1/* $NetBSD: sys_select.c,v 1.39 2014/04/25 15:52:45 pooka Exp $ */
2
3/*-
4 * Copyright (c) 2007, 2008, 2009, 2010 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran and Mindaugas Rasiukevicius.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copyright (c) 1982, 1986, 1989, 1993
34 * The Regents of the University of California. All rights reserved.
35 * (c) UNIX System Laboratories, Inc.
36 * All or some portions of this file are derived from material licensed
37 * to the University of California by American Telephone and Telegraph
38 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
39 * the permission of UNIX System Laboratories, Inc.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 * may be used to endorse or promote products derived from this software
51 * without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 * SUCH DAMAGE.
64 *
65 * @(#)sys_generic.c 8.9 (Berkeley) 2/14/95
66 */
67
68/*
69 * System calls of synchronous I/O multiplexing subsystem.
70 *
71 * Locking
72 *
73 * Two locks are used: <object-lock> and selcluster_t::sc_lock.
74 *
75 * The <object-lock> might be a device driver or another subsystem, e.g.
76 * socket or pipe. This lock is not exported, and thus invisible to this
77 * subsystem. Mainly, synchronisation between selrecord() and selnotify()
78 * routines depends on this lock, as it will be described in the comments.
79 *
80 * Lock order
81 *
82 * <object-lock> ->
83 * selcluster_t::sc_lock
84 */
85
86#include <sys/cdefs.h>
87__KERNEL_RCSID(0, "$NetBSD: sys_select.c,v 1.39 2014/04/25 15:52:45 pooka Exp $");
88
89#include <sys/param.h>
90#include <sys/systm.h>
91#include <sys/filedesc.h>
92#include <sys/file.h>
93#include <sys/proc.h>
94#include <sys/socketvar.h>
95#include <sys/signalvar.h>
96#include <sys/uio.h>
97#include <sys/kernel.h>
98#include <sys/lwp.h>
99#include <sys/poll.h>
100#include <sys/mount.h>
101#include <sys/syscallargs.h>
102#include <sys/cpu.h>
103#include <sys/atomic.h>
104#include <sys/socketvar.h>
105#include <sys/sleepq.h>
106#include <sys/sysctl.h>
107
108/* Flags for lwp::l_selflag. */
109#define SEL_RESET 0 /* awoken, interrupted, or not yet polling */
110#define SEL_SCANNING 1 /* polling descriptors */
111#define SEL_BLOCKING 2 /* blocking and waiting for event */
112#define SEL_EVENT 3 /* interrupted, events set directly */
113
114/* Operations: either select() or poll(). */
115#define SELOP_SELECT 1
116#define SELOP_POLL 2
117
118/*
119 * Per-cluster state for select()/poll(). For a system with fewer
120 * than 32 CPUs, this gives us per-CPU clusters.
121 */
122#define SELCLUSTERS 32
123#define SELCLUSTERMASK (SELCLUSTERS - 1)
124
125typedef struct selcluster {
126 kmutex_t *sc_lock;
127 sleepq_t sc_sleepq;
128 int sc_ncoll;
129 uint32_t sc_mask;
130} selcluster_t;
131
132static inline int selscan(char *, const int, const size_t, register_t *);
133static inline int pollscan(struct pollfd *, const int, register_t *);
134static void selclear(void);
135
136static const int sel_flag[] = {
137 POLLRDNORM | POLLHUP | POLLERR,
138 POLLWRNORM | POLLHUP | POLLERR,
139 POLLRDBAND
140};
141
142static syncobj_t select_sobj = {
143 SOBJ_SLEEPQ_FIFO,
144 sleepq_unsleep,
145 sleepq_changepri,
146 sleepq_lendpri,
147 syncobj_noowner,
148};
149
150static selcluster_t *selcluster[SELCLUSTERS] __read_mostly;
151static int direct_select __read_mostly = 0;
152
153/*
154 * Select system call.
155 */
156int
157sys___pselect50(struct lwp *l, const struct sys___pselect50_args *uap,
158 register_t *retval)
159{
160 /* {
161 syscallarg(int) nd;
162 syscallarg(fd_set *) in;
163 syscallarg(fd_set *) ou;
164 syscallarg(fd_set *) ex;
165 syscallarg(const struct timespec *) ts;
166 syscallarg(sigset_t *) mask;
167 } */
168 struct timespec ats, *ts = NULL;
169 sigset_t amask, *mask = NULL;
170 int error;
171
172 if (SCARG(uap, ts)) {
173 error = copyin(SCARG(uap, ts), &ats, sizeof(ats));
174 if (error)
175 return error;
176 ts = &ats;
177 }
178 if (SCARG(uap, mask) != NULL) {
179 error = copyin(SCARG(uap, mask), &amask, sizeof(amask));
180 if (error)
181 return error;
182 mask = &amask;
183 }
184
185 return selcommon(retval, SCARG(uap, nd), SCARG(uap, in),
186 SCARG(uap, ou), SCARG(uap, ex), ts, mask);
187}
188
189int
190sys___select50(struct lwp *l, const struct sys___select50_args *uap,
191 register_t *retval)
192{
193 /* {
194 syscallarg(int) nd;
195 syscallarg(fd_set *) in;
196 syscallarg(fd_set *) ou;
197 syscallarg(fd_set *) ex;
198 syscallarg(struct timeval *) tv;
199 } */
200 struct timeval atv;
201 struct timespec ats, *ts = NULL;
202 int error;
203
204 if (SCARG(uap, tv)) {
205 error = copyin(SCARG(uap, tv), (void *)&atv, sizeof(atv));
206 if (error)
207 return error;
208 TIMEVAL_TO_TIMESPEC(&atv, &ats);
209 ts = &ats;
210 }
211
212 return selcommon(retval, SCARG(uap, nd), SCARG(uap, in),
213 SCARG(uap, ou), SCARG(uap, ex), ts, NULL);
214}
215
216/*
217 * sel_do_scan: common code to perform the scan on descriptors.
218 */
219static int
220sel_do_scan(const int op, void *fds, const int nf, const size_t ni,
221 struct timespec *ts, sigset_t *mask, register_t *retval)
222{
223 lwp_t * const l = curlwp;
224 selcluster_t *sc;
225 kmutex_t *lock;
226 struct timespec sleepts;
227 int error, timo;
228
229 timo = 0;
230 if (ts && inittimeleft(ts, &sleepts) == -1) {
231 return EINVAL;
232 }
233
234 if (__predict_false(mask))
235 sigsuspendsetup(l, mask);
236
237 sc = curcpu()->ci_data.cpu_selcluster;
238 lock = sc->sc_lock;
239 l->l_selcluster = sc;
240 if (op == SELOP_SELECT) {
241 l->l_selbits = fds;
242 l->l_selni = ni;
243 } else {
244 l->l_selbits = NULL;
245 }
246
247 for (;;) {
248 int ncoll;
249
250 SLIST_INIT(&l->l_selwait);
251 l->l_selret = 0;
252
253 /*
254 * No need to lock. If this is overwritten by another value
255 * while scanning, we will retry below. We only need to see
256 * exact state from the descriptors that we are about to poll,
257 * and lock activity resulting from fo_poll is enough to
258 * provide an up to date value for new polling activity.
259 */
260 l->l_selflag = SEL_SCANNING;
261 ncoll = sc->sc_ncoll;
262
263 if (op == SELOP_SELECT) {
264 error = selscan((char *)fds, nf, ni, retval);
265 } else {
266 error = pollscan((struct pollfd *)fds, nf, retval);
267 }
268 if (error || *retval)
269 break;
270 if (ts && (timo = gettimeleft(ts, &sleepts)) <= 0)
271 break;
272 /*
273 * Acquire the lock and perform the (re)checks. Note, if
274 * collision has occured, then our state does not matter,
275 * as we must perform re-scan. Therefore, check it first.
276 */
277state_check:
278 mutex_spin_enter(lock);
279 if (__predict_false(sc->sc_ncoll != ncoll)) {
280 /* Collision: perform re-scan. */
281 mutex_spin_exit(lock);
282 selclear();
283 continue;
284 }
285 if (__predict_true(l->l_selflag == SEL_EVENT)) {
286 /* Events occured, they are set directly. */
287 mutex_spin_exit(lock);
288 break;
289 }
290 if (__predict_true(l->l_selflag == SEL_RESET)) {
291 /* Events occured, but re-scan is requested. */
292 mutex_spin_exit(lock);
293 selclear();
294 continue;
295 }
296 /* Nothing happen, therefore - sleep. */
297 l->l_selflag = SEL_BLOCKING;
298 l->l_kpriority = true;
299 sleepq_enter(&sc->sc_sleepq, l, lock);
300 sleepq_enqueue(&sc->sc_sleepq, sc, "select", &select_sobj);
301 error = sleepq_block(timo, true);
302 if (error != 0) {
303 break;
304 }
305 /* Awoken: need to check the state. */
306 goto state_check;
307 }
308 selclear();
309
310 /* Add direct events if any. */
311 if (l->l_selflag == SEL_EVENT) {
312 KASSERT(l->l_selret != 0);
313 *retval += l->l_selret;
314 }
315
316 if (__predict_false(mask))
317 sigsuspendteardown(l);
318
319 /* select and poll are not restarted after signals... */
320 if (error == ERESTART)
321 return EINTR;
322 if (error == EWOULDBLOCK)
323 return 0;
324 return error;
325}
326
327int
328selcommon(register_t *retval, int nd, fd_set *u_in, fd_set *u_ou,
329 fd_set *u_ex, struct timespec *ts, sigset_t *mask)
330{
331 char smallbits[howmany(FD_SETSIZE, NFDBITS) *
332 sizeof(fd_mask) * 6];
333 char *bits;
334 int error, nf;
335 size_t ni;
336
337 if (nd < 0)
338 return (EINVAL);
339 nf = curlwp->l_fd->fd_dt->dt_nfiles;
340 if (nd > nf) {
341 /* forgiving; slightly wrong */
342 nd = nf;
343 }
344 ni = howmany(nd, NFDBITS) * sizeof(fd_mask);
345 if (ni * 6 > sizeof(smallbits)) {
346 bits = kmem_alloc(ni * 6, KM_SLEEP);
347 if (bits == NULL)
348 return ENOMEM;
349 } else
350 bits = smallbits;
351
352#define getbits(name, x) \
353 if (u_ ## name) { \
354 error = copyin(u_ ## name, bits + ni * x, ni); \
355 if (error) \
356 goto fail; \
357 } else \
358 memset(bits + ni * x, 0, ni);
359 getbits(in, 0);
360 getbits(ou, 1);
361 getbits(ex, 2);
362#undef getbits
363
364 error = sel_do_scan(SELOP_SELECT, bits, nd, ni, ts, mask, retval);
365 if (error == 0 && u_in != NULL)
366 error = copyout(bits + ni * 3, u_in, ni);
367 if (error == 0 && u_ou != NULL)
368 error = copyout(bits + ni * 4, u_ou, ni);
369 if (error == 0 && u_ex != NULL)
370 error = copyout(bits + ni * 5, u_ex, ni);
371 fail:
372 if (bits != smallbits)
373 kmem_free(bits, ni * 6);
374 return (error);
375}
376
377static inline int
378selscan(char *bits, const int nfd, const size_t ni, register_t *retval)
379{
380 fd_mask *ibitp, *obitp;
381 int msk, i, j, fd, n;
382 file_t *fp;
383
384 ibitp = (fd_mask *)(bits + ni * 0);
385 obitp = (fd_mask *)(bits + ni * 3);
386 n = 0;
387
388 memset(obitp, 0, ni * 3);
389 for (msk = 0; msk < 3; msk++) {
390 for (i = 0; i < nfd; i += NFDBITS) {
391 fd_mask ibits, obits;
392
393 ibits = *ibitp;
394 obits = 0;
395 while ((j = ffs(ibits)) && (fd = i + --j) < nfd) {
396 ibits &= ~(1 << j);
397 if ((fp = fd_getfile(fd)) == NULL)
398 return (EBADF);
399 /*
400 * Setup an argument to selrecord(), which is
401 * a file descriptor number.
402 */
403 curlwp->l_selrec = fd;
404 if ((*fp->f_ops->fo_poll)(fp, sel_flag[msk])) {
405 obits |= (1 << j);
406 n++;
407 }
408 fd_putfile(fd);
409 }
410 if (obits != 0) {
411 if (direct_select) {
412 kmutex_t *lock;
413 lock = curlwp->l_selcluster->sc_lock;
414 mutex_spin_enter(lock);
415 *obitp |= obits;
416 mutex_spin_exit(lock);
417 } else {
418 *obitp |= obits;
419 }
420 }
421 ibitp++;
422 obitp++;
423 }
424 }
425 *retval = n;
426 return (0);
427}
428
429/*
430 * Poll system call.
431 */
432int
433sys_poll(struct lwp *l, const struct sys_poll_args *uap, register_t *retval)
434{
435 /* {
436 syscallarg(struct pollfd *) fds;
437 syscallarg(u_int) nfds;
438 syscallarg(int) timeout;
439 } */
440 struct timespec ats, *ts = NULL;
441
442 if (SCARG(uap, timeout) != INFTIM) {
443 ats.tv_sec = SCARG(uap, timeout) / 1000;
444 ats.tv_nsec = (SCARG(uap, timeout) % 1000) * 1000000;
445 ts = &ats;
446 }
447
448 return pollcommon(retval, SCARG(uap, fds), SCARG(uap, nfds), ts, NULL);
449}
450
451/*
452 * Poll system call.
453 */
454int
455sys___pollts50(struct lwp *l, const struct sys___pollts50_args *uap,
456 register_t *retval)
457{
458 /* {
459 syscallarg(struct pollfd *) fds;
460 syscallarg(u_int) nfds;
461 syscallarg(const struct timespec *) ts;
462 syscallarg(const sigset_t *) mask;
463 } */
464 struct timespec ats, *ts = NULL;
465 sigset_t amask, *mask = NULL;
466 int error;
467
468 if (SCARG(uap, ts)) {
469 error = copyin(SCARG(uap, ts), &ats, sizeof(ats));
470 if (error)
471 return error;
472 ts = &ats;
473 }
474 if (SCARG(uap, mask)) {
475 error = copyin(SCARG(uap, mask), &amask, sizeof(amask));
476 if (error)
477 return error;
478 mask = &amask;
479 }
480
481 return pollcommon(retval, SCARG(uap, fds), SCARG(uap, nfds), ts, mask);
482}
483
484int
485pollcommon(register_t *retval, struct pollfd *u_fds, u_int nfds,
486 struct timespec *ts, sigset_t *mask)
487{
488 struct pollfd smallfds[32];
489 struct pollfd *fds;
490 int error;
491 size_t ni;
492
493 if (nfds > 1000 + curlwp->l_fd->fd_dt->dt_nfiles) {
494 /*
495 * Either the user passed in a very sparse 'fds' or junk!
496 * The kmem_alloc() call below would be bad news.
497 * We could process the 'fds' array in chunks, but that
498 * is a lot of code that isn't normally useful.
499 * (Or just move the copyin/out into pollscan().)
500 * Historically the code silently truncated 'fds' to
501 * dt_nfiles entries - but that does cause issues.
502 */
503 return EINVAL;
504 }
505 ni = nfds * sizeof(struct pollfd);
506 if (ni > sizeof(smallfds)) {
507 fds = kmem_alloc(ni, KM_SLEEP);
508 if (fds == NULL)
509 return ENOMEM;
510 } else
511 fds = smallfds;
512
513 error = copyin(u_fds, fds, ni);
514 if (error)
515 goto fail;
516
517 error = sel_do_scan(SELOP_POLL, fds, nfds, ni, ts, mask, retval);
518 if (error == 0)
519 error = copyout(fds, u_fds, ni);
520 fail:
521 if (fds != smallfds)
522 kmem_free(fds, ni);
523 return (error);
524}
525
526static inline int
527pollscan(struct pollfd *fds, const int nfd, register_t *retval)
528{
529 file_t *fp;
530 int i, n = 0, revents;
531
532 for (i = 0; i < nfd; i++, fds++) {
533 fds->revents = 0;
534 if (fds->fd < 0) {
535 revents = 0;
536 } else if ((fp = fd_getfile(fds->fd)) == NULL) {
537 revents = POLLNVAL;
538 } else {
539 /*
540 * Perform poll: registers select request or returns
541 * the events which are set. Setup an argument for
542 * selrecord(), which is a pointer to struct pollfd.
543 */
544 curlwp->l_selrec = (uintptr_t)fds;
545 revents = (*fp->f_ops->fo_poll)(fp,
546 fds->events | POLLERR | POLLHUP);
547 fd_putfile(fds->fd);
548 }
549 if (revents) {
550 fds->revents = revents;
551 n++;
552 }
553 }
554 *retval = n;
555 return (0);
556}
557
558int
559seltrue(dev_t dev, int events, lwp_t *l)
560{
561
562 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
563}
564
565/*
566 * Record a select request. Concurrency issues:
567 *
568 * The caller holds the same lock across calls to selrecord() and
569 * selnotify(), so we don't need to consider a concurrent wakeup
570 * while in this routine.
571 *
572 * The only activity we need to guard against is selclear(), called by
573 * another thread that is exiting sel_do_scan().
574 * `sel_lwp' can only become non-NULL while the caller's lock is held,
575 * so it cannot become non-NULL due to a change made by another thread
576 * while we are in this routine. It can only become _NULL_ due to a
577 * call to selclear().
578 *
579 * If it is non-NULL and != selector there is the potential for
580 * selclear() to be called by another thread. If either of those
581 * conditions are true, we're not interested in touching the `named
582 * waiter' part of the selinfo record because we need to record a
583 * collision. Hence there is no need for additional locking in this
584 * routine.
585 */
586void
587selrecord(lwp_t *selector, struct selinfo *sip)
588{
589 selcluster_t *sc;
590 lwp_t *other;
591
592 KASSERT(selector == curlwp);
593
594 sc = selector->l_selcluster;
595 other = sip->sel_lwp;
596
597 if (other == selector) {
598 /* 1. We (selector) already claimed to be the first LWP. */
599 KASSERT(sip->sel_cluster == sc);
600 } else if (other == NULL) {
601 /*
602 * 2. No first LWP, therefore we (selector) are the first.
603 *
604 * There may be unnamed waiters (collisions). Issue a memory
605 * barrier to ensure that we access sel_lwp (above) before
606 * other fields - this guards against a call to selclear().
607 */
608 membar_enter();
609 sip->sel_lwp = selector;
610 SLIST_INSERT_HEAD(&selector->l_selwait, sip, sel_chain);
611 /* Copy the argument, which is for selnotify(). */
612 sip->sel_fdinfo = selector->l_selrec;
613 /* Replace selinfo's lock with the chosen cluster's lock. */
614 sip->sel_cluster = sc;
615 } else {
616 /* 3. Multiple waiters: record a collision. */
617 sip->sel_collision |= sc->sc_mask;
618 KASSERT(sip->sel_cluster != NULL);
619 }
620}
621
622/*
623 * sel_setevents: a helper function for selnotify(), to set the events
624 * for LWP sleeping in selcommon() or pollcommon().
625 */
626static inline bool
627sel_setevents(lwp_t *l, struct selinfo *sip, const int events)
628{
629 const int oflag = l->l_selflag;
630 int ret = 0;
631
632 /*
633 * If we require re-scan or it was required by somebody else,
634 * then just (re)set SEL_RESET and return.
635 */
636 if (__predict_false(events == 0 || oflag == SEL_RESET)) {
637 l->l_selflag = SEL_RESET;
638 return true;
639 }
640 /*
641 * Direct set. Note: select state of LWP is locked. First,
642 * determine whether it is selcommon() or pollcommon().
643 */
644 if (l->l_selbits != NULL) {
645 const size_t ni = l->l_selni;
646 fd_mask *fds = (fd_mask *)l->l_selbits;
647 fd_mask *ofds = (fd_mask *)((char *)fds + ni * 3);
648 const int fd = sip->sel_fdinfo, fbit = 1 << (fd & __NFDMASK);
649 const int idx = fd >> __NFDSHIFT;
650 int n;
651
652 for (n = 0; n < 3; n++) {
653 if ((fds[idx] & fbit) != 0 &&
654 (ofds[idx] & fbit) == 0 &&
655 (sel_flag[n] & events)) {
656 ofds[idx] |= fbit;
657 ret++;
658 }
659 fds = (fd_mask *)((char *)fds + ni);
660 ofds = (fd_mask *)((char *)ofds + ni);
661 }
662 } else {
663 struct pollfd *pfd = (void *)sip->sel_fdinfo;
664 int revents = events & (pfd->events | POLLERR | POLLHUP);
665
666 if (revents) {
667 if (pfd->revents == 0)
668 ret = 1;
669 pfd->revents |= revents;
670 }
671 }
672 /* Check whether there are any events to return. */
673 if (!ret) {
674 return false;
675 }
676 /* Indicate direct set and note the event (cluster lock is held). */
677 l->l_selflag = SEL_EVENT;
678 l->l_selret += ret;
679 return true;
680}
681
682/*
683 * Do a wakeup when a selectable event occurs. Concurrency issues:
684 *
685 * As per selrecord(), the caller's object lock is held. If there
686 * is a named waiter, we must acquire the associated selcluster's lock
687 * in order to synchronize with selclear() and pollers going to sleep
688 * in sel_do_scan().
689 *
690 * sip->sel_cluser cannot change at this point, as it is only changed
691 * in selrecord(), and concurrent calls to selrecord() are locked
692 * out by the caller.
693 */
694void
695selnotify(struct selinfo *sip, int events, long knhint)
696{
697 selcluster_t *sc;
698 uint32_t mask;
699 int index, oflag;
700 lwp_t *l;
701 kmutex_t *lock;
702
703 KNOTE(&sip->sel_klist, knhint);
704
705 if (sip->sel_lwp != NULL) {
706 /* One named LWP is waiting. */
707 sc = sip->sel_cluster;
708 lock = sc->sc_lock;
709 mutex_spin_enter(lock);
710 /* Still there? */
711 if (sip->sel_lwp != NULL) {
712 /*
713 * Set the events for our LWP and indicate that.
714 * Otherwise, request for a full re-scan.
715 */
716 l = sip->sel_lwp;
717 oflag = l->l_selflag;
718
719 if (!direct_select) {
720 l->l_selflag = SEL_RESET;
721 } else if (!sel_setevents(l, sip, events)) {
722 /* No events to return. */
723 mutex_spin_exit(lock);
724 return;
725 }
726
727 /*
728 * If thread is sleeping, wake it up. If it's not
729 * yet asleep, it will notice the change in state
730 * and will re-poll the descriptors.
731 */
732 if (oflag == SEL_BLOCKING && l->l_mutex == lock) {
733 KASSERT(l->l_wchan == sc);
734 sleepq_unsleep(l, false);
735 }
736 }
737 mutex_spin_exit(lock);
738 }
739
740 if ((mask = sip->sel_collision) != 0) {
741 /*
742 * There was a collision (multiple waiters): we must
743 * inform all potentially interested waiters.
744 */
745 sip->sel_collision = 0;
746 do {
747 index = ffs(mask) - 1;
748 mask &= ~(1 << index);
749 sc = selcluster[index];
750 lock = sc->sc_lock;
751 mutex_spin_enter(lock);
752 sc->sc_ncoll++;
753 sleepq_wake(&sc->sc_sleepq, sc, (u_int)-1, lock);
754 } while (__predict_false(mask != 0));
755 }
756}
757
758/*
759 * Remove an LWP from all objects that it is waiting for. Concurrency
760 * issues:
761 *
762 * The object owner's (e.g. device driver) lock is not held here. Calls
763 * can be made to selrecord() and we do not synchronize against those
764 * directly using locks. However, we use `sel_lwp' to lock out changes.
765 * Before clearing it we must use memory barriers to ensure that we can
766 * safely traverse the list of selinfo records.
767 */
768static void
769selclear(void)
770{
771 struct selinfo *sip, *next;
772 selcluster_t *sc;
773 lwp_t *l;
774 kmutex_t *lock;
775
776 l = curlwp;
777 sc = l->l_selcluster;
778 lock = sc->sc_lock;
779
780 mutex_spin_enter(lock);
781 for (sip = SLIST_FIRST(&l->l_selwait); sip != NULL; sip = next) {
782 KASSERT(sip->sel_lwp == l);
783 KASSERT(sip->sel_cluster == l->l_selcluster);
784
785 /*
786 * Read link to next selinfo record, if any.
787 * It's no longer safe to touch `sip' after clearing
788 * `sel_lwp', so ensure that the read of `sel_chain'
789 * completes before the clearing of sel_lwp becomes
790 * globally visible.
791 */
792 next = SLIST_NEXT(sip, sel_chain);
793 membar_exit();
794 /* Release the record for another named waiter to use. */
795 sip->sel_lwp = NULL;
796 }
797 mutex_spin_exit(lock);
798}
799
800/*
801 * Initialize the select/poll system calls. Called once for each
802 * CPU in the system, as they are attached.
803 */
804void
805selsysinit(struct cpu_info *ci)
806{
807 selcluster_t *sc;
808 u_int index;
809
810 /* If already a cluster in place for this bit, re-use. */
811 index = cpu_index(ci) & SELCLUSTERMASK;
812 sc = selcluster[index];
813 if (sc == NULL) {
814 sc = kmem_alloc(roundup2(sizeof(selcluster_t),
815 coherency_unit) + coherency_unit, KM_SLEEP);
816 sc = (void *)roundup2((uintptr_t)sc, coherency_unit);
817 sc->sc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED);
818 sleepq_init(&sc->sc_sleepq);
819 sc->sc_ncoll = 0;
820 sc->sc_mask = (1 << index);
821 selcluster[index] = sc;
822 }
823 ci->ci_data.cpu_selcluster = sc;
824}
825
826/*
827 * Initialize a selinfo record.
828 */
829void
830selinit(struct selinfo *sip)
831{
832
833 memset(sip, 0, sizeof(*sip));
834}
835
836/*
837 * Destroy a selinfo record. The owning object must not gain new
838 * references while this is in progress: all activity on the record
839 * must be stopped.
840 *
841 * Concurrency issues: we only need guard against a call to selclear()
842 * by a thread exiting sel_do_scan(). The caller has prevented further
843 * references being made to the selinfo record via selrecord(), and it
844 * will not call selnotify() again.
845 */
846void
847seldestroy(struct selinfo *sip)
848{
849 selcluster_t *sc;
850 kmutex_t *lock;
851 lwp_t *l;
852
853 if (sip->sel_lwp == NULL)
854 return;
855
856 /*
857 * Lock out selclear(). The selcluster pointer can't change while
858 * we are here since it is only ever changed in selrecord(),
859 * and that will not be entered again for this record because
860 * it is dying.
861 */
862 KASSERT(sip->sel_cluster != NULL);
863 sc = sip->sel_cluster;
864 lock = sc->sc_lock;
865 mutex_spin_enter(lock);
866 if ((l = sip->sel_lwp) != NULL) {
867 /*
868 * This should rarely happen, so although SLIST_REMOVE()
869 * is slow, using it here is not a problem.
870 */
871 KASSERT(l->l_selcluster == sc);
872 SLIST_REMOVE(&l->l_selwait, sip, selinfo, sel_chain);
873 sip->sel_lwp = NULL;
874 }
875 mutex_spin_exit(lock);
876}
877
878/*
879 * System control nodes.
880 */
881SYSCTL_SETUP(sysctl_select_setup, "sysctl select setup")
882{
883
884 sysctl_createv(clog, 0, NULL, NULL,
885 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
886 CTLTYPE_INT, "direct_select",
887 SYSCTL_DESCR("Enable/disable direct select (for testing)"),
888 NULL, 0, &direct_select, 0,
889 CTL_KERN, CTL_CREATE, CTL_EOL);
890}
891