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;
113 printk(KERN_INFO "msgmni has been set to %d for ipc namespace %p\n",
117 void msg_init_ns(struct ipc_namespace *ns)
119 ns->msg_ctlmax = MSGMAX;
120 ns->msg_ctlmnb = MSGMNB;
122 recompute_msgmni(ns);
124 atomic_set(&ns->msg_bytes, 0);
125 atomic_set(&ns->msg_hdrs, 0);
126 ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
130 void msg_exit_ns(struct ipc_namespace *ns)
132 free_ipcs(ns, &msg_ids(ns), freeque);
136 void __init msg_init(void)
138 msg_init_ns(&init_ipc_ns);
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 * This routine is called in the paths where the rw_mutex is held to protect
146 * access to the idr tree.
148 static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
151 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
154 return (struct msg_queue *)ipcp;
156 return container_of(ipcp, struct msg_queue, q_perm);
160 * msg_lock_(check_) routines are called in the paths where the rw_mutex
163 static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
165 struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
168 return (struct msg_queue *)ipcp;
170 return container_of(ipcp, struct msg_queue, q_perm);
173 static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
176 struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
179 return (struct msg_queue *)ipcp;
181 return container_of(ipcp, struct msg_queue, q_perm);
184 static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
186 ipc_rmid(&msg_ids(ns), &s->q_perm);
190 * newque - Create a new msg queue
192 * @params: ptr to the structure that contains the key and msgflg
194 * Called with msg_ids.rw_mutex held (writer)
196 static int newque(struct ipc_namespace *ns, struct ipc_params *params)
198 struct msg_queue *msq;
200 key_t key = params->key;
201 int msgflg = params->flg;
203 msq = ipc_rcu_alloc(sizeof(*msq));
207 msq->q_perm.mode = msgflg & S_IRWXUGO;
208 msq->q_perm.key = key;
210 msq->q_perm.security = NULL;
211 retval = security_msg_queue_alloc(msq);
218 * ipc_addid() locks msq
220 id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
222 security_msg_queue_free(msq);
227 msq->q_stime = msq->q_rtime = 0;
228 msq->q_ctime = get_seconds();
229 msq->q_cbytes = msq->q_qnum = 0;
230 msq->q_qbytes = ns->msg_ctlmnb;
231 msq->q_lspid = msq->q_lrpid = 0;
232 INIT_LIST_HEAD(&msq->q_messages);
233 INIT_LIST_HEAD(&msq->q_receivers);
234 INIT_LIST_HEAD(&msq->q_senders);
238 return msq->q_perm.id;
241 static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
244 current->state = TASK_INTERRUPTIBLE;
245 list_add_tail(&mss->list, &msq->q_senders);
248 static inline void ss_del(struct msg_sender *mss)
250 if (mss->list.next != NULL)
251 list_del(&mss->list);
254 static void ss_wakeup(struct list_head *h, int kill)
256 struct list_head *tmp;
260 struct msg_sender *mss;
262 mss = list_entry(tmp, struct msg_sender, list);
265 mss->list.next = NULL;
266 wake_up_process(mss->tsk);
270 static void expunge_all(struct msg_queue *msq, int res)
272 struct list_head *tmp;
274 tmp = msq->q_receivers.next;
275 while (tmp != &msq->q_receivers) {
276 struct msg_receiver *msr;
278 msr = list_entry(tmp, struct msg_receiver, r_list);
281 wake_up_process(msr->r_tsk);
283 msr->r_msg = ERR_PTR(res);
288 * freeque() wakes up waiters on the sender and receiver waiting queue,
289 * removes the message queue from message queue ID IDR, and cleans up all the
290 * messages associated with this queue.
292 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
293 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
295 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
297 struct list_head *tmp;
298 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
300 expunge_all(msq, -EIDRM);
301 ss_wakeup(&msq->q_senders, 1);
305 tmp = msq->q_messages.next;
306 while (tmp != &msq->q_messages) {
307 struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
310 atomic_dec(&ns->msg_hdrs);
313 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
314 security_msg_queue_free(msq);
319 * Called with msg_ids.rw_mutex and ipcp locked.
321 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
323 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
325 return security_msg_queue_associate(msq, msgflg);
328 asmlinkage long sys_msgget(key_t key, int msgflg)
330 struct ipc_namespace *ns;
331 struct ipc_ops msg_ops;
332 struct ipc_params msg_params;
334 ns = current->nsproxy->ipc_ns;
336 msg_ops.getnew = newque;
337 msg_ops.associate = msg_security;
338 msg_ops.more_checks = NULL;
340 msg_params.key = key;
341 msg_params.flg = msgflg;
343 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
346 static inline unsigned long
347 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
351 return copy_to_user(buf, in, sizeof(*in));
356 memset(&out, 0, sizeof(out));
358 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
360 out.msg_stime = in->msg_stime;
361 out.msg_rtime = in->msg_rtime;
362 out.msg_ctime = in->msg_ctime;
364 if (in->msg_cbytes > USHRT_MAX)
365 out.msg_cbytes = USHRT_MAX;
367 out.msg_cbytes = in->msg_cbytes;
368 out.msg_lcbytes = in->msg_cbytes;
370 if (in->msg_qnum > USHRT_MAX)
371 out.msg_qnum = USHRT_MAX;
373 out.msg_qnum = in->msg_qnum;
375 if (in->msg_qbytes > USHRT_MAX)
376 out.msg_qbytes = USHRT_MAX;
378 out.msg_qbytes = in->msg_qbytes;
379 out.msg_lqbytes = in->msg_qbytes;
381 out.msg_lspid = in->msg_lspid;
382 out.msg_lrpid = in->msg_lrpid;
384 return copy_to_user(buf, &out, sizeof(out));
391 static inline unsigned long
392 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
396 if (copy_from_user(out, buf, sizeof(*out)))
401 struct msqid_ds tbuf_old;
403 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
406 out->msg_perm.uid = tbuf_old.msg_perm.uid;
407 out->msg_perm.gid = tbuf_old.msg_perm.gid;
408 out->msg_perm.mode = tbuf_old.msg_perm.mode;
410 if (tbuf_old.msg_qbytes == 0)
411 out->msg_qbytes = tbuf_old.msg_lqbytes;
413 out->msg_qbytes = tbuf_old.msg_qbytes;
423 * This function handles some msgctl commands which require the rw_mutex
424 * to be held in write mode.
425 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
427 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
428 struct msqid_ds __user *buf, int version)
430 struct kern_ipc_perm *ipcp;
431 struct msqid64_ds msqid64;
432 struct msg_queue *msq;
435 if (cmd == IPC_SET) {
436 if (copy_msqid_from_user(&msqid64, buf, version))
440 down_write(&msg_ids(ns).rw_mutex);
441 msq = msg_lock_check_down(ns, msqid);
449 err = audit_ipc_obj(ipcp);
453 if (cmd == IPC_SET) {
454 err = audit_ipc_set_perm(msqid64.msg_qbytes,
455 msqid64.msg_perm.uid,
456 msqid64.msg_perm.gid,
457 msqid64.msg_perm.mode);
462 if (current->euid != ipcp->cuid &&
463 current->euid != ipcp->uid &&
464 !capable(CAP_SYS_ADMIN)) {
465 /* We _could_ check for CAP_CHOWN above, but we don't */
470 err = security_msg_queue_msgctl(msq, cmd);
479 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
480 !capable(CAP_SYS_RESOURCE)) {
485 msq->q_qbytes = msqid64.msg_qbytes;
487 ipc_update_perm(&msqid64.msg_perm, ipcp);
488 msq->q_ctime = get_seconds();
489 /* sleeping receivers might be excluded by
490 * stricter permissions.
492 expunge_all(msq, -EAGAIN);
493 /* sleeping senders might be able to send
494 * due to a larger queue size.
496 ss_wakeup(&msq->q_senders, 0);
504 up_write(&msg_ids(ns).rw_mutex);
508 asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
510 struct msg_queue *msq;
512 struct ipc_namespace *ns;
514 if (msqid < 0 || cmd < 0)
517 version = ipc_parse_version(&cmd);
518 ns = current->nsproxy->ipc_ns;
524 struct msginfo msginfo;
530 * We must not return kernel stack data.
531 * due to padding, it's not enough
532 * to set all member fields.
534 err = security_msg_queue_msgctl(NULL, cmd);
538 memset(&msginfo, 0, sizeof(msginfo));
539 msginfo.msgmni = ns->msg_ctlmni;
540 msginfo.msgmax = ns->msg_ctlmax;
541 msginfo.msgmnb = ns->msg_ctlmnb;
542 msginfo.msgssz = MSGSSZ;
543 msginfo.msgseg = MSGSEG;
544 down_read(&msg_ids(ns).rw_mutex);
545 if (cmd == MSG_INFO) {
546 msginfo.msgpool = msg_ids(ns).in_use;
547 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
548 msginfo.msgtql = atomic_read(&ns->msg_bytes);
550 msginfo.msgmap = MSGMAP;
551 msginfo.msgpool = MSGPOOL;
552 msginfo.msgtql = MSGTQL;
554 max_id = ipc_get_maxid(&msg_ids(ns));
555 up_read(&msg_ids(ns).rw_mutex);
556 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
558 return (max_id < 0) ? 0 : max_id;
560 case MSG_STAT: /* msqid is an index rather than a msg queue id */
563 struct msqid64_ds tbuf;
569 if (cmd == MSG_STAT) {
570 msq = msg_lock(ns, msqid);
573 success_return = msq->q_perm.id;
575 msq = msg_lock_check(ns, msqid);
581 if (ipcperms(&msq->q_perm, S_IRUGO))
584 err = security_msg_queue_msgctl(msq, cmd);
588 memset(&tbuf, 0, sizeof(tbuf));
590 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
591 tbuf.msg_stime = msq->q_stime;
592 tbuf.msg_rtime = msq->q_rtime;
593 tbuf.msg_ctime = msq->q_ctime;
594 tbuf.msg_cbytes = msq->q_cbytes;
595 tbuf.msg_qnum = msq->q_qnum;
596 tbuf.msg_qbytes = msq->q_qbytes;
597 tbuf.msg_lspid = msq->q_lspid;
598 tbuf.msg_lrpid = msq->q_lrpid;
600 if (copy_msqid_to_user(buf, &tbuf, version))
602 return success_return;
606 err = msgctl_down(ns, msqid, cmd, buf, version);
617 static int testmsg(struct msg_msg *msg, long type, int mode)
623 case SEARCH_LESSEQUAL:
624 if (msg->m_type <=type)
628 if (msg->m_type == type)
631 case SEARCH_NOTEQUAL:
632 if (msg->m_type != type)
639 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
641 struct list_head *tmp;
643 tmp = msq->q_receivers.next;
644 while (tmp != &msq->q_receivers) {
645 struct msg_receiver *msr;
647 msr = list_entry(tmp, struct msg_receiver, r_list);
649 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
650 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
651 msr->r_msgtype, msr->r_mode)) {
653 list_del(&msr->r_list);
654 if (msr->r_maxsize < msg->m_ts) {
656 wake_up_process(msr->r_tsk);
658 msr->r_msg = ERR_PTR(-E2BIG);
661 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
662 msq->q_rtime = get_seconds();
663 wake_up_process(msr->r_tsk);
674 long do_msgsnd(int msqid, long mtype, void __user *mtext,
675 size_t msgsz, int msgflg)
677 struct msg_queue *msq;
680 struct ipc_namespace *ns;
682 ns = current->nsproxy->ipc_ns;
684 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
689 msg = load_msg(mtext, msgsz);
696 msq = msg_lock_check(ns, msqid);
706 if (ipcperms(&msq->q_perm, S_IWUGO))
707 goto out_unlock_free;
709 err = security_msg_queue_msgsnd(msq, msg, msgflg);
711 goto out_unlock_free;
713 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
714 1 + msq->q_qnum <= msq->q_qbytes) {
718 /* queue full, wait: */
719 if (msgflg & IPC_NOWAIT) {
721 goto out_unlock_free;
728 ipc_lock_by_ptr(&msq->q_perm);
730 if (msq->q_perm.deleted) {
732 goto out_unlock_free;
736 if (signal_pending(current)) {
737 err = -ERESTARTNOHAND;
738 goto out_unlock_free;
742 msq->q_lspid = task_tgid_vnr(current);
743 msq->q_stime = get_seconds();
745 if (!pipelined_send(msq, msg)) {
746 /* noone is waiting for this message, enqueue it */
747 list_add_tail(&msg->m_list, &msq->q_messages);
748 msq->q_cbytes += msgsz;
750 atomic_add(msgsz, &ns->msg_bytes);
751 atomic_inc(&ns->msg_hdrs);
766 sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
770 if (get_user(mtype, &msgp->mtype))
772 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
775 static inline int convert_mode(long *msgtyp, int msgflg)
778 * find message of correct type.
779 * msgtyp = 0 => get first.
780 * msgtyp > 0 => get first message of matching type.
781 * msgtyp < 0 => get message with least type must be < abs(msgtype).
787 return SEARCH_LESSEQUAL;
789 if (msgflg & MSG_EXCEPT)
790 return SEARCH_NOTEQUAL;
794 long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
795 size_t msgsz, long msgtyp, int msgflg)
797 struct msg_queue *msq;
800 struct ipc_namespace *ns;
802 if (msqid < 0 || (long) msgsz < 0)
804 mode = convert_mode(&msgtyp, msgflg);
805 ns = current->nsproxy->ipc_ns;
807 msq = msg_lock_check(ns, msqid);
812 struct msg_receiver msr_d;
813 struct list_head *tmp;
815 msg = ERR_PTR(-EACCES);
816 if (ipcperms(&msq->q_perm, S_IRUGO))
819 msg = ERR_PTR(-EAGAIN);
820 tmp = msq->q_messages.next;
821 while (tmp != &msq->q_messages) {
822 struct msg_msg *walk_msg;
824 walk_msg = list_entry(tmp, struct msg_msg, m_list);
825 if (testmsg(walk_msg, msgtyp, mode) &&
826 !security_msg_queue_msgrcv(msq, walk_msg, current,
830 if (mode == SEARCH_LESSEQUAL &&
831 walk_msg->m_type != 1) {
833 msgtyp = walk_msg->m_type - 1;
843 * Found a suitable message.
844 * Unlink it from the queue.
846 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
847 msg = ERR_PTR(-E2BIG);
850 list_del(&msg->m_list);
852 msq->q_rtime = get_seconds();
853 msq->q_lrpid = task_tgid_vnr(current);
854 msq->q_cbytes -= msg->m_ts;
855 atomic_sub(msg->m_ts, &ns->msg_bytes);
856 atomic_dec(&ns->msg_hdrs);
857 ss_wakeup(&msq->q_senders, 0);
861 /* No message waiting. Wait for a message */
862 if (msgflg & IPC_NOWAIT) {
863 msg = ERR_PTR(-ENOMSG);
866 list_add_tail(&msr_d.r_list, &msq->q_receivers);
867 msr_d.r_tsk = current;
868 msr_d.r_msgtype = msgtyp;
870 if (msgflg & MSG_NOERROR)
871 msr_d.r_maxsize = INT_MAX;
873 msr_d.r_maxsize = msgsz;
874 msr_d.r_msg = ERR_PTR(-EAGAIN);
875 current->state = TASK_INTERRUPTIBLE;
880 /* Lockless receive, part 1:
881 * Disable preemption. We don't hold a reference to the queue
882 * and getting a reference would defeat the idea of a lockless
883 * operation, thus the code relies on rcu to guarantee the
885 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
886 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
887 * rcu_read_lock() prevents preemption between reading r_msg
888 * and the spin_lock() inside ipc_lock_by_ptr().
892 /* Lockless receive, part 2:
893 * Wait until pipelined_send or expunge_all are outside of
894 * wake_up_process(). There is a race with exit(), see
895 * ipc/mqueue.c for the details.
897 msg = (struct msg_msg*)msr_d.r_msg;
898 while (msg == NULL) {
900 msg = (struct msg_msg *)msr_d.r_msg;
903 /* Lockless receive, part 3:
904 * If there is a message or an error then accept it without
907 if (msg != ERR_PTR(-EAGAIN)) {
912 /* Lockless receive, part 3:
913 * Acquire the queue spinlock.
915 ipc_lock_by_ptr(&msq->q_perm);
918 /* Lockless receive, part 4:
919 * Repeat test after acquiring the spinlock.
921 msg = (struct msg_msg*)msr_d.r_msg;
922 if (msg != ERR_PTR(-EAGAIN))
925 list_del(&msr_d.r_list);
926 if (signal_pending(current)) {
927 msg = ERR_PTR(-ERESTARTNOHAND);
936 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
937 *pmtype = msg->m_type;
938 if (store_msg(mtext, msg, msgsz))
946 asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
947 long msgtyp, int msgflg)
951 err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
955 if (put_user(mtype, &msgp->mtype))
961 #ifdef CONFIG_PROC_FS
962 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
964 struct msg_queue *msq = it;
967 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",