aboutsummaryrefslogtreecommitdiff
blob: f288bac477df8c846c6b2498376f5a67f5907e80 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
/* Copyright (C) 2004-2020 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contribute by Ulrich Drepper <drepper@redhat.com>, 2004.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <https://www.gnu.org/licenses/>.  */

#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <mqueue.h>
#include <pthread.h>
#include <signal.h>
#include <stdlib.h>
#include <string.h>
#include <sysdep.h>
#include <unistd.h>
#include <sys/socket.h>
#include <not-cancel.h>
#include <nptl/pthreadP.h>


#ifdef __NR_mq_notify

/* Defined in the kernel headers: */
#define NOTIFY_COOKIE_LEN	32	/* Length of the cookie used.  */
#define NOTIFY_WOKENUP		1	/* Code for notifcation.  */
#define NOTIFY_REMOVED		2	/* Code for closed message queue
					   of de-notifcation.  */


/* Data structure for the queued notification requests.  */
union notify_data
{
  struct
  {
    void (*fct) (union sigval);	/* The function to run.  */
    union sigval param;		/* The parameter to pass.  */
    pthread_attr_t *attr;	/* Attributes to create the thread with.  */
    /* NB: on 64-bit machines the struct as a size of 24 bytes.  Which means
       byte 31 can still be used for returning the status.  */
  };
  char raw[NOTIFY_COOKIE_LEN];
};


/* Keep track of the initialization.  */
static pthread_once_t once = PTHREAD_ONCE_INIT;


/* The netlink socket.  */
static int netlink_socket = -1;


/* Barrier used to make sure data passed to the new thread is not
   resused by the parent.  */
static pthread_barrier_t notify_barrier;


/* Modify the signal mask.  We move this into a separate function so
   that the stack space needed for sigset_t is not deducted from what
   the thread can use.  */
static int
__attribute__ ((noinline))
change_sigmask (int how, sigset_t *oss)
{
  sigset_t ss;
  sigfillset (&ss);
  return pthread_sigmask (how, &ss, oss);
}


/* The function used for the notification.  */
static void *
notification_function (void *arg)
{
  /* Copy the function and parameter so that the parent thread can go
     on with its life.  */
  volatile union notify_data *data = (volatile union notify_data *) arg;
  void (*fct) (union sigval) = data->fct;
  union sigval param = data->param;

  /* Let the parent go.  */
  (void) __pthread_barrier_wait (&notify_barrier);

  /* Make the thread detached.  */
  (void) pthread_detach (pthread_self ());

  /* The parent thread has all signals blocked.  This is probably a
     bit surprising for this thread.  So we unblock all of them.  */
  (void) change_sigmask (SIG_UNBLOCK, NULL);

  /* Now run the user code.  */
  fct (param);

  /* And we are done.  */
  return NULL;
}


/* Helper thread.  */
static void *
helper_thread (void *arg)
{
  while (1)
    {
      union notify_data data;

      ssize_t n = __recv (netlink_socket, &data, sizeof (data),
			  MSG_NOSIGNAL | MSG_WAITALL);
      if (n < NOTIFY_COOKIE_LEN)
	continue;

      if (data.raw[NOTIFY_COOKIE_LEN - 1] == NOTIFY_WOKENUP)
	{
	  /* Just create the thread as instructed.  There is no way to
	     report a problem with creating a thread.  */
	  pthread_t th;
	  if (__builtin_expect (pthread_create (&th, data.attr,
						notification_function, &data)
				== 0, 0))
	    /* Since we passed a pointer to DATA to the new thread we have
	       to wait until it is done with it.  */
	    (void) __pthread_barrier_wait (&notify_barrier);
	}
      else if (data.raw[NOTIFY_COOKIE_LEN - 1] == NOTIFY_REMOVED)
	/* The only state we keep is the copy of the thread attributes.  */
	free (data.attr);
    }
  return NULL;
}


static void
reset_once (void)
{
  once = PTHREAD_ONCE_INIT;
}


static void
init_mq_netlink (void)
{
  /* This code might be called a second time after fork().  The file
     descriptor is inherited from the parent.  */
  if (netlink_socket == -1)
    {
      /* Just a normal netlink socket, not bound.  */
      netlink_socket = __socket (AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, 0);
      /* No need to do more if we have no socket.  */
      if (netlink_socket == -1)
	return;
    }

  int err = 1;

  /* Initialize the barrier.  */
  if (__builtin_expect (__pthread_barrier_init (&notify_barrier, NULL, 2) == 0,
			0))
    {
      /* Create the helper thread.  */
      pthread_attr_t attr;
      (void) pthread_attr_init (&attr);
      (void) pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);
      /* We do not need much stack space, the bare minimum will be enough.  */
      (void) pthread_attr_setstacksize (&attr, __pthread_get_minstack (&attr));

      /* Temporarily block all signals so that the newly created
	 thread inherits the mask.  */
      sigset_t oss;
      int have_no_oss = change_sigmask (SIG_BLOCK, &oss);

      pthread_t th;
      err = pthread_create (&th, &attr, helper_thread, NULL);

      /* Reset the signal mask.  */
      if (!have_no_oss)
	pthread_sigmask (SIG_SETMASK, &oss, NULL);

      (void) pthread_attr_destroy (&attr);

      if (err == 0)
	{
	  static int added_atfork;

	  if (added_atfork == 0
	      && pthread_atfork (NULL, NULL, reset_once) != 0)
	    {
	      /* The child thread will call recv() which is a
		 cancellation point.  */
	      (void) pthread_cancel (th);
	      err = 1;
	    }
	  else
	    added_atfork = 1;
	}
    }

  if (err != 0)
    {
      __close_nocancel_nostatus (netlink_socket);
      netlink_socket = -1;
    }
}


/* Register notification upon message arrival to an empty message queue
   MQDES.  */
int
mq_notify (mqd_t mqdes, const struct sigevent *notification)
{
  /* Make sure the type is correctly defined.  */
  assert (sizeof (union notify_data) == NOTIFY_COOKIE_LEN);

  /* Special treatment needed for SIGEV_THREAD.  */
  if (notification == NULL || notification->sigev_notify != SIGEV_THREAD)
    return INLINE_SYSCALL (mq_notify, 2, mqdes, notification);

  /* The kernel cannot directly start threads.  This will have to be
     done at userlevel.  Since we cannot start threads from signal
     handlers we have to create a dedicated thread which waits for
     notifications for arriving messages and creates threads in
     response.  */

  /* Initialize only once.  */
  pthread_once (&once, init_mq_netlink);

  /* If we cannot create the netlink socket we cannot provide
     SIGEV_THREAD support.  */
  if (__glibc_unlikely (netlink_socket == -1))
    {
      __set_errno (ENOSYS);
      return -1;
    }

  /* Create the cookie.  It will hold almost all the state.  */
  union notify_data data;
  memset (&data, '\0', sizeof (data));
  data.fct = notification->sigev_notify_function;
  data.param = notification->sigev_value;

  if (notification->sigev_notify_attributes != NULL)
    {
      /* The thread attribute has to be allocated separately.  */
      data.attr = (pthread_attr_t *) malloc (sizeof (pthread_attr_t));
      if (data.attr == NULL)
	return -1;

      memcpy (data.attr, notification->sigev_notify_attributes,
	      sizeof (pthread_attr_t));
    }

  /* Construct the new request.  */
  struct sigevent se;
  se.sigev_notify = SIGEV_THREAD;
  se.sigev_signo = netlink_socket;
  se.sigev_value.sival_ptr = &data;

  /* Tell the kernel.  */
  int retval = INLINE_SYSCALL (mq_notify, 2, mqdes, &se);

  /* If it failed, free the allocated memory.  */
  if (__glibc_unlikely (retval != 0))
    free (data.attr);

  return retval;
}

#else
# include <rt/mq_notify.c>
#endif