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_nr_ns(current,
509 ns_of_pid(info->notify_owner));
510 sig_i.si_uid = current_uid();
512 kill_pid_info(info->notify.sigev_signo,
513 &sig_i, info->notify_owner);
516 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
517 netlink_sendskb(info->notify_sock, info->notify_cookie);
520 /* after notification unregisters process */
521 put_pid(info->notify_owner);
522 info->notify_owner = NULL;
524 wake_up(&info->wait_q);
527 static long prepare_timeout(struct timespec *p)
529 struct timespec nowts;
533 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
534 || p->tv_nsec >= NSEC_PER_SEC))
536 nowts = CURRENT_TIME;
537 /* first subtract as jiffies can't be too big */
538 p->tv_sec -= nowts.tv_sec;
539 if (p->tv_nsec < nowts.tv_nsec) {
540 p->tv_nsec += NSEC_PER_SEC;
543 p->tv_nsec -= nowts.tv_nsec;
547 timeout = timespec_to_jiffies(p) + 1;
549 return MAX_SCHEDULE_TIMEOUT;
554 static void remove_notification(struct mqueue_inode_info *info)
556 if (info->notify_owner != NULL &&
557 info->notify.sigev_notify == SIGEV_THREAD) {
558 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
559 netlink_sendskb(info->notify_sock, info->notify_cookie);
561 put_pid(info->notify_owner);
562 info->notify_owner = NULL;
565 static int mq_attr_ok(struct mq_attr *attr)
567 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
569 if (capable(CAP_SYS_RESOURCE)) {
570 if (attr->mq_maxmsg > HARD_MSGMAX)
573 if (attr->mq_maxmsg > msg_max ||
574 attr->mq_msgsize > msgsize_max)
577 /* check for overflow */
578 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
580 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
581 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
582 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
588 * Invoked when creating a new queue via sys_mq_open
590 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
591 int oflag, mode_t mode, struct mq_attr *attr)
593 const struct cred *cred = current_cred();
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, cred);
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 const struct cred *cred = current_cred();
635 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
636 MAY_READ | MAY_WRITE };
638 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
641 return ERR_PTR(-EINVAL);
644 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
647 return ERR_PTR(-EACCES);
650 return dentry_open(dentry, mqueue_mnt, oflag, cred);
653 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
654 struct mq_attr __user *, u_attr)
656 struct dentry *dentry;
662 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
665 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
667 if (IS_ERR(name = getname(u_name)))
668 return PTR_ERR(name);
670 fd = get_unused_fd_flags(O_CLOEXEC);
674 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
675 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
676 if (IS_ERR(dentry)) {
677 error = PTR_ERR(dentry);
682 if (oflag & O_CREAT) {
683 if (dentry->d_inode) { /* entry already exists */
684 audit_inode(name, dentry);
688 filp = do_open(dentry, oflag);
690 filp = do_create(mqueue_mnt->mnt_root, dentry,
692 u_attr ? &attr : NULL);
696 if (!dentry->d_inode)
698 audit_inode(name, dentry);
699 filp = do_open(dentry, oflag);
703 error = PTR_ERR(filp);
707 fd_install(fd, filp);
718 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
724 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
728 struct dentry *dentry;
729 struct inode *inode = NULL;
731 name = getname(u_name);
733 return PTR_ERR(name);
735 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
737 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
738 if (IS_ERR(dentry)) {
739 err = PTR_ERR(dentry);
743 if (!dentry->d_inode) {
748 inode = dentry->d_inode;
750 atomic_inc(&inode->i_count);
751 err = mnt_want_write(mqueue_mnt);
754 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
755 mnt_drop_write(mqueue_mnt);
760 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
768 /* Pipelined send and receive functions.
770 * If a receiver finds no waiting message, then it registers itself in the
771 * list of waiting receivers. A sender checks that list before adding the new
772 * message into the message array. If there is a waiting receiver, then it
773 * bypasses the message array and directly hands the message over to the
775 * The receiver accepts the message and returns without grabbing the queue
776 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
777 * are necessary. The same algorithm is used for sysv semaphores, see
778 * ipc/sem.c for more details.
780 * The same algorithm is used for senders.
783 /* pipelined_send() - send a message directly to the task waiting in
784 * sys_mq_timedreceive() (without inserting message into a queue).
786 static inline void pipelined_send(struct mqueue_inode_info *info,
787 struct msg_msg *message,
788 struct ext_wait_queue *receiver)
790 receiver->msg = message;
791 list_del(&receiver->list);
792 receiver->state = STATE_PENDING;
793 wake_up_process(receiver->task);
795 receiver->state = STATE_READY;
798 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
799 * gets its message and put to the queue (we have one free place for sure). */
800 static inline void pipelined_receive(struct mqueue_inode_info *info)
802 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
806 wake_up_interruptible(&info->wait_q);
809 msg_insert(sender->msg, info);
810 list_del(&sender->list);
811 sender->state = STATE_PENDING;
812 wake_up_process(sender->task);
814 sender->state = STATE_READY;
817 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
818 size_t, msg_len, unsigned int, msg_prio,
819 const struct timespec __user *, u_abs_timeout)
823 struct ext_wait_queue wait;
824 struct ext_wait_queue *receiver;
825 struct msg_msg *msg_ptr;
826 struct mqueue_inode_info *info;
827 struct timespec ts, *p = NULL;
832 if (copy_from_user(&ts, u_abs_timeout,
833 sizeof(struct timespec)))
838 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
841 audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
842 timeout = prepare_timeout(p);
849 inode = filp->f_path.dentry->d_inode;
850 if (unlikely(filp->f_op != &mqueue_file_operations))
852 info = MQUEUE_I(inode);
853 audit_inode(NULL, filp->f_path.dentry);
855 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
858 if (unlikely(msg_len > info->attr.mq_msgsize)) {
863 /* First try to allocate memory, before doing anything with
864 * existing queues. */
865 msg_ptr = load_msg(u_msg_ptr, msg_len);
866 if (IS_ERR(msg_ptr)) {
867 ret = PTR_ERR(msg_ptr);
870 msg_ptr->m_ts = msg_len;
871 msg_ptr->m_type = msg_prio;
873 spin_lock(&info->lock);
875 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
876 if (filp->f_flags & O_NONBLOCK) {
877 spin_unlock(&info->lock);
879 } else if (unlikely(timeout < 0)) {
880 spin_unlock(&info->lock);
884 wait.msg = (void *) msg_ptr;
885 wait.state = STATE_NONE;
886 ret = wq_sleep(info, SEND, timeout, &wait);
891 receiver = wq_get_first_waiter(info, RECV);
893 pipelined_send(info, msg_ptr, receiver);
895 /* adds message to the queue */
896 msg_insert(msg_ptr, info);
899 inode->i_atime = inode->i_mtime = inode->i_ctime =
901 spin_unlock(&info->lock);
910 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
911 size_t, msg_len, unsigned int __user *, u_msg_prio,
912 const struct timespec __user *, u_abs_timeout)
916 struct msg_msg *msg_ptr;
919 struct mqueue_inode_info *info;
920 struct ext_wait_queue wait;
921 struct timespec ts, *p = NULL;
924 if (copy_from_user(&ts, u_abs_timeout,
925 sizeof(struct timespec)))
930 audit_mq_sendrecv(mqdes, msg_len, 0, p);
931 timeout = prepare_timeout(p);
938 inode = filp->f_path.dentry->d_inode;
939 if (unlikely(filp->f_op != &mqueue_file_operations))
941 info = MQUEUE_I(inode);
942 audit_inode(NULL, filp->f_path.dentry);
944 if (unlikely(!(filp->f_mode & FMODE_READ)))
947 /* checks if buffer is big enough */
948 if (unlikely(msg_len < info->attr.mq_msgsize)) {
953 spin_lock(&info->lock);
954 if (info->attr.mq_curmsgs == 0) {
955 if (filp->f_flags & O_NONBLOCK) {
956 spin_unlock(&info->lock);
959 } else if (unlikely(timeout < 0)) {
960 spin_unlock(&info->lock);
965 wait.state = STATE_NONE;
966 ret = wq_sleep(info, RECV, timeout, &wait);
970 msg_ptr = msg_get(info);
972 inode->i_atime = inode->i_mtime = inode->i_ctime =
975 /* There is now free space in queue. */
976 pipelined_receive(info);
977 spin_unlock(&info->lock);
983 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
984 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
996 * Notes: the case when user wants us to deregister (with NULL as pointer)
997 * and he isn't currently owner of notification, will be silently discarded.
998 * It isn't explicitly defined in the POSIX.
1000 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1001 const struct sigevent __user *, u_notification)
1006 struct inode *inode;
1007 struct sigevent notification;
1008 struct mqueue_inode_info *info;
1011 if (u_notification) {
1012 if (copy_from_user(¬ification, u_notification,
1013 sizeof(struct sigevent)))
1017 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1021 if (u_notification != NULL) {
1022 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1023 notification.sigev_notify != SIGEV_SIGNAL &&
1024 notification.sigev_notify != SIGEV_THREAD))
1026 if (notification.sigev_notify == SIGEV_SIGNAL &&
1027 !valid_signal(notification.sigev_signo)) {
1030 if (notification.sigev_notify == SIGEV_THREAD) {
1033 /* create the notify skb */
1034 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1039 if (copy_from_user(nc->data,
1040 notification.sigev_value.sival_ptr,
1041 NOTIFY_COOKIE_LEN)) {
1045 /* TODO: add a header? */
1046 skb_put(nc, NOTIFY_COOKIE_LEN);
1047 /* and attach it to the socket */
1049 filp = fget(notification.sigev_signo);
1053 sock = netlink_getsockbyfilp(filp);
1056 ret = PTR_ERR(sock);
1061 timeo = MAX_SCHEDULE_TIMEOUT;
1062 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1078 inode = filp->f_path.dentry->d_inode;
1079 if (unlikely(filp->f_op != &mqueue_file_operations))
1081 info = MQUEUE_I(inode);
1084 spin_lock(&info->lock);
1085 if (u_notification == NULL) {
1086 if (info->notify_owner == task_tgid(current)) {
1087 remove_notification(info);
1088 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1090 } else if (info->notify_owner != NULL) {
1093 switch (notification.sigev_notify) {
1095 info->notify.sigev_notify = SIGEV_NONE;
1098 info->notify_sock = sock;
1099 info->notify_cookie = nc;
1102 info->notify.sigev_notify = SIGEV_THREAD;
1105 info->notify.sigev_signo = notification.sigev_signo;
1106 info->notify.sigev_value = notification.sigev_value;
1107 info->notify.sigev_notify = SIGEV_SIGNAL;
1111 info->notify_owner = get_pid(task_tgid(current));
1112 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1114 spin_unlock(&info->lock);
1119 netlink_detachskb(sock, nc);
1126 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1127 const struct mq_attr __user *, u_mqstat,
1128 struct mq_attr __user *, u_omqstat)
1131 struct mq_attr mqstat, omqstat;
1133 struct inode *inode;
1134 struct mqueue_inode_info *info;
1136 if (u_mqstat != NULL) {
1137 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1139 if (mqstat.mq_flags & (~O_NONBLOCK))
1148 inode = filp->f_path.dentry->d_inode;
1149 if (unlikely(filp->f_op != &mqueue_file_operations))
1151 info = MQUEUE_I(inode);
1153 spin_lock(&info->lock);
1155 omqstat = info->attr;
1156 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1158 audit_mq_getsetattr(mqdes, &mqstat);
1159 if (mqstat.mq_flags & O_NONBLOCK)
1160 filp->f_flags |= O_NONBLOCK;
1162 filp->f_flags &= ~O_NONBLOCK;
1164 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1167 spin_unlock(&info->lock);
1170 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1171 sizeof(struct mq_attr)))
1180 static const struct inode_operations mqueue_dir_inode_operations = {
1181 .lookup = simple_lookup,
1182 .create = mqueue_create,
1183 .unlink = mqueue_unlink,
1186 static const struct file_operations mqueue_file_operations = {
1187 .flush = mqueue_flush_file,
1188 .poll = mqueue_poll_file,
1189 .read = mqueue_read_file,
1192 static struct super_operations mqueue_super_ops = {
1193 .alloc_inode = mqueue_alloc_inode,
1194 .destroy_inode = mqueue_destroy_inode,
1195 .statfs = simple_statfs,
1196 .delete_inode = mqueue_delete_inode,
1197 .drop_inode = generic_delete_inode,
1200 static struct file_system_type mqueue_fs_type = {
1202 .get_sb = mqueue_get_sb,
1203 .kill_sb = kill_litter_super,
1206 static int msg_max_limit_min = MIN_MSGMAX;
1207 static int msg_max_limit_max = MAX_MSGMAX;
1209 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1210 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1212 static ctl_table mq_sysctls[] = {
1214 .procname = "queues_max",
1215 .data = &queues_max,
1216 .maxlen = sizeof(int),
1218 .proc_handler = &proc_dointvec,
1221 .procname = "msg_max",
1223 .maxlen = sizeof(int),
1225 .proc_handler = &proc_dointvec_minmax,
1226 .extra1 = &msg_max_limit_min,
1227 .extra2 = &msg_max_limit_max,
1230 .procname = "msgsize_max",
1231 .data = &msgsize_max,
1232 .maxlen = sizeof(int),
1234 .proc_handler = &proc_dointvec_minmax,
1235 .extra1 = &msg_maxsize_limit_min,
1236 .extra2 = &msg_maxsize_limit_max,
1241 static ctl_table mq_sysctl_dir[] = {
1243 .procname = "mqueue",
1245 .child = mq_sysctls,
1250 static ctl_table mq_sysctl_root[] = {
1255 .child = mq_sysctl_dir,
1260 static int __init init_mqueue_fs(void)
1264 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1265 sizeof(struct mqueue_inode_info), 0,
1266 SLAB_HWCACHE_ALIGN, init_once);
1267 if (mqueue_inode_cachep == NULL)
1270 /* ignore failues - they are not fatal */
1271 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1273 error = register_filesystem(&mqueue_fs_type);
1277 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1278 error = PTR_ERR(mqueue_mnt);
1279 goto out_filesystem;
1282 /* internal initialization - not common for vfs */
1284 spin_lock_init(&mq_lock);
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);