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_unlock(msq) ipc_unlock(&(msq)->q_perm)
77 #define msg_buildid(ns, id, seq) \
78 ipc_buildid(&msg_ids(ns), id, seq)
80 static void freeque(struct ipc_namespace *, struct msg_queue *);
81 static int newque(struct ipc_namespace *, struct ipc_params *);
83 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
86 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
88 ns->ids[IPC_MSG_IDS] = ids;
89 ns->msg_ctlmax = MSGMAX;
90 ns->msg_ctlmnb = MSGMNB;
91 ns->msg_ctlmni = MSGMNI;
95 int msg_init_ns(struct ipc_namespace *ns)
99 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
103 __msg_init_ns(ns, ids);
107 void msg_exit_ns(struct ipc_namespace *ns)
109 struct msg_queue *msq;
113 mutex_lock(&msg_ids(ns).mutex);
115 in_use = msg_ids(ns).in_use;
117 for (total = 0, next_id = 0; total < in_use; next_id++) {
118 msq = idr_find(&msg_ids(ns).ipcs_idr, next_id);
121 ipc_lock_by_ptr(&msq->q_perm);
125 mutex_unlock(&msg_ids(ns).mutex);
127 kfree(ns->ids[IPC_MSG_IDS]);
128 ns->ids[IPC_MSG_IDS] = NULL;
131 void __init msg_init(void)
133 __msg_init_ns(&init_ipc_ns, &init_msg_ids);
134 ipc_init_proc_interface("sysvipc/msg",
135 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
136 IPC_MSG_IDS, sysvipc_msg_proc_show);
139 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
141 return (struct msg_queue *) ipc_lock(&msg_ids(ns), id);
144 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
147 return (struct msg_queue *) ipc_lock_check(&msg_ids(ns), id);
150 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
152 ipc_rmid(&msg_ids(ns), &s->q_perm);
155 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
157 struct msg_queue *msq;
159 key_t key = params->key;
160 int msgflg = params->flg;
162 msq = ipc_rcu_alloc(sizeof(*msq));
166 msq->q_perm.mode = msgflg & S_IRWXUGO;
167 msq->q_perm.key = key;
169 msq->q_perm.security = NULL;
170 retval = security_msg_queue_alloc(msq);
177 * ipc_addid() locks msq
179 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
181 security_msg_queue_free(msq);
186 msq->q_perm.id = msg_buildid(ns, id, msq->q_perm.seq);
187 msq->q_stime = msq->q_rtime = 0;
188 msq->q_ctime = get_seconds();
189 msq->q_cbytes = msq->q_qnum = 0;
190 msq->q_qbytes = ns->msg_ctlmnb;
191 msq->q_lspid = msq->q_lrpid = 0;
192 INIT_LIST_HEAD(&msq->q_messages);
193 INIT_LIST_HEAD(&msq->q_receivers);
194 INIT_LIST_HEAD(&msq->q_senders);
198 return msq->q_perm.id;
201 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
204 current->state = TASK_INTERRUPTIBLE;
205 list_add_tail(&mss->list, &msq->q_senders);
208 static inline void ss_del(struct msg_sender *mss)
210 if (mss->list.next != NULL)
211 list_del(&mss->list);
214 static void ss_wakeup(struct list_head *h, int kill)
216 struct list_head *tmp;
220 struct msg_sender *mss;
222 mss = list_entry(tmp, struct msg_sender, list);
225 mss->list.next = NULL;
226 wake_up_process(mss->tsk);
230 static void expunge_all(struct msg_queue *msq, int res)
232 struct list_head *tmp;
234 tmp = msq->q_receivers.next;
235 while (tmp != &msq->q_receivers) {
236 struct msg_receiver *msr;
238 msr = list_entry(tmp, struct msg_receiver, r_list);
241 wake_up_process(msr->r_tsk);
243 msr->r_msg = ERR_PTR(res);
248 * freeque() wakes up waiters on the sender and receiver waiting queue,
249 * removes the message queue from message queue ID
250 * IDR, and cleans up all the messages associated with this queue.
252 * msg_ids.mutex and the spinlock for this message queue are held
253 * before freeque() is called. msg_ids.mutex remains locked on exit.
255 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
257 struct list_head *tmp;
259 expunge_all(msq, -EIDRM);
260 ss_wakeup(&msq->q_senders, 1);
264 tmp = msq->q_messages.next;
265 while (tmp != &msq->q_messages) {
266 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
269 atomic_dec(&msg_hdrs);
272 atomic_sub(msq->q_cbytes, &msg_bytes);
273 security_msg_queue_free(msq);
277 static inline int msg_security(void *msq, int msgflg)
279 return security_msg_queue_associate((struct msg_queue *) msq, msgflg);
282 asmlinkage long sys_msgget(key_t key, int msgflg)
284 struct ipc_namespace *ns;
285 struct ipc_ops msg_ops;
286 struct ipc_params msg_params;
288 ns = current->nsproxy->ipc_ns;
290 msg_ops.getnew = newque;
291 msg_ops.associate = msg_security;
292 msg_ops.more_checks = NULL;
294 msg_params.key = key;
295 msg_params.flg = msgflg;
297 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
300 static inline unsigned long
301 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
305 return copy_to_user(buf, in, sizeof(*in));
310 memset(&out, 0, sizeof(out));
312 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
314 out.msg_stime = in->msg_stime;
315 out.msg_rtime = in->msg_rtime;
316 out.msg_ctime = in->msg_ctime;
318 if (in->msg_cbytes > USHRT_MAX)
319 out.msg_cbytes = USHRT_MAX;
321 out.msg_cbytes = in->msg_cbytes;
322 out.msg_lcbytes = in->msg_cbytes;
324 if (in->msg_qnum > USHRT_MAX)
325 out.msg_qnum = USHRT_MAX;
327 out.msg_qnum = in->msg_qnum;
329 if (in->msg_qbytes > USHRT_MAX)
330 out.msg_qbytes = USHRT_MAX;
332 out.msg_qbytes = in->msg_qbytes;
333 out.msg_lqbytes = in->msg_qbytes;
335 out.msg_lspid = in->msg_lspid;
336 out.msg_lrpid = in->msg_lrpid;
338 return copy_to_user(buf, &out, sizeof(out));
346 unsigned long qbytes;
352 static inline unsigned long
353 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
358 struct msqid64_ds tbuf;
360 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
363 out->qbytes = tbuf.msg_qbytes;
364 out->uid = tbuf.msg_perm.uid;
365 out->gid = tbuf.msg_perm.gid;
366 out->mode = tbuf.msg_perm.mode;
372 struct msqid_ds tbuf_old;
374 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
377 out->uid = tbuf_old.msg_perm.uid;
378 out->gid = tbuf_old.msg_perm.gid;
379 out->mode = tbuf_old.msg_perm.mode;
381 if (tbuf_old.msg_qbytes == 0)
382 out->qbytes = tbuf_old.msg_lqbytes;
384 out->qbytes = tbuf_old.msg_qbytes;
393 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
395 struct kern_ipc_perm *ipcp;
396 struct msq_setbuf uninitialized_var(setbuf);
397 struct msg_queue *msq;
399 struct ipc_namespace *ns;
401 if (msqid < 0 || cmd < 0)
404 version = ipc_parse_version(&cmd);
405 ns = current->nsproxy->ipc_ns;
411 struct msginfo msginfo;
417 * We must not return kernel stack data.
418 * due to padding, it's not enough
419 * to set all member fields.
421 err = security_msg_queue_msgctl(NULL, cmd);
425 memset(&msginfo, 0, sizeof(msginfo));
426 msginfo.msgmni = ns->msg_ctlmni;
427 msginfo.msgmax = ns->msg_ctlmax;
428 msginfo.msgmnb = ns->msg_ctlmnb;
429 msginfo.msgssz = MSGSSZ;
430 msginfo.msgseg = MSGSEG;
431 mutex_lock(&msg_ids(ns).mutex);
432 if (cmd == MSG_INFO) {
433 msginfo.msgpool = msg_ids(ns).in_use;
434 msginfo.msgmap = atomic_read(&msg_hdrs);
435 msginfo.msgtql = atomic_read(&msg_bytes);
437 msginfo.msgmap = MSGMAP;
438 msginfo.msgpool = MSGPOOL;
439 msginfo.msgtql = MSGTQL;
441 max_id = ipc_get_maxid(&msg_ids(ns));
442 mutex_unlock(&msg_ids(ns).mutex);
443 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
445 return (max_id < 0) ? 0 : max_id;
447 case MSG_STAT: /* msqid is an index rather than a msg queue id */
450 struct msqid64_ds tbuf;
456 if (cmd == MSG_STAT) {
457 msq = msg_lock(ns, msqid);
460 success_return = msq->q_perm.id;
462 msq = msg_lock_check(ns, msqid);
468 if (ipcperms(&msq->q_perm, S_IRUGO))
471 err = security_msg_queue_msgctl(msq, cmd);
475 memset(&tbuf, 0, sizeof(tbuf));
477 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
478 tbuf.msg_stime = msq->q_stime;
479 tbuf.msg_rtime = msq->q_rtime;
480 tbuf.msg_ctime = msq->q_ctime;
481 tbuf.msg_cbytes = msq->q_cbytes;
482 tbuf.msg_qnum = msq->q_qnum;
483 tbuf.msg_qbytes = msq->q_qbytes;
484 tbuf.msg_lspid = msq->q_lspid;
485 tbuf.msg_lrpid = msq->q_lrpid;
487 if (copy_msqid_to_user(buf, &tbuf, version))
489 return success_return;
494 if (copy_msqid_from_user(&setbuf, buf, version))
503 mutex_lock(&msg_ids(ns).mutex);
504 msq = msg_lock_check(ns, msqid);
512 err = audit_ipc_obj(ipcp);
515 if (cmd == IPC_SET) {
516 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
523 if (current->euid != ipcp->cuid &&
524 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
525 /* We _could_ check for CAP_CHOWN above, but we don't */
528 err = security_msg_queue_msgctl(msq, cmd);
536 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
539 msq->q_qbytes = setbuf.qbytes;
541 ipcp->uid = setbuf.uid;
542 ipcp->gid = setbuf.gid;
543 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
544 (S_IRWXUGO & setbuf.mode);
545 msq->q_ctime = get_seconds();
546 /* sleeping receivers might be excluded by
547 * stricter permissions.
549 expunge_all(msq, -EAGAIN);
550 /* sleeping senders might be able to send
551 * due to a larger queue size.
553 ss_wakeup(&msq->q_senders, 0);
563 mutex_unlock(&msg_ids(ns).mutex);
573 static int testmsg(struct msg_msg *msg, long type, int mode)
579 case SEARCH_LESSEQUAL:
580 if (msg->m_type <=type)
584 if (msg->m_type == type)
587 case SEARCH_NOTEQUAL:
588 if (msg->m_type != type)
595 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
597 struct list_head *tmp;
599 tmp = msq->q_receivers.next;
600 while (tmp != &msq->q_receivers) {
601 struct msg_receiver *msr;
603 msr = list_entry(tmp, struct msg_receiver, r_list);
605 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
606 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
607 msr->r_msgtype, msr->r_mode)) {
609 list_del(&msr->r_list);
610 if (msr->r_maxsize < msg->m_ts) {
612 wake_up_process(msr->r_tsk);
614 msr->r_msg = ERR_PTR(-E2BIG);
617 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
618 msq->q_rtime = get_seconds();
619 wake_up_process(msr->r_tsk);
630 long do_msgsnd(int msqid, long mtype, void __user *mtext,
631 size_t msgsz, int msgflg)
633 struct msg_queue *msq;
636 struct ipc_namespace *ns;
638 ns = current->nsproxy->ipc_ns;
640 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
645 msg = load_msg(mtext, msgsz);
652 msq = msg_lock_check(ns, msqid);
662 if (ipcperms(&msq->q_perm, S_IWUGO))
663 goto out_unlock_free;
665 err = security_msg_queue_msgsnd(msq, msg, msgflg);
667 goto out_unlock_free;
669 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
670 1 + msq->q_qnum <= msq->q_qbytes) {
674 /* queue full, wait: */
675 if (msgflg & IPC_NOWAIT) {
677 goto out_unlock_free;
684 ipc_lock_by_ptr(&msq->q_perm);
686 if (msq->q_perm.deleted) {
688 goto out_unlock_free;
692 if (signal_pending(current)) {
693 err = -ERESTARTNOHAND;
694 goto out_unlock_free;
698 msq->q_lspid = task_tgid_vnr(current);
699 msq->q_stime = get_seconds();
701 if (!pipelined_send(msq, msg)) {
702 /* noone is waiting for this message, enqueue it */
703 list_add_tail(&msg->m_list, &msq->q_messages);
704 msq->q_cbytes += msgsz;
706 atomic_add(msgsz, &msg_bytes);
707 atomic_inc(&msg_hdrs);
722 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
726 if (get_user(mtype, &msgp->mtype))
728 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
731 static inline int convert_mode(long *msgtyp, int msgflg)
734 * find message of correct type.
735 * msgtyp = 0 => get first.
736 * msgtyp > 0 => get first message of matching type.
737 * msgtyp < 0 => get message with least type must be < abs(msgtype).
743 return SEARCH_LESSEQUAL;
745 if (msgflg & MSG_EXCEPT)
746 return SEARCH_NOTEQUAL;
750 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
751 size_t msgsz, long msgtyp, int msgflg)
753 struct msg_queue *msq;
756 struct ipc_namespace *ns;
758 if (msqid < 0 || (long) msgsz < 0)
760 mode = convert_mode(&msgtyp, msgflg);
761 ns = current->nsproxy->ipc_ns;
763 msq = msg_lock_check(ns, msqid);
768 struct msg_receiver msr_d;
769 struct list_head *tmp;
771 msg = ERR_PTR(-EACCES);
772 if (ipcperms(&msq->q_perm, S_IRUGO))
775 msg = ERR_PTR(-EAGAIN);
776 tmp = msq->q_messages.next;
777 while (tmp != &msq->q_messages) {
778 struct msg_msg *walk_msg;
780 walk_msg = list_entry(tmp, struct msg_msg, m_list);
781 if (testmsg(walk_msg, msgtyp, mode) &&
782 !security_msg_queue_msgrcv(msq, walk_msg, current,
786 if (mode == SEARCH_LESSEQUAL &&
787 walk_msg->m_type != 1) {
789 msgtyp = walk_msg->m_type - 1;
799 * Found a suitable message.
800 * Unlink it from the queue.
802 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
803 msg = ERR_PTR(-E2BIG);
806 list_del(&msg->m_list);
808 msq->q_rtime = get_seconds();
809 msq->q_lrpid = task_tgid_vnr(current);
810 msq->q_cbytes -= msg->m_ts;
811 atomic_sub(msg->m_ts, &msg_bytes);
812 atomic_dec(&msg_hdrs);
813 ss_wakeup(&msq->q_senders, 0);
817 /* No message waiting. Wait for a message */
818 if (msgflg & IPC_NOWAIT) {
819 msg = ERR_PTR(-ENOMSG);
822 list_add_tail(&msr_d.r_list, &msq->q_receivers);
823 msr_d.r_tsk = current;
824 msr_d.r_msgtype = msgtyp;
826 if (msgflg & MSG_NOERROR)
827 msr_d.r_maxsize = INT_MAX;
829 msr_d.r_maxsize = msgsz;
830 msr_d.r_msg = ERR_PTR(-EAGAIN);
831 current->state = TASK_INTERRUPTIBLE;
836 /* Lockless receive, part 1:
837 * Disable preemption. We don't hold a reference to the queue
838 * and getting a reference would defeat the idea of a lockless
839 * operation, thus the code relies on rcu to guarantee the
841 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
842 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
843 * rcu_read_lock() prevents preemption between reading r_msg
844 * and the spin_lock() inside ipc_lock_by_ptr().
848 /* Lockless receive, part 2:
849 * Wait until pipelined_send or expunge_all are outside of
850 * wake_up_process(). There is a race with exit(), see
851 * ipc/mqueue.c for the details.
853 msg = (struct msg_msg*)msr_d.r_msg;
854 while (msg == NULL) {
856 msg = (struct msg_msg *)msr_d.r_msg;
859 /* Lockless receive, part 3:
860 * If there is a message or an error then accept it without
863 if (msg != ERR_PTR(-EAGAIN)) {
868 /* Lockless receive, part 3:
869 * Acquire the queue spinlock.
871 ipc_lock_by_ptr(&msq->q_perm);
874 /* Lockless receive, part 4:
875 * Repeat test after acquiring the spinlock.
877 msg = (struct msg_msg*)msr_d.r_msg;
878 if (msg != ERR_PTR(-EAGAIN))
881 list_del(&msr_d.r_list);
882 if (signal_pending(current)) {
883 msg = ERR_PTR(-ERESTARTNOHAND);
892 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
893 *pmtype = msg->m_type;
894 if (store_msg(mtext, msg, msgsz))
902 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
903 long msgtyp, int msgflg)
907 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
911 if (put_user(mtype, &msgp->mtype))
917 #ifdef CONFIG_PROC_FS
918 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
920 struct msg_queue *msq = it;
923 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",