Loading...
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#include <linux/percpu_counter.h>
43
44#include <asm/current.h>
45#include <linux/uaccess.h>
46#include "util.h"
47
48/* one msq_queue structure for each present queue on the system */
49struct msg_queue {
50 struct kern_ipc_perm q_perm;
51 time64_t q_stime; /* last msgsnd time */
52 time64_t q_rtime; /* last msgrcv time */
53 time64_t q_ctime; /* last change time */
54 unsigned long q_cbytes; /* current number of bytes on queue */
55 unsigned long q_qnum; /* number of messages in queue */
56 unsigned long q_qbytes; /* max number of bytes on queue */
57 struct pid *q_lspid; /* pid of last msgsnd */
58 struct pid *q_lrpid; /* last receive pid */
59
60 struct list_head q_messages;
61 struct list_head q_receivers;
62 struct list_head q_senders;
63} __randomize_layout;
64
65/*
66 * MSG_BARRIER Locking:
67 *
68 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
69 * does not acquire any locks when it sees that a message exists in
70 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
71 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
72 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
73 */
74
75/* one msg_receiver structure for each sleeping receiver */
76struct msg_receiver {
77 struct list_head r_list;
78 struct task_struct *r_tsk;
79
80 int r_mode;
81 long r_msgtype;
82 long r_maxsize;
83
84 struct msg_msg *r_msg;
85};
86
87/* one msg_sender for each sleeping sender */
88struct msg_sender {
89 struct list_head list;
90 struct task_struct *tsk;
91 size_t msgsz;
92};
93
94#define SEARCH_ANY 1
95#define SEARCH_EQUAL 2
96#define SEARCH_NOTEQUAL 3
97#define SEARCH_LESSEQUAL 4
98#define SEARCH_NUMBER 5
99
100#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
101
102static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
103{
104 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
105
106 if (IS_ERR(ipcp))
107 return ERR_CAST(ipcp);
108
109 return container_of(ipcp, struct msg_queue, q_perm);
110}
111
112static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
113 int id)
114{
115 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
116
117 if (IS_ERR(ipcp))
118 return ERR_CAST(ipcp);
119
120 return container_of(ipcp, struct msg_queue, q_perm);
121}
122
123static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
124{
125 ipc_rmid(&msg_ids(ns), &s->q_perm);
126}
127
128static void msg_rcu_free(struct rcu_head *head)
129{
130 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
131 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
132
133 security_msg_queue_free(&msq->q_perm);
134 kfree(msq);
135}
136
137/**
138 * newque - Create a new msg queue
139 * @ns: namespace
140 * @params: ptr to the structure that contains the key and msgflg
141 *
142 * Called with msg_ids.rwsem held (writer)
143 */
144static int newque(struct ipc_namespace *ns, struct ipc_params *params)
145{
146 struct msg_queue *msq;
147 int retval;
148 key_t key = params->key;
149 int msgflg = params->flg;
150
151 msq = kmalloc(sizeof(*msq), GFP_KERNEL_ACCOUNT);
152 if (unlikely(!msq))
153 return -ENOMEM;
154
155 msq->q_perm.mode = msgflg & S_IRWXUGO;
156 msq->q_perm.key = key;
157
158 msq->q_perm.security = NULL;
159 retval = security_msg_queue_alloc(&msq->q_perm);
160 if (retval) {
161 kfree(msq);
162 return retval;
163 }
164
165 msq->q_stime = msq->q_rtime = 0;
166 msq->q_ctime = ktime_get_real_seconds();
167 msq->q_cbytes = msq->q_qnum = 0;
168 msq->q_qbytes = ns->msg_ctlmnb;
169 msq->q_lspid = msq->q_lrpid = NULL;
170 INIT_LIST_HEAD(&msq->q_messages);
171 INIT_LIST_HEAD(&msq->q_receivers);
172 INIT_LIST_HEAD(&msq->q_senders);
173
174 /* ipc_addid() locks msq upon success. */
175 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
176 if (retval < 0) {
177 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
178 return retval;
179 }
180
181 ipc_unlock_object(&msq->q_perm);
182 rcu_read_unlock();
183
184 return msq->q_perm.id;
185}
186
187static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
188{
189 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
190 1 + msq->q_qnum <= msq->q_qbytes;
191}
192
193static inline void ss_add(struct msg_queue *msq,
194 struct msg_sender *mss, size_t msgsz)
195{
196 mss->tsk = current;
197 mss->msgsz = msgsz;
198 /*
199 * No memory barrier required: we did ipc_lock_object(),
200 * and the waker obtains that lock before calling wake_q_add().
201 */
202 __set_current_state(TASK_INTERRUPTIBLE);
203 list_add_tail(&mss->list, &msq->q_senders);
204}
205
206static inline void ss_del(struct msg_sender *mss)
207{
208 if (mss->list.next)
209 list_del(&mss->list);
210}
211
212static void ss_wakeup(struct msg_queue *msq,
213 struct wake_q_head *wake_q, bool kill)
214{
215 struct msg_sender *mss, *t;
216 struct task_struct *stop_tsk = NULL;
217 struct list_head *h = &msq->q_senders;
218
219 list_for_each_entry_safe(mss, t, h, list) {
220 if (kill)
221 mss->list.next = NULL;
222
223 /*
224 * Stop at the first task we don't wakeup,
225 * we've already iterated the original
226 * sender queue.
227 */
228 else if (stop_tsk == mss->tsk)
229 break;
230 /*
231 * We are not in an EIDRM scenario here, therefore
232 * verify that we really need to wakeup the task.
233 * To maintain current semantics and wakeup order,
234 * move the sender to the tail on behalf of the
235 * blocked task.
236 */
237 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
238 if (!stop_tsk)
239 stop_tsk = mss->tsk;
240
241 list_move_tail(&mss->list, &msq->q_senders);
242 continue;
243 }
244
245 wake_q_add(wake_q, mss->tsk);
246 }
247}
248
249static void expunge_all(struct msg_queue *msq, int res,
250 struct wake_q_head *wake_q)
251{
252 struct msg_receiver *msr, *t;
253
254 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
255 struct task_struct *r_tsk;
256
257 r_tsk = get_task_struct(msr->r_tsk);
258
259 /* see MSG_BARRIER for purpose/pairing */
260 smp_store_release(&msr->r_msg, ERR_PTR(res));
261 wake_q_add_safe(wake_q, r_tsk);
262 }
263}
264
265/*
266 * freeque() wakes up waiters on the sender and receiver waiting queue,
267 * removes the message queue from message queue ID IDR, and cleans up all the
268 * messages associated with this queue.
269 *
270 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
271 * before freeque() is called. msg_ids.rwsem remains locked on exit.
272 */
273static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
274 __releases(RCU)
275 __releases(&msq->q_perm)
276{
277 struct msg_msg *msg, *t;
278 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
279 DEFINE_WAKE_Q(wake_q);
280
281 expunge_all(msq, -EIDRM, &wake_q);
282 ss_wakeup(msq, &wake_q, true);
283 msg_rmid(ns, msq);
284 ipc_unlock_object(&msq->q_perm);
285 wake_up_q(&wake_q);
286 rcu_read_unlock();
287
288 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
289 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
290 free_msg(msg);
291 }
292 percpu_counter_sub_local(&ns->percpu_msg_bytes, msq->q_cbytes);
293 ipc_update_pid(&msq->q_lspid, NULL);
294 ipc_update_pid(&msq->q_lrpid, NULL);
295 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
296}
297
298long ksys_msgget(key_t key, int msgflg)
299{
300 struct ipc_namespace *ns;
301 static const struct ipc_ops msg_ops = {
302 .getnew = newque,
303 .associate = security_msg_queue_associate,
304 };
305 struct ipc_params msg_params;
306
307 ns = current->nsproxy->ipc_ns;
308
309 msg_params.key = key;
310 msg_params.flg = msgflg;
311
312 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
313}
314
315SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
316{
317 return ksys_msgget(key, msgflg);
318}
319
320static inline unsigned long
321copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
322{
323 switch (version) {
324 case IPC_64:
325 return copy_to_user(buf, in, sizeof(*in));
326 case IPC_OLD:
327 {
328 struct msqid_ds out;
329
330 memset(&out, 0, sizeof(out));
331
332 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
333
334 out.msg_stime = in->msg_stime;
335 out.msg_rtime = in->msg_rtime;
336 out.msg_ctime = in->msg_ctime;
337
338 if (in->msg_cbytes > USHRT_MAX)
339 out.msg_cbytes = USHRT_MAX;
340 else
341 out.msg_cbytes = in->msg_cbytes;
342 out.msg_lcbytes = in->msg_cbytes;
343
344 if (in->msg_qnum > USHRT_MAX)
345 out.msg_qnum = USHRT_MAX;
346 else
347 out.msg_qnum = in->msg_qnum;
348
349 if (in->msg_qbytes > USHRT_MAX)
350 out.msg_qbytes = USHRT_MAX;
351 else
352 out.msg_qbytes = in->msg_qbytes;
353 out.msg_lqbytes = in->msg_qbytes;
354
355 out.msg_lspid = in->msg_lspid;
356 out.msg_lrpid = in->msg_lrpid;
357
358 return copy_to_user(buf, &out, sizeof(out));
359 }
360 default:
361 return -EINVAL;
362 }
363}
364
365static inline unsigned long
366copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
367{
368 switch (version) {
369 case IPC_64:
370 if (copy_from_user(out, buf, sizeof(*out)))
371 return -EFAULT;
372 return 0;
373 case IPC_OLD:
374 {
375 struct msqid_ds tbuf_old;
376
377 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
378 return -EFAULT;
379
380 out->msg_perm.uid = tbuf_old.msg_perm.uid;
381 out->msg_perm.gid = tbuf_old.msg_perm.gid;
382 out->msg_perm.mode = tbuf_old.msg_perm.mode;
383
384 if (tbuf_old.msg_qbytes == 0)
385 out->msg_qbytes = tbuf_old.msg_lqbytes;
386 else
387 out->msg_qbytes = tbuf_old.msg_qbytes;
388
389 return 0;
390 }
391 default:
392 return -EINVAL;
393 }
394}
395
396/*
397 * This function handles some msgctl commands which require the rwsem
398 * to be held in write mode.
399 * NOTE: no locks must be held, the rwsem is taken inside this function.
400 */
401static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
402 struct ipc64_perm *perm, int msg_qbytes)
403{
404 struct kern_ipc_perm *ipcp;
405 struct msg_queue *msq;
406 int err;
407
408 down_write(&msg_ids(ns).rwsem);
409 rcu_read_lock();
410
411 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
412 perm, msg_qbytes);
413 if (IS_ERR(ipcp)) {
414 err = PTR_ERR(ipcp);
415 goto out_unlock1;
416 }
417
418 msq = container_of(ipcp, struct msg_queue, q_perm);
419
420 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
421 if (err)
422 goto out_unlock1;
423
424 switch (cmd) {
425 case IPC_RMID:
426 ipc_lock_object(&msq->q_perm);
427 /* freeque unlocks the ipc object and rcu */
428 freeque(ns, ipcp);
429 goto out_up;
430 case IPC_SET:
431 {
432 DEFINE_WAKE_Q(wake_q);
433
434 if (msg_qbytes > ns->msg_ctlmnb &&
435 !capable(CAP_SYS_RESOURCE)) {
436 err = -EPERM;
437 goto out_unlock1;
438 }
439
440 ipc_lock_object(&msq->q_perm);
441 err = ipc_update_perm(perm, ipcp);
442 if (err)
443 goto out_unlock0;
444
445 msq->q_qbytes = msg_qbytes;
446
447 msq->q_ctime = ktime_get_real_seconds();
448 /*
449 * Sleeping receivers might be excluded by
450 * stricter permissions.
451 */
452 expunge_all(msq, -EAGAIN, &wake_q);
453 /*
454 * Sleeping senders might be able to send
455 * due to a larger queue size.
456 */
457 ss_wakeup(msq, &wake_q, false);
458 ipc_unlock_object(&msq->q_perm);
459 wake_up_q(&wake_q);
460
461 goto out_unlock1;
462 }
463 default:
464 err = -EINVAL;
465 goto out_unlock1;
466 }
467
468out_unlock0:
469 ipc_unlock_object(&msq->q_perm);
470out_unlock1:
471 rcu_read_unlock();
472out_up:
473 up_write(&msg_ids(ns).rwsem);
474 return err;
475}
476
477static int msgctl_info(struct ipc_namespace *ns, int msqid,
478 int cmd, struct msginfo *msginfo)
479{
480 int err;
481 int max_idx;
482
483 /*
484 * We must not return kernel stack data.
485 * due to padding, it's not enough
486 * to set all member fields.
487 */
488 err = security_msg_queue_msgctl(NULL, cmd);
489 if (err)
490 return err;
491
492 memset(msginfo, 0, sizeof(*msginfo));
493 msginfo->msgmni = ns->msg_ctlmni;
494 msginfo->msgmax = ns->msg_ctlmax;
495 msginfo->msgmnb = ns->msg_ctlmnb;
496 msginfo->msgssz = MSGSSZ;
497 msginfo->msgseg = MSGSEG;
498 down_read(&msg_ids(ns).rwsem);
499 if (cmd == MSG_INFO)
500 msginfo->msgpool = msg_ids(ns).in_use;
501 max_idx = ipc_get_maxidx(&msg_ids(ns));
502 up_read(&msg_ids(ns).rwsem);
503 if (cmd == MSG_INFO) {
504 msginfo->msgmap = min_t(int,
505 percpu_counter_sum(&ns->percpu_msg_hdrs),
506 INT_MAX);
507 msginfo->msgtql = min_t(int,
508 percpu_counter_sum(&ns->percpu_msg_bytes),
509 INT_MAX);
510 } else {
511 msginfo->msgmap = MSGMAP;
512 msginfo->msgpool = MSGPOOL;
513 msginfo->msgtql = MSGTQL;
514 }
515 return (max_idx < 0) ? 0 : max_idx;
516}
517
518static int msgctl_stat(struct ipc_namespace *ns, int msqid,
519 int cmd, struct msqid64_ds *p)
520{
521 struct msg_queue *msq;
522 int err;
523
524 memset(p, 0, sizeof(*p));
525
526 rcu_read_lock();
527 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
528 msq = msq_obtain_object(ns, msqid);
529 if (IS_ERR(msq)) {
530 err = PTR_ERR(msq);
531 goto out_unlock;
532 }
533 } else { /* IPC_STAT */
534 msq = msq_obtain_object_check(ns, msqid);
535 if (IS_ERR(msq)) {
536 err = PTR_ERR(msq);
537 goto out_unlock;
538 }
539 }
540
541 /* see comment for SHM_STAT_ANY */
542 if (cmd == MSG_STAT_ANY)
543 audit_ipc_obj(&msq->q_perm);
544 else {
545 err = -EACCES;
546 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
547 goto out_unlock;
548 }
549
550 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
551 if (err)
552 goto out_unlock;
553
554 ipc_lock_object(&msq->q_perm);
555
556 if (!ipc_valid_object(&msq->q_perm)) {
557 ipc_unlock_object(&msq->q_perm);
558 err = -EIDRM;
559 goto out_unlock;
560 }
561
562 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
563 p->msg_stime = msq->q_stime;
564 p->msg_rtime = msq->q_rtime;
565 p->msg_ctime = msq->q_ctime;
566#ifndef CONFIG_64BIT
567 p->msg_stime_high = msq->q_stime >> 32;
568 p->msg_rtime_high = msq->q_rtime >> 32;
569 p->msg_ctime_high = msq->q_ctime >> 32;
570#endif
571 p->msg_cbytes = msq->q_cbytes;
572 p->msg_qnum = msq->q_qnum;
573 p->msg_qbytes = msq->q_qbytes;
574 p->msg_lspid = pid_vnr(msq->q_lspid);
575 p->msg_lrpid = pid_vnr(msq->q_lrpid);
576
577 if (cmd == IPC_STAT) {
578 /*
579 * As defined in SUS:
580 * Return 0 on success
581 */
582 err = 0;
583 } else {
584 /*
585 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
586 * Return the full id, including the sequence number
587 */
588 err = msq->q_perm.id;
589 }
590
591 ipc_unlock_object(&msq->q_perm);
592out_unlock:
593 rcu_read_unlock();
594 return err;
595}
596
597static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
598{
599 struct ipc_namespace *ns;
600 struct msqid64_ds msqid64;
601 int err;
602
603 if (msqid < 0 || cmd < 0)
604 return -EINVAL;
605
606 ns = current->nsproxy->ipc_ns;
607
608 switch (cmd) {
609 case IPC_INFO:
610 case MSG_INFO: {
611 struct msginfo msginfo;
612 err = msgctl_info(ns, msqid, cmd, &msginfo);
613 if (err < 0)
614 return err;
615 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
616 err = -EFAULT;
617 return err;
618 }
619 case MSG_STAT: /* msqid is an index rather than a msg queue id */
620 case MSG_STAT_ANY:
621 case IPC_STAT:
622 err = msgctl_stat(ns, msqid, cmd, &msqid64);
623 if (err < 0)
624 return err;
625 if (copy_msqid_to_user(buf, &msqid64, version))
626 err = -EFAULT;
627 return err;
628 case IPC_SET:
629 if (copy_msqid_from_user(&msqid64, buf, version))
630 return -EFAULT;
631 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
632 msqid64.msg_qbytes);
633 case IPC_RMID:
634 return msgctl_down(ns, msqid, cmd, NULL, 0);
635 default:
636 return -EINVAL;
637 }
638}
639
640SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
641{
642 return ksys_msgctl(msqid, cmd, buf, IPC_64);
643}
644
645#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
646long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
647{
648 int version = ipc_parse_version(&cmd);
649
650 return ksys_msgctl(msqid, cmd, buf, version);
651}
652
653SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
654{
655 return ksys_old_msgctl(msqid, cmd, buf);
656}
657#endif
658
659#ifdef CONFIG_COMPAT
660
661struct compat_msqid_ds {
662 struct compat_ipc_perm msg_perm;
663 compat_uptr_t msg_first;
664 compat_uptr_t msg_last;
665 old_time32_t msg_stime;
666 old_time32_t msg_rtime;
667 old_time32_t msg_ctime;
668 compat_ulong_t msg_lcbytes;
669 compat_ulong_t msg_lqbytes;
670 unsigned short msg_cbytes;
671 unsigned short msg_qnum;
672 unsigned short msg_qbytes;
673 compat_ipc_pid_t msg_lspid;
674 compat_ipc_pid_t msg_lrpid;
675};
676
677static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
678 int version)
679{
680 memset(out, 0, sizeof(*out));
681 if (version == IPC_64) {
682 struct compat_msqid64_ds __user *p = buf;
683 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
684 return -EFAULT;
685 if (get_user(out->msg_qbytes, &p->msg_qbytes))
686 return -EFAULT;
687 } else {
688 struct compat_msqid_ds __user *p = buf;
689 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
690 return -EFAULT;
691 if (get_user(out->msg_qbytes, &p->msg_qbytes))
692 return -EFAULT;
693 }
694 return 0;
695}
696
697static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
698 int version)
699{
700 if (version == IPC_64) {
701 struct compat_msqid64_ds v;
702 memset(&v, 0, sizeof(v));
703 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
704 v.msg_stime = lower_32_bits(in->msg_stime);
705 v.msg_stime_high = upper_32_bits(in->msg_stime);
706 v.msg_rtime = lower_32_bits(in->msg_rtime);
707 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
708 v.msg_ctime = lower_32_bits(in->msg_ctime);
709 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
710 v.msg_cbytes = in->msg_cbytes;
711 v.msg_qnum = in->msg_qnum;
712 v.msg_qbytes = in->msg_qbytes;
713 v.msg_lspid = in->msg_lspid;
714 v.msg_lrpid = in->msg_lrpid;
715 return copy_to_user(buf, &v, sizeof(v));
716 } else {
717 struct compat_msqid_ds v;
718 memset(&v, 0, sizeof(v));
719 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
720 v.msg_stime = in->msg_stime;
721 v.msg_rtime = in->msg_rtime;
722 v.msg_ctime = in->msg_ctime;
723 v.msg_cbytes = in->msg_cbytes;
724 v.msg_qnum = in->msg_qnum;
725 v.msg_qbytes = in->msg_qbytes;
726 v.msg_lspid = in->msg_lspid;
727 v.msg_lrpid = in->msg_lrpid;
728 return copy_to_user(buf, &v, sizeof(v));
729 }
730}
731
732static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
733{
734 struct ipc_namespace *ns;
735 int err;
736 struct msqid64_ds msqid64;
737
738 ns = current->nsproxy->ipc_ns;
739
740 if (msqid < 0 || cmd < 0)
741 return -EINVAL;
742
743 switch (cmd & (~IPC_64)) {
744 case IPC_INFO:
745 case MSG_INFO: {
746 struct msginfo msginfo;
747 err = msgctl_info(ns, msqid, cmd, &msginfo);
748 if (err < 0)
749 return err;
750 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
751 err = -EFAULT;
752 return err;
753 }
754 case IPC_STAT:
755 case MSG_STAT:
756 case MSG_STAT_ANY:
757 err = msgctl_stat(ns, msqid, cmd, &msqid64);
758 if (err < 0)
759 return err;
760 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
761 err = -EFAULT;
762 return err;
763 case IPC_SET:
764 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
765 return -EFAULT;
766 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
767 case IPC_RMID:
768 return msgctl_down(ns, msqid, cmd, NULL, 0);
769 default:
770 return -EINVAL;
771 }
772}
773
774COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
775{
776 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
777}
778
779#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
780long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
781{
782 int version = compat_ipc_parse_version(&cmd);
783
784 return compat_ksys_msgctl(msqid, cmd, uptr, version);
785}
786
787COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
788{
789 return compat_ksys_old_msgctl(msqid, cmd, uptr);
790}
791#endif
792#endif
793
794static int testmsg(struct msg_msg *msg, long type, int mode)
795{
796 switch (mode) {
797 case SEARCH_ANY:
798 case SEARCH_NUMBER:
799 return 1;
800 case SEARCH_LESSEQUAL:
801 if (msg->m_type <= type)
802 return 1;
803 break;
804 case SEARCH_EQUAL:
805 if (msg->m_type == type)
806 return 1;
807 break;
808 case SEARCH_NOTEQUAL:
809 if (msg->m_type != type)
810 return 1;
811 break;
812 }
813 return 0;
814}
815
816static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
817 struct wake_q_head *wake_q)
818{
819 struct msg_receiver *msr, *t;
820
821 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
822 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
823 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
824 msr->r_msgtype, msr->r_mode)) {
825
826 list_del(&msr->r_list);
827 if (msr->r_maxsize < msg->m_ts) {
828 wake_q_add(wake_q, msr->r_tsk);
829
830 /* See expunge_all regarding memory barrier */
831 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
832 } else {
833 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
834 msq->q_rtime = ktime_get_real_seconds();
835
836 wake_q_add(wake_q, msr->r_tsk);
837
838 /* See expunge_all regarding memory barrier */
839 smp_store_release(&msr->r_msg, msg);
840 return 1;
841 }
842 }
843 }
844
845 return 0;
846}
847
848static long do_msgsnd(int msqid, long mtype, void __user *mtext,
849 size_t msgsz, int msgflg)
850{
851 struct msg_queue *msq;
852 struct msg_msg *msg;
853 int err;
854 struct ipc_namespace *ns;
855 DEFINE_WAKE_Q(wake_q);
856
857 ns = current->nsproxy->ipc_ns;
858
859 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
860 return -EINVAL;
861 if (mtype < 1)
862 return -EINVAL;
863
864 msg = load_msg(mtext, msgsz);
865 if (IS_ERR(msg))
866 return PTR_ERR(msg);
867
868 msg->m_type = mtype;
869 msg->m_ts = msgsz;
870
871 rcu_read_lock();
872 msq = msq_obtain_object_check(ns, msqid);
873 if (IS_ERR(msq)) {
874 err = PTR_ERR(msq);
875 goto out_unlock1;
876 }
877
878 ipc_lock_object(&msq->q_perm);
879
880 for (;;) {
881 struct msg_sender s;
882
883 err = -EACCES;
884 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
885 goto out_unlock0;
886
887 /* raced with RMID? */
888 if (!ipc_valid_object(&msq->q_perm)) {
889 err = -EIDRM;
890 goto out_unlock0;
891 }
892
893 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
894 if (err)
895 goto out_unlock0;
896
897 if (msg_fits_inqueue(msq, msgsz))
898 break;
899
900 /* queue full, wait: */
901 if (msgflg & IPC_NOWAIT) {
902 err = -EAGAIN;
903 goto out_unlock0;
904 }
905
906 /* enqueue the sender and prepare to block */
907 ss_add(msq, &s, msgsz);
908
909 if (!ipc_rcu_getref(&msq->q_perm)) {
910 err = -EIDRM;
911 goto out_unlock0;
912 }
913
914 ipc_unlock_object(&msq->q_perm);
915 rcu_read_unlock();
916 schedule();
917
918 rcu_read_lock();
919 ipc_lock_object(&msq->q_perm);
920
921 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
922 /* raced with RMID? */
923 if (!ipc_valid_object(&msq->q_perm)) {
924 err = -EIDRM;
925 goto out_unlock0;
926 }
927 ss_del(&s);
928
929 if (signal_pending(current)) {
930 err = -ERESTARTNOHAND;
931 goto out_unlock0;
932 }
933
934 }
935
936 ipc_update_pid(&msq->q_lspid, task_tgid(current));
937 msq->q_stime = ktime_get_real_seconds();
938
939 if (!pipelined_send(msq, msg, &wake_q)) {
940 /* no one is waiting for this message, enqueue it */
941 list_add_tail(&msg->m_list, &msq->q_messages);
942 msq->q_cbytes += msgsz;
943 msq->q_qnum++;
944 percpu_counter_add_local(&ns->percpu_msg_bytes, msgsz);
945 percpu_counter_add_local(&ns->percpu_msg_hdrs, 1);
946 }
947
948 err = 0;
949 msg = NULL;
950
951out_unlock0:
952 ipc_unlock_object(&msq->q_perm);
953 wake_up_q(&wake_q);
954out_unlock1:
955 rcu_read_unlock();
956 if (msg != NULL)
957 free_msg(msg);
958 return err;
959}
960
961long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
962 int msgflg)
963{
964 long mtype;
965
966 if (get_user(mtype, &msgp->mtype))
967 return -EFAULT;
968 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
969}
970
971SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
972 int, msgflg)
973{
974 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
975}
976
977#ifdef CONFIG_COMPAT
978
979struct compat_msgbuf {
980 compat_long_t mtype;
981 char mtext[1];
982};
983
984long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
985 compat_ssize_t msgsz, int msgflg)
986{
987 struct compat_msgbuf __user *up = compat_ptr(msgp);
988 compat_long_t mtype;
989
990 if (get_user(mtype, &up->mtype))
991 return -EFAULT;
992 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
993}
994
995COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
996 compat_ssize_t, msgsz, int, msgflg)
997{
998 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
999}
1000#endif
1001
1002static inline int convert_mode(long *msgtyp, int msgflg)
1003{
1004 if (msgflg & MSG_COPY)
1005 return SEARCH_NUMBER;
1006 /*
1007 * find message of correct type.
1008 * msgtyp = 0 => get first.
1009 * msgtyp > 0 => get first message of matching type.
1010 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1011 */
1012 if (*msgtyp == 0)
1013 return SEARCH_ANY;
1014 if (*msgtyp < 0) {
1015 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1016 *msgtyp = LONG_MAX;
1017 else
1018 *msgtyp = -*msgtyp;
1019 return SEARCH_LESSEQUAL;
1020 }
1021 if (msgflg & MSG_EXCEPT)
1022 return SEARCH_NOTEQUAL;
1023 return SEARCH_EQUAL;
1024}
1025
1026static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1027{
1028 struct msgbuf __user *msgp = dest;
1029 size_t msgsz;
1030
1031 if (put_user(msg->m_type, &msgp->mtype))
1032 return -EFAULT;
1033
1034 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1035 if (store_msg(msgp->mtext, msg, msgsz))
1036 return -EFAULT;
1037 return msgsz;
1038}
1039
1040#ifdef CONFIG_CHECKPOINT_RESTORE
1041/*
1042 * This function creates new kernel message structure, large enough to store
1043 * bufsz message bytes.
1044 */
1045static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1046{
1047 struct msg_msg *copy;
1048
1049 /*
1050 * Create dummy message to copy real message to.
1051 */
1052 copy = load_msg(buf, bufsz);
1053 if (!IS_ERR(copy))
1054 copy->m_ts = bufsz;
1055 return copy;
1056}
1057
1058static inline void free_copy(struct msg_msg *copy)
1059{
1060 if (copy)
1061 free_msg(copy);
1062}
1063#else
1064static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1065{
1066 return ERR_PTR(-ENOSYS);
1067}
1068
1069static inline void free_copy(struct msg_msg *copy)
1070{
1071}
1072#endif
1073
1074static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1075{
1076 struct msg_msg *msg, *found = NULL;
1077 long count = 0;
1078
1079 list_for_each_entry(msg, &msq->q_messages, m_list) {
1080 if (testmsg(msg, *msgtyp, mode) &&
1081 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1082 *msgtyp, mode)) {
1083 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1084 *msgtyp = msg->m_type - 1;
1085 found = msg;
1086 } else if (mode == SEARCH_NUMBER) {
1087 if (*msgtyp == count)
1088 return msg;
1089 } else
1090 return msg;
1091 count++;
1092 }
1093 }
1094
1095 return found ?: ERR_PTR(-EAGAIN);
1096}
1097
1098static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1099 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1100{
1101 int mode;
1102 struct msg_queue *msq;
1103 struct ipc_namespace *ns;
1104 struct msg_msg *msg, *copy = NULL;
1105 DEFINE_WAKE_Q(wake_q);
1106
1107 ns = current->nsproxy->ipc_ns;
1108
1109 if (msqid < 0 || (long) bufsz < 0)
1110 return -EINVAL;
1111
1112 if (msgflg & MSG_COPY) {
1113 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1114 return -EINVAL;
1115 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1116 if (IS_ERR(copy))
1117 return PTR_ERR(copy);
1118 }
1119 mode = convert_mode(&msgtyp, msgflg);
1120
1121 rcu_read_lock();
1122 msq = msq_obtain_object_check(ns, msqid);
1123 if (IS_ERR(msq)) {
1124 rcu_read_unlock();
1125 free_copy(copy);
1126 return PTR_ERR(msq);
1127 }
1128
1129 for (;;) {
1130 struct msg_receiver msr_d;
1131
1132 msg = ERR_PTR(-EACCES);
1133 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1134 goto out_unlock1;
1135
1136 ipc_lock_object(&msq->q_perm);
1137
1138 /* raced with RMID? */
1139 if (!ipc_valid_object(&msq->q_perm)) {
1140 msg = ERR_PTR(-EIDRM);
1141 goto out_unlock0;
1142 }
1143
1144 msg = find_msg(msq, &msgtyp, mode);
1145 if (!IS_ERR(msg)) {
1146 /*
1147 * Found a suitable message.
1148 * Unlink it from the queue.
1149 */
1150 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1151 msg = ERR_PTR(-E2BIG);
1152 goto out_unlock0;
1153 }
1154 /*
1155 * If we are copying, then do not unlink message and do
1156 * not update queue parameters.
1157 */
1158 if (msgflg & MSG_COPY) {
1159 msg = copy_msg(msg, copy);
1160 goto out_unlock0;
1161 }
1162
1163 list_del(&msg->m_list);
1164 msq->q_qnum--;
1165 msq->q_rtime = ktime_get_real_seconds();
1166 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1167 msq->q_cbytes -= msg->m_ts;
1168 percpu_counter_sub_local(&ns->percpu_msg_bytes, msg->m_ts);
1169 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
1170 ss_wakeup(msq, &wake_q, false);
1171
1172 goto out_unlock0;
1173 }
1174
1175 /* No message waiting. Wait for a message */
1176 if (msgflg & IPC_NOWAIT) {
1177 msg = ERR_PTR(-ENOMSG);
1178 goto out_unlock0;
1179 }
1180
1181 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1182 msr_d.r_tsk = current;
1183 msr_d.r_msgtype = msgtyp;
1184 msr_d.r_mode = mode;
1185 if (msgflg & MSG_NOERROR)
1186 msr_d.r_maxsize = INT_MAX;
1187 else
1188 msr_d.r_maxsize = bufsz;
1189
1190 /* memory barrier not require due to ipc_lock_object() */
1191 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1192
1193 /* memory barrier not required, we own ipc_lock_object() */
1194 __set_current_state(TASK_INTERRUPTIBLE);
1195
1196 ipc_unlock_object(&msq->q_perm);
1197 rcu_read_unlock();
1198 schedule();
1199
1200 /*
1201 * Lockless receive, part 1:
1202 * We don't hold a reference to the queue and getting a
1203 * reference would defeat the idea of a lockless operation,
1204 * thus the code relies on rcu to guarantee the existence of
1205 * msq:
1206 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1207 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1208 */
1209 rcu_read_lock();
1210
1211 /*
1212 * Lockless receive, part 2:
1213 * The work in pipelined_send() and expunge_all():
1214 * - Set pointer to message
1215 * - Queue the receiver task for later wakeup
1216 * - Wake up the process after the lock is dropped.
1217 *
1218 * Should the process wake up before this wakeup (due to a
1219 * signal) it will either see the message and continue ...
1220 */
1221 msg = READ_ONCE(msr_d.r_msg);
1222 if (msg != ERR_PTR(-EAGAIN)) {
1223 /* see MSG_BARRIER for purpose/pairing */
1224 smp_acquire__after_ctrl_dep();
1225
1226 goto out_unlock1;
1227 }
1228
1229 /*
1230 * ... or see -EAGAIN, acquire the lock to check the message
1231 * again.
1232 */
1233 ipc_lock_object(&msq->q_perm);
1234
1235 msg = READ_ONCE(msr_d.r_msg);
1236 if (msg != ERR_PTR(-EAGAIN))
1237 goto out_unlock0;
1238
1239 list_del(&msr_d.r_list);
1240 if (signal_pending(current)) {
1241 msg = ERR_PTR(-ERESTARTNOHAND);
1242 goto out_unlock0;
1243 }
1244
1245 ipc_unlock_object(&msq->q_perm);
1246 }
1247
1248out_unlock0:
1249 ipc_unlock_object(&msq->q_perm);
1250 wake_up_q(&wake_q);
1251out_unlock1:
1252 rcu_read_unlock();
1253 if (IS_ERR(msg)) {
1254 free_copy(copy);
1255 return PTR_ERR(msg);
1256 }
1257
1258 bufsz = msg_handler(buf, msg, bufsz);
1259 free_msg(msg);
1260
1261 return bufsz;
1262}
1263
1264long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1265 long msgtyp, int msgflg)
1266{
1267 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1268}
1269
1270SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1271 long, msgtyp, int, msgflg)
1272{
1273 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1274}
1275
1276#ifdef CONFIG_COMPAT
1277static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1278{
1279 struct compat_msgbuf __user *msgp = dest;
1280 size_t msgsz;
1281
1282 if (put_user(msg->m_type, &msgp->mtype))
1283 return -EFAULT;
1284
1285 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1286 if (store_msg(msgp->mtext, msg, msgsz))
1287 return -EFAULT;
1288 return msgsz;
1289}
1290
1291long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1292 compat_long_t msgtyp, int msgflg)
1293{
1294 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1295 msgflg, compat_do_msg_fill);
1296}
1297
1298COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1299 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1300 int, msgflg)
1301{
1302 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1303}
1304#endif
1305
1306int msg_init_ns(struct ipc_namespace *ns)
1307{
1308 int ret;
1309
1310 ns->msg_ctlmax = MSGMAX;
1311 ns->msg_ctlmnb = MSGMNB;
1312 ns->msg_ctlmni = MSGMNI;
1313
1314 ret = percpu_counter_init(&ns->percpu_msg_bytes, 0, GFP_KERNEL);
1315 if (ret)
1316 goto fail_msg_bytes;
1317 ret = percpu_counter_init(&ns->percpu_msg_hdrs, 0, GFP_KERNEL);
1318 if (ret)
1319 goto fail_msg_hdrs;
1320 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1321 return 0;
1322
1323fail_msg_hdrs:
1324 percpu_counter_destroy(&ns->percpu_msg_bytes);
1325fail_msg_bytes:
1326 return ret;
1327}
1328
1329#ifdef CONFIG_IPC_NS
1330void msg_exit_ns(struct ipc_namespace *ns)
1331{
1332 free_ipcs(ns, &msg_ids(ns), freeque);
1333 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1334 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1335 percpu_counter_destroy(&ns->percpu_msg_bytes);
1336 percpu_counter_destroy(&ns->percpu_msg_hdrs);
1337}
1338#endif
1339
1340#ifdef CONFIG_PROC_FS
1341static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1342{
1343 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1344 struct user_namespace *user_ns = seq_user_ns(s);
1345 struct kern_ipc_perm *ipcp = it;
1346 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1347
1348 seq_printf(s,
1349 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1350 msq->q_perm.key,
1351 msq->q_perm.id,
1352 msq->q_perm.mode,
1353 msq->q_cbytes,
1354 msq->q_qnum,
1355 pid_nr_ns(msq->q_lspid, pid_ns),
1356 pid_nr_ns(msq->q_lrpid, pid_ns),
1357 from_kuid_munged(user_ns, msq->q_perm.uid),
1358 from_kgid_munged(user_ns, msq->q_perm.gid),
1359 from_kuid_munged(user_ns, msq->q_perm.cuid),
1360 from_kgid_munged(user_ns, msq->q_perm.cgid),
1361 msq->q_stime,
1362 msq->q_rtime,
1363 msq->q_ctime);
1364
1365 return 0;
1366}
1367#endif
1368
1369void __init msg_init(void)
1370{
1371 msg_init_ns(&init_ipc_ns);
1372
1373 ipc_init_proc_interface("sysvipc/msg",
1374 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1375 IPC_MSG_IDS, sysvipc_msg_proc_show);
1376}
1/*
2 * linux/ipc/msg.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 *
5 * Removed all the remaining kerneld mess
6 * Catch the -EFAULT stuff properly
7 * Use GFP_KERNEL for messages as in 1.2
8 * Fixed up the unchecked user space derefs
9 * Copyright (C) 1998 Alan Cox & Andi Kleen
10 *
11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
12 *
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
15 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 */
24
25#include <linux/capability.h>
26#include <linux/msg.h>
27#include <linux/spinlock.h>
28#include <linux/init.h>
29#include <linux/mm.h>
30#include <linux/proc_fs.h>
31#include <linux/list.h>
32#include <linux/security.h>
33#include <linux/sched.h>
34#include <linux/syscalls.h>
35#include <linux/audit.h>
36#include <linux/seq_file.h>
37#include <linux/rwsem.h>
38#include <linux/nsproxy.h>
39#include <linux/ipc_namespace.h>
40
41#include <asm/current.h>
42#include <linux/uaccess.h>
43#include "util.h"
44
45/* one msg_receiver structure for each sleeping receiver */
46struct msg_receiver {
47 struct list_head r_list;
48 struct task_struct *r_tsk;
49
50 int r_mode;
51 long r_msgtype;
52 long r_maxsize;
53
54 /*
55 * Mark r_msg volatile so that the compiler
56 * does not try to get smart and optimize
57 * it. We rely on this for the lockless
58 * receive algorithm.
59 */
60 struct msg_msg *volatile r_msg;
61};
62
63/* one msg_sender for each sleeping sender */
64struct msg_sender {
65 struct list_head list;
66 struct task_struct *tsk;
67};
68
69#define SEARCH_ANY 1
70#define SEARCH_EQUAL 2
71#define SEARCH_NOTEQUAL 3
72#define SEARCH_LESSEQUAL 4
73#define SEARCH_NUMBER 5
74
75#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
76
77static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
78{
79 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
80
81 if (IS_ERR(ipcp))
82 return ERR_CAST(ipcp);
83
84 return container_of(ipcp, struct msg_queue, q_perm);
85}
86
87static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
88 int id)
89{
90 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
91
92 if (IS_ERR(ipcp))
93 return ERR_CAST(ipcp);
94
95 return container_of(ipcp, struct msg_queue, q_perm);
96}
97
98static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
99{
100 ipc_rmid(&msg_ids(ns), &s->q_perm);
101}
102
103static void msg_rcu_free(struct rcu_head *head)
104{
105 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
106 struct msg_queue *msq = ipc_rcu_to_struct(p);
107
108 security_msg_queue_free(msq);
109 ipc_rcu_free(head);
110}
111
112/**
113 * newque - Create a new msg queue
114 * @ns: namespace
115 * @params: ptr to the structure that contains the key and msgflg
116 *
117 * Called with msg_ids.rwsem held (writer)
118 */
119static int newque(struct ipc_namespace *ns, struct ipc_params *params)
120{
121 struct msg_queue *msq;
122 int id, retval;
123 key_t key = params->key;
124 int msgflg = params->flg;
125
126 msq = ipc_rcu_alloc(sizeof(*msq));
127 if (!msq)
128 return -ENOMEM;
129
130 msq->q_perm.mode = msgflg & S_IRWXUGO;
131 msq->q_perm.key = key;
132
133 msq->q_perm.security = NULL;
134 retval = security_msg_queue_alloc(msq);
135 if (retval) {
136 ipc_rcu_putref(msq, ipc_rcu_free);
137 return retval;
138 }
139
140 msq->q_stime = msq->q_rtime = 0;
141 msq->q_ctime = get_seconds();
142 msq->q_cbytes = msq->q_qnum = 0;
143 msq->q_qbytes = ns->msg_ctlmnb;
144 msq->q_lspid = msq->q_lrpid = 0;
145 INIT_LIST_HEAD(&msq->q_messages);
146 INIT_LIST_HEAD(&msq->q_receivers);
147 INIT_LIST_HEAD(&msq->q_senders);
148
149 /* ipc_addid() locks msq upon success. */
150 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
151 if (id < 0) {
152 ipc_rcu_putref(msq, msg_rcu_free);
153 return id;
154 }
155
156 ipc_unlock_object(&msq->q_perm);
157 rcu_read_unlock();
158
159 return msq->q_perm.id;
160}
161
162static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
163{
164 mss->tsk = current;
165 __set_current_state(TASK_INTERRUPTIBLE);
166 list_add_tail(&mss->list, &msq->q_senders);
167}
168
169static inline void ss_del(struct msg_sender *mss)
170{
171 if (mss->list.next != NULL)
172 list_del(&mss->list);
173}
174
175static void ss_wakeup(struct list_head *h, int kill)
176{
177 struct msg_sender *mss, *t;
178
179 list_for_each_entry_safe(mss, t, h, list) {
180 if (kill)
181 mss->list.next = NULL;
182 wake_up_process(mss->tsk);
183 }
184}
185
186static void expunge_all(struct msg_queue *msq, int res)
187{
188 struct msg_receiver *msr, *t;
189
190 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
191 msr->r_msg = NULL; /* initialize expunge ordering */
192 wake_up_process(msr->r_tsk);
193 /*
194 * Ensure that the wakeup is visible before setting r_msg as
195 * the receiving end depends on it: either spinning on a nil,
196 * or dealing with -EAGAIN cases. See lockless receive part 1
197 * and 2 in do_msgrcv().
198 */
199 smp_wmb(); /* barrier (B) */
200 msr->r_msg = ERR_PTR(res);
201 }
202}
203
204/*
205 * freeque() wakes up waiters on the sender and receiver waiting queue,
206 * removes the message queue from message queue ID IDR, and cleans up all the
207 * messages associated with this queue.
208 *
209 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
210 * before freeque() is called. msg_ids.rwsem remains locked on exit.
211 */
212static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
213{
214 struct msg_msg *msg, *t;
215 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
216
217 expunge_all(msq, -EIDRM);
218 ss_wakeup(&msq->q_senders, 1);
219 msg_rmid(ns, msq);
220 ipc_unlock_object(&msq->q_perm);
221 rcu_read_unlock();
222
223 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
224 atomic_dec(&ns->msg_hdrs);
225 free_msg(msg);
226 }
227 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
228 ipc_rcu_putref(msq, msg_rcu_free);
229}
230
231/*
232 * Called with msg_ids.rwsem and ipcp locked.
233 */
234static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
235{
236 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
237
238 return security_msg_queue_associate(msq, msgflg);
239}
240
241SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
242{
243 struct ipc_namespace *ns;
244 static const struct ipc_ops msg_ops = {
245 .getnew = newque,
246 .associate = msg_security,
247 };
248 struct ipc_params msg_params;
249
250 ns = current->nsproxy->ipc_ns;
251
252 msg_params.key = key;
253 msg_params.flg = msgflg;
254
255 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
256}
257
258static inline unsigned long
259copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
260{
261 switch (version) {
262 case IPC_64:
263 return copy_to_user(buf, in, sizeof(*in));
264 case IPC_OLD:
265 {
266 struct msqid_ds out;
267
268 memset(&out, 0, sizeof(out));
269
270 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
271
272 out.msg_stime = in->msg_stime;
273 out.msg_rtime = in->msg_rtime;
274 out.msg_ctime = in->msg_ctime;
275
276 if (in->msg_cbytes > USHRT_MAX)
277 out.msg_cbytes = USHRT_MAX;
278 else
279 out.msg_cbytes = in->msg_cbytes;
280 out.msg_lcbytes = in->msg_cbytes;
281
282 if (in->msg_qnum > USHRT_MAX)
283 out.msg_qnum = USHRT_MAX;
284 else
285 out.msg_qnum = in->msg_qnum;
286
287 if (in->msg_qbytes > USHRT_MAX)
288 out.msg_qbytes = USHRT_MAX;
289 else
290 out.msg_qbytes = in->msg_qbytes;
291 out.msg_lqbytes = in->msg_qbytes;
292
293 out.msg_lspid = in->msg_lspid;
294 out.msg_lrpid = in->msg_lrpid;
295
296 return copy_to_user(buf, &out, sizeof(out));
297 }
298 default:
299 return -EINVAL;
300 }
301}
302
303static inline unsigned long
304copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
305{
306 switch (version) {
307 case IPC_64:
308 if (copy_from_user(out, buf, sizeof(*out)))
309 return -EFAULT;
310 return 0;
311 case IPC_OLD:
312 {
313 struct msqid_ds tbuf_old;
314
315 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
316 return -EFAULT;
317
318 out->msg_perm.uid = tbuf_old.msg_perm.uid;
319 out->msg_perm.gid = tbuf_old.msg_perm.gid;
320 out->msg_perm.mode = tbuf_old.msg_perm.mode;
321
322 if (tbuf_old.msg_qbytes == 0)
323 out->msg_qbytes = tbuf_old.msg_lqbytes;
324 else
325 out->msg_qbytes = tbuf_old.msg_qbytes;
326
327 return 0;
328 }
329 default:
330 return -EINVAL;
331 }
332}
333
334/*
335 * This function handles some msgctl commands which require the rwsem
336 * to be held in write mode.
337 * NOTE: no locks must be held, the rwsem is taken inside this function.
338 */
339static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
340 struct msqid_ds __user *buf, int version)
341{
342 struct kern_ipc_perm *ipcp;
343 struct msqid64_ds uninitialized_var(msqid64);
344 struct msg_queue *msq;
345 int err;
346
347 if (cmd == IPC_SET) {
348 if (copy_msqid_from_user(&msqid64, buf, version))
349 return -EFAULT;
350 }
351
352 down_write(&msg_ids(ns).rwsem);
353 rcu_read_lock();
354
355 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
356 &msqid64.msg_perm, msqid64.msg_qbytes);
357 if (IS_ERR(ipcp)) {
358 err = PTR_ERR(ipcp);
359 goto out_unlock1;
360 }
361
362 msq = container_of(ipcp, struct msg_queue, q_perm);
363
364 err = security_msg_queue_msgctl(msq, cmd);
365 if (err)
366 goto out_unlock1;
367
368 switch (cmd) {
369 case IPC_RMID:
370 ipc_lock_object(&msq->q_perm);
371 /* freeque unlocks the ipc object and rcu */
372 freeque(ns, ipcp);
373 goto out_up;
374 case IPC_SET:
375 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
376 !capable(CAP_SYS_RESOURCE)) {
377 err = -EPERM;
378 goto out_unlock1;
379 }
380
381 ipc_lock_object(&msq->q_perm);
382 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
383 if (err)
384 goto out_unlock0;
385
386 msq->q_qbytes = msqid64.msg_qbytes;
387
388 msq->q_ctime = get_seconds();
389 /* sleeping receivers might be excluded by
390 * stricter permissions.
391 */
392 expunge_all(msq, -EAGAIN);
393 /* sleeping senders might be able to send
394 * due to a larger queue size.
395 */
396 ss_wakeup(&msq->q_senders, 0);
397 break;
398 default:
399 err = -EINVAL;
400 goto out_unlock1;
401 }
402
403out_unlock0:
404 ipc_unlock_object(&msq->q_perm);
405out_unlock1:
406 rcu_read_unlock();
407out_up:
408 up_write(&msg_ids(ns).rwsem);
409 return err;
410}
411
412static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
413 int cmd, int version, void __user *buf)
414{
415 int err;
416 struct msg_queue *msq;
417
418 switch (cmd) {
419 case IPC_INFO:
420 case MSG_INFO:
421 {
422 struct msginfo msginfo;
423 int max_id;
424
425 if (!buf)
426 return -EFAULT;
427
428 /*
429 * We must not return kernel stack data.
430 * due to padding, it's not enough
431 * to set all member fields.
432 */
433 err = security_msg_queue_msgctl(NULL, cmd);
434 if (err)
435 return err;
436
437 memset(&msginfo, 0, sizeof(msginfo));
438 msginfo.msgmni = ns->msg_ctlmni;
439 msginfo.msgmax = ns->msg_ctlmax;
440 msginfo.msgmnb = ns->msg_ctlmnb;
441 msginfo.msgssz = MSGSSZ;
442 msginfo.msgseg = MSGSEG;
443 down_read(&msg_ids(ns).rwsem);
444 if (cmd == MSG_INFO) {
445 msginfo.msgpool = msg_ids(ns).in_use;
446 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
447 msginfo.msgtql = atomic_read(&ns->msg_bytes);
448 } else {
449 msginfo.msgmap = MSGMAP;
450 msginfo.msgpool = MSGPOOL;
451 msginfo.msgtql = MSGTQL;
452 }
453 max_id = ipc_get_maxid(&msg_ids(ns));
454 up_read(&msg_ids(ns).rwsem);
455 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
456 return -EFAULT;
457 return (max_id < 0) ? 0 : max_id;
458 }
459
460 case MSG_STAT:
461 case IPC_STAT:
462 {
463 struct msqid64_ds tbuf;
464 int success_return;
465
466 if (!buf)
467 return -EFAULT;
468
469 memset(&tbuf, 0, sizeof(tbuf));
470
471 rcu_read_lock();
472 if (cmd == MSG_STAT) {
473 msq = msq_obtain_object(ns, msqid);
474 if (IS_ERR(msq)) {
475 err = PTR_ERR(msq);
476 goto out_unlock;
477 }
478 success_return = msq->q_perm.id;
479 } else {
480 msq = msq_obtain_object_check(ns, msqid);
481 if (IS_ERR(msq)) {
482 err = PTR_ERR(msq);
483 goto out_unlock;
484 }
485 success_return = 0;
486 }
487
488 err = -EACCES;
489 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
490 goto out_unlock;
491
492 err = security_msg_queue_msgctl(msq, cmd);
493 if (err)
494 goto out_unlock;
495
496 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
497 tbuf.msg_stime = msq->q_stime;
498 tbuf.msg_rtime = msq->q_rtime;
499 tbuf.msg_ctime = msq->q_ctime;
500 tbuf.msg_cbytes = msq->q_cbytes;
501 tbuf.msg_qnum = msq->q_qnum;
502 tbuf.msg_qbytes = msq->q_qbytes;
503 tbuf.msg_lspid = msq->q_lspid;
504 tbuf.msg_lrpid = msq->q_lrpid;
505 rcu_read_unlock();
506
507 if (copy_msqid_to_user(buf, &tbuf, version))
508 return -EFAULT;
509 return success_return;
510 }
511
512 default:
513 return -EINVAL;
514 }
515
516 return err;
517out_unlock:
518 rcu_read_unlock();
519 return err;
520}
521
522SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
523{
524 int version;
525 struct ipc_namespace *ns;
526
527 if (msqid < 0 || cmd < 0)
528 return -EINVAL;
529
530 version = ipc_parse_version(&cmd);
531 ns = current->nsproxy->ipc_ns;
532
533 switch (cmd) {
534 case IPC_INFO:
535 case MSG_INFO:
536 case MSG_STAT: /* msqid is an index rather than a msg queue id */
537 case IPC_STAT:
538 return msgctl_nolock(ns, msqid, cmd, version, buf);
539 case IPC_SET:
540 case IPC_RMID:
541 return msgctl_down(ns, msqid, cmd, buf, version);
542 default:
543 return -EINVAL;
544 }
545}
546
547static int testmsg(struct msg_msg *msg, long type, int mode)
548{
549 switch (mode) {
550 case SEARCH_ANY:
551 case SEARCH_NUMBER:
552 return 1;
553 case SEARCH_LESSEQUAL:
554 if (msg->m_type <= type)
555 return 1;
556 break;
557 case SEARCH_EQUAL:
558 if (msg->m_type == type)
559 return 1;
560 break;
561 case SEARCH_NOTEQUAL:
562 if (msg->m_type != type)
563 return 1;
564 break;
565 }
566 return 0;
567}
568
569static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
570{
571 struct msg_receiver *msr, *t;
572
573 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
574 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
575 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
576 msr->r_msgtype, msr->r_mode)) {
577
578 list_del(&msr->r_list);
579 if (msr->r_maxsize < msg->m_ts) {
580 /* initialize pipelined send ordering */
581 msr->r_msg = NULL;
582 wake_up_process(msr->r_tsk);
583 /* barrier (B) see barrier comment below */
584 smp_wmb();
585 msr->r_msg = ERR_PTR(-E2BIG);
586 } else {
587 msr->r_msg = NULL;
588 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
589 msq->q_rtime = get_seconds();
590 wake_up_process(msr->r_tsk);
591 /*
592 * Ensure that the wakeup is visible before
593 * setting r_msg, as the receiving can otherwise
594 * exit - once r_msg is set, the receiver can
595 * continue. See lockless receive part 1 and 2
596 * in do_msgrcv(). Barrier (B).
597 */
598 smp_wmb();
599 msr->r_msg = msg;
600
601 return 1;
602 }
603 }
604 }
605
606 return 0;
607}
608
609long do_msgsnd(int msqid, long mtype, void __user *mtext,
610 size_t msgsz, int msgflg)
611{
612 struct msg_queue *msq;
613 struct msg_msg *msg;
614 int err;
615 struct ipc_namespace *ns;
616
617 ns = current->nsproxy->ipc_ns;
618
619 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
620 return -EINVAL;
621 if (mtype < 1)
622 return -EINVAL;
623
624 msg = load_msg(mtext, msgsz);
625 if (IS_ERR(msg))
626 return PTR_ERR(msg);
627
628 msg->m_type = mtype;
629 msg->m_ts = msgsz;
630
631 rcu_read_lock();
632 msq = msq_obtain_object_check(ns, msqid);
633 if (IS_ERR(msq)) {
634 err = PTR_ERR(msq);
635 goto out_unlock1;
636 }
637
638 ipc_lock_object(&msq->q_perm);
639
640 for (;;) {
641 struct msg_sender s;
642
643 err = -EACCES;
644 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
645 goto out_unlock0;
646
647 /* raced with RMID? */
648 if (!ipc_valid_object(&msq->q_perm)) {
649 err = -EIDRM;
650 goto out_unlock0;
651 }
652
653 err = security_msg_queue_msgsnd(msq, msg, msgflg);
654 if (err)
655 goto out_unlock0;
656
657 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
658 1 + msq->q_qnum <= msq->q_qbytes) {
659 break;
660 }
661
662 /* queue full, wait: */
663 if (msgflg & IPC_NOWAIT) {
664 err = -EAGAIN;
665 goto out_unlock0;
666 }
667
668 /* enqueue the sender and prepare to block */
669 ss_add(msq, &s);
670
671 if (!ipc_rcu_getref(msq)) {
672 err = -EIDRM;
673 goto out_unlock0;
674 }
675
676 ipc_unlock_object(&msq->q_perm);
677 rcu_read_unlock();
678 schedule();
679
680 rcu_read_lock();
681 ipc_lock_object(&msq->q_perm);
682
683 ipc_rcu_putref(msq, ipc_rcu_free);
684 /* raced with RMID? */
685 if (!ipc_valid_object(&msq->q_perm)) {
686 err = -EIDRM;
687 goto out_unlock0;
688 }
689
690 ss_del(&s);
691
692 if (signal_pending(current)) {
693 err = -ERESTARTNOHAND;
694 goto out_unlock0;
695 }
696
697 }
698 msq->q_lspid = task_tgid_vnr(current);
699 msq->q_stime = get_seconds();
700
701 if (!pipelined_send(msq, msg)) {
702 /* no one is waiting for this message, enqueue it */
703 list_add_tail(&msg->m_list, &msq->q_messages);
704 msq->q_cbytes += msgsz;
705 msq->q_qnum++;
706 atomic_add(msgsz, &ns->msg_bytes);
707 atomic_inc(&ns->msg_hdrs);
708 }
709
710 err = 0;
711 msg = NULL;
712
713out_unlock0:
714 ipc_unlock_object(&msq->q_perm);
715out_unlock1:
716 rcu_read_unlock();
717 if (msg != NULL)
718 free_msg(msg);
719 return err;
720}
721
722SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
723 int, msgflg)
724{
725 long mtype;
726
727 if (get_user(mtype, &msgp->mtype))
728 return -EFAULT;
729 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
730}
731
732static inline int convert_mode(long *msgtyp, int msgflg)
733{
734 if (msgflg & MSG_COPY)
735 return SEARCH_NUMBER;
736 /*
737 * find message of correct type.
738 * msgtyp = 0 => get first.
739 * msgtyp > 0 => get first message of matching type.
740 * msgtyp < 0 => get message with least type must be < abs(msgtype).
741 */
742 if (*msgtyp == 0)
743 return SEARCH_ANY;
744 if (*msgtyp < 0) {
745 *msgtyp = -*msgtyp;
746 return SEARCH_LESSEQUAL;
747 }
748 if (msgflg & MSG_EXCEPT)
749 return SEARCH_NOTEQUAL;
750 return SEARCH_EQUAL;
751}
752
753static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
754{
755 struct msgbuf __user *msgp = dest;
756 size_t msgsz;
757
758 if (put_user(msg->m_type, &msgp->mtype))
759 return -EFAULT;
760
761 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
762 if (store_msg(msgp->mtext, msg, msgsz))
763 return -EFAULT;
764 return msgsz;
765}
766
767#ifdef CONFIG_CHECKPOINT_RESTORE
768/*
769 * This function creates new kernel message structure, large enough to store
770 * bufsz message bytes.
771 */
772static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
773{
774 struct msg_msg *copy;
775
776 /*
777 * Create dummy message to copy real message to.
778 */
779 copy = load_msg(buf, bufsz);
780 if (!IS_ERR(copy))
781 copy->m_ts = bufsz;
782 return copy;
783}
784
785static inline void free_copy(struct msg_msg *copy)
786{
787 if (copy)
788 free_msg(copy);
789}
790#else
791static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
792{
793 return ERR_PTR(-ENOSYS);
794}
795
796static inline void free_copy(struct msg_msg *copy)
797{
798}
799#endif
800
801static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
802{
803 struct msg_msg *msg, *found = NULL;
804 long count = 0;
805
806 list_for_each_entry(msg, &msq->q_messages, m_list) {
807 if (testmsg(msg, *msgtyp, mode) &&
808 !security_msg_queue_msgrcv(msq, msg, current,
809 *msgtyp, mode)) {
810 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
811 *msgtyp = msg->m_type - 1;
812 found = msg;
813 } else if (mode == SEARCH_NUMBER) {
814 if (*msgtyp == count)
815 return msg;
816 } else
817 return msg;
818 count++;
819 }
820 }
821
822 return found ?: ERR_PTR(-EAGAIN);
823}
824
825long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
826 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
827{
828 int mode;
829 struct msg_queue *msq;
830 struct ipc_namespace *ns;
831 struct msg_msg *msg, *copy = NULL;
832
833 ns = current->nsproxy->ipc_ns;
834
835 if (msqid < 0 || (long) bufsz < 0)
836 return -EINVAL;
837
838 if (msgflg & MSG_COPY) {
839 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
840 return -EINVAL;
841 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
842 if (IS_ERR(copy))
843 return PTR_ERR(copy);
844 }
845 mode = convert_mode(&msgtyp, msgflg);
846
847 rcu_read_lock();
848 msq = msq_obtain_object_check(ns, msqid);
849 if (IS_ERR(msq)) {
850 rcu_read_unlock();
851 free_copy(copy);
852 return PTR_ERR(msq);
853 }
854
855 for (;;) {
856 struct msg_receiver msr_d;
857
858 msg = ERR_PTR(-EACCES);
859 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
860 goto out_unlock1;
861
862 ipc_lock_object(&msq->q_perm);
863
864 /* raced with RMID? */
865 if (!ipc_valid_object(&msq->q_perm)) {
866 msg = ERR_PTR(-EIDRM);
867 goto out_unlock0;
868 }
869
870 msg = find_msg(msq, &msgtyp, mode);
871 if (!IS_ERR(msg)) {
872 /*
873 * Found a suitable message.
874 * Unlink it from the queue.
875 */
876 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
877 msg = ERR_PTR(-E2BIG);
878 goto out_unlock0;
879 }
880 /*
881 * If we are copying, then do not unlink message and do
882 * not update queue parameters.
883 */
884 if (msgflg & MSG_COPY) {
885 msg = copy_msg(msg, copy);
886 goto out_unlock0;
887 }
888
889 list_del(&msg->m_list);
890 msq->q_qnum--;
891 msq->q_rtime = get_seconds();
892 msq->q_lrpid = task_tgid_vnr(current);
893 msq->q_cbytes -= msg->m_ts;
894 atomic_sub(msg->m_ts, &ns->msg_bytes);
895 atomic_dec(&ns->msg_hdrs);
896 ss_wakeup(&msq->q_senders, 0);
897
898 goto out_unlock0;
899 }
900
901 /* No message waiting. Wait for a message */
902 if (msgflg & IPC_NOWAIT) {
903 msg = ERR_PTR(-ENOMSG);
904 goto out_unlock0;
905 }
906
907 list_add_tail(&msr_d.r_list, &msq->q_receivers);
908 msr_d.r_tsk = current;
909 msr_d.r_msgtype = msgtyp;
910 msr_d.r_mode = mode;
911 if (msgflg & MSG_NOERROR)
912 msr_d.r_maxsize = INT_MAX;
913 else
914 msr_d.r_maxsize = bufsz;
915 msr_d.r_msg = ERR_PTR(-EAGAIN);
916 __set_current_state(TASK_INTERRUPTIBLE);
917
918 ipc_unlock_object(&msq->q_perm);
919 rcu_read_unlock();
920 schedule();
921
922 /* Lockless receive, part 1:
923 * Disable preemption. We don't hold a reference to the queue
924 * and getting a reference would defeat the idea of a lockless
925 * operation, thus the code relies on rcu to guarantee the
926 * existence of msq:
927 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
928 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
929 * rcu_read_lock() prevents preemption between reading r_msg
930 * and acquiring the q_perm.lock in ipc_lock_object().
931 */
932 rcu_read_lock();
933
934 /* Lockless receive, part 2:
935 * Wait until pipelined_send or expunge_all are outside of
936 * wake_up_process(). There is a race with exit(), see
937 * ipc/mqueue.c for the details. The correct serialization
938 * ensures that a receiver cannot continue without the wakeup
939 * being visibible _before_ setting r_msg:
940 *
941 * CPU 0 CPU 1
942 * <loop receiver>
943 * smp_rmb(); (A) <-- pair -. <waker thread>
944 * <load ->r_msg> | msr->r_msg = NULL;
945 * | wake_up_process();
946 * <continue> `------> smp_wmb(); (B)
947 * msr->r_msg = msg;
948 *
949 * Where (A) orders the message value read and where (B) orders
950 * the write to the r_msg -- done in both pipelined_send and
951 * expunge_all.
952 */
953 for (;;) {
954 /*
955 * Pairs with writer barrier in pipelined_send
956 * or expunge_all.
957 */
958 smp_rmb(); /* barrier (A) */
959 msg = (struct msg_msg *)msr_d.r_msg;
960 if (msg)
961 break;
962
963 /*
964 * The cpu_relax() call is a compiler barrier
965 * which forces everything in this loop to be
966 * re-loaded.
967 */
968 cpu_relax();
969 }
970
971 /* Lockless receive, part 3:
972 * If there is a message or an error then accept it without
973 * locking.
974 */
975 if (msg != ERR_PTR(-EAGAIN))
976 goto out_unlock1;
977
978 /* Lockless receive, part 3:
979 * Acquire the queue spinlock.
980 */
981 ipc_lock_object(&msq->q_perm);
982
983 /* Lockless receive, part 4:
984 * Repeat test after acquiring the spinlock.
985 */
986 msg = (struct msg_msg *)msr_d.r_msg;
987 if (msg != ERR_PTR(-EAGAIN))
988 goto out_unlock0;
989
990 list_del(&msr_d.r_list);
991 if (signal_pending(current)) {
992 msg = ERR_PTR(-ERESTARTNOHAND);
993 goto out_unlock0;
994 }
995
996 ipc_unlock_object(&msq->q_perm);
997 }
998
999out_unlock0:
1000 ipc_unlock_object(&msq->q_perm);
1001out_unlock1:
1002 rcu_read_unlock();
1003 if (IS_ERR(msg)) {
1004 free_copy(copy);
1005 return PTR_ERR(msg);
1006 }
1007
1008 bufsz = msg_handler(buf, msg, bufsz);
1009 free_msg(msg);
1010
1011 return bufsz;
1012}
1013
1014SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1015 long, msgtyp, int, msgflg)
1016{
1017 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1018}
1019
1020
1021void msg_init_ns(struct ipc_namespace *ns)
1022{
1023 ns->msg_ctlmax = MSGMAX;
1024 ns->msg_ctlmnb = MSGMNB;
1025 ns->msg_ctlmni = MSGMNI;
1026
1027 atomic_set(&ns->msg_bytes, 0);
1028 atomic_set(&ns->msg_hdrs, 0);
1029 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1030}
1031
1032#ifdef CONFIG_IPC_NS
1033void msg_exit_ns(struct ipc_namespace *ns)
1034{
1035 free_ipcs(ns, &msg_ids(ns), freeque);
1036 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1037}
1038#endif
1039
1040#ifdef CONFIG_PROC_FS
1041static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1042{
1043 struct user_namespace *user_ns = seq_user_ns(s);
1044 struct msg_queue *msq = it;
1045
1046 seq_printf(s,
1047 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1048 msq->q_perm.key,
1049 msq->q_perm.id,
1050 msq->q_perm.mode,
1051 msq->q_cbytes,
1052 msq->q_qnum,
1053 msq->q_lspid,
1054 msq->q_lrpid,
1055 from_kuid_munged(user_ns, msq->q_perm.uid),
1056 from_kgid_munged(user_ns, msq->q_perm.gid),
1057 from_kuid_munged(user_ns, msq->q_perm.cuid),
1058 from_kgid_munged(user_ns, msq->q_perm.cgid),
1059 msq->q_stime,
1060 msq->q_rtime,
1061 msq->q_ctime);
1062
1063 return 0;
1064}
1065#endif
1066
1067void __init msg_init(void)
1068{
1069 msg_init_ns(&init_ipc_ns);
1070
1071 ipc_init_proc_interface("sysvipc/msg",
1072 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1073 IPC_MSG_IDS, sysvipc_msg_proc_show);
1074}