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_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/capsicum.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/fcntl.h>
54 #include <sys/signalvar.h>
55 #include <sys/socketvar.h>
57 #include <sys/kernel.h>
59 #include <sys/limits.h>
60 #include <sys/malloc.h>
62 #include <sys/resourcevar.h>
63 #include <sys/selinfo.h>
64 #include <sys/sleepqueue.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
71 #include <sys/condvar.h>
73 #include <sys/ktrace.h>
76 #include <security/audit/audit.h>
79 * The following macro defines how many bytes will be allocated from
80 * the stack instead of memory allocated when passing the IOCTL data
81 * structures from userspace and to the kernel. Some IOCTLs having
82 * small data structures are used very frequently and this small
83 * buffer on the stack gives a significant speedup improvement for
84 * those requests. The value of this define should be greater or equal
85 * to 64 bytes and should also be power of two. The data structure is
86 * currently hard-aligned to a 8-byte boundary on the stack. This
87 * should currently be sufficient for all supported platforms.
89 #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */
90 #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */
92 int iosize_max_clamp = 1;
93 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
94 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
95 int devfs_iosize_max_clamp = 1;
96 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
97 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
100 * Assert that the return value of read(2) and write(2) syscalls fits
101 * into a register. If not, an architecture will need to provide the
102 * usermode wrappers to reconstruct the result.
104 CTASSERT(sizeof(register_t) >= sizeof(size_t));
106 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
107 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
108 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
110 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
112 static int pollscan(struct thread *, struct pollfd *, u_int);
113 static int pollrescan(struct thread *);
114 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
115 static int selrescan(struct thread *, fd_mask **, fd_mask **);
116 static void selfdalloc(struct thread *, void *);
117 static void selfdfree(struct seltd *, struct selfd *);
118 static int dofileread(struct thread *, int, struct file *, struct uio *,
120 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
122 static void doselwakeup(struct selinfo *, int);
123 static void seltdinit(struct thread *);
124 static int seltdwait(struct thread *, sbintime_t, sbintime_t);
125 static void seltdclear(struct thread *);
128 * One seltd per-thread allocated on demand as needed.
130 * t - protected by st_mtx
131 * k - Only accessed by curthread or read-only
134 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
135 struct selfd *st_free1; /* (k) free fd for read set. */
136 struct selfd *st_free2; /* (k) free fd for write set. */
137 struct mtx st_mtx; /* Protects struct seltd */
138 struct cv st_wait; /* (t) Wait channel. */
139 int st_flags; /* (t) SELTD_ flags. */
142 #define SELTD_PENDING 0x0001 /* We have pending events. */
143 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
146 * One selfd allocated per-thread per-file-descriptor.
147 * f - protected by sf_mtx
150 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
151 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
152 struct selinfo *sf_si; /* (f) selinfo when linked. */
153 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
154 struct seltd *sf_td; /* (k) owning seltd. */
155 void *sf_cookie; /* (k) fd or pollfd. */
158 static uma_zone_t selfd_zone;
159 static struct mtx_pool *mtxpool_select;
161 #ifndef _SYS_SYSPROTO_H_
171 struct read_args *uap;
177 if (uap->nbyte > IOSIZE_MAX)
179 aiov.iov_base = uap->buf;
180 aiov.iov_len = uap->nbyte;
181 auio.uio_iov = &aiov;
183 auio.uio_resid = uap->nbyte;
184 auio.uio_segflg = UIO_USERSPACE;
185 error = kern_readv(td, uap->fd, &auio);
190 * Positioned read system call
192 #ifndef _SYS_SYSPROTO_H_
204 struct pread_args *uap;
210 if (uap->nbyte > IOSIZE_MAX)
212 aiov.iov_base = uap->buf;
213 aiov.iov_len = uap->nbyte;
214 auio.uio_iov = &aiov;
216 auio.uio_resid = uap->nbyte;
217 auio.uio_segflg = UIO_USERSPACE;
218 error = kern_preadv(td, uap->fd, &auio, uap->offset);
223 freebsd6_pread(td, uap)
225 struct freebsd6_pread_args *uap;
227 struct pread_args oargs;
230 oargs.buf = uap->buf;
231 oargs.nbyte = uap->nbyte;
232 oargs.offset = uap->offset;
233 return (sys_pread(td, &oargs));
237 * Scatter read system call.
239 #ifndef _SYS_SYSPROTO_H_
247 sys_readv(struct thread *td, struct readv_args *uap)
252 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
255 error = kern_readv(td, uap->fd, auio);
261 kern_readv(struct thread *td, int fd, struct uio *auio)
267 error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp);
270 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
276 * Scatter positioned read system call.
278 #ifndef _SYS_SYSPROTO_H_
287 sys_preadv(struct thread *td, struct preadv_args *uap)
292 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
295 error = kern_preadv(td, uap->fd, auio, uap->offset);
301 kern_preadv(td, fd, auio, offset)
311 error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp);
314 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
316 else if (offset < 0 &&
317 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
320 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
326 * Common code for readv and preadv that reads data in
327 * from a file using the passed in uio, offset, and flags.
330 dofileread(td, fd, fp, auio, offset, flags)
341 struct uio *ktruio = NULL;
344 /* Finish zero length reads right here */
345 if (auio->uio_resid == 0) {
346 td->td_retval[0] = 0;
349 auio->uio_rw = UIO_READ;
350 auio->uio_offset = offset;
353 if (KTRPOINT(td, KTR_GENIO))
354 ktruio = cloneuio(auio);
356 cnt = auio->uio_resid;
357 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
358 if (auio->uio_resid != cnt && (error == ERESTART ||
359 error == EINTR || error == EWOULDBLOCK))
362 cnt -= auio->uio_resid;
364 if (ktruio != NULL) {
365 ktruio->uio_resid = cnt;
366 ktrgenio(fd, UIO_READ, ktruio, error);
369 td->td_retval[0] = cnt;
373 #ifndef _SYS_SYSPROTO_H_
383 struct write_args *uap;
389 if (uap->nbyte > IOSIZE_MAX)
391 aiov.iov_base = (void *)(uintptr_t)uap->buf;
392 aiov.iov_len = uap->nbyte;
393 auio.uio_iov = &aiov;
395 auio.uio_resid = uap->nbyte;
396 auio.uio_segflg = UIO_USERSPACE;
397 error = kern_writev(td, uap->fd, &auio);
402 * Positioned write system call.
404 #ifndef _SYS_SYSPROTO_H_
416 struct pwrite_args *uap;
422 if (uap->nbyte > IOSIZE_MAX)
424 aiov.iov_base = (void *)(uintptr_t)uap->buf;
425 aiov.iov_len = uap->nbyte;
426 auio.uio_iov = &aiov;
428 auio.uio_resid = uap->nbyte;
429 auio.uio_segflg = UIO_USERSPACE;
430 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
435 freebsd6_pwrite(td, uap)
437 struct freebsd6_pwrite_args *uap;
439 struct pwrite_args oargs;
442 oargs.buf = uap->buf;
443 oargs.nbyte = uap->nbyte;
444 oargs.offset = uap->offset;
445 return (sys_pwrite(td, &oargs));
449 * Gather write system call.
451 #ifndef _SYS_SYSPROTO_H_
459 sys_writev(struct thread *td, struct writev_args *uap)
464 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
467 error = kern_writev(td, uap->fd, auio);
473 kern_writev(struct thread *td, int fd, struct uio *auio)
479 error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp);
482 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
488 * Gather positioned write system call.
490 #ifndef _SYS_SYSPROTO_H_
491 struct pwritev_args {
499 sys_pwritev(struct thread *td, struct pwritev_args *uap)
504 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
507 error = kern_pwritev(td, uap->fd, auio, uap->offset);
513 kern_pwritev(td, fd, auio, offset)
523 error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp);
526 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
528 else if (offset < 0 &&
529 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
532 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
538 * Common code for writev and pwritev that writes data to
539 * a file using the passed in uio, offset, and flags.
542 dofilewrite(td, fd, fp, auio, offset, flags)
553 struct uio *ktruio = NULL;
556 auio->uio_rw = UIO_WRITE;
558 auio->uio_offset = offset;
560 if (KTRPOINT(td, KTR_GENIO))
561 ktruio = cloneuio(auio);
563 cnt = auio->uio_resid;
564 if (fp->f_type == DTYPE_VNODE &&
565 (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
567 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
568 if (auio->uio_resid != cnt && (error == ERESTART ||
569 error == EINTR || error == EWOULDBLOCK))
571 /* Socket layer is responsible for issuing SIGPIPE. */
572 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
573 PROC_LOCK(td->td_proc);
574 tdsignal(td, SIGPIPE);
575 PROC_UNLOCK(td->td_proc);
578 cnt -= auio->uio_resid;
580 if (ktruio != NULL) {
581 ktruio->uio_resid = cnt;
582 ktrgenio(fd, UIO_WRITE, ktruio, error);
585 td->td_retval[0] = cnt;
590 * Truncate a file given a file descriptor.
592 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
593 * descriptor isn't writable.
596 kern_ftruncate(td, fd, length)
608 error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp);
611 AUDIT_ARG_FILE(td->td_proc, fp);
612 if (!(fp->f_flag & FWRITE)) {
616 error = fo_truncate(fp, length, td->td_ucred, td);
621 #ifndef _SYS_SYSPROTO_H_
622 struct ftruncate_args {
629 sys_ftruncate(td, uap)
631 struct ftruncate_args *uap;
634 return (kern_ftruncate(td, uap->fd, uap->length));
637 #if defined(COMPAT_43)
638 #ifndef _SYS_SYSPROTO_H_
639 struct oftruncate_args {
647 struct oftruncate_args *uap;
650 return (kern_ftruncate(td, uap->fd, uap->length));
652 #endif /* COMPAT_43 */
654 #ifndef _SYS_SYSPROTO_H_
663 sys_ioctl(struct thread *td, struct ioctl_args *uap)
665 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
672 if (uap->com > 0xffffffff) {
674 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
675 td->td_proc->p_pid, td->td_name, uap->com);
678 com = (uint32_t)uap->com;
681 * Interpret high order word to find amount of data to be
682 * copied to/from the user's address space.
684 size = IOCPARM_LEN(com);
685 if ((size > IOCPARM_MAX) ||
686 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
687 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
688 ((com & IOC_OUT) && size == 0) ||
690 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
692 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
696 if (com & IOC_VOID) {
697 /* Integer argument. */
698 arg = (intptr_t)uap->data;
702 if (size > SYS_IOCTL_SMALL_SIZE)
703 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
708 data = (void *)&uap->data;
710 error = copyin(uap->data, data, (u_int)size);
713 } else if (com & IOC_OUT) {
715 * Zero the buffer so the user always
716 * gets back something deterministic.
721 error = kern_ioctl(td, uap->fd, com, data);
723 if (error == 0 && (com & IOC_OUT))
724 error = copyout(data, uap->data, (u_int)size);
727 if (size > SYS_IOCTL_SMALL_SIZE)
728 free(data, M_IOCTLOPS);
733 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
736 struct filedesc *fdp;
740 int error, tmp, locked;
745 fdp = td->td_proc->p_fd;
758 locked = LA_UNLOCKED;
764 if ((fp = fget_locked(fdp, fd)) == NULL) {
768 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
769 fp = NULL; /* fhold() was not called yet */
773 if (locked == LA_SLOCKED) {
774 FILEDESC_SUNLOCK(fdp);
775 locked = LA_UNLOCKED;
778 error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp);
784 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
791 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
794 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
797 if ((tmp = *(int *)data))
798 atomic_set_int(&fp->f_flag, FNONBLOCK);
800 atomic_clear_int(&fp->f_flag, FNONBLOCK);
804 if ((tmp = *(int *)data))
805 atomic_set_int(&fp->f_flag, FASYNC);
807 atomic_clear_int(&fp->f_flag, FASYNC);
812 error = fo_ioctl(fp, com, data, td->td_ucred, td);
816 FILEDESC_XUNLOCK(fdp);
820 FILEDESC_SUNLOCK(fdp);
824 FILEDESC_UNLOCK_ASSERT(fdp);
833 poll_no_poll(int events)
836 * Return true for read/write. If the user asked for something
837 * special, return POLLNVAL, so that clients have a way of
838 * determining reliably whether or not the extended
839 * functionality is present without hard-coding knowledge
840 * of specific filesystem implementations.
842 if (events & ~POLLSTANDARD)
845 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
849 sys_pselect(struct thread *td, struct pselect_args *uap)
852 struct timeval tv, *tvp;
856 if (uap->ts != NULL) {
857 error = copyin(uap->ts, &ts, sizeof(ts));
860 TIMESPEC_TO_TIMEVAL(&tv, &ts);
864 if (uap->sm != NULL) {
865 error = copyin(uap->sm, &set, sizeof(set));
871 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
876 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
877 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
882 error = kern_sigprocmask(td, SIG_SETMASK, uset,
883 &td->td_oldsigmask, 0);
886 td->td_pflags |= TDP_OLDMASK;
888 * Make sure that ast() is called on return to
889 * usermode and TDP_OLDMASK is cleared, restoring old
893 td->td_flags |= TDF_ASTPENDING;
896 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
900 #ifndef _SYS_SYSPROTO_H_
903 fd_set *in, *ou, *ex;
908 sys_select(struct thread *td, struct select_args *uap)
910 struct timeval tv, *tvp;
913 if (uap->tv != NULL) {
914 error = copyin(uap->tv, &tv, sizeof(tv));
921 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
926 * In the unlikely case when user specified n greater then the last
927 * open file descriptor, check that no bits are set after the last
928 * valid fd. We must return EBADF if any is set.
930 * There are applications that rely on the behaviour.
932 * nd is fd_lastfile + 1.
935 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
941 if (nd >= ndu || fd_in == NULL)
945 bits = 0; /* silence gcc */
946 for (i = nd; i < ndu; i++) {
948 #if BYTE_ORDER == LITTLE_ENDIAN
949 addr = (char *)fd_in + b;
951 addr = (char *)fd_in;
952 if (abi_nfdbits == NFDBITS) {
953 addr += rounddown(b, sizeof(fd_mask)) +
954 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
956 addr += rounddown(b, sizeof(uint32_t)) +
957 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
967 if ((bits & (1 << (i % NBBY))) != 0)
974 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
975 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
977 struct filedesc *fdp;
979 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
980 * infds with the new FD_SETSIZE of 1024, and more than enough for
981 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
984 fd_mask s_selbits[howmany(2048, NFDBITS)];
985 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
987 sbintime_t asbt, precision, rsbt;
988 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
993 fdp = td->td_proc->p_fd;
995 lf = fdp->fd_lastfile;
999 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1002 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1005 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1010 * Allocate just enough bits for the non-null fd_sets. Use the
1011 * preallocated auto buffer if possible.
1013 nfdbits = roundup(nd, NFDBITS);
1014 ncpbytes = nfdbits / NBBY;
1015 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1018 nbufbytes += 2 * ncpbytes;
1020 nbufbytes += 2 * ncpbytes;
1022 nbufbytes += 2 * ncpbytes;
1023 if (nbufbytes <= sizeof s_selbits)
1024 selbits = &s_selbits[0];
1026 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1029 * Assign pointers into the bit buffers and fetch the input bits.
1030 * Put the output buffers together so that they can be bzeroed
1034 #define getbits(name, x) \
1036 if (name == NULL) { \
1040 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
1042 sbp += ncpbytes / sizeof *sbp; \
1043 error = copyin(name, ibits[x], ncpubytes); \
1046 bzero((char *)ibits[x] + ncpubytes, \
1047 ncpbytes - ncpubytes); \
1055 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1057 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1058 * we are running under 32-bit emulation. This should be more
1061 #define swizzle_fdset(bits) \
1062 if (abi_nfdbits != NFDBITS && bits != NULL) { \
1064 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
1065 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
1068 #define swizzle_fdset(bits)
1071 /* Make sure the bit order makes it through an ABI transition */
1072 swizzle_fdset(ibits[0]);
1073 swizzle_fdset(ibits[1]);
1074 swizzle_fdset(ibits[2]);
1077 bzero(selbits, nbufbytes / 2);
1082 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1083 rtv.tv_usec >= 1000000) {
1087 if (!timevalisset(&rtv))
1089 else if (rtv.tv_sec <= INT32_MAX) {
1090 rsbt = tvtosbt(rtv);
1092 precision >>= tc_precexp;
1093 if (TIMESEL(&asbt, rsbt))
1094 asbt += tc_tick_sbt;
1095 if (asbt <= SBT_MAX - rsbt)
1104 /* Iterate until the timeout expires or descriptors become ready. */
1106 error = selscan(td, ibits, obits, nd);
1107 if (error || td->td_retval[0] != 0)
1109 error = seltdwait(td, asbt, precision);
1112 error = selrescan(td, ibits, obits);
1113 if (error || td->td_retval[0] != 0)
1119 /* select is not restarted after signals... */
1120 if (error == ERESTART)
1122 if (error == EWOULDBLOCK)
1125 /* swizzle bit order back, if necessary */
1126 swizzle_fdset(obits[0]);
1127 swizzle_fdset(obits[1]);
1128 swizzle_fdset(obits[2]);
1129 #undef swizzle_fdset
1131 #define putbits(name, x) \
1132 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1142 if (selbits != &s_selbits[0])
1143 free(selbits, M_SELECT);
1148 * Convert a select bit set to poll flags.
1150 * The backend always returns POLLHUP/POLLERR if appropriate and we
1151 * return this as a set bit in any set.
1153 static int select_flags[3] = {
1154 POLLRDNORM | POLLHUP | POLLERR,
1155 POLLWRNORM | POLLHUP | POLLERR,
1156 POLLRDBAND | POLLERR
1160 * Compute the fo_poll flags required for a fd given by the index and
1161 * bit position in the fd_mask array.
1164 selflags(fd_mask **ibits, int idx, fd_mask bit)
1170 for (msk = 0; msk < 3; msk++) {
1171 if (ibits[msk] == NULL)
1173 if ((ibits[msk][idx] & bit) == 0)
1175 flags |= select_flags[msk];
1181 * Set the appropriate output bits given a mask of fired events and the
1182 * input bits originally requested.
1185 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1191 for (msk = 0; msk < 3; msk++) {
1192 if ((events & select_flags[msk]) == 0)
1194 if (ibits[msk] == NULL)
1196 if ((ibits[msk][idx] & bit) == 0)
1199 * XXX Check for a duplicate set. This can occur because a
1200 * socket calls selrecord() twice for each poll() call
1201 * resulting in two selfds per real fd. selrescan() will
1202 * call selsetbits twice as a result.
1204 if ((obits[msk][idx] & bit) != 0)
1206 obits[msk][idx] |= bit;
1214 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1216 cap_rights_t rights;
1218 cap_rights_init(&rights, CAP_EVENT);
1220 return (fget_unlocked(fdp, fd, &rights, 0, fpp, NULL));
1224 * Traverse the list of fds attached to this thread's seltd and check for
1228 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1230 struct filedesc *fdp;
1240 fdp = td->td_proc->p_fd;
1243 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1244 fd = (int)(uintptr_t)sfp->sf_cookie;
1246 selfdfree(stp, sfp);
1247 /* If the selinfo wasn't cleared the event didn't fire. */
1250 error = getselfd_cap(fdp, fd, &fp);
1254 bit = (fd_mask)1 << (fd % NFDBITS);
1255 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1258 n += selsetbits(ibits, obits, idx, bit, ev);
1261 td->td_retval[0] = n;
1266 * Perform the initial filedescriptor scan and register ourselves with
1270 selscan(td, ibits, obits, nfd)
1272 fd_mask **ibits, **obits;
1275 struct filedesc *fdp;
1278 int ev, flags, end, fd;
1282 fdp = td->td_proc->p_fd;
1284 for (idx = 0, fd = 0; fd < nfd; idx++) {
1285 end = imin(fd + NFDBITS, nfd);
1286 for (bit = 1; fd < end; bit <<= 1, fd++) {
1287 /* Compute the list of events we're interested in. */
1288 flags = selflags(ibits, idx, bit);
1291 error = getselfd_cap(fdp, fd, &fp);
1294 selfdalloc(td, (void *)(uintptr_t)fd);
1295 ev = fo_poll(fp, flags, td->td_ucred, td);
1298 n += selsetbits(ibits, obits, idx, bit, ev);
1302 td->td_retval[0] = n;
1307 sys_poll(struct thread *td, struct poll_args *uap)
1309 struct timespec ts, *tsp;
1311 if (uap->timeout != INFTIM) {
1312 if (uap->timeout < 0)
1314 ts.tv_sec = uap->timeout / 1000;
1315 ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1320 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1324 kern_poll(struct thread *td, struct pollfd *fds, u_int nfds,
1325 struct timespec *tsp, sigset_t *uset)
1327 struct pollfd *bits;
1328 struct pollfd smallbits[32];
1329 sbintime_t sbt, precision, tmp;
1337 if (tsp->tv_sec < 0)
1339 if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1341 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1345 if (ts.tv_sec > INT32_MAX / 2) {
1346 over = ts.tv_sec - INT32_MAX / 2;
1352 precision >>= tc_precexp;
1353 if (TIMESEL(&sbt, tmp))
1360 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1362 ni = nfds * sizeof(struct pollfd);
1363 if (ni > sizeof(smallbits))
1364 bits = malloc(ni, M_TEMP, M_WAITOK);
1367 error = copyin(fds, bits, ni);
1372 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1373 &td->td_oldsigmask, 0);
1376 td->td_pflags |= TDP_OLDMASK;
1378 * Make sure that ast() is called on return to
1379 * usermode and TDP_OLDMASK is cleared, restoring old
1383 td->td_flags |= TDF_ASTPENDING;
1388 /* Iterate until the timeout expires or descriptors become ready. */
1390 error = pollscan(td, bits, nfds);
1391 if (error || td->td_retval[0] != 0)
1393 error = seltdwait(td, sbt, precision);
1396 error = pollrescan(td);
1397 if (error || td->td_retval[0] != 0)
1403 /* poll is not restarted after signals... */
1404 if (error == ERESTART)
1406 if (error == EWOULDBLOCK)
1409 error = pollout(td, bits, fds, nfds);
1414 if (ni > sizeof(smallbits))
1420 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1422 struct timespec ts, *tsp;
1426 if (uap->ts != NULL) {
1427 error = copyin(uap->ts, &ts, sizeof(ts));
1433 if (uap->set != NULL) {
1434 error = copyin(uap->set, &set, sizeof(set));
1441 * fds is still a pointer to user space. kern_poll() will
1442 * take care of copyin that array to the kernel space.
1445 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1449 pollrescan(struct thread *td)
1455 struct filedesc *fdp;
1459 cap_rights_t rights;
1464 fdp = td->td_proc->p_fd;
1466 FILEDESC_SLOCK(fdp);
1467 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1468 fd = (struct pollfd *)sfp->sf_cookie;
1470 selfdfree(stp, sfp);
1471 /* If the selinfo wasn't cleared the event didn't fire. */
1474 fp = fdp->fd_ofiles[fd->fd].fde_file;
1477 cap_check(cap_rights(fdp, fd->fd),
1478 cap_rights_init(&rights, CAP_EVENT)) != 0)
1483 fd->revents = POLLNVAL;
1489 * Note: backend also returns POLLHUP and
1490 * POLLERR if appropriate.
1492 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1493 if (fd->revents != 0)
1496 FILEDESC_SUNLOCK(fdp);
1498 td->td_retval[0] = n;
1504 pollout(td, fds, ufds, nfd)
1507 struct pollfd *ufds;
1514 for (i = 0; i < nfd; i++) {
1515 error = copyout(&fds->revents, &ufds->revents,
1516 sizeof(ufds->revents));
1519 if (fds->revents != 0)
1524 td->td_retval[0] = n;
1529 pollscan(td, fds, nfd)
1534 struct filedesc *fdp = td->td_proc->p_fd;
1537 cap_rights_t rights;
1541 FILEDESC_SLOCK(fdp);
1542 for (i = 0; i < nfd; i++, fds++) {
1543 if (fds->fd > fdp->fd_lastfile) {
1544 fds->revents = POLLNVAL;
1546 } else if (fds->fd < 0) {
1549 fp = fdp->fd_ofiles[fds->fd].fde_file;
1552 cap_check(cap_rights(fdp, fds->fd),
1553 cap_rights_init(&rights, CAP_EVENT)) != 0)
1558 fds->revents = POLLNVAL;
1562 * Note: backend also returns POLLHUP and
1563 * POLLERR if appropriate.
1565 selfdalloc(td, fds);
1566 fds->revents = fo_poll(fp, fds->events,
1569 * POSIX requires POLLOUT to be never
1570 * set simultaneously with POLLHUP.
1572 if ((fds->revents & POLLHUP) != 0)
1573 fds->revents &= ~POLLOUT;
1575 if (fds->revents != 0)
1580 FILEDESC_SUNLOCK(fdp);
1581 td->td_retval[0] = n;
1586 * OpenBSD poll system call.
1588 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct openbsd_poll_args {
1598 sys_openbsd_poll(td, uap)
1599 register struct thread *td;
1600 register struct openbsd_poll_args *uap;
1602 return (sys_poll(td, (struct poll_args *)uap));
1606 * XXX This was created specifically to support netncp and netsmb. This
1607 * allows the caller to specify a socket to wait for events on. It returns
1608 * 0 if any events matched and an error otherwise. There is no way to
1609 * determine which events fired.
1612 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1615 sbintime_t asbt, precision, rsbt;
1618 precision = 0; /* stupid gcc! */
1621 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1622 rtv.tv_usec >= 1000000)
1624 if (!timevalisset(&rtv))
1626 else if (rtv.tv_sec <= INT32_MAX) {
1627 rsbt = tvtosbt(rtv);
1629 precision >>= tc_precexp;
1630 if (TIMESEL(&asbt, rsbt))
1631 asbt += tc_tick_sbt;
1632 if (asbt <= SBT_MAX - rsbt)
1642 * Iterate until the timeout expires or the socket becomes ready.
1645 selfdalloc(td, NULL);
1646 error = sopoll(so, events, NULL, td);
1647 /* error here is actually the ready events. */
1650 error = seltdwait(td, asbt, precision);
1655 /* XXX Duplicates ncp/smb behavior. */
1656 if (error == ERESTART)
1662 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1663 * have two select sets, one for read and another for write.
1666 selfdalloc(struct thread *td, void *cookie)
1671 if (stp->st_free1 == NULL)
1672 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1673 stp->st_free1->sf_td = stp;
1674 stp->st_free1->sf_cookie = cookie;
1675 if (stp->st_free2 == NULL)
1676 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1677 stp->st_free2->sf_td = stp;
1678 stp->st_free2->sf_cookie = cookie;
1682 selfdfree(struct seltd *stp, struct selfd *sfp)
1684 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1685 mtx_lock(sfp->sf_mtx);
1687 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1688 mtx_unlock(sfp->sf_mtx);
1689 uma_zfree(selfd_zone, sfp);
1692 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1695 struct selinfo *sip;
1699 * This feature is already provided by doselwakeup(), thus it is
1700 * enough to go for it.
1701 * Eventually, the context, should take care to avoid races
1702 * between thread calling select()/poll() and file descriptor
1703 * detaching, but, again, the races are just the same as
1706 doselwakeup(sip, -1);
1710 * Record a select request.
1713 selrecord(selector, sip)
1714 struct thread *selector;
1715 struct selinfo *sip;
1721 stp = selector->td_sel;
1723 * Don't record when doing a rescan.
1725 if (stp->st_flags & SELTD_RESCAN)
1728 * Grab one of the preallocated descriptors.
1731 if ((sfp = stp->st_free1) != NULL)
1732 stp->st_free1 = NULL;
1733 else if ((sfp = stp->st_free2) != NULL)
1734 stp->st_free2 = NULL;
1736 panic("selrecord: No free selfd on selq");
1739 mtxp = mtx_pool_find(mtxpool_select, sip);
1741 * Initialize the sfp and queue it in the thread.
1745 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1747 * Now that we've locked the sip, check for initialization.
1750 if (sip->si_mtx == NULL) {
1752 TAILQ_INIT(&sip->si_tdlist);
1755 * Add this thread to the list of selfds listening on this selinfo.
1757 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1758 mtx_unlock(sip->si_mtx);
1761 /* Wake up a selecting thread. */
1764 struct selinfo *sip;
1766 doselwakeup(sip, -1);
1769 /* Wake up a selecting thread, and set its priority. */
1771 selwakeuppri(sip, pri)
1772 struct selinfo *sip;
1775 doselwakeup(sip, pri);
1779 * Do a wakeup when a selectable event occurs.
1782 doselwakeup(sip, pri)
1783 struct selinfo *sip;
1790 /* If it's not initialized there can't be any waiters. */
1791 if (sip->si_mtx == NULL)
1794 * Locking the selinfo locks all selfds associated with it.
1796 mtx_lock(sip->si_mtx);
1797 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1799 * Once we remove this sfp from the list and clear the
1800 * sf_si seltdclear will know to ignore this si.
1802 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1805 mtx_lock(&stp->st_mtx);
1806 stp->st_flags |= SELTD_PENDING;
1807 cv_broadcastpri(&stp->st_wait, pri);
1808 mtx_unlock(&stp->st_mtx);
1810 mtx_unlock(sip->si_mtx);
1814 seltdinit(struct thread *td)
1818 if ((stp = td->td_sel) != NULL)
1820 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1821 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1822 cv_init(&stp->st_wait, "select");
1825 STAILQ_INIT(&stp->st_selq);
1829 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1836 * An event of interest may occur while we do not hold the seltd
1837 * locked so check the pending flag before we sleep.
1839 mtx_lock(&stp->st_mtx);
1841 * Any further calls to selrecord will be a rescan.
1843 stp->st_flags |= SELTD_RESCAN;
1844 if (stp->st_flags & SELTD_PENDING) {
1845 mtx_unlock(&stp->st_mtx);
1849 error = EWOULDBLOCK;
1851 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1852 sbt, precision, C_ABSOLUTE);
1854 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1855 mtx_unlock(&stp->st_mtx);
1861 seltdfini(struct thread *td)
1869 uma_zfree(selfd_zone, stp->st_free1);
1871 uma_zfree(selfd_zone, stp->st_free2);
1873 free(stp, M_SELECT);
1877 * Remove the references to the thread from all of the objects we were
1881 seltdclear(struct thread *td)
1888 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1889 selfdfree(stp, sfp);
1893 static void selectinit(void *);
1894 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1896 selectinit(void *dummy __unused)
1899 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1900 NULL, NULL, UMA_ALIGN_PTR, 0);
1901 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);