xref: /linux/ipc/msg.c (revision eae3df7e82318d798f45dedf111e241805ec7a4a)
1 /*
2  * linux/ipc/msg.c
3  * Copyright (C) 1992 Krishna Balasubramanian
4  *
5  * Removed all the remaining kerneld mess
6  * Catch the -EFAULT stuff properly
7  * Use GFP_KERNEL for messages as in 1.2
8  * Fixed up the unchecked user space derefs
9  * Copyright (C) 1998 Alan Cox & Andi Kleen
10  *
11  * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
12  *
13  * mostly rewritten, threaded and wake-one semantics added
14  * MSGMAX limit removed, sysctl's added
15  * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
16  *
17  * support for audit of ipc object properties and permission changes
18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19  *
20  * namespaces support
21  * OpenVZ, SWsoft Inc.
22  * Pavel Emelianov <xemul@openvz.org>
23  */
24 
25 #include <linux/capability.h>
26 #include <linux/msg.h>
27 #include <linux/spinlock.h>
28 #include <linux/init.h>
29 #include <linux/mm.h>
30 #include <linux/proc_fs.h>
31 #include <linux/list.h>
32 #include <linux/security.h>
33 #include <linux/sched/wake_q.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/seq_file.h>
37 #include <linux/rwsem.h>
38 #include <linux/nsproxy.h>
39 #include <linux/ipc_namespace.h>
40 
41 #include <asm/current.h>
42 #include <linux/uaccess.h>
43 #include "util.h"
44 
45 /* one msg_receiver structure for each sleeping receiver */
46 struct msg_receiver {
47 	struct list_head	r_list;
48 	struct task_struct	*r_tsk;
49 
50 	int			r_mode;
51 	long			r_msgtype;
52 	long			r_maxsize;
53 
54 	struct msg_msg		*r_msg;
55 };
56 
57 /* one msg_sender for each sleeping sender */
58 struct msg_sender {
59 	struct list_head	list;
60 	struct task_struct	*tsk;
61 	size_t                  msgsz;
62 };
63 
64 #define SEARCH_ANY		1
65 #define SEARCH_EQUAL		2
66 #define SEARCH_NOTEQUAL		3
67 #define SEARCH_LESSEQUAL	4
68 #define SEARCH_NUMBER		5
69 
70 #define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
71 
72 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
73 {
74 	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
75 
76 	if (IS_ERR(ipcp))
77 		return ERR_CAST(ipcp);
78 
79 	return container_of(ipcp, struct msg_queue, q_perm);
80 }
81 
82 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
83 							int id)
84 {
85 	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
86 
87 	if (IS_ERR(ipcp))
88 		return ERR_CAST(ipcp);
89 
90 	return container_of(ipcp, struct msg_queue, q_perm);
91 }
92 
93 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
94 {
95 	ipc_rmid(&msg_ids(ns), &s->q_perm);
96 }
97 
98 static void msg_rcu_free(struct rcu_head *head)
99 {
100 	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
101 	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
102 
103 	security_msg_queue_free(msq);
104 	kvfree(msq);
105 }
106 
107 /**
108  * newque - Create a new msg queue
109  * @ns: namespace
110  * @params: ptr to the structure that contains the key and msgflg
111  *
112  * Called with msg_ids.rwsem held (writer)
113  */
114 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
115 {
116 	struct msg_queue *msq;
117 	int retval;
118 	key_t key = params->key;
119 	int msgflg = params->flg;
120 
121 	msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
122 	if (unlikely(!msq))
123 		return -ENOMEM;
124 
125 	msq->q_perm.mode = msgflg & S_IRWXUGO;
126 	msq->q_perm.key = key;
127 
128 	msq->q_perm.security = NULL;
129 	retval = security_msg_queue_alloc(msq);
130 	if (retval) {
131 		kvfree(msq);
132 		return retval;
133 	}
134 
135 	msq->q_stime = msq->q_rtime = 0;
136 	msq->q_ctime = ktime_get_real_seconds();
137 	msq->q_cbytes = msq->q_qnum = 0;
138 	msq->q_qbytes = ns->msg_ctlmnb;
139 	msq->q_lspid = msq->q_lrpid = 0;
140 	INIT_LIST_HEAD(&msq->q_messages);
141 	INIT_LIST_HEAD(&msq->q_receivers);
142 	INIT_LIST_HEAD(&msq->q_senders);
143 
144 	/* ipc_addid() locks msq upon success. */
145 	retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
146 	if (retval < 0) {
147 		call_rcu(&msq->q_perm.rcu, msg_rcu_free);
148 		return retval;
149 	}
150 
151 	ipc_unlock_object(&msq->q_perm);
152 	rcu_read_unlock();
153 
154 	return msq->q_perm.id;
155 }
156 
157 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
158 {
159 	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
160 		1 + msq->q_qnum <= msq->q_qbytes;
161 }
162 
163 static inline void ss_add(struct msg_queue *msq,
164 			  struct msg_sender *mss, size_t msgsz)
165 {
166 	mss->tsk = current;
167 	mss->msgsz = msgsz;
168 	__set_current_state(TASK_INTERRUPTIBLE);
169 	list_add_tail(&mss->list, &msq->q_senders);
170 }
171 
172 static inline void ss_del(struct msg_sender *mss)
173 {
174 	if (mss->list.next)
175 		list_del(&mss->list);
176 }
177 
178 static void ss_wakeup(struct msg_queue *msq,
179 		      struct wake_q_head *wake_q, bool kill)
180 {
181 	struct msg_sender *mss, *t;
182 	struct task_struct *stop_tsk = NULL;
183 	struct list_head *h = &msq->q_senders;
184 
185 	list_for_each_entry_safe(mss, t, h, list) {
186 		if (kill)
187 			mss->list.next = NULL;
188 
189 		/*
190 		 * Stop at the first task we don't wakeup,
191 		 * we've already iterated the original
192 		 * sender queue.
193 		 */
194 		else if (stop_tsk == mss->tsk)
195 			break;
196 		/*
197 		 * We are not in an EIDRM scenario here, therefore
198 		 * verify that we really need to wakeup the task.
199 		 * To maintain current semantics and wakeup order,
200 		 * move the sender to the tail on behalf of the
201 		 * blocked task.
202 		 */
203 		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
204 			if (!stop_tsk)
205 				stop_tsk = mss->tsk;
206 
207 			list_move_tail(&mss->list, &msq->q_senders);
208 			continue;
209 		}
210 
211 		wake_q_add(wake_q, mss->tsk);
212 	}
213 }
214 
215 static void expunge_all(struct msg_queue *msq, int res,
216 			struct wake_q_head *wake_q)
217 {
218 	struct msg_receiver *msr, *t;
219 
220 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
221 		wake_q_add(wake_q, msr->r_tsk);
222 		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
223 	}
224 }
225 
226 /*
227  * freeque() wakes up waiters on the sender and receiver waiting queue,
228  * removes the message queue from message queue ID IDR, and cleans up all the
229  * messages associated with this queue.
230  *
231  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
232  * before freeque() is called. msg_ids.rwsem remains locked on exit.
233  */
234 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
235 {
236 	struct msg_msg *msg, *t;
237 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
238 	DEFINE_WAKE_Q(wake_q);
239 
240 	expunge_all(msq, -EIDRM, &wake_q);
241 	ss_wakeup(msq, &wake_q, true);
242 	msg_rmid(ns, msq);
243 	ipc_unlock_object(&msq->q_perm);
244 	wake_up_q(&wake_q);
245 	rcu_read_unlock();
246 
247 	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
248 		atomic_dec(&ns->msg_hdrs);
249 		free_msg(msg);
250 	}
251 	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
252 	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
253 }
254 
255 /*
256  * Called with msg_ids.rwsem and ipcp locked.
257  */
258 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
259 {
260 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
261 
262 	return security_msg_queue_associate(msq, msgflg);
263 }
264 
265 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
266 {
267 	struct ipc_namespace *ns;
268 	static const struct ipc_ops msg_ops = {
269 		.getnew = newque,
270 		.associate = msg_security,
271 	};
272 	struct ipc_params msg_params;
273 
274 	ns = current->nsproxy->ipc_ns;
275 
276 	msg_params.key = key;
277 	msg_params.flg = msgflg;
278 
279 	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
280 }
281 
282 static inline unsigned long
283 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
284 {
285 	switch (version) {
286 	case IPC_64:
287 		return copy_to_user(buf, in, sizeof(*in));
288 	case IPC_OLD:
289 	{
290 		struct msqid_ds out;
291 
292 		memset(&out, 0, sizeof(out));
293 
294 		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
295 
296 		out.msg_stime		= in->msg_stime;
297 		out.msg_rtime		= in->msg_rtime;
298 		out.msg_ctime		= in->msg_ctime;
299 
300 		if (in->msg_cbytes > USHRT_MAX)
301 			out.msg_cbytes	= USHRT_MAX;
302 		else
303 			out.msg_cbytes	= in->msg_cbytes;
304 		out.msg_lcbytes		= in->msg_cbytes;
305 
306 		if (in->msg_qnum > USHRT_MAX)
307 			out.msg_qnum	= USHRT_MAX;
308 		else
309 			out.msg_qnum	= in->msg_qnum;
310 
311 		if (in->msg_qbytes > USHRT_MAX)
312 			out.msg_qbytes	= USHRT_MAX;
313 		else
314 			out.msg_qbytes	= in->msg_qbytes;
315 		out.msg_lqbytes		= in->msg_qbytes;
316 
317 		out.msg_lspid		= in->msg_lspid;
318 		out.msg_lrpid		= in->msg_lrpid;
319 
320 		return copy_to_user(buf, &out, sizeof(out));
321 	}
322 	default:
323 		return -EINVAL;
324 	}
325 }
326 
327 static inline unsigned long
328 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
329 {
330 	switch (version) {
331 	case IPC_64:
332 		if (copy_from_user(out, buf, sizeof(*out)))
333 			return -EFAULT;
334 		return 0;
335 	case IPC_OLD:
336 	{
337 		struct msqid_ds tbuf_old;
338 
339 		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
340 			return -EFAULT;
341 
342 		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
343 		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
344 		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
345 
346 		if (tbuf_old.msg_qbytes == 0)
347 			out->msg_qbytes	= tbuf_old.msg_lqbytes;
348 		else
349 			out->msg_qbytes	= tbuf_old.msg_qbytes;
350 
351 		return 0;
352 	}
353 	default:
354 		return -EINVAL;
355 	}
356 }
357 
358 /*
359  * This function handles some msgctl commands which require the rwsem
360  * to be held in write mode.
361  * NOTE: no locks must be held, the rwsem is taken inside this function.
362  */
363 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
364 			struct msqid64_ds *msqid64)
365 {
366 	struct kern_ipc_perm *ipcp;
367 	struct msg_queue *msq;
368 	int err;
369 
370 	down_write(&msg_ids(ns).rwsem);
371 	rcu_read_lock();
372 
373 	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
374 				      &msqid64->msg_perm, msqid64->msg_qbytes);
375 	if (IS_ERR(ipcp)) {
376 		err = PTR_ERR(ipcp);
377 		goto out_unlock1;
378 	}
379 
380 	msq = container_of(ipcp, struct msg_queue, q_perm);
381 
382 	err = security_msg_queue_msgctl(msq, cmd);
383 	if (err)
384 		goto out_unlock1;
385 
386 	switch (cmd) {
387 	case IPC_RMID:
388 		ipc_lock_object(&msq->q_perm);
389 		/* freeque unlocks the ipc object and rcu */
390 		freeque(ns, ipcp);
391 		goto out_up;
392 	case IPC_SET:
393 	{
394 		DEFINE_WAKE_Q(wake_q);
395 
396 		if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
397 		    !capable(CAP_SYS_RESOURCE)) {
398 			err = -EPERM;
399 			goto out_unlock1;
400 		}
401 
402 		ipc_lock_object(&msq->q_perm);
403 		err = ipc_update_perm(&msqid64->msg_perm, ipcp);
404 		if (err)
405 			goto out_unlock0;
406 
407 		msq->q_qbytes = msqid64->msg_qbytes;
408 
409 		msq->q_ctime = ktime_get_real_seconds();
410 		/*
411 		 * Sleeping receivers might be excluded by
412 		 * stricter permissions.
413 		 */
414 		expunge_all(msq, -EAGAIN, &wake_q);
415 		/*
416 		 * Sleeping senders might be able to send
417 		 * due to a larger queue size.
418 		 */
419 		ss_wakeup(msq, &wake_q, false);
420 		ipc_unlock_object(&msq->q_perm);
421 		wake_up_q(&wake_q);
422 
423 		goto out_unlock1;
424 	}
425 	default:
426 		err = -EINVAL;
427 		goto out_unlock1;
428 	}
429 
430 out_unlock0:
431 	ipc_unlock_object(&msq->q_perm);
432 out_unlock1:
433 	rcu_read_unlock();
434 out_up:
435 	up_write(&msg_ids(ns).rwsem);
436 	return err;
437 }
438 
439 static int msgctl_info(struct ipc_namespace *ns, int msqid,
440 			 int cmd, struct msginfo *msginfo)
441 {
442 	int err;
443 	int max_id;
444 
445 	/*
446 	 * We must not return kernel stack data.
447 	 * due to padding, it's not enough
448 	 * to set all member fields.
449 	 */
450 	err = security_msg_queue_msgctl(NULL, cmd);
451 	if (err)
452 		return err;
453 
454 	memset(msginfo, 0, sizeof(*msginfo));
455 	msginfo->msgmni = ns->msg_ctlmni;
456 	msginfo->msgmax = ns->msg_ctlmax;
457 	msginfo->msgmnb = ns->msg_ctlmnb;
458 	msginfo->msgssz = MSGSSZ;
459 	msginfo->msgseg = MSGSEG;
460 	down_read(&msg_ids(ns).rwsem);
461 	if (cmd == MSG_INFO) {
462 		msginfo->msgpool = msg_ids(ns).in_use;
463 		msginfo->msgmap = atomic_read(&ns->msg_hdrs);
464 		msginfo->msgtql = atomic_read(&ns->msg_bytes);
465 	} else {
466 		msginfo->msgmap = MSGMAP;
467 		msginfo->msgpool = MSGPOOL;
468 		msginfo->msgtql = MSGTQL;
469 	}
470 	max_id = ipc_get_maxid(&msg_ids(ns));
471 	up_read(&msg_ids(ns).rwsem);
472 	return (max_id < 0) ? 0 : max_id;
473 }
474 
475 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
476 			 int cmd, struct msqid64_ds *p)
477 {
478 	int err;
479 	struct msg_queue *msq;
480 	int success_return;
481 
482 	memset(p, 0, sizeof(*p));
483 
484 	rcu_read_lock();
485 	if (cmd == MSG_STAT) {
486 		msq = msq_obtain_object(ns, msqid);
487 		if (IS_ERR(msq)) {
488 			err = PTR_ERR(msq);
489 			goto out_unlock;
490 		}
491 		success_return = msq->q_perm.id;
492 	} else {
493 		msq = msq_obtain_object_check(ns, msqid);
494 		if (IS_ERR(msq)) {
495 			err = PTR_ERR(msq);
496 			goto out_unlock;
497 		}
498 		success_return = 0;
499 	}
500 
501 	err = -EACCES;
502 	if (ipcperms(ns, &msq->q_perm, S_IRUGO))
503 		goto out_unlock;
504 
505 	err = security_msg_queue_msgctl(msq, cmd);
506 	if (err)
507 		goto out_unlock;
508 
509 	kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
510 	p->msg_stime  = msq->q_stime;
511 	p->msg_rtime  = msq->q_rtime;
512 	p->msg_ctime  = msq->q_ctime;
513 	p->msg_cbytes = msq->q_cbytes;
514 	p->msg_qnum   = msq->q_qnum;
515 	p->msg_qbytes = msq->q_qbytes;
516 	p->msg_lspid  = msq->q_lspid;
517 	p->msg_lrpid  = msq->q_lrpid;
518 	rcu_read_unlock();
519 
520 	return success_return;
521 
522 out_unlock:
523 	rcu_read_unlock();
524 	return err;
525 }
526 
527 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
528 {
529 	int version;
530 	struct ipc_namespace *ns;
531 	struct msqid64_ds msqid64;
532 	int err;
533 
534 	if (msqid < 0 || cmd < 0)
535 		return -EINVAL;
536 
537 	version = ipc_parse_version(&cmd);
538 	ns = current->nsproxy->ipc_ns;
539 
540 	switch (cmd) {
541 	case IPC_INFO:
542 	case MSG_INFO: {
543 		struct msginfo msginfo;
544 		err = msgctl_info(ns, msqid, cmd, &msginfo);
545 		if (err < 0)
546 			return err;
547 		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
548 			err = -EFAULT;
549 		return err;
550 	}
551 	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
552 	case IPC_STAT:
553 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
554 		if (err < 0)
555 			return err;
556 		if (copy_msqid_to_user(buf, &msqid64, version))
557 			err = -EFAULT;
558 		return err;
559 	case IPC_SET:
560 		if (copy_msqid_from_user(&msqid64, buf, version))
561 			return -EFAULT;
562 		/* fallthru */
563 	case IPC_RMID:
564 		return msgctl_down(ns, msqid, cmd, &msqid64);
565 	default:
566 		return  -EINVAL;
567 	}
568 }
569 
570 #ifdef CONFIG_COMPAT
571 
572 struct compat_msqid_ds {
573 	struct compat_ipc_perm msg_perm;
574 	compat_uptr_t msg_first;
575 	compat_uptr_t msg_last;
576 	compat_time_t msg_stime;
577 	compat_time_t msg_rtime;
578 	compat_time_t msg_ctime;
579 	compat_ulong_t msg_lcbytes;
580 	compat_ulong_t msg_lqbytes;
581 	unsigned short msg_cbytes;
582 	unsigned short msg_qnum;
583 	unsigned short msg_qbytes;
584 	compat_ipc_pid_t msg_lspid;
585 	compat_ipc_pid_t msg_lrpid;
586 };
587 
588 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
589 					int version)
590 {
591 	memset(out, 0, sizeof(*out));
592 	if (version == IPC_64) {
593 		struct compat_msqid64_ds *p = buf;
594 		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
595 			return -EFAULT;
596 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
597 			return -EFAULT;
598 	} else {
599 		struct compat_msqid_ds *p = buf;
600 		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
601 			return -EFAULT;
602 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
603 			return -EFAULT;
604 	}
605 	return 0;
606 }
607 
608 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
609 					int version)
610 {
611 	if (version == IPC_64) {
612 		struct compat_msqid64_ds v;
613 		memset(&v, 0, sizeof(v));
614 		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
615 		v.msg_stime = in->msg_stime;
616 		v.msg_rtime = in->msg_rtime;
617 		v.msg_ctime = in->msg_ctime;
618 		v.msg_cbytes = in->msg_cbytes;
619 		v.msg_qnum = in->msg_qnum;
620 		v.msg_qbytes = in->msg_qbytes;
621 		v.msg_lspid = in->msg_lspid;
622 		v.msg_lrpid = in->msg_lrpid;
623 		return copy_to_user(buf, &v, sizeof(v));
624 	} else {
625 		struct compat_msqid_ds v;
626 		memset(&v, 0, sizeof(v));
627 		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
628 		v.msg_stime = in->msg_stime;
629 		v.msg_rtime = in->msg_rtime;
630 		v.msg_ctime = in->msg_ctime;
631 		v.msg_cbytes = in->msg_cbytes;
632 		v.msg_qnum = in->msg_qnum;
633 		v.msg_qbytes = in->msg_qbytes;
634 		v.msg_lspid = in->msg_lspid;
635 		v.msg_lrpid = in->msg_lrpid;
636 		return copy_to_user(buf, &v, sizeof(v));
637 	}
638 }
639 
640 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
641 {
642 	struct ipc_namespace *ns;
643 	int err;
644 	struct msqid64_ds msqid64;
645 	int version = compat_ipc_parse_version(&cmd);
646 
647 	ns = current->nsproxy->ipc_ns;
648 
649 	if (msqid < 0 || cmd < 0)
650 		return -EINVAL;
651 
652 	switch (cmd & (~IPC_64)) {
653 	case IPC_INFO:
654 	case MSG_INFO: {
655 		struct msginfo msginfo;
656 		err = msgctl_info(ns, msqid, cmd, &msginfo);
657 		if (err < 0)
658 			return err;
659 		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
660 			err = -EFAULT;
661 		return err;
662 	}
663 	case IPC_STAT:
664 	case MSG_STAT:
665 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
666 		if (err < 0)
667 			return err;
668 		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
669 			err = -EFAULT;
670 		return err;
671 	case IPC_SET:
672 		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
673 			return -EFAULT;
674 		/* fallthru */
675 	case IPC_RMID:
676 		return msgctl_down(ns, msqid, cmd, &msqid64);
677 	default:
678 		return -EINVAL;
679 	}
680 }
681 #endif
682 
683 static int testmsg(struct msg_msg *msg, long type, int mode)
684 {
685 	switch (mode) {
686 	case SEARCH_ANY:
687 	case SEARCH_NUMBER:
688 		return 1;
689 	case SEARCH_LESSEQUAL:
690 		if (msg->m_type <= type)
691 			return 1;
692 		break;
693 	case SEARCH_EQUAL:
694 		if (msg->m_type == type)
695 			return 1;
696 		break;
697 	case SEARCH_NOTEQUAL:
698 		if (msg->m_type != type)
699 			return 1;
700 		break;
701 	}
702 	return 0;
703 }
704 
705 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
706 				 struct wake_q_head *wake_q)
707 {
708 	struct msg_receiver *msr, *t;
709 
710 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
711 		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
712 		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
713 					       msr->r_msgtype, msr->r_mode)) {
714 
715 			list_del(&msr->r_list);
716 			if (msr->r_maxsize < msg->m_ts) {
717 				wake_q_add(wake_q, msr->r_tsk);
718 				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
719 			} else {
720 				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
721 				msq->q_rtime = get_seconds();
722 
723 				wake_q_add(wake_q, msr->r_tsk);
724 				WRITE_ONCE(msr->r_msg, msg);
725 				return 1;
726 			}
727 		}
728 	}
729 
730 	return 0;
731 }
732 
733 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
734 		size_t msgsz, int msgflg)
735 {
736 	struct msg_queue *msq;
737 	struct msg_msg *msg;
738 	int err;
739 	struct ipc_namespace *ns;
740 	DEFINE_WAKE_Q(wake_q);
741 
742 	ns = current->nsproxy->ipc_ns;
743 
744 	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
745 		return -EINVAL;
746 	if (mtype < 1)
747 		return -EINVAL;
748 
749 	msg = load_msg(mtext, msgsz);
750 	if (IS_ERR(msg))
751 		return PTR_ERR(msg);
752 
753 	msg->m_type = mtype;
754 	msg->m_ts = msgsz;
755 
756 	rcu_read_lock();
757 	msq = msq_obtain_object_check(ns, msqid);
758 	if (IS_ERR(msq)) {
759 		err = PTR_ERR(msq);
760 		goto out_unlock1;
761 	}
762 
763 	ipc_lock_object(&msq->q_perm);
764 
765 	for (;;) {
766 		struct msg_sender s;
767 
768 		err = -EACCES;
769 		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
770 			goto out_unlock0;
771 
772 		/* raced with RMID? */
773 		if (!ipc_valid_object(&msq->q_perm)) {
774 			err = -EIDRM;
775 			goto out_unlock0;
776 		}
777 
778 		err = security_msg_queue_msgsnd(msq, msg, msgflg);
779 		if (err)
780 			goto out_unlock0;
781 
782 		if (msg_fits_inqueue(msq, msgsz))
783 			break;
784 
785 		/* queue full, wait: */
786 		if (msgflg & IPC_NOWAIT) {
787 			err = -EAGAIN;
788 			goto out_unlock0;
789 		}
790 
791 		/* enqueue the sender and prepare to block */
792 		ss_add(msq, &s, msgsz);
793 
794 		if (!ipc_rcu_getref(&msq->q_perm)) {
795 			err = -EIDRM;
796 			goto out_unlock0;
797 		}
798 
799 		ipc_unlock_object(&msq->q_perm);
800 		rcu_read_unlock();
801 		schedule();
802 
803 		rcu_read_lock();
804 		ipc_lock_object(&msq->q_perm);
805 
806 		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
807 		/* raced with RMID? */
808 		if (!ipc_valid_object(&msq->q_perm)) {
809 			err = -EIDRM;
810 			goto out_unlock0;
811 		}
812 		ss_del(&s);
813 
814 		if (signal_pending(current)) {
815 			err = -ERESTARTNOHAND;
816 			goto out_unlock0;
817 		}
818 
819 	}
820 
821 	msq->q_lspid = task_tgid_vnr(current);
822 	msq->q_stime = get_seconds();
823 
824 	if (!pipelined_send(msq, msg, &wake_q)) {
825 		/* no one is waiting for this message, enqueue it */
826 		list_add_tail(&msg->m_list, &msq->q_messages);
827 		msq->q_cbytes += msgsz;
828 		msq->q_qnum++;
829 		atomic_add(msgsz, &ns->msg_bytes);
830 		atomic_inc(&ns->msg_hdrs);
831 	}
832 
833 	err = 0;
834 	msg = NULL;
835 
836 out_unlock0:
837 	ipc_unlock_object(&msq->q_perm);
838 	wake_up_q(&wake_q);
839 out_unlock1:
840 	rcu_read_unlock();
841 	if (msg != NULL)
842 		free_msg(msg);
843 	return err;
844 }
845 
846 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
847 		int, msgflg)
848 {
849 	long mtype;
850 
851 	if (get_user(mtype, &msgp->mtype))
852 		return -EFAULT;
853 	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
854 }
855 
856 #ifdef CONFIG_COMPAT
857 
858 struct compat_msgbuf {
859 	compat_long_t mtype;
860 	char mtext[1];
861 };
862 
863 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
864 		       compat_ssize_t, msgsz, int, msgflg)
865 {
866 	struct compat_msgbuf __user *up = compat_ptr(msgp);
867 	compat_long_t mtype;
868 
869 	if (get_user(mtype, &up->mtype))
870 		return -EFAULT;
871 	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
872 }
873 #endif
874 
875 static inline int convert_mode(long *msgtyp, int msgflg)
876 {
877 	if (msgflg & MSG_COPY)
878 		return SEARCH_NUMBER;
879 	/*
880 	 *  find message of correct type.
881 	 *  msgtyp = 0 => get first.
882 	 *  msgtyp > 0 => get first message of matching type.
883 	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
884 	 */
885 	if (*msgtyp == 0)
886 		return SEARCH_ANY;
887 	if (*msgtyp < 0) {
888 		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
889 			*msgtyp = LONG_MAX;
890 		else
891 			*msgtyp = -*msgtyp;
892 		return SEARCH_LESSEQUAL;
893 	}
894 	if (msgflg & MSG_EXCEPT)
895 		return SEARCH_NOTEQUAL;
896 	return SEARCH_EQUAL;
897 }
898 
899 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
900 {
901 	struct msgbuf __user *msgp = dest;
902 	size_t msgsz;
903 
904 	if (put_user(msg->m_type, &msgp->mtype))
905 		return -EFAULT;
906 
907 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
908 	if (store_msg(msgp->mtext, msg, msgsz))
909 		return -EFAULT;
910 	return msgsz;
911 }
912 
913 #ifdef CONFIG_CHECKPOINT_RESTORE
914 /*
915  * This function creates new kernel message structure, large enough to store
916  * bufsz message bytes.
917  */
918 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
919 {
920 	struct msg_msg *copy;
921 
922 	/*
923 	 * Create dummy message to copy real message to.
924 	 */
925 	copy = load_msg(buf, bufsz);
926 	if (!IS_ERR(copy))
927 		copy->m_ts = bufsz;
928 	return copy;
929 }
930 
931 static inline void free_copy(struct msg_msg *copy)
932 {
933 	if (copy)
934 		free_msg(copy);
935 }
936 #else
937 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
938 {
939 	return ERR_PTR(-ENOSYS);
940 }
941 
942 static inline void free_copy(struct msg_msg *copy)
943 {
944 }
945 #endif
946 
947 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
948 {
949 	struct msg_msg *msg, *found = NULL;
950 	long count = 0;
951 
952 	list_for_each_entry(msg, &msq->q_messages, m_list) {
953 		if (testmsg(msg, *msgtyp, mode) &&
954 		    !security_msg_queue_msgrcv(msq, msg, current,
955 					       *msgtyp, mode)) {
956 			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
957 				*msgtyp = msg->m_type - 1;
958 				found = msg;
959 			} else if (mode == SEARCH_NUMBER) {
960 				if (*msgtyp == count)
961 					return msg;
962 			} else
963 				return msg;
964 			count++;
965 		}
966 	}
967 
968 	return found ?: ERR_PTR(-EAGAIN);
969 }
970 
971 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
972 	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
973 {
974 	int mode;
975 	struct msg_queue *msq;
976 	struct ipc_namespace *ns;
977 	struct msg_msg *msg, *copy = NULL;
978 	DEFINE_WAKE_Q(wake_q);
979 
980 	ns = current->nsproxy->ipc_ns;
981 
982 	if (msqid < 0 || (long) bufsz < 0)
983 		return -EINVAL;
984 
985 	if (msgflg & MSG_COPY) {
986 		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
987 			return -EINVAL;
988 		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
989 		if (IS_ERR(copy))
990 			return PTR_ERR(copy);
991 	}
992 	mode = convert_mode(&msgtyp, msgflg);
993 
994 	rcu_read_lock();
995 	msq = msq_obtain_object_check(ns, msqid);
996 	if (IS_ERR(msq)) {
997 		rcu_read_unlock();
998 		free_copy(copy);
999 		return PTR_ERR(msq);
1000 	}
1001 
1002 	for (;;) {
1003 		struct msg_receiver msr_d;
1004 
1005 		msg = ERR_PTR(-EACCES);
1006 		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1007 			goto out_unlock1;
1008 
1009 		ipc_lock_object(&msq->q_perm);
1010 
1011 		/* raced with RMID? */
1012 		if (!ipc_valid_object(&msq->q_perm)) {
1013 			msg = ERR_PTR(-EIDRM);
1014 			goto out_unlock0;
1015 		}
1016 
1017 		msg = find_msg(msq, &msgtyp, mode);
1018 		if (!IS_ERR(msg)) {
1019 			/*
1020 			 * Found a suitable message.
1021 			 * Unlink it from the queue.
1022 			 */
1023 			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1024 				msg = ERR_PTR(-E2BIG);
1025 				goto out_unlock0;
1026 			}
1027 			/*
1028 			 * If we are copying, then do not unlink message and do
1029 			 * not update queue parameters.
1030 			 */
1031 			if (msgflg & MSG_COPY) {
1032 				msg = copy_msg(msg, copy);
1033 				goto out_unlock0;
1034 			}
1035 
1036 			list_del(&msg->m_list);
1037 			msq->q_qnum--;
1038 			msq->q_rtime = get_seconds();
1039 			msq->q_lrpid = task_tgid_vnr(current);
1040 			msq->q_cbytes -= msg->m_ts;
1041 			atomic_sub(msg->m_ts, &ns->msg_bytes);
1042 			atomic_dec(&ns->msg_hdrs);
1043 			ss_wakeup(msq, &wake_q, false);
1044 
1045 			goto out_unlock0;
1046 		}
1047 
1048 		/* No message waiting. Wait for a message */
1049 		if (msgflg & IPC_NOWAIT) {
1050 			msg = ERR_PTR(-ENOMSG);
1051 			goto out_unlock0;
1052 		}
1053 
1054 		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1055 		msr_d.r_tsk = current;
1056 		msr_d.r_msgtype = msgtyp;
1057 		msr_d.r_mode = mode;
1058 		if (msgflg & MSG_NOERROR)
1059 			msr_d.r_maxsize = INT_MAX;
1060 		else
1061 			msr_d.r_maxsize = bufsz;
1062 		msr_d.r_msg = ERR_PTR(-EAGAIN);
1063 		__set_current_state(TASK_INTERRUPTIBLE);
1064 
1065 		ipc_unlock_object(&msq->q_perm);
1066 		rcu_read_unlock();
1067 		schedule();
1068 
1069 		/*
1070 		 * Lockless receive, part 1:
1071 		 * We don't hold a reference to the queue and getting a
1072 		 * reference would defeat the idea of a lockless operation,
1073 		 * thus the code relies on rcu to guarantee the existence of
1074 		 * msq:
1075 		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1076 		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1077 		 */
1078 		rcu_read_lock();
1079 
1080 		/*
1081 		 * Lockless receive, part 2:
1082 		 * The work in pipelined_send() and expunge_all():
1083 		 * - Set pointer to message
1084 		 * - Queue the receiver task for later wakeup
1085 		 * - Wake up the process after the lock is dropped.
1086 		 *
1087 		 * Should the process wake up before this wakeup (due to a
1088 		 * signal) it will either see the message and continue ...
1089 		 */
1090 		msg = READ_ONCE(msr_d.r_msg);
1091 		if (msg != ERR_PTR(-EAGAIN))
1092 			goto out_unlock1;
1093 
1094 		 /*
1095 		  * ... or see -EAGAIN, acquire the lock to check the message
1096 		  * again.
1097 		  */
1098 		ipc_lock_object(&msq->q_perm);
1099 
1100 		msg = msr_d.r_msg;
1101 		if (msg != ERR_PTR(-EAGAIN))
1102 			goto out_unlock0;
1103 
1104 		list_del(&msr_d.r_list);
1105 		if (signal_pending(current)) {
1106 			msg = ERR_PTR(-ERESTARTNOHAND);
1107 			goto out_unlock0;
1108 		}
1109 
1110 		ipc_unlock_object(&msq->q_perm);
1111 	}
1112 
1113 out_unlock0:
1114 	ipc_unlock_object(&msq->q_perm);
1115 	wake_up_q(&wake_q);
1116 out_unlock1:
1117 	rcu_read_unlock();
1118 	if (IS_ERR(msg)) {
1119 		free_copy(copy);
1120 		return PTR_ERR(msg);
1121 	}
1122 
1123 	bufsz = msg_handler(buf, msg, bufsz);
1124 	free_msg(msg);
1125 
1126 	return bufsz;
1127 }
1128 
1129 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1130 		long, msgtyp, int, msgflg)
1131 {
1132 	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1133 }
1134 
1135 #ifdef CONFIG_COMPAT
1136 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1137 {
1138 	struct compat_msgbuf __user *msgp = dest;
1139 	size_t msgsz;
1140 
1141 	if (put_user(msg->m_type, &msgp->mtype))
1142 		return -EFAULT;
1143 
1144 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1145 	if (store_msg(msgp->mtext, msg, msgsz))
1146 		return -EFAULT;
1147 	return msgsz;
1148 }
1149 
1150 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1151 		       compat_ssize_t, msgsz, compat_long_t, msgtyp, int, msgflg)
1152 {
1153 	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1154 			 msgflg, compat_do_msg_fill);
1155 }
1156 #endif
1157 
1158 int msg_init_ns(struct ipc_namespace *ns)
1159 {
1160 	ns->msg_ctlmax = MSGMAX;
1161 	ns->msg_ctlmnb = MSGMNB;
1162 	ns->msg_ctlmni = MSGMNI;
1163 
1164 	atomic_set(&ns->msg_bytes, 0);
1165 	atomic_set(&ns->msg_hdrs, 0);
1166 	return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1167 }
1168 
1169 #ifdef CONFIG_IPC_NS
1170 void msg_exit_ns(struct ipc_namespace *ns)
1171 {
1172 	free_ipcs(ns, &msg_ids(ns), freeque);
1173 	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1174 	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1175 }
1176 #endif
1177 
1178 #ifdef CONFIG_PROC_FS
1179 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1180 {
1181 	struct user_namespace *user_ns = seq_user_ns(s);
1182 	struct kern_ipc_perm *ipcp = it;
1183 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1184 
1185 	seq_printf(s,
1186 		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1187 		   msq->q_perm.key,
1188 		   msq->q_perm.id,
1189 		   msq->q_perm.mode,
1190 		   msq->q_cbytes,
1191 		   msq->q_qnum,
1192 		   msq->q_lspid,
1193 		   msq->q_lrpid,
1194 		   from_kuid_munged(user_ns, msq->q_perm.uid),
1195 		   from_kgid_munged(user_ns, msq->q_perm.gid),
1196 		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1197 		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1198 		   msq->q_stime,
1199 		   msq->q_rtime,
1200 		   msq->q_ctime);
1201 
1202 	return 0;
1203 }
1204 #endif
1205 
1206 int __init msg_init(void)
1207 {
1208 	const int err = msg_init_ns(&init_ipc_ns);
1209 
1210 	ipc_init_proc_interface("sysvipc/msg",
1211 				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1212 				IPC_MSG_IDS, sysvipc_msg_proc_show);
1213 	return err;
1214 }
1215