1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/proc.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
53 #include <sys/uio.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/limits.h>
57 #include <sys/malloc.h>
58 #include <sys/poll.h>
59 #include <sys/resourcevar.h>
60 #include <sys/selinfo.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/vnode.h>
66 #include <sys/bio.h>
67 #include <sys/buf.h>
68 #include <sys/condvar.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72
73 #include <security/audit/audit.h>
74
75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
78
79 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
80 u_int);
81 static int pollscan(struct thread *, struct pollfd *, u_int);
82 static int pollrescan(struct thread *);
83 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
84 static int selrescan(struct thread *, fd_mask **, fd_mask **);
85 static void selfdalloc(struct thread *, void *);
86 static void selfdfree(struct seltd *, struct selfd *);
87 static int dofileread(struct thread *, int, struct file *, struct uio *,
88 off_t, int);
89 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
90 off_t, int);
91 static void doselwakeup(struct selinfo *, int);
92 static void seltdinit(struct thread *);
93 static int seltdwait(struct thread *, int);
94 static void seltdclear(struct thread *);
95
96 /*
97 * One seltd per-thread allocated on demand as needed.
98 *
99 * t - protected by st_mtx
100 * k - Only accessed by curthread or read-only
101 */
102 struct seltd {
103 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
104 struct selfd *st_free1; /* (k) free fd for read set. */
105 struct selfd *st_free2; /* (k) free fd for write set. */
106 struct mtx st_mtx; /* Protects struct seltd */
107 struct cv st_wait; /* (t) Wait channel. */
108 int st_flags; /* (t) SELTD_ flags. */
109 };
110
111 #define SELTD_PENDING 0x0001 /* We have pending events. */
112 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
113
114 /*
115 * One selfd allocated per-thread per-file-descriptor.
116 * f - protected by sf_mtx
117 */
118 struct selfd {
119 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
120 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
121 struct selinfo *sf_si; /* (f) selinfo when linked. */
122 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
123 struct seltd *sf_td; /* (k) owning seltd. */
124 void *sf_cookie; /* (k) fd or pollfd. */
125 };
126
127 static uma_zone_t selfd_zone;
128 static struct mtx_pool *mtxpool_select;
129
130 #ifndef _SYS_SYSPROTO_H_
131 struct read_args {
132 int fd;
133 void *buf;
134 size_t nbyte;
135 };
136 #endif
137 int
138 read(td, uap)
139 struct thread *td;
140 struct read_args *uap;
141 {
142 struct uio auio;
143 struct iovec aiov;
144 int error;
145
146 if (uap->nbyte > INT_MAX)
147 return (EINVAL);
148 aiov.iov_base = uap->buf;
149 aiov.iov_len = uap->nbyte;
150 auio.uio_iov = &aiov;
151 auio.uio_iovcnt = 1;
152 auio.uio_resid = uap->nbyte;
153 auio.uio_segflg = UIO_USERSPACE;
154 error = kern_readv(td, uap->fd, &auio);
155 return(error);
156 }
157
158 /*
159 * Positioned read system call
160 */
161 #ifndef _SYS_SYSPROTO_H_
162 struct pread_args {
163 int fd;
164 void *buf;
165 size_t nbyte;
166 int pad;
167 off_t offset;
168 };
169 #endif
170 int
171 pread(td, uap)
172 struct thread *td;
173 struct pread_args *uap;
174 {
175 struct uio auio;
176 struct iovec aiov;
177 int error;
178
179 if (uap->nbyte > INT_MAX)
180 return (EINVAL);
181 aiov.iov_base = uap->buf;
182 aiov.iov_len = uap->nbyte;
183 auio.uio_iov = &aiov;
184 auio.uio_iovcnt = 1;
185 auio.uio_resid = uap->nbyte;
186 auio.uio_segflg = UIO_USERSPACE;
187 error = kern_preadv(td, uap->fd, &auio, uap->offset);
188 return(error);
189 }
190
191 int
192 freebsd6_pread(td, uap)
193 struct thread *td;
194 struct freebsd6_pread_args *uap;
195 {
196 struct pread_args oargs;
197
198 oargs.fd = uap->fd;
199 oargs.buf = uap->buf;
200 oargs.nbyte = uap->nbyte;
201 oargs.offset = uap->offset;
202 return (pread(td, &oargs));
203 }
204
205 /*
206 * Scatter read system call.
207 */
208 #ifndef _SYS_SYSPROTO_H_
209 struct readv_args {
210 int fd;
211 struct iovec *iovp;
212 u_int iovcnt;
213 };
214 #endif
215 int
216 readv(struct thread *td, struct readv_args *uap)
217 {
218 struct uio *auio;
219 int error;
220
221 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
222 if (error)
223 return (error);
224 error = kern_readv(td, uap->fd, auio);
225 free(auio, M_IOV);
226 return (error);
227 }
228
229 int
230 kern_readv(struct thread *td, int fd, struct uio *auio)
231 {
232 struct file *fp;
233 int error;
234
235 error = fget_read(td, fd, &fp);
236 if (error)
237 return (error);
238 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
239 fdrop(fp, td);
240 return (error);
241 }
242
243 /*
244 * Scatter positioned read system call.
245 */
246 #ifndef _SYS_SYSPROTO_H_
247 struct preadv_args {
248 int fd;
249 struct iovec *iovp;
250 u_int iovcnt;
251 off_t offset;
252 };
253 #endif
254 int
255 preadv(struct thread *td, struct preadv_args *uap)
256 {
257 struct uio *auio;
258 int error;
259
260 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
261 if (error)
262 return (error);
263 error = kern_preadv(td, uap->fd, auio, uap->offset);
264 free(auio, M_IOV);
265 return (error);
266 }
267
268 int
269 kern_preadv(td, fd, auio, offset)
270 struct thread *td;
271 int fd;
272 struct uio *auio;
273 off_t offset;
274 {
275 struct file *fp;
276 int error;
277
278 error = fget_read(td, fd, &fp);
279 if (error)
280 return (error);
281 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
282 error = ESPIPE;
283 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
284 error = EINVAL;
285 else
286 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
287 fdrop(fp, td);
288 return (error);
289 }
290
291 /*
292 * Common code for readv and preadv that reads data in
293 * from a file using the passed in uio, offset, and flags.
294 */
295 static int
296 dofileread(td, fd, fp, auio, offset, flags)
297 struct thread *td;
298 int fd;
299 struct file *fp;
300 struct uio *auio;
301 off_t offset;
302 int flags;
303 {
304 ssize_t cnt;
305 int error;
306 #ifdef KTRACE
307 struct uio *ktruio = NULL;
308 #endif
309
310 /* Finish zero length reads right here */
311 if (auio->uio_resid == 0) {
312 td->td_retval[0] = 0;
313 return(0);
314 }
315 auio->uio_rw = UIO_READ;
316 auio->uio_offset = offset;
317 auio->uio_td = td;
318 #ifdef KTRACE
319 if (KTRPOINT(td, KTR_GENIO))
320 ktruio = cloneuio(auio);
321 #endif
322 cnt = auio->uio_resid;
323 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
324 if (auio->uio_resid != cnt && (error == ERESTART ||
325 error == EINTR || error == EWOULDBLOCK))
326 error = 0;
327 }
328 cnt -= auio->uio_resid;
329 #ifdef KTRACE
330 if (ktruio != NULL) {
331 ktruio->uio_resid = cnt;
332 ktrgenio(fd, UIO_READ, ktruio, error);
333 }
334 #endif
335 td->td_retval[0] = cnt;
336 return (error);
337 }
338
339 #ifndef _SYS_SYSPROTO_H_
340 struct write_args {
341 int fd;
342 const void *buf;
343 size_t nbyte;
344 };
345 #endif
346 int
347 write(td, uap)
348 struct thread *td;
349 struct write_args *uap;
350 {
351 struct uio auio;
352 struct iovec aiov;
353 int error;
354
355 if (uap->nbyte > INT_MAX)
356 return (EINVAL);
357 aiov.iov_base = (void *)(uintptr_t)uap->buf;
358 aiov.iov_len = uap->nbyte;
359 auio.uio_iov = &aiov;
360 auio.uio_iovcnt = 1;
361 auio.uio_resid = uap->nbyte;
362 auio.uio_segflg = UIO_USERSPACE;
363 error = kern_writev(td, uap->fd, &auio);
364 return(error);
365 }
366
367 /*
368 * Positioned write system call.
369 */
370 #ifndef _SYS_SYSPROTO_H_
371 struct pwrite_args {
372 int fd;
373 const void *buf;
374 size_t nbyte;
375 int pad;
376 off_t offset;
377 };
378 #endif
379 int
380 pwrite(td, uap)
381 struct thread *td;
382 struct pwrite_args *uap;
383 {
384 struct uio auio;
385 struct iovec aiov;
386 int error;
387
388 if (uap->nbyte > INT_MAX)
389 return (EINVAL);
390 aiov.iov_base = (void *)(uintptr_t)uap->buf;
391 aiov.iov_len = uap->nbyte;
392 auio.uio_iov = &aiov;
393 auio.uio_iovcnt = 1;
394 auio.uio_resid = uap->nbyte;
395 auio.uio_segflg = UIO_USERSPACE;
396 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
397 return(error);
398 }
399
400 int
401 freebsd6_pwrite(td, uap)
402 struct thread *td;
403 struct freebsd6_pwrite_args *uap;
404 {
405 struct pwrite_args oargs;
406
407 oargs.fd = uap->fd;
408 oargs.buf = uap->buf;
409 oargs.nbyte = uap->nbyte;
410 oargs.offset = uap->offset;
411 return (pwrite(td, &oargs));
412 }
413
414 /*
415 * Gather write system call.
416 */
417 #ifndef _SYS_SYSPROTO_H_
418 struct writev_args {
419 int fd;
420 struct iovec *iovp;
421 u_int iovcnt;
422 };
423 #endif
424 int
425 writev(struct thread *td, struct writev_args *uap)
426 {
427 struct uio *auio;
428 int error;
429
430 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
431 if (error)
432 return (error);
433 error = kern_writev(td, uap->fd, auio);
434 free(auio, M_IOV);
435 return (error);
436 }
437
438 int
439 kern_writev(struct thread *td, int fd, struct uio *auio)
440 {
441 struct file *fp;
442 int error;
443
444 error = fget_write(td, fd, &fp);
445 if (error)
446 return (error);
447 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
448 fdrop(fp, td);
449 return (error);
450 }
451
452 /*
453 * Gather positioned write system call.
454 */
455 #ifndef _SYS_SYSPROTO_H_
456 struct pwritev_args {
457 int fd;
458 struct iovec *iovp;
459 u_int iovcnt;
460 off_t offset;
461 };
462 #endif
463 int
464 pwritev(struct thread *td, struct pwritev_args *uap)
465 {
466 struct uio *auio;
467 int error;
468
469 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
470 if (error)
471 return (error);
472 error = kern_pwritev(td, uap->fd, auio, uap->offset);
473 free(auio, M_IOV);
474 return (error);
475 }
476
477 int
478 kern_pwritev(td, fd, auio, offset)
479 struct thread *td;
480 struct uio *auio;
481 int fd;
482 off_t offset;
483 {
484 struct file *fp;
485 int error;
486
487 error = fget_write(td, fd, &fp);
488 if (error)
489 return (error);
490 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
491 error = ESPIPE;
492 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
493 error = EINVAL;
494 else
495 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
496 fdrop(fp, td);
497 return (error);
498 }
499
500 /*
501 * Common code for writev and pwritev that writes data to
502 * a file using the passed in uio, offset, and flags.
503 */
504 static int
505 dofilewrite(td, fd, fp, auio, offset, flags)
506 struct thread *td;
507 int fd;
508 struct file *fp;
509 struct uio *auio;
510 off_t offset;
511 int flags;
512 {
513 ssize_t cnt;
514 int error;
515 #ifdef KTRACE
516 struct uio *ktruio = NULL;
517 #endif
518
519 auio->uio_rw = UIO_WRITE;
520 auio->uio_td = td;
521 auio->uio_offset = offset;
522 #ifdef KTRACE
523 if (KTRPOINT(td, KTR_GENIO))
524 ktruio = cloneuio(auio);
525 #endif
526 cnt = auio->uio_resid;
527 if (fp->f_type == DTYPE_VNODE)
528 bwillwrite();
529 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
530 if (auio->uio_resid != cnt && (error == ERESTART ||
531 error == EINTR || error == EWOULDBLOCK))
532 error = 0;
533 /* Socket layer is responsible for issuing SIGPIPE. */
534 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
535 PROC_LOCK(td->td_proc);
536 tdksignal(td, SIGPIPE, NULL);
537 PROC_UNLOCK(td->td_proc);
538 }
539 }
540 cnt -= auio->uio_resid;
541 #ifdef KTRACE
542 if (ktruio != NULL) {
543 ktruio->uio_resid = cnt;
544 ktrgenio(fd, UIO_WRITE, ktruio, error);
545 }
546 #endif
547 td->td_retval[0] = cnt;
548 return (error);
549 }
550
551 /*
552 * Truncate a file given a file descriptor.
553 *
554 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
555 * descriptor isn't writable.
556 */
557 int
558 kern_ftruncate(td, fd, length)
559 struct thread *td;
560 int fd;
561 off_t length;
562 {
563 struct file *fp;
564 int error;
565
566 AUDIT_ARG_FD(fd);
567 if (length < 0)
568 return (EINVAL);
569 error = fget(td, fd, &fp);
570 if (error)
571 return (error);
572 AUDIT_ARG_FILE(td->td_proc, fp);
573 if (!(fp->f_flag & FWRITE)) {
574 fdrop(fp, td);
575 return (EINVAL);
576 }
577 error = fo_truncate(fp, length, td->td_ucred, td);
578 fdrop(fp, td);
579 return (error);
580 }
581
582 #ifndef _SYS_SYSPROTO_H_
583 struct ftruncate_args {
584 int fd;
585 int pad;
586 off_t length;
587 };
588 #endif
589 int
590 ftruncate(td, uap)
591 struct thread *td;
592 struct ftruncate_args *uap;
593 {
594
595 return (kern_ftruncate(td, uap->fd, uap->length));
596 }
597
598 #if defined(COMPAT_43)
599 #ifndef _SYS_SYSPROTO_H_
600 struct oftruncate_args {
601 int fd;
602 long length;
603 };
604 #endif
605 int
606 oftruncate(td, uap)
607 struct thread *td;
608 struct oftruncate_args *uap;
609 {
610
611 return (kern_ftruncate(td, uap->fd, uap->length));
612 }
613 #endif /* COMPAT_43 */
614
615 #ifndef _SYS_SYSPROTO_H_
616 struct ioctl_args {
617 int fd;
618 u_long com;
619 caddr_t data;
620 };
621 #endif
622 /* ARGSUSED */
623 int
624 ioctl(struct thread *td, struct ioctl_args *uap)
625 {
626 u_long com;
627 int arg, error;
628 u_int size;
629 caddr_t data;
630
631 if (uap->com > 0xffffffff) {
632 printf(
633 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
634 td->td_proc->p_pid, td->td_name, uap->com);
635 uap->com &= 0xffffffff;
636 }
637 com = uap->com;
638
639 /*
640 * Interpret high order word to find amount of data to be
641 * copied to/from the user's address space.
642 */
643 size = IOCPARM_LEN(com);
644 if ((size > IOCPARM_MAX) ||
645 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
646 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
647 ((com & IOC_OUT) && size == 0) ||
648 #else
649 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
650 #endif
651 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
652 return (ENOTTY);
653
654 if (size > 0) {
655 if (com & IOC_VOID) {
656 /* Integer argument. */
657 arg = (intptr_t)uap->data;
658 data = (void *)&arg;
659 size = 0;
660 } else
661 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
662 } else
663 data = (void *)&uap->data;
664 if (com & IOC_IN) {
665 error = copyin(uap->data, data, (u_int)size);
666 if (error) {
667 if (size > 0)
668 free(data, M_IOCTLOPS);
669 return (error);
670 }
671 } else if (com & IOC_OUT) {
672 /*
673 * Zero the buffer so the user always
674 * gets back something deterministic.
675 */
676 bzero(data, size);
677 }
678
679 error = kern_ioctl(td, uap->fd, com, data);
680
681 if (error == 0 && (com & IOC_OUT))
682 error = copyout(data, uap->data, (u_int)size);
683
684 if (size > 0)
685 free(data, M_IOCTLOPS);
686 return (error);
687 }
688
689 int
690 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
691 {
692 struct file *fp;
693 struct filedesc *fdp;
694 int error;
695 int tmp;
696
697 AUDIT_ARG_FD(fd);
698 AUDIT_ARG_CMD(com);
699 if ((error = fget(td, fd, &fp)) != 0)
700 return (error);
701 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
702 fdrop(fp, td);
703 return (EBADF);
704 }
705 fdp = td->td_proc->p_fd;
706 switch (com) {
707 case FIONCLEX:
708 FILEDESC_XLOCK(fdp);
709 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
710 FILEDESC_XUNLOCK(fdp);
711 goto out;
712 case FIOCLEX:
713 FILEDESC_XLOCK(fdp);
714 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
715 FILEDESC_XUNLOCK(fdp);
716 goto out;
717 case FIONBIO:
718 if ((tmp = *(int *)data))
719 atomic_set_int(&fp->f_flag, FNONBLOCK);
720 else
721 atomic_clear_int(&fp->f_flag, FNONBLOCK);
722 data = (void *)&tmp;
723 break;
724 case FIOASYNC:
725 if ((tmp = *(int *)data))
726 atomic_set_int(&fp->f_flag, FASYNC);
727 else
728 atomic_clear_int(&fp->f_flag, FASYNC);
729 data = (void *)&tmp;
730 break;
731 }
732
733 error = fo_ioctl(fp, com, data, td->td_ucred, td);
734 out:
735 fdrop(fp, td);
736 return (error);
737 }
738
739 int
740 poll_no_poll(int events)
741 {
742 /*
743 * Return true for read/write. If the user asked for something
744 * special, return POLLNVAL, so that clients have a way of
745 * determining reliably whether or not the extended
746 * functionality is present without hard-coding knowledge
747 * of specific filesystem implementations.
748 */
749 if (events & ~POLLSTANDARD)
750 return (POLLNVAL);
751
752 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
753 }
754
755 int
756 pselect(struct thread *td, struct pselect_args *uap)
757 {
758 struct timespec ts;
759 struct timeval tv, *tvp;
760 sigset_t set, *uset;
761 int error;
762
763 if (uap->ts != NULL) {
764 error = copyin(uap->ts, &ts, sizeof(ts));
765 if (error != 0)
766 return (error);
767 TIMESPEC_TO_TIMEVAL(&tv, &ts);
768 tvp = &tv;
769 } else
770 tvp = NULL;
771 if (uap->sm != NULL) {
772 error = copyin(uap->sm, &set, sizeof(set));
773 if (error != 0)
774 return (error);
775 uset = &set;
776 } else
777 uset = NULL;
778 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
779 uset, NFDBITS));
780 }
781
782 int
783 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
784 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
785 {
786 int error;
787
788 if (uset != NULL) {
789 error = kern_sigprocmask(td, SIG_SETMASK, uset,
790 &td->td_oldsigmask, 0);
791 if (error != 0)
792 return (error);
793 td->td_pflags |= TDP_OLDMASK;
794 /*
795 * Make sure that ast() is called on return to
796 * usermode and TDP_OLDMASK is cleared, restoring old
797 * sigmask.
798 */
799 thread_lock(td);
800 td->td_flags |= TDF_ASTPENDING;
801 thread_unlock(td);
802 }
803 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
804 return (error);
805 }
806
807 #ifndef _SYS_SYSPROTO_H_
808 struct select_args {
809 int nd;
810 fd_set *in, *ou, *ex;
811 struct timeval *tv;
812 };
813 #endif
814 int
815 select(struct thread *td, struct select_args *uap)
816 {
817 struct timeval tv, *tvp;
818 int error;
819
820 if (uap->tv != NULL) {
821 error = copyin(uap->tv, &tv, sizeof(tv));
822 if (error)
823 return (error);
824 tvp = &tv;
825 } else
826 tvp = NULL;
827
828 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
829 NFDBITS));
830 }
831
832 /*
833 * In the unlikely case when user specified n greater then the last
834 * open file descriptor, check that no bits are set after the last
835 * valid fd. We must return EBADF if any is set.
836 *
837 * There are applications that rely on the behaviour.
838 *
839 * nd is fd_lastfile + 1.
840 */
841 static int
842 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
843 {
844 char *addr, *oaddr;
845 int b, i, res;
846 uint8_t bits;
847
848 if (nd >= ndu || fd_in == NULL)
849 return (0);
850
851 oaddr = NULL;
852 bits = 0; /* silence gcc */
853 for (i = nd; i < ndu; i++) {
854 b = i / NBBY;
855 #if BYTE_ORDER == LITTLE_ENDIAN
856 addr = (char *)fd_in + b;
857 #else
858 addr = (char *)fd_in;
859 if (abi_nfdbits == NFDBITS) {
860 addr += rounddown(b, sizeof(fd_mask)) +
861 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
862 } else {
863 addr += rounddown(b, sizeof(uint32_t)) +
864 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
865 }
866 #endif
867 if (addr != oaddr) {
868 res = fubyte(addr);
869 if (res == -1)
870 return (EFAULT);
871 oaddr = addr;
872 bits = res;
873 }
874 if ((bits & (1 << (i % NBBY))) != 0)
875 return (EBADF);
876 }
877 return (0);
878 }
879
880 int
881 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
882 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
883 {
884 struct filedesc *fdp;
885 /*
886 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
887 * infds with the new FD_SETSIZE of 1024, and more than enough for
888 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
889 * of 256.
890 */
891 fd_mask s_selbits[howmany(2048, NFDBITS)];
892 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
893 struct timeval atv, rtv, ttv;
894 int error, lf, ndu, timo;
895 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
896
897 if (nd < 0)
898 return (EINVAL);
899 fdp = td->td_proc->p_fd;
900 ndu = nd;
901 lf = fdp->fd_lastfile;
902 if (nd > lf + 1)
903 nd = lf + 1;
904
905 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
906 if (error != 0)
907 return (error);
908 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
909 if (error != 0)
910 return (error);
911 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
912 if (error != 0)
913 return (error);
914
915 /*
916 * Allocate just enough bits for the non-null fd_sets. Use the
917 * preallocated auto buffer if possible.
918 */
919 nfdbits = roundup(nd, NFDBITS);
920 ncpbytes = nfdbits / NBBY;
921 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
922 nbufbytes = 0;
923 if (fd_in != NULL)
924 nbufbytes += 2 * ncpbytes;
925 if (fd_ou != NULL)
926 nbufbytes += 2 * ncpbytes;
927 if (fd_ex != NULL)
928 nbufbytes += 2 * ncpbytes;
929 if (nbufbytes <= sizeof s_selbits)
930 selbits = &s_selbits[0];
931 else
932 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
933
934 /*
935 * Assign pointers into the bit buffers and fetch the input bits.
936 * Put the output buffers together so that they can be bzeroed
937 * together.
938 */
939 sbp = selbits;
940 #define getbits(name, x) \
941 do { \
942 if (name == NULL) { \
943 ibits[x] = NULL; \
944 obits[x] = NULL; \
945 } else { \
946 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
947 obits[x] = sbp; \
948 sbp += ncpbytes / sizeof *sbp; \
949 error = copyin(name, ibits[x], ncpubytes); \
950 if (error != 0) \
951 goto done; \
952 bzero((char *)ibits[x] + ncpubytes, \
953 ncpbytes - ncpubytes); \
954 } \
955 } while (0)
956 getbits(fd_in, 0);
957 getbits(fd_ou, 1);
958 getbits(fd_ex, 2);
959 #undef getbits
960
961 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
962 /*
963 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
964 * we are running under 32-bit emulation. This should be more
965 * generic.
966 */
967 #define swizzle_fdset(bits) \
968 if (abi_nfdbits != NFDBITS && bits != NULL) { \
969 int i; \
970 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
971 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
972 }
973 #else
974 #define swizzle_fdset(bits)
975 #endif
976
977 /* Make sure the bit order makes it through an ABI transition */
978 swizzle_fdset(ibits[0]);
979 swizzle_fdset(ibits[1]);
980 swizzle_fdset(ibits[2]);
981
982 if (nbufbytes != 0)
983 bzero(selbits, nbufbytes / 2);
984
985 if (tvp != NULL) {
986 atv = *tvp;
987 if (itimerfix(&atv)) {
988 error = EINVAL;
989 goto done;
990 }
991 getmicrouptime(&rtv);
992 timevaladd(&atv, &rtv);
993 } else {
994 atv.tv_sec = 0;
995 atv.tv_usec = 0;
996 }
997 timo = 0;
998 seltdinit(td);
999 /* Iterate until the timeout expires or descriptors become ready. */
1000 for (;;) {
1001 error = selscan(td, ibits, obits, nd);
1002 if (error || td->td_retval[0] != 0)
1003 break;
1004 if (atv.tv_sec || atv.tv_usec) {
1005 getmicrouptime(&rtv);
1006 if (timevalcmp(&rtv, &atv, >=))
1007 break;
1008 ttv = atv;
1009 timevalsub(&ttv, &rtv);
1010 timo = ttv.tv_sec > 24 * 60 * 60 ?
1011 24 * 60 * 60 * hz : tvtohz(&ttv);
1012 }
1013 error = seltdwait(td, timo);
1014 if (error)
1015 break;
1016 error = selrescan(td, ibits, obits);
1017 if (error || td->td_retval[0] != 0)
1018 break;
1019 }
1020 seltdclear(td);
1021
1022 done:
1023 /* select is not restarted after signals... */
1024 if (error == ERESTART)
1025 error = EINTR;
1026 if (error == EWOULDBLOCK)
1027 error = 0;
1028
1029 /* swizzle bit order back, if necessary */
1030 swizzle_fdset(obits[0]);
1031 swizzle_fdset(obits[1]);
1032 swizzle_fdset(obits[2]);
1033 #undef swizzle_fdset
1034
1035 #define putbits(name, x) \
1036 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1037 error = error2;
1038 if (error == 0) {
1039 int error2;
1040
1041 putbits(fd_in, 0);
1042 putbits(fd_ou, 1);
1043 putbits(fd_ex, 2);
1044 #undef putbits
1045 }
1046 if (selbits != &s_selbits[0])
1047 free(selbits, M_SELECT);
1048
1049 return (error);
1050 }
1051 /*
1052 * Convert a select bit set to poll flags.
1053 *
1054 * The backend always returns POLLHUP/POLLERR if appropriate and we
1055 * return this as a set bit in any set.
1056 */
1057 static int select_flags[3] = {
1058 POLLRDNORM | POLLHUP | POLLERR,
1059 POLLWRNORM | POLLHUP | POLLERR,
1060 POLLRDBAND | POLLERR
1061 };
1062
1063 /*
1064 * Compute the fo_poll flags required for a fd given by the index and
1065 * bit position in the fd_mask array.
1066 */
1067 static __inline int
1068 selflags(fd_mask **ibits, int idx, fd_mask bit)
1069 {
1070 int flags;
1071 int msk;
1072
1073 flags = 0;
1074 for (msk = 0; msk < 3; msk++) {
1075 if (ibits[msk] == NULL)
1076 continue;
1077 if ((ibits[msk][idx] & bit) == 0)
1078 continue;
1079 flags |= select_flags[msk];
1080 }
1081 return (flags);
1082 }
1083
1084 /*
1085 * Set the appropriate output bits given a mask of fired events and the
1086 * input bits originally requested.
1087 */
1088 static __inline int
1089 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1090 {
1091 int msk;
1092 int n;
1093
1094 n = 0;
1095 for (msk = 0; msk < 3; msk++) {
1096 if ((events & select_flags[msk]) == 0)
1097 continue;
1098 if (ibits[msk] == NULL)
1099 continue;
1100 if ((ibits[msk][idx] & bit) == 0)
1101 continue;
1102 /*
1103 * XXX Check for a duplicate set. This can occur because a
1104 * socket calls selrecord() twice for each poll() call
1105 * resulting in two selfds per real fd. selrescan() will
1106 * call selsetbits twice as a result.
1107 */
1108 if ((obits[msk][idx] & bit) != 0)
1109 continue;
1110 obits[msk][idx] |= bit;
1111 n++;
1112 }
1113
1114 return (n);
1115 }
1116
1117 /*
1118 * Traverse the list of fds attached to this thread's seltd and check for
1119 * completion.
1120 */
1121 static int
1122 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1123 {
1124 struct filedesc *fdp;
1125 struct selinfo *si;
1126 struct seltd *stp;
1127 struct selfd *sfp;
1128 struct selfd *sfn;
1129 struct file *fp;
1130 fd_mask bit;
1131 int fd, ev, n, idx;
1132
1133 fdp = td->td_proc->p_fd;
1134 stp = td->td_sel;
1135 n = 0;
1136 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1137 fd = (int)(uintptr_t)sfp->sf_cookie;
1138 si = sfp->sf_si;
1139 selfdfree(stp, sfp);
1140 /* If the selinfo wasn't cleared the event didn't fire. */
1141 if (si != NULL)
1142 continue;
1143 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1144 return (EBADF);
1145 idx = fd / NFDBITS;
1146 bit = (fd_mask)1 << (fd % NFDBITS);
1147 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1148 fdrop(fp, td);
1149 if (ev != 0)
1150 n += selsetbits(ibits, obits, idx, bit, ev);
1151 }
1152 stp->st_flags = 0;
1153 td->td_retval[0] = n;
1154 return (0);
1155 }
1156
1157 /*
1158 * Perform the initial filedescriptor scan and register ourselves with
1159 * each selinfo.
1160 */
1161 static int
1162 selscan(td, ibits, obits, nfd)
1163 struct thread *td;
1164 fd_mask **ibits, **obits;
1165 int nfd;
1166 {
1167 struct filedesc *fdp;
1168 struct file *fp;
1169 fd_mask bit;
1170 int ev, flags, end, fd;
1171 int n, idx;
1172
1173 fdp = td->td_proc->p_fd;
1174 n = 0;
1175 for (idx = 0, fd = 0; fd < nfd; idx++) {
1176 end = imin(fd + NFDBITS, nfd);
1177 for (bit = 1; fd < end; bit <<= 1, fd++) {
1178 /* Compute the list of events we're interested in. */
1179 flags = selflags(ibits, idx, bit);
1180 if (flags == 0)
1181 continue;
1182 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1183 return (EBADF);
1184 selfdalloc(td, (void *)(uintptr_t)fd);
1185 ev = fo_poll(fp, flags, td->td_ucred, td);
1186 fdrop(fp, td);
1187 if (ev != 0)
1188 n += selsetbits(ibits, obits, idx, bit, ev);
1189 }
1190 }
1191
1192 td->td_retval[0] = n;
1193 return (0);
1194 }
1195
1196 #ifndef _SYS_SYSPROTO_H_
1197 struct poll_args {
1198 struct pollfd *fds;
1199 u_int nfds;
1200 int timeout;
1201 };
1202 #endif
1203 int
1204 poll(td, uap)
1205 struct thread *td;
1206 struct poll_args *uap;
1207 {
1208 struct pollfd *bits;
1209 struct pollfd smallbits[32];
1210 struct timeval atv, rtv, ttv;
1211 int error, timo;
1212 u_int nfds;
1213 size_t ni;
1214
1215 nfds = uap->nfds;
1216 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1217 return (EINVAL);
1218 ni = nfds * sizeof(struct pollfd);
1219 if (ni > sizeof(smallbits))
1220 bits = malloc(ni, M_TEMP, M_WAITOK);
1221 else
1222 bits = smallbits;
1223 error = copyin(uap->fds, bits, ni);
1224 if (error)
1225 goto done;
1226 if (uap->timeout != INFTIM) {
1227 atv.tv_sec = uap->timeout / 1000;
1228 atv.tv_usec = (uap->timeout % 1000) * 1000;
1229 if (itimerfix(&atv)) {
1230 error = EINVAL;
1231 goto done;
1232 }
1233 getmicrouptime(&rtv);
1234 timevaladd(&atv, &rtv);
1235 } else {
1236 atv.tv_sec = 0;
1237 atv.tv_usec = 0;
1238 }
1239 timo = 0;
1240 seltdinit(td);
1241 /* Iterate until the timeout expires or descriptors become ready. */
1242 for (;;) {
1243 error = pollscan(td, bits, nfds);
1244 if (error || td->td_retval[0] != 0)
1245 break;
1246 if (atv.tv_sec || atv.tv_usec) {
1247 getmicrouptime(&rtv);
1248 if (timevalcmp(&rtv, &atv, >=))
1249 break;
1250 ttv = atv;
1251 timevalsub(&ttv, &rtv);
1252 timo = ttv.tv_sec > 24 * 60 * 60 ?
1253 24 * 60 * 60 * hz : tvtohz(&ttv);
1254 }
1255 error = seltdwait(td, timo);
1256 if (error)
1257 break;
1258 error = pollrescan(td);
1259 if (error || td->td_retval[0] != 0)
1260 break;
1261 }
1262 seltdclear(td);
1263
1264 done:
1265 /* poll is not restarted after signals... */
1266 if (error == ERESTART)
1267 error = EINTR;
1268 if (error == EWOULDBLOCK)
1269 error = 0;
1270 if (error == 0) {
1271 error = pollout(td, bits, uap->fds, nfds);
1272 if (error)
1273 goto out;
1274 }
1275 out:
1276 if (ni > sizeof(smallbits))
1277 free(bits, M_TEMP);
1278 return (error);
1279 }
1280
1281 static int
1282 pollrescan(struct thread *td)
1283 {
1284 struct seltd *stp;
1285 struct selfd *sfp;
1286 struct selfd *sfn;
1287 struct selinfo *si;
1288 struct filedesc *fdp;
1289 struct file *fp;
1290 struct pollfd *fd;
1291 int n;
1292
1293 n = 0;
1294 fdp = td->td_proc->p_fd;
1295 stp = td->td_sel;
1296 FILEDESC_SLOCK(fdp);
1297 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1298 fd = (struct pollfd *)sfp->sf_cookie;
1299 si = sfp->sf_si;
1300 selfdfree(stp, sfp);
1301 /* If the selinfo wasn't cleared the event didn't fire. */
1302 if (si != NULL)
1303 continue;
1304 fp = fdp->fd_ofiles[fd->fd];
1305 if (fp == NULL) {
1306 fd->revents = POLLNVAL;
1307 n++;
1308 continue;
1309 }
1310 /*
1311 * Note: backend also returns POLLHUP and
1312 * POLLERR if appropriate.
1313 */
1314 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1315 if (fd->revents != 0)
1316 n++;
1317 }
1318 FILEDESC_SUNLOCK(fdp);
1319 stp->st_flags = 0;
1320 td->td_retval[0] = n;
1321 return (0);
1322 }
1323
1324
1325 static int
1326 pollout(td, fds, ufds, nfd)
1327 struct thread *td;
1328 struct pollfd *fds;
1329 struct pollfd *ufds;
1330 u_int nfd;
1331 {
1332 int error = 0;
1333 u_int i = 0;
1334 u_int n = 0;
1335
1336 for (i = 0; i < nfd; i++) {
1337 error = copyout(&fds->revents, &ufds->revents,
1338 sizeof(ufds->revents));
1339 if (error)
1340 return (error);
1341 if (fds->revents != 0)
1342 n++;
1343 fds++;
1344 ufds++;
1345 }
1346 td->td_retval[0] = n;
1347 return (0);
1348 }
1349
1350 static int
1351 pollscan(td, fds, nfd)
1352 struct thread *td;
1353 struct pollfd *fds;
1354 u_int nfd;
1355 {
1356 struct filedesc *fdp = td->td_proc->p_fd;
1357 int i;
1358 struct file *fp;
1359 int n = 0;
1360
1361 FILEDESC_SLOCK(fdp);
1362 for (i = 0; i < nfd; i++, fds++) {
1363 if (fds->fd >= fdp->fd_nfiles) {
1364 fds->revents = POLLNVAL;
1365 n++;
1366 } else if (fds->fd < 0) {
1367 fds->revents = 0;
1368 } else {
1369 fp = fdp->fd_ofiles[fds->fd];
1370 if (fp == NULL) {
1371 fds->revents = POLLNVAL;
1372 n++;
1373 } else {
1374 /*
1375 * Note: backend also returns POLLHUP and
1376 * POLLERR if appropriate.
1377 */
1378 selfdalloc(td, fds);
1379 fds->revents = fo_poll(fp, fds->events,
1380 td->td_ucred, td);
1381 /*
1382 * POSIX requires POLLOUT to be never
1383 * set simultaneously with POLLHUP.
1384 */
1385 if ((fds->revents & POLLHUP) != 0)
1386 fds->revents &= ~POLLOUT;
1387
1388 if (fds->revents != 0)
1389 n++;
1390 }
1391 }
1392 }
1393 FILEDESC_SUNLOCK(fdp);
1394 td->td_retval[0] = n;
1395 return (0);
1396 }
1397
1398 /*
1399 * OpenBSD poll system call.
1400 *
1401 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1402 */
1403 #ifndef _SYS_SYSPROTO_H_
1404 struct openbsd_poll_args {
1405 struct pollfd *fds;
1406 u_int nfds;
1407 int timeout;
1408 };
1409 #endif
1410 int
1411 openbsd_poll(td, uap)
1412 register struct thread *td;
1413 register struct openbsd_poll_args *uap;
1414 {
1415 return (poll(td, (struct poll_args *)uap));
1416 }
1417
1418 /*
1419 * XXX This was created specifically to support netncp and netsmb. This
1420 * allows the caller to specify a socket to wait for events on. It returns
1421 * 0 if any events matched and an error otherwise. There is no way to
1422 * determine which events fired.
1423 */
1424 int
1425 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1426 {
1427 struct timeval atv, rtv, ttv;
1428 int error, timo;
1429
1430 if (tvp != NULL) {
1431 atv = *tvp;
1432 if (itimerfix(&atv))
1433 return (EINVAL);
1434 getmicrouptime(&rtv);
1435 timevaladd(&atv, &rtv);
1436 } else {
1437 atv.tv_sec = 0;
1438 atv.tv_usec = 0;
1439 }
1440
1441 timo = 0;
1442 seltdinit(td);
1443 /*
1444 * Iterate until the timeout expires or the socket becomes ready.
1445 */
1446 for (;;) {
1447 selfdalloc(td, NULL);
1448 error = sopoll(so, events, NULL, td);
1449 /* error here is actually the ready events. */
1450 if (error)
1451 return (0);
1452 if (atv.tv_sec || atv.tv_usec) {
1453 getmicrouptime(&rtv);
1454 if (timevalcmp(&rtv, &atv, >=)) {
1455 seltdclear(td);
1456 return (EWOULDBLOCK);
1457 }
1458 ttv = atv;
1459 timevalsub(&ttv, &rtv);
1460 timo = ttv.tv_sec > 24 * 60 * 60 ?
1461 24 * 60 * 60 * hz : tvtohz(&ttv);
1462 }
1463 error = seltdwait(td, timo);
1464 seltdclear(td);
1465 if (error)
1466 break;
1467 }
1468 /* XXX Duplicates ncp/smb behavior. */
1469 if (error == ERESTART)
1470 error = 0;
1471 return (error);
1472 }
1473
1474 /*
1475 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1476 * have two select sets, one for read and another for write.
1477 */
1478 static void
1479 selfdalloc(struct thread *td, void *cookie)
1480 {
1481 struct seltd *stp;
1482
1483 stp = td->td_sel;
1484 if (stp->st_free1 == NULL)
1485 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1486 stp->st_free1->sf_td = stp;
1487 stp->st_free1->sf_cookie = cookie;
1488 if (stp->st_free2 == NULL)
1489 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1490 stp->st_free2->sf_td = stp;
1491 stp->st_free2->sf_cookie = cookie;
1492 }
1493
1494 static void
1495 selfdfree(struct seltd *stp, struct selfd *sfp)
1496 {
1497 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1498 mtx_lock(sfp->sf_mtx);
1499 if (sfp->sf_si)
1500 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1501 mtx_unlock(sfp->sf_mtx);
1502 uma_zfree(selfd_zone, sfp);
1503 }
1504
1505 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1506 void
1507 seldrain(sip)
1508 struct selinfo *sip;
1509 {
1510
1511 /*
1512 * This feature is already provided by doselwakeup(), thus it is
1513 * enough to go for it.
1514 * Eventually, the context, should take care to avoid races
1515 * between thread calling select()/poll() and file descriptor
1516 * detaching, but, again, the races are just the same as
1517 * selwakeup().
1518 */
1519 doselwakeup(sip, -1);
1520 }
1521
1522 /*
1523 * Record a select request.
1524 */
1525 void
1526 selrecord(selector, sip)
1527 struct thread *selector;
1528 struct selinfo *sip;
1529 {
1530 struct selfd *sfp;
1531 struct seltd *stp;
1532 struct mtx *mtxp;
1533
1534 stp = selector->td_sel;
1535 /*
1536 * Don't record when doing a rescan.
1537 */
1538 if (stp->st_flags & SELTD_RESCAN)
1539 return;
1540 /*
1541 * Grab one of the preallocated descriptors.
1542 */
1543 sfp = NULL;
1544 if ((sfp = stp->st_free1) != NULL)
1545 stp->st_free1 = NULL;
1546 else if ((sfp = stp->st_free2) != NULL)
1547 stp->st_free2 = NULL;
1548 else
1549 panic("selrecord: No free selfd on selq");
1550 mtxp = sip->si_mtx;
1551 if (mtxp == NULL)
1552 mtxp = mtx_pool_find(mtxpool_select, sip);
1553 /*
1554 * Initialize the sfp and queue it in the thread.
1555 */
1556 sfp->sf_si = sip;
1557 sfp->sf_mtx = mtxp;
1558 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1559 /*
1560 * Now that we've locked the sip, check for initialization.
1561 */
1562 mtx_lock(mtxp);
1563 if (sip->si_mtx == NULL) {
1564 sip->si_mtx = mtxp;
1565 TAILQ_INIT(&sip->si_tdlist);
1566 }
1567 /*
1568 * Add this thread to the list of selfds listening on this selinfo.
1569 */
1570 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1571 mtx_unlock(sip->si_mtx);
1572 }
1573
1574 /* Wake up a selecting thread. */
1575 void
1576 selwakeup(sip)
1577 struct selinfo *sip;
1578 {
1579 doselwakeup(sip, -1);
1580 }
1581
1582 /* Wake up a selecting thread, and set its priority. */
1583 void
1584 selwakeuppri(sip, pri)
1585 struct selinfo *sip;
1586 int pri;
1587 {
1588 doselwakeup(sip, pri);
1589 }
1590
1591 /*
1592 * Do a wakeup when a selectable event occurs.
1593 */
1594 static void
1595 doselwakeup(sip, pri)
1596 struct selinfo *sip;
1597 int pri;
1598 {
1599 struct selfd *sfp;
1600 struct selfd *sfn;
1601 struct seltd *stp;
1602
1603 /* If it's not initialized there can't be any waiters. */
1604 if (sip->si_mtx == NULL)
1605 return;
1606 /*
1607 * Locking the selinfo locks all selfds associated with it.
1608 */
1609 mtx_lock(sip->si_mtx);
1610 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1611 /*
1612 * Once we remove this sfp from the list and clear the
1613 * sf_si seltdclear will know to ignore this si.
1614 */
1615 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1616 sfp->sf_si = NULL;
1617 stp = sfp->sf_td;
1618 mtx_lock(&stp->st_mtx);
1619 stp->st_flags |= SELTD_PENDING;
1620 cv_broadcastpri(&stp->st_wait, pri);
1621 mtx_unlock(&stp->st_mtx);
1622 }
1623 mtx_unlock(sip->si_mtx);
1624 }
1625
1626 static void
1627 seltdinit(struct thread *td)
1628 {
1629 struct seltd *stp;
1630
1631 if ((stp = td->td_sel) != NULL)
1632 goto out;
1633 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1634 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1635 cv_init(&stp->st_wait, "select");
1636 out:
1637 stp->st_flags = 0;
1638 STAILQ_INIT(&stp->st_selq);
1639 }
1640
1641 static int
1642 seltdwait(struct thread *td, int timo)
1643 {
1644 struct seltd *stp;
1645 int error;
1646
1647 stp = td->td_sel;
1648 /*
1649 * An event of interest may occur while we do not hold the seltd
1650 * locked so check the pending flag before we sleep.
1651 */
1652 mtx_lock(&stp->st_mtx);
1653 /*
1654 * Any further calls to selrecord will be a rescan.
1655 */
1656 stp->st_flags |= SELTD_RESCAN;
1657 if (stp->st_flags & SELTD_PENDING) {
1658 mtx_unlock(&stp->st_mtx);
1659 return (0);
1660 }
1661 if (timo > 0)
1662 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1663 else
1664 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1665 mtx_unlock(&stp->st_mtx);
1666
1667 return (error);
1668 }
1669
1670 void
1671 seltdfini(struct thread *td)
1672 {
1673 struct seltd *stp;
1674
1675 stp = td->td_sel;
1676 if (stp == NULL)
1677 return;
1678 if (stp->st_free1)
1679 uma_zfree(selfd_zone, stp->st_free1);
1680 if (stp->st_free2)
1681 uma_zfree(selfd_zone, stp->st_free2);
1682 td->td_sel = NULL;
1683 free(stp, M_SELECT);
1684 }
1685
1686 /*
1687 * Remove the references to the thread from all of the objects we were
1688 * polling.
1689 */
1690 static void
1691 seltdclear(struct thread *td)
1692 {
1693 struct seltd *stp;
1694 struct selfd *sfp;
1695 struct selfd *sfn;
1696
1697 stp = td->td_sel;
1698 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1699 selfdfree(stp, sfp);
1700 stp->st_flags = 0;
1701 }
1702
1703 static void selectinit(void *);
1704 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1705 static void
1706 selectinit(void *dummy __unused)
1707 {
1708
1709 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1710 NULL, NULL, UMA_ALIGN_PTR, 0);
1711 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1712 }
Cache object: a7d4a67b6c3168b56da5d69ae786fcee
|