hwtracing: stm: fix build error on some arches
[linux-stable.git] / ipc / msg.c
blobbce7ac1c809998762aab36fc84cf8c431184403c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/ipc/msg.c
4 * Copyright (C) 1992 Krishna Balasubramanian
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
12 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
14 * mostly rewritten, threaded and wake-one semantics added
15 * MSGMAX limit removed, sysctl's added
16 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
23 * Pavel Emelianov <xemul@openvz.org>
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>
42 #include <asm/current.h>
43 #include <linux/uaccess.h>
44 #include "util.h"
46 /* one msg_receiver structure for each sleeping receiver */
47 struct msg_receiver {
48 struct list_head r_list;
49 struct task_struct *r_tsk;
51 int r_mode;
52 long r_msgtype;
53 long r_maxsize;
55 struct msg_msg *r_msg;
58 /* one msg_sender for each sleeping sender */
59 struct msg_sender {
60 struct list_head list;
61 struct task_struct *tsk;
62 size_t msgsz;
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
71 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
73 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
75 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
77 if (IS_ERR(ipcp))
78 return ERR_CAST(ipcp);
80 return container_of(ipcp, struct msg_queue, q_perm);
83 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
84 int id)
86 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
88 if (IS_ERR(ipcp))
89 return ERR_CAST(ipcp);
91 return container_of(ipcp, struct msg_queue, q_perm);
94 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
96 ipc_rmid(&msg_ids(ns), &s->q_perm);
99 static void msg_rcu_free(struct rcu_head *head)
101 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
102 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
104 security_msg_queue_free(msq);
105 kvfree(msq);
109 * newque - Create a new msg queue
110 * @ns: namespace
111 * @params: ptr to the structure that contains the key and msgflg
113 * Called with msg_ids.rwsem held (writer)
115 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
117 struct msg_queue *msq;
118 int retval;
119 key_t key = params->key;
120 int msgflg = params->flg;
122 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
123 if (unlikely(!msq))
124 return -ENOMEM;
126 msq->q_perm.mode = msgflg & S_IRWXUGO;
127 msq->q_perm.key = key;
129 msq->q_perm.security = NULL;
130 retval = security_msg_queue_alloc(msq);
131 if (retval) {
132 kvfree(msq);
133 return retval;
136 msq->q_stime = msq->q_rtime = 0;
137 msq->q_ctime = ktime_get_real_seconds();
138 msq->q_cbytes = msq->q_qnum = 0;
139 msq->q_qbytes = ns->msg_ctlmnb;
140 msq->q_lspid = msq->q_lrpid = 0;
141 INIT_LIST_HEAD(&msq->q_messages);
142 INIT_LIST_HEAD(&msq->q_receivers);
143 INIT_LIST_HEAD(&msq->q_senders);
145 /* ipc_addid() locks msq upon success. */
146 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
147 if (retval < 0) {
148 call_rcu(&msq->q_perm.rcu, msg_rcu_free);
149 return retval;
152 ipc_unlock_object(&msq->q_perm);
153 rcu_read_unlock();
155 return msq->q_perm.id;
158 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
160 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
161 1 + msq->q_qnum <= msq->q_qbytes;
164 static inline void ss_add(struct msg_queue *msq,
165 struct msg_sender *mss, size_t msgsz)
167 mss->tsk = current;
168 mss->msgsz = msgsz;
169 __set_current_state(TASK_INTERRUPTIBLE);
170 list_add_tail(&mss->list, &msq->q_senders);
173 static inline void ss_del(struct msg_sender *mss)
175 if (mss->list.next)
176 list_del(&mss->list);
179 static void ss_wakeup(struct msg_queue *msq,
180 struct wake_q_head *wake_q, bool kill)
182 struct msg_sender *mss, *t;
183 struct task_struct *stop_tsk = NULL;
184 struct list_head *h = &msq->q_senders;
186 list_for_each_entry_safe(mss, t, h, list) {
187 if (kill)
188 mss->list.next = NULL;
191 * Stop at the first task we don't wakeup,
192 * we've already iterated the original
193 * sender queue.
195 else if (stop_tsk == mss->tsk)
196 break;
198 * We are not in an EIDRM scenario here, therefore
199 * verify that we really need to wakeup the task.
200 * To maintain current semantics and wakeup order,
201 * move the sender to the tail on behalf of the
202 * blocked task.
204 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
205 if (!stop_tsk)
206 stop_tsk = mss->tsk;
208 list_move_tail(&mss->list, &msq->q_senders);
209 continue;
212 wake_q_add(wake_q, mss->tsk);
216 static void expunge_all(struct msg_queue *msq, int res,
217 struct wake_q_head *wake_q)
219 struct msg_receiver *msr, *t;
221 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
222 wake_q_add(wake_q, msr->r_tsk);
223 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
228 * freeque() wakes up waiters on the sender and receiver waiting queue,
229 * removes the message queue from message queue ID IDR, and cleans up all the
230 * messages associated with this queue.
232 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
233 * before freeque() is called. msg_ids.rwsem remains locked on exit.
235 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
237 struct msg_msg *msg, *t;
238 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
239 DEFINE_WAKE_Q(wake_q);
241 expunge_all(msq, -EIDRM, &wake_q);
242 ss_wakeup(msq, &wake_q, true);
243 msg_rmid(ns, msq);
244 ipc_unlock_object(&msq->q_perm);
245 wake_up_q(&wake_q);
246 rcu_read_unlock();
248 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
249 atomic_dec(&ns->msg_hdrs);
250 free_msg(msg);
252 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
253 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
257 * Called with msg_ids.rwsem and ipcp locked.
259 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
261 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
263 return security_msg_queue_associate(msq, msgflg);
266 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
268 struct ipc_namespace *ns;
269 static const struct ipc_ops msg_ops = {
270 .getnew = newque,
271 .associate = msg_security,
273 struct ipc_params msg_params;
275 ns = current->nsproxy->ipc_ns;
277 msg_params.key = key;
278 msg_params.flg = msgflg;
280 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
283 static inline unsigned long
284 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
286 switch (version) {
287 case IPC_64:
288 return copy_to_user(buf, in, sizeof(*in));
289 case IPC_OLD:
291 struct msqid_ds out;
293 memset(&out, 0, sizeof(out));
295 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
297 out.msg_stime = in->msg_stime;
298 out.msg_rtime = in->msg_rtime;
299 out.msg_ctime = in->msg_ctime;
301 if (in->msg_cbytes > USHRT_MAX)
302 out.msg_cbytes = USHRT_MAX;
303 else
304 out.msg_cbytes = in->msg_cbytes;
305 out.msg_lcbytes = in->msg_cbytes;
307 if (in->msg_qnum > USHRT_MAX)
308 out.msg_qnum = USHRT_MAX;
309 else
310 out.msg_qnum = in->msg_qnum;
312 if (in->msg_qbytes > USHRT_MAX)
313 out.msg_qbytes = USHRT_MAX;
314 else
315 out.msg_qbytes = in->msg_qbytes;
316 out.msg_lqbytes = in->msg_qbytes;
318 out.msg_lspid = in->msg_lspid;
319 out.msg_lrpid = in->msg_lrpid;
321 return copy_to_user(buf, &out, sizeof(out));
323 default:
324 return -EINVAL;
328 static inline unsigned long
329 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
331 switch (version) {
332 case IPC_64:
333 if (copy_from_user(out, buf, sizeof(*out)))
334 return -EFAULT;
335 return 0;
336 case IPC_OLD:
338 struct msqid_ds tbuf_old;
340 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
341 return -EFAULT;
343 out->msg_perm.uid = tbuf_old.msg_perm.uid;
344 out->msg_perm.gid = tbuf_old.msg_perm.gid;
345 out->msg_perm.mode = tbuf_old.msg_perm.mode;
347 if (tbuf_old.msg_qbytes == 0)
348 out->msg_qbytes = tbuf_old.msg_lqbytes;
349 else
350 out->msg_qbytes = tbuf_old.msg_qbytes;
352 return 0;
354 default:
355 return -EINVAL;
360 * This function handles some msgctl commands which require the rwsem
361 * to be held in write mode.
362 * NOTE: no locks must be held, the rwsem is taken inside this function.
364 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
365 struct msqid64_ds *msqid64)
367 struct kern_ipc_perm *ipcp;
368 struct msg_queue *msq;
369 int err;
371 down_write(&msg_ids(ns).rwsem);
372 rcu_read_lock();
374 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
375 &msqid64->msg_perm, msqid64->msg_qbytes);
376 if (IS_ERR(ipcp)) {
377 err = PTR_ERR(ipcp);
378 goto out_unlock1;
381 msq = container_of(ipcp, struct msg_queue, q_perm);
383 err = security_msg_queue_msgctl(msq, cmd);
384 if (err)
385 goto out_unlock1;
387 switch (cmd) {
388 case IPC_RMID:
389 ipc_lock_object(&msq->q_perm);
390 /* freeque unlocks the ipc object and rcu */
391 freeque(ns, ipcp);
392 goto out_up;
393 case IPC_SET:
395 DEFINE_WAKE_Q(wake_q);
397 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
398 !capable(CAP_SYS_RESOURCE)) {
399 err = -EPERM;
400 goto out_unlock1;
403 ipc_lock_object(&msq->q_perm);
404 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
405 if (err)
406 goto out_unlock0;
408 msq->q_qbytes = msqid64->msg_qbytes;
410 msq->q_ctime = ktime_get_real_seconds();
412 * Sleeping receivers might be excluded by
413 * stricter permissions.
415 expunge_all(msq, -EAGAIN, &wake_q);
417 * Sleeping senders might be able to send
418 * due to a larger queue size.
420 ss_wakeup(msq, &wake_q, false);
421 ipc_unlock_object(&msq->q_perm);
422 wake_up_q(&wake_q);
424 goto out_unlock1;
426 default:
427 err = -EINVAL;
428 goto out_unlock1;
431 out_unlock0:
432 ipc_unlock_object(&msq->q_perm);
433 out_unlock1:
434 rcu_read_unlock();
435 out_up:
436 up_write(&msg_ids(ns).rwsem);
437 return err;
440 static int msgctl_info(struct ipc_namespace *ns, int msqid,
441 int cmd, struct msginfo *msginfo)
443 int err;
444 int max_id;
447 * We must not return kernel stack data.
448 * due to padding, it's not enough
449 * to set all member fields.
451 err = security_msg_queue_msgctl(NULL, cmd);
452 if (err)
453 return err;
455 memset(msginfo, 0, sizeof(*msginfo));
456 msginfo->msgmni = ns->msg_ctlmni;
457 msginfo->msgmax = ns->msg_ctlmax;
458 msginfo->msgmnb = ns->msg_ctlmnb;
459 msginfo->msgssz = MSGSSZ;
460 msginfo->msgseg = MSGSEG;
461 down_read(&msg_ids(ns).rwsem);
462 if (cmd == MSG_INFO) {
463 msginfo->msgpool = msg_ids(ns).in_use;
464 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
465 msginfo->msgtql = atomic_read(&ns->msg_bytes);
466 } else {
467 msginfo->msgmap = MSGMAP;
468 msginfo->msgpool = MSGPOOL;
469 msginfo->msgtql = MSGTQL;
471 max_id = ipc_get_maxid(&msg_ids(ns));
472 up_read(&msg_ids(ns).rwsem);
473 return (max_id < 0) ? 0 : max_id;
476 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
477 int cmd, struct msqid64_ds *p)
479 int err;
480 struct msg_queue *msq;
481 int success_return;
483 memset(p, 0, sizeof(*p));
485 rcu_read_lock();
486 if (cmd == MSG_STAT) {
487 msq = msq_obtain_object(ns, msqid);
488 if (IS_ERR(msq)) {
489 err = PTR_ERR(msq);
490 goto out_unlock;
492 success_return = msq->q_perm.id;
493 } else {
494 msq = msq_obtain_object_check(ns, msqid);
495 if (IS_ERR(msq)) {
496 err = PTR_ERR(msq);
497 goto out_unlock;
499 success_return = 0;
502 err = -EACCES;
503 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
504 goto out_unlock;
506 err = security_msg_queue_msgctl(msq, cmd);
507 if (err)
508 goto out_unlock;
510 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
511 p->msg_stime = msq->q_stime;
512 p->msg_rtime = msq->q_rtime;
513 p->msg_ctime = msq->q_ctime;
514 p->msg_cbytes = msq->q_cbytes;
515 p->msg_qnum = msq->q_qnum;
516 p->msg_qbytes = msq->q_qbytes;
517 p->msg_lspid = msq->q_lspid;
518 p->msg_lrpid = msq->q_lrpid;
519 rcu_read_unlock();
521 return success_return;
523 out_unlock:
524 rcu_read_unlock();
525 return err;
528 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
530 int version;
531 struct ipc_namespace *ns;
532 struct msqid64_ds msqid64;
533 int err;
535 if (msqid < 0 || cmd < 0)
536 return -EINVAL;
538 version = ipc_parse_version(&cmd);
539 ns = current->nsproxy->ipc_ns;
541 switch (cmd) {
542 case IPC_INFO:
543 case MSG_INFO: {
544 struct msginfo msginfo;
545 err = msgctl_info(ns, msqid, cmd, &msginfo);
546 if (err < 0)
547 return err;
548 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
549 err = -EFAULT;
550 return err;
552 case MSG_STAT: /* msqid is an index rather than a msg queue id */
553 case IPC_STAT:
554 err = msgctl_stat(ns, msqid, cmd, &msqid64);
555 if (err < 0)
556 return err;
557 if (copy_msqid_to_user(buf, &msqid64, version))
558 err = -EFAULT;
559 return err;
560 case IPC_SET:
561 if (copy_msqid_from_user(&msqid64, buf, version))
562 return -EFAULT;
563 /* fallthru */
564 case IPC_RMID:
565 return msgctl_down(ns, msqid, cmd, &msqid64);
566 default:
567 return -EINVAL;
571 #ifdef CONFIG_COMPAT
573 struct compat_msqid_ds {
574 struct compat_ipc_perm msg_perm;
575 compat_uptr_t msg_first;
576 compat_uptr_t msg_last;
577 compat_time_t msg_stime;
578 compat_time_t msg_rtime;
579 compat_time_t msg_ctime;
580 compat_ulong_t msg_lcbytes;
581 compat_ulong_t msg_lqbytes;
582 unsigned short msg_cbytes;
583 unsigned short msg_qnum;
584 unsigned short msg_qbytes;
585 compat_ipc_pid_t msg_lspid;
586 compat_ipc_pid_t msg_lrpid;
589 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
590 int version)
592 memset(out, 0, sizeof(*out));
593 if (version == IPC_64) {
594 struct compat_msqid64_ds *p = buf;
595 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
596 return -EFAULT;
597 if (get_user(out->msg_qbytes, &p->msg_qbytes))
598 return -EFAULT;
599 } else {
600 struct compat_msqid_ds *p = buf;
601 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
602 return -EFAULT;
603 if (get_user(out->msg_qbytes, &p->msg_qbytes))
604 return -EFAULT;
606 return 0;
609 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
610 int version)
612 if (version == IPC_64) {
613 struct compat_msqid64_ds v;
614 memset(&v, 0, sizeof(v));
615 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
616 v.msg_stime = in->msg_stime;
617 v.msg_rtime = in->msg_rtime;
618 v.msg_ctime = in->msg_ctime;
619 v.msg_cbytes = in->msg_cbytes;
620 v.msg_qnum = in->msg_qnum;
621 v.msg_qbytes = in->msg_qbytes;
622 v.msg_lspid = in->msg_lspid;
623 v.msg_lrpid = in->msg_lrpid;
624 return copy_to_user(buf, &v, sizeof(v));
625 } else {
626 struct compat_msqid_ds v;
627 memset(&v, 0, sizeof(v));
628 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
629 v.msg_stime = in->msg_stime;
630 v.msg_rtime = in->msg_rtime;
631 v.msg_ctime = in->msg_ctime;
632 v.msg_cbytes = in->msg_cbytes;
633 v.msg_qnum = in->msg_qnum;
634 v.msg_qbytes = in->msg_qbytes;
635 v.msg_lspid = in->msg_lspid;
636 v.msg_lrpid = in->msg_lrpid;
637 return copy_to_user(buf, &v, sizeof(v));
641 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
643 struct ipc_namespace *ns;
644 int err;
645 struct msqid64_ds msqid64;
646 int version = compat_ipc_parse_version(&cmd);
648 ns = current->nsproxy->ipc_ns;
650 if (msqid < 0 || cmd < 0)
651 return -EINVAL;
653 switch (cmd & (~IPC_64)) {
654 case IPC_INFO:
655 case MSG_INFO: {
656 struct msginfo msginfo;
657 err = msgctl_info(ns, msqid, cmd, &msginfo);
658 if (err < 0)
659 return err;
660 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
661 err = -EFAULT;
662 return err;
664 case IPC_STAT:
665 case MSG_STAT:
666 err = msgctl_stat(ns, msqid, cmd, &msqid64);
667 if (err < 0)
668 return err;
669 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
670 err = -EFAULT;
671 return err;
672 case IPC_SET:
673 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
674 return -EFAULT;
675 /* fallthru */
676 case IPC_RMID:
677 return msgctl_down(ns, msqid, cmd, &msqid64);
678 default:
679 return -EINVAL;
682 #endif
684 static int testmsg(struct msg_msg *msg, long type, int mode)
686 switch (mode) {
687 case SEARCH_ANY:
688 case SEARCH_NUMBER:
689 return 1;
690 case SEARCH_LESSEQUAL:
691 if (msg->m_type <= type)
692 return 1;
693 break;
694 case SEARCH_EQUAL:
695 if (msg->m_type == type)
696 return 1;
697 break;
698 case SEARCH_NOTEQUAL:
699 if (msg->m_type != type)
700 return 1;
701 break;
703 return 0;
706 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
707 struct wake_q_head *wake_q)
709 struct msg_receiver *msr, *t;
711 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
712 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
713 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
714 msr->r_msgtype, msr->r_mode)) {
716 list_del(&msr->r_list);
717 if (msr->r_maxsize < msg->m_ts) {
718 wake_q_add(wake_q, msr->r_tsk);
719 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
720 } else {
721 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
722 msq->q_rtime = get_seconds();
724 wake_q_add(wake_q, msr->r_tsk);
725 WRITE_ONCE(msr->r_msg, msg);
726 return 1;
731 return 0;
734 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
735 size_t msgsz, int msgflg)
737 struct msg_queue *msq;
738 struct msg_msg *msg;
739 int err;
740 struct ipc_namespace *ns;
741 DEFINE_WAKE_Q(wake_q);
743 ns = current->nsproxy->ipc_ns;
745 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
746 return -EINVAL;
747 if (mtype < 1)
748 return -EINVAL;
750 msg = load_msg(mtext, msgsz);
751 if (IS_ERR(msg))
752 return PTR_ERR(msg);
754 msg->m_type = mtype;
755 msg->m_ts = msgsz;
757 rcu_read_lock();
758 msq = msq_obtain_object_check(ns, msqid);
759 if (IS_ERR(msq)) {
760 err = PTR_ERR(msq);
761 goto out_unlock1;
764 ipc_lock_object(&msq->q_perm);
766 for (;;) {
767 struct msg_sender s;
769 err = -EACCES;
770 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
771 goto out_unlock0;
773 /* raced with RMID? */
774 if (!ipc_valid_object(&msq->q_perm)) {
775 err = -EIDRM;
776 goto out_unlock0;
779 err = security_msg_queue_msgsnd(msq, msg, msgflg);
780 if (err)
781 goto out_unlock0;
783 if (msg_fits_inqueue(msq, msgsz))
784 break;
786 /* queue full, wait: */
787 if (msgflg & IPC_NOWAIT) {
788 err = -EAGAIN;
789 goto out_unlock0;
792 /* enqueue the sender and prepare to block */
793 ss_add(msq, &s, msgsz);
795 if (!ipc_rcu_getref(&msq->q_perm)) {
796 err = -EIDRM;
797 goto out_unlock0;
800 ipc_unlock_object(&msq->q_perm);
801 rcu_read_unlock();
802 schedule();
804 rcu_read_lock();
805 ipc_lock_object(&msq->q_perm);
807 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
808 /* raced with RMID? */
809 if (!ipc_valid_object(&msq->q_perm)) {
810 err = -EIDRM;
811 goto out_unlock0;
813 ss_del(&s);
815 if (signal_pending(current)) {
816 err = -ERESTARTNOHAND;
817 goto out_unlock0;
822 msq->q_lspid = task_tgid_vnr(current);
823 msq->q_stime = get_seconds();
825 if (!pipelined_send(msq, msg, &wake_q)) {
826 /* no one is waiting for this message, enqueue it */
827 list_add_tail(&msg->m_list, &msq->q_messages);
828 msq->q_cbytes += msgsz;
829 msq->q_qnum++;
830 atomic_add(msgsz, &ns->msg_bytes);
831 atomic_inc(&ns->msg_hdrs);
834 err = 0;
835 msg = NULL;
837 out_unlock0:
838 ipc_unlock_object(&msq->q_perm);
839 wake_up_q(&wake_q);
840 out_unlock1:
841 rcu_read_unlock();
842 if (msg != NULL)
843 free_msg(msg);
844 return err;
847 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
848 int, msgflg)
850 long mtype;
852 if (get_user(mtype, &msgp->mtype))
853 return -EFAULT;
854 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
857 #ifdef CONFIG_COMPAT
859 struct compat_msgbuf {
860 compat_long_t mtype;
861 char mtext[1];
864 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
865 compat_ssize_t, msgsz, int, msgflg)
867 struct compat_msgbuf __user *up = compat_ptr(msgp);
868 compat_long_t mtype;
870 if (get_user(mtype, &up->mtype))
871 return -EFAULT;
872 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
874 #endif
876 static inline int convert_mode(long *msgtyp, int msgflg)
878 if (msgflg & MSG_COPY)
879 return SEARCH_NUMBER;
881 * find message of correct type.
882 * msgtyp = 0 => get first.
883 * msgtyp > 0 => get first message of matching type.
884 * msgtyp < 0 => get message with least type must be < abs(msgtype).
886 if (*msgtyp == 0)
887 return SEARCH_ANY;
888 if (*msgtyp < 0) {
889 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
890 *msgtyp = LONG_MAX;
891 else
892 *msgtyp = -*msgtyp;
893 return SEARCH_LESSEQUAL;
895 if (msgflg & MSG_EXCEPT)
896 return SEARCH_NOTEQUAL;
897 return SEARCH_EQUAL;
900 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
902 struct msgbuf __user *msgp = dest;
903 size_t msgsz;
905 if (put_user(msg->m_type, &msgp->mtype))
906 return -EFAULT;
908 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
909 if (store_msg(msgp->mtext, msg, msgsz))
910 return -EFAULT;
911 return msgsz;
914 #ifdef CONFIG_CHECKPOINT_RESTORE
916 * This function creates new kernel message structure, large enough to store
917 * bufsz message bytes.
919 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
921 struct msg_msg *copy;
924 * Create dummy message to copy real message to.
926 copy = load_msg(buf, bufsz);
927 if (!IS_ERR(copy))
928 copy->m_ts = bufsz;
929 return copy;
932 static inline void free_copy(struct msg_msg *copy)
934 if (copy)
935 free_msg(copy);
937 #else
938 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
940 return ERR_PTR(-ENOSYS);
943 static inline void free_copy(struct msg_msg *copy)
946 #endif
948 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
950 struct msg_msg *msg, *found = NULL;
951 long count = 0;
953 list_for_each_entry(msg, &msq->q_messages, m_list) {
954 if (testmsg(msg, *msgtyp, mode) &&
955 !security_msg_queue_msgrcv(msq, msg, current,
956 *msgtyp, mode)) {
957 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
958 *msgtyp = msg->m_type - 1;
959 found = msg;
960 } else if (mode == SEARCH_NUMBER) {
961 if (*msgtyp == count)
962 return msg;
963 } else
964 return msg;
965 count++;
969 return found ?: ERR_PTR(-EAGAIN);
972 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
973 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
975 int mode;
976 struct msg_queue *msq;
977 struct ipc_namespace *ns;
978 struct msg_msg *msg, *copy = NULL;
979 DEFINE_WAKE_Q(wake_q);
981 ns = current->nsproxy->ipc_ns;
983 if (msqid < 0 || (long) bufsz < 0)
984 return -EINVAL;
986 if (msgflg & MSG_COPY) {
987 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
988 return -EINVAL;
989 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
990 if (IS_ERR(copy))
991 return PTR_ERR(copy);
993 mode = convert_mode(&msgtyp, msgflg);
995 rcu_read_lock();
996 msq = msq_obtain_object_check(ns, msqid);
997 if (IS_ERR(msq)) {
998 rcu_read_unlock();
999 free_copy(copy);
1000 return PTR_ERR(msq);
1003 for (;;) {
1004 struct msg_receiver msr_d;
1006 msg = ERR_PTR(-EACCES);
1007 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1008 goto out_unlock1;
1010 ipc_lock_object(&msq->q_perm);
1012 /* raced with RMID? */
1013 if (!ipc_valid_object(&msq->q_perm)) {
1014 msg = ERR_PTR(-EIDRM);
1015 goto out_unlock0;
1018 msg = find_msg(msq, &msgtyp, mode);
1019 if (!IS_ERR(msg)) {
1021 * Found a suitable message.
1022 * Unlink it from the queue.
1024 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1025 msg = ERR_PTR(-E2BIG);
1026 goto out_unlock0;
1029 * If we are copying, then do not unlink message and do
1030 * not update queue parameters.
1032 if (msgflg & MSG_COPY) {
1033 msg = copy_msg(msg, copy);
1034 goto out_unlock0;
1037 list_del(&msg->m_list);
1038 msq->q_qnum--;
1039 msq->q_rtime = get_seconds();
1040 msq->q_lrpid = task_tgid_vnr(current);
1041 msq->q_cbytes -= msg->m_ts;
1042 atomic_sub(msg->m_ts, &ns->msg_bytes);
1043 atomic_dec(&ns->msg_hdrs);
1044 ss_wakeup(msq, &wake_q, false);
1046 goto out_unlock0;
1049 /* No message waiting. Wait for a message */
1050 if (msgflg & IPC_NOWAIT) {
1051 msg = ERR_PTR(-ENOMSG);
1052 goto out_unlock0;
1055 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1056 msr_d.r_tsk = current;
1057 msr_d.r_msgtype = msgtyp;
1058 msr_d.r_mode = mode;
1059 if (msgflg & MSG_NOERROR)
1060 msr_d.r_maxsize = INT_MAX;
1061 else
1062 msr_d.r_maxsize = bufsz;
1063 msr_d.r_msg = ERR_PTR(-EAGAIN);
1064 __set_current_state(TASK_INTERRUPTIBLE);
1066 ipc_unlock_object(&msq->q_perm);
1067 rcu_read_unlock();
1068 schedule();
1071 * Lockless receive, part 1:
1072 * We don't hold a reference to the queue and getting a
1073 * reference would defeat the idea of a lockless operation,
1074 * thus the code relies on rcu to guarantee the existence of
1075 * msq:
1076 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1077 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1079 rcu_read_lock();
1082 * Lockless receive, part 2:
1083 * The work in pipelined_send() and expunge_all():
1084 * - Set pointer to message
1085 * - Queue the receiver task for later wakeup
1086 * - Wake up the process after the lock is dropped.
1088 * Should the process wake up before this wakeup (due to a
1089 * signal) it will either see the message and continue ...
1091 msg = READ_ONCE(msr_d.r_msg);
1092 if (msg != ERR_PTR(-EAGAIN))
1093 goto out_unlock1;
1096 * ... or see -EAGAIN, acquire the lock to check the message
1097 * again.
1099 ipc_lock_object(&msq->q_perm);
1101 msg = msr_d.r_msg;
1102 if (msg != ERR_PTR(-EAGAIN))
1103 goto out_unlock0;
1105 list_del(&msr_d.r_list);
1106 if (signal_pending(current)) {
1107 msg = ERR_PTR(-ERESTARTNOHAND);
1108 goto out_unlock0;
1111 ipc_unlock_object(&msq->q_perm);
1114 out_unlock0:
1115 ipc_unlock_object(&msq->q_perm);
1116 wake_up_q(&wake_q);
1117 out_unlock1:
1118 rcu_read_unlock();
1119 if (IS_ERR(msg)) {
1120 free_copy(copy);
1121 return PTR_ERR(msg);
1124 bufsz = msg_handler(buf, msg, bufsz);
1125 free_msg(msg);
1127 return bufsz;
1130 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1131 long, msgtyp, int, msgflg)
1133 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1136 #ifdef CONFIG_COMPAT
1137 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1139 struct compat_msgbuf __user *msgp = dest;
1140 size_t msgsz;
1142 if (put_user(msg->m_type, &msgp->mtype))
1143 return -EFAULT;
1145 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1146 if (store_msg(msgp->mtext, msg, msgsz))
1147 return -EFAULT;
1148 return msgsz;
1151 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1152 compat_ssize_t, msgsz, compat_long_t, msgtyp, int, msgflg)
1154 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1155 msgflg, compat_do_msg_fill);
1157 #endif
1159 int msg_init_ns(struct ipc_namespace *ns)
1161 ns->msg_ctlmax = MSGMAX;
1162 ns->msg_ctlmnb = MSGMNB;
1163 ns->msg_ctlmni = MSGMNI;
1165 atomic_set(&ns->msg_bytes, 0);
1166 atomic_set(&ns->msg_hdrs, 0);
1167 return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1170 #ifdef CONFIG_IPC_NS
1171 void msg_exit_ns(struct ipc_namespace *ns)
1173 free_ipcs(ns, &msg_ids(ns), freeque);
1174 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1175 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1177 #endif
1179 #ifdef CONFIG_PROC_FS
1180 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1182 struct user_namespace *user_ns = seq_user_ns(s);
1183 struct kern_ipc_perm *ipcp = it;
1184 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1186 seq_printf(s,
1187 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1188 msq->q_perm.key,
1189 msq->q_perm.id,
1190 msq->q_perm.mode,
1191 msq->q_cbytes,
1192 msq->q_qnum,
1193 msq->q_lspid,
1194 msq->q_lrpid,
1195 from_kuid_munged(user_ns, msq->q_perm.uid),
1196 from_kgid_munged(user_ns, msq->q_perm.gid),
1197 from_kuid_munged(user_ns, msq->q_perm.cuid),
1198 from_kgid_munged(user_ns, msq->q_perm.cgid),
1199 msq->q_stime,
1200 msq->q_rtime,
1201 msq->q_ctime);
1203 return 0;
1205 #endif
1207 int __init msg_init(void)
1209 const int err = msg_init_ns(&init_ipc_ns);
1211 ipc_init_proc_interface("sysvipc/msg",
1212 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1213 IPC_MSG_IDS, sysvipc_msg_proc_show);
1214 return err;