xref: /linux/ipc/msg.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
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 	p->msg_cbytes = msq->q_cbytes;
541 	p->msg_qnum   = msq->q_qnum;
542 	p->msg_qbytes = msq->q_qbytes;
543 	p->msg_lspid  = pid_vnr(msq->q_lspid);
544 	p->msg_lrpid  = pid_vnr(msq->q_lrpid);
545 
546 	ipc_unlock_object(&msq->q_perm);
547 	rcu_read_unlock();
548 	return id;
549 
550 out_unlock:
551 	rcu_read_unlock();
552 	return err;
553 }
554 
555 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
556 {
557 	int version;
558 	struct ipc_namespace *ns;
559 	struct msqid64_ds msqid64;
560 	int err;
561 
562 	if (msqid < 0 || cmd < 0)
563 		return -EINVAL;
564 
565 	version = ipc_parse_version(&cmd);
566 	ns = current->nsproxy->ipc_ns;
567 
568 	switch (cmd) {
569 	case IPC_INFO:
570 	case MSG_INFO: {
571 		struct msginfo msginfo;
572 		err = msgctl_info(ns, msqid, cmd, &msginfo);
573 		if (err < 0)
574 			return err;
575 		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
576 			err = -EFAULT;
577 		return err;
578 	}
579 	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
580 	case MSG_STAT_ANY:
581 	case IPC_STAT:
582 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
583 		if (err < 0)
584 			return err;
585 		if (copy_msqid_to_user(buf, &msqid64, version))
586 			err = -EFAULT;
587 		return err;
588 	case IPC_SET:
589 		if (copy_msqid_from_user(&msqid64, buf, version))
590 			return -EFAULT;
591 		/* fallthru */
592 	case IPC_RMID:
593 		return msgctl_down(ns, msqid, cmd, &msqid64);
594 	default:
595 		return  -EINVAL;
596 	}
597 }
598 
599 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
600 {
601 	return ksys_msgctl(msqid, cmd, buf);
602 }
603 
604 #ifdef CONFIG_COMPAT
605 
606 struct compat_msqid_ds {
607 	struct compat_ipc_perm msg_perm;
608 	compat_uptr_t msg_first;
609 	compat_uptr_t msg_last;
610 	compat_time_t msg_stime;
611 	compat_time_t msg_rtime;
612 	compat_time_t msg_ctime;
613 	compat_ulong_t msg_lcbytes;
614 	compat_ulong_t msg_lqbytes;
615 	unsigned short msg_cbytes;
616 	unsigned short msg_qnum;
617 	unsigned short msg_qbytes;
618 	compat_ipc_pid_t msg_lspid;
619 	compat_ipc_pid_t msg_lrpid;
620 };
621 
622 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
623 					int version)
624 {
625 	memset(out, 0, sizeof(*out));
626 	if (version == IPC_64) {
627 		struct compat_msqid64_ds __user *p = buf;
628 		if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
629 			return -EFAULT;
630 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
631 			return -EFAULT;
632 	} else {
633 		struct compat_msqid_ds __user *p = buf;
634 		if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
635 			return -EFAULT;
636 		if (get_user(out->msg_qbytes, &p->msg_qbytes))
637 			return -EFAULT;
638 	}
639 	return 0;
640 }
641 
642 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
643 					int version)
644 {
645 	if (version == IPC_64) {
646 		struct compat_msqid64_ds v;
647 		memset(&v, 0, sizeof(v));
648 		to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
649 		v.msg_stime = in->msg_stime;
650 		v.msg_rtime = in->msg_rtime;
651 		v.msg_ctime = in->msg_ctime;
652 		v.msg_cbytes = in->msg_cbytes;
653 		v.msg_qnum = in->msg_qnum;
654 		v.msg_qbytes = in->msg_qbytes;
655 		v.msg_lspid = in->msg_lspid;
656 		v.msg_lrpid = in->msg_lrpid;
657 		return copy_to_user(buf, &v, sizeof(v));
658 	} else {
659 		struct compat_msqid_ds v;
660 		memset(&v, 0, sizeof(v));
661 		to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
662 		v.msg_stime = in->msg_stime;
663 		v.msg_rtime = in->msg_rtime;
664 		v.msg_ctime = in->msg_ctime;
665 		v.msg_cbytes = in->msg_cbytes;
666 		v.msg_qnum = in->msg_qnum;
667 		v.msg_qbytes = in->msg_qbytes;
668 		v.msg_lspid = in->msg_lspid;
669 		v.msg_lrpid = in->msg_lrpid;
670 		return copy_to_user(buf, &v, sizeof(v));
671 	}
672 }
673 
674 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
675 {
676 	struct ipc_namespace *ns;
677 	int err;
678 	struct msqid64_ds msqid64;
679 	int version = compat_ipc_parse_version(&cmd);
680 
681 	ns = current->nsproxy->ipc_ns;
682 
683 	if (msqid < 0 || cmd < 0)
684 		return -EINVAL;
685 
686 	switch (cmd & (~IPC_64)) {
687 	case IPC_INFO:
688 	case MSG_INFO: {
689 		struct msginfo msginfo;
690 		err = msgctl_info(ns, msqid, cmd, &msginfo);
691 		if (err < 0)
692 			return err;
693 		if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
694 			err = -EFAULT;
695 		return err;
696 	}
697 	case IPC_STAT:
698 	case MSG_STAT:
699 	case MSG_STAT_ANY:
700 		err = msgctl_stat(ns, msqid, cmd, &msqid64);
701 		if (err < 0)
702 			return err;
703 		if (copy_compat_msqid_to_user(uptr, &msqid64, version))
704 			err = -EFAULT;
705 		return err;
706 	case IPC_SET:
707 		if (copy_compat_msqid_from_user(&msqid64, uptr, version))
708 			return -EFAULT;
709 		/* fallthru */
710 	case IPC_RMID:
711 		return msgctl_down(ns, msqid, cmd, &msqid64);
712 	default:
713 		return -EINVAL;
714 	}
715 }
716 
717 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
718 {
719 	return compat_ksys_msgctl(msqid, cmd, uptr);
720 }
721 #endif
722 
723 static int testmsg(struct msg_msg *msg, long type, int mode)
724 {
725 	switch (mode) {
726 	case SEARCH_ANY:
727 	case SEARCH_NUMBER:
728 		return 1;
729 	case SEARCH_LESSEQUAL:
730 		if (msg->m_type <= type)
731 			return 1;
732 		break;
733 	case SEARCH_EQUAL:
734 		if (msg->m_type == type)
735 			return 1;
736 		break;
737 	case SEARCH_NOTEQUAL:
738 		if (msg->m_type != type)
739 			return 1;
740 		break;
741 	}
742 	return 0;
743 }
744 
745 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
746 				 struct wake_q_head *wake_q)
747 {
748 	struct msg_receiver *msr, *t;
749 
750 	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
751 		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
752 		    !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
753 					       msr->r_msgtype, msr->r_mode)) {
754 
755 			list_del(&msr->r_list);
756 			if (msr->r_maxsize < msg->m_ts) {
757 				wake_q_add(wake_q, msr->r_tsk);
758 				WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
759 			} else {
760 				ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
761 				msq->q_rtime = get_seconds();
762 
763 				wake_q_add(wake_q, msr->r_tsk);
764 				WRITE_ONCE(msr->r_msg, msg);
765 				return 1;
766 			}
767 		}
768 	}
769 
770 	return 0;
771 }
772 
773 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
774 		size_t msgsz, int msgflg)
775 {
776 	struct msg_queue *msq;
777 	struct msg_msg *msg;
778 	int err;
779 	struct ipc_namespace *ns;
780 	DEFINE_WAKE_Q(wake_q);
781 
782 	ns = current->nsproxy->ipc_ns;
783 
784 	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
785 		return -EINVAL;
786 	if (mtype < 1)
787 		return -EINVAL;
788 
789 	msg = load_msg(mtext, msgsz);
790 	if (IS_ERR(msg))
791 		return PTR_ERR(msg);
792 
793 	msg->m_type = mtype;
794 	msg->m_ts = msgsz;
795 
796 	rcu_read_lock();
797 	msq = msq_obtain_object_check(ns, msqid);
798 	if (IS_ERR(msq)) {
799 		err = PTR_ERR(msq);
800 		goto out_unlock1;
801 	}
802 
803 	ipc_lock_object(&msq->q_perm);
804 
805 	for (;;) {
806 		struct msg_sender s;
807 
808 		err = -EACCES;
809 		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
810 			goto out_unlock0;
811 
812 		/* raced with RMID? */
813 		if (!ipc_valid_object(&msq->q_perm)) {
814 			err = -EIDRM;
815 			goto out_unlock0;
816 		}
817 
818 		err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
819 		if (err)
820 			goto out_unlock0;
821 
822 		if (msg_fits_inqueue(msq, msgsz))
823 			break;
824 
825 		/* queue full, wait: */
826 		if (msgflg & IPC_NOWAIT) {
827 			err = -EAGAIN;
828 			goto out_unlock0;
829 		}
830 
831 		/* enqueue the sender and prepare to block */
832 		ss_add(msq, &s, msgsz);
833 
834 		if (!ipc_rcu_getref(&msq->q_perm)) {
835 			err = -EIDRM;
836 			goto out_unlock0;
837 		}
838 
839 		ipc_unlock_object(&msq->q_perm);
840 		rcu_read_unlock();
841 		schedule();
842 
843 		rcu_read_lock();
844 		ipc_lock_object(&msq->q_perm);
845 
846 		ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
847 		/* raced with RMID? */
848 		if (!ipc_valid_object(&msq->q_perm)) {
849 			err = -EIDRM;
850 			goto out_unlock0;
851 		}
852 		ss_del(&s);
853 
854 		if (signal_pending(current)) {
855 			err = -ERESTARTNOHAND;
856 			goto out_unlock0;
857 		}
858 
859 	}
860 
861 	ipc_update_pid(&msq->q_lspid, task_tgid(current));
862 	msq->q_stime = get_seconds();
863 
864 	if (!pipelined_send(msq, msg, &wake_q)) {
865 		/* no one is waiting for this message, enqueue it */
866 		list_add_tail(&msg->m_list, &msq->q_messages);
867 		msq->q_cbytes += msgsz;
868 		msq->q_qnum++;
869 		atomic_add(msgsz, &ns->msg_bytes);
870 		atomic_inc(&ns->msg_hdrs);
871 	}
872 
873 	err = 0;
874 	msg = NULL;
875 
876 out_unlock0:
877 	ipc_unlock_object(&msq->q_perm);
878 	wake_up_q(&wake_q);
879 out_unlock1:
880 	rcu_read_unlock();
881 	if (msg != NULL)
882 		free_msg(msg);
883 	return err;
884 }
885 
886 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
887 		 int msgflg)
888 {
889 	long mtype;
890 
891 	if (get_user(mtype, &msgp->mtype))
892 		return -EFAULT;
893 	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
894 }
895 
896 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
897 		int, msgflg)
898 {
899 	return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
900 }
901 
902 #ifdef CONFIG_COMPAT
903 
904 struct compat_msgbuf {
905 	compat_long_t mtype;
906 	char mtext[1];
907 };
908 
909 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
910 		       compat_ssize_t msgsz, int msgflg)
911 {
912 	struct compat_msgbuf __user *up = compat_ptr(msgp);
913 	compat_long_t mtype;
914 
915 	if (get_user(mtype, &up->mtype))
916 		return -EFAULT;
917 	return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
918 }
919 
920 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
921 		       compat_ssize_t, msgsz, int, msgflg)
922 {
923 	return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
924 }
925 #endif
926 
927 static inline int convert_mode(long *msgtyp, int msgflg)
928 {
929 	if (msgflg & MSG_COPY)
930 		return SEARCH_NUMBER;
931 	/*
932 	 *  find message of correct type.
933 	 *  msgtyp = 0 => get first.
934 	 *  msgtyp > 0 => get first message of matching type.
935 	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
936 	 */
937 	if (*msgtyp == 0)
938 		return SEARCH_ANY;
939 	if (*msgtyp < 0) {
940 		if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
941 			*msgtyp = LONG_MAX;
942 		else
943 			*msgtyp = -*msgtyp;
944 		return SEARCH_LESSEQUAL;
945 	}
946 	if (msgflg & MSG_EXCEPT)
947 		return SEARCH_NOTEQUAL;
948 	return SEARCH_EQUAL;
949 }
950 
951 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
952 {
953 	struct msgbuf __user *msgp = dest;
954 	size_t msgsz;
955 
956 	if (put_user(msg->m_type, &msgp->mtype))
957 		return -EFAULT;
958 
959 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
960 	if (store_msg(msgp->mtext, msg, msgsz))
961 		return -EFAULT;
962 	return msgsz;
963 }
964 
965 #ifdef CONFIG_CHECKPOINT_RESTORE
966 /*
967  * This function creates new kernel message structure, large enough to store
968  * bufsz message bytes.
969  */
970 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
971 {
972 	struct msg_msg *copy;
973 
974 	/*
975 	 * Create dummy message to copy real message to.
976 	 */
977 	copy = load_msg(buf, bufsz);
978 	if (!IS_ERR(copy))
979 		copy->m_ts = bufsz;
980 	return copy;
981 }
982 
983 static inline void free_copy(struct msg_msg *copy)
984 {
985 	if (copy)
986 		free_msg(copy);
987 }
988 #else
989 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
990 {
991 	return ERR_PTR(-ENOSYS);
992 }
993 
994 static inline void free_copy(struct msg_msg *copy)
995 {
996 }
997 #endif
998 
999 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1000 {
1001 	struct msg_msg *msg, *found = NULL;
1002 	long count = 0;
1003 
1004 	list_for_each_entry(msg, &msq->q_messages, m_list) {
1005 		if (testmsg(msg, *msgtyp, mode) &&
1006 		    !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1007 					       *msgtyp, mode)) {
1008 			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1009 				*msgtyp = msg->m_type - 1;
1010 				found = msg;
1011 			} else if (mode == SEARCH_NUMBER) {
1012 				if (*msgtyp == count)
1013 					return msg;
1014 			} else
1015 				return msg;
1016 			count++;
1017 		}
1018 	}
1019 
1020 	return found ?: ERR_PTR(-EAGAIN);
1021 }
1022 
1023 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1024 	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1025 {
1026 	int mode;
1027 	struct msg_queue *msq;
1028 	struct ipc_namespace *ns;
1029 	struct msg_msg *msg, *copy = NULL;
1030 	DEFINE_WAKE_Q(wake_q);
1031 
1032 	ns = current->nsproxy->ipc_ns;
1033 
1034 	if (msqid < 0 || (long) bufsz < 0)
1035 		return -EINVAL;
1036 
1037 	if (msgflg & MSG_COPY) {
1038 		if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1039 			return -EINVAL;
1040 		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1041 		if (IS_ERR(copy))
1042 			return PTR_ERR(copy);
1043 	}
1044 	mode = convert_mode(&msgtyp, msgflg);
1045 
1046 	rcu_read_lock();
1047 	msq = msq_obtain_object_check(ns, msqid);
1048 	if (IS_ERR(msq)) {
1049 		rcu_read_unlock();
1050 		free_copy(copy);
1051 		return PTR_ERR(msq);
1052 	}
1053 
1054 	for (;;) {
1055 		struct msg_receiver msr_d;
1056 
1057 		msg = ERR_PTR(-EACCES);
1058 		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1059 			goto out_unlock1;
1060 
1061 		ipc_lock_object(&msq->q_perm);
1062 
1063 		/* raced with RMID? */
1064 		if (!ipc_valid_object(&msq->q_perm)) {
1065 			msg = ERR_PTR(-EIDRM);
1066 			goto out_unlock0;
1067 		}
1068 
1069 		msg = find_msg(msq, &msgtyp, mode);
1070 		if (!IS_ERR(msg)) {
1071 			/*
1072 			 * Found a suitable message.
1073 			 * Unlink it from the queue.
1074 			 */
1075 			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1076 				msg = ERR_PTR(-E2BIG);
1077 				goto out_unlock0;
1078 			}
1079 			/*
1080 			 * If we are copying, then do not unlink message and do
1081 			 * not update queue parameters.
1082 			 */
1083 			if (msgflg & MSG_COPY) {
1084 				msg = copy_msg(msg, copy);
1085 				goto out_unlock0;
1086 			}
1087 
1088 			list_del(&msg->m_list);
1089 			msq->q_qnum--;
1090 			msq->q_rtime = get_seconds();
1091 			ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1092 			msq->q_cbytes -= msg->m_ts;
1093 			atomic_sub(msg->m_ts, &ns->msg_bytes);
1094 			atomic_dec(&ns->msg_hdrs);
1095 			ss_wakeup(msq, &wake_q, false);
1096 
1097 			goto out_unlock0;
1098 		}
1099 
1100 		/* No message waiting. Wait for a message */
1101 		if (msgflg & IPC_NOWAIT) {
1102 			msg = ERR_PTR(-ENOMSG);
1103 			goto out_unlock0;
1104 		}
1105 
1106 		list_add_tail(&msr_d.r_list, &msq->q_receivers);
1107 		msr_d.r_tsk = current;
1108 		msr_d.r_msgtype = msgtyp;
1109 		msr_d.r_mode = mode;
1110 		if (msgflg & MSG_NOERROR)
1111 			msr_d.r_maxsize = INT_MAX;
1112 		else
1113 			msr_d.r_maxsize = bufsz;
1114 		msr_d.r_msg = ERR_PTR(-EAGAIN);
1115 		__set_current_state(TASK_INTERRUPTIBLE);
1116 
1117 		ipc_unlock_object(&msq->q_perm);
1118 		rcu_read_unlock();
1119 		schedule();
1120 
1121 		/*
1122 		 * Lockless receive, part 1:
1123 		 * We don't hold a reference to the queue and getting a
1124 		 * reference would defeat the idea of a lockless operation,
1125 		 * thus the code relies on rcu to guarantee the existence of
1126 		 * msq:
1127 		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1128 		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1129 		 */
1130 		rcu_read_lock();
1131 
1132 		/*
1133 		 * Lockless receive, part 2:
1134 		 * The work in pipelined_send() and expunge_all():
1135 		 * - Set pointer to message
1136 		 * - Queue the receiver task for later wakeup
1137 		 * - Wake up the process after the lock is dropped.
1138 		 *
1139 		 * Should the process wake up before this wakeup (due to a
1140 		 * signal) it will either see the message and continue ...
1141 		 */
1142 		msg = READ_ONCE(msr_d.r_msg);
1143 		if (msg != ERR_PTR(-EAGAIN))
1144 			goto out_unlock1;
1145 
1146 		 /*
1147 		  * ... or see -EAGAIN, acquire the lock to check the message
1148 		  * again.
1149 		  */
1150 		ipc_lock_object(&msq->q_perm);
1151 
1152 		msg = msr_d.r_msg;
1153 		if (msg != ERR_PTR(-EAGAIN))
1154 			goto out_unlock0;
1155 
1156 		list_del(&msr_d.r_list);
1157 		if (signal_pending(current)) {
1158 			msg = ERR_PTR(-ERESTARTNOHAND);
1159 			goto out_unlock0;
1160 		}
1161 
1162 		ipc_unlock_object(&msq->q_perm);
1163 	}
1164 
1165 out_unlock0:
1166 	ipc_unlock_object(&msq->q_perm);
1167 	wake_up_q(&wake_q);
1168 out_unlock1:
1169 	rcu_read_unlock();
1170 	if (IS_ERR(msg)) {
1171 		free_copy(copy);
1172 		return PTR_ERR(msg);
1173 	}
1174 
1175 	bufsz = msg_handler(buf, msg, bufsz);
1176 	free_msg(msg);
1177 
1178 	return bufsz;
1179 }
1180 
1181 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1182 		 long msgtyp, int msgflg)
1183 {
1184 	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1185 }
1186 
1187 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1188 		long, msgtyp, int, msgflg)
1189 {
1190 	return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1191 }
1192 
1193 #ifdef CONFIG_COMPAT
1194 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1195 {
1196 	struct compat_msgbuf __user *msgp = dest;
1197 	size_t msgsz;
1198 
1199 	if (put_user(msg->m_type, &msgp->mtype))
1200 		return -EFAULT;
1201 
1202 	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1203 	if (store_msg(msgp->mtext, msg, msgsz))
1204 		return -EFAULT;
1205 	return msgsz;
1206 }
1207 
1208 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1209 			compat_long_t msgtyp, int msgflg)
1210 {
1211 	return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1212 			 msgflg, compat_do_msg_fill);
1213 }
1214 
1215 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1216 		       compat_ssize_t, msgsz, compat_long_t, msgtyp,
1217 		       int, msgflg)
1218 {
1219 	return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1220 }
1221 #endif
1222 
1223 int msg_init_ns(struct ipc_namespace *ns)
1224 {
1225 	ns->msg_ctlmax = MSGMAX;
1226 	ns->msg_ctlmnb = MSGMNB;
1227 	ns->msg_ctlmni = MSGMNI;
1228 
1229 	atomic_set(&ns->msg_bytes, 0);
1230 	atomic_set(&ns->msg_hdrs, 0);
1231 	return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1232 }
1233 
1234 #ifdef CONFIG_IPC_NS
1235 void msg_exit_ns(struct ipc_namespace *ns)
1236 {
1237 	free_ipcs(ns, &msg_ids(ns), freeque);
1238 	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1239 	rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1240 }
1241 #endif
1242 
1243 #ifdef CONFIG_PROC_FS
1244 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1245 {
1246 	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1247 	struct user_namespace *user_ns = seq_user_ns(s);
1248 	struct kern_ipc_perm *ipcp = it;
1249 	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1250 
1251 	seq_printf(s,
1252 		   "%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1253 		   msq->q_perm.key,
1254 		   msq->q_perm.id,
1255 		   msq->q_perm.mode,
1256 		   msq->q_cbytes,
1257 		   msq->q_qnum,
1258 		   pid_nr_ns(msq->q_lspid, pid_ns),
1259 		   pid_nr_ns(msq->q_lrpid, pid_ns),
1260 		   from_kuid_munged(user_ns, msq->q_perm.uid),
1261 		   from_kgid_munged(user_ns, msq->q_perm.gid),
1262 		   from_kuid_munged(user_ns, msq->q_perm.cuid),
1263 		   from_kgid_munged(user_ns, msq->q_perm.cgid),
1264 		   msq->q_stime,
1265 		   msq->q_rtime,
1266 		   msq->q_ctime);
1267 
1268 	return 0;
1269 }
1270 #endif
1271 
1272 int __init msg_init(void)
1273 {
1274 	const int err = msg_init_ns(&init_ipc_ns);
1275 
1276 	ipc_init_proc_interface("sysvipc/msg",
1277 				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
1278 				IPC_MSG_IDS, sysvipc_msg_proc_show);
1279 	return err;
1280 }
1281