msg.c revision 0d97a82b
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/*
65 * MSG_BARRIER Locking:
66 *
67 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
68 * does not acquire any locks when it sees that a message exists in
69 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
70 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
71 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
72 */
73
74/* one msg_receiver structure for each sleeping receiver */
75struct msg_receiver {
76	struct list_head	r_list;
77	struct task_struct	*r_tsk;
78
79	int			r_mode;
80	long			r_msgtype;
81	long			r_maxsize;
82
83	struct msg_msg		*r_msg;
84};
85
86/* one msg_sender for each sleeping sender */
87struct msg_sender {
88	struct list_head	list;
89	struct task_struct	*tsk;
90	size_t                  msgsz;
91};
92
93#define SEARCH_ANY		1
94#define SEARCH_EQUAL		2
95#define SEARCH_NOTEQUAL		3
96#define SEARCH_LESSEQUAL	4
97#define SEARCH_NUMBER		5
98
99#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
100
101static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
102{
103	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
104
105	if (IS_ERR(ipcp))
106		return ERR_CAST(ipcp);
107
108	return container_of(ipcp, struct msg_queue, q_perm);
109}
110
111static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
112							int id)
113{
114	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
115
116	if (IS_ERR(ipcp))
117		return ERR_CAST(ipcp);
118
119	return container_of(ipcp, struct msg_queue, q_perm);
120}
121
122static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
123{
124	ipc_rmid(&msg_ids(ns), &s->q_perm);
125}
126
127static void msg_rcu_free(struct rcu_head *head)
128{
129	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
130	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
131
132	security_msg_queue_free(&msq->q_perm);
133	kvfree(msq);
134}
135
136/**
137 * newque - Create a new msg queue
138 * @ns: namespace
139 * @params: ptr to the structure that contains the key and msgflg
140 *
141 * Called with msg_ids.rwsem held (writer)
142 */
143static int newque(struct ipc_namespace *ns, struct ipc_params *params)
144{
145	struct msg_queue *msq;
146	int retval;
147	key_t key = params->key;
148	int msgflg = params->flg;
149
150	msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
151	if (unlikely(!msq))
152		return -ENOMEM;
153
154	msq->q_perm.mode = msgflg & S_IRWXUGO;
155	msq->q_perm.key = key;
156
157	msq->q_perm.security = NULL;
158	retval = security_msg_queue_alloc(&msq->q_perm);
159	if (retval) {
160		kvfree(msq);
161		return retval;
162	}
163
164	msq->q_stime = msq->q_rtime = 0;
165	msq->q_ctime = ktime_get_real_seconds();
166	msq->q_cbytes = msq->q_qnum = 0;
167	msq->q_qbytes = ns->msg_ctlmnb;
168	msq->q_lspid = msq->q_lrpid = NULL;
169	INIT_LIST_HEAD(&msq->q_messages);
170	INIT_LIST_HEAD(&msq->q_receivers);
171	INIT_LIST_HEAD(&msq->q_senders);
172
173	/* ipc_addid() locks msq upon success. */
174	retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
175	if (retval < 0) {
176		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
177		return retval;
178	}
179
180	ipc_unlock_object(&msq->q_perm);
181	rcu_read_unlock();
182
183	return msq->q_perm.id;
184}
185
186static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
187{
188	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
189		1 + msq->q_qnum <= msq->q_qbytes;
190}
191
192static inline void ss_add(struct msg_queue *msq,
193			  struct msg_sender *mss, size_t msgsz)
194{
195	mss->tsk = current;
196	mss->msgsz = msgsz;
197	/*
198	 * No memory barrier required: we did ipc_lock_object(),
199	 * and the waker obtains that lock before calling wake_q_add().
200	 */
201	__set_current_state(TASK_INTERRUPTIBLE);
202	list_add_tail(&mss->list, &msq->q_senders);
203}
204
205static inline void ss_del(struct msg_sender *mss)
206{
207	if (mss->list.next)
208		list_del(&mss->list);
209}
210
211static void ss_wakeup(struct msg_queue *msq,
212		      struct wake_q_head *wake_q, bool kill)
213{
214	struct msg_sender *mss, *t;
215	struct task_struct *stop_tsk = NULL;
216	struct list_head *h = &msq->q_senders;
217
218	list_for_each_entry_safe(mss, t, h, list) {
219		if (kill)
220			mss->list.next = NULL;
221
222		/*
223		 * Stop at the first task we don't wakeup,
224		 * we've already iterated the original
225		 * sender queue.
226		 */
227		else if (stop_tsk == mss->tsk)
228			break;
229		/*
230		 * We are not in an EIDRM scenario here, therefore
231		 * verify that we really need to wakeup the task.
232		 * To maintain current semantics and wakeup order,
233		 * move the sender to the tail on behalf of the
234		 * blocked task.
235		 */
236		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
237			if (!stop_tsk)
238				stop_tsk = mss->tsk;
239
240			list_move_tail(&mss->list, &msq->q_senders);
241			continue;
242		}
243
244		wake_q_add(wake_q, mss->tsk);
245	}
246}
247
248static void expunge_all(struct msg_queue *msq, int res,
249			struct wake_q_head *wake_q)
250{
251	struct msg_receiver *msr, *t;
252
253	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
254		get_task_struct(msr->r_tsk);
255
256		/* see MSG_BARRIER for purpose/pairing */
257		smp_store_release(&msr->r_msg, ERR_PTR(res));
258		wake_q_add_safe(wake_q, msr->r_tsk);
259	}
260}
261
262/*
263 * freeque() wakes up waiters on the sender and receiver waiting queue,
264 * removes the message queue from message queue ID IDR, and cleans up all the
265 * messages associated with this queue.
266 *
267 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
268 * before freeque() is called. msg_ids.rwsem remains locked on exit.
269 */
270static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
271{
272	struct msg_msg *msg, *t;
273	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
274	DEFINE_WAKE_Q(wake_q);
275
276	expunge_all(msq, -EIDRM, &wake_q);
277	ss_wakeup(msq, &wake_q, true);
278	msg_rmid(ns, msq);
279	ipc_unlock_object(&msq->q_perm);
280	wake_up_q(&wake_q);
281	rcu_read_unlock();
282
283	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
284		atomic_dec(&ns->msg_hdrs);
285		free_msg(msg);
286	}
287	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
288	ipc_update_pid(&msq->q_lspid, NULL);
289	ipc_update_pid(&msq->q_lrpid, NULL);
290	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
291}
292
293long ksys_msgget(key_t key, int msgflg)
294{
295	struct ipc_namespace *ns;
296	static const struct ipc_ops msg_ops = {
297		.getnew = newque,
298		.associate = security_msg_queue_associate,
299	};
300	struct ipc_params msg_params;
301
302	ns = current->nsproxy->ipc_ns;
303
304	msg_params.key = key;
305	msg_params.flg = msgflg;
306
307	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
308}
309
310SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
311{
312	return ksys_msgget(key, msgflg);
313}
314
315static inline unsigned long
316copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
317{
318	switch (version) {
319	case IPC_64:
320		return copy_to_user(buf, in, sizeof(*in));
321	case IPC_OLD:
322	{
323		struct msqid_ds out;
324
325		memset(&out, 0, sizeof(out));
326
327		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
328
329		out.msg_stime		= in->msg_stime;
330		out.msg_rtime		= in->msg_rtime;
331		out.msg_ctime		= in->msg_ctime;
332
333		if (in->msg_cbytes > USHRT_MAX)
334			out.msg_cbytes	= USHRT_MAX;
335		else
336			out.msg_cbytes	= in->msg_cbytes;
337		out.msg_lcbytes		= in->msg_cbytes;
338
339		if (in->msg_qnum > USHRT_MAX)
340			out.msg_qnum	= USHRT_MAX;
341		else
342			out.msg_qnum	= in->msg_qnum;
343
344		if (in->msg_qbytes > USHRT_MAX)
345			out.msg_qbytes	= USHRT_MAX;
346		else
347			out.msg_qbytes	= in->msg_qbytes;
348		out.msg_lqbytes		= in->msg_qbytes;
349
350		out.msg_lspid		= in->msg_lspid;
351		out.msg_lrpid		= in->msg_lrpid;
352
353		return copy_to_user(buf, &out, sizeof(out));
354	}
355	default:
356		return -EINVAL;
357	}
358}
359
360static inline unsigned long
361copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
362{
363	switch (version) {
364	case IPC_64:
365		if (copy_from_user(out, buf, sizeof(*out)))
366			return -EFAULT;
367		return 0;
368	case IPC_OLD:
369	{
370		struct msqid_ds tbuf_old;
371
372		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
373			return -EFAULT;
374
375		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
376		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
377		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
378
379		if (tbuf_old.msg_qbytes == 0)
380			out->msg_qbytes	= tbuf_old.msg_lqbytes;
381		else
382			out->msg_qbytes	= tbuf_old.msg_qbytes;
383
384		return 0;
385	}
386	default:
387		return -EINVAL;
388	}
389}
390
391/*
392 * This function handles some msgctl commands which require the rwsem
393 * to be held in write mode.
394 * NOTE: no locks must be held, the rwsem is taken inside this function.
395 */
396static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
397			struct msqid64_ds *msqid64)
398{
399	struct kern_ipc_perm *ipcp;
400	struct msg_queue *msq;
401	int err;
402
403	down_write(&msg_ids(ns).rwsem);
404	rcu_read_lock();
405
406	ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
407				      &msqid64->msg_perm, msqid64->msg_qbytes);
408	if (IS_ERR(ipcp)) {
409		err = PTR_ERR(ipcp);
410		goto out_unlock1;
411	}
412
413	msq = container_of(ipcp, struct msg_queue, q_perm);
414
415	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
416	if (err)
417		goto out_unlock1;
418
419	switch (cmd) {
420	case IPC_RMID:
421		ipc_lock_object(&msq->q_perm);
422		/* freeque unlocks the ipc object and rcu */
423		freeque(ns, ipcp);
424		goto out_up;
425	case IPC_SET:
426	{
427		DEFINE_WAKE_Q(wake_q);
428
429		if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
430		    !capable(CAP_SYS_RESOURCE)) {
431			err = -EPERM;
432			goto out_unlock1;
433		}
434
435		ipc_lock_object(&msq->q_perm);
436		err = ipc_update_perm(&msqid64->msg_perm, ipcp);
437		if (err)
438			goto out_unlock0;
439
440		msq->q_qbytes = msqid64->msg_qbytes;
441
442		msq->q_ctime = ktime_get_real_seconds();
443		/*
444		 * Sleeping receivers might be excluded by
445		 * stricter permissions.
446		 */
447		expunge_all(msq, -EAGAIN, &wake_q);
448		/*
449		 * Sleeping senders might be able to send
450		 * due to a larger queue size.
451		 */
452		ss_wakeup(msq, &wake_q, false);
453		ipc_unlock_object(&msq->q_perm);
454		wake_up_q(&wake_q);
455
456		goto out_unlock1;
457	}
458	default:
459		err = -EINVAL;
460		goto out_unlock1;
461	}
462
463out_unlock0:
464	ipc_unlock_object(&msq->q_perm);
465out_unlock1:
466	rcu_read_unlock();
467out_up:
468	up_write(&msg_ids(ns).rwsem);
469	return err;
470}
471
472static int msgctl_info(struct ipc_namespace *ns, int msqid,
473			 int cmd, struct msginfo *msginfo)
474{
475	int err;
476	int max_idx;
477
478	/*
479	 * We must not return kernel stack data.
480	 * due to padding, it's not enough
481	 * to set all member fields.
482	 */
483	err = security_msg_queue_msgctl(NULL, cmd);
484	if (err)
485		return err;
486
487	memset(msginfo, 0, sizeof(*msginfo));
488	msginfo->msgmni = ns->msg_ctlmni;
489	msginfo->msgmax = ns->msg_ctlmax;
490	msginfo->msgmnb = ns->msg_ctlmnb;
491	msginfo->msgssz = MSGSSZ;
492	msginfo->msgseg = MSGSEG;
493	down_read(&msg_ids(ns).rwsem);
494	if (cmd == MSG_INFO) {
495		msginfo->msgpool = msg_ids(ns).in_use;
496		msginfo->msgmap = atomic_read(&ns->msg_hdrs);
497		msginfo->msgtql = atomic_read(&ns->msg_bytes);
498	} else {
499		msginfo->msgmap = MSGMAP;
500		msginfo->msgpool = MSGPOOL;
501		msginfo->msgtql = MSGTQL;
502	}
503	max_idx = ipc_get_maxidx(&msg_ids(ns));
504	up_read(&msg_ids(ns).rwsem);
505	return (max_idx < 0) ? 0 : max_idx;
506}
507
508static int msgctl_stat(struct ipc_namespace *ns, int msqid,
509			 int cmd, struct msqid64_ds *p)
510{
511	struct msg_queue *msq;
512	int err;
513
514	memset(p, 0, sizeof(*p));
515
516	rcu_read_lock();
517	if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
518		msq = msq_obtain_object(ns, msqid);
519		if (IS_ERR(msq)) {
520			err = PTR_ERR(msq);
521			goto out_unlock;
522		}
523	} else { /* IPC_STAT */
524		msq = msq_obtain_object_check(ns, msqid);
525		if (IS_ERR(msq)) {
526			err = PTR_ERR(msq);
527			goto out_unlock;
528		}
529	}
530
531	/* see comment for SHM_STAT_ANY */
532	if (cmd == MSG_STAT_ANY)
533		audit_ipc_obj(&msq->q_perm);
534	else {
535		err = -EACCES;
536		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
537			goto out_unlock;
538	}
539
540	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
541	if (err)
542		goto out_unlock;
543
544	ipc_lock_object(&msq->q_perm);
545
546	if (!ipc_valid_object(&msq->q_perm)) {
547		ipc_unlock_object(&msq->q_perm);
548		err = -EIDRM;
549		goto out_unlock;
550	}
551
552	kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
553	p->msg_stime  = msq->q_stime;
554	p->msg_rtime  = msq->q_rtime;
555	p->msg_ctime  = msq->q_ctime;
556#ifndef CONFIG_64BIT
557	p->msg_stime_high = msq->q_stime >> 32;
558	p->msg_rtime_high = msq->q_rtime >> 32;
559	p->msg_ctime_high = msq->q_ctime >> 32;
560#endif
561	p->msg_cbytes = msq->q_cbytes;
562	p->msg_qnum   = msq->q_qnum;
563	p->msg_qbytes = msq->q_qbytes;
564	p->msg_lspid  = pid_vnr(msq->q_lspid);
565	p->msg_lrpid  = pid_vnr(msq->q_lrpid);
566
567	if (cmd == IPC_STAT) {
568		/*
569		 * As defined in SUS:
570		 * Return 0 on success
571		 */
572		err = 0;
573	} else {
574		/*
575		 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
576		 * Return the full id, including the sequence number
577		 */
578		err = msq->q_perm.id;
579	}
580
581	ipc_unlock_object(&msq->q_perm);
582out_unlock:
583	rcu_read_unlock();
584	return err;
585}
586
587static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
588{
589	struct ipc_namespace *ns;
590	struct msqid64_ds msqid64;
591	int err;
592
593	if (msqid < 0 || cmd < 0)
594		return -EINVAL;
595
596	ns = current->nsproxy->ipc_ns;
597
598	switch (cmd) {
599	case IPC_INFO:
600	case MSG_INFO: {
601		struct msginfo msginfo;
602		err = msgctl_info(ns, msqid, cmd, &msginfo);
603		if (err < 0)
604			return err;
605		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
606			err = -EFAULT;
607		return err;
608	}
609	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
610	case MSG_STAT_ANY:
611	case IPC_STAT:
612		err = msgctl_stat(ns, msqid, cmd, &msqid64);
613		if (err < 0)
614			return err;
615		if (copy_msqid_to_user(buf, &msqid64, version))
616			err = -EFAULT;
617		return err;
618	case IPC_SET:
619		if (copy_msqid_from_user(&msqid64, buf, version))
620			return -EFAULT;
621		/* fallthru */
622	case IPC_RMID:
623		return msgctl_down(ns, msqid, cmd, &msqid64);
624	default:
625		return  -EINVAL;
626	}
627}
628
629SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
630{
631	return ksys_msgctl(msqid, cmd, buf, IPC_64);
632}
633
634#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
635long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
636{
637	int version = ipc_parse_version(&cmd);
638
639	return ksys_msgctl(msqid, cmd, buf, version);
640}
641
642SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
643{
644	return ksys_old_msgctl(msqid, cmd, buf);
645}
646#endif
647
648#ifdef CONFIG_COMPAT
649
650struct compat_msqid_ds {
651	struct compat_ipc_perm msg_perm;
652	compat_uptr_t msg_first;
653	compat_uptr_t msg_last;
654	old_time32_t msg_stime;
655	old_time32_t msg_rtime;
656	old_time32_t msg_ctime;
657	compat_ulong_t msg_lcbytes;
658	compat_ulong_t msg_lqbytes;
659	unsigned short msg_cbytes;
660	unsigned short msg_qnum;
661	unsigned short msg_qbytes;
662	compat_ipc_pid_t msg_lspid;
663	compat_ipc_pid_t msg_lrpid;
664};
665
666static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
667					int version)
668{
669	memset(out, 0, sizeof(*out));
670	if (version == IPC_64) {
671		struct compat_msqid64_ds __user *p = buf;
672		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
673			return -EFAULT;
674		if (get_user(out->msg_qbytes, &p->msg_qbytes))
675			return -EFAULT;
676	} else {
677		struct compat_msqid_ds __user *p = buf;
678		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
679			return -EFAULT;
680		if (get_user(out->msg_qbytes, &p->msg_qbytes))
681			return -EFAULT;
682	}
683	return 0;
684}
685
686static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
687					int version)
688{
689	if (version == IPC_64) {
690		struct compat_msqid64_ds v;
691		memset(&v, 0, sizeof(v));
692		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
693		v.msg_stime	 = lower_32_bits(in->msg_stime);
694		v.msg_stime_high = upper_32_bits(in->msg_stime);
695		v.msg_rtime	 = lower_32_bits(in->msg_rtime);
696		v.msg_rtime_high = upper_32_bits(in->msg_rtime);
697		v.msg_ctime	 = lower_32_bits(in->msg_ctime);
698		v.msg_ctime_high = upper_32_bits(in->msg_ctime);
699		v.msg_cbytes = in->msg_cbytes;
700		v.msg_qnum = in->msg_qnum;
701		v.msg_qbytes = in->msg_qbytes;
702		v.msg_lspid = in->msg_lspid;
703		v.msg_lrpid = in->msg_lrpid;
704		return copy_to_user(buf, &v, sizeof(v));
705	} else {
706		struct compat_msqid_ds v;
707		memset(&v, 0, sizeof(v));
708		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
709		v.msg_stime = in->msg_stime;
710		v.msg_rtime = in->msg_rtime;
711		v.msg_ctime = in->msg_ctime;
712		v.msg_cbytes = in->msg_cbytes;
713		v.msg_qnum = in->msg_qnum;
714		v.msg_qbytes = in->msg_qbytes;
715		v.msg_lspid = in->msg_lspid;
716		v.msg_lrpid = in->msg_lrpid;
717		return copy_to_user(buf, &v, sizeof(v));
718	}
719}
720
721static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
722{
723	struct ipc_namespace *ns;
724	int err;
725	struct msqid64_ds msqid64;
726
727	ns = current->nsproxy->ipc_ns;
728
729	if (msqid < 0 || cmd < 0)
730		return -EINVAL;
731
732	switch (cmd & (~IPC_64)) {
733	case IPC_INFO:
734	case MSG_INFO: {
735		struct msginfo msginfo;
736		err = msgctl_info(ns, msqid, cmd, &msginfo);
737		if (err < 0)
738			return err;
739		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
740			err = -EFAULT;
741		return err;
742	}
743	case IPC_STAT:
744	case MSG_STAT:
745	case MSG_STAT_ANY:
746		err = msgctl_stat(ns, msqid, cmd, &msqid64);
747		if (err < 0)
748			return err;
749		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
750			err = -EFAULT;
751		return err;
752	case IPC_SET:
753		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
754			return -EFAULT;
755		/* fallthru */
756	case IPC_RMID:
757		return msgctl_down(ns, msqid, cmd, &msqid64);
758	default:
759		return -EINVAL;
760	}
761}
762
763COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
764{
765	return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
766}
767
768#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
769long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
770{
771	int version = compat_ipc_parse_version(&cmd);
772
773	return compat_ksys_msgctl(msqid, cmd, uptr, version);
774}
775
776COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
777{
778	return compat_ksys_old_msgctl(msqid, cmd, uptr);
779}
780#endif
781#endif
782
783static int testmsg(struct msg_msg *msg, long type, int mode)
784{
785	switch (mode) {
786	case SEARCH_ANY:
787	case SEARCH_NUMBER:
788		return 1;
789	case SEARCH_LESSEQUAL:
790		if (msg->m_type <= type)
791			return 1;
792		break;
793	case SEARCH_EQUAL:
794		if (msg->m_type == type)
795			return 1;
796		break;
797	case SEARCH_NOTEQUAL:
798		if (msg->m_type != type)
799			return 1;
800		break;
801	}
802	return 0;
803}
804
805static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
806				 struct wake_q_head *wake_q)
807{
808	struct msg_receiver *msr, *t;
809
810	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
811		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
812		    !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
813					       msr->r_msgtype, msr->r_mode)) {
814
815			list_del(&msr->r_list);
816			if (msr->r_maxsize < msg->m_ts) {
817				wake_q_add(wake_q, msr->r_tsk);
818
819				/* See expunge_all regarding memory barrier */
820				smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
821			} else {
822				ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
823				msq->q_rtime = ktime_get_real_seconds();
824
825				wake_q_add(wake_q, msr->r_tsk);
826
827				/* See expunge_all regarding memory barrier */
828				smp_store_release(&msr->r_msg, msg);
829				return 1;
830			}
831		}
832	}
833
834	return 0;
835}
836
837static long do_msgsnd(int msqid, long mtype, void __user *mtext,
838		size_t msgsz, int msgflg)
839{
840	struct msg_queue *msq;
841	struct msg_msg *msg;
842	int err;
843	struct ipc_namespace *ns;
844	DEFINE_WAKE_Q(wake_q);
845
846	ns = current->nsproxy->ipc_ns;
847
848	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
849		return -EINVAL;
850	if (mtype < 1)
851		return -EINVAL;
852
853	msg = load_msg(mtext, msgsz);
854	if (IS_ERR(msg))
855		return PTR_ERR(msg);
856
857	msg->m_type = mtype;
858	msg->m_ts = msgsz;
859
860	rcu_read_lock();
861	msq = msq_obtain_object_check(ns, msqid);
862	if (IS_ERR(msq)) {
863		err = PTR_ERR(msq);
864		goto out_unlock1;
865	}
866
867	ipc_lock_object(&msq->q_perm);
868
869	for (;;) {
870		struct msg_sender s;
871
872		err = -EACCES;
873		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
874			goto out_unlock0;
875
876		/* raced with RMID? */
877		if (!ipc_valid_object(&msq->q_perm)) {
878			err = -EIDRM;
879			goto out_unlock0;
880		}
881
882		err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
883		if (err)
884			goto out_unlock0;
885
886		if (msg_fits_inqueue(msq, msgsz))
887			break;
888
889		/* queue full, wait: */
890		if (msgflg & IPC_NOWAIT) {
891			err = -EAGAIN;
892			goto out_unlock0;
893		}
894
895		/* enqueue the sender and prepare to block */
896		ss_add(msq, &s, msgsz);
897
898		if (!ipc_rcu_getref(&msq->q_perm)) {
899			err = -EIDRM;
900			goto out_unlock0;
901		}
902
903		ipc_unlock_object(&msq->q_perm);
904		rcu_read_unlock();
905		schedule();
906
907		rcu_read_lock();
908		ipc_lock_object(&msq->q_perm);
909
910		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
911		/* raced with RMID? */
912		if (!ipc_valid_object(&msq->q_perm)) {
913			err = -EIDRM;
914			goto out_unlock0;
915		}
916		ss_del(&s);
917
918		if (signal_pending(current)) {
919			err = -ERESTARTNOHAND;
920			goto out_unlock0;
921		}
922
923	}
924
925	ipc_update_pid(&msq->q_lspid, task_tgid(current));
926	msq->q_stime = ktime_get_real_seconds();
927
928	if (!pipelined_send(msq, msg, &wake_q)) {
929		/* no one is waiting for this message, enqueue it */
930		list_add_tail(&msg->m_list, &msq->q_messages);
931		msq->q_cbytes += msgsz;
932		msq->q_qnum++;
933		atomic_add(msgsz, &ns->msg_bytes);
934		atomic_inc(&ns->msg_hdrs);
935	}
936
937	err = 0;
938	msg = NULL;
939
940out_unlock0:
941	ipc_unlock_object(&msq->q_perm);
942	wake_up_q(&wake_q);
943out_unlock1:
944	rcu_read_unlock();
945	if (msg != NULL)
946		free_msg(msg);
947	return err;
948}
949
950long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
951		 int msgflg)
952{
953	long mtype;
954
955	if (get_user(mtype, &msgp->mtype))
956		return -EFAULT;
957	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
958}
959
960SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
961		int, msgflg)
962{
963	return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
964}
965
966#ifdef CONFIG_COMPAT
967
968struct compat_msgbuf {
969	compat_long_t mtype;
970	char mtext[1];
971};
972
973long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
974		       compat_ssize_t msgsz, int msgflg)
975{
976	struct compat_msgbuf __user *up = compat_ptr(msgp);
977	compat_long_t mtype;
978
979	if (get_user(mtype, &up->mtype))
980		return -EFAULT;
981	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
982}
983
984COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
985		       compat_ssize_t, msgsz, int, msgflg)
986{
987	return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
988}
989#endif
990
991static inline int convert_mode(long *msgtyp, int msgflg)
992{
993	if (msgflg & MSG_COPY)
994		return SEARCH_NUMBER;
995	/*
996	 *  find message of correct type.
997	 *  msgtyp = 0 => get first.
998	 *  msgtyp > 0 => get first message of matching type.
999	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
1000	 */
1001	if (*msgtyp == 0)
1002		return SEARCH_ANY;
1003	if (*msgtyp < 0) {
1004		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1005			*msgtyp = LONG_MAX;
1006		else
1007			*msgtyp = -*msgtyp;
1008		return SEARCH_LESSEQUAL;
1009	}
1010	if (msgflg & MSG_EXCEPT)
1011		return SEARCH_NOTEQUAL;
1012	return SEARCH_EQUAL;
1013}
1014
1015static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1016{
1017	struct msgbuf __user *msgp = dest;
1018	size_t msgsz;
1019
1020	if (put_user(msg->m_type, &msgp->mtype))
1021		return -EFAULT;
1022
1023	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1024	if (store_msg(msgp->mtext, msg, msgsz))
1025		return -EFAULT;
1026	return msgsz;
1027}
1028
1029#ifdef CONFIG_CHECKPOINT_RESTORE
1030/*
1031 * This function creates new kernel message structure, large enough to store
1032 * bufsz message bytes.
1033 */
1034static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1035{
1036	struct msg_msg *copy;
1037
1038	/*
1039	 * Create dummy message to copy real message to.
1040	 */
1041	copy = load_msg(buf, bufsz);
1042	if (!IS_ERR(copy))
1043		copy->m_ts = bufsz;
1044	return copy;
1045}
1046
1047static inline void free_copy(struct msg_msg *copy)
1048{
1049	if (copy)
1050		free_msg(copy);
1051}
1052#else
1053static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1054{
1055	return ERR_PTR(-ENOSYS);
1056}
1057
1058static inline void free_copy(struct msg_msg *copy)
1059{
1060}
1061#endif
1062
1063static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1064{
1065	struct msg_msg *msg, *found = NULL;
1066	long count = 0;
1067
1068	list_for_each_entry(msg, &msq->q_messages, m_list) {
1069		if (testmsg(msg, *msgtyp, mode) &&
1070		    !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1071					       *msgtyp, mode)) {
1072			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1073				*msgtyp = msg->m_type - 1;
1074				found = msg;
1075			} else if (mode == SEARCH_NUMBER) {
1076				if (*msgtyp == count)
1077					return msg;
1078			} else
1079				return msg;
1080			count++;
1081		}
1082	}
1083
1084	return found ?: ERR_PTR(-EAGAIN);
1085}
1086
1087static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1088	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1089{
1090	int mode;
1091	struct msg_queue *msq;
1092	struct ipc_namespace *ns;
1093	struct msg_msg *msg, *copy = NULL;
1094	DEFINE_WAKE_Q(wake_q);
1095
1096	ns = current->nsproxy->ipc_ns;
1097
1098	if (msqid < 0 || (long) bufsz < 0)
1099		return -EINVAL;
1100
1101	if (msgflg & MSG_COPY) {
1102		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1103			return -EINVAL;
1104		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1105		if (IS_ERR(copy))
1106			return PTR_ERR(copy);
1107	}
1108	mode = convert_mode(&msgtyp, msgflg);
1109
1110	rcu_read_lock();
1111	msq = msq_obtain_object_check(ns, msqid);
1112	if (IS_ERR(msq)) {
1113		rcu_read_unlock();
1114		free_copy(copy);
1115		return PTR_ERR(msq);
1116	}
1117
1118	for (;;) {
1119		struct msg_receiver msr_d;
1120
1121		msg = ERR_PTR(-EACCES);
1122		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1123			goto out_unlock1;
1124
1125		ipc_lock_object(&msq->q_perm);
1126
1127		/* raced with RMID? */
1128		if (!ipc_valid_object(&msq->q_perm)) {
1129			msg = ERR_PTR(-EIDRM);
1130			goto out_unlock0;
1131		}
1132
1133		msg = find_msg(msq, &msgtyp, mode);
1134		if (!IS_ERR(msg)) {
1135			/*
1136			 * Found a suitable message.
1137			 * Unlink it from the queue.
1138			 */
1139			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1140				msg = ERR_PTR(-E2BIG);
1141				goto out_unlock0;
1142			}
1143			/*
1144			 * If we are copying, then do not unlink message and do
1145			 * not update queue parameters.
1146			 */
1147			if (msgflg & MSG_COPY) {
1148				msg = copy_msg(msg, copy);
1149				goto out_unlock0;
1150			}
1151
1152			list_del(&msg->m_list);
1153			msq->q_qnum--;
1154			msq->q_rtime = ktime_get_real_seconds();
1155			ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1156			msq->q_cbytes -= msg->m_ts;
1157			atomic_sub(msg->m_ts, &ns->msg_bytes);
1158			atomic_dec(&ns->msg_hdrs);
1159			ss_wakeup(msq, &wake_q, false);
1160
1161			goto out_unlock0;
1162		}
1163
1164		/* No message waiting. Wait for a message */
1165		if (msgflg & IPC_NOWAIT) {
1166			msg = ERR_PTR(-ENOMSG);
1167			goto out_unlock0;
1168		}
1169
1170		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1171		msr_d.r_tsk = current;
1172		msr_d.r_msgtype = msgtyp;
1173		msr_d.r_mode = mode;
1174		if (msgflg & MSG_NOERROR)
1175			msr_d.r_maxsize = INT_MAX;
1176		else
1177			msr_d.r_maxsize = bufsz;
1178
1179		/* memory barrier not require due to ipc_lock_object() */
1180		WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1181
1182		/* memory barrier not required, we own ipc_lock_object() */
1183		__set_current_state(TASK_INTERRUPTIBLE);
1184
1185		ipc_unlock_object(&msq->q_perm);
1186		rcu_read_unlock();
1187		schedule();
1188
1189		/*
1190		 * Lockless receive, part 1:
1191		 * We don't hold a reference to the queue and getting a
1192		 * reference would defeat the idea of a lockless operation,
1193		 * thus the code relies on rcu to guarantee the existence of
1194		 * msq:
1195		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1196		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1197		 */
1198		rcu_read_lock();
1199
1200		/*
1201		 * Lockless receive, part 2:
1202		 * The work in pipelined_send() and expunge_all():
1203		 * - Set pointer to message
1204		 * - Queue the receiver task for later wakeup
1205		 * - Wake up the process after the lock is dropped.
1206		 *
1207		 * Should the process wake up before this wakeup (due to a
1208		 * signal) it will either see the message and continue ...
1209		 */
1210		msg = READ_ONCE(msr_d.r_msg);
1211		if (msg != ERR_PTR(-EAGAIN)) {
1212			/* see MSG_BARRIER for purpose/pairing */
1213			smp_acquire__after_ctrl_dep();
1214
1215			goto out_unlock1;
1216		}
1217
1218		 /*
1219		  * ... or see -EAGAIN, acquire the lock to check the message
1220		  * again.
1221		  */
1222		ipc_lock_object(&msq->q_perm);
1223
1224		msg = READ_ONCE(msr_d.r_msg);
1225		if (msg != ERR_PTR(-EAGAIN))
1226			goto out_unlock0;
1227
1228		list_del(&msr_d.r_list);
1229		if (signal_pending(current)) {
1230			msg = ERR_PTR(-ERESTARTNOHAND);
1231			goto out_unlock0;
1232		}
1233
1234		ipc_unlock_object(&msq->q_perm);
1235	}
1236
1237out_unlock0:
1238	ipc_unlock_object(&msq->q_perm);
1239	wake_up_q(&wake_q);
1240out_unlock1:
1241	rcu_read_unlock();
1242	if (IS_ERR(msg)) {
1243		free_copy(copy);
1244		return PTR_ERR(msg);
1245	}
1246
1247	bufsz = msg_handler(buf, msg, bufsz);
1248	free_msg(msg);
1249
1250	return bufsz;
1251}
1252
1253long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1254		 long msgtyp, int msgflg)
1255{
1256	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1257}
1258
1259SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1260		long, msgtyp, int, msgflg)
1261{
1262	return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1263}
1264
1265#ifdef CONFIG_COMPAT
1266static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1267{
1268	struct compat_msgbuf __user *msgp = dest;
1269	size_t msgsz;
1270
1271	if (put_user(msg->m_type, &msgp->mtype))
1272		return -EFAULT;
1273
1274	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1275	if (store_msg(msgp->mtext, msg, msgsz))
1276		return -EFAULT;
1277	return msgsz;
1278}
1279
1280long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1281			compat_long_t msgtyp, int msgflg)
1282{
1283	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1284			 msgflg, compat_do_msg_fill);
1285}
1286
1287COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1288		       compat_ssize_t, msgsz, compat_long_t, msgtyp,
1289		       int, msgflg)
1290{
1291	return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1292}
1293#endif
1294
1295void msg_init_ns(struct ipc_namespace *ns)
1296{
1297	ns->msg_ctlmax = MSGMAX;
1298	ns->msg_ctlmnb = MSGMNB;
1299	ns->msg_ctlmni = MSGMNI;
1300
1301	atomic_set(&ns->msg_bytes, 0);
1302	atomic_set(&ns->msg_hdrs, 0);
1303	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1304}
1305
1306#ifdef CONFIG_IPC_NS
1307void msg_exit_ns(struct ipc_namespace *ns)
1308{
1309	free_ipcs(ns, &msg_ids(ns), freeque);
1310	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1311	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1312}
1313#endif
1314
1315#ifdef CONFIG_PROC_FS
1316static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1317{
1318	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1319	struct user_namespace *user_ns = seq_user_ns(s);
1320	struct kern_ipc_perm *ipcp = it;
1321	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1322
1323	seq_printf(s,
1324		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1325		   msq->q_perm.key,
1326		   msq->q_perm.id,
1327		   msq->q_perm.mode,
1328		   msq->q_cbytes,
1329		   msq->q_qnum,
1330		   pid_nr_ns(msq->q_lspid, pid_ns),
1331		   pid_nr_ns(msq->q_lrpid, pid_ns),
1332		   from_kuid_munged(user_ns, msq->q_perm.uid),
1333		   from_kgid_munged(user_ns, msq->q_perm.gid),
1334		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1335		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1336		   msq->q_stime,
1337		   msq->q_rtime,
1338		   msq->q_ctime);
1339
1340	return 0;
1341}
1342#endif
1343
1344void __init msg_init(void)
1345{
1346	msg_init_ns(&init_ipc_ns);
1347
1348	ipc_init_proc_interface("sysvipc/msg",
1349				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1350				IPC_MSG_IDS, sysvipc_msg_proc_show);
1351}
1352