msg.c revision 9ef5932f
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 */
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	struct msg_msg		*r_msg;
55};
56
57/* one msg_sender for each sleeping sender */
58struct 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
72static 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
82static 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
93static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
94{
95	ipc_rmid(&msg_ids(ns), &s->q_perm);
96}
97
98static void __msg_free(struct msg_queue *msq)
99{
100	kvfree(msq);
101}
102
103static void msg_rcu_free(struct rcu_head *head)
104{
105	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
106	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
107
108	security_msg_queue_free(msq);
109	__msg_free(msq);
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	BUILD_BUG_ON(offsetof(struct msg_queue, q_perm) != 0);
127
128	msq = container_of(ipc_rcu_alloc(sizeof(*msq)), struct msg_queue,
129				q_perm);
130	if (!msq)
131		return -ENOMEM;
132
133	msq->q_perm.mode = msgflg & S_IRWXUGO;
134	msq->q_perm.key = key;
135
136	msq->q_perm.security = NULL;
137	retval = security_msg_queue_alloc(msq);
138	if (retval) {
139		__msg_free(msq);
140		return retval;
141	}
142
143	msq->q_stime = msq->q_rtime = 0;
144	msq->q_ctime = get_seconds();
145	msq->q_cbytes = msq->q_qnum = 0;
146	msq->q_qbytes = ns->msg_ctlmnb;
147	msq->q_lspid = msq->q_lrpid = 0;
148	INIT_LIST_HEAD(&msq->q_messages);
149	INIT_LIST_HEAD(&msq->q_receivers);
150	INIT_LIST_HEAD(&msq->q_senders);
151
152	/* ipc_addid() locks msq upon success. */
153	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
154	if (id < 0) {
155		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
156		return id;
157	}
158
159	ipc_unlock_object(&msq->q_perm);
160	rcu_read_unlock();
161
162	return msq->q_perm.id;
163}
164
165static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
166{
167	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
168		1 + msq->q_qnum <= msq->q_qbytes;
169}
170
171static inline void ss_add(struct msg_queue *msq,
172			  struct msg_sender *mss, size_t msgsz)
173{
174	mss->tsk = current;
175	mss->msgsz = msgsz;
176	__set_current_state(TASK_INTERRUPTIBLE);
177	list_add_tail(&mss->list, &msq->q_senders);
178}
179
180static inline void ss_del(struct msg_sender *mss)
181{
182	if (mss->list.next)
183		list_del(&mss->list);
184}
185
186static void ss_wakeup(struct msg_queue *msq,
187		      struct wake_q_head *wake_q, bool kill)
188{
189	struct msg_sender *mss, *t;
190	struct task_struct *stop_tsk = NULL;
191	struct list_head *h = &msq->q_senders;
192
193	list_for_each_entry_safe(mss, t, h, list) {
194		if (kill)
195			mss->list.next = NULL;
196
197		/*
198		 * Stop at the first task we don't wakeup,
199		 * we've already iterated the original
200		 * sender queue.
201		 */
202		else if (stop_tsk == mss->tsk)
203			break;
204		/*
205		 * We are not in an EIDRM scenario here, therefore
206		 * verify that we really need to wakeup the task.
207		 * To maintain current semantics and wakeup order,
208		 * move the sender to the tail on behalf of the
209		 * blocked task.
210		 */
211		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
212			if (!stop_tsk)
213				stop_tsk = mss->tsk;
214
215			list_move_tail(&mss->list, &msq->q_senders);
216			continue;
217		}
218
219		wake_q_add(wake_q, mss->tsk);
220	}
221}
222
223static void expunge_all(struct msg_queue *msq, int res,
224			struct wake_q_head *wake_q)
225{
226	struct msg_receiver *msr, *t;
227
228	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
229		wake_q_add(wake_q, msr->r_tsk);
230		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
231	}
232}
233
234/*
235 * freeque() wakes up waiters on the sender and receiver waiting queue,
236 * removes the message queue from message queue ID IDR, and cleans up all the
237 * messages associated with this queue.
238 *
239 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
240 * before freeque() is called. msg_ids.rwsem remains locked on exit.
241 */
242static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
243{
244	struct msg_msg *msg, *t;
245	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
246	DEFINE_WAKE_Q(wake_q);
247
248	expunge_all(msq, -EIDRM, &wake_q);
249	ss_wakeup(msq, &wake_q, true);
250	msg_rmid(ns, msq);
251	ipc_unlock_object(&msq->q_perm);
252	wake_up_q(&wake_q);
253	rcu_read_unlock();
254
255	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
256		atomic_dec(&ns->msg_hdrs);
257		free_msg(msg);
258	}
259	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
260	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
261}
262
263/*
264 * Called with msg_ids.rwsem and ipcp locked.
265 */
266static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
267{
268	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
269
270	return security_msg_queue_associate(msq, msgflg);
271}
272
273SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
274{
275	struct ipc_namespace *ns;
276	static const struct ipc_ops msg_ops = {
277		.getnew = newque,
278		.associate = msg_security,
279	};
280	struct ipc_params msg_params;
281
282	ns = current->nsproxy->ipc_ns;
283
284	msg_params.key = key;
285	msg_params.flg = msgflg;
286
287	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
288}
289
290static inline unsigned long
291copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
292{
293	switch (version) {
294	case IPC_64:
295		return copy_to_user(buf, in, sizeof(*in));
296	case IPC_OLD:
297	{
298		struct msqid_ds out;
299
300		memset(&out, 0, sizeof(out));
301
302		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
303
304		out.msg_stime		= in->msg_stime;
305		out.msg_rtime		= in->msg_rtime;
306		out.msg_ctime		= in->msg_ctime;
307
308		if (in->msg_cbytes > USHRT_MAX)
309			out.msg_cbytes	= USHRT_MAX;
310		else
311			out.msg_cbytes	= in->msg_cbytes;
312		out.msg_lcbytes		= in->msg_cbytes;
313
314		if (in->msg_qnum > USHRT_MAX)
315			out.msg_qnum	= USHRT_MAX;
316		else
317			out.msg_qnum	= in->msg_qnum;
318
319		if (in->msg_qbytes > USHRT_MAX)
320			out.msg_qbytes	= USHRT_MAX;
321		else
322			out.msg_qbytes	= in->msg_qbytes;
323		out.msg_lqbytes		= in->msg_qbytes;
324
325		out.msg_lspid		= in->msg_lspid;
326		out.msg_lrpid		= in->msg_lrpid;
327
328		return copy_to_user(buf, &out, sizeof(out));
329	}
330	default:
331		return -EINVAL;
332	}
333}
334
335static inline unsigned long
336copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
337{
338	switch (version) {
339	case IPC_64:
340		if (copy_from_user(out, buf, sizeof(*out)))
341			return -EFAULT;
342		return 0;
343	case IPC_OLD:
344	{
345		struct msqid_ds tbuf_old;
346
347		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
348			return -EFAULT;
349
350		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
351		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
352		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
353
354		if (tbuf_old.msg_qbytes == 0)
355			out->msg_qbytes	= tbuf_old.msg_lqbytes;
356		else
357			out->msg_qbytes	= tbuf_old.msg_qbytes;
358
359		return 0;
360	}
361	default:
362		return -EINVAL;
363	}
364}
365
366/*
367 * This function handles some msgctl commands which require the rwsem
368 * to be held in write mode.
369 * NOTE: no locks must be held, the rwsem is taken inside this function.
370 */
371static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
372		       struct msqid_ds __user *buf, int version)
373{
374	struct kern_ipc_perm *ipcp;
375	struct msqid64_ds uninitialized_var(msqid64);
376	struct msg_queue *msq;
377	int err;
378
379	if (cmd == IPC_SET) {
380		if (copy_msqid_from_user(&msqid64, buf, version))
381			return -EFAULT;
382	}
383
384	down_write(&msg_ids(ns).rwsem);
385	rcu_read_lock();
386
387	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
388				      &msqid64.msg_perm, msqid64.msg_qbytes);
389	if (IS_ERR(ipcp)) {
390		err = PTR_ERR(ipcp);
391		goto out_unlock1;
392	}
393
394	msq = container_of(ipcp, struct msg_queue, q_perm);
395
396	err = security_msg_queue_msgctl(msq, cmd);
397	if (err)
398		goto out_unlock1;
399
400	switch (cmd) {
401	case IPC_RMID:
402		ipc_lock_object(&msq->q_perm);
403		/* freeque unlocks the ipc object and rcu */
404		freeque(ns, ipcp);
405		goto out_up;
406	case IPC_SET:
407	{
408		DEFINE_WAKE_Q(wake_q);
409
410		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
411		    !capable(CAP_SYS_RESOURCE)) {
412			err = -EPERM;
413			goto out_unlock1;
414		}
415
416		ipc_lock_object(&msq->q_perm);
417		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
418		if (err)
419			goto out_unlock0;
420
421		msq->q_qbytes = msqid64.msg_qbytes;
422
423		msq->q_ctime = get_seconds();
424		/*
425		 * Sleeping receivers might be excluded by
426		 * stricter permissions.
427		 */
428		expunge_all(msq, -EAGAIN, &wake_q);
429		/*
430		 * Sleeping senders might be able to send
431		 * due to a larger queue size.
432		 */
433		ss_wakeup(msq, &wake_q, false);
434		ipc_unlock_object(&msq->q_perm);
435		wake_up_q(&wake_q);
436
437		goto out_unlock1;
438	}
439	default:
440		err = -EINVAL;
441		goto out_unlock1;
442	}
443
444out_unlock0:
445	ipc_unlock_object(&msq->q_perm);
446out_unlock1:
447	rcu_read_unlock();
448out_up:
449	up_write(&msg_ids(ns).rwsem);
450	return err;
451}
452
453static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
454			 int cmd, int version, void __user *buf)
455{
456	int err;
457	struct msg_queue *msq;
458
459	switch (cmd) {
460	case IPC_INFO:
461	case MSG_INFO:
462	{
463		struct msginfo msginfo;
464		int max_id;
465
466		if (!buf)
467			return -EFAULT;
468
469		/*
470		 * We must not return kernel stack data.
471		 * due to padding, it's not enough
472		 * to set all member fields.
473		 */
474		err = security_msg_queue_msgctl(NULL, cmd);
475		if (err)
476			return err;
477
478		memset(&msginfo, 0, sizeof(msginfo));
479		msginfo.msgmni = ns->msg_ctlmni;
480		msginfo.msgmax = ns->msg_ctlmax;
481		msginfo.msgmnb = ns->msg_ctlmnb;
482		msginfo.msgssz = MSGSSZ;
483		msginfo.msgseg = MSGSEG;
484		down_read(&msg_ids(ns).rwsem);
485		if (cmd == MSG_INFO) {
486			msginfo.msgpool = msg_ids(ns).in_use;
487			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
488			msginfo.msgtql = atomic_read(&ns->msg_bytes);
489		} else {
490			msginfo.msgmap = MSGMAP;
491			msginfo.msgpool = MSGPOOL;
492			msginfo.msgtql = MSGTQL;
493		}
494		max_id = ipc_get_maxid(&msg_ids(ns));
495		up_read(&msg_ids(ns).rwsem);
496		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
497			return -EFAULT;
498		return (max_id < 0) ? 0 : max_id;
499	}
500
501	case MSG_STAT:
502	case IPC_STAT:
503	{
504		struct msqid64_ds tbuf;
505		int success_return;
506
507		if (!buf)
508			return -EFAULT;
509
510		memset(&tbuf, 0, sizeof(tbuf));
511
512		rcu_read_lock();
513		if (cmd == MSG_STAT) {
514			msq = msq_obtain_object(ns, msqid);
515			if (IS_ERR(msq)) {
516				err = PTR_ERR(msq);
517				goto out_unlock;
518			}
519			success_return = msq->q_perm.id;
520		} else {
521			msq = msq_obtain_object_check(ns, msqid);
522			if (IS_ERR(msq)) {
523				err = PTR_ERR(msq);
524				goto out_unlock;
525			}
526			success_return = 0;
527		}
528
529		err = -EACCES;
530		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
531			goto out_unlock;
532
533		err = security_msg_queue_msgctl(msq, cmd);
534		if (err)
535			goto out_unlock;
536
537		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
538		tbuf.msg_stime  = msq->q_stime;
539		tbuf.msg_rtime  = msq->q_rtime;
540		tbuf.msg_ctime  = msq->q_ctime;
541		tbuf.msg_cbytes = msq->q_cbytes;
542		tbuf.msg_qnum   = msq->q_qnum;
543		tbuf.msg_qbytes = msq->q_qbytes;
544		tbuf.msg_lspid  = msq->q_lspid;
545		tbuf.msg_lrpid  = msq->q_lrpid;
546		rcu_read_unlock();
547
548		if (copy_msqid_to_user(buf, &tbuf, version))
549			return -EFAULT;
550		return success_return;
551	}
552
553	default:
554		return -EINVAL;
555	}
556
557	return err;
558out_unlock:
559	rcu_read_unlock();
560	return err;
561}
562
563SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
564{
565	int version;
566	struct ipc_namespace *ns;
567
568	if (msqid < 0 || cmd < 0)
569		return -EINVAL;
570
571	version = ipc_parse_version(&cmd);
572	ns = current->nsproxy->ipc_ns;
573
574	switch (cmd) {
575	case IPC_INFO:
576	case MSG_INFO:
577	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
578	case IPC_STAT:
579		return msgctl_nolock(ns, msqid, cmd, version, buf);
580	case IPC_SET:
581	case IPC_RMID:
582		return msgctl_down(ns, msqid, cmd, buf, version);
583	default:
584		return  -EINVAL;
585	}
586}
587
588static int testmsg(struct msg_msg *msg, long type, int mode)
589{
590	switch (mode) {
591	case SEARCH_ANY:
592	case SEARCH_NUMBER:
593		return 1;
594	case SEARCH_LESSEQUAL:
595		if (msg->m_type <= type)
596			return 1;
597		break;
598	case SEARCH_EQUAL:
599		if (msg->m_type == type)
600			return 1;
601		break;
602	case SEARCH_NOTEQUAL:
603		if (msg->m_type != type)
604			return 1;
605		break;
606	}
607	return 0;
608}
609
610static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
611				 struct wake_q_head *wake_q)
612{
613	struct msg_receiver *msr, *t;
614
615	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
616		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
617		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
618					       msr->r_msgtype, msr->r_mode)) {
619
620			list_del(&msr->r_list);
621			if (msr->r_maxsize < msg->m_ts) {
622				wake_q_add(wake_q, msr->r_tsk);
623				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
624			} else {
625				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
626				msq->q_rtime = get_seconds();
627
628				wake_q_add(wake_q, msr->r_tsk);
629				WRITE_ONCE(msr->r_msg, msg);
630				return 1;
631			}
632		}
633	}
634
635	return 0;
636}
637
638long do_msgsnd(int msqid, long mtype, void __user *mtext,
639		size_t msgsz, int msgflg)
640{
641	struct msg_queue *msq;
642	struct msg_msg *msg;
643	int err;
644	struct ipc_namespace *ns;
645	DEFINE_WAKE_Q(wake_q);
646
647	ns = current->nsproxy->ipc_ns;
648
649	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
650		return -EINVAL;
651	if (mtype < 1)
652		return -EINVAL;
653
654	msg = load_msg(mtext, msgsz);
655	if (IS_ERR(msg))
656		return PTR_ERR(msg);
657
658	msg->m_type = mtype;
659	msg->m_ts = msgsz;
660
661	rcu_read_lock();
662	msq = msq_obtain_object_check(ns, msqid);
663	if (IS_ERR(msq)) {
664		err = PTR_ERR(msq);
665		goto out_unlock1;
666	}
667
668	ipc_lock_object(&msq->q_perm);
669
670	for (;;) {
671		struct msg_sender s;
672
673		err = -EACCES;
674		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
675			goto out_unlock0;
676
677		/* raced with RMID? */
678		if (!ipc_valid_object(&msq->q_perm)) {
679			err = -EIDRM;
680			goto out_unlock0;
681		}
682
683		err = security_msg_queue_msgsnd(msq, msg, msgflg);
684		if (err)
685			goto out_unlock0;
686
687		if (msg_fits_inqueue(msq, msgsz))
688			break;
689
690		/* queue full, wait: */
691		if (msgflg & IPC_NOWAIT) {
692			err = -EAGAIN;
693			goto out_unlock0;
694		}
695
696		/* enqueue the sender and prepare to block */
697		ss_add(msq, &s, msgsz);
698
699		if (!ipc_rcu_getref(&msq->q_perm)) {
700			err = -EIDRM;
701			goto out_unlock0;
702		}
703
704		ipc_unlock_object(&msq->q_perm);
705		rcu_read_unlock();
706		schedule();
707
708		rcu_read_lock();
709		ipc_lock_object(&msq->q_perm);
710
711		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
712		/* raced with RMID? */
713		if (!ipc_valid_object(&msq->q_perm)) {
714			err = -EIDRM;
715			goto out_unlock0;
716		}
717		ss_del(&s);
718
719		if (signal_pending(current)) {
720			err = -ERESTARTNOHAND;
721			goto out_unlock0;
722		}
723
724	}
725
726	msq->q_lspid = task_tgid_vnr(current);
727	msq->q_stime = get_seconds();
728
729	if (!pipelined_send(msq, msg, &wake_q)) {
730		/* no one is waiting for this message, enqueue it */
731		list_add_tail(&msg->m_list, &msq->q_messages);
732		msq->q_cbytes += msgsz;
733		msq->q_qnum++;
734		atomic_add(msgsz, &ns->msg_bytes);
735		atomic_inc(&ns->msg_hdrs);
736	}
737
738	err = 0;
739	msg = NULL;
740
741out_unlock0:
742	ipc_unlock_object(&msq->q_perm);
743	wake_up_q(&wake_q);
744out_unlock1:
745	rcu_read_unlock();
746	if (msg != NULL)
747		free_msg(msg);
748	return err;
749}
750
751SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
752		int, msgflg)
753{
754	long mtype;
755
756	if (get_user(mtype, &msgp->mtype))
757		return -EFAULT;
758	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
759}
760
761static inline int convert_mode(long *msgtyp, int msgflg)
762{
763	if (msgflg & MSG_COPY)
764		return SEARCH_NUMBER;
765	/*
766	 *  find message of correct type.
767	 *  msgtyp = 0 => get first.
768	 *  msgtyp > 0 => get first message of matching type.
769	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
770	 */
771	if (*msgtyp == 0)
772		return SEARCH_ANY;
773	if (*msgtyp < 0) {
774		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
775			*msgtyp = LONG_MAX;
776		else
777			*msgtyp = -*msgtyp;
778		return SEARCH_LESSEQUAL;
779	}
780	if (msgflg & MSG_EXCEPT)
781		return SEARCH_NOTEQUAL;
782	return SEARCH_EQUAL;
783}
784
785static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
786{
787	struct msgbuf __user *msgp = dest;
788	size_t msgsz;
789
790	if (put_user(msg->m_type, &msgp->mtype))
791		return -EFAULT;
792
793	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
794	if (store_msg(msgp->mtext, msg, msgsz))
795		return -EFAULT;
796	return msgsz;
797}
798
799#ifdef CONFIG_CHECKPOINT_RESTORE
800/*
801 * This function creates new kernel message structure, large enough to store
802 * bufsz message bytes.
803 */
804static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
805{
806	struct msg_msg *copy;
807
808	/*
809	 * Create dummy message to copy real message to.
810	 */
811	copy = load_msg(buf, bufsz);
812	if (!IS_ERR(copy))
813		copy->m_ts = bufsz;
814	return copy;
815}
816
817static inline void free_copy(struct msg_msg *copy)
818{
819	if (copy)
820		free_msg(copy);
821}
822#else
823static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
824{
825	return ERR_PTR(-ENOSYS);
826}
827
828static inline void free_copy(struct msg_msg *copy)
829{
830}
831#endif
832
833static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
834{
835	struct msg_msg *msg, *found = NULL;
836	long count = 0;
837
838	list_for_each_entry(msg, &msq->q_messages, m_list) {
839		if (testmsg(msg, *msgtyp, mode) &&
840		    !security_msg_queue_msgrcv(msq, msg, current,
841					       *msgtyp, mode)) {
842			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
843				*msgtyp = msg->m_type - 1;
844				found = msg;
845			} else if (mode == SEARCH_NUMBER) {
846				if (*msgtyp == count)
847					return msg;
848			} else
849				return msg;
850			count++;
851		}
852	}
853
854	return found ?: ERR_PTR(-EAGAIN);
855}
856
857long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
858	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
859{
860	int mode;
861	struct msg_queue *msq;
862	struct ipc_namespace *ns;
863	struct msg_msg *msg, *copy = NULL;
864	DEFINE_WAKE_Q(wake_q);
865
866	ns = current->nsproxy->ipc_ns;
867
868	if (msqid < 0 || (long) bufsz < 0)
869		return -EINVAL;
870
871	if (msgflg & MSG_COPY) {
872		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
873			return -EINVAL;
874		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
875		if (IS_ERR(copy))
876			return PTR_ERR(copy);
877	}
878	mode = convert_mode(&msgtyp, msgflg);
879
880	rcu_read_lock();
881	msq = msq_obtain_object_check(ns, msqid);
882	if (IS_ERR(msq)) {
883		rcu_read_unlock();
884		free_copy(copy);
885		return PTR_ERR(msq);
886	}
887
888	for (;;) {
889		struct msg_receiver msr_d;
890
891		msg = ERR_PTR(-EACCES);
892		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
893			goto out_unlock1;
894
895		ipc_lock_object(&msq->q_perm);
896
897		/* raced with RMID? */
898		if (!ipc_valid_object(&msq->q_perm)) {
899			msg = ERR_PTR(-EIDRM);
900			goto out_unlock0;
901		}
902
903		msg = find_msg(msq, &msgtyp, mode);
904		if (!IS_ERR(msg)) {
905			/*
906			 * Found a suitable message.
907			 * Unlink it from the queue.
908			 */
909			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
910				msg = ERR_PTR(-E2BIG);
911				goto out_unlock0;
912			}
913			/*
914			 * If we are copying, then do not unlink message and do
915			 * not update queue parameters.
916			 */
917			if (msgflg & MSG_COPY) {
918				msg = copy_msg(msg, copy);
919				goto out_unlock0;
920			}
921
922			list_del(&msg->m_list);
923			msq->q_qnum--;
924			msq->q_rtime = get_seconds();
925			msq->q_lrpid = task_tgid_vnr(current);
926			msq->q_cbytes -= msg->m_ts;
927			atomic_sub(msg->m_ts, &ns->msg_bytes);
928			atomic_dec(&ns->msg_hdrs);
929			ss_wakeup(msq, &wake_q, false);
930
931			goto out_unlock0;
932		}
933
934		/* No message waiting. Wait for a message */
935		if (msgflg & IPC_NOWAIT) {
936			msg = ERR_PTR(-ENOMSG);
937			goto out_unlock0;
938		}
939
940		list_add_tail(&msr_d.r_list, &msq->q_receivers);
941		msr_d.r_tsk = current;
942		msr_d.r_msgtype = msgtyp;
943		msr_d.r_mode = mode;
944		if (msgflg & MSG_NOERROR)
945			msr_d.r_maxsize = INT_MAX;
946		else
947			msr_d.r_maxsize = bufsz;
948		msr_d.r_msg = ERR_PTR(-EAGAIN);
949		__set_current_state(TASK_INTERRUPTIBLE);
950
951		ipc_unlock_object(&msq->q_perm);
952		rcu_read_unlock();
953		schedule();
954
955		/*
956		 * Lockless receive, part 1:
957		 * We don't hold a reference to the queue and getting a
958		 * reference would defeat the idea of a lockless operation,
959		 * thus the code relies on rcu to guarantee the existence of
960		 * msq:
961		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
962		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
963		 */
964		rcu_read_lock();
965
966		/*
967		 * Lockless receive, part 2:
968		 * The work in pipelined_send() and expunge_all():
969		 * - Set pointer to message
970		 * - Queue the receiver task for later wakeup
971		 * - Wake up the process after the lock is dropped.
972		 *
973		 * Should the process wake up before this wakeup (due to a
974		 * signal) it will either see the message and continue ...
975		 */
976		msg = READ_ONCE(msr_d.r_msg);
977		if (msg != ERR_PTR(-EAGAIN))
978			goto out_unlock1;
979
980		 /*
981		  * ... or see -EAGAIN, acquire the lock to check the message
982		  * again.
983		  */
984		ipc_lock_object(&msq->q_perm);
985
986		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);
1001	wake_up_q(&wake_q);
1002out_unlock1:
1003	rcu_read_unlock();
1004	if (IS_ERR(msg)) {
1005		free_copy(copy);
1006		return PTR_ERR(msg);
1007	}
1008
1009	bufsz = msg_handler(buf, msg, bufsz);
1010	free_msg(msg);
1011
1012	return bufsz;
1013}
1014
1015SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1016		long, msgtyp, int, msgflg)
1017{
1018	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1019}
1020
1021
1022void msg_init_ns(struct ipc_namespace *ns)
1023{
1024	ns->msg_ctlmax = MSGMAX;
1025	ns->msg_ctlmnb = MSGMNB;
1026	ns->msg_ctlmni = MSGMNI;
1027
1028	atomic_set(&ns->msg_bytes, 0);
1029	atomic_set(&ns->msg_hdrs, 0);
1030	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1031}
1032
1033#ifdef CONFIG_IPC_NS
1034void msg_exit_ns(struct ipc_namespace *ns)
1035{
1036	free_ipcs(ns, &msg_ids(ns), freeque);
1037	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1038}
1039#endif
1040
1041#ifdef CONFIG_PROC_FS
1042static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1043{
1044	struct user_namespace *user_ns = seq_user_ns(s);
1045	struct msg_queue *msq = it;
1046
1047	seq_printf(s,
1048		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1049		   msq->q_perm.key,
1050		   msq->q_perm.id,
1051		   msq->q_perm.mode,
1052		   msq->q_cbytes,
1053		   msq->q_qnum,
1054		   msq->q_lspid,
1055		   msq->q_lrpid,
1056		   from_kuid_munged(user_ns, msq->q_perm.uid),
1057		   from_kgid_munged(user_ns, msq->q_perm.gid),
1058		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1059		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1060		   msq->q_stime,
1061		   msq->q_rtime,
1062		   msq->q_ctime);
1063
1064	return 0;
1065}
1066#endif
1067
1068void __init msg_init(void)
1069{
1070	msg_init_ns(&init_ipc_ns);
1071
1072	ipc_init_proc_interface("sysvipc/msg",
1073				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1074				IPC_MSG_IDS, sysvipc_msg_proc_show);
1075}
1076