msg.c revision cf9d5d78
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	/* ipc_addid() locks msq upon success. */
203	id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
204	if (id < 0) {
205		security_msg_queue_free(msq);
206		ipc_rcu_putref(msq);
207		return id;
208	}
209
210	msq->q_stime = msq->q_rtime = 0;
211	msq->q_ctime = get_seconds();
212	msq->q_cbytes = msq->q_qnum = 0;
213	msq->q_qbytes = ns->msg_ctlmnb;
214	msq->q_lspid = msq->q_lrpid = 0;
215	INIT_LIST_HEAD(&msq->q_messages);
216	INIT_LIST_HEAD(&msq->q_receivers);
217	INIT_LIST_HEAD(&msq->q_senders);
218
219	ipc_unlock_object(&msq->q_perm);
220	rcu_read_unlock();
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 msg_sender *mss, *t;
241
242	list_for_each_entry_safe(mss, t, h, list) {
243		if (kill)
244			mss->list.next = NULL;
245		wake_up_process(mss->tsk);
246	}
247}
248
249static void expunge_all(struct msg_queue *msq, int res)
250{
251	struct msg_receiver *msr, *t;
252
253	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
254		msr->r_msg = NULL;
255		wake_up_process(msr->r_tsk);
256		smp_mb();
257		msr->r_msg = ERR_PTR(res);
258	}
259}
260
261/*
262 * freeque() wakes up waiters on the sender and receiver waiting queue,
263 * removes the message queue from message queue ID IDR, and cleans up all the
264 * messages associated with this queue.
265 *
266 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
267 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
268 */
269static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
270{
271	struct msg_msg *msg, *t;
272	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
273
274	expunge_all(msq, -EIDRM);
275	ss_wakeup(&msq->q_senders, 1);
276	msg_rmid(ns, msq);
277	msg_unlock(msq);
278
279	list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
280		atomic_dec(&ns->msg_hdrs);
281		free_msg(msg);
282	}
283	atomic_sub(msq->q_cbytes, &ns->msg_bytes);
284	security_msg_queue_free(msq);
285	ipc_rcu_putref(msq);
286}
287
288/*
289 * Called with msg_ids.rw_mutex and ipcp locked.
290 */
291static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
292{
293	struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
294
295	return security_msg_queue_associate(msq, msgflg);
296}
297
298SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
299{
300	struct ipc_namespace *ns;
301	struct ipc_ops msg_ops;
302	struct ipc_params msg_params;
303
304	ns = current->nsproxy->ipc_ns;
305
306	msg_ops.getnew = newque;
307	msg_ops.associate = msg_security;
308	msg_ops.more_checks = NULL;
309
310	msg_params.key = key;
311	msg_params.flg = msgflg;
312
313	return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
314}
315
316static inline unsigned long
317copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
318{
319	switch(version) {
320	case IPC_64:
321		return copy_to_user(buf, in, sizeof(*in));
322	case IPC_OLD:
323	{
324		struct msqid_ds out;
325
326		memset(&out, 0, sizeof(out));
327
328		ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
329
330		out.msg_stime		= in->msg_stime;
331		out.msg_rtime		= in->msg_rtime;
332		out.msg_ctime		= in->msg_ctime;
333
334		if (in->msg_cbytes > USHRT_MAX)
335			out.msg_cbytes	= USHRT_MAX;
336		else
337			out.msg_cbytes	= in->msg_cbytes;
338		out.msg_lcbytes		= in->msg_cbytes;
339
340		if (in->msg_qnum > USHRT_MAX)
341			out.msg_qnum	= USHRT_MAX;
342		else
343			out.msg_qnum	= in->msg_qnum;
344
345		if (in->msg_qbytes > USHRT_MAX)
346			out.msg_qbytes	= USHRT_MAX;
347		else
348			out.msg_qbytes	= in->msg_qbytes;
349		out.msg_lqbytes		= in->msg_qbytes;
350
351		out.msg_lspid		= in->msg_lspid;
352		out.msg_lrpid		= in->msg_lrpid;
353
354		return copy_to_user(buf, &out, sizeof(out));
355	}
356	default:
357		return -EINVAL;
358	}
359}
360
361static inline unsigned long
362copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
363{
364	switch(version) {
365	case IPC_64:
366		if (copy_from_user(out, buf, sizeof(*out)))
367			return -EFAULT;
368		return 0;
369	case IPC_OLD:
370	{
371		struct msqid_ds tbuf_old;
372
373		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
374			return -EFAULT;
375
376		out->msg_perm.uid      	= tbuf_old.msg_perm.uid;
377		out->msg_perm.gid      	= tbuf_old.msg_perm.gid;
378		out->msg_perm.mode     	= tbuf_old.msg_perm.mode;
379
380		if (tbuf_old.msg_qbytes == 0)
381			out->msg_qbytes	= tbuf_old.msg_lqbytes;
382		else
383			out->msg_qbytes	= tbuf_old.msg_qbytes;
384
385		return 0;
386	}
387	default:
388		return -EINVAL;
389	}
390}
391
392/*
393 * This function handles some msgctl commands which require the rw_mutex
394 * to be held in write mode.
395 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
396 */
397static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
398		       struct msqid_ds __user *buf, int version)
399{
400	struct kern_ipc_perm *ipcp;
401	struct msqid64_ds uninitialized_var(msqid64);
402	struct msg_queue *msq;
403	int err;
404
405	if (cmd == IPC_SET) {
406		if (copy_msqid_from_user(&msqid64, buf, version))
407			return -EFAULT;
408	}
409
410	ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
411			       &msqid64.msg_perm, msqid64.msg_qbytes);
412	if (IS_ERR(ipcp))
413		return PTR_ERR(ipcp);
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_unlock;
420
421	switch (cmd) {
422	case IPC_RMID:
423		freeque(ns, ipcp);
424		goto out_up;
425	case IPC_SET:
426		if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
427		    !capable(CAP_SYS_RESOURCE)) {
428			err = -EPERM;
429			goto out_unlock;
430		}
431
432		err = ipc_update_perm(&msqid64.msg_perm, ipcp);
433		if (err)
434			goto out_unlock;
435
436		msq->q_qbytes = msqid64.msg_qbytes;
437
438		msq->q_ctime = get_seconds();
439		/* sleeping receivers might be excluded by
440		 * stricter permissions.
441		 */
442		expunge_all(msq, -EAGAIN);
443		/* sleeping senders might be able to send
444		 * due to a larger queue size.
445		 */
446		ss_wakeup(&msq->q_senders, 0);
447		break;
448	default:
449		err = -EINVAL;
450	}
451out_unlock:
452	msg_unlock(msq);
453out_up:
454	up_write(&msg_ids(ns).rw_mutex);
455	return err;
456}
457
458SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
459{
460	struct msg_queue *msq;
461	int err, version;
462	struct ipc_namespace *ns;
463
464	if (msqid < 0 || cmd < 0)
465		return -EINVAL;
466
467	version = ipc_parse_version(&cmd);
468	ns = current->nsproxy->ipc_ns;
469
470	switch (cmd) {
471	case IPC_INFO:
472	case MSG_INFO:
473	{
474		struct msginfo msginfo;
475		int max_id;
476
477		if (!buf)
478			return -EFAULT;
479		/*
480		 * We must not return kernel stack data.
481		 * due to padding, it's not enough
482		 * to set all member fields.
483		 */
484		err = security_msg_queue_msgctl(NULL, cmd);
485		if (err)
486			return err;
487
488		memset(&msginfo, 0, sizeof(msginfo));
489		msginfo.msgmni = ns->msg_ctlmni;
490		msginfo.msgmax = ns->msg_ctlmax;
491		msginfo.msgmnb = ns->msg_ctlmnb;
492		msginfo.msgssz = MSGSSZ;
493		msginfo.msgseg = MSGSEG;
494		down_read(&msg_ids(ns).rw_mutex);
495		if (cmd == MSG_INFO) {
496			msginfo.msgpool = msg_ids(ns).in_use;
497			msginfo.msgmap = atomic_read(&ns->msg_hdrs);
498			msginfo.msgtql = atomic_read(&ns->msg_bytes);
499		} else {
500			msginfo.msgmap = MSGMAP;
501			msginfo.msgpool = MSGPOOL;
502			msginfo.msgtql = MSGTQL;
503		}
504		max_id = ipc_get_maxid(&msg_ids(ns));
505		up_read(&msg_ids(ns).rw_mutex);
506		if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
507			return -EFAULT;
508		return (max_id < 0) ? 0 : max_id;
509	}
510	case MSG_STAT:	/* msqid is an index rather than a msg queue id */
511	case IPC_STAT:
512	{
513		struct msqid64_ds tbuf;
514		int success_return;
515
516		if (!buf)
517			return -EFAULT;
518
519		if (cmd == MSG_STAT) {
520			msq = msg_lock(ns, msqid);
521			if (IS_ERR(msq))
522				return PTR_ERR(msq);
523			success_return = msq->q_perm.id;
524		} else {
525			msq = msg_lock_check(ns, msqid);
526			if (IS_ERR(msq))
527				return PTR_ERR(msq);
528			success_return = 0;
529		}
530		err = -EACCES;
531		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
532			goto out_unlock;
533
534		err = security_msg_queue_msgctl(msq, cmd);
535		if (err)
536			goto out_unlock;
537
538		memset(&tbuf, 0, sizeof(tbuf));
539
540		kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
541		tbuf.msg_stime  = msq->q_stime;
542		tbuf.msg_rtime  = msq->q_rtime;
543		tbuf.msg_ctime  = msq->q_ctime;
544		tbuf.msg_cbytes = msq->q_cbytes;
545		tbuf.msg_qnum   = msq->q_qnum;
546		tbuf.msg_qbytes = msq->q_qbytes;
547		tbuf.msg_lspid  = msq->q_lspid;
548		tbuf.msg_lrpid  = msq->q_lrpid;
549		msg_unlock(msq);
550		if (copy_msqid_to_user(buf, &tbuf, version))
551			return -EFAULT;
552		return success_return;
553	}
554	case IPC_SET:
555	case IPC_RMID:
556		err = msgctl_down(ns, msqid, cmd, buf, version);
557		return err;
558	default:
559		return  -EINVAL;
560	}
561
562out_unlock:
563	msg_unlock(msq);
564	return err;
565}
566
567static int testmsg(struct msg_msg *msg, long type, int mode)
568{
569	switch(mode)
570	{
571		case SEARCH_ANY:
572		case SEARCH_NUMBER:
573			return 1;
574		case SEARCH_LESSEQUAL:
575			if (msg->m_type <=type)
576				return 1;
577			break;
578		case SEARCH_EQUAL:
579			if (msg->m_type == type)
580				return 1;
581			break;
582		case SEARCH_NOTEQUAL:
583			if (msg->m_type != type)
584				return 1;
585			break;
586	}
587	return 0;
588}
589
590static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
591{
592	struct msg_receiver *msr, *t;
593
594	list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
595		if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
596		    !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
597					       msr->r_msgtype, msr->r_mode)) {
598
599			list_del(&msr->r_list);
600			if (msr->r_maxsize < msg->m_ts) {
601				msr->r_msg = NULL;
602				wake_up_process(msr->r_tsk);
603				smp_mb();
604				msr->r_msg = ERR_PTR(-E2BIG);
605			} else {
606				msr->r_msg = NULL;
607				msq->q_lrpid = task_pid_vnr(msr->r_tsk);
608				msq->q_rtime = get_seconds();
609				wake_up_process(msr->r_tsk);
610				smp_mb();
611				msr->r_msg = msg;
612
613				return 1;
614			}
615		}
616	}
617	return 0;
618}
619
620long do_msgsnd(int msqid, long mtype, void __user *mtext,
621		size_t msgsz, int msgflg)
622{
623	struct msg_queue *msq;
624	struct msg_msg *msg;
625	int err;
626	struct ipc_namespace *ns;
627
628	ns = current->nsproxy->ipc_ns;
629
630	if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
631		return -EINVAL;
632	if (mtype < 1)
633		return -EINVAL;
634
635	msg = load_msg(mtext, msgsz);
636	if (IS_ERR(msg))
637		return PTR_ERR(msg);
638
639	msg->m_type = mtype;
640	msg->m_ts = msgsz;
641
642	msq = msg_lock_check(ns, msqid);
643	if (IS_ERR(msq)) {
644		err = PTR_ERR(msq);
645		goto out_free;
646	}
647
648	for (;;) {
649		struct msg_sender s;
650
651		err = -EACCES;
652		if (ipcperms(ns, &msq->q_perm, S_IWUGO))
653			goto out_unlock_free;
654
655		err = security_msg_queue_msgsnd(msq, msg, msgflg);
656		if (err)
657			goto out_unlock_free;
658
659		if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
660				1 + msq->q_qnum <= msq->q_qbytes) {
661			break;
662		}
663
664		/* queue full, wait: */
665		if (msgflg & IPC_NOWAIT) {
666			err = -EAGAIN;
667			goto out_unlock_free;
668		}
669		ss_add(msq, &s);
670
671		if (!ipc_rcu_getref(msq)) {
672			err = -EIDRM;
673			goto out_unlock_free;
674		}
675
676		msg_unlock(msq);
677		schedule();
678
679		ipc_lock_by_ptr(&msq->q_perm);
680		ipc_rcu_putref(msq);
681		if (msq->q_perm.deleted) {
682			err = -EIDRM;
683			goto out_unlock_free;
684		}
685		ss_del(&s);
686
687		if (signal_pending(current)) {
688			err = -ERESTARTNOHAND;
689			goto out_unlock_free;
690		}
691	}
692
693	msq->q_lspid = task_tgid_vnr(current);
694	msq->q_stime = get_seconds();
695
696	if (!pipelined_send(msq, msg)) {
697		/* no one is waiting for this message, enqueue it */
698		list_add_tail(&msg->m_list, &msq->q_messages);
699		msq->q_cbytes += msgsz;
700		msq->q_qnum++;
701		atomic_add(msgsz, &ns->msg_bytes);
702		atomic_inc(&ns->msg_hdrs);
703	}
704
705	err = 0;
706	msg = NULL;
707
708out_unlock_free:
709	msg_unlock(msq);
710out_free:
711	if (msg != NULL)
712		free_msg(msg);
713	return err;
714}
715
716SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
717		int, msgflg)
718{
719	long mtype;
720
721	if (get_user(mtype, &msgp->mtype))
722		return -EFAULT;
723	return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
724}
725
726static inline int convert_mode(long *msgtyp, int msgflg)
727{
728	if (msgflg & MSG_COPY)
729		return SEARCH_NUMBER;
730	/*
731	 *  find message of correct type.
732	 *  msgtyp = 0 => get first.
733	 *  msgtyp > 0 => get first message of matching type.
734	 *  msgtyp < 0 => get message with least type must be < abs(msgtype).
735	 */
736	if (*msgtyp == 0)
737		return SEARCH_ANY;
738	if (*msgtyp < 0) {
739		*msgtyp = -*msgtyp;
740		return SEARCH_LESSEQUAL;
741	}
742	if (msgflg & MSG_EXCEPT)
743		return SEARCH_NOTEQUAL;
744	return SEARCH_EQUAL;
745}
746
747static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
748{
749	struct msgbuf __user *msgp = dest;
750	size_t msgsz;
751
752	if (put_user(msg->m_type, &msgp->mtype))
753		return -EFAULT;
754
755	msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
756	if (store_msg(msgp->mtext, msg, msgsz))
757		return -EFAULT;
758	return msgsz;
759}
760
761#ifdef CONFIG_CHECKPOINT_RESTORE
762/*
763 * This function creates new kernel message structure, large enough to store
764 * bufsz message bytes.
765 */
766static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
767{
768	struct msg_msg *copy;
769
770	/*
771	 * Create dummy message to copy real message to.
772	 */
773	copy = load_msg(buf, bufsz);
774	if (!IS_ERR(copy))
775		copy->m_ts = bufsz;
776	return copy;
777}
778
779static inline void free_copy(struct msg_msg *copy)
780{
781	if (copy)
782		free_msg(copy);
783}
784#else
785static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
786{
787	return ERR_PTR(-ENOSYS);
788}
789
790static inline void free_copy(struct msg_msg *copy)
791{
792}
793#endif
794
795static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
796{
797	struct msg_msg *msg;
798	long count = 0;
799
800	list_for_each_entry(msg, &msq->q_messages, m_list) {
801		if (testmsg(msg, *msgtyp, mode) &&
802		    !security_msg_queue_msgrcv(msq, msg, current,
803					       *msgtyp, mode)) {
804			if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
805				*msgtyp = msg->m_type - 1;
806			} else if (mode == SEARCH_NUMBER) {
807				if (*msgtyp == count)
808					return msg;
809			} else
810				return msg;
811			count++;
812		}
813	}
814
815	return ERR_PTR(-EAGAIN);
816}
817
818
819long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
820	       int msgflg,
821	       long (*msg_handler)(void __user *, struct msg_msg *, size_t))
822{
823	struct msg_queue *msq;
824	struct msg_msg *msg;
825	int mode;
826	struct ipc_namespace *ns;
827	struct msg_msg *copy = NULL;
828
829	ns = current->nsproxy->ipc_ns;
830
831	if (msqid < 0 || (long) bufsz < 0)
832		return -EINVAL;
833	if (msgflg & MSG_COPY) {
834		copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
835		if (IS_ERR(copy))
836			return PTR_ERR(copy);
837	}
838	mode = convert_mode(&msgtyp, msgflg);
839
840	msq = msg_lock_check(ns, msqid);
841	if (IS_ERR(msq)) {
842		free_copy(copy);
843		return PTR_ERR(msq);
844	}
845
846	for (;;) {
847		struct msg_receiver msr_d;
848
849		msg = ERR_PTR(-EACCES);
850		if (ipcperms(ns, &msq->q_perm, S_IRUGO))
851			goto out_unlock;
852
853		msg = find_msg(msq, &msgtyp, mode);
854
855		if (!IS_ERR(msg)) {
856			/*
857			 * Found a suitable message.
858			 * Unlink it from the queue.
859			 */
860			if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
861				msg = ERR_PTR(-E2BIG);
862				goto out_unlock;
863			}
864			/*
865			 * If we are copying, then do not unlink message and do
866			 * not update queue parameters.
867			 */
868			if (msgflg & MSG_COPY) {
869				msg = copy_msg(msg, copy);
870				goto out_unlock;
871			}
872			list_del(&msg->m_list);
873			msq->q_qnum--;
874			msq->q_rtime = get_seconds();
875			msq->q_lrpid = task_tgid_vnr(current);
876			msq->q_cbytes -= msg->m_ts;
877			atomic_sub(msg->m_ts, &ns->msg_bytes);
878			atomic_dec(&ns->msg_hdrs);
879			ss_wakeup(&msq->q_senders, 0);
880			msg_unlock(msq);
881			break;
882		}
883		/* No message waiting. Wait for a message */
884		if (msgflg & IPC_NOWAIT) {
885			msg = ERR_PTR(-ENOMSG);
886			goto out_unlock;
887		}
888		list_add_tail(&msr_d.r_list, &msq->q_receivers);
889		msr_d.r_tsk = current;
890		msr_d.r_msgtype = msgtyp;
891		msr_d.r_mode = mode;
892		if (msgflg & MSG_NOERROR)
893			msr_d.r_maxsize = INT_MAX;
894		else
895			msr_d.r_maxsize = bufsz;
896		msr_d.r_msg = ERR_PTR(-EAGAIN);
897		current->state = TASK_INTERRUPTIBLE;
898		msg_unlock(msq);
899
900		schedule();
901
902		/* Lockless receive, part 1:
903		 * Disable preemption.  We don't hold a reference to the queue
904		 * and getting a reference would defeat the idea of a lockless
905		 * operation, thus the code relies on rcu to guarantee the
906		 * existence of msq:
907		 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
908		 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
909		 * rcu_read_lock() prevents preemption between reading r_msg
910		 * and the spin_lock() inside ipc_lock_by_ptr().
911		 */
912		rcu_read_lock();
913
914		/* Lockless receive, part 2:
915		 * Wait until pipelined_send or expunge_all are outside of
916		 * wake_up_process(). There is a race with exit(), see
917		 * ipc/mqueue.c for the details.
918		 */
919		msg = (struct msg_msg*)msr_d.r_msg;
920		while (msg == NULL) {
921			cpu_relax();
922			msg = (struct msg_msg *)msr_d.r_msg;
923		}
924
925		/* Lockless receive, part 3:
926		 * If there is a message or an error then accept it without
927		 * locking.
928		 */
929		if (msg != ERR_PTR(-EAGAIN)) {
930			rcu_read_unlock();
931			break;
932		}
933
934		/* Lockless receive, part 3:
935		 * Acquire the queue spinlock.
936		 */
937		ipc_lock_by_ptr(&msq->q_perm);
938		rcu_read_unlock();
939
940		/* Lockless receive, part 4:
941		 * Repeat test after acquiring the spinlock.
942		 */
943		msg = (struct msg_msg*)msr_d.r_msg;
944		if (msg != ERR_PTR(-EAGAIN))
945			goto out_unlock;
946
947		list_del(&msr_d.r_list);
948		if (signal_pending(current)) {
949			msg = ERR_PTR(-ERESTARTNOHAND);
950out_unlock:
951			msg_unlock(msq);
952			break;
953		}
954	}
955	if (IS_ERR(msg)) {
956		free_copy(copy);
957		return PTR_ERR(msg);
958	}
959
960	bufsz = msg_handler(buf, msg, bufsz);
961	free_msg(msg);
962
963	return bufsz;
964}
965
966SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
967		long, msgtyp, int, msgflg)
968{
969	return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
970}
971
972#ifdef CONFIG_PROC_FS
973static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
974{
975	struct user_namespace *user_ns = seq_user_ns(s);
976	struct msg_queue *msq = it;
977
978	return seq_printf(s,
979			"%10d %10d  %4o  %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
980			msq->q_perm.key,
981			msq->q_perm.id,
982			msq->q_perm.mode,
983			msq->q_cbytes,
984			msq->q_qnum,
985			msq->q_lspid,
986			msq->q_lrpid,
987			from_kuid_munged(user_ns, msq->q_perm.uid),
988			from_kgid_munged(user_ns, msq->q_perm.gid),
989			from_kuid_munged(user_ns, msq->q_perm.cuid),
990			from_kgid_munged(user_ns, msq->q_perm.cgid),
991			msq->q_stime,
992			msq->q_rtime,
993			msq->q_ctime);
994}
995#endif
996