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: releng/9.0/sys/kern/sys_generic.c 228034 2011-11-27 19:00:52Z kib $");
39
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ktrace.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/capability.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/proc.h>
53 #include <sys/signalvar.h>
54 #include <sys/socketvar.h>
55 #include <sys/uio.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/poll.h>
61 #include <sys/resourcevar.h>
62 #include <sys/selinfo.h>
63 #include <sys/sleepqueue.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/bio.h>
69 #include <sys/buf.h>
70 #include <sys/condvar.h>
71 #ifdef KTRACE
72 #include <sys/ktrace.h>
73 #endif
74
75 #include <security/audit/audit.h>
76
77 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
78 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
79 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
80
81 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
82 u_int);
83 static int pollscan(struct thread *, struct pollfd *, u_int);
84 static int pollrescan(struct thread *);
85 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
86 static int selrescan(struct thread *, fd_mask **, fd_mask **);
87 static void selfdalloc(struct thread *, void *);
88 static void selfdfree(struct seltd *, struct selfd *);
89 static int dofileread(struct thread *, int, struct file *, struct uio *,
90 off_t, int);
91 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
92 off_t, int);
93 static void doselwakeup(struct selinfo *, int);
94 static void seltdinit(struct thread *);
95 static int seltdwait(struct thread *, int);
96 static void seltdclear(struct thread *);
97
98 /*
99 * One seltd per-thread allocated on demand as needed.
100 *
101 * t - protected by st_mtx
102 * k - Only accessed by curthread or read-only
103 */
104 struct seltd {
105 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
106 struct selfd *st_free1; /* (k) free fd for read set. */
107 struct selfd *st_free2; /* (k) free fd for write set. */
108 struct mtx st_mtx; /* Protects struct seltd */
109 struct cv st_wait; /* (t) Wait channel. */
110 int st_flags; /* (t) SELTD_ flags. */
111 };
112
113 #define SELTD_PENDING 0x0001 /* We have pending events. */
114 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
115
116 /*
117 * One selfd allocated per-thread per-file-descriptor.
118 * f - protected by sf_mtx
119 */
120 struct selfd {
121 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
122 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
123 struct selinfo *sf_si; /* (f) selinfo when linked. */
124 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
125 struct seltd *sf_td; /* (k) owning seltd. */
126 void *sf_cookie; /* (k) fd or pollfd. */
127 };
128
129 static uma_zone_t selfd_zone;
130 static struct mtx_pool *mtxpool_select;
131
132 #ifndef _SYS_SYSPROTO_H_
133 struct read_args {
134 int fd;
135 void *buf;
136 size_t nbyte;
137 };
138 #endif
139 int
140 sys_read(td, uap)
141 struct thread *td;
142 struct read_args *uap;
143 {
144 struct uio auio;
145 struct iovec aiov;
146 int error;
147
148 if (uap->nbyte > INT_MAX)
149 return (EINVAL);
150 aiov.iov_base = uap->buf;
151 aiov.iov_len = uap->nbyte;
152 auio.uio_iov = &aiov;
153 auio.uio_iovcnt = 1;
154 auio.uio_resid = uap->nbyte;
155 auio.uio_segflg = UIO_USERSPACE;
156 error = kern_readv(td, uap->fd, &auio);
157 return(error);
158 }
159
160 /*
161 * Positioned read system call
162 */
163 #ifndef _SYS_SYSPROTO_H_
164 struct pread_args {
165 int fd;
166 void *buf;
167 size_t nbyte;
168 int pad;
169 off_t offset;
170 };
171 #endif
172 int
173 sys_pread(td, uap)
174 struct thread *td;
175 struct pread_args *uap;
176 {
177 struct uio auio;
178 struct iovec aiov;
179 int error;
180
181 if (uap->nbyte > INT_MAX)
182 return (EINVAL);
183 aiov.iov_base = uap->buf;
184 aiov.iov_len = uap->nbyte;
185 auio.uio_iov = &aiov;
186 auio.uio_iovcnt = 1;
187 auio.uio_resid = uap->nbyte;
188 auio.uio_segflg = UIO_USERSPACE;
189 error = kern_preadv(td, uap->fd, &auio, uap->offset);
190 return(error);
191 }
192
193 int
194 freebsd6_pread(td, uap)
195 struct thread *td;
196 struct freebsd6_pread_args *uap;
197 {
198 struct pread_args oargs;
199
200 oargs.fd = uap->fd;
201 oargs.buf = uap->buf;
202 oargs.nbyte = uap->nbyte;
203 oargs.offset = uap->offset;
204 return (sys_pread(td, &oargs));
205 }
206
207 /*
208 * Scatter read system call.
209 */
210 #ifndef _SYS_SYSPROTO_H_
211 struct readv_args {
212 int fd;
213 struct iovec *iovp;
214 u_int iovcnt;
215 };
216 #endif
217 int
218 sys_readv(struct thread *td, struct readv_args *uap)
219 {
220 struct uio *auio;
221 int error;
222
223 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
224 if (error)
225 return (error);
226 error = kern_readv(td, uap->fd, auio);
227 free(auio, M_IOV);
228 return (error);
229 }
230
231 int
232 kern_readv(struct thread *td, int fd, struct uio *auio)
233 {
234 struct file *fp;
235 int error;
236
237 error = fget_read(td, fd, CAP_READ | CAP_SEEK, &fp);
238 if (error)
239 return (error);
240 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
241 fdrop(fp, td);
242 return (error);
243 }
244
245 /*
246 * Scatter positioned read system call.
247 */
248 #ifndef _SYS_SYSPROTO_H_
249 struct preadv_args {
250 int fd;
251 struct iovec *iovp;
252 u_int iovcnt;
253 off_t offset;
254 };
255 #endif
256 int
257 sys_preadv(struct thread *td, struct preadv_args *uap)
258 {
259 struct uio *auio;
260 int error;
261
262 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
263 if (error)
264 return (error);
265 error = kern_preadv(td, uap->fd, auio, uap->offset);
266 free(auio, M_IOV);
267 return (error);
268 }
269
270 int
271 kern_preadv(td, fd, auio, offset)
272 struct thread *td;
273 int fd;
274 struct uio *auio;
275 off_t offset;
276 {
277 struct file *fp;
278 int error;
279
280 error = fget_read(td, fd, CAP_READ, &fp);
281 if (error)
282 return (error);
283 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
284 error = ESPIPE;
285 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
286 error = EINVAL;
287 else
288 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
289 fdrop(fp, td);
290 return (error);
291 }
292
293 /*
294 * Common code for readv and preadv that reads data in
295 * from a file using the passed in uio, offset, and flags.
296 */
297 static int
298 dofileread(td, fd, fp, auio, offset, flags)
299 struct thread *td;
300 int fd;
301 struct file *fp;
302 struct uio *auio;
303 off_t offset;
304 int flags;
305 {
306 ssize_t cnt;
307 int error;
308 #ifdef KTRACE
309 struct uio *ktruio = NULL;
310 #endif
311
312 /* Finish zero length reads right here */
313 if (auio->uio_resid == 0) {
314 td->td_retval[0] = 0;
315 return(0);
316 }
317 auio->uio_rw = UIO_READ;
318 auio->uio_offset = offset;
319 auio->uio_td = td;
320 #ifdef KTRACE
321 if (KTRPOINT(td, KTR_GENIO))
322 ktruio = cloneuio(auio);
323 #endif
324 cnt = auio->uio_resid;
325 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
326 if (auio->uio_resid != cnt && (error == ERESTART ||
327 error == EINTR || error == EWOULDBLOCK))
328 error = 0;
329 }
330 cnt -= auio->uio_resid;
331 #ifdef KTRACE
332 if (ktruio != NULL) {
333 ktruio->uio_resid = cnt;
334 ktrgenio(fd, UIO_READ, ktruio, error);
335 }
336 #endif
337 td->td_retval[0] = cnt;
338 return (error);
339 }
340
341 #ifndef _SYS_SYSPROTO_H_
342 struct write_args {
343 int fd;
344 const void *buf;
345 size_t nbyte;
346 };
347 #endif
348 int
349 sys_write(td, uap)
350 struct thread *td;
351 struct write_args *uap;
352 {
353 struct uio auio;
354 struct iovec aiov;
355 int error;
356
357 if (uap->nbyte > INT_MAX)
358 return (EINVAL);
359 aiov.iov_base = (void *)(uintptr_t)uap->buf;
360 aiov.iov_len = uap->nbyte;
361 auio.uio_iov = &aiov;
362 auio.uio_iovcnt = 1;
363 auio.uio_resid = uap->nbyte;
364 auio.uio_segflg = UIO_USERSPACE;
365 error = kern_writev(td, uap->fd, &auio);
366 return(error);
367 }
368
369 /*
370 * Positioned write system call.
371 */
372 #ifndef _SYS_SYSPROTO_H_
373 struct pwrite_args {
374 int fd;
375 const void *buf;
376 size_t nbyte;
377 int pad;
378 off_t offset;
379 };
380 #endif
381 int
382 sys_pwrite(td, uap)
383 struct thread *td;
384 struct pwrite_args *uap;
385 {
386 struct uio auio;
387 struct iovec aiov;
388 int error;
389
390 if (uap->nbyte > INT_MAX)
391 return (EINVAL);
392 aiov.iov_base = (void *)(uintptr_t)uap->buf;
393 aiov.iov_len = uap->nbyte;
394 auio.uio_iov = &aiov;
395 auio.uio_iovcnt = 1;
396 auio.uio_resid = uap->nbyte;
397 auio.uio_segflg = UIO_USERSPACE;
398 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
399 return(error);
400 }
401
402 int
403 freebsd6_pwrite(td, uap)
404 struct thread *td;
405 struct freebsd6_pwrite_args *uap;
406 {
407 struct pwrite_args oargs;
408
409 oargs.fd = uap->fd;
410 oargs.buf = uap->buf;
411 oargs.nbyte = uap->nbyte;
412 oargs.offset = uap->offset;
413 return (sys_pwrite(td, &oargs));
414 }
415
416 /*
417 * Gather write system call.
418 */
419 #ifndef _SYS_SYSPROTO_H_
420 struct writev_args {
421 int fd;
422 struct iovec *iovp;
423 u_int iovcnt;
424 };
425 #endif
426 int
427 sys_writev(struct thread *td, struct writev_args *uap)
428 {
429 struct uio *auio;
430 int error;
431
432 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
433 if (error)
434 return (error);
435 error = kern_writev(td, uap->fd, auio);
436 free(auio, M_IOV);
437 return (error);
438 }
439
440 int
441 kern_writev(struct thread *td, int fd, struct uio *auio)
442 {
443 struct file *fp;
444 int error;
445
446 error = fget_write(td, fd, CAP_WRITE | CAP_SEEK, &fp);
447 if (error)
448 return (error);
449 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
450 fdrop(fp, td);
451 return (error);
452 }
453
454 /*
455 * Gather positioned write system call.
456 */
457 #ifndef _SYS_SYSPROTO_H_
458 struct pwritev_args {
459 int fd;
460 struct iovec *iovp;
461 u_int iovcnt;
462 off_t offset;
463 };
464 #endif
465 int
466 sys_pwritev(struct thread *td, struct pwritev_args *uap)
467 {
468 struct uio *auio;
469 int error;
470
471 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
472 if (error)
473 return (error);
474 error = kern_pwritev(td, uap->fd, auio, uap->offset);
475 free(auio, M_IOV);
476 return (error);
477 }
478
479 int
480 kern_pwritev(td, fd, auio, offset)
481 struct thread *td;
482 struct uio *auio;
483 int fd;
484 off_t offset;
485 {
486 struct file *fp;
487 int error;
488
489 error = fget_write(td, fd, CAP_WRITE, &fp);
490 if (error)
491 return (error);
492 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
493 error = ESPIPE;
494 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
495 error = EINVAL;
496 else
497 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
498 fdrop(fp, td);
499 return (error);
500 }
501
502 /*
503 * Common code for writev and pwritev that writes data to
504 * a file using the passed in uio, offset, and flags.
505 */
506 static int
507 dofilewrite(td, fd, fp, auio, offset, flags)
508 struct thread *td;
509 int fd;
510 struct file *fp;
511 struct uio *auio;
512 off_t offset;
513 int flags;
514 {
515 ssize_t cnt;
516 int error;
517 #ifdef KTRACE
518 struct uio *ktruio = NULL;
519 #endif
520
521 auio->uio_rw = UIO_WRITE;
522 auio->uio_td = td;
523 auio->uio_offset = offset;
524 #ifdef KTRACE
525 if (KTRPOINT(td, KTR_GENIO))
526 ktruio = cloneuio(auio);
527 #endif
528 cnt = auio->uio_resid;
529 if (fp->f_type == DTYPE_VNODE)
530 bwillwrite();
531 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
532 if (auio->uio_resid != cnt && (error == ERESTART ||
533 error == EINTR || error == EWOULDBLOCK))
534 error = 0;
535 /* Socket layer is responsible for issuing SIGPIPE. */
536 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
537 PROC_LOCK(td->td_proc);
538 tdsignal(td, SIGPIPE);
539 PROC_UNLOCK(td->td_proc);
540 }
541 }
542 cnt -= auio->uio_resid;
543 #ifdef KTRACE
544 if (ktruio != NULL) {
545 ktruio->uio_resid = cnt;
546 ktrgenio(fd, UIO_WRITE, ktruio, error);
547 }
548 #endif
549 td->td_retval[0] = cnt;
550 return (error);
551 }
552
553 /*
554 * Truncate a file given a file descriptor.
555 *
556 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
557 * descriptor isn't writable.
558 */
559 int
560 kern_ftruncate(td, fd, length)
561 struct thread *td;
562 int fd;
563 off_t length;
564 {
565 struct file *fp;
566 int error;
567
568 AUDIT_ARG_FD(fd);
569 if (length < 0)
570 return (EINVAL);
571 error = fget(td, fd, CAP_FTRUNCATE, &fp);
572 if (error)
573 return (error);
574 AUDIT_ARG_FILE(td->td_proc, fp);
575 if (!(fp->f_flag & FWRITE)) {
576 fdrop(fp, td);
577 return (EINVAL);
578 }
579 error = fo_truncate(fp, length, td->td_ucred, td);
580 fdrop(fp, td);
581 return (error);
582 }
583
584 #ifndef _SYS_SYSPROTO_H_
585 struct ftruncate_args {
586 int fd;
587 int pad;
588 off_t length;
589 };
590 #endif
591 int
592 sys_ftruncate(td, uap)
593 struct thread *td;
594 struct ftruncate_args *uap;
595 {
596
597 return (kern_ftruncate(td, uap->fd, uap->length));
598 }
599
600 #if defined(COMPAT_43)
601 #ifndef _SYS_SYSPROTO_H_
602 struct oftruncate_args {
603 int fd;
604 long length;
605 };
606 #endif
607 int
608 oftruncate(td, uap)
609 struct thread *td;
610 struct oftruncate_args *uap;
611 {
612
613 return (kern_ftruncate(td, uap->fd, uap->length));
614 }
615 #endif /* COMPAT_43 */
616
617 #ifndef _SYS_SYSPROTO_H_
618 struct ioctl_args {
619 int fd;
620 u_long com;
621 caddr_t data;
622 };
623 #endif
624 /* ARGSUSED */
625 int
626 sys_ioctl(struct thread *td, struct ioctl_args *uap)
627 {
628 u_long com;
629 int arg, error;
630 u_int size;
631 caddr_t data;
632
633 if (uap->com > 0xffffffff) {
634 printf(
635 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
636 td->td_proc->p_pid, td->td_name, uap->com);
637 uap->com &= 0xffffffff;
638 }
639 com = uap->com;
640
641 /*
642 * Interpret high order word to find amount of data to be
643 * copied to/from the user's address space.
644 */
645 size = IOCPARM_LEN(com);
646 if ((size > IOCPARM_MAX) ||
647 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
648 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
649 ((com & IOC_OUT) && size == 0) ||
650 #else
651 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
652 #endif
653 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
654 return (ENOTTY);
655
656 if (size > 0) {
657 if (com & IOC_VOID) {
658 /* Integer argument. */
659 arg = (intptr_t)uap->data;
660 data = (void *)&arg;
661 size = 0;
662 } else
663 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
664 } else
665 data = (void *)&uap->data;
666 if (com & IOC_IN) {
667 error = copyin(uap->data, data, (u_int)size);
668 if (error) {
669 if (size > 0)
670 free(data, M_IOCTLOPS);
671 return (error);
672 }
673 } else if (com & IOC_OUT) {
674 /*
675 * Zero the buffer so the user always
676 * gets back something deterministic.
677 */
678 bzero(data, size);
679 }
680
681 error = kern_ioctl(td, uap->fd, com, data);
682
683 if (error == 0 && (com & IOC_OUT))
684 error = copyout(data, uap->data, (u_int)size);
685
686 if (size > 0)
687 free(data, M_IOCTLOPS);
688 return (error);
689 }
690
691 int
692 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
693 {
694 struct file *fp;
695 struct filedesc *fdp;
696 int error;
697 int tmp;
698
699 AUDIT_ARG_FD(fd);
700 AUDIT_ARG_CMD(com);
701 if ((error = fget(td, fd, CAP_IOCTL, &fp)) != 0)
702 return (error);
703 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
704 fdrop(fp, td);
705 return (EBADF);
706 }
707 fdp = td->td_proc->p_fd;
708 switch (com) {
709 case FIONCLEX:
710 FILEDESC_XLOCK(fdp);
711 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
712 FILEDESC_XUNLOCK(fdp);
713 goto out;
714 case FIOCLEX:
715 FILEDESC_XLOCK(fdp);
716 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
717 FILEDESC_XUNLOCK(fdp);
718 goto out;
719 case FIONBIO:
720 if ((tmp = *(int *)data))
721 atomic_set_int(&fp->f_flag, FNONBLOCK);
722 else
723 atomic_clear_int(&fp->f_flag, FNONBLOCK);
724 data = (void *)&tmp;
725 break;
726 case FIOASYNC:
727 if ((tmp = *(int *)data))
728 atomic_set_int(&fp->f_flag, FASYNC);
729 else
730 atomic_clear_int(&fp->f_flag, FASYNC);
731 data = (void *)&tmp;
732 break;
733 }
734
735 error = fo_ioctl(fp, com, data, td->td_ucred, td);
736 out:
737 fdrop(fp, td);
738 return (error);
739 }
740
741 int
742 poll_no_poll(int events)
743 {
744 /*
745 * Return true for read/write. If the user asked for something
746 * special, return POLLNVAL, so that clients have a way of
747 * determining reliably whether or not the extended
748 * functionality is present without hard-coding knowledge
749 * of specific filesystem implementations.
750 */
751 if (events & ~POLLSTANDARD)
752 return (POLLNVAL);
753
754 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
755 }
756
757 int
758 sys_pselect(struct thread *td, struct pselect_args *uap)
759 {
760 struct timespec ts;
761 struct timeval tv, *tvp;
762 sigset_t set, *uset;
763 int error;
764
765 if (uap->ts != NULL) {
766 error = copyin(uap->ts, &ts, sizeof(ts));
767 if (error != 0)
768 return (error);
769 TIMESPEC_TO_TIMEVAL(&tv, &ts);
770 tvp = &tv;
771 } else
772 tvp = NULL;
773 if (uap->sm != NULL) {
774 error = copyin(uap->sm, &set, sizeof(set));
775 if (error != 0)
776 return (error);
777 uset = &set;
778 } else
779 uset = NULL;
780 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
781 uset, NFDBITS));
782 }
783
784 int
785 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
786 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
787 {
788 int error;
789
790 if (uset != NULL) {
791 error = kern_sigprocmask(td, SIG_SETMASK, uset,
792 &td->td_oldsigmask, 0);
793 if (error != 0)
794 return (error);
795 td->td_pflags |= TDP_OLDMASK;
796 /*
797 * Make sure that ast() is called on return to
798 * usermode and TDP_OLDMASK is cleared, restoring old
799 * sigmask.
800 */
801 thread_lock(td);
802 td->td_flags |= TDF_ASTPENDING;
803 thread_unlock(td);
804 }
805 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
806 return (error);
807 }
808
809 #ifndef _SYS_SYSPROTO_H_
810 struct select_args {
811 int nd;
812 fd_set *in, *ou, *ex;
813 struct timeval *tv;
814 };
815 #endif
816 int
817 sys_select(struct thread *td, struct select_args *uap)
818 {
819 struct timeval tv, *tvp;
820 int error;
821
822 if (uap->tv != NULL) {
823 error = copyin(uap->tv, &tv, sizeof(tv));
824 if (error)
825 return (error);
826 tvp = &tv;
827 } else
828 tvp = NULL;
829
830 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
831 NFDBITS));
832 }
833
834 /*
835 * In the unlikely case when user specified n greater then the last
836 * open file descriptor, check that no bits are set after the last
837 * valid fd. We must return EBADF if any is set.
838 *
839 * There are applications that rely on the behaviour.
840 *
841 * nd is fd_lastfile + 1.
842 */
843 static int
844 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
845 {
846 char *addr, *oaddr;
847 int b, i, res;
848 uint8_t bits;
849
850 if (nd >= ndu || fd_in == NULL)
851 return (0);
852
853 oaddr = NULL;
854 bits = 0; /* silence gcc */
855 for (i = nd; i < ndu; i++) {
856 b = i / NBBY;
857 #if BYTE_ORDER == LITTLE_ENDIAN
858 addr = (char *)fd_in + b;
859 #else
860 addr = (char *)fd_in;
861 if (abi_nfdbits == NFDBITS) {
862 addr += rounddown(b, sizeof(fd_mask)) +
863 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
864 } else {
865 addr += rounddown(b, sizeof(uint32_t)) +
866 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
867 }
868 #endif
869 if (addr != oaddr) {
870 res = fubyte(addr);
871 if (res == -1)
872 return (EFAULT);
873 oaddr = addr;
874 bits = res;
875 }
876 if ((bits & (1 << (i % NBBY))) != 0)
877 return (EBADF);
878 }
879 return (0);
880 }
881
882 int
883 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
884 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
885 {
886 struct filedesc *fdp;
887 /*
888 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
889 * infds with the new FD_SETSIZE of 1024, and more than enough for
890 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
891 * of 256.
892 */
893 fd_mask s_selbits[howmany(2048, NFDBITS)];
894 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
895 struct timeval atv, rtv, ttv;
896 int error, lf, ndu, timo;
897 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
898
899 if (nd < 0)
900 return (EINVAL);
901 fdp = td->td_proc->p_fd;
902 ndu = nd;
903 lf = fdp->fd_lastfile;
904 if (nd > lf + 1)
905 nd = lf + 1;
906
907 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
908 if (error != 0)
909 return (error);
910 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
911 if (error != 0)
912 return (error);
913 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
914 if (error != 0)
915 return (error);
916
917 /*
918 * Allocate just enough bits for the non-null fd_sets. Use the
919 * preallocated auto buffer if possible.
920 */
921 nfdbits = roundup(nd, NFDBITS);
922 ncpbytes = nfdbits / NBBY;
923 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
924 nbufbytes = 0;
925 if (fd_in != NULL)
926 nbufbytes += 2 * ncpbytes;
927 if (fd_ou != NULL)
928 nbufbytes += 2 * ncpbytes;
929 if (fd_ex != NULL)
930 nbufbytes += 2 * ncpbytes;
931 if (nbufbytes <= sizeof s_selbits)
932 selbits = &s_selbits[0];
933 else
934 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
935
936 /*
937 * Assign pointers into the bit buffers and fetch the input bits.
938 * Put the output buffers together so that they can be bzeroed
939 * together.
940 */
941 sbp = selbits;
942 #define getbits(name, x) \
943 do { \
944 if (name == NULL) { \
945 ibits[x] = NULL; \
946 obits[x] = NULL; \
947 } else { \
948 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
949 obits[x] = sbp; \
950 sbp += ncpbytes / sizeof *sbp; \
951 error = copyin(name, ibits[x], ncpubytes); \
952 if (error != 0) \
953 goto done; \
954 bzero((char *)ibits[x] + ncpubytes, \
955 ncpbytes - ncpubytes); \
956 } \
957 } while (0)
958 getbits(fd_in, 0);
959 getbits(fd_ou, 1);
960 getbits(fd_ex, 2);
961 #undef getbits
962
963 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
964 /*
965 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
966 * we are running under 32-bit emulation. This should be more
967 * generic.
968 */
969 #define swizzle_fdset(bits) \
970 if (abi_nfdbits != NFDBITS && bits != NULL) { \
971 int i; \
972 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
973 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
974 }
975 #else
976 #define swizzle_fdset(bits)
977 #endif
978
979 /* Make sure the bit order makes it through an ABI transition */
980 swizzle_fdset(ibits[0]);
981 swizzle_fdset(ibits[1]);
982 swizzle_fdset(ibits[2]);
983
984 if (nbufbytes != 0)
985 bzero(selbits, nbufbytes / 2);
986
987 if (tvp != NULL) {
988 atv = *tvp;
989 if (itimerfix(&atv)) {
990 error = EINVAL;
991 goto done;
992 }
993 getmicrouptime(&rtv);
994 timevaladd(&atv, &rtv);
995 } else {
996 atv.tv_sec = 0;
997 atv.tv_usec = 0;
998 }
999 timo = 0;
1000 seltdinit(td);
1001 /* Iterate until the timeout expires or descriptors become ready. */
1002 for (;;) {
1003 error = selscan(td, ibits, obits, nd);
1004 if (error || td->td_retval[0] != 0)
1005 break;
1006 if (atv.tv_sec || atv.tv_usec) {
1007 getmicrouptime(&rtv);
1008 if (timevalcmp(&rtv, &atv, >=))
1009 break;
1010 ttv = atv;
1011 timevalsub(&ttv, &rtv);
1012 timo = ttv.tv_sec > 24 * 60 * 60 ?
1013 24 * 60 * 60 * hz : tvtohz(&ttv);
1014 }
1015 error = seltdwait(td, timo);
1016 if (error)
1017 break;
1018 error = selrescan(td, ibits, obits);
1019 if (error || td->td_retval[0] != 0)
1020 break;
1021 }
1022 seltdclear(td);
1023
1024 done:
1025 /* select is not restarted after signals... */
1026 if (error == ERESTART)
1027 error = EINTR;
1028 if (error == EWOULDBLOCK)
1029 error = 0;
1030
1031 /* swizzle bit order back, if necessary */
1032 swizzle_fdset(obits[0]);
1033 swizzle_fdset(obits[1]);
1034 swizzle_fdset(obits[2]);
1035 #undef swizzle_fdset
1036
1037 #define putbits(name, x) \
1038 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1039 error = error2;
1040 if (error == 0) {
1041 int error2;
1042
1043 putbits(fd_in, 0);
1044 putbits(fd_ou, 1);
1045 putbits(fd_ex, 2);
1046 #undef putbits
1047 }
1048 if (selbits != &s_selbits[0])
1049 free(selbits, M_SELECT);
1050
1051 return (error);
1052 }
1053 /*
1054 * Convert a select bit set to poll flags.
1055 *
1056 * The backend always returns POLLHUP/POLLERR if appropriate and we
1057 * return this as a set bit in any set.
1058 */
1059 static int select_flags[3] = {
1060 POLLRDNORM | POLLHUP | POLLERR,
1061 POLLWRNORM | POLLHUP | POLLERR,
1062 POLLRDBAND | POLLERR
1063 };
1064
1065 /*
1066 * Compute the fo_poll flags required for a fd given by the index and
1067 * bit position in the fd_mask array.
1068 */
1069 static __inline int
1070 selflags(fd_mask **ibits, int idx, fd_mask bit)
1071 {
1072 int flags;
1073 int msk;
1074
1075 flags = 0;
1076 for (msk = 0; msk < 3; msk++) {
1077 if (ibits[msk] == NULL)
1078 continue;
1079 if ((ibits[msk][idx] & bit) == 0)
1080 continue;
1081 flags |= select_flags[msk];
1082 }
1083 return (flags);
1084 }
1085
1086 /*
1087 * Set the appropriate output bits given a mask of fired events and the
1088 * input bits originally requested.
1089 */
1090 static __inline int
1091 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1092 {
1093 int msk;
1094 int n;
1095
1096 n = 0;
1097 for (msk = 0; msk < 3; msk++) {
1098 if ((events & select_flags[msk]) == 0)
1099 continue;
1100 if (ibits[msk] == NULL)
1101 continue;
1102 if ((ibits[msk][idx] & bit) == 0)
1103 continue;
1104 /*
1105 * XXX Check for a duplicate set. This can occur because a
1106 * socket calls selrecord() twice for each poll() call
1107 * resulting in two selfds per real fd. selrescan() will
1108 * call selsetbits twice as a result.
1109 */
1110 if ((obits[msk][idx] & bit) != 0)
1111 continue;
1112 obits[msk][idx] |= bit;
1113 n++;
1114 }
1115
1116 return (n);
1117 }
1118
1119 static __inline int
1120 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1121 {
1122 struct file *fp;
1123 #ifdef CAPABILITIES
1124 struct file *fp_fromcap;
1125 int error;
1126 #endif
1127
1128 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1129 return (EBADF);
1130 #ifdef CAPABILITIES
1131 /*
1132 * If the file descriptor is for a capability, test rights and use
1133 * the file descriptor references by the capability.
1134 */
1135 error = cap_funwrap(fp, CAP_POLL_EVENT, &fp_fromcap);
1136 if (error) {
1137 fdrop(fp, curthread);
1138 return (error);
1139 }
1140 if (fp != fp_fromcap) {
1141 fhold(fp_fromcap);
1142 fdrop(fp, curthread);
1143 fp = fp_fromcap;
1144 }
1145 #endif /* CAPABILITIES */
1146 *fpp = fp;
1147 return (0);
1148 }
1149
1150 /*
1151 * Traverse the list of fds attached to this thread's seltd and check for
1152 * completion.
1153 */
1154 static int
1155 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1156 {
1157 struct filedesc *fdp;
1158 struct selinfo *si;
1159 struct seltd *stp;
1160 struct selfd *sfp;
1161 struct selfd *sfn;
1162 struct file *fp;
1163 fd_mask bit;
1164 int fd, ev, n, idx;
1165 int error;
1166
1167 fdp = td->td_proc->p_fd;
1168 stp = td->td_sel;
1169 n = 0;
1170 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1171 fd = (int)(uintptr_t)sfp->sf_cookie;
1172 si = sfp->sf_si;
1173 selfdfree(stp, sfp);
1174 /* If the selinfo wasn't cleared the event didn't fire. */
1175 if (si != NULL)
1176 continue;
1177 error = getselfd_cap(fdp, fd, &fp);
1178 if (error)
1179 return (error);
1180 idx = fd / NFDBITS;
1181 bit = (fd_mask)1 << (fd % NFDBITS);
1182 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1183 fdrop(fp, td);
1184 if (ev != 0)
1185 n += selsetbits(ibits, obits, idx, bit, ev);
1186 }
1187 stp->st_flags = 0;
1188 td->td_retval[0] = n;
1189 return (0);
1190 }
1191
1192 /*
1193 * Perform the initial filedescriptor scan and register ourselves with
1194 * each selinfo.
1195 */
1196 static int
1197 selscan(td, ibits, obits, nfd)
1198 struct thread *td;
1199 fd_mask **ibits, **obits;
1200 int nfd;
1201 {
1202 struct filedesc *fdp;
1203 struct file *fp;
1204 fd_mask bit;
1205 int ev, flags, end, fd;
1206 int n, idx;
1207 int error;
1208
1209 fdp = td->td_proc->p_fd;
1210 n = 0;
1211 for (idx = 0, fd = 0; fd < nfd; idx++) {
1212 end = imin(fd + NFDBITS, nfd);
1213 for (bit = 1; fd < end; bit <<= 1, fd++) {
1214 /* Compute the list of events we're interested in. */
1215 flags = selflags(ibits, idx, bit);
1216 if (flags == 0)
1217 continue;
1218 error = getselfd_cap(fdp, fd, &fp);
1219 if (error)
1220 return (error);
1221 selfdalloc(td, (void *)(uintptr_t)fd);
1222 ev = fo_poll(fp, flags, td->td_ucred, td);
1223 fdrop(fp, td);
1224 if (ev != 0)
1225 n += selsetbits(ibits, obits, idx, bit, ev);
1226 }
1227 }
1228
1229 td->td_retval[0] = n;
1230 return (0);
1231 }
1232
1233 #ifndef _SYS_SYSPROTO_H_
1234 struct poll_args {
1235 struct pollfd *fds;
1236 u_int nfds;
1237 int timeout;
1238 };
1239 #endif
1240 int
1241 sys_poll(td, uap)
1242 struct thread *td;
1243 struct poll_args *uap;
1244 {
1245 struct pollfd *bits;
1246 struct pollfd smallbits[32];
1247 struct timeval atv, rtv, ttv;
1248 int error = 0, timo;
1249 u_int nfds;
1250 size_t ni;
1251
1252 nfds = uap->nfds;
1253 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1254 return (EINVAL);
1255 ni = nfds * sizeof(struct pollfd);
1256 if (ni > sizeof(smallbits))
1257 bits = malloc(ni, M_TEMP, M_WAITOK);
1258 else
1259 bits = smallbits;
1260 error = copyin(uap->fds, bits, ni);
1261 if (error)
1262 goto done;
1263 if (uap->timeout != INFTIM) {
1264 atv.tv_sec = uap->timeout / 1000;
1265 atv.tv_usec = (uap->timeout % 1000) * 1000;
1266 if (itimerfix(&atv)) {
1267 error = EINVAL;
1268 goto done;
1269 }
1270 getmicrouptime(&rtv);
1271 timevaladd(&atv, &rtv);
1272 } else {
1273 atv.tv_sec = 0;
1274 atv.tv_usec = 0;
1275 }
1276 timo = 0;
1277 seltdinit(td);
1278 /* Iterate until the timeout expires or descriptors become ready. */
1279 for (;;) {
1280 error = pollscan(td, bits, nfds);
1281 if (error || td->td_retval[0] != 0)
1282 break;
1283 if (atv.tv_sec || atv.tv_usec) {
1284 getmicrouptime(&rtv);
1285 if (timevalcmp(&rtv, &atv, >=))
1286 break;
1287 ttv = atv;
1288 timevalsub(&ttv, &rtv);
1289 timo = ttv.tv_sec > 24 * 60 * 60 ?
1290 24 * 60 * 60 * hz : tvtohz(&ttv);
1291 }
1292 error = seltdwait(td, timo);
1293 if (error)
1294 break;
1295 error = pollrescan(td);
1296 if (error || td->td_retval[0] != 0)
1297 break;
1298 }
1299 seltdclear(td);
1300
1301 done:
1302 /* poll is not restarted after signals... */
1303 if (error == ERESTART)
1304 error = EINTR;
1305 if (error == EWOULDBLOCK)
1306 error = 0;
1307 if (error == 0) {
1308 error = pollout(td, bits, uap->fds, nfds);
1309 if (error)
1310 goto out;
1311 }
1312 out:
1313 if (ni > sizeof(smallbits))
1314 free(bits, M_TEMP);
1315 return (error);
1316 }
1317
1318 static int
1319 pollrescan(struct thread *td)
1320 {
1321 struct seltd *stp;
1322 struct selfd *sfp;
1323 struct selfd *sfn;
1324 struct selinfo *si;
1325 struct filedesc *fdp;
1326 struct file *fp;
1327 struct pollfd *fd;
1328 int n;
1329
1330 n = 0;
1331 fdp = td->td_proc->p_fd;
1332 stp = td->td_sel;
1333 FILEDESC_SLOCK(fdp);
1334 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1335 fd = (struct pollfd *)sfp->sf_cookie;
1336 si = sfp->sf_si;
1337 selfdfree(stp, sfp);
1338 /* If the selinfo wasn't cleared the event didn't fire. */
1339 if (si != NULL)
1340 continue;
1341 fp = fdp->fd_ofiles[fd->fd];
1342 #ifdef CAPABILITIES
1343 if ((fp == NULL)
1344 || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
1345 #else
1346 if (fp == NULL) {
1347 #endif
1348 fd->revents = POLLNVAL;
1349 n++;
1350 continue;
1351 }
1352
1353 /*
1354 * Note: backend also returns POLLHUP and
1355 * POLLERR if appropriate.
1356 */
1357 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1358 if (fd->revents != 0)
1359 n++;
1360 }
1361 FILEDESC_SUNLOCK(fdp);
1362 stp->st_flags = 0;
1363 td->td_retval[0] = n;
1364 return (0);
1365 }
1366
1367
1368 static int
1369 pollout(td, fds, ufds, nfd)
1370 struct thread *td;
1371 struct pollfd *fds;
1372 struct pollfd *ufds;
1373 u_int nfd;
1374 {
1375 int error = 0;
1376 u_int i = 0;
1377 u_int n = 0;
1378
1379 for (i = 0; i < nfd; i++) {
1380 error = copyout(&fds->revents, &ufds->revents,
1381 sizeof(ufds->revents));
1382 if (error)
1383 return (error);
1384 if (fds->revents != 0)
1385 n++;
1386 fds++;
1387 ufds++;
1388 }
1389 td->td_retval[0] = n;
1390 return (0);
1391 }
1392
1393 static int
1394 pollscan(td, fds, nfd)
1395 struct thread *td;
1396 struct pollfd *fds;
1397 u_int nfd;
1398 {
1399 struct filedesc *fdp = td->td_proc->p_fd;
1400 int i;
1401 struct file *fp;
1402 int n = 0;
1403
1404 FILEDESC_SLOCK(fdp);
1405 for (i = 0; i < nfd; i++, fds++) {
1406 if (fds->fd >= fdp->fd_nfiles) {
1407 fds->revents = POLLNVAL;
1408 n++;
1409 } else if (fds->fd < 0) {
1410 fds->revents = 0;
1411 } else {
1412 fp = fdp->fd_ofiles[fds->fd];
1413 #ifdef CAPABILITIES
1414 if ((fp == NULL)
1415 || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
1416 #else
1417 if (fp == NULL) {
1418 #endif
1419 fds->revents = POLLNVAL;
1420 n++;
1421 } else {
1422 /*
1423 * Note: backend also returns POLLHUP and
1424 * POLLERR if appropriate.
1425 */
1426 selfdalloc(td, fds);
1427 fds->revents = fo_poll(fp, fds->events,
1428 td->td_ucred, td);
1429 /*
1430 * POSIX requires POLLOUT to be never
1431 * set simultaneously with POLLHUP.
1432 */
1433 if ((fds->revents & POLLHUP) != 0)
1434 fds->revents &= ~POLLOUT;
1435
1436 if (fds->revents != 0)
1437 n++;
1438 }
1439 }
1440 }
1441 FILEDESC_SUNLOCK(fdp);
1442 td->td_retval[0] = n;
1443 return (0);
1444 }
1445
1446 /*
1447 * OpenBSD poll system call.
1448 *
1449 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1450 */
1451 #ifndef _SYS_SYSPROTO_H_
1452 struct openbsd_poll_args {
1453 struct pollfd *fds;
1454 u_int nfds;
1455 int timeout;
1456 };
1457 #endif
1458 int
1459 sys_openbsd_poll(td, uap)
1460 register struct thread *td;
1461 register struct openbsd_poll_args *uap;
1462 {
1463 return (sys_poll(td, (struct poll_args *)uap));
1464 }
1465
1466 /*
1467 * XXX This was created specifically to support netncp and netsmb. This
1468 * allows the caller to specify a socket to wait for events on. It returns
1469 * 0 if any events matched and an error otherwise. There is no way to
1470 * determine which events fired.
1471 */
1472 int
1473 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1474 {
1475 struct timeval atv, rtv, ttv;
1476 int error, timo;
1477
1478 if (tvp != NULL) {
1479 atv = *tvp;
1480 if (itimerfix(&atv))
1481 return (EINVAL);
1482 getmicrouptime(&rtv);
1483 timevaladd(&atv, &rtv);
1484 } else {
1485 atv.tv_sec = 0;
1486 atv.tv_usec = 0;
1487 }
1488
1489 timo = 0;
1490 seltdinit(td);
1491 /*
1492 * Iterate until the timeout expires or the socket becomes ready.
1493 */
1494 for (;;) {
1495 selfdalloc(td, NULL);
1496 error = sopoll(so, events, NULL, td);
1497 /* error here is actually the ready events. */
1498 if (error)
1499 return (0);
1500 if (atv.tv_sec || atv.tv_usec) {
1501 getmicrouptime(&rtv);
1502 if (timevalcmp(&rtv, &atv, >=)) {
1503 seltdclear(td);
1504 return (EWOULDBLOCK);
1505 }
1506 ttv = atv;
1507 timevalsub(&ttv, &rtv);
1508 timo = ttv.tv_sec > 24 * 60 * 60 ?
1509 24 * 60 * 60 * hz : tvtohz(&ttv);
1510 }
1511 error = seltdwait(td, timo);
1512 seltdclear(td);
1513 if (error)
1514 break;
1515 }
1516 /* XXX Duplicates ncp/smb behavior. */
1517 if (error == ERESTART)
1518 error = 0;
1519 return (error);
1520 }
1521
1522 /*
1523 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1524 * have two select sets, one for read and another for write.
1525 */
1526 static void
1527 selfdalloc(struct thread *td, void *cookie)
1528 {
1529 struct seltd *stp;
1530
1531 stp = td->td_sel;
1532 if (stp->st_free1 == NULL)
1533 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1534 stp->st_free1->sf_td = stp;
1535 stp->st_free1->sf_cookie = cookie;
1536 if (stp->st_free2 == NULL)
1537 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1538 stp->st_free2->sf_td = stp;
1539 stp->st_free2->sf_cookie = cookie;
1540 }
1541
1542 static void
1543 selfdfree(struct seltd *stp, struct selfd *sfp)
1544 {
1545 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1546 mtx_lock(sfp->sf_mtx);
1547 if (sfp->sf_si)
1548 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1549 mtx_unlock(sfp->sf_mtx);
1550 uma_zfree(selfd_zone, sfp);
1551 }
1552
1553 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1554 void
1555 seldrain(sip)
1556 struct selinfo *sip;
1557 {
1558
1559 /*
1560 * This feature is already provided by doselwakeup(), thus it is
1561 * enough to go for it.
1562 * Eventually, the context, should take care to avoid races
1563 * between thread calling select()/poll() and file descriptor
1564 * detaching, but, again, the races are just the same as
1565 * selwakeup().
1566 */
1567 doselwakeup(sip, -1);
1568 }
1569
1570 /*
1571 * Record a select request.
1572 */
1573 void
1574 selrecord(selector, sip)
1575 struct thread *selector;
1576 struct selinfo *sip;
1577 {
1578 struct selfd *sfp;
1579 struct seltd *stp;
1580 struct mtx *mtxp;
1581
1582 stp = selector->td_sel;
1583 /*
1584 * Don't record when doing a rescan.
1585 */
1586 if (stp->st_flags & SELTD_RESCAN)
1587 return;
1588 /*
1589 * Grab one of the preallocated descriptors.
1590 */
1591 sfp = NULL;
1592 if ((sfp = stp->st_free1) != NULL)
1593 stp->st_free1 = NULL;
1594 else if ((sfp = stp->st_free2) != NULL)
1595 stp->st_free2 = NULL;
1596 else
1597 panic("selrecord: No free selfd on selq");
1598 mtxp = sip->si_mtx;
1599 if (mtxp == NULL)
1600 mtxp = mtx_pool_find(mtxpool_select, sip);
1601 /*
1602 * Initialize the sfp and queue it in the thread.
1603 */
1604 sfp->sf_si = sip;
1605 sfp->sf_mtx = mtxp;
1606 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1607 /*
1608 * Now that we've locked the sip, check for initialization.
1609 */
1610 mtx_lock(mtxp);
1611 if (sip->si_mtx == NULL) {
1612 sip->si_mtx = mtxp;
1613 TAILQ_INIT(&sip->si_tdlist);
1614 }
1615 /*
1616 * Add this thread to the list of selfds listening on this selinfo.
1617 */
1618 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1619 mtx_unlock(sip->si_mtx);
1620 }
1621
1622 /* Wake up a selecting thread. */
1623 void
1624 selwakeup(sip)
1625 struct selinfo *sip;
1626 {
1627 doselwakeup(sip, -1);
1628 }
1629
1630 /* Wake up a selecting thread, and set its priority. */
1631 void
1632 selwakeuppri(sip, pri)
1633 struct selinfo *sip;
1634 int pri;
1635 {
1636 doselwakeup(sip, pri);
1637 }
1638
1639 /*
1640 * Do a wakeup when a selectable event occurs.
1641 */
1642 static void
1643 doselwakeup(sip, pri)
1644 struct selinfo *sip;
1645 int pri;
1646 {
1647 struct selfd *sfp;
1648 struct selfd *sfn;
1649 struct seltd *stp;
1650
1651 /* If it's not initialized there can't be any waiters. */
1652 if (sip->si_mtx == NULL)
1653 return;
1654 /*
1655 * Locking the selinfo locks all selfds associated with it.
1656 */
1657 mtx_lock(sip->si_mtx);
1658 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1659 /*
1660 * Once we remove this sfp from the list and clear the
1661 * sf_si seltdclear will know to ignore this si.
1662 */
1663 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1664 sfp->sf_si = NULL;
1665 stp = sfp->sf_td;
1666 mtx_lock(&stp->st_mtx);
1667 stp->st_flags |= SELTD_PENDING;
1668 cv_broadcastpri(&stp->st_wait, pri);
1669 mtx_unlock(&stp->st_mtx);
1670 }
1671 mtx_unlock(sip->si_mtx);
1672 }
1673
1674 static void
1675 seltdinit(struct thread *td)
1676 {
1677 struct seltd *stp;
1678
1679 if ((stp = td->td_sel) != NULL)
1680 goto out;
1681 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1682 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1683 cv_init(&stp->st_wait, "select");
1684 out:
1685 stp->st_flags = 0;
1686 STAILQ_INIT(&stp->st_selq);
1687 }
1688
1689 static int
1690 seltdwait(struct thread *td, int timo)
1691 {
1692 struct seltd *stp;
1693 int error;
1694
1695 stp = td->td_sel;
1696 /*
1697 * An event of interest may occur while we do not hold the seltd
1698 * locked so check the pending flag before we sleep.
1699 */
1700 mtx_lock(&stp->st_mtx);
1701 /*
1702 * Any further calls to selrecord will be a rescan.
1703 */
1704 stp->st_flags |= SELTD_RESCAN;
1705 if (stp->st_flags & SELTD_PENDING) {
1706 mtx_unlock(&stp->st_mtx);
1707 return (0);
1708 }
1709 if (timo > 0)
1710 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1711 else
1712 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1713 mtx_unlock(&stp->st_mtx);
1714
1715 return (error);
1716 }
1717
1718 void
1719 seltdfini(struct thread *td)
1720 {
1721 struct seltd *stp;
1722
1723 stp = td->td_sel;
1724 if (stp == NULL)
1725 return;
1726 if (stp->st_free1)
1727 uma_zfree(selfd_zone, stp->st_free1);
1728 if (stp->st_free2)
1729 uma_zfree(selfd_zone, stp->st_free2);
1730 td->td_sel = NULL;
1731 free(stp, M_SELECT);
1732 }
1733
1734 /*
1735 * Remove the references to the thread from all of the objects we were
1736 * polling.
1737 */
1738 static void
1739 seltdclear(struct thread *td)
1740 {
1741 struct seltd *stp;
1742 struct selfd *sfp;
1743 struct selfd *sfn;
1744
1745 stp = td->td_sel;
1746 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1747 selfdfree(stp, sfp);
1748 stp->st_flags = 0;
1749 }
1750
1751 static void selectinit(void *);
1752 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1753 static void
1754 selectinit(void *dummy __unused)
1755 {
1756
1757 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1758 NULL, NULL, UMA_ALIGN_PTR, 0);
1759 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1760 }
Cache object: 9df3e289e95ba6e1cf39dd061b425e84
|