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 <asm/uaccess.h>
43#include "util.h"
44
45/*
46 * one msg_receiver structure for each sleeping receiver:
47 */
48struct msg_receiver {
49 struct list_head r_list;
50 struct task_struct *r_tsk;
51
52 int r_mode;
53 long r_msgtype;
54 long r_maxsize;
55
56 struct msg_msg *volatile r_msg;
57};
58
59/* one msg_sender for each sleeping sender */
60struct msg_sender {
61 struct list_head list;
62 struct task_struct *tsk;
63};
64
65#define SEARCH_ANY 1
66#define SEARCH_EQUAL 2
67#define SEARCH_NOTEQUAL 3
68#define SEARCH_LESSEQUAL 4
69#define SEARCH_NUMBER 5
70
71#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
72
73static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
74static int newque(struct ipc_namespace *, struct ipc_params *);
75#ifdef CONFIG_PROC_FS
76static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
77#endif
78
79/*
80 * Scale msgmni with the available lowmem size: the memory dedicated to msg
81 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
82 * Also take into account the number of nsproxies created so far.
83 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
84 */
85void recompute_msgmni(struct ipc_namespace *ns)
86{
87 struct sysinfo i;
88 unsigned long allowed;
89 int nb_ns;
90
91 si_meminfo(&i);
92 allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
93 / MSGMNB;
94 nb_ns = atomic_read(&nr_ipc_ns);
95 allowed /= nb_ns;
96
97 if (allowed < MSGMNI) {
98 ns->msg_ctlmni = MSGMNI;
99 return;
100 }
101
102 if (allowed > IPCMNI / nb_ns) {
103 ns->msg_ctlmni = IPCMNI / nb_ns;
104 return;
105 }
106
107 ns->msg_ctlmni = allowed;
108}
109
110void msg_init_ns(struct ipc_namespace *ns)
111{
112 ns->msg_ctlmax = MSGMAX;
113 ns->msg_ctlmnb = MSGMNB;
114
115 recompute_msgmni(ns);
116
117 atomic_set(&ns->msg_bytes, 0);
118 atomic_set(&ns->msg_hdrs, 0);
119 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
120}
121
122#ifdef CONFIG_IPC_NS
123void msg_exit_ns(struct ipc_namespace *ns)
124{
125 free_ipcs(ns, &msg_ids(ns), freeque);
126 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
127}
128#endif
129
130void __init msg_init(void)
131{
132 msg_init_ns(&init_ipc_ns);
133
134 printk(KERN_INFO "msgmni has been set to %d\n",
135 init_ipc_ns.msg_ctlmni);
136
137 ipc_init_proc_interface("sysvipc/msg",
138 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
139 IPC_MSG_IDS, sysvipc_msg_proc_show);
140}
141
142static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
143{
144 struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
145
146 if (IS_ERR(ipcp))
147 return ERR_CAST(ipcp);
148
149 return container_of(ipcp, struct msg_queue, q_perm);
150}
151
152static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
153 int id)
154{
155 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
156
157 if (IS_ERR(ipcp))
158 return ERR_CAST(ipcp);
159
160 return container_of(ipcp, struct msg_queue, q_perm);
161}
162
163static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
164{
165 ipc_rmid(&msg_ids(ns), &s->q_perm);
166}
167
168static void msg_rcu_free(struct rcu_head *head)
169{
170 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
171 struct msg_queue *msq = ipc_rcu_to_struct(p);
172
173 security_msg_queue_free(msq);
174 ipc_rcu_free(head);
175}
176
177/**
178 * newque - Create a new msg queue
179 * @ns: namespace
180 * @params: ptr to the structure that contains the key and msgflg
181 *
182 * Called with msg_ids.rwsem held (writer)
183 */
184static int newque(struct ipc_namespace *ns, struct ipc_params *params)
185{
186 struct msg_queue *msq;
187 int id, retval;
188 key_t key = params->key;
189 int msgflg = params->flg;
190
191 msq = ipc_rcu_alloc(sizeof(*msq));
192 if (!msq)
193 return -ENOMEM;
194
195 msq->q_perm.mode = msgflg & S_IRWXUGO;
196 msq->q_perm.key = key;
197
198 msq->q_perm.security = NULL;
199 retval = security_msg_queue_alloc(msq);
200 if (retval) {
201 ipc_rcu_putref(msq, ipc_rcu_free);
202 return retval;
203 }
204
205 /* ipc_addid() locks msq upon success. */
206 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
207 if (id < 0) {
208 ipc_rcu_putref(msq, msg_rcu_free);
209 return id;
210 }
211
212 msq->q_stime = msq->q_rtime = 0;
213 msq->q_ctime = get_seconds();
214 msq->q_cbytes = msq->q_qnum = 0;
215 msq->q_qbytes = ns->msg_ctlmnb;
216 msq->q_lspid = msq->q_lrpid = 0;
217 INIT_LIST_HEAD(&msq->q_messages);
218 INIT_LIST_HEAD(&msq->q_receivers);
219 INIT_LIST_HEAD(&msq->q_senders);
220
221 ipc_unlock_object(&msq->q_perm);
222 rcu_read_unlock();
223
224 return msq->q_perm.id;
225}
226
227static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
228{
229 mss->tsk = current;
230 current->state = TASK_INTERRUPTIBLE;
231 list_add_tail(&mss->list, &msq->q_senders);
232}
233
234static inline void ss_del(struct msg_sender *mss)
235{
236 if (mss->list.next != NULL)
237 list_del(&mss->list);
238}
239
240static void ss_wakeup(struct list_head *h, int kill)
241{
242 struct msg_sender *mss, *t;
243
244 list_for_each_entry_safe(mss, t, h, list) {
245 if (kill)
246 mss->list.next = NULL;
247 wake_up_process(mss->tsk);
248 }
249}
250
251static void expunge_all(struct msg_queue *msq, int res)
252{
253 struct msg_receiver *msr, *t;
254
255 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
256 msr->r_msg = NULL; /* initialize expunge ordering */
257 wake_up_process(msr->r_tsk);
258 /*
259 * Ensure that the wakeup is visible before setting r_msg as
260 * the receiving end depends on it: either spinning on a nil,
261 * or dealing with -EAGAIN cases. See lockless receive part 1
262 * and 2 in do_msgrcv().
263 */
264 smp_mb();
265 msr->r_msg = ERR_PTR(res);
266 }
267}
268
269/*
270 * freeque() wakes up waiters on the sender and receiver waiting queue,
271 * removes the message queue from message queue ID IDR, and cleans up all the
272 * messages associated with this queue.
273 *
274 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
275 * before freeque() is called. msg_ids.rwsem remains locked on exit.
276 */
277static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
278{
279 struct msg_msg *msg, *t;
280 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
281
282 expunge_all(msq, -EIDRM);
283 ss_wakeup(&msq->q_senders, 1);
284 msg_rmid(ns, msq);
285 ipc_unlock_object(&msq->q_perm);
286 rcu_read_unlock();
287
288 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
289 atomic_dec(&ns->msg_hdrs);
290 free_msg(msg);
291 }
292 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
293 ipc_rcu_putref(msq, msg_rcu_free);
294}
295
296/*
297 * Called with msg_ids.rwsem and ipcp locked.
298 */
299static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
300{
301 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
302
303 return security_msg_queue_associate(msq, msgflg);
304}
305
306SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
307{
308 struct ipc_namespace *ns;
309 struct ipc_ops msg_ops;
310 struct ipc_params msg_params;
311
312 ns = current->nsproxy->ipc_ns;
313
314 msg_ops.getnew = newque;
315 msg_ops.associate = msg_security;
316 msg_ops.more_checks = NULL;
317
318 msg_params.key = key;
319 msg_params.flg = msgflg;
320
321 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
322}
323
324static inline unsigned long
325copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
326{
327 switch (version) {
328 case IPC_64:
329 return copy_to_user(buf, in, sizeof(*in));
330 case IPC_OLD:
331 {
332 struct msqid_ds out;
333
334 memset(&out, 0, sizeof(out));
335
336 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
337
338 out.msg_stime = in->msg_stime;
339 out.msg_rtime = in->msg_rtime;
340 out.msg_ctime = in->msg_ctime;
341
342 if (in->msg_cbytes > USHRT_MAX)
343 out.msg_cbytes = USHRT_MAX;
344 else
345 out.msg_cbytes = in->msg_cbytes;
346 out.msg_lcbytes = in->msg_cbytes;
347
348 if (in->msg_qnum > USHRT_MAX)
349 out.msg_qnum = USHRT_MAX;
350 else
351 out.msg_qnum = in->msg_qnum;
352
353 if (in->msg_qbytes > USHRT_MAX)
354 out.msg_qbytes = USHRT_MAX;
355 else
356 out.msg_qbytes = in->msg_qbytes;
357 out.msg_lqbytes = in->msg_qbytes;
358
359 out.msg_lspid = in->msg_lspid;
360 out.msg_lrpid = in->msg_lrpid;
361
362 return copy_to_user(buf, &out, sizeof(out));
363 }
364 default:
365 return -EINVAL;
366 }
367}
368
369static inline unsigned long
370copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
371{
372 switch (version) {
373 case IPC_64:
374 if (copy_from_user(out, buf, sizeof(*out)))
375 return -EFAULT;
376 return 0;
377 case IPC_OLD:
378 {
379 struct msqid_ds tbuf_old;
380
381 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
382 return -EFAULT;
383
384 out->msg_perm.uid = tbuf_old.msg_perm.uid;
385 out->msg_perm.gid = tbuf_old.msg_perm.gid;
386 out->msg_perm.mode = tbuf_old.msg_perm.mode;
387
388 if (tbuf_old.msg_qbytes == 0)
389 out->msg_qbytes = tbuf_old.msg_lqbytes;
390 else
391 out->msg_qbytes = tbuf_old.msg_qbytes;
392
393 return 0;
394 }
395 default:
396 return -EINVAL;
397 }
398}
399
400/*
401 * This function handles some msgctl commands which require the rwsem
402 * to be held in write mode.
403 * NOTE: no locks must be held, the rwsem is taken inside this function.
404 */
405static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
406 struct msqid_ds __user *buf, int version)
407{
408 struct kern_ipc_perm *ipcp;
409 struct msqid64_ds uninitialized_var(msqid64);
410 struct msg_queue *msq;
411 int err;
412
413 if (cmd == IPC_SET) {
414 if (copy_msqid_from_user(&msqid64, buf, version))
415 return -EFAULT;
416 }
417
418 down_write(&msg_ids(ns).rwsem);
419 rcu_read_lock();
420
421 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
422 &msqid64.msg_perm, msqid64.msg_qbytes);
423 if (IS_ERR(ipcp)) {
424 err = PTR_ERR(ipcp);
425 goto out_unlock1;
426 }
427
428 msq = container_of(ipcp, struct msg_queue, q_perm);
429
430 err = security_msg_queue_msgctl(msq, cmd);
431 if (err)
432 goto out_unlock1;
433
434 switch (cmd) {
435 case IPC_RMID:
436 ipc_lock_object(&msq->q_perm);
437 /* freeque unlocks the ipc object and rcu */
438 freeque(ns, ipcp);
439 goto out_up;
440 case IPC_SET:
441 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
442 !capable(CAP_SYS_RESOURCE)) {
443 err = -EPERM;
444 goto out_unlock1;
445 }
446
447 ipc_lock_object(&msq->q_perm);
448 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
449 if (err)
450 goto out_unlock0;
451
452 msq->q_qbytes = msqid64.msg_qbytes;
453
454 msq->q_ctime = get_seconds();
455 /* sleeping receivers might be excluded by
456 * stricter permissions.
457 */
458 expunge_all(msq, -EAGAIN);
459 /* sleeping senders might be able to send
460 * due to a larger queue size.
461 */
462 ss_wakeup(&msq->q_senders, 0);
463 break;
464 default:
465 err = -EINVAL;
466 goto out_unlock1;
467 }
468
469out_unlock0:
470 ipc_unlock_object(&msq->q_perm);
471out_unlock1:
472 rcu_read_unlock();
473out_up:
474 up_write(&msg_ids(ns).rwsem);
475 return err;
476}
477
478static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
479 int cmd, int version, void __user *buf)
480{
481 int err;
482 struct msg_queue *msq;
483
484 switch (cmd) {
485 case IPC_INFO:
486 case MSG_INFO:
487 {
488 struct msginfo msginfo;
489 int max_id;
490
491 if (!buf)
492 return -EFAULT;
493
494 /*
495 * We must not return kernel stack data.
496 * due to padding, it's not enough
497 * to set all member fields.
498 */
499 err = security_msg_queue_msgctl(NULL, cmd);
500 if (err)
501 return err;
502
503 memset(&msginfo, 0, sizeof(msginfo));
504 msginfo.msgmni = ns->msg_ctlmni;
505 msginfo.msgmax = ns->msg_ctlmax;
506 msginfo.msgmnb = ns->msg_ctlmnb;
507 msginfo.msgssz = MSGSSZ;
508 msginfo.msgseg = MSGSEG;
509 down_read(&msg_ids(ns).rwsem);
510 if (cmd == MSG_INFO) {
511 msginfo.msgpool = msg_ids(ns).in_use;
512 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
513 msginfo.msgtql = atomic_read(&ns->msg_bytes);
514 } else {
515 msginfo.msgmap = MSGMAP;
516 msginfo.msgpool = MSGPOOL;
517 msginfo.msgtql = MSGTQL;
518 }
519 max_id = ipc_get_maxid(&msg_ids(ns));
520 up_read(&msg_ids(ns).rwsem);
521 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
522 return -EFAULT;
523 return (max_id < 0) ? 0 : max_id;
524 }
525
526 case MSG_STAT:
527 case IPC_STAT:
528 {
529 struct msqid64_ds tbuf;
530 int success_return;
531
532 if (!buf)
533 return -EFAULT;
534
535 memset(&tbuf, 0, sizeof(tbuf));
536
537 rcu_read_lock();
538 if (cmd == MSG_STAT) {
539 msq = msq_obtain_object(ns, msqid);
540 if (IS_ERR(msq)) {
541 err = PTR_ERR(msq);
542 goto out_unlock;
543 }
544 success_return = msq->q_perm.id;
545 } else {
546 msq = msq_obtain_object_check(ns, msqid);
547 if (IS_ERR(msq)) {
548 err = PTR_ERR(msq);
549 goto out_unlock;
550 }
551 success_return = 0;
552 }
553
554 err = -EACCES;
555 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
556 goto out_unlock;
557
558 err = security_msg_queue_msgctl(msq, cmd);
559 if (err)
560 goto out_unlock;
561
562 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
563 tbuf.msg_stime = msq->q_stime;
564 tbuf.msg_rtime = msq->q_rtime;
565 tbuf.msg_ctime = msq->q_ctime;
566 tbuf.msg_cbytes = msq->q_cbytes;
567 tbuf.msg_qnum = msq->q_qnum;
568 tbuf.msg_qbytes = msq->q_qbytes;
569 tbuf.msg_lspid = msq->q_lspid;
570 tbuf.msg_lrpid = msq->q_lrpid;
571 rcu_read_unlock();
572
573 if (copy_msqid_to_user(buf, &tbuf, version))
574 return -EFAULT;
575 return success_return;
576 }
577
578 default:
579 return -EINVAL;
580 }
581
582 return err;
583out_unlock:
584 rcu_read_unlock();
585 return err;
586}
587
588SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
589{
590 int version;
591 struct ipc_namespace *ns;
592
593 if (msqid < 0 || cmd < 0)
594 return -EINVAL;
595
596 version = ipc_parse_version(&cmd);
597 ns = current->nsproxy->ipc_ns;
598
599 switch (cmd) {
600 case IPC_INFO:
601 case MSG_INFO:
602 case MSG_STAT: /* msqid is an index rather than a msg queue id */
603 case IPC_STAT:
604 return msgctl_nolock(ns, msqid, cmd, version, buf);
605 case IPC_SET:
606 case IPC_RMID:
607 return msgctl_down(ns, msqid, cmd, buf, version);
608 default:
609 return -EINVAL;
610 }
611}
612
613static int testmsg(struct msg_msg *msg, long type, int mode)
614{
615 switch (mode)
616 {
617 case SEARCH_ANY:
618 case SEARCH_NUMBER:
619 return 1;
620 case SEARCH_LESSEQUAL:
621 if (msg->m_type <= type)
622 return 1;
623 break;
624 case SEARCH_EQUAL:
625 if (msg->m_type == type)
626 return 1;
627 break;
628 case SEARCH_NOTEQUAL:
629 if (msg->m_type != type)
630 return 1;
631 break;
632 }
633 return 0;
634}
635
636static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
637{
638 struct msg_receiver *msr, *t;
639
640 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
641 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
642 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
643 msr->r_msgtype, msr->r_mode)) {
644
645 list_del(&msr->r_list);
646 if (msr->r_maxsize < msg->m_ts) {
647 /* initialize pipelined send ordering */
648 msr->r_msg = NULL;
649 wake_up_process(msr->r_tsk);
650 smp_mb(); /* see barrier comment below */
651 msr->r_msg = ERR_PTR(-E2BIG);
652 } else {
653 msr->r_msg = NULL;
654 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
655 msq->q_rtime = get_seconds();
656 wake_up_process(msr->r_tsk);
657 /*
658 * Ensure that the wakeup is visible before
659 * setting r_msg, as the receiving end depends
660 * on it. See lockless receive part 1 and 2 in
661 * do_msgrcv().
662 */
663 smp_mb();
664 msr->r_msg = msg;
665
666 return 1;
667 }
668 }
669 }
670
671 return 0;
672}
673
674long do_msgsnd(int msqid, long mtype, void __user *mtext,
675 size_t msgsz, int msgflg)
676{
677 struct msg_queue *msq;
678 struct msg_msg *msg;
679 int err;
680 struct ipc_namespace *ns;
681
682 ns = current->nsproxy->ipc_ns;
683
684 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
685 return -EINVAL;
686 if (mtype < 1)
687 return -EINVAL;
688
689 msg = load_msg(mtext, msgsz);
690 if (IS_ERR(msg))
691 return PTR_ERR(msg);
692
693 msg->m_type = mtype;
694 msg->m_ts = msgsz;
695
696 rcu_read_lock();
697 msq = msq_obtain_object_check(ns, msqid);
698 if (IS_ERR(msq)) {
699 err = PTR_ERR(msq);
700 goto out_unlock1;
701 }
702
703 ipc_lock_object(&msq->q_perm);
704
705 for (;;) {
706 struct msg_sender s;
707
708 err = -EACCES;
709 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
710 goto out_unlock0;
711
712 /* raced with RMID? */
713 if (!ipc_valid_object(&msq->q_perm)) {
714 err = -EIDRM;
715 goto out_unlock0;
716 }
717
718 err = security_msg_queue_msgsnd(msq, msg, msgflg);
719 if (err)
720 goto out_unlock0;
721
722 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
723 1 + msq->q_qnum <= msq->q_qbytes) {
724 break;
725 }
726
727 /* queue full, wait: */
728 if (msgflg & IPC_NOWAIT) {
729 err = -EAGAIN;
730 goto out_unlock0;
731 }
732
733 /* enqueue the sender and prepare to block */
734 ss_add(msq, &s);
735
736 if (!ipc_rcu_getref(msq)) {
737 err = -EIDRM;
738 goto out_unlock0;
739 }
740
741 ipc_unlock_object(&msq->q_perm);
742 rcu_read_unlock();
743 schedule();
744
745 rcu_read_lock();
746 ipc_lock_object(&msq->q_perm);
747
748 ipc_rcu_putref(msq, ipc_rcu_free);
749 /* raced with RMID? */
750 if (!ipc_valid_object(&msq->q_perm)) {
751 err = -EIDRM;
752 goto out_unlock0;
753 }
754
755 ss_del(&s);
756
757 if (signal_pending(current)) {
758 err = -ERESTARTNOHAND;
759 goto out_unlock0;
760 }
761
762 }
763 msq->q_lspid = task_tgid_vnr(current);
764 msq->q_stime = get_seconds();
765
766 if (!pipelined_send(msq, msg)) {
767 /* no one is waiting for this message, enqueue it */
768 list_add_tail(&msg->m_list, &msq->q_messages);
769 msq->q_cbytes += msgsz;
770 msq->q_qnum++;
771 atomic_add(msgsz, &ns->msg_bytes);
772 atomic_inc(&ns->msg_hdrs);
773 }
774
775 err = 0;
776 msg = NULL;
777
778out_unlock0:
779 ipc_unlock_object(&msq->q_perm);
780out_unlock1:
781 rcu_read_unlock();
782 if (msg != NULL)
783 free_msg(msg);
784 return err;
785}
786
787SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
788 int, msgflg)
789{
790 long mtype;
791
792 if (get_user(mtype, &msgp->mtype))
793 return -EFAULT;
794 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
795}
796
797static inline int convert_mode(long *msgtyp, int msgflg)
798{
799 if (msgflg & MSG_COPY)
800 return SEARCH_NUMBER;
801 /*
802 * find message of correct type.
803 * msgtyp = 0 => get first.
804 * msgtyp > 0 => get first message of matching type.
805 * msgtyp < 0 => get message with least type must be < abs(msgtype).
806 */
807 if (*msgtyp == 0)
808 return SEARCH_ANY;
809 if (*msgtyp < 0) {
810 *msgtyp = -*msgtyp;
811 return SEARCH_LESSEQUAL;
812 }
813 if (msgflg & MSG_EXCEPT)
814 return SEARCH_NOTEQUAL;
815 return SEARCH_EQUAL;
816}
817
818static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
819{
820 struct msgbuf __user *msgp = dest;
821 size_t msgsz;
822
823 if (put_user(msg->m_type, &msgp->mtype))
824 return -EFAULT;
825
826 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
827 if (store_msg(msgp->mtext, msg, msgsz))
828 return -EFAULT;
829 return msgsz;
830}
831
832#ifdef CONFIG_CHECKPOINT_RESTORE
833/*
834 * This function creates new kernel message structure, large enough to store
835 * bufsz message bytes.
836 */
837static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
838{
839 struct msg_msg *copy;
840
841 /*
842 * Create dummy message to copy real message to.
843 */
844 copy = load_msg(buf, bufsz);
845 if (!IS_ERR(copy))
846 copy->m_ts = bufsz;
847 return copy;
848}
849
850static inline void free_copy(struct msg_msg *copy)
851{
852 if (copy)
853 free_msg(copy);
854}
855#else
856static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
857{
858 return ERR_PTR(-ENOSYS);
859}
860
861static inline void free_copy(struct msg_msg *copy)
862{
863}
864#endif
865
866static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
867{
868 struct msg_msg *msg, *found = NULL;
869 long count = 0;
870
871 list_for_each_entry(msg, &msq->q_messages, m_list) {
872 if (testmsg(msg, *msgtyp, mode) &&
873 !security_msg_queue_msgrcv(msq, msg, current,
874 *msgtyp, mode)) {
875 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
876 *msgtyp = msg->m_type - 1;
877 found = msg;
878 } else if (mode == SEARCH_NUMBER) {
879 if (*msgtyp == count)
880 return msg;
881 } else
882 return msg;
883 count++;
884 }
885 }
886
887 return found ?: ERR_PTR(-EAGAIN);
888}
889
890long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
891 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
892{
893 int mode;
894 struct msg_queue *msq;
895 struct ipc_namespace *ns;
896 struct msg_msg *msg, *copy = NULL;
897
898 ns = current->nsproxy->ipc_ns;
899
900 if (msqid < 0 || (long) bufsz < 0)
901 return -EINVAL;
902
903 if (msgflg & MSG_COPY) {
904 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
905 return -EINVAL;
906 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
907 if (IS_ERR(copy))
908 return PTR_ERR(copy);
909 }
910 mode = convert_mode(&msgtyp, msgflg);
911
912 rcu_read_lock();
913 msq = msq_obtain_object_check(ns, msqid);
914 if (IS_ERR(msq)) {
915 rcu_read_unlock();
916 free_copy(copy);
917 return PTR_ERR(msq);
918 }
919
920 for (;;) {
921 struct msg_receiver msr_d;
922
923 msg = ERR_PTR(-EACCES);
924 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
925 goto out_unlock1;
926
927 ipc_lock_object(&msq->q_perm);
928
929 /* raced with RMID? */
930 if (!ipc_valid_object(&msq->q_perm)) {
931 msg = ERR_PTR(-EIDRM);
932 goto out_unlock0;
933 }
934
935 msg = find_msg(msq, &msgtyp, mode);
936 if (!IS_ERR(msg)) {
937 /*
938 * Found a suitable message.
939 * Unlink it from the queue.
940 */
941 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
942 msg = ERR_PTR(-E2BIG);
943 goto out_unlock0;
944 }
945 /*
946 * If we are copying, then do not unlink message and do
947 * not update queue parameters.
948 */
949 if (msgflg & MSG_COPY) {
950 msg = copy_msg(msg, copy);
951 goto out_unlock0;
952 }
953
954 list_del(&msg->m_list);
955 msq->q_qnum--;
956 msq->q_rtime = get_seconds();
957 msq->q_lrpid = task_tgid_vnr(current);
958 msq->q_cbytes -= msg->m_ts;
959 atomic_sub(msg->m_ts, &ns->msg_bytes);
960 atomic_dec(&ns->msg_hdrs);
961 ss_wakeup(&msq->q_senders, 0);
962
963 goto out_unlock0;
964 }
965
966 /* No message waiting. Wait for a message */
967 if (msgflg & IPC_NOWAIT) {
968 msg = ERR_PTR(-ENOMSG);
969 goto out_unlock0;
970 }
971
972 list_add_tail(&msr_d.r_list, &msq->q_receivers);
973 msr_d.r_tsk = current;
974 msr_d.r_msgtype = msgtyp;
975 msr_d.r_mode = mode;
976 if (msgflg & MSG_NOERROR)
977 msr_d.r_maxsize = INT_MAX;
978 else
979 msr_d.r_maxsize = bufsz;
980 msr_d.r_msg = ERR_PTR(-EAGAIN);
981 current->state = TASK_INTERRUPTIBLE;
982
983 ipc_unlock_object(&msq->q_perm);
984 rcu_read_unlock();
985 schedule();
986
987 /* Lockless receive, part 1:
988 * Disable preemption. We don't hold a reference to the queue
989 * and getting a reference would defeat the idea of a lockless
990 * operation, thus the code relies on rcu to guarantee the
991 * existence of msq:
992 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
993 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
994 * rcu_read_lock() prevents preemption between reading r_msg
995 * and acquiring the q_perm.lock in ipc_lock_object().
996 */
997 rcu_read_lock();
998
999 /* Lockless receive, part 2:
1000 * Wait until pipelined_send or expunge_all are outside of
1001 * wake_up_process(). There is a race with exit(), see
1002 * ipc/mqueue.c for the details.
1003 */
1004 msg = (struct msg_msg *)msr_d.r_msg;
1005 while (msg == NULL) {
1006 cpu_relax();
1007 msg = (struct msg_msg *)msr_d.r_msg;
1008 }
1009
1010 /* Lockless receive, part 3:
1011 * If there is a message or an error then accept it without
1012 * locking.
1013 */
1014 if (msg != ERR_PTR(-EAGAIN))
1015 goto out_unlock1;
1016
1017 /* Lockless receive, part 3:
1018 * Acquire the queue spinlock.
1019 */
1020 ipc_lock_object(&msq->q_perm);
1021
1022 /* Lockless receive, part 4:
1023 * Repeat test after acquiring the spinlock.
1024 */
1025 msg = (struct msg_msg *)msr_d.r_msg;
1026 if (msg != ERR_PTR(-EAGAIN))
1027 goto out_unlock0;
1028
1029 list_del(&msr_d.r_list);
1030 if (signal_pending(current)) {
1031 msg = ERR_PTR(-ERESTARTNOHAND);
1032 goto out_unlock0;
1033 }
1034
1035 ipc_unlock_object(&msq->q_perm);
1036 }
1037
1038out_unlock0:
1039 ipc_unlock_object(&msq->q_perm);
1040out_unlock1:
1041 rcu_read_unlock();
1042 if (IS_ERR(msg)) {
1043 free_copy(copy);
1044 return PTR_ERR(msg);
1045 }
1046
1047 bufsz = msg_handler(buf, msg, bufsz);
1048 free_msg(msg);
1049
1050 return bufsz;
1051}
1052
1053SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1054 long, msgtyp, int, msgflg)
1055{
1056 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1057}
1058
1059#ifdef CONFIG_PROC_FS
1060static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1061{
1062 struct user_namespace *user_ns = seq_user_ns(s);
1063 struct msg_queue *msq = it;
1064
1065 return seq_printf(s,
1066 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1067 msq->q_perm.key,
1068 msq->q_perm.id,
1069 msq->q_perm.mode,
1070 msq->q_cbytes,
1071 msq->q_qnum,
1072 msq->q_lspid,
1073 msq->q_lrpid,
1074 from_kuid_munged(user_ns, msq->q_perm.uid),
1075 from_kgid_munged(user_ns, msq->q_perm.gid),
1076 from_kuid_munged(user_ns, msq->q_perm.cuid),
1077 from_kgid_munged(user_ns, msq->q_perm.cgid),
1078 msq->q_stime,
1079 msq->q_rtime,
1080 msq->q_ctime);
1081}
1082#endif
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}