msg.c revision 2dc958fa
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
70#define msg_ids(ns)	((ns)->ids[IPC_MSG_IDS])
71
72#define msg_unlock(msq)		ipc_unlock(&(msq)->q_perm)
73
74static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
75static int newque(struct ipc_namespace *, struct ipc_params *);
76#ifdef CONFIG_PROC_FS
77static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
78#endif
79
80/*
81 * Scale msgmni with the available lowmem size: the memory dedicated to msg
82 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
83 * Also take into account the number of nsproxies created so far.
84 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
85 */
86void recompute_msgmni(struct ipc_namespace *ns)
87{
88	struct sysinfo i;
89	unsigned long allowed;
90	int nb_ns;
91
92	si_meminfo(&i);
93	allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
94		/ MSGMNB;
95	nb_ns = atomic_read(&nr_ipc_ns);
96	allowed /= nb_ns;
97
98	if (allowed < MSGMNI) {
99		ns->msg_ctlmni = MSGMNI;
100		return;
101	}
102
103	if (allowed > IPCMNI / nb_ns) {
104		ns->msg_ctlmni = IPCMNI / nb_ns;
105		return;
106	}
107
108	ns->msg_ctlmni = allowed;
109}
110
111void msg_init_ns(struct ipc_namespace *ns)
112{
113	ns->msg_ctlmax = MSGMAX;
114	ns->msg_ctlmnb = MSGMNB;
115
116	recompute_msgmni(ns);
117
118	atomic_set(&ns->msg_bytes, 0);
119	atomic_set(&ns->msg_hdrs, 0);
120	ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
121}
122
123#ifdef CONFIG_IPC_NS
124void msg_exit_ns(struct ipc_namespace *ns)
125{
126	free_ipcs(ns, &msg_ids(ns), freeque);
127	idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
128}
129#endif
130
131void __init msg_init(void)
132{
133	msg_init_ns(&init_ipc_ns);
134
135	printk(KERN_INFO "msgmni has been set to %d\n",
136		init_ipc_ns.msg_ctlmni);
137
138	ipc_init_proc_interface("sysvipc/msg",
139				"       key      msqid perms      cbytes       qnum lspid lrpid   uid   gid  cuid  cgid      stime      rtime      ctime\n",
140				IPC_MSG_IDS, sysvipc_msg_proc_show);
141}
142
143/*
144 * msg_lock_(check_) routines are called in the paths where the rw_mutex
145 * is not held.
146 */
147static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
148{
149	struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
150
151	if (IS_ERR(ipcp))
152		return (struct msg_queue *)ipcp;
153
154	return container_of(ipcp, struct msg_queue, q_perm);
155}
156
157static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
158						int id)
159{
160	struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
161
162	if (IS_ERR(ipcp))
163		return (struct msg_queue *)ipcp;
164
165	return container_of(ipcp, struct msg_queue, q_perm);
166}
167
168static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
169{
170	ipc_rmid(&msg_ids(ns), &s->q_perm);
171}
172
173/**
174 * newque - Create a new msg queue
175 * @ns: namespace
176 * @params: ptr to the structure that contains the key and msgflg
177 *
178 * Called with msg_ids.rw_mutex held (writer)
179 */
180static int newque(struct ipc_namespace *ns, struct ipc_params *params)
181{
182	struct msg_queue *msq;
183	int id, retval;
184	key_t key = params->key;
185	int msgflg = params->flg;
186
187	msq = ipc_rcu_alloc(sizeof(*msq));
188	if (!msq)
189		return -ENOMEM;
190
191	msq->q_perm.mode = msgflg & S_IRWXUGO;
192	msq->q_perm.key = key;
193
194	msq->q_perm.security = NULL;
195	retval = security_msg_queue_alloc(msq);
196	if (retval) {
197		ipc_rcu_putref(msq);
198		return retval;
199	}
200
201	/*
202	 * ipc_addid() locks msq
203	 */
204	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
205	if (id < 0) {
206		security_msg_queue_free(msq);
207		ipc_rcu_putref(msq);
208		return id;
209	}
210
211	msq->q_stime = msq->q_rtime = 0;
212	msq->q_ctime = get_seconds();
213	msq->q_cbytes = msq->q_qnum = 0;
214	msq->q_qbytes = ns->msg_ctlmnb;
215	msq->q_lspid = msq->q_lrpid = 0;
216	INIT_LIST_HEAD(&msq->q_messages);
217	INIT_LIST_HEAD(&msq->q_receivers);
218	INIT_LIST_HEAD(&msq->q_senders);
219
220	msg_unlock(msq);
221
222	return msq->q_perm.id;
223}
224
225static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
226{
227	mss->tsk = current;
228	current->state = TASK_INTERRUPTIBLE;
229	list_add_tail(&mss->list, &msq->q_senders);
230}
231
232static inline void ss_del(struct msg_sender *mss)
233{
234	if (mss->list.next != NULL)
235		list_del(&mss->list);
236}
237
238static void ss_wakeup(struct list_head *h, int kill)
239{
240	struct list_head *tmp;
241
242	tmp = h->next;
243	while (tmp != h) {
244		struct msg_sender *mss;
245
246		mss = list_entry(tmp, struct msg_sender, list);
247		tmp = tmp->next;
248		if (kill)
249			mss->list.next = NULL;
250		wake_up_process(mss->tsk);
251	}
252}
253
254static void expunge_all(struct msg_queue *msq, int res)
255{
256	struct list_head *tmp;
257
258	tmp = msq->q_receivers.next;
259	while (tmp != &msq->q_receivers) {
260		struct msg_receiver *msr;
261
262		msr = list_entry(tmp, struct msg_receiver, r_list);
263		tmp = tmp->next;
264		msr->r_msg = NULL;
265		wake_up_process(msr->r_tsk);
266		smp_mb();
267		msr->r_msg = ERR_PTR(res);
268	}
269}
270
271/*
272 * freeque() wakes up waiters on the sender and receiver waiting queue,
273 * removes the message queue from message queue ID IDR, and cleans up all the
274 * messages associated with this queue.
275 *
276 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
277 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
278 */
279static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
280{
281	struct list_head *tmp;
282	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
283
284	expunge_all(msq, -EIDRM);
285	ss_wakeup(&msq->q_senders, 1);
286	msg_rmid(ns, msq);
287	msg_unlock(msq);
288
289	tmp = msq->q_messages.next;
290	while (tmp != &msq->q_messages) {
291		struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
292
293		tmp = tmp->next;
294		atomic_dec(&ns->msg_hdrs);
295		free_msg(msg);
296	}
297	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
298	security_msg_queue_free(msq);
299	ipc_rcu_putref(msq);
300}
301
302/*
303 * Called with msg_ids.rw_mutex and ipcp locked.
304 */
305static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
306{
307	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
308
309	return security_msg_queue_associate(msq, msgflg);
310}
311
312SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
313{
314	struct ipc_namespace *ns;
315	struct ipc_ops msg_ops;
316	struct ipc_params msg_params;
317
318	ns = current->nsproxy->ipc_ns;
319
320	msg_ops.getnew = newque;
321	msg_ops.associate = msg_security;
322	msg_ops.more_checks = NULL;
323
324	msg_params.key = key;
325	msg_params.flg = msgflg;
326
327	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
328}
329
330static inline unsigned long
331copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
332{
333	switch(version) {
334	case IPC_64:
335		return copy_to_user(buf, in, sizeof(*in));
336	case IPC_OLD:
337	{
338		struct msqid_ds out;
339
340		memset(&out, 0, sizeof(out));
341
342		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
343
344		out.msg_stime		= in->msg_stime;
345		out.msg_rtime		= in->msg_rtime;
346		out.msg_ctime		= in->msg_ctime;
347
348		if (in->msg_cbytes > USHRT_MAX)
349			out.msg_cbytes	= USHRT_MAX;
350		else
351			out.msg_cbytes	= in->msg_cbytes;
352		out.msg_lcbytes		= in->msg_cbytes;
353
354		if (in->msg_qnum > USHRT_MAX)
355			out.msg_qnum	= USHRT_MAX;
356		else
357			out.msg_qnum	= in->msg_qnum;
358
359		if (in->msg_qbytes > USHRT_MAX)
360			out.msg_qbytes	= USHRT_MAX;
361		else
362			out.msg_qbytes	= in->msg_qbytes;
363		out.msg_lqbytes		= in->msg_qbytes;
364
365		out.msg_lspid		= in->msg_lspid;
366		out.msg_lrpid		= in->msg_lrpid;
367
368		return copy_to_user(buf, &out, sizeof(out));
369	}
370	default:
371		return -EINVAL;
372	}
373}
374
375static inline unsigned long
376copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
377{
378	switch(version) {
379	case IPC_64:
380		if (copy_from_user(out, buf, sizeof(*out)))
381			return -EFAULT;
382		return 0;
383	case IPC_OLD:
384	{
385		struct msqid_ds tbuf_old;
386
387		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
388			return -EFAULT;
389
390		out->msg_perm.uid      	= tbuf_old.msg_perm.uid;
391		out->msg_perm.gid      	= tbuf_old.msg_perm.gid;
392		out->msg_perm.mode     	= tbuf_old.msg_perm.mode;
393
394		if (tbuf_old.msg_qbytes == 0)
395			out->msg_qbytes	= tbuf_old.msg_lqbytes;
396		else
397			out->msg_qbytes	= tbuf_old.msg_qbytes;
398
399		return 0;
400	}
401	default:
402		return -EINVAL;
403	}
404}
405
406/*
407 * This function handles some msgctl commands which require the rw_mutex
408 * to be held in write mode.
409 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
410 */
411static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
412		       struct msqid_ds __user *buf, int version)
413{
414	struct kern_ipc_perm *ipcp;
415	struct msqid64_ds uninitialized_var(msqid64);
416	struct msg_queue *msq;
417	int err;
418
419	if (cmd == IPC_SET) {
420		if (copy_msqid_from_user(&msqid64, buf, version))
421			return -EFAULT;
422	}
423
424	ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
425			       &msqid64.msg_perm, msqid64.msg_qbytes);
426	if (IS_ERR(ipcp))
427		return PTR_ERR(ipcp);
428
429	msq = container_of(ipcp, struct msg_queue, q_perm);
430
431	err = security_msg_queue_msgctl(msq, cmd);
432	if (err)
433		goto out_unlock;
434
435	switch (cmd) {
436	case IPC_RMID:
437		freeque(ns, ipcp);
438		goto out_up;
439	case IPC_SET:
440		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
441		    !capable(CAP_SYS_RESOURCE)) {
442			err = -EPERM;
443			goto out_unlock;
444		}
445
446		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
447		if (err)
448			goto out_unlock;
449
450		msq->q_qbytes = msqid64.msg_qbytes;
451
452		msq->q_ctime = get_seconds();
453		/* sleeping receivers might be excluded by
454		 * stricter permissions.
455		 */
456		expunge_all(msq, -EAGAIN);
457		/* sleeping senders might be able to send
458		 * due to a larger queue size.
459		 */
460		ss_wakeup(&msq->q_senders, 0);
461		break;
462	default:
463		err = -EINVAL;
464	}
465out_unlock:
466	msg_unlock(msq);
467out_up:
468	up_write(&msg_ids(ns).rw_mutex);
469	return err;
470}
471
472SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
473{
474	struct msg_queue *msq;
475	int err, version;
476	struct ipc_namespace *ns;
477
478	if (msqid < 0 || cmd < 0)
479		return -EINVAL;
480
481	version = ipc_parse_version(&cmd);
482	ns = current->nsproxy->ipc_ns;
483
484	switch (cmd) {
485	case IPC_INFO:
486	case MSG_INFO:
487	{
488		struct msginfo msginfo;
489		int max_id;
490
491		if (!buf)
492			return -EFAULT;
493		/*
494		 * We must not return kernel stack data.
495		 * due to padding, it's not enough
496		 * to set all member fields.
497		 */
498		err = security_msg_queue_msgctl(NULL, cmd);
499		if (err)
500			return err;
501
502		memset(&msginfo, 0, sizeof(msginfo));
503		msginfo.msgmni = ns->msg_ctlmni;
504		msginfo.msgmax = ns->msg_ctlmax;
505		msginfo.msgmnb = ns->msg_ctlmnb;
506		msginfo.msgssz = MSGSSZ;
507		msginfo.msgseg = MSGSEG;
508		down_read(&msg_ids(ns).rw_mutex);
509		if (cmd == MSG_INFO) {
510			msginfo.msgpool = msg_ids(ns).in_use;
511			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
512			msginfo.msgtql = atomic_read(&ns->msg_bytes);
513		} else {
514			msginfo.msgmap = MSGMAP;
515			msginfo.msgpool = MSGPOOL;
516			msginfo.msgtql = MSGTQL;
517		}
518		max_id = ipc_get_maxid(&msg_ids(ns));
519		up_read(&msg_ids(ns).rw_mutex);
520		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
521			return -EFAULT;
522		return (max_id < 0) ? 0 : max_id;
523	}
524	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
525	case IPC_STAT:
526	{
527		struct msqid64_ds tbuf;
528		int success_return;
529
530		if (!buf)
531			return -EFAULT;
532
533		if (cmd == MSG_STAT) {
534			msq = msg_lock(ns, msqid);
535			if (IS_ERR(msq))
536				return PTR_ERR(msq);
537			success_return = msq->q_perm.id;
538		} else {
539			msq = msg_lock_check(ns, msqid);
540			if (IS_ERR(msq))
541				return PTR_ERR(msq);
542			success_return = 0;
543		}
544		err = -EACCES;
545		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
546			goto out_unlock;
547
548		err = security_msg_queue_msgctl(msq, cmd);
549		if (err)
550			goto out_unlock;
551
552		memset(&tbuf, 0, sizeof(tbuf));
553
554		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
555		tbuf.msg_stime  = msq->q_stime;
556		tbuf.msg_rtime  = msq->q_rtime;
557		tbuf.msg_ctime  = msq->q_ctime;
558		tbuf.msg_cbytes = msq->q_cbytes;
559		tbuf.msg_qnum   = msq->q_qnum;
560		tbuf.msg_qbytes = msq->q_qbytes;
561		tbuf.msg_lspid  = msq->q_lspid;
562		tbuf.msg_lrpid  = msq->q_lrpid;
563		msg_unlock(msq);
564		if (copy_msqid_to_user(buf, &tbuf, version))
565			return -EFAULT;
566		return success_return;
567	}
568	case IPC_SET:
569	case IPC_RMID:
570		err = msgctl_down(ns, msqid, cmd, buf, version);
571		return err;
572	default:
573		return  -EINVAL;
574	}
575
576out_unlock:
577	msg_unlock(msq);
578	return err;
579}
580
581static int testmsg(struct msg_msg *msg, long type, int mode)
582{
583	switch(mode)
584	{
585		case SEARCH_ANY:
586			return 1;
587		case SEARCH_LESSEQUAL:
588			if (msg->m_type <=type)
589				return 1;
590			break;
591		case SEARCH_EQUAL:
592			if (msg->m_type == type)
593				return 1;
594			break;
595		case SEARCH_NOTEQUAL:
596			if (msg->m_type != type)
597				return 1;
598			break;
599	}
600	return 0;
601}
602
603static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
604{
605	struct list_head *tmp;
606
607	tmp = msq->q_receivers.next;
608	while (tmp != &msq->q_receivers) {
609		struct msg_receiver *msr;
610
611		msr = list_entry(tmp, struct msg_receiver, r_list);
612		tmp = tmp->next;
613		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
614		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
615					       msr->r_msgtype, msr->r_mode)) {
616
617			list_del(&msr->r_list);
618			if (msr->r_maxsize < msg->m_ts) {
619				msr->r_msg = NULL;
620				wake_up_process(msr->r_tsk);
621				smp_mb();
622				msr->r_msg = ERR_PTR(-E2BIG);
623			} else {
624				msr->r_msg = NULL;
625				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
626				msq->q_rtime = get_seconds();
627				wake_up_process(msr->r_tsk);
628				smp_mb();
629				msr->r_msg = msg;
630
631				return 1;
632			}
633		}
634	}
635	return 0;
636}
637
638long do_msgsnd(int msqid, long mtype, void __user *mtext,
639		size_t msgsz, int msgflg)
640{
641	struct msg_queue *msq;
642	struct msg_msg *msg;
643	int err;
644	struct ipc_namespace *ns;
645
646	ns = current->nsproxy->ipc_ns;
647
648	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
649		return -EINVAL;
650	if (mtype < 1)
651		return -EINVAL;
652
653	msg = load_msg(mtext, msgsz);
654	if (IS_ERR(msg))
655		return PTR_ERR(msg);
656
657	msg->m_type = mtype;
658	msg->m_ts = msgsz;
659
660	msq = msg_lock_check(ns, msqid);
661	if (IS_ERR(msq)) {
662		err = PTR_ERR(msq);
663		goto out_free;
664	}
665
666	for (;;) {
667		struct msg_sender s;
668
669		err = -EACCES;
670		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
671			goto out_unlock_free;
672
673		err = security_msg_queue_msgsnd(msq, msg, msgflg);
674		if (err)
675			goto out_unlock_free;
676
677		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
678				1 + msq->q_qnum <= msq->q_qbytes) {
679			break;
680		}
681
682		/* queue full, wait: */
683		if (msgflg & IPC_NOWAIT) {
684			err = -EAGAIN;
685			goto out_unlock_free;
686		}
687		ss_add(msq, &s);
688		ipc_rcu_getref(msq);
689		msg_unlock(msq);
690		schedule();
691
692		ipc_lock_by_ptr(&msq->q_perm);
693		ipc_rcu_putref(msq);
694		if (msq->q_perm.deleted) {
695			err = -EIDRM;
696			goto out_unlock_free;
697		}
698		ss_del(&s);
699
700		if (signal_pending(current)) {
701			err = -ERESTARTNOHAND;
702			goto out_unlock_free;
703		}
704	}
705
706	msq->q_lspid = task_tgid_vnr(current);
707	msq->q_stime = get_seconds();
708
709	if (!pipelined_send(msq, msg)) {
710		/* no one is waiting for this message, enqueue it */
711		list_add_tail(&msg->m_list, &msq->q_messages);
712		msq->q_cbytes += msgsz;
713		msq->q_qnum++;
714		atomic_add(msgsz, &ns->msg_bytes);
715		atomic_inc(&ns->msg_hdrs);
716	}
717
718	err = 0;
719	msg = NULL;
720
721out_unlock_free:
722	msg_unlock(msq);
723out_free:
724	if (msg != NULL)
725		free_msg(msg);
726	return err;
727}
728
729SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
730		int, msgflg)
731{
732	long mtype;
733
734	if (get_user(mtype, &msgp->mtype))
735		return -EFAULT;
736	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
737}
738
739static inline int convert_mode(long *msgtyp, int msgflg)
740{
741	/*
742	 *  find message of correct type.
743	 *  msgtyp = 0 => get first.
744	 *  msgtyp > 0 => get first message of matching type.
745	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
746	 */
747	if (*msgtyp == 0)
748		return SEARCH_ANY;
749	if (*msgtyp < 0) {
750		*msgtyp = -*msgtyp;
751		return SEARCH_LESSEQUAL;
752	}
753	if (msgflg & MSG_EXCEPT)
754		return SEARCH_NOTEQUAL;
755	return SEARCH_EQUAL;
756}
757
758static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
759{
760	struct msgbuf __user *msgp = dest;
761	size_t msgsz;
762
763	if (put_user(msg->m_type, &msgp->mtype))
764		return -EFAULT;
765
766	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
767	if (store_msg(msgp->mtext, msg, msgsz))
768		return -EFAULT;
769	return msgsz;
770}
771
772#ifdef CONFIG_CHECKPOINT_RESTORE
773/*
774 * This function creates new kernel message structure, large enough to store
775 * bufsz message bytes.
776 */
777static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz,
778					   int msgflg, long *msgtyp,
779					   unsigned long *copy_number)
780{
781	struct msg_msg *copy;
782
783	*copy_number = *msgtyp;
784	*msgtyp = 0;
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					   int msgflg, long *msgtyp,
802					   unsigned long *copy_number)
803{
804	return ERR_PTR(-ENOSYS);
805}
806
807static inline void free_copy(struct msg_msg *copy)
808{
809}
810#endif
811
812long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
813	       int msgflg,
814	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
815{
816	struct msg_queue *msq;
817	struct msg_msg *msg;
818	int mode;
819	struct ipc_namespace *ns;
820	struct msg_msg *copy = NULL;
821	unsigned long copy_number = 0;
822
823	ns = current->nsproxy->ipc_ns;
824
825	if (msqid < 0 || (long) bufsz < 0)
826		return -EINVAL;
827	if (msgflg & MSG_COPY) {
828		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax),
829				    msgflg, &msgtyp, &copy_number);
830		if (IS_ERR(copy))
831			return PTR_ERR(copy);
832	}
833	mode = convert_mode(&msgtyp, msgflg);
834
835	msq = msg_lock_check(ns, msqid);
836	if (IS_ERR(msq)) {
837		free_copy(copy);
838		return PTR_ERR(msq);
839	}
840
841	for (;;) {
842		struct msg_receiver msr_d;
843		struct list_head *tmp;
844		long msg_counter = 0;
845
846		msg = ERR_PTR(-EACCES);
847		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
848			goto out_unlock;
849
850		msg = ERR_PTR(-EAGAIN);
851		tmp = msq->q_messages.next;
852		while (tmp != &msq->q_messages) {
853			struct msg_msg *walk_msg;
854
855			walk_msg = list_entry(tmp, struct msg_msg, m_list);
856			if (testmsg(walk_msg, msgtyp, mode) &&
857			    !security_msg_queue_msgrcv(msq, walk_msg, current,
858						       msgtyp, mode)) {
859
860				msg = walk_msg;
861				if (mode == SEARCH_LESSEQUAL &&
862						walk_msg->m_type != 1) {
863					msgtyp = walk_msg->m_type - 1;
864				} else if (msgflg & MSG_COPY) {
865					if (copy_number == msg_counter) {
866						/*
867						 * Found requested message.
868						 * Copy it.
869						 */
870						msg = copy_msg(msg, copy);
871						if (IS_ERR(msg))
872							goto out_unlock;
873						break;
874					}
875					msg = ERR_PTR(-EAGAIN);
876				} else
877					break;
878				msg_counter++;
879			}
880			tmp = tmp->next;
881		}
882		if (!IS_ERR(msg)) {
883			/*
884			 * Found a suitable message.
885			 * Unlink it from the queue.
886			 */
887			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
888				msg = ERR_PTR(-E2BIG);
889				goto out_unlock;
890			}
891			/*
892			 * If we are copying, then do not unlink message and do
893			 * not update queue parameters.
894			 */
895			if (msgflg & MSG_COPY)
896				goto out_unlock;
897			list_del(&msg->m_list);
898			msq->q_qnum--;
899			msq->q_rtime = get_seconds();
900			msq->q_lrpid = task_tgid_vnr(current);
901			msq->q_cbytes -= msg->m_ts;
902			atomic_sub(msg->m_ts, &ns->msg_bytes);
903			atomic_dec(&ns->msg_hdrs);
904			ss_wakeup(&msq->q_senders, 0);
905			msg_unlock(msq);
906			break;
907		}
908		/* No message waiting. Wait for a message */
909		if (msgflg & IPC_NOWAIT) {
910			msg = ERR_PTR(-ENOMSG);
911			goto out_unlock;
912		}
913		list_add_tail(&msr_d.r_list, &msq->q_receivers);
914		msr_d.r_tsk = current;
915		msr_d.r_msgtype = msgtyp;
916		msr_d.r_mode = mode;
917		if (msgflg & MSG_NOERROR)
918			msr_d.r_maxsize = INT_MAX;
919		else
920			msr_d.r_maxsize = bufsz;
921		msr_d.r_msg = ERR_PTR(-EAGAIN);
922		current->state = TASK_INTERRUPTIBLE;
923		msg_unlock(msq);
924
925		schedule();
926
927		/* Lockless receive, part 1:
928		 * Disable preemption.  We don't hold a reference to the queue
929		 * and getting a reference would defeat the idea of a lockless
930		 * operation, thus the code relies on rcu to guarantee the
931		 * existence of msq:
932		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
933		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
934		 * rcu_read_lock() prevents preemption between reading r_msg
935		 * and the spin_lock() inside ipc_lock_by_ptr().
936		 */
937		rcu_read_lock();
938
939		/* Lockless receive, part 2:
940		 * Wait until pipelined_send or expunge_all are outside of
941		 * wake_up_process(). There is a race with exit(), see
942		 * ipc/mqueue.c for the details.
943		 */
944		msg = (struct msg_msg*)msr_d.r_msg;
945		while (msg == NULL) {
946			cpu_relax();
947			msg = (struct msg_msg *)msr_d.r_msg;
948		}
949
950		/* Lockless receive, part 3:
951		 * If there is a message or an error then accept it without
952		 * locking.
953		 */
954		if (msg != ERR_PTR(-EAGAIN)) {
955			rcu_read_unlock();
956			break;
957		}
958
959		/* Lockless receive, part 3:
960		 * Acquire the queue spinlock.
961		 */
962		ipc_lock_by_ptr(&msq->q_perm);
963		rcu_read_unlock();
964
965		/* Lockless receive, part 4:
966		 * Repeat test after acquiring the spinlock.
967		 */
968		msg = (struct msg_msg*)msr_d.r_msg;
969		if (msg != ERR_PTR(-EAGAIN))
970			goto out_unlock;
971
972		list_del(&msr_d.r_list);
973		if (signal_pending(current)) {
974			msg = ERR_PTR(-ERESTARTNOHAND);
975out_unlock:
976			msg_unlock(msq);
977			break;
978		}
979	}
980	if (IS_ERR(msg)) {
981		free_copy(copy);
982		return PTR_ERR(msg);
983	}
984
985	bufsz = msg_handler(buf, msg, bufsz);
986	free_msg(msg);
987
988	return bufsz;
989}
990
991SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
992		long, msgtyp, int, msgflg)
993{
994	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
995}
996
997#ifdef CONFIG_PROC_FS
998static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
999{
1000	struct user_namespace *user_ns = seq_user_ns(s);
1001	struct msg_queue *msq = it;
1002
1003	return seq_printf(s,
1004			"%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1005			msq->q_perm.key,
1006			msq->q_perm.id,
1007			msq->q_perm.mode,
1008			msq->q_cbytes,
1009			msq->q_qnum,
1010			msq->q_lspid,
1011			msq->q_lrpid,
1012			from_kuid_munged(user_ns, msq->q_perm.uid),
1013			from_kgid_munged(user_ns, msq->q_perm.gid),
1014			from_kuid_munged(user_ns, msq->q_perm.cuid),
1015			from_kgid_munged(user_ns, msq->q_perm.cgid),
1016			msq->q_stime,
1017			msq->q_rtime,
1018			msq->q_ctime);
1019}
1020#endif
1021