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>
22 * Pavel Emelianov <xemul@openvz.org>
25 #include <linux/capability.h>
26 #include <linux/slab.h>
27 #include <linux/msg.h>
28 #include <linux/spinlock.h>
29 #include <linux/init.h>
30 #include <linux/proc_fs.h>
31 #include <linux/list.h>
32 #include <linux/security.h>
33 #include <linux/sched.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/seq_file.h>
37 #include <linux/mutex.h>
38 #include <linux/nsproxy.h>
40 #include <asm/current.h>
41 #include <asm/uaccess.h>
45 * one msg_receiver structure for each sleeping receiver:
48 struct list_head r_list;
49 struct task_struct *r_tsk;
55 struct msg_msg *volatile 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 init_msg_ids;
74 #define msg_ids(ns) (*((ns)->ids[IPC_MSG_IDS]))
76 #define msg_lock(ns, id) ((struct msg_queue*)ipc_lock(&msg_ids(ns), id))
77 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
78 #define msg_rmid(ns, id) ((struct msg_queue*)ipc_rmid(&msg_ids(ns), id))
79 #define msg_checkid(ns, msq, msgid) \
80 ipc_checkid(&msg_ids(ns), &msq->q_perm, msgid)
81 #define msg_buildid(ns, id, seq) \
82 ipc_buildid(&msg_ids(ns), id, seq)
84 static void freeque (struct ipc_namespace *ns, struct msg_queue *msq, int id);
85 static int newque (struct ipc_namespace *ns, key_t key, int msgflg);
87 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
90 static void __ipc_init __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
92 ns->ids[IPC_MSG_IDS] = ids;
93 ns->msg_ctlmax = MSGMAX;
94 ns->msg_ctlmnb = MSGMNB;
95 ns->msg_ctlmni = MSGMNI;
96 ipc_init_ids(ids, ns->msg_ctlmni);
100 int msg_init_ns(struct ipc_namespace *ns)
104 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
108 __msg_init_ns(ns, ids);
112 void msg_exit_ns(struct ipc_namespace *ns)
115 struct msg_queue *msq;
117 mutex_lock(&msg_ids(ns).mutex);
118 for (i = 0; i <= msg_ids(ns).max_id; i++) {
119 msq = msg_lock(ns, i);
125 mutex_unlock(&msg_ids(ns).mutex);
127 ipc_fini_ids(ns->ids[IPC_MSG_IDS]);
128 kfree(ns->ids[IPC_MSG_IDS]);
129 ns->ids[IPC_MSG_IDS] = NULL;
133 void __init msg_init(void)
135 __msg_init_ns(&init_ipc_ns, &init_msg_ids);
136 ipc_init_proc_interface("sysvipc/msg",
137 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
138 IPC_MSG_IDS, sysvipc_msg_proc_show);
141 static int newque (struct ipc_namespace *ns, key_t key, int msgflg)
143 struct msg_queue *msq;
146 msq = ipc_rcu_alloc(sizeof(*msq));
150 msq->q_perm.mode = msgflg & S_IRWXUGO;
151 msq->q_perm.key = key;
153 msq->q_perm.security = NULL;
154 retval = security_msg_queue_alloc(msq);
160 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
162 security_msg_queue_free(msq);
167 msq->q_id = msg_buildid(ns, id, msq->q_perm.seq);
168 msq->q_stime = msq->q_rtime = 0;
169 msq->q_ctime = get_seconds();
170 msq->q_cbytes = msq->q_qnum = 0;
171 msq->q_qbytes = ns->msg_ctlmnb;
172 msq->q_lspid = msq->q_lrpid = 0;
173 INIT_LIST_HEAD(&msq->q_messages);
174 INIT_LIST_HEAD(&msq->q_receivers);
175 INIT_LIST_HEAD(&msq->q_senders);
181 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
184 current->state = TASK_INTERRUPTIBLE;
185 list_add_tail(&mss->list, &msq->q_senders);
188 static inline void ss_del(struct msg_sender *mss)
190 if (mss->list.next != NULL)
191 list_del(&mss->list);
194 static void ss_wakeup(struct list_head *h, int kill)
196 struct list_head *tmp;
200 struct msg_sender *mss;
202 mss = list_entry(tmp, struct msg_sender, list);
205 mss->list.next = NULL;
206 wake_up_process(mss->tsk);
210 static void expunge_all(struct msg_queue *msq, int res)
212 struct list_head *tmp;
214 tmp = msq->q_receivers.next;
215 while (tmp != &msq->q_receivers) {
216 struct msg_receiver *msr;
218 msr = list_entry(tmp, struct msg_receiver, r_list);
221 wake_up_process(msr->r_tsk);
223 msr->r_msg = ERR_PTR(res);
228 * freeque() wakes up waiters on the sender and receiver waiting queue,
229 * removes the message queue from message queue ID
230 * array, and cleans up all the messages associated with this queue.
232 * msg_ids.mutex and the spinlock for this message queue is hold
233 * before freeque() is called. msg_ids.mutex remains locked on exit.
235 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq, int id)
237 struct list_head *tmp;
239 expunge_all(msq, -EIDRM);
240 ss_wakeup(&msq->q_senders, 1);
241 msq = msg_rmid(ns, id);
244 tmp = msq->q_messages.next;
245 while (tmp != &msq->q_messages) {
246 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
249 atomic_dec(&msg_hdrs);
252 atomic_sub(msq->q_cbytes, &msg_bytes);
253 security_msg_queue_free(msq);
257 asmlinkage long sys_msgget(key_t key, int msgflg)
259 struct msg_queue *msq;
260 int id, ret = -EPERM;
261 struct ipc_namespace *ns;
263 ns = current->nsproxy->ipc_ns;
265 mutex_lock(&msg_ids(ns).mutex);
266 if (key == IPC_PRIVATE)
267 ret = newque(ns, key, msgflg);
268 else if ((id = ipc_findkey(&msg_ids(ns), key)) == -1) { /* key not used */
269 if (!(msgflg & IPC_CREAT))
272 ret = newque(ns, key, msgflg);
273 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
276 msq = msg_lock(ns, id);
278 if (ipcperms(&msq->q_perm, msgflg))
281 int qid = msg_buildid(ns, id, msq->q_perm.seq);
283 ret = security_msg_queue_associate(msq, msgflg);
289 mutex_unlock(&msg_ids(ns).mutex);
294 static inline unsigned long
295 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
299 return copy_to_user(buf, in, sizeof(*in));
304 memset(&out, 0, sizeof(out));
306 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
308 out.msg_stime = in->msg_stime;
309 out.msg_rtime = in->msg_rtime;
310 out.msg_ctime = in->msg_ctime;
312 if (in->msg_cbytes > USHRT_MAX)
313 out.msg_cbytes = USHRT_MAX;
315 out.msg_cbytes = in->msg_cbytes;
316 out.msg_lcbytes = in->msg_cbytes;
318 if (in->msg_qnum > USHRT_MAX)
319 out.msg_qnum = USHRT_MAX;
321 out.msg_qnum = in->msg_qnum;
323 if (in->msg_qbytes > USHRT_MAX)
324 out.msg_qbytes = USHRT_MAX;
326 out.msg_qbytes = in->msg_qbytes;
327 out.msg_lqbytes = in->msg_qbytes;
329 out.msg_lspid = in->msg_lspid;
330 out.msg_lrpid = in->msg_lrpid;
332 return copy_to_user(buf, &out, sizeof(out));
340 unsigned long qbytes;
346 static inline unsigned long
347 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
352 struct msqid64_ds tbuf;
354 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
357 out->qbytes = tbuf.msg_qbytes;
358 out->uid = tbuf.msg_perm.uid;
359 out->gid = tbuf.msg_perm.gid;
360 out->mode = tbuf.msg_perm.mode;
366 struct msqid_ds tbuf_old;
368 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
371 out->uid = tbuf_old.msg_perm.uid;
372 out->gid = tbuf_old.msg_perm.gid;
373 out->mode = tbuf_old.msg_perm.mode;
375 if (tbuf_old.msg_qbytes == 0)
376 out->qbytes = tbuf_old.msg_lqbytes;
378 out->qbytes = tbuf_old.msg_qbytes;
387 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
389 struct kern_ipc_perm *ipcp;
390 struct msq_setbuf setbuf;
391 struct msg_queue *msq;
393 struct ipc_namespace *ns;
395 if (msqid < 0 || cmd < 0)
398 version = ipc_parse_version(&cmd);
399 ns = current->nsproxy->ipc_ns;
405 struct msginfo msginfo;
411 * We must not return kernel stack data.
412 * due to padding, it's not enough
413 * to set all member fields.
415 err = security_msg_queue_msgctl(NULL, cmd);
419 memset(&msginfo, 0, sizeof(msginfo));
420 msginfo.msgmni = ns->msg_ctlmni;
421 msginfo.msgmax = ns->msg_ctlmax;
422 msginfo.msgmnb = ns->msg_ctlmnb;
423 msginfo.msgssz = MSGSSZ;
424 msginfo.msgseg = MSGSEG;
425 mutex_lock(&msg_ids(ns).mutex);
426 if (cmd == MSG_INFO) {
427 msginfo.msgpool = msg_ids(ns).in_use;
428 msginfo.msgmap = atomic_read(&msg_hdrs);
429 msginfo.msgtql = atomic_read(&msg_bytes);
431 msginfo.msgmap = MSGMAP;
432 msginfo.msgpool = MSGPOOL;
433 msginfo.msgtql = MSGTQL;
435 max_id = msg_ids(ns).max_id;
436 mutex_unlock(&msg_ids(ns).mutex);
437 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
439 return (max_id < 0) ? 0 : max_id;
444 struct msqid64_ds tbuf;
449 if (cmd == MSG_STAT && msqid >= msg_ids(ns).entries->size)
452 memset(&tbuf, 0, sizeof(tbuf));
454 msq = msg_lock(ns, msqid);
458 if (cmd == MSG_STAT) {
459 success_return = msg_buildid(ns, msqid, msq->q_perm.seq);
462 if (msg_checkid(ns, msq, msqid))
467 if (ipcperms(&msq->q_perm, S_IRUGO))
470 err = security_msg_queue_msgctl(msq, cmd);
474 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
475 tbuf.msg_stime = msq->q_stime;
476 tbuf.msg_rtime = msq->q_rtime;
477 tbuf.msg_ctime = msq->q_ctime;
478 tbuf.msg_cbytes = msq->q_cbytes;
479 tbuf.msg_qnum = msq->q_qnum;
480 tbuf.msg_qbytes = msq->q_qbytes;
481 tbuf.msg_lspid = msq->q_lspid;
482 tbuf.msg_lrpid = msq->q_lrpid;
484 if (copy_msqid_to_user(buf, &tbuf, version))
486 return success_return;
491 if (copy_msqid_from_user(&setbuf, buf, version))
500 mutex_lock(&msg_ids(ns).mutex);
501 msq = msg_lock(ns, msqid);
507 if (msg_checkid(ns, msq, msqid))
511 err = audit_ipc_obj(ipcp);
515 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
522 if (current->euid != ipcp->cuid &&
523 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
524 /* We _could_ check for CAP_CHOWN above, but we don't */
527 err = security_msg_queue_msgctl(msq, cmd);
535 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
538 msq->q_qbytes = setbuf.qbytes;
540 ipcp->uid = setbuf.uid;
541 ipcp->gid = setbuf.gid;
542 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
543 (S_IRWXUGO & setbuf.mode);
544 msq->q_ctime = get_seconds();
545 /* sleeping receivers might be excluded by
546 * stricter permissions.
548 expunge_all(msq, -EAGAIN);
549 /* sleeping senders might be able to send
550 * due to a larger queue size.
552 ss_wakeup(&msq->q_senders, 0);
557 freeque(ns, msq, msqid);
562 mutex_unlock(&msg_ids(ns).mutex);
572 static int testmsg(struct msg_msg *msg, long type, int mode)
578 case SEARCH_LESSEQUAL:
579 if (msg->m_type <=type)
583 if (msg->m_type == type)
586 case SEARCH_NOTEQUAL:
587 if (msg->m_type != type)
594 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
596 struct list_head *tmp;
598 tmp = msq->q_receivers.next;
599 while (tmp != &msq->q_receivers) {
600 struct msg_receiver *msr;
602 msr = list_entry(tmp, struct msg_receiver, r_list);
604 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
605 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
606 msr->r_msgtype, msr->r_mode)) {
608 list_del(&msr->r_list);
609 if (msr->r_maxsize < msg->m_ts) {
611 wake_up_process(msr->r_tsk);
613 msr->r_msg = ERR_PTR(-E2BIG);
616 msq->q_lrpid = msr->r_tsk->pid;
617 msq->q_rtime = get_seconds();
618 wake_up_process(msr->r_tsk);
629 long do_msgsnd(int msqid, long mtype, void __user *mtext,
630 size_t msgsz, int msgflg)
632 struct msg_queue *msq;
635 struct ipc_namespace *ns;
637 ns = current->nsproxy->ipc_ns;
639 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
644 msg = load_msg(mtext, msgsz);
651 msq = msg_lock(ns, msqid);
657 if (msg_checkid(ns, msq, msqid))
658 goto out_unlock_free;
664 if (ipcperms(&msq->q_perm, S_IWUGO))
665 goto out_unlock_free;
667 err = security_msg_queue_msgsnd(msq, msg, msgflg);
669 goto out_unlock_free;
671 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
672 1 + msq->q_qnum <= msq->q_qbytes) {
676 /* queue full, wait: */
677 if (msgflg & IPC_NOWAIT) {
679 goto out_unlock_free;
686 ipc_lock_by_ptr(&msq->q_perm);
688 if (msq->q_perm.deleted) {
690 goto out_unlock_free;
694 if (signal_pending(current)) {
695 err = -ERESTARTNOHAND;
696 goto out_unlock_free;
700 msq->q_lspid = current->tgid;
701 msq->q_stime = get_seconds();
703 if (!pipelined_send(msq, msg)) {
704 /* noone is waiting for this message, enqueue it */
705 list_add_tail(&msg->m_list, &msq->q_messages);
706 msq->q_cbytes += msgsz;
708 atomic_add(msgsz, &msg_bytes);
709 atomic_inc(&msg_hdrs);
724 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
728 if (get_user(mtype, &msgp->mtype))
730 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
733 static inline int convert_mode(long *msgtyp, int msgflg)
736 * find message of correct type.
737 * msgtyp = 0 => get first.
738 * msgtyp > 0 => get first message of matching type.
739 * msgtyp < 0 => get message with least type must be < abs(msgtype).
745 return SEARCH_LESSEQUAL;
747 if (msgflg & MSG_EXCEPT)
748 return SEARCH_NOTEQUAL;
752 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
753 size_t msgsz, long msgtyp, int msgflg)
755 struct msg_queue *msq;
758 struct ipc_namespace *ns;
760 if (msqid < 0 || (long) msgsz < 0)
762 mode = convert_mode(&msgtyp, msgflg);
763 ns = current->nsproxy->ipc_ns;
765 msq = msg_lock(ns, msqid);
769 msg = ERR_PTR(-EIDRM);
770 if (msg_checkid(ns, msq, msqid))
774 struct msg_receiver msr_d;
775 struct list_head *tmp;
777 msg = ERR_PTR(-EACCES);
778 if (ipcperms(&msq->q_perm, S_IRUGO))
781 msg = ERR_PTR(-EAGAIN);
782 tmp = msq->q_messages.next;
783 while (tmp != &msq->q_messages) {
784 struct msg_msg *walk_msg;
786 walk_msg = list_entry(tmp, struct msg_msg, m_list);
787 if (testmsg(walk_msg, msgtyp, mode) &&
788 !security_msg_queue_msgrcv(msq, walk_msg, current,
792 if (mode == SEARCH_LESSEQUAL &&
793 walk_msg->m_type != 1) {
795 msgtyp = walk_msg->m_type - 1;
805 * Found a suitable message.
806 * Unlink it from the queue.
808 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
809 msg = ERR_PTR(-E2BIG);
812 list_del(&msg->m_list);
814 msq->q_rtime = get_seconds();
815 msq->q_lrpid = current->tgid;
816 msq->q_cbytes -= msg->m_ts;
817 atomic_sub(msg->m_ts, &msg_bytes);
818 atomic_dec(&msg_hdrs);
819 ss_wakeup(&msq->q_senders, 0);
823 /* No message waiting. Wait for a message */
824 if (msgflg & IPC_NOWAIT) {
825 msg = ERR_PTR(-ENOMSG);
828 list_add_tail(&msr_d.r_list, &msq->q_receivers);
829 msr_d.r_tsk = current;
830 msr_d.r_msgtype = msgtyp;
832 if (msgflg & MSG_NOERROR)
833 msr_d.r_maxsize = INT_MAX;
835 msr_d.r_maxsize = msgsz;
836 msr_d.r_msg = ERR_PTR(-EAGAIN);
837 current->state = TASK_INTERRUPTIBLE;
842 /* Lockless receive, part 1:
843 * Disable preemption. We don't hold a reference to the queue
844 * and getting a reference would defeat the idea of a lockless
845 * operation, thus the code relies on rcu to guarantee the
847 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
848 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
849 * rcu_read_lock() prevents preemption between reading r_msg
850 * and the spin_lock() inside ipc_lock_by_ptr().
854 /* Lockless receive, part 2:
855 * Wait until pipelined_send or expunge_all are outside of
856 * wake_up_process(). There is a race with exit(), see
857 * ipc/mqueue.c for the details.
859 msg = (struct msg_msg*)msr_d.r_msg;
860 while (msg == NULL) {
862 msg = (struct msg_msg *)msr_d.r_msg;
865 /* Lockless receive, part 3:
866 * If there is a message or an error then accept it without
869 if (msg != ERR_PTR(-EAGAIN)) {
874 /* Lockless receive, part 3:
875 * Acquire the queue spinlock.
877 ipc_lock_by_ptr(&msq->q_perm);
880 /* Lockless receive, part 4:
881 * Repeat test after acquiring the spinlock.
883 msg = (struct msg_msg*)msr_d.r_msg;
884 if (msg != ERR_PTR(-EAGAIN))
887 list_del(&msr_d.r_list);
888 if (signal_pending(current)) {
889 msg = ERR_PTR(-ERESTARTNOHAND);
898 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
899 *pmtype = msg->m_type;
900 if (store_msg(mtext, msg, msgsz))
908 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
909 long msgtyp, int msgflg)
913 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
917 if (put_user(mtype, &msgp->mtype))
923 #ifdef CONFIG_PROC_FS
924 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
926 struct msg_queue *msq = it;
929 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",