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/rwsem.h>
38 #include <linux/nsproxy.h>
39 #include <linux/ipc_namespace.h>
41 #include <asm/current.h>
42 #include <asm/uaccess.h>
46 * one msg_receiver structure for each sleeping receiver:
49 struct list_head r_list;
50 struct task_struct *r_tsk;
56 struct msg_msg *volatile 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 struct ipc_ids init_msg_ids;
72 #define msg_ids(ns) (*((ns)->ids[IPC_MSG_IDS]))
74 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
75 #define msg_buildid(id, seq) ipc_buildid(id, seq)
77 static void freeque(struct ipc_namespace *, struct msg_queue *);
78 static int newque(struct ipc_namespace *, struct ipc_params *);
80 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
83 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
85 ns->ids[IPC_MSG_IDS] = ids;
86 ns->msg_ctlmax = MSGMAX;
87 ns->msg_ctlmnb = MSGMNB;
88 ns->msg_ctlmni = MSGMNI;
89 atomic_set(&ns->msg_bytes, 0);
90 atomic_set(&ns->msg_hdrs, 0);
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;
110 struct kern_ipc_perm *perm;
114 down_write(&msg_ids(ns).rw_mutex);
116 in_use = msg_ids(ns).in_use;
118 for (total = 0, next_id = 0; total < in_use; next_id++) {
119 perm = idr_find(&msg_ids(ns).ipcs_idr, next_id);
122 ipc_lock_by_ptr(perm);
123 msq = container_of(perm, struct msg_queue, q_perm);
128 up_write(&msg_ids(ns).rw_mutex);
130 kfree(ns->ids[IPC_MSG_IDS]);
131 ns->ids[IPC_MSG_IDS] = NULL;
135 void __init msg_init(void)
137 __msg_init_ns(&init_ipc_ns, &init_msg_ids);
138 ipc_init_proc_interface("sysvipc/msg",
139 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
140 IPC_MSG_IDS, sysvipc_msg_proc_show);
144 * This routine is called in the paths where the rw_mutex is held to protect
145 * access to the idr tree.
147 static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
150 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
153 return (struct msg_queue *)ipcp;
155 return container_of(ipcp, struct msg_queue, q_perm);
159 * msg_lock_(check_) routines are called in the paths where the rw_mutex
162 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
164 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
167 return (struct msg_queue *)ipcp;
169 return container_of(ipcp, struct msg_queue, q_perm);
172 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
175 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
178 return (struct msg_queue *)ipcp;
180 return container_of(ipcp, struct msg_queue, q_perm);
183 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
185 ipc_rmid(&msg_ids(ns), &s->q_perm);
189 * newque - Create a new msg queue
191 * @params: ptr to the structure that contains the key and msgflg
193 * Called with msg_ids.rw_mutex held (writer)
195 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
197 struct msg_queue *msq;
199 key_t key = params->key;
200 int msgflg = params->flg;
202 msq = ipc_rcu_alloc(sizeof(*msq));
206 msq->q_perm.mode = msgflg & S_IRWXUGO;
207 msq->q_perm.key = key;
209 msq->q_perm.security = NULL;
210 retval = security_msg_queue_alloc(msq);
217 * ipc_addid() locks msq
219 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
221 security_msg_queue_free(msq);
226 msq->q_perm.id = msg_buildid(id, msq->q_perm.seq);
227 msq->q_stime = msq->q_rtime = 0;
228 msq->q_ctime = get_seconds();
229 msq->q_cbytes = msq->q_qnum = 0;
230 msq->q_qbytes = ns->msg_ctlmnb;
231 msq->q_lspid = msq->q_lrpid = 0;
232 INIT_LIST_HEAD(&msq->q_messages);
233 INIT_LIST_HEAD(&msq->q_receivers);
234 INIT_LIST_HEAD(&msq->q_senders);
238 return msq->q_perm.id;
241 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
244 current->state = TASK_INTERRUPTIBLE;
245 list_add_tail(&mss->list, &msq->q_senders);
248 static inline void ss_del(struct msg_sender *mss)
250 if (mss->list.next != NULL)
251 list_del(&mss->list);
254 static void ss_wakeup(struct list_head *h, int kill)
256 struct list_head *tmp;
260 struct msg_sender *mss;
262 mss = list_entry(tmp, struct msg_sender, list);
265 mss->list.next = NULL;
266 wake_up_process(mss->tsk);
270 static void expunge_all(struct msg_queue *msq, int res)
272 struct list_head *tmp;
274 tmp = msq->q_receivers.next;
275 while (tmp != &msq->q_receivers) {
276 struct msg_receiver *msr;
278 msr = list_entry(tmp, struct msg_receiver, r_list);
281 wake_up_process(msr->r_tsk);
283 msr->r_msg = ERR_PTR(res);
288 * freeque() wakes up waiters on the sender and receiver waiting queue,
289 * removes the message queue from message queue ID IDR, and cleans up all the
290 * messages associated with this queue.
292 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
293 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
295 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
297 struct list_head *tmp;
299 expunge_all(msq, -EIDRM);
300 ss_wakeup(&msq->q_senders, 1);
304 tmp = msq->q_messages.next;
305 while (tmp != &msq->q_messages) {
306 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
309 atomic_dec(&ns->msg_hdrs);
312 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
313 security_msg_queue_free(msq);
318 * Called with msg_ids.rw_mutex and ipcp locked.
320 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
322 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
324 return security_msg_queue_associate(msq, msgflg);
327 asmlinkage long sys_msgget(key_t key, int msgflg)
329 struct ipc_namespace *ns;
330 struct ipc_ops msg_ops;
331 struct ipc_params msg_params;
333 ns = current->nsproxy->ipc_ns;
335 msg_ops.getnew = newque;
336 msg_ops.associate = msg_security;
337 msg_ops.more_checks = NULL;
339 msg_params.key = key;
340 msg_params.flg = msgflg;
342 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
345 static inline unsigned long
346 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
350 return copy_to_user(buf, in, sizeof(*in));
355 memset(&out, 0, sizeof(out));
357 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
359 out.msg_stime = in->msg_stime;
360 out.msg_rtime = in->msg_rtime;
361 out.msg_ctime = in->msg_ctime;
363 if (in->msg_cbytes > USHRT_MAX)
364 out.msg_cbytes = USHRT_MAX;
366 out.msg_cbytes = in->msg_cbytes;
367 out.msg_lcbytes = in->msg_cbytes;
369 if (in->msg_qnum > USHRT_MAX)
370 out.msg_qnum = USHRT_MAX;
372 out.msg_qnum = in->msg_qnum;
374 if (in->msg_qbytes > USHRT_MAX)
375 out.msg_qbytes = USHRT_MAX;
377 out.msg_qbytes = in->msg_qbytes;
378 out.msg_lqbytes = in->msg_qbytes;
380 out.msg_lspid = in->msg_lspid;
381 out.msg_lrpid = in->msg_lrpid;
383 return copy_to_user(buf, &out, sizeof(out));
391 unsigned long qbytes;
397 static inline unsigned long
398 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
403 struct msqid64_ds tbuf;
405 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
408 out->qbytes = tbuf.msg_qbytes;
409 out->uid = tbuf.msg_perm.uid;
410 out->gid = tbuf.msg_perm.gid;
411 out->mode = tbuf.msg_perm.mode;
417 struct msqid_ds tbuf_old;
419 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
422 out->uid = tbuf_old.msg_perm.uid;
423 out->gid = tbuf_old.msg_perm.gid;
424 out->mode = tbuf_old.msg_perm.mode;
426 if (tbuf_old.msg_qbytes == 0)
427 out->qbytes = tbuf_old.msg_lqbytes;
429 out->qbytes = tbuf_old.msg_qbytes;
438 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
440 struct kern_ipc_perm *ipcp;
441 struct msq_setbuf uninitialized_var(setbuf);
442 struct msg_queue *msq;
444 struct ipc_namespace *ns;
446 if (msqid < 0 || cmd < 0)
449 version = ipc_parse_version(&cmd);
450 ns = current->nsproxy->ipc_ns;
456 struct msginfo msginfo;
462 * We must not return kernel stack data.
463 * due to padding, it's not enough
464 * to set all member fields.
466 err = security_msg_queue_msgctl(NULL, cmd);
470 memset(&msginfo, 0, sizeof(msginfo));
471 msginfo.msgmni = ns->msg_ctlmni;
472 msginfo.msgmax = ns->msg_ctlmax;
473 msginfo.msgmnb = ns->msg_ctlmnb;
474 msginfo.msgssz = MSGSSZ;
475 msginfo.msgseg = MSGSEG;
476 down_read(&msg_ids(ns).rw_mutex);
477 if (cmd == MSG_INFO) {
478 msginfo.msgpool = msg_ids(ns).in_use;
479 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
480 msginfo.msgtql = atomic_read(&ns->msg_bytes);
482 msginfo.msgmap = MSGMAP;
483 msginfo.msgpool = MSGPOOL;
484 msginfo.msgtql = MSGTQL;
486 max_id = ipc_get_maxid(&msg_ids(ns));
487 up_read(&msg_ids(ns).rw_mutex);
488 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
490 return (max_id < 0) ? 0 : max_id;
492 case MSG_STAT: /* msqid is an index rather than a msg queue id */
495 struct msqid64_ds tbuf;
501 if (cmd == MSG_STAT) {
502 msq = msg_lock(ns, msqid);
505 success_return = msq->q_perm.id;
507 msq = msg_lock_check(ns, msqid);
513 if (ipcperms(&msq->q_perm, S_IRUGO))
516 err = security_msg_queue_msgctl(msq, cmd);
520 memset(&tbuf, 0, sizeof(tbuf));
522 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
523 tbuf.msg_stime = msq->q_stime;
524 tbuf.msg_rtime = msq->q_rtime;
525 tbuf.msg_ctime = msq->q_ctime;
526 tbuf.msg_cbytes = msq->q_cbytes;
527 tbuf.msg_qnum = msq->q_qnum;
528 tbuf.msg_qbytes = msq->q_qbytes;
529 tbuf.msg_lspid = msq->q_lspid;
530 tbuf.msg_lrpid = msq->q_lrpid;
532 if (copy_msqid_to_user(buf, &tbuf, version))
534 return success_return;
539 if (copy_msqid_from_user(&setbuf, buf, version))
548 down_write(&msg_ids(ns).rw_mutex);
549 msq = msg_lock_check_down(ns, msqid);
557 err = audit_ipc_obj(ipcp);
560 if (cmd == IPC_SET) {
561 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
568 if (current->euid != ipcp->cuid &&
569 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
570 /* We _could_ check for CAP_CHOWN above, but we don't */
573 err = security_msg_queue_msgctl(msq, cmd);
581 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
584 msq->q_qbytes = setbuf.qbytes;
586 ipcp->uid = setbuf.uid;
587 ipcp->gid = setbuf.gid;
588 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
589 (S_IRWXUGO & setbuf.mode);
590 msq->q_ctime = get_seconds();
591 /* sleeping receivers might be excluded by
592 * stricter permissions.
594 expunge_all(msq, -EAGAIN);
595 /* sleeping senders might be able to send
596 * due to a larger queue size.
598 ss_wakeup(&msq->q_senders, 0);
608 up_write(&msg_ids(ns).rw_mutex);
618 static int testmsg(struct msg_msg *msg, long type, int mode)
624 case SEARCH_LESSEQUAL:
625 if (msg->m_type <=type)
629 if (msg->m_type == type)
632 case SEARCH_NOTEQUAL:
633 if (msg->m_type != type)
640 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
642 struct list_head *tmp;
644 tmp = msq->q_receivers.next;
645 while (tmp != &msq->q_receivers) {
646 struct msg_receiver *msr;
648 msr = list_entry(tmp, struct msg_receiver, r_list);
650 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
651 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
652 msr->r_msgtype, msr->r_mode)) {
654 list_del(&msr->r_list);
655 if (msr->r_maxsize < msg->m_ts) {
657 wake_up_process(msr->r_tsk);
659 msr->r_msg = ERR_PTR(-E2BIG);
662 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
663 msq->q_rtime = get_seconds();
664 wake_up_process(msr->r_tsk);
675 long do_msgsnd(int msqid, long mtype, void __user *mtext,
676 size_t msgsz, int msgflg)
678 struct msg_queue *msq;
681 struct ipc_namespace *ns;
683 ns = current->nsproxy->ipc_ns;
685 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
690 msg = load_msg(mtext, msgsz);
697 msq = msg_lock_check(ns, msqid);
707 if (ipcperms(&msq->q_perm, S_IWUGO))
708 goto out_unlock_free;
710 err = security_msg_queue_msgsnd(msq, msg, msgflg);
712 goto out_unlock_free;
714 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
715 1 + msq->q_qnum <= msq->q_qbytes) {
719 /* queue full, wait: */
720 if (msgflg & IPC_NOWAIT) {
722 goto out_unlock_free;
729 ipc_lock_by_ptr(&msq->q_perm);
731 if (msq->q_perm.deleted) {
733 goto out_unlock_free;
737 if (signal_pending(current)) {
738 err = -ERESTARTNOHAND;
739 goto out_unlock_free;
743 msq->q_lspid = task_tgid_vnr(current);
744 msq->q_stime = get_seconds();
746 if (!pipelined_send(msq, msg)) {
747 /* noone is waiting for this message, enqueue it */
748 list_add_tail(&msg->m_list, &msq->q_messages);
749 msq->q_cbytes += msgsz;
751 atomic_add(msgsz, &ns->msg_bytes);
752 atomic_inc(&ns->msg_hdrs);
767 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
771 if (get_user(mtype, &msgp->mtype))
773 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
776 static inline int convert_mode(long *msgtyp, int msgflg)
779 * find message of correct type.
780 * msgtyp = 0 => get first.
781 * msgtyp > 0 => get first message of matching type.
782 * msgtyp < 0 => get message with least type must be < abs(msgtype).
788 return SEARCH_LESSEQUAL;
790 if (msgflg & MSG_EXCEPT)
791 return SEARCH_NOTEQUAL;
795 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
796 size_t msgsz, long msgtyp, int msgflg)
798 struct msg_queue *msq;
801 struct ipc_namespace *ns;
803 if (msqid < 0 || (long) msgsz < 0)
805 mode = convert_mode(&msgtyp, msgflg);
806 ns = current->nsproxy->ipc_ns;
808 msq = msg_lock_check(ns, msqid);
813 struct msg_receiver msr_d;
814 struct list_head *tmp;
816 msg = ERR_PTR(-EACCES);
817 if (ipcperms(&msq->q_perm, S_IRUGO))
820 msg = ERR_PTR(-EAGAIN);
821 tmp = msq->q_messages.next;
822 while (tmp != &msq->q_messages) {
823 struct msg_msg *walk_msg;
825 walk_msg = list_entry(tmp, struct msg_msg, m_list);
826 if (testmsg(walk_msg, msgtyp, mode) &&
827 !security_msg_queue_msgrcv(msq, walk_msg, current,
831 if (mode == SEARCH_LESSEQUAL &&
832 walk_msg->m_type != 1) {
834 msgtyp = walk_msg->m_type - 1;
844 * Found a suitable message.
845 * Unlink it from the queue.
847 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
848 msg = ERR_PTR(-E2BIG);
851 list_del(&msg->m_list);
853 msq->q_rtime = get_seconds();
854 msq->q_lrpid = task_tgid_vnr(current);
855 msq->q_cbytes -= msg->m_ts;
856 atomic_sub(msg->m_ts, &ns->msg_bytes);
857 atomic_dec(&ns->msg_hdrs);
858 ss_wakeup(&msq->q_senders, 0);
862 /* No message waiting. Wait for a message */
863 if (msgflg & IPC_NOWAIT) {
864 msg = ERR_PTR(-ENOMSG);
867 list_add_tail(&msr_d.r_list, &msq->q_receivers);
868 msr_d.r_tsk = current;
869 msr_d.r_msgtype = msgtyp;
871 if (msgflg & MSG_NOERROR)
872 msr_d.r_maxsize = INT_MAX;
874 msr_d.r_maxsize = msgsz;
875 msr_d.r_msg = ERR_PTR(-EAGAIN);
876 current->state = TASK_INTERRUPTIBLE;
881 /* Lockless receive, part 1:
882 * Disable preemption. We don't hold a reference to the queue
883 * and getting a reference would defeat the idea of a lockless
884 * operation, thus the code relies on rcu to guarantee the
886 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
887 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
888 * rcu_read_lock() prevents preemption between reading r_msg
889 * and the spin_lock() inside ipc_lock_by_ptr().
893 /* Lockless receive, part 2:
894 * Wait until pipelined_send or expunge_all are outside of
895 * wake_up_process(). There is a race with exit(), see
896 * ipc/mqueue.c for the details.
898 msg = (struct msg_msg*)msr_d.r_msg;
899 while (msg == NULL) {
901 msg = (struct msg_msg *)msr_d.r_msg;
904 /* Lockless receive, part 3:
905 * If there is a message or an error then accept it without
908 if (msg != ERR_PTR(-EAGAIN)) {
913 /* Lockless receive, part 3:
914 * Acquire the queue spinlock.
916 ipc_lock_by_ptr(&msq->q_perm);
919 /* Lockless receive, part 4:
920 * Repeat test after acquiring the spinlock.
922 msg = (struct msg_msg*)msr_d.r_msg;
923 if (msg != ERR_PTR(-EAGAIN))
926 list_del(&msr_d.r_list);
927 if (signal_pending(current)) {
928 msg = ERR_PTR(-ERESTARTNOHAND);
937 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
938 *pmtype = msg->m_type;
939 if (store_msg(mtext, msg, msgsz))
947 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
948 long msgtyp, int msgflg)
952 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
956 if (put_user(mtype, &msgp->mtype))
962 #ifdef CONFIG_PROC_FS
963 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
965 struct msg_queue *msq = it;
968 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",