Print this page
XXXX adding PID information to netstat output
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>
↓ open down ↓ |
69 lines elided |
↑ open up ↑ |
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 +#include <sys/fcntl.h>
80 81
81 82 /*
82 83 * This define helps improve the readability of streams code while
83 84 * still maintaining a very old streams performance enhancement. The
84 85 * performance enhancement basically involved having all callers
85 86 * of straccess() perform the first check that straccess() will do
86 87 * locally before actually calling straccess(). (There by reducing
87 88 * the number of unnecessary calls to straccess().)
88 89 */
89 90 #define i_straccess(x, y) ((stp->sd_sidp == NULL) ? 0 : \
90 91 (stp->sd_vnode->v_type == VFIFO) ? 0 : \
91 92 straccess((x), (y)))
92 93
93 94 /*
94 95 * what is mblk_pull_len?
95 96 *
96 97 * If a streams message consists of many short messages,
97 98 * a performance degradation occurs from copyout overhead.
98 99 * To decrease the per mblk overhead, messages that are
99 100 * likely to consist of many small mblks are pulled up into
100 101 * one continuous chunk of memory.
101 102 *
102 103 * To avoid the processing overhead of examining every
103 104 * mblk, a quick heuristic is used. If the first mblk in
104 105 * the message is shorter than mblk_pull_len, it is likely
105 106 * that the rest of the mblk will be short.
106 107 *
107 108 * This heuristic was decided upon after performance tests
108 109 * indicated that anything more complex slowed down the main
109 110 * code path.
110 111 */
111 112 #define MBLK_PULL_LEN 64
112 113 uint32_t mblk_pull_len = MBLK_PULL_LEN;
113 114
114 115 /*
115 116 * The sgttyb_handling flag controls the handling of the old BSD
116 117 * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
117 118 *
118 119 * 0 - Emit no warnings at all and retain old, broken behavior.
119 120 * 1 - Emit no warnings and silently handle new semantics.
120 121 * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
121 122 * (once per system invocation). Handle with new semantics.
122 123 * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
123 124 * made (so that offenders drop core and are easy to debug).
124 125 *
125 126 * The "new semantics" are that TIOCGETP returns B38400 for
126 127 * sg_[io]speed if the corresponding value is over B38400, and that
127 128 * TIOCSET[PN] accept B38400 in these cases to mean "retain current
128 129 * bit rate."
129 130 */
130 131 int sgttyb_handling = 1;
131 132 static boolean_t sgttyb_complaint;
132 133
133 134 /* don't push drcompat module by default on Style-2 streams */
134 135 static int push_drcompat = 0;
↓ open down ↓ |
45 lines elided |
↑ open up ↑ |
135 136
136 137 /*
137 138 * id value used to distinguish between different ioctl messages
138 139 */
139 140 static uint32_t ioc_id;
140 141
141 142 static void putback(struct stdata *, queue_t *, mblk_t *, int);
142 143 static void strcleanall(struct vnode *);
143 144 static int strwsrv(queue_t *);
144 145 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
146 +static boolean_t is_xti_str(const struct stdata *);
145 147
146 148 /*
147 149 * qinit and module_info structures for stream head read and write queues
148 150 */
149 151 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
150 152 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
151 153 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
152 154 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
153 155 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
154 156 FIFOLOWAT };
155 157 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
156 158 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
157 159 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
158 160
159 161 extern kmutex_t strresources; /* protects global resources */
160 162 extern kmutex_t muxifier; /* single-threads multiplexor creation */
161 163
162 164 static boolean_t msghasdata(mblk_t *bp);
163 165 #define msgnodata(bp) (!msghasdata(bp))
164 166
165 167 /*
166 168 * Stream head locking notes:
167 169 * There are four monitors associated with the stream head:
168 170 * 1. v_stream monitor: in stropen() and strclose() v_lock
169 171 * is held while the association of vnode and stream
170 172 * head is established or tested for.
171 173 * 2. open/close/push/pop monitor: sd_lock is held while each
172 174 * thread bids for exclusive access to this monitor
173 175 * for opening or closing a stream. In addition, this
174 176 * monitor is entered during pushes and pops. This
175 177 * guarantees that during plumbing operations there
176 178 * is only one thread trying to change the plumbing.
177 179 * Any other threads present in the stream are only
178 180 * using the plumbing.
179 181 * 3. read/write monitor: in the case of read, a thread holds
180 182 * sd_lock while trying to get data from the stream
181 183 * head queue. if there is none to fulfill a read
182 184 * request, it sets RSLEEP and calls cv_wait_sig() down
183 185 * in strwaitq() to await the arrival of new data.
184 186 * when new data arrives in strrput(), sd_lock is acquired
185 187 * before testing for RSLEEP and calling cv_broadcast().
186 188 * the behavior of strwrite(), strwsrv(), and WSLEEP
187 189 * mirror this.
188 190 * 4. ioctl monitor: sd_lock is gotten to ensure that only one
189 191 * thread is doing an ioctl at a time.
190 192 */
191 193
192 194 static int
193 195 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
194 196 int anchor, cred_t *crp, uint_t anchor_zoneid)
195 197 {
196 198 int error;
197 199 fmodsw_impl_t *fp;
198 200
199 201 if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
200 202 error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
201 203 return (error);
202 204 }
203 205 if (stp->sd_pushcnt >= nstrpush) {
204 206 return (EINVAL);
205 207 }
206 208
207 209 if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
208 210 stp->sd_flag |= STREOPENFAIL;
209 211 return (EINVAL);
210 212 }
211 213
212 214 /*
213 215 * push new module and call its open routine via qattach
214 216 */
215 217 if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
216 218 return (error);
217 219
218 220 /*
219 221 * Check to see if caller wants a STREAMS anchor
220 222 * put at this place in the stream, and add if so.
221 223 */
222 224 mutex_enter(&stp->sd_lock);
223 225 if (anchor == stp->sd_pushcnt) {
224 226 stp->sd_anchor = stp->sd_pushcnt;
225 227 stp->sd_anchorzone = anchor_zoneid;
226 228 }
227 229 mutex_exit(&stp->sd_lock);
228 230
229 231 return (0);
230 232 }
231 233
232 234 /*
233 235 * Open a stream device.
234 236 */
235 237 int
236 238 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
237 239 {
238 240 struct stdata *stp;
239 241 queue_t *qp;
240 242 int s;
241 243 dev_t dummydev, savedev;
242 244 struct autopush *ap;
243 245 struct dlautopush dlap;
244 246 int error = 0;
245 247 ssize_t rmin, rmax;
246 248 int cloneopen;
247 249 queue_t *brq;
248 250 major_t major;
249 251 str_stack_t *ss;
250 252 zoneid_t zoneid;
251 253 uint_t anchor;
252 254
253 255 /*
254 256 * If the stream already exists, wait for any open in progress
255 257 * to complete, then call the open function of each module and
256 258 * driver in the stream. Otherwise create the stream.
257 259 */
258 260 TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
259 261 retry:
260 262 mutex_enter(&vp->v_lock);
261 263 if ((stp = vp->v_stream) != NULL) {
262 264
263 265 /*
264 266 * Waiting for stream to be created to device
265 267 * due to another open.
266 268 */
267 269 mutex_exit(&vp->v_lock);
268 270
269 271 if (STRMATED(stp)) {
270 272 struct stdata *strmatep = stp->sd_mate;
271 273
272 274 STRLOCKMATES(stp);
273 275 if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
274 276 if (flag & (FNDELAY|FNONBLOCK)) {
275 277 error = EAGAIN;
276 278 mutex_exit(&strmatep->sd_lock);
277 279 goto ckreturn;
278 280 }
279 281 mutex_exit(&stp->sd_lock);
280 282 if (!cv_wait_sig(&strmatep->sd_monitor,
281 283 &strmatep->sd_lock)) {
282 284 error = EINTR;
283 285 mutex_exit(&strmatep->sd_lock);
284 286 mutex_enter(&stp->sd_lock);
285 287 goto ckreturn;
286 288 }
287 289 mutex_exit(&strmatep->sd_lock);
288 290 goto retry;
289 291 }
290 292 if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
291 293 if (flag & (FNDELAY|FNONBLOCK)) {
292 294 error = EAGAIN;
293 295 mutex_exit(&strmatep->sd_lock);
294 296 goto ckreturn;
295 297 }
296 298 mutex_exit(&strmatep->sd_lock);
297 299 if (!cv_wait_sig(&stp->sd_monitor,
298 300 &stp->sd_lock)) {
299 301 error = EINTR;
300 302 goto ckreturn;
301 303 }
302 304 mutex_exit(&stp->sd_lock);
303 305 goto retry;
304 306 }
305 307
306 308 if (stp->sd_flag & (STRDERR|STWRERR)) {
307 309 error = EIO;
308 310 mutex_exit(&strmatep->sd_lock);
309 311 goto ckreturn;
310 312 }
311 313
312 314 stp->sd_flag |= STWOPEN;
313 315 STRUNLOCKMATES(stp);
314 316 } else {
315 317 mutex_enter(&stp->sd_lock);
316 318 if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
317 319 if (flag & (FNDELAY|FNONBLOCK)) {
318 320 error = EAGAIN;
319 321 goto ckreturn;
320 322 }
321 323 if (!cv_wait_sig(&stp->sd_monitor,
322 324 &stp->sd_lock)) {
323 325 error = EINTR;
324 326 goto ckreturn;
325 327 }
326 328 mutex_exit(&stp->sd_lock);
327 329 goto retry; /* could be clone! */
328 330 }
329 331
330 332 if (stp->sd_flag & (STRDERR|STWRERR)) {
331 333 error = EIO;
332 334 goto ckreturn;
333 335 }
334 336
335 337 stp->sd_flag |= STWOPEN;
336 338 mutex_exit(&stp->sd_lock);
337 339 }
338 340
339 341 /*
340 342 * Open all modules and devices down stream to notify
341 343 * that another user is streaming. For modules, set the
342 344 * last argument to MODOPEN and do not pass any open flags.
343 345 * Ignore dummydev since this is not the first open.
344 346 */
345 347 claimstr(stp->sd_wrq);
346 348 qp = stp->sd_wrq;
347 349 while (_SAMESTR(qp)) {
348 350 qp = qp->q_next;
349 351 if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
350 352 break;
351 353 }
352 354 releasestr(stp->sd_wrq);
353 355 mutex_enter(&stp->sd_lock);
354 356 stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
355 357 stp->sd_rerror = 0;
356 358 stp->sd_werror = 0;
357 359 ckreturn:
358 360 cv_broadcast(&stp->sd_monitor);
359 361 mutex_exit(&stp->sd_lock);
360 362 return (error);
361 363 }
362 364
363 365 /*
364 366 * This vnode isn't streaming. SPECFS already
365 367 * checked for multiple vnodes pointing to the
366 368 * same stream, so create a stream to the driver.
367 369 */
368 370 qp = allocq();
369 371 stp = shalloc(qp);
370 372
371 373 /*
372 374 * Initialize stream head. shalloc() has given us
373 375 * exclusive access, and we have the vnode locked;
374 376 * we can do whatever we want with stp.
375 377 */
376 378 stp->sd_flag = STWOPEN;
377 379 stp->sd_siglist = NULL;
378 380 stp->sd_pollist.ph_list = NULL;
379 381 stp->sd_sigflags = 0;
380 382 stp->sd_mark = NULL;
381 383 stp->sd_closetime = STRTIMOUT;
382 384 stp->sd_sidp = NULL;
383 385 stp->sd_pgidp = NULL;
384 386 stp->sd_vnode = vp;
385 387 stp->sd_rerror = 0;
386 388 stp->sd_werror = 0;
387 389 stp->sd_wroff = 0;
388 390 stp->sd_tail = 0;
↓ open down ↓ |
234 lines elided |
↑ open up ↑ |
389 391 stp->sd_iocblk = NULL;
390 392 stp->sd_cmdblk = NULL;
391 393 stp->sd_pushcnt = 0;
392 394 stp->sd_qn_minpsz = 0;
393 395 stp->sd_qn_maxpsz = INFPSZ - 1; /* used to check for initialization */
394 396 stp->sd_maxblk = INFPSZ;
395 397 qp->q_ptr = _WR(qp)->q_ptr = stp;
396 398 STREAM(qp) = STREAM(_WR(qp)) = stp;
397 399 vp->v_stream = stp;
398 400 mutex_exit(&vp->v_lock);
401 +
402 + /*
403 + * If this is not a system process, then add it to
404 + * the list associated with the stream head.
405 + */
406 + if (!(curproc->p_flag & SSYS) && is_xti_str(stp))
407 + sh_insert_pid(stp, curproc->p_pidp->pid_id);
408 +
399 409 if (vp->v_type == VFIFO) {
400 410 stp->sd_flag |= OLDNDELAY;
401 411 /*
402 412 * This means, both for pipes and fifos
403 413 * strwrite will send SIGPIPE if the other
404 414 * end is closed. For putmsg it depends
405 415 * on whether it is a XPG4_2 application
406 416 * or not
407 417 */
408 418 stp->sd_wput_opt = SW_SIGPIPE;
409 419
410 420 /* setq might sleep in kmem_alloc - avoid holding locks. */
411 421 setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
412 422 SQ_CI|SQ_CO, B_FALSE);
413 423
414 424 set_qend(qp);
415 425 stp->sd_strtab = fifo_getinfo();
416 426 _WR(qp)->q_nfsrv = _WR(qp);
417 427 qp->q_nfsrv = qp;
418 428 /*
419 429 * Wake up others that are waiting for stream to be created.
420 430 */
421 431 mutex_enter(&stp->sd_lock);
422 432 /*
423 433 * nothing is be pushed on stream yet, so
424 434 * optimized stream head packetsizes are just that
425 435 * of the read queue
426 436 */
427 437 stp->sd_qn_minpsz = qp->q_minpsz;
428 438 stp->sd_qn_maxpsz = qp->q_maxpsz;
429 439 stp->sd_flag &= ~STWOPEN;
430 440 goto fifo_opendone;
431 441 }
432 442 /* setq might sleep in kmem_alloc - avoid holding locks. */
433 443 setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
434 444
435 445 set_qend(qp);
436 446
437 447 /*
438 448 * Open driver and create stream to it (via qattach).
439 449 */
440 450 savedev = *devp;
441 451 cloneopen = (getmajor(*devp) == clone_major);
442 452 if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
443 453 mutex_enter(&vp->v_lock);
444 454 vp->v_stream = NULL;
445 455 mutex_exit(&vp->v_lock);
446 456 mutex_enter(&stp->sd_lock);
447 457 cv_broadcast(&stp->sd_monitor);
448 458 mutex_exit(&stp->sd_lock);
449 459 freeq(_RD(qp));
450 460 shfree(stp);
451 461 return (error);
452 462 }
453 463 /*
454 464 * Set sd_strtab after open in order to handle clonable drivers
455 465 */
456 466 stp->sd_strtab = STREAMSTAB(getmajor(*devp));
457 467
458 468 /*
459 469 * Historical note: dummydev used to be be prior to the initial
460 470 * open (via qattach above), which made the value seen
461 471 * inconsistent between an I_PUSH and an autopush of a module.
462 472 */
463 473 dummydev = *devp;
464 474
465 475 /*
466 476 * For clone open of old style (Q not associated) network driver,
467 477 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
468 478 */
469 479 brq = _RD(_WR(qp)->q_next);
470 480 major = getmajor(*devp);
471 481 if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
472 482 ((brq->q_flag & _QASSOCIATED) == 0)) {
473 483 if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
474 484 cmn_err(CE_WARN, "cannot push " DRMODNAME
475 485 " streams module");
476 486 }
477 487
478 488 if (!NETWORK_DRV(major)) {
479 489 savedev = *devp;
480 490 } else {
481 491 /*
482 492 * For network devices, process differently based on the
483 493 * return value from dld_autopush():
484 494 *
485 495 * 0: the passed-in device points to a GLDv3 datalink with
486 496 * per-link autopush configuration; use that configuration
487 497 * and ignore any per-driver autopush configuration.
488 498 *
489 499 * 1: the passed-in device points to a physical GLDv3
490 500 * datalink without per-link autopush configuration. The
491 501 * passed in device was changed to refer to the actual
492 502 * physical device (if it's not already); we use that new
493 503 * device to look up any per-driver autopush configuration.
494 504 *
495 505 * -1: neither of the above cases applied; use the initial
496 506 * device to look up any per-driver autopush configuration.
497 507 */
498 508 switch (dld_autopush(&savedev, &dlap)) {
499 509 case 0:
500 510 zoneid = crgetzoneid(crp);
501 511 for (s = 0; s < dlap.dap_npush; s++) {
502 512 error = push_mod(qp, &dummydev, stp,
503 513 dlap.dap_aplist[s], dlap.dap_anchor, crp,
504 514 zoneid);
505 515 if (error != 0)
506 516 break;
507 517 }
508 518 goto opendone;
509 519 case 1:
510 520 break;
511 521 case -1:
512 522 savedev = *devp;
513 523 break;
514 524 }
515 525 }
516 526 /*
517 527 * Find the autopush configuration based on "savedev". Start with the
518 528 * global zone. If not found check in the local zone.
519 529 */
520 530 zoneid = GLOBAL_ZONEID;
521 531 retryap:
522 532 ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
523 533 netstack_str;
524 534 if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
525 535 netstack_rele(ss->ss_netstack);
526 536 if (zoneid == GLOBAL_ZONEID) {
527 537 /*
528 538 * None found. Also look in the zone's autopush table.
529 539 */
530 540 zoneid = crgetzoneid(crp);
531 541 if (zoneid != GLOBAL_ZONEID)
532 542 goto retryap;
533 543 }
534 544 goto opendone;
535 545 }
536 546 anchor = ap->ap_anchor;
537 547 zoneid = crgetzoneid(crp);
538 548 for (s = 0; s < ap->ap_npush; s++) {
539 549 error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
540 550 anchor, crp, zoneid);
541 551 if (error != 0)
542 552 break;
543 553 }
544 554 sad_ap_rele(ap, ss);
545 555 netstack_rele(ss->ss_netstack);
546 556
547 557 opendone:
548 558
549 559 /*
550 560 * let specfs know that open failed part way through
551 561 */
552 562 if (error) {
553 563 mutex_enter(&stp->sd_lock);
554 564 stp->sd_flag |= STREOPENFAIL;
555 565 mutex_exit(&stp->sd_lock);
556 566 }
557 567
558 568 /*
559 569 * Wake up others that are waiting for stream to be created.
560 570 */
561 571 mutex_enter(&stp->sd_lock);
562 572 stp->sd_flag &= ~STWOPEN;
563 573
564 574 /*
565 575 * As a performance concern we are caching the values of
566 576 * q_minpsz and q_maxpsz of the module below the stream
567 577 * head in the stream head.
568 578 */
569 579 mutex_enter(QLOCK(stp->sd_wrq->q_next));
570 580 rmin = stp->sd_wrq->q_next->q_minpsz;
571 581 rmax = stp->sd_wrq->q_next->q_maxpsz;
572 582 mutex_exit(QLOCK(stp->sd_wrq->q_next));
573 583
574 584 /* do this processing here as a performance concern */
575 585 if (strmsgsz != 0) {
576 586 if (rmax == INFPSZ)
577 587 rmax = strmsgsz;
578 588 else
579 589 rmax = MIN(strmsgsz, rmax);
580 590 }
581 591
582 592 mutex_enter(QLOCK(stp->sd_wrq));
583 593 stp->sd_qn_minpsz = rmin;
584 594 stp->sd_qn_maxpsz = rmax;
585 595 mutex_exit(QLOCK(stp->sd_wrq));
586 596
587 597 fifo_opendone:
588 598 cv_broadcast(&stp->sd_monitor);
589 599 mutex_exit(&stp->sd_lock);
590 600 return (error);
591 601 }
592 602
593 603 static int strsink(queue_t *, mblk_t *);
594 604 static struct qinit deadrend = {
595 605 strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
596 606 };
597 607 static struct qinit deadwend = {
598 608 NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
599 609 };
600 610
601 611 /*
602 612 * Close a stream.
603 613 * This is called from closef() on the last close of an open stream.
604 614 * Strclean() will already have removed the siglist and pollist
605 615 * information, so all that remains is to remove all multiplexor links
606 616 * for the stream, pop all the modules (and the driver), and free the
607 617 * stream structure.
608 618 */
609 619
610 620 int
611 621 strclose(struct vnode *vp, int flag, cred_t *crp)
612 622 {
613 623 struct stdata *stp;
614 624 queue_t *qp;
615 625 int rval;
616 626 int freestp = 1;
617 627 queue_t *rmq;
618 628
619 629 TRACE_1(TR_FAC_STREAMS_FR,
620 630 TR_STRCLOSE, "strclose:%p", vp);
621 631 ASSERT(vp->v_stream);
622 632
623 633 stp = vp->v_stream;
624 634 ASSERT(!(stp->sd_flag & STPLEX));
625 635 qp = stp->sd_wrq;
626 636
627 637 /*
628 638 * Needed so that strpoll will return non-zero for this fd.
629 639 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
630 640 */
631 641 mutex_enter(&stp->sd_lock);
632 642 stp->sd_flag |= STRHUP;
633 643 mutex_exit(&stp->sd_lock);
634 644
635 645 /*
636 646 * If the registered process or process group did not have an
637 647 * open instance of this stream then strclean would not be
638 648 * called. Thus at the time of closing all remaining siglist entries
639 649 * are removed.
640 650 */
641 651 if (stp->sd_siglist != NULL)
642 652 strcleanall(vp);
643 653
644 654 ASSERT(stp->sd_siglist == NULL);
645 655 ASSERT(stp->sd_sigflags == 0);
646 656
647 657 if (STRMATED(stp)) {
648 658 struct stdata *strmatep = stp->sd_mate;
649 659 int waited = 1;
650 660
651 661 STRLOCKMATES(stp);
652 662 while (waited) {
653 663 waited = 0;
654 664 while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
655 665 mutex_exit(&strmatep->sd_lock);
656 666 cv_wait(&stp->sd_monitor, &stp->sd_lock);
657 667 mutex_exit(&stp->sd_lock);
658 668 STRLOCKMATES(stp);
659 669 waited = 1;
660 670 }
661 671 while (strmatep->sd_flag &
662 672 (STWOPEN|STRCLOSE|STRPLUMB)) {
663 673 mutex_exit(&stp->sd_lock);
664 674 cv_wait(&strmatep->sd_monitor,
665 675 &strmatep->sd_lock);
666 676 mutex_exit(&strmatep->sd_lock);
667 677 STRLOCKMATES(stp);
668 678 waited = 1;
669 679 }
670 680 }
671 681 stp->sd_flag |= STRCLOSE;
672 682 STRUNLOCKMATES(stp);
673 683 } else {
674 684 mutex_enter(&stp->sd_lock);
675 685 stp->sd_flag |= STRCLOSE;
676 686 mutex_exit(&stp->sd_lock);
677 687 }
678 688
679 689 ASSERT(qp->q_first == NULL); /* No more delayed write */
680 690
681 691 /* Check if an I_LINK was ever done on this stream */
682 692 if (stp->sd_flag & STRHASLINKS) {
683 693 netstack_t *ns;
684 694 str_stack_t *ss;
685 695
686 696 ns = netstack_find_by_cred(crp);
687 697 ASSERT(ns != NULL);
688 698 ss = ns->netstack_str;
689 699 ASSERT(ss != NULL);
690 700
691 701 (void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
692 702 netstack_rele(ss->ss_netstack);
693 703 }
694 704
695 705 while (_SAMESTR(qp)) {
696 706 /*
697 707 * Holding sd_lock prevents q_next from changing in
698 708 * this stream.
699 709 */
700 710 mutex_enter(&stp->sd_lock);
701 711 if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
702 712
703 713 /*
704 714 * sleep until awakened by strwsrv() or timeout
705 715 */
706 716 for (;;) {
707 717 mutex_enter(QLOCK(qp->q_next));
708 718 if (!(qp->q_next->q_mblkcnt)) {
709 719 mutex_exit(QLOCK(qp->q_next));
710 720 break;
711 721 }
712 722 stp->sd_flag |= WSLEEP;
713 723
714 724 /* ensure strwsrv gets enabled */
715 725 qp->q_next->q_flag |= QWANTW;
716 726 mutex_exit(QLOCK(qp->q_next));
717 727 /* get out if we timed out or recv'd a signal */
718 728 if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
719 729 stp->sd_closetime, 0) <= 0) {
720 730 break;
721 731 }
722 732 }
723 733 stp->sd_flag &= ~WSLEEP;
724 734 }
725 735 mutex_exit(&stp->sd_lock);
726 736
727 737 rmq = qp->q_next;
728 738 if (rmq->q_flag & QISDRV) {
729 739 ASSERT(!_SAMESTR(rmq));
730 740 wait_sq_svc(_RD(qp)->q_syncq);
731 741 }
732 742
733 743 qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
734 744 }
735 745
736 746 /*
737 747 * Since we call pollwakeup in close() now, the poll list should
738 748 * be empty in most cases. The only exception is the layered devices
739 749 * (e.g. the console drivers with redirection modules pushed on top
740 750 * of it). We have to do this after calling qdetach() because
741 751 * the redirection module won't have torn down the console
742 752 * redirection until after qdetach() has been invoked.
743 753 */
744 754 if (stp->sd_pollist.ph_list != NULL) {
745 755 pollwakeup(&stp->sd_pollist, POLLERR);
746 756 pollhead_clean(&stp->sd_pollist);
747 757 }
748 758 ASSERT(stp->sd_pollist.ph_list == NULL);
749 759 ASSERT(stp->sd_sidp == NULL);
750 760 ASSERT(stp->sd_pgidp == NULL);
751 761
752 762 /* Prevent qenable from re-enabling the stream head queue */
753 763 disable_svc(_RD(qp));
754 764
755 765 /*
756 766 * Wait until service procedure of each queue is
757 767 * run, if QINSERVICE is set.
758 768 */
759 769 wait_svc(_RD(qp));
760 770
761 771 /*
762 772 * Now, flush both queues.
763 773 */
764 774 flushq(_RD(qp), FLUSHALL);
765 775 flushq(qp, FLUSHALL);
766 776
767 777 /*
768 778 * If the write queue of the stream head is pointing to a
769 779 * read queue, we have a twisted stream. If the read queue
770 780 * is alive, convert the stream head queues into a dead end.
771 781 * If the read queue is dead, free the dead pair.
772 782 */
773 783 if (qp->q_next && !_SAMESTR(qp)) {
774 784 if (qp->q_next->q_qinfo == &deadrend) { /* half-closed pipe */
775 785 flushq(qp->q_next, FLUSHALL); /* ensure no message */
776 786 shfree(qp->q_next->q_stream);
777 787 freeq(qp->q_next);
778 788 freeq(_RD(qp));
779 789 } else if (qp->q_next == _RD(qp)) { /* fifo */
780 790 freeq(_RD(qp));
781 791 } else { /* pipe */
782 792 freestp = 0;
783 793 /*
784 794 * The q_info pointers are never accessed when
785 795 * SQLOCK is held.
786 796 */
787 797 ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
788 798 mutex_enter(SQLOCK(qp->q_syncq));
789 799 qp->q_qinfo = &deadwend;
790 800 _RD(qp)->q_qinfo = &deadrend;
791 801 mutex_exit(SQLOCK(qp->q_syncq));
792 802 }
793 803 } else {
794 804 freeq(_RD(qp)); /* free stream head queue pair */
795 805 }
796 806
797 807 mutex_enter(&vp->v_lock);
798 808 if (stp->sd_iocblk) {
799 809 if (stp->sd_iocblk != (mblk_t *)-1) {
800 810 freemsg(stp->sd_iocblk);
801 811 }
802 812 stp->sd_iocblk = NULL;
803 813 }
804 814 stp->sd_vnode = NULL;
805 815 vp->v_stream = NULL;
806 816 mutex_exit(&vp->v_lock);
807 817 mutex_enter(&stp->sd_lock);
808 818 freemsg(stp->sd_cmdblk);
809 819 stp->sd_cmdblk = NULL;
810 820 stp->sd_flag &= ~STRCLOSE;
811 821 cv_broadcast(&stp->sd_monitor);
812 822 mutex_exit(&stp->sd_lock);
813 823
814 824 if (freestp)
815 825 shfree(stp);
816 826 return (0);
817 827 }
818 828
819 829 static int
820 830 strsink(queue_t *q, mblk_t *bp)
821 831 {
822 832 struct copyresp *resp;
823 833
824 834 switch (bp->b_datap->db_type) {
825 835 case M_FLUSH:
826 836 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
827 837 *bp->b_rptr &= ~FLUSHR;
828 838 bp->b_flag |= MSGNOLOOP;
829 839 /*
830 840 * Protect against the driver passing up
831 841 * messages after it has done a qprocsoff.
832 842 */
833 843 if (_OTHERQ(q)->q_next == NULL)
834 844 freemsg(bp);
835 845 else
836 846 qreply(q, bp);
837 847 } else {
838 848 freemsg(bp);
839 849 }
840 850 break;
841 851
842 852 case M_COPYIN:
843 853 case M_COPYOUT:
844 854 if (bp->b_cont) {
845 855 freemsg(bp->b_cont);
846 856 bp->b_cont = NULL;
847 857 }
848 858 bp->b_datap->db_type = M_IOCDATA;
849 859 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
850 860 resp = (struct copyresp *)bp->b_rptr;
851 861 resp->cp_rval = (caddr_t)1; /* failure */
852 862 /*
853 863 * Protect against the driver passing up
854 864 * messages after it has done a qprocsoff.
855 865 */
856 866 if (_OTHERQ(q)->q_next == NULL)
857 867 freemsg(bp);
858 868 else
859 869 qreply(q, bp);
860 870 break;
861 871
862 872 case M_IOCTL:
863 873 if (bp->b_cont) {
864 874 freemsg(bp->b_cont);
865 875 bp->b_cont = NULL;
866 876 }
867 877 bp->b_datap->db_type = M_IOCNAK;
868 878 /*
869 879 * Protect against the driver passing up
870 880 * messages after it has done a qprocsoff.
871 881 */
872 882 if (_OTHERQ(q)->q_next == NULL)
873 883 freemsg(bp);
874 884 else
875 885 qreply(q, bp);
876 886 break;
877 887
878 888 default:
879 889 freemsg(bp);
880 890 break;
881 891 }
882 892
883 893 return (0);
884 894 }
885 895
886 896 /*
887 897 * Clean up after a process when it closes a stream. This is called
888 898 * from closef for all closes, whereas strclose is called only for the
889 899 * last close on a stream. The siglist is scanned for entries for the
890 900 * current process, and these are removed.
891 901 */
892 902 void
893 903 strclean(struct vnode *vp)
894 904 {
895 905 strsig_t *ssp, *pssp, *tssp;
896 906 stdata_t *stp;
897 907 int update = 0;
898 908
899 909 TRACE_1(TR_FAC_STREAMS_FR,
900 910 TR_STRCLEAN, "strclean:%p", vp);
901 911 stp = vp->v_stream;
902 912 pssp = NULL;
903 913 mutex_enter(&stp->sd_lock);
904 914 ssp = stp->sd_siglist;
905 915 while (ssp) {
906 916 if (ssp->ss_pidp == curproc->p_pidp) {
907 917 tssp = ssp->ss_next;
908 918 if (pssp)
909 919 pssp->ss_next = tssp;
910 920 else
911 921 stp->sd_siglist = tssp;
912 922 mutex_enter(&pidlock);
913 923 PID_RELE(ssp->ss_pidp);
914 924 mutex_exit(&pidlock);
915 925 kmem_free(ssp, sizeof (strsig_t));
916 926 update = 1;
917 927 ssp = tssp;
918 928 } else {
919 929 pssp = ssp;
920 930 ssp = ssp->ss_next;
921 931 }
922 932 }
923 933 if (update) {
924 934 stp->sd_sigflags = 0;
925 935 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
926 936 stp->sd_sigflags |= ssp->ss_events;
927 937 }
928 938 mutex_exit(&stp->sd_lock);
929 939 }
930 940
931 941 /*
932 942 * Used on the last close to remove any remaining items on the siglist.
933 943 * These could be present on the siglist due to I_ESETSIG calls that
934 944 * use process groups or processed that do not have an open file descriptor
935 945 * for this stream (Such entries would not be removed by strclean).
936 946 */
937 947 static void
938 948 strcleanall(struct vnode *vp)
939 949 {
940 950 strsig_t *ssp, *nssp;
941 951 stdata_t *stp;
942 952
943 953 stp = vp->v_stream;
944 954 mutex_enter(&stp->sd_lock);
945 955 ssp = stp->sd_siglist;
946 956 stp->sd_siglist = NULL;
947 957 while (ssp) {
948 958 nssp = ssp->ss_next;
949 959 mutex_enter(&pidlock);
950 960 PID_RELE(ssp->ss_pidp);
951 961 mutex_exit(&pidlock);
952 962 kmem_free(ssp, sizeof (strsig_t));
953 963 ssp = nssp;
954 964 }
955 965 stp->sd_sigflags = 0;
956 966 mutex_exit(&stp->sd_lock);
957 967 }
958 968
959 969 /*
960 970 * Retrieve the next message from the logical stream head read queue
961 971 * using either rwnext (if sync stream) or getq_noenab.
962 972 * It is the callers responsibility to call qbackenable after
963 973 * it is finished with the message. The caller should not call
964 974 * qbackenable until after any putback calls to avoid spurious backenabling.
965 975 */
966 976 mblk_t *
967 977 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
968 978 int *errorp)
969 979 {
970 980 mblk_t *bp;
971 981 int error;
972 982 ssize_t rbytes = 0;
973 983
974 984 /* Holding sd_lock prevents the read queue from changing */
975 985 ASSERT(MUTEX_HELD(&stp->sd_lock));
976 986
977 987 if (uiop != NULL && stp->sd_struiordq != NULL &&
978 988 q->q_first == NULL &&
979 989 (!first || (stp->sd_wakeq & RSLEEP))) {
980 990 /*
981 991 * Stream supports rwnext() for the read side.
982 992 * If this is the first time we're called by e.g. strread
983 993 * only do the downcall if there is a deferred wakeup
984 994 * (registered in sd_wakeq).
985 995 */
986 996 struiod_t uiod;
987 997
988 998 if (first)
989 999 stp->sd_wakeq &= ~RSLEEP;
990 1000
991 1001 (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
992 1002 sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
993 1003 uiod.d_mp = 0;
994 1004 /*
995 1005 * Mark that a thread is in rwnext on the read side
996 1006 * to prevent strrput from nacking ioctls immediately.
997 1007 * When the last concurrent rwnext returns
998 1008 * the ioctls are nack'ed.
999 1009 */
1000 1010 ASSERT(MUTEX_HELD(&stp->sd_lock));
1001 1011 stp->sd_struiodnak++;
1002 1012 /*
1003 1013 * Note: rwnext will drop sd_lock.
1004 1014 */
1005 1015 error = rwnext(q, &uiod);
1006 1016 ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1007 1017 mutex_enter(&stp->sd_lock);
1008 1018 stp->sd_struiodnak--;
1009 1019 while (stp->sd_struiodnak == 0 &&
1010 1020 ((bp = stp->sd_struionak) != NULL)) {
1011 1021 stp->sd_struionak = bp->b_next;
1012 1022 bp->b_next = NULL;
1013 1023 bp->b_datap->db_type = M_IOCNAK;
1014 1024 /*
1015 1025 * Protect against the driver passing up
1016 1026 * messages after it has done a qprocsoff.
1017 1027 */
1018 1028 if (_OTHERQ(q)->q_next == NULL)
1019 1029 freemsg(bp);
1020 1030 else {
1021 1031 mutex_exit(&stp->sd_lock);
1022 1032 qreply(q, bp);
1023 1033 mutex_enter(&stp->sd_lock);
1024 1034 }
1025 1035 }
1026 1036 ASSERT(MUTEX_HELD(&stp->sd_lock));
1027 1037 if (error == 0 || error == EWOULDBLOCK) {
1028 1038 if ((bp = uiod.d_mp) != NULL) {
1029 1039 *errorp = 0;
1030 1040 ASSERT(MUTEX_HELD(&stp->sd_lock));
1031 1041 return (bp);
1032 1042 }
1033 1043 error = 0;
1034 1044 } else if (error == EINVAL) {
1035 1045 /*
1036 1046 * The stream plumbing must have
1037 1047 * changed while we were away, so
1038 1048 * just turn off rwnext()s.
1039 1049 */
1040 1050 error = 0;
1041 1051 } else if (error == EBUSY) {
1042 1052 /*
1043 1053 * The module might have data in transit using putnext
1044 1054 * Fall back on waiting + getq.
1045 1055 */
1046 1056 error = 0;
1047 1057 } else {
1048 1058 *errorp = error;
1049 1059 ASSERT(MUTEX_HELD(&stp->sd_lock));
1050 1060 return (NULL);
1051 1061 }
1052 1062 /*
1053 1063 * Try a getq in case a rwnext() generated mblk
1054 1064 * has bubbled up via strrput().
1055 1065 */
1056 1066 }
1057 1067 *errorp = 0;
1058 1068 ASSERT(MUTEX_HELD(&stp->sd_lock));
1059 1069
1060 1070 /*
1061 1071 * If we have a valid uio, try and use this as a guide for how
1062 1072 * many bytes to retrieve from the queue via getq_noenab().
1063 1073 * Doing this can avoid unneccesary counting of overlong
1064 1074 * messages in putback(). We currently only do this for sockets
1065 1075 * and only if there is no sd_rputdatafunc hook.
1066 1076 *
1067 1077 * The sd_rputdatafunc hook transforms the entire message
1068 1078 * before any bytes in it can be given to a client. So, rbytes
1069 1079 * must be 0 if there is a hook.
1070 1080 */
1071 1081 if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK) &&
1072 1082 (stp->sd_rputdatafunc == NULL))
1073 1083 rbytes = uiop->uio_resid;
1074 1084
1075 1085 return (getq_noenab(q, rbytes));
1076 1086 }
1077 1087
1078 1088 /*
1079 1089 * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1080 1090 * If the message does not fit in the uio the remainder of it is returned;
1081 1091 * otherwise NULL is returned. Any embedded zero-length mblk_t's are
1082 1092 * consumed, even if uio_resid reaches zero. On error, `*errorp' is set to
1083 1093 * the error code, the message is consumed, and NULL is returned.
1084 1094 */
1085 1095 static mblk_t *
1086 1096 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1087 1097 {
1088 1098 int error;
1089 1099 ptrdiff_t n;
1090 1100 mblk_t *nbp;
1091 1101
1092 1102 ASSERT(bp->b_wptr >= bp->b_rptr);
1093 1103
1094 1104 do {
1095 1105 if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1096 1106 ASSERT(n > 0);
1097 1107
1098 1108 error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1099 1109 if (error != 0) {
1100 1110 freemsg(bp);
1101 1111 *errorp = error;
1102 1112 return (NULL);
1103 1113 }
1104 1114 }
1105 1115
1106 1116 bp->b_rptr += n;
1107 1117 while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1108 1118 nbp = bp;
1109 1119 bp = bp->b_cont;
1110 1120 freeb(nbp);
1111 1121 }
1112 1122 } while (bp != NULL && uiop->uio_resid > 0);
1113 1123
1114 1124 *errorp = 0;
1115 1125 return (bp);
1116 1126 }
1117 1127
1118 1128 /*
1119 1129 * Read a stream according to the mode flags in sd_flag:
1120 1130 *
1121 1131 * (default mode) - Byte stream, msg boundaries are ignored
1122 1132 * RD_MSGDIS (msg discard) - Read on msg boundaries and throw away
1123 1133 * any data remaining in msg
1124 1134 * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1125 1135 * any remaining data on head of read queue
1126 1136 *
1127 1137 * Consume readable messages on the front of the queue until
1128 1138 * ttolwp(curthread)->lwp_count
1129 1139 * is satisfied, the readable messages are exhausted, or a message
1130 1140 * boundary is reached in a message mode. If no data was read and
1131 1141 * the stream was not opened with the NDELAY flag, block until data arrives.
1132 1142 * Otherwise return the data read and update the count.
1133 1143 *
1134 1144 * In default mode a 0 length message signifies end-of-file and terminates
1135 1145 * a read in progress. The 0 length message is removed from the queue
1136 1146 * only if it is the only message read (no data is read).
1137 1147 *
1138 1148 * An attempt to read an M_PROTO or M_PCPROTO message results in an
1139 1149 * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1140 1150 * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1141 1151 * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1142 1152 * are unlinked from and M_DATA blocks in the message, the protos are
1143 1153 * thrown away, and the data is read.
1144 1154 */
1145 1155 /* ARGSUSED */
1146 1156 int
1147 1157 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1148 1158 {
1149 1159 struct stdata *stp;
1150 1160 mblk_t *bp, *nbp;
1151 1161 queue_t *q;
1152 1162 int error = 0;
1153 1163 uint_t old_sd_flag;
1154 1164 int first;
1155 1165 char rflg;
1156 1166 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
1157 1167 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
1158 1168 short delim;
1159 1169 unsigned char pri = 0;
1160 1170 char waitflag;
1161 1171 unsigned char type;
1162 1172
1163 1173 TRACE_1(TR_FAC_STREAMS_FR,
1164 1174 TR_STRREAD_ENTER, "strread:%p", vp);
1165 1175 ASSERT(vp->v_stream);
1166 1176 stp = vp->v_stream;
1167 1177
1168 1178 mutex_enter(&stp->sd_lock);
1169 1179
1170 1180 if ((error = i_straccess(stp, JCREAD)) != 0) {
1171 1181 mutex_exit(&stp->sd_lock);
1172 1182 return (error);
1173 1183 }
1174 1184
1175 1185 if (stp->sd_flag & (STRDERR|STPLEX)) {
1176 1186 error = strgeterr(stp, STRDERR|STPLEX, 0);
1177 1187 if (error != 0) {
1178 1188 mutex_exit(&stp->sd_lock);
1179 1189 return (error);
1180 1190 }
1181 1191 }
1182 1192
1183 1193 /*
1184 1194 * Loop terminates when uiop->uio_resid == 0.
1185 1195 */
1186 1196 rflg = 0;
1187 1197 waitflag = READWAIT;
1188 1198 q = _RD(stp->sd_wrq);
1189 1199 for (;;) {
1190 1200 ASSERT(MUTEX_HELD(&stp->sd_lock));
1191 1201 old_sd_flag = stp->sd_flag;
1192 1202 mark = 0;
1193 1203 delim = 0;
1194 1204 first = 1;
1195 1205 while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1196 1206 int done = 0;
1197 1207
1198 1208 ASSERT(MUTEX_HELD(&stp->sd_lock));
1199 1209
1200 1210 if (error != 0)
1201 1211 goto oops;
1202 1212
1203 1213 if (stp->sd_flag & (STRHUP|STREOF)) {
1204 1214 goto oops;
1205 1215 }
1206 1216 if (rflg && !(stp->sd_flag & STRDELIM)) {
1207 1217 goto oops;
1208 1218 }
1209 1219 /*
1210 1220 * If a read(fd,buf,0) has been done, there is no
1211 1221 * need to sleep. We always have zero bytes to
1212 1222 * return.
1213 1223 */
1214 1224 if (uiop->uio_resid == 0) {
1215 1225 goto oops;
1216 1226 }
1217 1227
1218 1228 qbackenable(q, 0);
1219 1229
1220 1230 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1221 1231 "strread calls strwaitq:%p, %p, %p",
1222 1232 vp, uiop, crp);
1223 1233 if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1224 1234 uiop->uio_fmode, -1, &done)) != 0 || done) {
1225 1235 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1226 1236 "strread error or done:%p, %p, %p",
1227 1237 vp, uiop, crp);
1228 1238 if ((uiop->uio_fmode & FNDELAY) &&
1229 1239 (stp->sd_flag & OLDNDELAY) &&
1230 1240 (error == EAGAIN))
1231 1241 error = 0;
1232 1242 goto oops;
1233 1243 }
1234 1244 TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1235 1245 "strread awakes:%p, %p, %p", vp, uiop, crp);
1236 1246 if ((error = i_straccess(stp, JCREAD)) != 0) {
1237 1247 goto oops;
1238 1248 }
1239 1249 first = 0;
1240 1250 }
1241 1251
1242 1252 ASSERT(MUTEX_HELD(&stp->sd_lock));
1243 1253 ASSERT(bp);
1244 1254 pri = bp->b_band;
1245 1255 /*
1246 1256 * Extract any mark information. If the message is not
1247 1257 * completely consumed this information will be put in the mblk
1248 1258 * that is putback.
1249 1259 * If MSGMARKNEXT is set and the message is completely consumed
1250 1260 * the STRATMARK flag will be set below. Likewise, if
1251 1261 * MSGNOTMARKNEXT is set and the message is
1252 1262 * completely consumed STRNOTATMARK will be set.
1253 1263 *
1254 1264 * For some unknown reason strread only breaks the read at the
1255 1265 * last mark.
1256 1266 */
1257 1267 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1258 1268 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1259 1269 (MSGMARKNEXT|MSGNOTMARKNEXT));
1260 1270 if (mark != 0 && bp == stp->sd_mark) {
1261 1271 if (rflg) {
1262 1272 putback(stp, q, bp, pri);
1263 1273 goto oops;
1264 1274 }
1265 1275 mark |= _LASTMARK;
1266 1276 stp->sd_mark = NULL;
1267 1277 }
1268 1278 if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1269 1279 delim = 1;
1270 1280 mutex_exit(&stp->sd_lock);
1271 1281
1272 1282 if (STREAM_NEEDSERVICE(stp))
1273 1283 stream_runservice(stp);
1274 1284
1275 1285 type = bp->b_datap->db_type;
1276 1286
1277 1287 switch (type) {
1278 1288
1279 1289 case M_DATA:
1280 1290 ismdata:
1281 1291 if (msgnodata(bp)) {
1282 1292 if (mark || delim) {
1283 1293 freemsg(bp);
1284 1294 } else if (rflg) {
1285 1295
1286 1296 /*
1287 1297 * If already read data put zero
1288 1298 * length message back on queue else
1289 1299 * free msg and return 0.
1290 1300 */
1291 1301 bp->b_band = pri;
1292 1302 mutex_enter(&stp->sd_lock);
1293 1303 putback(stp, q, bp, pri);
1294 1304 mutex_exit(&stp->sd_lock);
1295 1305 } else {
1296 1306 freemsg(bp);
1297 1307 }
1298 1308 error = 0;
1299 1309 goto oops1;
1300 1310 }
1301 1311
1302 1312 rflg = 1;
1303 1313 waitflag |= NOINTR;
1304 1314 bp = struiocopyout(bp, uiop, &error);
1305 1315 if (error != 0)
1306 1316 goto oops1;
1307 1317
1308 1318 mutex_enter(&stp->sd_lock);
1309 1319 if (bp) {
1310 1320 /*
1311 1321 * Have remaining data in message.
1312 1322 * Free msg if in discard mode.
1313 1323 */
1314 1324 if (stp->sd_read_opt & RD_MSGDIS) {
1315 1325 freemsg(bp);
1316 1326 } else {
1317 1327 bp->b_band = pri;
1318 1328 if ((mark & _LASTMARK) &&
1319 1329 (stp->sd_mark == NULL))
1320 1330 stp->sd_mark = bp;
1321 1331 bp->b_flag |= mark & ~_LASTMARK;
1322 1332 if (delim)
1323 1333 bp->b_flag |= MSGDELIM;
1324 1334 if (msgnodata(bp))
1325 1335 freemsg(bp);
1326 1336 else
1327 1337 putback(stp, q, bp, pri);
1328 1338 }
1329 1339 } else {
1330 1340 /*
1331 1341 * Consumed the complete message.
1332 1342 * Move the MSG*MARKNEXT information
1333 1343 * to the stream head just in case
1334 1344 * the read queue becomes empty.
1335 1345 *
1336 1346 * If the stream head was at the mark
1337 1347 * (STRATMARK) before we dropped sd_lock above
1338 1348 * and some data was consumed then we have
1339 1349 * moved past the mark thus STRATMARK is
1340 1350 * cleared. However, if a message arrived in
1341 1351 * strrput during the copyout above causing
1342 1352 * STRATMARK to be set we can not clear that
1343 1353 * flag.
1344 1354 */
1345 1355 if (mark &
1346 1356 (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1347 1357 if (mark & MSGMARKNEXT) {
1348 1358 stp->sd_flag &= ~STRNOTATMARK;
1349 1359 stp->sd_flag |= STRATMARK;
1350 1360 } else if (mark & MSGNOTMARKNEXT) {
1351 1361 stp->sd_flag &= ~STRATMARK;
1352 1362 stp->sd_flag |= STRNOTATMARK;
1353 1363 } else {
1354 1364 stp->sd_flag &=
1355 1365 ~(STRATMARK|STRNOTATMARK);
1356 1366 }
1357 1367 } else if (rflg && (old_sd_flag & STRATMARK)) {
1358 1368 stp->sd_flag &= ~STRATMARK;
1359 1369 }
1360 1370 }
1361 1371
1362 1372 /*
1363 1373 * Check for signal messages at the front of the read
1364 1374 * queue and generate the signal(s) if appropriate.
1365 1375 * The only signal that can be on queue is M_SIG at
1366 1376 * this point.
1367 1377 */
1368 1378 while ((((bp = q->q_first)) != NULL) &&
1369 1379 (bp->b_datap->db_type == M_SIG)) {
1370 1380 bp = getq_noenab(q, 0);
1371 1381 /*
1372 1382 * sd_lock is held so the content of the
1373 1383 * read queue can not change.
1374 1384 */
1375 1385 ASSERT(bp != NULL && DB_TYPE(bp) == M_SIG);
1376 1386 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
1377 1387 mutex_exit(&stp->sd_lock);
1378 1388 freemsg(bp);
1379 1389 if (STREAM_NEEDSERVICE(stp))
1380 1390 stream_runservice(stp);
1381 1391 mutex_enter(&stp->sd_lock);
1382 1392 }
1383 1393
1384 1394 if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1385 1395 delim ||
1386 1396 (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1387 1397 goto oops;
1388 1398 }
1389 1399 continue;
1390 1400
1391 1401 case M_SIG:
1392 1402 strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1393 1403 freemsg(bp);
1394 1404 mutex_enter(&stp->sd_lock);
1395 1405 continue;
1396 1406
1397 1407 case M_PROTO:
1398 1408 case M_PCPROTO:
1399 1409 /*
1400 1410 * Only data messages are readable.
1401 1411 * Any others generate an error, unless
1402 1412 * RD_PROTDIS or RD_PROTDAT is set.
1403 1413 */
1404 1414 if (stp->sd_read_opt & RD_PROTDAT) {
1405 1415 for (nbp = bp; nbp; nbp = nbp->b_next) {
1406 1416 if ((nbp->b_datap->db_type ==
1407 1417 M_PROTO) ||
1408 1418 (nbp->b_datap->db_type ==
1409 1419 M_PCPROTO)) {
1410 1420 nbp->b_datap->db_type = M_DATA;
1411 1421 } else {
1412 1422 break;
1413 1423 }
1414 1424 }
1415 1425 /*
1416 1426 * clear stream head hi pri flag based on
1417 1427 * first message
1418 1428 */
1419 1429 if (type == M_PCPROTO) {
1420 1430 mutex_enter(&stp->sd_lock);
1421 1431 stp->sd_flag &= ~STRPRI;
1422 1432 mutex_exit(&stp->sd_lock);
1423 1433 }
1424 1434 goto ismdata;
1425 1435 } else if (stp->sd_read_opt & RD_PROTDIS) {
1426 1436 /*
1427 1437 * discard non-data messages
1428 1438 */
1429 1439 while (bp &&
1430 1440 ((bp->b_datap->db_type == M_PROTO) ||
1431 1441 (bp->b_datap->db_type == M_PCPROTO))) {
1432 1442 nbp = unlinkb(bp);
1433 1443 freeb(bp);
1434 1444 bp = nbp;
1435 1445 }
1436 1446 /*
1437 1447 * clear stream head hi pri flag based on
1438 1448 * first message
1439 1449 */
1440 1450 if (type == M_PCPROTO) {
1441 1451 mutex_enter(&stp->sd_lock);
1442 1452 stp->sd_flag &= ~STRPRI;
1443 1453 mutex_exit(&stp->sd_lock);
1444 1454 }
1445 1455 if (bp) {
1446 1456 bp->b_band = pri;
1447 1457 goto ismdata;
1448 1458 } else {
1449 1459 break;
1450 1460 }
1451 1461 }
1452 1462 /* FALLTHRU */
1453 1463 case M_PASSFP:
1454 1464 if ((bp->b_datap->db_type == M_PASSFP) &&
1455 1465 (stp->sd_read_opt & RD_PROTDIS)) {
1456 1466 freemsg(bp);
1457 1467 break;
1458 1468 }
1459 1469 mutex_enter(&stp->sd_lock);
1460 1470 putback(stp, q, bp, pri);
1461 1471 mutex_exit(&stp->sd_lock);
1462 1472 if (rflg == 0)
1463 1473 error = EBADMSG;
1464 1474 goto oops1;
1465 1475
1466 1476 default:
1467 1477 /*
1468 1478 * Garbage on stream head read queue.
1469 1479 */
1470 1480 cmn_err(CE_WARN, "bad %x found at stream head\n",
1471 1481 bp->b_datap->db_type);
1472 1482 freemsg(bp);
1473 1483 goto oops1;
1474 1484 }
1475 1485 mutex_enter(&stp->sd_lock);
1476 1486 }
1477 1487 oops:
1478 1488 mutex_exit(&stp->sd_lock);
1479 1489 oops1:
1480 1490 qbackenable(q, pri);
1481 1491 return (error);
1482 1492 #undef _LASTMARK
1483 1493 }
1484 1494
1485 1495 /*
1486 1496 * Default processing of M_PROTO/M_PCPROTO messages.
1487 1497 * Determine which wakeups and signals are needed.
1488 1498 * This can be replaced by a user-specified procedure for kernel users
1489 1499 * of STREAMS.
1490 1500 */
1491 1501 /* ARGSUSED */
1492 1502 mblk_t *
1493 1503 strrput_proto(vnode_t *vp, mblk_t *mp,
1494 1504 strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1495 1505 strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1496 1506 {
1497 1507 *wakeups = RSLEEP;
1498 1508 *allmsgsigs = 0;
1499 1509
1500 1510 switch (mp->b_datap->db_type) {
1501 1511 case M_PROTO:
1502 1512 if (mp->b_band == 0) {
1503 1513 *firstmsgsigs = S_INPUT | S_RDNORM;
1504 1514 *pollwakeups = POLLIN | POLLRDNORM;
1505 1515 } else {
1506 1516 *firstmsgsigs = S_INPUT | S_RDBAND;
1507 1517 *pollwakeups = POLLIN | POLLRDBAND;
1508 1518 }
1509 1519 break;
1510 1520 case M_PCPROTO:
1511 1521 *firstmsgsigs = S_HIPRI;
1512 1522 *pollwakeups = POLLPRI;
1513 1523 break;
1514 1524 }
1515 1525 return (mp);
1516 1526 }
1517 1527
1518 1528 /*
1519 1529 * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1520 1530 * M_PASSFP messages.
1521 1531 * Determine which wakeups and signals are needed.
1522 1532 * This can be replaced by a user-specified procedure for kernel users
1523 1533 * of STREAMS.
1524 1534 */
1525 1535 /* ARGSUSED */
1526 1536 mblk_t *
1527 1537 strrput_misc(vnode_t *vp, mblk_t *mp,
1528 1538 strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1529 1539 strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1530 1540 {
1531 1541 *wakeups = 0;
1532 1542 *firstmsgsigs = 0;
1533 1543 *allmsgsigs = 0;
1534 1544 *pollwakeups = 0;
1535 1545 return (mp);
1536 1546 }
1537 1547
1538 1548 /*
1539 1549 * Stream read put procedure. Called from downstream driver/module
1540 1550 * with messages for the stream head. Data, protocol, and in-stream
1541 1551 * signal messages are placed on the queue, others are handled directly.
1542 1552 */
1543 1553 int
1544 1554 strrput(queue_t *q, mblk_t *bp)
1545 1555 {
1546 1556 struct stdata *stp;
1547 1557 ulong_t rput_opt;
1548 1558 strwakeup_t wakeups;
1549 1559 strsigset_t firstmsgsigs; /* Signals if first message on queue */
1550 1560 strsigset_t allmsgsigs; /* Signals for all messages */
1551 1561 strsigset_t signals; /* Signals events to generate */
1552 1562 strpollset_t pollwakeups;
1553 1563 mblk_t *nextbp;
1554 1564 uchar_t band = 0;
1555 1565 int hipri_sig;
1556 1566
1557 1567 stp = (struct stdata *)q->q_ptr;
1558 1568 /*
1559 1569 * Use rput_opt for optimized access to the SR_ flags except
1560 1570 * SR_POLLIN. That flag has to be checked under sd_lock since it
1561 1571 * is modified by strpoll().
1562 1572 */
1563 1573 rput_opt = stp->sd_rput_opt;
1564 1574
1565 1575 ASSERT(qclaimed(q));
1566 1576 TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1567 1577 "strrput called with message type:q %p bp %p", q, bp);
1568 1578
1569 1579 /*
1570 1580 * Perform initial processing and pass to the parameterized functions.
1571 1581 */
1572 1582 ASSERT(bp->b_next == NULL);
1573 1583
1574 1584 switch (bp->b_datap->db_type) {
1575 1585 case M_DATA:
1576 1586 /*
1577 1587 * sockfs is the only consumer of STREOF and when it is set,
1578 1588 * it implies that the receiver is not interested in receiving
1579 1589 * any more data, hence the mblk is freed to prevent unnecessary
1580 1590 * message queueing at the stream head.
1581 1591 */
1582 1592 if (stp->sd_flag == STREOF) {
1583 1593 freemsg(bp);
1584 1594 return (0);
1585 1595 }
1586 1596 if ((rput_opt & SR_IGN_ZEROLEN) &&
1587 1597 bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1588 1598 /*
1589 1599 * Ignore zero-length M_DATA messages. These might be
1590 1600 * generated by some transports.
1591 1601 * The zero-length M_DATA messages, even if they
1592 1602 * are ignored, should effect the atmark tracking and
1593 1603 * should wake up a thread sleeping in strwaitmark.
1594 1604 */
1595 1605 mutex_enter(&stp->sd_lock);
1596 1606 if (bp->b_flag & MSGMARKNEXT) {
1597 1607 /*
1598 1608 * Record the position of the mark either
1599 1609 * in q_last or in STRATMARK.
1600 1610 */
1601 1611 if (q->q_last != NULL) {
1602 1612 q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1603 1613 q->q_last->b_flag |= MSGMARKNEXT;
1604 1614 } else {
1605 1615 stp->sd_flag &= ~STRNOTATMARK;
1606 1616 stp->sd_flag |= STRATMARK;
1607 1617 }
1608 1618 } else if (bp->b_flag & MSGNOTMARKNEXT) {
1609 1619 /*
1610 1620 * Record that this is not the position of
1611 1621 * the mark either in q_last or in
1612 1622 * STRNOTATMARK.
1613 1623 */
1614 1624 if (q->q_last != NULL) {
1615 1625 q->q_last->b_flag &= ~MSGMARKNEXT;
1616 1626 q->q_last->b_flag |= MSGNOTMARKNEXT;
1617 1627 } else {
1618 1628 stp->sd_flag &= ~STRATMARK;
1619 1629 stp->sd_flag |= STRNOTATMARK;
1620 1630 }
1621 1631 }
1622 1632 if (stp->sd_flag & RSLEEP) {
1623 1633 stp->sd_flag &= ~RSLEEP;
1624 1634 cv_broadcast(&q->q_wait);
1625 1635 }
1626 1636 mutex_exit(&stp->sd_lock);
1627 1637 freemsg(bp);
1628 1638 return (0);
1629 1639 }
1630 1640 wakeups = RSLEEP;
1631 1641 if (bp->b_band == 0) {
1632 1642 firstmsgsigs = S_INPUT | S_RDNORM;
1633 1643 pollwakeups = POLLIN | POLLRDNORM;
1634 1644 } else {
1635 1645 firstmsgsigs = S_INPUT | S_RDBAND;
1636 1646 pollwakeups = POLLIN | POLLRDBAND;
1637 1647 }
1638 1648 if (rput_opt & SR_SIGALLDATA)
1639 1649 allmsgsigs = firstmsgsigs;
1640 1650 else
1641 1651 allmsgsigs = 0;
1642 1652
1643 1653 mutex_enter(&stp->sd_lock);
1644 1654 if ((rput_opt & SR_CONSOL_DATA) &&
1645 1655 (q->q_last != NULL) &&
1646 1656 (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1647 1657 /*
1648 1658 * Consolidate an M_DATA message onto an M_DATA,
1649 1659 * M_PROTO, or M_PCPROTO by merging it with q_last.
1650 1660 * The consolidation does not take place if
1651 1661 * the old message is marked with either of the
1652 1662 * marks or the delim flag or if the new
1653 1663 * message is marked with MSGMARK. The MSGMARK
1654 1664 * check is needed to handle the odd semantics of
1655 1665 * MSGMARK where essentially the whole message
1656 1666 * is to be treated as marked.
1657 1667 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from the
1658 1668 * new message to the front of the b_cont chain.
1659 1669 */
1660 1670 mblk_t *lbp = q->q_last;
1661 1671 unsigned char db_type = lbp->b_datap->db_type;
1662 1672
1663 1673 if ((db_type == M_DATA || db_type == M_PROTO ||
1664 1674 db_type == M_PCPROTO) &&
1665 1675 !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1666 1676 rmvq_noenab(q, lbp);
1667 1677 /*
1668 1678 * The first message in the b_cont list
1669 1679 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1670 1680 * We need to handle the case where we
1671 1681 * are appending:
1672 1682 *
1673 1683 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1674 1684 * 2) a MSGMARKNEXT to a plain message.
1675 1685 * 3) a MSGNOTMARKNEXT to a plain message
1676 1686 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1677 1687 * message.
1678 1688 *
1679 1689 * Thus we never append a MSGMARKNEXT or
1680 1690 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1681 1691 */
1682 1692 if (bp->b_flag & MSGMARKNEXT) {
1683 1693 lbp->b_flag |= MSGMARKNEXT;
1684 1694 lbp->b_flag &= ~MSGNOTMARKNEXT;
1685 1695 bp->b_flag &= ~MSGMARKNEXT;
1686 1696 } else if (bp->b_flag & MSGNOTMARKNEXT) {
1687 1697 lbp->b_flag |= MSGNOTMARKNEXT;
1688 1698 bp->b_flag &= ~MSGNOTMARKNEXT;
1689 1699 }
1690 1700
1691 1701 linkb(lbp, bp);
1692 1702 bp = lbp;
1693 1703 /*
1694 1704 * The new message logically isn't the first
1695 1705 * even though the q_first check below thinks
1696 1706 * it is. Clear the firstmsgsigs to make it
1697 1707 * not appear to be first.
1698 1708 */
1699 1709 firstmsgsigs = 0;
1700 1710 }
1701 1711 }
1702 1712 break;
1703 1713
1704 1714 case M_PASSFP:
1705 1715 wakeups = RSLEEP;
1706 1716 allmsgsigs = 0;
1707 1717 if (bp->b_band == 0) {
1708 1718 firstmsgsigs = S_INPUT | S_RDNORM;
1709 1719 pollwakeups = POLLIN | POLLRDNORM;
1710 1720 } else {
1711 1721 firstmsgsigs = S_INPUT | S_RDBAND;
1712 1722 pollwakeups = POLLIN | POLLRDBAND;
1713 1723 }
1714 1724 mutex_enter(&stp->sd_lock);
1715 1725 break;
1716 1726
1717 1727 case M_PROTO:
1718 1728 case M_PCPROTO:
1719 1729 ASSERT(stp->sd_rprotofunc != NULL);
1720 1730 bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1721 1731 &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1722 1732 #define ALLSIG (S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1723 1733 S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1724 1734 #define ALLPOLL (POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1725 1735 POLLWRBAND)
1726 1736
1727 1737 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1728 1738 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1729 1739 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1730 1740 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1731 1741
1732 1742 mutex_enter(&stp->sd_lock);
1733 1743 break;
1734 1744
1735 1745 default:
1736 1746 ASSERT(stp->sd_rmiscfunc != NULL);
1737 1747 bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1738 1748 &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1739 1749 ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1740 1750 ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1741 1751 ASSERT((allmsgsigs & ~ALLSIG) == 0);
1742 1752 ASSERT((pollwakeups & ~ALLPOLL) == 0);
1743 1753 #undef ALLSIG
1744 1754 #undef ALLPOLL
1745 1755 mutex_enter(&stp->sd_lock);
1746 1756 break;
1747 1757 }
1748 1758 ASSERT(MUTEX_HELD(&stp->sd_lock));
1749 1759
1750 1760 /* By default generate superset of signals */
1751 1761 signals = (firstmsgsigs | allmsgsigs);
1752 1762
1753 1763 /*
1754 1764 * The proto and misc functions can return multiple messages
1755 1765 * as a b_next chain. Such messages are processed separately.
1756 1766 */
1757 1767 one_more:
1758 1768 hipri_sig = 0;
1759 1769 if (bp == NULL) {
1760 1770 nextbp = NULL;
1761 1771 } else {
1762 1772 nextbp = bp->b_next;
1763 1773 bp->b_next = NULL;
1764 1774
1765 1775 switch (bp->b_datap->db_type) {
1766 1776 case M_PCPROTO:
1767 1777 /*
1768 1778 * Only one priority protocol message is allowed at the
1769 1779 * stream head at a time.
1770 1780 */
1771 1781 if (stp->sd_flag & STRPRI) {
1772 1782 TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1773 1783 "M_PCPROTO already at head");
1774 1784 freemsg(bp);
1775 1785 mutex_exit(&stp->sd_lock);
1776 1786 goto done;
1777 1787 }
1778 1788 stp->sd_flag |= STRPRI;
1779 1789 hipri_sig = 1;
1780 1790 /* FALLTHRU */
1781 1791 case M_DATA:
1782 1792 case M_PROTO:
1783 1793 case M_PASSFP:
1784 1794 band = bp->b_band;
1785 1795 /*
1786 1796 * Marking doesn't work well when messages
1787 1797 * are marked in more than one band. We only
1788 1798 * remember the last message received, even if
1789 1799 * it is placed on the queue ahead of other
1790 1800 * marked messages.
1791 1801 */
1792 1802 if (bp->b_flag & MSGMARK)
1793 1803 stp->sd_mark = bp;
1794 1804 (void) putq(q, bp);
1795 1805
1796 1806 /*
1797 1807 * If message is a PCPROTO message, always use
1798 1808 * firstmsgsigs to determine if a signal should be
1799 1809 * sent as strrput is the only place to send
1800 1810 * signals for PCPROTO. Other messages are based on
1801 1811 * the STRGETINPROG flag. The flag determines if
1802 1812 * strrput or (k)strgetmsg will be responsible for
1803 1813 * sending the signals, in the firstmsgsigs case.
1804 1814 */
1805 1815 if ((hipri_sig == 1) ||
1806 1816 (((stp->sd_flag & STRGETINPROG) == 0) &&
1807 1817 (q->q_first == bp)))
1808 1818 signals = (firstmsgsigs | allmsgsigs);
1809 1819 else
1810 1820 signals = allmsgsigs;
1811 1821 break;
1812 1822
1813 1823 default:
1814 1824 mutex_exit(&stp->sd_lock);
1815 1825 (void) strrput_nondata(q, bp);
1816 1826 mutex_enter(&stp->sd_lock);
1817 1827 break;
1818 1828 }
1819 1829 }
1820 1830 ASSERT(MUTEX_HELD(&stp->sd_lock));
1821 1831 /*
1822 1832 * Wake sleeping read/getmsg and cancel deferred wakeup
1823 1833 */
1824 1834 if (wakeups & RSLEEP)
1825 1835 stp->sd_wakeq &= ~RSLEEP;
1826 1836
1827 1837 wakeups &= stp->sd_flag;
1828 1838 if (wakeups & RSLEEP) {
1829 1839 stp->sd_flag &= ~RSLEEP;
1830 1840 cv_broadcast(&q->q_wait);
1831 1841 }
1832 1842 if (wakeups & WSLEEP) {
1833 1843 stp->sd_flag &= ~WSLEEP;
1834 1844 cv_broadcast(&_WR(q)->q_wait);
1835 1845 }
1836 1846
1837 1847 if (pollwakeups != 0) {
1838 1848 if (pollwakeups == (POLLIN | POLLRDNORM)) {
1839 1849 /*
1840 1850 * Can't use rput_opt since it was not
1841 1851 * read when sd_lock was held and SR_POLLIN is changed
1842 1852 * by strpoll() under sd_lock.
1843 1853 */
1844 1854 if (!(stp->sd_rput_opt & SR_POLLIN))
1845 1855 goto no_pollwake;
1846 1856 stp->sd_rput_opt &= ~SR_POLLIN;
1847 1857 }
1848 1858 mutex_exit(&stp->sd_lock);
1849 1859 pollwakeup(&stp->sd_pollist, pollwakeups);
1850 1860 mutex_enter(&stp->sd_lock);
1851 1861 }
1852 1862 no_pollwake:
1853 1863
1854 1864 /*
1855 1865 * strsendsig can handle multiple signals with a
1856 1866 * single call.
1857 1867 */
1858 1868 if (stp->sd_sigflags & signals)
1859 1869 strsendsig(stp->sd_siglist, signals, band, 0);
1860 1870 mutex_exit(&stp->sd_lock);
1861 1871
1862 1872
1863 1873 done:
1864 1874 if (nextbp == NULL)
1865 1875 return (0);
1866 1876
1867 1877 /*
1868 1878 * Any signals were handled the first time.
1869 1879 * Wakeups and pollwakeups are redone to avoid any race
1870 1880 * conditions - all the messages are not queued until the
1871 1881 * last message has been processed by strrput.
1872 1882 */
1873 1883 bp = nextbp;
1874 1884 signals = firstmsgsigs = allmsgsigs = 0;
1875 1885 mutex_enter(&stp->sd_lock);
1876 1886 goto one_more;
1877 1887 }
1878 1888
1879 1889 static void
1880 1890 log_dupioc(queue_t *rq, mblk_t *bp)
1881 1891 {
1882 1892 queue_t *wq, *qp;
1883 1893 char *modnames, *mnp, *dname;
1884 1894 size_t maxmodstr;
1885 1895 boolean_t islast;
1886 1896
1887 1897 /*
1888 1898 * Allocate a buffer large enough to hold the names of nstrpush modules
1889 1899 * and one driver, with spaces between and NUL terminator. If we can't
1890 1900 * get memory, then we'll just log the driver name.
1891 1901 */
1892 1902 maxmodstr = nstrpush * (FMNAMESZ + 1);
1893 1903 mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1894 1904
1895 1905 /* march down write side to print log message down to the driver */
1896 1906 wq = WR(rq);
1897 1907
1898 1908 /* make sure q_next doesn't shift around while we're grabbing data */
1899 1909 claimstr(wq);
1900 1910 qp = wq->q_next;
1901 1911 do {
1902 1912 dname = Q2NAME(qp);
1903 1913 islast = !SAMESTR(qp) || qp->q_next == NULL;
1904 1914 if (modnames == NULL) {
1905 1915 /*
1906 1916 * If we don't have memory, then get the driver name in
1907 1917 * the log where we can see it. Note that memory
1908 1918 * pressure is a possible cause of these sorts of bugs.
1909 1919 */
1910 1920 if (islast) {
1911 1921 modnames = dname;
1912 1922 maxmodstr = 0;
1913 1923 }
1914 1924 } else {
1915 1925 mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1916 1926 if (!islast)
1917 1927 *mnp++ = ' ';
1918 1928 }
1919 1929 qp = qp->q_next;
1920 1930 } while (!islast);
1921 1931 releasestr(wq);
1922 1932 /* Cannot happen unless stream head is corrupt. */
1923 1933 ASSERT(modnames != NULL);
1924 1934 (void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1925 1935 SL_CONSOLE|SL_TRACE|SL_ERROR,
1926 1936 "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1927 1937 rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1928 1938 (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1929 1939 if (maxmodstr != 0)
1930 1940 kmem_free(modnames, maxmodstr);
1931 1941 }
1932 1942
1933 1943 int
1934 1944 strrput_nondata(queue_t *q, mblk_t *bp)
1935 1945 {
1936 1946 struct stdata *stp;
1937 1947 struct iocblk *iocbp;
1938 1948 struct stroptions *sop;
1939 1949 struct copyreq *reqp;
1940 1950 struct copyresp *resp;
1941 1951 unsigned char bpri;
1942 1952 unsigned char flushed_already = 0;
1943 1953
1944 1954 stp = (struct stdata *)q->q_ptr;
1945 1955
1946 1956 ASSERT(!(stp->sd_flag & STPLEX));
1947 1957 ASSERT(qclaimed(q));
1948 1958
1949 1959 switch (bp->b_datap->db_type) {
1950 1960 case M_ERROR:
1951 1961 /*
1952 1962 * An error has occurred downstream, the errno is in the first
1953 1963 * bytes of the message.
1954 1964 */
1955 1965 if ((bp->b_wptr - bp->b_rptr) == 2) { /* New flavor */
1956 1966 unsigned char rw = 0;
1957 1967
1958 1968 mutex_enter(&stp->sd_lock);
1959 1969 if (*bp->b_rptr != NOERROR) { /* read error */
1960 1970 if (*bp->b_rptr != 0) {
1961 1971 if (stp->sd_flag & STRDERR)
1962 1972 flushed_already |= FLUSHR;
1963 1973 stp->sd_flag |= STRDERR;
1964 1974 rw |= FLUSHR;
1965 1975 } else {
1966 1976 stp->sd_flag &= ~STRDERR;
1967 1977 }
1968 1978 stp->sd_rerror = *bp->b_rptr;
1969 1979 }
1970 1980 bp->b_rptr++;
1971 1981 if (*bp->b_rptr != NOERROR) { /* write error */
1972 1982 if (*bp->b_rptr != 0) {
1973 1983 if (stp->sd_flag & STWRERR)
1974 1984 flushed_already |= FLUSHW;
1975 1985 stp->sd_flag |= STWRERR;
1976 1986 rw |= FLUSHW;
1977 1987 } else {
1978 1988 stp->sd_flag &= ~STWRERR;
1979 1989 }
1980 1990 stp->sd_werror = *bp->b_rptr;
1981 1991 }
1982 1992 if (rw) {
1983 1993 TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1984 1994 "strrput cv_broadcast:q %p, bp %p",
1985 1995 q, bp);
1986 1996 cv_broadcast(&q->q_wait); /* readers */
1987 1997 cv_broadcast(&_WR(q)->q_wait); /* writers */
1988 1998 cv_broadcast(&stp->sd_monitor); /* ioctllers */
1989 1999
1990 2000 mutex_exit(&stp->sd_lock);
1991 2001 pollwakeup(&stp->sd_pollist, POLLERR);
1992 2002 mutex_enter(&stp->sd_lock);
1993 2003
1994 2004 if (stp->sd_sigflags & S_ERROR)
1995 2005 strsendsig(stp->sd_siglist, S_ERROR, 0,
1996 2006 ((rw & FLUSHR) ? stp->sd_rerror :
1997 2007 stp->sd_werror));
1998 2008 mutex_exit(&stp->sd_lock);
1999 2009 /*
2000 2010 * Send the M_FLUSH only
2001 2011 * for the first M_ERROR
2002 2012 * message on the stream
2003 2013 */
2004 2014 if (flushed_already == rw) {
2005 2015 freemsg(bp);
2006 2016 return (0);
2007 2017 }
2008 2018
2009 2019 bp->b_datap->db_type = M_FLUSH;
2010 2020 *bp->b_rptr = rw;
2011 2021 bp->b_wptr = bp->b_rptr + 1;
2012 2022 /*
2013 2023 * Protect against the driver
2014 2024 * passing up messages after
2015 2025 * it has done a qprocsoff
2016 2026 */
2017 2027 if (_OTHERQ(q)->q_next == NULL)
2018 2028 freemsg(bp);
2019 2029 else
2020 2030 qreply(q, bp);
2021 2031 return (0);
2022 2032 } else
2023 2033 mutex_exit(&stp->sd_lock);
2024 2034 } else if (*bp->b_rptr != 0) { /* Old flavor */
2025 2035 if (stp->sd_flag & (STRDERR|STWRERR))
2026 2036 flushed_already = FLUSHRW;
2027 2037 mutex_enter(&stp->sd_lock);
2028 2038 stp->sd_flag |= (STRDERR|STWRERR);
2029 2039 stp->sd_rerror = *bp->b_rptr;
2030 2040 stp->sd_werror = *bp->b_rptr;
2031 2041 TRACE_2(TR_FAC_STREAMS_FR,
2032 2042 TR_STRRPUT_WAKE2,
2033 2043 "strrput wakeup #2:q %p, bp %p", q, bp);
2034 2044 cv_broadcast(&q->q_wait); /* the readers */
2035 2045 cv_broadcast(&_WR(q)->q_wait); /* the writers */
2036 2046 cv_broadcast(&stp->sd_monitor); /* ioctllers */
2037 2047
2038 2048 mutex_exit(&stp->sd_lock);
2039 2049 pollwakeup(&stp->sd_pollist, POLLERR);
2040 2050 mutex_enter(&stp->sd_lock);
2041 2051
2042 2052 if (stp->sd_sigflags & S_ERROR)
2043 2053 strsendsig(stp->sd_siglist, S_ERROR, 0,
2044 2054 (stp->sd_werror ? stp->sd_werror :
2045 2055 stp->sd_rerror));
2046 2056 mutex_exit(&stp->sd_lock);
2047 2057
2048 2058 /*
2049 2059 * Send the M_FLUSH only
2050 2060 * for the first M_ERROR
2051 2061 * message on the stream
2052 2062 */
2053 2063 if (flushed_already != FLUSHRW) {
2054 2064 bp->b_datap->db_type = M_FLUSH;
2055 2065 *bp->b_rptr = FLUSHRW;
2056 2066 /*
2057 2067 * Protect against the driver passing up
2058 2068 * messages after it has done a
2059 2069 * qprocsoff.
2060 2070 */
2061 2071 if (_OTHERQ(q)->q_next == NULL)
2062 2072 freemsg(bp);
2063 2073 else
2064 2074 qreply(q, bp);
2065 2075 return (0);
2066 2076 }
2067 2077 }
2068 2078 freemsg(bp);
2069 2079 return (0);
2070 2080
2071 2081 case M_HANGUP:
2072 2082
2073 2083 freemsg(bp);
2074 2084 mutex_enter(&stp->sd_lock);
2075 2085 stp->sd_werror = ENXIO;
2076 2086 stp->sd_flag |= STRHUP;
2077 2087 stp->sd_flag &= ~(WSLEEP|RSLEEP);
2078 2088
2079 2089 /*
2080 2090 * send signal if controlling tty
2081 2091 */
2082 2092
2083 2093 if (stp->sd_sidp) {
2084 2094 prsignal(stp->sd_sidp, SIGHUP);
2085 2095 if (stp->sd_sidp != stp->sd_pgidp)
2086 2096 pgsignal(stp->sd_pgidp, SIGTSTP);
2087 2097 }
2088 2098
2089 2099 /*
2090 2100 * wake up read, write, and exception pollers and
2091 2101 * reset wakeup mechanism.
2092 2102 */
2093 2103 cv_broadcast(&q->q_wait); /* the readers */
2094 2104 cv_broadcast(&_WR(q)->q_wait); /* the writers */
2095 2105 cv_broadcast(&stp->sd_monitor); /* the ioctllers */
2096 2106 strhup(stp);
2097 2107 mutex_exit(&stp->sd_lock);
2098 2108 return (0);
2099 2109
2100 2110 case M_UNHANGUP:
2101 2111 freemsg(bp);
2102 2112 mutex_enter(&stp->sd_lock);
2103 2113 stp->sd_werror = 0;
2104 2114 stp->sd_flag &= ~STRHUP;
2105 2115 mutex_exit(&stp->sd_lock);
2106 2116 return (0);
2107 2117
2108 2118 case M_SIG:
2109 2119 /*
2110 2120 * Someone downstream wants to post a signal. The
2111 2121 * signal to post is contained in the first byte of the
2112 2122 * message. If the message would go on the front of
2113 2123 * the queue, send a signal to the process group
2114 2124 * (if not SIGPOLL) or to the siglist processes
2115 2125 * (SIGPOLL). If something is already on the queue,
2116 2126 * OR if we are delivering a delayed suspend (*sigh*
2117 2127 * another "tty" hack) and there's no one sleeping already,
2118 2128 * just enqueue the message.
2119 2129 */
2120 2130 mutex_enter(&stp->sd_lock);
2121 2131 if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2122 2132 !(stp->sd_flag & RSLEEP))) {
2123 2133 (void) putq(q, bp);
2124 2134 mutex_exit(&stp->sd_lock);
2125 2135 return (0);
2126 2136 }
2127 2137 mutex_exit(&stp->sd_lock);
2128 2138 /* FALLTHRU */
2129 2139
2130 2140 case M_PCSIG:
2131 2141 /*
2132 2142 * Don't enqueue, just post the signal.
2133 2143 */
2134 2144 strsignal(stp, *bp->b_rptr, 0L);
2135 2145 freemsg(bp);
2136 2146 return (0);
2137 2147
2138 2148 case M_CMD:
2139 2149 if (MBLKL(bp) != sizeof (cmdblk_t)) {
2140 2150 freemsg(bp);
2141 2151 return (0);
2142 2152 }
2143 2153
2144 2154 mutex_enter(&stp->sd_lock);
2145 2155 if (stp->sd_flag & STRCMDWAIT) {
2146 2156 ASSERT(stp->sd_cmdblk == NULL);
2147 2157 stp->sd_cmdblk = bp;
2148 2158 cv_broadcast(&stp->sd_monitor);
2149 2159 mutex_exit(&stp->sd_lock);
2150 2160 } else {
2151 2161 mutex_exit(&stp->sd_lock);
2152 2162 freemsg(bp);
2153 2163 }
2154 2164 return (0);
2155 2165
2156 2166 case M_FLUSH:
2157 2167 /*
2158 2168 * Flush queues. The indication of which queues to flush
2159 2169 * is in the first byte of the message. If the read queue
2160 2170 * is specified, then flush it. If FLUSHBAND is set, just
2161 2171 * flush the band specified by the second byte of the message.
2162 2172 *
2163 2173 * If a module has issued a M_SETOPT to not flush hi
2164 2174 * priority messages off of the stream head, then pass this
2165 2175 * flag into the flushq code to preserve such messages.
2166 2176 */
2167 2177
2168 2178 if (*bp->b_rptr & FLUSHR) {
2169 2179 mutex_enter(&stp->sd_lock);
2170 2180 if (*bp->b_rptr & FLUSHBAND) {
2171 2181 ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2172 2182 flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2173 2183 } else
2174 2184 flushq_common(q, FLUSHALL,
2175 2185 stp->sd_read_opt & RFLUSHPCPROT);
2176 2186 if ((q->q_first == NULL) ||
2177 2187 (q->q_first->b_datap->db_type < QPCTL))
2178 2188 stp->sd_flag &= ~STRPRI;
2179 2189 else {
2180 2190 ASSERT(stp->sd_flag & STRPRI);
2181 2191 }
2182 2192 mutex_exit(&stp->sd_lock);
2183 2193 }
2184 2194 if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2185 2195 *bp->b_rptr &= ~FLUSHR;
2186 2196 bp->b_flag |= MSGNOLOOP;
2187 2197 /*
2188 2198 * Protect against the driver passing up
2189 2199 * messages after it has done a qprocsoff.
2190 2200 */
2191 2201 if (_OTHERQ(q)->q_next == NULL)
2192 2202 freemsg(bp);
2193 2203 else
2194 2204 qreply(q, bp);
2195 2205 return (0);
2196 2206 }
2197 2207 freemsg(bp);
2198 2208 return (0);
2199 2209
2200 2210 case M_IOCACK:
2201 2211 case M_IOCNAK:
2202 2212 iocbp = (struct iocblk *)bp->b_rptr;
2203 2213 /*
2204 2214 * If not waiting for ACK or NAK then just free msg.
2205 2215 * If incorrect id sequence number then just free msg.
2206 2216 * If already have ACK or NAK for user then this is a
2207 2217 * duplicate, display a warning and free the msg.
2208 2218 */
2209 2219 mutex_enter(&stp->sd_lock);
2210 2220 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2211 2221 (stp->sd_iocid != iocbp->ioc_id)) {
2212 2222 /*
2213 2223 * If the ACK/NAK is a dup, display a message
2214 2224 * Dup is when sd_iocid == ioc_id, and
2215 2225 * sd_iocblk == <valid ptr> or -1 (the former
2216 2226 * is when an ioctl has been put on the stream
2217 2227 * head, but has not yet been consumed, the
2218 2228 * later is when it has been consumed).
2219 2229 */
2220 2230 if ((stp->sd_iocid == iocbp->ioc_id) &&
2221 2231 (stp->sd_iocblk != NULL)) {
2222 2232 log_dupioc(q, bp);
2223 2233 }
2224 2234 freemsg(bp);
2225 2235 mutex_exit(&stp->sd_lock);
2226 2236 return (0);
2227 2237 }
2228 2238
2229 2239 /*
2230 2240 * Assign ACK or NAK to user and wake up.
2231 2241 */
2232 2242 stp->sd_iocblk = bp;
2233 2243 cv_broadcast(&stp->sd_monitor);
2234 2244 mutex_exit(&stp->sd_lock);
2235 2245 return (0);
2236 2246
2237 2247 case M_COPYIN:
2238 2248 case M_COPYOUT:
2239 2249 reqp = (struct copyreq *)bp->b_rptr;
2240 2250
2241 2251 /*
2242 2252 * If not waiting for ACK or NAK then just fail request.
2243 2253 * If already have ACK, NAK, or copy request, then just
2244 2254 * fail request.
2245 2255 * If incorrect id sequence number then just fail request.
2246 2256 */
2247 2257 mutex_enter(&stp->sd_lock);
2248 2258 if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2249 2259 (stp->sd_iocid != reqp->cq_id)) {
2250 2260 if (bp->b_cont) {
2251 2261 freemsg(bp->b_cont);
2252 2262 bp->b_cont = NULL;
2253 2263 }
2254 2264 bp->b_datap->db_type = M_IOCDATA;
2255 2265 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2256 2266 resp = (struct copyresp *)bp->b_rptr;
2257 2267 resp->cp_rval = (caddr_t)1; /* failure */
2258 2268 mutex_exit(&stp->sd_lock);
2259 2269 putnext(stp->sd_wrq, bp);
2260 2270 return (0);
2261 2271 }
2262 2272
2263 2273 /*
2264 2274 * Assign copy request to user and wake up.
2265 2275 */
2266 2276 stp->sd_iocblk = bp;
2267 2277 cv_broadcast(&stp->sd_monitor);
2268 2278 mutex_exit(&stp->sd_lock);
2269 2279 return (0);
2270 2280
2271 2281 case M_SETOPTS:
2272 2282 /*
2273 2283 * Set stream head options (read option, write offset,
2274 2284 * min/max packet size, and/or high/low water marks for
2275 2285 * the read side only).
2276 2286 */
2277 2287
2278 2288 bpri = 0;
2279 2289 sop = (struct stroptions *)bp->b_rptr;
2280 2290 mutex_enter(&stp->sd_lock);
2281 2291 if (sop->so_flags & SO_READOPT) {
2282 2292 switch (sop->so_readopt & RMODEMASK) {
2283 2293 case RNORM:
2284 2294 stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2285 2295 break;
2286 2296
2287 2297 case RMSGD:
2288 2298 stp->sd_read_opt =
2289 2299 ((stp->sd_read_opt & ~RD_MSGNODIS) |
2290 2300 RD_MSGDIS);
2291 2301 break;
2292 2302
2293 2303 case RMSGN:
2294 2304 stp->sd_read_opt =
2295 2305 ((stp->sd_read_opt & ~RD_MSGDIS) |
2296 2306 RD_MSGNODIS);
2297 2307 break;
2298 2308 }
2299 2309 switch (sop->so_readopt & RPROTMASK) {
2300 2310 case RPROTNORM:
2301 2311 stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2302 2312 break;
2303 2313
2304 2314 case RPROTDAT:
2305 2315 stp->sd_read_opt =
2306 2316 ((stp->sd_read_opt & ~RD_PROTDIS) |
2307 2317 RD_PROTDAT);
2308 2318 break;
2309 2319
2310 2320 case RPROTDIS:
2311 2321 stp->sd_read_opt =
2312 2322 ((stp->sd_read_opt & ~RD_PROTDAT) |
2313 2323 RD_PROTDIS);
2314 2324 break;
2315 2325 }
2316 2326 switch (sop->so_readopt & RFLUSHMASK) {
2317 2327 case RFLUSHPCPROT:
2318 2328 /*
2319 2329 * This sets the stream head to NOT flush
2320 2330 * M_PCPROTO messages.
2321 2331 */
2322 2332 stp->sd_read_opt |= RFLUSHPCPROT;
2323 2333 break;
2324 2334 }
2325 2335 }
2326 2336 if (sop->so_flags & SO_ERROPT) {
2327 2337 switch (sop->so_erropt & RERRMASK) {
2328 2338 case RERRNORM:
2329 2339 stp->sd_flag &= ~STRDERRNONPERSIST;
2330 2340 break;
2331 2341 case RERRNONPERSIST:
2332 2342 stp->sd_flag |= STRDERRNONPERSIST;
2333 2343 break;
2334 2344 }
2335 2345 switch (sop->so_erropt & WERRMASK) {
2336 2346 case WERRNORM:
2337 2347 stp->sd_flag &= ~STWRERRNONPERSIST;
2338 2348 break;
2339 2349 case WERRNONPERSIST:
2340 2350 stp->sd_flag |= STWRERRNONPERSIST;
2341 2351 break;
2342 2352 }
2343 2353 }
2344 2354 if (sop->so_flags & SO_COPYOPT) {
2345 2355 if (sop->so_copyopt & ZCVMSAFE) {
2346 2356 stp->sd_copyflag |= STZCVMSAFE;
2347 2357 stp->sd_copyflag &= ~STZCVMUNSAFE;
2348 2358 } else if (sop->so_copyopt & ZCVMUNSAFE) {
2349 2359 stp->sd_copyflag |= STZCVMUNSAFE;
2350 2360 stp->sd_copyflag &= ~STZCVMSAFE;
2351 2361 }
2352 2362
2353 2363 if (sop->so_copyopt & COPYCACHED) {
2354 2364 stp->sd_copyflag |= STRCOPYCACHED;
2355 2365 }
2356 2366 }
2357 2367 if (sop->so_flags & SO_WROFF)
2358 2368 stp->sd_wroff = sop->so_wroff;
2359 2369 if (sop->so_flags & SO_TAIL)
2360 2370 stp->sd_tail = sop->so_tail;
2361 2371 if (sop->so_flags & SO_MINPSZ)
2362 2372 q->q_minpsz = sop->so_minpsz;
2363 2373 if (sop->so_flags & SO_MAXPSZ)
2364 2374 q->q_maxpsz = sop->so_maxpsz;
2365 2375 if (sop->so_flags & SO_MAXBLK)
2366 2376 stp->sd_maxblk = sop->so_maxblk;
2367 2377 if (sop->so_flags & SO_HIWAT) {
2368 2378 if (sop->so_flags & SO_BAND) {
2369 2379 if (strqset(q, QHIWAT,
2370 2380 sop->so_band, sop->so_hiwat)) {
2371 2381 cmn_err(CE_WARN, "strrput: could not "
2372 2382 "allocate qband\n");
2373 2383 } else {
2374 2384 bpri = sop->so_band;
2375 2385 }
2376 2386 } else {
2377 2387 q->q_hiwat = sop->so_hiwat;
2378 2388 }
2379 2389 }
2380 2390 if (sop->so_flags & SO_LOWAT) {
2381 2391 if (sop->so_flags & SO_BAND) {
2382 2392 if (strqset(q, QLOWAT,
2383 2393 sop->so_band, sop->so_lowat)) {
2384 2394 cmn_err(CE_WARN, "strrput: could not "
2385 2395 "allocate qband\n");
2386 2396 } else {
2387 2397 bpri = sop->so_band;
2388 2398 }
2389 2399 } else {
2390 2400 q->q_lowat = sop->so_lowat;
2391 2401 }
2392 2402 }
2393 2403 if (sop->so_flags & SO_MREADON)
2394 2404 stp->sd_flag |= SNDMREAD;
2395 2405 if (sop->so_flags & SO_MREADOFF)
2396 2406 stp->sd_flag &= ~SNDMREAD;
2397 2407 if (sop->so_flags & SO_NDELON)
2398 2408 stp->sd_flag |= OLDNDELAY;
2399 2409 if (sop->so_flags & SO_NDELOFF)
2400 2410 stp->sd_flag &= ~OLDNDELAY;
2401 2411 if (sop->so_flags & SO_ISTTY)
2402 2412 stp->sd_flag |= STRISTTY;
2403 2413 if (sop->so_flags & SO_ISNTTY)
2404 2414 stp->sd_flag &= ~STRISTTY;
2405 2415 if (sop->so_flags & SO_TOSTOP)
2406 2416 stp->sd_flag |= STRTOSTOP;
2407 2417 if (sop->so_flags & SO_TONSTOP)
2408 2418 stp->sd_flag &= ~STRTOSTOP;
2409 2419 if (sop->so_flags & SO_DELIM)
2410 2420 stp->sd_flag |= STRDELIM;
2411 2421 if (sop->so_flags & SO_NODELIM)
2412 2422 stp->sd_flag &= ~STRDELIM;
2413 2423
2414 2424 mutex_exit(&stp->sd_lock);
2415 2425 freemsg(bp);
2416 2426
2417 2427 /* Check backenable in case the water marks changed */
2418 2428 qbackenable(q, bpri);
2419 2429 return (0);
2420 2430
2421 2431 /*
2422 2432 * The following set of cases deal with situations where two stream
2423 2433 * heads are connected to each other (twisted streams). These messages
2424 2434 * have no meaning at the stream head.
2425 2435 */
2426 2436 case M_BREAK:
2427 2437 case M_CTL:
2428 2438 case M_DELAY:
2429 2439 case M_START:
2430 2440 case M_STOP:
2431 2441 case M_IOCDATA:
2432 2442 case M_STARTI:
2433 2443 case M_STOPI:
2434 2444 freemsg(bp);
2435 2445 return (0);
2436 2446
2437 2447 case M_IOCTL:
2438 2448 /*
2439 2449 * Always NAK this condition
2440 2450 * (makes no sense)
2441 2451 * If there is one or more threads in the read side
2442 2452 * rwnext we have to defer the nacking until that thread
2443 2453 * returns (in strget).
2444 2454 */
2445 2455 mutex_enter(&stp->sd_lock);
2446 2456 if (stp->sd_struiodnak != 0) {
2447 2457 /*
2448 2458 * Defer NAK to the streamhead. Queue at the end
2449 2459 * the list.
2450 2460 */
2451 2461 mblk_t *mp = stp->sd_struionak;
2452 2462
2453 2463 while (mp && mp->b_next)
2454 2464 mp = mp->b_next;
2455 2465 if (mp)
2456 2466 mp->b_next = bp;
2457 2467 else
2458 2468 stp->sd_struionak = bp;
2459 2469 bp->b_next = NULL;
2460 2470 mutex_exit(&stp->sd_lock);
2461 2471 return (0);
2462 2472 }
2463 2473 mutex_exit(&stp->sd_lock);
2464 2474
2465 2475 bp->b_datap->db_type = M_IOCNAK;
2466 2476 /*
2467 2477 * Protect against the driver passing up
2468 2478 * messages after it has done a qprocsoff.
2469 2479 */
2470 2480 if (_OTHERQ(q)->q_next == NULL)
2471 2481 freemsg(bp);
2472 2482 else
2473 2483 qreply(q, bp);
2474 2484 return (0);
2475 2485
2476 2486 default:
2477 2487 #ifdef DEBUG
2478 2488 cmn_err(CE_WARN,
2479 2489 "bad message type %x received at stream head\n",
2480 2490 bp->b_datap->db_type);
2481 2491 #endif
2482 2492 freemsg(bp);
2483 2493 return (0);
2484 2494 }
2485 2495
2486 2496 /* NOTREACHED */
2487 2497 }
2488 2498
2489 2499 /*
2490 2500 * Check if the stream pointed to by `stp' can be written to, and return an
2491 2501 * error code if not. If `eiohup' is set, then return EIO if STRHUP is set.
2492 2502 * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2493 2503 * then always return EPIPE and send a SIGPIPE to the invoking thread.
2494 2504 */
2495 2505 static int
2496 2506 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2497 2507 {
2498 2508 int error;
2499 2509
2500 2510 ASSERT(MUTEX_HELD(&stp->sd_lock));
2501 2511
2502 2512 /*
2503 2513 * For modem support, POSIX states that on writes, EIO should
2504 2514 * be returned if the stream has been hung up.
2505 2515 */
2506 2516 if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2507 2517 error = EIO;
2508 2518 else
2509 2519 error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2510 2520
2511 2521 if (error != 0) {
2512 2522 if (!(stp->sd_flag & STPLEX) &&
2513 2523 (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2514 2524 tsignal(curthread, SIGPIPE);
2515 2525 error = EPIPE;
2516 2526 }
2517 2527 }
2518 2528
2519 2529 return (error);
2520 2530 }
2521 2531
2522 2532 /*
2523 2533 * Copyin and send data down a stream.
2524 2534 * The caller will allocate and copyin any control part that precedes the
2525 2535 * message and pass that in as mctl.
2526 2536 *
2527 2537 * Caller should *not* hold sd_lock.
2528 2538 * When EWOULDBLOCK is returned the caller has to redo the canputnext
2529 2539 * under sd_lock in order to avoid missing a backenabling wakeup.
2530 2540 *
2531 2541 * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2532 2542 *
2533 2543 * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2534 2544 * For sync streams we can only ignore flow control by reverting to using
2535 2545 * putnext.
2536 2546 *
2537 2547 * If sd_maxblk is less than *iosize this routine might return without
2538 2548 * transferring all of *iosize. In all cases, on return *iosize will contain
2539 2549 * the amount of data that was transferred.
2540 2550 */
2541 2551 static int
2542 2552 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2543 2553 int b_flag, int pri, int flags)
2544 2554 {
2545 2555 struiod_t uiod;
2546 2556 mblk_t *mp;
2547 2557 queue_t *wqp = stp->sd_wrq;
2548 2558 int error = 0;
2549 2559 ssize_t count = *iosize;
2550 2560
2551 2561 ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2552 2562
2553 2563 if (uiop != NULL && count >= 0)
2554 2564 flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2555 2565
2556 2566 if (!(flags & STRUIO_POSTPONE)) {
2557 2567 /*
2558 2568 * Use regular canputnext, strmakedata, putnext sequence.
2559 2569 */
2560 2570 if (pri == 0) {
2561 2571 if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2562 2572 freemsg(mctl);
2563 2573 return (EWOULDBLOCK);
2564 2574 }
2565 2575 } else {
2566 2576 if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2567 2577 freemsg(mctl);
2568 2578 return (EWOULDBLOCK);
2569 2579 }
2570 2580 }
2571 2581
2572 2582 if ((error = strmakedata(iosize, uiop, stp, flags,
2573 2583 &mp)) != 0) {
2574 2584 freemsg(mctl);
2575 2585 /*
2576 2586 * need to change return code to ENOMEM
2577 2587 * so that this is not confused with
2578 2588 * flow control, EAGAIN.
2579 2589 */
2580 2590
2581 2591 if (error == EAGAIN)
2582 2592 return (ENOMEM);
2583 2593 else
2584 2594 return (error);
2585 2595 }
2586 2596 if (mctl != NULL) {
2587 2597 if (mctl->b_cont == NULL)
2588 2598 mctl->b_cont = mp;
2589 2599 else if (mp != NULL)
2590 2600 linkb(mctl, mp);
2591 2601 mp = mctl;
2592 2602 } else if (mp == NULL)
2593 2603 return (0);
2594 2604
2595 2605 mp->b_flag |= b_flag;
2596 2606 mp->b_band = (uchar_t)pri;
2597 2607
2598 2608 if (flags & MSG_IGNFLOW) {
2599 2609 /*
2600 2610 * XXX Hack: Don't get stuck running service
2601 2611 * procedures. This is needed for sockfs when
2602 2612 * sending the unbind message out of the rput
2603 2613 * procedure - we don't want a put procedure
2604 2614 * to run service procedures.
2605 2615 */
2606 2616 putnext(wqp, mp);
2607 2617 } else {
2608 2618 stream_willservice(stp);
2609 2619 putnext(wqp, mp);
2610 2620 stream_runservice(stp);
2611 2621 }
2612 2622 return (0);
2613 2623 }
2614 2624 /*
2615 2625 * Stream supports rwnext() for the write side.
2616 2626 */
2617 2627 if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2618 2628 freemsg(mctl);
2619 2629 /*
2620 2630 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2621 2631 */
2622 2632 return (error == EAGAIN ? ENOMEM : error);
2623 2633 }
2624 2634 if (mctl != NULL) {
2625 2635 if (mctl->b_cont == NULL)
2626 2636 mctl->b_cont = mp;
2627 2637 else if (mp != NULL)
2628 2638 linkb(mctl, mp);
2629 2639 mp = mctl;
2630 2640 } else if (mp == NULL) {
2631 2641 return (0);
2632 2642 }
2633 2643
2634 2644 mp->b_flag |= b_flag;
2635 2645 mp->b_band = (uchar_t)pri;
2636 2646
2637 2647 (void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2638 2648 sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2639 2649 uiod.d_uio.uio_offset = 0;
2640 2650 uiod.d_mp = mp;
2641 2651 error = rwnext(wqp, &uiod);
2642 2652 if (! uiod.d_mp) {
2643 2653 uioskip(uiop, *iosize);
2644 2654 return (error);
2645 2655 }
2646 2656 ASSERT(mp == uiod.d_mp);
2647 2657 if (error == EINVAL) {
2648 2658 /*
2649 2659 * The stream plumbing must have changed while
2650 2660 * we were away, so just turn off rwnext()s.
2651 2661 */
2652 2662 error = 0;
2653 2663 } else if (error == EBUSY || error == EWOULDBLOCK) {
2654 2664 /*
2655 2665 * Couldn't enter a perimeter or took a page fault,
2656 2666 * so fall-back to putnext().
2657 2667 */
2658 2668 error = 0;
2659 2669 } else {
2660 2670 freemsg(mp);
2661 2671 return (error);
2662 2672 }
2663 2673 /* Have to check canput before consuming data from the uio */
2664 2674 if (pri == 0) {
2665 2675 if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2666 2676 freemsg(mp);
2667 2677 return (EWOULDBLOCK);
2668 2678 }
2669 2679 } else {
2670 2680 if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2671 2681 freemsg(mp);
2672 2682 return (EWOULDBLOCK);
2673 2683 }
2674 2684 }
2675 2685 ASSERT(mp == uiod.d_mp);
2676 2686 /* Copyin data from the uio */
2677 2687 if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2678 2688 freemsg(mp);
2679 2689 return (error);
2680 2690 }
2681 2691 uioskip(uiop, *iosize);
2682 2692 if (flags & MSG_IGNFLOW) {
2683 2693 /*
2684 2694 * XXX Hack: Don't get stuck running service procedures.
2685 2695 * This is needed for sockfs when sending the unbind message
2686 2696 * out of the rput procedure - we don't want a put procedure
2687 2697 * to run service procedures.
2688 2698 */
2689 2699 putnext(wqp, mp);
2690 2700 } else {
2691 2701 stream_willservice(stp);
2692 2702 putnext(wqp, mp);
2693 2703 stream_runservice(stp);
2694 2704 }
2695 2705 return (0);
2696 2706 }
2697 2707
2698 2708 /*
2699 2709 * Write attempts to break the write request into messages conforming
2700 2710 * with the minimum and maximum packet sizes set downstream.
2701 2711 *
2702 2712 * Write will not block if downstream queue is full and
2703 2713 * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2704 2714 *
2705 2715 * A write of zero bytes gets packaged into a zero length message and sent
2706 2716 * downstream like any other message.
2707 2717 *
2708 2718 * If buffers of the requested sizes are not available, the write will
2709 2719 * sleep until the buffers become available.
2710 2720 *
2711 2721 * Write (if specified) will supply a write offset in a message if it
2712 2722 * makes sense. This can be specified by downstream modules as part of
2713 2723 * a M_SETOPTS message. Write will not supply the write offset if it
2714 2724 * cannot supply any data in a buffer. In other words, write will never
2715 2725 * send down an empty packet due to a write offset.
2716 2726 */
2717 2727 /* ARGSUSED2 */
2718 2728 int
2719 2729 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2720 2730 {
2721 2731 return (strwrite_common(vp, uiop, crp, 0));
2722 2732 }
2723 2733
2724 2734 /* ARGSUSED2 */
2725 2735 int
2726 2736 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2727 2737 {
2728 2738 struct stdata *stp;
2729 2739 struct queue *wqp;
2730 2740 ssize_t rmin, rmax;
2731 2741 ssize_t iosize;
2732 2742 int waitflag;
2733 2743 int tempmode;
2734 2744 int error = 0;
2735 2745 int b_flag;
2736 2746
2737 2747 ASSERT(vp->v_stream);
2738 2748 stp = vp->v_stream;
2739 2749
2740 2750 mutex_enter(&stp->sd_lock);
2741 2751
2742 2752 if ((error = i_straccess(stp, JCWRITE)) != 0) {
2743 2753 mutex_exit(&stp->sd_lock);
2744 2754 return (error);
2745 2755 }
2746 2756
2747 2757 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2748 2758 error = strwriteable(stp, B_TRUE, B_TRUE);
2749 2759 if (error != 0) {
2750 2760 mutex_exit(&stp->sd_lock);
2751 2761 return (error);
2752 2762 }
2753 2763 }
2754 2764
2755 2765 mutex_exit(&stp->sd_lock);
2756 2766
2757 2767 wqp = stp->sd_wrq;
2758 2768
2759 2769 /* get these values from them cached in the stream head */
2760 2770 rmin = stp->sd_qn_minpsz;
2761 2771 rmax = stp->sd_qn_maxpsz;
2762 2772
2763 2773 /*
2764 2774 * Check the min/max packet size constraints. If min packet size
2765 2775 * is non-zero, the write cannot be split into multiple messages
2766 2776 * and still guarantee the size constraints.
2767 2777 */
2768 2778 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2769 2779
2770 2780 ASSERT((rmax >= 0) || (rmax == INFPSZ));
2771 2781 if (rmax == 0) {
2772 2782 return (0);
2773 2783 }
2774 2784 if (rmin > 0) {
2775 2785 if (uiop->uio_resid < rmin) {
2776 2786 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2777 2787 "strwrite out:q %p out %d error %d",
2778 2788 wqp, 0, ERANGE);
2779 2789 return (ERANGE);
2780 2790 }
2781 2791 if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2782 2792 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2783 2793 "strwrite out:q %p out %d error %d",
2784 2794 wqp, 1, ERANGE);
2785 2795 return (ERANGE);
2786 2796 }
2787 2797 }
2788 2798
2789 2799 /*
2790 2800 * Do until count satisfied or error.
2791 2801 */
2792 2802 waitflag = WRITEWAIT | wflag;
2793 2803 if (stp->sd_flag & OLDNDELAY)
2794 2804 tempmode = uiop->uio_fmode & ~FNDELAY;
2795 2805 else
2796 2806 tempmode = uiop->uio_fmode;
2797 2807
2798 2808 if (rmax == INFPSZ)
2799 2809 rmax = uiop->uio_resid;
2800 2810
2801 2811 /*
2802 2812 * Note that tempmode does not get used in strput/strmakedata
2803 2813 * but only in strwaitq. The other routines use uio_fmode
2804 2814 * unmodified.
2805 2815 */
2806 2816
2807 2817 /* LINTED: constant in conditional context */
2808 2818 while (1) { /* breaks when uio_resid reaches zero */
2809 2819 /*
2810 2820 * Determine the size of the next message to be
2811 2821 * packaged. May have to break write into several
2812 2822 * messages based on max packet size.
2813 2823 */
2814 2824 iosize = MIN(uiop->uio_resid, rmax);
2815 2825
2816 2826 /*
2817 2827 * Put block downstream when flow control allows it.
2818 2828 */
2819 2829 if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2820 2830 b_flag = MSGDELIM;
2821 2831 else
2822 2832 b_flag = 0;
2823 2833
2824 2834 for (;;) {
2825 2835 int done = 0;
2826 2836
2827 2837 error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2828 2838 if (error == 0)
2829 2839 break;
2830 2840 if (error != EWOULDBLOCK)
2831 2841 goto out;
2832 2842
2833 2843 mutex_enter(&stp->sd_lock);
2834 2844 /*
2835 2845 * Check for a missed wakeup.
2836 2846 * Needed since strput did not hold sd_lock across
2837 2847 * the canputnext.
2838 2848 */
2839 2849 if (canputnext(wqp)) {
2840 2850 /* Try again */
2841 2851 mutex_exit(&stp->sd_lock);
2842 2852 continue;
2843 2853 }
2844 2854 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2845 2855 "strwrite wait:q %p wait", wqp);
2846 2856 if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2847 2857 tempmode, -1, &done)) != 0 || done) {
2848 2858 mutex_exit(&stp->sd_lock);
2849 2859 if ((vp->v_type == VFIFO) &&
2850 2860 (uiop->uio_fmode & FNDELAY) &&
2851 2861 (error == EAGAIN))
2852 2862 error = 0;
2853 2863 goto out;
2854 2864 }
2855 2865 TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2856 2866 "strwrite wake:q %p awakes", wqp);
2857 2867 if ((error = i_straccess(stp, JCWRITE)) != 0) {
2858 2868 mutex_exit(&stp->sd_lock);
2859 2869 goto out;
2860 2870 }
2861 2871 mutex_exit(&stp->sd_lock);
2862 2872 }
2863 2873 waitflag |= NOINTR;
2864 2874 TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2865 2875 "strwrite resid:q %p uiop %p", wqp, uiop);
2866 2876 if (uiop->uio_resid) {
2867 2877 /* Recheck for errors - needed for sockets */
2868 2878 if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2869 2879 (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2870 2880 mutex_enter(&stp->sd_lock);
2871 2881 error = strwriteable(stp, B_FALSE, B_TRUE);
2872 2882 mutex_exit(&stp->sd_lock);
2873 2883 if (error != 0)
2874 2884 return (error);
2875 2885 }
2876 2886 continue;
2877 2887 }
2878 2888 break;
2879 2889 }
2880 2890 out:
2881 2891 /*
2882 2892 * For historical reasons, applications expect EAGAIN when a data
2883 2893 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2884 2894 */
2885 2895 if (error == ENOMEM)
2886 2896 error = EAGAIN;
2887 2897 TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2888 2898 "strwrite out:q %p out %d error %d", wqp, 2, error);
2889 2899 return (error);
2890 2900 }
2891 2901
2892 2902 /*
2893 2903 * Stream head write service routine.
2894 2904 * Its job is to wake up any sleeping writers when a queue
2895 2905 * downstream needs data (part of the flow control in putq and getq).
2896 2906 * It also must wake anyone sleeping on a poll().
2897 2907 * For stream head right below mux module, it must also invoke put procedure
2898 2908 * of next downstream module.
2899 2909 */
2900 2910 int
2901 2911 strwsrv(queue_t *q)
2902 2912 {
2903 2913 struct stdata *stp;
2904 2914 queue_t *tq;
2905 2915 qband_t *qbp;
2906 2916 int i;
2907 2917 qband_t *myqbp;
2908 2918 int isevent;
2909 2919 unsigned char qbf[NBAND]; /* band flushing backenable flags */
2910 2920
2911 2921 TRACE_1(TR_FAC_STREAMS_FR,
2912 2922 TR_STRWSRV, "strwsrv:q %p", q);
2913 2923 stp = (struct stdata *)q->q_ptr;
2914 2924 ASSERT(qclaimed(q));
2915 2925 mutex_enter(&stp->sd_lock);
2916 2926 ASSERT(!(stp->sd_flag & STPLEX));
2917 2927
2918 2928 if (stp->sd_flag & WSLEEP) {
2919 2929 stp->sd_flag &= ~WSLEEP;
2920 2930 cv_broadcast(&q->q_wait);
2921 2931 }
2922 2932 mutex_exit(&stp->sd_lock);
2923 2933
2924 2934 /* The other end of a stream pipe went away. */
2925 2935 if ((tq = q->q_next) == NULL) {
2926 2936 return (0);
2927 2937 }
2928 2938
2929 2939 /* Find the next module forward that has a service procedure */
2930 2940 claimstr(q);
2931 2941 tq = q->q_nfsrv;
2932 2942 ASSERT(tq != NULL);
2933 2943
2934 2944 if ((q->q_flag & QBACK)) {
2935 2945 if ((tq->q_flag & QFULL)) {
2936 2946 mutex_enter(QLOCK(tq));
2937 2947 if (!(tq->q_flag & QFULL)) {
2938 2948 mutex_exit(QLOCK(tq));
2939 2949 goto wakeup;
2940 2950 }
2941 2951 /*
2942 2952 * The queue must have become full again. Set QWANTW
2943 2953 * again so strwsrv will be back enabled when
2944 2954 * the queue becomes non-full next time.
2945 2955 */
2946 2956 tq->q_flag |= QWANTW;
2947 2957 mutex_exit(QLOCK(tq));
2948 2958 } else {
2949 2959 wakeup:
2950 2960 pollwakeup(&stp->sd_pollist, POLLWRNORM);
2951 2961 mutex_enter(&stp->sd_lock);
2952 2962 if (stp->sd_sigflags & S_WRNORM)
2953 2963 strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2954 2964 mutex_exit(&stp->sd_lock);
2955 2965 }
2956 2966 }
2957 2967
2958 2968 isevent = 0;
2959 2969 i = 1;
2960 2970 bzero((caddr_t)qbf, NBAND);
2961 2971 mutex_enter(QLOCK(tq));
2962 2972 if ((myqbp = q->q_bandp) != NULL)
2963 2973 for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2964 2974 ASSERT(myqbp);
2965 2975 if ((myqbp->qb_flag & QB_BACK)) {
2966 2976 if (qbp->qb_flag & QB_FULL) {
2967 2977 /*
2968 2978 * The band must have become full again.
2969 2979 * Set QB_WANTW again so strwsrv will
2970 2980 * be back enabled when the band becomes
2971 2981 * non-full next time.
2972 2982 */
2973 2983 qbp->qb_flag |= QB_WANTW;
2974 2984 } else {
2975 2985 isevent = 1;
2976 2986 qbf[i] = 1;
2977 2987 }
2978 2988 }
2979 2989 myqbp = myqbp->qb_next;
2980 2990 i++;
2981 2991 }
2982 2992 mutex_exit(QLOCK(tq));
2983 2993
2984 2994 if (isevent) {
2985 2995 for (i = tq->q_nband; i; i--) {
2986 2996 if (qbf[i]) {
2987 2997 pollwakeup(&stp->sd_pollist, POLLWRBAND);
2988 2998 mutex_enter(&stp->sd_lock);
2989 2999 if (stp->sd_sigflags & S_WRBAND)
2990 3000 strsendsig(stp->sd_siglist, S_WRBAND,
2991 3001 (uchar_t)i, 0);
2992 3002 mutex_exit(&stp->sd_lock);
2993 3003 }
2994 3004 }
2995 3005 }
2996 3006
2997 3007 releasestr(q);
2998 3008 return (0);
2999 3009 }
3000 3010
3001 3011 /*
3002 3012 * Special case of strcopyin/strcopyout for copying
3003 3013 * struct strioctl that can deal with both data
3004 3014 * models.
3005 3015 */
3006 3016
3007 3017 #ifdef _LP64
3008 3018
3009 3019 static int
3010 3020 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3011 3021 {
3012 3022 struct strioctl32 strioc32;
3013 3023 struct strioctl *striocp;
3014 3024
3015 3025 if (copyflag & U_TO_K) {
3016 3026 ASSERT((copyflag & K_TO_K) == 0);
3017 3027
3018 3028 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3019 3029 if (copyin(from, &strioc32, sizeof (strioc32)))
3020 3030 return (EFAULT);
3021 3031
3022 3032 striocp = (struct strioctl *)to;
3023 3033 striocp->ic_cmd = strioc32.ic_cmd;
3024 3034 striocp->ic_timout = strioc32.ic_timout;
3025 3035 striocp->ic_len = strioc32.ic_len;
3026 3036 striocp->ic_dp = (char *)(uintptr_t)strioc32.ic_dp;
3027 3037
3028 3038 } else { /* NATIVE data model */
3029 3039 if (copyin(from, to, sizeof (struct strioctl))) {
3030 3040 return (EFAULT);
3031 3041 } else {
3032 3042 return (0);
3033 3043 }
3034 3044 }
3035 3045 } else {
3036 3046 ASSERT(copyflag & K_TO_K);
3037 3047 bcopy(from, to, sizeof (struct strioctl));
3038 3048 }
3039 3049 return (0);
3040 3050 }
3041 3051
3042 3052 static int
3043 3053 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3044 3054 {
3045 3055 struct strioctl32 strioc32;
3046 3056 struct strioctl *striocp;
3047 3057
3048 3058 if (copyflag & U_TO_K) {
3049 3059 ASSERT((copyflag & K_TO_K) == 0);
3050 3060
3051 3061 if ((flag & FMODELS) == DATAMODEL_ILP32) {
3052 3062 striocp = (struct strioctl *)from;
3053 3063 strioc32.ic_cmd = striocp->ic_cmd;
3054 3064 strioc32.ic_timout = striocp->ic_timout;
3055 3065 strioc32.ic_len = striocp->ic_len;
3056 3066 strioc32.ic_dp = (caddr32_t)(uintptr_t)striocp->ic_dp;
3057 3067 ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3058 3068 striocp->ic_dp);
3059 3069
3060 3070 if (copyout(&strioc32, to, sizeof (strioc32)))
3061 3071 return (EFAULT);
3062 3072
3063 3073 } else { /* NATIVE data model */
3064 3074 if (copyout(from, to, sizeof (struct strioctl))) {
3065 3075 return (EFAULT);
3066 3076 } else {
3067 3077 return (0);
3068 3078 }
3069 3079 }
3070 3080 } else {
3071 3081 ASSERT(copyflag & K_TO_K);
3072 3082 bcopy(from, to, sizeof (struct strioctl));
3073 3083 }
3074 3084 return (0);
3075 3085 }
3076 3086
3077 3087 #else /* ! _LP64 */
3078 3088
3079 3089 /* ARGSUSED2 */
3080 3090 static int
3081 3091 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3082 3092 {
3083 3093 return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3084 3094 }
3085 3095
3086 3096 /* ARGSUSED2 */
3087 3097 static int
3088 3098 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3089 3099 {
3090 3100 return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3091 3101 }
3092 3102
3093 3103 #endif /* _LP64 */
3094 3104
3095 3105 /*
3096 3106 * Determine type of job control semantics expected by user. The
3097 3107 * possibilities are:
3098 3108 * JCREAD - Behaves like read() on fd; send SIGTTIN
3099 3109 * JCWRITE - Behaves like write() on fd; send SIGTTOU if TOSTOP set
3100 3110 * JCSETP - Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3101 3111 * JCGETP - Gets a value in the stream; no signals.
3102 3112 * See straccess in strsubr.c for usage of these values.
3103 3113 *
3104 3114 * This routine also returns -1 for I_STR as a special case; the
3105 3115 * caller must call again with the real ioctl number for
3106 3116 * classification.
3107 3117 */
3108 3118 static int
3109 3119 job_control_type(int cmd)
3110 3120 {
3111 3121 switch (cmd) {
3112 3122 case I_STR:
3113 3123 return (-1);
3114 3124
3115 3125 case I_RECVFD:
3116 3126 case I_E_RECVFD:
3117 3127 return (JCREAD);
3118 3128
3119 3129 case I_FDINSERT:
3120 3130 case I_SENDFD:
3121 3131 return (JCWRITE);
3122 3132
3123 3133 case TCSETA:
3124 3134 case TCSETAW:
3125 3135 case TCSETAF:
3126 3136 case TCSBRK:
3127 3137 case TCXONC:
3128 3138 case TCFLSH:
3129 3139 case TCDSET: /* Obsolete */
3130 3140 case TIOCSWINSZ:
3131 3141 case TCSETS:
3132 3142 case TCSETSW:
3133 3143 case TCSETSF:
3134 3144 case TIOCSETD:
3135 3145 case TIOCHPCL:
3136 3146 case TIOCSETP:
3137 3147 case TIOCSETN:
3138 3148 case TIOCEXCL:
3139 3149 case TIOCNXCL:
3140 3150 case TIOCFLUSH:
3141 3151 case TIOCSETC:
3142 3152 case TIOCLBIS:
3143 3153 case TIOCLBIC:
3144 3154 case TIOCLSET:
3145 3155 case TIOCSBRK:
3146 3156 case TIOCCBRK:
3147 3157 case TIOCSDTR:
3148 3158 case TIOCCDTR:
3149 3159 case TIOCSLTC:
3150 3160 case TIOCSTOP:
3151 3161 case TIOCSTART:
3152 3162 case TIOCSTI:
3153 3163 case TIOCSPGRP:
3154 3164 case TIOCMSET:
3155 3165 case TIOCMBIS:
3156 3166 case TIOCMBIC:
3157 3167 case TIOCREMOTE:
3158 3168 case TIOCSIGNAL:
3159 3169 case LDSETT:
3160 3170 case LDSMAP: /* Obsolete */
3161 3171 case DIOCSETP:
3162 3172 case I_FLUSH:
3163 3173 case I_SRDOPT:
3164 3174 case I_SETSIG:
3165 3175 case I_SWROPT:
3166 3176 case I_FLUSHBAND:
3167 3177 case I_SETCLTIME:
3168 3178 case I_SERROPT:
3169 3179 case I_ESETSIG:
3170 3180 case FIONBIO:
3171 3181 case FIOASYNC:
3172 3182 case FIOSETOWN:
3173 3183 case JBOOT: /* Obsolete */
3174 3184 case JTERM: /* Obsolete */
3175 3185 case JTIMOM: /* Obsolete */
3176 3186 case JZOMBOOT: /* Obsolete */
3177 3187 case JAGENT: /* Obsolete */
3178 3188 case JTRUN: /* Obsolete */
3179 3189 case JXTPROTO: /* Obsolete */
3180 3190 return (JCSETP);
3181 3191 }
3182 3192
3183 3193 return (JCGETP);
3184 3194 }
3185 3195
3186 3196 /*
3187 3197 * ioctl for streams
3188 3198 */
3189 3199 int
3190 3200 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3191 3201 cred_t *crp, int *rvalp)
3192 3202 {
3193 3203 struct stdata *stp;
3194 3204 struct strcmd *scp;
3195 3205 struct strioctl strioc;
3196 3206 struct uio uio;
3197 3207 struct iovec iov;
3198 3208 int access;
3199 3209 mblk_t *mp;
3200 3210 int error = 0;
3201 3211 int done = 0;
3202 3212 ssize_t rmin, rmax;
3203 3213 queue_t *wrq;
3204 3214 queue_t *rdq;
3205 3215 boolean_t kioctl = B_FALSE;
3206 3216 uint32_t auditing = AU_AUDITING();
3207 3217
3208 3218 if (flag & FKIOCTL) {
3209 3219 copyflag = K_TO_K;
3210 3220 kioctl = B_TRUE;
3211 3221 }
3212 3222 ASSERT(vp->v_stream);
3213 3223 ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3214 3224 stp = vp->v_stream;
3215 3225
3216 3226 TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3217 3227 "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3218 3228
3219 3229 /*
3220 3230 * If the copy is kernel to kernel, make sure that the FNATIVE
3221 3231 * flag is set. After this it would be a serious error to have
3222 3232 * no model flag.
3223 3233 */
3224 3234 if (copyflag == K_TO_K)
3225 3235 flag = (flag & ~FMODELS) | FNATIVE;
3226 3236
3227 3237 ASSERT((flag & FMODELS) != 0);
3228 3238
3229 3239 wrq = stp->sd_wrq;
3230 3240 rdq = _RD(wrq);
3231 3241
3232 3242 access = job_control_type(cmd);
3233 3243
3234 3244 /* We should never see these here, should be handled by iwscn */
3235 3245 if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3236 3246 return (EINVAL);
3237 3247
3238 3248 mutex_enter(&stp->sd_lock);
3239 3249 if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3240 3250 mutex_exit(&stp->sd_lock);
3241 3251 return (error);
3242 3252 }
3243 3253 mutex_exit(&stp->sd_lock);
3244 3254
3245 3255 /*
3246 3256 * Check for sgttyb-related ioctls first, and complain as
3247 3257 * necessary.
3248 3258 */
3249 3259 switch (cmd) {
3250 3260 case TIOCGETP:
3251 3261 case TIOCSETP:
3252 3262 case TIOCSETN:
3253 3263 if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3254 3264 sgttyb_complaint = B_TRUE;
3255 3265 cmn_err(CE_NOTE,
3256 3266 "application used obsolete TIOC[GS]ET");
3257 3267 }
3258 3268 if (sgttyb_handling >= 3) {
3259 3269 tsignal(curthread, SIGSYS);
3260 3270 return (EIO);
3261 3271 }
3262 3272 break;
3263 3273 }
3264 3274
3265 3275 mutex_enter(&stp->sd_lock);
3266 3276
3267 3277 switch (cmd) {
3268 3278 case I_RECVFD:
3269 3279 case I_E_RECVFD:
3270 3280 case I_PEEK:
3271 3281 case I_NREAD:
3272 3282 case FIONREAD:
3273 3283 case FIORDCHK:
3274 3284 case I_ATMARK:
3275 3285 case FIONBIO:
3276 3286 case FIOASYNC:
3277 3287 if (stp->sd_flag & (STRDERR|STPLEX)) {
3278 3288 error = strgeterr(stp, STRDERR|STPLEX, 0);
3279 3289 if (error != 0) {
3280 3290 mutex_exit(&stp->sd_lock);
3281 3291 return (error);
3282 3292 }
3283 3293 }
3284 3294 break;
3285 3295
3286 3296 default:
3287 3297 if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3288 3298 error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3289 3299 if (error != 0) {
3290 3300 mutex_exit(&stp->sd_lock);
3291 3301 return (error);
3292 3302 }
3293 3303 }
3294 3304 }
3295 3305
3296 3306 mutex_exit(&stp->sd_lock);
3297 3307
3298 3308 switch (cmd) {
3299 3309 default:
3300 3310 /*
3301 3311 * The stream head has hardcoded knowledge of a
3302 3312 * miscellaneous collection of terminal-, keyboard- and
3303 3313 * mouse-related ioctls, enumerated below. This hardcoded
3304 3314 * knowledge allows the stream head to automatically
3305 3315 * convert transparent ioctl requests made by userland
3306 3316 * programs into I_STR ioctls which many old STREAMS
3307 3317 * modules and drivers require.
3308 3318 *
3309 3319 * No new ioctls should ever be added to this list.
3310 3320 * Instead, the STREAMS module or driver should be written
3311 3321 * to either handle transparent ioctls or require any
3312 3322 * userland programs to use I_STR ioctls (by returning
3313 3323 * EINVAL to any transparent ioctl requests).
3314 3324 *
3315 3325 * More importantly, removing ioctls from this list should
3316 3326 * be done with the utmost care, since our STREAMS modules
3317 3327 * and drivers *count* on the stream head performing this
3318 3328 * conversion, and thus may panic while processing
3319 3329 * transparent ioctl request for one of these ioctls (keep
3320 3330 * in mind that third party modules and drivers may have
3321 3331 * similar problems).
3322 3332 */
3323 3333 if (((cmd & IOCTYPE) == LDIOC) ||
3324 3334 ((cmd & IOCTYPE) == tIOC) ||
3325 3335 ((cmd & IOCTYPE) == TIOC) ||
3326 3336 ((cmd & IOCTYPE) == KIOC) ||
3327 3337 ((cmd & IOCTYPE) == MSIOC) ||
3328 3338 ((cmd & IOCTYPE) == VUIOC)) {
3329 3339 /*
3330 3340 * The ioctl is a tty ioctl - set up strioc buffer
3331 3341 * and call strdoioctl() to do the work.
3332 3342 */
3333 3343 if (stp->sd_flag & STRHUP)
3334 3344 return (ENXIO);
3335 3345 strioc.ic_cmd = cmd;
3336 3346 strioc.ic_timout = INFTIM;
3337 3347
3338 3348 switch (cmd) {
3339 3349
3340 3350 case TCXONC:
3341 3351 case TCSBRK:
3342 3352 case TCFLSH:
3343 3353 case TCDSET:
3344 3354 {
3345 3355 int native_arg = (int)arg;
3346 3356 strioc.ic_len = sizeof (int);
3347 3357 strioc.ic_dp = (char *)&native_arg;
3348 3358 return (strdoioctl(stp, &strioc, flag,
3349 3359 K_TO_K, crp, rvalp));
3350 3360 }
3351 3361
3352 3362 case TCSETA:
3353 3363 case TCSETAW:
3354 3364 case TCSETAF:
3355 3365 strioc.ic_len = sizeof (struct termio);
3356 3366 strioc.ic_dp = (char *)arg;
3357 3367 return (strdoioctl(stp, &strioc, flag,
3358 3368 copyflag, crp, rvalp));
3359 3369
3360 3370 case TCSETS:
3361 3371 case TCSETSW:
3362 3372 case TCSETSF:
3363 3373 strioc.ic_len = sizeof (struct termios);
3364 3374 strioc.ic_dp = (char *)arg;
3365 3375 return (strdoioctl(stp, &strioc, flag,
3366 3376 copyflag, crp, rvalp));
3367 3377
3368 3378 case LDSETT:
3369 3379 strioc.ic_len = sizeof (struct termcb);
3370 3380 strioc.ic_dp = (char *)arg;
3371 3381 return (strdoioctl(stp, &strioc, flag,
3372 3382 copyflag, crp, rvalp));
3373 3383
3374 3384 case TIOCSETP:
3375 3385 strioc.ic_len = sizeof (struct sgttyb);
3376 3386 strioc.ic_dp = (char *)arg;
3377 3387 return (strdoioctl(stp, &strioc, flag,
3378 3388 copyflag, crp, rvalp));
3379 3389
3380 3390 case TIOCSTI:
3381 3391 if ((flag & FREAD) == 0 &&
3382 3392 secpolicy_sti(crp) != 0) {
3383 3393 return (EPERM);
3384 3394 }
3385 3395 mutex_enter(&stp->sd_lock);
3386 3396 mutex_enter(&curproc->p_splock);
3387 3397 if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3388 3398 secpolicy_sti(crp) != 0) {
3389 3399 mutex_exit(&curproc->p_splock);
3390 3400 mutex_exit(&stp->sd_lock);
3391 3401 return (EACCES);
3392 3402 }
3393 3403 mutex_exit(&curproc->p_splock);
3394 3404 mutex_exit(&stp->sd_lock);
3395 3405
3396 3406 strioc.ic_len = sizeof (char);
3397 3407 strioc.ic_dp = (char *)arg;
3398 3408 return (strdoioctl(stp, &strioc, flag,
3399 3409 copyflag, crp, rvalp));
3400 3410
3401 3411 case TIOCSWINSZ:
3402 3412 strioc.ic_len = sizeof (struct winsize);
3403 3413 strioc.ic_dp = (char *)arg;
3404 3414 return (strdoioctl(stp, &strioc, flag,
3405 3415 copyflag, crp, rvalp));
3406 3416
3407 3417 case TIOCSSIZE:
3408 3418 strioc.ic_len = sizeof (struct ttysize);
3409 3419 strioc.ic_dp = (char *)arg;
3410 3420 return (strdoioctl(stp, &strioc, flag,
3411 3421 copyflag, crp, rvalp));
3412 3422
3413 3423 case TIOCSSOFTCAR:
3414 3424 case KIOCTRANS:
3415 3425 case KIOCTRANSABLE:
3416 3426 case KIOCCMD:
3417 3427 case KIOCSDIRECT:
3418 3428 case KIOCSCOMPAT:
3419 3429 case KIOCSKABORTEN:
3420 3430 case KIOCSRPTDELAY:
3421 3431 case KIOCSRPTRATE:
3422 3432 case VUIDSFORMAT:
3423 3433 case TIOCSPPS:
3424 3434 strioc.ic_len = sizeof (int);
3425 3435 strioc.ic_dp = (char *)arg;
3426 3436 return (strdoioctl(stp, &strioc, flag,
3427 3437 copyflag, crp, rvalp));
3428 3438
3429 3439 case KIOCSETKEY:
3430 3440 case KIOCGETKEY:
3431 3441 strioc.ic_len = sizeof (struct kiockey);
3432 3442 strioc.ic_dp = (char *)arg;
3433 3443 return (strdoioctl(stp, &strioc, flag,
3434 3444 copyflag, crp, rvalp));
3435 3445
3436 3446 case KIOCSKEY:
3437 3447 case KIOCGKEY:
3438 3448 strioc.ic_len = sizeof (struct kiockeymap);
3439 3449 strioc.ic_dp = (char *)arg;
3440 3450 return (strdoioctl(stp, &strioc, flag,
3441 3451 copyflag, crp, rvalp));
3442 3452
3443 3453 case KIOCSLED:
3444 3454 /* arg is a pointer to char */
3445 3455 strioc.ic_len = sizeof (char);
3446 3456 strioc.ic_dp = (char *)arg;
3447 3457 return (strdoioctl(stp, &strioc, flag,
3448 3458 copyflag, crp, rvalp));
3449 3459
3450 3460 case MSIOSETPARMS:
3451 3461 strioc.ic_len = sizeof (Ms_parms);
3452 3462 strioc.ic_dp = (char *)arg;
3453 3463 return (strdoioctl(stp, &strioc, flag,
3454 3464 copyflag, crp, rvalp));
3455 3465
3456 3466 case VUIDSADDR:
3457 3467 case VUIDGADDR:
3458 3468 strioc.ic_len = sizeof (struct vuid_addr_probe);
3459 3469 strioc.ic_dp = (char *)arg;
3460 3470 return (strdoioctl(stp, &strioc, flag,
3461 3471 copyflag, crp, rvalp));
3462 3472
3463 3473 /*
3464 3474 * These M_IOCTL's don't require any data to be sent
3465 3475 * downstream, and the driver will allocate and link
3466 3476 * on its own mblk_t upon M_IOCACK -- thus we set
3467 3477 * ic_len to zero and set ic_dp to arg so we know
3468 3478 * where to copyout to later.
3469 3479 */
3470 3480 case TIOCGSOFTCAR:
3471 3481 case TIOCGWINSZ:
3472 3482 case TIOCGSIZE:
3473 3483 case KIOCGTRANS:
3474 3484 case KIOCGTRANSABLE:
3475 3485 case KIOCTYPE:
3476 3486 case KIOCGDIRECT:
3477 3487 case KIOCGCOMPAT:
3478 3488 case KIOCLAYOUT:
3479 3489 case KIOCGLED:
3480 3490 case MSIOGETPARMS:
3481 3491 case MSIOBUTTONS:
3482 3492 case VUIDGFORMAT:
3483 3493 case TIOCGPPS:
3484 3494 case TIOCGPPSEV:
3485 3495 case TCGETA:
3486 3496 case TCGETS:
3487 3497 case LDGETT:
3488 3498 case TIOCGETP:
3489 3499 case KIOCGRPTDELAY:
3490 3500 case KIOCGRPTRATE:
3491 3501 strioc.ic_len = 0;
3492 3502 strioc.ic_dp = (char *)arg;
3493 3503 return (strdoioctl(stp, &strioc, flag,
3494 3504 copyflag, crp, rvalp));
3495 3505 }
3496 3506 }
3497 3507
3498 3508 /*
3499 3509 * Unknown cmd - send it down as a transparent ioctl.
3500 3510 */
3501 3511 strioc.ic_cmd = cmd;
3502 3512 strioc.ic_timout = INFTIM;
3503 3513 strioc.ic_len = TRANSPARENT;
3504 3514 strioc.ic_dp = (char *)&arg;
3505 3515
3506 3516 return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3507 3517
3508 3518 case I_STR:
3509 3519 /*
3510 3520 * Stream ioctl. Read in an strioctl buffer from the user
3511 3521 * along with any data specified and send it downstream.
3512 3522 * Strdoioctl will wait allow only one ioctl message at
3513 3523 * a time, and waits for the acknowledgement.
3514 3524 */
3515 3525
3516 3526 if (stp->sd_flag & STRHUP)
3517 3527 return (ENXIO);
3518 3528
3519 3529 error = strcopyin_strioctl((void *)arg, &strioc, flag,
3520 3530 copyflag);
3521 3531 if (error != 0)
3522 3532 return (error);
3523 3533
3524 3534 if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3525 3535 return (EINVAL);
3526 3536
3527 3537 access = job_control_type(strioc.ic_cmd);
3528 3538 mutex_enter(&stp->sd_lock);
3529 3539 if ((access != -1) &&
3530 3540 ((error = i_straccess(stp, access)) != 0)) {
3531 3541 mutex_exit(&stp->sd_lock);
3532 3542 return (error);
3533 3543 }
3534 3544 mutex_exit(&stp->sd_lock);
3535 3545
3536 3546 /*
3537 3547 * The I_STR facility provides a trap door for malicious
3538 3548 * code to send down bogus streamio(7I) ioctl commands to
3539 3549 * unsuspecting STREAMS modules and drivers which expect to
3540 3550 * only get these messages from the stream head.
3541 3551 * Explicitly prohibit any streamio ioctls which can be
3542 3552 * passed downstream by the stream head. Note that we do
3543 3553 * not block all streamio ioctls because the ioctl
3544 3554 * numberspace is not well managed and thus it's possible
3545 3555 * that a module or driver's ioctl numbers may accidentally
3546 3556 * collide with them.
3547 3557 */
3548 3558 switch (strioc.ic_cmd) {
3549 3559 case I_LINK:
3550 3560 case I_PLINK:
3551 3561 case I_UNLINK:
3552 3562 case I_PUNLINK:
3553 3563 case _I_GETPEERCRED:
3554 3564 case _I_PLINK_LH:
3555 3565 return (EINVAL);
3556 3566 }
3557 3567
3558 3568 error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3559 3569 if (error == 0) {
3560 3570 error = strcopyout_strioctl(&strioc, (void *)arg,
3561 3571 flag, copyflag);
3562 3572 }
3563 3573 return (error);
3564 3574
3565 3575 case _I_CMD:
3566 3576 /*
3567 3577 * Like I_STR, but without using M_IOC* messages and without
3568 3578 * copyins/copyouts beyond the passed-in argument.
3569 3579 */
3570 3580 if (stp->sd_flag & STRHUP)
3571 3581 return (ENXIO);
3572 3582
3573 3583 if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3574 3584 return (ENOMEM);
3575 3585
3576 3586 if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3577 3587 kmem_free(scp, sizeof (strcmd_t));
3578 3588 return (EFAULT);
3579 3589 }
3580 3590
3581 3591 access = job_control_type(scp->sc_cmd);
3582 3592 mutex_enter(&stp->sd_lock);
3583 3593 if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3584 3594 mutex_exit(&stp->sd_lock);
3585 3595 kmem_free(scp, sizeof (strcmd_t));
3586 3596 return (error);
3587 3597 }
3588 3598 mutex_exit(&stp->sd_lock);
3589 3599
3590 3600 *rvalp = 0;
3591 3601 if ((error = strdocmd(stp, scp, crp)) == 0) {
3592 3602 if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3593 3603 error = EFAULT;
3594 3604 }
3595 3605 kmem_free(scp, sizeof (strcmd_t));
3596 3606 return (error);
3597 3607
3598 3608 case I_NREAD:
3599 3609 /*
3600 3610 * Return number of bytes of data in first message
3601 3611 * in queue in "arg" and return the number of messages
3602 3612 * in queue in return value.
3603 3613 */
3604 3614 {
3605 3615 size_t size;
3606 3616 int retval;
3607 3617 int count = 0;
3608 3618
3609 3619 mutex_enter(QLOCK(rdq));
3610 3620
3611 3621 size = msgdsize(rdq->q_first);
3612 3622 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3613 3623 count++;
3614 3624
3615 3625 mutex_exit(QLOCK(rdq));
3616 3626 if (stp->sd_struiordq) {
3617 3627 infod_t infod;
3618 3628
3619 3629 infod.d_cmd = INFOD_COUNT;
3620 3630 infod.d_count = 0;
3621 3631 if (count == 0) {
3622 3632 infod.d_cmd |= INFOD_FIRSTBYTES;
3623 3633 infod.d_bytes = 0;
3624 3634 }
3625 3635 infod.d_res = 0;
3626 3636 (void) infonext(rdq, &infod);
3627 3637 count += infod.d_count;
3628 3638 if (infod.d_res & INFOD_FIRSTBYTES)
3629 3639 size = infod.d_bytes;
3630 3640 }
3631 3641
3632 3642 /*
3633 3643 * Drop down from size_t to the "int" required by the
3634 3644 * interface. Cap at INT_MAX.
3635 3645 */
3636 3646 retval = MIN(size, INT_MAX);
3637 3647 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3638 3648 copyflag);
3639 3649 if (!error)
3640 3650 *rvalp = count;
3641 3651 return (error);
3642 3652 }
3643 3653
3644 3654 case FIONREAD:
3645 3655 /*
3646 3656 * Return number of bytes of data in all data messages
3647 3657 * in queue in "arg".
3648 3658 */
3649 3659 {
3650 3660 size_t size = 0;
3651 3661 int retval;
3652 3662
3653 3663 mutex_enter(QLOCK(rdq));
3654 3664 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3655 3665 size += msgdsize(mp);
3656 3666 mutex_exit(QLOCK(rdq));
3657 3667
3658 3668 if (stp->sd_struiordq) {
3659 3669 infod_t infod;
3660 3670
3661 3671 infod.d_cmd = INFOD_BYTES;
3662 3672 infod.d_res = 0;
3663 3673 infod.d_bytes = 0;
3664 3674 (void) infonext(rdq, &infod);
3665 3675 size += infod.d_bytes;
3666 3676 }
3667 3677
3668 3678 /*
3669 3679 * Drop down from size_t to the "int" required by the
3670 3680 * interface. Cap at INT_MAX.
3671 3681 */
3672 3682 retval = MIN(size, INT_MAX);
3673 3683 error = strcopyout(&retval, (void *)arg, sizeof (retval),
3674 3684 copyflag);
3675 3685
3676 3686 *rvalp = 0;
3677 3687 return (error);
3678 3688 }
3679 3689 case FIORDCHK:
3680 3690 /*
3681 3691 * FIORDCHK does not use arg value (like FIONREAD),
3682 3692 * instead a count is returned. I_NREAD value may
3683 3693 * not be accurate but safe. The real thing to do is
3684 3694 * to add the msgdsizes of all data messages until
3685 3695 * a non-data message.
3686 3696 */
3687 3697 {
3688 3698 size_t size = 0;
3689 3699
3690 3700 mutex_enter(QLOCK(rdq));
3691 3701 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3692 3702 size += msgdsize(mp);
3693 3703 mutex_exit(QLOCK(rdq));
3694 3704
3695 3705 if (stp->sd_struiordq) {
3696 3706 infod_t infod;
3697 3707
3698 3708 infod.d_cmd = INFOD_BYTES;
3699 3709 infod.d_res = 0;
3700 3710 infod.d_bytes = 0;
3701 3711 (void) infonext(rdq, &infod);
3702 3712 size += infod.d_bytes;
3703 3713 }
3704 3714
3705 3715 /*
3706 3716 * Since ioctl returns an int, and memory sizes under
3707 3717 * LP64 may not fit, we return INT_MAX if the count was
3708 3718 * actually greater.
3709 3719 */
3710 3720 *rvalp = MIN(size, INT_MAX);
3711 3721 return (0);
3712 3722 }
3713 3723
3714 3724 case I_FIND:
3715 3725 /*
3716 3726 * Get module name.
3717 3727 */
3718 3728 {
3719 3729 char mname[FMNAMESZ + 1];
3720 3730 queue_t *q;
3721 3731
3722 3732 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3723 3733 mname, FMNAMESZ + 1, NULL);
3724 3734 if (error)
3725 3735 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3726 3736
3727 3737 /*
3728 3738 * Return EINVAL if we're handed a bogus module name.
3729 3739 */
3730 3740 if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3731 3741 TRACE_0(TR_FAC_STREAMS_FR,
3732 3742 TR_I_CANT_FIND, "couldn't I_FIND");
3733 3743 return (EINVAL);
3734 3744 }
3735 3745
3736 3746 *rvalp = 0;
3737 3747
3738 3748 /* Look downstream to see if module is there. */
3739 3749 claimstr(stp->sd_wrq);
3740 3750 for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3741 3751 if (q->q_flag & QREADR) {
3742 3752 q = NULL;
3743 3753 break;
3744 3754 }
3745 3755 if (strcmp(mname, Q2NAME(q)) == 0)
3746 3756 break;
3747 3757 }
3748 3758 releasestr(stp->sd_wrq);
3749 3759
3750 3760 *rvalp = (q ? 1 : 0);
3751 3761 return (error);
3752 3762 }
3753 3763
3754 3764 case I_PUSH:
3755 3765 case __I_PUSH_NOCTTY:
3756 3766 /*
3757 3767 * Push a module.
3758 3768 * For the case __I_PUSH_NOCTTY push a module but
3759 3769 * do not allocate controlling tty. See bugid 4025044
3760 3770 */
3761 3771
3762 3772 {
3763 3773 char mname[FMNAMESZ + 1];
3764 3774 fmodsw_impl_t *fp;
3765 3775 dev_t dummydev;
3766 3776
3767 3777 if (stp->sd_flag & STRHUP)
3768 3778 return (ENXIO);
3769 3779
3770 3780 /*
3771 3781 * Get module name and look up in fmodsw.
3772 3782 */
3773 3783 error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3774 3784 mname, FMNAMESZ + 1, NULL);
3775 3785 if (error)
3776 3786 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3777 3787
3778 3788 if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3779 3789 NULL)
3780 3790 return (EINVAL);
3781 3791
3782 3792 TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3783 3793 "I_PUSH:fp %p stp %p", fp, stp);
3784 3794
3785 3795 if (error = strstartplumb(stp, flag, cmd)) {
3786 3796 fmodsw_rele(fp);
3787 3797 return (error);
3788 3798 }
3789 3799
3790 3800 /*
3791 3801 * See if any more modules can be pushed on this stream.
3792 3802 * Note that this check must be done after strstartplumb()
3793 3803 * since otherwise multiple threads issuing I_PUSHes on
3794 3804 * the same stream will be able to exceed nstrpush.
3795 3805 */
3796 3806 mutex_enter(&stp->sd_lock);
3797 3807 if (stp->sd_pushcnt >= nstrpush) {
3798 3808 fmodsw_rele(fp);
3799 3809 strendplumb(stp);
3800 3810 mutex_exit(&stp->sd_lock);
3801 3811 return (EINVAL);
3802 3812 }
3803 3813 mutex_exit(&stp->sd_lock);
3804 3814
3805 3815 /*
3806 3816 * Push new module and call its open routine
3807 3817 * via qattach(). Modules don't change device
3808 3818 * numbers, so just ignore dummydev here.
3809 3819 */
3810 3820 dummydev = vp->v_rdev;
3811 3821 if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3812 3822 B_FALSE)) == 0) {
3813 3823 if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3814 3824 (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3815 3825 /*
3816 3826 * try to allocate it as a controlling terminal
3817 3827 */
3818 3828 (void) strctty(stp);
3819 3829 }
3820 3830 }
3821 3831
3822 3832 mutex_enter(&stp->sd_lock);
3823 3833
3824 3834 /*
3825 3835 * As a performance concern we are caching the values of
3826 3836 * q_minpsz and q_maxpsz of the module below the stream
3827 3837 * head in the stream head.
3828 3838 */
3829 3839 mutex_enter(QLOCK(stp->sd_wrq->q_next));
3830 3840 rmin = stp->sd_wrq->q_next->q_minpsz;
3831 3841 rmax = stp->sd_wrq->q_next->q_maxpsz;
3832 3842 mutex_exit(QLOCK(stp->sd_wrq->q_next));
3833 3843
3834 3844 /* Do this processing here as a performance concern */
3835 3845 if (strmsgsz != 0) {
3836 3846 if (rmax == INFPSZ)
3837 3847 rmax = strmsgsz;
3838 3848 else {
3839 3849 if (vp->v_type == VFIFO)
3840 3850 rmax = MIN(PIPE_BUF, rmax);
3841 3851 else rmax = MIN(strmsgsz, rmax);
3842 3852 }
3843 3853 }
3844 3854
3845 3855 mutex_enter(QLOCK(wrq));
3846 3856 stp->sd_qn_minpsz = rmin;
3847 3857 stp->sd_qn_maxpsz = rmax;
3848 3858 mutex_exit(QLOCK(wrq));
3849 3859
3850 3860 strendplumb(stp);
3851 3861 mutex_exit(&stp->sd_lock);
3852 3862 return (error);
3853 3863 }
3854 3864
3855 3865 case I_POP:
3856 3866 {
3857 3867 queue_t *q;
3858 3868
3859 3869 if (stp->sd_flag & STRHUP)
3860 3870 return (ENXIO);
3861 3871 if (!wrq->q_next) /* for broken pipes */
3862 3872 return (EINVAL);
3863 3873
3864 3874 if (error = strstartplumb(stp, flag, cmd))
3865 3875 return (error);
3866 3876
3867 3877 /*
3868 3878 * If there is an anchor on this stream and popping
3869 3879 * the current module would attempt to pop through the
3870 3880 * anchor, then disallow the pop unless we have sufficient
3871 3881 * privileges; take the cheapest (non-locking) check
3872 3882 * first.
3873 3883 */
3874 3884 if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3875 3885 (stp->sd_anchorzone != crgetzoneid(crp))) {
3876 3886 mutex_enter(&stp->sd_lock);
3877 3887 /*
3878 3888 * Anchors only apply if there's at least one
3879 3889 * module on the stream (sd_pushcnt > 0).
3880 3890 */
3881 3891 if (stp->sd_pushcnt > 0 &&
3882 3892 stp->sd_pushcnt == stp->sd_anchor &&
3883 3893 stp->sd_vnode->v_type != VFIFO) {
3884 3894 strendplumb(stp);
3885 3895 mutex_exit(&stp->sd_lock);
3886 3896 if (stp->sd_anchorzone != crgetzoneid(crp))
3887 3897 return (EINVAL);
3888 3898 /* Audit and report error */
3889 3899 return (secpolicy_ip_config(crp, B_FALSE));
3890 3900 }
3891 3901 mutex_exit(&stp->sd_lock);
3892 3902 }
3893 3903
3894 3904 q = wrq->q_next;
3895 3905 TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3896 3906 "I_POP:%p from %p", q, stp);
3897 3907 if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3898 3908 error = EINVAL;
3899 3909 } else {
3900 3910 qdetach(_RD(q), 1, flag, crp, B_FALSE);
3901 3911 error = 0;
3902 3912 }
3903 3913 mutex_enter(&stp->sd_lock);
3904 3914
3905 3915 /*
3906 3916 * As a performance concern we are caching the values of
3907 3917 * q_minpsz and q_maxpsz of the module below the stream
3908 3918 * head in the stream head.
3909 3919 */
3910 3920 mutex_enter(QLOCK(wrq->q_next));
3911 3921 rmin = wrq->q_next->q_minpsz;
3912 3922 rmax = wrq->q_next->q_maxpsz;
3913 3923 mutex_exit(QLOCK(wrq->q_next));
3914 3924
3915 3925 /* Do this processing here as a performance concern */
3916 3926 if (strmsgsz != 0) {
3917 3927 if (rmax == INFPSZ)
3918 3928 rmax = strmsgsz;
3919 3929 else {
3920 3930 if (vp->v_type == VFIFO)
3921 3931 rmax = MIN(PIPE_BUF, rmax);
3922 3932 else rmax = MIN(strmsgsz, rmax);
3923 3933 }
3924 3934 }
3925 3935
3926 3936 mutex_enter(QLOCK(wrq));
3927 3937 stp->sd_qn_minpsz = rmin;
3928 3938 stp->sd_qn_maxpsz = rmax;
3929 3939 mutex_exit(QLOCK(wrq));
3930 3940
3931 3941 /* If we popped through the anchor, then reset the anchor. */
3932 3942 if (stp->sd_pushcnt < stp->sd_anchor) {
3933 3943 stp->sd_anchor = 0;
3934 3944 stp->sd_anchorzone = 0;
3935 3945 }
3936 3946 strendplumb(stp);
3937 3947 mutex_exit(&stp->sd_lock);
3938 3948 return (error);
3939 3949 }
3940 3950
3941 3951 case _I_MUXID2FD:
3942 3952 {
3943 3953 /*
3944 3954 * Create a fd for a I_PLINK'ed lower stream with a given
3945 3955 * muxid. With the fd, application can send down ioctls,
3946 3956 * like I_LIST, to the previously I_PLINK'ed stream. Note
3947 3957 * that after getting the fd, the application has to do an
3948 3958 * I_PUNLINK on the muxid before it can do any operation
3949 3959 * on the lower stream. This is required by spec1170.
3950 3960 *
3951 3961 * The fd used to do this ioctl should point to the same
3952 3962 * controlling device used to do the I_PLINK. If it uses
3953 3963 * a different stream or an invalid muxid, I_MUXID2FD will
3954 3964 * fail. The error code is set to EINVAL.
3955 3965 *
3956 3966 * The intended use of this interface is the following.
3957 3967 * An application I_PLINK'ed a stream and exits. The fd
3958 3968 * to the lower stream is gone. Another application
3959 3969 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3960 3970 */
3961 3971 int muxid = (int)arg;
3962 3972 int fd;
3963 3973 linkinfo_t *linkp;
3964 3974 struct file *fp;
3965 3975 netstack_t *ns;
3966 3976 str_stack_t *ss;
3967 3977
3968 3978 /*
3969 3979 * Do not allow the wildcard muxid. This ioctl is not
3970 3980 * intended to find arbitrary link.
3971 3981 */
3972 3982 if (muxid == 0) {
3973 3983 return (EINVAL);
3974 3984 }
3975 3985
3976 3986 ns = netstack_find_by_cred(crp);
3977 3987 ASSERT(ns != NULL);
3978 3988 ss = ns->netstack_str;
3979 3989 ASSERT(ss != NULL);
3980 3990
3981 3991 mutex_enter(&muxifier);
3982 3992 linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
3983 3993 if (linkp == NULL) {
3984 3994 mutex_exit(&muxifier);
3985 3995 netstack_rele(ss->ss_netstack);
3986 3996 return (EINVAL);
3987 3997 }
3988 3998
3989 3999 if ((fd = ufalloc(0)) == -1) {
3990 4000 mutex_exit(&muxifier);
3991 4001 netstack_rele(ss->ss_netstack);
3992 4002 return (EMFILE);
3993 4003 }
3994 4004 fp = linkp->li_fpdown;
3995 4005 mutex_enter(&fp->f_tlock);
3996 4006 fp->f_count++;
3997 4007 mutex_exit(&fp->f_tlock);
3998 4008 mutex_exit(&muxifier);
3999 4009 setf(fd, fp);
4000 4010 *rvalp = fd;
4001 4011 netstack_rele(ss->ss_netstack);
4002 4012 return (0);
4003 4013 }
4004 4014
4005 4015 case _I_INSERT:
4006 4016 {
4007 4017 /*
4008 4018 * To insert a module to a given position in a stream.
4009 4019 * In the first release, only allow privileged user
4010 4020 * to use this ioctl. Furthermore, the insert is only allowed
4011 4021 * below an anchor if the zoneid is the same as the zoneid
4012 4022 * which created the anchor.
4013 4023 *
4014 4024 * Note that we do not plan to support this ioctl
4015 4025 * on pipes in the first release. We want to learn more
4016 4026 * about the implications of these ioctls before extending
4017 4027 * their support. And we do not think these features are
4018 4028 * valuable for pipes.
4019 4029 */
4020 4030 STRUCT_DECL(strmodconf, strmodinsert);
4021 4031 char mod_name[FMNAMESZ + 1];
4022 4032 fmodsw_impl_t *fp;
4023 4033 dev_t dummydev;
4024 4034 queue_t *tmp_wrq;
4025 4035 int pos;
4026 4036 boolean_t is_insert;
4027 4037
4028 4038 STRUCT_INIT(strmodinsert, flag);
4029 4039 if (stp->sd_flag & STRHUP)
4030 4040 return (ENXIO);
4031 4041 if (STRMATED(stp))
4032 4042 return (EINVAL);
4033 4043 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4034 4044 return (error);
4035 4045 if (stp->sd_anchor != 0 &&
4036 4046 stp->sd_anchorzone != crgetzoneid(crp))
4037 4047 return (EINVAL);
4038 4048
4039 4049 error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4040 4050 STRUCT_SIZE(strmodinsert), copyflag);
4041 4051 if (error)
4042 4052 return (error);
4043 4053
4044 4054 /*
4045 4055 * Get module name and look up in fmodsw.
4046 4056 */
4047 4057 error = (copyflag & U_TO_K ? copyinstr :
4048 4058 copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4049 4059 mod_name, FMNAMESZ + 1, NULL);
4050 4060 if (error)
4051 4061 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4052 4062
4053 4063 if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4054 4064 NULL)
4055 4065 return (EINVAL);
4056 4066
4057 4067 if (error = strstartplumb(stp, flag, cmd)) {
4058 4068 fmodsw_rele(fp);
4059 4069 return (error);
4060 4070 }
4061 4071
4062 4072 /*
4063 4073 * Is this _I_INSERT just like an I_PUSH? We need to know
4064 4074 * this because we do some optimizations if this is a
4065 4075 * module being pushed.
4066 4076 */
4067 4077 pos = STRUCT_FGET(strmodinsert, pos);
4068 4078 is_insert = (pos != 0);
4069 4079
4070 4080 /*
4071 4081 * Make sure pos is valid. Even though it is not an I_PUSH,
4072 4082 * we impose the same limit on the number of modules in a
4073 4083 * stream.
4074 4084 */
4075 4085 mutex_enter(&stp->sd_lock);
4076 4086 if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4077 4087 pos > stp->sd_pushcnt) {
4078 4088 fmodsw_rele(fp);
4079 4089 strendplumb(stp);
4080 4090 mutex_exit(&stp->sd_lock);
4081 4091 return (EINVAL);
4082 4092 }
4083 4093 if (stp->sd_anchor != 0) {
4084 4094 /*
4085 4095 * Is this insert below the anchor?
4086 4096 * Pushcnt hasn't been increased yet hence
4087 4097 * we test for greater than here, and greater or
4088 4098 * equal after qattach.
4089 4099 */
4090 4100 if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4091 4101 stp->sd_anchorzone != crgetzoneid(crp)) {
4092 4102 fmodsw_rele(fp);
4093 4103 strendplumb(stp);
4094 4104 mutex_exit(&stp->sd_lock);
4095 4105 return (EPERM);
4096 4106 }
4097 4107 }
4098 4108
4099 4109 mutex_exit(&stp->sd_lock);
4100 4110
4101 4111 /*
4102 4112 * First find the correct position this module to
4103 4113 * be inserted. We don't need to call claimstr()
4104 4114 * as the stream should not be changing at this point.
4105 4115 *
4106 4116 * Insert new module and call its open routine
4107 4117 * via qattach(). Modules don't change device
4108 4118 * numbers, so just ignore dummydev here.
4109 4119 */
4110 4120 for (tmp_wrq = stp->sd_wrq; pos > 0;
4111 4121 tmp_wrq = tmp_wrq->q_next, pos--) {
4112 4122 ASSERT(SAMESTR(tmp_wrq));
4113 4123 }
4114 4124 dummydev = vp->v_rdev;
4115 4125 if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4116 4126 fp, is_insert)) != 0) {
4117 4127 mutex_enter(&stp->sd_lock);
4118 4128 strendplumb(stp);
4119 4129 mutex_exit(&stp->sd_lock);
4120 4130 return (error);
4121 4131 }
4122 4132
4123 4133 mutex_enter(&stp->sd_lock);
4124 4134
4125 4135 /*
4126 4136 * As a performance concern we are caching the values of
4127 4137 * q_minpsz and q_maxpsz of the module below the stream
4128 4138 * head in the stream head.
4129 4139 */
4130 4140 if (!is_insert) {
4131 4141 mutex_enter(QLOCK(stp->sd_wrq->q_next));
4132 4142 rmin = stp->sd_wrq->q_next->q_minpsz;
4133 4143 rmax = stp->sd_wrq->q_next->q_maxpsz;
4134 4144 mutex_exit(QLOCK(stp->sd_wrq->q_next));
4135 4145
4136 4146 /* Do this processing here as a performance concern */
4137 4147 if (strmsgsz != 0) {
4138 4148 if (rmax == INFPSZ) {
4139 4149 rmax = strmsgsz;
4140 4150 } else {
4141 4151 rmax = MIN(strmsgsz, rmax);
4142 4152 }
4143 4153 }
4144 4154
4145 4155 mutex_enter(QLOCK(wrq));
4146 4156 stp->sd_qn_minpsz = rmin;
4147 4157 stp->sd_qn_maxpsz = rmax;
4148 4158 mutex_exit(QLOCK(wrq));
4149 4159 }
4150 4160
4151 4161 /*
4152 4162 * Need to update the anchor value if this module is
4153 4163 * inserted below the anchor point.
4154 4164 */
4155 4165 if (stp->sd_anchor != 0) {
4156 4166 pos = STRUCT_FGET(strmodinsert, pos);
4157 4167 if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4158 4168 stp->sd_anchor++;
4159 4169 }
4160 4170
4161 4171 strendplumb(stp);
4162 4172 mutex_exit(&stp->sd_lock);
4163 4173 return (0);
4164 4174 }
4165 4175
4166 4176 case _I_REMOVE:
4167 4177 {
4168 4178 /*
4169 4179 * To remove a module with a given name in a stream. The
4170 4180 * caller of this ioctl needs to provide both the name and
4171 4181 * the position of the module to be removed. This eliminates
4172 4182 * the ambiguity of removal if a module is inserted/pushed
4173 4183 * multiple times in a stream. In the first release, only
4174 4184 * allow privileged user to use this ioctl.
4175 4185 * Furthermore, the remove is only allowed
4176 4186 * below an anchor if the zoneid is the same as the zoneid
4177 4187 * which created the anchor.
4178 4188 *
4179 4189 * Note that we do not plan to support this ioctl
4180 4190 * on pipes in the first release. We want to learn more
4181 4191 * about the implications of these ioctls before extending
4182 4192 * their support. And we do not think these features are
4183 4193 * valuable for pipes.
4184 4194 *
4185 4195 * Also note that _I_REMOVE cannot be used to remove a
4186 4196 * driver or the stream head.
4187 4197 */
4188 4198 STRUCT_DECL(strmodconf, strmodremove);
4189 4199 queue_t *q;
4190 4200 int pos;
4191 4201 char mod_name[FMNAMESZ + 1];
4192 4202 boolean_t is_remove;
4193 4203
4194 4204 STRUCT_INIT(strmodremove, flag);
4195 4205 if (stp->sd_flag & STRHUP)
4196 4206 return (ENXIO);
4197 4207 if (STRMATED(stp))
4198 4208 return (EINVAL);
4199 4209 if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4200 4210 return (error);
4201 4211 if (stp->sd_anchor != 0 &&
4202 4212 stp->sd_anchorzone != crgetzoneid(crp))
4203 4213 return (EINVAL);
4204 4214
4205 4215 error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4206 4216 STRUCT_SIZE(strmodremove), copyflag);
4207 4217 if (error)
4208 4218 return (error);
4209 4219
4210 4220 error = (copyflag & U_TO_K ? copyinstr :
4211 4221 copystr)(STRUCT_FGETP(strmodremove, mod_name),
4212 4222 mod_name, FMNAMESZ + 1, NULL);
4213 4223 if (error)
4214 4224 return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4215 4225
4216 4226 if ((error = strstartplumb(stp, flag, cmd)) != 0)
4217 4227 return (error);
4218 4228
4219 4229 /*
4220 4230 * Match the name of given module to the name of module at
4221 4231 * the given position.
4222 4232 */
4223 4233 pos = STRUCT_FGET(strmodremove, pos);
4224 4234
4225 4235 is_remove = (pos != 0);
4226 4236 for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4227 4237 q = q->q_next, pos--)
4228 4238 ;
4229 4239 if (pos > 0 || !SAMESTR(q) ||
4230 4240 strcmp(Q2NAME(q), mod_name) != 0) {
4231 4241 mutex_enter(&stp->sd_lock);
4232 4242 strendplumb(stp);
4233 4243 mutex_exit(&stp->sd_lock);
4234 4244 return (EINVAL);
4235 4245 }
4236 4246
4237 4247 /*
4238 4248 * If the position is at or below an anchor, then the zoneid
4239 4249 * must match the zoneid that created the anchor.
4240 4250 */
4241 4251 if (stp->sd_anchor != 0) {
4242 4252 pos = STRUCT_FGET(strmodremove, pos);
4243 4253 if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4244 4254 stp->sd_anchorzone != crgetzoneid(crp)) {
4245 4255 mutex_enter(&stp->sd_lock);
4246 4256 strendplumb(stp);
4247 4257 mutex_exit(&stp->sd_lock);
4248 4258 return (EPERM);
4249 4259 }
4250 4260 }
4251 4261
4252 4262
4253 4263 ASSERT(!(q->q_flag & QREADR));
4254 4264 qdetach(_RD(q), 1, flag, crp, is_remove);
4255 4265
4256 4266 mutex_enter(&stp->sd_lock);
4257 4267
4258 4268 /*
4259 4269 * As a performance concern we are caching the values of
4260 4270 * q_minpsz and q_maxpsz of the module below the stream
4261 4271 * head in the stream head.
4262 4272 */
4263 4273 if (!is_remove) {
4264 4274 mutex_enter(QLOCK(wrq->q_next));
4265 4275 rmin = wrq->q_next->q_minpsz;
4266 4276 rmax = wrq->q_next->q_maxpsz;
4267 4277 mutex_exit(QLOCK(wrq->q_next));
4268 4278
4269 4279 /* Do this processing here as a performance concern */
4270 4280 if (strmsgsz != 0) {
4271 4281 if (rmax == INFPSZ)
4272 4282 rmax = strmsgsz;
4273 4283 else {
4274 4284 if (vp->v_type == VFIFO)
4275 4285 rmax = MIN(PIPE_BUF, rmax);
4276 4286 else rmax = MIN(strmsgsz, rmax);
4277 4287 }
4278 4288 }
4279 4289
4280 4290 mutex_enter(QLOCK(wrq));
4281 4291 stp->sd_qn_minpsz = rmin;
4282 4292 stp->sd_qn_maxpsz = rmax;
4283 4293 mutex_exit(QLOCK(wrq));
4284 4294 }
4285 4295
4286 4296 /*
4287 4297 * Need to update the anchor value if this module is removed
4288 4298 * at or below the anchor point. If the removed module is at
4289 4299 * the anchor point, remove the anchor for this stream if
4290 4300 * there is no module above the anchor point. Otherwise, if
4291 4301 * the removed module is below the anchor point, decrement the
4292 4302 * anchor point by 1.
4293 4303 */
4294 4304 if (stp->sd_anchor != 0) {
4295 4305 pos = STRUCT_FGET(strmodremove, pos);
4296 4306 if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4297 4307 stp->sd_anchor = 0;
4298 4308 else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4299 4309 stp->sd_anchor--;
4300 4310 }
4301 4311
4302 4312 strendplumb(stp);
4303 4313 mutex_exit(&stp->sd_lock);
4304 4314 return (0);
4305 4315 }
4306 4316
4307 4317 case I_ANCHOR:
4308 4318 /*
4309 4319 * Set the anchor position on the stream to reside at
4310 4320 * the top module (in other words, the top module
4311 4321 * cannot be popped). Anchors with a FIFO make no
4312 4322 * obvious sense, so they're not allowed.
4313 4323 */
4314 4324 mutex_enter(&stp->sd_lock);
4315 4325
4316 4326 if (stp->sd_vnode->v_type == VFIFO) {
4317 4327 mutex_exit(&stp->sd_lock);
4318 4328 return (EINVAL);
4319 4329 }
4320 4330 /* Only allow the same zoneid to update the anchor */
4321 4331 if (stp->sd_anchor != 0 &&
4322 4332 stp->sd_anchorzone != crgetzoneid(crp)) {
4323 4333 mutex_exit(&stp->sd_lock);
4324 4334 return (EINVAL);
4325 4335 }
4326 4336 stp->sd_anchor = stp->sd_pushcnt;
4327 4337 stp->sd_anchorzone = crgetzoneid(crp);
4328 4338 mutex_exit(&stp->sd_lock);
4329 4339 return (0);
4330 4340
4331 4341 case I_LOOK:
4332 4342 /*
4333 4343 * Get name of first module downstream.
4334 4344 * If no module, return an error.
4335 4345 */
4336 4346 claimstr(wrq);
4337 4347 if (_SAMESTR(wrq) && wrq->q_next->q_next != NULL) {
4338 4348 char *name = Q2NAME(wrq->q_next);
4339 4349
4340 4350 error = strcopyout(name, (void *)arg, strlen(name) + 1,
4341 4351 copyflag);
4342 4352 releasestr(wrq);
4343 4353 return (error);
4344 4354 }
4345 4355 releasestr(wrq);
4346 4356 return (EINVAL);
4347 4357
4348 4358 case I_LINK:
4349 4359 case I_PLINK:
4350 4360 /*
4351 4361 * Link a multiplexor.
4352 4362 */
4353 4363 return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4354 4364
4355 4365 case _I_PLINK_LH:
4356 4366 /*
4357 4367 * Link a multiplexor: Call must originate from kernel.
4358 4368 */
4359 4369 if (kioctl)
4360 4370 return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4361 4371
4362 4372 return (EINVAL);
4363 4373 case I_UNLINK:
4364 4374 case I_PUNLINK:
4365 4375 /*
4366 4376 * Unlink a multiplexor.
4367 4377 * If arg is -1, unlink all links for which this is the
4368 4378 * controlling stream. Otherwise, arg is an index number
4369 4379 * for a link to be removed.
4370 4380 */
4371 4381 {
4372 4382 struct linkinfo *linkp;
4373 4383 int native_arg = (int)arg;
4374 4384 int type;
4375 4385 netstack_t *ns;
4376 4386 str_stack_t *ss;
4377 4387
4378 4388 TRACE_1(TR_FAC_STREAMS_FR,
4379 4389 TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4380 4390 if (vp->v_type == VFIFO) {
4381 4391 return (EINVAL);
4382 4392 }
4383 4393 if (cmd == I_UNLINK)
4384 4394 type = LINKNORMAL;
4385 4395 else /* I_PUNLINK */
4386 4396 type = LINKPERSIST;
4387 4397 if (native_arg == 0) {
4388 4398 return (EINVAL);
4389 4399 }
4390 4400 ns = netstack_find_by_cred(crp);
4391 4401 ASSERT(ns != NULL);
4392 4402 ss = ns->netstack_str;
4393 4403 ASSERT(ss != NULL);
4394 4404
4395 4405 if (native_arg == MUXID_ALL)
4396 4406 error = munlinkall(stp, type, crp, rvalp, ss);
4397 4407 else {
4398 4408 mutex_enter(&muxifier);
4399 4409 if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4400 4410 /* invalid user supplied index number */
4401 4411 mutex_exit(&muxifier);
4402 4412 netstack_rele(ss->ss_netstack);
4403 4413 return (EINVAL);
4404 4414 }
4405 4415 /* munlink drops the muxifier lock */
4406 4416 error = munlink(stp, linkp, type, crp, rvalp, ss);
4407 4417 }
4408 4418 netstack_rele(ss->ss_netstack);
4409 4419 return (error);
4410 4420 }
4411 4421
4412 4422 case I_FLUSH:
4413 4423 /*
4414 4424 * send a flush message downstream
4415 4425 * flush message can indicate
4416 4426 * FLUSHR - flush read queue
4417 4427 * FLUSHW - flush write queue
4418 4428 * FLUSHRW - flush read/write queue
4419 4429 */
4420 4430 if (stp->sd_flag & STRHUP)
4421 4431 return (ENXIO);
4422 4432 if (arg & ~FLUSHRW)
4423 4433 return (EINVAL);
4424 4434
4425 4435 for (;;) {
4426 4436 if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4427 4437 break;
4428 4438 }
4429 4439 if (error = strwaitbuf(1, BPRI_HI)) {
4430 4440 return (error);
4431 4441 }
4432 4442 }
4433 4443
4434 4444 /*
4435 4445 * Send down an unsupported ioctl and wait for the nack
4436 4446 * in order to allow the M_FLUSH to propagate back
4437 4447 * up to the stream head.
4438 4448 * Replaces if (qready()) runqueues();
4439 4449 */
4440 4450 strioc.ic_cmd = -1; /* The unsupported ioctl */
4441 4451 strioc.ic_timout = 0;
4442 4452 strioc.ic_len = 0;
4443 4453 strioc.ic_dp = NULL;
4444 4454 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4445 4455 *rvalp = 0;
4446 4456 return (0);
4447 4457
4448 4458 case I_FLUSHBAND:
4449 4459 {
4450 4460 struct bandinfo binfo;
4451 4461
4452 4462 error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4453 4463 copyflag);
4454 4464 if (error)
4455 4465 return (error);
4456 4466 if (stp->sd_flag & STRHUP)
4457 4467 return (ENXIO);
4458 4468 if (binfo.bi_flag & ~FLUSHRW)
4459 4469 return (EINVAL);
4460 4470 while (!(mp = allocb(2, BPRI_HI))) {
4461 4471 if (error = strwaitbuf(2, BPRI_HI))
4462 4472 return (error);
4463 4473 }
4464 4474 mp->b_datap->db_type = M_FLUSH;
4465 4475 *mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4466 4476 *mp->b_wptr++ = binfo.bi_pri;
4467 4477 putnext(stp->sd_wrq, mp);
4468 4478 /*
4469 4479 * Send down an unsupported ioctl and wait for the nack
4470 4480 * in order to allow the M_FLUSH to propagate back
4471 4481 * up to the stream head.
4472 4482 * Replaces if (qready()) runqueues();
4473 4483 */
4474 4484 strioc.ic_cmd = -1; /* The unsupported ioctl */
4475 4485 strioc.ic_timout = 0;
4476 4486 strioc.ic_len = 0;
4477 4487 strioc.ic_dp = NULL;
4478 4488 (void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4479 4489 *rvalp = 0;
4480 4490 return (0);
4481 4491 }
4482 4492
4483 4493 case I_SRDOPT:
4484 4494 /*
4485 4495 * Set read options
4486 4496 *
4487 4497 * RNORM - default stream mode
4488 4498 * RMSGN - message no discard
4489 4499 * RMSGD - message discard
4490 4500 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4491 4501 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4492 4502 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4493 4503 */
4494 4504 if (arg & ~(RMODEMASK | RPROTMASK))
4495 4505 return (EINVAL);
4496 4506
4497 4507 if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4498 4508 return (EINVAL);
4499 4509
4500 4510 mutex_enter(&stp->sd_lock);
4501 4511 switch (arg & RMODEMASK) {
4502 4512 case RNORM:
4503 4513 stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4504 4514 break;
4505 4515 case RMSGD:
4506 4516 stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4507 4517 RD_MSGDIS;
4508 4518 break;
4509 4519 case RMSGN:
4510 4520 stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4511 4521 RD_MSGNODIS;
4512 4522 break;
4513 4523 }
4514 4524
4515 4525 switch (arg & RPROTMASK) {
4516 4526 case RPROTNORM:
4517 4527 stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4518 4528 break;
4519 4529
4520 4530 case RPROTDAT:
4521 4531 stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4522 4532 RD_PROTDAT);
4523 4533 break;
4524 4534
4525 4535 case RPROTDIS:
4526 4536 stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4527 4537 RD_PROTDIS);
4528 4538 break;
4529 4539 }
4530 4540 mutex_exit(&stp->sd_lock);
4531 4541 return (0);
4532 4542
4533 4543 case I_GRDOPT:
4534 4544 /*
4535 4545 * Get read option and return the value
4536 4546 * to spot pointed to by arg
4537 4547 */
4538 4548 {
4539 4549 int rdopt;
4540 4550
4541 4551 rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4542 4552 ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4543 4553 rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4544 4554 ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4545 4555
4546 4556 return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4547 4557 copyflag));
4548 4558 }
4549 4559
4550 4560 case I_SERROPT:
4551 4561 /*
4552 4562 * Set error options
4553 4563 *
4554 4564 * RERRNORM - persistent read errors
4555 4565 * RERRNONPERSIST - non-persistent read errors
4556 4566 * WERRNORM - persistent write errors
4557 4567 * WERRNONPERSIST - non-persistent write errors
4558 4568 */
4559 4569 if (arg & ~(RERRMASK | WERRMASK))
4560 4570 return (EINVAL);
4561 4571
4562 4572 mutex_enter(&stp->sd_lock);
4563 4573 switch (arg & RERRMASK) {
4564 4574 case RERRNORM:
4565 4575 stp->sd_flag &= ~STRDERRNONPERSIST;
4566 4576 break;
4567 4577 case RERRNONPERSIST:
4568 4578 stp->sd_flag |= STRDERRNONPERSIST;
4569 4579 break;
4570 4580 }
4571 4581 switch (arg & WERRMASK) {
4572 4582 case WERRNORM:
4573 4583 stp->sd_flag &= ~STWRERRNONPERSIST;
4574 4584 break;
4575 4585 case WERRNONPERSIST:
4576 4586 stp->sd_flag |= STWRERRNONPERSIST;
4577 4587 break;
4578 4588 }
4579 4589 mutex_exit(&stp->sd_lock);
4580 4590 return (0);
4581 4591
4582 4592 case I_GERROPT:
4583 4593 /*
4584 4594 * Get error option and return the value
4585 4595 * to spot pointed to by arg
4586 4596 */
4587 4597 {
4588 4598 int erropt = 0;
4589 4599
4590 4600 erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4591 4601 RERRNORM;
4592 4602 erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4593 4603 WERRNORM;
4594 4604 return (strcopyout(&erropt, (void *)arg, sizeof (int),
4595 4605 copyflag));
4596 4606 }
4597 4607
4598 4608 case I_SETSIG:
4599 4609 /*
4600 4610 * Register the calling proc to receive the SIGPOLL
4601 4611 * signal based on the events given in arg. If
4602 4612 * arg is zero, remove the proc from register list.
4603 4613 */
4604 4614 {
4605 4615 strsig_t *ssp, *pssp;
4606 4616 struct pid *pidp;
4607 4617
4608 4618 pssp = NULL;
4609 4619 pidp = curproc->p_pidp;
4610 4620 /*
4611 4621 * Hold sd_lock to prevent traversal of sd_siglist while
4612 4622 * it is modified.
4613 4623 */
4614 4624 mutex_enter(&stp->sd_lock);
4615 4625 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4616 4626 pssp = ssp, ssp = ssp->ss_next)
4617 4627 ;
4618 4628
4619 4629 if (arg) {
4620 4630 if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4621 4631 S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4622 4632 mutex_exit(&stp->sd_lock);
4623 4633 return (EINVAL);
4624 4634 }
4625 4635 if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4626 4636 mutex_exit(&stp->sd_lock);
4627 4637 return (EINVAL);
4628 4638 }
4629 4639
4630 4640 /*
4631 4641 * If proc not already registered, add it
4632 4642 * to list.
4633 4643 */
4634 4644 if (!ssp) {
4635 4645 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4636 4646 ssp->ss_pidp = pidp;
4637 4647 ssp->ss_pid = pidp->pid_id;
4638 4648 ssp->ss_next = NULL;
4639 4649 if (pssp)
4640 4650 pssp->ss_next = ssp;
4641 4651 else
4642 4652 stp->sd_siglist = ssp;
4643 4653 mutex_enter(&pidlock);
4644 4654 PID_HOLD(pidp);
4645 4655 mutex_exit(&pidlock);
4646 4656 }
4647 4657
4648 4658 /*
4649 4659 * Set events.
4650 4660 */
4651 4661 ssp->ss_events = (int)arg;
4652 4662 } else {
4653 4663 /*
4654 4664 * Remove proc from register list.
4655 4665 */
4656 4666 if (ssp) {
4657 4667 mutex_enter(&pidlock);
4658 4668 PID_RELE(pidp);
4659 4669 mutex_exit(&pidlock);
4660 4670 if (pssp)
4661 4671 pssp->ss_next = ssp->ss_next;
4662 4672 else
4663 4673 stp->sd_siglist = ssp->ss_next;
4664 4674 kmem_free(ssp, sizeof (strsig_t));
4665 4675 } else {
4666 4676 mutex_exit(&stp->sd_lock);
4667 4677 return (EINVAL);
4668 4678 }
4669 4679 }
4670 4680
4671 4681 /*
4672 4682 * Recalculate OR of sig events.
4673 4683 */
4674 4684 stp->sd_sigflags = 0;
4675 4685 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4676 4686 stp->sd_sigflags |= ssp->ss_events;
4677 4687 mutex_exit(&stp->sd_lock);
4678 4688 return (0);
4679 4689 }
4680 4690
4681 4691 case I_GETSIG:
4682 4692 /*
4683 4693 * Return (in arg) the current registration of events
4684 4694 * for which the calling proc is to be signaled.
4685 4695 */
4686 4696 {
4687 4697 struct strsig *ssp;
4688 4698 struct pid *pidp;
4689 4699
4690 4700 pidp = curproc->p_pidp;
4691 4701 mutex_enter(&stp->sd_lock);
4692 4702 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4693 4703 if (ssp->ss_pidp == pidp) {
4694 4704 error = strcopyout(&ssp->ss_events, (void *)arg,
4695 4705 sizeof (int), copyflag);
4696 4706 mutex_exit(&stp->sd_lock);
4697 4707 return (error);
4698 4708 }
4699 4709 mutex_exit(&stp->sd_lock);
4700 4710 return (EINVAL);
4701 4711 }
4702 4712
4703 4713 case I_ESETSIG:
4704 4714 /*
4705 4715 * Register the ss_pid to receive the SIGPOLL
4706 4716 * signal based on the events is ss_events arg. If
4707 4717 * ss_events is zero, remove the proc from register list.
4708 4718 */
4709 4719 {
4710 4720 struct strsig *ssp, *pssp;
4711 4721 struct proc *proc;
4712 4722 struct pid *pidp;
4713 4723 pid_t pid;
4714 4724 struct strsigset ss;
4715 4725
4716 4726 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4717 4727 if (error)
4718 4728 return (error);
4719 4729
4720 4730 pid = ss.ss_pid;
4721 4731
4722 4732 if (ss.ss_events != 0) {
4723 4733 /*
4724 4734 * Permissions check by sending signal 0.
4725 4735 * Note that when kill fails it does a set_errno
4726 4736 * causing the system call to fail.
4727 4737 */
4728 4738 error = kill(pid, 0);
4729 4739 if (error) {
4730 4740 return (error);
4731 4741 }
4732 4742 }
4733 4743 mutex_enter(&pidlock);
4734 4744 if (pid == 0)
4735 4745 proc = curproc;
4736 4746 else if (pid < 0)
4737 4747 proc = pgfind(-pid);
4738 4748 else
4739 4749 proc = prfind(pid);
4740 4750 if (proc == NULL) {
4741 4751 mutex_exit(&pidlock);
4742 4752 return (ESRCH);
4743 4753 }
4744 4754 if (pid < 0)
4745 4755 pidp = proc->p_pgidp;
4746 4756 else
4747 4757 pidp = proc->p_pidp;
4748 4758 ASSERT(pidp);
4749 4759 /*
4750 4760 * Get a hold on the pid structure while referencing it.
4751 4761 * There is a separate PID_HOLD should it be inserted
4752 4762 * in the list below.
4753 4763 */
4754 4764 PID_HOLD(pidp);
4755 4765 mutex_exit(&pidlock);
4756 4766
4757 4767 pssp = NULL;
4758 4768 /*
4759 4769 * Hold sd_lock to prevent traversal of sd_siglist while
4760 4770 * it is modified.
4761 4771 */
4762 4772 mutex_enter(&stp->sd_lock);
4763 4773 for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4764 4774 pssp = ssp, ssp = ssp->ss_next)
4765 4775 ;
4766 4776
4767 4777 if (ss.ss_events) {
4768 4778 if (ss.ss_events &
4769 4779 ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4770 4780 S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4771 4781 mutex_exit(&stp->sd_lock);
4772 4782 mutex_enter(&pidlock);
4773 4783 PID_RELE(pidp);
4774 4784 mutex_exit(&pidlock);
4775 4785 return (EINVAL);
4776 4786 }
4777 4787 if ((ss.ss_events & S_BANDURG) &&
4778 4788 !(ss.ss_events & S_RDBAND)) {
4779 4789 mutex_exit(&stp->sd_lock);
4780 4790 mutex_enter(&pidlock);
4781 4791 PID_RELE(pidp);
4782 4792 mutex_exit(&pidlock);
4783 4793 return (EINVAL);
4784 4794 }
4785 4795
4786 4796 /*
4787 4797 * If proc not already registered, add it
4788 4798 * to list.
4789 4799 */
4790 4800 if (!ssp) {
4791 4801 ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4792 4802 ssp->ss_pidp = pidp;
4793 4803 ssp->ss_pid = pid;
4794 4804 ssp->ss_next = NULL;
4795 4805 if (pssp)
4796 4806 pssp->ss_next = ssp;
4797 4807 else
4798 4808 stp->sd_siglist = ssp;
4799 4809 mutex_enter(&pidlock);
4800 4810 PID_HOLD(pidp);
4801 4811 mutex_exit(&pidlock);
4802 4812 }
4803 4813
4804 4814 /*
4805 4815 * Set events.
4806 4816 */
4807 4817 ssp->ss_events = ss.ss_events;
4808 4818 } else {
4809 4819 /*
4810 4820 * Remove proc from register list.
4811 4821 */
4812 4822 if (ssp) {
4813 4823 mutex_enter(&pidlock);
4814 4824 PID_RELE(pidp);
4815 4825 mutex_exit(&pidlock);
4816 4826 if (pssp)
4817 4827 pssp->ss_next = ssp->ss_next;
4818 4828 else
4819 4829 stp->sd_siglist = ssp->ss_next;
4820 4830 kmem_free(ssp, sizeof (strsig_t));
4821 4831 } else {
4822 4832 mutex_exit(&stp->sd_lock);
4823 4833 mutex_enter(&pidlock);
4824 4834 PID_RELE(pidp);
4825 4835 mutex_exit(&pidlock);
4826 4836 return (EINVAL);
4827 4837 }
4828 4838 }
4829 4839
4830 4840 /*
4831 4841 * Recalculate OR of sig events.
4832 4842 */
4833 4843 stp->sd_sigflags = 0;
4834 4844 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4835 4845 stp->sd_sigflags |= ssp->ss_events;
4836 4846 mutex_exit(&stp->sd_lock);
4837 4847 mutex_enter(&pidlock);
4838 4848 PID_RELE(pidp);
4839 4849 mutex_exit(&pidlock);
4840 4850 return (0);
4841 4851 }
4842 4852
4843 4853 case I_EGETSIG:
4844 4854 /*
4845 4855 * Return (in arg) the current registration of events
4846 4856 * for which the calling proc is to be signaled.
4847 4857 */
4848 4858 {
4849 4859 struct strsig *ssp;
4850 4860 struct proc *proc;
4851 4861 pid_t pid;
4852 4862 struct pid *pidp;
4853 4863 struct strsigset ss;
4854 4864
4855 4865 error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4856 4866 if (error)
4857 4867 return (error);
4858 4868
4859 4869 pid = ss.ss_pid;
4860 4870 mutex_enter(&pidlock);
4861 4871 if (pid == 0)
4862 4872 proc = curproc;
4863 4873 else if (pid < 0)
4864 4874 proc = pgfind(-pid);
4865 4875 else
4866 4876 proc = prfind(pid);
4867 4877 if (proc == NULL) {
4868 4878 mutex_exit(&pidlock);
4869 4879 return (ESRCH);
4870 4880 }
4871 4881 if (pid < 0)
4872 4882 pidp = proc->p_pgidp;
4873 4883 else
4874 4884 pidp = proc->p_pidp;
4875 4885
4876 4886 /* Prevent the pidp from being reassigned */
4877 4887 PID_HOLD(pidp);
4878 4888 mutex_exit(&pidlock);
4879 4889
4880 4890 mutex_enter(&stp->sd_lock);
4881 4891 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4882 4892 if (ssp->ss_pid == pid) {
4883 4893 ss.ss_pid = ssp->ss_pid;
4884 4894 ss.ss_events = ssp->ss_events;
4885 4895 error = strcopyout(&ss, (void *)arg,
4886 4896 sizeof (struct strsigset), copyflag);
4887 4897 mutex_exit(&stp->sd_lock);
4888 4898 mutex_enter(&pidlock);
4889 4899 PID_RELE(pidp);
4890 4900 mutex_exit(&pidlock);
4891 4901 return (error);
4892 4902 }
4893 4903 mutex_exit(&stp->sd_lock);
4894 4904 mutex_enter(&pidlock);
4895 4905 PID_RELE(pidp);
4896 4906 mutex_exit(&pidlock);
4897 4907 return (EINVAL);
4898 4908 }
4899 4909
4900 4910 case I_PEEK:
4901 4911 {
4902 4912 STRUCT_DECL(strpeek, strpeek);
4903 4913 size_t n;
4904 4914 mblk_t *fmp, *tmp_mp = NULL;
4905 4915
4906 4916 STRUCT_INIT(strpeek, flag);
4907 4917
4908 4918 error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4909 4919 STRUCT_SIZE(strpeek), copyflag);
4910 4920 if (error)
4911 4921 return (error);
4912 4922
4913 4923 mutex_enter(QLOCK(rdq));
4914 4924 /*
4915 4925 * Skip the invalid messages
4916 4926 */
4917 4927 for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4918 4928 if (mp->b_datap->db_type != M_SIG)
4919 4929 break;
4920 4930
4921 4931 /*
4922 4932 * If user has requested to peek at a high priority message
4923 4933 * and first message is not, return 0
4924 4934 */
4925 4935 if (mp != NULL) {
4926 4936 if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4927 4937 queclass(mp) == QNORM) {
4928 4938 *rvalp = 0;
4929 4939 mutex_exit(QLOCK(rdq));
4930 4940 return (0);
4931 4941 }
4932 4942 } else if (stp->sd_struiordq == NULL ||
4933 4943 (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4934 4944 /*
4935 4945 * No mblks to look at at the streamhead and
4936 4946 * 1). This isn't a synch stream or
4937 4947 * 2). This is a synch stream but caller wants high
4938 4948 * priority messages which is not supported by
4939 4949 * the synch stream. (it only supports QNORM)
4940 4950 */
4941 4951 *rvalp = 0;
4942 4952 mutex_exit(QLOCK(rdq));
4943 4953 return (0);
4944 4954 }
4945 4955
4946 4956 fmp = mp;
4947 4957
4948 4958 if (mp && mp->b_datap->db_type == M_PASSFP) {
4949 4959 mutex_exit(QLOCK(rdq));
4950 4960 return (EBADMSG);
4951 4961 }
4952 4962
4953 4963 ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4954 4964 mp->b_datap->db_type == M_PROTO ||
4955 4965 mp->b_datap->db_type == M_DATA);
4956 4966
4957 4967 if (mp && mp->b_datap->db_type == M_PCPROTO) {
4958 4968 STRUCT_FSET(strpeek, flags, RS_HIPRI);
4959 4969 } else {
4960 4970 STRUCT_FSET(strpeek, flags, 0);
4961 4971 }
4962 4972
4963 4973
4964 4974 if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
4965 4975 mutex_exit(QLOCK(rdq));
4966 4976 return (ENOSR);
4967 4977 }
4968 4978 mutex_exit(QLOCK(rdq));
4969 4979
4970 4980 /*
4971 4981 * set mp = tmp_mp, so that I_PEEK processing can continue.
4972 4982 * tmp_mp is used to free the dup'd message.
4973 4983 */
4974 4984 mp = tmp_mp;
4975 4985
4976 4986 uio.uio_fmode = 0;
4977 4987 uio.uio_extflg = UIO_COPY_CACHED;
4978 4988 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
4979 4989 UIO_SYSSPACE;
4980 4990 uio.uio_limit = 0;
4981 4991 /*
4982 4992 * First process PROTO blocks, if any.
4983 4993 * If user doesn't want to get ctl info by setting maxlen <= 0,
4984 4994 * then set len to -1/0 and skip control blocks part.
4985 4995 */
4986 4996 if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
4987 4997 STRUCT_FSET(strpeek, ctlbuf.len, -1);
4988 4998 else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
4989 4999 STRUCT_FSET(strpeek, ctlbuf.len, 0);
4990 5000 else {
4991 5001 int ctl_part = 0;
4992 5002
4993 5003 iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
4994 5004 iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
4995 5005 uio.uio_iov = &iov;
4996 5006 uio.uio_resid = iov.iov_len;
4997 5007 uio.uio_loffset = 0;
4998 5008 uio.uio_iovcnt = 1;
4999 5009 while (mp && mp->b_datap->db_type != M_DATA &&
5000 5010 uio.uio_resid >= 0) {
5001 5011 ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5002 5012 mp->b_datap->db_type == M_PROTO :
5003 5013 mp->b_datap->db_type == M_PCPROTO);
5004 5014
5005 5015 if ((n = MIN(uio.uio_resid,
5006 5016 mp->b_wptr - mp->b_rptr)) != 0 &&
5007 5017 (error = uiomove((char *)mp->b_rptr, n,
5008 5018 UIO_READ, &uio)) != 0) {
5009 5019 freemsg(tmp_mp);
5010 5020 return (error);
5011 5021 }
5012 5022 ctl_part = 1;
5013 5023 mp = mp->b_cont;
5014 5024 }
5015 5025 /* No ctl message */
5016 5026 if (ctl_part == 0)
5017 5027 STRUCT_FSET(strpeek, ctlbuf.len, -1);
5018 5028 else
5019 5029 STRUCT_FSET(strpeek, ctlbuf.len,
5020 5030 STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5021 5031 uio.uio_resid);
5022 5032 }
5023 5033
5024 5034 /*
5025 5035 * Now process DATA blocks, if any.
5026 5036 * If user doesn't want to get data info by setting maxlen <= 0,
5027 5037 * then set len to -1/0 and skip data blocks part.
5028 5038 */
5029 5039 if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5030 5040 STRUCT_FSET(strpeek, databuf.len, -1);
5031 5041 else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5032 5042 STRUCT_FSET(strpeek, databuf.len, 0);
5033 5043 else {
5034 5044 int data_part = 0;
5035 5045
5036 5046 iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5037 5047 iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5038 5048 uio.uio_iov = &iov;
5039 5049 uio.uio_resid = iov.iov_len;
5040 5050 uio.uio_loffset = 0;
5041 5051 uio.uio_iovcnt = 1;
5042 5052 while (mp && uio.uio_resid) {
5043 5053 if (mp->b_datap->db_type == M_DATA) {
5044 5054 if ((n = MIN(uio.uio_resid,
5045 5055 mp->b_wptr - mp->b_rptr)) != 0 &&
5046 5056 (error = uiomove((char *)mp->b_rptr,
5047 5057 n, UIO_READ, &uio)) != 0) {
5048 5058 freemsg(tmp_mp);
5049 5059 return (error);
5050 5060 }
5051 5061 data_part = 1;
5052 5062 }
5053 5063 ASSERT(data_part == 0 ||
5054 5064 mp->b_datap->db_type == M_DATA);
5055 5065 mp = mp->b_cont;
5056 5066 }
5057 5067 /* No data message */
5058 5068 if (data_part == 0)
5059 5069 STRUCT_FSET(strpeek, databuf.len, -1);
5060 5070 else
5061 5071 STRUCT_FSET(strpeek, databuf.len,
5062 5072 STRUCT_FGET(strpeek, databuf.maxlen) -
5063 5073 uio.uio_resid);
5064 5074 }
5065 5075 freemsg(tmp_mp);
5066 5076
5067 5077 /*
5068 5078 * It is a synch stream and user wants to get
5069 5079 * data (maxlen > 0).
5070 5080 * uio setup is done by the codes that process DATA
5071 5081 * blocks above.
5072 5082 */
5073 5083 if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5074 5084 infod_t infod;
5075 5085
5076 5086 infod.d_cmd = INFOD_COPYOUT;
5077 5087 infod.d_res = 0;
5078 5088 infod.d_uiop = &uio;
5079 5089 error = infonext(rdq, &infod);
5080 5090 if (error == EINVAL || error == EBUSY)
5081 5091 error = 0;
5082 5092 if (error)
5083 5093 return (error);
5084 5094 STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5085 5095 databuf.maxlen) - uio.uio_resid);
5086 5096 if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5087 5097 /*
5088 5098 * No data found by the infonext().
5089 5099 */
5090 5100 STRUCT_FSET(strpeek, databuf.len, -1);
5091 5101 }
5092 5102 }
5093 5103 error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5094 5104 STRUCT_SIZE(strpeek), copyflag);
5095 5105 if (error) {
5096 5106 return (error);
5097 5107 }
5098 5108 /*
5099 5109 * If there is no message retrieved, set return code to 0
5100 5110 * otherwise, set it to 1.
5101 5111 */
5102 5112 if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5103 5113 STRUCT_FGET(strpeek, databuf.len) == -1)
5104 5114 *rvalp = 0;
5105 5115 else
5106 5116 *rvalp = 1;
5107 5117 return (0);
5108 5118 }
5109 5119
5110 5120 case I_FDINSERT:
5111 5121 {
5112 5122 STRUCT_DECL(strfdinsert, strfdinsert);
5113 5123 struct file *resftp;
5114 5124 struct stdata *resstp;
5115 5125 t_uscalar_t ival;
5116 5126 ssize_t msgsize;
5117 5127 struct strbuf mctl;
5118 5128
5119 5129 STRUCT_INIT(strfdinsert, flag);
5120 5130 if (stp->sd_flag & STRHUP)
5121 5131 return (ENXIO);
5122 5132 /*
5123 5133 * STRDERR, STWRERR and STPLEX tested above.
5124 5134 */
5125 5135 error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5126 5136 STRUCT_SIZE(strfdinsert), copyflag);
5127 5137 if (error)
5128 5138 return (error);
5129 5139
5130 5140 if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5131 5141 (STRUCT_FGET(strfdinsert, offset) %
5132 5142 sizeof (t_uscalar_t)) != 0)
5133 5143 return (EINVAL);
5134 5144 if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5135 5145 if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5136 5146 releasef(STRUCT_FGET(strfdinsert, fildes));
5137 5147 return (EINVAL);
5138 5148 }
5139 5149 } else
5140 5150 return (EINVAL);
5141 5151
5142 5152 mutex_enter(&resstp->sd_lock);
5143 5153 if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5144 5154 error = strgeterr(resstp,
5145 5155 STRDERR|STWRERR|STRHUP|STPLEX, 0);
5146 5156 if (error != 0) {
5147 5157 mutex_exit(&resstp->sd_lock);
5148 5158 releasef(STRUCT_FGET(strfdinsert, fildes));
5149 5159 return (error);
5150 5160 }
5151 5161 }
5152 5162 mutex_exit(&resstp->sd_lock);
5153 5163
5154 5164 #ifdef _ILP32
5155 5165 {
5156 5166 queue_t *q;
5157 5167 queue_t *mate = NULL;
5158 5168
5159 5169 /* get read queue of stream terminus */
5160 5170 claimstr(resstp->sd_wrq);
5161 5171 for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5162 5172 q = q->q_next)
5163 5173 if (!STRMATED(resstp) && STREAM(q) != resstp &&
5164 5174 mate == NULL) {
5165 5175 ASSERT(q->q_qinfo->qi_srvp);
5166 5176 ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5167 5177 claimstr(q);
5168 5178 mate = q;
5169 5179 }
5170 5180 q = _RD(q);
5171 5181 if (mate)
5172 5182 releasestr(mate);
5173 5183 releasestr(resstp->sd_wrq);
5174 5184 ival = (t_uscalar_t)q;
5175 5185 }
5176 5186 #else
5177 5187 ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5178 5188 #endif /* _ILP32 */
5179 5189
5180 5190 if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5181 5191 STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5182 5192 releasef(STRUCT_FGET(strfdinsert, fildes));
5183 5193 return (EINVAL);
5184 5194 }
5185 5195
5186 5196 /*
5187 5197 * Check for legal flag value.
5188 5198 */
5189 5199 if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5190 5200 releasef(STRUCT_FGET(strfdinsert, fildes));
5191 5201 return (EINVAL);
5192 5202 }
5193 5203
5194 5204 /* get these values from those cached in the stream head */
5195 5205 mutex_enter(QLOCK(stp->sd_wrq));
5196 5206 rmin = stp->sd_qn_minpsz;
5197 5207 rmax = stp->sd_qn_maxpsz;
5198 5208 mutex_exit(QLOCK(stp->sd_wrq));
5199 5209
5200 5210 /*
5201 5211 * Make sure ctl and data sizes together fall within
5202 5212 * the limits of the max and min receive packet sizes
5203 5213 * and do not exceed system limit. A negative data
5204 5214 * length means that no data part is to be sent.
5205 5215 */
5206 5216 ASSERT((rmax >= 0) || (rmax == INFPSZ));
5207 5217 if (rmax == 0) {
5208 5218 releasef(STRUCT_FGET(strfdinsert, fildes));
5209 5219 return (ERANGE);
5210 5220 }
5211 5221 if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5212 5222 msgsize = 0;
5213 5223 if ((msgsize < rmin) ||
5214 5224 ((msgsize > rmax) && (rmax != INFPSZ)) ||
5215 5225 (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5216 5226 releasef(STRUCT_FGET(strfdinsert, fildes));
5217 5227 return (ERANGE);
5218 5228 }
5219 5229
5220 5230 mutex_enter(&stp->sd_lock);
5221 5231 while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5222 5232 !canputnext(stp->sd_wrq)) {
5223 5233 if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5224 5234 flag, -1, &done)) != 0 || done) {
5225 5235 mutex_exit(&stp->sd_lock);
5226 5236 releasef(STRUCT_FGET(strfdinsert, fildes));
5227 5237 return (error);
5228 5238 }
5229 5239 if ((error = i_straccess(stp, access)) != 0) {
5230 5240 mutex_exit(&stp->sd_lock);
5231 5241 releasef(
5232 5242 STRUCT_FGET(strfdinsert, fildes));
5233 5243 return (error);
5234 5244 }
5235 5245 }
5236 5246 mutex_exit(&stp->sd_lock);
5237 5247
5238 5248 /*
5239 5249 * Copy strfdinsert.ctlbuf into native form of
5240 5250 * ctlbuf to pass down into strmakemsg().
5241 5251 */
5242 5252 mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5243 5253 mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5244 5254 mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5245 5255
5246 5256 iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5247 5257 iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5248 5258 uio.uio_iov = &iov;
5249 5259 uio.uio_iovcnt = 1;
5250 5260 uio.uio_loffset = 0;
5251 5261 uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5252 5262 UIO_SYSSPACE;
5253 5263 uio.uio_fmode = 0;
5254 5264 uio.uio_extflg = UIO_COPY_CACHED;
5255 5265 uio.uio_resid = iov.iov_len;
5256 5266 if ((error = strmakemsg(&mctl,
5257 5267 &msgsize, &uio, stp,
5258 5268 STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5259 5269 STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5260 5270 releasef(STRUCT_FGET(strfdinsert, fildes));
5261 5271 return (error);
5262 5272 }
5263 5273
5264 5274 STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5265 5275
5266 5276 /*
5267 5277 * Place the possibly reencoded queue pointer 'offset' bytes
5268 5278 * from the start of the control portion of the message.
5269 5279 */
5270 5280 *((t_uscalar_t *)(mp->b_rptr +
5271 5281 STRUCT_FGET(strfdinsert, offset))) = ival;
5272 5282
5273 5283 /*
5274 5284 * Put message downstream.
5275 5285 */
5276 5286 stream_willservice(stp);
5277 5287 putnext(stp->sd_wrq, mp);
5278 5288 stream_runservice(stp);
5279 5289 releasef(STRUCT_FGET(strfdinsert, fildes));
5280 5290 return (error);
5281 5291 }
5282 5292
5283 5293 case I_SENDFD:
5284 5294 {
5285 5295 struct file *fp;
5286 5296
5287 5297 if ((fp = getf((int)arg)) == NULL)
5288 5298 return (EBADF);
5289 5299 error = do_sendfp(stp, fp, crp);
5290 5300 if (auditing) {
5291 5301 audit_fdsend((int)arg, fp, error);
5292 5302 }
5293 5303 releasef((int)arg);
5294 5304 return (error);
5295 5305 }
5296 5306
5297 5307 case I_RECVFD:
5298 5308 case I_E_RECVFD:
5299 5309 {
5300 5310 struct k_strrecvfd *srf;
5301 5311 int i, fd;
5302 5312
5303 5313 mutex_enter(&stp->sd_lock);
5304 5314 while (!(mp = getq(rdq))) {
5305 5315 if (stp->sd_flag & (STRHUP|STREOF)) {
5306 5316 mutex_exit(&stp->sd_lock);
5307 5317 return (ENXIO);
5308 5318 }
5309 5319 if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5310 5320 flag, -1, &done)) != 0 || done) {
5311 5321 mutex_exit(&stp->sd_lock);
5312 5322 return (error);
5313 5323 }
5314 5324 if ((error = i_straccess(stp, access)) != 0) {
5315 5325 mutex_exit(&stp->sd_lock);
5316 5326 return (error);
5317 5327 }
5318 5328 }
5319 5329 if (mp->b_datap->db_type != M_PASSFP) {
5320 5330 putback(stp, rdq, mp, mp->b_band);
5321 5331 mutex_exit(&stp->sd_lock);
5322 5332 return (EBADMSG);
5323 5333 }
5324 5334 mutex_exit(&stp->sd_lock);
5325 5335
5326 5336 srf = (struct k_strrecvfd *)mp->b_rptr;
5327 5337 if ((fd = ufalloc(0)) == -1) {
5328 5338 mutex_enter(&stp->sd_lock);
5329 5339 putback(stp, rdq, mp, mp->b_band);
5330 5340 mutex_exit(&stp->sd_lock);
5331 5341 return (EMFILE);
5332 5342 }
5333 5343 if (cmd == I_RECVFD) {
5334 5344 struct o_strrecvfd ostrfd;
5335 5345
5336 5346 /* check to see if uid/gid values are too large. */
5337 5347
5338 5348 if (srf->uid > (o_uid_t)USHRT_MAX ||
5339 5349 srf->gid > (o_gid_t)USHRT_MAX) {
5340 5350 mutex_enter(&stp->sd_lock);
5341 5351 putback(stp, rdq, mp, mp->b_band);
5342 5352 mutex_exit(&stp->sd_lock);
5343 5353 setf(fd, NULL); /* release fd entry */
5344 5354 return (EOVERFLOW);
5345 5355 }
5346 5356
5347 5357 ostrfd.fd = fd;
5348 5358 ostrfd.uid = (o_uid_t)srf->uid;
5349 5359 ostrfd.gid = (o_gid_t)srf->gid;
5350 5360
5351 5361 /* Null the filler bits */
5352 5362 for (i = 0; i < 8; i++)
5353 5363 ostrfd.fill[i] = 0;
5354 5364
5355 5365 error = strcopyout(&ostrfd, (void *)arg,
5356 5366 sizeof (struct o_strrecvfd), copyflag);
5357 5367 } else { /* I_E_RECVFD */
5358 5368 struct strrecvfd strfd;
5359 5369
5360 5370 strfd.fd = fd;
5361 5371 strfd.uid = srf->uid;
5362 5372 strfd.gid = srf->gid;
5363 5373
5364 5374 /* null the filler bits */
5365 5375 for (i = 0; i < 8; i++)
5366 5376 strfd.fill[i] = 0;
5367 5377
5368 5378 error = strcopyout(&strfd, (void *)arg,
5369 5379 sizeof (struct strrecvfd), copyflag);
5370 5380 }
5371 5381
5372 5382 if (error) {
5373 5383 setf(fd, NULL); /* release fd entry */
5374 5384 mutex_enter(&stp->sd_lock);
5375 5385 putback(stp, rdq, mp, mp->b_band);
5376 5386 mutex_exit(&stp->sd_lock);
5377 5387 return (error);
5378 5388 }
5379 5389 if (auditing) {
5380 5390 audit_fdrecv(fd, srf->fp);
5381 5391 }
5382 5392
5383 5393 /*
5384 5394 * Always increment f_count since the freemsg() below will
5385 5395 * always call free_passfp() which performs a closef().
5386 5396 */
5387 5397 mutex_enter(&srf->fp->f_tlock);
5388 5398 srf->fp->f_count++;
5389 5399 mutex_exit(&srf->fp->f_tlock);
5390 5400 setf(fd, srf->fp);
5391 5401 freemsg(mp);
5392 5402 return (0);
5393 5403 }
5394 5404
5395 5405 case I_SWROPT:
5396 5406 /*
5397 5407 * Set/clear the write options. arg is a bit
5398 5408 * mask with any of the following bits set...
5399 5409 * SNDZERO - send zero length message
5400 5410 * SNDPIPE - send sigpipe to process if
5401 5411 * sd_werror is set and process is
5402 5412 * doing a write or putmsg.
5403 5413 * The new stream head write options should reflect
5404 5414 * what is in arg.
5405 5415 */
5406 5416 if (arg & ~(SNDZERO|SNDPIPE))
5407 5417 return (EINVAL);
5408 5418
5409 5419 mutex_enter(&stp->sd_lock);
5410 5420 stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5411 5421 if (arg & SNDZERO)
5412 5422 stp->sd_wput_opt |= SW_SNDZERO;
5413 5423 if (arg & SNDPIPE)
5414 5424 stp->sd_wput_opt |= SW_SIGPIPE;
5415 5425 mutex_exit(&stp->sd_lock);
5416 5426 return (0);
5417 5427
5418 5428 case I_GWROPT:
5419 5429 {
5420 5430 int wropt = 0;
5421 5431
5422 5432 if (stp->sd_wput_opt & SW_SNDZERO)
5423 5433 wropt |= SNDZERO;
5424 5434 if (stp->sd_wput_opt & SW_SIGPIPE)
5425 5435 wropt |= SNDPIPE;
5426 5436 return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5427 5437 copyflag));
5428 5438 }
5429 5439
5430 5440 case I_LIST:
5431 5441 /*
5432 5442 * Returns all the modules found on this stream,
5433 5443 * upto the driver. If argument is NULL, return the
5434 5444 * number of modules (including driver). If argument
5435 5445 * is not NULL, copy the names into the structure
5436 5446 * provided.
5437 5447 */
5438 5448
5439 5449 {
5440 5450 queue_t *q;
5441 5451 char *qname;
5442 5452 int i, nmods;
5443 5453 struct str_mlist *mlist;
5444 5454 STRUCT_DECL(str_list, strlist);
5445 5455
5446 5456 if (arg == NULL) { /* Return number of modules plus driver */
5447 5457 if (stp->sd_vnode->v_type == VFIFO)
5448 5458 *rvalp = stp->sd_pushcnt;
5449 5459 else
5450 5460 *rvalp = stp->sd_pushcnt + 1;
5451 5461 return (0);
5452 5462 }
5453 5463
5454 5464 STRUCT_INIT(strlist, flag);
5455 5465
5456 5466 error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5457 5467 STRUCT_SIZE(strlist), copyflag);
5458 5468 if (error != 0)
5459 5469 return (error);
5460 5470
5461 5471 mlist = STRUCT_FGETP(strlist, sl_modlist);
5462 5472 nmods = STRUCT_FGET(strlist, sl_nmods);
5463 5473 if (nmods <= 0)
5464 5474 return (EINVAL);
5465 5475
5466 5476 claimstr(stp->sd_wrq);
5467 5477 q = stp->sd_wrq;
5468 5478 for (i = 0; i < nmods && _SAMESTR(q); i++, q = q->q_next) {
5469 5479 qname = Q2NAME(q->q_next);
5470 5480 error = strcopyout(qname, &mlist[i], strlen(qname) + 1,
5471 5481 copyflag);
5472 5482 if (error != 0) {
5473 5483 releasestr(stp->sd_wrq);
5474 5484 return (error);
5475 5485 }
5476 5486 }
5477 5487 releasestr(stp->sd_wrq);
5478 5488 return (strcopyout(&i, (void *)arg, sizeof (int), copyflag));
5479 5489 }
5480 5490
5481 5491 case I_CKBAND:
5482 5492 {
5483 5493 queue_t *q;
5484 5494 qband_t *qbp;
5485 5495
5486 5496 if ((arg < 0) || (arg >= NBAND))
5487 5497 return (EINVAL);
5488 5498 q = _RD(stp->sd_wrq);
5489 5499 mutex_enter(QLOCK(q));
5490 5500 if (arg > (int)q->q_nband) {
5491 5501 *rvalp = 0;
5492 5502 } else {
5493 5503 if (arg == 0) {
5494 5504 if (q->q_first)
5495 5505 *rvalp = 1;
5496 5506 else
5497 5507 *rvalp = 0;
5498 5508 } else {
5499 5509 qbp = q->q_bandp;
5500 5510 while (--arg > 0)
5501 5511 qbp = qbp->qb_next;
5502 5512 if (qbp->qb_first)
5503 5513 *rvalp = 1;
5504 5514 else
5505 5515 *rvalp = 0;
5506 5516 }
5507 5517 }
5508 5518 mutex_exit(QLOCK(q));
5509 5519 return (0);
5510 5520 }
5511 5521
5512 5522 case I_GETBAND:
5513 5523 {
5514 5524 int intpri;
5515 5525 queue_t *q;
5516 5526
5517 5527 q = _RD(stp->sd_wrq);
5518 5528 mutex_enter(QLOCK(q));
5519 5529 mp = q->q_first;
5520 5530 if (!mp) {
5521 5531 mutex_exit(QLOCK(q));
5522 5532 return (ENODATA);
5523 5533 }
5524 5534 intpri = (int)mp->b_band;
5525 5535 error = strcopyout(&intpri, (void *)arg, sizeof (int),
5526 5536 copyflag);
5527 5537 mutex_exit(QLOCK(q));
5528 5538 return (error);
5529 5539 }
5530 5540
5531 5541 case I_ATMARK:
5532 5542 {
5533 5543 queue_t *q;
5534 5544
5535 5545 if (arg & ~(ANYMARK|LASTMARK))
5536 5546 return (EINVAL);
5537 5547 q = _RD(stp->sd_wrq);
5538 5548 mutex_enter(&stp->sd_lock);
5539 5549 if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5540 5550 *rvalp = 1;
5541 5551 } else {
5542 5552 mutex_enter(QLOCK(q));
5543 5553 mp = q->q_first;
5544 5554
5545 5555 if (mp == NULL)
5546 5556 *rvalp = 0;
5547 5557 else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5548 5558 *rvalp = 1;
5549 5559 else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5550 5560 *rvalp = 1;
5551 5561 else
5552 5562 *rvalp = 0;
5553 5563 mutex_exit(QLOCK(q));
5554 5564 }
5555 5565 mutex_exit(&stp->sd_lock);
5556 5566 return (0);
5557 5567 }
5558 5568
5559 5569 case I_CANPUT:
5560 5570 {
5561 5571 char band;
5562 5572
5563 5573 if ((arg < 0) || (arg >= NBAND))
5564 5574 return (EINVAL);
5565 5575 band = (char)arg;
5566 5576 *rvalp = bcanputnext(stp->sd_wrq, band);
5567 5577 return (0);
5568 5578 }
5569 5579
5570 5580 case I_SETCLTIME:
5571 5581 {
5572 5582 int closetime;
5573 5583
5574 5584 error = strcopyin((void *)arg, &closetime, sizeof (int),
5575 5585 copyflag);
5576 5586 if (error)
5577 5587 return (error);
5578 5588 if (closetime < 0)
5579 5589 return (EINVAL);
5580 5590
5581 5591 stp->sd_closetime = closetime;
5582 5592 return (0);
5583 5593 }
5584 5594
5585 5595 case I_GETCLTIME:
5586 5596 {
5587 5597 int closetime;
5588 5598
5589 5599 closetime = stp->sd_closetime;
5590 5600 return (strcopyout(&closetime, (void *)arg, sizeof (int),
5591 5601 copyflag));
5592 5602 }
5593 5603
5594 5604 case TIOCGSID:
5595 5605 {
5596 5606 pid_t sid;
5597 5607
5598 5608 mutex_enter(&stp->sd_lock);
5599 5609 if (stp->sd_sidp == NULL) {
5600 5610 mutex_exit(&stp->sd_lock);
5601 5611 return (ENOTTY);
5602 5612 }
5603 5613 sid = stp->sd_sidp->pid_id;
5604 5614 mutex_exit(&stp->sd_lock);
5605 5615 return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5606 5616 copyflag));
5607 5617 }
5608 5618
5609 5619 case TIOCSPGRP:
5610 5620 {
5611 5621 pid_t pgrp;
5612 5622 proc_t *q;
5613 5623 pid_t sid, fg_pgid, bg_pgid;
5614 5624
5615 5625 if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5616 5626 copyflag))
5617 5627 return (error);
5618 5628 mutex_enter(&stp->sd_lock);
5619 5629 mutex_enter(&pidlock);
5620 5630 if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5621 5631 mutex_exit(&pidlock);
5622 5632 mutex_exit(&stp->sd_lock);
5623 5633 return (ENOTTY);
5624 5634 }
5625 5635 if (pgrp == stp->sd_pgidp->pid_id) {
5626 5636 mutex_exit(&pidlock);
5627 5637 mutex_exit(&stp->sd_lock);
5628 5638 return (0);
5629 5639 }
5630 5640 if (pgrp <= 0 || pgrp >= maxpid) {
5631 5641 mutex_exit(&pidlock);
5632 5642 mutex_exit(&stp->sd_lock);
5633 5643 return (EINVAL);
5634 5644 }
5635 5645 if ((q = pgfind(pgrp)) == NULL ||
5636 5646 q->p_sessp != ttoproc(curthread)->p_sessp) {
5637 5647 mutex_exit(&pidlock);
5638 5648 mutex_exit(&stp->sd_lock);
5639 5649 return (EPERM);
5640 5650 }
5641 5651 sid = stp->sd_sidp->pid_id;
5642 5652 fg_pgid = q->p_pgrp;
5643 5653 bg_pgid = stp->sd_pgidp->pid_id;
5644 5654 CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5645 5655 PID_RELE(stp->sd_pgidp);
5646 5656 ctty_clear_sighuped();
5647 5657 stp->sd_pgidp = q->p_pgidp;
5648 5658 PID_HOLD(stp->sd_pgidp);
5649 5659 mutex_exit(&pidlock);
5650 5660 mutex_exit(&stp->sd_lock);
5651 5661 return (0);
5652 5662 }
5653 5663
5654 5664 case TIOCGPGRP:
5655 5665 {
5656 5666 pid_t pgrp;
5657 5667
5658 5668 mutex_enter(&stp->sd_lock);
5659 5669 if (stp->sd_sidp == NULL) {
5660 5670 mutex_exit(&stp->sd_lock);
5661 5671 return (ENOTTY);
5662 5672 }
5663 5673 pgrp = stp->sd_pgidp->pid_id;
5664 5674 mutex_exit(&stp->sd_lock);
5665 5675 return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5666 5676 copyflag));
5667 5677 }
5668 5678
5669 5679 case TIOCSCTTY:
5670 5680 {
5671 5681 return (strctty(stp));
5672 5682 }
5673 5683
5674 5684 case TIOCNOTTY:
↓ open down ↓ |
5266 lines elided |
↑ open up ↑ |
5675 5685 {
5676 5686 /* freectty() always assumes curproc. */
5677 5687 if (freectty(B_FALSE) != 0)
5678 5688 return (0);
5679 5689 return (ENOTTY);
5680 5690 }
5681 5691
5682 5692 case FIONBIO:
5683 5693 case FIOASYNC:
5684 5694 return (0); /* handled by the upper layer */
5695 + case F_ASSOCI_PID:
5696 + {
5697 + if (crp != kcred)
5698 + return (EPERM);
5699 + if (is_xti_str(stp))
5700 + sh_insert_pid(stp, (pid_t)arg);
5701 + return (0);
5702 + }
5703 + case F_DASSOC_PID:
5704 + {
5705 + if (crp != kcred)
5706 + return (EPERM);
5707 + if (is_xti_str(stp))
5708 + sh_remove_pid(stp, (pid_t)arg);
5709 + return (0);
5710 + }
5685 5711 }
5686 5712 }
5687 5713
5688 5714 /*
5689 5715 * Custom free routine used for M_PASSFP messages.
5690 5716 */
5691 5717 static void
5692 5718 free_passfp(struct k_strrecvfd *srf)
5693 5719 {
5694 5720 (void) closef(srf->fp);
5695 5721 kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5696 5722 }
5697 5723
5698 5724 /* ARGSUSED */
5699 5725 int
5700 5726 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5701 5727 {
5702 5728 queue_t *qp, *nextqp;
5703 5729 struct k_strrecvfd *srf;
5704 5730 mblk_t *mp;
5705 5731 frtn_t *frtnp;
5706 5732 size_t bufsize;
5707 5733 queue_t *mate = NULL;
5708 5734 syncq_t *sq = NULL;
5709 5735 int retval = 0;
5710 5736
5711 5737 if (stp->sd_flag & STRHUP)
5712 5738 return (ENXIO);
5713 5739
5714 5740 claimstr(stp->sd_wrq);
5715 5741
5716 5742 /* Fastpath, we have a pipe, and we are already mated, use it. */
5717 5743 if (STRMATED(stp)) {
5718 5744 qp = _RD(stp->sd_mate->sd_wrq);
5719 5745 claimstr(qp);
5720 5746 mate = qp;
5721 5747 } else { /* Not already mated. */
5722 5748
5723 5749 /*
5724 5750 * Walk the stream to the end of this one.
5725 5751 * assumes that the claimstr() will prevent
5726 5752 * plumbing between the stream head and the
5727 5753 * driver from changing
5728 5754 */
5729 5755 qp = stp->sd_wrq;
5730 5756
5731 5757 /*
5732 5758 * Loop until we reach the end of this stream.
5733 5759 * On completion, qp points to the write queue
5734 5760 * at the end of the stream, or the read queue
5735 5761 * at the stream head if this is a fifo.
5736 5762 */
5737 5763 while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5738 5764 ;
5739 5765
5740 5766 /*
5741 5767 * Just in case we get a q_next which is NULL, but
5742 5768 * not at the end of the stream. This is actually
5743 5769 * broken, so we set an assert to catch it in
5744 5770 * debug, and set an error and return if not debug.
5745 5771 */
5746 5772 ASSERT(qp);
5747 5773 if (qp == NULL) {
5748 5774 releasestr(stp->sd_wrq);
5749 5775 return (EINVAL);
5750 5776 }
5751 5777
5752 5778 /*
5753 5779 * Enter the syncq for the driver, so (hopefully)
5754 5780 * the queue values will not change on us.
5755 5781 * XXXX - This will only prevent the race IFF only
5756 5782 * the write side modifies the q_next member, and
5757 5783 * the put procedure is protected by at least
5758 5784 * MT_PERQ.
5759 5785 */
5760 5786 if ((sq = qp->q_syncq) != NULL)
5761 5787 entersq(sq, SQ_PUT);
5762 5788
5763 5789 /* Now get the q_next value from this qp. */
5764 5790 nextqp = qp->q_next;
5765 5791
5766 5792 /*
5767 5793 * If nextqp exists and the other stream is different
5768 5794 * from this one claim the stream, set the mate, and
5769 5795 * get the read queue at the stream head of the other
5770 5796 * stream. Assumes that nextqp was at least valid when
5771 5797 * we got it. Hopefully the entersq of the driver
5772 5798 * will prevent it from changing on us.
5773 5799 */
5774 5800 if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5775 5801 ASSERT(qp->q_qinfo->qi_srvp);
5776 5802 ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5777 5803 ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5778 5804 claimstr(nextqp);
5779 5805
5780 5806 /* Make sure we still have a q_next */
5781 5807 if (nextqp != qp->q_next) {
5782 5808 releasestr(stp->sd_wrq);
5783 5809 releasestr(nextqp);
5784 5810 return (EINVAL);
5785 5811 }
5786 5812
5787 5813 qp = _RD(STREAM(nextqp)->sd_wrq);
5788 5814 mate = qp;
5789 5815 }
5790 5816 /* If we entered the synq above, leave it. */
5791 5817 if (sq != NULL)
5792 5818 leavesq(sq, SQ_PUT);
5793 5819 } /* STRMATED(STP) */
5794 5820
5795 5821 /* XXX prevents substitution of the ops vector */
5796 5822 if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5797 5823 retval = EINVAL;
5798 5824 goto out;
5799 5825 }
5800 5826
5801 5827 if (qp->q_flag & QFULL) {
5802 5828 retval = EAGAIN;
5803 5829 goto out;
5804 5830 }
5805 5831
5806 5832 /*
5807 5833 * Since M_PASSFP messages include a file descriptor, we use
5808 5834 * esballoc() and specify a custom free routine (free_passfp()) that
5809 5835 * will close the descriptor as part of freeing the message. For
5810 5836 * convenience, we stash the frtn_t right after the data block.
5811 5837 */
5812 5838 bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5813 5839 srf = kmem_alloc(bufsize, KM_NOSLEEP);
5814 5840 if (srf == NULL) {
5815 5841 retval = EAGAIN;
5816 5842 goto out;
5817 5843 }
5818 5844
5819 5845 frtnp = (frtn_t *)(srf + 1);
5820 5846 frtnp->free_arg = (caddr_t)srf;
5821 5847 frtnp->free_func = free_passfp;
5822 5848
5823 5849 mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5824 5850 if (mp == NULL) {
5825 5851 kmem_free(srf, bufsize);
5826 5852 retval = EAGAIN;
5827 5853 goto out;
5828 5854 }
5829 5855 mp->b_wptr += sizeof (struct k_strrecvfd);
5830 5856 mp->b_datap->db_type = M_PASSFP;
5831 5857
5832 5858 srf->fp = fp;
5833 5859 srf->uid = crgetuid(curthread->t_cred);
5834 5860 srf->gid = crgetgid(curthread->t_cred);
5835 5861 mutex_enter(&fp->f_tlock);
5836 5862 fp->f_count++;
5837 5863 mutex_exit(&fp->f_tlock);
5838 5864
5839 5865 put(qp, mp);
5840 5866 out:
5841 5867 releasestr(stp->sd_wrq);
5842 5868 if (mate)
5843 5869 releasestr(mate);
5844 5870 return (retval);
5845 5871 }
5846 5872
5847 5873 /*
5848 5874 * Send an ioctl message downstream and wait for acknowledgement.
5849 5875 * flags may be set to either U_TO_K or K_TO_K and a combination
5850 5876 * of STR_NOERROR or STR_NOSIG
5851 5877 * STR_NOSIG: Signals are essentially ignored or held and have
5852 5878 * no effect for the duration of the call.
5853 5879 * STR_NOERROR: Ignores stream head read, write and hup errors.
5854 5880 * Additionally, if an existing ioctl times out, it is assumed
5855 5881 * lost and and this ioctl will continue as if the previous ioctl had
5856 5882 * finished. ETIME may be returned if this ioctl times out (i.e.
5857 5883 * ic_timout is not INFTIM). Non-stream head errors may be returned if
5858 5884 * the ioc_error indicates that the driver/module had problems,
5859 5885 * an EFAULT was found when accessing user data, a lack of
5860 5886 * resources, etc.
5861 5887 */
5862 5888 int
5863 5889 strdoioctl(
5864 5890 struct stdata *stp,
5865 5891 struct strioctl *strioc,
5866 5892 int fflags, /* file flags with model info */
5867 5893 int flag,
5868 5894 cred_t *crp,
5869 5895 int *rvalp)
5870 5896 {
5871 5897 mblk_t *bp;
5872 5898 struct iocblk *iocbp;
5873 5899 struct copyreq *reqp;
5874 5900 struct copyresp *resp;
5875 5901 int id;
5876 5902 int transparent = 0;
5877 5903 int error = 0;
5878 5904 int len = 0;
5879 5905 caddr_t taddr;
5880 5906 int copyflag = (flag & (U_TO_K | K_TO_K));
5881 5907 int sigflag = (flag & STR_NOSIG);
5882 5908 int errs;
5883 5909 uint_t waitflags;
5884 5910 boolean_t set_iocwaitne = B_FALSE;
5885 5911
5886 5912 ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5887 5913 ASSERT((fflags & FMODELS) != 0);
5888 5914
5889 5915 TRACE_2(TR_FAC_STREAMS_FR,
5890 5916 TR_STRDOIOCTL,
5891 5917 "strdoioctl:stp %p strioc %p", stp, strioc);
5892 5918 if (strioc->ic_len == TRANSPARENT) { /* send arg in M_DATA block */
5893 5919 transparent = 1;
5894 5920 strioc->ic_len = sizeof (intptr_t);
5895 5921 }
5896 5922
5897 5923 if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5898 5924 return (EINVAL);
5899 5925
5900 5926 if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5901 5927 crp, curproc->p_pid)) == NULL)
5902 5928 return (error);
5903 5929
5904 5930 bzero(bp->b_wptr, sizeof (union ioctypes));
5905 5931
5906 5932 iocbp = (struct iocblk *)bp->b_wptr;
5907 5933 iocbp->ioc_count = strioc->ic_len;
5908 5934 iocbp->ioc_cmd = strioc->ic_cmd;
5909 5935 iocbp->ioc_flag = (fflags & FMODELS);
5910 5936
5911 5937 crhold(crp);
5912 5938 iocbp->ioc_cr = crp;
5913 5939 DB_TYPE(bp) = M_IOCTL;
5914 5940 bp->b_wptr += sizeof (struct iocblk);
5915 5941
5916 5942 if (flag & STR_NOERROR)
5917 5943 errs = STPLEX;
5918 5944 else
5919 5945 errs = STRHUP|STRDERR|STWRERR|STPLEX;
5920 5946
5921 5947 /*
5922 5948 * If there is data to copy into ioctl block, do so.
5923 5949 */
5924 5950 if (iocbp->ioc_count > 0) {
5925 5951 if (transparent)
5926 5952 /*
5927 5953 * Note: STR_NOERROR does not have an effect
5928 5954 * in putiocd()
5929 5955 */
5930 5956 id = K_TO_K | sigflag;
5931 5957 else
5932 5958 id = flag;
5933 5959 if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5934 5960 freemsg(bp);
5935 5961 crfree(crp);
5936 5962 return (error);
5937 5963 }
5938 5964
5939 5965 /*
5940 5966 * We could have slept copying in user pages.
5941 5967 * Recheck the stream head state (the other end
5942 5968 * of a pipe could have gone away).
5943 5969 */
5944 5970 if (stp->sd_flag & errs) {
5945 5971 mutex_enter(&stp->sd_lock);
5946 5972 error = strgeterr(stp, errs, 0);
5947 5973 mutex_exit(&stp->sd_lock);
5948 5974 if (error != 0) {
5949 5975 freemsg(bp);
5950 5976 crfree(crp);
5951 5977 return (error);
5952 5978 }
5953 5979 }
5954 5980 }
5955 5981 if (transparent)
5956 5982 iocbp->ioc_count = TRANSPARENT;
5957 5983
5958 5984 /*
5959 5985 * Block for up to STRTIMOUT milliseconds if there is an outstanding
5960 5986 * ioctl for this stream already running. All processes
5961 5987 * sleeping here will be awakened as a result of an ACK
5962 5988 * or NAK being received for the outstanding ioctl, or
5963 5989 * as a result of the timer expiring on the outstanding
5964 5990 * ioctl (a failure), or as a result of any waiting
5965 5991 * process's timer expiring (also a failure).
5966 5992 */
5967 5993
5968 5994 error = 0;
5969 5995 mutex_enter(&stp->sd_lock);
5970 5996 while ((stp->sd_flag & IOCWAIT) ||
5971 5997 (!set_iocwaitne && (stp->sd_flag & IOCWAITNE))) {
5972 5998 clock_t cv_rval;
5973 5999
5974 6000 TRACE_0(TR_FAC_STREAMS_FR,
5975 6001 TR_STRDOIOCTL_WAIT,
5976 6002 "strdoioctl sleeps - IOCWAIT");
5977 6003 cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
5978 6004 STRTIMOUT, sigflag);
5979 6005 if (cv_rval <= 0) {
5980 6006 if (cv_rval == 0) {
5981 6007 error = EINTR;
5982 6008 } else {
5983 6009 if (flag & STR_NOERROR) {
5984 6010 /*
5985 6011 * Terminating current ioctl in
5986 6012 * progress -- assume it got lost and
5987 6013 * wake up the other thread so that the
5988 6014 * operation completes.
5989 6015 */
5990 6016 if (!(stp->sd_flag & IOCWAITNE)) {
5991 6017 set_iocwaitne = B_TRUE;
5992 6018 stp->sd_flag |= IOCWAITNE;
5993 6019 cv_broadcast(&stp->sd_monitor);
5994 6020 }
5995 6021 /*
5996 6022 * Otherwise, there's a running
5997 6023 * STR_NOERROR -- we have no choice
5998 6024 * here but to wait forever (or until
5999 6025 * interrupted).
6000 6026 */
6001 6027 } else {
6002 6028 /*
6003 6029 * pending ioctl has caused
6004 6030 * us to time out
6005 6031 */
6006 6032 error = ETIME;
6007 6033 }
6008 6034 }
6009 6035 } else if ((stp->sd_flag & errs)) {
6010 6036 error = strgeterr(stp, errs, 0);
6011 6037 }
6012 6038 if (error) {
6013 6039 mutex_exit(&stp->sd_lock);
6014 6040 freemsg(bp);
6015 6041 crfree(crp);
6016 6042 return (error);
6017 6043 }
6018 6044 }
6019 6045
6020 6046 /*
6021 6047 * Have control of ioctl mechanism.
6022 6048 * Send down ioctl packet and wait for response.
6023 6049 */
6024 6050 if (stp->sd_iocblk != (mblk_t *)-1) {
6025 6051 freemsg(stp->sd_iocblk);
6026 6052 }
6027 6053 stp->sd_iocblk = NULL;
6028 6054
6029 6055 /*
6030 6056 * If this is marked with 'noerror' (internal; mostly
6031 6057 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6032 6058 * in here by setting IOCWAITNE.
6033 6059 */
6034 6060 waitflags = IOCWAIT;
6035 6061 if (flag & STR_NOERROR)
6036 6062 waitflags |= IOCWAITNE;
6037 6063
6038 6064 stp->sd_flag |= waitflags;
6039 6065
6040 6066 /*
6041 6067 * Assign sequence number.
6042 6068 */
6043 6069 iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6044 6070
6045 6071 mutex_exit(&stp->sd_lock);
6046 6072
6047 6073 TRACE_1(TR_FAC_STREAMS_FR,
6048 6074 TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6049 6075 stream_willservice(stp);
6050 6076 putnext(stp->sd_wrq, bp);
6051 6077 stream_runservice(stp);
6052 6078
6053 6079 /*
6054 6080 * Timed wait for acknowledgment. The wait time is limited by the
6055 6081 * timeout value, which must be a positive integer (number of
6056 6082 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6057 6083 * milliseconds), or -1 (wait forever). This will be awakened
6058 6084 * either by an ACK/NAK message arriving, the timer expiring, or
6059 6085 * the timer expiring on another ioctl waiting for control of the
6060 6086 * mechanism.
6061 6087 */
6062 6088 waitioc:
6063 6089 mutex_enter(&stp->sd_lock);
6064 6090
6065 6091
6066 6092 /*
6067 6093 * If the reply has already arrived, don't sleep. If awakened from
6068 6094 * the sleep, fail only if the reply has not arrived by then.
6069 6095 * Otherwise, process the reply.
6070 6096 */
6071 6097 while (!stp->sd_iocblk) {
6072 6098 clock_t cv_rval;
6073 6099
6074 6100 if (stp->sd_flag & errs) {
6075 6101 error = strgeterr(stp, errs, 0);
6076 6102 if (error != 0) {
6077 6103 stp->sd_flag &= ~waitflags;
6078 6104 cv_broadcast(&stp->sd_iocmonitor);
6079 6105 mutex_exit(&stp->sd_lock);
6080 6106 crfree(crp);
6081 6107 return (error);
6082 6108 }
6083 6109 }
6084 6110
6085 6111 TRACE_0(TR_FAC_STREAMS_FR,
6086 6112 TR_STRDOIOCTL_WAIT2,
6087 6113 "strdoioctl sleeps awaiting reply");
6088 6114 ASSERT(error == 0);
6089 6115
6090 6116 cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6091 6117 (strioc->ic_timout ?
6092 6118 strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6093 6119
6094 6120 /*
6095 6121 * There are four possible cases here: interrupt, timeout,
6096 6122 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6097 6123 * valid M_IOCTL reply).
6098 6124 *
6099 6125 * If we've been awakened by a STR_NOERROR ioctl on some other
6100 6126 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6101 6127 * will be set. Pretend as if we just timed out. Note that
6102 6128 * this other thread waited at least STRTIMOUT before trying to
6103 6129 * awaken our thread, so this is indistinguishable (even for
6104 6130 * INFTIM) from the case where we failed with ETIME waiting on
6105 6131 * IOCWAIT in the prior loop.
6106 6132 */
6107 6133 if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6108 6134 stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6109 6135 cv_rval = -1;
6110 6136 }
6111 6137
6112 6138 /*
6113 6139 * note: STR_NOERROR does not protect
6114 6140 * us here.. use ic_timout < 0
6115 6141 */
6116 6142 if (cv_rval <= 0) {
6117 6143 if (cv_rval == 0) {
6118 6144 error = EINTR;
6119 6145 } else {
6120 6146 error = ETIME;
6121 6147 }
6122 6148 /*
6123 6149 * A message could have come in after we were scheduled
6124 6150 * but before we were actually run.
6125 6151 */
6126 6152 bp = stp->sd_iocblk;
6127 6153 stp->sd_iocblk = NULL;
6128 6154 if (bp != NULL) {
6129 6155 if ((bp->b_datap->db_type == M_COPYIN) ||
6130 6156 (bp->b_datap->db_type == M_COPYOUT)) {
6131 6157 mutex_exit(&stp->sd_lock);
6132 6158 if (bp->b_cont) {
6133 6159 freemsg(bp->b_cont);
6134 6160 bp->b_cont = NULL;
6135 6161 }
6136 6162 bp->b_datap->db_type = M_IOCDATA;
6137 6163 bp->b_wptr = bp->b_rptr +
6138 6164 sizeof (struct copyresp);
6139 6165 resp = (struct copyresp *)bp->b_rptr;
6140 6166 resp->cp_rval =
6141 6167 (caddr_t)1; /* failure */
6142 6168 stream_willservice(stp);
6143 6169 putnext(stp->sd_wrq, bp);
6144 6170 stream_runservice(stp);
6145 6171 mutex_enter(&stp->sd_lock);
6146 6172 } else {
6147 6173 freemsg(bp);
6148 6174 }
6149 6175 }
6150 6176 stp->sd_flag &= ~waitflags;
6151 6177 cv_broadcast(&stp->sd_iocmonitor);
6152 6178 mutex_exit(&stp->sd_lock);
6153 6179 crfree(crp);
6154 6180 return (error);
6155 6181 }
6156 6182 }
6157 6183 bp = stp->sd_iocblk;
6158 6184 /*
6159 6185 * Note: it is strictly impossible to get here with sd_iocblk set to
6160 6186 * -1. This is because the initial loop above doesn't allow any new
6161 6187 * ioctls into the fray until all others have passed this point.
6162 6188 */
6163 6189 ASSERT(bp != NULL && bp != (mblk_t *)-1);
6164 6190 TRACE_1(TR_FAC_STREAMS_FR,
6165 6191 TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6166 6192 if ((bp->b_datap->db_type == M_IOCACK) ||
6167 6193 (bp->b_datap->db_type == M_IOCNAK)) {
6168 6194 /* for detection of duplicate ioctl replies */
6169 6195 stp->sd_iocblk = (mblk_t *)-1;
6170 6196 stp->sd_flag &= ~waitflags;
6171 6197 cv_broadcast(&stp->sd_iocmonitor);
6172 6198 mutex_exit(&stp->sd_lock);
6173 6199 } else {
6174 6200 /*
6175 6201 * flags not cleared here because we're still doing
6176 6202 * copy in/out for ioctl.
6177 6203 */
6178 6204 stp->sd_iocblk = NULL;
6179 6205 mutex_exit(&stp->sd_lock);
6180 6206 }
6181 6207
6182 6208
6183 6209 /*
6184 6210 * Have received acknowledgment.
6185 6211 */
6186 6212
6187 6213 switch (bp->b_datap->db_type) {
6188 6214 case M_IOCACK:
6189 6215 /*
6190 6216 * Positive ack.
6191 6217 */
6192 6218 iocbp = (struct iocblk *)bp->b_rptr;
6193 6219
6194 6220 /*
6195 6221 * Set error if indicated.
6196 6222 */
6197 6223 if (iocbp->ioc_error) {
6198 6224 error = iocbp->ioc_error;
6199 6225 break;
6200 6226 }
6201 6227
6202 6228 /*
6203 6229 * Set return value.
6204 6230 */
6205 6231 *rvalp = iocbp->ioc_rval;
6206 6232
6207 6233 /*
6208 6234 * Data may have been returned in ACK message (ioc_count > 0).
6209 6235 * If so, copy it out to the user's buffer.
6210 6236 */
6211 6237 if (iocbp->ioc_count && !transparent) {
6212 6238 if (error = getiocd(bp, strioc->ic_dp, copyflag))
6213 6239 break;
6214 6240 }
6215 6241 if (!transparent) {
6216 6242 if (len) /* an M_COPYOUT was used with I_STR */
6217 6243 strioc->ic_len = len;
6218 6244 else
6219 6245 strioc->ic_len = (int)iocbp->ioc_count;
6220 6246 }
6221 6247 break;
6222 6248
6223 6249 case M_IOCNAK:
6224 6250 /*
6225 6251 * Negative ack.
6226 6252 *
6227 6253 * The only thing to do is set error as specified
6228 6254 * in neg ack packet.
6229 6255 */
6230 6256 iocbp = (struct iocblk *)bp->b_rptr;
6231 6257
6232 6258 error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6233 6259 break;
6234 6260
6235 6261 case M_COPYIN:
6236 6262 /*
6237 6263 * Driver or module has requested user ioctl data.
6238 6264 */
6239 6265 reqp = (struct copyreq *)bp->b_rptr;
6240 6266
6241 6267 /*
6242 6268 * M_COPYIN should *never* have a message attached, though
6243 6269 * it's harmless if it does -- thus, panic on a DEBUG
6244 6270 * kernel and just free it on a non-DEBUG build.
6245 6271 */
6246 6272 ASSERT(bp->b_cont == NULL);
6247 6273 if (bp->b_cont != NULL) {
6248 6274 freemsg(bp->b_cont);
6249 6275 bp->b_cont = NULL;
6250 6276 }
6251 6277
6252 6278 error = putiocd(bp, reqp->cq_addr, flag, crp);
6253 6279 if (error && bp->b_cont) {
6254 6280 freemsg(bp->b_cont);
6255 6281 bp->b_cont = NULL;
6256 6282 }
6257 6283
6258 6284 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6259 6285 bp->b_datap->db_type = M_IOCDATA;
6260 6286
6261 6287 mblk_setcred(bp, crp, curproc->p_pid);
6262 6288 resp = (struct copyresp *)bp->b_rptr;
6263 6289 resp->cp_rval = (caddr_t)(uintptr_t)error;
6264 6290 resp->cp_flag = (fflags & FMODELS);
6265 6291
6266 6292 stream_willservice(stp);
6267 6293 putnext(stp->sd_wrq, bp);
6268 6294 stream_runservice(stp);
6269 6295
6270 6296 if (error) {
6271 6297 mutex_enter(&stp->sd_lock);
6272 6298 stp->sd_flag &= ~waitflags;
6273 6299 cv_broadcast(&stp->sd_iocmonitor);
6274 6300 mutex_exit(&stp->sd_lock);
6275 6301 crfree(crp);
6276 6302 return (error);
6277 6303 }
6278 6304
6279 6305 goto waitioc;
6280 6306
6281 6307 case M_COPYOUT:
6282 6308 /*
6283 6309 * Driver or module has ioctl data for a user.
6284 6310 */
6285 6311 reqp = (struct copyreq *)bp->b_rptr;
6286 6312 ASSERT(bp->b_cont != NULL);
6287 6313
6288 6314 /*
6289 6315 * Always (transparent or non-transparent )
6290 6316 * use the address specified in the request
6291 6317 */
6292 6318 taddr = reqp->cq_addr;
6293 6319 if (!transparent)
6294 6320 len = (int)reqp->cq_size;
6295 6321
6296 6322 /* copyout data to the provided address */
6297 6323 error = getiocd(bp, taddr, copyflag);
6298 6324
6299 6325 freemsg(bp->b_cont);
6300 6326 bp->b_cont = NULL;
6301 6327
6302 6328 bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6303 6329 bp->b_datap->db_type = M_IOCDATA;
6304 6330
6305 6331 mblk_setcred(bp, crp, curproc->p_pid);
6306 6332 resp = (struct copyresp *)bp->b_rptr;
6307 6333 resp->cp_rval = (caddr_t)(uintptr_t)error;
6308 6334 resp->cp_flag = (fflags & FMODELS);
6309 6335
6310 6336 stream_willservice(stp);
6311 6337 putnext(stp->sd_wrq, bp);
6312 6338 stream_runservice(stp);
6313 6339
6314 6340 if (error) {
6315 6341 mutex_enter(&stp->sd_lock);
6316 6342 stp->sd_flag &= ~waitflags;
6317 6343 cv_broadcast(&stp->sd_iocmonitor);
6318 6344 mutex_exit(&stp->sd_lock);
6319 6345 crfree(crp);
6320 6346 return (error);
6321 6347 }
6322 6348 goto waitioc;
6323 6349
6324 6350 default:
6325 6351 ASSERT(0);
6326 6352 mutex_enter(&stp->sd_lock);
6327 6353 stp->sd_flag &= ~waitflags;
6328 6354 cv_broadcast(&stp->sd_iocmonitor);
6329 6355 mutex_exit(&stp->sd_lock);
6330 6356 break;
6331 6357 }
6332 6358
6333 6359 freemsg(bp);
6334 6360 crfree(crp);
6335 6361 return (error);
6336 6362 }
6337 6363
6338 6364 /*
6339 6365 * Send an M_CMD message downstream and wait for a reply. This is a ptools
6340 6366 * special used to retrieve information from modules/drivers a stream without
6341 6367 * being subjected to flow control or interfering with pending messages on the
6342 6368 * stream (e.g. an ioctl in flight).
6343 6369 */
6344 6370 int
6345 6371 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6346 6372 {
6347 6373 mblk_t *mp;
6348 6374 struct cmdblk *cmdp;
6349 6375 int error = 0;
6350 6376 int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6351 6377 clock_t rval, timeout = STRTIMOUT;
6352 6378
6353 6379 if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6354 6380 scp->sc_timeout < -1)
6355 6381 return (EINVAL);
6356 6382
6357 6383 if (scp->sc_timeout > 0)
6358 6384 timeout = scp->sc_timeout * MILLISEC;
6359 6385
6360 6386 if ((mp = allocb_cred(sizeof (struct cmdblk), crp,
6361 6387 curproc->p_pid)) == NULL)
6362 6388 return (ENOMEM);
6363 6389
6364 6390 crhold(crp);
6365 6391
6366 6392 cmdp = (struct cmdblk *)mp->b_wptr;
6367 6393 cmdp->cb_cr = crp;
6368 6394 cmdp->cb_cmd = scp->sc_cmd;
6369 6395 cmdp->cb_len = scp->sc_len;
6370 6396 cmdp->cb_error = 0;
6371 6397 mp->b_wptr += sizeof (struct cmdblk);
6372 6398
6373 6399 DB_TYPE(mp) = M_CMD;
6374 6400 DB_CPID(mp) = curproc->p_pid;
6375 6401
6376 6402 /*
6377 6403 * Copy in the payload.
6378 6404 */
6379 6405 if (cmdp->cb_len > 0) {
6380 6406 mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp,
6381 6407 curproc->p_pid);
6382 6408 if (mp->b_cont == NULL) {
6383 6409 error = ENOMEM;
6384 6410 goto out;
6385 6411 }
6386 6412
6387 6413 /* cb_len comes from sc_len, which has already been checked */
6388 6414 ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6389 6415 (void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6390 6416 mp->b_cont->b_wptr += cmdp->cb_len;
6391 6417 DB_CPID(mp->b_cont) = curproc->p_pid;
6392 6418 }
6393 6419
6394 6420 /*
6395 6421 * Since this mechanism is strictly for ptools, and since only one
6396 6422 * process can be grabbed at a time, we simply fail if there's
6397 6423 * currently an operation pending.
6398 6424 */
6399 6425 mutex_enter(&stp->sd_lock);
6400 6426 if (stp->sd_flag & STRCMDWAIT) {
6401 6427 mutex_exit(&stp->sd_lock);
6402 6428 error = EBUSY;
6403 6429 goto out;
6404 6430 }
6405 6431 stp->sd_flag |= STRCMDWAIT;
6406 6432 ASSERT(stp->sd_cmdblk == NULL);
6407 6433 mutex_exit(&stp->sd_lock);
6408 6434
6409 6435 putnext(stp->sd_wrq, mp);
6410 6436 mp = NULL;
6411 6437
6412 6438 /*
6413 6439 * Timed wait for acknowledgment. If the reply has already arrived,
6414 6440 * don't sleep. If awakened from the sleep, fail only if the reply
6415 6441 * has not arrived by then. Otherwise, process the reply.
6416 6442 */
6417 6443 mutex_enter(&stp->sd_lock);
6418 6444 while (stp->sd_cmdblk == NULL) {
6419 6445 if (stp->sd_flag & errs) {
6420 6446 if ((error = strgeterr(stp, errs, 0)) != 0)
6421 6447 goto waitout;
6422 6448 }
6423 6449
6424 6450 rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6425 6451 if (stp->sd_cmdblk != NULL)
6426 6452 break;
6427 6453
6428 6454 if (rval <= 0) {
6429 6455 error = (rval == 0) ? EINTR : ETIME;
6430 6456 goto waitout;
6431 6457 }
6432 6458 }
6433 6459
6434 6460 /*
6435 6461 * We received a reply.
6436 6462 */
6437 6463 mp = stp->sd_cmdblk;
6438 6464 stp->sd_cmdblk = NULL;
6439 6465 ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6440 6466 ASSERT(stp->sd_flag & STRCMDWAIT);
6441 6467 stp->sd_flag &= ~STRCMDWAIT;
6442 6468 mutex_exit(&stp->sd_lock);
6443 6469
6444 6470 cmdp = (struct cmdblk *)mp->b_rptr;
6445 6471 if ((error = cmdp->cb_error) != 0)
6446 6472 goto out;
6447 6473
6448 6474 /*
6449 6475 * Data may have been returned in the reply (cb_len > 0).
6450 6476 * If so, copy it out to the user's buffer.
6451 6477 */
6452 6478 if (cmdp->cb_len > 0) {
6453 6479 if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6454 6480 error = EPROTO;
6455 6481 goto out;
6456 6482 }
6457 6483
6458 6484 cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6459 6485 (void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6460 6486 }
6461 6487 scp->sc_len = cmdp->cb_len;
6462 6488 out:
6463 6489 freemsg(mp);
6464 6490 crfree(crp);
6465 6491 return (error);
6466 6492 waitout:
6467 6493 ASSERT(stp->sd_cmdblk == NULL);
6468 6494 stp->sd_flag &= ~STRCMDWAIT;
6469 6495 mutex_exit(&stp->sd_lock);
6470 6496 crfree(crp);
6471 6497 return (error);
6472 6498 }
6473 6499
6474 6500 /*
6475 6501 * For the SunOS keyboard driver.
6476 6502 * Return the next available "ioctl" sequence number.
6477 6503 * Exported, so that streams modules can send "ioctl" messages
6478 6504 * downstream from their open routine.
6479 6505 */
6480 6506 int
6481 6507 getiocseqno(void)
6482 6508 {
6483 6509 int i;
6484 6510
6485 6511 mutex_enter(&strresources);
6486 6512 i = ++ioc_id;
6487 6513 mutex_exit(&strresources);
6488 6514 return (i);
6489 6515 }
6490 6516
6491 6517 /*
6492 6518 * Get the next message from the read queue. If the message is
6493 6519 * priority, STRPRI will have been set by strrput(). This flag
6494 6520 * should be reset only when the entire message at the front of the
6495 6521 * queue as been consumed.
6496 6522 *
6497 6523 * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6498 6524 */
6499 6525 int
6500 6526 strgetmsg(
6501 6527 struct vnode *vp,
6502 6528 struct strbuf *mctl,
6503 6529 struct strbuf *mdata,
6504 6530 unsigned char *prip,
6505 6531 int *flagsp,
6506 6532 int fmode,
6507 6533 rval_t *rvp)
6508 6534 {
6509 6535 struct stdata *stp;
6510 6536 mblk_t *bp, *nbp;
6511 6537 mblk_t *savemp = NULL;
6512 6538 mblk_t *savemptail = NULL;
6513 6539 uint_t old_sd_flag;
6514 6540 int flg;
6515 6541 int more = 0;
6516 6542 int error = 0;
6517 6543 char first = 1;
6518 6544 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
6519 6545 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
6520 6546 unsigned char pri = 0;
6521 6547 queue_t *q;
6522 6548 int pr = 0; /* Partial read successful */
6523 6549 struct uio uios;
6524 6550 struct uio *uiop = &uios;
6525 6551 struct iovec iovs;
6526 6552 unsigned char type;
6527 6553
6528 6554 TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6529 6555 "strgetmsg:%p", vp);
6530 6556
6531 6557 ASSERT(vp->v_stream);
6532 6558 stp = vp->v_stream;
6533 6559 rvp->r_val1 = 0;
6534 6560
6535 6561 mutex_enter(&stp->sd_lock);
6536 6562
6537 6563 if ((error = i_straccess(stp, JCREAD)) != 0) {
6538 6564 mutex_exit(&stp->sd_lock);
6539 6565 return (error);
6540 6566 }
6541 6567
6542 6568 if (stp->sd_flag & (STRDERR|STPLEX)) {
6543 6569 error = strgeterr(stp, STRDERR|STPLEX, 0);
6544 6570 if (error != 0) {
6545 6571 mutex_exit(&stp->sd_lock);
6546 6572 return (error);
6547 6573 }
6548 6574 }
6549 6575 mutex_exit(&stp->sd_lock);
6550 6576
6551 6577 switch (*flagsp) {
6552 6578 case MSG_HIPRI:
6553 6579 if (*prip != 0)
6554 6580 return (EINVAL);
6555 6581 break;
6556 6582
6557 6583 case MSG_ANY:
6558 6584 case MSG_BAND:
6559 6585 break;
6560 6586
6561 6587 default:
6562 6588 return (EINVAL);
6563 6589 }
6564 6590 /*
6565 6591 * Setup uio and iov for data part
6566 6592 */
6567 6593 iovs.iov_base = mdata->buf;
6568 6594 iovs.iov_len = mdata->maxlen;
6569 6595 uios.uio_iov = &iovs;
6570 6596 uios.uio_iovcnt = 1;
6571 6597 uios.uio_loffset = 0;
6572 6598 uios.uio_segflg = UIO_USERSPACE;
6573 6599 uios.uio_fmode = 0;
6574 6600 uios.uio_extflg = UIO_COPY_CACHED;
6575 6601 uios.uio_resid = mdata->maxlen;
6576 6602 uios.uio_offset = 0;
6577 6603
6578 6604 q = _RD(stp->sd_wrq);
6579 6605 mutex_enter(&stp->sd_lock);
6580 6606 old_sd_flag = stp->sd_flag;
6581 6607 mark = 0;
6582 6608 for (;;) {
6583 6609 int done = 0;
6584 6610 mblk_t *q_first = q->q_first;
6585 6611
6586 6612 /*
6587 6613 * Get the next message of appropriate priority
6588 6614 * from the stream head. If the caller is interested
6589 6615 * in band or hipri messages, then they should already
6590 6616 * be enqueued at the stream head. On the other hand
6591 6617 * if the caller wants normal (band 0) messages, they
6592 6618 * might be deferred in a synchronous stream and they
6593 6619 * will need to be pulled up.
6594 6620 *
6595 6621 * After we have dequeued a message, we might find that
6596 6622 * it was a deferred M_SIG that was enqueued at the
6597 6623 * stream head. It must now be posted as part of the
6598 6624 * read by calling strsignal_nolock().
6599 6625 *
6600 6626 * Also note that strrput does not enqueue an M_PCSIG,
6601 6627 * and there cannot be more than one hipri message,
6602 6628 * so there was no need to have the M_PCSIG case.
6603 6629 *
6604 6630 * At some time it might be nice to try and wrap the
6605 6631 * functionality of kstrgetmsg() and strgetmsg() into
6606 6632 * a common routine so to reduce the amount of replicated
6607 6633 * code (since they are extremely similar).
6608 6634 */
6609 6635 if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6610 6636 /* Asking for normal, band0 data */
6611 6637 bp = strget(stp, q, uiop, first, &error);
6612 6638 ASSERT(MUTEX_HELD(&stp->sd_lock));
6613 6639 if (bp != NULL) {
6614 6640 if (DB_TYPE(bp) == M_SIG) {
6615 6641 strsignal_nolock(stp, *bp->b_rptr,
6616 6642 bp->b_band);
6617 6643 freemsg(bp);
6618 6644 continue;
6619 6645 } else {
6620 6646 break;
6621 6647 }
6622 6648 }
6623 6649 if (error != 0)
6624 6650 goto getmout;
6625 6651
6626 6652 /*
6627 6653 * We can't depend on the value of STRPRI here because
6628 6654 * the stream head may be in transit. Therefore, we
6629 6655 * must look at the type of the first message to
6630 6656 * determine if a high priority messages is waiting
6631 6657 */
6632 6658 } else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6633 6659 DB_TYPE(q_first) >= QPCTL &&
6634 6660 (bp = getq_noenab(q, 0)) != NULL) {
6635 6661 /* Asked for HIPRI and got one */
6636 6662 ASSERT(DB_TYPE(bp) >= QPCTL);
6637 6663 break;
6638 6664 } else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6639 6665 ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
6640 6666 (bp = getq_noenab(q, 0)) != NULL) {
6641 6667 /*
6642 6668 * Asked for at least band "prip" and got either at
6643 6669 * least that band or a hipri message.
6644 6670 */
6645 6671 ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
6646 6672 if (DB_TYPE(bp) == M_SIG) {
6647 6673 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6648 6674 freemsg(bp);
6649 6675 continue;
6650 6676 } else {
6651 6677 break;
6652 6678 }
6653 6679 }
6654 6680
6655 6681 /* No data. Time to sleep? */
6656 6682 qbackenable(q, 0);
6657 6683
6658 6684 /*
6659 6685 * If STRHUP or STREOF, return 0 length control and data.
6660 6686 * If resid is 0, then a read(fd,buf,0) was done. Do not
6661 6687 * sleep to satisfy this request because by default we have
6662 6688 * zero bytes to return.
6663 6689 */
6664 6690 if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6665 6691 mdata->maxlen == 0)) {
6666 6692 mctl->len = mdata->len = 0;
6667 6693 *flagsp = 0;
6668 6694 mutex_exit(&stp->sd_lock);
6669 6695 return (0);
6670 6696 }
6671 6697 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6672 6698 "strgetmsg calls strwaitq:%p, %p",
6673 6699 vp, uiop);
6674 6700 if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6675 6701 &done)) != 0) || done) {
6676 6702 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6677 6703 "strgetmsg error or done:%p, %p",
6678 6704 vp, uiop);
6679 6705 mutex_exit(&stp->sd_lock);
6680 6706 return (error);
6681 6707 }
6682 6708 TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6683 6709 "strgetmsg awakes:%p, %p", vp, uiop);
6684 6710 if ((error = i_straccess(stp, JCREAD)) != 0) {
6685 6711 mutex_exit(&stp->sd_lock);
6686 6712 return (error);
6687 6713 }
6688 6714 first = 0;
6689 6715 }
6690 6716 ASSERT(bp != NULL);
6691 6717 /*
6692 6718 * Extract any mark information. If the message is not completely
6693 6719 * consumed this information will be put in the mblk
6694 6720 * that is putback.
6695 6721 * If MSGMARKNEXT is set and the message is completely consumed
6696 6722 * the STRATMARK flag will be set below. Likewise, if
6697 6723 * MSGNOTMARKNEXT is set and the message is
6698 6724 * completely consumed STRNOTATMARK will be set.
6699 6725 */
6700 6726 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6701 6727 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6702 6728 (MSGMARKNEXT|MSGNOTMARKNEXT));
6703 6729 if (mark != 0 && bp == stp->sd_mark) {
6704 6730 mark |= _LASTMARK;
6705 6731 stp->sd_mark = NULL;
6706 6732 }
6707 6733 /*
6708 6734 * keep track of the original message type and priority
6709 6735 */
6710 6736 pri = bp->b_band;
6711 6737 type = bp->b_datap->db_type;
6712 6738 if (type == M_PASSFP) {
6713 6739 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6714 6740 stp->sd_mark = bp;
6715 6741 bp->b_flag |= mark & ~_LASTMARK;
6716 6742 putback(stp, q, bp, pri);
6717 6743 qbackenable(q, pri);
6718 6744 mutex_exit(&stp->sd_lock);
6719 6745 return (EBADMSG);
6720 6746 }
6721 6747 ASSERT(type != M_SIG);
6722 6748
6723 6749 /*
6724 6750 * Set this flag so strrput will not generate signals. Need to
6725 6751 * make sure this flag is cleared before leaving this routine
6726 6752 * else signals will stop being sent.
6727 6753 */
6728 6754 stp->sd_flag |= STRGETINPROG;
6729 6755 mutex_exit(&stp->sd_lock);
6730 6756
6731 6757 if (STREAM_NEEDSERVICE(stp))
6732 6758 stream_runservice(stp);
6733 6759
6734 6760 /*
6735 6761 * Set HIPRI flag if message is priority.
6736 6762 */
6737 6763 if (type >= QPCTL)
6738 6764 flg = MSG_HIPRI;
6739 6765 else
6740 6766 flg = MSG_BAND;
6741 6767
6742 6768 /*
6743 6769 * First process PROTO or PCPROTO blocks, if any.
6744 6770 */
6745 6771 if (mctl->maxlen >= 0 && type != M_DATA) {
6746 6772 size_t n, bcnt;
6747 6773 char *ubuf;
6748 6774
6749 6775 bcnt = mctl->maxlen;
6750 6776 ubuf = mctl->buf;
6751 6777 while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6752 6778 if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6753 6779 copyout(bp->b_rptr, ubuf, n)) {
6754 6780 error = EFAULT;
6755 6781 mutex_enter(&stp->sd_lock);
6756 6782 /*
6757 6783 * clear stream head pri flag based on
6758 6784 * first message type
6759 6785 */
6760 6786 if (type >= QPCTL) {
6761 6787 ASSERT(type == M_PCPROTO);
6762 6788 stp->sd_flag &= ~STRPRI;
6763 6789 }
6764 6790 more = 0;
6765 6791 freemsg(bp);
6766 6792 goto getmout;
6767 6793 }
6768 6794 ubuf += n;
6769 6795 bp->b_rptr += n;
6770 6796 if (bp->b_rptr >= bp->b_wptr) {
6771 6797 nbp = bp;
6772 6798 bp = bp->b_cont;
6773 6799 freeb(nbp);
6774 6800 }
6775 6801 ASSERT(n <= bcnt);
6776 6802 bcnt -= n;
6777 6803 if (bcnt == 0)
6778 6804 break;
6779 6805 }
6780 6806 mctl->len = mctl->maxlen - bcnt;
6781 6807 } else
6782 6808 mctl->len = -1;
6783 6809
6784 6810 if (bp && bp->b_datap->db_type != M_DATA) {
6785 6811 /*
6786 6812 * More PROTO blocks in msg.
6787 6813 */
6788 6814 more |= MORECTL;
6789 6815 savemp = bp;
6790 6816 while (bp && bp->b_datap->db_type != M_DATA) {
6791 6817 savemptail = bp;
6792 6818 bp = bp->b_cont;
6793 6819 }
6794 6820 savemptail->b_cont = NULL;
6795 6821 }
6796 6822
6797 6823 /*
6798 6824 * Now process DATA blocks, if any.
6799 6825 */
6800 6826 if (mdata->maxlen >= 0 && bp) {
6801 6827 /*
6802 6828 * struiocopyout will consume a potential zero-length
6803 6829 * M_DATA even if uio_resid is zero.
6804 6830 */
6805 6831 size_t oldresid = uiop->uio_resid;
6806 6832
6807 6833 bp = struiocopyout(bp, uiop, &error);
6808 6834 if (error != 0) {
6809 6835 mutex_enter(&stp->sd_lock);
6810 6836 /*
6811 6837 * clear stream head hi pri flag based on
6812 6838 * first message
6813 6839 */
6814 6840 if (type >= QPCTL) {
6815 6841 ASSERT(type == M_PCPROTO);
6816 6842 stp->sd_flag &= ~STRPRI;
6817 6843 }
6818 6844 more = 0;
6819 6845 freemsg(savemp);
6820 6846 goto getmout;
6821 6847 }
6822 6848 /*
6823 6849 * (pr == 1) indicates a partial read.
6824 6850 */
6825 6851 if (oldresid > uiop->uio_resid)
6826 6852 pr = 1;
6827 6853 mdata->len = mdata->maxlen - uiop->uio_resid;
6828 6854 } else
6829 6855 mdata->len = -1;
6830 6856
6831 6857 if (bp) { /* more data blocks in msg */
6832 6858 more |= MOREDATA;
6833 6859 if (savemp)
6834 6860 savemptail->b_cont = bp;
6835 6861 else
6836 6862 savemp = bp;
6837 6863 }
6838 6864
6839 6865 mutex_enter(&stp->sd_lock);
6840 6866 if (savemp) {
6841 6867 if (pr && (savemp->b_datap->db_type == M_DATA) &&
6842 6868 msgnodata(savemp)) {
6843 6869 /*
6844 6870 * Avoid queuing a zero-length tail part of
6845 6871 * a message. pr=1 indicates that we read some of
6846 6872 * the message.
6847 6873 */
6848 6874 freemsg(savemp);
6849 6875 more &= ~MOREDATA;
6850 6876 /*
6851 6877 * clear stream head hi pri flag based on
6852 6878 * first message
6853 6879 */
6854 6880 if (type >= QPCTL) {
6855 6881 ASSERT(type == M_PCPROTO);
6856 6882 stp->sd_flag &= ~STRPRI;
6857 6883 }
6858 6884 } else {
6859 6885 savemp->b_band = pri;
6860 6886 /*
6861 6887 * If the first message was HIPRI and the one we're
6862 6888 * putting back isn't, then clear STRPRI, otherwise
6863 6889 * set STRPRI again. Note that we must set STRPRI
6864 6890 * again since the flush logic in strrput_nondata()
6865 6891 * may have cleared it while we had sd_lock dropped.
6866 6892 */
6867 6893 if (type >= QPCTL) {
6868 6894 ASSERT(type == M_PCPROTO);
6869 6895 if (queclass(savemp) < QPCTL)
6870 6896 stp->sd_flag &= ~STRPRI;
6871 6897 else
6872 6898 stp->sd_flag |= STRPRI;
6873 6899 } else if (queclass(savemp) >= QPCTL) {
6874 6900 /*
6875 6901 * The first message was not a HIPRI message,
6876 6902 * but the one we are about to putback is.
6877 6903 * For simplicitly, we do not allow for HIPRI
6878 6904 * messages to be embedded in the message
6879 6905 * body, so just force it to same type as
6880 6906 * first message.
6881 6907 */
6882 6908 ASSERT(type == M_DATA || type == M_PROTO);
6883 6909 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6884 6910 savemp->b_datap->db_type = type;
6885 6911 }
6886 6912 if (mark != 0) {
6887 6913 savemp->b_flag |= mark & ~_LASTMARK;
6888 6914 if ((mark & _LASTMARK) &&
6889 6915 (stp->sd_mark == NULL)) {
6890 6916 /*
6891 6917 * If another marked message arrived
6892 6918 * while sd_lock was not held sd_mark
6893 6919 * would be non-NULL.
6894 6920 */
6895 6921 stp->sd_mark = savemp;
6896 6922 }
6897 6923 }
6898 6924 putback(stp, q, savemp, pri);
6899 6925 }
6900 6926 } else {
6901 6927 /*
6902 6928 * The complete message was consumed.
6903 6929 *
6904 6930 * If another M_PCPROTO arrived while sd_lock was not held
6905 6931 * it would have been discarded since STRPRI was still set.
6906 6932 *
6907 6933 * Move the MSG*MARKNEXT information
6908 6934 * to the stream head just in case
6909 6935 * the read queue becomes empty.
6910 6936 * clear stream head hi pri flag based on
6911 6937 * first message
6912 6938 *
6913 6939 * If the stream head was at the mark
6914 6940 * (STRATMARK) before we dropped sd_lock above
6915 6941 * and some data was consumed then we have
6916 6942 * moved past the mark thus STRATMARK is
6917 6943 * cleared. However, if a message arrived in
6918 6944 * strrput during the copyout above causing
6919 6945 * STRATMARK to be set we can not clear that
6920 6946 * flag.
6921 6947 */
6922 6948 if (type >= QPCTL) {
6923 6949 ASSERT(type == M_PCPROTO);
6924 6950 stp->sd_flag &= ~STRPRI;
6925 6951 }
6926 6952 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6927 6953 if (mark & MSGMARKNEXT) {
6928 6954 stp->sd_flag &= ~STRNOTATMARK;
6929 6955 stp->sd_flag |= STRATMARK;
6930 6956 } else if (mark & MSGNOTMARKNEXT) {
6931 6957 stp->sd_flag &= ~STRATMARK;
6932 6958 stp->sd_flag |= STRNOTATMARK;
6933 6959 } else {
6934 6960 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6935 6961 }
6936 6962 } else if (pr && (old_sd_flag & STRATMARK)) {
6937 6963 stp->sd_flag &= ~STRATMARK;
6938 6964 }
6939 6965 }
6940 6966
6941 6967 *flagsp = flg;
6942 6968 *prip = pri;
6943 6969
6944 6970 /*
6945 6971 * Getmsg cleanup processing - if the state of the queue has changed
6946 6972 * some signals may need to be sent and/or poll awakened.
6947 6973 */
6948 6974 getmout:
6949 6975 qbackenable(q, pri);
6950 6976
6951 6977 /*
6952 6978 * We dropped the stream head lock above. Send all M_SIG messages
6953 6979 * before processing stream head for SIGPOLL messages.
6954 6980 */
6955 6981 ASSERT(MUTEX_HELD(&stp->sd_lock));
6956 6982 while ((bp = q->q_first) != NULL &&
6957 6983 (bp->b_datap->db_type == M_SIG)) {
6958 6984 /*
6959 6985 * sd_lock is held so the content of the read queue can not
6960 6986 * change.
6961 6987 */
6962 6988 bp = getq(q);
6963 6989 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
6964 6990
6965 6991 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6966 6992 mutex_exit(&stp->sd_lock);
6967 6993 freemsg(bp);
6968 6994 if (STREAM_NEEDSERVICE(stp))
6969 6995 stream_runservice(stp);
6970 6996 mutex_enter(&stp->sd_lock);
6971 6997 }
6972 6998
6973 6999 /*
6974 7000 * stream head cannot change while we make the determination
6975 7001 * whether or not to send a signal. Drop the flag to allow strrput
6976 7002 * to send firstmsgsigs again.
6977 7003 */
6978 7004 stp->sd_flag &= ~STRGETINPROG;
6979 7005
6980 7006 /*
6981 7007 * If the type of message at the front of the queue changed
6982 7008 * due to the receive the appropriate signals and pollwakeup events
6983 7009 * are generated. The type of changes are:
6984 7010 * Processed a hipri message, q_first is not hipri.
6985 7011 * Processed a band X message, and q_first is band Y.
6986 7012 * The generated signals and pollwakeups are identical to what
6987 7013 * strrput() generates should the message that is now on q_first
6988 7014 * arrive to an empty read queue.
6989 7015 *
6990 7016 * Note: only strrput will send a signal for a hipri message.
6991 7017 */
6992 7018 if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
6993 7019 strsigset_t signals = 0;
6994 7020 strpollset_t pollwakeups = 0;
6995 7021
6996 7022 if (flg & MSG_HIPRI) {
6997 7023 /*
6998 7024 * Removed a hipri message. Regular data at
6999 7025 * the front of the queue.
7000 7026 */
7001 7027 if (bp->b_band == 0) {
7002 7028 signals = S_INPUT | S_RDNORM;
7003 7029 pollwakeups = POLLIN | POLLRDNORM;
7004 7030 } else {
7005 7031 signals = S_INPUT | S_RDBAND;
7006 7032 pollwakeups = POLLIN | POLLRDBAND;
7007 7033 }
7008 7034 } else if (pri != bp->b_band) {
7009 7035 /*
7010 7036 * The band is different for the new q_first.
7011 7037 */
7012 7038 if (bp->b_band == 0) {
7013 7039 signals = S_RDNORM;
7014 7040 pollwakeups = POLLIN | POLLRDNORM;
7015 7041 } else {
7016 7042 signals = S_RDBAND;
7017 7043 pollwakeups = POLLIN | POLLRDBAND;
7018 7044 }
7019 7045 }
7020 7046
7021 7047 if (pollwakeups != 0) {
7022 7048 if (pollwakeups == (POLLIN | POLLRDNORM)) {
7023 7049 if (!(stp->sd_rput_opt & SR_POLLIN))
7024 7050 goto no_pollwake;
7025 7051 stp->sd_rput_opt &= ~SR_POLLIN;
7026 7052 }
7027 7053 mutex_exit(&stp->sd_lock);
7028 7054 pollwakeup(&stp->sd_pollist, pollwakeups);
7029 7055 mutex_enter(&stp->sd_lock);
7030 7056 }
7031 7057 no_pollwake:
7032 7058
7033 7059 if (stp->sd_sigflags & signals)
7034 7060 strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7035 7061 }
7036 7062 mutex_exit(&stp->sd_lock);
7037 7063
7038 7064 rvp->r_val1 = more;
7039 7065 return (error);
7040 7066 #undef _LASTMARK
7041 7067 }
7042 7068
7043 7069 /*
7044 7070 * Get the next message from the read queue. If the message is
7045 7071 * priority, STRPRI will have been set by strrput(). This flag
7046 7072 * should be reset only when the entire message at the front of the
7047 7073 * queue as been consumed.
7048 7074 *
7049 7075 * If uiop is NULL all data is returned in mctlp.
7050 7076 * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7051 7077 * not enabled.
7052 7078 * The timeout parameter is in milliseconds; -1 for infinity.
7053 7079 * This routine handles the consolidation private flags:
7054 7080 * MSG_IGNERROR Ignore any stream head error except STPLEX.
7055 7081 * MSG_DELAYERROR Defer the error check until the queue is empty.
7056 7082 * MSG_HOLDSIG Hold signals while waiting for data.
7057 7083 * MSG_IPEEK Only peek at messages.
7058 7084 * MSG_DISCARDTAIL Discard the tail M_DATA part of the message
7059 7085 * that doesn't fit.
7060 7086 * MSG_NOMARK If the message is marked leave it on the queue.
7061 7087 *
7062 7088 * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7063 7089 */
7064 7090 int
7065 7091 kstrgetmsg(
7066 7092 struct vnode *vp,
7067 7093 mblk_t **mctlp,
7068 7094 struct uio *uiop,
7069 7095 unsigned char *prip,
7070 7096 int *flagsp,
7071 7097 clock_t timout,
7072 7098 rval_t *rvp)
7073 7099 {
7074 7100 struct stdata *stp;
7075 7101 mblk_t *bp, *nbp;
7076 7102 mblk_t *savemp = NULL;
7077 7103 mblk_t *savemptail = NULL;
7078 7104 int flags;
7079 7105 uint_t old_sd_flag;
7080 7106 int flg;
7081 7107 int more = 0;
7082 7108 int error = 0;
7083 7109 char first = 1;
7084 7110 uint_t mark; /* Contains MSG*MARK and _LASTMARK */
7085 7111 #define _LASTMARK 0x8000 /* Distinct from MSG*MARK */
7086 7112 unsigned char pri = 0;
7087 7113 queue_t *q;
7088 7114 int pr = 0; /* Partial read successful */
7089 7115 unsigned char type;
7090 7116
7091 7117 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7092 7118 "kstrgetmsg:%p", vp);
7093 7119
7094 7120 ASSERT(vp->v_stream);
7095 7121 stp = vp->v_stream;
7096 7122 rvp->r_val1 = 0;
7097 7123
7098 7124 mutex_enter(&stp->sd_lock);
7099 7125
7100 7126 if ((error = i_straccess(stp, JCREAD)) != 0) {
7101 7127 mutex_exit(&stp->sd_lock);
7102 7128 return (error);
7103 7129 }
7104 7130
7105 7131 flags = *flagsp;
7106 7132 if (stp->sd_flag & (STRDERR|STPLEX)) {
7107 7133 if ((stp->sd_flag & STPLEX) ||
7108 7134 (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7109 7135 error = strgeterr(stp, STRDERR|STPLEX,
7110 7136 (flags & MSG_IPEEK));
7111 7137 if (error != 0) {
7112 7138 mutex_exit(&stp->sd_lock);
7113 7139 return (error);
7114 7140 }
7115 7141 }
7116 7142 }
7117 7143 mutex_exit(&stp->sd_lock);
7118 7144
7119 7145 switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7120 7146 case MSG_HIPRI:
7121 7147 if (*prip != 0)
7122 7148 return (EINVAL);
7123 7149 break;
7124 7150
7125 7151 case MSG_ANY:
7126 7152 case MSG_BAND:
7127 7153 break;
7128 7154
7129 7155 default:
7130 7156 return (EINVAL);
7131 7157 }
7132 7158
7133 7159 retry:
7134 7160 q = _RD(stp->sd_wrq);
7135 7161 mutex_enter(&stp->sd_lock);
7136 7162 old_sd_flag = stp->sd_flag;
7137 7163 mark = 0;
7138 7164 for (;;) {
7139 7165 int done = 0;
7140 7166 int waitflag;
7141 7167 int fmode;
7142 7168 mblk_t *q_first = q->q_first;
7143 7169
7144 7170 /*
7145 7171 * This section of the code operates just like the code
7146 7172 * in strgetmsg(). There is a comment there about what
7147 7173 * is going on here.
7148 7174 */
7149 7175 if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7150 7176 /* Asking for normal, band0 data */
7151 7177 bp = strget(stp, q, uiop, first, &error);
7152 7178 ASSERT(MUTEX_HELD(&stp->sd_lock));
7153 7179 if (bp != NULL) {
7154 7180 if (DB_TYPE(bp) == M_SIG) {
7155 7181 strsignal_nolock(stp, *bp->b_rptr,
7156 7182 bp->b_band);
7157 7183 freemsg(bp);
7158 7184 continue;
7159 7185 } else {
7160 7186 break;
7161 7187 }
7162 7188 }
7163 7189 if (error != 0) {
7164 7190 goto getmout;
7165 7191 }
7166 7192 /*
7167 7193 * We can't depend on the value of STRPRI here because
7168 7194 * the stream head may be in transit. Therefore, we
7169 7195 * must look at the type of the first message to
7170 7196 * determine if a high priority messages is waiting
7171 7197 */
7172 7198 } else if ((flags & MSG_HIPRI) && q_first != NULL &&
7173 7199 DB_TYPE(q_first) >= QPCTL &&
7174 7200 (bp = getq_noenab(q, 0)) != NULL) {
7175 7201 ASSERT(DB_TYPE(bp) >= QPCTL);
7176 7202 break;
7177 7203 } else if ((flags & MSG_BAND) && q_first != NULL &&
7178 7204 ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
7179 7205 (bp = getq_noenab(q, 0)) != NULL) {
7180 7206 /*
7181 7207 * Asked for at least band "prip" and got either at
7182 7208 * least that band or a hipri message.
7183 7209 */
7184 7210 ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
7185 7211 if (DB_TYPE(bp) == M_SIG) {
7186 7212 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7187 7213 freemsg(bp);
7188 7214 continue;
7189 7215 } else {
7190 7216 break;
7191 7217 }
7192 7218 }
7193 7219
7194 7220 /* No data. Time to sleep? */
7195 7221 qbackenable(q, 0);
7196 7222
7197 7223 /*
7198 7224 * Delayed error notification?
7199 7225 */
7200 7226 if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7201 7227 (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7202 7228 error = strgeterr(stp, STRDERR|STPLEX,
7203 7229 (flags & MSG_IPEEK));
7204 7230 if (error != 0) {
7205 7231 mutex_exit(&stp->sd_lock);
7206 7232 return (error);
7207 7233 }
7208 7234 }
7209 7235
7210 7236 /*
7211 7237 * If STRHUP or STREOF, return 0 length control and data.
7212 7238 * If a read(fd,buf,0) has been done, do not sleep, just
7213 7239 * return.
7214 7240 *
7215 7241 * If mctlp == NULL and uiop == NULL, then the code will
7216 7242 * do the strwaitq. This is an understood way of saying
7217 7243 * sleep "polling" until a message is received.
7218 7244 */
7219 7245 if ((stp->sd_flag & (STRHUP|STREOF)) ||
7220 7246 (uiop != NULL && uiop->uio_resid == 0)) {
7221 7247 if (mctlp != NULL)
7222 7248 *mctlp = NULL;
7223 7249 *flagsp = 0;
7224 7250 mutex_exit(&stp->sd_lock);
7225 7251 return (0);
7226 7252 }
7227 7253
7228 7254 waitflag = GETWAIT;
7229 7255 if (flags &
7230 7256 (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7231 7257 if (flags & MSG_HOLDSIG)
7232 7258 waitflag |= STR_NOSIG;
7233 7259 if (flags & MSG_IGNERROR)
7234 7260 waitflag |= STR_NOERROR;
7235 7261 if (flags & MSG_IPEEK)
7236 7262 waitflag |= STR_PEEK;
7237 7263 if (flags & MSG_DELAYERROR)
7238 7264 waitflag |= STR_DELAYERR;
7239 7265 }
7240 7266 if (uiop != NULL)
7241 7267 fmode = uiop->uio_fmode;
7242 7268 else
7243 7269 fmode = 0;
7244 7270
7245 7271 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7246 7272 "kstrgetmsg calls strwaitq:%p, %p",
7247 7273 vp, uiop);
7248 7274 if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7249 7275 fmode, timout, &done))) != 0 || done) {
7250 7276 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7251 7277 "kstrgetmsg error or done:%p, %p",
7252 7278 vp, uiop);
7253 7279 mutex_exit(&stp->sd_lock);
7254 7280 return (error);
7255 7281 }
7256 7282 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7257 7283 "kstrgetmsg awakes:%p, %p", vp, uiop);
7258 7284 if ((error = i_straccess(stp, JCREAD)) != 0) {
7259 7285 mutex_exit(&stp->sd_lock);
7260 7286 return (error);
7261 7287 }
7262 7288 first = 0;
7263 7289 }
7264 7290 ASSERT(bp != NULL);
7265 7291 /*
7266 7292 * Extract any mark information. If the message is not completely
7267 7293 * consumed this information will be put in the mblk
7268 7294 * that is putback.
7269 7295 * If MSGMARKNEXT is set and the message is completely consumed
7270 7296 * the STRATMARK flag will be set below. Likewise, if
7271 7297 * MSGNOTMARKNEXT is set and the message is
7272 7298 * completely consumed STRNOTATMARK will be set.
7273 7299 */
7274 7300 mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7275 7301 ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7276 7302 (MSGMARKNEXT|MSGNOTMARKNEXT));
7277 7303 pri = bp->b_band;
7278 7304 if (mark != 0) {
7279 7305 /*
7280 7306 * If the caller doesn't want the mark return.
7281 7307 * Used to implement MSG_WAITALL in sockets.
7282 7308 */
7283 7309 if (flags & MSG_NOMARK) {
7284 7310 putback(stp, q, bp, pri);
7285 7311 qbackenable(q, pri);
7286 7312 mutex_exit(&stp->sd_lock);
7287 7313 return (EWOULDBLOCK);
7288 7314 }
7289 7315 if (bp == stp->sd_mark) {
7290 7316 mark |= _LASTMARK;
7291 7317 stp->sd_mark = NULL;
7292 7318 }
7293 7319 }
7294 7320
7295 7321 /*
7296 7322 * keep track of the first message type
7297 7323 */
7298 7324 type = bp->b_datap->db_type;
7299 7325
7300 7326 if (bp->b_datap->db_type == M_PASSFP) {
7301 7327 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7302 7328 stp->sd_mark = bp;
7303 7329 bp->b_flag |= mark & ~_LASTMARK;
7304 7330 putback(stp, q, bp, pri);
7305 7331 qbackenable(q, pri);
7306 7332 mutex_exit(&stp->sd_lock);
7307 7333 return (EBADMSG);
7308 7334 }
7309 7335 ASSERT(type != M_SIG);
7310 7336
7311 7337 if (flags & MSG_IPEEK) {
7312 7338 /*
7313 7339 * Clear any struioflag - we do the uiomove over again
7314 7340 * when peeking since it simplifies the code.
7315 7341 *
7316 7342 * Dup the message and put the original back on the queue.
7317 7343 * If dupmsg() fails, try again with copymsg() to see if
7318 7344 * there is indeed a shortage of memory. dupmsg() may fail
7319 7345 * if db_ref in any of the messages reaches its limit.
7320 7346 */
7321 7347
7322 7348 if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7323 7349 /*
7324 7350 * Restore the state of the stream head since we
7325 7351 * need to drop sd_lock (strwaitbuf is sleeping).
7326 7352 */
7327 7353 size_t size = msgdsize(bp);
7328 7354
7329 7355 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7330 7356 stp->sd_mark = bp;
7331 7357 bp->b_flag |= mark & ~_LASTMARK;
7332 7358 putback(stp, q, bp, pri);
7333 7359 mutex_exit(&stp->sd_lock);
7334 7360 error = strwaitbuf(size, BPRI_HI);
7335 7361 if (error) {
7336 7362 /*
7337 7363 * There is no net change to the queue thus
7338 7364 * no need to qbackenable.
7339 7365 */
7340 7366 return (error);
7341 7367 }
7342 7368 goto retry;
7343 7369 }
7344 7370
7345 7371 if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7346 7372 stp->sd_mark = bp;
7347 7373 bp->b_flag |= mark & ~_LASTMARK;
7348 7374 putback(stp, q, bp, pri);
7349 7375 bp = nbp;
7350 7376 }
7351 7377
7352 7378 /*
7353 7379 * Set this flag so strrput will not generate signals. Need to
7354 7380 * make sure this flag is cleared before leaving this routine
7355 7381 * else signals will stop being sent.
7356 7382 */
7357 7383 stp->sd_flag |= STRGETINPROG;
7358 7384 mutex_exit(&stp->sd_lock);
7359 7385
7360 7386 if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7361 7387 mblk_t *tmp, *prevmp;
7362 7388
7363 7389 /*
7364 7390 * Put first non-data mblk back to stream head and
7365 7391 * cut the mblk chain so sd_rputdatafunc only sees
7366 7392 * M_DATA mblks. We can skip the first mblk since it
7367 7393 * is M_DATA according to the condition above.
7368 7394 */
7369 7395 for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7370 7396 prevmp = tmp, tmp = tmp->b_cont) {
7371 7397 if (DB_TYPE(tmp) != M_DATA) {
7372 7398 prevmp->b_cont = NULL;
7373 7399 mutex_enter(&stp->sd_lock);
7374 7400 putback(stp, q, tmp, tmp->b_band);
7375 7401 mutex_exit(&stp->sd_lock);
7376 7402 break;
7377 7403 }
7378 7404 }
7379 7405
7380 7406 bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7381 7407 NULL, NULL, NULL, NULL);
7382 7408
7383 7409 if (bp == NULL)
7384 7410 goto retry;
7385 7411 }
7386 7412
7387 7413 if (STREAM_NEEDSERVICE(stp))
7388 7414 stream_runservice(stp);
7389 7415
7390 7416 /*
7391 7417 * Set HIPRI flag if message is priority.
7392 7418 */
7393 7419 if (type >= QPCTL)
7394 7420 flg = MSG_HIPRI;
7395 7421 else
7396 7422 flg = MSG_BAND;
7397 7423
7398 7424 /*
7399 7425 * First process PROTO or PCPROTO blocks, if any.
7400 7426 */
7401 7427 if (mctlp != NULL && type != M_DATA) {
7402 7428 mblk_t *nbp;
7403 7429
7404 7430 *mctlp = bp;
7405 7431 while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7406 7432 bp = bp->b_cont;
7407 7433 nbp = bp->b_cont;
7408 7434 bp->b_cont = NULL;
7409 7435 bp = nbp;
7410 7436 }
7411 7437
7412 7438 if (bp && bp->b_datap->db_type != M_DATA) {
7413 7439 /*
7414 7440 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7415 7441 */
7416 7442 more |= MORECTL;
7417 7443 savemp = bp;
7418 7444 while (bp && bp->b_datap->db_type != M_DATA) {
7419 7445 savemptail = bp;
7420 7446 bp = bp->b_cont;
7421 7447 }
7422 7448 savemptail->b_cont = NULL;
7423 7449 }
7424 7450
7425 7451 /*
7426 7452 * Now process DATA blocks, if any.
7427 7453 */
7428 7454 if (uiop == NULL) {
7429 7455 /* Append data to tail of mctlp */
7430 7456
7431 7457 if (mctlp != NULL) {
7432 7458 mblk_t **mpp = mctlp;
7433 7459
7434 7460 while (*mpp != NULL)
7435 7461 mpp = &((*mpp)->b_cont);
7436 7462 *mpp = bp;
7437 7463 bp = NULL;
7438 7464 }
7439 7465 } else if (uiop->uio_resid >= 0 && bp) {
7440 7466 size_t oldresid = uiop->uio_resid;
7441 7467
7442 7468 /*
7443 7469 * If a streams message is likely to consist
7444 7470 * of many small mblks, it is pulled up into
7445 7471 * one continuous chunk of memory.
7446 7472 * The size of the first mblk may be bogus because
7447 7473 * successive read() calls on the socket reduce
7448 7474 * the size of this mblk until it is exhausted
7449 7475 * and then the code walks on to the next. Thus
7450 7476 * the size of the mblk may not be the original size
7451 7477 * that was passed up, it's simply a remainder
7452 7478 * and hence can be very small without any
7453 7479 * implication that the packet is badly fragmented.
7454 7480 * So the size of the possible second mblk is
7455 7481 * used to spot a badly fragmented packet.
7456 7482 * see longer comment at top of page
7457 7483 * by mblk_pull_len declaration.
7458 7484 */
7459 7485
7460 7486 if (bp->b_cont != NULL && MBLKL(bp->b_cont) < mblk_pull_len) {
7461 7487 (void) pullupmsg(bp, -1);
7462 7488 }
7463 7489
7464 7490 bp = struiocopyout(bp, uiop, &error);
7465 7491 if (error != 0) {
7466 7492 if (mctlp != NULL) {
7467 7493 freemsg(*mctlp);
7468 7494 *mctlp = NULL;
7469 7495 } else
7470 7496 freemsg(savemp);
7471 7497 mutex_enter(&stp->sd_lock);
7472 7498 /*
7473 7499 * clear stream head hi pri flag based on
7474 7500 * first message
7475 7501 */
7476 7502 if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7477 7503 ASSERT(type == M_PCPROTO);
7478 7504 stp->sd_flag &= ~STRPRI;
7479 7505 }
7480 7506 more = 0;
7481 7507 goto getmout;
7482 7508 }
7483 7509 /*
7484 7510 * (pr == 1) indicates a partial read.
7485 7511 */
7486 7512 if (oldresid > uiop->uio_resid)
7487 7513 pr = 1;
7488 7514 }
7489 7515
7490 7516 if (bp) { /* more data blocks in msg */
7491 7517 more |= MOREDATA;
7492 7518 if (savemp)
7493 7519 savemptail->b_cont = bp;
7494 7520 else
7495 7521 savemp = bp;
7496 7522 }
7497 7523
7498 7524 mutex_enter(&stp->sd_lock);
7499 7525 if (savemp) {
7500 7526 if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7501 7527 /*
7502 7528 * When MSG_DISCARDTAIL is set or
7503 7529 * when peeking discard any tail. When peeking this
7504 7530 * is the tail of the dup that was copied out - the
7505 7531 * message has already been putback on the queue.
7506 7532 * Return MOREDATA to the caller even though the data
7507 7533 * is discarded. This is used by sockets (to
7508 7534 * set MSG_TRUNC).
7509 7535 */
7510 7536 freemsg(savemp);
7511 7537 if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7512 7538 ASSERT(type == M_PCPROTO);
7513 7539 stp->sd_flag &= ~STRPRI;
7514 7540 }
7515 7541 } else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7516 7542 msgnodata(savemp)) {
7517 7543 /*
7518 7544 * Avoid queuing a zero-length tail part of
7519 7545 * a message. pr=1 indicates that we read some of
7520 7546 * the message.
7521 7547 */
7522 7548 freemsg(savemp);
7523 7549 more &= ~MOREDATA;
7524 7550 if (type >= QPCTL) {
7525 7551 ASSERT(type == M_PCPROTO);
7526 7552 stp->sd_flag &= ~STRPRI;
7527 7553 }
7528 7554 } else {
7529 7555 savemp->b_band = pri;
7530 7556 /*
7531 7557 * If the first message was HIPRI and the one we're
7532 7558 * putting back isn't, then clear STRPRI, otherwise
7533 7559 * set STRPRI again. Note that we must set STRPRI
7534 7560 * again since the flush logic in strrput_nondata()
7535 7561 * may have cleared it while we had sd_lock dropped.
7536 7562 */
7537 7563
7538 7564 if (type >= QPCTL) {
7539 7565 ASSERT(type == M_PCPROTO);
7540 7566 if (queclass(savemp) < QPCTL)
7541 7567 stp->sd_flag &= ~STRPRI;
7542 7568 else
7543 7569 stp->sd_flag |= STRPRI;
7544 7570 } else if (queclass(savemp) >= QPCTL) {
7545 7571 /*
7546 7572 * The first message was not a HIPRI message,
7547 7573 * but the one we are about to putback is.
7548 7574 * For simplicitly, we do not allow for HIPRI
7549 7575 * messages to be embedded in the message
7550 7576 * body, so just force it to same type as
7551 7577 * first message.
7552 7578 */
7553 7579 ASSERT(type == M_DATA || type == M_PROTO);
7554 7580 ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7555 7581 savemp->b_datap->db_type = type;
7556 7582 }
7557 7583 if (mark != 0) {
7558 7584 if ((mark & _LASTMARK) &&
7559 7585 (stp->sd_mark == NULL)) {
7560 7586 /*
7561 7587 * If another marked message arrived
7562 7588 * while sd_lock was not held sd_mark
7563 7589 * would be non-NULL.
7564 7590 */
7565 7591 stp->sd_mark = savemp;
7566 7592 }
7567 7593 savemp->b_flag |= mark & ~_LASTMARK;
7568 7594 }
7569 7595 putback(stp, q, savemp, pri);
7570 7596 }
7571 7597 } else if (!(flags & MSG_IPEEK)) {
7572 7598 /*
7573 7599 * The complete message was consumed.
7574 7600 *
7575 7601 * If another M_PCPROTO arrived while sd_lock was not held
7576 7602 * it would have been discarded since STRPRI was still set.
7577 7603 *
7578 7604 * Move the MSG*MARKNEXT information
7579 7605 * to the stream head just in case
7580 7606 * the read queue becomes empty.
7581 7607 * clear stream head hi pri flag based on
7582 7608 * first message
7583 7609 *
7584 7610 * If the stream head was at the mark
7585 7611 * (STRATMARK) before we dropped sd_lock above
7586 7612 * and some data was consumed then we have
7587 7613 * moved past the mark thus STRATMARK is
7588 7614 * cleared. However, if a message arrived in
7589 7615 * strrput during the copyout above causing
7590 7616 * STRATMARK to be set we can not clear that
7591 7617 * flag.
7592 7618 * XXX A "perimeter" would help by single-threading strrput,
7593 7619 * strread, strgetmsg and kstrgetmsg.
7594 7620 */
7595 7621 if (type >= QPCTL) {
7596 7622 ASSERT(type == M_PCPROTO);
7597 7623 stp->sd_flag &= ~STRPRI;
7598 7624 }
7599 7625 if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7600 7626 if (mark & MSGMARKNEXT) {
7601 7627 stp->sd_flag &= ~STRNOTATMARK;
7602 7628 stp->sd_flag |= STRATMARK;
7603 7629 } else if (mark & MSGNOTMARKNEXT) {
7604 7630 stp->sd_flag &= ~STRATMARK;
7605 7631 stp->sd_flag |= STRNOTATMARK;
7606 7632 } else {
7607 7633 stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7608 7634 }
7609 7635 } else if (pr && (old_sd_flag & STRATMARK)) {
7610 7636 stp->sd_flag &= ~STRATMARK;
7611 7637 }
7612 7638 }
7613 7639
7614 7640 *flagsp = flg;
7615 7641 *prip = pri;
7616 7642
7617 7643 /*
7618 7644 * Getmsg cleanup processing - if the state of the queue has changed
7619 7645 * some signals may need to be sent and/or poll awakened.
7620 7646 */
7621 7647 getmout:
7622 7648 qbackenable(q, pri);
7623 7649
7624 7650 /*
7625 7651 * We dropped the stream head lock above. Send all M_SIG messages
7626 7652 * before processing stream head for SIGPOLL messages.
7627 7653 */
7628 7654 ASSERT(MUTEX_HELD(&stp->sd_lock));
7629 7655 while ((bp = q->q_first) != NULL &&
7630 7656 (bp->b_datap->db_type == M_SIG)) {
7631 7657 /*
7632 7658 * sd_lock is held so the content of the read queue can not
7633 7659 * change.
7634 7660 */
7635 7661 bp = getq(q);
7636 7662 ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7637 7663
7638 7664 strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7639 7665 mutex_exit(&stp->sd_lock);
7640 7666 freemsg(bp);
7641 7667 if (STREAM_NEEDSERVICE(stp))
7642 7668 stream_runservice(stp);
7643 7669 mutex_enter(&stp->sd_lock);
7644 7670 }
7645 7671
7646 7672 /*
7647 7673 * stream head cannot change while we make the determination
7648 7674 * whether or not to send a signal. Drop the flag to allow strrput
7649 7675 * to send firstmsgsigs again.
7650 7676 */
7651 7677 stp->sd_flag &= ~STRGETINPROG;
7652 7678
7653 7679 /*
7654 7680 * If the type of message at the front of the queue changed
7655 7681 * due to the receive the appropriate signals and pollwakeup events
7656 7682 * are generated. The type of changes are:
7657 7683 * Processed a hipri message, q_first is not hipri.
7658 7684 * Processed a band X message, and q_first is band Y.
7659 7685 * The generated signals and pollwakeups are identical to what
7660 7686 * strrput() generates should the message that is now on q_first
7661 7687 * arrive to an empty read queue.
7662 7688 *
7663 7689 * Note: only strrput will send a signal for a hipri message.
7664 7690 */
7665 7691 if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7666 7692 strsigset_t signals = 0;
7667 7693 strpollset_t pollwakeups = 0;
7668 7694
7669 7695 if (flg & MSG_HIPRI) {
7670 7696 /*
7671 7697 * Removed a hipri message. Regular data at
7672 7698 * the front of the queue.
7673 7699 */
7674 7700 if (bp->b_band == 0) {
7675 7701 signals = S_INPUT | S_RDNORM;
7676 7702 pollwakeups = POLLIN | POLLRDNORM;
7677 7703 } else {
7678 7704 signals = S_INPUT | S_RDBAND;
7679 7705 pollwakeups = POLLIN | POLLRDBAND;
7680 7706 }
7681 7707 } else if (pri != bp->b_band) {
7682 7708 /*
7683 7709 * The band is different for the new q_first.
7684 7710 */
7685 7711 if (bp->b_band == 0) {
7686 7712 signals = S_RDNORM;
7687 7713 pollwakeups = POLLIN | POLLRDNORM;
7688 7714 } else {
7689 7715 signals = S_RDBAND;
7690 7716 pollwakeups = POLLIN | POLLRDBAND;
7691 7717 }
7692 7718 }
7693 7719
7694 7720 if (pollwakeups != 0) {
7695 7721 if (pollwakeups == (POLLIN | POLLRDNORM)) {
7696 7722 if (!(stp->sd_rput_opt & SR_POLLIN))
7697 7723 goto no_pollwake;
7698 7724 stp->sd_rput_opt &= ~SR_POLLIN;
7699 7725 }
7700 7726 mutex_exit(&stp->sd_lock);
7701 7727 pollwakeup(&stp->sd_pollist, pollwakeups);
7702 7728 mutex_enter(&stp->sd_lock);
7703 7729 }
7704 7730 no_pollwake:
7705 7731
7706 7732 if (stp->sd_sigflags & signals)
7707 7733 strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7708 7734 }
7709 7735 mutex_exit(&stp->sd_lock);
7710 7736
7711 7737 rvp->r_val1 = more;
7712 7738 return (error);
7713 7739 #undef _LASTMARK
7714 7740 }
7715 7741
7716 7742 /*
7717 7743 * Put a message downstream.
7718 7744 *
7719 7745 * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7720 7746 */
7721 7747 int
7722 7748 strputmsg(
7723 7749 struct vnode *vp,
7724 7750 struct strbuf *mctl,
7725 7751 struct strbuf *mdata,
7726 7752 unsigned char pri,
7727 7753 int flag,
7728 7754 int fmode)
7729 7755 {
7730 7756 struct stdata *stp;
7731 7757 queue_t *wqp;
7732 7758 mblk_t *mp;
7733 7759 ssize_t msgsize;
7734 7760 ssize_t rmin, rmax;
7735 7761 int error;
7736 7762 struct uio uios;
7737 7763 struct uio *uiop = &uios;
7738 7764 struct iovec iovs;
7739 7765 int xpg4 = 0;
7740 7766
7741 7767 ASSERT(vp->v_stream);
7742 7768 stp = vp->v_stream;
7743 7769 wqp = stp->sd_wrq;
7744 7770
7745 7771 /*
7746 7772 * If it is an XPG4 application, we need to send
7747 7773 * SIGPIPE below
7748 7774 */
7749 7775
7750 7776 xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7751 7777 flag &= ~MSG_XPG4;
7752 7778
7753 7779 if (AU_AUDITING())
7754 7780 audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7755 7781
7756 7782 mutex_enter(&stp->sd_lock);
7757 7783
7758 7784 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7759 7785 mutex_exit(&stp->sd_lock);
7760 7786 return (error);
7761 7787 }
7762 7788
7763 7789 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7764 7790 error = strwriteable(stp, B_FALSE, xpg4);
7765 7791 if (error != 0) {
7766 7792 mutex_exit(&stp->sd_lock);
7767 7793 return (error);
7768 7794 }
7769 7795 }
7770 7796
7771 7797 mutex_exit(&stp->sd_lock);
7772 7798
7773 7799 /*
7774 7800 * Check for legal flag value.
7775 7801 */
7776 7802 switch (flag) {
7777 7803 case MSG_HIPRI:
7778 7804 if ((mctl->len < 0) || (pri != 0))
7779 7805 return (EINVAL);
7780 7806 break;
7781 7807 case MSG_BAND:
7782 7808 break;
7783 7809
7784 7810 default:
7785 7811 return (EINVAL);
7786 7812 }
7787 7813
7788 7814 TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7789 7815 "strputmsg in:stp %p", stp);
7790 7816
7791 7817 /* get these values from those cached in the stream head */
7792 7818 rmin = stp->sd_qn_minpsz;
7793 7819 rmax = stp->sd_qn_maxpsz;
7794 7820
7795 7821 /*
7796 7822 * Make sure ctl and data sizes together fall within the
7797 7823 * limits of the max and min receive packet sizes and do
7798 7824 * not exceed system limit.
7799 7825 */
7800 7826 ASSERT((rmax >= 0) || (rmax == INFPSZ));
7801 7827 if (rmax == 0) {
7802 7828 return (ERANGE);
7803 7829 }
7804 7830 /*
7805 7831 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7806 7832 * Needed to prevent partial failures in the strmakedata loop.
7807 7833 */
7808 7834 if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7809 7835 rmax = stp->sd_maxblk;
7810 7836
7811 7837 if ((msgsize = mdata->len) < 0) {
7812 7838 msgsize = 0;
7813 7839 rmin = 0; /* no range check for NULL data part */
7814 7840 }
7815 7841 if ((msgsize < rmin) ||
7816 7842 ((msgsize > rmax) && (rmax != INFPSZ)) ||
7817 7843 (mctl->len > strctlsz)) {
7818 7844 return (ERANGE);
7819 7845 }
7820 7846
7821 7847 /*
7822 7848 * Setup uio and iov for data part
7823 7849 */
7824 7850 iovs.iov_base = mdata->buf;
7825 7851 iovs.iov_len = msgsize;
7826 7852 uios.uio_iov = &iovs;
7827 7853 uios.uio_iovcnt = 1;
7828 7854 uios.uio_loffset = 0;
7829 7855 uios.uio_segflg = UIO_USERSPACE;
7830 7856 uios.uio_fmode = fmode;
7831 7857 uios.uio_extflg = UIO_COPY_DEFAULT;
7832 7858 uios.uio_resid = msgsize;
7833 7859 uios.uio_offset = 0;
7834 7860
7835 7861 /* Ignore flow control in strput for HIPRI */
7836 7862 if (flag & MSG_HIPRI)
7837 7863 flag |= MSG_IGNFLOW;
7838 7864
7839 7865 for (;;) {
7840 7866 int done = 0;
7841 7867
7842 7868 /*
7843 7869 * strput will always free the ctl mblk - even when strput
7844 7870 * fails.
7845 7871 */
7846 7872 if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7847 7873 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7848 7874 "strputmsg out:stp %p out %d error %d",
7849 7875 stp, 1, error);
7850 7876 return (error);
7851 7877 }
7852 7878 /*
7853 7879 * Verify that the whole message can be transferred by
7854 7880 * strput.
7855 7881 */
7856 7882 ASSERT(stp->sd_maxblk == INFPSZ ||
7857 7883 stp->sd_maxblk >= mdata->len);
7858 7884
7859 7885 msgsize = mdata->len;
7860 7886 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7861 7887 mdata->len = msgsize;
7862 7888
7863 7889 if (error == 0)
7864 7890 break;
7865 7891
7866 7892 if (error != EWOULDBLOCK)
7867 7893 goto out;
7868 7894
7869 7895 mutex_enter(&stp->sd_lock);
7870 7896 /*
7871 7897 * Check for a missed wakeup.
7872 7898 * Needed since strput did not hold sd_lock across
7873 7899 * the canputnext.
7874 7900 */
7875 7901 if (bcanputnext(wqp, pri)) {
7876 7902 /* Try again */
7877 7903 mutex_exit(&stp->sd_lock);
7878 7904 continue;
7879 7905 }
7880 7906 TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7881 7907 "strputmsg wait:stp %p waits pri %d", stp, pri);
7882 7908 if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7883 7909 &done)) != 0) || done) {
7884 7910 mutex_exit(&stp->sd_lock);
7885 7911 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7886 7912 "strputmsg out:q %p out %d error %d",
7887 7913 stp, 0, error);
7888 7914 return (error);
7889 7915 }
7890 7916 TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7891 7917 "strputmsg wake:stp %p wakes", stp);
7892 7918 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7893 7919 mutex_exit(&stp->sd_lock);
7894 7920 return (error);
7895 7921 }
7896 7922 mutex_exit(&stp->sd_lock);
7897 7923 }
7898 7924 out:
7899 7925 /*
7900 7926 * For historic reasons, applications expect EAGAIN
7901 7927 * when data mblk could not be allocated. so change
7902 7928 * ENOMEM back to EAGAIN
7903 7929 */
7904 7930 if (error == ENOMEM)
7905 7931 error = EAGAIN;
7906 7932 TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7907 7933 "strputmsg out:stp %p out %d error %d", stp, 2, error);
7908 7934 return (error);
7909 7935 }
7910 7936
7911 7937 /*
7912 7938 * Put a message downstream.
7913 7939 * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7914 7940 * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7915 7941 * and the fmode parameter.
7916 7942 *
7917 7943 * This routine handles the consolidation private flags:
7918 7944 * MSG_IGNERROR Ignore any stream head error except STPLEX.
7919 7945 * MSG_HOLDSIG Hold signals while waiting for data.
7920 7946 * MSG_IGNFLOW Don't check streams flow control.
7921 7947 *
7922 7948 * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7923 7949 */
7924 7950 int
7925 7951 kstrputmsg(
7926 7952 struct vnode *vp,
7927 7953 mblk_t *mctl,
7928 7954 struct uio *uiop,
7929 7955 ssize_t msgsize,
7930 7956 unsigned char pri,
7931 7957 int flag,
7932 7958 int fmode)
7933 7959 {
7934 7960 struct stdata *stp;
7935 7961 queue_t *wqp;
7936 7962 ssize_t rmin, rmax;
7937 7963 int error;
7938 7964
7939 7965 ASSERT(vp->v_stream);
7940 7966 stp = vp->v_stream;
7941 7967 wqp = stp->sd_wrq;
7942 7968 if (AU_AUDITING())
7943 7969 audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7944 7970 if (mctl == NULL)
7945 7971 return (EINVAL);
7946 7972
7947 7973 mutex_enter(&stp->sd_lock);
7948 7974
7949 7975 if ((error = i_straccess(stp, JCWRITE)) != 0) {
7950 7976 mutex_exit(&stp->sd_lock);
7951 7977 freemsg(mctl);
7952 7978 return (error);
7953 7979 }
7954 7980
7955 7981 if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7956 7982 if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7957 7983 error = strwriteable(stp, B_FALSE, B_TRUE);
7958 7984 if (error != 0) {
7959 7985 mutex_exit(&stp->sd_lock);
7960 7986 freemsg(mctl);
7961 7987 return (error);
7962 7988 }
7963 7989 }
7964 7990 }
7965 7991
7966 7992 mutex_exit(&stp->sd_lock);
7967 7993
7968 7994 /*
7969 7995 * Check for legal flag value.
7970 7996 */
7971 7997 switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
7972 7998 case MSG_HIPRI:
7973 7999 if (pri != 0) {
7974 8000 freemsg(mctl);
7975 8001 return (EINVAL);
7976 8002 }
7977 8003 break;
7978 8004 case MSG_BAND:
7979 8005 break;
7980 8006 default:
7981 8007 freemsg(mctl);
7982 8008 return (EINVAL);
7983 8009 }
7984 8010
7985 8011 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
7986 8012 "kstrputmsg in:stp %p", stp);
7987 8013
7988 8014 /* get these values from those cached in the stream head */
7989 8015 rmin = stp->sd_qn_minpsz;
7990 8016 rmax = stp->sd_qn_maxpsz;
7991 8017
7992 8018 /*
7993 8019 * Make sure ctl and data sizes together fall within the
7994 8020 * limits of the max and min receive packet sizes and do
7995 8021 * not exceed system limit.
7996 8022 */
7997 8023 ASSERT((rmax >= 0) || (rmax == INFPSZ));
7998 8024 if (rmax == 0) {
7999 8025 freemsg(mctl);
8000 8026 return (ERANGE);
8001 8027 }
8002 8028 /*
8003 8029 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8004 8030 * Needed to prevent partial failures in the strmakedata loop.
8005 8031 */
8006 8032 if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8007 8033 rmax = stp->sd_maxblk;
8008 8034
8009 8035 if (uiop == NULL) {
8010 8036 msgsize = -1;
8011 8037 rmin = -1; /* no range check for NULL data part */
8012 8038 } else {
8013 8039 /* Use uio flags as well as the fmode parameter flags */
8014 8040 fmode |= uiop->uio_fmode;
8015 8041
8016 8042 if ((msgsize < rmin) ||
8017 8043 ((msgsize > rmax) && (rmax != INFPSZ))) {
8018 8044 freemsg(mctl);
8019 8045 return (ERANGE);
8020 8046 }
8021 8047 }
8022 8048
8023 8049 /* Ignore flow control in strput for HIPRI */
8024 8050 if (flag & MSG_HIPRI)
8025 8051 flag |= MSG_IGNFLOW;
8026 8052
8027 8053 for (;;) {
8028 8054 int done = 0;
8029 8055 int waitflag;
8030 8056 mblk_t *mp;
8031 8057
8032 8058 /*
8033 8059 * strput will always free the ctl mblk - even when strput
8034 8060 * fails. If MSG_IGNFLOW is set then any error returned
8035 8061 * will cause us to break the loop, so we don't need a copy
8036 8062 * of the message. If MSG_IGNFLOW is not set, then we can
8037 8063 * get hit by flow control and be forced to try again. In
8038 8064 * this case we need to have a copy of the message. We
8039 8065 * do this using copymsg since the message may get modified
8040 8066 * by something below us.
8041 8067 *
8042 8068 * We've observed that many TPI providers do not check db_ref
8043 8069 * on the control messages but blindly reuse them for the
8044 8070 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8045 8071 * friendly to such providers than using dupmsg. Also, note
8046 8072 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8047 8073 * Only data messages are subject to flow control, hence
8048 8074 * subject to this copymsg.
8049 8075 */
8050 8076 if (flag & MSG_IGNFLOW) {
8051 8077 mp = mctl;
8052 8078 mctl = NULL;
8053 8079 } else {
8054 8080 do {
8055 8081 /*
8056 8082 * If a message has a free pointer, the message
8057 8083 * must be dupmsg to maintain this pointer.
8058 8084 * Code using this facility must be sure
8059 8085 * that modules below will not change the
8060 8086 * contents of the dblk without checking db_ref
8061 8087 * first. If db_ref is > 1, then the module
8062 8088 * needs to do a copymsg first. Otherwise,
8063 8089 * the contents of the dblk may become
8064 8090 * inconsistent because the freesmg/freeb below
8065 8091 * may end up calling atomic_add_32_nv.
8066 8092 * The atomic_add_32_nv in freeb (accessing
8067 8093 * all of db_ref, db_type, db_flags, and
8068 8094 * db_struioflag) does not prevent other threads
8069 8095 * from concurrently trying to modify e.g.
8070 8096 * db_type.
8071 8097 */
8072 8098 if (mctl->b_datap->db_frtnp != NULL)
8073 8099 mp = dupmsg(mctl);
8074 8100 else
8075 8101 mp = copymsg(mctl);
8076 8102
8077 8103 if (mp != NULL)
8078 8104 break;
8079 8105
8080 8106 error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8081 8107 if (error) {
8082 8108 freemsg(mctl);
8083 8109 return (error);
8084 8110 }
8085 8111 } while (mp == NULL);
8086 8112 }
8087 8113 /*
8088 8114 * Verify that all of msgsize can be transferred by
8089 8115 * strput.
8090 8116 */
8091 8117 ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8092 8118 error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8093 8119 if (error == 0)
8094 8120 break;
8095 8121
8096 8122 if (error != EWOULDBLOCK)
8097 8123 goto out;
8098 8124
8099 8125 /*
8100 8126 * IF MSG_IGNFLOW is set we should have broken out of loop
8101 8127 * above.
8102 8128 */
8103 8129 ASSERT(!(flag & MSG_IGNFLOW));
8104 8130 mutex_enter(&stp->sd_lock);
8105 8131 /*
8106 8132 * Check for a missed wakeup.
8107 8133 * Needed since strput did not hold sd_lock across
8108 8134 * the canputnext.
8109 8135 */
8110 8136 if (bcanputnext(wqp, pri)) {
8111 8137 /* Try again */
8112 8138 mutex_exit(&stp->sd_lock);
8113 8139 continue;
8114 8140 }
8115 8141 TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8116 8142 "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8117 8143
8118 8144 waitflag = WRITEWAIT;
8119 8145 if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8120 8146 if (flag & MSG_HOLDSIG)
8121 8147 waitflag |= STR_NOSIG;
8122 8148 if (flag & MSG_IGNERROR)
8123 8149 waitflag |= STR_NOERROR;
8124 8150 }
8125 8151 if (((error = strwaitq(stp, waitflag,
8126 8152 (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8127 8153 mutex_exit(&stp->sd_lock);
8128 8154 TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8129 8155 "kstrputmsg out:stp %p out %d error %d",
8130 8156 stp, 0, error);
8131 8157 freemsg(mctl);
8132 8158 return (error);
8133 8159 }
8134 8160 TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8135 8161 "kstrputmsg wake:stp %p wakes", stp);
8136 8162 if ((error = i_straccess(stp, JCWRITE)) != 0) {
8137 8163 mutex_exit(&stp->sd_lock);
8138 8164 freemsg(mctl);
8139 8165 return (error);
8140 8166 }
8141 8167 mutex_exit(&stp->sd_lock);
8142 8168 }
8143 8169 out:
8144 8170 freemsg(mctl);
8145 8171 /*
8146 8172 * For historic reasons, applications expect EAGAIN
8147 8173 * when data mblk could not be allocated. so change
8148 8174 * ENOMEM back to EAGAIN
8149 8175 */
8150 8176 if (error == ENOMEM)
8151 8177 error = EAGAIN;
8152 8178 TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8153 8179 "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8154 8180 return (error);
8155 8181 }
8156 8182
8157 8183 /*
8158 8184 * Determines whether the necessary conditions are set on a stream
8159 8185 * for it to be readable, writeable, or have exceptions.
8160 8186 *
8161 8187 * strpoll handles the consolidation private events:
8162 8188 * POLLNOERR Do not return POLLERR even if there are stream
8163 8189 * head errors.
8164 8190 * Used by sockfs.
8165 8191 * POLLRDDATA Do not return POLLIN unless at least one message on
8166 8192 * the queue contains one or more M_DATA mblks. Thus
8167 8193 * when this flag is set a queue with only
8168 8194 * M_PROTO/M_PCPROTO mblks does not return POLLIN.
8169 8195 * Used by sockfs to ignore T_EXDATA_IND messages.
8170 8196 *
8171 8197 * Note: POLLRDDATA assumes that synch streams only return messages with
8172 8198 * an M_DATA attached (i.e. not messages consisting of only
8173 8199 * an M_PROTO/M_PCPROTO part).
8174 8200 */
8175 8201 int
8176 8202 strpoll(
8177 8203 struct stdata *stp,
8178 8204 short events_arg,
8179 8205 int anyyet,
8180 8206 short *reventsp,
8181 8207 struct pollhead **phpp)
8182 8208 {
8183 8209 int events = (ushort_t)events_arg;
8184 8210 int retevents = 0;
8185 8211 mblk_t *mp;
8186 8212 qband_t *qbp;
8187 8213 long sd_flags = stp->sd_flag;
8188 8214 int headlocked = 0;
8189 8215
8190 8216 /*
8191 8217 * For performance, a single 'if' tests for most possible edge
8192 8218 * conditions in one shot
8193 8219 */
8194 8220 if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8195 8221 if (sd_flags & STPLEX) {
8196 8222 *reventsp = POLLNVAL;
8197 8223 return (EINVAL);
8198 8224 }
8199 8225 if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8200 8226 (sd_flags & STRDERR)) ||
8201 8227 ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8202 8228 (sd_flags & STWRERR))) {
8203 8229 if (!(events & POLLNOERR)) {
8204 8230 *reventsp = POLLERR;
8205 8231 return (0);
8206 8232 }
8207 8233 }
8208 8234 }
8209 8235 if (sd_flags & STRHUP) {
8210 8236 retevents |= POLLHUP;
8211 8237 } else if (events & (POLLWRNORM | POLLWRBAND)) {
8212 8238 queue_t *tq;
8213 8239 queue_t *qp = stp->sd_wrq;
8214 8240
8215 8241 claimstr(qp);
8216 8242 /* Find next module forward that has a service procedure */
8217 8243 tq = qp->q_next->q_nfsrv;
8218 8244 ASSERT(tq != NULL);
8219 8245
8220 8246 polllock(&stp->sd_pollist, QLOCK(tq));
8221 8247 if (events & POLLWRNORM) {
8222 8248 queue_t *sqp;
8223 8249
8224 8250 if (tq->q_flag & QFULL)
8225 8251 /* ensure backq svc procedure runs */
8226 8252 tq->q_flag |= QWANTW;
8227 8253 else if ((sqp = stp->sd_struiowrq) != NULL) {
8228 8254 /* Check sync stream barrier write q */
8229 8255 mutex_exit(QLOCK(tq));
8230 8256 polllock(&stp->sd_pollist, QLOCK(sqp));
8231 8257 if (sqp->q_flag & QFULL)
8232 8258 /* ensure pollwakeup() is done */
8233 8259 sqp->q_flag |= QWANTWSYNC;
8234 8260 else
8235 8261 retevents |= POLLOUT;
8236 8262 /* More write events to process ??? */
8237 8263 if (! (events & POLLWRBAND)) {
8238 8264 mutex_exit(QLOCK(sqp));
8239 8265 releasestr(qp);
8240 8266 goto chkrd;
8241 8267 }
8242 8268 mutex_exit(QLOCK(sqp));
8243 8269 polllock(&stp->sd_pollist, QLOCK(tq));
8244 8270 } else
8245 8271 retevents |= POLLOUT;
8246 8272 }
8247 8273 if (events & POLLWRBAND) {
8248 8274 qbp = tq->q_bandp;
8249 8275 if (qbp) {
8250 8276 while (qbp) {
8251 8277 if (qbp->qb_flag & QB_FULL)
8252 8278 qbp->qb_flag |= QB_WANTW;
8253 8279 else
8254 8280 retevents |= POLLWRBAND;
8255 8281 qbp = qbp->qb_next;
8256 8282 }
8257 8283 } else {
8258 8284 retevents |= POLLWRBAND;
8259 8285 }
8260 8286 }
8261 8287 mutex_exit(QLOCK(tq));
8262 8288 releasestr(qp);
8263 8289 }
8264 8290 chkrd:
8265 8291 if (sd_flags & STRPRI) {
8266 8292 retevents |= (events & POLLPRI);
8267 8293 } else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8268 8294 queue_t *qp = _RD(stp->sd_wrq);
8269 8295 int normevents = (events & (POLLIN | POLLRDNORM));
8270 8296
8271 8297 /*
8272 8298 * Note: Need to do polllock() here since ps_lock may be
8273 8299 * held. See bug 4191544.
8274 8300 */
8275 8301 polllock(&stp->sd_pollist, &stp->sd_lock);
8276 8302 headlocked = 1;
8277 8303 mp = qp->q_first;
8278 8304 while (mp) {
8279 8305 /*
8280 8306 * For POLLRDDATA we scan b_cont and b_next until we
8281 8307 * find an M_DATA.
8282 8308 */
8283 8309 if ((events & POLLRDDATA) &&
8284 8310 mp->b_datap->db_type != M_DATA) {
8285 8311 mblk_t *nmp = mp->b_cont;
8286 8312
8287 8313 while (nmp != NULL &&
8288 8314 nmp->b_datap->db_type != M_DATA)
8289 8315 nmp = nmp->b_cont;
8290 8316 if (nmp == NULL) {
8291 8317 mp = mp->b_next;
8292 8318 continue;
8293 8319 }
8294 8320 }
8295 8321 if (mp->b_band == 0)
8296 8322 retevents |= normevents;
8297 8323 else
8298 8324 retevents |= (events & (POLLIN | POLLRDBAND));
8299 8325 break;
8300 8326 }
8301 8327 if (! (retevents & normevents) &&
8302 8328 (stp->sd_wakeq & RSLEEP)) {
8303 8329 /*
8304 8330 * Sync stream barrier read queue has data.
8305 8331 */
8306 8332 retevents |= normevents;
8307 8333 }
8308 8334 /* Treat eof as normal data */
8309 8335 if (sd_flags & STREOF)
8310 8336 retevents |= normevents;
8311 8337 }
8312 8338
8313 8339 *reventsp = (short)retevents;
8314 8340 if (retevents) {
8315 8341 if (headlocked)
8316 8342 mutex_exit(&stp->sd_lock);
8317 8343 return (0);
8318 8344 }
8319 8345
8320 8346 /*
8321 8347 * If poll() has not found any events yet, set up event cell
8322 8348 * to wake up the poll if a requested event occurs on this
8323 8349 * stream. Check for collisions with outstanding poll requests.
8324 8350 */
8325 8351 if (!anyyet) {
8326 8352 *phpp = &stp->sd_pollist;
8327 8353 if (headlocked == 0) {
8328 8354 polllock(&stp->sd_pollist, &stp->sd_lock);
8329 8355 headlocked = 1;
8330 8356 }
8331 8357 stp->sd_rput_opt |= SR_POLLIN;
8332 8358 }
8333 8359 if (headlocked)
8334 8360 mutex_exit(&stp->sd_lock);
8335 8361 return (0);
8336 8362 }
8337 8363
8338 8364 /*
8339 8365 * The purpose of putback() is to assure sleeping polls/reads
8340 8366 * are awakened when there are no new messages arriving at the,
8341 8367 * stream head, and a message is placed back on the read queue.
8342 8368 *
8343 8369 * sd_lock must be held when messages are placed back on stream
8344 8370 * head. (getq() holds sd_lock when it removes messages from
8345 8371 * the queue)
8346 8372 */
8347 8373
8348 8374 static void
8349 8375 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8350 8376 {
8351 8377 mblk_t *qfirst;
8352 8378 ASSERT(MUTEX_HELD(&stp->sd_lock));
8353 8379
8354 8380 /*
8355 8381 * As a result of lock-step ordering around q_lock and sd_lock,
8356 8382 * it's possible for function calls like putnext() and
8357 8383 * canputnext() to get an inaccurate picture of how much
8358 8384 * data is really being processed at the stream head.
8359 8385 * We only consolidate with existing messages on the queue
8360 8386 * if the length of the message we want to put back is smaller
8361 8387 * than the queue hiwater mark.
8362 8388 */
8363 8389 if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8364 8390 (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8365 8391 (DB_TYPE(qfirst) == M_DATA) &&
8366 8392 ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8367 8393 ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8368 8394 (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8369 8395 /*
8370 8396 * We use the same logic as defined in strrput()
8371 8397 * but in reverse as we are putting back onto the
8372 8398 * queue and want to retain byte ordering.
8373 8399 * Consolidate M_DATA messages with M_DATA ONLY.
8374 8400 * strrput() allows the consolidation of M_DATA onto
8375 8401 * M_PROTO | M_PCPROTO but not the other way round.
8376 8402 *
8377 8403 * The consolidation does not take place if the message
8378 8404 * we are returning to the queue is marked with either
8379 8405 * of the marks or the delim flag or if q_first
8380 8406 * is marked with MSGMARK. The MSGMARK check is needed to
8381 8407 * handle the odd semantics of MSGMARK where essentially
8382 8408 * the whole message is to be treated as marked.
8383 8409 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8384 8410 * to the front of the b_cont chain.
8385 8411 */
8386 8412 rmvq_noenab(q, qfirst);
8387 8413
8388 8414 /*
8389 8415 * The first message in the b_cont list
8390 8416 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8391 8417 * We need to handle the case where we
8392 8418 * are appending:
8393 8419 *
8394 8420 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8395 8421 * 2) a MSGMARKNEXT to a plain message.
8396 8422 * 3) a MSGNOTMARKNEXT to a plain message
8397 8423 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8398 8424 * message.
8399 8425 *
8400 8426 * Thus we never append a MSGMARKNEXT or
8401 8427 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8402 8428 */
8403 8429 if (qfirst->b_flag & MSGMARKNEXT) {
8404 8430 bp->b_flag |= MSGMARKNEXT;
8405 8431 bp->b_flag &= ~MSGNOTMARKNEXT;
8406 8432 qfirst->b_flag &= ~MSGMARKNEXT;
8407 8433 } else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8408 8434 bp->b_flag |= MSGNOTMARKNEXT;
8409 8435 qfirst->b_flag &= ~MSGNOTMARKNEXT;
8410 8436 }
8411 8437
8412 8438 linkb(bp, qfirst);
8413 8439 }
8414 8440 (void) putbq(q, bp);
8415 8441
8416 8442 /*
8417 8443 * A message may have come in when the sd_lock was dropped in the
8418 8444 * calling routine. If this is the case and STR*ATMARK info was
8419 8445 * received, need to move that from the stream head to the q_last
8420 8446 * so that SIOCATMARK can return the proper value.
8421 8447 */
8422 8448 if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8423 8449 unsigned short *flagp = &q->q_last->b_flag;
8424 8450 uint_t b_flag = (uint_t)*flagp;
8425 8451
8426 8452 if (stp->sd_flag & STRATMARK) {
8427 8453 b_flag &= ~MSGNOTMARKNEXT;
8428 8454 b_flag |= MSGMARKNEXT;
8429 8455 stp->sd_flag &= ~STRATMARK;
8430 8456 } else {
8431 8457 b_flag &= ~MSGMARKNEXT;
8432 8458 b_flag |= MSGNOTMARKNEXT;
8433 8459 stp->sd_flag &= ~STRNOTATMARK;
8434 8460 }
8435 8461 *flagp = (unsigned short) b_flag;
8436 8462 }
8437 8463
8438 8464 #ifdef DEBUG
8439 8465 /*
8440 8466 * Make sure that the flags are not messed up.
8441 8467 */
8442 8468 {
8443 8469 mblk_t *mp;
8444 8470 mp = q->q_last;
8445 8471 while (mp != NULL) {
8446 8472 ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8447 8473 (MSGMARKNEXT|MSGNOTMARKNEXT));
8448 8474 mp = mp->b_cont;
8449 8475 }
8450 8476 }
8451 8477 #endif
8452 8478 if (q->q_first == bp) {
8453 8479 short pollevents;
8454 8480
8455 8481 if (stp->sd_flag & RSLEEP) {
8456 8482 stp->sd_flag &= ~RSLEEP;
8457 8483 cv_broadcast(&q->q_wait);
8458 8484 }
8459 8485 if (stp->sd_flag & STRPRI) {
8460 8486 pollevents = POLLPRI;
8461 8487 } else {
8462 8488 if (band == 0) {
8463 8489 if (!(stp->sd_rput_opt & SR_POLLIN))
8464 8490 return;
8465 8491 stp->sd_rput_opt &= ~SR_POLLIN;
8466 8492 pollevents = POLLIN | POLLRDNORM;
8467 8493 } else {
8468 8494 pollevents = POLLIN | POLLRDBAND;
8469 8495 }
8470 8496 }
8471 8497 mutex_exit(&stp->sd_lock);
8472 8498 pollwakeup(&stp->sd_pollist, pollevents);
8473 8499 mutex_enter(&stp->sd_lock);
8474 8500 }
8475 8501 }
8476 8502
8477 8503 /*
8478 8504 * Return the held vnode attached to the stream head of a
8479 8505 * given queue
8480 8506 * It is the responsibility of the calling routine to ensure
8481 8507 * that the queue does not go away (e.g. pop).
8482 8508 */
8483 8509 vnode_t *
8484 8510 strq2vp(queue_t *qp)
8485 8511 {
8486 8512 vnode_t *vp;
8487 8513 vp = STREAM(qp)->sd_vnode;
8488 8514 ASSERT(vp != NULL);
8489 8515 VN_HOLD(vp);
8490 8516 return (vp);
8491 8517 }
8492 8518
8493 8519 /*
8494 8520 * return the stream head write queue for the given vp
8495 8521 * It is the responsibility of the calling routine to ensure
8496 8522 * that the stream or vnode do not close.
8497 8523 */
8498 8524 queue_t *
8499 8525 strvp2wq(vnode_t *vp)
8500 8526 {
8501 8527 ASSERT(vp->v_stream != NULL);
8502 8528 return (vp->v_stream->sd_wrq);
8503 8529 }
8504 8530
8505 8531 /*
8506 8532 * pollwakeup stream head
8507 8533 * It is the responsibility of the calling routine to ensure
8508 8534 * that the stream or vnode do not close.
8509 8535 */
8510 8536 void
8511 8537 strpollwakeup(vnode_t *vp, short event)
8512 8538 {
8513 8539 ASSERT(vp->v_stream);
8514 8540 pollwakeup(&vp->v_stream->sd_pollist, event);
8515 8541 }
8516 8542
8517 8543 /*
8518 8544 * Mate the stream heads of two vnodes together. If the two vnodes are the
8519 8545 * same, we just make the write-side point at the read-side -- otherwise,
8520 8546 * we do a full mate. Only works on vnodes associated with streams that are
8521 8547 * still being built and thus have only a stream head.
8522 8548 */
8523 8549 void
8524 8550 strmate(vnode_t *vp1, vnode_t *vp2)
8525 8551 {
8526 8552 queue_t *wrq1 = strvp2wq(vp1);
8527 8553 queue_t *wrq2 = strvp2wq(vp2);
8528 8554
8529 8555 /*
8530 8556 * Verify that there are no modules on the stream yet. We also
8531 8557 * rely on the stream head always having a service procedure to
8532 8558 * avoid tweaking q_nfsrv.
8533 8559 */
8534 8560 ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8535 8561 ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8536 8562 ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8537 8563
8538 8564 /*
8539 8565 * If the queues are the same, just twist; otherwise do a full mate.
8540 8566 */
8541 8567 if (wrq1 == wrq2) {
8542 8568 wrq1->q_next = _RD(wrq1);
8543 8569 } else {
8544 8570 wrq1->q_next = _RD(wrq2);
8545 8571 wrq2->q_next = _RD(wrq1);
8546 8572 STREAM(wrq1)->sd_mate = STREAM(wrq2);
8547 8573 STREAM(wrq1)->sd_flag |= STRMATE;
8548 8574 STREAM(wrq2)->sd_mate = STREAM(wrq1);
8549 8575 STREAM(wrq2)->sd_flag |= STRMATE;
8550 8576 }
8551 8577 }
8552 8578
8553 8579 /*
8554 8580 * XXX will go away when console is correctly fixed.
8555 8581 * Clean up the console PIDS, from previous I_SETSIG,
8556 8582 * called only for cnopen which never calls strclean().
8557 8583 */
8558 8584 void
8559 8585 str_cn_clean(struct vnode *vp)
8560 8586 {
8561 8587 strsig_t *ssp, *pssp, *tssp;
8562 8588 struct stdata *stp;
8563 8589 struct pid *pidp;
8564 8590 int update = 0;
8565 8591
8566 8592 ASSERT(vp->v_stream);
8567 8593 stp = vp->v_stream;
8568 8594 pssp = NULL;
8569 8595 mutex_enter(&stp->sd_lock);
8570 8596 ssp = stp->sd_siglist;
8571 8597 while (ssp) {
8572 8598 mutex_enter(&pidlock);
8573 8599 pidp = ssp->ss_pidp;
8574 8600 /*
8575 8601 * Get rid of PID if the proc is gone.
8576 8602 */
8577 8603 if (pidp->pid_prinactive) {
8578 8604 tssp = ssp->ss_next;
8579 8605 if (pssp)
8580 8606 pssp->ss_next = tssp;
8581 8607 else
8582 8608 stp->sd_siglist = tssp;
8583 8609 ASSERT(pidp->pid_ref <= 1);
8584 8610 PID_RELE(ssp->ss_pidp);
8585 8611 mutex_exit(&pidlock);
8586 8612 kmem_free(ssp, sizeof (strsig_t));
8587 8613 update = 1;
8588 8614 ssp = tssp;
8589 8615 continue;
8590 8616 } else
8591 8617 mutex_exit(&pidlock);
8592 8618 pssp = ssp;
8593 8619 ssp = ssp->ss_next;
8594 8620 }
8595 8621 if (update) {
8596 8622 stp->sd_sigflags = 0;
8597 8623 for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8598 8624 stp->sd_sigflags |= ssp->ss_events;
8599 8625 }
8600 8626 mutex_exit(&stp->sd_lock);
8601 8627 }
8602 8628
8603 8629 /*
8604 8630 * Return B_TRUE if there is data in the message, B_FALSE otherwise.
↓ open down ↓ |
2910 lines elided |
↑ open up ↑ |
8605 8631 */
8606 8632 static boolean_t
8607 8633 msghasdata(mblk_t *bp)
8608 8634 {
8609 8635 for (; bp; bp = bp->b_cont)
8610 8636 if (bp->b_datap->db_type == M_DATA) {
8611 8637 ASSERT(bp->b_wptr >= bp->b_rptr);
8612 8638 if (bp->b_wptr > bp->b_rptr)
8613 8639 return (B_TRUE);
8614 8640 }
8641 + return (B_FALSE);
8642 +}
8643 +
8644 +/*
8645 + * Check whether a stream is an XTI stream or not.
8646 + */
8647 +static boolean_t
8648 +is_xti_str(const struct stdata *stp)
8649 +{
8650 + struct devnames *dnp;
8651 + vnode_t *vn;
8652 + major_t major;
8653 + if ((vn = stp->sd_vnode) != NULL && vn->v_type == VCHR &&
8654 + vn->v_rdev != 0) {
8655 + major = getmajor(vn->v_rdev);
8656 + dnp = (major != DDI_MAJOR_T_NONE && major >= 0 &&
8657 + major < devcnt) ? &devnamesp[major] : NULL;
8658 + if (dnp != NULL && dnp->dn_name != NULL &&
8659 + (strcmp(dnp->dn_name, "ip") == 0 ||
8660 + strcmp(dnp->dn_name, "tcp") == 0 ||
8661 + strcmp(dnp->dn_name, "udp") == 0 ||
8662 + strcmp(dnp->dn_name, "icmp") == 0 ||
8663 + strcmp(dnp->dn_name, "tl") == 0 ||
8664 + strcmp(dnp->dn_name, "ip6") == 0 ||
8665 + strcmp(dnp->dn_name, "tcp6") == 0 ||
8666 + strcmp(dnp->dn_name, "udp6") == 0 ||
8667 + strcmp(dnp->dn_name, "icmp6") == 0)) {
8668 + return (B_TRUE);
8669 + }
8670 + }
8615 8671 return (B_FALSE);
8616 8672 }
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX