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 volatile struct msg_msg *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 kfree(ns->ids[IPC_MSG_IDS]);
128 ns->ids[IPC_MSG_IDS] = NULL;
132 void __init msg_init(void)
134 __msg_init_ns(&init_ipc_ns, &init_msg_ids);
135 ipc_init_proc_interface("sysvipc/msg",
136 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
137 IPC_MSG_IDS, sysvipc_msg_proc_show);
140 static int newque (struct ipc_namespace *ns, key_t key, int msgflg)
142 struct msg_queue *msq;
145 msq = ipc_rcu_alloc(sizeof(*msq));
149 msq->q_perm.mode = msgflg & S_IRWXUGO;
150 msq->q_perm.key = key;
152 msq->q_perm.security = NULL;
153 retval = security_msg_queue_alloc(msq);
159 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
161 security_msg_queue_free(msq);
166 msq->q_id = msg_buildid(ns, id, msq->q_perm.seq);
167 msq->q_stime = msq->q_rtime = 0;
168 msq->q_ctime = get_seconds();
169 msq->q_cbytes = msq->q_qnum = 0;
170 msq->q_qbytes = ns->msg_ctlmnb;
171 msq->q_lspid = msq->q_lrpid = 0;
172 INIT_LIST_HEAD(&msq->q_messages);
173 INIT_LIST_HEAD(&msq->q_receivers);
174 INIT_LIST_HEAD(&msq->q_senders);
180 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
183 current->state = TASK_INTERRUPTIBLE;
184 list_add_tail(&mss->list, &msq->q_senders);
187 static inline void ss_del(struct msg_sender *mss)
189 if (mss->list.next != NULL)
190 list_del(&mss->list);
193 static void ss_wakeup(struct list_head *h, int kill)
195 struct list_head *tmp;
199 struct msg_sender *mss;
201 mss = list_entry(tmp, struct msg_sender, list);
204 mss->list.next = NULL;
205 wake_up_process(mss->tsk);
209 static void expunge_all(struct msg_queue *msq, int res)
211 struct list_head *tmp;
213 tmp = msq->q_receivers.next;
214 while (tmp != &msq->q_receivers) {
215 struct msg_receiver *msr;
217 msr = list_entry(tmp, struct msg_receiver, r_list);
220 wake_up_process(msr->r_tsk);
222 msr->r_msg = ERR_PTR(res);
227 * freeque() wakes up waiters on the sender and receiver waiting queue,
228 * removes the message queue from message queue ID
229 * array, and cleans up all the messages associated with this queue.
231 * msg_ids.mutex and the spinlock for this message queue is hold
232 * before freeque() is called. msg_ids.mutex remains locked on exit.
234 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq, int id)
236 struct list_head *tmp;
238 expunge_all(msq, -EIDRM);
239 ss_wakeup(&msq->q_senders, 1);
240 msq = msg_rmid(ns, id);
243 tmp = msq->q_messages.next;
244 while (tmp != &msq->q_messages) {
245 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
248 atomic_dec(&msg_hdrs);
251 atomic_sub(msq->q_cbytes, &msg_bytes);
252 security_msg_queue_free(msq);
256 asmlinkage long sys_msgget(key_t key, int msgflg)
258 struct msg_queue *msq;
259 int id, ret = -EPERM;
260 struct ipc_namespace *ns;
262 ns = current->nsproxy->ipc_ns;
264 mutex_lock(&msg_ids(ns).mutex);
265 if (key == IPC_PRIVATE)
266 ret = newque(ns, key, msgflg);
267 else if ((id = ipc_findkey(&msg_ids(ns), key)) == -1) { /* key not used */
268 if (!(msgflg & IPC_CREAT))
271 ret = newque(ns, key, msgflg);
272 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
275 msq = msg_lock(ns, id);
277 if (ipcperms(&msq->q_perm, msgflg))
280 int qid = msg_buildid(ns, id, msq->q_perm.seq);
282 ret = security_msg_queue_associate(msq, msgflg);
288 mutex_unlock(&msg_ids(ns).mutex);
293 static inline unsigned long
294 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
298 return copy_to_user(buf, in, sizeof(*in));
303 memset(&out, 0, sizeof(out));
305 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
307 out.msg_stime = in->msg_stime;
308 out.msg_rtime = in->msg_rtime;
309 out.msg_ctime = in->msg_ctime;
311 if (in->msg_cbytes > USHRT_MAX)
312 out.msg_cbytes = USHRT_MAX;
314 out.msg_cbytes = in->msg_cbytes;
315 out.msg_lcbytes = in->msg_cbytes;
317 if (in->msg_qnum > USHRT_MAX)
318 out.msg_qnum = USHRT_MAX;
320 out.msg_qnum = in->msg_qnum;
322 if (in->msg_qbytes > USHRT_MAX)
323 out.msg_qbytes = USHRT_MAX;
325 out.msg_qbytes = in->msg_qbytes;
326 out.msg_lqbytes = in->msg_qbytes;
328 out.msg_lspid = in->msg_lspid;
329 out.msg_lrpid = in->msg_lrpid;
331 return copy_to_user(buf, &out, sizeof(out));
339 unsigned long qbytes;
345 static inline unsigned long
346 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
351 struct msqid64_ds tbuf;
353 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
356 out->qbytes = tbuf.msg_qbytes;
357 out->uid = tbuf.msg_perm.uid;
358 out->gid = tbuf.msg_perm.gid;
359 out->mode = tbuf.msg_perm.mode;
365 struct msqid_ds tbuf_old;
367 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
370 out->uid = tbuf_old.msg_perm.uid;
371 out->gid = tbuf_old.msg_perm.gid;
372 out->mode = tbuf_old.msg_perm.mode;
374 if (tbuf_old.msg_qbytes == 0)
375 out->qbytes = tbuf_old.msg_lqbytes;
377 out->qbytes = tbuf_old.msg_qbytes;
386 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
388 struct kern_ipc_perm *ipcp;
389 struct msq_setbuf setbuf;
390 struct msg_queue *msq;
392 struct ipc_namespace *ns;
394 if (msqid < 0 || cmd < 0)
397 version = ipc_parse_version(&cmd);
398 ns = current->nsproxy->ipc_ns;
404 struct msginfo msginfo;
410 * We must not return kernel stack data.
411 * due to padding, it's not enough
412 * to set all member fields.
414 err = security_msg_queue_msgctl(NULL, cmd);
418 memset(&msginfo, 0, sizeof(msginfo));
419 msginfo.msgmni = ns->msg_ctlmni;
420 msginfo.msgmax = ns->msg_ctlmax;
421 msginfo.msgmnb = ns->msg_ctlmnb;
422 msginfo.msgssz = MSGSSZ;
423 msginfo.msgseg = MSGSEG;
424 mutex_lock(&msg_ids(ns).mutex);
425 if (cmd == MSG_INFO) {
426 msginfo.msgpool = msg_ids(ns).in_use;
427 msginfo.msgmap = atomic_read(&msg_hdrs);
428 msginfo.msgtql = atomic_read(&msg_bytes);
430 msginfo.msgmap = MSGMAP;
431 msginfo.msgpool = MSGPOOL;
432 msginfo.msgtql = MSGTQL;
434 max_id = msg_ids(ns).max_id;
435 mutex_unlock(&msg_ids(ns).mutex);
436 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
438 return (max_id < 0) ? 0 : max_id;
443 struct msqid64_ds tbuf;
448 if (cmd == MSG_STAT && msqid >= msg_ids(ns).entries->size)
451 memset(&tbuf, 0, sizeof(tbuf));
453 msq = msg_lock(ns, msqid);
457 if (cmd == MSG_STAT) {
458 success_return = msg_buildid(ns, msqid, msq->q_perm.seq);
461 if (msg_checkid(ns, msq, msqid))
466 if (ipcperms(&msq->q_perm, S_IRUGO))
469 err = security_msg_queue_msgctl(msq, cmd);
473 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
474 tbuf.msg_stime = msq->q_stime;
475 tbuf.msg_rtime = msq->q_rtime;
476 tbuf.msg_ctime = msq->q_ctime;
477 tbuf.msg_cbytes = msq->q_cbytes;
478 tbuf.msg_qnum = msq->q_qnum;
479 tbuf.msg_qbytes = msq->q_qbytes;
480 tbuf.msg_lspid = msq->q_lspid;
481 tbuf.msg_lrpid = msq->q_lrpid;
483 if (copy_msqid_to_user(buf, &tbuf, version))
485 return success_return;
490 if (copy_msqid_from_user(&setbuf, buf, version))
499 mutex_lock(&msg_ids(ns).mutex);
500 msq = msg_lock(ns, msqid);
506 if (msg_checkid(ns, msq, msqid))
510 err = audit_ipc_obj(ipcp);
514 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
521 if (current->euid != ipcp->cuid &&
522 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
523 /* We _could_ check for CAP_CHOWN above, but we don't */
526 err = security_msg_queue_msgctl(msq, cmd);
534 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
537 msq->q_qbytes = setbuf.qbytes;
539 ipcp->uid = setbuf.uid;
540 ipcp->gid = setbuf.gid;
541 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
542 (S_IRWXUGO & setbuf.mode);
543 msq->q_ctime = get_seconds();
544 /* sleeping receivers might be excluded by
545 * stricter permissions.
547 expunge_all(msq, -EAGAIN);
548 /* sleeping senders might be able to send
549 * due to a larger queue size.
551 ss_wakeup(&msq->q_senders, 0);
556 freeque(ns, msq, msqid);
561 mutex_unlock(&msg_ids(ns).mutex);
571 static int testmsg(struct msg_msg *msg, long type, int mode)
577 case SEARCH_LESSEQUAL:
578 if (msg->m_type <=type)
582 if (msg->m_type == type)
585 case SEARCH_NOTEQUAL:
586 if (msg->m_type != type)
593 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
595 struct list_head *tmp;
597 tmp = msq->q_receivers.next;
598 while (tmp != &msq->q_receivers) {
599 struct msg_receiver *msr;
601 msr = list_entry(tmp, struct msg_receiver, r_list);
603 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
604 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
605 msr->r_msgtype, msr->r_mode)) {
607 list_del(&msr->r_list);
608 if (msr->r_maxsize < msg->m_ts) {
610 wake_up_process(msr->r_tsk);
612 msr->r_msg = ERR_PTR(-E2BIG);
615 msq->q_lrpid = msr->r_tsk->pid;
616 msq->q_rtime = get_seconds();
617 wake_up_process(msr->r_tsk);
629 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
631 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)
641 if (get_user(mtype, &msgp->mtype))
646 msg = load_msg(msgp->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);
725 static inline int convert_mode(long *msgtyp, int msgflg)
728 * find message of correct type.
729 * msgtyp = 0 => get first.
730 * msgtyp > 0 => get first message of matching type.
731 * msgtyp < 0 => get message with least type must be < abs(msgtype).
737 return SEARCH_LESSEQUAL;
739 if (msgflg & MSG_EXCEPT)
740 return SEARCH_NOTEQUAL;
744 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
745 long msgtyp, int msgflg)
747 struct msg_queue *msq;
750 struct ipc_namespace *ns;
752 if (msqid < 0 || (long) msgsz < 0)
754 mode = convert_mode(&msgtyp, msgflg);
755 ns = current->nsproxy->ipc_ns;
757 msq = msg_lock(ns, msqid);
761 msg = ERR_PTR(-EIDRM);
762 if (msg_checkid(ns, msq, msqid))
766 struct msg_receiver msr_d;
767 struct list_head *tmp;
769 msg = ERR_PTR(-EACCES);
770 if (ipcperms(&msq->q_perm, S_IRUGO))
773 msg = ERR_PTR(-EAGAIN);
774 tmp = msq->q_messages.next;
775 while (tmp != &msq->q_messages) {
776 struct msg_msg *walk_msg;
778 walk_msg = list_entry(tmp, struct msg_msg, m_list);
779 if (testmsg(walk_msg, msgtyp, mode) &&
780 !security_msg_queue_msgrcv(msq, walk_msg, current,
784 if (mode == SEARCH_LESSEQUAL &&
785 walk_msg->m_type != 1) {
787 msgtyp = walk_msg->m_type - 1;
797 * Found a suitable message.
798 * Unlink it from the queue.
800 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
801 msg = ERR_PTR(-E2BIG);
804 list_del(&msg->m_list);
806 msq->q_rtime = get_seconds();
807 msq->q_lrpid = current->tgid;
808 msq->q_cbytes -= msg->m_ts;
809 atomic_sub(msg->m_ts, &msg_bytes);
810 atomic_dec(&msg_hdrs);
811 ss_wakeup(&msq->q_senders, 0);
815 /* No message waiting. Wait for a message */
816 if (msgflg & IPC_NOWAIT) {
817 msg = ERR_PTR(-ENOMSG);
820 list_add_tail(&msr_d.r_list, &msq->q_receivers);
821 msr_d.r_tsk = current;
822 msr_d.r_msgtype = msgtyp;
824 if (msgflg & MSG_NOERROR)
825 msr_d.r_maxsize = INT_MAX;
827 msr_d.r_maxsize = msgsz;
828 msr_d.r_msg = ERR_PTR(-EAGAIN);
829 current->state = TASK_INTERRUPTIBLE;
834 /* Lockless receive, part 1:
835 * Disable preemption. We don't hold a reference to the queue
836 * and getting a reference would defeat the idea of a lockless
837 * operation, thus the code relies on rcu to guarantee the
839 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
840 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
841 * rcu_read_lock() prevents preemption between reading r_msg
842 * and the spin_lock() inside ipc_lock_by_ptr().
846 /* Lockless receive, part 2:
847 * Wait until pipelined_send or expunge_all are outside of
848 * wake_up_process(). There is a race with exit(), see
849 * ipc/mqueue.c for the details.
851 msg = (struct msg_msg*)msr_d.r_msg;
852 while (msg == NULL) {
854 msg = (struct msg_msg *)msr_d.r_msg;
857 /* Lockless receive, part 3:
858 * If there is a message or an error then accept it without
861 if (msg != ERR_PTR(-EAGAIN)) {
866 /* Lockless receive, part 3:
867 * Acquire the queue spinlock.
869 ipc_lock_by_ptr(&msq->q_perm);
872 /* Lockless receive, part 4:
873 * Repeat test after acquiring the spinlock.
875 msg = (struct msg_msg*)msr_d.r_msg;
876 if (msg != ERR_PTR(-EAGAIN))
879 list_del(&msr_d.r_list);
880 if (signal_pending(current)) {
881 msg = ERR_PTR(-ERESTARTNOHAND);
890 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
891 if (put_user (msg->m_type, &msgp->mtype) ||
892 store_msg(msgp->mtext, msg, msgsz)) {
900 #ifdef CONFIG_PROC_FS
901 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
903 struct msg_queue *msq = it;
906 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",