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>
38 #define MQUEUE_MAGIC 0x19800202
39 #define DIRENT_SIZE 20
40 #define FILENT_SIZE 80
46 #define STATE_PENDING 1
50 #define DFLT_QUEUESMAX 256 /* max number of message queues */
51 #define DFLT_MSGMAX 10 /* max number of messages in each queue */
52 #define HARD_MSGMAX (131072/sizeof(void*))
53 #define DFLT_MSGSIZEMAX 8192 /* max message size */
56 struct ext_wait_queue { /* queue of sleeping tasks */
57 struct task_struct *task;
58 struct list_head list;
59 struct msg_msg *msg; /* ptr of loaded message */
60 int state; /* one of STATE_* values */
63 struct mqueue_inode_info {
65 struct inode vfs_inode;
66 wait_queue_head_t wait_q;
68 struct msg_msg **messages;
71 struct sigevent notify;
72 struct pid* notify_owner;
73 struct user_struct *user; /* user who created, for accounting */
74 struct sock *notify_sock;
75 struct sk_buff *notify_cookie;
77 /* for tasks waiting for free space and messages, respectively */
78 struct ext_wait_queue e_wait_q[2];
80 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
83 static const struct inode_operations mqueue_dir_inode_operations;
84 static const struct file_operations mqueue_file_operations;
85 static struct super_operations mqueue_super_ops;
86 static void remove_notification(struct mqueue_inode_info *info);
88 static spinlock_t mq_lock;
89 static struct kmem_cache *mqueue_inode_cachep;
90 static struct vfsmount *mqueue_mnt;
92 static unsigned int queues_count;
93 static unsigned int queues_max = DFLT_QUEUESMAX;
94 static unsigned int msg_max = DFLT_MSGMAX;
95 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
97 static struct ctl_table_header * mq_sysctl_table;
99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
104 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
105 struct mq_attr *attr)
109 inode = new_inode(sb);
111 inode->i_mode = mode;
112 inode->i_uid = current->fsuid;
113 inode->i_gid = current->fsgid;
115 inode->i_mtime = inode->i_ctime = inode->i_atime =
119 struct mqueue_inode_info *info;
120 struct task_struct *p = current;
121 struct user_struct *u = p->user;
122 unsigned long mq_bytes, mq_msg_tblsz;
124 inode->i_fop = &mqueue_file_operations;
125 inode->i_size = FILENT_SIZE;
126 /* mqueue specific info */
127 info = MQUEUE_I(inode);
128 spin_lock_init(&info->lock);
129 init_waitqueue_head(&info->wait_q);
130 INIT_LIST_HEAD(&info->e_wait_q[0].list);
131 INIT_LIST_HEAD(&info->e_wait_q[1].list);
132 info->messages = NULL;
133 info->notify_owner = NULL;
135 info->user = NULL; /* set when all is ok */
136 memset(&info->attr, 0, sizeof(info->attr));
137 info->attr.mq_maxmsg = DFLT_MSGMAX;
138 info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
140 info->attr.mq_maxmsg = attr->mq_maxmsg;
141 info->attr.mq_msgsize = attr->mq_msgsize;
143 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
144 mq_bytes = (mq_msg_tblsz +
145 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
148 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
149 u->mq_bytes + mq_bytes >
150 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
151 spin_unlock(&mq_lock);
154 u->mq_bytes += mq_bytes;
155 spin_unlock(&mq_lock);
157 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
158 if (!info->messages) {
160 u->mq_bytes -= mq_bytes;
161 spin_unlock(&mq_lock);
165 info->user = get_uid(u);
166 } else if (S_ISDIR(mode)) {
168 /* Some things misbehave if size == 0 on a directory */
169 inode->i_size = 2 * DIRENT_SIZE;
170 inode->i_op = &mqueue_dir_inode_operations;
171 inode->i_fop = &simple_dir_operations;
176 make_bad_inode(inode);
181 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
185 sb->s_blocksize = PAGE_CACHE_SIZE;
186 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
187 sb->s_magic = MQUEUE_MAGIC;
188 sb->s_op = &mqueue_super_ops;
190 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
194 sb->s_root = d_alloc_root(inode);
203 static int mqueue_get_sb(struct file_system_type *fs_type,
204 int flags, const char *dev_name,
205 void *data, struct vfsmount *mnt)
207 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
210 static void init_once(void *foo)
212 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
214 inode_init_once(&p->vfs_inode);
217 static struct inode *mqueue_alloc_inode(struct super_block *sb)
219 struct mqueue_inode_info *ei;
221 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
224 return &ei->vfs_inode;
227 static void mqueue_destroy_inode(struct inode *inode)
229 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
232 static void mqueue_delete_inode(struct inode *inode)
234 struct mqueue_inode_info *info;
235 struct user_struct *user;
236 unsigned long mq_bytes;
239 if (S_ISDIR(inode->i_mode)) {
243 info = MQUEUE_I(inode);
244 spin_lock(&info->lock);
245 for (i = 0; i < info->attr.mq_curmsgs; i++)
246 free_msg(info->messages[i]);
247 kfree(info->messages);
248 spin_unlock(&info->lock);
252 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
253 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
257 user->mq_bytes -= mq_bytes;
259 spin_unlock(&mq_lock);
264 static int mqueue_create(struct inode *dir, struct dentry *dentry,
265 int mode, struct nameidata *nd)
268 struct mq_attr *attr = dentry->d_fsdata;
272 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
277 spin_unlock(&mq_lock);
279 inode = mqueue_get_inode(dir->i_sb, mode, attr);
287 dir->i_size += DIRENT_SIZE;
288 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
290 d_instantiate(dentry, inode);
294 spin_unlock(&mq_lock);
298 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
300 struct inode *inode = dentry->d_inode;
302 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
303 dir->i_size -= DIRENT_SIZE;
310 * This is routine for system read from queue file.
311 * To avoid mess with doing here some sort of mq_receive we allow
312 * to read only queue size & notification info (the only values
313 * that are interesting from user point of view and aren't accessible
314 * through std routines)
316 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
317 size_t count, loff_t *off)
319 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
320 char buffer[FILENT_SIZE];
323 spin_lock(&info->lock);
324 snprintf(buffer, sizeof(buffer),
325 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
327 info->notify_owner ? info->notify.sigev_notify : 0,
328 (info->notify_owner &&
329 info->notify.sigev_notify == SIGEV_SIGNAL) ?
330 info->notify.sigev_signo : 0,
331 pid_vnr(info->notify_owner));
332 spin_unlock(&info->lock);
333 buffer[sizeof(buffer)-1] = '\0';
335 ret = simple_read_from_buffer(u_data, count, off, buffer,
340 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
344 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
346 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
348 spin_lock(&info->lock);
349 if (task_tgid(current) == info->notify_owner)
350 remove_notification(info);
352 spin_unlock(&info->lock);
356 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
358 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
361 poll_wait(filp, &info->wait_q, poll_tab);
363 spin_lock(&info->lock);
364 if (info->attr.mq_curmsgs)
365 retval = POLLIN | POLLRDNORM;
367 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
368 retval |= POLLOUT | POLLWRNORM;
369 spin_unlock(&info->lock);
374 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
375 static void wq_add(struct mqueue_inode_info *info, int sr,
376 struct ext_wait_queue *ewp)
378 struct ext_wait_queue *walk;
382 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
383 if (walk->task->static_prio <= current->static_prio) {
384 list_add_tail(&ewp->list, &walk->list);
388 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
392 * Puts current task to sleep. Caller must hold queue lock. After return
396 static int wq_sleep(struct mqueue_inode_info *info, int sr,
397 long timeout, struct ext_wait_queue *ewp)
402 wq_add(info, sr, ewp);
405 set_current_state(TASK_INTERRUPTIBLE);
407 spin_unlock(&info->lock);
408 time = schedule_timeout(timeout);
410 while (ewp->state == STATE_PENDING)
413 if (ewp->state == STATE_READY) {
417 spin_lock(&info->lock);
418 if (ewp->state == STATE_READY) {
422 if (signal_pending(current)) {
423 retval = -ERESTARTSYS;
431 list_del(&ewp->list);
433 spin_unlock(&info->lock);
439 * Returns waiting task that should be serviced first or NULL if none exists
441 static struct ext_wait_queue *wq_get_first_waiter(
442 struct mqueue_inode_info *info, int sr)
444 struct list_head *ptr;
446 ptr = info->e_wait_q[sr].list.prev;
447 if (ptr == &info->e_wait_q[sr].list)
449 return list_entry(ptr, struct ext_wait_queue, list);
452 /* Auxiliary functions to manipulate messages' list */
453 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
457 k = info->attr.mq_curmsgs - 1;
458 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
459 info->messages[k + 1] = info->messages[k];
462 info->attr.mq_curmsgs++;
463 info->qsize += ptr->m_ts;
464 info->messages[k + 1] = ptr;
467 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
469 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
470 return info->messages[info->attr.mq_curmsgs];
473 static inline void set_cookie(struct sk_buff *skb, char code)
475 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
479 * The next function is only to split too long sys_mq_timedsend
481 static void __do_notify(struct mqueue_inode_info *info)
484 * invoked when there is registered process and there isn't process
485 * waiting synchronously for message AND state of queue changed from
486 * empty to not empty. Here we are sure that no one is waiting
488 if (info->notify_owner &&
489 info->attr.mq_curmsgs == 1) {
490 struct siginfo sig_i;
491 switch (info->notify.sigev_notify) {
497 sig_i.si_signo = info->notify.sigev_signo;
499 sig_i.si_code = SI_MESGQ;
500 sig_i.si_value = info->notify.sigev_value;
501 sig_i.si_pid = task_tgid_nr_ns(current,
502 ns_of_pid(info->notify_owner));
503 sig_i.si_uid = current->uid;
505 kill_pid_info(info->notify.sigev_signo,
506 &sig_i, info->notify_owner);
509 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
510 netlink_sendskb(info->notify_sock, info->notify_cookie);
513 /* after notification unregisters process */
514 put_pid(info->notify_owner);
515 info->notify_owner = NULL;
517 wake_up(&info->wait_q);
520 static long prepare_timeout(const struct timespec __user *u_arg)
522 struct timespec ts, nowts;
526 if (unlikely(copy_from_user(&ts, u_arg,
527 sizeof(struct timespec))))
530 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
531 || ts.tv_nsec >= NSEC_PER_SEC))
533 nowts = CURRENT_TIME;
534 /* first subtract as jiffies can't be too big */
535 ts.tv_sec -= nowts.tv_sec;
536 if (ts.tv_nsec < nowts.tv_nsec) {
537 ts.tv_nsec += NSEC_PER_SEC;
540 ts.tv_nsec -= nowts.tv_nsec;
544 timeout = timespec_to_jiffies(&ts) + 1;
546 return MAX_SCHEDULE_TIMEOUT;
551 static void remove_notification(struct mqueue_inode_info *info)
553 if (info->notify_owner != NULL &&
554 info->notify.sigev_notify == SIGEV_THREAD) {
555 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
556 netlink_sendskb(info->notify_sock, info->notify_cookie);
558 put_pid(info->notify_owner);
559 info->notify_owner = NULL;
562 static int mq_attr_ok(struct mq_attr *attr)
564 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
566 if (capable(CAP_SYS_RESOURCE)) {
567 if (attr->mq_maxmsg > HARD_MSGMAX)
570 if (attr->mq_maxmsg > msg_max ||
571 attr->mq_msgsize > msgsize_max)
574 /* check for overflow */
575 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
577 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
578 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
579 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
585 * Invoked when creating a new queue via sys_mq_open
587 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
588 int oflag, mode_t mode, struct mq_attr __user *u_attr)
596 if (copy_from_user(&attr, u_attr, sizeof(attr)))
599 if (!mq_attr_ok(&attr))
601 /* store for use during create */
602 dentry->d_fsdata = &attr;
605 mode &= ~current->fs->umask;
606 ret = mnt_want_write(mqueue_mnt);
609 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
610 dentry->d_fsdata = NULL;
614 result = dentry_open(dentry, mqueue_mnt, oflag);
616 * dentry_open() took a persistent mnt_want_write(),
617 * so we can now drop this one.
619 mnt_drop_write(mqueue_mnt);
623 mnt_drop_write(mqueue_mnt);
630 /* Opens existing queue */
631 static struct file *do_open(struct dentry *dentry, int oflag)
633 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
634 MAY_READ | MAY_WRITE };
636 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
639 return ERR_PTR(-EINVAL);
642 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
645 return ERR_PTR(-EACCES);
648 return dentry_open(dentry, mqueue_mnt, oflag);
651 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
652 struct mq_attr __user *, u_attr)
654 struct dentry *dentry;
659 error = audit_mq_open(oflag, mode, u_attr);
663 if (IS_ERR(name = getname(u_name)))
664 return PTR_ERR(name);
666 fd = get_unused_fd_flags(O_CLOEXEC);
670 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
671 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
672 if (IS_ERR(dentry)) {
673 error = PTR_ERR(dentry);
678 if (oflag & O_CREAT) {
679 if (dentry->d_inode) { /* entry already exists */
680 audit_inode(name, dentry);
684 filp = do_open(dentry, oflag);
686 filp = do_create(mqueue_mnt->mnt_root, dentry,
687 oflag, mode, u_attr);
691 if (!dentry->d_inode)
693 audit_inode(name, dentry);
694 filp = do_open(dentry, oflag);
698 error = PTR_ERR(filp);
702 fd_install(fd, filp);
713 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
719 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
723 struct dentry *dentry;
724 struct inode *inode = NULL;
726 name = getname(u_name);
728 return PTR_ERR(name);
730 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
732 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
733 if (IS_ERR(dentry)) {
734 err = PTR_ERR(dentry);
738 if (!dentry->d_inode) {
743 inode = dentry->d_inode;
745 atomic_inc(&inode->i_count);
746 err = mnt_want_write(mqueue_mnt);
749 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
750 mnt_drop_write(mqueue_mnt);
755 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
763 /* Pipelined send and receive functions.
765 * If a receiver finds no waiting message, then it registers itself in the
766 * list of waiting receivers. A sender checks that list before adding the new
767 * message into the message array. If there is a waiting receiver, then it
768 * bypasses the message array and directly hands the message over to the
770 * The receiver accepts the message and returns without grabbing the queue
771 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
772 * are necessary. The same algorithm is used for sysv semaphores, see
773 * ipc/sem.c for more details.
775 * The same algorithm is used for senders.
778 /* pipelined_send() - send a message directly to the task waiting in
779 * sys_mq_timedreceive() (without inserting message into a queue).
781 static inline void pipelined_send(struct mqueue_inode_info *info,
782 struct msg_msg *message,
783 struct ext_wait_queue *receiver)
785 receiver->msg = message;
786 list_del(&receiver->list);
787 receiver->state = STATE_PENDING;
788 wake_up_process(receiver->task);
790 receiver->state = STATE_READY;
793 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
794 * gets its message and put to the queue (we have one free place for sure). */
795 static inline void pipelined_receive(struct mqueue_inode_info *info)
797 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
801 wake_up_interruptible(&info->wait_q);
804 msg_insert(sender->msg, info);
805 list_del(&sender->list);
806 sender->state = STATE_PENDING;
807 wake_up_process(sender->task);
809 sender->state = STATE_READY;
812 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
813 size_t, msg_len, unsigned int, msg_prio,
814 const struct timespec __user *, u_abs_timeout)
818 struct ext_wait_queue wait;
819 struct ext_wait_queue *receiver;
820 struct msg_msg *msg_ptr;
821 struct mqueue_inode_info *info;
825 ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
829 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
832 timeout = prepare_timeout(u_abs_timeout);
839 inode = filp->f_path.dentry->d_inode;
840 if (unlikely(filp->f_op != &mqueue_file_operations))
842 info = MQUEUE_I(inode);
843 audit_inode(NULL, filp->f_path.dentry);
845 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
848 if (unlikely(msg_len > info->attr.mq_msgsize)) {
853 /* First try to allocate memory, before doing anything with
854 * existing queues. */
855 msg_ptr = load_msg(u_msg_ptr, msg_len);
856 if (IS_ERR(msg_ptr)) {
857 ret = PTR_ERR(msg_ptr);
860 msg_ptr->m_ts = msg_len;
861 msg_ptr->m_type = msg_prio;
863 spin_lock(&info->lock);
865 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
866 if (filp->f_flags & O_NONBLOCK) {
867 spin_unlock(&info->lock);
869 } else if (unlikely(timeout < 0)) {
870 spin_unlock(&info->lock);
874 wait.msg = (void *) msg_ptr;
875 wait.state = STATE_NONE;
876 ret = wq_sleep(info, SEND, timeout, &wait);
881 receiver = wq_get_first_waiter(info, RECV);
883 pipelined_send(info, msg_ptr, receiver);
885 /* adds message to the queue */
886 msg_insert(msg_ptr, info);
889 inode->i_atime = inode->i_mtime = inode->i_ctime =
891 spin_unlock(&info->lock);
900 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
901 size_t, msg_len, unsigned int __user *, u_msg_prio,
902 const struct timespec __user *, u_abs_timeout)
906 struct msg_msg *msg_ptr;
909 struct mqueue_inode_info *info;
910 struct ext_wait_queue wait;
912 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
916 timeout = prepare_timeout(u_abs_timeout);
923 inode = filp->f_path.dentry->d_inode;
924 if (unlikely(filp->f_op != &mqueue_file_operations))
926 info = MQUEUE_I(inode);
927 audit_inode(NULL, filp->f_path.dentry);
929 if (unlikely(!(filp->f_mode & FMODE_READ)))
932 /* checks if buffer is big enough */
933 if (unlikely(msg_len < info->attr.mq_msgsize)) {
938 spin_lock(&info->lock);
939 if (info->attr.mq_curmsgs == 0) {
940 if (filp->f_flags & O_NONBLOCK) {
941 spin_unlock(&info->lock);
944 } else if (unlikely(timeout < 0)) {
945 spin_unlock(&info->lock);
950 wait.state = STATE_NONE;
951 ret = wq_sleep(info, RECV, timeout, &wait);
955 msg_ptr = msg_get(info);
957 inode->i_atime = inode->i_mtime = inode->i_ctime =
960 /* There is now free space in queue. */
961 pipelined_receive(info);
962 spin_unlock(&info->lock);
968 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
969 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
981 * Notes: the case when user wants us to deregister (with NULL as pointer)
982 * and he isn't currently owner of notification, will be silently discarded.
983 * It isn't explicitly defined in the POSIX.
985 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
986 const struct sigevent __user *, u_notification)
992 struct sigevent notification;
993 struct mqueue_inode_info *info;
996 ret = audit_mq_notify(mqdes, u_notification);
1002 if (u_notification != NULL) {
1003 if (copy_from_user(¬ification, u_notification,
1004 sizeof(struct sigevent)))
1007 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1008 notification.sigev_notify != SIGEV_SIGNAL &&
1009 notification.sigev_notify != SIGEV_THREAD))
1011 if (notification.sigev_notify == SIGEV_SIGNAL &&
1012 !valid_signal(notification.sigev_signo)) {
1015 if (notification.sigev_notify == SIGEV_THREAD) {
1018 /* create the notify skb */
1019 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1024 if (copy_from_user(nc->data,
1025 notification.sigev_value.sival_ptr,
1026 NOTIFY_COOKIE_LEN)) {
1030 /* TODO: add a header? */
1031 skb_put(nc, NOTIFY_COOKIE_LEN);
1032 /* and attach it to the socket */
1034 filp = fget(notification.sigev_signo);
1038 sock = netlink_getsockbyfilp(filp);
1041 ret = PTR_ERR(sock);
1046 timeo = MAX_SCHEDULE_TIMEOUT;
1047 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1063 inode = filp->f_path.dentry->d_inode;
1064 if (unlikely(filp->f_op != &mqueue_file_operations))
1066 info = MQUEUE_I(inode);
1069 spin_lock(&info->lock);
1070 if (u_notification == NULL) {
1071 if (info->notify_owner == task_tgid(current)) {
1072 remove_notification(info);
1073 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1075 } else if (info->notify_owner != NULL) {
1078 switch (notification.sigev_notify) {
1080 info->notify.sigev_notify = SIGEV_NONE;
1083 info->notify_sock = sock;
1084 info->notify_cookie = nc;
1087 info->notify.sigev_notify = SIGEV_THREAD;
1090 info->notify.sigev_signo = notification.sigev_signo;
1091 info->notify.sigev_value = notification.sigev_value;
1092 info->notify.sigev_notify = SIGEV_SIGNAL;
1096 info->notify_owner = get_pid(task_tgid(current));
1097 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1099 spin_unlock(&info->lock);
1104 netlink_detachskb(sock, nc);
1111 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1112 const struct mq_attr __user *, u_mqstat,
1113 struct mq_attr __user *, u_omqstat)
1116 struct mq_attr mqstat, omqstat;
1118 struct inode *inode;
1119 struct mqueue_inode_info *info;
1121 if (u_mqstat != NULL) {
1122 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1124 if (mqstat.mq_flags & (~O_NONBLOCK))
1133 inode = filp->f_path.dentry->d_inode;
1134 if (unlikely(filp->f_op != &mqueue_file_operations))
1136 info = MQUEUE_I(inode);
1138 spin_lock(&info->lock);
1140 omqstat = info->attr;
1141 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1143 ret = audit_mq_getsetattr(mqdes, &mqstat);
1145 spin_unlock(&info->lock);
1148 if (mqstat.mq_flags & O_NONBLOCK)
1149 filp->f_flags |= O_NONBLOCK;
1151 filp->f_flags &= ~O_NONBLOCK;
1153 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1156 spin_unlock(&info->lock);
1159 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1160 sizeof(struct mq_attr)))
1169 static const struct inode_operations mqueue_dir_inode_operations = {
1170 .lookup = simple_lookup,
1171 .create = mqueue_create,
1172 .unlink = mqueue_unlink,
1175 static const struct file_operations mqueue_file_operations = {
1176 .flush = mqueue_flush_file,
1177 .poll = mqueue_poll_file,
1178 .read = mqueue_read_file,
1181 static struct super_operations mqueue_super_ops = {
1182 .alloc_inode = mqueue_alloc_inode,
1183 .destroy_inode = mqueue_destroy_inode,
1184 .statfs = simple_statfs,
1185 .delete_inode = mqueue_delete_inode,
1186 .drop_inode = generic_delete_inode,
1189 static struct file_system_type mqueue_fs_type = {
1191 .get_sb = mqueue_get_sb,
1192 .kill_sb = kill_litter_super,
1195 static int msg_max_limit_min = DFLT_MSGMAX;
1196 static int msg_max_limit_max = HARD_MSGMAX;
1198 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1199 static int msg_maxsize_limit_max = INT_MAX;
1201 static ctl_table mq_sysctls[] = {
1203 .procname = "queues_max",
1204 .data = &queues_max,
1205 .maxlen = sizeof(int),
1207 .proc_handler = &proc_dointvec,
1210 .procname = "msg_max",
1212 .maxlen = sizeof(int),
1214 .proc_handler = &proc_dointvec_minmax,
1215 .extra1 = &msg_max_limit_min,
1216 .extra2 = &msg_max_limit_max,
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 .procname = "mqueue",
1234 .child = mq_sysctls,
1239 static ctl_table mq_sysctl_root[] = {
1244 .child = mq_sysctl_dir,
1249 static int __init init_mqueue_fs(void)
1253 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1254 sizeof(struct mqueue_inode_info), 0,
1255 SLAB_HWCACHE_ALIGN, init_once);
1256 if (mqueue_inode_cachep == NULL)
1259 /* ignore failues - they are not fatal */
1260 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1262 error = register_filesystem(&mqueue_fs_type);
1266 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1267 error = PTR_ERR(mqueue_mnt);
1268 goto out_filesystem;
1271 /* internal initialization - not common for vfs */
1273 spin_lock_init(&mq_lock);
1278 unregister_filesystem(&mqueue_fs_type);
1280 if (mq_sysctl_table)
1281 unregister_sysctl_table(mq_sysctl_table);
1282 kmem_cache_destroy(mqueue_inode_cachep);
1286 __initcall(init_mqueue_fs);