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/vs_context.h>
33 #include <linux/vs_limit.h>
38 #define MQUEUE_MAGIC 0x19800202
39 #define DIRENT_SIZE 20
40 #define FILENT_SIZE 80
46 #define STATE_PENDING 1
51 #define CTL_QUEUESMAX 2
53 #define CTL_MSGSIZEMAX 4
56 #define DFLT_QUEUESMAX 256 /* max number of message queues */
57 #define DFLT_MSGMAX 10 /* max number of messages in each queue */
58 #define HARD_MSGMAX (131072/sizeof(void*))
59 #define DFLT_MSGSIZEMAX 8192 /* max message size */
62 struct ext_wait_queue { /* queue of sleeping tasks */
63 struct task_struct *task;
64 struct list_head list;
65 struct msg_msg *msg; /* ptr of loaded message */
66 int state; /* one of STATE_* values */
69 struct mqueue_inode_info {
71 struct inode vfs_inode;
72 wait_queue_head_t wait_q;
74 struct msg_msg **messages;
77 struct sigevent notify;
78 struct pid* notify_owner;
79 struct user_struct *user; /* user who created, for accounting */
80 struct sock *notify_sock;
81 struct sk_buff *notify_cookie;
83 /* for tasks waiting for free space and messages, respectively */
84 struct ext_wait_queue e_wait_q[2];
86 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
89 static struct inode_operations mqueue_dir_inode_operations;
90 static struct file_operations mqueue_file_operations;
91 static struct super_operations mqueue_super_ops;
92 static void remove_notification(struct mqueue_inode_info *info);
94 static spinlock_t mq_lock;
95 static struct kmem_cache *mqueue_inode_cachep;
96 static struct vfsmount *mqueue_mnt;
98 static unsigned int queues_count;
99 static unsigned int queues_max = DFLT_QUEUESMAX;
100 static unsigned int msg_max = DFLT_MSGMAX;
101 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
103 static struct ctl_table_header * mq_sysctl_table;
105 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
107 return container_of(inode, struct mqueue_inode_info, vfs_inode);
110 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
111 struct mq_attr *attr)
115 inode = new_inode(sb);
117 inode->i_mode = mode;
118 inode->i_uid = current->fsuid;
119 inode->i_gid = current->fsgid;
121 inode->i_mtime = inode->i_ctime = inode->i_atime =
125 struct mqueue_inode_info *info;
126 struct task_struct *p = current;
127 struct user_struct *u = p->user;
128 unsigned long mq_bytes, mq_msg_tblsz;
130 inode->i_fop = &mqueue_file_operations;
131 inode->i_size = FILENT_SIZE;
132 /* mqueue specific info */
133 info = MQUEUE_I(inode);
134 spin_lock_init(&info->lock);
135 init_waitqueue_head(&info->wait_q);
136 INIT_LIST_HEAD(&info->e_wait_q[0].list);
137 INIT_LIST_HEAD(&info->e_wait_q[1].list);
138 info->messages = NULL;
139 info->notify_owner = NULL;
141 info->user = NULL; /* set when all is ok */
142 memset(&info->attr, 0, sizeof(info->attr));
143 info->attr.mq_maxmsg = DFLT_MSGMAX;
144 info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
146 info->attr.mq_maxmsg = attr->mq_maxmsg;
147 info->attr.mq_msgsize = attr->mq_msgsize;
149 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
150 mq_bytes = (mq_msg_tblsz +
151 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
154 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
155 u->mq_bytes + mq_bytes >
156 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
157 !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
158 spin_unlock(&mq_lock);
161 u->mq_bytes += mq_bytes;
162 vx_ipcmsg_add(p->vx_info, u, mq_bytes);
163 spin_unlock(&mq_lock);
165 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
166 if (!info->messages) {
168 u->mq_bytes -= mq_bytes;
169 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
170 spin_unlock(&mq_lock);
174 info->user = get_uid(u);
175 } else if (S_ISDIR(mode)) {
177 /* Some things misbehave if size == 0 on a directory */
178 inode->i_size = 2 * DIRENT_SIZE;
179 inode->i_op = &mqueue_dir_inode_operations;
180 inode->i_fop = &simple_dir_operations;
185 make_bad_inode(inode);
190 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
194 sb->s_blocksize = PAGE_CACHE_SIZE;
195 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
196 sb->s_magic = MQUEUE_MAGIC;
197 sb->s_op = &mqueue_super_ops;
199 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
203 sb->s_root = d_alloc_root(inode);
212 static int mqueue_get_sb(struct file_system_type *fs_type,
213 int flags, const char *dev_name,
214 void *data, struct vfsmount *mnt)
216 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
219 static void init_once(void *foo, struct kmem_cache * cachep, unsigned long flags)
221 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
223 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
224 SLAB_CTOR_CONSTRUCTOR)
225 inode_init_once(&p->vfs_inode);
228 static struct inode *mqueue_alloc_inode(struct super_block *sb)
230 struct mqueue_inode_info *ei;
232 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
235 return &ei->vfs_inode;
238 static void mqueue_destroy_inode(struct inode *inode)
240 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
243 static void mqueue_delete_inode(struct inode *inode)
245 struct mqueue_inode_info *info;
246 struct user_struct *user;
247 unsigned long mq_bytes;
250 if (S_ISDIR(inode->i_mode)) {
254 info = MQUEUE_I(inode);
255 spin_lock(&info->lock);
256 for (i = 0; i < info->attr.mq_curmsgs; i++)
257 free_msg(info->messages[i]);
258 kfree(info->messages);
259 spin_unlock(&info->lock);
263 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
264 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
267 struct vx_info *vxi = lookup_vx_info(user->xid);
270 user->mq_bytes -= mq_bytes;
271 vx_ipcmsg_sub(vxi, user, mq_bytes);
273 spin_unlock(&mq_lock);
279 static int mqueue_create(struct inode *dir, struct dentry *dentry,
280 int mode, struct nameidata *nd)
283 struct mq_attr *attr = dentry->d_fsdata;
287 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
292 spin_unlock(&mq_lock);
294 inode = mqueue_get_inode(dir->i_sb, mode, attr);
302 dir->i_size += DIRENT_SIZE;
303 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
305 d_instantiate(dentry, inode);
309 spin_unlock(&mq_lock);
313 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
315 struct inode *inode = dentry->d_inode;
317 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
318 dir->i_size -= DIRENT_SIZE;
325 * This is routine for system read from queue file.
326 * To avoid mess with doing here some sort of mq_receive we allow
327 * to read only queue size & notification info (the only values
328 * that are interesting from user point of view and aren't accessible
329 * through std routines)
331 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
332 size_t count, loff_t * off)
334 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
335 char buffer[FILENT_SIZE];
342 spin_lock(&info->lock);
343 snprintf(buffer, sizeof(buffer),
344 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
346 info->notify_owner ? info->notify.sigev_notify : 0,
347 (info->notify_owner &&
348 info->notify.sigev_notify == SIGEV_SIGNAL) ?
349 info->notify.sigev_signo : 0,
350 pid_nr(info->notify_owner));
351 spin_unlock(&info->lock);
352 buffer[sizeof(buffer)-1] = '\0';
353 slen = strlen(buffer)+1;
359 if (o + count > slen)
362 if (copy_to_user(u_data, buffer + o, count))
366 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
370 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
372 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
374 spin_lock(&info->lock);
375 if (task_tgid(current) == info->notify_owner)
376 remove_notification(info);
378 spin_unlock(&info->lock);
382 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
384 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
387 poll_wait(filp, &info->wait_q, poll_tab);
389 spin_lock(&info->lock);
390 if (info->attr.mq_curmsgs)
391 retval = POLLIN | POLLRDNORM;
393 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
394 retval |= POLLOUT | POLLWRNORM;
395 spin_unlock(&info->lock);
400 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
401 static void wq_add(struct mqueue_inode_info *info, int sr,
402 struct ext_wait_queue *ewp)
404 struct ext_wait_queue *walk;
408 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
409 if (walk->task->static_prio <= current->static_prio) {
410 list_add_tail(&ewp->list, &walk->list);
414 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
418 * Puts current task to sleep. Caller must hold queue lock. After return
422 static int wq_sleep(struct mqueue_inode_info *info, int sr,
423 long timeout, struct ext_wait_queue *ewp)
428 wq_add(info, sr, ewp);
431 set_current_state(TASK_INTERRUPTIBLE);
433 spin_unlock(&info->lock);
434 time = schedule_timeout(timeout);
436 while (ewp->state == STATE_PENDING)
439 if (ewp->state == STATE_READY) {
443 spin_lock(&info->lock);
444 if (ewp->state == STATE_READY) {
448 if (signal_pending(current)) {
449 retval = -ERESTARTSYS;
457 list_del(&ewp->list);
459 spin_unlock(&info->lock);
465 * Returns waiting task that should be serviced first or NULL if none exists
467 static struct ext_wait_queue *wq_get_first_waiter(
468 struct mqueue_inode_info *info, int sr)
470 struct list_head *ptr;
472 ptr = info->e_wait_q[sr].list.prev;
473 if (ptr == &info->e_wait_q[sr].list)
475 return list_entry(ptr, struct ext_wait_queue, list);
478 /* Auxiliary functions to manipulate messages' list */
479 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
483 k = info->attr.mq_curmsgs - 1;
484 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
485 info->messages[k + 1] = info->messages[k];
488 info->attr.mq_curmsgs++;
489 info->qsize += ptr->m_ts;
490 info->messages[k + 1] = ptr;
493 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
495 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
496 return info->messages[info->attr.mq_curmsgs];
499 static inline void set_cookie(struct sk_buff *skb, char code)
501 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
505 * The next function is only to split too long sys_mq_timedsend
507 static void __do_notify(struct mqueue_inode_info *info)
510 * invoked when there is registered process and there isn't process
511 * waiting synchronously for message AND state of queue changed from
512 * empty to not empty. Here we are sure that no one is waiting
514 if (info->notify_owner &&
515 info->attr.mq_curmsgs == 1) {
516 struct siginfo sig_i;
517 switch (info->notify.sigev_notify) {
523 sig_i.si_signo = info->notify.sigev_signo;
525 sig_i.si_code = SI_MESGQ;
526 sig_i.si_value = info->notify.sigev_value;
527 sig_i.si_pid = current->tgid;
528 sig_i.si_uid = current->uid;
530 kill_pid_info(info->notify.sigev_signo,
531 &sig_i, info->notify_owner);
534 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
535 netlink_sendskb(info->notify_sock,
536 info->notify_cookie, 0);
539 /* after notification unregisters process */
540 put_pid(info->notify_owner);
541 info->notify_owner = NULL;
543 wake_up(&info->wait_q);
546 static long prepare_timeout(const struct timespec __user *u_arg)
548 struct timespec ts, nowts;
552 if (unlikely(copy_from_user(&ts, u_arg,
553 sizeof(struct timespec))))
556 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
557 || ts.tv_nsec >= NSEC_PER_SEC))
559 nowts = CURRENT_TIME;
560 /* first subtract as jiffies can't be too big */
561 ts.tv_sec -= nowts.tv_sec;
562 if (ts.tv_nsec < nowts.tv_nsec) {
563 ts.tv_nsec += NSEC_PER_SEC;
566 ts.tv_nsec -= nowts.tv_nsec;
570 timeout = timespec_to_jiffies(&ts) + 1;
572 return MAX_SCHEDULE_TIMEOUT;
577 static void remove_notification(struct mqueue_inode_info *info)
579 if (info->notify_owner != NULL &&
580 info->notify.sigev_notify == SIGEV_THREAD) {
581 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
582 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
584 put_pid(info->notify_owner);
585 info->notify_owner = NULL;
588 static int mq_attr_ok(struct mq_attr *attr)
590 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
592 if (capable(CAP_SYS_RESOURCE)) {
593 if (attr->mq_maxmsg > HARD_MSGMAX)
596 if (attr->mq_maxmsg > msg_max ||
597 attr->mq_msgsize > msgsize_max)
600 /* check for overflow */
601 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
603 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
604 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
605 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
611 * Invoked when creating a new queue via sys_mq_open
613 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
614 int oflag, mode_t mode, struct mq_attr __user *u_attr)
621 if (copy_from_user(&attr, u_attr, sizeof(attr)))
624 if (!mq_attr_ok(&attr))
626 /* store for use during create */
627 dentry->d_fsdata = &attr;
630 mode &= ~current->fs->umask;
631 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
632 dentry->d_fsdata = NULL;
636 return dentry_open(dentry, mqueue_mnt, oflag);
644 /* Opens existing queue */
645 static struct file *do_open(struct dentry *dentry, int oflag)
647 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
648 MAY_READ | MAY_WRITE };
650 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
653 return ERR_PTR(-EINVAL);
656 if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
659 return ERR_PTR(-EACCES);
662 return dentry_open(dentry, mqueue_mnt, oflag);
665 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
666 struct mq_attr __user *u_attr)
668 struct dentry *dentry;
673 error = audit_mq_open(oflag, mode, u_attr);
677 if (IS_ERR(name = getname(u_name)))
678 return PTR_ERR(name);
680 fd = get_unused_fd();
684 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
685 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
686 if (IS_ERR(dentry)) {
687 error = PTR_ERR(dentry);
692 if (oflag & O_CREAT) {
693 if (dentry->d_inode) { /* entry already exists */
697 filp = do_open(dentry, oflag);
699 filp = do_create(mqueue_mnt->mnt_root, dentry,
700 oflag, mode, u_attr);
704 if (!dentry->d_inode)
706 filp = do_open(dentry, oflag);
710 error = PTR_ERR(filp);
714 set_close_on_exec(fd, 1);
715 fd_install(fd, filp);
726 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
732 asmlinkage long sys_mq_unlink(const char __user *u_name)
736 struct dentry *dentry;
737 struct inode *inode = NULL;
739 name = getname(u_name);
741 return PTR_ERR(name);
743 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
744 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
745 if (IS_ERR(dentry)) {
746 err = PTR_ERR(dentry);
750 if (!dentry->d_inode) {
755 inode = dentry->d_inode;
757 atomic_inc(&inode->i_count);
759 err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
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);
853 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
856 if (unlikely(msg_len > info->attr.mq_msgsize)) {
861 /* First try to allocate memory, before doing anything with
862 * existing queues. */
863 msg_ptr = load_msg(u_msg_ptr, msg_len);
864 if (IS_ERR(msg_ptr)) {
865 ret = PTR_ERR(msg_ptr);
868 msg_ptr->m_ts = msg_len;
869 msg_ptr->m_type = msg_prio;
871 spin_lock(&info->lock);
873 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
874 if (filp->f_flags & O_NONBLOCK) {
875 spin_unlock(&info->lock);
877 } else if (unlikely(timeout < 0)) {
878 spin_unlock(&info->lock);
882 wait.msg = (void *) msg_ptr;
883 wait.state = STATE_NONE;
884 ret = wq_sleep(info, SEND, timeout, &wait);
889 receiver = wq_get_first_waiter(info, RECV);
891 pipelined_send(info, msg_ptr, receiver);
893 /* adds message to the queue */
894 msg_insert(msg_ptr, info);
897 inode->i_atime = inode->i_mtime = inode->i_ctime =
899 spin_unlock(&info->lock);
908 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
909 size_t msg_len, unsigned int __user *u_msg_prio,
910 const struct timespec __user *u_abs_timeout)
914 struct msg_msg *msg_ptr;
917 struct mqueue_inode_info *info;
918 struct ext_wait_queue wait;
920 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
924 timeout = prepare_timeout(u_abs_timeout);
931 inode = filp->f_path.dentry->d_inode;
932 if (unlikely(filp->f_op != &mqueue_file_operations))
934 info = MQUEUE_I(inode);
936 if (unlikely(!(filp->f_mode & FMODE_READ)))
939 /* checks if buffer is big enough */
940 if (unlikely(msg_len < info->attr.mq_msgsize)) {
945 spin_lock(&info->lock);
946 if (info->attr.mq_curmsgs == 0) {
947 if (filp->f_flags & O_NONBLOCK) {
948 spin_unlock(&info->lock);
951 } else if (unlikely(timeout < 0)) {
952 spin_unlock(&info->lock);
957 wait.state = STATE_NONE;
958 ret = wq_sleep(info, RECV, timeout, &wait);
962 msg_ptr = msg_get(info);
964 inode->i_atime = inode->i_mtime = inode->i_ctime =
967 /* There is now free space in queue. */
968 pipelined_receive(info);
969 spin_unlock(&info->lock);
975 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
976 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
988 * Notes: the case when user wants us to deregister (with NULL as pointer)
989 * and he isn't currently owner of notification, will be silently discarded.
990 * It isn't explicitly defined in the POSIX.
992 asmlinkage long sys_mq_notify(mqd_t mqdes,
993 const struct sigevent __user *u_notification)
999 struct sigevent notification;
1000 struct mqueue_inode_info *info;
1003 ret = audit_mq_notify(mqdes, u_notification);
1009 if (u_notification != NULL) {
1010 if (copy_from_user(¬ification, u_notification,
1011 sizeof(struct sigevent)))
1014 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1015 notification.sigev_notify != SIGEV_SIGNAL &&
1016 notification.sigev_notify != SIGEV_THREAD))
1018 if (notification.sigev_notify == SIGEV_SIGNAL &&
1019 !valid_signal(notification.sigev_signo)) {
1022 if (notification.sigev_notify == SIGEV_THREAD) {
1023 /* create the notify skb */
1024 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1029 if (copy_from_user(nc->data,
1030 notification.sigev_value.sival_ptr,
1031 NOTIFY_COOKIE_LEN)) {
1035 /* TODO: add a header? */
1036 skb_put(nc, NOTIFY_COOKIE_LEN);
1037 /* and attach it to the socket */
1039 filp = fget(notification.sigev_signo);
1043 sock = netlink_getsockbyfilp(filp);
1046 ret = PTR_ERR(sock);
1051 ret = netlink_attachskb(sock, nc, 0,
1052 MAX_SCHEDULE_TIMEOUT, NULL);
1068 inode = filp->f_path.dentry->d_inode;
1069 if (unlikely(filp->f_op != &mqueue_file_operations))
1071 info = MQUEUE_I(inode);
1074 spin_lock(&info->lock);
1075 if (u_notification == NULL) {
1076 if (info->notify_owner == task_tgid(current)) {
1077 remove_notification(info);
1078 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1080 } else if (info->notify_owner != NULL) {
1083 switch (notification.sigev_notify) {
1085 info->notify.sigev_notify = SIGEV_NONE;
1088 info->notify_sock = sock;
1089 info->notify_cookie = nc;
1092 info->notify.sigev_notify = SIGEV_THREAD;
1095 info->notify.sigev_signo = notification.sigev_signo;
1096 info->notify.sigev_value = notification.sigev_value;
1097 info->notify.sigev_notify = SIGEV_SIGNAL;
1101 info->notify_owner = get_pid(task_tgid(current));
1102 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1104 spin_unlock(&info->lock);
1109 netlink_detachskb(sock, nc);
1116 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1117 const struct mq_attr __user *u_mqstat,
1118 struct mq_attr __user *u_omqstat)
1121 struct mq_attr mqstat, omqstat;
1123 struct inode *inode;
1124 struct mqueue_inode_info *info;
1126 if (u_mqstat != NULL) {
1127 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1129 if (mqstat.mq_flags & (~O_NONBLOCK))
1138 inode = filp->f_path.dentry->d_inode;
1139 if (unlikely(filp->f_op != &mqueue_file_operations))
1141 info = MQUEUE_I(inode);
1143 spin_lock(&info->lock);
1145 omqstat = info->attr;
1146 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1148 ret = audit_mq_getsetattr(mqdes, &mqstat);
1151 if (mqstat.mq_flags & O_NONBLOCK)
1152 filp->f_flags |= O_NONBLOCK;
1154 filp->f_flags &= ~O_NONBLOCK;
1156 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1159 spin_unlock(&info->lock);
1162 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1163 sizeof(struct mq_attr)))
1172 static struct inode_operations mqueue_dir_inode_operations = {
1173 .lookup = simple_lookup,
1174 .create = mqueue_create,
1175 .unlink = mqueue_unlink,
1178 static struct file_operations mqueue_file_operations = {
1179 .flush = mqueue_flush_file,
1180 .poll = mqueue_poll_file,
1181 .read = mqueue_read_file,
1184 static struct super_operations mqueue_super_ops = {
1185 .alloc_inode = mqueue_alloc_inode,
1186 .destroy_inode = mqueue_destroy_inode,
1187 .statfs = simple_statfs,
1188 .delete_inode = mqueue_delete_inode,
1189 .drop_inode = generic_delete_inode,
1192 static struct file_system_type mqueue_fs_type = {
1194 .get_sb = mqueue_get_sb,
1195 .kill_sb = kill_litter_super,
1198 static int msg_max_limit_min = DFLT_MSGMAX;
1199 static int msg_max_limit_max = HARD_MSGMAX;
1201 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1202 static int msg_maxsize_limit_max = INT_MAX;
1204 static ctl_table mq_sysctls[] = {
1206 .ctl_name = CTL_QUEUESMAX,
1207 .procname = "queues_max",
1208 .data = &queues_max,
1209 .maxlen = sizeof(int),
1211 .proc_handler = &proc_dointvec,
1214 .ctl_name = CTL_MSGMAX,
1215 .procname = "msg_max",
1217 .maxlen = sizeof(int),
1219 .proc_handler = &proc_dointvec_minmax,
1220 .extra1 = &msg_max_limit_min,
1221 .extra2 = &msg_max_limit_max,
1224 .ctl_name = CTL_MSGSIZEMAX,
1225 .procname = "msgsize_max",
1226 .data = &msgsize_max,
1227 .maxlen = sizeof(int),
1229 .proc_handler = &proc_dointvec_minmax,
1230 .extra1 = &msg_maxsize_limit_min,
1231 .extra2 = &msg_maxsize_limit_max,
1236 static ctl_table mq_sysctl_dir[] = {
1238 .ctl_name = FS_MQUEUE,
1239 .procname = "mqueue",
1241 .child = mq_sysctls,
1246 static ctl_table mq_sysctl_root[] = {
1251 .child = mq_sysctl_dir,
1256 static int __init init_mqueue_fs(void)
1260 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1261 sizeof(struct mqueue_inode_info), 0,
1262 SLAB_HWCACHE_ALIGN, init_once, NULL);
1263 if (mqueue_inode_cachep == NULL)
1266 /* ignore failues - they are not fatal */
1267 mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1269 error = register_filesystem(&mqueue_fs_type);
1273 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1274 error = PTR_ERR(mqueue_mnt);
1275 goto out_filesystem;
1278 /* internal initialization - not common for vfs */
1280 spin_lock_init(&mq_lock);
1285 unregister_filesystem(&mqueue_fs_type);
1287 if (mq_sysctl_table)
1288 unregister_sysctl_table(mq_sysctl_table);
1289 kmem_cache_destroy(mqueue_inode_cachep);
1293 __initcall(init_mqueue_fs);