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>
35 #include <linux/vs_base.h>
37 #include <asm/current.h>
38 #include <asm/uaccess.h>
42 int msg_ctlmax = MSGMAX;
43 int msg_ctlmnb = MSGMNB;
44 int msg_ctlmni = MSGMNI;
46 /* one msg_receiver structure for each sleeping receiver */
48 struct list_head r_list;
49 struct task_struct* r_tsk;
55 struct msg_msg* volatile r_msg;
58 /* one msg_sender for each sleeping sender */
60 struct list_head list;
61 struct task_struct* tsk;
65 #define SEARCH_EQUAL 2
66 #define SEARCH_NOTEQUAL 3
67 #define SEARCH_LESSEQUAL 4
69 static atomic_t msg_bytes = ATOMIC_INIT(0);
70 static atomic_t msg_hdrs = ATOMIC_INIT(0);
72 static struct ipc_ids msg_ids;
74 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
75 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
76 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
77 #define msg_checkid(msq, msgid) \
78 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
79 #define msg_buildid(id, seq) \
80 ipc_buildid(&msg_ids, id, seq)
82 static void freeque (struct msg_queue *msq, int id);
83 static int newque (key_t key, int msgflg);
85 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
88 void __init msg_init (void)
90 ipc_init_ids(&msg_ids,msg_ctlmni);
91 ipc_init_proc_interface("sysvipc/msg",
92 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
94 sysvipc_msg_proc_show);
97 static int newque (key_t key, int msgflg)
101 struct msg_queue *msq;
103 msq = ipc_rcu_alloc(sizeof(*msq));
107 msq->q_perm.mode = (msgflg & S_IRWXUGO);
108 msq->q_perm.key = key;
109 msq->q_perm.xid = vx_current_xid();
111 msq->q_perm.security = NULL;
112 retval = security_msg_queue_alloc(msq);
118 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
120 security_msg_queue_free(msq);
125 msq->q_id = msg_buildid(id,msq->q_perm.seq);
126 msq->q_stime = msq->q_rtime = 0;
127 msq->q_ctime = get_seconds();
128 msq->q_cbytes = msq->q_qnum = 0;
129 msq->q_qbytes = msg_ctlmnb;
130 msq->q_lspid = msq->q_lrpid = 0;
131 INIT_LIST_HEAD(&msq->q_messages);
132 INIT_LIST_HEAD(&msq->q_receivers);
133 INIT_LIST_HEAD(&msq->q_senders);
139 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
142 current->state=TASK_INTERRUPTIBLE;
143 list_add_tail(&mss->list,&msq->q_senders);
146 static inline void ss_del(struct msg_sender* mss)
148 if(mss->list.next != NULL)
149 list_del(&mss->list);
152 static void ss_wakeup(struct list_head* h, int kill)
154 struct list_head *tmp;
158 struct msg_sender* mss;
160 mss = list_entry(tmp,struct msg_sender,list);
164 wake_up_process(mss->tsk);
168 static void expunge_all(struct msg_queue* msq, int res)
170 struct list_head *tmp;
172 tmp = msq->q_receivers.next;
173 while (tmp != &msq->q_receivers) {
174 struct msg_receiver* msr;
176 msr = list_entry(tmp,struct msg_receiver,r_list);
179 wake_up_process(msr->r_tsk);
181 msr->r_msg = ERR_PTR(res);
185 * freeque() wakes up waiters on the sender and receiver waiting queue,
186 * removes the message queue from message queue ID
187 * array, and cleans up all the messages associated with this queue.
189 * msg_ids.mutex and the spinlock for this message queue is hold
190 * before freeque() is called. msg_ids.mutex remains locked on exit.
192 static void freeque (struct msg_queue *msq, int id)
194 struct list_head *tmp;
196 expunge_all(msq,-EIDRM);
197 ss_wakeup(&msq->q_senders,1);
201 tmp = msq->q_messages.next;
202 while(tmp != &msq->q_messages) {
203 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
205 atomic_dec(&msg_hdrs);
208 atomic_sub(msq->q_cbytes, &msg_bytes);
209 security_msg_queue_free(msq);
213 asmlinkage long sys_msgget (key_t key, int msgflg)
215 int id, ret = -EPERM;
216 struct msg_queue *msq;
218 mutex_lock(&msg_ids.mutex);
219 if (key == IPC_PRIVATE)
220 ret = newque(key, msgflg);
221 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
222 if (!(msgflg & IPC_CREAT))
225 ret = newque(key, msgflg);
226 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
231 if (ipcperms(&msq->q_perm, msgflg))
234 int qid = msg_buildid(id, msq->q_perm.seq);
235 ret = security_msg_queue_associate(msq, msgflg);
241 mutex_unlock(&msg_ids.mutex);
245 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
249 return copy_to_user (buf, in, sizeof(*in));
254 memset(&out,0,sizeof(out));
256 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
258 out.msg_stime = in->msg_stime;
259 out.msg_rtime = in->msg_rtime;
260 out.msg_ctime = in->msg_ctime;
262 if(in->msg_cbytes > USHRT_MAX)
263 out.msg_cbytes = USHRT_MAX;
265 out.msg_cbytes = in->msg_cbytes;
266 out.msg_lcbytes = in->msg_cbytes;
268 if(in->msg_qnum > USHRT_MAX)
269 out.msg_qnum = USHRT_MAX;
271 out.msg_qnum = in->msg_qnum;
273 if(in->msg_qbytes > USHRT_MAX)
274 out.msg_qbytes = USHRT_MAX;
276 out.msg_qbytes = in->msg_qbytes;
277 out.msg_lqbytes = in->msg_qbytes;
279 out.msg_lspid = in->msg_lspid;
280 out.msg_lrpid = in->msg_lrpid;
282 return copy_to_user (buf, &out, sizeof(out));
290 unsigned long qbytes;
296 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
301 struct msqid64_ds tbuf;
303 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
306 out->qbytes = tbuf.msg_qbytes;
307 out->uid = tbuf.msg_perm.uid;
308 out->gid = tbuf.msg_perm.gid;
309 out->mode = tbuf.msg_perm.mode;
315 struct msqid_ds tbuf_old;
317 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
320 out->uid = tbuf_old.msg_perm.uid;
321 out->gid = tbuf_old.msg_perm.gid;
322 out->mode = tbuf_old.msg_perm.mode;
324 if(tbuf_old.msg_qbytes == 0)
325 out->qbytes = tbuf_old.msg_lqbytes;
327 out->qbytes = tbuf_old.msg_qbytes;
336 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
339 struct msg_queue *msq;
340 struct msq_setbuf setbuf;
341 struct kern_ipc_perm *ipcp;
343 if (msqid < 0 || cmd < 0)
346 version = ipc_parse_version(&cmd);
352 struct msginfo msginfo;
356 /* We must not return kernel stack data.
357 * due to padding, it's not enough
358 * to set all member fields.
361 err = security_msg_queue_msgctl(NULL, cmd);
365 memset(&msginfo,0,sizeof(msginfo));
366 msginfo.msgmni = msg_ctlmni;
367 msginfo.msgmax = msg_ctlmax;
368 msginfo.msgmnb = msg_ctlmnb;
369 msginfo.msgssz = MSGSSZ;
370 msginfo.msgseg = MSGSEG;
371 mutex_lock(&msg_ids.mutex);
372 if (cmd == MSG_INFO) {
373 msginfo.msgpool = msg_ids.in_use;
374 msginfo.msgmap = atomic_read(&msg_hdrs);
375 msginfo.msgtql = atomic_read(&msg_bytes);
377 msginfo.msgmap = MSGMAP;
378 msginfo.msgpool = MSGPOOL;
379 msginfo.msgtql = MSGTQL;
381 max_id = msg_ids.max_id;
382 mutex_unlock(&msg_ids.mutex);
383 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
385 return (max_id < 0) ? 0: max_id;
390 struct msqid64_ds tbuf;
394 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size)
397 memset(&tbuf,0,sizeof(tbuf));
399 msq = msg_lock(msqid);
403 if(cmd == MSG_STAT) {
404 success_return = msg_buildid(msqid, msq->q_perm.seq);
407 if (msg_checkid(msq,msqid))
412 if (ipcperms (&msq->q_perm, S_IRUGO))
415 err = security_msg_queue_msgctl(msq, cmd);
419 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
420 tbuf.msg_stime = msq->q_stime;
421 tbuf.msg_rtime = msq->q_rtime;
422 tbuf.msg_ctime = msq->q_ctime;
423 tbuf.msg_cbytes = msq->q_cbytes;
424 tbuf.msg_qnum = msq->q_qnum;
425 tbuf.msg_qbytes = msq->q_qbytes;
426 tbuf.msg_lspid = msq->q_lspid;
427 tbuf.msg_lrpid = msq->q_lrpid;
429 if (copy_msqid_to_user(buf, &tbuf, version))
431 return success_return;
436 if (copy_msqid_from_user (&setbuf, buf, version))
445 mutex_lock(&msg_ids.mutex);
446 msq = msg_lock(msqid);
452 if (msg_checkid(msq,msqid))
456 err = audit_ipc_obj(ipcp);
461 if (current->euid != ipcp->cuid &&
462 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
463 /* We _could_ check for CAP_CHOWN above, but we don't */
466 err = security_msg_queue_msgctl(msq, cmd);
473 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode, ipcp);
478 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
481 msq->q_qbytes = setbuf.qbytes;
483 ipcp->uid = setbuf.uid;
484 ipcp->gid = setbuf.gid;
485 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
486 (S_IRWXUGO & setbuf.mode);
487 msq->q_ctime = get_seconds();
488 /* sleeping receivers might be excluded by
489 * stricter permissions.
491 expunge_all(msq,-EAGAIN);
492 /* sleeping senders might be able to send
493 * due to a larger queue size.
495 ss_wakeup(&msq->q_senders,0);
500 freeque (msq, msqid);
505 mutex_unlock(&msg_ids.mutex);
515 static int testmsg(struct msg_msg* msg,long type,int mode)
521 case SEARCH_LESSEQUAL:
522 if(msg->m_type <=type)
526 if(msg->m_type == type)
529 case SEARCH_NOTEQUAL:
530 if(msg->m_type != type)
537 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
539 struct list_head* tmp;
541 tmp = msq->q_receivers.next;
542 while (tmp != &msq->q_receivers) {
543 struct msg_receiver* msr;
544 msr = list_entry(tmp,struct msg_receiver,r_list);
546 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
547 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
548 list_del(&msr->r_list);
549 if(msr->r_maxsize < msg->m_ts) {
551 wake_up_process(msr->r_tsk);
553 msr->r_msg = ERR_PTR(-E2BIG);
556 msq->q_lrpid = msr->r_tsk->pid;
557 msq->q_rtime = get_seconds();
558 wake_up_process(msr->r_tsk);
568 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
570 struct msg_queue *msq;
575 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
577 if (get_user(mtype, &msgp->mtype))
582 msg = load_msg(msgp->mtext, msgsz);
589 msq = msg_lock(msqid);
595 if (msg_checkid(msq,msqid))
596 goto out_unlock_free;
602 if (ipcperms(&msq->q_perm, S_IWUGO))
603 goto out_unlock_free;
605 err = security_msg_queue_msgsnd(msq, msg, msgflg);
607 goto out_unlock_free;
609 if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
610 1 + msq->q_qnum <= msq->q_qbytes) {
614 /* queue full, wait: */
615 if(msgflg&IPC_NOWAIT) {
617 goto out_unlock_free;
624 ipc_lock_by_ptr(&msq->q_perm);
626 if (msq->q_perm.deleted) {
628 goto out_unlock_free;
632 if (signal_pending(current)) {
634 goto out_unlock_free;
638 msq->q_lspid = current->tgid;
639 msq->q_stime = get_seconds();
641 if(!pipelined_send(msq,msg)) {
642 /* noone is waiting for this message, enqueue it */
643 list_add_tail(&msg->m_list,&msq->q_messages);
644 msq->q_cbytes += msgsz;
646 atomic_add(msgsz,&msg_bytes);
647 atomic_inc(&msg_hdrs);
661 static inline int convert_mode(long* msgtyp, int msgflg)
664 * find message of correct type.
665 * msgtyp = 0 => get first.
666 * msgtyp > 0 => get first message of matching type.
667 * msgtyp < 0 => get message with least type must be < abs(msgtype).
673 return SEARCH_LESSEQUAL;
675 if(msgflg & MSG_EXCEPT)
676 return SEARCH_NOTEQUAL;
680 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
681 long msgtyp, int msgflg)
683 struct msg_queue *msq;
687 if (msqid < 0 || (long) msgsz < 0)
689 mode = convert_mode(&msgtyp,msgflg);
691 msq = msg_lock(msqid);
695 msg = ERR_PTR(-EIDRM);
696 if (msg_checkid(msq,msqid))
700 struct msg_receiver msr_d;
701 struct list_head* tmp;
703 msg = ERR_PTR(-EACCES);
704 if (ipcperms (&msq->q_perm, S_IRUGO))
707 msg = ERR_PTR(-EAGAIN);
708 tmp = msq->q_messages.next;
709 while (tmp != &msq->q_messages) {
710 struct msg_msg *walk_msg;
711 walk_msg = list_entry(tmp,struct msg_msg,m_list);
712 if(testmsg(walk_msg,msgtyp,mode) &&
713 !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
715 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
717 msgtyp=walk_msg->m_type-1;
726 /* Found a suitable message. Unlink it from the queue. */
727 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
728 msg = ERR_PTR(-E2BIG);
731 list_del(&msg->m_list);
733 msq->q_rtime = get_seconds();
734 msq->q_lrpid = current->tgid;
735 msq->q_cbytes -= msg->m_ts;
736 atomic_sub(msg->m_ts,&msg_bytes);
737 atomic_dec(&msg_hdrs);
738 ss_wakeup(&msq->q_senders,0);
742 /* No message waiting. Wait for a message */
743 if (msgflg & IPC_NOWAIT) {
744 msg = ERR_PTR(-ENOMSG);
747 list_add_tail(&msr_d.r_list,&msq->q_receivers);
748 msr_d.r_tsk = current;
749 msr_d.r_msgtype = msgtyp;
751 if(msgflg & MSG_NOERROR)
752 msr_d.r_maxsize = INT_MAX;
754 msr_d.r_maxsize = msgsz;
755 msr_d.r_msg = ERR_PTR(-EAGAIN);
756 current->state = TASK_INTERRUPTIBLE;
761 /* Lockless receive, part 1:
762 * Disable preemption. We don't hold a reference to the queue
763 * and getting a reference would defeat the idea of a lockless
764 * operation, thus the code relies on rcu to guarantee the
766 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
767 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
768 * rcu_read_lock() prevents preemption between reading r_msg
769 * and the spin_lock() inside ipc_lock_by_ptr().
773 /* Lockless receive, part 2:
774 * Wait until pipelined_send or expunge_all are outside of
775 * wake_up_process(). There is a race with exit(), see
776 * ipc/mqueue.c for the details.
778 msg = (struct msg_msg*) msr_d.r_msg;
779 while (msg == NULL) {
781 msg = (struct msg_msg*) msr_d.r_msg;
784 /* Lockless receive, part 3:
785 * If there is a message or an error then accept it without
788 if(msg != ERR_PTR(-EAGAIN)) {
793 /* Lockless receive, part 3:
794 * Acquire the queue spinlock.
796 ipc_lock_by_ptr(&msq->q_perm);
799 /* Lockless receive, part 4:
800 * Repeat test after acquiring the spinlock.
802 msg = (struct msg_msg*)msr_d.r_msg;
803 if(msg != ERR_PTR(-EAGAIN))
806 list_del(&msr_d.r_list);
807 if (signal_pending(current)) {
808 msg = ERR_PTR(-ERESTARTNOHAND);
817 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
818 if (put_user (msg->m_type, &msgp->mtype) ||
819 store_msg(msgp->mtext, msg, msgsz)) {
826 #ifdef CONFIG_PROC_FS
827 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
829 struct msg_queue *msq = it;
831 if (!vx_check(msq->q_perm.xid, VX_IDENT))
835 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",