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>
31 #include <linux/proc_fs.h>
32 #include <linux/list.h>
33 #include <linux/security.h>
34 #include <linux/sched.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/ipc_namespace.h>
42 #include <asm/current.h>
43 #include <asm/uaccess.h>
47 * one msg_receiver structure for each sleeping receiver:
50 struct list_head r_list;
51 struct task_struct *r_tsk;
57 struct msg_msg *volatile r_msg;
60 /* one msg_sender for each sleeping sender */
62 struct list_head list;
63 struct task_struct *tsk;
67 #define SEARCH_EQUAL 2
68 #define SEARCH_NOTEQUAL 3
69 #define SEARCH_LESSEQUAL 4
71 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
73 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
75 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
76 static int newque(struct ipc_namespace *, struct ipc_params *);
78 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
82 * Scale msgmni with the available lowmem size: the memory dedicated to msg
83 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
84 * Also take into account the number of nsproxies created so far.
85 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
87 void recompute_msgmni(struct ipc_namespace *ns)
90 unsigned long allowed;
94 allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
96 nb_ns = atomic_read(&nr_ipc_ns);
99 if (allowed < MSGMNI) {
100 ns->msg_ctlmni = MSGMNI;
104 if (allowed > IPCMNI / nb_ns) {
105 ns->msg_ctlmni = IPCMNI / nb_ns;
109 ns->msg_ctlmni = allowed;
112 void msg_init_ns(struct ipc_namespace *ns)
114 ns->msg_ctlmax = MSGMAX;
115 ns->msg_ctlmnb = MSGMNB;
117 recompute_msgmni(ns);
119 atomic_set(&ns->msg_bytes, 0);
120 atomic_set(&ns->msg_hdrs, 0);
121 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
125 void msg_exit_ns(struct ipc_namespace *ns)
127 free_ipcs(ns, &msg_ids(ns), freeque);
128 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
132 void __init msg_init(void)
134 msg_init_ns(&init_ipc_ns);
136 printk(KERN_INFO "msgmni has been set to %d\n",
137 init_ipc_ns.msg_ctlmni);
139 ipc_init_proc_interface("sysvipc/msg",
140 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
141 IPC_MSG_IDS, sysvipc_msg_proc_show);
145 * msg_lock_(check_) routines are called in the paths where the rw_mutex
148 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
150 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
153 return (struct msg_queue *)ipcp;
155 return container_of(ipcp, struct msg_queue, q_perm);
158 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
161 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
164 return (struct msg_queue *)ipcp;
166 return container_of(ipcp, struct msg_queue, q_perm);
169 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
171 ipc_rmid(&msg_ids(ns), &s->q_perm);
175 * newque - Create a new msg queue
177 * @params: ptr to the structure that contains the key and msgflg
179 * Called with msg_ids.rw_mutex held (writer)
181 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
183 struct msg_queue *msq;
185 key_t key = params->key;
186 int msgflg = params->flg;
188 msq = ipc_rcu_alloc(sizeof(*msq));
192 msq->q_perm.mode = msgflg & S_IRWXUGO;
193 msq->q_perm.key = key;
195 msq->q_perm.security = NULL;
196 retval = security_msg_queue_alloc(msq);
203 * ipc_addid() locks msq
205 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
207 security_msg_queue_free(msq);
212 msq->q_stime = msq->q_rtime = 0;
213 msq->q_ctime = get_seconds();
214 msq->q_cbytes = msq->q_qnum = 0;
215 msq->q_qbytes = ns->msg_ctlmnb;
216 msq->q_lspid = msq->q_lrpid = 0;
217 INIT_LIST_HEAD(&msq->q_messages);
218 INIT_LIST_HEAD(&msq->q_receivers);
219 INIT_LIST_HEAD(&msq->q_senders);
223 return msq->q_perm.id;
226 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
229 current->state = TASK_INTERRUPTIBLE;
230 list_add_tail(&mss->list, &msq->q_senders);
233 static inline void ss_del(struct msg_sender *mss)
235 if (mss->list.next != NULL)
236 list_del(&mss->list);
239 static void ss_wakeup(struct list_head *h, int kill)
241 struct list_head *tmp;
245 struct msg_sender *mss;
247 mss = list_entry(tmp, struct msg_sender, list);
250 mss->list.next = NULL;
251 wake_up_process(mss->tsk);
255 static void expunge_all(struct msg_queue *msq, int res)
257 struct list_head *tmp;
259 tmp = msq->q_receivers.next;
260 while (tmp != &msq->q_receivers) {
261 struct msg_receiver *msr;
263 msr = list_entry(tmp, struct msg_receiver, r_list);
266 wake_up_process(msr->r_tsk);
268 msr->r_msg = ERR_PTR(res);
273 * freeque() wakes up waiters on the sender and receiver waiting queue,
274 * removes the message queue from message queue ID IDR, and cleans up all the
275 * messages associated with this queue.
277 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
278 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
280 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
282 struct list_head *tmp;
283 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
285 expunge_all(msq, -EIDRM);
286 ss_wakeup(&msq->q_senders, 1);
290 tmp = msq->q_messages.next;
291 while (tmp != &msq->q_messages) {
292 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
295 atomic_dec(&ns->msg_hdrs);
298 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
299 security_msg_queue_free(msq);
304 * Called with msg_ids.rw_mutex and ipcp locked.
306 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
308 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
310 return security_msg_queue_associate(msq, msgflg);
313 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
315 struct ipc_namespace *ns;
316 struct ipc_ops msg_ops;
317 struct ipc_params msg_params;
319 ns = current->nsproxy->ipc_ns;
321 msg_ops.getnew = newque;
322 msg_ops.associate = msg_security;
323 msg_ops.more_checks = NULL;
325 msg_params.key = key;
326 msg_params.flg = msgflg;
328 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
331 static inline unsigned long
332 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
336 return copy_to_user(buf, in, sizeof(*in));
341 memset(&out, 0, sizeof(out));
343 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
345 out.msg_stime = in->msg_stime;
346 out.msg_rtime = in->msg_rtime;
347 out.msg_ctime = in->msg_ctime;
349 if (in->msg_cbytes > USHORT_MAX)
350 out.msg_cbytes = USHORT_MAX;
352 out.msg_cbytes = in->msg_cbytes;
353 out.msg_lcbytes = in->msg_cbytes;
355 if (in->msg_qnum > USHORT_MAX)
356 out.msg_qnum = USHORT_MAX;
358 out.msg_qnum = in->msg_qnum;
360 if (in->msg_qbytes > USHORT_MAX)
361 out.msg_qbytes = USHORT_MAX;
363 out.msg_qbytes = in->msg_qbytes;
364 out.msg_lqbytes = in->msg_qbytes;
366 out.msg_lspid = in->msg_lspid;
367 out.msg_lrpid = in->msg_lrpid;
369 return copy_to_user(buf, &out, sizeof(out));
376 static inline unsigned long
377 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
381 if (copy_from_user(out, buf, sizeof(*out)))
386 struct msqid_ds tbuf_old;
388 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
391 out->msg_perm.uid = tbuf_old.msg_perm.uid;
392 out->msg_perm.gid = tbuf_old.msg_perm.gid;
393 out->msg_perm.mode = tbuf_old.msg_perm.mode;
395 if (tbuf_old.msg_qbytes == 0)
396 out->msg_qbytes = tbuf_old.msg_lqbytes;
398 out->msg_qbytes = tbuf_old.msg_qbytes;
408 * This function handles some msgctl commands which require the rw_mutex
409 * to be held in write mode.
410 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
412 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
413 struct msqid_ds __user *buf, int version)
415 struct kern_ipc_perm *ipcp;
416 struct msqid64_ds uninitialized_var(msqid64);
417 struct msg_queue *msq;
420 if (cmd == IPC_SET) {
421 if (copy_msqid_from_user(&msqid64, buf, version))
425 ipcp = ipcctl_pre_down(&msg_ids(ns), msqid, cmd,
426 &msqid64.msg_perm, msqid64.msg_qbytes);
428 return PTR_ERR(ipcp);
430 msq = container_of(ipcp, struct msg_queue, q_perm);
432 err = security_msg_queue_msgctl(msq, cmd);
441 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
442 !capable(CAP_SYS_RESOURCE)) {
447 msq->q_qbytes = msqid64.msg_qbytes;
449 ipc_update_perm(&msqid64.msg_perm, ipcp);
450 msq->q_ctime = get_seconds();
451 /* sleeping receivers might be excluded by
452 * stricter permissions.
454 expunge_all(msq, -EAGAIN);
455 /* sleeping senders might be able to send
456 * due to a larger queue size.
458 ss_wakeup(&msq->q_senders, 0);
466 up_write(&msg_ids(ns).rw_mutex);
470 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
472 struct msg_queue *msq;
474 struct ipc_namespace *ns;
476 if (msqid < 0 || cmd < 0)
479 version = ipc_parse_version(&cmd);
480 ns = current->nsproxy->ipc_ns;
486 struct msginfo msginfo;
492 * We must not return kernel stack data.
493 * due to padding, it's not enough
494 * to set all member fields.
496 err = security_msg_queue_msgctl(NULL, cmd);
500 memset(&msginfo, 0, sizeof(msginfo));
501 msginfo.msgmni = ns->msg_ctlmni;
502 msginfo.msgmax = ns->msg_ctlmax;
503 msginfo.msgmnb = ns->msg_ctlmnb;
504 msginfo.msgssz = MSGSSZ;
505 msginfo.msgseg = MSGSEG;
506 down_read(&msg_ids(ns).rw_mutex);
507 if (cmd == MSG_INFO) {
508 msginfo.msgpool = msg_ids(ns).in_use;
509 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
510 msginfo.msgtql = atomic_read(&ns->msg_bytes);
512 msginfo.msgmap = MSGMAP;
513 msginfo.msgpool = MSGPOOL;
514 msginfo.msgtql = MSGTQL;
516 max_id = ipc_get_maxid(&msg_ids(ns));
517 up_read(&msg_ids(ns).rw_mutex);
518 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
520 return (max_id < 0) ? 0 : max_id;
522 case MSG_STAT: /* msqid is an index rather than a msg queue id */
525 struct msqid64_ds tbuf;
531 if (cmd == MSG_STAT) {
532 msq = msg_lock(ns, msqid);
535 success_return = msq->q_perm.id;
537 msq = msg_lock_check(ns, msqid);
543 if (ipcperms(&msq->q_perm, S_IRUGO))
546 err = security_msg_queue_msgctl(msq, cmd);
550 memset(&tbuf, 0, sizeof(tbuf));
552 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
553 tbuf.msg_stime = msq->q_stime;
554 tbuf.msg_rtime = msq->q_rtime;
555 tbuf.msg_ctime = msq->q_ctime;
556 tbuf.msg_cbytes = msq->q_cbytes;
557 tbuf.msg_qnum = msq->q_qnum;
558 tbuf.msg_qbytes = msq->q_qbytes;
559 tbuf.msg_lspid = msq->q_lspid;
560 tbuf.msg_lrpid = msq->q_lrpid;
562 if (copy_msqid_to_user(buf, &tbuf, version))
564 return success_return;
568 err = msgctl_down(ns, msqid, cmd, buf, version);
579 static int testmsg(struct msg_msg *msg, long type, int mode)
585 case SEARCH_LESSEQUAL:
586 if (msg->m_type <=type)
590 if (msg->m_type == type)
593 case SEARCH_NOTEQUAL:
594 if (msg->m_type != type)
601 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
603 struct list_head *tmp;
605 tmp = msq->q_receivers.next;
606 while (tmp != &msq->q_receivers) {
607 struct msg_receiver *msr;
609 msr = list_entry(tmp, struct msg_receiver, r_list);
611 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
612 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
613 msr->r_msgtype, msr->r_mode)) {
615 list_del(&msr->r_list);
616 if (msr->r_maxsize < msg->m_ts) {
618 wake_up_process(msr->r_tsk);
620 msr->r_msg = ERR_PTR(-E2BIG);
623 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
624 msq->q_rtime = get_seconds();
625 wake_up_process(msr->r_tsk);
636 long do_msgsnd(int msqid, long mtype, void __user *mtext,
637 size_t msgsz, int msgflg)
639 struct msg_queue *msq;
642 struct ipc_namespace *ns;
644 ns = current->nsproxy->ipc_ns;
646 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
651 msg = load_msg(mtext, msgsz);
658 msq = msg_lock_check(ns, msqid);
668 if (ipcperms(&msq->q_perm, S_IWUGO))
669 goto out_unlock_free;
671 err = security_msg_queue_msgsnd(msq, msg, msgflg);
673 goto out_unlock_free;
675 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
676 1 + msq->q_qnum <= msq->q_qbytes) {
680 /* queue full, wait: */
681 if (msgflg & IPC_NOWAIT) {
683 goto out_unlock_free;
690 ipc_lock_by_ptr(&msq->q_perm);
692 if (msq->q_perm.deleted) {
694 goto out_unlock_free;
698 if (signal_pending(current)) {
699 err = -ERESTARTNOHAND;
700 goto out_unlock_free;
704 msq->q_lspid = task_tgid_vnr(current);
705 msq->q_stime = get_seconds();
707 if (!pipelined_send(msq, msg)) {
708 /* noone is waiting for this message, enqueue it */
709 list_add_tail(&msg->m_list, &msq->q_messages);
710 msq->q_cbytes += msgsz;
712 atomic_add(msgsz, &ns->msg_bytes);
713 atomic_inc(&ns->msg_hdrs);
727 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
732 if (get_user(mtype, &msgp->mtype))
734 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
737 static inline int convert_mode(long *msgtyp, int msgflg)
740 * find message of correct type.
741 * msgtyp = 0 => get first.
742 * msgtyp > 0 => get first message of matching type.
743 * msgtyp < 0 => get message with least type must be < abs(msgtype).
749 return SEARCH_LESSEQUAL;
751 if (msgflg & MSG_EXCEPT)
752 return SEARCH_NOTEQUAL;
756 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
757 size_t msgsz, long msgtyp, int msgflg)
759 struct msg_queue *msq;
762 struct ipc_namespace *ns;
764 if (msqid < 0 || (long) msgsz < 0)
766 mode = convert_mode(&msgtyp, msgflg);
767 ns = current->nsproxy->ipc_ns;
769 msq = msg_lock_check(ns, msqid);
774 struct msg_receiver msr_d;
775 struct list_head *tmp;
777 msg = ERR_PTR(-EACCES);
778 if (ipcperms(&msq->q_perm, S_IRUGO))
781 msg = ERR_PTR(-EAGAIN);
782 tmp = msq->q_messages.next;
783 while (tmp != &msq->q_messages) {
784 struct msg_msg *walk_msg;
786 walk_msg = list_entry(tmp, struct msg_msg, m_list);
787 if (testmsg(walk_msg, msgtyp, mode) &&
788 !security_msg_queue_msgrcv(msq, walk_msg, current,
792 if (mode == SEARCH_LESSEQUAL &&
793 walk_msg->m_type != 1) {
795 msgtyp = walk_msg->m_type - 1;
805 * Found a suitable message.
806 * Unlink it from the queue.
808 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
809 msg = ERR_PTR(-E2BIG);
812 list_del(&msg->m_list);
814 msq->q_rtime = get_seconds();
815 msq->q_lrpid = task_tgid_vnr(current);
816 msq->q_cbytes -= msg->m_ts;
817 atomic_sub(msg->m_ts, &ns->msg_bytes);
818 atomic_dec(&ns->msg_hdrs);
819 ss_wakeup(&msq->q_senders, 0);
823 /* No message waiting. Wait for a message */
824 if (msgflg & IPC_NOWAIT) {
825 msg = ERR_PTR(-ENOMSG);
828 list_add_tail(&msr_d.r_list, &msq->q_receivers);
829 msr_d.r_tsk = current;
830 msr_d.r_msgtype = msgtyp;
832 if (msgflg & MSG_NOERROR)
833 msr_d.r_maxsize = INT_MAX;
835 msr_d.r_maxsize = msgsz;
836 msr_d.r_msg = ERR_PTR(-EAGAIN);
837 current->state = TASK_INTERRUPTIBLE;
842 /* Lockless receive, part 1:
843 * Disable preemption. We don't hold a reference to the queue
844 * and getting a reference would defeat the idea of a lockless
845 * operation, thus the code relies on rcu to guarantee the
847 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
848 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
849 * rcu_read_lock() prevents preemption between reading r_msg
850 * and the spin_lock() inside ipc_lock_by_ptr().
854 /* Lockless receive, part 2:
855 * Wait until pipelined_send or expunge_all are outside of
856 * wake_up_process(). There is a race with exit(), see
857 * ipc/mqueue.c for the details.
859 msg = (struct msg_msg*)msr_d.r_msg;
860 while (msg == NULL) {
862 msg = (struct msg_msg *)msr_d.r_msg;
865 /* Lockless receive, part 3:
866 * If there is a message or an error then accept it without
869 if (msg != ERR_PTR(-EAGAIN)) {
874 /* Lockless receive, part 3:
875 * Acquire the queue spinlock.
877 ipc_lock_by_ptr(&msq->q_perm);
880 /* Lockless receive, part 4:
881 * Repeat test after acquiring the spinlock.
883 msg = (struct msg_msg*)msr_d.r_msg;
884 if (msg != ERR_PTR(-EAGAIN))
887 list_del(&msr_d.r_list);
888 if (signal_pending(current)) {
889 msg = ERR_PTR(-ERESTARTNOHAND);
898 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
899 *pmtype = msg->m_type;
900 if (store_msg(mtext, msg, msgsz))
908 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
909 long, msgtyp, int, msgflg)
913 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
917 if (put_user(mtype, &msgp->mtype))
923 #ifdef CONFIG_PROC_FS
924 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
926 struct msg_queue *msq = it;
929 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",