xref: /netbsd-src/sys/kern/sys_select.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*	$NetBSD: sys_select.c,v 1.41 2018/01/30 07:52:23 ozaki-r 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 1.41 2018/01/30 07:52:23 ozaki-r 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 
125 typedef 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 
132 static inline int	selscan(char *, const int, const size_t, register_t *);
133 static inline int	pollscan(struct pollfd *, const int, register_t *);
134 static void		selclear(void);
135 
136 static const int sel_flag[] = {
137 	POLLRDNORM | POLLHUP | POLLERR,
138 	POLLWRNORM | POLLHUP | POLLERR,
139 	POLLRDBAND
140 };
141 
142 static syncobj_t select_sobj = {
143 	.sobj_flag	= SOBJ_SLEEPQ_FIFO,
144 	.sobj_unsleep	= sleepq_unsleep,
145 	.sobj_changepri	= sleepq_changepri,
146 	.sobj_lendpri	= sleepq_lendpri,
147 	.sobj_owner	= syncobj_noowner,
148 };
149 
150 static selcluster_t	*selcluster[SELCLUSTERS] __read_mostly;
151 static int		direct_select __read_mostly = 0;
152 
153 /*
154  * Select system call.
155  */
156 int
157 sys___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 
189 int
190 sys___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  */
219 static int
220 sel_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 		 */
277 state_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 
327 int
328 selcommon(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 	else
348 		bits = smallbits;
349 
350 #define	getbits(name, x)						\
351 	if (u_ ## name) {						\
352 		error = copyin(u_ ## name, bits + ni * x, ni);		\
353 		if (error)						\
354 			goto fail;					\
355 	} else								\
356 		memset(bits + ni * x, 0, ni);
357 	getbits(in, 0);
358 	getbits(ou, 1);
359 	getbits(ex, 2);
360 #undef	getbits
361 
362 	error = sel_do_scan(SELOP_SELECT, bits, nd, ni, ts, mask, retval);
363 	if (error == 0 && u_in != NULL)
364 		error = copyout(bits + ni * 3, u_in, ni);
365 	if (error == 0 && u_ou != NULL)
366 		error = copyout(bits + ni * 4, u_ou, ni);
367 	if (error == 0 && u_ex != NULL)
368 		error = copyout(bits + ni * 5, u_ex, ni);
369  fail:
370 	if (bits != smallbits)
371 		kmem_free(bits, ni * 6);
372 	return (error);
373 }
374 
375 static inline int
376 selscan(char *bits, const int nfd, const size_t ni, register_t *retval)
377 {
378 	fd_mask *ibitp, *obitp;
379 	int msk, i, j, fd, n;
380 	file_t *fp;
381 
382 	ibitp = (fd_mask *)(bits + ni * 0);
383 	obitp = (fd_mask *)(bits + ni * 3);
384 	n = 0;
385 
386 	memset(obitp, 0, ni * 3);
387 	for (msk = 0; msk < 3; msk++) {
388 		for (i = 0; i < nfd; i += NFDBITS) {
389 			fd_mask ibits, obits;
390 
391 			ibits = *ibitp;
392 			obits = 0;
393 			while ((j = ffs(ibits)) && (fd = i + --j) < nfd) {
394 				ibits &= ~(1 << j);
395 				if ((fp = fd_getfile(fd)) == NULL)
396 					return (EBADF);
397 				/*
398 				 * Setup an argument to selrecord(), which is
399 				 * a file descriptor number.
400 				 */
401 				curlwp->l_selrec = fd;
402 				if ((*fp->f_ops->fo_poll)(fp, sel_flag[msk])) {
403 					obits |= (1 << j);
404 					n++;
405 				}
406 				fd_putfile(fd);
407 			}
408 			if (obits != 0) {
409 				if (direct_select) {
410 					kmutex_t *lock;
411 					lock = curlwp->l_selcluster->sc_lock;
412 					mutex_spin_enter(lock);
413 					*obitp |= obits;
414 					mutex_spin_exit(lock);
415 				} else {
416 					*obitp |= obits;
417 				}
418 			}
419 			ibitp++;
420 			obitp++;
421 		}
422 	}
423 	*retval = n;
424 	return (0);
425 }
426 
427 /*
428  * Poll system call.
429  */
430 int
431 sys_poll(struct lwp *l, const struct sys_poll_args *uap, register_t *retval)
432 {
433 	/* {
434 		syscallarg(struct pollfd *)	fds;
435 		syscallarg(u_int)		nfds;
436 		syscallarg(int)			timeout;
437 	} */
438 	struct timespec	ats, *ts = NULL;
439 
440 	if (SCARG(uap, timeout) != INFTIM) {
441 		ats.tv_sec = SCARG(uap, timeout) / 1000;
442 		ats.tv_nsec = (SCARG(uap, timeout) % 1000) * 1000000;
443 		ts = &ats;
444 	}
445 
446 	return pollcommon(retval, SCARG(uap, fds), SCARG(uap, nfds), ts, NULL);
447 }
448 
449 /*
450  * Poll system call.
451  */
452 int
453 sys___pollts50(struct lwp *l, const struct sys___pollts50_args *uap,
454     register_t *retval)
455 {
456 	/* {
457 		syscallarg(struct pollfd *)		fds;
458 		syscallarg(u_int)			nfds;
459 		syscallarg(const struct timespec *)	ts;
460 		syscallarg(const sigset_t *)		mask;
461 	} */
462 	struct timespec	ats, *ts = NULL;
463 	sigset_t	amask, *mask = NULL;
464 	int		error;
465 
466 	if (SCARG(uap, ts)) {
467 		error = copyin(SCARG(uap, ts), &ats, sizeof(ats));
468 		if (error)
469 			return error;
470 		ts = &ats;
471 	}
472 	if (SCARG(uap, mask)) {
473 		error = copyin(SCARG(uap, mask), &amask, sizeof(amask));
474 		if (error)
475 			return error;
476 		mask = &amask;
477 	}
478 
479 	return pollcommon(retval, SCARG(uap, fds), SCARG(uap, nfds), ts, mask);
480 }
481 
482 int
483 pollcommon(register_t *retval, struct pollfd *u_fds, u_int nfds,
484     struct timespec *ts, sigset_t *mask)
485 {
486 	struct pollfd	smallfds[32];
487 	struct pollfd	*fds;
488 	int		error;
489 	size_t		ni;
490 
491 	if (nfds > 1000 + curlwp->l_fd->fd_dt->dt_nfiles) {
492 		/*
493 		 * Either the user passed in a very sparse 'fds' or junk!
494 		 * The kmem_alloc() call below would be bad news.
495 		 * We could process the 'fds' array in chunks, but that
496 		 * is a lot of code that isn't normally useful.
497 		 * (Or just move the copyin/out into pollscan().)
498 		 * Historically the code silently truncated 'fds' to
499 		 * dt_nfiles entries - but that does cause issues.
500 		 */
501 		return EINVAL;
502 	}
503 	ni = nfds * sizeof(struct pollfd);
504 	if (ni > sizeof(smallfds))
505 		fds = kmem_alloc(ni, KM_SLEEP);
506 	else
507 		fds = smallfds;
508 
509 	error = copyin(u_fds, fds, ni);
510 	if (error)
511 		goto fail;
512 
513 	error = sel_do_scan(SELOP_POLL, fds, nfds, ni, ts, mask, retval);
514 	if (error == 0)
515 		error = copyout(fds, u_fds, ni);
516  fail:
517 	if (fds != smallfds)
518 		kmem_free(fds, ni);
519 	return (error);
520 }
521 
522 static inline int
523 pollscan(struct pollfd *fds, const int nfd, register_t *retval)
524 {
525 	file_t *fp;
526 	int i, n = 0, revents;
527 
528 	for (i = 0; i < nfd; i++, fds++) {
529 		fds->revents = 0;
530 		if (fds->fd < 0) {
531 			revents = 0;
532 		} else if ((fp = fd_getfile(fds->fd)) == NULL) {
533 			revents = POLLNVAL;
534 		} else {
535 			/*
536 			 * Perform poll: registers select request or returns
537 			 * the events which are set.  Setup an argument for
538 			 * selrecord(), which is a pointer to struct pollfd.
539 			 */
540 			curlwp->l_selrec = (uintptr_t)fds;
541 			revents = (*fp->f_ops->fo_poll)(fp,
542 			    fds->events | POLLERR | POLLHUP);
543 			fd_putfile(fds->fd);
544 		}
545 		if (revents) {
546 			fds->revents = revents;
547 			n++;
548 		}
549 	}
550 	*retval = n;
551 	return (0);
552 }
553 
554 int
555 seltrue(dev_t dev, int events, lwp_t *l)
556 {
557 
558 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
559 }
560 
561 /*
562  * Record a select request.  Concurrency issues:
563  *
564  * The caller holds the same lock across calls to selrecord() and
565  * selnotify(), so we don't need to consider a concurrent wakeup
566  * while in this routine.
567  *
568  * The only activity we need to guard against is selclear(), called by
569  * another thread that is exiting sel_do_scan().
570  * `sel_lwp' can only become non-NULL while the caller's lock is held,
571  * so it cannot become non-NULL due to a change made by another thread
572  * while we are in this routine.  It can only become _NULL_ due to a
573  * call to selclear().
574  *
575  * If it is non-NULL and != selector there is the potential for
576  * selclear() to be called by another thread.  If either of those
577  * conditions are true, we're not interested in touching the `named
578  * waiter' part of the selinfo record because we need to record a
579  * collision.  Hence there is no need for additional locking in this
580  * routine.
581  */
582 void
583 selrecord(lwp_t *selector, struct selinfo *sip)
584 {
585 	selcluster_t *sc;
586 	lwp_t *other;
587 
588 	KASSERT(selector == curlwp);
589 
590 	sc = selector->l_selcluster;
591 	other = sip->sel_lwp;
592 
593 	if (other == selector) {
594 		/* 1. We (selector) already claimed to be the first LWP. */
595 		KASSERT(sip->sel_cluster == sc);
596 	} else if (other == NULL) {
597 		/*
598 		 * 2. No first LWP, therefore we (selector) are the first.
599 		 *
600 		 * There may be unnamed waiters (collisions).  Issue a memory
601 		 * barrier to ensure that we access sel_lwp (above) before
602 		 * other fields - this guards against a call to selclear().
603 		 */
604 		membar_enter();
605 		sip->sel_lwp = selector;
606 		SLIST_INSERT_HEAD(&selector->l_selwait, sip, sel_chain);
607 		/* Copy the argument, which is for selnotify(). */
608 		sip->sel_fdinfo = selector->l_selrec;
609 		/* Replace selinfo's lock with the chosen cluster's lock. */
610 		sip->sel_cluster = sc;
611 	} else {
612 		/* 3. Multiple waiters: record a collision. */
613 		sip->sel_collision |= sc->sc_mask;
614 		KASSERT(sip->sel_cluster != NULL);
615 	}
616 }
617 
618 /*
619  * sel_setevents: a helper function for selnotify(), to set the events
620  * for LWP sleeping in selcommon() or pollcommon().
621  */
622 static inline bool
623 sel_setevents(lwp_t *l, struct selinfo *sip, const int events)
624 {
625 	const int oflag = l->l_selflag;
626 	int ret = 0;
627 
628 	/*
629 	 * If we require re-scan or it was required by somebody else,
630 	 * then just (re)set SEL_RESET and return.
631 	 */
632 	if (__predict_false(events == 0 || oflag == SEL_RESET)) {
633 		l->l_selflag = SEL_RESET;
634 		return true;
635 	}
636 	/*
637 	 * Direct set.  Note: select state of LWP is locked.  First,
638 	 * determine whether it is selcommon() or pollcommon().
639 	 */
640 	if (l->l_selbits != NULL) {
641 		const size_t ni = l->l_selni;
642 		fd_mask *fds = (fd_mask *)l->l_selbits;
643 		fd_mask *ofds = (fd_mask *)((char *)fds + ni * 3);
644 		const int fd = sip->sel_fdinfo, fbit = 1 << (fd & __NFDMASK);
645 		const int idx = fd >> __NFDSHIFT;
646 		int n;
647 
648 		for (n = 0; n < 3; n++) {
649 			if ((fds[idx] & fbit) != 0 &&
650 			    (ofds[idx] & fbit) == 0 &&
651 			    (sel_flag[n] & events)) {
652 				ofds[idx] |= fbit;
653 				ret++;
654 			}
655 			fds = (fd_mask *)((char *)fds + ni);
656 			ofds = (fd_mask *)((char *)ofds + ni);
657 		}
658 	} else {
659 		struct pollfd *pfd = (void *)sip->sel_fdinfo;
660 		int revents = events & (pfd->events | POLLERR | POLLHUP);
661 
662 		if (revents) {
663 			if (pfd->revents == 0)
664 				ret = 1;
665 			pfd->revents |= revents;
666 		}
667 	}
668 	/* Check whether there are any events to return. */
669 	if (!ret) {
670 		return false;
671 	}
672 	/* Indicate direct set and note the event (cluster lock is held). */
673 	l->l_selflag = SEL_EVENT;
674 	l->l_selret += ret;
675 	return true;
676 }
677 
678 /*
679  * Do a wakeup when a selectable event occurs.  Concurrency issues:
680  *
681  * As per selrecord(), the caller's object lock is held.  If there
682  * is a named waiter, we must acquire the associated selcluster's lock
683  * in order to synchronize with selclear() and pollers going to sleep
684  * in sel_do_scan().
685  *
686  * sip->sel_cluser cannot change at this point, as it is only changed
687  * in selrecord(), and concurrent calls to selrecord() are locked
688  * out by the caller.
689  */
690 void
691 selnotify(struct selinfo *sip, int events, long knhint)
692 {
693 	selcluster_t *sc;
694 	uint32_t mask;
695 	int index, oflag;
696 	lwp_t *l;
697 	kmutex_t *lock;
698 
699 	KNOTE(&sip->sel_klist, knhint);
700 
701 	if (sip->sel_lwp != NULL) {
702 		/* One named LWP is waiting. */
703 		sc = sip->sel_cluster;
704 		lock = sc->sc_lock;
705 		mutex_spin_enter(lock);
706 		/* Still there? */
707 		if (sip->sel_lwp != NULL) {
708 			/*
709 			 * Set the events for our LWP and indicate that.
710 			 * Otherwise, request for a full re-scan.
711 			 */
712 			l = sip->sel_lwp;
713 			oflag = l->l_selflag;
714 
715 			if (!direct_select) {
716 				l->l_selflag = SEL_RESET;
717 			} else if (!sel_setevents(l, sip, events)) {
718 				/* No events to return. */
719 				mutex_spin_exit(lock);
720 				return;
721 			}
722 
723 			/*
724 			 * If thread is sleeping, wake it up.  If it's not
725 			 * yet asleep, it will notice the change in state
726 			 * and will re-poll the descriptors.
727 			 */
728 			if (oflag == SEL_BLOCKING && l->l_mutex == lock) {
729 				KASSERT(l->l_wchan == sc);
730 				sleepq_unsleep(l, false);
731 			}
732 		}
733 		mutex_spin_exit(lock);
734 	}
735 
736 	if ((mask = sip->sel_collision) != 0) {
737 		/*
738 		 * There was a collision (multiple waiters): we must
739 		 * inform all potentially interested waiters.
740 		 */
741 		sip->sel_collision = 0;
742 		do {
743 			index = ffs(mask) - 1;
744 			mask &= ~(1 << index);
745 			sc = selcluster[index];
746 			lock = sc->sc_lock;
747 			mutex_spin_enter(lock);
748 			sc->sc_ncoll++;
749 			sleepq_wake(&sc->sc_sleepq, sc, (u_int)-1, lock);
750 		} while (__predict_false(mask != 0));
751 	}
752 }
753 
754 /*
755  * Remove an LWP from all objects that it is waiting for.  Concurrency
756  * issues:
757  *
758  * The object owner's (e.g. device driver) lock is not held here.  Calls
759  * can be made to selrecord() and we do not synchronize against those
760  * directly using locks.  However, we use `sel_lwp' to lock out changes.
761  * Before clearing it we must use memory barriers to ensure that we can
762  * safely traverse the list of selinfo records.
763  */
764 static void
765 selclear(void)
766 {
767 	struct selinfo *sip, *next;
768 	selcluster_t *sc;
769 	lwp_t *l;
770 	kmutex_t *lock;
771 
772 	l = curlwp;
773 	sc = l->l_selcluster;
774 	lock = sc->sc_lock;
775 
776 	mutex_spin_enter(lock);
777 	for (sip = SLIST_FIRST(&l->l_selwait); sip != NULL; sip = next) {
778 		KASSERT(sip->sel_lwp == l);
779 		KASSERT(sip->sel_cluster == l->l_selcluster);
780 
781 		/*
782 		 * Read link to next selinfo record, if any.
783 		 * It's no longer safe to touch `sip' after clearing
784 		 * `sel_lwp', so ensure that the read of `sel_chain'
785 		 * completes before the clearing of sel_lwp becomes
786 		 * globally visible.
787 		 */
788 		next = SLIST_NEXT(sip, sel_chain);
789 		membar_exit();
790 		/* Release the record for another named waiter to use. */
791 		sip->sel_lwp = NULL;
792 	}
793 	mutex_spin_exit(lock);
794 }
795 
796 /*
797  * Initialize the select/poll system calls.  Called once for each
798  * CPU in the system, as they are attached.
799  */
800 void
801 selsysinit(struct cpu_info *ci)
802 {
803 	selcluster_t *sc;
804 	u_int index;
805 
806 	/* If already a cluster in place for this bit, re-use. */
807 	index = cpu_index(ci) & SELCLUSTERMASK;
808 	sc = selcluster[index];
809 	if (sc == NULL) {
810 		sc = kmem_alloc(roundup2(sizeof(selcluster_t),
811 		    coherency_unit) + coherency_unit, KM_SLEEP);
812 		sc = (void *)roundup2((uintptr_t)sc, coherency_unit);
813 		sc->sc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED);
814 		sleepq_init(&sc->sc_sleepq);
815 		sc->sc_ncoll = 0;
816 		sc->sc_mask = (1 << index);
817 		selcluster[index] = sc;
818 	}
819 	ci->ci_data.cpu_selcluster = sc;
820 }
821 
822 /*
823  * Initialize a selinfo record.
824  */
825 void
826 selinit(struct selinfo *sip)
827 {
828 
829 	memset(sip, 0, sizeof(*sip));
830 }
831 
832 /*
833  * Destroy a selinfo record.  The owning object must not gain new
834  * references while this is in progress: all activity on the record
835  * must be stopped.
836  *
837  * Concurrency issues: we only need guard against a call to selclear()
838  * by a thread exiting sel_do_scan().  The caller has prevented further
839  * references being made to the selinfo record via selrecord(), and it
840  * will not call selnotify() again.
841  */
842 void
843 seldestroy(struct selinfo *sip)
844 {
845 	selcluster_t *sc;
846 	kmutex_t *lock;
847 	lwp_t *l;
848 
849 	if (sip->sel_lwp == NULL)
850 		return;
851 
852 	/*
853 	 * Lock out selclear().  The selcluster pointer can't change while
854 	 * we are here since it is only ever changed in selrecord(),
855 	 * and that will not be entered again for this record because
856 	 * it is dying.
857 	 */
858 	KASSERT(sip->sel_cluster != NULL);
859 	sc = sip->sel_cluster;
860 	lock = sc->sc_lock;
861 	mutex_spin_enter(lock);
862 	if ((l = sip->sel_lwp) != NULL) {
863 		/*
864 		 * This should rarely happen, so although SLIST_REMOVE()
865 		 * is slow, using it here is not a problem.
866 		 */
867 		KASSERT(l->l_selcluster == sc);
868 		SLIST_REMOVE(&l->l_selwait, sip, selinfo, sel_chain);
869 		sip->sel_lwp = NULL;
870 	}
871 	mutex_spin_exit(lock);
872 }
873 
874 /*
875  * System control nodes.
876  */
877 SYSCTL_SETUP(sysctl_select_setup, "sysctl select setup")
878 {
879 
880 	sysctl_createv(clog, 0, NULL, NULL,
881 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
882 		CTLTYPE_INT, "direct_select",
883 		SYSCTL_DESCR("Enable/disable direct select (for testing)"),
884 		NULL, 0, &direct_select, 0,
885 		CTL_KERN, CTL_CREATE, CTL_EOL);
886 }
887