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>
18 #include <linux/capability.h>
19 #include <linux/config.h>
20 #include <linux/slab.h>
21 #include <linux/msg.h>
22 #include <linux/spinlock.h>
23 #include <linux/init.h>
24 #include <linux/proc_fs.h>
25 #include <linux/list.h>
26 #include <linux/security.h>
27 #include <linux/sched.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/seq_file.h>
31 #include <asm/current.h>
32 #include <asm/uaccess.h>
36 int msg_ctlmax = MSGMAX;
37 int msg_ctlmnb = MSGMNB;
38 int msg_ctlmni = MSGMNI;
40 /* one msg_receiver structure for each sleeping receiver */
42 struct list_head r_list;
43 struct task_struct* r_tsk;
49 struct msg_msg* volatile r_msg;
52 /* one msg_sender for each sleeping sender */
54 struct list_head list;
55 struct task_struct* tsk;
59 #define SEARCH_EQUAL 2
60 #define SEARCH_NOTEQUAL 3
61 #define SEARCH_LESSEQUAL 4
63 static atomic_t msg_bytes = ATOMIC_INIT(0);
64 static atomic_t msg_hdrs = ATOMIC_INIT(0);
66 static struct ipc_ids msg_ids;
68 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
69 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
70 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
71 #define msg_checkid(msq, msgid) \
72 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
73 #define msg_buildid(id, seq) \
74 ipc_buildid(&msg_ids, id, seq)
76 static void freeque (struct msg_queue *msq, int id);
77 static int newque (key_t key, int msgflg);
79 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
82 void __init msg_init (void)
84 ipc_init_ids(&msg_ids,msg_ctlmni);
85 ipc_init_proc_interface("sysvipc/msg",
86 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
88 sysvipc_msg_proc_show);
91 static int newque (key_t key, int msgflg)
95 struct msg_queue *msq;
97 msq = ipc_rcu_alloc(sizeof(*msq));
101 msq->q_perm.mode = (msgflg & S_IRWXUGO);
102 msq->q_perm.key = key;
103 msq->q_perm.xid = vx_current_xid();
105 msq->q_perm.security = NULL;
106 retval = security_msg_queue_alloc(msq);
112 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
114 security_msg_queue_free(msq);
119 msq->q_id = msg_buildid(id,msq->q_perm.seq);
120 msq->q_stime = msq->q_rtime = 0;
121 msq->q_ctime = get_seconds();
122 msq->q_cbytes = msq->q_qnum = 0;
123 msq->q_qbytes = msg_ctlmnb;
124 msq->q_lspid = msq->q_lrpid = 0;
125 INIT_LIST_HEAD(&msq->q_messages);
126 INIT_LIST_HEAD(&msq->q_receivers);
127 INIT_LIST_HEAD(&msq->q_senders);
133 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
136 current->state=TASK_INTERRUPTIBLE;
137 list_add_tail(&mss->list,&msq->q_senders);
140 static inline void ss_del(struct msg_sender* mss)
142 if(mss->list.next != NULL)
143 list_del(&mss->list);
146 static void ss_wakeup(struct list_head* h, int kill)
148 struct list_head *tmp;
152 struct msg_sender* mss;
154 mss = list_entry(tmp,struct msg_sender,list);
158 wake_up_process(mss->tsk);
162 static void expunge_all(struct msg_queue* msq, int res)
164 struct list_head *tmp;
166 tmp = msq->q_receivers.next;
167 while (tmp != &msq->q_receivers) {
168 struct msg_receiver* msr;
170 msr = list_entry(tmp,struct msg_receiver,r_list);
173 wake_up_process(msr->r_tsk);
175 msr->r_msg = ERR_PTR(res);
179 * freeque() wakes up waiters on the sender and receiver waiting queue,
180 * removes the message queue from message queue ID
181 * array, and cleans up all the messages associated with this queue.
183 * msg_ids.sem and the spinlock for this message queue is hold
184 * before freeque() is called. msg_ids.sem remains locked on exit.
186 static void freeque (struct msg_queue *msq, int id)
188 struct list_head *tmp;
190 expunge_all(msq,-EIDRM);
191 ss_wakeup(&msq->q_senders,1);
195 tmp = msq->q_messages.next;
196 while(tmp != &msq->q_messages) {
197 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
199 atomic_dec(&msg_hdrs);
202 atomic_sub(msq->q_cbytes, &msg_bytes);
203 security_msg_queue_free(msq);
207 asmlinkage long sys_msgget (key_t key, int msgflg)
209 int id, ret = -EPERM;
210 struct msg_queue *msq;
213 if (key == IPC_PRIVATE)
214 ret = newque(key, msgflg);
215 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
216 if (!(msgflg & IPC_CREAT))
219 ret = newque(key, msgflg);
220 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
226 if (ipcperms(&msq->q_perm, msgflg))
229 int qid = msg_buildid(id, msq->q_perm.seq);
230 ret = security_msg_queue_associate(msq, msgflg);
240 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
244 return copy_to_user (buf, in, sizeof(*in));
249 memset(&out,0,sizeof(out));
251 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
253 out.msg_stime = in->msg_stime;
254 out.msg_rtime = in->msg_rtime;
255 out.msg_ctime = in->msg_ctime;
257 if(in->msg_cbytes > USHRT_MAX)
258 out.msg_cbytes = USHRT_MAX;
260 out.msg_cbytes = in->msg_cbytes;
261 out.msg_lcbytes = in->msg_cbytes;
263 if(in->msg_qnum > USHRT_MAX)
264 out.msg_qnum = USHRT_MAX;
266 out.msg_qnum = in->msg_qnum;
268 if(in->msg_qbytes > USHRT_MAX)
269 out.msg_qbytes = USHRT_MAX;
271 out.msg_qbytes = in->msg_qbytes;
272 out.msg_lqbytes = in->msg_qbytes;
274 out.msg_lspid = in->msg_lspid;
275 out.msg_lrpid = in->msg_lrpid;
277 return copy_to_user (buf, &out, sizeof(out));
285 unsigned long qbytes;
291 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
296 struct msqid64_ds tbuf;
298 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
301 out->qbytes = tbuf.msg_qbytes;
302 out->uid = tbuf.msg_perm.uid;
303 out->gid = tbuf.msg_perm.gid;
304 out->mode = tbuf.msg_perm.mode;
310 struct msqid_ds tbuf_old;
312 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
315 out->uid = tbuf_old.msg_perm.uid;
316 out->gid = tbuf_old.msg_perm.gid;
317 out->mode = tbuf_old.msg_perm.mode;
319 if(tbuf_old.msg_qbytes == 0)
320 out->qbytes = tbuf_old.msg_lqbytes;
322 out->qbytes = tbuf_old.msg_qbytes;
331 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
334 struct msg_queue *msq;
335 struct msq_setbuf setbuf;
336 struct kern_ipc_perm *ipcp;
338 if (msqid < 0 || cmd < 0)
341 version = ipc_parse_version(&cmd);
347 struct msginfo msginfo;
351 /* We must not return kernel stack data.
352 * due to padding, it's not enough
353 * to set all member fields.
356 err = security_msg_queue_msgctl(NULL, cmd);
360 memset(&msginfo,0,sizeof(msginfo));
361 msginfo.msgmni = msg_ctlmni;
362 msginfo.msgmax = msg_ctlmax;
363 msginfo.msgmnb = msg_ctlmnb;
364 msginfo.msgssz = MSGSSZ;
365 msginfo.msgseg = MSGSEG;
367 if (cmd == MSG_INFO) {
368 msginfo.msgpool = msg_ids.in_use;
369 msginfo.msgmap = atomic_read(&msg_hdrs);
370 msginfo.msgtql = atomic_read(&msg_bytes);
372 msginfo.msgmap = MSGMAP;
373 msginfo.msgpool = MSGPOOL;
374 msginfo.msgtql = MSGTQL;
376 max_id = msg_ids.max_id;
378 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
380 return (max_id < 0) ? 0: max_id;
385 struct msqid64_ds tbuf;
389 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size)
392 memset(&tbuf,0,sizeof(tbuf));
394 msq = msg_lock(msqid);
398 if(cmd == MSG_STAT) {
399 success_return = msg_buildid(msqid, msq->q_perm.seq);
402 if (msg_checkid(msq,msqid))
407 if (ipcperms (&msq->q_perm, S_IRUGO))
410 err = security_msg_queue_msgctl(msq, cmd);
414 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
415 tbuf.msg_stime = msq->q_stime;
416 tbuf.msg_rtime = msq->q_rtime;
417 tbuf.msg_ctime = msq->q_ctime;
418 tbuf.msg_cbytes = msq->q_cbytes;
419 tbuf.msg_qnum = msq->q_qnum;
420 tbuf.msg_qbytes = msq->q_qbytes;
421 tbuf.msg_lspid = msq->q_lspid;
422 tbuf.msg_lrpid = msq->q_lrpid;
424 if (copy_msqid_to_user(buf, &tbuf, version))
426 return success_return;
431 if (copy_msqid_from_user (&setbuf, buf, version))
433 if ((err = audit_ipc_perms(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode)))
443 msq = msg_lock(msqid);
449 if (msg_checkid(msq,msqid))
453 if (current->euid != ipcp->cuid &&
454 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
455 /* We _could_ check for CAP_CHOWN above, but we don't */
458 err = security_msg_queue_msgctl(msq, cmd);
466 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
469 msq->q_qbytes = setbuf.qbytes;
471 ipcp->uid = setbuf.uid;
472 ipcp->gid = setbuf.gid;
473 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
474 (S_IRWXUGO & setbuf.mode);
475 msq->q_ctime = get_seconds();
476 /* sleeping receivers might be excluded by
477 * stricter permissions.
479 expunge_all(msq,-EAGAIN);
480 /* sleeping senders might be able to send
481 * due to a larger queue size.
483 ss_wakeup(&msq->q_senders,0);
488 freeque (msq, msqid);
503 static int testmsg(struct msg_msg* msg,long type,int mode)
509 case SEARCH_LESSEQUAL:
510 if(msg->m_type <=type)
514 if(msg->m_type == type)
517 case SEARCH_NOTEQUAL:
518 if(msg->m_type != type)
525 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
527 struct list_head* tmp;
529 tmp = msq->q_receivers.next;
530 while (tmp != &msq->q_receivers) {
531 struct msg_receiver* msr;
532 msr = list_entry(tmp,struct msg_receiver,r_list);
534 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
535 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
536 list_del(&msr->r_list);
537 if(msr->r_maxsize < msg->m_ts) {
539 wake_up_process(msr->r_tsk);
541 msr->r_msg = ERR_PTR(-E2BIG);
544 msq->q_lrpid = msr->r_tsk->pid;
545 msq->q_rtime = get_seconds();
546 wake_up_process(msr->r_tsk);
556 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
558 struct msg_queue *msq;
563 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
565 if (get_user(mtype, &msgp->mtype))
570 msg = load_msg(msgp->mtext, msgsz);
577 msq = msg_lock(msqid);
583 if (msg_checkid(msq,msqid))
584 goto out_unlock_free;
590 if (ipcperms(&msq->q_perm, S_IWUGO))
591 goto out_unlock_free;
593 err = security_msg_queue_msgsnd(msq, msg, msgflg);
595 goto out_unlock_free;
597 if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
598 1 + msq->q_qnum <= msq->q_qbytes) {
602 /* queue full, wait: */
603 if(msgflg&IPC_NOWAIT) {
605 goto out_unlock_free;
612 ipc_lock_by_ptr(&msq->q_perm);
614 if (msq->q_perm.deleted) {
616 goto out_unlock_free;
620 if (signal_pending(current)) {
622 goto out_unlock_free;
626 msq->q_lspid = current->tgid;
627 msq->q_stime = get_seconds();
629 if(!pipelined_send(msq,msg)) {
630 /* noone is waiting for this message, enqueue it */
631 list_add_tail(&msg->m_list,&msq->q_messages);
632 msq->q_cbytes += msgsz;
634 atomic_add(msgsz,&msg_bytes);
635 atomic_inc(&msg_hdrs);
649 static inline int convert_mode(long* msgtyp, int msgflg)
652 * find message of correct type.
653 * msgtyp = 0 => get first.
654 * msgtyp > 0 => get first message of matching type.
655 * msgtyp < 0 => get message with least type must be < abs(msgtype).
661 return SEARCH_LESSEQUAL;
663 if(msgflg & MSG_EXCEPT)
664 return SEARCH_NOTEQUAL;
668 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
669 long msgtyp, int msgflg)
671 struct msg_queue *msq;
675 if (msqid < 0 || (long) msgsz < 0)
677 mode = convert_mode(&msgtyp,msgflg);
679 msq = msg_lock(msqid);
683 msg = ERR_PTR(-EIDRM);
684 if (msg_checkid(msq,msqid))
688 struct msg_receiver msr_d;
689 struct list_head* tmp;
691 msg = ERR_PTR(-EACCES);
692 if (ipcperms (&msq->q_perm, S_IRUGO))
695 msg = ERR_PTR(-EAGAIN);
696 tmp = msq->q_messages.next;
697 while (tmp != &msq->q_messages) {
698 struct msg_msg *walk_msg;
699 walk_msg = list_entry(tmp,struct msg_msg,m_list);
700 if(testmsg(walk_msg,msgtyp,mode) &&
701 !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
703 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
705 msgtyp=walk_msg->m_type-1;
714 /* Found a suitable message. Unlink it from the queue. */
715 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
716 msg = ERR_PTR(-E2BIG);
719 list_del(&msg->m_list);
721 msq->q_rtime = get_seconds();
722 msq->q_lrpid = current->tgid;
723 msq->q_cbytes -= msg->m_ts;
724 atomic_sub(msg->m_ts,&msg_bytes);
725 atomic_dec(&msg_hdrs);
726 ss_wakeup(&msq->q_senders,0);
730 /* No message waiting. Wait for a message */
731 if (msgflg & IPC_NOWAIT) {
732 msg = ERR_PTR(-ENOMSG);
735 list_add_tail(&msr_d.r_list,&msq->q_receivers);
736 msr_d.r_tsk = current;
737 msr_d.r_msgtype = msgtyp;
739 if(msgflg & MSG_NOERROR)
740 msr_d.r_maxsize = INT_MAX;
742 msr_d.r_maxsize = msgsz;
743 msr_d.r_msg = ERR_PTR(-EAGAIN);
744 current->state = TASK_INTERRUPTIBLE;
749 /* Lockless receive, part 1:
750 * Disable preemption. We don't hold a reference to the queue
751 * and getting a reference would defeat the idea of a lockless
752 * operation, thus the code relies on rcu to guarantee the
754 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
755 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
756 * rcu_read_lock() prevents preemption between reading r_msg
757 * and the spin_lock() inside ipc_lock_by_ptr().
761 /* Lockless receive, part 2:
762 * Wait until pipelined_send or expunge_all are outside of
763 * wake_up_process(). There is a race with exit(), see
764 * ipc/mqueue.c for the details.
766 msg = (struct msg_msg*) msr_d.r_msg;
767 while (msg == NULL) {
769 msg = (struct msg_msg*) msr_d.r_msg;
772 /* Lockless receive, part 3:
773 * If there is a message or an error then accept it without
776 if(msg != ERR_PTR(-EAGAIN)) {
781 /* Lockless receive, part 3:
782 * Acquire the queue spinlock.
784 ipc_lock_by_ptr(&msq->q_perm);
787 /* Lockless receive, part 4:
788 * Repeat test after acquiring the spinlock.
790 msg = (struct msg_msg*)msr_d.r_msg;
791 if(msg != ERR_PTR(-EAGAIN))
794 list_del(&msr_d.r_list);
795 if (signal_pending(current)) {
796 msg = ERR_PTR(-ERESTARTNOHAND);
805 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
806 if (put_user (msg->m_type, &msgp->mtype) ||
807 store_msg(msgp->mtext, msg, msgsz)) {
814 #ifdef CONFIG_PROC_FS
815 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
817 struct msg_queue *msq = it;
819 if (!vx_check(msq->q_perm.xid, VX_IDENT))
823 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",