msg.c 24.1 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2
/*
 * linux/ipc/msg.c
3
 * Copyright (C) 1992 Krishna Balasubramanian
Linus Torvalds's avatar
Linus Torvalds committed
4 5 6 7 8 9 10 11 12 13 14
 *
 * Removed all the remaining kerneld mess
 * Catch the -EFAULT stuff properly
 * Use GFP_KERNEL for messages as in 1.2
 * Fixed up the unchecked user space derefs
 * Copyright (C) 1998 Alan Cox & Andi Kleen
 *
 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
 *
 * mostly rewritten, threaded and wake-one semantics added
 * MSGMAX limit removed, sysctl's added
15
 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
16 17 18
 *
 * support for audit of ipc object properties and permission changes
 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 20 21 22
 *
 * namespaces support
 * OpenVZ, SWsoft Inc.
 * Pavel Emelianov <xemul@openvz.org>
Linus Torvalds's avatar
Linus Torvalds committed
23 24
 */

25
#include <linux/capability.h>
Linus Torvalds's avatar
Linus Torvalds committed
26 27 28
#include <linux/msg.h>
#include <linux/spinlock.h>
#include <linux/init.h>
29
#include <linux/mm.h>
Linus Torvalds's avatar
Linus Torvalds committed
30 31 32 33 34 35
#include <linux/proc_fs.h>
#include <linux/list.h>
#include <linux/security.h>
#include <linux/sched.h>
#include <linux/syscalls.h>
#include <linux/audit.h>
36
#include <linux/seq_file.h>
Nadia Derbey's avatar
Nadia Derbey committed
37
#include <linux/rwsem.h>
38
#include <linux/nsproxy.h>
39
#include <linux/ipc_namespace.h>
40

Linus Torvalds's avatar
Linus Torvalds committed
41
#include <asm/current.h>
42
#include <linux/uaccess.h>
Linus Torvalds's avatar
Linus Torvalds committed
43 44
#include "util.h"

45
/* one msg_receiver structure for each sleeping receiver */
Linus Torvalds's avatar
Linus Torvalds committed
46
struct msg_receiver {
47 48
	struct list_head	r_list;
	struct task_struct	*r_tsk;
Linus Torvalds's avatar
Linus Torvalds committed
49

50 51 52
	int			r_mode;
	long			r_msgtype;
	long			r_maxsize;
Linus Torvalds's avatar
Linus Torvalds committed
53

54 55 56 57 58 59
	/*
	 * Mark r_msg volatile so that the compiler
	 * does not try to get smart and optimize
	 * it. We rely on this for the lockless
	 * receive algorithm.
	 */
60
	struct msg_msg		*volatile r_msg;
Linus Torvalds's avatar
Linus Torvalds committed
61 62 63 64
};

/* one msg_sender for each sleeping sender */
struct msg_sender {
65 66
	struct list_head	list;
	struct task_struct	*tsk;
Linus Torvalds's avatar
Linus Torvalds committed
67 68 69 70 71 72
};

#define SEARCH_ANY		1
#define SEARCH_EQUAL		2
#define SEARCH_NOTEQUAL		3
#define SEARCH_LESSEQUAL	4
73
#define SEARCH_NUMBER		5
Linus Torvalds's avatar
Linus Torvalds committed
74

75
#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
Linus Torvalds's avatar
Linus Torvalds committed
76

77 78
static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
{
79
	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97

	if (IS_ERR(ipcp))
		return ERR_CAST(ipcp);

	return container_of(ipcp, struct msg_queue, q_perm);
}

static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
							int id)
{
	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);

	if (IS_ERR(ipcp))
		return ERR_CAST(ipcp);

	return container_of(ipcp, struct msg_queue, q_perm);
}

Nadia Derbey's avatar
Nadia Derbey committed
98 99 100 101 102
static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
{
	ipc_rmid(&msg_ids(ns), &s->q_perm);
}

103 104 105 106 107 108 109 110 111
static void msg_rcu_free(struct rcu_head *head)
{
	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
	struct msg_queue *msq = ipc_rcu_to_struct(p);

	security_msg_queue_free(msq);
	ipc_rcu_free(head);
}

Nadia Derbey's avatar
Nadia Derbey committed
112 113 114 115 116
/**
 * newque - Create a new msg queue
 * @ns: namespace
 * @params: ptr to the structure that contains the key and msgflg
 *
117
 * Called with msg_ids.rwsem held (writer)
Nadia Derbey's avatar
Nadia Derbey committed
118
 */
119
static int newque(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds's avatar
Linus Torvalds committed
120 121
{
	struct msg_queue *msq;
122
	int id, retval;
123 124
	key_t key = params->key;
	int msgflg = params->flg;
Linus Torvalds's avatar
Linus Torvalds committed
125

126 127
	msq = ipc_rcu_alloc(sizeof(*msq));
	if (!msq)
Linus Torvalds's avatar
Linus Torvalds committed
128 129
		return -ENOMEM;

130
	msq->q_perm.mode = msgflg & S_IRWXUGO;
Linus Torvalds's avatar
Linus Torvalds committed
131 132 133 134 135
	msq->q_perm.key = key;

	msq->q_perm.security = NULL;
	retval = security_msg_queue_alloc(msq);
	if (retval) {
136
		ipc_rcu_putref(msq, ipc_rcu_free);
Linus Torvalds's avatar
Linus Torvalds committed
137 138 139 140 141 142
		return retval;
	}

	msq->q_stime = msq->q_rtime = 0;
	msq->q_ctime = get_seconds();
	msq->q_cbytes = msq->q_qnum = 0;
143
	msq->q_qbytes = ns->msg_ctlmnb;
Linus Torvalds's avatar
Linus Torvalds committed
144 145 146 147
	msq->q_lspid = msq->q_lrpid = 0;
	INIT_LIST_HEAD(&msq->q_messages);
	INIT_LIST_HEAD(&msq->q_receivers);
	INIT_LIST_HEAD(&msq->q_senders);
Nadia Derbey's avatar
Nadia Derbey committed
148

149 150 151 152 153 154 155
	/* ipc_addid() locks msq upon success. */
	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
	if (id < 0) {
		ipc_rcu_putref(msq, msg_rcu_free);
		return id;
	}

156
	ipc_unlock_object(&msq->q_perm);
157
	rcu_read_unlock();
Linus Torvalds's avatar
Linus Torvalds committed
158

Nadia Derbey's avatar
Nadia Derbey committed
159
	return msq->q_perm.id;
Linus Torvalds's avatar
Linus Torvalds committed
160 161
}

162
static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
Linus Torvalds's avatar
Linus Torvalds committed
163
{
164
	mss->tsk = current;
165
	__set_current_state(TASK_INTERRUPTIBLE);
166
	list_add_tail(&mss->list, &msq->q_senders);
Linus Torvalds's avatar
Linus Torvalds committed
167 168
}

169
static inline void ss_del(struct msg_sender *mss)
Linus Torvalds's avatar
Linus Torvalds committed
170
{
171
	if (mss->list.next != NULL)
Linus Torvalds's avatar
Linus Torvalds committed
172 173 174
		list_del(&mss->list);
}

175
static void ss_wakeup(struct list_head *h, int kill)
Linus Torvalds's avatar
Linus Torvalds committed
176
{
177
	struct msg_sender *mss, *t;
Linus Torvalds's avatar
Linus Torvalds committed
178

179
	list_for_each_entry_safe(mss, t, h, list) {
180 181
		if (kill)
			mss->list.next = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
182 183 184 185
		wake_up_process(mss->tsk);
	}
}

186
static void expunge_all(struct msg_queue *msq, int res)
Linus Torvalds's avatar
Linus Torvalds committed
187
{
188
	struct msg_receiver *msr, *t;
189

190
	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
191
		msr->r_msg = NULL; /* initialize expunge ordering */
Linus Torvalds's avatar
Linus Torvalds committed
192
		wake_up_process(msr->r_tsk);
193 194 195 196 197 198
		/*
		 * Ensure that the wakeup is visible before setting r_msg as
		 * the receiving end depends on it: either spinning on a nil,
		 * or dealing with -EAGAIN cases. See lockless receive part 1
		 * and 2 in do_msgrcv().
		 */
199
		smp_wmb(); /* barrier (B) */
Linus Torvalds's avatar
Linus Torvalds committed
200 201 202
		msr->r_msg = ERR_PTR(res);
	}
}
203 204 205

/*
 * freeque() wakes up waiters on the sender and receiver waiting queue,
Nadia Derbey's avatar
Nadia Derbey committed
206 207
 * removes the message queue from message queue ID IDR, and cleans up all the
 * messages associated with this queue.
Linus Torvalds's avatar
Linus Torvalds committed
208
 *
209 210
 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
 * before freeque() is called. msg_ids.rwsem remains locked on exit.
Linus Torvalds's avatar
Linus Torvalds committed
211
 */
212
static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds's avatar
Linus Torvalds committed
213
{
214
	struct msg_msg *msg, *t;
215
	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
Linus Torvalds's avatar
Linus Torvalds committed
216

217 218
	expunge_all(msq, -EIDRM);
	ss_wakeup(&msq->q_senders, 1);
Nadia Derbey's avatar
Nadia Derbey committed
219
	msg_rmid(ns, msq);
220 221
	ipc_unlock_object(&msq->q_perm);
	rcu_read_unlock();
222

223
	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
224
		atomic_dec(&ns->msg_hdrs);
Linus Torvalds's avatar
Linus Torvalds committed
225 226
		free_msg(msg);
	}
227
	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
228
	ipc_rcu_putref(msq, msg_rcu_free);
Linus Torvalds's avatar
Linus Torvalds committed
229 230
}

Nadia Derbey's avatar
Nadia Derbey committed
231
/*
232
 * Called with msg_ids.rwsem and ipcp locked.
Nadia Derbey's avatar
Nadia Derbey committed
233
 */
Nadia Derbey's avatar
Nadia Derbey committed
234
static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
235
{
Nadia Derbey's avatar
Nadia Derbey committed
236 237 238
	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);

	return security_msg_queue_associate(msq, msgflg);
239 240
}

241
SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
Linus Torvalds's avatar
Linus Torvalds committed
242
{
243
	struct ipc_namespace *ns;
Mathias Krause's avatar
Mathias Krause committed
244 245 246 247
	static const struct ipc_ops msg_ops = {
		.getnew = newque,
		.associate = msg_security,
	};
248
	struct ipc_params msg_params;
249 250

	ns = current->nsproxy->ipc_ns;
Nadia Derbey's avatar
Nadia Derbey committed
251

252 253
	msg_params.key = key;
	msg_params.flg = msgflg;
254

255
	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
Linus Torvalds's avatar
Linus Torvalds committed
256 257
}

258 259
static inline unsigned long
copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
Linus Torvalds's avatar
Linus Torvalds committed
260
{
Manfred Spraul's avatar
Manfred Spraul committed
261
	switch (version) {
Linus Torvalds's avatar
Linus Torvalds committed
262
	case IPC_64:
263
		return copy_to_user(buf, in, sizeof(*in));
Linus Torvalds's avatar
Linus Torvalds committed
264
	case IPC_OLD:
265
	{
Linus Torvalds's avatar
Linus Torvalds committed
266 267
		struct msqid_ds out;

268
		memset(&out, 0, sizeof(out));
Linus Torvalds's avatar
Linus Torvalds committed
269 270 271 272 273 274 275

		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);

		out.msg_stime		= in->msg_stime;
		out.msg_rtime		= in->msg_rtime;
		out.msg_ctime		= in->msg_ctime;

276 277
		if (in->msg_cbytes > USHRT_MAX)
			out.msg_cbytes	= USHRT_MAX;
Linus Torvalds's avatar
Linus Torvalds committed
278 279 280 281
		else
			out.msg_cbytes	= in->msg_cbytes;
		out.msg_lcbytes		= in->msg_cbytes;

282 283
		if (in->msg_qnum > USHRT_MAX)
			out.msg_qnum	= USHRT_MAX;
Linus Torvalds's avatar
Linus Torvalds committed
284 285 286
		else
			out.msg_qnum	= in->msg_qnum;

287 288
		if (in->msg_qbytes > USHRT_MAX)
			out.msg_qbytes	= USHRT_MAX;
Linus Torvalds's avatar
Linus Torvalds committed
289 290 291 292 293 294 295
		else
			out.msg_qbytes	= in->msg_qbytes;
		out.msg_lqbytes		= in->msg_qbytes;

		out.msg_lspid		= in->msg_lspid;
		out.msg_lrpid		= in->msg_lrpid;

296 297
		return copy_to_user(buf, &out, sizeof(out));
	}
Linus Torvalds's avatar
Linus Torvalds committed
298 299 300 301 302
	default:
		return -EINVAL;
	}
}

303
static inline unsigned long
304
copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
Linus Torvalds's avatar
Linus Torvalds committed
305
{
Manfred Spraul's avatar
Manfred Spraul committed
306
	switch (version) {
Linus Torvalds's avatar
Linus Torvalds committed
307
	case IPC_64:
308
		if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds's avatar
Linus Torvalds committed
309 310 311
			return -EFAULT;
		return 0;
	case IPC_OLD:
312
	{
Linus Torvalds's avatar
Linus Torvalds committed
313 314
		struct msqid_ds tbuf_old;

315
		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
Linus Torvalds's avatar
Linus Torvalds committed
316 317
			return -EFAULT;

Manfred Spraul's avatar
Manfred Spraul committed
318 319 320
		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
Linus Torvalds's avatar
Linus Torvalds committed
321

322
		if (tbuf_old.msg_qbytes == 0)
323
			out->msg_qbytes	= tbuf_old.msg_lqbytes;
Linus Torvalds's avatar
Linus Torvalds committed
324
		else
325
			out->msg_qbytes	= tbuf_old.msg_qbytes;
Linus Torvalds's avatar
Linus Torvalds committed
326 327

		return 0;
328
	}
Linus Torvalds's avatar
Linus Torvalds committed
329 330 331 332 333
	default:
		return -EINVAL;
	}
}

334
/*
335
 * This function handles some msgctl commands which require the rwsem
336
 * to be held in write mode.
337
 * NOTE: no locks must be held, the rwsem is taken inside this function.
338 339 340
 */
static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
		       struct msqid_ds __user *buf, int version)
Linus Torvalds's avatar
Linus Torvalds committed
341 342
{
	struct kern_ipc_perm *ipcp;
343
	struct msqid64_ds uninitialized_var(msqid64);
344 345 346 347
	struct msg_queue *msq;
	int err;

	if (cmd == IPC_SET) {
348
		if (copy_msqid_from_user(&msqid64, buf, version))
349 350 351
			return -EFAULT;
	}

352
	down_write(&msg_ids(ns).rwsem);
353 354
	rcu_read_lock();

355 356
	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
				      &msqid64.msg_perm, msqid64.msg_qbytes);
357 358 359 360
	if (IS_ERR(ipcp)) {
		err = PTR_ERR(ipcp);
		goto out_unlock1;
	}
361

362
	msq = container_of(ipcp, struct msg_queue, q_perm);
363 364 365

	err = security_msg_queue_msgctl(msq, cmd);
	if (err)
366
		goto out_unlock1;
367 368 369

	switch (cmd) {
	case IPC_RMID:
370
		ipc_lock_object(&msq->q_perm);
371
		/* freeque unlocks the ipc object and rcu */
372 373 374
		freeque(ns, ipcp);
		goto out_up;
	case IPC_SET:
375
		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
376 377
		    !capable(CAP_SYS_RESOURCE)) {
			err = -EPERM;
378
			goto out_unlock1;
379 380
		}

381
		ipc_lock_object(&msq->q_perm);
382 383
		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
		if (err)
384
			goto out_unlock0;
385

386
		msq->q_qbytes = msqid64.msg_qbytes;
387 388 389 390 391 392 393 394 395 396 397 398 399

		msq->q_ctime = get_seconds();
		/* sleeping receivers might be excluded by
		 * stricter permissions.
		 */
		expunge_all(msq, -EAGAIN);
		/* sleeping senders might be able to send
		 * due to a larger queue size.
		 */
		ss_wakeup(&msq->q_senders, 0);
		break;
	default:
		err = -EINVAL;
400
		goto out_unlock1;
401
	}
402 403 404 405 406

out_unlock0:
	ipc_unlock_object(&msq->q_perm);
out_unlock1:
	rcu_read_unlock();
407
out_up:
408
	up_write(&msg_ids(ns).rwsem);
409 410 411
	return err;
}

412 413
static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
			 int cmd, int version, void __user *buf)
414
{
415
	int err;
416
	struct msg_queue *msq;
Linus Torvalds's avatar
Linus Torvalds committed
417 418

	switch (cmd) {
419 420 421
	case IPC_INFO:
	case MSG_INFO:
	{
Linus Torvalds's avatar
Linus Torvalds committed
422 423
		struct msginfo msginfo;
		int max_id;
424

Linus Torvalds's avatar
Linus Torvalds committed
425 426
		if (!buf)
			return -EFAULT;
427

428 429
		/*
		 * We must not return kernel stack data.
Linus Torvalds's avatar
Linus Torvalds committed
430 431 432 433 434 435 436
		 * due to padding, it's not enough
		 * to set all member fields.
		 */
		err = security_msg_queue_msgctl(NULL, cmd);
		if (err)
			return err;

437
		memset(&msginfo, 0, sizeof(msginfo));
438 439 440
		msginfo.msgmni = ns->msg_ctlmni;
		msginfo.msgmax = ns->msg_ctlmax;
		msginfo.msgmnb = ns->msg_ctlmnb;
Linus Torvalds's avatar
Linus Torvalds committed
441 442
		msginfo.msgssz = MSGSSZ;
		msginfo.msgseg = MSGSEG;
443
		down_read(&msg_ids(ns).rwsem);
Linus Torvalds's avatar
Linus Torvalds committed
444
		if (cmd == MSG_INFO) {
445
			msginfo.msgpool = msg_ids(ns).in_use;
446 447
			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
			msginfo.msgtql = atomic_read(&ns->msg_bytes);
Linus Torvalds's avatar
Linus Torvalds committed
448 449 450 451 452
		} else {
			msginfo.msgmap = MSGMAP;
			msginfo.msgpool = MSGPOOL;
			msginfo.msgtql = MSGTQL;
		}
Nadia Derbey's avatar
Nadia Derbey committed
453
		max_id = ipc_get_maxid(&msg_ids(ns));
454
		up_read(&msg_ids(ns).rwsem);
455
		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
Linus Torvalds's avatar
Linus Torvalds committed
456
			return -EFAULT;
457
		return (max_id < 0) ? 0 : max_id;
Linus Torvalds's avatar
Linus Torvalds committed
458
	}
459 460

	case MSG_STAT:
Linus Torvalds's avatar
Linus Torvalds committed
461 462 463 464
	case IPC_STAT:
	{
		struct msqid64_ds tbuf;
		int success_return;
465

Linus Torvalds's avatar
Linus Torvalds committed
466 467 468
		if (!buf)
			return -EFAULT;

469 470 471
		memset(&tbuf, 0, sizeof(tbuf));

		rcu_read_lock();
472
		if (cmd == MSG_STAT) {
473 474 475 476 477
			msq = msq_obtain_object(ns, msqid);
			if (IS_ERR(msq)) {
				err = PTR_ERR(msq);
				goto out_unlock;
			}
Nadia Derbey's avatar
Nadia Derbey committed
478
			success_return = msq->q_perm.id;
Linus Torvalds's avatar
Linus Torvalds committed
479
		} else {
480 481 482 483 484
			msq = msq_obtain_object_check(ns, msqid);
			if (IS_ERR(msq)) {
				err = PTR_ERR(msq);
				goto out_unlock;
			}
Linus Torvalds's avatar
Linus Torvalds committed
485 486
			success_return = 0;
		}
487

Linus Torvalds's avatar
Linus Torvalds committed
488
		err = -EACCES;
489
		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
Linus Torvalds's avatar
Linus Torvalds committed
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
			goto out_unlock;

		err = security_msg_queue_msgctl(msq, cmd);
		if (err)
			goto out_unlock;

		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
		tbuf.msg_stime  = msq->q_stime;
		tbuf.msg_rtime  = msq->q_rtime;
		tbuf.msg_ctime  = msq->q_ctime;
		tbuf.msg_cbytes = msq->q_cbytes;
		tbuf.msg_qnum   = msq->q_qnum;
		tbuf.msg_qbytes = msq->q_qbytes;
		tbuf.msg_lspid  = msq->q_lspid;
		tbuf.msg_lrpid  = msq->q_lrpid;
505 506
		rcu_read_unlock();

Linus Torvalds's avatar
Linus Torvalds committed
507 508 509 510
		if (copy_msqid_to_user(buf, &tbuf, version))
			return -EFAULT;
		return success_return;
	}
511

Linus Torvalds's avatar
Linus Torvalds committed
512
	default:
513
		return -EINVAL;
Linus Torvalds's avatar
Linus Torvalds committed
514 515
	}

516
	return err;
Linus Torvalds's avatar
Linus Torvalds committed
517
out_unlock:
518
	rcu_read_unlock();
Linus Torvalds's avatar
Linus Torvalds committed
519 520 521
	return err;
}

522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
{
	int version;
	struct ipc_namespace *ns;

	if (msqid < 0 || cmd < 0)
		return -EINVAL;

	version = ipc_parse_version(&cmd);
	ns = current->nsproxy->ipc_ns;

	switch (cmd) {
	case IPC_INFO:
	case MSG_INFO:
	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
	case IPC_STAT:
		return msgctl_nolock(ns, msqid, cmd, version, buf);
	case IPC_SET:
	case IPC_RMID:
		return msgctl_down(ns, msqid, cmd, buf, version);
	default:
		return  -EINVAL;
	}
}

547
static int testmsg(struct msg_msg *msg, long type, int mode)
Linus Torvalds's avatar
Linus Torvalds committed
548
{
549 550 551 552 553 554 555 556 557 558
	switch (mode) {
	case SEARCH_ANY:
	case SEARCH_NUMBER:
		return 1;
	case SEARCH_LESSEQUAL:
		if (msg->m_type <= type)
			return 1;
		break;
	case SEARCH_EQUAL:
		if (msg->m_type == type)
Linus Torvalds's avatar
Linus Torvalds committed
559
			return 1;
560 561 562 563 564
		break;
	case SEARCH_NOTEQUAL:
		if (msg->m_type != type)
			return 1;
		break;
Linus Torvalds's avatar
Linus Torvalds committed
565 566 567 568
	}
	return 0;
}

569
static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
Linus Torvalds's avatar
Linus Torvalds committed
570
{
571
	struct msg_receiver *msr, *t;
572

573
	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
574 575 576 577
		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
					       msr->r_msgtype, msr->r_mode)) {

Linus Torvalds's avatar
Linus Torvalds committed
578
			list_del(&msr->r_list);
579
			if (msr->r_maxsize < msg->m_ts) {
580
				/* initialize pipelined send ordering */
Linus Torvalds's avatar
Linus Torvalds committed
581 582
				msr->r_msg = NULL;
				wake_up_process(msr->r_tsk);
583 584
				/* barrier (B) see barrier comment below */
				smp_wmb();
Linus Torvalds's avatar
Linus Torvalds committed
585 586 587
				msr->r_msg = ERR_PTR(-E2BIG);
			} else {
				msr->r_msg = NULL;
588
				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
Linus Torvalds's avatar
Linus Torvalds committed
589 590
				msq->q_rtime = get_seconds();
				wake_up_process(msr->r_tsk);
591 592
				/*
				 * Ensure that the wakeup is visible before
593 594 595 596
				 * setting r_msg, as the receiving can otherwise
				 * exit - once r_msg is set, the receiver can
				 * continue. See lockless receive part 1 and 2
				 * in do_msgrcv(). Barrier (B).
597
				 */
598
				smp_wmb();
Linus Torvalds's avatar
Linus Torvalds committed
599
				msr->r_msg = msg;
600

Linus Torvalds's avatar
Linus Torvalds committed
601 602 603 604
				return 1;
			}
		}
	}
605

Linus Torvalds's avatar
Linus Torvalds committed
606 607 608
	return 0;
}

609 610
long do_msgsnd(int msqid, long mtype, void __user *mtext,
		size_t msgsz, int msgflg)
Linus Torvalds's avatar
Linus Torvalds committed
611 612 613 614
{
	struct msg_queue *msq;
	struct msg_msg *msg;
	int err;
615 616 617
	struct ipc_namespace *ns;

	ns = current->nsproxy->ipc_ns;
618

619
	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
Linus Torvalds's avatar
Linus Torvalds committed
620 621 622 623
		return -EINVAL;
	if (mtype < 1)
		return -EINVAL;

624
	msg = load_msg(mtext, msgsz);
625
	if (IS_ERR(msg))
Linus Torvalds's avatar
Linus Torvalds committed
626 627 628 629 630
		return PTR_ERR(msg);

	msg->m_type = mtype;
	msg->m_ts = msgsz;

631 632
	rcu_read_lock();
	msq = msq_obtain_object_check(ns, msqid);
633 634
	if (IS_ERR(msq)) {
		err = PTR_ERR(msq);
635
		goto out_unlock1;
636
	}
Linus Torvalds's avatar
Linus Torvalds committed
637

638 639
	ipc_lock_object(&msq->q_perm);

Linus Torvalds's avatar
Linus Torvalds committed
640 641 642
	for (;;) {
		struct msg_sender s;

643
		err = -EACCES;
644
		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
645
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
646

647
		/* raced with RMID? */
648
		if (!ipc_valid_object(&msq->q_perm)) {
649 650 651 652
			err = -EIDRM;
			goto out_unlock0;
		}

Linus Torvalds's avatar
Linus Torvalds committed
653 654
		err = security_msg_queue_msgsnd(msq, msg, msgflg);
		if (err)
655
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
656

657
		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
Linus Torvalds's avatar
Linus Torvalds committed
658 659 660 661 662
				1 + msq->q_qnum <= msq->q_qbytes) {
			break;
		}

		/* queue full, wait: */
663 664
		if (msgflg & IPC_NOWAIT) {
			err = -EAGAIN;
665
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
666
		}
667

668
		/* enqueue the sender and prepare to block */
Linus Torvalds's avatar
Linus Torvalds committed
669
		ss_add(msq, &s);
670 671 672

		if (!ipc_rcu_getref(msq)) {
			err = -EIDRM;
673
			goto out_unlock0;
674 675
		}

676 677
		ipc_unlock_object(&msq->q_perm);
		rcu_read_unlock();
Linus Torvalds's avatar
Linus Torvalds committed
678 679
		schedule();

680 681 682
		rcu_read_lock();
		ipc_lock_object(&msq->q_perm);

683
		ipc_rcu_putref(msq, msg_rcu_free);
684 685
		/* raced with RMID? */
		if (!ipc_valid_object(&msq->q_perm)) {
Linus Torvalds's avatar
Linus Torvalds committed
686
			err = -EIDRM;
687
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
688
		}
689

Linus Torvalds's avatar
Linus Torvalds committed
690
		ss_del(&s);
691

Linus Torvalds's avatar
Linus Torvalds committed
692
		if (signal_pending(current)) {
693
			err = -ERESTARTNOHAND;
694
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
695
		}
696

Linus Torvalds's avatar
Linus Torvalds committed
697
	}
698
	msq->q_lspid = task_tgid_vnr(current);
Linus Torvalds's avatar
Linus Torvalds committed
699 700
	msq->q_stime = get_seconds();

701
	if (!pipelined_send(msq, msg)) {
702
		/* no one is waiting for this message, enqueue it */
703
		list_add_tail(&msg->m_list, &msq->q_messages);
Linus Torvalds's avatar
Linus Torvalds committed
704 705
		msq->q_cbytes += msgsz;
		msq->q_qnum++;
706 707
		atomic_add(msgsz, &ns->msg_bytes);
		atomic_inc(&ns->msg_hdrs);
Linus Torvalds's avatar
Linus Torvalds committed
708
	}
709

Linus Torvalds's avatar
Linus Torvalds committed
710 711 712
	err = 0;
	msg = NULL;

713 714 715 716
out_unlock0:
	ipc_unlock_object(&msq->q_perm);
out_unlock1:
	rcu_read_unlock();
717
	if (msg != NULL)
Linus Torvalds's avatar
Linus Torvalds committed
718 719 720 721
		free_msg(msg);
	return err;
}

722 723
SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
		int, msgflg)
724 725 726 727 728 729 730 731
{
	long mtype;

	if (get_user(mtype, &msgp->mtype))
		return -EFAULT;
	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
}

732
static inline int convert_mode(long *msgtyp, int msgflg)
Linus Torvalds's avatar
Linus Torvalds committed
733
{
734 735
	if (msgflg & MSG_COPY)
		return SEARCH_NUMBER;
736
	/*
Linus Torvalds's avatar
Linus Torvalds committed
737 738 739
	 *  find message of correct type.
	 *  msgtyp = 0 => get first.
	 *  msgtyp > 0 => get first message of matching type.
740
	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
Linus Torvalds's avatar
Linus Torvalds committed
741
	 */
742
	if (*msgtyp == 0)
Linus Torvalds's avatar
Linus Torvalds committed
743
		return SEARCH_ANY;
744 745
	if (*msgtyp < 0) {
		*msgtyp = -*msgtyp;
Linus Torvalds's avatar
Linus Torvalds committed
746 747
		return SEARCH_LESSEQUAL;
	}
748
	if (msgflg & MSG_EXCEPT)
Linus Torvalds's avatar
Linus Torvalds committed
749 750 751 752
		return SEARCH_NOTEQUAL;
	return SEARCH_EQUAL;
}

753 754 755 756 757 758 759 760 761 762 763 764 765 766
static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
{
	struct msgbuf __user *msgp = dest;
	size_t msgsz;

	if (put_user(msg->m_type, &msgp->mtype))
		return -EFAULT;

	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
	if (store_msg(msgp->mtext, msg, msgsz))
		return -EFAULT;
	return msgsz;
}

767
#ifdef CONFIG_CHECKPOINT_RESTORE
768 769 770 771
/*
 * This function creates new kernel message structure, large enough to store
 * bufsz message bytes.
 */
772
static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
773 774 775 776 777 778 779 780 781 782 783 784
{
	struct msg_msg *copy;

	/*
	 * Create dummy message to copy real message to.
	 */
	copy = load_msg(buf, bufsz);
	if (!IS_ERR(copy))
		copy->m_ts = bufsz;
	return copy;
}

785
static inline void free_copy(struct msg_msg *copy)
786
{
787
	if (copy)
788 789 790
		free_msg(copy);
}
#else
791
static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
792 793 794 795
{
	return ERR_PTR(-ENOSYS);
}

796 797 798
static inline void free_copy(struct msg_msg *copy)
{
}
799 800
#endif

801 802
static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
{
803
	struct msg_msg *msg, *found = NULL;
804 805 806 807 808 809 810 811
	long count = 0;

	list_for_each_entry(msg, &msq->q_messages, m_list) {
		if (testmsg(msg, *msgtyp, mode) &&
		    !security_msg_queue_msgrcv(msq, msg, current,
					       *msgtyp, mode)) {
			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
				*msgtyp = msg->m_type - 1;
812
				found = msg;
813 814 815 816 817 818 819 820 821
			} else if (mode == SEARCH_NUMBER) {
				if (*msgtyp == count)
					return msg;
			} else
				return msg;
			count++;
		}
	}

822
	return found ?: ERR_PTR(-EAGAIN);
823 824
}

825
long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
826
	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
Linus Torvalds's avatar
Linus Torvalds committed
827 828
{
	int mode;
829
	struct msg_queue *msq;
830
	struct ipc_namespace *ns;
831
	struct msg_msg *msg, *copy = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
832

833 834
	ns = current->nsproxy->ipc_ns;

835
	if (msqid < 0 || (long) bufsz < 0)
Linus Torvalds's avatar
Linus Torvalds committed
836
		return -EINVAL;
837

838
	if (msgflg & MSG_COPY) {
839 840
		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
			return -EINVAL;
841
		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
842 843 844
		if (IS_ERR(copy))
			return PTR_ERR(copy);
	}
845
	mode = convert_mode(&msgtyp, msgflg);
Linus Torvalds's avatar
Linus Torvalds committed
846

847 848
	rcu_read_lock();
	msq = msq_obtain_object_check(ns, msqid);
849
	if (IS_ERR(msq)) {
850
		rcu_read_unlock();
851
		free_copy(copy);
852
		return PTR_ERR(msq);
853
	}
Linus Torvalds's avatar
Linus Torvalds committed
854 855 856 857 858

	for (;;) {
		struct msg_receiver msr_d;

		msg = ERR_PTR(-EACCES);
859
		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
860
			goto out_unlock1;
Linus Torvalds's avatar
Linus Torvalds committed
861

862
		ipc_lock_object(&msq->q_perm);
863 864

		/* raced with RMID? */
865
		if (!ipc_valid_object(&msq->q_perm)) {
866 867 868 869
			msg = ERR_PTR(-EIDRM);
			goto out_unlock0;
		}

870
		msg = find_msg(msq, &msgtyp, mode);
871 872 873 874 875
		if (!IS_ERR(msg)) {
			/*
			 * Found a suitable message.
			 * Unlink it from the queue.
			 */
876
			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
Linus Torvalds's avatar
Linus Torvalds committed
877
				msg = ERR_PTR(-E2BIG);
878
				goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
879
			}
880 881 882 883
			/*
			 * If we are copying, then do not unlink message and do
			 * not update queue parameters.
			 */
884 885
			if (msgflg & MSG_COPY) {
				msg = copy_msg(msg, copy);
886
				goto out_unlock0;
887
			}
888

Linus Torvalds's avatar
Linus Torvalds committed
889 890 891
			list_del(&msg->m_list);
			msq->q_qnum--;
			msq->q_rtime = get_seconds();
892
			msq->q_lrpid = task_tgid_vnr(current);
Linus Torvalds's avatar
Linus Torvalds committed
893
			msq->q_cbytes -= msg->m_ts;
894 895
			atomic_sub(msg->m_ts, &ns->msg_bytes);
			atomic_dec(&ns->msg_hdrs);
896
			ss_wakeup(&msq->q_senders, 0);
897 898

			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
899
		}
900

Linus Torvalds's avatar
Linus Torvalds committed
901 902 903
		/* No message waiting. Wait for a message */
		if (msgflg & IPC_NOWAIT) {
			msg = ERR_PTR(-ENOMSG);
904
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
905
		}
906

907
		list_add_tail(&msr_d.r_list, &msq->q_receivers);
Linus Torvalds's avatar
Linus Torvalds committed
908 909 910
		msr_d.r_tsk = current;
		msr_d.r_msgtype = msgtyp;
		msr_d.r_mode = mode;
911
		if (msgflg & MSG_NOERROR)
Linus Torvalds's avatar
Linus Torvalds committed
912
			msr_d.r_maxsize = INT_MAX;
913
		else
914
			msr_d.r_maxsize = bufsz;
Linus Torvalds's avatar
Linus Torvalds committed
915
		msr_d.r_msg = ERR_PTR(-EAGAIN);
916
		__set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds's avatar
Linus Torvalds committed
917

918 919
		ipc_unlock_object(&msq->q_perm);
		rcu_read_unlock();
Linus Torvalds's avatar
Linus Torvalds committed
920 921 922 923 924 925
		schedule();

		/* Lockless receive, part 1:
		 * Disable preemption.  We don't hold a reference to the queue
		 * and getting a reference would defeat the idea of a lockless
		 * operation, thus the code relies on rcu to guarantee the
926
		 * existence of msq:
Linus Torvalds's avatar
Linus Torvalds committed
927 928 929
		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
		 * rcu_read_lock() prevents preemption between reading r_msg
930
		 * and acquiring the q_perm.lock in ipc_lock_object().
Linus Torvalds's avatar
Linus Torvalds committed
931 932 933 934 935 936
		 */
		rcu_read_lock();

		/* Lockless receive, part 2:
		 * Wait until pipelined_send or expunge_all are outside of
		 * wake_up_process(). There is a race with exit(), see
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951
		 * ipc/mqueue.c for the details. The correct serialization
		 * ensures that a receiver cannot continue without the wakeup
		 * being visibible _before_ setting r_msg:
		 *
		 * CPU 0                             CPU 1
		 * <loop receiver>
		 *   smp_rmb(); (A) <-- pair -.      <waker thread>
		 *   <load ->r_msg>           |        msr->r_msg = NULL;
		 *                            |        wake_up_process();
		 * <continue>                 `------> smp_wmb(); (B)
		 *                                     msr->r_msg = msg;
		 *
		 * Where (A) orders the message value read and where (B) orders
		 * the write to the r_msg -- done in both pipelined_send and
		 * expunge_all.
Linus Torvalds's avatar
Linus Torvalds committed
952
		 */
953 954 955 956 957 958
		for (;;) {
			/*
			 * Pairs with writer barrier in pipelined_send
			 * or expunge_all.
			 */
			smp_rmb(); /* barrier (A) */
959
			msg = (struct msg_msg *)msr_d.r_msg;
960 961 962 963 964 965 966 967 968
			if (msg)
				break;

			/*
			 * The cpu_relax() call is a compiler barrier
			 * which forces everything in this loop to be
			 * re-loaded.
			 */
			cpu_relax();
Linus Torvalds's avatar
Linus Torvalds committed
969 970 971 972 973 974
		}

		/* Lockless receive, part 3:
		 * If there is a message or an error then accept it without
		 * locking.
		 */
975 976
		if (msg != ERR_PTR(-EAGAIN))
			goto out_unlock1;
Linus Torvalds's avatar
Linus Torvalds committed
977 978 979 980

		/* Lockless receive, part 3:
		 * Acquire the queue spinlock.
		 */
981
		ipc_lock_object(&msq->q_perm);
Linus Torvalds's avatar
Linus Torvalds committed
982 983 984 985

		/* Lockless receive, part 4:
		 * Repeat test after acquiring the spinlock.
		 */
Manfred Spraul's avatar
Manfred Spraul committed
986
		msg = (struct msg_msg *)msr_d.r_msg;
987
		if (msg != ERR_PTR(-EAGAIN))
988
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
989 990 991 992

		list_del(&msr_d.r_list);
		if (signal_pending(current)) {
			msg = ERR_PTR(-ERESTARTNOHAND);
993
			goto out_unlock0;
Linus Torvalds's avatar
Linus Torvalds committed
994
		}
995 996

		ipc_unlock_object(&msq->q_perm);
Linus Torvalds's avatar
Linus Torvalds committed
997
	}
998 999 1000 1001 1002

out_unlock0:
	ipc_unlock_object(&msq->q_perm);
out_unlock1:
	rcu_read_unlock();
1003
	if (IS_ERR(msg)) {
1004
		free_copy(copy);
1005
		return PTR_ERR(msg);
1006
	}
Linus Torvalds's avatar
Linus Torvalds committed
1007

1008
	bufsz = msg_handler(buf, msg, bufsz);
Linus Torvalds's avatar
Linus Torvalds committed
1009
	free_msg(msg);
1010

1011
	return bufsz;
Linus Torvalds's avatar
Linus Torvalds committed
1012 1013
}

1014 1015
SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
		long, msgtyp, int, msgflg)
1016
{
1017
	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1018 1019
}

1020 1021 1022 1023 1024

void msg_init_ns(struct ipc_namespace *ns)
{
	ns->msg_ctlmax = MSGMAX;
	ns->msg_ctlmnb = MSGMNB;
1025
	ns->msg_ctlmni = MSGMNI;
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039

	atomic_set(&ns->msg_bytes, 0);
	atomic_set(&ns->msg_hdrs, 0);
	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
}

#ifdef CONFIG_IPC_NS
void msg_exit_ns(struct ipc_namespace *ns)
{
	free_ipcs(ns, &msg_ids(ns), freeque);
	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
}
#endif

Linus Torvalds's avatar
Linus Torvalds committed
1040
#ifdef CONFIG_PROC_FS
1041
static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
Linus Torvalds's avatar
Linus Torvalds committed
1042
{
1043
	struct user_namespace *user_ns = seq_user_ns(s);
1044 1045
	struct msg_queue *msq = it;

1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
	seq_printf(s,
		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
		   msq->q_perm.key,
		   msq->q_perm.id,
		   msq->q_perm.mode,
		   msq->q_cbytes,
		   msq->q_qnum,
		   msq->q_lspid,
		   msq->q_lrpid,
		   from_kuid_munged(user_ns, msq->q_perm.uid),
		   from_kgid_munged(user_ns, msq->q_perm.gid),
		   from_kuid_munged(user_ns, msq->q_perm.cuid),
		   from_kgid_munged(user_ns, msq->q_perm.cgid),
		   msq->q_stime,
		   msq->q_rtime,
		   msq->q_ctime);

	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
1064 1065
}
#endif
1066 1067 1068 1069 1070 1071 1072 1073 1074

void __init msg_init(void)
{
	msg_init_ns(&init_ipc_ns);

	ipc_init_proc_interface("sysvipc/msg",
				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
				IPC_MSG_IDS, sysvipc_msg_proc_show);
}