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>
21 #include <linux/capability.h>
22 #include <linux/config.h>
23 #include <linux/slab.h>
24 #include <linux/msg.h>
25 #include <linux/spinlock.h>
26 #include <linux/init.h>
27 #include <linux/proc_fs.h>
28 #include <linux/list.h>
29 #include <linux/security.h>
30 #include <linux/sched.h>
31 #include <linux/syscalls.h>
32 #include <linux/audit.h>
33 #include <linux/seq_file.h>
34 #include <linux/mutex.h>
36 #include <asm/current.h>
37 #include <asm/uaccess.h>
41 int msg_ctlmax
= MSGMAX
;
42 int msg_ctlmnb
= MSGMNB
;
43 int msg_ctlmni
= MSGMNI
;
45 /* one msg_receiver structure for each sleeping receiver */
47 struct list_head r_list
;
48 struct task_struct
* r_tsk
;
54 struct msg_msg
* volatile r_msg
;
57 /* one msg_sender for each sleeping sender */
59 struct list_head list
;
60 struct task_struct
* tsk
;
64 #define SEARCH_EQUAL 2
65 #define SEARCH_NOTEQUAL 3
66 #define SEARCH_LESSEQUAL 4
68 static atomic_t msg_bytes
= ATOMIC_INIT(0);
69 static atomic_t msg_hdrs
= ATOMIC_INIT(0);
71 static struct ipc_ids msg_ids
;
73 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
74 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
75 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
76 #define msg_checkid(msq, msgid) \
77 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
78 #define msg_buildid(id, seq) \
79 ipc_buildid(&msg_ids, id, seq)
81 static void freeque (struct msg_queue
*msq
, int id
);
82 static int newque (key_t key
, int msgflg
);
84 static int sysvipc_msg_proc_show(struct seq_file
*s
, void *it
);
87 void __init
msg_init (void)
89 ipc_init_ids(&msg_ids
,msg_ctlmni
);
90 ipc_init_proc_interface("sysvipc/msg",
91 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
93 sysvipc_msg_proc_show
);
96 static int newque (key_t key
, int msgflg
)
100 struct msg_queue
*msq
;
102 msq
= ipc_rcu_alloc(sizeof(*msq
));
106 msq
->q_perm
.mode
= (msgflg
& S_IRWXUGO
);
107 msq
->q_perm
.key
= key
;
109 msq
->q_perm
.security
= NULL
;
110 retval
= security_msg_queue_alloc(msq
);
116 id
= ipc_addid(&msg_ids
, &msq
->q_perm
, msg_ctlmni
);
118 security_msg_queue_free(msq
);
123 msq
->q_id
= msg_buildid(id
,msq
->q_perm
.seq
);
124 msq
->q_stime
= msq
->q_rtime
= 0;
125 msq
->q_ctime
= get_seconds();
126 msq
->q_cbytes
= msq
->q_qnum
= 0;
127 msq
->q_qbytes
= msg_ctlmnb
;
128 msq
->q_lspid
= msq
->q_lrpid
= 0;
129 INIT_LIST_HEAD(&msq
->q_messages
);
130 INIT_LIST_HEAD(&msq
->q_receivers
);
131 INIT_LIST_HEAD(&msq
->q_senders
);
137 static inline void ss_add(struct msg_queue
* msq
, struct msg_sender
* mss
)
140 current
->state
=TASK_INTERRUPTIBLE
;
141 list_add_tail(&mss
->list
,&msq
->q_senders
);
144 static inline void ss_del(struct msg_sender
* mss
)
146 if(mss
->list
.next
!= NULL
)
147 list_del(&mss
->list
);
150 static void ss_wakeup(struct list_head
* h
, int kill
)
152 struct list_head
*tmp
;
156 struct msg_sender
* mss
;
158 mss
= list_entry(tmp
,struct msg_sender
,list
);
162 wake_up_process(mss
->tsk
);
166 static void expunge_all(struct msg_queue
* msq
, int res
)
168 struct list_head
*tmp
;
170 tmp
= msq
->q_receivers
.next
;
171 while (tmp
!= &msq
->q_receivers
) {
172 struct msg_receiver
* msr
;
174 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
177 wake_up_process(msr
->r_tsk
);
179 msr
->r_msg
= ERR_PTR(res
);
183 * freeque() wakes up waiters on the sender and receiver waiting queue,
184 * removes the message queue from message queue ID
185 * array, and cleans up all the messages associated with this queue.
187 * msg_ids.mutex and the spinlock for this message queue is hold
188 * before freeque() is called. msg_ids.mutex remains locked on exit.
190 static void freeque (struct msg_queue
*msq
, int id
)
192 struct list_head
*tmp
;
194 expunge_all(msq
,-EIDRM
);
195 ss_wakeup(&msq
->q_senders
,1);
199 tmp
= msq
->q_messages
.next
;
200 while(tmp
!= &msq
->q_messages
) {
201 struct msg_msg
* msg
= list_entry(tmp
,struct msg_msg
,m_list
);
203 atomic_dec(&msg_hdrs
);
206 atomic_sub(msq
->q_cbytes
, &msg_bytes
);
207 security_msg_queue_free(msq
);
211 asmlinkage
long sys_msgget (key_t key
, int msgflg
)
213 int id
, ret
= -EPERM
;
214 struct msg_queue
*msq
;
216 mutex_lock(&msg_ids
.mutex
);
217 if (key
== IPC_PRIVATE
)
218 ret
= newque(key
, msgflg
);
219 else if ((id
= ipc_findkey(&msg_ids
, key
)) == -1) { /* key not used */
220 if (!(msgflg
& IPC_CREAT
))
223 ret
= newque(key
, msgflg
);
224 } else if (msgflg
& IPC_CREAT
&& msgflg
& IPC_EXCL
) {
229 if (ipcperms(&msq
->q_perm
, msgflg
))
232 int qid
= msg_buildid(id
, msq
->q_perm
.seq
);
233 ret
= security_msg_queue_associate(msq
, msgflg
);
239 mutex_unlock(&msg_ids
.mutex
);
243 static inline unsigned long copy_msqid_to_user(void __user
*buf
, struct msqid64_ds
*in
, int version
)
247 return copy_to_user (buf
, in
, sizeof(*in
));
252 memset(&out
,0,sizeof(out
));
254 ipc64_perm_to_ipc_perm(&in
->msg_perm
, &out
.msg_perm
);
256 out
.msg_stime
= in
->msg_stime
;
257 out
.msg_rtime
= in
->msg_rtime
;
258 out
.msg_ctime
= in
->msg_ctime
;
260 if(in
->msg_cbytes
> USHRT_MAX
)
261 out
.msg_cbytes
= USHRT_MAX
;
263 out
.msg_cbytes
= in
->msg_cbytes
;
264 out
.msg_lcbytes
= in
->msg_cbytes
;
266 if(in
->msg_qnum
> USHRT_MAX
)
267 out
.msg_qnum
= USHRT_MAX
;
269 out
.msg_qnum
= in
->msg_qnum
;
271 if(in
->msg_qbytes
> USHRT_MAX
)
272 out
.msg_qbytes
= USHRT_MAX
;
274 out
.msg_qbytes
= in
->msg_qbytes
;
275 out
.msg_lqbytes
= in
->msg_qbytes
;
277 out
.msg_lspid
= in
->msg_lspid
;
278 out
.msg_lrpid
= in
->msg_lrpid
;
280 return copy_to_user (buf
, &out
, sizeof(out
));
288 unsigned long qbytes
;
294 static inline unsigned long copy_msqid_from_user(struct msq_setbuf
*out
, void __user
*buf
, int version
)
299 struct msqid64_ds tbuf
;
301 if (copy_from_user (&tbuf
, buf
, sizeof (tbuf
)))
304 out
->qbytes
= tbuf
.msg_qbytes
;
305 out
->uid
= tbuf
.msg_perm
.uid
;
306 out
->gid
= tbuf
.msg_perm
.gid
;
307 out
->mode
= tbuf
.msg_perm
.mode
;
313 struct msqid_ds tbuf_old
;
315 if (copy_from_user (&tbuf_old
, buf
, sizeof (tbuf_old
)))
318 out
->uid
= tbuf_old
.msg_perm
.uid
;
319 out
->gid
= tbuf_old
.msg_perm
.gid
;
320 out
->mode
= tbuf_old
.msg_perm
.mode
;
322 if(tbuf_old
.msg_qbytes
== 0)
323 out
->qbytes
= tbuf_old
.msg_lqbytes
;
325 out
->qbytes
= tbuf_old
.msg_qbytes
;
334 asmlinkage
long sys_msgctl (int msqid
, int cmd
, struct msqid_ds __user
*buf
)
337 struct msg_queue
*msq
;
338 struct msq_setbuf setbuf
;
339 struct kern_ipc_perm
*ipcp
;
341 if (msqid
< 0 || cmd
< 0)
344 version
= ipc_parse_version(&cmd
);
350 struct msginfo msginfo
;
354 /* We must not return kernel stack data.
355 * due to padding, it's not enough
356 * to set all member fields.
359 err
= security_msg_queue_msgctl(NULL
, cmd
);
363 memset(&msginfo
,0,sizeof(msginfo
));
364 msginfo
.msgmni
= msg_ctlmni
;
365 msginfo
.msgmax
= msg_ctlmax
;
366 msginfo
.msgmnb
= msg_ctlmnb
;
367 msginfo
.msgssz
= MSGSSZ
;
368 msginfo
.msgseg
= MSGSEG
;
369 mutex_lock(&msg_ids
.mutex
);
370 if (cmd
== MSG_INFO
) {
371 msginfo
.msgpool
= msg_ids
.in_use
;
372 msginfo
.msgmap
= atomic_read(&msg_hdrs
);
373 msginfo
.msgtql
= atomic_read(&msg_bytes
);
375 msginfo
.msgmap
= MSGMAP
;
376 msginfo
.msgpool
= MSGPOOL
;
377 msginfo
.msgtql
= MSGTQL
;
379 max_id
= msg_ids
.max_id
;
380 mutex_unlock(&msg_ids
.mutex
);
381 if (copy_to_user (buf
, &msginfo
, sizeof(struct msginfo
)))
383 return (max_id
< 0) ? 0: max_id
;
388 struct msqid64_ds tbuf
;
392 if(cmd
== MSG_STAT
&& msqid
>= msg_ids
.entries
->size
)
395 memset(&tbuf
,0,sizeof(tbuf
));
397 msq
= msg_lock(msqid
);
401 if(cmd
== MSG_STAT
) {
402 success_return
= msg_buildid(msqid
, msq
->q_perm
.seq
);
405 if (msg_checkid(msq
,msqid
))
410 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
413 err
= security_msg_queue_msgctl(msq
, cmd
);
417 kernel_to_ipc64_perm(&msq
->q_perm
, &tbuf
.msg_perm
);
418 tbuf
.msg_stime
= msq
->q_stime
;
419 tbuf
.msg_rtime
= msq
->q_rtime
;
420 tbuf
.msg_ctime
= msq
->q_ctime
;
421 tbuf
.msg_cbytes
= msq
->q_cbytes
;
422 tbuf
.msg_qnum
= msq
->q_qnum
;
423 tbuf
.msg_qbytes
= msq
->q_qbytes
;
424 tbuf
.msg_lspid
= msq
->q_lspid
;
425 tbuf
.msg_lrpid
= msq
->q_lrpid
;
427 if (copy_msqid_to_user(buf
, &tbuf
, version
))
429 return success_return
;
434 if (copy_msqid_from_user (&setbuf
, buf
, version
))
443 mutex_lock(&msg_ids
.mutex
);
444 msq
= msg_lock(msqid
);
450 if (msg_checkid(msq
,msqid
))
454 err
= audit_ipc_obj(ipcp
);
459 if (current
->euid
!= ipcp
->cuid
&&
460 current
->euid
!= ipcp
->uid
&& !capable(CAP_SYS_ADMIN
))
461 /* We _could_ check for CAP_CHOWN above, but we don't */
464 err
= security_msg_queue_msgctl(msq
, cmd
);
471 err
= audit_ipc_set_perm(setbuf
.qbytes
, setbuf
.uid
, setbuf
.gid
, setbuf
.mode
, ipcp
);
476 if (setbuf
.qbytes
> msg_ctlmnb
&& !capable(CAP_SYS_RESOURCE
))
479 msq
->q_qbytes
= setbuf
.qbytes
;
481 ipcp
->uid
= setbuf
.uid
;
482 ipcp
->gid
= setbuf
.gid
;
483 ipcp
->mode
= (ipcp
->mode
& ~S_IRWXUGO
) |
484 (S_IRWXUGO
& setbuf
.mode
);
485 msq
->q_ctime
= get_seconds();
486 /* sleeping receivers might be excluded by
487 * stricter permissions.
489 expunge_all(msq
,-EAGAIN
);
490 /* sleeping senders might be able to send
491 * due to a larger queue size.
493 ss_wakeup(&msq
->q_senders
,0);
498 freeque (msq
, msqid
);
503 mutex_unlock(&msg_ids
.mutex
);
513 static int testmsg(struct msg_msg
* msg
,long type
,int mode
)
519 case SEARCH_LESSEQUAL
:
520 if(msg
->m_type
<=type
)
524 if(msg
->m_type
== type
)
527 case SEARCH_NOTEQUAL
:
528 if(msg
->m_type
!= type
)
535 static inline int pipelined_send(struct msg_queue
* msq
, struct msg_msg
* msg
)
537 struct list_head
* tmp
;
539 tmp
= msq
->q_receivers
.next
;
540 while (tmp
!= &msq
->q_receivers
) {
541 struct msg_receiver
* msr
;
542 msr
= list_entry(tmp
,struct msg_receiver
,r_list
);
544 if(testmsg(msg
,msr
->r_msgtype
,msr
->r_mode
) &&
545 !security_msg_queue_msgrcv(msq
, msg
, msr
->r_tsk
, msr
->r_msgtype
, msr
->r_mode
)) {
546 list_del(&msr
->r_list
);
547 if(msr
->r_maxsize
< msg
->m_ts
) {
549 wake_up_process(msr
->r_tsk
);
551 msr
->r_msg
= ERR_PTR(-E2BIG
);
554 msq
->q_lrpid
= msr
->r_tsk
->pid
;
555 msq
->q_rtime
= get_seconds();
556 wake_up_process(msr
->r_tsk
);
566 asmlinkage
long sys_msgsnd (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
, int msgflg
)
568 struct msg_queue
*msq
;
573 if (msgsz
> msg_ctlmax
|| (long) msgsz
< 0 || msqid
< 0)
575 if (get_user(mtype
, &msgp
->mtype
))
580 msg
= load_msg(msgp
->mtext
, msgsz
);
587 msq
= msg_lock(msqid
);
593 if (msg_checkid(msq
,msqid
))
594 goto out_unlock_free
;
600 if (ipcperms(&msq
->q_perm
, S_IWUGO
))
601 goto out_unlock_free
;
603 err
= security_msg_queue_msgsnd(msq
, msg
, msgflg
);
605 goto out_unlock_free
;
607 if(msgsz
+ msq
->q_cbytes
<= msq
->q_qbytes
&&
608 1 + msq
->q_qnum
<= msq
->q_qbytes
) {
612 /* queue full, wait: */
613 if(msgflg
&IPC_NOWAIT
) {
615 goto out_unlock_free
;
622 ipc_lock_by_ptr(&msq
->q_perm
);
624 if (msq
->q_perm
.deleted
) {
626 goto out_unlock_free
;
630 if (signal_pending(current
)) {
632 goto out_unlock_free
;
636 msq
->q_lspid
= current
->tgid
;
637 msq
->q_stime
= get_seconds();
639 if(!pipelined_send(msq
,msg
)) {
640 /* noone is waiting for this message, enqueue it */
641 list_add_tail(&msg
->m_list
,&msq
->q_messages
);
642 msq
->q_cbytes
+= msgsz
;
644 atomic_add(msgsz
,&msg_bytes
);
645 atomic_inc(&msg_hdrs
);
659 static inline int convert_mode(long* msgtyp
, int msgflg
)
662 * find message of correct type.
663 * msgtyp = 0 => get first.
664 * msgtyp > 0 => get first message of matching type.
665 * msgtyp < 0 => get message with least type must be < abs(msgtype).
671 return SEARCH_LESSEQUAL
;
673 if(msgflg
& MSG_EXCEPT
)
674 return SEARCH_NOTEQUAL
;
678 asmlinkage
long sys_msgrcv (int msqid
, struct msgbuf __user
*msgp
, size_t msgsz
,
679 long msgtyp
, int msgflg
)
681 struct msg_queue
*msq
;
685 if (msqid
< 0 || (long) msgsz
< 0)
687 mode
= convert_mode(&msgtyp
,msgflg
);
689 msq
= msg_lock(msqid
);
693 msg
= ERR_PTR(-EIDRM
);
694 if (msg_checkid(msq
,msqid
))
698 struct msg_receiver msr_d
;
699 struct list_head
* tmp
;
701 msg
= ERR_PTR(-EACCES
);
702 if (ipcperms (&msq
->q_perm
, S_IRUGO
))
705 msg
= ERR_PTR(-EAGAIN
);
706 tmp
= msq
->q_messages
.next
;
707 while (tmp
!= &msq
->q_messages
) {
708 struct msg_msg
*walk_msg
;
709 walk_msg
= list_entry(tmp
,struct msg_msg
,m_list
);
710 if(testmsg(walk_msg
,msgtyp
,mode
) &&
711 !security_msg_queue_msgrcv(msq
, walk_msg
, current
, msgtyp
, mode
)) {
713 if(mode
== SEARCH_LESSEQUAL
&& walk_msg
->m_type
!= 1) {
715 msgtyp
=walk_msg
->m_type
-1;
724 /* Found a suitable message. Unlink it from the queue. */
725 if ((msgsz
< msg
->m_ts
) && !(msgflg
& MSG_NOERROR
)) {
726 msg
= ERR_PTR(-E2BIG
);
729 list_del(&msg
->m_list
);
731 msq
->q_rtime
= get_seconds();
732 msq
->q_lrpid
= current
->tgid
;
733 msq
->q_cbytes
-= msg
->m_ts
;
734 atomic_sub(msg
->m_ts
,&msg_bytes
);
735 atomic_dec(&msg_hdrs
);
736 ss_wakeup(&msq
->q_senders
,0);
740 /* No message waiting. Wait for a message */
741 if (msgflg
& IPC_NOWAIT
) {
742 msg
= ERR_PTR(-ENOMSG
);
745 list_add_tail(&msr_d
.r_list
,&msq
->q_receivers
);
746 msr_d
.r_tsk
= current
;
747 msr_d
.r_msgtype
= msgtyp
;
749 if(msgflg
& MSG_NOERROR
)
750 msr_d
.r_maxsize
= INT_MAX
;
752 msr_d
.r_maxsize
= msgsz
;
753 msr_d
.r_msg
= ERR_PTR(-EAGAIN
);
754 current
->state
= TASK_INTERRUPTIBLE
;
759 /* Lockless receive, part 1:
760 * Disable preemption. We don't hold a reference to the queue
761 * and getting a reference would defeat the idea of a lockless
762 * operation, thus the code relies on rcu to guarantee the
764 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
765 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
766 * rcu_read_lock() prevents preemption between reading r_msg
767 * and the spin_lock() inside ipc_lock_by_ptr().
771 /* Lockless receive, part 2:
772 * Wait until pipelined_send or expunge_all are outside of
773 * wake_up_process(). There is a race with exit(), see
774 * ipc/mqueue.c for the details.
776 msg
= (struct msg_msg
*) msr_d
.r_msg
;
777 while (msg
== NULL
) {
779 msg
= (struct msg_msg
*) msr_d
.r_msg
;
782 /* Lockless receive, part 3:
783 * If there is a message or an error then accept it without
786 if(msg
!= ERR_PTR(-EAGAIN
)) {
791 /* Lockless receive, part 3:
792 * Acquire the queue spinlock.
794 ipc_lock_by_ptr(&msq
->q_perm
);
797 /* Lockless receive, part 4:
798 * Repeat test after acquiring the spinlock.
800 msg
= (struct msg_msg
*)msr_d
.r_msg
;
801 if(msg
!= ERR_PTR(-EAGAIN
))
804 list_del(&msr_d
.r_list
);
805 if (signal_pending(current
)) {
806 msg
= ERR_PTR(-ERESTARTNOHAND
);
815 msgsz
= (msgsz
> msg
->m_ts
) ? msg
->m_ts
: msgsz
;
816 if (put_user (msg
->m_type
, &msgp
->mtype
) ||
817 store_msg(msgp
->mtext
, msg
, msgsz
)) {
824 #ifdef CONFIG_PROC_FS
825 static int sysvipc_msg_proc_show(struct seq_file
*s
, void *it
)
827 struct msg_queue
*msq
= it
;
830 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",