msg.c revision 4271b05a
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
168static void msg_rcu_free(struct rcu_head *head)
169{
170	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
171	struct msg_queue *msq = ipc_rcu_to_struct(p);
172
173	security_msg_queue_free(msq);
174	ipc_rcu_free(head);
175}
176
177/**
178 * newque - Create a new msg queue
179 * @ns: namespace
180 * @params: ptr to the structure that contains the key and msgflg
181 *
182 * Called with msg_ids.rwsem held (writer)
183 */
184static int newque(struct ipc_namespace *ns, struct ipc_params *params)
185{
186	struct msg_queue *msq;
187	int id, retval;
188	key_t key = params->key;
189	int msgflg = params->flg;
190
191	msq = ipc_rcu_alloc(sizeof(*msq));
192	if (!msq)
193		return -ENOMEM;
194
195	msq->q_perm.mode = msgflg & S_IRWXUGO;
196	msq->q_perm.key = key;
197
198	msq->q_perm.security = NULL;
199	retval = security_msg_queue_alloc(msq);
200	if (retval) {
201		ipc_rcu_putref(msq, ipc_rcu_free);
202		return retval;
203	}
204
205	/* ipc_addid() locks msq upon success. */
206	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
207	if (id < 0) {
208		ipc_rcu_putref(msq, msg_rcu_free);
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	ipc_unlock_object(&msq->q_perm);
222	rcu_read_unlock();
223
224	return msq->q_perm.id;
225}
226
227static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
228{
229	mss->tsk = current;
230	current->state = TASK_INTERRUPTIBLE;
231	list_add_tail(&mss->list, &msq->q_senders);
232}
233
234static inline void ss_del(struct msg_sender *mss)
235{
236	if (mss->list.next != NULL)
237		list_del(&mss->list);
238}
239
240static void ss_wakeup(struct list_head *h, int kill)
241{
242	struct msg_sender *mss, *t;
243
244	list_for_each_entry_safe(mss, t, h, list) {
245		if (kill)
246			mss->list.next = NULL;
247		wake_up_process(mss->tsk);
248	}
249}
250
251static void expunge_all(struct msg_queue *msq, int res)
252{
253	struct msg_receiver *msr, *t;
254
255	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
256		msr->r_msg = NULL;
257		wake_up_process(msr->r_tsk);
258		smp_mb();
259		msr->r_msg = ERR_PTR(res);
260	}
261}
262
263/*
264 * freeque() wakes up waiters on the sender and receiver waiting queue,
265 * removes the message queue from message queue ID IDR, and cleans up all the
266 * messages associated with this queue.
267 *
268 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
269 * before freeque() is called. msg_ids.rwsem remains locked on exit.
270 */
271static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
272{
273	struct msg_msg *msg, *t;
274	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
275
276	expunge_all(msq, -EIDRM);
277	ss_wakeup(&msq->q_senders, 1);
278	msg_rmid(ns, msq);
279	ipc_unlock_object(&msq->q_perm);
280	rcu_read_unlock();
281
282	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
283		atomic_dec(&ns->msg_hdrs);
284		free_msg(msg);
285	}
286	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
287	ipc_rcu_putref(msq, msg_rcu_free);
288}
289
290/*
291 * Called with msg_ids.rwsem and ipcp locked.
292 */
293static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
294{
295	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
296
297	return security_msg_queue_associate(msq, msgflg);
298}
299
300SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
301{
302	struct ipc_namespace *ns;
303	struct ipc_ops msg_ops;
304	struct ipc_params msg_params;
305
306	ns = current->nsproxy->ipc_ns;
307
308	msg_ops.getnew = newque;
309	msg_ops.associate = msg_security;
310	msg_ops.more_checks = NULL;
311
312	msg_params.key = key;
313	msg_params.flg = msgflg;
314
315	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
316}
317
318static inline unsigned long
319copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
320{
321	switch(version) {
322	case IPC_64:
323		return copy_to_user(buf, in, sizeof(*in));
324	case IPC_OLD:
325	{
326		struct msqid_ds out;
327
328		memset(&out, 0, sizeof(out));
329
330		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
331
332		out.msg_stime		= in->msg_stime;
333		out.msg_rtime		= in->msg_rtime;
334		out.msg_ctime		= in->msg_ctime;
335
336		if (in->msg_cbytes > USHRT_MAX)
337			out.msg_cbytes	= USHRT_MAX;
338		else
339			out.msg_cbytes	= in->msg_cbytes;
340		out.msg_lcbytes		= in->msg_cbytes;
341
342		if (in->msg_qnum > USHRT_MAX)
343			out.msg_qnum	= USHRT_MAX;
344		else
345			out.msg_qnum	= in->msg_qnum;
346
347		if (in->msg_qbytes > USHRT_MAX)
348			out.msg_qbytes	= USHRT_MAX;
349		else
350			out.msg_qbytes	= in->msg_qbytes;
351		out.msg_lqbytes		= in->msg_qbytes;
352
353		out.msg_lspid		= in->msg_lspid;
354		out.msg_lrpid		= in->msg_lrpid;
355
356		return copy_to_user(buf, &out, sizeof(out));
357	}
358	default:
359		return -EINVAL;
360	}
361}
362
363static inline unsigned long
364copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
365{
366	switch(version) {
367	case IPC_64:
368		if (copy_from_user(out, buf, sizeof(*out)))
369			return -EFAULT;
370		return 0;
371	case IPC_OLD:
372	{
373		struct msqid_ds tbuf_old;
374
375		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
376			return -EFAULT;
377
378		out->msg_perm.uid      	= tbuf_old.msg_perm.uid;
379		out->msg_perm.gid      	= tbuf_old.msg_perm.gid;
380		out->msg_perm.mode     	= tbuf_old.msg_perm.mode;
381
382		if (tbuf_old.msg_qbytes == 0)
383			out->msg_qbytes	= tbuf_old.msg_lqbytes;
384		else
385			out->msg_qbytes	= tbuf_old.msg_qbytes;
386
387		return 0;
388	}
389	default:
390		return -EINVAL;
391	}
392}
393
394/*
395 * This function handles some msgctl commands which require the rwsem
396 * to be held in write mode.
397 * NOTE: no locks must be held, the rwsem is taken inside this function.
398 */
399static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
400		       struct msqid_ds __user *buf, int version)
401{
402	struct kern_ipc_perm *ipcp;
403	struct msqid64_ds uninitialized_var(msqid64);
404	struct msg_queue *msq;
405	int err;
406
407	if (cmd == IPC_SET) {
408		if (copy_msqid_from_user(&msqid64, buf, version))
409			return -EFAULT;
410	}
411
412	down_write(&msg_ids(ns).rwsem);
413	rcu_read_lock();
414
415	ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
416				      &msqid64.msg_perm, msqid64.msg_qbytes);
417	if (IS_ERR(ipcp)) {
418		err = PTR_ERR(ipcp);
419		goto out_unlock1;
420	}
421
422	msq = container_of(ipcp, struct msg_queue, q_perm);
423
424	err = security_msg_queue_msgctl(msq, cmd);
425	if (err)
426		goto out_unlock1;
427
428	switch (cmd) {
429	case IPC_RMID:
430		ipc_lock_object(&msq->q_perm);
431		/* freeque unlocks the ipc object and rcu */
432		freeque(ns, ipcp);
433		goto out_up;
434	case IPC_SET:
435		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
436		    !capable(CAP_SYS_RESOURCE)) {
437			err = -EPERM;
438			goto out_unlock1;
439		}
440
441		ipc_lock_object(&msq->q_perm);
442		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
443		if (err)
444			goto out_unlock0;
445
446		msq->q_qbytes = msqid64.msg_qbytes;
447
448		msq->q_ctime = get_seconds();
449		/* sleeping receivers might be excluded by
450		 * stricter permissions.
451		 */
452		expunge_all(msq, -EAGAIN);
453		/* sleeping senders might be able to send
454		 * due to a larger queue size.
455		 */
456		ss_wakeup(&msq->q_senders, 0);
457		break;
458	default:
459		err = -EINVAL;
460		goto out_unlock1;
461	}
462
463out_unlock0:
464	ipc_unlock_object(&msq->q_perm);
465out_unlock1:
466	rcu_read_unlock();
467out_up:
468	up_write(&msg_ids(ns).rwsem);
469	return err;
470}
471
472static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
473			 int cmd, int version, void __user *buf)
474{
475	int err;
476	struct msg_queue *msq;
477
478	switch (cmd) {
479	case IPC_INFO:
480	case MSG_INFO:
481	{
482		struct msginfo msginfo;
483		int max_id;
484
485		if (!buf)
486			return -EFAULT;
487
488		/*
489		 * We must not return kernel stack data.
490		 * due to padding, it's not enough
491		 * to set all member fields.
492		 */
493		err = security_msg_queue_msgctl(NULL, cmd);
494		if (err)
495			return err;
496
497		memset(&msginfo, 0, sizeof(msginfo));
498		msginfo.msgmni = ns->msg_ctlmni;
499		msginfo.msgmax = ns->msg_ctlmax;
500		msginfo.msgmnb = ns->msg_ctlmnb;
501		msginfo.msgssz = MSGSSZ;
502		msginfo.msgseg = MSGSEG;
503		down_read(&msg_ids(ns).rwsem);
504		if (cmd == MSG_INFO) {
505			msginfo.msgpool = msg_ids(ns).in_use;
506			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
507			msginfo.msgtql = atomic_read(&ns->msg_bytes);
508		} else {
509			msginfo.msgmap = MSGMAP;
510			msginfo.msgpool = MSGPOOL;
511			msginfo.msgtql = MSGTQL;
512		}
513		max_id = ipc_get_maxid(&msg_ids(ns));
514		up_read(&msg_ids(ns).rwsem);
515		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
516			return -EFAULT;
517		return (max_id < 0) ? 0 : max_id;
518	}
519
520	case MSG_STAT:
521	case IPC_STAT:
522	{
523		struct msqid64_ds tbuf;
524		int success_return;
525
526		if (!buf)
527			return -EFAULT;
528
529		memset(&tbuf, 0, sizeof(tbuf));
530
531		rcu_read_lock();
532		if (cmd == MSG_STAT) {
533			msq = msq_obtain_object(ns, msqid);
534			if (IS_ERR(msq)) {
535				err = PTR_ERR(msq);
536				goto out_unlock;
537			}
538			success_return = msq->q_perm.id;
539		} else {
540			msq = msq_obtain_object_check(ns, msqid);
541			if (IS_ERR(msq)) {
542				err = PTR_ERR(msq);
543				goto out_unlock;
544			}
545			success_return = 0;
546		}
547
548		err = -EACCES;
549		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
550			goto out_unlock;
551
552		err = security_msg_queue_msgctl(msq, cmd);
553		if (err)
554			goto out_unlock;
555
556		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
557		tbuf.msg_stime  = msq->q_stime;
558		tbuf.msg_rtime  = msq->q_rtime;
559		tbuf.msg_ctime  = msq->q_ctime;
560		tbuf.msg_cbytes = msq->q_cbytes;
561		tbuf.msg_qnum   = msq->q_qnum;
562		tbuf.msg_qbytes = msq->q_qbytes;
563		tbuf.msg_lspid  = msq->q_lspid;
564		tbuf.msg_lrpid  = msq->q_lrpid;
565		rcu_read_unlock();
566
567		if (copy_msqid_to_user(buf, &tbuf, version))
568			return -EFAULT;
569		return success_return;
570	}
571
572	default:
573		return -EINVAL;
574	}
575
576	return err;
577out_unlock:
578	rcu_read_unlock();
579	return err;
580}
581
582SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
583{
584	int version;
585	struct ipc_namespace *ns;
586
587	if (msqid < 0 || cmd < 0)
588		return -EINVAL;
589
590	version = ipc_parse_version(&cmd);
591	ns = current->nsproxy->ipc_ns;
592
593	switch (cmd) {
594	case IPC_INFO:
595	case MSG_INFO:
596	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
597	case IPC_STAT:
598		return msgctl_nolock(ns, msqid, cmd, version, buf);
599	case IPC_SET:
600	case IPC_RMID:
601		return msgctl_down(ns, msqid, cmd, buf, version);
602	default:
603		return  -EINVAL;
604	}
605}
606
607static int testmsg(struct msg_msg *msg, long type, int mode)
608{
609	switch(mode)
610	{
611		case SEARCH_ANY:
612		case SEARCH_NUMBER:
613			return 1;
614		case SEARCH_LESSEQUAL:
615			if (msg->m_type <=type)
616				return 1;
617			break;
618		case SEARCH_EQUAL:
619			if (msg->m_type == type)
620				return 1;
621			break;
622		case SEARCH_NOTEQUAL:
623			if (msg->m_type != type)
624				return 1;
625			break;
626	}
627	return 0;
628}
629
630static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
631{
632	struct msg_receiver *msr, *t;
633
634	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
635		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
636		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
637					       msr->r_msgtype, msr->r_mode)) {
638
639			list_del(&msr->r_list);
640			if (msr->r_maxsize < msg->m_ts) {
641				msr->r_msg = NULL;
642				wake_up_process(msr->r_tsk);
643				smp_mb();
644				msr->r_msg = ERR_PTR(-E2BIG);
645			} else {
646				msr->r_msg = NULL;
647				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
648				msq->q_rtime = get_seconds();
649				wake_up_process(msr->r_tsk);
650				smp_mb();
651				msr->r_msg = msg;
652
653				return 1;
654			}
655		}
656	}
657	return 0;
658}
659
660long do_msgsnd(int msqid, long mtype, void __user *mtext,
661		size_t msgsz, int msgflg)
662{
663	struct msg_queue *msq;
664	struct msg_msg *msg;
665	int err;
666	struct ipc_namespace *ns;
667
668	ns = current->nsproxy->ipc_ns;
669
670	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
671		return -EINVAL;
672	if (mtype < 1)
673		return -EINVAL;
674
675	msg = load_msg(mtext, msgsz);
676	if (IS_ERR(msg))
677		return PTR_ERR(msg);
678
679	msg->m_type = mtype;
680	msg->m_ts = msgsz;
681
682	rcu_read_lock();
683	msq = msq_obtain_object_check(ns, msqid);
684	if (IS_ERR(msq)) {
685		err = PTR_ERR(msq);
686		goto out_unlock1;
687	}
688
689	ipc_lock_object(&msq->q_perm);
690
691	for (;;) {
692		struct msg_sender s;
693
694		err = -EACCES;
695		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
696			goto out_unlock0;
697
698		/* raced with RMID? */
699		if (msq->q_perm.deleted) {
700			err = -EIDRM;
701			goto out_unlock0;
702		}
703
704		err = security_msg_queue_msgsnd(msq, msg, msgflg);
705		if (err)
706			goto out_unlock0;
707
708		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
709				1 + msq->q_qnum <= msq->q_qbytes) {
710			break;
711		}
712
713		/* queue full, wait: */
714		if (msgflg & IPC_NOWAIT) {
715			err = -EAGAIN;
716			goto out_unlock0;
717		}
718
719		ss_add(msq, &s);
720
721		if (!ipc_rcu_getref(msq)) {
722			err = -EIDRM;
723			goto out_unlock0;
724		}
725
726		ipc_unlock_object(&msq->q_perm);
727		rcu_read_unlock();
728		schedule();
729
730		rcu_read_lock();
731		ipc_lock_object(&msq->q_perm);
732
733		ipc_rcu_putref(msq, ipc_rcu_free);
734		if (msq->q_perm.deleted) {
735			err = -EIDRM;
736			goto out_unlock0;
737		}
738
739		ss_del(&s);
740
741		if (signal_pending(current)) {
742			err = -ERESTARTNOHAND;
743			goto out_unlock0;
744		}
745
746	}
747	msq->q_lspid = task_tgid_vnr(current);
748	msq->q_stime = get_seconds();
749
750	if (!pipelined_send(msq, msg)) {
751		/* no one is waiting for this message, enqueue it */
752		list_add_tail(&msg->m_list, &msq->q_messages);
753		msq->q_cbytes += msgsz;
754		msq->q_qnum++;
755		atomic_add(msgsz, &ns->msg_bytes);
756		atomic_inc(&ns->msg_hdrs);
757	}
758
759	err = 0;
760	msg = NULL;
761
762out_unlock0:
763	ipc_unlock_object(&msq->q_perm);
764out_unlock1:
765	rcu_read_unlock();
766	if (msg != NULL)
767		free_msg(msg);
768	return err;
769}
770
771SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
772		int, msgflg)
773{
774	long mtype;
775
776	if (get_user(mtype, &msgp->mtype))
777		return -EFAULT;
778	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
779}
780
781static inline int convert_mode(long *msgtyp, int msgflg)
782{
783	if (msgflg & MSG_COPY)
784		return SEARCH_NUMBER;
785	/*
786	 *  find message of correct type.
787	 *  msgtyp = 0 => get first.
788	 *  msgtyp > 0 => get first message of matching type.
789	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
790	 */
791	if (*msgtyp == 0)
792		return SEARCH_ANY;
793	if (*msgtyp < 0) {
794		*msgtyp = -*msgtyp;
795		return SEARCH_LESSEQUAL;
796	}
797	if (msgflg & MSG_EXCEPT)
798		return SEARCH_NOTEQUAL;
799	return SEARCH_EQUAL;
800}
801
802static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
803{
804	struct msgbuf __user *msgp = dest;
805	size_t msgsz;
806
807	if (put_user(msg->m_type, &msgp->mtype))
808		return -EFAULT;
809
810	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
811	if (store_msg(msgp->mtext, msg, msgsz))
812		return -EFAULT;
813	return msgsz;
814}
815
816#ifdef CONFIG_CHECKPOINT_RESTORE
817/*
818 * This function creates new kernel message structure, large enough to store
819 * bufsz message bytes.
820 */
821static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
822{
823	struct msg_msg *copy;
824
825	/*
826	 * Create dummy message to copy real message to.
827	 */
828	copy = load_msg(buf, bufsz);
829	if (!IS_ERR(copy))
830		copy->m_ts = bufsz;
831	return copy;
832}
833
834static inline void free_copy(struct msg_msg *copy)
835{
836	if (copy)
837		free_msg(copy);
838}
839#else
840static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
841{
842	return ERR_PTR(-ENOSYS);
843}
844
845static inline void free_copy(struct msg_msg *copy)
846{
847}
848#endif
849
850static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
851{
852	struct msg_msg *msg, *found = NULL;
853	long count = 0;
854
855	list_for_each_entry(msg, &msq->q_messages, m_list) {
856		if (testmsg(msg, *msgtyp, mode) &&
857		    !security_msg_queue_msgrcv(msq, msg, current,
858					       *msgtyp, mode)) {
859			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
860				*msgtyp = msg->m_type - 1;
861				found = msg;
862			} else if (mode == SEARCH_NUMBER) {
863				if (*msgtyp == count)
864					return msg;
865			} else
866				return msg;
867			count++;
868		}
869	}
870
871	return found ?: ERR_PTR(-EAGAIN);
872}
873
874long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
875	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
876{
877	int mode;
878	struct msg_queue *msq;
879	struct ipc_namespace *ns;
880	struct msg_msg *msg, *copy = NULL;
881
882	ns = current->nsproxy->ipc_ns;
883
884	if (msqid < 0 || (long) bufsz < 0)
885		return -EINVAL;
886
887	if (msgflg & MSG_COPY) {
888		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
889		if (IS_ERR(copy))
890			return PTR_ERR(copy);
891	}
892	mode = convert_mode(&msgtyp, msgflg);
893
894	rcu_read_lock();
895	msq = msq_obtain_object_check(ns, msqid);
896	if (IS_ERR(msq)) {
897		rcu_read_unlock();
898		free_copy(copy);
899		return PTR_ERR(msq);
900	}
901
902	for (;;) {
903		struct msg_receiver msr_d;
904
905		msg = ERR_PTR(-EACCES);
906		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
907			goto out_unlock1;
908
909		ipc_lock_object(&msq->q_perm);
910
911		/* raced with RMID? */
912		if (msq->q_perm.deleted) {
913			msg = ERR_PTR(-EIDRM);
914			goto out_unlock0;
915		}
916
917		msg = find_msg(msq, &msgtyp, mode);
918		if (!IS_ERR(msg)) {
919			/*
920			 * Found a suitable message.
921			 * Unlink it from the queue.
922			 */
923			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
924				msg = ERR_PTR(-E2BIG);
925				goto out_unlock0;
926			}
927			/*
928			 * If we are copying, then do not unlink message and do
929			 * not update queue parameters.
930			 */
931			if (msgflg & MSG_COPY) {
932				msg = copy_msg(msg, copy);
933				goto out_unlock0;
934			}
935
936			list_del(&msg->m_list);
937			msq->q_qnum--;
938			msq->q_rtime = get_seconds();
939			msq->q_lrpid = task_tgid_vnr(current);
940			msq->q_cbytes -= msg->m_ts;
941			atomic_sub(msg->m_ts, &ns->msg_bytes);
942			atomic_dec(&ns->msg_hdrs);
943			ss_wakeup(&msq->q_senders, 0);
944
945			goto out_unlock0;
946		}
947
948		/* No message waiting. Wait for a message */
949		if (msgflg & IPC_NOWAIT) {
950			msg = ERR_PTR(-ENOMSG);
951			goto out_unlock0;
952		}
953
954		list_add_tail(&msr_d.r_list, &msq->q_receivers);
955		msr_d.r_tsk = current;
956		msr_d.r_msgtype = msgtyp;
957		msr_d.r_mode = mode;
958		if (msgflg & MSG_NOERROR)
959			msr_d.r_maxsize = INT_MAX;
960		else
961			msr_d.r_maxsize = bufsz;
962		msr_d.r_msg = ERR_PTR(-EAGAIN);
963		current->state = TASK_INTERRUPTIBLE;
964
965		ipc_unlock_object(&msq->q_perm);
966		rcu_read_unlock();
967		schedule();
968
969		/* Lockless receive, part 1:
970		 * Disable preemption.  We don't hold a reference to the queue
971		 * and getting a reference would defeat the idea of a lockless
972		 * operation, thus the code relies on rcu to guarantee the
973		 * existence of msq:
974		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
975		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
976		 * rcu_read_lock() prevents preemption between reading r_msg
977		 * and acquiring the q_perm.lock in ipc_lock_object().
978		 */
979		rcu_read_lock();
980
981		/* Lockless receive, part 2:
982		 * Wait until pipelined_send or expunge_all are outside of
983		 * wake_up_process(). There is a race with exit(), see
984		 * ipc/mqueue.c for the details.
985		 */
986		msg = (struct msg_msg*)msr_d.r_msg;
987		while (msg == NULL) {
988			cpu_relax();
989			msg = (struct msg_msg *)msr_d.r_msg;
990		}
991
992		/* Lockless receive, part 3:
993		 * If there is a message or an error then accept it without
994		 * locking.
995		 */
996		if (msg != ERR_PTR(-EAGAIN))
997			goto out_unlock1;
998
999		/* Lockless receive, part 3:
1000		 * Acquire the queue spinlock.
1001		 */
1002		ipc_lock_object(&msq->q_perm);
1003
1004		/* Lockless receive, part 4:
1005		 * Repeat test after acquiring the spinlock.
1006		 */
1007		msg = (struct msg_msg*)msr_d.r_msg;
1008		if (msg != ERR_PTR(-EAGAIN))
1009			goto out_unlock0;
1010
1011		list_del(&msr_d.r_list);
1012		if (signal_pending(current)) {
1013			msg = ERR_PTR(-ERESTARTNOHAND);
1014			goto out_unlock0;
1015		}
1016
1017		ipc_unlock_object(&msq->q_perm);
1018	}
1019
1020out_unlock0:
1021	ipc_unlock_object(&msq->q_perm);
1022out_unlock1:
1023	rcu_read_unlock();
1024	if (IS_ERR(msg)) {
1025		free_copy(copy);
1026		return PTR_ERR(msg);
1027	}
1028
1029	bufsz = msg_handler(buf, msg, bufsz);
1030	free_msg(msg);
1031
1032	return bufsz;
1033}
1034
1035SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1036		long, msgtyp, int, msgflg)
1037{
1038	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1039}
1040
1041#ifdef CONFIG_PROC_FS
1042static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1043{
1044	struct user_namespace *user_ns = seq_user_ns(s);
1045	struct msg_queue *msq = it;
1046
1047	return seq_printf(s,
1048			"%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1049			msq->q_perm.key,
1050			msq->q_perm.id,
1051			msq->q_perm.mode,
1052			msq->q_cbytes,
1053			msq->q_qnum,
1054			msq->q_lspid,
1055			msq->q_lrpid,
1056			from_kuid_munged(user_ns, msq->q_perm.uid),
1057			from_kgid_munged(user_ns, msq->q_perm.gid),
1058			from_kuid_munged(user_ns, msq->q_perm.cuid),
1059			from_kgid_munged(user_ns, msq->q_perm.cgid),
1060			msq->q_stime,
1061			msq->q_rtime,
1062			msq->q_ctime);
1063}
1064#endif
1065