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 * Define the ranges various user-specified maximum values can
59 #define MIN_MSGMAX 1 /* min value for msg_max */
60 #define MAX_MSGMAX HARD_MSGMAX /* max value for msg_max */
61 #define MIN_MSGSIZEMAX 128 /* min value for msgsize_max */
62 #define MAX_MSGSIZEMAX (8192*128) /* max value for msgsize_max */
64 struct ext_wait_queue { /* queue of sleeping tasks */
65 struct task_struct *task;
66 struct list_head list;
67 struct msg_msg *msg; /* ptr of loaded message */
68 int state; /* one of STATE_* values */
71 struct mqueue_inode_info {
73 struct inode vfs_inode;
74 wait_queue_head_t wait_q;
76 struct msg_msg **messages;
79 struct sigevent notify;
80 struct pid* notify_owner;
81 struct user_struct *user; /* user who created, for accounting */
82 struct sock *notify_sock;
83 struct sk_buff *notify_cookie;
85 /* for tasks waiting for free space and messages, respectively */
86 struct ext_wait_queue e_wait_q[2];
88 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
91 static const struct inode_operations mqueue_dir_inode_operations;
92 static const struct file_operations mqueue_file_operations;
93 static struct super_operations mqueue_super_ops;
94 static void remove_notification(struct mqueue_inode_info *info);
96 static spinlock_t mq_lock;
97 static struct kmem_cache *mqueue_inode_cachep;
98 static struct vfsmount *mqueue_mnt;
100 static unsigned int queues_count;
101 static unsigned int queues_max = DFLT_QUEUESMAX;
102 static unsigned int msg_max = DFLT_MSGMAX;
103 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
105 static struct ctl_table_header * mq_sysctl_table;
107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
109 return container_of(inode, struct mqueue_inode_info, vfs_inode);
112 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
113 struct mq_attr *attr)
115 struct user_struct *u = current_user();
118 inode = new_inode(sb);
120 inode->i_mode = mode;
121 inode->i_uid = current_fsuid();
122 inode->i_gid = current_fsgid();
123 inode->i_mtime = inode->i_ctime = inode->i_atime =
127 struct mqueue_inode_info *info;
128 struct task_struct *p = current;
129 unsigned long mq_bytes, mq_msg_tblsz;
131 inode->i_fop = &mqueue_file_operations;
132 inode->i_size = FILENT_SIZE;
133 /* mqueue specific info */
134 info = MQUEUE_I(inode);
135 spin_lock_init(&info->lock);
136 init_waitqueue_head(&info->wait_q);
137 INIT_LIST_HEAD(&info->e_wait_q[0].list);
138 INIT_LIST_HEAD(&info->e_wait_q[1].list);
139 info->messages = NULL;
140 info->notify_owner = NULL;
142 info->user = NULL; /* set when all is ok */
143 memset(&info->attr, 0, sizeof(info->attr));
144 info->attr.mq_maxmsg = msg_max;
145 info->attr.mq_msgsize = msgsize_max;
147 info->attr.mq_maxmsg = attr->mq_maxmsg;
148 info->attr.mq_msgsize = attr->mq_msgsize;
150 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
151 mq_bytes = (mq_msg_tblsz +
152 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
155 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
156 u->mq_bytes + mq_bytes >
157 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
158 spin_unlock(&mq_lock);
161 u->mq_bytes += mq_bytes;
162 spin_unlock(&mq_lock);
164 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
165 if (!info->messages) {
167 u->mq_bytes -= mq_bytes;
168 spin_unlock(&mq_lock);
172 info->user = get_uid(u);
173 } else if (S_ISDIR(mode)) {
175 /* Some things misbehave if size == 0 on a directory */
176 inode->i_size = 2 * DIRENT_SIZE;
177 inode->i_op = &mqueue_dir_inode_operations;
178 inode->i_fop = &simple_dir_operations;
183 make_bad_inode(inode);
188 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
192 sb->s_blocksize = PAGE_CACHE_SIZE;
193 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
194 sb->s_magic = MQUEUE_MAGIC;
195 sb->s_op = &mqueue_super_ops;
197 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
201 sb->s_root = d_alloc_root(inode);
210 static int mqueue_get_sb(struct file_system_type *fs_type,
211 int flags, const char *dev_name,
212 void *data, struct vfsmount *mnt)
214 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
217 static void init_once(void *foo)
219 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
221 inode_init_once(&p->vfs_inode);
224 static struct inode *mqueue_alloc_inode(struct super_block *sb)
226 struct mqueue_inode_info *ei;
228 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
231 return &ei->vfs_inode;
234 static void mqueue_destroy_inode(struct inode *inode)
236 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
239 static void mqueue_delete_inode(struct inode *inode)
241 struct mqueue_inode_info *info;
242 struct user_struct *user;
243 unsigned long mq_bytes;
246 if (S_ISDIR(inode->i_mode)) {
250 info = MQUEUE_I(inode);
251 spin_lock(&info->lock);
252 for (i = 0; i < info->attr.mq_curmsgs; i++)
253 free_msg(info->messages[i]);
254 kfree(info->messages);
255 spin_unlock(&info->lock);
259 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
260 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
264 user->mq_bytes -= mq_bytes;
266 spin_unlock(&mq_lock);
271 static int mqueue_create(struct inode *dir, struct dentry *dentry,
272 int mode, struct nameidata *nd)
275 struct mq_attr *attr = dentry->d_fsdata;
279 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
284 spin_unlock(&mq_lock);
286 inode = mqueue_get_inode(dir->i_sb, mode, attr);
294 dir->i_size += DIRENT_SIZE;
295 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
297 d_instantiate(dentry, inode);
301 spin_unlock(&mq_lock);
305 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
307 struct inode *inode = dentry->d_inode;
309 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
310 dir->i_size -= DIRENT_SIZE;
317 * This is routine for system read from queue file.
318 * To avoid mess with doing here some sort of mq_receive we allow
319 * to read only queue size & notification info (the only values
320 * that are interesting from user point of view and aren't accessible
321 * through std routines)
323 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
324 size_t count, loff_t *off)
326 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
327 char buffer[FILENT_SIZE];
330 spin_lock(&info->lock);
331 snprintf(buffer, sizeof(buffer),
332 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
334 info->notify_owner ? info->notify.sigev_notify : 0,
335 (info->notify_owner &&
336 info->notify.sigev_notify == SIGEV_SIGNAL) ?
337 info->notify.sigev_signo : 0,
338 pid_vnr(info->notify_owner));
339 spin_unlock(&info->lock);
340 buffer[sizeof(buffer)-1] = '\0';
342 ret = simple_read_from_buffer(u_data, count, off, buffer,
347 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
351 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
353 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
355 spin_lock(&info->lock);
356 if (task_tgid(current) == info->notify_owner)
357 remove_notification(info);
359 spin_unlock(&info->lock);
363 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
365 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
368 poll_wait(filp, &info->wait_q, poll_tab);
370 spin_lock(&info->lock);
371 if (info->attr.mq_curmsgs)
372 retval = POLLIN | POLLRDNORM;
374 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
375 retval |= POLLOUT | POLLWRNORM;
376 spin_unlock(&info->lock);
381 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
382 static void wq_add(struct mqueue_inode_info *info, int sr,
383 struct ext_wait_queue *ewp)
385 struct ext_wait_queue *walk;
389 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
390 if (walk->task->static_prio <= current->static_prio) {
391 list_add_tail(&ewp->list, &walk->list);
395 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
399 * Puts current task to sleep. Caller must hold queue lock. After return
403 static int wq_sleep(struct mqueue_inode_info *info, int sr,
404 long timeout, struct ext_wait_queue *ewp)
409 wq_add(info, sr, ewp);
412 set_current_state(TASK_INTERRUPTIBLE);
414 spin_unlock(&info->lock);
415 time = schedule_timeout(timeout);
417 while (ewp->state == STATE_PENDING)
420 if (ewp->state == STATE_READY) {
424 spin_lock(&info->lock);
425 if (ewp->state == STATE_READY) {
429 if (signal_pending(current)) {
430 retval = -ERESTARTSYS;
438 list_del(&ewp->list);
440 spin_unlock(&info->lock);
446 * Returns waiting task that should be serviced first or NULL if none exists
448 static struct ext_wait_queue *wq_get_first_waiter(
449 struct mqueue_inode_info *info, int sr)
451 struct list_head *ptr;
453 ptr = info->e_wait_q[sr].list.prev;
454 if (ptr == &info->e_wait_q[sr].list)
456 return list_entry(ptr, struct ext_wait_queue, list);
459 /* Auxiliary functions to manipulate messages' list */
460 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
464 k = info->attr.mq_curmsgs - 1;
465 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
466 info->messages[k + 1] = info->messages[k];
469 info->attr.mq_curmsgs++;
470 info->qsize += ptr->m_ts;
471 info->messages[k + 1] = ptr;
474 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
476 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
477 return info->messages[info->attr.mq_curmsgs];
480 static inline void set_cookie(struct sk_buff *skb, char code)
482 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
486 * The next function is only to split too long sys_mq_timedsend
488 static void __do_notify(struct mqueue_inode_info *info)
491 * invoked when there is registered process and there isn't process
492 * waiting synchronously for message AND state of queue changed from
493 * empty to not empty. Here we are sure that no one is waiting
495 if (info->notify_owner &&
496 info->attr.mq_curmsgs == 1) {
497 struct siginfo sig_i;
498 switch (info->notify.sigev_notify) {
504 sig_i.si_signo = info->notify.sigev_signo;
506 sig_i.si_code = SI_MESGQ;
507 sig_i.si_value = info->notify.sigev_value;
508 sig_i.si_pid = task_tgid_vnr(current);
509 sig_i.si_uid = current_uid();
511 kill_pid_info(info->notify.sigev_signo,
512 &sig_i, info->notify_owner);
515 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
516 netlink_sendskb(info->notify_sock, info->notify_cookie);
519 /* after notification unregisters process */
520 put_pid(info->notify_owner);
521 info->notify_owner = NULL;
523 wake_up(&info->wait_q);
526 static long prepare_timeout(const struct timespec __user *u_arg)
528 struct timespec ts, nowts;
532 if (unlikely(copy_from_user(&ts, u_arg,
533 sizeof(struct timespec))))
536 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
537 || ts.tv_nsec >= NSEC_PER_SEC))
539 nowts = CURRENT_TIME;
540 /* first subtract as jiffies can't be too big */
541 ts.tv_sec -= nowts.tv_sec;
542 if (ts.tv_nsec < nowts.tv_nsec) {
543 ts.tv_nsec += NSEC_PER_SEC;
546 ts.tv_nsec -= nowts.tv_nsec;
550 timeout = timespec_to_jiffies(&ts) + 1;
552 return MAX_SCHEDULE_TIMEOUT;
557 static void remove_notification(struct mqueue_inode_info *info)
559 if (info->notify_owner != NULL &&
560 info->notify.sigev_notify == SIGEV_THREAD) {
561 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
562 netlink_sendskb(info->notify_sock, info->notify_cookie);
564 put_pid(info->notify_owner);
565 info->notify_owner = NULL;
568 static int mq_attr_ok(struct mq_attr *attr)
570 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
572 if (capable(CAP_SYS_RESOURCE)) {
573 if (attr->mq_maxmsg > HARD_MSGMAX)
576 if (attr->mq_maxmsg > msg_max ||
577 attr->mq_msgsize > msgsize_max)
580 /* check for overflow */
581 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
583 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
584 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
585 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
591 * Invoked when creating a new queue via sys_mq_open
593 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
594 int oflag, mode_t mode, struct mq_attr __user *u_attr)
596 const struct cred *cred = current_cred();
603 if (copy_from_user(&attr, u_attr, sizeof(attr)))
606 if (!mq_attr_ok(&attr))
608 /* store for use during create */
609 dentry->d_fsdata = &attr;
612 mode &= ~current->fs->umask;
613 ret = mnt_want_write(mqueue_mnt);
616 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
617 dentry->d_fsdata = NULL;
621 result = dentry_open(dentry, mqueue_mnt, oflag, cred);
623 * dentry_open() took a persistent mnt_want_write(),
624 * so we can now drop this one.
626 mnt_drop_write(mqueue_mnt);
630 mnt_drop_write(mqueue_mnt);
637 /* Opens existing queue */
638 static struct file *do_open(struct dentry *dentry, int oflag)
640 const struct cred *cred = current_cred();
642 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
643 MAY_READ | MAY_WRITE };
645 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
648 return ERR_PTR(-EINVAL);
651 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
654 return ERR_PTR(-EACCES);
657 return dentry_open(dentry, mqueue_mnt, oflag, cred);
660 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
661 struct mq_attr __user *u_attr)
663 struct dentry *dentry;
668 error = audit_mq_open(oflag, mode, u_attr);
672 if (IS_ERR(name = getname(u_name)))
673 return PTR_ERR(name);
675 fd = get_unused_fd_flags(O_CLOEXEC);
679 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
680 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
681 if (IS_ERR(dentry)) {
682 error = PTR_ERR(dentry);
687 if (oflag & O_CREAT) {
688 if (dentry->d_inode) { /* entry already exists */
689 audit_inode(name, dentry);
693 filp = do_open(dentry, oflag);
695 filp = do_create(mqueue_mnt->mnt_root, dentry,
696 oflag, mode, u_attr);
700 if (!dentry->d_inode)
702 audit_inode(name, dentry);
703 filp = do_open(dentry, oflag);
707 error = PTR_ERR(filp);
711 fd_install(fd, filp);
722 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
728 asmlinkage long sys_mq_unlink(const char __user *u_name)
732 struct dentry *dentry;
733 struct inode *inode = NULL;
735 name = getname(u_name);
737 return PTR_ERR(name);
739 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
741 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
742 if (IS_ERR(dentry)) {
743 err = PTR_ERR(dentry);
747 if (!dentry->d_inode) {
752 inode = dentry->d_inode;
754 atomic_inc(&inode->i_count);
755 err = mnt_want_write(mqueue_mnt);
758 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
759 mnt_drop_write(mqueue_mnt);
764 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
772 /* Pipelined send and receive functions.
774 * If a receiver finds no waiting message, then it registers itself in the
775 * list of waiting receivers. A sender checks that list before adding the new
776 * message into the message array. If there is a waiting receiver, then it
777 * bypasses the message array and directly hands the message over to the
779 * The receiver accepts the message and returns without grabbing the queue
780 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
781 * are necessary. The same algorithm is used for sysv semaphores, see
782 * ipc/sem.c for more details.
784 * The same algorithm is used for senders.
787 /* pipelined_send() - send a message directly to the task waiting in
788 * sys_mq_timedreceive() (without inserting message into a queue).
790 static inline void pipelined_send(struct mqueue_inode_info *info,
791 struct msg_msg *message,
792 struct ext_wait_queue *receiver)
794 receiver->msg = message;
795 list_del(&receiver->list);
796 receiver->state = STATE_PENDING;
797 wake_up_process(receiver->task);
799 receiver->state = STATE_READY;
802 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
803 * gets its message and put to the queue (we have one free place for sure). */
804 static inline void pipelined_receive(struct mqueue_inode_info *info)
806 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
810 wake_up_interruptible(&info->wait_q);
813 msg_insert(sender->msg, info);
814 list_del(&sender->list);
815 sender->state = STATE_PENDING;
816 wake_up_process(sender->task);
818 sender->state = STATE_READY;
821 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
822 size_t msg_len, unsigned int msg_prio,
823 const struct timespec __user *u_abs_timeout)
827 struct ext_wait_queue wait;
828 struct ext_wait_queue *receiver;
829 struct msg_msg *msg_ptr;
830 struct mqueue_inode_info *info;
834 ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
838 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
841 timeout = prepare_timeout(u_abs_timeout);
848 inode = filp->f_path.dentry->d_inode;
849 if (unlikely(filp->f_op != &mqueue_file_operations))
851 info = MQUEUE_I(inode);
852 audit_inode(NULL, filp->f_path.dentry);
854 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
857 if (unlikely(msg_len > info->attr.mq_msgsize)) {
862 /* First try to allocate memory, before doing anything with
863 * existing queues. */
864 msg_ptr = load_msg(u_msg_ptr, msg_len);
865 if (IS_ERR(msg_ptr)) {
866 ret = PTR_ERR(msg_ptr);
869 msg_ptr->m_ts = msg_len;
870 msg_ptr->m_type = msg_prio;
872 spin_lock(&info->lock);
874 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
875 if (filp->f_flags & O_NONBLOCK) {
876 spin_unlock(&info->lock);
878 } else if (unlikely(timeout < 0)) {
879 spin_unlock(&info->lock);
883 wait.msg = (void *) msg_ptr;
884 wait.state = STATE_NONE;
885 ret = wq_sleep(info, SEND, timeout, &wait);
890 receiver = wq_get_first_waiter(info, RECV);
892 pipelined_send(info, msg_ptr, receiver);
894 /* adds message to the queue */
895 msg_insert(msg_ptr, info);
898 inode->i_atime = inode->i_mtime = inode->i_ctime =
900 spin_unlock(&info->lock);
909 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
910 size_t msg_len, unsigned int __user *u_msg_prio,
911 const struct timespec __user *u_abs_timeout)
915 struct msg_msg *msg_ptr;
918 struct mqueue_inode_info *info;
919 struct ext_wait_queue wait;
921 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
925 timeout = prepare_timeout(u_abs_timeout);
932 inode = filp->f_path.dentry->d_inode;
933 if (unlikely(filp->f_op != &mqueue_file_operations))
935 info = MQUEUE_I(inode);
936 audit_inode(NULL, filp->f_path.dentry);
938 if (unlikely(!(filp->f_mode & FMODE_READ)))
941 /* checks if buffer is big enough */
942 if (unlikely(msg_len < info->attr.mq_msgsize)) {
947 spin_lock(&info->lock);
948 if (info->attr.mq_curmsgs == 0) {
949 if (filp->f_flags & O_NONBLOCK) {
950 spin_unlock(&info->lock);
953 } else if (unlikely(timeout < 0)) {
954 spin_unlock(&info->lock);
959 wait.state = STATE_NONE;
960 ret = wq_sleep(info, RECV, timeout, &wait);
964 msg_ptr = msg_get(info);
966 inode->i_atime = inode->i_mtime = inode->i_ctime =
969 /* There is now free space in queue. */
970 pipelined_receive(info);
971 spin_unlock(&info->lock);
977 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
978 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
990 * Notes: the case when user wants us to deregister (with NULL as pointer)
991 * and he isn't currently owner of notification, will be silently discarded.
992 * It isn't explicitly defined in the POSIX.
994 asmlinkage long sys_mq_notify(mqd_t mqdes,
995 const struct sigevent __user *u_notification)
1000 struct inode *inode;
1001 struct sigevent notification;
1002 struct mqueue_inode_info *info;
1005 ret = audit_mq_notify(mqdes, u_notification);
1011 if (u_notification != NULL) {
1012 if (copy_from_user(¬ification, u_notification,
1013 sizeof(struct sigevent)))
1016 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1017 notification.sigev_notify != SIGEV_SIGNAL &&
1018 notification.sigev_notify != SIGEV_THREAD))
1020 if (notification.sigev_notify == SIGEV_SIGNAL &&
1021 !valid_signal(notification.sigev_signo)) {
1024 if (notification.sigev_notify == SIGEV_THREAD) {
1027 /* create the notify skb */
1028 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1033 if (copy_from_user(nc->data,
1034 notification.sigev_value.sival_ptr,
1035 NOTIFY_COOKIE_LEN)) {
1039 /* TODO: add a header? */
1040 skb_put(nc, NOTIFY_COOKIE_LEN);
1041 /* and attach it to the socket */
1043 filp = fget(notification.sigev_signo);
1047 sock = netlink_getsockbyfilp(filp);
1050 ret = PTR_ERR(sock);
1055 timeo = MAX_SCHEDULE_TIMEOUT;
1056 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1072 inode = filp->f_path.dentry->d_inode;
1073 if (unlikely(filp->f_op != &mqueue_file_operations))
1075 info = MQUEUE_I(inode);
1078 spin_lock(&info->lock);
1079 if (u_notification == NULL) {
1080 if (info->notify_owner == task_tgid(current)) {
1081 remove_notification(info);
1082 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1084 } else if (info->notify_owner != NULL) {
1087 switch (notification.sigev_notify) {
1089 info->notify.sigev_notify = SIGEV_NONE;
1092 info->notify_sock = sock;
1093 info->notify_cookie = nc;
1096 info->notify.sigev_notify = SIGEV_THREAD;
1099 info->notify.sigev_signo = notification.sigev_signo;
1100 info->notify.sigev_value = notification.sigev_value;
1101 info->notify.sigev_notify = SIGEV_SIGNAL;
1105 info->notify_owner = get_pid(task_tgid(current));
1106 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1108 spin_unlock(&info->lock);
1113 netlink_detachskb(sock, nc);
1120 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1121 const struct mq_attr __user *u_mqstat,
1122 struct mq_attr __user *u_omqstat)
1125 struct mq_attr mqstat, omqstat;
1127 struct inode *inode;
1128 struct mqueue_inode_info *info;
1130 if (u_mqstat != NULL) {
1131 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1133 if (mqstat.mq_flags & (~O_NONBLOCK))
1142 inode = filp->f_path.dentry->d_inode;
1143 if (unlikely(filp->f_op != &mqueue_file_operations))
1145 info = MQUEUE_I(inode);
1147 spin_lock(&info->lock);
1149 omqstat = info->attr;
1150 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1152 ret = audit_mq_getsetattr(mqdes, &mqstat);
1154 spin_unlock(&info->lock);
1157 if (mqstat.mq_flags & O_NONBLOCK)
1158 filp->f_flags |= O_NONBLOCK;
1160 filp->f_flags &= ~O_NONBLOCK;
1162 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1165 spin_unlock(&info->lock);
1168 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1169 sizeof(struct mq_attr)))
1178 static const struct inode_operations mqueue_dir_inode_operations = {
1179 .lookup = simple_lookup,
1180 .create = mqueue_create,
1181 .unlink = mqueue_unlink,
1184 static const struct file_operations mqueue_file_operations = {
1185 .flush = mqueue_flush_file,
1186 .poll = mqueue_poll_file,
1187 .read = mqueue_read_file,
1190 static struct super_operations mqueue_super_ops = {
1191 .alloc_inode = mqueue_alloc_inode,
1192 .destroy_inode = mqueue_destroy_inode,
1193 .statfs = simple_statfs,
1194 .delete_inode = mqueue_delete_inode,
1195 .drop_inode = generic_delete_inode,
1198 static struct file_system_type mqueue_fs_type = {
1200 .get_sb = mqueue_get_sb,
1201 .kill_sb = kill_litter_super,
1204 static int msg_max_limit_min = MIN_MSGMAX;
1205 static int msg_max_limit_max = MAX_MSGMAX;
1207 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1208 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1210 static ctl_table mq_sysctls[] = {
1212 .procname = "queues_max",
1213 .data = &queues_max,
1214 .maxlen = sizeof(int),
1216 .proc_handler = &proc_dointvec,
1219 .procname = "msg_max",
1221 .maxlen = sizeof(int),
1223 .proc_handler = &proc_dointvec_minmax,
1224 .extra1 = &msg_max_limit_min,
1225 .extra2 = &msg_max_limit_max,
1228 .procname = "msgsize_max",
1229 .data = &msgsize_max,
1230 .maxlen = sizeof(int),
1232 .proc_handler = &proc_dointvec_minmax,
1233 .extra1 = &msg_maxsize_limit_min,
1234 .extra2 = &msg_maxsize_limit_max,
1239 static ctl_table mq_sysctl_dir[] = {
1241 .procname = "mqueue",
1243 .child = mq_sysctls,
1248 static ctl_table mq_sysctl_root[] = {
1253 .child = mq_sysctl_dir,
1258 static int __init init_mqueue_fs(void)
1262 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1263 sizeof(struct mqueue_inode_info), 0,
1264 SLAB_HWCACHE_ALIGN, init_once);
1265 if (mqueue_inode_cachep == NULL)
1268 /* ignore failues - they are not fatal */
1269 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1271 error = register_filesystem(&mqueue_fs_type);
1275 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1276 error = PTR_ERR(mqueue_mnt);
1277 goto out_filesystem;
1280 /* internal initialization - not common for vfs */
1282 spin_lock_init(&mq_lock);
1287 unregister_filesystem(&mqueue_fs_type);
1289 if (mq_sysctl_table)
1290 unregister_sysctl_table(mq_sysctl_table);
1291 kmem_cache_destroy(mqueue_inode_cachep);
1295 __initcall(init_mqueue_fs);