?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Ignore machine-check MSRs
[freebsd-src/fkvm-freebsd.git] / sys / kern / sys_generic.c
blob2f5b2db277c777daa16ab140c2c78ab06f5eb334
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 pollscan(struct thread *, struct pollfd *, u_int);
80 static int pollrescan(struct thread *);
81 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
82 static int selrescan(struct thread *, fd_mask **, fd_mask **);
83 static void selfdalloc(struct thread *, void *);
84 static void selfdfree(struct seltd *, struct selfd *);
85 static int dofileread(struct thread *, int, struct file *, struct uio *,
86 off_t, int);
87 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
88 off_t, int);
89 static void doselwakeup(struct selinfo *, int);
90 static void seltdinit(struct thread *);
91 static int seltdwait(struct thread *, int);
92 static void seltdclear(struct thread *);
93
94 /*
95 * One seltd per-thread allocated on demand as needed.
96 *
97 * t - protected by st_mtx
98 * k - Only accessed by curthread or read-only
99 */
100 struct seltd {
101 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
102 struct selfd *st_free1; /* (k) free fd for read set. */
103 struct selfd *st_free2; /* (k) free fd for write set. */
104 struct mtx st_mtx; /* Protects struct seltd */
105 struct cv st_wait; /* (t) Wait channel. */
106 int st_flags; /* (t) SELTD_ flags. */
107 };
108
109 #define SELTD_PENDING 0x0001 /* We have pending events. */
110 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
111
112 /*
113 * One selfd allocated per-thread per-file-descriptor.
114 * f - protected by sf_mtx
115 */
116 struct selfd {
117 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
118 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
119 struct selinfo *sf_si; /* (f) selinfo when linked. */
120 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
121 struct seltd *sf_td; /* (k) owning seltd. */
122 void *sf_cookie; /* (k) fd or pollfd. */
123 };
124
125 static uma_zone_t selfd_zone;
126
127 #ifndef _SYS_SYSPROTO_H_
128 struct read_args {
129 int fd;
130 void *buf;
131 size_t nbyte;
132 };
133 #endif
134 int
135 read(td, uap)
136 struct thread *td;
137 struct read_args *uap;
138 {
139 struct uio auio;
140 struct iovec aiov;
141 int error;
142
143 if (uap->nbyte > INT_MAX)
144 return (EINVAL);
145 aiov.iov_base = uap->buf;
146 aiov.iov_len = uap->nbyte;
147 auio.uio_iov = &aiov;
148 auio.uio_iovcnt = 1;
149 auio.uio_resid = uap->nbyte;
150 auio.uio_segflg = UIO_USERSPACE;
151 error = kern_readv(td, uap->fd, &auio);
152 return(error);
153 }
154
155 /*
156 * Positioned read system call
157 */
158 #ifndef _SYS_SYSPROTO_H_
159 struct pread_args {
160 int fd;
161 void *buf;
162 size_t nbyte;
163 int pad;
164 off_t offset;
165 };
166 #endif
167 int
168 pread(td, uap)
169 struct thread *td;
170 struct pread_args *uap;
171 {
172 struct uio auio;
173 struct iovec aiov;
174 int error;
175
176 if (uap->nbyte > INT_MAX)
177 return (EINVAL);
178 aiov.iov_base = uap->buf;
179 aiov.iov_len = uap->nbyte;
180 auio.uio_iov = &aiov;
181 auio.uio_iovcnt = 1;
182 auio.uio_resid = uap->nbyte;
183 auio.uio_segflg = UIO_USERSPACE;
184 error = kern_preadv(td, uap->fd, &auio, uap->offset);
185 return(error);
186 }
187
188 int
189 freebsd6_pread(td, uap)
190 struct thread *td;
191 struct freebsd6_pread_args *uap;
192 {
193 struct pread_args oargs;
194
195 oargs.fd = uap->fd;
196 oargs.buf = uap->buf;
197 oargs.nbyte = uap->nbyte;
198 oargs.offset = uap->offset;
199 return (pread(td, &oargs));
200 }
201
202 /*
203 * Scatter read system call.
204 */
205 #ifndef _SYS_SYSPROTO_H_
206 struct readv_args {
207 int fd;
208 struct iovec *iovp;
209 u_int iovcnt;
210 };
211 #endif
212 int
213 readv(struct thread *td, struct readv_args *uap)
214 {
215 struct uio *auio;
216 int error;
217
218 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
219 if (error)
220 return (error);
221 error = kern_readv(td, uap->fd, auio);
222 free(auio, M_IOV);
223 return (error);
224 }
225
226 int
227 kern_readv(struct thread *td, int fd, struct uio *auio)
228 {
229 struct file *fp;
230 int error;
231
232 error = fget_read(td, fd, &fp);
233 if (error)
234 return (error);
235 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
236 fdrop(fp, td);
237 return (error);
238 }
239
240 /*
241 * Scatter positioned read system call.
242 */
243 #ifndef _SYS_SYSPROTO_H_
244 struct preadv_args {
245 int fd;
246 struct iovec *iovp;
247 u_int iovcnt;
248 off_t offset;
249 };
250 #endif
251 int
252 preadv(struct thread *td, struct preadv_args *uap)
253 {
254 struct uio *auio;
255 int error;
256
257 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
258 if (error)
259 return (error);
260 error = kern_preadv(td, uap->fd, auio, uap->offset);
261 free(auio, M_IOV);
262 return (error);
263 }
264
265 int
266 kern_preadv(td, fd, auio, offset)
267 struct thread *td;
268 int fd;
269 struct uio *auio;
270 off_t offset;
271 {
272 struct file *fp;
273 int error;
274
275 error = fget_read(td, fd, &fp);
276 if (error)
277 return (error);
278 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
279 error = ESPIPE;
280 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
281 error = EINVAL;
282 else
283 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
284 fdrop(fp, td);
285 return (error);
286 }
287
288 /*
289 * Common code for readv and preadv that reads data in
290 * from a file using the passed in uio, offset, and flags.
291 */
292 static int
293 dofileread(td, fd, fp, auio, offset, flags)
294 struct thread *td;
295 int fd;
296 struct file *fp;
297 struct uio *auio;
298 off_t offset;
299 int flags;
300 {
301 ssize_t cnt;
302 int error;
303 #ifdef KTRACE
304 struct uio *ktruio = NULL;
305 #endif
306
307 /* Finish zero length reads right here */
308 if (auio->uio_resid == 0) {
309 td->td_retval[0] = 0;
310 return(0);
311 }
312 auio->uio_rw = UIO_READ;
313 auio->uio_offset = offset;
314 auio->uio_td = td;
315 #ifdef KTRACE
316 if (KTRPOINT(td, KTR_GENIO))
317 ktruio = cloneuio(auio);
318 #endif
319 cnt = auio->uio_resid;
320 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
321 if (auio->uio_resid != cnt && (error == ERESTART ||
322 error == EINTR || error == EWOULDBLOCK))
323 error = 0;
324 }
325 cnt -= auio->uio_resid;
326 #ifdef KTRACE
327 if (ktruio != NULL) {
328 ktruio->uio_resid = cnt;
329 ktrgenio(fd, UIO_READ, ktruio, error);
330 }
331 #endif
332 td->td_retval[0] = cnt;
333 return (error);
334 }
335
336 #ifndef _SYS_SYSPROTO_H_
337 struct write_args {
338 int fd;
339 const void *buf;
340 size_t nbyte;
341 };
342 #endif
343 int
344 write(td, uap)
345 struct thread *td;
346 struct write_args *uap;
347 {
348 struct uio auio;
349 struct iovec aiov;
350 int error;
351
352 if (uap->nbyte > INT_MAX)
353 return (EINVAL);
354 aiov.iov_base = (void *)(uintptr_t)uap->buf;
355 aiov.iov_len = uap->nbyte;
356 auio.uio_iov = &aiov;
357 auio.uio_iovcnt = 1;
358 auio.uio_resid = uap->nbyte;
359 auio.uio_segflg = UIO_USERSPACE;
360 error = kern_writev(td, uap->fd, &auio);
361 return(error);
362 }
363
364 /*
365 * Positioned write system call.
366 */
367 #ifndef _SYS_SYSPROTO_H_
368 struct pwrite_args {
369 int fd;
370 const void *buf;
371 size_t nbyte;
372 int pad;
373 off_t offset;
374 };
375 #endif
376 int
377 pwrite(td, uap)
378 struct thread *td;
379 struct pwrite_args *uap;
380 {
381 struct uio auio;
382 struct iovec aiov;
383 int error;
384
385 if (uap->nbyte > INT_MAX)
386 return (EINVAL);
387 aiov.iov_base = (void *)(uintptr_t)uap->buf;
388 aiov.iov_len = uap->nbyte;
389 auio.uio_iov = &aiov;
390 auio.uio_iovcnt = 1;
391 auio.uio_resid = uap->nbyte;
392 auio.uio_segflg = UIO_USERSPACE;
393 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
394 return(error);
395 }
396
397 int
398 freebsd6_pwrite(td, uap)
399 struct thread *td;
400 struct freebsd6_pwrite_args *uap;
401 {
402 struct pwrite_args oargs;
403
404 oargs.fd = uap->fd;
405 oargs.buf = uap->buf;
406 oargs.nbyte = uap->nbyte;
407 oargs.offset = uap->offset;
408 return (pwrite(td, &oargs));
409 }
410
411 /*
412 * Gather write system call.
413 */
414 #ifndef _SYS_SYSPROTO_H_
415 struct writev_args {
416 int fd;
417 struct iovec *iovp;
418 u_int iovcnt;
419 };
420 #endif
421 int
422 writev(struct thread *td, struct writev_args *uap)
423 {
424 struct uio *auio;
425 int error;
426
427 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
428 if (error)
429 return (error);
430 error = kern_writev(td, uap->fd, auio);
431 free(auio, M_IOV);
432 return (error);
433 }
434
435 int
436 kern_writev(struct thread *td, int fd, struct uio *auio)
437 {
438 struct file *fp;
439 int error;
440
441 error = fget_write(td, fd, &fp);
442 if (error)
443 return (error);
444 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
445 fdrop(fp, td);
446 return (error);
447 }
448
449 /*
450 * Gather positioned write system call.
451 */
452 #ifndef _SYS_SYSPROTO_H_
453 struct pwritev_args {
454 int fd;
455 struct iovec *iovp;
456 u_int iovcnt;
457 off_t offset;
458 };
459 #endif
460 int
461 pwritev(struct thread *td, struct pwritev_args *uap)
462 {
463 struct uio *auio;
464 int error;
465
466 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
467 if (error)
468 return (error);
469 error = kern_pwritev(td, uap->fd, auio, uap->offset);
470 free(auio, M_IOV);
471 return (error);
472 }
473
474 int
475 kern_pwritev(td, fd, auio, offset)
476 struct thread *td;
477 struct uio *auio;
478 int fd;
479 off_t offset;
480 {
481 struct file *fp;
482 int error;
483
484 error = fget_write(td, fd, &fp);
485 if (error)
486 return (error);
487 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
488 error = ESPIPE;
489 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
490 error = EINVAL;
491 else
492 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
493 fdrop(fp, td);
494 return (error);
495 }
496
497 /*
498 * Common code for writev and pwritev that writes data to
499 * a file using the passed in uio, offset, and flags.
500 */
501 static int
502 dofilewrite(td, fd, fp, auio, offset, flags)
503 struct thread *td;
504 int fd;
505 struct file *fp;
506 struct uio *auio;
507 off_t offset;
508 int flags;
509 {
510 ssize_t cnt;
511 int error;
512 #ifdef KTRACE
513 struct uio *ktruio = NULL;
514 #endif
515
516 auio->uio_rw = UIO_WRITE;
517 auio->uio_td = td;
518 auio->uio_offset = offset;
519 #ifdef KTRACE
520 if (KTRPOINT(td, KTR_GENIO))
521 ktruio = cloneuio(auio);
522 #endif
523 cnt = auio->uio_resid;
524 if (fp->f_type == DTYPE_VNODE)
525 bwillwrite();
526 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
527 if (auio->uio_resid != cnt && (error == ERESTART ||
528 error == EINTR || error == EWOULDBLOCK))
529 error = 0;
530 /* Socket layer is responsible for issuing SIGPIPE. */
531 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
532 PROC_LOCK(td->td_proc);
533 psignal(td->td_proc, SIGPIPE);
534 PROC_UNLOCK(td->td_proc);
535 }
536 }
537 cnt -= auio->uio_resid;
538 #ifdef KTRACE
539 if (ktruio != NULL) {
540 ktruio->uio_resid = cnt;
541 ktrgenio(fd, UIO_WRITE, ktruio, error);
542 }
543 #endif
544 td->td_retval[0] = cnt;
545 return (error);
546 }
547
548 /*
549 * Truncate a file given a file descriptor.
550 *
551 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
552 * descriptor isn't writable.
553 */
554 int
555 kern_ftruncate(td, fd, length)
556 struct thread *td;
557 int fd;
558 off_t length;
559 {
560 struct file *fp;
561 int error;
562
563 AUDIT_ARG(fd, fd);
564 if (length < 0)
565 return (EINVAL);
566 error = fget(td, fd, &fp);
567 if (error)
568 return (error);
569 AUDIT_ARG(file, td->td_proc, fp);
570 if (!(fp->f_flag & FWRITE)) {
571 fdrop(fp, td);
572 return (EINVAL);
573 }
574 error = fo_truncate(fp, length, td->td_ucred, td);
575 fdrop(fp, td);
576 return (error);
577 }
578
579 #ifndef _SYS_SYSPROTO_H_
580 struct ftruncate_args {
581 int fd;
582 int pad;
583 off_t length;
584 };
585 #endif
586 int
587 ftruncate(td, uap)
588 struct thread *td;
589 struct ftruncate_args *uap;
590 {
591
592 return (kern_ftruncate(td, uap->fd, uap->length));
593 }
594
595 #if defined(COMPAT_43)
596 #ifndef _SYS_SYSPROTO_H_
597 struct oftruncate_args {
598 int fd;
599 long length;
600 };
601 #endif
602 int
603 oftruncate(td, uap)
604 struct thread *td;
605 struct oftruncate_args *uap;
606 {
607
608 return (kern_ftruncate(td, uap->fd, uap->length));
609 }
610 #endif /* COMPAT_43 */
611
612 #ifndef _SYS_SYSPROTO_H_
613 struct ioctl_args {
614 int fd;
615 u_long com;
616 caddr_t data;
617 };
618 #endif
619 /* ARGSUSED */
620 int
621 ioctl(struct thread *td, struct ioctl_args *uap)
622 {
623 u_long com;
624 int arg, error;
625 u_int size;
626 caddr_t data;
627
628 if (uap->com > 0xffffffff) {
629 printf(
630 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
631 td->td_proc->p_pid, td->td_name, uap->com);
632 uap->com &= 0xffffffff;
633 }
634 com = uap->com;
635
636 /*
637 * Interpret high order word to find amount of data to be
638 * copied to/from the user's address space.
639 */
640 size = IOCPARM_LEN(com);
641 if ((size > IOCPARM_MAX) ||
642 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
643 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
644 ((com & IOC_OUT) && size == 0) ||
645 #else
646 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
647 #endif
648 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
649 return (ENOTTY);
650
651 if (size > 0) {
652 if (!(com & IOC_VOID))
653 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
654 else {
655 /* Integer argument. */
656 arg = (intptr_t)uap->data;
657 data = (void *)&arg;
658 size = 0;
659 }
660 } else
661 data = (void *)&uap->data;
662 if (com & IOC_IN) {
663 error = copyin(uap->data, data, (u_int)size);
664 if (error) {
665 if (size > 0)
666 free(data, M_IOCTLOPS);
667 return (error);
668 }
669 } else if (com & IOC_OUT) {
670 /*
671 * Zero the buffer so the user always
672 * gets back something deterministic.
673 */
674 bzero(data, size);
675 }
676
677 error = kern_ioctl(td, uap->fd, com, data);
678
679 if (error == 0 && (com & IOC_OUT))
680 error = copyout(data, uap->data, (u_int)size);
681
682 if (size > 0)
683 free(data, M_IOCTLOPS);
684 return (error);
685 }
686
687 int
688 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
689 {
690 struct file *fp;
691 struct filedesc *fdp;
692 int error;
693 int tmp;
694
695 if ((error = fget(td, fd, &fp)) != 0)
696 return (error);
697 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
698 fdrop(fp, td);
699 return (EBADF);
700 }
701 fdp = td->td_proc->p_fd;
702 switch (com) {
703 case FIONCLEX:
704 FILEDESC_XLOCK(fdp);
705 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
706 FILEDESC_XUNLOCK(fdp);
707 goto out;
708 case FIOCLEX:
709 FILEDESC_XLOCK(fdp);
710 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
711 FILEDESC_XUNLOCK(fdp);
712 goto out;
713 case FIONBIO:
714 if ((tmp = *(int *)data))
715 atomic_set_int(&fp->f_flag, FNONBLOCK);
716 else
717 atomic_clear_int(&fp->f_flag, FNONBLOCK);
718 data = (void *)&tmp;
719 break;
720 case FIOASYNC:
721 if ((tmp = *(int *)data))
722 atomic_set_int(&fp->f_flag, FASYNC);
723 else
724 atomic_clear_int(&fp->f_flag, FASYNC);
725 data = (void *)&tmp;
726 break;
727 }
728
729 error = fo_ioctl(fp, com, data, td->td_ucred, td);
730 out:
731 fdrop(fp, td);
732 return (error);
733 }
734
735 #ifndef _SYS_SYSPROTO_H_
736 struct select_args {
737 int nd;
738 fd_set *in, *ou, *ex;
739 struct timeval *tv;
740 };
741 #endif
742 int
743 select(td, uap)
744 register struct thread *td;
745 register struct select_args *uap;
746 {
747 struct timeval tv, *tvp;
748 int error;
749
750 if (uap->tv != NULL) {
751 error = copyin(uap->tv, &tv, sizeof(tv));
752 if (error)
753 return (error);
754 tvp = &tv;
755 } else
756 tvp = NULL;
757
758 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp));
759 }
760
761 int
762 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
763 fd_set *fd_ex, struct timeval *tvp)
764 {
765 struct filedesc *fdp;
766 /*
767 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
768 * infds with the new FD_SETSIZE of 1024, and more than enough for
769 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
770 * of 256.
771 */
772 fd_mask s_selbits[howmany(2048, NFDBITS)];
773 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
774 struct timeval atv, rtv, ttv;
775 int error, timo;
776 u_int nbufbytes, ncpbytes, nfdbits;
777
778 if (nd < 0)
779 return (EINVAL);
780 fdp = td->td_proc->p_fd;
781
782 FILEDESC_SLOCK(fdp);
783 if (nd > td->td_proc->p_fd->fd_nfiles)
784 nd = td->td_proc->p_fd->fd_nfiles; /* forgiving; slightly wrong */
785 FILEDESC_SUNLOCK(fdp);
786
787 /*
788 * Allocate just enough bits for the non-null fd_sets. Use the
789 * preallocated auto buffer if possible.
790 */
791 nfdbits = roundup(nd, NFDBITS);
792 ncpbytes = nfdbits / NBBY;
793 nbufbytes = 0;
794 if (fd_in != NULL)
795 nbufbytes += 2 * ncpbytes;
796 if (fd_ou != NULL)
797 nbufbytes += 2 * ncpbytes;
798 if (fd_ex != NULL)
799 nbufbytes += 2 * ncpbytes;
800 if (nbufbytes <= sizeof s_selbits)
801 selbits = &s_selbits[0];
802 else
803 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
804
805 /*
806 * Assign pointers into the bit buffers and fetch the input bits.
807 * Put the output buffers together so that they can be bzeroed
808 * together.
809 */
810 sbp = selbits;
811 #define getbits(name, x) \
812 do { \
813 if (name == NULL) \
814 ibits[x] = NULL; \
815 else { \
816 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
817 obits[x] = sbp; \
818 sbp += ncpbytes / sizeof *sbp; \
819 error = copyin(name, ibits[x], ncpbytes); \
820 if (error != 0) \
821 goto done; \
822 } \
823 } while (0)
824 getbits(fd_in, 0);
825 getbits(fd_ou, 1);
826 getbits(fd_ex, 2);
827 #undef getbits
828 if (nbufbytes != 0)
829 bzero(selbits, nbufbytes / 2);
830
831 if (tvp != NULL) {
832 atv = *tvp;
833 if (itimerfix(&atv)) {
834 error = EINVAL;
835 goto done;
836 }
837 getmicrouptime(&rtv);
838 timevaladd(&atv, &rtv);
839 } else {
840 atv.tv_sec = 0;
841 atv.tv_usec = 0;
842 }
843 timo = 0;
844 seltdinit(td);
845 /* Iterate until the timeout expires or descriptors become ready. */
846 for (;;) {
847 error = selscan(td, ibits, obits, nd);
848 if (error || td->td_retval[0] != 0)
849 break;
850 if (atv.tv_sec || atv.tv_usec) {
851 getmicrouptime(&rtv);
852 if (timevalcmp(&rtv, &atv, >=))
853 break;
854 ttv = atv;
855 timevalsub(&ttv, &rtv);
856 timo = ttv.tv_sec > 24 * 60 * 60 ?
857 24 * 60 * 60 * hz : tvtohz(&ttv);
858 }
859 error = seltdwait(td, timo);
860 if (error)
861 break;
862 error = selrescan(td, ibits, obits);
863 if (error || td->td_retval[0] != 0)
864 break;
865 }
866 seltdclear(td);
867
868 done:
869 /* select is not restarted after signals... */
870 if (error == ERESTART)
871 error = EINTR;
872 if (error == EWOULDBLOCK)
873 error = 0;
874 #define putbits(name, x) \
875 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
876 error = error2;
877 if (error == 0) {
878 int error2;
879
880 putbits(fd_in, 0);
881 putbits(fd_ou, 1);
882 putbits(fd_ex, 2);
883 #undef putbits
884 }
885 if (selbits != &s_selbits[0])
886 free(selbits, M_SELECT);
887
888 return (error);
889 }
890
891 /*
892 * Traverse the list of fds attached to this thread's seltd and check for
893 * completion.
894 */
895 static int
896 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
897 {
898 struct seltd *stp;
899 struct selfd *sfp;
900 struct selfd *sfn;
901 struct selinfo *si;
902 struct file *fp;
903 int msk, fd;
904 int n = 0;
905 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
906 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
907 struct filedesc *fdp = td->td_proc->p_fd;
908
909 stp = td->td_sel;
910 FILEDESC_SLOCK(fdp);
911 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
912 fd = (int)(uintptr_t)sfp->sf_cookie;
913 si = sfp->sf_si;
914 selfdfree(stp, sfp);
915 /* If the selinfo wasn't cleared the event didn't fire. */
916 if (si != NULL)
917 continue;
918 if ((fp = fget_locked(fdp, fd)) == NULL) {
919 FILEDESC_SUNLOCK(fdp);
920 return (EBADF);
921 }
922 for (msk = 0; msk < 3; msk++) {
923 if (ibits[msk] == NULL)
924 continue;
925 if ((ibits[msk][fd/NFDBITS] &
926 ((fd_mask) 1 << (fd % NFDBITS))) == 0)
927 continue;
928 if (fo_poll(fp, flag[msk], td->td_ucred, td)) {
929 obits[msk][(fd)/NFDBITS] |=
930 ((fd_mask)1 << ((fd) % NFDBITS));
931 n++;
932 }
933 }
934 }
935 FILEDESC_SUNLOCK(fdp);
936 stp->st_flags = 0;
937 td->td_retval[0] = n;
938 return (0);
939 }
940
941 /*
942 * Perform the initial filedescriptor scan and register ourselves with
943 * each selinfo.
944 */
945 static int
946 selscan(td, ibits, obits, nfd)
947 struct thread *td;
948 fd_mask **ibits, **obits;
949 int nfd;
950 {
951 int msk, i, fd;
952 fd_mask bits;
953 struct file *fp;
954 int n = 0;
955 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
956 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
957 struct filedesc *fdp = td->td_proc->p_fd;
958
959 FILEDESC_SLOCK(fdp);
960 for (msk = 0; msk < 3; msk++) {
961 if (ibits[msk] == NULL)
962 continue;
963 for (i = 0; i < nfd; i += NFDBITS) {
964 bits = ibits[msk][i/NFDBITS];
965 /* ffs(int mask) not portable, fd_mask is long */
966 for (fd = i; bits && fd < nfd; fd++, bits >>= 1) {
967 if (!(bits & 1))
968 continue;
969 if ((fp = fget_locked(fdp, fd)) == NULL) {
970 FILEDESC_SUNLOCK(fdp);
971 return (EBADF);
972 }
973 selfdalloc(td, (void *)(uintptr_t)fd);
974 if (fo_poll(fp, flag[msk], td->td_ucred,
975 td)) {
976 obits[msk][(fd)/NFDBITS] |=
977 ((fd_mask)1 << ((fd) % NFDBITS));
978 n++;
979 }
980 }
981 }
982 }
983 FILEDESC_SUNLOCK(fdp);
984 td->td_retval[0] = n;
985 return (0);
986 }
987
988 #ifndef _SYS_SYSPROTO_H_
989 struct poll_args {
990 struct pollfd *fds;
991 u_int nfds;
992 int timeout;
993 };
994 #endif
995 int
996 poll(td, uap)
997 struct thread *td;
998 struct poll_args *uap;
999 {
1000 struct pollfd *bits;
1001 struct pollfd smallbits[32];
1002 struct timeval atv, rtv, ttv;
1003 int error = 0, timo;
1004 u_int nfds;
1005 size_t ni;
1006
1007 nfds = uap->nfds;
1008 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1009 return (EINVAL);
1010 ni = nfds * sizeof(struct pollfd);
1011 if (ni > sizeof(smallbits))
1012 bits = malloc(ni, M_TEMP, M_WAITOK);
1013 else
1014 bits = smallbits;
1015 error = copyin(uap->fds, bits, ni);
1016 if (error)
1017 goto done;
1018 if (uap->timeout != INFTIM) {
1019 atv.tv_sec = uap->timeout / 1000;
1020 atv.tv_usec = (uap->timeout % 1000) * 1000;
1021 if (itimerfix(&atv)) {
1022 error = EINVAL;
1023 goto done;
1024 }
1025 getmicrouptime(&rtv);
1026 timevaladd(&atv, &rtv);
1027 } else {
1028 atv.tv_sec = 0;
1029 atv.tv_usec = 0;
1030 }
1031 timo = 0;
1032 seltdinit(td);
1033 /* Iterate until the timeout expires or descriptors become ready. */
1034 for (;;) {
1035 error = pollscan(td, bits, nfds);
1036 if (error || td->td_retval[0] != 0)
1037 break;
1038 if (atv.tv_sec || atv.tv_usec) {
1039 getmicrouptime(&rtv);
1040 if (timevalcmp(&rtv, &atv, >=))
1041 break;
1042 ttv = atv;
1043 timevalsub(&ttv, &rtv);
1044 timo = ttv.tv_sec > 24 * 60 * 60 ?
1045 24 * 60 * 60 * hz : tvtohz(&ttv);
1046 }
1047 error = seltdwait(td, timo);
1048 if (error)
1049 break;
1050 error = pollrescan(td);
1051 if (error || td->td_retval[0] != 0)
1052 break;
1053 }
1054 seltdclear(td);
1055
1056 done:
1057 /* poll is not restarted after signals... */
1058 if (error == ERESTART)
1059 error = EINTR;
1060 if (error == EWOULDBLOCK)
1061 error = 0;
1062 if (error == 0) {
1063 error = copyout(bits, uap->fds, ni);
1064 if (error)
1065 goto out;
1066 }
1067 out:
1068 if (ni > sizeof(smallbits))
1069 free(bits, M_TEMP);
1070 return (error);
1071 }
1072
1073 static int
1074 pollrescan(struct thread *td)
1075 {
1076 struct seltd *stp;
1077 struct selfd *sfp;
1078 struct selfd *sfn;
1079 struct selinfo *si;
1080 struct filedesc *fdp;
1081 struct file *fp;
1082 struct pollfd *fd;
1083 int n;
1084
1085 n = 0;
1086 fdp = td->td_proc->p_fd;
1087 stp = td->td_sel;
1088 FILEDESC_SLOCK(fdp);
1089 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1090 fd = (struct pollfd *)sfp->sf_cookie;
1091 si = sfp->sf_si;
1092 selfdfree(stp, sfp);
1093 /* If the selinfo wasn't cleared the event didn't fire. */
1094 if (si != NULL)
1095 continue;
1096 fp = fdp->fd_ofiles[fd->fd];
1097 if (fp == NULL) {
1098 fd->revents = POLLNVAL;
1099 n++;
1100 continue;
1101 }
1102 /*
1103 * Note: backend also returns POLLHUP and
1104 * POLLERR if appropriate.
1105 */
1106 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1107 if (fd->revents != 0)
1108 n++;
1109 }
1110 FILEDESC_SUNLOCK(fdp);
1111 stp->st_flags = 0;
1112 td->td_retval[0] = n;
1113 return (0);
1114 }
1115
1116
1117 static int
1118 pollscan(td, fds, nfd)
1119 struct thread *td;
1120 struct pollfd *fds;
1121 u_int nfd;
1122 {
1123 struct filedesc *fdp = td->td_proc->p_fd;
1124 int i;
1125 struct file *fp;
1126 int n = 0;
1127
1128 FILEDESC_SLOCK(fdp);
1129 for (i = 0; i < nfd; i++, fds++) {
1130 if (fds->fd >= fdp->fd_nfiles) {
1131 fds->revents = POLLNVAL;
1132 n++;
1133 } else if (fds->fd < 0) {
1134 fds->revents = 0;
1135 } else {
1136 fp = fdp->fd_ofiles[fds->fd];
1137 if (fp == NULL) {
1138 fds->revents = POLLNVAL;
1139 n++;
1140 } else {
1141 /*
1142 * Note: backend also returns POLLHUP and
1143 * POLLERR if appropriate.
1144 */
1145 selfdalloc(td, fds);
1146 fds->revents = fo_poll(fp, fds->events,
1147 td->td_ucred, td);
1148 if (fds->revents != 0)
1149 n++;
1150 }
1151 }
1152 }
1153 FILEDESC_SUNLOCK(fdp);
1154 td->td_retval[0] = n;
1155 return (0);
1156 }
1157
1158 /*
1159 * OpenBSD poll system call.
1160 *
1161 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1162 */
1163 #ifndef _SYS_SYSPROTO_H_
1164 struct openbsd_poll_args {
1165 struct pollfd *fds;
1166 u_int nfds;
1167 int timeout;
1168 };
1169 #endif
1170 int
1171 openbsd_poll(td, uap)
1172 register struct thread *td;
1173 register struct openbsd_poll_args *uap;
1174 {
1175 return (poll(td, (struct poll_args *)uap));
1176 }
1177
1178 /*
1179 * XXX This was created specifically to support netncp and netsmb. This
1180 * allows the caller to specify a socket to wait for events on. It returns
1181 * 0 if any events matched and an error otherwise. There is no way to
1182 * determine which events fired.
1183 */
1184 int
1185 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1186 {
1187 struct timeval atv, rtv, ttv;
1188 int error, timo;
1189
1190 if (tvp != NULL) {
1191 atv = *tvp;
1192 if (itimerfix(&atv))
1193 return (EINVAL);
1194 getmicrouptime(&rtv);
1195 timevaladd(&atv, &rtv);
1196 } else {
1197 atv.tv_sec = 0;
1198 atv.tv_usec = 0;
1199 }
1200
1201 timo = 0;
1202 seltdinit(td);
1203 /*
1204 * Iterate until the timeout expires or the socket becomes ready.
1205 */
1206 for (;;) {
1207 selfdalloc(td, NULL);
1208 error = sopoll(so, events, NULL, td);
1209 /* error here is actually the ready events. */
1210 if (error)
1211 return (0);
1212 if (atv.tv_sec || atv.tv_usec) {
1213 getmicrouptime(&rtv);
1214 if (timevalcmp(&rtv, &atv, >=)) {
1215 seltdclear(td);
1216 return (EWOULDBLOCK);
1217 }
1218 ttv = atv;
1219 timevalsub(&ttv, &rtv);
1220 timo = ttv.tv_sec > 24 * 60 * 60 ?
1221 24 * 60 * 60 * hz : tvtohz(&ttv);
1222 }
1223 error = seltdwait(td, timo);
1224 seltdclear(td);
1225 if (error)
1226 break;
1227 }
1228 /* XXX Duplicates ncp/smb behavior. */
1229 if (error == ERESTART)
1230 error = 0;
1231 return (error);
1232 }
1233
1234 /*
1235 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1236 * have two select sets, one for read and another for write.
1237 */
1238 static void
1239 selfdalloc(struct thread *td, void *cookie)
1240 {
1241 struct seltd *stp;
1242
1243 stp = td->td_sel;
1244 if (stp->st_free1 == NULL)
1245 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1246 stp->st_free1->sf_td = stp;
1247 stp->st_free1->sf_cookie = cookie;
1248 if (stp->st_free2 == NULL)
1249 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1250 stp->st_free2->sf_td = stp;
1251 stp->st_free2->sf_cookie = cookie;
1252 }
1253
1254 static void
1255 selfdfree(struct seltd *stp, struct selfd *sfp)
1256 {
1257 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1258 mtx_lock(sfp->sf_mtx);
1259 if (sfp->sf_si)
1260 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1261 mtx_unlock(sfp->sf_mtx);
1262 uma_zfree(selfd_zone, sfp);
1263 }
1264
1265 /*
1266 * Record a select request.
1267 */
1268 void
1269 selrecord(selector, sip)
1270 struct thread *selector;
1271 struct selinfo *sip;
1272 {
1273 struct selfd *sfp;
1274 struct seltd *stp;
1275 struct mtx *mtxp;
1276
1277 stp = selector->td_sel;
1278 /*
1279 * Don't record when doing a rescan.
1280 */
1281 if (stp->st_flags & SELTD_RESCAN)
1282 return;
1283 /*
1284 * Grab one of the preallocated descriptors.
1285 */
1286 sfp = NULL;
1287 if ((sfp = stp->st_free1) != NULL)
1288 stp->st_free1 = NULL;
1289 else if ((sfp = stp->st_free2) != NULL)
1290 stp->st_free2 = NULL;
1291 else
1292 panic("selrecord: No free selfd on selq");
1293 mtxp = mtx_pool_find(mtxpool_sleep, sip);
1294 /*
1295 * Initialize the sfp and queue it in the thread.
1296 */
1297 sfp->sf_si = sip;
1298 sfp->sf_mtx = mtxp;
1299 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1300 /*
1301 * Now that we've locked the sip, check for initialization.
1302 */
1303 mtx_lock(mtxp);
1304 if (sip->si_mtx == NULL) {
1305 sip->si_mtx = mtxp;
1306 TAILQ_INIT(&sip->si_tdlist);
1307 }
1308 /*
1309 * Add this thread to the list of selfds listening on this selinfo.
1310 */
1311 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1312 mtx_unlock(sip->si_mtx);
1313 }
1314
1315 /* Wake up a selecting thread. */
1316 void
1317 selwakeup(sip)
1318 struct selinfo *sip;
1319 {
1320 doselwakeup(sip, -1);
1321 }
1322
1323 /* Wake up a selecting thread, and set its priority. */
1324 void
1325 selwakeuppri(sip, pri)
1326 struct selinfo *sip;
1327 int pri;
1328 {
1329 doselwakeup(sip, pri);
1330 }
1331
1332 /*
1333 * Do a wakeup when a selectable event occurs.
1334 */
1335 static void
1336 doselwakeup(sip, pri)
1337 struct selinfo *sip;
1338 int pri;
1339 {
1340 struct selfd *sfp;
1341 struct selfd *sfn;
1342 struct seltd *stp;
1343
1344 /* If it's not initialized there can't be any waiters. */
1345 if (sip->si_mtx == NULL)
1346 return;
1347 /*
1348 * Locking the selinfo locks all selfds associated with it.
1349 */
1350 mtx_lock(sip->si_mtx);
1351 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1352 /*
1353 * Once we remove this sfp from the list and clear the
1354 * sf_si seltdclear will know to ignore this si.
1355 */
1356 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1357 sfp->sf_si = NULL;
1358 stp = sfp->sf_td;
1359 mtx_lock(&stp->st_mtx);
1360 stp->st_flags |= SELTD_PENDING;
1361 cv_broadcastpri(&stp->st_wait, pri);
1362 mtx_unlock(&stp->st_mtx);
1363 }
1364 mtx_unlock(sip->si_mtx);
1365 }
1366
1367 static void
1368 seltdinit(struct thread *td)
1369 {
1370 struct seltd *stp;
1371
1372 if ((stp = td->td_sel) != NULL)
1373 goto out;
1374 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1375 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1376 cv_init(&stp->st_wait, "select");
1377 out:
1378 stp->st_flags = 0;
1379 STAILQ_INIT(&stp->st_selq);
1380 }
1381
1382 static int
1383 seltdwait(struct thread *td, int timo)
1384 {
1385 struct seltd *stp;
1386 int error;
1387
1388 stp = td->td_sel;
1389 /*
1390 * An event of interest may occur while we do not hold the seltd
1391 * locked so check the pending flag before we sleep.
1392 */
1393 mtx_lock(&stp->st_mtx);
1394 /*
1395 * Any further calls to selrecord will be a rescan.
1396 */
1397 stp->st_flags |= SELTD_RESCAN;
1398 if (stp->st_flags & SELTD_PENDING) {
1399 mtx_unlock(&stp->st_mtx);
1400 return (0);
1401 }
1402 if (timo > 0)
1403 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1404 else
1405 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1406 mtx_unlock(&stp->st_mtx);
1407
1408 return (error);
1409 }
1410
1411 void
1412 seltdfini(struct thread *td)
1413 {
1414 struct seltd *stp;
1415
1416 stp = td->td_sel;
1417 if (stp == NULL)
1418 return;
1419 if (stp->st_free1)
1420 uma_zfree(selfd_zone, stp->st_free1);
1421 if (stp->st_free2)
1422 uma_zfree(selfd_zone, stp->st_free2);
1423 td->td_sel = NULL;
1424 free(stp, M_SELECT);
1425 }
1426
1427 /*
1428 * Remove the references to the thread from all of the objects we were
1429 * polling.
1430 */
1431 static void
1432 seltdclear(struct thread *td)
1433 {
1434 struct seltd *stp;
1435 struct selfd *sfp;
1436 struct selfd *sfn;
1437
1438 stp = td->td_sel;
1439 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1440 selfdfree(stp, sfp);
1441 stp->st_flags = 0;
1442 }
1443
1444 static void selectinit(void *);
1445 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1446 static void
1447 selectinit(void *dummy __unused)
1448 {
1449 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1450 NULL, NULL, UMA_ALIGN_PTR, 0);
1451 }
FreeBSD kvm: FreeBSD sources