1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25 /*
26 * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
27 * Copyright 2019 Joyent, Inc.
28 */
29
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/signal.h>
33 #include <sys/cmn_err.h>
34
35 #include <sys/stropts.h>
36 #include <sys/socket.h>
37 #include <sys/socketvar.h>
38 #include <sys/sockio.h>
39 #include <sys/strsubr.h>
40 #include <sys/strsun.h>
41 #include <sys/atomic.h>
42 #include <sys/tihdr.h>
43
44 #include <fs/sockfs/sockcommon.h>
45 #include <fs/sockfs/sockfilter_impl.h>
46 #include <fs/sockfs/socktpi.h>
47 #include <fs/sockfs/sodirect.h>
48 #include <sys/ddi.h>
49 #include <inet/ip.h>
50 #include <sys/time.h>
51 #include <sys/cmn_err.h>
52
53 #ifdef SOCK_TEST
54 extern int do_useracc;
55 extern clock_t sock_test_timelimit;
56 #endif /* SOCK_TEST */
57
58 #define MBLK_PULL_LEN 64
59 uint32_t so_mblk_pull_len = MBLK_PULL_LEN;
60
61 #ifdef DEBUG
62 boolean_t so_debug_length = B_FALSE;
63 static boolean_t so_check_length(sonode_t *so);
64 #endif
65
66 static int
so_acceptq_dequeue_locked(struct sonode * so,boolean_t dontblock,struct sonode ** nsop)67 so_acceptq_dequeue_locked(struct sonode *so, boolean_t dontblock,
68 struct sonode **nsop)
69 {
70 struct sonode *nso = NULL;
71
72 *nsop = NULL;
73 ASSERT(MUTEX_HELD(&so->so_acceptq_lock));
74 while ((nso = list_remove_head(&so->so_acceptq_list)) == NULL) {
75 /*
76 * No need to check so_error here, because it is not
77 * possible for a listening socket to be reset or otherwise
78 * disconnected.
79 *
80 * So now we just need check if it's ok to wait.
81 */
82 if (dontblock)
83 return (EWOULDBLOCK);
84 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
85 return (EINTR);
86
87 if (cv_wait_sig_swap(&so->so_acceptq_cv,
88 &so->so_acceptq_lock) == 0)
89 return (EINTR);
90 }
91
92 ASSERT(nso != NULL);
93 ASSERT(so->so_acceptq_len > 0);
94 so->so_acceptq_len--;
95 nso->so_listener = NULL;
96
97 *nsop = nso;
98
99 return (0);
100 }
101
102 /*
103 * int so_acceptq_dequeue(struct sonode *, boolean_t, struct sonode **)
104 *
105 * Pulls a connection off of the accept queue.
106 *
107 * Arguments:
108 * so - listening socket
109 * dontblock - indicate whether it's ok to sleep if there are no
110 * connections on the queue
111 * nsop - Value-return argument
112 *
113 * Return values:
114 * 0 when a connection is successfully dequeued, in which case nsop
115 * is set to point to the new connection. Upon failure a non-zero
116 * value is returned, and the value of nsop is set to NULL.
117 *
118 * Note:
119 * so_acceptq_dequeue() may return prematurly if the socket is falling
120 * back to TPI.
121 */
122 int
so_acceptq_dequeue(struct sonode * so,boolean_t dontblock,struct sonode ** nsop)123 so_acceptq_dequeue(struct sonode *so, boolean_t dontblock,
124 struct sonode **nsop)
125 {
126 int error;
127
128 mutex_enter(&so->so_acceptq_lock);
129 error = so_acceptq_dequeue_locked(so, dontblock, nsop);
130 mutex_exit(&so->so_acceptq_lock);
131
132 return (error);
133 }
134
135 static void
so_acceptq_flush_impl(struct sonode * so,list_t * list,boolean_t doclose)136 so_acceptq_flush_impl(struct sonode *so, list_t *list, boolean_t doclose)
137 {
138 struct sonode *nso;
139
140 while ((nso = list_remove_head(list)) != NULL) {
141 nso->so_listener = NULL;
142 if (doclose) {
143 (void) socket_close(nso, 0, CRED());
144 } else {
145 /*
146 * Only used for fallback - not possible when filters
147 * are present.
148 */
149 ASSERT(so->so_filter_active == 0);
150 /*
151 * Since the socket is on the accept queue, there can
152 * only be one reference. We drop the reference and
153 * just blow off the socket.
154 */
155 ASSERT(nso->so_count == 1);
156 nso->so_count--;
157 /* drop the proto ref */
158 VN_RELE(SOTOV(nso));
159 }
160 socket_destroy(nso);
161 }
162 }
163 /*
164 * void so_acceptq_flush(struct sonode *so)
165 *
166 * Removes all pending connections from a listening socket, and
167 * frees the associated resources.
168 *
169 * Arguments
170 * so - listening socket
171 * doclose - make a close downcall for each socket on the accept queue
172 *
173 * Return values:
174 * None.
175 *
176 * Note:
177 * The caller has to ensure that no calls to so_acceptq_enqueue() or
178 * so_acceptq_dequeue() occur while the accept queue is being flushed.
179 * So either the socket needs to be in a state where no operations
180 * would come in, or so_lock needs to be obtained.
181 */
182 void
so_acceptq_flush(struct sonode * so,boolean_t doclose)183 so_acceptq_flush(struct sonode *so, boolean_t doclose)
184 {
185 so_acceptq_flush_impl(so, &so->so_acceptq_list, doclose);
186 so_acceptq_flush_impl(so, &so->so_acceptq_defer, doclose);
187
188 so->so_acceptq_len = 0;
189 }
190
191 int
so_wait_connected_locked(struct sonode * so,boolean_t nonblock,sock_connid_t id)192 so_wait_connected_locked(struct sonode *so, boolean_t nonblock,
193 sock_connid_t id)
194 {
195 ASSERT(MUTEX_HELD(&so->so_lock));
196
197 /*
198 * The protocol has notified us that a connection attempt is being
199 * made, so before we wait for a notification to arrive we must
200 * clear out any errors associated with earlier connection attempts.
201 */
202 if (so->so_error != 0 && SOCK_CONNID_LT(so->so_proto_connid, id))
203 so->so_error = 0;
204
205 while (SOCK_CONNID_LT(so->so_proto_connid, id)) {
206 if (nonblock)
207 return (EINPROGRESS);
208
209 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
210 return (EINTR);
211
212 if (cv_wait_sig_swap(&so->so_state_cv, &so->so_lock) == 0)
213 return (EINTR);
214 }
215
216 if (so->so_error != 0)
217 return (sogeterr(so, B_TRUE));
218 /*
219 * Under normal circumstances, so_error should contain an error
220 * in case the connect failed. However, it is possible for another
221 * thread to come in a consume the error, so generate a sensible
222 * error in that case.
223 */
224 if ((so->so_state & SS_ISCONNECTED) == 0)
225 return (ECONNREFUSED);
226
227 return (0);
228 }
229
230 /*
231 * int so_wait_connected(struct sonode *so, boolean_t nonblock,
232 * sock_connid_t id)
233 *
234 * Wait until the socket is connected or an error has occured.
235 *
236 * Arguments:
237 * so - socket
238 * nonblock - indicate whether it's ok to sleep if the connection has
239 * not yet been established
240 * gen - generation number that was returned by the protocol
241 * when the operation was started
242 *
243 * Returns:
244 * 0 if the connection attempt was successful, or an error indicating why
245 * the connection attempt failed.
246 */
247 int
so_wait_connected(struct sonode * so,boolean_t nonblock,sock_connid_t id)248 so_wait_connected(struct sonode *so, boolean_t nonblock, sock_connid_t id)
249 {
250 int error;
251
252 mutex_enter(&so->so_lock);
253 error = so_wait_connected_locked(so, nonblock, id);
254 mutex_exit(&so->so_lock);
255
256 return (error);
257 }
258
259 int
so_snd_wait_qnotfull_locked(struct sonode * so,boolean_t dontblock)260 so_snd_wait_qnotfull_locked(struct sonode *so, boolean_t dontblock)
261 {
262 int error;
263
264 ASSERT(MUTEX_HELD(&so->so_lock));
265 while (SO_SND_FLOWCTRLD(so)) {
266 if (so->so_state & SS_CANTSENDMORE)
267 return (EPIPE);
268 if (dontblock)
269 return (EWOULDBLOCK);
270
271 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
272 return (EINTR);
273
274 if (so->so_sndtimeo == 0) {
275 /*
276 * Zero means disable timeout.
277 */
278 error = cv_wait_sig(&so->so_snd_cv, &so->so_lock);
279 } else {
280 error = cv_reltimedwait_sig(&so->so_snd_cv,
281 &so->so_lock, so->so_sndtimeo, TR_CLOCK_TICK);
282 }
283 if (error == 0)
284 return (EINTR);
285 else if (error == -1)
286 return (EAGAIN);
287 }
288 return (0);
289 }
290
291 /*
292 * int so_wait_sendbuf(struct sonode *so, boolean_t dontblock)
293 *
294 * Wait for the transport to notify us about send buffers becoming
295 * available.
296 */
297 int
so_snd_wait_qnotfull(struct sonode * so,boolean_t dontblock)298 so_snd_wait_qnotfull(struct sonode *so, boolean_t dontblock)
299 {
300 int error = 0;
301
302 mutex_enter(&so->so_lock);
303 so->so_snd_wakeup = B_TRUE;
304 error = so_snd_wait_qnotfull_locked(so, dontblock);
305 so->so_snd_wakeup = B_FALSE;
306 mutex_exit(&so->so_lock);
307
308 return (error);
309 }
310
311 void
so_snd_qfull(struct sonode * so)312 so_snd_qfull(struct sonode *so)
313 {
314 mutex_enter(&so->so_lock);
315 so->so_snd_qfull = B_TRUE;
316 mutex_exit(&so->so_lock);
317 }
318
319 void
so_snd_qnotfull(struct sonode * so)320 so_snd_qnotfull(struct sonode *so)
321 {
322 mutex_enter(&so->so_lock);
323 so->so_snd_qfull = B_FALSE;
324 /* wake up everyone waiting for buffers */
325 cv_broadcast(&so->so_snd_cv);
326 mutex_exit(&so->so_lock);
327 }
328
329 /*
330 * Change the process/process group to which SIGIO is sent.
331 */
332 int
socket_chgpgrp(struct sonode * so,pid_t pid)333 socket_chgpgrp(struct sonode *so, pid_t pid)
334 {
335 int error;
336
337 ASSERT(MUTEX_HELD(&so->so_lock));
338 if (pid != 0) {
339 /*
340 * Permissions check by sending signal 0.
341 * Note that when kill fails it does a
342 * set_errno causing the system call to fail.
343 */
344 error = kill(pid, 0);
345 if (error != 0) {
346 return (error);
347 }
348 }
349 so->so_pgrp = pid;
350 return (0);
351 }
352
353
354 /*
355 * Generate a SIGIO, for 'writable' events include siginfo structure,
356 * for read events just send the signal.
357 */
358 /*ARGSUSED*/
359 static void
socket_sigproc(proc_t * proc,int event)360 socket_sigproc(proc_t *proc, int event)
361 {
362 k_siginfo_t info;
363
364 ASSERT(event & (SOCKETSIG_WRITE | SOCKETSIG_READ | SOCKETSIG_URG));
365
366 if (event & SOCKETSIG_WRITE) {
367 info.si_signo = SIGPOLL;
368 info.si_code = POLL_OUT;
369 info.si_errno = 0;
370 info.si_fd = 0;
371 info.si_band = 0;
372 sigaddq(proc, NULL, &info, KM_NOSLEEP);
373 }
374 if (event & SOCKETSIG_READ) {
375 sigtoproc(proc, NULL, SIGPOLL);
376 }
377 if (event & SOCKETSIG_URG) {
378 sigtoproc(proc, NULL, SIGURG);
379 }
380 }
381
382 void
socket_sendsig(struct sonode * so,int event)383 socket_sendsig(struct sonode *so, int event)
384 {
385 proc_t *proc;
386
387 ASSERT(MUTEX_HELD(&so->so_lock));
388
389 if (so->so_pgrp == 0 || (!(so->so_state & SS_ASYNC) &&
390 event != SOCKETSIG_URG)) {
391 return;
392 }
393
394 dprint(3, ("sending sig %d to %d\n", event, so->so_pgrp));
395
396 if (so->so_pgrp > 0) {
397 /*
398 * XXX This unfortunately still generates
399 * a signal when a fd is closed but
400 * the proc is active.
401 */
402 mutex_enter(&pidlock);
403 /*
404 * Even if the thread started in another zone, we're receiving
405 * on behalf of this socket's zone, so find the proc using the
406 * socket's zone ID.
407 */
408 proc = prfind_zone(so->so_pgrp, so->so_zoneid);
409 if (proc == NULL) {
410 mutex_exit(&pidlock);
411 return;
412 }
413 mutex_enter(&proc->p_lock);
414 mutex_exit(&pidlock);
415 socket_sigproc(proc, event);
416 mutex_exit(&proc->p_lock);
417 } else {
418 /*
419 * Send to process group. Hold pidlock across
420 * calls to socket_sigproc().
421 */
422 pid_t pgrp = -so->so_pgrp;
423
424 mutex_enter(&pidlock);
425 /*
426 * Even if the thread started in another zone, we're receiving
427 * on behalf of this socket's zone, so find the pgrp using the
428 * socket's zone ID.
429 */
430 proc = pgfind_zone(pgrp, so->so_zoneid);
431 while (proc != NULL) {
432 mutex_enter(&proc->p_lock);
433 socket_sigproc(proc, event);
434 mutex_exit(&proc->p_lock);
435 proc = proc->p_pglink;
436 }
437 mutex_exit(&pidlock);
438 }
439 }
440
441 #define MIN(a, b) ((a) < (b) ? (a) : (b))
442 /* Copy userdata into a new mblk_t */
443 mblk_t *
socopyinuio(uio_t * uiop,ssize_t iosize,size_t wroff,ssize_t maxblk,size_t tail_len,int * errorp)444 socopyinuio(uio_t *uiop, ssize_t iosize, size_t wroff, ssize_t maxblk,
445 size_t tail_len, int *errorp)
446 {
447 mblk_t *head = NULL, **tail = &head;
448
449 ASSERT(iosize == INFPSZ || iosize > 0);
450
451 if (iosize == INFPSZ || iosize > uiop->uio_resid)
452 iosize = uiop->uio_resid;
453
454 if (maxblk == INFPSZ)
455 maxblk = iosize;
456
457 /* Nothing to do in these cases, so we're done */
458 if (iosize < 0 || maxblk < 0 || (maxblk == 0 && iosize > 0))
459 goto done;
460
461 /*
462 * We will enter the loop below if iosize is 0; it will allocate an
463 * empty message block and call uiomove(9F) which will just return.
464 * We could avoid that with an extra check but would only slow
465 * down the much more likely case where iosize is larger than 0.
466 */
467 do {
468 ssize_t blocksize;
469 mblk_t *mp;
470
471 blocksize = MIN(iosize, maxblk);
472 ASSERT(blocksize >= 0);
473 mp = allocb(wroff + blocksize + tail_len, BPRI_MED);
474 if (mp == NULL) {
475 *errorp = ENOMEM;
476 return (head);
477 }
478 mp->b_rptr += wroff;
479 mp->b_wptr = mp->b_rptr + blocksize;
480
481 *tail = mp;
482 tail = &mp->b_cont;
483
484 /* uiomove(9F) either returns 0 or EFAULT */
485 if ((*errorp = uiomove(mp->b_rptr, (size_t)blocksize,
486 UIO_WRITE, uiop)) != 0) {
487 ASSERT(*errorp != ENOMEM);
488 freemsg(head);
489 return (NULL);
490 }
491
492 iosize -= blocksize;
493 } while (iosize > 0);
494
495 done:
496 *errorp = 0;
497 return (head);
498 }
499
500 mblk_t *
socopyoutuio(mblk_t * mp,struct uio * uiop,ssize_t max_read,int * errorp)501 socopyoutuio(mblk_t *mp, struct uio *uiop, ssize_t max_read, int *errorp)
502 {
503 int error;
504 ptrdiff_t n;
505 mblk_t *nmp;
506
507 ASSERT(mp->b_wptr >= mp->b_rptr);
508
509 /*
510 * max_read is the offset of the oobmark and read can not go pass
511 * the oobmark.
512 */
513 if (max_read == INFPSZ || max_read > uiop->uio_resid)
514 max_read = uiop->uio_resid;
515
516 do {
517 if ((n = MIN(max_read, MBLKL(mp))) != 0) {
518 ASSERT(n > 0);
519
520 error = uiomove(mp->b_rptr, n, UIO_READ, uiop);
521 if (error != 0) {
522 freemsg(mp);
523 *errorp = error;
524 return (NULL);
525 }
526 }
527
528 mp->b_rptr += n;
529 max_read -= n;
530 while (mp != NULL && (mp->b_rptr >= mp->b_wptr)) {
531 /*
532 * get rid of zero length mblks
533 */
534 nmp = mp;
535 mp = mp->b_cont;
536 freeb(nmp);
537 }
538 } while (mp != NULL && max_read > 0);
539
540 *errorp = 0;
541 return (mp);
542 }
543
544 static void
so_prepend_msg(struct sonode * so,mblk_t * mp,mblk_t * last_tail)545 so_prepend_msg(struct sonode *so, mblk_t *mp, mblk_t *last_tail)
546 {
547 ASSERT(last_tail != NULL);
548 mp->b_next = so->so_rcv_q_head;
549 mp->b_prev = last_tail;
550 ASSERT(!(DB_FLAGS(mp) & DBLK_UIOA));
551
552 if (so->so_rcv_q_head == NULL) {
553 ASSERT(so->so_rcv_q_last_head == NULL);
554 so->so_rcv_q_last_head = mp;
555 #ifdef DEBUG
556 } else {
557 ASSERT(!(DB_FLAGS(so->so_rcv_q_head) & DBLK_UIOA));
558 #endif
559 }
560 so->so_rcv_q_head = mp;
561
562 #ifdef DEBUG
563 if (so_debug_length) {
564 mutex_enter(&so->so_lock);
565 ASSERT(so_check_length(so));
566 mutex_exit(&so->so_lock);
567 }
568 #endif
569 }
570
571 /*
572 * Move a mblk chain (mp_head, mp_last_head) to the sonode's rcv queue so it
573 * can be processed by so_dequeue_msg().
574 */
575 void
so_process_new_message(struct sonode * so,mblk_t * mp_head,mblk_t * mp_last_head)576 so_process_new_message(struct sonode *so, mblk_t *mp_head, mblk_t *mp_last_head)
577 {
578 if (so->so_filter_active > 0 &&
579 (mp_head = sof_filter_data_in_proc(so, mp_head,
580 &mp_last_head)) == NULL)
581 return;
582
583 ASSERT(mp_head->b_prev != NULL);
584 if (so->so_rcv_q_head == NULL) {
585 so->so_rcv_q_head = mp_head;
586 so->so_rcv_q_last_head = mp_last_head;
587 ASSERT(so->so_rcv_q_last_head->b_prev != NULL);
588 } else {
589 boolean_t flag_equal = ((DB_FLAGS(mp_head) & DBLK_UIOA) ==
590 (DB_FLAGS(so->so_rcv_q_last_head) & DBLK_UIOA));
591
592 if (mp_head->b_next == NULL &&
593 DB_TYPE(mp_head) == M_DATA &&
594 DB_TYPE(so->so_rcv_q_last_head) == M_DATA && flag_equal) {
595 so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
596 so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
597 mp_head->b_prev = NULL;
598 } else if (flag_equal && (DB_FLAGS(mp_head) & DBLK_UIOA)) {
599 /*
600 * Append to last_head if more than one mblks, and both
601 * mp_head and last_head are I/OAT mblks.
602 */
603 ASSERT(mp_head->b_next != NULL);
604 so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
605 so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
606 mp_head->b_prev = NULL;
607
608 so->so_rcv_q_last_head->b_next = mp_head->b_next;
609 mp_head->b_next = NULL;
610 so->so_rcv_q_last_head = mp_last_head;
611 } else {
612 #ifdef DEBUG
613 {
614 mblk_t *tmp_mblk;
615 tmp_mblk = mp_head;
616 while (tmp_mblk != NULL) {
617 ASSERT(tmp_mblk->b_prev != NULL);
618 tmp_mblk = tmp_mblk->b_next;
619 }
620 }
621 #endif
622 so->so_rcv_q_last_head->b_next = mp_head;
623 so->so_rcv_q_last_head = mp_last_head;
624 }
625 }
626 }
627
628 /*
629 * Check flow control on a given sonode. Must have so_lock held, and
630 * this function will release the hold. Return true if flow control
631 * is cleared.
632 */
633 boolean_t
so_check_flow_control(struct sonode * so)634 so_check_flow_control(struct sonode *so)
635 {
636 ASSERT(MUTEX_HELD(&so->so_lock));
637
638 if (so->so_flowctrld && (so->so_rcv_queued < so->so_rcvlowat &&
639 !(so->so_state & SS_FIL_RCV_FLOWCTRL))) {
640 so->so_flowctrld = B_FALSE;
641 mutex_exit(&so->so_lock);
642 /*
643 * Open up flow control. SCTP does not have any downcalls, and
644 * it will clr flow ctrl in sosctp_recvmsg().
645 */
646 if (so->so_downcalls != NULL &&
647 so->so_downcalls->sd_clr_flowctrl != NULL) {
648 (*so->so_downcalls->sd_clr_flowctrl)
649 (so->so_proto_handle);
650 }
651 /* filters can start injecting data */
652 sof_sonode_notify_filters(so, SOF_EV_INJECT_DATA_IN_OK, 0);
653 return (B_TRUE);
654 } else {
655 mutex_exit(&so->so_lock);
656 return (B_FALSE);
657 }
658 }
659
660 int
so_dequeue_msg(struct sonode * so,mblk_t ** mctlp,struct uio * uiop,rval_t * rvalp,int flags)661 so_dequeue_msg(struct sonode *so, mblk_t **mctlp, struct uio *uiop,
662 rval_t *rvalp, int flags)
663 {
664 mblk_t *mp, *nmp;
665 mblk_t *savemp, *savemptail;
666 mblk_t *new_msg_head;
667 mblk_t *new_msg_last_head;
668 mblk_t *last_tail = NULL;
669 boolean_t partial_read;
670 boolean_t reset_atmark = B_FALSE;
671 int more = 0;
672 int error;
673 ssize_t oobmark;
674 sodirect_t *sodp = so->so_direct;
675
676 partial_read = B_FALSE;
677 *mctlp = NULL;
678 again:
679 mutex_enter(&so->so_lock);
680 again1:
681 #ifdef DEBUG
682 if (so_debug_length) {
683 ASSERT(so_check_length(so));
684 }
685 #endif
686 if (so->so_state & SS_RCVATMARK) {
687 /* Check whether the caller is OK to read past the mark */
688 if (flags & MSG_NOMARK) {
689 mutex_exit(&so->so_lock);
690 return (EWOULDBLOCK);
691 }
692 reset_atmark = B_TRUE;
693 }
694 /*
695 * First move messages from the dump area to processing area
696 */
697 if (sodp != NULL) {
698 if (sodp->sod_enabled) {
699 if (sodp->sod_uioa.uioa_state & UIOA_ALLOC) {
700 /* nothing to uioamove */
701 sodp = NULL;
702 } else if (sodp->sod_uioa.uioa_state & UIOA_INIT) {
703 sodp->sod_uioa.uioa_state &= UIOA_CLR;
704 sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
705 /*
706 * try to uioamove() the data that
707 * has already queued.
708 */
709 sod_uioa_so_init(so, sodp, uiop);
710 }
711 } else {
712 sodp = NULL;
713 }
714 }
715 new_msg_head = so->so_rcv_head;
716 new_msg_last_head = so->so_rcv_last_head;
717 so->so_rcv_head = NULL;
718 so->so_rcv_last_head = NULL;
719 oobmark = so->so_oobmark;
720 /*
721 * We can release the lock as there can only be one reader
722 */
723 mutex_exit(&so->so_lock);
724
725 if (new_msg_head != NULL) {
726 so_process_new_message(so, new_msg_head, new_msg_last_head);
727 }
728 savemp = savemptail = NULL;
729 rvalp->r_vals = 0;
730 error = 0;
731 mp = so->so_rcv_q_head;
732
733 if (mp != NULL &&
734 (so->so_rcv_timer_tid == 0 ||
735 so->so_rcv_queued >= so->so_rcv_thresh)) {
736 partial_read = B_FALSE;
737
738 if (flags & MSG_PEEK) {
739 if ((nmp = dupmsg(mp)) == NULL &&
740 (nmp = copymsg(mp)) == NULL) {
741 size_t size = msgsize(mp);
742
743 error = strwaitbuf(size, BPRI_HI);
744 if (error) {
745 return (error);
746 }
747 goto again;
748 }
749 mp = nmp;
750 } else {
751 ASSERT(mp->b_prev != NULL);
752 last_tail = mp->b_prev;
753 mp->b_prev = NULL;
754 so->so_rcv_q_head = mp->b_next;
755 if (so->so_rcv_q_head == NULL) {
756 so->so_rcv_q_last_head = NULL;
757 }
758 mp->b_next = NULL;
759 }
760
761 ASSERT(mctlp != NULL);
762 /*
763 * First process PROTO or PCPROTO blocks, if any.
764 */
765 if (DB_TYPE(mp) != M_DATA) {
766 *mctlp = mp;
767 savemp = mp;
768 savemptail = mp;
769 ASSERT(DB_TYPE(mp) == M_PROTO ||
770 DB_TYPE(mp) == M_PCPROTO);
771 while (mp->b_cont != NULL &&
772 DB_TYPE(mp->b_cont) != M_DATA) {
773 ASSERT(DB_TYPE(mp->b_cont) == M_PROTO ||
774 DB_TYPE(mp->b_cont) == M_PCPROTO);
775 mp = mp->b_cont;
776 savemptail = mp;
777 }
778 mp = savemptail->b_cont;
779 savemptail->b_cont = NULL;
780 }
781
782 ASSERT(DB_TYPE(mp) == M_DATA);
783 /*
784 * Now process DATA blocks, if any. Note that for sodirect
785 * enabled socket, uio_resid can be 0.
786 */
787 if (uiop->uio_resid >= 0) {
788 ssize_t copied = 0;
789
790 if (sodp != NULL && (DB_FLAGS(mp) & DBLK_UIOA)) {
791 mutex_enter(&so->so_lock);
792 ASSERT(uiop == (uio_t *)&sodp->sod_uioa);
793 copied = sod_uioa_mblk(so, mp);
794 if (copied > 0)
795 partial_read = B_TRUE;
796 mutex_exit(&so->so_lock);
797 /* mark this mblk as processed */
798 mp = NULL;
799 } else {
800 ssize_t oldresid = uiop->uio_resid;
801
802 if (MBLKL(mp) < so_mblk_pull_len) {
803 if (pullupmsg(mp, -1) == 1) {
804 last_tail = mp;
805 }
806 }
807 /*
808 * Can not read beyond the oobmark
809 */
810 mp = socopyoutuio(mp, uiop,
811 oobmark == 0 ? INFPSZ : oobmark, &error);
812 if (error != 0) {
813 freemsg(*mctlp);
814 *mctlp = NULL;
815 more = 0;
816 goto done;
817 }
818 ASSERT(oldresid >= uiop->uio_resid);
819 copied = oldresid - uiop->uio_resid;
820 if (oldresid > uiop->uio_resid)
821 partial_read = B_TRUE;
822 }
823 ASSERT(copied >= 0);
824 if (copied > 0 && !(flags & MSG_PEEK)) {
825 mutex_enter(&so->so_lock);
826 so->so_rcv_queued -= copied;
827 ASSERT(so->so_oobmark >= 0);
828 if (so->so_oobmark > 0) {
829 so->so_oobmark -= copied;
830 ASSERT(so->so_oobmark >= 0);
831 if (so->so_oobmark == 0) {
832 ASSERT(so->so_state &
833 SS_OOBPEND);
834 so->so_oobmark = 0;
835 so->so_state |= SS_RCVATMARK;
836 }
837 }
838 /*
839 * so_check_flow_control() will drop
840 * so->so_lock.
841 */
842 rvalp->r_val2 = so_check_flow_control(so);
843 }
844 }
845 if (mp != NULL) { /* more data blocks in msg */
846 more |= MOREDATA;
847 if ((flags & (MSG_PEEK|MSG_TRUNC))) {
848 if (flags & MSG_PEEK) {
849 freemsg(mp);
850 } else {
851 unsigned int msize = msgdsize(mp);
852
853 freemsg(mp);
854 mutex_enter(&so->so_lock);
855 so->so_rcv_queued -= msize;
856 /*
857 * so_check_flow_control() will drop
858 * so->so_lock.
859 */
860 rvalp->r_val2 =
861 so_check_flow_control(so);
862 }
863 } else if (partial_read && !somsghasdata(mp)) {
864 /*
865 * Avoid queuing a zero-length tail part of
866 * a message. partial_read == 1 indicates that
867 * we read some of the message.
868 */
869 freemsg(mp);
870 more &= ~MOREDATA;
871 } else {
872 if (savemp != NULL &&
873 (flags & MSG_DUPCTRL)) {
874 mblk_t *nmp;
875 /*
876 * There should only be non data mblks
877 */
878 ASSERT(DB_TYPE(savemp) != M_DATA &&
879 DB_TYPE(savemptail) != M_DATA);
880 try_again:
881 if ((nmp = dupmsg(savemp)) == NULL &&
882 (nmp = copymsg(savemp)) == NULL) {
883
884 size_t size = msgsize(savemp);
885
886 error = strwaitbuf(size,
887 BPRI_HI);
888 if (error != 0) {
889 /*
890 * In case we
891 * cannot copy
892 * control data
893 * free the remaining
894 * data.
895 */
896 freemsg(mp);
897 goto done;
898 }
899 goto try_again;
900 }
901
902 ASSERT(nmp != NULL);
903 ASSERT(DB_TYPE(nmp) != M_DATA);
904 savemptail->b_cont = mp;
905 *mctlp = nmp;
906 mp = savemp;
907 }
908 /*
909 * putback mp
910 */
911 so_prepend_msg(so, mp, last_tail);
912 }
913 }
914
915 /* fast check so_rcv_head if there is more data */
916 if (partial_read && !(so->so_state & SS_RCVATMARK) &&
917 *mctlp == NULL && uiop->uio_resid > 0 &&
918 !(flags & MSG_PEEK) && so->so_rcv_head != NULL) {
919 goto again;
920 }
921 } else if (!partial_read) {
922 mutex_enter(&so->so_lock);
923 if (so->so_error != 0) {
924 error = sogeterr(so, !(flags & MSG_PEEK));
925 mutex_exit(&so->so_lock);
926 return (error);
927 }
928
929 /* See if new data has arrived in the meantime */
930 if (so->so_rcv_head != NULL)
931 goto again1;
932
933 /*
934 * No pending data. Return right away for nonblocking
935 * socket, otherwise sleep waiting for data.
936 */
937 if (!(so->so_state & SS_CANTRCVMORE) && uiop->uio_resid > 0) {
938 if ((uiop->uio_fmode & (FNDELAY|FNONBLOCK)) ||
939 (flags & MSG_DONTWAIT)) {
940 error = EWOULDBLOCK;
941 } else {
942 if (so->so_state & (SS_CLOSING |
943 SS_FALLBACK_PENDING)) {
944 mutex_exit(&so->so_lock);
945 error = EINTR;
946 goto done;
947 }
948
949 so->so_rcv_wakeup = B_TRUE;
950 so->so_rcv_wanted = uiop->uio_resid;
951 if (so->so_rcvtimeo == 0) {
952 /*
953 * Zero means disable timeout.
954 */
955 error = cv_wait_sig(&so->so_rcv_cv,
956 &so->so_lock);
957 } else {
958 error = cv_reltimedwait_sig(
959 &so->so_rcv_cv, &so->so_lock,
960 so->so_rcvtimeo, TR_CLOCK_TICK);
961 }
962 so->so_rcv_wakeup = B_FALSE;
963 so->so_rcv_wanted = 0;
964
965 if (error == 0) {
966 error = EINTR;
967 } else if (error == -1) {
968 error = EAGAIN;
969 } else {
970 goto again1;
971 }
972 }
973 }
974 mutex_exit(&so->so_lock);
975 }
976 if (reset_atmark && partial_read && !(flags & MSG_PEEK)) {
977 /*
978 * We are passed the mark, update state
979 * 4.3BSD and 4.4BSD clears the mark when peeking across it.
980 * The draft Posix socket spec states that the mark should
981 * not be cleared when peeking. We follow the latter.
982 */
983 mutex_enter(&so->so_lock);
984 ASSERT(so_verify_oobstate(so));
985 so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK);
986 freemsg(so->so_oobmsg);
987 so->so_oobmsg = NULL;
988 ASSERT(so_verify_oobstate(so));
989 mutex_exit(&so->so_lock);
990 }
991 ASSERT(so->so_rcv_wakeup == B_FALSE);
992 done:
993 if (sodp != NULL) {
994 mutex_enter(&so->so_lock);
995 if (sodp->sod_enabled &&
996 (sodp->sod_uioa.uioa_state & UIOA_ENABLED)) {
997 SOD_UIOAFINI(sodp);
998 if (sodp->sod_uioa.uioa_mbytes > 0) {
999 ASSERT(so->so_rcv_q_head != NULL ||
1000 so->so_rcv_head != NULL);
1001 so->so_rcv_queued -= sod_uioa_mblk(so, NULL);
1002 if (error == EWOULDBLOCK)
1003 error = 0;
1004 }
1005 }
1006 mutex_exit(&so->so_lock);
1007 }
1008 #ifdef DEBUG
1009 if (so_debug_length) {
1010 mutex_enter(&so->so_lock);
1011 ASSERT(so_check_length(so));
1012 mutex_exit(&so->so_lock);
1013 }
1014 #endif
1015 rvalp->r_val1 = more;
1016 ASSERT(MUTEX_NOT_HELD(&so->so_lock));
1017 return (error);
1018 }
1019
1020 /*
1021 * Enqueue data from the protocol on the socket's rcv queue.
1022 *
1023 * We try to hook new M_DATA mblks onto an existing chain, however,
1024 * that cannot be done if the existing chain has already been
1025 * processed by I/OAT. Non-M_DATA mblks are just linked together via
1026 * b_next. In all cases the b_prev of the enqueued mblk is set to
1027 * point to the last mblk in its b_cont chain.
1028 */
1029 void
so_enqueue_msg(struct sonode * so,mblk_t * mp,size_t msg_size)1030 so_enqueue_msg(struct sonode *so, mblk_t *mp, size_t msg_size)
1031 {
1032 ASSERT(MUTEX_HELD(&so->so_lock));
1033
1034 #ifdef DEBUG
1035 if (so_debug_length) {
1036 ASSERT(so_check_length(so));
1037 }
1038 #endif
1039 so->so_rcv_queued += msg_size;
1040
1041 if (so->so_rcv_head == NULL) {
1042 ASSERT(so->so_rcv_last_head == NULL);
1043 so->so_rcv_head = mp;
1044 so->so_rcv_last_head = mp;
1045 } else if ((DB_TYPE(mp) == M_DATA &&
1046 DB_TYPE(so->so_rcv_last_head) == M_DATA) &&
1047 ((DB_FLAGS(mp) & DBLK_UIOA) ==
1048 (DB_FLAGS(so->so_rcv_last_head) & DBLK_UIOA))) {
1049 /* Added to the end */
1050 ASSERT(so->so_rcv_last_head != NULL);
1051 ASSERT(so->so_rcv_last_head->b_prev != NULL);
1052 so->so_rcv_last_head->b_prev->b_cont = mp;
1053 } else {
1054 /* Start a new end */
1055 so->so_rcv_last_head->b_next = mp;
1056 so->so_rcv_last_head = mp;
1057 }
1058 while (mp->b_cont != NULL)
1059 mp = mp->b_cont;
1060
1061 so->so_rcv_last_head->b_prev = mp;
1062 #ifdef DEBUG
1063 if (so_debug_length) {
1064 ASSERT(so_check_length(so));
1065 }
1066 #endif
1067 }
1068
1069 /*
1070 * Return B_TRUE if there is data in the message, B_FALSE otherwise.
1071 */
1072 boolean_t
somsghasdata(mblk_t * mp)1073 somsghasdata(mblk_t *mp)
1074 {
1075 for (; mp; mp = mp->b_cont)
1076 if (mp->b_datap->db_type == M_DATA) {
1077 ASSERT(mp->b_wptr >= mp->b_rptr);
1078 if (mp->b_wptr > mp->b_rptr)
1079 return (B_TRUE);
1080 }
1081 return (B_FALSE);
1082 }
1083
1084 /*
1085 * Flush the read side of sockfs.
1086 *
1087 * The caller must be sure that a reader is not already active when the
1088 * buffer is being flushed.
1089 */
1090 void
so_rcv_flush(struct sonode * so)1091 so_rcv_flush(struct sonode *so)
1092 {
1093 mblk_t *mp;
1094
1095 ASSERT(MUTEX_HELD(&so->so_lock));
1096
1097 if (so->so_oobmsg != NULL) {
1098 freemsg(so->so_oobmsg);
1099 so->so_oobmsg = NULL;
1100 so->so_oobmark = 0;
1101 so->so_state &=
1102 ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA|SS_RCVATMARK);
1103 }
1104
1105 /*
1106 * Free messages sitting in the recv queues
1107 */
1108 while (so->so_rcv_q_head != NULL) {
1109 mp = so->so_rcv_q_head;
1110 so->so_rcv_q_head = mp->b_next;
1111 mp->b_next = mp->b_prev = NULL;
1112 freemsg(mp);
1113 }
1114 while (so->so_rcv_head != NULL) {
1115 mp = so->so_rcv_head;
1116 so->so_rcv_head = mp->b_next;
1117 mp->b_next = mp->b_prev = NULL;
1118 freemsg(mp);
1119 }
1120 so->so_rcv_queued = 0;
1121 so->so_rcv_q_head = NULL;
1122 so->so_rcv_q_last_head = NULL;
1123 so->so_rcv_head = NULL;
1124 so->so_rcv_last_head = NULL;
1125 }
1126
1127 /*
1128 * Handle recv* calls that set MSG_OOB or MSG_OOB together with MSG_PEEK.
1129 */
1130 int
sorecvoob(struct sonode * so,struct nmsghdr * msg,struct uio * uiop,int flags,boolean_t oob_inline)1131 sorecvoob(struct sonode *so, struct nmsghdr *msg, struct uio *uiop, int flags,
1132 boolean_t oob_inline)
1133 {
1134 mblk_t *mp, *nmp;
1135 int error;
1136
1137 dprintso(so, 1, ("sorecvoob(%p, %p, 0x%x)\n", (void *)so, (void *)msg,
1138 flags));
1139
1140 if (msg != NULL) {
1141 /*
1142 * There is never any oob data with addresses or control since
1143 * the T_EXDATA_IND does not carry any options.
1144 */
1145 msg->msg_controllen = 0;
1146 msg->msg_namelen = 0;
1147 msg->msg_flags = 0;
1148 }
1149
1150 mutex_enter(&so->so_lock);
1151 ASSERT(so_verify_oobstate(so));
1152 if (oob_inline ||
1153 (so->so_state & (SS_OOBPEND|SS_HADOOBDATA)) != SS_OOBPEND) {
1154 dprintso(so, 1, ("sorecvoob: inline or data consumed\n"));
1155 mutex_exit(&so->so_lock);
1156 return (EINVAL);
1157 }
1158 if (!(so->so_state & SS_HAVEOOBDATA)) {
1159 dprintso(so, 1, ("sorecvoob: no data yet\n"));
1160 mutex_exit(&so->so_lock);
1161 return (EWOULDBLOCK);
1162 }
1163 ASSERT(so->so_oobmsg != NULL);
1164 mp = so->so_oobmsg;
1165 if (flags & MSG_PEEK) {
1166 /*
1167 * Since recv* can not return ENOBUFS we can not use dupmsg.
1168 * Instead we revert to the consolidation private
1169 * allocb_wait plus bcopy.
1170 */
1171 mblk_t *mp1;
1172
1173 mp1 = allocb_wait(msgdsize(mp), BPRI_MED, STR_NOSIG, NULL);
1174 ASSERT(mp1);
1175
1176 while (mp != NULL) {
1177 ssize_t size;
1178
1179 size = MBLKL(mp);
1180 bcopy(mp->b_rptr, mp1->b_wptr, size);
1181 mp1->b_wptr += size;
1182 ASSERT(mp1->b_wptr <= mp1->b_datap->db_lim);
1183 mp = mp->b_cont;
1184 }
1185 mp = mp1;
1186 } else {
1187 /*
1188 * Update the state indicating that the data has been consumed.
1189 * Keep SS_OOBPEND set until data is consumed past the mark.
1190 */
1191 so->so_oobmsg = NULL;
1192 so->so_state ^= SS_HAVEOOBDATA|SS_HADOOBDATA;
1193 }
1194 ASSERT(so_verify_oobstate(so));
1195 mutex_exit(&so->so_lock);
1196
1197 error = 0;
1198 nmp = mp;
1199 while (nmp != NULL && uiop->uio_resid > 0) {
1200 ssize_t n = MBLKL(nmp);
1201
1202 n = MIN(n, uiop->uio_resid);
1203 if (n > 0)
1204 error = uiomove(nmp->b_rptr, n,
1205 UIO_READ, uiop);
1206 if (error)
1207 break;
1208 nmp = nmp->b_cont;
1209 }
1210 ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
1211 freemsg(mp);
1212 return (error);
1213 }
1214
1215 /*
1216 * Allocate and initializ sonode
1217 */
1218 /* ARGSUSED */
1219 struct sonode *
socket_sonode_create(struct sockparams * sp,int family,int type,int protocol,int version,int sflags,int * errorp,struct cred * cr)1220 socket_sonode_create(struct sockparams *sp, int family, int type,
1221 int protocol, int version, int sflags, int *errorp, struct cred *cr)
1222 {
1223 sonode_t *so;
1224 int kmflags;
1225
1226 /*
1227 * Choose the right set of sonodeops based on the upcall and
1228 * down call version that the protocol has provided
1229 */
1230 if (SOCK_UC_VERSION != sp->sp_smod_info->smod_uc_version ||
1231 SOCK_DC_VERSION != sp->sp_smod_info->smod_dc_version) {
1232 /*
1233 * mismatch
1234 */
1235 #ifdef DEBUG
1236 cmn_err(CE_CONT, "protocol and socket module version mismatch");
1237 #endif
1238 *errorp = EINVAL;
1239 return (NULL);
1240 }
1241
1242 kmflags = (sflags & SOCKET_NOSLEEP) ? KM_NOSLEEP : KM_SLEEP;
1243
1244 so = kmem_cache_alloc(socket_cache, kmflags);
1245 if (so == NULL) {
1246 *errorp = ENOMEM;
1247 return (NULL);
1248 }
1249
1250 sonode_init(so, sp, family, type, protocol, &so_sonodeops);
1251
1252 if (version == SOV_DEFAULT)
1253 version = so_default_version;
1254
1255 so->so_version = (short)version;
1256
1257 /*
1258 * set the default values to be INFPSZ
1259 * if a protocol desires it can change the value later
1260 */
1261 so->so_proto_props.sopp_rxhiwat = SOCKET_RECVHIWATER;
1262 so->so_proto_props.sopp_rxlowat = SOCKET_RECVLOWATER;
1263 so->so_proto_props.sopp_maxpsz = INFPSZ;
1264 so->so_proto_props.sopp_maxblk = INFPSZ;
1265
1266 return (so);
1267 }
1268
1269 int
socket_init_common(struct sonode * so,struct sonode * pso,int flags,cred_t * cr)1270 socket_init_common(struct sonode *so, struct sonode *pso, int flags, cred_t *cr)
1271 {
1272 int error = 0;
1273
1274 if (pso != NULL) {
1275 /*
1276 * We have a passive open, so inherit basic state from
1277 * the parent (listener).
1278 *
1279 * No need to grab the new sonode's lock, since there is no
1280 * one that can have a reference to it.
1281 */
1282 mutex_enter(&pso->so_lock);
1283
1284 so->so_state |= SS_ISCONNECTED | (pso->so_state & SS_ASYNC);
1285 so->so_pgrp = pso->so_pgrp;
1286 so->so_rcvtimeo = pso->so_rcvtimeo;
1287 so->so_sndtimeo = pso->so_sndtimeo;
1288 so->so_xpg_rcvbuf = pso->so_xpg_rcvbuf;
1289 /*
1290 * Make note of the socket level options. TCP and IP level
1291 * options are already inherited. We could do all this after
1292 * accept is successful but doing it here simplifies code and
1293 * no harm done for error case.
1294 */
1295 so->so_options = pso->so_options & (SO_DEBUG|SO_REUSEADDR|
1296 SO_KEEPALIVE|SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK|
1297 SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER);
1298 so->so_proto_props = pso->so_proto_props;
1299 so->so_mode = pso->so_mode;
1300 so->so_pollev = pso->so_pollev & SO_POLLEV_ALWAYS;
1301
1302 mutex_exit(&pso->so_lock);
1303
1304 /*
1305 * If the parent has any filters, try to inherit them.
1306 */
1307 if (pso->so_filter_active > 0 &&
1308 (error = sof_sonode_inherit_filters(so, pso)) != 0)
1309 return (error);
1310
1311 } else {
1312 struct sockparams *sp = so->so_sockparams;
1313 sock_upcalls_t *upcalls_to_use;
1314
1315 /*
1316 * Attach automatic filters, if there are any.
1317 */
1318 if (!list_is_empty(&sp->sp_auto_filters) &&
1319 (error = sof_sonode_autoattach_filters(so, cr)) != 0)
1320 return (error);
1321
1322 /* OK to attach filters */
1323 so->so_state |= SS_FILOP_OK;
1324
1325 /*
1326 * Based on the version number select the right upcalls to
1327 * pass down. Currently we only have one version so choose
1328 * default
1329 */
1330 upcalls_to_use = &so_upcalls;
1331
1332 /* active open, so create a lower handle */
1333 so->so_proto_handle =
1334 sp->sp_smod_info->smod_proto_create_func(so->so_family,
1335 so->so_type, so->so_protocol, &so->so_downcalls,
1336 &so->so_mode, &error, flags, cr);
1337
1338 if (so->so_proto_handle == NULL) {
1339 ASSERT(error != 0);
1340 /*
1341 * To be safe; if a lower handle cannot be created, and
1342 * the proto does not give a reason why, assume there
1343 * was a lack of memory.
1344 */
1345 return ((error == 0) ? ENOMEM : error);
1346 }
1347 ASSERT(so->so_downcalls != NULL);
1348 ASSERT(so->so_downcalls->sd_send != NULL ||
1349 so->so_downcalls->sd_send_uio != NULL);
1350 if (so->so_downcalls->sd_recv_uio != NULL) {
1351 ASSERT(so->so_downcalls->sd_poll != NULL);
1352 so->so_pollev |= SO_POLLEV_ALWAYS;
1353 }
1354
1355 (*so->so_downcalls->sd_activate)(so->so_proto_handle,
1356 (sock_upper_handle_t)so, upcalls_to_use, 0, cr);
1357
1358 /* Wildcard */
1359
1360 /*
1361 * FIXME No need for this, the protocol can deal with it in
1362 * sd_create(). Should update ICMP.
1363 */
1364 if (so->so_protocol != so->so_sockparams->sp_protocol) {
1365 int protocol = so->so_protocol;
1366 int error;
1367 /*
1368 * Issue SO_PROTOTYPE setsockopt.
1369 */
1370 error = socket_setsockopt(so, SOL_SOCKET, SO_PROTOTYPE,
1371 &protocol, (t_uscalar_t)sizeof (protocol), cr);
1372 if (error) {
1373 (void) (*so->so_downcalls->sd_close)
1374 (so->so_proto_handle, 0, cr);
1375
1376 mutex_enter(&so->so_lock);
1377 so_rcv_flush(so);
1378 mutex_exit(&so->so_lock);
1379 /*
1380 * Setsockopt often fails with ENOPROTOOPT but
1381 * socket() should fail with
1382 * EPROTONOSUPPORT/EPROTOTYPE.
1383 */
1384 return (EPROTONOSUPPORT);
1385 }
1386 }
1387 }
1388
1389 if (uioasync.enabled)
1390 sod_sock_init(so);
1391
1392 /* put an extra reference on the socket for the protocol */
1393 VN_HOLD(SOTOV(so));
1394
1395 return (0);
1396 }
1397
1398 /*
1399 * int socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1400 * struct cred *cr, int32_t *rvalp)
1401 *
1402 * Handle ioctls that manipulate basic socket state; non-blocking,
1403 * async, etc.
1404 *
1405 * Returns:
1406 * < 0 - ioctl was not handle
1407 * >= 0 - ioctl was handled, if > 0, then it is an errno
1408 *
1409 * Notes:
1410 * Assumes the standard receive buffer is used to obtain info for
1411 * NREAD.
1412 */
1413 /* ARGSUSED */
1414 int
socket_ioctl_common(struct sonode * so,int cmd,intptr_t arg,int mode,struct cred * cr,int32_t * rvalp)1415 socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1416 struct cred *cr, int32_t *rvalp)
1417 {
1418 switch (cmd) {
1419 case SIOCSQPTR:
1420 /*
1421 * SIOCSQPTR is valid only when helper stream is created
1422 * by the protocol.
1423 */
1424
1425 return (EOPNOTSUPP);
1426 case FIONBIO: {
1427 int32_t value;
1428
1429 if (so_copyin((void *)arg, &value, sizeof (int32_t),
1430 (mode & (int)FKIOCTL)))
1431 return (EFAULT);
1432
1433 mutex_enter(&so->so_lock);
1434 if (value) {
1435 so->so_state |= SS_NDELAY;
1436 } else {
1437 so->so_state &= ~SS_NDELAY;
1438 }
1439 mutex_exit(&so->so_lock);
1440 return (0);
1441 }
1442 case FIOASYNC: {
1443 int32_t value;
1444
1445 if (so_copyin((void *)arg, &value, sizeof (int32_t),
1446 (mode & (int)FKIOCTL)))
1447 return (EFAULT);
1448
1449 mutex_enter(&so->so_lock);
1450
1451 if (value) {
1452 /* Turn on SIGIO */
1453 so->so_state |= SS_ASYNC;
1454 } else {
1455 /* Turn off SIGIO */
1456 so->so_state &= ~SS_ASYNC;
1457 }
1458 mutex_exit(&so->so_lock);
1459
1460 return (0);
1461 }
1462
1463 case SIOCSPGRP:
1464 case FIOSETOWN: {
1465 int error;
1466 pid_t pid;
1467
1468 if (so_copyin((void *)arg, &pid, sizeof (pid_t),
1469 (mode & (int)FKIOCTL)))
1470 return (EFAULT);
1471
1472 mutex_enter(&so->so_lock);
1473 error = (pid != so->so_pgrp) ? socket_chgpgrp(so, pid) : 0;
1474 mutex_exit(&so->so_lock);
1475 return (error);
1476 }
1477 case SIOCGPGRP:
1478 case FIOGETOWN:
1479 if (so_copyout(&so->so_pgrp, (void *)arg,
1480 sizeof (pid_t), (mode & (int)FKIOCTL)))
1481 return (EFAULT);
1482
1483 return (0);
1484 case SIOCATMARK: {
1485 int retval;
1486
1487 /*
1488 * Only protocols that support urgent data can handle ATMARK.
1489 */
1490 if ((so->so_mode & SM_EXDATA) == 0)
1491 return (EINVAL);
1492
1493 /*
1494 * If the protocol is maintaining its own buffer, then the
1495 * request must be passed down.
1496 */
1497 if (so->so_downcalls->sd_recv_uio != NULL)
1498 return (-1);
1499
1500 retval = (so->so_state & SS_RCVATMARK) != 0;
1501
1502 if (so_copyout(&retval, (void *)arg, sizeof (int),
1503 (mode & (int)FKIOCTL))) {
1504 return (EFAULT);
1505 }
1506 return (0);
1507 }
1508
1509 case FIONREAD: {
1510 int retval;
1511
1512 /*
1513 * If the protocol is maintaining its own buffer, then the
1514 * request must be passed down.
1515 */
1516 if (so->so_downcalls->sd_recv_uio != NULL)
1517 return (-1);
1518
1519 retval = MIN(so->so_rcv_queued, INT_MAX);
1520
1521 if (so_copyout(&retval, (void *)arg,
1522 sizeof (retval), (mode & (int)FKIOCTL))) {
1523 return (EFAULT);
1524 }
1525 return (0);
1526 }
1527
1528 case _I_GETPEERCRED: {
1529 int error = 0;
1530
1531 if ((mode & FKIOCTL) == 0)
1532 return (EINVAL);
1533
1534 mutex_enter(&so->so_lock);
1535 if ((so->so_mode & SM_CONNREQUIRED) == 0) {
1536 error = ENOTSUP;
1537 } else if ((so->so_state & SS_ISCONNECTED) == 0) {
1538 error = ENOTCONN;
1539 } else if (so->so_peercred != NULL) {
1540 k_peercred_t *kp = (k_peercred_t *)arg;
1541 kp->pc_cr = so->so_peercred;
1542 kp->pc_cpid = so->so_cpid;
1543 crhold(so->so_peercred);
1544 } else {
1545 error = EINVAL;
1546 }
1547 mutex_exit(&so->so_lock);
1548 return (error);
1549 }
1550 default:
1551 return (-1);
1552 }
1553 }
1554
1555 /*
1556 * Handle the I_NREAD STREAM ioctl.
1557 */
1558 static int
so_strioc_nread(struct sonode * so,intptr_t arg,int mode,int32_t * rvalp)1559 so_strioc_nread(struct sonode *so, intptr_t arg, int mode, int32_t *rvalp)
1560 {
1561 size_t size = 0;
1562 int retval;
1563 int count = 0;
1564 mblk_t *mp;
1565 clock_t wakeup = drv_usectohz(10);
1566
1567 if (so->so_downcalls == NULL ||
1568 so->so_downcalls->sd_recv_uio != NULL)
1569 return (EINVAL);
1570
1571 mutex_enter(&so->so_lock);
1572 /* Wait for reader to get out of the way. */
1573 while (so->so_flag & SOREADLOCKED) {
1574 /*
1575 * If reader is waiting for data, then there should be nothing
1576 * on the rcv queue.
1577 */
1578 if (so->so_rcv_wakeup)
1579 goto out;
1580
1581 /* Do a timed sleep, in case the reader goes to sleep. */
1582 (void) cv_reltimedwait(&so->so_read_cv, &so->so_lock, wakeup,
1583 TR_CLOCK_TICK);
1584 }
1585
1586 /*
1587 * Since we are holding so_lock no new reader will come in, and the
1588 * protocol will not be able to enqueue data. So it's safe to walk
1589 * both rcv queues.
1590 */
1591 mp = so->so_rcv_q_head;
1592 if (mp != NULL) {
1593 size = msgdsize(so->so_rcv_q_head);
1594 for (; mp != NULL; mp = mp->b_next)
1595 count++;
1596 } else {
1597 /*
1598 * In case the processing list was empty, get the size of the
1599 * next msg in line.
1600 */
1601 size = msgdsize(so->so_rcv_head);
1602 }
1603
1604 for (mp = so->so_rcv_head; mp != NULL; mp = mp->b_next)
1605 count++;
1606 out:
1607 mutex_exit(&so->so_lock);
1608
1609 /*
1610 * Drop down from size_t to the "int" required by the
1611 * interface. Cap at INT_MAX.
1612 */
1613 retval = MIN(size, INT_MAX);
1614 if (so_copyout(&retval, (void *)arg, sizeof (retval),
1615 (mode & (int)FKIOCTL))) {
1616 return (EFAULT);
1617 } else {
1618 *rvalp = count;
1619 return (0);
1620 }
1621 }
1622
1623 /*
1624 * Process STREAM ioctls.
1625 *
1626 * Returns:
1627 * < 0 - ioctl was not handle
1628 * >= 0 - ioctl was handled, if > 0, then it is an errno
1629 */
1630 int
socket_strioc_common(struct sonode * so,int cmd,intptr_t arg,int mode,struct cred * cr,int32_t * rvalp)1631 socket_strioc_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1632 struct cred *cr, int32_t *rvalp)
1633 {
1634 int retval;
1635
1636 /* Only STREAM iotcls are handled here */
1637 if ((cmd & 0xffffff00U) != STR)
1638 return (-1);
1639
1640 switch (cmd) {
1641 case I_CANPUT:
1642 /*
1643 * We return an error for I_CANPUT so that isastream(3C) will
1644 * not report the socket as being a STREAM.
1645 */
1646 return (EOPNOTSUPP);
1647 case I_NREAD:
1648 /* Avoid doing a fallback for I_NREAD. */
1649 return (so_strioc_nread(so, arg, mode, rvalp));
1650 case I_LOOK:
1651 /* Avoid doing a fallback for I_LOOK. */
1652 if (so_copyout("sockmod", (void *)arg, strlen("sockmod") + 1,
1653 (mode & (int)FKIOCTL))) {
1654 return (EFAULT);
1655 }
1656 return (0);
1657 default:
1658 break;
1659 }
1660
1661 /*
1662 * Try to fall back to TPI, and if successful, reissue the ioctl.
1663 */
1664 if ((retval = so_tpi_fallback(so, cr)) == 0) {
1665 /* Reissue the ioctl */
1666 ASSERT(so->so_rcv_q_head == NULL);
1667 return (SOP_IOCTL(so, cmd, arg, mode, cr, rvalp));
1668 } else {
1669 return (retval);
1670 }
1671 }
1672
1673 /*
1674 * This is called for all socket types to verify that the buffer size is large
1675 * enough for the option, and if we can, handle the request as well. Most
1676 * options will be forwarded to the protocol.
1677 */
1678 int
socket_getopt_common(struct sonode * so,int level,int option_name,void * optval,socklen_t * optlenp,int flags)1679 socket_getopt_common(struct sonode *so, int level, int option_name,
1680 void *optval, socklen_t *optlenp, int flags)
1681 {
1682 if (level != SOL_SOCKET)
1683 return (-1);
1684
1685 switch (option_name) {
1686 case SO_ERROR:
1687 case SO_DOMAIN:
1688 case SO_TYPE:
1689 case SO_ACCEPTCONN: {
1690 int32_t value;
1691 socklen_t optlen = *optlenp;
1692
1693 if (optlen < (t_uscalar_t)sizeof (int32_t)) {
1694 return (EINVAL);
1695 }
1696
1697 switch (option_name) {
1698 case SO_ERROR:
1699 mutex_enter(&so->so_lock);
1700 value = sogeterr(so, B_TRUE);
1701 mutex_exit(&so->so_lock);
1702 break;
1703 /*
1704 * While SO_DOMAIN and SO_TYPE are here, SO_PROTOCOL (aka
1705 * SO_PROTOYPE) is not implemented in the common layer because
1706 * some socket modules support setting the protocol and
1707 * therefore we must ask the module directly.
1708 */
1709 case SO_DOMAIN:
1710 value = so->so_family;
1711 break;
1712 case SO_TYPE:
1713 value = so->so_type;
1714 break;
1715 case SO_ACCEPTCONN:
1716 if (so->so_state & SS_ACCEPTCONN)
1717 value = SO_ACCEPTCONN;
1718 else
1719 value = 0;
1720 break;
1721 }
1722
1723 bcopy(&value, optval, sizeof (value));
1724 *optlenp = sizeof (value);
1725
1726 return (0);
1727 }
1728 case SO_SNDTIMEO:
1729 case SO_RCVTIMEO: {
1730 clock_t value;
1731 socklen_t optlen = *optlenp;
1732
1733 if (get_udatamodel() == DATAMODEL_NONE ||
1734 get_udatamodel() == DATAMODEL_NATIVE) {
1735 if (optlen < sizeof (struct timeval))
1736 return (EINVAL);
1737 } else {
1738 if (optlen < sizeof (struct timeval32))
1739 return (EINVAL);
1740 }
1741 if (option_name == SO_RCVTIMEO)
1742 value = drv_hztousec(so->so_rcvtimeo);
1743 else
1744 value = drv_hztousec(so->so_sndtimeo);
1745
1746 if (get_udatamodel() == DATAMODEL_NONE ||
1747 get_udatamodel() == DATAMODEL_NATIVE) {
1748 ((struct timeval *)(optval))->tv_sec =
1749 value / (1000 * 1000);
1750 ((struct timeval *)(optval))->tv_usec =
1751 value % (1000 * 1000);
1752 *optlenp = sizeof (struct timeval);
1753 } else {
1754 ((struct timeval32 *)(optval))->tv_sec =
1755 value / (1000 * 1000);
1756 ((struct timeval32 *)(optval))->tv_usec =
1757 value % (1000 * 1000);
1758 *optlenp = sizeof (struct timeval32);
1759 }
1760 return (0);
1761 }
1762 case SO_DEBUG:
1763 case SO_REUSEADDR:
1764 case SO_KEEPALIVE:
1765 case SO_DONTROUTE:
1766 case SO_BROADCAST:
1767 case SO_USELOOPBACK:
1768 case SO_OOBINLINE:
1769 case SO_SNDBUF:
1770 #ifdef notyet
1771 case SO_SNDLOWAT:
1772 case SO_RCVLOWAT:
1773 #endif /* notyet */
1774 case SO_DGRAM_ERRIND: {
1775 socklen_t optlen = *optlenp;
1776
1777 if (optlen < (t_uscalar_t)sizeof (int32_t))
1778 return (EINVAL);
1779 break;
1780 }
1781 case SO_RCVBUF: {
1782 socklen_t optlen = *optlenp;
1783
1784 if (optlen < (t_uscalar_t)sizeof (int32_t))
1785 return (EINVAL);
1786
1787 if ((flags & _SOGETSOCKOPT_XPG4_2) && so->so_xpg_rcvbuf != 0) {
1788 /*
1789 * XXX If SO_RCVBUF has been set and this is an
1790 * XPG 4.2 application then do not ask the transport
1791 * since the transport might adjust the value and not
1792 * return exactly what was set by the application.
1793 * For non-XPG 4.2 application we return the value
1794 * that the transport is actually using.
1795 */
1796 *(int32_t *)optval = so->so_xpg_rcvbuf;
1797 *optlenp = sizeof (so->so_xpg_rcvbuf);
1798 return (0);
1799 }
1800 /*
1801 * If the option has not been set then get a default
1802 * value from the transport.
1803 */
1804 break;
1805 }
1806 case SO_LINGER: {
1807 socklen_t optlen = *optlenp;
1808
1809 if (optlen < (t_uscalar_t)sizeof (struct linger))
1810 return (EINVAL);
1811 break;
1812 }
1813 case SO_SND_BUFINFO: {
1814 socklen_t optlen = *optlenp;
1815
1816 if (optlen < (t_uscalar_t)sizeof (struct so_snd_bufinfo))
1817 return (EINVAL);
1818 ((struct so_snd_bufinfo *)(optval))->sbi_wroff =
1819 (so->so_proto_props).sopp_wroff;
1820 ((struct so_snd_bufinfo *)(optval))->sbi_maxblk =
1821 (so->so_proto_props).sopp_maxblk;
1822 ((struct so_snd_bufinfo *)(optval))->sbi_maxpsz =
1823 (so->so_proto_props).sopp_maxpsz;
1824 ((struct so_snd_bufinfo *)(optval))->sbi_tail =
1825 (so->so_proto_props).sopp_tail;
1826 *optlenp = sizeof (struct so_snd_bufinfo);
1827 return (0);
1828 }
1829 case SO_SND_COPYAVOID: {
1830 sof_instance_t *inst;
1831
1832 /*
1833 * Avoid zero-copy if there is a filter with a data_out
1834 * callback. We could let the operation succeed, but then
1835 * the filter would have to copy the data anyway.
1836 */
1837 for (inst = so->so_filter_top; inst != NULL;
1838 inst = inst->sofi_next) {
1839 if (SOF_INTERESTED(inst, data_out))
1840 return (EOPNOTSUPP);
1841 }
1842 break;
1843 }
1844
1845 default:
1846 break;
1847 }
1848
1849 /* Unknown Option */
1850 return (-1);
1851 }
1852
1853 void
socket_sonode_destroy(struct sonode * so)1854 socket_sonode_destroy(struct sonode *so)
1855 {
1856 sonode_fini(so);
1857 kmem_cache_free(socket_cache, so);
1858 }
1859
1860 int
so_zcopy_wait(struct sonode * so)1861 so_zcopy_wait(struct sonode *so)
1862 {
1863 int error = 0;
1864
1865 mutex_enter(&so->so_lock);
1866 while (!(so->so_copyflag & STZCNOTIFY)) {
1867 if (so->so_state & SS_CLOSING) {
1868 mutex_exit(&so->so_lock);
1869 return (EINTR);
1870 }
1871 if (cv_wait_sig(&so->so_copy_cv, &so->so_lock) == 0) {
1872 error = EINTR;
1873 break;
1874 }
1875 }
1876 so->so_copyflag &= ~STZCNOTIFY;
1877 mutex_exit(&so->so_lock);
1878 return (error);
1879 }
1880
1881 void
so_timer_callback(void * arg)1882 so_timer_callback(void *arg)
1883 {
1884 struct sonode *so = (struct sonode *)arg;
1885
1886 mutex_enter(&so->so_lock);
1887
1888 so->so_rcv_timer_tid = 0;
1889 if (so->so_rcv_queued > 0) {
1890 so_notify_data(so, so->so_rcv_queued);
1891 } else {
1892 mutex_exit(&so->so_lock);
1893 }
1894 }
1895
1896 #ifdef DEBUG
1897 /*
1898 * Verify that the length stored in so_rcv_queued and the length of data blocks
1899 * queued is same.
1900 */
1901 static boolean_t
so_check_length(sonode_t * so)1902 so_check_length(sonode_t *so)
1903 {
1904 mblk_t *mp = so->so_rcv_q_head;
1905 int len = 0;
1906
1907 ASSERT(MUTEX_HELD(&so->so_lock));
1908
1909 if (mp != NULL) {
1910 len = msgdsize(mp);
1911 while ((mp = mp->b_next) != NULL)
1912 len += msgdsize(mp);
1913 }
1914 mp = so->so_rcv_head;
1915 if (mp != NULL) {
1916 len += msgdsize(mp);
1917 while ((mp = mp->b_next) != NULL)
1918 len += msgdsize(mp);
1919 }
1920 return ((len == so->so_rcv_queued) ? B_TRUE : B_FALSE);
1921 }
1922 #endif
1923
1924 int
so_get_mod_version(struct sockparams * sp)1925 so_get_mod_version(struct sockparams *sp)
1926 {
1927 ASSERT(sp != NULL && sp->sp_smod_info != NULL);
1928 return (sp->sp_smod_info->smod_version);
1929 }
1930
1931 /*
1932 * so_start_fallback()
1933 *
1934 * Block new socket operations from coming in, and wait for active operations
1935 * to complete. Threads that are sleeping will be woken up so they can get
1936 * out of the way.
1937 *
1938 * The caller must be a reader on so_fallback_rwlock.
1939 */
1940 static boolean_t
so_start_fallback(struct sonode * so)1941 so_start_fallback(struct sonode *so)
1942 {
1943 ASSERT(RW_READ_HELD(&so->so_fallback_rwlock));
1944
1945 mutex_enter(&so->so_lock);
1946 if (so->so_state & SS_FALLBACK_PENDING) {
1947 mutex_exit(&so->so_lock);
1948 return (B_FALSE);
1949 }
1950 so->so_state |= SS_FALLBACK_PENDING;
1951 /*
1952 * Poke all threads that might be sleeping. Any operation that comes
1953 * in after the cv_broadcast will observe the fallback pending flag
1954 * which cause the call to return where it would normally sleep.
1955 */
1956 cv_broadcast(&so->so_state_cv); /* threads in connect() */
1957 cv_broadcast(&so->so_rcv_cv); /* threads in recvmsg() */
1958 cv_broadcast(&so->so_snd_cv); /* threads in sendmsg() */
1959 mutex_enter(&so->so_acceptq_lock);
1960 cv_broadcast(&so->so_acceptq_cv); /* threads in accept() */
1961 mutex_exit(&so->so_acceptq_lock);
1962 mutex_exit(&so->so_lock);
1963
1964 /*
1965 * The main reason for the rw_tryupgrade call is to provide
1966 * observability during the fallback process. We want to
1967 * be able to see if there are pending operations.
1968 */
1969 if (rw_tryupgrade(&so->so_fallback_rwlock) == 0) {
1970 /*
1971 * It is safe to drop and reaquire the fallback lock, because
1972 * we are guaranteed that another fallback cannot take place.
1973 */
1974 rw_exit(&so->so_fallback_rwlock);
1975 DTRACE_PROBE1(pending__ops__wait, (struct sonode *), so);
1976 rw_enter(&so->so_fallback_rwlock, RW_WRITER);
1977 DTRACE_PROBE1(pending__ops__complete, (struct sonode *), so);
1978 }
1979
1980 return (B_TRUE);
1981 }
1982
1983 /*
1984 * so_end_fallback()
1985 *
1986 * Allow socket opertions back in.
1987 *
1988 * The caller must be a writer on so_fallback_rwlock.
1989 */
1990 static void
so_end_fallback(struct sonode * so)1991 so_end_fallback(struct sonode *so)
1992 {
1993 ASSERT(RW_ISWRITER(&so->so_fallback_rwlock));
1994
1995 mutex_enter(&so->so_lock);
1996 so->so_state &= ~(SS_FALLBACK_PENDING|SS_FALLBACK_DRAIN);
1997 mutex_exit(&so->so_lock);
1998
1999 rw_downgrade(&so->so_fallback_rwlock);
2000 }
2001
2002 /*
2003 * so_quiesced_cb()
2004 *
2005 * Callback passed to the protocol during fallback. It is called once
2006 * the endpoint is quiescent.
2007 *
2008 * No requests from the user, no notifications from the protocol, so it
2009 * is safe to synchronize the state. Data can also be moved without
2010 * risk for reordering.
2011 *
2012 * We do not need to hold so_lock, since there can be only one thread
2013 * operating on the sonode.
2014 */
2015 static mblk_t *
so_quiesced_cb(sock_upper_handle_t sock_handle,sock_quiesce_arg_t * arg,struct T_capability_ack * tcap,struct sockaddr * laddr,socklen_t laddrlen,struct sockaddr * faddr,socklen_t faddrlen,short opts)2016 so_quiesced_cb(sock_upper_handle_t sock_handle, sock_quiesce_arg_t *arg,
2017 struct T_capability_ack *tcap,
2018 struct sockaddr *laddr, socklen_t laddrlen,
2019 struct sockaddr *faddr, socklen_t faddrlen, short opts)
2020 {
2021 struct sonode *so = (struct sonode *)sock_handle;
2022 boolean_t atmark;
2023 mblk_t *retmp = NULL, **tailmpp = &retmp;
2024
2025 if (tcap != NULL)
2026 sotpi_update_state(so, tcap, laddr, laddrlen, faddr, faddrlen,
2027 opts);
2028
2029 /*
2030 * Some protocols do not quiece the data path during fallback. Once
2031 * we set the SS_FALLBACK_DRAIN flag any attempt to queue data will
2032 * fail and the protocol is responsible for saving the data for later
2033 * delivery (i.e., once the fallback has completed).
2034 */
2035 mutex_enter(&so->so_lock);
2036 so->so_state |= SS_FALLBACK_DRAIN;
2037 SOCKET_TIMER_CANCEL(so);
2038 mutex_exit(&so->so_lock);
2039
2040 if (so->so_rcv_head != NULL) {
2041 if (so->so_rcv_q_last_head == NULL)
2042 so->so_rcv_q_head = so->so_rcv_head;
2043 else
2044 so->so_rcv_q_last_head->b_next = so->so_rcv_head;
2045 so->so_rcv_q_last_head = so->so_rcv_last_head;
2046 }
2047
2048 atmark = (so->so_state & SS_RCVATMARK) != 0;
2049 /*
2050 * Clear any OOB state having to do with pending data. The TPI
2051 * code path will set the appropriate oob state when we move the
2052 * oob data to the STREAM head. We leave SS_HADOOBDATA since the oob
2053 * data has already been consumed.
2054 */
2055 so->so_state &= ~(SS_RCVATMARK|SS_OOBPEND|SS_HAVEOOBDATA);
2056
2057 ASSERT(so->so_oobmsg != NULL || so->so_oobmark <= so->so_rcv_queued);
2058
2059 /*
2060 * Move data to the STREAM head.
2061 */
2062 while (so->so_rcv_q_head != NULL) {
2063 mblk_t *mp = so->so_rcv_q_head;
2064 size_t mlen = msgdsize(mp);
2065
2066 so->so_rcv_q_head = mp->b_next;
2067 mp->b_next = NULL;
2068 mp->b_prev = NULL;
2069
2070 /*
2071 * Send T_EXDATA_IND if we are at the oob mark.
2072 */
2073 if (atmark) {
2074 struct T_exdata_ind *tei;
2075 mblk_t *mp1 = arg->soqa_exdata_mp;
2076
2077 arg->soqa_exdata_mp = NULL;
2078 ASSERT(mp1 != NULL);
2079 mp1->b_datap->db_type = M_PROTO;
2080 tei = (struct T_exdata_ind *)mp1->b_rptr;
2081 tei->PRIM_type = T_EXDATA_IND;
2082 tei->MORE_flag = 0;
2083 mp1->b_wptr = (uchar_t *)&tei[1];
2084
2085 if (IS_SO_OOB_INLINE(so)) {
2086 mp1->b_cont = mp;
2087 } else {
2088 ASSERT(so->so_oobmsg != NULL);
2089 mp1->b_cont = so->so_oobmsg;
2090 so->so_oobmsg = NULL;
2091
2092 /* process current mp next time around */
2093 mp->b_next = so->so_rcv_q_head;
2094 so->so_rcv_q_head = mp;
2095 mlen = 0;
2096 }
2097 mp = mp1;
2098
2099 /* we have consumed the oob mark */
2100 atmark = B_FALSE;
2101 } else if (so->so_oobmark > 0) {
2102 /*
2103 * Check if the OOB mark is within the current
2104 * mblk chain. In that case we have to split it up.
2105 */
2106 if (so->so_oobmark < mlen) {
2107 mblk_t *urg_mp = mp;
2108
2109 atmark = B_TRUE;
2110 mp = NULL;
2111 mlen = so->so_oobmark;
2112
2113 /*
2114 * It is assumed that the OOB mark does
2115 * not land within a mblk.
2116 */
2117 do {
2118 so->so_oobmark -= MBLKL(urg_mp);
2119 mp = urg_mp;
2120 urg_mp = urg_mp->b_cont;
2121 } while (so->so_oobmark > 0);
2122 mp->b_cont = NULL;
2123 if (urg_mp != NULL) {
2124 urg_mp->b_next = so->so_rcv_q_head;
2125 so->so_rcv_q_head = urg_mp;
2126 }
2127 } else {
2128 so->so_oobmark -= mlen;
2129 if (so->so_oobmark == 0)
2130 atmark = B_TRUE;
2131 }
2132 }
2133
2134 /*
2135 * Queue data on the STREAM head.
2136 */
2137 so->so_rcv_queued -= mlen;
2138 *tailmpp = mp;
2139 tailmpp = &mp->b_next;
2140 }
2141 so->so_rcv_head = NULL;
2142 so->so_rcv_last_head = NULL;
2143 so->so_rcv_q_head = NULL;
2144 so->so_rcv_q_last_head = NULL;
2145
2146 /*
2147 * Check if the oob byte is at the end of the data stream, or if the
2148 * oob byte has not yet arrived. In the latter case we have to send a
2149 * SIGURG and a mark indicator to the STREAM head. The mark indicator
2150 * is needed to guarantee correct behavior for SIOCATMARK. See block
2151 * comment in socktpi.h for more details.
2152 */
2153 if (atmark || so->so_oobmark > 0) {
2154 mblk_t *mp;
2155
2156 if (atmark && so->so_oobmsg != NULL) {
2157 struct T_exdata_ind *tei;
2158
2159 mp = arg->soqa_exdata_mp;
2160 arg->soqa_exdata_mp = NULL;
2161 ASSERT(mp != NULL);
2162 mp->b_datap->db_type = M_PROTO;
2163 tei = (struct T_exdata_ind *)mp->b_rptr;
2164 tei->PRIM_type = T_EXDATA_IND;
2165 tei->MORE_flag = 0;
2166 mp->b_wptr = (uchar_t *)&tei[1];
2167
2168 mp->b_cont = so->so_oobmsg;
2169 so->so_oobmsg = NULL;
2170
2171 *tailmpp = mp;
2172 tailmpp = &mp->b_next;
2173 } else {
2174 /* Send up the signal */
2175 mp = arg->soqa_exdata_mp;
2176 arg->soqa_exdata_mp = NULL;
2177 ASSERT(mp != NULL);
2178 DB_TYPE(mp) = M_PCSIG;
2179 *mp->b_wptr++ = (uchar_t)SIGURG;
2180 *tailmpp = mp;
2181 tailmpp = &mp->b_next;
2182
2183 /* Send up the mark indicator */
2184 mp = arg->soqa_urgmark_mp;
2185 arg->soqa_urgmark_mp = NULL;
2186 mp->b_flag = atmark ? MSGMARKNEXT : MSGNOTMARKNEXT;
2187 *tailmpp = mp;
2188 tailmpp = &mp->b_next;
2189
2190 so->so_oobmark = 0;
2191 }
2192 }
2193 ASSERT(so->so_oobmark == 0);
2194 ASSERT(so->so_rcv_queued == 0);
2195
2196 return (retmp);
2197 }
2198
2199 #ifdef DEBUG
2200 /*
2201 * Do an integrity check of the sonode. This should be done if a
2202 * fallback fails after sonode has initially been converted to use
2203 * TPI and subsequently have to be reverted.
2204 *
2205 * Failure to pass the integrity check will panic the system.
2206 */
2207 void
so_integrity_check(struct sonode * cur,struct sonode * orig)2208 so_integrity_check(struct sonode *cur, struct sonode *orig)
2209 {
2210 VERIFY(cur->so_vnode == orig->so_vnode);
2211 VERIFY(cur->so_ops == orig->so_ops);
2212 /*
2213 * For so_state we can only VERIFY the state flags in CHECK_STATE.
2214 * The other state flags might be affected by a notification from the
2215 * protocol.
2216 */
2217 #define CHECK_STATE (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_NDELAY|SS_NONBLOCK| \
2218 SS_ASYNC|SS_ACCEPTCONN|SS_SAVEDEOR|SS_RCVATMARK|SS_OOBPEND| \
2219 SS_HAVEOOBDATA|SS_HADOOBDATA|SS_SENTLASTREADSIG|SS_SENTLASTWRITESIG)
2220 VERIFY((cur->so_state & (orig->so_state & CHECK_STATE)) ==
2221 (orig->so_state & CHECK_STATE));
2222 VERIFY(cur->so_mode == orig->so_mode);
2223 VERIFY(cur->so_flag == orig->so_flag);
2224 VERIFY(cur->so_count == orig->so_count);
2225 /* Cannot VERIFY so_proto_connid; proto can update it */
2226 VERIFY(cur->so_sockparams == orig->so_sockparams);
2227 /* an error might have been recorded, but it can not be lost */
2228 VERIFY(cur->so_error != 0 || orig->so_error == 0);
2229 VERIFY(cur->so_family == orig->so_family);
2230 VERIFY(cur->so_type == orig->so_type);
2231 VERIFY(cur->so_protocol == orig->so_protocol);
2232 VERIFY(cur->so_version == orig->so_version);
2233 /* New conns might have arrived, but none should have been lost */
2234 VERIFY(cur->so_acceptq_len >= orig->so_acceptq_len);
2235 VERIFY(list_head(&cur->so_acceptq_list) ==
2236 list_head(&orig->so_acceptq_list));
2237 VERIFY(cur->so_backlog == orig->so_backlog);
2238 /* New OOB migth have arrived, but mark should not have been lost */
2239 VERIFY(cur->so_oobmark >= orig->so_oobmark);
2240 /* Cannot VERIFY so_oobmsg; the proto might have sent up a new one */
2241 VERIFY(cur->so_pgrp == orig->so_pgrp);
2242 VERIFY(cur->so_peercred == orig->so_peercred);
2243 VERIFY(cur->so_cpid == orig->so_cpid);
2244 VERIFY(cur->so_zoneid == orig->so_zoneid);
2245 /* New data migth have arrived, but none should have been lost */
2246 VERIFY(cur->so_rcv_queued >= orig->so_rcv_queued);
2247 VERIFY(cur->so_rcv_q_head == orig->so_rcv_q_head);
2248 VERIFY(cur->so_rcv_head == orig->so_rcv_head);
2249 VERIFY(cur->so_proto_handle == orig->so_proto_handle);
2250 VERIFY(cur->so_downcalls == orig->so_downcalls);
2251 /* Cannot VERIFY so_proto_props; they can be updated by proto */
2252 }
2253 #endif
2254
2255 /*
2256 * so_tpi_fallback()
2257 *
2258 * This is the fallback initation routine; things start here.
2259 *
2260 * Basic strategy:
2261 * o Block new socket operations from coming in
2262 * o Allocate/initate info needed by TPI
2263 * o Quiesce the connection, at which point we sync
2264 * state and move data
2265 * o Change operations (sonodeops) associated with the socket
2266 * o Unblock threads waiting for the fallback to finish
2267 */
2268 int
so_tpi_fallback(struct sonode * so,struct cred * cr)2269 so_tpi_fallback(struct sonode *so, struct cred *cr)
2270 {
2271 int error;
2272 queue_t *q;
2273 struct sockparams *sp;
2274 struct sockparams *newsp = NULL;
2275 so_proto_fallback_func_t fbfunc;
2276 const char *devpath;
2277 boolean_t direct;
2278 struct sonode *nso;
2279 sock_quiesce_arg_t arg = { NULL, NULL };
2280 #ifdef DEBUG
2281 struct sonode origso;
2282 #endif
2283 error = 0;
2284 sp = so->so_sockparams;
2285 fbfunc = sp->sp_smod_info->smod_proto_fallback_func;
2286
2287 /*
2288 * Cannot fallback if the socket has active filters or a krecv callback.
2289 */
2290 if (so->so_filter_active > 0 || so->so_krecv_cb != NULL)
2291 return (EINVAL);
2292
2293 switch (so->so_family) {
2294 case AF_INET:
2295 devpath = sp->sp_smod_info->smod_fallback_devpath_v4;
2296 break;
2297 case AF_INET6:
2298 devpath = sp->sp_smod_info->smod_fallback_devpath_v6;
2299 break;
2300 default:
2301 return (EINVAL);
2302 }
2303
2304 /*
2305 * Fallback can only happen if the socket module has a TPI device
2306 * and fallback function.
2307 */
2308 if (devpath == NULL || fbfunc == NULL)
2309 return (EINVAL);
2310
2311 /*
2312 * Initiate fallback; upon success we know that no new requests
2313 * will come in from the user.
2314 */
2315 if (!so_start_fallback(so))
2316 return (EAGAIN);
2317 #ifdef DEBUG
2318 /*
2319 * Make a copy of the sonode in case we need to make an integrity
2320 * check later on.
2321 */
2322 bcopy(so, &origso, sizeof (*so));
2323 #endif
2324
2325 sp->sp_stats.sps_nfallback.value.ui64++;
2326
2327 newsp = sockparams_hold_ephemeral_bydev(so->so_family, so->so_type,
2328 so->so_protocol, devpath, KM_SLEEP, &error);
2329 if (error != 0)
2330 goto out;
2331
2332 if (so->so_direct != NULL) {
2333 sodirect_t *sodp = so->so_direct;
2334 mutex_enter(&so->so_lock);
2335
2336 so->so_direct->sod_enabled = B_FALSE;
2337 so->so_state &= ~SS_SODIRECT;
2338 ASSERT(sodp->sod_uioafh == NULL);
2339 mutex_exit(&so->so_lock);
2340 }
2341
2342 /* Turn sonode into a TPI socket */
2343 error = sotpi_convert_sonode(so, newsp, &direct, &q, cr);
2344 if (error != 0)
2345 goto out;
2346 /*
2347 * When it comes to urgent data we have two cases to deal with;
2348 * (1) The oob byte has already arrived, or (2) the protocol has
2349 * notified that oob data is pending, but it has not yet arrived.
2350 *
2351 * For (1) all we need to do is send a T_EXDATA_IND to indicate were
2352 * in the byte stream the oob byte is. For (2) we have to send a
2353 * SIGURG (M_PCSIG), followed by a zero-length mblk indicating whether
2354 * the oob byte will be the next byte from the protocol.
2355 *
2356 * So in the worst case we need two mblks, one for the signal, another
2357 * for mark indication. In that case we use the exdata_mp for the sig.
2358 */
2359 arg.soqa_exdata_mp = allocb_wait(sizeof (struct T_exdata_ind),
2360 BPRI_MED, STR_NOSIG, NULL);
2361 arg.soqa_urgmark_mp = allocb_wait(0, BPRI_MED, STR_NOSIG, NULL);
2362
2363 /*
2364 * Now tell the protocol to start using TPI. so_quiesced_cb be
2365 * called once it's safe to synchronize state.
2366 */
2367 DTRACE_PROBE1(proto__fallback__begin, struct sonode *, so);
2368 error = (*fbfunc)(so->so_proto_handle, q, direct, so_quiesced_cb,
2369 &arg);
2370 DTRACE_PROBE1(proto__fallback__end, struct sonode *, so);
2371
2372 if (error != 0) {
2373 /* protocol was unable to do a fallback, revert the sonode */
2374 sotpi_revert_sonode(so, cr);
2375 goto out;
2376 }
2377
2378 /*
2379 * Walk the accept queue and notify the proto that they should
2380 * fall back to TPI. The protocol will send up the T_CONN_IND.
2381 */
2382 nso = list_head(&so->so_acceptq_list);
2383 while (nso != NULL) {
2384 int rval;
2385 struct sonode *next;
2386
2387 if (arg.soqa_exdata_mp == NULL) {
2388 arg.soqa_exdata_mp =
2389 allocb_wait(sizeof (struct T_exdata_ind),
2390 BPRI_MED, STR_NOSIG, NULL);
2391 }
2392 if (arg.soqa_urgmark_mp == NULL) {
2393 arg.soqa_urgmark_mp = allocb_wait(0, BPRI_MED,
2394 STR_NOSIG, NULL);
2395 }
2396
2397 DTRACE_PROBE1(proto__fallback__begin, struct sonode *, nso);
2398 rval = (*fbfunc)(nso->so_proto_handle, NULL, direct,
2399 so_quiesced_cb, &arg);
2400 DTRACE_PROBE1(proto__fallback__end, struct sonode *, nso);
2401 if (rval != 0) {
2402 /* Abort the connection */
2403 zcmn_err(getzoneid(), CE_WARN,
2404 "Failed to convert socket in accept queue to TPI. "
2405 "Pid = %d\n", curproc->p_pid);
2406 next = list_next(&so->so_acceptq_list, nso);
2407 list_remove(&so->so_acceptq_list, nso);
2408 so->so_acceptq_len--;
2409
2410 (void) socket_close(nso, 0, CRED());
2411 socket_destroy(nso);
2412 nso = next;
2413 } else {
2414 nso = list_next(&so->so_acceptq_list, nso);
2415 }
2416 }
2417
2418 /*
2419 * Now flush the acceptq, this will destroy all sockets. They will
2420 * be recreated in sotpi_accept().
2421 */
2422 so_acceptq_flush(so, B_FALSE);
2423
2424 mutex_enter(&so->so_lock);
2425 so->so_state |= SS_FALLBACK_COMP;
2426 mutex_exit(&so->so_lock);
2427
2428 /*
2429 * Swap the sonode ops. Socket opertations that come in once this
2430 * is done will proceed without blocking.
2431 */
2432 so->so_ops = &sotpi_sonodeops;
2433
2434 /*
2435 * Wake up any threads stuck in poll. This is needed since the poll
2436 * head changes when the fallback happens (moves from the sonode to
2437 * the STREAMS head).
2438 */
2439 pollwakeup(&so->so_poll_list, POLLERR);
2440
2441 /*
2442 * When this non-STREAM socket was created we placed an extra ref on
2443 * the associated vnode to support asynchronous close. Drop that ref
2444 * here.
2445 */
2446 ASSERT(SOTOV(so)->v_count >= 2);
2447 VN_RELE(SOTOV(so));
2448 out:
2449 so_end_fallback(so);
2450
2451 if (error != 0) {
2452 #ifdef DEBUG
2453 so_integrity_check(so, &origso);
2454 #endif
2455 zcmn_err(getzoneid(), CE_WARN,
2456 "Failed to convert socket to TPI (err=%d). Pid = %d\n",
2457 error, curproc->p_pid);
2458 if (newsp != NULL)
2459 SOCKPARAMS_DEC_REF(newsp);
2460 }
2461 if (arg.soqa_exdata_mp != NULL)
2462 freemsg(arg.soqa_exdata_mp);
2463 if (arg.soqa_urgmark_mp != NULL)
2464 freemsg(arg.soqa_urgmark_mp);
2465
2466 return (error);
2467 }
2468
2469 int
so_krecv_set(sonode_t * so,so_krecv_f cb,void * arg)2470 so_krecv_set(sonode_t *so, so_krecv_f cb, void *arg)
2471 {
2472 int ret;
2473
2474 if (cb == NULL && arg != NULL)
2475 return (EINVAL);
2476
2477 SO_BLOCK_FALLBACK(so, so_krecv_set(so, cb, arg));
2478
2479 mutex_enter(&so->so_lock);
2480 if (so->so_state & SS_FALLBACK_COMP) {
2481 mutex_exit(&so->so_lock);
2482 SO_UNBLOCK_FALLBACK(so);
2483 return (ENOTSUP);
2484 }
2485
2486 ret = so_lock_read(so, 0);
2487 VERIFY(ret == 0);
2488 /*
2489 * Other consumers may actually care about getting extant data delivered
2490 * to them, when they come along, they should figure out the best API
2491 * for that.
2492 */
2493 so_rcv_flush(so);
2494
2495 so->so_krecv_cb = cb;
2496 so->so_krecv_arg = arg;
2497
2498 so_unlock_read(so);
2499 mutex_exit(&so->so_lock);
2500 SO_UNBLOCK_FALLBACK(so);
2501
2502 return (0);
2503 }
2504
2505 void
so_krecv_unblock(sonode_t * so)2506 so_krecv_unblock(sonode_t *so)
2507 {
2508 mutex_enter(&so->so_lock);
2509 VERIFY(so->so_krecv_cb != NULL);
2510
2511 so->so_rcv_queued = 0;
2512 (void) so_check_flow_control(so);
2513 /*
2514 * so_check_flow_control() always drops so->so_lock, so we won't
2515 * need to drop it ourselves.
2516 */
2517 }
2518