xref: /dflybsd-src/sys/kern/sys_generic.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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  * 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.
25  *
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.
37  *
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 $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.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/kern_syscall.h>
58 #include <sys/malloc.h>
59 #include <sys/mapped_ioctl.h>
60 #include <sys/poll.h>
61 #include <sys/queue.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/buf.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 #include <vm/vm.h>
70 #include <vm/vm_page.h>
71 #include <sys/file2.h>
72 
73 #include <machine/limits.h>
74 
75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
76 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
77 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
78 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
79 
80 static int 	doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
81 			struct timeval *tv, int *res);
82 static int	pollscan (struct proc *, struct pollfd *, u_int, int *);
83 static int	selscan (struct proc *, fd_mask **, fd_mask **,
84 			int, int *);
85 static int	dofileread(int, struct file *, struct uio *, int, size_t *);
86 static int	dofilewrite(int, struct file *, struct uio *, int, size_t *);
87 
88 /*
89  * Read system call.
90  *
91  * MPSAFE
92  */
93 int
94 sys_read(struct read_args *uap)
95 {
96 	struct thread *td = curthread;
97 	struct uio auio;
98 	struct iovec aiov;
99 	int error;
100 
101 	if ((ssize_t)uap->nbyte < 0)
102 		error = EINVAL;
103 
104 	aiov.iov_base = uap->buf;
105 	aiov.iov_len = uap->nbyte;
106 	auio.uio_iov = &aiov;
107 	auio.uio_iovcnt = 1;
108 	auio.uio_offset = -1;
109 	auio.uio_resid = uap->nbyte;
110 	auio.uio_rw = UIO_READ;
111 	auio.uio_segflg = UIO_USERSPACE;
112 	auio.uio_td = td;
113 
114 	error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
115 	return(error);
116 }
117 
118 /*
119  * Positioned (Pread) read system call
120  *
121  * MPSAFE
122  */
123 int
124 sys_extpread(struct extpread_args *uap)
125 {
126 	struct thread *td = curthread;
127 	struct uio auio;
128 	struct iovec aiov;
129 	int error;
130 	int flags;
131 
132 	if ((ssize_t)uap->nbyte < 0)
133 		return(EINVAL);
134 
135 	aiov.iov_base = uap->buf;
136 	aiov.iov_len = uap->nbyte;
137 	auio.uio_iov = &aiov;
138 	auio.uio_iovcnt = 1;
139 	auio.uio_offset = uap->offset;
140 	auio.uio_resid = uap->nbyte;
141 	auio.uio_rw = UIO_READ;
142 	auio.uio_segflg = UIO_USERSPACE;
143 	auio.uio_td = td;
144 
145 	flags = uap->flags & O_FMASK;
146 	if (uap->offset != (off_t)-1)
147 		flags |= O_FOFFSET;
148 
149 	error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
150 	return(error);
151 }
152 
153 /*
154  * Scatter read system call.
155  *
156  * MPSAFE
157  */
158 int
159 sys_readv(struct readv_args *uap)
160 {
161 	struct thread *td = curthread;
162 	struct uio auio;
163 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
164 	int error;
165 
166 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
167 			     &auio.uio_resid);
168 	if (error)
169 		return (error);
170 	auio.uio_iov = iov;
171 	auio.uio_iovcnt = uap->iovcnt;
172 	auio.uio_offset = -1;
173 	auio.uio_rw = UIO_READ;
174 	auio.uio_segflg = UIO_USERSPACE;
175 	auio.uio_td = td;
176 
177 	error = kern_preadv(uap->fd, &auio, 0, &uap->sysmsg_szresult);
178 
179 	iovec_free(&iov, aiov);
180 	return (error);
181 }
182 
183 
184 /*
185  * Scatter positioned read system call.
186  *
187  * MPSAFE
188  */
189 int
190 sys_extpreadv(struct extpreadv_args *uap)
191 {
192 	struct thread *td = curthread;
193 	struct uio auio;
194 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
195 	int error;
196 	int flags;
197 
198 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
199 			     &auio.uio_resid);
200 	if (error)
201 		return (error);
202 	auio.uio_iov = iov;
203 	auio.uio_iovcnt = uap->iovcnt;
204 	auio.uio_offset = uap->offset;
205 	auio.uio_rw = UIO_READ;
206 	auio.uio_segflg = UIO_USERSPACE;
207 	auio.uio_td = td;
208 
209 	flags = uap->flags & O_FMASK;
210 	if (uap->offset != (off_t)-1)
211 		flags |= O_FOFFSET;
212 
213 	error = kern_preadv(uap->fd, &auio, flags, &uap->sysmsg_szresult);
214 
215 	iovec_free(&iov, aiov);
216 	return(error);
217 }
218 
219 /*
220  * MPSAFE
221  */
222 int
223 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
224 {
225 	struct thread *td = curthread;
226 	struct proc *p = td->td_proc;
227 	struct file *fp;
228 	int error;
229 
230 	KKASSERT(p);
231 
232 	fp = holdfp(p->p_fd, fd, FREAD);
233 	if (fp == NULL)
234 		return (EBADF);
235 	if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
236 		error = ESPIPE;
237 	} else {
238 		error = dofileread(fd, fp, auio, flags, res);
239 	}
240 	fdrop(fp);
241 	return(error);
242 }
243 
244 /*
245  * Common code for readv and preadv that reads data in
246  * from a file using the passed in uio, offset, and flags.
247  *
248  * MPALMOSTSAFE - ktrace needs help
249  */
250 static int
251 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
252 {
253 	struct thread *td = curthread;
254 	int error;
255 	size_t len;
256 #ifdef KTRACE
257 	struct iovec *ktriov = NULL;
258 	struct uio ktruio;
259 #endif
260 
261 #ifdef KTRACE
262 	/*
263 	 * if tracing, save a copy of iovec
264 	 */
265 	if (KTRPOINT(td, KTR_GENIO))  {
266 		int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
267 
268 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
269 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
270 		ktruio = *auio;
271 	}
272 #endif
273 	len = auio->uio_resid;
274 	error = fo_read(fp, auio, fp->f_cred, flags);
275 	if (error) {
276 		if (auio->uio_resid != len && (error == ERESTART ||
277 		    error == EINTR || error == EWOULDBLOCK))
278 			error = 0;
279 	}
280 #ifdef KTRACE
281 	if (ktriov != NULL) {
282 		if (error == 0) {
283 			ktruio.uio_iov = ktriov;
284 			ktruio.uio_resid = len - auio->uio_resid;
285 			get_mplock();
286 			ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
287 			rel_mplock();
288 		}
289 		FREE(ktriov, M_TEMP);
290 	}
291 #endif
292 	if (error == 0)
293 		*res = len - auio->uio_resid;
294 
295 	return(error);
296 }
297 
298 /*
299  * Write system call
300  *
301  * MPSAFE
302  */
303 int
304 sys_write(struct write_args *uap)
305 {
306 	struct thread *td = curthread;
307 	struct uio auio;
308 	struct iovec aiov;
309 	int error;
310 
311 	if ((ssize_t)uap->nbyte < 0)
312 		error = EINVAL;
313 
314 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
315 	aiov.iov_len = uap->nbyte;
316 	auio.uio_iov = &aiov;
317 	auio.uio_iovcnt = 1;
318 	auio.uio_offset = -1;
319 	auio.uio_resid = uap->nbyte;
320 	auio.uio_rw = UIO_WRITE;
321 	auio.uio_segflg = UIO_USERSPACE;
322 	auio.uio_td = td;
323 
324 	error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
325 
326 	return(error);
327 }
328 
329 /*
330  * Pwrite system call
331  *
332  * MPSAFE
333  */
334 int
335 sys_extpwrite(struct extpwrite_args *uap)
336 {
337 	struct thread *td = curthread;
338 	struct uio auio;
339 	struct iovec aiov;
340 	int error;
341 	int flags;
342 
343 	if ((ssize_t)uap->nbyte < 0)
344 		error = EINVAL;
345 
346 	aiov.iov_base = (void *)(uintptr_t)uap->buf;
347 	aiov.iov_len = uap->nbyte;
348 	auio.uio_iov = &aiov;
349 	auio.uio_iovcnt = 1;
350 	auio.uio_offset = uap->offset;
351 	auio.uio_resid = uap->nbyte;
352 	auio.uio_rw = UIO_WRITE;
353 	auio.uio_segflg = UIO_USERSPACE;
354 	auio.uio_td = td;
355 
356 	flags = uap->flags & O_FMASK;
357 	if (uap->offset != (off_t)-1)
358 		flags |= O_FOFFSET;
359 	error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
360 	return(error);
361 }
362 
363 /*
364  * MPSAFE
365  */
366 int
367 sys_writev(struct writev_args *uap)
368 {
369 	struct thread *td = curthread;
370 	struct uio auio;
371 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
372 	int error;
373 
374 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
375 			     &auio.uio_resid);
376 	if (error)
377 		return (error);
378 	auio.uio_iov = iov;
379 	auio.uio_iovcnt = uap->iovcnt;
380 	auio.uio_offset = -1;
381 	auio.uio_rw = UIO_WRITE;
382 	auio.uio_segflg = UIO_USERSPACE;
383 	auio.uio_td = td;
384 
385 	error = kern_pwritev(uap->fd, &auio, 0, &uap->sysmsg_szresult);
386 
387 	iovec_free(&iov, aiov);
388 	return (error);
389 }
390 
391 
392 /*
393  * Gather positioned write system call
394  *
395  * MPSAFE
396  */
397 int
398 sys_extpwritev(struct extpwritev_args *uap)
399 {
400 	struct thread *td = curthread;
401 	struct uio auio;
402 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
403 	int error;
404 	int flags;
405 
406 	error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
407 			     &auio.uio_resid);
408 	if (error)
409 		return (error);
410 	auio.uio_iov = iov;
411 	auio.uio_iovcnt = uap->iovcnt;
412 	auio.uio_offset = uap->offset;
413 	auio.uio_rw = UIO_WRITE;
414 	auio.uio_segflg = UIO_USERSPACE;
415 	auio.uio_td = td;
416 
417 	flags = uap->flags & O_FMASK;
418 	if (uap->offset != (off_t)-1)
419 		flags |= O_FOFFSET;
420 
421 	error = kern_pwritev(uap->fd, &auio, flags, &uap->sysmsg_szresult);
422 
423 	iovec_free(&iov, aiov);
424 	return(error);
425 }
426 
427 /*
428  * MPSAFE
429  */
430 int
431 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
432 {
433 	struct thread *td = curthread;
434 	struct proc *p = td->td_proc;
435 	struct file *fp;
436 	int error;
437 
438 	KKASSERT(p);
439 
440 	fp = holdfp(p->p_fd, fd, FWRITE);
441 	if (fp == NULL)
442 		return (EBADF);
443 	else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
444 		error = ESPIPE;
445 	} else {
446 		error = dofilewrite(fd, fp, auio, flags, res);
447 	}
448 
449 	fdrop(fp);
450 	return (error);
451 }
452 
453 /*
454  * Common code for writev and pwritev that writes data to
455  * a file using the passed in uio, offset, and flags.
456  *
457  * MPALMOSTSAFE - ktrace needs help
458  */
459 static int
460 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
461 {
462 	struct thread *td = curthread;
463 	struct lwp *lp = td->td_lwp;
464 	int error;
465 	size_t len;
466 #ifdef KTRACE
467 	struct iovec *ktriov = NULL;
468 	struct uio ktruio;
469 #endif
470 
471 #ifdef KTRACE
472 	/*
473 	 * if tracing, save a copy of iovec and uio
474 	 */
475 	if (KTRPOINT(td, KTR_GENIO))  {
476 		int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
477 
478 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
479 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
480 		ktruio = *auio;
481 	}
482 #endif
483 	len = auio->uio_resid;
484 	error = fo_write(fp, auio, fp->f_cred, flags);
485 	if (error) {
486 		if (auio->uio_resid != len && (error == ERESTART ||
487 		    error == EINTR || error == EWOULDBLOCK))
488 			error = 0;
489 		/* Socket layer is responsible for issuing SIGPIPE. */
490 		if (error == EPIPE) {
491 			get_mplock();
492 			lwpsignal(lp->lwp_proc, lp, SIGPIPE);
493 			rel_mplock();
494 		}
495 	}
496 #ifdef KTRACE
497 	if (ktriov != NULL) {
498 		if (error == 0) {
499 			ktruio.uio_iov = ktriov;
500 			ktruio.uio_resid = len - auio->uio_resid;
501 			get_mplock();
502 			ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
503 			rel_mplock();
504 		}
505 		FREE(ktriov, M_TEMP);
506 	}
507 #endif
508 	if (error == 0)
509 		*res = len - auio->uio_resid;
510 
511 	return(error);
512 }
513 
514 /*
515  * Ioctl system call
516  *
517  * MPALMOSTSAFE
518  */
519 int
520 sys_ioctl(struct ioctl_args *uap)
521 {
522 	int error;
523 
524 	get_mplock();
525 	error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, &uap->sysmsg);
526 	rel_mplock();
527 	return (error);
528 }
529 
530 struct ioctl_map_entry {
531 	const char *subsys;
532 	struct ioctl_map_range *cmd_ranges;
533 	LIST_ENTRY(ioctl_map_entry) entries;
534 };
535 
536 /*
537  * The true heart of all ioctl syscall handlers (native, emulation).
538  * If map != NULL, it will be searched for a matching entry for com,
539  * and appropriate conversions/conversion functions will be utilized.
540  */
541 int
542 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
543 	     struct sysmsg *msg)
544 {
545 	struct thread *td = curthread;
546 	struct proc *p = td->td_proc;
547 	struct ucred *cred;
548 	struct file *fp;
549 	struct ioctl_map_range *iomc = NULL;
550 	int error;
551 	u_int size;
552 	u_long ocom = com;
553 	caddr_t data, memp;
554 	int tmp;
555 #define STK_PARAMS	128
556 	union {
557 	    char stkbuf[STK_PARAMS];
558 	    long align;
559 	} ubuf;
560 
561 	KKASSERT(p);
562 	cred = td->td_ucred;
563 
564 	fp = holdfp(p->p_fd, fd, FREAD|FWRITE);
565 	if (fp == NULL)
566 		return(EBADF);
567 
568 	if (map != NULL) {	/* obey translation map */
569 		u_long maskcmd;
570 		struct ioctl_map_entry *e;
571 
572 		maskcmd = com & map->mask;
573 
574 		LIST_FOREACH(e, &map->mapping, entries) {
575 			for (iomc = e->cmd_ranges; iomc->start != 0 ||
576 			     iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
577 			     iomc->mapfunc != NULL;
578 			     iomc++) {
579 				if (maskcmd >= iomc->start &&
580 				    maskcmd <= iomc->end)
581 					break;
582 			}
583 
584 			/* Did we find a match? */
585 			if (iomc->start != 0 || iomc->maptocmd != 0 ||
586 			    iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
587 				break;
588 		}
589 
590 		if (iomc == NULL ||
591 		    (iomc->start == 0 && iomc->maptocmd == 0
592 		     && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
593 			kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
594 			       map->sys, fd, maskcmd,
595 			       (int)((maskcmd >> 8) & 0xff),
596 			       (int)(maskcmd & 0xff));
597 			error = EINVAL;
598 			goto done;
599 		}
600 
601 		/*
602 		 * If it's a non-range one to one mapping, maptocmd should be
603 		 * correct. If it's a ranged one to one mapping, we pass the
604 		 * original value of com, and for a range mapped to a different
605 		 * range, we always need a mapping function to translate the
606 		 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
607 		 */
608 		if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
609 			com = iomc->maptocmd;
610 		} else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
611 			if (iomc->mapfunc != NULL)
612 				com = iomc->mapfunc(iomc->start, iomc->end,
613 						    iomc->start, iomc->end,
614 						    com, com);
615 		} else {
616 			if (iomc->mapfunc != NULL) {
617 				com = iomc->mapfunc(iomc->start, iomc->end,
618 						    iomc->maptocmd, iomc->maptoend,
619 						    com, ocom);
620 			} else {
621 				kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
622 				       map->sys, fd, maskcmd,
623 				       (int)((maskcmd >> 8) & 0xff),
624 				       (int)(maskcmd & 0xff));
625 				error = EINVAL;
626 				goto done;
627 			}
628 		}
629 	}
630 
631 	switch (com) {
632 	case FIONCLEX:
633 		error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
634 		goto done;
635 	case FIOCLEX:
636 		error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
637 		goto done;
638 	}
639 
640 	/*
641 	 * Interpret high order word to find amount of data to be
642 	 * copied to/from the user's address space.
643 	 */
644 	size = IOCPARM_LEN(com);
645 	if (size > IOCPARM_MAX) {
646 		error = ENOTTY;
647 		goto done;
648 	}
649 
650 	memp = NULL;
651 	if (size > sizeof (ubuf.stkbuf)) {
652 		memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
653 		data = memp;
654 	} else {
655 		data = ubuf.stkbuf;
656 	}
657 	if ((com & IOC_IN) != 0) {
658 		if (size != 0) {
659 			error = copyin(uspc_data, data, (size_t)size);
660 			if (error) {
661 				if (memp != NULL)
662 					kfree(memp, M_IOCTLOPS);
663 				goto done;
664 			}
665 		} else {
666 			*(caddr_t *)data = uspc_data;
667 		}
668 	} else if ((com & IOC_OUT) != 0 && size) {
669 		/*
670 		 * Zero the buffer so the user always
671 		 * gets back something deterministic.
672 		 */
673 		bzero(data, (size_t)size);
674 	} else if ((com & IOC_VOID) != 0) {
675 		*(caddr_t *)data = uspc_data;
676 	}
677 
678 	switch (com) {
679 	case FIONBIO:
680 		if ((tmp = *(int *)data))
681 			fp->f_flag |= FNONBLOCK;
682 		else
683 			fp->f_flag &= ~FNONBLOCK;
684 		error = 0;
685 		break;
686 
687 	case FIOASYNC:
688 		if ((tmp = *(int *)data))
689 			fp->f_flag |= FASYNC;
690 		else
691 			fp->f_flag &= ~FASYNC;
692 		error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
693 		break;
694 
695 	default:
696 		/*
697 		 *  If there is a override function,
698 		 *  call it instead of directly routing the call
699 		 */
700 		if (map != NULL && iomc->wrapfunc != NULL)
701 			error = iomc->wrapfunc(fp, com, ocom, data, cred);
702 		else
703 			error = fo_ioctl(fp, com, data, cred, msg);
704 		/*
705 		 * Copy any data to user, size was
706 		 * already set and checked above.
707 		 */
708 		if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
709 			error = copyout(data, uspc_data, (size_t)size);
710 		break;
711 	}
712 	if (memp != NULL)
713 		kfree(memp, M_IOCTLOPS);
714 done:
715 	fdrop(fp);
716 	return(error);
717 }
718 
719 int
720 mapped_ioctl_register_handler(struct ioctl_map_handler *he)
721 {
722 	struct ioctl_map_entry *ne;
723 
724 	KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL &&
725 		 he->subsys != NULL && *he->subsys != '\0');
726 
727 	ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP, M_WAITOK);
728 
729 	ne->subsys = he->subsys;
730 	ne->cmd_ranges = he->cmd_ranges;
731 
732 	LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
733 
734 	return(0);
735 }
736 
737 int
738 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
739 {
740 	struct ioctl_map_entry *ne;
741 
742 	KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
743 
744 	LIST_FOREACH(ne, &he->map->mapping, entries) {
745 		if (ne->cmd_ranges != he->cmd_ranges)
746 			continue;
747 		LIST_REMOVE(ne, entries);
748 		kfree(ne, M_IOCTLMAP);
749 		return(0);
750 	}
751 	return(EINVAL);
752 }
753 
754 static int	nselcoll;	/* Select collisions since boot */
755 int	selwait;
756 SYSCTL_INT(_kern, OID_AUTO, nselcoll, CTLFLAG_RD, &nselcoll, 0, "");
757 
758 /*
759  * Select system call.
760  *
761  * MPALMOSTSAFE
762  */
763 int
764 sys_select(struct select_args *uap)
765 {
766 	struct timeval ktv;
767 	struct timeval *ktvp;
768 	int error;
769 
770 	/*
771 	 * Get timeout if any.
772 	 */
773 	if (uap->tv != NULL) {
774 		error = copyin(uap->tv, &ktv, sizeof (ktv));
775 		if (error)
776 			return (error);
777 		error = itimerfix(&ktv);
778 		if (error)
779 			return (error);
780 		ktvp = &ktv;
781 	} else {
782 		ktvp = NULL;
783 	}
784 
785 	/*
786 	 * Do real work.
787 	 */
788 	get_mplock();
789 	error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktvp,
790 			&uap->sysmsg_result);
791 	rel_mplock();
792 
793 	return (error);
794 }
795 
796 
797 /*
798  * Pselect system call.
799  *
800  * MPALMOSTSAFE
801  */
802 int
803 sys_pselect(struct pselect_args *uap)
804 {
805 	struct thread *td = curthread;
806 	struct lwp *lp = td->td_lwp;
807 	struct timespec kts;
808 	struct timeval ktv;
809 	struct timeval *ktvp;
810 	sigset_t sigmask;
811 	int error;
812 
813 	/*
814 	 * Get timeout if any and convert it.
815 	 * Round up during conversion to avoid timeout going off early.
816 	 */
817 	if (uap->ts != NULL) {
818 		error = copyin(uap->ts, &kts, sizeof (kts));
819 		if (error)
820 			return (error);
821 		ktv.tv_sec = kts.tv_sec;
822 		ktv.tv_usec = (kts.tv_nsec + 999) / 1000;
823 		error = itimerfix(&ktv);
824 		if (error)
825 			return (error);
826 		ktvp = &ktv;
827 	} else {
828 		ktvp = NULL;
829 	}
830 
831 	/*
832 	 * Install temporary signal mask if any provided.
833 	 */
834 	if (uap->sigmask != NULL) {
835 		error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
836 		if (error)
837 			return (error);
838 		get_mplock();
839 		lp->lwp_oldsigmask = lp->lwp_sigmask;
840 		SIG_CANTMASK(sigmask);
841 		lp->lwp_sigmask = sigmask;
842 	} else {
843 		get_mplock();
844 	}
845 
846 	/*
847 	 * Do real job.
848 	 */
849 	error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktvp,
850 			&uap->sysmsg_result);
851 
852 	if (uap->sigmask != NULL) {
853 		/* doselect() responsible for turning ERESTART into EINTR */
854 		KKASSERT(error != ERESTART);
855 		if (error == EINTR) {
856 			/*
857 			 * We can't restore the previous signal mask now
858 			 * because it could block the signal that interrupted
859 			 * us.  So make a note to restore it after executing
860 			 * the handler.
861 			 */
862 			lp->lwp_flag |= LWP_OLDMASK;
863 		} else {
864 			/*
865 			 * No handler to run. Restore previous mask immediately.
866 			 */
867 			lp->lwp_sigmask = lp->lwp_oldsigmask;
868 		}
869 	}
870 	rel_mplock();
871 
872 	return (error);
873 }
874 
875 /*
876  * Common code for sys_select() and sys_pselect().
877  *
878  * in, out and ex are userland pointers.  tv must point to validated
879  * kernel-side timeout value or NULL for infinite timeout.  res must
880  * point to syscall return value.
881  */
882 static int
883 doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex, struct timeval *tv,
884 		int *res)
885 {
886 	struct lwp *lp = curthread->td_lwp;
887 	struct proc *p = curproc;
888 
889 	/*
890 	 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
891 	 * infds with the new FD_SETSIZE of 1024, and more than enough for
892 	 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
893 	 * of 256.
894 	 */
895 	fd_mask s_selbits[howmany(2048, NFDBITS)];
896 	fd_mask *ibits[3], *obits[3], *selbits, *sbp;
897 	struct timeval atv, rtv, ttv;
898 	int ncoll, error, timo;
899 	u_int nbufbytes, ncpbytes, nfdbits;
900 
901 	if (nd < 0)
902 		return (EINVAL);
903 	if (nd > p->p_fd->fd_nfiles)
904 		nd = p->p_fd->fd_nfiles;   /* forgiving; slightly wrong */
905 
906 	/*
907 	 * Allocate just enough bits for the non-null fd_sets.  Use the
908 	 * preallocated auto buffer if possible.
909 	 */
910 	nfdbits = roundup(nd, NFDBITS);
911 	ncpbytes = nfdbits / NBBY;
912 	nbufbytes = 0;
913 	if (in != NULL)
914 		nbufbytes += 2 * ncpbytes;
915 	if (ou != NULL)
916 		nbufbytes += 2 * ncpbytes;
917 	if (ex != NULL)
918 		nbufbytes += 2 * ncpbytes;
919 	if (nbufbytes <= sizeof s_selbits)
920 		selbits = &s_selbits[0];
921 	else
922 		selbits = kmalloc(nbufbytes, M_SELECT, M_WAITOK);
923 
924 	/*
925 	 * Assign pointers into the bit buffers and fetch the input bits.
926 	 * Put the output buffers together so that they can be bzeroed
927 	 * together.
928 	 */
929 	sbp = selbits;
930 #define	getbits(name, x) \
931 	do {								\
932 		if (name == NULL)					\
933 			ibits[x] = NULL;				\
934 		else {							\
935 			ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;	\
936 			obits[x] = sbp;					\
937 			sbp += ncpbytes / sizeof *sbp;			\
938 			error = copyin(name, ibits[x], ncpbytes);	\
939 			if (error != 0)					\
940 				goto done;				\
941 		}							\
942 	} while (0)
943 	getbits(in, 0);
944 	getbits(ou, 1);
945 	getbits(ex, 2);
946 #undef	getbits
947 	if (nbufbytes != 0)
948 		bzero(selbits, nbufbytes / 2);
949 
950 	if (tv != NULL) {
951 		atv = *tv;
952 		getmicrouptime(&rtv);
953 		timevaladd(&atv, &rtv);
954 	} else {
955 		atv.tv_sec = 0;
956 		atv.tv_usec = 0;
957 	}
958 	timo = 0;
959 retry:
960 	ncoll = nselcoll;
961 	lp->lwp_flag |= LWP_SELECT;
962 	error = selscan(p, ibits, obits, nd, res);
963 	if (error || *res)
964 		goto done;
965 	if (atv.tv_sec || atv.tv_usec) {
966 		getmicrouptime(&rtv);
967 		if (timevalcmp(&rtv, &atv, >=))
968 			goto done;
969 		ttv = atv;
970 		timevalsub(&ttv, &rtv);
971 		timo = ttv.tv_sec > 24 * 60 * 60 ?
972 		    24 * 60 * 60 * hz : tvtohz_high(&ttv);
973 	}
974 	crit_enter();
975 	if ((lp->lwp_flag & LWP_SELECT) == 0 || nselcoll != ncoll) {
976 		crit_exit();
977 		goto retry;
978 	}
979 	lp->lwp_flag &= ~LWP_SELECT;
980 
981 	error = tsleep((caddr_t)&selwait, PCATCH, "select", timo);
982 
983 	crit_exit();
984 	if (error == 0)
985 		goto retry;
986 done:
987 	lp->lwp_flag &= ~LWP_SELECT;
988 	/* select is not restarted after signals... */
989 	if (error == ERESTART)
990 		error = EINTR;
991 	if (error == EWOULDBLOCK)
992 		error = 0;
993 #define	putbits(name, x) \
994 	if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
995 		error = error2;
996 	if (error == 0) {
997 		int error2;
998 
999 		putbits(in, 0);
1000 		putbits(ou, 1);
1001 		putbits(ex, 2);
1002 #undef putbits
1003 	}
1004 	if (selbits != &s_selbits[0])
1005 		kfree(selbits, M_SELECT);
1006 	return (error);
1007 }
1008 
1009 static int
1010 selscan(struct proc *p, fd_mask **ibits, fd_mask **obits, int nfd, int *res)
1011 {
1012 	int msk, i, fd;
1013 	fd_mask bits;
1014 	struct file *fp;
1015 	int n = 0;
1016 	/* Note: backend also returns POLLHUP/POLLERR if appropriate. */
1017 	static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
1018 
1019 	for (msk = 0; msk < 3; msk++) {
1020 		if (ibits[msk] == NULL)
1021 			continue;
1022 		for (i = 0; i < nfd; i += NFDBITS) {
1023 			bits = ibits[msk][i/NFDBITS];
1024 			/* ffs(int mask) not portable, fd_mask is long */
1025 			for (fd = i; bits && fd < nfd; fd++, bits >>= 1) {
1026 				if (!(bits & 1))
1027 					continue;
1028 				fp = holdfp(p->p_fd, fd, -1);
1029 				if (fp == NULL)
1030 					return (EBADF);
1031 				if (fo_poll(fp, flag[msk], fp->f_cred)) {
1032 					obits[msk][(fd)/NFDBITS] |=
1033 					    ((fd_mask)1 << ((fd) % NFDBITS));
1034 					n++;
1035 				}
1036 				fdrop(fp);
1037 			}
1038 		}
1039 	}
1040 	*res = n;
1041 	return (0);
1042 }
1043 
1044 /*
1045  * Poll system call.
1046  *
1047  * MPALMOSTSAFE
1048  */
1049 int
1050 sys_poll(struct poll_args *uap)
1051 {
1052 	struct pollfd *bits;
1053 	struct pollfd smallbits[32];
1054 	struct timeval atv, rtv, ttv;
1055 	int ncoll, error = 0, timo;
1056 	u_int nfds;
1057 	size_t ni;
1058 	struct lwp *lp = curthread->td_lwp;
1059 	struct proc *p = curproc;
1060 
1061 	nfds = uap->nfds;
1062 	/*
1063 	 * This is kinda bogus.  We have fd limits, but that is not
1064 	 * really related to the size of the pollfd array.  Make sure
1065 	 * we let the process use at least FD_SETSIZE entries and at
1066 	 * least enough for the current limits.  We want to be reasonably
1067 	 * safe, but not overly restrictive.
1068 	 */
1069 	if (nfds > p->p_rlimit[RLIMIT_NOFILE].rlim_cur && nfds > FD_SETSIZE)
1070 		return (EINVAL);
1071 	ni = nfds * sizeof(struct pollfd);
1072 	if (ni > sizeof(smallbits))
1073 		bits = kmalloc(ni, M_TEMP, M_WAITOK);
1074 	else
1075 		bits = smallbits;
1076 	error = copyin(uap->fds, bits, ni);
1077 	if (error)
1078 		goto done;
1079 	if (uap->timeout != INFTIM) {
1080 		atv.tv_sec = uap->timeout / 1000;
1081 		atv.tv_usec = (uap->timeout % 1000) * 1000;
1082 		if (itimerfix(&atv)) {
1083 			error = EINVAL;
1084 			goto done;
1085 		}
1086 		getmicrouptime(&rtv);
1087 		timevaladd(&atv, &rtv);
1088 	} else {
1089 		atv.tv_sec = 0;
1090 		atv.tv_usec = 0;
1091 	}
1092 	timo = 0;
1093 retry:
1094 	ncoll = nselcoll;
1095 	lp->lwp_flag |= LWP_SELECT;
1096 	get_mplock();
1097 	error = pollscan(p, bits, nfds, &uap->sysmsg_result);
1098 	rel_mplock();
1099 	if (error || uap->sysmsg_result)
1100 		goto done;
1101 	if (atv.tv_sec || atv.tv_usec) {
1102 		getmicrouptime(&rtv);
1103 		if (timevalcmp(&rtv, &atv, >=))
1104 			goto done;
1105 		ttv = atv;
1106 		timevalsub(&ttv, &rtv);
1107 		timo = ttv.tv_sec > 24 * 60 * 60 ?
1108 		    24 * 60 * 60 * hz : tvtohz_high(&ttv);
1109 	}
1110 	crit_enter();
1111 	tsleep_interlock(&selwait, PCATCH);
1112 	if ((lp->lwp_flag & LWP_SELECT) == 0 || nselcoll != ncoll) {
1113 		crit_exit();
1114 		goto retry;
1115 	}
1116 	lp->lwp_flag &= ~LWP_SELECT;
1117 	error = tsleep(&selwait, PCATCH | PINTERLOCKED, "poll", timo);
1118 	crit_exit();
1119 	if (error == 0)
1120 		goto retry;
1121 done:
1122 	lp->lwp_flag &= ~LWP_SELECT;
1123 	/* poll is not restarted after signals... */
1124 	if (error == ERESTART)
1125 		error = EINTR;
1126 	if (error == EWOULDBLOCK)
1127 		error = 0;
1128 	if (error == 0) {
1129 		error = copyout(bits, uap->fds, ni);
1130 		if (error)
1131 			goto out;
1132 	}
1133 out:
1134 	if (ni > sizeof(smallbits))
1135 		kfree(bits, M_TEMP);
1136 	return (error);
1137 }
1138 
1139 static int
1140 pollscan(struct proc *p, struct pollfd *fds, u_int nfd, int *res)
1141 {
1142 	int i;
1143 	struct file *fp;
1144 	int n = 0;
1145 
1146 	for (i = 0; i < nfd; i++, fds++) {
1147 		if (fds->fd >= p->p_fd->fd_nfiles) {
1148 			fds->revents = POLLNVAL;
1149 			n++;
1150 		} else if (fds->fd < 0) {
1151 			fds->revents = 0;
1152 		} else {
1153 			fp = holdfp(p->p_fd, fds->fd, -1);
1154 			if (fp == NULL) {
1155 				fds->revents = POLLNVAL;
1156 				n++;
1157 			} else {
1158 				/*
1159 				 * Note: backend also returns POLLHUP and
1160 				 * POLLERR if appropriate.
1161 				 */
1162 				fds->revents = fo_poll(fp, fds->events,
1163 							fp->f_cred);
1164 				if (fds->revents != 0)
1165 					n++;
1166 				fdrop(fp);
1167 			}
1168 		}
1169 	}
1170 	*res = n;
1171 	return (0);
1172 }
1173 
1174 /*
1175  * OpenBSD poll system call.
1176  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
1177  *
1178  * MPSAFE
1179  */
1180 int
1181 sys_openbsd_poll(struct openbsd_poll_args *uap)
1182 {
1183 	return (sys_poll((struct poll_args *)uap));
1184 }
1185 
1186 /*ARGSUSED*/
1187 int
1188 seltrue(cdev_t dev, int events)
1189 {
1190 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1191 }
1192 
1193 /*
1194  * Record a select request.  A global wait must be used since a process/thread
1195  * might go away after recording its request.
1196  */
1197 void
1198 selrecord(struct thread *selector, struct selinfo *sip)
1199 {
1200 	struct proc *p;
1201 	struct lwp *lp = NULL;
1202 
1203 	if (selector->td_lwp == NULL)
1204 		panic("selrecord: thread needs a process");
1205 
1206 	if (sip->si_pid == selector->td_proc->p_pid &&
1207 	    sip->si_tid == selector->td_lwp->lwp_tid)
1208 		return;
1209 	if (sip->si_pid && (p = pfind(sip->si_pid)))
1210 		lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, sip->si_tid);
1211 	if (lp != NULL && lp->lwp_wchan == (caddr_t)&selwait) {
1212 		sip->si_flags |= SI_COLL;
1213 	} else {
1214 		sip->si_pid = selector->td_proc->p_pid;
1215 		sip->si_tid = selector->td_lwp->lwp_tid;
1216 	}
1217 }
1218 
1219 /*
1220  * Do a wakeup when a selectable event occurs.
1221  */
1222 void
1223 selwakeup(struct selinfo *sip)
1224 {
1225 	struct proc *p;
1226 	struct lwp *lp = NULL;
1227 
1228 	if (sip->si_pid == 0)
1229 		return;
1230 	if (sip->si_flags & SI_COLL) {
1231 		nselcoll++;
1232 		sip->si_flags &= ~SI_COLL;
1233 		wakeup((caddr_t)&selwait);	/* YYY fixable */
1234 	}
1235 	p = pfind(sip->si_pid);
1236 	sip->si_pid = 0;
1237 	if (p == NULL)
1238 		return;
1239 	lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, sip->si_tid);
1240 	if (lp == NULL)
1241 		return;
1242 
1243 	crit_enter();
1244 	if (lp->lwp_wchan == (caddr_t)&selwait) {
1245 		/*
1246 		 * Flag the process to break the tsleep when
1247 		 * setrunnable is called, but only call setrunnable
1248 		 * here if the process is not in a stopped state.
1249 		 */
1250 		lp->lwp_flag |= LWP_BREAKTSLEEP;
1251 		if (p->p_stat != SSTOP)
1252 			setrunnable(lp);
1253 	} else if (lp->lwp_flag & LWP_SELECT) {
1254 		lp->lwp_flag &= ~LWP_SELECT;
1255 	}
1256 	crit_exit();
1257 }
1258 
1259