2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 #include <sys/cdefs.h>
38 #include "opt_capsicum.h"
39 #include "opt_ktrace.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/capsicum.h>
45 #include <sys/filedesc.h>
46 #include <sys/filio.h>
47 #include <sys/fcntl.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
54 #include <sys/eventfd.h>
55 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/malloc.h>
60 #include <sys/resourcevar.h>
61 #include <sys/selinfo.h>
62 #include <sys/sleepqueue.h>
63 #include <sys/specialfd.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/unistd.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. */
160 MALLOC_DEFINE(M_SELFD, "selfd", "selfd");
161 static struct mtx_pool *mtxpool_select;
165 devfs_iosize_max(void)
168 return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
169 INT_MAX : SSIZE_MAX);
176 return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ?
177 INT_MAX : SSIZE_MAX);
181 #ifndef _SYS_SYSPROTO_H_
189 sys_read(struct thread *td, struct read_args *uap)
195 if (uap->nbyte > IOSIZE_MAX)
197 aiov.iov_base = uap->buf;
198 aiov.iov_len = uap->nbyte;
199 auio.uio_iov = &aiov;
201 auio.uio_resid = uap->nbyte;
202 auio.uio_segflg = UIO_USERSPACE;
203 error = kern_readv(td, uap->fd, &auio);
208 * Positioned read system call
210 #ifndef _SYS_SYSPROTO_H_
220 sys_pread(struct thread *td, struct pread_args *uap)
223 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
227 kern_pread(struct thread *td, int fd, void *buf, size_t nbyte, off_t offset)
233 if (nbyte > IOSIZE_MAX)
236 aiov.iov_len = nbyte;
237 auio.uio_iov = &aiov;
239 auio.uio_resid = nbyte;
240 auio.uio_segflg = UIO_USERSPACE;
241 error = kern_preadv(td, fd, &auio, offset);
245 #if defined(COMPAT_FREEBSD6)
247 freebsd6_pread(struct thread *td, struct freebsd6_pread_args *uap)
250 return (kern_pread(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
255 * Scatter read system call.
257 #ifndef _SYS_SYSPROTO_H_
265 sys_readv(struct thread *td, struct readv_args *uap)
270 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
273 error = kern_readv(td, uap->fd, auio);
279 kern_readv(struct thread *td, int fd, struct uio *auio)
284 error = fget_read(td, fd, &cap_read_rights, &fp);
287 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
293 * Scatter positioned read system call.
295 #ifndef _SYS_SYSPROTO_H_
304 sys_preadv(struct thread *td, struct preadv_args *uap)
309 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
312 error = kern_preadv(td, uap->fd, auio, uap->offset);
318 kern_preadv(struct thread *td, int fd, struct uio *auio, off_t offset)
323 error = fget_read(td, fd, &cap_pread_rights, &fp);
326 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
328 else if (offset < 0 &&
329 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
332 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
338 * Common code for readv and preadv that reads data in
339 * from a file using the passed in uio, offset, and flags.
342 dofileread(struct thread *td, int fd, struct file *fp, struct uio *auio,
343 off_t offset, int flags)
348 struct uio *ktruio = NULL;
353 /* Finish zero length reads right here */
354 if (auio->uio_resid == 0) {
355 td->td_retval[0] = 0;
358 auio->uio_rw = UIO_READ;
359 auio->uio_offset = offset;
362 if (KTRPOINT(td, KTR_GENIO))
363 ktruio = cloneuio(auio);
365 cnt = auio->uio_resid;
366 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
367 if (auio->uio_resid != cnt && (error == ERESTART ||
368 error == EINTR || error == EWOULDBLOCK))
371 cnt -= auio->uio_resid;
373 if (ktruio != NULL) {
374 ktruio->uio_resid = cnt;
375 ktrgenio(fd, UIO_READ, ktruio, error);
378 td->td_retval[0] = cnt;
382 #ifndef _SYS_SYSPROTO_H_
390 sys_write(struct thread *td, struct write_args *uap)
396 if (uap->nbyte > IOSIZE_MAX)
398 aiov.iov_base = (void *)(uintptr_t)uap->buf;
399 aiov.iov_len = uap->nbyte;
400 auio.uio_iov = &aiov;
402 auio.uio_resid = uap->nbyte;
403 auio.uio_segflg = UIO_USERSPACE;
404 error = kern_writev(td, uap->fd, &auio);
409 * Positioned write system call.
411 #ifndef _SYS_SYSPROTO_H_
421 sys_pwrite(struct thread *td, struct pwrite_args *uap)
424 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
428 kern_pwrite(struct thread *td, int fd, const void *buf, size_t nbyte,
435 if (nbyte > IOSIZE_MAX)
437 aiov.iov_base = (void *)(uintptr_t)buf;
438 aiov.iov_len = nbyte;
439 auio.uio_iov = &aiov;
441 auio.uio_resid = nbyte;
442 auio.uio_segflg = UIO_USERSPACE;
443 error = kern_pwritev(td, fd, &auio, offset);
447 #if defined(COMPAT_FREEBSD6)
449 freebsd6_pwrite(struct thread *td, struct freebsd6_pwrite_args *uap)
452 return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, uap->offset));
457 * Gather write system call.
459 #ifndef _SYS_SYSPROTO_H_
467 sys_writev(struct thread *td, struct writev_args *uap)
472 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
475 error = kern_writev(td, uap->fd, auio);
481 kern_writev(struct thread *td, int fd, struct uio *auio)
486 error = fget_write(td, fd, &cap_write_rights, &fp);
489 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
495 * Gather positioned write system call.
497 #ifndef _SYS_SYSPROTO_H_
498 struct pwritev_args {
506 sys_pwritev(struct thread *td, struct pwritev_args *uap)
511 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
514 error = kern_pwritev(td, uap->fd, auio, uap->offset);
520 kern_pwritev(struct thread *td, int fd, struct uio *auio, off_t offset)
525 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
528 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
530 else if (offset < 0 &&
531 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR))
534 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
540 * Common code for writev and pwritev that writes data to
541 * a file using the passed in uio, offset, and flags.
544 dofilewrite(struct thread *td, int fd, struct file *fp, struct uio *auio,
545 off_t offset, int flags)
550 struct uio *ktruio = NULL;
554 auio->uio_rw = UIO_WRITE;
556 auio->uio_offset = offset;
558 if (KTRPOINT(td, KTR_GENIO))
559 ktruio = cloneuio(auio);
561 cnt = auio->uio_resid;
562 error = fo_write(fp, auio, td->td_ucred, flags, td);
564 * Socket layer is responsible for special error handling,
567 if (error != 0 && fp->f_type != DTYPE_SOCKET) {
568 if (auio->uio_resid != cnt && (error == ERESTART ||
569 error == EINTR || error == EWOULDBLOCK))
571 if (error == EPIPE) {
572 PROC_LOCK(td->td_proc);
573 tdsignal(td, SIGPIPE);
574 PROC_UNLOCK(td->td_proc);
577 cnt -= auio->uio_resid;
579 if (ktruio != NULL) {
580 ktruio->uio_resid = cnt;
581 ktrgenio(fd, UIO_WRITE, ktruio, error);
584 td->td_retval[0] = cnt;
589 * Truncate a file given a file descriptor.
591 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
592 * descriptor isn't writable.
595 kern_ftruncate(struct thread *td, int fd, off_t length)
603 error = fget(td, fd, &cap_ftruncate_rights, &fp);
606 AUDIT_ARG_FILE(td->td_proc, fp);
607 if (!(fp->f_flag & FWRITE)) {
611 error = fo_truncate(fp, length, td->td_ucred, td);
616 #ifndef _SYS_SYSPROTO_H_
617 struct ftruncate_args {
624 sys_ftruncate(struct thread *td, struct ftruncate_args *uap)
627 return (kern_ftruncate(td, uap->fd, uap->length));
630 #if defined(COMPAT_43)
631 #ifndef _SYS_SYSPROTO_H_
632 struct oftruncate_args {
638 oftruncate(struct thread *td, struct oftruncate_args *uap)
641 return (kern_ftruncate(td, uap->fd, uap->length));
643 #endif /* COMPAT_43 */
645 #ifndef _SYS_SYSPROTO_H_
654 sys_ioctl(struct thread *td, struct ioctl_args *uap)
656 u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
663 if (uap->com > 0xffffffff) {
665 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
666 td->td_proc->p_pid, td->td_name, uap->com);
669 com = (uint32_t)uap->com;
672 * Interpret high order word to find amount of data to be
673 * copied to/from the user's address space.
675 size = IOCPARM_LEN(com);
676 if ((size > IOCPARM_MAX) ||
677 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
678 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
679 ((com & IOC_OUT) && size == 0) ||
681 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
683 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
687 if (com & IOC_VOID) {
688 /* Integer argument. */
689 arg = (intptr_t)uap->data;
693 if (size > SYS_IOCTL_SMALL_SIZE)
694 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
699 data = (void *)&uap->data;
701 error = copyin(uap->data, data, (u_int)size);
704 } else if (com & IOC_OUT) {
706 * Zero the buffer so the user always
707 * gets back something deterministic.
712 error = kern_ioctl(td, uap->fd, com, data);
714 if (error == 0 && (com & IOC_OUT))
715 error = copyout(data, uap->data, (u_int)size);
718 if (size > SYS_IOCTL_SMALL_SIZE)
719 free(data, M_IOCTLOPS);
724 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
727 struct filedesc *fdp;
728 int error, tmp, locked;
733 fdp = td->td_proc->p_fd;
746 locked = LA_UNLOCKED;
752 if ((fp = fget_noref(fdp, fd)) == NULL) {
756 if ((error = cap_ioctl_check(fdp, fd, com)) != 0) {
757 fp = NULL; /* fhold() was not called yet */
765 if (locked == LA_SLOCKED) {
766 FILEDESC_SUNLOCK(fdp);
767 locked = LA_UNLOCKED;
770 error = fget(td, fd, &cap_ioctl_rights, &fp);
776 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
783 fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE;
786 fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE;
789 if ((tmp = *(int *)data))
790 atomic_set_int(&fp->f_flag, FNONBLOCK);
792 atomic_clear_int(&fp->f_flag, FNONBLOCK);
796 if ((tmp = *(int *)data))
797 atomic_set_int(&fp->f_flag, FASYNC);
799 atomic_clear_int(&fp->f_flag, FASYNC);
804 error = fo_ioctl(fp, com, data, td->td_ucred, td);
808 FILEDESC_XUNLOCK(fdp);
812 FILEDESC_SUNLOCK(fdp);
816 FILEDESC_UNLOCK_ASSERT(fdp);
825 sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap)
829 error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len);
830 return (kern_posix_error(td, error));
834 kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len)
840 if (offset < 0 || len <= 0)
842 /* Check for wrap. */
843 if (offset > OFF_MAX - len)
846 error = fget(td, fd, &cap_pwrite_rights, &fp);
849 AUDIT_ARG_FILE(td->td_proc, fp);
850 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
854 if ((fp->f_flag & FWRITE) == 0) {
859 error = fo_fallocate(fp, offset, len, td);
866 sys_fspacectl(struct thread *td, struct fspacectl_args *uap)
868 struct spacectl_range rqsr, rmsr;
871 error = copyin(uap->rqsr, &rqsr, sizeof(rqsr));
875 error = kern_fspacectl(td, uap->fd, uap->cmd, &rqsr, uap->flags,
877 if (uap->rmsr != NULL) {
878 cerror = copyout(&rmsr, uap->rmsr, sizeof(rmsr));
886 kern_fspacectl(struct thread *td, int fd, int cmd,
887 const struct spacectl_range *rqsr, int flags, struct spacectl_range *rmsrp)
890 struct spacectl_range rmsr;
895 AUDIT_ARG_FFLAGS(flags);
903 if (cmd != SPACECTL_DEALLOC ||
904 rqsr->r_offset < 0 || rqsr->r_len <= 0 ||
905 rqsr->r_offset > OFF_MAX - rqsr->r_len ||
906 (flags & ~SPACECTL_F_SUPPORTED) != 0)
909 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
912 AUDIT_ARG_FILE(td->td_proc, fp);
913 if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
917 if ((fp->f_flag & FWRITE) == 0) {
922 error = fo_fspacectl(fp, cmd, &rmsr.r_offset, &rmsr.r_len, flags,
924 /* fspacectl is not restarted after signals if the file is modified. */
925 if (rmsr.r_len != rqsr->r_len && (error == ERESTART ||
926 error == EINTR || error == EWOULDBLOCK))
936 kern_specialfd(struct thread *td, int type, void *arg)
939 struct specialfd_eventfd *ae;
940 int error, fd, fflags;
943 error = falloc_noinstall(td, &fp);
948 case SPECIALFD_EVENTFD:
950 if ((ae->flags & EFD_CLOEXEC) != 0)
952 error = eventfd_create_file(td, fp, ae->initval, ae->flags);
960 error = finstall(td, fp, &fd, fflags, NULL);
963 td->td_retval[0] = fd;
968 sys___specialfd(struct thread *td, struct __specialfd_args *args)
970 struct specialfd_eventfd ae;
973 switch (args->type) {
974 case SPECIALFD_EVENTFD:
975 if (args->len != sizeof(struct specialfd_eventfd)) {
979 error = copyin(args->req, &ae, sizeof(ae));
982 if ((ae.flags & ~(EFD_CLOEXEC | EFD_NONBLOCK |
983 EFD_SEMAPHORE)) != 0) {
987 error = kern_specialfd(td, args->type, &ae);
997 poll_no_poll(int events)
1000 * Return true for read/write. If the user asked for something
1001 * special, return POLLNVAL, so that clients have a way of
1002 * determining reliably whether or not the extended
1003 * functionality is present without hard-coding knowledge
1004 * of specific filesystem implementations.
1006 if (events & ~POLLSTANDARD)
1009 return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
1013 sys_pselect(struct thread *td, struct pselect_args *uap)
1016 struct timeval tv, *tvp;
1017 sigset_t set, *uset;
1020 if (uap->ts != NULL) {
1021 error = copyin(uap->ts, &ts, sizeof(ts));
1024 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1028 if (uap->sm != NULL) {
1029 error = copyin(uap->sm, &set, sizeof(set));
1035 return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1040 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
1041 struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
1046 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1047 &td->td_oldsigmask, 0);
1050 td->td_pflags |= TDP_OLDMASK;
1052 * Make sure that ast() is called on return to
1053 * usermode and TDP_OLDMASK is cleared, restoring old
1056 ast_sched(td, TDA_SIGSUSPEND);
1058 error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
1062 #ifndef _SYS_SYSPROTO_H_
1063 struct select_args {
1065 fd_set *in, *ou, *ex;
1070 sys_select(struct thread *td, struct select_args *uap)
1072 struct timeval tv, *tvp;
1075 if (uap->tv != NULL) {
1076 error = copyin(uap->tv, &tv, sizeof(tv));
1083 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
1088 * In the unlikely case when user specified n greater then the last
1089 * open file descriptor, check that no bits are set after the last
1090 * valid fd. We must return EBADF if any is set.
1092 * There are applications that rely on the behaviour.
1097 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
1103 if (nd >= ndu || fd_in == NULL)
1107 bits = 0; /* silence gcc */
1108 for (i = nd; i < ndu; i++) {
1110 #if BYTE_ORDER == LITTLE_ENDIAN
1111 addr = (char *)fd_in + b;
1113 addr = (char *)fd_in;
1114 if (abi_nfdbits == NFDBITS) {
1115 addr += rounddown(b, sizeof(fd_mask)) +
1116 sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
1118 addr += rounddown(b, sizeof(uint32_t)) +
1119 sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
1122 if (addr != oaddr) {
1129 if ((bits & (1 << (i % NBBY))) != 0)
1136 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
1137 fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
1139 struct filedesc *fdp;
1141 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
1142 * infds with the new FD_SETSIZE of 1024, and more than enough for
1143 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
1146 fd_mask s_selbits[howmany(2048, NFDBITS)];
1147 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
1149 sbintime_t asbt, precision, rsbt;
1150 u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
1155 fdp = td->td_proc->p_fd;
1157 lf = fdp->fd_nfiles;
1161 error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
1164 error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
1167 error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
1172 * Allocate just enough bits for the non-null fd_sets. Use the
1173 * preallocated auto buffer if possible.
1175 nfdbits = roundup(nd, NFDBITS);
1176 ncpbytes = nfdbits / NBBY;
1177 ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
1180 nbufbytes += 2 * ncpbytes;
1182 nbufbytes += 2 * ncpbytes;
1184 nbufbytes += 2 * ncpbytes;
1185 if (nbufbytes <= sizeof s_selbits)
1186 selbits = &s_selbits[0];
1188 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
1191 * Assign pointers into the bit buffers and fetch the input bits.
1192 * Put the output buffers together so that they can be bzeroed
1196 #define getbits(name, x) \
1198 if (name == NULL) { \
1202 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
1204 sbp += ncpbytes / sizeof *sbp; \
1205 error = copyin(name, ibits[x], ncpubytes); \
1208 if (ncpbytes != ncpubytes) \
1209 bzero((char *)ibits[x] + ncpubytes, \
1210 ncpbytes - ncpubytes); \
1218 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
1220 * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
1221 * we are running under 32-bit emulation. This should be more
1224 #define swizzle_fdset(bits) \
1225 if (abi_nfdbits != NFDBITS && bits != NULL) { \
1227 for (i = 0; i < ncpbytes / sizeof *sbp; i++) \
1228 bits[i] = (bits[i] >> 32) | (bits[i] << 32); \
1231 #define swizzle_fdset(bits)
1234 /* Make sure the bit order makes it through an ABI transition */
1235 swizzle_fdset(ibits[0]);
1236 swizzle_fdset(ibits[1]);
1237 swizzle_fdset(ibits[2]);
1240 bzero(selbits, nbufbytes / 2);
1245 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1246 rtv.tv_usec >= 1000000) {
1250 if (!timevalisset(&rtv))
1252 else if (rtv.tv_sec <= INT32_MAX) {
1253 rsbt = tvtosbt(rtv);
1255 precision >>= tc_precexp;
1256 if (TIMESEL(&asbt, rsbt))
1257 asbt += tc_tick_sbt;
1258 if (asbt <= SBT_MAX - rsbt)
1267 /* Iterate until the timeout expires or descriptors become ready. */
1269 error = selscan(td, ibits, obits, nd);
1270 if (error || td->td_retval[0] != 0)
1272 error = seltdwait(td, asbt, precision);
1275 error = selrescan(td, ibits, obits);
1276 if (error || td->td_retval[0] != 0)
1282 /* select is not restarted after signals... */
1283 if (error == ERESTART)
1285 if (error == EWOULDBLOCK)
1288 /* swizzle bit order back, if necessary */
1289 swizzle_fdset(obits[0]);
1290 swizzle_fdset(obits[1]);
1291 swizzle_fdset(obits[2]);
1292 #undef swizzle_fdset
1294 #define putbits(name, x) \
1295 if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
1305 if (selbits != &s_selbits[0])
1306 free(selbits, M_SELECT);
1311 * Convert a select bit set to poll flags.
1313 * The backend always returns POLLHUP/POLLERR if appropriate and we
1314 * return this as a set bit in any set.
1316 static const int select_flags[3] = {
1317 POLLRDNORM | POLLHUP | POLLERR,
1318 POLLWRNORM | POLLHUP | POLLERR,
1319 POLLRDBAND | POLLERR
1323 * Compute the fo_poll flags required for a fd given by the index and
1324 * bit position in the fd_mask array.
1327 selflags(fd_mask **ibits, int idx, fd_mask bit)
1333 for (msk = 0; msk < 3; msk++) {
1334 if (ibits[msk] == NULL)
1336 if ((ibits[msk][idx] & bit) == 0)
1338 flags |= select_flags[msk];
1344 * Set the appropriate output bits given a mask of fired events and the
1345 * input bits originally requested.
1348 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
1354 for (msk = 0; msk < 3; msk++) {
1355 if ((events & select_flags[msk]) == 0)
1357 if (ibits[msk] == NULL)
1359 if ((ibits[msk][idx] & bit) == 0)
1362 * XXX Check for a duplicate set. This can occur because a
1363 * socket calls selrecord() twice for each poll() call
1364 * resulting in two selfds per real fd. selrescan() will
1365 * call selsetbits twice as a result.
1367 if ((obits[msk][idx] & bit) != 0)
1369 obits[msk][idx] |= bit;
1377 * Traverse the list of fds attached to this thread's seltd and check for
1381 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
1383 struct filedesc *fdp;
1394 fdp = td->td_proc->p_fd;
1397 only_user = FILEDESC_IS_ONLY_USER(fdp);
1398 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1399 fd = (int)(uintptr_t)sfp->sf_cookie;
1401 selfdfree(stp, sfp);
1402 /* If the selinfo wasn't cleared the event didn't fire. */
1406 error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1408 error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1409 if (__predict_false(error != 0))
1412 bit = (fd_mask)1 << (fd % NFDBITS);
1413 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
1415 fput_only_user(fdp, fp);
1419 n += selsetbits(ibits, obits, idx, bit, ev);
1422 td->td_retval[0] = n;
1427 * Perform the initial filedescriptor scan and register ourselves with
1431 selscan(struct thread *td, fd_mask **ibits, fd_mask **obits, int nfd)
1433 struct filedesc *fdp;
1436 int ev, flags, end, fd;
1441 fdp = td->td_proc->p_fd;
1443 only_user = FILEDESC_IS_ONLY_USER(fdp);
1444 for (idx = 0, fd = 0; fd < nfd; idx++) {
1445 end = imin(fd + NFDBITS, nfd);
1446 for (bit = 1; fd < end; bit <<= 1, fd++) {
1447 /* Compute the list of events we're interested in. */
1448 flags = selflags(ibits, idx, bit);
1452 error = fget_only_user(fdp, fd, &cap_event_rights, &fp);
1454 error = fget_unlocked(td, fd, &cap_event_rights, &fp);
1455 if (__predict_false(error != 0))
1457 selfdalloc(td, (void *)(uintptr_t)fd);
1458 ev = fo_poll(fp, flags, td->td_ucred, td);
1460 fput_only_user(fdp, fp);
1464 n += selsetbits(ibits, obits, idx, bit, ev);
1468 td->td_retval[0] = n;
1473 sys_poll(struct thread *td, struct poll_args *uap)
1475 struct timespec ts, *tsp;
1477 if (uap->timeout != INFTIM) {
1478 if (uap->timeout < 0)
1480 ts.tv_sec = uap->timeout / 1000;
1481 ts.tv_nsec = (uap->timeout % 1000) * 1000000;
1486 return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL));
1490 * kfds points to an array in the kernel.
1493 kern_poll_kfds(struct thread *td, struct pollfd *kfds, u_int nfds,
1494 struct timespec *tsp, sigset_t *uset)
1496 sbintime_t sbt, precision, tmp;
1503 if (!timespecvalid_interval(tsp))
1505 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1509 if (ts.tv_sec > INT32_MAX / 2) {
1510 over = ts.tv_sec - INT32_MAX / 2;
1516 precision >>= tc_precexp;
1517 if (TIMESEL(&sbt, tmp))
1525 error = kern_sigprocmask(td, SIG_SETMASK, uset,
1526 &td->td_oldsigmask, 0);
1529 td->td_pflags |= TDP_OLDMASK;
1531 * Make sure that ast() is called on return to
1532 * usermode and TDP_OLDMASK is cleared, restoring old
1535 ast_sched(td, TDA_SIGSUSPEND);
1539 /* Iterate until the timeout expires or descriptors become ready. */
1541 error = pollscan(td, kfds, nfds);
1542 if (error || td->td_retval[0] != 0)
1544 error = seltdwait(td, sbt, precision);
1547 error = pollrescan(td);
1548 if (error || td->td_retval[0] != 0)
1553 /* poll is not restarted after signals... */
1554 if (error == ERESTART)
1556 if (error == EWOULDBLOCK)
1562 sys_ppoll(struct thread *td, struct ppoll_args *uap)
1564 struct timespec ts, *tsp;
1568 if (uap->ts != NULL) {
1569 error = copyin(uap->ts, &ts, sizeof(ts));
1575 if (uap->set != NULL) {
1576 error = copyin(uap->set, &set, sizeof(set));
1582 return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp));
1586 * ufds points to an array in user space.
1589 kern_poll(struct thread *td, struct pollfd *ufds, u_int nfds,
1590 struct timespec *tsp, sigset_t *set)
1592 struct pollfd *kfds;
1593 struct pollfd stackfds[32];
1596 if (kern_poll_maxfds(nfds))
1598 if (nfds > nitems(stackfds))
1599 kfds = mallocarray(nfds, sizeof(*kfds), M_TEMP, M_WAITOK);
1602 error = copyin(ufds, kfds, nfds * sizeof(*kfds));
1606 error = kern_poll_kfds(td, kfds, nfds, tsp, set);
1608 error = pollout(td, kfds, ufds, nfds);
1611 if (nfds > nitems(stackfds))
1617 kern_poll_maxfds(u_int nfds)
1621 * This is kinda bogus. We have fd limits, but that is not
1622 * really related to the size of the pollfd array. Make sure
1623 * we let the process use at least FD_SETSIZE entries and at
1624 * least enough for the system-wide limits. We want to be reasonably
1625 * safe, but not overly restrictive.
1627 return (nfds > maxfilesperproc && nfds > FD_SETSIZE);
1631 pollrescan(struct thread *td)
1637 struct filedesc *fdp;
1644 fdp = td->td_proc->p_fd;
1646 only_user = FILEDESC_IS_ONLY_USER(fdp);
1647 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1648 fd = (struct pollfd *)sfp->sf_cookie;
1650 selfdfree(stp, sfp);
1651 /* If the selinfo wasn't cleared the event didn't fire. */
1655 error = fget_only_user(fdp, fd->fd, &cap_event_rights, &fp);
1657 error = fget_unlocked(td, fd->fd, &cap_event_rights, &fp);
1658 if (__predict_false(error != 0)) {
1659 fd->revents = POLLNVAL;
1664 * Note: backend also returns POLLHUP and
1665 * POLLERR if appropriate.
1667 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1669 fput_only_user(fdp, fp);
1672 if (fd->revents != 0)
1676 td->td_retval[0] = n;
1681 pollout(struct thread *td, struct pollfd *fds, struct pollfd *ufds, u_int nfd)
1687 for (i = 0; i < nfd; i++) {
1688 error = copyout(&fds->revents, &ufds->revents,
1689 sizeof(ufds->revents));
1692 if (fds->revents != 0)
1697 td->td_retval[0] = n;
1702 pollscan(struct thread *td, struct pollfd *fds, u_int nfd)
1704 struct filedesc *fdp;
1710 fdp = td->td_proc->p_fd;
1711 only_user = FILEDESC_IS_ONLY_USER(fdp);
1712 for (i = 0; i < nfd; i++, fds++) {
1718 error = fget_only_user(fdp, fds->fd, &cap_event_rights, &fp);
1720 error = fget_unlocked(td, fds->fd, &cap_event_rights, &fp);
1721 if (__predict_false(error != 0)) {
1722 fds->revents = POLLNVAL;
1727 * Note: backend also returns POLLHUP and
1728 * POLLERR if appropriate.
1730 selfdalloc(td, fds);
1731 fds->revents = fo_poll(fp, fds->events,
1734 fput_only_user(fdp, fp);
1738 * POSIX requires POLLOUT to be never
1739 * set simultaneously with POLLHUP.
1741 if ((fds->revents & POLLHUP) != 0)
1742 fds->revents &= ~POLLOUT;
1744 if (fds->revents != 0)
1747 td->td_retval[0] = n;
1752 * XXX This was created specifically to support netncp and netsmb. This
1753 * allows the caller to specify a socket to wait for events on. It returns
1754 * 0 if any events matched and an error otherwise. There is no way to
1755 * determine which events fired.
1758 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1761 sbintime_t asbt, precision, rsbt;
1764 precision = 0; /* stupid gcc! */
1767 if (rtv.tv_sec < 0 || rtv.tv_usec < 0 ||
1768 rtv.tv_usec >= 1000000)
1770 if (!timevalisset(&rtv))
1772 else if (rtv.tv_sec <= INT32_MAX) {
1773 rsbt = tvtosbt(rtv);
1775 precision >>= tc_precexp;
1776 if (TIMESEL(&asbt, rsbt))
1777 asbt += tc_tick_sbt;
1778 if (asbt <= SBT_MAX - rsbt)
1788 * Iterate until the timeout expires or the socket becomes ready.
1791 selfdalloc(td, NULL);
1792 if (sopoll(so, events, NULL, td) != 0) {
1796 error = seltdwait(td, asbt, precision);
1801 /* XXX Duplicates ncp/smb behavior. */
1802 if (error == ERESTART)
1808 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1809 * have two select sets, one for read and another for write.
1812 selfdalloc(struct thread *td, void *cookie)
1817 if (stp->st_free1 == NULL)
1818 stp->st_free1 = malloc(sizeof(*stp->st_free1), M_SELFD, M_WAITOK|M_ZERO);
1819 stp->st_free1->sf_td = stp;
1820 stp->st_free1->sf_cookie = cookie;
1821 if (stp->st_free2 == NULL)
1822 stp->st_free2 = malloc(sizeof(*stp->st_free2), M_SELFD, M_WAITOK|M_ZERO);
1823 stp->st_free2->sf_td = stp;
1824 stp->st_free2->sf_cookie = cookie;
1828 selfdfree(struct seltd *stp, struct selfd *sfp)
1830 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1832 * Paired with doselwakeup.
1834 if (atomic_load_acq_ptr((uintptr_t *)&sfp->sf_si) != (uintptr_t)NULL) {
1835 mtx_lock(sfp->sf_mtx);
1836 if (sfp->sf_si != NULL) {
1837 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1839 mtx_unlock(sfp->sf_mtx);
1844 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
1846 seldrain(struct selinfo *sip)
1850 * This feature is already provided by doselwakeup(), thus it is
1851 * enough to go for it.
1852 * Eventually, the context, should take care to avoid races
1853 * between thread calling select()/poll() and file descriptor
1854 * detaching, but, again, the races are just the same as
1857 doselwakeup(sip, -1);
1861 * Record a select request.
1864 selrecord(struct thread *selector, struct selinfo *sip)
1870 stp = selector->td_sel;
1872 * Don't record when doing a rescan.
1874 if (stp->st_flags & SELTD_RESCAN)
1877 * Grab one of the preallocated descriptors.
1880 if ((sfp = stp->st_free1) != NULL)
1881 stp->st_free1 = NULL;
1882 else if ((sfp = stp->st_free2) != NULL)
1883 stp->st_free2 = NULL;
1885 panic("selrecord: No free selfd on selq");
1888 mtxp = mtx_pool_find(mtxpool_select, sip);
1890 * Initialize the sfp and queue it in the thread.
1894 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1896 * Now that we've locked the sip, check for initialization.
1899 if (sip->si_mtx == NULL) {
1901 TAILQ_INIT(&sip->si_tdlist);
1904 * Add this thread to the list of selfds listening on this selinfo.
1906 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1907 mtx_unlock(sip->si_mtx);
1910 /* Wake up a selecting thread. */
1912 selwakeup(struct selinfo *sip)
1914 doselwakeup(sip, -1);
1917 /* Wake up a selecting thread, and set its priority. */
1919 selwakeuppri(struct selinfo *sip, int pri)
1921 doselwakeup(sip, pri);
1925 * Do a wakeup when a selectable event occurs.
1928 doselwakeup(struct selinfo *sip, int pri)
1934 /* If it's not initialized there can't be any waiters. */
1935 if (sip->si_mtx == NULL)
1938 * Locking the selinfo locks all selfds associated with it.
1940 mtx_lock(sip->si_mtx);
1941 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1943 * Once we remove this sfp from the list and clear the
1944 * sf_si seltdclear will know to ignore this si.
1946 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1948 mtx_lock(&stp->st_mtx);
1949 stp->st_flags |= SELTD_PENDING;
1950 cv_broadcastpri(&stp->st_wait, pri);
1951 mtx_unlock(&stp->st_mtx);
1953 * Paired with selfdfree.
1955 * Storing this only after the wakeup provides an invariant that
1956 * stp is not used after selfdfree returns.
1958 atomic_store_rel_ptr((uintptr_t *)&sfp->sf_si, (uintptr_t)NULL);
1960 mtx_unlock(sip->si_mtx);
1964 seltdinit(struct thread *td)
1970 MPASS(stp->st_flags == 0);
1971 MPASS(STAILQ_EMPTY(&stp->st_selq));
1974 stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1975 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1976 cv_init(&stp->st_wait, "select");
1978 STAILQ_INIT(&stp->st_selq);
1983 seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision)
1990 * An event of interest may occur while we do not hold the seltd
1991 * locked so check the pending flag before we sleep.
1993 mtx_lock(&stp->st_mtx);
1995 * Any further calls to selrecord will be a rescan.
1997 stp->st_flags |= SELTD_RESCAN;
1998 if (stp->st_flags & SELTD_PENDING) {
1999 mtx_unlock(&stp->st_mtx);
2003 error = EWOULDBLOCK;
2005 error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx,
2006 sbt, precision, C_ABSOLUTE);
2008 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
2009 mtx_unlock(&stp->st_mtx);
2015 seltdfini(struct thread *td)
2022 MPASS(stp->st_flags == 0);
2023 MPASS(STAILQ_EMPTY(&stp->st_selq));
2025 free(stp->st_free1, M_SELFD);
2027 free(stp->st_free2, M_SELFD);
2029 cv_destroy(&stp->st_wait);
2030 mtx_destroy(&stp->st_mtx);
2031 free(stp, M_SELECT);
2035 * Remove the references to the thread from all of the objects we were
2039 seltdclear(struct thread *td)
2046 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
2047 selfdfree(stp, sfp);
2051 static void selectinit(void *);
2052 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
2054 selectinit(void *dummy __unused)
2057 mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
2061 * Set up a syscall return value that follows the convention specified for
2062 * posix_* functions.
2065 kern_posix_error(struct thread *td, int error)
2070 td->td_errno = error;
2071 td->td_pflags |= TDP_NERRNO;
2072 td->td_retval[0] = error;
2077 kcmp_cmp(uintptr_t a, uintptr_t b)
2087 kcmp_pget(struct thread *td, pid_t pid, struct proc **pp)
2089 if (pid == td->td_proc->p_pid) {
2093 return (pget(pid, PGET_CANDEBUG | PGET_NOTWEXIT | PGET_HOLD, pp));
2097 kern_kcmp(struct thread *td, pid_t pid1, pid_t pid2, int type,
2098 uintptr_t idx1, uintptr_t idx2)
2100 struct proc *p1, *p2;
2101 struct file *fp1, *fp2;
2106 error = kcmp_pget(td, pid1, &p1);
2108 error = kcmp_pget(td, pid2, &p2);
2115 error = fget_remote(td, p1, idx1, &fp1);
2117 error = fget_remote(td, p2, idx2, &fp2);
2119 if (type == KCMP_FILEOBJ)
2120 res = fo_cmp(fp1, fp2, td);
2122 res = kcmp_cmp((uintptr_t)fp1,
2130 res = kcmp_cmp((uintptr_t)p1->p_fd, (uintptr_t)p2->p_fd);
2133 res = kcmp_cmp((uintptr_t)p1->p_sigacts,
2134 (uintptr_t)p2->p_sigacts);
2137 res = kcmp_cmp((uintptr_t)p1->p_vmspace,
2138 (uintptr_t)p2->p_vmspace);
2146 if (p1 != NULL && p1 != td->td_proc)
2148 if (p2 != NULL && p2 != td->td_proc)
2151 td->td_retval[0] = res;
2156 sys_kcmp(struct thread *td, struct kcmp_args *uap)
2158 return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type,
2159 uap->idx1, uap->idx2));
2163 file_kcmp_generic(struct file *fp1, struct file *fp2, struct thread *td)
2165 if (fp1->f_type != fp2->f_type)
2167 return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));