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();
124 inode->i_mtime = inode->i_ctime = inode->i_atime =
128 struct mqueue_inode_info *info;
129 struct task_struct *p = current;
130 unsigned long mq_bytes, mq_msg_tblsz;
132 inode->i_fop = &mqueue_file_operations;
133 inode->i_size = FILENT_SIZE;
134 /* mqueue specific info */
135 info = MQUEUE_I(inode);
136 spin_lock_init(&info->lock);
137 init_waitqueue_head(&info->wait_q);
138 INIT_LIST_HEAD(&info->e_wait_q[0].list);
139 INIT_LIST_HEAD(&info->e_wait_q[1].list);
140 info->messages = NULL;
141 info->notify_owner = NULL;
143 info->user = NULL; /* set when all is ok */
144 memset(&info->attr, 0, sizeof(info->attr));
145 info->attr.mq_maxmsg = msg_max;
146 info->attr.mq_msgsize = msgsize_max;
148 info->attr.mq_maxmsg = attr->mq_maxmsg;
149 info->attr.mq_msgsize = attr->mq_msgsize;
151 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
152 mq_bytes = (mq_msg_tblsz +
153 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
156 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
157 u->mq_bytes + mq_bytes >
158 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
159 spin_unlock(&mq_lock);
162 u->mq_bytes += 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 spin_unlock(&mq_lock);
173 info->user = get_uid(u);
174 } else if (S_ISDIR(mode)) {
176 /* Some things misbehave if size == 0 on a directory */
177 inode->i_size = 2 * DIRENT_SIZE;
178 inode->i_op = &mqueue_dir_inode_operations;
179 inode->i_fop = &simple_dir_operations;
184 make_bad_inode(inode);
189 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
193 sb->s_blocksize = PAGE_CACHE_SIZE;
194 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
195 sb->s_magic = MQUEUE_MAGIC;
196 sb->s_op = &mqueue_super_ops;
198 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
202 sb->s_root = d_alloc_root(inode);
211 static int mqueue_get_sb(struct file_system_type *fs_type,
212 int flags, const char *dev_name,
213 void *data, struct vfsmount *mnt)
215 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
218 static void init_once(void *foo)
220 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
222 inode_init_once(&p->vfs_inode);
225 static struct inode *mqueue_alloc_inode(struct super_block *sb)
227 struct mqueue_inode_info *ei;
229 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
232 return &ei->vfs_inode;
235 static void mqueue_destroy_inode(struct inode *inode)
237 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
240 static void mqueue_delete_inode(struct inode *inode)
242 struct mqueue_inode_info *info;
243 struct user_struct *user;
244 unsigned long mq_bytes;
247 if (S_ISDIR(inode->i_mode)) {
251 info = MQUEUE_I(inode);
252 spin_lock(&info->lock);
253 for (i = 0; i < info->attr.mq_curmsgs; i++)
254 free_msg(info->messages[i]);
255 kfree(info->messages);
256 spin_unlock(&info->lock);
260 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
261 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
265 user->mq_bytes -= mq_bytes;
267 spin_unlock(&mq_lock);
272 static int mqueue_create(struct inode *dir, struct dentry *dentry,
273 int mode, struct nameidata *nd)
276 struct mq_attr *attr = dentry->d_fsdata;
280 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
285 spin_unlock(&mq_lock);
287 inode = mqueue_get_inode(dir->i_sb, mode, attr);
295 dir->i_size += DIRENT_SIZE;
296 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
298 d_instantiate(dentry, inode);
302 spin_unlock(&mq_lock);
306 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
308 struct inode *inode = dentry->d_inode;
310 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
311 dir->i_size -= DIRENT_SIZE;
318 * This is routine for system read from queue file.
319 * To avoid mess with doing here some sort of mq_receive we allow
320 * to read only queue size & notification info (the only values
321 * that are interesting from user point of view and aren't accessible
322 * through std routines)
324 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
325 size_t count, loff_t *off)
327 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
328 char buffer[FILENT_SIZE];
331 spin_lock(&info->lock);
332 snprintf(buffer, sizeof(buffer),
333 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
335 info->notify_owner ? info->notify.sigev_notify : 0,
336 (info->notify_owner &&
337 info->notify.sigev_notify == SIGEV_SIGNAL) ?
338 info->notify.sigev_signo : 0,
339 pid_vnr(info->notify_owner));
340 spin_unlock(&info->lock);
341 buffer[sizeof(buffer)-1] = '\0';
343 ret = simple_read_from_buffer(u_data, count, off, buffer,
348 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
352 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
354 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
356 spin_lock(&info->lock);
357 if (task_tgid(current) == info->notify_owner)
358 remove_notification(info);
360 spin_unlock(&info->lock);
364 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
366 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
369 poll_wait(filp, &info->wait_q, poll_tab);
371 spin_lock(&info->lock);
372 if (info->attr.mq_curmsgs)
373 retval = POLLIN | POLLRDNORM;
375 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
376 retval |= POLLOUT | POLLWRNORM;
377 spin_unlock(&info->lock);
382 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
383 static void wq_add(struct mqueue_inode_info *info, int sr,
384 struct ext_wait_queue *ewp)
386 struct ext_wait_queue *walk;
390 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
391 if (walk->task->static_prio <= current->static_prio) {
392 list_add_tail(&ewp->list, &walk->list);
396 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
400 * Puts current task to sleep. Caller must hold queue lock. After return
404 static int wq_sleep(struct mqueue_inode_info *info, int sr,
405 long timeout, struct ext_wait_queue *ewp)
410 wq_add(info, sr, ewp);
413 set_current_state(TASK_INTERRUPTIBLE);
415 spin_unlock(&info->lock);
416 time = schedule_timeout(timeout);
418 while (ewp->state == STATE_PENDING)
421 if (ewp->state == STATE_READY) {
425 spin_lock(&info->lock);
426 if (ewp->state == STATE_READY) {
430 if (signal_pending(current)) {
431 retval = -ERESTARTSYS;
439 list_del(&ewp->list);
441 spin_unlock(&info->lock);
447 * Returns waiting task that should be serviced first or NULL if none exists
449 static struct ext_wait_queue *wq_get_first_waiter(
450 struct mqueue_inode_info *info, int sr)
452 struct list_head *ptr;
454 ptr = info->e_wait_q[sr].list.prev;
455 if (ptr == &info->e_wait_q[sr].list)
457 return list_entry(ptr, struct ext_wait_queue, list);
460 /* Auxiliary functions to manipulate messages' list */
461 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
465 k = info->attr.mq_curmsgs - 1;
466 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
467 info->messages[k + 1] = info->messages[k];
470 info->attr.mq_curmsgs++;
471 info->qsize += ptr->m_ts;
472 info->messages[k + 1] = ptr;
475 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
477 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
478 return info->messages[info->attr.mq_curmsgs];
481 static inline void set_cookie(struct sk_buff *skb, char code)
483 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
487 * The next function is only to split too long sys_mq_timedsend
489 static void __do_notify(struct mqueue_inode_info *info)
492 * invoked when there is registered process and there isn't process
493 * waiting synchronously for message AND state of queue changed from
494 * empty to not empty. Here we are sure that no one is waiting
496 if (info->notify_owner &&
497 info->attr.mq_curmsgs == 1) {
498 struct siginfo sig_i;
499 switch (info->notify.sigev_notify) {
505 sig_i.si_signo = info->notify.sigev_signo;
507 sig_i.si_code = SI_MESGQ;
508 sig_i.si_value = info->notify.sigev_value;
509 sig_i.si_pid = task_tgid_vnr(current);
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 __user *u_attr)
593 const struct cred *cred = current_cred();
600 if (copy_from_user(&attr, u_attr, sizeof(attr)))
603 if (!mq_attr_ok(&attr))
605 /* store for use during create */
606 dentry->d_fsdata = &attr;
609 mode &= ~current->fs->umask;
610 ret = mnt_want_write(mqueue_mnt);
613 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
614 dentry->d_fsdata = NULL;
618 result = dentry_open(dentry, mqueue_mnt, oflag, cred);
620 * dentry_open() took a persistent mnt_want_write(),
621 * so we can now drop this one.
623 mnt_drop_write(mqueue_mnt);
627 mnt_drop_write(mqueue_mnt);
634 /* Opens existing queue */
635 static struct file *do_open(struct dentry *dentry, int oflag)
637 const struct cred *cred = current_cred();
639 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
640 MAY_READ | MAY_WRITE };
642 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
645 return ERR_PTR(-EINVAL);
648 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
651 return ERR_PTR(-EACCES);
654 return dentry_open(dentry, mqueue_mnt, oflag, cred);
657 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
658 struct mq_attr __user *u_attr)
660 struct dentry *dentry;
665 error = audit_mq_open(oflag, mode, u_attr);
669 if (IS_ERR(name = getname(u_name)))
670 return PTR_ERR(name);
672 fd = get_unused_fd_flags(O_CLOEXEC);
676 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
677 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
678 if (IS_ERR(dentry)) {
679 error = PTR_ERR(dentry);
684 if (oflag & O_CREAT) {
685 if (dentry->d_inode) { /* entry already exists */
686 audit_inode(name, dentry);
690 filp = do_open(dentry, oflag);
692 filp = do_create(mqueue_mnt->mnt_root, dentry,
693 oflag, mode, u_attr);
697 if (!dentry->d_inode)
699 audit_inode(name, dentry);
700 filp = do_open(dentry, oflag);
704 error = PTR_ERR(filp);
708 fd_install(fd, filp);
719 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
725 asmlinkage long sys_mq_unlink(const char __user *u_name)
729 struct dentry *dentry;
730 struct inode *inode = NULL;
732 name = getname(u_name);
734 return PTR_ERR(name);
736 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
738 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
739 if (IS_ERR(dentry)) {
740 err = PTR_ERR(dentry);
744 if (!dentry->d_inode) {
749 inode = dentry->d_inode;
751 atomic_inc(&inode->i_count);
752 err = mnt_want_write(mqueue_mnt);
755 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
756 mnt_drop_write(mqueue_mnt);
761 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
769 /* Pipelined send and receive functions.
771 * If a receiver finds no waiting message, then it registers itself in the
772 * list of waiting receivers. A sender checks that list before adding the new
773 * message into the message array. If there is a waiting receiver, then it
774 * bypasses the message array and directly hands the message over to the
776 * The receiver accepts the message and returns without grabbing the queue
777 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
778 * are necessary. The same algorithm is used for sysv semaphores, see
779 * ipc/sem.c for more details.
781 * The same algorithm is used for senders.
784 /* pipelined_send() - send a message directly to the task waiting in
785 * sys_mq_timedreceive() (without inserting message into a queue).
787 static inline void pipelined_send(struct mqueue_inode_info *info,
788 struct msg_msg *message,
789 struct ext_wait_queue *receiver)
791 receiver->msg = message;
792 list_del(&receiver->list);
793 receiver->state = STATE_PENDING;
794 wake_up_process(receiver->task);
796 receiver->state = STATE_READY;
799 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
800 * gets its message and put to the queue (we have one free place for sure). */
801 static inline void pipelined_receive(struct mqueue_inode_info *info)
803 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
807 wake_up_interruptible(&info->wait_q);
810 msg_insert(sender->msg, info);
811 list_del(&sender->list);
812 sender->state = STATE_PENDING;
813 wake_up_process(sender->task);
815 sender->state = STATE_READY;
818 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
819 size_t msg_len, unsigned int msg_prio,
820 const struct timespec __user *u_abs_timeout)
824 struct ext_wait_queue wait;
825 struct ext_wait_queue *receiver;
826 struct msg_msg *msg_ptr;
827 struct mqueue_inode_info *info;
828 struct timespec ts, *p = NULL;
833 if (copy_from_user(&ts, u_abs_timeout,
834 sizeof(struct timespec)))
839 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
842 audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
843 timeout = prepare_timeout(p);
850 inode = filp->f_path.dentry->d_inode;
851 if (unlikely(filp->f_op != &mqueue_file_operations))
853 info = MQUEUE_I(inode);
854 audit_inode(NULL, filp->f_path.dentry);
856 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
859 if (unlikely(msg_len > info->attr.mq_msgsize)) {
864 /* First try to allocate memory, before doing anything with
865 * existing queues. */
866 msg_ptr = load_msg(u_msg_ptr, msg_len);
867 if (IS_ERR(msg_ptr)) {
868 ret = PTR_ERR(msg_ptr);
871 msg_ptr->m_ts = msg_len;
872 msg_ptr->m_type = msg_prio;
874 spin_lock(&info->lock);
876 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
877 if (filp->f_flags & O_NONBLOCK) {
878 spin_unlock(&info->lock);
880 } else if (unlikely(timeout < 0)) {
881 spin_unlock(&info->lock);
885 wait.msg = (void *) msg_ptr;
886 wait.state = STATE_NONE;
887 ret = wq_sleep(info, SEND, timeout, &wait);
892 receiver = wq_get_first_waiter(info, RECV);
894 pipelined_send(info, msg_ptr, receiver);
896 /* adds message to the queue */
897 msg_insert(msg_ptr, info);
900 inode->i_atime = inode->i_mtime = inode->i_ctime =
902 spin_unlock(&info->lock);
911 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
912 size_t msg_len, unsigned int __user *u_msg_prio,
913 const struct timespec __user *u_abs_timeout)
917 struct msg_msg *msg_ptr;
920 struct mqueue_inode_info *info;
921 struct ext_wait_queue wait;
922 struct timespec ts, *p = NULL;
925 if (copy_from_user(&ts, u_abs_timeout,
926 sizeof(struct timespec)))
931 audit_mq_sendrecv(mqdes, msg_len, 0, p);
932 timeout = prepare_timeout(p);
939 inode = filp->f_path.dentry->d_inode;
940 if (unlikely(filp->f_op != &mqueue_file_operations))
942 info = MQUEUE_I(inode);
943 audit_inode(NULL, filp->f_path.dentry);
945 if (unlikely(!(filp->f_mode & FMODE_READ)))
948 /* checks if buffer is big enough */
949 if (unlikely(msg_len < info->attr.mq_msgsize)) {
954 spin_lock(&info->lock);
955 if (info->attr.mq_curmsgs == 0) {
956 if (filp->f_flags & O_NONBLOCK) {
957 spin_unlock(&info->lock);
960 } else if (unlikely(timeout < 0)) {
961 spin_unlock(&info->lock);
966 wait.state = STATE_NONE;
967 ret = wq_sleep(info, RECV, timeout, &wait);
971 msg_ptr = msg_get(info);
973 inode->i_atime = inode->i_mtime = inode->i_ctime =
976 /* There is now free space in queue. */
977 pipelined_receive(info);
978 spin_unlock(&info->lock);
984 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
985 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
997 * Notes: the case when user wants us to deregister (with NULL as pointer)
998 * and he isn't currently owner of notification, will be silently discarded.
999 * It isn't explicitly defined in the POSIX.
1001 asmlinkage long sys_mq_notify(mqd_t mqdes,
1002 const struct sigevent __user *u_notification)
1007 struct inode *inode;
1008 struct sigevent notification;
1009 struct mqueue_inode_info *info;
1012 if (u_notification) {
1013 if (copy_from_user(¬ification, u_notification,
1014 sizeof(struct sigevent)))
1018 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1022 if (u_notification != NULL) {
1023 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1024 notification.sigev_notify != SIGEV_SIGNAL &&
1025 notification.sigev_notify != SIGEV_THREAD))
1027 if (notification.sigev_notify == SIGEV_SIGNAL &&
1028 !valid_signal(notification.sigev_signo)) {
1031 if (notification.sigev_notify == SIGEV_THREAD) {
1034 /* create the notify skb */
1035 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1040 if (copy_from_user(nc->data,
1041 notification.sigev_value.sival_ptr,
1042 NOTIFY_COOKIE_LEN)) {
1046 /* TODO: add a header? */
1047 skb_put(nc, NOTIFY_COOKIE_LEN);
1048 /* and attach it to the socket */
1050 filp = fget(notification.sigev_signo);
1054 sock = netlink_getsockbyfilp(filp);
1057 ret = PTR_ERR(sock);
1062 timeo = MAX_SCHEDULE_TIMEOUT;
1063 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1079 inode = filp->f_path.dentry->d_inode;
1080 if (unlikely(filp->f_op != &mqueue_file_operations))
1082 info = MQUEUE_I(inode);
1085 spin_lock(&info->lock);
1086 if (u_notification == NULL) {
1087 if (info->notify_owner == task_tgid(current)) {
1088 remove_notification(info);
1089 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1091 } else if (info->notify_owner != NULL) {
1094 switch (notification.sigev_notify) {
1096 info->notify.sigev_notify = SIGEV_NONE;
1099 info->notify_sock = sock;
1100 info->notify_cookie = nc;
1103 info->notify.sigev_notify = SIGEV_THREAD;
1106 info->notify.sigev_signo = notification.sigev_signo;
1107 info->notify.sigev_value = notification.sigev_value;
1108 info->notify.sigev_notify = SIGEV_SIGNAL;
1112 info->notify_owner = get_pid(task_tgid(current));
1113 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1115 spin_unlock(&info->lock);
1120 netlink_detachskb(sock, nc);
1127 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1128 const struct mq_attr __user *u_mqstat,
1129 struct mq_attr __user *u_omqstat)
1132 struct mq_attr mqstat, omqstat;
1134 struct inode *inode;
1135 struct mqueue_inode_info *info;
1137 if (u_mqstat != NULL) {
1138 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1140 if (mqstat.mq_flags & (~O_NONBLOCK))
1149 inode = filp->f_path.dentry->d_inode;
1150 if (unlikely(filp->f_op != &mqueue_file_operations))
1152 info = MQUEUE_I(inode);
1154 spin_lock(&info->lock);
1156 omqstat = info->attr;
1157 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1159 audit_mq_getsetattr(mqdes, &mqstat);
1160 if (mqstat.mq_flags & O_NONBLOCK)
1161 filp->f_flags |= O_NONBLOCK;
1163 filp->f_flags &= ~O_NONBLOCK;
1165 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1168 spin_unlock(&info->lock);
1171 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1172 sizeof(struct mq_attr)))
1181 static const struct inode_operations mqueue_dir_inode_operations = {
1182 .lookup = simple_lookup,
1183 .create = mqueue_create,
1184 .unlink = mqueue_unlink,
1187 static const struct file_operations mqueue_file_operations = {
1188 .flush = mqueue_flush_file,
1189 .poll = mqueue_poll_file,
1190 .read = mqueue_read_file,
1193 static struct super_operations mqueue_super_ops = {
1194 .alloc_inode = mqueue_alloc_inode,
1195 .destroy_inode = mqueue_destroy_inode,
1196 .statfs = simple_statfs,
1197 .delete_inode = mqueue_delete_inode,
1198 .drop_inode = generic_delete_inode,
1201 static struct file_system_type mqueue_fs_type = {
1203 .get_sb = mqueue_get_sb,
1204 .kill_sb = kill_litter_super,
1207 static int msg_max_limit_min = MIN_MSGMAX;
1208 static int msg_max_limit_max = MAX_MSGMAX;
1210 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1211 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1213 static ctl_table mq_sysctls[] = {
1215 .procname = "queues_max",
1216 .data = &queues_max,
1217 .maxlen = sizeof(int),
1219 .proc_handler = &proc_dointvec,
1222 .procname = "msg_max",
1224 .maxlen = sizeof(int),
1226 .proc_handler = &proc_dointvec_minmax,
1227 .extra1 = &msg_max_limit_min,
1228 .extra2 = &msg_max_limit_max,
1231 .procname = "msgsize_max",
1232 .data = &msgsize_max,
1233 .maxlen = sizeof(int),
1235 .proc_handler = &proc_dointvec_minmax,
1236 .extra1 = &msg_maxsize_limit_min,
1237 .extra2 = &msg_maxsize_limit_max,
1242 static ctl_table mq_sysctl_dir[] = {
1244 .procname = "mqueue",
1246 .child = mq_sysctls,
1251 static ctl_table mq_sysctl_root[] = {
1256 .child = mq_sysctl_dir,
1261 static int __init init_mqueue_fs(void)
1265 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1266 sizeof(struct mqueue_inode_info), 0,
1267 SLAB_HWCACHE_ALIGN, init_once);
1268 if (mqueue_inode_cachep == NULL)
1271 /* ignore failues - they are not fatal */
1272 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1274 error = register_filesystem(&mqueue_fs_type);
1278 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1279 error = PTR_ERR(mqueue_mnt);
1280 goto out_filesystem;
1283 /* internal initialization - not common for vfs */
1285 spin_lock_init(&mq_lock);
1290 unregister_filesystem(&mqueue_fs_type);
1292 if (mq_sysctl_table)
1293 unregister_sysctl_table(mq_sysctl_table);
1294 kmem_cache_destroy(mqueue_inode_cachep);
1298 __initcall(init_mqueue_fs);