kernel - make poll return only explicitly asked for events
[dragonfly.git] / sys / kern / sys_generic.c
blob2e31710a8857dd9496447f38e05b3e065d725c71
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.
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
38 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
39 * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
40 * $DragonFly: src/sys/kern/sys_generic.c,v 1.49 2008/05/05 22:09:44 dillon Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/event.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/socketvar.h>
56 #include <sys/uio.h>
57 #include <sys/kernel.h>
58 #include <sys/kern_syscall.h>
59 #include <sys/malloc.h>
60 #include <sys/mapped_ioctl.h>
61 #include <sys/poll.h>
62 #include <sys/queue.h>
63 #include <sys/resourcevar.h>
64 #include <sys/socketops.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/buf.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71 #include <vm/vm.h>
72 #include <vm/vm_page.h>
74 #include <sys/file2.h>
75 #include <sys/mplock2.h>
76 #include <sys/spinlock2.h>
78 #include <machine/limits.h>
80 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
81 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
82 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
83 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
85 typedef struct kfd_set {
86 fd_mask fds_bits[2];
87 } kfd_set;
89 enum select_copyin_states {
90 COPYIN_READ, COPYIN_WRITE, COPYIN_EXCEPT, COPYIN_DONE };
92 struct select_kevent_copyin_args {
93 kfd_set *read_set;
94 kfd_set *write_set;
95 kfd_set *except_set;
96 int active_set; /* One of select_copyin_states */
97 struct lwp *lwp; /* Pointer to our lwp */
98 int num_fds; /* Number of file descriptors (syscall arg) */
99 int proc_fds; /* Processed fd's (wraps) */
100 int error; /* Returned to userland */
103 struct poll_kevent_copyin_args {
104 struct lwp *lwp;
105 struct pollfd *fds;
106 int nfds;
107 int pfds;
108 int error;
111 static int doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
112 struct timespec *ts, int *res);
113 static int dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
114 int *res);
115 static int dofileread(int, struct file *, struct uio *, int, size_t *);
116 static int dofilewrite(int, struct file *, struct uio *, int, size_t *);
119 * Read system call.
121 * MPSAFE
124 sys_read(struct read_args *uap)
126 struct thread *td = curthread;
127 struct uio auio;
128 struct iovec aiov;
129 int error;
131 if ((ssize_t)uap->nbyte < 0)
132 error = EINVAL;
134 aiov.iov_base = uap->buf;
135 aiov.iov_len = uap->nbyte;
136 auio.uio_iov = &aiov;
137 auio.uio_iovcnt = 1;
138 auio.uio_offset = -1;
139 auio.uio_resid = uap->nbyte;
140 auio.uio_rw = UIO_READ;
141 auio.uio_segflg = UIO_USERSPACE;
142 auio.uio_td = td;
144 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
145 return(error);
149 * Positioned (Pread) read system call
151 * MPSAFE
154 sys_extpread(struct extpread_args *uap)
156 struct thread *td = curthread;
157 struct uio auio;
158 struct iovec aiov;
159 int error;
160 int flags;
162 if ((ssize_t)uap->nbyte < 0)
163 return(EINVAL);
165 aiov.iov_base = uap->buf;
166 aiov.iov_len = uap->nbyte;
167 auio.uio_iov = &aiov;
168 auio.uio_iovcnt = 1;
169 auio.uio_offset = uap->offset;
170 auio.uio_resid = uap->nbyte;
171 auio.uio_rw = UIO_READ;
172 auio.uio_segflg = UIO_USERSPACE;
173 auio.uio_td = td;
175 flags = uap->flags & O_FMASK;
176 if (uap->offset != (off_t)-1)
177 flags |= O_FOFFSET;
179 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
180 return(error);
184 * Scatter read system call.
186 * MPSAFE
189 sys_readv(struct readv_args *uap)
191 struct thread *td = curthread;
192 struct uio auio;
193 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
194 int error;
196 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
197 &auio.uio_resid);
198 if (error)
199 return (error);
200 auio.uio_iov = iov;
201 auio.uio_iovcnt = uap->iovcnt;
202 auio.uio_offset = -1;
203 auio.uio_rw = UIO_READ;
204 auio.uio_segflg = UIO_USERSPACE;
205 auio.uio_td = td;
207 error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
209 iovec_free(&iov, aiov);
210 return (error);
215 * Scatter positioned read system call.
217 * MPSAFE
220 sys_extpreadv(struct extpreadv_args *uap)
222 struct thread *td = curthread;
223 struct uio auio;
224 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
225 int error;
226 int flags;
228 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
229 &auio.uio_resid);
230 if (error)
231 return (error);
232 auio.uio_iov = iov;
233 auio.uio_iovcnt = uap->iovcnt;
234 auio.uio_offset = uap->offset;
235 auio.uio_rw = UIO_READ;
236 auio.uio_segflg = UIO_USERSPACE;
237 auio.uio_td = td;
239 flags = uap->flags & O_FMASK;
240 if (uap->offset != (off_t)-1)
241 flags |= O_FOFFSET;
243 error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
245 iovec_free(&iov, aiov);
246 return(error);
250 * MPSAFE
253 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
255 struct thread *td = curthread;
256 struct proc *p = td->td_proc;
257 struct file *fp;
258 int error;
260 KKASSERT(p);
262 fp = holdfp(p->p_fd, fd, FREAD);
263 if (fp == NULL)
264 return (EBADF);
265 if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
266 error = ESPIPE;
267 } else {
268 error = dofileread(fd, fp, auio, flags, res);
270 fdrop(fp);
271 return(error);
275 * Common code for readv and preadv that reads data in
276 * from a file using the passed in uio, offset, and flags.
278 * MPALMOSTSAFE - ktrace needs help
280 static int
281 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
283 int error;
284 size_t len;
285 #ifdef KTRACE
286 struct thread *td = curthread;
287 struct iovec *ktriov = NULL;
288 struct uio ktruio;
289 #endif
291 #ifdef KTRACE
293 * if tracing, save a copy of iovec
295 if (KTRPOINT(td, KTR_GENIO)) {
296 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
298 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
299 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
300 ktruio = *auio;
302 #endif
303 len = auio->uio_resid;
304 error = fo_read(fp, auio, fp->f_cred, flags);
305 if (error) {
306 if (auio->uio_resid != len && (error == ERESTART ||
307 error == EINTR || error == EWOULDBLOCK))
308 error = 0;
310 #ifdef KTRACE
311 if (ktriov != NULL) {
312 if (error == 0) {
313 ktruio.uio_iov = ktriov;
314 ktruio.uio_resid = len - auio->uio_resid;
315 get_mplock();
316 ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
317 rel_mplock();
319 FREE(ktriov, M_TEMP);
321 #endif
322 if (error == 0)
323 *res = len - auio->uio_resid;
325 return(error);
329 * Write system call
331 * MPSAFE
334 sys_write(struct write_args *uap)
336 struct thread *td = curthread;
337 struct uio auio;
338 struct iovec aiov;
339 int error;
341 if ((ssize_t)uap->nbyte < 0)
342 error = EINVAL;
344 aiov.iov_base = (void *)(uintptr_t)uap->buf;
345 aiov.iov_len = uap->nbyte;
346 auio.uio_iov = &aiov;
347 auio.uio_iovcnt = 1;
348 auio.uio_offset = -1;
349 auio.uio_resid = uap->nbyte;
350 auio.uio_rw = UIO_WRITE;
351 auio.uio_segflg = UIO_USERSPACE;
352 auio.uio_td = td;
354 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
356 return(error);
360 * Pwrite system call
362 * MPSAFE
365 sys_extpwrite(struct extpwrite_args *uap)
367 struct thread *td = curthread;
368 struct uio auio;
369 struct iovec aiov;
370 int error;
371 int flags;
373 if ((ssize_t)uap->nbyte < 0)
374 error = EINVAL;
376 aiov.iov_base = (void *)(uintptr_t)uap->buf;
377 aiov.iov_len = uap->nbyte;
378 auio.uio_iov = &aiov;
379 auio.uio_iovcnt = 1;
380 auio.uio_offset = uap->offset;
381 auio.uio_resid = uap->nbyte;
382 auio.uio_rw = UIO_WRITE;
383 auio.uio_segflg = UIO_USERSPACE;
384 auio.uio_td = td;
386 flags = uap->flags & O_FMASK;
387 if (uap->offset != (off_t)-1)
388 flags |= O_FOFFSET;
389 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
390 return(error);
394 * MPSAFE
397 sys_writev(struct writev_args *uap)
399 struct thread *td = curthread;
400 struct uio auio;
401 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
402 int error;
404 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
405 &auio.uio_resid);
406 if (error)
407 return (error);
408 auio.uio_iov = iov;
409 auio.uio_iovcnt = uap->iovcnt;
410 auio.uio_offset = -1;
411 auio.uio_rw = UIO_WRITE;
412 auio.uio_segflg = UIO_USERSPACE;
413 auio.uio_td = td;
415 error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
417 iovec_free(&iov, aiov);
418 return (error);
423 * Gather positioned write system call
425 * MPSAFE
428 sys_extpwritev(struct extpwritev_args *uap)
430 struct thread *td = curthread;
431 struct uio auio;
432 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
433 int error;
434 int flags;
436 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
437 &auio.uio_resid);
438 if (error)
439 return (error);
440 auio.uio_iov = iov;
441 auio.uio_iovcnt = uap->iovcnt;
442 auio.uio_offset = uap->offset;
443 auio.uio_rw = UIO_WRITE;
444 auio.uio_segflg = UIO_USERSPACE;
445 auio.uio_td = td;
447 flags = uap->flags & O_FMASK;
448 if (uap->offset != (off_t)-1)
449 flags |= O_FOFFSET;
451 error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
453 iovec_free(&iov, aiov);
454 return(error);
458 * MPSAFE
461 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
463 struct thread *td = curthread;
464 struct proc *p = td->td_proc;
465 struct file *fp;
466 int error;
468 KKASSERT(p);
470 fp = holdfp(p->p_fd, fd, FWRITE);
471 if (fp == NULL)
472 return (EBADF);
473 else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
474 error = ESPIPE;
475 } else {
476 error = dofilewrite(fd, fp, auio, flags, res);
479 fdrop(fp);
480 return (error);
484 * Common code for writev and pwritev that writes data to
485 * a file using the passed in uio, offset, and flags.
487 * MPALMOSTSAFE - ktrace needs help
489 static int
490 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
492 struct thread *td = curthread;
493 struct lwp *lp = td->td_lwp;
494 int error;
495 size_t len;
496 #ifdef KTRACE
497 struct iovec *ktriov = NULL;
498 struct uio ktruio;
499 #endif
501 #ifdef KTRACE
503 * if tracing, save a copy of iovec and uio
505 if (KTRPOINT(td, KTR_GENIO)) {
506 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
508 MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
509 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
510 ktruio = *auio;
512 #endif
513 len = auio->uio_resid;
514 error = fo_write(fp, auio, fp->f_cred, flags);
515 if (error) {
516 if (auio->uio_resid != len && (error == ERESTART ||
517 error == EINTR || error == EWOULDBLOCK))
518 error = 0;
519 /* Socket layer is responsible for issuing SIGPIPE. */
520 if (error == EPIPE) {
521 get_mplock();
522 lwpsignal(lp->lwp_proc, lp, SIGPIPE);
523 rel_mplock();
526 #ifdef KTRACE
527 if (ktriov != NULL) {
528 if (error == 0) {
529 ktruio.uio_iov = ktriov;
530 ktruio.uio_resid = len - auio->uio_resid;
531 get_mplock();
532 ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
533 rel_mplock();
535 FREE(ktriov, M_TEMP);
537 #endif
538 if (error == 0)
539 *res = len - auio->uio_resid;
541 return(error);
545 * Ioctl system call
547 * MPALMOSTSAFE
550 sys_ioctl(struct ioctl_args *uap)
552 int error;
554 get_mplock();
555 error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, &uap->sysmsg);
556 rel_mplock();
557 return (error);
560 struct ioctl_map_entry {
561 const char *subsys;
562 struct ioctl_map_range *cmd_ranges;
563 LIST_ENTRY(ioctl_map_entry) entries;
567 * The true heart of all ioctl syscall handlers (native, emulation).
568 * If map != NULL, it will be searched for a matching entry for com,
569 * and appropriate conversions/conversion functions will be utilized.
572 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
573 struct sysmsg *msg)
575 struct thread *td = curthread;
576 struct proc *p = td->td_proc;
577 struct ucred *cred;
578 struct file *fp;
579 struct ioctl_map_range *iomc = NULL;
580 int error;
581 u_int size;
582 u_long ocom = com;
583 caddr_t data, memp;
584 int tmp;
585 #define STK_PARAMS 128
586 union {
587 char stkbuf[STK_PARAMS];
588 long align;
589 } ubuf;
591 KKASSERT(p);
592 cred = td->td_ucred;
594 fp = holdfp(p->p_fd, fd, FREAD|FWRITE);
595 if (fp == NULL)
596 return(EBADF);
598 if (map != NULL) { /* obey translation map */
599 u_long maskcmd;
600 struct ioctl_map_entry *e;
602 maskcmd = com & map->mask;
604 LIST_FOREACH(e, &map->mapping, entries) {
605 for (iomc = e->cmd_ranges; iomc->start != 0 ||
606 iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
607 iomc->mapfunc != NULL;
608 iomc++) {
609 if (maskcmd >= iomc->start &&
610 maskcmd <= iomc->end)
611 break;
614 /* Did we find a match? */
615 if (iomc->start != 0 || iomc->maptocmd != 0 ||
616 iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
617 break;
620 if (iomc == NULL ||
621 (iomc->start == 0 && iomc->maptocmd == 0
622 && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
623 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
624 map->sys, fd, maskcmd,
625 (int)((maskcmd >> 8) & 0xff),
626 (int)(maskcmd & 0xff));
627 error = EINVAL;
628 goto done;
632 * If it's a non-range one to one mapping, maptocmd should be
633 * correct. If it's a ranged one to one mapping, we pass the
634 * original value of com, and for a range mapped to a different
635 * range, we always need a mapping function to translate the
636 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
638 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
639 com = iomc->maptocmd;
640 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
641 if (iomc->mapfunc != NULL)
642 com = iomc->mapfunc(iomc->start, iomc->end,
643 iomc->start, iomc->end,
644 com, com);
645 } else {
646 if (iomc->mapfunc != NULL) {
647 com = iomc->mapfunc(iomc->start, iomc->end,
648 iomc->maptocmd, iomc->maptoend,
649 com, ocom);
650 } else {
651 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
652 map->sys, fd, maskcmd,
653 (int)((maskcmd >> 8) & 0xff),
654 (int)(maskcmd & 0xff));
655 error = EINVAL;
656 goto done;
661 switch (com) {
662 case FIONCLEX:
663 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
664 goto done;
665 case FIOCLEX:
666 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
667 goto done;
671 * Interpret high order word to find amount of data to be
672 * copied to/from the user's address space.
674 size = IOCPARM_LEN(com);
675 if (size > IOCPARM_MAX) {
676 error = ENOTTY;
677 goto done;
680 memp = NULL;
681 if (size > sizeof (ubuf.stkbuf)) {
682 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
683 data = memp;
684 } else {
685 data = ubuf.stkbuf;
687 if ((com & IOC_IN) != 0) {
688 if (size != 0) {
689 error = copyin(uspc_data, data, (size_t)size);
690 if (error) {
691 if (memp != NULL)
692 kfree(memp, M_IOCTLOPS);
693 goto done;
695 } else {
696 *(caddr_t *)data = uspc_data;
698 } else if ((com & IOC_OUT) != 0 && size) {
700 * Zero the buffer so the user always
701 * gets back something deterministic.
703 bzero(data, (size_t)size);
704 } else if ((com & IOC_VOID) != 0) {
705 *(caddr_t *)data = uspc_data;
708 switch (com) {
709 case FIONBIO:
710 if ((tmp = *(int *)data))
711 fp->f_flag |= FNONBLOCK;
712 else
713 fp->f_flag &= ~FNONBLOCK;
714 error = 0;
715 break;
717 case FIOASYNC:
718 if ((tmp = *(int *)data))
719 fp->f_flag |= FASYNC;
720 else
721 fp->f_flag &= ~FASYNC;
722 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
723 break;
725 default:
727 * If there is a override function,
728 * call it instead of directly routing the call
730 if (map != NULL && iomc->wrapfunc != NULL)
731 error = iomc->wrapfunc(fp, com, ocom, data, cred);
732 else
733 error = fo_ioctl(fp, com, data, cred, msg);
735 * Copy any data to user, size was
736 * already set and checked above.
738 if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
739 error = copyout(data, uspc_data, (size_t)size);
740 break;
742 if (memp != NULL)
743 kfree(memp, M_IOCTLOPS);
744 done:
745 fdrop(fp);
746 return(error);
750 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
752 struct ioctl_map_entry *ne;
754 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
755 he->subsys != NULL && *he->subsys != '\0');
757 ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP, M_WAITOK);
759 ne->subsys = he->subsys;
760 ne->cmd_ranges = he->cmd_ranges;
762 LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
764 return(0);
768 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
770 struct ioctl_map_entry *ne;
772 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
774 LIST_FOREACH(ne, &he->map->mapping, entries) {
775 if (ne->cmd_ranges != he->cmd_ranges)
776 continue;
777 LIST_REMOVE(ne, entries);
778 kfree(ne, M_IOCTLMAP);
779 return(0);
781 return(EINVAL);
784 static int nselcoll; /* Select collisions since boot */
785 int selwait;
786 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, "");
787 static int nseldebug;
788 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
791 * Select system call.
793 * MPSAFE
796 sys_select(struct select_args *uap)
798 struct timeval ktv;
799 struct timespec *ktsp, kts;
800 int error;
803 * Get timeout if any.
805 if (uap->tv != NULL) {
806 error = copyin(uap->tv, &ktv, sizeof (ktv));
807 if (error)
808 return (error);
809 TIMEVAL_TO_TIMESPEC(&ktv, &kts);
810 ktsp = &kts;
811 } else {
812 ktsp = NULL;
816 * Do real work.
818 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
819 &uap->sysmsg_result);
821 return (error);
826 * Pselect system call.
828 * MPALMOSTSAFE
831 sys_pselect(struct pselect_args *uap)
833 struct thread *td = curthread;
834 struct lwp *lp = td->td_lwp;
835 struct timespec *ktsp, kts;
836 sigset_t sigmask;
837 int error;
840 * Get timeout if any.
842 if (uap->ts != NULL) {
843 error = copyin(uap->ts, &kts, sizeof (kts));
844 if (error)
845 return (error);
846 ktsp = &kts;
847 } else {
848 ktsp = NULL;
852 * Install temporary signal mask if any provided.
854 if (uap->sigmask != NULL) {
855 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
856 if (error)
857 return (error);
858 get_mplock();
859 lp->lwp_oldsigmask = lp->lwp_sigmask;
860 SIG_CANTMASK(sigmask);
861 lp->lwp_sigmask = sigmask;
862 } else {
863 get_mplock();
867 * Do real job.
869 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
870 &uap->sysmsg_result);
872 if (uap->sigmask != NULL) {
873 /* doselect() responsible for turning ERESTART into EINTR */
874 KKASSERT(error != ERESTART);
875 if (error == EINTR) {
877 * We can't restore the previous signal mask now
878 * because it could block the signal that interrupted
879 * us. So make a note to restore it after executing
880 * the handler.
882 lp->lwp_flag |= LWP_OLDMASK;
883 } else {
885 * No handler to run. Restore previous mask immediately.
887 lp->lwp_sigmask = lp->lwp_oldsigmask;
890 rel_mplock();
892 return (error);
895 static int
896 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
898 struct select_kevent_copyin_args *skap = NULL;
899 struct kevent *kev;
900 int fd;
901 kfd_set *fdp = NULL;
902 short filter = 0;
903 u_int fflags = 0;
905 skap = (struct select_kevent_copyin_args *)arg;
907 if (*events == maxevents)
908 return (0);
910 while (skap->active_set < COPYIN_DONE) {
911 switch (skap->active_set) {
912 case COPYIN_READ:
914 * Register descriptors for the read filter
916 fdp = skap->read_set;
917 filter = EVFILT_READ;
918 fflags = 0;
919 if (fdp)
920 break;
921 ++skap->active_set;
922 skap->proc_fds = 0;
923 /* fall through */
924 case COPYIN_WRITE:
926 * Register descriptors for the write filter
928 fdp = skap->write_set;
929 filter = EVFILT_WRITE;
930 fflags = 0;
931 if (fdp)
932 break;
933 ++skap->active_set;
934 skap->proc_fds = 0;
935 /* fall through */
936 case COPYIN_EXCEPT:
938 * Register descriptors for the exception filter
940 fdp = skap->except_set;
941 filter = EVFILT_EXCEPT;
942 fflags = NOTE_OOB;
943 if (fdp)
944 break;
945 ++skap->active_set;
946 skap->proc_fds = 0;
947 /* fall through */
948 case COPYIN_DONE:
950 * Nothing left to register
952 return(0);
953 /* NOT REACHED */
956 while (skap->proc_fds < skap->num_fds) {
957 fd = skap->proc_fds;
958 if (FD_ISSET(fd, fdp)) {
959 kev = &kevp[*events];
960 EV_SET(kev, fd, filter,
961 EV_ADD|EV_ENABLE,
962 fflags, 0,
963 (void *)(uintptr_t)
964 skap->lwp->lwp_kqueue_serial);
965 FD_CLR(fd, fdp);
966 ++*events;
968 if (nseldebug)
969 kprintf("select fd %d filter %d serial %d\n",
970 fd, filter, skap->lwp->lwp_kqueue_serial);
972 ++skap->proc_fds;
973 if (*events == maxevents)
974 return (0);
976 skap->active_set++;
977 skap->proc_fds = 0;
980 return (0);
983 static int
984 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
986 struct select_kevent_copyin_args *skap;
987 struct kevent kev;
988 int i = 0;
990 skap = (struct select_kevent_copyin_args *)arg;
992 for (i = 0; i < count; ++i) {
994 * Filter out and delete spurious events
996 if ((u_int)(uintptr_t)kevp[i].udata !=
997 skap->lwp->lwp_kqueue_serial) {
998 kev = kevp[i];
999 kev.flags = EV_DISABLE|EV_DELETE;
1000 kqueue_register(&skap->lwp->lwp_kqueue, &kev);
1001 if (nseldebug)
1002 kprintf("select fd %ju mismatched serial %d\n",
1003 (uintmax_t)kevp[i].ident,
1004 skap->lwp->lwp_kqueue_serial);
1005 continue;
1009 * Handle errors
1011 if (kevp[i].flags & EV_ERROR) {
1012 switch(kevp[i].data) {
1013 case EBADF:
1015 * A bad file descriptor is considered a
1016 * fatal error for select, bail out.
1018 skap->error = EBADF;
1019 *res = 0;
1020 return (1);
1021 break;
1022 default:
1024 * Select silently swallows any unknown errors
1025 * for descriptors in the read or write sets.
1027 if (kevp[i].filter != EVFILT_READ &&
1028 kevp[i].filter != EVFILT_WRITE) {
1029 skap->error = kevp[i].data;
1030 *res = 0;
1031 return (1);
1033 break;
1035 if (nseldebug)
1036 kprintf("select fd %ju filter %d error %jd\n",
1037 (uintmax_t)kevp[i].ident,
1038 kevp[i].filter,
1039 (intmax_t)kevp[i].data);
1040 continue;
1043 switch (kevp[i].filter) {
1044 case EVFILT_READ:
1045 FD_SET(kevp[i].ident, skap->read_set);
1046 break;
1047 case EVFILT_WRITE:
1048 FD_SET(kevp[i].ident, skap->write_set);
1049 break;
1050 case EVFILT_EXCEPT:
1051 FD_SET(kevp[i].ident, skap->except_set);
1052 break;
1055 ++*res;
1058 return (0);
1062 * Copy select bits in from userland. Allocate kernel memory if the
1063 * set is large.
1065 static int
1066 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1068 int error;
1070 if (in_set) {
1071 if (bytes < sizeof(*tmp_set))
1072 *out_set = tmp_set;
1073 else
1074 *out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1075 error = copyin(in_set, *out_set, bytes);
1076 } else {
1077 *out_set = NULL;
1078 error = 0;
1080 return (error);
1084 * Copy returned select bits back out to userland.
1086 static int
1087 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1089 int error;
1091 if (in_set) {
1092 error = copyout(in_set, out_set, bytes);
1093 } else {
1094 error = 0;
1096 return (error);
1100 * Common code for sys_select() and sys_pselect().
1102 * in, out and ex are userland pointers. ts must point to validated
1103 * kernel-side timeout value or NULL for infinite timeout. res must
1104 * point to syscall return value.
1106 static int
1107 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1108 struct timespec *ts, int *res)
1110 struct proc *p = curproc;
1111 struct select_kevent_copyin_args *kap, ka;
1112 int bytes, error;
1113 kfd_set read_tmp;
1114 kfd_set write_tmp;
1115 kfd_set except_tmp;
1117 *res = 0;
1118 if (nd < 0)
1119 return (EINVAL);
1120 if (nd > p->p_fd->fd_nfiles) /* limit kmalloc */
1121 nd = p->p_fd->fd_nfiles;
1123 kap = &ka;
1124 kap->lwp = curthread->td_lwp;
1125 kap->num_fds = nd;
1126 kap->proc_fds = 0;
1127 kap->error = 0;
1128 kap->active_set = COPYIN_READ;
1131 * Calculate bytes based on the number of __fd_mask[] array entries
1132 * multiplied by the size of __fd_mask.
1134 bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1136 error = getbits(bytes, read, &kap->read_set, &read_tmp);
1137 if (error == 0)
1138 error = getbits(bytes, write, &kap->write_set, &write_tmp);
1139 if (error == 0)
1140 error = getbits(bytes, except, &kap->except_set, &except_tmp);
1141 if (error)
1142 goto done;
1145 * NOTE: Make sure the max events passed to kern_kevent() is
1146 * effectively unlimited. (nd * 3) accomplishes this.
1148 * (*res) continues to increment as returned events are
1149 * loaded in.
1151 error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1152 select_copyin, select_copyout, ts);
1153 if (error == 0)
1154 error = putbits(bytes, kap->read_set, read);
1155 if (error == 0)
1156 error = putbits(bytes, kap->write_set, write);
1157 if (error == 0)
1158 error = putbits(bytes, kap->except_set, except);
1161 * An error from an individual event that should be passed
1162 * back to userland (EBADF)
1164 if (kap->error)
1165 error = kap->error;
1168 * Clean up.
1170 done:
1171 if (kap->read_set && kap->read_set != &read_tmp)
1172 kfree(kap->read_set, M_SELECT);
1173 if (kap->write_set && kap->write_set != &write_tmp)
1174 kfree(kap->write_set, M_SELECT);
1175 if (kap->except_set && kap->except_set != &except_tmp)
1176 kfree(kap->except_set, M_SELECT);
1178 kap->lwp->lwp_kqueue_serial += kap->num_fds;
1180 return (error);
1184 * Poll system call.
1186 * MPSAFE
1189 sys_poll(struct poll_args *uap)
1191 struct timespec ts, *tsp;
1192 int error;
1194 if (uap->timeout != INFTIM) {
1195 ts.tv_sec = uap->timeout / 1000;
1196 ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1197 tsp = &ts;
1198 } else {
1199 tsp = NULL;
1202 error = dopoll(uap->nfds, uap->fds, tsp, &uap->sysmsg_result);
1204 return (error);
1207 static int
1208 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1210 struct poll_kevent_copyin_args *pkap;
1211 struct pollfd *pfd;
1212 struct kevent *kev;
1213 int kev_count;
1215 pkap = (struct poll_kevent_copyin_args *)arg;
1217 while (pkap->pfds < pkap->nfds) {
1218 pfd = &pkap->fds[pkap->pfds];
1220 /* Clear return events */
1221 pfd->revents = 0;
1223 /* Do not check if fd is equal to -1 */
1224 if (pfd->fd == -1) {
1225 ++pkap->pfds;
1226 continue;
1229 kev_count = 0;
1230 if (pfd->events & (POLLIN | POLLRDNORM))
1231 kev_count++;
1232 if (pfd->events & (POLLOUT | POLLWRNORM))
1233 kev_count++;
1234 if (pfd->events & (POLLPRI | POLLRDBAND))
1235 kev_count++;
1237 if (*events + kev_count > maxevents)
1238 return (0);
1241 * NOTE: A combined serial number and poll array index is
1242 * stored in kev->udata.
1244 kev = &kevp[*events];
1245 if (pfd->events & (POLLIN | POLLRDNORM)) {
1246 EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1247 0, 0, (void *)(uintptr_t)
1248 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1250 if (pfd->events & (POLLOUT | POLLWRNORM)) {
1251 EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1252 0, 0, (void *)(uintptr_t)
1253 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1255 if (pfd->events & (POLLPRI | POLLRDBAND)) {
1256 EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1257 NOTE_OOB, 0,
1258 (void *)(uintptr_t)
1259 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1262 if (nseldebug) {
1263 kprintf("poll index %d/%d fd %d events %08x serial %d\n",
1264 pkap->pfds, pkap->nfds-1, pfd->fd, pfd->events,
1265 pkap->lwp->lwp_kqueue_serial);
1268 ++pkap->pfds;
1269 (*events) += kev_count;
1272 return (0);
1275 static int
1276 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1278 struct poll_kevent_copyin_args *pkap;
1279 struct pollfd *pfd;
1280 struct kevent kev;
1281 int i;
1282 u_int pi;
1284 pkap = (struct poll_kevent_copyin_args *)arg;
1286 for (i = 0; i < count; ++i) {
1288 * Extract the poll array index and delete spurious events.
1289 * We can easily tell if the serial number is incorrect
1290 * by checking whether the extracted index is out of range.
1292 pi = (u_int)(uintptr_t)kevp[i].udata -
1293 (u_int)pkap->lwp->lwp_kqueue_serial;
1295 if (pi >= pkap->nfds) {
1296 kev = kevp[i];
1297 kev.flags = EV_DISABLE|EV_DELETE;
1298 kqueue_register(&pkap->lwp->lwp_kqueue, &kev);
1299 if (nseldebug)
1300 kprintf("poll index %d out of range against serial %d\n",
1301 pi, pkap->lwp->lwp_kqueue_serial);
1302 continue;
1304 pfd = &pkap->fds[pi];
1305 if (kevp[i].ident == pfd->fd) {
1307 * A single descriptor may generate an error against
1308 * more than one filter, make sure to set the appropriate
1309 * flags but do not increment (*res) more than once.
1311 if (kevp[i].flags & EV_ERROR) {
1312 switch(kevp[i].data) {
1313 case EBADF:
1314 /* Bad file descriptor */
1315 if (pfd->revents == 0)
1316 ++*res;
1317 pfd->revents |= POLLNVAL;
1318 break;
1319 default:
1321 * Poll silently swallows any unknown
1322 * errors except in the case of POLLPRI
1323 * (OOB/urgent data).
1325 * ALWAYS filter out EOPNOTSUPP errors
1326 * from filters, common applications
1327 * set POLLPRI|POLLRDBAND and most
1328 * filters do not support EVFILT_EXCEPT.
1330 if (kevp[i].filter != EVFILT_READ &&
1331 kevp[i].filter != EVFILT_WRITE &&
1332 kevp[i].data != EOPNOTSUPP) {
1333 if (pfd->revents == 0)
1334 ++*res;
1335 pfd->revents |= POLLERR;
1337 break;
1339 if (nseldebug) {
1340 kprintf("poll index %d fd %d "
1341 "filter %d error %jd\n",
1342 pi, pfd->fd,
1343 kevp[i].filter,
1344 (intmax_t)kevp[i].data);
1346 continue;
1349 if (kevp[i].flags & EV_EOF) {
1350 if (pfd->revents == 0)
1351 ++*res;
1352 pfd->revents |= POLLHUP;
1353 continue;
1356 switch (kevp[i].filter) {
1357 case EVFILT_READ:
1358 if (pfd->events & POLLIN)
1359 pfd->revents |= POLLIN;
1360 if (pfd->events & POLLRDNORM)
1361 pfd->revents |= POLLRDNORM;
1362 break;
1363 case EVFILT_WRITE:
1364 if (pfd->events & POLLOUT)
1365 pfd->revents |= POLLOUT;
1366 if (pfd->events & POLLWRNORM)
1367 pfd->revents |= POLLWRNORM;
1368 break;
1369 case EVFILT_EXCEPT:
1370 if (pfd->events & POLLPRI)
1371 pfd->revents |= POLLPRI;
1372 if (pfd->events & POLLRDBAND)
1373 pfd->revents |= POLLRDBAND;
1374 break;
1377 if (nseldebug) {
1378 kprintf("poll index %d/%d fd %d revents %08x\n",
1379 pi, pkap->nfds, pfd->fd, pfd->revents);
1382 ++*res;
1383 continue;
1384 } else {
1385 if (nseldebug) {
1386 kprintf("poll index %d mismatch %ju/%d\n",
1387 pi, (uintmax_t)kevp[i].ident, pfd->fd);
1392 return (0);
1395 static int
1396 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res)
1398 struct poll_kevent_copyin_args ka;
1399 struct pollfd sfds[64];
1400 int bytes;
1401 int error;
1403 *res = 0;
1404 if (nfds < 0)
1405 return (EINVAL);
1408 * This is a bit arbitrary but we need to limit internal kmallocs.
1410 if (nfds > maxfilesperproc * 2)
1411 nfds = maxfilesperproc * 2;
1412 bytes = sizeof(struct pollfd) * nfds;
1414 ka.lwp = curthread->td_lwp;
1415 ka.nfds = nfds;
1416 ka.pfds = 0;
1417 ka.error = 0;
1419 if (ka.nfds < 64)
1420 ka.fds = sfds;
1421 else
1422 ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1424 error = copyin(fds, ka.fds, bytes);
1425 if (error == 0)
1426 error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1427 poll_copyin, poll_copyout, ts);
1429 if (error == 0)
1430 error = copyout(ka.fds, fds, bytes);
1432 if (ka.fds != sfds)
1433 kfree(ka.fds, M_SELECT);
1435 ka.lwp->lwp_kqueue_serial += nfds;
1437 return (error);
1440 static int
1441 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1443 return (0);
1446 static int
1447 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1449 ++*res;
1450 return (0);
1453 extern struct fileops socketops;
1455 socket_wait(struct socket *so, struct timespec *ts, int *res)
1457 struct thread *td = curthread;
1458 struct file *fp;
1459 struct kqueue kq;
1460 struct kevent kev;
1461 int error, fd;
1463 if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1464 return (error);
1466 fp->f_type = DTYPE_SOCKET;
1467 fp->f_flag = FREAD | FWRITE;
1468 fp->f_ops = &socketops;
1469 fp->f_data = so;
1470 fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1472 kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1473 EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1474 if ((error = kqueue_register(&kq, &kev)) != 0) {
1475 fdrop(fp);
1476 return (error);
1479 error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1480 socket_wait_copyout, ts);
1482 EV_SET(&kev, fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
1483 kqueue_register(&kq, &kev);
1484 fp->f_ops = &badfileops;
1485 fdrop(fp);
1487 return (error);
1491 * OpenBSD poll system call.
1492 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1494 * MPSAFE
1497 sys_openbsd_poll(struct openbsd_poll_args *uap)
1499 return (sys_poll((struct poll_args *)uap));
1502 /*ARGSUSED*/
1504 seltrue(cdev_t dev, int events)
1506 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));