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/msg.h>
27 #include <linux/spinlock.h>
28 #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>
39 #include <linux/ipc_namespace.h>
41 #include <asm/current.h>
42 #include <asm/uaccess.h>
46 * one msg_receiver structure for each sleeping receiver:
49 struct list_head r_list;
50 struct task_struct *r_tsk;
56 struct msg_msg *volatile r_msg;
59 /* one msg_sender for each sleeping sender */
61 struct list_head list;
62 struct task_struct *tsk;
66 #define SEARCH_EQUAL 2
67 #define SEARCH_NOTEQUAL 3
68 #define SEARCH_LESSEQUAL 4
70 #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
72 #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
74 static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
75 static int newque(struct ipc_namespace *, struct ipc_params *);
77 static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
81 * Scale msgmni with the available lowmem size: the memory dedicated to msg
82 * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
83 * Also take into account the number of nsproxies created so far.
84 * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
86 void recompute_msgmni(struct ipc_namespace *ns)
89 unsigned long allowed;
93 allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
95 nb_ns = atomic_read(&nr_ipc_ns);
98 if (allowed < MSGMNI) {
99 ns->msg_ctlmni = MSGMNI;
103 if (allowed > IPCMNI / nb_ns) {
104 ns->msg_ctlmni = IPCMNI / nb_ns;
108 ns->msg_ctlmni = allowed;
111 void msg_init_ns(struct ipc_namespace *ns)
113 ns->msg_ctlmax = MSGMAX;
114 ns->msg_ctlmnb = MSGMNB;
116 recompute_msgmni(ns);
118 atomic_set(&ns->msg_bytes, 0);
119 atomic_set(&ns->msg_hdrs, 0);
120 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
124 void msg_exit_ns(struct ipc_namespace *ns)
126 free_ipcs(ns, &msg_ids(ns), freeque);
127 idr_destroy(&ns->ids[IPC_MSG_IDS].ipcs_idr);
131 void __init msg_init(void)
133 msg_init_ns(&init_ipc_ns);
135 printk(KERN_INFO "msgmni has been set to %d\n",
136 init_ipc_ns.msg_ctlmni);
138 ipc_init_proc_interface("sysvipc/msg",
139 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
140 IPC_MSG_IDS, sysvipc_msg_proc_show);
144 * msg_lock_(check_) routines are called in the paths where the rw_mutex
147 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
149 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
152 return (struct msg_queue *)ipcp;
154 return container_of(ipcp, struct msg_queue, q_perm);
157 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
160 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
163 return (struct msg_queue *)ipcp;
165 return container_of(ipcp, struct msg_queue, q_perm);
168 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
170 ipc_rmid(&msg_ids(ns), &s->q_perm);
174 * newque - Create a new msg queue
176 * @params: ptr to the structure that contains the key and msgflg
178 * Called with msg_ids.rw_mutex held (writer)
180 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
182 struct msg_queue *msq;
184 key_t key = params->key;
185 int msgflg = params->flg;
187 msq = ipc_rcu_alloc(sizeof(*msq));
191 msq->q_perm.mode = msgflg & S_IRWXUGO;
192 msq->q_perm.key = key;
194 msq->q_perm.security = NULL;
195 retval = security_msg_queue_alloc(msq);
202 * ipc_addid() locks msq
204 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
206 security_msg_queue_free(msq);
211 msq->q_stime = msq->q_rtime = 0;
212 msq->q_ctime = get_seconds();
213 msq->q_cbytes = msq->q_qnum = 0;
214 msq->q_qbytes = ns->msg_ctlmnb;
215 msq->q_lspid = msq->q_lrpid = 0;
216 INIT_LIST_HEAD(&msq->q_messages);
217 INIT_LIST_HEAD(&msq->q_receivers);
218 INIT_LIST_HEAD(&msq->q_senders);
222 return msq->q_perm.id;
225 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
228 current->state = TASK_INTERRUPTIBLE;
229 list_add_tail(&mss->list, &msq->q_senders);
232 static inline void ss_del(struct msg_sender *mss)
234 if (mss->list.next != NULL)
235 list_del(&mss->list);
238 static void ss_wakeup(struct list_head *h, int kill)
240 struct list_head *tmp;
244 struct msg_sender *mss;
246 mss = list_entry(tmp, struct msg_sender, list);
249 mss->list.next = NULL;
250 wake_up_process(mss->tsk);
254 static void expunge_all(struct msg_queue *msq, int res)
256 struct list_head *tmp;
258 tmp = msq->q_receivers.next;
259 while (tmp != &msq->q_receivers) {
260 struct msg_receiver *msr;
262 msr = list_entry(tmp, struct msg_receiver, r_list);
265 wake_up_process(msr->r_tsk);
267 msr->r_msg = ERR_PTR(res);
272 * freeque() wakes up waiters on the sender and receiver waiting queue,
273 * removes the message queue from message queue ID IDR, and cleans up all the
274 * messages associated with this queue.
276 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
277 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
279 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
281 struct list_head *tmp;
282 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
284 expunge_all(msq, -EIDRM);
285 ss_wakeup(&msq->q_senders, 1);
289 tmp = msq->q_messages.next;
290 while (tmp != &msq->q_messages) {
291 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
294 atomic_dec(&ns->msg_hdrs);
297 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
298 security_msg_queue_free(msq);
303 * Called with msg_ids.rw_mutex and ipcp locked.
305 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
307 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
309 return security_msg_queue_associate(msq, msgflg);
312 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
314 struct ipc_namespace *ns;
315 struct ipc_ops msg_ops;
316 struct ipc_params msg_params;
318 ns = current->nsproxy->ipc_ns;
320 msg_ops.getnew = newque;
321 msg_ops.associate = msg_security;
322 msg_ops.more_checks = NULL;
324 msg_params.key = key;
325 msg_params.flg = msgflg;
327 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
330 static inline unsigned long
331 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
335 return copy_to_user(buf, in, sizeof(*in));
340 memset(&out, 0, sizeof(out));
342 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
344 out.msg_stime = in->msg_stime;
345 out.msg_rtime = in->msg_rtime;
346 out.msg_ctime = in->msg_ctime;
348 if (in->msg_cbytes > USHRT_MAX)
349 out.msg_cbytes = USHRT_MAX;
351 out.msg_cbytes = in->msg_cbytes;
352 out.msg_lcbytes = in->msg_cbytes;
354 if (in->msg_qnum > USHRT_MAX)
355 out.msg_qnum = USHRT_MAX;
357 out.msg_qnum = in->msg_qnum;
359 if (in->msg_qbytes > USHRT_MAX)
360 out.msg_qbytes = USHRT_MAX;
362 out.msg_qbytes = in->msg_qbytes;
363 out.msg_lqbytes = in->msg_qbytes;
365 out.msg_lspid = in->msg_lspid;
366 out.msg_lrpid = in->msg_lrpid;
368 return copy_to_user(buf, &out, sizeof(out));
375 static inline unsigned long
376 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
380 if (copy_from_user(out, buf, sizeof(*out)))
385 struct msqid_ds tbuf_old;
387 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
390 out->msg_perm.uid = tbuf_old.msg_perm.uid;
391 out->msg_perm.gid = tbuf_old.msg_perm.gid;
392 out->msg_perm.mode = tbuf_old.msg_perm.mode;
394 if (tbuf_old.msg_qbytes == 0)
395 out->msg_qbytes = tbuf_old.msg_lqbytes;
397 out->msg_qbytes = tbuf_old.msg_qbytes;
407 * This function handles some msgctl commands which require the rw_mutex
408 * to be held in write mode.
409 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
411 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
412 struct msqid_ds __user *buf, int version)
414 struct kern_ipc_perm *ipcp;
415 struct msqid64_ds uninitialized_var(msqid64);
416 struct msg_queue *msq;
419 if (cmd == IPC_SET) {
420 if (copy_msqid_from_user(&msqid64, buf, version))
424 ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
425 &msqid64.msg_perm, msqid64.msg_qbytes);
427 return PTR_ERR(ipcp);
429 msq = container_of(ipcp, struct msg_queue, q_perm);
431 err = security_msg_queue_msgctl(msq, cmd);
440 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
441 !capable(CAP_SYS_RESOURCE)) {
446 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
450 msq->q_qbytes = msqid64.msg_qbytes;
452 msq->q_ctime = get_seconds();
453 /* sleeping receivers might be excluded by
454 * stricter permissions.
456 expunge_all(msq, -EAGAIN);
457 /* sleeping senders might be able to send
458 * due to a larger queue size.
460 ss_wakeup(&msq->q_senders, 0);
468 up_write(&msg_ids(ns).rw_mutex);
472 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
474 struct msg_queue *msq;
476 struct ipc_namespace *ns;
478 if (msqid < 0 || cmd < 0)
481 version = ipc_parse_version(&cmd);
482 ns = current->nsproxy->ipc_ns;
488 struct msginfo msginfo;
494 * We must not return kernel stack data.
495 * due to padding, it's not enough
496 * to set all member fields.
498 err = security_msg_queue_msgctl(NULL, cmd);
502 memset(&msginfo, 0, sizeof(msginfo));
503 msginfo.msgmni = ns->msg_ctlmni;
504 msginfo.msgmax = ns->msg_ctlmax;
505 msginfo.msgmnb = ns->msg_ctlmnb;
506 msginfo.msgssz = MSGSSZ;
507 msginfo.msgseg = MSGSEG;
508 down_read(&msg_ids(ns).rw_mutex);
509 if (cmd == MSG_INFO) {
510 msginfo.msgpool = msg_ids(ns).in_use;
511 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
512 msginfo.msgtql = atomic_read(&ns->msg_bytes);
514 msginfo.msgmap = MSGMAP;
515 msginfo.msgpool = MSGPOOL;
516 msginfo.msgtql = MSGTQL;
518 max_id = ipc_get_maxid(&msg_ids(ns));
519 up_read(&msg_ids(ns).rw_mutex);
520 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
522 return (max_id < 0) ? 0 : max_id;
524 case MSG_STAT: /* msqid is an index rather than a msg queue id */
527 struct msqid64_ds tbuf;
533 if (cmd == MSG_STAT) {
534 msq = msg_lock(ns, msqid);
537 success_return = msq->q_perm.id;
539 msq = msg_lock_check(ns, msqid);
545 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
548 err = security_msg_queue_msgctl(msq, cmd);
552 memset(&tbuf, 0, sizeof(tbuf));
554 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
555 tbuf.msg_stime = msq->q_stime;
556 tbuf.msg_rtime = msq->q_rtime;
557 tbuf.msg_ctime = msq->q_ctime;
558 tbuf.msg_cbytes = msq->q_cbytes;
559 tbuf.msg_qnum = msq->q_qnum;
560 tbuf.msg_qbytes = msq->q_qbytes;
561 tbuf.msg_lspid = msq->q_lspid;
562 tbuf.msg_lrpid = msq->q_lrpid;
564 if (copy_msqid_to_user(buf, &tbuf, version))
566 return success_return;
570 err = msgctl_down(ns, msqid, cmd, buf, version);
581 static int testmsg(struct msg_msg *msg, long type, int mode)
587 case SEARCH_LESSEQUAL:
588 if (msg->m_type <=type)
592 if (msg->m_type == type)
595 case SEARCH_NOTEQUAL:
596 if (msg->m_type != type)
603 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
605 struct list_head *tmp;
607 tmp = msq->q_receivers.next;
608 while (tmp != &msq->q_receivers) {
609 struct msg_receiver *msr;
611 msr = list_entry(tmp, struct msg_receiver, r_list);
613 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
614 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
615 msr->r_msgtype, msr->r_mode)) {
617 list_del(&msr->r_list);
618 if (msr->r_maxsize < msg->m_ts) {
620 wake_up_process(msr->r_tsk);
622 msr->r_msg = ERR_PTR(-E2BIG);
625 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
626 msq->q_rtime = get_seconds();
627 wake_up_process(msr->r_tsk);
638 long do_msgsnd(int msqid, long mtype, void __user *mtext,
639 size_t msgsz, int msgflg)
641 struct msg_queue *msq;
644 struct ipc_namespace *ns;
646 ns = current->nsproxy->ipc_ns;
648 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
653 msg = load_msg(mtext, msgsz);
660 msq = msg_lock_check(ns, msqid);
670 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
671 goto out_unlock_free;
673 err = security_msg_queue_msgsnd(msq, msg, msgflg);
675 goto out_unlock_free;
677 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
678 1 + msq->q_qnum <= msq->q_qbytes) {
682 /* queue full, wait: */
683 if (msgflg & IPC_NOWAIT) {
685 goto out_unlock_free;
692 ipc_lock_by_ptr(&msq->q_perm);
694 if (msq->q_perm.deleted) {
696 goto out_unlock_free;
700 if (signal_pending(current)) {
701 err = -ERESTARTNOHAND;
702 goto out_unlock_free;
706 msq->q_lspid = task_tgid_vnr(current);
707 msq->q_stime = get_seconds();
709 if (!pipelined_send(msq, msg)) {
710 /* no one is waiting for this message, enqueue it */
711 list_add_tail(&msg->m_list, &msq->q_messages);
712 msq->q_cbytes += msgsz;
714 atomic_add(msgsz, &ns->msg_bytes);
715 atomic_inc(&ns->msg_hdrs);
729 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
734 if (get_user(mtype, &msgp->mtype))
736 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
739 static inline int convert_mode(long *msgtyp, int msgflg)
742 * find message of correct type.
743 * msgtyp = 0 => get first.
744 * msgtyp > 0 => get first message of matching type.
745 * msgtyp < 0 => get message with least type must be < abs(msgtype).
751 return SEARCH_LESSEQUAL;
753 if (msgflg & MSG_EXCEPT)
754 return SEARCH_NOTEQUAL;
758 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
759 size_t msgsz, long msgtyp, int msgflg)
761 struct msg_queue *msq;
764 struct ipc_namespace *ns;
766 if (msqid < 0 || (long) msgsz < 0)
768 mode = convert_mode(&msgtyp, msgflg);
769 ns = current->nsproxy->ipc_ns;
771 msq = msg_lock_check(ns, msqid);
776 struct msg_receiver msr_d;
777 struct list_head *tmp;
779 msg = ERR_PTR(-EACCES);
780 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
783 msg = ERR_PTR(-EAGAIN);
784 tmp = msq->q_messages.next;
785 while (tmp != &msq->q_messages) {
786 struct msg_msg *walk_msg;
788 walk_msg = list_entry(tmp, struct msg_msg, m_list);
789 if (testmsg(walk_msg, msgtyp, mode) &&
790 !security_msg_queue_msgrcv(msq, walk_msg, current,
794 if (mode == SEARCH_LESSEQUAL &&
795 walk_msg->m_type != 1) {
797 msgtyp = walk_msg->m_type - 1;
807 * Found a suitable message.
808 * Unlink it from the queue.
810 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
811 msg = ERR_PTR(-E2BIG);
814 list_del(&msg->m_list);
816 msq->q_rtime = get_seconds();
817 msq->q_lrpid = task_tgid_vnr(current);
818 msq->q_cbytes -= msg->m_ts;
819 atomic_sub(msg->m_ts, &ns->msg_bytes);
820 atomic_dec(&ns->msg_hdrs);
821 ss_wakeup(&msq->q_senders, 0);
825 /* No message waiting. Wait for a message */
826 if (msgflg & IPC_NOWAIT) {
827 msg = ERR_PTR(-ENOMSG);
830 list_add_tail(&msr_d.r_list, &msq->q_receivers);
831 msr_d.r_tsk = current;
832 msr_d.r_msgtype = msgtyp;
834 if (msgflg & MSG_NOERROR)
835 msr_d.r_maxsize = INT_MAX;
837 msr_d.r_maxsize = msgsz;
838 msr_d.r_msg = ERR_PTR(-EAGAIN);
839 current->state = TASK_INTERRUPTIBLE;
844 /* Lockless receive, part 1:
845 * Disable preemption. We don't hold a reference to the queue
846 * and getting a reference would defeat the idea of a lockless
847 * operation, thus the code relies on rcu to guarantee the
849 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
850 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
851 * rcu_read_lock() prevents preemption between reading r_msg
852 * and the spin_lock() inside ipc_lock_by_ptr().
856 /* Lockless receive, part 2:
857 * Wait until pipelined_send or expunge_all are outside of
858 * wake_up_process(). There is a race with exit(), see
859 * ipc/mqueue.c for the details.
861 msg = (struct msg_msg*)msr_d.r_msg;
862 while (msg == NULL) {
864 msg = (struct msg_msg *)msr_d.r_msg;
867 /* Lockless receive, part 3:
868 * If there is a message or an error then accept it without
871 if (msg != ERR_PTR(-EAGAIN)) {
876 /* Lockless receive, part 3:
877 * Acquire the queue spinlock.
879 ipc_lock_by_ptr(&msq->q_perm);
882 /* Lockless receive, part 4:
883 * Repeat test after acquiring the spinlock.
885 msg = (struct msg_msg*)msr_d.r_msg;
886 if (msg != ERR_PTR(-EAGAIN))
889 list_del(&msr_d.r_list);
890 if (signal_pending(current)) {
891 msg = ERR_PTR(-ERESTARTNOHAND);
900 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
901 *pmtype = msg->m_type;
902 if (store_msg(mtext, msg, msgsz))
910 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
911 long, msgtyp, int, msgflg)
915 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
919 if (put_user(mtype, &msgp->mtype))
925 #ifdef CONFIG_PROC_FS
926 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
928 struct user_namespace *user_ns = seq_user_ns(s);
929 struct msg_queue *msq = it;
932 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
940 from_kuid_munged(user_ns, msq->q_perm.uid),
941 from_kgid_munged(user_ns, msq->q_perm.gid),
942 from_kuid_munged(user_ns, msq->q_perm.cuid),
943 from_kgid_munged(user_ns, msq->q_perm.cgid),