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