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>
35 #include <asm/current.h>
36 #include <asm/uaccess.h>
40 int msg_ctlmax = MSGMAX;
41 int msg_ctlmnb = MSGMNB;
42 int msg_ctlmni = MSGMNI;
45 * one msg_receiver structure for each sleeping receiver:
48 struct list_head r_list;
49 struct task_struct *r_tsk;
55 volatile struct msg_msg *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) ipc_checkid(&msg_ids, &msq->q_perm, msgid)
78 #define msg_buildid(id, seq) ipc_buildid(&msg_ids, id, seq)
80 static void freeque(struct msg_queue *msq, int id);
81 static int newque(key_t key, int msgflg);
83 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
86 void __init msg_init(void)
88 ipc_init_ids(&msg_ids, msg_ctlmni);
89 ipc_init_proc_interface("sysvipc/msg",
90 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
92 sysvipc_msg_proc_show);
95 static int newque(key_t key, int msgflg)
97 struct msg_queue *msq;
100 msq = ipc_rcu_alloc(sizeof(*msq));
104 msq->q_perm.mode = msgflg & S_IRWXUGO;
105 msq->q_perm.key = key;
107 msq->q_perm.security = NULL;
108 retval = security_msg_queue_alloc(msq);
114 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
116 security_msg_queue_free(msq);
121 msq->q_id = msg_buildid(id, msq->q_perm.seq);
122 msq->q_stime = msq->q_rtime = 0;
123 msq->q_ctime = get_seconds();
124 msq->q_cbytes = msq->q_qnum = 0;
125 msq->q_qbytes = msg_ctlmnb;
126 msq->q_lspid = msq->q_lrpid = 0;
127 INIT_LIST_HEAD(&msq->q_messages);
128 INIT_LIST_HEAD(&msq->q_receivers);
129 INIT_LIST_HEAD(&msq->q_senders);
135 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
138 current->state = TASK_INTERRUPTIBLE;
139 list_add_tail(&mss->list, &msq->q_senders);
142 static inline void ss_del(struct msg_sender *mss)
144 if (mss->list.next != NULL)
145 list_del(&mss->list);
148 static void ss_wakeup(struct list_head *h, int kill)
150 struct list_head *tmp;
154 struct msg_sender *mss;
156 mss = list_entry(tmp, struct msg_sender, list);
159 mss->list.next = NULL;
160 wake_up_process(mss->tsk);
164 static void expunge_all(struct msg_queue *msq, int res)
166 struct list_head *tmp;
168 tmp = msq->q_receivers.next;
169 while (tmp != &msq->q_receivers) {
170 struct msg_receiver *msr;
172 msr = list_entry(tmp, struct msg_receiver, r_list);
175 wake_up_process(msr->r_tsk);
177 msr->r_msg = ERR_PTR(res);
182 * freeque() wakes up waiters on the sender and receiver waiting queue,
183 * removes the message queue from message queue ID
184 * array, and cleans up all the messages associated with this queue.
186 * msg_ids.mutex and the spinlock for this message queue is hold
187 * before freeque() is called. msg_ids.mutex remains locked on exit.
189 static void freeque(struct msg_queue *msq, int id)
191 struct list_head *tmp;
193 expunge_all(msq, -EIDRM);
194 ss_wakeup(&msq->q_senders, 1);
198 tmp = msq->q_messages.next;
199 while (tmp != &msq->q_messages) {
200 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
203 atomic_dec(&msg_hdrs);
206 atomic_sub(msq->q_cbytes, &msg_bytes);
207 security_msg_queue_free(msq);
211 asmlinkage long sys_msgget(key_t key, int msgflg)
213 struct msg_queue *msq;
214 int id, ret = -EPERM;
216 mutex_lock(&msg_ids.mutex);
217 if (key == IPC_PRIVATE)
218 ret = newque(key, msgflg);
219 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
220 if (!(msgflg & IPC_CREAT))
223 ret = newque(key, msgflg);
224 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
229 if (ipcperms(&msq->q_perm, msgflg))
232 int qid = msg_buildid(id, msq->q_perm.seq);
234 ret = security_msg_queue_associate(msq, msgflg);
240 mutex_unlock(&msg_ids.mutex);
245 static inline unsigned long
246 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
250 return copy_to_user(buf, in, sizeof(*in));
255 memset(&out, 0, sizeof(out));
257 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
259 out.msg_stime = in->msg_stime;
260 out.msg_rtime = in->msg_rtime;
261 out.msg_ctime = in->msg_ctime;
263 if (in->msg_cbytes > USHRT_MAX)
264 out.msg_cbytes = USHRT_MAX;
266 out.msg_cbytes = in->msg_cbytes;
267 out.msg_lcbytes = in->msg_cbytes;
269 if (in->msg_qnum > USHRT_MAX)
270 out.msg_qnum = USHRT_MAX;
272 out.msg_qnum = in->msg_qnum;
274 if (in->msg_qbytes > USHRT_MAX)
275 out.msg_qbytes = USHRT_MAX;
277 out.msg_qbytes = in->msg_qbytes;
278 out.msg_lqbytes = in->msg_qbytes;
280 out.msg_lspid = in->msg_lspid;
281 out.msg_lrpid = in->msg_lrpid;
283 return copy_to_user(buf, &out, sizeof(out));
291 unsigned long qbytes;
297 static inline unsigned long
298 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
303 struct msqid64_ds tbuf;
305 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
308 out->qbytes = tbuf.msg_qbytes;
309 out->uid = tbuf.msg_perm.uid;
310 out->gid = tbuf.msg_perm.gid;
311 out->mode = tbuf.msg_perm.mode;
317 struct msqid_ds tbuf_old;
319 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
322 out->uid = tbuf_old.msg_perm.uid;
323 out->gid = tbuf_old.msg_perm.gid;
324 out->mode = tbuf_old.msg_perm.mode;
326 if (tbuf_old.msg_qbytes == 0)
327 out->qbytes = tbuf_old.msg_lqbytes;
329 out->qbytes = tbuf_old.msg_qbytes;
338 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
340 struct kern_ipc_perm *ipcp;
341 struct msq_setbuf setbuf;
342 struct msg_queue *msq;
345 if (msqid < 0 || cmd < 0)
348 version = ipc_parse_version(&cmd);
354 struct msginfo msginfo;
360 * We must not return kernel stack data.
361 * due to padding, it's not enough
362 * to set all member fields.
364 err = security_msg_queue_msgctl(NULL, cmd);
368 memset(&msginfo, 0, sizeof(msginfo));
369 msginfo.msgmni = msg_ctlmni;
370 msginfo.msgmax = msg_ctlmax;
371 msginfo.msgmnb = msg_ctlmnb;
372 msginfo.msgssz = MSGSSZ;
373 msginfo.msgseg = MSGSEG;
374 mutex_lock(&msg_ids.mutex);
375 if (cmd == MSG_INFO) {
376 msginfo.msgpool = msg_ids.in_use;
377 msginfo.msgmap = atomic_read(&msg_hdrs);
378 msginfo.msgtql = atomic_read(&msg_bytes);
380 msginfo.msgmap = MSGMAP;
381 msginfo.msgpool = MSGPOOL;
382 msginfo.msgtql = MSGTQL;
384 max_id = msg_ids.max_id;
385 mutex_unlock(&msg_ids.mutex);
386 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
388 return (max_id < 0) ? 0 : max_id;
393 struct msqid64_ds tbuf;
398 if (cmd == MSG_STAT && msqid >= msg_ids.entries->size)
401 memset(&tbuf, 0, sizeof(tbuf));
403 msq = msg_lock(msqid);
407 if (cmd == MSG_STAT) {
408 success_return = msg_buildid(msqid, msq->q_perm.seq);
411 if (msg_checkid(msq, msqid))
416 if (ipcperms(&msq->q_perm, S_IRUGO))
419 err = security_msg_queue_msgctl(msq, cmd);
423 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
424 tbuf.msg_stime = msq->q_stime;
425 tbuf.msg_rtime = msq->q_rtime;
426 tbuf.msg_ctime = msq->q_ctime;
427 tbuf.msg_cbytes = msq->q_cbytes;
428 tbuf.msg_qnum = msq->q_qnum;
429 tbuf.msg_qbytes = msq->q_qbytes;
430 tbuf.msg_lspid = msq->q_lspid;
431 tbuf.msg_lrpid = msq->q_lrpid;
433 if (copy_msqid_to_user(buf, &tbuf, version))
435 return success_return;
440 if (copy_msqid_from_user(&setbuf, buf, version))
449 mutex_lock(&msg_ids.mutex);
450 msq = msg_lock(msqid);
456 if (msg_checkid(msq, msqid))
460 err = audit_ipc_obj(ipcp);
464 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
471 if (current->euid != ipcp->cuid &&
472 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
473 /* We _could_ check for CAP_CHOWN above, but we don't */
476 err = security_msg_queue_msgctl(msq, cmd);
484 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
487 msq->q_qbytes = setbuf.qbytes;
489 ipcp->uid = setbuf.uid;
490 ipcp->gid = setbuf.gid;
491 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
492 (S_IRWXUGO & setbuf.mode);
493 msq->q_ctime = get_seconds();
494 /* sleeping receivers might be excluded by
495 * stricter permissions.
497 expunge_all(msq, -EAGAIN);
498 /* sleeping senders might be able to send
499 * due to a larger queue size.
501 ss_wakeup(&msq->q_senders, 0);
511 mutex_unlock(&msg_ids.mutex);
521 static int testmsg(struct msg_msg *msg, long type, int mode)
527 case SEARCH_LESSEQUAL:
528 if (msg->m_type <=type)
532 if (msg->m_type == type)
535 case SEARCH_NOTEQUAL:
536 if (msg->m_type != type)
543 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
545 struct list_head *tmp;
547 tmp = msq->q_receivers.next;
548 while (tmp != &msq->q_receivers) {
549 struct msg_receiver *msr;
551 msr = list_entry(tmp, struct msg_receiver, r_list);
553 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
554 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
555 msr->r_msgtype, msr->r_mode)) {
557 list_del(&msr->r_list);
558 if (msr->r_maxsize < msg->m_ts) {
560 wake_up_process(msr->r_tsk);
562 msr->r_msg = ERR_PTR(-E2BIG);
565 msq->q_lrpid = msr->r_tsk->pid;
566 msq->q_rtime = get_seconds();
567 wake_up_process(msr->r_tsk);
579 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
581 struct msg_queue *msq;
586 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
588 if (get_user(mtype, &msgp->mtype))
593 msg = load_msg(msgp->mtext, msgsz);
600 msq = msg_lock(msqid);
606 if (msg_checkid(msq, msqid))
607 goto out_unlock_free;
613 if (ipcperms(&msq->q_perm, S_IWUGO))
614 goto out_unlock_free;
616 err = security_msg_queue_msgsnd(msq, msg, msgflg);
618 goto out_unlock_free;
620 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
621 1 + msq->q_qnum <= msq->q_qbytes) {
625 /* queue full, wait: */
626 if (msgflg & IPC_NOWAIT) {
628 goto out_unlock_free;
635 ipc_lock_by_ptr(&msq->q_perm);
637 if (msq->q_perm.deleted) {
639 goto out_unlock_free;
643 if (signal_pending(current)) {
644 err = -ERESTARTNOHAND;
645 goto out_unlock_free;
649 msq->q_lspid = current->tgid;
650 msq->q_stime = get_seconds();
652 if (!pipelined_send(msq, msg)) {
653 /* noone is waiting for this message, enqueue it */
654 list_add_tail(&msg->m_list, &msq->q_messages);
655 msq->q_cbytes += msgsz;
657 atomic_add(msgsz, &msg_bytes);
658 atomic_inc(&msg_hdrs);
672 static inline int convert_mode(long *msgtyp, int msgflg)
675 * find message of correct type.
676 * msgtyp = 0 => get first.
677 * msgtyp > 0 => get first message of matching type.
678 * msgtyp < 0 => get message with least type must be < abs(msgtype).
684 return SEARCH_LESSEQUAL;
686 if (msgflg & MSG_EXCEPT)
687 return SEARCH_NOTEQUAL;
691 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
692 long msgtyp, int msgflg)
694 struct msg_queue *msq;
698 if (msqid < 0 || (long) msgsz < 0)
700 mode = convert_mode(&msgtyp, msgflg);
702 msq = msg_lock(msqid);
706 msg = ERR_PTR(-EIDRM);
707 if (msg_checkid(msq, msqid))
711 struct msg_receiver msr_d;
712 struct list_head *tmp;
714 msg = ERR_PTR(-EACCES);
715 if (ipcperms(&msq->q_perm, S_IRUGO))
718 msg = ERR_PTR(-EAGAIN);
719 tmp = msq->q_messages.next;
720 while (tmp != &msq->q_messages) {
721 struct msg_msg *walk_msg;
723 walk_msg = list_entry(tmp, struct msg_msg, m_list);
724 if (testmsg(walk_msg, msgtyp, mode) &&
725 !security_msg_queue_msgrcv(msq, walk_msg, current,
729 if (mode == SEARCH_LESSEQUAL &&
730 walk_msg->m_type != 1) {
732 msgtyp = walk_msg->m_type - 1;
742 * Found a suitable message.
743 * Unlink it from the queue.
745 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
746 msg = ERR_PTR(-E2BIG);
749 list_del(&msg->m_list);
751 msq->q_rtime = get_seconds();
752 msq->q_lrpid = current->tgid;
753 msq->q_cbytes -= msg->m_ts;
754 atomic_sub(msg->m_ts, &msg_bytes);
755 atomic_dec(&msg_hdrs);
756 ss_wakeup(&msq->q_senders, 0);
760 /* No message waiting. Wait for a message */
761 if (msgflg & IPC_NOWAIT) {
762 msg = ERR_PTR(-ENOMSG);
765 list_add_tail(&msr_d.r_list, &msq->q_receivers);
766 msr_d.r_tsk = current;
767 msr_d.r_msgtype = msgtyp;
769 if (msgflg & MSG_NOERROR)
770 msr_d.r_maxsize = INT_MAX;
772 msr_d.r_maxsize = msgsz;
773 msr_d.r_msg = ERR_PTR(-EAGAIN);
774 current->state = TASK_INTERRUPTIBLE;
779 /* Lockless receive, part 1:
780 * Disable preemption. We don't hold a reference to the queue
781 * and getting a reference would defeat the idea of a lockless
782 * operation, thus the code relies on rcu to guarantee the
784 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
785 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
786 * rcu_read_lock() prevents preemption between reading r_msg
787 * and the spin_lock() inside ipc_lock_by_ptr().
791 /* Lockless receive, part 2:
792 * Wait until pipelined_send or expunge_all are outside of
793 * wake_up_process(). There is a race with exit(), see
794 * ipc/mqueue.c for the details.
796 msg = (struct msg_msg*)msr_d.r_msg;
797 while (msg == NULL) {
799 msg = (struct msg_msg *)msr_d.r_msg;
802 /* Lockless receive, part 3:
803 * If there is a message or an error then accept it without
806 if (msg != ERR_PTR(-EAGAIN)) {
811 /* Lockless receive, part 3:
812 * Acquire the queue spinlock.
814 ipc_lock_by_ptr(&msq->q_perm);
817 /* Lockless receive, part 4:
818 * Repeat test after acquiring the spinlock.
820 msg = (struct msg_msg*)msr_d.r_msg;
821 if (msg != ERR_PTR(-EAGAIN))
824 list_del(&msr_d.r_list);
825 if (signal_pending(current)) {
826 msg = ERR_PTR(-ERESTARTNOHAND);
835 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
836 if (put_user (msg->m_type, &msgp->mtype) ||
837 store_msg(msgp->mtext, msg, msgsz)) {
845 #ifdef CONFIG_PROC_FS
846 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
848 struct msg_queue *msq = it;
851 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",