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>
39 #include <linux/vs_base.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 atomic_t msg_bytes = ATOMIC_INIT(0);
71 static atomic_t msg_hdrs = ATOMIC_INIT(0);
73 static struct ipc_ids init_msg_ids;
75 #define msg_ids(ns) (*((ns)->ids[IPC_MSG_IDS]))
77 #define msg_lock(ns, id) ((struct msg_queue*)ipc_lock(&msg_ids(ns), id))
78 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
79 #define msg_rmid(ns, id) ((struct msg_queue*)ipc_rmid(&msg_ids(ns), id))
80 #define msg_checkid(ns, msq, msgid) \
81 ipc_checkid(&msg_ids(ns), &msq->q_perm, msgid)
82 #define msg_buildid(ns, id, seq) \
83 ipc_buildid(&msg_ids(ns), id, seq)
85 static void freeque (struct ipc_namespace *ns, struct msg_queue *msq, int id);
86 static int newque (struct ipc_namespace *ns, key_t key, int msgflg);
88 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
91 static void __ipc_init __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
93 ns->ids[IPC_MSG_IDS] = ids;
94 ns->msg_ctlmax = MSGMAX;
95 ns->msg_ctlmnb = MSGMNB;
96 ns->msg_ctlmni = MSGMNI;
97 ipc_init_ids(ids, ns->msg_ctlmni);
101 int msg_init_ns(struct ipc_namespace *ns)
105 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
109 __msg_init_ns(ns, ids);
113 void msg_exit_ns(struct ipc_namespace *ns)
116 struct msg_queue *msq;
118 mutex_lock(&msg_ids(ns).mutex);
119 for (i = 0; i <= msg_ids(ns).max_id; i++) {
120 msq = msg_lock(ns, i);
126 mutex_unlock(&msg_ids(ns).mutex);
128 ipc_fini_ids(ns->ids[IPC_MSG_IDS]);
129 kfree(ns->ids[IPC_MSG_IDS]);
130 ns->ids[IPC_MSG_IDS] = NULL;
134 void __init msg_init(void)
136 __msg_init_ns(&init_ipc_ns, &init_msg_ids);
137 ipc_init_proc_interface("sysvipc/msg",
138 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
139 IPC_MSG_IDS, sysvipc_msg_proc_show);
142 static int newque (struct ipc_namespace *ns, key_t key, int msgflg)
144 struct msg_queue *msq;
147 msq = ipc_rcu_alloc(sizeof(*msq));
151 msq->q_perm.mode = msgflg & S_IRWXUGO;
152 msq->q_perm.key = key;
153 msq->q_perm.xid = vx_current_xid();
155 msq->q_perm.security = NULL;
156 retval = security_msg_queue_alloc(msq);
162 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
164 security_msg_queue_free(msq);
169 msq->q_id = msg_buildid(ns, id, msq->q_perm.seq);
170 msq->q_stime = msq->q_rtime = 0;
171 msq->q_ctime = get_seconds();
172 msq->q_cbytes = msq->q_qnum = 0;
173 msq->q_qbytes = ns->msg_ctlmnb;
174 msq->q_lspid = msq->q_lrpid = 0;
175 INIT_LIST_HEAD(&msq->q_messages);
176 INIT_LIST_HEAD(&msq->q_receivers);
177 INIT_LIST_HEAD(&msq->q_senders);
183 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
186 current->state = TASK_INTERRUPTIBLE;
187 list_add_tail(&mss->list, &msq->q_senders);
190 static inline void ss_del(struct msg_sender *mss)
192 if (mss->list.next != NULL)
193 list_del(&mss->list);
196 static void ss_wakeup(struct list_head *h, int kill)
198 struct list_head *tmp;
202 struct msg_sender *mss;
204 mss = list_entry(tmp, struct msg_sender, list);
207 mss->list.next = NULL;
208 wake_up_process(mss->tsk);
212 static void expunge_all(struct msg_queue *msq, int res)
214 struct list_head *tmp;
216 tmp = msq->q_receivers.next;
217 while (tmp != &msq->q_receivers) {
218 struct msg_receiver *msr;
220 msr = list_entry(tmp, struct msg_receiver, r_list);
223 wake_up_process(msr->r_tsk);
225 msr->r_msg = ERR_PTR(res);
230 * freeque() wakes up waiters on the sender and receiver waiting queue,
231 * removes the message queue from message queue ID
232 * array, and cleans up all the messages associated with this queue.
234 * msg_ids.mutex and the spinlock for this message queue is hold
235 * before freeque() is called. msg_ids.mutex remains locked on exit.
237 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq, int id)
239 struct list_head *tmp;
241 expunge_all(msq, -EIDRM);
242 ss_wakeup(&msq->q_senders, 1);
243 msq = msg_rmid(ns, id);
246 tmp = msq->q_messages.next;
247 while (tmp != &msq->q_messages) {
248 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
251 atomic_dec(&msg_hdrs);
254 atomic_sub(msq->q_cbytes, &msg_bytes);
255 security_msg_queue_free(msq);
259 asmlinkage long sys_msgget(key_t key, int msgflg)
261 struct msg_queue *msq;
262 int id, ret = -EPERM;
263 struct ipc_namespace *ns;
265 ns = current->nsproxy->ipc_ns;
267 mutex_lock(&msg_ids(ns).mutex);
268 if (key == IPC_PRIVATE)
269 ret = newque(ns, key, msgflg);
270 else if ((id = ipc_findkey(&msg_ids(ns), key)) == -1) { /* key not used */
271 if (!(msgflg & IPC_CREAT))
274 ret = newque(ns, key, msgflg);
275 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
278 msq = msg_lock(ns, id);
280 if (ipcperms(&msq->q_perm, msgflg))
283 int qid = msg_buildid(ns, id, msq->q_perm.seq);
285 ret = security_msg_queue_associate(msq, msgflg);
291 mutex_unlock(&msg_ids(ns).mutex);
296 static inline unsigned long
297 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
301 return copy_to_user(buf, in, sizeof(*in));
306 memset(&out, 0, sizeof(out));
308 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
310 out.msg_stime = in->msg_stime;
311 out.msg_rtime = in->msg_rtime;
312 out.msg_ctime = in->msg_ctime;
314 if (in->msg_cbytes > USHRT_MAX)
315 out.msg_cbytes = USHRT_MAX;
317 out.msg_cbytes = in->msg_cbytes;
318 out.msg_lcbytes = in->msg_cbytes;
320 if (in->msg_qnum > USHRT_MAX)
321 out.msg_qnum = USHRT_MAX;
323 out.msg_qnum = in->msg_qnum;
325 if (in->msg_qbytes > USHRT_MAX)
326 out.msg_qbytes = USHRT_MAX;
328 out.msg_qbytes = in->msg_qbytes;
329 out.msg_lqbytes = in->msg_qbytes;
331 out.msg_lspid = in->msg_lspid;
332 out.msg_lrpid = in->msg_lrpid;
334 return copy_to_user(buf, &out, sizeof(out));
342 unsigned long qbytes;
348 static inline unsigned long
349 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
354 struct msqid64_ds tbuf;
356 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
359 out->qbytes = tbuf.msg_qbytes;
360 out->uid = tbuf.msg_perm.uid;
361 out->gid = tbuf.msg_perm.gid;
362 out->mode = tbuf.msg_perm.mode;
368 struct msqid_ds tbuf_old;
370 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
373 out->uid = tbuf_old.msg_perm.uid;
374 out->gid = tbuf_old.msg_perm.gid;
375 out->mode = tbuf_old.msg_perm.mode;
377 if (tbuf_old.msg_qbytes == 0)
378 out->qbytes = tbuf_old.msg_lqbytes;
380 out->qbytes = tbuf_old.msg_qbytes;
389 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
391 struct kern_ipc_perm *ipcp;
392 struct msq_setbuf setbuf;
393 struct msg_queue *msq;
395 struct ipc_namespace *ns;
397 if (msqid < 0 || cmd < 0)
400 version = ipc_parse_version(&cmd);
401 ns = current->nsproxy->ipc_ns;
407 struct msginfo msginfo;
413 * We must not return kernel stack data.
414 * due to padding, it's not enough
415 * to set all member fields.
417 err = security_msg_queue_msgctl(NULL, cmd);
421 memset(&msginfo, 0, sizeof(msginfo));
422 msginfo.msgmni = ns->msg_ctlmni;
423 msginfo.msgmax = ns->msg_ctlmax;
424 msginfo.msgmnb = ns->msg_ctlmnb;
425 msginfo.msgssz = MSGSSZ;
426 msginfo.msgseg = MSGSEG;
427 mutex_lock(&msg_ids(ns).mutex);
428 if (cmd == MSG_INFO) {
429 msginfo.msgpool = msg_ids(ns).in_use;
430 msginfo.msgmap = atomic_read(&msg_hdrs);
431 msginfo.msgtql = atomic_read(&msg_bytes);
433 msginfo.msgmap = MSGMAP;
434 msginfo.msgpool = MSGPOOL;
435 msginfo.msgtql = MSGTQL;
437 max_id = msg_ids(ns).max_id;
438 mutex_unlock(&msg_ids(ns).mutex);
439 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
441 return (max_id < 0) ? 0 : max_id;
446 struct msqid64_ds tbuf;
451 if (cmd == MSG_STAT && msqid >= msg_ids(ns).entries->size)
454 memset(&tbuf, 0, sizeof(tbuf));
456 msq = msg_lock(ns, msqid);
460 if (cmd == MSG_STAT) {
461 success_return = msg_buildid(ns, msqid, msq->q_perm.seq);
464 if (msg_checkid(ns, msq, msqid))
469 if (ipcperms(&msq->q_perm, S_IRUGO))
472 err = security_msg_queue_msgctl(msq, cmd);
476 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
477 tbuf.msg_stime = msq->q_stime;
478 tbuf.msg_rtime = msq->q_rtime;
479 tbuf.msg_ctime = msq->q_ctime;
480 tbuf.msg_cbytes = msq->q_cbytes;
481 tbuf.msg_qnum = msq->q_qnum;
482 tbuf.msg_qbytes = msq->q_qbytes;
483 tbuf.msg_lspid = msq->q_lspid;
484 tbuf.msg_lrpid = msq->q_lrpid;
486 if (copy_msqid_to_user(buf, &tbuf, version))
488 return success_return;
493 if (copy_msqid_from_user(&setbuf, buf, version))
502 mutex_lock(&msg_ids(ns).mutex);
503 msq = msg_lock(ns, msqid);
509 if (msg_checkid(ns, msq, msqid))
513 err = audit_ipc_obj(ipcp);
517 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
524 if (current->euid != ipcp->cuid &&
525 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
526 /* We _could_ check for CAP_CHOWN above, but we don't */
529 err = security_msg_queue_msgctl(msq, cmd);
537 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
540 msq->q_qbytes = setbuf.qbytes;
542 ipcp->uid = setbuf.uid;
543 ipcp->gid = setbuf.gid;
544 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
545 (S_IRWXUGO & setbuf.mode);
546 msq->q_ctime = get_seconds();
547 /* sleeping receivers might be excluded by
548 * stricter permissions.
550 expunge_all(msq, -EAGAIN);
551 /* sleeping senders might be able to send
552 * due to a larger queue size.
554 ss_wakeup(&msq->q_senders, 0);
559 freeque(ns, msq, msqid);
564 mutex_unlock(&msg_ids(ns).mutex);
574 static int testmsg(struct msg_msg *msg, long type, int mode)
580 case SEARCH_LESSEQUAL:
581 if (msg->m_type <=type)
585 if (msg->m_type == type)
588 case SEARCH_NOTEQUAL:
589 if (msg->m_type != type)
596 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
598 struct list_head *tmp;
600 tmp = msq->q_receivers.next;
601 while (tmp != &msq->q_receivers) {
602 struct msg_receiver *msr;
604 msr = list_entry(tmp, struct msg_receiver, r_list);
606 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
607 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
608 msr->r_msgtype, msr->r_mode)) {
610 list_del(&msr->r_list);
611 if (msr->r_maxsize < msg->m_ts) {
613 wake_up_process(msr->r_tsk);
615 msr->r_msg = ERR_PTR(-E2BIG);
618 msq->q_lrpid = msr->r_tsk->pid;
619 msq->q_rtime = get_seconds();
620 wake_up_process(msr->r_tsk);
631 long do_msgsnd(int msqid, long mtype, void __user *mtext,
632 size_t msgsz, int msgflg)
634 struct msg_queue *msq;
637 struct ipc_namespace *ns;
639 ns = current->nsproxy->ipc_ns;
641 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
646 msg = load_msg(mtext, msgsz);
653 msq = msg_lock(ns, msqid);
659 if (msg_checkid(ns, msq, msqid))
660 goto out_unlock_free;
666 if (ipcperms(&msq->q_perm, S_IWUGO))
667 goto out_unlock_free;
669 err = security_msg_queue_msgsnd(msq, msg, msgflg);
671 goto out_unlock_free;
673 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
674 1 + msq->q_qnum <= msq->q_qbytes) {
678 /* queue full, wait: */
679 if (msgflg & IPC_NOWAIT) {
681 goto out_unlock_free;
688 ipc_lock_by_ptr(&msq->q_perm);
690 if (msq->q_perm.deleted) {
692 goto out_unlock_free;
696 if (signal_pending(current)) {
697 err = -ERESTARTNOHAND;
698 goto out_unlock_free;
702 msq->q_lspid = current->tgid;
703 msq->q_stime = get_seconds();
705 if (!pipelined_send(msq, msg)) {
706 /* noone is waiting for this message, enqueue it */
707 list_add_tail(&msg->m_list, &msq->q_messages);
708 msq->q_cbytes += msgsz;
710 atomic_add(msgsz, &msg_bytes);
711 atomic_inc(&msg_hdrs);
726 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
730 if (get_user(mtype, &msgp->mtype))
732 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
735 static inline int convert_mode(long *msgtyp, int msgflg)
738 * find message of correct type.
739 * msgtyp = 0 => get first.
740 * msgtyp > 0 => get first message of matching type.
741 * msgtyp < 0 => get message with least type must be < abs(msgtype).
747 return SEARCH_LESSEQUAL;
749 if (msgflg & MSG_EXCEPT)
750 return SEARCH_NOTEQUAL;
754 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
755 size_t msgsz, long msgtyp, int msgflg)
757 struct msg_queue *msq;
760 struct ipc_namespace *ns;
762 if (msqid < 0 || (long) msgsz < 0)
764 mode = convert_mode(&msgtyp, msgflg);
765 ns = current->nsproxy->ipc_ns;
767 msq = msg_lock(ns, msqid);
771 msg = ERR_PTR(-EIDRM);
772 if (msg_checkid(ns, msq, msqid))
776 struct msg_receiver msr_d;
777 struct list_head *tmp;
779 msg = ERR_PTR(-EACCES);
780 if (ipcperms(&msq->q_perm, S_IRUGO))
783 msg = ERR_PTR(-EAGAIN);
784 tmp = msq->q_messages.next;
785 while (tmp != &msq->q_messages) {
786 struct msg_msg *walk_msg;
788 walk_msg = list_entry(tmp, struct msg_msg, m_list);
789 if (testmsg(walk_msg, msgtyp, mode) &&
790 !security_msg_queue_msgrcv(msq, walk_msg, current,
794 if (mode == SEARCH_LESSEQUAL &&
795 walk_msg->m_type != 1) {
797 msgtyp = walk_msg->m_type - 1;
807 * Found a suitable message.
808 * Unlink it from the queue.
810 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
811 msg = ERR_PTR(-E2BIG);
814 list_del(&msg->m_list);
816 msq->q_rtime = get_seconds();
817 msq->q_lrpid = current->tgid;
818 msq->q_cbytes -= msg->m_ts;
819 atomic_sub(msg->m_ts, &msg_bytes);
820 atomic_dec(&msg_hdrs);
821 ss_wakeup(&msq->q_senders, 0);
825 /* No message waiting. Wait for a message */
826 if (msgflg & IPC_NOWAIT) {
827 msg = ERR_PTR(-ENOMSG);
830 list_add_tail(&msr_d.r_list, &msq->q_receivers);
831 msr_d.r_tsk = current;
832 msr_d.r_msgtype = msgtyp;
834 if (msgflg & MSG_NOERROR)
835 msr_d.r_maxsize = INT_MAX;
837 msr_d.r_maxsize = msgsz;
838 msr_d.r_msg = ERR_PTR(-EAGAIN);
839 current->state = TASK_INTERRUPTIBLE;
844 /* Lockless receive, part 1:
845 * Disable preemption. We don't hold a reference to the queue
846 * and getting a reference would defeat the idea of a lockless
847 * operation, thus the code relies on rcu to guarantee the
849 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
850 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
851 * rcu_read_lock() prevents preemption between reading r_msg
852 * and the spin_lock() inside ipc_lock_by_ptr().
856 /* Lockless receive, part 2:
857 * Wait until pipelined_send or expunge_all are outside of
858 * wake_up_process(). There is a race with exit(), see
859 * ipc/mqueue.c for the details.
861 msg = (struct msg_msg*)msr_d.r_msg;
862 while (msg == NULL) {
864 msg = (struct msg_msg *)msr_d.r_msg;
867 /* Lockless receive, part 3:
868 * If there is a message or an error then accept it without
871 if (msg != ERR_PTR(-EAGAIN)) {
876 /* Lockless receive, part 3:
877 * Acquire the queue spinlock.
879 ipc_lock_by_ptr(&msq->q_perm);
882 /* Lockless receive, part 4:
883 * Repeat test after acquiring the spinlock.
885 msg = (struct msg_msg*)msr_d.r_msg;
886 if (msg != ERR_PTR(-EAGAIN))
889 list_del(&msr_d.r_list);
890 if (signal_pending(current)) {
891 msg = ERR_PTR(-ERESTARTNOHAND);
900 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
901 *pmtype = msg->m_type;
902 if (store_msg(mtext, msg, msgsz))
910 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
911 long msgtyp, int msgflg)
915 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
919 if (put_user(mtype, &msgp->mtype))
925 #ifdef CONFIG_PROC_FS
926 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
928 struct msg_queue *msq = it;
931 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",