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>
42 #include <linux/percpu_counter.h>
44 #include <asm/current.h>
45 #include <linux/uaccess.h>
48 /* one msq_queue structure for each present queue on the system */
50 struct kern_ipc_perm q_perm;
51 time64_t q_stime; /* last msgsnd time */
52 time64_t q_rtime; /* last msgrcv time */
53 time64_t q_ctime; /* last change time */
54 unsigned long q_cbytes; /* current number of bytes on queue */
55 unsigned long q_qnum; /* number of messages in queue */
56 unsigned long q_qbytes; /* max number of bytes on queue */
57 struct pid *q_lspid; /* pid of last msgsnd */
58 struct pid *q_lrpid; /* last receive pid */
60 struct list_head q_messages;
61 struct list_head q_receivers;
62 struct list_head q_senders;
66 * MSG_BARRIER Locking:
68 * Similar to the optimization used in ipc/mqueue.c, one syscall return path
69 * does not acquire any locks when it sees that a message exists in
70 * msg_receiver.r_msg. Therefore r_msg is set using smp_store_release()
71 * and accessed using READ_ONCE()+smp_acquire__after_ctrl_dep(). In addition,
72 * wake_q_add_safe() is used. See ipc/mqueue.c for more details
75 /* one msg_receiver structure for each sleeping receiver */
77 struct list_head r_list;
78 struct task_struct *r_tsk;
84 struct msg_msg *r_msg;
87 /* one msg_sender for each sleeping sender */
89 struct list_head list;
90 struct task_struct *tsk;
95 #define SEARCH_EQUAL 2
96 #define SEARCH_NOTEQUAL 3
97 #define SEARCH_LESSEQUAL 4
98 #define SEARCH_NUMBER 5
100 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
102 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
104 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
107 return ERR_CAST(ipcp);
109 return container_of(ipcp, struct msg_queue, q_perm);
112 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
115 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
118 return ERR_CAST(ipcp);
120 return container_of(ipcp, struct msg_queue, q_perm);
123 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
125 ipc_rmid(&msg_ids(ns), &s->q_perm);
128 static void msg_rcu_free(struct rcu_head *head)
130 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
131 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
133 security_msg_queue_free(&msq->q_perm);
138 * newque - Create a new msg queue
140 * @params: ptr to the structure that contains the key and msgflg
142 * Called with msg_ids.rwsem held (writer)
144 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
146 struct msg_queue *msq;
148 key_t key = params->key;
149 int msgflg = params->flg;
151 msq = kmalloc(sizeof(*msq), GFP_KERNEL_ACCOUNT);
155 msq->q_perm.mode = msgflg & S_IRWXUGO;
156 msq->q_perm.key = key;
158 msq->q_perm.security = NULL;
159 retval = security_msg_queue_alloc(&msq->q_perm);
165 msq->q_stime = msq->q_rtime = 0;
166 msq->q_ctime = ktime_get_real_seconds();
167 msq->q_cbytes = msq->q_qnum = 0;
168 msq->q_qbytes = ns->msg_ctlmnb;
169 msq->q_lspid = msq->q_lrpid = NULL;
170 INIT_LIST_HEAD(&msq->q_messages);
171 INIT_LIST_HEAD(&msq->q_receivers);
172 INIT_LIST_HEAD(&msq->q_senders);
174 /* ipc_addid() locks msq upon success. */
175 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
177 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
181 ipc_unlock_object(&msq->q_perm);
184 return msq->q_perm.id;
187 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
189 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
190 1 + msq->q_qnum <= msq->q_qbytes;
193 static inline void ss_add(struct msg_queue *msq,
194 struct msg_sender *mss, size_t msgsz)
199 * No memory barrier required: we did ipc_lock_object(),
200 * and the waker obtains that lock before calling wake_q_add().
202 __set_current_state(TASK_INTERRUPTIBLE);
203 list_add_tail(&mss->list, &msq->q_senders);
206 static inline void ss_del(struct msg_sender *mss)
209 list_del(&mss->list);
212 static void ss_wakeup(struct msg_queue *msq,
213 struct wake_q_head *wake_q, bool kill)
215 struct msg_sender *mss, *t;
216 struct task_struct *stop_tsk = NULL;
217 struct list_head *h = &msq->q_senders;
219 list_for_each_entry_safe(mss, t, h, list) {
221 mss->list.next = NULL;
224 * Stop at the first task we don't wakeup,
225 * we've already iterated the original
228 else if (stop_tsk == mss->tsk)
231 * We are not in an EIDRM scenario here, therefore
232 * verify that we really need to wakeup the task.
233 * To maintain current semantics and wakeup order,
234 * move the sender to the tail on behalf of the
237 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
241 list_move_tail(&mss->list, &msq->q_senders);
245 wake_q_add(wake_q, mss->tsk);
249 static void expunge_all(struct msg_queue *msq, int res,
250 struct wake_q_head *wake_q)
252 struct msg_receiver *msr, *t;
254 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
255 struct task_struct *r_tsk;
257 r_tsk = get_task_struct(msr->r_tsk);
259 /* see MSG_BARRIER for purpose/pairing */
260 smp_store_release(&msr->r_msg, ERR_PTR(res));
261 wake_q_add_safe(wake_q, r_tsk);
266 * freeque() wakes up waiters on the sender and receiver waiting queue,
267 * removes the message queue from message queue ID IDR, and cleans up all the
268 * messages associated with this queue.
270 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
271 * before freeque() is called. msg_ids.rwsem remains locked on exit.
273 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
275 __releases(&msq->q_perm)
277 struct msg_msg *msg, *t;
278 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
279 DEFINE_WAKE_Q(wake_q);
281 expunge_all(msq, -EIDRM, &wake_q);
282 ss_wakeup(msq, &wake_q, true);
284 ipc_unlock_object(&msq->q_perm);
288 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
289 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
292 percpu_counter_sub_local(&ns->percpu_msg_bytes, msq->q_cbytes);
293 ipc_update_pid(&msq->q_lspid, NULL);
294 ipc_update_pid(&msq->q_lrpid, NULL);
295 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
298 long ksys_msgget(key_t key, int msgflg)
300 struct ipc_namespace *ns;
301 static const struct ipc_ops msg_ops = {
303 .associate = security_msg_queue_associate,
305 struct ipc_params msg_params;
307 ns = current->nsproxy->ipc_ns;
309 msg_params.key = key;
310 msg_params.flg = msgflg;
312 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
315 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
317 return ksys_msgget(key, msgflg);
320 static inline unsigned long
321 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
325 return copy_to_user(buf, in, sizeof(*in));
330 memset(&out, 0, sizeof(out));
332 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
334 out.msg_stime = in->msg_stime;
335 out.msg_rtime = in->msg_rtime;
336 out.msg_ctime = in->msg_ctime;
338 if (in->msg_cbytes > USHRT_MAX)
339 out.msg_cbytes = USHRT_MAX;
341 out.msg_cbytes = in->msg_cbytes;
342 out.msg_lcbytes = in->msg_cbytes;
344 if (in->msg_qnum > USHRT_MAX)
345 out.msg_qnum = USHRT_MAX;
347 out.msg_qnum = in->msg_qnum;
349 if (in->msg_qbytes > USHRT_MAX)
350 out.msg_qbytes = USHRT_MAX;
352 out.msg_qbytes = in->msg_qbytes;
353 out.msg_lqbytes = in->msg_qbytes;
355 out.msg_lspid = in->msg_lspid;
356 out.msg_lrpid = in->msg_lrpid;
358 return copy_to_user(buf, &out, sizeof(out));
365 static inline unsigned long
366 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
370 if (copy_from_user(out, buf, sizeof(*out)))
375 struct msqid_ds tbuf_old;
377 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
380 out->msg_perm.uid = tbuf_old.msg_perm.uid;
381 out->msg_perm.gid = tbuf_old.msg_perm.gid;
382 out->msg_perm.mode = tbuf_old.msg_perm.mode;
384 if (tbuf_old.msg_qbytes == 0)
385 out->msg_qbytes = tbuf_old.msg_lqbytes;
387 out->msg_qbytes = tbuf_old.msg_qbytes;
397 * This function handles some msgctl commands which require the rwsem
398 * to be held in write mode.
399 * NOTE: no locks must be held, the rwsem is taken inside this function.
401 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
402 struct ipc64_perm *perm, int msg_qbytes)
404 struct kern_ipc_perm *ipcp;
405 struct msg_queue *msq;
408 down_write(&msg_ids(ns).rwsem);
411 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
418 msq = container_of(ipcp, struct msg_queue, q_perm);
420 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
426 ipc_lock_object(&msq->q_perm);
427 /* freeque unlocks the ipc object and rcu */
432 DEFINE_WAKE_Q(wake_q);
434 if (msg_qbytes > ns->msg_ctlmnb &&
435 !capable(CAP_SYS_RESOURCE)) {
440 ipc_lock_object(&msq->q_perm);
441 err = ipc_update_perm(perm, ipcp);
445 msq->q_qbytes = msg_qbytes;
447 msq->q_ctime = ktime_get_real_seconds();
449 * Sleeping receivers might be excluded by
450 * stricter permissions.
452 expunge_all(msq, -EAGAIN, &wake_q);
454 * Sleeping senders might be able to send
455 * due to a larger queue size.
457 ss_wakeup(msq, &wake_q, false);
458 ipc_unlock_object(&msq->q_perm);
469 ipc_unlock_object(&msq->q_perm);
473 up_write(&msg_ids(ns).rwsem);
477 static int msgctl_info(struct ipc_namespace *ns, int msqid,
478 int cmd, struct msginfo *msginfo)
484 * We must not return kernel stack data.
485 * due to padding, it's not enough
486 * to set all member fields.
488 err = security_msg_queue_msgctl(NULL, cmd);
492 memset(msginfo, 0, sizeof(*msginfo));
493 msginfo->msgmni = ns->msg_ctlmni;
494 msginfo->msgmax = ns->msg_ctlmax;
495 msginfo->msgmnb = ns->msg_ctlmnb;
496 msginfo->msgssz = MSGSSZ;
497 msginfo->msgseg = MSGSEG;
498 down_read(&msg_ids(ns).rwsem);
500 msginfo->msgpool = msg_ids(ns).in_use;
501 max_idx = ipc_get_maxidx(&msg_ids(ns));
502 up_read(&msg_ids(ns).rwsem);
503 if (cmd == MSG_INFO) {
504 msginfo->msgmap = min_t(int,
505 percpu_counter_sum(&ns->percpu_msg_hdrs),
507 msginfo->msgtql = min_t(int,
508 percpu_counter_sum(&ns->percpu_msg_bytes),
511 msginfo->msgmap = MSGMAP;
512 msginfo->msgpool = MSGPOOL;
513 msginfo->msgtql = MSGTQL;
515 return (max_idx < 0) ? 0 : max_idx;
518 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
519 int cmd, struct msqid64_ds *p)
521 struct msg_queue *msq;
524 memset(p, 0, sizeof(*p));
527 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
528 msq = msq_obtain_object(ns, msqid);
533 } else { /* IPC_STAT */
534 msq = msq_obtain_object_check(ns, msqid);
541 /* see comment for SHM_STAT_ANY */
542 if (cmd == MSG_STAT_ANY)
543 audit_ipc_obj(&msq->q_perm);
546 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
550 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
554 ipc_lock_object(&msq->q_perm);
556 if (!ipc_valid_object(&msq->q_perm)) {
557 ipc_unlock_object(&msq->q_perm);
562 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
563 p->msg_stime = msq->q_stime;
564 p->msg_rtime = msq->q_rtime;
565 p->msg_ctime = msq->q_ctime;
567 p->msg_stime_high = msq->q_stime >> 32;
568 p->msg_rtime_high = msq->q_rtime >> 32;
569 p->msg_ctime_high = msq->q_ctime >> 32;
571 p->msg_cbytes = msq->q_cbytes;
572 p->msg_qnum = msq->q_qnum;
573 p->msg_qbytes = msq->q_qbytes;
574 p->msg_lspid = pid_vnr(msq->q_lspid);
575 p->msg_lrpid = pid_vnr(msq->q_lrpid);
577 if (cmd == IPC_STAT) {
580 * Return 0 on success
585 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
586 * Return the full id, including the sequence number
588 err = msq->q_perm.id;
591 ipc_unlock_object(&msq->q_perm);
597 static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
599 struct ipc_namespace *ns;
600 struct msqid64_ds msqid64;
603 if (msqid < 0 || cmd < 0)
606 ns = current->nsproxy->ipc_ns;
611 struct msginfo msginfo;
612 err = msgctl_info(ns, msqid, cmd, &msginfo);
615 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
619 case MSG_STAT: /* msqid is an index rather than a msg queue id */
622 err = msgctl_stat(ns, msqid, cmd, &msqid64);
625 if (copy_msqid_to_user(buf, &msqid64, version))
629 if (copy_msqid_from_user(&msqid64, buf, version))
631 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
634 return msgctl_down(ns, msqid, cmd, NULL, 0);
640 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
642 return ksys_msgctl(msqid, cmd, buf, IPC_64);
645 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
646 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
648 int version = ipc_parse_version(&cmd);
650 return ksys_msgctl(msqid, cmd, buf, version);
653 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
655 return ksys_old_msgctl(msqid, cmd, buf);
661 struct compat_msqid_ds {
662 struct compat_ipc_perm msg_perm;
663 compat_uptr_t msg_first;
664 compat_uptr_t msg_last;
665 old_time32_t msg_stime;
666 old_time32_t msg_rtime;
667 old_time32_t msg_ctime;
668 compat_ulong_t msg_lcbytes;
669 compat_ulong_t msg_lqbytes;
670 unsigned short msg_cbytes;
671 unsigned short msg_qnum;
672 unsigned short msg_qbytes;
673 compat_ipc_pid_t msg_lspid;
674 compat_ipc_pid_t msg_lrpid;
677 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
680 memset(out, 0, sizeof(*out));
681 if (version == IPC_64) {
682 struct compat_msqid64_ds __user *p = buf;
683 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
685 if (get_user(out->msg_qbytes, &p->msg_qbytes))
688 struct compat_msqid_ds __user *p = buf;
689 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
691 if (get_user(out->msg_qbytes, &p->msg_qbytes))
697 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
700 if (version == IPC_64) {
701 struct compat_msqid64_ds v;
702 memset(&v, 0, sizeof(v));
703 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
704 v.msg_stime = lower_32_bits(in->msg_stime);
705 v.msg_stime_high = upper_32_bits(in->msg_stime);
706 v.msg_rtime = lower_32_bits(in->msg_rtime);
707 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
708 v.msg_ctime = lower_32_bits(in->msg_ctime);
709 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
710 v.msg_cbytes = in->msg_cbytes;
711 v.msg_qnum = in->msg_qnum;
712 v.msg_qbytes = in->msg_qbytes;
713 v.msg_lspid = in->msg_lspid;
714 v.msg_lrpid = in->msg_lrpid;
715 return copy_to_user(buf, &v, sizeof(v));
717 struct compat_msqid_ds v;
718 memset(&v, 0, sizeof(v));
719 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
720 v.msg_stime = in->msg_stime;
721 v.msg_rtime = in->msg_rtime;
722 v.msg_ctime = in->msg_ctime;
723 v.msg_cbytes = in->msg_cbytes;
724 v.msg_qnum = in->msg_qnum;
725 v.msg_qbytes = in->msg_qbytes;
726 v.msg_lspid = in->msg_lspid;
727 v.msg_lrpid = in->msg_lrpid;
728 return copy_to_user(buf, &v, sizeof(v));
732 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
734 struct ipc_namespace *ns;
736 struct msqid64_ds msqid64;
738 ns = current->nsproxy->ipc_ns;
740 if (msqid < 0 || cmd < 0)
743 switch (cmd & (~IPC_64)) {
746 struct msginfo msginfo;
747 err = msgctl_info(ns, msqid, cmd, &msginfo);
750 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
757 err = msgctl_stat(ns, msqid, cmd, &msqid64);
760 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
764 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
766 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
768 return msgctl_down(ns, msqid, cmd, NULL, 0);
774 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
776 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
779 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
780 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
782 int version = compat_ipc_parse_version(&cmd);
784 return compat_ksys_msgctl(msqid, cmd, uptr, version);
787 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
789 return compat_ksys_old_msgctl(msqid, cmd, uptr);
794 static int testmsg(struct msg_msg *msg, long type, int mode)
800 case SEARCH_LESSEQUAL:
801 if (msg->m_type <= type)
805 if (msg->m_type == type)
808 case SEARCH_NOTEQUAL:
809 if (msg->m_type != type)
816 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
817 struct wake_q_head *wake_q)
819 struct msg_receiver *msr, *t;
821 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
822 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
823 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
824 msr->r_msgtype, msr->r_mode)) {
826 list_del(&msr->r_list);
827 if (msr->r_maxsize < msg->m_ts) {
828 wake_q_add(wake_q, msr->r_tsk);
830 /* See expunge_all regarding memory barrier */
831 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
833 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
834 msq->q_rtime = ktime_get_real_seconds();
836 wake_q_add(wake_q, msr->r_tsk);
838 /* See expunge_all regarding memory barrier */
839 smp_store_release(&msr->r_msg, msg);
848 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
849 size_t msgsz, int msgflg)
851 struct msg_queue *msq;
854 struct ipc_namespace *ns;
855 DEFINE_WAKE_Q(wake_q);
857 ns = current->nsproxy->ipc_ns;
859 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
864 msg = load_msg(mtext, msgsz);
872 msq = msq_obtain_object_check(ns, msqid);
878 ipc_lock_object(&msq->q_perm);
884 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
887 /* raced with RMID? */
888 if (!ipc_valid_object(&msq->q_perm)) {
893 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
897 if (msg_fits_inqueue(msq, msgsz))
900 /* queue full, wait: */
901 if (msgflg & IPC_NOWAIT) {
906 /* enqueue the sender and prepare to block */
907 ss_add(msq, &s, msgsz);
909 if (!ipc_rcu_getref(&msq->q_perm)) {
914 ipc_unlock_object(&msq->q_perm);
919 ipc_lock_object(&msq->q_perm);
921 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
922 /* raced with RMID? */
923 if (!ipc_valid_object(&msq->q_perm)) {
929 if (signal_pending(current)) {
930 err = -ERESTARTNOHAND;
936 ipc_update_pid(&msq->q_lspid, task_tgid(current));
937 msq->q_stime = ktime_get_real_seconds();
939 if (!pipelined_send(msq, msg, &wake_q)) {
940 /* no one is waiting for this message, enqueue it */
941 list_add_tail(&msg->m_list, &msq->q_messages);
942 msq->q_cbytes += msgsz;
944 percpu_counter_add_local(&ns->percpu_msg_bytes, msgsz);
945 percpu_counter_add_local(&ns->percpu_msg_hdrs, 1);
952 ipc_unlock_object(&msq->q_perm);
961 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
966 if (get_user(mtype, &msgp->mtype))
968 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
971 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
974 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
979 struct compat_msgbuf {
984 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
985 compat_ssize_t msgsz, int msgflg)
987 struct compat_msgbuf __user *up = compat_ptr(msgp);
990 if (get_user(mtype, &up->mtype))
992 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
995 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
996 compat_ssize_t, msgsz, int, msgflg)
998 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
1002 static inline int convert_mode(long *msgtyp, int msgflg)
1004 if (msgflg & MSG_COPY)
1005 return SEARCH_NUMBER;
1007 * find message of correct type.
1008 * msgtyp = 0 => get first.
1009 * msgtyp > 0 => get first message of matching type.
1010 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1015 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1019 return SEARCH_LESSEQUAL;
1021 if (msgflg & MSG_EXCEPT)
1022 return SEARCH_NOTEQUAL;
1023 return SEARCH_EQUAL;
1026 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1028 struct msgbuf __user *msgp = dest;
1031 if (put_user(msg->m_type, &msgp->mtype))
1034 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1035 if (store_msg(msgp->mtext, msg, msgsz))
1040 #ifdef CONFIG_CHECKPOINT_RESTORE
1042 * This function creates new kernel message structure, large enough to store
1043 * bufsz message bytes.
1045 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1047 struct msg_msg *copy;
1050 * Create dummy message to copy real message to.
1052 copy = load_msg(buf, bufsz);
1058 static inline void free_copy(struct msg_msg *copy)
1064 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1066 return ERR_PTR(-ENOSYS);
1069 static inline void free_copy(struct msg_msg *copy)
1074 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1076 struct msg_msg *msg, *found = NULL;
1079 list_for_each_entry(msg, &msq->q_messages, m_list) {
1080 if (testmsg(msg, *msgtyp, mode) &&
1081 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1083 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1084 *msgtyp = msg->m_type - 1;
1086 } else if (mode == SEARCH_NUMBER) {
1087 if (*msgtyp == count)
1095 return found ?: ERR_PTR(-EAGAIN);
1098 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1099 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1102 struct msg_queue *msq;
1103 struct ipc_namespace *ns;
1104 struct msg_msg *msg, *copy = NULL;
1105 DEFINE_WAKE_Q(wake_q);
1107 ns = current->nsproxy->ipc_ns;
1109 if (msqid < 0 || (long) bufsz < 0)
1112 if (msgflg & MSG_COPY) {
1113 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1115 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1117 return PTR_ERR(copy);
1119 mode = convert_mode(&msgtyp, msgflg);
1122 msq = msq_obtain_object_check(ns, msqid);
1126 return PTR_ERR(msq);
1130 struct msg_receiver msr_d;
1132 msg = ERR_PTR(-EACCES);
1133 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1136 ipc_lock_object(&msq->q_perm);
1138 /* raced with RMID? */
1139 if (!ipc_valid_object(&msq->q_perm)) {
1140 msg = ERR_PTR(-EIDRM);
1144 msg = find_msg(msq, &msgtyp, mode);
1147 * Found a suitable message.
1148 * Unlink it from the queue.
1150 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1151 msg = ERR_PTR(-E2BIG);
1155 * If we are copying, then do not unlink message and do
1156 * not update queue parameters.
1158 if (msgflg & MSG_COPY) {
1159 msg = copy_msg(msg, copy);
1163 list_del(&msg->m_list);
1165 msq->q_rtime = ktime_get_real_seconds();
1166 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1167 msq->q_cbytes -= msg->m_ts;
1168 percpu_counter_sub_local(&ns->percpu_msg_bytes, msg->m_ts);
1169 percpu_counter_sub_local(&ns->percpu_msg_hdrs, 1);
1170 ss_wakeup(msq, &wake_q, false);
1175 /* No message waiting. Wait for a message */
1176 if (msgflg & IPC_NOWAIT) {
1177 msg = ERR_PTR(-ENOMSG);
1181 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1182 msr_d.r_tsk = current;
1183 msr_d.r_msgtype = msgtyp;
1184 msr_d.r_mode = mode;
1185 if (msgflg & MSG_NOERROR)
1186 msr_d.r_maxsize = INT_MAX;
1188 msr_d.r_maxsize = bufsz;
1190 /* memory barrier not require due to ipc_lock_object() */
1191 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1193 /* memory barrier not required, we own ipc_lock_object() */
1194 __set_current_state(TASK_INTERRUPTIBLE);
1196 ipc_unlock_object(&msq->q_perm);
1201 * Lockless receive, part 1:
1202 * We don't hold a reference to the queue and getting a
1203 * reference would defeat the idea of a lockless operation,
1204 * thus the code relies on rcu to guarantee the existence of
1206 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1207 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1212 * Lockless receive, part 2:
1213 * The work in pipelined_send() and expunge_all():
1214 * - Set pointer to message
1215 * - Queue the receiver task for later wakeup
1216 * - Wake up the process after the lock is dropped.
1218 * Should the process wake up before this wakeup (due to a
1219 * signal) it will either see the message and continue ...
1221 msg = READ_ONCE(msr_d.r_msg);
1222 if (msg != ERR_PTR(-EAGAIN)) {
1223 /* see MSG_BARRIER for purpose/pairing */
1224 smp_acquire__after_ctrl_dep();
1230 * ... or see -EAGAIN, acquire the lock to check the message
1233 ipc_lock_object(&msq->q_perm);
1235 msg = READ_ONCE(msr_d.r_msg);
1236 if (msg != ERR_PTR(-EAGAIN))
1239 list_del(&msr_d.r_list);
1240 if (signal_pending(current)) {
1241 msg = ERR_PTR(-ERESTARTNOHAND);
1245 ipc_unlock_object(&msq->q_perm);
1249 ipc_unlock_object(&msq->q_perm);
1255 return PTR_ERR(msg);
1258 bufsz = msg_handler(buf, msg, bufsz);
1264 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1265 long msgtyp, int msgflg)
1267 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1270 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1271 long, msgtyp, int, msgflg)
1273 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1276 #ifdef CONFIG_COMPAT
1277 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1279 struct compat_msgbuf __user *msgp = dest;
1282 if (put_user(msg->m_type, &msgp->mtype))
1285 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1286 if (store_msg(msgp->mtext, msg, msgsz))
1291 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1292 compat_long_t msgtyp, int msgflg)
1294 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1295 msgflg, compat_do_msg_fill);
1298 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1299 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1302 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1306 int msg_init_ns(struct ipc_namespace *ns)
1310 ns->msg_ctlmax = MSGMAX;
1311 ns->msg_ctlmnb = MSGMNB;
1312 ns->msg_ctlmni = MSGMNI;
1314 ret = percpu_counter_init(&ns->percpu_msg_bytes, 0, GFP_KERNEL);
1316 goto fail_msg_bytes;
1317 ret = percpu_counter_init(&ns->percpu_msg_hdrs, 0, GFP_KERNEL);
1320 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1324 percpu_counter_destroy(&ns->percpu_msg_bytes);
1329 #ifdef CONFIG_IPC_NS
1330 void msg_exit_ns(struct ipc_namespace *ns)
1332 percpu_counter_destroy(&ns->percpu_msg_bytes);
1333 percpu_counter_destroy(&ns->percpu_msg_hdrs);
1334 free_ipcs(ns, &msg_ids(ns), freeque);
1335 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1336 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1340 #ifdef CONFIG_PROC_FS
1341 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1343 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1344 struct user_namespace *user_ns = seq_user_ns(s);
1345 struct kern_ipc_perm *ipcp = it;
1346 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1349 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1355 pid_nr_ns(msq->q_lspid, pid_ns),
1356 pid_nr_ns(msq->q_lrpid, pid_ns),
1357 from_kuid_munged(user_ns, msq->q_perm.uid),
1358 from_kgid_munged(user_ns, msq->q_perm.gid),
1359 from_kuid_munged(user_ns, msq->q_perm.cuid),
1360 from_kgid_munged(user_ns, msq->q_perm.cgid),
1369 void __init msg_init(void)
1371 msg_init_ns(&init_ipc_ns);
1373 ipc_init_proc_interface("sysvipc/msg",
1374 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1375 IPC_MSG_IDS, sysvipc_msg_proc_show);