xref: /linux/ipc/msg.c (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
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 msq_queue structure for each present queue on the system */
47 struct msg_queue {
48 	struct kern_ipc_perm q_perm;
49 	time64_t q_stime;		/* last msgsnd time */
50 	time64_t q_rtime;		/* last msgrcv time */
51 	time64_t q_ctime;		/* last change time */
52 	unsigned long q_cbytes;		/* current number of bytes on queue */
53 	unsigned long q_qnum;		/* number of messages in queue */
54 	unsigned long q_qbytes;		/* max number of bytes on queue */
55 	struct pid *q_lspid;		/* pid of last msgsnd */
56 	struct pid *q_lrpid;		/* last receive pid */
57 
58 	struct list_head q_messages;
59 	struct list_head q_receivers;
60 	struct list_head q_senders;
61 } __randomize_layout;
62 
63 /* one msg_receiver structure for each sleeping receiver */
64 struct msg_receiver {
65 	struct list_head	r_list;
66 	struct task_struct	*r_tsk;
67 
68 	int			r_mode;
69 	long			r_msgtype;
70 	long			r_maxsize;
71 
72 	struct msg_msg		*r_msg;
73 };
74 
75 /* one msg_sender for each sleeping sender */
76 struct msg_sender {
77 	struct list_head	list;
78 	struct task_struct	*tsk;
79 	size_t                  msgsz;
80 };
81 
82 #define SEARCH_ANY		1
83 #define SEARCH_EQUAL		2
84 #define SEARCH_NOTEQUAL		3
85 #define SEARCH_LESSEQUAL	4
86 #define SEARCH_NUMBER		5
87 
88 #define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
89 
90 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
91 {
92 	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
93 
94 	if (IS_ERR(ipcp))
95 		return ERR_CAST(ipcp);
96 
97 	return container_of(ipcp, struct msg_queue, q_perm);
98 }
99 
100 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
101 							int id)
102 {
103 	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&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 
111 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
112 {
113 	ipc_rmid(&msg_ids(ns), &s->q_perm);
114 }
115 
116 static void msg_rcu_free(struct rcu_head *head)
117 {
118 	struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
119 	struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
120 
121 	security_msg_queue_free(&msq->q_perm);
122 	kvfree(msq);
123 }
124 
125 /**
126  * newque - Create a new msg queue
127  * @ns: namespace
128  * @params: ptr to the structure that contains the key and msgflg
129  *
130  * Called with msg_ids.rwsem held (writer)
131  */
132 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
133 {
134 	struct msg_queue *msq;
135 	int retval;
136 	key_t key = params->key;
137 	int msgflg = params->flg;
138 
139 	msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
140 	if (unlikely(!msq))
141 		return -ENOMEM;
142 
143 	msq->q_perm.mode = msgflg & S_IRWXUGO;
144 	msq->q_perm.key = key;
145 
146 	msq->q_perm.security = NULL;
147 	retval = security_msg_queue_alloc(&msq->q_perm);
148 	if (retval) {
149 		kvfree(msq);
150 		return retval;
151 	}
152 
153 	msq->q_stime = msq->q_rtime = 0;
154 	msq->q_ctime = ktime_get_real_seconds();
155 	msq->q_cbytes = msq->q_qnum = 0;
156 	msq->q_qbytes = ns->msg_ctlmnb;
157 	msq->q_lspid = msq->q_lrpid = NULL;
158 	INIT_LIST_HEAD(&msq->q_messages);
159 	INIT_LIST_HEAD(&msq->q_receivers);
160 	INIT_LIST_HEAD(&msq->q_senders);
161 
162 	/* ipc_addid() locks msq upon success. */
163 	retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
164 	if (retval < 0) {
165 		call_rcu(&msq->q_perm.rcu, msg_rcu_free);
166 		return retval;
167 	}
168 
169 	ipc_unlock_object(&msq->q_perm);
170 	rcu_read_unlock();
171 
172 	return msq->q_perm.id;
173 }
174 
175 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
176 {
177 	return msgsz + msq->q_cbytes <= msq->q_qbytes &&
178 		1 + msq->q_qnum <= msq->q_qbytes;
179 }
180 
181 static inline void ss_add(struct msg_queue *msq,
182 			  struct msg_sender *mss, size_t msgsz)
183 {
184 	mss->tsk = current;
185 	mss->msgsz = msgsz;
186 	__set_current_state(TASK_INTERRUPTIBLE);
187 	list_add_tail(&mss->list, &msq->q_senders);
188 }
189 
190 static inline void ss_del(struct msg_sender *mss)
191 {
192 	if (mss->list.next)
193 		list_del(&mss->list);
194 }
195 
196 static void ss_wakeup(struct msg_queue *msq,
197 		      struct wake_q_head *wake_q, bool kill)
198 {
199 	struct msg_sender *mss, *t;
200 	struct task_struct *stop_tsk = NULL;
201 	struct list_head *h = &msq->q_senders;
202 
203 	list_for_each_entry_safe(mss, t, h, list) {
204 		if (kill)
205 			mss->list.next = NULL;
206 
207 		/*
208 		 * Stop at the first task we don't wakeup,
209 		 * we've already iterated the original
210 		 * sender queue.
211 		 */
212 		else if (stop_tsk == mss->tsk)
213 			break;
214 		/*
215 		 * We are not in an EIDRM scenario here, therefore
216 		 * verify that we really need to wakeup the task.
217 		 * To maintain current semantics and wakeup order,
218 		 * move the sender to the tail on behalf of the
219 		 * blocked task.
220 		 */
221 		else if (!msg_fits_inqueue(msq, mss->msgsz)) {
222 			if (!stop_tsk)
223 				stop_tsk = mss->tsk;
224 
225 			list_move_tail(&mss->list, &msq->q_senders);
226 			continue;
227 		}
228 
229 		wake_q_add(wake_q, mss->tsk);
230 	}
231 }
232 
233 static void expunge_all(struct msg_queue *msq, int res,
234 			struct wake_q_head *wake_q)
235 {
236 	struct msg_receiver *msr, *t;
237 
238 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
239 		wake_q_add(wake_q, msr->r_tsk);
240 		WRITE_ONCE(msr->r_msg, ERR_PTR(res));
241 	}
242 }
243 
244 /*
245  * freeque() wakes up waiters on the sender and receiver waiting queue,
246  * removes the message queue from message queue ID IDR, and cleans up all the
247  * messages associated with this queue.
248  *
249  * msg_ids.rwsem (writer) and the spinlock for this message queue are held
250  * before freeque() is called. msg_ids.rwsem remains locked on exit.
251  */
252 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
253 {
254 	struct msg_msg *msg, *t;
255 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
256 	DEFINE_WAKE_Q(wake_q);
257 
258 	expunge_all(msq, -EIDRM, &wake_q);
259 	ss_wakeup(msq, &wake_q, true);
260 	msg_rmid(ns, msq);
261 	ipc_unlock_object(&msq->q_perm);
262 	wake_up_q(&wake_q);
263 	rcu_read_unlock();
264 
265 	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
266 		atomic_dec(&ns->msg_hdrs);
267 		free_msg(msg);
268 	}
269 	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
270 	ipc_update_pid(&msq->q_lspid, NULL);
271 	ipc_update_pid(&msq->q_lrpid, NULL);
272 	ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
273 }
274 
275 long ksys_msgget(key_t key, int msgflg)
276 {
277 	struct ipc_namespace *ns;
278 	static const struct ipc_ops msg_ops = {
279 		.getnew = newque,
280 		.associate = security_msg_queue_associate,
281 	};
282 	struct ipc_params msg_params;
283 
284 	ns = current->nsproxy->ipc_ns;
285 
286 	msg_params.key = key;
287 	msg_params.flg = msgflg;
288 
289 	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
290 }
291 
292 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
293 {
294 	return ksys_msgget(key, msgflg);
295 }
296 
297 static inline unsigned long
298 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
299 {
300 	switch (version) {
301 	case IPC_64:
302 		return copy_to_user(buf, in, sizeof(*in));
303 	case IPC_OLD:
304 	{
305 		struct msqid_ds out;
306 
307 		memset(&out, 0, sizeof(out));
308 
309 		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
310 
311 		out.msg_stime		= in->msg_stime;
312 		out.msg_rtime		= in->msg_rtime;
313 		out.msg_ctime		= in->msg_ctime;
314 
315 		if (in->msg_cbytes > USHRT_MAX)
316 			out.msg_cbytes	= USHRT_MAX;
317 		else
318 			out.msg_cbytes	= in->msg_cbytes;
319 		out.msg_lcbytes		= in->msg_cbytes;
320 
321 		if (in->msg_qnum > USHRT_MAX)
322 			out.msg_qnum	= USHRT_MAX;
323 		else
324 			out.msg_qnum	= in->msg_qnum;
325 
326 		if (in->msg_qbytes > USHRT_MAX)
327 			out.msg_qbytes	= USHRT_MAX;
328 		else
329 			out.msg_qbytes	= in->msg_qbytes;
330 		out.msg_lqbytes		= in->msg_qbytes;
331 
332 		out.msg_lspid		= in->msg_lspid;
333 		out.msg_lrpid		= in->msg_lrpid;
334 
335 		return copy_to_user(buf, &out, sizeof(out));
336 	}
337 	default:
338 		return -EINVAL;
339 	}
340 }
341 
342 static inline unsigned long
343 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
344 {
345 	switch (version) {
346 	case IPC_64:
347 		if (copy_from_user(out, buf, sizeof(*out)))
348 			return -EFAULT;
349 		return 0;
350 	case IPC_OLD:
351 	{
352 		struct msqid_ds tbuf_old;
353 
354 		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
355 			return -EFAULT;
356 
357 		out->msg_perm.uid	= tbuf_old.msg_perm.uid;
358 		out->msg_perm.gid	= tbuf_old.msg_perm.gid;
359 		out->msg_perm.mode	= tbuf_old.msg_perm.mode;
360 
361 		if (tbuf_old.msg_qbytes == 0)
362 			out->msg_qbytes	= tbuf_old.msg_lqbytes;
363 		else
364 			out->msg_qbytes	= tbuf_old.msg_qbytes;
365 
366 		return 0;
367 	}
368 	default:
369 		return -EINVAL;
370 	}
371 }
372 
373 /*
374  * This function handles some msgctl commands which require the rwsem
375  * to be held in write mode.
376  * NOTE: no locks must be held, the rwsem is taken inside this function.
377  */
378 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
379 			struct msqid64_ds *msqid64)
380 {
381 	struct kern_ipc_perm *ipcp;
382 	struct msg_queue *msq;
383 	int err;
384 
385 	down_write(&msg_ids(ns).rwsem);
386 	rcu_read_lock();
387 
388 	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
389 				      &msqid64->msg_perm, msqid64->msg_qbytes);
390 	if (IS_ERR(ipcp)) {
391 		err = PTR_ERR(ipcp);
392 		goto out_unlock1;
393 	}
394 
395 	msq = container_of(ipcp, struct msg_queue, q_perm);
396 
397 	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
398 	if (err)
399 		goto out_unlock1;
400 
401 	switch (cmd) {
402 	case IPC_RMID:
403 		ipc_lock_object(&msq->q_perm);
404 		/* freeque unlocks the ipc object and rcu */
405 		freeque(ns, ipcp);
406 		goto out_up;
407 	case IPC_SET:
408 	{
409 		DEFINE_WAKE_Q(wake_q);
410 
411 		if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
412 		    !capable(CAP_SYS_RESOURCE)) {
413 			err = -EPERM;
414 			goto out_unlock1;
415 		}
416 
417 		ipc_lock_object(&msq->q_perm);
418 		err = ipc_update_perm(&msqid64->msg_perm, ipcp);
419 		if (err)
420 			goto out_unlock0;
421 
422 		msq->q_qbytes = msqid64->msg_qbytes;
423 
424 		msq->q_ctime = ktime_get_real_seconds();
425 		/*
426 		 * Sleeping receivers might be excluded by
427 		 * stricter permissions.
428 		 */
429 		expunge_all(msq, -EAGAIN, &wake_q);
430 		/*
431 		 * Sleeping senders might be able to send
432 		 * due to a larger queue size.
433 		 */
434 		ss_wakeup(msq, &wake_q, false);
435 		ipc_unlock_object(&msq->q_perm);
436 		wake_up_q(&wake_q);
437 
438 		goto out_unlock1;
439 	}
440 	default:
441 		err = -EINVAL;
442 		goto out_unlock1;
443 	}
444 
445 out_unlock0:
446 	ipc_unlock_object(&msq->q_perm);
447 out_unlock1:
448 	rcu_read_unlock();
449 out_up:
450 	up_write(&msg_ids(ns).rwsem);
451 	return err;
452 }
453 
454 static int msgctl_info(struct ipc_namespace *ns, int msqid,
455 			 int cmd, struct msginfo *msginfo)
456 {
457 	int err;
458 	int max_id;
459 
460 	/*
461 	 * We must not return kernel stack data.
462 	 * due to padding, it's not enough
463 	 * to set all member fields.
464 	 */
465 	err = security_msg_queue_msgctl(NULL, cmd);
466 	if (err)
467 		return err;
468 
469 	memset(msginfo, 0, sizeof(*msginfo));
470 	msginfo->msgmni = ns->msg_ctlmni;
471 	msginfo->msgmax = ns->msg_ctlmax;
472 	msginfo->msgmnb = ns->msg_ctlmnb;
473 	msginfo->msgssz = MSGSSZ;
474 	msginfo->msgseg = MSGSEG;
475 	down_read(&msg_ids(ns).rwsem);
476 	if (cmd == MSG_INFO) {
477 		msginfo->msgpool = msg_ids(ns).in_use;
478 		msginfo->msgmap = atomic_read(&ns->msg_hdrs);
479 		msginfo->msgtql = atomic_read(&ns->msg_bytes);
480 	} else {
481 		msginfo->msgmap = MSGMAP;
482 		msginfo->msgpool = MSGPOOL;
483 		msginfo->msgtql = MSGTQL;
484 	}
485 	max_id = ipc_get_maxid(&msg_ids(ns));
486 	up_read(&msg_ids(ns).rwsem);
487 	return (max_id < 0) ? 0 : max_id;
488 }
489 
490 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
491 			 int cmd, struct msqid64_ds *p)
492 {
493 	struct msg_queue *msq;
494 	int id = 0;
495 	int err;
496 
497 	memset(p, 0, sizeof(*p));
498 
499 	rcu_read_lock();
500 	if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
501 		msq = msq_obtain_object(ns, msqid);
502 		if (IS_ERR(msq)) {
503 			err = PTR_ERR(msq);
504 			goto out_unlock;
505 		}
506 		id = msq->q_perm.id;
507 	} else { /* IPC_STAT */
508 		msq = msq_obtain_object_check(ns, msqid);
509 		if (IS_ERR(msq)) {
510 			err = PTR_ERR(msq);
511 			goto out_unlock;
512 		}
513 	}
514 
515 	/* see comment for SHM_STAT_ANY */
516 	if (cmd == MSG_STAT_ANY)
517 		audit_ipc_obj(&msq->q_perm);
518 	else {
519 		err = -EACCES;
520 		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
521 			goto out_unlock;
522 	}
523 
524 	err = security_msg_queue_msgctl(&msq->q_perm, cmd);
525 	if (err)
526 		goto out_unlock;
527 
528 	ipc_lock_object(&msq->q_perm);
529 
530 	if (!ipc_valid_object(&msq->q_perm)) {
531 		ipc_unlock_object(&msq->q_perm);
532 		err = -EIDRM;
533 		goto out_unlock;
534 	}
535 
536 	kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
537 	p->msg_stime  = msq->q_stime;
538 	p->msg_rtime  = msq->q_rtime;
539 	p->msg_ctime  = msq->q_ctime;
540 #ifndef CONFIG_64BIT
541 	p->msg_stime_high = msq->q_stime >> 32;
542 	p->msg_rtime_high = msq->q_rtime >> 32;
543 	p->msg_ctime_high = msq->q_ctime >> 32;
544 #endif
545 	p->msg_cbytes = msq->q_cbytes;
546 	p->msg_qnum   = msq->q_qnum;
547 	p->msg_qbytes = msq->q_qbytes;
548 	p->msg_lspid  = pid_vnr(msq->q_lspid);
549 	p->msg_lrpid  = pid_vnr(msq->q_lrpid);
550 
551 	ipc_unlock_object(&msq->q_perm);
552 	rcu_read_unlock();
553 	return id;
554 
555 out_unlock:
556 	rcu_read_unlock();
557 	return err;
558 }
559 
560 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
561 {
562 	int version;
563 	struct ipc_namespace *ns;
564 	struct msqid64_ds msqid64;
565 	int err;
566 
567 	if (msqid < 0 || cmd < 0)
568 		return -EINVAL;
569 
570 	version = ipc_parse_version(&cmd);
571 	ns = current->nsproxy->ipc_ns;
572 
573 	switch (cmd) {
574 	case IPC_INFO:
575 	case MSG_INFO: {
576 		struct msginfo msginfo;
577 		err = msgctl_info(ns, msqid, cmd, &msginfo);
578 		if (err < 0)
579 			return err;
580 		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
581 			err = -EFAULT;
582 		return err;
583 	}
584 	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
585 	case MSG_STAT_ANY:
586 	case IPC_STAT:
587 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
588 		if (err < 0)
589 			return err;
590 		if (copy_msqid_to_user(buf, &msqid64, version))
591 			err = -EFAULT;
592 		return err;
593 	case IPC_SET:
594 		if (copy_msqid_from_user(&msqid64, buf, version))
595 			return -EFAULT;
596 		/* fallthru */
597 	case IPC_RMID:
598 		return msgctl_down(ns, msqid, cmd, &msqid64);
599 	default:
600 		return  -EINVAL;
601 	}
602 }
603 
604 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
605 {
606 	return ksys_msgctl(msqid, cmd, buf);
607 }
608 
609 #ifdef CONFIG_COMPAT
610 
611 struct compat_msqid_ds {
612 	struct compat_ipc_perm msg_perm;
613 	compat_uptr_t msg_first;
614 	compat_uptr_t msg_last;
615 	compat_time_t msg_stime;
616 	compat_time_t msg_rtime;
617 	compat_time_t msg_ctime;
618 	compat_ulong_t msg_lcbytes;
619 	compat_ulong_t msg_lqbytes;
620 	unsigned short msg_cbytes;
621 	unsigned short msg_qnum;
622 	unsigned short msg_qbytes;
623 	compat_ipc_pid_t msg_lspid;
624 	compat_ipc_pid_t msg_lrpid;
625 };
626 
627 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
628 					int version)
629 {
630 	memset(out, 0, sizeof(*out));
631 	if (version == IPC_64) {
632 		struct compat_msqid64_ds __user *p = buf;
633 		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
634 			return -EFAULT;
635 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
636 			return -EFAULT;
637 	} else {
638 		struct compat_msqid_ds __user *p = buf;
639 		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
640 			return -EFAULT;
641 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
642 			return -EFAULT;
643 	}
644 	return 0;
645 }
646 
647 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
648 					int version)
649 {
650 	if (version == IPC_64) {
651 		struct compat_msqid64_ds v;
652 		memset(&v, 0, sizeof(v));
653 		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
654 		v.msg_stime	 = lower_32_bits(in->msg_stime);
655 		v.msg_stime_high = upper_32_bits(in->msg_stime);
656 		v.msg_rtime	 = lower_32_bits(in->msg_rtime);
657 		v.msg_rtime_high = upper_32_bits(in->msg_rtime);
658 		v.msg_ctime	 = lower_32_bits(in->msg_ctime);
659 		v.msg_ctime_high = upper_32_bits(in->msg_ctime);
660 		v.msg_cbytes = in->msg_cbytes;
661 		v.msg_qnum = in->msg_qnum;
662 		v.msg_qbytes = in->msg_qbytes;
663 		v.msg_lspid = in->msg_lspid;
664 		v.msg_lrpid = in->msg_lrpid;
665 		return copy_to_user(buf, &v, sizeof(v));
666 	} else {
667 		struct compat_msqid_ds v;
668 		memset(&v, 0, sizeof(v));
669 		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
670 		v.msg_stime = in->msg_stime;
671 		v.msg_rtime = in->msg_rtime;
672 		v.msg_ctime = in->msg_ctime;
673 		v.msg_cbytes = in->msg_cbytes;
674 		v.msg_qnum = in->msg_qnum;
675 		v.msg_qbytes = in->msg_qbytes;
676 		v.msg_lspid = in->msg_lspid;
677 		v.msg_lrpid = in->msg_lrpid;
678 		return copy_to_user(buf, &v, sizeof(v));
679 	}
680 }
681 
682 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
683 {
684 	struct ipc_namespace *ns;
685 	int err;
686 	struct msqid64_ds msqid64;
687 	int version = compat_ipc_parse_version(&cmd);
688 
689 	ns = current->nsproxy->ipc_ns;
690 
691 	if (msqid < 0 || cmd < 0)
692 		return -EINVAL;
693 
694 	switch (cmd & (~IPC_64)) {
695 	case IPC_INFO:
696 	case MSG_INFO: {
697 		struct msginfo msginfo;
698 		err = msgctl_info(ns, msqid, cmd, &msginfo);
699 		if (err < 0)
700 			return err;
701 		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
702 			err = -EFAULT;
703 		return err;
704 	}
705 	case IPC_STAT:
706 	case MSG_STAT:
707 	case MSG_STAT_ANY:
708 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
709 		if (err < 0)
710 			return err;
711 		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
712 			err = -EFAULT;
713 		return err;
714 	case IPC_SET:
715 		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
716 			return -EFAULT;
717 		/* fallthru */
718 	case IPC_RMID:
719 		return msgctl_down(ns, msqid, cmd, &msqid64);
720 	default:
721 		return -EINVAL;
722 	}
723 }
724 
725 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
726 {
727 	return compat_ksys_msgctl(msqid, cmd, uptr);
728 }
729 #endif
730 
731 static int testmsg(struct msg_msg *msg, long type, int mode)
732 {
733 	switch (mode) {
734 	case SEARCH_ANY:
735 	case SEARCH_NUMBER:
736 		return 1;
737 	case SEARCH_LESSEQUAL:
738 		if (msg->m_type <= type)
739 			return 1;
740 		break;
741 	case SEARCH_EQUAL:
742 		if (msg->m_type == type)
743 			return 1;
744 		break;
745 	case SEARCH_NOTEQUAL:
746 		if (msg->m_type != type)
747 			return 1;
748 		break;
749 	}
750 	return 0;
751 }
752 
753 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
754 				 struct wake_q_head *wake_q)
755 {
756 	struct msg_receiver *msr, *t;
757 
758 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
759 		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
760 		    !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
761 					       msr->r_msgtype, msr->r_mode)) {
762 
763 			list_del(&msr->r_list);
764 			if (msr->r_maxsize < msg->m_ts) {
765 				wake_q_add(wake_q, msr->r_tsk);
766 				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
767 			} else {
768 				ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
769 				msq->q_rtime = ktime_get_real_seconds();
770 
771 				wake_q_add(wake_q, msr->r_tsk);
772 				WRITE_ONCE(msr->r_msg, msg);
773 				return 1;
774 			}
775 		}
776 	}
777 
778 	return 0;
779 }
780 
781 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
782 		size_t msgsz, int msgflg)
783 {
784 	struct msg_queue *msq;
785 	struct msg_msg *msg;
786 	int err;
787 	struct ipc_namespace *ns;
788 	DEFINE_WAKE_Q(wake_q);
789 
790 	ns = current->nsproxy->ipc_ns;
791 
792 	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
793 		return -EINVAL;
794 	if (mtype < 1)
795 		return -EINVAL;
796 
797 	msg = load_msg(mtext, msgsz);
798 	if (IS_ERR(msg))
799 		return PTR_ERR(msg);
800 
801 	msg->m_type = mtype;
802 	msg->m_ts = msgsz;
803 
804 	rcu_read_lock();
805 	msq = msq_obtain_object_check(ns, msqid);
806 	if (IS_ERR(msq)) {
807 		err = PTR_ERR(msq);
808 		goto out_unlock1;
809 	}
810 
811 	ipc_lock_object(&msq->q_perm);
812 
813 	for (;;) {
814 		struct msg_sender s;
815 
816 		err = -EACCES;
817 		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
818 			goto out_unlock0;
819 
820 		/* raced with RMID? */
821 		if (!ipc_valid_object(&msq->q_perm)) {
822 			err = -EIDRM;
823 			goto out_unlock0;
824 		}
825 
826 		err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
827 		if (err)
828 			goto out_unlock0;
829 
830 		if (msg_fits_inqueue(msq, msgsz))
831 			break;
832 
833 		/* queue full, wait: */
834 		if (msgflg & IPC_NOWAIT) {
835 			err = -EAGAIN;
836 			goto out_unlock0;
837 		}
838 
839 		/* enqueue the sender and prepare to block */
840 		ss_add(msq, &s, msgsz);
841 
842 		if (!ipc_rcu_getref(&msq->q_perm)) {
843 			err = -EIDRM;
844 			goto out_unlock0;
845 		}
846 
847 		ipc_unlock_object(&msq->q_perm);
848 		rcu_read_unlock();
849 		schedule();
850 
851 		rcu_read_lock();
852 		ipc_lock_object(&msq->q_perm);
853 
854 		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
855 		/* raced with RMID? */
856 		if (!ipc_valid_object(&msq->q_perm)) {
857 			err = -EIDRM;
858 			goto out_unlock0;
859 		}
860 		ss_del(&s);
861 
862 		if (signal_pending(current)) {
863 			err = -ERESTARTNOHAND;
864 			goto out_unlock0;
865 		}
866 
867 	}
868 
869 	ipc_update_pid(&msq->q_lspid, task_tgid(current));
870 	msq->q_stime = ktime_get_real_seconds();
871 
872 	if (!pipelined_send(msq, msg, &wake_q)) {
873 		/* no one is waiting for this message, enqueue it */
874 		list_add_tail(&msg->m_list, &msq->q_messages);
875 		msq->q_cbytes += msgsz;
876 		msq->q_qnum++;
877 		atomic_add(msgsz, &ns->msg_bytes);
878 		atomic_inc(&ns->msg_hdrs);
879 	}
880 
881 	err = 0;
882 	msg = NULL;
883 
884 out_unlock0:
885 	ipc_unlock_object(&msq->q_perm);
886 	wake_up_q(&wake_q);
887 out_unlock1:
888 	rcu_read_unlock();
889 	if (msg != NULL)
890 		free_msg(msg);
891 	return err;
892 }
893 
894 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
895 		 int msgflg)
896 {
897 	long mtype;
898 
899 	if (get_user(mtype, &msgp->mtype))
900 		return -EFAULT;
901 	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
902 }
903 
904 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
905 		int, msgflg)
906 {
907 	return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
908 }
909 
910 #ifdef CONFIG_COMPAT
911 
912 struct compat_msgbuf {
913 	compat_long_t mtype;
914 	char mtext[1];
915 };
916 
917 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
918 		       compat_ssize_t msgsz, int msgflg)
919 {
920 	struct compat_msgbuf __user *up = compat_ptr(msgp);
921 	compat_long_t mtype;
922 
923 	if (get_user(mtype, &up->mtype))
924 		return -EFAULT;
925 	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
926 }
927 
928 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
929 		       compat_ssize_t, msgsz, int, msgflg)
930 {
931 	return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
932 }
933 #endif
934 
935 static inline int convert_mode(long *msgtyp, int msgflg)
936 {
937 	if (msgflg & MSG_COPY)
938 		return SEARCH_NUMBER;
939 	/*
940 	 *  find message of correct type.
941 	 *  msgtyp = 0 => get first.
942 	 *  msgtyp > 0 => get first message of matching type.
943 	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
944 	 */
945 	if (*msgtyp == 0)
946 		return SEARCH_ANY;
947 	if (*msgtyp < 0) {
948 		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
949 			*msgtyp = LONG_MAX;
950 		else
951 			*msgtyp = -*msgtyp;
952 		return SEARCH_LESSEQUAL;
953 	}
954 	if (msgflg & MSG_EXCEPT)
955 		return SEARCH_NOTEQUAL;
956 	return SEARCH_EQUAL;
957 }
958 
959 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
960 {
961 	struct msgbuf __user *msgp = dest;
962 	size_t msgsz;
963 
964 	if (put_user(msg->m_type, &msgp->mtype))
965 		return -EFAULT;
966 
967 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
968 	if (store_msg(msgp->mtext, msg, msgsz))
969 		return -EFAULT;
970 	return msgsz;
971 }
972 
973 #ifdef CONFIG_CHECKPOINT_RESTORE
974 /*
975  * This function creates new kernel message structure, large enough to store
976  * bufsz message bytes.
977  */
978 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
979 {
980 	struct msg_msg *copy;
981 
982 	/*
983 	 * Create dummy message to copy real message to.
984 	 */
985 	copy = load_msg(buf, bufsz);
986 	if (!IS_ERR(copy))
987 		copy->m_ts = bufsz;
988 	return copy;
989 }
990 
991 static inline void free_copy(struct msg_msg *copy)
992 {
993 	if (copy)
994 		free_msg(copy);
995 }
996 #else
997 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
998 {
999 	return ERR_PTR(-ENOSYS);
1000 }
1001 
1002 static inline void free_copy(struct msg_msg *copy)
1003 {
1004 }
1005 #endif
1006 
1007 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1008 {
1009 	struct msg_msg *msg, *found = NULL;
1010 	long count = 0;
1011 
1012 	list_for_each_entry(msg, &msq->q_messages, m_list) {
1013 		if (testmsg(msg, *msgtyp, mode) &&
1014 		    !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1015 					       *msgtyp, mode)) {
1016 			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1017 				*msgtyp = msg->m_type - 1;
1018 				found = msg;
1019 			} else if (mode == SEARCH_NUMBER) {
1020 				if (*msgtyp == count)
1021 					return msg;
1022 			} else
1023 				return msg;
1024 			count++;
1025 		}
1026 	}
1027 
1028 	return found ?: ERR_PTR(-EAGAIN);
1029 }
1030 
1031 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1032 	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1033 {
1034 	int mode;
1035 	struct msg_queue *msq;
1036 	struct ipc_namespace *ns;
1037 	struct msg_msg *msg, *copy = NULL;
1038 	DEFINE_WAKE_Q(wake_q);
1039 
1040 	ns = current->nsproxy->ipc_ns;
1041 
1042 	if (msqid < 0 || (long) bufsz < 0)
1043 		return -EINVAL;
1044 
1045 	if (msgflg & MSG_COPY) {
1046 		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1047 			return -EINVAL;
1048 		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1049 		if (IS_ERR(copy))
1050 			return PTR_ERR(copy);
1051 	}
1052 	mode = convert_mode(&msgtyp, msgflg);
1053 
1054 	rcu_read_lock();
1055 	msq = msq_obtain_object_check(ns, msqid);
1056 	if (IS_ERR(msq)) {
1057 		rcu_read_unlock();
1058 		free_copy(copy);
1059 		return PTR_ERR(msq);
1060 	}
1061 
1062 	for (;;) {
1063 		struct msg_receiver msr_d;
1064 
1065 		msg = ERR_PTR(-EACCES);
1066 		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1067 			goto out_unlock1;
1068 
1069 		ipc_lock_object(&msq->q_perm);
1070 
1071 		/* raced with RMID? */
1072 		if (!ipc_valid_object(&msq->q_perm)) {
1073 			msg = ERR_PTR(-EIDRM);
1074 			goto out_unlock0;
1075 		}
1076 
1077 		msg = find_msg(msq, &msgtyp, mode);
1078 		if (!IS_ERR(msg)) {
1079 			/*
1080 			 * Found a suitable message.
1081 			 * Unlink it from the queue.
1082 			 */
1083 			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1084 				msg = ERR_PTR(-E2BIG);
1085 				goto out_unlock0;
1086 			}
1087 			/*
1088 			 * If we are copying, then do not unlink message and do
1089 			 * not update queue parameters.
1090 			 */
1091 			if (msgflg & MSG_COPY) {
1092 				msg = copy_msg(msg, copy);
1093 				goto out_unlock0;
1094 			}
1095 
1096 			list_del(&msg->m_list);
1097 			msq->q_qnum--;
1098 			msq->q_rtime = ktime_get_real_seconds();
1099 			ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1100 			msq->q_cbytes -= msg->m_ts;
1101 			atomic_sub(msg->m_ts, &ns->msg_bytes);
1102 			atomic_dec(&ns->msg_hdrs);
1103 			ss_wakeup(msq, &wake_q, false);
1104 
1105 			goto out_unlock0;
1106 		}
1107 
1108 		/* No message waiting. Wait for a message */
1109 		if (msgflg & IPC_NOWAIT) {
1110 			msg = ERR_PTR(-ENOMSG);
1111 			goto out_unlock0;
1112 		}
1113 
1114 		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1115 		msr_d.r_tsk = current;
1116 		msr_d.r_msgtype = msgtyp;
1117 		msr_d.r_mode = mode;
1118 		if (msgflg & MSG_NOERROR)
1119 			msr_d.r_maxsize = INT_MAX;
1120 		else
1121 			msr_d.r_maxsize = bufsz;
1122 		msr_d.r_msg = ERR_PTR(-EAGAIN);
1123 		__set_current_state(TASK_INTERRUPTIBLE);
1124 
1125 		ipc_unlock_object(&msq->q_perm);
1126 		rcu_read_unlock();
1127 		schedule();
1128 
1129 		/*
1130 		 * Lockless receive, part 1:
1131 		 * We don't hold a reference to the queue and getting a
1132 		 * reference would defeat the idea of a lockless operation,
1133 		 * thus the code relies on rcu to guarantee the existence of
1134 		 * msq:
1135 		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1136 		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1137 		 */
1138 		rcu_read_lock();
1139 
1140 		/*
1141 		 * Lockless receive, part 2:
1142 		 * The work in pipelined_send() and expunge_all():
1143 		 * - Set pointer to message
1144 		 * - Queue the receiver task for later wakeup
1145 		 * - Wake up the process after the lock is dropped.
1146 		 *
1147 		 * Should the process wake up before this wakeup (due to a
1148 		 * signal) it will either see the message and continue ...
1149 		 */
1150 		msg = READ_ONCE(msr_d.r_msg);
1151 		if (msg != ERR_PTR(-EAGAIN))
1152 			goto out_unlock1;
1153 
1154 		 /*
1155 		  * ... or see -EAGAIN, acquire the lock to check the message
1156 		  * again.
1157 		  */
1158 		ipc_lock_object(&msq->q_perm);
1159 
1160 		msg = msr_d.r_msg;
1161 		if (msg != ERR_PTR(-EAGAIN))
1162 			goto out_unlock0;
1163 
1164 		list_del(&msr_d.r_list);
1165 		if (signal_pending(current)) {
1166 			msg = ERR_PTR(-ERESTARTNOHAND);
1167 			goto out_unlock0;
1168 		}
1169 
1170 		ipc_unlock_object(&msq->q_perm);
1171 	}
1172 
1173 out_unlock0:
1174 	ipc_unlock_object(&msq->q_perm);
1175 	wake_up_q(&wake_q);
1176 out_unlock1:
1177 	rcu_read_unlock();
1178 	if (IS_ERR(msg)) {
1179 		free_copy(copy);
1180 		return PTR_ERR(msg);
1181 	}
1182 
1183 	bufsz = msg_handler(buf, msg, bufsz);
1184 	free_msg(msg);
1185 
1186 	return bufsz;
1187 }
1188 
1189 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1190 		 long msgtyp, int msgflg)
1191 {
1192 	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1193 }
1194 
1195 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1196 		long, msgtyp, int, msgflg)
1197 {
1198 	return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1199 }
1200 
1201 #ifdef CONFIG_COMPAT
1202 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1203 {
1204 	struct compat_msgbuf __user *msgp = dest;
1205 	size_t msgsz;
1206 
1207 	if (put_user(msg->m_type, &msgp->mtype))
1208 		return -EFAULT;
1209 
1210 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1211 	if (store_msg(msgp->mtext, msg, msgsz))
1212 		return -EFAULT;
1213 	return msgsz;
1214 }
1215 
1216 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1217 			compat_long_t msgtyp, int msgflg)
1218 {
1219 	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1220 			 msgflg, compat_do_msg_fill);
1221 }
1222 
1223 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1224 		       compat_ssize_t, msgsz, compat_long_t, msgtyp,
1225 		       int, msgflg)
1226 {
1227 	return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1228 }
1229 #endif
1230 
1231 int msg_init_ns(struct ipc_namespace *ns)
1232 {
1233 	ns->msg_ctlmax = MSGMAX;
1234 	ns->msg_ctlmnb = MSGMNB;
1235 	ns->msg_ctlmni = MSGMNI;
1236 
1237 	atomic_set(&ns->msg_bytes, 0);
1238 	atomic_set(&ns->msg_hdrs, 0);
1239 	return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1240 }
1241 
1242 #ifdef CONFIG_IPC_NS
1243 void msg_exit_ns(struct ipc_namespace *ns)
1244 {
1245 	free_ipcs(ns, &msg_ids(ns), freeque);
1246 	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1247 	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1248 }
1249 #endif
1250 
1251 #ifdef CONFIG_PROC_FS
1252 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1253 {
1254 	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1255 	struct user_namespace *user_ns = seq_user_ns(s);
1256 	struct kern_ipc_perm *ipcp = it;
1257 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1258 
1259 	seq_printf(s,
1260 		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1261 		   msq->q_perm.key,
1262 		   msq->q_perm.id,
1263 		   msq->q_perm.mode,
1264 		   msq->q_cbytes,
1265 		   msq->q_qnum,
1266 		   pid_nr_ns(msq->q_lspid, pid_ns),
1267 		   pid_nr_ns(msq->q_lrpid, pid_ns),
1268 		   from_kuid_munged(user_ns, msq->q_perm.uid),
1269 		   from_kgid_munged(user_ns, msq->q_perm.gid),
1270 		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1271 		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1272 		   msq->q_stime,
1273 		   msq->q_rtime,
1274 		   msq->q_ctime);
1275 
1276 	return 0;
1277 }
1278 #endif
1279 
1280 int __init msg_init(void)
1281 {
1282 	const int err = msg_init_ns(&init_ipc_ns);
1283 
1284 	ipc_init_proc_interface("sysvipc/msg",
1285 				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1286 				IPC_MSG_IDS, sysvipc_msg_proc_show);
1287 	return err;
1288 }
1289