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>
41 #include <linux/rhashtable.h>
43 #include <asm/current.h>
44 #include <linux/uaccess.h>
47 /* one msq_queue structure for each present queue on the system */
49 struct kern_ipc_perm q_perm;
50 time64_t q_stime; /* last msgsnd time */
51 time64_t q_rtime; /* last msgrcv time */
52 time64_t q_ctime; /* last change time */
53 unsigned long q_cbytes; /* current number of bytes on queue */
54 unsigned long q_qnum; /* number of messages in queue */
55 unsigned long q_qbytes; /* max number of bytes on queue */
56 struct pid *q_lspid; /* pid of last msgsnd */
57 struct pid *q_lrpid; /* last receive pid */
59 struct list_head q_messages;
60 struct list_head q_receivers;
61 struct list_head q_senders;
64 /* one msg_receiver structure for each sleeping receiver */
66 struct list_head r_list;
67 struct task_struct *r_tsk;
73 struct msg_msg *r_msg;
76 /* one msg_sender for each sleeping sender */
78 struct list_head list;
79 struct task_struct *tsk;
84 #define SEARCH_EQUAL 2
85 #define SEARCH_NOTEQUAL 3
86 #define SEARCH_LESSEQUAL 4
87 #define SEARCH_NUMBER 5
89 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
91 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
93 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
96 return ERR_CAST(ipcp);
98 return container_of(ipcp, struct msg_queue, q_perm);
101 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
104 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
107 return ERR_CAST(ipcp);
109 return container_of(ipcp, struct msg_queue, q_perm);
112 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
114 ipc_rmid(&msg_ids(ns), &s->q_perm);
117 static void msg_rcu_free(struct rcu_head *head)
119 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
120 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
122 security_msg_queue_free(&msq->q_perm);
127 * newque - Create a new msg queue
129 * @params: ptr to the structure that contains the key and msgflg
131 * Called with msg_ids.rwsem held (writer)
133 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
135 struct msg_queue *msq;
137 key_t key = params->key;
138 int msgflg = params->flg;
140 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
144 msq->q_perm.mode = msgflg & S_IRWXUGO;
145 msq->q_perm.key = key;
147 msq->q_perm.security = NULL;
148 retval = security_msg_queue_alloc(&msq->q_perm);
154 msq->q_stime = msq->q_rtime = 0;
155 msq->q_ctime = ktime_get_real_seconds();
156 msq->q_cbytes = msq->q_qnum = 0;
157 msq->q_qbytes = ns->msg_ctlmnb;
158 msq->q_lspid = msq->q_lrpid = NULL;
159 INIT_LIST_HEAD(&msq->q_messages);
160 INIT_LIST_HEAD(&msq->q_receivers);
161 INIT_LIST_HEAD(&msq->q_senders);
163 /* ipc_addid() locks msq upon success. */
164 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
166 call_rcu(&msq->q_perm.rcu, msg_rcu_free);
170 ipc_unlock_object(&msq->q_perm);
173 return msq->q_perm.id;
176 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
178 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
179 1 + msq->q_qnum <= msq->q_qbytes;
182 static inline void ss_add(struct msg_queue *msq,
183 struct msg_sender *mss, size_t msgsz)
187 __set_current_state(TASK_INTERRUPTIBLE);
188 list_add_tail(&mss->list, &msq->q_senders);
191 static inline void ss_del(struct msg_sender *mss)
194 list_del(&mss->list);
197 static void ss_wakeup(struct msg_queue *msq,
198 struct wake_q_head *wake_q, bool kill)
200 struct msg_sender *mss, *t;
201 struct task_struct *stop_tsk = NULL;
202 struct list_head *h = &msq->q_senders;
204 list_for_each_entry_safe(mss, t, h, list) {
206 mss->list.next = NULL;
209 * Stop at the first task we don't wakeup,
210 * we've already iterated the original
213 else if (stop_tsk == mss->tsk)
216 * We are not in an EIDRM scenario here, therefore
217 * verify that we really need to wakeup the task.
218 * To maintain current semantics and wakeup order,
219 * move the sender to the tail on behalf of the
222 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
226 list_move_tail(&mss->list, &msq->q_senders);
230 wake_q_add(wake_q, mss->tsk);
234 static void expunge_all(struct msg_queue *msq, int res,
235 struct wake_q_head *wake_q)
237 struct msg_receiver *msr, *t;
239 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
240 wake_q_add(wake_q, msr->r_tsk);
241 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
246 * freeque() wakes up waiters on the sender and receiver waiting queue,
247 * removes the message queue from message queue ID IDR, and cleans up all the
248 * messages associated with this queue.
250 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
251 * before freeque() is called. msg_ids.rwsem remains locked on exit.
253 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
255 struct msg_msg *msg, *t;
256 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
257 DEFINE_WAKE_Q(wake_q);
259 expunge_all(msq, -EIDRM, &wake_q);
260 ss_wakeup(msq, &wake_q, true);
262 ipc_unlock_object(&msq->q_perm);
266 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
267 atomic_dec(&ns->msg_hdrs);
270 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
271 ipc_update_pid(&msq->q_lspid, NULL);
272 ipc_update_pid(&msq->q_lrpid, NULL);
273 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
276 long ksys_msgget(key_t key, int msgflg)
278 struct ipc_namespace *ns;
279 static const struct ipc_ops msg_ops = {
281 .associate = security_msg_queue_associate,
283 struct ipc_params msg_params;
285 ns = current->nsproxy->ipc_ns;
287 msg_params.key = key;
288 msg_params.flg = msgflg;
290 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
293 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
295 return ksys_msgget(key, msgflg);
298 static inline unsigned long
299 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
303 return copy_to_user(buf, in, sizeof(*in));
308 memset(&out, 0, sizeof(out));
310 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
312 out.msg_stime = in->msg_stime;
313 out.msg_rtime = in->msg_rtime;
314 out.msg_ctime = in->msg_ctime;
316 if (in->msg_cbytes > USHRT_MAX)
317 out.msg_cbytes = USHRT_MAX;
319 out.msg_cbytes = in->msg_cbytes;
320 out.msg_lcbytes = in->msg_cbytes;
322 if (in->msg_qnum > USHRT_MAX)
323 out.msg_qnum = USHRT_MAX;
325 out.msg_qnum = in->msg_qnum;
327 if (in->msg_qbytes > USHRT_MAX)
328 out.msg_qbytes = USHRT_MAX;
330 out.msg_qbytes = in->msg_qbytes;
331 out.msg_lqbytes = in->msg_qbytes;
333 out.msg_lspid = in->msg_lspid;
334 out.msg_lrpid = in->msg_lrpid;
336 return copy_to_user(buf, &out, sizeof(out));
343 static inline unsigned long
344 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
348 if (copy_from_user(out, buf, sizeof(*out)))
353 struct msqid_ds tbuf_old;
355 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
358 out->msg_perm.uid = tbuf_old.msg_perm.uid;
359 out->msg_perm.gid = tbuf_old.msg_perm.gid;
360 out->msg_perm.mode = tbuf_old.msg_perm.mode;
362 if (tbuf_old.msg_qbytes == 0)
363 out->msg_qbytes = tbuf_old.msg_lqbytes;
365 out->msg_qbytes = tbuf_old.msg_qbytes;
375 * This function handles some msgctl commands which require the rwsem
376 * to be held in write mode.
377 * NOTE: no locks must be held, the rwsem is taken inside this function.
379 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
380 struct msqid64_ds *msqid64)
382 struct kern_ipc_perm *ipcp;
383 struct msg_queue *msq;
386 down_write(&msg_ids(ns).rwsem);
389 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
390 &msqid64->msg_perm, msqid64->msg_qbytes);
396 msq = container_of(ipcp, struct msg_queue, q_perm);
398 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
404 ipc_lock_object(&msq->q_perm);
405 /* freeque unlocks the ipc object and rcu */
410 DEFINE_WAKE_Q(wake_q);
412 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
413 !capable(CAP_SYS_RESOURCE)) {
418 ipc_lock_object(&msq->q_perm);
419 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
423 msq->q_qbytes = msqid64->msg_qbytes;
425 msq->q_ctime = ktime_get_real_seconds();
427 * Sleeping receivers might be excluded by
428 * stricter permissions.
430 expunge_all(msq, -EAGAIN, &wake_q);
432 * Sleeping senders might be able to send
433 * due to a larger queue size.
435 ss_wakeup(msq, &wake_q, false);
436 ipc_unlock_object(&msq->q_perm);
447 ipc_unlock_object(&msq->q_perm);
451 up_write(&msg_ids(ns).rwsem);
455 static int msgctl_info(struct ipc_namespace *ns, int msqid,
456 int cmd, struct msginfo *msginfo)
462 * We must not return kernel stack data.
463 * due to padding, it's not enough
464 * to set all member fields.
466 err = security_msg_queue_msgctl(NULL, cmd);
470 memset(msginfo, 0, sizeof(*msginfo));
471 msginfo->msgmni = ns->msg_ctlmni;
472 msginfo->msgmax = ns->msg_ctlmax;
473 msginfo->msgmnb = ns->msg_ctlmnb;
474 msginfo->msgssz = MSGSSZ;
475 msginfo->msgseg = MSGSEG;
476 down_read(&msg_ids(ns).rwsem);
477 if (cmd == MSG_INFO) {
478 msginfo->msgpool = msg_ids(ns).in_use;
479 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
480 msginfo->msgtql = atomic_read(&ns->msg_bytes);
482 msginfo->msgmap = MSGMAP;
483 msginfo->msgpool = MSGPOOL;
484 msginfo->msgtql = MSGTQL;
486 max_id = ipc_get_maxid(&msg_ids(ns));
487 up_read(&msg_ids(ns).rwsem);
488 return (max_id < 0) ? 0 : max_id;
491 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
492 int cmd, struct msqid64_ds *p)
494 struct msg_queue *msq;
498 memset(p, 0, sizeof(*p));
501 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
502 msq = msq_obtain_object(ns, msqid);
508 } else { /* IPC_STAT */
509 msq = msq_obtain_object_check(ns, msqid);
516 /* see comment for SHM_STAT_ANY */
517 if (cmd == MSG_STAT_ANY)
518 audit_ipc_obj(&msq->q_perm);
521 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
525 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
529 ipc_lock_object(&msq->q_perm);
531 if (!ipc_valid_object(&msq->q_perm)) {
532 ipc_unlock_object(&msq->q_perm);
537 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
538 p->msg_stime = msq->q_stime;
539 p->msg_rtime = msq->q_rtime;
540 p->msg_ctime = msq->q_ctime;
542 p->msg_stime_high = msq->q_stime >> 32;
543 p->msg_rtime_high = msq->q_rtime >> 32;
544 p->msg_ctime_high = msq->q_ctime >> 32;
546 p->msg_cbytes = msq->q_cbytes;
547 p->msg_qnum = msq->q_qnum;
548 p->msg_qbytes = msq->q_qbytes;
549 p->msg_lspid = pid_vnr(msq->q_lspid);
550 p->msg_lrpid = pid_vnr(msq->q_lrpid);
552 ipc_unlock_object(&msq->q_perm);
561 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
564 struct ipc_namespace *ns;
565 struct msqid64_ds msqid64;
568 if (msqid < 0 || cmd < 0)
571 version = ipc_parse_version(&cmd);
572 ns = current->nsproxy->ipc_ns;
577 struct msginfo msginfo;
578 err = msgctl_info(ns, msqid, cmd, &msginfo);
581 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
585 case MSG_STAT: /* msqid is an index rather than a msg queue id */
588 err = msgctl_stat(ns, msqid, cmd, &msqid64);
591 if (copy_msqid_to_user(buf, &msqid64, version))
595 if (copy_msqid_from_user(&msqid64, buf, version))
599 return msgctl_down(ns, msqid, cmd, &msqid64);
605 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
607 return ksys_msgctl(msqid, cmd, buf);
612 struct compat_msqid_ds {
613 struct compat_ipc_perm msg_perm;
614 compat_uptr_t msg_first;
615 compat_uptr_t msg_last;
616 compat_time_t msg_stime;
617 compat_time_t msg_rtime;
618 compat_time_t msg_ctime;
619 compat_ulong_t msg_lcbytes;
620 compat_ulong_t msg_lqbytes;
621 unsigned short msg_cbytes;
622 unsigned short msg_qnum;
623 unsigned short msg_qbytes;
624 compat_ipc_pid_t msg_lspid;
625 compat_ipc_pid_t msg_lrpid;
628 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
631 memset(out, 0, sizeof(*out));
632 if (version == IPC_64) {
633 struct compat_msqid64_ds __user *p = buf;
634 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
636 if (get_user(out->msg_qbytes, &p->msg_qbytes))
639 struct compat_msqid_ds __user *p = buf;
640 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
642 if (get_user(out->msg_qbytes, &p->msg_qbytes))
648 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
651 if (version == IPC_64) {
652 struct compat_msqid64_ds v;
653 memset(&v, 0, sizeof(v));
654 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
655 v.msg_stime = lower_32_bits(in->msg_stime);
656 v.msg_stime_high = upper_32_bits(in->msg_stime);
657 v.msg_rtime = lower_32_bits(in->msg_rtime);
658 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
659 v.msg_ctime = lower_32_bits(in->msg_ctime);
660 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
661 v.msg_cbytes = in->msg_cbytes;
662 v.msg_qnum = in->msg_qnum;
663 v.msg_qbytes = in->msg_qbytes;
664 v.msg_lspid = in->msg_lspid;
665 v.msg_lrpid = in->msg_lrpid;
666 return copy_to_user(buf, &v, sizeof(v));
668 struct compat_msqid_ds v;
669 memset(&v, 0, sizeof(v));
670 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
671 v.msg_stime = in->msg_stime;
672 v.msg_rtime = in->msg_rtime;
673 v.msg_ctime = in->msg_ctime;
674 v.msg_cbytes = in->msg_cbytes;
675 v.msg_qnum = in->msg_qnum;
676 v.msg_qbytes = in->msg_qbytes;
677 v.msg_lspid = in->msg_lspid;
678 v.msg_lrpid = in->msg_lrpid;
679 return copy_to_user(buf, &v, sizeof(v));
683 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
685 struct ipc_namespace *ns;
687 struct msqid64_ds msqid64;
688 int version = compat_ipc_parse_version(&cmd);
690 ns = current->nsproxy->ipc_ns;
692 if (msqid < 0 || cmd < 0)
695 switch (cmd & (~IPC_64)) {
698 struct msginfo msginfo;
699 err = msgctl_info(ns, msqid, cmd, &msginfo);
702 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
709 err = msgctl_stat(ns, msqid, cmd, &msqid64);
712 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
716 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
720 return msgctl_down(ns, msqid, cmd, &msqid64);
726 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
728 return compat_ksys_msgctl(msqid, cmd, uptr);
732 static int testmsg(struct msg_msg *msg, long type, int mode)
738 case SEARCH_LESSEQUAL:
739 if (msg->m_type <= type)
743 if (msg->m_type == type)
746 case SEARCH_NOTEQUAL:
747 if (msg->m_type != type)
754 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
755 struct wake_q_head *wake_q)
757 struct msg_receiver *msr, *t;
759 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
760 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
761 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
762 msr->r_msgtype, msr->r_mode)) {
764 list_del(&msr->r_list);
765 if (msr->r_maxsize < msg->m_ts) {
766 wake_q_add(wake_q, msr->r_tsk);
767 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
769 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
770 msq->q_rtime = ktime_get_real_seconds();
772 wake_q_add(wake_q, msr->r_tsk);
773 WRITE_ONCE(msr->r_msg, msg);
782 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
783 size_t msgsz, int msgflg)
785 struct msg_queue *msq;
788 struct ipc_namespace *ns;
789 DEFINE_WAKE_Q(wake_q);
791 ns = current->nsproxy->ipc_ns;
793 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
798 msg = load_msg(mtext, msgsz);
806 msq = msq_obtain_object_check(ns, msqid);
812 ipc_lock_object(&msq->q_perm);
818 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
821 /* raced with RMID? */
822 if (!ipc_valid_object(&msq->q_perm)) {
827 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
831 if (msg_fits_inqueue(msq, msgsz))
834 /* queue full, wait: */
835 if (msgflg & IPC_NOWAIT) {
840 /* enqueue the sender and prepare to block */
841 ss_add(msq, &s, msgsz);
843 if (!ipc_rcu_getref(&msq->q_perm)) {
848 ipc_unlock_object(&msq->q_perm);
853 ipc_lock_object(&msq->q_perm);
855 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
856 /* raced with RMID? */
857 if (!ipc_valid_object(&msq->q_perm)) {
863 if (signal_pending(current)) {
864 err = -ERESTARTNOHAND;
870 ipc_update_pid(&msq->q_lspid, task_tgid(current));
871 msq->q_stime = ktime_get_real_seconds();
873 if (!pipelined_send(msq, msg, &wake_q)) {
874 /* no one is waiting for this message, enqueue it */
875 list_add_tail(&msg->m_list, &msq->q_messages);
876 msq->q_cbytes += msgsz;
878 atomic_add(msgsz, &ns->msg_bytes);
879 atomic_inc(&ns->msg_hdrs);
886 ipc_unlock_object(&msq->q_perm);
895 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
900 if (get_user(mtype, &msgp->mtype))
902 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
905 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
908 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
913 struct compat_msgbuf {
918 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
919 compat_ssize_t msgsz, int msgflg)
921 struct compat_msgbuf __user *up = compat_ptr(msgp);
924 if (get_user(mtype, &up->mtype))
926 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
929 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
930 compat_ssize_t, msgsz, int, msgflg)
932 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
936 static inline int convert_mode(long *msgtyp, int msgflg)
938 if (msgflg & MSG_COPY)
939 return SEARCH_NUMBER;
941 * find message of correct type.
942 * msgtyp = 0 => get first.
943 * msgtyp > 0 => get first message of matching type.
944 * msgtyp < 0 => get message with least type must be < abs(msgtype).
949 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
953 return SEARCH_LESSEQUAL;
955 if (msgflg & MSG_EXCEPT)
956 return SEARCH_NOTEQUAL;
960 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
962 struct msgbuf __user *msgp = dest;
965 if (put_user(msg->m_type, &msgp->mtype))
968 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
969 if (store_msg(msgp->mtext, msg, msgsz))
974 #ifdef CONFIG_CHECKPOINT_RESTORE
976 * This function creates new kernel message structure, large enough to store
977 * bufsz message bytes.
979 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
981 struct msg_msg *copy;
984 * Create dummy message to copy real message to.
986 copy = load_msg(buf, bufsz);
992 static inline void free_copy(struct msg_msg *copy)
998 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1000 return ERR_PTR(-ENOSYS);
1003 static inline void free_copy(struct msg_msg *copy)
1008 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1010 struct msg_msg *msg, *found = NULL;
1013 list_for_each_entry(msg, &msq->q_messages, m_list) {
1014 if (testmsg(msg, *msgtyp, mode) &&
1015 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1017 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1018 *msgtyp = msg->m_type - 1;
1020 } else if (mode == SEARCH_NUMBER) {
1021 if (*msgtyp == count)
1029 return found ?: ERR_PTR(-EAGAIN);
1032 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1033 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1036 struct msg_queue *msq;
1037 struct ipc_namespace *ns;
1038 struct msg_msg *msg, *copy = NULL;
1039 DEFINE_WAKE_Q(wake_q);
1041 ns = current->nsproxy->ipc_ns;
1043 if (msqid < 0 || (long) bufsz < 0)
1046 if (msgflg & MSG_COPY) {
1047 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1049 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1051 return PTR_ERR(copy);
1053 mode = convert_mode(&msgtyp, msgflg);
1056 msq = msq_obtain_object_check(ns, msqid);
1060 return PTR_ERR(msq);
1064 struct msg_receiver msr_d;
1066 msg = ERR_PTR(-EACCES);
1067 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1070 ipc_lock_object(&msq->q_perm);
1072 /* raced with RMID? */
1073 if (!ipc_valid_object(&msq->q_perm)) {
1074 msg = ERR_PTR(-EIDRM);
1078 msg = find_msg(msq, &msgtyp, mode);
1081 * Found a suitable message.
1082 * Unlink it from the queue.
1084 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1085 msg = ERR_PTR(-E2BIG);
1089 * If we are copying, then do not unlink message and do
1090 * not update queue parameters.
1092 if (msgflg & MSG_COPY) {
1093 msg = copy_msg(msg, copy);
1097 list_del(&msg->m_list);
1099 msq->q_rtime = ktime_get_real_seconds();
1100 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1101 msq->q_cbytes -= msg->m_ts;
1102 atomic_sub(msg->m_ts, &ns->msg_bytes);
1103 atomic_dec(&ns->msg_hdrs);
1104 ss_wakeup(msq, &wake_q, false);
1109 /* No message waiting. Wait for a message */
1110 if (msgflg & IPC_NOWAIT) {
1111 msg = ERR_PTR(-ENOMSG);
1115 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1116 msr_d.r_tsk = current;
1117 msr_d.r_msgtype = msgtyp;
1118 msr_d.r_mode = mode;
1119 if (msgflg & MSG_NOERROR)
1120 msr_d.r_maxsize = INT_MAX;
1122 msr_d.r_maxsize = bufsz;
1123 msr_d.r_msg = ERR_PTR(-EAGAIN);
1124 __set_current_state(TASK_INTERRUPTIBLE);
1126 ipc_unlock_object(&msq->q_perm);
1131 * Lockless receive, part 1:
1132 * We don't hold a reference to the queue and getting a
1133 * reference would defeat the idea of a lockless operation,
1134 * thus the code relies on rcu to guarantee the existence of
1136 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1137 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1142 * Lockless receive, part 2:
1143 * The work in pipelined_send() and expunge_all():
1144 * - Set pointer to message
1145 * - Queue the receiver task for later wakeup
1146 * - Wake up the process after the lock is dropped.
1148 * Should the process wake up before this wakeup (due to a
1149 * signal) it will either see the message and continue ...
1151 msg = READ_ONCE(msr_d.r_msg);
1152 if (msg != ERR_PTR(-EAGAIN))
1156 * ... or see -EAGAIN, acquire the lock to check the message
1159 ipc_lock_object(&msq->q_perm);
1162 if (msg != ERR_PTR(-EAGAIN))
1165 list_del(&msr_d.r_list);
1166 if (signal_pending(current)) {
1167 msg = ERR_PTR(-ERESTARTNOHAND);
1171 ipc_unlock_object(&msq->q_perm);
1175 ipc_unlock_object(&msq->q_perm);
1181 return PTR_ERR(msg);
1184 bufsz = msg_handler(buf, msg, bufsz);
1190 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1191 long msgtyp, int msgflg)
1193 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1196 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1197 long, msgtyp, int, msgflg)
1199 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1202 #ifdef CONFIG_COMPAT
1203 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1205 struct compat_msgbuf __user *msgp = dest;
1208 if (put_user(msg->m_type, &msgp->mtype))
1211 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1212 if (store_msg(msgp->mtext, msg, msgsz))
1217 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1218 compat_long_t msgtyp, int msgflg)
1220 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1221 msgflg, compat_do_msg_fill);
1224 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1225 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1228 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1232 int msg_init_ns(struct ipc_namespace *ns)
1234 ns->msg_ctlmax = MSGMAX;
1235 ns->msg_ctlmnb = MSGMNB;
1236 ns->msg_ctlmni = MSGMNI;
1238 atomic_set(&ns->msg_bytes, 0);
1239 atomic_set(&ns->msg_hdrs, 0);
1240 return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1243 #ifdef CONFIG_IPC_NS
1244 void msg_exit_ns(struct ipc_namespace *ns)
1246 free_ipcs(ns, &msg_ids(ns), freeque);
1247 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1248 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1252 #ifdef CONFIG_PROC_FS
1253 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1255 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1256 struct user_namespace *user_ns = seq_user_ns(s);
1257 struct kern_ipc_perm *ipcp = it;
1258 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1261 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1267 pid_nr_ns(msq->q_lspid, pid_ns),
1268 pid_nr_ns(msq->q_lrpid, pid_ns),
1269 from_kuid_munged(user_ns, msq->q_perm.uid),
1270 from_kgid_munged(user_ns, msq->q_perm.gid),
1271 from_kuid_munged(user_ns, msq->q_perm.cuid),
1272 from_kgid_munged(user_ns, msq->q_perm.cgid),
1281 int __init msg_init(void)
1283 const int err = msg_init_ns(&init_ipc_ns);
1285 ipc_init_proc_interface("sysvipc/msg",
1286 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1287 IPC_MSG_IDS, sysvipc_msg_proc_show);