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