RDMA/ipoib: Fix ABBA deadlock with ipoib_reap_ah()
[linux/fpc-iii.git] / ipc / msg.c
blob767587ab45a39315a541fd9df502f853acfd23a1
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 ipc64_perm *perm, int msg_qbytes)
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 perm, 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 (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(perm, ipcp);
420 if (err)
421 goto out_unlock0;
423 msq->q_qbytes = 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 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm,
605 msqid64.msg_qbytes);
606 case IPC_RMID:
607 return msgctl_down(ns, msqid, cmd, NULL, 0);
608 default:
609 return -EINVAL;
613 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
615 return ksys_msgctl(msqid, cmd, buf, IPC_64);
618 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
619 long ksys_old_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
621 int version = ipc_parse_version(&cmd);
623 return ksys_msgctl(msqid, cmd, buf, version);
626 SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
628 return ksys_old_msgctl(msqid, cmd, buf);
630 #endif
632 #ifdef CONFIG_COMPAT
634 struct compat_msqid_ds {
635 struct compat_ipc_perm msg_perm;
636 compat_uptr_t msg_first;
637 compat_uptr_t msg_last;
638 old_time32_t msg_stime;
639 old_time32_t msg_rtime;
640 old_time32_t msg_ctime;
641 compat_ulong_t msg_lcbytes;
642 compat_ulong_t msg_lqbytes;
643 unsigned short msg_cbytes;
644 unsigned short msg_qnum;
645 unsigned short msg_qbytes;
646 compat_ipc_pid_t msg_lspid;
647 compat_ipc_pid_t msg_lrpid;
650 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
651 int version)
653 memset(out, 0, sizeof(*out));
654 if (version == IPC_64) {
655 struct compat_msqid64_ds __user *p = buf;
656 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
657 return -EFAULT;
658 if (get_user(out->msg_qbytes, &p->msg_qbytes))
659 return -EFAULT;
660 } else {
661 struct compat_msqid_ds __user *p = buf;
662 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
663 return -EFAULT;
664 if (get_user(out->msg_qbytes, &p->msg_qbytes))
665 return -EFAULT;
667 return 0;
670 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
671 int version)
673 if (version == IPC_64) {
674 struct compat_msqid64_ds v;
675 memset(&v, 0, sizeof(v));
676 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
677 v.msg_stime = lower_32_bits(in->msg_stime);
678 v.msg_stime_high = upper_32_bits(in->msg_stime);
679 v.msg_rtime = lower_32_bits(in->msg_rtime);
680 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
681 v.msg_ctime = lower_32_bits(in->msg_ctime);
682 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
683 v.msg_cbytes = in->msg_cbytes;
684 v.msg_qnum = in->msg_qnum;
685 v.msg_qbytes = in->msg_qbytes;
686 v.msg_lspid = in->msg_lspid;
687 v.msg_lrpid = in->msg_lrpid;
688 return copy_to_user(buf, &v, sizeof(v));
689 } else {
690 struct compat_msqid_ds v;
691 memset(&v, 0, sizeof(v));
692 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
693 v.msg_stime = in->msg_stime;
694 v.msg_rtime = in->msg_rtime;
695 v.msg_ctime = in->msg_ctime;
696 v.msg_cbytes = in->msg_cbytes;
697 v.msg_qnum = in->msg_qnum;
698 v.msg_qbytes = in->msg_qbytes;
699 v.msg_lspid = in->msg_lspid;
700 v.msg_lrpid = in->msg_lrpid;
701 return copy_to_user(buf, &v, sizeof(v));
705 static long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr, int version)
707 struct ipc_namespace *ns;
708 int err;
709 struct msqid64_ds msqid64;
711 ns = current->nsproxy->ipc_ns;
713 if (msqid < 0 || cmd < 0)
714 return -EINVAL;
716 switch (cmd & (~IPC_64)) {
717 case IPC_INFO:
718 case MSG_INFO: {
719 struct msginfo msginfo;
720 err = msgctl_info(ns, msqid, cmd, &msginfo);
721 if (err < 0)
722 return err;
723 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
724 err = -EFAULT;
725 return err;
727 case IPC_STAT:
728 case MSG_STAT:
729 case MSG_STAT_ANY:
730 err = msgctl_stat(ns, msqid, cmd, &msqid64);
731 if (err < 0)
732 return err;
733 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
734 err = -EFAULT;
735 return err;
736 case IPC_SET:
737 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
738 return -EFAULT;
739 return msgctl_down(ns, msqid, cmd, &msqid64.msg_perm, msqid64.msg_qbytes);
740 case IPC_RMID:
741 return msgctl_down(ns, msqid, cmd, NULL, 0);
742 default:
743 return -EINVAL;
747 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
749 return compat_ksys_msgctl(msqid, cmd, uptr, IPC_64);
752 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
753 long compat_ksys_old_msgctl(int msqid, int cmd, void __user *uptr)
755 int version = compat_ipc_parse_version(&cmd);
757 return compat_ksys_msgctl(msqid, cmd, uptr, version);
760 COMPAT_SYSCALL_DEFINE3(old_msgctl, int, msqid, int, cmd, void __user *, uptr)
762 return compat_ksys_old_msgctl(msqid, cmd, uptr);
764 #endif
765 #endif
767 static int testmsg(struct msg_msg *msg, long type, int mode)
769 switch (mode) {
770 case SEARCH_ANY:
771 case SEARCH_NUMBER:
772 return 1;
773 case SEARCH_LESSEQUAL:
774 if (msg->m_type <= type)
775 return 1;
776 break;
777 case SEARCH_EQUAL:
778 if (msg->m_type == type)
779 return 1;
780 break;
781 case SEARCH_NOTEQUAL:
782 if (msg->m_type != type)
783 return 1;
784 break;
786 return 0;
789 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
790 struct wake_q_head *wake_q)
792 struct msg_receiver *msr, *t;
794 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
795 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
796 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
797 msr->r_msgtype, msr->r_mode)) {
799 list_del(&msr->r_list);
800 if (msr->r_maxsize < msg->m_ts) {
801 wake_q_add(wake_q, msr->r_tsk);
802 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
803 } else {
804 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
805 msq->q_rtime = ktime_get_real_seconds();
807 wake_q_add(wake_q, msr->r_tsk);
808 WRITE_ONCE(msr->r_msg, msg);
809 return 1;
814 return 0;
817 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
818 size_t msgsz, int msgflg)
820 struct msg_queue *msq;
821 struct msg_msg *msg;
822 int err;
823 struct ipc_namespace *ns;
824 DEFINE_WAKE_Q(wake_q);
826 ns = current->nsproxy->ipc_ns;
828 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
829 return -EINVAL;
830 if (mtype < 1)
831 return -EINVAL;
833 msg = load_msg(mtext, msgsz);
834 if (IS_ERR(msg))
835 return PTR_ERR(msg);
837 msg->m_type = mtype;
838 msg->m_ts = msgsz;
840 rcu_read_lock();
841 msq = msq_obtain_object_check(ns, msqid);
842 if (IS_ERR(msq)) {
843 err = PTR_ERR(msq);
844 goto out_unlock1;
847 ipc_lock_object(&msq->q_perm);
849 for (;;) {
850 struct msg_sender s;
852 err = -EACCES;
853 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
854 goto out_unlock0;
856 /* raced with RMID? */
857 if (!ipc_valid_object(&msq->q_perm)) {
858 err = -EIDRM;
859 goto out_unlock0;
862 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
863 if (err)
864 goto out_unlock0;
866 if (msg_fits_inqueue(msq, msgsz))
867 break;
869 /* queue full, wait: */
870 if (msgflg & IPC_NOWAIT) {
871 err = -EAGAIN;
872 goto out_unlock0;
875 /* enqueue the sender and prepare to block */
876 ss_add(msq, &s, msgsz);
878 if (!ipc_rcu_getref(&msq->q_perm)) {
879 err = -EIDRM;
880 goto out_unlock0;
883 ipc_unlock_object(&msq->q_perm);
884 rcu_read_unlock();
885 schedule();
887 rcu_read_lock();
888 ipc_lock_object(&msq->q_perm);
890 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
891 /* raced with RMID? */
892 if (!ipc_valid_object(&msq->q_perm)) {
893 err = -EIDRM;
894 goto out_unlock0;
896 ss_del(&s);
898 if (signal_pending(current)) {
899 err = -ERESTARTNOHAND;
900 goto out_unlock0;
905 ipc_update_pid(&msq->q_lspid, task_tgid(current));
906 msq->q_stime = ktime_get_real_seconds();
908 if (!pipelined_send(msq, msg, &wake_q)) {
909 /* no one is waiting for this message, enqueue it */
910 list_add_tail(&msg->m_list, &msq->q_messages);
911 msq->q_cbytes += msgsz;
912 msq->q_qnum++;
913 atomic_add(msgsz, &ns->msg_bytes);
914 atomic_inc(&ns->msg_hdrs);
917 err = 0;
918 msg = NULL;
920 out_unlock0:
921 ipc_unlock_object(&msq->q_perm);
922 wake_up_q(&wake_q);
923 out_unlock1:
924 rcu_read_unlock();
925 if (msg != NULL)
926 free_msg(msg);
927 return err;
930 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
931 int msgflg)
933 long mtype;
935 if (get_user(mtype, &msgp->mtype))
936 return -EFAULT;
937 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
940 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
941 int, msgflg)
943 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
946 #ifdef CONFIG_COMPAT
948 struct compat_msgbuf {
949 compat_long_t mtype;
950 char mtext[1];
953 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
954 compat_ssize_t msgsz, int msgflg)
956 struct compat_msgbuf __user *up = compat_ptr(msgp);
957 compat_long_t mtype;
959 if (get_user(mtype, &up->mtype))
960 return -EFAULT;
961 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
964 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
965 compat_ssize_t, msgsz, int, msgflg)
967 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
969 #endif
971 static inline int convert_mode(long *msgtyp, int msgflg)
973 if (msgflg & MSG_COPY)
974 return SEARCH_NUMBER;
976 * find message of correct type.
977 * msgtyp = 0 => get first.
978 * msgtyp > 0 => get first message of matching type.
979 * msgtyp < 0 => get message with least type must be < abs(msgtype).
981 if (*msgtyp == 0)
982 return SEARCH_ANY;
983 if (*msgtyp < 0) {
984 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
985 *msgtyp = LONG_MAX;
986 else
987 *msgtyp = -*msgtyp;
988 return SEARCH_LESSEQUAL;
990 if (msgflg & MSG_EXCEPT)
991 return SEARCH_NOTEQUAL;
992 return SEARCH_EQUAL;
995 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
997 struct msgbuf __user *msgp = dest;
998 size_t msgsz;
1000 if (put_user(msg->m_type, &msgp->mtype))
1001 return -EFAULT;
1003 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1004 if (store_msg(msgp->mtext, msg, msgsz))
1005 return -EFAULT;
1006 return msgsz;
1009 #ifdef CONFIG_CHECKPOINT_RESTORE
1011 * This function creates new kernel message structure, large enough to store
1012 * bufsz message bytes.
1014 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1016 struct msg_msg *copy;
1019 * Create dummy message to copy real message to.
1021 copy = load_msg(buf, bufsz);
1022 if (!IS_ERR(copy))
1023 copy->m_ts = bufsz;
1024 return copy;
1027 static inline void free_copy(struct msg_msg *copy)
1029 if (copy)
1030 free_msg(copy);
1032 #else
1033 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1035 return ERR_PTR(-ENOSYS);
1038 static inline void free_copy(struct msg_msg *copy)
1041 #endif
1043 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1045 struct msg_msg *msg, *found = NULL;
1046 long count = 0;
1048 list_for_each_entry(msg, &msq->q_messages, m_list) {
1049 if (testmsg(msg, *msgtyp, mode) &&
1050 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1051 *msgtyp, mode)) {
1052 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1053 *msgtyp = msg->m_type - 1;
1054 found = msg;
1055 } else if (mode == SEARCH_NUMBER) {
1056 if (*msgtyp == count)
1057 return msg;
1058 } else
1059 return msg;
1060 count++;
1064 return found ?: ERR_PTR(-EAGAIN);
1067 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1068 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1070 int mode;
1071 struct msg_queue *msq;
1072 struct ipc_namespace *ns;
1073 struct msg_msg *msg, *copy = NULL;
1074 DEFINE_WAKE_Q(wake_q);
1076 ns = current->nsproxy->ipc_ns;
1078 if (msqid < 0 || (long) bufsz < 0)
1079 return -EINVAL;
1081 if (msgflg & MSG_COPY) {
1082 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1083 return -EINVAL;
1084 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1085 if (IS_ERR(copy))
1086 return PTR_ERR(copy);
1088 mode = convert_mode(&msgtyp, msgflg);
1090 rcu_read_lock();
1091 msq = msq_obtain_object_check(ns, msqid);
1092 if (IS_ERR(msq)) {
1093 rcu_read_unlock();
1094 free_copy(copy);
1095 return PTR_ERR(msq);
1098 for (;;) {
1099 struct msg_receiver msr_d;
1101 msg = ERR_PTR(-EACCES);
1102 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1103 goto out_unlock1;
1105 ipc_lock_object(&msq->q_perm);
1107 /* raced with RMID? */
1108 if (!ipc_valid_object(&msq->q_perm)) {
1109 msg = ERR_PTR(-EIDRM);
1110 goto out_unlock0;
1113 msg = find_msg(msq, &msgtyp, mode);
1114 if (!IS_ERR(msg)) {
1116 * Found a suitable message.
1117 * Unlink it from the queue.
1119 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1120 msg = ERR_PTR(-E2BIG);
1121 goto out_unlock0;
1124 * If we are copying, then do not unlink message and do
1125 * not update queue parameters.
1127 if (msgflg & MSG_COPY) {
1128 msg = copy_msg(msg, copy);
1129 goto out_unlock0;
1132 list_del(&msg->m_list);
1133 msq->q_qnum--;
1134 msq->q_rtime = ktime_get_real_seconds();
1135 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1136 msq->q_cbytes -= msg->m_ts;
1137 atomic_sub(msg->m_ts, &ns->msg_bytes);
1138 atomic_dec(&ns->msg_hdrs);
1139 ss_wakeup(msq, &wake_q, false);
1141 goto out_unlock0;
1144 /* No message waiting. Wait for a message */
1145 if (msgflg & IPC_NOWAIT) {
1146 msg = ERR_PTR(-ENOMSG);
1147 goto out_unlock0;
1150 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1151 msr_d.r_tsk = current;
1152 msr_d.r_msgtype = msgtyp;
1153 msr_d.r_mode = mode;
1154 if (msgflg & MSG_NOERROR)
1155 msr_d.r_maxsize = INT_MAX;
1156 else
1157 msr_d.r_maxsize = bufsz;
1158 msr_d.r_msg = ERR_PTR(-EAGAIN);
1159 __set_current_state(TASK_INTERRUPTIBLE);
1161 ipc_unlock_object(&msq->q_perm);
1162 rcu_read_unlock();
1163 schedule();
1166 * Lockless receive, part 1:
1167 * We don't hold a reference to the queue and getting a
1168 * reference would defeat the idea of a lockless operation,
1169 * thus the code relies on rcu to guarantee the existence of
1170 * msq:
1171 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1172 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1174 rcu_read_lock();
1177 * Lockless receive, part 2:
1178 * The work in pipelined_send() and expunge_all():
1179 * - Set pointer to message
1180 * - Queue the receiver task for later wakeup
1181 * - Wake up the process after the lock is dropped.
1183 * Should the process wake up before this wakeup (due to a
1184 * signal) it will either see the message and continue ...
1186 msg = READ_ONCE(msr_d.r_msg);
1187 if (msg != ERR_PTR(-EAGAIN))
1188 goto out_unlock1;
1191 * ... or see -EAGAIN, acquire the lock to check the message
1192 * again.
1194 ipc_lock_object(&msq->q_perm);
1196 msg = msr_d.r_msg;
1197 if (msg != ERR_PTR(-EAGAIN))
1198 goto out_unlock0;
1200 list_del(&msr_d.r_list);
1201 if (signal_pending(current)) {
1202 msg = ERR_PTR(-ERESTARTNOHAND);
1203 goto out_unlock0;
1206 ipc_unlock_object(&msq->q_perm);
1209 out_unlock0:
1210 ipc_unlock_object(&msq->q_perm);
1211 wake_up_q(&wake_q);
1212 out_unlock1:
1213 rcu_read_unlock();
1214 if (IS_ERR(msg)) {
1215 free_copy(copy);
1216 return PTR_ERR(msg);
1219 bufsz = msg_handler(buf, msg, bufsz);
1220 free_msg(msg);
1222 return bufsz;
1225 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1226 long msgtyp, int msgflg)
1228 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1231 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1232 long, msgtyp, int, msgflg)
1234 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1237 #ifdef CONFIG_COMPAT
1238 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1240 struct compat_msgbuf __user *msgp = dest;
1241 size_t msgsz;
1243 if (put_user(msg->m_type, &msgp->mtype))
1244 return -EFAULT;
1246 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1247 if (store_msg(msgp->mtext, msg, msgsz))
1248 return -EFAULT;
1249 return msgsz;
1252 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1253 compat_long_t msgtyp, int msgflg)
1255 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1256 msgflg, compat_do_msg_fill);
1259 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1260 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1261 int, msgflg)
1263 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1265 #endif
1267 void msg_init_ns(struct ipc_namespace *ns)
1269 ns->msg_ctlmax = MSGMAX;
1270 ns->msg_ctlmnb = MSGMNB;
1271 ns->msg_ctlmni = MSGMNI;
1273 atomic_set(&ns->msg_bytes, 0);
1274 atomic_set(&ns->msg_hdrs, 0);
1275 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1278 #ifdef CONFIG_IPC_NS
1279 void msg_exit_ns(struct ipc_namespace *ns)
1281 free_ipcs(ns, &msg_ids(ns), freeque);
1282 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1283 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1285 #endif
1287 #ifdef CONFIG_PROC_FS
1288 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1290 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1291 struct user_namespace *user_ns = seq_user_ns(s);
1292 struct kern_ipc_perm *ipcp = it;
1293 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1295 seq_printf(s,
1296 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1297 msq->q_perm.key,
1298 msq->q_perm.id,
1299 msq->q_perm.mode,
1300 msq->q_cbytes,
1301 msq->q_qnum,
1302 pid_nr_ns(msq->q_lspid, pid_ns),
1303 pid_nr_ns(msq->q_lrpid, pid_ns),
1304 from_kuid_munged(user_ns, msq->q_perm.uid),
1305 from_kgid_munged(user_ns, msq->q_perm.gid),
1306 from_kuid_munged(user_ns, msq->q_perm.cuid),
1307 from_kgid_munged(user_ns, msq->q_perm.cgid),
1308 msq->q_stime,
1309 msq->q_rtime,
1310 msq->q_ctime);
1312 return 0;
1314 #endif
1316 void __init msg_init(void)
1318 msg_init_ns(&init_ipc_ns);
1320 ipc_init_proc_interface("sysvipc/msg",
1321 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1322 IPC_MSG_IDS, sysvipc_msg_proc_show);