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 pollout(struct pollfd *, struct pollfd *, u_int);
80 static int pollscan(struct thread *, struct pollfd *, u_int);
81 static int pollrescan(struct thread *);
82 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
83 static int selrescan(struct thread *, fd_mask **, fd_mask **);
84 static void selfdalloc(struct thread *, void *);
85 static void selfdfree(struct seltd *, struct selfd *);
86 static int dofileread(struct thread *, int, struct file *, struct uio *,
88 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
90 static void doselwakeup(struct selinfo *, int);
91 static void seltdinit(struct thread *);
92 static int seltdwait(struct thread *, int);
93 static void seltdclear(struct thread *);
96 * One seltd per-thread allocated on demand as needed.
98 * t - protected by st_mtx
99 * k - Only accessed by curthread or read-only
102 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
103 struct selfd *st_free1; /* (k) free fd for read set. */
104 struct selfd *st_free2; /* (k) free fd for write set. */
105 struct mtx st_mtx; /* Protects struct seltd */
106 struct cv st_wait; /* (t) Wait channel. */
107 int st_flags; /* (t) SELTD_ flags. */
110 #define SELTD_PENDING 0x0001 /* We have pending events. */
111 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
114 * One selfd allocated per-thread per-file-descriptor.
115 * f - protected by sf_mtx
118 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
119 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
120 struct selinfo *sf_si; /* (f) selinfo when linked. */
121 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
122 struct seltd *sf_td; /* (k) owning seltd. */
123 void *sf_cookie; /* (k) fd or pollfd. */
126 static uma_zone_t selfd_zone;
127 static struct mtx_pool *mtxpool_select;
129 #ifndef _SYS_SYSPROTO_H_
139 struct read_args *uap;
145 if (uap->nbyte > INT_MAX)
147 aiov.iov_base = uap->buf;
148 aiov.iov_len = uap->nbyte;
149 auio.uio_iov = &aiov;
151 auio.uio_resid = uap->nbyte;
152 auio.uio_segflg = UIO_USERSPACE;
153 error = kern_readv(td, uap->fd, &auio);
158 * Positioned read system call
160 #ifndef _SYS_SYSPROTO_H_
172 struct pread_args *uap;
178 if (uap->nbyte > INT_MAX)
180 aiov.iov_base = uap->buf;
181 aiov.iov_len = uap->nbyte;
182 auio.uio_iov = &aiov;
184 auio.uio_resid = uap->nbyte;
185 auio.uio_segflg = UIO_USERSPACE;
186 error = kern_preadv(td, uap->fd, &auio, uap->offset);
191 freebsd6_pread(td, uap)
193 struct freebsd6_pread_args *uap;
195 struct pread_args oargs;
198 oargs.buf = uap->buf;
199 oargs.nbyte = uap->nbyte;
200 oargs.offset = uap->offset;
201 return (pread(td, &oargs));
205 * Scatter read system call.
207 #ifndef _SYS_SYSPROTO_H_
215 readv(struct thread *td, struct readv_args *uap)
220 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
223 error = kern_readv(td, uap->fd, auio);
229 kern_readv(struct thread *td, int fd, struct uio *auio)
234 error = fget_read(td, fd, &fp);
237 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
243 * Scatter positioned read system call.
245 #ifndef _SYS_SYSPROTO_H_
254 preadv(struct thread *td, struct preadv_args *uap)
259 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
262 error = kern_preadv(td, uap->fd, auio, uap->offset);
268 kern_preadv(td, fd, auio, offset)
277 error = fget_read(td, fd, &fp);
280 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
282 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
285 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
291 * Common code for readv and preadv that reads data in
292 * from a file using the passed in uio, offset, and flags.
295 dofileread(td, fd, fp, auio, offset, flags)
306 struct uio *ktruio = NULL;
309 /* Finish zero length reads right here */
310 if (auio->uio_resid == 0) {
311 td->td_retval[0] = 0;
314 auio->uio_rw = UIO_READ;
315 auio->uio_offset = offset;
318 if (KTRPOINT(td, KTR_GENIO))
319 ktruio = cloneuio(auio);
321 cnt = auio->uio_resid;
322 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
323 if (auio->uio_resid != cnt && (error == ERESTART ||
324 error == EINTR || error == EWOULDBLOCK))
327 cnt -= auio->uio_resid;
329 if (ktruio != NULL) {
330 ktruio->uio_resid = cnt;
331 ktrgenio(fd, UIO_READ, ktruio, error);
334 td->td_retval[0] = cnt;
338 #ifndef _SYS_SYSPROTO_H_
348 struct write_args *uap;
354 if (uap->nbyte > INT_MAX)
356 aiov.iov_base = (void *)(uintptr_t)uap->buf;
357 aiov.iov_len = uap->nbyte;
358 auio.uio_iov = &aiov;
360 auio.uio_resid = uap->nbyte;
361 auio.uio_segflg = UIO_USERSPACE;
362 error = kern_writev(td, uap->fd, &auio);
367 * Positioned write system call.
369 #ifndef _SYS_SYSPROTO_H_
381 struct pwrite_args *uap;
387 if (uap->nbyte > INT_MAX)
389 aiov.iov_base = (void *)(uintptr_t)uap->buf;
390 aiov.iov_len = uap->nbyte;
391 auio.uio_iov = &aiov;
393 auio.uio_resid = uap->nbyte;
394 auio.uio_segflg = UIO_USERSPACE;
395 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
400 freebsd6_pwrite(td, uap)
402 struct freebsd6_pwrite_args *uap;
404 struct pwrite_args oargs;
407 oargs.buf = uap->buf;
408 oargs.nbyte = uap->nbyte;
409 oargs.offset = uap->offset;
410 return (pwrite(td, &oargs));
414 * Gather write system call.
416 #ifndef _SYS_SYSPROTO_H_
424 writev(struct thread *td, struct writev_args *uap)
429 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
432 error = kern_writev(td, uap->fd, auio);
438 kern_writev(struct thread *td, int fd, struct uio *auio)
443 error = fget_write(td, fd, &fp);
446 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
452 * Gather positioned write system call.
454 #ifndef _SYS_SYSPROTO_H_
455 struct pwritev_args {
463 pwritev(struct thread *td, struct pwritev_args *uap)
468 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
471 error = kern_pwritev(td, uap->fd, auio, uap->offset);
477 kern_pwritev(td, fd, auio, offset)
486 error = fget_write(td, fd, &fp);
489 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
491 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
494 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
500 * Common code for writev and pwritev that writes data to
501 * a file using the passed in uio, offset, and flags.
504 dofilewrite(td, fd, fp, auio, offset, flags)
515 struct uio *ktruio = NULL;
518 auio->uio_rw = UIO_WRITE;
520 auio->uio_offset = offset;
522 if (KTRPOINT(td, KTR_GENIO))
523 ktruio = cloneuio(auio);
525 cnt = auio->uio_resid;
526 if (fp->f_type == DTYPE_VNODE)
528 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
529 if (auio->uio_resid != cnt && (error == ERESTART ||
530 error == EINTR || error == EWOULDBLOCK))
532 /* Socket layer is responsible for issuing SIGPIPE. */
533 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
534 PROC_LOCK(td->td_proc);
535 psignal(td->td_proc, SIGPIPE);
536 PROC_UNLOCK(td->td_proc);
539 cnt -= auio->uio_resid;
541 if (ktruio != NULL) {
542 ktruio->uio_resid = cnt;
543 ktrgenio(fd, UIO_WRITE, ktruio, error);
546 td->td_retval[0] = cnt;
551 * Truncate a file given a file descriptor.
553 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
554 * descriptor isn't writable.
557 kern_ftruncate(td, fd, length)
568 error = fget(td, fd, &fp);
571 AUDIT_ARG_FILE(td->td_proc, fp);
572 if (!(fp->f_flag & FWRITE)) {
576 error = fo_truncate(fp, length, td->td_ucred, td);
581 #ifndef _SYS_SYSPROTO_H_
582 struct ftruncate_args {
591 struct ftruncate_args *uap;
594 return (kern_ftruncate(td, uap->fd, uap->length));
597 #if defined(COMPAT_43)
598 #ifndef _SYS_SYSPROTO_H_
599 struct oftruncate_args {
607 struct oftruncate_args *uap;
610 return (kern_ftruncate(td, uap->fd, uap->length));
612 #endif /* COMPAT_43 */
614 #ifndef _SYS_SYSPROTO_H_
623 ioctl(struct thread *td, struct ioctl_args *uap)
630 if (uap->com > 0xffffffff) {
632 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
633 td->td_proc->p_pid, td->td_name, uap->com);
634 uap->com &= 0xffffffff;
639 * Interpret high order word to find amount of data to be
640 * copied to/from the user's address space.
642 size = IOCPARM_LEN(com);
643 if ((size > IOCPARM_MAX) ||
644 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
645 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
646 ((com & IOC_OUT) && size == 0) ||
648 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
650 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
654 if (com & IOC_VOID) {
655 /* Integer argument. */
656 arg = (intptr_t)uap->data;
660 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
662 data = (void *)&uap->data;
664 error = copyin(uap->data, data, (u_int)size);
667 free(data, M_IOCTLOPS);
670 } else if (com & IOC_OUT) {
672 * Zero the buffer so the user always
673 * gets back something deterministic.
678 error = kern_ioctl(td, uap->fd, com, data);
680 if (error == 0 && (com & IOC_OUT))
681 error = copyout(data, uap->data, (u_int)size);
684 free(data, M_IOCTLOPS);
689 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
692 struct filedesc *fdp;
698 if ((error = fget(td, fd, &fp)) != 0)
700 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
704 fdp = td->td_proc->p_fd;
708 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
709 FILEDESC_XUNLOCK(fdp);
713 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
714 FILEDESC_XUNLOCK(fdp);
717 if ((tmp = *(int *)data))
718 atomic_set_int(&fp->f_flag, FNONBLOCK);
720 atomic_clear_int(&fp->f_flag, FNONBLOCK);
724 if ((tmp = *(int *)data))
725 atomic_set_int(&fp->f_flag, FASYNC);
727 atomic_clear_int(&fp->f_flag, FASYNC);
732 error = fo_ioctl(fp, com, data, td->td_ucred, td);
739 poll_no_poll(int events)
742 * Return true for read/write. If the user asked for something
743 * special, return POLLNVAL, so that clients have a way of
744 * determining reliably whether or not the extended
745 * functionality is present without hard-coding knowledge
746 * of specific filesystem implementations.
748 if (events & ~POLLSTANDARD)
751 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
754 #ifndef _SYS_SYSPROTO_H_
757 fd_set *in, *ou, *ex;
763 register struct thread *td;
764 register struct select_args *uap;
766 struct timeval tv, *tvp;
769 if (uap->tv != NULL) {
770 error = copyin(uap->tv, &tv, sizeof(tv));
777 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp));
781 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
782 fd_set *fd_ex, struct timeval *tvp)
784 struct filedesc *fdp;
786 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
787 * infds with the new FD_SETSIZE of 1024, and more than enough for
788 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
791 fd_mask s_selbits[howmany(2048, NFDBITS)];
792 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
793 struct timeval atv, rtv, ttv;
795 u_int nbufbytes, ncpbytes, nfdbits;
799 fdp = td->td_proc->p_fd;
800 if (nd > fdp->fd_lastfile + 1)
801 nd = fdp->fd_lastfile + 1;
804 * Allocate just enough bits for the non-null fd_sets. Use the
805 * preallocated auto buffer if possible.
807 nfdbits = roundup(nd, NFDBITS);
808 ncpbytes = nfdbits / NBBY;
811 nbufbytes += 2 * ncpbytes;
813 nbufbytes += 2 * ncpbytes;
815 nbufbytes += 2 * ncpbytes;
816 if (nbufbytes <= sizeof s_selbits)
817 selbits = &s_selbits[0];
819 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
822 * Assign pointers into the bit buffers and fetch the input bits.
823 * Put the output buffers together so that they can be bzeroed
827 #define getbits(name, x) \
832 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
834 sbp += ncpbytes / sizeof *sbp; \
835 error = copyin(name, ibits[x], ncpbytes); \
845 bzero(selbits, nbufbytes / 2);
849 if (itimerfix(&atv)) {
853 getmicrouptime(&rtv);
854 timevaladd(&atv, &rtv);
861 /* Iterate until the timeout expires or descriptors become ready. */
863 error = selscan(td, ibits, obits, nd);
864 if (error || td->td_retval[0] != 0)
866 if (atv.tv_sec || atv.tv_usec) {
867 getmicrouptime(&rtv);
868 if (timevalcmp(&rtv, &atv, >=))
871 timevalsub(&ttv, &rtv);
872 timo = ttv.tv_sec > 24 * 60 * 60 ?
873 24 * 60 * 60 * hz : tvtohz(&ttv);
875 error = seltdwait(td, timo);
878 error = selrescan(td, ibits, obits);
879 if (error || td->td_retval[0] != 0)
885 /* select is not restarted after signals... */
886 if (error == ERESTART)
888 if (error == EWOULDBLOCK)
890 #define putbits(name, x) \
891 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
901 if (selbits != &s_selbits[0])
902 free(selbits, M_SELECT);
907 * Convert a select bit set to poll flags.
909 * The backend always returns POLLHUP/POLLERR if appropriate and we
910 * return this as a set bit in any set.
912 static int select_flags[3] = {
913 POLLRDNORM | POLLHUP | POLLERR,
914 POLLWRNORM | POLLHUP | POLLERR,
915 POLLRDBAND | POLLHUP | POLLERR
919 * Compute the fo_poll flags required for a fd given by the index and
920 * bit position in the fd_mask array.
923 selflags(fd_mask **ibits, int idx, fd_mask bit)
929 for (msk = 0; msk < 3; msk++) {
930 if (ibits[msk] == NULL)
932 if ((ibits[msk][idx] & bit) == 0)
934 flags |= select_flags[msk];
940 * Set the appropriate output bits given a mask of fired events and the
941 * input bits originally requested.
944 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
950 for (msk = 0; msk < 3; msk++) {
951 if ((events & select_flags[msk]) == 0)
953 if (ibits[msk] == NULL)
955 if ((ibits[msk][idx] & bit) == 0)
958 * XXX Check for a duplicate set. This can occur because a
959 * socket calls selrecord() twice for each poll() call
960 * resulting in two selfds per real fd. selrescan() will
961 * call selsetbits twice as a result.
963 if ((obits[msk][idx] & bit) != 0)
965 obits[msk][idx] |= bit;
973 * Traverse the list of fds attached to this thread's seltd and check for
977 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
979 struct filedesc *fdp;
988 fdp = td->td_proc->p_fd;
991 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
992 fd = (int)(uintptr_t)sfp->sf_cookie;
995 /* If the selinfo wasn't cleared the event didn't fire. */
998 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1001 bit = (fd_mask)1 << (fd % NFDBITS);
1002 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1005 n += selsetbits(ibits, obits, idx, bit, ev);
1008 td->td_retval[0] = n;
1013 * Perform the initial filedescriptor scan and register ourselves with
1017 selscan(td, ibits, obits, nfd)
1019 fd_mask **ibits, **obits;
1022 struct filedesc *fdp;
1025 int ev, flags, end, fd;
1028 fdp = td->td_proc->p_fd;
1030 for (idx = 0, fd = 0; fd < nfd; idx++) {
1031 end = imin(fd + NFDBITS, nfd);
1032 for (bit = 1; fd < end; bit <<= 1, fd++) {
1033 /* Compute the list of events we're interested in. */
1034 flags = selflags(ibits, idx, bit);
1037 if ((fp = fget_unlocked(fdp, fd)) == NULL)
1039 selfdalloc(td, (void *)(uintptr_t)fd);
1040 ev = fo_poll(fp, flags, td->td_ucred, td);
1043 n += selsetbits(ibits, obits, idx, bit, ev);
1047 td->td_retval[0] = n;
1051 #ifndef _SYS_SYSPROTO_H_
1061 struct poll_args *uap;
1063 struct pollfd *bits;
1064 struct pollfd smallbits[32];
1065 struct timeval atv, rtv, ttv;
1066 int error = 0, timo;
1071 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1073 ni = nfds * sizeof(struct pollfd);
1074 if (ni > sizeof(smallbits))
1075 bits = malloc(ni, M_TEMP, M_WAITOK);
1078 error = copyin(uap->fds, bits, ni);
1081 if (uap->timeout != INFTIM) {
1082 atv.tv_sec = uap->timeout / 1000;
1083 atv.tv_usec = (uap->timeout % 1000) * 1000;
1084 if (itimerfix(&atv)) {
1088 getmicrouptime(&rtv);
1089 timevaladd(&atv, &rtv);
1096 /* Iterate until the timeout expires or descriptors become ready. */
1098 error = pollscan(td, bits, nfds);
1099 if (error || td->td_retval[0] != 0)
1101 if (atv.tv_sec || atv.tv_usec) {
1102 getmicrouptime(&rtv);
1103 if (timevalcmp(&rtv, &atv, >=))
1106 timevalsub(&ttv, &rtv);
1107 timo = ttv.tv_sec > 24 * 60 * 60 ?
1108 24 * 60 * 60 * hz : tvtohz(&ttv);
1110 error = seltdwait(td, timo);
1113 error = pollrescan(td);
1114 if (error || td->td_retval[0] != 0)
1120 /* poll is not restarted after signals... */
1121 if (error == ERESTART)
1123 if (error == EWOULDBLOCK)
1126 error = pollout(bits, uap->fds, nfds);
1131 if (ni > sizeof(smallbits))
1137 pollrescan(struct thread *td)
1143 struct filedesc *fdp;
1149 fdp = td->td_proc->p_fd;
1151 FILEDESC_SLOCK(fdp);
1152 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1153 fd = (struct pollfd *)sfp->sf_cookie;
1155 selfdfree(stp, sfp);
1156 /* If the selinfo wasn't cleared the event didn't fire. */
1159 fp = fdp->fd_ofiles[fd->fd];
1161 fd->revents = POLLNVAL;
1166 * Note: backend also returns POLLHUP and
1167 * POLLERR if appropriate.
1169 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1170 if (fd->revents != 0)
1173 FILEDESC_SUNLOCK(fdp);
1175 td->td_retval[0] = n;
1181 pollout(fds, ufds, nfd)
1183 struct pollfd *ufds;
1189 for (i = 0; i < nfd; i++) {
1190 error = copyout(&fds->revents, &ufds->revents,
1191 sizeof(ufds->revents));
1201 pollscan(td, fds, nfd)
1206 struct filedesc *fdp = td->td_proc->p_fd;
1211 FILEDESC_SLOCK(fdp);
1212 for (i = 0; i < nfd; i++, fds++) {
1213 if (fds->fd >= fdp->fd_nfiles) {
1214 fds->revents = POLLNVAL;
1216 } else if (fds->fd < 0) {
1219 fp = fdp->fd_ofiles[fds->fd];
1221 fds->revents = POLLNVAL;
1225 * Note: backend also returns POLLHUP and
1226 * POLLERR if appropriate.
1228 selfdalloc(td, fds);
1229 fds->revents = fo_poll(fp, fds->events,
1231 if (fds->revents != 0)
1236 FILEDESC_SUNLOCK(fdp);
1237 td->td_retval[0] = n;
1242 * OpenBSD poll system call.
1244 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1246 #ifndef _SYS_SYSPROTO_H_
1247 struct openbsd_poll_args {
1254 openbsd_poll(td, uap)
1255 register struct thread *td;
1256 register struct openbsd_poll_args *uap;
1258 return (poll(td, (struct poll_args *)uap));
1262 * XXX This was created specifically to support netncp and netsmb. This
1263 * allows the caller to specify a socket to wait for events on. It returns
1264 * 0 if any events matched and an error otherwise. There is no way to
1265 * determine which events fired.
1268 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1270 struct timeval atv, rtv, ttv;
1275 if (itimerfix(&atv))
1277 getmicrouptime(&rtv);
1278 timevaladd(&atv, &rtv);
1287 * Iterate until the timeout expires or the socket becomes ready.
1290 selfdalloc(td, NULL);
1291 error = sopoll(so, events, NULL, td);
1292 /* error here is actually the ready events. */
1295 if (atv.tv_sec || atv.tv_usec) {
1296 getmicrouptime(&rtv);
1297 if (timevalcmp(&rtv, &atv, >=)) {
1299 return (EWOULDBLOCK);
1302 timevalsub(&ttv, &rtv);
1303 timo = ttv.tv_sec > 24 * 60 * 60 ?
1304 24 * 60 * 60 * hz : tvtohz(&ttv);
1306 error = seltdwait(td, timo);
1311 /* XXX Duplicates ncp/smb behavior. */
1312 if (error == ERESTART)
1318 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1319 * have two select sets, one for read and another for write.
1322 selfdalloc(struct thread *td, void *cookie)
1327 if (stp->st_free1 == NULL)
1328 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1329 stp->st_free1->sf_td = stp;
1330 stp->st_free1->sf_cookie = cookie;
1331 if (stp->st_free2 == NULL)
1332 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1333 stp->st_free2->sf_td = stp;
1334 stp->st_free2->sf_cookie = cookie;
1338 selfdfree(struct seltd *stp, struct selfd *sfp)
1340 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1341 mtx_lock(sfp->sf_mtx);
1343 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1344 mtx_unlock(sfp->sf_mtx);
1345 uma_zfree(selfd_zone, sfp);
1349 * Record a select request.
1352 selrecord(selector, sip)
1353 struct thread *selector;
1354 struct selinfo *sip;
1360 stp = selector->td_sel;
1362 * Don't record when doing a rescan.
1364 if (stp->st_flags & SELTD_RESCAN)
1367 * Grab one of the preallocated descriptors.
1370 if ((sfp = stp->st_free1) != NULL)
1371 stp->st_free1 = NULL;
1372 else if ((sfp = stp->st_free2) != NULL)
1373 stp->st_free2 = NULL;
1375 panic("selrecord: No free selfd on selq");
1378 mtxp = mtx_pool_find(mtxpool_select, sip);
1380 * Initialize the sfp and queue it in the thread.
1384 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1386 * Now that we've locked the sip, check for initialization.
1389 if (sip->si_mtx == NULL) {
1391 TAILQ_INIT(&sip->si_tdlist);
1394 * Add this thread to the list of selfds listening on this selinfo.
1396 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1397 mtx_unlock(sip->si_mtx);
1400 /* Wake up a selecting thread. */
1403 struct selinfo *sip;
1405 doselwakeup(sip, -1);
1408 /* Wake up a selecting thread, and set its priority. */
1410 selwakeuppri(sip, pri)
1411 struct selinfo *sip;
1414 doselwakeup(sip, pri);
1418 * Do a wakeup when a selectable event occurs.
1421 doselwakeup(sip, pri)
1422 struct selinfo *sip;
1429 /* If it's not initialized there can't be any waiters. */
1430 if (sip->si_mtx == NULL)
1433 * Locking the selinfo locks all selfds associated with it.
1435 mtx_lock(sip->si_mtx);
1436 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1438 * Once we remove this sfp from the list and clear the
1439 * sf_si seltdclear will know to ignore this si.
1441 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1444 mtx_lock(&stp->st_mtx);
1445 stp->st_flags |= SELTD_PENDING;
1446 cv_broadcastpri(&stp->st_wait, pri);
1447 mtx_unlock(&stp->st_mtx);
1449 mtx_unlock(sip->si_mtx);
1453 seltdinit(struct thread *td)
1457 if ((stp = td->td_sel) != NULL)
1459 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1460 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1461 cv_init(&stp->st_wait, "select");
1464 STAILQ_INIT(&stp->st_selq);
1468 seltdwait(struct thread *td, int timo)
1475 * An event of interest may occur while we do not hold the seltd
1476 * locked so check the pending flag before we sleep.
1478 mtx_lock(&stp->st_mtx);
1480 * Any further calls to selrecord will be a rescan.
1482 stp->st_flags |= SELTD_RESCAN;
1483 if (stp->st_flags & SELTD_PENDING) {
1484 mtx_unlock(&stp->st_mtx);
1488 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1490 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1491 mtx_unlock(&stp->st_mtx);
1497 seltdfini(struct thread *td)
1505 uma_zfree(selfd_zone, stp->st_free1);
1507 uma_zfree(selfd_zone, stp->st_free2);
1509 free(stp, M_SELECT);
1513 * Remove the references to the thread from all of the objects we were
1517 seltdclear(struct thread *td)
1524 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1525 selfdfree(stp, sfp);
1529 static void selectinit(void *);
1530 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1532 selectinit(void *dummy __unused)
1535 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1536 NULL, NULL, UMA_ALIGN_PTR, 0);
1537 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);