msg.c revision 55b7ae50
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/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 */
46struct 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	/*
55	 * Mark r_msg volatile so that the compiler
56	 * does not try to get smart and optimize
57	 * it. We rely on this for the lockless
58	 * receive algorithm.
59	 */
60	struct msg_msg		*volatile r_msg;
61};
62
63/* one msg_sender for each sleeping sender */
64struct msg_sender {
65	struct list_head	list;
66	struct task_struct	*tsk;
67};
68
69#define SEARCH_ANY		1
70#define SEARCH_EQUAL		2
71#define SEARCH_NOTEQUAL		3
72#define SEARCH_LESSEQUAL	4
73#define SEARCH_NUMBER		5
74
75#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
76
77static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
78{
79	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
80
81	if (IS_ERR(ipcp))
82		return ERR_CAST(ipcp);
83
84	return container_of(ipcp, struct msg_queue, q_perm);
85}
86
87static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
88							int id)
89{
90	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
91
92	if (IS_ERR(ipcp))
93		return ERR_CAST(ipcp);
94
95	return container_of(ipcp, struct msg_queue, q_perm);
96}
97
98static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
99{
100	ipc_rmid(&msg_ids(ns), &s->q_perm);
101}
102
103static void msg_rcu_free(struct rcu_head *head)
104{
105	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
106	struct msg_queue *msq = ipc_rcu_to_struct(p);
107
108	security_msg_queue_free(msq);
109	ipc_rcu_free(head);
110}
111
112/**
113 * newque - Create a new msg queue
114 * @ns: namespace
115 * @params: ptr to the structure that contains the key and msgflg
116 *
117 * Called with msg_ids.rwsem held (writer)
118 */
119static int newque(struct ipc_namespace *ns, struct ipc_params *params)
120{
121	struct msg_queue *msq;
122	int id, retval;
123	key_t key = params->key;
124	int msgflg = params->flg;
125
126	msq = ipc_rcu_alloc(sizeof(*msq));
127	if (!msq)
128		return -ENOMEM;
129
130	msq->q_perm.mode = msgflg & S_IRWXUGO;
131	msq->q_perm.key = key;
132
133	msq->q_perm.security = NULL;
134	retval = security_msg_queue_alloc(msq);
135	if (retval) {
136		ipc_rcu_putref(msq, ipc_rcu_free);
137		return retval;
138	}
139
140	/* ipc_addid() locks msq upon success. */
141	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
142	if (id < 0) {
143		ipc_rcu_putref(msq, msg_rcu_free);
144		return id;
145	}
146
147	msq->q_stime = msq->q_rtime = 0;
148	msq->q_ctime = get_seconds();
149	msq->q_cbytes = msq->q_qnum = 0;
150	msq->q_qbytes = ns->msg_ctlmnb;
151	msq->q_lspid = msq->q_lrpid = 0;
152	INIT_LIST_HEAD(&msq->q_messages);
153	INIT_LIST_HEAD(&msq->q_receivers);
154	INIT_LIST_HEAD(&msq->q_senders);
155
156	ipc_unlock_object(&msq->q_perm);
157	rcu_read_unlock();
158
159	return msq->q_perm.id;
160}
161
162static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
163{
164	mss->tsk = current;
165	__set_current_state(TASK_INTERRUPTIBLE);
166	list_add_tail(&mss->list, &msq->q_senders);
167}
168
169static inline void ss_del(struct msg_sender *mss)
170{
171	if (mss->list.next != NULL)
172		list_del(&mss->list);
173}
174
175static void ss_wakeup(struct list_head *h, int kill)
176{
177	struct msg_sender *mss, *t;
178
179	list_for_each_entry_safe(mss, t, h, list) {
180		if (kill)
181			mss->list.next = NULL;
182		wake_up_process(mss->tsk);
183	}
184}
185
186static void expunge_all(struct msg_queue *msq, int res)
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 */
192		wake_up_process(msr->r_tsk);
193		/*
194		 * Ensure that the wakeup is visible before setting r_msg as
195		 * the receiving end depends on it: either spinning on a nil,
196		 * or dealing with -EAGAIN cases. See lockless receive part 1
197		 * and 2 in do_msgrcv().
198		 */
199		smp_wmb(); /* barrier (B) */
200		msr->r_msg = ERR_PTR(res);
201	}
202}
203
204/*
205 * freeque() wakes up waiters on the sender and receiver waiting queue,
206 * removes the message queue from message queue ID IDR, and cleans up all the
207 * messages associated with this queue.
208 *
209 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
210 * before freeque() is called. msg_ids.rwsem remains locked on exit.
211 */
212static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
213{
214	struct msg_msg *msg, *t;
215	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
216
217	expunge_all(msq, -EIDRM);
218	ss_wakeup(&msq->q_senders, 1);
219	msg_rmid(ns, msq);
220	ipc_unlock_object(&msq->q_perm);
221	rcu_read_unlock();
222
223	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
224		atomic_dec(&ns->msg_hdrs);
225		free_msg(msg);
226	}
227	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
228	ipc_rcu_putref(msq, msg_rcu_free);
229}
230
231/*
232 * Called with msg_ids.rwsem and ipcp locked.
233 */
234static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
235{
236	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
237
238	return security_msg_queue_associate(msq, msgflg);
239}
240
241SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
242{
243	struct ipc_namespace *ns;
244	static const struct ipc_ops msg_ops = {
245		.getnew = newque,
246		.associate = msg_security,
247	};
248	struct ipc_params msg_params;
249
250	ns = current->nsproxy->ipc_ns;
251
252	msg_params.key = key;
253	msg_params.flg = msgflg;
254
255	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
256}
257
258static inline unsigned long
259copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
260{
261	switch (version) {
262	case IPC_64:
263		return copy_to_user(buf, in, sizeof(*in));
264	case IPC_OLD:
265	{
266		struct msqid_ds out;
267
268		memset(&out, 0, sizeof(out));
269
270		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
271
272		out.msg_stime		= in->msg_stime;
273		out.msg_rtime		= in->msg_rtime;
274		out.msg_ctime		= in->msg_ctime;
275
276		if (in->msg_cbytes > USHRT_MAX)
277			out.msg_cbytes	= USHRT_MAX;
278		else
279			out.msg_cbytes	= in->msg_cbytes;
280		out.msg_lcbytes		= in->msg_cbytes;
281
282		if (in->msg_qnum > USHRT_MAX)
283			out.msg_qnum	= USHRT_MAX;
284		else
285			out.msg_qnum	= in->msg_qnum;
286
287		if (in->msg_qbytes > USHRT_MAX)
288			out.msg_qbytes	= USHRT_MAX;
289		else
290			out.msg_qbytes	= in->msg_qbytes;
291		out.msg_lqbytes		= in->msg_qbytes;
292
293		out.msg_lspid		= in->msg_lspid;
294		out.msg_lrpid		= in->msg_lrpid;
295
296		return copy_to_user(buf, &out, sizeof(out));
297	}
298	default:
299		return -EINVAL;
300	}
301}
302
303static inline unsigned long
304copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
305{
306	switch (version) {
307	case IPC_64:
308		if (copy_from_user(out, buf, sizeof(*out)))
309			return -EFAULT;
310		return 0;
311	case IPC_OLD:
312	{
313		struct msqid_ds tbuf_old;
314
315		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
316			return -EFAULT;
317
318		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
319		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
320		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
321
322		if (tbuf_old.msg_qbytes == 0)
323			out->msg_qbytes	= tbuf_old.msg_lqbytes;
324		else
325			out->msg_qbytes	= tbuf_old.msg_qbytes;
326
327		return 0;
328	}
329	default:
330		return -EINVAL;
331	}
332}
333
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 */
339static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
340		       struct msqid_ds __user *buf, int version)
341{
342	struct kern_ipc_perm *ipcp;
343	struct msqid64_ds uninitialized_var(msqid64);
344	struct msg_queue *msq;
345	int err;
346
347	if (cmd == IPC_SET) {
348		if (copy_msqid_from_user(&msqid64, buf, version))
349			return -EFAULT;
350	}
351
352	down_write(&msg_ids(ns).rwsem);
353	rcu_read_lock();
354
355	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
356				      &msqid64.msg_perm, msqid64.msg_qbytes);
357	if (IS_ERR(ipcp)) {
358		err = PTR_ERR(ipcp);
359		goto out_unlock1;
360	}
361
362	msq = container_of(ipcp, struct msg_queue, q_perm);
363
364	err = security_msg_queue_msgctl(msq, cmd);
365	if (err)
366		goto out_unlock1;
367
368	switch (cmd) {
369	case IPC_RMID:
370		ipc_lock_object(&msq->q_perm);
371		/* freeque unlocks the ipc object and rcu */
372		freeque(ns, ipcp);
373		goto out_up;
374	case IPC_SET:
375		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
376		    !capable(CAP_SYS_RESOURCE)) {
377			err = -EPERM;
378			goto out_unlock1;
379		}
380
381		ipc_lock_object(&msq->q_perm);
382		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
383		if (err)
384			goto out_unlock0;
385
386		msq->q_qbytes = msqid64.msg_qbytes;
387
388		msq->q_ctime = get_seconds();
389		/* sleeping receivers might be excluded by
390		 * stricter permissions.
391		 */
392		expunge_all(msq, -EAGAIN);
393		/* sleeping senders might be able to send
394		 * due to a larger queue size.
395		 */
396		ss_wakeup(&msq->q_senders, 0);
397		break;
398	default:
399		err = -EINVAL;
400		goto out_unlock1;
401	}
402
403out_unlock0:
404	ipc_unlock_object(&msq->q_perm);
405out_unlock1:
406	rcu_read_unlock();
407out_up:
408	up_write(&msg_ids(ns).rwsem);
409	return err;
410}
411
412static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
413			 int cmd, int version, void __user *buf)
414{
415	int err;
416	struct msg_queue *msq;
417
418	switch (cmd) {
419	case IPC_INFO:
420	case MSG_INFO:
421	{
422		struct msginfo msginfo;
423		int max_id;
424
425		if (!buf)
426			return -EFAULT;
427
428		/*
429		 * We must not return kernel stack data.
430		 * due to padding, it's not enough
431		 * to set all member fields.
432		 */
433		err = security_msg_queue_msgctl(NULL, cmd);
434		if (err)
435			return err;
436
437		memset(&msginfo, 0, sizeof(msginfo));
438		msginfo.msgmni = ns->msg_ctlmni;
439		msginfo.msgmax = ns->msg_ctlmax;
440		msginfo.msgmnb = ns->msg_ctlmnb;
441		msginfo.msgssz = MSGSSZ;
442		msginfo.msgseg = MSGSEG;
443		down_read(&msg_ids(ns).rwsem);
444		if (cmd == MSG_INFO) {
445			msginfo.msgpool = msg_ids(ns).in_use;
446			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
447			msginfo.msgtql = atomic_read(&ns->msg_bytes);
448		} else {
449			msginfo.msgmap = MSGMAP;
450			msginfo.msgpool = MSGPOOL;
451			msginfo.msgtql = MSGTQL;
452		}
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)))
456			return -EFAULT;
457		return (max_id < 0) ? 0 : max_id;
458	}
459
460	case MSG_STAT:
461	case IPC_STAT:
462	{
463		struct msqid64_ds tbuf;
464		int success_return;
465
466		if (!buf)
467			return -EFAULT;
468
469		memset(&tbuf, 0, sizeof(tbuf));
470
471		rcu_read_lock();
472		if (cmd == MSG_STAT) {
473			msq = msq_obtain_object(ns, msqid);
474			if (IS_ERR(msq)) {
475				err = PTR_ERR(msq);
476				goto out_unlock;
477			}
478			success_return = msq->q_perm.id;
479		} else {
480			msq = msq_obtain_object_check(ns, msqid);
481			if (IS_ERR(msq)) {
482				err = PTR_ERR(msq);
483				goto out_unlock;
484			}
485			success_return = 0;
486		}
487
488		err = -EACCES;
489		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
490			goto out_unlock;
491
492		err = security_msg_queue_msgctl(msq, cmd);
493		if (err)
494			goto out_unlock;
495
496		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
497		tbuf.msg_stime  = msq->q_stime;
498		tbuf.msg_rtime  = msq->q_rtime;
499		tbuf.msg_ctime  = msq->q_ctime;
500		tbuf.msg_cbytes = msq->q_cbytes;
501		tbuf.msg_qnum   = msq->q_qnum;
502		tbuf.msg_qbytes = msq->q_qbytes;
503		tbuf.msg_lspid  = msq->q_lspid;
504		tbuf.msg_lrpid  = msq->q_lrpid;
505		rcu_read_unlock();
506
507		if (copy_msqid_to_user(buf, &tbuf, version))
508			return -EFAULT;
509		return success_return;
510	}
511
512	default:
513		return -EINVAL;
514	}
515
516	return err;
517out_unlock:
518	rcu_read_unlock();
519	return err;
520}
521
522SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
523{
524	int version;
525	struct ipc_namespace *ns;
526
527	if (msqid < 0 || cmd < 0)
528		return -EINVAL;
529
530	version = ipc_parse_version(&cmd);
531	ns = current->nsproxy->ipc_ns;
532
533	switch (cmd) {
534	case IPC_INFO:
535	case MSG_INFO:
536	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
537	case IPC_STAT:
538		return msgctl_nolock(ns, msqid, cmd, version, buf);
539	case IPC_SET:
540	case IPC_RMID:
541		return msgctl_down(ns, msqid, cmd, buf, version);
542	default:
543		return  -EINVAL;
544	}
545}
546
547static int testmsg(struct msg_msg *msg, long type, int mode)
548{
549	switch (mode) {
550	case SEARCH_ANY:
551	case SEARCH_NUMBER:
552		return 1;
553	case SEARCH_LESSEQUAL:
554		if (msg->m_type <= type)
555			return 1;
556		break;
557	case SEARCH_EQUAL:
558		if (msg->m_type == type)
559			return 1;
560		break;
561	case SEARCH_NOTEQUAL:
562		if (msg->m_type != type)
563			return 1;
564		break;
565	}
566	return 0;
567}
568
569static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
570{
571	struct msg_receiver *msr, *t;
572
573	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
574		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
575		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
576					       msr->r_msgtype, msr->r_mode)) {
577
578			list_del(&msr->r_list);
579			if (msr->r_maxsize < msg->m_ts) {
580				/* initialize pipelined send ordering */
581				msr->r_msg = NULL;
582				wake_up_process(msr->r_tsk);
583				/* barrier (B) see barrier comment below */
584				smp_wmb();
585				msr->r_msg = ERR_PTR(-E2BIG);
586			} else {
587				msr->r_msg = NULL;
588				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
589				msq->q_rtime = get_seconds();
590				wake_up_process(msr->r_tsk);
591				/*
592				 * Ensure that the wakeup is visible before
593				 * setting r_msg, as the receiving can otherwise
594				 * exit - once r_msg is set, the receiver can
595				 * continue. See lockless receive part 1 and 2
596				 * in do_msgrcv(). Barrier (B).
597				 */
598				smp_wmb();
599				msr->r_msg = msg;
600
601				return 1;
602			}
603		}
604	}
605
606	return 0;
607}
608
609long do_msgsnd(int msqid, long mtype, void __user *mtext,
610		size_t msgsz, int msgflg)
611{
612	struct msg_queue *msq;
613	struct msg_msg *msg;
614	int err;
615	struct ipc_namespace *ns;
616
617	ns = current->nsproxy->ipc_ns;
618
619	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
620		return -EINVAL;
621	if (mtype < 1)
622		return -EINVAL;
623
624	msg = load_msg(mtext, msgsz);
625	if (IS_ERR(msg))
626		return PTR_ERR(msg);
627
628	msg->m_type = mtype;
629	msg->m_ts = msgsz;
630
631	rcu_read_lock();
632	msq = msq_obtain_object_check(ns, msqid);
633	if (IS_ERR(msq)) {
634		err = PTR_ERR(msq);
635		goto out_unlock1;
636	}
637
638	ipc_lock_object(&msq->q_perm);
639
640	for (;;) {
641		struct msg_sender s;
642
643		err = -EACCES;
644		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
645			goto out_unlock0;
646
647		/* raced with RMID? */
648		if (!ipc_valid_object(&msq->q_perm)) {
649			err = -EIDRM;
650			goto out_unlock0;
651		}
652
653		err = security_msg_queue_msgsnd(msq, msg, msgflg);
654		if (err)
655			goto out_unlock0;
656
657		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
658				1 + msq->q_qnum <= msq->q_qbytes) {
659			break;
660		}
661
662		/* queue full, wait: */
663		if (msgflg & IPC_NOWAIT) {
664			err = -EAGAIN;
665			goto out_unlock0;
666		}
667
668		/* enqueue the sender and prepare to block */
669		ss_add(msq, &s);
670
671		if (!ipc_rcu_getref(msq)) {
672			err = -EIDRM;
673			goto out_unlock0;
674		}
675
676		ipc_unlock_object(&msq->q_perm);
677		rcu_read_unlock();
678		schedule();
679
680		rcu_read_lock();
681		ipc_lock_object(&msq->q_perm);
682
683		ipc_rcu_putref(msq, ipc_rcu_free);
684		/* raced with RMID? */
685		if (!ipc_valid_object(&msq->q_perm)) {
686			err = -EIDRM;
687			goto out_unlock0;
688		}
689
690		ss_del(&s);
691
692		if (signal_pending(current)) {
693			err = -ERESTARTNOHAND;
694			goto out_unlock0;
695		}
696
697	}
698	msq->q_lspid = task_tgid_vnr(current);
699	msq->q_stime = get_seconds();
700
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);
704		msq->q_cbytes += msgsz;
705		msq->q_qnum++;
706		atomic_add(msgsz, &ns->msg_bytes);
707		atomic_inc(&ns->msg_hdrs);
708	}
709
710	err = 0;
711	msg = NULL;
712
713out_unlock0:
714	ipc_unlock_object(&msq->q_perm);
715out_unlock1:
716	rcu_read_unlock();
717	if (msg != NULL)
718		free_msg(msg);
719	return err;
720}
721
722SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
723		int, msgflg)
724{
725	long mtype;
726
727	if (get_user(mtype, &msgp->mtype))
728		return -EFAULT;
729	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
730}
731
732static inline int convert_mode(long *msgtyp, int msgflg)
733{
734	if (msgflg & MSG_COPY)
735		return SEARCH_NUMBER;
736	/*
737	 *  find message of correct type.
738	 *  msgtyp = 0 => get first.
739	 *  msgtyp > 0 => get first message of matching type.
740	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
741	 */
742	if (*msgtyp == 0)
743		return SEARCH_ANY;
744	if (*msgtyp < 0) {
745		*msgtyp = -*msgtyp;
746		return SEARCH_LESSEQUAL;
747	}
748	if (msgflg & MSG_EXCEPT)
749		return SEARCH_NOTEQUAL;
750	return SEARCH_EQUAL;
751}
752
753static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
754{
755	struct msgbuf __user *msgp = dest;
756	size_t msgsz;
757
758	if (put_user(msg->m_type, &msgp->mtype))
759		return -EFAULT;
760
761	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
762	if (store_msg(msgp->mtext, msg, msgsz))
763		return -EFAULT;
764	return msgsz;
765}
766
767#ifdef CONFIG_CHECKPOINT_RESTORE
768/*
769 * This function creates new kernel message structure, large enough to store
770 * bufsz message bytes.
771 */
772static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
773{
774	struct msg_msg *copy;
775
776	/*
777	 * Create dummy message to copy real message to.
778	 */
779	copy = load_msg(buf, bufsz);
780	if (!IS_ERR(copy))
781		copy->m_ts = bufsz;
782	return copy;
783}
784
785static inline void free_copy(struct msg_msg *copy)
786{
787	if (copy)
788		free_msg(copy);
789}
790#else
791static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
792{
793	return ERR_PTR(-ENOSYS);
794}
795
796static inline void free_copy(struct msg_msg *copy)
797{
798}
799#endif
800
801static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
802{
803	struct msg_msg *msg, *found = NULL;
804	long count = 0;
805
806	list_for_each_entry(msg, &msq->q_messages, m_list) {
807		if (testmsg(msg, *msgtyp, mode) &&
808		    !security_msg_queue_msgrcv(msq, msg, current,
809					       *msgtyp, mode)) {
810			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
811				*msgtyp = msg->m_type - 1;
812				found = msg;
813			} else if (mode == SEARCH_NUMBER) {
814				if (*msgtyp == count)
815					return msg;
816			} else
817				return msg;
818			count++;
819		}
820	}
821
822	return found ?: ERR_PTR(-EAGAIN);
823}
824
825long 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))
827{
828	int mode;
829	struct msg_queue *msq;
830	struct ipc_namespace *ns;
831	struct msg_msg *msg, *copy = NULL;
832
833	ns = current->nsproxy->ipc_ns;
834
835	if (msqid < 0 || (long) bufsz < 0)
836		return -EINVAL;
837
838	if (msgflg & MSG_COPY) {
839		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
840			return -EINVAL;
841		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
842		if (IS_ERR(copy))
843			return PTR_ERR(copy);
844	}
845	mode = convert_mode(&msgtyp, msgflg);
846
847	rcu_read_lock();
848	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	}
854
855	for (;;) {
856		struct msg_receiver msr_d;
857
858		msg = ERR_PTR(-EACCES);
859		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
860			goto out_unlock1;
861
862		ipc_lock_object(&msq->q_perm);
863
864		/* raced with RMID? */
865		if (!ipc_valid_object(&msq->q_perm)) {
866			msg = ERR_PTR(-EIDRM);
867			goto out_unlock0;
868		}
869
870		msg = find_msg(msq, &msgtyp, mode);
871		if (!IS_ERR(msg)) {
872			/*
873			 * Found a suitable message.
874			 * Unlink it from the queue.
875			 */
876			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
877				msg = ERR_PTR(-E2BIG);
878				goto out_unlock0;
879			}
880			/*
881			 * If we are copying, then do not unlink message and do
882			 * not update queue parameters.
883			 */
884			if (msgflg & MSG_COPY) {
885				msg = copy_msg(msg, copy);
886				goto out_unlock0;
887			}
888
889			list_del(&msg->m_list);
890			msq->q_qnum--;
891			msq->q_rtime = get_seconds();
892			msq->q_lrpid = task_tgid_vnr(current);
893			msq->q_cbytes -= msg->m_ts;
894			atomic_sub(msg->m_ts, &ns->msg_bytes);
895			atomic_dec(&ns->msg_hdrs);
896			ss_wakeup(&msq->q_senders, 0);
897
898			goto out_unlock0;
899		}
900
901		/* No message waiting. Wait for a message */
902		if (msgflg & IPC_NOWAIT) {
903			msg = ERR_PTR(-ENOMSG);
904			goto out_unlock0;
905		}
906
907		list_add_tail(&msr_d.r_list, &msq->q_receivers);
908		msr_d.r_tsk = current;
909		msr_d.r_msgtype = msgtyp;
910		msr_d.r_mode = mode;
911		if (msgflg & MSG_NOERROR)
912			msr_d.r_maxsize = INT_MAX;
913		else
914			msr_d.r_maxsize = bufsz;
915		msr_d.r_msg = ERR_PTR(-EAGAIN);
916		__set_current_state(TASK_INTERRUPTIBLE);
917
918		ipc_unlock_object(&msq->q_perm);
919		rcu_read_unlock();
920		schedule();
921
922		/* Lockless receive, part 1:
923		 * Disable preemption.  We don't hold a reference to the queue
924		 * and getting a reference would defeat the idea of a lockless
925		 * operation, thus the code relies on rcu to guarantee the
926		 * existence of msq:
927		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
928		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
929		 * rcu_read_lock() prevents preemption between reading r_msg
930		 * and acquiring the q_perm.lock in ipc_lock_object().
931		 */
932		rcu_read_lock();
933
934		/* Lockless receive, part 2:
935		 * Wait until pipelined_send or expunge_all are outside of
936		 * wake_up_process(). There is a race with exit(), see
937		 * ipc/mqueue.c for the details. The correct serialization
938		 * ensures that a receiver cannot continue without the wakeup
939		 * being visibible _before_ setting r_msg:
940		 *
941		 * CPU 0                             CPU 1
942		 * <loop receiver>
943		 *   smp_rmb(); (A) <-- pair -.      <waker thread>
944		 *   <load ->r_msg>           |        msr->r_msg = NULL;
945		 *                            |        wake_up_process();
946		 * <continue>                 `------> smp_wmb(); (B)
947		 *                                     msr->r_msg = msg;
948		 *
949		 * Where (A) orders the message value read and where (B) orders
950		 * the write to the r_msg -- done in both pipelined_send and
951		 * expunge_all.
952		 */
953		for (;;) {
954			/*
955			 * Pairs with writer barrier in pipelined_send
956			 * or expunge_all.
957			 */
958			smp_rmb(); /* barrier (A) */
959			msg = (struct msg_msg *)msr_d.r_msg;
960			if (msg)
961				break;
962
963			/*
964			 * The cpu_relax() call is a compiler barrier
965			 * which forces everything in this loop to be
966			 * re-loaded.
967			 */
968			cpu_relax();
969		}
970
971		/* Lockless receive, part 3:
972		 * If there is a message or an error then accept it without
973		 * locking.
974		 */
975		if (msg != ERR_PTR(-EAGAIN))
976			goto out_unlock1;
977
978		/* Lockless receive, part 3:
979		 * Acquire the queue spinlock.
980		 */
981		ipc_lock_object(&msq->q_perm);
982
983		/* Lockless receive, part 4:
984		 * Repeat test after acquiring the spinlock.
985		 */
986		msg = (struct msg_msg *)msr_d.r_msg;
987		if (msg != ERR_PTR(-EAGAIN))
988			goto out_unlock0;
989
990		list_del(&msr_d.r_list);
991		if (signal_pending(current)) {
992			msg = ERR_PTR(-ERESTARTNOHAND);
993			goto out_unlock0;
994		}
995
996		ipc_unlock_object(&msq->q_perm);
997	}
998
999out_unlock0:
1000	ipc_unlock_object(&msq->q_perm);
1001out_unlock1:
1002	rcu_read_unlock();
1003	if (IS_ERR(msg)) {
1004		free_copy(copy);
1005		return PTR_ERR(msg);
1006	}
1007
1008	bufsz = msg_handler(buf, msg, bufsz);
1009	free_msg(msg);
1010
1011	return bufsz;
1012}
1013
1014SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1015		long, msgtyp, int, msgflg)
1016{
1017	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1018}
1019
1020
1021void msg_init_ns(struct ipc_namespace *ns)
1022{
1023	ns->msg_ctlmax = MSGMAX;
1024	ns->msg_ctlmnb = MSGMNB;
1025	ns->msg_ctlmni = MSGMNI;
1026
1027	atomic_set(&ns->msg_bytes, 0);
1028	atomic_set(&ns->msg_hdrs, 0);
1029	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1030}
1031
1032#ifdef CONFIG_IPC_NS
1033void msg_exit_ns(struct ipc_namespace *ns)
1034{
1035	free_ipcs(ns, &msg_ids(ns), freeque);
1036	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1037}
1038#endif
1039
1040#ifdef CONFIG_PROC_FS
1041static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1042{
1043	struct user_namespace *user_ns = seq_user_ns(s);
1044	struct msg_queue *msq = it;
1045
1046	seq_printf(s,
1047		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1048		   msq->q_perm.key,
1049		   msq->q_perm.id,
1050		   msq->q_perm.mode,
1051		   msq->q_cbytes,
1052		   msq->q_qnum,
1053		   msq->q_lspid,
1054		   msq->q_lrpid,
1055		   from_kuid_munged(user_ns, msq->q_perm.uid),
1056		   from_kgid_munged(user_ns, msq->q_perm.gid),
1057		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1058		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1059		   msq->q_stime,
1060		   msq->q_rtime,
1061		   msq->q_ctime);
1062
1063	return 0;
1064}
1065#endif
1066
1067void __init msg_init(void)
1068{
1069	msg_init_ns(&init_ipc_ns);
1070
1071	ipc_init_proc_interface("sysvipc/msg",
1072				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1073				IPC_MSG_IDS, sysvipc_msg_proc_show);
1074}
1075