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 __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);
99 int msg_init_ns(struct ipc_namespace *ns)
103 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
107 __msg_init_ns(ns, ids);
111 void msg_exit_ns(struct ipc_namespace *ns)
114 struct msg_queue *msq;
116 mutex_lock(&msg_ids(ns).mutex);
117 for (i = 0; i <= msg_ids(ns).max_id; i++) {
118 msq = msg_lock(ns, i);
124 mutex_unlock(&msg_ids(ns).mutex);
126 ipc_fini_ids(ns->ids[IPC_MSG_IDS]);
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 int newque (struct ipc_namespace *ns, key_t key, int msgflg)
141 struct msg_queue *msq;
144 msq = ipc_rcu_alloc(sizeof(*msq));
148 msq->q_perm.mode = msgflg & S_IRWXUGO;
149 msq->q_perm.key = key;
151 msq->q_perm.security = NULL;
152 retval = security_msg_queue_alloc(msq);
158 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
160 security_msg_queue_free(msq);
165 msq->q_id = msg_buildid(ns, id, msq->q_perm.seq);
166 msq->q_stime = msq->q_rtime = 0;
167 msq->q_ctime = get_seconds();
168 msq->q_cbytes = msq->q_qnum = 0;
169 msq->q_qbytes = ns->msg_ctlmnb;
170 msq->q_lspid = msq->q_lrpid = 0;
171 INIT_LIST_HEAD(&msq->q_messages);
172 INIT_LIST_HEAD(&msq->q_receivers);
173 INIT_LIST_HEAD(&msq->q_senders);
179 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
182 current->state = TASK_INTERRUPTIBLE;
183 list_add_tail(&mss->list, &msq->q_senders);
186 static inline void ss_del(struct msg_sender *mss)
188 if (mss->list.next != NULL)
189 list_del(&mss->list);
192 static void ss_wakeup(struct list_head *h, int kill)
194 struct list_head *tmp;
198 struct msg_sender *mss;
200 mss = list_entry(tmp, struct msg_sender, list);
203 mss->list.next = NULL;
204 wake_up_process(mss->tsk);
208 static void expunge_all(struct msg_queue *msq, int res)
210 struct list_head *tmp;
212 tmp = msq->q_receivers.next;
213 while (tmp != &msq->q_receivers) {
214 struct msg_receiver *msr;
216 msr = list_entry(tmp, struct msg_receiver, r_list);
219 wake_up_process(msr->r_tsk);
221 msr->r_msg = ERR_PTR(res);
226 * freeque() wakes up waiters on the sender and receiver waiting queue,
227 * removes the message queue from message queue ID
228 * array, and cleans up all the messages associated with this queue.
230 * msg_ids.mutex and the spinlock for this message queue is hold
231 * before freeque() is called. msg_ids.mutex remains locked on exit.
233 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq, int id)
235 struct list_head *tmp;
237 expunge_all(msq, -EIDRM);
238 ss_wakeup(&msq->q_senders, 1);
239 msq = msg_rmid(ns, id);
242 tmp = msq->q_messages.next;
243 while (tmp != &msq->q_messages) {
244 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
247 atomic_dec(&msg_hdrs);
250 atomic_sub(msq->q_cbytes, &msg_bytes);
251 security_msg_queue_free(msq);
255 asmlinkage long sys_msgget(key_t key, int msgflg)
257 struct msg_queue *msq;
258 int id, ret = -EPERM;
259 struct ipc_namespace *ns;
261 ns = current->nsproxy->ipc_ns;
263 mutex_lock(&msg_ids(ns).mutex);
264 if (key == IPC_PRIVATE)
265 ret = newque(ns, key, msgflg);
266 else if ((id = ipc_findkey(&msg_ids(ns), key)) == -1) { /* key not used */
267 if (!(msgflg & IPC_CREAT))
270 ret = newque(ns, key, msgflg);
271 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
274 msq = msg_lock(ns, id);
276 if (ipcperms(&msq->q_perm, msgflg))
279 int qid = msg_buildid(ns, id, msq->q_perm.seq);
281 ret = security_msg_queue_associate(msq, msgflg);
287 mutex_unlock(&msg_ids(ns).mutex);
292 static inline unsigned long
293 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
297 return copy_to_user(buf, in, sizeof(*in));
302 memset(&out, 0, sizeof(out));
304 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
306 out.msg_stime = in->msg_stime;
307 out.msg_rtime = in->msg_rtime;
308 out.msg_ctime = in->msg_ctime;
310 if (in->msg_cbytes > USHRT_MAX)
311 out.msg_cbytes = USHRT_MAX;
313 out.msg_cbytes = in->msg_cbytes;
314 out.msg_lcbytes = in->msg_cbytes;
316 if (in->msg_qnum > USHRT_MAX)
317 out.msg_qnum = USHRT_MAX;
319 out.msg_qnum = in->msg_qnum;
321 if (in->msg_qbytes > USHRT_MAX)
322 out.msg_qbytes = USHRT_MAX;
324 out.msg_qbytes = in->msg_qbytes;
325 out.msg_lqbytes = in->msg_qbytes;
327 out.msg_lspid = in->msg_lspid;
328 out.msg_lrpid = in->msg_lrpid;
330 return copy_to_user(buf, &out, sizeof(out));
338 unsigned long qbytes;
344 static inline unsigned long
345 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
350 struct msqid64_ds tbuf;
352 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
355 out->qbytes = tbuf.msg_qbytes;
356 out->uid = tbuf.msg_perm.uid;
357 out->gid = tbuf.msg_perm.gid;
358 out->mode = tbuf.msg_perm.mode;
364 struct msqid_ds tbuf_old;
366 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
369 out->uid = tbuf_old.msg_perm.uid;
370 out->gid = tbuf_old.msg_perm.gid;
371 out->mode = tbuf_old.msg_perm.mode;
373 if (tbuf_old.msg_qbytes == 0)
374 out->qbytes = tbuf_old.msg_lqbytes;
376 out->qbytes = tbuf_old.msg_qbytes;
385 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
387 struct kern_ipc_perm *ipcp;
388 struct msq_setbuf setbuf;
389 struct msg_queue *msq;
391 struct ipc_namespace *ns;
393 if (msqid < 0 || cmd < 0)
396 version = ipc_parse_version(&cmd);
397 ns = current->nsproxy->ipc_ns;
403 struct msginfo msginfo;
409 * We must not return kernel stack data.
410 * due to padding, it's not enough
411 * to set all member fields.
413 err = security_msg_queue_msgctl(NULL, cmd);
417 memset(&msginfo, 0, sizeof(msginfo));
418 msginfo.msgmni = ns->msg_ctlmni;
419 msginfo.msgmax = ns->msg_ctlmax;
420 msginfo.msgmnb = ns->msg_ctlmnb;
421 msginfo.msgssz = MSGSSZ;
422 msginfo.msgseg = MSGSEG;
423 mutex_lock(&msg_ids(ns).mutex);
424 if (cmd == MSG_INFO) {
425 msginfo.msgpool = msg_ids(ns).in_use;
426 msginfo.msgmap = atomic_read(&msg_hdrs);
427 msginfo.msgtql = atomic_read(&msg_bytes);
429 msginfo.msgmap = MSGMAP;
430 msginfo.msgpool = MSGPOOL;
431 msginfo.msgtql = MSGTQL;
433 max_id = msg_ids(ns).max_id;
434 mutex_unlock(&msg_ids(ns).mutex);
435 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
437 return (max_id < 0) ? 0 : max_id;
442 struct msqid64_ds tbuf;
447 if (cmd == MSG_STAT && msqid >= msg_ids(ns).entries->size)
450 memset(&tbuf, 0, sizeof(tbuf));
452 msq = msg_lock(ns, msqid);
456 if (cmd == MSG_STAT) {
457 success_return = msg_buildid(ns, msqid, msq->q_perm.seq);
460 if (msg_checkid(ns, msq, msqid))
465 if (ipcperms(&msq->q_perm, S_IRUGO))
468 err = security_msg_queue_msgctl(msq, cmd);
472 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
473 tbuf.msg_stime = msq->q_stime;
474 tbuf.msg_rtime = msq->q_rtime;
475 tbuf.msg_ctime = msq->q_ctime;
476 tbuf.msg_cbytes = msq->q_cbytes;
477 tbuf.msg_qnum = msq->q_qnum;
478 tbuf.msg_qbytes = msq->q_qbytes;
479 tbuf.msg_lspid = msq->q_lspid;
480 tbuf.msg_lrpid = msq->q_lrpid;
482 if (copy_msqid_to_user(buf, &tbuf, version))
484 return success_return;
489 if (copy_msqid_from_user(&setbuf, buf, version))
498 mutex_lock(&msg_ids(ns).mutex);
499 msq = msg_lock(ns, msqid);
505 if (msg_checkid(ns, msq, msqid))
509 err = audit_ipc_obj(ipcp);
513 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
520 if (current->euid != ipcp->cuid &&
521 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
522 /* We _could_ check for CAP_CHOWN above, but we don't */
525 err = security_msg_queue_msgctl(msq, cmd);
533 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
536 msq->q_qbytes = setbuf.qbytes;
538 ipcp->uid = setbuf.uid;
539 ipcp->gid = setbuf.gid;
540 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
541 (S_IRWXUGO & setbuf.mode);
542 msq->q_ctime = get_seconds();
543 /* sleeping receivers might be excluded by
544 * stricter permissions.
546 expunge_all(msq, -EAGAIN);
547 /* sleeping senders might be able to send
548 * due to a larger queue size.
550 ss_wakeup(&msq->q_senders, 0);
555 freeque(ns, msq, msqid);
560 mutex_unlock(&msg_ids(ns).mutex);
570 static int testmsg(struct msg_msg *msg, long type, int mode)
576 case SEARCH_LESSEQUAL:
577 if (msg->m_type <=type)
581 if (msg->m_type == type)
584 case SEARCH_NOTEQUAL:
585 if (msg->m_type != type)
592 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
594 struct list_head *tmp;
596 tmp = msq->q_receivers.next;
597 while (tmp != &msq->q_receivers) {
598 struct msg_receiver *msr;
600 msr = list_entry(tmp, struct msg_receiver, r_list);
602 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
603 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
604 msr->r_msgtype, msr->r_mode)) {
606 list_del(&msr->r_list);
607 if (msr->r_maxsize < msg->m_ts) {
609 wake_up_process(msr->r_tsk);
611 msr->r_msg = ERR_PTR(-E2BIG);
614 msq->q_lrpid = msr->r_tsk->pid;
615 msq->q_rtime = get_seconds();
616 wake_up_process(msr->r_tsk);
627 long do_msgsnd(int msqid, long mtype, void __user *mtext,
628 size_t msgsz, int msgflg)
630 struct msg_queue *msq;
633 struct ipc_namespace *ns;
635 ns = current->nsproxy->ipc_ns;
637 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
642 msg = load_msg(mtext, msgsz);
649 msq = msg_lock(ns, msqid);
655 if (msg_checkid(ns, msq, msqid))
656 goto out_unlock_free;
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 = current->tgid;
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(ns, msqid);
767 msg = ERR_PTR(-EIDRM);
768 if (msg_checkid(ns, msq, msqid))
772 struct msg_receiver msr_d;
773 struct list_head *tmp;
775 msg = ERR_PTR(-EACCES);
776 if (ipcperms(&msq->q_perm, S_IRUGO))
779 msg = ERR_PTR(-EAGAIN);
780 tmp = msq->q_messages.next;
781 while (tmp != &msq->q_messages) {
782 struct msg_msg *walk_msg;
784 walk_msg = list_entry(tmp, struct msg_msg, m_list);
785 if (testmsg(walk_msg, msgtyp, mode) &&
786 !security_msg_queue_msgrcv(msq, walk_msg, current,
790 if (mode == SEARCH_LESSEQUAL &&
791 walk_msg->m_type != 1) {
793 msgtyp = walk_msg->m_type - 1;
803 * Found a suitable message.
804 * Unlink it from the queue.
806 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
807 msg = ERR_PTR(-E2BIG);
810 list_del(&msg->m_list);
812 msq->q_rtime = get_seconds();
813 msq->q_lrpid = current->tgid;
814 msq->q_cbytes -= msg->m_ts;
815 atomic_sub(msg->m_ts, &msg_bytes);
816 atomic_dec(&msg_hdrs);
817 ss_wakeup(&msq->q_senders, 0);
821 /* No message waiting. Wait for a message */
822 if (msgflg & IPC_NOWAIT) {
823 msg = ERR_PTR(-ENOMSG);
826 list_add_tail(&msr_d.r_list, &msq->q_receivers);
827 msr_d.r_tsk = current;
828 msr_d.r_msgtype = msgtyp;
830 if (msgflg & MSG_NOERROR)
831 msr_d.r_maxsize = INT_MAX;
833 msr_d.r_maxsize = msgsz;
834 msr_d.r_msg = ERR_PTR(-EAGAIN);
835 current->state = TASK_INTERRUPTIBLE;
840 /* Lockless receive, part 1:
841 * Disable preemption. We don't hold a reference to the queue
842 * and getting a reference would defeat the idea of a lockless
843 * operation, thus the code relies on rcu to guarantee the
845 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
846 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
847 * rcu_read_lock() prevents preemption between reading r_msg
848 * and the spin_lock() inside ipc_lock_by_ptr().
852 /* Lockless receive, part 2:
853 * Wait until pipelined_send or expunge_all are outside of
854 * wake_up_process(). There is a race with exit(), see
855 * ipc/mqueue.c for the details.
857 msg = (struct msg_msg*)msr_d.r_msg;
858 while (msg == NULL) {
860 msg = (struct msg_msg *)msr_d.r_msg;
863 /* Lockless receive, part 3:
864 * If there is a message or an error then accept it without
867 if (msg != ERR_PTR(-EAGAIN)) {
872 /* Lockless receive, part 3:
873 * Acquire the queue spinlock.
875 ipc_lock_by_ptr(&msq->q_perm);
878 /* Lockless receive, part 4:
879 * Repeat test after acquiring the spinlock.
881 msg = (struct msg_msg*)msr_d.r_msg;
882 if (msg != ERR_PTR(-EAGAIN))
885 list_del(&msr_d.r_list);
886 if (signal_pending(current)) {
887 msg = ERR_PTR(-ERESTARTNOHAND);
896 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
897 *pmtype = msg->m_type;
898 if (store_msg(mtext, msg, msgsz))
906 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
907 long msgtyp, int msgflg)
911 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
915 if (put_user(mtype, &msgp->mtype))
921 #ifdef CONFIG_PROC_FS
922 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
924 struct msg_queue *msq = it;
927 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",