call kthread_stop from workqueue, do not reset conn if connect cmd has been retransmitted
[cor.git] / ipc / msg.c
blob8dec945fa030b099b5becf08b4350b23852203d2
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;
64 /* one msg_receiver structure for each sleeping receiver */
65 struct msg_receiver {
66 struct list_head r_list;
67 struct task_struct *r_tsk;
69 int r_mode;
70 long r_msgtype;
71 long r_maxsize;
73 struct msg_msg *r_msg;
76 /* one msg_sender for each sleeping sender */
77 struct msg_sender {
78 struct list_head list;
79 struct task_struct *tsk;
80 size_t msgsz;
83 #define SEARCH_ANY 1
84 #define SEARCH_EQUAL 2
85 #define SEARCH_NOTEQUAL 3
86 #define SEARCH_LESSEQUAL 4
87 #define SEARCH_NUMBER 5
89 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
91 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
93 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&msg_ids(ns), id);
95 if (IS_ERR(ipcp))
96 return ERR_CAST(ipcp);
98 return container_of(ipcp, struct msg_queue, q_perm);
101 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
102 int id)
104 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
106 if (IS_ERR(ipcp))
107 return ERR_CAST(ipcp);
109 return container_of(ipcp, struct msg_queue, q_perm);
112 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
114 ipc_rmid(&msg_ids(ns), &s->q_perm);
117 static void msg_rcu_free(struct rcu_head *head)
119 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
120 struct msg_queue *msq = container_of(p, struct msg_queue, q_perm);
122 security_msg_queue_free(&msq->q_perm);
123 kvfree(msq);
127 * newque - Create a new msg queue
128 * @ns: namespace
129 * @params: ptr to the structure that contains the key and msgflg
131 * Called with msg_ids.rwsem held (writer)
133 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
135 struct msg_queue *msq;
136 int retval;
137 key_t key = params->key;
138 int msgflg = params->flg;
140 msq = kvmalloc(sizeof(*msq), GFP_KERNEL);
141 if (unlikely(!msq))
142 return -ENOMEM;
144 msq->q_perm.mode = msgflg & S_IRWXUGO;
145 msq->q_perm.key = key;
147 msq->q_perm.security = NULL;
148 retval = security_msg_queue_alloc(&msq->q_perm);
149 if (retval) {
150 kvfree(msq);
151 return retval;
154 msq->q_stime = msq->q_rtime = 0;
155 msq->q_ctime = ktime_get_real_seconds();
156 msq->q_cbytes = msq->q_qnum = 0;
157 msq->q_qbytes = ns->msg_ctlmnb;
158 msq->q_lspid = msq->q_lrpid = NULL;
159 INIT_LIST_HEAD(&msq->q_messages);
160 INIT_LIST_HEAD(&msq->q_receivers);
161 INIT_LIST_HEAD(&msq->q_senders);
163 /* ipc_addid() locks msq upon success. */
164 retval = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
165 if (retval < 0) {
166 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
167 return retval;
170 ipc_unlock_object(&msq->q_perm);
171 rcu_read_unlock();
173 return msq->q_perm.id;
176 static inline bool msg_fits_inqueue(struct msg_queue *msq, size_t msgsz)
178 return msgsz + msq->q_cbytes <= msq->q_qbytes &&
179 1 + msq->q_qnum <= msq->q_qbytes;
182 static inline void ss_add(struct msg_queue *msq,
183 struct msg_sender *mss, size_t msgsz)
185 mss->tsk = current;
186 mss->msgsz = msgsz;
187 __set_current_state(TASK_INTERRUPTIBLE);
188 list_add_tail(&mss->list, &msq->q_senders);
191 static inline void ss_del(struct msg_sender *mss)
193 if (mss->list.next)
194 list_del(&mss->list);
197 static void ss_wakeup(struct msg_queue *msq,
198 struct wake_q_head *wake_q, bool kill)
200 struct msg_sender *mss, *t;
201 struct task_struct *stop_tsk = NULL;
202 struct list_head *h = &msq->q_senders;
204 list_for_each_entry_safe(mss, t, h, list) {
205 if (kill)
206 mss->list.next = NULL;
209 * Stop at the first task we don't wakeup,
210 * we've already iterated the original
211 * sender queue.
213 else if (stop_tsk == mss->tsk)
214 break;
216 * We are not in an EIDRM scenario here, therefore
217 * verify that we really need to wakeup the task.
218 * To maintain current semantics and wakeup order,
219 * move the sender to the tail on behalf of the
220 * blocked task.
222 else if (!msg_fits_inqueue(msq, mss->msgsz)) {
223 if (!stop_tsk)
224 stop_tsk = mss->tsk;
226 list_move_tail(&mss->list, &msq->q_senders);
227 continue;
230 wake_q_add(wake_q, mss->tsk);
234 static void expunge_all(struct msg_queue *msq, int res,
235 struct wake_q_head *wake_q)
237 struct msg_receiver *msr, *t;
239 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
240 wake_q_add(wake_q, msr->r_tsk);
241 WRITE_ONCE(msr->r_msg, ERR_PTR(res));
246 * freeque() wakes up waiters on the sender and receiver waiting queue,
247 * removes the message queue from message queue ID IDR, and cleans up all the
248 * messages associated with this queue.
250 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
251 * before freeque() is called. msg_ids.rwsem remains locked on exit.
253 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
255 struct msg_msg *msg, *t;
256 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
257 DEFINE_WAKE_Q(wake_q);
259 expunge_all(msq, -EIDRM, &wake_q);
260 ss_wakeup(msq, &wake_q, true);
261 msg_rmid(ns, msq);
262 ipc_unlock_object(&msq->q_perm);
263 wake_up_q(&wake_q);
264 rcu_read_unlock();
266 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
267 atomic_dec(&ns->msg_hdrs);
268 free_msg(msg);
270 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
271 ipc_update_pid(&msq->q_lspid, NULL);
272 ipc_update_pid(&msq->q_lrpid, NULL);
273 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
276 long ksys_msgget(key_t key, int msgflg)
278 struct ipc_namespace *ns;
279 static const struct ipc_ops msg_ops = {
280 .getnew = newque,
281 .associate = security_msg_queue_associate,
283 struct ipc_params msg_params;
285 ns = current->nsproxy->ipc_ns;
287 msg_params.key = key;
288 msg_params.flg = msgflg;
290 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
293 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
295 return ksys_msgget(key, msgflg);
298 static inline unsigned long
299 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
301 switch (version) {
302 case IPC_64:
303 return copy_to_user(buf, in, sizeof(*in));
304 case IPC_OLD:
306 struct msqid_ds out;
308 memset(&out, 0, sizeof(out));
310 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
312 out.msg_stime = in->msg_stime;
313 out.msg_rtime = in->msg_rtime;
314 out.msg_ctime = in->msg_ctime;
316 if (in->msg_cbytes > USHRT_MAX)
317 out.msg_cbytes = USHRT_MAX;
318 else
319 out.msg_cbytes = in->msg_cbytes;
320 out.msg_lcbytes = in->msg_cbytes;
322 if (in->msg_qnum > USHRT_MAX)
323 out.msg_qnum = USHRT_MAX;
324 else
325 out.msg_qnum = in->msg_qnum;
327 if (in->msg_qbytes > USHRT_MAX)
328 out.msg_qbytes = USHRT_MAX;
329 else
330 out.msg_qbytes = in->msg_qbytes;
331 out.msg_lqbytes = in->msg_qbytes;
333 out.msg_lspid = in->msg_lspid;
334 out.msg_lrpid = in->msg_lrpid;
336 return copy_to_user(buf, &out, sizeof(out));
338 default:
339 return -EINVAL;
343 static inline unsigned long
344 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
346 switch (version) {
347 case IPC_64:
348 if (copy_from_user(out, buf, sizeof(*out)))
349 return -EFAULT;
350 return 0;
351 case IPC_OLD:
353 struct msqid_ds tbuf_old;
355 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
356 return -EFAULT;
358 out->msg_perm.uid = tbuf_old.msg_perm.uid;
359 out->msg_perm.gid = tbuf_old.msg_perm.gid;
360 out->msg_perm.mode = tbuf_old.msg_perm.mode;
362 if (tbuf_old.msg_qbytes == 0)
363 out->msg_qbytes = tbuf_old.msg_lqbytes;
364 else
365 out->msg_qbytes = tbuf_old.msg_qbytes;
367 return 0;
369 default:
370 return -EINVAL;
375 * This function handles some msgctl commands which require the rwsem
376 * to be held in write mode.
377 * NOTE: no locks must be held, the rwsem is taken inside this function.
379 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
380 struct msqid64_ds *msqid64)
382 struct kern_ipc_perm *ipcp;
383 struct msg_queue *msq;
384 int err;
386 down_write(&msg_ids(ns).rwsem);
387 rcu_read_lock();
389 ipcp = ipcctl_obtain_check(ns, &msg_ids(ns), msqid, cmd,
390 &msqid64->msg_perm, msqid64->msg_qbytes);
391 if (IS_ERR(ipcp)) {
392 err = PTR_ERR(ipcp);
393 goto out_unlock1;
396 msq = container_of(ipcp, struct msg_queue, q_perm);
398 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
399 if (err)
400 goto out_unlock1;
402 switch (cmd) {
403 case IPC_RMID:
404 ipc_lock_object(&msq->q_perm);
405 /* freeque unlocks the ipc object and rcu */
406 freeque(ns, ipcp);
407 goto out_up;
408 case IPC_SET:
410 DEFINE_WAKE_Q(wake_q);
412 if (msqid64->msg_qbytes > ns->msg_ctlmnb &&
413 !capable(CAP_SYS_RESOURCE)) {
414 err = -EPERM;
415 goto out_unlock1;
418 ipc_lock_object(&msq->q_perm);
419 err = ipc_update_perm(&msqid64->msg_perm, ipcp);
420 if (err)
421 goto out_unlock0;
423 msq->q_qbytes = msqid64->msg_qbytes;
425 msq->q_ctime = ktime_get_real_seconds();
427 * Sleeping receivers might be excluded by
428 * stricter permissions.
430 expunge_all(msq, -EAGAIN, &wake_q);
432 * Sleeping senders might be able to send
433 * due to a larger queue size.
435 ss_wakeup(msq, &wake_q, false);
436 ipc_unlock_object(&msq->q_perm);
437 wake_up_q(&wake_q);
439 goto out_unlock1;
441 default:
442 err = -EINVAL;
443 goto out_unlock1;
446 out_unlock0:
447 ipc_unlock_object(&msq->q_perm);
448 out_unlock1:
449 rcu_read_unlock();
450 out_up:
451 up_write(&msg_ids(ns).rwsem);
452 return err;
455 static int msgctl_info(struct ipc_namespace *ns, int msqid,
456 int cmd, struct msginfo *msginfo)
458 int err;
459 int max_idx;
462 * We must not return kernel stack data.
463 * due to padding, it's not enough
464 * to set all member fields.
466 err = security_msg_queue_msgctl(NULL, cmd);
467 if (err)
468 return err;
470 memset(msginfo, 0, sizeof(*msginfo));
471 msginfo->msgmni = ns->msg_ctlmni;
472 msginfo->msgmax = ns->msg_ctlmax;
473 msginfo->msgmnb = ns->msg_ctlmnb;
474 msginfo->msgssz = MSGSSZ;
475 msginfo->msgseg = MSGSEG;
476 down_read(&msg_ids(ns).rwsem);
477 if (cmd == MSG_INFO) {
478 msginfo->msgpool = msg_ids(ns).in_use;
479 msginfo->msgmap = atomic_read(&ns->msg_hdrs);
480 msginfo->msgtql = atomic_read(&ns->msg_bytes);
481 } else {
482 msginfo->msgmap = MSGMAP;
483 msginfo->msgpool = MSGPOOL;
484 msginfo->msgtql = MSGTQL;
486 max_idx = ipc_get_maxidx(&msg_ids(ns));
487 up_read(&msg_ids(ns).rwsem);
488 return (max_idx < 0) ? 0 : max_idx;
491 static int msgctl_stat(struct ipc_namespace *ns, int msqid,
492 int cmd, struct msqid64_ds *p)
494 struct msg_queue *msq;
495 int err;
497 memset(p, 0, sizeof(*p));
499 rcu_read_lock();
500 if (cmd == MSG_STAT || cmd == MSG_STAT_ANY) {
501 msq = msq_obtain_object(ns, msqid);
502 if (IS_ERR(msq)) {
503 err = PTR_ERR(msq);
504 goto out_unlock;
506 } else { /* IPC_STAT */
507 msq = msq_obtain_object_check(ns, msqid);
508 if (IS_ERR(msq)) {
509 err = PTR_ERR(msq);
510 goto out_unlock;
514 /* see comment for SHM_STAT_ANY */
515 if (cmd == MSG_STAT_ANY)
516 audit_ipc_obj(&msq->q_perm);
517 else {
518 err = -EACCES;
519 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
520 goto out_unlock;
523 err = security_msg_queue_msgctl(&msq->q_perm, cmd);
524 if (err)
525 goto out_unlock;
527 ipc_lock_object(&msq->q_perm);
529 if (!ipc_valid_object(&msq->q_perm)) {
530 ipc_unlock_object(&msq->q_perm);
531 err = -EIDRM;
532 goto out_unlock;
535 kernel_to_ipc64_perm(&msq->q_perm, &p->msg_perm);
536 p->msg_stime = msq->q_stime;
537 p->msg_rtime = msq->q_rtime;
538 p->msg_ctime = msq->q_ctime;
539 #ifndef CONFIG_64BIT
540 p->msg_stime_high = msq->q_stime >> 32;
541 p->msg_rtime_high = msq->q_rtime >> 32;
542 p->msg_ctime_high = msq->q_ctime >> 32;
543 #endif
544 p->msg_cbytes = msq->q_cbytes;
545 p->msg_qnum = msq->q_qnum;
546 p->msg_qbytes = msq->q_qbytes;
547 p->msg_lspid = pid_vnr(msq->q_lspid);
548 p->msg_lrpid = pid_vnr(msq->q_lrpid);
550 if (cmd == IPC_STAT) {
552 * As defined in SUS:
553 * Return 0 on success
555 err = 0;
556 } else {
558 * MSG_STAT and MSG_STAT_ANY (both Linux specific)
559 * Return the full id, including the sequence number
561 err = msq->q_perm.id;
564 ipc_unlock_object(&msq->q_perm);
565 out_unlock:
566 rcu_read_unlock();
567 return err;
570 static long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf, int version)
572 struct ipc_namespace *ns;
573 struct msqid64_ds msqid64;
574 int err;
576 if (msqid < 0 || cmd < 0)
577 return -EINVAL;
579 ns = current->nsproxy->ipc_ns;
581 switch (cmd) {
582 case IPC_INFO:
583 case MSG_INFO: {
584 struct msginfo msginfo;
585 err = msgctl_info(ns, msqid, cmd, &msginfo);
586 if (err < 0)
587 return err;
588 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
589 err = -EFAULT;
590 return err;
592 case MSG_STAT: /* msqid is an index rather than a msg queue id */
593 case MSG_STAT_ANY:
594 case IPC_STAT:
595 err = msgctl_stat(ns, msqid, cmd, &msqid64);
596 if (err < 0)
597 return err;
598 if (copy_msqid_to_user(buf, &msqid64, version))
599 err = -EFAULT;
600 return err;
601 case IPC_SET:
602 if (copy_msqid_from_user(&msqid64, buf, version))
603 return -EFAULT;
604 /* fallthru */
605 case IPC_RMID:
606 return msgctl_down(ns, msqid, cmd, &msqid64);
607 default:
608 return -EINVAL;
612 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
614 return ksys_msgctl(msqid, cmd, buf, IPC_64);
617 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
618 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
620 int version = ipc_parse_version(&cmd);
622 return ksys_msgctl(msqid, cmd, buf, version);
625 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
627 return ksys_old_msgctl(msqid, cmd, buf);
629 #endif
631 #ifdef CONFIG_COMPAT
633 struct compat_msqid_ds {
634 struct compat_ipc_perm msg_perm;
635 compat_uptr_t msg_first;
636 compat_uptr_t msg_last;
637 old_time32_t msg_stime;
638 old_time32_t msg_rtime;
639 old_time32_t msg_ctime;
640 compat_ulong_t msg_lcbytes;
641 compat_ulong_t msg_lqbytes;
642 unsigned short msg_cbytes;
643 unsigned short msg_qnum;
644 unsigned short msg_qbytes;
645 compat_ipc_pid_t msg_lspid;
646 compat_ipc_pid_t msg_lrpid;
649 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
650 int version)
652 memset(out, 0, sizeof(*out));
653 if (version == IPC_64) {
654 struct compat_msqid64_ds __user *p = buf;
655 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
656 return -EFAULT;
657 if (get_user(out->msg_qbytes, &p->msg_qbytes))
658 return -EFAULT;
659 } else {
660 struct compat_msqid_ds __user *p = buf;
661 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
662 return -EFAULT;
663 if (get_user(out->msg_qbytes, &p->msg_qbytes))
664 return -EFAULT;
666 return 0;
669 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
670 int version)
672 if (version == IPC_64) {
673 struct compat_msqid64_ds v;
674 memset(&v, 0, sizeof(v));
675 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
676 v.msg_stime = lower_32_bits(in->msg_stime);
677 v.msg_stime_high = upper_32_bits(in->msg_stime);
678 v.msg_rtime = lower_32_bits(in->msg_rtime);
679 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
680 v.msg_ctime = lower_32_bits(in->msg_ctime);
681 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
682 v.msg_cbytes = in->msg_cbytes;
683 v.msg_qnum = in->msg_qnum;
684 v.msg_qbytes = in->msg_qbytes;
685 v.msg_lspid = in->msg_lspid;
686 v.msg_lrpid = in->msg_lrpid;
687 return copy_to_user(buf, &v, sizeof(v));
688 } else {
689 struct compat_msqid_ds v;
690 memset(&v, 0, sizeof(v));
691 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
692 v.msg_stime = in->msg_stime;
693 v.msg_rtime = in->msg_rtime;
694 v.msg_ctime = in->msg_ctime;
695 v.msg_cbytes = in->msg_cbytes;
696 v.msg_qnum = in->msg_qnum;
697 v.msg_qbytes = in->msg_qbytes;
698 v.msg_lspid = in->msg_lspid;
699 v.msg_lrpid = in->msg_lrpid;
700 return copy_to_user(buf, &v, sizeof(v));
704 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
706 struct ipc_namespace *ns;
707 int err;
708 struct msqid64_ds msqid64;
710 ns = current->nsproxy->ipc_ns;
712 if (msqid < 0 || cmd < 0)
713 return -EINVAL;
715 switch (cmd & (~IPC_64)) {
716 case IPC_INFO:
717 case MSG_INFO: {
718 struct msginfo msginfo;
719 err = msgctl_info(ns, msqid, cmd, &msginfo);
720 if (err < 0)
721 return err;
722 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
723 err = -EFAULT;
724 return err;
726 case IPC_STAT:
727 case MSG_STAT:
728 case MSG_STAT_ANY:
729 err = msgctl_stat(ns, msqid, cmd, &msqid64);
730 if (err < 0)
731 return err;
732 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
733 err = -EFAULT;
734 return err;
735 case IPC_SET:
736 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
737 return -EFAULT;
738 /* fallthru */
739 case IPC_RMID:
740 return msgctl_down(ns, msqid, cmd, &msqid64);
741 default:
742 return -EINVAL;
746 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
748 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
751 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
752 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
754 int version = compat_ipc_parse_version(&cmd);
756 return compat_ksys_msgctl(msqid, cmd, uptr, version);
759 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
761 return compat_ksys_old_msgctl(msqid, cmd, uptr);
763 #endif
764 #endif
766 static int testmsg(struct msg_msg *msg, long type, int mode)
768 switch (mode) {
769 case SEARCH_ANY:
770 case SEARCH_NUMBER:
771 return 1;
772 case SEARCH_LESSEQUAL:
773 if (msg->m_type <= type)
774 return 1;
775 break;
776 case SEARCH_EQUAL:
777 if (msg->m_type == type)
778 return 1;
779 break;
780 case SEARCH_NOTEQUAL:
781 if (msg->m_type != type)
782 return 1;
783 break;
785 return 0;
788 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
789 struct wake_q_head *wake_q)
791 struct msg_receiver *msr, *t;
793 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
794 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
795 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
796 msr->r_msgtype, msr->r_mode)) {
798 list_del(&msr->r_list);
799 if (msr->r_maxsize < msg->m_ts) {
800 wake_q_add(wake_q, msr->r_tsk);
801 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
802 } else {
803 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
804 msq->q_rtime = ktime_get_real_seconds();
806 wake_q_add(wake_q, msr->r_tsk);
807 WRITE_ONCE(msr->r_msg, msg);
808 return 1;
813 return 0;
816 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
817 size_t msgsz, int msgflg)
819 struct msg_queue *msq;
820 struct msg_msg *msg;
821 int err;
822 struct ipc_namespace *ns;
823 DEFINE_WAKE_Q(wake_q);
825 ns = current->nsproxy->ipc_ns;
827 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
828 return -EINVAL;
829 if (mtype < 1)
830 return -EINVAL;
832 msg = load_msg(mtext, msgsz);
833 if (IS_ERR(msg))
834 return PTR_ERR(msg);
836 msg->m_type = mtype;
837 msg->m_ts = msgsz;
839 rcu_read_lock();
840 msq = msq_obtain_object_check(ns, msqid);
841 if (IS_ERR(msq)) {
842 err = PTR_ERR(msq);
843 goto out_unlock1;
846 ipc_lock_object(&msq->q_perm);
848 for (;;) {
849 struct msg_sender s;
851 err = -EACCES;
852 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
853 goto out_unlock0;
855 /* raced with RMID? */
856 if (!ipc_valid_object(&msq->q_perm)) {
857 err = -EIDRM;
858 goto out_unlock0;
861 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
862 if (err)
863 goto out_unlock0;
865 if (msg_fits_inqueue(msq, msgsz))
866 break;
868 /* queue full, wait: */
869 if (msgflg & IPC_NOWAIT) {
870 err = -EAGAIN;
871 goto out_unlock0;
874 /* enqueue the sender and prepare to block */
875 ss_add(msq, &s, msgsz);
877 if (!ipc_rcu_getref(&msq->q_perm)) {
878 err = -EIDRM;
879 goto out_unlock0;
882 ipc_unlock_object(&msq->q_perm);
883 rcu_read_unlock();
884 schedule();
886 rcu_read_lock();
887 ipc_lock_object(&msq->q_perm);
889 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
890 /* raced with RMID? */
891 if (!ipc_valid_object(&msq->q_perm)) {
892 err = -EIDRM;
893 goto out_unlock0;
895 ss_del(&s);
897 if (signal_pending(current)) {
898 err = -ERESTARTNOHAND;
899 goto out_unlock0;
904 ipc_update_pid(&msq->q_lspid, task_tgid(current));
905 msq->q_stime = ktime_get_real_seconds();
907 if (!pipelined_send(msq, msg, &wake_q)) {
908 /* no one is waiting for this message, enqueue it */
909 list_add_tail(&msg->m_list, &msq->q_messages);
910 msq->q_cbytes += msgsz;
911 msq->q_qnum++;
912 atomic_add(msgsz, &ns->msg_bytes);
913 atomic_inc(&ns->msg_hdrs);
916 err = 0;
917 msg = NULL;
919 out_unlock0:
920 ipc_unlock_object(&msq->q_perm);
921 wake_up_q(&wake_q);
922 out_unlock1:
923 rcu_read_unlock();
924 if (msg != NULL)
925 free_msg(msg);
926 return err;
929 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
930 int msgflg)
932 long mtype;
934 if (get_user(mtype, &msgp->mtype))
935 return -EFAULT;
936 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
939 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
940 int, msgflg)
942 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
945 #ifdef CONFIG_COMPAT
947 struct compat_msgbuf {
948 compat_long_t mtype;
949 char mtext[1];
952 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
953 compat_ssize_t msgsz, int msgflg)
955 struct compat_msgbuf __user *up = compat_ptr(msgp);
956 compat_long_t mtype;
958 if (get_user(mtype, &up->mtype))
959 return -EFAULT;
960 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
963 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
964 compat_ssize_t, msgsz, int, msgflg)
966 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
968 #endif
970 static inline int convert_mode(long *msgtyp, int msgflg)
972 if (msgflg & MSG_COPY)
973 return SEARCH_NUMBER;
975 * find message of correct type.
976 * msgtyp = 0 => get first.
977 * msgtyp > 0 => get first message of matching type.
978 * msgtyp < 0 => get message with least type must be < abs(msgtype).
980 if (*msgtyp == 0)
981 return SEARCH_ANY;
982 if (*msgtyp < 0) {
983 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
984 *msgtyp = LONG_MAX;
985 else
986 *msgtyp = -*msgtyp;
987 return SEARCH_LESSEQUAL;
989 if (msgflg & MSG_EXCEPT)
990 return SEARCH_NOTEQUAL;
991 return SEARCH_EQUAL;
994 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
996 struct msgbuf __user *msgp = dest;
997 size_t msgsz;
999 if (put_user(msg->m_type, &msgp->mtype))
1000 return -EFAULT;
1002 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1003 if (store_msg(msgp->mtext, msg, msgsz))
1004 return -EFAULT;
1005 return msgsz;
1008 #ifdef CONFIG_CHECKPOINT_RESTORE
1010 * This function creates new kernel message structure, large enough to store
1011 * bufsz message bytes.
1013 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1015 struct msg_msg *copy;
1018 * Create dummy message to copy real message to.
1020 copy = load_msg(buf, bufsz);
1021 if (!IS_ERR(copy))
1022 copy->m_ts = bufsz;
1023 return copy;
1026 static inline void free_copy(struct msg_msg *copy)
1028 if (copy)
1029 free_msg(copy);
1031 #else
1032 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1034 return ERR_PTR(-ENOSYS);
1037 static inline void free_copy(struct msg_msg *copy)
1040 #endif
1042 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1044 struct msg_msg *msg, *found = NULL;
1045 long count = 0;
1047 list_for_each_entry(msg, &msq->q_messages, m_list) {
1048 if (testmsg(msg, *msgtyp, mode) &&
1049 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1050 *msgtyp, mode)) {
1051 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1052 *msgtyp = msg->m_type - 1;
1053 found = msg;
1054 } else if (mode == SEARCH_NUMBER) {
1055 if (*msgtyp == count)
1056 return msg;
1057 } else
1058 return msg;
1059 count++;
1063 return found ?: ERR_PTR(-EAGAIN);
1066 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1067 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1069 int mode;
1070 struct msg_queue *msq;
1071 struct ipc_namespace *ns;
1072 struct msg_msg *msg, *copy = NULL;
1073 DEFINE_WAKE_Q(wake_q);
1075 ns = current->nsproxy->ipc_ns;
1077 if (msqid < 0 || (long) bufsz < 0)
1078 return -EINVAL;
1080 if (msgflg & MSG_COPY) {
1081 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1082 return -EINVAL;
1083 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1084 if (IS_ERR(copy))
1085 return PTR_ERR(copy);
1087 mode = convert_mode(&msgtyp, msgflg);
1089 rcu_read_lock();
1090 msq = msq_obtain_object_check(ns, msqid);
1091 if (IS_ERR(msq)) {
1092 rcu_read_unlock();
1093 free_copy(copy);
1094 return PTR_ERR(msq);
1097 for (;;) {
1098 struct msg_receiver msr_d;
1100 msg = ERR_PTR(-EACCES);
1101 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1102 goto out_unlock1;
1104 ipc_lock_object(&msq->q_perm);
1106 /* raced with RMID? */
1107 if (!ipc_valid_object(&msq->q_perm)) {
1108 msg = ERR_PTR(-EIDRM);
1109 goto out_unlock0;
1112 msg = find_msg(msq, &msgtyp, mode);
1113 if (!IS_ERR(msg)) {
1115 * Found a suitable message.
1116 * Unlink it from the queue.
1118 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1119 msg = ERR_PTR(-E2BIG);
1120 goto out_unlock0;
1123 * If we are copying, then do not unlink message and do
1124 * not update queue parameters.
1126 if (msgflg & MSG_COPY) {
1127 msg = copy_msg(msg, copy);
1128 goto out_unlock0;
1131 list_del(&msg->m_list);
1132 msq->q_qnum--;
1133 msq->q_rtime = ktime_get_real_seconds();
1134 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1135 msq->q_cbytes -= msg->m_ts;
1136 atomic_sub(msg->m_ts, &ns->msg_bytes);
1137 atomic_dec(&ns->msg_hdrs);
1138 ss_wakeup(msq, &wake_q, false);
1140 goto out_unlock0;
1143 /* No message waiting. Wait for a message */
1144 if (msgflg & IPC_NOWAIT) {
1145 msg = ERR_PTR(-ENOMSG);
1146 goto out_unlock0;
1149 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1150 msr_d.r_tsk = current;
1151 msr_d.r_msgtype = msgtyp;
1152 msr_d.r_mode = mode;
1153 if (msgflg & MSG_NOERROR)
1154 msr_d.r_maxsize = INT_MAX;
1155 else
1156 msr_d.r_maxsize = bufsz;
1157 msr_d.r_msg = ERR_PTR(-EAGAIN);
1158 __set_current_state(TASK_INTERRUPTIBLE);
1160 ipc_unlock_object(&msq->q_perm);
1161 rcu_read_unlock();
1162 schedule();
1165 * Lockless receive, part 1:
1166 * We don't hold a reference to the queue and getting a
1167 * reference would defeat the idea of a lockless operation,
1168 * thus the code relies on rcu to guarantee the existence of
1169 * msq:
1170 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1171 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1173 rcu_read_lock();
1176 * Lockless receive, part 2:
1177 * The work in pipelined_send() and expunge_all():
1178 * - Set pointer to message
1179 * - Queue the receiver task for later wakeup
1180 * - Wake up the process after the lock is dropped.
1182 * Should the process wake up before this wakeup (due to a
1183 * signal) it will either see the message and continue ...
1185 msg = READ_ONCE(msr_d.r_msg);
1186 if (msg != ERR_PTR(-EAGAIN))
1187 goto out_unlock1;
1190 * ... or see -EAGAIN, acquire the lock to check the message
1191 * again.
1193 ipc_lock_object(&msq->q_perm);
1195 msg = msr_d.r_msg;
1196 if (msg != ERR_PTR(-EAGAIN))
1197 goto out_unlock0;
1199 list_del(&msr_d.r_list);
1200 if (signal_pending(current)) {
1201 msg = ERR_PTR(-ERESTARTNOHAND);
1202 goto out_unlock0;
1205 ipc_unlock_object(&msq->q_perm);
1208 out_unlock0:
1209 ipc_unlock_object(&msq->q_perm);
1210 wake_up_q(&wake_q);
1211 out_unlock1:
1212 rcu_read_unlock();
1213 if (IS_ERR(msg)) {
1214 free_copy(copy);
1215 return PTR_ERR(msg);
1218 bufsz = msg_handler(buf, msg, bufsz);
1219 free_msg(msg);
1221 return bufsz;
1224 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1225 long msgtyp, int msgflg)
1227 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1230 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1231 long, msgtyp, int, msgflg)
1233 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1236 #ifdef CONFIG_COMPAT
1237 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1239 struct compat_msgbuf __user *msgp = dest;
1240 size_t msgsz;
1242 if (put_user(msg->m_type, &msgp->mtype))
1243 return -EFAULT;
1245 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1246 if (store_msg(msgp->mtext, msg, msgsz))
1247 return -EFAULT;
1248 return msgsz;
1251 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1252 compat_long_t msgtyp, int msgflg)
1254 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1255 msgflg, compat_do_msg_fill);
1258 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1259 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1260 int, msgflg)
1262 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1264 #endif
1266 void msg_init_ns(struct ipc_namespace *ns)
1268 ns->msg_ctlmax = MSGMAX;
1269 ns->msg_ctlmnb = MSGMNB;
1270 ns->msg_ctlmni = MSGMNI;
1272 atomic_set(&ns->msg_bytes, 0);
1273 atomic_set(&ns->msg_hdrs, 0);
1274 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1277 #ifdef CONFIG_IPC_NS
1278 void msg_exit_ns(struct ipc_namespace *ns)
1280 free_ipcs(ns, &msg_ids(ns), freeque);
1281 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1282 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1284 #endif
1286 #ifdef CONFIG_PROC_FS
1287 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1289 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1290 struct user_namespace *user_ns = seq_user_ns(s);
1291 struct kern_ipc_perm *ipcp = it;
1292 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1294 seq_printf(s,
1295 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1296 msq->q_perm.key,
1297 msq->q_perm.id,
1298 msq->q_perm.mode,
1299 msq->q_cbytes,
1300 msq->q_qnum,
1301 pid_nr_ns(msq->q_lspid, pid_ns),
1302 pid_nr_ns(msq->q_lrpid, pid_ns),
1303 from_kuid_munged(user_ns, msq->q_perm.uid),
1304 from_kgid_munged(user_ns, msq->q_perm.gid),
1305 from_kuid_munged(user_ns, msq->q_perm.cuid),
1306 from_kgid_munged(user_ns, msq->q_perm.cgid),
1307 msq->q_stime,
1308 msq->q_rtime,
1309 msq->q_ctime);
1311 return 0;
1313 #endif
1315 void __init msg_init(void)
1317 msg_init_ns(&init_ipc_ns);
1319 ipc_init_proc_interface("sysvipc/msg",
1320 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1321 IPC_MSG_IDS, sysvipc_msg_proc_show);