2 * POSIX message queues filesystem for Linux.
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
13 * This file is released under the GPL.
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
34 #include <linux/ipc_namespace.h>
39 #define MQUEUE_MAGIC 0x19800202
40 #define DIRENT_SIZE 20
41 #define FILENT_SIZE 80
47 #define STATE_PENDING 1
50 struct ext_wait_queue { /* queue of sleeping tasks */
51 struct task_struct *task;
52 struct list_head list;
53 struct msg_msg *msg; /* ptr of loaded message */
54 int state; /* one of STATE_* values */
57 struct mqueue_inode_info {
59 struct inode vfs_inode;
60 wait_queue_head_t wait_q;
62 struct msg_msg **messages;
65 struct sigevent notify;
66 struct pid* notify_owner;
67 struct user_struct *user; /* user who created, for accounting */
68 struct sock *notify_sock;
69 struct sk_buff *notify_cookie;
71 /* for tasks waiting for free space and messages, respectively */
72 struct ext_wait_queue e_wait_q[2];
74 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
77 static const struct inode_operations mqueue_dir_inode_operations;
78 static const struct file_operations mqueue_file_operations;
79 static const struct super_operations mqueue_super_ops;
80 static void remove_notification(struct mqueue_inode_info *info);
82 static struct kmem_cache *mqueue_inode_cachep;
84 static struct ctl_table_header * mq_sysctl_table;
86 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
88 return container_of(inode, struct mqueue_inode_info, vfs_inode);
92 * This routine should be called with the mq_lock held.
94 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
96 return get_ipc_ns(inode->i_sb->s_fs_info);
99 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
101 struct ipc_namespace *ns;
104 ns = __get_ns_from_inode(inode);
105 spin_unlock(&mq_lock);
109 static struct inode *mqueue_get_inode(struct super_block *sb,
110 struct ipc_namespace *ipc_ns, int mode,
111 struct mq_attr *attr)
113 struct user_struct *u = current_user();
116 inode = new_inode(sb);
118 inode->i_mode = mode;
119 inode->i_uid = current_fsuid();
120 inode->i_gid = current_fsgid();
121 inode->i_mtime = inode->i_ctime = inode->i_atime =
125 struct mqueue_inode_info *info;
126 struct task_struct *p = current;
127 unsigned long mq_bytes, mq_msg_tblsz;
129 inode->i_fop = &mqueue_file_operations;
130 inode->i_size = FILENT_SIZE;
131 /* mqueue specific info */
132 info = MQUEUE_I(inode);
133 spin_lock_init(&info->lock);
134 init_waitqueue_head(&info->wait_q);
135 INIT_LIST_HEAD(&info->e_wait_q[0].list);
136 INIT_LIST_HEAD(&info->e_wait_q[1].list);
137 info->notify_owner = NULL;
139 info->user = NULL; /* set when all is ok */
140 memset(&info->attr, 0, sizeof(info->attr));
141 info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
142 info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
144 info->attr.mq_maxmsg = attr->mq_maxmsg;
145 info->attr.mq_msgsize = attr->mq_msgsize;
147 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
148 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
152 mq_bytes = (mq_msg_tblsz +
153 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
156 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
157 u->mq_bytes + mq_bytes >
158 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
159 spin_unlock(&mq_lock);
160 kfree(info->messages);
163 u->mq_bytes += mq_bytes;
164 spin_unlock(&mq_lock);
167 info->user = get_uid(u);
168 } else if (S_ISDIR(mode)) {
170 /* Some things misbehave if size == 0 on a directory */
171 inode->i_size = 2 * DIRENT_SIZE;
172 inode->i_op = &mqueue_dir_inode_operations;
173 inode->i_fop = &simple_dir_operations;
178 make_bad_inode(inode);
183 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
186 struct ipc_namespace *ns = data;
189 sb->s_blocksize = PAGE_CACHE_SIZE;
190 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
191 sb->s_magic = MQUEUE_MAGIC;
192 sb->s_op = &mqueue_super_ops;
194 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
201 sb->s_root = d_alloc_root(inode);
211 static int mqueue_get_sb(struct file_system_type *fs_type,
212 int flags, const char *dev_name,
213 void *data, struct vfsmount *mnt)
215 if (!(flags & MS_KERNMOUNT))
216 data = current->nsproxy->ipc_ns;
217 return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
220 static void init_once(void *foo)
222 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
224 inode_init_once(&p->vfs_inode);
227 static struct inode *mqueue_alloc_inode(struct super_block *sb)
229 struct mqueue_inode_info *ei;
231 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
234 return &ei->vfs_inode;
237 static void mqueue_destroy_inode(struct inode *inode)
239 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
242 static void mqueue_delete_inode(struct inode *inode)
244 struct mqueue_inode_info *info;
245 struct user_struct *user;
246 unsigned long mq_bytes;
248 struct ipc_namespace *ipc_ns;
250 if (S_ISDIR(inode->i_mode)) {
254 ipc_ns = get_ns_from_inode(inode);
255 info = MQUEUE_I(inode);
256 spin_lock(&info->lock);
257 for (i = 0; i < info->attr.mq_curmsgs; i++)
258 free_msg(info->messages[i]);
259 kfree(info->messages);
260 spin_unlock(&info->lock);
264 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
265 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
269 user->mq_bytes -= mq_bytes;
271 * get_ns_from_inode() ensures that the
272 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
273 * to which we now hold a reference, or it is NULL.
274 * We can't put it here under mq_lock, though.
277 ipc_ns->mq_queues_count--;
278 spin_unlock(&mq_lock);
285 static int mqueue_create(struct inode *dir, struct dentry *dentry,
286 int mode, struct nameidata *nd)
289 struct mq_attr *attr = dentry->d_fsdata;
291 struct ipc_namespace *ipc_ns;
294 ipc_ns = __get_ns_from_inode(dir);
299 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
300 !capable(CAP_SYS_RESOURCE)) {
304 ipc_ns->mq_queues_count++;
305 spin_unlock(&mq_lock);
307 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
311 ipc_ns->mq_queues_count--;
316 dir->i_size += DIRENT_SIZE;
317 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
319 d_instantiate(dentry, inode);
323 spin_unlock(&mq_lock);
329 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
331 struct inode *inode = dentry->d_inode;
333 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
334 dir->i_size -= DIRENT_SIZE;
341 * This is routine for system read from queue file.
342 * To avoid mess with doing here some sort of mq_receive we allow
343 * to read only queue size & notification info (the only values
344 * that are interesting from user point of view and aren't accessible
345 * through std routines)
347 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
348 size_t count, loff_t *off)
350 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
351 char buffer[FILENT_SIZE];
354 spin_lock(&info->lock);
355 snprintf(buffer, sizeof(buffer),
356 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
358 info->notify_owner ? info->notify.sigev_notify : 0,
359 (info->notify_owner &&
360 info->notify.sigev_notify == SIGEV_SIGNAL) ?
361 info->notify.sigev_signo : 0,
362 pid_vnr(info->notify_owner));
363 spin_unlock(&info->lock);
364 buffer[sizeof(buffer)-1] = '\0';
366 ret = simple_read_from_buffer(u_data, count, off, buffer,
371 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
375 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
377 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
379 spin_lock(&info->lock);
380 if (task_tgid(current) == info->notify_owner)
381 remove_notification(info);
383 spin_unlock(&info->lock);
387 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
389 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
392 poll_wait(filp, &info->wait_q, poll_tab);
394 spin_lock(&info->lock);
395 if (info->attr.mq_curmsgs)
396 retval = POLLIN | POLLRDNORM;
398 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
399 retval |= POLLOUT | POLLWRNORM;
400 spin_unlock(&info->lock);
405 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
406 static void wq_add(struct mqueue_inode_info *info, int sr,
407 struct ext_wait_queue *ewp)
409 struct ext_wait_queue *walk;
413 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
414 if (walk->task->static_prio <= current->static_prio) {
415 list_add_tail(&ewp->list, &walk->list);
419 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
423 * Puts current task to sleep. Caller must hold queue lock. After return
427 static int wq_sleep(struct mqueue_inode_info *info, int sr,
428 long timeout, struct ext_wait_queue *ewp)
433 wq_add(info, sr, ewp);
436 set_current_state(TASK_INTERRUPTIBLE);
438 spin_unlock(&info->lock);
439 time = schedule_timeout(timeout);
441 while (ewp->state == STATE_PENDING)
444 if (ewp->state == STATE_READY) {
448 spin_lock(&info->lock);
449 if (ewp->state == STATE_READY) {
453 if (signal_pending(current)) {
454 retval = -ERESTARTSYS;
462 list_del(&ewp->list);
464 spin_unlock(&info->lock);
470 * Returns waiting task that should be serviced first or NULL if none exists
472 static struct ext_wait_queue *wq_get_first_waiter(
473 struct mqueue_inode_info *info, int sr)
475 struct list_head *ptr;
477 ptr = info->e_wait_q[sr].list.prev;
478 if (ptr == &info->e_wait_q[sr].list)
480 return list_entry(ptr, struct ext_wait_queue, list);
483 /* Auxiliary functions to manipulate messages' list */
484 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
488 k = info->attr.mq_curmsgs - 1;
489 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
490 info->messages[k + 1] = info->messages[k];
493 info->attr.mq_curmsgs++;
494 info->qsize += ptr->m_ts;
495 info->messages[k + 1] = ptr;
498 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
500 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
501 return info->messages[info->attr.mq_curmsgs];
504 static inline void set_cookie(struct sk_buff *skb, char code)
506 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
510 * The next function is only to split too long sys_mq_timedsend
512 static void __do_notify(struct mqueue_inode_info *info)
515 * invoked when there is registered process and there isn't process
516 * waiting synchronously for message AND state of queue changed from
517 * empty to not empty. Here we are sure that no one is waiting
519 if (info->notify_owner &&
520 info->attr.mq_curmsgs == 1) {
521 struct siginfo sig_i;
522 switch (info->notify.sigev_notify) {
528 sig_i.si_signo = info->notify.sigev_signo;
530 sig_i.si_code = SI_MESGQ;
531 sig_i.si_value = info->notify.sigev_value;
532 sig_i.si_pid = task_tgid_nr_ns(current,
533 ns_of_pid(info->notify_owner));
534 sig_i.si_uid = current_uid();
536 kill_pid_info(info->notify.sigev_signo,
537 &sig_i, info->notify_owner);
540 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
541 netlink_sendskb(info->notify_sock, info->notify_cookie);
544 /* after notification unregisters process */
545 put_pid(info->notify_owner);
546 info->notify_owner = NULL;
548 wake_up(&info->wait_q);
551 static long prepare_timeout(struct timespec *p)
553 struct timespec nowts;
557 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
558 || p->tv_nsec >= NSEC_PER_SEC))
560 nowts = CURRENT_TIME;
561 /* first subtract as jiffies can't be too big */
562 p->tv_sec -= nowts.tv_sec;
563 if (p->tv_nsec < nowts.tv_nsec) {
564 p->tv_nsec += NSEC_PER_SEC;
567 p->tv_nsec -= nowts.tv_nsec;
571 timeout = timespec_to_jiffies(p) + 1;
573 return MAX_SCHEDULE_TIMEOUT;
578 static void remove_notification(struct mqueue_inode_info *info)
580 if (info->notify_owner != NULL &&
581 info->notify.sigev_notify == SIGEV_THREAD) {
582 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
583 netlink_sendskb(info->notify_sock, info->notify_cookie);
585 put_pid(info->notify_owner);
586 info->notify_owner = NULL;
589 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
591 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
593 if (capable(CAP_SYS_RESOURCE)) {
594 if (attr->mq_maxmsg > HARD_MSGMAX)
597 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
598 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
601 /* check for overflow */
602 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
604 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
605 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
606 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
612 * Invoked when creating a new queue via sys_mq_open
614 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
615 struct dentry *dentry, int oflag, mode_t mode,
616 struct mq_attr *attr)
618 const struct cred *cred = current_cred();
624 if (!mq_attr_ok(ipc_ns, attr))
626 /* store for use during create */
627 dentry->d_fsdata = attr;
630 mode &= ~current_umask();
631 ret = mnt_want_write(ipc_ns->mq_mnt);
634 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
635 dentry->d_fsdata = NULL;
639 result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
641 * dentry_open() took a persistent mnt_want_write(),
642 * so we can now drop this one.
644 mnt_drop_write(ipc_ns->mq_mnt);
648 mnt_drop_write(ipc_ns->mq_mnt);
651 mntput(ipc_ns->mq_mnt);
655 /* Opens existing queue */
656 static struct file *do_open(struct ipc_namespace *ipc_ns,
657 struct dentry *dentry, int oflag)
659 const struct cred *cred = current_cred();
661 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
662 MAY_READ | MAY_WRITE };
664 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
666 mntput(ipc_ns->mq_mnt);
667 return ERR_PTR(-EINVAL);
670 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
672 mntput(ipc_ns->mq_mnt);
673 return ERR_PTR(-EACCES);
676 return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
679 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
680 struct mq_attr __user *, u_attr)
682 struct dentry *dentry;
687 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
689 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
692 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
694 if (IS_ERR(name = getname(u_name)))
695 return PTR_ERR(name);
697 fd = get_unused_fd_flags(O_CLOEXEC);
701 mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
702 dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
703 if (IS_ERR(dentry)) {
704 error = PTR_ERR(dentry);
707 mntget(ipc_ns->mq_mnt);
709 if (oflag & O_CREAT) {
710 if (dentry->d_inode) { /* entry already exists */
711 audit_inode(name, dentry);
715 filp = do_open(ipc_ns, dentry, oflag);
717 filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
719 u_attr ? &attr : NULL);
723 if (!dentry->d_inode)
725 audit_inode(name, dentry);
726 filp = do_open(ipc_ns, dentry, oflag);
730 error = PTR_ERR(filp);
734 fd_install(fd, filp);
739 mntput(ipc_ns->mq_mnt);
744 mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
750 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
754 struct dentry *dentry;
755 struct inode *inode = NULL;
756 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
758 name = getname(u_name);
760 return PTR_ERR(name);
762 mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
764 dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
765 if (IS_ERR(dentry)) {
766 err = PTR_ERR(dentry);
770 if (!dentry->d_inode) {
775 inode = dentry->d_inode;
777 atomic_inc(&inode->i_count);
778 err = mnt_want_write(ipc_ns->mq_mnt);
781 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
782 mnt_drop_write(ipc_ns->mq_mnt);
787 mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
795 /* Pipelined send and receive functions.
797 * If a receiver finds no waiting message, then it registers itself in the
798 * list of waiting receivers. A sender checks that list before adding the new
799 * message into the message array. If there is a waiting receiver, then it
800 * bypasses the message array and directly hands the message over to the
802 * The receiver accepts the message and returns without grabbing the queue
803 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
804 * are necessary. The same algorithm is used for sysv semaphores, see
805 * ipc/sem.c for more details.
807 * The same algorithm is used for senders.
810 /* pipelined_send() - send a message directly to the task waiting in
811 * sys_mq_timedreceive() (without inserting message into a queue).
813 static inline void pipelined_send(struct mqueue_inode_info *info,
814 struct msg_msg *message,
815 struct ext_wait_queue *receiver)
817 receiver->msg = message;
818 list_del(&receiver->list);
819 receiver->state = STATE_PENDING;
820 wake_up_process(receiver->task);
822 receiver->state = STATE_READY;
825 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
826 * gets its message and put to the queue (we have one free place for sure). */
827 static inline void pipelined_receive(struct mqueue_inode_info *info)
829 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
833 wake_up_interruptible(&info->wait_q);
836 msg_insert(sender->msg, info);
837 list_del(&sender->list);
838 sender->state = STATE_PENDING;
839 wake_up_process(sender->task);
841 sender->state = STATE_READY;
844 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
845 size_t, msg_len, unsigned int, msg_prio,
846 const struct timespec __user *, u_abs_timeout)
850 struct ext_wait_queue wait;
851 struct ext_wait_queue *receiver;
852 struct msg_msg *msg_ptr;
853 struct mqueue_inode_info *info;
854 struct timespec ts, *p = NULL;
859 if (copy_from_user(&ts, u_abs_timeout,
860 sizeof(struct timespec)))
865 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
868 audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
869 timeout = prepare_timeout(p);
876 inode = filp->f_path.dentry->d_inode;
877 if (unlikely(filp->f_op != &mqueue_file_operations))
879 info = MQUEUE_I(inode);
880 audit_inode(NULL, filp->f_path.dentry);
882 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
885 if (unlikely(msg_len > info->attr.mq_msgsize)) {
890 /* First try to allocate memory, before doing anything with
891 * existing queues. */
892 msg_ptr = load_msg(u_msg_ptr, msg_len);
893 if (IS_ERR(msg_ptr)) {
894 ret = PTR_ERR(msg_ptr);
897 msg_ptr->m_ts = msg_len;
898 msg_ptr->m_type = msg_prio;
900 spin_lock(&info->lock);
902 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
903 if (filp->f_flags & O_NONBLOCK) {
904 spin_unlock(&info->lock);
906 } else if (unlikely(timeout < 0)) {
907 spin_unlock(&info->lock);
911 wait.msg = (void *) msg_ptr;
912 wait.state = STATE_NONE;
913 ret = wq_sleep(info, SEND, timeout, &wait);
918 receiver = wq_get_first_waiter(info, RECV);
920 pipelined_send(info, msg_ptr, receiver);
922 /* adds message to the queue */
923 msg_insert(msg_ptr, info);
926 inode->i_atime = inode->i_mtime = inode->i_ctime =
928 spin_unlock(&info->lock);
937 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
938 size_t, msg_len, unsigned int __user *, u_msg_prio,
939 const struct timespec __user *, u_abs_timeout)
943 struct msg_msg *msg_ptr;
946 struct mqueue_inode_info *info;
947 struct ext_wait_queue wait;
948 struct timespec ts, *p = NULL;
951 if (copy_from_user(&ts, u_abs_timeout,
952 sizeof(struct timespec)))
957 audit_mq_sendrecv(mqdes, msg_len, 0, p);
958 timeout = prepare_timeout(p);
965 inode = filp->f_path.dentry->d_inode;
966 if (unlikely(filp->f_op != &mqueue_file_operations))
968 info = MQUEUE_I(inode);
969 audit_inode(NULL, filp->f_path.dentry);
971 if (unlikely(!(filp->f_mode & FMODE_READ)))
974 /* checks if buffer is big enough */
975 if (unlikely(msg_len < info->attr.mq_msgsize)) {
980 spin_lock(&info->lock);
981 if (info->attr.mq_curmsgs == 0) {
982 if (filp->f_flags & O_NONBLOCK) {
983 spin_unlock(&info->lock);
986 } else if (unlikely(timeout < 0)) {
987 spin_unlock(&info->lock);
992 wait.state = STATE_NONE;
993 ret = wq_sleep(info, RECV, timeout, &wait);
997 msg_ptr = msg_get(info);
999 inode->i_atime = inode->i_mtime = inode->i_ctime =
1002 /* There is now free space in queue. */
1003 pipelined_receive(info);
1004 spin_unlock(&info->lock);
1008 ret = msg_ptr->m_ts;
1010 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1011 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1023 * Notes: the case when user wants us to deregister (with NULL as pointer)
1024 * and he isn't currently owner of notification, will be silently discarded.
1025 * It isn't explicitly defined in the POSIX.
1027 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1028 const struct sigevent __user *, u_notification)
1033 struct inode *inode;
1034 struct sigevent notification;
1035 struct mqueue_inode_info *info;
1038 if (u_notification) {
1039 if (copy_from_user(¬ification, u_notification,
1040 sizeof(struct sigevent)))
1044 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1048 if (u_notification != NULL) {
1049 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1050 notification.sigev_notify != SIGEV_SIGNAL &&
1051 notification.sigev_notify != SIGEV_THREAD))
1053 if (notification.sigev_notify == SIGEV_SIGNAL &&
1054 !valid_signal(notification.sigev_signo)) {
1057 if (notification.sigev_notify == SIGEV_THREAD) {
1060 /* create the notify skb */
1061 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1066 if (copy_from_user(nc->data,
1067 notification.sigev_value.sival_ptr,
1068 NOTIFY_COOKIE_LEN)) {
1072 /* TODO: add a header? */
1073 skb_put(nc, NOTIFY_COOKIE_LEN);
1074 /* and attach it to the socket */
1076 filp = fget(notification.sigev_signo);
1080 sock = netlink_getsockbyfilp(filp);
1083 ret = PTR_ERR(sock);
1088 timeo = MAX_SCHEDULE_TIMEOUT;
1089 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1105 inode = filp->f_path.dentry->d_inode;
1106 if (unlikely(filp->f_op != &mqueue_file_operations))
1108 info = MQUEUE_I(inode);
1111 spin_lock(&info->lock);
1112 if (u_notification == NULL) {
1113 if (info->notify_owner == task_tgid(current)) {
1114 remove_notification(info);
1115 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1117 } else if (info->notify_owner != NULL) {
1120 switch (notification.sigev_notify) {
1122 info->notify.sigev_notify = SIGEV_NONE;
1125 info->notify_sock = sock;
1126 info->notify_cookie = nc;
1129 info->notify.sigev_notify = SIGEV_THREAD;
1132 info->notify.sigev_signo = notification.sigev_signo;
1133 info->notify.sigev_value = notification.sigev_value;
1134 info->notify.sigev_notify = SIGEV_SIGNAL;
1138 info->notify_owner = get_pid(task_tgid(current));
1139 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1141 spin_unlock(&info->lock);
1146 netlink_detachskb(sock, nc);
1153 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1154 const struct mq_attr __user *, u_mqstat,
1155 struct mq_attr __user *, u_omqstat)
1158 struct mq_attr mqstat, omqstat;
1160 struct inode *inode;
1161 struct mqueue_inode_info *info;
1163 if (u_mqstat != NULL) {
1164 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1166 if (mqstat.mq_flags & (~O_NONBLOCK))
1175 inode = filp->f_path.dentry->d_inode;
1176 if (unlikely(filp->f_op != &mqueue_file_operations))
1178 info = MQUEUE_I(inode);
1180 spin_lock(&info->lock);
1182 omqstat = info->attr;
1183 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1185 audit_mq_getsetattr(mqdes, &mqstat);
1186 spin_lock(&filp->f_lock);
1187 if (mqstat.mq_flags & O_NONBLOCK)
1188 filp->f_flags |= O_NONBLOCK;
1190 filp->f_flags &= ~O_NONBLOCK;
1191 spin_unlock(&filp->f_lock);
1193 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1196 spin_unlock(&info->lock);
1199 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1200 sizeof(struct mq_attr)))
1209 static const struct inode_operations mqueue_dir_inode_operations = {
1210 .lookup = simple_lookup,
1211 .create = mqueue_create,
1212 .unlink = mqueue_unlink,
1215 static const struct file_operations mqueue_file_operations = {
1216 .flush = mqueue_flush_file,
1217 .poll = mqueue_poll_file,
1218 .read = mqueue_read_file,
1221 static const struct super_operations mqueue_super_ops = {
1222 .alloc_inode = mqueue_alloc_inode,
1223 .destroy_inode = mqueue_destroy_inode,
1224 .statfs = simple_statfs,
1225 .delete_inode = mqueue_delete_inode,
1226 .drop_inode = generic_delete_inode,
1229 static struct file_system_type mqueue_fs_type = {
1231 .get_sb = mqueue_get_sb,
1232 .kill_sb = kill_litter_super,
1235 int mq_init_ns(struct ipc_namespace *ns)
1237 ns->mq_queues_count = 0;
1238 ns->mq_queues_max = DFLT_QUEUESMAX;
1239 ns->mq_msg_max = DFLT_MSGMAX;
1240 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1242 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1243 if (IS_ERR(ns->mq_mnt)) {
1244 int err = PTR_ERR(ns->mq_mnt);
1251 void mq_clear_sbinfo(struct ipc_namespace *ns)
1253 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1256 void mq_put_mnt(struct ipc_namespace *ns)
1261 static int __init init_mqueue_fs(void)
1265 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1266 sizeof(struct mqueue_inode_info), 0,
1267 SLAB_HWCACHE_ALIGN, init_once);
1268 if (mqueue_inode_cachep == NULL)
1271 /* ignore failues - they are not fatal */
1272 mq_sysctl_table = mq_register_sysctl_table();
1274 error = register_filesystem(&mqueue_fs_type);
1278 spin_lock_init(&mq_lock);
1280 init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1281 if (IS_ERR(init_ipc_ns.mq_mnt)) {
1282 error = PTR_ERR(init_ipc_ns.mq_mnt);
1283 goto out_filesystem;
1289 unregister_filesystem(&mqueue_fs_type);
1291 if (mq_sysctl_table)
1292 unregister_sysctl_table(mq_sysctl_table);
1293 kmem_cache_destroy(mqueue_inode_cachep);
1297 __initcall(init_mqueue_fs);