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 = kmalloc(sizeof(*msq), GFP_KERNEL_ACCOUNT);
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 struct task_struct *r_tsk;
256 r_tsk = get_task_struct(msr->r_tsk);
258 /* see MSG_BARRIER for purpose/pairing */
259 smp_store_release(&msr->r_msg, ERR_PTR(res));
260 wake_q_add_safe(wake_q, r_tsk);
265 * freeque() wakes up waiters on the sender and receiver waiting queue,
266 * removes the message queue from message queue ID IDR, and cleans up all the
267 * messages associated with this queue.
269 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
270 * before freeque() is called. msg_ids.rwsem remains locked on exit.
272 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
274 __releases(&msq->q_perm)
276 struct msg_msg *msg, *t;
277 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
278 DEFINE_WAKE_Q(wake_q);
280 expunge_all(msq, -EIDRM, &wake_q);
281 ss_wakeup(msq, &wake_q, true);
283 ipc_unlock_object(&msq->q_perm);
287 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
288 atomic_dec(&ns->msg_hdrs);
291 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
292 ipc_update_pid(&msq->q_lspid, NULL);
293 ipc_update_pid(&msq->q_lrpid, NULL);
294 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
297 long ksys_msgget(key_t key, int msgflg)
299 struct ipc_namespace *ns;
300 static const struct ipc_ops msg_ops = {
302 .associate = security_msg_queue_associate,
304 struct ipc_params msg_params;
306 ns = current->nsproxy->ipc_ns;
308 msg_params.key = key;
309 msg_params.flg = msgflg;
311 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
314 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
316 return ksys_msgget(key, msgflg);
319 static inline unsigned long
320 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
324 return copy_to_user(buf, in, sizeof(*in));
329 memset(&out, 0, sizeof(out));
331 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
333 out.msg_stime = in->msg_stime;
334 out.msg_rtime = in->msg_rtime;
335 out.msg_ctime = in->msg_ctime;
337 if (in->msg_cbytes > USHRT_MAX)
338 out.msg_cbytes = USHRT_MAX;
340 out.msg_cbytes = in->msg_cbytes;
341 out.msg_lcbytes = in->msg_cbytes;
343 if (in->msg_qnum > USHRT_MAX)
344 out.msg_qnum = USHRT_MAX;
346 out.msg_qnum = in->msg_qnum;
348 if (in->msg_qbytes > USHRT_MAX)
349 out.msg_qbytes = USHRT_MAX;
351 out.msg_qbytes = in->msg_qbytes;
352 out.msg_lqbytes = in->msg_qbytes;
354 out.msg_lspid = in->msg_lspid;
355 out.msg_lrpid = in->msg_lrpid;
357 return copy_to_user(buf, &out, sizeof(out));
364 static inline unsigned long
365 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
369 if (copy_from_user(out, buf, sizeof(*out)))
374 struct msqid_ds tbuf_old;
376 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
379 out->msg_perm.uid = tbuf_old.msg_perm.uid;
380 out->msg_perm.gid = tbuf_old.msg_perm.gid;
381 out->msg_perm.mode = tbuf_old.msg_perm.mode;
383 if (tbuf_old.msg_qbytes == 0)
384 out->msg_qbytes = tbuf_old.msg_lqbytes;
386 out->msg_qbytes = tbuf_old.msg_qbytes;
396 * This function handles some msgctl commands which require the rwsem
397 * to be held in write mode.
398 * NOTE: no locks must be held, the rwsem is taken inside this function.
400 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
401 struct ipc64_perm *perm, int msg_qbytes)
403 struct kern_ipc_perm *ipcp;
404 struct msg_queue *msq;
407 down_write(&msg_ids(ns).rwsem);
410 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
417 msq = container_of(ipcp, struct msg_queue, q_perm);
419 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
425 ipc_lock_object(&msq->q_perm);
426 /* freeque unlocks the ipc object and rcu */
431 DEFINE_WAKE_Q(wake_q);
433 if (msg_qbytes > ns->msg_ctlmnb &&
434 !capable(CAP_SYS_RESOURCE)) {
439 ipc_lock_object(&msq->q_perm);
440 err = ipc_update_perm(perm, ipcp);
444 msq->q_qbytes = msg_qbytes;
446 msq->q_ctime = ktime_get_real_seconds();
448 * Sleeping receivers might be excluded by
449 * stricter permissions.
451 expunge_all(msq, -EAGAIN, &wake_q);
453 * Sleeping senders might be able to send
454 * due to a larger queue size.
456 ss_wakeup(msq, &wake_q, false);
457 ipc_unlock_object(&msq->q_perm);
468 ipc_unlock_object(&msq->q_perm);
472 up_write(&msg_ids(ns).rwsem);
476 static int msgctl_info(struct ipc_namespace *ns, int msqid,
477 int cmd, struct msginfo *msginfo)
483 * We must not return kernel stack data.
484 * due to padding, it's not enough
485 * to set all member fields.
487 err = security_msg_queue_msgctl(NULL, cmd);
491 memset(msginfo, 0, sizeof(*msginfo));
492 msginfo->msgmni = ns->msg_ctlmni;
493 msginfo->msgmax = ns->msg_ctlmax;
494 msginfo->msgmnb = ns->msg_ctlmnb;
495 msginfo->msgssz = MSGSSZ;
496 msginfo->msgseg = MSGSEG;
497 down_read(&msg_ids(ns).rwsem);
498 if (cmd == MSG_INFO) {
499 msginfo->msgpool = msg_ids(ns).in_use;
500 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
501 msginfo->msgtql = atomic_read(&ns->msg_bytes);
503 msginfo->msgmap = MSGMAP;
504 msginfo->msgpool = MSGPOOL;
505 msginfo->msgtql = MSGTQL;
507 max_idx = ipc_get_maxidx(&msg_ids(ns));
508 up_read(&msg_ids(ns).rwsem);
509 return (max_idx < 0) ? 0 : max_idx;
512 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
513 int cmd, struct msqid64_ds *p)
515 struct msg_queue *msq;
518 memset(p, 0, sizeof(*p));
521 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
522 msq = msq_obtain_object(ns, msqid);
527 } else { /* IPC_STAT */
528 msq = msq_obtain_object_check(ns, msqid);
535 /* see comment for SHM_STAT_ANY */
536 if (cmd == MSG_STAT_ANY)
537 audit_ipc_obj(&msq->q_perm);
540 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
544 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
548 ipc_lock_object(&msq->q_perm);
550 if (!ipc_valid_object(&msq->q_perm)) {
551 ipc_unlock_object(&msq->q_perm);
556 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
557 p->msg_stime = msq->q_stime;
558 p->msg_rtime = msq->q_rtime;
559 p->msg_ctime = msq->q_ctime;
561 p->msg_stime_high = msq->q_stime >> 32;
562 p->msg_rtime_high = msq->q_rtime >> 32;
563 p->msg_ctime_high = msq->q_ctime >> 32;
565 p->msg_cbytes = msq->q_cbytes;
566 p->msg_qnum = msq->q_qnum;
567 p->msg_qbytes = msq->q_qbytes;
568 p->msg_lspid = pid_vnr(msq->q_lspid);
569 p->msg_lrpid = pid_vnr(msq->q_lrpid);
571 if (cmd == IPC_STAT) {
574 * Return 0 on success
579 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
580 * Return the full id, including the sequence number
582 err = msq->q_perm.id;
585 ipc_unlock_object(&msq->q_perm);
591 static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
593 struct ipc_namespace *ns;
594 struct msqid64_ds msqid64;
597 if (msqid < 0 || cmd < 0)
600 ns = current->nsproxy->ipc_ns;
605 struct msginfo msginfo;
606 err = msgctl_info(ns, msqid, cmd, &msginfo);
609 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
613 case MSG_STAT: /* msqid is an index rather than a msg queue id */
616 err = msgctl_stat(ns, msqid, cmd, &msqid64);
619 if (copy_msqid_to_user(buf, &msqid64, version))
623 if (copy_msqid_from_user(&msqid64, buf, version))
625 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
628 return msgctl_down(ns, msqid, cmd, NULL, 0);
634 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
636 return ksys_msgctl(msqid, cmd, buf, IPC_64);
639 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
640 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
642 int version = ipc_parse_version(&cmd);
644 return ksys_msgctl(msqid, cmd, buf, version);
647 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
649 return ksys_old_msgctl(msqid, cmd, buf);
655 struct compat_msqid_ds {
656 struct compat_ipc_perm msg_perm;
657 compat_uptr_t msg_first;
658 compat_uptr_t msg_last;
659 old_time32_t msg_stime;
660 old_time32_t msg_rtime;
661 old_time32_t msg_ctime;
662 compat_ulong_t msg_lcbytes;
663 compat_ulong_t msg_lqbytes;
664 unsigned short msg_cbytes;
665 unsigned short msg_qnum;
666 unsigned short msg_qbytes;
667 compat_ipc_pid_t msg_lspid;
668 compat_ipc_pid_t msg_lrpid;
671 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
674 memset(out, 0, sizeof(*out));
675 if (version == IPC_64) {
676 struct compat_msqid64_ds __user *p = buf;
677 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
679 if (get_user(out->msg_qbytes, &p->msg_qbytes))
682 struct compat_msqid_ds __user *p = buf;
683 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
685 if (get_user(out->msg_qbytes, &p->msg_qbytes))
691 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
694 if (version == IPC_64) {
695 struct compat_msqid64_ds v;
696 memset(&v, 0, sizeof(v));
697 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
698 v.msg_stime = lower_32_bits(in->msg_stime);
699 v.msg_stime_high = upper_32_bits(in->msg_stime);
700 v.msg_rtime = lower_32_bits(in->msg_rtime);
701 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
702 v.msg_ctime = lower_32_bits(in->msg_ctime);
703 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
704 v.msg_cbytes = in->msg_cbytes;
705 v.msg_qnum = in->msg_qnum;
706 v.msg_qbytes = in->msg_qbytes;
707 v.msg_lspid = in->msg_lspid;
708 v.msg_lrpid = in->msg_lrpid;
709 return copy_to_user(buf, &v, sizeof(v));
711 struct compat_msqid_ds v;
712 memset(&v, 0, sizeof(v));
713 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
714 v.msg_stime = in->msg_stime;
715 v.msg_rtime = in->msg_rtime;
716 v.msg_ctime = in->msg_ctime;
717 v.msg_cbytes = in->msg_cbytes;
718 v.msg_qnum = in->msg_qnum;
719 v.msg_qbytes = in->msg_qbytes;
720 v.msg_lspid = in->msg_lspid;
721 v.msg_lrpid = in->msg_lrpid;
722 return copy_to_user(buf, &v, sizeof(v));
726 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
728 struct ipc_namespace *ns;
730 struct msqid64_ds msqid64;
732 ns = current->nsproxy->ipc_ns;
734 if (msqid < 0 || cmd < 0)
737 switch (cmd & (~IPC_64)) {
740 struct msginfo msginfo;
741 err = msgctl_info(ns, msqid, cmd, &msginfo);
744 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
751 err = msgctl_stat(ns, msqid, cmd, &msqid64);
754 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
758 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
760 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
762 return msgctl_down(ns, msqid, cmd, NULL, 0);
768 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
770 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
773 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
774 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
776 int version = compat_ipc_parse_version(&cmd);
778 return compat_ksys_msgctl(msqid, cmd, uptr, version);
781 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
783 return compat_ksys_old_msgctl(msqid, cmd, uptr);
788 static int testmsg(struct msg_msg *msg, long type, int mode)
794 case SEARCH_LESSEQUAL:
795 if (msg->m_type <= type)
799 if (msg->m_type == type)
802 case SEARCH_NOTEQUAL:
803 if (msg->m_type != type)
810 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
811 struct wake_q_head *wake_q)
813 struct msg_receiver *msr, *t;
815 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
816 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
817 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
818 msr->r_msgtype, msr->r_mode)) {
820 list_del(&msr->r_list);
821 if (msr->r_maxsize < msg->m_ts) {
822 wake_q_add(wake_q, msr->r_tsk);
824 /* See expunge_all regarding memory barrier */
825 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
827 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
828 msq->q_rtime = ktime_get_real_seconds();
830 wake_q_add(wake_q, msr->r_tsk);
832 /* See expunge_all regarding memory barrier */
833 smp_store_release(&msr->r_msg, msg);
842 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
843 size_t msgsz, int msgflg)
845 struct msg_queue *msq;
848 struct ipc_namespace *ns;
849 DEFINE_WAKE_Q(wake_q);
851 ns = current->nsproxy->ipc_ns;
853 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
858 msg = load_msg(mtext, msgsz);
866 msq = msq_obtain_object_check(ns, msqid);
872 ipc_lock_object(&msq->q_perm);
878 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
881 /* raced with RMID? */
882 if (!ipc_valid_object(&msq->q_perm)) {
887 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
891 if (msg_fits_inqueue(msq, msgsz))
894 /* queue full, wait: */
895 if (msgflg & IPC_NOWAIT) {
900 /* enqueue the sender and prepare to block */
901 ss_add(msq, &s, msgsz);
903 if (!ipc_rcu_getref(&msq->q_perm)) {
908 ipc_unlock_object(&msq->q_perm);
913 ipc_lock_object(&msq->q_perm);
915 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
916 /* raced with RMID? */
917 if (!ipc_valid_object(&msq->q_perm)) {
923 if (signal_pending(current)) {
924 err = -ERESTARTNOHAND;
930 ipc_update_pid(&msq->q_lspid, task_tgid(current));
931 msq->q_stime = ktime_get_real_seconds();
933 if (!pipelined_send(msq, msg, &wake_q)) {
934 /* no one is waiting for this message, enqueue it */
935 list_add_tail(&msg->m_list, &msq->q_messages);
936 msq->q_cbytes += msgsz;
938 atomic_add(msgsz, &ns->msg_bytes);
939 atomic_inc(&ns->msg_hdrs);
946 ipc_unlock_object(&msq->q_perm);
955 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
960 if (get_user(mtype, &msgp->mtype))
962 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
965 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
968 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
973 struct compat_msgbuf {
978 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
979 compat_ssize_t msgsz, int msgflg)
981 struct compat_msgbuf __user *up = compat_ptr(msgp);
984 if (get_user(mtype, &up->mtype))
986 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
989 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
990 compat_ssize_t, msgsz, int, msgflg)
992 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
996 static inline int convert_mode(long *msgtyp, int msgflg)
998 if (msgflg & MSG_COPY)
999 return SEARCH_NUMBER;
1001 * find message of correct type.
1002 * msgtyp = 0 => get first.
1003 * msgtyp > 0 => get first message of matching type.
1004 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1009 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1013 return SEARCH_LESSEQUAL;
1015 if (msgflg & MSG_EXCEPT)
1016 return SEARCH_NOTEQUAL;
1017 return SEARCH_EQUAL;
1020 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1022 struct msgbuf __user *msgp = dest;
1025 if (put_user(msg->m_type, &msgp->mtype))
1028 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1029 if (store_msg(msgp->mtext, msg, msgsz))
1034 #ifdef CONFIG_CHECKPOINT_RESTORE
1036 * This function creates new kernel message structure, large enough to store
1037 * bufsz message bytes.
1039 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1041 struct msg_msg *copy;
1044 * Create dummy message to copy real message to.
1046 copy = load_msg(buf, bufsz);
1052 static inline void free_copy(struct msg_msg *copy)
1058 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1060 return ERR_PTR(-ENOSYS);
1063 static inline void free_copy(struct msg_msg *copy)
1068 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1070 struct msg_msg *msg, *found = NULL;
1073 list_for_each_entry(msg, &msq->q_messages, m_list) {
1074 if (testmsg(msg, *msgtyp, mode) &&
1075 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1077 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1078 *msgtyp = msg->m_type - 1;
1080 } else if (mode == SEARCH_NUMBER) {
1081 if (*msgtyp == count)
1089 return found ?: ERR_PTR(-EAGAIN);
1092 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1093 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1096 struct msg_queue *msq;
1097 struct ipc_namespace *ns;
1098 struct msg_msg *msg, *copy = NULL;
1099 DEFINE_WAKE_Q(wake_q);
1101 ns = current->nsproxy->ipc_ns;
1103 if (msqid < 0 || (long) bufsz < 0)
1106 if (msgflg & MSG_COPY) {
1107 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1109 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1111 return PTR_ERR(copy);
1113 mode = convert_mode(&msgtyp, msgflg);
1116 msq = msq_obtain_object_check(ns, msqid);
1120 return PTR_ERR(msq);
1124 struct msg_receiver msr_d;
1126 msg = ERR_PTR(-EACCES);
1127 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1130 ipc_lock_object(&msq->q_perm);
1132 /* raced with RMID? */
1133 if (!ipc_valid_object(&msq->q_perm)) {
1134 msg = ERR_PTR(-EIDRM);
1138 msg = find_msg(msq, &msgtyp, mode);
1141 * Found a suitable message.
1142 * Unlink it from the queue.
1144 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1145 msg = ERR_PTR(-E2BIG);
1149 * If we are copying, then do not unlink message and do
1150 * not update queue parameters.
1152 if (msgflg & MSG_COPY) {
1153 msg = copy_msg(msg, copy);
1157 list_del(&msg->m_list);
1159 msq->q_rtime = ktime_get_real_seconds();
1160 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1161 msq->q_cbytes -= msg->m_ts;
1162 atomic_sub(msg->m_ts, &ns->msg_bytes);
1163 atomic_dec(&ns->msg_hdrs);
1164 ss_wakeup(msq, &wake_q, false);
1169 /* No message waiting. Wait for a message */
1170 if (msgflg & IPC_NOWAIT) {
1171 msg = ERR_PTR(-ENOMSG);
1175 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1176 msr_d.r_tsk = current;
1177 msr_d.r_msgtype = msgtyp;
1178 msr_d.r_mode = mode;
1179 if (msgflg & MSG_NOERROR)
1180 msr_d.r_maxsize = INT_MAX;
1182 msr_d.r_maxsize = bufsz;
1184 /* memory barrier not require due to ipc_lock_object() */
1185 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1187 /* memory barrier not required, we own ipc_lock_object() */
1188 __set_current_state(TASK_INTERRUPTIBLE);
1190 ipc_unlock_object(&msq->q_perm);
1195 * Lockless receive, part 1:
1196 * We don't hold a reference to the queue and getting a
1197 * reference would defeat the idea of a lockless operation,
1198 * thus the code relies on rcu to guarantee the existence of
1200 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1201 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1206 * Lockless receive, part 2:
1207 * The work in pipelined_send() and expunge_all():
1208 * - Set pointer to message
1209 * - Queue the receiver task for later wakeup
1210 * - Wake up the process after the lock is dropped.
1212 * Should the process wake up before this wakeup (due to a
1213 * signal) it will either see the message and continue ...
1215 msg = READ_ONCE(msr_d.r_msg);
1216 if (msg != ERR_PTR(-EAGAIN)) {
1217 /* see MSG_BARRIER for purpose/pairing */
1218 smp_acquire__after_ctrl_dep();
1224 * ... or see -EAGAIN, acquire the lock to check the message
1227 ipc_lock_object(&msq->q_perm);
1229 msg = READ_ONCE(msr_d.r_msg);
1230 if (msg != ERR_PTR(-EAGAIN))
1233 list_del(&msr_d.r_list);
1234 if (signal_pending(current)) {
1235 msg = ERR_PTR(-ERESTARTNOHAND);
1239 ipc_unlock_object(&msq->q_perm);
1243 ipc_unlock_object(&msq->q_perm);
1249 return PTR_ERR(msg);
1252 bufsz = msg_handler(buf, msg, bufsz);
1258 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1259 long msgtyp, int msgflg)
1261 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1264 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1265 long, msgtyp, int, msgflg)
1267 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1270 #ifdef CONFIG_COMPAT
1271 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1273 struct compat_msgbuf __user *msgp = dest;
1276 if (put_user(msg->m_type, &msgp->mtype))
1279 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1280 if (store_msg(msgp->mtext, msg, msgsz))
1285 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1286 compat_long_t msgtyp, int msgflg)
1288 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1289 msgflg, compat_do_msg_fill);
1292 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1293 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1296 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1300 void msg_init_ns(struct ipc_namespace *ns)
1302 ns->msg_ctlmax = MSGMAX;
1303 ns->msg_ctlmnb = MSGMNB;
1304 ns->msg_ctlmni = MSGMNI;
1306 atomic_set(&ns->msg_bytes, 0);
1307 atomic_set(&ns->msg_hdrs, 0);
1308 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1311 #ifdef CONFIG_IPC_NS
1312 void msg_exit_ns(struct ipc_namespace *ns)
1314 free_ipcs(ns, &msg_ids(ns), freeque);
1315 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1316 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1320 #ifdef CONFIG_PROC_FS
1321 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1323 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1324 struct user_namespace *user_ns = seq_user_ns(s);
1325 struct kern_ipc_perm *ipcp = it;
1326 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1329 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1335 pid_nr_ns(msq->q_lspid, pid_ns),
1336 pid_nr_ns(msq->q_lrpid, pid_ns),
1337 from_kuid_munged(user_ns, msq->q_perm.uid),
1338 from_kgid_munged(user_ns, msq->q_perm.gid),
1339 from_kuid_munged(user_ns, msq->q_perm.cuid),
1340 from_kgid_munged(user_ns, msq->q_perm.cgid),
1349 void __init msg_init(void)
1351 msg_init_ns(&init_ipc_ns);
1353 ipc_init_proc_interface("sysvipc/msg",
1354 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1355 IPC_MSG_IDS, sysvipc_msg_proc_show);