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 <asm/current.h>
28 #include <asm/uaccess.h>
32 int msg_ctlmax = MSGMAX;
33 int msg_ctlmnb = MSGMNB;
34 int msg_ctlmni = MSGMNI;
36 /* one msg_receiver structure for each sleeping receiver */
38 struct list_head r_list;
39 struct task_struct* r_tsk;
45 struct msg_msg* volatile r_msg;
48 /* one msg_sender for each sleeping sender */
50 struct list_head list;
51 struct task_struct* tsk;
55 #define SEARCH_EQUAL 2
56 #define SEARCH_NOTEQUAL 3
57 #define SEARCH_LESSEQUAL 4
59 static atomic_t msg_bytes = ATOMIC_INIT(0);
60 static atomic_t msg_hdrs = ATOMIC_INIT(0);
62 static struct ipc_ids msg_ids;
64 #define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
65 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
66 #define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
67 #define msg_checkid(msq, msgid) \
68 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
69 #define msg_buildid(id, seq) \
70 ipc_buildid(&msg_ids, id, seq)
72 static void freeque (struct msg_queue *msq, int id);
73 static int newque (key_t key, int msgflg);
75 static int sysvipc_msg_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
78 void __init msg_init (void)
80 ipc_init_ids(&msg_ids,msg_ctlmni);
83 create_proc_read_entry("sysvipc/msg", 0, 0, sysvipc_msg_read_proc, NULL);
87 static int newque (key_t key, int msgflg)
91 struct msg_queue *msq;
93 msq = ipc_rcu_alloc(sizeof(*msq));
97 msq->q_perm.mode = (msgflg & S_IRWXUGO);
98 msq->q_perm.key = key;
100 msq->q_perm.security = NULL;
101 retval = security_msg_queue_alloc(msq);
103 ipc_rcu_free(msq, sizeof(*msq));
107 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
109 security_msg_queue_free(msq);
110 ipc_rcu_free(msq, sizeof(*msq));
114 msq->q_stime = msq->q_rtime = 0;
115 msq->q_ctime = get_seconds();
116 msq->q_cbytes = msq->q_qnum = 0;
117 msq->q_qbytes = msg_ctlmnb;
118 msq->q_lspid = msq->q_lrpid = 0;
119 INIT_LIST_HEAD(&msq->q_messages);
120 INIT_LIST_HEAD(&msq->q_receivers);
121 INIT_LIST_HEAD(&msq->q_senders);
124 return msg_buildid(id,msq->q_perm.seq);
127 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
130 current->state=TASK_INTERRUPTIBLE;
131 list_add_tail(&mss->list,&msq->q_senders);
134 static inline void ss_del(struct msg_sender* mss)
136 if(mss->list.next != NULL)
137 list_del(&mss->list);
140 static void ss_wakeup(struct list_head* h, int kill)
142 struct list_head *tmp;
146 struct msg_sender* mss;
148 mss = list_entry(tmp,struct msg_sender,list);
152 wake_up_process(mss->tsk);
156 static void expunge_all(struct msg_queue* msq, int res)
158 struct list_head *tmp;
160 tmp = msq->q_receivers.next;
161 while (tmp != &msq->q_receivers) {
162 struct msg_receiver* msr;
164 msr = list_entry(tmp,struct msg_receiver,r_list);
166 msr->r_msg = ERR_PTR(res);
167 wake_up_process(msr->r_tsk);
171 * freeque() wakes up waiters on the sender and receiver waiting queue,
172 * removes the message queue from message queue ID
173 * array, and cleans up all the messages associated with this queue.
175 * msg_ids.sem and the spinlock for this message queue is hold
176 * before freeque() is called. msg_ids.sem remains locked on exit.
178 static void freeque (struct msg_queue *msq, int id)
180 struct list_head *tmp;
182 expunge_all(msq,-EIDRM);
183 ss_wakeup(&msq->q_senders,1);
187 tmp = msq->q_messages.next;
188 while(tmp != &msq->q_messages) {
189 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
191 atomic_dec(&msg_hdrs);
194 atomic_sub(msq->q_cbytes, &msg_bytes);
195 security_msg_queue_free(msq);
196 ipc_rcu_free(msq, sizeof(struct msg_queue));
199 asmlinkage long sys_msgget (key_t key, int msgflg)
201 int id, ret = -EPERM;
202 struct msg_queue *msq;
205 if (key == IPC_PRIVATE)
206 ret = newque(key, msgflg);
207 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
208 if (!(msgflg & IPC_CREAT))
211 ret = newque(key, msgflg);
212 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
218 if (ipcperms(&msq->q_perm, msgflg))
221 int qid = msg_buildid(id, msq->q_perm.seq);
222 ret = security_msg_queue_associate(msq, msgflg);
232 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
236 return copy_to_user (buf, in, sizeof(*in));
241 memset(&out,0,sizeof(out));
243 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
245 out.msg_stime = in->msg_stime;
246 out.msg_rtime = in->msg_rtime;
247 out.msg_ctime = in->msg_ctime;
249 if(in->msg_cbytes > USHRT_MAX)
250 out.msg_cbytes = USHRT_MAX;
252 out.msg_cbytes = in->msg_cbytes;
253 out.msg_lcbytes = in->msg_cbytes;
255 if(in->msg_qnum > USHRT_MAX)
256 out.msg_qnum = USHRT_MAX;
258 out.msg_qnum = in->msg_qnum;
260 if(in->msg_qbytes > USHRT_MAX)
261 out.msg_qbytes = USHRT_MAX;
263 out.msg_qbytes = in->msg_qbytes;
264 out.msg_lqbytes = in->msg_qbytes;
266 out.msg_lspid = in->msg_lspid;
267 out.msg_lrpid = in->msg_lrpid;
269 return copy_to_user (buf, &out, sizeof(out));
277 unsigned long qbytes;
283 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
288 struct msqid64_ds tbuf;
290 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
293 out->qbytes = tbuf.msg_qbytes;
294 out->uid = tbuf.msg_perm.uid;
295 out->gid = tbuf.msg_perm.gid;
296 out->mode = tbuf.msg_perm.mode;
302 struct msqid_ds tbuf_old;
304 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
307 out->uid = tbuf_old.msg_perm.uid;
308 out->gid = tbuf_old.msg_perm.gid;
309 out->mode = tbuf_old.msg_perm.mode;
311 if(tbuf_old.msg_qbytes == 0)
312 out->qbytes = tbuf_old.msg_lqbytes;
314 out->qbytes = tbuf_old.msg_qbytes;
323 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
326 struct msg_queue *msq;
327 struct msq_setbuf setbuf;
328 struct kern_ipc_perm *ipcp;
330 if (msqid < 0 || cmd < 0)
333 version = ipc_parse_version(&cmd);
339 struct msginfo msginfo;
343 /* We must not return kernel stack data.
344 * due to padding, it's not enough
345 * to set all member fields.
348 err = security_msg_queue_msgctl(NULL, cmd);
352 memset(&msginfo,0,sizeof(msginfo));
353 msginfo.msgmni = msg_ctlmni;
354 msginfo.msgmax = msg_ctlmax;
355 msginfo.msgmnb = msg_ctlmnb;
356 msginfo.msgssz = MSGSSZ;
357 msginfo.msgseg = MSGSEG;
359 if (cmd == MSG_INFO) {
360 msginfo.msgpool = msg_ids.in_use;
361 msginfo.msgmap = atomic_read(&msg_hdrs);
362 msginfo.msgtql = atomic_read(&msg_bytes);
364 msginfo.msgmap = MSGMAP;
365 msginfo.msgpool = MSGPOOL;
366 msginfo.msgtql = MSGTQL;
368 max_id = msg_ids.max_id;
370 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
372 return (max_id < 0) ? 0: max_id;
377 struct msqid64_ds tbuf;
381 if(cmd == MSG_STAT && msqid >= msg_ids.size)
384 memset(&tbuf,0,sizeof(tbuf));
386 msq = msg_lock(msqid);
390 if(cmd == MSG_STAT) {
391 success_return = msg_buildid(msqid, msq->q_perm.seq);
394 if (msg_checkid(msq,msqid))
399 if (ipcperms (&msq->q_perm, S_IRUGO))
402 err = security_msg_queue_msgctl(msq, cmd);
406 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
407 tbuf.msg_stime = msq->q_stime;
408 tbuf.msg_rtime = msq->q_rtime;
409 tbuf.msg_ctime = msq->q_ctime;
410 tbuf.msg_cbytes = msq->q_cbytes;
411 tbuf.msg_qnum = msq->q_qnum;
412 tbuf.msg_qbytes = msq->q_qbytes;
413 tbuf.msg_lspid = msq->q_lspid;
414 tbuf.msg_lrpid = msq->q_lrpid;
416 if (copy_msqid_to_user(buf, &tbuf, version))
418 return success_return;
423 if (copy_msqid_from_user (&setbuf, buf, version))
433 msq = msg_lock(msqid);
439 if (msg_checkid(msq,msqid))
443 if (current->euid != ipcp->cuid &&
444 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
445 /* We _could_ check for CAP_CHOWN above, but we don't */
448 err = security_msg_queue_msgctl(msq, cmd);
456 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
459 msq->q_qbytes = setbuf.qbytes;
461 ipcp->uid = setbuf.uid;
462 ipcp->gid = setbuf.gid;
463 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
464 (S_IRWXUGO & setbuf.mode);
465 msq->q_ctime = get_seconds();
466 /* sleeping receivers might be excluded by
467 * stricter permissions.
469 expunge_all(msq,-EAGAIN);
470 /* sleeping senders might be able to send
471 * due to a larger queue size.
473 ss_wakeup(&msq->q_senders,0);
478 freeque (msq, msqid);
493 static int testmsg(struct msg_msg* msg,long type,int mode)
499 case SEARCH_LESSEQUAL:
500 if(msg->m_type <=type)
504 if(msg->m_type == type)
507 case SEARCH_NOTEQUAL:
508 if(msg->m_type != type)
515 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
517 struct list_head* tmp;
519 tmp = msq->q_receivers.next;
520 while (tmp != &msq->q_receivers) {
521 struct msg_receiver* msr;
522 msr = list_entry(tmp,struct msg_receiver,r_list);
524 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
525 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
526 list_del(&msr->r_list);
527 if(msr->r_maxsize < msg->m_ts) {
528 msr->r_msg = ERR_PTR(-E2BIG);
529 wake_up_process(msr->r_tsk);
532 msq->q_lrpid = msr->r_tsk->pid;
533 msq->q_rtime = get_seconds();
534 wake_up_process(msr->r_tsk);
542 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
544 struct msg_queue *msq;
549 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
551 if (get_user(mtype, &msgp->mtype))
556 msg = load_msg(msgp->mtext, msgsz);
563 msq = msg_lock(msqid);
569 if (msg_checkid(msq,msqid))
570 goto out_unlock_free;
573 if (ipcperms(&msq->q_perm, S_IWUGO))
574 goto out_unlock_free;
576 err = security_msg_queue_msgsnd(msq, msg, msgflg);
578 goto out_unlock_free;
580 if(msgsz + msq->q_cbytes > msq->q_qbytes ||
581 1 + msq->q_qnum > msq->q_qbytes) {
584 if(msgflg&IPC_NOWAIT) {
586 goto out_unlock_free;
591 current->state= TASK_RUNNING;
593 msq = msg_lock(msqid);
599 if (signal_pending(current)) {
601 goto out_unlock_free;
606 msq->q_lspid = current->tgid;
607 msq->q_stime = get_seconds();
609 if(!pipelined_send(msq,msg)) {
610 /* noone is waiting for this message, enqueue it */
611 list_add_tail(&msg->m_list,&msq->q_messages);
612 msq->q_cbytes += msgsz;
614 atomic_add(msgsz,&msg_bytes);
615 atomic_inc(&msg_hdrs);
629 static inline int convert_mode(long* msgtyp, int msgflg)
632 * find message of correct type.
633 * msgtyp = 0 => get first.
634 * msgtyp > 0 => get first message of matching type.
635 * msgtyp < 0 => get message with least type must be < abs(msgtype).
641 return SEARCH_LESSEQUAL;
643 if(msgflg & MSG_EXCEPT)
644 return SEARCH_NOTEQUAL;
648 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
649 long msgtyp, int msgflg)
651 struct msg_queue *msq;
652 struct msg_receiver msr_d;
653 struct list_head* tmp;
654 struct msg_msg* msg, *found_msg;
658 if (msqid < 0 || (long) msgsz < 0)
660 mode = convert_mode(&msgtyp,msgflg);
662 msq = msg_lock(msqid);
667 if (msg_checkid(msq,msqid))
671 if (ipcperms (&msq->q_perm, S_IRUGO))
674 tmp = msq->q_messages.next;
676 while (tmp != &msq->q_messages) {
677 msg = list_entry(tmp,struct msg_msg,m_list);
678 if(testmsg(msg,msgtyp,mode) &&
679 !security_msg_queue_msgrcv(msq, msg, current, msgtyp, mode)) {
681 if(mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
683 msgtyp=msg->m_type-1;
693 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
697 list_del(&msg->m_list);
699 msq->q_rtime = get_seconds();
700 msq->q_lrpid = current->tgid;
701 msq->q_cbytes -= msg->m_ts;
702 atomic_sub(msg->m_ts,&msg_bytes);
703 atomic_dec(&msg_hdrs);
704 ss_wakeup(&msq->q_senders,0);
707 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
708 if (put_user (msg->m_type, &msgp->mtype) ||
709 store_msg(msgp->mtext, msg, msgsz)) {
716 /* no message waiting. Prepare for pipelined
719 if (msgflg & IPC_NOWAIT) {
723 list_add_tail(&msr_d.r_list,&msq->q_receivers);
724 msr_d.r_tsk = current;
725 msr_d.r_msgtype = msgtyp;
727 if(msgflg & MSG_NOERROR)
728 msr_d.r_maxsize = INT_MAX;
730 msr_d.r_maxsize = msgsz;
731 msr_d.r_msg = ERR_PTR(-EAGAIN);
732 current->state = TASK_INTERRUPTIBLE;
738 * The below optimisation is buggy. A sleeping thread that is
739 * woken up checks if it got a message and if so, copies it to
740 * userspace and just returns without taking any locks.
741 * But this return to user space can be faster than the message
742 * send, and if the receiver immediately exits the
743 * wake_up_process performed by the sender will oops.
746 msg = (struct msg_msg*) msr_d.r_msg;
751 msq = msg_lock(msqid);
752 msg = (struct msg_msg*)msr_d.r_msg;
754 /* our message arived while we waited for
755 * the spinlock. Process it.
765 list_del(&msr_d.r_list);
766 if (signal_pending(current))
778 #ifdef CONFIG_PROC_FS
779 static int sysvipc_msg_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
786 len += sprintf(buffer, " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n");
788 for(i = 0; i <= msg_ids.max_id; i++) {
789 struct msg_queue * msq;
792 len += sprintf(buffer + len, "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
794 msg_buildid(i,msq->q_perm.seq),
814 if(pos > offset + length)
822 *start = buffer + (offset - begin);
823 len -= (offset - begin);