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;
261 /* flags related to what event is registered */
262 if ((levents & LINUX_EPOLL_EVRD) != 0) {
263 EV_SET(kevent++, fd, EVFILT_READ, *kev_flags, 0, 0, 0);
266 if ((levents & LINUX_EPOLL_EVWR) != 0) {
267 EV_SET(kevent++, fd, EVFILT_WRITE, *kev_flags, 0, 0, 0);
271 if ((levents & ~(LINUX_EPOLL_EVSUP)) != 0) {
275 KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
276 KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n"));
278 LINUX_PEM_XLOCK(pem);
279 if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) {
280 pem->flags |= LINUX_XUNSUP_EPOLL;
281 LINUX_PEM_XUNLOCK(pem);
282 linux_msg(td, "epoll_ctl unsupported flags: 0x%x\n",
285 LINUX_PEM_XUNLOCK(pem);
293 * Structure converting function from kevent to epoll. In a case
294 * this is called on error in registration we store the error in
295 * event->data and pick it up later in linux_epoll_ctl().
298 kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event)
301 if ((kevent->flags & EV_ERROR) != 0) {
302 l_event->events = LINUX_EPOLLERR;
306 switch (kevent->filter) {
308 l_event->events = LINUX_EPOLLIN|LINUX_EPOLLRDNORM|LINUX_EPOLLPRI;
311 l_event->events = LINUX_EPOLLOUT|LINUX_EPOLLWRNORM;
317 * Copyout callback used by kevent. This converts kevent
318 * events to epoll events and copies them back to the
319 * userspace. This is also called on error on registering
323 epoll_kev_copyout(void *arg, struct kevent *kevp, int count)
325 struct epoll_copyout_args *args;
326 struct linux_pemuldata *pem;
327 struct epoll_emuldata *emd;
328 struct epoll_event *eep;
331 args = (struct epoll_copyout_args*) arg;
332 eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO);
334 pem = pem_find(args->p);
335 KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
336 LINUX_PEM_SLOCK(pem);
338 KASSERT(emd != NULL, ("epoll proc epolldata not found.\n"));
340 for (i = 0; i < count; i++) {
341 kevent_to_epoll(&kevp[i], &eep[i]);
344 KASSERT(fd <= emd->fdc, ("epoll user data vector"
345 " is too small.\n"));
346 eep[i].data = emd->udata[fd];
348 LINUX_PEM_SUNLOCK(pem);
350 error = copyout(eep, args->leventlist, count * sizeof(*eep));
352 args->leventlist += count;
353 args->count += count;
354 } else if (args->error == 0)
362 * Copyin callback used by kevent. This copies already
363 * converted filters from kernel memory to the kevent
364 * internal kernel memory. Hence the memcpy instead of
368 epoll_kev_copyin(void *arg, struct kevent *kevp, int count)
370 struct epoll_copyin_args *args;
372 args = (struct epoll_copyin_args*) arg;
374 memcpy(kevp, args->changelist, count * sizeof(*kevp));
375 args->changelist += count;
381 * Load epoll filter, convert it to kevent filter
382 * and load it into kevent subsystem.
385 linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args)
387 struct file *epfp, *fp;
388 struct epoll_copyin_args ciargs;
389 struct kevent kev[2];
390 struct kevent_copyops k_ops = { &ciargs,
393 struct epoll_event le;
399 if (args->op != LINUX_EPOLL_CTL_DEL) {
400 error = copyin(args->event, &le, sizeof(le));
405 error = fget(td, args->epfd,
406 cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &epfp);
409 if (epfp->f_type != DTYPE_KQUEUE)
412 /* Protect user data vector from incorrectly supplied fd. */
413 error = fget(td, args->fd, cap_rights_init(&rights, CAP_POLL_EVENT), &fp);
417 /* Linux disallows spying on himself */
423 ciargs.changelist = kev;
426 case LINUX_EPOLL_CTL_MOD:
428 * We don't memorize which events were set for this FD
429 * on this level, so just delete all we could have set:
430 * EVFILT_READ and EVFILT_WRITE, ignoring any errors
432 error = epoll_delete_all_events(td, epfp, args->fd);
437 case LINUX_EPOLL_CTL_ADD:
438 kev_flags = EV_ADD | EV_ENABLE;
441 case LINUX_EPOLL_CTL_DEL:
442 /* CTL_DEL means unregister this fd with this epoll */
443 error = epoll_delete_all_events(td, epfp, args->fd);
451 error = epoll_to_kevent(td, epfp, args->fd, &le, &kev_flags,
456 epoll_fd_install(td, args->fd, le.data);
458 error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL);
469 * Wait for a filter to be triggered on the epoll file descriptor.
472 linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args)
475 struct timespec ts, *tsp;
477 struct epoll_copyout_args coargs;
478 struct kevent_copyops k_ops = { &coargs,
483 if (args->maxevents <= 0 || args->maxevents > LINUX_MAX_EVENTS)
486 error = fget(td, args->epfd,
487 cap_rights_init(&rights, CAP_KQUEUE_EVENT), &epfp);
491 coargs.leventlist = args->events;
492 coargs.p = td->td_proc;
496 if (args->timeout != -1) {
497 if (args->timeout < 0) {
501 /* Convert from milliseconds to timespec. */
502 ts.tv_sec = args->timeout / 1000;
503 ts.tv_nsec = (args->timeout % 1000) * 1000000;
509 error = kern_kevent_fp(td, epfp, 0, args->maxevents, &k_ops, tsp);
510 if (error == 0 && coargs.error != 0)
511 error = coargs.error;
514 * kern_kevent might return ENOMEM which is not expected from epoll_wait.
515 * Maybe we should translate that but I don't think it matters at all.
518 td->td_retval[0] = coargs.count;
525 epoll_delete_event(struct thread *td, struct file *epfp, int fd, int filter)
527 struct epoll_copyin_args ciargs;
529 struct kevent_copyops k_ops = { &ciargs,
534 ciargs.changelist = &kev;
535 EV_SET(&kev, fd, filter, EV_DELETE | EV_DISABLE, 0, 0, 0);
537 error = kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL);
540 * here we ignore ENONT, because we don't keep track of events here
548 epoll_delete_all_events(struct thread *td, struct file *epfp, int fd)
552 error1 = epoll_delete_event(td, epfp, fd, EVFILT_READ);
553 error2 = epoll_delete_event(td, epfp, fd, EVFILT_WRITE);
555 /* report any errors we got */
556 return (error1 == 0 ? error2 : error1);
560 eventfd_create(struct thread *td, uint32_t initval, int flags)
562 struct filedesc *fdp;
565 int fflags, fd, error;
568 if ((flags & LINUX_O_CLOEXEC) != 0)
571 fdp = td->td_proc->p_fd;
572 error = falloc(td, &fp, &fd, fflags);
576 efd = malloc(sizeof(*efd), M_EPOLL, M_WAITOK | M_ZERO);
577 efd->efd_flags = flags;
578 efd->efd_count = initval;
579 mtx_init(&efd->efd_lock, "eventfd", NULL, MTX_DEF);
581 knlist_init_mtx(&efd->efd_sel.si_note, &efd->efd_lock);
583 fflags = FREAD | FWRITE;
584 if ((flags & LINUX_O_NONBLOCK) != 0)
587 finit(fp, fflags, DTYPE_LINUXEFD, efd, &eventfdops);
590 td->td_retval[0] = fd;
595 linux_eventfd(struct thread *td, struct linux_eventfd_args *args)
598 return (eventfd_create(td, args->initval, 0));
602 linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args)
605 if ((args->flags & ~(LINUX_O_CLOEXEC|LINUX_O_NONBLOCK|LINUX_EFD_SEMAPHORE)) != 0)
608 return (eventfd_create(td, args->initval, args->flags));
612 eventfd_close(struct file *fp, struct thread *td)
617 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
620 seldrain(&efd->efd_sel);
621 knlist_destroy(&efd->efd_sel.si_note);
623 fp->f_ops = &badfileops;
624 mtx_destroy(&efd->efd_lock);
631 eventfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
632 int flags, struct thread *td)
639 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
642 if (uio->uio_resid < sizeof(eventfd_t))
646 mtx_lock(&efd->efd_lock);
648 if (efd->efd_count == 0) {
649 if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
650 mtx_unlock(&efd->efd_lock);
653 error = mtx_sleep(&efd->efd_count, &efd->efd_lock, PCATCH, "lefdrd", 0);
658 if ((efd->efd_flags & LINUX_EFD_SEMAPHORE) != 0) {
662 count = efd->efd_count;
665 KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
666 selwakeup(&efd->efd_sel);
667 wakeup(&efd->efd_count);
668 mtx_unlock(&efd->efd_lock);
669 error = uiomove(&count, sizeof(eventfd_t), uio);
671 mtx_unlock(&efd->efd_lock);
677 eventfd_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
678 int flags, struct thread *td)
685 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
688 if (uio->uio_resid < sizeof(eventfd_t))
691 error = uiomove(&count, sizeof(eventfd_t), uio);
694 if (count == UINT64_MAX)
697 mtx_lock(&efd->efd_lock);
699 if (UINT64_MAX - efd->efd_count <= count) {
700 if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
701 mtx_unlock(&efd->efd_lock);
704 error = mtx_sleep(&efd->efd_count, &efd->efd_lock,
705 PCATCH, "lefdwr", 0);
710 efd->efd_count += count;
711 KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
712 selwakeup(&efd->efd_sel);
713 wakeup(&efd->efd_count);
715 mtx_unlock(&efd->efd_lock);
721 eventfd_poll(struct file *fp, int events, struct ucred *active_cred,
728 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
731 mtx_lock(&efd->efd_lock);
732 if ((events & (POLLIN|POLLRDNORM)) && efd->efd_count > 0)
733 revents |= events & (POLLIN|POLLRDNORM);
734 if ((events & (POLLOUT|POLLWRNORM)) && UINT64_MAX - 1 > efd->efd_count)
735 revents |= events & (POLLOUT|POLLWRNORM);
737 selrecord(td, &efd->efd_sel);
738 mtx_unlock(&efd->efd_lock);
745 eventfd_kqfilter(struct file *fp, struct knote *kn)
750 if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
753 mtx_lock(&efd->efd_lock);
754 switch (kn->kn_filter) {
756 kn->kn_fop = &eventfd_rfiltops;
759 kn->kn_fop = &eventfd_wfiltops;
762 mtx_unlock(&efd->efd_lock);
767 knlist_add(&efd->efd_sel.si_note, kn, 1);
768 mtx_unlock(&efd->efd_lock);
774 filt_eventfddetach(struct knote *kn)
776 struct eventfd *efd = kn->kn_hook;
778 mtx_lock(&efd->efd_lock);
779 knlist_remove(&efd->efd_sel.si_note, kn, 1);
780 mtx_unlock(&efd->efd_lock);
785 filt_eventfdread(struct knote *kn, long hint)
787 struct eventfd *efd = kn->kn_hook;
790 mtx_assert(&efd->efd_lock, MA_OWNED);
791 ret = (efd->efd_count > 0);
798 filt_eventfdwrite(struct knote *kn, long hint)
800 struct eventfd *efd = kn->kn_hook;
803 mtx_assert(&efd->efd_lock, MA_OWNED);
804 ret = (UINT64_MAX - 1 > efd->efd_count);
811 eventfd_truncate(struct file *fp, off_t length, struct ucred *active_cred,
820 eventfd_ioctl(struct file *fp, u_long cmd, void *data,
821 struct ucred *active_cred, struct thread *td)
829 eventfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred,