msg.c revision 84f001e1
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.h>
34#include <linux/sched/wake_q.h>
35#include <linux/syscalls.h>
36#include <linux/audit.h>
37#include <linux/seq_file.h>
38#include <linux/rwsem.h>
39#include <linux/nsproxy.h>
40#include <linux/ipc_namespace.h>
41
42#include <asm/current.h>
43#include <linux/uaccess.h>
44#include "util.h"
45
46/* one msg_receiver structure for each sleeping receiver */
47struct msg_receiver {
48	struct list_head	r_list;
49	struct task_struct	*r_tsk;
50
51	int			r_mode;
52	long			r_msgtype;
53	long			r_maxsize;
54
55	struct msg_msg		*r_msg;
56};
57
58/* one msg_sender for each sleeping sender */
59struct msg_sender {
60	struct list_head	list;
61	struct task_struct	*tsk;
62	size_t                  msgsz;
63};
64
65#define SEARCH_ANY		1
66#define SEARCH_EQUAL		2
67#define SEARCH_NOTEQUAL		3
68#define SEARCH_LESSEQUAL	4
69#define SEARCH_NUMBER		5
70
71#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
72
73static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
74{
75	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
76
77	if (IS_ERR(ipcp))
78		return ERR_CAST(ipcp);
79
80	return container_of(ipcp, struct msg_queue, q_perm);
81}
82
83static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
84							int id)
85{
86	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
87
88	if (IS_ERR(ipcp))
89		return ERR_CAST(ipcp);
90
91	return container_of(ipcp, struct msg_queue, q_perm);
92}
93
94static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
95{
96	ipc_rmid(&msg_ids(ns), &s->q_perm);
97}
98
99static void msg_rcu_free(struct rcu_head *head)
100{
101	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
102	struct msg_queue *msq = ipc_rcu_to_struct(p);
103
104	security_msg_queue_free(msq);
105	ipc_rcu_free(head);
106}
107
108/**
109 * newque - Create a new msg queue
110 * @ns: namespace
111 * @params: ptr to the structure that contains the key and msgflg
112 *
113 * Called with msg_ids.rwsem held (writer)
114 */
115static int newque(struct ipc_namespace *ns, struct ipc_params *params)
116{
117	struct msg_queue *msq;
118	int id, retval;
119	key_t key = params->key;
120	int msgflg = params->flg;
121
122	msq = ipc_rcu_alloc(sizeof(*msq));
123	if (!msq)
124		return -ENOMEM;
125
126	msq->q_perm.mode = msgflg & S_IRWXUGO;
127	msq->q_perm.key = key;
128
129	msq->q_perm.security = NULL;
130	retval = security_msg_queue_alloc(msq);
131	if (retval) {
132		ipc_rcu_putref(msq, ipc_rcu_free);
133		return retval;
134	}
135
136	msq->q_stime = msq->q_rtime = 0;
137	msq->q_ctime = get_seconds();
138	msq->q_cbytes = msq->q_qnum = 0;
139	msq->q_qbytes = ns->msg_ctlmnb;
140	msq->q_lspid = msq->q_lrpid = 0;
141	INIT_LIST_HEAD(&msq->q_messages);
142	INIT_LIST_HEAD(&msq->q_receivers);
143	INIT_LIST_HEAD(&msq->q_senders);
144
145	/* ipc_addid() locks msq upon success. */
146	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
147	if (id < 0) {
148		ipc_rcu_putref(msq, msg_rcu_free);
149		return id;
150	}
151
152	ipc_unlock_object(&msq->q_perm);
153	rcu_read_unlock();
154
155	return msq->q_perm.id;
156}
157
158static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
159{
160	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
161		1 + msq->q_qnum <= msq->q_qbytes;
162}
163
164static inline void ss_add(struct msg_queue *msq,
165			  struct msg_sender *mss, size_t msgsz)
166{
167	mss->tsk = current;
168	mss->msgsz = msgsz;
169	__set_current_state(TASK_INTERRUPTIBLE);
170	list_add_tail(&mss->list, &msq->q_senders);
171}
172
173static inline void ss_del(struct msg_sender *mss)
174{
175	if (mss->list.next)
176		list_del(&mss->list);
177}
178
179static void ss_wakeup(struct msg_queue *msq,
180		      struct wake_q_head *wake_q, bool kill)
181{
182	struct msg_sender *mss, *t;
183	struct task_struct *stop_tsk = NULL;
184	struct list_head *h = &msq->q_senders;
185
186	list_for_each_entry_safe(mss, t, h, list) {
187		if (kill)
188			mss->list.next = NULL;
189
190		/*
191		 * Stop at the first task we don't wakeup,
192		 * we've already iterated the original
193		 * sender queue.
194		 */
195		else if (stop_tsk == mss->tsk)
196			break;
197		/*
198		 * We are not in an EIDRM scenario here, therefore
199		 * verify that we really need to wakeup the task.
200		 * To maintain current semantics and wakeup order,
201		 * move the sender to the tail on behalf of the
202		 * blocked task.
203		 */
204		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
205			if (!stop_tsk)
206				stop_tsk = mss->tsk;
207
208			list_move_tail(&mss->list, &msq->q_senders);
209			continue;
210		}
211
212		wake_q_add(wake_q, mss->tsk);
213	}
214}
215
216static void expunge_all(struct msg_queue *msq, int res,
217			struct wake_q_head *wake_q)
218{
219	struct msg_receiver *msr, *t;
220
221	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
222		wake_q_add(wake_q, msr->r_tsk);
223		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
224	}
225}
226
227/*
228 * freeque() wakes up waiters on the sender and receiver waiting queue,
229 * removes the message queue from message queue ID IDR, and cleans up all the
230 * messages associated with this queue.
231 *
232 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
233 * before freeque() is called. msg_ids.rwsem remains locked on exit.
234 */
235static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
236{
237	struct msg_msg *msg, *t;
238	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
239	DEFINE_WAKE_Q(wake_q);
240
241	expunge_all(msq, -EIDRM, &wake_q);
242	ss_wakeup(msq, &wake_q, true);
243	msg_rmid(ns, msq);
244	ipc_unlock_object(&msq->q_perm);
245	wake_up_q(&wake_q);
246	rcu_read_unlock();
247
248	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
249		atomic_dec(&ns->msg_hdrs);
250		free_msg(msg);
251	}
252	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
253	ipc_rcu_putref(msq, msg_rcu_free);
254}
255
256/*
257 * Called with msg_ids.rwsem and ipcp locked.
258 */
259static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
260{
261	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
262
263	return security_msg_queue_associate(msq, msgflg);
264}
265
266SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
267{
268	struct ipc_namespace *ns;
269	static const struct ipc_ops msg_ops = {
270		.getnew = newque,
271		.associate = msg_security,
272	};
273	struct ipc_params msg_params;
274
275	ns = current->nsproxy->ipc_ns;
276
277	msg_params.key = key;
278	msg_params.flg = msgflg;
279
280	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
281}
282
283static inline unsigned long
284copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
285{
286	switch (version) {
287	case IPC_64:
288		return copy_to_user(buf, in, sizeof(*in));
289	case IPC_OLD:
290	{
291		struct msqid_ds out;
292
293		memset(&out, 0, sizeof(out));
294
295		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
296
297		out.msg_stime		= in->msg_stime;
298		out.msg_rtime		= in->msg_rtime;
299		out.msg_ctime		= in->msg_ctime;
300
301		if (in->msg_cbytes > USHRT_MAX)
302			out.msg_cbytes	= USHRT_MAX;
303		else
304			out.msg_cbytes	= in->msg_cbytes;
305		out.msg_lcbytes		= in->msg_cbytes;
306
307		if (in->msg_qnum > USHRT_MAX)
308			out.msg_qnum	= USHRT_MAX;
309		else
310			out.msg_qnum	= in->msg_qnum;
311
312		if (in->msg_qbytes > USHRT_MAX)
313			out.msg_qbytes	= USHRT_MAX;
314		else
315			out.msg_qbytes	= in->msg_qbytes;
316		out.msg_lqbytes		= in->msg_qbytes;
317
318		out.msg_lspid		= in->msg_lspid;
319		out.msg_lrpid		= in->msg_lrpid;
320
321		return copy_to_user(buf, &out, sizeof(out));
322	}
323	default:
324		return -EINVAL;
325	}
326}
327
328static inline unsigned long
329copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
330{
331	switch (version) {
332	case IPC_64:
333		if (copy_from_user(out, buf, sizeof(*out)))
334			return -EFAULT;
335		return 0;
336	case IPC_OLD:
337	{
338		struct msqid_ds tbuf_old;
339
340		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
341			return -EFAULT;
342
343		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
344		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
345		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
346
347		if (tbuf_old.msg_qbytes == 0)
348			out->msg_qbytes	= tbuf_old.msg_lqbytes;
349		else
350			out->msg_qbytes	= tbuf_old.msg_qbytes;
351
352		return 0;
353	}
354	default:
355		return -EINVAL;
356	}
357}
358
359/*
360 * This function handles some msgctl commands which require the rwsem
361 * to be held in write mode.
362 * NOTE: no locks must be held, the rwsem is taken inside this function.
363 */
364static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
365		       struct msqid_ds __user *buf, int version)
366{
367	struct kern_ipc_perm *ipcp;
368	struct msqid64_ds uninitialized_var(msqid64);
369	struct msg_queue *msq;
370	int err;
371
372	if (cmd == IPC_SET) {
373		if (copy_msqid_from_user(&msqid64, buf, version))
374			return -EFAULT;
375	}
376
377	down_write(&msg_ids(ns).rwsem);
378	rcu_read_lock();
379
380	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
381				      &msqid64.msg_perm, msqid64.msg_qbytes);
382	if (IS_ERR(ipcp)) {
383		err = PTR_ERR(ipcp);
384		goto out_unlock1;
385	}
386
387	msq = container_of(ipcp, struct msg_queue, q_perm);
388
389	err = security_msg_queue_msgctl(msq, cmd);
390	if (err)
391		goto out_unlock1;
392
393	switch (cmd) {
394	case IPC_RMID:
395		ipc_lock_object(&msq->q_perm);
396		/* freeque unlocks the ipc object and rcu */
397		freeque(ns, ipcp);
398		goto out_up;
399	case IPC_SET:
400	{
401		DEFINE_WAKE_Q(wake_q);
402
403		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
404		    !capable(CAP_SYS_RESOURCE)) {
405			err = -EPERM;
406			goto out_unlock1;
407		}
408
409		ipc_lock_object(&msq->q_perm);
410		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
411		if (err)
412			goto out_unlock0;
413
414		msq->q_qbytes = msqid64.msg_qbytes;
415
416		msq->q_ctime = get_seconds();
417		/*
418		 * Sleeping receivers might be excluded by
419		 * stricter permissions.
420		 */
421		expunge_all(msq, -EAGAIN, &wake_q);
422		/*
423		 * Sleeping senders might be able to send
424		 * due to a larger queue size.
425		 */
426		ss_wakeup(msq, &wake_q, false);
427		ipc_unlock_object(&msq->q_perm);
428		wake_up_q(&wake_q);
429
430		goto out_unlock1;
431	}
432	default:
433		err = -EINVAL;
434		goto out_unlock1;
435	}
436
437out_unlock0:
438	ipc_unlock_object(&msq->q_perm);
439out_unlock1:
440	rcu_read_unlock();
441out_up:
442	up_write(&msg_ids(ns).rwsem);
443	return err;
444}
445
446static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
447			 int cmd, int version, void __user *buf)
448{
449	int err;
450	struct msg_queue *msq;
451
452	switch (cmd) {
453	case IPC_INFO:
454	case MSG_INFO:
455	{
456		struct msginfo msginfo;
457		int max_id;
458
459		if (!buf)
460			return -EFAULT;
461
462		/*
463		 * We must not return kernel stack data.
464		 * due to padding, it's not enough
465		 * to set all member fields.
466		 */
467		err = security_msg_queue_msgctl(NULL, cmd);
468		if (err)
469			return err;
470
471		memset(&msginfo, 0, sizeof(msginfo));
472		msginfo.msgmni = ns->msg_ctlmni;
473		msginfo.msgmax = ns->msg_ctlmax;
474		msginfo.msgmnb = ns->msg_ctlmnb;
475		msginfo.msgssz = MSGSSZ;
476		msginfo.msgseg = MSGSEG;
477		down_read(&msg_ids(ns).rwsem);
478		if (cmd == MSG_INFO) {
479			msginfo.msgpool = msg_ids(ns).in_use;
480			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
481			msginfo.msgtql = atomic_read(&ns->msg_bytes);
482		} else {
483			msginfo.msgmap = MSGMAP;
484			msginfo.msgpool = MSGPOOL;
485			msginfo.msgtql = MSGTQL;
486		}
487		max_id = ipc_get_maxid(&msg_ids(ns));
488		up_read(&msg_ids(ns).rwsem);
489		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
490			return -EFAULT;
491		return (max_id < 0) ? 0 : max_id;
492	}
493
494	case MSG_STAT:
495	case IPC_STAT:
496	{
497		struct msqid64_ds tbuf;
498		int success_return;
499
500		if (!buf)
501			return -EFAULT;
502
503		memset(&tbuf, 0, sizeof(tbuf));
504
505		rcu_read_lock();
506		if (cmd == MSG_STAT) {
507			msq = msq_obtain_object(ns, msqid);
508			if (IS_ERR(msq)) {
509				err = PTR_ERR(msq);
510				goto out_unlock;
511			}
512			success_return = msq->q_perm.id;
513		} else {
514			msq = msq_obtain_object_check(ns, msqid);
515			if (IS_ERR(msq)) {
516				err = PTR_ERR(msq);
517				goto out_unlock;
518			}
519			success_return = 0;
520		}
521
522		err = -EACCES;
523		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
524			goto out_unlock;
525
526		err = security_msg_queue_msgctl(msq, cmd);
527		if (err)
528			goto out_unlock;
529
530		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
531		tbuf.msg_stime  = msq->q_stime;
532		tbuf.msg_rtime  = msq->q_rtime;
533		tbuf.msg_ctime  = msq->q_ctime;
534		tbuf.msg_cbytes = msq->q_cbytes;
535		tbuf.msg_qnum   = msq->q_qnum;
536		tbuf.msg_qbytes = msq->q_qbytes;
537		tbuf.msg_lspid  = msq->q_lspid;
538		tbuf.msg_lrpid  = msq->q_lrpid;
539		rcu_read_unlock();
540
541		if (copy_msqid_to_user(buf, &tbuf, version))
542			return -EFAULT;
543		return success_return;
544	}
545
546	default:
547		return -EINVAL;
548	}
549
550	return err;
551out_unlock:
552	rcu_read_unlock();
553	return err;
554}
555
556SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
557{
558	int version;
559	struct ipc_namespace *ns;
560
561	if (msqid < 0 || cmd < 0)
562		return -EINVAL;
563
564	version = ipc_parse_version(&cmd);
565	ns = current->nsproxy->ipc_ns;
566
567	switch (cmd) {
568	case IPC_INFO:
569	case MSG_INFO:
570	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
571	case IPC_STAT:
572		return msgctl_nolock(ns, msqid, cmd, version, buf);
573	case IPC_SET:
574	case IPC_RMID:
575		return msgctl_down(ns, msqid, cmd, buf, version);
576	default:
577		return  -EINVAL;
578	}
579}
580
581static int testmsg(struct msg_msg *msg, long type, int mode)
582{
583	switch (mode) {
584	case SEARCH_ANY:
585	case SEARCH_NUMBER:
586		return 1;
587	case SEARCH_LESSEQUAL:
588		if (msg->m_type <= type)
589			return 1;
590		break;
591	case SEARCH_EQUAL:
592		if (msg->m_type == type)
593			return 1;
594		break;
595	case SEARCH_NOTEQUAL:
596		if (msg->m_type != type)
597			return 1;
598		break;
599	}
600	return 0;
601}
602
603static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
604				 struct wake_q_head *wake_q)
605{
606	struct msg_receiver *msr, *t;
607
608	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
609		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
610		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
611					       msr->r_msgtype, msr->r_mode)) {
612
613			list_del(&msr->r_list);
614			if (msr->r_maxsize < msg->m_ts) {
615				wake_q_add(wake_q, msr->r_tsk);
616				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
617			} else {
618				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
619				msq->q_rtime = get_seconds();
620
621				wake_q_add(wake_q, msr->r_tsk);
622				WRITE_ONCE(msr->r_msg, msg);
623				return 1;
624			}
625		}
626	}
627
628	return 0;
629}
630
631long do_msgsnd(int msqid, long mtype, void __user *mtext,
632		size_t msgsz, int msgflg)
633{
634	struct msg_queue *msq;
635	struct msg_msg *msg;
636	int err;
637	struct ipc_namespace *ns;
638	DEFINE_WAKE_Q(wake_q);
639
640	ns = current->nsproxy->ipc_ns;
641
642	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
643		return -EINVAL;
644	if (mtype < 1)
645		return -EINVAL;
646
647	msg = load_msg(mtext, msgsz);
648	if (IS_ERR(msg))
649		return PTR_ERR(msg);
650
651	msg->m_type = mtype;
652	msg->m_ts = msgsz;
653
654	rcu_read_lock();
655	msq = msq_obtain_object_check(ns, msqid);
656	if (IS_ERR(msq)) {
657		err = PTR_ERR(msq);
658		goto out_unlock1;
659	}
660
661	ipc_lock_object(&msq->q_perm);
662
663	for (;;) {
664		struct msg_sender s;
665
666		err = -EACCES;
667		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
668			goto out_unlock0;
669
670		/* raced with RMID? */
671		if (!ipc_valid_object(&msq->q_perm)) {
672			err = -EIDRM;
673			goto out_unlock0;
674		}
675
676		err = security_msg_queue_msgsnd(msq, msg, msgflg);
677		if (err)
678			goto out_unlock0;
679
680		if (msg_fits_inqueue(msq, msgsz))
681			break;
682
683		/* queue full, wait: */
684		if (msgflg & IPC_NOWAIT) {
685			err = -EAGAIN;
686			goto out_unlock0;
687		}
688
689		/* enqueue the sender and prepare to block */
690		ss_add(msq, &s, msgsz);
691
692		if (!ipc_rcu_getref(msq)) {
693			err = -EIDRM;
694			goto out_unlock0;
695		}
696
697		ipc_unlock_object(&msq->q_perm);
698		rcu_read_unlock();
699		schedule();
700
701		rcu_read_lock();
702		ipc_lock_object(&msq->q_perm);
703
704		ipc_rcu_putref(msq, msg_rcu_free);
705		/* raced with RMID? */
706		if (!ipc_valid_object(&msq->q_perm)) {
707			err = -EIDRM;
708			goto out_unlock0;
709		}
710		ss_del(&s);
711
712		if (signal_pending(current)) {
713			err = -ERESTARTNOHAND;
714			goto out_unlock0;
715		}
716
717	}
718
719	msq->q_lspid = task_tgid_vnr(current);
720	msq->q_stime = get_seconds();
721
722	if (!pipelined_send(msq, msg, &wake_q)) {
723		/* no one is waiting for this message, enqueue it */
724		list_add_tail(&msg->m_list, &msq->q_messages);
725		msq->q_cbytes += msgsz;
726		msq->q_qnum++;
727		atomic_add(msgsz, &ns->msg_bytes);
728		atomic_inc(&ns->msg_hdrs);
729	}
730
731	err = 0;
732	msg = NULL;
733
734out_unlock0:
735	ipc_unlock_object(&msq->q_perm);
736	wake_up_q(&wake_q);
737out_unlock1:
738	rcu_read_unlock();
739	if (msg != NULL)
740		free_msg(msg);
741	return err;
742}
743
744SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
745		int, msgflg)
746{
747	long mtype;
748
749	if (get_user(mtype, &msgp->mtype))
750		return -EFAULT;
751	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
752}
753
754static inline int convert_mode(long *msgtyp, int msgflg)
755{
756	if (msgflg & MSG_COPY)
757		return SEARCH_NUMBER;
758	/*
759	 *  find message of correct type.
760	 *  msgtyp = 0 => get first.
761	 *  msgtyp > 0 => get first message of matching type.
762	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
763	 */
764	if (*msgtyp == 0)
765		return SEARCH_ANY;
766	if (*msgtyp < 0) {
767		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
768			*msgtyp = LONG_MAX;
769		else
770			*msgtyp = -*msgtyp;
771		return SEARCH_LESSEQUAL;
772	}
773	if (msgflg & MSG_EXCEPT)
774		return SEARCH_NOTEQUAL;
775	return SEARCH_EQUAL;
776}
777
778static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
779{
780	struct msgbuf __user *msgp = dest;
781	size_t msgsz;
782
783	if (put_user(msg->m_type, &msgp->mtype))
784		return -EFAULT;
785
786	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
787	if (store_msg(msgp->mtext, msg, msgsz))
788		return -EFAULT;
789	return msgsz;
790}
791
792#ifdef CONFIG_CHECKPOINT_RESTORE
793/*
794 * This function creates new kernel message structure, large enough to store
795 * bufsz message bytes.
796 */
797static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
798{
799	struct msg_msg *copy;
800
801	/*
802	 * Create dummy message to copy real message to.
803	 */
804	copy = load_msg(buf, bufsz);
805	if (!IS_ERR(copy))
806		copy->m_ts = bufsz;
807	return copy;
808}
809
810static inline void free_copy(struct msg_msg *copy)
811{
812	if (copy)
813		free_msg(copy);
814}
815#else
816static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
817{
818	return ERR_PTR(-ENOSYS);
819}
820
821static inline void free_copy(struct msg_msg *copy)
822{
823}
824#endif
825
826static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
827{
828	struct msg_msg *msg, *found = NULL;
829	long count = 0;
830
831	list_for_each_entry(msg, &msq->q_messages, m_list) {
832		if (testmsg(msg, *msgtyp, mode) &&
833		    !security_msg_queue_msgrcv(msq, msg, current,
834					       *msgtyp, mode)) {
835			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
836				*msgtyp = msg->m_type - 1;
837				found = msg;
838			} else if (mode == SEARCH_NUMBER) {
839				if (*msgtyp == count)
840					return msg;
841			} else
842				return msg;
843			count++;
844		}
845	}
846
847	return found ?: ERR_PTR(-EAGAIN);
848}
849
850long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
851	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
852{
853	int mode;
854	struct msg_queue *msq;
855	struct ipc_namespace *ns;
856	struct msg_msg *msg, *copy = NULL;
857	DEFINE_WAKE_Q(wake_q);
858
859	ns = current->nsproxy->ipc_ns;
860
861	if (msqid < 0 || (long) bufsz < 0)
862		return -EINVAL;
863
864	if (msgflg & MSG_COPY) {
865		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
866			return -EINVAL;
867		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
868		if (IS_ERR(copy))
869			return PTR_ERR(copy);
870	}
871	mode = convert_mode(&msgtyp, msgflg);
872
873	rcu_read_lock();
874	msq = msq_obtain_object_check(ns, msqid);
875	if (IS_ERR(msq)) {
876		rcu_read_unlock();
877		free_copy(copy);
878		return PTR_ERR(msq);
879	}
880
881	for (;;) {
882		struct msg_receiver msr_d;
883
884		msg = ERR_PTR(-EACCES);
885		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
886			goto out_unlock1;
887
888		ipc_lock_object(&msq->q_perm);
889
890		/* raced with RMID? */
891		if (!ipc_valid_object(&msq->q_perm)) {
892			msg = ERR_PTR(-EIDRM);
893			goto out_unlock0;
894		}
895
896		msg = find_msg(msq, &msgtyp, mode);
897		if (!IS_ERR(msg)) {
898			/*
899			 * Found a suitable message.
900			 * Unlink it from the queue.
901			 */
902			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
903				msg = ERR_PTR(-E2BIG);
904				goto out_unlock0;
905			}
906			/*
907			 * If we are copying, then do not unlink message and do
908			 * not update queue parameters.
909			 */
910			if (msgflg & MSG_COPY) {
911				msg = copy_msg(msg, copy);
912				goto out_unlock0;
913			}
914
915			list_del(&msg->m_list);
916			msq->q_qnum--;
917			msq->q_rtime = get_seconds();
918			msq->q_lrpid = task_tgid_vnr(current);
919			msq->q_cbytes -= msg->m_ts;
920			atomic_sub(msg->m_ts, &ns->msg_bytes);
921			atomic_dec(&ns->msg_hdrs);
922			ss_wakeup(msq, &wake_q, false);
923
924			goto out_unlock0;
925		}
926
927		/* No message waiting. Wait for a message */
928		if (msgflg & IPC_NOWAIT) {
929			msg = ERR_PTR(-ENOMSG);
930			goto out_unlock0;
931		}
932
933		list_add_tail(&msr_d.r_list, &msq->q_receivers);
934		msr_d.r_tsk = current;
935		msr_d.r_msgtype = msgtyp;
936		msr_d.r_mode = mode;
937		if (msgflg & MSG_NOERROR)
938			msr_d.r_maxsize = INT_MAX;
939		else
940			msr_d.r_maxsize = bufsz;
941		msr_d.r_msg = ERR_PTR(-EAGAIN);
942		__set_current_state(TASK_INTERRUPTIBLE);
943
944		ipc_unlock_object(&msq->q_perm);
945		rcu_read_unlock();
946		schedule();
947
948		/*
949		 * Lockless receive, part 1:
950		 * We don't hold a reference to the queue and getting a
951		 * reference would defeat the idea of a lockless operation,
952		 * thus the code relies on rcu to guarantee the existence of
953		 * msq:
954		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
955		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
956		 */
957		rcu_read_lock();
958
959		/*
960		 * Lockless receive, part 2:
961		 * The work in pipelined_send() and expunge_all():
962		 * - Set pointer to message
963		 * - Queue the receiver task for later wakeup
964		 * - Wake up the process after the lock is dropped.
965		 *
966		 * Should the process wake up before this wakeup (due to a
967		 * signal) it will either see the message and continue ...
968		 */
969		msg = READ_ONCE(msr_d.r_msg);
970		if (msg != ERR_PTR(-EAGAIN))
971			goto out_unlock1;
972
973		 /*
974		  * ... or see -EAGAIN, acquire the lock to check the message
975		  * again.
976		  */
977		ipc_lock_object(&msq->q_perm);
978
979		msg = msr_d.r_msg;
980		if (msg != ERR_PTR(-EAGAIN))
981			goto out_unlock0;
982
983		list_del(&msr_d.r_list);
984		if (signal_pending(current)) {
985			msg = ERR_PTR(-ERESTARTNOHAND);
986			goto out_unlock0;
987		}
988
989		ipc_unlock_object(&msq->q_perm);
990	}
991
992out_unlock0:
993	ipc_unlock_object(&msq->q_perm);
994	wake_up_q(&wake_q);
995out_unlock1:
996	rcu_read_unlock();
997	if (IS_ERR(msg)) {
998		free_copy(copy);
999		return PTR_ERR(msg);
1000	}
1001
1002	bufsz = msg_handler(buf, msg, bufsz);
1003	free_msg(msg);
1004
1005	return bufsz;
1006}
1007
1008SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1009		long, msgtyp, int, msgflg)
1010{
1011	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1012}
1013
1014
1015void msg_init_ns(struct ipc_namespace *ns)
1016{
1017	ns->msg_ctlmax = MSGMAX;
1018	ns->msg_ctlmnb = MSGMNB;
1019	ns->msg_ctlmni = MSGMNI;
1020
1021	atomic_set(&ns->msg_bytes, 0);
1022	atomic_set(&ns->msg_hdrs, 0);
1023	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1024}
1025
1026#ifdef CONFIG_IPC_NS
1027void msg_exit_ns(struct ipc_namespace *ns)
1028{
1029	free_ipcs(ns, &msg_ids(ns), freeque);
1030	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1031}
1032#endif
1033
1034#ifdef CONFIG_PROC_FS
1035static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1036{
1037	struct user_namespace *user_ns = seq_user_ns(s);
1038	struct msg_queue *msq = it;
1039
1040	seq_printf(s,
1041		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1042		   msq->q_perm.key,
1043		   msq->q_perm.id,
1044		   msq->q_perm.mode,
1045		   msq->q_cbytes,
1046		   msq->q_qnum,
1047		   msq->q_lspid,
1048		   msq->q_lrpid,
1049		   from_kuid_munged(user_ns, msq->q_perm.uid),
1050		   from_kgid_munged(user_ns, msq->q_perm.gid),
1051		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1052		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1053		   msq->q_stime,
1054		   msq->q_rtime,
1055		   msq->q_ctime);
1056
1057	return 0;
1058}
1059#endif
1060
1061void __init msg_init(void)
1062{
1063	msg_init_ns(&init_ipc_ns);
1064
1065	ipc_init_proc_interface("sysvipc/msg",
1066				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1067				IPC_MSG_IDS, sysvipc_msg_proc_show);
1068}
1069