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/vmalloc.h>
28 #include <linux/netlink.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/signal.h>
32 #include <linux/mutex.h>
33 #include <linux/nsproxy.h>
34 #include <linux/pid.h>
35 #include <linux/ipc_namespace.h>
36 #include <linux/user_namespace.h>
37 #include <linux/slab.h>
42 #define MQUEUE_MAGIC 0x19800202
43 #define DIRENT_SIZE 20
44 #define FILENT_SIZE 80
50 #define STATE_PENDING 1
53 struct posix_msg_tree_node {
54 struct rb_node rb_node;
55 struct list_head msg_list;
59 struct ext_wait_queue { /* queue of sleeping tasks */
60 struct task_struct *task;
61 struct list_head list;
62 struct msg_msg *msg; /* ptr of loaded message */
63 int state; /* one of STATE_* values */
66 struct mqueue_inode_info {
68 struct inode vfs_inode;
69 wait_queue_head_t wait_q;
71 struct rb_root msg_tree;
72 struct posix_msg_tree_node *node_cache;
75 struct sigevent notify;
76 struct pid* notify_owner;
77 struct user_namespace *notify_user_ns;
78 struct user_struct *user; /* user who created, for accounting */
79 struct sock *notify_sock;
80 struct sk_buff *notify_cookie;
82 /* for tasks waiting for free space and messages, respectively */
83 struct ext_wait_queue e_wait_q[2];
85 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88 static const struct inode_operations mqueue_dir_inode_operations;
89 static const struct file_operations mqueue_file_operations;
90 static const struct super_operations mqueue_super_ops;
91 static void remove_notification(struct mqueue_inode_info *info);
93 static struct kmem_cache *mqueue_inode_cachep;
95 static struct ctl_table_header * mq_sysctl_table;
97 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
99 return container_of(inode, struct mqueue_inode_info, vfs_inode);
103 * This routine should be called with the mq_lock held.
105 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
107 return get_ipc_ns(inode->i_sb->s_fs_info);
110 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
112 struct ipc_namespace *ns;
115 ns = __get_ns_from_inode(inode);
116 spin_unlock(&mq_lock);
120 /* Auxiliary functions to manipulate messages' list */
121 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
123 struct rb_node **p, *parent = NULL;
124 struct posix_msg_tree_node *leaf;
126 p = &info->msg_tree.rb_node;
129 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
131 if (likely(leaf->priority == msg->m_type))
133 else if (msg->m_type < leaf->priority)
138 if (info->node_cache) {
139 leaf = info->node_cache;
140 info->node_cache = NULL;
142 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145 rb_init_node(&leaf->rb_node);
146 INIT_LIST_HEAD(&leaf->msg_list);
147 info->qsize += sizeof(*leaf);
149 leaf->priority = msg->m_type;
150 rb_link_node(&leaf->rb_node, parent, p);
151 rb_insert_color(&leaf->rb_node, &info->msg_tree);
153 info->attr.mq_curmsgs++;
154 info->qsize += msg->m_ts;
155 list_add_tail(&msg->m_list, &leaf->msg_list);
159 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
161 struct rb_node **p, *parent = NULL;
162 struct posix_msg_tree_node *leaf;
166 p = &info->msg_tree.rb_node;
170 * During insert, low priorities go to the left and high to the
171 * right. On receive, we want the highest priorities first, so
172 * walk all the way to the right.
177 if (info->attr.mq_curmsgs) {
178 pr_warn_once("Inconsistency in POSIX message queue, "
179 "no tree element, but supposedly messages "
181 info->attr.mq_curmsgs = 0;
185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186 if (unlikely(list_empty(&leaf->msg_list))) {
187 pr_warn_once("Inconsistency in POSIX message queue, "
188 "empty leaf node but we haven't implemented "
189 "lazy leaf delete!\n");
190 rb_erase(&leaf->rb_node, &info->msg_tree);
191 if (info->node_cache) {
192 info->qsize -= sizeof(*leaf);
195 info->node_cache = leaf;
199 msg = list_first_entry(&leaf->msg_list,
200 struct msg_msg, m_list);
201 list_del(&msg->m_list);
202 if (list_empty(&leaf->msg_list)) {
203 rb_erase(&leaf->rb_node, &info->msg_tree);
204 if (info->node_cache) {
205 info->qsize -= sizeof(*leaf);
208 info->node_cache = leaf;
212 info->attr.mq_curmsgs--;
213 info->qsize -= msg->m_ts;
217 static struct inode *mqueue_get_inode(struct super_block *sb,
218 struct ipc_namespace *ipc_ns, umode_t mode,
219 struct mq_attr *attr)
221 struct user_struct *u = current_user();
225 inode = new_inode(sb);
229 inode->i_ino = get_next_ino();
230 inode->i_mode = mode;
231 inode->i_uid = current_fsuid();
232 inode->i_gid = current_fsgid();
233 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
236 struct mqueue_inode_info *info;
237 unsigned long mq_bytes, mq_treesize;
239 inode->i_fop = &mqueue_file_operations;
240 inode->i_size = FILENT_SIZE;
241 /* mqueue specific info */
242 info = MQUEUE_I(inode);
243 spin_lock_init(&info->lock);
244 init_waitqueue_head(&info->wait_q);
245 INIT_LIST_HEAD(&info->e_wait_q[0].list);
246 INIT_LIST_HEAD(&info->e_wait_q[1].list);
247 info->notify_owner = NULL;
248 info->notify_user_ns = NULL;
250 info->user = NULL; /* set when all is ok */
251 info->msg_tree = RB_ROOT;
252 info->node_cache = NULL;
253 memset(&info->attr, 0, sizeof(info->attr));
254 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
255 ipc_ns->mq_msg_default);
256 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
257 ipc_ns->mq_msgsize_default);
259 info->attr.mq_maxmsg = attr->mq_maxmsg;
260 info->attr.mq_msgsize = attr->mq_msgsize;
263 * We used to allocate a static array of pointers and account
264 * the size of that array as well as one msg_msg struct per
265 * possible message into the queue size. That's no longer
266 * accurate as the queue is now an rbtree and will grow and
267 * shrink depending on usage patterns. We can, however, still
268 * account one msg_msg struct per message, but the nodes are
269 * allocated depending on priority usage, and most programs
270 * only use one, or a handful, of priorities. However, since
271 * this is pinned memory, we need to assume worst case, so
272 * that means the min(mq_maxmsg, max_priorities) * struct
273 * posix_msg_tree_node.
275 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
276 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
277 sizeof(struct posix_msg_tree_node);
279 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
280 info->attr.mq_msgsize);
283 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
284 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
285 spin_unlock(&mq_lock);
286 /* mqueue_evict_inode() releases info->messages */
290 u->mq_bytes += mq_bytes;
291 spin_unlock(&mq_lock);
294 info->user = get_uid(u);
295 } else if (S_ISDIR(mode)) {
297 /* Some things misbehave if size == 0 on a directory */
298 inode->i_size = 2 * DIRENT_SIZE;
299 inode->i_op = &mqueue_dir_inode_operations;
300 inode->i_fop = &simple_dir_operations;
310 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
313 struct ipc_namespace *ns = data;
315 sb->s_blocksize = PAGE_CACHE_SIZE;
316 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
317 sb->s_magic = MQUEUE_MAGIC;
318 sb->s_op = &mqueue_super_ops;
320 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
322 return PTR_ERR(inode);
324 sb->s_root = d_make_root(inode);
330 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
331 int flags, const char *dev_name,
334 if (!(flags & MS_KERNMOUNT))
335 data = current->nsproxy->ipc_ns;
336 return mount_ns(fs_type, flags, data, mqueue_fill_super);
339 static void init_once(void *foo)
341 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
343 inode_init_once(&p->vfs_inode);
346 static struct inode *mqueue_alloc_inode(struct super_block *sb)
348 struct mqueue_inode_info *ei;
350 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
353 return &ei->vfs_inode;
356 static void mqueue_i_callback(struct rcu_head *head)
358 struct inode *inode = container_of(head, struct inode, i_rcu);
359 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
362 static void mqueue_destroy_inode(struct inode *inode)
364 call_rcu(&inode->i_rcu, mqueue_i_callback);
367 static void mqueue_evict_inode(struct inode *inode)
369 struct mqueue_inode_info *info;
370 struct user_struct *user;
371 unsigned long mq_bytes, mq_treesize;
372 struct ipc_namespace *ipc_ns;
377 if (S_ISDIR(inode->i_mode))
380 ipc_ns = get_ns_from_inode(inode);
381 info = MQUEUE_I(inode);
382 spin_lock(&info->lock);
383 while ((msg = msg_get(info)) != NULL)
385 kfree(info->node_cache);
386 spin_unlock(&info->lock);
388 /* Total amount of bytes accounted for the mqueue */
389 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
390 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
391 sizeof(struct posix_msg_tree_node);
393 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
394 info->attr.mq_msgsize);
399 user->mq_bytes -= mq_bytes;
401 * get_ns_from_inode() ensures that the
402 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
403 * to which we now hold a reference, or it is NULL.
404 * We can't put it here under mq_lock, though.
407 ipc_ns->mq_queues_count--;
408 spin_unlock(&mq_lock);
415 static int mqueue_create(struct inode *dir, struct dentry *dentry,
416 umode_t mode, bool excl)
419 struct mq_attr *attr = dentry->d_fsdata;
421 struct ipc_namespace *ipc_ns;
424 ipc_ns = __get_ns_from_inode(dir);
429 if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
430 (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
431 !capable(CAP_SYS_RESOURCE))) {
435 ipc_ns->mq_queues_count++;
436 spin_unlock(&mq_lock);
438 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
440 error = PTR_ERR(inode);
442 ipc_ns->mq_queues_count--;
447 dir->i_size += DIRENT_SIZE;
448 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
450 d_instantiate(dentry, inode);
454 spin_unlock(&mq_lock);
460 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
462 struct inode *inode = dentry->d_inode;
464 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
465 dir->i_size -= DIRENT_SIZE;
472 * This is routine for system read from queue file.
473 * To avoid mess with doing here some sort of mq_receive we allow
474 * to read only queue size & notification info (the only values
475 * that are interesting from user point of view and aren't accessible
476 * through std routines)
478 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
479 size_t count, loff_t *off)
481 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
482 char buffer[FILENT_SIZE];
485 spin_lock(&info->lock);
486 snprintf(buffer, sizeof(buffer),
487 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
489 info->notify_owner ? info->notify.sigev_notify : 0,
490 (info->notify_owner &&
491 info->notify.sigev_notify == SIGEV_SIGNAL) ?
492 info->notify.sigev_signo : 0,
493 pid_vnr(info->notify_owner));
494 spin_unlock(&info->lock);
495 buffer[sizeof(buffer)-1] = '\0';
497 ret = simple_read_from_buffer(u_data, count, off, buffer,
502 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
506 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
508 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
510 spin_lock(&info->lock);
511 if (task_tgid(current) == info->notify_owner)
512 remove_notification(info);
514 spin_unlock(&info->lock);
518 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
520 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
523 poll_wait(filp, &info->wait_q, poll_tab);
525 spin_lock(&info->lock);
526 if (info->attr.mq_curmsgs)
527 retval = POLLIN | POLLRDNORM;
529 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
530 retval |= POLLOUT | POLLWRNORM;
531 spin_unlock(&info->lock);
536 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
537 static void wq_add(struct mqueue_inode_info *info, int sr,
538 struct ext_wait_queue *ewp)
540 struct ext_wait_queue *walk;
544 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
545 if (walk->task->static_prio <= current->static_prio) {
546 list_add_tail(&ewp->list, &walk->list);
550 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
554 * Puts current task to sleep. Caller must hold queue lock. After return
558 static int wq_sleep(struct mqueue_inode_info *info, int sr,
559 ktime_t *timeout, struct ext_wait_queue *ewp)
564 wq_add(info, sr, ewp);
567 set_current_state(TASK_INTERRUPTIBLE);
569 spin_unlock(&info->lock);
570 time = schedule_hrtimeout_range_clock(timeout, 0,
571 HRTIMER_MODE_ABS, CLOCK_REALTIME);
573 while (ewp->state == STATE_PENDING)
576 if (ewp->state == STATE_READY) {
580 spin_lock(&info->lock);
581 if (ewp->state == STATE_READY) {
585 if (signal_pending(current)) {
586 retval = -ERESTARTSYS;
594 list_del(&ewp->list);
596 spin_unlock(&info->lock);
602 * Returns waiting task that should be serviced first or NULL if none exists
604 static struct ext_wait_queue *wq_get_first_waiter(
605 struct mqueue_inode_info *info, int sr)
607 struct list_head *ptr;
609 ptr = info->e_wait_q[sr].list.prev;
610 if (ptr == &info->e_wait_q[sr].list)
612 return list_entry(ptr, struct ext_wait_queue, list);
616 static inline void set_cookie(struct sk_buff *skb, char code)
618 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
622 * The next function is only to split too long sys_mq_timedsend
624 static void __do_notify(struct mqueue_inode_info *info)
627 * invoked when there is registered process and there isn't process
628 * waiting synchronously for message AND state of queue changed from
629 * empty to not empty. Here we are sure that no one is waiting
631 if (info->notify_owner &&
632 info->attr.mq_curmsgs == 1) {
633 struct siginfo sig_i;
634 switch (info->notify.sigev_notify) {
640 sig_i.si_signo = info->notify.sigev_signo;
642 sig_i.si_code = SI_MESGQ;
643 sig_i.si_value = info->notify.sigev_value;
644 /* map current pid/uid into info->owner's namespaces */
646 sig_i.si_pid = task_tgid_nr_ns(current,
647 ns_of_pid(info->notify_owner));
648 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
651 kill_pid_info(info->notify.sigev_signo,
652 &sig_i, info->notify_owner);
655 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
656 netlink_sendskb(info->notify_sock, info->notify_cookie);
659 /* after notification unregisters process */
660 put_pid(info->notify_owner);
661 put_user_ns(info->notify_user_ns);
662 info->notify_owner = NULL;
663 info->notify_user_ns = NULL;
665 wake_up(&info->wait_q);
668 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
669 ktime_t *expires, struct timespec *ts)
671 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
673 if (!timespec_valid(ts))
676 *expires = timespec_to_ktime(*ts);
680 static void remove_notification(struct mqueue_inode_info *info)
682 if (info->notify_owner != NULL &&
683 info->notify.sigev_notify == SIGEV_THREAD) {
684 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
685 netlink_sendskb(info->notify_sock, info->notify_cookie);
687 put_pid(info->notify_owner);
688 put_user_ns(info->notify_user_ns);
689 info->notify_owner = NULL;
690 info->notify_user_ns = NULL;
693 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
696 unsigned long total_size;
698 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
700 if (capable(CAP_SYS_RESOURCE)) {
701 if (attr->mq_maxmsg > HARD_MSGMAX ||
702 attr->mq_msgsize > HARD_MSGSIZEMAX)
705 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
706 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
709 /* check for overflow */
710 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
712 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
713 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
714 sizeof(struct posix_msg_tree_node);
715 total_size = attr->mq_maxmsg * attr->mq_msgsize;
716 if (total_size + mq_treesize < total_size)
722 * Invoked when creating a new queue via sys_mq_open
724 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
725 struct path *path, int oflag, umode_t mode,
726 struct mq_attr *attr)
728 const struct cred *cred = current_cred();
733 ret = mq_attr_ok(ipc_ns, attr);
736 /* store for use during create */
737 path->dentry->d_fsdata = attr;
739 struct mq_attr def_attr;
741 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
742 ipc_ns->mq_msg_default);
743 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
744 ipc_ns->mq_msgsize_default);
745 ret = mq_attr_ok(ipc_ns, &def_attr);
750 mode &= ~current_umask();
751 ret = mnt_want_write(path->mnt);
754 ret = vfs_create(dir, path->dentry, mode, true);
755 path->dentry->d_fsdata = NULL;
757 result = dentry_open(path, oflag, cred);
759 result = ERR_PTR(ret);
761 * dentry_open() took a persistent mnt_want_write(),
762 * so we can now drop this one.
764 mnt_drop_write(path->mnt);
768 /* Opens existing queue */
769 static struct file *do_open(struct path *path, int oflag)
771 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
772 MAY_READ | MAY_WRITE };
774 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
775 return ERR_PTR(-EINVAL);
776 acc = oflag2acc[oflag & O_ACCMODE];
777 if (inode_permission(path->dentry->d_inode, acc))
778 return ERR_PTR(-EACCES);
779 return dentry_open(path, oflag, current_cred());
782 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
783 struct mq_attr __user *, u_attr)
790 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
791 struct dentry *root = ipc_ns->mq_mnt->mnt_root;
793 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
796 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
798 if (IS_ERR(name = getname(u_name)))
799 return PTR_ERR(name);
801 fd = get_unused_fd_flags(O_CLOEXEC);
806 mutex_lock(&root->d_inode->i_mutex);
807 path.dentry = lookup_one_len(name, root, strlen(name));
808 if (IS_ERR(path.dentry)) {
809 error = PTR_ERR(path.dentry);
812 path.mnt = mntget(ipc_ns->mq_mnt);
814 if (oflag & O_CREAT) {
815 if (path.dentry->d_inode) { /* entry already exists */
816 audit_inode(name, path.dentry);
817 if (oflag & O_EXCL) {
821 filp = do_open(&path, oflag);
823 filp = do_create(ipc_ns, root->d_inode,
825 u_attr ? &attr : NULL);
828 if (!path.dentry->d_inode) {
832 audit_inode(name, path.dentry);
833 filp = do_open(&path, oflag);
837 fd_install(fd, filp);
839 error = PTR_ERR(filp);
847 mutex_unlock(&root->d_inode->i_mutex);
853 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
857 struct dentry *dentry;
858 struct inode *inode = NULL;
859 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
861 name = getname(u_name);
863 return PTR_ERR(name);
865 mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
867 dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
868 if (IS_ERR(dentry)) {
869 err = PTR_ERR(dentry);
873 if (!dentry->d_inode) {
878 inode = dentry->d_inode;
881 err = mnt_want_write(ipc_ns->mq_mnt);
884 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
885 mnt_drop_write(ipc_ns->mq_mnt);
890 mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
898 /* Pipelined send and receive functions.
900 * If a receiver finds no waiting message, then it registers itself in the
901 * list of waiting receivers. A sender checks that list before adding the new
902 * message into the message array. If there is a waiting receiver, then it
903 * bypasses the message array and directly hands the message over to the
905 * The receiver accepts the message and returns without grabbing the queue
906 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
907 * are necessary. The same algorithm is used for sysv semaphores, see
908 * ipc/sem.c for more details.
910 * The same algorithm is used for senders.
913 /* pipelined_send() - send a message directly to the task waiting in
914 * sys_mq_timedreceive() (without inserting message into a queue).
916 static inline void pipelined_send(struct mqueue_inode_info *info,
917 struct msg_msg *message,
918 struct ext_wait_queue *receiver)
920 receiver->msg = message;
921 list_del(&receiver->list);
922 receiver->state = STATE_PENDING;
923 wake_up_process(receiver->task);
925 receiver->state = STATE_READY;
928 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
929 * gets its message and put to the queue (we have one free place for sure). */
930 static inline void pipelined_receive(struct mqueue_inode_info *info)
932 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
936 wake_up_interruptible(&info->wait_q);
939 if (msg_insert(sender->msg, info))
941 list_del(&sender->list);
942 sender->state = STATE_PENDING;
943 wake_up_process(sender->task);
945 sender->state = STATE_READY;
948 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
949 size_t, msg_len, unsigned int, msg_prio,
950 const struct timespec __user *, u_abs_timeout)
954 struct ext_wait_queue wait;
955 struct ext_wait_queue *receiver;
956 struct msg_msg *msg_ptr;
957 struct mqueue_inode_info *info;
958 ktime_t expires, *timeout = NULL;
960 struct posix_msg_tree_node *new_leaf = NULL;
964 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
970 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
973 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
976 if (unlikely(!filp)) {
981 inode = filp->f_path.dentry->d_inode;
982 if (unlikely(filp->f_op != &mqueue_file_operations)) {
986 info = MQUEUE_I(inode);
987 audit_inode(NULL, filp->f_path.dentry);
989 if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
994 if (unlikely(msg_len > info->attr.mq_msgsize)) {
999 /* First try to allocate memory, before doing anything with
1000 * existing queues. */
1001 msg_ptr = load_msg(u_msg_ptr, msg_len);
1002 if (IS_ERR(msg_ptr)) {
1003 ret = PTR_ERR(msg_ptr);
1006 msg_ptr->m_ts = msg_len;
1007 msg_ptr->m_type = msg_prio;
1010 * msg_insert really wants us to have a valid, spare node struct so
1011 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1012 * fall back to that if necessary.
1014 if (!info->node_cache)
1015 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1017 spin_lock(&info->lock);
1019 if (!info->node_cache && new_leaf) {
1020 /* Save our speculative allocation into the cache */
1021 rb_init_node(&new_leaf->rb_node);
1022 INIT_LIST_HEAD(&new_leaf->msg_list);
1023 info->node_cache = new_leaf;
1024 info->qsize += sizeof(*new_leaf);
1030 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1031 if (filp->f_flags & O_NONBLOCK) {
1034 wait.task = current;
1035 wait.msg = (void *) msg_ptr;
1036 wait.state = STATE_NONE;
1037 ret = wq_sleep(info, SEND, timeout, &wait);
1039 * wq_sleep must be called with info->lock held, and
1040 * returns with the lock released
1045 receiver = wq_get_first_waiter(info, RECV);
1047 pipelined_send(info, msg_ptr, receiver);
1049 /* adds message to the queue */
1050 ret = msg_insert(msg_ptr, info);
1055 inode->i_atime = inode->i_mtime = inode->i_ctime =
1059 spin_unlock(&info->lock);
1069 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1070 size_t, msg_len, unsigned int __user *, u_msg_prio,
1071 const struct timespec __user *, u_abs_timeout)
1074 struct msg_msg *msg_ptr;
1076 struct inode *inode;
1077 struct mqueue_inode_info *info;
1078 struct ext_wait_queue wait;
1079 ktime_t expires, *timeout = NULL;
1081 struct posix_msg_tree_node *new_leaf = NULL;
1083 if (u_abs_timeout) {
1084 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1090 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1093 if (unlikely(!filp)) {
1098 inode = filp->f_path.dentry->d_inode;
1099 if (unlikely(filp->f_op != &mqueue_file_operations)) {
1103 info = MQUEUE_I(inode);
1104 audit_inode(NULL, filp->f_path.dentry);
1106 if (unlikely(!(filp->f_mode & FMODE_READ))) {
1111 /* checks if buffer is big enough */
1112 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1118 * msg_insert really wants us to have a valid, spare node struct so
1119 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1120 * fall back to that if necessary.
1122 if (!info->node_cache)
1123 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1125 spin_lock(&info->lock);
1127 if (!info->node_cache && new_leaf) {
1128 /* Save our speculative allocation into the cache */
1129 rb_init_node(&new_leaf->rb_node);
1130 INIT_LIST_HEAD(&new_leaf->msg_list);
1131 info->node_cache = new_leaf;
1132 info->qsize += sizeof(*new_leaf);
1137 if (info->attr.mq_curmsgs == 0) {
1138 if (filp->f_flags & O_NONBLOCK) {
1139 spin_unlock(&info->lock);
1142 wait.task = current;
1143 wait.state = STATE_NONE;
1144 ret = wq_sleep(info, RECV, timeout, &wait);
1148 msg_ptr = msg_get(info);
1150 inode->i_atime = inode->i_mtime = inode->i_ctime =
1153 /* There is now free space in queue. */
1154 pipelined_receive(info);
1155 spin_unlock(&info->lock);
1159 ret = msg_ptr->m_ts;
1161 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1162 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1174 * Notes: the case when user wants us to deregister (with NULL as pointer)
1175 * and he isn't currently owner of notification, will be silently discarded.
1176 * It isn't explicitly defined in the POSIX.
1178 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1179 const struct sigevent __user *, u_notification)
1184 struct inode *inode;
1185 struct sigevent notification;
1186 struct mqueue_inode_info *info;
1189 if (u_notification) {
1190 if (copy_from_user(¬ification, u_notification,
1191 sizeof(struct sigevent)))
1195 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1199 if (u_notification != NULL) {
1200 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1201 notification.sigev_notify != SIGEV_SIGNAL &&
1202 notification.sigev_notify != SIGEV_THREAD))
1204 if (notification.sigev_notify == SIGEV_SIGNAL &&
1205 !valid_signal(notification.sigev_signo)) {
1208 if (notification.sigev_notify == SIGEV_THREAD) {
1211 /* create the notify skb */
1212 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1217 if (copy_from_user(nc->data,
1218 notification.sigev_value.sival_ptr,
1219 NOTIFY_COOKIE_LEN)) {
1224 /* TODO: add a header? */
1225 skb_put(nc, NOTIFY_COOKIE_LEN);
1226 /* and attach it to the socket */
1228 filp = fget(notification.sigev_signo);
1233 sock = netlink_getsockbyfilp(filp);
1236 ret = PTR_ERR(sock);
1241 timeo = MAX_SCHEDULE_TIMEOUT;
1242 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1259 inode = filp->f_path.dentry->d_inode;
1260 if (unlikely(filp->f_op != &mqueue_file_operations)) {
1264 info = MQUEUE_I(inode);
1267 spin_lock(&info->lock);
1268 if (u_notification == NULL) {
1269 if (info->notify_owner == task_tgid(current)) {
1270 remove_notification(info);
1271 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1273 } else if (info->notify_owner != NULL) {
1276 switch (notification.sigev_notify) {
1278 info->notify.sigev_notify = SIGEV_NONE;
1281 info->notify_sock = sock;
1282 info->notify_cookie = nc;
1285 info->notify.sigev_notify = SIGEV_THREAD;
1288 info->notify.sigev_signo = notification.sigev_signo;
1289 info->notify.sigev_value = notification.sigev_value;
1290 info->notify.sigev_notify = SIGEV_SIGNAL;
1294 info->notify_owner = get_pid(task_tgid(current));
1295 info->notify_user_ns = get_user_ns(current_user_ns());
1296 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1298 spin_unlock(&info->lock);
1303 netlink_detachskb(sock, nc);
1310 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1311 const struct mq_attr __user *, u_mqstat,
1312 struct mq_attr __user *, u_omqstat)
1315 struct mq_attr mqstat, omqstat;
1317 struct inode *inode;
1318 struct mqueue_inode_info *info;
1320 if (u_mqstat != NULL) {
1321 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1323 if (mqstat.mq_flags & (~O_NONBLOCK))
1333 inode = filp->f_path.dentry->d_inode;
1334 if (unlikely(filp->f_op != &mqueue_file_operations)) {
1338 info = MQUEUE_I(inode);
1340 spin_lock(&info->lock);
1342 omqstat = info->attr;
1343 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1345 audit_mq_getsetattr(mqdes, &mqstat);
1346 spin_lock(&filp->f_lock);
1347 if (mqstat.mq_flags & O_NONBLOCK)
1348 filp->f_flags |= O_NONBLOCK;
1350 filp->f_flags &= ~O_NONBLOCK;
1351 spin_unlock(&filp->f_lock);
1353 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1356 spin_unlock(&info->lock);
1359 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1360 sizeof(struct mq_attr)))
1369 static const struct inode_operations mqueue_dir_inode_operations = {
1370 .lookup = simple_lookup,
1371 .create = mqueue_create,
1372 .unlink = mqueue_unlink,
1375 static const struct file_operations mqueue_file_operations = {
1376 .flush = mqueue_flush_file,
1377 .poll = mqueue_poll_file,
1378 .read = mqueue_read_file,
1379 .llseek = default_llseek,
1382 static const struct super_operations mqueue_super_ops = {
1383 .alloc_inode = mqueue_alloc_inode,
1384 .destroy_inode = mqueue_destroy_inode,
1385 .evict_inode = mqueue_evict_inode,
1386 .statfs = simple_statfs,
1389 static struct file_system_type mqueue_fs_type = {
1391 .mount = mqueue_mount,
1392 .kill_sb = kill_litter_super,
1395 int mq_init_ns(struct ipc_namespace *ns)
1397 ns->mq_queues_count = 0;
1398 ns->mq_queues_max = DFLT_QUEUESMAX;
1399 ns->mq_msg_max = DFLT_MSGMAX;
1400 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1401 ns->mq_msg_default = DFLT_MSG;
1402 ns->mq_msgsize_default = DFLT_MSGSIZE;
1404 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1405 if (IS_ERR(ns->mq_mnt)) {
1406 int err = PTR_ERR(ns->mq_mnt);
1413 void mq_clear_sbinfo(struct ipc_namespace *ns)
1415 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1418 void mq_put_mnt(struct ipc_namespace *ns)
1420 kern_unmount(ns->mq_mnt);
1423 static int __init init_mqueue_fs(void)
1427 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1428 sizeof(struct mqueue_inode_info), 0,
1429 SLAB_HWCACHE_ALIGN, init_once);
1430 if (mqueue_inode_cachep == NULL)
1433 /* ignore failures - they are not fatal */
1434 mq_sysctl_table = mq_register_sysctl_table();
1436 error = register_filesystem(&mqueue_fs_type);
1440 spin_lock_init(&mq_lock);
1442 error = mq_init_ns(&init_ipc_ns);
1444 goto out_filesystem;
1449 unregister_filesystem(&mqueue_fs_type);
1451 if (mq_sysctl_table)
1452 unregister_sysctl_table(mq_sysctl_table);
1453 kmem_cache_destroy(mqueue_inode_cachep);
1457 __initcall(init_mqueue_fs);