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/9.1/sys/kern/sys_generic.c 237485 2012-06-23 17:39:40Z davide $");
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 int iosize_max_clamp = 1;
78 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
79 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
80 /*
81 * Assert that the return value of read(2) and write(2) syscalls fits
82 * into a register. If not, an architecture will need to provide the
83 * usermode wrappers to reconstruct the result.
84 */
85 CTASSERT(sizeof(register_t) >= sizeof(size_t));
86
87 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
88 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
89 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
90
91 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
92 u_int);
93 static int pollscan(struct thread *, struct pollfd *, u_int);
94 static int pollrescan(struct thread *);
95 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
96 static int selrescan(struct thread *, fd_mask **, fd_mask **);
97 static void selfdalloc(struct thread *, void *);
98 static void selfdfree(struct seltd *, struct selfd *);
99 static int dofileread(struct thread *, int, struct file *, struct uio *,
100 off_t, int);
101 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
102 off_t, int);
103 static void doselwakeup(struct selinfo *, int);
104 static void seltdinit(struct thread *);
105 static int seltdwait(struct thread *, int);
106 static void seltdclear(struct thread *);
107
108 /*
109 * One seltd per-thread allocated on demand as needed.
110 *
111 * t - protected by st_mtx
112 * k - Only accessed by curthread or read-only
113 */
114 struct seltd {
115 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
116 struct selfd *st_free1; /* (k) free fd for read set. */
117 struct selfd *st_free2; /* (k) free fd for write set. */
118 struct mtx st_mtx; /* Protects struct seltd */
119 struct cv st_wait; /* (t) Wait channel. */
120 int st_flags; /* (t) SELTD_ flags. */
121 };
122
123 #define SELTD_PENDING 0x0001 /* We have pending events. */
124 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
125
126 /*
127 * One selfd allocated per-thread per-file-descriptor.
128 * f - protected by sf_mtx
129 */
130 struct selfd {
131 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
132 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
133 struct selinfo *sf_si; /* (f) selinfo when linked. */
134 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
135 struct seltd *sf_td; /* (k) owning seltd. */
136 void *sf_cookie; /* (k) fd or pollfd. */
137 };
138
139 static uma_zone_t selfd_zone;
140 static struct mtx_pool *mtxpool_select;
141
142 #ifndef _SYS_SYSPROTO_H_
143 struct read_args {
144 int fd;
145 void *buf;
146 size_t nbyte;
147 };
148 #endif
149 int
150 sys_read(td, uap)
151 struct thread *td;
152 struct read_args *uap;
153 {
154 struct uio auio;
155 struct iovec aiov;
156 int error;
157
158 if (uap->nbyte > IOSIZE_MAX)
159 return (EINVAL);
160 aiov.iov_base = uap->buf;
161 aiov.iov_len = uap->nbyte;
162 auio.uio_iov = &aiov;
163 auio.uio_iovcnt = 1;
164 auio.uio_resid = uap->nbyte;
165 auio.uio_segflg = UIO_USERSPACE;
166 error = kern_readv(td, uap->fd, &auio);
167 return(error);
168 }
169
170 /*
171 * Positioned read system call
172 */
173 #ifndef _SYS_SYSPROTO_H_
174 struct pread_args {
175 int fd;
176 void *buf;
177 size_t nbyte;
178 int pad;
179 off_t offset;
180 };
181 #endif
182 int
183 sys_pread(td, uap)
184 struct thread *td;
185 struct pread_args *uap;
186 {
187 struct uio auio;
188 struct iovec aiov;
189 int error;
190
191 if (uap->nbyte > IOSIZE_MAX)
192 return (EINVAL);
193 aiov.iov_base = uap->buf;
194 aiov.iov_len = uap->nbyte;
195 auio.uio_iov = &aiov;
196 auio.uio_iovcnt = 1;
197 auio.uio_resid = uap->nbyte;
198 auio.uio_segflg = UIO_USERSPACE;
199 error = kern_preadv(td, uap->fd, &auio, uap->offset);
200 return(error);
201 }
202
203 int
204 freebsd6_pread(td, uap)
205 struct thread *td;
206 struct freebsd6_pread_args *uap;
207 {
208 struct pread_args oargs;
209
210 oargs.fd = uap->fd;
211 oargs.buf = uap->buf;
212 oargs.nbyte = uap->nbyte;
213 oargs.offset = uap->offset;
214 return (sys_pread(td, &oargs));
215 }
216
217 /*
218 * Scatter read system call.
219 */
220 #ifndef _SYS_SYSPROTO_H_
221 struct readv_args {
222 int fd;
223 struct iovec *iovp;
224 u_int iovcnt;
225 };
226 #endif
227 int
228 sys_readv(struct thread *td, struct readv_args *uap)
229 {
230 struct uio *auio;
231 int error;
232
233 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
234 if (error)
235 return (error);
236 error = kern_readv(td, uap->fd, auio);
237 free(auio, M_IOV);
238 return (error);
239 }
240
241 int
242 kern_readv(struct thread *td, int fd, struct uio *auio)
243 {
244 struct file *fp;
245 int error;
246
247 error = fget_read(td, fd, CAP_READ | CAP_SEEK, &fp);
248 if (error)
249 return (error);
250 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
251 fdrop(fp, td);
252 return (error);
253 }
254
255 /*
256 * Scatter positioned read system call.
257 */
258 #ifndef _SYS_SYSPROTO_H_
259 struct preadv_args {
260 int fd;
261 struct iovec *iovp;
262 u_int iovcnt;
263 off_t offset;
264 };
265 #endif
266 int
267 sys_preadv(struct thread *td, struct preadv_args *uap)
268 {
269 struct uio *auio;
270 int error;
271
272 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
273 if (error)
274 return (error);
275 error = kern_preadv(td, uap->fd, auio, uap->offset);
276 free(auio, M_IOV);
277 return (error);
278 }
279
280 int
281 kern_preadv(td, fd, auio, offset)
282 struct thread *td;
283 int fd;
284 struct uio *auio;
285 off_t offset;
286 {
287 struct file *fp;
288 int error;
289
290 error = fget_read(td, fd, CAP_READ, &fp);
291 if (error)
292 return (error);
293 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
294 error = ESPIPE;
295 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
296 error = EINVAL;
297 else
298 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
299 fdrop(fp, td);
300 return (error);
301 }
302
303 /*
304 * Common code for readv and preadv that reads data in
305 * from a file using the passed in uio, offset, and flags.
306 */
307 static int
308 dofileread(td, fd, fp, auio, offset, flags)
309 struct thread *td;
310 int fd;
311 struct file *fp;
312 struct uio *auio;
313 off_t offset;
314 int flags;
315 {
316 ssize_t cnt;
317 int error;
318 #ifdef KTRACE
319 struct uio *ktruio = NULL;
320 #endif
321
322 /* Finish zero length reads right here */
323 if (auio->uio_resid == 0) {
324 td->td_retval[0] = 0;
325 return(0);
326 }
327 auio->uio_rw = UIO_READ;
328 auio->uio_offset = offset;
329 auio->uio_td = td;
330 #ifdef KTRACE
331 if (KTRPOINT(td, KTR_GENIO))
332 ktruio = cloneuio(auio);
333 #endif
334 cnt = auio->uio_resid;
335 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
336 if (auio->uio_resid != cnt && (error == ERESTART ||
337 error == EINTR || error == EWOULDBLOCK))
338 error = 0;
339 }
340 cnt -= auio->uio_resid;
341 #ifdef KTRACE
342 if (ktruio != NULL) {
343 ktruio->uio_resid = cnt;
344 ktrgenio(fd, UIO_READ, ktruio, error);
345 }
346 #endif
347 td->td_retval[0] = cnt;
348 return (error);
349 }
350
351 #ifndef _SYS_SYSPROTO_H_
352 struct write_args {
353 int fd;
354 const void *buf;
355 size_t nbyte;
356 };
357 #endif
358 int
359 sys_write(td, uap)
360 struct thread *td;
361 struct write_args *uap;
362 {
363 struct uio auio;
364 struct iovec aiov;
365 int error;
366
367 if (uap->nbyte > IOSIZE_MAX)
368 return (EINVAL);
369 aiov.iov_base = (void *)(uintptr_t)uap->buf;
370 aiov.iov_len = uap->nbyte;
371 auio.uio_iov = &aiov;
372 auio.uio_iovcnt = 1;
373 auio.uio_resid = uap->nbyte;
374 auio.uio_segflg = UIO_USERSPACE;
375 error = kern_writev(td, uap->fd, &auio);
376 return(error);
377 }
378
379 /*
380 * Positioned write system call.
381 */
382 #ifndef _SYS_SYSPROTO_H_
383 struct pwrite_args {
384 int fd;
385 const void *buf;
386 size_t nbyte;
387 int pad;
388 off_t offset;
389 };
390 #endif
391 int
392 sys_pwrite(td, uap)
393 struct thread *td;
394 struct pwrite_args *uap;
395 {
396 struct uio auio;
397 struct iovec aiov;
398 int error;
399
400 if (uap->nbyte > IOSIZE_MAX)
401 return (EINVAL);
402 aiov.iov_base = (void *)(uintptr_t)uap->buf;
403 aiov.iov_len = uap->nbyte;
404 auio.uio_iov = &aiov;
405 auio.uio_iovcnt = 1;
406 auio.uio_resid = uap->nbyte;
407 auio.uio_segflg = UIO_USERSPACE;
408 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
409 return(error);
410 }
411
412 int
413 freebsd6_pwrite(td, uap)
414 struct thread *td;
415 struct freebsd6_pwrite_args *uap;
416 {
417 struct pwrite_args oargs;
418
419 oargs.fd = uap->fd;
420 oargs.buf = uap->buf;
421 oargs.nbyte = uap->nbyte;
422 oargs.offset = uap->offset;
423 return (sys_pwrite(td, &oargs));
424 }
425
426 /*
427 * Gather write system call.
428 */
429 #ifndef _SYS_SYSPROTO_H_
430 struct writev_args {
431 int fd;
432 struct iovec *iovp;
433 u_int iovcnt;
434 };
435 #endif
436 int
437 sys_writev(struct thread *td, struct writev_args *uap)
438 {
439 struct uio *auio;
440 int error;
441
442 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
443 if (error)
444 return (error);
445 error = kern_writev(td, uap->fd, auio);
446 free(auio, M_IOV);
447 return (error);
448 }
449
450 int
451 kern_writev(struct thread *td, int fd, struct uio *auio)
452 {
453 struct file *fp;
454 int error;
455
456 error = fget_write(td, fd, CAP_WRITE | CAP_SEEK, &fp);
457 if (error)
458 return (error);
459 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
460 fdrop(fp, td);
461 return (error);
462 }
463
464 /*
465 * Gather positioned write system call.
466 */
467 #ifndef _SYS_SYSPROTO_H_
468 struct pwritev_args {
469 int fd;
470 struct iovec *iovp;
471 u_int iovcnt;
472 off_t offset;
473 };
474 #endif
475 int
476 sys_pwritev(struct thread *td, struct pwritev_args *uap)
477 {
478 struct uio *auio;
479 int error;
480
481 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
482 if (error)
483 return (error);
484 error = kern_pwritev(td, uap->fd, auio, uap->offset);
485 free(auio, M_IOV);
486 return (error);
487 }
488
489 int
490 kern_pwritev(td, fd, auio, offset)
491 struct thread *td;
492 struct uio *auio;
493 int fd;
494 off_t offset;
495 {
496 struct file *fp;
497 int error;
498
499 error = fget_write(td, fd, CAP_WRITE, &fp);
500 if (error)
501 return (error);
502 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
503 error = ESPIPE;
504 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
505 error = EINVAL;
506 else
507 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
508 fdrop(fp, td);
509 return (error);
510 }
511
512 /*
513 * Common code for writev and pwritev that writes data to
514 * a file using the passed in uio, offset, and flags.
515 */
516 static int
517 dofilewrite(td, fd, fp, auio, offset, flags)
518 struct thread *td;
519 int fd;
520 struct file *fp;
521 struct uio *auio;
522 off_t offset;
523 int flags;
524 {
525 ssize_t cnt;
526 int error;
527 #ifdef KTRACE
528 struct uio *ktruio = NULL;
529 #endif
530
531 auio->uio_rw = UIO_WRITE;
532 auio->uio_td = td;
533 auio->uio_offset = offset;
534 #ifdef KTRACE
535 if (KTRPOINT(td, KTR_GENIO))
536 ktruio = cloneuio(auio);
537 #endif
538 cnt = auio->uio_resid;
539 if (fp->f_type == DTYPE_VNODE)
540 bwillwrite();
541 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
542 if (auio->uio_resid != cnt && (error == ERESTART ||
543 error == EINTR || error == EWOULDBLOCK))
544 error = 0;
545 /* Socket layer is responsible for issuing SIGPIPE. */
546 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
547 PROC_LOCK(td->td_proc);
548 tdsignal(td, SIGPIPE);
549 PROC_UNLOCK(td->td_proc);
550 }
551 }
552 cnt -= auio->uio_resid;
553 #ifdef KTRACE
554 if (ktruio != NULL) {
555 ktruio->uio_resid = cnt;
556 ktrgenio(fd, UIO_WRITE, ktruio, error);
557 }
558 #endif
559 td->td_retval[0] = cnt;
560 return (error);
561 }
562
563 /*
564 * Truncate a file given a file descriptor.
565 *
566 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
567 * descriptor isn't writable.
568 */
569 int
570 kern_ftruncate(td, fd, length)
571 struct thread *td;
572 int fd;
573 off_t length;
574 {
575 struct file *fp;
576 int error;
577
578 AUDIT_ARG_FD(fd);
579 if (length < 0)
580 return (EINVAL);
581 error = fget(td, fd, CAP_FTRUNCATE, &fp);
582 if (error)
583 return (error);
584 AUDIT_ARG_FILE(td->td_proc, fp);
585 if (!(fp->f_flag & FWRITE)) {
586 fdrop(fp, td);
587 return (EINVAL);
588 }
589 error = fo_truncate(fp, length, td->td_ucred, td);
590 fdrop(fp, td);
591 return (error);
592 }
593
594 #ifndef _SYS_SYSPROTO_H_
595 struct ftruncate_args {
596 int fd;
597 int pad;
598 off_t length;
599 };
600 #endif
601 int
602 sys_ftruncate(td, uap)
603 struct thread *td;
604 struct ftruncate_args *uap;
605 {
606
607 return (kern_ftruncate(td, uap->fd, uap->length));
608 }
609
610 #if defined(COMPAT_43)
611 #ifndef _SYS_SYSPROTO_H_
612 struct oftruncate_args {
613 int fd;
614 long length;
615 };
616 #endif
617 int
618 oftruncate(td, uap)
619 struct thread *td;
620 struct oftruncate_args *uap;
621 {
622
623 return (kern_ftruncate(td, uap->fd, uap->length));
624 }
625 #endif /* COMPAT_43 */
626
627 #ifndef _SYS_SYSPROTO_H_
628 struct ioctl_args {
629 int fd;
630 u_long com;
631 caddr_t data;
632 };
633 #endif
634 /* ARGSUSED */
635 int
636 sys_ioctl(struct thread *td, struct ioctl_args *uap)
637 {
638 u_long com;
639 int arg, error;
640 u_int size;
641 caddr_t data;
642
643 if (uap->com > 0xffffffff) {
644 printf(
645 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
646 td->td_proc->p_pid, td->td_name, uap->com);
647 uap->com &= 0xffffffff;
648 }
649 com = uap->com;
650
651 /*
652 * Interpret high order word to find amount of data to be
653 * copied to/from the user's address space.
654 */
655 size = IOCPARM_LEN(com);
656 if ((size > IOCPARM_MAX) ||
657 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
658 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
659 ((com & IOC_OUT) && size == 0) ||
660 #else
661 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
662 #endif
663 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
664 return (ENOTTY);
665
666 if (size > 0) {
667 if (com & IOC_VOID) {
668 /* Integer argument. */
669 arg = (intptr_t)uap->data;
670 data = (void *)&arg;
671 size = 0;
672 } else
673 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
674 } else
675 data = (void *)&uap->data;
676 if (com & IOC_IN) {
677 error = copyin(uap->data, data, (u_int)size);
678 if (error) {
679 if (size > 0)
680 free(data, M_IOCTLOPS);
681 return (error);
682 }
683 } else if (com & IOC_OUT) {
684 /*
685 * Zero the buffer so the user always
686 * gets back something deterministic.
687 */
688 bzero(data, size);
689 }
690
691 error = kern_ioctl(td, uap->fd, com, data);
692
693 if (error == 0 && (com & IOC_OUT))
694 error = copyout(data, uap->data, (u_int)size);
695
696 if (size > 0)
697 free(data, M_IOCTLOPS);
698 return (error);
699 }
700
701 int
702 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
703 {
704 struct file *fp;
705 struct filedesc *fdp;
706 int error;
707 int tmp;
708
709 AUDIT_ARG_FD(fd);
710 AUDIT_ARG_CMD(com);
711 if ((error = fget(td, fd, CAP_IOCTL, &fp)) != 0)
712 return (error);
713 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
714 fdrop(fp, td);
715 return (EBADF);
716 }
717 fdp = td->td_proc->p_fd;
718 switch (com) {
719 case FIONCLEX:
720 FILEDESC_XLOCK(fdp);
721 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
722 FILEDESC_XUNLOCK(fdp);
723 goto out;
724 case FIOCLEX:
725 FILEDESC_XLOCK(fdp);
726 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
727 FILEDESC_XUNLOCK(fdp);
728 goto out;
729 case FIONBIO:
730 if ((tmp = *(int *)data))
731 atomic_set_int(&fp->f_flag, FNONBLOCK);
732 else
733 atomic_clear_int(&fp->f_flag, FNONBLOCK);
734 data = (void *)&tmp;
735 break;
736 case FIOASYNC:
737 if ((tmp = *(int *)data))
738 atomic_set_int(&fp->f_flag, FASYNC);
739 else
740 atomic_clear_int(&fp->f_flag, FASYNC);
741 data = (void *)&tmp;
742 break;
743 }
744
745 error = fo_ioctl(fp, com, data, td->td_ucred, td);
746 out:
747 fdrop(fp, td);
748 return (error);
749 }
750
751 int
752 poll_no_poll(int events)
753 {
754 /*
755 * Return true for read/write. If the user asked for something
756 * special, return POLLNVAL, so that clients have a way of
757 * determining reliably whether or not the extended
758 * functionality is present without hard-coding knowledge
759 * of specific filesystem implementations.
760 */
761 if (events & ~POLLSTANDARD)
762 return (POLLNVAL);
763
764 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
765 }
766
767 int
768 sys_pselect(struct thread *td, struct pselect_args *uap)
769 {
770 struct timespec ts;
771 struct timeval tv, *tvp;
772 sigset_t set, *uset;
773 int error;
774
775 if (uap->ts != NULL) {
776 error = copyin(uap->ts, &ts, sizeof(ts));
777 if (error != 0)
778 return (error);
779 TIMESPEC_TO_TIMEVAL(&tv, &ts);
780 tvp = &tv;
781 } else
782 tvp = NULL;
783 if (uap->sm != NULL) {
784 error = copyin(uap->sm, &set, sizeof(set));
785 if (error != 0)
786 return (error);
787 uset = &set;
788 } else
789 uset = NULL;
790 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
791 uset, NFDBITS));
792 }
793
794 int
795 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
796 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
797 {
798 int error;
799
800 if (uset != NULL) {
801 error = kern_sigprocmask(td, SIG_SETMASK, uset,
802 &td->td_oldsigmask, 0);
803 if (error != 0)
804 return (error);
805 td->td_pflags |= TDP_OLDMASK;
806 /*
807 * Make sure that ast() is called on return to
808 * usermode and TDP_OLDMASK is cleared, restoring old
809 * sigmask.
810 */
811 thread_lock(td);
812 td->td_flags |= TDF_ASTPENDING;
813 thread_unlock(td);
814 }
815 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
816 return (error);
817 }
818
819 #ifndef _SYS_SYSPROTO_H_
820 struct select_args {
821 int nd;
822 fd_set *in, *ou, *ex;
823 struct timeval *tv;
824 };
825 #endif
826 int
827 sys_select(struct thread *td, struct select_args *uap)
828 {
829 struct timeval tv, *tvp;
830 int error;
831
832 if (uap->tv != NULL) {
833 error = copyin(uap->tv, &tv, sizeof(tv));
834 if (error)
835 return (error);
836 tvp = &tv;
837 } else
838 tvp = NULL;
839
840 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
841 NFDBITS));
842 }
843
844 /*
845 * In the unlikely case when user specified n greater then the last
846 * open file descriptor, check that no bits are set after the last
847 * valid fd. We must return EBADF if any is set.
848 *
849 * There are applications that rely on the behaviour.
850 *
851 * nd is fd_lastfile + 1.
852 */
853 static int
854 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
855 {
856 char *addr, *oaddr;
857 int b, i, res;
858 uint8_t bits;
859
860 if (nd >= ndu || fd_in == NULL)
861 return (0);
862
863 oaddr = NULL;
864 bits = 0; /* silence gcc */
865 for (i = nd; i < ndu; i++) {
866 b = i / NBBY;
867 #if BYTE_ORDER == LITTLE_ENDIAN
868 addr = (char *)fd_in + b;
869 #else
870 addr = (char *)fd_in;
871 if (abi_nfdbits == NFDBITS) {
872 addr += rounddown(b, sizeof(fd_mask)) +
873 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
874 } else {
875 addr += rounddown(b, sizeof(uint32_t)) +
876 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
877 }
878 #endif
879 if (addr != oaddr) {
880 res = fubyte(addr);
881 if (res == -1)
882 return (EFAULT);
883 oaddr = addr;
884 bits = res;
885 }
886 if ((bits & (1 << (i % NBBY))) != 0)
887 return (EBADF);
888 }
889 return (0);
890 }
891
892 int
893 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
894 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
895 {
896 struct filedesc *fdp;
897 /*
898 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
899 * infds with the new FD_SETSIZE of 1024, and more than enough for
900 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
901 * of 256.
902 */
903 fd_mask s_selbits[howmany(2048, NFDBITS)];
904 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
905 struct timeval atv, rtv, ttv;
906 int error, lf, ndu, timo;
907 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
908
909 if (nd < 0)
910 return (EINVAL);
911 fdp = td->td_proc->p_fd;
912 ndu = nd;
913 lf = fdp->fd_lastfile;
914 if (nd > lf + 1)
915 nd = lf + 1;
916
917 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
918 if (error != 0)
919 return (error);
920 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
921 if (error != 0)
922 return (error);
923 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
924 if (error != 0)
925 return (error);
926
927 /*
928 * Allocate just enough bits for the non-null fd_sets. Use the
929 * preallocated auto buffer if possible.
930 */
931 nfdbits = roundup(nd, NFDBITS);
932 ncpbytes = nfdbits / NBBY;
933 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
934 nbufbytes = 0;
935 if (fd_in != NULL)
936 nbufbytes += 2 * ncpbytes;
937 if (fd_ou != NULL)
938 nbufbytes += 2 * ncpbytes;
939 if (fd_ex != NULL)
940 nbufbytes += 2 * ncpbytes;
941 if (nbufbytes <= sizeof s_selbits)
942 selbits = &s_selbits[0];
943 else
944 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
945
946 /*
947 * Assign pointers into the bit buffers and fetch the input bits.
948 * Put the output buffers together so that they can be bzeroed
949 * together.
950 */
951 sbp = selbits;
952 #define getbits(name, x) \
953 do { \
954 if (name == NULL) { \
955 ibits[x] = NULL; \
956 obits[x] = NULL; \
957 } else { \
958 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
959 obits[x] = sbp; \
960 sbp += ncpbytes / sizeof *sbp; \
961 error = copyin(name, ibits[x], ncpubytes); \
962 if (error != 0) \
963 goto done; \
964 bzero((char *)ibits[x] + ncpubytes, \
965 ncpbytes - ncpubytes); \
966 } \
967 } while (0)
968 getbits(fd_in, 0);
969 getbits(fd_ou, 1);
970 getbits(fd_ex, 2);
971 #undef getbits
972
973 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
974 /*
975 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
976 * we are running under 32-bit emulation. This should be more
977 * generic.
978 */
979 #define swizzle_fdset(bits) \
980 if (abi_nfdbits != NFDBITS && bits != NULL) { \
981 int i; \
982 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
983 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
984 }
985 #else
986 #define swizzle_fdset(bits)
987 #endif
988
989 /* Make sure the bit order makes it through an ABI transition */
990 swizzle_fdset(ibits[0]);
991 swizzle_fdset(ibits[1]);
992 swizzle_fdset(ibits[2]);
993
994 if (nbufbytes != 0)
995 bzero(selbits, nbufbytes / 2);
996
997 if (tvp != NULL) {
998 atv = *tvp;
999 if (itimerfix(&atv)) {
1000 error = EINVAL;
1001 goto done;
1002 }
1003 getmicrouptime(&rtv);
1004 timevaladd(&atv, &rtv);
1005 } else {
1006 atv.tv_sec = 0;
1007 atv.tv_usec = 0;
1008 }
1009 timo = 0;
1010 seltdinit(td);
1011 /* Iterate until the timeout expires or descriptors become ready. */
1012 for (;;) {
1013 error = selscan(td, ibits, obits, nd);
1014 if (error || td->td_retval[0] != 0)
1015 break;
1016 if (atv.tv_sec || atv.tv_usec) {
1017 getmicrouptime(&rtv);
1018 if (timevalcmp(&rtv, &atv, >=))
1019 break;
1020 ttv = atv;
1021 timevalsub(&ttv, &rtv);
1022 timo = ttv.tv_sec > 24 * 60 * 60 ?
1023 24 * 60 * 60 * hz : tvtohz(&ttv);
1024 }
1025 error = seltdwait(td, timo);
1026 if (error)
1027 break;
1028 error = selrescan(td, ibits, obits);
1029 if (error || td->td_retval[0] != 0)
1030 break;
1031 }
1032 seltdclear(td);
1033
1034 done:
1035 /* select is not restarted after signals... */
1036 if (error == ERESTART)
1037 error = EINTR;
1038 if (error == EWOULDBLOCK)
1039 error = 0;
1040
1041 /* swizzle bit order back, if necessary */
1042 swizzle_fdset(obits[0]);
1043 swizzle_fdset(obits[1]);
1044 swizzle_fdset(obits[2]);
1045 #undef swizzle_fdset
1046
1047 #define putbits(name, x) \
1048 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1049 error = error2;
1050 if (error == 0) {
1051 int error2;
1052
1053 putbits(fd_in, 0);
1054 putbits(fd_ou, 1);
1055 putbits(fd_ex, 2);
1056 #undef putbits
1057 }
1058 if (selbits != &s_selbits[0])
1059 free(selbits, M_SELECT);
1060
1061 return (error);
1062 }
1063 /*
1064 * Convert a select bit set to poll flags.
1065 *
1066 * The backend always returns POLLHUP/POLLERR if appropriate and we
1067 * return this as a set bit in any set.
1068 */
1069 static int select_flags[3] = {
1070 POLLRDNORM | POLLHUP | POLLERR,
1071 POLLWRNORM | POLLHUP | POLLERR,
1072 POLLRDBAND | POLLERR
1073 };
1074
1075 /*
1076 * Compute the fo_poll flags required for a fd given by the index and
1077 * bit position in the fd_mask array.
1078 */
1079 static __inline int
1080 selflags(fd_mask **ibits, int idx, fd_mask bit)
1081 {
1082 int flags;
1083 int msk;
1084
1085 flags = 0;
1086 for (msk = 0; msk < 3; msk++) {
1087 if (ibits[msk] == NULL)
1088 continue;
1089 if ((ibits[msk][idx] & bit) == 0)
1090 continue;
1091 flags |= select_flags[msk];
1092 }
1093 return (flags);
1094 }
1095
1096 /*
1097 * Set the appropriate output bits given a mask of fired events and the
1098 * input bits originally requested.
1099 */
1100 static __inline int
1101 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1102 {
1103 int msk;
1104 int n;
1105
1106 n = 0;
1107 for (msk = 0; msk < 3; msk++) {
1108 if ((events & select_flags[msk]) == 0)
1109 continue;
1110 if (ibits[msk] == NULL)
1111 continue;
1112 if ((ibits[msk][idx] & bit) == 0)
1113 continue;
1114 /*
1115 * XXX Check for a duplicate set. This can occur because a
1116 * socket calls selrecord() twice for each poll() call
1117 * resulting in two selfds per real fd. selrescan() will
1118 * call selsetbits twice as a result.
1119 */
1120 if ((obits[msk][idx] & bit) != 0)
1121 continue;
1122 obits[msk][idx] |= bit;
1123 n++;
1124 }
1125
1126 return (n);
1127 }
1128
1129 static __inline int
1130 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1131 {
1132 struct file *fp;
1133 #ifdef CAPABILITIES
1134 struct file *fp_fromcap;
1135 int error;
1136 #endif
1137
1138 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1139 return (EBADF);
1140 #ifdef CAPABILITIES
1141 /*
1142 * If the file descriptor is for a capability, test rights and use
1143 * the file descriptor references by the capability.
1144 */
1145 error = cap_funwrap(fp, CAP_POLL_EVENT, &fp_fromcap);
1146 if (error) {
1147 fdrop(fp, curthread);
1148 return (error);
1149 }
1150 if (fp != fp_fromcap) {
1151 fhold(fp_fromcap);
1152 fdrop(fp, curthread);
1153 fp = fp_fromcap;
1154 }
1155 #endif /* CAPABILITIES */
1156 *fpp = fp;
1157 return (0);
1158 }
1159
1160 /*
1161 * Traverse the list of fds attached to this thread's seltd and check for
1162 * completion.
1163 */
1164 static int
1165 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1166 {
1167 struct filedesc *fdp;
1168 struct selinfo *si;
1169 struct seltd *stp;
1170 struct selfd *sfp;
1171 struct selfd *sfn;
1172 struct file *fp;
1173 fd_mask bit;
1174 int fd, ev, n, idx;
1175 int error;
1176
1177 fdp = td->td_proc->p_fd;
1178 stp = td->td_sel;
1179 n = 0;
1180 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1181 fd = (int)(uintptr_t)sfp->sf_cookie;
1182 si = sfp->sf_si;
1183 selfdfree(stp, sfp);
1184 /* If the selinfo wasn't cleared the event didn't fire. */
1185 if (si != NULL)
1186 continue;
1187 error = getselfd_cap(fdp, fd, &fp);
1188 if (error)
1189 return (error);
1190 idx = fd / NFDBITS;
1191 bit = (fd_mask)1 << (fd % NFDBITS);
1192 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1193 fdrop(fp, td);
1194 if (ev != 0)
1195 n += selsetbits(ibits, obits, idx, bit, ev);
1196 }
1197 stp->st_flags = 0;
1198 td->td_retval[0] = n;
1199 return (0);
1200 }
1201
1202 /*
1203 * Perform the initial filedescriptor scan and register ourselves with
1204 * each selinfo.
1205 */
1206 static int
1207 selscan(td, ibits, obits, nfd)
1208 struct thread *td;
1209 fd_mask **ibits, **obits;
1210 int nfd;
1211 {
1212 struct filedesc *fdp;
1213 struct file *fp;
1214 fd_mask bit;
1215 int ev, flags, end, fd;
1216 int n, idx;
1217 int error;
1218
1219 fdp = td->td_proc->p_fd;
1220 n = 0;
1221 for (idx = 0, fd = 0; fd < nfd; idx++) {
1222 end = imin(fd + NFDBITS, nfd);
1223 for (bit = 1; fd < end; bit <<= 1, fd++) {
1224 /* Compute the list of events we're interested in. */
1225 flags = selflags(ibits, idx, bit);
1226 if (flags == 0)
1227 continue;
1228 error = getselfd_cap(fdp, fd, &fp);
1229 if (error)
1230 return (error);
1231 selfdalloc(td, (void *)(uintptr_t)fd);
1232 ev = fo_poll(fp, flags, td->td_ucred, td);
1233 fdrop(fp, td);
1234 if (ev != 0)
1235 n += selsetbits(ibits, obits, idx, bit, ev);
1236 }
1237 }
1238
1239 td->td_retval[0] = n;
1240 return (0);
1241 }
1242
1243 #ifndef _SYS_SYSPROTO_H_
1244 struct poll_args {
1245 struct pollfd *fds;
1246 u_int nfds;
1247 int timeout;
1248 };
1249 #endif
1250 int
1251 sys_poll(td, uap)
1252 struct thread *td;
1253 struct poll_args *uap;
1254 {
1255 struct pollfd *bits;
1256 struct pollfd smallbits[32];
1257 struct timeval atv, rtv, ttv;
1258 int error, timo;
1259 u_int nfds;
1260 size_t ni;
1261
1262 nfds = uap->nfds;
1263 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1264 return (EINVAL);
1265 ni = nfds * sizeof(struct pollfd);
1266 if (ni > sizeof(smallbits))
1267 bits = malloc(ni, M_TEMP, M_WAITOK);
1268 else
1269 bits = smallbits;
1270 error = copyin(uap->fds, bits, ni);
1271 if (error)
1272 goto done;
1273 if (uap->timeout != INFTIM) {
1274 atv.tv_sec = uap->timeout / 1000;
1275 atv.tv_usec = (uap->timeout % 1000) * 1000;
1276 if (itimerfix(&atv)) {
1277 error = EINVAL;
1278 goto done;
1279 }
1280 getmicrouptime(&rtv);
1281 timevaladd(&atv, &rtv);
1282 } else {
1283 atv.tv_sec = 0;
1284 atv.tv_usec = 0;
1285 }
1286 timo = 0;
1287 seltdinit(td);
1288 /* Iterate until the timeout expires or descriptors become ready. */
1289 for (;;) {
1290 error = pollscan(td, bits, nfds);
1291 if (error || td->td_retval[0] != 0)
1292 break;
1293 if (atv.tv_sec || atv.tv_usec) {
1294 getmicrouptime(&rtv);
1295 if (timevalcmp(&rtv, &atv, >=))
1296 break;
1297 ttv = atv;
1298 timevalsub(&ttv, &rtv);
1299 timo = ttv.tv_sec > 24 * 60 * 60 ?
1300 24 * 60 * 60 * hz : tvtohz(&ttv);
1301 }
1302 error = seltdwait(td, timo);
1303 if (error)
1304 break;
1305 error = pollrescan(td);
1306 if (error || td->td_retval[0] != 0)
1307 break;
1308 }
1309 seltdclear(td);
1310
1311 done:
1312 /* poll is not restarted after signals... */
1313 if (error == ERESTART)
1314 error = EINTR;
1315 if (error == EWOULDBLOCK)
1316 error = 0;
1317 if (error == 0) {
1318 error = pollout(td, bits, uap->fds, nfds);
1319 if (error)
1320 goto out;
1321 }
1322 out:
1323 if (ni > sizeof(smallbits))
1324 free(bits, M_TEMP);
1325 return (error);
1326 }
1327
1328 static int
1329 pollrescan(struct thread *td)
1330 {
1331 struct seltd *stp;
1332 struct selfd *sfp;
1333 struct selfd *sfn;
1334 struct selinfo *si;
1335 struct filedesc *fdp;
1336 struct file *fp;
1337 struct pollfd *fd;
1338 int n;
1339
1340 n = 0;
1341 fdp = td->td_proc->p_fd;
1342 stp = td->td_sel;
1343 FILEDESC_SLOCK(fdp);
1344 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1345 fd = (struct pollfd *)sfp->sf_cookie;
1346 si = sfp->sf_si;
1347 selfdfree(stp, sfp);
1348 /* If the selinfo wasn't cleared the event didn't fire. */
1349 if (si != NULL)
1350 continue;
1351 fp = fdp->fd_ofiles[fd->fd];
1352 #ifdef CAPABILITIES
1353 if ((fp == NULL)
1354 || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
1355 #else
1356 if (fp == NULL) {
1357 #endif
1358 fd->revents = POLLNVAL;
1359 n++;
1360 continue;
1361 }
1362
1363 /*
1364 * Note: backend also returns POLLHUP and
1365 * POLLERR if appropriate.
1366 */
1367 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1368 if (fd->revents != 0)
1369 n++;
1370 }
1371 FILEDESC_SUNLOCK(fdp);
1372 stp->st_flags = 0;
1373 td->td_retval[0] = n;
1374 return (0);
1375 }
1376
1377
1378 static int
1379 pollout(td, fds, ufds, nfd)
1380 struct thread *td;
1381 struct pollfd *fds;
1382 struct pollfd *ufds;
1383 u_int nfd;
1384 {
1385 int error = 0;
1386 u_int i = 0;
1387 u_int n = 0;
1388
1389 for (i = 0; i < nfd; i++) {
1390 error = copyout(&fds->revents, &ufds->revents,
1391 sizeof(ufds->revents));
1392 if (error)
1393 return (error);
1394 if (fds->revents != 0)
1395 n++;
1396 fds++;
1397 ufds++;
1398 }
1399 td->td_retval[0] = n;
1400 return (0);
1401 }
1402
1403 static int
1404 pollscan(td, fds, nfd)
1405 struct thread *td;
1406 struct pollfd *fds;
1407 u_int nfd;
1408 {
1409 struct filedesc *fdp = td->td_proc->p_fd;
1410 int i;
1411 struct file *fp;
1412 int n = 0;
1413
1414 FILEDESC_SLOCK(fdp);
1415 for (i = 0; i < nfd; i++, fds++) {
1416 if (fds->fd >= fdp->fd_nfiles) {
1417 fds->revents = POLLNVAL;
1418 n++;
1419 } else if (fds->fd < 0) {
1420 fds->revents = 0;
1421 } else {
1422 fp = fdp->fd_ofiles[fds->fd];
1423 #ifdef CAPABILITIES
1424 if ((fp == NULL)
1425 || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
1426 #else
1427 if (fp == NULL) {
1428 #endif
1429 fds->revents = POLLNVAL;
1430 n++;
1431 } else {
1432 /*
1433 * Note: backend also returns POLLHUP and
1434 * POLLERR if appropriate.
1435 */
1436 selfdalloc(td, fds);
1437 fds->revents = fo_poll(fp, fds->events,
1438 td->td_ucred, td);
1439 /*
1440 * POSIX requires POLLOUT to be never
1441 * set simultaneously with POLLHUP.
1442 */
1443 if ((fds->revents & POLLHUP) != 0)
1444 fds->revents &= ~POLLOUT;
1445
1446 if (fds->revents != 0)
1447 n++;
1448 }
1449 }
1450 }
1451 FILEDESC_SUNLOCK(fdp);
1452 td->td_retval[0] = n;
1453 return (0);
1454 }
1455
1456 /*
1457 * OpenBSD poll system call.
1458 *
1459 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1460 */
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct openbsd_poll_args {
1463 struct pollfd *fds;
1464 u_int nfds;
1465 int timeout;
1466 };
1467 #endif
1468 int
1469 sys_openbsd_poll(td, uap)
1470 register struct thread *td;
1471 register struct openbsd_poll_args *uap;
1472 {
1473 return (sys_poll(td, (struct poll_args *)uap));
1474 }
1475
1476 /*
1477 * XXX This was created specifically to support netncp and netsmb. This
1478 * allows the caller to specify a socket to wait for events on. It returns
1479 * 0 if any events matched and an error otherwise. There is no way to
1480 * determine which events fired.
1481 */
1482 int
1483 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1484 {
1485 struct timeval atv, rtv, ttv;
1486 int error, timo;
1487
1488 if (tvp != NULL) {
1489 atv = *tvp;
1490 if (itimerfix(&atv))
1491 return (EINVAL);
1492 getmicrouptime(&rtv);
1493 timevaladd(&atv, &rtv);
1494 } else {
1495 atv.tv_sec = 0;
1496 atv.tv_usec = 0;
1497 }
1498
1499 timo = 0;
1500 seltdinit(td);
1501 /*
1502 * Iterate until the timeout expires or the socket becomes ready.
1503 */
1504 for (;;) {
1505 selfdalloc(td, NULL);
1506 error = sopoll(so, events, NULL, td);
1507 /* error here is actually the ready events. */
1508 if (error)
1509 return (0);
1510 if (atv.tv_sec || atv.tv_usec) {
1511 getmicrouptime(&rtv);
1512 if (timevalcmp(&rtv, &atv, >=)) {
1513 seltdclear(td);
1514 return (EWOULDBLOCK);
1515 }
1516 ttv = atv;
1517 timevalsub(&ttv, &rtv);
1518 timo = ttv.tv_sec > 24 * 60 * 60 ?
1519 24 * 60 * 60 * hz : tvtohz(&ttv);
1520 }
1521 error = seltdwait(td, timo);
1522 seltdclear(td);
1523 if (error)
1524 break;
1525 }
1526 /* XXX Duplicates ncp/smb behavior. */
1527 if (error == ERESTART)
1528 error = 0;
1529 return (error);
1530 }
1531
1532 /*
1533 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1534 * have two select sets, one for read and another for write.
1535 */
1536 static void
1537 selfdalloc(struct thread *td, void *cookie)
1538 {
1539 struct seltd *stp;
1540
1541 stp = td->td_sel;
1542 if (stp->st_free1 == NULL)
1543 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1544 stp->st_free1->sf_td = stp;
1545 stp->st_free1->sf_cookie = cookie;
1546 if (stp->st_free2 == NULL)
1547 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1548 stp->st_free2->sf_td = stp;
1549 stp->st_free2->sf_cookie = cookie;
1550 }
1551
1552 static void
1553 selfdfree(struct seltd *stp, struct selfd *sfp)
1554 {
1555 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1556 mtx_lock(sfp->sf_mtx);
1557 if (sfp->sf_si)
1558 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1559 mtx_unlock(sfp->sf_mtx);
1560 uma_zfree(selfd_zone, sfp);
1561 }
1562
1563 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1564 void
1565 seldrain(sip)
1566 struct selinfo *sip;
1567 {
1568
1569 /*
1570 * This feature is already provided by doselwakeup(), thus it is
1571 * enough to go for it.
1572 * Eventually, the context, should take care to avoid races
1573 * between thread calling select()/poll() and file descriptor
1574 * detaching, but, again, the races are just the same as
1575 * selwakeup().
1576 */
1577 doselwakeup(sip, -1);
1578 }
1579
1580 /*
1581 * Record a select request.
1582 */
1583 void
1584 selrecord(selector, sip)
1585 struct thread *selector;
1586 struct selinfo *sip;
1587 {
1588 struct selfd *sfp;
1589 struct seltd *stp;
1590 struct mtx *mtxp;
1591
1592 stp = selector->td_sel;
1593 /*
1594 * Don't record when doing a rescan.
1595 */
1596 if (stp->st_flags & SELTD_RESCAN)
1597 return;
1598 /*
1599 * Grab one of the preallocated descriptors.
1600 */
1601 sfp = NULL;
1602 if ((sfp = stp->st_free1) != NULL)
1603 stp->st_free1 = NULL;
1604 else if ((sfp = stp->st_free2) != NULL)
1605 stp->st_free2 = NULL;
1606 else
1607 panic("selrecord: No free selfd on selq");
1608 mtxp = sip->si_mtx;
1609 if (mtxp == NULL)
1610 mtxp = mtx_pool_find(mtxpool_select, sip);
1611 /*
1612 * Initialize the sfp and queue it in the thread.
1613 */
1614 sfp->sf_si = sip;
1615 sfp->sf_mtx = mtxp;
1616 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1617 /*
1618 * Now that we've locked the sip, check for initialization.
1619 */
1620 mtx_lock(mtxp);
1621 if (sip->si_mtx == NULL) {
1622 sip->si_mtx = mtxp;
1623 TAILQ_INIT(&sip->si_tdlist);
1624 }
1625 /*
1626 * Add this thread to the list of selfds listening on this selinfo.
1627 */
1628 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1629 mtx_unlock(sip->si_mtx);
1630 }
1631
1632 /* Wake up a selecting thread. */
1633 void
1634 selwakeup(sip)
1635 struct selinfo *sip;
1636 {
1637 doselwakeup(sip, -1);
1638 }
1639
1640 /* Wake up a selecting thread, and set its priority. */
1641 void
1642 selwakeuppri(sip, pri)
1643 struct selinfo *sip;
1644 int pri;
1645 {
1646 doselwakeup(sip, pri);
1647 }
1648
1649 /*
1650 * Do a wakeup when a selectable event occurs.
1651 */
1652 static void
1653 doselwakeup(sip, pri)
1654 struct selinfo *sip;
1655 int pri;
1656 {
1657 struct selfd *sfp;
1658 struct selfd *sfn;
1659 struct seltd *stp;
1660
1661 /* If it's not initialized there can't be any waiters. */
1662 if (sip->si_mtx == NULL)
1663 return;
1664 /*
1665 * Locking the selinfo locks all selfds associated with it.
1666 */
1667 mtx_lock(sip->si_mtx);
1668 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1669 /*
1670 * Once we remove this sfp from the list and clear the
1671 * sf_si seltdclear will know to ignore this si.
1672 */
1673 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1674 sfp->sf_si = NULL;
1675 stp = sfp->sf_td;
1676 mtx_lock(&stp->st_mtx);
1677 stp->st_flags |= SELTD_PENDING;
1678 cv_broadcastpri(&stp->st_wait, pri);
1679 mtx_unlock(&stp->st_mtx);
1680 }
1681 mtx_unlock(sip->si_mtx);
1682 }
1683
1684 static void
1685 seltdinit(struct thread *td)
1686 {
1687 struct seltd *stp;
1688
1689 if ((stp = td->td_sel) != NULL)
1690 goto out;
1691 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1692 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1693 cv_init(&stp->st_wait, "select");
1694 out:
1695 stp->st_flags = 0;
1696 STAILQ_INIT(&stp->st_selq);
1697 }
1698
1699 static int
1700 seltdwait(struct thread *td, int timo)
1701 {
1702 struct seltd *stp;
1703 int error;
1704
1705 stp = td->td_sel;
1706 /*
1707 * An event of interest may occur while we do not hold the seltd
1708 * locked so check the pending flag before we sleep.
1709 */
1710 mtx_lock(&stp->st_mtx);
1711 /*
1712 * Any further calls to selrecord will be a rescan.
1713 */
1714 stp->st_flags |= SELTD_RESCAN;
1715 if (stp->st_flags & SELTD_PENDING) {
1716 mtx_unlock(&stp->st_mtx);
1717 return (0);
1718 }
1719 if (timo > 0)
1720 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1721 else
1722 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1723 mtx_unlock(&stp->st_mtx);
1724
1725 return (error);
1726 }
1727
1728 void
1729 seltdfini(struct thread *td)
1730 {
1731 struct seltd *stp;
1732
1733 stp = td->td_sel;
1734 if (stp == NULL)
1735 return;
1736 if (stp->st_free1)
1737 uma_zfree(selfd_zone, stp->st_free1);
1738 if (stp->st_free2)
1739 uma_zfree(selfd_zone, stp->st_free2);
1740 td->td_sel = NULL;
1741 free(stp, M_SELECT);
1742 }
1743
1744 /*
1745 * Remove the references to the thread from all of the objects we were
1746 * polling.
1747 */
1748 static void
1749 seltdclear(struct thread *td)
1750 {
1751 struct seltd *stp;
1752 struct selfd *sfp;
1753 struct selfd *sfn;
1754
1755 stp = td->td_sel;
1756 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1757 selfdfree(stp, sfp);
1758 stp->st_flags = 0;
1759 }
1760
1761 static void selectinit(void *);
1762 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1763 static void
1764 selectinit(void *dummy __unused)
1765 {
1766
1767 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1768 NULL, NULL, UMA_ALIGN_PTR, 0);
1769 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1770 }
Cache object: 887504ac8c09da4f5f79c48ef954d7f6
|