1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992 Krishna Balasubramanian
6 * Removed all the remaining kerneld mess
7 * Catch the -EFAULT stuff properly
8 * Use GFP_KERNEL for messages as in 1.2
9 * Fixed up the unchecked user space derefs
10 * Copyright (C) 1998 Alan Cox & Andi Kleen
12 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
14 * mostly rewritten, threaded and wake-one semantics added
15 * MSGMAX limit removed, sysctl's added
16 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
23 * Pavel Emelianov <xemul@openvz.org>
26 #include <linux/capability.h>
27 #include <linux/msg.h>
28 #include <linux/spinlock.h>
29 #include <linux/init.h>
31 #include <linux/proc_fs.h>
32 #include <linux/list.h>
33 #include <linux/security.h>
34 #include <linux/sched/wake_q.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/ipc_namespace.h>
42 #include <asm/current.h>
43 #include <linux/uaccess.h>
46 /* one msg_receiver structure for each sleeping receiver */
48 struct list_head r_list;
49 struct task_struct *r_tsk;
55 struct msg_msg *r_msg;
58 /* one msg_sender for each sleeping sender */
60 struct list_head list;
61 struct task_struct *tsk;
66 #define SEARCH_EQUAL 2
67 #define SEARCH_NOTEQUAL 3
68 #define SEARCH_LESSEQUAL 4
69 #define SEARCH_NUMBER 5
71 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
73 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
75 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
78 return ERR_CAST(ipcp);
80 return container_of(ipcp, struct msg_queue, q_perm);
83 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
86 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
89 return ERR_CAST(ipcp);
91 return container_of(ipcp, struct msg_queue, q_perm);
94 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
96 ipc_rmid(&msg_ids(ns), &s->q_perm);
99 static void msg_rcu_free(struct rcu_head *head)
101 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
102 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
104 security_msg_queue_free(msq);
109 * newque - Create a new msg queue
111 * @params: ptr to the structure that contains the key and msgflg
113 * Called with msg_ids.rwsem held (writer)
115 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
117 struct msg_queue *msq;
119 key_t key = params->key;
120 int msgflg = params->flg;
122 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
126 msq->q_perm.mode = msgflg & S_IRWXUGO;
127 msq->q_perm.key = key;
129 msq->q_perm.security = NULL;
130 retval = security_msg_queue_alloc(msq);
136 msq->q_stime = msq->q_rtime = 0;
137 msq->q_ctime = ktime_get_real_seconds();
138 msq->q_cbytes = msq->q_qnum = 0;
139 msq->q_qbytes = ns->msg_ctlmnb;
140 msq->q_lspid = msq->q_lrpid = 0;
141 INIT_LIST_HEAD(&msq->q_messages);
142 INIT_LIST_HEAD(&msq->q_receivers);
143 INIT_LIST_HEAD(&msq->q_senders);
145 /* ipc_addid() locks msq upon success. */
146 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
148 call_rcu(&msq->q_perm.rcu, msg_rcu_free);
152 ipc_unlock_object(&msq->q_perm);
155 return msq->q_perm.id;
158 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
160 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
161 1 + msq->q_qnum <= msq->q_qbytes;
164 static inline void ss_add(struct msg_queue *msq,
165 struct msg_sender *mss, size_t msgsz)
169 __set_current_state(TASK_INTERRUPTIBLE);
170 list_add_tail(&mss->list, &msq->q_senders);
173 static inline void ss_del(struct msg_sender *mss)
176 list_del(&mss->list);
179 static void ss_wakeup(struct msg_queue *msq,
180 struct wake_q_head *wake_q, bool kill)
182 struct msg_sender *mss, *t;
183 struct task_struct *stop_tsk = NULL;
184 struct list_head *h = &msq->q_senders;
186 list_for_each_entry_safe(mss, t, h, list) {
188 mss->list.next = NULL;
191 * Stop at the first task we don't wakeup,
192 * we've already iterated the original
195 else if (stop_tsk == mss->tsk)
198 * We are not in an EIDRM scenario here, therefore
199 * verify that we really need to wakeup the task.
200 * To maintain current semantics and wakeup order,
201 * move the sender to the tail on behalf of the
204 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
208 list_move_tail(&mss->list, &msq->q_senders);
212 wake_q_add(wake_q, mss->tsk);
216 static void expunge_all(struct msg_queue *msq, int res,
217 struct wake_q_head *wake_q)
219 struct msg_receiver *msr, *t;
221 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
222 wake_q_add(wake_q, msr->r_tsk);
223 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
228 * freeque() wakes up waiters on the sender and receiver waiting queue,
229 * removes the message queue from message queue ID IDR, and cleans up all the
230 * messages associated with this queue.
232 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
233 * before freeque() is called. msg_ids.rwsem remains locked on exit.
235 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
237 struct msg_msg *msg, *t;
238 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
239 DEFINE_WAKE_Q(wake_q);
241 expunge_all(msq, -EIDRM, &wake_q);
242 ss_wakeup(msq, &wake_q, true);
244 ipc_unlock_object(&msq->q_perm);
248 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
249 atomic_dec(&ns->msg_hdrs);
252 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
253 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
257 * Called with msg_ids.rwsem and ipcp locked.
259 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
261 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
263 return security_msg_queue_associate(msq, msgflg);
266 long ksys_msgget(key_t key, int msgflg)
268 struct ipc_namespace *ns;
269 static const struct ipc_ops msg_ops = {
271 .associate = msg_security,
273 struct ipc_params msg_params;
275 ns = current->nsproxy->ipc_ns;
277 msg_params.key = key;
278 msg_params.flg = msgflg;
280 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
283 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
285 return ksys_msgget(key, msgflg);
288 static inline unsigned long
289 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
293 return copy_to_user(buf, in, sizeof(*in));
298 memset(&out, 0, sizeof(out));
300 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
302 out.msg_stime = in->msg_stime;
303 out.msg_rtime = in->msg_rtime;
304 out.msg_ctime = in->msg_ctime;
306 if (in->msg_cbytes > USHRT_MAX)
307 out.msg_cbytes = USHRT_MAX;
309 out.msg_cbytes = in->msg_cbytes;
310 out.msg_lcbytes = in->msg_cbytes;
312 if (in->msg_qnum > USHRT_MAX)
313 out.msg_qnum = USHRT_MAX;
315 out.msg_qnum = in->msg_qnum;
317 if (in->msg_qbytes > USHRT_MAX)
318 out.msg_qbytes = USHRT_MAX;
320 out.msg_qbytes = in->msg_qbytes;
321 out.msg_lqbytes = in->msg_qbytes;
323 out.msg_lspid = in->msg_lspid;
324 out.msg_lrpid = in->msg_lrpid;
326 return copy_to_user(buf, &out, sizeof(out));
333 static inline unsigned long
334 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
338 if (copy_from_user(out, buf, sizeof(*out)))
343 struct msqid_ds tbuf_old;
345 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
348 out->msg_perm.uid = tbuf_old.msg_perm.uid;
349 out->msg_perm.gid = tbuf_old.msg_perm.gid;
350 out->msg_perm.mode = tbuf_old.msg_perm.mode;
352 if (tbuf_old.msg_qbytes == 0)
353 out->msg_qbytes = tbuf_old.msg_lqbytes;
355 out->msg_qbytes = tbuf_old.msg_qbytes;
365 * This function handles some msgctl commands which require the rwsem
366 * to be held in write mode.
367 * NOTE: no locks must be held, the rwsem is taken inside this function.
369 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
370 struct msqid64_ds *msqid64)
372 struct kern_ipc_perm *ipcp;
373 struct msg_queue *msq;
376 down_write(&msg_ids(ns).rwsem);
379 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
380 &msqid64->msg_perm, msqid64->msg_qbytes);
386 msq = container_of(ipcp, struct msg_queue, q_perm);
388 err = security_msg_queue_msgctl(msq, cmd);
394 ipc_lock_object(&msq->q_perm);
395 /* freeque unlocks the ipc object and rcu */
400 DEFINE_WAKE_Q(wake_q);
402 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
403 !capable(CAP_SYS_RESOURCE)) {
408 ipc_lock_object(&msq->q_perm);
409 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
413 msq->q_qbytes = msqid64->msg_qbytes;
415 msq->q_ctime = ktime_get_real_seconds();
417 * Sleeping receivers might be excluded by
418 * stricter permissions.
420 expunge_all(msq, -EAGAIN, &wake_q);
422 * Sleeping senders might be able to send
423 * due to a larger queue size.
425 ss_wakeup(msq, &wake_q, false);
426 ipc_unlock_object(&msq->q_perm);
437 ipc_unlock_object(&msq->q_perm);
441 up_write(&msg_ids(ns).rwsem);
445 static int msgctl_info(struct ipc_namespace *ns, int msqid,
446 int cmd, struct msginfo *msginfo)
452 * We must not return kernel stack data.
453 * due to padding, it's not enough
454 * to set all member fields.
456 err = security_msg_queue_msgctl(NULL, cmd);
460 memset(msginfo, 0, sizeof(*msginfo));
461 msginfo->msgmni = ns->msg_ctlmni;
462 msginfo->msgmax = ns->msg_ctlmax;
463 msginfo->msgmnb = ns->msg_ctlmnb;
464 msginfo->msgssz = MSGSSZ;
465 msginfo->msgseg = MSGSEG;
466 down_read(&msg_ids(ns).rwsem);
467 if (cmd == MSG_INFO) {
468 msginfo->msgpool = msg_ids(ns).in_use;
469 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
470 msginfo->msgtql = atomic_read(&ns->msg_bytes);
472 msginfo->msgmap = MSGMAP;
473 msginfo->msgpool = MSGPOOL;
474 msginfo->msgtql = MSGTQL;
476 max_id = ipc_get_maxid(&msg_ids(ns));
477 up_read(&msg_ids(ns).rwsem);
478 return (max_id < 0) ? 0 : max_id;
481 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
482 int cmd, struct msqid64_ds *p)
484 struct msg_queue *msq;
488 memset(p, 0, sizeof(*p));
491 if (cmd == MSG_STAT) {
492 msq = msq_obtain_object(ns, msqid);
499 msq = msq_obtain_object_check(ns, msqid);
507 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
510 err = security_msg_queue_msgctl(msq, cmd);
514 ipc_lock_object(&msq->q_perm);
516 if (!ipc_valid_object(&msq->q_perm)) {
517 ipc_unlock_object(&msq->q_perm);
522 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
523 p->msg_stime = msq->q_stime;
524 p->msg_rtime = msq->q_rtime;
525 p->msg_ctime = msq->q_ctime;
526 p->msg_cbytes = msq->q_cbytes;
527 p->msg_qnum = msq->q_qnum;
528 p->msg_qbytes = msq->q_qbytes;
529 p->msg_lspid = msq->q_lspid;
530 p->msg_lrpid = msq->q_lrpid;
532 ipc_unlock_object(&msq->q_perm);
541 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
544 struct ipc_namespace *ns;
545 struct msqid64_ds msqid64;
548 if (msqid < 0 || cmd < 0)
551 version = ipc_parse_version(&cmd);
552 ns = current->nsproxy->ipc_ns;
557 struct msginfo msginfo;
558 err = msgctl_info(ns, msqid, cmd, &msginfo);
561 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
565 case MSG_STAT: /* msqid is an index rather than a msg queue id */
567 err = msgctl_stat(ns, msqid, cmd, &msqid64);
570 if (copy_msqid_to_user(buf, &msqid64, version))
574 if (copy_msqid_from_user(&msqid64, buf, version))
578 return msgctl_down(ns, msqid, cmd, &msqid64);
584 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
586 return ksys_msgctl(msqid, cmd, buf);
591 struct compat_msqid_ds {
592 struct compat_ipc_perm msg_perm;
593 compat_uptr_t msg_first;
594 compat_uptr_t msg_last;
595 compat_time_t msg_stime;
596 compat_time_t msg_rtime;
597 compat_time_t msg_ctime;
598 compat_ulong_t msg_lcbytes;
599 compat_ulong_t msg_lqbytes;
600 unsigned short msg_cbytes;
601 unsigned short msg_qnum;
602 unsigned short msg_qbytes;
603 compat_ipc_pid_t msg_lspid;
604 compat_ipc_pid_t msg_lrpid;
607 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
610 memset(out, 0, sizeof(*out));
611 if (version == IPC_64) {
612 struct compat_msqid64_ds __user *p = buf;
613 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
615 if (get_user(out->msg_qbytes, &p->msg_qbytes))
618 struct compat_msqid_ds __user *p = buf;
619 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
621 if (get_user(out->msg_qbytes, &p->msg_qbytes))
627 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
630 if (version == IPC_64) {
631 struct compat_msqid64_ds v;
632 memset(&v, 0, sizeof(v));
633 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
634 v.msg_stime = in->msg_stime;
635 v.msg_rtime = in->msg_rtime;
636 v.msg_ctime = in->msg_ctime;
637 v.msg_cbytes = in->msg_cbytes;
638 v.msg_qnum = in->msg_qnum;
639 v.msg_qbytes = in->msg_qbytes;
640 v.msg_lspid = in->msg_lspid;
641 v.msg_lrpid = in->msg_lrpid;
642 return copy_to_user(buf, &v, sizeof(v));
644 struct compat_msqid_ds v;
645 memset(&v, 0, sizeof(v));
646 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
647 v.msg_stime = in->msg_stime;
648 v.msg_rtime = in->msg_rtime;
649 v.msg_ctime = in->msg_ctime;
650 v.msg_cbytes = in->msg_cbytes;
651 v.msg_qnum = in->msg_qnum;
652 v.msg_qbytes = in->msg_qbytes;
653 v.msg_lspid = in->msg_lspid;
654 v.msg_lrpid = in->msg_lrpid;
655 return copy_to_user(buf, &v, sizeof(v));
659 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
661 struct ipc_namespace *ns;
663 struct msqid64_ds msqid64;
664 int version = compat_ipc_parse_version(&cmd);
666 ns = current->nsproxy->ipc_ns;
668 if (msqid < 0 || cmd < 0)
671 switch (cmd & (~IPC_64)) {
674 struct msginfo msginfo;
675 err = msgctl_info(ns, msqid, cmd, &msginfo);
678 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
684 err = msgctl_stat(ns, msqid, cmd, &msqid64);
687 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
691 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
695 return msgctl_down(ns, msqid, cmd, &msqid64);
701 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
703 return compat_ksys_msgctl(msqid, cmd, uptr);
707 static int testmsg(struct msg_msg *msg, long type, int mode)
713 case SEARCH_LESSEQUAL:
714 if (msg->m_type <= type)
718 if (msg->m_type == type)
721 case SEARCH_NOTEQUAL:
722 if (msg->m_type != type)
729 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
730 struct wake_q_head *wake_q)
732 struct msg_receiver *msr, *t;
734 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
735 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
736 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
737 msr->r_msgtype, msr->r_mode)) {
739 list_del(&msr->r_list);
740 if (msr->r_maxsize < msg->m_ts) {
741 wake_q_add(wake_q, msr->r_tsk);
742 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
744 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
745 msq->q_rtime = get_seconds();
747 wake_q_add(wake_q, msr->r_tsk);
748 WRITE_ONCE(msr->r_msg, msg);
757 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
758 size_t msgsz, int msgflg)
760 struct msg_queue *msq;
763 struct ipc_namespace *ns;
764 DEFINE_WAKE_Q(wake_q);
766 ns = current->nsproxy->ipc_ns;
768 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
773 msg = load_msg(mtext, msgsz);
781 msq = msq_obtain_object_check(ns, msqid);
787 ipc_lock_object(&msq->q_perm);
793 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
796 /* raced with RMID? */
797 if (!ipc_valid_object(&msq->q_perm)) {
802 err = security_msg_queue_msgsnd(msq, msg, msgflg);
806 if (msg_fits_inqueue(msq, msgsz))
809 /* queue full, wait: */
810 if (msgflg & IPC_NOWAIT) {
815 /* enqueue the sender and prepare to block */
816 ss_add(msq, &s, msgsz);
818 if (!ipc_rcu_getref(&msq->q_perm)) {
823 ipc_unlock_object(&msq->q_perm);
828 ipc_lock_object(&msq->q_perm);
830 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
831 /* raced with RMID? */
832 if (!ipc_valid_object(&msq->q_perm)) {
838 if (signal_pending(current)) {
839 err = -ERESTARTNOHAND;
845 msq->q_lspid = task_tgid_vnr(current);
846 msq->q_stime = get_seconds();
848 if (!pipelined_send(msq, msg, &wake_q)) {
849 /* no one is waiting for this message, enqueue it */
850 list_add_tail(&msg->m_list, &msq->q_messages);
851 msq->q_cbytes += msgsz;
853 atomic_add(msgsz, &ns->msg_bytes);
854 atomic_inc(&ns->msg_hdrs);
861 ipc_unlock_object(&msq->q_perm);
870 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
875 if (get_user(mtype, &msgp->mtype))
877 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
882 struct compat_msgbuf {
887 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
888 compat_ssize_t, msgsz, int, msgflg)
890 struct compat_msgbuf __user *up = compat_ptr(msgp);
893 if (get_user(mtype, &up->mtype))
895 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
899 static inline int convert_mode(long *msgtyp, int msgflg)
901 if (msgflg & MSG_COPY)
902 return SEARCH_NUMBER;
904 * find message of correct type.
905 * msgtyp = 0 => get first.
906 * msgtyp > 0 => get first message of matching type.
907 * msgtyp < 0 => get message with least type must be < abs(msgtype).
912 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
916 return SEARCH_LESSEQUAL;
918 if (msgflg & MSG_EXCEPT)
919 return SEARCH_NOTEQUAL;
923 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
925 struct msgbuf __user *msgp = dest;
928 if (put_user(msg->m_type, &msgp->mtype))
931 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
932 if (store_msg(msgp->mtext, msg, msgsz))
937 #ifdef CONFIG_CHECKPOINT_RESTORE
939 * This function creates new kernel message structure, large enough to store
940 * bufsz message bytes.
942 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
944 struct msg_msg *copy;
947 * Create dummy message to copy real message to.
949 copy = load_msg(buf, bufsz);
955 static inline void free_copy(struct msg_msg *copy)
961 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
963 return ERR_PTR(-ENOSYS);
966 static inline void free_copy(struct msg_msg *copy)
971 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
973 struct msg_msg *msg, *found = NULL;
976 list_for_each_entry(msg, &msq->q_messages, m_list) {
977 if (testmsg(msg, *msgtyp, mode) &&
978 !security_msg_queue_msgrcv(msq, msg, current,
980 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
981 *msgtyp = msg->m_type - 1;
983 } else if (mode == SEARCH_NUMBER) {
984 if (*msgtyp == count)
992 return found ?: ERR_PTR(-EAGAIN);
995 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
996 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
999 struct msg_queue *msq;
1000 struct ipc_namespace *ns;
1001 struct msg_msg *msg, *copy = NULL;
1002 DEFINE_WAKE_Q(wake_q);
1004 ns = current->nsproxy->ipc_ns;
1006 if (msqid < 0 || (long) bufsz < 0)
1009 if (msgflg & MSG_COPY) {
1010 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1012 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1014 return PTR_ERR(copy);
1016 mode = convert_mode(&msgtyp, msgflg);
1019 msq = msq_obtain_object_check(ns, msqid);
1023 return PTR_ERR(msq);
1027 struct msg_receiver msr_d;
1029 msg = ERR_PTR(-EACCES);
1030 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1033 ipc_lock_object(&msq->q_perm);
1035 /* raced with RMID? */
1036 if (!ipc_valid_object(&msq->q_perm)) {
1037 msg = ERR_PTR(-EIDRM);
1041 msg = find_msg(msq, &msgtyp, mode);
1044 * Found a suitable message.
1045 * Unlink it from the queue.
1047 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1048 msg = ERR_PTR(-E2BIG);
1052 * If we are copying, then do not unlink message and do
1053 * not update queue parameters.
1055 if (msgflg & MSG_COPY) {
1056 msg = copy_msg(msg, copy);
1060 list_del(&msg->m_list);
1062 msq->q_rtime = get_seconds();
1063 msq->q_lrpid = task_tgid_vnr(current);
1064 msq->q_cbytes -= msg->m_ts;
1065 atomic_sub(msg->m_ts, &ns->msg_bytes);
1066 atomic_dec(&ns->msg_hdrs);
1067 ss_wakeup(msq, &wake_q, false);
1072 /* No message waiting. Wait for a message */
1073 if (msgflg & IPC_NOWAIT) {
1074 msg = ERR_PTR(-ENOMSG);
1078 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1079 msr_d.r_tsk = current;
1080 msr_d.r_msgtype = msgtyp;
1081 msr_d.r_mode = mode;
1082 if (msgflg & MSG_NOERROR)
1083 msr_d.r_maxsize = INT_MAX;
1085 msr_d.r_maxsize = bufsz;
1086 msr_d.r_msg = ERR_PTR(-EAGAIN);
1087 __set_current_state(TASK_INTERRUPTIBLE);
1089 ipc_unlock_object(&msq->q_perm);
1094 * Lockless receive, part 1:
1095 * We don't hold a reference to the queue and getting a
1096 * reference would defeat the idea of a lockless operation,
1097 * thus the code relies on rcu to guarantee the existence of
1099 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1100 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1105 * Lockless receive, part 2:
1106 * The work in pipelined_send() and expunge_all():
1107 * - Set pointer to message
1108 * - Queue the receiver task for later wakeup
1109 * - Wake up the process after the lock is dropped.
1111 * Should the process wake up before this wakeup (due to a
1112 * signal) it will either see the message and continue ...
1114 msg = READ_ONCE(msr_d.r_msg);
1115 if (msg != ERR_PTR(-EAGAIN))
1119 * ... or see -EAGAIN, acquire the lock to check the message
1122 ipc_lock_object(&msq->q_perm);
1125 if (msg != ERR_PTR(-EAGAIN))
1128 list_del(&msr_d.r_list);
1129 if (signal_pending(current)) {
1130 msg = ERR_PTR(-ERESTARTNOHAND);
1134 ipc_unlock_object(&msq->q_perm);
1138 ipc_unlock_object(&msq->q_perm);
1144 return PTR_ERR(msg);
1147 bufsz = msg_handler(buf, msg, bufsz);
1153 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1154 long, msgtyp, int, msgflg)
1156 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1159 #ifdef CONFIG_COMPAT
1160 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1162 struct compat_msgbuf __user *msgp = dest;
1165 if (put_user(msg->m_type, &msgp->mtype))
1168 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1169 if (store_msg(msgp->mtext, msg, msgsz))
1174 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1175 compat_ssize_t, msgsz, compat_long_t, msgtyp, int, msgflg)
1177 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1178 msgflg, compat_do_msg_fill);
1182 int msg_init_ns(struct ipc_namespace *ns)
1184 ns->msg_ctlmax = MSGMAX;
1185 ns->msg_ctlmnb = MSGMNB;
1186 ns->msg_ctlmni = MSGMNI;
1188 atomic_set(&ns->msg_bytes, 0);
1189 atomic_set(&ns->msg_hdrs, 0);
1190 return ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1193 #ifdef CONFIG_IPC_NS
1194 void msg_exit_ns(struct ipc_namespace *ns)
1196 free_ipcs(ns, &msg_ids(ns), freeque);
1197 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1198 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1202 #ifdef CONFIG_PROC_FS
1203 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1205 struct user_namespace *user_ns = seq_user_ns(s);
1206 struct kern_ipc_perm *ipcp = it;
1207 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1210 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1218 from_kuid_munged(user_ns, msq->q_perm.uid),
1219 from_kgid_munged(user_ns, msq->q_perm.gid),
1220 from_kuid_munged(user_ns, msq->q_perm.cuid),
1221 from_kgid_munged(user_ns, msq->q_perm.cgid),
1230 int __init msg_init(void)
1232 const int err = msg_init_ns(&init_ipc_ns);
1234 ipc_init_proc_interface("sysvipc/msg",
1235 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1236 IPC_MSG_IDS, sysvipc_msg_proc_show);