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;
65 * MSG_BARRIER Locking:
67 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
68 * does not acquire any locks when it sees that a message exists in
69 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
70 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
71 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
74 /* one msg_receiver structure for each sleeping receiver */
76 struct list_head r_list;
77 struct task_struct *r_tsk;
83 struct msg_msg *r_msg;
86 /* one msg_sender for each sleeping sender */
88 struct list_head list;
89 struct task_struct *tsk;
94 #define SEARCH_EQUAL 2
95 #define SEARCH_NOTEQUAL 3
96 #define SEARCH_LESSEQUAL 4
97 #define SEARCH_NUMBER 5
99 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
101 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
103 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
106 return ERR_CAST(ipcp);
108 return container_of(ipcp, struct msg_queue, q_perm);
111 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
114 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
117 return ERR_CAST(ipcp);
119 return container_of(ipcp, struct msg_queue, q_perm);
122 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
124 ipc_rmid(&msg_ids(ns), &s->q_perm);
127 static void msg_rcu_free(struct rcu_head *head)
129 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
130 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
132 security_msg_queue_free(&msq->q_perm);
137 * newque - Create a new msg queue
139 * @params: ptr to the structure that contains the key and msgflg
141 * Called with msg_ids.rwsem held (writer)
143 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
145 struct msg_queue *msq;
147 key_t key = params->key;
148 int msgflg = params->flg;
150 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
154 msq->q_perm.mode = msgflg & S_IRWXUGO;
155 msq->q_perm.key = key;
157 msq->q_perm.security = NULL;
158 retval = security_msg_queue_alloc(&msq->q_perm);
164 msq->q_stime = msq->q_rtime = 0;
165 msq->q_ctime = ktime_get_real_seconds();
166 msq->q_cbytes = msq->q_qnum = 0;
167 msq->q_qbytes = ns->msg_ctlmnb;
168 msq->q_lspid = msq->q_lrpid = NULL;
169 INIT_LIST_HEAD(&msq->q_messages);
170 INIT_LIST_HEAD(&msq->q_receivers);
171 INIT_LIST_HEAD(&msq->q_senders);
173 /* ipc_addid() locks msq upon success. */
174 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
176 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
180 ipc_unlock_object(&msq->q_perm);
183 return msq->q_perm.id;
186 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
188 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
189 1 + msq->q_qnum <= msq->q_qbytes;
192 static inline void ss_add(struct msg_queue *msq,
193 struct msg_sender *mss, size_t msgsz)
198 * No memory barrier required: we did ipc_lock_object(),
199 * and the waker obtains that lock before calling wake_q_add().
201 __set_current_state(TASK_INTERRUPTIBLE);
202 list_add_tail(&mss->list, &msq->q_senders);
205 static inline void ss_del(struct msg_sender *mss)
208 list_del(&mss->list);
211 static void ss_wakeup(struct msg_queue *msq,
212 struct wake_q_head *wake_q, bool kill)
214 struct msg_sender *mss, *t;
215 struct task_struct *stop_tsk = NULL;
216 struct list_head *h = &msq->q_senders;
218 list_for_each_entry_safe(mss, t, h, list) {
220 mss->list.next = NULL;
223 * Stop at the first task we don't wakeup,
224 * we've already iterated the original
227 else if (stop_tsk == mss->tsk)
230 * We are not in an EIDRM scenario here, therefore
231 * verify that we really need to wakeup the task.
232 * To maintain current semantics and wakeup order,
233 * move the sender to the tail on behalf of the
236 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
240 list_move_tail(&mss->list, &msq->q_senders);
244 wake_q_add(wake_q, mss->tsk);
248 static void expunge_all(struct msg_queue *msq, int res,
249 struct wake_q_head *wake_q)
251 struct msg_receiver *msr, *t;
253 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
254 get_task_struct(msr->r_tsk);
256 /* see MSG_BARRIER for purpose/pairing */
257 smp_store_release(&msr->r_msg, ERR_PTR(res));
258 wake_q_add_safe(wake_q, msr->r_tsk);
263 * freeque() wakes up waiters on the sender and receiver waiting queue,
264 * removes the message queue from message queue ID IDR, and cleans up all the
265 * messages associated with this queue.
267 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
268 * before freeque() is called. msg_ids.rwsem remains locked on exit.
270 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
272 struct msg_msg *msg, *t;
273 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
274 DEFINE_WAKE_Q(wake_q);
276 expunge_all(msq, -EIDRM, &wake_q);
277 ss_wakeup(msq, &wake_q, true);
279 ipc_unlock_object(&msq->q_perm);
283 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
284 atomic_dec(&ns->msg_hdrs);
287 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
288 ipc_update_pid(&msq->q_lspid, NULL);
289 ipc_update_pid(&msq->q_lrpid, NULL);
290 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
293 long ksys_msgget(key_t key, int msgflg)
295 struct ipc_namespace *ns;
296 static const struct ipc_ops msg_ops = {
298 .associate = security_msg_queue_associate,
300 struct ipc_params msg_params;
302 ns = current->nsproxy->ipc_ns;
304 msg_params.key = key;
305 msg_params.flg = msgflg;
307 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
310 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
312 return ksys_msgget(key, msgflg);
315 static inline unsigned long
316 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
320 return copy_to_user(buf, in, sizeof(*in));
325 memset(&out, 0, sizeof(out));
327 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
329 out.msg_stime = in->msg_stime;
330 out.msg_rtime = in->msg_rtime;
331 out.msg_ctime = in->msg_ctime;
333 if (in->msg_cbytes > USHRT_MAX)
334 out.msg_cbytes = USHRT_MAX;
336 out.msg_cbytes = in->msg_cbytes;
337 out.msg_lcbytes = in->msg_cbytes;
339 if (in->msg_qnum > USHRT_MAX)
340 out.msg_qnum = USHRT_MAX;
342 out.msg_qnum = in->msg_qnum;
344 if (in->msg_qbytes > USHRT_MAX)
345 out.msg_qbytes = USHRT_MAX;
347 out.msg_qbytes = in->msg_qbytes;
348 out.msg_lqbytes = in->msg_qbytes;
350 out.msg_lspid = in->msg_lspid;
351 out.msg_lrpid = in->msg_lrpid;
353 return copy_to_user(buf, &out, sizeof(out));
360 static inline unsigned long
361 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
365 if (copy_from_user(out, buf, sizeof(*out)))
370 struct msqid_ds tbuf_old;
372 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
375 out->msg_perm.uid = tbuf_old.msg_perm.uid;
376 out->msg_perm.gid = tbuf_old.msg_perm.gid;
377 out->msg_perm.mode = tbuf_old.msg_perm.mode;
379 if (tbuf_old.msg_qbytes == 0)
380 out->msg_qbytes = tbuf_old.msg_lqbytes;
382 out->msg_qbytes = tbuf_old.msg_qbytes;
392 * This function handles some msgctl commands which require the rwsem
393 * to be held in write mode.
394 * NOTE: no locks must be held, the rwsem is taken inside this function.
396 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
397 struct ipc64_perm *perm, int msg_qbytes)
399 struct kern_ipc_perm *ipcp;
400 struct msg_queue *msq;
403 down_write(&msg_ids(ns).rwsem);
406 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
413 msq = container_of(ipcp, struct msg_queue, q_perm);
415 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
421 ipc_lock_object(&msq->q_perm);
422 /* freeque unlocks the ipc object and rcu */
427 DEFINE_WAKE_Q(wake_q);
429 if (msg_qbytes > ns->msg_ctlmnb &&
430 !capable(CAP_SYS_RESOURCE)) {
435 ipc_lock_object(&msq->q_perm);
436 err = ipc_update_perm(perm, ipcp);
440 msq->q_qbytes = msg_qbytes;
442 msq->q_ctime = ktime_get_real_seconds();
444 * Sleeping receivers might be excluded by
445 * stricter permissions.
447 expunge_all(msq, -EAGAIN, &wake_q);
449 * Sleeping senders might be able to send
450 * due to a larger queue size.
452 ss_wakeup(msq, &wake_q, false);
453 ipc_unlock_object(&msq->q_perm);
464 ipc_unlock_object(&msq->q_perm);
468 up_write(&msg_ids(ns).rwsem);
472 static int msgctl_info(struct ipc_namespace *ns, int msqid,
473 int cmd, struct msginfo *msginfo)
479 * We must not return kernel stack data.
480 * due to padding, it's not enough
481 * to set all member fields.
483 err = security_msg_queue_msgctl(NULL, cmd);
487 memset(msginfo, 0, sizeof(*msginfo));
488 msginfo->msgmni = ns->msg_ctlmni;
489 msginfo->msgmax = ns->msg_ctlmax;
490 msginfo->msgmnb = ns->msg_ctlmnb;
491 msginfo->msgssz = MSGSSZ;
492 msginfo->msgseg = MSGSEG;
493 down_read(&msg_ids(ns).rwsem);
494 if (cmd == MSG_INFO) {
495 msginfo->msgpool = msg_ids(ns).in_use;
496 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
497 msginfo->msgtql = atomic_read(&ns->msg_bytes);
499 msginfo->msgmap = MSGMAP;
500 msginfo->msgpool = MSGPOOL;
501 msginfo->msgtql = MSGTQL;
503 max_idx = ipc_get_maxidx(&msg_ids(ns));
504 up_read(&msg_ids(ns).rwsem);
505 return (max_idx < 0) ? 0 : max_idx;
508 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
509 int cmd, struct msqid64_ds *p)
511 struct msg_queue *msq;
514 memset(p, 0, sizeof(*p));
517 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
518 msq = msq_obtain_object(ns, msqid);
523 } else { /* IPC_STAT */
524 msq = msq_obtain_object_check(ns, msqid);
531 /* see comment for SHM_STAT_ANY */
532 if (cmd == MSG_STAT_ANY)
533 audit_ipc_obj(&msq->q_perm);
536 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
540 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
544 ipc_lock_object(&msq->q_perm);
546 if (!ipc_valid_object(&msq->q_perm)) {
547 ipc_unlock_object(&msq->q_perm);
552 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
553 p->msg_stime = msq->q_stime;
554 p->msg_rtime = msq->q_rtime;
555 p->msg_ctime = msq->q_ctime;
557 p->msg_stime_high = msq->q_stime >> 32;
558 p->msg_rtime_high = msq->q_rtime >> 32;
559 p->msg_ctime_high = msq->q_ctime >> 32;
561 p->msg_cbytes = msq->q_cbytes;
562 p->msg_qnum = msq->q_qnum;
563 p->msg_qbytes = msq->q_qbytes;
564 p->msg_lspid = pid_vnr(msq->q_lspid);
565 p->msg_lrpid = pid_vnr(msq->q_lrpid);
567 if (cmd == IPC_STAT) {
570 * Return 0 on success
575 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
576 * Return the full id, including the sequence number
578 err = msq->q_perm.id;
581 ipc_unlock_object(&msq->q_perm);
587 static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
589 struct ipc_namespace *ns;
590 struct msqid64_ds msqid64;
593 if (msqid < 0 || cmd < 0)
596 ns = current->nsproxy->ipc_ns;
601 struct msginfo msginfo;
602 err = msgctl_info(ns, msqid, cmd, &msginfo);
605 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
609 case MSG_STAT: /* msqid is an index rather than a msg queue id */
612 err = msgctl_stat(ns, msqid, cmd, &msqid64);
615 if (copy_msqid_to_user(buf, &msqid64, version))
619 if (copy_msqid_from_user(&msqid64, buf, version))
621 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
624 return msgctl_down(ns, msqid, cmd, NULL, 0);
630 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
632 return ksys_msgctl(msqid, cmd, buf, IPC_64);
635 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
636 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
638 int version = ipc_parse_version(&cmd);
640 return ksys_msgctl(msqid, cmd, buf, version);
643 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
645 return ksys_old_msgctl(msqid, cmd, buf);
651 struct compat_msqid_ds {
652 struct compat_ipc_perm msg_perm;
653 compat_uptr_t msg_first;
654 compat_uptr_t msg_last;
655 old_time32_t msg_stime;
656 old_time32_t msg_rtime;
657 old_time32_t msg_ctime;
658 compat_ulong_t msg_lcbytes;
659 compat_ulong_t msg_lqbytes;
660 unsigned short msg_cbytes;
661 unsigned short msg_qnum;
662 unsigned short msg_qbytes;
663 compat_ipc_pid_t msg_lspid;
664 compat_ipc_pid_t msg_lrpid;
667 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
670 memset(out, 0, sizeof(*out));
671 if (version == IPC_64) {
672 struct compat_msqid64_ds __user *p = buf;
673 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
675 if (get_user(out->msg_qbytes, &p->msg_qbytes))
678 struct compat_msqid_ds __user *p = buf;
679 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
681 if (get_user(out->msg_qbytes, &p->msg_qbytes))
687 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
690 if (version == IPC_64) {
691 struct compat_msqid64_ds v;
692 memset(&v, 0, sizeof(v));
693 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
694 v.msg_stime = lower_32_bits(in->msg_stime);
695 v.msg_stime_high = upper_32_bits(in->msg_stime);
696 v.msg_rtime = lower_32_bits(in->msg_rtime);
697 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
698 v.msg_ctime = lower_32_bits(in->msg_ctime);
699 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
700 v.msg_cbytes = in->msg_cbytes;
701 v.msg_qnum = in->msg_qnum;
702 v.msg_qbytes = in->msg_qbytes;
703 v.msg_lspid = in->msg_lspid;
704 v.msg_lrpid = in->msg_lrpid;
705 return copy_to_user(buf, &v, sizeof(v));
707 struct compat_msqid_ds v;
708 memset(&v, 0, sizeof(v));
709 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
710 v.msg_stime = in->msg_stime;
711 v.msg_rtime = in->msg_rtime;
712 v.msg_ctime = in->msg_ctime;
713 v.msg_cbytes = in->msg_cbytes;
714 v.msg_qnum = in->msg_qnum;
715 v.msg_qbytes = in->msg_qbytes;
716 v.msg_lspid = in->msg_lspid;
717 v.msg_lrpid = in->msg_lrpid;
718 return copy_to_user(buf, &v, sizeof(v));
722 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
724 struct ipc_namespace *ns;
726 struct msqid64_ds msqid64;
728 ns = current->nsproxy->ipc_ns;
730 if (msqid < 0 || cmd < 0)
733 switch (cmd & (~IPC_64)) {
736 struct msginfo msginfo;
737 err = msgctl_info(ns, msqid, cmd, &msginfo);
740 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
747 err = msgctl_stat(ns, msqid, cmd, &msqid64);
750 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
754 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
756 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
758 return msgctl_down(ns, msqid, cmd, NULL, 0);
764 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
766 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
769 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
770 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
772 int version = compat_ipc_parse_version(&cmd);
774 return compat_ksys_msgctl(msqid, cmd, uptr, version);
777 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
779 return compat_ksys_old_msgctl(msqid, cmd, uptr);
784 static int testmsg(struct msg_msg *msg, long type, int mode)
790 case SEARCH_LESSEQUAL:
791 if (msg->m_type <= type)
795 if (msg->m_type == type)
798 case SEARCH_NOTEQUAL:
799 if (msg->m_type != type)
806 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
807 struct wake_q_head *wake_q)
809 struct msg_receiver *msr, *t;
811 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
812 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
813 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
814 msr->r_msgtype, msr->r_mode)) {
816 list_del(&msr->r_list);
817 if (msr->r_maxsize < msg->m_ts) {
818 wake_q_add(wake_q, msr->r_tsk);
820 /* See expunge_all regarding memory barrier */
821 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
823 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
824 msq->q_rtime = ktime_get_real_seconds();
826 wake_q_add(wake_q, msr->r_tsk);
828 /* See expunge_all regarding memory barrier */
829 smp_store_release(&msr->r_msg, msg);
838 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
839 size_t msgsz, int msgflg)
841 struct msg_queue *msq;
844 struct ipc_namespace *ns;
845 DEFINE_WAKE_Q(wake_q);
847 ns = current->nsproxy->ipc_ns;
849 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
854 msg = load_msg(mtext, msgsz);
862 msq = msq_obtain_object_check(ns, msqid);
868 ipc_lock_object(&msq->q_perm);
874 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
877 /* raced with RMID? */
878 if (!ipc_valid_object(&msq->q_perm)) {
883 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
887 if (msg_fits_inqueue(msq, msgsz))
890 /* queue full, wait: */
891 if (msgflg & IPC_NOWAIT) {
896 /* enqueue the sender and prepare to block */
897 ss_add(msq, &s, msgsz);
899 if (!ipc_rcu_getref(&msq->q_perm)) {
904 ipc_unlock_object(&msq->q_perm);
909 ipc_lock_object(&msq->q_perm);
911 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
912 /* raced with RMID? */
913 if (!ipc_valid_object(&msq->q_perm)) {
919 if (signal_pending(current)) {
920 err = -ERESTARTNOHAND;
926 ipc_update_pid(&msq->q_lspid, task_tgid(current));
927 msq->q_stime = ktime_get_real_seconds();
929 if (!pipelined_send(msq, msg, &wake_q)) {
930 /* no one is waiting for this message, enqueue it */
931 list_add_tail(&msg->m_list, &msq->q_messages);
932 msq->q_cbytes += msgsz;
934 atomic_add(msgsz, &ns->msg_bytes);
935 atomic_inc(&ns->msg_hdrs);
942 ipc_unlock_object(&msq->q_perm);
951 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
956 if (get_user(mtype, &msgp->mtype))
958 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
961 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
964 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
969 struct compat_msgbuf {
974 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
975 compat_ssize_t msgsz, int msgflg)
977 struct compat_msgbuf __user *up = compat_ptr(msgp);
980 if (get_user(mtype, &up->mtype))
982 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
985 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
986 compat_ssize_t, msgsz, int, msgflg)
988 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
992 static inline int convert_mode(long *msgtyp, int msgflg)
994 if (msgflg & MSG_COPY)
995 return SEARCH_NUMBER;
997 * find message of correct type.
998 * msgtyp = 0 => get first.
999 * msgtyp > 0 => get first message of matching type.
1000 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1005 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1009 return SEARCH_LESSEQUAL;
1011 if (msgflg & MSG_EXCEPT)
1012 return SEARCH_NOTEQUAL;
1013 return SEARCH_EQUAL;
1016 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1018 struct msgbuf __user *msgp = dest;
1021 if (put_user(msg->m_type, &msgp->mtype))
1024 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1025 if (store_msg(msgp->mtext, msg, msgsz))
1030 #ifdef CONFIG_CHECKPOINT_RESTORE
1032 * This function creates new kernel message structure, large enough to store
1033 * bufsz message bytes.
1035 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1037 struct msg_msg *copy;
1040 * Create dummy message to copy real message to.
1042 copy = load_msg(buf, bufsz);
1048 static inline void free_copy(struct msg_msg *copy)
1054 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1056 return ERR_PTR(-ENOSYS);
1059 static inline void free_copy(struct msg_msg *copy)
1064 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1066 struct msg_msg *msg, *found = NULL;
1069 list_for_each_entry(msg, &msq->q_messages, m_list) {
1070 if (testmsg(msg, *msgtyp, mode) &&
1071 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1073 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1074 *msgtyp = msg->m_type - 1;
1076 } else if (mode == SEARCH_NUMBER) {
1077 if (*msgtyp == count)
1085 return found ?: ERR_PTR(-EAGAIN);
1088 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1089 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1092 struct msg_queue *msq;
1093 struct ipc_namespace *ns;
1094 struct msg_msg *msg, *copy = NULL;
1095 DEFINE_WAKE_Q(wake_q);
1097 ns = current->nsproxy->ipc_ns;
1099 if (msqid < 0 || (long) bufsz < 0)
1102 if (msgflg & MSG_COPY) {
1103 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1105 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1107 return PTR_ERR(copy);
1109 mode = convert_mode(&msgtyp, msgflg);
1112 msq = msq_obtain_object_check(ns, msqid);
1116 return PTR_ERR(msq);
1120 struct msg_receiver msr_d;
1122 msg = ERR_PTR(-EACCES);
1123 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1126 ipc_lock_object(&msq->q_perm);
1128 /* raced with RMID? */
1129 if (!ipc_valid_object(&msq->q_perm)) {
1130 msg = ERR_PTR(-EIDRM);
1134 msg = find_msg(msq, &msgtyp, mode);
1137 * Found a suitable message.
1138 * Unlink it from the queue.
1140 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1141 msg = ERR_PTR(-E2BIG);
1145 * If we are copying, then do not unlink message and do
1146 * not update queue parameters.
1148 if (msgflg & MSG_COPY) {
1149 msg = copy_msg(msg, copy);
1153 list_del(&msg->m_list);
1155 msq->q_rtime = ktime_get_real_seconds();
1156 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1157 msq->q_cbytes -= msg->m_ts;
1158 atomic_sub(msg->m_ts, &ns->msg_bytes);
1159 atomic_dec(&ns->msg_hdrs);
1160 ss_wakeup(msq, &wake_q, false);
1165 /* No message waiting. Wait for a message */
1166 if (msgflg & IPC_NOWAIT) {
1167 msg = ERR_PTR(-ENOMSG);
1171 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1172 msr_d.r_tsk = current;
1173 msr_d.r_msgtype = msgtyp;
1174 msr_d.r_mode = mode;
1175 if (msgflg & MSG_NOERROR)
1176 msr_d.r_maxsize = INT_MAX;
1178 msr_d.r_maxsize = bufsz;
1180 /* memory barrier not require due to ipc_lock_object() */
1181 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1183 /* memory barrier not required, we own ipc_lock_object() */
1184 __set_current_state(TASK_INTERRUPTIBLE);
1186 ipc_unlock_object(&msq->q_perm);
1191 * Lockless receive, part 1:
1192 * We don't hold a reference to the queue and getting a
1193 * reference would defeat the idea of a lockless operation,
1194 * thus the code relies on rcu to guarantee the existence of
1196 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1197 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1202 * Lockless receive, part 2:
1203 * The work in pipelined_send() and expunge_all():
1204 * - Set pointer to message
1205 * - Queue the receiver task for later wakeup
1206 * - Wake up the process after the lock is dropped.
1208 * Should the process wake up before this wakeup (due to a
1209 * signal) it will either see the message and continue ...
1211 msg = READ_ONCE(msr_d.r_msg);
1212 if (msg != ERR_PTR(-EAGAIN)) {
1213 /* see MSG_BARRIER for purpose/pairing */
1214 smp_acquire__after_ctrl_dep();
1220 * ... or see -EAGAIN, acquire the lock to check the message
1223 ipc_lock_object(&msq->q_perm);
1225 msg = READ_ONCE(msr_d.r_msg);
1226 if (msg != ERR_PTR(-EAGAIN))
1229 list_del(&msr_d.r_list);
1230 if (signal_pending(current)) {
1231 msg = ERR_PTR(-ERESTARTNOHAND);
1235 ipc_unlock_object(&msq->q_perm);
1239 ipc_unlock_object(&msq->q_perm);
1245 return PTR_ERR(msg);
1248 bufsz = msg_handler(buf, msg, bufsz);
1254 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1255 long msgtyp, int msgflg)
1257 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1260 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1261 long, msgtyp, int, msgflg)
1263 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1266 #ifdef CONFIG_COMPAT
1267 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1269 struct compat_msgbuf __user *msgp = dest;
1272 if (put_user(msg->m_type, &msgp->mtype))
1275 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1276 if (store_msg(msgp->mtext, msg, msgsz))
1281 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1282 compat_long_t msgtyp, int msgflg)
1284 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1285 msgflg, compat_do_msg_fill);
1288 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1289 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1292 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1296 void msg_init_ns(struct ipc_namespace *ns)
1298 ns->msg_ctlmax = MSGMAX;
1299 ns->msg_ctlmnb = MSGMNB;
1300 ns->msg_ctlmni = MSGMNI;
1302 atomic_set(&ns->msg_bytes, 0);
1303 atomic_set(&ns->msg_hdrs, 0);
1304 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1307 #ifdef CONFIG_IPC_NS
1308 void msg_exit_ns(struct ipc_namespace *ns)
1310 free_ipcs(ns, &msg_ids(ns), freeque);
1311 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1312 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1316 #ifdef CONFIG_PROC_FS
1317 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1319 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1320 struct user_namespace *user_ns = seq_user_ns(s);
1321 struct kern_ipc_perm *ipcp = it;
1322 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1325 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1331 pid_nr_ns(msq->q_lspid, pid_ns),
1332 pid_nr_ns(msq->q_lrpid, pid_ns),
1333 from_kuid_munged(user_ns, msq->q_perm.uid),
1334 from_kgid_munged(user_ns, msq->q_perm.gid),
1335 from_kuid_munged(user_ns, msq->q_perm.cuid),
1336 from_kgid_munged(user_ns, msq->q_perm.cgid),
1345 void __init msg_init(void)
1347 msg_init_ns(&init_ipc_ns);
1349 ipc_init_proc_interface("sysvipc/msg",
1350 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1351 IPC_MSG_IDS, sysvipc_msg_proc_show);