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 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
143 return container_of(ipcp, struct msg_queue, q_perm);
146 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
149 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
151 return container_of(ipcp, struct msg_queue, q_perm);
154 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
156 ipc_rmid(&msg_ids(ns), &s->q_perm);
160 * newque - Create a new msg queue
162 * @params: ptr to the structure that contains the key and msgflg
164 * Called with msg_ids.mutex held
166 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
168 struct msg_queue *msq;
170 key_t key = params->key;
171 int msgflg = params->flg;
173 msq = ipc_rcu_alloc(sizeof(*msq));
177 msq->q_perm.mode = msgflg & S_IRWXUGO;
178 msq->q_perm.key = key;
180 msq->q_perm.security = NULL;
181 retval = security_msg_queue_alloc(msq);
188 * ipc_addid() locks msq
190 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
192 security_msg_queue_free(msq);
197 msq->q_perm.id = msg_buildid(ns, id, msq->q_perm.seq);
198 msq->q_stime = msq->q_rtime = 0;
199 msq->q_ctime = get_seconds();
200 msq->q_cbytes = msq->q_qnum = 0;
201 msq->q_qbytes = ns->msg_ctlmnb;
202 msq->q_lspid = msq->q_lrpid = 0;
203 INIT_LIST_HEAD(&msq->q_messages);
204 INIT_LIST_HEAD(&msq->q_receivers);
205 INIT_LIST_HEAD(&msq->q_senders);
209 return msq->q_perm.id;
212 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
215 current->state = TASK_INTERRUPTIBLE;
216 list_add_tail(&mss->list, &msq->q_senders);
219 static inline void ss_del(struct msg_sender *mss)
221 if (mss->list.next != NULL)
222 list_del(&mss->list);
225 static void ss_wakeup(struct list_head *h, int kill)
227 struct list_head *tmp;
231 struct msg_sender *mss;
233 mss = list_entry(tmp, struct msg_sender, list);
236 mss->list.next = NULL;
237 wake_up_process(mss->tsk);
241 static void expunge_all(struct msg_queue *msq, int res)
243 struct list_head *tmp;
245 tmp = msq->q_receivers.next;
246 while (tmp != &msq->q_receivers) {
247 struct msg_receiver *msr;
249 msr = list_entry(tmp, struct msg_receiver, r_list);
252 wake_up_process(msr->r_tsk);
254 msr->r_msg = ERR_PTR(res);
259 * freeque() wakes up waiters on the sender and receiver waiting queue,
260 * removes the message queue from message queue ID IDR, and cleans up all the
261 * messages associated with this queue.
263 * msg_ids.mutex and the spinlock for this message queue are held
264 * before freeque() is called. msg_ids.mutex remains locked on exit.
266 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
268 struct list_head *tmp;
270 expunge_all(msq, -EIDRM);
271 ss_wakeup(&msq->q_senders, 1);
275 tmp = msq->q_messages.next;
276 while (tmp != &msq->q_messages) {
277 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
280 atomic_dec(&msg_hdrs);
283 atomic_sub(msq->q_cbytes, &msg_bytes);
284 security_msg_queue_free(msq);
289 * Called with msg_ids.mutex and ipcp locked.
291 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
293 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
295 return security_msg_queue_associate(msq, msgflg);
298 asmlinkage long sys_msgget(key_t key, int msgflg)
300 struct ipc_namespace *ns;
301 struct ipc_ops msg_ops;
302 struct ipc_params msg_params;
304 ns = current->nsproxy->ipc_ns;
306 msg_ops.getnew = newque;
307 msg_ops.associate = msg_security;
308 msg_ops.more_checks = NULL;
310 msg_params.key = key;
311 msg_params.flg = msgflg;
313 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
316 static inline unsigned long
317 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
321 return copy_to_user(buf, in, sizeof(*in));
326 memset(&out, 0, sizeof(out));
328 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
330 out.msg_stime = in->msg_stime;
331 out.msg_rtime = in->msg_rtime;
332 out.msg_ctime = in->msg_ctime;
334 if (in->msg_cbytes > USHRT_MAX)
335 out.msg_cbytes = USHRT_MAX;
337 out.msg_cbytes = in->msg_cbytes;
338 out.msg_lcbytes = in->msg_cbytes;
340 if (in->msg_qnum > USHRT_MAX)
341 out.msg_qnum = USHRT_MAX;
343 out.msg_qnum = in->msg_qnum;
345 if (in->msg_qbytes > USHRT_MAX)
346 out.msg_qbytes = USHRT_MAX;
348 out.msg_qbytes = in->msg_qbytes;
349 out.msg_lqbytes = in->msg_qbytes;
351 out.msg_lspid = in->msg_lspid;
352 out.msg_lrpid = in->msg_lrpid;
354 return copy_to_user(buf, &out, sizeof(out));
362 unsigned long qbytes;
368 static inline unsigned long
369 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
374 struct msqid64_ds tbuf;
376 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
379 out->qbytes = tbuf.msg_qbytes;
380 out->uid = tbuf.msg_perm.uid;
381 out->gid = tbuf.msg_perm.gid;
382 out->mode = tbuf.msg_perm.mode;
388 struct msqid_ds tbuf_old;
390 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
393 out->uid = tbuf_old.msg_perm.uid;
394 out->gid = tbuf_old.msg_perm.gid;
395 out->mode = tbuf_old.msg_perm.mode;
397 if (tbuf_old.msg_qbytes == 0)
398 out->qbytes = tbuf_old.msg_lqbytes;
400 out->qbytes = tbuf_old.msg_qbytes;
409 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
411 struct kern_ipc_perm *ipcp;
412 struct msq_setbuf uninitialized_var(setbuf);
413 struct msg_queue *msq;
415 struct ipc_namespace *ns;
417 if (msqid < 0 || cmd < 0)
420 version = ipc_parse_version(&cmd);
421 ns = current->nsproxy->ipc_ns;
427 struct msginfo msginfo;
433 * We must not return kernel stack data.
434 * due to padding, it's not enough
435 * to set all member fields.
437 err = security_msg_queue_msgctl(NULL, cmd);
441 memset(&msginfo, 0, sizeof(msginfo));
442 msginfo.msgmni = ns->msg_ctlmni;
443 msginfo.msgmax = ns->msg_ctlmax;
444 msginfo.msgmnb = ns->msg_ctlmnb;
445 msginfo.msgssz = MSGSSZ;
446 msginfo.msgseg = MSGSEG;
447 mutex_lock(&msg_ids(ns).mutex);
448 if (cmd == MSG_INFO) {
449 msginfo.msgpool = msg_ids(ns).in_use;
450 msginfo.msgmap = atomic_read(&msg_hdrs);
451 msginfo.msgtql = atomic_read(&msg_bytes);
453 msginfo.msgmap = MSGMAP;
454 msginfo.msgpool = MSGPOOL;
455 msginfo.msgtql = MSGTQL;
457 max_id = ipc_get_maxid(&msg_ids(ns));
458 mutex_unlock(&msg_ids(ns).mutex);
459 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
461 return (max_id < 0) ? 0 : max_id;
463 case MSG_STAT: /* msqid is an index rather than a msg queue id */
466 struct msqid64_ds tbuf;
472 if (cmd == MSG_STAT) {
473 msq = msg_lock(ns, msqid);
476 success_return = msq->q_perm.id;
478 msq = msg_lock_check(ns, msqid);
484 if (ipcperms(&msq->q_perm, S_IRUGO))
487 err = security_msg_queue_msgctl(msq, cmd);
491 memset(&tbuf, 0, sizeof(tbuf));
493 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
494 tbuf.msg_stime = msq->q_stime;
495 tbuf.msg_rtime = msq->q_rtime;
496 tbuf.msg_ctime = msq->q_ctime;
497 tbuf.msg_cbytes = msq->q_cbytes;
498 tbuf.msg_qnum = msq->q_qnum;
499 tbuf.msg_qbytes = msq->q_qbytes;
500 tbuf.msg_lspid = msq->q_lspid;
501 tbuf.msg_lrpid = msq->q_lrpid;
503 if (copy_msqid_to_user(buf, &tbuf, version))
505 return success_return;
510 if (copy_msqid_from_user(&setbuf, buf, version))
519 mutex_lock(&msg_ids(ns).mutex);
520 msq = msg_lock_check(ns, msqid);
528 err = audit_ipc_obj(ipcp);
531 if (cmd == IPC_SET) {
532 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
539 if (current->euid != ipcp->cuid &&
540 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
541 /* We _could_ check for CAP_CHOWN above, but we don't */
544 err = security_msg_queue_msgctl(msq, cmd);
552 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
555 msq->q_qbytes = setbuf.qbytes;
557 ipcp->uid = setbuf.uid;
558 ipcp->gid = setbuf.gid;
559 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
560 (S_IRWXUGO & setbuf.mode);
561 msq->q_ctime = get_seconds();
562 /* sleeping receivers might be excluded by
563 * stricter permissions.
565 expunge_all(msq, -EAGAIN);
566 /* sleeping senders might be able to send
567 * due to a larger queue size.
569 ss_wakeup(&msq->q_senders, 0);
579 mutex_unlock(&msg_ids(ns).mutex);
589 static int testmsg(struct msg_msg *msg, long type, int mode)
595 case SEARCH_LESSEQUAL:
596 if (msg->m_type <=type)
600 if (msg->m_type == type)
603 case SEARCH_NOTEQUAL:
604 if (msg->m_type != type)
611 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
613 struct list_head *tmp;
615 tmp = msq->q_receivers.next;
616 while (tmp != &msq->q_receivers) {
617 struct msg_receiver *msr;
619 msr = list_entry(tmp, struct msg_receiver, r_list);
621 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
622 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
623 msr->r_msgtype, msr->r_mode)) {
625 list_del(&msr->r_list);
626 if (msr->r_maxsize < msg->m_ts) {
628 wake_up_process(msr->r_tsk);
630 msr->r_msg = ERR_PTR(-E2BIG);
633 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
634 msq->q_rtime = get_seconds();
635 wake_up_process(msr->r_tsk);
646 long do_msgsnd(int msqid, long mtype, void __user *mtext,
647 size_t msgsz, int msgflg)
649 struct msg_queue *msq;
652 struct ipc_namespace *ns;
654 ns = current->nsproxy->ipc_ns;
656 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
661 msg = load_msg(mtext, msgsz);
668 msq = msg_lock_check(ns, msqid);
678 if (ipcperms(&msq->q_perm, S_IWUGO))
679 goto out_unlock_free;
681 err = security_msg_queue_msgsnd(msq, msg, msgflg);
683 goto out_unlock_free;
685 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
686 1 + msq->q_qnum <= msq->q_qbytes) {
690 /* queue full, wait: */
691 if (msgflg & IPC_NOWAIT) {
693 goto out_unlock_free;
700 ipc_lock_by_ptr(&msq->q_perm);
702 if (msq->q_perm.deleted) {
704 goto out_unlock_free;
708 if (signal_pending(current)) {
709 err = -ERESTARTNOHAND;
710 goto out_unlock_free;
714 msq->q_lspid = task_tgid_vnr(current);
715 msq->q_stime = get_seconds();
717 if (!pipelined_send(msq, msg)) {
718 /* noone is waiting for this message, enqueue it */
719 list_add_tail(&msg->m_list, &msq->q_messages);
720 msq->q_cbytes += msgsz;
722 atomic_add(msgsz, &msg_bytes);
723 atomic_inc(&msg_hdrs);
738 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
742 if (get_user(mtype, &msgp->mtype))
744 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
747 static inline int convert_mode(long *msgtyp, int msgflg)
750 * find message of correct type.
751 * msgtyp = 0 => get first.
752 * msgtyp > 0 => get first message of matching type.
753 * msgtyp < 0 => get message with least type must be < abs(msgtype).
759 return SEARCH_LESSEQUAL;
761 if (msgflg & MSG_EXCEPT)
762 return SEARCH_NOTEQUAL;
766 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
767 size_t msgsz, long msgtyp, int msgflg)
769 struct msg_queue *msq;
772 struct ipc_namespace *ns;
774 if (msqid < 0 || (long) msgsz < 0)
776 mode = convert_mode(&msgtyp, msgflg);
777 ns = current->nsproxy->ipc_ns;
779 msq = msg_lock_check(ns, msqid);
784 struct msg_receiver msr_d;
785 struct list_head *tmp;
787 msg = ERR_PTR(-EACCES);
788 if (ipcperms(&msq->q_perm, S_IRUGO))
791 msg = ERR_PTR(-EAGAIN);
792 tmp = msq->q_messages.next;
793 while (tmp != &msq->q_messages) {
794 struct msg_msg *walk_msg;
796 walk_msg = list_entry(tmp, struct msg_msg, m_list);
797 if (testmsg(walk_msg, msgtyp, mode) &&
798 !security_msg_queue_msgrcv(msq, walk_msg, current,
802 if (mode == SEARCH_LESSEQUAL &&
803 walk_msg->m_type != 1) {
805 msgtyp = walk_msg->m_type - 1;
815 * Found a suitable message.
816 * Unlink it from the queue.
818 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
819 msg = ERR_PTR(-E2BIG);
822 list_del(&msg->m_list);
824 msq->q_rtime = get_seconds();
825 msq->q_lrpid = task_tgid_vnr(current);
826 msq->q_cbytes -= msg->m_ts;
827 atomic_sub(msg->m_ts, &msg_bytes);
828 atomic_dec(&msg_hdrs);
829 ss_wakeup(&msq->q_senders, 0);
833 /* No message waiting. Wait for a message */
834 if (msgflg & IPC_NOWAIT) {
835 msg = ERR_PTR(-ENOMSG);
838 list_add_tail(&msr_d.r_list, &msq->q_receivers);
839 msr_d.r_tsk = current;
840 msr_d.r_msgtype = msgtyp;
842 if (msgflg & MSG_NOERROR)
843 msr_d.r_maxsize = INT_MAX;
845 msr_d.r_maxsize = msgsz;
846 msr_d.r_msg = ERR_PTR(-EAGAIN);
847 current->state = TASK_INTERRUPTIBLE;
852 /* Lockless receive, part 1:
853 * Disable preemption. We don't hold a reference to the queue
854 * and getting a reference would defeat the idea of a lockless
855 * operation, thus the code relies on rcu to guarantee the
857 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
858 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
859 * rcu_read_lock() prevents preemption between reading r_msg
860 * and the spin_lock() inside ipc_lock_by_ptr().
864 /* Lockless receive, part 2:
865 * Wait until pipelined_send or expunge_all are outside of
866 * wake_up_process(). There is a race with exit(), see
867 * ipc/mqueue.c for the details.
869 msg = (struct msg_msg*)msr_d.r_msg;
870 while (msg == NULL) {
872 msg = (struct msg_msg *)msr_d.r_msg;
875 /* Lockless receive, part 3:
876 * If there is a message or an error then accept it without
879 if (msg != ERR_PTR(-EAGAIN)) {
884 /* Lockless receive, part 3:
885 * Acquire the queue spinlock.
887 ipc_lock_by_ptr(&msq->q_perm);
890 /* Lockless receive, part 4:
891 * Repeat test after acquiring the spinlock.
893 msg = (struct msg_msg*)msr_d.r_msg;
894 if (msg != ERR_PTR(-EAGAIN))
897 list_del(&msr_d.r_list);
898 if (signal_pending(current)) {
899 msg = ERR_PTR(-ERESTARTNOHAND);
908 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
909 *pmtype = msg->m_type;
910 if (store_msg(mtext, msg, msgsz))
918 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
919 long msgtyp, int msgflg)
923 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
927 if (put_user(mtype, &msgp->mtype))
933 #ifdef CONFIG_PROC_FS
934 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
936 struct msg_queue *msq = it;
939 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",