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/slab.h>
23 #include <linux/msg.h>
24 #include <linux/spinlock.h>
25 #include <linux/init.h>
26 #include <linux/proc_fs.h>
27 #include <linux/list.h>
28 #include <linux/security.h>
29 #include <linux/sched.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32 #include <linux/seq_file.h>
33 #include <linux/mutex.h>
34 #include <linux/vs_base.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;
46 * one msg_receiver structure for each sleeping receiver:
49 struct list_head r_list;
50 struct task_struct *r_tsk;
56 volatile struct msg_msg *r_msg;
59 /* one msg_sender for each sleeping sender */
61 struct list_head list;
62 struct task_struct *tsk;
66 #define SEARCH_EQUAL 2
67 #define SEARCH_NOTEQUAL 3
68 #define SEARCH_LESSEQUAL 4
70 static atomic_t msg_bytes = ATOMIC_INIT(0);
71 static atomic_t msg_hdrs = ATOMIC_INIT(0);
73 static struct ipc_ids msg_ids;
75 #define msg_lock(id) ((struct msg_queue *)ipc_lock(&msg_ids, id))
76 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
77 #define msg_rmid(id) ((struct msg_queue *)ipc_rmid(&msg_ids, id))
78 #define msg_checkid(msq, msgid) ipc_checkid(&msg_ids, &msq->q_perm, msgid)
79 #define msg_buildid(id, seq) 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)
98 struct msg_queue *msq;
101 msq = ipc_rcu_alloc(sizeof(*msq));
105 msq->q_perm.mode = msgflg & S_IRWXUGO;
106 msq->q_perm.key = key;
107 msq->q_perm.xid = vx_current_xid();
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);
161 mss->list.next = NULL;
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);
184 * freeque() wakes up waiters on the sender and receiver waiting queue,
185 * removes the message queue from message queue ID
186 * array, and cleans up all the messages associated with this queue.
188 * msg_ids.mutex and the spinlock for this message queue is hold
189 * before freeque() is called. msg_ids.mutex remains locked on exit.
191 static void freeque(struct msg_queue *msq, int id)
193 struct list_head *tmp;
195 expunge_all(msq, -EIDRM);
196 ss_wakeup(&msq->q_senders, 1);
200 tmp = msq->q_messages.next;
201 while (tmp != &msq->q_messages) {
202 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 struct msg_queue *msq;
216 int id, ret = -EPERM;
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);
236 ret = security_msg_queue_associate(msq, msgflg);
242 mutex_unlock(&msg_ids.mutex);
247 static inline unsigned long
248 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
252 return copy_to_user(buf, in, sizeof(*in));
257 memset(&out, 0, sizeof(out));
259 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
261 out.msg_stime = in->msg_stime;
262 out.msg_rtime = in->msg_rtime;
263 out.msg_ctime = in->msg_ctime;
265 if (in->msg_cbytes > USHRT_MAX)
266 out.msg_cbytes = USHRT_MAX;
268 out.msg_cbytes = in->msg_cbytes;
269 out.msg_lcbytes = in->msg_cbytes;
271 if (in->msg_qnum > USHRT_MAX)
272 out.msg_qnum = USHRT_MAX;
274 out.msg_qnum = in->msg_qnum;
276 if (in->msg_qbytes > USHRT_MAX)
277 out.msg_qbytes = USHRT_MAX;
279 out.msg_qbytes = in->msg_qbytes;
280 out.msg_lqbytes = in->msg_qbytes;
282 out.msg_lspid = in->msg_lspid;
283 out.msg_lrpid = in->msg_lrpid;
285 return copy_to_user(buf, &out, sizeof(out));
293 unsigned long qbytes;
299 static inline unsigned long
300 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
305 struct msqid64_ds tbuf;
307 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
310 out->qbytes = tbuf.msg_qbytes;
311 out->uid = tbuf.msg_perm.uid;
312 out->gid = tbuf.msg_perm.gid;
313 out->mode = tbuf.msg_perm.mode;
319 struct msqid_ds tbuf_old;
321 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
324 out->uid = tbuf_old.msg_perm.uid;
325 out->gid = tbuf_old.msg_perm.gid;
326 out->mode = tbuf_old.msg_perm.mode;
328 if (tbuf_old.msg_qbytes == 0)
329 out->qbytes = tbuf_old.msg_lqbytes;
331 out->qbytes = tbuf_old.msg_qbytes;
340 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
342 struct kern_ipc_perm *ipcp;
343 struct msq_setbuf setbuf;
344 struct msg_queue *msq;
347 if (msqid < 0 || cmd < 0)
350 version = ipc_parse_version(&cmd);
356 struct msginfo msginfo;
362 * We must not return kernel stack data.
363 * due to padding, it's not enough
364 * to set all member fields.
366 err = security_msg_queue_msgctl(NULL, cmd);
370 memset(&msginfo, 0, sizeof(msginfo));
371 msginfo.msgmni = msg_ctlmni;
372 msginfo.msgmax = msg_ctlmax;
373 msginfo.msgmnb = msg_ctlmnb;
374 msginfo.msgssz = MSGSSZ;
375 msginfo.msgseg = MSGSEG;
376 mutex_lock(&msg_ids.mutex);
377 if (cmd == MSG_INFO) {
378 msginfo.msgpool = msg_ids.in_use;
379 msginfo.msgmap = atomic_read(&msg_hdrs);
380 msginfo.msgtql = atomic_read(&msg_bytes);
382 msginfo.msgmap = MSGMAP;
383 msginfo.msgpool = MSGPOOL;
384 msginfo.msgtql = MSGTQL;
386 max_id = msg_ids.max_id;
387 mutex_unlock(&msg_ids.mutex);
388 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
390 return (max_id < 0) ? 0 : max_id;
395 struct msqid64_ds tbuf;
400 if (cmd == MSG_STAT && msqid >= msg_ids.entries->size)
403 memset(&tbuf, 0, sizeof(tbuf));
405 msq = msg_lock(msqid);
409 if (cmd == MSG_STAT) {
410 success_return = msg_buildid(msqid, msq->q_perm.seq);
413 if (msg_checkid(msq, msqid))
418 if (ipcperms(&msq->q_perm, S_IRUGO))
421 err = security_msg_queue_msgctl(msq, cmd);
425 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
426 tbuf.msg_stime = msq->q_stime;
427 tbuf.msg_rtime = msq->q_rtime;
428 tbuf.msg_ctime = msq->q_ctime;
429 tbuf.msg_cbytes = msq->q_cbytes;
430 tbuf.msg_qnum = msq->q_qnum;
431 tbuf.msg_qbytes = msq->q_qbytes;
432 tbuf.msg_lspid = msq->q_lspid;
433 tbuf.msg_lrpid = msq->q_lrpid;
435 if (copy_msqid_to_user(buf, &tbuf, version))
437 return success_return;
442 if (copy_msqid_from_user(&setbuf, buf, version))
451 mutex_lock(&msg_ids.mutex);
452 msq = msg_lock(msqid);
458 if (msg_checkid(msq, msqid))
462 err = audit_ipc_obj(ipcp);
466 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
473 if (current->euid != ipcp->cuid &&
474 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
475 /* We _could_ check for CAP_CHOWN above, but we don't */
478 err = security_msg_queue_msgctl(msq, cmd);
486 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
489 msq->q_qbytes = setbuf.qbytes;
491 ipcp->uid = setbuf.uid;
492 ipcp->gid = setbuf.gid;
493 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
494 (S_IRWXUGO & setbuf.mode);
495 msq->q_ctime = get_seconds();
496 /* sleeping receivers might be excluded by
497 * stricter permissions.
499 expunge_all(msq, -EAGAIN);
500 /* sleeping senders might be able to send
501 * due to a larger queue size.
503 ss_wakeup(&msq->q_senders, 0);
513 mutex_unlock(&msg_ids.mutex);
523 static int testmsg(struct msg_msg *msg, long type, int mode)
529 case SEARCH_LESSEQUAL:
530 if (msg->m_type <=type)
534 if (msg->m_type == type)
537 case SEARCH_NOTEQUAL:
538 if (msg->m_type != type)
545 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
547 struct list_head *tmp;
549 tmp = msq->q_receivers.next;
550 while (tmp != &msq->q_receivers) {
551 struct msg_receiver *msr;
553 msr = list_entry(tmp, struct msg_receiver, r_list);
555 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
556 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
557 msr->r_msgtype, msr->r_mode)) {
559 list_del(&msr->r_list);
560 if (msr->r_maxsize < msg->m_ts) {
562 wake_up_process(msr->r_tsk);
564 msr->r_msg = ERR_PTR(-E2BIG);
567 msq->q_lrpid = msr->r_tsk->pid;
568 msq->q_rtime = get_seconds();
569 wake_up_process(msr->r_tsk);
581 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
583 struct msg_queue *msq;
588 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
590 if (get_user(mtype, &msgp->mtype))
595 msg = load_msg(msgp->mtext, msgsz);
602 msq = msg_lock(msqid);
608 if (msg_checkid(msq, msqid))
609 goto out_unlock_free;
615 if (ipcperms(&msq->q_perm, S_IWUGO))
616 goto out_unlock_free;
618 err = security_msg_queue_msgsnd(msq, msg, msgflg);
620 goto out_unlock_free;
622 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
623 1 + msq->q_qnum <= msq->q_qbytes) {
627 /* queue full, wait: */
628 if (msgflg & IPC_NOWAIT) {
630 goto out_unlock_free;
637 ipc_lock_by_ptr(&msq->q_perm);
639 if (msq->q_perm.deleted) {
641 goto out_unlock_free;
645 if (signal_pending(current)) {
646 err = -ERESTARTNOHAND;
647 goto out_unlock_free;
651 msq->q_lspid = current->tgid;
652 msq->q_stime = get_seconds();
654 if (!pipelined_send(msq, msg)) {
655 /* noone is waiting for this message, enqueue it */
656 list_add_tail(&msg->m_list, &msq->q_messages);
657 msq->q_cbytes += msgsz;
659 atomic_add(msgsz, &msg_bytes);
660 atomic_inc(&msg_hdrs);
674 static inline int convert_mode(long *msgtyp, int msgflg)
677 * find message of correct type.
678 * msgtyp = 0 => get first.
679 * msgtyp > 0 => get first message of matching type.
680 * msgtyp < 0 => get message with least type must be < abs(msgtype).
686 return SEARCH_LESSEQUAL;
688 if (msgflg & MSG_EXCEPT)
689 return SEARCH_NOTEQUAL;
693 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
694 long msgtyp, int msgflg)
696 struct msg_queue *msq;
700 if (msqid < 0 || (long) msgsz < 0)
702 mode = convert_mode(&msgtyp, msgflg);
704 msq = msg_lock(msqid);
708 msg = ERR_PTR(-EIDRM);
709 if (msg_checkid(msq, msqid))
713 struct msg_receiver msr_d;
714 struct list_head *tmp;
716 msg = ERR_PTR(-EACCES);
717 if (ipcperms(&msq->q_perm, S_IRUGO))
720 msg = ERR_PTR(-EAGAIN);
721 tmp = msq->q_messages.next;
722 while (tmp != &msq->q_messages) {
723 struct msg_msg *walk_msg;
725 walk_msg = list_entry(tmp, struct msg_msg, m_list);
726 if (testmsg(walk_msg, msgtyp, mode) &&
727 !security_msg_queue_msgrcv(msq, walk_msg, current,
731 if (mode == SEARCH_LESSEQUAL &&
732 walk_msg->m_type != 1) {
734 msgtyp = walk_msg->m_type - 1;
744 * Found a suitable message.
745 * Unlink it from the queue.
747 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
748 msg = ERR_PTR(-E2BIG);
751 list_del(&msg->m_list);
753 msq->q_rtime = get_seconds();
754 msq->q_lrpid = current->tgid;
755 msq->q_cbytes -= msg->m_ts;
756 atomic_sub(msg->m_ts, &msg_bytes);
757 atomic_dec(&msg_hdrs);
758 ss_wakeup(&msq->q_senders, 0);
762 /* No message waiting. Wait for a message */
763 if (msgflg & IPC_NOWAIT) {
764 msg = ERR_PTR(-ENOMSG);
767 list_add_tail(&msr_d.r_list, &msq->q_receivers);
768 msr_d.r_tsk = current;
769 msr_d.r_msgtype = msgtyp;
771 if (msgflg & MSG_NOERROR)
772 msr_d.r_maxsize = INT_MAX;
774 msr_d.r_maxsize = msgsz;
775 msr_d.r_msg = ERR_PTR(-EAGAIN);
776 current->state = TASK_INTERRUPTIBLE;
781 /* Lockless receive, part 1:
782 * Disable preemption. We don't hold a reference to the queue
783 * and getting a reference would defeat the idea of a lockless
784 * operation, thus the code relies on rcu to guarantee the
786 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
787 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
788 * rcu_read_lock() prevents preemption between reading r_msg
789 * and the spin_lock() inside ipc_lock_by_ptr().
793 /* Lockless receive, part 2:
794 * Wait until pipelined_send or expunge_all are outside of
795 * wake_up_process(). There is a race with exit(), see
796 * ipc/mqueue.c for the details.
798 msg = (struct msg_msg*)msr_d.r_msg;
799 while (msg == NULL) {
801 msg = (struct msg_msg *)msr_d.r_msg;
804 /* Lockless receive, part 3:
805 * If there is a message or an error then accept it without
808 if (msg != ERR_PTR(-EAGAIN)) {
813 /* Lockless receive, part 3:
814 * Acquire the queue spinlock.
816 ipc_lock_by_ptr(&msq->q_perm);
819 /* Lockless receive, part 4:
820 * Repeat test after acquiring the spinlock.
822 msg = (struct msg_msg*)msr_d.r_msg;
823 if (msg != ERR_PTR(-EAGAIN))
826 list_del(&msr_d.r_list);
827 if (signal_pending(current)) {
828 msg = ERR_PTR(-ERESTARTNOHAND);
837 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
838 if (put_user (msg->m_type, &msgp->mtype) ||
839 store_msg(msgp->mtext, msg, msgsz)) {
847 #ifdef CONFIG_PROC_FS
848 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
850 struct msg_queue *msq = it;
852 if (!vx_check(msq->q_perm.xid, VX_IDENT))
856 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",