sys/emulation/linux/linux_epoll.c

   1 /*-
   2  * Copyright (c) 2007 Roman Divacky
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  */
  26
  27 #include "opt_compat.h"
  28
  29 #include <sys/param.h>
  30 #include <sys/systm.h>
  31 #include <sys/kern_syscall.h>
  32 #include <sys/event.h>
  33 #include <sys/lock.h>
  34 #include <sys/mplock2.h>
  35 #include <sys/malloc.h>
  36 #include <sys/ptrace.h>
  37 #include <sys/proc.h>
  38 #include <sys/signalvar.h>
  39 #include <sys/sysent.h>
  40 #include <sys/sysproto.h>
  41
  42 #include <vm/vm.h>
  43 #include <vm/vm_param.h>
  44 #include <vm/vm_page.h>
  45 #include <vm/vm_extern.h>
  46 #include <sys/exec.h>
  47 #include <sys/kernel.h>
  48 #include <sys/module.h>
  49 #include <machine/cpu.h>
  50
  51 #include "i386/linux.h"
  52 #include "i386/linux_proto.h"
  53 #include "linux_signal.h"
  54 #include "linux_util.h"
  55 #include "linux_epoll.h"
  56
  57
  58 /* Create a new epoll file descriptor. */
  59 int
  60 sys_linux_epoll_create(struct linux_epoll_create_args *args)
  61 {
  62         struct kqueue_args k_args;
  63
  64         if (args->size <= 0)
  65                 return (EINVAL);
  66         /* args->size is unused. Linux ignores it as well. */
  67
  68         return (sys_kqueue(&k_args));
  69 }
  70
  71 /* Structure converting function from epoll to kevent. */
  72 static void
  73 linux_epoll_to_kevent(int fd, struct linux_epoll_event *event, struct kevent *kevent)
  74 {
  75         int filter = 0;
  76         int flags = kevent->flags;
  77
  78         if (event->events & LINUX_EPOLLIN)
  79                 filter |= EVFILT_READ;
  80         if (event->events & LINUX_EPOLLOUT)
  81                 filter |= EVFILT_WRITE;
  82         if (event->events & LINUX_EPOLLPRI)
  83                 filter |= EVFILT_READ;
  84         if (event->events & LINUX_EPOLLET)
  85                 flags |= EV_CLEAR;
  86         if (event->events & LINUX_EPOLLONESHOT)
  87                 flags |= EV_ONESHOT;
  88
  89         EV_SET(kevent, fd, filter, flags, 0, 0, NULL);
  90 }
  91
  92 /*
  93  * Structure converting function from kevent to epoll. In a case
  94  * this is called on error in registration we store the error in
  95  * event->data and pick it up later in linux_epoll_ctl().
  96  */
  97 static void
  98 linux_kevent_to_epoll(struct kevent *kevent, struct linux_epoll_event *event)
  99 {
 100         if (kevent->flags & EV_ERROR) {
 101                 event->data = kevent->data;
 102                 return;
 103         }
 104         switch (kevent->filter) {
 105                 case EVFILT_READ:
 106                         if (kevent->data > 0)
 107                                 event->events = LINUX_EPOLLIN;
 108                         event->data = kevent->ident;
 109                 break;
 110                 case EVFILT_WRITE:
 111                         if (kevent->data > 0)
 112                                 event->events = LINUX_EPOLLOUT;
 113                         event->data = kevent->ident;
 114                 break;
 115         }
 116 }
 117
 118 /*
 119  * Copyout callback used by kevent. This converts kevent
 120  * events to epoll events and copies them back to the
 121  * userspace. This is also called on error on registering
 122  * of the filter.
 123  */
 124 static int
 125 linux_kev_copyout(void *arg, struct kevent *kevp, int count)
 126 {
 127         struct kevent_args *uap;
 128         struct linux_epoll_event *eep;
 129         int error, i;
 130
 131         uap = (struct kevent_args*) arg;
 132
 133         eep = kmalloc(sizeof(*eep) * count, M_TEMP, M_WAITOK | M_ZERO);
 134
 135         for (i = 0; i < count; i++) {
 136                 linux_kevent_to_epoll(&kevp[i], &eep[i]);
 137         }
 138
 139         error = copyout(eep, uap->eventlist, count * sizeof(*eep));
 140         if (error)
 141                 uap->eventlist = (struct kevent *)((char *)uap->eventlist + count * sizeof(*eep));
 142
 143         kfree(eep, M_TEMP);
 144         return (0);
 145 }
 146
 147 /*
 148  * Copyin callback used by kevent. This copies already
 149  * converted filters to the kevent internal memory.
 150  */
 151 static int
 152 linux_kev_copyin(void *arg, struct kevent *kevp, int count)
 153 {
 154         struct kevent_args *uap;
 155
 156         uap = (struct kevent_args*) arg;
 157
 158         memcpy(kevp, uap->changelist, count * sizeof(*kevp));
 159
 160         uap->changelist += count;
 161
 162         return (0);
 163 }
 164
 165 /*
 166  * Load epoll filter, convert it to kevent filter
 167  * and load it into kevent subsystem.
 168  */
 169 int
 170 sys_linux_epoll_ctl(struct linux_epoll_ctl_args *args)
 171 {
 172         struct kevent_args k_args;
 173         struct kevent kev;
 174         struct linux_epoll_event le;
 175         int error;
 176
 177         error = copyin(args->event, &le, sizeof(le));
 178         if (error)
 179                 return (error);
 180 #ifdef DEBUG
 181         if (ldebug(epoll_ctl))
 182                 kprintf(ARGS(epoll_ctl,"%i, %i, %i, %u"), args->epfd, args->op,
 183                         args->fd, le.events);
 184 #endif
 185         k_args.fd = args->epfd;
 186         k_args.changelist = &kev;
 187         /* The epoll can register only 1 filter at once. */
 188         k_args.nchanges = 1;
 189         k_args.eventlist = NULL;
 190         k_args.nevents = 0;
 191         k_args.timeout = NULL;
 192
 193         switch (args->op) {
 194         case LINUX_EPOLL_CTL_ADD:
 195                         kev.flags = EV_ADD | EV_ENABLE;
 196                 break;
 197         case LINUX_EPOLL_CTL_MOD:
 198                         /* TODO: DELETE && ADD maybe? */
 199                         return (EINVAL);
 200                 break;
 201         case LINUX_EPOLL_CTL_DEL:
 202                         kev.flags = EV_DELETE | EV_DISABLE;
 203                 break;
 204         }
 205         linux_epoll_to_kevent(args->fd, &le, &kev);
 206
 207         error = kern_kevent(args->epfd, 1, 0, &k_args, linux_kev_copyin,
 208             linux_kev_copyout, NULL);
 209         /* Check if there was an error during registration. */
 210         if (error == 0 && k_args.sysmsg_result != 0) {
 211                 /* The copyout callback stored the error there. */
 212                 error = le.data;
 213         }
 214
 215         return (error);
 216 }
 217
 218 /*
 219  * Wait for a filter to be triggered on the epoll file descriptor. */
 220 int
 221 sys_linux_epoll_wait(struct linux_epoll_wait_args *args)
 222 {
 223         struct timespec ts;
 224         struct kevent_args k_args;
 225         int error;
 226
 227         /* Convert from miliseconds to timespec. */
 228         ts.tv_sec = args->timeout / 1000000;
 229         ts.tv_nsec = (args->timeout % 1000000) * 1000;
 230
 231         k_args.fd = args->epfd;
 232         k_args.changelist = NULL;
 233         k_args.nchanges = 0;
 234         /*
 235          * We don't mind the bogus type-cast because
 236          * our copyout function knows about this and
 237          * handles it correctly.
 238          */
 239         k_args.eventlist = (struct kevent *)args->events;
 240         k_args.nevents = args->maxevents;
 241         k_args.timeout = &ts;
 242
 243         error = kern_kevent(args->epfd, 0, args->maxevents, &k_args,
 244             linux_kev_copyin, linux_kev_copyout, &ts);
 245
 246         /* translation? */
 247         return (error);
 248 }