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>
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(id, seq) ipc_buildid(id, seq)
79 static void freeque(struct ipc_namespace *, struct msg_queue *);
80 static int newque(struct ipc_namespace *, struct ipc_params *);
82 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
85 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
87 ns->ids[IPC_MSG_IDS] = ids;
88 ns->msg_ctlmax = MSGMAX;
89 ns->msg_ctlmnb = MSGMNB;
90 ns->msg_ctlmni = MSGMNI;
94 int msg_init_ns(struct ipc_namespace *ns)
98 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
102 __msg_init_ns(ns, ids);
106 void msg_exit_ns(struct ipc_namespace *ns)
108 struct msg_queue *msq;
112 down_write(&msg_ids(ns).rw_mutex);
114 in_use = msg_ids(ns).in_use;
116 for (total = 0, next_id = 0; total < in_use; next_id++) {
117 msq = idr_find(&msg_ids(ns).ipcs_idr, next_id);
120 ipc_lock_by_ptr(&msq->q_perm);
125 up_write(&msg_ids(ns).rw_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);
140 * This routine is called in the paths where the rw_mutex is held to protect
141 * access to the idr tree.
143 static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
146 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
148 return container_of(ipcp, struct msg_queue, q_perm);
152 * msg_lock_(check_) routines are called in the paths where the rw_mutex
155 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
157 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
159 return container_of(ipcp, struct msg_queue, q_perm);
162 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
165 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
167 return container_of(ipcp, struct msg_queue, q_perm);
170 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
172 ipc_rmid(&msg_ids(ns), &s->q_perm);
176 * newque - Create a new msg queue
178 * @params: ptr to the structure that contains the key and msgflg
180 * Called with msg_ids.rw_mutex held (writer)
182 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
184 struct msg_queue *msq;
186 key_t key = params->key;
187 int msgflg = params->flg;
189 msq = ipc_rcu_alloc(sizeof(*msq));
193 msq->q_perm.mode = msgflg & S_IRWXUGO;
194 msq->q_perm.key = key;
196 msq->q_perm.security = NULL;
197 retval = security_msg_queue_alloc(msq);
204 * ipc_addid() locks msq
206 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
208 security_msg_queue_free(msq);
213 msq->q_perm.id = msg_buildid(id, msq->q_perm.seq);
214 msq->q_stime = msq->q_rtime = 0;
215 msq->q_ctime = get_seconds();
216 msq->q_cbytes = msq->q_qnum = 0;
217 msq->q_qbytes = ns->msg_ctlmnb;
218 msq->q_lspid = msq->q_lrpid = 0;
219 INIT_LIST_HEAD(&msq->q_messages);
220 INIT_LIST_HEAD(&msq->q_receivers);
221 INIT_LIST_HEAD(&msq->q_senders);
225 return msq->q_perm.id;
228 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
231 current->state = TASK_INTERRUPTIBLE;
232 list_add_tail(&mss->list, &msq->q_senders);
235 static inline void ss_del(struct msg_sender *mss)
237 if (mss->list.next != NULL)
238 list_del(&mss->list);
241 static void ss_wakeup(struct list_head *h, int kill)
243 struct list_head *tmp;
247 struct msg_sender *mss;
249 mss = list_entry(tmp, struct msg_sender, list);
252 mss->list.next = NULL;
253 wake_up_process(mss->tsk);
257 static void expunge_all(struct msg_queue *msq, int res)
259 struct list_head *tmp;
261 tmp = msq->q_receivers.next;
262 while (tmp != &msq->q_receivers) {
263 struct msg_receiver *msr;
265 msr = list_entry(tmp, struct msg_receiver, r_list);
268 wake_up_process(msr->r_tsk);
270 msr->r_msg = ERR_PTR(res);
275 * freeque() wakes up waiters on the sender and receiver waiting queue,
276 * removes the message queue from message queue ID IDR, and cleans up all the
277 * messages associated with this queue.
279 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
280 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
282 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
284 struct list_head *tmp;
286 expunge_all(msq, -EIDRM);
287 ss_wakeup(&msq->q_senders, 1);
291 tmp = msq->q_messages.next;
292 while (tmp != &msq->q_messages) {
293 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
296 atomic_dec(&msg_hdrs);
299 atomic_sub(msq->q_cbytes, &msg_bytes);
300 security_msg_queue_free(msq);
305 * Called with msg_ids.rw_mutex and ipcp locked.
307 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
309 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
311 return security_msg_queue_associate(msq, msgflg);
314 asmlinkage long sys_msgget(key_t key, int msgflg)
316 struct ipc_namespace *ns;
317 struct ipc_ops msg_ops;
318 struct ipc_params msg_params;
320 ns = current->nsproxy->ipc_ns;
322 msg_ops.getnew = newque;
323 msg_ops.associate = msg_security;
324 msg_ops.more_checks = NULL;
326 msg_params.key = key;
327 msg_params.flg = msgflg;
329 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
332 static inline unsigned long
333 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
337 return copy_to_user(buf, in, sizeof(*in));
342 memset(&out, 0, sizeof(out));
344 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
346 out.msg_stime = in->msg_stime;
347 out.msg_rtime = in->msg_rtime;
348 out.msg_ctime = in->msg_ctime;
350 if (in->msg_cbytes > USHRT_MAX)
351 out.msg_cbytes = USHRT_MAX;
353 out.msg_cbytes = in->msg_cbytes;
354 out.msg_lcbytes = in->msg_cbytes;
356 if (in->msg_qnum > USHRT_MAX)
357 out.msg_qnum = USHRT_MAX;
359 out.msg_qnum = in->msg_qnum;
361 if (in->msg_qbytes > USHRT_MAX)
362 out.msg_qbytes = USHRT_MAX;
364 out.msg_qbytes = in->msg_qbytes;
365 out.msg_lqbytes = in->msg_qbytes;
367 out.msg_lspid = in->msg_lspid;
368 out.msg_lrpid = in->msg_lrpid;
370 return copy_to_user(buf, &out, sizeof(out));
378 unsigned long qbytes;
384 static inline unsigned long
385 copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
390 struct msqid64_ds tbuf;
392 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
395 out->qbytes = tbuf.msg_qbytes;
396 out->uid = tbuf.msg_perm.uid;
397 out->gid = tbuf.msg_perm.gid;
398 out->mode = tbuf.msg_perm.mode;
404 struct msqid_ds tbuf_old;
406 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
409 out->uid = tbuf_old.msg_perm.uid;
410 out->gid = tbuf_old.msg_perm.gid;
411 out->mode = tbuf_old.msg_perm.mode;
413 if (tbuf_old.msg_qbytes == 0)
414 out->qbytes = tbuf_old.msg_lqbytes;
416 out->qbytes = tbuf_old.msg_qbytes;
425 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
427 struct kern_ipc_perm *ipcp;
428 struct msq_setbuf uninitialized_var(setbuf);
429 struct msg_queue *msq;
431 struct ipc_namespace *ns;
433 if (msqid < 0 || cmd < 0)
436 version = ipc_parse_version(&cmd);
437 ns = current->nsproxy->ipc_ns;
443 struct msginfo msginfo;
449 * We must not return kernel stack data.
450 * due to padding, it's not enough
451 * to set all member fields.
453 err = security_msg_queue_msgctl(NULL, cmd);
457 memset(&msginfo, 0, sizeof(msginfo));
458 msginfo.msgmni = ns->msg_ctlmni;
459 msginfo.msgmax = ns->msg_ctlmax;
460 msginfo.msgmnb = ns->msg_ctlmnb;
461 msginfo.msgssz = MSGSSZ;
462 msginfo.msgseg = MSGSEG;
463 down_read(&msg_ids(ns).rw_mutex);
464 if (cmd == MSG_INFO) {
465 msginfo.msgpool = msg_ids(ns).in_use;
466 msginfo.msgmap = atomic_read(&msg_hdrs);
467 msginfo.msgtql = atomic_read(&msg_bytes);
469 msginfo.msgmap = MSGMAP;
470 msginfo.msgpool = MSGPOOL;
471 msginfo.msgtql = MSGTQL;
473 max_id = ipc_get_maxid(&msg_ids(ns));
474 up_read(&msg_ids(ns).rw_mutex);
475 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
477 return (max_id < 0) ? 0 : max_id;
479 case MSG_STAT: /* msqid is an index rather than a msg queue id */
482 struct msqid64_ds tbuf;
488 if (cmd == MSG_STAT) {
489 msq = msg_lock(ns, msqid);
492 success_return = msq->q_perm.id;
494 msq = msg_lock_check(ns, msqid);
500 if (ipcperms(&msq->q_perm, S_IRUGO))
503 err = security_msg_queue_msgctl(msq, cmd);
507 memset(&tbuf, 0, sizeof(tbuf));
509 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
510 tbuf.msg_stime = msq->q_stime;
511 tbuf.msg_rtime = msq->q_rtime;
512 tbuf.msg_ctime = msq->q_ctime;
513 tbuf.msg_cbytes = msq->q_cbytes;
514 tbuf.msg_qnum = msq->q_qnum;
515 tbuf.msg_qbytes = msq->q_qbytes;
516 tbuf.msg_lspid = msq->q_lspid;
517 tbuf.msg_lrpid = msq->q_lrpid;
519 if (copy_msqid_to_user(buf, &tbuf, version))
521 return success_return;
526 if (copy_msqid_from_user(&setbuf, buf, version))
535 down_write(&msg_ids(ns).rw_mutex);
536 msq = msg_lock_check_down(ns, msqid);
544 err = audit_ipc_obj(ipcp);
547 if (cmd == IPC_SET) {
548 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
555 if (current->euid != ipcp->cuid &&
556 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
557 /* We _could_ check for CAP_CHOWN above, but we don't */
560 err = security_msg_queue_msgctl(msq, cmd);
568 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
571 msq->q_qbytes = setbuf.qbytes;
573 ipcp->uid = setbuf.uid;
574 ipcp->gid = setbuf.gid;
575 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
576 (S_IRWXUGO & setbuf.mode);
577 msq->q_ctime = get_seconds();
578 /* sleeping receivers might be excluded by
579 * stricter permissions.
581 expunge_all(msq, -EAGAIN);
582 /* sleeping senders might be able to send
583 * due to a larger queue size.
585 ss_wakeup(&msq->q_senders, 0);
595 up_write(&msg_ids(ns).rw_mutex);
605 static int testmsg(struct msg_msg *msg, long type, int mode)
611 case SEARCH_LESSEQUAL:
612 if (msg->m_type <=type)
616 if (msg->m_type == type)
619 case SEARCH_NOTEQUAL:
620 if (msg->m_type != type)
627 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
629 struct list_head *tmp;
631 tmp = msq->q_receivers.next;
632 while (tmp != &msq->q_receivers) {
633 struct msg_receiver *msr;
635 msr = list_entry(tmp, struct msg_receiver, r_list);
637 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
638 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
639 msr->r_msgtype, msr->r_mode)) {
641 list_del(&msr->r_list);
642 if (msr->r_maxsize < msg->m_ts) {
644 wake_up_process(msr->r_tsk);
646 msr->r_msg = ERR_PTR(-E2BIG);
649 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
650 msq->q_rtime = get_seconds();
651 wake_up_process(msr->r_tsk);
662 long do_msgsnd(int msqid, long mtype, void __user *mtext,
663 size_t msgsz, int msgflg)
665 struct msg_queue *msq;
668 struct ipc_namespace *ns;
670 ns = current->nsproxy->ipc_ns;
672 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
677 msg = load_msg(mtext, msgsz);
684 msq = msg_lock_check(ns, msqid);
694 if (ipcperms(&msq->q_perm, S_IWUGO))
695 goto out_unlock_free;
697 err = security_msg_queue_msgsnd(msq, msg, msgflg);
699 goto out_unlock_free;
701 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
702 1 + msq->q_qnum <= msq->q_qbytes) {
706 /* queue full, wait: */
707 if (msgflg & IPC_NOWAIT) {
709 goto out_unlock_free;
716 ipc_lock_by_ptr(&msq->q_perm);
718 if (msq->q_perm.deleted) {
720 goto out_unlock_free;
724 if (signal_pending(current)) {
725 err = -ERESTARTNOHAND;
726 goto out_unlock_free;
730 msq->q_lspid = task_tgid_vnr(current);
731 msq->q_stime = get_seconds();
733 if (!pipelined_send(msq, msg)) {
734 /* noone is waiting for this message, enqueue it */
735 list_add_tail(&msg->m_list, &msq->q_messages);
736 msq->q_cbytes += msgsz;
738 atomic_add(msgsz, &msg_bytes);
739 atomic_inc(&msg_hdrs);
754 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
758 if (get_user(mtype, &msgp->mtype))
760 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
763 static inline int convert_mode(long *msgtyp, int msgflg)
766 * find message of correct type.
767 * msgtyp = 0 => get first.
768 * msgtyp > 0 => get first message of matching type.
769 * msgtyp < 0 => get message with least type must be < abs(msgtype).
775 return SEARCH_LESSEQUAL;
777 if (msgflg & MSG_EXCEPT)
778 return SEARCH_NOTEQUAL;
782 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
783 size_t msgsz, long msgtyp, int msgflg)
785 struct msg_queue *msq;
788 struct ipc_namespace *ns;
790 if (msqid < 0 || (long) msgsz < 0)
792 mode = convert_mode(&msgtyp, msgflg);
793 ns = current->nsproxy->ipc_ns;
795 msq = msg_lock_check(ns, msqid);
800 struct msg_receiver msr_d;
801 struct list_head *tmp;
803 msg = ERR_PTR(-EACCES);
804 if (ipcperms(&msq->q_perm, S_IRUGO))
807 msg = ERR_PTR(-EAGAIN);
808 tmp = msq->q_messages.next;
809 while (tmp != &msq->q_messages) {
810 struct msg_msg *walk_msg;
812 walk_msg = list_entry(tmp, struct msg_msg, m_list);
813 if (testmsg(walk_msg, msgtyp, mode) &&
814 !security_msg_queue_msgrcv(msq, walk_msg, current,
818 if (mode == SEARCH_LESSEQUAL &&
819 walk_msg->m_type != 1) {
821 msgtyp = walk_msg->m_type - 1;
831 * Found a suitable message.
832 * Unlink it from the queue.
834 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
835 msg = ERR_PTR(-E2BIG);
838 list_del(&msg->m_list);
840 msq->q_rtime = get_seconds();
841 msq->q_lrpid = task_tgid_vnr(current);
842 msq->q_cbytes -= msg->m_ts;
843 atomic_sub(msg->m_ts, &msg_bytes);
844 atomic_dec(&msg_hdrs);
845 ss_wakeup(&msq->q_senders, 0);
849 /* No message waiting. Wait for a message */
850 if (msgflg & IPC_NOWAIT) {
851 msg = ERR_PTR(-ENOMSG);
854 list_add_tail(&msr_d.r_list, &msq->q_receivers);
855 msr_d.r_tsk = current;
856 msr_d.r_msgtype = msgtyp;
858 if (msgflg & MSG_NOERROR)
859 msr_d.r_maxsize = INT_MAX;
861 msr_d.r_maxsize = msgsz;
862 msr_d.r_msg = ERR_PTR(-EAGAIN);
863 current->state = TASK_INTERRUPTIBLE;
868 /* Lockless receive, part 1:
869 * Disable preemption. We don't hold a reference to the queue
870 * and getting a reference would defeat the idea of a lockless
871 * operation, thus the code relies on rcu to guarantee the
873 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
874 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
875 * rcu_read_lock() prevents preemption between reading r_msg
876 * and the spin_lock() inside ipc_lock_by_ptr().
880 /* Lockless receive, part 2:
881 * Wait until pipelined_send or expunge_all are outside of
882 * wake_up_process(). There is a race with exit(), see
883 * ipc/mqueue.c for the details.
885 msg = (struct msg_msg*)msr_d.r_msg;
886 while (msg == NULL) {
888 msg = (struct msg_msg *)msr_d.r_msg;
891 /* Lockless receive, part 3:
892 * If there is a message or an error then accept it without
895 if (msg != ERR_PTR(-EAGAIN)) {
900 /* Lockless receive, part 3:
901 * Acquire the queue spinlock.
903 ipc_lock_by_ptr(&msq->q_perm);
906 /* Lockless receive, part 4:
907 * Repeat test after acquiring the spinlock.
909 msg = (struct msg_msg*)msr_d.r_msg;
910 if (msg != ERR_PTR(-EAGAIN))
913 list_del(&msr_d.r_list);
914 if (signal_pending(current)) {
915 msg = ERR_PTR(-ERESTARTNOHAND);
924 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
925 *pmtype = msg->m_type;
926 if (store_msg(mtext, msg, msgsz))
934 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
935 long msgtyp, int msgflg)
939 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
943 if (put_user(mtype, &msgp->mtype))
949 #ifdef CONFIG_PROC_FS
950 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
952 struct msg_queue *msq = it;
955 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",