2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
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
34 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
54 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/malloc.h>
59 #include <sys/resourcevar.h>
60 #include <sys/selinfo.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/vnode.h>
68 #include <sys/condvar.h>
70 #include <sys/ktrace.h>
73 #include <security/audit/audit.h>
75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
79 static int pollscan(struct thread *, struct pollfd *, u_int);
80 static int pollrescan(struct thread *);
81 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
82 static int selrescan(struct thread *, fd_mask **, fd_mask **);
83 static void selfdalloc(struct thread *, void *);
84 static void selfdfree(struct seltd *, struct selfd *);
85 static int dofileread(struct thread *, int, struct file *, struct uio *,
87 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
89 static void doselwakeup(struct selinfo *, int);
90 static void seltdinit(struct thread *);
91 static int seltdwait(struct thread *, int);
92 static void seltdclear(struct thread *);
95 * One seltd per-thread allocated on demand as needed.
97 * t - protected by st_mtx
98 * k - Only accessed by curthread or read-only
101 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
102 struct selfd *st_free1; /* (k) free fd for read set. */
103 struct selfd *st_free2; /* (k) free fd for write set. */
104 struct mtx st_mtx; /* Protects struct seltd */
105 struct cv st_wait; /* (t) Wait channel. */
106 int st_flags; /* (t) SELTD_ flags. */
109 #define SELTD_PENDING 0x0001 /* We have pending events. */
110 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
113 * One selfd allocated per-thread per-file-descriptor.
114 * f - protected by sf_mtx
117 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
118 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
119 struct selinfo *sf_si; /* (f) selinfo when linked. */
120 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
121 struct seltd *sf_td; /* (k) owning seltd. */
122 void *sf_cookie; /* (k) fd or pollfd. */
125 static uma_zone_t selfd_zone;
127 #ifndef _SYS_SYSPROTO_H_
137 struct read_args *uap;
143 if (uap->nbyte > INT_MAX)
145 aiov.iov_base = uap->buf;
146 aiov.iov_len = uap->nbyte;
147 auio.uio_iov = &aiov;
149 auio.uio_resid = uap->nbyte;
150 auio.uio_segflg = UIO_USERSPACE;
151 error = kern_readv(td, uap->fd, &auio);
156 * Positioned read system call
158 #ifndef _SYS_SYSPROTO_H_
170 struct pread_args *uap;
176 if (uap->nbyte > INT_MAX)
178 aiov.iov_base = uap->buf;
179 aiov.iov_len = uap->nbyte;
180 auio.uio_iov = &aiov;
182 auio.uio_resid = uap->nbyte;
183 auio.uio_segflg = UIO_USERSPACE;
184 error = kern_preadv(td, uap->fd, &auio, uap->offset);
189 freebsd6_pread(td, uap)
191 struct freebsd6_pread_args *uap;
193 struct pread_args oargs;
196 oargs.buf = uap->buf;
197 oargs.nbyte = uap->nbyte;
198 oargs.offset = uap->offset;
199 return (pread(td, &oargs));
203 * Scatter read system call.
205 #ifndef _SYS_SYSPROTO_H_
213 readv(struct thread *td, struct readv_args *uap)
218 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
221 error = kern_readv(td, uap->fd, auio);
227 kern_readv(struct thread *td, int fd, struct uio *auio)
232 error = fget_read(td, fd, &fp);
235 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
241 * Scatter positioned read system call.
243 #ifndef _SYS_SYSPROTO_H_
252 preadv(struct thread *td, struct preadv_args *uap)
257 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
260 error = kern_preadv(td, uap->fd, auio, uap->offset);
266 kern_preadv(td, fd, auio, offset)
275 error = fget_read(td, fd, &fp);
278 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
280 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
283 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
289 * Common code for readv and preadv that reads data in
290 * from a file using the passed in uio, offset, and flags.
293 dofileread(td, fd, fp, auio, offset, flags)
304 struct uio *ktruio = NULL;
307 /* Finish zero length reads right here */
308 if (auio->uio_resid == 0) {
309 td->td_retval[0] = 0;
312 auio->uio_rw = UIO_READ;
313 auio->uio_offset = offset;
316 if (KTRPOINT(td, KTR_GENIO))
317 ktruio = cloneuio(auio);
319 cnt = auio->uio_resid;
320 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
321 if (auio->uio_resid != cnt && (error == ERESTART ||
322 error == EINTR || error == EWOULDBLOCK))
325 cnt -= auio->uio_resid;
327 if (ktruio != NULL) {
328 ktruio->uio_resid = cnt;
329 ktrgenio(fd, UIO_READ, ktruio, error);
332 td->td_retval[0] = cnt;
336 #ifndef _SYS_SYSPROTO_H_
346 struct write_args *uap;
352 if (uap->nbyte > INT_MAX)
354 aiov.iov_base = (void *)(uintptr_t)uap->buf;
355 aiov.iov_len = uap->nbyte;
356 auio.uio_iov = &aiov;
358 auio.uio_resid = uap->nbyte;
359 auio.uio_segflg = UIO_USERSPACE;
360 error = kern_writev(td, uap->fd, &auio);
365 * Positioned write system call.
367 #ifndef _SYS_SYSPROTO_H_
379 struct pwrite_args *uap;
385 if (uap->nbyte > INT_MAX)
387 aiov.iov_base = (void *)(uintptr_t)uap->buf;
388 aiov.iov_len = uap->nbyte;
389 auio.uio_iov = &aiov;
391 auio.uio_resid = uap->nbyte;
392 auio.uio_segflg = UIO_USERSPACE;
393 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
398 freebsd6_pwrite(td, uap)
400 struct freebsd6_pwrite_args *uap;
402 struct pwrite_args oargs;
405 oargs.buf = uap->buf;
406 oargs.nbyte = uap->nbyte;
407 oargs.offset = uap->offset;
408 return (pwrite(td, &oargs));
412 * Gather write system call.
414 #ifndef _SYS_SYSPROTO_H_
422 writev(struct thread *td, struct writev_args *uap)
427 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
430 error = kern_writev(td, uap->fd, auio);
436 kern_writev(struct thread *td, int fd, struct uio *auio)
441 error = fget_write(td, fd, &fp);
444 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
450 * Gather positioned write system call.
452 #ifndef _SYS_SYSPROTO_H_
453 struct pwritev_args {
461 pwritev(struct thread *td, struct pwritev_args *uap)
466 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
469 error = kern_pwritev(td, uap->fd, auio, uap->offset);
475 kern_pwritev(td, fd, auio, offset)
484 error = fget_write(td, fd, &fp);
487 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
489 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
492 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
498 * Common code for writev and pwritev that writes data to
499 * a file using the passed in uio, offset, and flags.
502 dofilewrite(td, fd, fp, auio, offset, flags)
513 struct uio *ktruio = NULL;
516 auio->uio_rw = UIO_WRITE;
518 auio->uio_offset = offset;
520 if (KTRPOINT(td, KTR_GENIO))
521 ktruio = cloneuio(auio);
523 cnt = auio->uio_resid;
524 if (fp->f_type == DTYPE_VNODE)
526 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
527 if (auio->uio_resid != cnt && (error == ERESTART ||
528 error == EINTR || error == EWOULDBLOCK))
530 /* Socket layer is responsible for issuing SIGPIPE. */
531 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
532 PROC_LOCK(td->td_proc);
533 psignal(td->td_proc, SIGPIPE);
534 PROC_UNLOCK(td->td_proc);
537 cnt -= auio->uio_resid;
539 if (ktruio != NULL) {
540 ktruio->uio_resid = cnt;
541 ktrgenio(fd, UIO_WRITE, ktruio, error);
544 td->td_retval[0] = cnt;
549 * Truncate a file given a file descriptor.
551 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
552 * descriptor isn't writable.
555 kern_ftruncate(td, fd, length)
566 error = fget(td, fd, &fp);
569 AUDIT_ARG(file, td->td_proc, fp);
570 if (!(fp->f_flag & FWRITE)) {
574 error = fo_truncate(fp, length, td->td_ucred, td);
579 #ifndef _SYS_SYSPROTO_H_
580 struct ftruncate_args {
589 struct ftruncate_args *uap;
592 return (kern_ftruncate(td, uap->fd, uap->length));
595 #if defined(COMPAT_43)
596 #ifndef _SYS_SYSPROTO_H_
597 struct oftruncate_args {
605 struct oftruncate_args *uap;
608 return (kern_ftruncate(td, uap->fd, uap->length));
610 #endif /* COMPAT_43 */
612 #ifndef _SYS_SYSPROTO_H_
621 ioctl(struct thread *td, struct ioctl_args *uap)
628 if (uap->com > 0xffffffff) {
630 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
631 td->td_proc->p_pid, td->td_name, uap->com);
632 uap->com &= 0xffffffff;
637 * Interpret high order word to find amount of data to be
638 * copied to/from the user's address space.
640 size = IOCPARM_LEN(com);
641 if ((size > IOCPARM_MAX) ||
642 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
643 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
644 ((com & IOC_OUT) && size == 0) ||
646 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
648 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
652 if (com & IOC_VOID) {
653 /* Integer argument. */
654 arg = (intptr_t)uap->data;
658 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
660 data = (void *)&uap->data;
662 error = copyin(uap->data, data, (u_int)size);
665 free(data, M_IOCTLOPS);
668 } else if (com & IOC_OUT) {
670 * Zero the buffer so the user always
671 * gets back something deterministic.
676 error = kern_ioctl(td, uap->fd, com, data);
678 if (error == 0 && (com & IOC_OUT))
679 error = copyout(data, uap->data, (u_int)size);
682 free(data, M_IOCTLOPS);
687 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
690 struct filedesc *fdp;
694 if ((error = fget(td, fd, &fp)) != 0)
696 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
700 fdp = td->td_proc->p_fd;
704 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
705 FILEDESC_XUNLOCK(fdp);
709 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
710 FILEDESC_XUNLOCK(fdp);
713 if ((tmp = *(int *)data))
714 atomic_set_int(&fp->f_flag, FNONBLOCK);
716 atomic_clear_int(&fp->f_flag, FNONBLOCK);
720 if ((tmp = *(int *)data))
721 atomic_set_int(&fp->f_flag, FASYNC);
723 atomic_clear_int(&fp->f_flag, FASYNC);
728 error = fo_ioctl(fp, com, data, td->td_ucred, td);
734 #ifndef _SYS_SYSPROTO_H_
737 fd_set *in, *ou, *ex;
743 register struct thread *td;
744 register struct select_args *uap;
746 struct timeval tv, *tvp;
749 if (uap->tv != NULL) {
750 error = copyin(uap->tv, &tv, sizeof(tv));
757 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp));
761 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
762 fd_set *fd_ex, struct timeval *tvp)
764 struct filedesc *fdp;
766 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
767 * infds with the new FD_SETSIZE of 1024, and more than enough for
768 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
771 fd_mask s_selbits[howmany(2048, NFDBITS)];
772 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
773 struct timeval atv, rtv, ttv;
775 u_int nbufbytes, ncpbytes, nfdbits;
779 fdp = td->td_proc->p_fd;
782 if (nd > td->td_proc->p_fd->fd_nfiles)
783 nd = td->td_proc->p_fd->fd_nfiles; /* forgiving; slightly wrong */
784 FILEDESC_SUNLOCK(fdp);
787 * Allocate just enough bits for the non-null fd_sets. Use the
788 * preallocated auto buffer if possible.
790 nfdbits = roundup(nd, NFDBITS);
791 ncpbytes = nfdbits / NBBY;
794 nbufbytes += 2 * ncpbytes;
796 nbufbytes += 2 * ncpbytes;
798 nbufbytes += 2 * ncpbytes;
799 if (nbufbytes <= sizeof s_selbits)
800 selbits = &s_selbits[0];
802 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
805 * Assign pointers into the bit buffers and fetch the input bits.
806 * Put the output buffers together so that they can be bzeroed
810 #define getbits(name, x) \
815 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
817 sbp += ncpbytes / sizeof *sbp; \
818 error = copyin(name, ibits[x], ncpbytes); \
828 bzero(selbits, nbufbytes / 2);
832 if (itimerfix(&atv)) {
836 getmicrouptime(&rtv);
837 timevaladd(&atv, &rtv);
844 /* Iterate until the timeout expires or descriptors become ready. */
846 error = selscan(td, ibits, obits, nd);
847 if (error || td->td_retval[0] != 0)
849 if (atv.tv_sec || atv.tv_usec) {
850 getmicrouptime(&rtv);
851 if (timevalcmp(&rtv, &atv, >=))
854 timevalsub(&ttv, &rtv);
855 timo = ttv.tv_sec > 24 * 60 * 60 ?
856 24 * 60 * 60 * hz : tvtohz(&ttv);
858 error = seltdwait(td, timo);
861 error = selrescan(td, ibits, obits);
862 if (error || td->td_retval[0] != 0)
868 /* select is not restarted after signals... */
869 if (error == ERESTART)
871 if (error == EWOULDBLOCK)
873 #define putbits(name, x) \
874 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
884 if (selbits != &s_selbits[0])
885 free(selbits, M_SELECT);
891 * Traverse the list of fds attached to this thread's seltd and check for
895 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
904 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
905 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
906 struct filedesc *fdp = td->td_proc->p_fd;
910 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
911 fd = (int)(uintptr_t)sfp->sf_cookie;
914 /* If the selinfo wasn't cleared the event didn't fire. */
917 if ((fp = fget_locked(fdp, fd)) == NULL) {
918 FILEDESC_SUNLOCK(fdp);
921 for (msk = 0; msk < 3; msk++) {
922 if (ibits[msk] == NULL)
924 if ((ibits[msk][fd/NFDBITS] &
925 ((fd_mask) 1 << (fd % NFDBITS))) == 0)
927 if (fo_poll(fp, flag[msk], td->td_ucred, td)) {
928 obits[msk][(fd)/NFDBITS] |=
929 ((fd_mask)1 << ((fd) % NFDBITS));
934 FILEDESC_SUNLOCK(fdp);
936 td->td_retval[0] = n;
941 * Perform the initial filedescriptor scan and register ourselves with
945 selscan(td, ibits, obits, nfd)
947 fd_mask **ibits, **obits;
954 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
955 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
956 struct filedesc *fdp = td->td_proc->p_fd;
959 for (msk = 0; msk < 3; msk++) {
960 if (ibits[msk] == NULL)
962 for (i = 0; i < nfd; i += NFDBITS) {
963 bits = ibits[msk][i/NFDBITS];
964 /* ffs(int mask) not portable, fd_mask is long */
965 for (fd = i; bits && fd < nfd; fd++, bits >>= 1) {
968 if ((fp = fget_locked(fdp, fd)) == NULL) {
969 FILEDESC_SUNLOCK(fdp);
972 selfdalloc(td, (void *)(uintptr_t)fd);
973 if (fo_poll(fp, flag[msk], td->td_ucred,
975 obits[msk][(fd)/NFDBITS] |=
976 ((fd_mask)1 << ((fd) % NFDBITS));
982 FILEDESC_SUNLOCK(fdp);
983 td->td_retval[0] = n;
987 #ifndef _SYS_SYSPROTO_H_
997 struct poll_args *uap;
1000 struct pollfd smallbits[32];
1001 struct timeval atv, rtv, ttv;
1002 int error = 0, timo;
1007 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1009 ni = nfds * sizeof(struct pollfd);
1010 if (ni > sizeof(smallbits))
1011 bits = malloc(ni, M_TEMP, M_WAITOK);
1014 error = copyin(uap->fds, bits, ni);
1017 if (uap->timeout != INFTIM) {
1018 atv.tv_sec = uap->timeout / 1000;
1019 atv.tv_usec = (uap->timeout % 1000) * 1000;
1020 if (itimerfix(&atv)) {
1024 getmicrouptime(&rtv);
1025 timevaladd(&atv, &rtv);
1032 /* Iterate until the timeout expires or descriptors become ready. */
1034 error = pollscan(td, bits, nfds);
1035 if (error || td->td_retval[0] != 0)
1037 if (atv.tv_sec || atv.tv_usec) {
1038 getmicrouptime(&rtv);
1039 if (timevalcmp(&rtv, &atv, >=))
1042 timevalsub(&ttv, &rtv);
1043 timo = ttv.tv_sec > 24 * 60 * 60 ?
1044 24 * 60 * 60 * hz : tvtohz(&ttv);
1046 error = seltdwait(td, timo);
1049 error = pollrescan(td);
1050 if (error || td->td_retval[0] != 0)
1056 /* poll is not restarted after signals... */
1057 if (error == ERESTART)
1059 if (error == EWOULDBLOCK)
1062 error = copyout(bits, uap->fds, ni);
1067 if (ni > sizeof(smallbits))
1073 pollrescan(struct thread *td)
1079 struct filedesc *fdp;
1085 fdp = td->td_proc->p_fd;
1087 FILEDESC_SLOCK(fdp);
1088 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1089 fd = (struct pollfd *)sfp->sf_cookie;
1091 selfdfree(stp, sfp);
1092 /* If the selinfo wasn't cleared the event didn't fire. */
1095 fp = fdp->fd_ofiles[fd->fd];
1097 fd->revents = POLLNVAL;
1102 * Note: backend also returns POLLHUP and
1103 * POLLERR if appropriate.
1105 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1106 if (fd->revents != 0)
1109 FILEDESC_SUNLOCK(fdp);
1111 td->td_retval[0] = n;
1117 pollscan(td, fds, nfd)
1122 struct filedesc *fdp = td->td_proc->p_fd;
1127 FILEDESC_SLOCK(fdp);
1128 for (i = 0; i < nfd; i++, fds++) {
1129 if (fds->fd >= fdp->fd_nfiles) {
1130 fds->revents = POLLNVAL;
1132 } else if (fds->fd < 0) {
1135 fp = fdp->fd_ofiles[fds->fd];
1137 fds->revents = POLLNVAL;
1141 * Note: backend also returns POLLHUP and
1142 * POLLERR if appropriate.
1144 selfdalloc(td, fds);
1145 fds->revents = fo_poll(fp, fds->events,
1147 if (fds->revents != 0)
1152 FILEDESC_SUNLOCK(fdp);
1153 td->td_retval[0] = n;
1158 * OpenBSD poll system call.
1160 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1162 #ifndef _SYS_SYSPROTO_H_
1163 struct openbsd_poll_args {
1170 openbsd_poll(td, uap)
1171 register struct thread *td;
1172 register struct openbsd_poll_args *uap;
1174 return (poll(td, (struct poll_args *)uap));
1178 * XXX This was created specifically to support netncp and netsmb. This
1179 * allows the caller to specify a socket to wait for events on. It returns
1180 * 0 if any events matched and an error otherwise. There is no way to
1181 * determine which events fired.
1184 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1186 struct timeval atv, rtv, ttv;
1191 if (itimerfix(&atv))
1193 getmicrouptime(&rtv);
1194 timevaladd(&atv, &rtv);
1203 * Iterate until the timeout expires or the socket becomes ready.
1206 selfdalloc(td, NULL);
1207 error = sopoll(so, events, NULL, td);
1208 /* error here is actually the ready events. */
1211 if (atv.tv_sec || atv.tv_usec) {
1212 getmicrouptime(&rtv);
1213 if (timevalcmp(&rtv, &atv, >=)) {
1215 return (EWOULDBLOCK);
1218 timevalsub(&ttv, &rtv);
1219 timo = ttv.tv_sec > 24 * 60 * 60 ?
1220 24 * 60 * 60 * hz : tvtohz(&ttv);
1222 error = seltdwait(td, timo);
1227 /* XXX Duplicates ncp/smb behavior. */
1228 if (error == ERESTART)
1234 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1235 * have two select sets, one for read and another for write.
1238 selfdalloc(struct thread *td, void *cookie)
1243 if (stp->st_free1 == NULL)
1244 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1245 stp->st_free1->sf_td = stp;
1246 stp->st_free1->sf_cookie = cookie;
1247 if (stp->st_free2 == NULL)
1248 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1249 stp->st_free2->sf_td = stp;
1250 stp->st_free2->sf_cookie = cookie;
1254 selfdfree(struct seltd *stp, struct selfd *sfp)
1256 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1257 mtx_lock(sfp->sf_mtx);
1259 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1260 mtx_unlock(sfp->sf_mtx);
1261 uma_zfree(selfd_zone, sfp);
1265 * Record a select request.
1268 selrecord(selector, sip)
1269 struct thread *selector;
1270 struct selinfo *sip;
1276 stp = selector->td_sel;
1278 * Don't record when doing a rescan.
1280 if (stp->st_flags & SELTD_RESCAN)
1283 * Grab one of the preallocated descriptors.
1286 if ((sfp = stp->st_free1) != NULL)
1287 stp->st_free1 = NULL;
1288 else if ((sfp = stp->st_free2) != NULL)
1289 stp->st_free2 = NULL;
1291 panic("selrecord: No free selfd on selq");
1292 mtxp = mtx_pool_find(mtxpool_sleep, sip);
1294 * Initialize the sfp and queue it in the thread.
1298 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1300 * Now that we've locked the sip, check for initialization.
1303 if (sip->si_mtx == NULL) {
1305 TAILQ_INIT(&sip->si_tdlist);
1308 * Add this thread to the list of selfds listening on this selinfo.
1310 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1311 mtx_unlock(sip->si_mtx);
1314 /* Wake up a selecting thread. */
1317 struct selinfo *sip;
1319 doselwakeup(sip, -1);
1322 /* Wake up a selecting thread, and set its priority. */
1324 selwakeuppri(sip, pri)
1325 struct selinfo *sip;
1328 doselwakeup(sip, pri);
1332 * Do a wakeup when a selectable event occurs.
1335 doselwakeup(sip, pri)
1336 struct selinfo *sip;
1343 /* If it's not initialized there can't be any waiters. */
1344 if (sip->si_mtx == NULL)
1347 * Locking the selinfo locks all selfds associated with it.
1349 mtx_lock(sip->si_mtx);
1350 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1352 * Once we remove this sfp from the list and clear the
1353 * sf_si seltdclear will know to ignore this si.
1355 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1358 mtx_lock(&stp->st_mtx);
1359 stp->st_flags |= SELTD_PENDING;
1360 cv_broadcastpri(&stp->st_wait, pri);
1361 mtx_unlock(&stp->st_mtx);
1363 mtx_unlock(sip->si_mtx);
1367 seltdinit(struct thread *td)
1371 if ((stp = td->td_sel) != NULL)
1373 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1374 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1375 cv_init(&stp->st_wait, "select");
1378 STAILQ_INIT(&stp->st_selq);
1382 seltdwait(struct thread *td, int timo)
1389 * An event of interest may occur while we do not hold the seltd
1390 * locked so check the pending flag before we sleep.
1392 mtx_lock(&stp->st_mtx);
1394 * Any further calls to selrecord will be a rescan.
1396 stp->st_flags |= SELTD_RESCAN;
1397 if (stp->st_flags & SELTD_PENDING) {
1398 mtx_unlock(&stp->st_mtx);
1402 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1404 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1405 mtx_unlock(&stp->st_mtx);
1411 seltdfini(struct thread *td)
1419 uma_zfree(selfd_zone, stp->st_free1);
1421 uma_zfree(selfd_zone, stp->st_free2);
1423 free(stp, M_SELECT);
1427 * Remove the references to the thread from all of the objects we were
1431 seltdclear(struct thread *td)
1438 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1439 selfdfree(stp, sfp);
1443 static void selectinit(void *);
1444 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1446 selectinit(void *dummy __unused)
1448 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1449 NULL, NULL, UMA_ALIGN_PTR, 0);