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