xref: /openbmc/linux/ipc/msg.c (revision 82e6fdd6)
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 
42 #include <asm/current.h>
43 #include <linux/uaccess.h>
44 #include "util.h"
45 
46 /* one msg_receiver structure for each sleeping receiver */
47 struct msg_receiver {
48 	struct list_head	r_list;
49 	struct task_struct	*r_tsk;
50 
51 	int			r_mode;
52 	long			r_msgtype;
53 	long			r_maxsize;
54 
55 	struct msg_msg		*r_msg;
56 };
57 
58 /* one msg_sender for each sleeping sender */
59 struct msg_sender {
60 	struct list_head	list;
61 	struct task_struct	*tsk;
62 	size_t                  msgsz;
63 };
64 
65 #define SEARCH_ANY		1
66 #define SEARCH_EQUAL		2
67 #define SEARCH_NOTEQUAL		3
68 #define SEARCH_LESSEQUAL	4
69 #define SEARCH_NUMBER		5
70 
71 #define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
72 
73 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
74 {
75 	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
76 
77 	if (IS_ERR(ipcp))
78 		return ERR_CAST(ipcp);
79 
80 	return container_of(ipcp, struct msg_queue, q_perm);
81 }
82 
83 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
84 							int id)
85 {
86 	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
87 
88 	if (IS_ERR(ipcp))
89 		return ERR_CAST(ipcp);
90 
91 	return container_of(ipcp, struct msg_queue, q_perm);
92 }
93 
94 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
95 {
96 	ipc_rmid(&msg_ids(ns), &s->q_perm);
97 }
98 
99 static void msg_rcu_free(struct rcu_head *head)
100 {
101 	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
102 	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
103 
104 	security_msg_queue_free(msq);
105 	kvfree(msq);
106 }
107 
108 /**
109  * newque - Create a new msg queue
110  * @ns: namespace
111  * @params: ptr to the structure that contains the key and msgflg
112  *
113  * Called with msg_ids.rwsem held (writer)
114  */
115 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
116 {
117 	struct msg_queue *msq;
118 	int retval;
119 	key_t key = params->key;
120 	int msgflg = params->flg;
121 
122 	msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
123 	if (unlikely(!msq))
124 		return -ENOMEM;
125 
126 	msq->q_perm.mode = msgflg & S_IRWXUGO;
127 	msq->q_perm.key = key;
128 
129 	msq->q_perm.security = NULL;
130 	retval = security_msg_queue_alloc(msq);
131 	if (retval) {
132 		kvfree(msq);
133 		return retval;
134 	}
135 
136 	msq->q_stime = msq->q_rtime = 0;
137 	msq->q_ctime = ktime_get_real_seconds();
138 	msq->q_cbytes = msq->q_qnum = 0;
139 	msq->q_qbytes = ns->msg_ctlmnb;
140 	msq->q_lspid = msq->q_lrpid = 0;
141 	INIT_LIST_HEAD(&msq->q_messages);
142 	INIT_LIST_HEAD(&msq->q_receivers);
143 	INIT_LIST_HEAD(&msq->q_senders);
144 
145 	/* ipc_addid() locks msq upon success. */
146 	retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
147 	if (retval < 0) {
148 		call_rcu(&msq->q_perm.rcu, msg_rcu_free);
149 		return retval;
150 	}
151 
152 	ipc_unlock_object(&msq->q_perm);
153 	rcu_read_unlock();
154 
155 	return msq->q_perm.id;
156 }
157 
158 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
159 {
160 	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
161 		1 + msq->q_qnum <= msq->q_qbytes;
162 }
163 
164 static inline void ss_add(struct msg_queue *msq,
165 			  struct msg_sender *mss, size_t msgsz)
166 {
167 	mss->tsk = current;
168 	mss->msgsz = msgsz;
169 	__set_current_state(TASK_INTERRUPTIBLE);
170 	list_add_tail(&mss->list, &msq->q_senders);
171 }
172 
173 static inline void ss_del(struct msg_sender *mss)
174 {
175 	if (mss->list.next)
176 		list_del(&mss->list);
177 }
178 
179 static void ss_wakeup(struct msg_queue *msq,
180 		      struct wake_q_head *wake_q, bool kill)
181 {
182 	struct msg_sender *mss, *t;
183 	struct task_struct *stop_tsk = NULL;
184 	struct list_head *h = &msq->q_senders;
185 
186 	list_for_each_entry_safe(mss, t, h, list) {
187 		if (kill)
188 			mss->list.next = NULL;
189 
190 		/*
191 		 * Stop at the first task we don't wakeup,
192 		 * we've already iterated the original
193 		 * sender queue.
194 		 */
195 		else if (stop_tsk == mss->tsk)
196 			break;
197 		/*
198 		 * We are not in an EIDRM scenario here, therefore
199 		 * verify that we really need to wakeup the task.
200 		 * To maintain current semantics and wakeup order,
201 		 * move the sender to the tail on behalf of the
202 		 * blocked task.
203 		 */
204 		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
205 			if (!stop_tsk)
206 				stop_tsk = mss->tsk;
207 
208 			list_move_tail(&mss->list, &msq->q_senders);
209 			continue;
210 		}
211 
212 		wake_q_add(wake_q, mss->tsk);
213 	}
214 }
215 
216 static void expunge_all(struct msg_queue *msq, int res,
217 			struct wake_q_head *wake_q)
218 {
219 	struct msg_receiver *msr, *t;
220 
221 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
222 		wake_q_add(wake_q, msr->r_tsk);
223 		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
224 	}
225 }
226 
227 /*
228  * freeque() wakes up waiters on the sender and receiver waiting queue,
229  * removes the message queue from message queue ID IDR, and cleans up all the
230  * messages associated with this queue.
231  *
232  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
233  * before freeque() is called. msg_ids.rwsem remains locked on exit.
234  */
235 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
236 {
237 	struct msg_msg *msg, *t;
238 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
239 	DEFINE_WAKE_Q(wake_q);
240 
241 	expunge_all(msq, -EIDRM, &wake_q);
242 	ss_wakeup(msq, &wake_q, true);
243 	msg_rmid(ns, msq);
244 	ipc_unlock_object(&msq->q_perm);
245 	wake_up_q(&wake_q);
246 	rcu_read_unlock();
247 
248 	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
249 		atomic_dec(&ns->msg_hdrs);
250 		free_msg(msg);
251 	}
252 	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
253 	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
254 }
255 
256 /*
257  * Called with msg_ids.rwsem and ipcp locked.
258  */
259 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
260 {
261 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
262 
263 	return security_msg_queue_associate(msq, msgflg);
264 }
265 
266 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
267 {
268 	struct ipc_namespace *ns;
269 	static const struct ipc_ops msg_ops = {
270 		.getnew = newque,
271 		.associate = msg_security,
272 	};
273 	struct ipc_params msg_params;
274 
275 	ns = current->nsproxy->ipc_ns;
276 
277 	msg_params.key = key;
278 	msg_params.flg = msgflg;
279 
280 	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
281 }
282 
283 static inline unsigned long
284 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
285 {
286 	switch (version) {
287 	case IPC_64:
288 		return copy_to_user(buf, in, sizeof(*in));
289 	case IPC_OLD:
290 	{
291 		struct msqid_ds out;
292 
293 		memset(&out, 0, sizeof(out));
294 
295 		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
296 
297 		out.msg_stime		= in->msg_stime;
298 		out.msg_rtime		= in->msg_rtime;
299 		out.msg_ctime		= in->msg_ctime;
300 
301 		if (in->msg_cbytes > USHRT_MAX)
302 			out.msg_cbytes	= USHRT_MAX;
303 		else
304 			out.msg_cbytes	= in->msg_cbytes;
305 		out.msg_lcbytes		= in->msg_cbytes;
306 
307 		if (in->msg_qnum > USHRT_MAX)
308 			out.msg_qnum	= USHRT_MAX;
309 		else
310 			out.msg_qnum	= in->msg_qnum;
311 
312 		if (in->msg_qbytes > USHRT_MAX)
313 			out.msg_qbytes	= USHRT_MAX;
314 		else
315 			out.msg_qbytes	= in->msg_qbytes;
316 		out.msg_lqbytes		= in->msg_qbytes;
317 
318 		out.msg_lspid		= in->msg_lspid;
319 		out.msg_lrpid		= in->msg_lrpid;
320 
321 		return copy_to_user(buf, &out, sizeof(out));
322 	}
323 	default:
324 		return -EINVAL;
325 	}
326 }
327 
328 static inline unsigned long
329 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
330 {
331 	switch (version) {
332 	case IPC_64:
333 		if (copy_from_user(out, buf, sizeof(*out)))
334 			return -EFAULT;
335 		return 0;
336 	case IPC_OLD:
337 	{
338 		struct msqid_ds tbuf_old;
339 
340 		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
341 			return -EFAULT;
342 
343 		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
344 		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
345 		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
346 
347 		if (tbuf_old.msg_qbytes == 0)
348 			out->msg_qbytes	= tbuf_old.msg_lqbytes;
349 		else
350 			out->msg_qbytes	= tbuf_old.msg_qbytes;
351 
352 		return 0;
353 	}
354 	default:
355 		return -EINVAL;
356 	}
357 }
358 
359 /*
360  * This function handles some msgctl commands which require the rwsem
361  * to be held in write mode.
362  * NOTE: no locks must be held, the rwsem is taken inside this function.
363  */
364 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
365 			struct msqid64_ds *msqid64)
366 {
367 	struct kern_ipc_perm *ipcp;
368 	struct msg_queue *msq;
369 	int err;
370 
371 	down_write(&msg_ids(ns).rwsem);
372 	rcu_read_lock();
373 
374 	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
375 				      &msqid64->msg_perm, msqid64->msg_qbytes);
376 	if (IS_ERR(ipcp)) {
377 		err = PTR_ERR(ipcp);
378 		goto out_unlock1;
379 	}
380 
381 	msq = container_of(ipcp, struct msg_queue, q_perm);
382 
383 	err = security_msg_queue_msgctl(msq, cmd);
384 	if (err)
385 		goto out_unlock1;
386 
387 	switch (cmd) {
388 	case IPC_RMID:
389 		ipc_lock_object(&msq->q_perm);
390 		/* freeque unlocks the ipc object and rcu */
391 		freeque(ns, ipcp);
392 		goto out_up;
393 	case IPC_SET:
394 	{
395 		DEFINE_WAKE_Q(wake_q);
396 
397 		if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
398 		    !capable(CAP_SYS_RESOURCE)) {
399 			err = -EPERM;
400 			goto out_unlock1;
401 		}
402 
403 		ipc_lock_object(&msq->q_perm);
404 		err = ipc_update_perm(&msqid64->msg_perm, ipcp);
405 		if (err)
406 			goto out_unlock0;
407 
408 		msq->q_qbytes = msqid64->msg_qbytes;
409 
410 		msq->q_ctime = ktime_get_real_seconds();
411 		/*
412 		 * Sleeping receivers might be excluded by
413 		 * stricter permissions.
414 		 */
415 		expunge_all(msq, -EAGAIN, &wake_q);
416 		/*
417 		 * Sleeping senders might be able to send
418 		 * due to a larger queue size.
419 		 */
420 		ss_wakeup(msq, &wake_q, false);
421 		ipc_unlock_object(&msq->q_perm);
422 		wake_up_q(&wake_q);
423 
424 		goto out_unlock1;
425 	}
426 	default:
427 		err = -EINVAL;
428 		goto out_unlock1;
429 	}
430 
431 out_unlock0:
432 	ipc_unlock_object(&msq->q_perm);
433 out_unlock1:
434 	rcu_read_unlock();
435 out_up:
436 	up_write(&msg_ids(ns).rwsem);
437 	return err;
438 }
439 
440 static int msgctl_info(struct ipc_namespace *ns, int msqid,
441 			 int cmd, struct msginfo *msginfo)
442 {
443 	int err;
444 	int max_id;
445 
446 	/*
447 	 * We must not return kernel stack data.
448 	 * due to padding, it's not enough
449 	 * to set all member fields.
450 	 */
451 	err = security_msg_queue_msgctl(NULL, cmd);
452 	if (err)
453 		return err;
454 
455 	memset(msginfo, 0, sizeof(*msginfo));
456 	msginfo->msgmni = ns->msg_ctlmni;
457 	msginfo->msgmax = ns->msg_ctlmax;
458 	msginfo->msgmnb = ns->msg_ctlmnb;
459 	msginfo->msgssz = MSGSSZ;
460 	msginfo->msgseg = MSGSEG;
461 	down_read(&msg_ids(ns).rwsem);
462 	if (cmd == MSG_INFO) {
463 		msginfo->msgpool = msg_ids(ns).in_use;
464 		msginfo->msgmap = atomic_read(&ns->msg_hdrs);
465 		msginfo->msgtql = atomic_read(&ns->msg_bytes);
466 	} else {
467 		msginfo->msgmap = MSGMAP;
468 		msginfo->msgpool = MSGPOOL;
469 		msginfo->msgtql = MSGTQL;
470 	}
471 	max_id = ipc_get_maxid(&msg_ids(ns));
472 	up_read(&msg_ids(ns).rwsem);
473 	return (max_id < 0) ? 0 : max_id;
474 }
475 
476 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
477 			 int cmd, struct msqid64_ds *p)
478 {
479 	struct msg_queue *msq;
480 	int id = 0;
481 	int err;
482 
483 	memset(p, 0, sizeof(*p));
484 
485 	rcu_read_lock();
486 	if (cmd == MSG_STAT) {
487 		msq = msq_obtain_object(ns, msqid);
488 		if (IS_ERR(msq)) {
489 			err = PTR_ERR(msq);
490 			goto out_unlock;
491 		}
492 		id = msq->q_perm.id;
493 	} else {
494 		msq = msq_obtain_object_check(ns, msqid);
495 		if (IS_ERR(msq)) {
496 			err = PTR_ERR(msq);
497 			goto out_unlock;
498 		}
499 	}
500 
501 	err = -EACCES;
502 	if (ipcperms(ns, &msq->q_perm, S_IRUGO))
503 		goto out_unlock;
504 
505 	err = security_msg_queue_msgctl(msq, cmd);
506 	if (err)
507 		goto out_unlock;
508 
509 	ipc_lock_object(&msq->q_perm);
510 
511 	if (!ipc_valid_object(&msq->q_perm)) {
512 		ipc_unlock_object(&msq->q_perm);
513 		err = -EIDRM;
514 		goto out_unlock;
515 	}
516 
517 	kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
518 	p->msg_stime  = msq->q_stime;
519 	p->msg_rtime  = msq->q_rtime;
520 	p->msg_ctime  = msq->q_ctime;
521 	p->msg_cbytes = msq->q_cbytes;
522 	p->msg_qnum   = msq->q_qnum;
523 	p->msg_qbytes = msq->q_qbytes;
524 	p->msg_lspid  = msq->q_lspid;
525 	p->msg_lrpid  = msq->q_lrpid;
526 
527 	ipc_unlock_object(&msq->q_perm);
528 	rcu_read_unlock();
529 	return id;
530 
531 out_unlock:
532 	rcu_read_unlock();
533 	return err;
534 }
535 
536 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
537 {
538 	int version;
539 	struct ipc_namespace *ns;
540 	struct msqid64_ds msqid64;
541 	int err;
542 
543 	if (msqid < 0 || cmd < 0)
544 		return -EINVAL;
545 
546 	version = ipc_parse_version(&cmd);
547 	ns = current->nsproxy->ipc_ns;
548 
549 	switch (cmd) {
550 	case IPC_INFO:
551 	case MSG_INFO: {
552 		struct msginfo msginfo;
553 		err = msgctl_info(ns, msqid, cmd, &msginfo);
554 		if (err < 0)
555 			return err;
556 		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
557 			err = -EFAULT;
558 		return err;
559 	}
560 	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
561 	case IPC_STAT:
562 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
563 		if (err < 0)
564 			return err;
565 		if (copy_msqid_to_user(buf, &msqid64, version))
566 			err = -EFAULT;
567 		return err;
568 	case IPC_SET:
569 		if (copy_msqid_from_user(&msqid64, buf, version))
570 			return -EFAULT;
571 		/* fallthru */
572 	case IPC_RMID:
573 		return msgctl_down(ns, msqid, cmd, &msqid64);
574 	default:
575 		return  -EINVAL;
576 	}
577 }
578 
579 #ifdef CONFIG_COMPAT
580 
581 struct compat_msqid_ds {
582 	struct compat_ipc_perm msg_perm;
583 	compat_uptr_t msg_first;
584 	compat_uptr_t msg_last;
585 	compat_time_t msg_stime;
586 	compat_time_t msg_rtime;
587 	compat_time_t msg_ctime;
588 	compat_ulong_t msg_lcbytes;
589 	compat_ulong_t msg_lqbytes;
590 	unsigned short msg_cbytes;
591 	unsigned short msg_qnum;
592 	unsigned short msg_qbytes;
593 	compat_ipc_pid_t msg_lspid;
594 	compat_ipc_pid_t msg_lrpid;
595 };
596 
597 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
598 					int version)
599 {
600 	memset(out, 0, sizeof(*out));
601 	if (version == IPC_64) {
602 		struct compat_msqid64_ds __user *p = buf;
603 		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
604 			return -EFAULT;
605 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
606 			return -EFAULT;
607 	} else {
608 		struct compat_msqid_ds __user *p = buf;
609 		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
610 			return -EFAULT;
611 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
612 			return -EFAULT;
613 	}
614 	return 0;
615 }
616 
617 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
618 					int version)
619 {
620 	if (version == IPC_64) {
621 		struct compat_msqid64_ds v;
622 		memset(&v, 0, sizeof(v));
623 		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
624 		v.msg_stime = in->msg_stime;
625 		v.msg_rtime = in->msg_rtime;
626 		v.msg_ctime = in->msg_ctime;
627 		v.msg_cbytes = in->msg_cbytes;
628 		v.msg_qnum = in->msg_qnum;
629 		v.msg_qbytes = in->msg_qbytes;
630 		v.msg_lspid = in->msg_lspid;
631 		v.msg_lrpid = in->msg_lrpid;
632 		return copy_to_user(buf, &v, sizeof(v));
633 	} else {
634 		struct compat_msqid_ds v;
635 		memset(&v, 0, sizeof(v));
636 		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
637 		v.msg_stime = in->msg_stime;
638 		v.msg_rtime = in->msg_rtime;
639 		v.msg_ctime = in->msg_ctime;
640 		v.msg_cbytes = in->msg_cbytes;
641 		v.msg_qnum = in->msg_qnum;
642 		v.msg_qbytes = in->msg_qbytes;
643 		v.msg_lspid = in->msg_lspid;
644 		v.msg_lrpid = in->msg_lrpid;
645 		return copy_to_user(buf, &v, sizeof(v));
646 	}
647 }
648 
649 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
650 {
651 	struct ipc_namespace *ns;
652 	int err;
653 	struct msqid64_ds msqid64;
654 	int version = compat_ipc_parse_version(&cmd);
655 
656 	ns = current->nsproxy->ipc_ns;
657 
658 	if (msqid < 0 || cmd < 0)
659 		return -EINVAL;
660 
661 	switch (cmd & (~IPC_64)) {
662 	case IPC_INFO:
663 	case MSG_INFO: {
664 		struct msginfo msginfo;
665 		err = msgctl_info(ns, msqid, cmd, &msginfo);
666 		if (err < 0)
667 			return err;
668 		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
669 			err = -EFAULT;
670 		return err;
671 	}
672 	case IPC_STAT:
673 	case MSG_STAT:
674 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
675 		if (err < 0)
676 			return err;
677 		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
678 			err = -EFAULT;
679 		return err;
680 	case IPC_SET:
681 		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
682 			return -EFAULT;
683 		/* fallthru */
684 	case IPC_RMID:
685 		return msgctl_down(ns, msqid, cmd, &msqid64);
686 	default:
687 		return -EINVAL;
688 	}
689 }
690 #endif
691 
692 static int testmsg(struct msg_msg *msg, long type, int mode)
693 {
694 	switch (mode) {
695 	case SEARCH_ANY:
696 	case SEARCH_NUMBER:
697 		return 1;
698 	case SEARCH_LESSEQUAL:
699 		if (msg->m_type <= type)
700 			return 1;
701 		break;
702 	case SEARCH_EQUAL:
703 		if (msg->m_type == type)
704 			return 1;
705 		break;
706 	case SEARCH_NOTEQUAL:
707 		if (msg->m_type != type)
708 			return 1;
709 		break;
710 	}
711 	return 0;
712 }
713 
714 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
715 				 struct wake_q_head *wake_q)
716 {
717 	struct msg_receiver *msr, *t;
718 
719 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
720 		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
721 		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
722 					       msr->r_msgtype, msr->r_mode)) {
723 
724 			list_del(&msr->r_list);
725 			if (msr->r_maxsize < msg->m_ts) {
726 				wake_q_add(wake_q, msr->r_tsk);
727 				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
728 			} else {
729 				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
730 				msq->q_rtime = get_seconds();
731 
732 				wake_q_add(wake_q, msr->r_tsk);
733 				WRITE_ONCE(msr->r_msg, msg);
734 				return 1;
735 			}
736 		}
737 	}
738 
739 	return 0;
740 }
741 
742 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
743 		size_t msgsz, int msgflg)
744 {
745 	struct msg_queue *msq;
746 	struct msg_msg *msg;
747 	int err;
748 	struct ipc_namespace *ns;
749 	DEFINE_WAKE_Q(wake_q);
750 
751 	ns = current->nsproxy->ipc_ns;
752 
753 	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
754 		return -EINVAL;
755 	if (mtype < 1)
756 		return -EINVAL;
757 
758 	msg = load_msg(mtext, msgsz);
759 	if (IS_ERR(msg))
760 		return PTR_ERR(msg);
761 
762 	msg->m_type = mtype;
763 	msg->m_ts = msgsz;
764 
765 	rcu_read_lock();
766 	msq = msq_obtain_object_check(ns, msqid);
767 	if (IS_ERR(msq)) {
768 		err = PTR_ERR(msq);
769 		goto out_unlock1;
770 	}
771 
772 	ipc_lock_object(&msq->q_perm);
773 
774 	for (;;) {
775 		struct msg_sender s;
776 
777 		err = -EACCES;
778 		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
779 			goto out_unlock0;
780 
781 		/* raced with RMID? */
782 		if (!ipc_valid_object(&msq->q_perm)) {
783 			err = -EIDRM;
784 			goto out_unlock0;
785 		}
786 
787 		err = security_msg_queue_msgsnd(msq, msg, msgflg);
788 		if (err)
789 			goto out_unlock0;
790 
791 		if (msg_fits_inqueue(msq, msgsz))
792 			break;
793 
794 		/* queue full, wait: */
795 		if (msgflg & IPC_NOWAIT) {
796 			err = -EAGAIN;
797 			goto out_unlock0;
798 		}
799 
800 		/* enqueue the sender and prepare to block */
801 		ss_add(msq, &s, msgsz);
802 
803 		if (!ipc_rcu_getref(&msq->q_perm)) {
804 			err = -EIDRM;
805 			goto out_unlock0;
806 		}
807 
808 		ipc_unlock_object(&msq->q_perm);
809 		rcu_read_unlock();
810 		schedule();
811 
812 		rcu_read_lock();
813 		ipc_lock_object(&msq->q_perm);
814 
815 		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
816 		/* raced with RMID? */
817 		if (!ipc_valid_object(&msq->q_perm)) {
818 			err = -EIDRM;
819 			goto out_unlock0;
820 		}
821 		ss_del(&s);
822 
823 		if (signal_pending(current)) {
824 			err = -ERESTARTNOHAND;
825 			goto out_unlock0;
826 		}
827 
828 	}
829 
830 	msq->q_lspid = task_tgid_vnr(current);
831 	msq->q_stime = get_seconds();
832 
833 	if (!pipelined_send(msq, msg, &wake_q)) {
834 		/* no one is waiting for this message, enqueue it */
835 		list_add_tail(&msg->m_list, &msq->q_messages);
836 		msq->q_cbytes += msgsz;
837 		msq->q_qnum++;
838 		atomic_add(msgsz, &ns->msg_bytes);
839 		atomic_inc(&ns->msg_hdrs);
840 	}
841 
842 	err = 0;
843 	msg = NULL;
844 
845 out_unlock0:
846 	ipc_unlock_object(&msq->q_perm);
847 	wake_up_q(&wake_q);
848 out_unlock1:
849 	rcu_read_unlock();
850 	if (msg != NULL)
851 		free_msg(msg);
852 	return err;
853 }
854 
855 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
856 		int, msgflg)
857 {
858 	long mtype;
859 
860 	if (get_user(mtype, &msgp->mtype))
861 		return -EFAULT;
862 	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
863 }
864 
865 #ifdef CONFIG_COMPAT
866 
867 struct compat_msgbuf {
868 	compat_long_t mtype;
869 	char mtext[1];
870 };
871 
872 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
873 		       compat_ssize_t, msgsz, int, msgflg)
874 {
875 	struct compat_msgbuf __user *up = compat_ptr(msgp);
876 	compat_long_t mtype;
877 
878 	if (get_user(mtype, &up->mtype))
879 		return -EFAULT;
880 	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
881 }
882 #endif
883 
884 static inline int convert_mode(long *msgtyp, int msgflg)
885 {
886 	if (msgflg & MSG_COPY)
887 		return SEARCH_NUMBER;
888 	/*
889 	 *  find message of correct type.
890 	 *  msgtyp = 0 => get first.
891 	 *  msgtyp > 0 => get first message of matching type.
892 	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
893 	 */
894 	if (*msgtyp == 0)
895 		return SEARCH_ANY;
896 	if (*msgtyp < 0) {
897 		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
898 			*msgtyp = LONG_MAX;
899 		else
900 			*msgtyp = -*msgtyp;
901 		return SEARCH_LESSEQUAL;
902 	}
903 	if (msgflg & MSG_EXCEPT)
904 		return SEARCH_NOTEQUAL;
905 	return SEARCH_EQUAL;
906 }
907 
908 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
909 {
910 	struct msgbuf __user *msgp = dest;
911 	size_t msgsz;
912 
913 	if (put_user(msg->m_type, &msgp->mtype))
914 		return -EFAULT;
915 
916 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
917 	if (store_msg(msgp->mtext, msg, msgsz))
918 		return -EFAULT;
919 	return msgsz;
920 }
921 
922 #ifdef CONFIG_CHECKPOINT_RESTORE
923 /*
924  * This function creates new kernel message structure, large enough to store
925  * bufsz message bytes.
926  */
927 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
928 {
929 	struct msg_msg *copy;
930 
931 	/*
932 	 * Create dummy message to copy real message to.
933 	 */
934 	copy = load_msg(buf, bufsz);
935 	if (!IS_ERR(copy))
936 		copy->m_ts = bufsz;
937 	return copy;
938 }
939 
940 static inline void free_copy(struct msg_msg *copy)
941 {
942 	if (copy)
943 		free_msg(copy);
944 }
945 #else
946 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
947 {
948 	return ERR_PTR(-ENOSYS);
949 }
950 
951 static inline void free_copy(struct msg_msg *copy)
952 {
953 }
954 #endif
955 
956 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
957 {
958 	struct msg_msg *msg, *found = NULL;
959 	long count = 0;
960 
961 	list_for_each_entry(msg, &msq->q_messages, m_list) {
962 		if (testmsg(msg, *msgtyp, mode) &&
963 		    !security_msg_queue_msgrcv(msq, msg, current,
964 					       *msgtyp, mode)) {
965 			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
966 				*msgtyp = msg->m_type - 1;
967 				found = msg;
968 			} else if (mode == SEARCH_NUMBER) {
969 				if (*msgtyp == count)
970 					return msg;
971 			} else
972 				return msg;
973 			count++;
974 		}
975 	}
976 
977 	return found ?: ERR_PTR(-EAGAIN);
978 }
979 
980 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
981 	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
982 {
983 	int mode;
984 	struct msg_queue *msq;
985 	struct ipc_namespace *ns;
986 	struct msg_msg *msg, *copy = NULL;
987 	DEFINE_WAKE_Q(wake_q);
988 
989 	ns = current->nsproxy->ipc_ns;
990 
991 	if (msqid < 0 || (long) bufsz < 0)
992 		return -EINVAL;
993 
994 	if (msgflg & MSG_COPY) {
995 		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
996 			return -EINVAL;
997 		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
998 		if (IS_ERR(copy))
999 			return PTR_ERR(copy);
1000 	}
1001 	mode = convert_mode(&msgtyp, msgflg);
1002 
1003 	rcu_read_lock();
1004 	msq = msq_obtain_object_check(ns, msqid);
1005 	if (IS_ERR(msq)) {
1006 		rcu_read_unlock();
1007 		free_copy(copy);
1008 		return PTR_ERR(msq);
1009 	}
1010 
1011 	for (;;) {
1012 		struct msg_receiver msr_d;
1013 
1014 		msg = ERR_PTR(-EACCES);
1015 		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1016 			goto out_unlock1;
1017 
1018 		ipc_lock_object(&msq->q_perm);
1019 
1020 		/* raced with RMID? */
1021 		if (!ipc_valid_object(&msq->q_perm)) {
1022 			msg = ERR_PTR(-EIDRM);
1023 			goto out_unlock0;
1024 		}
1025 
1026 		msg = find_msg(msq, &msgtyp, mode);
1027 		if (!IS_ERR(msg)) {
1028 			/*
1029 			 * Found a suitable message.
1030 			 * Unlink it from the queue.
1031 			 */
1032 			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1033 				msg = ERR_PTR(-E2BIG);
1034 				goto out_unlock0;
1035 			}
1036 			/*
1037 			 * If we are copying, then do not unlink message and do
1038 			 * not update queue parameters.
1039 			 */
1040 			if (msgflg & MSG_COPY) {
1041 				msg = copy_msg(msg, copy);
1042 				goto out_unlock0;
1043 			}
1044 
1045 			list_del(&msg->m_list);
1046 			msq->q_qnum--;
1047 			msq->q_rtime = get_seconds();
1048 			msq->q_lrpid = task_tgid_vnr(current);
1049 			msq->q_cbytes -= msg->m_ts;
1050 			atomic_sub(msg->m_ts, &ns->msg_bytes);
1051 			atomic_dec(&ns->msg_hdrs);
1052 			ss_wakeup(msq, &wake_q, false);
1053 
1054 			goto out_unlock0;
1055 		}
1056 
1057 		/* No message waiting. Wait for a message */
1058 		if (msgflg & IPC_NOWAIT) {
1059 			msg = ERR_PTR(-ENOMSG);
1060 			goto out_unlock0;
1061 		}
1062 
1063 		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1064 		msr_d.r_tsk = current;
1065 		msr_d.r_msgtype = msgtyp;
1066 		msr_d.r_mode = mode;
1067 		if (msgflg & MSG_NOERROR)
1068 			msr_d.r_maxsize = INT_MAX;
1069 		else
1070 			msr_d.r_maxsize = bufsz;
1071 		msr_d.r_msg = ERR_PTR(-EAGAIN);
1072 		__set_current_state(TASK_INTERRUPTIBLE);
1073 
1074 		ipc_unlock_object(&msq->q_perm);
1075 		rcu_read_unlock();
1076 		schedule();
1077 
1078 		/*
1079 		 * Lockless receive, part 1:
1080 		 * We don't hold a reference to the queue and getting a
1081 		 * reference would defeat the idea of a lockless operation,
1082 		 * thus the code relies on rcu to guarantee the existence of
1083 		 * msq:
1084 		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1085 		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1086 		 */
1087 		rcu_read_lock();
1088 
1089 		/*
1090 		 * Lockless receive, part 2:
1091 		 * The work in pipelined_send() and expunge_all():
1092 		 * - Set pointer to message
1093 		 * - Queue the receiver task for later wakeup
1094 		 * - Wake up the process after the lock is dropped.
1095 		 *
1096 		 * Should the process wake up before this wakeup (due to a
1097 		 * signal) it will either see the message and continue ...
1098 		 */
1099 		msg = READ_ONCE(msr_d.r_msg);
1100 		if (msg != ERR_PTR(-EAGAIN))
1101 			goto out_unlock1;
1102 
1103 		 /*
1104 		  * ... or see -EAGAIN, acquire the lock to check the message
1105 		  * again.
1106 		  */
1107 		ipc_lock_object(&msq->q_perm);
1108 
1109 		msg = msr_d.r_msg;
1110 		if (msg != ERR_PTR(-EAGAIN))
1111 			goto out_unlock0;
1112 
1113 		list_del(&msr_d.r_list);
1114 		if (signal_pending(current)) {
1115 			msg = ERR_PTR(-ERESTARTNOHAND);
1116 			goto out_unlock0;
1117 		}
1118 
1119 		ipc_unlock_object(&msq->q_perm);
1120 	}
1121 
1122 out_unlock0:
1123 	ipc_unlock_object(&msq->q_perm);
1124 	wake_up_q(&wake_q);
1125 out_unlock1:
1126 	rcu_read_unlock();
1127 	if (IS_ERR(msg)) {
1128 		free_copy(copy);
1129 		return PTR_ERR(msg);
1130 	}
1131 
1132 	bufsz = msg_handler(buf, msg, bufsz);
1133 	free_msg(msg);
1134 
1135 	return bufsz;
1136 }
1137 
1138 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1139 		long, msgtyp, int, msgflg)
1140 {
1141 	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1142 }
1143 
1144 #ifdef CONFIG_COMPAT
1145 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1146 {
1147 	struct compat_msgbuf __user *msgp = dest;
1148 	size_t msgsz;
1149 
1150 	if (put_user(msg->m_type, &msgp->mtype))
1151 		return -EFAULT;
1152 
1153 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1154 	if (store_msg(msgp->mtext, msg, msgsz))
1155 		return -EFAULT;
1156 	return msgsz;
1157 }
1158 
1159 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1160 		       compat_ssize_t, msgsz, compat_long_t, msgtyp, int, msgflg)
1161 {
1162 	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1163 			 msgflg, compat_do_msg_fill);
1164 }
1165 #endif
1166 
1167 int msg_init_ns(struct ipc_namespace *ns)
1168 {
1169 	ns->msg_ctlmax = MSGMAX;
1170 	ns->msg_ctlmnb = MSGMNB;
1171 	ns->msg_ctlmni = MSGMNI;
1172 
1173 	atomic_set(&ns->msg_bytes, 0);
1174 	atomic_set(&ns->msg_hdrs, 0);
1175 	return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1176 }
1177 
1178 #ifdef CONFIG_IPC_NS
1179 void msg_exit_ns(struct ipc_namespace *ns)
1180 {
1181 	free_ipcs(ns, &msg_ids(ns), freeque);
1182 	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1183 	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1184 }
1185 #endif
1186 
1187 #ifdef CONFIG_PROC_FS
1188 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1189 {
1190 	struct user_namespace *user_ns = seq_user_ns(s);
1191 	struct kern_ipc_perm *ipcp = it;
1192 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1193 
1194 	seq_printf(s,
1195 		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1196 		   msq->q_perm.key,
1197 		   msq->q_perm.id,
1198 		   msq->q_perm.mode,
1199 		   msq->q_cbytes,
1200 		   msq->q_qnum,
1201 		   msq->q_lspid,
1202 		   msq->q_lrpid,
1203 		   from_kuid_munged(user_ns, msq->q_perm.uid),
1204 		   from_kgid_munged(user_ns, msq->q_perm.gid),
1205 		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1206 		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1207 		   msq->q_stime,
1208 		   msq->q_rtime,
1209 		   msq->q_ctime);
1210 
1211 	return 0;
1212 }
1213 #endif
1214 
1215 int __init msg_init(void)
1216 {
1217 	const int err = msg_init_ns(&init_ipc_ns);
1218 
1219 	ipc_init_proc_interface("sysvipc/msg",
1220 				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1221 				IPC_MSG_IDS, sysvipc_msg_proc_show);
1222 	return err;
1223 }
1224