1 // SPDX-License-Identifier: GPL-2.0
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>
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>
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>
46 /* one msq_queue structure for each present queue on the system */
48 struct kern_ipc_perm q_perm;
49 time64_t q_stime; /* last msgsnd time */
50 time64_t q_rtime; /* last msgrcv time */
51 time64_t q_ctime; /* last change time */
52 unsigned long q_cbytes; /* current number of bytes on queue */
53 unsigned long q_qnum; /* number of messages in queue */
54 unsigned long q_qbytes; /* max number of bytes on queue */
55 struct pid *q_lspid; /* pid of last msgsnd */
56 struct pid *q_lrpid; /* last receive pid */
58 struct list_head q_messages;
59 struct list_head q_receivers;
60 struct list_head q_senders;
63 /* one msg_receiver structure for each sleeping receiver */
65 struct list_head r_list;
66 struct task_struct *r_tsk;
72 struct msg_msg *r_msg;
75 /* one msg_sender for each sleeping sender */
77 struct list_head list;
78 struct task_struct *tsk;
83 #define SEARCH_EQUAL 2
84 #define SEARCH_NOTEQUAL 3
85 #define SEARCH_LESSEQUAL 4
86 #define SEARCH_NUMBER 5
88 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
90 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
92 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
95 return ERR_CAST(ipcp);
97 return container_of(ipcp, struct msg_queue, q_perm);
100 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
103 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
106 return ERR_CAST(ipcp);
108 return container_of(ipcp, struct msg_queue, q_perm);
111 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
113 ipc_rmid(&msg_ids(ns), &s->q_perm);
116 static void msg_rcu_free(struct rcu_head *head)
118 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
119 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
121 security_msg_queue_free(&msq->q_perm);
126 * newque - Create a new msg queue
128 * @params: ptr to the structure that contains the key and msgflg
130 * Called with msg_ids.rwsem held (writer)
132 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
134 struct msg_queue *msq;
136 key_t key = params->key;
137 int msgflg = params->flg;
139 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
143 msq->q_perm.mode = msgflg & S_IRWXUGO;
144 msq->q_perm.key = key;
146 msq->q_perm.security = NULL;
147 retval = security_msg_queue_alloc(&msq->q_perm);
153 msq->q_stime = msq->q_rtime = 0;
154 msq->q_ctime = ktime_get_real_seconds();
155 msq->q_cbytes = msq->q_qnum = 0;
156 msq->q_qbytes = ns->msg_ctlmnb;
157 msq->q_lspid = msq->q_lrpid = NULL;
158 INIT_LIST_HEAD(&msq->q_messages);
159 INIT_LIST_HEAD(&msq->q_receivers);
160 INIT_LIST_HEAD(&msq->q_senders);
162 /* ipc_addid() locks msq upon success. */
163 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
165 call_rcu(&msq->q_perm.rcu, msg_rcu_free);
169 ipc_unlock_object(&msq->q_perm);
172 return msq->q_perm.id;
175 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
177 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
178 1 + msq->q_qnum <= msq->q_qbytes;
181 static inline void ss_add(struct msg_queue *msq,
182 struct msg_sender *mss, size_t msgsz)
186 __set_current_state(TASK_INTERRUPTIBLE);
187 list_add_tail(&mss->list, &msq->q_senders);
190 static inline void ss_del(struct msg_sender *mss)
193 list_del(&mss->list);
196 static void ss_wakeup(struct msg_queue *msq,
197 struct wake_q_head *wake_q, bool kill)
199 struct msg_sender *mss, *t;
200 struct task_struct *stop_tsk = NULL;
201 struct list_head *h = &msq->q_senders;
203 list_for_each_entry_safe(mss, t, h, list) {
205 mss->list.next = NULL;
208 * Stop at the first task we don't wakeup,
209 * we've already iterated the original
212 else if (stop_tsk == mss->tsk)
215 * We are not in an EIDRM scenario here, therefore
216 * verify that we really need to wakeup the task.
217 * To maintain current semantics and wakeup order,
218 * move the sender to the tail on behalf of the
221 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
225 list_move_tail(&mss->list, &msq->q_senders);
229 wake_q_add(wake_q, mss->tsk);
233 static void expunge_all(struct msg_queue *msq, int res,
234 struct wake_q_head *wake_q)
236 struct msg_receiver *msr, *t;
238 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
239 wake_q_add(wake_q, msr->r_tsk);
240 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
245 * freeque() wakes up waiters on the sender and receiver waiting queue,
246 * removes the message queue from message queue ID IDR, and cleans up all the
247 * messages associated with this queue.
249 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
250 * before freeque() is called. msg_ids.rwsem remains locked on exit.
252 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
254 struct msg_msg *msg, *t;
255 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
256 DEFINE_WAKE_Q(wake_q);
258 expunge_all(msq, -EIDRM, &wake_q);
259 ss_wakeup(msq, &wake_q, true);
261 ipc_unlock_object(&msq->q_perm);
265 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
266 atomic_dec(&ns->msg_hdrs);
269 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
270 ipc_update_pid(&msq->q_lspid, NULL);
271 ipc_update_pid(&msq->q_lrpid, NULL);
272 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
275 long ksys_msgget(key_t key, int msgflg)
277 struct ipc_namespace *ns;
278 static const struct ipc_ops msg_ops = {
280 .associate = security_msg_queue_associate,
282 struct ipc_params msg_params;
284 ns = current->nsproxy->ipc_ns;
286 msg_params.key = key;
287 msg_params.flg = msgflg;
289 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
292 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
294 return ksys_msgget(key, msgflg);
297 static inline unsigned long
298 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
302 return copy_to_user(buf, in, sizeof(*in));
307 memset(&out, 0, sizeof(out));
309 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
311 out.msg_stime = in->msg_stime;
312 out.msg_rtime = in->msg_rtime;
313 out.msg_ctime = in->msg_ctime;
315 if (in->msg_cbytes > USHRT_MAX)
316 out.msg_cbytes = USHRT_MAX;
318 out.msg_cbytes = in->msg_cbytes;
319 out.msg_lcbytes = in->msg_cbytes;
321 if (in->msg_qnum > USHRT_MAX)
322 out.msg_qnum = USHRT_MAX;
324 out.msg_qnum = in->msg_qnum;
326 if (in->msg_qbytes > USHRT_MAX)
327 out.msg_qbytes = USHRT_MAX;
329 out.msg_qbytes = in->msg_qbytes;
330 out.msg_lqbytes = in->msg_qbytes;
332 out.msg_lspid = in->msg_lspid;
333 out.msg_lrpid = in->msg_lrpid;
335 return copy_to_user(buf, &out, sizeof(out));
342 static inline unsigned long
343 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
347 if (copy_from_user(out, buf, sizeof(*out)))
352 struct msqid_ds tbuf_old;
354 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
357 out->msg_perm.uid = tbuf_old.msg_perm.uid;
358 out->msg_perm.gid = tbuf_old.msg_perm.gid;
359 out->msg_perm.mode = tbuf_old.msg_perm.mode;
361 if (tbuf_old.msg_qbytes == 0)
362 out->msg_qbytes = tbuf_old.msg_lqbytes;
364 out->msg_qbytes = tbuf_old.msg_qbytes;
374 * This function handles some msgctl commands which require the rwsem
375 * to be held in write mode.
376 * NOTE: no locks must be held, the rwsem is taken inside this function.
378 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
379 struct msqid64_ds *msqid64)
381 struct kern_ipc_perm *ipcp;
382 struct msg_queue *msq;
385 down_write(&msg_ids(ns).rwsem);
388 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
389 &msqid64->msg_perm, msqid64->msg_qbytes);
395 msq = container_of(ipcp, struct msg_queue, q_perm);
397 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
403 ipc_lock_object(&msq->q_perm);
404 /* freeque unlocks the ipc object and rcu */
409 DEFINE_WAKE_Q(wake_q);
411 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
412 !capable(CAP_SYS_RESOURCE)) {
417 ipc_lock_object(&msq->q_perm);
418 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
422 msq->q_qbytes = msqid64->msg_qbytes;
424 msq->q_ctime = ktime_get_real_seconds();
426 * Sleeping receivers might be excluded by
427 * stricter permissions.
429 expunge_all(msq, -EAGAIN, &wake_q);
431 * Sleeping senders might be able to send
432 * due to a larger queue size.
434 ss_wakeup(msq, &wake_q, false);
435 ipc_unlock_object(&msq->q_perm);
446 ipc_unlock_object(&msq->q_perm);
450 up_write(&msg_ids(ns).rwsem);
454 static int msgctl_info(struct ipc_namespace *ns, int msqid,
455 int cmd, struct msginfo *msginfo)
461 * We must not return kernel stack data.
462 * due to padding, it's not enough
463 * to set all member fields.
465 err = security_msg_queue_msgctl(NULL, cmd);
469 memset(msginfo, 0, sizeof(*msginfo));
470 msginfo->msgmni = ns->msg_ctlmni;
471 msginfo->msgmax = ns->msg_ctlmax;
472 msginfo->msgmnb = ns->msg_ctlmnb;
473 msginfo->msgssz = MSGSSZ;
474 msginfo->msgseg = MSGSEG;
475 down_read(&msg_ids(ns).rwsem);
476 if (cmd == MSG_INFO) {
477 msginfo->msgpool = msg_ids(ns).in_use;
478 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
479 msginfo->msgtql = atomic_read(&ns->msg_bytes);
481 msginfo->msgmap = MSGMAP;
482 msginfo->msgpool = MSGPOOL;
483 msginfo->msgtql = MSGTQL;
485 max_id = ipc_get_maxid(&msg_ids(ns));
486 up_read(&msg_ids(ns).rwsem);
487 return (max_id < 0) ? 0 : max_id;
490 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
491 int cmd, struct msqid64_ds *p)
493 struct msg_queue *msq;
497 memset(p, 0, sizeof(*p));
500 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
501 msq = msq_obtain_object(ns, msqid);
507 } else { /* IPC_STAT */
508 msq = msq_obtain_object_check(ns, msqid);
515 /* see comment for SHM_STAT_ANY */
516 if (cmd == MSG_STAT_ANY)
517 audit_ipc_obj(&msq->q_perm);
520 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
524 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
528 ipc_lock_object(&msq->q_perm);
530 if (!ipc_valid_object(&msq->q_perm)) {
531 ipc_unlock_object(&msq->q_perm);
536 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
537 p->msg_stime = msq->q_stime;
538 p->msg_rtime = msq->q_rtime;
539 p->msg_ctime = msq->q_ctime;
541 p->msg_stime_high = msq->q_stime >> 32;
542 p->msg_rtime_high = msq->q_rtime >> 32;
543 p->msg_ctime_high = msq->q_ctime >> 32;
545 p->msg_cbytes = msq->q_cbytes;
546 p->msg_qnum = msq->q_qnum;
547 p->msg_qbytes = msq->q_qbytes;
548 p->msg_lspid = pid_vnr(msq->q_lspid);
549 p->msg_lrpid = pid_vnr(msq->q_lrpid);
551 ipc_unlock_object(&msq->q_perm);
560 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
563 struct ipc_namespace *ns;
564 struct msqid64_ds msqid64;
567 if (msqid < 0 || cmd < 0)
570 version = ipc_parse_version(&cmd);
571 ns = current->nsproxy->ipc_ns;
576 struct msginfo msginfo;
577 err = msgctl_info(ns, msqid, cmd, &msginfo);
580 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
584 case MSG_STAT: /* msqid is an index rather than a msg queue id */
587 err = msgctl_stat(ns, msqid, cmd, &msqid64);
590 if (copy_msqid_to_user(buf, &msqid64, version))
594 if (copy_msqid_from_user(&msqid64, buf, version))
598 return msgctl_down(ns, msqid, cmd, &msqid64);
604 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
606 return ksys_msgctl(msqid, cmd, buf);
611 struct compat_msqid_ds {
612 struct compat_ipc_perm msg_perm;
613 compat_uptr_t msg_first;
614 compat_uptr_t msg_last;
615 compat_time_t msg_stime;
616 compat_time_t msg_rtime;
617 compat_time_t msg_ctime;
618 compat_ulong_t msg_lcbytes;
619 compat_ulong_t msg_lqbytes;
620 unsigned short msg_cbytes;
621 unsigned short msg_qnum;
622 unsigned short msg_qbytes;
623 compat_ipc_pid_t msg_lspid;
624 compat_ipc_pid_t msg_lrpid;
627 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
630 memset(out, 0, sizeof(*out));
631 if (version == IPC_64) {
632 struct compat_msqid64_ds __user *p = buf;
633 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
635 if (get_user(out->msg_qbytes, &p->msg_qbytes))
638 struct compat_msqid_ds __user *p = buf;
639 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
641 if (get_user(out->msg_qbytes, &p->msg_qbytes))
647 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
650 if (version == IPC_64) {
651 struct compat_msqid64_ds v;
652 memset(&v, 0, sizeof(v));
653 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
654 v.msg_stime = lower_32_bits(in->msg_stime);
655 v.msg_stime_high = upper_32_bits(in->msg_stime);
656 v.msg_rtime = lower_32_bits(in->msg_rtime);
657 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
658 v.msg_ctime = lower_32_bits(in->msg_ctime);
659 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
660 v.msg_cbytes = in->msg_cbytes;
661 v.msg_qnum = in->msg_qnum;
662 v.msg_qbytes = in->msg_qbytes;
663 v.msg_lspid = in->msg_lspid;
664 v.msg_lrpid = in->msg_lrpid;
665 return copy_to_user(buf, &v, sizeof(v));
667 struct compat_msqid_ds v;
668 memset(&v, 0, sizeof(v));
669 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
670 v.msg_stime = in->msg_stime;
671 v.msg_rtime = in->msg_rtime;
672 v.msg_ctime = in->msg_ctime;
673 v.msg_cbytes = in->msg_cbytes;
674 v.msg_qnum = in->msg_qnum;
675 v.msg_qbytes = in->msg_qbytes;
676 v.msg_lspid = in->msg_lspid;
677 v.msg_lrpid = in->msg_lrpid;
678 return copy_to_user(buf, &v, sizeof(v));
682 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
684 struct ipc_namespace *ns;
686 struct msqid64_ds msqid64;
687 int version = compat_ipc_parse_version(&cmd);
689 ns = current->nsproxy->ipc_ns;
691 if (msqid < 0 || cmd < 0)
694 switch (cmd & (~IPC_64)) {
697 struct msginfo msginfo;
698 err = msgctl_info(ns, msqid, cmd, &msginfo);
701 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
708 err = msgctl_stat(ns, msqid, cmd, &msqid64);
711 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
715 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
719 return msgctl_down(ns, msqid, cmd, &msqid64);
725 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
727 return compat_ksys_msgctl(msqid, cmd, uptr);
731 static int testmsg(struct msg_msg *msg, long type, int mode)
737 case SEARCH_LESSEQUAL:
738 if (msg->m_type <= type)
742 if (msg->m_type == type)
745 case SEARCH_NOTEQUAL:
746 if (msg->m_type != type)
753 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
754 struct wake_q_head *wake_q)
756 struct msg_receiver *msr, *t;
758 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
759 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
760 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
761 msr->r_msgtype, msr->r_mode)) {
763 list_del(&msr->r_list);
764 if (msr->r_maxsize < msg->m_ts) {
765 wake_q_add(wake_q, msr->r_tsk);
766 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
768 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
769 msq->q_rtime = ktime_get_real_seconds();
771 wake_q_add(wake_q, msr->r_tsk);
772 WRITE_ONCE(msr->r_msg, msg);
781 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
782 size_t msgsz, int msgflg)
784 struct msg_queue *msq;
787 struct ipc_namespace *ns;
788 DEFINE_WAKE_Q(wake_q);
790 ns = current->nsproxy->ipc_ns;
792 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
797 msg = load_msg(mtext, msgsz);
805 msq = msq_obtain_object_check(ns, msqid);
811 ipc_lock_object(&msq->q_perm);
817 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
820 /* raced with RMID? */
821 if (!ipc_valid_object(&msq->q_perm)) {
826 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
830 if (msg_fits_inqueue(msq, msgsz))
833 /* queue full, wait: */
834 if (msgflg & IPC_NOWAIT) {
839 /* enqueue the sender and prepare to block */
840 ss_add(msq, &s, msgsz);
842 if (!ipc_rcu_getref(&msq->q_perm)) {
847 ipc_unlock_object(&msq->q_perm);
852 ipc_lock_object(&msq->q_perm);
854 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
855 /* raced with RMID? */
856 if (!ipc_valid_object(&msq->q_perm)) {
862 if (signal_pending(current)) {
863 err = -ERESTARTNOHAND;
869 ipc_update_pid(&msq->q_lspid, task_tgid(current));
870 msq->q_stime = ktime_get_real_seconds();
872 if (!pipelined_send(msq, msg, &wake_q)) {
873 /* no one is waiting for this message, enqueue it */
874 list_add_tail(&msg->m_list, &msq->q_messages);
875 msq->q_cbytes += msgsz;
877 atomic_add(msgsz, &ns->msg_bytes);
878 atomic_inc(&ns->msg_hdrs);
885 ipc_unlock_object(&msq->q_perm);
894 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
899 if (get_user(mtype, &msgp->mtype))
901 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
904 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
907 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
912 struct compat_msgbuf {
917 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
918 compat_ssize_t msgsz, int msgflg)
920 struct compat_msgbuf __user *up = compat_ptr(msgp);
923 if (get_user(mtype, &up->mtype))
925 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
928 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
929 compat_ssize_t, msgsz, int, msgflg)
931 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
935 static inline int convert_mode(long *msgtyp, int msgflg)
937 if (msgflg & MSG_COPY)
938 return SEARCH_NUMBER;
940 * find message of correct type.
941 * msgtyp = 0 => get first.
942 * msgtyp > 0 => get first message of matching type.
943 * msgtyp < 0 => get message with least type must be < abs(msgtype).
948 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
952 return SEARCH_LESSEQUAL;
954 if (msgflg & MSG_EXCEPT)
955 return SEARCH_NOTEQUAL;
959 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
961 struct msgbuf __user *msgp = dest;
964 if (put_user(msg->m_type, &msgp->mtype))
967 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
968 if (store_msg(msgp->mtext, msg, msgsz))
973 #ifdef CONFIG_CHECKPOINT_RESTORE
975 * This function creates new kernel message structure, large enough to store
976 * bufsz message bytes.
978 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
980 struct msg_msg *copy;
983 * Create dummy message to copy real message to.
985 copy = load_msg(buf, bufsz);
991 static inline void free_copy(struct msg_msg *copy)
997 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
999 return ERR_PTR(-ENOSYS);
1002 static inline void free_copy(struct msg_msg *copy)
1007 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1009 struct msg_msg *msg, *found = NULL;
1012 list_for_each_entry(msg, &msq->q_messages, m_list) {
1013 if (testmsg(msg, *msgtyp, mode) &&
1014 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1016 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1017 *msgtyp = msg->m_type - 1;
1019 } else if (mode == SEARCH_NUMBER) {
1020 if (*msgtyp == count)
1028 return found ?: ERR_PTR(-EAGAIN);
1031 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1032 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1035 struct msg_queue *msq;
1036 struct ipc_namespace *ns;
1037 struct msg_msg *msg, *copy = NULL;
1038 DEFINE_WAKE_Q(wake_q);
1040 ns = current->nsproxy->ipc_ns;
1042 if (msqid < 0 || (long) bufsz < 0)
1045 if (msgflg & MSG_COPY) {
1046 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1048 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1050 return PTR_ERR(copy);
1052 mode = convert_mode(&msgtyp, msgflg);
1055 msq = msq_obtain_object_check(ns, msqid);
1059 return PTR_ERR(msq);
1063 struct msg_receiver msr_d;
1065 msg = ERR_PTR(-EACCES);
1066 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1069 ipc_lock_object(&msq->q_perm);
1071 /* raced with RMID? */
1072 if (!ipc_valid_object(&msq->q_perm)) {
1073 msg = ERR_PTR(-EIDRM);
1077 msg = find_msg(msq, &msgtyp, mode);
1080 * Found a suitable message.
1081 * Unlink it from the queue.
1083 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1084 msg = ERR_PTR(-E2BIG);
1088 * If we are copying, then do not unlink message and do
1089 * not update queue parameters.
1091 if (msgflg & MSG_COPY) {
1092 msg = copy_msg(msg, copy);
1096 list_del(&msg->m_list);
1098 msq->q_rtime = ktime_get_real_seconds();
1099 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1100 msq->q_cbytes -= msg->m_ts;
1101 atomic_sub(msg->m_ts, &ns->msg_bytes);
1102 atomic_dec(&ns->msg_hdrs);
1103 ss_wakeup(msq, &wake_q, false);
1108 /* No message waiting. Wait for a message */
1109 if (msgflg & IPC_NOWAIT) {
1110 msg = ERR_PTR(-ENOMSG);
1114 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1115 msr_d.r_tsk = current;
1116 msr_d.r_msgtype = msgtyp;
1117 msr_d.r_mode = mode;
1118 if (msgflg & MSG_NOERROR)
1119 msr_d.r_maxsize = INT_MAX;
1121 msr_d.r_maxsize = bufsz;
1122 msr_d.r_msg = ERR_PTR(-EAGAIN);
1123 __set_current_state(TASK_INTERRUPTIBLE);
1125 ipc_unlock_object(&msq->q_perm);
1130 * Lockless receive, part 1:
1131 * We don't hold a reference to the queue and getting a
1132 * reference would defeat the idea of a lockless operation,
1133 * thus the code relies on rcu to guarantee the existence of
1135 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1136 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1141 * Lockless receive, part 2:
1142 * The work in pipelined_send() and expunge_all():
1143 * - Set pointer to message
1144 * - Queue the receiver task for later wakeup
1145 * - Wake up the process after the lock is dropped.
1147 * Should the process wake up before this wakeup (due to a
1148 * signal) it will either see the message and continue ...
1150 msg = READ_ONCE(msr_d.r_msg);
1151 if (msg != ERR_PTR(-EAGAIN))
1155 * ... or see -EAGAIN, acquire the lock to check the message
1158 ipc_lock_object(&msq->q_perm);
1161 if (msg != ERR_PTR(-EAGAIN))
1164 list_del(&msr_d.r_list);
1165 if (signal_pending(current)) {
1166 msg = ERR_PTR(-ERESTARTNOHAND);
1170 ipc_unlock_object(&msq->q_perm);
1174 ipc_unlock_object(&msq->q_perm);
1180 return PTR_ERR(msg);
1183 bufsz = msg_handler(buf, msg, bufsz);
1189 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1190 long msgtyp, int msgflg)
1192 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1195 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1196 long, msgtyp, int, msgflg)
1198 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1201 #ifdef CONFIG_COMPAT
1202 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1204 struct compat_msgbuf __user *msgp = dest;
1207 if (put_user(msg->m_type, &msgp->mtype))
1210 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1211 if (store_msg(msgp->mtext, msg, msgsz))
1216 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1217 compat_long_t msgtyp, int msgflg)
1219 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1220 msgflg, compat_do_msg_fill);
1223 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1224 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1227 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1231 int msg_init_ns(struct ipc_namespace *ns)
1233 ns->msg_ctlmax = MSGMAX;
1234 ns->msg_ctlmnb = MSGMNB;
1235 ns->msg_ctlmni = MSGMNI;
1237 atomic_set(&ns->msg_bytes, 0);
1238 atomic_set(&ns->msg_hdrs, 0);
1239 return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1242 #ifdef CONFIG_IPC_NS
1243 void msg_exit_ns(struct ipc_namespace *ns)
1245 free_ipcs(ns, &msg_ids(ns), freeque);
1246 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1247 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1251 #ifdef CONFIG_PROC_FS
1252 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1254 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1255 struct user_namespace *user_ns = seq_user_ns(s);
1256 struct kern_ipc_perm *ipcp = it;
1257 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1260 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1266 pid_nr_ns(msq->q_lspid, pid_ns),
1267 pid_nr_ns(msq->q_lrpid, pid_ns),
1268 from_kuid_munged(user_ns, msq->q_perm.uid),
1269 from_kgid_munged(user_ns, msq->q_perm.gid),
1270 from_kuid_munged(user_ns, msq->q_perm.cuid),
1271 from_kgid_munged(user_ns, msq->q_perm.cgid),
1280 int __init msg_init(void)
1282 const int err = msg_init_ns(&init_ipc_ns);
1284 ipc_init_proc_interface("sysvipc/msg",
1285 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1286 IPC_MSG_IDS, sysvipc_msg_proc_show);