2 * Copyright (c) 2007 Roman Divacky
3 * Copyright (c) 2014 Dmitry Chagin
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include "opt_compat.h"
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/imgact.h>
36 #include <sys/kernel.h>
37 #include <sys/limits.h>
39 #include <sys/mutex.h>
40 #include <sys/capability.h>
41 #include <sys/types.h>
43 #include <sys/filedesc.h>
44 #include <sys/errno.h>
45 #include <sys/event.h>
48 #include <sys/selinfo.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/timespec.h>
54 #include <machine/../linux32/linux.h>
55 #include <machine/../linux32/linux32_proto.h>
57 #include <machine/../linux/linux.h>
58 #include <machine/../linux/linux_proto.h>
61 #include <compat/linux/linux_emul.h>
62 #include <compat/linux/linux_event.h>
63 #include <compat/linux/linux_file.h>
64 #include <compat/linux/linux_util.h>
67 * epoll defines 'struct epoll_event' with the field 'data' as 64 bits
68 * on all architectures. But on 32 bit architectures BSD 'struct kevent' only
69 * has 32 bit opaque pointer as 'udata' field. So we can't pass epoll supplied
70 * data verbatuim. Therefore we allocate 64-bit memory block to pass
71 * user supplied data for every file descriptor.
74 typedef uint64_t epoll_udata_t;
76 struct epoll_emuldata {
77 uint32_t fdc; /* epoll udata max index */
78 epoll_udata_t udata[1]; /* epoll user data vector */
81 #define EPOLL_DEF_SZ 16
82 #define EPOLL_SIZE(fdn) \
83 (sizeof(struct epoll_emuldata)+(fdn) * sizeof(epoll_udata_t))
89 #if defined(__amd64__)
90 __attribute__((packed))
94 #define LINUX_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
96 static void epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata);
97 static int epoll_to_kevent(struct thread *td, struct file *epfp,
98 int fd, struct epoll_event *l_event, int *kev_flags,
99 struct kevent *kevent, int *nkevents);
100 static void kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event);
101 static int epoll_kev_copyout(void *arg, struct kevent *kevp, int count);
102 static int epoll_kev_copyin(void *arg, struct kevent *kevp, int count);
103 static int epoll_delete_event(struct thread *td, struct file *epfp,
105 static int epoll_delete_all_events(struct thread *td, struct file *epfp,
108 struct epoll_copyin_args {
109 struct kevent *changelist;
112 struct epoll_copyout_args {
113 struct epoll_event *leventlist;
120 typedef uint64_t eventfd_t;
122 static fo_rdwr_t eventfd_read;
123 static fo_rdwr_t eventfd_write;
124 static fo_truncate_t eventfd_truncate;
125 static fo_ioctl_t eventfd_ioctl;
126 static fo_poll_t eventfd_poll;
127 static fo_kqfilter_t eventfd_kqfilter;
128 static fo_stat_t eventfd_stat;
129 static fo_close_t eventfd_close;
131 static struct fileops eventfdops = {
132 .fo_read = eventfd_read,
133 .fo_write = eventfd_write,
134 .fo_truncate = eventfd_truncate,
135 .fo_ioctl = eventfd_ioctl,
136 .fo_poll = eventfd_poll,
137 .fo_kqfilter = eventfd_kqfilter,
138 .fo_stat = eventfd_stat,
139 .fo_close = eventfd_close,
140 .fo_chmod = invfo_chmod,
141 .fo_chown = invfo_chown,
142 .fo_sendfile = invfo_sendfile,
143 .fo_flags = DFLAG_PASSABLE
146 static void filt_eventfddetach(struct knote *kn);
147 static int filt_eventfdread(struct knote *kn, long hint);
148 static int filt_eventfdwrite(struct knote *kn, long hint);
150 static struct filterops eventfd_rfiltops = {
152 .f_detach = filt_eventfddetach,
153 .f_event = filt_eventfdread
155 static struct filterops eventfd_wfiltops = {
157 .f_detach = filt_eventfddetach,
158 .f_event = filt_eventfdwrite
164 struct selinfo efd_sel;
168 static int eventfd_create(struct thread *td, uint32_t initval, int flags);
172 epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata)
174 struct linux_pemuldata *pem;
175 struct epoll_emuldata *emd;
181 KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
183 LINUX_PEM_XLOCK(pem);
184 if (pem->epoll == NULL) {
185 emd = malloc(EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
191 emd = realloc(emd, EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
196 emd->udata[fd] = udata;
197 LINUX_PEM_XUNLOCK(pem);
201 epoll_create_common(struct thread *td, int flags)
205 error = kern_kqueue(td, flags);
209 epoll_fd_install(td, EPOLL_DEF_SZ, 0);
215 linux_epoll_create(struct thread *td, struct linux_epoll_create_args *args)
219 * args->size is unused. Linux just tests it
220 * and then forgets it as well.
225 return (epoll_create_common(td, 0));
229 linux_epoll_create1(struct thread *td, struct linux_epoll_create1_args *args)
233 if ((args->flags & ~(LINUX_O_CLOEXEC)) != 0)
237 if ((args->flags & LINUX_O_CLOEXEC) != 0)
240 return (epoll_create_common(td, flags));
243 /* Structure converting function from epoll to kevent. */
245 epoll_to_kevent(struct thread *td, struct file *epfp,
246 int fd, struct epoll_event *l_event, int *kev_flags,
247 struct kevent *kevent, int *nkevents)
249 uint32_t levents = l_event->events;
250 struct linux_pemuldata *pem;
253 /* flags related to how event is registered */
254 if ((levents & LINUX_EPOLLONESHOT) != 0)
255 *kev_flags |= EV_ONESHOT;
256 if ((levents & LINUX_EPOLLET) != 0)
257 *kev_flags |= EV_CLEAR;
258 if ((levents & LINUX_EPOLLERR) != 0)
259 *kev_flags |= EV_ERROR;
260 if ((levents & LINUX_EPOLLRDHUP) != 0)
261 *kev_flags |= EV_EOF;
263 /* flags related to what event is registered */
264 if ((levents & LINUX_EPOLL_EVRD) != 0) {
265 EV_SET(kevent++, fd, EVFILT_READ, *kev_flags, 0, 0, 0);
268 if ((levents & LINUX_EPOLL_EVWR) != 0) {
269 EV_SET(kevent++, fd, EVFILT_WRITE, *kev_flags, 0, 0, 0);
273 if ((levents & ~(LINUX_EPOLL_EVSUP)) != 0) {
277 KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
278 KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n"));
280 LINUX_PEM_XLOCK(pem);
281 if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) {
282 pem->flags |= LINUX_XUNSUP_EPOLL;
283 LINUX_PEM_XUNLOCK(pem);
284 linux_msg(td, "epoll_ctl unsupported flags: 0x%x\n",
287 LINUX_PEM_XUNLOCK(pem);
295 * Structure converting function from kevent to epoll. In a case
296 * this is called on error in registration we store the error in
297 * event->data and pick it up later in linux_epoll_ctl().
300 kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event)
303 if ((kevent->flags & EV_ERROR) != 0) {
304 l_event->events = LINUX_EPOLLERR;
308 switch (kevent->filter) {
310 l_event->events = LINUX_EPOLLIN|LINUX_EPOLLRDNORM|LINUX_EPOLLPRI;
311 if ((kevent->flags & EV_EOF) != 0)
312 l_event->events |= LINUX_EPOLLRDHUP;
315 l_event->events = LINUX_EPOLLOUT|LINUX_EPOLLWRNORM;
321 * Copyout callback used by kevent. This converts kevent
322 * events to epoll events and copies them back to the
323 * userspace. This is also called on error on registering
327 epoll_kev_copyout(void *arg, struct kevent *kevp, int count)
329 struct epoll_copyout_args *args;
330 struct linux_pemuldata *pem;
331 struct epoll_emuldata *emd;
332 struct epoll_event *eep;
335 args = (struct epoll_copyout_args*) arg;
336 eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO);
338 pem = pem_find(args->p);
339 KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
340 LINUX_PEM_SLOCK(pem);
342 KASSERT(emd != NULL, ("epoll proc epolldata not found.\n"));
344 for (i = 0; i < count; i++) {
345 kevent_to_epoll(&kevp[i], &eep[i]);
348 KASSERT(fd <= emd->fdc, ("epoll user data vector"
349 " is too small.\n"));
350 eep[i].data = emd->udata[fd];
352 LINUX_PEM_SUNLOCK(pem);
354 error = copyout(eep, args->leventlist, count * sizeof(*eep));
356 args->leventlist += count;
357 args->count += count;
358 } else if (args->error == 0)
366 * Copyin callback used by kevent. This copies already
367 * converted filters from kernel memory to the kevent
368 * internal kernel memory. Hence the memcpy instead of
372 epoll_kev_copyin(void *arg, struct kevent *kevp, int count)
374 struct epoll_copyin_args *args;
376 args = (struct epoll_copyin_args*) arg;
378 memcpy(kevp, args->changelist, count * sizeof(*kevp));
379 args->changelist += count;
385 * Load epoll filter, convert it to kevent filter
386 * and load it into kevent subsystem.
389 linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args)
391 struct file *epfp, *fp;
392 struct epoll_copyin_args ciargs;
393 struct kevent kev[2];
394 struct kevent_copyops k_ops = { &ciargs,
397 struct epoll_event le;
403 if (args->op != LINUX_EPOLL_CTL_DEL) {
404 error = copyin(args->event, &le, sizeof(le));
409 error = fget(td, args->epfd,
410 cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &epfp);
413 if (epfp->f_type != DTYPE_KQUEUE)
416 /* Protect user data vector from incorrectly supplied fd. */
417 error = fget(td, args->fd, cap_rights_init(&rights, CAP_POLL_EVENT), &fp);
421 /* Linux disallows spying on himself */
427 ciargs.changelist = kev;
430 case LINUX_EPOLL_CTL_MOD:
432 * We don't memorize which events were set for this FD
433 * on this level, so just delete all we could have set:
434 * EVFILT_READ and EVFILT_WRITE, ignoring any errors
436 error = epoll_delete_all_events(td, epfp, args->fd);
441 case LINUX_EPOLL_CTL_ADD:
442 kev_flags = EV_ADD | EV_ENABLE;
445 case LINUX_EPOLL_CTL_DEL:
446 /* CTL_DEL means unregister this fd with this epoll */
447 error = epoll_delete_all_events(td, epfp, args->fd);
455 error = epoll_to_kevent(td, epfp, args->fd, &le, &kev_flags,
460 epoll_fd_install(td, args->fd, le.data);
462 error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL);
473 * Wait for a filter to be triggered on the epoll file descriptor.
476 linux_epoll_wait_common(struct thread *td, int epfd, struct epoll_event *events,
477 int maxevents, int timeout, sigset_t *uset)
480 struct timespec ts, *tsp;
482 struct epoll_copyout_args coargs;
483 struct kevent_copyops k_ops = { &coargs,
488 if (maxevents <= 0 || maxevents > LINUX_MAX_EVENTS)
492 error = kern_sigprocmask(td, SIG_SETMASK, uset,
493 &td->td_oldsigmask, 0);
496 td->td_pflags |= TDP_OLDMASK;
498 * Make sure that ast() is called on return to
499 * usermode and TDP_OLDMASK is cleared, restoring old
503 td->td_flags |= TDF_ASTPENDING;
507 error = fget(td, epfd,
508 cap_rights_init(&rights, CAP_KQUEUE_EVENT), &epfp);
512 coargs.leventlist = events;
513 coargs.p = td->td_proc;
522 /* Convert from milliseconds to timespec. */
523 ts.tv_sec = timeout / 1000;
524 ts.tv_nsec = (timeout % 1000) * 1000000;
530 error = kern_kevent_fp(td, epfp, 0, maxevents, &k_ops, tsp);
531 if (error == 0 && coargs.error != 0)
532 error = coargs.error;
535 * kern_kevent might return ENOMEM which is not expected from epoll_wait.
536 * Maybe we should translate that but I don't think it matters at all.
539 td->td_retval[0] = coargs.count;
546 linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args)
549 return (linux_epoll_wait_common(td, args->epfd, args->events,
550 args->maxevents, args->timeout, NULL));
554 linux_epoll_pwait(struct thread *td, struct linux_epoll_pwait_args *args)
556 sigset_t mask, *pmask;
560 if (args->mask != NULL) {
561 error = copyin(args->mask, &lmask, sizeof(l_sigset_t));
564 linux_to_bsd_sigset(&lmask, &mask);
568 return (linux_epoll_wait_common(td, args->epfd, args->events,
569 args->maxevents, args->timeout, pmask));
573 epoll_delete_event(struct thread *td, struct file *epfp, int fd, int filter)
575 struct epoll_copyin_args ciargs;
577 struct kevent_copyops k_ops = { &ciargs,
582 ciargs.changelist = &kev;
583 EV_SET(&kev, fd, filter, EV_DELETE | EV_DISABLE, 0, 0, 0);
585 error = kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL);
588 * here we ignore ENONT, because we don't keep track of events here
596 epoll_delete_all_events(struct thread *td, struct file *epfp, int fd)
600 error1 = epoll_delete_event(td, epfp, fd, EVFILT_READ);
601 error2 = epoll_delete_event(td, epfp, fd, EVFILT_WRITE);
603 /* report any errors we got */
604 return (error1 == 0 ? error2 : error1);
608 eventfd_create(struct thread *td, uint32_t initval, int flags)
610 struct filedesc *fdp;
613 int fflags, fd, error;
616 if ((flags & LINUX_O_CLOEXEC) != 0)
619 fdp = td->td_proc->p_fd;
620 error = falloc(td, &fp, &fd, fflags);
624 efd = malloc(sizeof(*efd), M_EPOLL, M_WAITOK | M_ZERO);
625 efd->efd_flags = flags;
626 efd->efd_count = initval;
627 mtx_init(&efd->efd_lock, "eventfd", NULL, MTX_DEF);
629 knlist_init_mtx(&efd->efd_sel.si_note, &efd->efd_lock);
631 fflags = FREAD | FWRITE;
632 if ((flags & LINUX_O_NONBLOCK) != 0)
635 finit(fp, fflags, DTYPE_LINUXEFD, efd, &eventfdops);
638 td->td_retval[0] = fd;
643 linux_eventfd(struct thread *td, struct linux_eventfd_args *args)
646 return (eventfd_create(td, args->initval, 0));
650 linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args)
653 if ((args->flags & ~(LINUX_O_CLOEXEC|LINUX_O_NONBLOCK|LINUX_EFD_SEMAPHORE)) != 0)
656 return (eventfd_create(td, args->initval, args->flags));
660 eventfd_close(struct file *fp, struct thread *td)
665 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
668 seldrain(&efd->efd_sel);
669 knlist_destroy(&efd->efd_sel.si_note);
671 fp->f_ops = &badfileops;
672 mtx_destroy(&efd->efd_lock);
679 eventfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
680 int flags, struct thread *td)
687 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
690 if (uio->uio_resid < sizeof(eventfd_t))
694 mtx_lock(&efd->efd_lock);
696 if (efd->efd_count == 0) {
697 if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
698 mtx_unlock(&efd->efd_lock);
701 error = mtx_sleep(&efd->efd_count, &efd->efd_lock, PCATCH, "lefdrd", 0);
706 if ((efd->efd_flags & LINUX_EFD_SEMAPHORE) != 0) {
710 count = efd->efd_count;
713 KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
714 selwakeup(&efd->efd_sel);
715 wakeup(&efd->efd_count);
716 mtx_unlock(&efd->efd_lock);
717 error = uiomove(&count, sizeof(eventfd_t), uio);
719 mtx_unlock(&efd->efd_lock);
725 eventfd_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
726 int flags, struct thread *td)
733 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
736 if (uio->uio_resid < sizeof(eventfd_t))
739 error = uiomove(&count, sizeof(eventfd_t), uio);
742 if (count == UINT64_MAX)
745 mtx_lock(&efd->efd_lock);
747 if (UINT64_MAX - efd->efd_count <= count) {
748 if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
749 mtx_unlock(&efd->efd_lock);
752 error = mtx_sleep(&efd->efd_count, &efd->efd_lock,
753 PCATCH, "lefdwr", 0);
758 efd->efd_count += count;
759 KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
760 selwakeup(&efd->efd_sel);
761 wakeup(&efd->efd_count);
763 mtx_unlock(&efd->efd_lock);
769 eventfd_poll(struct file *fp, int events, struct ucred *active_cred,
776 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
779 mtx_lock(&efd->efd_lock);
780 if ((events & (POLLIN|POLLRDNORM)) && efd->efd_count > 0)
781 revents |= events & (POLLIN|POLLRDNORM);
782 if ((events & (POLLOUT|POLLWRNORM)) && UINT64_MAX - 1 > efd->efd_count)
783 revents |= events & (POLLOUT|POLLWRNORM);
785 selrecord(td, &efd->efd_sel);
786 mtx_unlock(&efd->efd_lock);
793 eventfd_kqfilter(struct file *fp, struct knote *kn)
798 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
801 mtx_lock(&efd->efd_lock);
802 switch (kn->kn_filter) {
804 kn->kn_fop = &eventfd_rfiltops;
807 kn->kn_fop = &eventfd_wfiltops;
810 mtx_unlock(&efd->efd_lock);
815 knlist_add(&efd->efd_sel.si_note, kn, 1);
816 mtx_unlock(&efd->efd_lock);
822 filt_eventfddetach(struct knote *kn)
824 struct eventfd *efd = kn->kn_hook;
826 mtx_lock(&efd->efd_lock);
827 knlist_remove(&efd->efd_sel.si_note, kn, 1);
828 mtx_unlock(&efd->efd_lock);
833 filt_eventfdread(struct knote *kn, long hint)
835 struct eventfd *efd = kn->kn_hook;
838 mtx_assert(&efd->efd_lock, MA_OWNED);
839 ret = (efd->efd_count > 0);
846 filt_eventfdwrite(struct knote *kn, long hint)
848 struct eventfd *efd = kn->kn_hook;
851 mtx_assert(&efd->efd_lock, MA_OWNED);
852 ret = (UINT64_MAX - 1 > efd->efd_count);
859 eventfd_truncate(struct file *fp, off_t length, struct ucred *active_cred,
868 eventfd_ioctl(struct file *fp, u_long cmd, void *data,
869 struct ucred *active_cred, struct thread *td)
877 eventfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred,