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. 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 * $FreeBSD: src/sys/kern/sys_generic.c,v 1.55.2.10 2001/03/17 10:39:32 peter Exp $
36 */
37
38 #include "opt_ktrace.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysmsg.h>
43 #include <sys/event.h>
44 #include <sys/filedesc.h>
45 #include <sys/filio.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/proc.h>
49 #include <sys/signalvar.h>
50 #include <sys/socketvar.h>
51 #include <sys/malloc.h>
52 #include <sys/uio.h>
53 #include <sys/kernel.h>
54 #include <sys/kern_syscall.h>
55 #include <sys/mapped_ioctl.h>
56 #include <sys/poll.h>
57 #include <sys/queue.h>
58 #include <sys/resourcevar.h>
59 #include <sys/socketops.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/buf.h>
63 #ifdef KTRACE
64 #include <sys/ktrace.h>
65 #endif
66 #include <vm/vm.h>
67 #include <vm/vm_page.h>
68
69 #include <sys/file2.h>
70 #include <sys/spinlock2.h>
71
72 #include <machine/limits.h>
73
74 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
75 static MALLOC_DEFINE(M_IOCTLMAP, "ioctlmap", "mapped ioctl handler buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
78
79 typedef struct kfd_set {
80 fd_mask fds_bits[2];
81 } kfd_set;
82
83 enum select_copyin_states {
84 COPYIN_READ, COPYIN_WRITE, COPYIN_EXCEPT, COPYIN_DONE };
85
86 struct select_kevent_copyin_args {
87 kfd_set *read_set;
88 kfd_set *write_set;
89 kfd_set *except_set;
90 int active_set; /* One of select_copyin_states */
91 struct lwp *lwp; /* Pointer to our lwp */
92 int num_fds; /* Number of file descriptors (syscall arg) */
93 int proc_fds; /* Processed fd's (wraps) */
94 int error; /* Returned to userland */
95 };
96
97 struct poll_kevent_copyin_args {
98 struct lwp *lwp;
99 struct pollfd *fds;
100 int nfds;
101 int pfds;
102 int error;
103 };
104
105 static struct lwkt_token mioctl_token = LWKT_TOKEN_INITIALIZER(mioctl_token);
106
107 static int doselect(int nd, fd_set *in, fd_set *ou, fd_set *ex,
108 struct timespec *ts, int *res);
109 static int dopoll(int nfds, struct pollfd *fds, struct timespec *ts,
110 int *res, int flags);
111 static int dofileread(int, struct file *, struct uio *, int, size_t *);
112 static int dofilewrite(int, struct file *, struct uio *, int, size_t *);
113
114 /*
115 * Read system call.
116 *
117 * MPSAFE
118 */
119 int
sys_read(struct sysmsg * sysmsg,const struct read_args * uap)120 sys_read(struct sysmsg *sysmsg, const struct read_args *uap)
121 {
122 struct thread *td = curthread;
123 struct uio auio;
124 struct iovec aiov;
125 int error;
126
127 if ((ssize_t)uap->nbyte < 0)
128 error = EINVAL;
129
130 aiov.iov_base = uap->buf;
131 aiov.iov_len = uap->nbyte;
132 auio.uio_iov = &aiov;
133 auio.uio_iovcnt = 1;
134 auio.uio_offset = -1;
135 auio.uio_resid = uap->nbyte;
136 auio.uio_rw = UIO_READ;
137 auio.uio_segflg = UIO_USERSPACE;
138 auio.uio_td = td;
139
140 error = kern_preadv(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
141 return(error);
142 }
143
144 /*
145 * Positioned (Pread) read system call
146 *
147 * MPSAFE
148 */
149 int
sys_extpread(struct sysmsg * sysmsg,const struct extpread_args * uap)150 sys_extpread(struct sysmsg *sysmsg, const struct extpread_args *uap)
151 {
152 struct thread *td = curthread;
153 struct uio auio;
154 struct iovec aiov;
155 int error;
156 int flags;
157
158 if ((ssize_t)uap->nbyte < 0)
159 return(EINVAL);
160
161 aiov.iov_base = uap->buf;
162 aiov.iov_len = uap->nbyte;
163 auio.uio_iov = &aiov;
164 auio.uio_iovcnt = 1;
165 auio.uio_offset = uap->offset;
166 auio.uio_resid = uap->nbyte;
167 auio.uio_rw = UIO_READ;
168 auio.uio_segflg = UIO_USERSPACE;
169 auio.uio_td = td;
170
171 flags = uap->flags & O_FMASK;
172 if (uap->offset != (off_t)-1)
173 flags |= O_FOFFSET;
174
175 error = kern_preadv(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
176 return(error);
177 }
178
179 /*
180 * Scatter read system call.
181 *
182 * MPSAFE
183 */
184 int
sys_readv(struct sysmsg * sysmsg,const struct readv_args * uap)185 sys_readv(struct sysmsg *sysmsg, const struct readv_args *uap)
186 {
187 struct thread *td = curthread;
188 struct uio auio;
189 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
190 int error;
191
192 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
193 &auio.uio_resid);
194 if (error)
195 return (error);
196 auio.uio_iov = iov;
197 auio.uio_iovcnt = uap->iovcnt;
198 auio.uio_offset = -1;
199 auio.uio_rw = UIO_READ;
200 auio.uio_segflg = UIO_USERSPACE;
201 auio.uio_td = td;
202
203 error = kern_preadv(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
204
205 iovec_free(&iov, aiov);
206 return (error);
207 }
208
209
210 /*
211 * Scatter positioned read system call.
212 *
213 * MPSAFE
214 */
215 int
sys_extpreadv(struct sysmsg * sysmsg,const struct extpreadv_args * uap)216 sys_extpreadv(struct sysmsg *sysmsg, const struct extpreadv_args *uap)
217 {
218 struct thread *td = curthread;
219 struct uio auio;
220 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
221 int error;
222 int flags;
223
224 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
225 &auio.uio_resid);
226 if (error)
227 return (error);
228 auio.uio_iov = iov;
229 auio.uio_iovcnt = uap->iovcnt;
230 auio.uio_offset = uap->offset;
231 auio.uio_rw = UIO_READ;
232 auio.uio_segflg = UIO_USERSPACE;
233 auio.uio_td = td;
234
235 flags = uap->flags & O_FMASK;
236 if (uap->offset != (off_t)-1)
237 flags |= O_FOFFSET;
238
239 error = kern_preadv(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
240
241 iovec_free(&iov, aiov);
242 return(error);
243 }
244
245 /*
246 * MPSAFE
247 */
248 int
kern_preadv(int fd,struct uio * auio,int flags,size_t * res)249 kern_preadv(int fd, struct uio *auio, int flags, size_t *res)
250 {
251 struct thread *td = curthread;
252 struct file *fp;
253 int error;
254
255 fp = holdfp(td, fd, FREAD);
256 if (fp == NULL)
257 return (EBADF);
258 if (flags & O_FOFFSET && fp->f_type != DTYPE_VNODE) {
259 error = ESPIPE;
260 } else {
261 error = dofileread(fd, fp, auio, flags, res);
262 }
263 dropfp(td, fd, fp);
264
265 return(error);
266 }
267
268 /*
269 * Common code for readv and preadv that reads data in
270 * from a file using the passed in uio, offset, and flags.
271 *
272 * MPALMOSTSAFE - ktrace needs help
273 */
274 static int
dofileread(int fd,struct file * fp,struct uio * auio,int flags,size_t * res)275 dofileread(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
276 {
277 int error;
278 size_t len;
279 #ifdef KTRACE
280 struct thread *td = curthread;
281 struct iovec *ktriov = NULL;
282 struct uio ktruio;
283 #endif
284
285 #ifdef KTRACE
286 /*
287 * if tracing, save a copy of iovec
288 */
289 if (KTRPOINT(td, KTR_GENIO)) {
290 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
291
292 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
293 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
294 ktruio = *auio;
295 }
296 #endif
297 len = auio->uio_resid;
298 error = fo_read(fp, auio, fp->f_cred, flags);
299 if (error) {
300 if (auio->uio_resid != len && (error == ERESTART ||
301 error == EINTR || error == EWOULDBLOCK))
302 error = 0;
303 }
304 #ifdef KTRACE
305 if (ktriov != NULL) {
306 if (error == 0) {
307 ktruio.uio_iov = ktriov;
308 ktruio.uio_resid = len - auio->uio_resid;
309 ktrgenio(td->td_lwp, fd, UIO_READ, &ktruio, error);
310 }
311 kfree(ktriov, M_TEMP);
312 }
313 #endif
314 if (error == 0)
315 *res = len - auio->uio_resid;
316
317 return(error);
318 }
319
320 /*
321 * Write system call
322 *
323 * MPSAFE
324 */
325 int
sys_write(struct sysmsg * sysmsg,const struct write_args * uap)326 sys_write(struct sysmsg *sysmsg, const struct write_args *uap)
327 {
328 struct thread *td = curthread;
329 struct uio auio;
330 struct iovec aiov;
331 int error;
332
333 if ((ssize_t)uap->nbyte < 0)
334 error = EINVAL;
335
336 aiov.iov_base = (void *)(uintptr_t)uap->buf;
337 aiov.iov_len = uap->nbyte;
338 auio.uio_iov = &aiov;
339 auio.uio_iovcnt = 1;
340 auio.uio_offset = -1;
341 auio.uio_resid = uap->nbyte;
342 auio.uio_rw = UIO_WRITE;
343 auio.uio_segflg = UIO_USERSPACE;
344 auio.uio_td = td;
345
346 error = kern_pwritev(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
347
348 return(error);
349 }
350
351 /*
352 * Pwrite system call
353 *
354 * MPSAFE
355 */
356 int
sys_extpwrite(struct sysmsg * sysmsg,const struct extpwrite_args * uap)357 sys_extpwrite(struct sysmsg *sysmsg, const struct extpwrite_args *uap)
358 {
359 struct thread *td = curthread;
360 struct uio auio;
361 struct iovec aiov;
362 int error;
363 int flags;
364
365 if ((ssize_t)uap->nbyte < 0)
366 error = EINVAL;
367
368 aiov.iov_base = (void *)(uintptr_t)uap->buf;
369 aiov.iov_len = uap->nbyte;
370 auio.uio_iov = &aiov;
371 auio.uio_iovcnt = 1;
372 auio.uio_offset = uap->offset;
373 auio.uio_resid = uap->nbyte;
374 auio.uio_rw = UIO_WRITE;
375 auio.uio_segflg = UIO_USERSPACE;
376 auio.uio_td = td;
377
378 flags = uap->flags & O_FMASK;
379 if (uap->offset != (off_t)-1)
380 flags |= O_FOFFSET;
381 error = kern_pwritev(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
382 return(error);
383 }
384
385 /*
386 * MPSAFE
387 */
388 int
sys_writev(struct sysmsg * sysmsg,const struct writev_args * uap)389 sys_writev(struct sysmsg *sysmsg, const struct writev_args *uap)
390 {
391 struct thread *td = curthread;
392 struct uio auio;
393 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
394 int error;
395
396 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
397 &auio.uio_resid);
398 if (error)
399 return (error);
400 auio.uio_iov = iov;
401 auio.uio_iovcnt = uap->iovcnt;
402 auio.uio_offset = -1;
403 auio.uio_rw = UIO_WRITE;
404 auio.uio_segflg = UIO_USERSPACE;
405 auio.uio_td = td;
406
407 error = kern_pwritev(uap->fd, &auio, 0, &sysmsg->sysmsg_szresult);
408
409 iovec_free(&iov, aiov);
410 return (error);
411 }
412
413
414 /*
415 * Gather positioned write system call
416 *
417 * MPSAFE
418 */
419 int
sys_extpwritev(struct sysmsg * sysmsg,const struct extpwritev_args * uap)420 sys_extpwritev(struct sysmsg *sysmsg, const struct extpwritev_args *uap)
421 {
422 struct thread *td = curthread;
423 struct uio auio;
424 struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
425 int error;
426 int flags;
427
428 error = iovec_copyin(uap->iovp, &iov, aiov, uap->iovcnt,
429 &auio.uio_resid);
430 if (error)
431 return (error);
432 auio.uio_iov = iov;
433 auio.uio_iovcnt = uap->iovcnt;
434 auio.uio_offset = uap->offset;
435 auio.uio_rw = UIO_WRITE;
436 auio.uio_segflg = UIO_USERSPACE;
437 auio.uio_td = td;
438
439 flags = uap->flags & O_FMASK;
440 if (uap->offset != (off_t)-1)
441 flags |= O_FOFFSET;
442
443 error = kern_pwritev(uap->fd, &auio, flags, &sysmsg->sysmsg_szresult);
444
445 iovec_free(&iov, aiov);
446 return(error);
447 }
448
449 /*
450 * MPSAFE
451 */
452 int
kern_pwritev(int fd,struct uio * auio,int flags,size_t * res)453 kern_pwritev(int fd, struct uio *auio, int flags, size_t *res)
454 {
455 struct thread *td = curthread;
456 struct file *fp;
457 int error;
458
459 fp = holdfp(td, fd, FWRITE);
460 if (fp == NULL)
461 return (EBADF);
462 else if ((flags & O_FOFFSET) && fp->f_type != DTYPE_VNODE) {
463 error = ESPIPE;
464 } else {
465 error = dofilewrite(fd, fp, auio, flags, res);
466 }
467 dropfp(td, fd, fp);
468
469 return(error);
470 }
471
472 /*
473 * Common code for writev and pwritev that writes data to
474 * a file using the passed in uio, offset, and flags.
475 *
476 * MPALMOSTSAFE - ktrace needs help
477 */
478 static int
dofilewrite(int fd,struct file * fp,struct uio * auio,int flags,size_t * res)479 dofilewrite(int fd, struct file *fp, struct uio *auio, int flags, size_t *res)
480 {
481 struct thread *td = curthread;
482 struct lwp *lp = td->td_lwp;
483 int error;
484 size_t len;
485 #ifdef KTRACE
486 struct iovec *ktriov = NULL;
487 struct uio ktruio;
488 #endif
489
490 #ifdef KTRACE
491 /*
492 * if tracing, save a copy of iovec and uio
493 */
494 if (KTRPOINT(td, KTR_GENIO)) {
495 int iovlen = auio->uio_iovcnt * sizeof(struct iovec);
496
497 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK);
498 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
499 ktruio = *auio;
500 }
501 #endif
502 len = auio->uio_resid;
503 error = fo_write(fp, auio, fp->f_cred, flags);
504 if (error) {
505 if (auio->uio_resid != len && (error == ERESTART ||
506 error == EINTR || error == EWOULDBLOCK))
507 error = 0;
508 /* Socket layer is responsible for issuing SIGPIPE. */
509 if (error == EPIPE && fp->f_type != DTYPE_SOCKET)
510 lwpsignal(lp->lwp_proc, lp, SIGPIPE);
511 }
512 #ifdef KTRACE
513 if (ktriov != NULL) {
514 if (error == 0) {
515 ktruio.uio_iov = ktriov;
516 ktruio.uio_resid = len - auio->uio_resid;
517 ktrgenio(lp, fd, UIO_WRITE, &ktruio, error);
518 }
519 kfree(ktriov, M_TEMP);
520 }
521 #endif
522 if (error == 0)
523 *res = len - auio->uio_resid;
524
525 return(error);
526 }
527
528 /*
529 * Ioctl system call
530 *
531 * MPSAFE
532 */
533 int
sys_ioctl(struct sysmsg * sysmsg,const struct ioctl_args * uap)534 sys_ioctl(struct sysmsg *sysmsg, const struct ioctl_args *uap)
535 {
536 int error;
537
538 error = mapped_ioctl(uap->fd, uap->com, uap->data, NULL, sysmsg);
539 return (error);
540 }
541
542 struct ioctl_map_entry {
543 const char *subsys;
544 struct ioctl_map_range *cmd_ranges;
545 LIST_ENTRY(ioctl_map_entry) entries;
546 };
547
548 /*
549 * The true heart of all ioctl syscall handlers (native, emulation).
550 * If map != NULL, it will be searched for a matching entry for com,
551 * and appropriate conversions/conversion functions will be utilized.
552 *
553 * MPSAFE
554 */
555 int
mapped_ioctl(int fd,u_long com,caddr_t uspc_data,struct ioctl_map * map,struct sysmsg * msg)556 mapped_ioctl(int fd, u_long com, caddr_t uspc_data, struct ioctl_map *map,
557 struct sysmsg *msg)
558 {
559 struct thread *td = curthread;
560 struct proc *p = td->td_proc;
561 struct ucred *cred;
562 struct file *fp;
563 struct ioctl_map_range *iomc = NULL;
564 int error;
565 u_int size;
566 u_long ocom = com;
567 caddr_t data, memp;
568 int tmp;
569 #define STK_PARAMS 128
570 union {
571 char stkbuf[STK_PARAMS];
572 long align;
573 } ubuf;
574
575 KKASSERT(p);
576 cred = td->td_ucred;
577 memp = NULL;
578
579 fp = holdfp(td, fd, FREAD|FWRITE);
580 if (fp == NULL)
581 return(EBADF);
582
583 if (map != NULL) { /* obey translation map */
584 u_long maskcmd;
585 struct ioctl_map_entry *e;
586
587 maskcmd = com & map->mask;
588
589 lwkt_gettoken(&mioctl_token);
590 LIST_FOREACH(e, &map->mapping, entries) {
591 for (iomc = e->cmd_ranges; iomc->start != 0 ||
592 iomc->maptocmd != 0 || iomc->wrapfunc != NULL ||
593 iomc->mapfunc != NULL;
594 iomc++) {
595 if (maskcmd >= iomc->start &&
596 maskcmd <= iomc->end)
597 break;
598 }
599
600 /* Did we find a match? */
601 if (iomc->start != 0 || iomc->maptocmd != 0 ||
602 iomc->wrapfunc != NULL || iomc->mapfunc != NULL)
603 break;
604 }
605 lwkt_reltoken(&mioctl_token);
606
607 if (iomc == NULL ||
608 (iomc->start == 0 && iomc->maptocmd == 0
609 && iomc->wrapfunc == NULL && iomc->mapfunc == NULL)) {
610 kprintf("%s: 'ioctl' fd=%d, cmd=0x%lx ('%c',%d) not implemented\n",
611 map->sys, fd, maskcmd,
612 (int)((maskcmd >> 8) & 0xff),
613 (int)(maskcmd & 0xff));
614 error = EINVAL;
615 goto done;
616 }
617
618 /*
619 * If it's a non-range one to one mapping, maptocmd should be
620 * correct. If it's a ranged one to one mapping, we pass the
621 * original value of com, and for a range mapped to a different
622 * range, we always need a mapping function to translate the
623 * ioctl to our native ioctl. Ex. 6500-65ff <-> 9500-95ff
624 */
625 if (iomc->start == iomc->end && iomc->maptocmd == iomc->maptoend) {
626 com = iomc->maptocmd;
627 } else if (iomc->start == iomc->maptocmd && iomc->end == iomc->maptoend) {
628 if (iomc->mapfunc != NULL)
629 com = iomc->mapfunc(iomc->start, iomc->end,
630 iomc->start, iomc->end,
631 com, com);
632 } else {
633 if (iomc->mapfunc != NULL) {
634 com = iomc->mapfunc(iomc->start, iomc->end,
635 iomc->maptocmd, iomc->maptoend,
636 com, ocom);
637 } else {
638 kprintf("%s: Invalid mapping for fd=%d, cmd=%#lx ('%c',%d)\n",
639 map->sys, fd, maskcmd,
640 (int)((maskcmd >> 8) & 0xff),
641 (int)(maskcmd & 0xff));
642 error = EINVAL;
643 goto done;
644 }
645 }
646 }
647
648 switch (com) {
649 case FIONCLEX:
650 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
651 goto done;
652 case FIOCLEX:
653 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
654 goto done;
655 }
656
657 /*
658 * Interpret high order word to find amount of data to be
659 * copied to/from the user's address space.
660 */
661 size = IOCPARM_LEN(com);
662 if (size > IOCPARM_MAX) {
663 error = ENOTTY;
664 goto done;
665 }
666
667 if ((com & IOC_VOID) == 0 && size > sizeof(ubuf.stkbuf)) {
668 memp = kmalloc(size, M_IOCTLOPS, M_WAITOK);
669 data = memp;
670 } else {
671 memp = NULL;
672 data = ubuf.stkbuf;
673 }
674 if (com & IOC_VOID) {
675 *(caddr_t *)data = uspc_data;
676 } else if (com & IOC_IN) {
677 if (size != 0) {
678 error = copyin(uspc_data, data, (size_t)size);
679 if (error)
680 goto done;
681 } else {
682 *(caddr_t *)data = uspc_data;
683 }
684 } else if ((com & IOC_OUT) != 0 && size) {
685 /*
686 * Zero the buffer so the user always
687 * gets back something deterministic.
688 */
689 bzero(data, (size_t)size);
690 }
691
692 switch (com) {
693 case FIONBIO:
694 if ((tmp = *(int *)data))
695 atomic_set_int(&fp->f_flag, FNONBLOCK);
696 else
697 atomic_clear_int(&fp->f_flag, FNONBLOCK);
698 error = 0;
699 break;
700
701 case FIOASYNC:
702 if ((tmp = *(int *)data))
703 atomic_set_int(&fp->f_flag, FASYNC);
704 else
705 atomic_clear_int(&fp->f_flag, FASYNC);
706 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred, msg);
707 break;
708
709 default:
710 /*
711 * If there is a override function,
712 * call it instead of directly routing the call
713 */
714 if (map != NULL && iomc->wrapfunc != NULL)
715 error = iomc->wrapfunc(fp, com, ocom, data, cred);
716 else
717 error = fo_ioctl(fp, com, data, cred, msg);
718 /*
719 * Copy any data to user, size was
720 * already set and checked above.
721 */
722 if (error == 0 && (com & IOC_OUT) != 0 && size != 0)
723 error = copyout(data, uspc_data, (size_t)size);
724 break;
725 }
726 done:
727 if (memp != NULL)
728 kfree(memp, M_IOCTLOPS);
729 dropfp(td, fd, fp);
730
731 return(error);
732 }
733
734 /*
735 * MPSAFE
736 */
737 int
mapped_ioctl_register_handler(struct ioctl_map_handler * he)738 mapped_ioctl_register_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 he->subsys != NULL && *he->subsys != '\0');
744
745 ne = kmalloc(sizeof(struct ioctl_map_entry), M_IOCTLMAP,
746 M_WAITOK | M_ZERO);
747
748 ne->subsys = he->subsys;
749 ne->cmd_ranges = he->cmd_ranges;
750
751 lwkt_gettoken(&mioctl_token);
752 LIST_INSERT_HEAD(&he->map->mapping, ne, entries);
753 lwkt_reltoken(&mioctl_token);
754
755 return(0);
756 }
757
758 /*
759 * MPSAFE
760 */
761 int
mapped_ioctl_unregister_handler(struct ioctl_map_handler * he)762 mapped_ioctl_unregister_handler(struct ioctl_map_handler *he)
763 {
764 struct ioctl_map_entry *ne;
765 int error = EINVAL;
766
767 KKASSERT(he != NULL && he->map != NULL && he->cmd_ranges != NULL);
768
769 lwkt_gettoken(&mioctl_token);
770 LIST_FOREACH(ne, &he->map->mapping, entries) {
771 if (ne->cmd_ranges == he->cmd_ranges) {
772 LIST_REMOVE(ne, entries);
773 kfree(ne, M_IOCTLMAP);
774 error = 0;
775 break;
776 }
777 }
778 lwkt_reltoken(&mioctl_token);
779 return(error);
780 }
781
782 static int nseldebug;
783 SYSCTL_INT(_kern, OID_AUTO, nseldebug, CTLFLAG_RW, &nseldebug, 0, "");
784
785 /*
786 * Select system call.
787 *
788 * MPSAFE
789 */
790 int
sys_select(struct sysmsg * sysmsg,const struct select_args * uap)791 sys_select(struct sysmsg *sysmsg, const struct select_args *uap)
792 {
793 struct timeval ktv;
794 struct timespec *ktsp, kts;
795 int error;
796
797 /*
798 * Get timeout if any.
799 */
800 if (uap->tv != NULL) {
801 error = copyin(uap->tv, &ktv, sizeof (ktv));
802 if (error)
803 return (error);
804 TIMEVAL_TO_TIMESPEC(&ktv, &kts);
805 ktsp = &kts;
806 } else {
807 ktsp = NULL;
808 }
809
810 /*
811 * Do real work.
812 */
813 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
814 &sysmsg->sysmsg_result);
815
816 return (error);
817 }
818
819
820 /*
821 * Pselect system call.
822 */
823 int
sys_pselect(struct sysmsg * sysmsg,const struct pselect_args * uap)824 sys_pselect(struct sysmsg *sysmsg, const struct pselect_args *uap)
825 {
826 struct thread *td = curthread;
827 struct lwp *lp = td->td_lwp;
828 struct timespec *ktsp, kts;
829 sigset_t sigmask;
830 int error;
831
832 /*
833 * Get timeout if any.
834 */
835 if (uap->ts != NULL) {
836 error = copyin(uap->ts, &kts, sizeof (kts));
837 if (error)
838 return (error);
839 ktsp = &kts;
840 } else {
841 ktsp = NULL;
842 }
843
844 /*
845 * Install temporary signal mask if any provided.
846 */
847 if (uap->sigmask != NULL) {
848 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
849 if (error)
850 return (error);
851 lwkt_gettoken(&lp->lwp_proc->p_token);
852 lp->lwp_oldsigmask = lp->lwp_sigmask;
853 SIG_CANTMASK(sigmask);
854 lp->lwp_sigmask = sigmask;
855 lwkt_reltoken(&lp->lwp_proc->p_token);
856 }
857
858 /*
859 * Do real job.
860 */
861 error = doselect(uap->nd, uap->in, uap->ou, uap->ex, ktsp,
862 &sysmsg->sysmsg_result);
863
864 if (uap->sigmask != NULL) {
865 lwkt_gettoken(&lp->lwp_proc->p_token);
866 /* doselect() responsible for turning ERESTART into EINTR */
867 KKASSERT(error != ERESTART);
868 if (error == EINTR) {
869 /*
870 * We can't restore the previous signal mask now
871 * because it could block the signal that interrupted
872 * us. So make a note to restore it after executing
873 * the handler.
874 */
875 lp->lwp_flags |= LWP_OLDMASK;
876 } else {
877 /*
878 * No handler to run. Restore previous mask immediately.
879 */
880 lp->lwp_sigmask = lp->lwp_oldsigmask;
881 }
882 lwkt_reltoken(&lp->lwp_proc->p_token);
883 }
884
885 return (error);
886 }
887
888 static int
select_copyin(void * arg,struct kevent * kevp,int maxevents,int * events)889 select_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
890 {
891 struct select_kevent_copyin_args *skap = NULL;
892 struct kevent *kev;
893 int fd;
894 kfd_set *fdp = NULL;
895 short filter = 0;
896 u_int fflags = 0;
897
898 skap = (struct select_kevent_copyin_args *)arg;
899
900 if (*events == maxevents)
901 return (0);
902
903 while (skap->active_set < COPYIN_DONE) {
904 switch (skap->active_set) {
905 case COPYIN_READ:
906 /*
907 * Register descriptors for the read filter
908 */
909 fdp = skap->read_set;
910 filter = EVFILT_READ;
911 fflags = NOTE_OLDAPI;
912 if (fdp)
913 break;
914 ++skap->active_set;
915 skap->proc_fds = 0;
916 /* fall through */
917 case COPYIN_WRITE:
918 /*
919 * Register descriptors for the write filter
920 */
921 fdp = skap->write_set;
922 filter = EVFILT_WRITE;
923 fflags = NOTE_OLDAPI;
924 if (fdp)
925 break;
926 ++skap->active_set;
927 skap->proc_fds = 0;
928 /* fall through */
929 case COPYIN_EXCEPT:
930 /*
931 * Register descriptors for the exception filter
932 */
933 fdp = skap->except_set;
934 filter = EVFILT_EXCEPT;
935 fflags = NOTE_OLDAPI | NOTE_OOB;
936 if (fdp)
937 break;
938 ++skap->active_set;
939 skap->proc_fds = 0;
940 /* fall through */
941 case COPYIN_DONE:
942 /*
943 * Nothing left to register
944 */
945 return(0);
946 /* NOT REACHED */
947 }
948
949 while (skap->proc_fds < skap->num_fds) {
950 fd = skap->proc_fds;
951 if (FD_ISSET(fd, fdp)) {
952 kev = &kevp[*events];
953 EV_SET(kev, fd, filter,
954 EV_ADD|EV_ENABLE,
955 fflags, 0,
956 (void *)(uintptr_t)
957 skap->lwp->lwp_kqueue_serial);
958 FD_CLR(fd, fdp);
959 ++*events;
960
961 if (nseldebug) {
962 kprintf("select fd %d filter %d "
963 "serial %ju\n", fd, filter,
964 (uintmax_t)
965 skap->lwp->lwp_kqueue_serial);
966 }
967 }
968 ++skap->proc_fds;
969 if (*events == maxevents)
970 return (0);
971 }
972 skap->active_set++;
973 skap->proc_fds = 0;
974 }
975
976 return (0);
977 }
978
979 static int
select_copyout(void * arg,struct kevent * kevp,int count,int * res)980 select_copyout(void *arg, struct kevent *kevp, int count, int *res)
981 {
982 struct select_kevent_copyin_args *skap;
983 struct kevent kev;
984 int i;
985 int n;
986
987 skap = (struct select_kevent_copyin_args *)arg;
988
989 for (i = 0; i < count; ++i) {
990 /*
991 * Filter out and delete spurious events
992 */
993 if ((uint64_t)(uintptr_t)kevp[i].udata !=
994 skap->lwp->lwp_kqueue_serial)
995 {
996 panic("select_copyout: unexpected udata");
997 deregister:
998 kev = kevp[i];
999 kev.flags = EV_DISABLE|EV_DELETE;
1000 n = 1;
1001 kqueue_register(&skap->lwp->lwp_kqueue, &kev, &n, 0);
1002 if (nseldebug) {
1003 kprintf("select fd %ju mismatched serial %ju\n",
1004 (uintmax_t)kevp[i].ident,
1005 (uintmax_t)skap->lwp->lwp_kqueue_serial);
1006 }
1007 continue;
1008 }
1009
1010 /*
1011 * Handle errors
1012 */
1013 if (kevp[i].flags & EV_ERROR) {
1014 int error = kevp[i].data;
1015
1016 switch (error) {
1017 case EBADF:
1018 /*
1019 * A bad file descriptor is considered a
1020 * fatal error for select, bail out.
1021 */
1022 skap->error = error;
1023 *res = -1;
1024 return error;
1025
1026 default:
1027 /*
1028 * Select silently swallows any unknown errors
1029 * for descriptors in the read or write sets.
1030 *
1031 * ALWAYS filter out EOPNOTSUPP errors from
1032 * filters (at least until all filters support
1033 * EVFILT_EXCEPT)
1034 *
1035 * We also filter out ENODEV since dev_dkqfilter
1036 * returns ENODEV if EOPNOTSUPP is returned in an
1037 * inner call.
1038 *
1039 * XXX: fix this
1040 */
1041 if (kevp[i].filter != EVFILT_READ &&
1042 kevp[i].filter != EVFILT_WRITE &&
1043 error != EOPNOTSUPP &&
1044 error != ENODEV) {
1045 skap->error = error;
1046 *res = -1;
1047 return error;
1048 }
1049 break;
1050 }
1051
1052 /*
1053 * We must deregister any unsupported select events
1054 * to avoid a live-lock.
1055 */
1056 if (nseldebug) {
1057 kprintf("select fd %ju filter %d error %d\n",
1058 (uintmax_t)kevp[i].ident,
1059 kevp[i].filter, error);
1060 }
1061 goto deregister;
1062 }
1063
1064 switch (kevp[i].filter) {
1065 case EVFILT_READ:
1066 FD_SET(kevp[i].ident, skap->read_set);
1067 break;
1068 case EVFILT_WRITE:
1069 FD_SET(kevp[i].ident, skap->write_set);
1070 break;
1071 case EVFILT_EXCEPT:
1072 FD_SET(kevp[i].ident, skap->except_set);
1073 break;
1074 }
1075
1076 ++*res;
1077 }
1078
1079 return (0);
1080 }
1081
1082 /*
1083 * Copy select bits in from userland. Allocate kernel memory if the
1084 * set is large.
1085 */
1086 static int
getbits(int bytes,fd_set * in_set,kfd_set ** out_set,kfd_set * tmp_set)1087 getbits(int bytes, fd_set *in_set, kfd_set **out_set, kfd_set *tmp_set)
1088 {
1089 int error;
1090
1091 if (in_set) {
1092 if (bytes < sizeof(*tmp_set))
1093 *out_set = tmp_set;
1094 else
1095 *out_set = kmalloc(bytes, M_SELECT, M_WAITOK);
1096 error = copyin(in_set, *out_set, bytes);
1097 } else {
1098 *out_set = NULL;
1099 error = 0;
1100 }
1101 return (error);
1102 }
1103
1104 /*
1105 * Copy returned select bits back out to userland.
1106 */
1107 static int
putbits(int bytes,kfd_set * in_set,fd_set * out_set)1108 putbits(int bytes, kfd_set *in_set, fd_set *out_set)
1109 {
1110 int error;
1111
1112 if (in_set) {
1113 error = copyout(in_set, out_set, bytes);
1114 } else {
1115 error = 0;
1116 }
1117 return (error);
1118 }
1119
1120 static int
dotimeout_only(struct timespec * ts)1121 dotimeout_only(struct timespec *ts)
1122 {
1123 return(nanosleep1(ts, NULL));
1124 }
1125
1126 /*
1127 * Common code for sys_select() and sys_pselect().
1128 *
1129 * in, out and ex are userland pointers. ts must point to validated
1130 * kernel-side timeout value or NULL for infinite timeout. res must
1131 * point to syscall return value.
1132 */
1133 static int
doselect(int nd,fd_set * read,fd_set * write,fd_set * except,struct timespec * ts,int * res)1134 doselect(int nd, fd_set *read, fd_set *write, fd_set *except,
1135 struct timespec *ts, int *res)
1136 {
1137 struct proc *p = curproc;
1138 struct select_kevent_copyin_args *kap, ka;
1139 int bytes, error;
1140 kfd_set read_tmp;
1141 kfd_set write_tmp;
1142 kfd_set except_tmp;
1143
1144 *res = 0;
1145 if (nd < 0)
1146 return (EINVAL);
1147 if (nd == 0 && ts)
1148 return (dotimeout_only(ts));
1149
1150 if (nd > p->p_fd->fd_nfiles) /* limit kmalloc */
1151 nd = p->p_fd->fd_nfiles;
1152
1153 kap = &ka;
1154 kap->lwp = curthread->td_lwp;
1155 kap->num_fds = nd;
1156 kap->proc_fds = 0;
1157 kap->error = 0;
1158 kap->active_set = COPYIN_READ;
1159
1160 /*
1161 * Calculate bytes based on the number of __fd_mask[] array entries
1162 * multiplied by the size of __fd_mask.
1163 */
1164 bytes = howmany(nd, __NFDBITS) * sizeof(__fd_mask);
1165
1166 /* kap->read_set = NULL; not needed */
1167 kap->write_set = NULL;
1168 kap->except_set = NULL;
1169
1170 error = getbits(bytes, read, &kap->read_set, &read_tmp);
1171 if (error == 0)
1172 error = getbits(bytes, write, &kap->write_set, &write_tmp);
1173 if (error == 0)
1174 error = getbits(bytes, except, &kap->except_set, &except_tmp);
1175 if (error)
1176 goto done;
1177
1178 /*
1179 * NOTE: Make sure the max events passed to kern_kevent() is
1180 * effectively unlimited. (nd * 3) accomplishes this.
1181 *
1182 * (*res) continues to increment as returned events are
1183 * loaded in.
1184 */
1185 error = kern_kevent(&kap->lwp->lwp_kqueue, 0x7FFFFFFF, res, kap,
1186 select_copyin, select_copyout, ts,
1187 KEVENT_AUTO_STALE);
1188 if (error == 0)
1189 error = putbits(bytes, kap->read_set, read);
1190 if (error == 0)
1191 error = putbits(bytes, kap->write_set, write);
1192 if (error == 0)
1193 error = putbits(bytes, kap->except_set, except);
1194
1195 /*
1196 * An error from an individual event that should be passed
1197 * back to userland (EBADF)
1198 */
1199 if (kap->error)
1200 error = kap->error;
1201
1202 /*
1203 * Clean up.
1204 */
1205 done:
1206 if (kap->read_set && kap->read_set != &read_tmp)
1207 kfree(kap->read_set, M_SELECT);
1208 if (kap->write_set && kap->write_set != &write_tmp)
1209 kfree(kap->write_set, M_SELECT);
1210 if (kap->except_set && kap->except_set != &except_tmp)
1211 kfree(kap->except_set, M_SELECT);
1212
1213 kap->lwp->lwp_kqueue_serial += kap->num_fds;
1214
1215 return (error);
1216 }
1217
1218 /*
1219 * Poll system call.
1220 *
1221 * MPSAFE
1222 */
1223 int
sys_poll(struct sysmsg * sysmsg,const struct poll_args * uap)1224 sys_poll(struct sysmsg *sysmsg, const struct poll_args *uap)
1225 {
1226 struct timespec ts, *tsp;
1227 int error;
1228
1229 if (uap->timeout != INFTIM) {
1230 if (uap->timeout < 0)
1231 return (EINVAL);
1232 ts.tv_sec = uap->timeout / 1000;
1233 ts.tv_nsec = (uap->timeout % 1000) * 1000 * 1000;
1234 tsp = &ts;
1235 } else {
1236 tsp = NULL;
1237 }
1238
1239 error = dopoll(uap->nfds, uap->fds, tsp, &sysmsg->sysmsg_result, 0);
1240
1241 return (error);
1242 }
1243
1244 /*
1245 * Ppoll system call.
1246 *
1247 * MPSAFE
1248 */
1249 int
sys_ppoll(struct sysmsg * sysmsg,const struct ppoll_args * uap)1250 sys_ppoll(struct sysmsg *sysmsg, const struct ppoll_args *uap)
1251 {
1252 struct thread *td = curthread;
1253 struct lwp *lp = td->td_lwp;
1254 struct timespec *ktsp, kts;
1255 sigset_t sigmask;
1256 int error;
1257
1258 /*
1259 * Get timeout if any.
1260 */
1261 if (uap->ts != NULL) {
1262 error = copyin(uap->ts, &kts, sizeof (kts));
1263 if (error)
1264 return (error);
1265 ktsp = &kts;
1266 } else {
1267 ktsp = NULL;
1268 }
1269
1270 /*
1271 * Install temporary signal mask if any provided.
1272 */
1273 if (uap->sigmask != NULL) {
1274 error = copyin(uap->sigmask, &sigmask, sizeof(sigmask));
1275 if (error)
1276 return (error);
1277 lwkt_gettoken(&lp->lwp_proc->p_token);
1278 lp->lwp_oldsigmask = lp->lwp_sigmask;
1279 SIG_CANTMASK(sigmask);
1280 lp->lwp_sigmask = sigmask;
1281 lwkt_reltoken(&lp->lwp_proc->p_token);
1282 }
1283
1284 error = dopoll(uap->nfds, uap->fds, ktsp, &sysmsg->sysmsg_result,
1285 ktsp != NULL ? KEVENT_TIMEOUT_PRECISE : 0);
1286
1287 if (uap->sigmask != NULL) {
1288 lwkt_gettoken(&lp->lwp_proc->p_token);
1289 /* dopoll() responsible for turning ERESTART into EINTR */
1290 KKASSERT(error != ERESTART);
1291 if (error == EINTR) {
1292 /*
1293 * We can't restore the previous signal mask now
1294 * because it could block the signal that interrupted
1295 * us. So make a note to restore it after executing
1296 * the handler.
1297 */
1298 lp->lwp_flags |= LWP_OLDMASK;
1299 } else {
1300 /*
1301 * No handler to run. Restore previous mask immediately.
1302 */
1303 lp->lwp_sigmask = lp->lwp_oldsigmask;
1304 }
1305 lwkt_reltoken(&lp->lwp_proc->p_token);
1306 }
1307
1308 return (error);
1309 }
1310
1311 static int
poll_copyin(void * arg,struct kevent * kevp,int maxevents,int * events)1312 poll_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1313 {
1314 struct poll_kevent_copyin_args *pkap;
1315 struct pollfd *pfd;
1316 struct kevent *kev;
1317 int kev_count;
1318
1319 pkap = (struct poll_kevent_copyin_args *)arg;
1320
1321 while (pkap->pfds < pkap->nfds) {
1322 pfd = &pkap->fds[pkap->pfds];
1323
1324 /* Clear return events */
1325 pfd->revents = 0;
1326
1327 /* Do not check if fd is equal to -1 */
1328 if (pfd->fd == -1) {
1329 ++pkap->pfds;
1330 continue;
1331 }
1332
1333 /*
1334 * NOTE: pfd->events == 0 implies POLLHUP in BSDs. Used
1335 * by at least sshd and X11 udev support.
1336 */
1337 kev_count = 0;
1338 if (pfd->events == 0)
1339 kev_count++;
1340 if (pfd->events & (POLLIN | POLLHUP | POLLRDNORM))
1341 kev_count++;
1342 if (pfd->events & (POLLOUT | POLLWRNORM))
1343 kev_count++;
1344 if (pfd->events & (POLLPRI | POLLRDBAND))
1345 kev_count++;
1346
1347 if (*events + kev_count > maxevents)
1348 return (0);
1349
1350 /*
1351 * NOTE: A combined serial number and poll array index is
1352 * stored in kev->udata.
1353 *
1354 * NOTE: Events will be registered with KEVENT_UNIQUE_NOTES
1355 * set, using kev->data for the uniqifier. kev->data
1356 * is an implied in the actual registration.
1357 */
1358 kev = &kevp[*events];
1359
1360 /*
1361 * Implied POLLHUP
1362 */
1363 if (pfd->events == 0) {
1364 int notes = NOTE_OLDAPI | NOTE_HUPONLY;
1365
1366 EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1367 notes, pkap->pfds, (void *)(uintptr_t)
1368 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1369 }
1370
1371 /*
1372 * Nominal read events
1373 */
1374 if (pfd->events & (POLLIN | POLLHUP | POLLRDNORM)) {
1375 int notes = NOTE_OLDAPI;
1376 if ((pfd->events & (POLLIN | POLLRDNORM)) == 0)
1377 notes |= NOTE_HUPONLY;
1378
1379 EV_SET(kev++, pfd->fd, EVFILT_READ, EV_ADD|EV_ENABLE,
1380 notes, pkap->pfds, (void *)(uintptr_t)
1381 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1382 }
1383
1384 /*
1385 * Nominal write events
1386 */
1387 if (pfd->events & (POLLOUT | POLLWRNORM)) {
1388 EV_SET(kev++, pfd->fd, EVFILT_WRITE, EV_ADD|EV_ENABLE,
1389 NOTE_OLDAPI, pkap->pfds, (void *)(uintptr_t)
1390 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1391 }
1392
1393 /*
1394 * Nominal exceptional events
1395 */
1396 if (pfd->events & (POLLPRI | POLLRDBAND)) {
1397 EV_SET(kev++, pfd->fd, EVFILT_EXCEPT, EV_ADD|EV_ENABLE,
1398 NOTE_OLDAPI | NOTE_OOB, pkap->pfds,
1399 (void *)(uintptr_t)
1400 (pkap->lwp->lwp_kqueue_serial + pkap->pfds));
1401 }
1402
1403 if (nseldebug) {
1404 kprintf("poll index %d/%d fd %d events %08x "
1405 "serial %ju\n", pkap->pfds, pkap->nfds-1,
1406 pfd->fd, pfd->events,
1407 (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1408 }
1409
1410 ++pkap->pfds;
1411 (*events) += kev_count;
1412 }
1413
1414 return (0);
1415 }
1416
1417 static int
poll_copyout(void * arg,struct kevent * kevp,int count,int * res)1418 poll_copyout(void *arg, struct kevent *kevp, int count, int *res)
1419 {
1420 struct poll_kevent_copyin_args *pkap;
1421 struct pollfd *pfd;
1422 struct kevent kev;
1423 int count_res;
1424 int i;
1425 int n;
1426 uint64_t pi;
1427
1428 pkap = (struct poll_kevent_copyin_args *)arg;
1429
1430 for (i = 0; i < count; ++i) {
1431 /*
1432 * Extract the poll array index and delete spurious events.
1433 * We can easily tell if the serial number is incorrect
1434 * by checking whether the extracted index is out of range.
1435 */
1436 pi = (uint64_t)(uintptr_t)kevp[i].udata -
1437 pkap->lwp->lwp_kqueue_serial;
1438 if (pi >= pkap->nfds) {
1439 panic("poll_copyout: unexpected udata");
1440 deregister:
1441 kev = kevp[i];
1442 kev.flags = EV_DISABLE|EV_DELETE;
1443 kev.data = pi; /* uniquifier */
1444 n = 1;
1445 kqueue_register(&pkap->lwp->lwp_kqueue, &kev, &n,
1446 KEVENT_UNIQUE_NOTES);
1447 if (nseldebug) {
1448 kprintf("poll index %ju out of range against "
1449 "serial %ju\n", (uintmax_t)pi,
1450 (uintmax_t)pkap->lwp->lwp_kqueue_serial);
1451 }
1452 continue;
1453 }
1454
1455 /*
1456 * Locate the pollfd and process events
1457 */
1458 pfd = &pkap->fds[pi];
1459 if (kevp[i].ident == pfd->fd) {
1460 /*
1461 * A single descriptor may generate an error against
1462 * more than one filter, make sure to set the
1463 * appropriate flags but do not increment (*res)
1464 * more than once.
1465 */
1466 count_res = (pfd->revents == 0);
1467 if (kevp[i].flags & EV_ERROR) {
1468 switch(kevp[i].data) {
1469 case EBADF:
1470 case POLLNVAL:
1471 /* Bad file descriptor */
1472 if (count_res)
1473 ++*res;
1474 pfd->revents |= POLLNVAL;
1475 break;
1476 default:
1477 /*
1478 * Poll silently swallows any unknown
1479 * errors except in the case of POLLPRI
1480 * (OOB/urgent data).
1481 *
1482 * ALWAYS filter out EOPNOTSUPP errors
1483 * from filters, common applications
1484 * set POLLPRI|POLLRDBAND and most
1485 * filters do not support EVFILT_EXCEPT.
1486 *
1487 * We also filter out ENODEV since
1488 * dev_dkqfilter returns ENODEV if
1489 * EOPNOTSUPP is returned in an
1490 * inner call.
1491 *
1492 * XXX: fix this
1493 */
1494 if (kevp[i].filter != EVFILT_READ &&
1495 kevp[i].filter != EVFILT_WRITE &&
1496 kevp[i].data != EOPNOTSUPP &&
1497 kevp[i].data != ENODEV) {
1498 if (count_res)
1499 ++*res;
1500 pfd->revents |= POLLERR;
1501 }
1502 break;
1503 }
1504 if (pfd->revents == 0 && nseldebug) {
1505 kprintf("poll index EV_ERROR %ju fd %d "
1506 "filter %d error %jd\n",
1507 (uintmax_t)pi, pfd->fd,
1508 kevp[i].filter,
1509 (intmax_t)kevp[i].data);
1510 }
1511
1512 /*
1513 * Silently deregister any unhandled EV_ERROR
1514 * condition (usually EOPNOTSUPP).
1515 */
1516 if (pfd->revents == 0)
1517 goto deregister;
1518 continue;
1519 }
1520
1521 switch (kevp[i].filter) {
1522 case EVFILT_READ:
1523 /*
1524 * NODATA on the read side can indicate a
1525 * half-closed situation and not necessarily
1526 * a disconnect, so depend on the user
1527 * issuing a read() and getting 0 bytes back.
1528 *
1529 * If EV_HUP is set the peer completely
1530 * disconnected and we can set POLLHUP.
1531 * Linux can return POLLHUP even if read
1532 * data has not been drained, so we should
1533 * too.
1534 */
1535 /* if (kevp[i].flags & EV_NODATA) */ {
1536 if (kevp[i].flags & EV_HUP)
1537 pfd->revents |= POLLHUP;
1538 }
1539 if ((kevp[i].flags & EV_EOF) &&
1540 kevp[i].fflags != 0)
1541 pfd->revents |= POLLERR;
1542 if (pfd->events & POLLIN)
1543 pfd->revents |= POLLIN;
1544 if (pfd->events & POLLRDNORM)
1545 pfd->revents |= POLLRDNORM;
1546 break;
1547 case EVFILT_WRITE:
1548 /*
1549 * As per the OpenGroup POLLHUP is mutually
1550 * exclusive with the writability flags. I
1551 * consider this a bit broken but...
1552 *
1553 * In this case a disconnect is implied even
1554 * for a half-closed (write side) situation.
1555 */
1556 if (kevp[i].flags & EV_EOF) {
1557 pfd->revents |= POLLHUP;
1558 if (kevp[i].fflags != 0)
1559 pfd->revents |= POLLERR;
1560 } else {
1561 if (pfd->events & POLLOUT)
1562 pfd->revents |= POLLOUT;
1563 if (pfd->events & POLLWRNORM)
1564 pfd->revents |= POLLWRNORM;
1565 }
1566 break;
1567 case EVFILT_EXCEPT:
1568 /*
1569 * EV_NODATA should never be tagged for this
1570 * filter.
1571 */
1572 if (pfd->events & POLLPRI)
1573 pfd->revents |= POLLPRI;
1574 if (pfd->events & POLLRDBAND)
1575 pfd->revents |= POLLRDBAND;
1576 break;
1577 }
1578
1579 if (nseldebug) {
1580 kprintf("poll index %ju/%d fd %d "
1581 "revents %08x\n", (uintmax_t)pi, pkap->nfds,
1582 pfd->fd, pfd->revents);
1583 }
1584
1585 if (count_res && pfd->revents)
1586 ++*res;
1587 }
1588
1589 /*
1590 * We must deregister any kqueue poll event that does not
1591 * set poll return bits to prevent a live-lock.
1592 */
1593 if (pfd->revents == 0) {
1594 kprintf("poll index %ju no-action %ju/%d "
1595 "events=%08x kevpfilt=%d/%08x\n",
1596 (uintmax_t)pi, (uintmax_t)kevp[i].ident,
1597 pfd->fd, pfd->events,
1598 kevp[i].filter, kevp[i].flags);
1599 goto deregister;
1600 }
1601 }
1602
1603 return (0);
1604 }
1605
1606 static int
dopoll(int nfds,struct pollfd * fds,struct timespec * ts,int * res,int flags)1607 dopoll(int nfds, struct pollfd *fds, struct timespec *ts, int *res, int flags)
1608 {
1609 struct poll_kevent_copyin_args ka;
1610 struct pollfd sfds[64];
1611 int bytes;
1612 int error;
1613
1614 flags |= KEVENT_AUTO_STALE | KEVENT_UNIQUE_NOTES;
1615
1616 *res = 0;
1617 if (nfds < 0)
1618 return (EINVAL);
1619
1620 if (nfds == 0 && ts)
1621 return (dotimeout_only(ts));
1622
1623 /*
1624 * This is a bit arbitrary but we need to limit internal kmallocs.
1625 */
1626 if (nfds > maxfilesperproc * 2)
1627 nfds = maxfilesperproc * 2;
1628 bytes = sizeof(struct pollfd) * nfds;
1629
1630 ka.lwp = curthread->td_lwp;
1631 ka.nfds = nfds;
1632 ka.pfds = 0;
1633 ka.error = 0;
1634
1635 if (ka.nfds < 64)
1636 ka.fds = sfds;
1637 else
1638 ka.fds = kmalloc(bytes, M_SELECT, M_WAITOK);
1639
1640 error = copyin(fds, ka.fds, bytes);
1641
1642 if (error == 0)
1643 error = kern_kevent(&ka.lwp->lwp_kqueue, 0x7FFFFFFF, res, &ka,
1644 poll_copyin, poll_copyout, ts, flags);
1645
1646 if (error == 0)
1647 error = copyout(ka.fds, fds, bytes);
1648
1649 if (ka.fds != sfds)
1650 kfree(ka.fds, M_SELECT);
1651
1652 ka.lwp->lwp_kqueue_serial += nfds;
1653
1654 return (error);
1655 }
1656
1657 static int
socket_wait_copyin(void * arg,struct kevent * kevp,int maxevents,int * events)1658 socket_wait_copyin(void *arg, struct kevent *kevp, int maxevents, int *events)
1659 {
1660 return (0);
1661 }
1662
1663 static int
socket_wait_copyout(void * arg,struct kevent * kevp,int count,int * res)1664 socket_wait_copyout(void *arg, struct kevent *kevp, int count, int *res)
1665 {
1666 ++*res;
1667 return (0);
1668 }
1669
1670 extern struct fileops socketops;
1671
1672 /*
1673 * NOTE: Callers of socket_wait() must already have a reference on the
1674 * socket.
1675 */
1676 int
socket_wait(struct socket * so,struct timespec * ts,int * res)1677 socket_wait(struct socket *so, struct timespec *ts, int *res)
1678 {
1679 struct thread *td = curthread;
1680 struct file *fp;
1681 struct kqueue kq;
1682 struct kevent kev;
1683 int error, fd;
1684 int n;
1685
1686 if ((error = falloc(td->td_lwp, &fp, &fd)) != 0)
1687 return (error);
1688
1689 fp->f_type = DTYPE_SOCKET;
1690 fp->f_flag = FREAD | FWRITE;
1691 fp->f_ops = &socketops;
1692 fp->f_data = so;
1693 fsetfd(td->td_lwp->lwp_proc->p_fd, fp, fd);
1694 fsetfdflags(td->td_proc->p_fd, fd, UF_EXCLOSE);
1695
1696 bzero(&kq, sizeof(kq));
1697 kqueue_init(&kq, td->td_lwp->lwp_proc->p_fd);
1698 EV_SET(&kev, fd, EVFILT_READ, EV_ADD|EV_ENABLE, 0, 0, NULL);
1699 n = 1;
1700 if ((error = kqueue_register(&kq, &kev, &n, 0)) != 0) {
1701 fdrop(fp);
1702 return (error);
1703 }
1704
1705 error = kern_kevent(&kq, 1, res, NULL, socket_wait_copyin,
1706 socket_wait_copyout, ts, 0);
1707
1708 EV_SET(&kev, fd, EVFILT_READ, EV_DELETE|EV_DISABLE, 0, 0, NULL);
1709 n = 1;
1710 kqueue_register(&kq, &kev, &n, 0);
1711 fp->f_ops = &badfileops;
1712 fdrop(fp);
1713
1714 return (error);
1715 }
1716
1717 /*
1718 * OpenBSD poll system call.
1719 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1720 *
1721 * MPSAFE
1722 */
1723 int
sys_openbsd_poll(struct sysmsg * sysmsg,const struct openbsd_poll_args * uap)1724 sys_openbsd_poll(struct sysmsg *sysmsg, const struct openbsd_poll_args *uap)
1725 {
1726 return (sys_poll(sysmsg, (const struct poll_args *)uap));
1727 }
1728
1729 /*ARGSUSED*/
1730 int
seltrue(cdev_t dev,int events)1731 seltrue(cdev_t dev, int events)
1732 {
1733 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1734 }
1735