msg.c revision d076ac91
1/*
2 * linux/ipc/msg.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 *
5 * Removed all the remaining kerneld mess
6 * Catch the -EFAULT stuff properly
7 * Use GFP_KERNEL for messages as in 1.2
8 * Fixed up the unchecked user space derefs
9 * Copyright (C) 1998 Alan Cox & Andi Kleen
10 *
11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
12 *
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
15 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 */
24
25#include <linux/capability.h>
26#include <linux/msg.h>
27#include <linux/spinlock.h>
28#include <linux/init.h>
29#include <linux/mm.h>
30#include <linux/proc_fs.h>
31#include <linux/list.h>
32#include <linux/security.h>
33#include <linux/sched.h>
34#include <linux/syscalls.h>
35#include <linux/audit.h>
36#include <linux/seq_file.h>
37#include <linux/rwsem.h>
38#include <linux/nsproxy.h>
39#include <linux/ipc_namespace.h>
40
41#include <asm/current.h>
42#include <asm/uaccess.h>
43#include "util.h"
44
45/*
46 * one msg_receiver structure for each sleeping receiver:
47 */
48struct msg_receiver {
49	struct list_head	r_list;
50	struct task_struct	*r_tsk;
51
52	int			r_mode;
53	long			r_msgtype;
54	long			r_maxsize;
55
56	struct msg_msg		*volatile r_msg;
57};
58
59/* one msg_sender for each sleeping sender */
60struct msg_sender {
61	struct list_head	list;
62	struct task_struct	*tsk;
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#define msg_unlock(msq)		ipc_unlock(&(msq)->q_perm)
74
75static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
76static int newque(struct ipc_namespace *, struct ipc_params *);
77#ifdef CONFIG_PROC_FS
78static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
79#endif
80
81/*
82 * Scale msgmni with the available lowmem size: the memory dedicated to msg
83 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
84 * Also take into account the number of nsproxies created so far.
85 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
86 */
87void recompute_msgmni(struct ipc_namespace *ns)
88{
89	struct sysinfo i;
90	unsigned long allowed;
91	int nb_ns;
92
93	si_meminfo(&i);
94	allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
95		/ MSGMNB;
96	nb_ns = atomic_read(&nr_ipc_ns);
97	allowed /= nb_ns;
98
99	if (allowed < MSGMNI) {
100		ns->msg_ctlmni = MSGMNI;
101		return;
102	}
103
104	if (allowed > IPCMNI / nb_ns) {
105		ns->msg_ctlmni = IPCMNI / nb_ns;
106		return;
107	}
108
109	ns->msg_ctlmni = allowed;
110}
111
112void msg_init_ns(struct ipc_namespace *ns)
113{
114	ns->msg_ctlmax = MSGMAX;
115	ns->msg_ctlmnb = MSGMNB;
116
117	recompute_msgmni(ns);
118
119	atomic_set(&ns->msg_bytes, 0);
120	atomic_set(&ns->msg_hdrs, 0);
121	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
122}
123
124#ifdef CONFIG_IPC_NS
125void msg_exit_ns(struct ipc_namespace *ns)
126{
127	free_ipcs(ns, &msg_ids(ns), freeque);
128	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
129}
130#endif
131
132void __init msg_init(void)
133{
134	msg_init_ns(&init_ipc_ns);
135
136	printk(KERN_INFO "msgmni has been set to %d\n",
137		init_ipc_ns.msg_ctlmni);
138
139	ipc_init_proc_interface("sysvipc/msg",
140				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
141				IPC_MSG_IDS, sysvipc_msg_proc_show);
142}
143
144/*
145 * msg_lock_(check_) routines are called in the paths where the rw_mutex
146 * is not held.
147 */
148static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
149{
150	struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
151
152	if (IS_ERR(ipcp))
153		return (struct msg_queue *)ipcp;
154
155	return container_of(ipcp, struct msg_queue, q_perm);
156}
157
158static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
159						int id)
160{
161	struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
162
163	if (IS_ERR(ipcp))
164		return (struct msg_queue *)ipcp;
165
166	return container_of(ipcp, struct msg_queue, q_perm);
167}
168
169static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
170{
171	ipc_rmid(&msg_ids(ns), &s->q_perm);
172}
173
174/**
175 * newque - Create a new msg queue
176 * @ns: namespace
177 * @params: ptr to the structure that contains the key and msgflg
178 *
179 * Called with msg_ids.rw_mutex held (writer)
180 */
181static int newque(struct ipc_namespace *ns, struct ipc_params *params)
182{
183	struct msg_queue *msq;
184	int id, retval;
185	key_t key = params->key;
186	int msgflg = params->flg;
187
188	msq = ipc_rcu_alloc(sizeof(*msq));
189	if (!msq)
190		return -ENOMEM;
191
192	msq->q_perm.mode = msgflg & S_IRWXUGO;
193	msq->q_perm.key = key;
194
195	msq->q_perm.security = NULL;
196	retval = security_msg_queue_alloc(msq);
197	if (retval) {
198		ipc_rcu_putref(msq);
199		return retval;
200	}
201
202	/*
203	 * ipc_addid() locks msq
204	 */
205	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
206	if (id < 0) {
207		security_msg_queue_free(msq);
208		ipc_rcu_putref(msq);
209		return id;
210	}
211
212	msq->q_stime = msq->q_rtime = 0;
213	msq->q_ctime = get_seconds();
214	msq->q_cbytes = msq->q_qnum = 0;
215	msq->q_qbytes = ns->msg_ctlmnb;
216	msq->q_lspid = msq->q_lrpid = 0;
217	INIT_LIST_HEAD(&msq->q_messages);
218	INIT_LIST_HEAD(&msq->q_receivers);
219	INIT_LIST_HEAD(&msq->q_senders);
220
221	msg_unlock(msq);
222
223	return msq->q_perm.id;
224}
225
226static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
227{
228	mss->tsk = current;
229	current->state = TASK_INTERRUPTIBLE;
230	list_add_tail(&mss->list, &msq->q_senders);
231}
232
233static inline void ss_del(struct msg_sender *mss)
234{
235	if (mss->list.next != NULL)
236		list_del(&mss->list);
237}
238
239static void ss_wakeup(struct list_head *h, int kill)
240{
241	struct list_head *tmp;
242
243	tmp = h->next;
244	while (tmp != h) {
245		struct msg_sender *mss;
246
247		mss = list_entry(tmp, struct msg_sender, list);
248		tmp = tmp->next;
249		if (kill)
250			mss->list.next = NULL;
251		wake_up_process(mss->tsk);
252	}
253}
254
255static void expunge_all(struct msg_queue *msq, int res)
256{
257	struct list_head *tmp;
258
259	tmp = msq->q_receivers.next;
260	while (tmp != &msq->q_receivers) {
261		struct msg_receiver *msr;
262
263		msr = list_entry(tmp, struct msg_receiver, r_list);
264		tmp = tmp->next;
265		msr->r_msg = NULL;
266		wake_up_process(msr->r_tsk);
267		smp_mb();
268		msr->r_msg = ERR_PTR(res);
269	}
270}
271
272/*
273 * freeque() wakes up waiters on the sender and receiver waiting queue,
274 * removes the message queue from message queue ID IDR, and cleans up all the
275 * messages associated with this queue.
276 *
277 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
278 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
279 */
280static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
281{
282	struct list_head *tmp;
283	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
284
285	expunge_all(msq, -EIDRM);
286	ss_wakeup(&msq->q_senders, 1);
287	msg_rmid(ns, msq);
288	msg_unlock(msq);
289
290	tmp = msq->q_messages.next;
291	while (tmp != &msq->q_messages) {
292		struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
293
294		tmp = tmp->next;
295		atomic_dec(&ns->msg_hdrs);
296		free_msg(msg);
297	}
298	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
299	security_msg_queue_free(msq);
300	ipc_rcu_putref(msq);
301}
302
303/*
304 * Called with msg_ids.rw_mutex and ipcp locked.
305 */
306static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
307{
308	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
309
310	return security_msg_queue_associate(msq, msgflg);
311}
312
313SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
314{
315	struct ipc_namespace *ns;
316	struct ipc_ops msg_ops;
317	struct ipc_params msg_params;
318
319	ns = current->nsproxy->ipc_ns;
320
321	msg_ops.getnew = newque;
322	msg_ops.associate = msg_security;
323	msg_ops.more_checks = NULL;
324
325	msg_params.key = key;
326	msg_params.flg = msgflg;
327
328	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
329}
330
331static inline unsigned long
332copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
333{
334	switch(version) {
335	case IPC_64:
336		return copy_to_user(buf, in, sizeof(*in));
337	case IPC_OLD:
338	{
339		struct msqid_ds out;
340
341		memset(&out, 0, sizeof(out));
342
343		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
344
345		out.msg_stime		= in->msg_stime;
346		out.msg_rtime		= in->msg_rtime;
347		out.msg_ctime		= in->msg_ctime;
348
349		if (in->msg_cbytes > USHRT_MAX)
350			out.msg_cbytes	= USHRT_MAX;
351		else
352			out.msg_cbytes	= in->msg_cbytes;
353		out.msg_lcbytes		= in->msg_cbytes;
354
355		if (in->msg_qnum > USHRT_MAX)
356			out.msg_qnum	= USHRT_MAX;
357		else
358			out.msg_qnum	= in->msg_qnum;
359
360		if (in->msg_qbytes > USHRT_MAX)
361			out.msg_qbytes	= USHRT_MAX;
362		else
363			out.msg_qbytes	= in->msg_qbytes;
364		out.msg_lqbytes		= in->msg_qbytes;
365
366		out.msg_lspid		= in->msg_lspid;
367		out.msg_lrpid		= in->msg_lrpid;
368
369		return copy_to_user(buf, &out, sizeof(out));
370	}
371	default:
372		return -EINVAL;
373	}
374}
375
376static inline unsigned long
377copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
378{
379	switch(version) {
380	case IPC_64:
381		if (copy_from_user(out, buf, sizeof(*out)))
382			return -EFAULT;
383		return 0;
384	case IPC_OLD:
385	{
386		struct msqid_ds tbuf_old;
387
388		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
389			return -EFAULT;
390
391		out->msg_perm.uid      	= tbuf_old.msg_perm.uid;
392		out->msg_perm.gid      	= tbuf_old.msg_perm.gid;
393		out->msg_perm.mode     	= tbuf_old.msg_perm.mode;
394
395		if (tbuf_old.msg_qbytes == 0)
396			out->msg_qbytes	= tbuf_old.msg_lqbytes;
397		else
398			out->msg_qbytes	= tbuf_old.msg_qbytes;
399
400		return 0;
401	}
402	default:
403		return -EINVAL;
404	}
405}
406
407/*
408 * This function handles some msgctl commands which require the rw_mutex
409 * to be held in write mode.
410 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
411 */
412static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
413		       struct msqid_ds __user *buf, int version)
414{
415	struct kern_ipc_perm *ipcp;
416	struct msqid64_ds uninitialized_var(msqid64);
417	struct msg_queue *msq;
418	int err;
419
420	if (cmd == IPC_SET) {
421		if (copy_msqid_from_user(&msqid64, buf, version))
422			return -EFAULT;
423	}
424
425	ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
426			       &msqid64.msg_perm, msqid64.msg_qbytes);
427	if (IS_ERR(ipcp))
428		return PTR_ERR(ipcp);
429
430	msq = container_of(ipcp, struct msg_queue, q_perm);
431
432	err = security_msg_queue_msgctl(msq, cmd);
433	if (err)
434		goto out_unlock;
435
436	switch (cmd) {
437	case IPC_RMID:
438		freeque(ns, ipcp);
439		goto out_up;
440	case IPC_SET:
441		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
442		    !capable(CAP_SYS_RESOURCE)) {
443			err = -EPERM;
444			goto out_unlock;
445		}
446
447		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
448		if (err)
449			goto out_unlock;
450
451		msq->q_qbytes = msqid64.msg_qbytes;
452
453		msq->q_ctime = get_seconds();
454		/* sleeping receivers might be excluded by
455		 * stricter permissions.
456		 */
457		expunge_all(msq, -EAGAIN);
458		/* sleeping senders might be able to send
459		 * due to a larger queue size.
460		 */
461		ss_wakeup(&msq->q_senders, 0);
462		break;
463	default:
464		err = -EINVAL;
465	}
466out_unlock:
467	msg_unlock(msq);
468out_up:
469	up_write(&msg_ids(ns).rw_mutex);
470	return err;
471}
472
473SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
474{
475	struct msg_queue *msq;
476	int err, version;
477	struct ipc_namespace *ns;
478
479	if (msqid < 0 || cmd < 0)
480		return -EINVAL;
481
482	version = ipc_parse_version(&cmd);
483	ns = current->nsproxy->ipc_ns;
484
485	switch (cmd) {
486	case IPC_INFO:
487	case MSG_INFO:
488	{
489		struct msginfo msginfo;
490		int max_id;
491
492		if (!buf)
493			return -EFAULT;
494		/*
495		 * We must not return kernel stack data.
496		 * due to padding, it's not enough
497		 * to set all member fields.
498		 */
499		err = security_msg_queue_msgctl(NULL, cmd);
500		if (err)
501			return err;
502
503		memset(&msginfo, 0, sizeof(msginfo));
504		msginfo.msgmni = ns->msg_ctlmni;
505		msginfo.msgmax = ns->msg_ctlmax;
506		msginfo.msgmnb = ns->msg_ctlmnb;
507		msginfo.msgssz = MSGSSZ;
508		msginfo.msgseg = MSGSEG;
509		down_read(&msg_ids(ns).rw_mutex);
510		if (cmd == MSG_INFO) {
511			msginfo.msgpool = msg_ids(ns).in_use;
512			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
513			msginfo.msgtql = atomic_read(&ns->msg_bytes);
514		} else {
515			msginfo.msgmap = MSGMAP;
516			msginfo.msgpool = MSGPOOL;
517			msginfo.msgtql = MSGTQL;
518		}
519		max_id = ipc_get_maxid(&msg_ids(ns));
520		up_read(&msg_ids(ns).rw_mutex);
521		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
522			return -EFAULT;
523		return (max_id < 0) ? 0 : max_id;
524	}
525	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
526	case IPC_STAT:
527	{
528		struct msqid64_ds tbuf;
529		int success_return;
530
531		if (!buf)
532			return -EFAULT;
533
534		if (cmd == MSG_STAT) {
535			msq = msg_lock(ns, msqid);
536			if (IS_ERR(msq))
537				return PTR_ERR(msq);
538			success_return = msq->q_perm.id;
539		} else {
540			msq = msg_lock_check(ns, msqid);
541			if (IS_ERR(msq))
542				return PTR_ERR(msq);
543			success_return = 0;
544		}
545		err = -EACCES;
546		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
547			goto out_unlock;
548
549		err = security_msg_queue_msgctl(msq, cmd);
550		if (err)
551			goto out_unlock;
552
553		memset(&tbuf, 0, sizeof(tbuf));
554
555		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
556		tbuf.msg_stime  = msq->q_stime;
557		tbuf.msg_rtime  = msq->q_rtime;
558		tbuf.msg_ctime  = msq->q_ctime;
559		tbuf.msg_cbytes = msq->q_cbytes;
560		tbuf.msg_qnum   = msq->q_qnum;
561		tbuf.msg_qbytes = msq->q_qbytes;
562		tbuf.msg_lspid  = msq->q_lspid;
563		tbuf.msg_lrpid  = msq->q_lrpid;
564		msg_unlock(msq);
565		if (copy_msqid_to_user(buf, &tbuf, version))
566			return -EFAULT;
567		return success_return;
568	}
569	case IPC_SET:
570	case IPC_RMID:
571		err = msgctl_down(ns, msqid, cmd, buf, version);
572		return err;
573	default:
574		return  -EINVAL;
575	}
576
577out_unlock:
578	msg_unlock(msq);
579	return err;
580}
581
582static int testmsg(struct msg_msg *msg, long type, int mode)
583{
584	switch(mode)
585	{
586		case SEARCH_ANY:
587		case SEARCH_NUMBER:
588			return 1;
589		case SEARCH_LESSEQUAL:
590			if (msg->m_type <=type)
591				return 1;
592			break;
593		case SEARCH_EQUAL:
594			if (msg->m_type == type)
595				return 1;
596			break;
597		case SEARCH_NOTEQUAL:
598			if (msg->m_type != type)
599				return 1;
600			break;
601	}
602	return 0;
603}
604
605static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
606{
607	struct list_head *tmp;
608
609	tmp = msq->q_receivers.next;
610	while (tmp != &msq->q_receivers) {
611		struct msg_receiver *msr;
612
613		msr = list_entry(tmp, struct msg_receiver, r_list);
614		tmp = tmp->next;
615		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
616		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
617					       msr->r_msgtype, msr->r_mode)) {
618
619			list_del(&msr->r_list);
620			if (msr->r_maxsize < msg->m_ts) {
621				msr->r_msg = NULL;
622				wake_up_process(msr->r_tsk);
623				smp_mb();
624				msr->r_msg = ERR_PTR(-E2BIG);
625			} else {
626				msr->r_msg = NULL;
627				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
628				msq->q_rtime = get_seconds();
629				wake_up_process(msr->r_tsk);
630				smp_mb();
631				msr->r_msg = msg;
632
633				return 1;
634			}
635		}
636	}
637	return 0;
638}
639
640long do_msgsnd(int msqid, long mtype, void __user *mtext,
641		size_t msgsz, int msgflg)
642{
643	struct msg_queue *msq;
644	struct msg_msg *msg;
645	int err;
646	struct ipc_namespace *ns;
647
648	ns = current->nsproxy->ipc_ns;
649
650	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
651		return -EINVAL;
652	if (mtype < 1)
653		return -EINVAL;
654
655	msg = load_msg(mtext, msgsz);
656	if (IS_ERR(msg))
657		return PTR_ERR(msg);
658
659	msg->m_type = mtype;
660	msg->m_ts = msgsz;
661
662	msq = msg_lock_check(ns, msqid);
663	if (IS_ERR(msq)) {
664		err = PTR_ERR(msq);
665		goto out_free;
666	}
667
668	for (;;) {
669		struct msg_sender s;
670
671		err = -EACCES;
672		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
673			goto out_unlock_free;
674
675		err = security_msg_queue_msgsnd(msq, msg, msgflg);
676		if (err)
677			goto out_unlock_free;
678
679		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
680				1 + msq->q_qnum <= msq->q_qbytes) {
681			break;
682		}
683
684		/* queue full, wait: */
685		if (msgflg & IPC_NOWAIT) {
686			err = -EAGAIN;
687			goto out_unlock_free;
688		}
689		ss_add(msq, &s);
690		ipc_rcu_getref(msq);
691		msg_unlock(msq);
692		schedule();
693
694		ipc_lock_by_ptr(&msq->q_perm);
695		ipc_rcu_putref(msq);
696		if (msq->q_perm.deleted) {
697			err = -EIDRM;
698			goto out_unlock_free;
699		}
700		ss_del(&s);
701
702		if (signal_pending(current)) {
703			err = -ERESTARTNOHAND;
704			goto out_unlock_free;
705		}
706	}
707
708	msq->q_lspid = task_tgid_vnr(current);
709	msq->q_stime = get_seconds();
710
711	if (!pipelined_send(msq, msg)) {
712		/* no one is waiting for this message, enqueue it */
713		list_add_tail(&msg->m_list, &msq->q_messages);
714		msq->q_cbytes += msgsz;
715		msq->q_qnum++;
716		atomic_add(msgsz, &ns->msg_bytes);
717		atomic_inc(&ns->msg_hdrs);
718	}
719
720	err = 0;
721	msg = NULL;
722
723out_unlock_free:
724	msg_unlock(msq);
725out_free:
726	if (msg != NULL)
727		free_msg(msg);
728	return err;
729}
730
731SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
732		int, msgflg)
733{
734	long mtype;
735
736	if (get_user(mtype, &msgp->mtype))
737		return -EFAULT;
738	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
739}
740
741static inline int convert_mode(long *msgtyp, int msgflg)
742{
743	if (msgflg & MSG_COPY)
744		return SEARCH_NUMBER;
745	/*
746	 *  find message of correct type.
747	 *  msgtyp = 0 => get first.
748	 *  msgtyp > 0 => get first message of matching type.
749	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
750	 */
751	if (*msgtyp == 0)
752		return SEARCH_ANY;
753	if (*msgtyp < 0) {
754		*msgtyp = -*msgtyp;
755		return SEARCH_LESSEQUAL;
756	}
757	if (msgflg & MSG_EXCEPT)
758		return SEARCH_NOTEQUAL;
759	return SEARCH_EQUAL;
760}
761
762static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
763{
764	struct msgbuf __user *msgp = dest;
765	size_t msgsz;
766
767	if (put_user(msg->m_type, &msgp->mtype))
768		return -EFAULT;
769
770	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
771	if (store_msg(msgp->mtext, msg, msgsz))
772		return -EFAULT;
773	return msgsz;
774}
775
776#ifdef CONFIG_CHECKPOINT_RESTORE
777/*
778 * This function creates new kernel message structure, large enough to store
779 * bufsz message bytes.
780 */
781static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
782{
783	struct msg_msg *copy;
784
785	/*
786	 * Create dummy message to copy real message to.
787	 */
788	copy = load_msg(buf, bufsz);
789	if (!IS_ERR(copy))
790		copy->m_ts = bufsz;
791	return copy;
792}
793
794static inline void free_copy(struct msg_msg *copy)
795{
796	if (copy)
797		free_msg(copy);
798}
799#else
800static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
801{
802	return ERR_PTR(-ENOSYS);
803}
804
805static inline void free_copy(struct msg_msg *copy)
806{
807}
808#endif
809
810long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
811	       int msgflg,
812	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
813{
814	struct msg_queue *msq;
815	struct msg_msg *msg;
816	int mode;
817	struct ipc_namespace *ns;
818	struct msg_msg *copy = NULL;
819
820	ns = current->nsproxy->ipc_ns;
821
822	if (msqid < 0 || (long) bufsz < 0)
823		return -EINVAL;
824	if (msgflg & MSG_COPY) {
825		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
826		if (IS_ERR(copy))
827			return PTR_ERR(copy);
828	}
829	mode = convert_mode(&msgtyp, msgflg);
830
831	msq = msg_lock_check(ns, msqid);
832	if (IS_ERR(msq)) {
833		free_copy(copy);
834		return PTR_ERR(msq);
835	}
836
837	for (;;) {
838		struct msg_receiver msr_d;
839		struct msg_msg *walk_msg;
840		long msg_counter = 0;
841
842		msg = ERR_PTR(-EACCES);
843		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
844			goto out_unlock;
845
846		msg = ERR_PTR(-EAGAIN);
847		list_for_each_entry(walk_msg, &msq->q_messages, m_list) {
848
849			if (testmsg(walk_msg, msgtyp, mode) &&
850			    !security_msg_queue_msgrcv(msq, walk_msg, current,
851						       msgtyp, mode)) {
852
853				msg = walk_msg;
854				if (mode == SEARCH_LESSEQUAL &&
855						walk_msg->m_type != 1) {
856					msgtyp = walk_msg->m_type - 1;
857				} else if (mode == SEARCH_NUMBER) {
858					if (msgtyp == msg_counter)
859						break;
860					msg = ERR_PTR(-EAGAIN);
861				} else
862					break;
863				msg_counter++;
864			}
865		}
866		if (!IS_ERR(msg)) {
867			/*
868			 * Found a suitable message.
869			 * Unlink it from the queue.
870			 */
871			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
872				msg = ERR_PTR(-E2BIG);
873				goto out_unlock;
874			}
875			/*
876			 * If we are copying, then do not unlink message and do
877			 * not update queue parameters.
878			 */
879			if (msgflg & MSG_COPY) {
880				msg = copy_msg(msg, copy);
881				goto out_unlock;
882			}
883			list_del(&msg->m_list);
884			msq->q_qnum--;
885			msq->q_rtime = get_seconds();
886			msq->q_lrpid = task_tgid_vnr(current);
887			msq->q_cbytes -= msg->m_ts;
888			atomic_sub(msg->m_ts, &ns->msg_bytes);
889			atomic_dec(&ns->msg_hdrs);
890			ss_wakeup(&msq->q_senders, 0);
891			msg_unlock(msq);
892			break;
893		}
894		/* No message waiting. Wait for a message */
895		if (msgflg & IPC_NOWAIT) {
896			msg = ERR_PTR(-ENOMSG);
897			goto out_unlock;
898		}
899		list_add_tail(&msr_d.r_list, &msq->q_receivers);
900		msr_d.r_tsk = current;
901		msr_d.r_msgtype = msgtyp;
902		msr_d.r_mode = mode;
903		if (msgflg & MSG_NOERROR)
904			msr_d.r_maxsize = INT_MAX;
905		else
906			msr_d.r_maxsize = bufsz;
907		msr_d.r_msg = ERR_PTR(-EAGAIN);
908		current->state = TASK_INTERRUPTIBLE;
909		msg_unlock(msq);
910
911		schedule();
912
913		/* Lockless receive, part 1:
914		 * Disable preemption.  We don't hold a reference to the queue
915		 * and getting a reference would defeat the idea of a lockless
916		 * operation, thus the code relies on rcu to guarantee the
917		 * existence of msq:
918		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
919		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
920		 * rcu_read_lock() prevents preemption between reading r_msg
921		 * and the spin_lock() inside ipc_lock_by_ptr().
922		 */
923		rcu_read_lock();
924
925		/* Lockless receive, part 2:
926		 * Wait until pipelined_send or expunge_all are outside of
927		 * wake_up_process(). There is a race with exit(), see
928		 * ipc/mqueue.c for the details.
929		 */
930		msg = (struct msg_msg*)msr_d.r_msg;
931		while (msg == NULL) {
932			cpu_relax();
933			msg = (struct msg_msg *)msr_d.r_msg;
934		}
935
936		/* Lockless receive, part 3:
937		 * If there is a message or an error then accept it without
938		 * locking.
939		 */
940		if (msg != ERR_PTR(-EAGAIN)) {
941			rcu_read_unlock();
942			break;
943		}
944
945		/* Lockless receive, part 3:
946		 * Acquire the queue spinlock.
947		 */
948		ipc_lock_by_ptr(&msq->q_perm);
949		rcu_read_unlock();
950
951		/* Lockless receive, part 4:
952		 * Repeat test after acquiring the spinlock.
953		 */
954		msg = (struct msg_msg*)msr_d.r_msg;
955		if (msg != ERR_PTR(-EAGAIN))
956			goto out_unlock;
957
958		list_del(&msr_d.r_list);
959		if (signal_pending(current)) {
960			msg = ERR_PTR(-ERESTARTNOHAND);
961out_unlock:
962			msg_unlock(msq);
963			break;
964		}
965	}
966	if (IS_ERR(msg)) {
967		free_copy(copy);
968		return PTR_ERR(msg);
969	}
970
971	bufsz = msg_handler(buf, msg, bufsz);
972	free_msg(msg);
973
974	return bufsz;
975}
976
977SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
978		long, msgtyp, int, msgflg)
979{
980	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
981}
982
983#ifdef CONFIG_PROC_FS
984static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
985{
986	struct user_namespace *user_ns = seq_user_ns(s);
987	struct msg_queue *msq = it;
988
989	return seq_printf(s,
990			"%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
991			msq->q_perm.key,
992			msq->q_perm.id,
993			msq->q_perm.mode,
994			msq->q_cbytes,
995			msq->q_qnum,
996			msq->q_lspid,
997			msq->q_lrpid,
998			from_kuid_munged(user_ns, msq->q_perm.uid),
999			from_kgid_munged(user_ns, msq->q_perm.gid),
1000			from_kuid_munged(user_ns, msq->q_perm.cuid),
1001			from_kgid_munged(user_ns, msq->q_perm.cgid),
1002			msq->q_stime,
1003			msq->q_rtime,
1004			msq->q_ctime);
1005}
1006#endif
1007