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