msg.c revision 0eb71a9d
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/msg.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 *
6 * Removed all the remaining kerneld mess
7 * Catch the -EFAULT stuff properly
8 * Use GFP_KERNEL for messages as in 1.2
9 * Fixed up the unchecked user space derefs
10 * Copyright (C) 1998 Alan Cox & Andi Kleen
11 *
12 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
13 *
14 * mostly rewritten, threaded and wake-one semantics added
15 * MSGMAX limit removed, sysctl's added
16 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
17 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
20 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
24 */
25
26#include <linux/capability.h>
27#include <linux/msg.h>
28#include <linux/spinlock.h>
29#include <linux/init.h>
30#include <linux/mm.h>
31#include <linux/proc_fs.h>
32#include <linux/list.h>
33#include <linux/security.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#include <linux/rhashtable.h>
42
43#include <asm/current.h>
44#include <linux/uaccess.h>
45#include "util.h"
46
47/* one msq_queue structure for each present queue on the system */
48struct msg_queue {
49	struct kern_ipc_perm q_perm;
50	time64_t q_stime;		/* last msgsnd time */
51	time64_t q_rtime;		/* last msgrcv time */
52	time64_t q_ctime;		/* last change time */
53	unsigned long q_cbytes;		/* current number of bytes on queue */
54	unsigned long q_qnum;		/* number of messages in queue */
55	unsigned long q_qbytes;		/* max number of bytes on queue */
56	struct pid *q_lspid;		/* pid of last msgsnd */
57	struct pid *q_lrpid;		/* last receive pid */
58
59	struct list_head q_messages;
60	struct list_head q_receivers;
61	struct list_head q_senders;
62} __randomize_layout;
63
64/* one msg_receiver structure for each sleeping receiver */
65struct msg_receiver {
66	struct list_head	r_list;
67	struct task_struct	*r_tsk;
68
69	int			r_mode;
70	long			r_msgtype;
71	long			r_maxsize;
72
73	struct msg_msg		*r_msg;
74};
75
76/* one msg_sender for each sleeping sender */
77struct msg_sender {
78	struct list_head	list;
79	struct task_struct	*tsk;
80	size_t                  msgsz;
81};
82
83#define SEARCH_ANY		1
84#define SEARCH_EQUAL		2
85#define SEARCH_NOTEQUAL		3
86#define SEARCH_LESSEQUAL	4
87#define SEARCH_NUMBER		5
88
89#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
90
91static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
92{
93	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
94
95	if (IS_ERR(ipcp))
96		return ERR_CAST(ipcp);
97
98	return container_of(ipcp, struct msg_queue, q_perm);
99}
100
101static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
102							int id)
103{
104	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
105
106	if (IS_ERR(ipcp))
107		return ERR_CAST(ipcp);
108
109	return container_of(ipcp, struct msg_queue, q_perm);
110}
111
112static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
113{
114	ipc_rmid(&msg_ids(ns), &s->q_perm);
115}
116
117static void msg_rcu_free(struct rcu_head *head)
118{
119	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
120	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
121
122	security_msg_queue_free(&msq->q_perm);
123	kvfree(msq);
124}
125
126/**
127 * newque - Create a new msg queue
128 * @ns: namespace
129 * @params: ptr to the structure that contains the key and msgflg
130 *
131 * Called with msg_ids.rwsem held (writer)
132 */
133static int newque(struct ipc_namespace *ns, struct ipc_params *params)
134{
135	struct msg_queue *msq;
136	int retval;
137	key_t key = params->key;
138	int msgflg = params->flg;
139
140	msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
141	if (unlikely(!msq))
142		return -ENOMEM;
143
144	msq->q_perm.mode = msgflg & S_IRWXUGO;
145	msq->q_perm.key = key;
146
147	msq->q_perm.security = NULL;
148	retval = security_msg_queue_alloc(&msq->q_perm);
149	if (retval) {
150		kvfree(msq);
151		return retval;
152	}
153
154	msq->q_stime = msq->q_rtime = 0;
155	msq->q_ctime = ktime_get_real_seconds();
156	msq->q_cbytes = msq->q_qnum = 0;
157	msq->q_qbytes = ns->msg_ctlmnb;
158	msq->q_lspid = msq->q_lrpid = NULL;
159	INIT_LIST_HEAD(&msq->q_messages);
160	INIT_LIST_HEAD(&msq->q_receivers);
161	INIT_LIST_HEAD(&msq->q_senders);
162
163	/* ipc_addid() locks msq upon success. */
164	retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
165	if (retval < 0) {
166		call_rcu(&msq->q_perm.rcu, msg_rcu_free);
167		return retval;
168	}
169
170	ipc_unlock_object(&msq->q_perm);
171	rcu_read_unlock();
172
173	return msq->q_perm.id;
174}
175
176static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
177{
178	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
179		1 + msq->q_qnum <= msq->q_qbytes;
180}
181
182static inline void ss_add(struct msg_queue *msq,
183			  struct msg_sender *mss, size_t msgsz)
184{
185	mss->tsk = current;
186	mss->msgsz = msgsz;
187	__set_current_state(TASK_INTERRUPTIBLE);
188	list_add_tail(&mss->list, &msq->q_senders);
189}
190
191static inline void ss_del(struct msg_sender *mss)
192{
193	if (mss->list.next)
194		list_del(&mss->list);
195}
196
197static void ss_wakeup(struct msg_queue *msq,
198		      struct wake_q_head *wake_q, bool kill)
199{
200	struct msg_sender *mss, *t;
201	struct task_struct *stop_tsk = NULL;
202	struct list_head *h = &msq->q_senders;
203
204	list_for_each_entry_safe(mss, t, h, list) {
205		if (kill)
206			mss->list.next = NULL;
207
208		/*
209		 * Stop at the first task we don't wakeup,
210		 * we've already iterated the original
211		 * sender queue.
212		 */
213		else if (stop_tsk == mss->tsk)
214			break;
215		/*
216		 * We are not in an EIDRM scenario here, therefore
217		 * verify that we really need to wakeup the task.
218		 * To maintain current semantics and wakeup order,
219		 * move the sender to the tail on behalf of the
220		 * blocked task.
221		 */
222		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
223			if (!stop_tsk)
224				stop_tsk = mss->tsk;
225
226			list_move_tail(&mss->list, &msq->q_senders);
227			continue;
228		}
229
230		wake_q_add(wake_q, mss->tsk);
231	}
232}
233
234static void expunge_all(struct msg_queue *msq, int res,
235			struct wake_q_head *wake_q)
236{
237	struct msg_receiver *msr, *t;
238
239	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
240		wake_q_add(wake_q, msr->r_tsk);
241		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
242	}
243}
244
245/*
246 * freeque() wakes up waiters on the sender and receiver waiting queue,
247 * removes the message queue from message queue ID IDR, and cleans up all the
248 * messages associated with this queue.
249 *
250 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
251 * before freeque() is called. msg_ids.rwsem remains locked on exit.
252 */
253static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
254{
255	struct msg_msg *msg, *t;
256	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
257	DEFINE_WAKE_Q(wake_q);
258
259	expunge_all(msq, -EIDRM, &wake_q);
260	ss_wakeup(msq, &wake_q, true);
261	msg_rmid(ns, msq);
262	ipc_unlock_object(&msq->q_perm);
263	wake_up_q(&wake_q);
264	rcu_read_unlock();
265
266	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
267		atomic_dec(&ns->msg_hdrs);
268		free_msg(msg);
269	}
270	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
271	ipc_update_pid(&msq->q_lspid, NULL);
272	ipc_update_pid(&msq->q_lrpid, NULL);
273	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
274}
275
276long ksys_msgget(key_t key, int msgflg)
277{
278	struct ipc_namespace *ns;
279	static const struct ipc_ops msg_ops = {
280		.getnew = newque,
281		.associate = security_msg_queue_associate,
282	};
283	struct ipc_params msg_params;
284
285	ns = current->nsproxy->ipc_ns;
286
287	msg_params.key = key;
288	msg_params.flg = msgflg;
289
290	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
291}
292
293SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
294{
295	return ksys_msgget(key, msgflg);
296}
297
298static inline unsigned long
299copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
300{
301	switch (version) {
302	case IPC_64:
303		return copy_to_user(buf, in, sizeof(*in));
304	case IPC_OLD:
305	{
306		struct msqid_ds out;
307
308		memset(&out, 0, sizeof(out));
309
310		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
311
312		out.msg_stime		= in->msg_stime;
313		out.msg_rtime		= in->msg_rtime;
314		out.msg_ctime		= in->msg_ctime;
315
316		if (in->msg_cbytes > USHRT_MAX)
317			out.msg_cbytes	= USHRT_MAX;
318		else
319			out.msg_cbytes	= in->msg_cbytes;
320		out.msg_lcbytes		= in->msg_cbytes;
321
322		if (in->msg_qnum > USHRT_MAX)
323			out.msg_qnum	= USHRT_MAX;
324		else
325			out.msg_qnum	= in->msg_qnum;
326
327		if (in->msg_qbytes > USHRT_MAX)
328			out.msg_qbytes	= USHRT_MAX;
329		else
330			out.msg_qbytes	= in->msg_qbytes;
331		out.msg_lqbytes		= in->msg_qbytes;
332
333		out.msg_lspid		= in->msg_lspid;
334		out.msg_lrpid		= in->msg_lrpid;
335
336		return copy_to_user(buf, &out, sizeof(out));
337	}
338	default:
339		return -EINVAL;
340	}
341}
342
343static inline unsigned long
344copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
345{
346	switch (version) {
347	case IPC_64:
348		if (copy_from_user(out, buf, sizeof(*out)))
349			return -EFAULT;
350		return 0;
351	case IPC_OLD:
352	{
353		struct msqid_ds tbuf_old;
354
355		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
356			return -EFAULT;
357
358		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
359		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
360		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
361
362		if (tbuf_old.msg_qbytes == 0)
363			out->msg_qbytes	= tbuf_old.msg_lqbytes;
364		else
365			out->msg_qbytes	= tbuf_old.msg_qbytes;
366
367		return 0;
368	}
369	default:
370		return -EINVAL;
371	}
372}
373
374/*
375 * This function handles some msgctl commands which require the rwsem
376 * to be held in write mode.
377 * NOTE: no locks must be held, the rwsem is taken inside this function.
378 */
379static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
380			struct msqid64_ds *msqid64)
381{
382	struct kern_ipc_perm *ipcp;
383	struct msg_queue *msq;
384	int err;
385
386	down_write(&msg_ids(ns).rwsem);
387	rcu_read_lock();
388
389	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
390				      &msqid64->msg_perm, msqid64->msg_qbytes);
391	if (IS_ERR(ipcp)) {
392		err = PTR_ERR(ipcp);
393		goto out_unlock1;
394	}
395
396	msq = container_of(ipcp, struct msg_queue, q_perm);
397
398	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
399	if (err)
400		goto out_unlock1;
401
402	switch (cmd) {
403	case IPC_RMID:
404		ipc_lock_object(&msq->q_perm);
405		/* freeque unlocks the ipc object and rcu */
406		freeque(ns, ipcp);
407		goto out_up;
408	case IPC_SET:
409	{
410		DEFINE_WAKE_Q(wake_q);
411
412		if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
413		    !capable(CAP_SYS_RESOURCE)) {
414			err = -EPERM;
415			goto out_unlock1;
416		}
417
418		ipc_lock_object(&msq->q_perm);
419		err = ipc_update_perm(&msqid64->msg_perm, ipcp);
420		if (err)
421			goto out_unlock0;
422
423		msq->q_qbytes = msqid64->msg_qbytes;
424
425		msq->q_ctime = ktime_get_real_seconds();
426		/*
427		 * Sleeping receivers might be excluded by
428		 * stricter permissions.
429		 */
430		expunge_all(msq, -EAGAIN, &wake_q);
431		/*
432		 * Sleeping senders might be able to send
433		 * due to a larger queue size.
434		 */
435		ss_wakeup(msq, &wake_q, false);
436		ipc_unlock_object(&msq->q_perm);
437		wake_up_q(&wake_q);
438
439		goto out_unlock1;
440	}
441	default:
442		err = -EINVAL;
443		goto out_unlock1;
444	}
445
446out_unlock0:
447	ipc_unlock_object(&msq->q_perm);
448out_unlock1:
449	rcu_read_unlock();
450out_up:
451	up_write(&msg_ids(ns).rwsem);
452	return err;
453}
454
455static int msgctl_info(struct ipc_namespace *ns, int msqid,
456			 int cmd, struct msginfo *msginfo)
457{
458	int err;
459	int max_id;
460
461	/*
462	 * We must not return kernel stack data.
463	 * due to padding, it's not enough
464	 * to set all member fields.
465	 */
466	err = security_msg_queue_msgctl(NULL, cmd);
467	if (err)
468		return err;
469
470	memset(msginfo, 0, sizeof(*msginfo));
471	msginfo->msgmni = ns->msg_ctlmni;
472	msginfo->msgmax = ns->msg_ctlmax;
473	msginfo->msgmnb = ns->msg_ctlmnb;
474	msginfo->msgssz = MSGSSZ;
475	msginfo->msgseg = MSGSEG;
476	down_read(&msg_ids(ns).rwsem);
477	if (cmd == MSG_INFO) {
478		msginfo->msgpool = msg_ids(ns).in_use;
479		msginfo->msgmap = atomic_read(&ns->msg_hdrs);
480		msginfo->msgtql = atomic_read(&ns->msg_bytes);
481	} else {
482		msginfo->msgmap = MSGMAP;
483		msginfo->msgpool = MSGPOOL;
484		msginfo->msgtql = MSGTQL;
485	}
486	max_id = ipc_get_maxid(&msg_ids(ns));
487	up_read(&msg_ids(ns).rwsem);
488	return (max_id < 0) ? 0 : max_id;
489}
490
491static int msgctl_stat(struct ipc_namespace *ns, int msqid,
492			 int cmd, struct msqid64_ds *p)
493{
494	struct msg_queue *msq;
495	int id = 0;
496	int err;
497
498	memset(p, 0, sizeof(*p));
499
500	rcu_read_lock();
501	if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
502		msq = msq_obtain_object(ns, msqid);
503		if (IS_ERR(msq)) {
504			err = PTR_ERR(msq);
505			goto out_unlock;
506		}
507		id = msq->q_perm.id;
508	} else { /* IPC_STAT */
509		msq = msq_obtain_object_check(ns, msqid);
510		if (IS_ERR(msq)) {
511			err = PTR_ERR(msq);
512			goto out_unlock;
513		}
514	}
515
516	/* see comment for SHM_STAT_ANY */
517	if (cmd == MSG_STAT_ANY)
518		audit_ipc_obj(&msq->q_perm);
519	else {
520		err = -EACCES;
521		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
522			goto out_unlock;
523	}
524
525	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
526	if (err)
527		goto out_unlock;
528
529	ipc_lock_object(&msq->q_perm);
530
531	if (!ipc_valid_object(&msq->q_perm)) {
532		ipc_unlock_object(&msq->q_perm);
533		err = -EIDRM;
534		goto out_unlock;
535	}
536
537	kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
538	p->msg_stime  = msq->q_stime;
539	p->msg_rtime  = msq->q_rtime;
540	p->msg_ctime  = msq->q_ctime;
541#ifndef CONFIG_64BIT
542	p->msg_stime_high = msq->q_stime >> 32;
543	p->msg_rtime_high = msq->q_rtime >> 32;
544	p->msg_ctime_high = msq->q_ctime >> 32;
545#endif
546	p->msg_cbytes = msq->q_cbytes;
547	p->msg_qnum   = msq->q_qnum;
548	p->msg_qbytes = msq->q_qbytes;
549	p->msg_lspid  = pid_vnr(msq->q_lspid);
550	p->msg_lrpid  = pid_vnr(msq->q_lrpid);
551
552	ipc_unlock_object(&msq->q_perm);
553	rcu_read_unlock();
554	return id;
555
556out_unlock:
557	rcu_read_unlock();
558	return err;
559}
560
561long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
562{
563	int version;
564	struct ipc_namespace *ns;
565	struct msqid64_ds msqid64;
566	int err;
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		struct msginfo msginfo;
578		err = msgctl_info(ns, msqid, cmd, &msginfo);
579		if (err < 0)
580			return err;
581		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
582			err = -EFAULT;
583		return err;
584	}
585	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
586	case MSG_STAT_ANY:
587	case IPC_STAT:
588		err = msgctl_stat(ns, msqid, cmd, &msqid64);
589		if (err < 0)
590			return err;
591		if (copy_msqid_to_user(buf, &msqid64, version))
592			err = -EFAULT;
593		return err;
594	case IPC_SET:
595		if (copy_msqid_from_user(&msqid64, buf, version))
596			return -EFAULT;
597		/* fallthru */
598	case IPC_RMID:
599		return msgctl_down(ns, msqid, cmd, &msqid64);
600	default:
601		return  -EINVAL;
602	}
603}
604
605SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
606{
607	return ksys_msgctl(msqid, cmd, buf);
608}
609
610#ifdef CONFIG_COMPAT
611
612struct compat_msqid_ds {
613	struct compat_ipc_perm msg_perm;
614	compat_uptr_t msg_first;
615	compat_uptr_t msg_last;
616	compat_time_t msg_stime;
617	compat_time_t msg_rtime;
618	compat_time_t msg_ctime;
619	compat_ulong_t msg_lcbytes;
620	compat_ulong_t msg_lqbytes;
621	unsigned short msg_cbytes;
622	unsigned short msg_qnum;
623	unsigned short msg_qbytes;
624	compat_ipc_pid_t msg_lspid;
625	compat_ipc_pid_t msg_lrpid;
626};
627
628static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
629					int version)
630{
631	memset(out, 0, sizeof(*out));
632	if (version == IPC_64) {
633		struct compat_msqid64_ds __user *p = buf;
634		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
635			return -EFAULT;
636		if (get_user(out->msg_qbytes, &p->msg_qbytes))
637			return -EFAULT;
638	} else {
639		struct compat_msqid_ds __user *p = buf;
640		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
641			return -EFAULT;
642		if (get_user(out->msg_qbytes, &p->msg_qbytes))
643			return -EFAULT;
644	}
645	return 0;
646}
647
648static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
649					int version)
650{
651	if (version == IPC_64) {
652		struct compat_msqid64_ds v;
653		memset(&v, 0, sizeof(v));
654		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
655		v.msg_stime	 = lower_32_bits(in->msg_stime);
656		v.msg_stime_high = upper_32_bits(in->msg_stime);
657		v.msg_rtime	 = lower_32_bits(in->msg_rtime);
658		v.msg_rtime_high = upper_32_bits(in->msg_rtime);
659		v.msg_ctime	 = lower_32_bits(in->msg_ctime);
660		v.msg_ctime_high = upper_32_bits(in->msg_ctime);
661		v.msg_cbytes = in->msg_cbytes;
662		v.msg_qnum = in->msg_qnum;
663		v.msg_qbytes = in->msg_qbytes;
664		v.msg_lspid = in->msg_lspid;
665		v.msg_lrpid = in->msg_lrpid;
666		return copy_to_user(buf, &v, sizeof(v));
667	} else {
668		struct compat_msqid_ds v;
669		memset(&v, 0, sizeof(v));
670		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
671		v.msg_stime = in->msg_stime;
672		v.msg_rtime = in->msg_rtime;
673		v.msg_ctime = in->msg_ctime;
674		v.msg_cbytes = in->msg_cbytes;
675		v.msg_qnum = in->msg_qnum;
676		v.msg_qbytes = in->msg_qbytes;
677		v.msg_lspid = in->msg_lspid;
678		v.msg_lrpid = in->msg_lrpid;
679		return copy_to_user(buf, &v, sizeof(v));
680	}
681}
682
683long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
684{
685	struct ipc_namespace *ns;
686	int err;
687	struct msqid64_ds msqid64;
688	int version = compat_ipc_parse_version(&cmd);
689
690	ns = current->nsproxy->ipc_ns;
691
692	if (msqid < 0 || cmd < 0)
693		return -EINVAL;
694
695	switch (cmd & (~IPC_64)) {
696	case IPC_INFO:
697	case MSG_INFO: {
698		struct msginfo msginfo;
699		err = msgctl_info(ns, msqid, cmd, &msginfo);
700		if (err < 0)
701			return err;
702		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
703			err = -EFAULT;
704		return err;
705	}
706	case IPC_STAT:
707	case MSG_STAT:
708	case MSG_STAT_ANY:
709		err = msgctl_stat(ns, msqid, cmd, &msqid64);
710		if (err < 0)
711			return err;
712		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
713			err = -EFAULT;
714		return err;
715	case IPC_SET:
716		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
717			return -EFAULT;
718		/* fallthru */
719	case IPC_RMID:
720		return msgctl_down(ns, msqid, cmd, &msqid64);
721	default:
722		return -EINVAL;
723	}
724}
725
726COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
727{
728	return compat_ksys_msgctl(msqid, cmd, uptr);
729}
730#endif
731
732static int testmsg(struct msg_msg *msg, long type, int mode)
733{
734	switch (mode) {
735	case SEARCH_ANY:
736	case SEARCH_NUMBER:
737		return 1;
738	case SEARCH_LESSEQUAL:
739		if (msg->m_type <= type)
740			return 1;
741		break;
742	case SEARCH_EQUAL:
743		if (msg->m_type == type)
744			return 1;
745		break;
746	case SEARCH_NOTEQUAL:
747		if (msg->m_type != type)
748			return 1;
749		break;
750	}
751	return 0;
752}
753
754static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
755				 struct wake_q_head *wake_q)
756{
757	struct msg_receiver *msr, *t;
758
759	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
760		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
761		    !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
762					       msr->r_msgtype, msr->r_mode)) {
763
764			list_del(&msr->r_list);
765			if (msr->r_maxsize < msg->m_ts) {
766				wake_q_add(wake_q, msr->r_tsk);
767				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
768			} else {
769				ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
770				msq->q_rtime = ktime_get_real_seconds();
771
772				wake_q_add(wake_q, msr->r_tsk);
773				WRITE_ONCE(msr->r_msg, msg);
774				return 1;
775			}
776		}
777	}
778
779	return 0;
780}
781
782static long do_msgsnd(int msqid, long mtype, void __user *mtext,
783		size_t msgsz, int msgflg)
784{
785	struct msg_queue *msq;
786	struct msg_msg *msg;
787	int err;
788	struct ipc_namespace *ns;
789	DEFINE_WAKE_Q(wake_q);
790
791	ns = current->nsproxy->ipc_ns;
792
793	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
794		return -EINVAL;
795	if (mtype < 1)
796		return -EINVAL;
797
798	msg = load_msg(mtext, msgsz);
799	if (IS_ERR(msg))
800		return PTR_ERR(msg);
801
802	msg->m_type = mtype;
803	msg->m_ts = msgsz;
804
805	rcu_read_lock();
806	msq = msq_obtain_object_check(ns, msqid);
807	if (IS_ERR(msq)) {
808		err = PTR_ERR(msq);
809		goto out_unlock1;
810	}
811
812	ipc_lock_object(&msq->q_perm);
813
814	for (;;) {
815		struct msg_sender s;
816
817		err = -EACCES;
818		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
819			goto out_unlock0;
820
821		/* raced with RMID? */
822		if (!ipc_valid_object(&msq->q_perm)) {
823			err = -EIDRM;
824			goto out_unlock0;
825		}
826
827		err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
828		if (err)
829			goto out_unlock0;
830
831		if (msg_fits_inqueue(msq, msgsz))
832			break;
833
834		/* queue full, wait: */
835		if (msgflg & IPC_NOWAIT) {
836			err = -EAGAIN;
837			goto out_unlock0;
838		}
839
840		/* enqueue the sender and prepare to block */
841		ss_add(msq, &s, msgsz);
842
843		if (!ipc_rcu_getref(&msq->q_perm)) {
844			err = -EIDRM;
845			goto out_unlock0;
846		}
847
848		ipc_unlock_object(&msq->q_perm);
849		rcu_read_unlock();
850		schedule();
851
852		rcu_read_lock();
853		ipc_lock_object(&msq->q_perm);
854
855		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
856		/* raced with RMID? */
857		if (!ipc_valid_object(&msq->q_perm)) {
858			err = -EIDRM;
859			goto out_unlock0;
860		}
861		ss_del(&s);
862
863		if (signal_pending(current)) {
864			err = -ERESTARTNOHAND;
865			goto out_unlock0;
866		}
867
868	}
869
870	ipc_update_pid(&msq->q_lspid, task_tgid(current));
871	msq->q_stime = ktime_get_real_seconds();
872
873	if (!pipelined_send(msq, msg, &wake_q)) {
874		/* no one is waiting for this message, enqueue it */
875		list_add_tail(&msg->m_list, &msq->q_messages);
876		msq->q_cbytes += msgsz;
877		msq->q_qnum++;
878		atomic_add(msgsz, &ns->msg_bytes);
879		atomic_inc(&ns->msg_hdrs);
880	}
881
882	err = 0;
883	msg = NULL;
884
885out_unlock0:
886	ipc_unlock_object(&msq->q_perm);
887	wake_up_q(&wake_q);
888out_unlock1:
889	rcu_read_unlock();
890	if (msg != NULL)
891		free_msg(msg);
892	return err;
893}
894
895long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
896		 int msgflg)
897{
898	long mtype;
899
900	if (get_user(mtype, &msgp->mtype))
901		return -EFAULT;
902	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
903}
904
905SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
906		int, msgflg)
907{
908	return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
909}
910
911#ifdef CONFIG_COMPAT
912
913struct compat_msgbuf {
914	compat_long_t mtype;
915	char mtext[1];
916};
917
918long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
919		       compat_ssize_t msgsz, int msgflg)
920{
921	struct compat_msgbuf __user *up = compat_ptr(msgp);
922	compat_long_t mtype;
923
924	if (get_user(mtype, &up->mtype))
925		return -EFAULT;
926	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
927}
928
929COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
930		       compat_ssize_t, msgsz, int, msgflg)
931{
932	return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
933}
934#endif
935
936static inline int convert_mode(long *msgtyp, int msgflg)
937{
938	if (msgflg & MSG_COPY)
939		return SEARCH_NUMBER;
940	/*
941	 *  find message of correct type.
942	 *  msgtyp = 0 => get first.
943	 *  msgtyp > 0 => get first message of matching type.
944	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
945	 */
946	if (*msgtyp == 0)
947		return SEARCH_ANY;
948	if (*msgtyp < 0) {
949		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
950			*msgtyp = LONG_MAX;
951		else
952			*msgtyp = -*msgtyp;
953		return SEARCH_LESSEQUAL;
954	}
955	if (msgflg & MSG_EXCEPT)
956		return SEARCH_NOTEQUAL;
957	return SEARCH_EQUAL;
958}
959
960static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
961{
962	struct msgbuf __user *msgp = dest;
963	size_t msgsz;
964
965	if (put_user(msg->m_type, &msgp->mtype))
966		return -EFAULT;
967
968	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
969	if (store_msg(msgp->mtext, msg, msgsz))
970		return -EFAULT;
971	return msgsz;
972}
973
974#ifdef CONFIG_CHECKPOINT_RESTORE
975/*
976 * This function creates new kernel message structure, large enough to store
977 * bufsz message bytes.
978 */
979static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
980{
981	struct msg_msg *copy;
982
983	/*
984	 * Create dummy message to copy real message to.
985	 */
986	copy = load_msg(buf, bufsz);
987	if (!IS_ERR(copy))
988		copy->m_ts = bufsz;
989	return copy;
990}
991
992static inline void free_copy(struct msg_msg *copy)
993{
994	if (copy)
995		free_msg(copy);
996}
997#else
998static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
999{
1000	return ERR_PTR(-ENOSYS);
1001}
1002
1003static inline void free_copy(struct msg_msg *copy)
1004{
1005}
1006#endif
1007
1008static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1009{
1010	struct msg_msg *msg, *found = NULL;
1011	long count = 0;
1012
1013	list_for_each_entry(msg, &msq->q_messages, m_list) {
1014		if (testmsg(msg, *msgtyp, mode) &&
1015		    !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1016					       *msgtyp, mode)) {
1017			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1018				*msgtyp = msg->m_type - 1;
1019				found = msg;
1020			} else if (mode == SEARCH_NUMBER) {
1021				if (*msgtyp == count)
1022					return msg;
1023			} else
1024				return msg;
1025			count++;
1026		}
1027	}
1028
1029	return found ?: ERR_PTR(-EAGAIN);
1030}
1031
1032static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1033	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1034{
1035	int mode;
1036	struct msg_queue *msq;
1037	struct ipc_namespace *ns;
1038	struct msg_msg *msg, *copy = NULL;
1039	DEFINE_WAKE_Q(wake_q);
1040
1041	ns = current->nsproxy->ipc_ns;
1042
1043	if (msqid < 0 || (long) bufsz < 0)
1044		return -EINVAL;
1045
1046	if (msgflg & MSG_COPY) {
1047		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1048			return -EINVAL;
1049		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1050		if (IS_ERR(copy))
1051			return PTR_ERR(copy);
1052	}
1053	mode = convert_mode(&msgtyp, msgflg);
1054
1055	rcu_read_lock();
1056	msq = msq_obtain_object_check(ns, msqid);
1057	if (IS_ERR(msq)) {
1058		rcu_read_unlock();
1059		free_copy(copy);
1060		return PTR_ERR(msq);
1061	}
1062
1063	for (;;) {
1064		struct msg_receiver msr_d;
1065
1066		msg = ERR_PTR(-EACCES);
1067		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1068			goto out_unlock1;
1069
1070		ipc_lock_object(&msq->q_perm);
1071
1072		/* raced with RMID? */
1073		if (!ipc_valid_object(&msq->q_perm)) {
1074			msg = ERR_PTR(-EIDRM);
1075			goto out_unlock0;
1076		}
1077
1078		msg = find_msg(msq, &msgtyp, mode);
1079		if (!IS_ERR(msg)) {
1080			/*
1081			 * Found a suitable message.
1082			 * Unlink it from the queue.
1083			 */
1084			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1085				msg = ERR_PTR(-E2BIG);
1086				goto out_unlock0;
1087			}
1088			/*
1089			 * If we are copying, then do not unlink message and do
1090			 * not update queue parameters.
1091			 */
1092			if (msgflg & MSG_COPY) {
1093				msg = copy_msg(msg, copy);
1094				goto out_unlock0;
1095			}
1096
1097			list_del(&msg->m_list);
1098			msq->q_qnum--;
1099			msq->q_rtime = ktime_get_real_seconds();
1100			ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1101			msq->q_cbytes -= msg->m_ts;
1102			atomic_sub(msg->m_ts, &ns->msg_bytes);
1103			atomic_dec(&ns->msg_hdrs);
1104			ss_wakeup(msq, &wake_q, false);
1105
1106			goto out_unlock0;
1107		}
1108
1109		/* No message waiting. Wait for a message */
1110		if (msgflg & IPC_NOWAIT) {
1111			msg = ERR_PTR(-ENOMSG);
1112			goto out_unlock0;
1113		}
1114
1115		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1116		msr_d.r_tsk = current;
1117		msr_d.r_msgtype = msgtyp;
1118		msr_d.r_mode = mode;
1119		if (msgflg & MSG_NOERROR)
1120			msr_d.r_maxsize = INT_MAX;
1121		else
1122			msr_d.r_maxsize = bufsz;
1123		msr_d.r_msg = ERR_PTR(-EAGAIN);
1124		__set_current_state(TASK_INTERRUPTIBLE);
1125
1126		ipc_unlock_object(&msq->q_perm);
1127		rcu_read_unlock();
1128		schedule();
1129
1130		/*
1131		 * Lockless receive, part 1:
1132		 * We don't hold a reference to the queue and getting a
1133		 * reference would defeat the idea of a lockless operation,
1134		 * thus the code relies on rcu to guarantee the existence of
1135		 * msq:
1136		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1137		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1138		 */
1139		rcu_read_lock();
1140
1141		/*
1142		 * Lockless receive, part 2:
1143		 * The work in pipelined_send() and expunge_all():
1144		 * - Set pointer to message
1145		 * - Queue the receiver task for later wakeup
1146		 * - Wake up the process after the lock is dropped.
1147		 *
1148		 * Should the process wake up before this wakeup (due to a
1149		 * signal) it will either see the message and continue ...
1150		 */
1151		msg = READ_ONCE(msr_d.r_msg);
1152		if (msg != ERR_PTR(-EAGAIN))
1153			goto out_unlock1;
1154
1155		 /*
1156		  * ... or see -EAGAIN, acquire the lock to check the message
1157		  * again.
1158		  */
1159		ipc_lock_object(&msq->q_perm);
1160
1161		msg = msr_d.r_msg;
1162		if (msg != ERR_PTR(-EAGAIN))
1163			goto out_unlock0;
1164
1165		list_del(&msr_d.r_list);
1166		if (signal_pending(current)) {
1167			msg = ERR_PTR(-ERESTARTNOHAND);
1168			goto out_unlock0;
1169		}
1170
1171		ipc_unlock_object(&msq->q_perm);
1172	}
1173
1174out_unlock0:
1175	ipc_unlock_object(&msq->q_perm);
1176	wake_up_q(&wake_q);
1177out_unlock1:
1178	rcu_read_unlock();
1179	if (IS_ERR(msg)) {
1180		free_copy(copy);
1181		return PTR_ERR(msg);
1182	}
1183
1184	bufsz = msg_handler(buf, msg, bufsz);
1185	free_msg(msg);
1186
1187	return bufsz;
1188}
1189
1190long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1191		 long msgtyp, int msgflg)
1192{
1193	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1194}
1195
1196SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1197		long, msgtyp, int, msgflg)
1198{
1199	return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1200}
1201
1202#ifdef CONFIG_COMPAT
1203static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1204{
1205	struct compat_msgbuf __user *msgp = dest;
1206	size_t msgsz;
1207
1208	if (put_user(msg->m_type, &msgp->mtype))
1209		return -EFAULT;
1210
1211	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1212	if (store_msg(msgp->mtext, msg, msgsz))
1213		return -EFAULT;
1214	return msgsz;
1215}
1216
1217long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1218			compat_long_t msgtyp, int msgflg)
1219{
1220	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1221			 msgflg, compat_do_msg_fill);
1222}
1223
1224COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1225		       compat_ssize_t, msgsz, compat_long_t, msgtyp,
1226		       int, msgflg)
1227{
1228	return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1229}
1230#endif
1231
1232int msg_init_ns(struct ipc_namespace *ns)
1233{
1234	ns->msg_ctlmax = MSGMAX;
1235	ns->msg_ctlmnb = MSGMNB;
1236	ns->msg_ctlmni = MSGMNI;
1237
1238	atomic_set(&ns->msg_bytes, 0);
1239	atomic_set(&ns->msg_hdrs, 0);
1240	return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1241}
1242
1243#ifdef CONFIG_IPC_NS
1244void msg_exit_ns(struct ipc_namespace *ns)
1245{
1246	free_ipcs(ns, &msg_ids(ns), freeque);
1247	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1248	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1249}
1250#endif
1251
1252#ifdef CONFIG_PROC_FS
1253static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1254{
1255	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1256	struct user_namespace *user_ns = seq_user_ns(s);
1257	struct kern_ipc_perm *ipcp = it;
1258	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1259
1260	seq_printf(s,
1261		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1262		   msq->q_perm.key,
1263		   msq->q_perm.id,
1264		   msq->q_perm.mode,
1265		   msq->q_cbytes,
1266		   msq->q_qnum,
1267		   pid_nr_ns(msq->q_lspid, pid_ns),
1268		   pid_nr_ns(msq->q_lrpid, pid_ns),
1269		   from_kuid_munged(user_ns, msq->q_perm.uid),
1270		   from_kgid_munged(user_ns, msq->q_perm.gid),
1271		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1272		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1273		   msq->q_stime,
1274		   msq->q_rtime,
1275		   msq->q_ctime);
1276
1277	return 0;
1278}
1279#endif
1280
1281int __init msg_init(void)
1282{
1283	const int err = msg_init_ns(&init_ipc_ns);
1284
1285	ipc_init_proc_interface("sysvipc/msg",
1286				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1287				IPC_MSG_IDS, sysvipc_msg_proc_show);
1288	return err;
1289}
1290