Loading...
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 <linux/uaccess.h>
43#include "util.h"
44
45/* one msg_receiver structure for each sleeping receiver */
46struct msg_receiver {
47 struct list_head r_list;
48 struct task_struct *r_tsk;
49
50 int r_mode;
51 long r_msgtype;
52 long r_maxsize;
53
54 /*
55 * Mark r_msg volatile so that the compiler
56 * does not try to get smart and optimize
57 * it. We rely on this for the lockless
58 * receive algorithm.
59 */
60 struct msg_msg *volatile r_msg;
61};
62
63/* one msg_sender for each sleeping sender */
64struct msg_sender {
65 struct list_head list;
66 struct task_struct *tsk;
67};
68
69#define SEARCH_ANY 1
70#define SEARCH_EQUAL 2
71#define SEARCH_NOTEQUAL 3
72#define SEARCH_LESSEQUAL 4
73#define SEARCH_NUMBER 5
74
75#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
76
77static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
78{
79 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
80
81 if (IS_ERR(ipcp))
82 return ERR_CAST(ipcp);
83
84 return container_of(ipcp, struct msg_queue, q_perm);
85}
86
87static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
88 int id)
89{
90 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
91
92 if (IS_ERR(ipcp))
93 return ERR_CAST(ipcp);
94
95 return container_of(ipcp, struct msg_queue, q_perm);
96}
97
98static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
99{
100 ipc_rmid(&msg_ids(ns), &s->q_perm);
101}
102
103static void msg_rcu_free(struct rcu_head *head)
104{
105 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
106 struct msg_queue *msq = ipc_rcu_to_struct(p);
107
108 security_msg_queue_free(msq);
109 ipc_rcu_free(head);
110}
111
112/**
113 * newque - Create a new msg queue
114 * @ns: namespace
115 * @params: ptr to the structure that contains the key and msgflg
116 *
117 * Called with msg_ids.rwsem held (writer)
118 */
119static int newque(struct ipc_namespace *ns, struct ipc_params *params)
120{
121 struct msg_queue *msq;
122 int id, retval;
123 key_t key = params->key;
124 int msgflg = params->flg;
125
126 msq = ipc_rcu_alloc(sizeof(*msq));
127 if (!msq)
128 return -ENOMEM;
129
130 msq->q_perm.mode = msgflg & S_IRWXUGO;
131 msq->q_perm.key = key;
132
133 msq->q_perm.security = NULL;
134 retval = security_msg_queue_alloc(msq);
135 if (retval) {
136 ipc_rcu_putref(msq, ipc_rcu_free);
137 return retval;
138 }
139
140 msq->q_stime = msq->q_rtime = 0;
141 msq->q_ctime = get_seconds();
142 msq->q_cbytes = msq->q_qnum = 0;
143 msq->q_qbytes = ns->msg_ctlmnb;
144 msq->q_lspid = msq->q_lrpid = 0;
145 INIT_LIST_HEAD(&msq->q_messages);
146 INIT_LIST_HEAD(&msq->q_receivers);
147 INIT_LIST_HEAD(&msq->q_senders);
148
149 /* ipc_addid() locks msq upon success. */
150 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
151 if (id < 0) {
152 ipc_rcu_putref(msq, msg_rcu_free);
153 return id;
154 }
155
156 ipc_unlock_object(&msq->q_perm);
157 rcu_read_unlock();
158
159 return msq->q_perm.id;
160}
161
162static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
163{
164 mss->tsk = current;
165 __set_current_state(TASK_INTERRUPTIBLE);
166 list_add_tail(&mss->list, &msq->q_senders);
167}
168
169static inline void ss_del(struct msg_sender *mss)
170{
171 if (mss->list.next != NULL)
172 list_del(&mss->list);
173}
174
175static void ss_wakeup(struct list_head *h, int kill)
176{
177 struct msg_sender *mss, *t;
178
179 list_for_each_entry_safe(mss, t, h, list) {
180 if (kill)
181 mss->list.next = NULL;
182 wake_up_process(mss->tsk);
183 }
184}
185
186static void expunge_all(struct msg_queue *msq, int res)
187{
188 struct msg_receiver *msr, *t;
189
190 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
191 msr->r_msg = NULL; /* initialize expunge ordering */
192 wake_up_process(msr->r_tsk);
193 /*
194 * Ensure that the wakeup is visible before setting r_msg as
195 * the receiving end depends on it: either spinning on a nil,
196 * or dealing with -EAGAIN cases. See lockless receive part 1
197 * and 2 in do_msgrcv().
198 */
199 smp_wmb(); /* barrier (B) */
200 msr->r_msg = ERR_PTR(res);
201 }
202}
203
204/*
205 * freeque() wakes up waiters on the sender and receiver waiting queue,
206 * removes the message queue from message queue ID IDR, and cleans up all the
207 * messages associated with this queue.
208 *
209 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
210 * before freeque() is called. msg_ids.rwsem remains locked on exit.
211 */
212static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
213{
214 struct msg_msg *msg, *t;
215 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
216
217 expunge_all(msq, -EIDRM);
218 ss_wakeup(&msq->q_senders, 1);
219 msg_rmid(ns, msq);
220 ipc_unlock_object(&msq->q_perm);
221 rcu_read_unlock();
222
223 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
224 atomic_dec(&ns->msg_hdrs);
225 free_msg(msg);
226 }
227 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
228 ipc_rcu_putref(msq, msg_rcu_free);
229}
230
231/*
232 * Called with msg_ids.rwsem and ipcp locked.
233 */
234static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
235{
236 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
237
238 return security_msg_queue_associate(msq, msgflg);
239}
240
241SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
242{
243 struct ipc_namespace *ns;
244 static const struct ipc_ops msg_ops = {
245 .getnew = newque,
246 .associate = msg_security,
247 };
248 struct ipc_params msg_params;
249
250 ns = current->nsproxy->ipc_ns;
251
252 msg_params.key = key;
253 msg_params.flg = msgflg;
254
255 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
256}
257
258static inline unsigned long
259copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
260{
261 switch (version) {
262 case IPC_64:
263 return copy_to_user(buf, in, sizeof(*in));
264 case IPC_OLD:
265 {
266 struct msqid_ds out;
267
268 memset(&out, 0, sizeof(out));
269
270 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
271
272 out.msg_stime = in->msg_stime;
273 out.msg_rtime = in->msg_rtime;
274 out.msg_ctime = in->msg_ctime;
275
276 if (in->msg_cbytes > USHRT_MAX)
277 out.msg_cbytes = USHRT_MAX;
278 else
279 out.msg_cbytes = in->msg_cbytes;
280 out.msg_lcbytes = in->msg_cbytes;
281
282 if (in->msg_qnum > USHRT_MAX)
283 out.msg_qnum = USHRT_MAX;
284 else
285 out.msg_qnum = in->msg_qnum;
286
287 if (in->msg_qbytes > USHRT_MAX)
288 out.msg_qbytes = USHRT_MAX;
289 else
290 out.msg_qbytes = in->msg_qbytes;
291 out.msg_lqbytes = in->msg_qbytes;
292
293 out.msg_lspid = in->msg_lspid;
294 out.msg_lrpid = in->msg_lrpid;
295
296 return copy_to_user(buf, &out, sizeof(out));
297 }
298 default:
299 return -EINVAL;
300 }
301}
302
303static inline unsigned long
304copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
305{
306 switch (version) {
307 case IPC_64:
308 if (copy_from_user(out, buf, sizeof(*out)))
309 return -EFAULT;
310 return 0;
311 case IPC_OLD:
312 {
313 struct msqid_ds tbuf_old;
314
315 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
316 return -EFAULT;
317
318 out->msg_perm.uid = tbuf_old.msg_perm.uid;
319 out->msg_perm.gid = tbuf_old.msg_perm.gid;
320 out->msg_perm.mode = tbuf_old.msg_perm.mode;
321
322 if (tbuf_old.msg_qbytes == 0)
323 out->msg_qbytes = tbuf_old.msg_lqbytes;
324 else
325 out->msg_qbytes = tbuf_old.msg_qbytes;
326
327 return 0;
328 }
329 default:
330 return -EINVAL;
331 }
332}
333
334/*
335 * This function handles some msgctl commands which require the rwsem
336 * to be held in write mode.
337 * NOTE: no locks must be held, the rwsem is taken inside this function.
338 */
339static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
340 struct msqid_ds __user *buf, int version)
341{
342 struct kern_ipc_perm *ipcp;
343 struct msqid64_ds uninitialized_var(msqid64);
344 struct msg_queue *msq;
345 int err;
346
347 if (cmd == IPC_SET) {
348 if (copy_msqid_from_user(&msqid64, buf, version))
349 return -EFAULT;
350 }
351
352 down_write(&msg_ids(ns).rwsem);
353 rcu_read_lock();
354
355 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
356 &msqid64.msg_perm, msqid64.msg_qbytes);
357 if (IS_ERR(ipcp)) {
358 err = PTR_ERR(ipcp);
359 goto out_unlock1;
360 }
361
362 msq = container_of(ipcp, struct msg_queue, q_perm);
363
364 err = security_msg_queue_msgctl(msq, cmd);
365 if (err)
366 goto out_unlock1;
367
368 switch (cmd) {
369 case IPC_RMID:
370 ipc_lock_object(&msq->q_perm);
371 /* freeque unlocks the ipc object and rcu */
372 freeque(ns, ipcp);
373 goto out_up;
374 case IPC_SET:
375 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
376 !capable(CAP_SYS_RESOURCE)) {
377 err = -EPERM;
378 goto out_unlock1;
379 }
380
381 ipc_lock_object(&msq->q_perm);
382 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
383 if (err)
384 goto out_unlock0;
385
386 msq->q_qbytes = msqid64.msg_qbytes;
387
388 msq->q_ctime = get_seconds();
389 /* sleeping receivers might be excluded by
390 * stricter permissions.
391 */
392 expunge_all(msq, -EAGAIN);
393 /* sleeping senders might be able to send
394 * due to a larger queue size.
395 */
396 ss_wakeup(&msq->q_senders, 0);
397 break;
398 default:
399 err = -EINVAL;
400 goto out_unlock1;
401 }
402
403out_unlock0:
404 ipc_unlock_object(&msq->q_perm);
405out_unlock1:
406 rcu_read_unlock();
407out_up:
408 up_write(&msg_ids(ns).rwsem);
409 return err;
410}
411
412static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
413 int cmd, int version, void __user *buf)
414{
415 int err;
416 struct msg_queue *msq;
417
418 switch (cmd) {
419 case IPC_INFO:
420 case MSG_INFO:
421 {
422 struct msginfo msginfo;
423 int max_id;
424
425 if (!buf)
426 return -EFAULT;
427
428 /*
429 * We must not return kernel stack data.
430 * due to padding, it's not enough
431 * to set all member fields.
432 */
433 err = security_msg_queue_msgctl(NULL, cmd);
434 if (err)
435 return err;
436
437 memset(&msginfo, 0, sizeof(msginfo));
438 msginfo.msgmni = ns->msg_ctlmni;
439 msginfo.msgmax = ns->msg_ctlmax;
440 msginfo.msgmnb = ns->msg_ctlmnb;
441 msginfo.msgssz = MSGSSZ;
442 msginfo.msgseg = MSGSEG;
443 down_read(&msg_ids(ns).rwsem);
444 if (cmd == MSG_INFO) {
445 msginfo.msgpool = msg_ids(ns).in_use;
446 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
447 msginfo.msgtql = atomic_read(&ns->msg_bytes);
448 } else {
449 msginfo.msgmap = MSGMAP;
450 msginfo.msgpool = MSGPOOL;
451 msginfo.msgtql = MSGTQL;
452 }
453 max_id = ipc_get_maxid(&msg_ids(ns));
454 up_read(&msg_ids(ns).rwsem);
455 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
456 return -EFAULT;
457 return (max_id < 0) ? 0 : max_id;
458 }
459
460 case MSG_STAT:
461 case IPC_STAT:
462 {
463 struct msqid64_ds tbuf;
464 int success_return;
465
466 if (!buf)
467 return -EFAULT;
468
469 memset(&tbuf, 0, sizeof(tbuf));
470
471 rcu_read_lock();
472 if (cmd == MSG_STAT) {
473 msq = msq_obtain_object(ns, msqid);
474 if (IS_ERR(msq)) {
475 err = PTR_ERR(msq);
476 goto out_unlock;
477 }
478 success_return = msq->q_perm.id;
479 } else {
480 msq = msq_obtain_object_check(ns, msqid);
481 if (IS_ERR(msq)) {
482 err = PTR_ERR(msq);
483 goto out_unlock;
484 }
485 success_return = 0;
486 }
487
488 err = -EACCES;
489 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
490 goto out_unlock;
491
492 err = security_msg_queue_msgctl(msq, cmd);
493 if (err)
494 goto out_unlock;
495
496 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
497 tbuf.msg_stime = msq->q_stime;
498 tbuf.msg_rtime = msq->q_rtime;
499 tbuf.msg_ctime = msq->q_ctime;
500 tbuf.msg_cbytes = msq->q_cbytes;
501 tbuf.msg_qnum = msq->q_qnum;
502 tbuf.msg_qbytes = msq->q_qbytes;
503 tbuf.msg_lspid = msq->q_lspid;
504 tbuf.msg_lrpid = msq->q_lrpid;
505 rcu_read_unlock();
506
507 if (copy_msqid_to_user(buf, &tbuf, version))
508 return -EFAULT;
509 return success_return;
510 }
511
512 default:
513 return -EINVAL;
514 }
515
516 return err;
517out_unlock:
518 rcu_read_unlock();
519 return err;
520}
521
522SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
523{
524 int version;
525 struct ipc_namespace *ns;
526
527 if (msqid < 0 || cmd < 0)
528 return -EINVAL;
529
530 version = ipc_parse_version(&cmd);
531 ns = current->nsproxy->ipc_ns;
532
533 switch (cmd) {
534 case IPC_INFO:
535 case MSG_INFO:
536 case MSG_STAT: /* msqid is an index rather than a msg queue id */
537 case IPC_STAT:
538 return msgctl_nolock(ns, msqid, cmd, version, buf);
539 case IPC_SET:
540 case IPC_RMID:
541 return msgctl_down(ns, msqid, cmd, buf, version);
542 default:
543 return -EINVAL;
544 }
545}
546
547static int testmsg(struct msg_msg *msg, long type, int mode)
548{
549 switch (mode) {
550 case SEARCH_ANY:
551 case SEARCH_NUMBER:
552 return 1;
553 case SEARCH_LESSEQUAL:
554 if (msg->m_type <= type)
555 return 1;
556 break;
557 case SEARCH_EQUAL:
558 if (msg->m_type == type)
559 return 1;
560 break;
561 case SEARCH_NOTEQUAL:
562 if (msg->m_type != type)
563 return 1;
564 break;
565 }
566 return 0;
567}
568
569static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
570{
571 struct msg_receiver *msr, *t;
572
573 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
574 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
575 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
576 msr->r_msgtype, msr->r_mode)) {
577
578 list_del(&msr->r_list);
579 if (msr->r_maxsize < msg->m_ts) {
580 /* initialize pipelined send ordering */
581 msr->r_msg = NULL;
582 wake_up_process(msr->r_tsk);
583 /* barrier (B) see barrier comment below */
584 smp_wmb();
585 msr->r_msg = ERR_PTR(-E2BIG);
586 } else {
587 msr->r_msg = NULL;
588 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
589 msq->q_rtime = get_seconds();
590 wake_up_process(msr->r_tsk);
591 /*
592 * Ensure that the wakeup is visible before
593 * setting r_msg, as the receiving can otherwise
594 * exit - once r_msg is set, the receiver can
595 * continue. See lockless receive part 1 and 2
596 * in do_msgrcv(). Barrier (B).
597 */
598 smp_wmb();
599 msr->r_msg = msg;
600
601 return 1;
602 }
603 }
604 }
605
606 return 0;
607}
608
609long do_msgsnd(int msqid, long mtype, void __user *mtext,
610 size_t msgsz, int msgflg)
611{
612 struct msg_queue *msq;
613 struct msg_msg *msg;
614 int err;
615 struct ipc_namespace *ns;
616
617 ns = current->nsproxy->ipc_ns;
618
619 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
620 return -EINVAL;
621 if (mtype < 1)
622 return -EINVAL;
623
624 msg = load_msg(mtext, msgsz);
625 if (IS_ERR(msg))
626 return PTR_ERR(msg);
627
628 msg->m_type = mtype;
629 msg->m_ts = msgsz;
630
631 rcu_read_lock();
632 msq = msq_obtain_object_check(ns, msqid);
633 if (IS_ERR(msq)) {
634 err = PTR_ERR(msq);
635 goto out_unlock1;
636 }
637
638 ipc_lock_object(&msq->q_perm);
639
640 for (;;) {
641 struct msg_sender s;
642
643 err = -EACCES;
644 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
645 goto out_unlock0;
646
647 /* raced with RMID? */
648 if (!ipc_valid_object(&msq->q_perm)) {
649 err = -EIDRM;
650 goto out_unlock0;
651 }
652
653 err = security_msg_queue_msgsnd(msq, msg, msgflg);
654 if (err)
655 goto out_unlock0;
656
657 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
658 1 + msq->q_qnum <= msq->q_qbytes) {
659 break;
660 }
661
662 /* queue full, wait: */
663 if (msgflg & IPC_NOWAIT) {
664 err = -EAGAIN;
665 goto out_unlock0;
666 }
667
668 /* enqueue the sender and prepare to block */
669 ss_add(msq, &s);
670
671 if (!ipc_rcu_getref(msq)) {
672 err = -EIDRM;
673 goto out_unlock0;
674 }
675
676 ipc_unlock_object(&msq->q_perm);
677 rcu_read_unlock();
678 schedule();
679
680 rcu_read_lock();
681 ipc_lock_object(&msq->q_perm);
682
683 ipc_rcu_putref(msq, ipc_rcu_free);
684 /* raced with RMID? */
685 if (!ipc_valid_object(&msq->q_perm)) {
686 err = -EIDRM;
687 goto out_unlock0;
688 }
689
690 ss_del(&s);
691
692 if (signal_pending(current)) {
693 err = -ERESTARTNOHAND;
694 goto out_unlock0;
695 }
696
697 }
698 msq->q_lspid = task_tgid_vnr(current);
699 msq->q_stime = get_seconds();
700
701 if (!pipelined_send(msq, msg)) {
702 /* no one is waiting for this message, enqueue it */
703 list_add_tail(&msg->m_list, &msq->q_messages);
704 msq->q_cbytes += msgsz;
705 msq->q_qnum++;
706 atomic_add(msgsz, &ns->msg_bytes);
707 atomic_inc(&ns->msg_hdrs);
708 }
709
710 err = 0;
711 msg = NULL;
712
713out_unlock0:
714 ipc_unlock_object(&msq->q_perm);
715out_unlock1:
716 rcu_read_unlock();
717 if (msg != NULL)
718 free_msg(msg);
719 return err;
720}
721
722SYSCALL_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
732static 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
753static 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 */
772static 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
785static inline void free_copy(struct msg_msg *copy)
786{
787 if (copy)
788 free_msg(copy);
789}
790#else
791static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
792{
793 return ERR_PTR(-ENOSYS);
794}
795
796static inline void free_copy(struct msg_msg *copy)
797{
798}
799#endif
800
801static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
802{
803 struct msg_msg *msg, *found = NULL;
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 found = msg;
813 } else if (mode == SEARCH_NUMBER) {
814 if (*msgtyp == count)
815 return msg;
816 } else
817 return msg;
818 count++;
819 }
820 }
821
822 return found ?: ERR_PTR(-EAGAIN);
823}
824
825long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
826 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
827{
828 int mode;
829 struct msg_queue *msq;
830 struct ipc_namespace *ns;
831 struct msg_msg *msg, *copy = NULL;
832
833 ns = current->nsproxy->ipc_ns;
834
835 if (msqid < 0 || (long) bufsz < 0)
836 return -EINVAL;
837
838 if (msgflg & MSG_COPY) {
839 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
840 return -EINVAL;
841 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
842 if (IS_ERR(copy))
843 return PTR_ERR(copy);
844 }
845 mode = convert_mode(&msgtyp, msgflg);
846
847 rcu_read_lock();
848 msq = msq_obtain_object_check(ns, msqid);
849 if (IS_ERR(msq)) {
850 rcu_read_unlock();
851 free_copy(copy);
852 return PTR_ERR(msq);
853 }
854
855 for (;;) {
856 struct msg_receiver msr_d;
857
858 msg = ERR_PTR(-EACCES);
859 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
860 goto out_unlock1;
861
862 ipc_lock_object(&msq->q_perm);
863
864 /* raced with RMID? */
865 if (!ipc_valid_object(&msq->q_perm)) {
866 msg = ERR_PTR(-EIDRM);
867 goto out_unlock0;
868 }
869
870 msg = find_msg(msq, &msgtyp, mode);
871 if (!IS_ERR(msg)) {
872 /*
873 * Found a suitable message.
874 * Unlink it from the queue.
875 */
876 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
877 msg = ERR_PTR(-E2BIG);
878 goto out_unlock0;
879 }
880 /*
881 * If we are copying, then do not unlink message and do
882 * not update queue parameters.
883 */
884 if (msgflg & MSG_COPY) {
885 msg = copy_msg(msg, copy);
886 goto out_unlock0;
887 }
888
889 list_del(&msg->m_list);
890 msq->q_qnum--;
891 msq->q_rtime = get_seconds();
892 msq->q_lrpid = task_tgid_vnr(current);
893 msq->q_cbytes -= msg->m_ts;
894 atomic_sub(msg->m_ts, &ns->msg_bytes);
895 atomic_dec(&ns->msg_hdrs);
896 ss_wakeup(&msq->q_senders, 0);
897
898 goto out_unlock0;
899 }
900
901 /* No message waiting. Wait for a message */
902 if (msgflg & IPC_NOWAIT) {
903 msg = ERR_PTR(-ENOMSG);
904 goto out_unlock0;
905 }
906
907 list_add_tail(&msr_d.r_list, &msq->q_receivers);
908 msr_d.r_tsk = current;
909 msr_d.r_msgtype = msgtyp;
910 msr_d.r_mode = mode;
911 if (msgflg & MSG_NOERROR)
912 msr_d.r_maxsize = INT_MAX;
913 else
914 msr_d.r_maxsize = bufsz;
915 msr_d.r_msg = ERR_PTR(-EAGAIN);
916 __set_current_state(TASK_INTERRUPTIBLE);
917
918 ipc_unlock_object(&msq->q_perm);
919 rcu_read_unlock();
920 schedule();
921
922 /* Lockless receive, part 1:
923 * Disable preemption. We don't hold a reference to the queue
924 * and getting a reference would defeat the idea of a lockless
925 * operation, thus the code relies on rcu to guarantee the
926 * existence of msq:
927 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
928 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
929 * rcu_read_lock() prevents preemption between reading r_msg
930 * and acquiring the q_perm.lock in ipc_lock_object().
931 */
932 rcu_read_lock();
933
934 /* Lockless receive, part 2:
935 * Wait until pipelined_send or expunge_all are outside of
936 * wake_up_process(). There is a race with exit(), see
937 * ipc/mqueue.c for the details. The correct serialization
938 * ensures that a receiver cannot continue without the wakeup
939 * being visibible _before_ setting r_msg:
940 *
941 * CPU 0 CPU 1
942 * <loop receiver>
943 * smp_rmb(); (A) <-- pair -. <waker thread>
944 * <load ->r_msg> | msr->r_msg = NULL;
945 * | wake_up_process();
946 * <continue> `------> smp_wmb(); (B)
947 * msr->r_msg = msg;
948 *
949 * Where (A) orders the message value read and where (B) orders
950 * the write to the r_msg -- done in both pipelined_send and
951 * expunge_all.
952 */
953 for (;;) {
954 /*
955 * Pairs with writer barrier in pipelined_send
956 * or expunge_all.
957 */
958 smp_rmb(); /* barrier (A) */
959 msg = (struct msg_msg *)msr_d.r_msg;
960 if (msg)
961 break;
962
963 /*
964 * The cpu_relax() call is a compiler barrier
965 * which forces everything in this loop to be
966 * re-loaded.
967 */
968 cpu_relax();
969 }
970
971 /* Lockless receive, part 3:
972 * If there is a message or an error then accept it without
973 * locking.
974 */
975 if (msg != ERR_PTR(-EAGAIN))
976 goto out_unlock1;
977
978 /* Lockless receive, part 3:
979 * Acquire the queue spinlock.
980 */
981 ipc_lock_object(&msq->q_perm);
982
983 /* Lockless receive, part 4:
984 * Repeat test after acquiring the spinlock.
985 */
986 msg = (struct msg_msg *)msr_d.r_msg;
987 if (msg != ERR_PTR(-EAGAIN))
988 goto out_unlock0;
989
990 list_del(&msr_d.r_list);
991 if (signal_pending(current)) {
992 msg = ERR_PTR(-ERESTARTNOHAND);
993 goto out_unlock0;
994 }
995
996 ipc_unlock_object(&msq->q_perm);
997 }
998
999out_unlock0:
1000 ipc_unlock_object(&msq->q_perm);
1001out_unlock1:
1002 rcu_read_unlock();
1003 if (IS_ERR(msg)) {
1004 free_copy(copy);
1005 return PTR_ERR(msg);
1006 }
1007
1008 bufsz = msg_handler(buf, msg, bufsz);
1009 free_msg(msg);
1010
1011 return bufsz;
1012}
1013
1014SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1015 long, msgtyp, int, msgflg)
1016{
1017 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1018}
1019
1020
1021void msg_init_ns(struct ipc_namespace *ns)
1022{
1023 ns->msg_ctlmax = MSGMAX;
1024 ns->msg_ctlmnb = MSGMNB;
1025 ns->msg_ctlmni = MSGMNI;
1026
1027 atomic_set(&ns->msg_bytes, 0);
1028 atomic_set(&ns->msg_hdrs, 0);
1029 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1030}
1031
1032#ifdef CONFIG_IPC_NS
1033void msg_exit_ns(struct ipc_namespace *ns)
1034{
1035 free_ipcs(ns, &msg_ids(ns), freeque);
1036 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1037}
1038#endif
1039
1040#ifdef CONFIG_PROC_FS
1041static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1042{
1043 struct user_namespace *user_ns = seq_user_ns(s);
1044 struct msg_queue *msq = it;
1045
1046 seq_printf(s,
1047 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1048 msq->q_perm.key,
1049 msq->q_perm.id,
1050 msq->q_perm.mode,
1051 msq->q_cbytes,
1052 msq->q_qnum,
1053 msq->q_lspid,
1054 msq->q_lrpid,
1055 from_kuid_munged(user_ns, msq->q_perm.uid),
1056 from_kgid_munged(user_ns, msq->q_perm.gid),
1057 from_kuid_munged(user_ns, msq->q_perm.cuid),
1058 from_kgid_munged(user_ns, msq->q_perm.cgid),
1059 msq->q_stime,
1060 msq->q_rtime,
1061 msq->q_ctime);
1062
1063 return 0;
1064}
1065#endif
1066
1067void __init msg_init(void)
1068{
1069 msg_init_ns(&init_ipc_ns);
1070
1071 ipc_init_proc_interface("sysvipc/msg",
1072 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1073 IPC_MSG_IDS, sysvipc_msg_proc_show);
1074}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/msg.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 *
6 * Removed all the remaining kerneld mess
7 * Catch the -EFAULT stuff properly
8 * Use GFP_KERNEL for messages as in 1.2
9 * Fixed up the unchecked user space derefs
10 * Copyright (C) 1998 Alan Cox & Andi Kleen
11 *
12 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
13 *
14 * mostly rewritten, threaded and wake-one semantics added
15 * MSGMAX limit removed, sysctl's added
16 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
17 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
20 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
24 */
25
26#include <linux/capability.h>
27#include <linux/msg.h>
28#include <linux/spinlock.h>
29#include <linux/init.h>
30#include <linux/mm.h>
31#include <linux/proc_fs.h>
32#include <linux/list.h>
33#include <linux/security.h>
34#include <linux/sched/wake_q.h>
35#include <linux/syscalls.h>
36#include <linux/audit.h>
37#include <linux/seq_file.h>
38#include <linux/rwsem.h>
39#include <linux/nsproxy.h>
40#include <linux/ipc_namespace.h>
41#include <linux/rhashtable.h>
42
43#include <asm/current.h>
44#include <linux/uaccess.h>
45#include "util.h"
46
47/* one msq_queue structure for each present queue on the system */
48struct msg_queue {
49 struct kern_ipc_perm q_perm;
50 time64_t q_stime; /* last msgsnd time */
51 time64_t q_rtime; /* last msgrcv time */
52 time64_t q_ctime; /* last change time */
53 unsigned long q_cbytes; /* current number of bytes on queue */
54 unsigned long q_qnum; /* number of messages in queue */
55 unsigned long q_qbytes; /* max number of bytes on queue */
56 struct pid *q_lspid; /* pid of last msgsnd */
57 struct pid *q_lrpid; /* last receive pid */
58
59 struct list_head q_messages;
60 struct list_head q_receivers;
61 struct list_head q_senders;
62} __randomize_layout;
63
64/*
65 * MSG_BARRIER Locking:
66 *
67 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
68 * does not acquire any locks when it sees that a message exists in
69 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
70 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
71 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
72 */
73
74/* one msg_receiver structure for each sleeping receiver */
75struct msg_receiver {
76 struct list_head r_list;
77 struct task_struct *r_tsk;
78
79 int r_mode;
80 long r_msgtype;
81 long r_maxsize;
82
83 struct msg_msg *r_msg;
84};
85
86/* one msg_sender for each sleeping sender */
87struct msg_sender {
88 struct list_head list;
89 struct task_struct *tsk;
90 size_t msgsz;
91};
92
93#define SEARCH_ANY 1
94#define SEARCH_EQUAL 2
95#define SEARCH_NOTEQUAL 3
96#define SEARCH_LESSEQUAL 4
97#define SEARCH_NUMBER 5
98
99#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
100
101static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
102{
103 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
104
105 if (IS_ERR(ipcp))
106 return ERR_CAST(ipcp);
107
108 return container_of(ipcp, struct msg_queue, q_perm);
109}
110
111static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
112 int id)
113{
114 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
115
116 if (IS_ERR(ipcp))
117 return ERR_CAST(ipcp);
118
119 return container_of(ipcp, struct msg_queue, q_perm);
120}
121
122static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
123{
124 ipc_rmid(&msg_ids(ns), &s->q_perm);
125}
126
127static void msg_rcu_free(struct rcu_head *head)
128{
129 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
130 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
131
132 security_msg_queue_free(&msq->q_perm);
133 kfree(msq);
134}
135
136/**
137 * newque - Create a new msg queue
138 * @ns: namespace
139 * @params: ptr to the structure that contains the key and msgflg
140 *
141 * Called with msg_ids.rwsem held (writer)
142 */
143static int newque(struct ipc_namespace *ns, struct ipc_params *params)
144{
145 struct msg_queue *msq;
146 int retval;
147 key_t key = params->key;
148 int msgflg = params->flg;
149
150 msq = kmalloc(sizeof(*msq), GFP_KERNEL);
151 if (unlikely(!msq))
152 return -ENOMEM;
153
154 msq->q_perm.mode = msgflg & S_IRWXUGO;
155 msq->q_perm.key = key;
156
157 msq->q_perm.security = NULL;
158 retval = security_msg_queue_alloc(&msq->q_perm);
159 if (retval) {
160 kfree(msq);
161 return retval;
162 }
163
164 msq->q_stime = msq->q_rtime = 0;
165 msq->q_ctime = ktime_get_real_seconds();
166 msq->q_cbytes = msq->q_qnum = 0;
167 msq->q_qbytes = ns->msg_ctlmnb;
168 msq->q_lspid = msq->q_lrpid = NULL;
169 INIT_LIST_HEAD(&msq->q_messages);
170 INIT_LIST_HEAD(&msq->q_receivers);
171 INIT_LIST_HEAD(&msq->q_senders);
172
173 /* ipc_addid() locks msq upon success. */
174 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
175 if (retval < 0) {
176 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
177 return retval;
178 }
179
180 ipc_unlock_object(&msq->q_perm);
181 rcu_read_unlock();
182
183 return msq->q_perm.id;
184}
185
186static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
187{
188 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
189 1 + msq->q_qnum <= msq->q_qbytes;
190}
191
192static inline void ss_add(struct msg_queue *msq,
193 struct msg_sender *mss, size_t msgsz)
194{
195 mss->tsk = current;
196 mss->msgsz = msgsz;
197 /*
198 * No memory barrier required: we did ipc_lock_object(),
199 * and the waker obtains that lock before calling wake_q_add().
200 */
201 __set_current_state(TASK_INTERRUPTIBLE);
202 list_add_tail(&mss->list, &msq->q_senders);
203}
204
205static inline void ss_del(struct msg_sender *mss)
206{
207 if (mss->list.next)
208 list_del(&mss->list);
209}
210
211static void ss_wakeup(struct msg_queue *msq,
212 struct wake_q_head *wake_q, bool kill)
213{
214 struct msg_sender *mss, *t;
215 struct task_struct *stop_tsk = NULL;
216 struct list_head *h = &msq->q_senders;
217
218 list_for_each_entry_safe(mss, t, h, list) {
219 if (kill)
220 mss->list.next = NULL;
221
222 /*
223 * Stop at the first task we don't wakeup,
224 * we've already iterated the original
225 * sender queue.
226 */
227 else if (stop_tsk == mss->tsk)
228 break;
229 /*
230 * We are not in an EIDRM scenario here, therefore
231 * verify that we really need to wakeup the task.
232 * To maintain current semantics and wakeup order,
233 * move the sender to the tail on behalf of the
234 * blocked task.
235 */
236 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
237 if (!stop_tsk)
238 stop_tsk = mss->tsk;
239
240 list_move_tail(&mss->list, &msq->q_senders);
241 continue;
242 }
243
244 wake_q_add(wake_q, mss->tsk);
245 }
246}
247
248static void expunge_all(struct msg_queue *msq, int res,
249 struct wake_q_head *wake_q)
250{
251 struct msg_receiver *msr, *t;
252
253 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
254 struct task_struct *r_tsk;
255
256 r_tsk = get_task_struct(msr->r_tsk);
257
258 /* see MSG_BARRIER for purpose/pairing */
259 smp_store_release(&msr->r_msg, ERR_PTR(res));
260 wake_q_add_safe(wake_q, r_tsk);
261 }
262}
263
264/*
265 * freeque() wakes up waiters on the sender and receiver waiting queue,
266 * removes the message queue from message queue ID IDR, and cleans up all the
267 * messages associated with this queue.
268 *
269 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
270 * before freeque() is called. msg_ids.rwsem remains locked on exit.
271 */
272static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
273 __releases(RCU)
274 __releases(&msq->q_perm)
275{
276 struct msg_msg *msg, *t;
277 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
278 DEFINE_WAKE_Q(wake_q);
279
280 expunge_all(msq, -EIDRM, &wake_q);
281 ss_wakeup(msq, &wake_q, true);
282 msg_rmid(ns, msq);
283 ipc_unlock_object(&msq->q_perm);
284 wake_up_q(&wake_q);
285 rcu_read_unlock();
286
287 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
288 atomic_dec(&ns->msg_hdrs);
289 free_msg(msg);
290 }
291 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
292 ipc_update_pid(&msq->q_lspid, NULL);
293 ipc_update_pid(&msq->q_lrpid, NULL);
294 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
295}
296
297long ksys_msgget(key_t key, int msgflg)
298{
299 struct ipc_namespace *ns;
300 static const struct ipc_ops msg_ops = {
301 .getnew = newque,
302 .associate = security_msg_queue_associate,
303 };
304 struct ipc_params msg_params;
305
306 ns = current->nsproxy->ipc_ns;
307
308 msg_params.key = key;
309 msg_params.flg = msgflg;
310
311 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
312}
313
314SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
315{
316 return ksys_msgget(key, msgflg);
317}
318
319static inline unsigned long
320copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
321{
322 switch (version) {
323 case IPC_64:
324 return copy_to_user(buf, in, sizeof(*in));
325 case IPC_OLD:
326 {
327 struct msqid_ds out;
328
329 memset(&out, 0, sizeof(out));
330
331 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
332
333 out.msg_stime = in->msg_stime;
334 out.msg_rtime = in->msg_rtime;
335 out.msg_ctime = in->msg_ctime;
336
337 if (in->msg_cbytes > USHRT_MAX)
338 out.msg_cbytes = USHRT_MAX;
339 else
340 out.msg_cbytes = in->msg_cbytes;
341 out.msg_lcbytes = in->msg_cbytes;
342
343 if (in->msg_qnum > USHRT_MAX)
344 out.msg_qnum = USHRT_MAX;
345 else
346 out.msg_qnum = in->msg_qnum;
347
348 if (in->msg_qbytes > USHRT_MAX)
349 out.msg_qbytes = USHRT_MAX;
350 else
351 out.msg_qbytes = in->msg_qbytes;
352 out.msg_lqbytes = in->msg_qbytes;
353
354 out.msg_lspid = in->msg_lspid;
355 out.msg_lrpid = in->msg_lrpid;
356
357 return copy_to_user(buf, &out, sizeof(out));
358 }
359 default:
360 return -EINVAL;
361 }
362}
363
364static inline unsigned long
365copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
366{
367 switch (version) {
368 case IPC_64:
369 if (copy_from_user(out, buf, sizeof(*out)))
370 return -EFAULT;
371 return 0;
372 case IPC_OLD:
373 {
374 struct msqid_ds tbuf_old;
375
376 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
377 return -EFAULT;
378
379 out->msg_perm.uid = tbuf_old.msg_perm.uid;
380 out->msg_perm.gid = tbuf_old.msg_perm.gid;
381 out->msg_perm.mode = tbuf_old.msg_perm.mode;
382
383 if (tbuf_old.msg_qbytes == 0)
384 out->msg_qbytes = tbuf_old.msg_lqbytes;
385 else
386 out->msg_qbytes = tbuf_old.msg_qbytes;
387
388 return 0;
389 }
390 default:
391 return -EINVAL;
392 }
393}
394
395/*
396 * This function handles some msgctl commands which require the rwsem
397 * to be held in write mode.
398 * NOTE: no locks must be held, the rwsem is taken inside this function.
399 */
400static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
401 struct ipc64_perm *perm, int msg_qbytes)
402{
403 struct kern_ipc_perm *ipcp;
404 struct msg_queue *msq;
405 int err;
406
407 down_write(&msg_ids(ns).rwsem);
408 rcu_read_lock();
409
410 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
411 perm, msg_qbytes);
412 if (IS_ERR(ipcp)) {
413 err = PTR_ERR(ipcp);
414 goto out_unlock1;
415 }
416
417 msq = container_of(ipcp, struct msg_queue, q_perm);
418
419 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
420 if (err)
421 goto out_unlock1;
422
423 switch (cmd) {
424 case IPC_RMID:
425 ipc_lock_object(&msq->q_perm);
426 /* freeque unlocks the ipc object and rcu */
427 freeque(ns, ipcp);
428 goto out_up;
429 case IPC_SET:
430 {
431 DEFINE_WAKE_Q(wake_q);
432
433 if (msg_qbytes > ns->msg_ctlmnb &&
434 !capable(CAP_SYS_RESOURCE)) {
435 err = -EPERM;
436 goto out_unlock1;
437 }
438
439 ipc_lock_object(&msq->q_perm);
440 err = ipc_update_perm(perm, ipcp);
441 if (err)
442 goto out_unlock0;
443
444 msq->q_qbytes = msg_qbytes;
445
446 msq->q_ctime = ktime_get_real_seconds();
447 /*
448 * Sleeping receivers might be excluded by
449 * stricter permissions.
450 */
451 expunge_all(msq, -EAGAIN, &wake_q);
452 /*
453 * Sleeping senders might be able to send
454 * due to a larger queue size.
455 */
456 ss_wakeup(msq, &wake_q, false);
457 ipc_unlock_object(&msq->q_perm);
458 wake_up_q(&wake_q);
459
460 goto out_unlock1;
461 }
462 default:
463 err = -EINVAL;
464 goto out_unlock1;
465 }
466
467out_unlock0:
468 ipc_unlock_object(&msq->q_perm);
469out_unlock1:
470 rcu_read_unlock();
471out_up:
472 up_write(&msg_ids(ns).rwsem);
473 return err;
474}
475
476static int msgctl_info(struct ipc_namespace *ns, int msqid,
477 int cmd, struct msginfo *msginfo)
478{
479 int err;
480 int max_idx;
481
482 /*
483 * We must not return kernel stack data.
484 * due to padding, it's not enough
485 * to set all member fields.
486 */
487 err = security_msg_queue_msgctl(NULL, cmd);
488 if (err)
489 return err;
490
491 memset(msginfo, 0, sizeof(*msginfo));
492 msginfo->msgmni = ns->msg_ctlmni;
493 msginfo->msgmax = ns->msg_ctlmax;
494 msginfo->msgmnb = ns->msg_ctlmnb;
495 msginfo->msgssz = MSGSSZ;
496 msginfo->msgseg = MSGSEG;
497 down_read(&msg_ids(ns).rwsem);
498 if (cmd == MSG_INFO) {
499 msginfo->msgpool = msg_ids(ns).in_use;
500 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
501 msginfo->msgtql = atomic_read(&ns->msg_bytes);
502 } else {
503 msginfo->msgmap = MSGMAP;
504 msginfo->msgpool = MSGPOOL;
505 msginfo->msgtql = MSGTQL;
506 }
507 max_idx = ipc_get_maxidx(&msg_ids(ns));
508 up_read(&msg_ids(ns).rwsem);
509 return (max_idx < 0) ? 0 : max_idx;
510}
511
512static int msgctl_stat(struct ipc_namespace *ns, int msqid,
513 int cmd, struct msqid64_ds *p)
514{
515 struct msg_queue *msq;
516 int err;
517
518 memset(p, 0, sizeof(*p));
519
520 rcu_read_lock();
521 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
522 msq = msq_obtain_object(ns, msqid);
523 if (IS_ERR(msq)) {
524 err = PTR_ERR(msq);
525 goto out_unlock;
526 }
527 } else { /* IPC_STAT */
528 msq = msq_obtain_object_check(ns, msqid);
529 if (IS_ERR(msq)) {
530 err = PTR_ERR(msq);
531 goto out_unlock;
532 }
533 }
534
535 /* see comment for SHM_STAT_ANY */
536 if (cmd == MSG_STAT_ANY)
537 audit_ipc_obj(&msq->q_perm);
538 else {
539 err = -EACCES;
540 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
541 goto out_unlock;
542 }
543
544 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
545 if (err)
546 goto out_unlock;
547
548 ipc_lock_object(&msq->q_perm);
549
550 if (!ipc_valid_object(&msq->q_perm)) {
551 ipc_unlock_object(&msq->q_perm);
552 err = -EIDRM;
553 goto out_unlock;
554 }
555
556 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
557 p->msg_stime = msq->q_stime;
558 p->msg_rtime = msq->q_rtime;
559 p->msg_ctime = msq->q_ctime;
560#ifndef CONFIG_64BIT
561 p->msg_stime_high = msq->q_stime >> 32;
562 p->msg_rtime_high = msq->q_rtime >> 32;
563 p->msg_ctime_high = msq->q_ctime >> 32;
564#endif
565 p->msg_cbytes = msq->q_cbytes;
566 p->msg_qnum = msq->q_qnum;
567 p->msg_qbytes = msq->q_qbytes;
568 p->msg_lspid = pid_vnr(msq->q_lspid);
569 p->msg_lrpid = pid_vnr(msq->q_lrpid);
570
571 if (cmd == IPC_STAT) {
572 /*
573 * As defined in SUS:
574 * Return 0 on success
575 */
576 err = 0;
577 } else {
578 /*
579 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
580 * Return the full id, including the sequence number
581 */
582 err = msq->q_perm.id;
583 }
584
585 ipc_unlock_object(&msq->q_perm);
586out_unlock:
587 rcu_read_unlock();
588 return err;
589}
590
591static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
592{
593 struct ipc_namespace *ns;
594 struct msqid64_ds msqid64;
595 int err;
596
597 if (msqid < 0 || cmd < 0)
598 return -EINVAL;
599
600 ns = current->nsproxy->ipc_ns;
601
602 switch (cmd) {
603 case IPC_INFO:
604 case MSG_INFO: {
605 struct msginfo msginfo;
606 err = msgctl_info(ns, msqid, cmd, &msginfo);
607 if (err < 0)
608 return err;
609 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
610 err = -EFAULT;
611 return err;
612 }
613 case MSG_STAT: /* msqid is an index rather than a msg queue id */
614 case MSG_STAT_ANY:
615 case IPC_STAT:
616 err = msgctl_stat(ns, msqid, cmd, &msqid64);
617 if (err < 0)
618 return err;
619 if (copy_msqid_to_user(buf, &msqid64, version))
620 err = -EFAULT;
621 return err;
622 case IPC_SET:
623 if (copy_msqid_from_user(&msqid64, buf, version))
624 return -EFAULT;
625 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
626 msqid64.msg_qbytes);
627 case IPC_RMID:
628 return msgctl_down(ns, msqid, cmd, NULL, 0);
629 default:
630 return -EINVAL;
631 }
632}
633
634SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
635{
636 return ksys_msgctl(msqid, cmd, buf, IPC_64);
637}
638
639#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
640long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
641{
642 int version = ipc_parse_version(&cmd);
643
644 return ksys_msgctl(msqid, cmd, buf, version);
645}
646
647SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
648{
649 return ksys_old_msgctl(msqid, cmd, buf);
650}
651#endif
652
653#ifdef CONFIG_COMPAT
654
655struct compat_msqid_ds {
656 struct compat_ipc_perm msg_perm;
657 compat_uptr_t msg_first;
658 compat_uptr_t msg_last;
659 old_time32_t msg_stime;
660 old_time32_t msg_rtime;
661 old_time32_t msg_ctime;
662 compat_ulong_t msg_lcbytes;
663 compat_ulong_t msg_lqbytes;
664 unsigned short msg_cbytes;
665 unsigned short msg_qnum;
666 unsigned short msg_qbytes;
667 compat_ipc_pid_t msg_lspid;
668 compat_ipc_pid_t msg_lrpid;
669};
670
671static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
672 int version)
673{
674 memset(out, 0, sizeof(*out));
675 if (version == IPC_64) {
676 struct compat_msqid64_ds __user *p = buf;
677 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
678 return -EFAULT;
679 if (get_user(out->msg_qbytes, &p->msg_qbytes))
680 return -EFAULT;
681 } else {
682 struct compat_msqid_ds __user *p = buf;
683 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
684 return -EFAULT;
685 if (get_user(out->msg_qbytes, &p->msg_qbytes))
686 return -EFAULT;
687 }
688 return 0;
689}
690
691static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
692 int version)
693{
694 if (version == IPC_64) {
695 struct compat_msqid64_ds v;
696 memset(&v, 0, sizeof(v));
697 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
698 v.msg_stime = lower_32_bits(in->msg_stime);
699 v.msg_stime_high = upper_32_bits(in->msg_stime);
700 v.msg_rtime = lower_32_bits(in->msg_rtime);
701 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
702 v.msg_ctime = lower_32_bits(in->msg_ctime);
703 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
704 v.msg_cbytes = in->msg_cbytes;
705 v.msg_qnum = in->msg_qnum;
706 v.msg_qbytes = in->msg_qbytes;
707 v.msg_lspid = in->msg_lspid;
708 v.msg_lrpid = in->msg_lrpid;
709 return copy_to_user(buf, &v, sizeof(v));
710 } else {
711 struct compat_msqid_ds v;
712 memset(&v, 0, sizeof(v));
713 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
714 v.msg_stime = in->msg_stime;
715 v.msg_rtime = in->msg_rtime;
716 v.msg_ctime = in->msg_ctime;
717 v.msg_cbytes = in->msg_cbytes;
718 v.msg_qnum = in->msg_qnum;
719 v.msg_qbytes = in->msg_qbytes;
720 v.msg_lspid = in->msg_lspid;
721 v.msg_lrpid = in->msg_lrpid;
722 return copy_to_user(buf, &v, sizeof(v));
723 }
724}
725
726static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
727{
728 struct ipc_namespace *ns;
729 int err;
730 struct msqid64_ds msqid64;
731
732 ns = current->nsproxy->ipc_ns;
733
734 if (msqid < 0 || cmd < 0)
735 return -EINVAL;
736
737 switch (cmd & (~IPC_64)) {
738 case IPC_INFO:
739 case MSG_INFO: {
740 struct msginfo msginfo;
741 err = msgctl_info(ns, msqid, cmd, &msginfo);
742 if (err < 0)
743 return err;
744 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
745 err = -EFAULT;
746 return err;
747 }
748 case IPC_STAT:
749 case MSG_STAT:
750 case MSG_STAT_ANY:
751 err = msgctl_stat(ns, msqid, cmd, &msqid64);
752 if (err < 0)
753 return err;
754 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
755 err = -EFAULT;
756 return err;
757 case IPC_SET:
758 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
759 return -EFAULT;
760 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
761 case IPC_RMID:
762 return msgctl_down(ns, msqid, cmd, NULL, 0);
763 default:
764 return -EINVAL;
765 }
766}
767
768COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
769{
770 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
771}
772
773#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
774long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
775{
776 int version = compat_ipc_parse_version(&cmd);
777
778 return compat_ksys_msgctl(msqid, cmd, uptr, version);
779}
780
781COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
782{
783 return compat_ksys_old_msgctl(msqid, cmd, uptr);
784}
785#endif
786#endif
787
788static int testmsg(struct msg_msg *msg, long type, int mode)
789{
790 switch (mode) {
791 case SEARCH_ANY:
792 case SEARCH_NUMBER:
793 return 1;
794 case SEARCH_LESSEQUAL:
795 if (msg->m_type <= type)
796 return 1;
797 break;
798 case SEARCH_EQUAL:
799 if (msg->m_type == type)
800 return 1;
801 break;
802 case SEARCH_NOTEQUAL:
803 if (msg->m_type != type)
804 return 1;
805 break;
806 }
807 return 0;
808}
809
810static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
811 struct wake_q_head *wake_q)
812{
813 struct msg_receiver *msr, *t;
814
815 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
816 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
817 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
818 msr->r_msgtype, msr->r_mode)) {
819
820 list_del(&msr->r_list);
821 if (msr->r_maxsize < msg->m_ts) {
822 wake_q_add(wake_q, msr->r_tsk);
823
824 /* See expunge_all regarding memory barrier */
825 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
826 } else {
827 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
828 msq->q_rtime = ktime_get_real_seconds();
829
830 wake_q_add(wake_q, msr->r_tsk);
831
832 /* See expunge_all regarding memory barrier */
833 smp_store_release(&msr->r_msg, msg);
834 return 1;
835 }
836 }
837 }
838
839 return 0;
840}
841
842static long do_msgsnd(int msqid, long mtype, void __user *mtext,
843 size_t msgsz, int msgflg)
844{
845 struct msg_queue *msq;
846 struct msg_msg *msg;
847 int err;
848 struct ipc_namespace *ns;
849 DEFINE_WAKE_Q(wake_q);
850
851 ns = current->nsproxy->ipc_ns;
852
853 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
854 return -EINVAL;
855 if (mtype < 1)
856 return -EINVAL;
857
858 msg = load_msg(mtext, msgsz);
859 if (IS_ERR(msg))
860 return PTR_ERR(msg);
861
862 msg->m_type = mtype;
863 msg->m_ts = msgsz;
864
865 rcu_read_lock();
866 msq = msq_obtain_object_check(ns, msqid);
867 if (IS_ERR(msq)) {
868 err = PTR_ERR(msq);
869 goto out_unlock1;
870 }
871
872 ipc_lock_object(&msq->q_perm);
873
874 for (;;) {
875 struct msg_sender s;
876
877 err = -EACCES;
878 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
879 goto out_unlock0;
880
881 /* raced with RMID? */
882 if (!ipc_valid_object(&msq->q_perm)) {
883 err = -EIDRM;
884 goto out_unlock0;
885 }
886
887 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
888 if (err)
889 goto out_unlock0;
890
891 if (msg_fits_inqueue(msq, msgsz))
892 break;
893
894 /* queue full, wait: */
895 if (msgflg & IPC_NOWAIT) {
896 err = -EAGAIN;
897 goto out_unlock0;
898 }
899
900 /* enqueue the sender and prepare to block */
901 ss_add(msq, &s, msgsz);
902
903 if (!ipc_rcu_getref(&msq->q_perm)) {
904 err = -EIDRM;
905 goto out_unlock0;
906 }
907
908 ipc_unlock_object(&msq->q_perm);
909 rcu_read_unlock();
910 schedule();
911
912 rcu_read_lock();
913 ipc_lock_object(&msq->q_perm);
914
915 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
916 /* raced with RMID? */
917 if (!ipc_valid_object(&msq->q_perm)) {
918 err = -EIDRM;
919 goto out_unlock0;
920 }
921 ss_del(&s);
922
923 if (signal_pending(current)) {
924 err = -ERESTARTNOHAND;
925 goto out_unlock0;
926 }
927
928 }
929
930 ipc_update_pid(&msq->q_lspid, task_tgid(current));
931 msq->q_stime = ktime_get_real_seconds();
932
933 if (!pipelined_send(msq, msg, &wake_q)) {
934 /* no one is waiting for this message, enqueue it */
935 list_add_tail(&msg->m_list, &msq->q_messages);
936 msq->q_cbytes += msgsz;
937 msq->q_qnum++;
938 atomic_add(msgsz, &ns->msg_bytes);
939 atomic_inc(&ns->msg_hdrs);
940 }
941
942 err = 0;
943 msg = NULL;
944
945out_unlock0:
946 ipc_unlock_object(&msq->q_perm);
947 wake_up_q(&wake_q);
948out_unlock1:
949 rcu_read_unlock();
950 if (msg != NULL)
951 free_msg(msg);
952 return err;
953}
954
955long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
956 int msgflg)
957{
958 long mtype;
959
960 if (get_user(mtype, &msgp->mtype))
961 return -EFAULT;
962 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
963}
964
965SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
966 int, msgflg)
967{
968 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
969}
970
971#ifdef CONFIG_COMPAT
972
973struct compat_msgbuf {
974 compat_long_t mtype;
975 char mtext[1];
976};
977
978long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
979 compat_ssize_t msgsz, int msgflg)
980{
981 struct compat_msgbuf __user *up = compat_ptr(msgp);
982 compat_long_t mtype;
983
984 if (get_user(mtype, &up->mtype))
985 return -EFAULT;
986 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
987}
988
989COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
990 compat_ssize_t, msgsz, int, msgflg)
991{
992 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
993}
994#endif
995
996static inline int convert_mode(long *msgtyp, int msgflg)
997{
998 if (msgflg & MSG_COPY)
999 return SEARCH_NUMBER;
1000 /*
1001 * find message of correct type.
1002 * msgtyp = 0 => get first.
1003 * msgtyp > 0 => get first message of matching type.
1004 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1005 */
1006 if (*msgtyp == 0)
1007 return SEARCH_ANY;
1008 if (*msgtyp < 0) {
1009 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1010 *msgtyp = LONG_MAX;
1011 else
1012 *msgtyp = -*msgtyp;
1013 return SEARCH_LESSEQUAL;
1014 }
1015 if (msgflg & MSG_EXCEPT)
1016 return SEARCH_NOTEQUAL;
1017 return SEARCH_EQUAL;
1018}
1019
1020static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1021{
1022 struct msgbuf __user *msgp = dest;
1023 size_t msgsz;
1024
1025 if (put_user(msg->m_type, &msgp->mtype))
1026 return -EFAULT;
1027
1028 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1029 if (store_msg(msgp->mtext, msg, msgsz))
1030 return -EFAULT;
1031 return msgsz;
1032}
1033
1034#ifdef CONFIG_CHECKPOINT_RESTORE
1035/*
1036 * This function creates new kernel message structure, large enough to store
1037 * bufsz message bytes.
1038 */
1039static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1040{
1041 struct msg_msg *copy;
1042
1043 /*
1044 * Create dummy message to copy real message to.
1045 */
1046 copy = load_msg(buf, bufsz);
1047 if (!IS_ERR(copy))
1048 copy->m_ts = bufsz;
1049 return copy;
1050}
1051
1052static inline void free_copy(struct msg_msg *copy)
1053{
1054 if (copy)
1055 free_msg(copy);
1056}
1057#else
1058static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1059{
1060 return ERR_PTR(-ENOSYS);
1061}
1062
1063static inline void free_copy(struct msg_msg *copy)
1064{
1065}
1066#endif
1067
1068static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1069{
1070 struct msg_msg *msg, *found = NULL;
1071 long count = 0;
1072
1073 list_for_each_entry(msg, &msq->q_messages, m_list) {
1074 if (testmsg(msg, *msgtyp, mode) &&
1075 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1076 *msgtyp, mode)) {
1077 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1078 *msgtyp = msg->m_type - 1;
1079 found = msg;
1080 } else if (mode == SEARCH_NUMBER) {
1081 if (*msgtyp == count)
1082 return msg;
1083 } else
1084 return msg;
1085 count++;
1086 }
1087 }
1088
1089 return found ?: ERR_PTR(-EAGAIN);
1090}
1091
1092static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1093 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1094{
1095 int mode;
1096 struct msg_queue *msq;
1097 struct ipc_namespace *ns;
1098 struct msg_msg *msg, *copy = NULL;
1099 DEFINE_WAKE_Q(wake_q);
1100
1101 ns = current->nsproxy->ipc_ns;
1102
1103 if (msqid < 0 || (long) bufsz < 0)
1104 return -EINVAL;
1105
1106 if (msgflg & MSG_COPY) {
1107 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1108 return -EINVAL;
1109 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1110 if (IS_ERR(copy))
1111 return PTR_ERR(copy);
1112 }
1113 mode = convert_mode(&msgtyp, msgflg);
1114
1115 rcu_read_lock();
1116 msq = msq_obtain_object_check(ns, msqid);
1117 if (IS_ERR(msq)) {
1118 rcu_read_unlock();
1119 free_copy(copy);
1120 return PTR_ERR(msq);
1121 }
1122
1123 for (;;) {
1124 struct msg_receiver msr_d;
1125
1126 msg = ERR_PTR(-EACCES);
1127 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1128 goto out_unlock1;
1129
1130 ipc_lock_object(&msq->q_perm);
1131
1132 /* raced with RMID? */
1133 if (!ipc_valid_object(&msq->q_perm)) {
1134 msg = ERR_PTR(-EIDRM);
1135 goto out_unlock0;
1136 }
1137
1138 msg = find_msg(msq, &msgtyp, mode);
1139 if (!IS_ERR(msg)) {
1140 /*
1141 * Found a suitable message.
1142 * Unlink it from the queue.
1143 */
1144 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1145 msg = ERR_PTR(-E2BIG);
1146 goto out_unlock0;
1147 }
1148 /*
1149 * If we are copying, then do not unlink message and do
1150 * not update queue parameters.
1151 */
1152 if (msgflg & MSG_COPY) {
1153 msg = copy_msg(msg, copy);
1154 goto out_unlock0;
1155 }
1156
1157 list_del(&msg->m_list);
1158 msq->q_qnum--;
1159 msq->q_rtime = ktime_get_real_seconds();
1160 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1161 msq->q_cbytes -= msg->m_ts;
1162 atomic_sub(msg->m_ts, &ns->msg_bytes);
1163 atomic_dec(&ns->msg_hdrs);
1164 ss_wakeup(msq, &wake_q, false);
1165
1166 goto out_unlock0;
1167 }
1168
1169 /* No message waiting. Wait for a message */
1170 if (msgflg & IPC_NOWAIT) {
1171 msg = ERR_PTR(-ENOMSG);
1172 goto out_unlock0;
1173 }
1174
1175 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1176 msr_d.r_tsk = current;
1177 msr_d.r_msgtype = msgtyp;
1178 msr_d.r_mode = mode;
1179 if (msgflg & MSG_NOERROR)
1180 msr_d.r_maxsize = INT_MAX;
1181 else
1182 msr_d.r_maxsize = bufsz;
1183
1184 /* memory barrier not require due to ipc_lock_object() */
1185 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1186
1187 /* memory barrier not required, we own ipc_lock_object() */
1188 __set_current_state(TASK_INTERRUPTIBLE);
1189
1190 ipc_unlock_object(&msq->q_perm);
1191 rcu_read_unlock();
1192 schedule();
1193
1194 /*
1195 * Lockless receive, part 1:
1196 * We don't hold a reference to the queue and getting a
1197 * reference would defeat the idea of a lockless operation,
1198 * thus the code relies on rcu to guarantee the existence of
1199 * msq:
1200 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1201 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1202 */
1203 rcu_read_lock();
1204
1205 /*
1206 * Lockless receive, part 2:
1207 * The work in pipelined_send() and expunge_all():
1208 * - Set pointer to message
1209 * - Queue the receiver task for later wakeup
1210 * - Wake up the process after the lock is dropped.
1211 *
1212 * Should the process wake up before this wakeup (due to a
1213 * signal) it will either see the message and continue ...
1214 */
1215 msg = READ_ONCE(msr_d.r_msg);
1216 if (msg != ERR_PTR(-EAGAIN)) {
1217 /* see MSG_BARRIER for purpose/pairing */
1218 smp_acquire__after_ctrl_dep();
1219
1220 goto out_unlock1;
1221 }
1222
1223 /*
1224 * ... or see -EAGAIN, acquire the lock to check the message
1225 * again.
1226 */
1227 ipc_lock_object(&msq->q_perm);
1228
1229 msg = READ_ONCE(msr_d.r_msg);
1230 if (msg != ERR_PTR(-EAGAIN))
1231 goto out_unlock0;
1232
1233 list_del(&msr_d.r_list);
1234 if (signal_pending(current)) {
1235 msg = ERR_PTR(-ERESTARTNOHAND);
1236 goto out_unlock0;
1237 }
1238
1239 ipc_unlock_object(&msq->q_perm);
1240 }
1241
1242out_unlock0:
1243 ipc_unlock_object(&msq->q_perm);
1244 wake_up_q(&wake_q);
1245out_unlock1:
1246 rcu_read_unlock();
1247 if (IS_ERR(msg)) {
1248 free_copy(copy);
1249 return PTR_ERR(msg);
1250 }
1251
1252 bufsz = msg_handler(buf, msg, bufsz);
1253 free_msg(msg);
1254
1255 return bufsz;
1256}
1257
1258long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1259 long msgtyp, int msgflg)
1260{
1261 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1262}
1263
1264SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1265 long, msgtyp, int, msgflg)
1266{
1267 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1268}
1269
1270#ifdef CONFIG_COMPAT
1271static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1272{
1273 struct compat_msgbuf __user *msgp = dest;
1274 size_t msgsz;
1275
1276 if (put_user(msg->m_type, &msgp->mtype))
1277 return -EFAULT;
1278
1279 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1280 if (store_msg(msgp->mtext, msg, msgsz))
1281 return -EFAULT;
1282 return msgsz;
1283}
1284
1285long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1286 compat_long_t msgtyp, int msgflg)
1287{
1288 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1289 msgflg, compat_do_msg_fill);
1290}
1291
1292COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1293 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1294 int, msgflg)
1295{
1296 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1297}
1298#endif
1299
1300void msg_init_ns(struct ipc_namespace *ns)
1301{
1302 ns->msg_ctlmax = MSGMAX;
1303 ns->msg_ctlmnb = MSGMNB;
1304 ns->msg_ctlmni = MSGMNI;
1305
1306 atomic_set(&ns->msg_bytes, 0);
1307 atomic_set(&ns->msg_hdrs, 0);
1308 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1309}
1310
1311#ifdef CONFIG_IPC_NS
1312void msg_exit_ns(struct ipc_namespace *ns)
1313{
1314 free_ipcs(ns, &msg_ids(ns), freeque);
1315 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1316 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1317}
1318#endif
1319
1320#ifdef CONFIG_PROC_FS
1321static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1322{
1323 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1324 struct user_namespace *user_ns = seq_user_ns(s);
1325 struct kern_ipc_perm *ipcp = it;
1326 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1327
1328 seq_printf(s,
1329 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1330 msq->q_perm.key,
1331 msq->q_perm.id,
1332 msq->q_perm.mode,
1333 msq->q_cbytes,
1334 msq->q_qnum,
1335 pid_nr_ns(msq->q_lspid, pid_ns),
1336 pid_nr_ns(msq->q_lrpid, pid_ns),
1337 from_kuid_munged(user_ns, msq->q_perm.uid),
1338 from_kgid_munged(user_ns, msq->q_perm.gid),
1339 from_kuid_munged(user_ns, msq->q_perm.cuid),
1340 from_kgid_munged(user_ns, msq->q_perm.cgid),
1341 msq->q_stime,
1342 msq->q_rtime,
1343 msq->q_ctime);
1344
1345 return 0;
1346}
1347#endif
1348
1349void __init msg_init(void)
1350{
1351 msg_init_ns(&init_ipc_ns);
1352
1353 ipc_init_proc_interface("sysvipc/msg",
1354 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1355 IPC_MSG_IDS, sysvipc_msg_proc_show);
1356}