msg.c revision 4718787d
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
73static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
74static int newque(struct ipc_namespace *, struct ipc_params *);
75#ifdef CONFIG_PROC_FS
76static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
77#endif
78
79/*
80 * Scale msgmni with the available lowmem size: the memory dedicated to msg
81 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
82 * Also take into account the number of nsproxies created so far.
83 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
84 */
85void recompute_msgmni(struct ipc_namespace *ns)
86{
87	struct sysinfo i;
88	unsigned long allowed;
89	int nb_ns;
90
91	si_meminfo(&i);
92	allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
93		/ MSGMNB;
94	nb_ns = atomic_read(&nr_ipc_ns);
95	allowed /= nb_ns;
96
97	if (allowed < MSGMNI) {
98		ns->msg_ctlmni = MSGMNI;
99		return;
100	}
101
102	if (allowed > IPCMNI / nb_ns) {
103		ns->msg_ctlmni = IPCMNI / nb_ns;
104		return;
105	}
106
107	ns->msg_ctlmni = allowed;
108}
109
110void msg_init_ns(struct ipc_namespace *ns)
111{
112	ns->msg_ctlmax = MSGMAX;
113	ns->msg_ctlmnb = MSGMNB;
114
115	recompute_msgmni(ns);
116
117	atomic_set(&ns->msg_bytes, 0);
118	atomic_set(&ns->msg_hdrs, 0);
119	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
120}
121
122#ifdef CONFIG_IPC_NS
123void msg_exit_ns(struct ipc_namespace *ns)
124{
125	free_ipcs(ns, &msg_ids(ns), freeque);
126	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
127}
128#endif
129
130void __init msg_init(void)
131{
132	msg_init_ns(&init_ipc_ns);
133
134	printk(KERN_INFO "msgmni has been set to %d\n",
135		init_ipc_ns.msg_ctlmni);
136
137	ipc_init_proc_interface("sysvipc/msg",
138				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
139				IPC_MSG_IDS, sysvipc_msg_proc_show);
140}
141
142static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
143{
144	struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
145
146	if (IS_ERR(ipcp))
147		return ERR_CAST(ipcp);
148
149	return container_of(ipcp, struct msg_queue, q_perm);
150}
151
152static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
153							int id)
154{
155	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
156
157	if (IS_ERR(ipcp))
158		return ERR_CAST(ipcp);
159
160	return container_of(ipcp, struct msg_queue, q_perm);
161}
162
163static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
164{
165	ipc_rmid(&msg_ids(ns), &s->q_perm);
166}
167
168/**
169 * newque - Create a new msg queue
170 * @ns: namespace
171 * @params: ptr to the structure that contains the key and msgflg
172 *
173 * Called with msg_ids.rwsem held (writer)
174 */
175static int newque(struct ipc_namespace *ns, struct ipc_params *params)
176{
177	struct msg_queue *msq;
178	int id, retval;
179	key_t key = params->key;
180	int msgflg = params->flg;
181
182	msq = ipc_rcu_alloc(sizeof(*msq));
183	if (!msq)
184		return -ENOMEM;
185
186	msq->q_perm.mode = msgflg & S_IRWXUGO;
187	msq->q_perm.key = key;
188
189	msq->q_perm.security = NULL;
190	retval = security_msg_queue_alloc(msq);
191	if (retval) {
192		ipc_rcu_putref(msq);
193		return retval;
194	}
195
196	/* ipc_addid() locks msq upon success. */
197	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
198	if (id < 0) {
199		security_msg_queue_free(msq);
200		ipc_rcu_putref(msq);
201		return id;
202	}
203
204	msq->q_stime = msq->q_rtime = 0;
205	msq->q_ctime = get_seconds();
206	msq->q_cbytes = msq->q_qnum = 0;
207	msq->q_qbytes = ns->msg_ctlmnb;
208	msq->q_lspid = msq->q_lrpid = 0;
209	INIT_LIST_HEAD(&msq->q_messages);
210	INIT_LIST_HEAD(&msq->q_receivers);
211	INIT_LIST_HEAD(&msq->q_senders);
212
213	ipc_unlock_object(&msq->q_perm);
214	rcu_read_unlock();
215
216	return msq->q_perm.id;
217}
218
219static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
220{
221	mss->tsk = current;
222	current->state = TASK_INTERRUPTIBLE;
223	list_add_tail(&mss->list, &msq->q_senders);
224}
225
226static inline void ss_del(struct msg_sender *mss)
227{
228	if (mss->list.next != NULL)
229		list_del(&mss->list);
230}
231
232static void ss_wakeup(struct list_head *h, int kill)
233{
234	struct msg_sender *mss, *t;
235
236	list_for_each_entry_safe(mss, t, h, list) {
237		if (kill)
238			mss->list.next = NULL;
239		wake_up_process(mss->tsk);
240	}
241}
242
243static void expunge_all(struct msg_queue *msq, int res)
244{
245	struct msg_receiver *msr, *t;
246
247	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
248		msr->r_msg = NULL;
249		wake_up_process(msr->r_tsk);
250		smp_mb();
251		msr->r_msg = ERR_PTR(res);
252	}
253}
254
255/*
256 * freeque() wakes up waiters on the sender and receiver waiting queue,
257 * removes the message queue from message queue ID IDR, and cleans up all the
258 * messages associated with this queue.
259 *
260 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
261 * before freeque() is called. msg_ids.rwsem remains locked on exit.
262 */
263static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
264{
265	struct msg_msg *msg, *t;
266	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
267
268	expunge_all(msq, -EIDRM);
269	ss_wakeup(&msq->q_senders, 1);
270	msg_rmid(ns, msq);
271	ipc_unlock_object(&msq->q_perm);
272	rcu_read_unlock();
273
274	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
275		atomic_dec(&ns->msg_hdrs);
276		free_msg(msg);
277	}
278	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
279	security_msg_queue_free(msq);
280	ipc_rcu_putref(msq);
281}
282
283/*
284 * Called with msg_ids.rwsem and ipcp locked.
285 */
286static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
287{
288	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
289
290	return security_msg_queue_associate(msq, msgflg);
291}
292
293SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
294{
295	struct ipc_namespace *ns;
296	struct ipc_ops msg_ops;
297	struct ipc_params msg_params;
298
299	ns = current->nsproxy->ipc_ns;
300
301	msg_ops.getnew = newque;
302	msg_ops.associate = msg_security;
303	msg_ops.more_checks = NULL;
304
305	msg_params.key = key;
306	msg_params.flg = msgflg;
307
308	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
309}
310
311static inline unsigned long
312copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
313{
314	switch(version) {
315	case IPC_64:
316		return copy_to_user(buf, in, sizeof(*in));
317	case IPC_OLD:
318	{
319		struct msqid_ds out;
320
321		memset(&out, 0, sizeof(out));
322
323		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
324
325		out.msg_stime		= in->msg_stime;
326		out.msg_rtime		= in->msg_rtime;
327		out.msg_ctime		= in->msg_ctime;
328
329		if (in->msg_cbytes > USHRT_MAX)
330			out.msg_cbytes	= USHRT_MAX;
331		else
332			out.msg_cbytes	= in->msg_cbytes;
333		out.msg_lcbytes		= in->msg_cbytes;
334
335		if (in->msg_qnum > USHRT_MAX)
336			out.msg_qnum	= USHRT_MAX;
337		else
338			out.msg_qnum	= in->msg_qnum;
339
340		if (in->msg_qbytes > USHRT_MAX)
341			out.msg_qbytes	= USHRT_MAX;
342		else
343			out.msg_qbytes	= in->msg_qbytes;
344		out.msg_lqbytes		= in->msg_qbytes;
345
346		out.msg_lspid		= in->msg_lspid;
347		out.msg_lrpid		= in->msg_lrpid;
348
349		return copy_to_user(buf, &out, sizeof(out));
350	}
351	default:
352		return -EINVAL;
353	}
354}
355
356static inline unsigned long
357copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
358{
359	switch(version) {
360	case IPC_64:
361		if (copy_from_user(out, buf, sizeof(*out)))
362			return -EFAULT;
363		return 0;
364	case IPC_OLD:
365	{
366		struct msqid_ds tbuf_old;
367
368		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
369			return -EFAULT;
370
371		out->msg_perm.uid      	= tbuf_old.msg_perm.uid;
372		out->msg_perm.gid      	= tbuf_old.msg_perm.gid;
373		out->msg_perm.mode     	= tbuf_old.msg_perm.mode;
374
375		if (tbuf_old.msg_qbytes == 0)
376			out->msg_qbytes	= tbuf_old.msg_lqbytes;
377		else
378			out->msg_qbytes	= tbuf_old.msg_qbytes;
379
380		return 0;
381	}
382	default:
383		return -EINVAL;
384	}
385}
386
387/*
388 * This function handles some msgctl commands which require the rwsem
389 * to be held in write mode.
390 * NOTE: no locks must be held, the rwsem is taken inside this function.
391 */
392static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
393		       struct msqid_ds __user *buf, int version)
394{
395	struct kern_ipc_perm *ipcp;
396	struct msqid64_ds uninitialized_var(msqid64);
397	struct msg_queue *msq;
398	int err;
399
400	if (cmd == IPC_SET) {
401		if (copy_msqid_from_user(&msqid64, buf, version))
402			return -EFAULT;
403	}
404
405	down_write(&msg_ids(ns).rwsem);
406	rcu_read_lock();
407
408	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
409				      &msqid64.msg_perm, msqid64.msg_qbytes);
410	if (IS_ERR(ipcp)) {
411		err = PTR_ERR(ipcp);
412		goto out_unlock1;
413	}
414
415	msq = container_of(ipcp, struct msg_queue, q_perm);
416
417	err = security_msg_queue_msgctl(msq, cmd);
418	if (err)
419		goto out_unlock1;
420
421	switch (cmd) {
422	case IPC_RMID:
423		ipc_lock_object(&msq->q_perm);
424		/* freeque unlocks the ipc object and rcu */
425		freeque(ns, ipcp);
426		goto out_up;
427	case IPC_SET:
428		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
429		    !capable(CAP_SYS_RESOURCE)) {
430			err = -EPERM;
431			goto out_unlock1;
432		}
433
434		ipc_lock_object(&msq->q_perm);
435		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
436		if (err)
437			goto out_unlock0;
438
439		msq->q_qbytes = msqid64.msg_qbytes;
440
441		msq->q_ctime = get_seconds();
442		/* sleeping receivers might be excluded by
443		 * stricter permissions.
444		 */
445		expunge_all(msq, -EAGAIN);
446		/* sleeping senders might be able to send
447		 * due to a larger queue size.
448		 */
449		ss_wakeup(&msq->q_senders, 0);
450		break;
451	default:
452		err = -EINVAL;
453		goto out_unlock1;
454	}
455
456out_unlock0:
457	ipc_unlock_object(&msq->q_perm);
458out_unlock1:
459	rcu_read_unlock();
460out_up:
461	up_write(&msg_ids(ns).rwsem);
462	return err;
463}
464
465static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
466			 int cmd, int version, void __user *buf)
467{
468	int err;
469	struct msg_queue *msq;
470
471	switch (cmd) {
472	case IPC_INFO:
473	case MSG_INFO:
474	{
475		struct msginfo msginfo;
476		int max_id;
477
478		if (!buf)
479			return -EFAULT;
480
481		/*
482		 * We must not return kernel stack data.
483		 * due to padding, it's not enough
484		 * to set all member fields.
485		 */
486		err = security_msg_queue_msgctl(NULL, cmd);
487		if (err)
488			return err;
489
490		memset(&msginfo, 0, sizeof(msginfo));
491		msginfo.msgmni = ns->msg_ctlmni;
492		msginfo.msgmax = ns->msg_ctlmax;
493		msginfo.msgmnb = ns->msg_ctlmnb;
494		msginfo.msgssz = MSGSSZ;
495		msginfo.msgseg = MSGSEG;
496		down_read(&msg_ids(ns).rwsem);
497		if (cmd == MSG_INFO) {
498			msginfo.msgpool = msg_ids(ns).in_use;
499			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
500			msginfo.msgtql = atomic_read(&ns->msg_bytes);
501		} else {
502			msginfo.msgmap = MSGMAP;
503			msginfo.msgpool = MSGPOOL;
504			msginfo.msgtql = MSGTQL;
505		}
506		max_id = ipc_get_maxid(&msg_ids(ns));
507		up_read(&msg_ids(ns).rwsem);
508		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
509			return -EFAULT;
510		return (max_id < 0) ? 0 : max_id;
511	}
512
513	case MSG_STAT:
514	case IPC_STAT:
515	{
516		struct msqid64_ds tbuf;
517		int success_return;
518
519		if (!buf)
520			return -EFAULT;
521
522		memset(&tbuf, 0, sizeof(tbuf));
523
524		rcu_read_lock();
525		if (cmd == MSG_STAT) {
526			msq = msq_obtain_object(ns, msqid);
527			if (IS_ERR(msq)) {
528				err = PTR_ERR(msq);
529				goto out_unlock;
530			}
531			success_return = msq->q_perm.id;
532		} else {
533			msq = msq_obtain_object_check(ns, msqid);
534			if (IS_ERR(msq)) {
535				err = PTR_ERR(msq);
536				goto out_unlock;
537			}
538			success_return = 0;
539		}
540
541		err = -EACCES;
542		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
543			goto out_unlock;
544
545		err = security_msg_queue_msgctl(msq, cmd);
546		if (err)
547			goto out_unlock;
548
549		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
550		tbuf.msg_stime  = msq->q_stime;
551		tbuf.msg_rtime  = msq->q_rtime;
552		tbuf.msg_ctime  = msq->q_ctime;
553		tbuf.msg_cbytes = msq->q_cbytes;
554		tbuf.msg_qnum   = msq->q_qnum;
555		tbuf.msg_qbytes = msq->q_qbytes;
556		tbuf.msg_lspid  = msq->q_lspid;
557		tbuf.msg_lrpid  = msq->q_lrpid;
558		rcu_read_unlock();
559
560		if (copy_msqid_to_user(buf, &tbuf, version))
561			return -EFAULT;
562		return success_return;
563	}
564
565	default:
566		return -EINVAL;
567	}
568
569	return err;
570out_unlock:
571	rcu_read_unlock();
572	return err;
573}
574
575SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
576{
577	int version;
578	struct ipc_namespace *ns;
579
580	if (msqid < 0 || cmd < 0)
581		return -EINVAL;
582
583	version = ipc_parse_version(&cmd);
584	ns = current->nsproxy->ipc_ns;
585
586	switch (cmd) {
587	case IPC_INFO:
588	case MSG_INFO:
589	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
590	case IPC_STAT:
591		return msgctl_nolock(ns, msqid, cmd, version, buf);
592	case IPC_SET:
593	case IPC_RMID:
594		return msgctl_down(ns, msqid, cmd, buf, version);
595	default:
596		return  -EINVAL;
597	}
598}
599
600static int testmsg(struct msg_msg *msg, long type, int mode)
601{
602	switch(mode)
603	{
604		case SEARCH_ANY:
605		case SEARCH_NUMBER:
606			return 1;
607		case SEARCH_LESSEQUAL:
608			if (msg->m_type <=type)
609				return 1;
610			break;
611		case SEARCH_EQUAL:
612			if (msg->m_type == type)
613				return 1;
614			break;
615		case SEARCH_NOTEQUAL:
616			if (msg->m_type != type)
617				return 1;
618			break;
619	}
620	return 0;
621}
622
623static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
624{
625	struct msg_receiver *msr, *t;
626
627	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
628		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
629		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
630					       msr->r_msgtype, msr->r_mode)) {
631
632			list_del(&msr->r_list);
633			if (msr->r_maxsize < msg->m_ts) {
634				msr->r_msg = NULL;
635				wake_up_process(msr->r_tsk);
636				smp_mb();
637				msr->r_msg = ERR_PTR(-E2BIG);
638			} else {
639				msr->r_msg = NULL;
640				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
641				msq->q_rtime = get_seconds();
642				wake_up_process(msr->r_tsk);
643				smp_mb();
644				msr->r_msg = msg;
645
646				return 1;
647			}
648		}
649	}
650	return 0;
651}
652
653long do_msgsnd(int msqid, long mtype, void __user *mtext,
654		size_t msgsz, int msgflg)
655{
656	struct msg_queue *msq;
657	struct msg_msg *msg;
658	int err;
659	struct ipc_namespace *ns;
660
661	ns = current->nsproxy->ipc_ns;
662
663	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
664		return -EINVAL;
665	if (mtype < 1)
666		return -EINVAL;
667
668	msg = load_msg(mtext, msgsz);
669	if (IS_ERR(msg))
670		return PTR_ERR(msg);
671
672	msg->m_type = mtype;
673	msg->m_ts = msgsz;
674
675	rcu_read_lock();
676	msq = msq_obtain_object_check(ns, msqid);
677	if (IS_ERR(msq)) {
678		err = PTR_ERR(msq);
679		goto out_unlock1;
680	}
681
682	ipc_lock_object(&msq->q_perm);
683
684	for (;;) {
685		struct msg_sender s;
686
687		err = -EACCES;
688		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
689			goto out_unlock0;
690
691		err = security_msg_queue_msgsnd(msq, msg, msgflg);
692		if (err)
693			goto out_unlock0;
694
695		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
696				1 + msq->q_qnum <= msq->q_qbytes) {
697			break;
698		}
699
700		/* queue full, wait: */
701		if (msgflg & IPC_NOWAIT) {
702			err = -EAGAIN;
703			goto out_unlock0;
704		}
705
706		ss_add(msq, &s);
707
708		if (!ipc_rcu_getref(msq)) {
709			err = -EIDRM;
710			goto out_unlock0;
711		}
712
713		ipc_unlock_object(&msq->q_perm);
714		rcu_read_unlock();
715		schedule();
716
717		rcu_read_lock();
718		ipc_lock_object(&msq->q_perm);
719
720		ipc_rcu_putref(msq);
721		if (msq->q_perm.deleted) {
722			err = -EIDRM;
723			goto out_unlock0;
724		}
725
726		ss_del(&s);
727
728		if (signal_pending(current)) {
729			err = -ERESTARTNOHAND;
730			goto out_unlock0;
731		}
732
733	}
734	msq->q_lspid = task_tgid_vnr(current);
735	msq->q_stime = get_seconds();
736
737	if (!pipelined_send(msq, msg)) {
738		/* no one is waiting for this message, enqueue it */
739		list_add_tail(&msg->m_list, &msq->q_messages);
740		msq->q_cbytes += msgsz;
741		msq->q_qnum++;
742		atomic_add(msgsz, &ns->msg_bytes);
743		atomic_inc(&ns->msg_hdrs);
744	}
745
746	err = 0;
747	msg = NULL;
748
749out_unlock0:
750	ipc_unlock_object(&msq->q_perm);
751out_unlock1:
752	rcu_read_unlock();
753	if (msg != NULL)
754		free_msg(msg);
755	return err;
756}
757
758SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
759		int, msgflg)
760{
761	long mtype;
762
763	if (get_user(mtype, &msgp->mtype))
764		return -EFAULT;
765	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
766}
767
768static inline int convert_mode(long *msgtyp, int msgflg)
769{
770	if (msgflg & MSG_COPY)
771		return SEARCH_NUMBER;
772	/*
773	 *  find message of correct type.
774	 *  msgtyp = 0 => get first.
775	 *  msgtyp > 0 => get first message of matching type.
776	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
777	 */
778	if (*msgtyp == 0)
779		return SEARCH_ANY;
780	if (*msgtyp < 0) {
781		*msgtyp = -*msgtyp;
782		return SEARCH_LESSEQUAL;
783	}
784	if (msgflg & MSG_EXCEPT)
785		return SEARCH_NOTEQUAL;
786	return SEARCH_EQUAL;
787}
788
789static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
790{
791	struct msgbuf __user *msgp = dest;
792	size_t msgsz;
793
794	if (put_user(msg->m_type, &msgp->mtype))
795		return -EFAULT;
796
797	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
798	if (store_msg(msgp->mtext, msg, msgsz))
799		return -EFAULT;
800	return msgsz;
801}
802
803#ifdef CONFIG_CHECKPOINT_RESTORE
804/*
805 * This function creates new kernel message structure, large enough to store
806 * bufsz message bytes.
807 */
808static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
809{
810	struct msg_msg *copy;
811
812	/*
813	 * Create dummy message to copy real message to.
814	 */
815	copy = load_msg(buf, bufsz);
816	if (!IS_ERR(copy))
817		copy->m_ts = bufsz;
818	return copy;
819}
820
821static inline void free_copy(struct msg_msg *copy)
822{
823	if (copy)
824		free_msg(copy);
825}
826#else
827static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
828{
829	return ERR_PTR(-ENOSYS);
830}
831
832static inline void free_copy(struct msg_msg *copy)
833{
834}
835#endif
836
837static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
838{
839	struct msg_msg *msg, *found = NULL;
840	long count = 0;
841
842	list_for_each_entry(msg, &msq->q_messages, m_list) {
843		if (testmsg(msg, *msgtyp, mode) &&
844		    !security_msg_queue_msgrcv(msq, msg, current,
845					       *msgtyp, mode)) {
846			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
847				*msgtyp = msg->m_type - 1;
848				found = msg;
849			} else if (mode == SEARCH_NUMBER) {
850				if (*msgtyp == count)
851					return msg;
852			} else
853				return msg;
854			count++;
855		}
856	}
857
858	return found ?: ERR_PTR(-EAGAIN);
859}
860
861long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
862	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
863{
864	int mode;
865	struct msg_queue *msq;
866	struct ipc_namespace *ns;
867	struct msg_msg *msg, *copy = NULL;
868
869	ns = current->nsproxy->ipc_ns;
870
871	if (msqid < 0 || (long) bufsz < 0)
872		return -EINVAL;
873
874	if (msgflg & MSG_COPY) {
875		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
876		if (IS_ERR(copy))
877			return PTR_ERR(copy);
878	}
879	mode = convert_mode(&msgtyp, msgflg);
880
881	rcu_read_lock();
882	msq = msq_obtain_object_check(ns, msqid);
883	if (IS_ERR(msq)) {
884		rcu_read_unlock();
885		free_copy(copy);
886		return PTR_ERR(msq);
887	}
888
889	for (;;) {
890		struct msg_receiver msr_d;
891
892		msg = ERR_PTR(-EACCES);
893		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
894			goto out_unlock1;
895
896		ipc_lock_object(&msq->q_perm);
897		msg = find_msg(msq, &msgtyp, mode);
898		if (!IS_ERR(msg)) {
899			/*
900			 * Found a suitable message.
901			 * Unlink it from the queue.
902			 */
903			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
904				msg = ERR_PTR(-E2BIG);
905				goto out_unlock0;
906			}
907			/*
908			 * If we are copying, then do not unlink message and do
909			 * not update queue parameters.
910			 */
911			if (msgflg & MSG_COPY) {
912				msg = copy_msg(msg, copy);
913				goto out_unlock0;
914			}
915
916			list_del(&msg->m_list);
917			msq->q_qnum--;
918			msq->q_rtime = get_seconds();
919			msq->q_lrpid = task_tgid_vnr(current);
920			msq->q_cbytes -= msg->m_ts;
921			atomic_sub(msg->m_ts, &ns->msg_bytes);
922			atomic_dec(&ns->msg_hdrs);
923			ss_wakeup(&msq->q_senders, 0);
924
925			goto out_unlock0;
926		}
927
928		/* No message waiting. Wait for a message */
929		if (msgflg & IPC_NOWAIT) {
930			msg = ERR_PTR(-ENOMSG);
931			goto out_unlock0;
932		}
933
934		list_add_tail(&msr_d.r_list, &msq->q_receivers);
935		msr_d.r_tsk = current;
936		msr_d.r_msgtype = msgtyp;
937		msr_d.r_mode = mode;
938		if (msgflg & MSG_NOERROR)
939			msr_d.r_maxsize = INT_MAX;
940		else
941			msr_d.r_maxsize = bufsz;
942		msr_d.r_msg = ERR_PTR(-EAGAIN);
943		current->state = TASK_INTERRUPTIBLE;
944
945		ipc_unlock_object(&msq->q_perm);
946		rcu_read_unlock();
947		schedule();
948
949		/* Lockless receive, part 1:
950		 * Disable preemption.  We don't hold a reference to the queue
951		 * and getting a reference would defeat the idea of a lockless
952		 * operation, thus the code relies on rcu to guarantee the
953		 * existence of msq:
954		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
955		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
956		 * rcu_read_lock() prevents preemption between reading r_msg
957		 * and acquiring the q_perm.lock in ipc_lock_object().
958		 */
959		rcu_read_lock();
960
961		/* Lockless receive, part 2:
962		 * Wait until pipelined_send or expunge_all are outside of
963		 * wake_up_process(). There is a race with exit(), see
964		 * ipc/mqueue.c for the details.
965		 */
966		msg = (struct msg_msg*)msr_d.r_msg;
967		while (msg == NULL) {
968			cpu_relax();
969			msg = (struct msg_msg *)msr_d.r_msg;
970		}
971
972		/* Lockless receive, part 3:
973		 * If there is a message or an error then accept it without
974		 * locking.
975		 */
976		if (msg != ERR_PTR(-EAGAIN))
977			goto out_unlock1;
978
979		/* Lockless receive, part 3:
980		 * Acquire the queue spinlock.
981		 */
982		ipc_lock_object(&msq->q_perm);
983
984		/* Lockless receive, part 4:
985		 * Repeat test after acquiring the spinlock.
986		 */
987		msg = (struct msg_msg*)msr_d.r_msg;
988		if (msg != ERR_PTR(-EAGAIN))
989			goto out_unlock0;
990
991		list_del(&msr_d.r_list);
992		if (signal_pending(current)) {
993			msg = ERR_PTR(-ERESTARTNOHAND);
994			goto out_unlock0;
995		}
996
997		ipc_unlock_object(&msq->q_perm);
998	}
999
1000out_unlock0:
1001	ipc_unlock_object(&msq->q_perm);
1002out_unlock1:
1003	rcu_read_unlock();
1004	if (IS_ERR(msg)) {
1005		free_copy(copy);
1006		return PTR_ERR(msg);
1007	}
1008
1009	bufsz = msg_handler(buf, msg, bufsz);
1010	free_msg(msg);
1011
1012	return bufsz;
1013}
1014
1015SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1016		long, msgtyp, int, msgflg)
1017{
1018	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1019}
1020
1021#ifdef CONFIG_PROC_FS
1022static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1023{
1024	struct user_namespace *user_ns = seq_user_ns(s);
1025	struct msg_queue *msq = it;
1026
1027	return seq_printf(s,
1028			"%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1029			msq->q_perm.key,
1030			msq->q_perm.id,
1031			msq->q_perm.mode,
1032			msq->q_cbytes,
1033			msq->q_qnum,
1034			msq->q_lspid,
1035			msq->q_lrpid,
1036			from_kuid_munged(user_ns, msq->q_perm.uid),
1037			from_kgid_munged(user_ns, msq->q_perm.gid),
1038			from_kuid_munged(user_ns, msq->q_perm.cuid),
1039			from_kgid_munged(user_ns, msq->q_perm.cgid),
1040			msq->q_stime,
1041			msq->q_rtime,
1042			msq->q_ctime);
1043}
1044#endif
1045