Print this page
Bring back LX zones.
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/os/streamio.c
+++ new/usr/src/uts/common/os/streamio.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
22 22 /* All Rights Reserved */
23 23
24 24
25 25 /*
26 26 * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
27 27 */
28 28
29 29 #include <sys/types.h>
30 30 #include <sys/sysmacros.h>
31 31 #include <sys/param.h>
32 32 #include <sys/errno.h>
33 33 #include <sys/signal.h>
34 34 #include <sys/stat.h>
35 35 #include <sys/proc.h>
36 36 #include <sys/cred.h>
37 37 #include <sys/user.h>
38 38 #include <sys/vnode.h>
39 39 #include <sys/file.h>
40 40 #include <sys/stream.h>
41 41 #include <sys/strsubr.h>
42 42 #include <sys/stropts.h>
43 43 #include <sys/tihdr.h>
44 44 #include <sys/var.h>
45 45 #include <sys/poll.h>
46 46 #include <sys/termio.h>
47 47 #include <sys/ttold.h>
48 48 #include <sys/systm.h>
49 49 #include <sys/uio.h>
50 50 #include <sys/cmn_err.h>
51 51 #include <sys/sad.h>
52 52 #include <sys/netstack.h>
53 53 #include <sys/priocntl.h>
54 54 #include <sys/jioctl.h>
55 55 #include <sys/procset.h>
56 56 #include <sys/session.h>
57 57 #include <sys/kmem.h>
58 58 #include <sys/filio.h>
59 59 #include <sys/vtrace.h>
60 60 #include <sys/debug.h>
61 61 #include <sys/strredir.h>
62 62 #include <sys/fs/fifonode.h>
63 63 #include <sys/fs/snode.h>
64 64 #include <sys/strlog.h>
65 65 #include <sys/strsun.h>
66 66 #include <sys/project.h>
67 67 #include <sys/kbio.h>
68 68 #include <sys/msio.h>
69 69 #include <sys/tty.h>
70 70 #include <sys/ptyvar.h>
71 71 #include <sys/vuid_event.h>
72 72 #include <sys/modctl.h>
73 73 #include <sys/sunddi.h>
74 74 #include <sys/sunldi_impl.h>
75 75 #include <sys/autoconf.h>
76 76 #include <sys/policy.h>
77 77 #include <sys/dld.h>
78 78 #include <sys/zone.h>
79 79 #include <c2/audit.h>
80 80
81 81 /*
82 82 * This define helps improve the readability of streams code while
83 83 * still maintaining a very old streams performance enhancement. The
84 84 * performance enhancement basically involved having all callers
85 85 * of straccess() perform the first check that straccess() will do
86 86 * locally before actually calling straccess(). (There by reducing
87 87 * the number of unnecessary calls to straccess().)
88 88 */
89 89 #define i_straccess(x, y) ((stp->sd_sidp == NULL) ? 0 : \
90 90 (stp->sd_vnode->v_type == VFIFO) ? 0 : \
91 91 straccess((x), (y)))
92 92
93 93 /*
94 94 * what is mblk_pull_len?
95 95 *
96 96 * If a streams message consists of many short messages,
97 97 * a performance degradation occurs from copyout overhead.
98 98 * To decrease the per mblk overhead, messages that are
99 99 * likely to consist of many small mblks are pulled up into
100 100 * one continuous chunk of memory.
101 101 *
102 102 * To avoid the processing overhead of examining every
103 103 * mblk, a quick heuristic is used. If the first mblk in
104 104 * the message is shorter than mblk_pull_len, it is likely
105 105 * that the rest of the mblk will be short.
106 106 *
107 107 * This heuristic was decided upon after performance tests
108 108 * indicated that anything more complex slowed down the main
109 109 * code path.
110 110 */
111 111 #define MBLK_PULL_LEN 64
112 112 uint32_t mblk_pull_len = MBLK_PULL_LEN;
113 113
114 114 /*
115 115 * The sgttyb_handling flag controls the handling of the old BSD
116 116 * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
117 117 *
118 118 * 0 - Emit no warnings at all and retain old, broken behavior.
119 119 * 1 - Emit no warnings and silently handle new semantics.
120 120 * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
121 121 * (once per system invocation). Handle with new semantics.
122 122 * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
123 123 * made (so that offenders drop core and are easy to debug).
124 124 *
125 125 * The "new semantics" are that TIOCGETP returns B38400 for
126 126 * sg_[io]speed if the corresponding value is over B38400, and that
127 127 * TIOCSET[PN] accept B38400 in these cases to mean "retain current
128 128 * bit rate."
129 129 */
130 130 int sgttyb_handling = 1;
131 131 static boolean_t sgttyb_complaint;
132 132
133 133 /* don't push drcompat module by default on Style-2 streams */
134 134 static int push_drcompat = 0;
135 135
136 136 /*
137 137 * id value used to distinguish between different ioctl messages
138 138 */
139 139 static uint32_t ioc_id;
140 140
141 141 static void putback(struct stdata *, queue_t *, mblk_t *, int);
142 142 static void strcleanall(struct vnode *);
143 143 static int strwsrv(queue_t *);
144 144 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
145 145
146 146 /*
147 147 * qinit and module_info structures for stream head read and write queues
148 148 */
149 149 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
150 150 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
151 151 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
152 152 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
153 153 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
154 154 FIFOLOWAT };
155 155 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
156 156 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
157 157 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
158 158
159 159 extern kmutex_t strresources; /* protects global resources */
160 160 extern kmutex_t muxifier; /* single-threads multiplexor creation */
161 161
162 162 static boolean_t msghasdata(mblk_t *bp);
163 163 #define msgnodata(bp) (!msghasdata(bp))
164 164
165 165 /*
166 166 * Stream head locking notes:
167 167 * There are four monitors associated with the stream head:
168 168 * 1. v_stream monitor: in stropen() and strclose() v_lock
169 169 * is held while the association of vnode and stream
170 170 * head is established or tested for.
171 171 * 2. open/close/push/pop monitor: sd_lock is held while each
172 172 * thread bids for exclusive access to this monitor
173 173 * for opening or closing a stream. In addition, this
174 174 * monitor is entered during pushes and pops. This
175 175 * guarantees that during plumbing operations there
176 176 * is only one thread trying to change the plumbing.
177 177 * Any other threads present in the stream are only
178 178 * using the plumbing.
179 179 * 3. read/write monitor: in the case of read, a thread holds
180 180 * sd_lock while trying to get data from the stream
181 181 * head queue. if there is none to fulfill a read
182 182 * request, it sets RSLEEP and calls cv_wait_sig() down
183 183 * in strwaitq() to await the arrival of new data.
184 184 * when new data arrives in strrput(), sd_lock is acquired
185 185 * before testing for RSLEEP and calling cv_broadcast().
186 186 * the behavior of strwrite(), strwsrv(), and WSLEEP
187 187 * mirror this.
188 188 * 4. ioctl monitor: sd_lock is gotten to ensure that only one
189 189 * thread is doing an ioctl at a time.
190 190 */
191 191
192 192 static int
193 193 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
194 194 int anchor, cred_t *crp, uint_t anchor_zoneid)
195 195 {
196 196 int error;
197 197 fmodsw_impl_t *fp;
198 198
199 199 if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
200 200 error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
201 201 return (error);
202 202 }
203 203 if (stp->sd_pushcnt >= nstrpush) {
204 204 return (EINVAL);
205 205 }
206 206
207 207 if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
208 208 stp->sd_flag |= STREOPENFAIL;
209 209 return (EINVAL);
210 210 }
211 211
212 212 /*
213 213 * push new module and call its open routine via qattach
214 214 */
215 215 if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
216 216 return (error);
217 217
218 218 /*
219 219 * Check to see if caller wants a STREAMS anchor
220 220 * put at this place in the stream, and add if so.
221 221 */
222 222 mutex_enter(&stp->sd_lock);
223 223 if (anchor == stp->sd_pushcnt) {
224 224 stp->sd_anchor = stp->sd_pushcnt;
225 225 stp->sd_anchorzone = anchor_zoneid;
226 226 }
227 227 mutex_exit(&stp->sd_lock);
228 228
229 229 return (0);
230 230 }
231 231
232 232 /*
233 233 * Open a stream device.
234 234 */
235 235 int
236 236 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
237 237 {
238 238 struct stdata *stp;
239 239 queue_t *qp;
240 240 int s;
241 241 dev_t dummydev, savedev;
242 242 struct autopush *ap;
243 243 struct dlautopush dlap;
244 244 int error = 0;
245 245 ssize_t rmin, rmax;
246 246 int cloneopen;
247 247 queue_t *brq;
248 248 major_t major;
249 249 str_stack_t *ss;
250 250 zoneid_t zoneid;
251 251 uint_t anchor;
252 252
253 253 /*
254 254 * If the stream already exists, wait for any open in progress
255 255 * to complete, then call the open function of each module and
256 256 * driver in the stream. Otherwise create the stream.
257 257 */
258 258 TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
259 259 retry:
260 260 mutex_enter(&vp->v_lock);
261 261 if ((stp = vp->v_stream) != NULL) {
262 262
263 263 /*
264 264 * Waiting for stream to be created to device
265 265 * due to another open.
266 266 */
267 267 mutex_exit(&vp->v_lock);
268 268
269 269 if (STRMATED(stp)) {
270 270 struct stdata *strmatep = stp->sd_mate;
271 271
272 272 STRLOCKMATES(stp);
273 273 if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
274 274 if (flag & (FNDELAY|FNONBLOCK)) {
275 275 error = EAGAIN;
276 276 mutex_exit(&strmatep->sd_lock);
277 277 goto ckreturn;
278 278 }
279 279 mutex_exit(&stp->sd_lock);
280 280 if (!cv_wait_sig(&strmatep->sd_monitor,
281 281 &strmatep->sd_lock)) {
282 282 error = EINTR;
283 283 mutex_exit(&strmatep->sd_lock);
284 284 mutex_enter(&stp->sd_lock);
285 285 goto ckreturn;
286 286 }
287 287 mutex_exit(&strmatep->sd_lock);
288 288 goto retry;
289 289 }
290 290 if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
291 291 if (flag & (FNDELAY|FNONBLOCK)) {
292 292 error = EAGAIN;
293 293 mutex_exit(&strmatep->sd_lock);
294 294 goto ckreturn;
295 295 }
296 296 mutex_exit(&strmatep->sd_lock);
297 297 if (!cv_wait_sig(&stp->sd_monitor,
298 298 &stp->sd_lock)) {
299 299 error = EINTR;
300 300 goto ckreturn;
301 301 }
302 302 mutex_exit(&stp->sd_lock);
303 303 goto retry;
304 304 }
305 305
306 306 if (stp->sd_flag & (STRDERR|STWRERR)) {
307 307 error = EIO;
308 308 mutex_exit(&strmatep->sd_lock);
309 309 goto ckreturn;
310 310 }
311 311
312 312 stp->sd_flag |= STWOPEN;
313 313 STRUNLOCKMATES(stp);
314 314 } else {
315 315 mutex_enter(&stp->sd_lock);
316 316 if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
317 317 if (flag & (FNDELAY|FNONBLOCK)) {
318 318 error = EAGAIN;
319 319 goto ckreturn;
320 320 }
321 321 if (!cv_wait_sig(&stp->sd_monitor,
322 322 &stp->sd_lock)) {
323 323 error = EINTR;
324 324 goto ckreturn;
325 325 }
326 326 mutex_exit(&stp->sd_lock);
327 327 goto retry; /* could be clone! */
328 328 }
329 329
330 330 if (stp->sd_flag & (STRDERR|STWRERR)) {
331 331 error = EIO;
332 332 goto ckreturn;
333 333 }
334 334
335 335 stp->sd_flag |= STWOPEN;
336 336 mutex_exit(&stp->sd_lock);
337 337 }
338 338
339 339 /*
340 340 * Open all modules and devices down stream to notify
341 341 * that another user is streaming. For modules, set the
342 342 * last argument to MODOPEN and do not pass any open flags.
343 343 * Ignore dummydev since this is not the first open.
344 344 */
345 345 claimstr(stp->sd_wrq);
346 346 qp = stp->sd_wrq;
347 347 while (_SAMESTR(qp)) {
348 348 qp = qp->q_next;
349 349 if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
350 350 break;
351 351 }
352 352 releasestr(stp->sd_wrq);
353 353 mutex_enter(&stp->sd_lock);
354 354 stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
355 355 stp->sd_rerror = 0;
356 356 stp->sd_werror = 0;
357 357 ckreturn:
358 358 cv_broadcast(&stp->sd_monitor);
359 359 mutex_exit(&stp->sd_lock);
360 360 return (error);
361 361 }
362 362
363 363 /*
364 364 * This vnode isn't streaming. SPECFS already
365 365 * checked for multiple vnodes pointing to the
366 366 * same stream, so create a stream to the driver.
367 367 */
368 368 qp = allocq();
369 369 stp = shalloc(qp);
370 370
371 371 /*
372 372 * Initialize stream head. shalloc() has given us
373 373 * exclusive access, and we have the vnode locked;
374 374 * we can do whatever we want with stp.
375 375 */
376 376 stp->sd_flag = STWOPEN;
377 377 stp->sd_siglist = NULL;
378 378 stp->sd_pollist.ph_list = NULL;
379 379 stp->sd_sigflags = 0;
380 380 stp->sd_mark = NULL;
381 381 stp->sd_closetime = STRTIMOUT;
382 382 stp->sd_sidp = NULL;
383 383 stp->sd_pgidp = NULL;
384 384 stp->sd_vnode = vp;
385 385 stp->sd_rerror = 0;
386 386 stp->sd_werror = 0;
387 387 stp->sd_wroff = 0;
388 388 stp->sd_tail = 0;
389 389 stp->sd_iocblk = NULL;
390 390 stp->sd_cmdblk = NULL;
391 391 stp->sd_pushcnt = 0;
392 392 stp->sd_qn_minpsz = 0;
393 393 stp->sd_qn_maxpsz = INFPSZ - 1; /* used to check for initialization */
394 394 stp->sd_maxblk = INFPSZ;
395 395 qp->q_ptr = _WR(qp)->q_ptr = stp;
396 396 STREAM(qp) = STREAM(_WR(qp)) = stp;
397 397 vp->v_stream = stp;
398 398 mutex_exit(&vp->v_lock);
399 399 if (vp->v_type == VFIFO) {
400 400 stp->sd_flag |= OLDNDELAY;
401 401 /*
402 402 * This means, both for pipes and fifos
403 403 * strwrite will send SIGPIPE if the other
404 404 * end is closed. For putmsg it depends
405 405 * on whether it is a XPG4_2 application
406 406 * or not
407 407 */
408 408 stp->sd_wput_opt = SW_SIGPIPE;
409 409
410 410 /* setq might sleep in kmem_alloc - avoid holding locks. */
411 411 setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
412 412 SQ_CI|SQ_CO, B_FALSE);
413 413
414 414 set_qend(qp);
415 415 stp->sd_strtab = fifo_getinfo();
416 416 _WR(qp)->q_nfsrv = _WR(qp);
417 417 qp->q_nfsrv = qp;
418 418 /*
419 419 * Wake up others that are waiting for stream to be created.
420 420 */
421 421 mutex_enter(&stp->sd_lock);
422 422 /*
423 423 * nothing is be pushed on stream yet, so
424 424 * optimized stream head packetsizes are just that
425 425 * of the read queue
426 426 */
427 427 stp->sd_qn_minpsz = qp->q_minpsz;
428 428 stp->sd_qn_maxpsz = qp->q_maxpsz;
429 429 stp->sd_flag &= ~STWOPEN;
430 430 goto fifo_opendone;
431 431 }
432 432 /* setq might sleep in kmem_alloc - avoid holding locks. */
433 433 setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
434 434
435 435 set_qend(qp);
436 436
437 437 /*
438 438 * Open driver and create stream to it (via qattach).
439 439 */
440 440 savedev = *devp;
441 441 cloneopen = (getmajor(*devp) == clone_major);
442 442 if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
443 443 mutex_enter(&vp->v_lock);
444 444 vp->v_stream = NULL;
445 445 mutex_exit(&vp->v_lock);
446 446 mutex_enter(&stp->sd_lock);
447 447 cv_broadcast(&stp->sd_monitor);
448 448 mutex_exit(&stp->sd_lock);
449 449 freeq(_RD(qp));
450 450 shfree(stp);
451 451 return (error);
452 452 }
453 453 /*
454 454 * Set sd_strtab after open in order to handle clonable drivers
455 455 */
456 456 stp->sd_strtab = STREAMSTAB(getmajor(*devp));
457 457
458 458 /*
459 459 * Historical note: dummydev used to be be prior to the initial
460 460 * open (via qattach above), which made the value seen
461 461 * inconsistent between an I_PUSH and an autopush of a module.
462 462 */
463 463 dummydev = *devp;
464 464
465 465 /*
466 466 * For clone open of old style (Q not associated) network driver,
467 467 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
468 468 */
469 469 brq = _RD(_WR(qp)->q_next);
470 470 major = getmajor(*devp);
471 471 if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
472 472 ((brq->q_flag & _QASSOCIATED) == 0)) {
473 473 if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
474 474 cmn_err(CE_WARN, "cannot push " DRMODNAME
475 475 " streams module");
476 476 }
477 477
478 478 if (!NETWORK_DRV(major)) {
479 479 savedev = *devp;
480 480 } else {
481 481 /*
482 482 * For network devices, process differently based on the
483 483 * return value from dld_autopush():
484 484 *
485 485 * 0: the passed-in device points to a GLDv3 datalink with
486 486 * per-link autopush configuration; use that configuration
487 487 * and ignore any per-driver autopush configuration.
488 488 *
489 489 * 1: the passed-in device points to a physical GLDv3
490 490 * datalink without per-link autopush configuration. The
491 491 * passed in device was changed to refer to the actual
492 492 * physical device (if it's not already); we use that new
493 493 * device to look up any per-driver autopush configuration.
494 494 *
495 495 * -1: neither of the above cases applied; use the initial
496 496 * device to look up any per-driver autopush configuration.
497 497 */
498 498 switch (dld_autopush(&savedev, &dlap)) {
499 499 case 0:
500 500 zoneid = crgetzoneid(crp);
501 501 for (s = 0; s < dlap.dap_npush; s++) {
502 502 error = push_mod(qp, &dummydev, stp,
503 503 dlap.dap_aplist[s], dlap.dap_anchor, crp,
504 504 zoneid);
505 505 if (error != 0)
506 506 break;
507 507 }
508 508 goto opendone;
509 509 case 1:
510 510 break;
511 511 case -1:
512 512 savedev = *devp;
513 513 break;
514 514 }
515 515 }
516 516 /*
517 517 * Find the autopush configuration based on "savedev". Start with the
518 518 * global zone. If not found check in the local zone.
519 519 */
520 520 zoneid = GLOBAL_ZONEID;
521 521 retryap:
522 522 ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
523 523 netstack_str;
524 524 if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
525 525 netstack_rele(ss->ss_netstack);
526 526 if (zoneid == GLOBAL_ZONEID) {
527 527 /*
528 528 * None found. Also look in the zone's autopush table.
529 529 */
530 530 zoneid = crgetzoneid(crp);
531 531 if (zoneid != GLOBAL_ZONEID)
532 532 goto retryap;
533 533 }
534 534 goto opendone;
535 535 }
536 536 anchor = ap->ap_anchor;
537 537 zoneid = crgetzoneid(crp);
538 538 for (s = 0; s < ap->ap_npush; s++) {
539 539 error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
540 540 anchor, crp, zoneid);
541 541 if (error != 0)
542 542 break;
543 543 }
544 544 sad_ap_rele(ap, ss);
545 545 netstack_rele(ss->ss_netstack);
546 546
547 547 opendone:
548 548
549 549 /*
550 550 * let specfs know that open failed part way through
551 551 */
552 552 if (error) {
553 553 mutex_enter(&stp->sd_lock);
554 554 stp->sd_flag |= STREOPENFAIL;
555 555 mutex_exit(&stp->sd_lock);
556 556 }
557 557
558 558 /*
559 559 * Wake up others that are waiting for stream to be created.
560 560 */
561 561 mutex_enter(&stp->sd_lock);
562 562 stp->sd_flag &= ~STWOPEN;
563 563
564 564 /*
565 565 * As a performance concern we are caching the values of
566 566 * q_minpsz and q_maxpsz of the module below the stream
567 567 * head in the stream head.
568 568 */
569 569 mutex_enter(QLOCK(stp->sd_wrq->q_next));
570 570 rmin = stp->sd_wrq->q_next->q_minpsz;
571 571 rmax = stp->sd_wrq->q_next->q_maxpsz;
572 572 mutex_exit(QLOCK(stp->sd_wrq->q_next));
573 573
574 574 /* do this processing here as a performance concern */
575 575 if (strmsgsz != 0) {
576 576 if (rmax == INFPSZ)
577 577 rmax = strmsgsz;
578 578 else
579 579 rmax = MIN(strmsgsz, rmax);
580 580 }
581 581
582 582 mutex_enter(QLOCK(stp->sd_wrq));
583 583 stp->sd_qn_minpsz = rmin;
584 584 stp->sd_qn_maxpsz = rmax;
585 585 mutex_exit(QLOCK(stp->sd_wrq));
586 586
587 587 fifo_opendone:
588 588 cv_broadcast(&stp->sd_monitor);
589 589 mutex_exit(&stp->sd_lock);
590 590 return (error);
591 591 }
592 592
593 593 static int strsink(queue_t *, mblk_t *);
594 594 static struct qinit deadrend = {
595 595 strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
596 596 };
597 597 static struct qinit deadwend = {
598 598 NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
599 599 };
600 600
601 601 /*
602 602 * Close a stream.
603 603 * This is called from closef() on the last close of an open stream.
604 604 * Strclean() will already have removed the siglist and pollist
605 605 * information, so all that remains is to remove all multiplexor links
606 606 * for the stream, pop all the modules (and the driver), and free the
607 607 * stream structure.
608 608 */
609 609
610 610 int
611 611 strclose(struct vnode *vp, int flag, cred_t *crp)
612 612 {
613 613 struct stdata *stp;
614 614 queue_t *qp;
615 615 int rval;
616 616 int freestp = 1;
617 617 queue_t *rmq;
618 618
619 619 TRACE_1(TR_FAC_STREAMS_FR,
620 620 TR_STRCLOSE, "strclose:%p", vp);
621 621 ASSERT(vp->v_stream);
622 622
623 623 stp = vp->v_stream;
624 624 ASSERT(!(stp->sd_flag & STPLEX));
625 625 qp = stp->sd_wrq;
626 626
627 627 /*
628 628 * Needed so that strpoll will return non-zero for this fd.
629 629 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
630 630 */
631 631 mutex_enter(&stp->sd_lock);
632 632 stp->sd_flag |= STRHUP;
633 633 mutex_exit(&stp->sd_lock);
634 634
635 635 /*
636 636 * If the registered process or process group did not have an
637 637 * open instance of this stream then strclean would not be
638 638 * called. Thus at the time of closing all remaining siglist entries
639 639 * are removed.
640 640 */
641 641 if (stp->sd_siglist != NULL)
642 642 strcleanall(vp);
643 643
644 644 ASSERT(stp->sd_siglist == NULL);
645 645 ASSERT(stp->sd_sigflags == 0);
646 646
647 647 if (STRMATED(stp)) {
648 648 struct stdata *strmatep = stp->sd_mate;
649 649 int waited = 1;
650 650
651 651 STRLOCKMATES(stp);
652 652 while (waited) {
653 653 waited = 0;
654 654 while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
655 655 mutex_exit(&strmatep->sd_lock);
656 656 cv_wait(&stp->sd_monitor, &stp->sd_lock);
657 657 mutex_exit(&stp->sd_lock);
658 658 STRLOCKMATES(stp);
659 659 waited = 1;
660 660 }
661 661 while (strmatep->sd_flag &
662 662 (STWOPEN|STRCLOSE|STRPLUMB)) {
663 663 mutex_exit(&stp->sd_lock);
664 664 cv_wait(&strmatep->sd_monitor,
665 665 &strmatep->sd_lock);
666 666 mutex_exit(&strmatep->sd_lock);
667 667 STRLOCKMATES(stp);
668 668 waited = 1;
669 669 }
670 670 }
671 671 stp->sd_flag |= STRCLOSE;
672 672 STRUNLOCKMATES(stp);
673 673 } else {
674 674 mutex_enter(&stp->sd_lock);
675 675 stp->sd_flag |= STRCLOSE;
676 676 mutex_exit(&stp->sd_lock);
677 677 }
678 678
679 679 ASSERT(qp->q_first == NULL); /* No more delayed write */
680 680
681 681 /* Check if an I_LINK was ever done on this stream */
682 682 if (stp->sd_flag & STRHASLINKS) {
683 683 netstack_t *ns;
684 684 str_stack_t *ss;
685 685
686 686 ns = netstack_find_by_cred(crp);
687 687 ASSERT(ns != NULL);
688 688 ss = ns->netstack_str;
689 689 ASSERT(ss != NULL);
690 690
691 691 (void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
692 692 netstack_rele(ss->ss_netstack);
693 693 }
694 694
695 695 while (_SAMESTR(qp)) {
696 696 /*
697 697 * Holding sd_lock prevents q_next from changing in
698 698 * this stream.
699 699 */
700 700 mutex_enter(&stp->sd_lock);
701 701 if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
702 702
703 703 /*
704 704 * sleep until awakened by strwsrv() or timeout
705 705 */
706 706 for (;;) {
707 707 mutex_enter(QLOCK(qp->q_next));
708 708 if (!(qp->q_next->q_mblkcnt)) {
709 709 mutex_exit(QLOCK(qp->q_next));
710 710 break;
711 711 }
712 712 stp->sd_flag |= WSLEEP;
713 713
714 714 /* ensure strwsrv gets enabled */
715 715 qp->q_next->q_flag |= QWANTW;
716 716 mutex_exit(QLOCK(qp->q_next));
717 717 /* get out if we timed out or recv'd a signal */
718 718 if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
719 719 stp->sd_closetime, 0) <= 0) {
720 720 break;
721 721 }
722 722 }
723 723 stp->sd_flag &= ~WSLEEP;
724 724 }
725 725 mutex_exit(&stp->sd_lock);
726 726
727 727 rmq = qp->q_next;
728 728 if (rmq->q_flag & QISDRV) {
729 729 ASSERT(!_SAMESTR(rmq));
730 730 wait_sq_svc(_RD(qp)->q_syncq);
731 731 }
732 732
733 733 qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
734 734 }
735 735
736 736 /*
737 737 * Since we call pollwakeup in close() now, the poll list should
738 738 * be empty in most cases. The only exception is the layered devices
739 739 * (e.g. the console drivers with redirection modules pushed on top
740 740 * of it). We have to do this after calling qdetach() because
741 741 * the redirection module won't have torn down the console
742 742 * redirection until after qdetach() has been invoked.
743 743 */
744 744 if (stp->sd_pollist.ph_list != NULL) {
745 745 pollwakeup(&stp->sd_pollist, POLLERR);
746 746 pollhead_clean(&stp->sd_pollist);
747 747 }
748 748 ASSERT(stp->sd_pollist.ph_list == NULL);
749 749 ASSERT(stp->sd_sidp == NULL);
750 750 ASSERT(stp->sd_pgidp == NULL);
751 751
752 752 /* Prevent qenable from re-enabling the stream head queue */
753 753 disable_svc(_RD(qp));
754 754
755 755 /*
756 756 * Wait until service procedure of each queue is
757 757 * run, if QINSERVICE is set.
758 758 */
759 759 wait_svc(_RD(qp));
760 760
761 761 /*
762 762 * Now, flush both queues.
763 763 */
764 764 flushq(_RD(qp), FLUSHALL);
765 765 flushq(qp, FLUSHALL);
766 766
767 767 /*
768 768 * If the write queue of the stream head is pointing to a
769 769 * read queue, we have a twisted stream. If the read queue
770 770 * is alive, convert the stream head queues into a dead end.
771 771 * If the read queue is dead, free the dead pair.
772 772 */
773 773 if (qp->q_next && !_SAMESTR(qp)) {
774 774 if (qp->q_next->q_qinfo == &deadrend) { /* half-closed pipe */
775 775 flushq(qp->q_next, FLUSHALL); /* ensure no message */
776 776 shfree(qp->q_next->q_stream);
777 777 freeq(qp->q_next);
778 778 freeq(_RD(qp));
779 779 } else if (qp->q_next == _RD(qp)) { /* fifo */
780 780 freeq(_RD(qp));
781 781 } else { /* pipe */
782 782 freestp = 0;
783 783 /*
784 784 * The q_info pointers are never accessed when
785 785 * SQLOCK is held.
786 786 */
787 787 ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
788 788 mutex_enter(SQLOCK(qp->q_syncq));
789 789 qp->q_qinfo = &deadwend;
790 790 _RD(qp)->q_qinfo = &deadrend;
791 791 mutex_exit(SQLOCK(qp->q_syncq));
792 792 }
793 793 } else {
794 794 freeq(_RD(qp)); /* free stream head queue pair */
795 795 }
796 796
797 797 mutex_enter(&vp->v_lock);
798 798 if (stp->sd_iocblk) {
799 799 if (stp->sd_iocblk != (mblk_t *)-1) {
800 800 freemsg(stp->sd_iocblk);
801 801 }
802 802 stp->sd_iocblk = NULL;
803 803 }
804 804 stp->sd_vnode = NULL;
805 805 vp->v_stream = NULL;
806 806 mutex_exit(&vp->v_lock);
807 807 mutex_enter(&stp->sd_lock);
808 808 freemsg(stp->sd_cmdblk);
809 809 stp->sd_cmdblk = NULL;
810 810 stp->sd_flag &= ~STRCLOSE;
811 811 cv_broadcast(&stp->sd_monitor);
812 812 mutex_exit(&stp->sd_lock);
813 813
814 814 if (freestp)
815 815 shfree(stp);
816 816 return (0);
817 817 }
818 818
819 819 static int
820 820 strsink(queue_t *q, mblk_t *bp)
821 821 {
822 822 struct copyresp *resp;
823 823
824 824 switch (bp->b_datap->db_type) {
825 825 case M_FLUSH:
826 826 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
827 827 *bp->b_rptr &= ~FLUSHR;
828 828 bp->b_flag |= MSGNOLOOP;
829 829 /*
830 830 * Protect against the driver passing up
831 831 * messages after it has done a qprocsoff.
832 832 */
833 833 if (_OTHERQ(q)->q_next == NULL)
834 834 freemsg(bp);
835 835 else
836 836 qreply(q, bp);
837 837 } else {
838 838 freemsg(bp);
839 839 }
840 840 break;
841 841
842 842 case M_COPYIN:
843 843 case M_COPYOUT:
844 844 if (bp->b_cont) {
845 845 freemsg(bp->b_cont);
846 846 bp->b_cont = NULL;
847 847 }
848 848 bp->b_datap->db_type = M_IOCDATA;
849 849 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
850 850 resp = (struct copyresp *)bp->b_rptr;
851 851 resp->cp_rval = (caddr_t)1; /* failure */
852 852 /*
853 853 * Protect against the driver passing up
854 854 * messages after it has done a qprocsoff.
855 855 */
856 856 if (_OTHERQ(q)->q_next == NULL)
857 857 freemsg(bp);
858 858 else
859 859 qreply(q, bp);
860 860 break;
861 861
862 862 case M_IOCTL:
863 863 if (bp->b_cont) {
864 864 freemsg(bp->b_cont);
865 865 bp->b_cont = NULL;
866 866 }
867 867 bp->b_datap->db_type = M_IOCNAK;
868 868 /*
869 869 * Protect against the driver passing up
870 870 * messages after it has done a qprocsoff.
871 871 */
872 872 if (_OTHERQ(q)->q_next == NULL)
873 873 freemsg(bp);
874 874 else
875 875 qreply(q, bp);
876 876 break;
877 877
878 878 default:
879 879 freemsg(bp);
880 880 break;
881 881 }
882 882
883 883 return (0);
884 884 }
885 885
886 886 /*
887 887 * Clean up after a process when it closes a stream. This is called
888 888 * from closef for all closes, whereas strclose is called only for the
889 889 * last close on a stream. The siglist is scanned for entries for the
890 890 * current process, and these are removed.
891 891 */
892 892 void
893 893 strclean(struct vnode *vp)
894 894 {
895 895 strsig_t *ssp, *pssp, *tssp;
896 896 stdata_t *stp;
897 897 int update = 0;
898 898
899 899 TRACE_1(TR_FAC_STREAMS_FR,
900 900 TR_STRCLEAN, "strclean:%p", vp);
901 901 stp = vp->v_stream;
902 902 pssp = NULL;
903 903 mutex_enter(&stp->sd_lock);
904 904 ssp = stp->sd_siglist;
905 905 while (ssp) {
906 906 if (ssp->ss_pidp == curproc->p_pidp) {
907 907 tssp = ssp->ss_next;
908 908 if (pssp)
909 909 pssp->ss_next = tssp;
910 910 else
911 911 stp->sd_siglist = tssp;
912 912 mutex_enter(&pidlock);
913 913 PID_RELE(ssp->ss_pidp);
914 914 mutex_exit(&pidlock);
915 915 kmem_free(ssp, sizeof (strsig_t));
916 916 update = 1;
917 917 ssp = tssp;
918 918 } else {
919 919 pssp = ssp;
920 920 ssp = ssp->ss_next;
921 921 }
922 922 }
923 923 if (update) {
924 924 stp->sd_sigflags = 0;
925 925 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
926 926 stp->sd_sigflags |= ssp->ss_events;
927 927 }
928 928 mutex_exit(&stp->sd_lock);
929 929 }
930 930
931 931 /*
932 932 * Used on the last close to remove any remaining items on the siglist.
933 933 * These could be present on the siglist due to I_ESETSIG calls that
934 934 * use process groups or processed that do not have an open file descriptor
935 935 * for this stream (Such entries would not be removed by strclean).
936 936 */
937 937 static void
938 938 strcleanall(struct vnode *vp)
939 939 {
940 940 strsig_t *ssp, *nssp;
941 941 stdata_t *stp;
942 942
943 943 stp = vp->v_stream;
944 944 mutex_enter(&stp->sd_lock);
945 945 ssp = stp->sd_siglist;
946 946 stp->sd_siglist = NULL;
947 947 while (ssp) {
948 948 nssp = ssp->ss_next;
949 949 mutex_enter(&pidlock);
950 950 PID_RELE(ssp->ss_pidp);
951 951 mutex_exit(&pidlock);
952 952 kmem_free(ssp, sizeof (strsig_t));
953 953 ssp = nssp;
954 954 }
955 955 stp->sd_sigflags = 0;
956 956 mutex_exit(&stp->sd_lock);
957 957 }
958 958
959 959 /*
960 960 * Retrieve the next message from the logical stream head read queue
961 961 * using either rwnext (if sync stream) or getq_noenab.
962 962 * It is the callers responsibility to call qbackenable after
963 963 * it is finished with the message. The caller should not call
964 964 * qbackenable until after any putback calls to avoid spurious backenabling.
965 965 */
966 966 mblk_t *
967 967 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
968 968 int *errorp)
969 969 {
970 970 mblk_t *bp;
971 971 int error;
972 972 ssize_t rbytes = 0;
973 973
974 974 /* Holding sd_lock prevents the read queue from changing */
975 975 ASSERT(MUTEX_HELD(&stp->sd_lock));
976 976
977 977 if (uiop != NULL && stp->sd_struiordq != NULL &&
978 978 q->q_first == NULL &&
979 979 (!first || (stp->sd_wakeq & RSLEEP))) {
980 980 /*
981 981 * Stream supports rwnext() for the read side.
982 982 * If this is the first time we're called by e.g. strread
983 983 * only do the downcall if there is a deferred wakeup
984 984 * (registered in sd_wakeq).
985 985 */
986 986 struiod_t uiod;
987 987
988 988 if (first)
989 989 stp->sd_wakeq &= ~RSLEEP;
990 990
991 991 (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
992 992 sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
993 993 uiod.d_mp = 0;
994 994 /*
995 995 * Mark that a thread is in rwnext on the read side
996 996 * to prevent strrput from nacking ioctls immediately.
997 997 * When the last concurrent rwnext returns
998 998 * the ioctls are nack'ed.
999 999 */
1000 1000 ASSERT(MUTEX_HELD(&stp->sd_lock));
1001 1001 stp->sd_struiodnak++;
1002 1002 /*
1003 1003 * Note: rwnext will drop sd_lock.
1004 1004 */
1005 1005 error = rwnext(q, &uiod);
1006 1006 ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1007 1007 mutex_enter(&stp->sd_lock);
1008 1008 stp->sd_struiodnak--;
1009 1009 while (stp->sd_struiodnak == 0 &&
1010 1010 ((bp = stp->sd_struionak) != NULL)) {
1011 1011 stp->sd_struionak = bp->b_next;
1012 1012 bp->b_next = NULL;
1013 1013 bp->b_datap->db_type = M_IOCNAK;
1014 1014 /*
1015 1015 * Protect against the driver passing up
1016 1016 * messages after it has done a qprocsoff.
1017 1017 */
1018 1018 if (_OTHERQ(q)->q_next == NULL)
1019 1019 freemsg(bp);
1020 1020 else {
1021 1021 mutex_exit(&stp->sd_lock);
1022 1022 qreply(q, bp);
1023 1023 mutex_enter(&stp->sd_lock);
1024 1024 }
1025 1025 }
1026 1026 ASSERT(MUTEX_HELD(&stp->sd_lock));
1027 1027 if (error == 0 || error == EWOULDBLOCK) {
1028 1028 if ((bp = uiod.d_mp) != NULL) {
1029 1029 *errorp = 0;
1030 1030 ASSERT(MUTEX_HELD(&stp->sd_lock));
1031 1031 return (bp);
1032 1032 }
1033 1033 error = 0;
1034 1034 } else if (error == EINVAL) {
1035 1035 /*
1036 1036 * The stream plumbing must have
1037 1037 * changed while we were away, so
1038 1038 * just turn off rwnext()s.
1039 1039 */
1040 1040 error = 0;
1041 1041 } else if (error == EBUSY) {
1042 1042 /*
1043 1043 * The module might have data in transit using putnext
1044 1044 * Fall back on waiting + getq.
1045 1045 */
1046 1046 error = 0;
1047 1047 } else {
1048 1048 *errorp = error;
1049 1049 ASSERT(MUTEX_HELD(&stp->sd_lock));
1050 1050 return (NULL);
1051 1051 }
1052 1052 /*
1053 1053 * Try a getq in case a rwnext() generated mblk
1054 1054 * has bubbled up via strrput().
1055 1055 */
1056 1056 }
1057 1057 *errorp = 0;
1058 1058 ASSERT(MUTEX_HELD(&stp->sd_lock));
1059 1059
1060 1060 /*
1061 1061 * If we have a valid uio, try and use this as a guide for how
1062 1062 * many bytes to retrieve from the queue via getq_noenab().
1063 1063 * Doing this can avoid unneccesary counting of overlong
1064 1064 * messages in putback(). We currently only do this for sockets
1065 1065 * and only if there is no sd_rputdatafunc hook.
1066 1066 *
1067 1067 * The sd_rputdatafunc hook transforms the entire message
1068 1068 * before any bytes in it can be given to a client. So, rbytes
1069 1069 * must be 0 if there is a hook.
1070 1070 */
1071 1071 if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK) &&
1072 1072 (stp->sd_rputdatafunc == NULL))
1073 1073 rbytes = uiop->uio_resid;
1074 1074
1075 1075 return (getq_noenab(q, rbytes));
1076 1076 }
1077 1077
1078 1078 /*
1079 1079 * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1080 1080 * If the message does not fit in the uio the remainder of it is returned;
1081 1081 * otherwise NULL is returned. Any embedded zero-length mblk_t's are
1082 1082 * consumed, even if uio_resid reaches zero. On error, `*errorp' is set to
1083 1083 * the error code, the message is consumed, and NULL is returned.
1084 1084 */
1085 1085 static mblk_t *
1086 1086 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1087 1087 {
1088 1088 int error;
1089 1089 ptrdiff_t n;
1090 1090 mblk_t *nbp;
1091 1091
1092 1092 ASSERT(bp->b_wptr >= bp->b_rptr);
1093 1093
1094 1094 do {
1095 1095 if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1096 1096 ASSERT(n > 0);
1097 1097
1098 1098 error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1099 1099 if (error != 0) {
1100 1100 freemsg(bp);
1101 1101 *errorp = error;
1102 1102 return (NULL);
1103 1103 }
1104 1104 }
1105 1105
1106 1106 bp->b_rptr += n;
1107 1107 while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1108 1108 nbp = bp;
1109 1109 bp = bp->b_cont;
1110 1110 freeb(nbp);
1111 1111 }
1112 1112 } while (bp != NULL && uiop->uio_resid > 0);
1113 1113
1114 1114 *errorp = 0;
1115 1115 return (bp);
1116 1116 }
1117 1117
1118 1118 /*
1119 1119 * Read a stream according to the mode flags in sd_flag:
1120 1120 *
1121 1121 * (default mode) - Byte stream, msg boundaries are ignored
1122 1122 * RD_MSGDIS (msg discard) - Read on msg boundaries and throw away
1123 1123 * any data remaining in msg
1124 1124 * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1125 1125 * any remaining data on head of read queue
1126 1126 *
1127 1127 * Consume readable messages on the front of the queue until
1128 1128 * ttolwp(curthread)->lwp_count
1129 1129 * is satisfied, the readable messages are exhausted, or a message
1130 1130 * boundary is reached in a message mode. If no data was read and
1131 1131 * the stream was not opened with the NDELAY flag, block until data arrives.
1132 1132 * Otherwise return the data read and update the count.
1133 1133 *
1134 1134 * In default mode a 0 length message signifies end-of-file and terminates
1135 1135 * a read in progress. The 0 length message is removed from the queue
1136 1136 * only if it is the only message read (no data is read).
1137 1137 *
1138 1138 * An attempt to read an M_PROTO or M_PCPROTO message results in an
1139 1139 * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1140 1140 * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1141 1141 * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1142 1142 * are unlinked from and M_DATA blocks in the message, the protos are
1143 1143 * thrown away, and the data is read.
1144 1144 */
1145 1145 /* ARGSUSED */
1146 1146 int
1147 1147 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1148 1148 {
1149 1149 struct stdata *stp;
1150 1150 mblk_t *bp, *nbp;
1151 1151 queue_t *q;
1152 1152 int error = 0;
1153 1153 uint_t old_sd_flag;
1154 1154 int first;
1155 1155 char rflg;
1156 1156 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
1157 1157 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
1158 1158 short delim;
1159 1159 unsigned char pri = 0;
1160 1160 char waitflag;
1161 1161 unsigned char type;
1162 1162
1163 1163 TRACE_1(TR_FAC_STREAMS_FR,
1164 1164 TR_STRREAD_ENTER, "strread:%p", vp);
1165 1165 ASSERT(vp->v_stream);
1166 1166 stp = vp->v_stream;
1167 1167
1168 1168 mutex_enter(&stp->sd_lock);
1169 1169
1170 1170 if ((error = i_straccess(stp, JCREAD)) != 0) {
1171 1171 mutex_exit(&stp->sd_lock);
1172 1172 return (error);
1173 1173 }
1174 1174
1175 1175 if (stp->sd_flag & (STRDERR|STPLEX)) {
1176 1176 error = strgeterr(stp, STRDERR|STPLEX, 0);
1177 1177 if (error != 0) {
1178 1178 mutex_exit(&stp->sd_lock);
1179 1179 return (error);
1180 1180 }
1181 1181 }
1182 1182
1183 1183 /*
1184 1184 * Loop terminates when uiop->uio_resid == 0.
1185 1185 */
1186 1186 rflg = 0;
1187 1187 waitflag = READWAIT;
1188 1188 q = _RD(stp->sd_wrq);
1189 1189 for (;;) {
1190 1190 ASSERT(MUTEX_HELD(&stp->sd_lock));
1191 1191 old_sd_flag = stp->sd_flag;
1192 1192 mark = 0;
1193 1193 delim = 0;
1194 1194 first = 1;
1195 1195 while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1196 1196 int done = 0;
1197 1197
1198 1198 ASSERT(MUTEX_HELD(&stp->sd_lock));
1199 1199
1200 1200 if (error != 0)
1201 1201 goto oops;
1202 1202
1203 1203 if (stp->sd_flag & (STRHUP|STREOF)) {
1204 1204 goto oops;
1205 1205 }
1206 1206 if (rflg && !(stp->sd_flag & STRDELIM)) {
1207 1207 goto oops;
1208 1208 }
1209 1209 /*
1210 1210 * If a read(fd,buf,0) has been done, there is no
1211 1211 * need to sleep. We always have zero bytes to
1212 1212 * return.
1213 1213 */
1214 1214 if (uiop->uio_resid == 0) {
1215 1215 goto oops;
1216 1216 }
1217 1217
1218 1218 qbackenable(q, 0);
1219 1219
1220 1220 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1221 1221 "strread calls strwaitq:%p, %p, %p",
1222 1222 vp, uiop, crp);
1223 1223 if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1224 1224 uiop->uio_fmode, -1, &done)) != 0 || done) {
1225 1225 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1226 1226 "strread error or done:%p, %p, %p",
1227 1227 vp, uiop, crp);
1228 1228 if ((uiop->uio_fmode & FNDELAY) &&
1229 1229 (stp->sd_flag & OLDNDELAY) &&
1230 1230 (error == EAGAIN))
1231 1231 error = 0;
1232 1232 goto oops;
1233 1233 }
1234 1234 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1235 1235 "strread awakes:%p, %p, %p", vp, uiop, crp);
1236 1236 if ((error = i_straccess(stp, JCREAD)) != 0) {
1237 1237 goto oops;
1238 1238 }
1239 1239 first = 0;
1240 1240 }
1241 1241
1242 1242 ASSERT(MUTEX_HELD(&stp->sd_lock));
1243 1243 ASSERT(bp);
1244 1244 pri = bp->b_band;
1245 1245 /*
1246 1246 * Extract any mark information. If the message is not
1247 1247 * completely consumed this information will be put in the mblk
1248 1248 * that is putback.
1249 1249 * If MSGMARKNEXT is set and the message is completely consumed
1250 1250 * the STRATMARK flag will be set below. Likewise, if
1251 1251 * MSGNOTMARKNEXT is set and the message is
1252 1252 * completely consumed STRNOTATMARK will be set.
1253 1253 *
1254 1254 * For some unknown reason strread only breaks the read at the
1255 1255 * last mark.
1256 1256 */
1257 1257 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1258 1258 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1259 1259 (MSGMARKNEXT|MSGNOTMARKNEXT));
1260 1260 if (mark != 0 && bp == stp->sd_mark) {
1261 1261 if (rflg) {
1262 1262 putback(stp, q, bp, pri);
1263 1263 goto oops;
1264 1264 }
1265 1265 mark |= _LASTMARK;
1266 1266 stp->sd_mark = NULL;
1267 1267 }
1268 1268 if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1269 1269 delim = 1;
1270 1270 mutex_exit(&stp->sd_lock);
1271 1271
1272 1272 if (STREAM_NEEDSERVICE(stp))
1273 1273 stream_runservice(stp);
1274 1274
1275 1275 type = bp->b_datap->db_type;
1276 1276
1277 1277 switch (type) {
1278 1278
1279 1279 case M_DATA:
1280 1280 ismdata:
1281 1281 if (msgnodata(bp)) {
1282 1282 if (mark || delim) {
1283 1283 freemsg(bp);
1284 1284 } else if (rflg) {
1285 1285
1286 1286 /*
1287 1287 * If already read data put zero
1288 1288 * length message back on queue else
1289 1289 * free msg and return 0.
1290 1290 */
1291 1291 bp->b_band = pri;
1292 1292 mutex_enter(&stp->sd_lock);
1293 1293 putback(stp, q, bp, pri);
1294 1294 mutex_exit(&stp->sd_lock);
1295 1295 } else {
1296 1296 freemsg(bp);
1297 1297 }
1298 1298 error = 0;
1299 1299 goto oops1;
1300 1300 }
1301 1301
1302 1302 rflg = 1;
1303 1303 waitflag |= NOINTR;
1304 1304 bp = struiocopyout(bp, uiop, &error);
1305 1305 if (error != 0)
1306 1306 goto oops1;
1307 1307
1308 1308 mutex_enter(&stp->sd_lock);
1309 1309 if (bp) {
1310 1310 /*
1311 1311 * Have remaining data in message.
1312 1312 * Free msg if in discard mode.
1313 1313 */
1314 1314 if (stp->sd_read_opt & RD_MSGDIS) {
1315 1315 freemsg(bp);
1316 1316 } else {
1317 1317 bp->b_band = pri;
1318 1318 if ((mark & _LASTMARK) &&
1319 1319 (stp->sd_mark == NULL))
1320 1320 stp->sd_mark = bp;
1321 1321 bp->b_flag |= mark & ~_LASTMARK;
1322 1322 if (delim)
1323 1323 bp->b_flag |= MSGDELIM;
1324 1324 if (msgnodata(bp))
1325 1325 freemsg(bp);
1326 1326 else
1327 1327 putback(stp, q, bp, pri);
1328 1328 }
1329 1329 } else {
1330 1330 /*
1331 1331 * Consumed the complete message.
1332 1332 * Move the MSG*MARKNEXT information
1333 1333 * to the stream head just in case
1334 1334 * the read queue becomes empty.
1335 1335 *
1336 1336 * If the stream head was at the mark
1337 1337 * (STRATMARK) before we dropped sd_lock above
1338 1338 * and some data was consumed then we have
1339 1339 * moved past the mark thus STRATMARK is
1340 1340 * cleared. However, if a message arrived in
1341 1341 * strrput during the copyout above causing
1342 1342 * STRATMARK to be set we can not clear that
1343 1343 * flag.
1344 1344 */
1345 1345 if (mark &
1346 1346 (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1347 1347 if (mark & MSGMARKNEXT) {
1348 1348 stp->sd_flag &= ~STRNOTATMARK;
1349 1349 stp->sd_flag |= STRATMARK;
1350 1350 } else if (mark & MSGNOTMARKNEXT) {
1351 1351 stp->sd_flag &= ~STRATMARK;
1352 1352 stp->sd_flag |= STRNOTATMARK;
1353 1353 } else {
1354 1354 stp->sd_flag &=
1355 1355 ~(STRATMARK|STRNOTATMARK);
1356 1356 }
1357 1357 } else if (rflg && (old_sd_flag & STRATMARK)) {
1358 1358 stp->sd_flag &= ~STRATMARK;
1359 1359 }
1360 1360 }
1361 1361
1362 1362 /*
1363 1363 * Check for signal messages at the front of the read
1364 1364 * queue and generate the signal(s) if appropriate.
1365 1365 * The only signal that can be on queue is M_SIG at
1366 1366 * this point.
1367 1367 */
1368 1368 while ((((bp = q->q_first)) != NULL) &&
1369 1369 (bp->b_datap->db_type == M_SIG)) {
1370 1370 bp = getq_noenab(q, 0);
1371 1371 /*
1372 1372 * sd_lock is held so the content of the
1373 1373 * read queue can not change.
1374 1374 */
1375 1375 ASSERT(bp != NULL && DB_TYPE(bp) == M_SIG);
1376 1376 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
1377 1377 mutex_exit(&stp->sd_lock);
1378 1378 freemsg(bp);
1379 1379 if (STREAM_NEEDSERVICE(stp))
1380 1380 stream_runservice(stp);
1381 1381 mutex_enter(&stp->sd_lock);
1382 1382 }
1383 1383
1384 1384 if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1385 1385 delim ||
1386 1386 (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1387 1387 goto oops;
1388 1388 }
1389 1389 continue;
1390 1390
1391 1391 case M_SIG:
1392 1392 strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1393 1393 freemsg(bp);
1394 1394 mutex_enter(&stp->sd_lock);
1395 1395 continue;
1396 1396
1397 1397 case M_PROTO:
1398 1398 case M_PCPROTO:
1399 1399 /*
1400 1400 * Only data messages are readable.
1401 1401 * Any others generate an error, unless
1402 1402 * RD_PROTDIS or RD_PROTDAT is set.
1403 1403 */
1404 1404 if (stp->sd_read_opt & RD_PROTDAT) {
1405 1405 for (nbp = bp; nbp; nbp = nbp->b_next) {
1406 1406 if ((nbp->b_datap->db_type ==
1407 1407 M_PROTO) ||
1408 1408 (nbp->b_datap->db_type ==
1409 1409 M_PCPROTO)) {
1410 1410 nbp->b_datap->db_type = M_DATA;
1411 1411 } else {
1412 1412 break;
1413 1413 }
1414 1414 }
1415 1415 /*
1416 1416 * clear stream head hi pri flag based on
1417 1417 * first message
1418 1418 */
1419 1419 if (type == M_PCPROTO) {
1420 1420 mutex_enter(&stp->sd_lock);
1421 1421 stp->sd_flag &= ~STRPRI;
1422 1422 mutex_exit(&stp->sd_lock);
1423 1423 }
1424 1424 goto ismdata;
1425 1425 } else if (stp->sd_read_opt & RD_PROTDIS) {
1426 1426 /*
1427 1427 * discard non-data messages
1428 1428 */
1429 1429 while (bp &&
1430 1430 ((bp->b_datap->db_type == M_PROTO) ||
1431 1431 (bp->b_datap->db_type == M_PCPROTO))) {
1432 1432 nbp = unlinkb(bp);
1433 1433 freeb(bp);
1434 1434 bp = nbp;
1435 1435 }
1436 1436 /*
1437 1437 * clear stream head hi pri flag based on
1438 1438 * first message
1439 1439 */
1440 1440 if (type == M_PCPROTO) {
1441 1441 mutex_enter(&stp->sd_lock);
1442 1442 stp->sd_flag &= ~STRPRI;
1443 1443 mutex_exit(&stp->sd_lock);
1444 1444 }
1445 1445 if (bp) {
1446 1446 bp->b_band = pri;
1447 1447 goto ismdata;
1448 1448 } else {
1449 1449 break;
1450 1450 }
1451 1451 }
1452 1452 /* FALLTHRU */
1453 1453 case M_PASSFP:
1454 1454 if ((bp->b_datap->db_type == M_PASSFP) &&
1455 1455 (stp->sd_read_opt & RD_PROTDIS)) {
1456 1456 freemsg(bp);
1457 1457 break;
1458 1458 }
1459 1459 mutex_enter(&stp->sd_lock);
1460 1460 putback(stp, q, bp, pri);
1461 1461 mutex_exit(&stp->sd_lock);
1462 1462 if (rflg == 0)
1463 1463 error = EBADMSG;
1464 1464 goto oops1;
1465 1465
1466 1466 default:
1467 1467 /*
1468 1468 * Garbage on stream head read queue.
1469 1469 */
1470 1470 cmn_err(CE_WARN, "bad %x found at stream head\n",
1471 1471 bp->b_datap->db_type);
1472 1472 freemsg(bp);
1473 1473 goto oops1;
1474 1474 }
1475 1475 mutex_enter(&stp->sd_lock);
1476 1476 }
1477 1477 oops:
1478 1478 mutex_exit(&stp->sd_lock);
1479 1479 oops1:
1480 1480 qbackenable(q, pri);
1481 1481 return (error);
1482 1482 #undef _LASTMARK
1483 1483 }
1484 1484
1485 1485 /*
1486 1486 * Default processing of M_PROTO/M_PCPROTO messages.
1487 1487 * Determine which wakeups and signals are needed.
1488 1488 * This can be replaced by a user-specified procedure for kernel users
1489 1489 * of STREAMS.
1490 1490 */
1491 1491 /* ARGSUSED */
1492 1492 mblk_t *
1493 1493 strrput_proto(vnode_t *vp, mblk_t *mp,
1494 1494 strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1495 1495 strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1496 1496 {
1497 1497 *wakeups = RSLEEP;
1498 1498 *allmsgsigs = 0;
1499 1499
1500 1500 switch (mp->b_datap->db_type) {
1501 1501 case M_PROTO:
1502 1502 if (mp->b_band == 0) {
1503 1503 *firstmsgsigs = S_INPUT | S_RDNORM;
1504 1504 *pollwakeups = POLLIN | POLLRDNORM;
1505 1505 } else {
1506 1506 *firstmsgsigs = S_INPUT | S_RDBAND;
1507 1507 *pollwakeups = POLLIN | POLLRDBAND;
1508 1508 }
1509 1509 break;
1510 1510 case M_PCPROTO:
1511 1511 *firstmsgsigs = S_HIPRI;
1512 1512 *pollwakeups = POLLPRI;
1513 1513 break;
1514 1514 }
1515 1515 return (mp);
1516 1516 }
1517 1517
1518 1518 /*
1519 1519 * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1520 1520 * M_PASSFP messages.
1521 1521 * Determine which wakeups and signals are needed.
1522 1522 * This can be replaced by a user-specified procedure for kernel users
1523 1523 * of STREAMS.
1524 1524 */
1525 1525 /* ARGSUSED */
1526 1526 mblk_t *
1527 1527 strrput_misc(vnode_t *vp, mblk_t *mp,
1528 1528 strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1529 1529 strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1530 1530 {
1531 1531 *wakeups = 0;
1532 1532 *firstmsgsigs = 0;
1533 1533 *allmsgsigs = 0;
1534 1534 *pollwakeups = 0;
1535 1535 return (mp);
1536 1536 }
1537 1537
1538 1538 /*
1539 1539 * Stream read put procedure. Called from downstream driver/module
1540 1540 * with messages for the stream head. Data, protocol, and in-stream
1541 1541 * signal messages are placed on the queue, others are handled directly.
1542 1542 */
1543 1543 int
1544 1544 strrput(queue_t *q, mblk_t *bp)
1545 1545 {
1546 1546 struct stdata *stp;
1547 1547 ulong_t rput_opt;
1548 1548 strwakeup_t wakeups;
1549 1549 strsigset_t firstmsgsigs; /* Signals if first message on queue */
1550 1550 strsigset_t allmsgsigs; /* Signals for all messages */
1551 1551 strsigset_t signals; /* Signals events to generate */
1552 1552 strpollset_t pollwakeups;
1553 1553 mblk_t *nextbp;
1554 1554 uchar_t band = 0;
1555 1555 int hipri_sig;
1556 1556
1557 1557 stp = (struct stdata *)q->q_ptr;
1558 1558 /*
1559 1559 * Use rput_opt for optimized access to the SR_ flags except
1560 1560 * SR_POLLIN. That flag has to be checked under sd_lock since it
1561 1561 * is modified by strpoll().
1562 1562 */
1563 1563 rput_opt = stp->sd_rput_opt;
1564 1564
1565 1565 ASSERT(qclaimed(q));
1566 1566 TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1567 1567 "strrput called with message type:q %p bp %p", q, bp);
1568 1568
1569 1569 /*
1570 1570 * Perform initial processing and pass to the parameterized functions.
1571 1571 */
1572 1572 ASSERT(bp->b_next == NULL);
1573 1573
1574 1574 switch (bp->b_datap->db_type) {
1575 1575 case M_DATA:
1576 1576 /*
1577 1577 * sockfs is the only consumer of STREOF and when it is set,
1578 1578 * it implies that the receiver is not interested in receiving
1579 1579 * any more data, hence the mblk is freed to prevent unnecessary
1580 1580 * message queueing at the stream head.
1581 1581 */
1582 1582 if (stp->sd_flag == STREOF) {
1583 1583 freemsg(bp);
1584 1584 return (0);
1585 1585 }
1586 1586 if ((rput_opt & SR_IGN_ZEROLEN) &&
1587 1587 bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1588 1588 /*
1589 1589 * Ignore zero-length M_DATA messages. These might be
1590 1590 * generated by some transports.
1591 1591 * The zero-length M_DATA messages, even if they
1592 1592 * are ignored, should effect the atmark tracking and
1593 1593 * should wake up a thread sleeping in strwaitmark.
1594 1594 */
1595 1595 mutex_enter(&stp->sd_lock);
1596 1596 if (bp->b_flag & MSGMARKNEXT) {
1597 1597 /*
1598 1598 * Record the position of the mark either
1599 1599 * in q_last or in STRATMARK.
1600 1600 */
1601 1601 if (q->q_last != NULL) {
1602 1602 q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1603 1603 q->q_last->b_flag |= MSGMARKNEXT;
1604 1604 } else {
1605 1605 stp->sd_flag &= ~STRNOTATMARK;
1606 1606 stp->sd_flag |= STRATMARK;
1607 1607 }
1608 1608 } else if (bp->b_flag & MSGNOTMARKNEXT) {
1609 1609 /*
1610 1610 * Record that this is not the position of
1611 1611 * the mark either in q_last or in
1612 1612 * STRNOTATMARK.
1613 1613 */
1614 1614 if (q->q_last != NULL) {
1615 1615 q->q_last->b_flag &= ~MSGMARKNEXT;
1616 1616 q->q_last->b_flag |= MSGNOTMARKNEXT;
1617 1617 } else {
1618 1618 stp->sd_flag &= ~STRATMARK;
1619 1619 stp->sd_flag |= STRNOTATMARK;
1620 1620 }
1621 1621 }
1622 1622 if (stp->sd_flag & RSLEEP) {
1623 1623 stp->sd_flag &= ~RSLEEP;
1624 1624 cv_broadcast(&q->q_wait);
1625 1625 }
1626 1626 mutex_exit(&stp->sd_lock);
1627 1627 freemsg(bp);
1628 1628 return (0);
1629 1629 }
1630 1630 wakeups = RSLEEP;
1631 1631 if (bp->b_band == 0) {
1632 1632 firstmsgsigs = S_INPUT | S_RDNORM;
1633 1633 pollwakeups = POLLIN | POLLRDNORM;
1634 1634 } else {
1635 1635 firstmsgsigs = S_INPUT | S_RDBAND;
1636 1636 pollwakeups = POLLIN | POLLRDBAND;
1637 1637 }
1638 1638 if (rput_opt & SR_SIGALLDATA)
1639 1639 allmsgsigs = firstmsgsigs;
1640 1640 else
1641 1641 allmsgsigs = 0;
1642 1642
1643 1643 mutex_enter(&stp->sd_lock);
1644 1644 if ((rput_opt & SR_CONSOL_DATA) &&
1645 1645 (q->q_last != NULL) &&
1646 1646 (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1647 1647 /*
1648 1648 * Consolidate an M_DATA message onto an M_DATA,
1649 1649 * M_PROTO, or M_PCPROTO by merging it with q_last.
1650 1650 * The consolidation does not take place if
1651 1651 * the old message is marked with either of the
1652 1652 * marks or the delim flag or if the new
1653 1653 * message is marked with MSGMARK. The MSGMARK
1654 1654 * check is needed to handle the odd semantics of
1655 1655 * MSGMARK where essentially the whole message
1656 1656 * is to be treated as marked.
1657 1657 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from the
1658 1658 * new message to the front of the b_cont chain.
1659 1659 */
1660 1660 mblk_t *lbp = q->q_last;
1661 1661 unsigned char db_type = lbp->b_datap->db_type;
1662 1662
1663 1663 if ((db_type == M_DATA || db_type == M_PROTO ||
1664 1664 db_type == M_PCPROTO) &&
1665 1665 !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1666 1666 rmvq_noenab(q, lbp);
1667 1667 /*
1668 1668 * The first message in the b_cont list
1669 1669 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1670 1670 * We need to handle the case where we
1671 1671 * are appending:
1672 1672 *
1673 1673 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1674 1674 * 2) a MSGMARKNEXT to a plain message.
1675 1675 * 3) a MSGNOTMARKNEXT to a plain message
1676 1676 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1677 1677 * message.
1678 1678 *
1679 1679 * Thus we never append a MSGMARKNEXT or
1680 1680 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1681 1681 */
1682 1682 if (bp->b_flag & MSGMARKNEXT) {
1683 1683 lbp->b_flag |= MSGMARKNEXT;
1684 1684 lbp->b_flag &= ~MSGNOTMARKNEXT;
1685 1685 bp->b_flag &= ~MSGMARKNEXT;
1686 1686 } else if (bp->b_flag & MSGNOTMARKNEXT) {
1687 1687 lbp->b_flag |= MSGNOTMARKNEXT;
1688 1688 bp->b_flag &= ~MSGNOTMARKNEXT;
1689 1689 }
1690 1690
1691 1691 linkb(lbp, bp);
1692 1692 bp = lbp;
1693 1693 /*
1694 1694 * The new message logically isn't the first
1695 1695 * even though the q_first check below thinks
1696 1696 * it is. Clear the firstmsgsigs to make it
1697 1697 * not appear to be first.
1698 1698 */
1699 1699 firstmsgsigs = 0;
1700 1700 }
1701 1701 }
1702 1702 break;
1703 1703
1704 1704 case M_PASSFP:
1705 1705 wakeups = RSLEEP;
1706 1706 allmsgsigs = 0;
1707 1707 if (bp->b_band == 0) {
1708 1708 firstmsgsigs = S_INPUT | S_RDNORM;
1709 1709 pollwakeups = POLLIN | POLLRDNORM;
1710 1710 } else {
1711 1711 firstmsgsigs = S_INPUT | S_RDBAND;
1712 1712 pollwakeups = POLLIN | POLLRDBAND;
1713 1713 }
1714 1714 mutex_enter(&stp->sd_lock);
1715 1715 break;
1716 1716
1717 1717 case M_PROTO:
1718 1718 case M_PCPROTO:
1719 1719 ASSERT(stp->sd_rprotofunc != NULL);
1720 1720 bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1721 1721 &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1722 1722 #define ALLSIG (S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1723 1723 S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1724 1724 #define ALLPOLL (POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1725 1725 POLLWRBAND)
1726 1726
1727 1727 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1728 1728 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1729 1729 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1730 1730 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1731 1731
1732 1732 mutex_enter(&stp->sd_lock);
1733 1733 break;
1734 1734
1735 1735 default:
1736 1736 ASSERT(stp->sd_rmiscfunc != NULL);
1737 1737 bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1738 1738 &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1739 1739 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1740 1740 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1741 1741 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1742 1742 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1743 1743 #undef ALLSIG
1744 1744 #undef ALLPOLL
1745 1745 mutex_enter(&stp->sd_lock);
1746 1746 break;
1747 1747 }
1748 1748 ASSERT(MUTEX_HELD(&stp->sd_lock));
1749 1749
1750 1750 /* By default generate superset of signals */
1751 1751 signals = (firstmsgsigs | allmsgsigs);
1752 1752
1753 1753 /*
1754 1754 * The proto and misc functions can return multiple messages
1755 1755 * as a b_next chain. Such messages are processed separately.
1756 1756 */
1757 1757 one_more:
1758 1758 hipri_sig = 0;
1759 1759 if (bp == NULL) {
1760 1760 nextbp = NULL;
1761 1761 } else {
1762 1762 nextbp = bp->b_next;
1763 1763 bp->b_next = NULL;
1764 1764
1765 1765 switch (bp->b_datap->db_type) {
1766 1766 case M_PCPROTO:
1767 1767 /*
1768 1768 * Only one priority protocol message is allowed at the
1769 1769 * stream head at a time.
1770 1770 */
1771 1771 if (stp->sd_flag & STRPRI) {
1772 1772 TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1773 1773 "M_PCPROTO already at head");
1774 1774 freemsg(bp);
1775 1775 mutex_exit(&stp->sd_lock);
1776 1776 goto done;
1777 1777 }
1778 1778 stp->sd_flag |= STRPRI;
1779 1779 hipri_sig = 1;
1780 1780 /* FALLTHRU */
1781 1781 case M_DATA:
1782 1782 case M_PROTO:
1783 1783 case M_PASSFP:
1784 1784 band = bp->b_band;
1785 1785 /*
1786 1786 * Marking doesn't work well when messages
1787 1787 * are marked in more than one band. We only
1788 1788 * remember the last message received, even if
1789 1789 * it is placed on the queue ahead of other
1790 1790 * marked messages.
1791 1791 */
1792 1792 if (bp->b_flag & MSGMARK)
1793 1793 stp->sd_mark = bp;
1794 1794 (void) putq(q, bp);
1795 1795
1796 1796 /*
1797 1797 * If message is a PCPROTO message, always use
1798 1798 * firstmsgsigs to determine if a signal should be
1799 1799 * sent as strrput is the only place to send
1800 1800 * signals for PCPROTO. Other messages are based on
1801 1801 * the STRGETINPROG flag. The flag determines if
1802 1802 * strrput or (k)strgetmsg will be responsible for
1803 1803 * sending the signals, in the firstmsgsigs case.
1804 1804 */
1805 1805 if ((hipri_sig == 1) ||
1806 1806 (((stp->sd_flag & STRGETINPROG) == 0) &&
1807 1807 (q->q_first == bp)))
1808 1808 signals = (firstmsgsigs | allmsgsigs);
1809 1809 else
1810 1810 signals = allmsgsigs;
1811 1811 break;
1812 1812
1813 1813 default:
1814 1814 mutex_exit(&stp->sd_lock);
1815 1815 (void) strrput_nondata(q, bp);
1816 1816 mutex_enter(&stp->sd_lock);
1817 1817 break;
1818 1818 }
1819 1819 }
1820 1820 ASSERT(MUTEX_HELD(&stp->sd_lock));
1821 1821 /*
1822 1822 * Wake sleeping read/getmsg and cancel deferred wakeup
1823 1823 */
1824 1824 if (wakeups & RSLEEP)
1825 1825 stp->sd_wakeq &= ~RSLEEP;
1826 1826
1827 1827 wakeups &= stp->sd_flag;
1828 1828 if (wakeups & RSLEEP) {
1829 1829 stp->sd_flag &= ~RSLEEP;
1830 1830 cv_broadcast(&q->q_wait);
1831 1831 }
1832 1832 if (wakeups & WSLEEP) {
1833 1833 stp->sd_flag &= ~WSLEEP;
1834 1834 cv_broadcast(&_WR(q)->q_wait);
1835 1835 }
1836 1836
1837 1837 if (pollwakeups != 0) {
1838 1838 if (pollwakeups == (POLLIN | POLLRDNORM)) {
1839 1839 /*
1840 1840 * Can't use rput_opt since it was not
1841 1841 * read when sd_lock was held and SR_POLLIN is changed
1842 1842 * by strpoll() under sd_lock.
1843 1843 */
1844 1844 if (!(stp->sd_rput_opt & SR_POLLIN))
1845 1845 goto no_pollwake;
1846 1846 stp->sd_rput_opt &= ~SR_POLLIN;
1847 1847 }
1848 1848 mutex_exit(&stp->sd_lock);
1849 1849 pollwakeup(&stp->sd_pollist, pollwakeups);
1850 1850 mutex_enter(&stp->sd_lock);
1851 1851 }
1852 1852 no_pollwake:
1853 1853
1854 1854 /*
1855 1855 * strsendsig can handle multiple signals with a
1856 1856 * single call.
1857 1857 */
1858 1858 if (stp->sd_sigflags & signals)
1859 1859 strsendsig(stp->sd_siglist, signals, band, 0);
1860 1860 mutex_exit(&stp->sd_lock);
1861 1861
1862 1862
1863 1863 done:
1864 1864 if (nextbp == NULL)
1865 1865 return (0);
1866 1866
1867 1867 /*
1868 1868 * Any signals were handled the first time.
1869 1869 * Wakeups and pollwakeups are redone to avoid any race
1870 1870 * conditions - all the messages are not queued until the
1871 1871 * last message has been processed by strrput.
1872 1872 */
1873 1873 bp = nextbp;
1874 1874 signals = firstmsgsigs = allmsgsigs = 0;
1875 1875 mutex_enter(&stp->sd_lock);
1876 1876 goto one_more;
1877 1877 }
1878 1878
1879 1879 static void
1880 1880 log_dupioc(queue_t *rq, mblk_t *bp)
1881 1881 {
1882 1882 queue_t *wq, *qp;
1883 1883 char *modnames, *mnp, *dname;
1884 1884 size_t maxmodstr;
1885 1885 boolean_t islast;
1886 1886
1887 1887 /*
1888 1888 * Allocate a buffer large enough to hold the names of nstrpush modules
1889 1889 * and one driver, with spaces between and NUL terminator. If we can't
1890 1890 * get memory, then we'll just log the driver name.
1891 1891 */
1892 1892 maxmodstr = nstrpush * (FMNAMESZ + 1);
1893 1893 mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1894 1894
1895 1895 /* march down write side to print log message down to the driver */
1896 1896 wq = WR(rq);
1897 1897
1898 1898 /* make sure q_next doesn't shift around while we're grabbing data */
1899 1899 claimstr(wq);
1900 1900 qp = wq->q_next;
1901 1901 do {
1902 1902 dname = Q2NAME(qp);
1903 1903 islast = !SAMESTR(qp) || qp->q_next == NULL;
1904 1904 if (modnames == NULL) {
1905 1905 /*
1906 1906 * If we don't have memory, then get the driver name in
1907 1907 * the log where we can see it. Note that memory
1908 1908 * pressure is a possible cause of these sorts of bugs.
1909 1909 */
1910 1910 if (islast) {
1911 1911 modnames = dname;
1912 1912 maxmodstr = 0;
1913 1913 }
1914 1914 } else {
1915 1915 mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1916 1916 if (!islast)
1917 1917 *mnp++ = ' ';
1918 1918 }
1919 1919 qp = qp->q_next;
1920 1920 } while (!islast);
1921 1921 releasestr(wq);
1922 1922 /* Cannot happen unless stream head is corrupt. */
1923 1923 ASSERT(modnames != NULL);
1924 1924 (void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1925 1925 SL_CONSOLE|SL_TRACE|SL_ERROR,
1926 1926 "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1927 1927 rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1928 1928 (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1929 1929 if (maxmodstr != 0)
1930 1930 kmem_free(modnames, maxmodstr);
1931 1931 }
1932 1932
1933 1933 int
1934 1934 strrput_nondata(queue_t *q, mblk_t *bp)
1935 1935 {
1936 1936 struct stdata *stp;
1937 1937 struct iocblk *iocbp;
1938 1938 struct stroptions *sop;
1939 1939 struct copyreq *reqp;
1940 1940 struct copyresp *resp;
1941 1941 unsigned char bpri;
1942 1942 unsigned char flushed_already = 0;
1943 1943
1944 1944 stp = (struct stdata *)q->q_ptr;
1945 1945
1946 1946 ASSERT(!(stp->sd_flag & STPLEX));
1947 1947 ASSERT(qclaimed(q));
1948 1948
1949 1949 switch (bp->b_datap->db_type) {
1950 1950 case M_ERROR:
1951 1951 /*
1952 1952 * An error has occurred downstream, the errno is in the first
1953 1953 * bytes of the message.
1954 1954 */
1955 1955 if ((bp->b_wptr - bp->b_rptr) == 2) { /* New flavor */
1956 1956 unsigned char rw = 0;
1957 1957
1958 1958 mutex_enter(&stp->sd_lock);
1959 1959 if (*bp->b_rptr != NOERROR) { /* read error */
1960 1960 if (*bp->b_rptr != 0) {
1961 1961 if (stp->sd_flag & STRDERR)
1962 1962 flushed_already |= FLUSHR;
1963 1963 stp->sd_flag |= STRDERR;
1964 1964 rw |= FLUSHR;
1965 1965 } else {
1966 1966 stp->sd_flag &= ~STRDERR;
1967 1967 }
1968 1968 stp->sd_rerror = *bp->b_rptr;
1969 1969 }
1970 1970 bp->b_rptr++;
1971 1971 if (*bp->b_rptr != NOERROR) { /* write error */
1972 1972 if (*bp->b_rptr != 0) {
1973 1973 if (stp->sd_flag & STWRERR)
1974 1974 flushed_already |= FLUSHW;
1975 1975 stp->sd_flag |= STWRERR;
1976 1976 rw |= FLUSHW;
1977 1977 } else {
1978 1978 stp->sd_flag &= ~STWRERR;
1979 1979 }
1980 1980 stp->sd_werror = *bp->b_rptr;
1981 1981 }
1982 1982 if (rw) {
1983 1983 TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1984 1984 "strrput cv_broadcast:q %p, bp %p",
1985 1985 q, bp);
1986 1986 cv_broadcast(&q->q_wait); /* readers */
1987 1987 cv_broadcast(&_WR(q)->q_wait); /* writers */
1988 1988 cv_broadcast(&stp->sd_monitor); /* ioctllers */
1989 1989
1990 1990 mutex_exit(&stp->sd_lock);
1991 1991 pollwakeup(&stp->sd_pollist, POLLERR);
1992 1992 mutex_enter(&stp->sd_lock);
1993 1993
1994 1994 if (stp->sd_sigflags & S_ERROR)
1995 1995 strsendsig(stp->sd_siglist, S_ERROR, 0,
1996 1996 ((rw & FLUSHR) ? stp->sd_rerror :
1997 1997 stp->sd_werror));
1998 1998 mutex_exit(&stp->sd_lock);
1999 1999 /*
2000 2000 * Send the M_FLUSH only
2001 2001 * for the first M_ERROR
2002 2002 * message on the stream
2003 2003 */
2004 2004 if (flushed_already == rw) {
2005 2005 freemsg(bp);
2006 2006 return (0);
2007 2007 }
2008 2008
2009 2009 bp->b_datap->db_type = M_FLUSH;
2010 2010 *bp->b_rptr = rw;
2011 2011 bp->b_wptr = bp->b_rptr + 1;
2012 2012 /*
2013 2013 * Protect against the driver
2014 2014 * passing up messages after
2015 2015 * it has done a qprocsoff
2016 2016 */
2017 2017 if (_OTHERQ(q)->q_next == NULL)
2018 2018 freemsg(bp);
2019 2019 else
2020 2020 qreply(q, bp);
2021 2021 return (0);
2022 2022 } else
2023 2023 mutex_exit(&stp->sd_lock);
2024 2024 } else if (*bp->b_rptr != 0) { /* Old flavor */
2025 2025 if (stp->sd_flag & (STRDERR|STWRERR))
2026 2026 flushed_already = FLUSHRW;
2027 2027 mutex_enter(&stp->sd_lock);
2028 2028 stp->sd_flag |= (STRDERR|STWRERR);
2029 2029 stp->sd_rerror = *bp->b_rptr;
2030 2030 stp->sd_werror = *bp->b_rptr;
2031 2031 TRACE_2(TR_FAC_STREAMS_FR,
2032 2032 TR_STRRPUT_WAKE2,
2033 2033 "strrput wakeup #2:q %p, bp %p", q, bp);
2034 2034 cv_broadcast(&q->q_wait); /* the readers */
2035 2035 cv_broadcast(&_WR(q)->q_wait); /* the writers */
2036 2036 cv_broadcast(&stp->sd_monitor); /* ioctllers */
2037 2037
2038 2038 mutex_exit(&stp->sd_lock);
2039 2039 pollwakeup(&stp->sd_pollist, POLLERR);
2040 2040 mutex_enter(&stp->sd_lock);
2041 2041
2042 2042 if (stp->sd_sigflags & S_ERROR)
2043 2043 strsendsig(stp->sd_siglist, S_ERROR, 0,
2044 2044 (stp->sd_werror ? stp->sd_werror :
2045 2045 stp->sd_rerror));
2046 2046 mutex_exit(&stp->sd_lock);
2047 2047
2048 2048 /*
2049 2049 * Send the M_FLUSH only
2050 2050 * for the first M_ERROR
2051 2051 * message on the stream
2052 2052 */
2053 2053 if (flushed_already != FLUSHRW) {
2054 2054 bp->b_datap->db_type = M_FLUSH;
2055 2055 *bp->b_rptr = FLUSHRW;
2056 2056 /*
2057 2057 * Protect against the driver passing up
2058 2058 * messages after it has done a
2059 2059 * qprocsoff.
2060 2060 */
2061 2061 if (_OTHERQ(q)->q_next == NULL)
2062 2062 freemsg(bp);
2063 2063 else
2064 2064 qreply(q, bp);
2065 2065 return (0);
2066 2066 }
2067 2067 }
2068 2068 freemsg(bp);
2069 2069 return (0);
2070 2070
2071 2071 case M_HANGUP:
2072 2072
2073 2073 freemsg(bp);
2074 2074 mutex_enter(&stp->sd_lock);
2075 2075 stp->sd_werror = ENXIO;
2076 2076 stp->sd_flag |= STRHUP;
2077 2077 stp->sd_flag &= ~(WSLEEP|RSLEEP);
2078 2078
2079 2079 /*
2080 2080 * send signal if controlling tty
2081 2081 */
2082 2082
2083 2083 if (stp->sd_sidp) {
2084 2084 prsignal(stp->sd_sidp, SIGHUP);
2085 2085 if (stp->sd_sidp != stp->sd_pgidp)
2086 2086 pgsignal(stp->sd_pgidp, SIGTSTP);
2087 2087 }
2088 2088
2089 2089 /*
2090 2090 * wake up read, write, and exception pollers and
2091 2091 * reset wakeup mechanism.
2092 2092 */
2093 2093 cv_broadcast(&q->q_wait); /* the readers */
2094 2094 cv_broadcast(&_WR(q)->q_wait); /* the writers */
2095 2095 cv_broadcast(&stp->sd_monitor); /* the ioctllers */
2096 2096 strhup(stp);
2097 2097 mutex_exit(&stp->sd_lock);
2098 2098 return (0);
2099 2099
2100 2100 case M_UNHANGUP:
2101 2101 freemsg(bp);
2102 2102 mutex_enter(&stp->sd_lock);
2103 2103 stp->sd_werror = 0;
2104 2104 stp->sd_flag &= ~STRHUP;
2105 2105 mutex_exit(&stp->sd_lock);
2106 2106 return (0);
2107 2107
2108 2108 case M_SIG:
2109 2109 /*
2110 2110 * Someone downstream wants to post a signal. The
2111 2111 * signal to post is contained in the first byte of the
2112 2112 * message. If the message would go on the front of
2113 2113 * the queue, send a signal to the process group
2114 2114 * (if not SIGPOLL) or to the siglist processes
2115 2115 * (SIGPOLL). If something is already on the queue,
2116 2116 * OR if we are delivering a delayed suspend (*sigh*
2117 2117 * another "tty" hack) and there's no one sleeping already,
2118 2118 * just enqueue the message.
2119 2119 */
2120 2120 mutex_enter(&stp->sd_lock);
2121 2121 if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2122 2122 !(stp->sd_flag & RSLEEP))) {
2123 2123 (void) putq(q, bp);
2124 2124 mutex_exit(&stp->sd_lock);
2125 2125 return (0);
2126 2126 }
2127 2127 mutex_exit(&stp->sd_lock);
2128 2128 /* FALLTHRU */
2129 2129
2130 2130 case M_PCSIG:
2131 2131 /*
2132 2132 * Don't enqueue, just post the signal.
2133 2133 */
2134 2134 strsignal(stp, *bp->b_rptr, 0L);
2135 2135 freemsg(bp);
2136 2136 return (0);
2137 2137
2138 2138 case M_CMD:
2139 2139 if (MBLKL(bp) != sizeof (cmdblk_t)) {
2140 2140 freemsg(bp);
2141 2141 return (0);
2142 2142 }
2143 2143
2144 2144 mutex_enter(&stp->sd_lock);
2145 2145 if (stp->sd_flag & STRCMDWAIT) {
2146 2146 ASSERT(stp->sd_cmdblk == NULL);
2147 2147 stp->sd_cmdblk = bp;
2148 2148 cv_broadcast(&stp->sd_monitor);
2149 2149 mutex_exit(&stp->sd_lock);
2150 2150 } else {
2151 2151 mutex_exit(&stp->sd_lock);
2152 2152 freemsg(bp);
2153 2153 }
2154 2154 return (0);
2155 2155
2156 2156 case M_FLUSH:
2157 2157 /*
2158 2158 * Flush queues. The indication of which queues to flush
2159 2159 * is in the first byte of the message. If the read queue
2160 2160 * is specified, then flush it. If FLUSHBAND is set, just
2161 2161 * flush the band specified by the second byte of the message.
2162 2162 *
2163 2163 * If a module has issued a M_SETOPT to not flush hi
2164 2164 * priority messages off of the stream head, then pass this
2165 2165 * flag into the flushq code to preserve such messages.
2166 2166 */
2167 2167
2168 2168 if (*bp->b_rptr & FLUSHR) {
2169 2169 mutex_enter(&stp->sd_lock);
2170 2170 if (*bp->b_rptr & FLUSHBAND) {
2171 2171 ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2172 2172 flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2173 2173 } else
2174 2174 flushq_common(q, FLUSHALL,
2175 2175 stp->sd_read_opt & RFLUSHPCPROT);
2176 2176 if ((q->q_first == NULL) ||
2177 2177 (q->q_first->b_datap->db_type < QPCTL))
2178 2178 stp->sd_flag &= ~STRPRI;
2179 2179 else {
2180 2180 ASSERT(stp->sd_flag & STRPRI);
2181 2181 }
2182 2182 mutex_exit(&stp->sd_lock);
2183 2183 }
2184 2184 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2185 2185 *bp->b_rptr &= ~FLUSHR;
2186 2186 bp->b_flag |= MSGNOLOOP;
2187 2187 /*
2188 2188 * Protect against the driver passing up
2189 2189 * messages after it has done a qprocsoff.
2190 2190 */
2191 2191 if (_OTHERQ(q)->q_next == NULL)
2192 2192 freemsg(bp);
2193 2193 else
2194 2194 qreply(q, bp);
2195 2195 return (0);
2196 2196 }
2197 2197 freemsg(bp);
2198 2198 return (0);
2199 2199
2200 2200 case M_IOCACK:
2201 2201 case M_IOCNAK:
2202 2202 iocbp = (struct iocblk *)bp->b_rptr;
2203 2203 /*
2204 2204 * If not waiting for ACK or NAK then just free msg.
2205 2205 * If incorrect id sequence number then just free msg.
2206 2206 * If already have ACK or NAK for user then this is a
2207 2207 * duplicate, display a warning and free the msg.
2208 2208 */
2209 2209 mutex_enter(&stp->sd_lock);
2210 2210 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2211 2211 (stp->sd_iocid != iocbp->ioc_id)) {
2212 2212 /*
2213 2213 * If the ACK/NAK is a dup, display a message
2214 2214 * Dup is when sd_iocid == ioc_id, and
2215 2215 * sd_iocblk == <valid ptr> or -1 (the former
2216 2216 * is when an ioctl has been put on the stream
2217 2217 * head, but has not yet been consumed, the
2218 2218 * later is when it has been consumed).
2219 2219 */
2220 2220 if ((stp->sd_iocid == iocbp->ioc_id) &&
2221 2221 (stp->sd_iocblk != NULL)) {
2222 2222 log_dupioc(q, bp);
2223 2223 }
2224 2224 freemsg(bp);
2225 2225 mutex_exit(&stp->sd_lock);
2226 2226 return (0);
2227 2227 }
2228 2228
2229 2229 /*
2230 2230 * Assign ACK or NAK to user and wake up.
2231 2231 */
2232 2232 stp->sd_iocblk = bp;
2233 2233 cv_broadcast(&stp->sd_monitor);
2234 2234 mutex_exit(&stp->sd_lock);
2235 2235 return (0);
2236 2236
2237 2237 case M_COPYIN:
2238 2238 case M_COPYOUT:
2239 2239 reqp = (struct copyreq *)bp->b_rptr;
2240 2240
2241 2241 /*
2242 2242 * If not waiting for ACK or NAK then just fail request.
2243 2243 * If already have ACK, NAK, or copy request, then just
2244 2244 * fail request.
2245 2245 * If incorrect id sequence number then just fail request.
2246 2246 */
2247 2247 mutex_enter(&stp->sd_lock);
2248 2248 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2249 2249 (stp->sd_iocid != reqp->cq_id)) {
2250 2250 if (bp->b_cont) {
2251 2251 freemsg(bp->b_cont);
2252 2252 bp->b_cont = NULL;
2253 2253 }
2254 2254 bp->b_datap->db_type = M_IOCDATA;
2255 2255 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2256 2256 resp = (struct copyresp *)bp->b_rptr;
2257 2257 resp->cp_rval = (caddr_t)1; /* failure */
2258 2258 mutex_exit(&stp->sd_lock);
2259 2259 putnext(stp->sd_wrq, bp);
2260 2260 return (0);
2261 2261 }
2262 2262
2263 2263 /*
2264 2264 * Assign copy request to user and wake up.
2265 2265 */
2266 2266 stp->sd_iocblk = bp;
2267 2267 cv_broadcast(&stp->sd_monitor);
2268 2268 mutex_exit(&stp->sd_lock);
2269 2269 return (0);
2270 2270
2271 2271 case M_SETOPTS:
2272 2272 /*
2273 2273 * Set stream head options (read option, write offset,
2274 2274 * min/max packet size, and/or high/low water marks for
2275 2275 * the read side only).
2276 2276 */
2277 2277
2278 2278 bpri = 0;
2279 2279 sop = (struct stroptions *)bp->b_rptr;
2280 2280 mutex_enter(&stp->sd_lock);
2281 2281 if (sop->so_flags & SO_READOPT) {
2282 2282 switch (sop->so_readopt & RMODEMASK) {
2283 2283 case RNORM:
2284 2284 stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2285 2285 break;
2286 2286
2287 2287 case RMSGD:
2288 2288 stp->sd_read_opt =
2289 2289 ((stp->sd_read_opt & ~RD_MSGNODIS) |
2290 2290 RD_MSGDIS);
2291 2291 break;
2292 2292
2293 2293 case RMSGN:
2294 2294 stp->sd_read_opt =
2295 2295 ((stp->sd_read_opt & ~RD_MSGDIS) |
2296 2296 RD_MSGNODIS);
2297 2297 break;
2298 2298 }
2299 2299 switch (sop->so_readopt & RPROTMASK) {
2300 2300 case RPROTNORM:
2301 2301 stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2302 2302 break;
2303 2303
2304 2304 case RPROTDAT:
2305 2305 stp->sd_read_opt =
2306 2306 ((stp->sd_read_opt & ~RD_PROTDIS) |
2307 2307 RD_PROTDAT);
2308 2308 break;
2309 2309
2310 2310 case RPROTDIS:
2311 2311 stp->sd_read_opt =
2312 2312 ((stp->sd_read_opt & ~RD_PROTDAT) |
2313 2313 RD_PROTDIS);
2314 2314 break;
2315 2315 }
2316 2316 switch (sop->so_readopt & RFLUSHMASK) {
2317 2317 case RFLUSHPCPROT:
2318 2318 /*
2319 2319 * This sets the stream head to NOT flush
2320 2320 * M_PCPROTO messages.
2321 2321 */
2322 2322 stp->sd_read_opt |= RFLUSHPCPROT;
2323 2323 break;
2324 2324 }
2325 2325 }
2326 2326 if (sop->so_flags & SO_ERROPT) {
2327 2327 switch (sop->so_erropt & RERRMASK) {
2328 2328 case RERRNORM:
2329 2329 stp->sd_flag &= ~STRDERRNONPERSIST;
2330 2330 break;
2331 2331 case RERRNONPERSIST:
2332 2332 stp->sd_flag |= STRDERRNONPERSIST;
2333 2333 break;
2334 2334 }
2335 2335 switch (sop->so_erropt & WERRMASK) {
2336 2336 case WERRNORM:
2337 2337 stp->sd_flag &= ~STWRERRNONPERSIST;
2338 2338 break;
2339 2339 case WERRNONPERSIST:
2340 2340 stp->sd_flag |= STWRERRNONPERSIST;
2341 2341 break;
2342 2342 }
2343 2343 }
2344 2344 if (sop->so_flags & SO_COPYOPT) {
2345 2345 if (sop->so_copyopt & ZCVMSAFE) {
2346 2346 stp->sd_copyflag |= STZCVMSAFE;
2347 2347 stp->sd_copyflag &= ~STZCVMUNSAFE;
2348 2348 } else if (sop->so_copyopt & ZCVMUNSAFE) {
2349 2349 stp->sd_copyflag |= STZCVMUNSAFE;
2350 2350 stp->sd_copyflag &= ~STZCVMSAFE;
2351 2351 }
2352 2352
2353 2353 if (sop->so_copyopt & COPYCACHED) {
2354 2354 stp->sd_copyflag |= STRCOPYCACHED;
2355 2355 }
2356 2356 }
2357 2357 if (sop->so_flags & SO_WROFF)
2358 2358 stp->sd_wroff = sop->so_wroff;
2359 2359 if (sop->so_flags & SO_TAIL)
2360 2360 stp->sd_tail = sop->so_tail;
2361 2361 if (sop->so_flags & SO_MINPSZ)
2362 2362 q->q_minpsz = sop->so_minpsz;
2363 2363 if (sop->so_flags & SO_MAXPSZ)
2364 2364 q->q_maxpsz = sop->so_maxpsz;
2365 2365 if (sop->so_flags & SO_MAXBLK)
2366 2366 stp->sd_maxblk = sop->so_maxblk;
2367 2367 if (sop->so_flags & SO_HIWAT) {
2368 2368 if (sop->so_flags & SO_BAND) {
2369 2369 if (strqset(q, QHIWAT,
2370 2370 sop->so_band, sop->so_hiwat)) {
2371 2371 cmn_err(CE_WARN, "strrput: could not "
2372 2372 "allocate qband\n");
2373 2373 } else {
2374 2374 bpri = sop->so_band;
2375 2375 }
2376 2376 } else {
2377 2377 q->q_hiwat = sop->so_hiwat;
2378 2378 }
2379 2379 }
2380 2380 if (sop->so_flags & SO_LOWAT) {
2381 2381 if (sop->so_flags & SO_BAND) {
2382 2382 if (strqset(q, QLOWAT,
2383 2383 sop->so_band, sop->so_lowat)) {
2384 2384 cmn_err(CE_WARN, "strrput: could not "
2385 2385 "allocate qband\n");
2386 2386 } else {
2387 2387 bpri = sop->so_band;
2388 2388 }
2389 2389 } else {
2390 2390 q->q_lowat = sop->so_lowat;
2391 2391 }
2392 2392 }
2393 2393 if (sop->so_flags & SO_MREADON)
2394 2394 stp->sd_flag |= SNDMREAD;
2395 2395 if (sop->so_flags & SO_MREADOFF)
2396 2396 stp->sd_flag &= ~SNDMREAD;
2397 2397 if (sop->so_flags & SO_NDELON)
2398 2398 stp->sd_flag |= OLDNDELAY;
2399 2399 if (sop->so_flags & SO_NDELOFF)
2400 2400 stp->sd_flag &= ~OLDNDELAY;
2401 2401 if (sop->so_flags & SO_ISTTY)
2402 2402 stp->sd_flag |= STRISTTY;
2403 2403 if (sop->so_flags & SO_ISNTTY)
2404 2404 stp->sd_flag &= ~STRISTTY;
2405 2405 if (sop->so_flags & SO_TOSTOP)
2406 2406 stp->sd_flag |= STRTOSTOP;
2407 2407 if (sop->so_flags & SO_TONSTOP)
2408 2408 stp->sd_flag &= ~STRTOSTOP;
2409 2409 if (sop->so_flags & SO_DELIM)
2410 2410 stp->sd_flag |= STRDELIM;
2411 2411 if (sop->so_flags & SO_NODELIM)
2412 2412 stp->sd_flag &= ~STRDELIM;
2413 2413
2414 2414 mutex_exit(&stp->sd_lock);
2415 2415 freemsg(bp);
2416 2416
2417 2417 /* Check backenable in case the water marks changed */
2418 2418 qbackenable(q, bpri);
2419 2419 return (0);
2420 2420
2421 2421 /*
2422 2422 * The following set of cases deal with situations where two stream
2423 2423 * heads are connected to each other (twisted streams). These messages
2424 2424 * have no meaning at the stream head.
2425 2425 */
2426 2426 case M_BREAK:
2427 2427 case M_CTL:
2428 2428 case M_DELAY:
2429 2429 case M_START:
2430 2430 case M_STOP:
2431 2431 case M_IOCDATA:
2432 2432 case M_STARTI:
2433 2433 case M_STOPI:
2434 2434 freemsg(bp);
2435 2435 return (0);
2436 2436
2437 2437 case M_IOCTL:
2438 2438 /*
2439 2439 * Always NAK this condition
2440 2440 * (makes no sense)
2441 2441 * If there is one or more threads in the read side
2442 2442 * rwnext we have to defer the nacking until that thread
2443 2443 * returns (in strget).
2444 2444 */
2445 2445 mutex_enter(&stp->sd_lock);
2446 2446 if (stp->sd_struiodnak != 0) {
2447 2447 /*
2448 2448 * Defer NAK to the streamhead. Queue at the end
2449 2449 * the list.
2450 2450 */
2451 2451 mblk_t *mp = stp->sd_struionak;
2452 2452
2453 2453 while (mp && mp->b_next)
2454 2454 mp = mp->b_next;
2455 2455 if (mp)
2456 2456 mp->b_next = bp;
2457 2457 else
2458 2458 stp->sd_struionak = bp;
2459 2459 bp->b_next = NULL;
2460 2460 mutex_exit(&stp->sd_lock);
2461 2461 return (0);
2462 2462 }
2463 2463 mutex_exit(&stp->sd_lock);
2464 2464
2465 2465 bp->b_datap->db_type = M_IOCNAK;
2466 2466 /*
2467 2467 * Protect against the driver passing up
2468 2468 * messages after it has done a qprocsoff.
2469 2469 */
2470 2470 if (_OTHERQ(q)->q_next == NULL)
2471 2471 freemsg(bp);
2472 2472 else
2473 2473 qreply(q, bp);
2474 2474 return (0);
2475 2475
2476 2476 default:
2477 2477 #ifdef DEBUG
2478 2478 cmn_err(CE_WARN,
2479 2479 "bad message type %x received at stream head\n",
2480 2480 bp->b_datap->db_type);
2481 2481 #endif
2482 2482 freemsg(bp);
2483 2483 return (0);
2484 2484 }
2485 2485
2486 2486 /* NOTREACHED */
2487 2487 }
2488 2488
2489 2489 /*
2490 2490 * Check if the stream pointed to by `stp' can be written to, and return an
2491 2491 * error code if not. If `eiohup' is set, then return EIO if STRHUP is set.
2492 2492 * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2493 2493 * then always return EPIPE and send a SIGPIPE to the invoking thread.
2494 2494 */
2495 2495 static int
2496 2496 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2497 2497 {
2498 2498 int error;
2499 2499
2500 2500 ASSERT(MUTEX_HELD(&stp->sd_lock));
2501 2501
2502 2502 /*
2503 2503 * For modem support, POSIX states that on writes, EIO should
2504 2504 * be returned if the stream has been hung up.
2505 2505 */
2506 2506 if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2507 2507 error = EIO;
2508 2508 else
2509 2509 error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2510 2510
2511 2511 if (error != 0) {
2512 2512 if (!(stp->sd_flag & STPLEX) &&
2513 2513 (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2514 2514 tsignal(curthread, SIGPIPE);
2515 2515 error = EPIPE;
2516 2516 }
2517 2517 }
2518 2518
2519 2519 return (error);
2520 2520 }
2521 2521
2522 2522 /*
2523 2523 * Copyin and send data down a stream.
2524 2524 * The caller will allocate and copyin any control part that precedes the
2525 2525 * message and pass that in as mctl.
2526 2526 *
2527 2527 * Caller should *not* hold sd_lock.
2528 2528 * When EWOULDBLOCK is returned the caller has to redo the canputnext
2529 2529 * under sd_lock in order to avoid missing a backenabling wakeup.
2530 2530 *
2531 2531 * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2532 2532 *
2533 2533 * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2534 2534 * For sync streams we can only ignore flow control by reverting to using
2535 2535 * putnext.
2536 2536 *
2537 2537 * If sd_maxblk is less than *iosize this routine might return without
2538 2538 * transferring all of *iosize. In all cases, on return *iosize will contain
2539 2539 * the amount of data that was transferred.
2540 2540 */
2541 2541 static int
2542 2542 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2543 2543 int b_flag, int pri, int flags)
2544 2544 {
2545 2545 struiod_t uiod;
2546 2546 mblk_t *mp;
2547 2547 queue_t *wqp = stp->sd_wrq;
2548 2548 int error = 0;
2549 2549 ssize_t count = *iosize;
2550 2550
2551 2551 ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2552 2552
2553 2553 if (uiop != NULL && count >= 0)
2554 2554 flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2555 2555
2556 2556 if (!(flags & STRUIO_POSTPONE)) {
2557 2557 /*
2558 2558 * Use regular canputnext, strmakedata, putnext sequence.
2559 2559 */
2560 2560 if (pri == 0) {
2561 2561 if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2562 2562 freemsg(mctl);
2563 2563 return (EWOULDBLOCK);
2564 2564 }
2565 2565 } else {
2566 2566 if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2567 2567 freemsg(mctl);
2568 2568 return (EWOULDBLOCK);
2569 2569 }
2570 2570 }
2571 2571
2572 2572 if ((error = strmakedata(iosize, uiop, stp, flags,
2573 2573 &mp)) != 0) {
2574 2574 freemsg(mctl);
2575 2575 /*
2576 2576 * need to change return code to ENOMEM
2577 2577 * so that this is not confused with
2578 2578 * flow control, EAGAIN.
2579 2579 */
2580 2580
2581 2581 if (error == EAGAIN)
2582 2582 return (ENOMEM);
2583 2583 else
2584 2584 return (error);
2585 2585 }
2586 2586 if (mctl != NULL) {
2587 2587 if (mctl->b_cont == NULL)
2588 2588 mctl->b_cont = mp;
2589 2589 else if (mp != NULL)
2590 2590 linkb(mctl, mp);
2591 2591 mp = mctl;
2592 2592 } else if (mp == NULL)
2593 2593 return (0);
2594 2594
2595 2595 mp->b_flag |= b_flag;
2596 2596 mp->b_band = (uchar_t)pri;
2597 2597
2598 2598 if (flags & MSG_IGNFLOW) {
2599 2599 /*
2600 2600 * XXX Hack: Don't get stuck running service
2601 2601 * procedures. This is needed for sockfs when
2602 2602 * sending the unbind message out of the rput
2603 2603 * procedure - we don't want a put procedure
2604 2604 * to run service procedures.
2605 2605 */
2606 2606 putnext(wqp, mp);
2607 2607 } else {
2608 2608 stream_willservice(stp);
2609 2609 putnext(wqp, mp);
2610 2610 stream_runservice(stp);
2611 2611 }
2612 2612 return (0);
2613 2613 }
2614 2614 /*
2615 2615 * Stream supports rwnext() for the write side.
2616 2616 */
2617 2617 if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2618 2618 freemsg(mctl);
2619 2619 /*
2620 2620 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2621 2621 */
2622 2622 return (error == EAGAIN ? ENOMEM : error);
2623 2623 }
2624 2624 if (mctl != NULL) {
2625 2625 if (mctl->b_cont == NULL)
2626 2626 mctl->b_cont = mp;
2627 2627 else if (mp != NULL)
2628 2628 linkb(mctl, mp);
2629 2629 mp = mctl;
2630 2630 } else if (mp == NULL) {
2631 2631 return (0);
2632 2632 }
2633 2633
2634 2634 mp->b_flag |= b_flag;
2635 2635 mp->b_band = (uchar_t)pri;
2636 2636
2637 2637 (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2638 2638 sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2639 2639 uiod.d_uio.uio_offset = 0;
2640 2640 uiod.d_mp = mp;
2641 2641 error = rwnext(wqp, &uiod);
2642 2642 if (! uiod.d_mp) {
2643 2643 uioskip(uiop, *iosize);
2644 2644 return (error);
2645 2645 }
2646 2646 ASSERT(mp == uiod.d_mp);
2647 2647 if (error == EINVAL) {
2648 2648 /*
2649 2649 * The stream plumbing must have changed while
2650 2650 * we were away, so just turn off rwnext()s.
2651 2651 */
2652 2652 error = 0;
2653 2653 } else if (error == EBUSY || error == EWOULDBLOCK) {
2654 2654 /*
2655 2655 * Couldn't enter a perimeter or took a page fault,
2656 2656 * so fall-back to putnext().
2657 2657 */
2658 2658 error = 0;
2659 2659 } else {
2660 2660 freemsg(mp);
2661 2661 return (error);
2662 2662 }
2663 2663 /* Have to check canput before consuming data from the uio */
2664 2664 if (pri == 0) {
2665 2665 if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2666 2666 freemsg(mp);
2667 2667 return (EWOULDBLOCK);
2668 2668 }
2669 2669 } else {
2670 2670 if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2671 2671 freemsg(mp);
2672 2672 return (EWOULDBLOCK);
2673 2673 }
2674 2674 }
2675 2675 ASSERT(mp == uiod.d_mp);
2676 2676 /* Copyin data from the uio */
2677 2677 if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2678 2678 freemsg(mp);
2679 2679 return (error);
2680 2680 }
2681 2681 uioskip(uiop, *iosize);
2682 2682 if (flags & MSG_IGNFLOW) {
2683 2683 /*
2684 2684 * XXX Hack: Don't get stuck running service procedures.
2685 2685 * This is needed for sockfs when sending the unbind message
2686 2686 * out of the rput procedure - we don't want a put procedure
2687 2687 * to run service procedures.
2688 2688 */
2689 2689 putnext(wqp, mp);
2690 2690 } else {
2691 2691 stream_willservice(stp);
2692 2692 putnext(wqp, mp);
2693 2693 stream_runservice(stp);
2694 2694 }
2695 2695 return (0);
2696 2696 }
2697 2697
2698 2698 /*
2699 2699 * Write attempts to break the write request into messages conforming
2700 2700 * with the minimum and maximum packet sizes set downstream.
2701 2701 *
2702 2702 * Write will not block if downstream queue is full and
2703 2703 * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2704 2704 *
2705 2705 * A write of zero bytes gets packaged into a zero length message and sent
2706 2706 * downstream like any other message.
2707 2707 *
2708 2708 * If buffers of the requested sizes are not available, the write will
2709 2709 * sleep until the buffers become available.
2710 2710 *
2711 2711 * Write (if specified) will supply a write offset in a message if it
2712 2712 * makes sense. This can be specified by downstream modules as part of
2713 2713 * a M_SETOPTS message. Write will not supply the write offset if it
2714 2714 * cannot supply any data in a buffer. In other words, write will never
2715 2715 * send down an empty packet due to a write offset.
2716 2716 */
2717 2717 /* ARGSUSED2 */
2718 2718 int
2719 2719 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2720 2720 {
2721 2721 return (strwrite_common(vp, uiop, crp, 0));
2722 2722 }
2723 2723
2724 2724 /* ARGSUSED2 */
2725 2725 int
2726 2726 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2727 2727 {
2728 2728 struct stdata *stp;
2729 2729 struct queue *wqp;
2730 2730 ssize_t rmin, rmax;
2731 2731 ssize_t iosize;
2732 2732 int waitflag;
2733 2733 int tempmode;
2734 2734 int error = 0;
2735 2735 int b_flag;
2736 2736
2737 2737 ASSERT(vp->v_stream);
2738 2738 stp = vp->v_stream;
2739 2739
2740 2740 mutex_enter(&stp->sd_lock);
2741 2741
2742 2742 if ((error = i_straccess(stp, JCWRITE)) != 0) {
2743 2743 mutex_exit(&stp->sd_lock);
2744 2744 return (error);
2745 2745 }
2746 2746
2747 2747 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2748 2748 error = strwriteable(stp, B_TRUE, B_TRUE);
2749 2749 if (error != 0) {
2750 2750 mutex_exit(&stp->sd_lock);
2751 2751 return (error);
2752 2752 }
2753 2753 }
2754 2754
2755 2755 mutex_exit(&stp->sd_lock);
2756 2756
2757 2757 wqp = stp->sd_wrq;
2758 2758
2759 2759 /* get these values from them cached in the stream head */
2760 2760 rmin = stp->sd_qn_minpsz;
2761 2761 rmax = stp->sd_qn_maxpsz;
2762 2762
2763 2763 /*
2764 2764 * Check the min/max packet size constraints. If min packet size
2765 2765 * is non-zero, the write cannot be split into multiple messages
2766 2766 * and still guarantee the size constraints.
2767 2767 */
2768 2768 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2769 2769
2770 2770 ASSERT((rmax >= 0) || (rmax == INFPSZ));
2771 2771 if (rmax == 0) {
2772 2772 return (0);
2773 2773 }
2774 2774 if (rmin > 0) {
2775 2775 if (uiop->uio_resid < rmin) {
2776 2776 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2777 2777 "strwrite out:q %p out %d error %d",
2778 2778 wqp, 0, ERANGE);
2779 2779 return (ERANGE);
2780 2780 }
2781 2781 if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2782 2782 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2783 2783 "strwrite out:q %p out %d error %d",
2784 2784 wqp, 1, ERANGE);
2785 2785 return (ERANGE);
2786 2786 }
2787 2787 }
2788 2788
2789 2789 /*
2790 2790 * Do until count satisfied or error.
2791 2791 */
2792 2792 waitflag = WRITEWAIT | wflag;
2793 2793 if (stp->sd_flag & OLDNDELAY)
2794 2794 tempmode = uiop->uio_fmode & ~FNDELAY;
2795 2795 else
2796 2796 tempmode = uiop->uio_fmode;
2797 2797
2798 2798 if (rmax == INFPSZ)
2799 2799 rmax = uiop->uio_resid;
2800 2800
2801 2801 /*
2802 2802 * Note that tempmode does not get used in strput/strmakedata
2803 2803 * but only in strwaitq. The other routines use uio_fmode
2804 2804 * unmodified.
2805 2805 */
2806 2806
2807 2807 /* LINTED: constant in conditional context */
2808 2808 while (1) { /* breaks when uio_resid reaches zero */
2809 2809 /*
2810 2810 * Determine the size of the next message to be
2811 2811 * packaged. May have to break write into several
2812 2812 * messages based on max packet size.
2813 2813 */
2814 2814 iosize = MIN(uiop->uio_resid, rmax);
2815 2815
2816 2816 /*
2817 2817 * Put block downstream when flow control allows it.
2818 2818 */
2819 2819 if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2820 2820 b_flag = MSGDELIM;
2821 2821 else
2822 2822 b_flag = 0;
2823 2823
2824 2824 for (;;) {
2825 2825 int done = 0;
2826 2826
2827 2827 error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2828 2828 if (error == 0)
2829 2829 break;
2830 2830 if (error != EWOULDBLOCK)
2831 2831 goto out;
2832 2832
2833 2833 mutex_enter(&stp->sd_lock);
2834 2834 /*
2835 2835 * Check for a missed wakeup.
2836 2836 * Needed since strput did not hold sd_lock across
2837 2837 * the canputnext.
2838 2838 */
2839 2839 if (canputnext(wqp)) {
2840 2840 /* Try again */
2841 2841 mutex_exit(&stp->sd_lock);
2842 2842 continue;
2843 2843 }
2844 2844 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2845 2845 "strwrite wait:q %p wait", wqp);
2846 2846 if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2847 2847 tempmode, -1, &done)) != 0 || done) {
2848 2848 mutex_exit(&stp->sd_lock);
2849 2849 if ((vp->v_type == VFIFO) &&
2850 2850 (uiop->uio_fmode & FNDELAY) &&
2851 2851 (error == EAGAIN))
2852 2852 error = 0;
2853 2853 goto out;
2854 2854 }
2855 2855 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2856 2856 "strwrite wake:q %p awakes", wqp);
2857 2857 if ((error = i_straccess(stp, JCWRITE)) != 0) {
2858 2858 mutex_exit(&stp->sd_lock);
2859 2859 goto out;
2860 2860 }
2861 2861 mutex_exit(&stp->sd_lock);
2862 2862 }
2863 2863 waitflag |= NOINTR;
2864 2864 TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2865 2865 "strwrite resid:q %p uiop %p", wqp, uiop);
2866 2866 if (uiop->uio_resid) {
2867 2867 /* Recheck for errors - needed for sockets */
2868 2868 if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2869 2869 (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2870 2870 mutex_enter(&stp->sd_lock);
2871 2871 error = strwriteable(stp, B_FALSE, B_TRUE);
2872 2872 mutex_exit(&stp->sd_lock);
2873 2873 if (error != 0)
2874 2874 return (error);
2875 2875 }
2876 2876 continue;
2877 2877 }
2878 2878 break;
2879 2879 }
2880 2880 out:
2881 2881 /*
2882 2882 * For historical reasons, applications expect EAGAIN when a data
2883 2883 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2884 2884 */
2885 2885 if (error == ENOMEM)
2886 2886 error = EAGAIN;
2887 2887 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2888 2888 "strwrite out:q %p out %d error %d", wqp, 2, error);
2889 2889 return (error);
2890 2890 }
2891 2891
2892 2892 /*
2893 2893 * Stream head write service routine.
2894 2894 * Its job is to wake up any sleeping writers when a queue
2895 2895 * downstream needs data (part of the flow control in putq and getq).
2896 2896 * It also must wake anyone sleeping on a poll().
2897 2897 * For stream head right below mux module, it must also invoke put procedure
2898 2898 * of next downstream module.
2899 2899 */
2900 2900 int
2901 2901 strwsrv(queue_t *q)
2902 2902 {
2903 2903 struct stdata *stp;
2904 2904 queue_t *tq;
2905 2905 qband_t *qbp;
2906 2906 int i;
2907 2907 qband_t *myqbp;
2908 2908 int isevent;
2909 2909 unsigned char qbf[NBAND]; /* band flushing backenable flags */
2910 2910
2911 2911 TRACE_1(TR_FAC_STREAMS_FR,
2912 2912 TR_STRWSRV, "strwsrv:q %p", q);
2913 2913 stp = (struct stdata *)q->q_ptr;
2914 2914 ASSERT(qclaimed(q));
2915 2915 mutex_enter(&stp->sd_lock);
2916 2916 ASSERT(!(stp->sd_flag & STPLEX));
2917 2917
2918 2918 if (stp->sd_flag & WSLEEP) {
2919 2919 stp->sd_flag &= ~WSLEEP;
2920 2920 cv_broadcast(&q->q_wait);
2921 2921 }
2922 2922 mutex_exit(&stp->sd_lock);
2923 2923
2924 2924 /* The other end of a stream pipe went away. */
2925 2925 if ((tq = q->q_next) == NULL) {
2926 2926 return (0);
2927 2927 }
2928 2928
2929 2929 /* Find the next module forward that has a service procedure */
2930 2930 claimstr(q);
2931 2931 tq = q->q_nfsrv;
2932 2932 ASSERT(tq != NULL);
2933 2933
2934 2934 if ((q->q_flag & QBACK)) {
2935 2935 if ((tq->q_flag & QFULL)) {
2936 2936 mutex_enter(QLOCK(tq));
2937 2937 if (!(tq->q_flag & QFULL)) {
2938 2938 mutex_exit(QLOCK(tq));
2939 2939 goto wakeup;
2940 2940 }
2941 2941 /*
2942 2942 * The queue must have become full again. Set QWANTW
2943 2943 * again so strwsrv will be back enabled when
2944 2944 * the queue becomes non-full next time.
2945 2945 */
2946 2946 tq->q_flag |= QWANTW;
2947 2947 mutex_exit(QLOCK(tq));
2948 2948 } else {
2949 2949 wakeup:
2950 2950 pollwakeup(&stp->sd_pollist, POLLWRNORM);
2951 2951 mutex_enter(&stp->sd_lock);
2952 2952 if (stp->sd_sigflags & S_WRNORM)
2953 2953 strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2954 2954 mutex_exit(&stp->sd_lock);
2955 2955 }
2956 2956 }
2957 2957
2958 2958 isevent = 0;
2959 2959 i = 1;
2960 2960 bzero((caddr_t)qbf, NBAND);
2961 2961 mutex_enter(QLOCK(tq));
2962 2962 if ((myqbp = q->q_bandp) != NULL)
2963 2963 for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2964 2964 ASSERT(myqbp);
2965 2965 if ((myqbp->qb_flag & QB_BACK)) {
2966 2966 if (qbp->qb_flag & QB_FULL) {
2967 2967 /*
2968 2968 * The band must have become full again.
2969 2969 * Set QB_WANTW again so strwsrv will
2970 2970 * be back enabled when the band becomes
2971 2971 * non-full next time.
2972 2972 */
2973 2973 qbp->qb_flag |= QB_WANTW;
2974 2974 } else {
2975 2975 isevent = 1;
2976 2976 qbf[i] = 1;
2977 2977 }
2978 2978 }
2979 2979 myqbp = myqbp->qb_next;
2980 2980 i++;
2981 2981 }
2982 2982 mutex_exit(QLOCK(tq));
2983 2983
2984 2984 if (isevent) {
2985 2985 for (i = tq->q_nband; i; i--) {
2986 2986 if (qbf[i]) {
2987 2987 pollwakeup(&stp->sd_pollist, POLLWRBAND);
2988 2988 mutex_enter(&stp->sd_lock);
2989 2989 if (stp->sd_sigflags & S_WRBAND)
2990 2990 strsendsig(stp->sd_siglist, S_WRBAND,
2991 2991 (uchar_t)i, 0);
2992 2992 mutex_exit(&stp->sd_lock);
2993 2993 }
2994 2994 }
2995 2995 }
2996 2996
2997 2997 releasestr(q);
2998 2998 return (0);
2999 2999 }
3000 3000
3001 3001 /*
3002 3002 * Special case of strcopyin/strcopyout for copying
3003 3003 * struct strioctl that can deal with both data
3004 3004 * models.
3005 3005 */
3006 3006
3007 3007 #ifdef _LP64
3008 3008
3009 3009 static int
3010 3010 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3011 3011 {
3012 3012 struct strioctl32 strioc32;
3013 3013 struct strioctl *striocp;
3014 3014
3015 3015 if (copyflag & U_TO_K) {
3016 3016 ASSERT((copyflag & K_TO_K) == 0);
3017 3017
3018 3018 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3019 3019 if (copyin(from, &strioc32, sizeof (strioc32)))
3020 3020 return (EFAULT);
3021 3021
3022 3022 striocp = (struct strioctl *)to;
3023 3023 striocp->ic_cmd = strioc32.ic_cmd;
3024 3024 striocp->ic_timout = strioc32.ic_timout;
3025 3025 striocp->ic_len = strioc32.ic_len;
3026 3026 striocp->ic_dp = (char *)(uintptr_t)strioc32.ic_dp;
3027 3027
3028 3028 } else { /* NATIVE data model */
3029 3029 if (copyin(from, to, sizeof (struct strioctl))) {
3030 3030 return (EFAULT);
3031 3031 } else {
3032 3032 return (0);
3033 3033 }
3034 3034 }
3035 3035 } else {
3036 3036 ASSERT(copyflag & K_TO_K);
3037 3037 bcopy(from, to, sizeof (struct strioctl));
3038 3038 }
3039 3039 return (0);
3040 3040 }
3041 3041
3042 3042 static int
3043 3043 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3044 3044 {
3045 3045 struct strioctl32 strioc32;
3046 3046 struct strioctl *striocp;
3047 3047
3048 3048 if (copyflag & U_TO_K) {
3049 3049 ASSERT((copyflag & K_TO_K) == 0);
3050 3050
3051 3051 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3052 3052 striocp = (struct strioctl *)from;
3053 3053 strioc32.ic_cmd = striocp->ic_cmd;
3054 3054 strioc32.ic_timout = striocp->ic_timout;
3055 3055 strioc32.ic_len = striocp->ic_len;
3056 3056 strioc32.ic_dp = (caddr32_t)(uintptr_t)striocp->ic_dp;
3057 3057 ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3058 3058 striocp->ic_dp);
3059 3059
3060 3060 if (copyout(&strioc32, to, sizeof (strioc32)))
3061 3061 return (EFAULT);
3062 3062
3063 3063 } else { /* NATIVE data model */
3064 3064 if (copyout(from, to, sizeof (struct strioctl))) {
3065 3065 return (EFAULT);
3066 3066 } else {
3067 3067 return (0);
3068 3068 }
3069 3069 }
3070 3070 } else {
3071 3071 ASSERT(copyflag & K_TO_K);
3072 3072 bcopy(from, to, sizeof (struct strioctl));
3073 3073 }
3074 3074 return (0);
3075 3075 }
3076 3076
3077 3077 #else /* ! _LP64 */
3078 3078
3079 3079 /* ARGSUSED2 */
3080 3080 static int
3081 3081 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3082 3082 {
3083 3083 return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3084 3084 }
3085 3085
3086 3086 /* ARGSUSED2 */
3087 3087 static int
3088 3088 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3089 3089 {
3090 3090 return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3091 3091 }
3092 3092
3093 3093 #endif /* _LP64 */
3094 3094
3095 3095 /*
3096 3096 * Determine type of job control semantics expected by user. The
3097 3097 * possibilities are:
3098 3098 * JCREAD - Behaves like read() on fd; send SIGTTIN
3099 3099 * JCWRITE - Behaves like write() on fd; send SIGTTOU if TOSTOP set
3100 3100 * JCSETP - Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3101 3101 * JCGETP - Gets a value in the stream; no signals.
3102 3102 * See straccess in strsubr.c for usage of these values.
3103 3103 *
3104 3104 * This routine also returns -1 for I_STR as a special case; the
3105 3105 * caller must call again with the real ioctl number for
3106 3106 * classification.
3107 3107 */
3108 3108 static int
3109 3109 job_control_type(int cmd)
3110 3110 {
3111 3111 switch (cmd) {
3112 3112 case I_STR:
3113 3113 return (-1);
3114 3114
3115 3115 case I_RECVFD:
3116 3116 case I_E_RECVFD:
3117 3117 return (JCREAD);
3118 3118
3119 3119 case I_FDINSERT:
3120 3120 case I_SENDFD:
3121 3121 return (JCWRITE);
3122 3122
3123 3123 case TCSETA:
3124 3124 case TCSETAW:
3125 3125 case TCSETAF:
3126 3126 case TCSBRK:
3127 3127 case TCXONC:
3128 3128 case TCFLSH:
3129 3129 case TCDSET: /* Obsolete */
3130 3130 case TIOCSWINSZ:
3131 3131 case TCSETS:
3132 3132 case TCSETSW:
3133 3133 case TCSETSF:
3134 3134 case TIOCSETD:
3135 3135 case TIOCHPCL:
3136 3136 case TIOCSETP:
3137 3137 case TIOCSETN:
3138 3138 case TIOCEXCL:
3139 3139 case TIOCNXCL:
3140 3140 case TIOCFLUSH:
3141 3141 case TIOCSETC:
3142 3142 case TIOCLBIS:
3143 3143 case TIOCLBIC:
3144 3144 case TIOCLSET:
3145 3145 case TIOCSBRK:
3146 3146 case TIOCCBRK:
3147 3147 case TIOCSDTR:
3148 3148 case TIOCCDTR:
3149 3149 case TIOCSLTC:
3150 3150 case TIOCSTOP:
3151 3151 case TIOCSTART:
3152 3152 case TIOCSTI:
3153 3153 case TIOCSPGRP:
3154 3154 case TIOCMSET:
3155 3155 case TIOCMBIS:
3156 3156 case TIOCMBIC:
3157 3157 case TIOCREMOTE:
3158 3158 case TIOCSIGNAL:
3159 3159 case LDSETT:
3160 3160 case LDSMAP: /* Obsolete */
3161 3161 case DIOCSETP:
3162 3162 case I_FLUSH:
3163 3163 case I_SRDOPT:
3164 3164 case I_SETSIG:
3165 3165 case I_SWROPT:
3166 3166 case I_FLUSHBAND:
3167 3167 case I_SETCLTIME:
3168 3168 case I_SERROPT:
3169 3169 case I_ESETSIG:
|
↓ open down ↓ |
3169 lines elided |
↑ open up ↑ |
3170 3170 case FIONBIO:
3171 3171 case FIOASYNC:
3172 3172 case FIOSETOWN:
3173 3173 case JBOOT: /* Obsolete */
3174 3174 case JTERM: /* Obsolete */
3175 3175 case JTIMOM: /* Obsolete */
3176 3176 case JZOMBOOT: /* Obsolete */
3177 3177 case JAGENT: /* Obsolete */
3178 3178 case JTRUN: /* Obsolete */
3179 3179 case JXTPROTO: /* Obsolete */
3180 + case TIOCSETLD:
3180 3181 return (JCSETP);
3181 3182 }
3182 3183
3183 3184 return (JCGETP);
3184 3185 }
3185 3186
3186 3187 /*
3187 3188 * ioctl for streams
3188 3189 */
3189 3190 int
3190 3191 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3191 3192 cred_t *crp, int *rvalp)
3192 3193 {
3193 3194 struct stdata *stp;
3194 3195 struct strcmd *scp;
3195 3196 struct strioctl strioc;
3196 3197 struct uio uio;
3197 3198 struct iovec iov;
3198 3199 int access;
3199 3200 mblk_t *mp;
3200 3201 int error = 0;
3201 3202 int done = 0;
3202 3203 ssize_t rmin, rmax;
3203 3204 queue_t *wrq;
3204 3205 queue_t *rdq;
3205 3206 boolean_t kioctl = B_FALSE;
3206 3207 uint32_t auditing = AU_AUDITING();
3207 3208
3208 3209 if (flag & FKIOCTL) {
3209 3210 copyflag = K_TO_K;
3210 3211 kioctl = B_TRUE;
3211 3212 }
3212 3213 ASSERT(vp->v_stream);
3213 3214 ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3214 3215 stp = vp->v_stream;
3215 3216
3216 3217 TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3217 3218 "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3218 3219
3219 3220 /*
3220 3221 * If the copy is kernel to kernel, make sure that the FNATIVE
3221 3222 * flag is set. After this it would be a serious error to have
3222 3223 * no model flag.
3223 3224 */
3224 3225 if (copyflag == K_TO_K)
3225 3226 flag = (flag & ~FMODELS) | FNATIVE;
3226 3227
3227 3228 ASSERT((flag & FMODELS) != 0);
3228 3229
3229 3230 wrq = stp->sd_wrq;
3230 3231 rdq = _RD(wrq);
3231 3232
3232 3233 access = job_control_type(cmd);
3233 3234
3234 3235 /* We should never see these here, should be handled by iwscn */
3235 3236 if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3236 3237 return (EINVAL);
3237 3238
3238 3239 mutex_enter(&stp->sd_lock);
3239 3240 if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3240 3241 mutex_exit(&stp->sd_lock);
3241 3242 return (error);
3242 3243 }
3243 3244 mutex_exit(&stp->sd_lock);
3244 3245
3245 3246 /*
3246 3247 * Check for sgttyb-related ioctls first, and complain as
3247 3248 * necessary.
3248 3249 */
3249 3250 switch (cmd) {
3250 3251 case TIOCGETP:
3251 3252 case TIOCSETP:
3252 3253 case TIOCSETN:
3253 3254 if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3254 3255 sgttyb_complaint = B_TRUE;
3255 3256 cmn_err(CE_NOTE,
3256 3257 "application used obsolete TIOC[GS]ET");
3257 3258 }
3258 3259 if (sgttyb_handling >= 3) {
3259 3260 tsignal(curthread, SIGSYS);
3260 3261 return (EIO);
3261 3262 }
3262 3263 break;
3263 3264 }
3264 3265
3265 3266 mutex_enter(&stp->sd_lock);
3266 3267
3267 3268 switch (cmd) {
3268 3269 case I_RECVFD:
3269 3270 case I_E_RECVFD:
3270 3271 case I_PEEK:
3271 3272 case I_NREAD:
3272 3273 case FIONREAD:
3273 3274 case FIORDCHK:
3274 3275 case I_ATMARK:
3275 3276 case FIONBIO:
3276 3277 case FIOASYNC:
3277 3278 if (stp->sd_flag & (STRDERR|STPLEX)) {
3278 3279 error = strgeterr(stp, STRDERR|STPLEX, 0);
3279 3280 if (error != 0) {
3280 3281 mutex_exit(&stp->sd_lock);
3281 3282 return (error);
3282 3283 }
3283 3284 }
3284 3285 break;
3285 3286
3286 3287 default:
3287 3288 if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3288 3289 error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3289 3290 if (error != 0) {
3290 3291 mutex_exit(&stp->sd_lock);
3291 3292 return (error);
3292 3293 }
3293 3294 }
3294 3295 }
3295 3296
3296 3297 mutex_exit(&stp->sd_lock);
3297 3298
3298 3299 switch (cmd) {
3299 3300 default:
3300 3301 /*
3301 3302 * The stream head has hardcoded knowledge of a
3302 3303 * miscellaneous collection of terminal-, keyboard- and
3303 3304 * mouse-related ioctls, enumerated below. This hardcoded
3304 3305 * knowledge allows the stream head to automatically
3305 3306 * convert transparent ioctl requests made by userland
3306 3307 * programs into I_STR ioctls which many old STREAMS
3307 3308 * modules and drivers require.
3308 3309 *
3309 3310 * No new ioctls should ever be added to this list.
3310 3311 * Instead, the STREAMS module or driver should be written
3311 3312 * to either handle transparent ioctls or require any
3312 3313 * userland programs to use I_STR ioctls (by returning
3313 3314 * EINVAL to any transparent ioctl requests).
3314 3315 *
3315 3316 * More importantly, removing ioctls from this list should
3316 3317 * be done with the utmost care, since our STREAMS modules
3317 3318 * and drivers *count* on the stream head performing this
3318 3319 * conversion, and thus may panic while processing
3319 3320 * transparent ioctl request for one of these ioctls (keep
3320 3321 * in mind that third party modules and drivers may have
3321 3322 * similar problems).
3322 3323 */
3323 3324 if (((cmd & IOCTYPE) == LDIOC) ||
3324 3325 ((cmd & IOCTYPE) == tIOC) ||
3325 3326 ((cmd & IOCTYPE) == TIOC) ||
3326 3327 ((cmd & IOCTYPE) == KIOC) ||
3327 3328 ((cmd & IOCTYPE) == MSIOC) ||
3328 3329 ((cmd & IOCTYPE) == VUIOC)) {
3329 3330 /*
3330 3331 * The ioctl is a tty ioctl - set up strioc buffer
3331 3332 * and call strdoioctl() to do the work.
3332 3333 */
3333 3334 if (stp->sd_flag & STRHUP)
3334 3335 return (ENXIO);
3335 3336 strioc.ic_cmd = cmd;
3336 3337 strioc.ic_timout = INFTIM;
3337 3338
3338 3339 switch (cmd) {
3339 3340
3340 3341 case TCXONC:
3341 3342 case TCSBRK:
3342 3343 case TCFLSH:
3343 3344 case TCDSET:
3344 3345 {
3345 3346 int native_arg = (int)arg;
3346 3347 strioc.ic_len = sizeof (int);
3347 3348 strioc.ic_dp = (char *)&native_arg;
3348 3349 return (strdoioctl(stp, &strioc, flag,
3349 3350 K_TO_K, crp, rvalp));
3350 3351 }
3351 3352
3352 3353 case TCSETA:
3353 3354 case TCSETAW:
3354 3355 case TCSETAF:
3355 3356 strioc.ic_len = sizeof (struct termio);
3356 3357 strioc.ic_dp = (char *)arg;
3357 3358 return (strdoioctl(stp, &strioc, flag,
3358 3359 copyflag, crp, rvalp));
3359 3360
3360 3361 case TCSETS:
3361 3362 case TCSETSW:
3362 3363 case TCSETSF:
3363 3364 strioc.ic_len = sizeof (struct termios);
3364 3365 strioc.ic_dp = (char *)arg;
3365 3366 return (strdoioctl(stp, &strioc, flag,
3366 3367 copyflag, crp, rvalp));
3367 3368
3368 3369 case LDSETT:
3369 3370 strioc.ic_len = sizeof (struct termcb);
3370 3371 strioc.ic_dp = (char *)arg;
3371 3372 return (strdoioctl(stp, &strioc, flag,
3372 3373 copyflag, crp, rvalp));
3373 3374
3374 3375 case TIOCSETP:
3375 3376 strioc.ic_len = sizeof (struct sgttyb);
3376 3377 strioc.ic_dp = (char *)arg;
3377 3378 return (strdoioctl(stp, &strioc, flag,
3378 3379 copyflag, crp, rvalp));
3379 3380
3380 3381 case TIOCSTI:
3381 3382 if ((flag & FREAD) == 0 &&
3382 3383 secpolicy_sti(crp) != 0) {
3383 3384 return (EPERM);
3384 3385 }
3385 3386 mutex_enter(&stp->sd_lock);
3386 3387 mutex_enter(&curproc->p_splock);
3387 3388 if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3388 3389 secpolicy_sti(crp) != 0) {
3389 3390 mutex_exit(&curproc->p_splock);
3390 3391 mutex_exit(&stp->sd_lock);
3391 3392 return (EACCES);
3392 3393 }
3393 3394 mutex_exit(&curproc->p_splock);
3394 3395 mutex_exit(&stp->sd_lock);
3395 3396
3396 3397 strioc.ic_len = sizeof (char);
3397 3398 strioc.ic_dp = (char *)arg;
3398 3399 return (strdoioctl(stp, &strioc, flag,
3399 3400 copyflag, crp, rvalp));
3400 3401
3401 3402 case TIOCSWINSZ:
3402 3403 strioc.ic_len = sizeof (struct winsize);
3403 3404 strioc.ic_dp = (char *)arg;
3404 3405 return (strdoioctl(stp, &strioc, flag,
3405 3406 copyflag, crp, rvalp));
3406 3407
3407 3408 case TIOCSSIZE:
3408 3409 strioc.ic_len = sizeof (struct ttysize);
3409 3410 strioc.ic_dp = (char *)arg;
3410 3411 return (strdoioctl(stp, &strioc, flag,
3411 3412 copyflag, crp, rvalp));
3412 3413
3413 3414 case TIOCSSOFTCAR:
3414 3415 case KIOCTRANS:
3415 3416 case KIOCTRANSABLE:
3416 3417 case KIOCCMD:
3417 3418 case KIOCSDIRECT:
3418 3419 case KIOCSCOMPAT:
3419 3420 case KIOCSKABORTEN:
3420 3421 case KIOCSRPTDELAY:
3421 3422 case KIOCSRPTRATE:
3422 3423 case VUIDSFORMAT:
3423 3424 case TIOCSPPS:
3424 3425 strioc.ic_len = sizeof (int);
3425 3426 strioc.ic_dp = (char *)arg;
3426 3427 return (strdoioctl(stp, &strioc, flag,
3427 3428 copyflag, crp, rvalp));
3428 3429
3429 3430 case KIOCSETKEY:
3430 3431 case KIOCGETKEY:
3431 3432 strioc.ic_len = sizeof (struct kiockey);
3432 3433 strioc.ic_dp = (char *)arg;
3433 3434 return (strdoioctl(stp, &strioc, flag,
3434 3435 copyflag, crp, rvalp));
3435 3436
3436 3437 case KIOCSKEY:
3437 3438 case KIOCGKEY:
3438 3439 strioc.ic_len = sizeof (struct kiockeymap);
3439 3440 strioc.ic_dp = (char *)arg;
3440 3441 return (strdoioctl(stp, &strioc, flag,
3441 3442 copyflag, crp, rvalp));
3442 3443
3443 3444 case KIOCSLED:
3444 3445 /* arg is a pointer to char */
3445 3446 strioc.ic_len = sizeof (char);
3446 3447 strioc.ic_dp = (char *)arg;
3447 3448 return (strdoioctl(stp, &strioc, flag,
3448 3449 copyflag, crp, rvalp));
3449 3450
3450 3451 case MSIOSETPARMS:
3451 3452 strioc.ic_len = sizeof (Ms_parms);
3452 3453 strioc.ic_dp = (char *)arg;
3453 3454 return (strdoioctl(stp, &strioc, flag,
3454 3455 copyflag, crp, rvalp));
3455 3456
3456 3457 case VUIDSADDR:
3457 3458 case VUIDGADDR:
3458 3459 strioc.ic_len = sizeof (struct vuid_addr_probe);
3459 3460 strioc.ic_dp = (char *)arg;
3460 3461 return (strdoioctl(stp, &strioc, flag,
3461 3462 copyflag, crp, rvalp));
3462 3463
3463 3464 /*
3464 3465 * These M_IOCTL's don't require any data to be sent
3465 3466 * downstream, and the driver will allocate and link
3466 3467 * on its own mblk_t upon M_IOCACK -- thus we set
3467 3468 * ic_len to zero and set ic_dp to arg so we know
3468 3469 * where to copyout to later.
3469 3470 */
3470 3471 case TIOCGSOFTCAR:
3471 3472 case TIOCGWINSZ:
3472 3473 case TIOCGSIZE:
3473 3474 case KIOCGTRANS:
3474 3475 case KIOCGTRANSABLE:
3475 3476 case KIOCTYPE:
3476 3477 case KIOCGDIRECT:
3477 3478 case KIOCGCOMPAT:
3478 3479 case KIOCLAYOUT:
3479 3480 case KIOCGLED:
3480 3481 case MSIOGETPARMS:
3481 3482 case MSIOBUTTONS:
3482 3483 case VUIDGFORMAT:
3483 3484 case TIOCGPPS:
3484 3485 case TIOCGPPSEV:
3485 3486 case TCGETA:
3486 3487 case TCGETS:
3487 3488 case LDGETT:
3488 3489 case TIOCGETP:
3489 3490 case KIOCGRPTDELAY:
3490 3491 case KIOCGRPTRATE:
3491 3492 strioc.ic_len = 0;
3492 3493 strioc.ic_dp = (char *)arg;
3493 3494 return (strdoioctl(stp, &strioc, flag,
3494 3495 copyflag, crp, rvalp));
3495 3496 }
3496 3497 }
3497 3498
3498 3499 /*
3499 3500 * Unknown cmd - send it down as a transparent ioctl.
3500 3501 */
3501 3502 strioc.ic_cmd = cmd;
3502 3503 strioc.ic_timout = INFTIM;
3503 3504 strioc.ic_len = TRANSPARENT;
3504 3505 strioc.ic_dp = (char *)&arg;
3505 3506
3506 3507 return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3507 3508
3508 3509 case I_STR:
3509 3510 /*
3510 3511 * Stream ioctl. Read in an strioctl buffer from the user
3511 3512 * along with any data specified and send it downstream.
3512 3513 * Strdoioctl will wait allow only one ioctl message at
3513 3514 * a time, and waits for the acknowledgement.
3514 3515 */
3515 3516
3516 3517 if (stp->sd_flag & STRHUP)
3517 3518 return (ENXIO);
3518 3519
3519 3520 error = strcopyin_strioctl((void *)arg, &strioc, flag,
3520 3521 copyflag);
3521 3522 if (error != 0)
3522 3523 return (error);
3523 3524
3524 3525 if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3525 3526 return (EINVAL);
3526 3527
3527 3528 access = job_control_type(strioc.ic_cmd);
3528 3529 mutex_enter(&stp->sd_lock);
3529 3530 if ((access != -1) &&
3530 3531 ((error = i_straccess(stp, access)) != 0)) {
3531 3532 mutex_exit(&stp->sd_lock);
3532 3533 return (error);
3533 3534 }
3534 3535 mutex_exit(&stp->sd_lock);
3535 3536
3536 3537 /*
3537 3538 * The I_STR facility provides a trap door for malicious
3538 3539 * code to send down bogus streamio(7I) ioctl commands to
3539 3540 * unsuspecting STREAMS modules and drivers which expect to
3540 3541 * only get these messages from the stream head.
3541 3542 * Explicitly prohibit any streamio ioctls which can be
3542 3543 * passed downstream by the stream head. Note that we do
3543 3544 * not block all streamio ioctls because the ioctl
3544 3545 * numberspace is not well managed and thus it's possible
3545 3546 * that a module or driver's ioctl numbers may accidentally
3546 3547 * collide with them.
3547 3548 */
3548 3549 switch (strioc.ic_cmd) {
3549 3550 case I_LINK:
3550 3551 case I_PLINK:
3551 3552 case I_UNLINK:
3552 3553 case I_PUNLINK:
3553 3554 case _I_GETPEERCRED:
3554 3555 case _I_PLINK_LH:
3555 3556 return (EINVAL);
3556 3557 }
3557 3558
3558 3559 error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3559 3560 if (error == 0) {
3560 3561 error = strcopyout_strioctl(&strioc, (void *)arg,
3561 3562 flag, copyflag);
3562 3563 }
3563 3564 return (error);
3564 3565
3565 3566 case _I_CMD:
3566 3567 /*
3567 3568 * Like I_STR, but without using M_IOC* messages and without
3568 3569 * copyins/copyouts beyond the passed-in argument.
3569 3570 */
3570 3571 if (stp->sd_flag & STRHUP)
3571 3572 return (ENXIO);
3572 3573
3573 3574 if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3574 3575 return (ENOMEM);
3575 3576
3576 3577 if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3577 3578 kmem_free(scp, sizeof (strcmd_t));
3578 3579 return (EFAULT);
3579 3580 }
3580 3581
3581 3582 access = job_control_type(scp->sc_cmd);
3582 3583 mutex_enter(&stp->sd_lock);
3583 3584 if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3584 3585 mutex_exit(&stp->sd_lock);
3585 3586 kmem_free(scp, sizeof (strcmd_t));
3586 3587 return (error);
3587 3588 }
3588 3589 mutex_exit(&stp->sd_lock);
3589 3590
3590 3591 *rvalp = 0;
3591 3592 if ((error = strdocmd(stp, scp, crp)) == 0) {
3592 3593 if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3593 3594 error = EFAULT;
3594 3595 }
3595 3596 kmem_free(scp, sizeof (strcmd_t));
3596 3597 return (error);
3597 3598
3598 3599 case I_NREAD:
3599 3600 /*
3600 3601 * Return number of bytes of data in first message
3601 3602 * in queue in "arg" and return the number of messages
3602 3603 * in queue in return value.
3603 3604 */
3604 3605 {
3605 3606 size_t size;
3606 3607 int retval;
3607 3608 int count = 0;
3608 3609
3609 3610 mutex_enter(QLOCK(rdq));
3610 3611
3611 3612 size = msgdsize(rdq->q_first);
3612 3613 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3613 3614 count++;
3614 3615
3615 3616 mutex_exit(QLOCK(rdq));
3616 3617 if (stp->sd_struiordq) {
3617 3618 infod_t infod;
3618 3619
3619 3620 infod.d_cmd = INFOD_COUNT;
3620 3621 infod.d_count = 0;
3621 3622 if (count == 0) {
3622 3623 infod.d_cmd |= INFOD_FIRSTBYTES;
3623 3624 infod.d_bytes = 0;
3624 3625 }
3625 3626 infod.d_res = 0;
3626 3627 (void) infonext(rdq, &infod);
3627 3628 count += infod.d_count;
3628 3629 if (infod.d_res & INFOD_FIRSTBYTES)
3629 3630 size = infod.d_bytes;
3630 3631 }
3631 3632
3632 3633 /*
3633 3634 * Drop down from size_t to the "int" required by the
3634 3635 * interface. Cap at INT_MAX.
3635 3636 */
3636 3637 retval = MIN(size, INT_MAX);
3637 3638 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3638 3639 copyflag);
3639 3640 if (!error)
3640 3641 *rvalp = count;
3641 3642 return (error);
3642 3643 }
3643 3644
3644 3645 case FIONREAD:
3645 3646 /*
3646 3647 * Return number of bytes of data in all data messages
3647 3648 * in queue in "arg".
3648 3649 */
3649 3650 {
3650 3651 size_t size = 0;
3651 3652 int retval;
3652 3653
3653 3654 mutex_enter(QLOCK(rdq));
3654 3655 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3655 3656 size += msgdsize(mp);
3656 3657 mutex_exit(QLOCK(rdq));
3657 3658
3658 3659 if (stp->sd_struiordq) {
3659 3660 infod_t infod;
3660 3661
3661 3662 infod.d_cmd = INFOD_BYTES;
3662 3663 infod.d_res = 0;
3663 3664 infod.d_bytes = 0;
3664 3665 (void) infonext(rdq, &infod);
3665 3666 size += infod.d_bytes;
3666 3667 }
3667 3668
3668 3669 /*
3669 3670 * Drop down from size_t to the "int" required by the
3670 3671 * interface. Cap at INT_MAX.
3671 3672 */
3672 3673 retval = MIN(size, INT_MAX);
3673 3674 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3674 3675 copyflag);
3675 3676
3676 3677 *rvalp = 0;
3677 3678 return (error);
3678 3679 }
3679 3680 case FIORDCHK:
3680 3681 /*
3681 3682 * FIORDCHK does not use arg value (like FIONREAD),
3682 3683 * instead a count is returned. I_NREAD value may
3683 3684 * not be accurate but safe. The real thing to do is
3684 3685 * to add the msgdsizes of all data messages until
3685 3686 * a non-data message.
3686 3687 */
3687 3688 {
3688 3689 size_t size = 0;
3689 3690
3690 3691 mutex_enter(QLOCK(rdq));
3691 3692 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3692 3693 size += msgdsize(mp);
3693 3694 mutex_exit(QLOCK(rdq));
3694 3695
3695 3696 if (stp->sd_struiordq) {
3696 3697 infod_t infod;
3697 3698
3698 3699 infod.d_cmd = INFOD_BYTES;
3699 3700 infod.d_res = 0;
3700 3701 infod.d_bytes = 0;
3701 3702 (void) infonext(rdq, &infod);
3702 3703 size += infod.d_bytes;
3703 3704 }
3704 3705
3705 3706 /*
3706 3707 * Since ioctl returns an int, and memory sizes under
3707 3708 * LP64 may not fit, we return INT_MAX if the count was
3708 3709 * actually greater.
3709 3710 */
3710 3711 *rvalp = MIN(size, INT_MAX);
3711 3712 return (0);
3712 3713 }
3713 3714
3714 3715 case I_FIND:
3715 3716 /*
3716 3717 * Get module name.
3717 3718 */
3718 3719 {
3719 3720 char mname[FMNAMESZ + 1];
3720 3721 queue_t *q;
3721 3722
3722 3723 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3723 3724 mname, FMNAMESZ + 1, NULL);
3724 3725 if (error)
3725 3726 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3726 3727
3727 3728 /*
3728 3729 * Return EINVAL if we're handed a bogus module name.
3729 3730 */
3730 3731 if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3731 3732 TRACE_0(TR_FAC_STREAMS_FR,
3732 3733 TR_I_CANT_FIND, "couldn't I_FIND");
3733 3734 return (EINVAL);
3734 3735 }
3735 3736
3736 3737 *rvalp = 0;
3737 3738
3738 3739 /* Look downstream to see if module is there. */
3739 3740 claimstr(stp->sd_wrq);
3740 3741 for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3741 3742 if (q->q_flag & QREADR) {
3742 3743 q = NULL;
3743 3744 break;
3744 3745 }
3745 3746 if (strcmp(mname, Q2NAME(q)) == 0)
3746 3747 break;
3747 3748 }
3748 3749 releasestr(stp->sd_wrq);
3749 3750
3750 3751 *rvalp = (q ? 1 : 0);
3751 3752 return (error);
3752 3753 }
3753 3754
3754 3755 case I_PUSH:
3755 3756 case __I_PUSH_NOCTTY:
3756 3757 /*
3757 3758 * Push a module.
3758 3759 * For the case __I_PUSH_NOCTTY push a module but
3759 3760 * do not allocate controlling tty. See bugid 4025044
3760 3761 */
3761 3762
3762 3763 {
3763 3764 char mname[FMNAMESZ + 1];
3764 3765 fmodsw_impl_t *fp;
3765 3766 dev_t dummydev;
3766 3767
3767 3768 if (stp->sd_flag & STRHUP)
3768 3769 return (ENXIO);
3769 3770
3770 3771 /*
3771 3772 * Get module name and look up in fmodsw.
3772 3773 */
3773 3774 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3774 3775 mname, FMNAMESZ + 1, NULL);
3775 3776 if (error)
3776 3777 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3777 3778
3778 3779 if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3779 3780 NULL)
3780 3781 return (EINVAL);
3781 3782
3782 3783 TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3783 3784 "I_PUSH:fp %p stp %p", fp, stp);
3784 3785
3785 3786 if (error = strstartplumb(stp, flag, cmd)) {
3786 3787 fmodsw_rele(fp);
3787 3788 return (error);
3788 3789 }
3789 3790
3790 3791 /*
3791 3792 * See if any more modules can be pushed on this stream.
3792 3793 * Note that this check must be done after strstartplumb()
3793 3794 * since otherwise multiple threads issuing I_PUSHes on
3794 3795 * the same stream will be able to exceed nstrpush.
3795 3796 */
3796 3797 mutex_enter(&stp->sd_lock);
3797 3798 if (stp->sd_pushcnt >= nstrpush) {
3798 3799 fmodsw_rele(fp);
3799 3800 strendplumb(stp);
3800 3801 mutex_exit(&stp->sd_lock);
3801 3802 return (EINVAL);
3802 3803 }
3803 3804 mutex_exit(&stp->sd_lock);
3804 3805
3805 3806 /*
3806 3807 * Push new module and call its open routine
3807 3808 * via qattach(). Modules don't change device
3808 3809 * numbers, so just ignore dummydev here.
3809 3810 */
3810 3811 dummydev = vp->v_rdev;
3811 3812 if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3812 3813 B_FALSE)) == 0) {
3813 3814 if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3814 3815 (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3815 3816 /*
3816 3817 * try to allocate it as a controlling terminal
3817 3818 */
3818 3819 (void) strctty(stp);
3819 3820 }
3820 3821 }
3821 3822
3822 3823 mutex_enter(&stp->sd_lock);
3823 3824
3824 3825 /*
3825 3826 * As a performance concern we are caching the values of
3826 3827 * q_minpsz and q_maxpsz of the module below the stream
3827 3828 * head in the stream head.
3828 3829 */
3829 3830 mutex_enter(QLOCK(stp->sd_wrq->q_next));
3830 3831 rmin = stp->sd_wrq->q_next->q_minpsz;
3831 3832 rmax = stp->sd_wrq->q_next->q_maxpsz;
3832 3833 mutex_exit(QLOCK(stp->sd_wrq->q_next));
3833 3834
3834 3835 /* Do this processing here as a performance concern */
3835 3836 if (strmsgsz != 0) {
3836 3837 if (rmax == INFPSZ)
3837 3838 rmax = strmsgsz;
3838 3839 else {
3839 3840 if (vp->v_type == VFIFO)
3840 3841 rmax = MIN(PIPE_BUF, rmax);
3841 3842 else rmax = MIN(strmsgsz, rmax);
3842 3843 }
3843 3844 }
3844 3845
3845 3846 mutex_enter(QLOCK(wrq));
3846 3847 stp->sd_qn_minpsz = rmin;
3847 3848 stp->sd_qn_maxpsz = rmax;
3848 3849 mutex_exit(QLOCK(wrq));
3849 3850
3850 3851 strendplumb(stp);
3851 3852 mutex_exit(&stp->sd_lock);
3852 3853 return (error);
3853 3854 }
3854 3855
3855 3856 case I_POP:
3856 3857 {
3857 3858 queue_t *q;
3858 3859
3859 3860 if (stp->sd_flag & STRHUP)
3860 3861 return (ENXIO);
3861 3862 if (!wrq->q_next) /* for broken pipes */
3862 3863 return (EINVAL);
3863 3864
3864 3865 if (error = strstartplumb(stp, flag, cmd))
3865 3866 return (error);
3866 3867
3867 3868 /*
3868 3869 * If there is an anchor on this stream and popping
3869 3870 * the current module would attempt to pop through the
3870 3871 * anchor, then disallow the pop unless we have sufficient
3871 3872 * privileges; take the cheapest (non-locking) check
3872 3873 * first.
3873 3874 */
3874 3875 if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3875 3876 (stp->sd_anchorzone != crgetzoneid(crp))) {
3876 3877 mutex_enter(&stp->sd_lock);
3877 3878 /*
3878 3879 * Anchors only apply if there's at least one
3879 3880 * module on the stream (sd_pushcnt > 0).
3880 3881 */
3881 3882 if (stp->sd_pushcnt > 0 &&
3882 3883 stp->sd_pushcnt == stp->sd_anchor &&
3883 3884 stp->sd_vnode->v_type != VFIFO) {
3884 3885 strendplumb(stp);
3885 3886 mutex_exit(&stp->sd_lock);
3886 3887 if (stp->sd_anchorzone != crgetzoneid(crp))
3887 3888 return (EINVAL);
3888 3889 /* Audit and report error */
3889 3890 return (secpolicy_ip_config(crp, B_FALSE));
3890 3891 }
3891 3892 mutex_exit(&stp->sd_lock);
3892 3893 }
3893 3894
3894 3895 q = wrq->q_next;
3895 3896 TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3896 3897 "I_POP:%p from %p", q, stp);
3897 3898 if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3898 3899 error = EINVAL;
3899 3900 } else {
3900 3901 qdetach(_RD(q), 1, flag, crp, B_FALSE);
3901 3902 error = 0;
3902 3903 }
3903 3904 mutex_enter(&stp->sd_lock);
3904 3905
3905 3906 /*
3906 3907 * As a performance concern we are caching the values of
3907 3908 * q_minpsz and q_maxpsz of the module below the stream
3908 3909 * head in the stream head.
3909 3910 */
3910 3911 mutex_enter(QLOCK(wrq->q_next));
3911 3912 rmin = wrq->q_next->q_minpsz;
3912 3913 rmax = wrq->q_next->q_maxpsz;
3913 3914 mutex_exit(QLOCK(wrq->q_next));
3914 3915
3915 3916 /* Do this processing here as a performance concern */
3916 3917 if (strmsgsz != 0) {
3917 3918 if (rmax == INFPSZ)
3918 3919 rmax = strmsgsz;
3919 3920 else {
3920 3921 if (vp->v_type == VFIFO)
3921 3922 rmax = MIN(PIPE_BUF, rmax);
3922 3923 else rmax = MIN(strmsgsz, rmax);
3923 3924 }
3924 3925 }
3925 3926
3926 3927 mutex_enter(QLOCK(wrq));
3927 3928 stp->sd_qn_minpsz = rmin;
3928 3929 stp->sd_qn_maxpsz = rmax;
3929 3930 mutex_exit(QLOCK(wrq));
3930 3931
3931 3932 /* If we popped through the anchor, then reset the anchor. */
3932 3933 if (stp->sd_pushcnt < stp->sd_anchor) {
3933 3934 stp->sd_anchor = 0;
3934 3935 stp->sd_anchorzone = 0;
3935 3936 }
3936 3937 strendplumb(stp);
3937 3938 mutex_exit(&stp->sd_lock);
3938 3939 return (error);
3939 3940 }
3940 3941
3941 3942 case _I_MUXID2FD:
3942 3943 {
3943 3944 /*
3944 3945 * Create a fd for a I_PLINK'ed lower stream with a given
3945 3946 * muxid. With the fd, application can send down ioctls,
3946 3947 * like I_LIST, to the previously I_PLINK'ed stream. Note
3947 3948 * that after getting the fd, the application has to do an
3948 3949 * I_PUNLINK on the muxid before it can do any operation
3949 3950 * on the lower stream. This is required by spec1170.
3950 3951 *
3951 3952 * The fd used to do this ioctl should point to the same
3952 3953 * controlling device used to do the I_PLINK. If it uses
3953 3954 * a different stream or an invalid muxid, I_MUXID2FD will
3954 3955 * fail. The error code is set to EINVAL.
3955 3956 *
3956 3957 * The intended use of this interface is the following.
3957 3958 * An application I_PLINK'ed a stream and exits. The fd
3958 3959 * to the lower stream is gone. Another application
3959 3960 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3960 3961 */
3961 3962 int muxid = (int)arg;
3962 3963 int fd;
3963 3964 linkinfo_t *linkp;
3964 3965 struct file *fp;
3965 3966 netstack_t *ns;
3966 3967 str_stack_t *ss;
3967 3968
3968 3969 /*
3969 3970 * Do not allow the wildcard muxid. This ioctl is not
3970 3971 * intended to find arbitrary link.
3971 3972 */
3972 3973 if (muxid == 0) {
3973 3974 return (EINVAL);
3974 3975 }
3975 3976
3976 3977 ns = netstack_find_by_cred(crp);
3977 3978 ASSERT(ns != NULL);
3978 3979 ss = ns->netstack_str;
3979 3980 ASSERT(ss != NULL);
3980 3981
3981 3982 mutex_enter(&muxifier);
3982 3983 linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
3983 3984 if (linkp == NULL) {
3984 3985 mutex_exit(&muxifier);
3985 3986 netstack_rele(ss->ss_netstack);
3986 3987 return (EINVAL);
3987 3988 }
3988 3989
3989 3990 if ((fd = ufalloc(0)) == -1) {
3990 3991 mutex_exit(&muxifier);
3991 3992 netstack_rele(ss->ss_netstack);
3992 3993 return (EMFILE);
3993 3994 }
3994 3995 fp = linkp->li_fpdown;
3995 3996 mutex_enter(&fp->f_tlock);
3996 3997 fp->f_count++;
3997 3998 mutex_exit(&fp->f_tlock);
3998 3999 mutex_exit(&muxifier);
3999 4000 setf(fd, fp);
4000 4001 *rvalp = fd;
4001 4002 netstack_rele(ss->ss_netstack);
4002 4003 return (0);
4003 4004 }
4004 4005
4005 4006 case _I_INSERT:
4006 4007 {
4007 4008 /*
4008 4009 * To insert a module to a given position in a stream.
4009 4010 * In the first release, only allow privileged user
4010 4011 * to use this ioctl. Furthermore, the insert is only allowed
4011 4012 * below an anchor if the zoneid is the same as the zoneid
4012 4013 * which created the anchor.
4013 4014 *
4014 4015 * Note that we do not plan to support this ioctl
4015 4016 * on pipes in the first release. We want to learn more
4016 4017 * about the implications of these ioctls before extending
4017 4018 * their support. And we do not think these features are
4018 4019 * valuable for pipes.
4019 4020 */
4020 4021 STRUCT_DECL(strmodconf, strmodinsert);
4021 4022 char mod_name[FMNAMESZ + 1];
4022 4023 fmodsw_impl_t *fp;
4023 4024 dev_t dummydev;
4024 4025 queue_t *tmp_wrq;
4025 4026 int pos;
4026 4027 boolean_t is_insert;
4027 4028
4028 4029 STRUCT_INIT(strmodinsert, flag);
4029 4030 if (stp->sd_flag & STRHUP)
4030 4031 return (ENXIO);
4031 4032 if (STRMATED(stp))
4032 4033 return (EINVAL);
4033 4034 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4034 4035 return (error);
4035 4036 if (stp->sd_anchor != 0 &&
4036 4037 stp->sd_anchorzone != crgetzoneid(crp))
4037 4038 return (EINVAL);
4038 4039
4039 4040 error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4040 4041 STRUCT_SIZE(strmodinsert), copyflag);
4041 4042 if (error)
4042 4043 return (error);
4043 4044
4044 4045 /*
4045 4046 * Get module name and look up in fmodsw.
4046 4047 */
4047 4048 error = (copyflag & U_TO_K ? copyinstr :
4048 4049 copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4049 4050 mod_name, FMNAMESZ + 1, NULL);
4050 4051 if (error)
4051 4052 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4052 4053
4053 4054 if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4054 4055 NULL)
4055 4056 return (EINVAL);
4056 4057
4057 4058 if (error = strstartplumb(stp, flag, cmd)) {
4058 4059 fmodsw_rele(fp);
4059 4060 return (error);
4060 4061 }
4061 4062
4062 4063 /*
4063 4064 * Is this _I_INSERT just like an I_PUSH? We need to know
4064 4065 * this because we do some optimizations if this is a
4065 4066 * module being pushed.
4066 4067 */
4067 4068 pos = STRUCT_FGET(strmodinsert, pos);
4068 4069 is_insert = (pos != 0);
4069 4070
4070 4071 /*
4071 4072 * Make sure pos is valid. Even though it is not an I_PUSH,
4072 4073 * we impose the same limit on the number of modules in a
4073 4074 * stream.
4074 4075 */
4075 4076 mutex_enter(&stp->sd_lock);
4076 4077 if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4077 4078 pos > stp->sd_pushcnt) {
4078 4079 fmodsw_rele(fp);
4079 4080 strendplumb(stp);
4080 4081 mutex_exit(&stp->sd_lock);
4081 4082 return (EINVAL);
4082 4083 }
4083 4084 if (stp->sd_anchor != 0) {
4084 4085 /*
4085 4086 * Is this insert below the anchor?
4086 4087 * Pushcnt hasn't been increased yet hence
4087 4088 * we test for greater than here, and greater or
4088 4089 * equal after qattach.
4089 4090 */
4090 4091 if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4091 4092 stp->sd_anchorzone != crgetzoneid(crp)) {
4092 4093 fmodsw_rele(fp);
4093 4094 strendplumb(stp);
4094 4095 mutex_exit(&stp->sd_lock);
4095 4096 return (EPERM);
4096 4097 }
4097 4098 }
4098 4099
4099 4100 mutex_exit(&stp->sd_lock);
4100 4101
4101 4102 /*
4102 4103 * First find the correct position this module to
4103 4104 * be inserted. We don't need to call claimstr()
4104 4105 * as the stream should not be changing at this point.
4105 4106 *
4106 4107 * Insert new module and call its open routine
4107 4108 * via qattach(). Modules don't change device
4108 4109 * numbers, so just ignore dummydev here.
4109 4110 */
4110 4111 for (tmp_wrq = stp->sd_wrq; pos > 0;
4111 4112 tmp_wrq = tmp_wrq->q_next, pos--) {
4112 4113 ASSERT(SAMESTR(tmp_wrq));
4113 4114 }
4114 4115 dummydev = vp->v_rdev;
4115 4116 if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4116 4117 fp, is_insert)) != 0) {
4117 4118 mutex_enter(&stp->sd_lock);
4118 4119 strendplumb(stp);
4119 4120 mutex_exit(&stp->sd_lock);
4120 4121 return (error);
4121 4122 }
4122 4123
4123 4124 mutex_enter(&stp->sd_lock);
4124 4125
4125 4126 /*
4126 4127 * As a performance concern we are caching the values of
4127 4128 * q_minpsz and q_maxpsz of the module below the stream
4128 4129 * head in the stream head.
4129 4130 */
4130 4131 if (!is_insert) {
4131 4132 mutex_enter(QLOCK(stp->sd_wrq->q_next));
4132 4133 rmin = stp->sd_wrq->q_next->q_minpsz;
4133 4134 rmax = stp->sd_wrq->q_next->q_maxpsz;
4134 4135 mutex_exit(QLOCK(stp->sd_wrq->q_next));
4135 4136
4136 4137 /* Do this processing here as a performance concern */
4137 4138 if (strmsgsz != 0) {
4138 4139 if (rmax == INFPSZ) {
4139 4140 rmax = strmsgsz;
4140 4141 } else {
4141 4142 rmax = MIN(strmsgsz, rmax);
4142 4143 }
4143 4144 }
4144 4145
4145 4146 mutex_enter(QLOCK(wrq));
4146 4147 stp->sd_qn_minpsz = rmin;
4147 4148 stp->sd_qn_maxpsz = rmax;
4148 4149 mutex_exit(QLOCK(wrq));
4149 4150 }
4150 4151
4151 4152 /*
4152 4153 * Need to update the anchor value if this module is
4153 4154 * inserted below the anchor point.
4154 4155 */
4155 4156 if (stp->sd_anchor != 0) {
4156 4157 pos = STRUCT_FGET(strmodinsert, pos);
4157 4158 if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4158 4159 stp->sd_anchor++;
4159 4160 }
4160 4161
4161 4162 strendplumb(stp);
4162 4163 mutex_exit(&stp->sd_lock);
4163 4164 return (0);
4164 4165 }
4165 4166
4166 4167 case _I_REMOVE:
4167 4168 {
4168 4169 /*
4169 4170 * To remove a module with a given name in a stream. The
4170 4171 * caller of this ioctl needs to provide both the name and
4171 4172 * the position of the module to be removed. This eliminates
4172 4173 * the ambiguity of removal if a module is inserted/pushed
4173 4174 * multiple times in a stream. In the first release, only
4174 4175 * allow privileged user to use this ioctl.
4175 4176 * Furthermore, the remove is only allowed
4176 4177 * below an anchor if the zoneid is the same as the zoneid
4177 4178 * which created the anchor.
4178 4179 *
4179 4180 * Note that we do not plan to support this ioctl
4180 4181 * on pipes in the first release. We want to learn more
4181 4182 * about the implications of these ioctls before extending
4182 4183 * their support. And we do not think these features are
4183 4184 * valuable for pipes.
4184 4185 *
4185 4186 * Also note that _I_REMOVE cannot be used to remove a
4186 4187 * driver or the stream head.
4187 4188 */
4188 4189 STRUCT_DECL(strmodconf, strmodremove);
4189 4190 queue_t *q;
4190 4191 int pos;
4191 4192 char mod_name[FMNAMESZ + 1];
4192 4193 boolean_t is_remove;
4193 4194
4194 4195 STRUCT_INIT(strmodremove, flag);
4195 4196 if (stp->sd_flag & STRHUP)
4196 4197 return (ENXIO);
4197 4198 if (STRMATED(stp))
4198 4199 return (EINVAL);
4199 4200 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4200 4201 return (error);
4201 4202 if (stp->sd_anchor != 0 &&
4202 4203 stp->sd_anchorzone != crgetzoneid(crp))
4203 4204 return (EINVAL);
4204 4205
4205 4206 error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4206 4207 STRUCT_SIZE(strmodremove), copyflag);
4207 4208 if (error)
4208 4209 return (error);
4209 4210
4210 4211 error = (copyflag & U_TO_K ? copyinstr :
4211 4212 copystr)(STRUCT_FGETP(strmodremove, mod_name),
4212 4213 mod_name, FMNAMESZ + 1, NULL);
4213 4214 if (error)
4214 4215 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4215 4216
4216 4217 if ((error = strstartplumb(stp, flag, cmd)) != 0)
4217 4218 return (error);
4218 4219
4219 4220 /*
4220 4221 * Match the name of given module to the name of module at
4221 4222 * the given position.
4222 4223 */
4223 4224 pos = STRUCT_FGET(strmodremove, pos);
4224 4225
4225 4226 is_remove = (pos != 0);
4226 4227 for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4227 4228 q = q->q_next, pos--)
4228 4229 ;
4229 4230 if (pos > 0 || !SAMESTR(q) ||
4230 4231 strcmp(Q2NAME(q), mod_name) != 0) {
4231 4232 mutex_enter(&stp->sd_lock);
4232 4233 strendplumb(stp);
4233 4234 mutex_exit(&stp->sd_lock);
4234 4235 return (EINVAL);
4235 4236 }
4236 4237
4237 4238 /*
4238 4239 * If the position is at or below an anchor, then the zoneid
4239 4240 * must match the zoneid that created the anchor.
4240 4241 */
4241 4242 if (stp->sd_anchor != 0) {
4242 4243 pos = STRUCT_FGET(strmodremove, pos);
4243 4244 if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4244 4245 stp->sd_anchorzone != crgetzoneid(crp)) {
4245 4246 mutex_enter(&stp->sd_lock);
4246 4247 strendplumb(stp);
4247 4248 mutex_exit(&stp->sd_lock);
4248 4249 return (EPERM);
4249 4250 }
4250 4251 }
4251 4252
4252 4253
4253 4254 ASSERT(!(q->q_flag & QREADR));
4254 4255 qdetach(_RD(q), 1, flag, crp, is_remove);
4255 4256
4256 4257 mutex_enter(&stp->sd_lock);
4257 4258
4258 4259 /*
4259 4260 * As a performance concern we are caching the values of
4260 4261 * q_minpsz and q_maxpsz of the module below the stream
4261 4262 * head in the stream head.
4262 4263 */
4263 4264 if (!is_remove) {
4264 4265 mutex_enter(QLOCK(wrq->q_next));
4265 4266 rmin = wrq->q_next->q_minpsz;
4266 4267 rmax = wrq->q_next->q_maxpsz;
4267 4268 mutex_exit(QLOCK(wrq->q_next));
4268 4269
4269 4270 /* Do this processing here as a performance concern */
4270 4271 if (strmsgsz != 0) {
4271 4272 if (rmax == INFPSZ)
4272 4273 rmax = strmsgsz;
4273 4274 else {
4274 4275 if (vp->v_type == VFIFO)
4275 4276 rmax = MIN(PIPE_BUF, rmax);
4276 4277 else rmax = MIN(strmsgsz, rmax);
4277 4278 }
4278 4279 }
4279 4280
4280 4281 mutex_enter(QLOCK(wrq));
4281 4282 stp->sd_qn_minpsz = rmin;
4282 4283 stp->sd_qn_maxpsz = rmax;
4283 4284 mutex_exit(QLOCK(wrq));
4284 4285 }
4285 4286
4286 4287 /*
4287 4288 * Need to update the anchor value if this module is removed
4288 4289 * at or below the anchor point. If the removed module is at
4289 4290 * the anchor point, remove the anchor for this stream if
4290 4291 * there is no module above the anchor point. Otherwise, if
4291 4292 * the removed module is below the anchor point, decrement the
4292 4293 * anchor point by 1.
4293 4294 */
4294 4295 if (stp->sd_anchor != 0) {
4295 4296 pos = STRUCT_FGET(strmodremove, pos);
4296 4297 if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4297 4298 stp->sd_anchor = 0;
4298 4299 else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4299 4300 stp->sd_anchor--;
4300 4301 }
4301 4302
4302 4303 strendplumb(stp);
4303 4304 mutex_exit(&stp->sd_lock);
4304 4305 return (0);
4305 4306 }
4306 4307
4307 4308 case I_ANCHOR:
4308 4309 /*
4309 4310 * Set the anchor position on the stream to reside at
4310 4311 * the top module (in other words, the top module
4311 4312 * cannot be popped). Anchors with a FIFO make no
4312 4313 * obvious sense, so they're not allowed.
4313 4314 */
4314 4315 mutex_enter(&stp->sd_lock);
4315 4316
4316 4317 if (stp->sd_vnode->v_type == VFIFO) {
4317 4318 mutex_exit(&stp->sd_lock);
4318 4319 return (EINVAL);
4319 4320 }
4320 4321 /* Only allow the same zoneid to update the anchor */
4321 4322 if (stp->sd_anchor != 0 &&
4322 4323 stp->sd_anchorzone != crgetzoneid(crp)) {
4323 4324 mutex_exit(&stp->sd_lock);
4324 4325 return (EINVAL);
4325 4326 }
4326 4327 stp->sd_anchor = stp->sd_pushcnt;
4327 4328 stp->sd_anchorzone = crgetzoneid(crp);
4328 4329 mutex_exit(&stp->sd_lock);
4329 4330 return (0);
4330 4331
4331 4332 case I_LOOK:
4332 4333 /*
4333 4334 * Get name of first module downstream.
4334 4335 * If no module, return an error.
4335 4336 */
4336 4337 claimstr(wrq);
4337 4338 if (_SAMESTR(wrq) && wrq->q_next->q_next != NULL) {
4338 4339 char *name = Q2NAME(wrq->q_next);
4339 4340
4340 4341 error = strcopyout(name, (void *)arg, strlen(name) + 1,
4341 4342 copyflag);
4342 4343 releasestr(wrq);
4343 4344 return (error);
4344 4345 }
4345 4346 releasestr(wrq);
4346 4347 return (EINVAL);
4347 4348
4348 4349 case I_LINK:
4349 4350 case I_PLINK:
4350 4351 /*
4351 4352 * Link a multiplexor.
4352 4353 */
4353 4354 return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4354 4355
4355 4356 case _I_PLINK_LH:
4356 4357 /*
4357 4358 * Link a multiplexor: Call must originate from kernel.
4358 4359 */
4359 4360 if (kioctl)
4360 4361 return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4361 4362
4362 4363 return (EINVAL);
4363 4364 case I_UNLINK:
4364 4365 case I_PUNLINK:
4365 4366 /*
4366 4367 * Unlink a multiplexor.
4367 4368 * If arg is -1, unlink all links for which this is the
4368 4369 * controlling stream. Otherwise, arg is an index number
4369 4370 * for a link to be removed.
4370 4371 */
4371 4372 {
4372 4373 struct linkinfo *linkp;
4373 4374 int native_arg = (int)arg;
4374 4375 int type;
4375 4376 netstack_t *ns;
4376 4377 str_stack_t *ss;
4377 4378
4378 4379 TRACE_1(TR_FAC_STREAMS_FR,
4379 4380 TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4380 4381 if (vp->v_type == VFIFO) {
4381 4382 return (EINVAL);
4382 4383 }
4383 4384 if (cmd == I_UNLINK)
4384 4385 type = LINKNORMAL;
4385 4386 else /* I_PUNLINK */
4386 4387 type = LINKPERSIST;
4387 4388 if (native_arg == 0) {
4388 4389 return (EINVAL);
4389 4390 }
4390 4391 ns = netstack_find_by_cred(crp);
4391 4392 ASSERT(ns != NULL);
4392 4393 ss = ns->netstack_str;
4393 4394 ASSERT(ss != NULL);
4394 4395
4395 4396 if (native_arg == MUXID_ALL)
4396 4397 error = munlinkall(stp, type, crp, rvalp, ss);
4397 4398 else {
4398 4399 mutex_enter(&muxifier);
4399 4400 if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4400 4401 /* invalid user supplied index number */
4401 4402 mutex_exit(&muxifier);
4402 4403 netstack_rele(ss->ss_netstack);
4403 4404 return (EINVAL);
4404 4405 }
4405 4406 /* munlink drops the muxifier lock */
4406 4407 error = munlink(stp, linkp, type, crp, rvalp, ss);
4407 4408 }
4408 4409 netstack_rele(ss->ss_netstack);
4409 4410 return (error);
4410 4411 }
4411 4412
4412 4413 case I_FLUSH:
4413 4414 /*
4414 4415 * send a flush message downstream
4415 4416 * flush message can indicate
4416 4417 * FLUSHR - flush read queue
4417 4418 * FLUSHW - flush write queue
4418 4419 * FLUSHRW - flush read/write queue
4419 4420 */
4420 4421 if (stp->sd_flag & STRHUP)
4421 4422 return (ENXIO);
4422 4423 if (arg & ~FLUSHRW)
4423 4424 return (EINVAL);
4424 4425
4425 4426 for (;;) {
4426 4427 if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4427 4428 break;
4428 4429 }
4429 4430 if (error = strwaitbuf(1, BPRI_HI)) {
4430 4431 return (error);
4431 4432 }
4432 4433 }
4433 4434
4434 4435 /*
4435 4436 * Send down an unsupported ioctl and wait for the nack
4436 4437 * in order to allow the M_FLUSH to propagate back
4437 4438 * up to the stream head.
4438 4439 * Replaces if (qready()) runqueues();
4439 4440 */
4440 4441 strioc.ic_cmd = -1; /* The unsupported ioctl */
4441 4442 strioc.ic_timout = 0;
4442 4443 strioc.ic_len = 0;
4443 4444 strioc.ic_dp = NULL;
4444 4445 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4445 4446 *rvalp = 0;
4446 4447 return (0);
4447 4448
4448 4449 case I_FLUSHBAND:
4449 4450 {
4450 4451 struct bandinfo binfo;
4451 4452
4452 4453 error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4453 4454 copyflag);
4454 4455 if (error)
4455 4456 return (error);
4456 4457 if (stp->sd_flag & STRHUP)
4457 4458 return (ENXIO);
4458 4459 if (binfo.bi_flag & ~FLUSHRW)
4459 4460 return (EINVAL);
4460 4461 while (!(mp = allocb(2, BPRI_HI))) {
4461 4462 if (error = strwaitbuf(2, BPRI_HI))
4462 4463 return (error);
4463 4464 }
4464 4465 mp->b_datap->db_type = M_FLUSH;
4465 4466 *mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4466 4467 *mp->b_wptr++ = binfo.bi_pri;
4467 4468 putnext(stp->sd_wrq, mp);
4468 4469 /*
4469 4470 * Send down an unsupported ioctl and wait for the nack
4470 4471 * in order to allow the M_FLUSH to propagate back
4471 4472 * up to the stream head.
4472 4473 * Replaces if (qready()) runqueues();
4473 4474 */
4474 4475 strioc.ic_cmd = -1; /* The unsupported ioctl */
4475 4476 strioc.ic_timout = 0;
4476 4477 strioc.ic_len = 0;
4477 4478 strioc.ic_dp = NULL;
4478 4479 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4479 4480 *rvalp = 0;
4480 4481 return (0);
4481 4482 }
4482 4483
4483 4484 case I_SRDOPT:
4484 4485 /*
4485 4486 * Set read options
4486 4487 *
4487 4488 * RNORM - default stream mode
4488 4489 * RMSGN - message no discard
4489 4490 * RMSGD - message discard
4490 4491 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4491 4492 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4492 4493 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4493 4494 */
4494 4495 if (arg & ~(RMODEMASK | RPROTMASK))
4495 4496 return (EINVAL);
4496 4497
4497 4498 if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4498 4499 return (EINVAL);
4499 4500
4500 4501 mutex_enter(&stp->sd_lock);
4501 4502 switch (arg & RMODEMASK) {
4502 4503 case RNORM:
4503 4504 stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4504 4505 break;
4505 4506 case RMSGD:
4506 4507 stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4507 4508 RD_MSGDIS;
4508 4509 break;
4509 4510 case RMSGN:
4510 4511 stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4511 4512 RD_MSGNODIS;
4512 4513 break;
4513 4514 }
4514 4515
4515 4516 switch (arg & RPROTMASK) {
4516 4517 case RPROTNORM:
4517 4518 stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4518 4519 break;
4519 4520
4520 4521 case RPROTDAT:
4521 4522 stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4522 4523 RD_PROTDAT);
4523 4524 break;
4524 4525
4525 4526 case RPROTDIS:
4526 4527 stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4527 4528 RD_PROTDIS);
4528 4529 break;
4529 4530 }
4530 4531 mutex_exit(&stp->sd_lock);
4531 4532 return (0);
4532 4533
4533 4534 case I_GRDOPT:
4534 4535 /*
4535 4536 * Get read option and return the value
4536 4537 * to spot pointed to by arg
4537 4538 */
4538 4539 {
4539 4540 int rdopt;
4540 4541
4541 4542 rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4542 4543 ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4543 4544 rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4544 4545 ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4545 4546
4546 4547 return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4547 4548 copyflag));
4548 4549 }
4549 4550
4550 4551 case I_SERROPT:
4551 4552 /*
4552 4553 * Set error options
4553 4554 *
4554 4555 * RERRNORM - persistent read errors
4555 4556 * RERRNONPERSIST - non-persistent read errors
4556 4557 * WERRNORM - persistent write errors
4557 4558 * WERRNONPERSIST - non-persistent write errors
4558 4559 */
4559 4560 if (arg & ~(RERRMASK | WERRMASK))
4560 4561 return (EINVAL);
4561 4562
4562 4563 mutex_enter(&stp->sd_lock);
4563 4564 switch (arg & RERRMASK) {
4564 4565 case RERRNORM:
4565 4566 stp->sd_flag &= ~STRDERRNONPERSIST;
4566 4567 break;
4567 4568 case RERRNONPERSIST:
4568 4569 stp->sd_flag |= STRDERRNONPERSIST;
4569 4570 break;
4570 4571 }
4571 4572 switch (arg & WERRMASK) {
4572 4573 case WERRNORM:
4573 4574 stp->sd_flag &= ~STWRERRNONPERSIST;
4574 4575 break;
4575 4576 case WERRNONPERSIST:
4576 4577 stp->sd_flag |= STWRERRNONPERSIST;
4577 4578 break;
4578 4579 }
4579 4580 mutex_exit(&stp->sd_lock);
4580 4581 return (0);
4581 4582
4582 4583 case I_GERROPT:
4583 4584 /*
4584 4585 * Get error option and return the value
4585 4586 * to spot pointed to by arg
4586 4587 */
4587 4588 {
4588 4589 int erropt = 0;
4589 4590
4590 4591 erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4591 4592 RERRNORM;
4592 4593 erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4593 4594 WERRNORM;
4594 4595 return (strcopyout(&erropt, (void *)arg, sizeof (int),
4595 4596 copyflag));
4596 4597 }
4597 4598
4598 4599 case I_SETSIG:
4599 4600 /*
4600 4601 * Register the calling proc to receive the SIGPOLL
4601 4602 * signal based on the events given in arg. If
4602 4603 * arg is zero, remove the proc from register list.
4603 4604 */
4604 4605 {
4605 4606 strsig_t *ssp, *pssp;
4606 4607 struct pid *pidp;
4607 4608
4608 4609 pssp = NULL;
4609 4610 pidp = curproc->p_pidp;
4610 4611 /*
4611 4612 * Hold sd_lock to prevent traversal of sd_siglist while
4612 4613 * it is modified.
4613 4614 */
4614 4615 mutex_enter(&stp->sd_lock);
4615 4616 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4616 4617 pssp = ssp, ssp = ssp->ss_next)
4617 4618 ;
4618 4619
4619 4620 if (arg) {
4620 4621 if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4621 4622 S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4622 4623 mutex_exit(&stp->sd_lock);
4623 4624 return (EINVAL);
4624 4625 }
4625 4626 if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4626 4627 mutex_exit(&stp->sd_lock);
4627 4628 return (EINVAL);
4628 4629 }
4629 4630
4630 4631 /*
4631 4632 * If proc not already registered, add it
4632 4633 * to list.
4633 4634 */
4634 4635 if (!ssp) {
4635 4636 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4636 4637 ssp->ss_pidp = pidp;
4637 4638 ssp->ss_pid = pidp->pid_id;
4638 4639 ssp->ss_next = NULL;
4639 4640 if (pssp)
4640 4641 pssp->ss_next = ssp;
4641 4642 else
4642 4643 stp->sd_siglist = ssp;
4643 4644 mutex_enter(&pidlock);
4644 4645 PID_HOLD(pidp);
4645 4646 mutex_exit(&pidlock);
4646 4647 }
4647 4648
4648 4649 /*
4649 4650 * Set events.
4650 4651 */
4651 4652 ssp->ss_events = (int)arg;
4652 4653 } else {
4653 4654 /*
4654 4655 * Remove proc from register list.
4655 4656 */
4656 4657 if (ssp) {
4657 4658 mutex_enter(&pidlock);
4658 4659 PID_RELE(pidp);
4659 4660 mutex_exit(&pidlock);
4660 4661 if (pssp)
4661 4662 pssp->ss_next = ssp->ss_next;
4662 4663 else
4663 4664 stp->sd_siglist = ssp->ss_next;
4664 4665 kmem_free(ssp, sizeof (strsig_t));
4665 4666 } else {
4666 4667 mutex_exit(&stp->sd_lock);
4667 4668 return (EINVAL);
4668 4669 }
4669 4670 }
4670 4671
4671 4672 /*
4672 4673 * Recalculate OR of sig events.
4673 4674 */
4674 4675 stp->sd_sigflags = 0;
4675 4676 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4676 4677 stp->sd_sigflags |= ssp->ss_events;
4677 4678 mutex_exit(&stp->sd_lock);
4678 4679 return (0);
4679 4680 }
4680 4681
4681 4682 case I_GETSIG:
4682 4683 /*
4683 4684 * Return (in arg) the current registration of events
4684 4685 * for which the calling proc is to be signaled.
4685 4686 */
4686 4687 {
4687 4688 struct strsig *ssp;
4688 4689 struct pid *pidp;
4689 4690
4690 4691 pidp = curproc->p_pidp;
4691 4692 mutex_enter(&stp->sd_lock);
4692 4693 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4693 4694 if (ssp->ss_pidp == pidp) {
4694 4695 error = strcopyout(&ssp->ss_events, (void *)arg,
4695 4696 sizeof (int), copyflag);
4696 4697 mutex_exit(&stp->sd_lock);
4697 4698 return (error);
4698 4699 }
4699 4700 mutex_exit(&stp->sd_lock);
4700 4701 return (EINVAL);
4701 4702 }
4702 4703
4703 4704 case I_ESETSIG:
4704 4705 /*
4705 4706 * Register the ss_pid to receive the SIGPOLL
4706 4707 * signal based on the events is ss_events arg. If
4707 4708 * ss_events is zero, remove the proc from register list.
4708 4709 */
4709 4710 {
4710 4711 struct strsig *ssp, *pssp;
4711 4712 struct proc *proc;
4712 4713 struct pid *pidp;
4713 4714 pid_t pid;
4714 4715 struct strsigset ss;
4715 4716
4716 4717 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4717 4718 if (error)
4718 4719 return (error);
4719 4720
4720 4721 pid = ss.ss_pid;
4721 4722
4722 4723 if (ss.ss_events != 0) {
4723 4724 /*
4724 4725 * Permissions check by sending signal 0.
4725 4726 * Note that when kill fails it does a set_errno
4726 4727 * causing the system call to fail.
4727 4728 */
4728 4729 error = kill(pid, 0);
4729 4730 if (error) {
4730 4731 return (error);
4731 4732 }
4732 4733 }
4733 4734 mutex_enter(&pidlock);
4734 4735 if (pid == 0)
4735 4736 proc = curproc;
4736 4737 else if (pid < 0)
4737 4738 proc = pgfind(-pid);
4738 4739 else
4739 4740 proc = prfind(pid);
4740 4741 if (proc == NULL) {
4741 4742 mutex_exit(&pidlock);
4742 4743 return (ESRCH);
4743 4744 }
4744 4745 if (pid < 0)
4745 4746 pidp = proc->p_pgidp;
4746 4747 else
4747 4748 pidp = proc->p_pidp;
4748 4749 ASSERT(pidp);
4749 4750 /*
4750 4751 * Get a hold on the pid structure while referencing it.
4751 4752 * There is a separate PID_HOLD should it be inserted
4752 4753 * in the list below.
4753 4754 */
4754 4755 PID_HOLD(pidp);
4755 4756 mutex_exit(&pidlock);
4756 4757
4757 4758 pssp = NULL;
4758 4759 /*
4759 4760 * Hold sd_lock to prevent traversal of sd_siglist while
4760 4761 * it is modified.
4761 4762 */
4762 4763 mutex_enter(&stp->sd_lock);
4763 4764 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4764 4765 pssp = ssp, ssp = ssp->ss_next)
4765 4766 ;
4766 4767
4767 4768 if (ss.ss_events) {
4768 4769 if (ss.ss_events &
4769 4770 ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4770 4771 S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4771 4772 mutex_exit(&stp->sd_lock);
4772 4773 mutex_enter(&pidlock);
4773 4774 PID_RELE(pidp);
4774 4775 mutex_exit(&pidlock);
4775 4776 return (EINVAL);
4776 4777 }
4777 4778 if ((ss.ss_events & S_BANDURG) &&
4778 4779 !(ss.ss_events & S_RDBAND)) {
4779 4780 mutex_exit(&stp->sd_lock);
4780 4781 mutex_enter(&pidlock);
4781 4782 PID_RELE(pidp);
4782 4783 mutex_exit(&pidlock);
4783 4784 return (EINVAL);
4784 4785 }
4785 4786
4786 4787 /*
4787 4788 * If proc not already registered, add it
4788 4789 * to list.
4789 4790 */
4790 4791 if (!ssp) {
4791 4792 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4792 4793 ssp->ss_pidp = pidp;
4793 4794 ssp->ss_pid = pid;
4794 4795 ssp->ss_next = NULL;
4795 4796 if (pssp)
4796 4797 pssp->ss_next = ssp;
4797 4798 else
4798 4799 stp->sd_siglist = ssp;
4799 4800 mutex_enter(&pidlock);
4800 4801 PID_HOLD(pidp);
4801 4802 mutex_exit(&pidlock);
4802 4803 }
4803 4804
4804 4805 /*
4805 4806 * Set events.
4806 4807 */
4807 4808 ssp->ss_events = ss.ss_events;
4808 4809 } else {
4809 4810 /*
4810 4811 * Remove proc from register list.
4811 4812 */
4812 4813 if (ssp) {
4813 4814 mutex_enter(&pidlock);
4814 4815 PID_RELE(pidp);
4815 4816 mutex_exit(&pidlock);
4816 4817 if (pssp)
4817 4818 pssp->ss_next = ssp->ss_next;
4818 4819 else
4819 4820 stp->sd_siglist = ssp->ss_next;
4820 4821 kmem_free(ssp, sizeof (strsig_t));
4821 4822 } else {
4822 4823 mutex_exit(&stp->sd_lock);
4823 4824 mutex_enter(&pidlock);
4824 4825 PID_RELE(pidp);
4825 4826 mutex_exit(&pidlock);
4826 4827 return (EINVAL);
4827 4828 }
4828 4829 }
4829 4830
4830 4831 /*
4831 4832 * Recalculate OR of sig events.
4832 4833 */
4833 4834 stp->sd_sigflags = 0;
4834 4835 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4835 4836 stp->sd_sigflags |= ssp->ss_events;
4836 4837 mutex_exit(&stp->sd_lock);
4837 4838 mutex_enter(&pidlock);
4838 4839 PID_RELE(pidp);
4839 4840 mutex_exit(&pidlock);
4840 4841 return (0);
4841 4842 }
4842 4843
4843 4844 case I_EGETSIG:
4844 4845 /*
4845 4846 * Return (in arg) the current registration of events
4846 4847 * for which the calling proc is to be signaled.
4847 4848 */
4848 4849 {
4849 4850 struct strsig *ssp;
4850 4851 struct proc *proc;
4851 4852 pid_t pid;
4852 4853 struct pid *pidp;
4853 4854 struct strsigset ss;
4854 4855
4855 4856 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4856 4857 if (error)
4857 4858 return (error);
4858 4859
4859 4860 pid = ss.ss_pid;
4860 4861 mutex_enter(&pidlock);
4861 4862 if (pid == 0)
4862 4863 proc = curproc;
4863 4864 else if (pid < 0)
4864 4865 proc = pgfind(-pid);
4865 4866 else
4866 4867 proc = prfind(pid);
4867 4868 if (proc == NULL) {
4868 4869 mutex_exit(&pidlock);
4869 4870 return (ESRCH);
4870 4871 }
4871 4872 if (pid < 0)
4872 4873 pidp = proc->p_pgidp;
4873 4874 else
4874 4875 pidp = proc->p_pidp;
4875 4876
4876 4877 /* Prevent the pidp from being reassigned */
4877 4878 PID_HOLD(pidp);
4878 4879 mutex_exit(&pidlock);
4879 4880
4880 4881 mutex_enter(&stp->sd_lock);
4881 4882 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4882 4883 if (ssp->ss_pid == pid) {
4883 4884 ss.ss_pid = ssp->ss_pid;
4884 4885 ss.ss_events = ssp->ss_events;
4885 4886 error = strcopyout(&ss, (void *)arg,
4886 4887 sizeof (struct strsigset), copyflag);
4887 4888 mutex_exit(&stp->sd_lock);
4888 4889 mutex_enter(&pidlock);
4889 4890 PID_RELE(pidp);
4890 4891 mutex_exit(&pidlock);
4891 4892 return (error);
4892 4893 }
4893 4894 mutex_exit(&stp->sd_lock);
4894 4895 mutex_enter(&pidlock);
4895 4896 PID_RELE(pidp);
4896 4897 mutex_exit(&pidlock);
4897 4898 return (EINVAL);
4898 4899 }
4899 4900
4900 4901 case I_PEEK:
4901 4902 {
4902 4903 STRUCT_DECL(strpeek, strpeek);
4903 4904 size_t n;
4904 4905 mblk_t *fmp, *tmp_mp = NULL;
4905 4906
4906 4907 STRUCT_INIT(strpeek, flag);
4907 4908
4908 4909 error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4909 4910 STRUCT_SIZE(strpeek), copyflag);
4910 4911 if (error)
4911 4912 return (error);
4912 4913
4913 4914 mutex_enter(QLOCK(rdq));
4914 4915 /*
4915 4916 * Skip the invalid messages
4916 4917 */
4917 4918 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4918 4919 if (mp->b_datap->db_type != M_SIG)
4919 4920 break;
4920 4921
4921 4922 /*
4922 4923 * If user has requested to peek at a high priority message
4923 4924 * and first message is not, return 0
4924 4925 */
4925 4926 if (mp != NULL) {
4926 4927 if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4927 4928 queclass(mp) == QNORM) {
4928 4929 *rvalp = 0;
4929 4930 mutex_exit(QLOCK(rdq));
4930 4931 return (0);
4931 4932 }
4932 4933 } else if (stp->sd_struiordq == NULL ||
4933 4934 (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4934 4935 /*
4935 4936 * No mblks to look at at the streamhead and
4936 4937 * 1). This isn't a synch stream or
4937 4938 * 2). This is a synch stream but caller wants high
4938 4939 * priority messages which is not supported by
4939 4940 * the synch stream. (it only supports QNORM)
4940 4941 */
4941 4942 *rvalp = 0;
4942 4943 mutex_exit(QLOCK(rdq));
4943 4944 return (0);
4944 4945 }
4945 4946
4946 4947 fmp = mp;
4947 4948
4948 4949 if (mp && mp->b_datap->db_type == M_PASSFP) {
4949 4950 mutex_exit(QLOCK(rdq));
4950 4951 return (EBADMSG);
4951 4952 }
4952 4953
4953 4954 ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4954 4955 mp->b_datap->db_type == M_PROTO ||
4955 4956 mp->b_datap->db_type == M_DATA);
4956 4957
4957 4958 if (mp && mp->b_datap->db_type == M_PCPROTO) {
4958 4959 STRUCT_FSET(strpeek, flags, RS_HIPRI);
4959 4960 } else {
4960 4961 STRUCT_FSET(strpeek, flags, 0);
4961 4962 }
4962 4963
4963 4964
4964 4965 if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
4965 4966 mutex_exit(QLOCK(rdq));
4966 4967 return (ENOSR);
4967 4968 }
4968 4969 mutex_exit(QLOCK(rdq));
4969 4970
4970 4971 /*
4971 4972 * set mp = tmp_mp, so that I_PEEK processing can continue.
4972 4973 * tmp_mp is used to free the dup'd message.
4973 4974 */
4974 4975 mp = tmp_mp;
4975 4976
4976 4977 uio.uio_fmode = 0;
4977 4978 uio.uio_extflg = UIO_COPY_CACHED;
4978 4979 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
4979 4980 UIO_SYSSPACE;
4980 4981 uio.uio_limit = 0;
4981 4982 /*
4982 4983 * First process PROTO blocks, if any.
4983 4984 * If user doesn't want to get ctl info by setting maxlen <= 0,
4984 4985 * then set len to -1/0 and skip control blocks part.
4985 4986 */
4986 4987 if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
4987 4988 STRUCT_FSET(strpeek, ctlbuf.len, -1);
4988 4989 else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
4989 4990 STRUCT_FSET(strpeek, ctlbuf.len, 0);
4990 4991 else {
4991 4992 int ctl_part = 0;
4992 4993
4993 4994 iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
4994 4995 iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
4995 4996 uio.uio_iov = &iov;
4996 4997 uio.uio_resid = iov.iov_len;
4997 4998 uio.uio_loffset = 0;
4998 4999 uio.uio_iovcnt = 1;
4999 5000 while (mp && mp->b_datap->db_type != M_DATA &&
5000 5001 uio.uio_resid >= 0) {
5001 5002 ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5002 5003 mp->b_datap->db_type == M_PROTO :
5003 5004 mp->b_datap->db_type == M_PCPROTO);
5004 5005
5005 5006 if ((n = MIN(uio.uio_resid,
5006 5007 mp->b_wptr - mp->b_rptr)) != 0 &&
5007 5008 (error = uiomove((char *)mp->b_rptr, n,
5008 5009 UIO_READ, &uio)) != 0) {
5009 5010 freemsg(tmp_mp);
5010 5011 return (error);
5011 5012 }
5012 5013 ctl_part = 1;
5013 5014 mp = mp->b_cont;
5014 5015 }
5015 5016 /* No ctl message */
5016 5017 if (ctl_part == 0)
5017 5018 STRUCT_FSET(strpeek, ctlbuf.len, -1);
5018 5019 else
5019 5020 STRUCT_FSET(strpeek, ctlbuf.len,
5020 5021 STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5021 5022 uio.uio_resid);
5022 5023 }
5023 5024
5024 5025 /*
5025 5026 * Now process DATA blocks, if any.
5026 5027 * If user doesn't want to get data info by setting maxlen <= 0,
5027 5028 * then set len to -1/0 and skip data blocks part.
5028 5029 */
5029 5030 if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5030 5031 STRUCT_FSET(strpeek, databuf.len, -1);
5031 5032 else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5032 5033 STRUCT_FSET(strpeek, databuf.len, 0);
5033 5034 else {
5034 5035 int data_part = 0;
5035 5036
5036 5037 iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5037 5038 iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5038 5039 uio.uio_iov = &iov;
5039 5040 uio.uio_resid = iov.iov_len;
5040 5041 uio.uio_loffset = 0;
5041 5042 uio.uio_iovcnt = 1;
5042 5043 while (mp && uio.uio_resid) {
5043 5044 if (mp->b_datap->db_type == M_DATA) {
5044 5045 if ((n = MIN(uio.uio_resid,
5045 5046 mp->b_wptr - mp->b_rptr)) != 0 &&
5046 5047 (error = uiomove((char *)mp->b_rptr,
5047 5048 n, UIO_READ, &uio)) != 0) {
5048 5049 freemsg(tmp_mp);
5049 5050 return (error);
5050 5051 }
5051 5052 data_part = 1;
5052 5053 }
5053 5054 ASSERT(data_part == 0 ||
5054 5055 mp->b_datap->db_type == M_DATA);
5055 5056 mp = mp->b_cont;
5056 5057 }
5057 5058 /* No data message */
5058 5059 if (data_part == 0)
5059 5060 STRUCT_FSET(strpeek, databuf.len, -1);
5060 5061 else
5061 5062 STRUCT_FSET(strpeek, databuf.len,
5062 5063 STRUCT_FGET(strpeek, databuf.maxlen) -
5063 5064 uio.uio_resid);
5064 5065 }
5065 5066 freemsg(tmp_mp);
5066 5067
5067 5068 /*
5068 5069 * It is a synch stream and user wants to get
5069 5070 * data (maxlen > 0).
5070 5071 * uio setup is done by the codes that process DATA
5071 5072 * blocks above.
5072 5073 */
5073 5074 if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5074 5075 infod_t infod;
5075 5076
5076 5077 infod.d_cmd = INFOD_COPYOUT;
5077 5078 infod.d_res = 0;
5078 5079 infod.d_uiop = &uio;
5079 5080 error = infonext(rdq, &infod);
5080 5081 if (error == EINVAL || error == EBUSY)
5081 5082 error = 0;
5082 5083 if (error)
5083 5084 return (error);
5084 5085 STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5085 5086 databuf.maxlen) - uio.uio_resid);
5086 5087 if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5087 5088 /*
5088 5089 * No data found by the infonext().
5089 5090 */
5090 5091 STRUCT_FSET(strpeek, databuf.len, -1);
5091 5092 }
5092 5093 }
5093 5094 error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5094 5095 STRUCT_SIZE(strpeek), copyflag);
5095 5096 if (error) {
5096 5097 return (error);
5097 5098 }
5098 5099 /*
5099 5100 * If there is no message retrieved, set return code to 0
5100 5101 * otherwise, set it to 1.
5101 5102 */
5102 5103 if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5103 5104 STRUCT_FGET(strpeek, databuf.len) == -1)
5104 5105 *rvalp = 0;
5105 5106 else
5106 5107 *rvalp = 1;
5107 5108 return (0);
5108 5109 }
5109 5110
5110 5111 case I_FDINSERT:
5111 5112 {
5112 5113 STRUCT_DECL(strfdinsert, strfdinsert);
5113 5114 struct file *resftp;
5114 5115 struct stdata *resstp;
5115 5116 t_uscalar_t ival;
5116 5117 ssize_t msgsize;
5117 5118 struct strbuf mctl;
5118 5119
5119 5120 STRUCT_INIT(strfdinsert, flag);
5120 5121 if (stp->sd_flag & STRHUP)
5121 5122 return (ENXIO);
5122 5123 /*
5123 5124 * STRDERR, STWRERR and STPLEX tested above.
5124 5125 */
5125 5126 error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5126 5127 STRUCT_SIZE(strfdinsert), copyflag);
5127 5128 if (error)
5128 5129 return (error);
5129 5130
5130 5131 if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5131 5132 (STRUCT_FGET(strfdinsert, offset) %
5132 5133 sizeof (t_uscalar_t)) != 0)
5133 5134 return (EINVAL);
5134 5135 if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5135 5136 if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5136 5137 releasef(STRUCT_FGET(strfdinsert, fildes));
5137 5138 return (EINVAL);
5138 5139 }
5139 5140 } else
5140 5141 return (EINVAL);
5141 5142
5142 5143 mutex_enter(&resstp->sd_lock);
5143 5144 if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5144 5145 error = strgeterr(resstp,
5145 5146 STRDERR|STWRERR|STRHUP|STPLEX, 0);
5146 5147 if (error != 0) {
5147 5148 mutex_exit(&resstp->sd_lock);
5148 5149 releasef(STRUCT_FGET(strfdinsert, fildes));
5149 5150 return (error);
5150 5151 }
5151 5152 }
5152 5153 mutex_exit(&resstp->sd_lock);
5153 5154
5154 5155 #ifdef _ILP32
5155 5156 {
5156 5157 queue_t *q;
5157 5158 queue_t *mate = NULL;
5158 5159
5159 5160 /* get read queue of stream terminus */
5160 5161 claimstr(resstp->sd_wrq);
5161 5162 for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5162 5163 q = q->q_next)
5163 5164 if (!STRMATED(resstp) && STREAM(q) != resstp &&
5164 5165 mate == NULL) {
5165 5166 ASSERT(q->q_qinfo->qi_srvp);
5166 5167 ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5167 5168 claimstr(q);
5168 5169 mate = q;
5169 5170 }
5170 5171 q = _RD(q);
5171 5172 if (mate)
5172 5173 releasestr(mate);
5173 5174 releasestr(resstp->sd_wrq);
5174 5175 ival = (t_uscalar_t)q;
5175 5176 }
5176 5177 #else
5177 5178 ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5178 5179 #endif /* _ILP32 */
5179 5180
5180 5181 if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5181 5182 STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5182 5183 releasef(STRUCT_FGET(strfdinsert, fildes));
5183 5184 return (EINVAL);
5184 5185 }
5185 5186
5186 5187 /*
5187 5188 * Check for legal flag value.
5188 5189 */
5189 5190 if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5190 5191 releasef(STRUCT_FGET(strfdinsert, fildes));
5191 5192 return (EINVAL);
5192 5193 }
5193 5194
5194 5195 /* get these values from those cached in the stream head */
5195 5196 mutex_enter(QLOCK(stp->sd_wrq));
5196 5197 rmin = stp->sd_qn_minpsz;
5197 5198 rmax = stp->sd_qn_maxpsz;
5198 5199 mutex_exit(QLOCK(stp->sd_wrq));
5199 5200
5200 5201 /*
5201 5202 * Make sure ctl and data sizes together fall within
5202 5203 * the limits of the max and min receive packet sizes
5203 5204 * and do not exceed system limit. A negative data
5204 5205 * length means that no data part is to be sent.
5205 5206 */
5206 5207 ASSERT((rmax >= 0) || (rmax == INFPSZ));
5207 5208 if (rmax == 0) {
5208 5209 releasef(STRUCT_FGET(strfdinsert, fildes));
5209 5210 return (ERANGE);
5210 5211 }
5211 5212 if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5212 5213 msgsize = 0;
5213 5214 if ((msgsize < rmin) ||
5214 5215 ((msgsize > rmax) && (rmax != INFPSZ)) ||
5215 5216 (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5216 5217 releasef(STRUCT_FGET(strfdinsert, fildes));
5217 5218 return (ERANGE);
5218 5219 }
5219 5220
5220 5221 mutex_enter(&stp->sd_lock);
5221 5222 while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5222 5223 !canputnext(stp->sd_wrq)) {
5223 5224 if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5224 5225 flag, -1, &done)) != 0 || done) {
5225 5226 mutex_exit(&stp->sd_lock);
5226 5227 releasef(STRUCT_FGET(strfdinsert, fildes));
5227 5228 return (error);
5228 5229 }
5229 5230 if ((error = i_straccess(stp, access)) != 0) {
5230 5231 mutex_exit(&stp->sd_lock);
5231 5232 releasef(
5232 5233 STRUCT_FGET(strfdinsert, fildes));
5233 5234 return (error);
5234 5235 }
5235 5236 }
5236 5237 mutex_exit(&stp->sd_lock);
5237 5238
5238 5239 /*
5239 5240 * Copy strfdinsert.ctlbuf into native form of
5240 5241 * ctlbuf to pass down into strmakemsg().
5241 5242 */
5242 5243 mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5243 5244 mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5244 5245 mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5245 5246
5246 5247 iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5247 5248 iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5248 5249 uio.uio_iov = &iov;
5249 5250 uio.uio_iovcnt = 1;
5250 5251 uio.uio_loffset = 0;
5251 5252 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5252 5253 UIO_SYSSPACE;
5253 5254 uio.uio_fmode = 0;
5254 5255 uio.uio_extflg = UIO_COPY_CACHED;
5255 5256 uio.uio_resid = iov.iov_len;
5256 5257 if ((error = strmakemsg(&mctl,
5257 5258 &msgsize, &uio, stp,
5258 5259 STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5259 5260 STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5260 5261 releasef(STRUCT_FGET(strfdinsert, fildes));
5261 5262 return (error);
5262 5263 }
5263 5264
5264 5265 STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5265 5266
5266 5267 /*
5267 5268 * Place the possibly reencoded queue pointer 'offset' bytes
5268 5269 * from the start of the control portion of the message.
5269 5270 */
5270 5271 *((t_uscalar_t *)(mp->b_rptr +
5271 5272 STRUCT_FGET(strfdinsert, offset))) = ival;
5272 5273
5273 5274 /*
5274 5275 * Put message downstream.
5275 5276 */
5276 5277 stream_willservice(stp);
5277 5278 putnext(stp->sd_wrq, mp);
5278 5279 stream_runservice(stp);
5279 5280 releasef(STRUCT_FGET(strfdinsert, fildes));
5280 5281 return (error);
5281 5282 }
5282 5283
5283 5284 case I_SENDFD:
5284 5285 {
5285 5286 struct file *fp;
5286 5287
5287 5288 if ((fp = getf((int)arg)) == NULL)
5288 5289 return (EBADF);
5289 5290 error = do_sendfp(stp, fp, crp);
5290 5291 if (auditing) {
5291 5292 audit_fdsend((int)arg, fp, error);
5292 5293 }
5293 5294 releasef((int)arg);
5294 5295 return (error);
5295 5296 }
5296 5297
5297 5298 case I_RECVFD:
5298 5299 case I_E_RECVFD:
5299 5300 {
5300 5301 struct k_strrecvfd *srf;
5301 5302 int i, fd;
5302 5303
5303 5304 mutex_enter(&stp->sd_lock);
5304 5305 while (!(mp = getq(rdq))) {
5305 5306 if (stp->sd_flag & (STRHUP|STREOF)) {
5306 5307 mutex_exit(&stp->sd_lock);
5307 5308 return (ENXIO);
5308 5309 }
5309 5310 if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5310 5311 flag, -1, &done)) != 0 || done) {
5311 5312 mutex_exit(&stp->sd_lock);
5312 5313 return (error);
5313 5314 }
5314 5315 if ((error = i_straccess(stp, access)) != 0) {
5315 5316 mutex_exit(&stp->sd_lock);
5316 5317 return (error);
5317 5318 }
5318 5319 }
5319 5320 if (mp->b_datap->db_type != M_PASSFP) {
5320 5321 putback(stp, rdq, mp, mp->b_band);
5321 5322 mutex_exit(&stp->sd_lock);
5322 5323 return (EBADMSG);
5323 5324 }
5324 5325 mutex_exit(&stp->sd_lock);
5325 5326
5326 5327 srf = (struct k_strrecvfd *)mp->b_rptr;
5327 5328 if ((fd = ufalloc(0)) == -1) {
5328 5329 mutex_enter(&stp->sd_lock);
5329 5330 putback(stp, rdq, mp, mp->b_band);
5330 5331 mutex_exit(&stp->sd_lock);
5331 5332 return (EMFILE);
5332 5333 }
5333 5334 if (cmd == I_RECVFD) {
5334 5335 struct o_strrecvfd ostrfd;
5335 5336
5336 5337 /* check to see if uid/gid values are too large. */
5337 5338
5338 5339 if (srf->uid > (o_uid_t)USHRT_MAX ||
5339 5340 srf->gid > (o_gid_t)USHRT_MAX) {
5340 5341 mutex_enter(&stp->sd_lock);
5341 5342 putback(stp, rdq, mp, mp->b_band);
5342 5343 mutex_exit(&stp->sd_lock);
5343 5344 setf(fd, NULL); /* release fd entry */
5344 5345 return (EOVERFLOW);
5345 5346 }
5346 5347
5347 5348 ostrfd.fd = fd;
5348 5349 ostrfd.uid = (o_uid_t)srf->uid;
5349 5350 ostrfd.gid = (o_gid_t)srf->gid;
5350 5351
5351 5352 /* Null the filler bits */
5352 5353 for (i = 0; i < 8; i++)
5353 5354 ostrfd.fill[i] = 0;
5354 5355
5355 5356 error = strcopyout(&ostrfd, (void *)arg,
5356 5357 sizeof (struct o_strrecvfd), copyflag);
5357 5358 } else { /* I_E_RECVFD */
5358 5359 struct strrecvfd strfd;
5359 5360
5360 5361 strfd.fd = fd;
5361 5362 strfd.uid = srf->uid;
5362 5363 strfd.gid = srf->gid;
5363 5364
5364 5365 /* null the filler bits */
5365 5366 for (i = 0; i < 8; i++)
5366 5367 strfd.fill[i] = 0;
5367 5368
5368 5369 error = strcopyout(&strfd, (void *)arg,
5369 5370 sizeof (struct strrecvfd), copyflag);
5370 5371 }
5371 5372
5372 5373 if (error) {
5373 5374 setf(fd, NULL); /* release fd entry */
5374 5375 mutex_enter(&stp->sd_lock);
5375 5376 putback(stp, rdq, mp, mp->b_band);
5376 5377 mutex_exit(&stp->sd_lock);
5377 5378 return (error);
5378 5379 }
5379 5380 if (auditing) {
5380 5381 audit_fdrecv(fd, srf->fp);
5381 5382 }
5382 5383
5383 5384 /*
5384 5385 * Always increment f_count since the freemsg() below will
5385 5386 * always call free_passfp() which performs a closef().
5386 5387 */
5387 5388 mutex_enter(&srf->fp->f_tlock);
5388 5389 srf->fp->f_count++;
5389 5390 mutex_exit(&srf->fp->f_tlock);
5390 5391 setf(fd, srf->fp);
5391 5392 freemsg(mp);
5392 5393 return (0);
5393 5394 }
5394 5395
5395 5396 case I_SWROPT:
5396 5397 /*
5397 5398 * Set/clear the write options. arg is a bit
5398 5399 * mask with any of the following bits set...
5399 5400 * SNDZERO - send zero length message
5400 5401 * SNDPIPE - send sigpipe to process if
5401 5402 * sd_werror is set and process is
5402 5403 * doing a write or putmsg.
5403 5404 * The new stream head write options should reflect
5404 5405 * what is in arg.
5405 5406 */
5406 5407 if (arg & ~(SNDZERO|SNDPIPE))
5407 5408 return (EINVAL);
5408 5409
5409 5410 mutex_enter(&stp->sd_lock);
5410 5411 stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5411 5412 if (arg & SNDZERO)
5412 5413 stp->sd_wput_opt |= SW_SNDZERO;
5413 5414 if (arg & SNDPIPE)
5414 5415 stp->sd_wput_opt |= SW_SIGPIPE;
5415 5416 mutex_exit(&stp->sd_lock);
5416 5417 return (0);
5417 5418
5418 5419 case I_GWROPT:
5419 5420 {
5420 5421 int wropt = 0;
5421 5422
5422 5423 if (stp->sd_wput_opt & SW_SNDZERO)
5423 5424 wropt |= SNDZERO;
5424 5425 if (stp->sd_wput_opt & SW_SIGPIPE)
5425 5426 wropt |= SNDPIPE;
5426 5427 return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5427 5428 copyflag));
5428 5429 }
5429 5430
5430 5431 case I_LIST:
5431 5432 /*
5432 5433 * Returns all the modules found on this stream,
5433 5434 * upto the driver. If argument is NULL, return the
5434 5435 * number of modules (including driver). If argument
5435 5436 * is not NULL, copy the names into the structure
5436 5437 * provided.
5437 5438 */
5438 5439
5439 5440 {
5440 5441 queue_t *q;
5441 5442 char *qname;
5442 5443 int i, nmods;
5443 5444 struct str_mlist *mlist;
5444 5445 STRUCT_DECL(str_list, strlist);
5445 5446
5446 5447 if (arg == NULL) { /* Return number of modules plus driver */
5447 5448 if (stp->sd_vnode->v_type == VFIFO)
5448 5449 *rvalp = stp->sd_pushcnt;
5449 5450 else
5450 5451 *rvalp = stp->sd_pushcnt + 1;
5451 5452 return (0);
5452 5453 }
5453 5454
5454 5455 STRUCT_INIT(strlist, flag);
5455 5456
5456 5457 error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5457 5458 STRUCT_SIZE(strlist), copyflag);
5458 5459 if (error != 0)
5459 5460 return (error);
5460 5461
5461 5462 mlist = STRUCT_FGETP(strlist, sl_modlist);
5462 5463 nmods = STRUCT_FGET(strlist, sl_nmods);
5463 5464 if (nmods <= 0)
5464 5465 return (EINVAL);
5465 5466
5466 5467 claimstr(stp->sd_wrq);
5467 5468 q = stp->sd_wrq;
5468 5469 for (i = 0; i < nmods && _SAMESTR(q); i++, q = q->q_next) {
5469 5470 qname = Q2NAME(q->q_next);
5470 5471 error = strcopyout(qname, &mlist[i], strlen(qname) + 1,
5471 5472 copyflag);
5472 5473 if (error != 0) {
5473 5474 releasestr(stp->sd_wrq);
5474 5475 return (error);
5475 5476 }
5476 5477 }
5477 5478 releasestr(stp->sd_wrq);
5478 5479 return (strcopyout(&i, (void *)arg, sizeof (int), copyflag));
5479 5480 }
5480 5481
5481 5482 case I_CKBAND:
5482 5483 {
5483 5484 queue_t *q;
5484 5485 qband_t *qbp;
5485 5486
5486 5487 if ((arg < 0) || (arg >= NBAND))
5487 5488 return (EINVAL);
5488 5489 q = _RD(stp->sd_wrq);
5489 5490 mutex_enter(QLOCK(q));
5490 5491 if (arg > (int)q->q_nband) {
5491 5492 *rvalp = 0;
5492 5493 } else {
5493 5494 if (arg == 0) {
5494 5495 if (q->q_first)
5495 5496 *rvalp = 1;
5496 5497 else
5497 5498 *rvalp = 0;
5498 5499 } else {
5499 5500 qbp = q->q_bandp;
5500 5501 while (--arg > 0)
5501 5502 qbp = qbp->qb_next;
5502 5503 if (qbp->qb_first)
5503 5504 *rvalp = 1;
5504 5505 else
5505 5506 *rvalp = 0;
5506 5507 }
5507 5508 }
5508 5509 mutex_exit(QLOCK(q));
5509 5510 return (0);
5510 5511 }
5511 5512
5512 5513 case I_GETBAND:
5513 5514 {
5514 5515 int intpri;
5515 5516 queue_t *q;
5516 5517
5517 5518 q = _RD(stp->sd_wrq);
5518 5519 mutex_enter(QLOCK(q));
5519 5520 mp = q->q_first;
5520 5521 if (!mp) {
5521 5522 mutex_exit(QLOCK(q));
5522 5523 return (ENODATA);
5523 5524 }
5524 5525 intpri = (int)mp->b_band;
5525 5526 error = strcopyout(&intpri, (void *)arg, sizeof (int),
5526 5527 copyflag);
5527 5528 mutex_exit(QLOCK(q));
5528 5529 return (error);
5529 5530 }
5530 5531
5531 5532 case I_ATMARK:
5532 5533 {
5533 5534 queue_t *q;
5534 5535
5535 5536 if (arg & ~(ANYMARK|LASTMARK))
5536 5537 return (EINVAL);
5537 5538 q = _RD(stp->sd_wrq);
5538 5539 mutex_enter(&stp->sd_lock);
5539 5540 if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5540 5541 *rvalp = 1;
5541 5542 } else {
5542 5543 mutex_enter(QLOCK(q));
5543 5544 mp = q->q_first;
5544 5545
5545 5546 if (mp == NULL)
5546 5547 *rvalp = 0;
5547 5548 else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5548 5549 *rvalp = 1;
5549 5550 else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5550 5551 *rvalp = 1;
5551 5552 else
5552 5553 *rvalp = 0;
5553 5554 mutex_exit(QLOCK(q));
5554 5555 }
5555 5556 mutex_exit(&stp->sd_lock);
5556 5557 return (0);
5557 5558 }
5558 5559
5559 5560 case I_CANPUT:
5560 5561 {
5561 5562 char band;
5562 5563
5563 5564 if ((arg < 0) || (arg >= NBAND))
5564 5565 return (EINVAL);
5565 5566 band = (char)arg;
5566 5567 *rvalp = bcanputnext(stp->sd_wrq, band);
5567 5568 return (0);
5568 5569 }
5569 5570
5570 5571 case I_SETCLTIME:
5571 5572 {
5572 5573 int closetime;
5573 5574
5574 5575 error = strcopyin((void *)arg, &closetime, sizeof (int),
5575 5576 copyflag);
5576 5577 if (error)
5577 5578 return (error);
5578 5579 if (closetime < 0)
5579 5580 return (EINVAL);
5580 5581
5581 5582 stp->sd_closetime = closetime;
5582 5583 return (0);
5583 5584 }
5584 5585
5585 5586 case I_GETCLTIME:
5586 5587 {
5587 5588 int closetime;
5588 5589
5589 5590 closetime = stp->sd_closetime;
5590 5591 return (strcopyout(&closetime, (void *)arg, sizeof (int),
5591 5592 copyflag));
5592 5593 }
5593 5594
5594 5595 case TIOCGSID:
5595 5596 {
5596 5597 pid_t sid;
5597 5598
5598 5599 mutex_enter(&stp->sd_lock);
5599 5600 if (stp->sd_sidp == NULL) {
5600 5601 mutex_exit(&stp->sd_lock);
5601 5602 return (ENOTTY);
5602 5603 }
5603 5604 sid = stp->sd_sidp->pid_id;
5604 5605 mutex_exit(&stp->sd_lock);
5605 5606 return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5606 5607 copyflag));
5607 5608 }
5608 5609
5609 5610 case TIOCSPGRP:
5610 5611 {
5611 5612 pid_t pgrp;
5612 5613 proc_t *q;
5613 5614 pid_t sid, fg_pgid, bg_pgid;
5614 5615
5615 5616 if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5616 5617 copyflag))
5617 5618 return (error);
5618 5619 mutex_enter(&stp->sd_lock);
5619 5620 mutex_enter(&pidlock);
5620 5621 if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5621 5622 mutex_exit(&pidlock);
5622 5623 mutex_exit(&stp->sd_lock);
5623 5624 return (ENOTTY);
5624 5625 }
5625 5626 if (pgrp == stp->sd_pgidp->pid_id) {
5626 5627 mutex_exit(&pidlock);
5627 5628 mutex_exit(&stp->sd_lock);
5628 5629 return (0);
5629 5630 }
5630 5631 if (pgrp <= 0 || pgrp >= maxpid) {
5631 5632 mutex_exit(&pidlock);
5632 5633 mutex_exit(&stp->sd_lock);
5633 5634 return (EINVAL);
5634 5635 }
5635 5636 if ((q = pgfind(pgrp)) == NULL ||
5636 5637 q->p_sessp != ttoproc(curthread)->p_sessp) {
5637 5638 mutex_exit(&pidlock);
5638 5639 mutex_exit(&stp->sd_lock);
5639 5640 return (EPERM);
5640 5641 }
5641 5642 sid = stp->sd_sidp->pid_id;
5642 5643 fg_pgid = q->p_pgrp;
5643 5644 bg_pgid = stp->sd_pgidp->pid_id;
5644 5645 CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5645 5646 PID_RELE(stp->sd_pgidp);
5646 5647 ctty_clear_sighuped();
5647 5648 stp->sd_pgidp = q->p_pgidp;
5648 5649 PID_HOLD(stp->sd_pgidp);
5649 5650 mutex_exit(&pidlock);
5650 5651 mutex_exit(&stp->sd_lock);
5651 5652 return (0);
5652 5653 }
5653 5654
5654 5655 case TIOCGPGRP:
5655 5656 {
5656 5657 pid_t pgrp;
5657 5658
5658 5659 mutex_enter(&stp->sd_lock);
5659 5660 if (stp->sd_sidp == NULL) {
5660 5661 mutex_exit(&stp->sd_lock);
5661 5662 return (ENOTTY);
5662 5663 }
5663 5664 pgrp = stp->sd_pgidp->pid_id;
5664 5665 mutex_exit(&stp->sd_lock);
5665 5666 return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5666 5667 copyflag));
5667 5668 }
5668 5669
5669 5670 case TIOCSCTTY:
5670 5671 {
5671 5672 return (strctty(stp));
5672 5673 }
5673 5674
5674 5675 case TIOCNOTTY:
5675 5676 {
5676 5677 /* freectty() always assumes curproc. */
5677 5678 if (freectty(B_FALSE) != 0)
5678 5679 return (0);
5679 5680 return (ENOTTY);
5680 5681 }
5681 5682
5682 5683 case FIONBIO:
5683 5684 case FIOASYNC:
5684 5685 return (0); /* handled by the upper layer */
5685 5686 }
5686 5687 }
5687 5688
5688 5689 /*
5689 5690 * Custom free routine used for M_PASSFP messages.
5690 5691 */
5691 5692 static void
5692 5693 free_passfp(struct k_strrecvfd *srf)
5693 5694 {
5694 5695 (void) closef(srf->fp);
5695 5696 kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5696 5697 }
5697 5698
5698 5699 /* ARGSUSED */
5699 5700 int
5700 5701 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5701 5702 {
5702 5703 queue_t *qp, *nextqp;
5703 5704 struct k_strrecvfd *srf;
5704 5705 mblk_t *mp;
5705 5706 frtn_t *frtnp;
5706 5707 size_t bufsize;
5707 5708 queue_t *mate = NULL;
5708 5709 syncq_t *sq = NULL;
5709 5710 int retval = 0;
5710 5711
5711 5712 if (stp->sd_flag & STRHUP)
5712 5713 return (ENXIO);
5713 5714
5714 5715 claimstr(stp->sd_wrq);
5715 5716
5716 5717 /* Fastpath, we have a pipe, and we are already mated, use it. */
5717 5718 if (STRMATED(stp)) {
5718 5719 qp = _RD(stp->sd_mate->sd_wrq);
5719 5720 claimstr(qp);
5720 5721 mate = qp;
5721 5722 } else { /* Not already mated. */
5722 5723
5723 5724 /*
5724 5725 * Walk the stream to the end of this one.
5725 5726 * assumes that the claimstr() will prevent
5726 5727 * plumbing between the stream head and the
5727 5728 * driver from changing
5728 5729 */
5729 5730 qp = stp->sd_wrq;
5730 5731
5731 5732 /*
5732 5733 * Loop until we reach the end of this stream.
5733 5734 * On completion, qp points to the write queue
5734 5735 * at the end of the stream, or the read queue
5735 5736 * at the stream head if this is a fifo.
5736 5737 */
5737 5738 while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5738 5739 ;
5739 5740
5740 5741 /*
5741 5742 * Just in case we get a q_next which is NULL, but
5742 5743 * not at the end of the stream. This is actually
5743 5744 * broken, so we set an assert to catch it in
5744 5745 * debug, and set an error and return if not debug.
5745 5746 */
5746 5747 ASSERT(qp);
5747 5748 if (qp == NULL) {
5748 5749 releasestr(stp->sd_wrq);
5749 5750 return (EINVAL);
5750 5751 }
5751 5752
5752 5753 /*
5753 5754 * Enter the syncq for the driver, so (hopefully)
5754 5755 * the queue values will not change on us.
5755 5756 * XXXX - This will only prevent the race IFF only
5756 5757 * the write side modifies the q_next member, and
5757 5758 * the put procedure is protected by at least
5758 5759 * MT_PERQ.
5759 5760 */
5760 5761 if ((sq = qp->q_syncq) != NULL)
5761 5762 entersq(sq, SQ_PUT);
5762 5763
5763 5764 /* Now get the q_next value from this qp. */
5764 5765 nextqp = qp->q_next;
5765 5766
5766 5767 /*
5767 5768 * If nextqp exists and the other stream is different
5768 5769 * from this one claim the stream, set the mate, and
5769 5770 * get the read queue at the stream head of the other
5770 5771 * stream. Assumes that nextqp was at least valid when
5771 5772 * we got it. Hopefully the entersq of the driver
5772 5773 * will prevent it from changing on us.
5773 5774 */
5774 5775 if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5775 5776 ASSERT(qp->q_qinfo->qi_srvp);
5776 5777 ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5777 5778 ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5778 5779 claimstr(nextqp);
5779 5780
5780 5781 /* Make sure we still have a q_next */
5781 5782 if (nextqp != qp->q_next) {
5782 5783 releasestr(stp->sd_wrq);
5783 5784 releasestr(nextqp);
5784 5785 return (EINVAL);
5785 5786 }
5786 5787
5787 5788 qp = _RD(STREAM(nextqp)->sd_wrq);
5788 5789 mate = qp;
5789 5790 }
5790 5791 /* If we entered the synq above, leave it. */
5791 5792 if (sq != NULL)
5792 5793 leavesq(sq, SQ_PUT);
5793 5794 } /* STRMATED(STP) */
5794 5795
5795 5796 /* XXX prevents substitution of the ops vector */
5796 5797 if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5797 5798 retval = EINVAL;
5798 5799 goto out;
5799 5800 }
5800 5801
5801 5802 if (qp->q_flag & QFULL) {
5802 5803 retval = EAGAIN;
5803 5804 goto out;
5804 5805 }
5805 5806
5806 5807 /*
5807 5808 * Since M_PASSFP messages include a file descriptor, we use
5808 5809 * esballoc() and specify a custom free routine (free_passfp()) that
5809 5810 * will close the descriptor as part of freeing the message. For
5810 5811 * convenience, we stash the frtn_t right after the data block.
5811 5812 */
5812 5813 bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5813 5814 srf = kmem_alloc(bufsize, KM_NOSLEEP);
5814 5815 if (srf == NULL) {
5815 5816 retval = EAGAIN;
5816 5817 goto out;
5817 5818 }
5818 5819
5819 5820 frtnp = (frtn_t *)(srf + 1);
5820 5821 frtnp->free_arg = (caddr_t)srf;
5821 5822 frtnp->free_func = free_passfp;
5822 5823
5823 5824 mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5824 5825 if (mp == NULL) {
5825 5826 kmem_free(srf, bufsize);
5826 5827 retval = EAGAIN;
5827 5828 goto out;
5828 5829 }
5829 5830 mp->b_wptr += sizeof (struct k_strrecvfd);
5830 5831 mp->b_datap->db_type = M_PASSFP;
5831 5832
5832 5833 srf->fp = fp;
5833 5834 srf->uid = crgetuid(curthread->t_cred);
5834 5835 srf->gid = crgetgid(curthread->t_cred);
5835 5836 mutex_enter(&fp->f_tlock);
5836 5837 fp->f_count++;
5837 5838 mutex_exit(&fp->f_tlock);
5838 5839
5839 5840 put(qp, mp);
5840 5841 out:
5841 5842 releasestr(stp->sd_wrq);
5842 5843 if (mate)
5843 5844 releasestr(mate);
5844 5845 return (retval);
5845 5846 }
5846 5847
5847 5848 /*
5848 5849 * Send an ioctl message downstream and wait for acknowledgement.
5849 5850 * flags may be set to either U_TO_K or K_TO_K and a combination
5850 5851 * of STR_NOERROR or STR_NOSIG
5851 5852 * STR_NOSIG: Signals are essentially ignored or held and have
5852 5853 * no effect for the duration of the call.
5853 5854 * STR_NOERROR: Ignores stream head read, write and hup errors.
5854 5855 * Additionally, if an existing ioctl times out, it is assumed
5855 5856 * lost and and this ioctl will continue as if the previous ioctl had
5856 5857 * finished. ETIME may be returned if this ioctl times out (i.e.
5857 5858 * ic_timout is not INFTIM). Non-stream head errors may be returned if
5858 5859 * the ioc_error indicates that the driver/module had problems,
5859 5860 * an EFAULT was found when accessing user data, a lack of
5860 5861 * resources, etc.
5861 5862 */
5862 5863 int
5863 5864 strdoioctl(
5864 5865 struct stdata *stp,
5865 5866 struct strioctl *strioc,
5866 5867 int fflags, /* file flags with model info */
5867 5868 int flag,
5868 5869 cred_t *crp,
5869 5870 int *rvalp)
5870 5871 {
5871 5872 mblk_t *bp;
5872 5873 struct iocblk *iocbp;
5873 5874 struct copyreq *reqp;
5874 5875 struct copyresp *resp;
5875 5876 int id;
5876 5877 int transparent = 0;
5877 5878 int error = 0;
5878 5879 int len = 0;
5879 5880 caddr_t taddr;
5880 5881 int copyflag = (flag & (U_TO_K | K_TO_K));
5881 5882 int sigflag = (flag & STR_NOSIG);
5882 5883 int errs;
5883 5884 uint_t waitflags;
5884 5885 boolean_t set_iocwaitne = B_FALSE;
5885 5886
5886 5887 ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5887 5888 ASSERT((fflags & FMODELS) != 0);
5888 5889
5889 5890 TRACE_2(TR_FAC_STREAMS_FR,
5890 5891 TR_STRDOIOCTL,
5891 5892 "strdoioctl:stp %p strioc %p", stp, strioc);
5892 5893 if (strioc->ic_len == TRANSPARENT) { /* send arg in M_DATA block */
5893 5894 transparent = 1;
5894 5895 strioc->ic_len = sizeof (intptr_t);
5895 5896 }
5896 5897
5897 5898 if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5898 5899 return (EINVAL);
5899 5900
5900 5901 if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5901 5902 crp, curproc->p_pid)) == NULL)
5902 5903 return (error);
5903 5904
5904 5905 bzero(bp->b_wptr, sizeof (union ioctypes));
5905 5906
5906 5907 iocbp = (struct iocblk *)bp->b_wptr;
5907 5908 iocbp->ioc_count = strioc->ic_len;
5908 5909 iocbp->ioc_cmd = strioc->ic_cmd;
5909 5910 iocbp->ioc_flag = (fflags & FMODELS);
5910 5911
5911 5912 crhold(crp);
5912 5913 iocbp->ioc_cr = crp;
5913 5914 DB_TYPE(bp) = M_IOCTL;
5914 5915 bp->b_wptr += sizeof (struct iocblk);
5915 5916
5916 5917 if (flag & STR_NOERROR)
5917 5918 errs = STPLEX;
5918 5919 else
5919 5920 errs = STRHUP|STRDERR|STWRERR|STPLEX;
5920 5921
5921 5922 /*
5922 5923 * If there is data to copy into ioctl block, do so.
5923 5924 */
5924 5925 if (iocbp->ioc_count > 0) {
5925 5926 if (transparent)
5926 5927 /*
5927 5928 * Note: STR_NOERROR does not have an effect
5928 5929 * in putiocd()
5929 5930 */
5930 5931 id = K_TO_K | sigflag;
5931 5932 else
5932 5933 id = flag;
5933 5934 if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5934 5935 freemsg(bp);
5935 5936 crfree(crp);
5936 5937 return (error);
5937 5938 }
5938 5939
5939 5940 /*
5940 5941 * We could have slept copying in user pages.
5941 5942 * Recheck the stream head state (the other end
5942 5943 * of a pipe could have gone away).
5943 5944 */
5944 5945 if (stp->sd_flag & errs) {
5945 5946 mutex_enter(&stp->sd_lock);
5946 5947 error = strgeterr(stp, errs, 0);
5947 5948 mutex_exit(&stp->sd_lock);
5948 5949 if (error != 0) {
5949 5950 freemsg(bp);
5950 5951 crfree(crp);
5951 5952 return (error);
5952 5953 }
5953 5954 }
5954 5955 }
5955 5956 if (transparent)
5956 5957 iocbp->ioc_count = TRANSPARENT;
5957 5958
5958 5959 /*
5959 5960 * Block for up to STRTIMOUT milliseconds if there is an outstanding
5960 5961 * ioctl for this stream already running. All processes
5961 5962 * sleeping here will be awakened as a result of an ACK
5962 5963 * or NAK being received for the outstanding ioctl, or
5963 5964 * as a result of the timer expiring on the outstanding
5964 5965 * ioctl (a failure), or as a result of any waiting
5965 5966 * process's timer expiring (also a failure).
5966 5967 */
5967 5968
5968 5969 error = 0;
5969 5970 mutex_enter(&stp->sd_lock);
5970 5971 while ((stp->sd_flag & IOCWAIT) ||
5971 5972 (!set_iocwaitne && (stp->sd_flag & IOCWAITNE))) {
5972 5973 clock_t cv_rval;
5973 5974
5974 5975 TRACE_0(TR_FAC_STREAMS_FR,
5975 5976 TR_STRDOIOCTL_WAIT,
5976 5977 "strdoioctl sleeps - IOCWAIT");
5977 5978 cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
5978 5979 STRTIMOUT, sigflag);
5979 5980 if (cv_rval <= 0) {
5980 5981 if (cv_rval == 0) {
5981 5982 error = EINTR;
5982 5983 } else {
5983 5984 if (flag & STR_NOERROR) {
5984 5985 /*
5985 5986 * Terminating current ioctl in
5986 5987 * progress -- assume it got lost and
5987 5988 * wake up the other thread so that the
5988 5989 * operation completes.
5989 5990 */
5990 5991 if (!(stp->sd_flag & IOCWAITNE)) {
5991 5992 set_iocwaitne = B_TRUE;
5992 5993 stp->sd_flag |= IOCWAITNE;
5993 5994 cv_broadcast(&stp->sd_monitor);
5994 5995 }
5995 5996 /*
5996 5997 * Otherwise, there's a running
5997 5998 * STR_NOERROR -- we have no choice
5998 5999 * here but to wait forever (or until
5999 6000 * interrupted).
6000 6001 */
6001 6002 } else {
6002 6003 /*
6003 6004 * pending ioctl has caused
6004 6005 * us to time out
6005 6006 */
6006 6007 error = ETIME;
6007 6008 }
6008 6009 }
6009 6010 } else if ((stp->sd_flag & errs)) {
6010 6011 error = strgeterr(stp, errs, 0);
6011 6012 }
6012 6013 if (error) {
6013 6014 mutex_exit(&stp->sd_lock);
6014 6015 freemsg(bp);
6015 6016 crfree(crp);
6016 6017 return (error);
6017 6018 }
6018 6019 }
6019 6020
6020 6021 /*
6021 6022 * Have control of ioctl mechanism.
6022 6023 * Send down ioctl packet and wait for response.
6023 6024 */
6024 6025 if (stp->sd_iocblk != (mblk_t *)-1) {
6025 6026 freemsg(stp->sd_iocblk);
6026 6027 }
6027 6028 stp->sd_iocblk = NULL;
6028 6029
6029 6030 /*
6030 6031 * If this is marked with 'noerror' (internal; mostly
6031 6032 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6032 6033 * in here by setting IOCWAITNE.
6033 6034 */
6034 6035 waitflags = IOCWAIT;
6035 6036 if (flag & STR_NOERROR)
6036 6037 waitflags |= IOCWAITNE;
6037 6038
6038 6039 stp->sd_flag |= waitflags;
6039 6040
6040 6041 /*
6041 6042 * Assign sequence number.
6042 6043 */
6043 6044 iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6044 6045
6045 6046 mutex_exit(&stp->sd_lock);
6046 6047
6047 6048 TRACE_1(TR_FAC_STREAMS_FR,
6048 6049 TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6049 6050 stream_willservice(stp);
6050 6051 putnext(stp->sd_wrq, bp);
6051 6052 stream_runservice(stp);
6052 6053
6053 6054 /*
6054 6055 * Timed wait for acknowledgment. The wait time is limited by the
6055 6056 * timeout value, which must be a positive integer (number of
6056 6057 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6057 6058 * milliseconds), or -1 (wait forever). This will be awakened
6058 6059 * either by an ACK/NAK message arriving, the timer expiring, or
6059 6060 * the timer expiring on another ioctl waiting for control of the
6060 6061 * mechanism.
6061 6062 */
6062 6063 waitioc:
6063 6064 mutex_enter(&stp->sd_lock);
6064 6065
6065 6066
6066 6067 /*
6067 6068 * If the reply has already arrived, don't sleep. If awakened from
6068 6069 * the sleep, fail only if the reply has not arrived by then.
6069 6070 * Otherwise, process the reply.
6070 6071 */
6071 6072 while (!stp->sd_iocblk) {
6072 6073 clock_t cv_rval;
6073 6074
6074 6075 if (stp->sd_flag & errs) {
6075 6076 error = strgeterr(stp, errs, 0);
6076 6077 if (error != 0) {
6077 6078 stp->sd_flag &= ~waitflags;
6078 6079 cv_broadcast(&stp->sd_iocmonitor);
6079 6080 mutex_exit(&stp->sd_lock);
6080 6081 crfree(crp);
6081 6082 return (error);
6082 6083 }
6083 6084 }
6084 6085
6085 6086 TRACE_0(TR_FAC_STREAMS_FR,
6086 6087 TR_STRDOIOCTL_WAIT2,
6087 6088 "strdoioctl sleeps awaiting reply");
6088 6089 ASSERT(error == 0);
6089 6090
6090 6091 cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6091 6092 (strioc->ic_timout ?
6092 6093 strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6093 6094
6094 6095 /*
6095 6096 * There are four possible cases here: interrupt, timeout,
6096 6097 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6097 6098 * valid M_IOCTL reply).
6098 6099 *
6099 6100 * If we've been awakened by a STR_NOERROR ioctl on some other
6100 6101 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6101 6102 * will be set. Pretend as if we just timed out. Note that
6102 6103 * this other thread waited at least STRTIMOUT before trying to
6103 6104 * awaken our thread, so this is indistinguishable (even for
6104 6105 * INFTIM) from the case where we failed with ETIME waiting on
6105 6106 * IOCWAIT in the prior loop.
6106 6107 */
6107 6108 if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6108 6109 stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6109 6110 cv_rval = -1;
6110 6111 }
6111 6112
6112 6113 /*
6113 6114 * note: STR_NOERROR does not protect
6114 6115 * us here.. use ic_timout < 0
6115 6116 */
6116 6117 if (cv_rval <= 0) {
6117 6118 if (cv_rval == 0) {
6118 6119 error = EINTR;
6119 6120 } else {
6120 6121 error = ETIME;
6121 6122 }
6122 6123 /*
6123 6124 * A message could have come in after we were scheduled
6124 6125 * but before we were actually run.
6125 6126 */
6126 6127 bp = stp->sd_iocblk;
6127 6128 stp->sd_iocblk = NULL;
6128 6129 if (bp != NULL) {
6129 6130 if ((bp->b_datap->db_type == M_COPYIN) ||
6130 6131 (bp->b_datap->db_type == M_COPYOUT)) {
6131 6132 mutex_exit(&stp->sd_lock);
6132 6133 if (bp->b_cont) {
6133 6134 freemsg(bp->b_cont);
6134 6135 bp->b_cont = NULL;
6135 6136 }
6136 6137 bp->b_datap->db_type = M_IOCDATA;
6137 6138 bp->b_wptr = bp->b_rptr +
6138 6139 sizeof (struct copyresp);
6139 6140 resp = (struct copyresp *)bp->b_rptr;
6140 6141 resp->cp_rval =
6141 6142 (caddr_t)1; /* failure */
6142 6143 stream_willservice(stp);
6143 6144 putnext(stp->sd_wrq, bp);
6144 6145 stream_runservice(stp);
6145 6146 mutex_enter(&stp->sd_lock);
6146 6147 } else {
6147 6148 freemsg(bp);
6148 6149 }
6149 6150 }
6150 6151 stp->sd_flag &= ~waitflags;
6151 6152 cv_broadcast(&stp->sd_iocmonitor);
6152 6153 mutex_exit(&stp->sd_lock);
6153 6154 crfree(crp);
6154 6155 return (error);
6155 6156 }
6156 6157 }
6157 6158 bp = stp->sd_iocblk;
6158 6159 /*
6159 6160 * Note: it is strictly impossible to get here with sd_iocblk set to
6160 6161 * -1. This is because the initial loop above doesn't allow any new
6161 6162 * ioctls into the fray until all others have passed this point.
6162 6163 */
6163 6164 ASSERT(bp != NULL && bp != (mblk_t *)-1);
6164 6165 TRACE_1(TR_FAC_STREAMS_FR,
6165 6166 TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6166 6167 if ((bp->b_datap->db_type == M_IOCACK) ||
6167 6168 (bp->b_datap->db_type == M_IOCNAK)) {
6168 6169 /* for detection of duplicate ioctl replies */
6169 6170 stp->sd_iocblk = (mblk_t *)-1;
6170 6171 stp->sd_flag &= ~waitflags;
6171 6172 cv_broadcast(&stp->sd_iocmonitor);
6172 6173 mutex_exit(&stp->sd_lock);
6173 6174 } else {
6174 6175 /*
6175 6176 * flags not cleared here because we're still doing
6176 6177 * copy in/out for ioctl.
6177 6178 */
6178 6179 stp->sd_iocblk = NULL;
6179 6180 mutex_exit(&stp->sd_lock);
6180 6181 }
6181 6182
6182 6183
6183 6184 /*
6184 6185 * Have received acknowledgment.
6185 6186 */
6186 6187
6187 6188 switch (bp->b_datap->db_type) {
6188 6189 case M_IOCACK:
6189 6190 /*
6190 6191 * Positive ack.
6191 6192 */
6192 6193 iocbp = (struct iocblk *)bp->b_rptr;
6193 6194
6194 6195 /*
6195 6196 * Set error if indicated.
6196 6197 */
6197 6198 if (iocbp->ioc_error) {
6198 6199 error = iocbp->ioc_error;
6199 6200 break;
6200 6201 }
6201 6202
6202 6203 /*
6203 6204 * Set return value.
6204 6205 */
6205 6206 *rvalp = iocbp->ioc_rval;
6206 6207
6207 6208 /*
6208 6209 * Data may have been returned in ACK message (ioc_count > 0).
6209 6210 * If so, copy it out to the user's buffer.
6210 6211 */
6211 6212 if (iocbp->ioc_count && !transparent) {
6212 6213 if (error = getiocd(bp, strioc->ic_dp, copyflag))
6213 6214 break;
6214 6215 }
6215 6216 if (!transparent) {
6216 6217 if (len) /* an M_COPYOUT was used with I_STR */
6217 6218 strioc->ic_len = len;
6218 6219 else
6219 6220 strioc->ic_len = (int)iocbp->ioc_count;
6220 6221 }
6221 6222 break;
6222 6223
6223 6224 case M_IOCNAK:
6224 6225 /*
6225 6226 * Negative ack.
6226 6227 *
6227 6228 * The only thing to do is set error as specified
6228 6229 * in neg ack packet.
6229 6230 */
6230 6231 iocbp = (struct iocblk *)bp->b_rptr;
6231 6232
6232 6233 error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6233 6234 break;
6234 6235
6235 6236 case M_COPYIN:
6236 6237 /*
6237 6238 * Driver or module has requested user ioctl data.
6238 6239 */
6239 6240 reqp = (struct copyreq *)bp->b_rptr;
6240 6241
6241 6242 /*
6242 6243 * M_COPYIN should *never* have a message attached, though
6243 6244 * it's harmless if it does -- thus, panic on a DEBUG
6244 6245 * kernel and just free it on a non-DEBUG build.
6245 6246 */
6246 6247 ASSERT(bp->b_cont == NULL);
6247 6248 if (bp->b_cont != NULL) {
6248 6249 freemsg(bp->b_cont);
6249 6250 bp->b_cont = NULL;
6250 6251 }
6251 6252
6252 6253 error = putiocd(bp, reqp->cq_addr, flag, crp);
6253 6254 if (error && bp->b_cont) {
6254 6255 freemsg(bp->b_cont);
6255 6256 bp->b_cont = NULL;
6256 6257 }
6257 6258
6258 6259 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6259 6260 bp->b_datap->db_type = M_IOCDATA;
6260 6261
6261 6262 mblk_setcred(bp, crp, curproc->p_pid);
6262 6263 resp = (struct copyresp *)bp->b_rptr;
6263 6264 resp->cp_rval = (caddr_t)(uintptr_t)error;
6264 6265 resp->cp_flag = (fflags & FMODELS);
6265 6266
6266 6267 stream_willservice(stp);
6267 6268 putnext(stp->sd_wrq, bp);
6268 6269 stream_runservice(stp);
6269 6270
6270 6271 if (error) {
6271 6272 mutex_enter(&stp->sd_lock);
6272 6273 stp->sd_flag &= ~waitflags;
6273 6274 cv_broadcast(&stp->sd_iocmonitor);
6274 6275 mutex_exit(&stp->sd_lock);
6275 6276 crfree(crp);
6276 6277 return (error);
6277 6278 }
6278 6279
6279 6280 goto waitioc;
6280 6281
6281 6282 case M_COPYOUT:
6282 6283 /*
6283 6284 * Driver or module has ioctl data for a user.
6284 6285 */
6285 6286 reqp = (struct copyreq *)bp->b_rptr;
6286 6287 ASSERT(bp->b_cont != NULL);
6287 6288
6288 6289 /*
6289 6290 * Always (transparent or non-transparent )
6290 6291 * use the address specified in the request
6291 6292 */
6292 6293 taddr = reqp->cq_addr;
6293 6294 if (!transparent)
6294 6295 len = (int)reqp->cq_size;
6295 6296
6296 6297 /* copyout data to the provided address */
6297 6298 error = getiocd(bp, taddr, copyflag);
6298 6299
6299 6300 freemsg(bp->b_cont);
6300 6301 bp->b_cont = NULL;
6301 6302
6302 6303 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6303 6304 bp->b_datap->db_type = M_IOCDATA;
6304 6305
6305 6306 mblk_setcred(bp, crp, curproc->p_pid);
6306 6307 resp = (struct copyresp *)bp->b_rptr;
6307 6308 resp->cp_rval = (caddr_t)(uintptr_t)error;
6308 6309 resp->cp_flag = (fflags & FMODELS);
6309 6310
6310 6311 stream_willservice(stp);
6311 6312 putnext(stp->sd_wrq, bp);
6312 6313 stream_runservice(stp);
6313 6314
6314 6315 if (error) {
6315 6316 mutex_enter(&stp->sd_lock);
6316 6317 stp->sd_flag &= ~waitflags;
6317 6318 cv_broadcast(&stp->sd_iocmonitor);
6318 6319 mutex_exit(&stp->sd_lock);
6319 6320 crfree(crp);
6320 6321 return (error);
6321 6322 }
6322 6323 goto waitioc;
6323 6324
6324 6325 default:
6325 6326 ASSERT(0);
6326 6327 mutex_enter(&stp->sd_lock);
6327 6328 stp->sd_flag &= ~waitflags;
6328 6329 cv_broadcast(&stp->sd_iocmonitor);
6329 6330 mutex_exit(&stp->sd_lock);
6330 6331 break;
6331 6332 }
6332 6333
6333 6334 freemsg(bp);
6334 6335 crfree(crp);
6335 6336 return (error);
6336 6337 }
6337 6338
6338 6339 /*
6339 6340 * Send an M_CMD message downstream and wait for a reply. This is a ptools
6340 6341 * special used to retrieve information from modules/drivers a stream without
6341 6342 * being subjected to flow control or interfering with pending messages on the
6342 6343 * stream (e.g. an ioctl in flight).
6343 6344 */
6344 6345 int
6345 6346 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6346 6347 {
6347 6348 mblk_t *mp;
6348 6349 struct cmdblk *cmdp;
6349 6350 int error = 0;
6350 6351 int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6351 6352 clock_t rval, timeout = STRTIMOUT;
6352 6353
6353 6354 if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6354 6355 scp->sc_timeout < -1)
6355 6356 return (EINVAL);
6356 6357
6357 6358 if (scp->sc_timeout > 0)
6358 6359 timeout = scp->sc_timeout * MILLISEC;
6359 6360
6360 6361 if ((mp = allocb_cred(sizeof (struct cmdblk), crp,
6361 6362 curproc->p_pid)) == NULL)
6362 6363 return (ENOMEM);
6363 6364
6364 6365 crhold(crp);
6365 6366
6366 6367 cmdp = (struct cmdblk *)mp->b_wptr;
6367 6368 cmdp->cb_cr = crp;
6368 6369 cmdp->cb_cmd = scp->sc_cmd;
6369 6370 cmdp->cb_len = scp->sc_len;
6370 6371 cmdp->cb_error = 0;
6371 6372 mp->b_wptr += sizeof (struct cmdblk);
6372 6373
6373 6374 DB_TYPE(mp) = M_CMD;
6374 6375 DB_CPID(mp) = curproc->p_pid;
6375 6376
6376 6377 /*
6377 6378 * Copy in the payload.
6378 6379 */
6379 6380 if (cmdp->cb_len > 0) {
6380 6381 mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp,
6381 6382 curproc->p_pid);
6382 6383 if (mp->b_cont == NULL) {
6383 6384 error = ENOMEM;
6384 6385 goto out;
6385 6386 }
6386 6387
6387 6388 /* cb_len comes from sc_len, which has already been checked */
6388 6389 ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6389 6390 (void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6390 6391 mp->b_cont->b_wptr += cmdp->cb_len;
6391 6392 DB_CPID(mp->b_cont) = curproc->p_pid;
6392 6393 }
6393 6394
6394 6395 /*
6395 6396 * Since this mechanism is strictly for ptools, and since only one
6396 6397 * process can be grabbed at a time, we simply fail if there's
6397 6398 * currently an operation pending.
6398 6399 */
6399 6400 mutex_enter(&stp->sd_lock);
6400 6401 if (stp->sd_flag & STRCMDWAIT) {
6401 6402 mutex_exit(&stp->sd_lock);
6402 6403 error = EBUSY;
6403 6404 goto out;
6404 6405 }
6405 6406 stp->sd_flag |= STRCMDWAIT;
6406 6407 ASSERT(stp->sd_cmdblk == NULL);
6407 6408 mutex_exit(&stp->sd_lock);
6408 6409
6409 6410 putnext(stp->sd_wrq, mp);
6410 6411 mp = NULL;
6411 6412
6412 6413 /*
6413 6414 * Timed wait for acknowledgment. If the reply has already arrived,
6414 6415 * don't sleep. If awakened from the sleep, fail only if the reply
6415 6416 * has not arrived by then. Otherwise, process the reply.
6416 6417 */
6417 6418 mutex_enter(&stp->sd_lock);
6418 6419 while (stp->sd_cmdblk == NULL) {
6419 6420 if (stp->sd_flag & errs) {
6420 6421 if ((error = strgeterr(stp, errs, 0)) != 0)
6421 6422 goto waitout;
6422 6423 }
6423 6424
6424 6425 rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6425 6426 if (stp->sd_cmdblk != NULL)
6426 6427 break;
6427 6428
6428 6429 if (rval <= 0) {
6429 6430 error = (rval == 0) ? EINTR : ETIME;
6430 6431 goto waitout;
6431 6432 }
6432 6433 }
6433 6434
6434 6435 /*
6435 6436 * We received a reply.
6436 6437 */
6437 6438 mp = stp->sd_cmdblk;
6438 6439 stp->sd_cmdblk = NULL;
6439 6440 ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6440 6441 ASSERT(stp->sd_flag & STRCMDWAIT);
6441 6442 stp->sd_flag &= ~STRCMDWAIT;
6442 6443 mutex_exit(&stp->sd_lock);
6443 6444
6444 6445 cmdp = (struct cmdblk *)mp->b_rptr;
6445 6446 if ((error = cmdp->cb_error) != 0)
6446 6447 goto out;
6447 6448
6448 6449 /*
6449 6450 * Data may have been returned in the reply (cb_len > 0).
6450 6451 * If so, copy it out to the user's buffer.
6451 6452 */
6452 6453 if (cmdp->cb_len > 0) {
6453 6454 if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6454 6455 error = EPROTO;
6455 6456 goto out;
6456 6457 }
6457 6458
6458 6459 cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6459 6460 (void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6460 6461 }
6461 6462 scp->sc_len = cmdp->cb_len;
6462 6463 out:
6463 6464 freemsg(mp);
6464 6465 crfree(crp);
6465 6466 return (error);
6466 6467 waitout:
6467 6468 ASSERT(stp->sd_cmdblk == NULL);
6468 6469 stp->sd_flag &= ~STRCMDWAIT;
6469 6470 mutex_exit(&stp->sd_lock);
6470 6471 crfree(crp);
6471 6472 return (error);
6472 6473 }
6473 6474
6474 6475 /*
6475 6476 * For the SunOS keyboard driver.
6476 6477 * Return the next available "ioctl" sequence number.
6477 6478 * Exported, so that streams modules can send "ioctl" messages
6478 6479 * downstream from their open routine.
6479 6480 */
6480 6481 int
6481 6482 getiocseqno(void)
6482 6483 {
6483 6484 int i;
6484 6485
6485 6486 mutex_enter(&strresources);
6486 6487 i = ++ioc_id;
6487 6488 mutex_exit(&strresources);
6488 6489 return (i);
6489 6490 }
6490 6491
6491 6492 /*
6492 6493 * Get the next message from the read queue. If the message is
6493 6494 * priority, STRPRI will have been set by strrput(). This flag
6494 6495 * should be reset only when the entire message at the front of the
6495 6496 * queue as been consumed.
6496 6497 *
6497 6498 * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6498 6499 */
6499 6500 int
6500 6501 strgetmsg(
6501 6502 struct vnode *vp,
6502 6503 struct strbuf *mctl,
6503 6504 struct strbuf *mdata,
6504 6505 unsigned char *prip,
6505 6506 int *flagsp,
6506 6507 int fmode,
6507 6508 rval_t *rvp)
6508 6509 {
6509 6510 struct stdata *stp;
6510 6511 mblk_t *bp, *nbp;
6511 6512 mblk_t *savemp = NULL;
6512 6513 mblk_t *savemptail = NULL;
6513 6514 uint_t old_sd_flag;
6514 6515 int flg;
6515 6516 int more = 0;
6516 6517 int error = 0;
6517 6518 char first = 1;
6518 6519 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
6519 6520 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
6520 6521 unsigned char pri = 0;
6521 6522 queue_t *q;
6522 6523 int pr = 0; /* Partial read successful */
6523 6524 struct uio uios;
6524 6525 struct uio *uiop = &uios;
6525 6526 struct iovec iovs;
6526 6527 unsigned char type;
6527 6528
6528 6529 TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6529 6530 "strgetmsg:%p", vp);
6530 6531
6531 6532 ASSERT(vp->v_stream);
6532 6533 stp = vp->v_stream;
6533 6534 rvp->r_val1 = 0;
6534 6535
6535 6536 mutex_enter(&stp->sd_lock);
6536 6537
6537 6538 if ((error = i_straccess(stp, JCREAD)) != 0) {
6538 6539 mutex_exit(&stp->sd_lock);
6539 6540 return (error);
6540 6541 }
6541 6542
6542 6543 if (stp->sd_flag & (STRDERR|STPLEX)) {
6543 6544 error = strgeterr(stp, STRDERR|STPLEX, 0);
6544 6545 if (error != 0) {
6545 6546 mutex_exit(&stp->sd_lock);
6546 6547 return (error);
6547 6548 }
6548 6549 }
6549 6550 mutex_exit(&stp->sd_lock);
6550 6551
6551 6552 switch (*flagsp) {
6552 6553 case MSG_HIPRI:
6553 6554 if (*prip != 0)
6554 6555 return (EINVAL);
6555 6556 break;
6556 6557
6557 6558 case MSG_ANY:
6558 6559 case MSG_BAND:
6559 6560 break;
6560 6561
6561 6562 default:
6562 6563 return (EINVAL);
6563 6564 }
6564 6565 /*
6565 6566 * Setup uio and iov for data part
6566 6567 */
6567 6568 iovs.iov_base = mdata->buf;
6568 6569 iovs.iov_len = mdata->maxlen;
6569 6570 uios.uio_iov = &iovs;
6570 6571 uios.uio_iovcnt = 1;
6571 6572 uios.uio_loffset = 0;
6572 6573 uios.uio_segflg = UIO_USERSPACE;
6573 6574 uios.uio_fmode = 0;
6574 6575 uios.uio_extflg = UIO_COPY_CACHED;
6575 6576 uios.uio_resid = mdata->maxlen;
6576 6577 uios.uio_offset = 0;
6577 6578
6578 6579 q = _RD(stp->sd_wrq);
6579 6580 mutex_enter(&stp->sd_lock);
6580 6581 old_sd_flag = stp->sd_flag;
6581 6582 mark = 0;
6582 6583 for (;;) {
6583 6584 int done = 0;
6584 6585 mblk_t *q_first = q->q_first;
6585 6586
6586 6587 /*
6587 6588 * Get the next message of appropriate priority
6588 6589 * from the stream head. If the caller is interested
6589 6590 * in band or hipri messages, then they should already
6590 6591 * be enqueued at the stream head. On the other hand
6591 6592 * if the caller wants normal (band 0) messages, they
6592 6593 * might be deferred in a synchronous stream and they
6593 6594 * will need to be pulled up.
6594 6595 *
6595 6596 * After we have dequeued a message, we might find that
6596 6597 * it was a deferred M_SIG that was enqueued at the
6597 6598 * stream head. It must now be posted as part of the
6598 6599 * read by calling strsignal_nolock().
6599 6600 *
6600 6601 * Also note that strrput does not enqueue an M_PCSIG,
6601 6602 * and there cannot be more than one hipri message,
6602 6603 * so there was no need to have the M_PCSIG case.
6603 6604 *
6604 6605 * At some time it might be nice to try and wrap the
6605 6606 * functionality of kstrgetmsg() and strgetmsg() into
6606 6607 * a common routine so to reduce the amount of replicated
6607 6608 * code (since they are extremely similar).
6608 6609 */
6609 6610 if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6610 6611 /* Asking for normal, band0 data */
6611 6612 bp = strget(stp, q, uiop, first, &error);
6612 6613 ASSERT(MUTEX_HELD(&stp->sd_lock));
6613 6614 if (bp != NULL) {
6614 6615 if (DB_TYPE(bp) == M_SIG) {
6615 6616 strsignal_nolock(stp, *bp->b_rptr,
6616 6617 bp->b_band);
6617 6618 freemsg(bp);
6618 6619 continue;
6619 6620 } else {
6620 6621 break;
6621 6622 }
6622 6623 }
6623 6624 if (error != 0)
6624 6625 goto getmout;
6625 6626
6626 6627 /*
6627 6628 * We can't depend on the value of STRPRI here because
6628 6629 * the stream head may be in transit. Therefore, we
6629 6630 * must look at the type of the first message to
6630 6631 * determine if a high priority messages is waiting
6631 6632 */
6632 6633 } else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6633 6634 DB_TYPE(q_first) >= QPCTL &&
6634 6635 (bp = getq_noenab(q, 0)) != NULL) {
6635 6636 /* Asked for HIPRI and got one */
6636 6637 ASSERT(DB_TYPE(bp) >= QPCTL);
6637 6638 break;
6638 6639 } else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6639 6640 ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
6640 6641 (bp = getq_noenab(q, 0)) != NULL) {
6641 6642 /*
6642 6643 * Asked for at least band "prip" and got either at
6643 6644 * least that band or a hipri message.
6644 6645 */
6645 6646 ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
6646 6647 if (DB_TYPE(bp) == M_SIG) {
6647 6648 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6648 6649 freemsg(bp);
6649 6650 continue;
6650 6651 } else {
6651 6652 break;
6652 6653 }
6653 6654 }
6654 6655
6655 6656 /* No data. Time to sleep? */
6656 6657 qbackenable(q, 0);
6657 6658
6658 6659 /*
6659 6660 * If STRHUP or STREOF, return 0 length control and data.
6660 6661 * If resid is 0, then a read(fd,buf,0) was done. Do not
6661 6662 * sleep to satisfy this request because by default we have
6662 6663 * zero bytes to return.
6663 6664 */
6664 6665 if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6665 6666 mdata->maxlen == 0)) {
6666 6667 mctl->len = mdata->len = 0;
6667 6668 *flagsp = 0;
6668 6669 mutex_exit(&stp->sd_lock);
6669 6670 return (0);
6670 6671 }
6671 6672 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6672 6673 "strgetmsg calls strwaitq:%p, %p",
6673 6674 vp, uiop);
6674 6675 if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6675 6676 &done)) != 0) || done) {
6676 6677 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6677 6678 "strgetmsg error or done:%p, %p",
6678 6679 vp, uiop);
6679 6680 mutex_exit(&stp->sd_lock);
6680 6681 return (error);
6681 6682 }
6682 6683 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6683 6684 "strgetmsg awakes:%p, %p", vp, uiop);
6684 6685 if ((error = i_straccess(stp, JCREAD)) != 0) {
6685 6686 mutex_exit(&stp->sd_lock);
6686 6687 return (error);
6687 6688 }
6688 6689 first = 0;
6689 6690 }
6690 6691 ASSERT(bp != NULL);
6691 6692 /*
6692 6693 * Extract any mark information. If the message is not completely
6693 6694 * consumed this information will be put in the mblk
6694 6695 * that is putback.
6695 6696 * If MSGMARKNEXT is set and the message is completely consumed
6696 6697 * the STRATMARK flag will be set below. Likewise, if
6697 6698 * MSGNOTMARKNEXT is set and the message is
6698 6699 * completely consumed STRNOTATMARK will be set.
6699 6700 */
6700 6701 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6701 6702 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6702 6703 (MSGMARKNEXT|MSGNOTMARKNEXT));
6703 6704 if (mark != 0 && bp == stp->sd_mark) {
6704 6705 mark |= _LASTMARK;
6705 6706 stp->sd_mark = NULL;
6706 6707 }
6707 6708 /*
6708 6709 * keep track of the original message type and priority
6709 6710 */
6710 6711 pri = bp->b_band;
6711 6712 type = bp->b_datap->db_type;
6712 6713 if (type == M_PASSFP) {
6713 6714 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6714 6715 stp->sd_mark = bp;
6715 6716 bp->b_flag |= mark & ~_LASTMARK;
6716 6717 putback(stp, q, bp, pri);
6717 6718 qbackenable(q, pri);
6718 6719 mutex_exit(&stp->sd_lock);
6719 6720 return (EBADMSG);
6720 6721 }
6721 6722 ASSERT(type != M_SIG);
6722 6723
6723 6724 /*
6724 6725 * Set this flag so strrput will not generate signals. Need to
6725 6726 * make sure this flag is cleared before leaving this routine
6726 6727 * else signals will stop being sent.
6727 6728 */
6728 6729 stp->sd_flag |= STRGETINPROG;
6729 6730 mutex_exit(&stp->sd_lock);
6730 6731
6731 6732 if (STREAM_NEEDSERVICE(stp))
6732 6733 stream_runservice(stp);
6733 6734
6734 6735 /*
6735 6736 * Set HIPRI flag if message is priority.
6736 6737 */
6737 6738 if (type >= QPCTL)
6738 6739 flg = MSG_HIPRI;
6739 6740 else
6740 6741 flg = MSG_BAND;
6741 6742
6742 6743 /*
6743 6744 * First process PROTO or PCPROTO blocks, if any.
6744 6745 */
6745 6746 if (mctl->maxlen >= 0 && type != M_DATA) {
6746 6747 size_t n, bcnt;
6747 6748 char *ubuf;
6748 6749
6749 6750 bcnt = mctl->maxlen;
6750 6751 ubuf = mctl->buf;
6751 6752 while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6752 6753 if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6753 6754 copyout(bp->b_rptr, ubuf, n)) {
6754 6755 error = EFAULT;
6755 6756 mutex_enter(&stp->sd_lock);
6756 6757 /*
6757 6758 * clear stream head pri flag based on
6758 6759 * first message type
6759 6760 */
6760 6761 if (type >= QPCTL) {
6761 6762 ASSERT(type == M_PCPROTO);
6762 6763 stp->sd_flag &= ~STRPRI;
6763 6764 }
6764 6765 more = 0;
6765 6766 freemsg(bp);
6766 6767 goto getmout;
6767 6768 }
6768 6769 ubuf += n;
6769 6770 bp->b_rptr += n;
6770 6771 if (bp->b_rptr >= bp->b_wptr) {
6771 6772 nbp = bp;
6772 6773 bp = bp->b_cont;
6773 6774 freeb(nbp);
6774 6775 }
6775 6776 ASSERT(n <= bcnt);
6776 6777 bcnt -= n;
6777 6778 if (bcnt == 0)
6778 6779 break;
6779 6780 }
6780 6781 mctl->len = mctl->maxlen - bcnt;
6781 6782 } else
6782 6783 mctl->len = -1;
6783 6784
6784 6785 if (bp && bp->b_datap->db_type != M_DATA) {
6785 6786 /*
6786 6787 * More PROTO blocks in msg.
6787 6788 */
6788 6789 more |= MORECTL;
6789 6790 savemp = bp;
6790 6791 while (bp && bp->b_datap->db_type != M_DATA) {
6791 6792 savemptail = bp;
6792 6793 bp = bp->b_cont;
6793 6794 }
6794 6795 savemptail->b_cont = NULL;
6795 6796 }
6796 6797
6797 6798 /*
6798 6799 * Now process DATA blocks, if any.
6799 6800 */
6800 6801 if (mdata->maxlen >= 0 && bp) {
6801 6802 /*
6802 6803 * struiocopyout will consume a potential zero-length
6803 6804 * M_DATA even if uio_resid is zero.
6804 6805 */
6805 6806 size_t oldresid = uiop->uio_resid;
6806 6807
6807 6808 bp = struiocopyout(bp, uiop, &error);
6808 6809 if (error != 0) {
6809 6810 mutex_enter(&stp->sd_lock);
6810 6811 /*
6811 6812 * clear stream head hi pri flag based on
6812 6813 * first message
6813 6814 */
6814 6815 if (type >= QPCTL) {
6815 6816 ASSERT(type == M_PCPROTO);
6816 6817 stp->sd_flag &= ~STRPRI;
6817 6818 }
6818 6819 more = 0;
6819 6820 freemsg(savemp);
6820 6821 goto getmout;
6821 6822 }
6822 6823 /*
6823 6824 * (pr == 1) indicates a partial read.
6824 6825 */
6825 6826 if (oldresid > uiop->uio_resid)
6826 6827 pr = 1;
6827 6828 mdata->len = mdata->maxlen - uiop->uio_resid;
6828 6829 } else
6829 6830 mdata->len = -1;
6830 6831
6831 6832 if (bp) { /* more data blocks in msg */
6832 6833 more |= MOREDATA;
6833 6834 if (savemp)
6834 6835 savemptail->b_cont = bp;
6835 6836 else
6836 6837 savemp = bp;
6837 6838 }
6838 6839
6839 6840 mutex_enter(&stp->sd_lock);
6840 6841 if (savemp) {
6841 6842 if (pr && (savemp->b_datap->db_type == M_DATA) &&
6842 6843 msgnodata(savemp)) {
6843 6844 /*
6844 6845 * Avoid queuing a zero-length tail part of
6845 6846 * a message. pr=1 indicates that we read some of
6846 6847 * the message.
6847 6848 */
6848 6849 freemsg(savemp);
6849 6850 more &= ~MOREDATA;
6850 6851 /*
6851 6852 * clear stream head hi pri flag based on
6852 6853 * first message
6853 6854 */
6854 6855 if (type >= QPCTL) {
6855 6856 ASSERT(type == M_PCPROTO);
6856 6857 stp->sd_flag &= ~STRPRI;
6857 6858 }
6858 6859 } else {
6859 6860 savemp->b_band = pri;
6860 6861 /*
6861 6862 * If the first message was HIPRI and the one we're
6862 6863 * putting back isn't, then clear STRPRI, otherwise
6863 6864 * set STRPRI again. Note that we must set STRPRI
6864 6865 * again since the flush logic in strrput_nondata()
6865 6866 * may have cleared it while we had sd_lock dropped.
6866 6867 */
6867 6868 if (type >= QPCTL) {
6868 6869 ASSERT(type == M_PCPROTO);
6869 6870 if (queclass(savemp) < QPCTL)
6870 6871 stp->sd_flag &= ~STRPRI;
6871 6872 else
6872 6873 stp->sd_flag |= STRPRI;
6873 6874 } else if (queclass(savemp) >= QPCTL) {
6874 6875 /*
6875 6876 * The first message was not a HIPRI message,
6876 6877 * but the one we are about to putback is.
6877 6878 * For simplicitly, we do not allow for HIPRI
6878 6879 * messages to be embedded in the message
6879 6880 * body, so just force it to same type as
6880 6881 * first message.
6881 6882 */
6882 6883 ASSERT(type == M_DATA || type == M_PROTO);
6883 6884 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6884 6885 savemp->b_datap->db_type = type;
6885 6886 }
6886 6887 if (mark != 0) {
6887 6888 savemp->b_flag |= mark & ~_LASTMARK;
6888 6889 if ((mark & _LASTMARK) &&
6889 6890 (stp->sd_mark == NULL)) {
6890 6891 /*
6891 6892 * If another marked message arrived
6892 6893 * while sd_lock was not held sd_mark
6893 6894 * would be non-NULL.
6894 6895 */
6895 6896 stp->sd_mark = savemp;
6896 6897 }
6897 6898 }
6898 6899 putback(stp, q, savemp, pri);
6899 6900 }
6900 6901 } else {
6901 6902 /*
6902 6903 * The complete message was consumed.
6903 6904 *
6904 6905 * If another M_PCPROTO arrived while sd_lock was not held
6905 6906 * it would have been discarded since STRPRI was still set.
6906 6907 *
6907 6908 * Move the MSG*MARKNEXT information
6908 6909 * to the stream head just in case
6909 6910 * the read queue becomes empty.
6910 6911 * clear stream head hi pri flag based on
6911 6912 * first message
6912 6913 *
6913 6914 * If the stream head was at the mark
6914 6915 * (STRATMARK) before we dropped sd_lock above
6915 6916 * and some data was consumed then we have
6916 6917 * moved past the mark thus STRATMARK is
6917 6918 * cleared. However, if a message arrived in
6918 6919 * strrput during the copyout above causing
6919 6920 * STRATMARK to be set we can not clear that
6920 6921 * flag.
6921 6922 */
6922 6923 if (type >= QPCTL) {
6923 6924 ASSERT(type == M_PCPROTO);
6924 6925 stp->sd_flag &= ~STRPRI;
6925 6926 }
6926 6927 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6927 6928 if (mark & MSGMARKNEXT) {
6928 6929 stp->sd_flag &= ~STRNOTATMARK;
6929 6930 stp->sd_flag |= STRATMARK;
6930 6931 } else if (mark & MSGNOTMARKNEXT) {
6931 6932 stp->sd_flag &= ~STRATMARK;
6932 6933 stp->sd_flag |= STRNOTATMARK;
6933 6934 } else {
6934 6935 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6935 6936 }
6936 6937 } else if (pr && (old_sd_flag & STRATMARK)) {
6937 6938 stp->sd_flag &= ~STRATMARK;
6938 6939 }
6939 6940 }
6940 6941
6941 6942 *flagsp = flg;
6942 6943 *prip = pri;
6943 6944
6944 6945 /*
6945 6946 * Getmsg cleanup processing - if the state of the queue has changed
6946 6947 * some signals may need to be sent and/or poll awakened.
6947 6948 */
6948 6949 getmout:
6949 6950 qbackenable(q, pri);
6950 6951
6951 6952 /*
6952 6953 * We dropped the stream head lock above. Send all M_SIG messages
6953 6954 * before processing stream head for SIGPOLL messages.
6954 6955 */
6955 6956 ASSERT(MUTEX_HELD(&stp->sd_lock));
6956 6957 while ((bp = q->q_first) != NULL &&
6957 6958 (bp->b_datap->db_type == M_SIG)) {
6958 6959 /*
6959 6960 * sd_lock is held so the content of the read queue can not
6960 6961 * change.
6961 6962 */
6962 6963 bp = getq(q);
6963 6964 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
6964 6965
6965 6966 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6966 6967 mutex_exit(&stp->sd_lock);
6967 6968 freemsg(bp);
6968 6969 if (STREAM_NEEDSERVICE(stp))
6969 6970 stream_runservice(stp);
6970 6971 mutex_enter(&stp->sd_lock);
6971 6972 }
6972 6973
6973 6974 /*
6974 6975 * stream head cannot change while we make the determination
6975 6976 * whether or not to send a signal. Drop the flag to allow strrput
6976 6977 * to send firstmsgsigs again.
6977 6978 */
6978 6979 stp->sd_flag &= ~STRGETINPROG;
6979 6980
6980 6981 /*
6981 6982 * If the type of message at the front of the queue changed
6982 6983 * due to the receive the appropriate signals and pollwakeup events
6983 6984 * are generated. The type of changes are:
6984 6985 * Processed a hipri message, q_first is not hipri.
6985 6986 * Processed a band X message, and q_first is band Y.
6986 6987 * The generated signals and pollwakeups are identical to what
6987 6988 * strrput() generates should the message that is now on q_first
6988 6989 * arrive to an empty read queue.
6989 6990 *
6990 6991 * Note: only strrput will send a signal for a hipri message.
6991 6992 */
6992 6993 if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
6993 6994 strsigset_t signals = 0;
6994 6995 strpollset_t pollwakeups = 0;
6995 6996
6996 6997 if (flg & MSG_HIPRI) {
6997 6998 /*
6998 6999 * Removed a hipri message. Regular data at
6999 7000 * the front of the queue.
7000 7001 */
7001 7002 if (bp->b_band == 0) {
7002 7003 signals = S_INPUT | S_RDNORM;
7003 7004 pollwakeups = POLLIN | POLLRDNORM;
7004 7005 } else {
7005 7006 signals = S_INPUT | S_RDBAND;
7006 7007 pollwakeups = POLLIN | POLLRDBAND;
7007 7008 }
7008 7009 } else if (pri != bp->b_band) {
7009 7010 /*
7010 7011 * The band is different for the new q_first.
7011 7012 */
7012 7013 if (bp->b_band == 0) {
7013 7014 signals = S_RDNORM;
7014 7015 pollwakeups = POLLIN | POLLRDNORM;
7015 7016 } else {
7016 7017 signals = S_RDBAND;
7017 7018 pollwakeups = POLLIN | POLLRDBAND;
7018 7019 }
7019 7020 }
7020 7021
7021 7022 if (pollwakeups != 0) {
7022 7023 if (pollwakeups == (POLLIN | POLLRDNORM)) {
7023 7024 if (!(stp->sd_rput_opt & SR_POLLIN))
7024 7025 goto no_pollwake;
7025 7026 stp->sd_rput_opt &= ~SR_POLLIN;
7026 7027 }
7027 7028 mutex_exit(&stp->sd_lock);
7028 7029 pollwakeup(&stp->sd_pollist, pollwakeups);
7029 7030 mutex_enter(&stp->sd_lock);
7030 7031 }
7031 7032 no_pollwake:
7032 7033
7033 7034 if (stp->sd_sigflags & signals)
7034 7035 strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7035 7036 }
7036 7037 mutex_exit(&stp->sd_lock);
7037 7038
7038 7039 rvp->r_val1 = more;
7039 7040 return (error);
7040 7041 #undef _LASTMARK
7041 7042 }
7042 7043
7043 7044 /*
7044 7045 * Get the next message from the read queue. If the message is
7045 7046 * priority, STRPRI will have been set by strrput(). This flag
7046 7047 * should be reset only when the entire message at the front of the
7047 7048 * queue as been consumed.
7048 7049 *
7049 7050 * If uiop is NULL all data is returned in mctlp.
7050 7051 * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7051 7052 * not enabled.
7052 7053 * The timeout parameter is in milliseconds; -1 for infinity.
7053 7054 * This routine handles the consolidation private flags:
7054 7055 * MSG_IGNERROR Ignore any stream head error except STPLEX.
7055 7056 * MSG_DELAYERROR Defer the error check until the queue is empty.
7056 7057 * MSG_HOLDSIG Hold signals while waiting for data.
7057 7058 * MSG_IPEEK Only peek at messages.
7058 7059 * MSG_DISCARDTAIL Discard the tail M_DATA part of the message
7059 7060 * that doesn't fit.
7060 7061 * MSG_NOMARK If the message is marked leave it on the queue.
7061 7062 *
7062 7063 * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7063 7064 */
7064 7065 int
7065 7066 kstrgetmsg(
7066 7067 struct vnode *vp,
7067 7068 mblk_t **mctlp,
7068 7069 struct uio *uiop,
7069 7070 unsigned char *prip,
7070 7071 int *flagsp,
7071 7072 clock_t timout,
7072 7073 rval_t *rvp)
7073 7074 {
7074 7075 struct stdata *stp;
7075 7076 mblk_t *bp, *nbp;
7076 7077 mblk_t *savemp = NULL;
7077 7078 mblk_t *savemptail = NULL;
7078 7079 int flags;
7079 7080 uint_t old_sd_flag;
7080 7081 int flg;
7081 7082 int more = 0;
7082 7083 int error = 0;
7083 7084 char first = 1;
7084 7085 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
7085 7086 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
7086 7087 unsigned char pri = 0;
7087 7088 queue_t *q;
7088 7089 int pr = 0; /* Partial read successful */
7089 7090 unsigned char type;
7090 7091
7091 7092 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7092 7093 "kstrgetmsg:%p", vp);
7093 7094
7094 7095 ASSERT(vp->v_stream);
7095 7096 stp = vp->v_stream;
7096 7097 rvp->r_val1 = 0;
7097 7098
7098 7099 mutex_enter(&stp->sd_lock);
7099 7100
7100 7101 if ((error = i_straccess(stp, JCREAD)) != 0) {
7101 7102 mutex_exit(&stp->sd_lock);
7102 7103 return (error);
7103 7104 }
7104 7105
7105 7106 flags = *flagsp;
7106 7107 if (stp->sd_flag & (STRDERR|STPLEX)) {
7107 7108 if ((stp->sd_flag & STPLEX) ||
7108 7109 (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7109 7110 error = strgeterr(stp, STRDERR|STPLEX,
7110 7111 (flags & MSG_IPEEK));
7111 7112 if (error != 0) {
7112 7113 mutex_exit(&stp->sd_lock);
7113 7114 return (error);
7114 7115 }
7115 7116 }
7116 7117 }
7117 7118 mutex_exit(&stp->sd_lock);
7118 7119
7119 7120 switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7120 7121 case MSG_HIPRI:
7121 7122 if (*prip != 0)
7122 7123 return (EINVAL);
7123 7124 break;
7124 7125
7125 7126 case MSG_ANY:
7126 7127 case MSG_BAND:
7127 7128 break;
7128 7129
7129 7130 default:
7130 7131 return (EINVAL);
7131 7132 }
7132 7133
7133 7134 retry:
7134 7135 q = _RD(stp->sd_wrq);
7135 7136 mutex_enter(&stp->sd_lock);
7136 7137 old_sd_flag = stp->sd_flag;
7137 7138 mark = 0;
7138 7139 for (;;) {
7139 7140 int done = 0;
7140 7141 int waitflag;
7141 7142 int fmode;
7142 7143 mblk_t *q_first = q->q_first;
7143 7144
7144 7145 /*
7145 7146 * This section of the code operates just like the code
7146 7147 * in strgetmsg(). There is a comment there about what
7147 7148 * is going on here.
7148 7149 */
7149 7150 if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7150 7151 /* Asking for normal, band0 data */
7151 7152 bp = strget(stp, q, uiop, first, &error);
7152 7153 ASSERT(MUTEX_HELD(&stp->sd_lock));
7153 7154 if (bp != NULL) {
7154 7155 if (DB_TYPE(bp) == M_SIG) {
7155 7156 strsignal_nolock(stp, *bp->b_rptr,
7156 7157 bp->b_band);
7157 7158 freemsg(bp);
7158 7159 continue;
7159 7160 } else {
7160 7161 break;
7161 7162 }
7162 7163 }
7163 7164 if (error != 0) {
7164 7165 goto getmout;
7165 7166 }
7166 7167 /*
7167 7168 * We can't depend on the value of STRPRI here because
7168 7169 * the stream head may be in transit. Therefore, we
7169 7170 * must look at the type of the first message to
7170 7171 * determine if a high priority messages is waiting
7171 7172 */
7172 7173 } else if ((flags & MSG_HIPRI) && q_first != NULL &&
7173 7174 DB_TYPE(q_first) >= QPCTL &&
7174 7175 (bp = getq_noenab(q, 0)) != NULL) {
7175 7176 ASSERT(DB_TYPE(bp) >= QPCTL);
7176 7177 break;
7177 7178 } else if ((flags & MSG_BAND) && q_first != NULL &&
7178 7179 ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
7179 7180 (bp = getq_noenab(q, 0)) != NULL) {
7180 7181 /*
7181 7182 * Asked for at least band "prip" and got either at
7182 7183 * least that band or a hipri message.
7183 7184 */
7184 7185 ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
7185 7186 if (DB_TYPE(bp) == M_SIG) {
7186 7187 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7187 7188 freemsg(bp);
7188 7189 continue;
7189 7190 } else {
7190 7191 break;
7191 7192 }
7192 7193 }
7193 7194
7194 7195 /* No data. Time to sleep? */
7195 7196 qbackenable(q, 0);
7196 7197
7197 7198 /*
7198 7199 * Delayed error notification?
7199 7200 */
7200 7201 if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7201 7202 (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7202 7203 error = strgeterr(stp, STRDERR|STPLEX,
7203 7204 (flags & MSG_IPEEK));
7204 7205 if (error != 0) {
7205 7206 mutex_exit(&stp->sd_lock);
7206 7207 return (error);
7207 7208 }
7208 7209 }
7209 7210
7210 7211 /*
7211 7212 * If STRHUP or STREOF, return 0 length control and data.
7212 7213 * If a read(fd,buf,0) has been done, do not sleep, just
7213 7214 * return.
7214 7215 *
7215 7216 * If mctlp == NULL and uiop == NULL, then the code will
7216 7217 * do the strwaitq. This is an understood way of saying
7217 7218 * sleep "polling" until a message is received.
7218 7219 */
7219 7220 if ((stp->sd_flag & (STRHUP|STREOF)) ||
7220 7221 (uiop != NULL && uiop->uio_resid == 0)) {
7221 7222 if (mctlp != NULL)
7222 7223 *mctlp = NULL;
7223 7224 *flagsp = 0;
7224 7225 mutex_exit(&stp->sd_lock);
7225 7226 return (0);
7226 7227 }
7227 7228
7228 7229 waitflag = GETWAIT;
7229 7230 if (flags &
7230 7231 (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7231 7232 if (flags & MSG_HOLDSIG)
7232 7233 waitflag |= STR_NOSIG;
7233 7234 if (flags & MSG_IGNERROR)
7234 7235 waitflag |= STR_NOERROR;
7235 7236 if (flags & MSG_IPEEK)
7236 7237 waitflag |= STR_PEEK;
7237 7238 if (flags & MSG_DELAYERROR)
7238 7239 waitflag |= STR_DELAYERR;
7239 7240 }
7240 7241 if (uiop != NULL)
7241 7242 fmode = uiop->uio_fmode;
7242 7243 else
7243 7244 fmode = 0;
7244 7245
7245 7246 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7246 7247 "kstrgetmsg calls strwaitq:%p, %p",
7247 7248 vp, uiop);
7248 7249 if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7249 7250 fmode, timout, &done))) != 0 || done) {
7250 7251 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7251 7252 "kstrgetmsg error or done:%p, %p",
7252 7253 vp, uiop);
7253 7254 mutex_exit(&stp->sd_lock);
7254 7255 return (error);
7255 7256 }
7256 7257 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7257 7258 "kstrgetmsg awakes:%p, %p", vp, uiop);
7258 7259 if ((error = i_straccess(stp, JCREAD)) != 0) {
7259 7260 mutex_exit(&stp->sd_lock);
7260 7261 return (error);
7261 7262 }
7262 7263 first = 0;
7263 7264 }
7264 7265 ASSERT(bp != NULL);
7265 7266 /*
7266 7267 * Extract any mark information. If the message is not completely
7267 7268 * consumed this information will be put in the mblk
7268 7269 * that is putback.
7269 7270 * If MSGMARKNEXT is set and the message is completely consumed
7270 7271 * the STRATMARK flag will be set below. Likewise, if
7271 7272 * MSGNOTMARKNEXT is set and the message is
7272 7273 * completely consumed STRNOTATMARK will be set.
7273 7274 */
7274 7275 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7275 7276 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7276 7277 (MSGMARKNEXT|MSGNOTMARKNEXT));
7277 7278 pri = bp->b_band;
7278 7279 if (mark != 0) {
7279 7280 /*
7280 7281 * If the caller doesn't want the mark return.
7281 7282 * Used to implement MSG_WAITALL in sockets.
7282 7283 */
7283 7284 if (flags & MSG_NOMARK) {
7284 7285 putback(stp, q, bp, pri);
7285 7286 qbackenable(q, pri);
7286 7287 mutex_exit(&stp->sd_lock);
7287 7288 return (EWOULDBLOCK);
7288 7289 }
7289 7290 if (bp == stp->sd_mark) {
7290 7291 mark |= _LASTMARK;
7291 7292 stp->sd_mark = NULL;
7292 7293 }
7293 7294 }
7294 7295
7295 7296 /*
7296 7297 * keep track of the first message type
7297 7298 */
7298 7299 type = bp->b_datap->db_type;
7299 7300
7300 7301 if (bp->b_datap->db_type == M_PASSFP) {
7301 7302 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7302 7303 stp->sd_mark = bp;
7303 7304 bp->b_flag |= mark & ~_LASTMARK;
7304 7305 putback(stp, q, bp, pri);
7305 7306 qbackenable(q, pri);
7306 7307 mutex_exit(&stp->sd_lock);
7307 7308 return (EBADMSG);
7308 7309 }
7309 7310 ASSERT(type != M_SIG);
7310 7311
7311 7312 if (flags & MSG_IPEEK) {
7312 7313 /*
7313 7314 * Clear any struioflag - we do the uiomove over again
7314 7315 * when peeking since it simplifies the code.
7315 7316 *
7316 7317 * Dup the message and put the original back on the queue.
7317 7318 * If dupmsg() fails, try again with copymsg() to see if
7318 7319 * there is indeed a shortage of memory. dupmsg() may fail
7319 7320 * if db_ref in any of the messages reaches its limit.
7320 7321 */
7321 7322
7322 7323 if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7323 7324 /*
7324 7325 * Restore the state of the stream head since we
7325 7326 * need to drop sd_lock (strwaitbuf is sleeping).
7326 7327 */
7327 7328 size_t size = msgdsize(bp);
7328 7329
7329 7330 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7330 7331 stp->sd_mark = bp;
7331 7332 bp->b_flag |= mark & ~_LASTMARK;
7332 7333 putback(stp, q, bp, pri);
7333 7334 mutex_exit(&stp->sd_lock);
7334 7335 error = strwaitbuf(size, BPRI_HI);
7335 7336 if (error) {
7336 7337 /*
7337 7338 * There is no net change to the queue thus
7338 7339 * no need to qbackenable.
7339 7340 */
7340 7341 return (error);
7341 7342 }
7342 7343 goto retry;
7343 7344 }
7344 7345
7345 7346 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7346 7347 stp->sd_mark = bp;
7347 7348 bp->b_flag |= mark & ~_LASTMARK;
7348 7349 putback(stp, q, bp, pri);
7349 7350 bp = nbp;
7350 7351 }
7351 7352
7352 7353 /*
7353 7354 * Set this flag so strrput will not generate signals. Need to
7354 7355 * make sure this flag is cleared before leaving this routine
7355 7356 * else signals will stop being sent.
7356 7357 */
7357 7358 stp->sd_flag |= STRGETINPROG;
7358 7359 mutex_exit(&stp->sd_lock);
7359 7360
7360 7361 if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7361 7362 mblk_t *tmp, *prevmp;
7362 7363
7363 7364 /*
7364 7365 * Put first non-data mblk back to stream head and
7365 7366 * cut the mblk chain so sd_rputdatafunc only sees
7366 7367 * M_DATA mblks. We can skip the first mblk since it
7367 7368 * is M_DATA according to the condition above.
7368 7369 */
7369 7370 for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7370 7371 prevmp = tmp, tmp = tmp->b_cont) {
7371 7372 if (DB_TYPE(tmp) != M_DATA) {
7372 7373 prevmp->b_cont = NULL;
7373 7374 mutex_enter(&stp->sd_lock);
7374 7375 putback(stp, q, tmp, tmp->b_band);
7375 7376 mutex_exit(&stp->sd_lock);
7376 7377 break;
7377 7378 }
7378 7379 }
7379 7380
7380 7381 bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7381 7382 NULL, NULL, NULL, NULL);
7382 7383
7383 7384 if (bp == NULL)
7384 7385 goto retry;
7385 7386 }
7386 7387
7387 7388 if (STREAM_NEEDSERVICE(stp))
7388 7389 stream_runservice(stp);
7389 7390
7390 7391 /*
7391 7392 * Set HIPRI flag if message is priority.
7392 7393 */
7393 7394 if (type >= QPCTL)
7394 7395 flg = MSG_HIPRI;
7395 7396 else
7396 7397 flg = MSG_BAND;
7397 7398
7398 7399 /*
7399 7400 * First process PROTO or PCPROTO blocks, if any.
7400 7401 */
7401 7402 if (mctlp != NULL && type != M_DATA) {
7402 7403 mblk_t *nbp;
7403 7404
7404 7405 *mctlp = bp;
7405 7406 while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7406 7407 bp = bp->b_cont;
7407 7408 nbp = bp->b_cont;
7408 7409 bp->b_cont = NULL;
7409 7410 bp = nbp;
7410 7411 }
7411 7412
7412 7413 if (bp && bp->b_datap->db_type != M_DATA) {
7413 7414 /*
7414 7415 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7415 7416 */
7416 7417 more |= MORECTL;
7417 7418 savemp = bp;
7418 7419 while (bp && bp->b_datap->db_type != M_DATA) {
7419 7420 savemptail = bp;
7420 7421 bp = bp->b_cont;
7421 7422 }
7422 7423 savemptail->b_cont = NULL;
7423 7424 }
7424 7425
7425 7426 /*
7426 7427 * Now process DATA blocks, if any.
7427 7428 */
7428 7429 if (uiop == NULL) {
7429 7430 /* Append data to tail of mctlp */
7430 7431
7431 7432 if (mctlp != NULL) {
7432 7433 mblk_t **mpp = mctlp;
7433 7434
7434 7435 while (*mpp != NULL)
7435 7436 mpp = &((*mpp)->b_cont);
7436 7437 *mpp = bp;
7437 7438 bp = NULL;
7438 7439 }
7439 7440 } else if (uiop->uio_resid >= 0 && bp) {
7440 7441 size_t oldresid = uiop->uio_resid;
7441 7442
7442 7443 /*
7443 7444 * If a streams message is likely to consist
7444 7445 * of many small mblks, it is pulled up into
7445 7446 * one continuous chunk of memory.
7446 7447 * The size of the first mblk may be bogus because
7447 7448 * successive read() calls on the socket reduce
7448 7449 * the size of this mblk until it is exhausted
7449 7450 * and then the code walks on to the next. Thus
7450 7451 * the size of the mblk may not be the original size
7451 7452 * that was passed up, it's simply a remainder
7452 7453 * and hence can be very small without any
7453 7454 * implication that the packet is badly fragmented.
7454 7455 * So the size of the possible second mblk is
7455 7456 * used to spot a badly fragmented packet.
7456 7457 * see longer comment at top of page
7457 7458 * by mblk_pull_len declaration.
7458 7459 */
7459 7460
7460 7461 if (bp->b_cont != NULL && MBLKL(bp->b_cont) < mblk_pull_len) {
7461 7462 (void) pullupmsg(bp, -1);
7462 7463 }
7463 7464
7464 7465 bp = struiocopyout(bp, uiop, &error);
7465 7466 if (error != 0) {
7466 7467 if (mctlp != NULL) {
7467 7468 freemsg(*mctlp);
7468 7469 *mctlp = NULL;
7469 7470 } else
7470 7471 freemsg(savemp);
7471 7472 mutex_enter(&stp->sd_lock);
7472 7473 /*
7473 7474 * clear stream head hi pri flag based on
7474 7475 * first message
7475 7476 */
7476 7477 if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7477 7478 ASSERT(type == M_PCPROTO);
7478 7479 stp->sd_flag &= ~STRPRI;
7479 7480 }
7480 7481 more = 0;
7481 7482 goto getmout;
7482 7483 }
7483 7484 /*
7484 7485 * (pr == 1) indicates a partial read.
7485 7486 */
7486 7487 if (oldresid > uiop->uio_resid)
7487 7488 pr = 1;
7488 7489 }
7489 7490
7490 7491 if (bp) { /* more data blocks in msg */
7491 7492 more |= MOREDATA;
7492 7493 if (savemp)
7493 7494 savemptail->b_cont = bp;
7494 7495 else
7495 7496 savemp = bp;
7496 7497 }
7497 7498
7498 7499 mutex_enter(&stp->sd_lock);
7499 7500 if (savemp) {
7500 7501 if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7501 7502 /*
7502 7503 * When MSG_DISCARDTAIL is set or
7503 7504 * when peeking discard any tail. When peeking this
7504 7505 * is the tail of the dup that was copied out - the
7505 7506 * message has already been putback on the queue.
7506 7507 * Return MOREDATA to the caller even though the data
7507 7508 * is discarded. This is used by sockets (to
7508 7509 * set MSG_TRUNC).
7509 7510 */
7510 7511 freemsg(savemp);
7511 7512 if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7512 7513 ASSERT(type == M_PCPROTO);
7513 7514 stp->sd_flag &= ~STRPRI;
7514 7515 }
7515 7516 } else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7516 7517 msgnodata(savemp)) {
7517 7518 /*
7518 7519 * Avoid queuing a zero-length tail part of
7519 7520 * a message. pr=1 indicates that we read some of
7520 7521 * the message.
7521 7522 */
7522 7523 freemsg(savemp);
7523 7524 more &= ~MOREDATA;
7524 7525 if (type >= QPCTL) {
7525 7526 ASSERT(type == M_PCPROTO);
7526 7527 stp->sd_flag &= ~STRPRI;
7527 7528 }
7528 7529 } else {
7529 7530 savemp->b_band = pri;
7530 7531 /*
7531 7532 * If the first message was HIPRI and the one we're
7532 7533 * putting back isn't, then clear STRPRI, otherwise
7533 7534 * set STRPRI again. Note that we must set STRPRI
7534 7535 * again since the flush logic in strrput_nondata()
7535 7536 * may have cleared it while we had sd_lock dropped.
7536 7537 */
7537 7538
7538 7539 if (type >= QPCTL) {
7539 7540 ASSERT(type == M_PCPROTO);
7540 7541 if (queclass(savemp) < QPCTL)
7541 7542 stp->sd_flag &= ~STRPRI;
7542 7543 else
7543 7544 stp->sd_flag |= STRPRI;
7544 7545 } else if (queclass(savemp) >= QPCTL) {
7545 7546 /*
7546 7547 * The first message was not a HIPRI message,
7547 7548 * but the one we are about to putback is.
7548 7549 * For simplicitly, we do not allow for HIPRI
7549 7550 * messages to be embedded in the message
7550 7551 * body, so just force it to same type as
7551 7552 * first message.
7552 7553 */
7553 7554 ASSERT(type == M_DATA || type == M_PROTO);
7554 7555 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7555 7556 savemp->b_datap->db_type = type;
7556 7557 }
7557 7558 if (mark != 0) {
7558 7559 if ((mark & _LASTMARK) &&
7559 7560 (stp->sd_mark == NULL)) {
7560 7561 /*
7561 7562 * If another marked message arrived
7562 7563 * while sd_lock was not held sd_mark
7563 7564 * would be non-NULL.
7564 7565 */
7565 7566 stp->sd_mark = savemp;
7566 7567 }
7567 7568 savemp->b_flag |= mark & ~_LASTMARK;
7568 7569 }
7569 7570 putback(stp, q, savemp, pri);
7570 7571 }
7571 7572 } else if (!(flags & MSG_IPEEK)) {
7572 7573 /*
7573 7574 * The complete message was consumed.
7574 7575 *
7575 7576 * If another M_PCPROTO arrived while sd_lock was not held
7576 7577 * it would have been discarded since STRPRI was still set.
7577 7578 *
7578 7579 * Move the MSG*MARKNEXT information
7579 7580 * to the stream head just in case
7580 7581 * the read queue becomes empty.
7581 7582 * clear stream head hi pri flag based on
7582 7583 * first message
7583 7584 *
7584 7585 * If the stream head was at the mark
7585 7586 * (STRATMARK) before we dropped sd_lock above
7586 7587 * and some data was consumed then we have
7587 7588 * moved past the mark thus STRATMARK is
7588 7589 * cleared. However, if a message arrived in
7589 7590 * strrput during the copyout above causing
7590 7591 * STRATMARK to be set we can not clear that
7591 7592 * flag.
7592 7593 * XXX A "perimeter" would help by single-threading strrput,
7593 7594 * strread, strgetmsg and kstrgetmsg.
7594 7595 */
7595 7596 if (type >= QPCTL) {
7596 7597 ASSERT(type == M_PCPROTO);
7597 7598 stp->sd_flag &= ~STRPRI;
7598 7599 }
7599 7600 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7600 7601 if (mark & MSGMARKNEXT) {
7601 7602 stp->sd_flag &= ~STRNOTATMARK;
7602 7603 stp->sd_flag |= STRATMARK;
7603 7604 } else if (mark & MSGNOTMARKNEXT) {
7604 7605 stp->sd_flag &= ~STRATMARK;
7605 7606 stp->sd_flag |= STRNOTATMARK;
7606 7607 } else {
7607 7608 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7608 7609 }
7609 7610 } else if (pr && (old_sd_flag & STRATMARK)) {
7610 7611 stp->sd_flag &= ~STRATMARK;
7611 7612 }
7612 7613 }
7613 7614
7614 7615 *flagsp = flg;
7615 7616 *prip = pri;
7616 7617
7617 7618 /*
7618 7619 * Getmsg cleanup processing - if the state of the queue has changed
7619 7620 * some signals may need to be sent and/or poll awakened.
7620 7621 */
7621 7622 getmout:
7622 7623 qbackenable(q, pri);
7623 7624
7624 7625 /*
7625 7626 * We dropped the stream head lock above. Send all M_SIG messages
7626 7627 * before processing stream head for SIGPOLL messages.
7627 7628 */
7628 7629 ASSERT(MUTEX_HELD(&stp->sd_lock));
7629 7630 while ((bp = q->q_first) != NULL &&
7630 7631 (bp->b_datap->db_type == M_SIG)) {
7631 7632 /*
7632 7633 * sd_lock is held so the content of the read queue can not
7633 7634 * change.
7634 7635 */
7635 7636 bp = getq(q);
7636 7637 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7637 7638
7638 7639 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7639 7640 mutex_exit(&stp->sd_lock);
7640 7641 freemsg(bp);
7641 7642 if (STREAM_NEEDSERVICE(stp))
7642 7643 stream_runservice(stp);
7643 7644 mutex_enter(&stp->sd_lock);
7644 7645 }
7645 7646
7646 7647 /*
7647 7648 * stream head cannot change while we make the determination
7648 7649 * whether or not to send a signal. Drop the flag to allow strrput
7649 7650 * to send firstmsgsigs again.
7650 7651 */
7651 7652 stp->sd_flag &= ~STRGETINPROG;
7652 7653
7653 7654 /*
7654 7655 * If the type of message at the front of the queue changed
7655 7656 * due to the receive the appropriate signals and pollwakeup events
7656 7657 * are generated. The type of changes are:
7657 7658 * Processed a hipri message, q_first is not hipri.
7658 7659 * Processed a band X message, and q_first is band Y.
7659 7660 * The generated signals and pollwakeups are identical to what
7660 7661 * strrput() generates should the message that is now on q_first
7661 7662 * arrive to an empty read queue.
7662 7663 *
7663 7664 * Note: only strrput will send a signal for a hipri message.
7664 7665 */
7665 7666 if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7666 7667 strsigset_t signals = 0;
7667 7668 strpollset_t pollwakeups = 0;
7668 7669
7669 7670 if (flg & MSG_HIPRI) {
7670 7671 /*
7671 7672 * Removed a hipri message. Regular data at
7672 7673 * the front of the queue.
7673 7674 */
7674 7675 if (bp->b_band == 0) {
7675 7676 signals = S_INPUT | S_RDNORM;
7676 7677 pollwakeups = POLLIN | POLLRDNORM;
7677 7678 } else {
7678 7679 signals = S_INPUT | S_RDBAND;
7679 7680 pollwakeups = POLLIN | POLLRDBAND;
7680 7681 }
7681 7682 } else if (pri != bp->b_band) {
7682 7683 /*
7683 7684 * The band is different for the new q_first.
7684 7685 */
7685 7686 if (bp->b_band == 0) {
7686 7687 signals = S_RDNORM;
7687 7688 pollwakeups = POLLIN | POLLRDNORM;
7688 7689 } else {
7689 7690 signals = S_RDBAND;
7690 7691 pollwakeups = POLLIN | POLLRDBAND;
7691 7692 }
7692 7693 }
7693 7694
7694 7695 if (pollwakeups != 0) {
7695 7696 if (pollwakeups == (POLLIN | POLLRDNORM)) {
7696 7697 if (!(stp->sd_rput_opt & SR_POLLIN))
7697 7698 goto no_pollwake;
7698 7699 stp->sd_rput_opt &= ~SR_POLLIN;
7699 7700 }
7700 7701 mutex_exit(&stp->sd_lock);
7701 7702 pollwakeup(&stp->sd_pollist, pollwakeups);
7702 7703 mutex_enter(&stp->sd_lock);
7703 7704 }
7704 7705 no_pollwake:
7705 7706
7706 7707 if (stp->sd_sigflags & signals)
7707 7708 strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7708 7709 }
7709 7710 mutex_exit(&stp->sd_lock);
7710 7711
7711 7712 rvp->r_val1 = more;
7712 7713 return (error);
7713 7714 #undef _LASTMARK
7714 7715 }
7715 7716
7716 7717 /*
7717 7718 * Put a message downstream.
7718 7719 *
7719 7720 * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7720 7721 */
7721 7722 int
7722 7723 strputmsg(
7723 7724 struct vnode *vp,
7724 7725 struct strbuf *mctl,
7725 7726 struct strbuf *mdata,
7726 7727 unsigned char pri,
7727 7728 int flag,
7728 7729 int fmode)
7729 7730 {
7730 7731 struct stdata *stp;
7731 7732 queue_t *wqp;
7732 7733 mblk_t *mp;
7733 7734 ssize_t msgsize;
7734 7735 ssize_t rmin, rmax;
7735 7736 int error;
7736 7737 struct uio uios;
7737 7738 struct uio *uiop = &uios;
7738 7739 struct iovec iovs;
7739 7740 int xpg4 = 0;
7740 7741
7741 7742 ASSERT(vp->v_stream);
7742 7743 stp = vp->v_stream;
7743 7744 wqp = stp->sd_wrq;
7744 7745
7745 7746 /*
7746 7747 * If it is an XPG4 application, we need to send
7747 7748 * SIGPIPE below
7748 7749 */
7749 7750
7750 7751 xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7751 7752 flag &= ~MSG_XPG4;
7752 7753
7753 7754 if (AU_AUDITING())
7754 7755 audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7755 7756
7756 7757 mutex_enter(&stp->sd_lock);
7757 7758
7758 7759 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7759 7760 mutex_exit(&stp->sd_lock);
7760 7761 return (error);
7761 7762 }
7762 7763
7763 7764 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7764 7765 error = strwriteable(stp, B_FALSE, xpg4);
7765 7766 if (error != 0) {
7766 7767 mutex_exit(&stp->sd_lock);
7767 7768 return (error);
7768 7769 }
7769 7770 }
7770 7771
7771 7772 mutex_exit(&stp->sd_lock);
7772 7773
7773 7774 /*
7774 7775 * Check for legal flag value.
7775 7776 */
7776 7777 switch (flag) {
7777 7778 case MSG_HIPRI:
7778 7779 if ((mctl->len < 0) || (pri != 0))
7779 7780 return (EINVAL);
7780 7781 break;
7781 7782 case MSG_BAND:
7782 7783 break;
7783 7784
7784 7785 default:
7785 7786 return (EINVAL);
7786 7787 }
7787 7788
7788 7789 TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7789 7790 "strputmsg in:stp %p", stp);
7790 7791
7791 7792 /* get these values from those cached in the stream head */
7792 7793 rmin = stp->sd_qn_minpsz;
7793 7794 rmax = stp->sd_qn_maxpsz;
7794 7795
7795 7796 /*
7796 7797 * Make sure ctl and data sizes together fall within the
7797 7798 * limits of the max and min receive packet sizes and do
7798 7799 * not exceed system limit.
7799 7800 */
7800 7801 ASSERT((rmax >= 0) || (rmax == INFPSZ));
7801 7802 if (rmax == 0) {
7802 7803 return (ERANGE);
7803 7804 }
7804 7805 /*
7805 7806 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7806 7807 * Needed to prevent partial failures in the strmakedata loop.
7807 7808 */
7808 7809 if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7809 7810 rmax = stp->sd_maxblk;
7810 7811
7811 7812 if ((msgsize = mdata->len) < 0) {
7812 7813 msgsize = 0;
7813 7814 rmin = 0; /* no range check for NULL data part */
7814 7815 }
7815 7816 if ((msgsize < rmin) ||
7816 7817 ((msgsize > rmax) && (rmax != INFPSZ)) ||
7817 7818 (mctl->len > strctlsz)) {
7818 7819 return (ERANGE);
7819 7820 }
7820 7821
7821 7822 /*
7822 7823 * Setup uio and iov for data part
7823 7824 */
7824 7825 iovs.iov_base = mdata->buf;
7825 7826 iovs.iov_len = msgsize;
7826 7827 uios.uio_iov = &iovs;
7827 7828 uios.uio_iovcnt = 1;
7828 7829 uios.uio_loffset = 0;
7829 7830 uios.uio_segflg = UIO_USERSPACE;
7830 7831 uios.uio_fmode = fmode;
7831 7832 uios.uio_extflg = UIO_COPY_DEFAULT;
7832 7833 uios.uio_resid = msgsize;
7833 7834 uios.uio_offset = 0;
7834 7835
7835 7836 /* Ignore flow control in strput for HIPRI */
7836 7837 if (flag & MSG_HIPRI)
7837 7838 flag |= MSG_IGNFLOW;
7838 7839
7839 7840 for (;;) {
7840 7841 int done = 0;
7841 7842
7842 7843 /*
7843 7844 * strput will always free the ctl mblk - even when strput
7844 7845 * fails.
7845 7846 */
7846 7847 if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7847 7848 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7848 7849 "strputmsg out:stp %p out %d error %d",
7849 7850 stp, 1, error);
7850 7851 return (error);
7851 7852 }
7852 7853 /*
7853 7854 * Verify that the whole message can be transferred by
7854 7855 * strput.
7855 7856 */
7856 7857 ASSERT(stp->sd_maxblk == INFPSZ ||
7857 7858 stp->sd_maxblk >= mdata->len);
7858 7859
7859 7860 msgsize = mdata->len;
7860 7861 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7861 7862 mdata->len = msgsize;
7862 7863
7863 7864 if (error == 0)
7864 7865 break;
7865 7866
7866 7867 if (error != EWOULDBLOCK)
7867 7868 goto out;
7868 7869
7869 7870 mutex_enter(&stp->sd_lock);
7870 7871 /*
7871 7872 * Check for a missed wakeup.
7872 7873 * Needed since strput did not hold sd_lock across
7873 7874 * the canputnext.
7874 7875 */
7875 7876 if (bcanputnext(wqp, pri)) {
7876 7877 /* Try again */
7877 7878 mutex_exit(&stp->sd_lock);
7878 7879 continue;
7879 7880 }
7880 7881 TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7881 7882 "strputmsg wait:stp %p waits pri %d", stp, pri);
7882 7883 if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7883 7884 &done)) != 0) || done) {
7884 7885 mutex_exit(&stp->sd_lock);
7885 7886 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7886 7887 "strputmsg out:q %p out %d error %d",
7887 7888 stp, 0, error);
7888 7889 return (error);
7889 7890 }
7890 7891 TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7891 7892 "strputmsg wake:stp %p wakes", stp);
7892 7893 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7893 7894 mutex_exit(&stp->sd_lock);
7894 7895 return (error);
7895 7896 }
7896 7897 mutex_exit(&stp->sd_lock);
7897 7898 }
7898 7899 out:
7899 7900 /*
7900 7901 * For historic reasons, applications expect EAGAIN
7901 7902 * when data mblk could not be allocated. so change
7902 7903 * ENOMEM back to EAGAIN
7903 7904 */
7904 7905 if (error == ENOMEM)
7905 7906 error = EAGAIN;
7906 7907 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7907 7908 "strputmsg out:stp %p out %d error %d", stp, 2, error);
7908 7909 return (error);
7909 7910 }
7910 7911
7911 7912 /*
7912 7913 * Put a message downstream.
7913 7914 * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7914 7915 * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7915 7916 * and the fmode parameter.
7916 7917 *
7917 7918 * This routine handles the consolidation private flags:
7918 7919 * MSG_IGNERROR Ignore any stream head error except STPLEX.
7919 7920 * MSG_HOLDSIG Hold signals while waiting for data.
7920 7921 * MSG_IGNFLOW Don't check streams flow control.
7921 7922 *
7922 7923 * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7923 7924 */
7924 7925 int
7925 7926 kstrputmsg(
7926 7927 struct vnode *vp,
7927 7928 mblk_t *mctl,
7928 7929 struct uio *uiop,
7929 7930 ssize_t msgsize,
7930 7931 unsigned char pri,
7931 7932 int flag,
7932 7933 int fmode)
7933 7934 {
7934 7935 struct stdata *stp;
7935 7936 queue_t *wqp;
7936 7937 ssize_t rmin, rmax;
7937 7938 int error;
7938 7939
7939 7940 ASSERT(vp->v_stream);
7940 7941 stp = vp->v_stream;
7941 7942 wqp = stp->sd_wrq;
7942 7943 if (AU_AUDITING())
7943 7944 audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7944 7945 if (mctl == NULL)
7945 7946 return (EINVAL);
7946 7947
7947 7948 mutex_enter(&stp->sd_lock);
7948 7949
7949 7950 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7950 7951 mutex_exit(&stp->sd_lock);
7951 7952 freemsg(mctl);
7952 7953 return (error);
7953 7954 }
7954 7955
7955 7956 if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7956 7957 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7957 7958 error = strwriteable(stp, B_FALSE, B_TRUE);
7958 7959 if (error != 0) {
7959 7960 mutex_exit(&stp->sd_lock);
7960 7961 freemsg(mctl);
7961 7962 return (error);
7962 7963 }
7963 7964 }
7964 7965 }
7965 7966
7966 7967 mutex_exit(&stp->sd_lock);
7967 7968
7968 7969 /*
7969 7970 * Check for legal flag value.
7970 7971 */
7971 7972 switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
7972 7973 case MSG_HIPRI:
7973 7974 if (pri != 0) {
7974 7975 freemsg(mctl);
7975 7976 return (EINVAL);
7976 7977 }
7977 7978 break;
7978 7979 case MSG_BAND:
7979 7980 break;
7980 7981 default:
7981 7982 freemsg(mctl);
7982 7983 return (EINVAL);
7983 7984 }
7984 7985
7985 7986 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
7986 7987 "kstrputmsg in:stp %p", stp);
7987 7988
7988 7989 /* get these values from those cached in the stream head */
7989 7990 rmin = stp->sd_qn_minpsz;
7990 7991 rmax = stp->sd_qn_maxpsz;
7991 7992
7992 7993 /*
7993 7994 * Make sure ctl and data sizes together fall within the
7994 7995 * limits of the max and min receive packet sizes and do
7995 7996 * not exceed system limit.
7996 7997 */
7997 7998 ASSERT((rmax >= 0) || (rmax == INFPSZ));
7998 7999 if (rmax == 0) {
7999 8000 freemsg(mctl);
8000 8001 return (ERANGE);
8001 8002 }
8002 8003 /*
8003 8004 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8004 8005 * Needed to prevent partial failures in the strmakedata loop.
8005 8006 */
8006 8007 if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8007 8008 rmax = stp->sd_maxblk;
8008 8009
8009 8010 if (uiop == NULL) {
8010 8011 msgsize = -1;
8011 8012 rmin = -1; /* no range check for NULL data part */
8012 8013 } else {
8013 8014 /* Use uio flags as well as the fmode parameter flags */
8014 8015 fmode |= uiop->uio_fmode;
8015 8016
8016 8017 if ((msgsize < rmin) ||
8017 8018 ((msgsize > rmax) && (rmax != INFPSZ))) {
8018 8019 freemsg(mctl);
8019 8020 return (ERANGE);
8020 8021 }
8021 8022 }
8022 8023
8023 8024 /* Ignore flow control in strput for HIPRI */
8024 8025 if (flag & MSG_HIPRI)
8025 8026 flag |= MSG_IGNFLOW;
8026 8027
8027 8028 for (;;) {
8028 8029 int done = 0;
8029 8030 int waitflag;
8030 8031 mblk_t *mp;
8031 8032
8032 8033 /*
8033 8034 * strput will always free the ctl mblk - even when strput
8034 8035 * fails. If MSG_IGNFLOW is set then any error returned
8035 8036 * will cause us to break the loop, so we don't need a copy
8036 8037 * of the message. If MSG_IGNFLOW is not set, then we can
8037 8038 * get hit by flow control and be forced to try again. In
8038 8039 * this case we need to have a copy of the message. We
8039 8040 * do this using copymsg since the message may get modified
8040 8041 * by something below us.
8041 8042 *
8042 8043 * We've observed that many TPI providers do not check db_ref
8043 8044 * on the control messages but blindly reuse them for the
8044 8045 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8045 8046 * friendly to such providers than using dupmsg. Also, note
8046 8047 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8047 8048 * Only data messages are subject to flow control, hence
8048 8049 * subject to this copymsg.
8049 8050 */
8050 8051 if (flag & MSG_IGNFLOW) {
8051 8052 mp = mctl;
8052 8053 mctl = NULL;
8053 8054 } else {
8054 8055 do {
8055 8056 /*
8056 8057 * If a message has a free pointer, the message
8057 8058 * must be dupmsg to maintain this pointer.
8058 8059 * Code using this facility must be sure
8059 8060 * that modules below will not change the
8060 8061 * contents of the dblk without checking db_ref
8061 8062 * first. If db_ref is > 1, then the module
8062 8063 * needs to do a copymsg first. Otherwise,
8063 8064 * the contents of the dblk may become
8064 8065 * inconsistent because the freesmg/freeb below
8065 8066 * may end up calling atomic_add_32_nv.
8066 8067 * The atomic_add_32_nv in freeb (accessing
8067 8068 * all of db_ref, db_type, db_flags, and
8068 8069 * db_struioflag) does not prevent other threads
8069 8070 * from concurrently trying to modify e.g.
8070 8071 * db_type.
8071 8072 */
8072 8073 if (mctl->b_datap->db_frtnp != NULL)
8073 8074 mp = dupmsg(mctl);
8074 8075 else
8075 8076 mp = copymsg(mctl);
8076 8077
8077 8078 if (mp != NULL)
8078 8079 break;
8079 8080
8080 8081 error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8081 8082 if (error) {
8082 8083 freemsg(mctl);
8083 8084 return (error);
8084 8085 }
8085 8086 } while (mp == NULL);
8086 8087 }
8087 8088 /*
8088 8089 * Verify that all of msgsize can be transferred by
8089 8090 * strput.
8090 8091 */
8091 8092 ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8092 8093 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8093 8094 if (error == 0)
8094 8095 break;
8095 8096
8096 8097 if (error != EWOULDBLOCK)
8097 8098 goto out;
8098 8099
8099 8100 /*
8100 8101 * IF MSG_IGNFLOW is set we should have broken out of loop
8101 8102 * above.
8102 8103 */
8103 8104 ASSERT(!(flag & MSG_IGNFLOW));
8104 8105 mutex_enter(&stp->sd_lock);
8105 8106 /*
8106 8107 * Check for a missed wakeup.
8107 8108 * Needed since strput did not hold sd_lock across
8108 8109 * the canputnext.
8109 8110 */
8110 8111 if (bcanputnext(wqp, pri)) {
8111 8112 /* Try again */
8112 8113 mutex_exit(&stp->sd_lock);
8113 8114 continue;
8114 8115 }
8115 8116 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8116 8117 "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8117 8118
8118 8119 waitflag = WRITEWAIT;
8119 8120 if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8120 8121 if (flag & MSG_HOLDSIG)
8121 8122 waitflag |= STR_NOSIG;
8122 8123 if (flag & MSG_IGNERROR)
8123 8124 waitflag |= STR_NOERROR;
8124 8125 }
8125 8126 if (((error = strwaitq(stp, waitflag,
8126 8127 (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8127 8128 mutex_exit(&stp->sd_lock);
8128 8129 TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8129 8130 "kstrputmsg out:stp %p out %d error %d",
8130 8131 stp, 0, error);
8131 8132 freemsg(mctl);
8132 8133 return (error);
8133 8134 }
8134 8135 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8135 8136 "kstrputmsg wake:stp %p wakes", stp);
8136 8137 if ((error = i_straccess(stp, JCWRITE)) != 0) {
8137 8138 mutex_exit(&stp->sd_lock);
8138 8139 freemsg(mctl);
8139 8140 return (error);
8140 8141 }
8141 8142 mutex_exit(&stp->sd_lock);
8142 8143 }
8143 8144 out:
8144 8145 freemsg(mctl);
8145 8146 /*
8146 8147 * For historic reasons, applications expect EAGAIN
8147 8148 * when data mblk could not be allocated. so change
8148 8149 * ENOMEM back to EAGAIN
8149 8150 */
8150 8151 if (error == ENOMEM)
8151 8152 error = EAGAIN;
8152 8153 TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8153 8154 "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8154 8155 return (error);
8155 8156 }
8156 8157
8157 8158 /*
8158 8159 * Determines whether the necessary conditions are set on a stream
8159 8160 * for it to be readable, writeable, or have exceptions.
8160 8161 *
8161 8162 * strpoll handles the consolidation private events:
8162 8163 * POLLNOERR Do not return POLLERR even if there are stream
8163 8164 * head errors.
8164 8165 * Used by sockfs.
8165 8166 * POLLRDDATA Do not return POLLIN unless at least one message on
8166 8167 * the queue contains one or more M_DATA mblks. Thus
8167 8168 * when this flag is set a queue with only
8168 8169 * M_PROTO/M_PCPROTO mblks does not return POLLIN.
8169 8170 * Used by sockfs to ignore T_EXDATA_IND messages.
8170 8171 *
8171 8172 * Note: POLLRDDATA assumes that synch streams only return messages with
8172 8173 * an M_DATA attached (i.e. not messages consisting of only
8173 8174 * an M_PROTO/M_PCPROTO part).
8174 8175 */
8175 8176 int
8176 8177 strpoll(
8177 8178 struct stdata *stp,
8178 8179 short events_arg,
8179 8180 int anyyet,
8180 8181 short *reventsp,
8181 8182 struct pollhead **phpp)
8182 8183 {
8183 8184 int events = (ushort_t)events_arg;
8184 8185 int retevents = 0;
8185 8186 mblk_t *mp;
8186 8187 qband_t *qbp;
8187 8188 long sd_flags = stp->sd_flag;
8188 8189 int headlocked = 0;
8189 8190
8190 8191 /*
8191 8192 * For performance, a single 'if' tests for most possible edge
8192 8193 * conditions in one shot
8193 8194 */
8194 8195 if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8195 8196 if (sd_flags & STPLEX) {
8196 8197 *reventsp = POLLNVAL;
8197 8198 return (EINVAL);
8198 8199 }
8199 8200 if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8200 8201 (sd_flags & STRDERR)) ||
8201 8202 ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8202 8203 (sd_flags & STWRERR))) {
8203 8204 if (!(events & POLLNOERR)) {
8204 8205 *reventsp = POLLERR;
8205 8206 return (0);
8206 8207 }
8207 8208 }
8208 8209 }
8209 8210 if (sd_flags & STRHUP) {
8210 8211 retevents |= POLLHUP;
8211 8212 } else if (events & (POLLWRNORM | POLLWRBAND)) {
8212 8213 queue_t *tq;
8213 8214 queue_t *qp = stp->sd_wrq;
8214 8215
8215 8216 claimstr(qp);
8216 8217 /* Find next module forward that has a service procedure */
8217 8218 tq = qp->q_next->q_nfsrv;
8218 8219 ASSERT(tq != NULL);
8219 8220
8220 8221 polllock(&stp->sd_pollist, QLOCK(tq));
8221 8222 if (events & POLLWRNORM) {
8222 8223 queue_t *sqp;
8223 8224
8224 8225 if (tq->q_flag & QFULL)
8225 8226 /* ensure backq svc procedure runs */
8226 8227 tq->q_flag |= QWANTW;
8227 8228 else if ((sqp = stp->sd_struiowrq) != NULL) {
8228 8229 /* Check sync stream barrier write q */
8229 8230 mutex_exit(QLOCK(tq));
8230 8231 polllock(&stp->sd_pollist, QLOCK(sqp));
8231 8232 if (sqp->q_flag & QFULL)
8232 8233 /* ensure pollwakeup() is done */
8233 8234 sqp->q_flag |= QWANTWSYNC;
8234 8235 else
8235 8236 retevents |= POLLOUT;
8236 8237 /* More write events to process ??? */
8237 8238 if (! (events & POLLWRBAND)) {
8238 8239 mutex_exit(QLOCK(sqp));
8239 8240 releasestr(qp);
8240 8241 goto chkrd;
8241 8242 }
8242 8243 mutex_exit(QLOCK(sqp));
8243 8244 polllock(&stp->sd_pollist, QLOCK(tq));
8244 8245 } else
8245 8246 retevents |= POLLOUT;
8246 8247 }
8247 8248 if (events & POLLWRBAND) {
8248 8249 qbp = tq->q_bandp;
8249 8250 if (qbp) {
8250 8251 while (qbp) {
8251 8252 if (qbp->qb_flag & QB_FULL)
8252 8253 qbp->qb_flag |= QB_WANTW;
8253 8254 else
8254 8255 retevents |= POLLWRBAND;
8255 8256 qbp = qbp->qb_next;
8256 8257 }
8257 8258 } else {
8258 8259 retevents |= POLLWRBAND;
8259 8260 }
8260 8261 }
8261 8262 mutex_exit(QLOCK(tq));
8262 8263 releasestr(qp);
8263 8264 }
8264 8265 chkrd:
8265 8266 if (sd_flags & STRPRI) {
8266 8267 retevents |= (events & POLLPRI);
8267 8268 } else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8268 8269 queue_t *qp = _RD(stp->sd_wrq);
8269 8270 int normevents = (events & (POLLIN | POLLRDNORM));
8270 8271
8271 8272 /*
8272 8273 * Note: Need to do polllock() here since ps_lock may be
8273 8274 * held. See bug 4191544.
8274 8275 */
8275 8276 polllock(&stp->sd_pollist, &stp->sd_lock);
8276 8277 headlocked = 1;
8277 8278 mp = qp->q_first;
8278 8279 while (mp) {
8279 8280 /*
8280 8281 * For POLLRDDATA we scan b_cont and b_next until we
8281 8282 * find an M_DATA.
8282 8283 */
8283 8284 if ((events & POLLRDDATA) &&
8284 8285 mp->b_datap->db_type != M_DATA) {
8285 8286 mblk_t *nmp = mp->b_cont;
8286 8287
8287 8288 while (nmp != NULL &&
8288 8289 nmp->b_datap->db_type != M_DATA)
8289 8290 nmp = nmp->b_cont;
8290 8291 if (nmp == NULL) {
8291 8292 mp = mp->b_next;
8292 8293 continue;
8293 8294 }
8294 8295 }
8295 8296 if (mp->b_band == 0)
8296 8297 retevents |= normevents;
8297 8298 else
8298 8299 retevents |= (events & (POLLIN | POLLRDBAND));
8299 8300 break;
8300 8301 }
8301 8302 if (! (retevents & normevents) &&
8302 8303 (stp->sd_wakeq & RSLEEP)) {
8303 8304 /*
8304 8305 * Sync stream barrier read queue has data.
8305 8306 */
8306 8307 retevents |= normevents;
8307 8308 }
8308 8309 /* Treat eof as normal data */
8309 8310 if (sd_flags & STREOF)
8310 8311 retevents |= normevents;
8311 8312 }
8312 8313
8313 8314 *reventsp = (short)retevents;
8314 8315 if (retevents) {
8315 8316 if (headlocked)
8316 8317 mutex_exit(&stp->sd_lock);
8317 8318 return (0);
8318 8319 }
8319 8320
8320 8321 /*
8321 8322 * If poll() has not found any events yet, set up event cell
8322 8323 * to wake up the poll if a requested event occurs on this
8323 8324 * stream. Check for collisions with outstanding poll requests.
8324 8325 */
8325 8326 if (!anyyet) {
8326 8327 *phpp = &stp->sd_pollist;
8327 8328 if (headlocked == 0) {
8328 8329 polllock(&stp->sd_pollist, &stp->sd_lock);
8329 8330 headlocked = 1;
8330 8331 }
8331 8332 stp->sd_rput_opt |= SR_POLLIN;
8332 8333 }
8333 8334 if (headlocked)
8334 8335 mutex_exit(&stp->sd_lock);
8335 8336 return (0);
8336 8337 }
8337 8338
8338 8339 /*
8339 8340 * The purpose of putback() is to assure sleeping polls/reads
8340 8341 * are awakened when there are no new messages arriving at the,
8341 8342 * stream head, and a message is placed back on the read queue.
8342 8343 *
8343 8344 * sd_lock must be held when messages are placed back on stream
8344 8345 * head. (getq() holds sd_lock when it removes messages from
8345 8346 * the queue)
8346 8347 */
8347 8348
8348 8349 static void
8349 8350 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8350 8351 {
8351 8352 mblk_t *qfirst;
8352 8353 ASSERT(MUTEX_HELD(&stp->sd_lock));
8353 8354
8354 8355 /*
8355 8356 * As a result of lock-step ordering around q_lock and sd_lock,
8356 8357 * it's possible for function calls like putnext() and
8357 8358 * canputnext() to get an inaccurate picture of how much
8358 8359 * data is really being processed at the stream head.
8359 8360 * We only consolidate with existing messages on the queue
8360 8361 * if the length of the message we want to put back is smaller
8361 8362 * than the queue hiwater mark.
8362 8363 */
8363 8364 if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8364 8365 (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8365 8366 (DB_TYPE(qfirst) == M_DATA) &&
8366 8367 ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8367 8368 ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8368 8369 (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8369 8370 /*
8370 8371 * We use the same logic as defined in strrput()
8371 8372 * but in reverse as we are putting back onto the
8372 8373 * queue and want to retain byte ordering.
8373 8374 * Consolidate M_DATA messages with M_DATA ONLY.
8374 8375 * strrput() allows the consolidation of M_DATA onto
8375 8376 * M_PROTO | M_PCPROTO but not the other way round.
8376 8377 *
8377 8378 * The consolidation does not take place if the message
8378 8379 * we are returning to the queue is marked with either
8379 8380 * of the marks or the delim flag or if q_first
8380 8381 * is marked with MSGMARK. The MSGMARK check is needed to
8381 8382 * handle the odd semantics of MSGMARK where essentially
8382 8383 * the whole message is to be treated as marked.
8383 8384 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8384 8385 * to the front of the b_cont chain.
8385 8386 */
8386 8387 rmvq_noenab(q, qfirst);
8387 8388
8388 8389 /*
8389 8390 * The first message in the b_cont list
8390 8391 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8391 8392 * We need to handle the case where we
8392 8393 * are appending:
8393 8394 *
8394 8395 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8395 8396 * 2) a MSGMARKNEXT to a plain message.
8396 8397 * 3) a MSGNOTMARKNEXT to a plain message
8397 8398 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8398 8399 * message.
8399 8400 *
8400 8401 * Thus we never append a MSGMARKNEXT or
8401 8402 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8402 8403 */
8403 8404 if (qfirst->b_flag & MSGMARKNEXT) {
8404 8405 bp->b_flag |= MSGMARKNEXT;
8405 8406 bp->b_flag &= ~MSGNOTMARKNEXT;
8406 8407 qfirst->b_flag &= ~MSGMARKNEXT;
8407 8408 } else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8408 8409 bp->b_flag |= MSGNOTMARKNEXT;
8409 8410 qfirst->b_flag &= ~MSGNOTMARKNEXT;
8410 8411 }
8411 8412
8412 8413 linkb(bp, qfirst);
8413 8414 }
8414 8415 (void) putbq(q, bp);
8415 8416
8416 8417 /*
8417 8418 * A message may have come in when the sd_lock was dropped in the
8418 8419 * calling routine. If this is the case and STR*ATMARK info was
8419 8420 * received, need to move that from the stream head to the q_last
8420 8421 * so that SIOCATMARK can return the proper value.
8421 8422 */
8422 8423 if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8423 8424 unsigned short *flagp = &q->q_last->b_flag;
8424 8425 uint_t b_flag = (uint_t)*flagp;
8425 8426
8426 8427 if (stp->sd_flag & STRATMARK) {
8427 8428 b_flag &= ~MSGNOTMARKNEXT;
8428 8429 b_flag |= MSGMARKNEXT;
8429 8430 stp->sd_flag &= ~STRATMARK;
8430 8431 } else {
8431 8432 b_flag &= ~MSGMARKNEXT;
8432 8433 b_flag |= MSGNOTMARKNEXT;
8433 8434 stp->sd_flag &= ~STRNOTATMARK;
8434 8435 }
8435 8436 *flagp = (unsigned short) b_flag;
8436 8437 }
8437 8438
8438 8439 #ifdef DEBUG
8439 8440 /*
8440 8441 * Make sure that the flags are not messed up.
8441 8442 */
8442 8443 {
8443 8444 mblk_t *mp;
8444 8445 mp = q->q_last;
8445 8446 while (mp != NULL) {
8446 8447 ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8447 8448 (MSGMARKNEXT|MSGNOTMARKNEXT));
8448 8449 mp = mp->b_cont;
8449 8450 }
8450 8451 }
8451 8452 #endif
8452 8453 if (q->q_first == bp) {
8453 8454 short pollevents;
8454 8455
8455 8456 if (stp->sd_flag & RSLEEP) {
8456 8457 stp->sd_flag &= ~RSLEEP;
8457 8458 cv_broadcast(&q->q_wait);
8458 8459 }
8459 8460 if (stp->sd_flag & STRPRI) {
8460 8461 pollevents = POLLPRI;
8461 8462 } else {
8462 8463 if (band == 0) {
8463 8464 if (!(stp->sd_rput_opt & SR_POLLIN))
8464 8465 return;
8465 8466 stp->sd_rput_opt &= ~SR_POLLIN;
8466 8467 pollevents = POLLIN | POLLRDNORM;
8467 8468 } else {
8468 8469 pollevents = POLLIN | POLLRDBAND;
8469 8470 }
8470 8471 }
8471 8472 mutex_exit(&stp->sd_lock);
8472 8473 pollwakeup(&stp->sd_pollist, pollevents);
8473 8474 mutex_enter(&stp->sd_lock);
8474 8475 }
8475 8476 }
8476 8477
8477 8478 /*
8478 8479 * Return the held vnode attached to the stream head of a
8479 8480 * given queue
8480 8481 * It is the responsibility of the calling routine to ensure
8481 8482 * that the queue does not go away (e.g. pop).
8482 8483 */
8483 8484 vnode_t *
8484 8485 strq2vp(queue_t *qp)
8485 8486 {
8486 8487 vnode_t *vp;
8487 8488 vp = STREAM(qp)->sd_vnode;
8488 8489 ASSERT(vp != NULL);
8489 8490 VN_HOLD(vp);
8490 8491 return (vp);
8491 8492 }
8492 8493
8493 8494 /*
8494 8495 * return the stream head write queue for the given vp
8495 8496 * It is the responsibility of the calling routine to ensure
8496 8497 * that the stream or vnode do not close.
8497 8498 */
8498 8499 queue_t *
8499 8500 strvp2wq(vnode_t *vp)
8500 8501 {
8501 8502 ASSERT(vp->v_stream != NULL);
8502 8503 return (vp->v_stream->sd_wrq);
8503 8504 }
8504 8505
8505 8506 /*
8506 8507 * pollwakeup stream head
8507 8508 * It is the responsibility of the calling routine to ensure
8508 8509 * that the stream or vnode do not close.
8509 8510 */
8510 8511 void
8511 8512 strpollwakeup(vnode_t *vp, short event)
8512 8513 {
8513 8514 ASSERT(vp->v_stream);
8514 8515 pollwakeup(&vp->v_stream->sd_pollist, event);
8515 8516 }
8516 8517
8517 8518 /*
8518 8519 * Mate the stream heads of two vnodes together. If the two vnodes are the
8519 8520 * same, we just make the write-side point at the read-side -- otherwise,
8520 8521 * we do a full mate. Only works on vnodes associated with streams that are
8521 8522 * still being built and thus have only a stream head.
8522 8523 */
8523 8524 void
8524 8525 strmate(vnode_t *vp1, vnode_t *vp2)
8525 8526 {
8526 8527 queue_t *wrq1 = strvp2wq(vp1);
8527 8528 queue_t *wrq2 = strvp2wq(vp2);
8528 8529
8529 8530 /*
8530 8531 * Verify that there are no modules on the stream yet. We also
8531 8532 * rely on the stream head always having a service procedure to
8532 8533 * avoid tweaking q_nfsrv.
8533 8534 */
8534 8535 ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8535 8536 ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8536 8537 ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8537 8538
8538 8539 /*
8539 8540 * If the queues are the same, just twist; otherwise do a full mate.
8540 8541 */
8541 8542 if (wrq1 == wrq2) {
8542 8543 wrq1->q_next = _RD(wrq1);
8543 8544 } else {
8544 8545 wrq1->q_next = _RD(wrq2);
8545 8546 wrq2->q_next = _RD(wrq1);
8546 8547 STREAM(wrq1)->sd_mate = STREAM(wrq2);
8547 8548 STREAM(wrq1)->sd_flag |= STRMATE;
8548 8549 STREAM(wrq2)->sd_mate = STREAM(wrq1);
8549 8550 STREAM(wrq2)->sd_flag |= STRMATE;
8550 8551 }
8551 8552 }
8552 8553
8553 8554 /*
8554 8555 * XXX will go away when console is correctly fixed.
8555 8556 * Clean up the console PIDS, from previous I_SETSIG,
8556 8557 * called only for cnopen which never calls strclean().
8557 8558 */
8558 8559 void
8559 8560 str_cn_clean(struct vnode *vp)
8560 8561 {
8561 8562 strsig_t *ssp, *pssp, *tssp;
8562 8563 struct stdata *stp;
8563 8564 struct pid *pidp;
8564 8565 int update = 0;
8565 8566
8566 8567 ASSERT(vp->v_stream);
8567 8568 stp = vp->v_stream;
8568 8569 pssp = NULL;
8569 8570 mutex_enter(&stp->sd_lock);
8570 8571 ssp = stp->sd_siglist;
8571 8572 while (ssp) {
8572 8573 mutex_enter(&pidlock);
8573 8574 pidp = ssp->ss_pidp;
8574 8575 /*
8575 8576 * Get rid of PID if the proc is gone.
8576 8577 */
8577 8578 if (pidp->pid_prinactive) {
8578 8579 tssp = ssp->ss_next;
8579 8580 if (pssp)
8580 8581 pssp->ss_next = tssp;
8581 8582 else
8582 8583 stp->sd_siglist = tssp;
8583 8584 ASSERT(pidp->pid_ref <= 1);
8584 8585 PID_RELE(ssp->ss_pidp);
8585 8586 mutex_exit(&pidlock);
8586 8587 kmem_free(ssp, sizeof (strsig_t));
8587 8588 update = 1;
8588 8589 ssp = tssp;
8589 8590 continue;
8590 8591 } else
8591 8592 mutex_exit(&pidlock);
8592 8593 pssp = ssp;
8593 8594 ssp = ssp->ss_next;
8594 8595 }
8595 8596 if (update) {
8596 8597 stp->sd_sigflags = 0;
8597 8598 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8598 8599 stp->sd_sigflags |= ssp->ss_events;
8599 8600 }
8600 8601 mutex_exit(&stp->sd_lock);
8601 8602 }
8602 8603
8603 8604 /*
8604 8605 * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8605 8606 */
8606 8607 static boolean_t
8607 8608 msghasdata(mblk_t *bp)
8608 8609 {
8609 8610 for (; bp; bp = bp->b_cont)
8610 8611 if (bp->b_datap->db_type == M_DATA) {
8611 8612 ASSERT(bp->b_wptr >= bp->b_rptr);
8612 8613 if (bp->b_wptr > bp->b_rptr)
8613 8614 return (B_TRUE);
8614 8615 }
8615 8616 return (B_FALSE);
8616 8617 }
|
↓ open down ↓ |
5427 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX