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/vs_base.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;
101 msq->q_perm.xid = current->xid;
103 msq->q_perm.security = NULL;
104 retval = security_msg_queue_alloc(msq);
106 ipc_rcu_free(msq, sizeof(*msq));
110 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
112 security_msg_queue_free(msq);
113 ipc_rcu_free(msq, sizeof(*msq));
117 msq->q_stime = msq->q_rtime = 0;
118 msq->q_ctime = get_seconds();
119 msq->q_cbytes = msq->q_qnum = 0;
120 msq->q_qbytes = msg_ctlmnb;
121 msq->q_lspid = msq->q_lrpid = 0;
122 INIT_LIST_HEAD(&msq->q_messages);
123 INIT_LIST_HEAD(&msq->q_receivers);
124 INIT_LIST_HEAD(&msq->q_senders);
127 return msg_buildid(id,msq->q_perm.seq);
130 static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
133 current->state=TASK_INTERRUPTIBLE;
134 list_add_tail(&mss->list,&msq->q_senders);
137 static inline void ss_del(struct msg_sender* mss)
139 if(mss->list.next != NULL)
140 list_del(&mss->list);
143 static void ss_wakeup(struct list_head* h, int kill)
145 struct list_head *tmp;
149 struct msg_sender* mss;
151 mss = list_entry(tmp,struct msg_sender,list);
155 wake_up_process(mss->tsk);
159 static void expunge_all(struct msg_queue* msq, int res)
161 struct list_head *tmp;
163 tmp = msq->q_receivers.next;
164 while (tmp != &msq->q_receivers) {
165 struct msg_receiver* msr;
167 msr = list_entry(tmp,struct msg_receiver,r_list);
169 msr->r_msg = ERR_PTR(res);
170 wake_up_process(msr->r_tsk);
174 * freeque() wakes up waiters on the sender and receiver waiting queue,
175 * removes the message queue from message queue ID
176 * array, and cleans up all the messages associated with this queue.
178 * msg_ids.sem and the spinlock for this message queue is hold
179 * before freeque() is called. msg_ids.sem remains locked on exit.
181 static void freeque (struct msg_queue *msq, int id)
183 struct list_head *tmp;
185 expunge_all(msq,-EIDRM);
186 ss_wakeup(&msq->q_senders,1);
190 tmp = msq->q_messages.next;
191 while(tmp != &msq->q_messages) {
192 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
194 atomic_dec(&msg_hdrs);
197 atomic_sub(msq->q_cbytes, &msg_bytes);
198 security_msg_queue_free(msq);
199 ipc_rcu_free(msq, sizeof(struct msg_queue));
202 asmlinkage long sys_msgget (key_t key, int msgflg)
204 int id, ret = -EPERM;
205 struct msg_queue *msq;
208 if (key == IPC_PRIVATE)
209 ret = newque(key, msgflg);
210 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
211 if (!(msgflg & IPC_CREAT))
214 ret = newque(key, msgflg);
215 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
221 if (ipcperms(&msq->q_perm, msgflg))
224 int qid = msg_buildid(id, msq->q_perm.seq);
225 ret = security_msg_queue_associate(msq, msgflg);
235 static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
239 return copy_to_user (buf, in, sizeof(*in));
244 memset(&out,0,sizeof(out));
246 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
248 out.msg_stime = in->msg_stime;
249 out.msg_rtime = in->msg_rtime;
250 out.msg_ctime = in->msg_ctime;
252 if(in->msg_cbytes > USHRT_MAX)
253 out.msg_cbytes = USHRT_MAX;
255 out.msg_cbytes = in->msg_cbytes;
256 out.msg_lcbytes = in->msg_cbytes;
258 if(in->msg_qnum > USHRT_MAX)
259 out.msg_qnum = USHRT_MAX;
261 out.msg_qnum = in->msg_qnum;
263 if(in->msg_qbytes > USHRT_MAX)
264 out.msg_qbytes = USHRT_MAX;
266 out.msg_qbytes = in->msg_qbytes;
267 out.msg_lqbytes = in->msg_qbytes;
269 out.msg_lspid = in->msg_lspid;
270 out.msg_lrpid = in->msg_lrpid;
272 return copy_to_user (buf, &out, sizeof(out));
280 unsigned long qbytes;
286 static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
291 struct msqid64_ds tbuf;
293 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
296 out->qbytes = tbuf.msg_qbytes;
297 out->uid = tbuf.msg_perm.uid;
298 out->gid = tbuf.msg_perm.gid;
299 out->mode = tbuf.msg_perm.mode;
305 struct msqid_ds tbuf_old;
307 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
310 out->uid = tbuf_old.msg_perm.uid;
311 out->gid = tbuf_old.msg_perm.gid;
312 out->mode = tbuf_old.msg_perm.mode;
314 if(tbuf_old.msg_qbytes == 0)
315 out->qbytes = tbuf_old.msg_lqbytes;
317 out->qbytes = tbuf_old.msg_qbytes;
326 asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
329 struct msg_queue *msq;
330 struct msq_setbuf setbuf;
331 struct kern_ipc_perm *ipcp;
333 if (msqid < 0 || cmd < 0)
336 version = ipc_parse_version(&cmd);
342 struct msginfo msginfo;
346 /* We must not return kernel stack data.
347 * due to padding, it's not enough
348 * to set all member fields.
351 err = security_msg_queue_msgctl(NULL, cmd);
355 memset(&msginfo,0,sizeof(msginfo));
356 msginfo.msgmni = msg_ctlmni;
357 msginfo.msgmax = msg_ctlmax;
358 msginfo.msgmnb = msg_ctlmnb;
359 msginfo.msgssz = MSGSSZ;
360 msginfo.msgseg = MSGSEG;
362 if (cmd == MSG_INFO) {
363 msginfo.msgpool = msg_ids.in_use;
364 msginfo.msgmap = atomic_read(&msg_hdrs);
365 msginfo.msgtql = atomic_read(&msg_bytes);
367 msginfo.msgmap = MSGMAP;
368 msginfo.msgpool = MSGPOOL;
369 msginfo.msgtql = MSGTQL;
371 max_id = msg_ids.max_id;
373 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
375 return (max_id < 0) ? 0: max_id;
380 struct msqid64_ds tbuf;
384 if(cmd == MSG_STAT && msqid >= msg_ids.size)
387 memset(&tbuf,0,sizeof(tbuf));
389 msq = msg_lock(msqid);
393 if(cmd == MSG_STAT) {
394 success_return = msg_buildid(msqid, msq->q_perm.seq);
397 if (msg_checkid(msq,msqid))
402 if (ipcperms (&msq->q_perm, S_IRUGO))
405 err = security_msg_queue_msgctl(msq, cmd);
409 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
410 tbuf.msg_stime = msq->q_stime;
411 tbuf.msg_rtime = msq->q_rtime;
412 tbuf.msg_ctime = msq->q_ctime;
413 tbuf.msg_cbytes = msq->q_cbytes;
414 tbuf.msg_qnum = msq->q_qnum;
415 tbuf.msg_qbytes = msq->q_qbytes;
416 tbuf.msg_lspid = msq->q_lspid;
417 tbuf.msg_lrpid = msq->q_lrpid;
419 if (copy_msqid_to_user(buf, &tbuf, version))
421 return success_return;
426 if (copy_msqid_from_user (&setbuf, buf, version))
436 msq = msg_lock(msqid);
442 if (msg_checkid(msq,msqid))
446 if (current->euid != ipcp->cuid &&
447 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
448 /* We _could_ check for CAP_CHOWN above, but we don't */
451 err = security_msg_queue_msgctl(msq, cmd);
459 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
462 msq->q_qbytes = setbuf.qbytes;
464 ipcp->uid = setbuf.uid;
465 ipcp->gid = setbuf.gid;
466 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
467 (S_IRWXUGO & setbuf.mode);
468 msq->q_ctime = get_seconds();
469 /* sleeping receivers might be excluded by
470 * stricter permissions.
472 expunge_all(msq,-EAGAIN);
473 /* sleeping senders might be able to send
474 * due to a larger queue size.
476 ss_wakeup(&msq->q_senders,0);
481 freeque (msq, msqid);
496 static int testmsg(struct msg_msg* msg,long type,int mode)
502 case SEARCH_LESSEQUAL:
503 if(msg->m_type <=type)
507 if(msg->m_type == type)
510 case SEARCH_NOTEQUAL:
511 if(msg->m_type != type)
518 static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
520 struct list_head* tmp;
522 tmp = msq->q_receivers.next;
523 while (tmp != &msq->q_receivers) {
524 struct msg_receiver* msr;
525 msr = list_entry(tmp,struct msg_receiver,r_list);
527 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
528 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
529 list_del(&msr->r_list);
530 if(msr->r_maxsize < msg->m_ts) {
531 msr->r_msg = ERR_PTR(-E2BIG);
532 wake_up_process(msr->r_tsk);
535 msq->q_lrpid = msr->r_tsk->pid;
536 msq->q_rtime = get_seconds();
537 wake_up_process(msr->r_tsk);
545 asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
547 struct msg_queue *msq;
552 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
554 if (get_user(mtype, &msgp->mtype))
559 msg = load_msg(msgp->mtext, msgsz);
566 msq = msg_lock(msqid);
572 if (msg_checkid(msq,msqid))
573 goto out_unlock_free;
576 if (ipcperms(&msq->q_perm, S_IWUGO))
577 goto out_unlock_free;
579 err = security_msg_queue_msgsnd(msq, msg, msgflg);
581 goto out_unlock_free;
583 if(msgsz + msq->q_cbytes > msq->q_qbytes ||
584 1 + msq->q_qnum > msq->q_qbytes) {
587 if(msgflg&IPC_NOWAIT) {
589 goto out_unlock_free;
594 current->state= TASK_RUNNING;
596 msq = msg_lock(msqid);
602 if (signal_pending(current)) {
604 goto out_unlock_free;
609 msq->q_lspid = current->tgid;
610 msq->q_stime = get_seconds();
612 if(!pipelined_send(msq,msg)) {
613 /* noone is waiting for this message, enqueue it */
614 list_add_tail(&msg->m_list,&msq->q_messages);
615 msq->q_cbytes += msgsz;
617 atomic_add(msgsz,&msg_bytes);
618 atomic_inc(&msg_hdrs);
632 static inline int convert_mode(long* msgtyp, int msgflg)
635 * find message of correct type.
636 * msgtyp = 0 => get first.
637 * msgtyp > 0 => get first message of matching type.
638 * msgtyp < 0 => get message with least type must be < abs(msgtype).
644 return SEARCH_LESSEQUAL;
646 if(msgflg & MSG_EXCEPT)
647 return SEARCH_NOTEQUAL;
651 asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
652 long msgtyp, int msgflg)
654 struct msg_queue *msq;
655 struct msg_receiver msr_d;
656 struct list_head* tmp;
657 struct msg_msg* msg, *found_msg;
661 if (msqid < 0 || (long) msgsz < 0)
663 mode = convert_mode(&msgtyp,msgflg);
665 msq = msg_lock(msqid);
670 if (msg_checkid(msq,msqid))
674 if (ipcperms (&msq->q_perm, S_IRUGO))
677 tmp = msq->q_messages.next;
679 while (tmp != &msq->q_messages) {
680 msg = list_entry(tmp,struct msg_msg,m_list);
681 if(testmsg(msg,msgtyp,mode) &&
682 !security_msg_queue_msgrcv(msq, msg, current, msgtyp, mode)) {
684 if(mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
686 msgtyp=msg->m_type-1;
696 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
700 list_del(&msg->m_list);
702 msq->q_rtime = get_seconds();
703 msq->q_lrpid = current->tgid;
704 msq->q_cbytes -= msg->m_ts;
705 atomic_sub(msg->m_ts,&msg_bytes);
706 atomic_dec(&msg_hdrs);
707 ss_wakeup(&msq->q_senders,0);
710 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
711 if (put_user (msg->m_type, &msgp->mtype) ||
712 store_msg(msgp->mtext, msg, msgsz)) {
719 /* no message waiting. Prepare for pipelined
722 if (msgflg & IPC_NOWAIT) {
726 list_add_tail(&msr_d.r_list,&msq->q_receivers);
727 msr_d.r_tsk = current;
728 msr_d.r_msgtype = msgtyp;
730 if(msgflg & MSG_NOERROR)
731 msr_d.r_maxsize = INT_MAX;
733 msr_d.r_maxsize = msgsz;
734 msr_d.r_msg = ERR_PTR(-EAGAIN);
735 current->state = TASK_INTERRUPTIBLE;
741 * The below optimisation is buggy. A sleeping thread that is
742 * woken up checks if it got a message and if so, copies it to
743 * userspace and just returns without taking any locks.
744 * But this return to user space can be faster than the message
745 * send, and if the receiver immediately exits the
746 * wake_up_process performed by the sender will oops.
749 msg = (struct msg_msg*) msr_d.r_msg;
754 msq = msg_lock(msqid);
755 msg = (struct msg_msg*)msr_d.r_msg;
757 /* our message arived while we waited for
758 * the spinlock. Process it.
768 list_del(&msr_d.r_list);
769 if (signal_pending(current))
781 #ifdef CONFIG_PROC_FS
782 static int sysvipc_msg_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
789 len += sprintf(buffer, " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n");
791 for(i = 0; i <= msg_ids.max_id; i++) {
792 struct msg_queue * msq;
795 if (!vx_check(msq->q_perm.xid, VX_IDENT)) {
799 len += sprintf(buffer + len, "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
801 msg_buildid(i,msq->q_perm.seq),
821 if(pos > offset + length)
829 *start = buffer + (offset - begin);
830 len -= (offset - begin);