memcg, oom: notify on oom killer invocation from the charge path
[linux/fpc-iii.git] / ipc / msg.c
blob0833c640591561dc907a443ce4b9dc61cfbab638
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 long ksys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
572 int version;
573 struct ipc_namespace *ns;
574 struct msqid64_ds msqid64;
575 int err;
577 if (msqid < 0 || cmd < 0)
578 return -EINVAL;
580 version = ipc_parse_version(&cmd);
581 ns = current->nsproxy->ipc_ns;
583 switch (cmd) {
584 case IPC_INFO:
585 case MSG_INFO: {
586 struct msginfo msginfo;
587 err = msgctl_info(ns, msqid, cmd, &msginfo);
588 if (err < 0)
589 return err;
590 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
591 err = -EFAULT;
592 return err;
594 case MSG_STAT: /* msqid is an index rather than a msg queue id */
595 case MSG_STAT_ANY:
596 case IPC_STAT:
597 err = msgctl_stat(ns, msqid, cmd, &msqid64);
598 if (err < 0)
599 return err;
600 if (copy_msqid_to_user(buf, &msqid64, version))
601 err = -EFAULT;
602 return err;
603 case IPC_SET:
604 if (copy_msqid_from_user(&msqid64, buf, version))
605 return -EFAULT;
606 /* fallthru */
607 case IPC_RMID:
608 return msgctl_down(ns, msqid, cmd, &msqid64);
609 default:
610 return -EINVAL;
614 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
616 return ksys_msgctl(msqid, cmd, buf);
619 #ifdef CONFIG_COMPAT
621 struct compat_msqid_ds {
622 struct compat_ipc_perm msg_perm;
623 compat_uptr_t msg_first;
624 compat_uptr_t msg_last;
625 old_time32_t msg_stime;
626 old_time32_t msg_rtime;
627 old_time32_t msg_ctime;
628 compat_ulong_t msg_lcbytes;
629 compat_ulong_t msg_lqbytes;
630 unsigned short msg_cbytes;
631 unsigned short msg_qnum;
632 unsigned short msg_qbytes;
633 compat_ipc_pid_t msg_lspid;
634 compat_ipc_pid_t msg_lrpid;
637 static int copy_compat_msqid_from_user(struct msqid64_ds *out, void __user *buf,
638 int version)
640 memset(out, 0, sizeof(*out));
641 if (version == IPC_64) {
642 struct compat_msqid64_ds __user *p = buf;
643 if (get_compat_ipc64_perm(&out->msg_perm, &p->msg_perm))
644 return -EFAULT;
645 if (get_user(out->msg_qbytes, &p->msg_qbytes))
646 return -EFAULT;
647 } else {
648 struct compat_msqid_ds __user *p = buf;
649 if (get_compat_ipc_perm(&out->msg_perm, &p->msg_perm))
650 return -EFAULT;
651 if (get_user(out->msg_qbytes, &p->msg_qbytes))
652 return -EFAULT;
654 return 0;
657 static int copy_compat_msqid_to_user(void __user *buf, struct msqid64_ds *in,
658 int version)
660 if (version == IPC_64) {
661 struct compat_msqid64_ds v;
662 memset(&v, 0, sizeof(v));
663 to_compat_ipc64_perm(&v.msg_perm, &in->msg_perm);
664 v.msg_stime = lower_32_bits(in->msg_stime);
665 v.msg_stime_high = upper_32_bits(in->msg_stime);
666 v.msg_rtime = lower_32_bits(in->msg_rtime);
667 v.msg_rtime_high = upper_32_bits(in->msg_rtime);
668 v.msg_ctime = lower_32_bits(in->msg_ctime);
669 v.msg_ctime_high = upper_32_bits(in->msg_ctime);
670 v.msg_cbytes = in->msg_cbytes;
671 v.msg_qnum = in->msg_qnum;
672 v.msg_qbytes = in->msg_qbytes;
673 v.msg_lspid = in->msg_lspid;
674 v.msg_lrpid = in->msg_lrpid;
675 return copy_to_user(buf, &v, sizeof(v));
676 } else {
677 struct compat_msqid_ds v;
678 memset(&v, 0, sizeof(v));
679 to_compat_ipc_perm(&v.msg_perm, &in->msg_perm);
680 v.msg_stime = in->msg_stime;
681 v.msg_rtime = in->msg_rtime;
682 v.msg_ctime = 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));
692 long compat_ksys_msgctl(int msqid, int cmd, void __user *uptr)
694 struct ipc_namespace *ns;
695 int err;
696 struct msqid64_ds msqid64;
697 int version = compat_ipc_parse_version(&cmd);
699 ns = current->nsproxy->ipc_ns;
701 if (msqid < 0 || cmd < 0)
702 return -EINVAL;
704 switch (cmd & (~IPC_64)) {
705 case IPC_INFO:
706 case MSG_INFO: {
707 struct msginfo msginfo;
708 err = msgctl_info(ns, msqid, cmd, &msginfo);
709 if (err < 0)
710 return err;
711 if (copy_to_user(uptr, &msginfo, sizeof(struct msginfo)))
712 err = -EFAULT;
713 return err;
715 case IPC_STAT:
716 case MSG_STAT:
717 case MSG_STAT_ANY:
718 err = msgctl_stat(ns, msqid, cmd, &msqid64);
719 if (err < 0)
720 return err;
721 if (copy_compat_msqid_to_user(uptr, &msqid64, version))
722 err = -EFAULT;
723 return err;
724 case IPC_SET:
725 if (copy_compat_msqid_from_user(&msqid64, uptr, version))
726 return -EFAULT;
727 /* fallthru */
728 case IPC_RMID:
729 return msgctl_down(ns, msqid, cmd, &msqid64);
730 default:
731 return -EINVAL;
735 COMPAT_SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, void __user *, uptr)
737 return compat_ksys_msgctl(msqid, cmd, uptr);
739 #endif
741 static int testmsg(struct msg_msg *msg, long type, int mode)
743 switch (mode) {
744 case SEARCH_ANY:
745 case SEARCH_NUMBER:
746 return 1;
747 case SEARCH_LESSEQUAL:
748 if (msg->m_type <= type)
749 return 1;
750 break;
751 case SEARCH_EQUAL:
752 if (msg->m_type == type)
753 return 1;
754 break;
755 case SEARCH_NOTEQUAL:
756 if (msg->m_type != type)
757 return 1;
758 break;
760 return 0;
763 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
764 struct wake_q_head *wake_q)
766 struct msg_receiver *msr, *t;
768 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
769 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
770 !security_msg_queue_msgrcv(&msq->q_perm, msg, msr->r_tsk,
771 msr->r_msgtype, msr->r_mode)) {
773 list_del(&msr->r_list);
774 if (msr->r_maxsize < msg->m_ts) {
775 wake_q_add(wake_q, msr->r_tsk);
776 WRITE_ONCE(msr->r_msg, ERR_PTR(-E2BIG));
777 } else {
778 ipc_update_pid(&msq->q_lrpid, task_pid(msr->r_tsk));
779 msq->q_rtime = ktime_get_real_seconds();
781 wake_q_add(wake_q, msr->r_tsk);
782 WRITE_ONCE(msr->r_msg, msg);
783 return 1;
788 return 0;
791 static long do_msgsnd(int msqid, long mtype, void __user *mtext,
792 size_t msgsz, int msgflg)
794 struct msg_queue *msq;
795 struct msg_msg *msg;
796 int err;
797 struct ipc_namespace *ns;
798 DEFINE_WAKE_Q(wake_q);
800 ns = current->nsproxy->ipc_ns;
802 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
803 return -EINVAL;
804 if (mtype < 1)
805 return -EINVAL;
807 msg = load_msg(mtext, msgsz);
808 if (IS_ERR(msg))
809 return PTR_ERR(msg);
811 msg->m_type = mtype;
812 msg->m_ts = msgsz;
814 rcu_read_lock();
815 msq = msq_obtain_object_check(ns, msqid);
816 if (IS_ERR(msq)) {
817 err = PTR_ERR(msq);
818 goto out_unlock1;
821 ipc_lock_object(&msq->q_perm);
823 for (;;) {
824 struct msg_sender s;
826 err = -EACCES;
827 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
828 goto out_unlock0;
830 /* raced with RMID? */
831 if (!ipc_valid_object(&msq->q_perm)) {
832 err = -EIDRM;
833 goto out_unlock0;
836 err = security_msg_queue_msgsnd(&msq->q_perm, msg, msgflg);
837 if (err)
838 goto out_unlock0;
840 if (msg_fits_inqueue(msq, msgsz))
841 break;
843 /* queue full, wait: */
844 if (msgflg & IPC_NOWAIT) {
845 err = -EAGAIN;
846 goto out_unlock0;
849 /* enqueue the sender and prepare to block */
850 ss_add(msq, &s, msgsz);
852 if (!ipc_rcu_getref(&msq->q_perm)) {
853 err = -EIDRM;
854 goto out_unlock0;
857 ipc_unlock_object(&msq->q_perm);
858 rcu_read_unlock();
859 schedule();
861 rcu_read_lock();
862 ipc_lock_object(&msq->q_perm);
864 ipc_rcu_putref(&msq->q_perm, msg_rcu_free);
865 /* raced with RMID? */
866 if (!ipc_valid_object(&msq->q_perm)) {
867 err = -EIDRM;
868 goto out_unlock0;
870 ss_del(&s);
872 if (signal_pending(current)) {
873 err = -ERESTARTNOHAND;
874 goto out_unlock0;
879 ipc_update_pid(&msq->q_lspid, task_tgid(current));
880 msq->q_stime = ktime_get_real_seconds();
882 if (!pipelined_send(msq, msg, &wake_q)) {
883 /* no one is waiting for this message, enqueue it */
884 list_add_tail(&msg->m_list, &msq->q_messages);
885 msq->q_cbytes += msgsz;
886 msq->q_qnum++;
887 atomic_add(msgsz, &ns->msg_bytes);
888 atomic_inc(&ns->msg_hdrs);
891 err = 0;
892 msg = NULL;
894 out_unlock0:
895 ipc_unlock_object(&msq->q_perm);
896 wake_up_q(&wake_q);
897 out_unlock1:
898 rcu_read_unlock();
899 if (msg != NULL)
900 free_msg(msg);
901 return err;
904 long ksys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz,
905 int msgflg)
907 long mtype;
909 if (get_user(mtype, &msgp->mtype))
910 return -EFAULT;
911 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
914 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
915 int, msgflg)
917 return ksys_msgsnd(msqid, msgp, msgsz, msgflg);
920 #ifdef CONFIG_COMPAT
922 struct compat_msgbuf {
923 compat_long_t mtype;
924 char mtext[1];
927 long compat_ksys_msgsnd(int msqid, compat_uptr_t msgp,
928 compat_ssize_t msgsz, int msgflg)
930 struct compat_msgbuf __user *up = compat_ptr(msgp);
931 compat_long_t mtype;
933 if (get_user(mtype, &up->mtype))
934 return -EFAULT;
935 return do_msgsnd(msqid, mtype, up->mtext, (ssize_t)msgsz, msgflg);
938 COMPAT_SYSCALL_DEFINE4(msgsnd, int, msqid, compat_uptr_t, msgp,
939 compat_ssize_t, msgsz, int, msgflg)
941 return compat_ksys_msgsnd(msqid, msgp, msgsz, msgflg);
943 #endif
945 static inline int convert_mode(long *msgtyp, int msgflg)
947 if (msgflg & MSG_COPY)
948 return SEARCH_NUMBER;
950 * find message of correct type.
951 * msgtyp = 0 => get first.
952 * msgtyp > 0 => get first message of matching type.
953 * msgtyp < 0 => get message with least type must be < abs(msgtype).
955 if (*msgtyp == 0)
956 return SEARCH_ANY;
957 if (*msgtyp < 0) {
958 if (*msgtyp == LONG_MIN) /* -LONG_MIN is undefined */
959 *msgtyp = LONG_MAX;
960 else
961 *msgtyp = -*msgtyp;
962 return SEARCH_LESSEQUAL;
964 if (msgflg & MSG_EXCEPT)
965 return SEARCH_NOTEQUAL;
966 return SEARCH_EQUAL;
969 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
971 struct msgbuf __user *msgp = dest;
972 size_t msgsz;
974 if (put_user(msg->m_type, &msgp->mtype))
975 return -EFAULT;
977 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
978 if (store_msg(msgp->mtext, msg, msgsz))
979 return -EFAULT;
980 return msgsz;
983 #ifdef CONFIG_CHECKPOINT_RESTORE
985 * This function creates new kernel message structure, large enough to store
986 * bufsz message bytes.
988 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
990 struct msg_msg *copy;
993 * Create dummy message to copy real message to.
995 copy = load_msg(buf, bufsz);
996 if (!IS_ERR(copy))
997 copy->m_ts = bufsz;
998 return copy;
1001 static inline void free_copy(struct msg_msg *copy)
1003 if (copy)
1004 free_msg(copy);
1006 #else
1007 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
1009 return ERR_PTR(-ENOSYS);
1012 static inline void free_copy(struct msg_msg *copy)
1015 #endif
1017 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
1019 struct msg_msg *msg, *found = NULL;
1020 long count = 0;
1022 list_for_each_entry(msg, &msq->q_messages, m_list) {
1023 if (testmsg(msg, *msgtyp, mode) &&
1024 !security_msg_queue_msgrcv(&msq->q_perm, msg, current,
1025 *msgtyp, mode)) {
1026 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
1027 *msgtyp = msg->m_type - 1;
1028 found = msg;
1029 } else if (mode == SEARCH_NUMBER) {
1030 if (*msgtyp == count)
1031 return msg;
1032 } else
1033 return msg;
1034 count++;
1038 return found ?: ERR_PTR(-EAGAIN);
1041 static long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
1042 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
1044 int mode;
1045 struct msg_queue *msq;
1046 struct ipc_namespace *ns;
1047 struct msg_msg *msg, *copy = NULL;
1048 DEFINE_WAKE_Q(wake_q);
1050 ns = current->nsproxy->ipc_ns;
1052 if (msqid < 0 || (long) bufsz < 0)
1053 return -EINVAL;
1055 if (msgflg & MSG_COPY) {
1056 if ((msgflg & MSG_EXCEPT) || !(msgflg & IPC_NOWAIT))
1057 return -EINVAL;
1058 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
1059 if (IS_ERR(copy))
1060 return PTR_ERR(copy);
1062 mode = convert_mode(&msgtyp, msgflg);
1064 rcu_read_lock();
1065 msq = msq_obtain_object_check(ns, msqid);
1066 if (IS_ERR(msq)) {
1067 rcu_read_unlock();
1068 free_copy(copy);
1069 return PTR_ERR(msq);
1072 for (;;) {
1073 struct msg_receiver msr_d;
1075 msg = ERR_PTR(-EACCES);
1076 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
1077 goto out_unlock1;
1079 ipc_lock_object(&msq->q_perm);
1081 /* raced with RMID? */
1082 if (!ipc_valid_object(&msq->q_perm)) {
1083 msg = ERR_PTR(-EIDRM);
1084 goto out_unlock0;
1087 msg = find_msg(msq, &msgtyp, mode);
1088 if (!IS_ERR(msg)) {
1090 * Found a suitable message.
1091 * Unlink it from the queue.
1093 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
1094 msg = ERR_PTR(-E2BIG);
1095 goto out_unlock0;
1098 * If we are copying, then do not unlink message and do
1099 * not update queue parameters.
1101 if (msgflg & MSG_COPY) {
1102 msg = copy_msg(msg, copy);
1103 goto out_unlock0;
1106 list_del(&msg->m_list);
1107 msq->q_qnum--;
1108 msq->q_rtime = ktime_get_real_seconds();
1109 ipc_update_pid(&msq->q_lrpid, task_tgid(current));
1110 msq->q_cbytes -= msg->m_ts;
1111 atomic_sub(msg->m_ts, &ns->msg_bytes);
1112 atomic_dec(&ns->msg_hdrs);
1113 ss_wakeup(msq, &wake_q, false);
1115 goto out_unlock0;
1118 /* No message waiting. Wait for a message */
1119 if (msgflg & IPC_NOWAIT) {
1120 msg = ERR_PTR(-ENOMSG);
1121 goto out_unlock0;
1124 list_add_tail(&msr_d.r_list, &msq->q_receivers);
1125 msr_d.r_tsk = current;
1126 msr_d.r_msgtype = msgtyp;
1127 msr_d.r_mode = mode;
1128 if (msgflg & MSG_NOERROR)
1129 msr_d.r_maxsize = INT_MAX;
1130 else
1131 msr_d.r_maxsize = bufsz;
1132 msr_d.r_msg = ERR_PTR(-EAGAIN);
1133 __set_current_state(TASK_INTERRUPTIBLE);
1135 ipc_unlock_object(&msq->q_perm);
1136 rcu_read_unlock();
1137 schedule();
1140 * Lockless receive, part 1:
1141 * We don't hold a reference to the queue and getting a
1142 * reference would defeat the idea of a lockless operation,
1143 * thus the code relies on rcu to guarantee the existence of
1144 * msq:
1145 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
1146 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
1148 rcu_read_lock();
1151 * Lockless receive, part 2:
1152 * The work in pipelined_send() and expunge_all():
1153 * - Set pointer to message
1154 * - Queue the receiver task for later wakeup
1155 * - Wake up the process after the lock is dropped.
1157 * Should the process wake up before this wakeup (due to a
1158 * signal) it will either see the message and continue ...
1160 msg = READ_ONCE(msr_d.r_msg);
1161 if (msg != ERR_PTR(-EAGAIN))
1162 goto out_unlock1;
1165 * ... or see -EAGAIN, acquire the lock to check the message
1166 * again.
1168 ipc_lock_object(&msq->q_perm);
1170 msg = msr_d.r_msg;
1171 if (msg != ERR_PTR(-EAGAIN))
1172 goto out_unlock0;
1174 list_del(&msr_d.r_list);
1175 if (signal_pending(current)) {
1176 msg = ERR_PTR(-ERESTARTNOHAND);
1177 goto out_unlock0;
1180 ipc_unlock_object(&msq->q_perm);
1183 out_unlock0:
1184 ipc_unlock_object(&msq->q_perm);
1185 wake_up_q(&wake_q);
1186 out_unlock1:
1187 rcu_read_unlock();
1188 if (IS_ERR(msg)) {
1189 free_copy(copy);
1190 return PTR_ERR(msg);
1193 bufsz = msg_handler(buf, msg, bufsz);
1194 free_msg(msg);
1196 return bufsz;
1199 long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
1200 long msgtyp, int msgflg)
1202 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1205 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1206 long, msgtyp, int, msgflg)
1208 return ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1211 #ifdef CONFIG_COMPAT
1212 static long compat_do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
1214 struct compat_msgbuf __user *msgp = dest;
1215 size_t msgsz;
1217 if (put_user(msg->m_type, &msgp->mtype))
1218 return -EFAULT;
1220 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
1221 if (store_msg(msgp->mtext, msg, msgsz))
1222 return -EFAULT;
1223 return msgsz;
1226 long compat_ksys_msgrcv(int msqid, compat_uptr_t msgp, compat_ssize_t msgsz,
1227 compat_long_t msgtyp, int msgflg)
1229 return do_msgrcv(msqid, compat_ptr(msgp), (ssize_t)msgsz, (long)msgtyp,
1230 msgflg, compat_do_msg_fill);
1233 COMPAT_SYSCALL_DEFINE5(msgrcv, int, msqid, compat_uptr_t, msgp,
1234 compat_ssize_t, msgsz, compat_long_t, msgtyp,
1235 int, msgflg)
1237 return compat_ksys_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg);
1239 #endif
1241 void msg_init_ns(struct ipc_namespace *ns)
1243 ns->msg_ctlmax = MSGMAX;
1244 ns->msg_ctlmnb = MSGMNB;
1245 ns->msg_ctlmni = MSGMNI;
1247 atomic_set(&ns->msg_bytes, 0);
1248 atomic_set(&ns->msg_hdrs, 0);
1249 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
1252 #ifdef CONFIG_IPC_NS
1253 void msg_exit_ns(struct ipc_namespace *ns)
1255 free_ipcs(ns, &msg_ids(ns), freeque);
1256 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
1257 rhashtable_destroy(&ns->ids[IPC_MSG_IDS].key_ht);
1259 #endif
1261 #ifdef CONFIG_PROC_FS
1262 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1264 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1265 struct user_namespace *user_ns = seq_user_ns(s);
1266 struct kern_ipc_perm *ipcp = it;
1267 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
1269 seq_printf(s,
1270 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10llu %10llu %10llu\n",
1271 msq->q_perm.key,
1272 msq->q_perm.id,
1273 msq->q_perm.mode,
1274 msq->q_cbytes,
1275 msq->q_qnum,
1276 pid_nr_ns(msq->q_lspid, pid_ns),
1277 pid_nr_ns(msq->q_lrpid, pid_ns),
1278 from_kuid_munged(user_ns, msq->q_perm.uid),
1279 from_kgid_munged(user_ns, msq->q_perm.gid),
1280 from_kuid_munged(user_ns, msq->q_perm.cuid),
1281 from_kgid_munged(user_ns, msq->q_perm.cgid),
1282 msq->q_stime,
1283 msq->q_rtime,
1284 msq->q_ctime);
1286 return 0;
1288 #endif
1290 void __init msg_init(void)
1292 msg_init_ns(&init_ipc_ns);
1294 ipc_init_proc_interface("sysvipc/msg",
1295 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
1296 IPC_MSG_IDS, sysvipc_msg_proc_show);