Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/msg.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 *
6 * Removed all the remaining kerneld mess
7 * Catch the -EFAULT stuff properly
8 * Use GFP_KERNEL for messages as in 1.2
9 * Fixed up the unchecked user space derefs
10 * Copyright (C) 1998 Alan Cox & Andi Kleen
11 *
12 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
13 *
14 * mostly rewritten, threaded and wake-one semantics added
15 * MSGMAX limit removed, sysctl's added
16 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
17 *
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
20 *
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
24 */
25
26#include <linux/capability.h>
27#include <linux/msg.h>
28#include <linux/spinlock.h>
29#include <linux/init.h>
30#include <linux/mm.h>
31#include <linux/proc_fs.h>
32#include <linux/list.h>
33#include <linux/security.h>
34#include <linux/sched/wake_q.h>
35#include <linux/syscalls.h>
36#include <linux/audit.h>
37#include <linux/seq_file.h>
38#include <linux/rwsem.h>
39#include <linux/nsproxy.h>
40#include <linux/ipc_namespace.h>
41#include <linux/rhashtable.h>
42#include <linux/percpu_counter.h>
43
44#include <asm/current.h>
45#include <linux/uaccess.h>
46#include "util.h"
47
48/* one msq_queue structure for each present queue on the system */
49struct msg_queue {
50 struct kern_ipc_perm q_perm;
51 time64_t q_stime; /* last msgsnd time */
52 time64_t q_rtime; /* last msgrcv time */
53 time64_t q_ctime; /* last change time */
54 unsigned long q_cbytes; /* current number of bytes on queue */
55 unsigned long q_qnum; /* number of messages in queue */
56 unsigned long q_qbytes; /* max number of bytes on queue */
57 struct pid *q_lspid; /* pid of last msgsnd */
58 struct pid *q_lrpid; /* last receive pid */
59
60 struct list_head q_messages;
61 struct list_head q_receivers;
62 struct list_head q_senders;
63} __randomize_layout;
64
65/*
66 * MSG_BARRIER Locking:
67 *
68 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
69 * does not acquire any locks when it sees that a message exists in
70 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
71 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
72 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
73 */
74
75/* one msg_receiver structure for each sleeping receiver */
76struct msg_receiver {
77 struct list_head r_list;
78 struct task_struct *r_tsk;
79
80 int r_mode;
81 long r_msgtype;
82 long r_maxsize;
83
84 struct msg_msg *r_msg;
85};
86
87/* one msg_sender for each sleeping sender */
88struct msg_sender {
89 struct list_head list;
90 struct task_struct *tsk;
91 size_t msgsz;
92};
93
94#define SEARCH_ANY 1
95#define SEARCH_EQUAL 2
96#define SEARCH_NOTEQUAL 3
97#define SEARCH_LESSEQUAL 4
98#define SEARCH_NUMBER 5
99
100#define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
101
102static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
103{
104 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
105
106 if (IS_ERR(ipcp))
107 return ERR_CAST(ipcp);
108
109 return container_of(ipcp, struct msg_queue, q_perm);
110}
111
112static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
113 int id)
114{
115 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
116
117 if (IS_ERR(ipcp))
118 return ERR_CAST(ipcp);
119
120 return container_of(ipcp, struct msg_queue, q_perm);
121}
122
123static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
124{
125 ipc_rmid(&msg_ids(ns), &s->q_perm);
126}
127
128static void msg_rcu_free(struct rcu_head *head)
129{
130 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
131 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
132
133 security_msg_queue_free(&msq->q_perm);
134 kfree(msq);
135}
136
137/**
138 * newque - Create a new msg queue
139 * @ns: namespace
140 * @params: ptr to the structure that contains the key and msgflg
141 *
142 * Called with msg_ids.rwsem held (writer)
143 */
144static int newque(struct ipc_namespace *ns, struct ipc_params *params)
145{
146 struct msg_queue *msq;
147 int retval;
148 key_t key = params->key;
149 int msgflg = params->flg;
150
151 msq = kmalloc(sizeof(*msq), GFP_KERNEL_ACCOUNT);
152 if (unlikely(!msq))
153 return -ENOMEM;
154
155 msq->q_perm.mode = msgflg & S_IRWXUGO;
156 msq->q_perm.key = key;
157
158 msq->q_perm.security = NULL;
159 retval = security_msg_queue_alloc(&msq->q_perm);
160 if (retval) {
161 kfree(msq);
162 return retval;
163 }
164
165 msq->q_stime = msq->q_rtime = 0;
166 msq->q_ctime = ktime_get_real_seconds();
167 msq->q_cbytes = msq->q_qnum = 0;
168 msq->q_qbytes = ns->msg_ctlmnb;
169 msq->q_lspid = msq->q_lrpid = NULL;
170 INIT_LIST_HEAD(&msq->q_messages);
171 INIT_LIST_HEAD(&msq->q_receivers);
172 INIT_LIST_HEAD(&msq->q_senders);
173
174 /* ipc_addid() locks msq upon success. */
175 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
176 if (retval < 0) {
177 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
178 return retval;
179 }
180
181 ipc_unlock_object(&msq->q_perm);
182 rcu_read_unlock();
183
184 return msq->q_perm.id;
185}
186
187static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
188{
189 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
190 1 + msq->q_qnum <= msq->q_qbytes;
191}
192
193static inline void ss_add(struct msg_queue *msq,
194 struct msg_sender *mss, size_t msgsz)
195{
196 mss->tsk = current;
197 mss->msgsz = msgsz;
198 /*
199 * No memory barrier required: we did ipc_lock_object(),
200 * and the waker obtains that lock before calling wake_q_add().
201 */
202 __set_current_state(TASK_INTERRUPTIBLE);
203 list_add_tail(&mss->list, &msq->q_senders);
204}
205
206static inline void ss_del(struct msg_sender *mss)
207{
208 if (mss->list.next)
209 list_del(&mss->list);
210}
211
212static void ss_wakeup(struct msg_queue *msq,
213 struct wake_q_head *wake_q, bool kill)
214{
215 struct msg_sender *mss, *t;
216 struct task_struct *stop_tsk = NULL;
217 struct list_head *h = &msq->q_senders;
218
219 list_for_each_entry_safe(mss, t, h, list) {
220 if (kill)
221 mss->list.next = NULL;
222
223 /*
224 * Stop at the first task we don't wakeup,
225 * we've already iterated the original
226 * sender queue.
227 */
228 else if (stop_tsk == mss->tsk)
229 break;
230 /*
231 * We are not in an EIDRM scenario here, therefore
232 * verify that we really need to wakeup the task.
233 * To maintain current semantics and wakeup order,
234 * move the sender to the tail on behalf of the
235 * blocked task.
236 */
237 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
238 if (!stop_tsk)
239 stop_tsk = mss->tsk;
240
241 list_move_tail(&mss->list, &msq->q_senders);
242 continue;
243 }
244
245 wake_q_add(wake_q, mss->tsk);
246 }
247}
248
249static void expunge_all(struct msg_queue *msq, int res,
250 struct wake_q_head *wake_q)
251{
252 struct msg_receiver *msr, *t;
253
254 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
255 struct task_struct *r_tsk;
256
257 r_tsk = get_task_struct(msr->r_tsk);
258
259 /* see MSG_BARRIER for purpose/pairing */
260 smp_store_release(&msr->r_msg, ERR_PTR(res));
261 wake_q_add_safe(wake_q, r_tsk);
262 }
263}
264
265/*
266 * freeque() wakes up waiters on the sender and receiver waiting queue,
267 * removes the message queue from message queue ID IDR, and cleans up all the
268 * messages associated with this queue.
269 *
270 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
271 * before freeque() is called. msg_ids.rwsem remains locked on exit.
272 */
273static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
274 __releases(RCU)
275 __releases(&msq->q_perm)
276{
277 struct msg_msg *msg, *t;
278 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
279 DEFINE_WAKE_Q(wake_q);
280
281 expunge_all(msq, -EIDRM, &wake_q);
282 ss_wakeup(msq, &wake_q, true);
283 msg_rmid(ns, msq);
284 ipc_unlock_object(&msq->q_perm);
285 wake_up_q(&wake_q);
286 rcu_read_unlock();
287
288 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
289 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
290 free_msg(msg);
291 }
292 percpu_counter_sub_local(&ns->percpu_msg_bytes, msq->q_cbytes);
293 ipc_update_pid(&msq->q_lspid, NULL);
294 ipc_update_pid(&msq->q_lrpid, NULL);
295 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
296}
297
298long ksys_msgget(key_t key, int msgflg)
299{
300 struct ipc_namespace *ns;
301 static const struct ipc_ops msg_ops = {
302 .getnew = newque,
303 .associate = security_msg_queue_associate,
304 };
305 struct ipc_params msg_params;
306
307 ns = current->nsproxy->ipc_ns;
308
309 msg_params.key = key;
310 msg_params.flg = msgflg;
311
312 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
313}
314
315SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
316{
317 return ksys_msgget(key, msgflg);
318}
319
320static inline unsigned long
321copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
322{
323 switch (version) {
324 case IPC_64:
325 return copy_to_user(buf, in, sizeof(*in));
326 case IPC_OLD:
327 {
328 struct msqid_ds out;
329
330 memset(&out, 0, sizeof(out));
331
332 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
333
334 out.msg_stime = in->msg_stime;
335 out.msg_rtime = in->msg_rtime;
336 out.msg_ctime = in->msg_ctime;
337
338 if (in->msg_cbytes > USHRT_MAX)
339 out.msg_cbytes = USHRT_MAX;
340 else
341 out.msg_cbytes = in->msg_cbytes;
342 out.msg_lcbytes = in->msg_cbytes;
343
344 if (in->msg_qnum > USHRT_MAX)
345 out.msg_qnum = USHRT_MAX;
346 else
347 out.msg_qnum = in->msg_qnum;
348
349 if (in->msg_qbytes > USHRT_MAX)
350 out.msg_qbytes = USHRT_MAX;
351 else
352 out.msg_qbytes = in->msg_qbytes;
353 out.msg_lqbytes = in->msg_qbytes;
354
355 out.msg_lspid = in->msg_lspid;
356 out.msg_lrpid = in->msg_lrpid;
357
358 return copy_to_user(buf, &out, sizeof(out));
359 }
360 default:
361 return -EINVAL;
362 }
363}
364
365static inline unsigned long
366copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
367{
368 switch (version) {
369 case IPC_64:
370 if (copy_from_user(out, buf, sizeof(*out)))
371 return -EFAULT;
372 return 0;
373 case IPC_OLD:
374 {
375 struct msqid_ds tbuf_old;
376
377 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
378 return -EFAULT;
379
380 out->msg_perm.uid = tbuf_old.msg_perm.uid;
381 out->msg_perm.gid = tbuf_old.msg_perm.gid;
382 out->msg_perm.mode = tbuf_old.msg_perm.mode;
383
384 if (tbuf_old.msg_qbytes == 0)
385 out->msg_qbytes = tbuf_old.msg_lqbytes;
386 else
387 out->msg_qbytes = tbuf_old.msg_qbytes;
388
389 return 0;
390 }
391 default:
392 return -EINVAL;
393 }
394}
395
396/*
397 * This function handles some msgctl commands which require the rwsem
398 * to be held in write mode.
399 * NOTE: no locks must be held, the rwsem is taken inside this function.
400 */
401static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
402 struct ipc64_perm *perm, int msg_qbytes)
403{
404 struct kern_ipc_perm *ipcp;
405 struct msg_queue *msq;
406 int err;
407
408 down_write(&msg_ids(ns).rwsem);
409 rcu_read_lock();
410
411 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
412 perm, msg_qbytes);
413 if (IS_ERR(ipcp)) {
414 err = PTR_ERR(ipcp);
415 goto out_unlock1;
416 }
417
418 msq = container_of(ipcp, struct msg_queue, q_perm);
419
420 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
421 if (err)
422 goto out_unlock1;
423
424 switch (cmd) {
425 case IPC_RMID:
426 ipc_lock_object(&msq->q_perm);
427 /* freeque unlocks the ipc object and rcu */
428 freeque(ns, ipcp);
429 goto out_up;
430 case IPC_SET:
431 {
432 DEFINE_WAKE_Q(wake_q);
433
434 if (msg_qbytes > ns->msg_ctlmnb &&
435 !capable(CAP_SYS_RESOURCE)) {
436 err = -EPERM;
437 goto out_unlock1;
438 }
439
440 ipc_lock_object(&msq->q_perm);
441 err = ipc_update_perm(perm, ipcp);
442 if (err)
443 goto out_unlock0;
444
445 msq->q_qbytes = msg_qbytes;
446
447 msq->q_ctime = ktime_get_real_seconds();
448 /*
449 * Sleeping receivers might be excluded by
450 * stricter permissions.
451 */
452 expunge_all(msq, -EAGAIN, &wake_q);
453 /*
454 * Sleeping senders might be able to send
455 * due to a larger queue size.
456 */
457 ss_wakeup(msq, &wake_q, false);
458 ipc_unlock_object(&msq->q_perm);
459 wake_up_q(&wake_q);
460
461 goto out_unlock1;
462 }
463 default:
464 err = -EINVAL;
465 goto out_unlock1;
466 }
467
468out_unlock0:
469 ipc_unlock_object(&msq->q_perm);
470out_unlock1:
471 rcu_read_unlock();
472out_up:
473 up_write(&msg_ids(ns).rwsem);
474 return err;
475}
476
477static int msgctl_info(struct ipc_namespace *ns, int msqid,
478 int cmd, struct msginfo *msginfo)
479{
480 int err;
481 int max_idx;
482
483 /*
484 * We must not return kernel stack data.
485 * due to padding, it's not enough
486 * to set all member fields.
487 */
488 err = security_msg_queue_msgctl(NULL, cmd);
489 if (err)
490 return err;
491
492 memset(msginfo, 0, sizeof(*msginfo));
493 msginfo->msgmni = ns->msg_ctlmni;
494 msginfo->msgmax = ns->msg_ctlmax;
495 msginfo->msgmnb = ns->msg_ctlmnb;
496 msginfo->msgssz = MSGSSZ;
497 msginfo->msgseg = MSGSEG;
498 down_read(&msg_ids(ns).rwsem);
499 if (cmd == MSG_INFO)
500 msginfo->msgpool = msg_ids(ns).in_use;
501 max_idx = ipc_get_maxidx(&msg_ids(ns));
502 up_read(&msg_ids(ns).rwsem);
503 if (cmd == MSG_INFO) {
504 msginfo->msgmap = min_t(int,
505 percpu_counter_sum(&ns->percpu_msg_hdrs),
506 INT_MAX);
507 msginfo->msgtql = min_t(int,
508 percpu_counter_sum(&ns->percpu_msg_bytes),
509 INT_MAX);
510 } else {
511 msginfo->msgmap = MSGMAP;
512 msginfo->msgpool = MSGPOOL;
513 msginfo->msgtql = MSGTQL;
514 }
515 return (max_idx < 0) ? 0 : max_idx;
516}
517
518static int msgctl_stat(struct ipc_namespace *ns, int msqid,
519 int cmd, struct msqid64_ds *p)
520{
521 struct msg_queue *msq;
522 int err;
523
524 memset(p, 0, sizeof(*p));
525
526 rcu_read_lock();
527 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
528 msq = msq_obtain_object(ns, msqid);
529 if (IS_ERR(msq)) {
530 err = PTR_ERR(msq);
531 goto out_unlock;
532 }
533 } else { /* IPC_STAT */
534 msq = msq_obtain_object_check(ns, msqid);
535 if (IS_ERR(msq)) {
536 err = PTR_ERR(msq);
537 goto out_unlock;
538 }
539 }
540
541 /* see comment for SHM_STAT_ANY */
542 if (cmd == MSG_STAT_ANY)
543 audit_ipc_obj(&msq->q_perm);
544 else {
545 err = -EACCES;
546 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
547 goto out_unlock;
548 }
549
550 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
551 if (err)
552 goto out_unlock;
553
554 ipc_lock_object(&msq->q_perm);
555
556 if (!ipc_valid_object(&msq->q_perm)) {
557 ipc_unlock_object(&msq->q_perm);
558 err = -EIDRM;
559 goto out_unlock;
560 }
561
562 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
563 p->msg_stime = msq->q_stime;
564 p->msg_rtime = msq->q_rtime;
565 p->msg_ctime = msq->q_ctime;
566#ifndef CONFIG_64BIT
567 p->msg_stime_high = msq->q_stime >> 32;
568 p->msg_rtime_high = msq->q_rtime >> 32;
569 p->msg_ctime_high = msq->q_ctime >> 32;
570#endif
571 p->msg_cbytes = msq->q_cbytes;
572 p->msg_qnum = msq->q_qnum;
573 p->msg_qbytes = msq->q_qbytes;
574 p->msg_lspid = pid_vnr(msq->q_lspid);
575 p->msg_lrpid = pid_vnr(msq->q_lrpid);
576
577 if (cmd == IPC_STAT) {
578 /*
579 * As defined in SUS:
580 * Return 0 on success
581 */
582 err = 0;
583 } else {
584 /*
585 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
586 * Return the full id, including the sequence number
587 */
588 err = msq->q_perm.id;
589 }
590
591 ipc_unlock_object(&msq->q_perm);
592out_unlock:
593 rcu_read_unlock();
594 return err;
595}
596
597static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
598{
599 struct ipc_namespace *ns;
600 struct msqid64_ds msqid64;
601 int err;
602
603 if (msqid < 0 || cmd < 0)
604 return -EINVAL;
605
606 ns = current->nsproxy->ipc_ns;
607
608 switch (cmd) {
609 case IPC_INFO:
610 case MSG_INFO: {
611 struct msginfo msginfo;
612 err = msgctl_info(ns, msqid, cmd, &msginfo);
613 if (err < 0)
614 return err;
615 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
616 err = -EFAULT;
617 return err;
618 }
619 case MSG_STAT: /* msqid is an index rather than a msg queue id */
620 case MSG_STAT_ANY:
621 case IPC_STAT:
622 err = msgctl_stat(ns, msqid, cmd, &msqid64);
623 if (err < 0)
624 return err;
625 if (copy_msqid_to_user(buf, &msqid64, version))
626 err = -EFAULT;
627 return err;
628 case IPC_SET:
629 if (copy_msqid_from_user(&msqid64, buf, version))
630 return -EFAULT;
631 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
632 msqid64.msg_qbytes);
633 case IPC_RMID:
634 return msgctl_down(ns, msqid, cmd, NULL, 0);
635 default:
636 return -EINVAL;
637 }
638}
639
640SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
641{
642 return ksys_msgctl(msqid, cmd, buf, IPC_64);
643}
644
645#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
646long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
647{
648 int version = ipc_parse_version(&cmd);
649
650 return ksys_msgctl(msqid, cmd, buf, version);
651}
652
653SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
654{
655 return ksys_old_msgctl(msqid, cmd, buf);
656}
657#endif
658
659#ifdef CONFIG_COMPAT
660
661struct compat_msqid_ds {
662 struct compat_ipc_perm msg_perm;
663 compat_uptr_t msg_first;
664 compat_uptr_t msg_last;
665 old_time32_t msg_stime;
666 old_time32_t msg_rtime;
667 old_time32_t msg_ctime;
668 compat_ulong_t msg_lcbytes;
669 compat_ulong_t msg_lqbytes;
670 unsigned short msg_cbytes;
671 unsigned short msg_qnum;
672 unsigned short msg_qbytes;
673 compat_ipc_pid_t msg_lspid;
674 compat_ipc_pid_t msg_lrpid;
675};
676
677static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
678 int version)
679{
680 memset(out, 0, sizeof(*out));
681 if (version == IPC_64) {
682 struct compat_msqid64_ds __user *p = buf;
683 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
684 return -EFAULT;
685 if (get_user(out->msg_qbytes, &p->msg_qbytes))
686 return -EFAULT;
687 } else {
688 struct compat_msqid_ds __user *p = buf;
689 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
690 return -EFAULT;
691 if (get_user(out->msg_qbytes, &p->msg_qbytes))
692 return -EFAULT;
693 }
694 return 0;
695}
696
697static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
698 int version)
699{
700 if (version == IPC_64) {
701 struct compat_msqid64_ds v;
702 memset(&v, 0, sizeof(v));
703 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
704 v.msg_stime = lower_32_bits(in->msg_stime);
705 v.msg_stime_high = upper_32_bits(in->msg_stime);
706 v.msg_rtime = lower_32_bits(in->msg_rtime);
707 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
708 v.msg_ctime = lower_32_bits(in->msg_ctime);
709 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
710 v.msg_cbytes = in->msg_cbytes;
711 v.msg_qnum = in->msg_qnum;
712 v.msg_qbytes = in->msg_qbytes;
713 v.msg_lspid = in->msg_lspid;
714 v.msg_lrpid = in->msg_lrpid;
715 return copy_to_user(buf, &v, sizeof(v));
716 } else {
717 struct compat_msqid_ds v;
718 memset(&v, 0, sizeof(v));
719 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
720 v.msg_stime = in->msg_stime;
721 v.msg_rtime = in->msg_rtime;
722 v.msg_ctime = in->msg_ctime;
723 v.msg_cbytes = in->msg_cbytes;
724 v.msg_qnum = in->msg_qnum;
725 v.msg_qbytes = in->msg_qbytes;
726 v.msg_lspid = in->msg_lspid;
727 v.msg_lrpid = in->msg_lrpid;
728 return copy_to_user(buf, &v, sizeof(v));
729 }
730}
731
732static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
733{
734 struct ipc_namespace *ns;
735 int err;
736 struct msqid64_ds msqid64;
737
738 ns = current->nsproxy->ipc_ns;
739
740 if (msqid < 0 || cmd < 0)
741 return -EINVAL;
742
743 switch (cmd & (~IPC_64)) {
744 case IPC_INFO:
745 case MSG_INFO: {
746 struct msginfo msginfo;
747 err = msgctl_info(ns, msqid, cmd, &msginfo);
748 if (err < 0)
749 return err;
750 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
751 err = -EFAULT;
752 return err;
753 }
754 case IPC_STAT:
755 case MSG_STAT:
756 case MSG_STAT_ANY:
757 err = msgctl_stat(ns, msqid, cmd, &msqid64);
758 if (err < 0)
759 return err;
760 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
761 err = -EFAULT;
762 return err;
763 case IPC_SET:
764 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
765 return -EFAULT;
766 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
767 case IPC_RMID:
768 return msgctl_down(ns, msqid, cmd, NULL, 0);
769 default:
770 return -EINVAL;
771 }
772}
773
774COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
775{
776 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
777}
778
779#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
780long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
781{
782 int version = compat_ipc_parse_version(&cmd);
783
784 return compat_ksys_msgctl(msqid, cmd, uptr, version);
785}
786
787COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
788{
789 return compat_ksys_old_msgctl(msqid, cmd, uptr);
790}
791#endif
792#endif
793
794static int testmsg(struct msg_msg *msg, long type, int mode)
795{
796 switch (mode) {
797 case SEARCH_ANY:
798 case SEARCH_NUMBER:
799 return 1;
800 case SEARCH_LESSEQUAL:
801 if (msg->m_type <= type)
802 return 1;
803 break;
804 case SEARCH_EQUAL:
805 if (msg->m_type == type)
806 return 1;
807 break;
808 case SEARCH_NOTEQUAL:
809 if (msg->m_type != type)
810 return 1;
811 break;
812 }
813 return 0;
814}
815
816static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
817 struct wake_q_head *wake_q)
818{
819 struct msg_receiver *msr, *t;
820
821 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
822 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
823 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
824 msr->r_msgtype, msr->r_mode)) {
825
826 list_del(&msr->r_list);
827 if (msr->r_maxsize < msg->m_ts) {
828 wake_q_add(wake_q, msr->r_tsk);
829
830 /* See expunge_all regarding memory barrier */
831 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
832 } else {
833 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
834 msq->q_rtime = ktime_get_real_seconds();
835
836 wake_q_add(wake_q, msr->r_tsk);
837
838 /* See expunge_all regarding memory barrier */
839 smp_store_release(&msr->r_msg, msg);
840 return 1;
841 }
842 }
843 }
844
845 return 0;
846}
847
848static long do_msgsnd(int msqid, long mtype, void __user *mtext,
849 size_t msgsz, int msgflg)
850{
851 struct msg_queue *msq;
852 struct msg_msg *msg;
853 int err;
854 struct ipc_namespace *ns;
855 DEFINE_WAKE_Q(wake_q);
856
857 ns = current->nsproxy->ipc_ns;
858
859 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
860 return -EINVAL;
861 if (mtype < 1)
862 return -EINVAL;
863
864 msg = load_msg(mtext, msgsz);
865 if (IS_ERR(msg))
866 return PTR_ERR(msg);
867
868 msg->m_type = mtype;
869 msg->m_ts = msgsz;
870
871 rcu_read_lock();
872 msq = msq_obtain_object_check(ns, msqid);
873 if (IS_ERR(msq)) {
874 err = PTR_ERR(msq);
875 goto out_unlock1;
876 }
877
878 ipc_lock_object(&msq->q_perm);
879
880 for (;;) {
881 struct msg_sender s;
882
883 err = -EACCES;
884 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
885 goto out_unlock0;
886
887 /* raced with RMID? */
888 if (!ipc_valid_object(&msq->q_perm)) {
889 err = -EIDRM;
890 goto out_unlock0;
891 }
892
893 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
894 if (err)
895 goto out_unlock0;
896
897 if (msg_fits_inqueue(msq, msgsz))
898 break;
899
900 /* queue full, wait: */
901 if (msgflg & IPC_NOWAIT) {
902 err = -EAGAIN;
903 goto out_unlock0;
904 }
905
906 /* enqueue the sender and prepare to block */
907 ss_add(msq, &s, msgsz);
908
909 if (!ipc_rcu_getref(&msq->q_perm)) {
910 err = -EIDRM;
911 goto out_unlock0;
912 }
913
914 ipc_unlock_object(&msq->q_perm);
915 rcu_read_unlock();
916 schedule();
917
918 rcu_read_lock();
919 ipc_lock_object(&msq->q_perm);
920
921 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
922 /* raced with RMID? */
923 if (!ipc_valid_object(&msq->q_perm)) {
924 err = -EIDRM;
925 goto out_unlock0;
926 }
927 ss_del(&s);
928
929 if (signal_pending(current)) {
930 err = -ERESTARTNOHAND;
931 goto out_unlock0;
932 }
933
934 }
935
936 ipc_update_pid(&msq->q_lspid, task_tgid(current));
937 msq->q_stime = ktime_get_real_seconds();
938
939 if (!pipelined_send(msq, msg, &wake_q)) {
940 /* no one is waiting for this message, enqueue it */
941 list_add_tail(&msg->m_list, &msq->q_messages);
942 msq->q_cbytes += msgsz;
943 msq->q_qnum++;
944 percpu_counter_add_local(&ns->percpu_msg_bytes, msgsz);
945 percpu_counter_add_local(&ns->percpu_msg_hdrs, 1);
946 }
947
948 err = 0;
949 msg = NULL;
950
951out_unlock0:
952 ipc_unlock_object(&msq->q_perm);
953 wake_up_q(&wake_q);
954out_unlock1:
955 rcu_read_unlock();
956 if (msg != NULL)
957 free_msg(msg);
958 return err;
959}
960
961long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
962 int msgflg)
963{
964 long mtype;
965
966 if (get_user(mtype, &msgp->mtype))
967 return -EFAULT;
968 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
969}
970
971SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
972 int, msgflg)
973{
974 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
975}
976
977#ifdef CONFIG_COMPAT
978
979struct compat_msgbuf {
980 compat_long_t mtype;
981 char mtext[1];
982};
983
984long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
985 compat_ssize_t msgsz, int msgflg)
986{
987 struct compat_msgbuf __user *up = compat_ptr(msgp);
988 compat_long_t mtype;
989
990 if (get_user(mtype, &up->mtype))
991 return -EFAULT;
992 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
993}
994
995COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
996 compat_ssize_t, msgsz, int, msgflg)
997{
998 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
999}
1000#endif
1001
1002static inline int convert_mode(long *msgtyp, int msgflg)
1003{
1004 if (msgflg & MSG_COPY)
1005 return SEARCH_NUMBER;
1006 /*
1007 * find message of correct type.
1008 * msgtyp = 0 => get first.
1009 * msgtyp > 0 => get first message of matching type.
1010 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1011 */
1012 if (*msgtyp == 0)
1013 return SEARCH_ANY;
1014 if (*msgtyp < 0) {
1015 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1016 *msgtyp = LONG_MAX;
1017 else
1018 *msgtyp = -*msgtyp;
1019 return SEARCH_LESSEQUAL;
1020 }
1021 if (msgflg & MSG_EXCEPT)
1022 return SEARCH_NOTEQUAL;
1023 return SEARCH_EQUAL;
1024}
1025
1026static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1027{
1028 struct msgbuf __user *msgp = dest;
1029 size_t msgsz;
1030
1031 if (put_user(msg->m_type, &msgp->mtype))
1032 return -EFAULT;
1033
1034 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1035 if (store_msg(msgp->mtext, msg, msgsz))
1036 return -EFAULT;
1037 return msgsz;
1038}
1039
1040#ifdef CONFIG_CHECKPOINT_RESTORE
1041/*
1042 * This function creates new kernel message structure, large enough to store
1043 * bufsz message bytes.
1044 */
1045static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1046{
1047 struct msg_msg *copy;
1048
1049 /*
1050 * Create dummy message to copy real message to.
1051 */
1052 copy = load_msg(buf, bufsz);
1053 if (!IS_ERR(copy))
1054 copy->m_ts = bufsz;
1055 return copy;
1056}
1057
1058static inline void free_copy(struct msg_msg *copy)
1059{
1060 if (copy)
1061 free_msg(copy);
1062}
1063#else
1064static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1065{
1066 return ERR_PTR(-ENOSYS);
1067}
1068
1069static inline void free_copy(struct msg_msg *copy)
1070{
1071}
1072#endif
1073
1074static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1075{
1076 struct msg_msg *msg, *found = NULL;
1077 long count = 0;
1078
1079 list_for_each_entry(msg, &msq->q_messages, m_list) {
1080 if (testmsg(msg, *msgtyp, mode) &&
1081 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1082 *msgtyp, mode)) {
1083 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1084 *msgtyp = msg->m_type - 1;
1085 found = msg;
1086 } else if (mode == SEARCH_NUMBER) {
1087 if (*msgtyp == count)
1088 return msg;
1089 } else
1090 return msg;
1091 count++;
1092 }
1093 }
1094
1095 return found ?: ERR_PTR(-EAGAIN);
1096}
1097
1098static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1099 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1100{
1101 int mode;
1102 struct msg_queue *msq;
1103 struct ipc_namespace *ns;
1104 struct msg_msg *msg, *copy = NULL;
1105 DEFINE_WAKE_Q(wake_q);
1106
1107 ns = current->nsproxy->ipc_ns;
1108
1109 if (msqid < 0 || (long) bufsz < 0)
1110 return -EINVAL;
1111
1112 if (msgflg & MSG_COPY) {
1113 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1114 return -EINVAL;
1115 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1116 if (IS_ERR(copy))
1117 return PTR_ERR(copy);
1118 }
1119 mode = convert_mode(&msgtyp, msgflg);
1120
1121 rcu_read_lock();
1122 msq = msq_obtain_object_check(ns, msqid);
1123 if (IS_ERR(msq)) {
1124 rcu_read_unlock();
1125 free_copy(copy);
1126 return PTR_ERR(msq);
1127 }
1128
1129 for (;;) {
1130 struct msg_receiver msr_d;
1131
1132 msg = ERR_PTR(-EACCES);
1133 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1134 goto out_unlock1;
1135
1136 ipc_lock_object(&msq->q_perm);
1137
1138 /* raced with RMID? */
1139 if (!ipc_valid_object(&msq->q_perm)) {
1140 msg = ERR_PTR(-EIDRM);
1141 goto out_unlock0;
1142 }
1143
1144 msg = find_msg(msq, &msgtyp, mode);
1145 if (!IS_ERR(msg)) {
1146 /*
1147 * Found a suitable message.
1148 * Unlink it from the queue.
1149 */
1150 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1151 msg = ERR_PTR(-E2BIG);
1152 goto out_unlock0;
1153 }
1154 /*
1155 * If we are copying, then do not unlink message and do
1156 * not update queue parameters.
1157 */
1158 if (msgflg & MSG_COPY) {
1159 msg = copy_msg(msg, copy);
1160 goto out_unlock0;
1161 }
1162
1163 list_del(&msg->m_list);
1164 msq->q_qnum--;
1165 msq->q_rtime = ktime_get_real_seconds();
1166 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1167 msq->q_cbytes -= msg->m_ts;
1168 percpu_counter_sub_local(&ns->percpu_msg_bytes, msg->m_ts);
1169 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
1170 ss_wakeup(msq, &wake_q, false);
1171
1172 goto out_unlock0;
1173 }
1174
1175 /* No message waiting. Wait for a message */
1176 if (msgflg & IPC_NOWAIT) {
1177 msg = ERR_PTR(-ENOMSG);
1178 goto out_unlock0;
1179 }
1180
1181 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1182 msr_d.r_tsk = current;
1183 msr_d.r_msgtype = msgtyp;
1184 msr_d.r_mode = mode;
1185 if (msgflg & MSG_NOERROR)
1186 msr_d.r_maxsize = INT_MAX;
1187 else
1188 msr_d.r_maxsize = bufsz;
1189
1190 /* memory barrier not require due to ipc_lock_object() */
1191 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1192
1193 /* memory barrier not required, we own ipc_lock_object() */
1194 __set_current_state(TASK_INTERRUPTIBLE);
1195
1196 ipc_unlock_object(&msq->q_perm);
1197 rcu_read_unlock();
1198 schedule();
1199
1200 /*
1201 * Lockless receive, part 1:
1202 * We don't hold a reference to the queue and getting a
1203 * reference would defeat the idea of a lockless operation,
1204 * thus the code relies on rcu to guarantee the existence of
1205 * msq:
1206 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1207 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1208 */
1209 rcu_read_lock();
1210
1211 /*
1212 * Lockless receive, part 2:
1213 * The work in pipelined_send() and expunge_all():
1214 * - Set pointer to message
1215 * - Queue the receiver task for later wakeup
1216 * - Wake up the process after the lock is dropped.
1217 *
1218 * Should the process wake up before this wakeup (due to a
1219 * signal) it will either see the message and continue ...
1220 */
1221 msg = READ_ONCE(msr_d.r_msg);
1222 if (msg != ERR_PTR(-EAGAIN)) {
1223 /* see MSG_BARRIER for purpose/pairing */
1224 smp_acquire__after_ctrl_dep();
1225
1226 goto out_unlock1;
1227 }
1228
1229 /*
1230 * ... or see -EAGAIN, acquire the lock to check the message
1231 * again.
1232 */
1233 ipc_lock_object(&msq->q_perm);
1234
1235 msg = READ_ONCE(msr_d.r_msg);
1236 if (msg != ERR_PTR(-EAGAIN))
1237 goto out_unlock0;
1238
1239 list_del(&msr_d.r_list);
1240 if (signal_pending(current)) {
1241 msg = ERR_PTR(-ERESTARTNOHAND);
1242 goto out_unlock0;
1243 }
1244
1245 ipc_unlock_object(&msq->q_perm);
1246 }
1247
1248out_unlock0:
1249 ipc_unlock_object(&msq->q_perm);
1250 wake_up_q(&wake_q);
1251out_unlock1:
1252 rcu_read_unlock();
1253 if (IS_ERR(msg)) {
1254 free_copy(copy);
1255 return PTR_ERR(msg);
1256 }
1257
1258 bufsz = msg_handler(buf, msg, bufsz);
1259 free_msg(msg);
1260
1261 return bufsz;
1262}
1263
1264long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1265 long msgtyp, int msgflg)
1266{
1267 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1268}
1269
1270SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1271 long, msgtyp, int, msgflg)
1272{
1273 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1274}
1275
1276#ifdef CONFIG_COMPAT
1277static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1278{
1279 struct compat_msgbuf __user *msgp = dest;
1280 size_t msgsz;
1281
1282 if (put_user(msg->m_type, &msgp->mtype))
1283 return -EFAULT;
1284
1285 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1286 if (store_msg(msgp->mtext, msg, msgsz))
1287 return -EFAULT;
1288 return msgsz;
1289}
1290
1291long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1292 compat_long_t msgtyp, int msgflg)
1293{
1294 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1295 msgflg, compat_do_msg_fill);
1296}
1297
1298COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1299 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1300 int, msgflg)
1301{
1302 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1303}
1304#endif
1305
1306int msg_init_ns(struct ipc_namespace *ns)
1307{
1308 int ret;
1309
1310 ns->msg_ctlmax = MSGMAX;
1311 ns->msg_ctlmnb = MSGMNB;
1312 ns->msg_ctlmni = MSGMNI;
1313
1314 ret = percpu_counter_init(&ns->percpu_msg_bytes, 0, GFP_KERNEL);
1315 if (ret)
1316 goto fail_msg_bytes;
1317 ret = percpu_counter_init(&ns->percpu_msg_hdrs, 0, GFP_KERNEL);
1318 if (ret)
1319 goto fail_msg_hdrs;
1320 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1321 return 0;
1322
1323fail_msg_hdrs:
1324 percpu_counter_destroy(&ns->percpu_msg_bytes);
1325fail_msg_bytes:
1326 return ret;
1327}
1328
1329#ifdef CONFIG_IPC_NS
1330void msg_exit_ns(struct ipc_namespace *ns)
1331{
1332 free_ipcs(ns, &msg_ids(ns), freeque);
1333 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1334 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1335 percpu_counter_destroy(&ns->percpu_msg_bytes);
1336 percpu_counter_destroy(&ns->percpu_msg_hdrs);
1337}
1338#endif
1339
1340#ifdef CONFIG_PROC_FS
1341static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1342{
1343 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1344 struct user_namespace *user_ns = seq_user_ns(s);
1345 struct kern_ipc_perm *ipcp = it;
1346 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1347
1348 seq_printf(s,
1349 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1350 msq->q_perm.key,
1351 msq->q_perm.id,
1352 msq->q_perm.mode,
1353 msq->q_cbytes,
1354 msq->q_qnum,
1355 pid_nr_ns(msq->q_lspid, pid_ns),
1356 pid_nr_ns(msq->q_lrpid, pid_ns),
1357 from_kuid_munged(user_ns, msq->q_perm.uid),
1358 from_kgid_munged(user_ns, msq->q_perm.gid),
1359 from_kuid_munged(user_ns, msq->q_perm.cuid),
1360 from_kgid_munged(user_ns, msq->q_perm.cgid),
1361 msq->q_stime,
1362 msq->q_rtime,
1363 msq->q_ctime);
1364
1365 return 0;
1366}
1367#endif
1368
1369void __init msg_init(void)
1370{
1371 msg_init_ns(&init_ipc_ns);
1372
1373 ipc_init_proc_interface("sysvipc/msg",
1374 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1375 IPC_MSG_IDS, sysvipc_msg_proc_show);
1376}
1/*
2 * linux/ipc/msg.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 *
5 * Removed all the remaining kerneld mess
6 * Catch the -EFAULT stuff properly
7 * Use GFP_KERNEL for messages as in 1.2
8 * Fixed up the unchecked user space derefs
9 * Copyright (C) 1998 Alan Cox & Andi Kleen
10 *
11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
12 *
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
15 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 */
24
25#include <linux/capability.h>
26#include <linux/msg.h>
27#include <linux/spinlock.h>
28#include <linux/init.h>
29#include <linux/mm.h>
30#include <linux/proc_fs.h>
31#include <linux/list.h>
32#include <linux/security.h>
33#include <linux/sched.h>
34#include <linux/syscalls.h>
35#include <linux/audit.h>
36#include <linux/seq_file.h>
37#include <linux/rwsem.h>
38#include <linux/nsproxy.h>
39#include <linux/ipc_namespace.h>
40
41#include <asm/current.h>
42#include <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