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>
36 #define MQUEUE_MAGIC 0x19800202
37 #define DIRENT_SIZE 20
38 #define FILENT_SIZE 80
44 #define STATE_PENDING 1
49 #define CTL_QUEUESMAX 2
51 #define CTL_MSGSIZEMAX 4
54 #define DFLT_QUEUESMAX 256 /* max number of message queues */
55 #define DFLT_MSGMAX 10 /* max number of messages in each queue */
56 #define HARD_MSGMAX (131072/sizeof(void*))
57 #define DFLT_MSGSIZEMAX 8192 /* max message size */
60 struct ext_wait_queue { /* queue of sleeping tasks */
61 struct task_struct *task;
62 struct list_head list;
63 struct msg_msg *msg; /* ptr of loaded message */
64 int state; /* one of STATE_* values */
67 struct mqueue_inode_info {
69 struct inode vfs_inode;
70 wait_queue_head_t wait_q;
72 struct msg_msg **messages;
75 struct sigevent notify;
76 struct pid* notify_owner;
77 struct user_struct *user; /* user who created, for accounting */
78 struct sock *notify_sock;
79 struct sk_buff *notify_cookie;
81 /* for tasks waiting for free space and messages, respectively */
82 struct ext_wait_queue e_wait_q[2];
84 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
92 static spinlock_t mq_lock;
93 static struct kmem_cache *mqueue_inode_cachep;
94 static struct vfsmount *mqueue_mnt;
96 static unsigned int queues_count;
97 static unsigned int queues_max = DFLT_QUEUESMAX;
98 static unsigned int msg_max = DFLT_MSGMAX;
99 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
101 static struct ctl_table_header * mq_sysctl_table;
103 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
105 return container_of(inode, struct mqueue_inode_info, vfs_inode);
108 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
109 struct mq_attr *attr)
113 inode = new_inode(sb);
115 inode->i_mode = mode;
116 inode->i_uid = current->fsuid;
117 inode->i_gid = current->fsgid;
119 inode->i_mtime = inode->i_ctime = inode->i_atime =
123 struct mqueue_inode_info *info;
124 struct task_struct *p = current;
125 struct user_struct *u = p->user;
126 unsigned long mq_bytes, mq_msg_tblsz;
128 inode->i_fop = &mqueue_file_operations;
129 inode->i_size = FILENT_SIZE;
130 /* mqueue specific info */
131 info = MQUEUE_I(inode);
132 spin_lock_init(&info->lock);
133 init_waitqueue_head(&info->wait_q);
134 INIT_LIST_HEAD(&info->e_wait_q[0].list);
135 INIT_LIST_HEAD(&info->e_wait_q[1].list);
136 info->messages = NULL;
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 = DFLT_MSGMAX;
142 info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
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 mq_bytes = (mq_msg_tblsz +
149 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
152 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
153 u->mq_bytes + mq_bytes >
154 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
155 spin_unlock(&mq_lock);
158 u->mq_bytes += mq_bytes;
159 spin_unlock(&mq_lock);
161 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
162 if (!info->messages) {
164 u->mq_bytes -= mq_bytes;
165 spin_unlock(&mq_lock);
169 info->user = get_uid(u);
170 } else if (S_ISDIR(mode)) {
172 /* Some things misbehave if size == 0 on a directory */
173 inode->i_size = 2 * DIRENT_SIZE;
174 inode->i_op = &mqueue_dir_inode_operations;
175 inode->i_fop = &simple_dir_operations;
180 make_bad_inode(inode);
185 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
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, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
198 sb->s_root = d_alloc_root(inode);
207 static int mqueue_get_sb(struct file_system_type *fs_type,
208 int flags, const char *dev_name,
209 void *data, struct vfsmount *mnt)
211 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
214 static void init_once(void *foo, struct kmem_cache * cachep, unsigned long flags)
216 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
218 inode_init_once(&p->vfs_inode);
221 static struct inode *mqueue_alloc_inode(struct super_block *sb)
223 struct mqueue_inode_info *ei;
225 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
228 return &ei->vfs_inode;
231 static void mqueue_destroy_inode(struct inode *inode)
233 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
236 static void mqueue_delete_inode(struct inode *inode)
238 struct mqueue_inode_info *info;
239 struct user_struct *user;
240 unsigned long mq_bytes;
243 if (S_ISDIR(inode->i_mode)) {
247 info = MQUEUE_I(inode);
248 spin_lock(&info->lock);
249 for (i = 0; i < info->attr.mq_curmsgs; i++)
250 free_msg(info->messages[i]);
251 kfree(info->messages);
252 spin_unlock(&info->lock);
256 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
257 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
261 user->mq_bytes -= mq_bytes;
263 spin_unlock(&mq_lock);
268 static int mqueue_create(struct inode *dir, struct dentry *dentry,
269 int mode, struct nameidata *nd)
272 struct mq_attr *attr = dentry->d_fsdata;
276 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
281 spin_unlock(&mq_lock);
283 inode = mqueue_get_inode(dir->i_sb, mode, attr);
291 dir->i_size += DIRENT_SIZE;
292 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
294 d_instantiate(dentry, inode);
298 spin_unlock(&mq_lock);
302 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
304 struct inode *inode = dentry->d_inode;
306 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
307 dir->i_size -= DIRENT_SIZE;
314 * This is routine for system read from queue file.
315 * To avoid mess with doing here some sort of mq_receive we allow
316 * to read only queue size & notification info (the only values
317 * that are interesting from user point of view and aren't accessible
318 * through std routines)
320 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
321 size_t count, loff_t * off)
323 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
324 char buffer[FILENT_SIZE];
331 spin_lock(&info->lock);
332 snprintf(buffer, sizeof(buffer),
333 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
335 info->notify_owner ? info->notify.sigev_notify : 0,
336 (info->notify_owner &&
337 info->notify.sigev_notify == SIGEV_SIGNAL) ?
338 info->notify.sigev_signo : 0,
339 pid_nr(info->notify_owner));
340 spin_unlock(&info->lock);
341 buffer[sizeof(buffer)-1] = '\0';
342 slen = strlen(buffer)+1;
348 if (o + count > slen)
351 if (copy_to_user(u_data, buffer + o, count))
355 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
359 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
361 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
363 spin_lock(&info->lock);
364 if (task_tgid(current) == info->notify_owner)
365 remove_notification(info);
367 spin_unlock(&info->lock);
371 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
373 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
376 poll_wait(filp, &info->wait_q, poll_tab);
378 spin_lock(&info->lock);
379 if (info->attr.mq_curmsgs)
380 retval = POLLIN | POLLRDNORM;
382 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
383 retval |= POLLOUT | POLLWRNORM;
384 spin_unlock(&info->lock);
389 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
390 static void wq_add(struct mqueue_inode_info *info, int sr,
391 struct ext_wait_queue *ewp)
393 struct ext_wait_queue *walk;
397 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
398 if (walk->task->static_prio <= current->static_prio) {
399 list_add_tail(&ewp->list, &walk->list);
403 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
407 * Puts current task to sleep. Caller must hold queue lock. After return
411 static int wq_sleep(struct mqueue_inode_info *info, int sr,
412 long timeout, struct ext_wait_queue *ewp)
417 wq_add(info, sr, ewp);
420 set_current_state(TASK_INTERRUPTIBLE);
422 spin_unlock(&info->lock);
423 time = schedule_timeout(timeout);
425 while (ewp->state == STATE_PENDING)
428 if (ewp->state == STATE_READY) {
432 spin_lock(&info->lock);
433 if (ewp->state == STATE_READY) {
437 if (signal_pending(current)) {
438 retval = -ERESTARTSYS;
446 list_del(&ewp->list);
448 spin_unlock(&info->lock);
454 * Returns waiting task that should be serviced first or NULL if none exists
456 static struct ext_wait_queue *wq_get_first_waiter(
457 struct mqueue_inode_info *info, int sr)
459 struct list_head *ptr;
461 ptr = info->e_wait_q[sr].list.prev;
462 if (ptr == &info->e_wait_q[sr].list)
464 return list_entry(ptr, struct ext_wait_queue, list);
467 /* Auxiliary functions to manipulate messages' list */
468 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
472 k = info->attr.mq_curmsgs - 1;
473 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
474 info->messages[k + 1] = info->messages[k];
477 info->attr.mq_curmsgs++;
478 info->qsize += ptr->m_ts;
479 info->messages[k + 1] = ptr;
482 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
484 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
485 return info->messages[info->attr.mq_curmsgs];
488 static inline void set_cookie(struct sk_buff *skb, char code)
490 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
494 * The next function is only to split too long sys_mq_timedsend
496 static void __do_notify(struct mqueue_inode_info *info)
499 * invoked when there is registered process and there isn't process
500 * waiting synchronously for message AND state of queue changed from
501 * empty to not empty. Here we are sure that no one is waiting
503 if (info->notify_owner &&
504 info->attr.mq_curmsgs == 1) {
505 struct siginfo sig_i;
506 switch (info->notify.sigev_notify) {
512 sig_i.si_signo = info->notify.sigev_signo;
514 sig_i.si_code = SI_MESGQ;
515 sig_i.si_value = info->notify.sigev_value;
516 sig_i.si_pid = current->tgid;
517 sig_i.si_uid = current->uid;
519 kill_pid_info(info->notify.sigev_signo,
520 &sig_i, info->notify_owner);
523 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
524 netlink_sendskb(info->notify_sock,
525 info->notify_cookie, 0);
528 /* after notification unregisters process */
529 put_pid(info->notify_owner);
530 info->notify_owner = NULL;
532 wake_up(&info->wait_q);
535 static long prepare_timeout(const struct timespec __user *u_arg)
537 struct timespec ts, nowts;
541 if (unlikely(copy_from_user(&ts, u_arg,
542 sizeof(struct timespec))))
545 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
546 || ts.tv_nsec >= NSEC_PER_SEC))
548 nowts = CURRENT_TIME;
549 /* first subtract as jiffies can't be too big */
550 ts.tv_sec -= nowts.tv_sec;
551 if (ts.tv_nsec < nowts.tv_nsec) {
552 ts.tv_nsec += NSEC_PER_SEC;
555 ts.tv_nsec -= nowts.tv_nsec;
559 timeout = timespec_to_jiffies(&ts) + 1;
561 return MAX_SCHEDULE_TIMEOUT;
566 static void remove_notification(struct mqueue_inode_info *info)
568 if (info->notify_owner != NULL &&
569 info->notify.sigev_notify == SIGEV_THREAD) {
570 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
571 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
573 put_pid(info->notify_owner);
574 info->notify_owner = NULL;
577 static int mq_attr_ok(struct mq_attr *attr)
579 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
581 if (capable(CAP_SYS_RESOURCE)) {
582 if (attr->mq_maxmsg > HARD_MSGMAX)
585 if (attr->mq_maxmsg > msg_max ||
586 attr->mq_msgsize > msgsize_max)
589 /* check for overflow */
590 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
592 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
593 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
594 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
600 * Invoked when creating a new queue via sys_mq_open
602 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
603 int oflag, mode_t mode, struct mq_attr __user *u_attr)
610 if (copy_from_user(&attr, u_attr, sizeof(attr)))
613 if (!mq_attr_ok(&attr))
615 /* store for use during create */
616 dentry->d_fsdata = &attr;
619 mode &= ~current->fs->umask;
620 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
621 dentry->d_fsdata = NULL;
625 return dentry_open(dentry, mqueue_mnt, oflag);
633 /* Opens existing queue */
634 static struct file *do_open(struct dentry *dentry, int oflag)
636 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
637 MAY_READ | MAY_WRITE };
639 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
642 return ERR_PTR(-EINVAL);
645 if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
648 return ERR_PTR(-EACCES);
651 return dentry_open(dentry, mqueue_mnt, oflag);
654 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
655 struct mq_attr __user *u_attr)
657 struct dentry *dentry;
662 error = audit_mq_open(oflag, mode, u_attr);
666 if (IS_ERR(name = getname(u_name)))
667 return PTR_ERR(name);
669 fd = get_unused_fd();
673 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
674 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
675 if (IS_ERR(dentry)) {
676 error = PTR_ERR(dentry);
681 if (oflag & O_CREAT) {
682 if (dentry->d_inode) { /* entry already exists */
683 audit_inode(name, dentry->d_inode);
687 filp = do_open(dentry, oflag);
689 filp = do_create(mqueue_mnt->mnt_root, dentry,
690 oflag, mode, u_attr);
694 if (!dentry->d_inode)
696 audit_inode(name, dentry->d_inode);
697 filp = do_open(dentry, oflag);
701 error = PTR_ERR(filp);
705 set_close_on_exec(fd, 1);
706 fd_install(fd, filp);
717 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
723 asmlinkage long sys_mq_unlink(const char __user *u_name)
727 struct dentry *dentry;
728 struct inode *inode = NULL;
730 name = getname(u_name);
732 return PTR_ERR(name);
734 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
736 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
737 if (IS_ERR(dentry)) {
738 err = PTR_ERR(dentry);
742 if (!dentry->d_inode) {
747 inode = dentry->d_inode;
749 atomic_inc(&inode->i_count);
751 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
756 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
764 /* Pipelined send and receive functions.
766 * If a receiver finds no waiting message, then it registers itself in the
767 * list of waiting receivers. A sender checks that list before adding the new
768 * message into the message array. If there is a waiting receiver, then it
769 * bypasses the message array and directly hands the message over to the
771 * The receiver accepts the message and returns without grabbing the queue
772 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
773 * are necessary. The same algorithm is used for sysv semaphores, see
774 * ipc/sem.c for more details.
776 * The same algorithm is used for senders.
779 /* pipelined_send() - send a message directly to the task waiting in
780 * sys_mq_timedreceive() (without inserting message into a queue).
782 static inline void pipelined_send(struct mqueue_inode_info *info,
783 struct msg_msg *message,
784 struct ext_wait_queue *receiver)
786 receiver->msg = message;
787 list_del(&receiver->list);
788 receiver->state = STATE_PENDING;
789 wake_up_process(receiver->task);
791 receiver->state = STATE_READY;
794 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
795 * gets its message and put to the queue (we have one free place for sure). */
796 static inline void pipelined_receive(struct mqueue_inode_info *info)
798 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
802 wake_up_interruptible(&info->wait_q);
805 msg_insert(sender->msg, info);
806 list_del(&sender->list);
807 sender->state = STATE_PENDING;
808 wake_up_process(sender->task);
810 sender->state = STATE_READY;
813 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
814 size_t msg_len, unsigned int msg_prio,
815 const struct timespec __user *u_abs_timeout)
819 struct ext_wait_queue wait;
820 struct ext_wait_queue *receiver;
821 struct msg_msg *msg_ptr;
822 struct mqueue_inode_info *info;
826 ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
830 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
833 timeout = prepare_timeout(u_abs_timeout);
840 inode = filp->f_path.dentry->d_inode;
841 if (unlikely(filp->f_op != &mqueue_file_operations))
843 info = MQUEUE_I(inode);
844 audit_inode(NULL, inode);
846 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
849 if (unlikely(msg_len > info->attr.mq_msgsize)) {
854 /* First try to allocate memory, before doing anything with
855 * existing queues. */
856 msg_ptr = load_msg(u_msg_ptr, msg_len);
857 if (IS_ERR(msg_ptr)) {
858 ret = PTR_ERR(msg_ptr);
861 msg_ptr->m_ts = msg_len;
862 msg_ptr->m_type = msg_prio;
864 spin_lock(&info->lock);
866 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
867 if (filp->f_flags & O_NONBLOCK) {
868 spin_unlock(&info->lock);
870 } else if (unlikely(timeout < 0)) {
871 spin_unlock(&info->lock);
875 wait.msg = (void *) msg_ptr;
876 wait.state = STATE_NONE;
877 ret = wq_sleep(info, SEND, timeout, &wait);
882 receiver = wq_get_first_waiter(info, RECV);
884 pipelined_send(info, msg_ptr, receiver);
886 /* adds message to the queue */
887 msg_insert(msg_ptr, info);
890 inode->i_atime = inode->i_mtime = inode->i_ctime =
892 spin_unlock(&info->lock);
901 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
902 size_t msg_len, unsigned int __user *u_msg_prio,
903 const struct timespec __user *u_abs_timeout)
907 struct msg_msg *msg_ptr;
910 struct mqueue_inode_info *info;
911 struct ext_wait_queue wait;
913 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
917 timeout = prepare_timeout(u_abs_timeout);
924 inode = filp->f_path.dentry->d_inode;
925 if (unlikely(filp->f_op != &mqueue_file_operations))
927 info = MQUEUE_I(inode);
928 audit_inode(NULL, inode);
930 if (unlikely(!(filp->f_mode & FMODE_READ)))
933 /* checks if buffer is big enough */
934 if (unlikely(msg_len < info->attr.mq_msgsize)) {
939 spin_lock(&info->lock);
940 if (info->attr.mq_curmsgs == 0) {
941 if (filp->f_flags & O_NONBLOCK) {
942 spin_unlock(&info->lock);
945 } else if (unlikely(timeout < 0)) {
946 spin_unlock(&info->lock);
951 wait.state = STATE_NONE;
952 ret = wq_sleep(info, RECV, timeout, &wait);
956 msg_ptr = msg_get(info);
958 inode->i_atime = inode->i_mtime = inode->i_ctime =
961 /* There is now free space in queue. */
962 pipelined_receive(info);
963 spin_unlock(&info->lock);
969 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
970 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
982 * Notes: the case when user wants us to deregister (with NULL as pointer)
983 * and he isn't currently owner of notification, will be silently discarded.
984 * It isn't explicitly defined in the POSIX.
986 asmlinkage long sys_mq_notify(mqd_t mqdes,
987 const struct sigevent __user *u_notification)
993 struct sigevent notification;
994 struct mqueue_inode_info *info;
997 ret = audit_mq_notify(mqdes, u_notification);
1003 if (u_notification != NULL) {
1004 if (copy_from_user(¬ification, u_notification,
1005 sizeof(struct sigevent)))
1008 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1009 notification.sigev_notify != SIGEV_SIGNAL &&
1010 notification.sigev_notify != SIGEV_THREAD))
1012 if (notification.sigev_notify == SIGEV_SIGNAL &&
1013 !valid_signal(notification.sigev_signo)) {
1016 if (notification.sigev_notify == SIGEV_THREAD) {
1017 /* create the notify skb */
1018 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1023 if (copy_from_user(nc->data,
1024 notification.sigev_value.sival_ptr,
1025 NOTIFY_COOKIE_LEN)) {
1029 /* TODO: add a header? */
1030 skb_put(nc, NOTIFY_COOKIE_LEN);
1031 /* and attach it to the socket */
1033 filp = fget(notification.sigev_signo);
1037 sock = netlink_getsockbyfilp(filp);
1040 ret = PTR_ERR(sock);
1045 ret = netlink_attachskb(sock, nc, 0,
1046 MAX_SCHEDULE_TIMEOUT, NULL);
1062 inode = filp->f_path.dentry->d_inode;
1063 if (unlikely(filp->f_op != &mqueue_file_operations))
1065 info = MQUEUE_I(inode);
1068 spin_lock(&info->lock);
1069 if (u_notification == NULL) {
1070 if (info->notify_owner == task_tgid(current)) {
1071 remove_notification(info);
1072 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1074 } else if (info->notify_owner != NULL) {
1077 switch (notification.sigev_notify) {
1079 info->notify.sigev_notify = SIGEV_NONE;
1082 info->notify_sock = sock;
1083 info->notify_cookie = nc;
1086 info->notify.sigev_notify = SIGEV_THREAD;
1089 info->notify.sigev_signo = notification.sigev_signo;
1090 info->notify.sigev_value = notification.sigev_value;
1091 info->notify.sigev_notify = SIGEV_SIGNAL;
1095 info->notify_owner = get_pid(task_tgid(current));
1096 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1098 spin_unlock(&info->lock);
1103 netlink_detachskb(sock, nc);
1110 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1111 const struct mq_attr __user *u_mqstat,
1112 struct mq_attr __user *u_omqstat)
1115 struct mq_attr mqstat, omqstat;
1117 struct inode *inode;
1118 struct mqueue_inode_info *info;
1120 if (u_mqstat != NULL) {
1121 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1123 if (mqstat.mq_flags & (~O_NONBLOCK))
1132 inode = filp->f_path.dentry->d_inode;
1133 if (unlikely(filp->f_op != &mqueue_file_operations))
1135 info = MQUEUE_I(inode);
1137 spin_lock(&info->lock);
1139 omqstat = info->attr;
1140 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1142 ret = audit_mq_getsetattr(mqdes, &mqstat);
1145 if (mqstat.mq_flags & O_NONBLOCK)
1146 filp->f_flags |= O_NONBLOCK;
1148 filp->f_flags &= ~O_NONBLOCK;
1150 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1153 spin_unlock(&info->lock);
1156 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1157 sizeof(struct mq_attr)))
1166 static const struct inode_operations mqueue_dir_inode_operations = {
1167 .lookup = simple_lookup,
1168 .create = mqueue_create,
1169 .unlink = mqueue_unlink,
1172 static const struct file_operations mqueue_file_operations = {
1173 .flush = mqueue_flush_file,
1174 .poll = mqueue_poll_file,
1175 .read = mqueue_read_file,
1178 static struct super_operations mqueue_super_ops = {
1179 .alloc_inode = mqueue_alloc_inode,
1180 .destroy_inode = mqueue_destroy_inode,
1181 .statfs = simple_statfs,
1182 .delete_inode = mqueue_delete_inode,
1183 .drop_inode = generic_delete_inode,
1186 static struct file_system_type mqueue_fs_type = {
1188 .get_sb = mqueue_get_sb,
1189 .kill_sb = kill_litter_super,
1192 static int msg_max_limit_min = DFLT_MSGMAX;
1193 static int msg_max_limit_max = HARD_MSGMAX;
1195 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1196 static int msg_maxsize_limit_max = INT_MAX;
1198 static ctl_table mq_sysctls[] = {
1200 .ctl_name = CTL_QUEUESMAX,
1201 .procname = "queues_max",
1202 .data = &queues_max,
1203 .maxlen = sizeof(int),
1205 .proc_handler = &proc_dointvec,
1208 .ctl_name = CTL_MSGMAX,
1209 .procname = "msg_max",
1211 .maxlen = sizeof(int),
1213 .proc_handler = &proc_dointvec_minmax,
1214 .extra1 = &msg_max_limit_min,
1215 .extra2 = &msg_max_limit_max,
1218 .ctl_name = CTL_MSGSIZEMAX,
1219 .procname = "msgsize_max",
1220 .data = &msgsize_max,
1221 .maxlen = sizeof(int),
1223 .proc_handler = &proc_dointvec_minmax,
1224 .extra1 = &msg_maxsize_limit_min,
1225 .extra2 = &msg_maxsize_limit_max,
1230 static ctl_table mq_sysctl_dir[] = {
1232 .ctl_name = FS_MQUEUE,
1233 .procname = "mqueue",
1235 .child = mq_sysctls,
1240 static ctl_table mq_sysctl_root[] = {
1245 .child = mq_sysctl_dir,
1250 static int __init init_mqueue_fs(void)
1254 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1255 sizeof(struct mqueue_inode_info), 0,
1256 SLAB_HWCACHE_ALIGN, init_once, NULL);
1257 if (mqueue_inode_cachep == NULL)
1260 /* ignore failues - they are not fatal */
1261 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1263 error = register_filesystem(&mqueue_fs_type);
1267 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1268 error = PTR_ERR(mqueue_mnt);
1269 goto out_filesystem;
1272 /* internal initialization - not common for vfs */
1274 spin_lock_init(&mq_lock);
1279 unregister_filesystem(&mqueue_fs_type);
1281 if (mq_sysctl_table)
1282 unregister_sysctl_table(mq_sysctl_table);
1283 kmem_cache_destroy(mqueue_inode_cachep);
1287 __initcall(init_mqueue_fs);