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
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
17 * support for audit of ipc object properties and permission changes
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);
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 msg_sender *mss, *t;
243 list_for_each_entry_safe(mss, t, h, list) {
245 mss->list.next = NULL;
246 wake_up_process(mss->tsk);
250 static void expunge_all(struct msg_queue *msq, int res)
252 struct msg_receiver *msr, *t;
254 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
256 wake_up_process(msr->r_tsk);
258 msr->r_msg = ERR_PTR(res);
263 * freeque() wakes up waiters on the sender and receiver waiting queue,
264 * removes the message queue from message queue ID IDR, and cleans up all the
265 * messages associated with this queue.
267 * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
268 * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
270 static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
272 struct msg_msg *msg, *t;
273 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
275 expunge_all(msq, -EIDRM);
276 ss_wakeup(&msq->q_senders, 1);
280 list_for_each_entry_safe(msg, t, &msq->q_messages, m_list) {
281 atomic_dec(&ns->msg_hdrs);
284 atomic_sub(msq->q_cbytes, &ns->msg_bytes);
285 security_msg_queue_free(msq);
290 * Called with msg_ids.rw_mutex and ipcp locked.
292 static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
294 struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
296 return security_msg_queue_associate(msq, msgflg);
299 SYSCALL_DEFINE2(msgget, key_t, key, int, msgflg)
301 struct ipc_namespace *ns;
302 struct ipc_ops msg_ops;
303 struct ipc_params msg_params;
305 ns = current->nsproxy->ipc_ns;
307 msg_ops.getnew = newque;
308 msg_ops.associate = msg_security;
309 msg_ops.more_checks = NULL;
311 msg_params.key = key;
312 msg_params.flg = msgflg;
314 return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
317 static inline unsigned long
318 copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
322 return copy_to_user(buf, in, sizeof(*in));
327 memset(&out, 0, sizeof(out));
329 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
331 out.msg_stime = in->msg_stime;
332 out.msg_rtime = in->msg_rtime;
333 out.msg_ctime = in->msg_ctime;
335 if (in->msg_cbytes > USHRT_MAX)
336 out.msg_cbytes = USHRT_MAX;
338 out.msg_cbytes = in->msg_cbytes;
339 out.msg_lcbytes = in->msg_cbytes;
341 if (in->msg_qnum > USHRT_MAX)
342 out.msg_qnum = USHRT_MAX;
344 out.msg_qnum = in->msg_qnum;
346 if (in->msg_qbytes > USHRT_MAX)
347 out.msg_qbytes = USHRT_MAX;
349 out.msg_qbytes = in->msg_qbytes;
350 out.msg_lqbytes = in->msg_qbytes;
352 out.msg_lspid = in->msg_lspid;
353 out.msg_lrpid = in->msg_lrpid;
355 return copy_to_user(buf, &out, sizeof(out));
362 static inline unsigned long
363 copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
367 if (copy_from_user(out, buf, sizeof(*out)))
372 struct msqid_ds tbuf_old;
374 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
377 out->msg_perm.uid = tbuf_old.msg_perm.uid;
378 out->msg_perm.gid = tbuf_old.msg_perm.gid;
379 out->msg_perm.mode = tbuf_old.msg_perm.mode;
381 if (tbuf_old.msg_qbytes == 0)
382 out->msg_qbytes = tbuf_old.msg_lqbytes;
384 out->msg_qbytes = tbuf_old.msg_qbytes;
394 * This function handles some msgctl commands which require the rw_mutex
395 * to be held in write mode.
396 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
398 static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
399 struct msqid_ds __user *buf, int version)
401 struct kern_ipc_perm *ipcp;
402 struct msqid64_ds uninitialized_var(msqid64);
403 struct msg_queue *msq;
406 if (cmd == IPC_SET) {
407 if (copy_msqid_from_user(&msqid64, buf, version))
411 ipcp = ipcctl_pre_down(ns, &msg_ids(ns), msqid, cmd,
412 &msqid64.msg_perm, msqid64.msg_qbytes);
414 return PTR_ERR(ipcp);
416 msq = container_of(ipcp, struct msg_queue, q_perm);
418 err = security_msg_queue_msgctl(msq, cmd);
427 if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
428 !capable(CAP_SYS_RESOURCE)) {
433 err = ipc_update_perm(&msqid64.msg_perm, ipcp);
437 msq->q_qbytes = msqid64.msg_qbytes;
439 msq->q_ctime = get_seconds();
440 /* sleeping receivers might be excluded by
441 * stricter permissions.
443 expunge_all(msq, -EAGAIN);
444 /* sleeping senders might be able to send
445 * due to a larger queue size.
447 ss_wakeup(&msq->q_senders, 0);
455 up_write(&msg_ids(ns).rw_mutex);
459 SYSCALL_DEFINE3(msgctl, int, msqid, int, cmd, struct msqid_ds __user *, buf)
461 struct msg_queue *msq;
463 struct ipc_namespace *ns;
465 if (msqid < 0 || cmd < 0)
468 version = ipc_parse_version(&cmd);
469 ns = current->nsproxy->ipc_ns;
475 struct msginfo msginfo;
481 * We must not return kernel stack data.
482 * due to padding, it's not enough
483 * to set all member fields.
485 err = security_msg_queue_msgctl(NULL, cmd);
489 memset(&msginfo, 0, sizeof(msginfo));
490 msginfo.msgmni = ns->msg_ctlmni;
491 msginfo.msgmax = ns->msg_ctlmax;
492 msginfo.msgmnb = ns->msg_ctlmnb;
493 msginfo.msgssz = MSGSSZ;
494 msginfo.msgseg = MSGSEG;
495 down_read(&msg_ids(ns).rw_mutex);
496 if (cmd == MSG_INFO) {
497 msginfo.msgpool = msg_ids(ns).in_use;
498 msginfo.msgmap = atomic_read(&ns->msg_hdrs);
499 msginfo.msgtql = atomic_read(&ns->msg_bytes);
501 msginfo.msgmap = MSGMAP;
502 msginfo.msgpool = MSGPOOL;
503 msginfo.msgtql = MSGTQL;
505 max_id = ipc_get_maxid(&msg_ids(ns));
506 up_read(&msg_ids(ns).rw_mutex);
507 if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
509 return (max_id < 0) ? 0 : max_id;
511 case MSG_STAT: /* msqid is an index rather than a msg queue id */
514 struct msqid64_ds tbuf;
520 if (cmd == MSG_STAT) {
521 msq = msg_lock(ns, msqid);
524 success_return = msq->q_perm.id;
526 msq = msg_lock_check(ns, msqid);
532 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
535 err = security_msg_queue_msgctl(msq, cmd);
539 memset(&tbuf, 0, sizeof(tbuf));
541 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
542 tbuf.msg_stime = msq->q_stime;
543 tbuf.msg_rtime = msq->q_rtime;
544 tbuf.msg_ctime = msq->q_ctime;
545 tbuf.msg_cbytes = msq->q_cbytes;
546 tbuf.msg_qnum = msq->q_qnum;
547 tbuf.msg_qbytes = msq->q_qbytes;
548 tbuf.msg_lspid = msq->q_lspid;
549 tbuf.msg_lrpid = msq->q_lrpid;
551 if (copy_msqid_to_user(buf, &tbuf, version))
553 return success_return;
557 err = msgctl_down(ns, msqid, cmd, buf, version);
568 static int testmsg(struct msg_msg *msg, long type, int mode)
575 case SEARCH_LESSEQUAL:
576 if (msg->m_type <=type)
580 if (msg->m_type == type)
583 case SEARCH_NOTEQUAL:
584 if (msg->m_type != type)
591 static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
593 struct msg_receiver *msr, *t;
595 list_for_each_entry_safe(msr, t, &msq->q_receivers, r_list) {
596 if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
597 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
598 msr->r_msgtype, msr->r_mode)) {
600 list_del(&msr->r_list);
601 if (msr->r_maxsize < msg->m_ts) {
603 wake_up_process(msr->r_tsk);
605 msr->r_msg = ERR_PTR(-E2BIG);
608 msq->q_lrpid = task_pid_vnr(msr->r_tsk);
609 msq->q_rtime = get_seconds();
610 wake_up_process(msr->r_tsk);
621 long do_msgsnd(int msqid, long mtype, void __user *mtext,
622 size_t msgsz, int msgflg)
624 struct msg_queue *msq;
627 struct ipc_namespace *ns;
629 ns = current->nsproxy->ipc_ns;
631 if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
636 msg = load_msg(mtext, msgsz);
643 msq = msg_lock_check(ns, msqid);
653 if (ipcperms(ns, &msq->q_perm, S_IWUGO))
654 goto out_unlock_free;
656 err = security_msg_queue_msgsnd(msq, msg, msgflg);
658 goto out_unlock_free;
660 if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
661 1 + msq->q_qnum <= msq->q_qbytes) {
665 /* queue full, wait: */
666 if (msgflg & IPC_NOWAIT) {
668 goto out_unlock_free;
672 if (!ipc_rcu_getref(msq)) {
674 goto out_unlock_free;
680 ipc_lock_by_ptr(&msq->q_perm);
682 if (msq->q_perm.deleted) {
684 goto out_unlock_free;
688 if (signal_pending(current)) {
689 err = -ERESTARTNOHAND;
690 goto out_unlock_free;
694 msq->q_lspid = task_tgid_vnr(current);
695 msq->q_stime = get_seconds();
697 if (!pipelined_send(msq, msg)) {
698 /* no one is waiting for this message, enqueue it */
699 list_add_tail(&msg->m_list, &msq->q_messages);
700 msq->q_cbytes += msgsz;
702 atomic_add(msgsz, &ns->msg_bytes);
703 atomic_inc(&ns->msg_hdrs);
717 SYSCALL_DEFINE4(msgsnd, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
722 if (get_user(mtype, &msgp->mtype))
724 return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
727 static inline int convert_mode(long *msgtyp, int msgflg)
729 if (msgflg & MSG_COPY)
730 return SEARCH_NUMBER;
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 static long do_msg_fill(void __user *dest, struct msg_msg *msg, size_t bufsz)
750 struct msgbuf __user *msgp = dest;
753 if (put_user(msg->m_type, &msgp->mtype))
756 msgsz = (bufsz > msg->m_ts) ? msg->m_ts : bufsz;
757 if (store_msg(msgp->mtext, msg, msgsz))
762 #ifdef CONFIG_CHECKPOINT_RESTORE
764 * This function creates new kernel message structure, large enough to store
765 * bufsz message bytes.
767 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
769 struct msg_msg *copy;
772 * Create dummy message to copy real message to.
774 copy = load_msg(buf, bufsz);
780 static inline void free_copy(struct msg_msg *copy)
786 static inline struct msg_msg *prepare_copy(void __user *buf, size_t bufsz)
788 return ERR_PTR(-ENOSYS);
791 static inline void free_copy(struct msg_msg *copy)
796 static struct msg_msg *find_msg(struct msg_queue *msq, long *msgtyp, int mode)
801 list_for_each_entry(msg, &msq->q_messages, m_list) {
802 if (testmsg(msg, *msgtyp, mode) &&
803 !security_msg_queue_msgrcv(msq, msg, current,
805 if (mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
806 *msgtyp = msg->m_type - 1;
807 } else if (mode == SEARCH_NUMBER) {
808 if (*msgtyp == count)
816 return ERR_PTR(-EAGAIN);
820 long do_msgrcv(int msqid, void __user *buf, size_t bufsz, long msgtyp,
822 long (*msg_handler)(void __user *, struct msg_msg *, size_t))
824 struct msg_queue *msq;
827 struct ipc_namespace *ns;
828 struct msg_msg *copy = NULL;
830 ns = current->nsproxy->ipc_ns;
832 if (msqid < 0 || (long) bufsz < 0)
834 if (msgflg & MSG_COPY) {
835 copy = prepare_copy(buf, min_t(size_t, bufsz, ns->msg_ctlmax));
837 return PTR_ERR(copy);
839 mode = convert_mode(&msgtyp, msgflg);
841 msq = msg_lock_check(ns, msqid);
848 struct msg_receiver msr_d;
850 msg = ERR_PTR(-EACCES);
851 if (ipcperms(ns, &msq->q_perm, S_IRUGO))
854 msg = find_msg(msq, &msgtyp, mode);
858 * Found a suitable message.
859 * Unlink it from the queue.
861 if ((bufsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
862 msg = ERR_PTR(-E2BIG);
866 * If we are copying, then do not unlink message and do
867 * not update queue parameters.
869 if (msgflg & MSG_COPY) {
870 msg = copy_msg(msg, copy);
873 list_del(&msg->m_list);
875 msq->q_rtime = get_seconds();
876 msq->q_lrpid = task_tgid_vnr(current);
877 msq->q_cbytes -= msg->m_ts;
878 atomic_sub(msg->m_ts, &ns->msg_bytes);
879 atomic_dec(&ns->msg_hdrs);
880 ss_wakeup(&msq->q_senders, 0);
884 /* No message waiting. Wait for a message */
885 if (msgflg & IPC_NOWAIT) {
886 msg = ERR_PTR(-ENOMSG);
889 list_add_tail(&msr_d.r_list, &msq->q_receivers);
890 msr_d.r_tsk = current;
891 msr_d.r_msgtype = msgtyp;
893 if (msgflg & MSG_NOERROR)
894 msr_d.r_maxsize = INT_MAX;
896 msr_d.r_maxsize = bufsz;
897 msr_d.r_msg = ERR_PTR(-EAGAIN);
898 current->state = TASK_INTERRUPTIBLE;
903 /* Lockless receive, part 1:
904 * Disable preemption. We don't hold a reference to the queue
905 * and getting a reference would defeat the idea of a lockless
906 * operation, thus the code relies on rcu to guarantee the
908 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
909 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
910 * rcu_read_lock() prevents preemption between reading r_msg
911 * and the spin_lock() inside ipc_lock_by_ptr().
915 /* Lockless receive, part 2:
916 * Wait until pipelined_send or expunge_all are outside of
917 * wake_up_process(). There is a race with exit(), see
918 * ipc/mqueue.c for the details.
920 msg = (struct msg_msg*)msr_d.r_msg;
921 while (msg == NULL) {
923 msg = (struct msg_msg *)msr_d.r_msg;
926 /* Lockless receive, part 3:
927 * If there is a message or an error then accept it without
930 if (msg != ERR_PTR(-EAGAIN)) {
935 /* Lockless receive, part 3:
936 * Acquire the queue spinlock.
938 ipc_lock_by_ptr(&msq->q_perm);
941 /* Lockless receive, part 4:
942 * Repeat test after acquiring the spinlock.
944 msg = (struct msg_msg*)msr_d.r_msg;
945 if (msg != ERR_PTR(-EAGAIN))
948 list_del(&msr_d.r_list);
949 if (signal_pending(current)) {
950 msg = ERR_PTR(-ERESTARTNOHAND);
961 bufsz = msg_handler(buf, msg, bufsz);
967 SYSCALL_DEFINE5(msgrcv, int, msqid, struct msgbuf __user *, msgp, size_t, msgsz,
968 long, msgtyp, int, msgflg)
970 return do_msgrcv(msqid, msgp, msgsz, msgtyp, msgflg, do_msg_fill);
973 #ifdef CONFIG_PROC_FS
974 static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
976 struct user_namespace *user_ns = seq_user_ns(s);
977 struct msg_queue *msq = it;
980 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
988 from_kuid_munged(user_ns, msq->q_perm.uid),
989 from_kgid_munged(user_ns, msq->q_perm.gid),
990 from_kuid_munged(user_ns, msq->q_perm.cuid),
991 from_kgid_munged(user_ns, msq->q_perm.cgid),