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