arm64: KVM: trap VM system registers until MMU and caches are ON
[linux/fpc-iii.git] / ipc / msg.c
blob245db1140ad66a2be47744f02ef0d80deea5006f
1 /*
2 * linux/ipc/msg.c
3 * Copyright (C) 1992 Krishna Balasubramanian
5 * Removed all the remaining kerneld mess
6 * Catch the -EFAULT stuff properly
7 * Use GFP_KERNEL for messages as in 1.2
8 * Fixed up the unchecked user space derefs
9 * Copyright (C) 1998 Alan Cox & Andi Kleen
11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
15 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
25 #include <linux/capability.h>
26 #include <linux/msg.h>
27 #include <linux/spinlock.h>
28 #include <linux/init.h>
29 #include <linux/mm.h>
30 #include <linux/proc_fs.h>
31 #include <linux/list.h>
32 #include <linux/security.h>
33 #include <linux/sched.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/seq_file.h>
37 #include <linux/rwsem.h>
38 #include <linux/nsproxy.h>
39 #include <linux/ipc_namespace.h>
41 #include <asm/current.h>
42 #include <asm/uaccess.h>
43 #include "util.h"
46 * one msg_receiver structure for each sleeping receiver:
48 struct msg_receiver {
49 struct list_head r_list;
50 struct task_struct *r_tsk;
52 int r_mode;
53 long r_msgtype;
54 long r_maxsize;
56 struct msg_msg *volatile r_msg;
59 /* one msg_sender for each sleeping sender */
60 struct msg_sender {
61 struct list_head list;
62 struct task_struct *tsk;
65 #define SEARCH_ANY 1
66 #define SEARCH_EQUAL 2
67 #define SEARCH_NOTEQUAL 3
68 #define SEARCH_LESSEQUAL 4
69 #define SEARCH_NUMBER 5
71 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
73 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
74 static int newque(struct ipc_namespace *, struct ipc_params *);
75 #ifdef CONFIG_PROC_FS
76 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
77 #endif
80 * Scale msgmni with the available lowmem size: the memory dedicated to msg
81 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
82 * Also take into account the number of nsproxies created so far.
83 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
85 void recompute_msgmni(struct ipc_namespace *ns)
87 struct sysinfo i;
88 unsigned long allowed;
89 int nb_ns;
91 si_meminfo(&i);
92 allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
93 / MSGMNB;
94 nb_ns = atomic_read(&nr_ipc_ns);
95 allowed /= nb_ns;
97 if (allowed < MSGMNI) {
98 ns->msg_ctlmni = MSGMNI;
99 return;
102 if (allowed > IPCMNI / nb_ns) {
103 ns->msg_ctlmni = IPCMNI / nb_ns;
104 return;
107 ns->msg_ctlmni = allowed;
110 void msg_init_ns(struct ipc_namespace *ns)
112 ns->msg_ctlmax = MSGMAX;
113 ns->msg_ctlmnb = MSGMNB;
115 recompute_msgmni(ns);
117 atomic_set(&ns->msg_bytes, 0);
118 atomic_set(&ns->msg_hdrs, 0);
119 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
122 #ifdef CONFIG_IPC_NS
123 void msg_exit_ns(struct ipc_namespace *ns)
125 free_ipcs(ns, &msg_ids(ns), freeque);
126 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
128 #endif
130 void __init msg_init(void)
132 msg_init_ns(&init_ipc_ns);
134 printk(KERN_INFO "msgmni has been set to %d\n",
135 init_ipc_ns.msg_ctlmni);
137 ipc_init_proc_interface("sysvipc/msg",
138 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
139 IPC_MSG_IDS, sysvipc_msg_proc_show);
142 static inline struct msg_queue *msq_obtain_object(struct ipc_namespace *ns, int id)
144 struct kern_ipc_perm *ipcp = ipc_obtain_object(&msg_ids(ns), id);
146 if (IS_ERR(ipcp))
147 return ERR_CAST(ipcp);
149 return container_of(ipcp, struct msg_queue, q_perm);
152 static inline struct msg_queue *msq_obtain_object_check(struct ipc_namespace *ns,
153 int id)
155 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&msg_ids(ns), id);
157 if (IS_ERR(ipcp))
158 return ERR_CAST(ipcp);
160 return container_of(ipcp, struct msg_queue, q_perm);
163 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
165 ipc_rmid(&msg_ids(ns), &s->q_perm);
168 static void msg_rcu_free(struct rcu_head *head)
170 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
171 struct msg_queue *msq = ipc_rcu_to_struct(p);
173 security_msg_queue_free(msq);
174 ipc_rcu_free(head);
178 * newque - Create a new msg queue
179 * @ns: namespace
180 * @params: ptr to the structure that contains the key and msgflg
182 * Called with msg_ids.rwsem held (writer)
184 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
186 struct msg_queue *msq;
187 int id, retval;
188 key_t key = params->key;
189 int msgflg = params->flg;
191 msq = ipc_rcu_alloc(sizeof(*msq));
192 if (!msq)
193 return -ENOMEM;
195 msq->q_perm.mode = msgflg & S_IRWXUGO;
196 msq->q_perm.key = key;
198 msq->q_perm.security = NULL;
199 retval = security_msg_queue_alloc(msq);
200 if (retval) {
201 ipc_rcu_putref(msq, ipc_rcu_free);
202 return retval;
205 /* ipc_addid() locks msq upon success. */
206 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
207 if (id < 0) {
208 ipc_rcu_putref(msq, msg_rcu_free);
209 return id;
212 msq->q_stime = msq->q_rtime = 0;
213 msq->q_ctime = get_seconds();
214 msq->q_cbytes = msq->q_qnum = 0;
215 msq->q_qbytes = ns->msg_ctlmnb;
216 msq->q_lspid = msq->q_lrpid = 0;
217 INIT_LIST_HEAD(&msq->q_messages);
218 INIT_LIST_HEAD(&msq->q_receivers);
219 INIT_LIST_HEAD(&msq->q_senders);
221 ipc_unlock_object(&msq->q_perm);
222 rcu_read_unlock();
224 return msq->q_perm.id;
227 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
229 mss->tsk = current;
230 current->state = TASK_INTERRUPTIBLE;
231 list_add_tail(&mss->list, &msq->q_senders);
234 static inline void ss_del(struct msg_sender *mss)
236 if (mss->list.next != NULL)
237 list_del(&mss->list);
240 static void ss_wakeup(struct list_head *h, int kill)
242 struct msg_sender *mss, *t;
244 list_for_each_entry_safe(mss, t, h, list) {
245 if (kill)
246 mss->list.next = NULL;
247 wake_up_process(mss->tsk);
251 static void expunge_all(struct msg_queue *msq, int res)
253 struct msg_receiver *msr, *t;
255 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
256 msr->r_msg = NULL; /* initialize expunge ordering */
257 wake_up_process(msr->r_tsk);
259 * Ensure that the wakeup is visible before setting r_msg as
260 * the receiving end depends on it: either spinning on a nil,
261 * or dealing with -EAGAIN cases. See lockless receive part 1
262 * and 2 in do_msgrcv().
264 smp_mb();
265 msr->r_msg = ERR_PTR(res);
270 * freeque() wakes up waiters on the sender and receiver waiting queue,
271 * removes the message queue from message queue ID IDR, and cleans up all the
272 * messages associated with this queue.
274 * msg_ids.rwsem (writer) and the spinlock for this message queue are held
275 * before freeque() is called. msg_ids.rwsem remains locked on exit.
277 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
279 struct msg_msg *msg, *t;
280 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
282 expunge_all(msq, -EIDRM);
283 ss_wakeup(&msq->q_senders, 1);
284 msg_rmid(ns, msq);
285 ipc_unlock_object(&msq->q_perm);
286 rcu_read_unlock();
288 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
289 atomic_dec(&ns->msg_hdrs);
290 free_msg(msg);
292 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
293 ipc_rcu_putref(msq, msg_rcu_free);
297 * Called with msg_ids.rwsem and ipcp locked.
299 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
301 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
303 return security_msg_queue_associate(msq, msgflg);
306 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
308 struct ipc_namespace *ns;
309 struct ipc_ops msg_ops;
310 struct ipc_params msg_params;
312 ns = current->nsproxy->ipc_ns;
314 msg_ops.getnew = newque;
315 msg_ops.associate = msg_security;
316 msg_ops.more_checks = NULL;
318 msg_params.key = key;
319 msg_params.flg = msgflg;
321 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
324 static inline unsigned long
325 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
327 switch (version) {
328 case IPC_64:
329 return copy_to_user(buf, in, sizeof(*in));
330 case IPC_OLD:
332 struct msqid_ds out;
334 memset(&out, 0, sizeof(out));
336 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
338 out.msg_stime = in->msg_stime;
339 out.msg_rtime = in->msg_rtime;
340 out.msg_ctime = in->msg_ctime;
342 if (in->msg_cbytes > USHRT_MAX)
343 out.msg_cbytes = USHRT_MAX;
344 else
345 out.msg_cbytes = in->msg_cbytes;
346 out.msg_lcbytes = in->msg_cbytes;
348 if (in->msg_qnum > USHRT_MAX)
349 out.msg_qnum = USHRT_MAX;
350 else
351 out.msg_qnum = in->msg_qnum;
353 if (in->msg_qbytes > USHRT_MAX)
354 out.msg_qbytes = USHRT_MAX;
355 else
356 out.msg_qbytes = in->msg_qbytes;
357 out.msg_lqbytes = in->msg_qbytes;
359 out.msg_lspid = in->msg_lspid;
360 out.msg_lrpid = in->msg_lrpid;
362 return copy_to_user(buf, &out, sizeof(out));
364 default:
365 return -EINVAL;
369 static inline unsigned long
370 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
372 switch (version) {
373 case IPC_64:
374 if (copy_from_user(out, buf, sizeof(*out)))
375 return -EFAULT;
376 return 0;
377 case IPC_OLD:
379 struct msqid_ds tbuf_old;
381 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
382 return -EFAULT;
384 out->msg_perm.uid = tbuf_old.msg_perm.uid;
385 out->msg_perm.gid = tbuf_old.msg_perm.gid;
386 out->msg_perm.mode = tbuf_old.msg_perm.mode;
388 if (tbuf_old.msg_qbytes == 0)
389 out->msg_qbytes = tbuf_old.msg_lqbytes;
390 else
391 out->msg_qbytes = tbuf_old.msg_qbytes;
393 return 0;
395 default:
396 return -EINVAL;
401 * This function handles some msgctl commands which require the rwsem
402 * to be held in write mode.
403 * NOTE: no locks must be held, the rwsem is taken inside this function.
405 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
406 struct msqid_ds __user *buf, int version)
408 struct kern_ipc_perm *ipcp;
409 struct msqid64_ds uninitialized_var(msqid64);
410 struct msg_queue *msq;
411 int err;
413 if (cmd == IPC_SET) {
414 if (copy_msqid_from_user(&msqid64, buf, version))
415 return -EFAULT;
418 down_write(&msg_ids(ns).rwsem);
419 rcu_read_lock();
421 ipcp = ipcctl_pre_down_nolock(ns, &msg_ids(ns), msqid, cmd,
422 &msqid64.msg_perm, msqid64.msg_qbytes);
423 if (IS_ERR(ipcp)) {
424 err = PTR_ERR(ipcp);
425 goto out_unlock1;
428 msq = container_of(ipcp, struct msg_queue, q_perm);
430 err = security_msg_queue_msgctl(msq, cmd);
431 if (err)
432 goto out_unlock1;
434 switch (cmd) {
435 case IPC_RMID:
436 ipc_lock_object(&msq->q_perm);
437 /* freeque unlocks the ipc object and rcu */
438 freeque(ns, ipcp);
439 goto out_up;
440 case IPC_SET:
441 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
442 !capable(CAP_SYS_RESOURCE)) {
443 err = -EPERM;
444 goto out_unlock1;
447 ipc_lock_object(&msq->q_perm);
448 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
449 if (err)
450 goto out_unlock0;
452 msq->q_qbytes = msqid64.msg_qbytes;
454 msq->q_ctime = get_seconds();
455 /* sleeping receivers might be excluded by
456 * stricter permissions.
458 expunge_all(msq, -EAGAIN);
459 /* sleeping senders might be able to send
460 * due to a larger queue size.
462 ss_wakeup(&msq->q_senders, 0);
463 break;
464 default:
465 err = -EINVAL;
466 goto out_unlock1;
469 out_unlock0:
470 ipc_unlock_object(&msq->q_perm);
471 out_unlock1:
472 rcu_read_unlock();
473 out_up:
474 up_write(&msg_ids(ns).rwsem);
475 return err;
478 static int msgctl_nolock(struct ipc_namespace *ns, int msqid,
479 int cmd, int version, void __user *buf)
481 int err;
482 struct msg_queue *msq;
484 switch (cmd) {
485 case IPC_INFO:
486 case MSG_INFO:
488 struct msginfo msginfo;
489 int max_id;
491 if (!buf)
492 return -EFAULT;
495 * We must not return kernel stack data.
496 * due to padding, it's not enough
497 * to set all member fields.
499 err = security_msg_queue_msgctl(NULL, cmd);
500 if (err)
501 return err;
503 memset(&msginfo, 0, sizeof(msginfo));
504 msginfo.msgmni = ns->msg_ctlmni;
505 msginfo.msgmax = ns->msg_ctlmax;
506 msginfo.msgmnb = ns->msg_ctlmnb;
507 msginfo.msgssz = MSGSSZ;
508 msginfo.msgseg = MSGSEG;
509 down_read(&msg_ids(ns).rwsem);
510 if (cmd == MSG_INFO) {
511 msginfo.msgpool = msg_ids(ns).in_use;
512 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
513 msginfo.msgtql = atomic_read(&ns->msg_bytes);
514 } else {
515 msginfo.msgmap = MSGMAP;
516 msginfo.msgpool = MSGPOOL;
517 msginfo.msgtql = MSGTQL;
519 max_id = ipc_get_maxid(&msg_ids(ns));
520 up_read(&msg_ids(ns).rwsem);
521 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
522 return -EFAULT;
523 return (max_id < 0) ? 0 : max_id;
526 case MSG_STAT:
527 case IPC_STAT:
529 struct msqid64_ds tbuf;
530 int success_return;
532 if (!buf)
533 return -EFAULT;
535 memset(&tbuf, 0, sizeof(tbuf));
537 rcu_read_lock();
538 if (cmd == MSG_STAT) {
539 msq = msq_obtain_object(ns, msqid);
540 if (IS_ERR(msq)) {
541 err = PTR_ERR(msq);
542 goto out_unlock;
544 success_return = msq->q_perm.id;
545 } else {
546 msq = msq_obtain_object_check(ns, msqid);
547 if (IS_ERR(msq)) {
548 err = PTR_ERR(msq);
549 goto out_unlock;
551 success_return = 0;
554 err = -EACCES;
555 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
556 goto out_unlock;
558 err = security_msg_queue_msgctl(msq, cmd);
559 if (err)
560 goto out_unlock;
562 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
563 tbuf.msg_stime = msq->q_stime;
564 tbuf.msg_rtime = msq->q_rtime;
565 tbuf.msg_ctime = msq->q_ctime;
566 tbuf.msg_cbytes = msq->q_cbytes;
567 tbuf.msg_qnum = msq->q_qnum;
568 tbuf.msg_qbytes = msq->q_qbytes;
569 tbuf.msg_lspid = msq->q_lspid;
570 tbuf.msg_lrpid = msq->q_lrpid;
571 rcu_read_unlock();
573 if (copy_msqid_to_user(buf, &tbuf, version))
574 return -EFAULT;
575 return success_return;
578 default:
579 return -EINVAL;
582 return err;
583 out_unlock:
584 rcu_read_unlock();
585 return err;
588 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
590 int version;
591 struct ipc_namespace *ns;
593 if (msqid < 0 || cmd < 0)
594 return -EINVAL;
596 version = ipc_parse_version(&cmd);
597 ns = current->nsproxy->ipc_ns;
599 switch (cmd) {
600 case IPC_INFO:
601 case MSG_INFO:
602 case MSG_STAT: /* msqid is an index rather than a msg queue id */
603 case IPC_STAT:
604 return msgctl_nolock(ns, msqid, cmd, version, buf);
605 case IPC_SET:
606 case IPC_RMID:
607 return msgctl_down(ns, msqid, cmd, buf, version);
608 default:
609 return -EINVAL;
613 static int testmsg(struct msg_msg *msg, long type, int mode)
615 switch (mode)
617 case SEARCH_ANY:
618 case SEARCH_NUMBER:
619 return 1;
620 case SEARCH_LESSEQUAL:
621 if (msg->m_type <= type)
622 return 1;
623 break;
624 case SEARCH_EQUAL:
625 if (msg->m_type == type)
626 return 1;
627 break;
628 case SEARCH_NOTEQUAL:
629 if (msg->m_type != type)
630 return 1;
631 break;
633 return 0;
636 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
638 struct msg_receiver *msr, *t;
640 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
641 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
642 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
643 msr->r_msgtype, msr->r_mode)) {
645 list_del(&msr->r_list);
646 if (msr->r_maxsize < msg->m_ts) {
647 /* initialize pipelined send ordering */
648 msr->r_msg = NULL;
649 wake_up_process(msr->r_tsk);
650 smp_mb(); /* see barrier comment below */
651 msr->r_msg = ERR_PTR(-E2BIG);
652 } else {
653 msr->r_msg = NULL;
654 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
655 msq->q_rtime = get_seconds();
656 wake_up_process(msr->r_tsk);
658 * Ensure that the wakeup is visible before
659 * setting r_msg, as the receiving end depends
660 * on it. See lockless receive part 1 and 2 in
661 * do_msgrcv().
663 smp_mb();
664 msr->r_msg = msg;
666 return 1;
671 return 0;
674 long do_msgsnd(int msqid, long mtype, void __user *mtext,
675 size_t msgsz, int msgflg)
677 struct msg_queue *msq;
678 struct msg_msg *msg;
679 int err;
680 struct ipc_namespace *ns;
682 ns = current->nsproxy->ipc_ns;
684 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
685 return -EINVAL;
686 if (mtype < 1)
687 return -EINVAL;
689 msg = load_msg(mtext, msgsz);
690 if (IS_ERR(msg))
691 return PTR_ERR(msg);
693 msg->m_type = mtype;
694 msg->m_ts = msgsz;
696 rcu_read_lock();
697 msq = msq_obtain_object_check(ns, msqid);
698 if (IS_ERR(msq)) {
699 err = PTR_ERR(msq);
700 goto out_unlock1;
703 ipc_lock_object(&msq->q_perm);
705 for (;;) {
706 struct msg_sender s;
708 err = -EACCES;
709 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
710 goto out_unlock0;
712 /* raced with RMID? */
713 if (!ipc_valid_object(&msq->q_perm)) {
714 err = -EIDRM;
715 goto out_unlock0;
718 err = security_msg_queue_msgsnd(msq, msg, msgflg);
719 if (err)
720 goto out_unlock0;
722 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
723 1 + msq->q_qnum <= msq->q_qbytes) {
724 break;
727 /* queue full, wait: */
728 if (msgflg & IPC_NOWAIT) {
729 err = -EAGAIN;
730 goto out_unlock0;
733 /* enqueue the sender and prepare to block */
734 ss_add(msq, &s);
736 if (!ipc_rcu_getref(msq)) {
737 err = -EIDRM;
738 goto out_unlock0;
741 ipc_unlock_object(&msq->q_perm);
742 rcu_read_unlock();
743 schedule();
745 rcu_read_lock();
746 ipc_lock_object(&msq->q_perm);
748 ipc_rcu_putref(msq, ipc_rcu_free);
749 /* raced with RMID? */
750 if (!ipc_valid_object(&msq->q_perm)) {
751 err = -EIDRM;
752 goto out_unlock0;
755 ss_del(&s);
757 if (signal_pending(current)) {
758 err = -ERESTARTNOHAND;
759 goto out_unlock0;
763 msq->q_lspid = task_tgid_vnr(current);
764 msq->q_stime = get_seconds();
766 if (!pipelined_send(msq, msg)) {
767 /* no one is waiting for this message, enqueue it */
768 list_add_tail(&msg->m_list, &msq->q_messages);
769 msq->q_cbytes += msgsz;
770 msq->q_qnum++;
771 atomic_add(msgsz, &ns->msg_bytes);
772 atomic_inc(&ns->msg_hdrs);
775 err = 0;
776 msg = NULL;
778 out_unlock0:
779 ipc_unlock_object(&msq->q_perm);
780 out_unlock1:
781 rcu_read_unlock();
782 if (msg != NULL)
783 free_msg(msg);
784 return err;
787 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
788 int, msgflg)
790 long mtype;
792 if (get_user(mtype, &msgp->mtype))
793 return -EFAULT;
794 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
797 static inline int convert_mode(long *msgtyp, int msgflg)
799 if (msgflg & MSG_COPY)
800 return SEARCH_NUMBER;
802 * find message of correct type.
803 * msgtyp = 0 => get first.
804 * msgtyp > 0 => get first message of matching type.
805 * msgtyp < 0 => get message with least type must be < abs(msgtype).
807 if (*msgtyp == 0)
808 return SEARCH_ANY;
809 if (*msgtyp < 0) {
810 *msgtyp = -*msgtyp;
811 return SEARCH_LESSEQUAL;
813 if (msgflg & MSG_EXCEPT)
814 return SEARCH_NOTEQUAL;
815 return SEARCH_EQUAL;
818 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
820 struct msgbuf __user *msgp = dest;
821 size_t msgsz;
823 if (put_user(msg->m_type, &msgp->mtype))
824 return -EFAULT;
826 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
827 if (store_msg(msgp->mtext, msg, msgsz))
828 return -EFAULT;
829 return msgsz;
832 #ifdef CONFIG_CHECKPOINT_RESTORE
834 * This function creates new kernel message structure, large enough to store
835 * bufsz message bytes.
837 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
839 struct msg_msg *copy;
842 * Create dummy message to copy real message to.
844 copy = load_msg(buf, bufsz);
845 if (!IS_ERR(copy))
846 copy->m_ts = bufsz;
847 return copy;
850 static inline void free_copy(struct msg_msg *copy)
852 if (copy)
853 free_msg(copy);
855 #else
856 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
858 return ERR_PTR(-ENOSYS);
861 static inline void free_copy(struct msg_msg *copy)
864 #endif
866 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
868 struct msg_msg *msg, *found = NULL;
869 long count = 0;
871 list_for_each_entry(msg, &msq->q_messages, m_list) {
872 if (testmsg(msg, *msgtyp, mode) &&
873 !security_msg_queue_msgrcv(msq, msg, current,
874 *msgtyp, mode)) {
875 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
876 *msgtyp = msg->m_type - 1;
877 found = msg;
878 } else if (mode == SEARCH_NUMBER) {
879 if (*msgtyp == count)
880 return msg;
881 } else
882 return msg;
883 count++;
887 return found ?: ERR_PTR(-EAGAIN);
890 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp, int msgflg,
891 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
893 int mode;
894 struct msg_queue *msq;
895 struct ipc_namespace *ns;
896 struct msg_msg *msg, *copy = NULL;
898 ns = current->nsproxy->ipc_ns;
900 if (msqid < 0 || (long) bufsz < 0)
901 return -EINVAL;
903 if (msgflg & MSG_COPY) {
904 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
905 if (IS_ERR(copy))
906 return PTR_ERR(copy);
908 mode = convert_mode(&msgtyp, msgflg);
910 rcu_read_lock();
911 msq = msq_obtain_object_check(ns, msqid);
912 if (IS_ERR(msq)) {
913 rcu_read_unlock();
914 free_copy(copy);
915 return PTR_ERR(msq);
918 for (;;) {
919 struct msg_receiver msr_d;
921 msg = ERR_PTR(-EACCES);
922 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
923 goto out_unlock1;
925 ipc_lock_object(&msq->q_perm);
927 /* raced with RMID? */
928 if (!ipc_valid_object(&msq->q_perm)) {
929 msg = ERR_PTR(-EIDRM);
930 goto out_unlock0;
933 msg = find_msg(msq, &msgtyp, mode);
934 if (!IS_ERR(msg)) {
936 * Found a suitable message.
937 * Unlink it from the queue.
939 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
940 msg = ERR_PTR(-E2BIG);
941 goto out_unlock0;
944 * If we are copying, then do not unlink message and do
945 * not update queue parameters.
947 if (msgflg & MSG_COPY) {
948 msg = copy_msg(msg, copy);
949 goto out_unlock0;
952 list_del(&msg->m_list);
953 msq->q_qnum--;
954 msq->q_rtime = get_seconds();
955 msq->q_lrpid = task_tgid_vnr(current);
956 msq->q_cbytes -= msg->m_ts;
957 atomic_sub(msg->m_ts, &ns->msg_bytes);
958 atomic_dec(&ns->msg_hdrs);
959 ss_wakeup(&msq->q_senders, 0);
961 goto out_unlock0;
964 /* No message waiting. Wait for a message */
965 if (msgflg & IPC_NOWAIT) {
966 msg = ERR_PTR(-ENOMSG);
967 goto out_unlock0;
970 list_add_tail(&msr_d.r_list, &msq->q_receivers);
971 msr_d.r_tsk = current;
972 msr_d.r_msgtype = msgtyp;
973 msr_d.r_mode = mode;
974 if (msgflg & MSG_NOERROR)
975 msr_d.r_maxsize = INT_MAX;
976 else
977 msr_d.r_maxsize = bufsz;
978 msr_d.r_msg = ERR_PTR(-EAGAIN);
979 current->state = TASK_INTERRUPTIBLE;
981 ipc_unlock_object(&msq->q_perm);
982 rcu_read_unlock();
983 schedule();
985 /* Lockless receive, part 1:
986 * Disable preemption. We don't hold a reference to the queue
987 * and getting a reference would defeat the idea of a lockless
988 * operation, thus the code relies on rcu to guarantee the
989 * existence of msq:
990 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
991 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
992 * rcu_read_lock() prevents preemption between reading r_msg
993 * and acquiring the q_perm.lock in ipc_lock_object().
995 rcu_read_lock();
997 /* Lockless receive, part 2:
998 * Wait until pipelined_send or expunge_all are outside of
999 * wake_up_process(). There is a race with exit(), see
1000 * ipc/mqueue.c for the details.
1002 msg = (struct msg_msg *)msr_d.r_msg;
1003 while (msg == NULL) {
1004 cpu_relax();
1005 msg = (struct msg_msg *)msr_d.r_msg;
1008 /* Lockless receive, part 3:
1009 * If there is a message or an error then accept it without
1010 * locking.
1012 if (msg != ERR_PTR(-EAGAIN))
1013 goto out_unlock1;
1015 /* Lockless receive, part 3:
1016 * Acquire the queue spinlock.
1018 ipc_lock_object(&msq->q_perm);
1020 /* Lockless receive, part 4:
1021 * Repeat test after acquiring the spinlock.
1023 msg = (struct msg_msg *)msr_d.r_msg;
1024 if (msg != ERR_PTR(-EAGAIN))
1025 goto out_unlock0;
1027 list_del(&msr_d.r_list);
1028 if (signal_pending(current)) {
1029 msg = ERR_PTR(-ERESTARTNOHAND);
1030 goto out_unlock0;
1033 ipc_unlock_object(&msq->q_perm);
1036 out_unlock0:
1037 ipc_unlock_object(&msq->q_perm);
1038 out_unlock1:
1039 rcu_read_unlock();
1040 if (IS_ERR(msg)) {
1041 free_copy(copy);
1042 return PTR_ERR(msg);
1045 bufsz = msg_handler(buf, msg, bufsz);
1046 free_msg(msg);
1048 return bufsz;
1051 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
1052 long, msgtyp, int, msgflg)
1054 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
1057 #ifdef CONFIG_PROC_FS
1058 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
1060 struct user_namespace *user_ns = seq_user_ns(s);
1061 struct msg_queue *msq = it;
1063 return seq_printf(s,
1064 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
1065 msq->q_perm.key,
1066 msq->q_perm.id,
1067 msq->q_perm.mode,
1068 msq->q_cbytes,
1069 msq->q_qnum,
1070 msq->q_lspid,
1071 msq->q_lrpid,
1072 from_kuid_munged(user_ns, msq->q_perm.uid),
1073 from_kgid_munged(user_ns, msq->q_perm.gid),
1074 from_kuid_munged(user_ns, msq->q_perm.cuid),
1075 from_kgid_munged(user_ns, msq->q_perm.cgid),
1076 msq->q_stime,
1077 msq->q_rtime,
1078 msq->q_ctime);
1080 #endif