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_checkid(ns, msq, msgid) \
79 ipc_checkid(&msg_ids(ns), &msq->q_perm, msgid)
80 #define msg_buildid(ns, id, seq) \
81 ipc_buildid(&msg_ids(ns), id, seq)
83 static void freeque(struct ipc_namespace *, struct msg_queue *);
84 static int newque(struct ipc_namespace *, struct ipc_params *);
86 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
89 static void __msg_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
91 ns->ids[IPC_MSG_IDS] = ids;
92 ns->msg_ctlmax = MSGMAX;
93 ns->msg_ctlmnb = MSGMNB;
94 ns->msg_ctlmni = MSGMNI;
98 int msg_init_ns(struct ipc_namespace *ns)
102 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
106 __msg_init_ns(ns, ids);
110 void msg_exit_ns(struct ipc_namespace *ns)
112 struct msg_queue *msq;
116 mutex_lock(&msg_ids(ns).mutex);
118 in_use = msg_ids(ns).in_use;
120 for (total = 0, next_id = 0; total < in_use; next_id++) {
121 msq = idr_find(&msg_ids(ns).ipcs_idr, next_id);
124 ipc_lock_by_ptr(&msq->q_perm);
128 mutex_unlock(&msg_ids(ns).mutex);
130 kfree(ns->ids[IPC_MSG_IDS]);
131 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 inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
144 ipc_rmid(&msg_ids(ns), &s->q_perm);
147 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
149 struct msg_queue *msq;
151 key_t key = params->key;
152 int msgflg = params->flg;
154 msq = ipc_rcu_alloc(sizeof(*msq));
158 msq->q_perm.mode = msgflg & S_IRWXUGO;
159 msq->q_perm.key = key;
161 msq->q_perm.security = NULL;
162 retval = security_msg_queue_alloc(msq);
169 * ipc_addid() locks msq
171 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
173 security_msg_queue_free(msq);
178 msq->q_perm.id = msg_buildid(ns, id, msq->q_perm.seq);
179 msq->q_stime = msq->q_rtime = 0;
180 msq->q_ctime = get_seconds();
181 msq->q_cbytes = msq->q_qnum = 0;
182 msq->q_qbytes = ns->msg_ctlmnb;
183 msq->q_lspid = msq->q_lrpid = 0;
184 INIT_LIST_HEAD(&msq->q_messages);
185 INIT_LIST_HEAD(&msq->q_receivers);
186 INIT_LIST_HEAD(&msq->q_senders);
190 return msq->q_perm.id;
193 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
196 current->state = TASK_INTERRUPTIBLE;
197 list_add_tail(&mss->list, &msq->q_senders);
200 static inline void ss_del(struct msg_sender *mss)
202 if (mss->list.next != NULL)
203 list_del(&mss->list);
206 static void ss_wakeup(struct list_head *h, int kill)
208 struct list_head *tmp;
212 struct msg_sender *mss;
214 mss = list_entry(tmp, struct msg_sender, list);
217 mss->list.next = NULL;
218 wake_up_process(mss->tsk);
222 static void expunge_all(struct msg_queue *msq, int res)
224 struct list_head *tmp;
226 tmp = msq->q_receivers.next;
227 while (tmp != &msq->q_receivers) {
228 struct msg_receiver *msr;
230 msr = list_entry(tmp, struct msg_receiver, r_list);
233 wake_up_process(msr->r_tsk);
235 msr->r_msg = ERR_PTR(res);
240 * freeque() wakes up waiters on the sender and receiver waiting queue,
241 * removes the message queue from message queue ID
242 * IDR, and cleans up all the messages associated with this queue.
244 * msg_ids.mutex and the spinlock for this message queue are held
245 * before freeque() is called. msg_ids.mutex remains locked on exit.
247 static void freeque(struct ipc_namespace *ns, struct msg_queue *msq)
249 struct list_head *tmp;
251 expunge_all(msq, -EIDRM);
252 ss_wakeup(&msq->q_senders, 1);
256 tmp = msq->q_messages.next;
257 while (tmp != &msq->q_messages) {
258 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
261 atomic_dec(&msg_hdrs);
264 atomic_sub(msq->q_cbytes, &msg_bytes);
265 security_msg_queue_free(msq);
269 static inline int msg_security(void *msq, int msgflg)
271 return security_msg_queue_associate((struct msg_queue *) msq, msgflg);
274 asmlinkage long sys_msgget(key_t key, int msgflg)
276 struct ipc_namespace *ns;
277 struct ipc_ops msg_ops;
278 struct ipc_params msg_params;
280 ns = current->nsproxy->ipc_ns;
282 msg_ops.getnew = newque;
283 msg_ops.associate = msg_security;
284 msg_ops.more_checks = NULL;
286 msg_params.key = key;
287 msg_params.flg = msgflg;
289 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
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 uninitialized_var(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 = ipc_get_maxid(&msg_ids(ns));
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;
439 case MSG_STAT: /* msqid is an index rather than a msg queue id */
442 struct msqid64_ds tbuf;
448 memset(&tbuf, 0, sizeof(tbuf));
450 msq = msg_lock(ns, msqid);
454 if (cmd == MSG_STAT) {
455 success_return = msq->q_perm.id;
458 if (msg_checkid(ns, msq, msqid))
463 if (ipcperms(&msq->q_perm, S_IRUGO))
466 err = security_msg_queue_msgctl(msq, cmd);
470 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
471 tbuf.msg_stime = msq->q_stime;
472 tbuf.msg_rtime = msq->q_rtime;
473 tbuf.msg_ctime = msq->q_ctime;
474 tbuf.msg_cbytes = msq->q_cbytes;
475 tbuf.msg_qnum = msq->q_qnum;
476 tbuf.msg_qbytes = msq->q_qbytes;
477 tbuf.msg_lspid = msq->q_lspid;
478 tbuf.msg_lrpid = msq->q_lrpid;
480 if (copy_msqid_to_user(buf, &tbuf, version))
482 return success_return;
487 if (copy_msqid_from_user(&setbuf, buf, version))
496 mutex_lock(&msg_ids(ns).mutex);
497 msq = msg_lock(ns, msqid);
503 if (msg_checkid(ns, msq, msqid))
507 err = audit_ipc_obj(ipcp);
510 if (cmd == IPC_SET) {
511 err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid,
518 if (current->euid != ipcp->cuid &&
519 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
520 /* We _could_ check for CAP_CHOWN above, but we don't */
523 err = security_msg_queue_msgctl(msq, cmd);
531 if (setbuf.qbytes > ns->msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
534 msq->q_qbytes = setbuf.qbytes;
536 ipcp->uid = setbuf.uid;
537 ipcp->gid = setbuf.gid;
538 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
539 (S_IRWXUGO & setbuf.mode);
540 msq->q_ctime = get_seconds();
541 /* sleeping receivers might be excluded by
542 * stricter permissions.
544 expunge_all(msq, -EAGAIN);
545 /* sleeping senders might be able to send
546 * due to a larger queue size.
548 ss_wakeup(&msq->q_senders, 0);
558 mutex_unlock(&msg_ids(ns).mutex);
568 static int testmsg(struct msg_msg *msg, long type, int mode)
574 case SEARCH_LESSEQUAL:
575 if (msg->m_type <=type)
579 if (msg->m_type == type)
582 case SEARCH_NOTEQUAL:
583 if (msg->m_type != type)
590 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
592 struct list_head *tmp;
594 tmp = msq->q_receivers.next;
595 while (tmp != &msq->q_receivers) {
596 struct msg_receiver *msr;
598 msr = list_entry(tmp, struct msg_receiver, r_list);
600 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
601 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
602 msr->r_msgtype, msr->r_mode)) {
604 list_del(&msr->r_list);
605 if (msr->r_maxsize < msg->m_ts) {
607 wake_up_process(msr->r_tsk);
609 msr->r_msg = ERR_PTR(-E2BIG);
612 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
613 msq->q_rtime = get_seconds();
614 wake_up_process(msr->r_tsk);
625 long do_msgsnd(int msqid, long mtype, void __user *mtext,
626 size_t msgsz, int msgflg)
628 struct msg_queue *msq;
631 struct ipc_namespace *ns;
633 ns = current->nsproxy->ipc_ns;
635 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
640 msg = load_msg(mtext, msgsz);
647 msq = msg_lock(ns, msqid);
653 if (msg_checkid(ns, msq, msqid))
654 goto out_unlock_free;
660 if (ipcperms(&msq->q_perm, S_IWUGO))
661 goto out_unlock_free;
663 err = security_msg_queue_msgsnd(msq, msg, msgflg);
665 goto out_unlock_free;
667 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
668 1 + msq->q_qnum <= msq->q_qbytes) {
672 /* queue full, wait: */
673 if (msgflg & IPC_NOWAIT) {
675 goto out_unlock_free;
682 ipc_lock_by_ptr(&msq->q_perm);
684 if (msq->q_perm.deleted) {
686 goto out_unlock_free;
690 if (signal_pending(current)) {
691 err = -ERESTARTNOHAND;
692 goto out_unlock_free;
696 msq->q_lspid = task_tgid_vnr(current);
697 msq->q_stime = get_seconds();
699 if (!pipelined_send(msq, msg)) {
700 /* noone is waiting for this message, enqueue it */
701 list_add_tail(&msg->m_list, &msq->q_messages);
702 msq->q_cbytes += msgsz;
704 atomic_add(msgsz, &msg_bytes);
705 atomic_inc(&msg_hdrs);
720 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
724 if (get_user(mtype, &msgp->mtype))
726 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
729 static inline int convert_mode(long *msgtyp, int msgflg)
732 * find message of correct type.
733 * msgtyp = 0 => get first.
734 * msgtyp > 0 => get first message of matching type.
735 * msgtyp < 0 => get message with least type must be < abs(msgtype).
741 return SEARCH_LESSEQUAL;
743 if (msgflg & MSG_EXCEPT)
744 return SEARCH_NOTEQUAL;
748 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
749 size_t msgsz, long msgtyp, int msgflg)
751 struct msg_queue *msq;
754 struct ipc_namespace *ns;
756 if (msqid < 0 || (long) msgsz < 0)
758 mode = convert_mode(&msgtyp, msgflg);
759 ns = current->nsproxy->ipc_ns;
761 msq = msg_lock(ns, msqid);
765 msg = ERR_PTR(-EIDRM);
766 if (msg_checkid(ns, msq, msqid))
770 struct msg_receiver msr_d;
771 struct list_head *tmp;
773 msg = ERR_PTR(-EACCES);
774 if (ipcperms(&msq->q_perm, S_IRUGO))
777 msg = ERR_PTR(-EAGAIN);
778 tmp = msq->q_messages.next;
779 while (tmp != &msq->q_messages) {
780 struct msg_msg *walk_msg;
782 walk_msg = list_entry(tmp, struct msg_msg, m_list);
783 if (testmsg(walk_msg, msgtyp, mode) &&
784 !security_msg_queue_msgrcv(msq, walk_msg, current,
788 if (mode == SEARCH_LESSEQUAL &&
789 walk_msg->m_type != 1) {
791 msgtyp = walk_msg->m_type - 1;
801 * Found a suitable message.
802 * Unlink it from the queue.
804 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
805 msg = ERR_PTR(-E2BIG);
808 list_del(&msg->m_list);
810 msq->q_rtime = get_seconds();
811 msq->q_lrpid = task_tgid_vnr(current);
812 msq->q_cbytes -= msg->m_ts;
813 atomic_sub(msg->m_ts, &msg_bytes);
814 atomic_dec(&msg_hdrs);
815 ss_wakeup(&msq->q_senders, 0);
819 /* No message waiting. Wait for a message */
820 if (msgflg & IPC_NOWAIT) {
821 msg = ERR_PTR(-ENOMSG);
824 list_add_tail(&msr_d.r_list, &msq->q_receivers);
825 msr_d.r_tsk = current;
826 msr_d.r_msgtype = msgtyp;
828 if (msgflg & MSG_NOERROR)
829 msr_d.r_maxsize = INT_MAX;
831 msr_d.r_maxsize = msgsz;
832 msr_d.r_msg = ERR_PTR(-EAGAIN);
833 current->state = TASK_INTERRUPTIBLE;
838 /* Lockless receive, part 1:
839 * Disable preemption. We don't hold a reference to the queue
840 * and getting a reference would defeat the idea of a lockless
841 * operation, thus the code relies on rcu to guarantee the
843 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
844 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
845 * rcu_read_lock() prevents preemption between reading r_msg
846 * and the spin_lock() inside ipc_lock_by_ptr().
850 /* Lockless receive, part 2:
851 * Wait until pipelined_send or expunge_all are outside of
852 * wake_up_process(). There is a race with exit(), see
853 * ipc/mqueue.c for the details.
855 msg = (struct msg_msg*)msr_d.r_msg;
856 while (msg == NULL) {
858 msg = (struct msg_msg *)msr_d.r_msg;
861 /* Lockless receive, part 3:
862 * If there is a message or an error then accept it without
865 if (msg != ERR_PTR(-EAGAIN)) {
870 /* Lockless receive, part 3:
871 * Acquire the queue spinlock.
873 ipc_lock_by_ptr(&msq->q_perm);
876 /* Lockless receive, part 4:
877 * Repeat test after acquiring the spinlock.
879 msg = (struct msg_msg*)msr_d.r_msg;
880 if (msg != ERR_PTR(-EAGAIN))
883 list_del(&msr_d.r_list);
884 if (signal_pending(current)) {
885 msg = ERR_PTR(-ERESTARTNOHAND);
894 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
895 *pmtype = msg->m_type;
896 if (store_msg(mtext, msg, msgsz))
904 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
905 long msgtyp, int msgflg)
909 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
913 if (put_user(mtype, &msgp->mtype))
919 #ifdef CONFIG_PROC_FS
920 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
922 struct msg_queue *msq = it;
925 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",