Merge tag 'locking-urgent-2020-12-27' of git://git.kernel.org/pub/scm/linux/kernel...
[linux/fpc-iii.git] / ipc / msg.c
blobacd1bc7af55a26b7d680348ef576e0e5ed51a83d
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/ipc/msg.c
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>
21 * namespaces support
22 * OpenVZ, SWsoft Inc.
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>
30 #include <linux/mm.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>
45 #include "util.h"
47 /* one msq_queue structure for each present queue on the system */
48 struct msg_queue {
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;
62 } __randomize_layout;
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 */
75 struct msg_receiver {
76 struct list_head r_list;
77 struct task_struct *r_tsk;
79 int r_mode;
80 long r_msgtype;
81 long r_maxsize;
83 struct msg_msg *r_msg;
86 /* one msg_sender for each sleeping sender */
87 struct msg_sender {
88 struct list_head list;
89 struct task_struct *tsk;
90 size_t msgsz;
93 #define SEARCH_ANY 1
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);
105 if (IS_ERR(ipcp))
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,
112 int id)
114 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
116 if (IS_ERR(ipcp))
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);
133 kvfree(msq);
137 * newque - Create a new msg queue
138 * @ns: namespace
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;
146 int retval;
147 key_t key = params->key;
148 int msgflg = params->flg;
150 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
151 if (unlikely(!msq))
152 return -ENOMEM;
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);
159 if (retval) {
160 kvfree(msq);
161 return retval;
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);
175 if (retval < 0) {
176 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
177 return retval;
180 ipc_unlock_object(&msq->q_perm);
181 rcu_read_unlock();
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)
195 mss->tsk = current;
196 mss->msgsz = 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)
207 if (mss->list.next)
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) {
219 if (kill)
220 mss->list.next = NULL;
223 * Stop at the first task we don't wakeup,
224 * we've already iterated the original
225 * sender queue.
227 else if (stop_tsk == mss->tsk)
228 break;
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
234 * blocked task.
236 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
237 if (!stop_tsk)
238 stop_tsk = mss->tsk;
240 list_move_tail(&mss->list, &msq->q_senders);
241 continue;
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)
271 __releases(RCU)
272 __releases(&msq->q_perm)
274 struct msg_msg *msg, *t;
275 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
276 DEFINE_WAKE_Q(wake_q);
278 expunge_all(msq, -EIDRM, &wake_q);
279 ss_wakeup(msq, &wake_q, true);
280 msg_rmid(ns, msq);
281 ipc_unlock_object(&msq->q_perm);
282 wake_up_q(&wake_q);
283 rcu_read_unlock();
285 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
286 atomic_dec(&ns->msg_hdrs);
287 free_msg(msg);
289 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
290 ipc_update_pid(&msq->q_lspid, NULL);
291 ipc_update_pid(&msq->q_lrpid, NULL);
292 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
295 long ksys_msgget(key_t key, int msgflg)
297 struct ipc_namespace *ns;
298 static const struct ipc_ops msg_ops = {
299 .getnew = newque,
300 .associate = security_msg_queue_associate,
302 struct ipc_params msg_params;
304 ns = current->nsproxy->ipc_ns;
306 msg_params.key = key;
307 msg_params.flg = msgflg;
309 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
312 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
314 return ksys_msgget(key, msgflg);
317 static inline unsigned long
318 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
320 switch (version) {
321 case IPC_64:
322 return copy_to_user(buf, in, sizeof(*in));
323 case IPC_OLD:
325 struct msqid_ds out;
327 memset(&out, 0, sizeof(out));
329 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
331 out.msg_stime = in->msg_stime;
332 out.msg_rtime = in->msg_rtime;
333 out.msg_ctime = in->msg_ctime;
335 if (in->msg_cbytes > USHRT_MAX)
336 out.msg_cbytes = USHRT_MAX;
337 else
338 out.msg_cbytes = in->msg_cbytes;
339 out.msg_lcbytes = in->msg_cbytes;
341 if (in->msg_qnum > USHRT_MAX)
342 out.msg_qnum = USHRT_MAX;
343 else
344 out.msg_qnum = in->msg_qnum;
346 if (in->msg_qbytes > USHRT_MAX)
347 out.msg_qbytes = USHRT_MAX;
348 else
349 out.msg_qbytes = in->msg_qbytes;
350 out.msg_lqbytes = in->msg_qbytes;
352 out.msg_lspid = in->msg_lspid;
353 out.msg_lrpid = in->msg_lrpid;
355 return copy_to_user(buf, &out, sizeof(out));
357 default:
358 return -EINVAL;
362 static inline unsigned long
363 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
365 switch (version) {
366 case IPC_64:
367 if (copy_from_user(out, buf, sizeof(*out)))
368 return -EFAULT;
369 return 0;
370 case IPC_OLD:
372 struct msqid_ds tbuf_old;
374 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
375 return -EFAULT;
377 out->msg_perm.uid = tbuf_old.msg_perm.uid;
378 out->msg_perm.gid = tbuf_old.msg_perm.gid;
379 out->msg_perm.mode = tbuf_old.msg_perm.mode;
381 if (tbuf_old.msg_qbytes == 0)
382 out->msg_qbytes = tbuf_old.msg_lqbytes;
383 else
384 out->msg_qbytes = tbuf_old.msg_qbytes;
386 return 0;
388 default:
389 return -EINVAL;
394 * This function handles some msgctl commands which require the rwsem
395 * to be held in write mode.
396 * NOTE: no locks must be held, the rwsem is taken inside this function.
398 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
399 struct ipc64_perm *perm, int msg_qbytes)
401 struct kern_ipc_perm *ipcp;
402 struct msg_queue *msq;
403 int err;
405 down_write(&msg_ids(ns).rwsem);
406 rcu_read_lock();
408 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
409 perm, msg_qbytes);
410 if (IS_ERR(ipcp)) {
411 err = PTR_ERR(ipcp);
412 goto out_unlock1;
415 msq = container_of(ipcp, struct msg_queue, q_perm);
417 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
418 if (err)
419 goto out_unlock1;
421 switch (cmd) {
422 case IPC_RMID:
423 ipc_lock_object(&msq->q_perm);
424 /* freeque unlocks the ipc object and rcu */
425 freeque(ns, ipcp);
426 goto out_up;
427 case IPC_SET:
429 DEFINE_WAKE_Q(wake_q);
431 if (msg_qbytes > ns->msg_ctlmnb &&
432 !capable(CAP_SYS_RESOURCE)) {
433 err = -EPERM;
434 goto out_unlock1;
437 ipc_lock_object(&msq->q_perm);
438 err = ipc_update_perm(perm, ipcp);
439 if (err)
440 goto out_unlock0;
442 msq->q_qbytes = msg_qbytes;
444 msq->q_ctime = ktime_get_real_seconds();
446 * Sleeping receivers might be excluded by
447 * stricter permissions.
449 expunge_all(msq, -EAGAIN, &wake_q);
451 * Sleeping senders might be able to send
452 * due to a larger queue size.
454 ss_wakeup(msq, &wake_q, false);
455 ipc_unlock_object(&msq->q_perm);
456 wake_up_q(&wake_q);
458 goto out_unlock1;
460 default:
461 err = -EINVAL;
462 goto out_unlock1;
465 out_unlock0:
466 ipc_unlock_object(&msq->q_perm);
467 out_unlock1:
468 rcu_read_unlock();
469 out_up:
470 up_write(&msg_ids(ns).rwsem);
471 return err;
474 static int msgctl_info(struct ipc_namespace *ns, int msqid,
475 int cmd, struct msginfo *msginfo)
477 int err;
478 int max_idx;
481 * We must not return kernel stack data.
482 * due to padding, it's not enough
483 * to set all member fields.
485 err = security_msg_queue_msgctl(NULL, cmd);
486 if (err)
487 return err;
489 memset(msginfo, 0, sizeof(*msginfo));
490 msginfo->msgmni = ns->msg_ctlmni;
491 msginfo->msgmax = ns->msg_ctlmax;
492 msginfo->msgmnb = ns->msg_ctlmnb;
493 msginfo->msgssz = MSGSSZ;
494 msginfo->msgseg = MSGSEG;
495 down_read(&msg_ids(ns).rwsem);
496 if (cmd == MSG_INFO) {
497 msginfo->msgpool = msg_ids(ns).in_use;
498 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
499 msginfo->msgtql = atomic_read(&ns->msg_bytes);
500 } else {
501 msginfo->msgmap = MSGMAP;
502 msginfo->msgpool = MSGPOOL;
503 msginfo->msgtql = MSGTQL;
505 max_idx = ipc_get_maxidx(&msg_ids(ns));
506 up_read(&msg_ids(ns).rwsem);
507 return (max_idx < 0) ? 0 : max_idx;
510 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
511 int cmd, struct msqid64_ds *p)
513 struct msg_queue *msq;
514 int err;
516 memset(p, 0, sizeof(*p));
518 rcu_read_lock();
519 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
520 msq = msq_obtain_object(ns, msqid);
521 if (IS_ERR(msq)) {
522 err = PTR_ERR(msq);
523 goto out_unlock;
525 } else { /* IPC_STAT */
526 msq = msq_obtain_object_check(ns, msqid);
527 if (IS_ERR(msq)) {
528 err = PTR_ERR(msq);
529 goto out_unlock;
533 /* see comment for SHM_STAT_ANY */
534 if (cmd == MSG_STAT_ANY)
535 audit_ipc_obj(&msq->q_perm);
536 else {
537 err = -EACCES;
538 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
539 goto out_unlock;
542 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
543 if (err)
544 goto out_unlock;
546 ipc_lock_object(&msq->q_perm);
548 if (!ipc_valid_object(&msq->q_perm)) {
549 ipc_unlock_object(&msq->q_perm);
550 err = -EIDRM;
551 goto out_unlock;
554 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
555 p->msg_stime = msq->q_stime;
556 p->msg_rtime = msq->q_rtime;
557 p->msg_ctime = msq->q_ctime;
558 #ifndef CONFIG_64BIT
559 p->msg_stime_high = msq->q_stime >> 32;
560 p->msg_rtime_high = msq->q_rtime >> 32;
561 p->msg_ctime_high = msq->q_ctime >> 32;
562 #endif
563 p->msg_cbytes = msq->q_cbytes;
564 p->msg_qnum = msq->q_qnum;
565 p->msg_qbytes = msq->q_qbytes;
566 p->msg_lspid = pid_vnr(msq->q_lspid);
567 p->msg_lrpid = pid_vnr(msq->q_lrpid);
569 if (cmd == IPC_STAT) {
571 * As defined in SUS:
572 * Return 0 on success
574 err = 0;
575 } else {
577 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
578 * Return the full id, including the sequence number
580 err = msq->q_perm.id;
583 ipc_unlock_object(&msq->q_perm);
584 out_unlock:
585 rcu_read_unlock();
586 return err;
589 static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
591 struct ipc_namespace *ns;
592 struct msqid64_ds msqid64;
593 int err;
595 if (msqid < 0 || cmd < 0)
596 return -EINVAL;
598 ns = current->nsproxy->ipc_ns;
600 switch (cmd) {
601 case IPC_INFO:
602 case MSG_INFO: {
603 struct msginfo msginfo;
604 err = msgctl_info(ns, msqid, cmd, &msginfo);
605 if (err < 0)
606 return err;
607 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
608 err = -EFAULT;
609 return err;
611 case MSG_STAT: /* msqid is an index rather than a msg queue id */
612 case MSG_STAT_ANY:
613 case IPC_STAT:
614 err = msgctl_stat(ns, msqid, cmd, &msqid64);
615 if (err < 0)
616 return err;
617 if (copy_msqid_to_user(buf, &msqid64, version))
618 err = -EFAULT;
619 return err;
620 case IPC_SET:
621 if (copy_msqid_from_user(&msqid64, buf, version))
622 return -EFAULT;
623 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
624 msqid64.msg_qbytes);
625 case IPC_RMID:
626 return msgctl_down(ns, msqid, cmd, NULL, 0);
627 default:
628 return -EINVAL;
632 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
634 return ksys_msgctl(msqid, cmd, buf, IPC_64);
637 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
638 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
640 int version = ipc_parse_version(&cmd);
642 return ksys_msgctl(msqid, cmd, buf, version);
645 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
647 return ksys_old_msgctl(msqid, cmd, buf);
649 #endif
651 #ifdef CONFIG_COMPAT
653 struct compat_msqid_ds {
654 struct compat_ipc_perm msg_perm;
655 compat_uptr_t msg_first;
656 compat_uptr_t msg_last;
657 old_time32_t msg_stime;
658 old_time32_t msg_rtime;
659 old_time32_t msg_ctime;
660 compat_ulong_t msg_lcbytes;
661 compat_ulong_t msg_lqbytes;
662 unsigned short msg_cbytes;
663 unsigned short msg_qnum;
664 unsigned short msg_qbytes;
665 compat_ipc_pid_t msg_lspid;
666 compat_ipc_pid_t msg_lrpid;
669 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
670 int version)
672 memset(out, 0, sizeof(*out));
673 if (version == IPC_64) {
674 struct compat_msqid64_ds __user *p = buf;
675 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
676 return -EFAULT;
677 if (get_user(out->msg_qbytes, &p->msg_qbytes))
678 return -EFAULT;
679 } else {
680 struct compat_msqid_ds __user *p = buf;
681 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
682 return -EFAULT;
683 if (get_user(out->msg_qbytes, &p->msg_qbytes))
684 return -EFAULT;
686 return 0;
689 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
690 int version)
692 if (version == IPC_64) {
693 struct compat_msqid64_ds v;
694 memset(&v, 0, sizeof(v));
695 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
696 v.msg_stime = lower_32_bits(in->msg_stime);
697 v.msg_stime_high = upper_32_bits(in->msg_stime);
698 v.msg_rtime = lower_32_bits(in->msg_rtime);
699 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
700 v.msg_ctime = lower_32_bits(in->msg_ctime);
701 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
702 v.msg_cbytes = in->msg_cbytes;
703 v.msg_qnum = in->msg_qnum;
704 v.msg_qbytes = in->msg_qbytes;
705 v.msg_lspid = in->msg_lspid;
706 v.msg_lrpid = in->msg_lrpid;
707 return copy_to_user(buf, &v, sizeof(v));
708 } else {
709 struct compat_msqid_ds v;
710 memset(&v, 0, sizeof(v));
711 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
712 v.msg_stime = in->msg_stime;
713 v.msg_rtime = in->msg_rtime;
714 v.msg_ctime = in->msg_ctime;
715 v.msg_cbytes = in->msg_cbytes;
716 v.msg_qnum = in->msg_qnum;
717 v.msg_qbytes = in->msg_qbytes;
718 v.msg_lspid = in->msg_lspid;
719 v.msg_lrpid = in->msg_lrpid;
720 return copy_to_user(buf, &v, sizeof(v));
724 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
726 struct ipc_namespace *ns;
727 int err;
728 struct msqid64_ds msqid64;
730 ns = current->nsproxy->ipc_ns;
732 if (msqid < 0 || cmd < 0)
733 return -EINVAL;
735 switch (cmd & (~IPC_64)) {
736 case IPC_INFO:
737 case MSG_INFO: {
738 struct msginfo msginfo;
739 err = msgctl_info(ns, msqid, cmd, &msginfo);
740 if (err < 0)
741 return err;
742 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
743 err = -EFAULT;
744 return err;
746 case IPC_STAT:
747 case MSG_STAT:
748 case MSG_STAT_ANY:
749 err = msgctl_stat(ns, msqid, cmd, &msqid64);
750 if (err < 0)
751 return err;
752 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
753 err = -EFAULT;
754 return err;
755 case IPC_SET:
756 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
757 return -EFAULT;
758 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
759 case IPC_RMID:
760 return msgctl_down(ns, msqid, cmd, NULL, 0);
761 default:
762 return -EINVAL;
766 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
768 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
771 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
772 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
774 int version = compat_ipc_parse_version(&cmd);
776 return compat_ksys_msgctl(msqid, cmd, uptr, version);
779 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
781 return compat_ksys_old_msgctl(msqid, cmd, uptr);
783 #endif
784 #endif
786 static int testmsg(struct msg_msg *msg, long type, int mode)
788 switch (mode) {
789 case SEARCH_ANY:
790 case SEARCH_NUMBER:
791 return 1;
792 case SEARCH_LESSEQUAL:
793 if (msg->m_type <= type)
794 return 1;
795 break;
796 case SEARCH_EQUAL:
797 if (msg->m_type == type)
798 return 1;
799 break;
800 case SEARCH_NOTEQUAL:
801 if (msg->m_type != type)
802 return 1;
803 break;
805 return 0;
808 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
809 struct wake_q_head *wake_q)
811 struct msg_receiver *msr, *t;
813 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
814 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
815 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
816 msr->r_msgtype, msr->r_mode)) {
818 list_del(&msr->r_list);
819 if (msr->r_maxsize < msg->m_ts) {
820 wake_q_add(wake_q, msr->r_tsk);
822 /* See expunge_all regarding memory barrier */
823 smp_store_release(&msr->r_msg, ERR_PTR(-E2BIG));
824 } else {
825 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
826 msq->q_rtime = ktime_get_real_seconds();
828 wake_q_add(wake_q, msr->r_tsk);
830 /* See expunge_all regarding memory barrier */
831 smp_store_release(&msr->r_msg, msg);
832 return 1;
837 return 0;
840 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
841 size_t msgsz, int msgflg)
843 struct msg_queue *msq;
844 struct msg_msg *msg;
845 int err;
846 struct ipc_namespace *ns;
847 DEFINE_WAKE_Q(wake_q);
849 ns = current->nsproxy->ipc_ns;
851 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
852 return -EINVAL;
853 if (mtype < 1)
854 return -EINVAL;
856 msg = load_msg(mtext, msgsz);
857 if (IS_ERR(msg))
858 return PTR_ERR(msg);
860 msg->m_type = mtype;
861 msg->m_ts = msgsz;
863 rcu_read_lock();
864 msq = msq_obtain_object_check(ns, msqid);
865 if (IS_ERR(msq)) {
866 err = PTR_ERR(msq);
867 goto out_unlock1;
870 ipc_lock_object(&msq->q_perm);
872 for (;;) {
873 struct msg_sender s;
875 err = -EACCES;
876 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
877 goto out_unlock0;
879 /* raced with RMID? */
880 if (!ipc_valid_object(&msq->q_perm)) {
881 err = -EIDRM;
882 goto out_unlock0;
885 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
886 if (err)
887 goto out_unlock0;
889 if (msg_fits_inqueue(msq, msgsz))
890 break;
892 /* queue full, wait: */
893 if (msgflg & IPC_NOWAIT) {
894 err = -EAGAIN;
895 goto out_unlock0;
898 /* enqueue the sender and prepare to block */
899 ss_add(msq, &s, msgsz);
901 if (!ipc_rcu_getref(&msq->q_perm)) {
902 err = -EIDRM;
903 goto out_unlock0;
906 ipc_unlock_object(&msq->q_perm);
907 rcu_read_unlock();
908 schedule();
910 rcu_read_lock();
911 ipc_lock_object(&msq->q_perm);
913 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
914 /* raced with RMID? */
915 if (!ipc_valid_object(&msq->q_perm)) {
916 err = -EIDRM;
917 goto out_unlock0;
919 ss_del(&s);
921 if (signal_pending(current)) {
922 err = -ERESTARTNOHAND;
923 goto out_unlock0;
928 ipc_update_pid(&msq->q_lspid, task_tgid(current));
929 msq->q_stime = ktime_get_real_seconds();
931 if (!pipelined_send(msq, msg, &wake_q)) {
932 /* no one is waiting for this message, enqueue it */
933 list_add_tail(&msg->m_list, &msq->q_messages);
934 msq->q_cbytes += msgsz;
935 msq->q_qnum++;
936 atomic_add(msgsz, &ns->msg_bytes);
937 atomic_inc(&ns->msg_hdrs);
940 err = 0;
941 msg = NULL;
943 out_unlock0:
944 ipc_unlock_object(&msq->q_perm);
945 wake_up_q(&wake_q);
946 out_unlock1:
947 rcu_read_unlock();
948 if (msg != NULL)
949 free_msg(msg);
950 return err;
953 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
954 int msgflg)
956 long mtype;
958 if (get_user(mtype, &msgp->mtype))
959 return -EFAULT;
960 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
963 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
964 int, msgflg)
966 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
969 #ifdef CONFIG_COMPAT
971 struct compat_msgbuf {
972 compat_long_t mtype;
973 char mtext[1];
976 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
977 compat_ssize_t msgsz, int msgflg)
979 struct compat_msgbuf __user *up = compat_ptr(msgp);
980 compat_long_t mtype;
982 if (get_user(mtype, &up->mtype))
983 return -EFAULT;
984 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
987 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
988 compat_ssize_t, msgsz, int, msgflg)
990 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
992 #endif
994 static inline int convert_mode(long *msgtyp, int msgflg)
996 if (msgflg & MSG_COPY)
997 return SEARCH_NUMBER;
999 * find message of correct type.
1000 * msgtyp = 0 => get first.
1001 * msgtyp > 0 => get first message of matching type.
1002 * msgtyp < 0 => get message with least type must be < abs(msgtype).
1004 if (*msgtyp == 0)
1005 return SEARCH_ANY;
1006 if (*msgtyp < 0) {
1007 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
1008 *msgtyp = LONG_MAX;
1009 else
1010 *msgtyp = -*msgtyp;
1011 return SEARCH_LESSEQUAL;
1013 if (msgflg & MSG_EXCEPT)
1014 return SEARCH_NOTEQUAL;
1015 return SEARCH_EQUAL;
1018 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1020 struct msgbuf __user *msgp = dest;
1021 size_t msgsz;
1023 if (put_user(msg->m_type, &msgp->mtype))
1024 return -EFAULT;
1026 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1027 if (store_msg(msgp->mtext, msg, msgsz))
1028 return -EFAULT;
1029 return msgsz;
1032 #ifdef CONFIG_CHECKPOINT_RESTORE
1034 * This function creates new kernel message structure, large enough to store
1035 * bufsz message bytes.
1037 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1039 struct msg_msg *copy;
1042 * Create dummy message to copy real message to.
1044 copy = load_msg(buf, bufsz);
1045 if (!IS_ERR(copy))
1046 copy->m_ts = bufsz;
1047 return copy;
1050 static inline void free_copy(struct msg_msg *copy)
1052 if (copy)
1053 free_msg(copy);
1055 #else
1056 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1058 return ERR_PTR(-ENOSYS);
1061 static inline void free_copy(struct msg_msg *copy)
1064 #endif
1066 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1068 struct msg_msg *msg, *found = NULL;
1069 long count = 0;
1071 list_for_each_entry(msg, &msq->q_messages, m_list) {
1072 if (testmsg(msg, *msgtyp, mode) &&
1073 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1074 *msgtyp, mode)) {
1075 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1076 *msgtyp = msg->m_type - 1;
1077 found = msg;
1078 } else if (mode == SEARCH_NUMBER) {
1079 if (*msgtyp == count)
1080 return msg;
1081 } else
1082 return msg;
1083 count++;
1087 return found ?: ERR_PTR(-EAGAIN);
1090 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1091 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1093 int mode;
1094 struct msg_queue *msq;
1095 struct ipc_namespace *ns;
1096 struct msg_msg *msg, *copy = NULL;
1097 DEFINE_WAKE_Q(wake_q);
1099 ns = current->nsproxy->ipc_ns;
1101 if (msqid < 0 || (long) bufsz < 0)
1102 return -EINVAL;
1104 if (msgflg & MSG_COPY) {
1105 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1106 return -EINVAL;
1107 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1108 if (IS_ERR(copy))
1109 return PTR_ERR(copy);
1111 mode = convert_mode(&msgtyp, msgflg);
1113 rcu_read_lock();
1114 msq = msq_obtain_object_check(ns, msqid);
1115 if (IS_ERR(msq)) {
1116 rcu_read_unlock();
1117 free_copy(copy);
1118 return PTR_ERR(msq);
1121 for (;;) {
1122 struct msg_receiver msr_d;
1124 msg = ERR_PTR(-EACCES);
1125 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1126 goto out_unlock1;
1128 ipc_lock_object(&msq->q_perm);
1130 /* raced with RMID? */
1131 if (!ipc_valid_object(&msq->q_perm)) {
1132 msg = ERR_PTR(-EIDRM);
1133 goto out_unlock0;
1136 msg = find_msg(msq, &msgtyp, mode);
1137 if (!IS_ERR(msg)) {
1139 * Found a suitable message.
1140 * Unlink it from the queue.
1142 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1143 msg = ERR_PTR(-E2BIG);
1144 goto out_unlock0;
1147 * If we are copying, then do not unlink message and do
1148 * not update queue parameters.
1150 if (msgflg & MSG_COPY) {
1151 msg = copy_msg(msg, copy);
1152 goto out_unlock0;
1155 list_del(&msg->m_list);
1156 msq->q_qnum--;
1157 msq->q_rtime = ktime_get_real_seconds();
1158 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1159 msq->q_cbytes -= msg->m_ts;
1160 atomic_sub(msg->m_ts, &ns->msg_bytes);
1161 atomic_dec(&ns->msg_hdrs);
1162 ss_wakeup(msq, &wake_q, false);
1164 goto out_unlock0;
1167 /* No message waiting. Wait for a message */
1168 if (msgflg & IPC_NOWAIT) {
1169 msg = ERR_PTR(-ENOMSG);
1170 goto out_unlock0;
1173 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1174 msr_d.r_tsk = current;
1175 msr_d.r_msgtype = msgtyp;
1176 msr_d.r_mode = mode;
1177 if (msgflg & MSG_NOERROR)
1178 msr_d.r_maxsize = INT_MAX;
1179 else
1180 msr_d.r_maxsize = bufsz;
1182 /* memory barrier not require due to ipc_lock_object() */
1183 WRITE_ONCE(msr_d.r_msg, ERR_PTR(-EAGAIN));
1185 /* memory barrier not required, we own ipc_lock_object() */
1186 __set_current_state(TASK_INTERRUPTIBLE);
1188 ipc_unlock_object(&msq->q_perm);
1189 rcu_read_unlock();
1190 schedule();
1193 * Lockless receive, part 1:
1194 * We don't hold a reference to the queue and getting a
1195 * reference would defeat the idea of a lockless operation,
1196 * thus the code relies on rcu to guarantee the existence of
1197 * msq:
1198 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1199 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1201 rcu_read_lock();
1204 * Lockless receive, part 2:
1205 * The work in pipelined_send() and expunge_all():
1206 * - Set pointer to message
1207 * - Queue the receiver task for later wakeup
1208 * - Wake up the process after the lock is dropped.
1210 * Should the process wake up before this wakeup (due to a
1211 * signal) it will either see the message and continue ...
1213 msg = READ_ONCE(msr_d.r_msg);
1214 if (msg != ERR_PTR(-EAGAIN)) {
1215 /* see MSG_BARRIER for purpose/pairing */
1216 smp_acquire__after_ctrl_dep();
1218 goto out_unlock1;
1222 * ... or see -EAGAIN, acquire the lock to check the message
1223 * again.
1225 ipc_lock_object(&msq->q_perm);
1227 msg = READ_ONCE(msr_d.r_msg);
1228 if (msg != ERR_PTR(-EAGAIN))
1229 goto out_unlock0;
1231 list_del(&msr_d.r_list);
1232 if (signal_pending(current)) {
1233 msg = ERR_PTR(-ERESTARTNOHAND);
1234 goto out_unlock0;
1237 ipc_unlock_object(&msq->q_perm);
1240 out_unlock0:
1241 ipc_unlock_object(&msq->q_perm);
1242 wake_up_q(&wake_q);
1243 out_unlock1:
1244 rcu_read_unlock();
1245 if (IS_ERR(msg)) {
1246 free_copy(copy);
1247 return PTR_ERR(msg);
1250 bufsz = msg_handler(buf, msg, bufsz);
1251 free_msg(msg);
1253 return bufsz;
1256 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1257 long msgtyp, int msgflg)
1259 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1262 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1263 long, msgtyp, int, msgflg)
1265 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1268 #ifdef CONFIG_COMPAT
1269 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1271 struct compat_msgbuf __user *msgp = dest;
1272 size_t msgsz;
1274 if (put_user(msg->m_type, &msgp->mtype))
1275 return -EFAULT;
1277 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1278 if (store_msg(msgp->mtext, msg, msgsz))
1279 return -EFAULT;
1280 return msgsz;
1283 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1284 compat_long_t msgtyp, int msgflg)
1286 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1287 msgflg, compat_do_msg_fill);
1290 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1291 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1292 int, msgflg)
1294 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1296 #endif
1298 void msg_init_ns(struct ipc_namespace *ns)
1300 ns->msg_ctlmax = MSGMAX;
1301 ns->msg_ctlmnb = MSGMNB;
1302 ns->msg_ctlmni = MSGMNI;
1304 atomic_set(&ns->msg_bytes, 0);
1305 atomic_set(&ns->msg_hdrs, 0);
1306 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1309 #ifdef CONFIG_IPC_NS
1310 void msg_exit_ns(struct ipc_namespace *ns)
1312 free_ipcs(ns, &msg_ids(ns), freeque);
1313 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1314 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1316 #endif
1318 #ifdef CONFIG_PROC_FS
1319 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1321 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1322 struct user_namespace *user_ns = seq_user_ns(s);
1323 struct kern_ipc_perm *ipcp = it;
1324 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1326 seq_printf(s,
1327 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1328 msq->q_perm.key,
1329 msq->q_perm.id,
1330 msq->q_perm.mode,
1331 msq->q_cbytes,
1332 msq->q_qnum,
1333 pid_nr_ns(msq->q_lspid, pid_ns),
1334 pid_nr_ns(msq->q_lrpid, pid_ns),
1335 from_kuid_munged(user_ns, msq->q_perm.uid),
1336 from_kgid_munged(user_ns, msq->q_perm.gid),
1337 from_kuid_munged(user_ns, msq->q_perm.cuid),
1338 from_kgid_munged(user_ns, msq->q_perm.cgid),
1339 msq->q_stime,
1340 msq->q_rtime,
1341 msq->q_ctime);
1343 return 0;
1345 #endif
1347 void __init msg_init(void)
1349 msg_init_ns(&init_ipc_ns);
1351 ipc_init_proc_interface("sysvipc/msg",
1352 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1353 IPC_MSG_IDS, sysvipc_msg_proc_show);