1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capsicum.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/lock.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/socketvar.h>
57 #include <sys/uio.h>
58 #include <sys/eventfd.h>
59 #include <sys/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/limits.h>
62 #include <sys/malloc.h>
63 #include <sys/poll.h>
64 #include <sys/resourcevar.h>
65 #include <sys/selinfo.h>
66 #include <sys/sleepqueue.h>
67 #include <sys/specialfd.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/vnode.h>
72 #include <sys/bio.h>
73 #include <sys/buf.h>
74 #include <sys/condvar.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78
79 #include <security/audit/audit.h>
80
81 /*
82 * The following macro defines how many bytes will be allocated from
83 * the stack instead of memory allocated when passing the IOCTL data
84 * structures from userspace and to the kernel. Some IOCTLs having
85 * small data structures are used very frequently and this small
86 * buffer on the stack gives a significant speedup improvement for
87 * those requests. The value of this define should be greater or equal
88 * to 64 bytes and should also be power of two. The data structure is
89 * currently hard-aligned to a 8-byte boundary on the stack. This
90 * should currently be sufficient for all supported platforms.
91 */
92 #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */
93 #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */
94
95 #ifdef __LP64__
96 static int iosize_max_clamp = 0;
97 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
98 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
99 static int devfs_iosize_max_clamp = 1;
100 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
101 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
102 #endif
103
104 /*
105 * Assert that the return value of read(2) and write(2) syscalls fits
106 * into a register. If not, an architecture will need to provide the
107 * usermode wrappers to reconstruct the result.
108 */
109 CTASSERT(sizeof(register_t) >= sizeof(size_t));
110
111 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
112 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
113 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
114
115 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
116 u_int);
117 static int pollscan(struct thread *, struct pollfd *, u_int);
118 static int pollrescan(struct thread *);
119 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
120 static int selrescan(struct thread *, fd_mask **, fd_mask **);
121 static void selfdalloc(struct thread *, void *);
122 static void selfdfree(struct seltd *, struct selfd *);
123 static int dofileread(struct thread *, int, struct file *, struct uio *,
124 off_t, int);
125 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
126 off_t, int);
127 static void doselwakeup(struct selinfo *, int);
128 static void seltdinit(struct thread *);
129 static int seltdwait(struct thread *, sbintime_t, sbintime_t);
130 static void seltdclear(struct thread *);
131
132 /*
133 * One seltd per-thread allocated on demand as needed.
134 *
135 * t - protected by st_mtx
136 * k - Only accessed by curthread or read-only
137 */
138 struct seltd {
139 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
140 struct selfd *st_free1; /* (k) free fd for read set. */
141 struct selfd *st_free2; /* (k) free fd for write set. */
142 struct mtx st_mtx; /* Protects struct seltd */
143 struct cv st_wait; /* (t) Wait channel. */
144 int st_flags; /* (t) SELTD_ flags. */
145 };
146
147 #define SELTD_PENDING 0x0001 /* We have pending events. */
148 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
149
150 /*
151 * One selfd allocated per-thread per-file-descriptor.
152 * f - protected by sf_mtx
153 */
154 struct selfd {
155 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
156 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
157 struct selinfo *sf_si; /* (f) selinfo when linked. */
158 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
159 struct seltd *sf_td; /* (k) owning seltd. */
160 void *sf_cookie; /* (k) fd or pollfd. */
161 };
162
163 MALLOC_DEFINE(M_SELFD, "selfd", "selfd");
164 static struct mtx_pool *mtxpool_select;
165
166 #ifdef __LP64__
167 size_t
168 devfs_iosize_max(void)
169 {
170
171 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
172 INT_MAX : SSIZE_MAX);
173 }
174
175 size_t
176 iosize_max(void)
177 {
178
179 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
180 INT_MAX : SSIZE_MAX);
181 }
182 #endif
183
184 #ifndef _SYS_SYSPROTO_H_
185 struct read_args {
186 int fd;
187 void *buf;
188 size_t nbyte;
189 };
190 #endif
191 int
192 sys_read(struct thread *td, struct read_args *uap)
193 {
194 struct uio auio;
195 struct iovec aiov;
196 int error;
197
198 if (uap->nbyte > IOSIZE_MAX)
199 return (EINVAL);
200 aiov.iov_base = uap->buf;
201 aiov.iov_len = uap->nbyte;
202 auio.uio_iov = &aiov;
203 auio.uio_iovcnt = 1;
204 auio.uio_resid = uap->nbyte;
205 auio.uio_segflg = UIO_USERSPACE;
206 error = kern_readv(td, uap->fd, &auio);
207 return (error);
208 }
209
210 /*
211 * Positioned read system call
212 */
213 #ifndef _SYS_SYSPROTO_H_
214 struct pread_args {
215 int fd;
216 void *buf;
217 size_t nbyte;
218 int pad;
219 off_t offset;
220 };
221 #endif
222 int
223 sys_pread(struct thread *td, struct pread_args *uap)
224 {
225
226 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
227 }
228
229 int
230 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
231 {
232 struct uio auio;
233 struct iovec aiov;
234 int error;
235
236 if (nbyte > IOSIZE_MAX)
237 return (EINVAL);
238 aiov.iov_base = buf;
239 aiov.iov_len = nbyte;
240 auio.uio_iov = &aiov;
241 auio.uio_iovcnt = 1;
242 auio.uio_resid = nbyte;
243 auio.uio_segflg = UIO_USERSPACE;
244 error = kern_preadv(td, fd, &auio, offset);
245 return (error);
246 }
247
248 #if defined(COMPAT_FREEBSD6)
249 int
250 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
251 {
252
253 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
254 }
255 #endif
256
257 /*
258 * Scatter read system call.
259 */
260 #ifndef _SYS_SYSPROTO_H_
261 struct readv_args {
262 int fd;
263 struct iovec *iovp;
264 u_int iovcnt;
265 };
266 #endif
267 int
268 sys_readv(struct thread *td, struct readv_args *uap)
269 {
270 struct uio *auio;
271 int error;
272
273 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
274 if (error)
275 return (error);
276 error = kern_readv(td, uap->fd, auio);
277 free(auio, M_IOV);
278 return (error);
279 }
280
281 int
282 kern_readv(struct thread *td, int fd, struct uio *auio)
283 {
284 struct file *fp;
285 int error;
286
287 error = fget_read(td, fd, &cap_read_rights, &fp);
288 if (error)
289 return (error);
290 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
291 fdrop(fp, td);
292 return (error);
293 }
294
295 /*
296 * Scatter positioned read system call.
297 */
298 #ifndef _SYS_SYSPROTO_H_
299 struct preadv_args {
300 int fd;
301 struct iovec *iovp;
302 u_int iovcnt;
303 off_t offset;
304 };
305 #endif
306 int
307 sys_preadv(struct thread *td, struct preadv_args *uap)
308 {
309 struct uio *auio;
310 int error;
311
312 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
313 if (error)
314 return (error);
315 error = kern_preadv(td, uap->fd, auio, uap->offset);
316 free(auio, M_IOV);
317 return (error);
318 }
319
320 int
321 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
322 {
323 struct file *fp;
324 int error;
325
326 error = fget_read(td, fd, &cap_pread_rights, &fp);
327 if (error)
328 return (error);
329 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
330 error = ESPIPE;
331 else if (offset < 0 &&
332 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
333 error = EINVAL;
334 else
335 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
336 fdrop(fp, td);
337 return (error);
338 }
339
340 /*
341 * Common code for readv and preadv that reads data in
342 * from a file using the passed in uio, offset, and flags.
343 */
344 static int
345 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
346 off_t offset, int flags)
347 {
348 ssize_t cnt;
349 int error;
350 #ifdef KTRACE
351 struct uio *ktruio = NULL;
352 #endif
353
354 AUDIT_ARG_FD(fd);
355
356 /* Finish zero length reads right here */
357 if (auio->uio_resid == 0) {
358 td->td_retval[0] = 0;
359 return (0);
360 }
361 auio->uio_rw = UIO_READ;
362 auio->uio_offset = offset;
363 auio->uio_td = td;
364 #ifdef KTRACE
365 if (KTRPOINT(td, KTR_GENIO))
366 ktruio = cloneuio(auio);
367 #endif
368 cnt = auio->uio_resid;
369 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
370 if (auio->uio_resid != cnt && (error == ERESTART ||
371 error == EINTR || error == EWOULDBLOCK))
372 error = 0;
373 }
374 cnt -= auio->uio_resid;
375 #ifdef KTRACE
376 if (ktruio != NULL) {
377 ktruio->uio_resid = cnt;
378 ktrgenio(fd, UIO_READ, ktruio, error);
379 }
380 #endif
381 td->td_retval[0] = cnt;
382 return (error);
383 }
384
385 #ifndef _SYS_SYSPROTO_H_
386 struct write_args {
387 int fd;
388 const void *buf;
389 size_t nbyte;
390 };
391 #endif
392 int
393 sys_write(struct thread *td, struct write_args *uap)
394 {
395 struct uio auio;
396 struct iovec aiov;
397 int error;
398
399 if (uap->nbyte > IOSIZE_MAX)
400 return (EINVAL);
401 aiov.iov_base = (void *)(uintptr_t)uap->buf;
402 aiov.iov_len = uap->nbyte;
403 auio.uio_iov = &aiov;
404 auio.uio_iovcnt = 1;
405 auio.uio_resid = uap->nbyte;
406 auio.uio_segflg = UIO_USERSPACE;
407 error = kern_writev(td, uap->fd, &auio);
408 return (error);
409 }
410
411 /*
412 * Positioned write system call.
413 */
414 #ifndef _SYS_SYSPROTO_H_
415 struct pwrite_args {
416 int fd;
417 const void *buf;
418 size_t nbyte;
419 int pad;
420 off_t offset;
421 };
422 #endif
423 int
424 sys_pwrite(struct thread *td, struct pwrite_args *uap)
425 {
426
427 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
428 }
429
430 int
431 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
432 off_t offset)
433 {
434 struct uio auio;
435 struct iovec aiov;
436 int error;
437
438 if (nbyte > IOSIZE_MAX)
439 return (EINVAL);
440 aiov.iov_base = (void *)(uintptr_t)buf;
441 aiov.iov_len = nbyte;
442 auio.uio_iov = &aiov;
443 auio.uio_iovcnt = 1;
444 auio.uio_resid = nbyte;
445 auio.uio_segflg = UIO_USERSPACE;
446 error = kern_pwritev(td, fd, &auio, offset);
447 return (error);
448 }
449
450 #if defined(COMPAT_FREEBSD6)
451 int
452 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
453 {
454
455 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
456 }
457 #endif
458
459 /*
460 * Gather write system call.
461 */
462 #ifndef _SYS_SYSPROTO_H_
463 struct writev_args {
464 int fd;
465 struct iovec *iovp;
466 u_int iovcnt;
467 };
468 #endif
469 int
470 sys_writev(struct thread *td, struct writev_args *uap)
471 {
472 struct uio *auio;
473 int error;
474
475 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
476 if (error)
477 return (error);
478 error = kern_writev(td, uap->fd, auio);
479 free(auio, M_IOV);
480 return (error);
481 }
482
483 int
484 kern_writev(struct thread *td, int fd, struct uio *auio)
485 {
486 struct file *fp;
487 int error;
488
489 error = fget_write(td, fd, &cap_write_rights, &fp);
490 if (error)
491 return (error);
492 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
493 fdrop(fp, td);
494 return (error);
495 }
496
497 /*
498 * Gather positioned write system call.
499 */
500 #ifndef _SYS_SYSPROTO_H_
501 struct pwritev_args {
502 int fd;
503 struct iovec *iovp;
504 u_int iovcnt;
505 off_t offset;
506 };
507 #endif
508 int
509 sys_pwritev(struct thread *td, struct pwritev_args *uap)
510 {
511 struct uio *auio;
512 int error;
513
514 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
515 if (error)
516 return (error);
517 error = kern_pwritev(td, uap->fd, auio, uap->offset);
518 free(auio, M_IOV);
519 return (error);
520 }
521
522 int
523 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
524 {
525 struct file *fp;
526 int error;
527
528 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
529 if (error)
530 return (error);
531 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
532 error = ESPIPE;
533 else if (offset < 0 &&
534 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
535 error = EINVAL;
536 else
537 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
538 fdrop(fp, td);
539 return (error);
540 }
541
542 /*
543 * Common code for writev and pwritev that writes data to
544 * a file using the passed in uio, offset, and flags.
545 */
546 static int
547 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
548 off_t offset, int flags)
549 {
550 ssize_t cnt;
551 int error;
552 #ifdef KTRACE
553 struct uio *ktruio = NULL;
554 #endif
555
556 AUDIT_ARG_FD(fd);
557 auio->uio_rw = UIO_WRITE;
558 auio->uio_td = td;
559 auio->uio_offset = offset;
560 #ifdef KTRACE
561 if (KTRPOINT(td, KTR_GENIO))
562 ktruio = cloneuio(auio);
563 #endif
564 cnt = auio->uio_resid;
565 error = fo_write(fp, auio, td->td_ucred, flags, td);
566 /*
567 * Socket layer is responsible for special error handling,
568 * see sousrsend().
569 */
570 if (error != 0 && fp->f_type != DTYPE_SOCKET) {
571 if (auio->uio_resid != cnt && (error == ERESTART ||
572 error == EINTR || error == EWOULDBLOCK))
573 error = 0;
574 if (error == EPIPE) {
575 PROC_LOCK(td->td_proc);
576 tdsignal(td, SIGPIPE);
577 PROC_UNLOCK(td->td_proc);
578 }
579 }
580 cnt -= auio->uio_resid;
581 #ifdef KTRACE
582 if (ktruio != NULL) {
583 ktruio->uio_resid = cnt;
584 ktrgenio(fd, UIO_WRITE, ktruio, error);
585 }
586 #endif
587 td->td_retval[0] = cnt;
588 return (error);
589 }
590
591 /*
592 * Truncate a file given a file descriptor.
593 *
594 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
595 * descriptor isn't writable.
596 */
597 int
598 kern_ftruncate(struct thread *td, int fd, off_t length)
599 {
600 struct file *fp;
601 int error;
602
603 AUDIT_ARG_FD(fd);
604 if (length < 0)
605 return (EINVAL);
606 error = fget(td, fd, &cap_ftruncate_rights, &fp);
607 if (error)
608 return (error);
609 AUDIT_ARG_FILE(td->td_proc, fp);
610 if (!(fp->f_flag & FWRITE)) {
611 fdrop(fp, td);
612 return (EINVAL);
613 }
614 error = fo_truncate(fp, length, td->td_ucred, td);
615 fdrop(fp, td);
616 return (error);
617 }
618
619 #ifndef _SYS_SYSPROTO_H_
620 struct ftruncate_args {
621 int fd;
622 int pad;
623 off_t length;
624 };
625 #endif
626 int
627 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
628 {
629
630 return (kern_ftruncate(td, uap->fd, uap->length));
631 }
632
633 #if defined(COMPAT_43)
634 #ifndef _SYS_SYSPROTO_H_
635 struct oftruncate_args {
636 int fd;
637 long length;
638 };
639 #endif
640 int
641 oftruncate(struct thread *td, struct oftruncate_args *uap)
642 {
643
644 return (kern_ftruncate(td, uap->fd, uap->length));
645 }
646 #endif /* COMPAT_43 */
647
648 #ifndef _SYS_SYSPROTO_H_
649 struct ioctl_args {
650 int fd;
651 u_long com;
652 caddr_t data;
653 };
654 #endif
655 /* ARGSUSED */
656 int
657 sys_ioctl(struct thread *td, struct ioctl_args *uap)
658 {
659 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
660 uint32_t com;
661 int arg, error;
662 u_int size;
663 caddr_t data;
664
665 #ifdef INVARIANTS
666 if (uap->com > 0xffffffff) {
667 printf(
668 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
669 td->td_proc->p_pid, td->td_name, uap->com);
670 }
671 #endif
672 com = (uint32_t)uap->com;
673
674 /*
675 * Interpret high order word to find amount of data to be
676 * copied to/from the user's address space.
677 */
678 size = IOCPARM_LEN(com);
679 if ((size > IOCPARM_MAX) ||
680 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
681 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
682 ((com & IOC_OUT) && size == 0) ||
683 #else
684 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
685 #endif
686 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
687 return (ENOTTY);
688
689 if (size > 0) {
690 if (com & IOC_VOID) {
691 /* Integer argument. */
692 arg = (intptr_t)uap->data;
693 data = (void *)&arg;
694 size = 0;
695 } else {
696 if (size > SYS_IOCTL_SMALL_SIZE)
697 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
698 else
699 data = smalldata;
700 }
701 } else
702 data = (void *)&uap->data;
703 if (com & IOC_IN) {
704 error = copyin(uap->data, data, (u_int)size);
705 if (error != 0)
706 goto out;
707 } else if (com & IOC_OUT) {
708 /*
709 * Zero the buffer so the user always
710 * gets back something deterministic.
711 */
712 bzero(data, size);
713 }
714
715 error = kern_ioctl(td, uap->fd, com, data);
716
717 if (error == 0 && (com & IOC_OUT))
718 error = copyout(data, uap->data, (u_int)size);
719
720 out:
721 if (size > SYS_IOCTL_SMALL_SIZE)
722 free(data, M_IOCTLOPS);
723 return (error);
724 }
725
726 int
727 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
728 {
729 struct file *fp;
730 struct filedesc *fdp;
731 int error, tmp, locked;
732
733 AUDIT_ARG_FD(fd);
734 AUDIT_ARG_CMD(com);
735
736 fdp = td->td_proc->p_fd;
737
738 switch (com) {
739 case FIONCLEX:
740 case FIOCLEX:
741 FILEDESC_XLOCK(fdp);
742 locked = LA_XLOCKED;
743 break;
744 default:
745 #ifdef CAPABILITIES
746 FILEDESC_SLOCK(fdp);
747 locked = LA_SLOCKED;
748 #else
749 locked = LA_UNLOCKED;
750 #endif
751 break;
752 }
753
754 #ifdef CAPABILITIES
755 if ((fp = fget_noref(fdp, fd)) == NULL) {
756 error = EBADF;
757 goto out;
758 }
759 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
760 fp = NULL; /* fhold() was not called yet */
761 goto out;
762 }
763 if (!fhold(fp)) {
764 error = EBADF;
765 fp = NULL;
766 goto out;
767 }
768 if (locked == LA_SLOCKED) {
769 FILEDESC_SUNLOCK(fdp);
770 locked = LA_UNLOCKED;
771 }
772 #else
773 error = fget(td, fd, &cap_ioctl_rights, &fp);
774 if (error != 0) {
775 fp = NULL;
776 goto out;
777 }
778 #endif
779 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
780 error = EBADF;
781 goto out;
782 }
783
784 switch (com) {
785 case FIONCLEX:
786 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
787 goto out;
788 case FIOCLEX:
789 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
790 goto out;
791 case FIONBIO:
792 if ((tmp = *(int *)data))
793 atomic_set_int(&fp->f_flag, FNONBLOCK);
794 else
795 atomic_clear_int(&fp->f_flag, FNONBLOCK);
796 data = (void *)&tmp;
797 break;
798 case FIOASYNC:
799 if ((tmp = *(int *)data))
800 atomic_set_int(&fp->f_flag, FASYNC);
801 else
802 atomic_clear_int(&fp->f_flag, FASYNC);
803 data = (void *)&tmp;
804 break;
805 }
806
807 error = fo_ioctl(fp, com, data, td->td_ucred, td);
808 out:
809 switch (locked) {
810 case LA_XLOCKED:
811 FILEDESC_XUNLOCK(fdp);
812 break;
813 #ifdef CAPABILITIES
814 case LA_SLOCKED:
815 FILEDESC_SUNLOCK(fdp);
816 break;
817 #endif
818 default:
819 FILEDESC_UNLOCK_ASSERT(fdp);
820 break;
821 }
822 if (fp != NULL)
823 fdrop(fp, td);
824 return (error);
825 }
826
827 int
828 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
829 {
830 int error;
831
832 error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
833 return (kern_posix_error(td, error));
834 }
835
836 int
837 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
838 {
839 struct file *fp;
840 int error;
841
842 AUDIT_ARG_FD(fd);
843 if (offset < 0 || len <= 0)
844 return (EINVAL);
845 /* Check for wrap. */
846 if (offset > OFF_MAX - len)
847 return (EFBIG);
848 AUDIT_ARG_FD(fd);
849 error = fget(td, fd, &cap_pwrite_rights, &fp);
850 if (error != 0)
851 return (error);
852 AUDIT_ARG_FILE(td->td_proc, fp);
853 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
854 error = ESPIPE;
855 goto out;
856 }
857 if ((fp->f_flag & FWRITE) == 0) {
858 error = EBADF;
859 goto out;
860 }
861
862 error = fo_fallocate(fp, offset, len, td);
863 out:
864 fdrop(fp, td);
865 return (error);
866 }
867
868 int
869 sys_fspacectl(struct thread *td, struct fspacectl_args *uap)
870 {
871 struct spacectl_range rqsr, rmsr;
872 int error, cerror;
873
874 error = copyin(uap->rqsr, &rqsr, sizeof(rqsr));
875 if (error != 0)
876 return (error);
877
878 error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags,
879 &rmsr);
880 if (uap->rmsr != NULL) {
881 cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr));
882 if (error == 0)
883 error = cerror;
884 }
885 return (error);
886 }
887
888 int
889 kern_fspacectl(struct thread *td, int fd, int cmd,
890 const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp)
891 {
892 struct file *fp;
893 struct spacectl_range rmsr;
894 int error;
895
896 AUDIT_ARG_FD(fd);
897 AUDIT_ARG_CMD(cmd);
898 AUDIT_ARG_FFLAGS(flags);
899
900 if (rqsr == NULL)
901 return (EINVAL);
902 rmsr = *rqsr;
903 if (rmsrp != NULL)
904 *rmsrp = rmsr;
905
906 if (cmd != SPACECTL_DEALLOC ||
907 rqsr->r_offset < 0 || rqsr->r_len <= 0 ||
908 rqsr->r_offset > OFF_MAX - rqsr->r_len ||
909 (flags & ~SPACECTL_F_SUPPORTED) != 0)
910 return (EINVAL);
911
912 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
913 if (error != 0)
914 return (error);
915 AUDIT_ARG_FILE(td->td_proc, fp);
916 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
917 error = ESPIPE;
918 goto out;
919 }
920 if ((fp->f_flag & FWRITE) == 0) {
921 error = EBADF;
922 goto out;
923 }
924
925 error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags,
926 td->td_ucred, td);
927 /* fspacectl is not restarted after signals if the file is modified. */
928 if (rmsr.r_len != rqsr->r_len && (error == ERESTART ||
929 error == EINTR || error == EWOULDBLOCK))
930 error = 0;
931 if (rmsrp != NULL)
932 *rmsrp = rmsr;
933 out:
934 fdrop(fp, td);
935 return (error);
936 }
937
938 int
939 kern_specialfd(struct thread *td, int type, void *arg)
940 {
941 struct file *fp;
942 struct specialfd_eventfd *ae;
943 int error, fd, fflags;
944
945 fflags = 0;
946 error = falloc_noinstall(td, &fp);
947 if (error != 0)
948 return (error);
949
950 switch (type) {
951 case SPECIALFD_EVENTFD:
952 ae = arg;
953 if ((ae->flags & EFD_CLOEXEC) != 0)
954 fflags |= O_CLOEXEC;
955 error = eventfd_create_file(td, fp, ae->initval, ae->flags);
956 break;
957 default:
958 error = EINVAL;
959 break;
960 }
961
962 if (error == 0)
963 error = finstall(td, fp, &fd, fflags, NULL);
964 fdrop(fp, td);
965 if (error == 0)
966 td->td_retval[0] = fd;
967 return (error);
968 }
969
970 int
971 sys___specialfd(struct thread *td, struct __specialfd_args *args)
972 {
973 struct specialfd_eventfd ae;
974 int error;
975
976 switch (args->type) {
977 case SPECIALFD_EVENTFD:
978 if (args->len != sizeof(struct specialfd_eventfd)) {
979 error = EINVAL;
980 break;
981 }
982 error = copyin(args->req, &ae, sizeof(ae));
983 if (error != 0)
984 break;
985 if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK |
986 EFD_SEMAPHORE)) != 0) {
987 error = EINVAL;
988 break;
989 }
990 error = kern_specialfd(td, args->type, &ae);
991 break;
992 default:
993 error = EINVAL;
994 break;
995 }
996 return (error);
997 }
998
999 int
1000 poll_no_poll(int events)
1001 {
1002 /*
1003 * Return true for read/write. If the user asked for something
1004 * special, return POLLNVAL, so that clients have a way of
1005 * determining reliably whether or not the extended
1006 * functionality is present without hard-coding knowledge
1007 * of specific filesystem implementations.
1008 */
1009 if (events & ~POLLSTANDARD)
1010 return (POLLNVAL);
1011
1012 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1013 }
1014
1015 int
1016 sys_pselect(struct thread *td, struct pselect_args *uap)
1017 {
1018 struct timespec ts;
1019 struct timeval tv, *tvp;
1020 sigset_t set, *uset;
1021 int error;
1022
1023 if (uap->ts != NULL) {
1024 error = copyin(uap->ts, &ts, sizeof(ts));
1025 if (error != 0)
1026 return (error);
1027 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1028 tvp = &tv;
1029 } else
1030 tvp = NULL;
1031 if (uap->sm != NULL) {
1032 error = copyin(uap->sm, &set, sizeof(set));
1033 if (error != 0)
1034 return (error);
1035 uset = &set;
1036 } else
1037 uset = NULL;
1038 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1039 uset, NFDBITS));
1040 }
1041
1042 int
1043 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
1044 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
1045 {
1046 int error;
1047
1048 if (uset != NULL) {
1049 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1050 &td->td_oldsigmask, 0);
1051 if (error != 0)
1052 return (error);
1053 td->td_pflags |= TDP_OLDMASK;
1054 /*
1055 * Make sure that ast() is called on return to
1056 * usermode and TDP_OLDMASK is cleared, restoring old
1057 * sigmask.
1058 */
1059 ast_sched(td, TDA_SIGSUSPEND);
1060 }
1061 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
1062 return (error);
1063 }
1064
1065 #ifndef _SYS_SYSPROTO_H_
1066 struct select_args {
1067 int nd;
1068 fd_set *in, *ou, *ex;
1069 struct timeval *tv;
1070 };
1071 #endif
1072 int
1073 sys_select(struct thread *td, struct select_args *uap)
1074 {
1075 struct timeval tv, *tvp;
1076 int error;
1077
1078 if (uap->tv != NULL) {
1079 error = copyin(uap->tv, &tv, sizeof(tv));
1080 if (error)
1081 return (error);
1082 tvp = &tv;
1083 } else
1084 tvp = NULL;
1085
1086 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1087 NFDBITS));
1088 }
1089
1090 /*
1091 * In the unlikely case when user specified n greater then the last
1092 * open file descriptor, check that no bits are set after the last
1093 * valid fd. We must return EBADF if any is set.
1094 *
1095 * There are applications that rely on the behaviour.
1096 *
1097 * nd is fd_nfiles.
1098 */
1099 static int
1100 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
1101 {
1102 char *addr, *oaddr;
1103 int b, i, res;
1104 uint8_t bits;
1105
1106 if (nd >= ndu || fd_in == NULL)
1107 return (0);
1108
1109 oaddr = NULL;
1110 bits = 0; /* silence gcc */
1111 for (i = nd; i < ndu; i++) {
1112 b = i / NBBY;
1113 #if BYTE_ORDER == LITTLE_ENDIAN
1114 addr = (char *)fd_in + b;
1115 #else
1116 addr = (char *)fd_in;
1117 if (abi_nfdbits == NFDBITS) {
1118 addr += rounddown(b, sizeof(fd_mask)) +
1119 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
1120 } else {
1121 addr += rounddown(b, sizeof(uint32_t)) +
1122 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
1123 }
1124 #endif
1125 if (addr != oaddr) {
1126 res = fubyte(addr);
1127 if (res == -1)
1128 return (EFAULT);
1129 oaddr = addr;
1130 bits = res;
1131 }
1132 if ((bits & (1 << (i % NBBY))) != 0)
1133 return (EBADF);
1134 }
1135 return (0);
1136 }
1137
1138 int
1139 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1140 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1141 {
1142 struct filedesc *fdp;
1143 /*
1144 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1145 * infds with the new FD_SETSIZE of 1024, and more than enough for
1146 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1147 * of 256.
1148 */
1149 fd_mask s_selbits[howmany(2048, NFDBITS)];
1150 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1151 struct timeval rtv;
1152 sbintime_t asbt, precision, rsbt;
1153 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1154 int error, lf, ndu;
1155
1156 if (nd < 0)
1157 return (EINVAL);
1158 fdp = td->td_proc->p_fd;
1159 ndu = nd;
1160 lf = fdp->fd_nfiles;
1161 if (nd > lf)
1162 nd = lf;
1163
1164 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1165 if (error != 0)
1166 return (error);
1167 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1168 if (error != 0)
1169 return (error);
1170 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1171 if (error != 0)
1172 return (error);
1173
1174 /*
1175 * Allocate just enough bits for the non-null fd_sets. Use the
1176 * preallocated auto buffer if possible.
1177 */
1178 nfdbits = roundup(nd, NFDBITS);
1179 ncpbytes = nfdbits / NBBY;
1180 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1181 nbufbytes = 0;
1182 if (fd_in != NULL)
1183 nbufbytes += 2 * ncpbytes;
1184 if (fd_ou != NULL)
1185 nbufbytes += 2 * ncpbytes;
1186 if (fd_ex != NULL)
1187 nbufbytes += 2 * ncpbytes;
1188 if (nbufbytes <= sizeof s_selbits)
1189 selbits = &s_selbits[0];
1190 else
1191 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1192
1193 /*
1194 * Assign pointers into the bit buffers and fetch the input bits.
1195 * Put the output buffers together so that they can be bzeroed
1196 * together.
1197 */
1198 sbp = selbits;
1199 #define getbits(name, x) \
1200 do { \
1201 if (name == NULL) { \
1202 ibits[x] = NULL; \
1203 obits[x] = NULL; \
1204 } else { \
1205 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
1206 obits[x] = sbp; \
1207 sbp += ncpbytes / sizeof *sbp; \
1208 error = copyin(name, ibits[x], ncpubytes); \
1209 if (error != 0) \
1210 goto done; \
1211 if (ncpbytes != ncpubytes) \
1212 bzero((char *)ibits[x] + ncpubytes, \
1213 ncpbytes - ncpubytes); \
1214 } \
1215 } while (0)
1216 getbits(fd_in, 0);
1217 getbits(fd_ou, 1);
1218 getbits(fd_ex, 2);
1219 #undef getbits
1220
1221 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1222 /*
1223 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1224 * we are running under 32-bit emulation. This should be more
1225 * generic.
1226 */
1227 #define swizzle_fdset(bits) \
1228 if (abi_nfdbits != NFDBITS && bits != NULL) { \
1229 int i; \
1230 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
1231 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
1232 }
1233 #else
1234 #define swizzle_fdset(bits)
1235 #endif
1236
1237 /* Make sure the bit order makes it through an ABI transition */
1238 swizzle_fdset(ibits[0]);
1239 swizzle_fdset(ibits[1]);
1240 swizzle_fdset(ibits[2]);
1241
1242 if (nbufbytes != 0)
1243 bzero(selbits, nbufbytes / 2);
1244
1245 precision = 0;
1246 if (tvp != NULL) {
1247 rtv = *tvp;
1248 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1249 rtv.tv_usec >= 1000000) {
1250 error = EINVAL;
1251 goto done;
1252 }
1253 if (!timevalisset(&rtv))
1254 asbt = 0;
1255 else if (rtv.tv_sec <= INT32_MAX) {
1256 rsbt = tvtosbt(rtv);
1257 precision = rsbt;
1258 precision >>= tc_precexp;
1259 if (TIMESEL(&asbt, rsbt))
1260 asbt += tc_tick_sbt;
1261 if (asbt <= SBT_MAX - rsbt)
1262 asbt += rsbt;
1263 else
1264 asbt = -1;
1265 } else
1266 asbt = -1;
1267 } else
1268 asbt = -1;
1269 seltdinit(td);
1270 /* Iterate until the timeout expires or descriptors become ready. */
1271 for (;;) {
1272 error = selscan(td, ibits, obits, nd);
1273 if (error || td->td_retval[0] != 0)
1274 break;
1275 error = seltdwait(td, asbt, precision);
1276 if (error)
1277 break;
1278 error = selrescan(td, ibits, obits);
1279 if (error || td->td_retval[0] != 0)
1280 break;
1281 }
1282 seltdclear(td);
1283
1284 done:
1285 /* select is not restarted after signals... */
1286 if (error == ERESTART)
1287 error = EINTR;
1288 if (error == EWOULDBLOCK)
1289 error = 0;
1290
1291 /* swizzle bit order back, if necessary */
1292 swizzle_fdset(obits[0]);
1293 swizzle_fdset(obits[1]);
1294 swizzle_fdset(obits[2]);
1295 #undef swizzle_fdset
1296
1297 #define putbits(name, x) \
1298 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1299 error = error2;
1300 if (error == 0) {
1301 int error2;
1302
1303 putbits(fd_in, 0);
1304 putbits(fd_ou, 1);
1305 putbits(fd_ex, 2);
1306 #undef putbits
1307 }
1308 if (selbits != &s_selbits[0])
1309 free(selbits, M_SELECT);
1310
1311 return (error);
1312 }
1313 /*
1314 * Convert a select bit set to poll flags.
1315 *
1316 * The backend always returns POLLHUP/POLLERR if appropriate and we
1317 * return this as a set bit in any set.
1318 */
1319 static const int select_flags[3] = {
1320 POLLRDNORM | POLLHUP | POLLERR,
1321 POLLWRNORM | POLLHUP | POLLERR,
1322 POLLRDBAND | POLLERR
1323 };
1324
1325 /*
1326 * Compute the fo_poll flags required for a fd given by the index and
1327 * bit position in the fd_mask array.
1328 */
1329 static __inline int
1330 selflags(fd_mask **ibits, int idx, fd_mask bit)
1331 {
1332 int flags;
1333 int msk;
1334
1335 flags = 0;
1336 for (msk = 0; msk < 3; msk++) {
1337 if (ibits[msk] == NULL)
1338 continue;
1339 if ((ibits[msk][idx] & bit) == 0)
1340 continue;
1341 flags |= select_flags[msk];
1342 }
1343 return (flags);
1344 }
1345
1346 /*
1347 * Set the appropriate output bits given a mask of fired events and the
1348 * input bits originally requested.
1349 */
1350 static __inline int
1351 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1352 {
1353 int msk;
1354 int n;
1355
1356 n = 0;
1357 for (msk = 0; msk < 3; msk++) {
1358 if ((events & select_flags[msk]) == 0)
1359 continue;
1360 if (ibits[msk] == NULL)
1361 continue;
1362 if ((ibits[msk][idx] & bit) == 0)
1363 continue;
1364 /*
1365 * XXX Check for a duplicate set. This can occur because a
1366 * socket calls selrecord() twice for each poll() call
1367 * resulting in two selfds per real fd. selrescan() will
1368 * call selsetbits twice as a result.
1369 */
1370 if ((obits[msk][idx] & bit) != 0)
1371 continue;
1372 obits[msk][idx] |= bit;
1373 n++;
1374 }
1375
1376 return (n);
1377 }
1378
1379 /*
1380 * Traverse the list of fds attached to this thread's seltd and check for
1381 * completion.
1382 */
1383 static int
1384 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1385 {
1386 struct filedesc *fdp;
1387 struct selinfo *si;
1388 struct seltd *stp;
1389 struct selfd *sfp;
1390 struct selfd *sfn;
1391 struct file *fp;
1392 fd_mask bit;
1393 int fd, ev, n, idx;
1394 int error;
1395 bool only_user;
1396
1397 fdp = td->td_proc->p_fd;
1398 stp = td->td_sel;
1399 n = 0;
1400 only_user = FILEDESC_IS_ONLY_USER(fdp);
1401 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1402 fd = (int)(uintptr_t)sfp->sf_cookie;
1403 si = sfp->sf_si;
1404 selfdfree(stp, sfp);
1405 /* If the selinfo wasn't cleared the event didn't fire. */
1406 if (si != NULL)
1407 continue;
1408 if (only_user)
1409 error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1410 else
1411 error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1412 if (__predict_false(error != 0))
1413 return (error);
1414 idx = fd / NFDBITS;
1415 bit = (fd_mask)1 << (fd % NFDBITS);
1416 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1417 if (only_user)
1418 fput_only_user(fdp, fp);
1419 else
1420 fdrop(fp, td);
1421 if (ev != 0)
1422 n += selsetbits(ibits, obits, idx, bit, ev);
1423 }
1424 stp->st_flags = 0;
1425 td->td_retval[0] = n;
1426 return (0);
1427 }
1428
1429 /*
1430 * Perform the initial filedescriptor scan and register ourselves with
1431 * each selinfo.
1432 */
1433 static int
1434 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1435 {
1436 struct filedesc *fdp;
1437 struct file *fp;
1438 fd_mask bit;
1439 int ev, flags, end, fd;
1440 int n, idx;
1441 int error;
1442 bool only_user;
1443
1444 fdp = td->td_proc->p_fd;
1445 n = 0;
1446 only_user = FILEDESC_IS_ONLY_USER(fdp);
1447 for (idx = 0, fd = 0; fd < nfd; idx++) {
1448 end = imin(fd + NFDBITS, nfd);
1449 for (bit = 1; fd < end; bit <<= 1, fd++) {
1450 /* Compute the list of events we're interested in. */
1451 flags = selflags(ibits, idx, bit);
1452 if (flags == 0)
1453 continue;
1454 if (only_user)
1455 error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1456 else
1457 error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1458 if (__predict_false(error != 0))
1459 return (error);
1460 selfdalloc(td, (void *)(uintptr_t)fd);
1461 ev = fo_poll(fp, flags, td->td_ucred, td);
1462 if (only_user)
1463 fput_only_user(fdp, fp);
1464 else
1465 fdrop(fp, td);
1466 if (ev != 0)
1467 n += selsetbits(ibits, obits, idx, bit, ev);
1468 }
1469 }
1470
1471 td->td_retval[0] = n;
1472 return (0);
1473 }
1474
1475 int
1476 sys_poll(struct thread *td, struct poll_args *uap)
1477 {
1478 struct timespec ts, *tsp;
1479
1480 if (uap->timeout != INFTIM) {
1481 if (uap->timeout < 0)
1482 return (EINVAL);
1483 ts.tv_sec = uap->timeout / 1000;
1484 ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1485 tsp = &ts;
1486 } else
1487 tsp = NULL;
1488
1489 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1490 }
1491
1492 /*
1493 * kfds points to an array in the kernel.
1494 */
1495 int
1496 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds,
1497 struct timespec *tsp, sigset_t *uset)
1498 {
1499 sbintime_t sbt, precision, tmp;
1500 time_t over;
1501 struct timespec ts;
1502 int error;
1503
1504 precision = 0;
1505 if (tsp != NULL) {
1506 if (!timespecvalid_interval(tsp))
1507 return (EINVAL);
1508 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1509 sbt = 0;
1510 else {
1511 ts = *tsp;
1512 if (ts.tv_sec > INT32_MAX / 2) {
1513 over = ts.tv_sec - INT32_MAX / 2;
1514 ts.tv_sec -= over;
1515 } else
1516 over = 0;
1517 tmp = tstosbt(ts);
1518 precision = tmp;
1519 precision >>= tc_precexp;
1520 if (TIMESEL(&sbt, tmp))
1521 sbt += tc_tick_sbt;
1522 sbt += tmp;
1523 }
1524 } else
1525 sbt = -1;
1526
1527 if (uset != NULL) {
1528 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1529 &td->td_oldsigmask, 0);
1530 if (error)
1531 return (error);
1532 td->td_pflags |= TDP_OLDMASK;
1533 /*
1534 * Make sure that ast() is called on return to
1535 * usermode and TDP_OLDMASK is cleared, restoring old
1536 * sigmask.
1537 */
1538 ast_sched(td, TDA_SIGSUSPEND);
1539 }
1540
1541 seltdinit(td);
1542 /* Iterate until the timeout expires or descriptors become ready. */
1543 for (;;) {
1544 error = pollscan(td, kfds, nfds);
1545 if (error || td->td_retval[0] != 0)
1546 break;
1547 error = seltdwait(td, sbt, precision);
1548 if (error)
1549 break;
1550 error = pollrescan(td);
1551 if (error || td->td_retval[0] != 0)
1552 break;
1553 }
1554 seltdclear(td);
1555
1556 /* poll is not restarted after signals... */
1557 if (error == ERESTART)
1558 error = EINTR;
1559 if (error == EWOULDBLOCK)
1560 error = 0;
1561 return (error);
1562 }
1563
1564 int
1565 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1566 {
1567 struct timespec ts, *tsp;
1568 sigset_t set, *ssp;
1569 int error;
1570
1571 if (uap->ts != NULL) {
1572 error = copyin(uap->ts, &ts, sizeof(ts));
1573 if (error)
1574 return (error);
1575 tsp = &ts;
1576 } else
1577 tsp = NULL;
1578 if (uap->set != NULL) {
1579 error = copyin(uap->set, &set, sizeof(set));
1580 if (error)
1581 return (error);
1582 ssp = &set;
1583 } else
1584 ssp = NULL;
1585 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1586 }
1587
1588 /*
1589 * ufds points to an array in user space.
1590 */
1591 int
1592 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1593 struct timespec *tsp, sigset_t *set)
1594 {
1595 struct pollfd *kfds;
1596 struct pollfd stackfds[32];
1597 int error;
1598
1599 if (kern_poll_maxfds(nfds))
1600 return (EINVAL);
1601 if (nfds > nitems(stackfds))
1602 kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1603 else
1604 kfds = stackfds;
1605 error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1606 if (error != 0)
1607 goto out;
1608
1609 error = kern_poll_kfds(td, kfds, nfds, tsp, set);
1610 if (error == 0)
1611 error = pollout(td, kfds, ufds, nfds);
1612
1613 out:
1614 if (nfds > nitems(stackfds))
1615 free(kfds, M_TEMP);
1616 return (error);
1617 }
1618
1619 bool
1620 kern_poll_maxfds(u_int nfds)
1621 {
1622
1623 /*
1624 * This is kinda bogus. We have fd limits, but that is not
1625 * really related to the size of the pollfd array. Make sure
1626 * we let the process use at least FD_SETSIZE entries and at
1627 * least enough for the system-wide limits. We want to be reasonably
1628 * safe, but not overly restrictive.
1629 */
1630 return (nfds > maxfilesperproc && nfds > FD_SETSIZE);
1631 }
1632
1633 static int
1634 pollrescan(struct thread *td)
1635 {
1636 struct seltd *stp;
1637 struct selfd *sfp;
1638 struct selfd *sfn;
1639 struct selinfo *si;
1640 struct filedesc *fdp;
1641 struct file *fp;
1642 struct pollfd *fd;
1643 int n, error;
1644 bool only_user;
1645
1646 n = 0;
1647 fdp = td->td_proc->p_fd;
1648 stp = td->td_sel;
1649 only_user = FILEDESC_IS_ONLY_USER(fdp);
1650 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1651 fd = (struct pollfd *)sfp->sf_cookie;
1652 si = sfp->sf_si;
1653 selfdfree(stp, sfp);
1654 /* If the selinfo wasn't cleared the event didn't fire. */
1655 if (si != NULL)
1656 continue;
1657 if (only_user)
1658 error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp);
1659 else
1660 error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp);
1661 if (__predict_false(error != 0)) {
1662 fd->revents = POLLNVAL;
1663 n++;
1664 continue;
1665 }
1666 /*
1667 * Note: backend also returns POLLHUP and
1668 * POLLERR if appropriate.
1669 */
1670 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1671 if (only_user)
1672 fput_only_user(fdp, fp);
1673 else
1674 fdrop(fp, td);
1675 if (fd->revents != 0)
1676 n++;
1677 }
1678 stp->st_flags = 0;
1679 td->td_retval[0] = n;
1680 return (0);
1681 }
1682
1683 static int
1684 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1685 {
1686 int error = 0;
1687 u_int i = 0;
1688 u_int n = 0;
1689
1690 for (i = 0; i < nfd; i++) {
1691 error = copyout(&fds->revents, &ufds->revents,
1692 sizeof(ufds->revents));
1693 if (error)
1694 return (error);
1695 if (fds->revents != 0)
1696 n++;
1697 fds++;
1698 ufds++;
1699 }
1700 td->td_retval[0] = n;
1701 return (0);
1702 }
1703
1704 static int
1705 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1706 {
1707 struct filedesc *fdp;
1708 struct file *fp;
1709 int i, n, error;
1710 bool only_user;
1711
1712 n = 0;
1713 fdp = td->td_proc->p_fd;
1714 only_user = FILEDESC_IS_ONLY_USER(fdp);
1715 for (i = 0; i < nfd; i++, fds++) {
1716 if (fds->fd < 0) {
1717 fds->revents = 0;
1718 continue;
1719 }
1720 if (only_user)
1721 error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp);
1722 else
1723 error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp);
1724 if (__predict_false(error != 0)) {
1725 fds->revents = POLLNVAL;
1726 n++;
1727 continue;
1728 }
1729 /*
1730 * Note: backend also returns POLLHUP and
1731 * POLLERR if appropriate.
1732 */
1733 selfdalloc(td, fds);
1734 fds->revents = fo_poll(fp, fds->events,
1735 td->td_ucred, td);
1736 if (only_user)
1737 fput_only_user(fdp, fp);
1738 else
1739 fdrop(fp, td);
1740 /*
1741 * POSIX requires POLLOUT to be never
1742 * set simultaneously with POLLHUP.
1743 */
1744 if ((fds->revents & POLLHUP) != 0)
1745 fds->revents &= ~POLLOUT;
1746
1747 if (fds->revents != 0)
1748 n++;
1749 }
1750 td->td_retval[0] = n;
1751 return (0);
1752 }
1753
1754 /*
1755 * XXX This was created specifically to support netncp and netsmb. This
1756 * allows the caller to specify a socket to wait for events on. It returns
1757 * 0 if any events matched and an error otherwise. There is no way to
1758 * determine which events fired.
1759 */
1760 int
1761 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1762 {
1763 struct timeval rtv;
1764 sbintime_t asbt, precision, rsbt;
1765 int error;
1766
1767 precision = 0; /* stupid gcc! */
1768 if (tvp != NULL) {
1769 rtv = *tvp;
1770 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1771 rtv.tv_usec >= 1000000)
1772 return (EINVAL);
1773 if (!timevalisset(&rtv))
1774 asbt = 0;
1775 else if (rtv.tv_sec <= INT32_MAX) {
1776 rsbt = tvtosbt(rtv);
1777 precision = rsbt;
1778 precision >>= tc_precexp;
1779 if (TIMESEL(&asbt, rsbt))
1780 asbt += tc_tick_sbt;
1781 if (asbt <= SBT_MAX - rsbt)
1782 asbt += rsbt;
1783 else
1784 asbt = -1;
1785 } else
1786 asbt = -1;
1787 } else
1788 asbt = -1;
1789 seltdinit(td);
1790 /*
1791 * Iterate until the timeout expires or the socket becomes ready.
1792 */
1793 for (;;) {
1794 selfdalloc(td, NULL);
1795 if (sopoll(so, events, NULL, td) != 0) {
1796 error = 0;
1797 break;
1798 }
1799 error = seltdwait(td, asbt, precision);
1800 if (error)
1801 break;
1802 }
1803 seltdclear(td);
1804 /* XXX Duplicates ncp/smb behavior. */
1805 if (error == ERESTART)
1806 error = 0;
1807 return (error);
1808 }
1809
1810 /*
1811 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1812 * have two select sets, one for read and another for write.
1813 */
1814 static void
1815 selfdalloc(struct thread *td, void *cookie)
1816 {
1817 struct seltd *stp;
1818
1819 stp = td->td_sel;
1820 if (stp->st_free1 == NULL)
1821 stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO);
1822 stp->st_free1->sf_td = stp;
1823 stp->st_free1->sf_cookie = cookie;
1824 if (stp->st_free2 == NULL)
1825 stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO);
1826 stp->st_free2->sf_td = stp;
1827 stp->st_free2->sf_cookie = cookie;
1828 }
1829
1830 static void
1831 selfdfree(struct seltd *stp, struct selfd *sfp)
1832 {
1833 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1834 /*
1835 * Paired with doselwakeup.
1836 */
1837 if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) {
1838 mtx_lock(sfp->sf_mtx);
1839 if (sfp->sf_si != NULL) {
1840 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1841 }
1842 mtx_unlock(sfp->sf_mtx);
1843 }
1844 free(sfp, M_SELFD);
1845 }
1846
1847 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1848 void
1849 seldrain(struct selinfo *sip)
1850 {
1851
1852 /*
1853 * This feature is already provided by doselwakeup(), thus it is
1854 * enough to go for it.
1855 * Eventually, the context, should take care to avoid races
1856 * between thread calling select()/poll() and file descriptor
1857 * detaching, but, again, the races are just the same as
1858 * selwakeup().
1859 */
1860 doselwakeup(sip, -1);
1861 }
1862
1863 /*
1864 * Record a select request.
1865 */
1866 void
1867 selrecord(struct thread *selector, struct selinfo *sip)
1868 {
1869 struct selfd *sfp;
1870 struct seltd *stp;
1871 struct mtx *mtxp;
1872
1873 stp = selector->td_sel;
1874 /*
1875 * Don't record when doing a rescan.
1876 */
1877 if (stp->st_flags & SELTD_RESCAN)
1878 return;
1879 /*
1880 * Grab one of the preallocated descriptors.
1881 */
1882 sfp = NULL;
1883 if ((sfp = stp->st_free1) != NULL)
1884 stp->st_free1 = NULL;
1885 else if ((sfp = stp->st_free2) != NULL)
1886 stp->st_free2 = NULL;
1887 else
1888 panic("selrecord: No free selfd on selq");
1889 mtxp = sip->si_mtx;
1890 if (mtxp == NULL)
1891 mtxp = mtx_pool_find(mtxpool_select, sip);
1892 /*
1893 * Initialize the sfp and queue it in the thread.
1894 */
1895 sfp->sf_si = sip;
1896 sfp->sf_mtx = mtxp;
1897 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1898 /*
1899 * Now that we've locked the sip, check for initialization.
1900 */
1901 mtx_lock(mtxp);
1902 if (sip->si_mtx == NULL) {
1903 sip->si_mtx = mtxp;
1904 TAILQ_INIT(&sip->si_tdlist);
1905 }
1906 /*
1907 * Add this thread to the list of selfds listening on this selinfo.
1908 */
1909 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1910 mtx_unlock(sip->si_mtx);
1911 }
1912
1913 /* Wake up a selecting thread. */
1914 void
1915 selwakeup(struct selinfo *sip)
1916 {
1917 doselwakeup(sip, -1);
1918 }
1919
1920 /* Wake up a selecting thread, and set its priority. */
1921 void
1922 selwakeuppri(struct selinfo *sip, int pri)
1923 {
1924 doselwakeup(sip, pri);
1925 }
1926
1927 /*
1928 * Do a wakeup when a selectable event occurs.
1929 */
1930 static void
1931 doselwakeup(struct selinfo *sip, int pri)
1932 {
1933 struct selfd *sfp;
1934 struct selfd *sfn;
1935 struct seltd *stp;
1936
1937 /* If it's not initialized there can't be any waiters. */
1938 if (sip->si_mtx == NULL)
1939 return;
1940 /*
1941 * Locking the selinfo locks all selfds associated with it.
1942 */
1943 mtx_lock(sip->si_mtx);
1944 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1945 /*
1946 * Once we remove this sfp from the list and clear the
1947 * sf_si seltdclear will know to ignore this si.
1948 */
1949 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1950 stp = sfp->sf_td;
1951 mtx_lock(&stp->st_mtx);
1952 stp->st_flags |= SELTD_PENDING;
1953 cv_broadcastpri(&stp->st_wait, pri);
1954 mtx_unlock(&stp->st_mtx);
1955 /*
1956 * Paired with selfdfree.
1957 *
1958 * Storing this only after the wakeup provides an invariant that
1959 * stp is not used after selfdfree returns.
1960 */
1961 atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL);
1962 }
1963 mtx_unlock(sip->si_mtx);
1964 }
1965
1966 static void
1967 seltdinit(struct thread *td)
1968 {
1969 struct seltd *stp;
1970
1971 stp = td->td_sel;
1972 if (stp != NULL) {
1973 MPASS(stp->st_flags == 0);
1974 MPASS(STAILQ_EMPTY(&stp->st_selq));
1975 return;
1976 }
1977 stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1978 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1979 cv_init(&stp->st_wait, "select");
1980 stp->st_flags = 0;
1981 STAILQ_INIT(&stp->st_selq);
1982 td->td_sel = stp;
1983 }
1984
1985 static int
1986 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1987 {
1988 struct seltd *stp;
1989 int error;
1990
1991 stp = td->td_sel;
1992 /*
1993 * An event of interest may occur while we do not hold the seltd
1994 * locked so check the pending flag before we sleep.
1995 */
1996 mtx_lock(&stp->st_mtx);
1997 /*
1998 * Any further calls to selrecord will be a rescan.
1999 */
2000 stp->st_flags |= SELTD_RESCAN;
2001 if (stp->st_flags & SELTD_PENDING) {
2002 mtx_unlock(&stp->st_mtx);
2003 return (0);
2004 }
2005 if (sbt == 0)
2006 error = EWOULDBLOCK;
2007 else if (sbt != -1)
2008 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
2009 sbt, precision, C_ABSOLUTE);
2010 else
2011 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
2012 mtx_unlock(&stp->st_mtx);
2013
2014 return (error);
2015 }
2016
2017 void
2018 seltdfini(struct thread *td)
2019 {
2020 struct seltd *stp;
2021
2022 stp = td->td_sel;
2023 if (stp == NULL)
2024 return;
2025 MPASS(stp->st_flags == 0);
2026 MPASS(STAILQ_EMPTY(&stp->st_selq));
2027 if (stp->st_free1)
2028 free(stp->st_free1, M_SELFD);
2029 if (stp->st_free2)
2030 free(stp->st_free2, M_SELFD);
2031 td->td_sel = NULL;
2032 cv_destroy(&stp->st_wait);
2033 mtx_destroy(&stp->st_mtx);
2034 free(stp, M_SELECT);
2035 }
2036
2037 /*
2038 * Remove the references to the thread from all of the objects we were
2039 * polling.
2040 */
2041 static void
2042 seltdclear(struct thread *td)
2043 {
2044 struct seltd *stp;
2045 struct selfd *sfp;
2046 struct selfd *sfn;
2047
2048 stp = td->td_sel;
2049 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
2050 selfdfree(stp, sfp);
2051 stp->st_flags = 0;
2052 }
2053
2054 static void selectinit(void *);
2055 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
2056 static void
2057 selectinit(void *dummy __unused)
2058 {
2059
2060 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
2061 }
2062
2063 /*
2064 * Set up a syscall return value that follows the convention specified for
2065 * posix_* functions.
2066 */
2067 int
2068 kern_posix_error(struct thread *td, int error)
2069 {
2070
2071 if (error <= 0)
2072 return (error);
2073 td->td_errno = error;
2074 td->td_pflags |= TDP_NERRNO;
2075 td->td_retval[0] = error;
2076 return (0);
2077 }
Cache object: 865f2cafa227c7a0170ea80f0e898e0a
|