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 <manfreds@colorfullife.com>
18 #include <linux/config.h>
19 #include <linux/slab.h>
20 #include <linux/msg.h>
21 #include <linux/spinlock.h>
22 #include <linux/init.h>
23 #include <linux/proc_fs.h>
24 #include <linux/list.h>
25 #include <linux/security.h>
26 #include <linux/sched.h>
27 #include <linux/syscalls.h>
28 #include <linux/audit.h>
29 #include <linux/seq_file.h>
30 #include <asm/current.h>
31 #include <asm/uaccess.h>
35 int msg_ctlmax
= MSGMAX
;
36 int msg_ctlmnb
= MSGMNB
;
37 int msg_ctlmni
= MSGMNI
;
39 /* one msg_receiver structure for each sleeping receiver */
41 struct list_head r_list
;
42 struct task_struct
* r_tsk
;
48 struct msg_msg
* volatile r_msg
;
51 /* one msg_sender for each sleeping sender */
53 struct list_head list
;
54 struct task_struct
* tsk
;
58 #define SEARCH_EQUAL 2
59 #define SEARCH_NOTEQUAL 3
60 #define SEARCH_LESSEQUAL 4
62 static atomic_t msg_bytes
= ATOMIC_INIT(0);
63 static atomic_t msg_hdrs
= ATOMIC_INIT(0);
65 static struct ipc_ids msg_ids
;
67 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
68 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
69 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
70 #define msg_checkid(msq, msgid) \
71 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
72 #define msg_buildid(id, seq) \
73 ipc_buildid(&msg_ids, id, seq)
75 static void freeque (struct msg_queue
*msq
, int id
);
76 static int newque (key_t key
, int msgflg
);
78 static int sysvipc_msg_proc_show(struct seq_file
*s
, void *it
);
81 void __init
msg_init (void)
83 ipc_init_ids(&msg_ids
,msg_ctlmni
);
84 ipc_init_proc_interface("sysvipc/msg",
85 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
87 sysvipc_msg_proc_show
);
90 static int newque (key_t key
, int msgflg
)
94 struct msg_queue
*msq
;
96 msq
= ipc_rcu_alloc(sizeof(*msq
));
100 msq
->q_perm
.mode
= (msgflg
& S_IRWXUGO
);
101 msq
->q_perm
.key
= key
;
103 msq
->q_perm
.security
= NULL
;
104 retval
= security_msg_queue_alloc(msq
);
110 id
= ipc_addid(&msg_ids
, &msq
->q_perm
, msg_ctlmni
);
112 security_msg_queue_free(msq
);
117 msq
->q_id
= msg_buildid(id
,msq
->q_perm
.seq
);
118 msq
->q_stime
= msq
->q_rtime
= 0;
119 msq
->q_ctime
= get_seconds();
120 msq
->q_cbytes
= msq
->q_qnum
= 0;
121 msq
->q_qbytes
= msg_ctlmnb
;
122 msq
->q_lspid
= msq
->q_lrpid
= 0;
123 INIT_LIST_HEAD(&msq
->q_messages
);
124 INIT_LIST_HEAD(&msq
->q_receivers
);
125 INIT_LIST_HEAD(&msq
->q_senders
);
131 static inline void ss_add(struct msg_queue
* msq
, struct msg_sender
* mss
)
134 current
->state
=TASK_INTERRUPTIBLE
;
135 list_add_tail(&mss
->list
,&msq
->q_senders
);
138 static inline void ss_del(struct msg_sender
* mss
)
140 if(mss
->list
.next
!= NULL
)
141 list_del(&mss
->list
);
144 static void ss_wakeup(struct list_head
* h
, int kill
)
146 struct list_head
*tmp
;
150 struct msg_sender
* mss
;
152 mss
= list_entry(tmp
,struct msg_sender
,list
);
156 wake_up_process(mss
->tsk
);
160 static void expunge_all(struct msg_queue
* msq
, int res
)
162 struct list_head
*tmp
;
164 tmp
= msq
->q_receivers
.next
;
165 while (tmp
!= &msq
->q_receivers
) {
166 struct msg_receiver
* msr
;
168 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
171 wake_up_process(msr
->r_tsk
);
173 msr
->r_msg
= ERR_PTR(res
);
177 * freeque() wakes up waiters on the sender and receiver waiting queue,
178 * removes the message queue from message queue ID
179 * array, and cleans up all the messages associated with this queue.
181 * msg_ids.sem and the spinlock for this message queue is hold
182 * before freeque() is called. msg_ids.sem remains locked on exit.
184 static void freeque (struct msg_queue
*msq
, int id
)
186 struct list_head
*tmp
;
188 expunge_all(msq
,-EIDRM
);
189 ss_wakeup(&msq
->q_senders
,1);
193 tmp
= msq
->q_messages
.next
;
194 while(tmp
!= &msq
->q_messages
) {
195 struct msg_msg
* msg
= list_entry(tmp
,struct msg_msg
,m_list
);
197 atomic_dec(&msg_hdrs
);
200 atomic_sub(msq
->q_cbytes
, &msg_bytes
);
201 security_msg_queue_free(msq
);
205 asmlinkage
long sys_msgget (key_t key
, int msgflg
)
207 int id
, ret
= -EPERM
;
208 struct msg_queue
*msq
;
211 if (key
== IPC_PRIVATE
)
212 ret
= newque(key
, msgflg
);
213 else if ((id
= ipc_findkey(&msg_ids
, key
)) == -1) { /* key not used */
214 if (!(msgflg
& IPC_CREAT
))
217 ret
= newque(key
, msgflg
);
218 } else if (msgflg
& IPC_CREAT
&& msgflg
& IPC_EXCL
) {
224 if (ipcperms(&msq
->q_perm
, msgflg
))
227 int qid
= msg_buildid(id
, msq
->q_perm
.seq
);
228 ret
= security_msg_queue_associate(msq
, msgflg
);
238 static inline unsigned long copy_msqid_to_user(void __user
*buf
, struct msqid64_ds
*in
, int version
)
242 return copy_to_user (buf
, in
, sizeof(*in
));
247 memset(&out
,0,sizeof(out
));
249 ipc64_perm_to_ipc_perm(&in
->msg_perm
, &out
.msg_perm
);
251 out
.msg_stime
= in
->msg_stime
;
252 out
.msg_rtime
= in
->msg_rtime
;
253 out
.msg_ctime
= in
->msg_ctime
;
255 if(in
->msg_cbytes
> USHRT_MAX
)
256 out
.msg_cbytes
= USHRT_MAX
;
258 out
.msg_cbytes
= in
->msg_cbytes
;
259 out
.msg_lcbytes
= in
->msg_cbytes
;
261 if(in
->msg_qnum
> USHRT_MAX
)
262 out
.msg_qnum
= USHRT_MAX
;
264 out
.msg_qnum
= in
->msg_qnum
;
266 if(in
->msg_qbytes
> USHRT_MAX
)
267 out
.msg_qbytes
= USHRT_MAX
;
269 out
.msg_qbytes
= in
->msg_qbytes
;
270 out
.msg_lqbytes
= in
->msg_qbytes
;
272 out
.msg_lspid
= in
->msg_lspid
;
273 out
.msg_lrpid
= in
->msg_lrpid
;
275 return copy_to_user (buf
, &out
, sizeof(out
));
283 unsigned long qbytes
;
289 static inline unsigned long copy_msqid_from_user(struct msq_setbuf
*out
, void __user
*buf
, int version
)
294 struct msqid64_ds tbuf
;
296 if (copy_from_user (&tbuf
, buf
, sizeof (tbuf
)))
299 out
->qbytes
= tbuf
.msg_qbytes
;
300 out
->uid
= tbuf
.msg_perm
.uid
;
301 out
->gid
= tbuf
.msg_perm
.gid
;
302 out
->mode
= tbuf
.msg_perm
.mode
;
308 struct msqid_ds tbuf_old
;
310 if (copy_from_user (&tbuf_old
, buf
, sizeof (tbuf_old
)))
313 out
->uid
= tbuf_old
.msg_perm
.uid
;
314 out
->gid
= tbuf_old
.msg_perm
.gid
;
315 out
->mode
= tbuf_old
.msg_perm
.mode
;
317 if(tbuf_old
.msg_qbytes
== 0)
318 out
->qbytes
= tbuf_old
.msg_lqbytes
;
320 out
->qbytes
= tbuf_old
.msg_qbytes
;
329 asmlinkage
long sys_msgctl (int msqid
, int cmd
, struct msqid_ds __user
*buf
)
332 struct msg_queue
*msq
;
333 struct msq_setbuf setbuf
;
334 struct kern_ipc_perm
*ipcp
;
336 if (msqid
< 0 || cmd
< 0)
339 version
= ipc_parse_version(&cmd
);
345 struct msginfo msginfo
;
349 /* We must not return kernel stack data.
350 * due to padding, it's not enough
351 * to set all member fields.
354 err
= security_msg_queue_msgctl(NULL
, cmd
);
358 memset(&msginfo
,0,sizeof(msginfo
));
359 msginfo
.msgmni
= msg_ctlmni
;
360 msginfo
.msgmax
= msg_ctlmax
;
361 msginfo
.msgmnb
= msg_ctlmnb
;
362 msginfo
.msgssz
= MSGSSZ
;
363 msginfo
.msgseg
= MSGSEG
;
365 if (cmd
== MSG_INFO
) {
366 msginfo
.msgpool
= msg_ids
.in_use
;
367 msginfo
.msgmap
= atomic_read(&msg_hdrs
);
368 msginfo
.msgtql
= atomic_read(&msg_bytes
);
370 msginfo
.msgmap
= MSGMAP
;
371 msginfo
.msgpool
= MSGPOOL
;
372 msginfo
.msgtql
= MSGTQL
;
374 max_id
= msg_ids
.max_id
;
376 if (copy_to_user (buf
, &msginfo
, sizeof(struct msginfo
)))
378 return (max_id
< 0) ? 0: max_id
;
383 struct msqid64_ds tbuf
;
387 if(cmd
== MSG_STAT
&& msqid
>= msg_ids
.entries
->size
)
390 memset(&tbuf
,0,sizeof(tbuf
));
392 msq
= msg_lock(msqid
);
396 if(cmd
== MSG_STAT
) {
397 success_return
= msg_buildid(msqid
, msq
->q_perm
.seq
);
400 if (msg_checkid(msq
,msqid
))
405 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
408 err
= security_msg_queue_msgctl(msq
, cmd
);
412 kernel_to_ipc64_perm(&msq
->q_perm
, &tbuf
.msg_perm
);
413 tbuf
.msg_stime
= msq
->q_stime
;
414 tbuf
.msg_rtime
= msq
->q_rtime
;
415 tbuf
.msg_ctime
= msq
->q_ctime
;
416 tbuf
.msg_cbytes
= msq
->q_cbytes
;
417 tbuf
.msg_qnum
= msq
->q_qnum
;
418 tbuf
.msg_qbytes
= msq
->q_qbytes
;
419 tbuf
.msg_lspid
= msq
->q_lspid
;
420 tbuf
.msg_lrpid
= msq
->q_lrpid
;
422 if (copy_msqid_to_user(buf
, &tbuf
, version
))
424 return success_return
;
429 if (copy_msqid_from_user (&setbuf
, buf
, version
))
431 if ((err
= audit_ipc_perms(setbuf
.qbytes
, setbuf
.uid
, setbuf
.gid
, setbuf
.mode
)))
441 msq
= msg_lock(msqid
);
447 if (msg_checkid(msq
,msqid
))
451 if (current
->euid
!= ipcp
->cuid
&&
452 current
->euid
!= ipcp
->uid
&& !capable(CAP_SYS_ADMIN
))
453 /* We _could_ check for CAP_CHOWN above, but we don't */
456 err
= security_msg_queue_msgctl(msq
, cmd
);
464 if (setbuf
.qbytes
> msg_ctlmnb
&& !capable(CAP_SYS_RESOURCE
))
467 msq
->q_qbytes
= setbuf
.qbytes
;
469 ipcp
->uid
= setbuf
.uid
;
470 ipcp
->gid
= setbuf
.gid
;
471 ipcp
->mode
= (ipcp
->mode
& ~S_IRWXUGO
) |
472 (S_IRWXUGO
& setbuf
.mode
);
473 msq
->q_ctime
= get_seconds();
474 /* sleeping receivers might be excluded by
475 * stricter permissions.
477 expunge_all(msq
,-EAGAIN
);
478 /* sleeping senders might be able to send
479 * due to a larger queue size.
481 ss_wakeup(&msq
->q_senders
,0);
486 freeque (msq
, msqid
);
501 static int testmsg(struct msg_msg
* msg
,long type
,int mode
)
507 case SEARCH_LESSEQUAL
:
508 if(msg
->m_type
<=type
)
512 if(msg
->m_type
== type
)
515 case SEARCH_NOTEQUAL
:
516 if(msg
->m_type
!= type
)
523 static inline int pipelined_send(struct msg_queue
* msq
, struct msg_msg
* msg
)
525 struct list_head
* tmp
;
527 tmp
= msq
->q_receivers
.next
;
528 while (tmp
!= &msq
->q_receivers
) {
529 struct msg_receiver
* msr
;
530 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
532 if(testmsg(msg
,msr
->r_msgtype
,msr
->r_mode
) &&
533 !security_msg_queue_msgrcv(msq
, msg
, msr
->r_tsk
, msr
->r_msgtype
, msr
->r_mode
)) {
534 list_del(&msr
->r_list
);
535 if(msr
->r_maxsize
< msg
->m_ts
) {
537 wake_up_process(msr
->r_tsk
);
539 msr
->r_msg
= ERR_PTR(-E2BIG
);
542 msq
->q_lrpid
= msr
->r_tsk
->pid
;
543 msq
->q_rtime
= get_seconds();
544 wake_up_process(msr
->r_tsk
);
554 asmlinkage
long sys_msgsnd (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
, int msgflg
)
556 struct msg_queue
*msq
;
561 if (msgsz
> msg_ctlmax
|| (long) msgsz
< 0 || msqid
< 0)
563 if (get_user(mtype
, &msgp
->mtype
))
568 msg
= load_msg(msgp
->mtext
, msgsz
);
575 msq
= msg_lock(msqid
);
581 if (msg_checkid(msq
,msqid
))
582 goto out_unlock_free
;
588 if (ipcperms(&msq
->q_perm
, S_IWUGO
))
589 goto out_unlock_free
;
591 err
= security_msg_queue_msgsnd(msq
, msg
, msgflg
);
593 goto out_unlock_free
;
595 if(msgsz
+ msq
->q_cbytes
<= msq
->q_qbytes
&&
596 1 + msq
->q_qnum
<= msq
->q_qbytes
) {
600 /* queue full, wait: */
601 if(msgflg
&IPC_NOWAIT
) {
603 goto out_unlock_free
;
610 ipc_lock_by_ptr(&msq
->q_perm
);
612 if (msq
->q_perm
.deleted
) {
614 goto out_unlock_free
;
618 if (signal_pending(current
)) {
620 goto out_unlock_free
;
624 msq
->q_lspid
= current
->tgid
;
625 msq
->q_stime
= get_seconds();
627 if(!pipelined_send(msq
,msg
)) {
628 /* noone is waiting for this message, enqueue it */
629 list_add_tail(&msg
->m_list
,&msq
->q_messages
);
630 msq
->q_cbytes
+= msgsz
;
632 atomic_add(msgsz
,&msg_bytes
);
633 atomic_inc(&msg_hdrs
);
647 static inline int convert_mode(long* msgtyp
, int msgflg
)
650 * find message of correct type.
651 * msgtyp = 0 => get first.
652 * msgtyp > 0 => get first message of matching type.
653 * msgtyp < 0 => get message with least type must be < abs(msgtype).
659 return SEARCH_LESSEQUAL
;
661 if(msgflg
& MSG_EXCEPT
)
662 return SEARCH_NOTEQUAL
;
666 asmlinkage
long sys_msgrcv (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
,
667 long msgtyp
, int msgflg
)
669 struct msg_queue
*msq
;
673 if (msqid
< 0 || (long) msgsz
< 0)
675 mode
= convert_mode(&msgtyp
,msgflg
);
677 msq
= msg_lock(msqid
);
681 msg
= ERR_PTR(-EIDRM
);
682 if (msg_checkid(msq
,msqid
))
686 struct msg_receiver msr_d
;
687 struct list_head
* tmp
;
689 msg
= ERR_PTR(-EACCES
);
690 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
693 msg
= ERR_PTR(-EAGAIN
);
694 tmp
= msq
->q_messages
.next
;
695 while (tmp
!= &msq
->q_messages
) {
696 struct msg_msg
*walk_msg
;
697 walk_msg
= list_entry(tmp
,struct msg_msg
,m_list
);
698 if(testmsg(walk_msg
,msgtyp
,mode
) &&
699 !security_msg_queue_msgrcv(msq
, walk_msg
, current
, msgtyp
, mode
)) {
701 if(mode
== SEARCH_LESSEQUAL
&& walk_msg
->m_type
!= 1) {
703 msgtyp
=walk_msg
->m_type
-1;
712 /* Found a suitable message. Unlink it from the queue. */
713 if ((msgsz
< msg
->m_ts
) && !(msgflg
& MSG_NOERROR
)) {
714 msg
= ERR_PTR(-E2BIG
);
717 list_del(&msg
->m_list
);
719 msq
->q_rtime
= get_seconds();
720 msq
->q_lrpid
= current
->tgid
;
721 msq
->q_cbytes
-= msg
->m_ts
;
722 atomic_sub(msg
->m_ts
,&msg_bytes
);
723 atomic_dec(&msg_hdrs
);
724 ss_wakeup(&msq
->q_senders
,0);
728 /* No message waiting. Wait for a message */
729 if (msgflg
& IPC_NOWAIT
) {
730 msg
= ERR_PTR(-ENOMSG
);
733 list_add_tail(&msr_d
.r_list
,&msq
->q_receivers
);
734 msr_d
.r_tsk
= current
;
735 msr_d
.r_msgtype
= msgtyp
;
737 if(msgflg
& MSG_NOERROR
)
738 msr_d
.r_maxsize
= INT_MAX
;
740 msr_d
.r_maxsize
= msgsz
;
741 msr_d
.r_msg
= ERR_PTR(-EAGAIN
);
742 current
->state
= TASK_INTERRUPTIBLE
;
747 /* Lockless receive, part 1:
748 * Disable preemption. We don't hold a reference to the queue
749 * and getting a reference would defeat the idea of a lockless
750 * operation, thus the code relies on rcu to guarantee the
752 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
753 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
754 * rcu_read_lock() prevents preemption between reading r_msg
755 * and the spin_lock() inside ipc_lock_by_ptr().
759 /* Lockless receive, part 2:
760 * Wait until pipelined_send or expunge_all are outside of
761 * wake_up_process(). There is a race with exit(), see
762 * ipc/mqueue.c for the details.
764 msg
= (struct msg_msg
*) msr_d
.r_msg
;
765 while (msg
== NULL
) {
767 msg
= (struct msg_msg
*) msr_d
.r_msg
;
770 /* Lockless receive, part 3:
771 * If there is a message or an error then accept it without
774 if(msg
!= ERR_PTR(-EAGAIN
)) {
779 /* Lockless receive, part 3:
780 * Acquire the queue spinlock.
782 ipc_lock_by_ptr(&msq
->q_perm
);
785 /* Lockless receive, part 4:
786 * Repeat test after acquiring the spinlock.
788 msg
= (struct msg_msg
*)msr_d
.r_msg
;
789 if(msg
!= ERR_PTR(-EAGAIN
))
792 list_del(&msr_d
.r_list
);
793 if (signal_pending(current
)) {
794 msg
= ERR_PTR(-ERESTARTNOHAND
);
803 msgsz
= (msgsz
> msg
->m_ts
) ? msg
->m_ts
: msgsz
;
804 if (put_user (msg
->m_type
, &msgp
->mtype
) ||
805 store_msg(msgp
->mtext
, msg
, msgsz
)) {
812 #ifdef CONFIG_PROC_FS
813 static int sysvipc_msg_proc_show(struct seq_file
*s
, void *it
)
815 struct msg_queue
*msq
= it
;
818 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",