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 <asm/current.h>
30 #include <asm/uaccess.h>
34 int msg_ctlmax
= MSGMAX
;
35 int msg_ctlmnb
= MSGMNB
;
36 int msg_ctlmni
= MSGMNI
;
38 /* one msg_receiver structure for each sleeping receiver */
40 struct list_head r_list
;
41 struct task_struct
* r_tsk
;
47 struct msg_msg
* volatile r_msg
;
50 /* one msg_sender for each sleeping sender */
52 struct list_head list
;
53 struct task_struct
* tsk
;
57 #define SEARCH_EQUAL 2
58 #define SEARCH_NOTEQUAL 3
59 #define SEARCH_LESSEQUAL 4
61 static atomic_t msg_bytes
= ATOMIC_INIT(0);
62 static atomic_t msg_hdrs
= ATOMIC_INIT(0);
64 static struct ipc_ids msg_ids
;
66 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
67 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
68 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
69 #define msg_checkid(msq, msgid) \
70 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
71 #define msg_buildid(id, seq) \
72 ipc_buildid(&msg_ids, id, seq)
74 static void freeque (struct msg_queue
*msq
, int id
);
75 static int newque (key_t key
, int msgflg
);
77 static int sysvipc_msg_read_proc(char *buffer
, char **start
, off_t offset
, int length
, int *eof
, void *data
);
80 void __init
msg_init (void)
82 ipc_init_ids(&msg_ids
,msg_ctlmni
);
85 create_proc_read_entry("sysvipc/msg", 0, NULL
, sysvipc_msg_read_proc
, NULL
);
89 static int newque (key_t key
, int msgflg
)
93 struct msg_queue
*msq
;
95 msq
= ipc_rcu_alloc(sizeof(*msq
));
99 msq
->q_perm
.mode
= (msgflg
& S_IRWXUGO
);
100 msq
->q_perm
.key
= key
;
102 msq
->q_perm
.security
= NULL
;
103 retval
= security_msg_queue_alloc(msq
);
109 id
= ipc_addid(&msg_ids
, &msq
->q_perm
, msg_ctlmni
);
111 security_msg_queue_free(msq
);
116 msq
->q_stime
= msq
->q_rtime
= 0;
117 msq
->q_ctime
= get_seconds();
118 msq
->q_cbytes
= msq
->q_qnum
= 0;
119 msq
->q_qbytes
= msg_ctlmnb
;
120 msq
->q_lspid
= msq
->q_lrpid
= 0;
121 INIT_LIST_HEAD(&msq
->q_messages
);
122 INIT_LIST_HEAD(&msq
->q_receivers
);
123 INIT_LIST_HEAD(&msq
->q_senders
);
126 return msg_buildid(id
,msq
->q_perm
.seq
);
129 static inline void ss_add(struct msg_queue
* msq
, struct msg_sender
* mss
)
132 current
->state
=TASK_INTERRUPTIBLE
;
133 list_add_tail(&mss
->list
,&msq
->q_senders
);
136 static inline void ss_del(struct msg_sender
* mss
)
138 if(mss
->list
.next
!= NULL
)
139 list_del(&mss
->list
);
142 static void ss_wakeup(struct list_head
* h
, int kill
)
144 struct list_head
*tmp
;
148 struct msg_sender
* mss
;
150 mss
= list_entry(tmp
,struct msg_sender
,list
);
154 wake_up_process(mss
->tsk
);
158 static void expunge_all(struct msg_queue
* msq
, int res
)
160 struct list_head
*tmp
;
162 tmp
= msq
->q_receivers
.next
;
163 while (tmp
!= &msq
->q_receivers
) {
164 struct msg_receiver
* msr
;
166 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
169 wake_up_process(msr
->r_tsk
);
171 msr
->r_msg
= ERR_PTR(res
);
175 * freeque() wakes up waiters on the sender and receiver waiting queue,
176 * removes the message queue from message queue ID
177 * array, and cleans up all the messages associated with this queue.
179 * msg_ids.sem and the spinlock for this message queue is hold
180 * before freeque() is called. msg_ids.sem remains locked on exit.
182 static void freeque (struct msg_queue
*msq
, int id
)
184 struct list_head
*tmp
;
186 expunge_all(msq
,-EIDRM
);
187 ss_wakeup(&msq
->q_senders
,1);
191 tmp
= msq
->q_messages
.next
;
192 while(tmp
!= &msq
->q_messages
) {
193 struct msg_msg
* msg
= list_entry(tmp
,struct msg_msg
,m_list
);
195 atomic_dec(&msg_hdrs
);
198 atomic_sub(msq
->q_cbytes
, &msg_bytes
);
199 security_msg_queue_free(msq
);
203 asmlinkage
long sys_msgget (key_t key
, int msgflg
)
205 int id
, ret
= -EPERM
;
206 struct msg_queue
*msq
;
209 if (key
== IPC_PRIVATE
)
210 ret
= newque(key
, msgflg
);
211 else if ((id
= ipc_findkey(&msg_ids
, key
)) == -1) { /* key not used */
212 if (!(msgflg
& IPC_CREAT
))
215 ret
= newque(key
, msgflg
);
216 } else if (msgflg
& IPC_CREAT
&& msgflg
& IPC_EXCL
) {
222 if (ipcperms(&msq
->q_perm
, msgflg
))
225 int qid
= msg_buildid(id
, msq
->q_perm
.seq
);
226 ret
= security_msg_queue_associate(msq
, msgflg
);
236 static inline unsigned long copy_msqid_to_user(void __user
*buf
, struct msqid64_ds
*in
, int version
)
240 return copy_to_user (buf
, in
, sizeof(*in
));
245 memset(&out
,0,sizeof(out
));
247 ipc64_perm_to_ipc_perm(&in
->msg_perm
, &out
.msg_perm
);
249 out
.msg_stime
= in
->msg_stime
;
250 out
.msg_rtime
= in
->msg_rtime
;
251 out
.msg_ctime
= in
->msg_ctime
;
253 if(in
->msg_cbytes
> USHRT_MAX
)
254 out
.msg_cbytes
= USHRT_MAX
;
256 out
.msg_cbytes
= in
->msg_cbytes
;
257 out
.msg_lcbytes
= in
->msg_cbytes
;
259 if(in
->msg_qnum
> USHRT_MAX
)
260 out
.msg_qnum
= USHRT_MAX
;
262 out
.msg_qnum
= in
->msg_qnum
;
264 if(in
->msg_qbytes
> USHRT_MAX
)
265 out
.msg_qbytes
= USHRT_MAX
;
267 out
.msg_qbytes
= in
->msg_qbytes
;
268 out
.msg_lqbytes
= in
->msg_qbytes
;
270 out
.msg_lspid
= in
->msg_lspid
;
271 out
.msg_lrpid
= in
->msg_lrpid
;
273 return copy_to_user (buf
, &out
, sizeof(out
));
281 unsigned long qbytes
;
287 static inline unsigned long copy_msqid_from_user(struct msq_setbuf
*out
, void __user
*buf
, int version
)
292 struct msqid64_ds tbuf
;
294 if (copy_from_user (&tbuf
, buf
, sizeof (tbuf
)))
297 out
->qbytes
= tbuf
.msg_qbytes
;
298 out
->uid
= tbuf
.msg_perm
.uid
;
299 out
->gid
= tbuf
.msg_perm
.gid
;
300 out
->mode
= tbuf
.msg_perm
.mode
;
306 struct msqid_ds tbuf_old
;
308 if (copy_from_user (&tbuf_old
, buf
, sizeof (tbuf_old
)))
311 out
->uid
= tbuf_old
.msg_perm
.uid
;
312 out
->gid
= tbuf_old
.msg_perm
.gid
;
313 out
->mode
= tbuf_old
.msg_perm
.mode
;
315 if(tbuf_old
.msg_qbytes
== 0)
316 out
->qbytes
= tbuf_old
.msg_lqbytes
;
318 out
->qbytes
= tbuf_old
.msg_qbytes
;
327 asmlinkage
long sys_msgctl (int msqid
, int cmd
, struct msqid_ds __user
*buf
)
330 struct msg_queue
*msq
;
331 struct msq_setbuf setbuf
;
332 struct kern_ipc_perm
*ipcp
;
334 if (msqid
< 0 || cmd
< 0)
337 version
= ipc_parse_version(&cmd
);
343 struct msginfo msginfo
;
347 /* We must not return kernel stack data.
348 * due to padding, it's not enough
349 * to set all member fields.
352 err
= security_msg_queue_msgctl(NULL
, cmd
);
356 memset(&msginfo
,0,sizeof(msginfo
));
357 msginfo
.msgmni
= msg_ctlmni
;
358 msginfo
.msgmax
= msg_ctlmax
;
359 msginfo
.msgmnb
= msg_ctlmnb
;
360 msginfo
.msgssz
= MSGSSZ
;
361 msginfo
.msgseg
= MSGSEG
;
363 if (cmd
== MSG_INFO
) {
364 msginfo
.msgpool
= msg_ids
.in_use
;
365 msginfo
.msgmap
= atomic_read(&msg_hdrs
);
366 msginfo
.msgtql
= atomic_read(&msg_bytes
);
368 msginfo
.msgmap
= MSGMAP
;
369 msginfo
.msgpool
= MSGPOOL
;
370 msginfo
.msgtql
= MSGTQL
;
372 max_id
= msg_ids
.max_id
;
374 if (copy_to_user (buf
, &msginfo
, sizeof(struct msginfo
)))
376 return (max_id
< 0) ? 0: max_id
;
381 struct msqid64_ds tbuf
;
385 if(cmd
== MSG_STAT
&& msqid
>= msg_ids
.entries
->size
)
388 memset(&tbuf
,0,sizeof(tbuf
));
390 msq
= msg_lock(msqid
);
394 if(cmd
== MSG_STAT
) {
395 success_return
= msg_buildid(msqid
, msq
->q_perm
.seq
);
398 if (msg_checkid(msq
,msqid
))
403 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
406 err
= security_msg_queue_msgctl(msq
, cmd
);
410 kernel_to_ipc64_perm(&msq
->q_perm
, &tbuf
.msg_perm
);
411 tbuf
.msg_stime
= msq
->q_stime
;
412 tbuf
.msg_rtime
= msq
->q_rtime
;
413 tbuf
.msg_ctime
= msq
->q_ctime
;
414 tbuf
.msg_cbytes
= msq
->q_cbytes
;
415 tbuf
.msg_qnum
= msq
->q_qnum
;
416 tbuf
.msg_qbytes
= msq
->q_qbytes
;
417 tbuf
.msg_lspid
= msq
->q_lspid
;
418 tbuf
.msg_lrpid
= msq
->q_lrpid
;
420 if (copy_msqid_to_user(buf
, &tbuf
, version
))
422 return success_return
;
427 if (copy_msqid_from_user (&setbuf
, buf
, version
))
429 if ((err
= audit_ipc_perms(setbuf
.qbytes
, setbuf
.uid
, setbuf
.gid
, setbuf
.mode
)))
439 msq
= msg_lock(msqid
);
445 if (msg_checkid(msq
,msqid
))
449 if (current
->euid
!= ipcp
->cuid
&&
450 current
->euid
!= ipcp
->uid
&& !capable(CAP_SYS_ADMIN
))
451 /* We _could_ check for CAP_CHOWN above, but we don't */
454 err
= security_msg_queue_msgctl(msq
, cmd
);
462 if (setbuf
.qbytes
> msg_ctlmnb
&& !capable(CAP_SYS_RESOURCE
))
465 msq
->q_qbytes
= setbuf
.qbytes
;
467 ipcp
->uid
= setbuf
.uid
;
468 ipcp
->gid
= setbuf
.gid
;
469 ipcp
->mode
= (ipcp
->mode
& ~S_IRWXUGO
) |
470 (S_IRWXUGO
& setbuf
.mode
);
471 msq
->q_ctime
= get_seconds();
472 /* sleeping receivers might be excluded by
473 * stricter permissions.
475 expunge_all(msq
,-EAGAIN
);
476 /* sleeping senders might be able to send
477 * due to a larger queue size.
479 ss_wakeup(&msq
->q_senders
,0);
484 freeque (msq
, msqid
);
499 static int testmsg(struct msg_msg
* msg
,long type
,int mode
)
505 case SEARCH_LESSEQUAL
:
506 if(msg
->m_type
<=type
)
510 if(msg
->m_type
== type
)
513 case SEARCH_NOTEQUAL
:
514 if(msg
->m_type
!= type
)
521 static inline int pipelined_send(struct msg_queue
* msq
, struct msg_msg
* msg
)
523 struct list_head
* tmp
;
525 tmp
= msq
->q_receivers
.next
;
526 while (tmp
!= &msq
->q_receivers
) {
527 struct msg_receiver
* msr
;
528 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
530 if(testmsg(msg
,msr
->r_msgtype
,msr
->r_mode
) &&
531 !security_msg_queue_msgrcv(msq
, msg
, msr
->r_tsk
, msr
->r_msgtype
, msr
->r_mode
)) {
532 list_del(&msr
->r_list
);
533 if(msr
->r_maxsize
< msg
->m_ts
) {
535 wake_up_process(msr
->r_tsk
);
537 msr
->r_msg
= ERR_PTR(-E2BIG
);
540 msq
->q_lrpid
= msr
->r_tsk
->pid
;
541 msq
->q_rtime
= get_seconds();
542 wake_up_process(msr
->r_tsk
);
552 asmlinkage
long sys_msgsnd (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
, int msgflg
)
554 struct msg_queue
*msq
;
559 if (msgsz
> msg_ctlmax
|| (long) msgsz
< 0 || msqid
< 0)
561 if (get_user(mtype
, &msgp
->mtype
))
566 msg
= load_msg(msgp
->mtext
, msgsz
);
573 msq
= msg_lock(msqid
);
579 if (msg_checkid(msq
,msqid
))
580 goto out_unlock_free
;
586 if (ipcperms(&msq
->q_perm
, S_IWUGO
))
587 goto out_unlock_free
;
589 err
= security_msg_queue_msgsnd(msq
, msg
, msgflg
);
591 goto out_unlock_free
;
593 if(msgsz
+ msq
->q_cbytes
<= msq
->q_qbytes
&&
594 1 + msq
->q_qnum
<= msq
->q_qbytes
) {
598 /* queue full, wait: */
599 if(msgflg
&IPC_NOWAIT
) {
601 goto out_unlock_free
;
608 ipc_lock_by_ptr(&msq
->q_perm
);
610 if (msq
->q_perm
.deleted
) {
612 goto out_unlock_free
;
616 if (signal_pending(current
)) {
618 goto out_unlock_free
;
622 msq
->q_lspid
= current
->tgid
;
623 msq
->q_stime
= get_seconds();
625 if(!pipelined_send(msq
,msg
)) {
626 /* noone is waiting for this message, enqueue it */
627 list_add_tail(&msg
->m_list
,&msq
->q_messages
);
628 msq
->q_cbytes
+= msgsz
;
630 atomic_add(msgsz
,&msg_bytes
);
631 atomic_inc(&msg_hdrs
);
645 static inline int convert_mode(long* msgtyp
, int msgflg
)
648 * find message of correct type.
649 * msgtyp = 0 => get first.
650 * msgtyp > 0 => get first message of matching type.
651 * msgtyp < 0 => get message with least type must be < abs(msgtype).
657 return SEARCH_LESSEQUAL
;
659 if(msgflg
& MSG_EXCEPT
)
660 return SEARCH_NOTEQUAL
;
664 asmlinkage
long sys_msgrcv (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
,
665 long msgtyp
, int msgflg
)
667 struct msg_queue
*msq
;
671 if (msqid
< 0 || (long) msgsz
< 0)
673 mode
= convert_mode(&msgtyp
,msgflg
);
675 msq
= msg_lock(msqid
);
679 msg
= ERR_PTR(-EIDRM
);
680 if (msg_checkid(msq
,msqid
))
684 struct msg_receiver msr_d
;
685 struct list_head
* tmp
;
687 msg
= ERR_PTR(-EACCES
);
688 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
691 msg
= ERR_PTR(-EAGAIN
);
692 tmp
= msq
->q_messages
.next
;
693 while (tmp
!= &msq
->q_messages
) {
694 struct msg_msg
*walk_msg
;
695 walk_msg
= list_entry(tmp
,struct msg_msg
,m_list
);
696 if(testmsg(walk_msg
,msgtyp
,mode
) &&
697 !security_msg_queue_msgrcv(msq
, walk_msg
, current
, msgtyp
, mode
)) {
699 if(mode
== SEARCH_LESSEQUAL
&& walk_msg
->m_type
!= 1) {
701 msgtyp
=walk_msg
->m_type
-1;
710 /* Found a suitable message. Unlink it from the queue. */
711 if ((msgsz
< msg
->m_ts
) && !(msgflg
& MSG_NOERROR
)) {
712 msg
= ERR_PTR(-E2BIG
);
715 list_del(&msg
->m_list
);
717 msq
->q_rtime
= get_seconds();
718 msq
->q_lrpid
= current
->tgid
;
719 msq
->q_cbytes
-= msg
->m_ts
;
720 atomic_sub(msg
->m_ts
,&msg_bytes
);
721 atomic_dec(&msg_hdrs
);
722 ss_wakeup(&msq
->q_senders
,0);
726 /* No message waiting. Wait for a message */
727 if (msgflg
& IPC_NOWAIT
) {
728 msg
= ERR_PTR(-ENOMSG
);
731 list_add_tail(&msr_d
.r_list
,&msq
->q_receivers
);
732 msr_d
.r_tsk
= current
;
733 msr_d
.r_msgtype
= msgtyp
;
735 if(msgflg
& MSG_NOERROR
)
736 msr_d
.r_maxsize
= INT_MAX
;
738 msr_d
.r_maxsize
= msgsz
;
739 msr_d
.r_msg
= ERR_PTR(-EAGAIN
);
740 current
->state
= TASK_INTERRUPTIBLE
;
745 /* Lockless receive, part 1:
746 * Disable preemption. We don't hold a reference to the queue
747 * and getting a reference would defeat the idea of a lockless
748 * operation, thus the code relies on rcu to guarantee the
750 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
751 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
752 * rcu_read_lock() prevents preemption between reading r_msg
753 * and the spin_lock() inside ipc_lock_by_ptr().
757 /* Lockless receive, part 2:
758 * Wait until pipelined_send or expunge_all are outside of
759 * wake_up_process(). There is a race with exit(), see
760 * ipc/mqueue.c for the details.
762 msg
= (struct msg_msg
*) msr_d
.r_msg
;
763 while (msg
== NULL
) {
765 msg
= (struct msg_msg
*) msr_d
.r_msg
;
768 /* Lockless receive, part 3:
769 * If there is a message or an error then accept it without
772 if(msg
!= ERR_PTR(-EAGAIN
)) {
777 /* Lockless receive, part 3:
778 * Acquire the queue spinlock.
780 ipc_lock_by_ptr(&msq
->q_perm
);
783 /* Lockless receive, part 4:
784 * Repeat test after acquiring the spinlock.
786 msg
= (struct msg_msg
*)msr_d
.r_msg
;
787 if(msg
!= ERR_PTR(-EAGAIN
))
790 list_del(&msr_d
.r_list
);
791 if (signal_pending(current
)) {
792 msg
= ERR_PTR(-ERESTARTNOHAND
);
801 msgsz
= (msgsz
> msg
->m_ts
) ? msg
->m_ts
: msgsz
;
802 if (put_user (msg
->m_type
, &msgp
->mtype
) ||
803 store_msg(msgp
->mtext
, msg
, msgsz
)) {
810 #ifdef CONFIG_PROC_FS
811 static int sysvipc_msg_read_proc(char *buffer
, char **start
, off_t offset
, int length
, int *eof
, void *data
)
818 len
+= sprintf(buffer
, " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n");
820 for(i
= 0; i
<= msg_ids
.max_id
; i
++) {
821 struct msg_queue
* msq
;
824 len
+= sprintf(buffer
+ len
, "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
826 msg_buildid(i
,msq
->q_perm
.seq
),
846 if(pos
> offset
+ length
)
854 *start
= buffer
+ (offset
- begin
);
855 len
-= (offset
- begin
);