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
69 #define SEARCH_NUMBER 5
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);
202 /* ipc_addid() locks msq upon success. */
203 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
205 security_msg_queue_free(msq);
210 msq->q_stime = msq->q_rtime = 0;
211 msq->q_ctime = get_seconds();
212 msq->q_cbytes = msq->q_qnum = 0;
213 msq->q_qbytes = ns->msg_ctlmnb;
214 msq->q_lspid = msq->q_lrpid = 0;
215 INIT_LIST_HEAD(&msq->q_messages);
216 INIT_LIST_HEAD(&msq->q_receivers);
217 INIT_LIST_HEAD(&msq->q_senders);
219 ipc_unlock_object(&msq->q_perm);
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 msg_sender *mss, *t;
242 list_for_each_entry_safe(mss, t, h, list) {
244 mss->list.next = NULL;
245 wake_up_process(mss->tsk);
249 static void expunge_all(struct msg_queue *msq, int res)
251 struct msg_receiver *msr, *t;
253 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
255 wake_up_process(msr->r_tsk);
257 msr->r_msg = ERR_PTR(res);
262 * freeque() wakes up waiters on the sender and receiver waiting queue,
263 * removes the message queue from message queue ID IDR, and cleans up all the
264 * messages associated with this queue.
266 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
267 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
269 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
271 struct msg_msg *msg, *t;
272 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
274 expunge_all(msq, -EIDRM);
275 ss_wakeup(&msq->q_senders, 1);
279 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
280 atomic_dec(&ns->msg_hdrs);
283 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
284 security_msg_queue_free(msq);
289 * Called with msg_ids.rw_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 SYSCALL_DEFINE2(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));
361 static inline unsigned long
362 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
366 if (copy_from_user(out, buf, sizeof(*out)))
371 struct msqid_ds tbuf_old;
373 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
376 out->msg_perm.uid = tbuf_old.msg_perm.uid;
377 out->msg_perm.gid = tbuf_old.msg_perm.gid;
378 out->msg_perm.mode = tbuf_old.msg_perm.mode;
380 if (tbuf_old.msg_qbytes == 0)
381 out->msg_qbytes = tbuf_old.msg_lqbytes;
383 out->msg_qbytes = tbuf_old.msg_qbytes;
393 * This function handles some msgctl commands which require the rw_mutex
394 * to be held in write mode.
395 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
397 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
398 struct msqid_ds __user *buf, int version)
400 struct kern_ipc_perm *ipcp;
401 struct msqid64_ds uninitialized_var(msqid64);
402 struct msg_queue *msq;
405 if (cmd == IPC_SET) {
406 if (copy_msqid_from_user(&msqid64, buf, version))
410 ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
411 &msqid64.msg_perm, msqid64.msg_qbytes);
413 return PTR_ERR(ipcp);
415 msq = container_of(ipcp, struct msg_queue, q_perm);
417 err = security_msg_queue_msgctl(msq, cmd);
426 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
427 !capable(CAP_SYS_RESOURCE)) {
432 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
436 msq->q_qbytes = msqid64.msg_qbytes;
438 msq->q_ctime = get_seconds();
439 /* sleeping receivers might be excluded by
440 * stricter permissions.
442 expunge_all(msq, -EAGAIN);
443 /* sleeping senders might be able to send
444 * due to a larger queue size.
446 ss_wakeup(&msq->q_senders, 0);
454 up_write(&msg_ids(ns).rw_mutex);
458 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
460 struct msg_queue *msq;
462 struct ipc_namespace *ns;
464 if (msqid < 0 || cmd < 0)
467 version = ipc_parse_version(&cmd);
468 ns = current->nsproxy->ipc_ns;
474 struct msginfo msginfo;
480 * We must not return kernel stack data.
481 * due to padding, it's not enough
482 * to set all member fields.
484 err = security_msg_queue_msgctl(NULL, cmd);
488 memset(&msginfo, 0, sizeof(msginfo));
489 msginfo.msgmni = ns->msg_ctlmni;
490 msginfo.msgmax = ns->msg_ctlmax;
491 msginfo.msgmnb = ns->msg_ctlmnb;
492 msginfo.msgssz = MSGSSZ;
493 msginfo.msgseg = MSGSEG;
494 down_read(&msg_ids(ns).rw_mutex);
495 if (cmd == MSG_INFO) {
496 msginfo.msgpool = msg_ids(ns).in_use;
497 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
498 msginfo.msgtql = atomic_read(&ns->msg_bytes);
500 msginfo.msgmap = MSGMAP;
501 msginfo.msgpool = MSGPOOL;
502 msginfo.msgtql = MSGTQL;
504 max_id = ipc_get_maxid(&msg_ids(ns));
505 up_read(&msg_ids(ns).rw_mutex);
506 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
508 return (max_id < 0) ? 0 : max_id;
510 case MSG_STAT: /* msqid is an index rather than a msg queue id */
513 struct msqid64_ds tbuf;
519 if (cmd == MSG_STAT) {
520 msq = msg_lock(ns, msqid);
523 success_return = msq->q_perm.id;
525 msq = msg_lock_check(ns, msqid);
531 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
534 err = security_msg_queue_msgctl(msq, cmd);
538 memset(&tbuf, 0, sizeof(tbuf));
540 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
541 tbuf.msg_stime = msq->q_stime;
542 tbuf.msg_rtime = msq->q_rtime;
543 tbuf.msg_ctime = msq->q_ctime;
544 tbuf.msg_cbytes = msq->q_cbytes;
545 tbuf.msg_qnum = msq->q_qnum;
546 tbuf.msg_qbytes = msq->q_qbytes;
547 tbuf.msg_lspid = msq->q_lspid;
548 tbuf.msg_lrpid = msq->q_lrpid;
550 if (copy_msqid_to_user(buf, &tbuf, version))
552 return success_return;
556 err = msgctl_down(ns, msqid, cmd, buf, version);
567 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 msg_receiver *msr, *t;
594 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
595 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
596 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
597 msr->r_msgtype, msr->r_mode)) {
599 list_del(&msr->r_list);
600 if (msr->r_maxsize < msg->m_ts) {
602 wake_up_process(msr->r_tsk);
604 msr->r_msg = ERR_PTR(-E2BIG);
607 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
608 msq->q_rtime = get_seconds();
609 wake_up_process(msr->r_tsk);
620 long do_msgsnd(int msqid, long mtype, void __user *mtext,
621 size_t msgsz, int msgflg)
623 struct msg_queue *msq;
626 struct ipc_namespace *ns;
628 ns = current->nsproxy->ipc_ns;
630 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
635 msg = load_msg(mtext, msgsz);
642 msq = msg_lock_check(ns, msqid);
652 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
653 goto out_unlock_free;
655 err = security_msg_queue_msgsnd(msq, msg, msgflg);
657 goto out_unlock_free;
659 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
660 1 + msq->q_qnum <= msq->q_qbytes) {
664 /* queue full, wait: */
665 if (msgflg & IPC_NOWAIT) {
667 goto out_unlock_free;
671 if (!ipc_rcu_getref(msq)) {
673 goto out_unlock_free;
679 ipc_lock_by_ptr(&msq->q_perm);
681 if (msq->q_perm.deleted) {
683 goto out_unlock_free;
687 if (signal_pending(current)) {
688 err = -ERESTARTNOHAND;
689 goto out_unlock_free;
693 msq->q_lspid = task_tgid_vnr(current);
694 msq->q_stime = get_seconds();
696 if (!pipelined_send(msq, msg)) {
697 /* no one is waiting for this message, enqueue it */
698 list_add_tail(&msg->m_list, &msq->q_messages);
699 msq->q_cbytes += msgsz;
701 atomic_add(msgsz, &ns->msg_bytes);
702 atomic_inc(&ns->msg_hdrs);
716 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
721 if (get_user(mtype, &msgp->mtype))
723 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
726 static inline int convert_mode(long *msgtyp, int msgflg)
728 if (msgflg & MSG_COPY)
729 return SEARCH_NUMBER;
731 * find message of correct type.
732 * msgtyp = 0 => get first.
733 * msgtyp > 0 => get first message of matching type.
734 * msgtyp < 0 => get message with least type must be < abs(msgtype).
740 return SEARCH_LESSEQUAL;
742 if (msgflg & MSG_EXCEPT)
743 return SEARCH_NOTEQUAL;
747 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
749 struct msgbuf __user *msgp = dest;
752 if (put_user(msg->m_type, &msgp->mtype))
755 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
756 if (store_msg(msgp->mtext, msg, msgsz))
761 #ifdef CONFIG_CHECKPOINT_RESTORE
763 * This function creates new kernel message structure, large enough to store
764 * bufsz message bytes.
766 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
768 struct msg_msg *copy;
771 * Create dummy message to copy real message to.
773 copy = load_msg(buf, bufsz);
779 static inline void free_copy(struct msg_msg *copy)
785 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
787 return ERR_PTR(-ENOSYS);
790 static inline void free_copy(struct msg_msg *copy)
795 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
800 list_for_each_entry(msg, &msq->q_messages, m_list) {
801 if (testmsg(msg, *msgtyp, mode) &&
802 !security_msg_queue_msgrcv(msq, msg, current,
804 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
805 *msgtyp = msg->m_type - 1;
806 } else if (mode == SEARCH_NUMBER) {
807 if (*msgtyp == count)
815 return ERR_PTR(-EAGAIN);
819 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
821 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
823 struct msg_queue *msq;
826 struct ipc_namespace *ns;
827 struct msg_msg *copy = NULL;
829 ns = current->nsproxy->ipc_ns;
831 if (msqid < 0 || (long) bufsz < 0)
833 if (msgflg & MSG_COPY) {
834 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
836 return PTR_ERR(copy);
838 mode = convert_mode(&msgtyp, msgflg);
840 msq = msg_lock_check(ns, msqid);
847 struct msg_receiver msr_d;
849 msg = ERR_PTR(-EACCES);
850 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
853 msg = find_msg(msq, &msgtyp, mode);
857 * Found a suitable message.
858 * Unlink it from the queue.
860 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
861 msg = ERR_PTR(-E2BIG);
865 * If we are copying, then do not unlink message and do
866 * not update queue parameters.
868 if (msgflg & MSG_COPY) {
869 msg = copy_msg(msg, copy);
872 list_del(&msg->m_list);
874 msq->q_rtime = get_seconds();
875 msq->q_lrpid = task_tgid_vnr(current);
876 msq->q_cbytes -= msg->m_ts;
877 atomic_sub(msg->m_ts, &ns->msg_bytes);
878 atomic_dec(&ns->msg_hdrs);
879 ss_wakeup(&msq->q_senders, 0);
883 /* No message waiting. Wait for a message */
884 if (msgflg & IPC_NOWAIT) {
885 msg = ERR_PTR(-ENOMSG);
888 list_add_tail(&msr_d.r_list, &msq->q_receivers);
889 msr_d.r_tsk = current;
890 msr_d.r_msgtype = msgtyp;
892 if (msgflg & MSG_NOERROR)
893 msr_d.r_maxsize = INT_MAX;
895 msr_d.r_maxsize = bufsz;
896 msr_d.r_msg = ERR_PTR(-EAGAIN);
897 current->state = TASK_INTERRUPTIBLE;
902 /* Lockless receive, part 1:
903 * Disable preemption. We don't hold a reference to the queue
904 * and getting a reference would defeat the idea of a lockless
905 * operation, thus the code relies on rcu to guarantee the
907 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
908 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
909 * rcu_read_lock() prevents preemption between reading r_msg
910 * and the spin_lock() inside ipc_lock_by_ptr().
914 /* Lockless receive, part 2:
915 * Wait until pipelined_send or expunge_all are outside of
916 * wake_up_process(). There is a race with exit(), see
917 * ipc/mqueue.c for the details.
919 msg = (struct msg_msg*)msr_d.r_msg;
920 while (msg == NULL) {
922 msg = (struct msg_msg *)msr_d.r_msg;
925 /* Lockless receive, part 3:
926 * If there is a message or an error then accept it without
929 if (msg != ERR_PTR(-EAGAIN)) {
934 /* Lockless receive, part 3:
935 * Acquire the queue spinlock.
937 ipc_lock_by_ptr(&msq->q_perm);
940 /* Lockless receive, part 4:
941 * Repeat test after acquiring the spinlock.
943 msg = (struct msg_msg*)msr_d.r_msg;
944 if (msg != ERR_PTR(-EAGAIN))
947 list_del(&msr_d.r_list);
948 if (signal_pending(current)) {
949 msg = ERR_PTR(-ERESTARTNOHAND);
960 bufsz = msg_handler(buf, msg, bufsz);
966 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
967 long, msgtyp, int, msgflg)
969 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
972 #ifdef CONFIG_PROC_FS
973 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
975 struct user_namespace *user_ns = seq_user_ns(s);
976 struct msg_queue *msq = it;
979 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
987 from_kuid_munged(user_ns, msq->q_perm.uid),
988 from_kgid_munged(user_ns, msq->q_perm.gid),
989 from_kuid_munged(user_ns, msq->q_perm.cuid),
990 from_kgid_munged(user_ns, msq->q_perm.cgid),