NBAND);
2963         mutex_enter(QLOCK(tq));
2964         if ((myqbp = q->q_bandp) != NULL)
2965                 for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2966                         ASSERT(myqbp);
2967                         if ((myqbp->qb_flag & QB_BACK)) {
2968                                 if (qbp->qb_flag & QB_FULL) {
2969                                         /*
2970                                          * The band must have become full again.
2971                                          * Set QB_WANTW again so strwsrv will
2972                                          * be back enabled when the band becomes
2973                                          * non-full next time.
2974                                          */
2975                                         qbp->qb_flag |= QB_WANTW;
2976                                 } else {
2977                                         isevent = 1;
2978                                         qbf[i] = 1;
2979                                 }
2980                         }
2981                         myqbp = myqbp->qb_next;
2982                         i++;
2983                 }
2984         mutex_exit(QLOCK(tq));
2985 
2986         if (isevent) {
2987                 for (i = tq->q_nband; i; i--) {
2988                         if (qbf[i]) {
2989                                 pollwakeup(&stp->sd_pollist, POLLWRBAND);
2990                                 mutex_enter(&stp->sd_lock);
2991                                 if (stp->sd_sigflags & S_WRBAND)
2992                                         strsendsig(stp->sd_siglist, S_WRBAND,
2993                                             (uchar_t)i, 0);
2994                                 mutex_exit(&stp->sd_lock);
2995                         }
2996                 }
2997         }
2998 
2999         releasestr(q);
3000         return (0);
3001 }
3002 
3003 /*
3004  * Special case of strcopyin/strcopyout for copying
3005  * struct strioctl that can deal with both data
3006  * models.
3007  */
3008 
3009 #ifdef  _LP64
3010 
3011 static int
3012 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3013 {
3014         struct  strioctl32 strioc32;
3015         struct  strioctl *striocp;
3016 
3017         if (copyflag & U_TO_K) {
3018                 ASSERT((copyflag & K_TO_K) == 0);
3019 
3020                 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3021                         if (copyin(from, &strioc32, sizeof (strioc32)))
3022                                 return (EFAULT);
3023 
3024                         striocp = (struct strioctl *)to;
3025                         striocp->ic_cmd      = strioc32.ic_cmd;
3026                         striocp->ic_timout = strioc32.ic_timout;
3027                         striocp->ic_len      = strioc32.ic_len;
3028                         striocp->ic_dp       = (char *)(uintptr_t)strioc32.ic_dp;
3029 
3030                 } else { /* NATIVE data model */
3031                         if (copyin(from, to, sizeof (struct strioctl))) {
3032                                 return (EFAULT);
3033                         } else {
3034                                 return (0);
3035                         }
3036                 }
3037         } else {
3038                 ASSERT(copyflag & K_TO_K);
3039                 bcopy(from, to, sizeof (struct strioctl));
3040         }
3041         return (0);
3042 }
3043 
3044 static int
3045 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3046 {
3047         struct  strioctl32 strioc32;
3048         struct  strioctl *striocp;
3049 
3050         if (copyflag & U_TO_K) {
3051                 ASSERT((copyflag & K_TO_K) == 0);
3052 
3053                 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3054                         striocp = (struct strioctl *)from;
3055                         strioc32.ic_cmd = striocp->ic_cmd;
3056                         strioc32.ic_timout = striocp->ic_timout;
3057                         strioc32.ic_len = striocp->ic_len;
3058                         strioc32.ic_dp  = (caddr32_t)(uintptr_t)striocp->ic_dp;
3059                         ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3060                             striocp->ic_dp);
3061 
3062                         if (copyout(&strioc32, to, sizeof (strioc32)))
3063                                 return (EFAULT);
3064 
3065                 } else { /* NATIVE data model */
3066                         if (copyout(from, to, sizeof (struct strioctl))) {
3067                                 return (EFAULT);
3068                         } else {
3069                                 return (0);
3070                         }
3071                 }
3072         } else {
3073                 ASSERT(copyflag & K_TO_K);
3074                 bcopy(from, to, sizeof (struct strioctl));
3075         }
3076         return (0);
3077 }
3078 
3079 #else   /* ! _LP64 */
3080 
3081 /* ARGSUSED2 */
3082 static int
3083 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3084 {
3085         return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3086 }
3087 
3088 /* ARGSUSED2 */
3089 static int
3090 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3091 {
3092         return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3093 }
3094 
3095 #endif  /* _LP64 */
3096 
3097 /*
3098  * Determine type of job control semantics expected by user.  The
3099  * possibilities are:
3100  *      JCREAD  - Behaves like read() on fd; send SIGTTIN
3101  *      JCWRITE - Behaves like write() on fd; send SIGTTOU if TOSTOP set
3102  *      JCSETP  - Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3103  *      JCGETP  - Gets a value in the stream; no signals.
3104  * See straccess in strsubr.c for usage of these values.
3105  *
3106  * This routine also returns -1 for I_STR as a special case; the
3107  * caller must call again with the real ioctl number for
3108  * classification.
3109  */
3110 static int
3111 job_control_type(int cmd)
3112 {
3113         switch (cmd) {
3114         case I_STR:
3115                 return (-1);
3116 
3117         case I_RECVFD:
3118         case I_E_RECVFD:
3119                 return (JCREAD);
3120 
3121         case I_FDINSERT:
3122         case I_SENDFD:
3123                 return (JCWRITE);
3124 
3125         case TCSETA:
3126         case TCSETAW:
3127         case TCSETAF:
3128         case TCSBRK:
3129         case TCXONC:
3130         case TCFLSH:
3131         case TCDSET:    /* Obsolete */
3132         case TIOCSWINSZ:
3133         case TCSETS:
3134         case TCSETSW:
3135         case TCSETSF:
3136         case TIOCSETD:
3137         case TIOCHPCL:
3138         case TIOCSETP:
3139         case TIOCSETN:
3140         case TIOCEXCL:
3141         case TIOCNXCL:
3142         case TIOCFLUSH:
3143         case TIOCSETC:
3144         case TIOCLBIS:
3145         case TIOCLBIC:
3146         case TIOCLSET:
3147         case TIOCSBRK:
3148         case TIOCCBRK:
3149         case TIOCSDTR:
3150         case TIOCCDTR:
3151         case TIOCSLTC:
3152         case TIOCSTOP:
3153         case TIOCSTART:
3154         case TIOCSTI:
3155         case TIOCSPGRP:
3156         case TIOCMSET:
3157         case TIOCMBIS:
3158         case TIOCMBIC:
3159         case TIOCREMOTE:
3160         case TIOCSIGNAL:
3161         case LDSETT:
3162         case LDSMAP:    /* Obsolete */
3163         case DIOCSETP:
3164         case I_FLUSH:
3165         case I_SRDOPT:
3166         case I_SETSIG:
3167         case I_SWROPT:
3168         case I_FLUSHBAND:
3169         case I_SETCLTIME:
3170         case I_SERROPT:
3171         case I_ESETSIG:
3172         case FIONBIO:
3173         case FIOASYNC:
3174         case FIOSETOWN:
3175         case JBOOT:     /* Obsolete */
3176         case JTERM:     /* Obsolete */
3177         case JTIMOM:    /* Obsolete */
3178         case JZOMBOOT:  /* Obsolete */
3179         case JAGENT:    /* Obsolete */
3180         case JTRUN:     /* Obsolete */
3181         case JXTPROTO:  /* Obsolete */
3182                 return (JCSETP);
3183         }
3184 
3185         return (JCGETP);
3186 }
3187 
3188 /*
3189  * ioctl for streams
3190  */
3191 int
3192 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3193     cred_t *crp, int *rvalp)
3194 {
3195         struct stdata *stp;
3196         struct strcmd *scp;
3197         struct strioctl strioc;
3198         struct uio uio;
3199         struct iovec iov;
3200         int access;
3201         mblk_t *mp;
3202         int error = 0;
3203         int done = 0;
3204         ssize_t rmin, rmax;
3205         queue_t *wrq;
3206         queue_t *rdq;
3207         boolean_t kioctl = B_FALSE;
3208         uint32_t auditing = AU_AUDITING();
3209 
3210         if (flag & FKIOCTL) {
3211                 copyflag = K_TO_K;
3212                 kioctl = B_TRUE;
3213         }
3214         ASSERT(vp->v_stream);
3215         ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3216         stp = vp->v_stream;
3217 
3218         TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3219             "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3220 
3221         /*
3222          * If the copy is kernel to kernel, make sure that the FNATIVE
3223          * flag is set.  After this it would be a serious error to have
3224          * no model flag.
3225          */
3226         if (copyflag == K_TO_K)
3227                 flag = (flag & ~FMODELS) | FNATIVE;
3228 
3229         ASSERT((flag & FMODELS) != 0);
3230 
3231         wrq = stp->sd_wrq;
3232         rdq = _RD(wrq);
3233 
3234         access = job_control_type(cmd);
3235 
3236         /* We should never see these here, should be handled by iwscn */
3237         if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3238                 return (EINVAL);
3239 
3240         mutex_enter(&stp->sd_lock);
3241         if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3242                 mutex_exit(&stp->sd_lock);
3243                 return (error);
3244         }
3245         mutex_exit(&stp->sd_lock);
3246 
3247         /*
3248          * Check for sgttyb-related ioctls first, and complain as
3249          * necessary.
3250          */
3251         switch (cmd) {
3252         case TIOCGETP:
3253         case TIOCSETP:
3254         case TIOCSETN:
3255                 if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3256                         sgttyb_complaint = B_TRUE;
3257                         cmn_err(CE_NOTE,
3258                             "application used obsolete TIOC[GS]ET");
3259                 }
3260                 if (sgttyb_handling >= 3) {
3261                         tsignal(curthread, SIGSYS);
3262                         return (EIO);
3263                 }
3264                 break;
3265         }
3266 
3267         mutex_enter(&stp->sd_lock);
3268 
3269         switch (cmd) {
3270         case I_RECVFD:
3271         case I_E_RECVFD:
3272         case I_PEEK:
3273         case I_NREAD:
3274         case FIONREAD:
3275         case FIORDCHK:
3276         case I_ATMARK:
3277         case FIONBIO:
3278         case FIOASYNC:
3279                 if (stp->sd_flag & (STRDERR|STPLEX)) {
3280                         error = strgeterr(stp, STRDERR|STPLEX, 0);
3281                         if (error != 0) {
3282                                 mutex_exit(&stp->sd_lock);
3283                                 return (error);
3284                         }
3285                 }
3286                 break;
3287 
3288         default:
3289                 if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3290                         error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3291                         if (error != 0) {
3292                                 mutex_exit(&stp->sd_lock);
3293                                 return (error);
3294                         }
3295                 }
3296         }
3297 
3298         mutex_exit(&stp->sd_lock);
3299 
3300         switch (cmd) {
3301         default:
3302                 /*
3303                  * The stream head has hardcoded knowledge of a
3304                  * miscellaneous collection of terminal-, keyboard- and
3305                  * mouse-related ioctls, enumerated below.  This hardcoded
3306                  * knowledge allows the stream head to automatically
3307                  * convert transparent ioctl requests made by userland
3308                  * programs into I_STR ioctls which many old STREAMS
3309                  * modules and drivers require.
3310                  *
3311                  * No new ioctls should ever be added to this list.
3312                  * Instead, the STREAMS module or driver should be written
3313                  * to either handle transparent ioctls or require any
3314                  * userland programs to use I_STR ioctls (by returning
3315                  * EINVAL to any transparent ioctl requests).
3316                  *
3317                  * More importantly, removing ioctls from this list should
3318                  * be done with the utmost care, since our STREAMS modules
3319                  * and drivers *count* on the stream head performing this
3320                  * conversion, and thus may panic while processing
3321                  * transparent ioctl request for one of these ioctls (keep
3322                  * in mind that third party modules and drivers may have
3323                  * similar problems).
3324                  */
3325                 if (((cmd & IOCTYPE) == LDIOC) ||
3326                     ((cmd & IOCTYPE) == tIOC) ||
3327                     ((cmd & IOCTYPE) == TIOC) ||
3328                     ((cmd & IOCTYPE) == KIOC) ||
3329                     ((cmd & IOCTYPE) == MSIOC) ||
3330                     ((cmd & IOCTYPE) == VUIOC)) {
3331                         /*
3332                          * The ioctl is a tty ioctl - set up strioc buffer
3333                          * and call strdoioctl() to do the work.
3334                          */
3335                         if (stp->sd_flag & STRHUP)
3336                                 return (ENXIO);
3337                         strioc.ic_cmd = cmd;
3338                         strioc.ic_timout = INFTIM;
3339 
3340                         switch (cmd) {
3341 
3342                         case TCXONC:
3343                         case TCSBRK:
3344                         case TCFLSH:
3345                         case TCDSET:
3346                                 {
3347                                 int native_arg = (int)arg;
3348                                 strioc.ic_len = sizeof (int);
3349                                 strioc.ic_dp = (char *)&native_arg;
3350                                 return (strdoioctl(stp, &strioc, flag,
3351                                     K_TO_K, crp, rvalp));
3352                                 }
3353 
3354                         case TCSETA:
3355                         case TCSETAW:
3356                         case TCSETAF:
3357                                 strioc.ic_len = sizeof (struct termio);
3358                                 strioc.ic_dp = (char *)arg;
3359                                 return (strdoioctl(stp, &strioc, flag,
3360                                     copyflag, crp, rvalp));
3361 
3362                         case TCSETS:
3363                         case TCSETSW:
3364                         case TCSETSF:
3365                                 strioc.ic_len = sizeof (struct termios);
3366                                 strioc.ic_dp = (char *)arg;
3367                                 return (strdoioctl(stp, &strioc, flag,
3368                                     copyflag, crp, rvalp));
3369 
3370                         case LDSETT:
3371                                 strioc.ic_len = sizeof (struct termcb);
3372                                 strioc.ic_dp = (char *)arg;
3373                                 return (strdoioctl(stp, &strioc, flag,
3374                                     copyflag, crp, rvalp));
3375 
3376                         case TIOCSETP:
3377                                 strioc.ic_len = sizeof (struct sgttyb);
3378                                 strioc.ic_dp = (char *)arg;
3379                                 return (strdoioctl(stp, &strioc, flag,
3380                                     copyflag, crp, rvalp));
3381 
3382                         case TIOCSTI:
3383                                 if ((flag & FREAD) == 0 &&
3384                                     secpolicy_sti(crp) != 0) {
3385                                         return (EPERM);
3386                                 }
3387                                 mutex_enter(&stp->sd_lock);
3388                                 mutex_enter(&curproc->p_splock);
3389                                 if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3390                                     secpolicy_sti(crp) != 0) {
3391                                         mutex_exit(&curproc->p_splock);
3392                                         mutex_exit(&stp->sd_lock);
3393                                         return (EACCES);
3394                                 }
3395                                 mutex_exit(&curproc->p_splock);
3396                                 mutex_exit(&stp->sd_lock);
3397 
3398                                 strioc.ic_len = sizeof (char);
3399                                 strioc.ic_dp = (char *)arg;
3400                                 return (strdoioctl(stp, &strioc, flag,
3401                                     copyflag, crp, rvalp));
3402 
3403                         case TIOCSWINSZ:
3404                                 strioc.ic_len = sizeof (struct winsize);
3405                                 strioc.ic_dp = (char *)arg;
3406                                 return (strdoioctl(stp, &strioc, flag,
3407                                     copyflag, crp, rvalp));
3408 
3409                         case TIOCSSIZE:
3410                                 strioc.ic_len = sizeof (struct ttysize);
3411                                 strioc.ic_dp = (char *)arg;
3412                                 return (strdoioctl(stp, &strioc, flag,
3413                                     copyflag, crp, rvalp));
3414 
3415                         case TIOCSSOFTCAR:
3416                         case KIOCTRANS:
3417                         case KIOCTRANSABLE:
3418                         case KIOCCMD:
3419                         case KIOCSDIRECT:
3420                         case KIOCSCOMPAT:
3421                         case KIOCSKABORTEN:
3422                         case KIOCSRPTDELAY:
3423                         case KIOCSRPTRATE:
3424                         case VUIDSFORMAT:
3425                         case TIOCSPPS:
3426                                 strioc.ic_len = sizeof (int);
3427                                 strioc.ic_dp = (char *)arg;
3428                                 return (strdoioctl(stp, &strioc, flag,
3429                                     copyflag, crp, rvalp));
3430 
3431                         case KIOCSETKEY:
3432                         case KIOCGETKEY:
3433                                 strioc.ic_len = sizeof (struct kiockey);
3434                                 strioc.ic_dp = (char *)arg;
3435                                 return (strdoioctl(stp, &strioc, flag,
3436                                     copyflag, crp, rvalp));
3437 
3438                         case KIOCSKEY:
3439                         case KIOCGKEY:
3440                                 strioc.ic_len = sizeof (struct kiockeymap);
3441                                 strioc.ic_dp = (char *)arg;
3442                                 return (strdoioctl(stp, &strioc, flag,
3443                                     copyflag, crp, rvalp));
3444 
3445                         case KIOCSLED:
3446                                 /* arg is a pointer to char */
3447                                 strioc.ic_len = sizeof (char);
3448                                 strioc.ic_dp = (char *)arg;
3449                                 return (strdoioctl(stp, &strioc, flag,
3450                                     copyflag, crp, rvalp));
3451 
3452                         case MSIOSETPARMS:
3453                                 strioc.ic_len = sizeof (Ms_parms);
3454                                 strioc.ic_dp = (char *)arg;
3455                                 return (strdoioctl(stp, &strioc, flag,
3456                                     copyflag, crp, rvalp));
3457 
3458                         case VUIDSADDR:
3459                         case VUIDGADDR:
3460                                 strioc.ic_len = sizeof (struct vuid_addr_probe);
3461                                 strioc.ic_dp = (char *)arg;
3462                                 return (strdoioctl(stp, &strioc, flag,
3463                                     copyflag, crp, rvalp));
3464 
3465                         /*
3466                          * These M_IOCTL's don't require any data to be sent
3467                          * downstream, and the driver will allocate and link
3468                          * on its own mblk_t upon M_IOCACK -- thus we set
3469                          * ic_len to zero and set ic_dp to arg so we know
3470                          * where to copyout to later.
3471                          */
3472                         case TIOCGSOFTCAR:
3473                         case TIOCGWINSZ:
3474                         case TIOCGSIZE:
3475                         case KIOCGTRANS:
3476                         case KIOCGTRANSABLE:
3477                         case KIOCTYPE:
3478                         case KIOCGDIRECT:
3479                         case KIOCGCOMPAT:
3480                         case KIOCLAYOUT:
3481                         case KIOCGLED:
3482                         case MSIOGETPARMS:
3483                         case MSIOBUTTONS:
3484                         case VUIDGFORMAT:
3485                         case TIOCGPPS:
3486                         case TIOCGPPSEV:
3487                         case TCGETA:
3488                         case TCGETS:
3489                         case LDGETT:
3490                         case TIOCGETP:
3491                         case KIOCGRPTDELAY:
3492                         case KIOCGRPTRATE:
3493                                 strioc.ic_len = 0;
3494                                 strioc.ic_dp = (char *)arg;
3495                                 return (strdoioctl(stp, &strioc, flag,
3496                                     copyflag, crp, rvalp));
3497                         }
3498                 }
3499 
3500                 /*
3501                  * Unknown cmd - send it down as a transparent ioctl.
3502                  */
3503                 strioc.ic_cmd = cmd;
3504                 strioc.ic_timout = INFTIM;
3505                 strioc.ic_len = TRANSPARENT;
3506                 strioc.ic_dp = (char *)&arg;
3507 
3508                 return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3509 
3510         case I_STR:
3511                 /*
3512                  * Stream ioctl.  Read in an strioctl buffer from the user
3513                  * along with any data specified and send it downstream.
3514                  * Strdoioctl will wait allow only one ioctl message at
3515                  * a time, and waits for the acknowledgement.
3516                  */
3517 
3518                 if (stp->sd_flag & STRHUP)
3519                         return (ENXIO);
3520 
3521                 error = strcopyin_strioctl((void *)arg, &strioc, flag,
3522                     copyflag);
3523                 if (error != 0)
3524                         return (error);
3525 
3526                 if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3527                         return (EINVAL);
3528 
3529                 access = job_control_type(strioc.ic_cmd);
3530                 mutex_enter(&stp->sd_lock);
3531                 if ((access != -1) &&
3532                     ((error = i_straccess(stp, access)) != 0)) {
3533                         mutex_exit(&stp->sd_lock);
3534                         return (error);
3535                 }
3536                 mutex_exit(&stp->sd_lock);
3537 
3538                 /*
3539                  * The I_STR facility provides a trap door for malicious
3540                  * code to send down bogus streamio(7I) ioctl commands to
3541                  * unsuspecting STREAMS modules and drivers which expect to
3542                  * only get these messages from the stream head.
3543                  * Explicitly prohibit any streamio ioctls which can be
3544                  * passed downstream by the stream head.  Note that we do
3545                  * not block all streamio ioctls because the ioctl
3546                  * numberspace is not well managed and thus it's possible
3547                  * that a module or driver's ioctl numbers may accidentally
3548                  * collide with them.
3549                  */
3550                 switch (strioc.ic_cmd) {
3551                 case I_LINK:
3552                 case I_PLINK:
3553                 case I_UNLINK:
3554                 case I_PUNLINK:
3555                 case _I_GETPEERCRED:
3556                 case _I_PLINK_LH:
3557                         return (EINVAL);
3558                 }
3559 
3560                 error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3561                 if (error == 0) {
3562                         error = strcopyout_strioctl(&strioc, (void *)arg,
3563                             flag, copyflag);
3564                 }
3565                 return (error);
3566 
3567         case _I_CMD:
3568                 /*
3569                  * Like I_STR, but without using M_IOC* messages and without
3570                  * copyins/copyouts beyond the passed-in argument.
3571                  */
3572                 if (stp->sd_flag & STRHUP)
3573                         return (ENXIO);
3574 
3575                 if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3576                         return (ENOMEM);
3577 
3578                 if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3579                         kmem_free(scp, sizeof (strcmd_t));
3580                         return (EFAULT);
3581                 }
3582 
3583                 access = job_control_type(scp->sc_cmd);
3584                 mutex_enter(&stp->sd_lock);
3585                 if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3586                         mutex_exit(&stp->sd_lock);
3587                         kmem_free(scp, sizeof (strcmd_t));
3588                         return (error);
3589                 }
3590                 mutex_exit(&stp->sd_lock);
3591 
3592                 *rvalp = 0;
3593                 if ((error = strdocmd(stp, scp, crp)) == 0) {
3594                         if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3595                                 error = EFAULT;
3596                 }
3597                 kmem_free(scp, sizeof (strcmd_t));
3598                 return (error);
3599 
3600         case I_NREAD:
3601                 /*
3602                  * Return number of bytes of data in first message
3603                  * in queue in "arg" and return the number of messages
3604                  * in queue in return value.
3605                  */
3606         {
3607                 size_t  size;
3608                 int     retval;
3609                 int     count = 0;
3610 
3611                 mutex_enter(QLOCK(rdq));
3612 
3613                 size = msgdsize(rdq->q_first);
3614                 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3615                         count++;
3616 
3617                 mutex_exit(QLOCK(rdq));
3618                 if (stp->sd_struiordq) {
3619                         infod_t infod;
3620 
3621                         infod.d_cmd = INFOD_COUNT;
3622                         infod.d_count = 0;
3623                         if (count == 0) {
3624                                 infod.d_cmd |= INFOD_FIRSTBYTES;
3625                                 infod.d_bytes = 0;
3626                         }
3627                         infod.d_res = 0;
3628                         (void) infonext(rdq, &infod);
3629                         count += infod.d_count;
3630                         if (infod.d_res & INFOD_FIRSTBYTES)
3631                                 size = infod.d_bytes;
3632                 }
3633 
3634                 /*
3635                  * Drop down from size_t to the "int" required by the
3636                  * interface.  Cap at INT_MAX.
3637                  */
3638                 retval = MIN(size, INT_MAX);
3639                 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3640                     copyflag);
3641                 if (!error)
3642                         *rvalp = count;
3643                 return (error);
3644         }
3645 
3646         case FIONREAD:
3647                 /*
3648                  * Return number of bytes of data in all data messages
3649                  * in queue in "arg".
3650                  */
3651         {
3652                 size_t  size = 0;
3653                 int     retval;
3654 
3655                 mutex_enter(QLOCK(rdq));
3656                 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3657                         size += msgdsize(mp);
3658                 mutex_exit(QLOCK(rdq));
3659 
3660                 if (stp->sd_struiordq) {
3661                         infod_t infod;
3662 
3663                         infod.d_cmd = INFOD_BYTES;
3664                         infod.d_res = 0;
3665                         infod.d_bytes = 0;
3666                         (void) infonext(rdq, &infod);
3667                         size += infod.d_bytes;
3668                 }
3669 
3670                 /*
3671                  * Drop down from size_t to the "int" required by the
3672                  * interface.  Cap at INT_MAX.
3673                  */
3674                 retval = MIN(size, INT_MAX);
3675                 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3676                     copyflag);
3677 
3678                 *rvalp = 0;
3679                 return (error);
3680         }
3681         case FIORDCHK:
3682                 /*
3683                  * FIORDCHK does not use arg value (like FIONREAD),
3684                  * instead a count is returned. I_NREAD value may
3685                  * not be accurate but safe. The real thing to do is
3686                  * to add the msgdsizes of all data  messages until
3687                  * a non-data message.
3688                  */
3689         {
3690                 size_t size = 0;
3691 
3692                 mutex_enter(QLOCK(rdq));
3693                 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3694                         size += msgdsize(mp);
3695                 mutex_exit(QLOCK(rdq));
3696 
3697                 if (stp->sd_struiordq) {
3698                         infod_t infod;
3699 
3700                         infod.d_cmd = INFOD_BYTES;
3701                         infod.d_res = 0;
3702                         infod.d_bytes = 0;
3703                         (void) infonext(rdq, &infod);
3704                         size += infod.d_bytes;
3705                 }
3706 
3707                 /*
3708                  * Since ioctl returns an int, and memory sizes under
3709                  * LP64 may not fit, we return INT_MAX if the count was
3710                  * actually greater.
3711                  */
3712                 *rvalp = MIN(size, INT_MAX);
3713                 return (0);
3714         }
3715 
3716         case I_FIND:
3717                 /*
3718                  * Get module name.
3719                  */
3720         {
3721                 char mname[FMNAMESZ + 1];
3722                 queue_t *q;
3723 
3724                 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3725                     mname, FMNAMESZ + 1, NULL);
3726                 if (error)
3727                         return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3728 
3729                 /*
3730                  * Return EINVAL if we're handed a bogus module name.
3731                  */
3732                 if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3733                         TRACE_0(TR_FAC_STREAMS_FR,
3734                             TR_I_CANT_FIND, "couldn't I_FIND");
3735                         return (EINVAL);
3736                 }
3737 
3738                 *rvalp = 0;
3739 
3740                 /* Look downstream to see if module is there. */
3741                 claimstr(stp->sd_wrq);
3742                 for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3743                         if (q->q_flag & QREADR) {
3744                                 q = NULL;
3745                                 break;
3746                         }
3747                         if (strcmp(mname, Q2NAME(q)) == 0)
3748                                 break;
3749                 }
3750                 releasestr(stp->sd_wrq);
3751 
3752                 *rvalp = (q ? 1 : 0);
3753                 return (error);
3754         }
3755 
3756         case I_PUSH:
3757         case __I_PUSH_NOCTTY:
3758                 /*
3759                  * Push a module.
3760                  * For the case __I_PUSH_NOCTTY push a module but
3761                  * do not allocate controlling tty. See bugid 4025044
3762                  */
3763 
3764         {
3765                 char mname[FMNAMESZ + 1];
3766                 fmodsw_impl_t *fp;
3767                 dev_t dummydev;
3768 
3769                 if (stp->sd_flag & STRHUP)
3770                         return (ENXIO);
3771 
3772                 /*
3773                  * Get module name and look up in fmodsw.
3774                  */
3775                 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3776                     mname, FMNAMESZ + 1, NULL);
3777                 if (error)
3778                         return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3779 
3780                 if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3781                     NULL)
3782                         return (EINVAL);
3783 
3784                 TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3785                     "I_PUSH:fp %p stp %p", fp, stp);
3786 
3787                 /*
3788                  * If the module is flagged as single-instance, then check
3789                  * to see if the module is already pushed. If it is, return
3790                  * as if the push was successful.
3791                  */
3792                 if (fp->f_qflag & _QSINGLE_INSTANCE) {
3793                         queue_t *q;
3794 
3795                         claimstr(stp->sd_wrq);
3796                         for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3797                                 if (q->q_flag & QREADR) {
3798                                         q = NULL;
3799                                         break;
3800                                 }
3801                                 if (strcmp(mname, Q2NAME(q)) == 0)
3802                                         break;
3803                         }
3804                         releasestr(stp->sd_wrq);
3805                         if (q != NULL) {
3806                                 fmodsw_rele(fp);
3807                                 return (0);
3808                         }
3809                 }
3810 
3811                 if (error = strstartplumb(stp, flag, cmd)) {
3812                         fmodsw_rele(fp);
3813                         return (error);
3814                 }
3815 
3816                 /*
3817                  * See if any more modules can be pushed on this stream.
3818                  * Note that this check must be done after strstartplumb()
3819                  * since otherwise multiple threads issuing I_PUSHes on
3820                  * the same stream will be able to exceed nstrpush.
3821                  */
3822                 mutex_enter(&stp->sd_lock);
3823                 if (stp->sd_pushcnt >= nstrpush) {
3824                         fmodsw_rele(fp);
3825                         strendplumb(stp);
3826                         mutex_exit(&stp->sd_lock);
3827                         return (EINVAL);
3828                 }
3829                 mutex_exit(&stp->sd_lock);
3830 
3831                 /*
3832                  * Push new module and call its open routine
3833                  * via qattach().  Modules don't change device
3834                  * numbers, so just ignore dummydev here.
3835                  */
3836                 dummydev = vp->v_rdev;
3837                 if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3838                     B_FALSE)) == 0) {
3839                         if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3840                             (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3841                                 /*
3842                                  * try to allocate it as a controlling terminal
3843                                  */
3844                                 (void) strctty(stp);
3845                         }
3846                 }
3847 
3848                 mutex_enter(&stp->sd_lock);
3849 
3850                 /*
3851                  * As a performance concern we are caching the values of
3852                  * q_minpsz and q_maxpsz of the module below the stream
3853                  * head in the stream head.
3854                  */
3855                 mutex_enter(QLOCK(stp->sd_wrq->q_next));
3856                 rmin = stp->sd_wrq->q_next->q_minpsz;
3857                 rmax = stp->sd_wrq->q_next->q_maxpsz;
3858                 mutex_exit(QLOCK(stp->sd_wrq->q_next));
3859 
3860                 /* Do this processing here as a performance concern */
3861                 if (strmsgsz != 0) {
3862                         if (rmax == INFPSZ)
3863                                 rmax = strmsgsz;
3864                         else  {
3865                                 if (vp->v_type == VFIFO)
3866                                         rmax = MIN(PIPE_BUF, rmax);
3867                                 else    rmax = MIN(strmsgsz, rmax);
3868                         }
3869                 }
3870 
3871                 mutex_enter(QLOCK(wrq));
3872                 stp->sd_qn_minpsz = rmin;
3873                 stp->sd_qn_maxpsz = rmax;
3874                 mutex_exit(QLOCK(wrq));
3875 
3876                 strendplumb(stp);
3877                 mutex_exit(&stp->sd_lock);
3878                 return (error);
3879         }
3880 
3881         case I_POP:
3882         {
3883                 queue_t *q;
3884 
3885                 if (stp->sd_flag & STRHUP)
3886                         return (ENXIO);
3887                 if (!wrq->q_next)    /* for broken pipes */
3888                         return (EINVAL);
3889 
3890                 if (error = strstartplumb(stp, flag, cmd))
3891                         return (error);
3892 
3893                 /*
3894                  * If there is an anchor on this stream and popping
3895                  * the current module would attempt to pop through the
3896                  * anchor, then disallow the pop unless we have sufficient
3897                  * privileges; take the cheapest (non-locking) check
3898                  * first.
3899                  */
3900                 if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3901                     (stp->sd_anchorzone != crgetzoneid(crp))) {
3902                         mutex_enter(&stp->sd_lock);
3903                         /*
3904                          * Anchors only apply if there's at least one
3905                          * module on the stream (sd_pushcnt > 0).
3906                          */
3907                         if (stp->sd_pushcnt > 0 &&
3908                             stp->sd_pushcnt == stp->sd_anchor &&
3909                             stp->sd_vnode->v_type != VFIFO) {
3910                                 strendplumb(stp);
3911                                 mutex_exit(&stp->sd_lock);
3912                                 if (stp->sd_anchorzone != crgetzoneid(crp))
3913                                         return (EINVAL);
3914                                 /* Audit and report error */
3915                                 return (secpolicy_ip_config(crp, B_FALSE));
3916                         }
3917                         mutex_exit(&stp->sd_lock);
3918                 }
3919 
3920                 q = wrq->q_next;
3921                 TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3922                     "I_POP:%p from %p", q, stp);
3923                 if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3924                         error = EINVAL;
3925                 } else {
3926                         qdetach(_RD(q), 1, flag, crp, B_FALSE);
3927                         error = 0;
3928                 }
3929                 mutex_enter(&stp->sd_lock);
3930 
3931                 /*
3932                  * As a performance concern we are caching the values of
3933                  * q_minpsz and q_maxpsz of the module below the stream
3934                  * head in the stream head.
3935                  */
3936                 mutex_enter(QLOCK(wrq->q_next));
3937                 rmin = wrq->q_next->q_minpsz;
3938                 rmax = wrq->q_next->q_maxpsz;
3939                 mutex_exit(QLOCK(wrq->q_next));
3940 
3941                 /* Do this processing here as a performance concern */
3942                 if (strmsgsz != 0) {
3943                         if (rmax == INFPSZ)
3944                                 rmax = strmsgsz;
3945                         else  {
3946                                 if (vp->v_type == VFIFO)
3947                                         rmax = MIN(PIPE_BUF, rmax);
3948                                 else    rmax = MIN(strmsgsz, rmax);
3949                         }
3950                 }
3951 
3952                 mutex_enter(QLOCK(wrq));
3953                 stp->sd_qn_minpsz = rmin;
3954                 stp->sd_qn_maxpsz = rmax;
3955                 mutex_exit(QLOCK(wrq));
3956 
3957                 /* If we popped through the anchor, then reset the anchor. */
3958                 if (stp->sd_pushcnt < stp->sd_anchor) {
3959                         stp->sd_anchor = 0;
3960                         stp->sd_anchorzone = 0;
3961                 }
3962                 strendplumb(stp);
3963                 mutex_exit(&stp->sd_lock);
3964                 return (error);
3965         }
3966 
3967         case _I_MUXID2FD:
3968         {
3969                 /*
3970                  * Create a fd for a I_PLINK'ed lower stream with a given
3971                  * muxid.  With the fd, application can send down ioctls,
3972                  * like I_LIST, to the previously I_PLINK'ed stream.  Note
3973                  * that after getting the fd, the application has to do an
3974                  * I_PUNLINK on the muxid before it can do any operation
3975                  * on the lower stream.  This is required by spec1170.
3976                  *
3977                  * The fd used to do this ioctl should point to the same
3978                  * controlling device used to do the I_PLINK.  If it uses
3979                  * a different stream or an invalid muxid, I_MUXID2FD will
3980                  * fail.  The error code is set to EINVAL.
3981                  *
3982                  * The intended use of this interface is the following.
3983                  * An application I_PLINK'ed a stream and exits.  The fd
3984                  * to the lower stream is gone.  Another application
3985                  * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3986                  */
3987                 int muxid = (int)arg;
3988                 int fd;
3989                 linkinfo_t *linkp;
3990                 struct file *fp;
3991                 netstack_t *ns;
3992                 str_stack_t *ss;
3993 
3994                 /*
3995                  * Do not allow the wildcard muxid.  This ioctl is not
3996                  * intended to find arbitrary link.
3997                  */
3998                 if (muxid == 0) {
3999                         return (EINVAL);
4000                 }
4001 
4002                 ns = netstack_find_by_cred(crp);
4003                 ASSERT(ns != NULL);
4004                 ss = ns->netstack_str;
4005                 ASSERT(ss != NULL);
4006 
4007                 mutex_enter(&muxifier);
4008                 linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
4009                 if (linkp == NULL) {
4010                         mutex_exit(&muxifier);
4011                         netstack_rele(ss->ss_netstack);
4012                         return (EINVAL);
4013                 }
4014 
4015                 if ((fd = ufalloc(0)) == -1) {
4016                         mutex_exit(&muxifier);
4017                         netstack_rele(ss->ss_netstack);
4018                         return (EMFILE);
4019                 }
4020                 fp = linkp->li_fpdown;
4021                 mutex_enter(&fp->f_tlock);
4022                 fp->f_count++;
4023                 mutex_exit(&fp->f_tlock);
4024                 mutex_exit(&muxifier);
4025                 setf(fd, fp);
4026                 *rvalp = fd;
4027                 netstack_rele(ss->ss_netstack);
4028                 return (0);
4029         }
4030 
4031         case _I_INSERT:
4032         {
4033                 /*
4034                  * To insert a module to a given position in a stream.
4035                  * In the first release, only allow privileged user
4036                  * to use this ioctl. Furthermore, the insert is only allowed
4037                  * below an anchor if the zoneid is the same as the zoneid
4038                  * which created the anchor.
4039                  *
4040                  * Note that we do not plan to support this ioctl
4041                  * on pipes in the first release.  We want to learn more
4042                  * about the implications of these ioctls before extending
4043                  * their support.  And we do not think these features are
4044                  * valuable for pipes.
4045                  */
4046                 STRUCT_DECL(strmodconf, strmodinsert);
4047                 char mod_name[FMNAMESZ + 1];
4048                 fmodsw_impl_t *fp;
4049                 dev_t dummydev;
4050                 queue_t *tmp_wrq;
4051                 int pos;
4052                 boolean_t is_insert;
4053 
4054                 STRUCT_INIT(strmodinsert, flag);
4055                 if (stp->sd_flag & STRHUP)
4056                         return (ENXIO);
4057                 if (STRMATED(stp))
4058                         return (EINVAL);
4059                 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4060                         return (error);
4061                 if (stp->sd_anchor != 0 &&
4062                     stp->sd_anchorzone != crgetzoneid(crp))
4063                         return (EINVAL);
4064 
4065                 error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4066                     STRUCT_SIZE(strmodinsert), copyflag);
4067                 if (error)
4068                         return (error);
4069 
4070                 /*
4071                  * Get module name and look up in fmodsw.
4072                  */
4073                 error = (copyflag & U_TO_K ? copyinstr :
4074                     copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4075                     mod_name, FMNAMESZ + 1, NULL);
4076                 if (error)
4077                         return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4078 
4079                 if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4080                     NULL)
4081                         return (EINVAL);
4082 
4083                 if (error = strstartplumb(stp, flag, cmd)) {
4084                         fmodsw_rele(fp);
4085                         return (error);
4086                 }
4087 
4088                 /*
4089                  * Is this _I_INSERT just like an I_PUSH?  We need to know
4090                  * this because we do some optimizations if this is a
4091                  * module being pushed.
4092                  */
4093                 pos = STRUCT_FGET(strmodinsert, pos);
4094                 is_insert = (pos != 0);
4095 
4096                 /*
4097                  * Make sure pos is valid.  Even though it is not an I_PUSH,
4098                  * we impose the same limit on the number of modules in a
4099                  * stream.
4100                  */
4101                 mutex_enter(&stp->sd_lock);
4102                 if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4103                     pos > stp->sd_pushcnt) {
4104                         fmodsw_rele(fp);
4105                         strendplumb(stp);
4106                         mutex_exit(&stp->sd_lock);
4107                         return (EINVAL);
4108                 }
4109                 if (stp->sd_anchor != 0) {
4110                         /*
4111                          * Is this insert below the anchor?
4112                          * Pushcnt hasn't been increased yet hence
4113                          * we test for greater than here, and greater or
4114                          * equal after qattach.
4115                          */
4116                         if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4117                             stp->sd_anchorzone != crgetzoneid(crp)) {
4118                                 fmodsw_rele(fp);
4119                                 strendplumb(stp);
4120                                 mutex_exit(&stp->sd_lock);
4121                                 return (EPERM);
4122                         }
4123                 }
4124 
4125                 mutex_exit(&stp->sd_lock);
4126 
4127                 /*
4128                  * First find the correct position this module to
4129                  * be inserted.  We don't need to call claimstr()
4130                  * as the stream should not be changing at this point.
4131                  *
4132                  * Insert new module and call its open routine
4133                  * via qattach().  Modules don't change device
4134                  * numbers, so just ignore dummydev here.
4135                  */
4136                 for (tmp_wrq = stp->sd_wrq; pos > 0;
4137                     tmp_wrq = tmp_wrq->q_next, pos--) {
4138                         ASSERT(SAMESTR(tmp_wrq));
4139                 }
4140                 dummydev = vp->v_rdev;
4141                 if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4142                     fp, is_insert)) != 0) {
4143                         mutex_enter(&stp->sd_lock);
4144                         strendplumb(stp);
4145                         mutex_exit(&stp->sd_lock);
4146                         return (error);
4147                 }
4148 
4149                 mutex_enter(&stp->sd_lock);
4150 
4151                 /*
4152                  * As a performance concern we are caching the values of
4153                  * q_minpsz and q_maxpsz of the module below the stream
4154                  * head in the stream head.
4155                  */
4156                 if (!is_insert) {
4157                         mutex_enter(QLOCK(stp->sd_wrq->q_next));
4158                         rmin = stp->sd_wrq->q_next->q_minpsz;
4159                         rmax = stp->sd_wrq->q_next->q_maxpsz;
4160                         mutex_exit(QLOCK(stp->sd_wrq->q_next));
4161 
4162                         /* Do this processing here as a performance concern */
4163                         if (strmsgsz != 0) {
4164                                 if (rmax == INFPSZ) {
4165                                         rmax = strmsgsz;
4166                                 } else  {
4167                                         rmax = MIN(strmsgsz, rmax);
4168                                 }
4169                         }
4170 
4171                         mutex_enter(QLOCK(wrq));
4172                         stp->sd_qn_minpsz = rmin;
4173                         stp->sd_qn_maxpsz = rmax;
4174                         mutex_exit(QLOCK(wrq));
4175                 }
4176 
4177                 /*
4178                  * Need to update the anchor value if this module is
4179                  * inserted below the anchor point.
4180                  */
4181                 if (stp->sd_anchor != 0) {
4182                         pos = STRUCT_FGET(strmodinsert, pos);
4183                         if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4184                                 stp->sd_anchor++;
4185                 }
4186 
4187                 strendplumb(stp);
4188                 mutex_exit(&stp->sd_lock);
4189                 return (0);
4190         }
4191 
4192         case _I_REMOVE:
4193         {
4194                 /*
4195                  * To remove a module with a given name in a stream.  The
4196                  * caller of this ioctl needs to provide both the name and
4197                  * the position of the module to be removed.  This eliminates
4198                  * the ambiguity of removal if a module is inserted/pushed
4199                  * multiple times in a stream.  In the first release, only
4200                  * allow privileged user to use this ioctl.
4201                  * Furthermore, the remove is only allowed
4202                  * below an anchor if the zoneid is the same as the zoneid
4203                  * which created the anchor.
4204                  *
4205                  * Note that we do not plan to support this ioctl
4206                  * on pipes in the first release.  We want to learn more
4207                  * about the implications of these ioctls before extending
4208                  * their support.  And we do not think these features are
4209                  * valuable for pipes.
4210                  *
4211                  * Also note that _I_REMOVE cannot be used to remove a
4212                  * driver or the stream head.
4213                  */
4214                 STRUCT_DECL(strmodconf, strmodremove);
4215                 queue_t *q;
4216                 int pos;
4217                 char mod_name[FMNAMESZ + 1];
4218                 boolean_t is_remove;
4219 
4220                 STRUCT_INIT(strmodremove, flag);
4221                 if (stp->sd_flag & STRHUP)
4222                         return (ENXIO);
4223                 if (STRMATED(stp))
4224                         return (EINVAL);
4225                 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4226                         return (error);
4227                 if (stp->sd_anchor != 0 &&
4228                     stp->sd_anchorzone != crgetzoneid(crp))
4229                         return (EINVAL);
4230 
4231                 error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4232                     STRUCT_SIZE(strmodremove), copyflag);
4233                 if (error)
4234                         return (error);
4235 
4236                 error = (copyflag & U_TO_K ? copyinstr :
4237                     copystr)(STRUCT_FGETP(strmodremove, mod_name),
4238                     mod_name, FMNAMESZ + 1, NULL);
4239                 if (error)
4240                         return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4241 
4242                 if ((error = strstartplumb(stp, flag, cmd)) != 0)
4243                         return (error);
4244 
4245                 /*
4246                  * Match the name of given module to the name of module at
4247                  * the given position.
4248                  */
4249                 pos = STRUCT_FGET(strmodremove, pos);
4250 
4251                 is_remove = (pos != 0);
4252                 for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4253                     q = q->q_next, pos--)
4254                         ;
4255                 if (pos > 0 || !SAMESTR(q) ||
4256                     strcmp(Q2NAME(q), mod_name) != 0) {
4257                         mutex_enter(&stp->sd_lock);
4258                         strendplumb(stp);
4259                         mutex_exit(&stp->sd_lock);
4260                         return (EINVAL);
4261                 }
4262 
4263                 /*
4264                  * If the position is at or below an anchor, then the zoneid
4265                  * must match the zoneid that created the anchor.
4266                  */
4267                 if (stp->sd_anchor != 0) {
4268                         pos = STRUCT_FGET(strmodremove, pos);
4269                         if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4270                             stp->sd_anchorzone != crgetzoneid(crp)) {
4271                                 mutex_enter(&stp->sd_lock);
4272                                 strendplumb(stp);
4273                                 mutex_exit(&stp->sd_lock);
4274                                 return (EPERM);
4275                         }
4276                 }
4277 
4278 
4279                 ASSERT(!(q->q_flag & QREADR));
4280                 qdetach(_RD(q), 1, flag, crp, is_remove);
4281 
4282                 mutex_enter(&stp->sd_lock);
4283 
4284                 /*
4285                  * As a performance concern we are caching the values of
4286                  * q_minpsz and q_maxpsz of the module below the stream
4287                  * head in the stream head.
4288                  */
4289                 if (!is_remove) {
4290                         mutex_enter(QLOCK(wrq->q_next));
4291                         rmin = wrq->q_next->q_minpsz;
4292                         rmax = wrq->q_next->q_maxpsz;
4293                         mutex_exit(QLOCK(wrq->q_next));
4294 
4295                         /* Do this processing here as a performance concern */
4296                         if (strmsgsz != 0) {
4297                                 if (rmax == INFPSZ)
4298                                         rmax = strmsgsz;
4299                                 else  {
4300                                         if (vp->v_type == VFIFO)
4301                                                 rmax = MIN(PIPE_BUF, rmax);
4302                                         else    rmax = MIN(strmsgsz, rmax);
4303                                 }
4304                         }
4305 
4306                         mutex_enter(QLOCK(wrq));
4307                         stp->sd_qn_minpsz = rmin;
4308                         stp->sd_qn_maxpsz = rmax;
4309                         mutex_exit(QLOCK(wrq));
4310                 }
4311 
4312                 /*
4313                  * Need to update the anchor value if this module is removed
4314                  * at or below the anchor point.  If the removed module is at
4315                  * the anchor point, remove the anchor for this stream if
4316                  * there is no module above the anchor point.  Otherwise, if
4317                  * the removed module is below the anchor point, decrement the
4318                  * anchor point by 1.
4319                  */
4320                 if (stp->sd_anchor != 0) {
4321                         pos = STRUCT_FGET(strmodremove, pos);
4322                         if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4323                                 stp->sd_anchor = 0;
4324                         else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4325                                 stp->sd_anchor--;
4326                 }
4327 
4328                 strendplumb(stp);
4329                 mutex_exit(&stp->sd_lock);
4330                 return (0);
4331         }
4332 
4333         case I_ANCHOR:
4334                 /*
4335                  * Set the anchor position on the stream to reside at
4336                  * the top module (in other words, the top module
4337                  * cannot be popped).  Anchors with a FIFO make no
4338                  * obvious sense, so they're not allowed.
4339                  */
4340                 mutex_enter(&stp->sd_lock);
4341 
4342                 if (stp->sd_vnode->v_type == VFIFO) {
4343                         mutex_exit(&stp->sd_lock);
4344                         return (EINVAL);
4345                 }
4346                 /* Only allow the same zoneid to update the anchor */
4347                 if (stp->sd_anchor != 0 &&
4348                     stp->sd_anchorzone != crgetzoneid(crp)) {
4349                         mutex_exit(&stp->sd_lock);
4350                         return (EINVAL);
4351                 }
4352                 stp->sd_anchor = stp->sd_pushcnt;
4353                 stp->sd_anchorzone = crgetzoneid(crp);
4354                 mutex_exit(&stp->sd_lock);
4355                 return (0);
4356 
4357         case I_LOOK:
4358                 /*
4359                  * Get name of first module downstream.
4360                  * If no module, return an error.
4361                  */
4362                 claimstr(wrq);
4363                 if (_SAMESTR(wrq) && wrq->q_next->q_next != NULL) {
4364                         char *name = Q2NAME(wrq->q_next);
4365 
4366                         error = strcopyout(name, (void *)arg, strlen(name) + 1,
4367                             copyflag);
4368                         releasestr(wrq);
4369                         return (error);
4370                 }
4371                 releasestr(wrq);
4372                 return (EINVAL);
4373 
4374         case I_LINK:
4375         case I_PLINK:
4376                 /*
4377                  * Link a multiplexor.
4378                  */
4379                 return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4380 
4381         case _I_PLINK_LH:
4382                 /*
4383                  * Link a multiplexor: Call must originate from kernel.
4384                  */
4385                 if (kioctl)
4386                         return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4387 
4388                 return (EINVAL);
4389         case I_UNLINK:
4390         case I_PUNLINK:
4391                 /*
4392                  * Unlink a multiplexor.
4393                  * If arg is -1, unlink all links for which this is the
4394                  * controlling stream.  Otherwise, arg is an index number
4395                  * for a link to be removed.
4396                  */
4397         {
4398                 struct linkinfo *linkp;
4399                 int native_arg = (int)arg;
4400                 int type;
4401                 netstack_t *ns;
4402                 str_stack_t *ss;
4403 
4404                 TRACE_1(TR_FAC_STREAMS_FR,
4405                     TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4406                 if (vp->v_type == VFIFO) {
4407                         return (EINVAL);
4408                 }
4409                 if (cmd == I_UNLINK)
4410                         type = LINKNORMAL;
4411                 else    /* I_PUNLINK */
4412                         type = LINKPERSIST;
4413                 if (native_arg == 0) {
4414                         return (EINVAL);
4415                 }
4416                 ns = netstack_find_by_cred(crp);
4417                 ASSERT(ns != NULL);
4418                 ss = ns->netstack_str;
4419                 ASSERT(ss != NULL);
4420 
4421                 if (native_arg == MUXID_ALL)
4422                         error = munlinkall(stp, type, crp, rvalp, ss);
4423                 else {
4424                         mutex_enter(&muxifier);
4425                         if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4426                                 /* invalid user supplied index number */
4427                                 mutex_exit(&muxifier);
4428                                 netstack_rele(ss->ss_netstack);
4429                                 return (EINVAL);
4430                         }
4431                         /* munlink drops the muxifier lock */
4432                         error = munlink(stp, linkp, type, crp, rvalp, ss);
4433                 }
4434                 netstack_rele(ss->ss_netstack);
4435                 return (error);
4436         }
4437 
4438         case I_FLUSH:
4439                 /*
4440                  * send a flush message downstream
4441                  * flush message can indicate
4442                  * FLUSHR - flush read queue
4443                  * FLUSHW - flush write queue
4444                  * FLUSHRW - flush read/write queue
4445                  */
4446                 if (stp->sd_flag & STRHUP)
4447                         return (ENXIO);
4448                 if (arg & ~FLUSHRW)
4449                         return (EINVAL);
4450 
4451                 for (;;) {
4452                         if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4453                                 break;
4454                         }
4455                         if (error = strwaitbuf(1, BPRI_HI)) {
4456                                 return (error);
4457                         }
4458                 }
4459 
4460                 /*
4461                  * Send down an unsupported ioctl and wait for the nack
4462                  * in order to allow the M_FLUSH to propagate back
4463                  * up to the stream head.
4464                  * Replaces if (qready()) runqueues();
4465                  */
4466                 strioc.ic_cmd = -1;     /* The unsupported ioctl */
4467                 strioc.ic_timout = 0;
4468                 strioc.ic_len = 0;
4469                 strioc.ic_dp = NULL;
4470                 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4471                 *rvalp = 0;
4472                 return (0);
4473 
4474         case I_FLUSHBAND:
4475         {
4476                 struct bandinfo binfo;
4477 
4478                 error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4479                     copyflag);
4480                 if (error)
4481                         return (error);
4482                 if (stp->sd_flag & STRHUP)
4483                         return (ENXIO);
4484                 if (binfo.bi_flag & ~FLUSHRW)
4485                         return (EINVAL);
4486                 while (!(mp = allocb(2, BPRI_HI))) {
4487                         if (error = strwaitbuf(2, BPRI_HI))
4488                                 return (error);
4489                 }
4490                 mp->b_datap->db_type = M_FLUSH;
4491                 *mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4492                 *mp->b_wptr++ = binfo.bi_pri;
4493                 putnext(stp->sd_wrq, mp);
4494                 /*
4495                  * Send down an unsupported ioctl and wait for the nack
4496                  * in order to allow the M_FLUSH to propagate back
4497                  * up to the stream head.
4498                  * Replaces if (qready()) runqueues();
4499                  */
4500                 strioc.ic_cmd = -1;     /* The unsupported ioctl */
4501                 strioc.ic_timout = 0;
4502                 strioc.ic_len = 0;
4503                 strioc.ic_dp = NULL;
4504                 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4505                 *rvalp = 0;
4506                 return (0);
4507         }
4508 
4509         case I_SRDOPT:
4510                 /*
4511                  * Set read options
4512                  *
4513                  * RNORM - default stream mode
4514                  * RMSGN - message no discard
4515                  * RMSGD - message discard
4516                  * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4517                  * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4518                  * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4519                  */
4520                 if (arg & ~(RMODEMASK | RPROTMASK))
4521                         return (EINVAL);
4522 
4523                 if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4524                         return (EINVAL);
4525 
4526                 mutex_enter(&stp->sd_lock);
4527                 switch (arg & RMODEMASK) {
4528                 case RNORM:
4529                         stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4530                         break;
4531                 case RMSGD:
4532                         stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4533                             RD_MSGDIS;
4534                         break;
4535                 case RMSGN:
4536                         stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4537                             RD_MSGNODIS;
4538                         break;
4539                 }
4540 
4541                 switch (arg & RPROTMASK) {
4542                 case RPROTNORM:
4543                         stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4544                         break;
4545 
4546                 case RPROTDAT:
4547                         stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4548                             RD_PROTDAT);
4549                         break;
4550 
4551                 case RPROTDIS:
4552                         stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4553                             RD_PROTDIS);
4554                         break;
4555                 }
4556                 mutex_exit(&stp->sd_lock);
4557                 return (0);
4558 
4559         case I_GRDOPT:
4560                 /*
4561                  * Get read option and return the value
4562                  * to spot pointed to by arg
4563                  */
4564         {
4565                 int rdopt;
4566 
4567                 rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4568                     ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4569                 rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4570                     ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4571 
4572                 return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4573                     copyflag));
4574         }
4575 
4576         case I_SERROPT:
4577                 /*
4578                  * Set error options
4579                  *
4580                  * RERRNORM - persistent read errors
4581                  * RERRNONPERSIST - non-persistent read errors
4582                  * WERRNORM - persistent write errors
4583                  * WERRNONPERSIST - non-persistent write errors
4584                  */
4585                 if (arg & ~(RERRMASK | WERRMASK))
4586                         return (EINVAL);
4587 
4588                 mutex_enter(&stp->sd_lock);
4589                 switch (arg & RERRMASK) {
4590                 case RERRNORM:
4591                         stp->sd_flag &= ~STRDERRNONPERSIST;
4592                         break;
4593                 case RERRNONPERSIST:
4594                         stp->sd_flag |= STRDERRNONPERSIST;
4595                         break;
4596                 }
4597                 switch (arg & WERRMASK) {
4598                 case WERRNORM:
4599                         stp->sd_flag &= ~STWRERRNONPERSIST;
4600                         break;
4601                 case WERRNONPERSIST:
4602                         stp->sd_flag |= STWRERRNONPERSIST;
4603                         break;
4604                 }
4605                 mutex_exit(&stp->sd_lock);
4606                 return (0);
4607 
4608         case I_GERROPT:
4609                 /*
4610                  * Get error option and return the value
4611                  * to spot pointed to by arg
4612                  */
4613         {
4614                 int erropt = 0;
4615 
4616                 erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4617                     RERRNORM;
4618                 erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4619                     WERRNORM;
4620                 return (strcopyout(&erropt, (void *)arg, sizeof (int),
4621                     copyflag));
4622         }
4623 
4624         case I_SETSIG:
4625                 /*
4626                  * Register the calling proc to receive the SIGPOLL
4627                  * signal based on the events given in arg.  If
4628                  * arg is zero, remove the proc from register list.
4629                  */
4630         {
4631                 strsig_t *ssp, *pssp;
4632                 struct pid *pidp;
4633 
4634                 pssp = NULL;
4635                 pidp = curproc->p_pidp;
4636                 /*
4637                  * Hold sd_lock to prevent traversal of sd_siglist while
4638                  * it is modified.
4639                  */
4640                 mutex_enter(&stp->sd_lock);
4641                 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4642                     pssp = ssp, ssp = ssp->ss_next)
4643                         ;
4644 
4645                 if (arg) {
4646                         if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4647                             S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4648                                 mutex_exit(&stp->sd_lock);
4649                                 return (EINVAL);
4650                         }
4651                         if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4652                                 mutex_exit(&stp->sd_lock);
4653                                 return (EINVAL);
4654                         }
4655 
4656                         /*
4657                          * If proc not already registered, add it
4658                          * to list.
4659                          */
4660                         if (!ssp) {
4661                                 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4662                                 ssp->ss_pidp = pidp;
4663                                 ssp->ss_pid = pidp->pid_id;
4664                                 ssp->ss_next = NULL;
4665                                 if (pssp)
4666                                         pssp->ss_next = ssp;
4667                                 else
4668                                         stp->sd_siglist = ssp;
4669                                 mutex_enter(&pidlock);
4670                                 PID_HOLD(pidp);
4671                                 mutex_exit(&pidlock);
4672                         }
4673 
4674                         /*
4675                          * Set events.
4676                          */
4677                         ssp->ss_events = (int)arg;
4678                 } else {
4679                         /*
4680                          * Remove proc from register list.
4681                          */
4682                         if (ssp) {
4683                                 mutex_enter(&pidlock);
4684                                 PID_RELE(pidp);
4685                                 mutex_exit(&pidlock);
4686                                 if (pssp)
4687                                         pssp->ss_next = ssp->ss_next;
4688                                 else
4689                                         stp->sd_siglist = ssp->ss_next;
4690                                 kmem_free(ssp, sizeof (strsig_t));
4691                         } else {
4692                                 mutex_exit(&stp->sd_lock);
4693                                 return (EINVAL);
4694                         }
4695                 }
4696 
4697                 /*
4698                  * Recalculate OR of sig events.
4699                  */
4700                 stp->sd_sigflags = 0;
4701                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4702                         stp->sd_sigflags |= ssp->ss_events;
4703                 mutex_exit(&stp->sd_lock);
4704                 return (0);
4705         }
4706 
4707         case I_GETSIG:
4708                 /*
4709                  * Return (in arg) the current registration of events
4710                  * for which the calling proc is to be signaled.
4711                  */
4712         {
4713                 struct strsig *ssp;
4714                 struct pid  *pidp;
4715 
4716                 pidp = curproc->p_pidp;
4717                 mutex_enter(&stp->sd_lock);
4718                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4719                         if (ssp->ss_pidp == pidp) {
4720                                 error = strcopyout(&ssp->ss_events, (void *)arg,
4721                                     sizeof (int), copyflag);
4722                                 mutex_exit(&stp->sd_lock);
4723                                 return (error);
4724                         }
4725                 mutex_exit(&stp->sd_lock);
4726                 return (EINVAL);
4727         }
4728 
4729         case I_ESETSIG:
4730                 /*
4731                  * Register the ss_pid to receive the SIGPOLL
4732                  * signal based on the events is ss_events arg.  If
4733                  * ss_events is zero, remove the proc from register list.
4734                  */
4735         {
4736                 struct strsig *ssp, *pssp;
4737                 struct proc *proc;
4738                 struct pid  *pidp;
4739                 pid_t pid;
4740                 struct strsigset ss;
4741 
4742                 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4743                 if (error)
4744                         return (error);
4745 
4746                 pid = ss.ss_pid;
4747 
4748                 if (ss.ss_events != 0) {
4749                         /*
4750                          * Permissions check by sending signal 0.
4751                          * Note that when kill fails it does a set_errno
4752                          * causing the system call to fail.
4753                          */
4754                         error = kill(pid, 0);
4755                         if (error) {
4756                                 return (error);
4757                         }
4758                 }
4759                 mutex_enter(&pidlock);
4760                 if (pid == 0)
4761                         proc = curproc;
4762                 else if (pid < 0)
4763                         proc = pgfind(-pid);
4764                 else
4765                         proc = prfind(pid);
4766                 if (proc == NULL) {
4767                         mutex_exit(&pidlock);
4768                         return (ESRCH);
4769                 }
4770                 if (pid < 0)
4771                         pidp = proc->p_pgidp;
4772                 else
4773                         pidp = proc->p_pidp;
4774                 ASSERT(pidp);
4775                 /*
4776                  * Get a hold on the pid structure while referencing it.
4777                  * There is a separate PID_HOLD should it be inserted
4778                  * in the list below.
4779                  */
4780                 PID_HOLD(pidp);
4781                 mutex_exit(&pidlock);
4782 
4783                 pssp = NULL;
4784                 /*
4785                  * Hold sd_lock to prevent traversal of sd_siglist while
4786                  * it is modified.
4787                  */
4788                 mutex_enter(&stp->sd_lock);
4789                 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4790                     pssp = ssp, ssp = ssp->ss_next)
4791                         ;
4792 
4793                 if (ss.ss_events) {
4794                         if (ss.ss_events &
4795                             ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4796                             S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4797                                 mutex_exit(&stp->sd_lock);
4798                                 mutex_enter(&pidlock);
4799                                 PID_RELE(pidp);
4800                                 mutex_exit(&pidlock);
4801                                 return (EINVAL);
4802                         }
4803                         if ((ss.ss_events & S_BANDURG) &&
4804                             !(ss.ss_events & S_RDBAND)) {
4805                                 mutex_exit(&stp->sd_lock);
4806                                 mutex_enter(&pidlock);
4807                                 PID_RELE(pidp);
4808                                 mutex_exit(&pidlock);
4809                                 return (EINVAL);
4810                         }
4811 
4812                         /*
4813                          * If proc not already registered, add it
4814                          * to list.
4815                          */
4816                         if (!ssp) {
4817                                 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4818                                 ssp->ss_pidp = pidp;
4819                                 ssp->ss_pid = pid;
4820                                 ssp->ss_next = NULL;
4821                                 if (pssp)
4822                                         pssp->ss_next = ssp;
4823                                 else
4824                                         stp->sd_siglist = ssp;
4825                                 mutex_enter(&pidlock);
4826                                 PID_HOLD(pidp);
4827                                 mutex_exit(&pidlock);
4828                         }
4829 
4830                         /*
4831                          * Set events.
4832                          */
4833                         ssp->ss_events = ss.ss_events;
4834                 } else {
4835                         /*
4836                          * Remove proc from register list.
4837                          */
4838                         if (ssp) {
4839                                 mutex_enter(&pidlock);
4840                                 PID_RELE(pidp);
4841                                 mutex_exit(&pidlock);
4842                                 if (pssp)
4843                                         pssp->ss_next = ssp->ss_next;
4844                                 else
4845                                         stp->sd_siglist = ssp->ss_next;
4846                                 kmem_free(ssp, sizeof (strsig_t));
4847                         } else {
4848                                 mutex_exit(&stp->sd_lock);
4849                                 mutex_enter(&pidlock);
4850                                 PID_RELE(pidp);
4851                                 mutex_exit(&pidlock);
4852                                 return (EINVAL);
4853                         }
4854                 }
4855 
4856                 /*
4857                  * Recalculate OR of sig events.
4858                  */
4859                 stp->sd_sigflags = 0;
4860                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4861                         stp->sd_sigflags |= ssp->ss_events;
4862                 mutex_exit(&stp->sd_lock);
4863                 mutex_enter(&pidlock);
4864                 PID_RELE(pidp);
4865                 mutex_exit(&pidlock);
4866                 return (0);
4867         }
4868 
4869         case I_EGETSIG:
4870                 /*
4871                  * Return (in arg) the current registration of events
4872                  * for which the calling proc is to be signaled.
4873                  */
4874         {
4875                 struct strsig *ssp;
4876                 struct proc *proc;
4877                 pid_t pid;
4878                 struct pid  *pidp;
4879                 struct strsigset ss;
4880 
4881                 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4882                 if (error)
4883                         return (error);
4884 
4885                 pid = ss.ss_pid;
4886                 mutex_enter(&pidlock);
4887                 if (pid == 0)
4888                         proc = curproc;
4889                 else if (pid < 0)
4890                         proc = pgfind(-pid);
4891                 else
4892                         proc = prfind(pid);
4893                 if (proc == NULL) {
4894                         mutex_exit(&pidlock);
4895                         return (ESRCH);
4896                 }
4897                 if (pid < 0)
4898                         pidp = proc->p_pgidp;
4899                 else
4900                         pidp = proc->p_pidp;
4901 
4902                 /* Prevent the pidp from being reassigned */
4903                 PID_HOLD(pidp);
4904                 mutex_exit(&pidlock);
4905 
4906                 mutex_enter(&stp->sd_lock);
4907                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4908                         if (ssp->ss_pid == pid) {
4909                                 ss.ss_pid = ssp->ss_pid;
4910                                 ss.ss_events = ssp->ss_events;
4911                                 error = strcopyout(&ss, (void *)arg,
4912                                     sizeof (struct strsigset), copyflag);
4913                                 mutex_exit(&stp->sd_lock);
4914                                 mutex_enter(&pidlock);
4915                                 PID_RELE(pidp);
4916                                 mutex_exit(&pidlock);
4917                                 return (error);
4918                         }
4919                 mutex_exit(&stp->sd_lock);
4920                 mutex_enter(&pidlock);
4921                 PID_RELE(pidp);
4922                 mutex_exit(&pidlock);
4923                 return (EINVAL);
4924         }
4925 
4926         case I_PEEK:
4927         {
4928                 STRUCT_DECL(strpeek, strpeek);
4929                 size_t n;
4930                 mblk_t *fmp, *tmp_mp = NULL;
4931 
4932                 STRUCT_INIT(strpeek, flag);
4933 
4934                 error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4935                     STRUCT_SIZE(strpeek), copyflag);
4936                 if (error)
4937                         return (error);
4938 
4939                 mutex_enter(QLOCK(rdq));
4940                 /*
4941                  * Skip the invalid messages
4942                  */
4943                 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4944                         if (mp->b_datap->db_type != M_SIG)
4945                                 break;
4946 
4947                 /*
4948                  * If user has requested to peek at a high priority message
4949                  * and first message is not, return 0
4950                  */
4951                 if (mp != NULL) {
4952                         if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4953                             queclass(mp) == QNORM) {
4954                                 *rvalp = 0;
4955                                 mutex_exit(QLOCK(rdq));
4956                                 return (0);
4957                         }
4958                 } else if (stp->sd_struiordq == NULL ||
4959                     (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4960                         /*
4961                          * No mblks to look at at the streamhead and
4962                          * 1). This isn't a synch stream or
4963                          * 2). This is a synch stream but caller wants high
4964                          *      priority messages which is not supported by
4965                          *      the synch stream. (it only supports QNORM)
4966                          */
4967                         *rvalp = 0;
4968                         mutex_exit(QLOCK(rdq));
4969                         return (0);
4970                 }
4971 
4972                 fmp = mp;
4973 
4974                 if (mp && mp->b_datap->db_type == M_PASSFP) {
4975                         mutex_exit(QLOCK(rdq));
4976                         return (EBADMSG);
4977                 }
4978 
4979                 ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4980                     mp->b_datap->db_type == M_PROTO ||
4981                     mp->b_datap->db_type == M_DATA);
4982 
4983                 if (mp && mp->b_datap->db_type == M_PCPROTO) {
4984                         STRUCT_FSET(strpeek, flags, RS_HIPRI);
4985                 } else {
4986                         STRUCT_FSET(strpeek, flags, 0);
4987                 }
4988 
4989 
4990                 if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
4991                         mutex_exit(QLOCK(rdq));
4992                         return (ENOSR);
4993                 }
4994                 mutex_exit(QLOCK(rdq));
4995 
4996                 /*
4997                  * set mp = tmp_mp, so that I_PEEK processing can continue.
4998                  * tmp_mp is used to free the dup'd message.
4999                  */
5000                 mp = tmp_mp;
5001 
5002                 uio.uio_fmode = 0;
5003                 uio.uio_extflg = UIO_COPY_CACHED;
5004                 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5005                     UIO_SYSSPACE;
5006                 uio.uio_limit = 0;
5007                 /*
5008                  * First process PROTO blocks, if any.
5009                  * If user doesn't want to get ctl info by setting maxlen <= 0,
5010                  * then set len to -1/0 and skip control blocks part.
5011                  */
5012                 if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
5013                         STRUCT_FSET(strpeek, ctlbuf.len, -1);
5014                 else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
5015                         STRUCT_FSET(strpeek, ctlbuf.len, 0);
5016                 else {
5017                         int     ctl_part = 0;
5018 
5019                         iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
5020                         iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
5021                         uio.uio_iov = &iov;
5022                         uio.uio_resid = iov.iov_len;
5023                         uio.uio_loffset = 0;
5024                         uio.uio_iovcnt = 1;
5025                         while (mp && mp->b_datap->db_type != M_DATA &&
5026                             uio.uio_resid >= 0) {
5027                                 ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5028                                     mp->b_datap->db_type == M_PROTO :
5029                                     mp->b_datap->db_type == M_PCPROTO);
5030 
5031                                 if ((n = MIN(uio.uio_resid,
5032                                     mp->b_wptr - mp->b_rptr)) != 0 &&
5033                                     (error = uiomove((char *)mp->b_rptr, n,
5034                                     UIO_READ, &uio)) != 0) {
5035                                         freemsg(tmp_mp);
5036                                         return (error);
5037                                 }
5038                                 ctl_part = 1;
5039                                 mp = mp->b_cont;
5040                         }
5041                         /* No ctl message */
5042                         if (ctl_part == 0)
5043                                 STRUCT_FSET(strpeek, ctlbuf.len, -1);
5044                         else
5045                                 STRUCT_FSET(strpeek, ctlbuf.len,
5046                                     STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5047                                     uio.uio_resid);
5048                 }
5049 
5050                 /*
5051                  * Now process DATA blocks, if any.
5052                  * If user doesn't want to get data info by setting maxlen <= 0,
5053                  * then set len to -1/0 and skip data blocks part.
5054                  */
5055                 if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5056                         STRUCT_FSET(strpeek, databuf.len, -1);
5057                 else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5058                         STRUCT_FSET(strpeek, databuf.len, 0);
5059                 else {
5060                         int     data_part = 0;
5061 
5062                         iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5063                         iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5064                         uio.uio_iov = &iov;
5065                         uio.uio_resid = iov.iov_len;
5066                         uio.uio_loffset = 0;
5067                         uio.uio_iovcnt = 1;
5068                         while (mp && uio.uio_resid) {
5069                                 if (mp->b_datap->db_type == M_DATA) {
5070                                         if ((n = MIN(uio.uio_resid,
5071                                             mp->b_wptr - mp->b_rptr)) != 0 &&
5072                                             (error = uiomove((char *)mp->b_rptr,
5073                                             n, UIO_READ, &uio)) != 0) {
5074                                                 freemsg(tmp_mp);
5075                                                 return (error);
5076                                         }
5077                                         data_part = 1;
5078                                 }
5079                                 ASSERT(data_part == 0 ||
5080                                     mp->b_datap->db_type == M_DATA);
5081                                 mp = mp->b_cont;
5082                         }
5083                         /* No data message */
5084                         if (data_part == 0)
5085                                 STRUCT_FSET(strpeek, databuf.len, -1);
5086                         else
5087                                 STRUCT_FSET(strpeek, databuf.len,
5088                                     STRUCT_FGET(strpeek, databuf.maxlen) -
5089                                     uio.uio_resid);
5090                 }
5091                 freemsg(tmp_mp);
5092 
5093                 /*
5094                  * It is a synch stream and user wants to get
5095                  * data (maxlen > 0).
5096                  * uio setup is done by the codes that process DATA
5097                  * blocks above.
5098                  */
5099                 if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5100                         infod_t infod;
5101 
5102                         infod.d_cmd = INFOD_COPYOUT;
5103                         infod.d_res = 0;
5104                         infod.d_uiop = &uio;
5105                         error = infonext(rdq, &infod);
5106                         if (error == EINVAL || error == EBUSY)
5107                                 error = 0;
5108                         if (error)
5109                                 return (error);
5110                         STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5111                             databuf.maxlen) - uio.uio_resid);
5112                         if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5113                                 /*
5114                                  * No data found by the infonext().
5115                                  */
5116                                 STRUCT_FSET(strpeek, databuf.len, -1);
5117                         }
5118                 }
5119                 error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5120                     STRUCT_SIZE(strpeek), copyflag);
5121                 if (error) {
5122                         return (error);
5123                 }
5124                 /*
5125                  * If there is no message retrieved, set return code to 0
5126                  * otherwise, set it to 1.
5127                  */
5128                 if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5129                     STRUCT_FGET(strpeek, databuf.len) == -1)
5130                         *rvalp = 0;
5131                 else
5132                         *rvalp = 1;
5133                 return (0);
5134         }
5135 
5136         case I_FDINSERT:
5137         {
5138                 STRUCT_DECL(strfdinsert, strfdinsert);
5139                 struct file *resftp;
5140                 struct stdata *resstp;
5141                 t_uscalar_t     ival;
5142                 ssize_t msgsize;
5143                 struct strbuf mctl;
5144 
5145                 STRUCT_INIT(strfdinsert, flag);
5146                 if (stp->sd_flag & STRHUP)
5147                         return (ENXIO);
5148                 /*
5149                  * STRDERR, STWRERR and STPLEX tested above.
5150                  */
5151                 error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5152                     STRUCT_SIZE(strfdinsert), copyflag);
5153                 if (error)
5154                         return (error);
5155 
5156                 if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5157                     (STRUCT_FGET(strfdinsert, offset) %
5158                     sizeof (t_uscalar_t)) != 0)
5159                         return (EINVAL);
5160                 if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5161                         if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5162                                 releasef(STRUCT_FGET(strfdinsert, fildes));
5163                                 return (EINVAL);
5164                         }
5165                 } else
5166                         return (EINVAL);
5167 
5168                 mutex_enter(&resstp->sd_lock);
5169                 if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5170                         error = strgeterr(resstp,
5171                             STRDERR|STWRERR|STRHUP|STPLEX, 0);
5172                         if (error != 0) {
5173                                 mutex_exit(&resstp->sd_lock);
5174                                 releasef(STRUCT_FGET(strfdinsert, fildes));
5175                                 return (error);
5176                         }
5177                 }
5178                 mutex_exit(&resstp->sd_lock);
5179 
5180 #ifdef  _ILP32
5181                 {
5182                         queue_t *q;
5183                         queue_t *mate = NULL;
5184 
5185                         /* get read queue of stream terminus */
5186                         claimstr(resstp->sd_wrq);
5187                         for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5188                             q = q->q_next)
5189                                 if (!STRMATED(resstp) && STREAM(q) != resstp &&
5190                                     mate == NULL) {
5191                                         ASSERT(q->q_qinfo->qi_srvp);
5192                                         ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5193                                         claimstr(q);
5194                                         mate = q;
5195                                 }
5196                         q = _RD(q);
5197                         if (mate)
5198                                 releasestr(mate);
5199                         releasestr(resstp->sd_wrq);
5200                         ival = (t_uscalar_t)q;
5201                 }
5202 #else
5203                 ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5204 #endif  /* _ILP32 */
5205 
5206                 if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5207                     STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5208                         releasef(STRUCT_FGET(strfdinsert, fildes));
5209                         return (EINVAL);
5210                 }
5211 
5212                 /*
5213                  * Check for legal flag value.
5214                  */
5215                 if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5216                         releasef(STRUCT_FGET(strfdinsert, fildes));
5217                         return (EINVAL);
5218                 }
5219 
5220                 /* get these values from those cached in the stream head */
5221                 mutex_enter(QLOCK(stp->sd_wrq));
5222                 rmin = stp->sd_qn_minpsz;
5223                 rmax = stp->sd_qn_maxpsz;
5224                 mutex_exit(QLOCK(stp->sd_wrq));
5225 
5226                 /*
5227                  * Make sure ctl and data sizes together fall within
5228                  * the limits of the max and min receive packet sizes
5229                  * and do not exceed system limit.  A negative data
5230                  * length means that no data part is to be sent.
5231                  */
5232                 ASSERT((rmax >= 0) || (rmax == INFPSZ));
5233                 if (rmax == 0) {
5234                         releasef(STRUCT_FGET(strfdinsert, fildes));
5235                         return (ERANGE);
5236                 }
5237                 if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5238                         msgsize = 0;
5239                 if ((msgsize < rmin) ||
5240                     ((msgsize > rmax) && (rmax != INFPSZ)) ||
5241                     (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5242                         releasef(STRUCT_FGET(strfdinsert, fildes));
5243                         return (ERANGE);
5244                 }
5245 
5246                 mutex_enter(&stp->sd_lock);
5247                 while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5248                     !canputnext(stp->sd_wrq)) {
5249                         if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5250                             flag, -1, &done)) != 0 || done) {
5251                                 mutex_exit(&stp->sd_lock);
5252                                 releasef(STRUCT_FGET(strfdinsert, fildes));
5253                                 return (error);
5254                         }
5255                         if ((error = i_straccess(stp, access)) != 0) {
5256                                 mutex_exit(&stp->sd_lock);
5257                                 releasef(
5258                                     STRUCT_FGET(strfdinsert, fildes));
5259                                 return (error);
5260                         }
5261                 }
5262                 mutex_exit(&stp->sd_lock);
5263 
5264                 /*
5265                  * Copy strfdinsert.ctlbuf into native form of
5266                  * ctlbuf to pass down into strmakemsg().
5267                  */
5268                 mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5269                 mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5270                 mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5271 
5272                 iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5273                 iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5274                 uio.uio_iov = &iov;
5275                 uio.uio_iovcnt = 1;
5276                 uio.uio_loffset = 0;
5277                 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5278                     UIO_SYSSPACE;
5279                 uio.uio_fmode = 0;
5280                 uio.uio_extflg = UIO_COPY_CACHED;
5281                 uio.uio_resid = iov.iov_len;
5282                 if ((error = strmakemsg(&mctl,
5283                     &msgsize, &uio, stp,
5284                     STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5285                         STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5286                         releasef(STRUCT_FGET(strfdinsert, fildes));
5287                         return (error);
5288                 }
5289 
5290                 STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5291 
5292                 /*
5293                  * Place the possibly reencoded queue pointer 'offset' bytes
5294                  * from the start of the control portion of the message.
5295                  */
5296                 *((t_uscalar_t *)(mp->b_rptr +
5297                     STRUCT_FGET(strfdinsert, offset))) = ival;
5298 
5299                 /*
5300                  * Put message downstream.
5301                  */
5302                 stream_willservice(stp);
5303                 putnext(stp->sd_wrq, mp);
5304                 stream_runservice(stp);
5305                 releasef(STRUCT_FGET(strfdinsert, fildes));
5306                 return (error);
5307         }
5308 
5309         case I_SENDFD:
5310         {
5311                 struct file *fp;
5312 
5313                 if ((fp = getf((int)arg)) == NULL)
5314                         return (EBADF);
5315                 error = do_sendfp(stp, fp, crp);
5316                 if (auditing) {
5317                         audit_fdsend((int)arg, fp, error);
5318                 }
5319                 releasef((int)arg);
5320                 return (error);
5321         }
5322 
5323         case I_RECVFD:
5324         case I_E_RECVFD:
5325         {
5326                 struct k_strrecvfd *srf;
5327                 int i, fd;
5328 
5329                 mutex_enter(&stp->sd_lock);
5330                 while (!(mp = getq(rdq))) {
5331                         if (stp->sd_flag & (STRHUP|STREOF)) {
5332                                 mutex_exit(&stp->sd_lock);
5333                                 return (ENXIO);
5334                         }
5335                         if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5336                             flag, -1, &done)) != 0 || done) {
5337                                 mutex_exit(&stp->sd_lock);
5338                                 return (error);
5339                         }
5340                         if ((error = i_straccess(stp, access)) != 0) {
5341                                 mutex_exit(&stp->sd_lock);
5342                                 return (error);
5343                         }
5344                 }
5345                 if (mp->b_datap->db_type != M_PASSFP) {
5346                         putback(stp, rdq, mp, mp->b_band);
5347                         mutex_exit(&stp->sd_lock);
5348                         return (EBADMSG);
5349                 }
5350                 mutex_exit(&stp->sd_lock);
5351 
5352                 srf = (struct k_strrecvfd *)mp->b_rptr;
5353                 if ((fd = ufalloc(0)) == -1) {
5354                         mutex_enter(&stp->sd_lock);
5355                         putback(stp, rdq, mp, mp->b_band);
5356                         mutex_exit(&stp->sd_lock);
5357                         return (EMFILE);
5358                 }
5359                 if (cmd == I_RECVFD) {
5360                         struct o_strrecvfd      ostrfd;
5361 
5362                         /* check to see if uid/gid values are too large. */
5363 
5364                         if (srf->uid > (o_uid_t)USHRT_MAX ||
5365                             srf->gid > (o_gid_t)USHRT_MAX) {
5366                                 mutex_enter(&stp->sd_lock);
5367                                 putback(stp, rdq, mp, mp->b_band);
5368                                 mutex_exit(&stp->sd_lock);
5369                                 setf(fd, NULL); /* release fd entry */
5370                                 return (EOVERFLOW);
5371                         }
5372 
5373                         ostrfd.fd = fd;
5374                         ostrfd.uid = (o_uid_t)srf->uid;
5375                         ostrfd.gid = (o_gid_t)srf->gid;
5376 
5377                         /* Null the filler bits */
5378                         for (i = 0; i < 8; i++)
5379                                 ostrfd.fill[i] = 0;
5380 
5381                         error = strcopyout(&ostrfd, (void *)arg,
5382                             sizeof (struct o_strrecvfd), copyflag);
5383                 } else {                /* I_E_RECVFD */
5384                         struct strrecvfd        strfd;
5385 
5386                         strfd.fd = fd;
5387                         strfd.uid = srf->uid;
5388                         strfd.gid = srf->gid;
5389 
5390                         /* null the filler bits */
5391                         for (i = 0; i < 8; i++)
5392                                 strfd.fill[i] = 0;
5393 
5394                         error = strcopyout(&strfd, (void *)arg,
5395                             sizeof (struct strrecvfd), copyflag);
5396                 }
5397 
5398                 if (error) {
5399                         setf(fd, NULL); /* release fd entry */
5400                         mutex_enter(&stp->sd_lock);
5401                         putback(stp, rdq, mp, mp->b_band);
5402                         mutex_exit(&stp->sd_lock);
5403                         return (error);
5404                 }
5405                 if (auditing) {
5406                         audit_fdrecv(fd, srf->fp);
5407                 }
5408 
5409                 /*
5410                  * Always increment f_count since the freemsg() below will
5411                  * always call free_passfp() which performs a closef().
5412                  */
5413                 mutex_enter(&srf->fp->f_tlock);
5414                 srf->fp->f_count++;
5415                 mutex_exit(&srf->fp->f_tlock);
5416                 setf(fd, srf->fp);
5417                 freemsg(mp);
5418                 return (0);
5419         }
5420 
5421         case I_SWROPT:
5422                 /*
5423                  * Set/clear the write options. arg is a bit
5424                  * mask with any of the following bits set...
5425                  *      SNDZERO - send zero length message
5426                  *      SNDPIPE - send sigpipe to process if
5427                  *              sd_werror is set and process is
5428                  *              doing a write or putmsg.
5429                  * The new stream head write options should reflect
5430                  * what is in arg.
5431                  */
5432                 if (arg & ~(SNDZERO|SNDPIPE))
5433                         return (EINVAL);
5434 
5435                 mutex_enter(&stp->sd_lock);
5436                 stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5437                 if (arg & SNDZERO)
5438                         stp->sd_wput_opt |= SW_SNDZERO;
5439                 if (arg & SNDPIPE)
5440                         stp->sd_wput_opt |= SW_SIGPIPE;
5441                 mutex_exit(&stp->sd_lock);
5442                 return (0);
5443 
5444         case I_GWROPT:
5445         {
5446                 int wropt = 0;
5447 
5448                 if (stp->sd_wput_opt & SW_SNDZERO)
5449                         wropt |= SNDZERO;
5450                 if (stp->sd_wput_opt & SW_SIGPIPE)
5451                         wropt |= SNDPIPE;
5452                 return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5453                     copyflag));
5454         }
5455 
5456         case I_LIST:
5457                 /*
5458                  * Returns all the modules found on this stream,
5459                  * upto the driver. If argument is NULL, return the
5460                  * number of modules (including driver). If argument
5461                  * is not NULL, copy the names into the structure
5462                  * provided.
5463                  */
5464 
5465         {
5466                 queue_t *q;
5467                 char *qname;
5468                 int i, nmods;
5469                 struct str_mlist *mlist;
5470                 STRUCT_DECL(str_list, strlist);
5471 
5472                 if (arg == NULL) { /* Return number of modules plus driver */
5473                         if (stp->sd_vnode->v_type == VFIFO)
5474                                 *rvalp = stp->sd_pushcnt;
5475                         else
5476                                 *rvalp = stp->sd_pushcnt + 1;
5477                         return (0);
5478                 }
5479 
5480                 STRUCT_INIT(strlist, flag);
5481 
5482                 error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5483                     STRUCT_SIZE(strlist), copyflag);
5484                 if (error != 0)
5485                         return (error);
5486 
5487                 mlist = STRUCT_FGETP(strlist, sl_modlist);
5488                 nmods = STRUCT_FGET(strlist, sl_nmods);
5489                 if (nmods <= 0)
5490                         return (EINVAL);
5491 
5492                 claimstr(stp->sd_wrq);
5493                 q = stp->sd_wrq;
5494                 for (i = 0; i < nmods && _SAMESTR(q); i++, q = q->q_next) {
5495                         qname = Q2NAME(q->q_next);
5496                         error = strcopyout(qname, &mlist[i], strlen(qname) + 1,
5497                             copyflag);
5498                         if (error != 0) {
5499                                 releasestr(stp->sd_wrq);
5500                                 return (error);
5501                         }
5502                 }
5503                 releasestr(stp->sd_wrq);
5504                 return (strcopyout(&i, (void *)arg, sizeof (int), copyflag));
5505         }
5506 
5507         case I_CKBAND:
5508         {
5509                 queue_t *q;
5510                 qband_t *qbp;
5511 
5512                 if ((arg < 0) || (arg >= NBAND))
5513                         return (EINVAL);
5514                 q = _RD(stp->sd_wrq);
5515                 mutex_enter(QLOCK(q));
5516                 if (arg > (int)q->q_nband) {
5517                         *rvalp = 0;
5518                 } else {
5519                         if (arg == 0) {
5520                                 if (q->q_first)
5521                                         *rvalp = 1;
5522                                 else
5523                                         *rvalp = 0;
5524                         } else {
5525                                 qbp = q->q_bandp;
5526                                 while (--arg > 0)
5527                                         qbp = qbp->qb_next;
5528                                 if (qbp->qb_first)
5529                                         *rvalp = 1;
5530                                 else
5531                                         *rvalp = 0;
5532                         }
5533                 }
5534                 mutex_exit(QLOCK(q));
5535                 return (0);
5536         }
5537 
5538         case I_GETBAND:
5539         {
5540                 int intpri;
5541                 queue_t *q;
5542 
5543                 q = _RD(stp->sd_wrq);
5544                 mutex_enter(QLOCK(q));
5545                 mp = q->q_first;
5546                 if (!mp) {
5547                         mutex_exit(QLOCK(q));
5548                         return (ENODATA);
5549                 }
5550                 intpri = (int)mp->b_band;
5551                 error = strcopyout(&intpri, (void *)arg, sizeof (int),
5552                     copyflag);
5553                 mutex_exit(QLOCK(q));
5554                 return (error);
5555         }
5556 
5557         case I_ATMARK:
5558         {
5559                 queue_t *q;
5560 
5561                 if (arg & ~(ANYMARK|LASTMARK))
5562                         return (EINVAL);
5563                 q = _RD(stp->sd_wrq);
5564                 mutex_enter(&stp->sd_lock);
5565                 if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5566                         *rvalp = 1;
5567                 } else {
5568                         mutex_enter(QLOCK(q));
5569                         mp = q->q_first;
5570 
5571                         if (mp == NULL)
5572                                 *rvalp = 0;
5573                         else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5574                                 *rvalp = 1;
5575                         else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5576                                 *rvalp = 1;
5577                         else
5578                                 *rvalp = 0;
5579                         mutex_exit(QLOCK(q));
5580                 }
5581                 mutex_exit(&stp->sd_lock);
5582                 return (0);
5583         }
5584 
5585         case I_CANPUT:
5586         {
5587                 char band;
5588 
5589                 if ((arg < 0) || (arg >= NBAND))
5590                         return (EINVAL);
5591                 band = (char)arg;
5592                 *rvalp = bcanputnext(stp->sd_wrq, band);
5593                 return (0);
5594         }
5595 
5596         case I_SETCLTIME:
5597         {
5598                 int closetime;
5599 
5600                 error = strcopyin((void *)arg, &closetime, sizeof (int),
5601                     copyflag);
5602                 if (error)
5603                         return (error);
5604                 if (closetime < 0)
5605                         return (EINVAL);
5606 
5607                 stp->sd_closetime = closetime;
5608                 return (0);
5609         }
5610 
5611         case I_GETCLTIME:
5612         {
5613                 int closetime;
5614 
5615                 closetime = stp->sd_closetime;
5616                 return (strcopyout(&closetime, (void *)arg, sizeof (int),
5617                     copyflag));
5618         }
5619 
5620         case TIOCGSID:
5621         {
5622                 pid_t sid;
5623 
5624                 mutex_enter(&stp->sd_lock);
5625                 if (stp->sd_sidp == NULL) {
5626                         mutex_exit(&stp->sd_lock);
5627                         return (ENOTTY);
5628                 }
5629                 sid = stp->sd_sidp->pid_id;
5630                 mutex_exit(&stp->sd_lock);
5631                 return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5632                     copyflag));
5633         }
5634 
5635         case TIOCSPGRP:
5636         {
5637                 pid_t pgrp;
5638                 proc_t *q;
5639                 pid_t   sid, fg_pgid, bg_pgid;
5640 
5641                 if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5642                     copyflag))
5643                         return (error);
5644                 mutex_enter(&stp->sd_lock);
5645                 mutex_enter(&pidlock);
5646                 if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5647                         mutex_exit(&pidlock);
5648                         mutex_exit(&stp->sd_lock);
5649                         return (ENOTTY);
5650                 }
5651                 if (pgrp == stp->sd_pgidp->pid_id) {
5652                         mutex_exit(&pidlock);
5653                         mutex_exit(&stp->sd_lock);
5654                         return (0);
5655                 }
5656                 if (pgrp <= 0 || pgrp >= maxpid) {
5657                         mutex_exit(&pidlock);
5658                         mutex_exit(&stp->sd_lock);
5659                         return (EINVAL);
5660                 }
5661                 if ((q = pgfind(pgrp)) == NULL ||
5662                     q->p_sessp != ttoproc(curthread)->p_sessp) {
5663                         mutex_exit(&pidlock);
5664                         mutex_exit(&stp->sd_lock);
5665                         return (EPERM);
5666                 }
5667                 sid = stp->sd_sidp->pid_id;
5668                 fg_pgid = q->p_pgrp;
5669                 bg_pgid = stp->sd_pgidp->pid_id;
5670                 CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5671                 PID_RELE(stp->sd_pgidp);
5672                 ctty_clear_sighuped();
5673                 stp->sd_pgidp = q->p_pgidp;
5674                 PID_HOLD(stp->sd_pgidp);
5675                 mutex_exit(&pidlock);
5676                 mutex_exit(&stp->sd_lock);
5677                 return (0);
5678         }
5679 
5680         case TIOCGPGRP:
5681         {
5682                 pid_t pgrp;
5683 
5684                 mutex_enter(&stp->sd_lock);
5685                 if (stp->sd_sidp == NULL) {
5686                         mutex_exit(&stp->sd_lock);
5687                         return (ENOTTY);
5688                 }
5689                 pgrp = stp->sd_pgidp->pid_id;
5690                 mutex_exit(&stp->sd_lock);
5691                 return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5692                     copyflag));
5693         }
5694 
5695         case TIOCSCTTY:
5696         {
5697                 return (strctty(stp));
5698         }
5699 
5700         case TIOCNOTTY:
5701         {
5702                 /* freectty() always assumes curproc. */
5703                 if (freectty(B_FALSE) != 0)
5704                         return (0);
5705                 return (ENOTTY);
5706         }
5707 
5708         case FIONBIO:
5709         case FIOASYNC:
5710                 return (0);     /* handled by the upper layer */
5711         }
5712 }
5713 
5714 /*
5715  * Custom free routine used for M_PASSFP messages.
5716  */
5717 static void
5718 free_passfp(struct k_strrecvfd *srf)
5719 {
5720         (void) closef(srf->fp);
5721         kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5722 }
5723 
5724 /* ARGSUSED */
5725 int
5726 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5727 {
5728         queue_t *qp, *nextqp;
5729         struct k_strrecvfd *srf;
5730         mblk_t *mp;
5731         frtn_t *frtnp;
5732         size_t bufsize;
5733         queue_t *mate = NULL;
5734         syncq_t *sq = NULL;
5735         int retval = 0;
5736 
5737         if (stp->sd_flag & STRHUP)
5738                 return (ENXIO);
5739 
5740         claimstr(stp->sd_wrq);
5741 
5742         /* Fastpath, we have a pipe, and we are already mated, use it. */
5743         if (STRMATED(stp)) {
5744                 qp = _RD(stp->sd_mate->sd_wrq);
5745                 claimstr(qp);
5746                 mate = qp;
5747         } else { /* Not already mated. */
5748 
5749                 /*
5750                  * Walk the stream to the end of this one.
5751                  * assumes that the claimstr() will prevent
5752                  * plumbing between the stream head and the
5753                  * driver from changing
5754                  */
5755                 qp = stp->sd_wrq;
5756 
5757                 /*
5758                  * Loop until we reach the end of this stream.
5759                  * On completion, qp points to the write queue
5760                  * at the end of the stream, or the read queue
5761                  * at the stream head if this is a fifo.
5762                  */
5763                 while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5764                         ;
5765 
5766                 /*
5767                  * Just in case we get a q_next which is NULL, but
5768                  * not at the end of the stream.  This is actually
5769                  * broken, so we set an assert to catch it in
5770                  * debug, and set an error and return if not debug.
5771                  */
5772                 ASSERT(qp);
5773                 if (qp == NULL) {
5774                         releasestr(stp->sd_wrq);
5775                         return (EINVAL);
5776                 }
5777 
5778                 /*
5779                  * Enter the syncq for the driver, so (hopefully)
5780                  * the queue values will not change on us.
5781                  * XXXX - This will only prevent the race IFF only
5782                  *   the write side modifies the q_next member, and
5783                  *   the put procedure is protected by at least
5784                  *   MT_PERQ.
5785                  */
5786                 if ((sq = qp->q_syncq) != NULL)
5787                         entersq(sq, SQ_PUT);
5788 
5789                 /* Now get the q_next value from this qp. */
5790                 nextqp = qp->q_next;
5791 
5792                 /*
5793                  * If nextqp exists and the other stream is different
5794                  * from this one claim the stream, set the mate, and
5795                  * get the read queue at the stream head of the other
5796                  * stream.  Assumes that nextqp was at least valid when
5797                  * we got it.  Hopefully the entersq of the driver
5798                  * will prevent it from changing on us.
5799                  */
5800                 if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5801                         ASSERT(qp->q_qinfo->qi_srvp);
5802                         ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5803                         ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5804                         claimstr(nextqp);
5805 
5806                         /* Make sure we still have a q_next */
5807                         if (nextqp != qp->q_next) {
5808                                 releasestr(stp->sd_wrq);
5809                                 releasestr(nextqp);
5810                                 return (EINVAL);
5811                         }
5812 
5813                         qp = _RD(STREAM(nextqp)->sd_wrq);
5814                         mate = qp;
5815                 }
5816                 /* If we entered the synq above, leave it. */
5817                 if (sq != NULL)
5818                         leavesq(sq, SQ_PUT);
5819         } /*  STRMATED(STP)  */
5820 
5821         /* XXX prevents substitution of the ops vector */
5822         if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5823                 retval = EINVAL;
5824                 goto out;
5825         }
5826 
5827         if (qp->q_flag & QFULL) {
5828                 retval = EAGAIN;
5829                 goto out;
5830         }
5831 
5832         /*
5833          * Since M_PASSFP messages include a file descriptor, we use
5834          * esballoc() and specify a custom free routine (free_passfp()) that
5835          * will close the descriptor as part of freeing the message.  For
5836          * convenience, we stash the frtn_t right after the data block.
5837          */
5838         bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5839         srf = kmem_alloc(bufsize, KM_NOSLEEP);
5840         if (srf == NULL) {
5841                 retval = EAGAIN;
5842                 goto out;
5843         }
5844 
5845         frtnp = (frtn_t *)(srf + 1);
5846         frtnp->free_arg = (caddr_t)srf;
5847         frtnp->free_func = free_passfp;
5848 
5849         mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5850         if (mp == NULL) {
5851                 kmem_free(srf, bufsize);
5852                 retval = EAGAIN;
5853                 goto out;
5854         }
5855         mp->b_wptr += sizeof (struct k_strrecvfd);
5856         mp->b_datap->db_type = M_PASSFP;
5857 
5858         srf->fp = fp;
5859         srf->uid = crgetuid(curthread->t_cred);
5860         srf->gid = crgetgid(curthread->t_cred);
5861         mutex_enter(&fp->f_tlock);
5862         fp->f_count++;
5863         mutex_exit(&fp->f_tlock);
5864 
5865         put(qp, mp);
5866 out:
5867         releasestr(stp->sd_wrq);
5868         if (mate)
5869                 releasestr(mate);
5870         return (retval);
5871 }
5872 
5873 /*
5874  * Send an ioctl message downstream and wait for acknowledgement.
5875  * flags may be set to either U_TO_K or K_TO_K and a combination
5876  * of STR_NOERROR or STR_NOSIG
5877  * STR_NOSIG: Signals are essentially ignored or held and have
5878  *      no effect for the duration of the call.
5879  * STR_NOERROR: Ignores stream head read, write and hup errors.
5880  *      Additionally, if an existing ioctl times out, it is assumed
5881  *      lost and and this ioctl will continue as if the previous ioctl had
5882  *      finished.  ETIME may be returned if this ioctl times out (i.e.
5883  *      ic_timout is not INFTIM).  Non-stream head errors may be returned if
5884  *      the ioc_error indicates that the driver/module had problems,
5885  *      an EFAULT was found when accessing user data, a lack of
5886  *      resources, etc.
5887  */
5888 int
5889 strdoioctl(
5890         struct stdata *stp,
5891         struct strioctl *strioc,
5892         int fflags,             /* file flags with model info */
5893         int flag,
5894         cred_t *crp,
5895         int *rvalp)
5896 {
5897         mblk_t *bp;
5898         struct iocblk *iocbp;
5899         struct copyreq *reqp;
5900         struct copyresp *resp;
5901         int id;
5902         int transparent = 0;
5903         int error = 0;
5904         int len = 0;
5905         caddr_t taddr;
5906         int copyflag = (flag & (U_TO_K | K_TO_K));
5907         int sigflag = (flag & STR_NOSIG);
5908         int errs;
5909         uint_t waitflags;
5910         boolean_t set_iocwaitne = B_FALSE;
5911 
5912         ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5913         ASSERT((fflags & FMODELS) != 0);
5914 
5915         TRACE_2(TR_FAC_STREAMS_FR,
5916             TR_STRDOIOCTL,
5917             "strdoioctl:stp %p strioc %p", stp, strioc);
5918         if (strioc->ic_len == TRANSPARENT) { /* send arg in M_DATA block */
5919                 transparent = 1;
5920                 strioc->ic_len = sizeof (intptr_t);
5921         }
5922 
5923         if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5924                 return (EINVAL);
5925 
5926         if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5927             crp, curproc->p_pid)) == NULL)
5928                         return (error);
5929 
5930         bzero(bp->b_wptr, sizeof (union ioctypes));
5931 
5932         iocbp = (struct iocblk *)bp->b_wptr;
5933         iocbp->ioc_count = strioc->ic_len;
5934         iocbp->ioc_cmd = strioc->ic_cmd;
5935         iocbp->ioc_flag = (fflags & FMODELS);
5936 
5937         crhold(crp);
5938         iocbp->ioc_cr = crp;
5939         DB_TYPE(bp) = M_IOCTL;
5940         bp->b_wptr += sizeof (struct iocblk);
5941 
5942         if (flag & STR_NOERROR)
5943                 errs = STPLEX;
5944         else
5945                 errs = STRHUP|STRDERR|STWRERR|STPLEX;
5946 
5947         /*
5948          * If there is data to copy into ioctl block, do so.
5949          */
5950         if (iocbp->ioc_count > 0) {
5951                 if (transparent)
5952                         /*
5953                          * Note: STR_NOERROR does not have an effect
5954                          * in putiocd()
5955                          */
5956                         id = K_TO_K | sigflag;
5957                 else
5958                         id = flag;
5959                 if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5960                         freemsg(bp);
5961                         crfree(crp);
5962                         return (error);
5963                 }
5964 
5965                 /*
5966                  * We could have slept copying in user pages.
5967                  * Recheck the stream head state (the other end
5968                  * of a pipe could have gone away).
5969                  */
5970                 if (stp->sd_flag & errs) {
5971                         mutex_enter(&stp->sd_lock);
5972                         error = strgeterr(stp, errs, 0);
5973                         mutex_exit(&stp->sd_lock);
5974                         if (error != 0) {
5975                                 freemsg(bp);
5976                                 crfree(crp);
5977                                 return (error);
5978                         }
5979                 }
5980         }
5981         if (transparent)
5982                 iocbp->ioc_count = TRANSPARENT;
5983 
5984         /*
5985          * Block for up to STRTIMOUT milliseconds if there is an outstanding
5986          * ioctl for this stream already running.  All processes
5987          * sleeping here will be awakened as a result of an ACK
5988          * or NAK being received for the outstanding ioctl, or
5989          * as a result of the timer expiring on the outstanding
5990          * ioctl (a failure), or as a result of any waiting
5991          * process's timer expiring (also a failure).
5992          */
5993 
5994         error = 0;
5995         mutex_enter(&stp->sd_lock);
5996         while ((stp->sd_flag & IOCWAIT) ||
5997             (!set_iocwaitne && (stp->sd_flag & IOCWAITNE))) {
5998                 clock_t cv_rval;
5999 
6000                 TRACE_0(TR_FAC_STREAMS_FR,
6001                     TR_STRDOIOCTL_WAIT,
6002                     "strdoioctl sleeps - IOCWAIT");
6003                 cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
6004                     STRTIMOUT, sigflag);
6005                 if (cv_rval <= 0) {
6006                         if (cv_rval == 0) {
6007                                 error = EINTR;
6008                         } else {
6009                                 if (flag & STR_NOERROR) {
6010                                         /*
6011                                          * Terminating current ioctl in
6012                                          * progress -- assume it got lost and
6013                                          * wake up the other thread so that the
6014                                          * operation completes.
6015                                          */
6016                                         if (!(stp->sd_flag & IOCWAITNE)) {
6017                                                 set_iocwaitne = B_TRUE;
6018                                                 stp->sd_flag |= IOCWAITNE;
6019                                                 cv_broadcast(&stp->sd_monitor);
6020                                         }
6021                                         /*
6022                                          * Otherwise, there's a running
6023                                          * STR_NOERROR -- we have no choice
6024                                          * here but to wait forever (or until
6025                                          * interrupted).
6026                                          */
6027                                 } else {
6028                                         /*
6029                                          * pending ioctl has caused
6030                                          * us to time out
6031                                          */
6032                                         error = ETIME;
6033                                 }
6034                         }
6035                 } else if ((stp->sd_flag & errs)) {
6036                         error = strgeterr(stp, errs, 0);
6037                 }
6038                 if (error) {
6039                         mutex_exit(&stp->sd_lock);
6040                         freemsg(bp);
6041                         crfree(crp);
6042                         return (error);
6043                 }
6044         }
6045 
6046         /*
6047          * Have control of ioctl mechanism.
6048          * Send down ioctl packet and wait for response.
6049          */
6050         if (stp->sd_iocblk != (mblk_t *)-1) {
6051                 freemsg(stp->sd_iocblk);
6052         }
6053         stp->sd_iocblk = NULL;
6054 
6055         /*
6056          * If this is marked with 'noerror' (internal; mostly
6057          * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6058          * in here by setting IOCWAITNE.
6059          */
6060         waitflags = IOCWAIT;
6061         if (flag & STR_NOERROR)
6062                 waitflags |= IOCWAITNE;
6063 
6064         stp->sd_flag |= waitflags;
6065 
6066         /*
6067          * Assign sequence number.
6068          */
6069         iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6070 
6071         mutex_exit(&stp->sd_lock);
6072 
6073         TRACE_1(TR_FAC_STREAMS_FR,
6074             TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6075         stream_willservice(stp);
6076         putnext(stp->sd_wrq, bp);
6077         stream_runservice(stp);
6078 
6079         /*
6080          * Timed wait for acknowledgment.  The wait time is limited by the
6081          * timeout value, which must be a positive integer (number of
6082          * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6083          * milliseconds), or -1 (wait forever).  This will be awakened
6084          * either by an ACK/NAK message arriving, the timer expiring, or
6085          * the timer expiring on another ioctl waiting for control of the
6086          * mechanism.
6087          */
6088 waitioc:
6089         mutex_enter(&stp->sd_lock);
6090 
6091 
6092         /*
6093          * If the reply has already arrived, don't sleep.  If awakened from
6094          * the sleep, fail only if the reply has not arrived by then.
6095          * Otherwise, process the reply.
6096          */
6097         while (!stp->sd_iocblk) {
6098                 clock_t cv_rval;
6099 
6100                 if (stp->sd_flag & errs) {
6101                         error = strgeterr(stp, errs, 0);
6102                         if (error != 0) {
6103                                 stp->sd_flag &= ~waitflags;
6104                                 cv_broadcast(&stp->sd_iocmonitor);
6105                                 mutex_exit(&stp->sd_lock);
6106                                 crfree(crp);
6107                                 return (error);
6108                         }
6109                 }
6110 
6111                 TRACE_0(TR_FAC_STREAMS_FR,
6112                     TR_STRDOIOCTL_WAIT2,
6113                     "strdoioctl sleeps awaiting reply");
6114                 ASSERT(error == 0);
6115 
6116                 cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6117                     (strioc->ic_timout ?
6118                     strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6119 
6120                 /*
6121                  * There are four possible cases here: interrupt, timeout,
6122                  * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6123                  * valid M_IOCTL reply).
6124                  *
6125                  * If we've been awakened by a STR_NOERROR ioctl on some other
6126                  * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6127                  * will be set.  Pretend as if we just timed out.  Note that
6128                  * this other thread waited at least STRTIMOUT before trying to
6129                  * awaken our thread, so this is indistinguishable (even for
6130                  * INFTIM) from the case where we failed with ETIME waiting on
6131                  * IOCWAIT in the prior loop.
6132                  */
6133                 if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6134                     stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6135                         cv_rval = -1;
6136                 }
6137 
6138                 /*
6139                  * note: STR_NOERROR does not protect
6140                  * us here.. use ic_timout < 0
6141                  */
6142                 if (cv_rval <= 0) {
6143                         if (cv_rval == 0) {
6144                                 error = EINTR;
6145                         } else {
6146                                 error =  ETIME;
6147                         }
6148                         /*
6149                          * A message could have come in after we were scheduled
6150                          * but before we were actually run.
6151                          */
6152                         bp = stp->sd_iocblk;
6153                         stp->sd_iocblk = NULL;
6154                         if (bp != NULL) {
6155                                 if ((bp->b_datap->db_type == M_COPYIN) ||
6156                                     (bp->b_datap->db_type == M_COPYOUT)) {
6157                                         mutex_exit(&stp->sd_lock);
6158                                         if (bp->b_cont) {
6159                                                 freemsg(bp->b_cont);
6160                                                 bp->b_cont = NULL;
6161                                         }
6162                                         bp->b_datap->db_type = M_IOCDATA;
6163                                         bp->b_wptr = bp->b_rptr +
6164                                             sizeof (struct copyresp);
6165                                         resp = (struct copyresp *)bp->b_rptr;
6166                                         resp->cp_rval =
6167                                             (caddr_t)1; /* failure */
6168                                         stream_willservice(stp);
6169                                         putnext(stp->sd_wrq, bp);
6170                                         stream_runservice(stp);
6171                                         mutex_enter(&stp->sd_lock);
6172                                 } else {
6173                                         freemsg(bp);
6174                                 }
6175                         }
6176                         stp->sd_flag &= ~waitflags;
6177                         cv_broadcast(&stp->sd_iocmonitor);
6178                         mutex_exit(&stp->sd_lock);
6179                         crfree(crp);
6180                         return (error);
6181                 }
6182         }
6183         bp = stp->sd_iocblk;
6184         /*
6185          * Note: it is strictly impossible to get here with sd_iocblk set to
6186          * -1.  This is because the initial loop above doesn't allow any new
6187          * ioctls into the fray until all others have passed this point.
6188          */
6189         ASSERT(bp != NULL && bp != (mblk_t *)-1);
6190         TRACE_1(TR_FAC_STREAMS_FR,
6191             TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6192         if ((bp->b_datap->db_type == M_IOCACK) ||
6193             (bp->b_datap->db_type == M_IOCNAK)) {
6194                 /* for detection of duplicate ioctl replies */
6195                 stp->sd_iocblk = (mblk_t *)-1;
6196                 stp->sd_flag &= ~waitflags;
6197                 cv_broadcast(&stp->sd_iocmonitor);
6198                 mutex_exit(&stp->sd_lock);
6199         } else {
6200                 /*
6201                  * flags not cleared here because we're still doing
6202                  * copy in/out for ioctl.
6203                  */
6204                 stp->sd_iocblk = NULL;
6205                 mutex_exit(&stp->sd_lock);
6206         }
6207 
6208 
6209         /*
6210          * Have received acknowledgment.
6211          */
6212 
6213         switch (bp->b_datap->db_type) {
6214         case M_IOCACK:
6215                 /*
6216                  * Positive ack.
6217                  */
6218                 iocbp = (struct iocblk *)bp->b_rptr;
6219 
6220                 /*
6221                  * Set error if indicated.
6222                  */
6223                 if (iocbp->ioc_error) {
6224                         error = iocbp->ioc_error;
6225                         break;
6226                 }
6227 
6228                 /*
6229                  * Set return value.
6230                  */
6231                 *rvalp = iocbp->ioc_rval;
6232 
6233                 /*
6234                  * Data may have been returned in ACK message (ioc_count > 0).
6235                  * If so, copy it out to the user's buffer.
6236                  */
6237                 if (iocbp->ioc_count && !transparent) {
6238                         if (error = getiocd(bp, strioc->ic_dp, copyflag))
6239                                 break;
6240                 }
6241                 if (!transparent) {
6242                         if (len)        /* an M_COPYOUT was used with I_STR */
6243                                 strioc->ic_len = len;
6244                         else
6245                                 strioc->ic_len = (int)iocbp->ioc_count;
6246                 }
6247                 break;
6248 
6249         case M_IOCNAK:
6250                 /*
6251                  * Negative ack.
6252                  *
6253                  * The only thing to do is set error as specified
6254                  * in neg ack packet.
6255                  */
6256                 iocbp = (struct iocblk *)bp->b_rptr;
6257 
6258                 error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6259                 break;
6260 
6261         case M_COPYIN:
6262                 /*
6263                  * Driver or module has requested user ioctl data.
6264                  */
6265                 reqp = (struct copyreq *)bp->b_rptr;
6266 
6267                 /*
6268                  * M_COPYIN should *never* have a message attached, though
6269                  * it's harmless if it does -- thus, panic on a DEBUG
6270                  * kernel and just free it on a non-DEBUG build.
6271                  */
6272                 ASSERT(bp->b_cont == NULL);
6273                 if (bp->b_cont != NULL) {
6274                         freemsg(bp->b_cont);
6275                         bp->b_cont = NULL;
6276                 }
6277 
6278                 error = putiocd(bp, reqp->cq_addr, flag, crp);
6279                 if (error && bp->b_cont) {
6280                         freemsg(bp->b_cont);
6281                         bp->b_cont = NULL;
6282                 }
6283 
6284                 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6285                 bp->b_datap->db_type = M_IOCDATA;
6286 
6287                 mblk_setcred(bp, crp, curproc->p_pid);
6288                 resp = (struct copyresp *)bp->b_rptr;
6289                 resp->cp_rval = (caddr_t)(uintptr_t)error;
6290                 resp->cp_flag = (fflags & FMODELS);
6291 
6292                 stream_willservice(stp);
6293                 putnext(stp->sd_wrq, bp);
6294                 stream_runservice(stp);
6295 
6296                 if (error) {
6297                         mutex_enter(&stp->sd_lock);
6298                         stp->sd_flag &= ~waitflags;
6299                         cv_broadcast(&stp->sd_iocmonitor);
6300                         mutex_exit(&stp->sd_lock);
6301                         crfree(crp);
6302                         return (error);
6303                 }
6304 
6305                 goto waitioc;
6306 
6307         case M_COPYOUT:
6308                 /*
6309                  * Driver or module has ioctl data for a user.
6310                  */
6311                 reqp = (struct copyreq *)bp->b_rptr;
6312                 ASSERT(bp->b_cont != NULL);
6313 
6314                 /*
6315                  * Always (transparent or non-transparent )
6316                  * use the address specified in the request
6317                  */
6318                 taddr = reqp->cq_addr;
6319                 if (!transparent)
6320                         len = (int)reqp->cq_size;
6321 
6322                 /* copyout data to the provided address */
6323                 error = getiocd(bp, taddr, copyflag);
6324 
6325                 freemsg(bp->b_cont);
6326                 bp->b_cont = NULL;
6327 
6328                 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6329                 bp->b_datap->db_type = M_IOCDATA;
6330 
6331                 mblk_setcred(bp, crp, curproc->p_pid);
6332                 resp = (struct copyresp *)bp->b_rptr;
6333                 resp->cp_rval = (caddr_t)(uintptr_t)error;
6334                 resp->cp_flag = (fflags & FMODELS);
6335 
6336                 stream_willservice(stp);
6337                 putnext(stp->sd_wrq, bp);
6338                 stream_runservice(stp);
6339 
6340                 if (error) {
6341                         mutex_enter(&stp->sd_lock);
6342                         stp->sd_flag &= ~waitflags;
6343                         cv_broadcast(&stp->sd_iocmonitor);
6344                         mutex_exit(&stp->sd_lock);
6345                         crfree(crp);
6346                         return (error);
6347                 }
6348                 goto waitioc;
6349 
6350         default:
6351                 ASSERT(0);
6352                 mutex_enter(&stp->sd_lock);
6353                 stp->sd_flag &= ~waitflags;
6354                 cv_broadcast(&stp->sd_iocmonitor);
6355                 mutex_exit(&stp->sd_lock);
6356                 break;
6357         }
6358 
6359         freemsg(bp);
6360         crfree(crp);
6361         return (error);
6362 }
6363 
6364 /*
6365  * Send an M_CMD message downstream and wait for a reply.  This is a ptools
6366  * special used to retrieve information from modules/drivers a stream without
6367  * being subjected to flow control or interfering with pending messages on the
6368  * stream (e.g. an ioctl in flight).
6369  */
6370 int
6371 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6372 {
6373         mblk_t *mp;
6374         struct cmdblk *cmdp;
6375         int error = 0;
6376         int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6377         clock_t rval, timeout = STRTIMOUT;
6378 
6379         if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6380             scp->sc_timeout < -1)
6381                 return (EINVAL);
6382 
6383         if (scp->sc_timeout > 0)
6384                 timeout = scp->sc_timeout * MILLISEC;
6385 
6386         if ((mp = allocb_cred(sizeof (struct cmdblk), crp,
6387             curproc->p_pid)) == NULL)
6388                 return (ENOMEM);
6389 
6390         crhold(crp);
6391 
6392         cmdp = (struct cmdblk *)mp->b_wptr;
6393         cmdp->cb_cr = crp;
6394         cmdp->cb_cmd = scp->sc_cmd;
6395         cmdp->cb_len = scp->sc_len;
6396         cmdp->cb_error = 0;
6397         mp->b_wptr += sizeof (struct cmdblk);
6398 
6399         DB_TYPE(mp) = M_CMD;
6400         DB_CPID(mp) = curproc->p_pid;
6401 
6402         /*
6403          * Copy in the payload.
6404          */
6405         if (cmdp->cb_len > 0) {
6406                 mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp,
6407                     curproc->p_pid);
6408                 if (mp->b_cont == NULL) {
6409                         error = ENOMEM;
6410                         goto out;
6411                 }
6412 
6413                 /* cb_len comes from sc_len, which has already been checked */
6414                 ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6415                 (void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6416                 mp->b_cont->b_wptr += cmdp->cb_len;
6417                 DB_CPID(mp->b_cont) = curproc->p_pid;
6418         }
6419 
6420         /*
6421          * Since this mechanism is strictly for ptools, and since only one
6422          * process can be grabbed at a time, we simply fail if there's
6423          * currently an operation pending.
6424          */
6425         mutex_enter(&stp->sd_lock);
6426         if (stp->sd_flag & STRCMDWAIT) {
6427                 mutex_exit(&stp->sd_lock);
6428                 error = EBUSY;
6429                 goto out;
6430         }
6431         stp->sd_flag |= STRCMDWAIT;
6432         ASSERT(stp->sd_cmdblk == NULL);
6433         mutex_exit(&stp->sd_lock);
6434 
6435         putnext(stp->sd_wrq, mp);
6436         mp = NULL;
6437 
6438         /*
6439          * Timed wait for acknowledgment.  If the reply has already arrived,
6440          * don't sleep.  If awakened from the sleep, fail only if the reply
6441          * has not arrived by then.  Otherwise, process the reply.
6442          */
6443         mutex_enter(&stp->sd_lock);
6444         while (stp->sd_cmdblk == NULL) {
6445                 if (stp->sd_flag & errs) {
6446                         if ((error = strgeterr(stp, errs, 0)) != 0)
6447                                 goto waitout;
6448                 }
6449 
6450                 rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6451                 if (stp->sd_cmdblk != NULL)
6452                         break;
6453 
6454                 if (rval <= 0) {
6455                         error = (rval == 0) ? EINTR : ETIME;
6456                         goto waitout;
6457                 }
6458         }
6459 
6460         /*
6461          * We received a reply.
6462          */
6463         mp = stp->sd_cmdblk;
6464         stp->sd_cmdblk = NULL;
6465         ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6466         ASSERT(stp->sd_flag & STRCMDWAIT);
6467         stp->sd_flag &= ~STRCMDWAIT;
6468         mutex_exit(&stp->sd_lock);
6469 
6470         cmdp = (struct cmdblk *)mp->b_rptr;
6471         if ((error = cmdp->cb_error) != 0)
6472                 goto out;
6473 
6474         /*
6475          * Data may have been returned in the reply (cb_len > 0).
6476          * If so, copy it out to the user's buffer.
6477          */
6478         if (cmdp->cb_len > 0) {
6479                 if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6480                         error = EPROTO;
6481                         goto out;
6482                 }
6483 
6484                 cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6485                 (void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6486         }
6487         scp->sc_len = cmdp->cb_len;
6488 out:
6489         freemsg(mp);
6490         crfree(crp);
6491         return (error);
6492 waitout:
6493         ASSERT(stp->sd_cmdblk == NULL);
6494         stp->sd_flag &= ~STRCMDWAIT;
6495         mutex_exit(&stp->sd_lock);
6496         crfree(crp);
6497         return (error);
6498 }
6499 
6500 /*
6501  * For the SunOS keyboard driver.
6502  * Return the next available "ioctl" sequence number.
6503  * Exported, so that streams modules can send "ioctl" messages
6504  * downstream from their open routine.
6505  */
6506 int
6507 getiocseqno(void)
6508 {
6509         int     i;
6510 
6511         mutex_enter(&strresources);
6512         i = ++ioc_id;
6513         mutex_exit(&strresources);
6514         return (i);
6515 }
6516 
6517 /*
6518  * Get the next message from the read queue.  If the message is
6519  * priority, STRPRI will have been set by strrput().  This flag
6520  * should be reset only when the entire message at the front of the
6521  * queue as been consumed.
6522  *
6523  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6524  */
6525 int
6526 strgetmsg(
6527         struct vnode *vp,
6528         struct strbuf *mctl,
6529         struct strbuf *mdata,
6530         unsigned char *prip,
6531         int *flagsp,
6532         int fmode,
6533         rval_t *rvp)
6534 {
6535         struct stdata *stp;
6536         mblk_t *bp, *nbp;
6537         mblk_t *savemp = NULL;
6538         mblk_t *savemptail = NULL;
6539         uint_t old_sd_flag;
6540         int flg;
6541         int more = 0;
6542         int error = 0;
6543         char first = 1;
6544         uint_t mark;            /* Contains MSG*MARK and _LASTMARK */
6545 #define _LASTMARK       0x8000  /* Distinct from MSG*MARK */
6546         unsigned char pri = 0;
6547         queue_t *q;
6548         int     pr = 0;                 /* Partial read successful */
6549         struct uio uios;
6550         struct uio *uiop = &uios;
6551         struct iovec iovs;
6552         unsigned char type;
6553 
6554         TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6555             "strgetmsg:%p", vp);
6556 
6557         ASSERT(vp->v_stream);
6558         stp = vp->v_stream;
6559         rvp->r_val1 = 0;
6560 
6561         mutex_enter(&stp->sd_lock);
6562 
6563         if ((error = i_straccess(stp, JCREAD)) != 0) {
6564                 mutex_exit(&stp->sd_lock);
6565                 return (error);
6566         }
6567 
6568         if (stp->sd_flag & (STRDERR|STPLEX)) {
6569                 error = strgeterr(stp, STRDERR|STPLEX, 0);
6570                 if (error != 0) {
6571                         mutex_exit(&stp->sd_lock);
6572                         return (error);
6573                 }
6574         }
6575         mutex_exit(&stp->sd_lock);
6576 
6577         switch (*flagsp) {
6578         case MSG_HIPRI:
6579                 if (*prip != 0)
6580                         return (EINVAL);
6581                 break;
6582 
6583         case MSG_ANY:
6584         case MSG_BAND:
6585                 break;
6586 
6587         default:
6588                 return (EINVAL);
6589         }
6590         /*
6591          * Setup uio and iov for data part
6592          */
6593         iovs.iov_base = mdata->buf;
6594         iovs.iov_len = mdata->maxlen;
6595         uios.uio_iov = &iovs;
6596         uios.uio_iovcnt = 1;
6597         uios.uio_loffset = 0;
6598         uios.uio_segflg = UIO_USERSPACE;
6599         uios.uio_fmode = 0;
6600         uios.uio_extflg = UIO_COPY_CACHED;
6601         uios.uio_resid = mdata->maxlen;
6602         uios.uio_offset = 0;
6603 
6604         q = _RD(stp->sd_wrq);
6605         mutex_enter(&stp->sd_lock);
6606         old_sd_flag = stp->sd_flag;
6607         mark = 0;
6608         for (;;) {
6609                 int done = 0;
6610                 mblk_t *q_first = q->q_first;
6611 
6612                 /*
6613                  * Get the next message of appropriate priority
6614                  * from the stream head.  If the caller is interested
6615                  * in band or hipri messages, then they should already
6616                  * be enqueued at the stream head.  On the other hand
6617                  * if the caller wants normal (band 0) messages, they
6618                  * might be deferred in a synchronous stream and they
6619                  * will need to be pulled up.
6620                  *
6621                  * After we have dequeued a message, we might find that
6622                  * it was a deferred M_SIG that was enqueued at the
6623                  * stream head.  It must now be posted as part of the
6624                  * read by calling strsignal_nolock().
6625                  *
6626                  * Also note that strrput does not enqueue an M_PCSIG,
6627                  * and there cannot be more than one hipri message,
6628                  * so there was no need to have the M_PCSIG case.
6629                  *
6630                  * At some time it might be nice to try and wrap the
6631                  * functionality of kstrgetmsg() and strgetmsg() into
6632                  * a common routine so to reduce the amount of replicated
6633                  * code (since they are extremely similar).
6634                  */
6635                 if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6636                         /* Asking for normal, band0 data */
6637                         bp = strget(stp, q, uiop, first, &error);
6638                         ASSERT(MUTEX_HELD(&stp->sd_lock));
6639                         if (bp != NULL) {
6640                                 if (DB_TYPE(bp) == M_SIG) {
6641                                         strsignal_nolock(stp, *bp->b_rptr,
6642                                             bp->b_band);
6643                                         freemsg(bp);
6644                                         continue;
6645                                 } else {
6646                                         break;
6647                                 }
6648                         }
6649                         if (error != 0)
6650                                 goto getmout;
6651 
6652                 /*
6653                  * We can't depend on the value of STRPRI here because
6654                  * the stream head may be in transit. Therefore, we
6655                  * must look at the type of the first message to
6656                  * determine if a high priority messages is waiting
6657                  */
6658                 } else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6659                     DB_TYPE(q_first) >= QPCTL &&
6660                     (bp = getq_noenab(q, 0)) != NULL) {
6661                         /* Asked for HIPRI and got one */
6662                         ASSERT(DB_TYPE(bp) >= QPCTL);
6663                         break;
6664                 } else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6665                     ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
6666                     (bp = getq_noenab(q, 0)) != NULL) {
6667                         /*
6668                          * Asked for at least band "prip" and got either at
6669                          * least that band or a hipri message.
6670                          */
6671                         ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
6672                         if (DB_TYPE(bp) == M_SIG) {
6673                                 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6674                                 freemsg(bp);
6675                                 continue;
6676                         } else {
6677                                 break;
6678                         }
6679                 }
6680 
6681                 /* No data. Time to sleep? */
6682                 qbackenable(q, 0);
6683 
6684                 /*
6685                  * If STRHUP or STREOF, return 0 length control and data.
6686                  * If resid is 0, then a read(fd,buf,0) was done. Do not
6687                  * sleep to satisfy this request because by default we have
6688                  * zero bytes to return.
6689                  */
6690                 if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6691                     mdata->maxlen == 0)) {
6692                         mctl->len = mdata->len = 0;
6693                         *flagsp = 0;
6694                         mutex_exit(&stp->sd_lock);
6695                         return (0);
6696                 }
6697                 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6698                     "strgetmsg calls strwaitq:%p, %p",
6699                     vp, uiop);
6700                 if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6701                     &done)) != 0) || done) {
6702                         TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6703                             "strgetmsg error or done:%p, %p",
6704                             vp, uiop);
6705                         mutex_exit(&stp->sd_lock);
6706                         return (error);
6707                 }
6708                 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6709                     "strgetmsg awakes:%p, %p", vp, uiop);
6710                 if ((error = i_straccess(stp, JCREAD)) != 0) {
6711                         mutex_exit(&stp->sd_lock);
6712                         return (error);
6713                 }
6714                 first = 0;
6715         }
6716         ASSERT(bp != NULL);
6717         /*
6718          * Extract any mark information. If the message is not completely
6719          * consumed this information will be put in the mblk
6720          * that is putback.
6721          * If MSGMARKNEXT is set and the message is completely consumed
6722          * the STRATMARK flag will be set below. Likewise, if
6723          * MSGNOTMARKNEXT is set and the message is
6724          * completely consumed STRNOTATMARK will be set.
6725          */
6726         mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6727         ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6728             (MSGMARKNEXT|MSGNOTMARKNEXT));
6729         if (mark != 0 && bp == stp->sd_mark) {
6730                 mark |= _LASTMARK;
6731                 stp->sd_mark = NULL;
6732         }
6733         /*
6734          * keep track of the original message type and priority
6735          */
6736         pri = bp->b_band;
6737         type = bp->b_datap->db_type;
6738         if (type == M_PASSFP) {
6739                 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6740                         stp->sd_mark = bp;
6741                 bp->b_flag |= mark & ~_LASTMARK;
6742                 putback(stp, q, bp, pri);
6743                 qbackenable(q, pri);
6744                 mutex_exit(&stp->sd_lock);
6745                 return (EBADMSG);
6746         }
6747         ASSERT(type != M_SIG);
6748 
6749         /*
6750          * Set this flag so strrput will not generate signals. Need to
6751          * make sure this flag is cleared before leaving this routine
6752          * else signals will stop being sent.
6753          */
6754         stp->sd_flag |= STRGETINPROG;
6755         mutex_exit(&stp->sd_lock);
6756 
6757         if (STREAM_NEEDSERVICE(stp))
6758                 stream_runservice(stp);
6759 
6760         /*
6761          * Set HIPRI flag if message is priority.
6762          */
6763         if (type >= QPCTL)
6764                 flg = MSG_HIPRI;
6765         else
6766                 flg = MSG_BAND;
6767 
6768         /*
6769          * First process PROTO or PCPROTO blocks, if any.
6770          */
6771         if (mctl->maxlen >= 0 && type != M_DATA) {
6772                 size_t  n, bcnt;
6773                 char    *ubuf;
6774 
6775                 bcnt = mctl->maxlen;
6776                 ubuf = mctl->buf;
6777                 while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6778                         if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6779                             copyout(bp->b_rptr, ubuf, n)) {
6780                                 error = EFAULT;
6781                                 mutex_enter(&stp->sd_lock);
6782                                 /*
6783                                  * clear stream head pri flag based on
6784                                  * first message type
6785                                  */
6786                                 if (type >= QPCTL) {
6787                                         ASSERT(type == M_PCPROTO);
6788                                         stp->sd_flag &= ~STRPRI;
6789                                 }
6790                                 more = 0;
6791                                 freemsg(bp);
6792                                 goto getmout;
6793                         }
6794                         ubuf += n;
6795                         bp->b_rptr += n;
6796                         if (bp->b_rptr >= bp->b_wptr) {
6797                                 nbp = bp;
6798                                 bp = bp->b_cont;
6799                                 freeb(nbp);
6800                         }
6801                         ASSERT(n <= bcnt);
6802                         bcnt -= n;
6803                         if (bcnt == 0)
6804                                 break;
6805                 }
6806                 mctl->len = mctl->maxlen - bcnt;
6807         } else
6808                 mctl->len = -1;
6809 
6810         if (bp && bp->b_datap->db_type != M_DATA) {
6811                 /*
6812                  * More PROTO blocks in msg.
6813                  */
6814                 more |= MORECTL;
6815                 savemp = bp;
6816                 while (bp && bp->b_datap->db_type != M_DATA) {
6817                         savemptail = bp;
6818                         bp = bp->b_cont;
6819                 }
6820                 savemptail->b_cont = NULL;
6821         }
6822 
6823         /*
6824          * Now process DATA blocks, if any.
6825          */
6826         if (mdata->maxlen >= 0 && bp) {
6827                 /*
6828                  * struiocopyout will consume a potential zero-length
6829                  * M_DATA even if uio_resid is zero.
6830                  */
6831                 size_t oldresid = uiop->uio_resid;
6832 
6833                 bp = struiocopyout(bp, uiop, &error);
6834                 if (error != 0) {
6835                         mutex_enter(&stp->sd_lock);
6836                         /*
6837                          * clear stream head hi pri flag based on
6838                          * first message
6839                          */
6840                         if (type >= QPCTL) {
6841                                 ASSERT(type == M_PCPROTO);
6842                                 stp->sd_flag &= ~STRPRI;
6843                         }
6844                         more = 0;
6845                         freemsg(savemp);
6846                         goto getmout;
6847                 }
6848                 /*
6849                  * (pr == 1) indicates a partial read.
6850                  */
6851                 if (oldresid > uiop->uio_resid)
6852                         pr = 1;
6853                 mdata->len = mdata->maxlen - uiop->uio_resid;
6854         } else
6855                 mdata->len = -1;
6856 
6857         if (bp) {                       /* more data blocks in msg */
6858                 more |= MOREDATA;
6859                 if (savemp)
6860                         savemptail->b_cont = bp;
6861                 else
6862                         savemp = bp;
6863         }
6864 
6865         mutex_enter(&stp->sd_lock);
6866         if (savemp) {
6867                 if (pr && (savemp->b_datap->db_type == M_DATA) &&
6868                     msgnodata(savemp)) {
6869                         /*
6870                          * Avoid queuing a zero-length tail part of
6871                          * a message. pr=1 indicates that we read some of
6872                          * the message.
6873                          */
6874                         freemsg(savemp);
6875                         more &= ~MOREDATA;
6876                         /*
6877                          * clear stream head hi pri flag based on
6878                          * first message
6879                          */
6880                         if (type >= QPCTL) {
6881                                 ASSERT(type == M_PCPROTO);
6882                                 stp->sd_flag &= ~STRPRI;
6883                         }
6884                 } else {
6885                         savemp->b_band = pri;
6886                         /*
6887                          * If the first message was HIPRI and the one we're
6888                          * putting back isn't, then clear STRPRI, otherwise
6889                          * set STRPRI again.  Note that we must set STRPRI
6890                          * again since the flush logic in strrput_nondata()
6891                          * may have cleared it while we had sd_lock dropped.
6892                          */
6893                         if (type >= QPCTL) {
6894                                 ASSERT(type == M_PCPROTO);
6895                                 if (queclass(savemp) < QPCTL)
6896                                         stp->sd_flag &= ~STRPRI;
6897                                 else
6898                                         stp->sd_flag |= STRPRI;
6899                         } else if (queclass(savemp) >= QPCTL) {
6900                                 /*
6901                                  * The first message was not a HIPRI message,
6902                                  * but the one we are about to putback is.
6903                                  * For simplicitly, we do not allow for HIPRI
6904                                  * messages to be embedded in the message
6905                                  * body, so just force it to same type as
6906                                  * first message.
6907                                  */
6908                                 ASSERT(type == M_DATA || type == M_PROTO);
6909                                 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6910                                 savemp->b_datap->db_type = type;
6911                         }
6912                         if (mark != 0) {
6913                                 savemp->b_flag |= mark & ~_LASTMARK;
6914                                 if ((mark & _LASTMARK) &&
6915                                     (stp->sd_mark == NULL)) {
6916                                         /*
6917                                          * If another marked message arrived
6918                                          * while sd_lock was not held sd_mark
6919                                          * would be non-NULL.
6920                                          */
6921                                         stp->sd_mark = savemp;
6922                                 }
6923                         }
6924                         putback(stp, q, savemp, pri);
6925                 }
6926         } else {
6927                 /*
6928                  * The complete message was consumed.
6929                  *
6930                  * If another M_PCPROTO arrived while sd_lock was not held
6931                  * it would have been discarded since STRPRI was still set.
6932                  *
6933                  * Move the MSG*MARKNEXT information
6934                  * to the stream head just in case
6935                  * the read queue becomes empty.
6936                  * clear stream head hi pri flag based on
6937                  * first message
6938                  *
6939                  * If the stream head was at the mark
6940                  * (STRATMARK) before we dropped sd_lock above
6941                  * and some data was consumed then we have
6942                  * moved past the mark thus STRATMARK is
6943                  * cleared. However, if a message arrived in
6944                  * strrput during the copyout above causing
6945                  * STRATMARK to be set we can not clear that
6946                  * flag.
6947                  */
6948                 if (type >= QPCTL) {
6949                         ASSERT(type == M_PCPROTO);
6950                         stp->sd_flag &= ~STRPRI;
6951                 }
6952                 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6953                         if (mark & MSGMARKNEXT) {
6954                                 stp->sd_flag &= ~STRNOTATMARK;
6955                                 stp->sd_flag |= STRATMARK;
6956                         } else if (mark & MSGNOTMARKNEXT) {
6957                                 stp->sd_flag &= ~STRATMARK;
6958                                 stp->sd_flag |= STRNOTATMARK;
6959                         } else {
6960                                 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6961                         }
6962                 } else if (pr && (old_sd_flag & STRATMARK)) {
6963                         stp->sd_flag &= ~STRATMARK;
6964                 }
6965         }
6966 
6967         *flagsp = flg;
6968         *prip = pri;
6969 
6970         /*
6971          * Getmsg cleanup processing - if the state of the queue has changed
6972          * some signals may need to be sent and/or poll awakened.
6973          */
6974 getmout:
6975         qbackenable(q, pri);
6976 
6977         /*
6978          * We dropped the stream head lock above. Send all M_SIG messages
6979          * before processing stream head for SIGPOLL messages.
6980          */
6981         ASSERT(MUTEX_HELD(&stp->sd_lock));
6982         while ((bp = q->q_first) != NULL &&
6983             (bp->b_datap->db_type == M_SIG)) {
6984                 /*
6985                  * sd_lock is held so the content of the read queue can not
6986                  * change.
6987                  */
6988                 bp = getq(q);
6989                 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
6990 
6991                 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6992                 mutex_exit(&stp->sd_lock);
6993                 freemsg(bp);
6994                 if (STREAM_NEEDSERVICE(stp))
6995                         stream_runservice(stp);
6996                 mutex_enter(&stp->sd_lock);
6997         }
6998 
6999         /*
7000          * stream head cannot change while we make the determination
7001          * whether or not to send a signal. Drop the flag to allow strrput
7002          * to send firstmsgsigs again.
7003          */
7004         stp->sd_flag &= ~STRGETINPROG;
7005 
7006         /*
7007          * If the type of message at the front of the queue changed
7008          * due to the receive the appropriate signals and pollwakeup events
7009          * are generated. The type of changes are:
7010          *      Processed a hipri message, q_first is not hipri.
7011          *      Processed a band X message, and q_first is band Y.
7012          * The generated signals and pollwakeups are identical to what
7013          * strrput() generates should the message that is now on q_first
7014          * arrive to an empty read queue.
7015          *
7016          * Note: only strrput will send a signal for a hipri message.
7017          */
7018         if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7019                 strsigset_t signals = 0;
7020                 strpollset_t pollwakeups = 0;
7021 
7022                 if (flg & MSG_HIPRI) {
7023                         /*
7024                          * Removed a hipri message. Regular data at
7025                          * the front of  the queue.
7026                          */
7027                         if (bp->b_band == 0) {
7028                                 signals = S_INPUT | S_RDNORM;
7029                                 pollwakeups = POLLIN | POLLRDNORM;
7030                         } else {
7031                                 signals = S_INPUT | S_RDBAND;
7032                                 pollwakeups = POLLIN | POLLRDBAND;
7033                         }
7034                 } else if (pri != bp->b_band) {
7035                         /*
7036                          * The band is different for the new q_first.
7037                          */
7038                         if (bp->b_band == 0) {
7039                                 signals = S_RDNORM;
7040                                 pollwakeups = POLLIN | POLLRDNORM;
7041                         } else {
7042                                 signals = S_RDBAND;
7043                                 pollwakeups = POLLIN | POLLRDBAND;
7044                         }
7045                 }
7046 
7047                 if (pollwakeups != 0) {
7048                         if (pollwakeups == (POLLIN | POLLRDNORM)) {
7049                                 if (!(stp->sd_rput_opt & SR_POLLIN))
7050                                         goto no_pollwake;
7051                                 stp->sd_rput_opt &= ~SR_POLLIN;
7052                         }
7053                         mutex_exit(&stp->sd_lock);
7054                         pollwakeup(&stp->sd_pollist, pollwakeups);
7055                         mutex_enter(&stp->sd_lock);
7056                 }
7057 no_pollwake:
7058 
7059                 if (stp->sd_sigflags & signals)
7060                         strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7061         }
7062         mutex_exit(&stp->sd_lock);
7063 
7064         rvp->r_val1 = more;
7065         return (error);
7066 #undef  _LASTMARK
7067 }
7068 
7069 /*
7070  * Get the next message from the read queue.  If the message is
7071  * priority, STRPRI will have been set by strrput().  This flag
7072  * should be reset only when the entire message at the front of the
7073  * queue as been consumed.
7074  *
7075  * If uiop is NULL all data is returned in mctlp.
7076  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7077  * not enabled.
7078  * The timeout parameter is in milliseconds; -1 for infinity.
7079  * This routine handles the consolidation private flags:
7080  *      MSG_IGNERROR    Ignore any stream head error except STPLEX.
7081  *      MSG_DELAYERROR  Defer the error check until the queue is empty.
7082  *      MSG_HOLDSIG     Hold signals while waiting for data.
7083  *      MSG_IPEEK       Only peek at messages.
7084  *      MSG_DISCARDTAIL Discard the tail M_DATA part of the message
7085  *                      that doesn't fit.
7086  *      MSG_NOMARK      If the message is marked leave it on the queue.
7087  *
7088  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7089  */
7090 int
7091 kstrgetmsg(
7092         struct vnode *vp,
7093         mblk_t **mctlp,
7094         struct uio *uiop,
7095         unsigned char *prip,
7096         int *flagsp,
7097         clock_t timout,
7098         rval_t *rvp)
7099 {
7100         struct stdata *stp;
7101         mblk_t *bp, *nbp;
7102         mblk_t *savemp = NULL;
7103         mblk_t *savemptail = NULL;
7104         int flags;
7105         uint_t old_sd_flag;
7106         int flg;
7107         int more = 0;
7108         int error = 0;
7109         char first = 1;
7110         uint_t mark;            /* Contains MSG*MARK and _LASTMARK */
7111 #define _LASTMARK       0x8000  /* Distinct from MSG*MARK */
7112         unsigned char pri = 0;
7113         queue_t *q;
7114         int     pr = 0;                 /* Partial read successful */
7115         unsigned char type;
7116 
7117         TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7118             "kstrgetmsg:%p", vp);
7119 
7120         ASSERT(vp->v_stream);
7121         stp = vp->v_stream;
7122         rvp->r_val1 = 0;
7123 
7124         mutex_enter(&stp->sd_lock);
7125 
7126         if ((error = i_straccess(stp, JCREAD)) != 0) {
7127                 mutex_exit(&stp->sd_lock);
7128                 return (error);
7129         }
7130 
7131         flags = *flagsp;
7132         if (stp->sd_flag & (STRDERR|STPLEX)) {
7133                 if ((stp->sd_flag & STPLEX) ||
7134                     (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7135                         error = strgeterr(stp, STRDERR|STPLEX,
7136                             (flags & MSG_IPEEK));
7137                         if (error != 0) {
7138                                 mutex_exit(&stp->sd_lock);
7139                                 return (error);
7140                         }
7141                 }
7142         }
7143         mutex_exit(&stp->sd_lock);
7144 
7145         switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7146         case MSG_HIPRI:
7147                 if (*prip != 0)
7148                         return (EINVAL);
7149                 break;
7150 
7151         case MSG_ANY:
7152         case MSG_BAND:
7153                 break;
7154 
7155         default:
7156                 return (EINVAL);
7157         }
7158 
7159 retry:
7160         q = _RD(stp->sd_wrq);
7161         mutex_enter(&stp->sd_lock);
7162         old_sd_flag = stp->sd_flag;
7163         mark = 0;
7164         for (;;) {
7165                 int done = 0;
7166                 int waitflag;
7167                 int fmode;
7168                 mblk_t *q_first = q->q_first;
7169 
7170                 /*
7171                  * This section of the code operates just like the code
7172                  * in strgetmsg().  There is a comment there about what
7173                  * is going on here.
7174                  */
7175                 if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7176                         /* Asking for normal, band0 data */
7177                         bp = strget(stp, q, uiop, first, &error);
7178                         ASSERT(MUTEX_HELD(&stp->sd_lock));
7179                         if (bp != NULL) {
7180                                 if (DB_TYPE(bp) == M_SIG) {
7181                                         strsignal_nolock(stp, *bp->b_rptr,
7182                                             bp->b_band);
7183                                         freemsg(bp);
7184                                         continue;
7185                                 } else {
7186                                         break;
7187                                 }
7188                         }
7189                         if (error != 0) {
7190                                 goto getmout;
7191                         }
7192                 /*
7193                  * We can't depend on the value of STRPRI here because
7194                  * the stream head may be in transit. Therefore, we
7195                  * must look at the type of the first message to
7196                  * determine if a high priority messages is waiting
7197                  */
7198                 } else if ((flags & MSG_HIPRI) && q_first != NULL &&
7199                     DB_TYPE(q_first) >= QPCTL &&
7200                     (bp = getq_noenab(q, 0)) != NULL) {
7201                         ASSERT(DB_TYPE(bp) >= QPCTL);
7202                         break;
7203                 } else if ((flags & MSG_BAND) && q_first != NULL &&
7204                     ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
7205                     (bp = getq_noenab(q, 0)) != NULL) {
7206                         /*
7207                          * Asked for at least band "prip" and got either at
7208                          * least that band or a hipri message.
7209                          */
7210                         ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
7211                         if (DB_TYPE(bp) == M_SIG) {
7212                                 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7213                                 freemsg(bp);
7214                                 continue;
7215                         } else {
7216                                 break;
7217                         }
7218                 }
7219 
7220                 /* No data. Time to sleep? */
7221                 qbackenable(q, 0);
7222 
7223                 /*
7224                  * Delayed error notification?
7225                  */
7226                 if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7227                     (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7228                         error = strgeterr(stp, STRDERR|STPLEX,
7229                             (flags & MSG_IPEEK));
7230                         if (error != 0) {
7231                                 mutex_exit(&stp->sd_lock);
7232                                 return (error);
7233                         }
7234                 }
7235 
7236                 /*
7237                  * If STRHUP or STREOF, return 0 length control and data.
7238                  * If a read(fd,buf,0) has been done, do not sleep, just
7239                  * return.
7240                  *
7241                  * If mctlp == NULL and uiop == NULL, then the code will
7242                  * do the strwaitq. This is an understood way of saying
7243                  * sleep "polling" until a message is received.
7244                  */
7245                 if ((stp->sd_flag & (STRHUP|STREOF)) ||
7246                     (uiop != NULL && uiop->uio_resid == 0)) {
7247                         if (mctlp != NULL)
7248                                 *mctlp = NULL;
7249                         *flagsp = 0;
7250                         mutex_exit(&stp->sd_lock);
7251                         return (0);
7252                 }
7253 
7254                 waitflag = GETWAIT;
7255                 if (flags &
7256                     (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7257                         if (flags & MSG_HOLDSIG)
7258                                 waitflag |= STR_NOSIG;
7259                         if (flags & MSG_IGNERROR)
7260                                 waitflag |= STR_NOERROR;
7261                         if (flags & MSG_IPEEK)
7262                                 waitflag |= STR_PEEK;
7263                         if (flags & MSG_DELAYERROR)
7264                                 waitflag |= STR_DELAYERR;
7265                 }
7266                 if (uiop != NULL)
7267                         fmode = uiop->uio_fmode;
7268                 else
7269                         fmode = 0;
7270 
7271                 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7272                     "kstrgetmsg calls strwaitq:%p, %p",
7273                     vp, uiop);
7274                 if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7275                     fmode, timout, &done))) != 0 || done) {
7276                         TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7277                             "kstrgetmsg error or done:%p, %p",
7278                             vp, uiop);
7279                         mutex_exit(&stp->sd_lock);
7280                         return (error);
7281                 }
7282                 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7283                     "kstrgetmsg awakes:%p, %p", vp, uiop);
7284                 if ((error = i_straccess(stp, JCREAD)) != 0) {
7285                         mutex_exit(&stp->sd_lock);
7286                         return (error);
7287                 }
7288                 first = 0;
7289         }
7290         ASSERT(bp != NULL);
7291         /*
7292          * Extract any mark information. If the message is not completely
7293          * consumed this information will be put in the mblk
7294          * that is putback.
7295          * If MSGMARKNEXT is set and the message is completely consumed
7296          * the STRATMARK flag will be set below. Likewise, if
7297          * MSGNOTMARKNEXT is set and the message is
7298          * completely consumed STRNOTATMARK will be set.
7299          */
7300         mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7301         ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7302             (MSGMARKNEXT|MSGNOTMARKNEXT));
7303         pri = bp->b_band;
7304         if (mark != 0) {
7305                 /*
7306                  * If the caller doesn't want the mark return.
7307                  * Used to implement MSG_WAITALL in sockets.
7308                  */
7309                 if (flags & MSG_NOMARK) {
7310                         putback(stp, q, bp, pri);
7311                         qbackenable(q, pri);
7312                         mutex_exit(&stp->sd_lock);
7313                         return (EWOULDBLOCK);
7314                 }
7315                 if (bp == stp->sd_mark) {
7316                         mark |= _LASTMARK;
7317                         stp->sd_mark = NULL;
7318                 }
7319         }
7320 
7321         /*
7322          * keep track of the first message type
7323          */
7324         type = bp->b_datap->db_type;
7325 
7326         if (bp->b_datap->db_type == M_PASSFP) {
7327                 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7328                         stp->sd_mark = bp;
7329                 bp->b_flag |= mark & ~_LASTMARK;
7330                 putback(stp, q, bp, pri);
7331                 qbackenable(q, pri);
7332                 mutex_exit(&stp->sd_lock);
7333                 return (EBADMSG);
7334         }
7335         ASSERT(type != M_SIG);
7336 
7337         if (flags & MSG_IPEEK) {
7338                 /*
7339                  * Clear any struioflag - we do the uiomove over again
7340                  * when peeking since it simplifies the code.
7341                  *
7342                  * Dup the message and put the original back on the queue.
7343                  * If dupmsg() fails, try again with copymsg() to see if
7344                  * there is indeed a shortage of memory.  dupmsg() may fail
7345                  * if db_ref in any of the messages reaches its limit.
7346                  */
7347 
7348                 if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7349                         /*
7350                          * Restore the state of the stream head since we
7351                          * need to drop sd_lock (strwaitbuf is sleeping).
7352                          */
7353                         size_t size = msgdsize(bp);
7354 
7355                         if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7356                                 stp->sd_mark = bp;
7357                         bp->b_flag |= mark & ~_LASTMARK;
7358                         putback(stp, q, bp, pri);
7359                         mutex_exit(&stp->sd_lock);
7360                         error = strwaitbuf(size, BPRI_HI);
7361                         if (error) {
7362                                 /*
7363                                  * There is no net change to the queue thus
7364                                  * no need to qbackenable.
7365                                  */
7366                                 return (error);
7367                         }
7368                         goto retry;
7369                 }
7370 
7371                 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7372                         stp->sd_mark = bp;
7373                 bp->b_flag |= mark & ~_LASTMARK;
7374                 putback(stp, q, bp, pri);
7375                 bp = nbp;
7376         }
7377 
7378         /*
7379          * Set this flag so strrput will not generate signals. Need to
7380          * make sure this flag is cleared before leaving this routine
7381          * else signals will stop being sent.
7382          */
7383         stp->sd_flag |= STRGETINPROG;
7384         mutex_exit(&stp->sd_lock);
7385 
7386         if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7387                 mblk_t *tmp, *prevmp;
7388 
7389                 /*
7390                  * Put first non-data mblk back to stream head and
7391                  * cut the mblk chain so sd_rputdatafunc only sees
7392                  * M_DATA mblks. We can skip the first mblk since it
7393                  * is M_DATA according to the condition above.
7394                  */
7395                 for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7396                     prevmp = tmp, tmp = tmp->b_cont) {
7397                         if (DB_TYPE(tmp) != M_DATA) {
7398                                 prevmp->b_cont = NULL;
7399                                 mutex_enter(&stp->sd_lock);
7400                                 putback(stp, q, tmp, tmp->b_band);
7401                                 mutex_exit(&stp->sd_lock);
7402                                 break;
7403                         }
7404                 }
7405 
7406                 bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7407                     NULL, NULL, NULL, NULL);
7408 
7409                 if (bp == NULL)
7410                         goto retry;
7411         }
7412 
7413         if (STREAM_NEEDSERVICE(stp))
7414                 stream_runservice(stp);
7415 
7416         /*
7417          * Set HIPRI flag if message is priority.
7418          */
7419         if (type >= QPCTL)
7420                 flg = MSG_HIPRI;
7421         else
7422                 flg = MSG_BAND;
7423 
7424         /*
7425          * First process PROTO or PCPROTO blocks, if any.
7426          */
7427         if (mctlp != NULL && type != M_DATA) {
7428                 mblk_t *nbp;
7429 
7430                 *mctlp = bp;
7431                 while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7432                         bp = bp->b_cont;
7433                 nbp = bp->b_cont;
7434                 bp->b_cont = NULL;
7435                 bp = nbp;
7436         }
7437 
7438         if (bp && bp->b_datap->db_type != M_DATA) {
7439                 /*
7440                  * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7441                  */
7442                 more |= MORECTL;
7443                 savemp = bp;
7444                 while (bp && bp->b_datap->db_type != M_DATA) {
7445                         savemptail = bp;
7446                         bp = bp->b_cont;
7447                 }
7448                 savemptail->b_cont = NULL;
7449         }
7450 
7451         /*
7452          * Now process DATA blocks, if any.
7453          */
7454         if (uiop == NULL) {
7455                 /* Append data to tail of mctlp */
7456 
7457                 if (mctlp != NULL) {
7458                         mblk_t **mpp = mctlp;
7459 
7460                         while (*mpp != NULL)
7461                                 mpp = &((*mpp)->b_cont);
7462                         *mpp = bp;
7463                         bp = NULL;
7464                 }
7465         } else if (uiop->uio_resid >= 0 && bp) {
7466                 size_t oldresid = uiop->uio_resid;
7467 
7468                 /*
7469                  * If a streams message is likely to consist
7470                  * of many small mblks, it is pulled up into
7471                  * one continuous chunk of memory.
7472                  * The size of the first mblk may be bogus because
7473                  * successive read() calls on the socket reduce
7474                  * the size of this mblk until it is exhausted
7475                  * and then the code walks on to the next. Thus
7476                  * the size of the mblk may not be the original size
7477                  * that was passed up, it's simply a remainder
7478                  * and hence can be very small without any
7479                  * implication that the packet is badly fragmented.
7480                  * So the size of the possible second mblk is
7481                  * used to spot a badly fragmented packet.
7482                  * see longer comment at top of page
7483                  * by mblk_pull_len declaration.
7484                  */
7485 
7486                 if (bp->b_cont != NULL && MBLKL(bp->b_cont) < mblk_pull_len) {
7487                         (void) pullupmsg(bp, -1);
7488                 }
7489 
7490                 bp = struiocopyout(bp, uiop, &error);
7491                 if (error != 0) {
7492                         if (mctlp != NULL) {
7493                                 freemsg(*mctlp);
7494                                 *mctlp = NULL;
7495                         } else
7496                                 freemsg(savemp);
7497                         mutex_enter(&stp->sd_lock);
7498                         /*
7499                          * clear stream head hi pri flag based on
7500                          * first message
7501                          */
7502                         if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7503                                 ASSERT(type == M_PCPROTO);
7504                                 stp->sd_flag &= ~STRPRI;
7505                         }
7506                         more = 0;
7507                         goto getmout;
7508                 }
7509                 /*
7510                  * (pr == 1) indicates a partial read.
7511                  */
7512                 if (oldresid > uiop->uio_resid)
7513                         pr = 1;
7514         }
7515 
7516         if (bp) {                       /* more data blocks in msg */
7517                 more |= MOREDATA;
7518                 if (savemp)
7519                         savemptail->b_cont = bp;
7520                 else
7521                         savemp = bp;
7522         }
7523 
7524         mutex_enter(&stp->sd_lock);
7525         if (savemp) {
7526                 if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7527                         /*
7528                          * When MSG_DISCARDTAIL is set or
7529                          * when peeking discard any tail. When peeking this
7530                          * is the tail of the dup that was copied out - the
7531                          * message has already been putback on the queue.
7532                          * Return MOREDATA to the caller even though the data
7533                          * is discarded. This is used by sockets (to
7534                          * set MSG_TRUNC).
7535                          */
7536                         freemsg(savemp);
7537                         if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7538                                 ASSERT(type == M_PCPROTO);
7539                                 stp->sd_flag &= ~STRPRI;
7540                         }
7541                 } else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7542                     msgnodata(savemp)) {
7543                         /*
7544                          * Avoid queuing a zero-length tail part of
7545                          * a message. pr=1 indicates that we read some of
7546                          * the message.
7547                          */
7548                         freemsg(savemp);
7549                         more &= ~MOREDATA;
7550                         if (type >= QPCTL) {
7551                                 ASSERT(type == M_PCPROTO);
7552                                 stp->sd_flag &= ~STRPRI;
7553                         }
7554                 } else {
7555                         savemp->b_band = pri;
7556                         /*
7557                          * If the first message was HIPRI and the one we're
7558                          * putting back isn't, then clear STRPRI, otherwise
7559                          * set STRPRI again.  Note that we must set STRPRI
7560                          * again since the flush logic in strrput_nondata()
7561                          * may have cleared it while we had sd_lock dropped.
7562                          */
7563 
7564                         if (type >= QPCTL) {
7565                                 ASSERT(type == M_PCPROTO);
7566                                 if (queclass(savemp) < QPCTL)
7567                                         stp->sd_flag &= ~STRPRI;
7568                                 else
7569                                         stp->sd_flag |= STRPRI;
7570                         } else if (queclass(savemp) >= QPCTL) {
7571                                 /*
7572                                  * The first message was not a HIPRI message,
7573                                  * but the one we are about to putback is.
7574                                  * For simplicitly, we do not allow for HIPRI
7575                                  * messages to be embedded in the message
7576                                  * body, so just force it to same type as
7577                                  * first message.
7578                                  */
7579                                 ASSERT(type == M_DATA || type == M_PROTO);
7580                                 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7581                                 savemp->b_datap->db_type = type;
7582                         }
7583                         if (mark != 0) {
7584                                 if ((mark & _LASTMARK) &&
7585                                     (stp->sd_mark == NULL)) {
7586                                         /*
7587                                          * If another marked message arrived
7588                                          * while sd_lock was not held sd_mark
7589                                          * would be non-NULL.
7590                                          */
7591                                         stp->sd_mark = savemp;
7592                                 }
7593                                 savemp->b_flag |= mark & ~_LASTMARK;
7594                         }
7595                         putback(stp, q, savemp, pri);
7596                 }
7597         } else if (!(flags & MSG_IPEEK)) {
7598                 /*
7599                  * The complete message was consumed.
7600                  *
7601                  * If another M_PCPROTO arrived while sd_lock was not held
7602                  * it would have been discarded since STRPRI was still set.
7603                  *
7604                  * Move the MSG*MARKNEXT information
7605                  * to the stream head just in case
7606                  * the read queue becomes empty.
7607                  * clear stream head hi pri flag based on
7608                  * first message
7609                  *
7610                  * If the stream head was at the mark
7611                  * (STRATMARK) before we dropped sd_lock above
7612                  * and some data was consumed then we have
7613                  * moved past the mark thus STRATMARK is
7614                  * cleared. However, if a message arrived in
7615                  * strrput during the copyout above causing
7616                  * STRATMARK to be set we can not clear that
7617                  * flag.
7618                  * XXX A "perimeter" would help by single-threading strrput,
7619                  * strread, strgetmsg and kstrgetmsg.
7620                  */
7621                 if (type >= QPCTL) {
7622                         ASSERT(type == M_PCPROTO);
7623                         stp->sd_flag &= ~STRPRI;
7624                 }
7625                 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7626                         if (mark & MSGMARKNEXT) {
7627                                 stp->sd_flag &= ~STRNOTATMARK;
7628                                 stp->sd_flag |= STRATMARK;
7629                         } else if (mark & MSGNOTMARKNEXT) {
7630                                 stp->sd_flag &= ~STRATMARK;
7631                                 stp->sd_flag |= STRNOTATMARK;
7632                         } else {
7633                                 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7634                         }
7635                 } else if (pr && (old_sd_flag & STRATMARK)) {
7636                         stp->sd_flag &= ~STRATMARK;
7637                 }
7638         }
7639 
7640         *flagsp = flg;
7641         *prip = pri;
7642 
7643         /*
7644          * Getmsg cleanup processing - if the state of the queue has changed
7645          * some signals may need to be sent and/or poll awakened.
7646          */
7647 getmout:
7648         qbackenable(q, pri);
7649 
7650         /*
7651          * We dropped the stream head lock above. Send all M_SIG messages
7652          * before processing stream head for SIGPOLL messages.
7653          */
7654         ASSERT(MUTEX_HELD(&stp->sd_lock));
7655         while ((bp = q->q_first) != NULL &&
7656             (bp->b_datap->db_type == M_SIG)) {
7657                 /*
7658                  * sd_lock is held so the content of the read queue can not
7659                  * change.
7660                  */
7661                 bp = getq(q);
7662                 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7663 
7664                 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7665                 mutex_exit(&stp->sd_lock);
7666                 freemsg(bp);
7667                 if (STREAM_NEEDSERVICE(stp))
7668                         stream_runservice(stp);
7669                 mutex_enter(&stp->sd_lock);
7670         }
7671 
7672         /*
7673          * stream head cannot change while we make the determination
7674          * whether or not to send a signal. Drop the flag to allow strrput
7675          * to send firstmsgsigs again.
7676          */
7677         stp->sd_flag &= ~STRGETINPROG;
7678 
7679         /*
7680          * If the type of message at the front of the queue changed
7681          * due to the receive the appropriate signals and pollwakeup events
7682          * are generated. The type of changes are:
7683          *      Processed a hipri message, q_first is not hipri.
7684          *      Processed a band X message, and q_first is band Y.
7685          * The generated signals and pollwakeups are identical to what
7686          * strrput() generates should the message that is now on q_first
7687          * arrive to an empty read queue.
7688          *
7689          * Note: only strrput will send a signal for a hipri message.
7690          */
7691         if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7692                 strsigset_t signals = 0;
7693                 strpollset_t pollwakeups = 0;
7694 
7695                 if (flg & MSG_HIPRI) {
7696                         /*
7697                          * Removed a hipri message. Regular data at
7698                          * the front of  the queue.
7699                          */
7700                         if (bp->b_band == 0) {
7701                                 signals = S_INPUT | S_RDNORM;
7702                                 pollwakeups = POLLIN | POLLRDNORM;
7703                         } else {
7704                                 signals = S_INPUT | S_RDBAND;
7705                                 pollwakeups = POLLIN | POLLRDBAND;
7706                         }
7707                 } else if (pri != bp->b_band) {
7708                         /*
7709                          * The band is different for the new q_first.
7710                          */
7711                         if (bp->b_band == 0) {
7712                                 signals = S_RDNORM;
7713                                 pollwakeups = POLLIN | POLLRDNORM;
7714                         } else {
7715                                 signals = S_RDBAND;
7716                                 pollwakeups = POLLIN | POLLRDBAND;
7717                         }
7718                 }
7719 
7720                 if (pollwakeups != 0) {
7721                         if (pollwakeups == (POLLIN | POLLRDNORM)) {
7722                                 if (!(stp->sd_rput_opt & SR_POLLIN))
7723                                         goto no_pollwake;
7724                                 stp->sd_rput_opt &= ~SR_POLLIN;
7725                         }
7726                         mutex_exit(&stp->sd_lock);
7727                         pollwakeup(&stp->sd_pollist, pollwakeups);
7728                         mutex_enter(&stp->sd_lock);
7729                 }
7730 no_pollwake:
7731 
7732                 if (stp->sd_sigflags & signals)
7733                         strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7734         }
7735         mutex_exit(&stp->sd_lock);
7736 
7737         rvp->r_val1 = more;
7738         return (error);
7739 #undef  _LASTMARK
7740 }
7741 
7742 /*
7743  * Put a message downstream.
7744  *
7745  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7746  */
7747 int
7748 strputmsg(
7749         struct vnode *vp,
7750         struct strbuf *mctl,
7751         struct strbuf *mdata,
7752         unsigned char pri,
7753         int flag,
7754         int fmode)
7755 {
7756         struct stdata *stp;
7757         queue_t *wqp;
7758         mblk_t *mp;
7759         ssize_t msgsize;
7760         ssize_t rmin, rmax;
7761         int error;
7762         struct uio uios;
7763         struct uio *uiop = &uios;
7764         struct iovec iovs;
7765         int xpg4 = 0;
7766 
7767         ASSERT(vp->v_stream);
7768         stp = vp->v_stream;
7769         wqp = stp->sd_wrq;
7770 
7771         /*
7772          * If it is an XPG4 application, we need to send
7773          * SIGPIPE below
7774          */
7775 
7776         xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7777         flag &= ~MSG_XPG4;
7778 
7779         if (AU_AUDITING())
7780                 audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7781 
7782         mutex_enter(&stp->sd_lock);
7783 
7784         if ((error = i_straccess(stp, JCWRITE)) != 0) {
7785                 mutex_exit(&stp->sd_lock);
7786                 return (error);
7787         }
7788 
7789         if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7790                 error = strwriteable(stp, B_FALSE, xpg4);
7791                 if (error != 0) {
7792                         mutex_exit(&stp->sd_lock);
7793                         return (error);
7794                 }
7795         }
7796 
7797         mutex_exit(&stp->sd_lock);
7798 
7799         /*
7800          * Check for legal flag value.
7801          */
7802         switch (flag) {
7803         case MSG_HIPRI:
7804                 if ((mctl->len < 0) || (pri != 0))
7805                         return (EINVAL);
7806                 break;
7807         case MSG_BAND:
7808                 break;
7809 
7810         default:
7811                 return (EINVAL);
7812         }
7813 
7814         TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7815             "strputmsg in:stp %p", stp);
7816 
7817         /* get these values from those cached in the stream head */
7818         rmin = stp->sd_qn_minpsz;
7819         rmax = stp->sd_qn_maxpsz;
7820 
7821         /*
7822          * Make sure ctl and data sizes together fall within the
7823          * limits of the max and min receive packet sizes and do
7824          * not exceed system limit.
7825          */
7826         ASSERT((rmax >= 0) || (rmax == INFPSZ));
7827         if (rmax == 0) {
7828                 return (ERANGE);
7829         }
7830         /*
7831          * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7832          * Needed to prevent partial failures in the strmakedata loop.
7833          */
7834         if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7835                 rmax = stp->sd_maxblk;
7836 
7837         if ((msgsize = mdata->len) < 0) {
7838                 msgsize = 0;
7839                 rmin = 0;       /* no range check for NULL data part */
7840         }
7841         if ((msgsize < rmin) ||
7842             ((msgsize > rmax) && (rmax != INFPSZ)) ||
7843             (mctl->len > strctlsz)) {
7844                 return (ERANGE);
7845         }
7846 
7847         /*
7848          * Setup uio and iov for data part
7849          */
7850         iovs.iov_base = mdata->buf;
7851         iovs.iov_len = msgsize;
7852         uios.uio_iov = &iovs;
7853         uios.uio_iovcnt = 1;
7854         uios.uio_loffset = 0;
7855         uios.uio_segflg = UIO_USERSPACE;
7856         uios.uio_fmode = fmode;
7857         uios.uio_extflg = UIO_COPY_DEFAULT;
7858         uios.uio_resid = msgsize;
7859         uios.uio_offset = 0;
7860 
7861         /* Ignore flow control in strput for HIPRI */
7862         if (flag & MSG_HIPRI)
7863                 flag |= MSG_IGNFLOW;
7864 
7865         for (;;) {
7866                 int done = 0;
7867 
7868                 /*
7869                  * strput will always free the ctl mblk - even when strput
7870                  * fails.
7871                  */
7872                 if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7873                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7874                             "strputmsg out:stp %p out %d error %d",
7875                             stp, 1, error);
7876                         return (error);
7877                 }
7878                 /*
7879                  * Verify that the whole message can be transferred by
7880                  * strput.
7881                  */
7882                 ASSERT(stp->sd_maxblk == INFPSZ ||
7883                     stp->sd_maxblk >= mdata->len);
7884 
7885                 msgsize = mdata->len;
7886                 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7887                 mdata->len = msgsize;
7888 
7889                 if (error == 0)
7890                         break;
7891 
7892                 if (error != EWOULDBLOCK)
7893                         goto out;
7894 
7895                 mutex_enter(&stp->sd_lock);
7896                 /*
7897                  * Check for a missed wakeup.
7898                  * Needed since strput did not hold sd_lock across
7899                  * the canputnext.
7900                  */
7901                 if (bcanputnext(wqp, pri)) {
7902                         /* Try again */
7903                         mutex_exit(&stp->sd_lock);
7904                         continue;
7905                 }
7906                 TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7907                     "strputmsg wait:stp %p waits pri %d", stp, pri);
7908                 if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7909                     &done)) != 0) || done) {
7910                         mutex_exit(&stp->sd_lock);
7911                         TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7912                             "strputmsg out:q %p out %d error %d",
7913                             stp, 0, error);
7914                         return (error);
7915                 }
7916                 TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7917                     "strputmsg wake:stp %p wakes", stp);
7918                 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7919                         mutex_exit(&stp->sd_lock);
7920                         return (error);
7921                 }
7922                 mutex_exit(&stp->sd_lock);
7923         }
7924 out:
7925         /*
7926          * For historic reasons, applications expect EAGAIN
7927          * when data mblk could not be allocated. so change
7928          * ENOMEM back to EAGAIN
7929          */
7930         if (error == ENOMEM)
7931                 error = EAGAIN;
7932         TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7933             "strputmsg out:stp %p out %d error %d", stp, 2, error);
7934         return (error);
7935 }
7936 
7937 /*
7938  * Put a message downstream.
7939  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7940  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7941  * and the fmode parameter.
7942  *
7943  * This routine handles the consolidation private flags:
7944  *      MSG_IGNERROR    Ignore any stream head error except STPLEX.
7945  *      MSG_HOLDSIG     Hold signals while waiting for data.
7946  *      MSG_IGNFLOW     Don't check streams flow control.
7947  *
7948  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7949  */
7950 int
7951 kstrputmsg(
7952         struct vnode *vp,
7953         mblk_t *mctl,
7954         struct uio *uiop,
7955         ssize_t msgsize,
7956         unsigned char pri,
7957         int flag,
7958         int fmode)
7959 {
7960         struct stdata *stp;
7961         queue_t *wqp;
7962         ssize_t rmin, rmax;
7963         int error;
7964 
7965         ASSERT(vp->v_stream);
7966         stp = vp->v_stream;
7967         wqp = stp->sd_wrq;
7968         if (AU_AUDITING())
7969                 audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7970         if (mctl == NULL)
7971                 return (EINVAL);
7972 
7973         mutex_enter(&stp->sd_lock);
7974 
7975         if ((error = i_straccess(stp, JCWRITE)) != 0) {
7976                 mutex_exit(&stp->sd_lock);
7977                 freemsg(mctl);
7978                 return (error);
7979         }
7980 
7981         if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7982                 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7983                         error = strwriteable(stp, B_FALSE, B_TRUE);
7984                         if (error != 0) {
7985                                 mutex_exit(&stp->sd_lock);
7986                                 freemsg(mctl);
7987                                 return (error);
7988                         }
7989                 }
7990         }
7991 
7992         mutex_exit(&stp->sd_lock);
7993 
7994         /*
7995          * Check for legal flag value.
7996          */
7997         switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
7998         case MSG_HIPRI:
7999                 if (pri != 0) {
8000                         freemsg(mctl);
8001                         return (EINVAL);
8002                 }
8003                 break;
8004         case MSG_BAND:
8005                 break;
8006         default:
8007                 freemsg(mctl);
8008                 return (EINVAL);
8009         }
8010 
8011         TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
8012             "kstrputmsg in:stp %p", stp);
8013 
8014         /* get these values from those cached in the stream head */
8015         rmin = stp->sd_qn_minpsz;
8016         rmax = stp->sd_qn_maxpsz;
8017 
8018         /*
8019          * Make sure ctl and data sizes together fall within the
8020          * limits of the max and min receive packet sizes and do
8021          * not exceed system limit.
8022          */
8023         ASSERT((rmax >= 0) || (rmax == INFPSZ));
8024         if (rmax == 0) {
8025                 freemsg(mctl);
8026                 return (ERANGE);
8027         }
8028         /*
8029          * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8030          * Needed to prevent partial failures in the strmakedata loop.
8031          */
8032         if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8033                 rmax = stp->sd_maxblk;
8034 
8035         if (uiop == NULL) {
8036                 msgsize = -1;
8037                 rmin = -1;      /* no range check for NULL data part */
8038         } else {
8039                 /* Use uio flags as well as the fmode parameter flags */
8040                 fmode |= uiop->uio_fmode;
8041 
8042                 if ((msgsize < rmin) ||
8043                     ((msgsize > rmax) && (rmax != INFPSZ))) {
8044                         freemsg(mctl);
8045                         return (ERANGE);
8046                 }
8047         }
8048 
8049         /* Ignore flow control in strput for HIPRI */
8050         if (flag & MSG_HIPRI)
8051                 flag |= MSG_IGNFLOW;
8052 
8053         for (;;) {
8054                 int done = 0;
8055                 int waitflag;
8056                 mblk_t *mp;
8057 
8058                 /*
8059                  * strput will always free the ctl mblk - even when strput
8060                  * fails. If MSG_IGNFLOW is set then any error returned
8061                  * will cause us to break the loop, so we don't need a copy
8062                  * of the message. If MSG_IGNFLOW is not set, then we can
8063                  * get hit by flow control and be forced to try again. In
8064                  * this case we need to have a copy of the message. We
8065                  * do this using copymsg since the message may get modified
8066                  * by something below us.
8067                  *
8068                  * We've observed that many TPI providers do not check db_ref
8069                  * on the control messages but blindly reuse them for the
8070                  * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8071                  * friendly to such providers than using dupmsg. Also, note
8072                  * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8073                  * Only data messages are subject to flow control, hence
8074                  * subject to this copymsg.
8075                  */
8076                 if (flag & MSG_IGNFLOW) {
8077                         mp = mctl;
8078                         mctl = NULL;
8079                 } else {
8080                         do {
8081                                 /*
8082                                  * If a message has a free pointer, the message
8083                                  * must be dupmsg to maintain this pointer.
8084                                  * Code using this facility must be sure
8085                                  * that modules below will not change the
8086                                  * contents of the dblk without checking db_ref
8087                                  * first. If db_ref is > 1, then the module
8088                                  * needs to do a copymsg first. Otherwise,
8089                                  * the contents of the dblk may become
8090                                  * inconsistent because the freesmg/freeb below
8091                                  * may end up calling atomic_add_32_nv.
8092                                  * The atomic_add_32_nv in freeb (accessing
8093                                  * all of db_ref, db_type, db_flags, and
8094                                  * db_struioflag) does not prevent other threads
8095                                  * from concurrently trying to modify e.g.
8096                                  * db_type.
8097                                  */
8098                                 if (mctl->b_datap->db_frtnp != NULL)
8099                                         mp = dupmsg(mctl);
8100                                 else
8101                                         mp = copymsg(mctl);
8102 
8103                                 if (mp != NULL)
8104                                         break;
8105 
8106                                 error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8107                                 if (error) {
8108                                         freemsg(mctl);
8109                                         return (error);
8110                                 }
8111                         } while (mp == NULL);
8112                 }
8113                 /*
8114                  * Verify that all of msgsize can be transferred by
8115                  * strput.
8116                  */
8117                 ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8118                 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8119                 if (error == 0)
8120                         break;
8121 
8122                 if (error != EWOULDBLOCK)
8123                         goto out;
8124 
8125                 /*
8126                  * IF MSG_IGNFLOW is set we should have broken out of loop
8127                  * above.
8128                  */
8129                 ASSERT(!(flag & MSG_IGNFLOW));
8130                 mutex_enter(&stp->sd_lock);
8131                 /*
8132                  * Check for a missed wakeup.
8133                  * Needed since strput did not hold sd_lock across
8134                  * the canputnext.
8135                  */
8136                 if (bcanputnext(wqp, pri)) {
8137                         /* Try again */
8138                         mutex_exit(&stp->sd_lock);
8139                         continue;
8140                 }
8141                 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8142                     "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8143 
8144                 waitflag = WRITEWAIT;
8145                 if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8146                         if (flag & MSG_HOLDSIG)
8147                                 waitflag |= STR_NOSIG;
8148                         if (flag & MSG_IGNERROR)
8149                                 waitflag |= STR_NOERROR;
8150                 }
8151                 if (((error = strwaitq(stp, waitflag,
8152                     (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8153                         mutex_exit(&stp->sd_lock);
8154                         TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8155                             "kstrputmsg out:stp %p out %d error %d",
8156                             stp, 0, error);
8157                         freemsg(mctl);
8158                         return (error);
8159                 }
8160                 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8161                     "kstrputmsg wake:stp %p wakes", stp);
8162                 if ((error = i_straccess(stp, JCWRITE)) != 0) {
8163                         mutex_exit(&stp->sd_lock);
8164                         freemsg(mctl);
8165                         return (error);
8166                 }
8167                 mutex_exit(&stp->sd_lock);
8168         }
8169 out:
8170         freemsg(mctl);
8171         /*
8172          * For historic reasons, applications expect EAGAIN
8173          * when data mblk could not be allocated. so change
8174          * ENOMEM back to EAGAIN
8175          */
8176         if (error == ENOMEM)
8177                 error = EAGAIN;
8178         TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8179             "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8180         return (error);
8181 }
8182 
8183 /*
8184  * Determines whether the necessary conditions are set on a stream
8185  * for it to be readable, writeable, or have exceptions.
8186  *
8187  * strpoll handles the consolidation private events:
8188  *      POLLNOERR       Do not return POLLERR even if there are stream
8189  *                      head errors.
8190  *                      Used by sockfs.
8191  *      POLLRDDATA      Do not return POLLIN unless at least one message on
8192  *                      the queue contains one or more M_DATA mblks. Thus
8193  *                      when this flag is set a queue with only
8194  *                      M_PROTO/M_PCPROTO mblks does not return POLLIN.
8195  *                      Used by sockfs to ignore T_EXDATA_IND messages.
8196  *
8197  * Note: POLLRDDATA assumes that synch streams only return messages with
8198  * an M_DATA attached (i.e. not messages consisting of only
8199  * an M_PROTO/M_PCPROTO part).
8200  */
8201 int
8202 strpoll(struct stdata *stp, short events_arg, int anyyet, short *reventsp,
8203     struct pollhead **phpp)
8204 {
8205         int events = (ushort_t)events_arg;
8206         int retevents = 0;
8207         mblk_t *mp;
8208         qband_t *qbp;
8209         long sd_flags = stp->sd_flag;
8210         int headlocked = 0;
8211 
8212         /*
8213          * For performance, a single 'if' tests for most possible edge
8214          * conditions in one shot
8215          */
8216         if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8217                 if (sd_flags & STPLEX) {
8218                         *reventsp = POLLNVAL;
8219                         return (EINVAL);
8220                 }
8221                 if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8222                     (sd_flags & STRDERR)) ||
8223                     ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8224                     (sd_flags & STWRERR))) {
8225                         if (!(events & POLLNOERR)) {
8226                                 *reventsp = POLLERR;
8227                                 return (0);
8228                         }
8229                 }
8230         }
8231         if (sd_flags & STRHUP) {
8232                 retevents |= POLLHUP;
8233         } else if (events & (POLLWRNORM | POLLWRBAND)) {
8234                 queue_t *tq;
8235                 queue_t *qp = stp->sd_wrq;
8236 
8237                 claimstr(qp);
8238                 /* Find next module forward that has a service procedure */
8239                 tq = qp->q_next->q_nfsrv;
8240                 ASSERT(tq != NULL);
8241 
8242                 if (polllock(&stp->sd_pollist, QLOCK(tq)) != 0) {
8243                         releasestr(qp);
8244                         *reventsp = POLLNVAL;
8245                         return (0);
8246                 }
8247                 if (events & POLLWRNORM) {
8248                         queue_t *sqp;
8249 
8250                         if (tq->q_flag & QFULL)
8251                                 /* ensure backq svc procedure runs */
8252                                 tq->q_flag |= QWANTW;
8253                         else if ((sqp = stp->sd_struiowrq) != NULL) {
8254                                 /* Check sync stream barrier write q */
8255                                 mutex_exit(QLOCK(tq));
8256                                 if (polllock(&stp->sd_pollist,
8257                                     QLOCK(sqp)) != 0) {
8258                                         releasestr(qp);
8259                                         *reventsp = POLLNVAL;
8260                                         return (0);
8261                                 }
8262                                 if (sqp->q_flag & QFULL)
8263                                         /* ensure pollwakeup() is done */
8264                                         sqp->q_flag |= QWANTWSYNC;
8265                                 else
8266                                         retevents |= POLLOUT;
8267                                 /* More write events to process ??? */
8268                                 if (! (events & POLLWRBAND)) {
8269                                         mutex_exit(QLOCK(sqp));
8270                                         releasestr(qp);
8271                                         goto chkrd;
8272                                 }
8273                                 mutex_exit(QLOCK(sqp));
8274                                 if (polllock(&stp->sd_pollist,
8275                                     QLOCK(tq)) != 0) {
8276                                         releasestr(qp);
8277                                         *reventsp = POLLNVAL;
8278                                         return (0);
8279                                 }
8280                         } else
8281                                 retevents |= POLLOUT;
8282                 }
8283                 if (events & POLLWRBAND) {
8284                         qbp = tq->q_bandp;
8285                         if (qbp) {
8286                                 while (qbp) {
8287                                         if (qbp->qb_flag & QB_FULL)
8288                                                 qbp->qb_flag |= QB_WANTW;
8289                                         else
8290                                                 retevents |= POLLWRBAND;
8291                                         qbp = qbp->qb_next;
8292                                 }
8293                         } else {
8294                                 retevents |= POLLWRBAND;
8295                         }
8296                 }
8297                 mutex_exit(QLOCK(tq));
8298                 releasestr(qp);
8299         }
8300 chkrd:
8301         if (sd_flags & STRPRI) {
8302                 retevents |= (events & POLLPRI);
8303         } else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8304                 queue_t *qp = _RD(stp->sd_wrq);
8305                 int normevents = (events & (POLLIN | POLLRDNORM));
8306 
8307                 /*
8308                  * Note: Need to do polllock() here since ps_lock may be
8309                  * held. See bug 4191544.
8310                  */
8311                 if (polllock(&stp->sd_pollist, &stp->sd_lock) != 0) {
8312                         *reventsp = POLLNVAL;
8313                         return (0);
8314                 }
8315                 headlocked = 1;
8316                 mp = qp->q_first;
8317                 while (mp) {
8318                         /*
8319                          * For POLLRDDATA we scan b_cont and b_next until we
8320                          * find an M_DATA.
8321                          */
8322                         if ((events & POLLRDDATA) &&
8323                             mp->b_datap->db_type != M_DATA) {
8324                                 mblk_t *nmp = mp->b_cont;
8325 
8326                                 while (nmp != NULL &&
8327                                     nmp->b_datap->db_type != M_DATA)
8328                                         nmp = nmp->b_cont;
8329                                 if (nmp == NULL) {
8330                                         mp = mp->b_next;
8331                                         continue;
8332                                 }
8333                         }
8334                         if (mp->b_band == 0)
8335                                 retevents |= normevents;
8336                         else
8337                                 retevents |= (events & (POLLIN | POLLRDBAND));
8338                         break;
8339                 }
8340                 if (!(retevents & normevents) && (stp->sd_wakeq & RSLEEP)) {
8341                         /*
8342                          * Sync stream barrier read queue has data.
8343                          */
8344                         retevents |= normevents;
8345                 }
8346                 /* Treat eof as normal data */
8347                 if (sd_flags & STREOF)
8348                         retevents |= normevents;
8349         }
8350 
8351         /*
8352          * Pass back a pollhead if no events are pending or if edge-triggering
8353          * has been configured on this resource.
8354          */
8355         if ((retevents == 0 && !anyyet) || (events & POLLET)) {
8356                 *phpp = &stp->sd_pollist;
8357                 if (headlocked == 0) {
8358                         if (polllock(&stp->sd_pollist, &stp->sd_lock) != 0) {
8359                                 *reventsp = POLLNVAL;
8360                                 return (0);
8361                         }
8362                         headlocked = 1;
8363                 }
8364                 stp->sd_rput_opt |= SR_POLLIN;
8365         }
8366 
8367         *reventsp = (short)retevents;
8368         if (headlocked)
8369                 mutex_exit(&stp->sd_lock);
8370         return (0);
8371 }
8372 
8373 /*
8374  * The purpose of putback() is to assure sleeping polls/reads
8375  * are awakened when there are no new messages arriving at the,
8376  * stream head, and a message is placed back on the read queue.
8377  *
8378  * sd_lock must be held when messages are placed back on stream
8379  * head.  (getq() holds sd_lock when it removes messages from
8380  * the queue)
8381  */
8382 
8383 static void
8384 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8385 {
8386         mblk_t  *qfirst;
8387         ASSERT(MUTEX_HELD(&stp->sd_lock));
8388 
8389         /*
8390          * As a result of lock-step ordering around q_lock and sd_lock,
8391          * it's possible for function calls like putnext() and
8392          * canputnext() to get an inaccurate picture of how much
8393          * data is really being processed at the stream head.
8394          * We only consolidate with existing messages on the queue
8395          * if the length of the message we want to put back is smaller
8396          * than the queue hiwater mark.
8397          */
8398         if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8399             (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8400             (DB_TYPE(qfirst) == M_DATA) &&
8401             ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8402             ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8403             (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8404                 /*
8405                  * We use the same logic as defined in strrput()
8406                  * but in reverse as we are putting back onto the
8407                  * queue and want to retain byte ordering.
8408                  * Consolidate M_DATA messages with M_DATA ONLY.
8409                  * strrput() allows the consolidation of M_DATA onto
8410                  * M_PROTO | M_PCPROTO but not the other way round.
8411                  *
8412                  * The consolidation does not take place if the message
8413                  * we are returning to the queue is marked with either
8414                  * of the marks or the delim flag or if q_first
8415                  * is marked with MSGMARK. The MSGMARK check is needed to
8416                  * handle the odd semantics of MSGMARK where essentially
8417                  * the whole message is to be treated as marked.
8418                  * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8419                  * to the front of the b_cont chain.
8420                  */
8421                 rmvq_noenab(q, qfirst);
8422 
8423                 /*
8424                  * The first message in the b_cont list
8425                  * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8426                  * We need to handle the case where we
8427                  * are appending:
8428                  *
8429                  * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8430                  * 2) a MSGMARKNEXT to a plain message.
8431                  * 3) a MSGNOTMARKNEXT to a plain message
8432                  * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8433                  *    message.
8434                  *
8435                  * Thus we never append a MSGMARKNEXT or
8436                  * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8437                  */
8438                 if (qfirst->b_flag & MSGMARKNEXT) {
8439                         bp->b_flag |= MSGMARKNEXT;
8440                         bp->b_flag &= ~MSGNOTMARKNEXT;
8441                         qfirst->b_flag &= ~MSGMARKNEXT;
8442                 } else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8443                         bp->b_flag |= MSGNOTMARKNEXT;
8444                         qfirst->b_flag &= ~MSGNOTMARKNEXT;
8445                 }
8446 
8447                 linkb(bp, qfirst);
8448         }
8449         (void) putbq(q, bp);
8450 
8451         /*
8452          * A message may have come in when the sd_lock was dropped in the
8453          * calling routine. If this is the case and STR*ATMARK info was
8454          * received, need to move that from the stream head to the q_last
8455          * so that SIOCATMARK can return the proper value.
8456          */
8457         if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8458                 unsigned short *flagp = &q->q_last->b_flag;
8459                 uint_t b_flag = (uint_t)*flagp;
8460 
8461                 if (stp->sd_flag & STRATMARK) {
8462                         b_flag &= ~MSGNOTMARKNEXT;
8463                         b_flag |= MSGMARKNEXT;
8464                         stp->sd_flag &= ~STRATMARK;
8465                 } else {
8466                         b_flag &= ~MSGMARKNEXT;
8467                         b_flag |= MSGNOTMARKNEXT;
8468                         stp->sd_flag &= ~STRNOTATMARK;
8469                 }
8470                 *flagp = (unsigned short) b_flag;
8471         }
8472 
8473 #ifdef  DEBUG
8474         /*
8475          * Make sure that the flags are not messed up.
8476          */
8477         {
8478                 mblk_t *mp;
8479                 mp = q->q_last;
8480                 while (mp != NULL) {
8481                         ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8482                             (MSGMARKNEXT|MSGNOTMARKNEXT));
8483                         mp = mp->b_cont;
8484                 }
8485         }
8486 #endif
8487         if (q->q_first == bp) {
8488                 short pollevents;
8489 
8490                 if (stp->sd_flag & RSLEEP) {
8491                         stp->sd_flag &= ~RSLEEP;
8492                         cv_broadcast(&q->q_wait);
8493                 }
8494                 if (stp->sd_flag & STRPRI) {
8495                         pollevents = POLLPRI;
8496                 } else {
8497                         if (band == 0) {
8498                                 if (!(stp->sd_rput_opt & SR_POLLIN))
8499                                         return;
8500                                 stp->sd_rput_opt &= ~SR_POLLIN;
8501                                 pollevents = POLLIN | POLLRDNORM;
8502                         } else {
8503                                 pollevents = POLLIN | POLLRDBAND;
8504                         }
8505                 }
8506                 mutex_exit(&stp->sd_lock);
8507                 pollwakeup(&stp->sd_pollist, pollevents);
8508                 mutex_enter(&stp->sd_lock);
8509         }
8510 }
8511 
8512 /*
8513  * Return the held vnode attached to the stream head of a
8514  * given queue
8515  * It is the responsibility of the calling routine to ensure
8516  * that the queue does not go away (e.g. pop).
8517  */
8518 vnode_t *
8519 strq2vp(queue_t *qp)
8520 {
8521         vnode_t *vp;
8522         vp = STREAM(qp)->sd_vnode;
8523         ASSERT(vp != NULL);
8524         VN_HOLD(vp);
8525         return (vp);
8526 }
8527 
8528 /*
8529  * return the stream head write queue for the given vp
8530  * It is the responsibility of the calling routine to ensure
8531  * that the stream or vnode do not close.
8532  */
8533 queue_t *
8534 strvp2wq(vnode_t *vp)
8535 {
8536         ASSERT(vp->v_stream != NULL);
8537         return (vp->v_stream->sd_wrq);
8538 }
8539 
8540 /*
8541  * pollwakeup stream head
8542  * It is the responsibility of the calling routine to ensure
8543  * that the stream or vnode do not close.
8544  */
8545 void
8546 strpollwakeup(vnode_t *vp, short event)
8547 {
8548         ASSERT(vp->v_stream);
8549         pollwakeup(&vp->v_stream->sd_pollist, event);
8550 }
8551 
8552 /*
8553  * Mate the stream heads of two vnodes together. If the two vnodes are the
8554  * same, we just make the write-side point at the read-side -- otherwise,
8555  * we do a full mate.  Only works on vnodes associated with streams that are
8556  * still being built and thus have only a stream head.
8557  */
8558 void
8559 strmate(vnode_t *vp1, vnode_t *vp2)
8560 {
8561         queue_t *wrq1 = strvp2wq(vp1);
8562         queue_t *wrq2 = strvp2wq(vp2);
8563 
8564         /*
8565          * Verify that there are no modules on the stream yet.  We also
8566          * rely on the stream head always having a service procedure to
8567          * avoid tweaking q_nfsrv.
8568          */
8569         ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8570         ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8571         ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8572 
8573         /*
8574          * If the queues are the same, just twist; otherwise do a full mate.
8575          */
8576         if (wrq1 == wrq2) {
8577                 wrq1->q_next = _RD(wrq1);
8578         } else {
8579                 wrq1->q_next = _RD(wrq2);
8580                 wrq2->q_next = _RD(wrq1);
8581                 STREAM(wrq1)->sd_mate = STREAM(wrq2);
8582                 STREAM(wrq1)->sd_flag |= STRMATE;
8583                 STREAM(wrq2)->sd_mate = STREAM(wrq1);
8584                 STREAM(wrq2)->sd_flag |= STRMATE;
8585         }
8586 }
8587 
8588 /*
8589  * XXX will go away when console is correctly fixed.
8590  * Clean up the console PIDS, from previous I_SETSIG,
8591  * called only for cnopen which never calls strclean().
8592  */
8593 void
8594 str_cn_clean(struct vnode *vp)
8595 {
8596         strsig_t *ssp, *pssp, *tssp;
8597         struct stdata *stp;
8598         struct pid  *pidp;
8599         int update = 0;
8600 
8601         ASSERT(vp->v_stream);
8602         stp = vp->v_stream;
8603         pssp = NULL;
8604         mutex_enter(&stp->sd_lock);
8605         ssp = stp->sd_siglist;
8606         while (ssp) {
8607                 mutex_enter(&pidlock);
8608                 pidp = ssp->ss_pidp;
8609                 /*
8610                  * Get rid of PID if the proc is gone.
8611                  */
8612                 if (pidp->pid_prinactive) {
8613                         tssp = ssp->ss_next;
8614                         if (pssp)
8615                                 pssp->ss_next = tssp;
8616                         else
8617                                 stp->sd_siglist = tssp;
8618                         ASSERT(pidp->pid_ref <= 1);
8619                         PID_RELE(ssp->ss_pidp);
8620                         mutex_exit(&pidlock);
8621                         kmem_free(ssp, sizeof (strsig_t));
8622                         update = 1;
8623                         ssp = tssp;
8624                         continue;
8625                 } else
8626                         mutex_exit(&pidlock);
8627                 pssp = ssp;
8628                 ssp = ssp->ss_next;
8629         }
8630         if (update) {
8631                 stp->sd_sigflags = 0;
8632                 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8633                         stp->sd_sigflags |= ssp->ss_events;
8634         }
8635         mutex_exit(&stp->sd_lock);
8636 }
8637 
8638 /*
8639  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8640  */
8641 static boolean_t
8642 msghasdata(mblk_t *bp)
8643 {
8644         for (; bp; bp = bp->b_cont)
8645                 if (bp->b_datap->db_type == M_DATA) {
8646                         ASSERT(bp->b_wptr >= bp->b_rptr);
8647                         if (bp->b_wptr > bp->b_rptr)
8648                                 return (B_TRUE);
8649                 }
8650         return (B_FALSE);
8651 }