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 */
93 static int iosize_max_clamp = 0;
94 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
95 &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
96 static int devfs_iosize_max_clamp = 1;
97 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
98 &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
102 * Assert that the return value of read(2) and write(2) syscalls fits
103 * into a register. If not, an architecture will need to provide the
104 * usermode wrappers to reconstruct the result.
106 CTASSERT(sizeof(register_t) >= sizeof(size_t));
108 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
109 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
110 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
112 static int pollout(struct thread *, struct pollfd *, struct pollfd *,
114 static int pollscan(struct thread *, struct pollfd *, u_int);
115 static int pollrescan(struct thread *);
116 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
117 static int selrescan(struct thread *, fd_mask **, fd_mask **);
118 static void selfdalloc(struct thread *, void *);
119 static void selfdfree(struct seltd *, struct selfd *);
120 static int dofileread(struct thread *, int, struct file *, struct uio *,
122 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
124 static void doselwakeup(struct selinfo *, int);
125 static void seltdinit(struct thread *);
126 static int seltdwait(struct thread *, sbintime_t, sbintime_t);
127 static void seltdclear(struct thread *);
130 * One seltd per-thread allocated on demand as needed.
132 * t - protected by st_mtx
133 * k - Only accessed by curthread or read-only
136 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
137 struct selfd *st_free1; /* (k) free fd for read set. */
138 struct selfd *st_free2; /* (k) free fd for write set. */
139 struct mtx st_mtx; /* Protects struct seltd */
140 struct cv st_wait; /* (t) Wait channel. */
141 int st_flags; /* (t) SELTD_ flags. */
144 #define SELTD_PENDING 0x0001 /* We have pending events. */
145 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
148 * One selfd allocated per-thread per-file-descriptor.
149 * f - protected by sf_mtx
152 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
153 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
154 struct selinfo *sf_si; /* (f) selinfo when linked. */
155 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
156 struct seltd *sf_td; /* (k) owning seltd. */
157 void *sf_cookie; /* (k) fd or pollfd. */
161 static uma_zone_t selfd_zone;
162 static struct mtx_pool *mtxpool_select;
166 devfs_iosize_max(void)
169 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
170 INT_MAX : SSIZE_MAX);
177 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
178 INT_MAX : SSIZE_MAX);
182 #ifndef _SYS_SYSPROTO_H_
192 struct read_args *uap;
198 if (uap->nbyte > IOSIZE_MAX)
200 aiov.iov_base = uap->buf;
201 aiov.iov_len = uap->nbyte;
202 auio.uio_iov = &aiov;
204 auio.uio_resid = uap->nbyte;
205 auio.uio_segflg = UIO_USERSPACE;
206 error = kern_readv(td, uap->fd, &auio);
211 * Positioned read system call
213 #ifndef _SYS_SYSPROTO_H_
225 struct pread_args *uap;
231 if (uap->nbyte > IOSIZE_MAX)
233 aiov.iov_base = uap->buf;
234 aiov.iov_len = uap->nbyte;
235 auio.uio_iov = &aiov;
237 auio.uio_resid = uap->nbyte;
238 auio.uio_segflg = UIO_USERSPACE;
239 error = kern_preadv(td, uap->fd, &auio, uap->offset);
243 #if defined(COMPAT_FREEBSD6)
245 freebsd6_pread(td, uap)
247 struct freebsd6_pread_args *uap;
249 struct pread_args oargs;
252 oargs.buf = uap->buf;
253 oargs.nbyte = uap->nbyte;
254 oargs.offset = uap->offset;
255 return (sys_pread(td, &oargs));
260 * Scatter read system call.
262 #ifndef _SYS_SYSPROTO_H_
270 sys_readv(struct thread *td, struct readv_args *uap)
275 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
278 error = kern_readv(td, uap->fd, auio);
284 kern_readv(struct thread *td, int fd, struct uio *auio)
290 error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp);
293 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
299 * Scatter positioned read system call.
301 #ifndef _SYS_SYSPROTO_H_
310 sys_preadv(struct thread *td, struct preadv_args *uap)
315 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
318 error = kern_preadv(td, uap->fd, auio, uap->offset);
324 kern_preadv(td, fd, auio, offset)
334 error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp);
337 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
339 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
342 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
348 * Common code for readv and preadv that reads data in
349 * from a file using the passed in uio, offset, and flags.
352 dofileread(td, fd, fp, auio, offset, flags)
363 struct uio *ktruio = NULL;
368 /* Finish zero length reads right here */
369 if (auio->uio_resid == 0) {
370 td->td_retval[0] = 0;
373 auio->uio_rw = UIO_READ;
374 auio->uio_offset = offset;
377 if (KTRPOINT(td, KTR_GENIO))
378 ktruio = cloneuio(auio);
380 cnt = auio->uio_resid;
381 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
382 if (auio->uio_resid != cnt && (error == ERESTART ||
383 error == EINTR || error == EWOULDBLOCK))
386 cnt -= auio->uio_resid;
388 if (ktruio != NULL) {
389 ktruio->uio_resid = cnt;
390 ktrgenio(fd, UIO_READ, ktruio, error);
393 td->td_retval[0] = cnt;
397 #ifndef _SYS_SYSPROTO_H_
407 struct write_args *uap;
413 if (uap->nbyte > IOSIZE_MAX)
415 aiov.iov_base = (void *)(uintptr_t)uap->buf;
416 aiov.iov_len = uap->nbyte;
417 auio.uio_iov = &aiov;
419 auio.uio_resid = uap->nbyte;
420 auio.uio_segflg = UIO_USERSPACE;
421 error = kern_writev(td, uap->fd, &auio);
426 * Positioned write system call.
428 #ifndef _SYS_SYSPROTO_H_
440 struct pwrite_args *uap;
446 if (uap->nbyte > IOSIZE_MAX)
448 aiov.iov_base = (void *)(uintptr_t)uap->buf;
449 aiov.iov_len = uap->nbyte;
450 auio.uio_iov = &aiov;
452 auio.uio_resid = uap->nbyte;
453 auio.uio_segflg = UIO_USERSPACE;
454 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
458 #if defined(COMPAT_FREEBSD6)
460 freebsd6_pwrite(td, uap)
462 struct freebsd6_pwrite_args *uap;
464 struct pwrite_args oargs;
467 oargs.buf = uap->buf;
468 oargs.nbyte = uap->nbyte;
469 oargs.offset = uap->offset;
470 return (sys_pwrite(td, &oargs));
475 * Gather write system call.
477 #ifndef _SYS_SYSPROTO_H_
485 sys_writev(struct thread *td, struct writev_args *uap)
490 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
493 error = kern_writev(td, uap->fd, auio);
499 kern_writev(struct thread *td, int fd, struct uio *auio)
505 error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp);
508 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
514 * Gather positioned write system call.
516 #ifndef _SYS_SYSPROTO_H_
517 struct pwritev_args {
525 sys_pwritev(struct thread *td, struct pwritev_args *uap)
530 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
533 error = kern_pwritev(td, uap->fd, auio, uap->offset);
539 kern_pwritev(td, fd, auio, offset)
549 error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp);
552 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
554 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
557 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
563 * Common code for writev and pwritev that writes data to
564 * a file using the passed in uio, offset, and flags.
567 dofilewrite(td, fd, fp, auio, offset, flags)
578 struct uio *ktruio = NULL;
582 auio->uio_rw = UIO_WRITE;
584 auio->uio_offset = offset;
586 if (KTRPOINT(td, KTR_GENIO))
587 ktruio = cloneuio(auio);
589 cnt = auio->uio_resid;
590 if (fp->f_type == DTYPE_VNODE &&
591 (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
593 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
594 if (auio->uio_resid != cnt && (error == ERESTART ||
595 error == EINTR || error == EWOULDBLOCK))
597 /* Socket layer is responsible for issuing SIGPIPE. */
598 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
599 PROC_LOCK(td->td_proc);
600 tdsignal(td, SIGPIPE);
601 PROC_UNLOCK(td->td_proc);
604 cnt -= auio->uio_resid;
606 if (ktruio != NULL) {
607 ktruio->uio_resid = cnt;
608 ktrgenio(fd, UIO_WRITE, ktruio, error);
611 td->td_retval[0] = cnt;
616 * Truncate a file given a file descriptor.
618 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
619 * descriptor isn't writable.
622 kern_ftruncate(td, fd, length)
634 error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp);
637 AUDIT_ARG_FILE(td->td_proc, fp);
638 if (!(fp->f_flag & FWRITE)) {
642 error = fo_truncate(fp, length, td->td_ucred, td);
647 #ifndef _SYS_SYSPROTO_H_
648 struct ftruncate_args {
655 sys_ftruncate(td, uap)
657 struct ftruncate_args *uap;
660 return (kern_ftruncate(td, uap->fd, uap->length));
663 #if defined(COMPAT_43)
664 #ifndef _SYS_SYSPROTO_H_
665 struct oftruncate_args {
673 struct oftruncate_args *uap;
676 return (kern_ftruncate(td, uap->fd, uap->length));
678 #endif /* COMPAT_43 */
680 #ifndef _SYS_SYSPROTO_H_
689 sys_ioctl(struct thread *td, struct ioctl_args *uap)
691 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
697 if (uap->com > 0xffffffff) {
699 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
700 td->td_proc->p_pid, td->td_name, uap->com);
701 uap->com &= 0xffffffff;
706 * Interpret high order word to find amount of data to be
707 * copied to/from the user's address space.
709 size = IOCPARM_LEN(com);
710 if ((size > IOCPARM_MAX) ||
711 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
712 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
713 ((com & IOC_OUT) && size == 0) ||
715 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
717 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
721 if (com & IOC_VOID) {
722 /* Integer argument. */
723 arg = (intptr_t)uap->data;
727 if (size > SYS_IOCTL_SMALL_SIZE)
728 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
733 data = (void *)&uap->data;
735 error = copyin(uap->data, data, (u_int)size);
738 } else if (com & IOC_OUT) {
740 * Zero the buffer so the user always
741 * gets back something deterministic.
746 error = kern_ioctl(td, uap->fd, com, data);
748 if (error == 0 && (com & IOC_OUT))
749 error = copyout(data, uap->data, (u_int)size);
752 if (size > SYS_IOCTL_SMALL_SIZE)
753 free(data, M_IOCTLOPS);
758 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
761 struct filedesc *fdp;
765 int error, tmp, locked;
770 fdp = td->td_proc->p_fd;
783 locked = LA_UNLOCKED;
789 if ((fp = fget_locked(fdp, fd)) == NULL) {
793 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
794 fp = NULL; /* fhold() was not called yet */
798 if (locked == LA_SLOCKED) {
799 FILEDESC_SUNLOCK(fdp);
800 locked = LA_UNLOCKED;
803 error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp);
809 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
816 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
819 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
822 if ((tmp = *(int *)data))
823 atomic_set_int(&fp->f_flag, FNONBLOCK);
825 atomic_clear_int(&fp->f_flag, FNONBLOCK);
829 if ((tmp = *(int *)data))
830 atomic_set_int(&fp->f_flag, FASYNC);
832 atomic_clear_int(&fp->f_flag, FASYNC);
837 error = fo_ioctl(fp, com, data, td->td_ucred, td);
841 FILEDESC_XUNLOCK(fdp);
845 FILEDESC_SUNLOCK(fdp);
849 FILEDESC_UNLOCK_ASSERT(fdp);
858 poll_no_poll(int events)
861 * Return true for read/write. If the user asked for something
862 * special, return POLLNVAL, so that clients have a way of
863 * determining reliably whether or not the extended
864 * functionality is present without hard-coding knowledge
865 * of specific filesystem implementations.
867 if (events & ~POLLSTANDARD)
870 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
874 sys_pselect(struct thread *td, struct pselect_args *uap)
877 struct timeval tv, *tvp;
881 if (uap->ts != NULL) {
882 error = copyin(uap->ts, &ts, sizeof(ts));
885 TIMESPEC_TO_TIMEVAL(&tv, &ts);
889 if (uap->sm != NULL) {
890 error = copyin(uap->sm, &set, sizeof(set));
896 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
901 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
902 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
907 error = kern_sigprocmask(td, SIG_SETMASK, uset,
908 &td->td_oldsigmask, 0);
911 td->td_pflags |= TDP_OLDMASK;
913 * Make sure that ast() is called on return to
914 * usermode and TDP_OLDMASK is cleared, restoring old
918 td->td_flags |= TDF_ASTPENDING;
921 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
925 #ifndef _SYS_SYSPROTO_H_
928 fd_set *in, *ou, *ex;
933 sys_select(struct thread *td, struct select_args *uap)
935 struct timeval tv, *tvp;
938 if (uap->tv != NULL) {
939 error = copyin(uap->tv, &tv, sizeof(tv));
946 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
951 * In the unlikely case when user specified n greater then the last
952 * open file descriptor, check that no bits are set after the last
953 * valid fd. We must return EBADF if any is set.
955 * There are applications that rely on the behaviour.
957 * nd is fd_lastfile + 1.
960 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
966 if (nd >= ndu || fd_in == NULL)
970 bits = 0; /* silence gcc */
971 for (i = nd; i < ndu; i++) {
973 #if BYTE_ORDER == LITTLE_ENDIAN
974 addr = (char *)fd_in + b;
976 addr = (char *)fd_in;
977 if (abi_nfdbits == NFDBITS) {
978 addr += rounddown(b, sizeof(fd_mask)) +
979 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
981 addr += rounddown(b, sizeof(uint32_t)) +
982 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
992 if ((bits & (1 << (i % NBBY))) != 0)
999 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1000 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1002 struct filedesc *fdp;
1004 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1005 * infds with the new FD_SETSIZE of 1024, and more than enough for
1006 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1009 fd_mask s_selbits[howmany(2048, NFDBITS)];
1010 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1012 sbintime_t asbt, precision, rsbt;
1013 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1018 fdp = td->td_proc->p_fd;
1020 lf = fdp->fd_lastfile;
1024 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1027 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1030 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1035 * Allocate just enough bits for the non-null fd_sets. Use the
1036 * preallocated auto buffer if possible.
1038 nfdbits = roundup(nd, NFDBITS);
1039 ncpbytes = nfdbits / NBBY;
1040 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1043 nbufbytes += 2 * ncpbytes;
1045 nbufbytes += 2 * ncpbytes;
1047 nbufbytes += 2 * ncpbytes;
1048 if (nbufbytes <= sizeof s_selbits)
1049 selbits = &s_selbits[0];
1051 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1054 * Assign pointers into the bit buffers and fetch the input bits.
1055 * Put the output buffers together so that they can be bzeroed
1059 #define getbits(name, x) \
1061 if (name == NULL) { \
1065 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
1067 sbp += ncpbytes / sizeof *sbp; \
1068 error = copyin(name, ibits[x], ncpubytes); \
1071 bzero((char *)ibits[x] + ncpubytes, \
1072 ncpbytes - ncpubytes); \
1080 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1082 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1083 * we are running under 32-bit emulation. This should be more
1086 #define swizzle_fdset(bits) \
1087 if (abi_nfdbits != NFDBITS && bits != NULL) { \
1089 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
1090 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
1093 #define swizzle_fdset(bits)
1096 /* Make sure the bit order makes it through an ABI transition */
1097 swizzle_fdset(ibits[0]);
1098 swizzle_fdset(ibits[1]);
1099 swizzle_fdset(ibits[2]);
1102 bzero(selbits, nbufbytes / 2);
1107 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1108 rtv.tv_usec >= 1000000) {
1112 if (!timevalisset(&rtv))
1114 else if (rtv.tv_sec <= INT32_MAX) {
1115 rsbt = tvtosbt(rtv);
1117 precision >>= tc_precexp;
1118 if (TIMESEL(&asbt, rsbt))
1119 asbt += tc_tick_sbt;
1120 if (asbt <= SBT_MAX - rsbt)
1129 /* Iterate until the timeout expires or descriptors become ready. */
1131 error = selscan(td, ibits, obits, nd);
1132 if (error || td->td_retval[0] != 0)
1134 error = seltdwait(td, asbt, precision);
1137 error = selrescan(td, ibits, obits);
1138 if (error || td->td_retval[0] != 0)
1144 /* select is not restarted after signals... */
1145 if (error == ERESTART)
1147 if (error == EWOULDBLOCK)
1150 /* swizzle bit order back, if necessary */
1151 swizzle_fdset(obits[0]);
1152 swizzle_fdset(obits[1]);
1153 swizzle_fdset(obits[2]);
1154 #undef swizzle_fdset
1156 #define putbits(name, x) \
1157 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1167 if (selbits != &s_selbits[0])
1168 free(selbits, M_SELECT);
1173 * Convert a select bit set to poll flags.
1175 * The backend always returns POLLHUP/POLLERR if appropriate and we
1176 * return this as a set bit in any set.
1178 static int select_flags[3] = {
1179 POLLRDNORM | POLLHUP | POLLERR,
1180 POLLWRNORM | POLLHUP | POLLERR,
1181 POLLRDBAND | POLLERR
1185 * Compute the fo_poll flags required for a fd given by the index and
1186 * bit position in the fd_mask array.
1189 selflags(fd_mask **ibits, int idx, fd_mask bit)
1195 for (msk = 0; msk < 3; msk++) {
1196 if (ibits[msk] == NULL)
1198 if ((ibits[msk][idx] & bit) == 0)
1200 flags |= select_flags[msk];
1206 * Set the appropriate output bits given a mask of fired events and the
1207 * input bits originally requested.
1210 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1216 for (msk = 0; msk < 3; msk++) {
1217 if ((events & select_flags[msk]) == 0)
1219 if (ibits[msk] == NULL)
1221 if ((ibits[msk][idx] & bit) == 0)
1224 * XXX Check for a duplicate set. This can occur because a
1225 * socket calls selrecord() twice for each poll() call
1226 * resulting in two selfds per real fd. selrescan() will
1227 * call selsetbits twice as a result.
1229 if ((obits[msk][idx] & bit) != 0)
1231 obits[msk][idx] |= bit;
1239 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
1241 cap_rights_t rights;
1243 cap_rights_init(&rights, CAP_EVENT);
1245 return (fget_unlocked(fdp, fd, &rights, fpp, NULL));
1249 * Traverse the list of fds attached to this thread's seltd and check for
1253 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1255 struct filedesc *fdp;
1265 fdp = td->td_proc->p_fd;
1268 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1269 fd = (int)(uintptr_t)sfp->sf_cookie;
1271 selfdfree(stp, sfp);
1272 /* If the selinfo wasn't cleared the event didn't fire. */
1275 error = getselfd_cap(fdp, fd, &fp);
1279 bit = (fd_mask)1 << (fd % NFDBITS);
1280 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1283 n += selsetbits(ibits, obits, idx, bit, ev);
1286 td->td_retval[0] = n;
1291 * Perform the initial filedescriptor scan and register ourselves with
1295 selscan(td, ibits, obits, nfd)
1297 fd_mask **ibits, **obits;
1300 struct filedesc *fdp;
1303 int ev, flags, end, fd;
1307 fdp = td->td_proc->p_fd;
1309 for (idx = 0, fd = 0; fd < nfd; idx++) {
1310 end = imin(fd + NFDBITS, nfd);
1311 for (bit = 1; fd < end; bit <<= 1, fd++) {
1312 /* Compute the list of events we're interested in. */
1313 flags = selflags(ibits, idx, bit);
1316 error = getselfd_cap(fdp, fd, &fp);
1319 selfdalloc(td, (void *)(uintptr_t)fd);
1320 ev = fo_poll(fp, flags, td->td_ucred, td);
1323 n += selsetbits(ibits, obits, idx, bit, ev);
1327 td->td_retval[0] = n;
1332 sys_poll(struct thread *td, struct poll_args *uap)
1334 struct timespec ts, *tsp;
1336 if (uap->timeout != INFTIM) {
1337 if (uap->timeout < 0)
1339 ts.tv_sec = uap->timeout / 1000;
1340 ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1345 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1349 kern_poll(struct thread *td, struct pollfd *fds, u_int nfds,
1350 struct timespec *tsp, sigset_t *uset)
1352 struct pollfd *bits;
1353 struct pollfd smallbits[32];
1354 sbintime_t sbt, precision, tmp;
1362 if (tsp->tv_sec < 0)
1364 if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000)
1366 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1370 if (ts.tv_sec > INT32_MAX / 2) {
1371 over = ts.tv_sec - INT32_MAX / 2;
1377 precision >>= tc_precexp;
1378 if (TIMESEL(&sbt, tmp))
1385 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1387 ni = nfds * sizeof(struct pollfd);
1388 if (ni > sizeof(smallbits))
1389 bits = malloc(ni, M_TEMP, M_WAITOK);
1392 error = copyin(fds, bits, ni);
1397 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1398 &td->td_oldsigmask, 0);
1401 td->td_pflags |= TDP_OLDMASK;
1403 * Make sure that ast() is called on return to
1404 * usermode and TDP_OLDMASK is cleared, restoring old
1408 td->td_flags |= TDF_ASTPENDING;
1413 /* Iterate until the timeout expires or descriptors become ready. */
1415 error = pollscan(td, bits, nfds);
1416 if (error || td->td_retval[0] != 0)
1418 error = seltdwait(td, sbt, precision);
1421 error = pollrescan(td);
1422 if (error || td->td_retval[0] != 0)
1428 /* poll is not restarted after signals... */
1429 if (error == ERESTART)
1431 if (error == EWOULDBLOCK)
1434 error = pollout(td, bits, fds, nfds);
1439 if (ni > sizeof(smallbits))
1445 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1447 struct timespec ts, *tsp;
1451 if (uap->ts != NULL) {
1452 error = copyin(uap->ts, &ts, sizeof(ts));
1458 if (uap->set != NULL) {
1459 error = copyin(uap->set, &set, sizeof(set));
1466 * fds is still a pointer to user space. kern_poll() will
1467 * take care of copyin that array to the kernel space.
1470 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1474 pollrescan(struct thread *td)
1480 struct filedesc *fdp;
1484 cap_rights_t rights;
1489 fdp = td->td_proc->p_fd;
1491 FILEDESC_SLOCK(fdp);
1492 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1493 fd = (struct pollfd *)sfp->sf_cookie;
1495 selfdfree(stp, sfp);
1496 /* If the selinfo wasn't cleared the event didn't fire. */
1499 fp = fdp->fd_ofiles[fd->fd].fde_file;
1502 cap_check(cap_rights(fdp, fd->fd),
1503 cap_rights_init(&rights, CAP_EVENT)) != 0)
1508 fd->revents = POLLNVAL;
1514 * Note: backend also returns POLLHUP and
1515 * POLLERR if appropriate.
1517 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1518 if (fd->revents != 0)
1521 FILEDESC_SUNLOCK(fdp);
1523 td->td_retval[0] = n;
1529 pollout(td, fds, ufds, nfd)
1532 struct pollfd *ufds;
1539 for (i = 0; i < nfd; i++) {
1540 error = copyout(&fds->revents, &ufds->revents,
1541 sizeof(ufds->revents));
1544 if (fds->revents != 0)
1549 td->td_retval[0] = n;
1554 pollscan(td, fds, nfd)
1559 struct filedesc *fdp = td->td_proc->p_fd;
1562 cap_rights_t rights;
1566 FILEDESC_SLOCK(fdp);
1567 for (i = 0; i < nfd; i++, fds++) {
1568 if (fds->fd > fdp->fd_lastfile) {
1569 fds->revents = POLLNVAL;
1571 } else if (fds->fd < 0) {
1574 fp = fdp->fd_ofiles[fds->fd].fde_file;
1577 cap_check(cap_rights(fdp, fds->fd),
1578 cap_rights_init(&rights, CAP_EVENT)) != 0)
1583 fds->revents = POLLNVAL;
1587 * Note: backend also returns POLLHUP and
1588 * POLLERR if appropriate.
1590 selfdalloc(td, fds);
1591 fds->revents = fo_poll(fp, fds->events,
1594 * POSIX requires POLLOUT to be never
1595 * set simultaneously with POLLHUP.
1597 if ((fds->revents & POLLHUP) != 0)
1598 fds->revents &= ~POLLOUT;
1600 if (fds->revents != 0)
1605 FILEDESC_SUNLOCK(fdp);
1606 td->td_retval[0] = n;
1611 * XXX This was created specifically to support netncp and netsmb. This
1612 * allows the caller to specify a socket to wait for events on. It returns
1613 * 0 if any events matched and an error otherwise. There is no way to
1614 * determine which events fired.
1617 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1620 sbintime_t asbt, precision, rsbt;
1623 precision = 0; /* stupid gcc! */
1626 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1627 rtv.tv_usec >= 1000000)
1629 if (!timevalisset(&rtv))
1631 else if (rtv.tv_sec <= INT32_MAX) {
1632 rsbt = tvtosbt(rtv);
1634 precision >>= tc_precexp;
1635 if (TIMESEL(&asbt, rsbt))
1636 asbt += tc_tick_sbt;
1637 if (asbt <= SBT_MAX - rsbt)
1647 * Iterate until the timeout expires or the socket becomes ready.
1650 selfdalloc(td, NULL);
1651 error = sopoll(so, events, NULL, td);
1652 /* error here is actually the ready events. */
1655 error = seltdwait(td, asbt, precision);
1660 /* XXX Duplicates ncp/smb behavior. */
1661 if (error == ERESTART)
1667 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1668 * have two select sets, one for read and another for write.
1671 selfdalloc(struct thread *td, void *cookie)
1676 if (stp->st_free1 == NULL)
1677 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1678 stp->st_free1->sf_td = stp;
1679 stp->st_free1->sf_cookie = cookie;
1680 if (stp->st_free2 == NULL)
1681 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1682 stp->st_free2->sf_td = stp;
1683 stp->st_free2->sf_cookie = cookie;
1687 selfdfree(struct seltd *stp, struct selfd *sfp)
1689 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1690 if (sfp->sf_si != NULL) {
1691 mtx_lock(sfp->sf_mtx);
1692 if (sfp->sf_si != NULL) {
1693 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1694 refcount_release(&sfp->sf_refs);
1696 mtx_unlock(sfp->sf_mtx);
1698 if (refcount_release(&sfp->sf_refs))
1699 uma_zfree(selfd_zone, sfp);
1702 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1705 struct selinfo *sip;
1709 * This feature is already provided by doselwakeup(), thus it is
1710 * enough to go for it.
1711 * Eventually, the context, should take care to avoid races
1712 * between thread calling select()/poll() and file descriptor
1713 * detaching, but, again, the races are just the same as
1716 doselwakeup(sip, -1);
1720 * Record a select request.
1723 selrecord(selector, sip)
1724 struct thread *selector;
1725 struct selinfo *sip;
1731 stp = selector->td_sel;
1733 * Don't record when doing a rescan.
1735 if (stp->st_flags & SELTD_RESCAN)
1738 * Grab one of the preallocated descriptors.
1741 if ((sfp = stp->st_free1) != NULL)
1742 stp->st_free1 = NULL;
1743 else if ((sfp = stp->st_free2) != NULL)
1744 stp->st_free2 = NULL;
1746 panic("selrecord: No free selfd on selq");
1749 mtxp = mtx_pool_find(mtxpool_select, sip);
1751 * Initialize the sfp and queue it in the thread.
1755 refcount_init(&sfp->sf_refs, 2);
1756 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1758 * Now that we've locked the sip, check for initialization.
1761 if (sip->si_mtx == NULL) {
1763 TAILQ_INIT(&sip->si_tdlist);
1766 * Add this thread to the list of selfds listening on this selinfo.
1768 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1769 mtx_unlock(sip->si_mtx);
1772 /* Wake up a selecting thread. */
1775 struct selinfo *sip;
1777 doselwakeup(sip, -1);
1780 /* Wake up a selecting thread, and set its priority. */
1782 selwakeuppri(sip, pri)
1783 struct selinfo *sip;
1786 doselwakeup(sip, pri);
1790 * Do a wakeup when a selectable event occurs.
1793 doselwakeup(sip, pri)
1794 struct selinfo *sip;
1801 /* If it's not initialized there can't be any waiters. */
1802 if (sip->si_mtx == NULL)
1805 * Locking the selinfo locks all selfds associated with it.
1807 mtx_lock(sip->si_mtx);
1808 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1810 * Once we remove this sfp from the list and clear the
1811 * sf_si seltdclear will know to ignore this si.
1813 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1816 mtx_lock(&stp->st_mtx);
1817 stp->st_flags |= SELTD_PENDING;
1818 cv_broadcastpri(&stp->st_wait, pri);
1819 mtx_unlock(&stp->st_mtx);
1820 if (refcount_release(&sfp->sf_refs))
1821 uma_zfree(selfd_zone, sfp);
1823 mtx_unlock(sip->si_mtx);
1827 seltdinit(struct thread *td)
1831 if ((stp = td->td_sel) != NULL)
1833 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1834 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1835 cv_init(&stp->st_wait, "select");
1838 STAILQ_INIT(&stp->st_selq);
1842 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1849 * An event of interest may occur while we do not hold the seltd
1850 * locked so check the pending flag before we sleep.
1852 mtx_lock(&stp->st_mtx);
1854 * Any further calls to selrecord will be a rescan.
1856 stp->st_flags |= SELTD_RESCAN;
1857 if (stp->st_flags & SELTD_PENDING) {
1858 mtx_unlock(&stp->st_mtx);
1862 error = EWOULDBLOCK;
1864 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
1865 sbt, precision, C_ABSOLUTE);
1867 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1868 mtx_unlock(&stp->st_mtx);
1874 seltdfini(struct thread *td)
1882 uma_zfree(selfd_zone, stp->st_free1);
1884 uma_zfree(selfd_zone, stp->st_free2);
1886 free(stp, M_SELECT);
1890 * Remove the references to the thread from all of the objects we were
1894 seltdclear(struct thread *td)
1901 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1902 selfdfree(stp, sfp);
1906 static void selectinit(void *);
1907 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1909 selectinit(void *dummy __unused)
1912 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1913 NULL, NULL, UMA_ALIGN_PTR, 0);
1914 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
1918 * Set up a syscall return value that follows the convention specified for
1919 * posix_* functions.
1922 kern_posix_error(struct thread *td, int error)
1927 td->td_errno = error;
1928 td->td_pflags |= TDP_NERRNO;
1929 td->td_retval[0] = error;