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 INIT_LIST_HEAD(&leaf->msg_list);
146 info->qsize += sizeof(*leaf);
148 leaf->priority = msg->m_type;
149 rb_link_node(&leaf->rb_node, parent, p);
150 rb_insert_color(&leaf->rb_node, &info->msg_tree);
152 info->attr.mq_curmsgs++;
153 info->qsize += msg->m_ts;
154 list_add_tail(&msg->m_list, &leaf->msg_list);
158 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160 struct rb_node **p, *parent = NULL;
161 struct posix_msg_tree_node *leaf;
165 p = &info->msg_tree.rb_node;
169 * During insert, low priorities go to the left and high to the
170 * right. On receive, we want the highest priorities first, so
171 * walk all the way to the right.
176 if (info->attr.mq_curmsgs) {
177 pr_warn_once("Inconsistency in POSIX message queue, "
178 "no tree element, but supposedly messages "
180 info->attr.mq_curmsgs = 0;
184 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
185 if (unlikely(list_empty(&leaf->msg_list))) {
186 pr_warn_once("Inconsistency in POSIX message queue, "
187 "empty leaf node but we haven't implemented "
188 "lazy leaf delete!\n");
189 rb_erase(&leaf->rb_node, &info->msg_tree);
190 if (info->node_cache) {
191 info->qsize -= sizeof(*leaf);
194 info->node_cache = leaf;
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
204 info->qsize -= sizeof(*leaf);
207 info->node_cache = leaf;
211 info->attr.mq_curmsgs--;
212 info->qsize -= msg->m_ts;
216 static struct inode *mqueue_get_inode(struct super_block *sb,
217 struct ipc_namespace *ipc_ns, umode_t mode,
218 struct mq_attr *attr)
220 struct user_struct *u = current_user();
224 inode = new_inode(sb);
228 inode->i_ino = get_next_ino();
229 inode->i_mode = mode;
230 inode->i_uid = current_fsuid();
231 inode->i_gid = current_fsgid();
232 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
235 struct mqueue_inode_info *info;
236 unsigned long mq_bytes, mq_treesize;
238 inode->i_fop = &mqueue_file_operations;
239 inode->i_size = FILENT_SIZE;
240 /* mqueue specific info */
241 info = MQUEUE_I(inode);
242 spin_lock_init(&info->lock);
243 init_waitqueue_head(&info->wait_q);
244 INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 info->notify_owner = NULL;
247 info->notify_user_ns = NULL;
249 info->user = NULL; /* set when all is ok */
250 info->msg_tree = RB_ROOT;
251 info->node_cache = NULL;
252 memset(&info->attr, 0, sizeof(info->attr));
253 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 ipc_ns->mq_msg_default);
255 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 ipc_ns->mq_msgsize_default);
258 info->attr.mq_maxmsg = attr->mq_maxmsg;
259 info->attr.mq_msgsize = attr->mq_msgsize;
262 * We used to allocate a static array of pointers and account
263 * the size of that array as well as one msg_msg struct per
264 * possible message into the queue size. That's no longer
265 * accurate as the queue is now an rbtree and will grow and
266 * shrink depending on usage patterns. We can, however, still
267 * account one msg_msg struct per message, but the nodes are
268 * allocated depending on priority usage, and most programs
269 * only use one, or a handful, of priorities. However, since
270 * this is pinned memory, we need to assume worst case, so
271 * that means the min(mq_maxmsg, max_priorities) * struct
272 * posix_msg_tree_node.
274 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 sizeof(struct posix_msg_tree_node);
278 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 info->attr.mq_msgsize);
282 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
283 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
284 spin_unlock(&mq_lock);
285 /* mqueue_evict_inode() releases info->messages */
289 u->mq_bytes += mq_bytes;
290 spin_unlock(&mq_lock);
293 info->user = get_uid(u);
294 } else if (S_ISDIR(mode)) {
296 /* Some things misbehave if size == 0 on a directory */
297 inode->i_size = 2 * DIRENT_SIZE;
298 inode->i_op = &mqueue_dir_inode_operations;
299 inode->i_fop = &simple_dir_operations;
309 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
312 struct ipc_namespace *ns = data;
314 sb->s_blocksize = PAGE_CACHE_SIZE;
315 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
321 return PTR_ERR(inode);
323 sb->s_root = d_make_root(inode);
329 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 int flags, const char *dev_name,
333 if (!(flags & MS_KERNMOUNT)) {
334 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
335 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
336 * over the ipc namespace.
338 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
339 return ERR_PTR(-EPERM);
343 return mount_ns(fs_type, flags, data, mqueue_fill_super);
346 static void init_once(void *foo)
348 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
350 inode_init_once(&p->vfs_inode);
353 static struct inode *mqueue_alloc_inode(struct super_block *sb)
355 struct mqueue_inode_info *ei;
357 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
360 return &ei->vfs_inode;
363 static void mqueue_i_callback(struct rcu_head *head)
365 struct inode *inode = container_of(head, struct inode, i_rcu);
366 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
369 static void mqueue_destroy_inode(struct inode *inode)
371 call_rcu(&inode->i_rcu, mqueue_i_callback);
374 static void mqueue_evict_inode(struct inode *inode)
376 struct mqueue_inode_info *info;
377 struct user_struct *user;
378 unsigned long mq_bytes, mq_treesize;
379 struct ipc_namespace *ipc_ns;
384 if (S_ISDIR(inode->i_mode))
387 ipc_ns = get_ns_from_inode(inode);
388 info = MQUEUE_I(inode);
389 spin_lock(&info->lock);
390 while ((msg = msg_get(info)) != NULL)
392 kfree(info->node_cache);
393 spin_unlock(&info->lock);
395 /* Total amount of bytes accounted for the mqueue */
396 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
397 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
398 sizeof(struct posix_msg_tree_node);
400 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
401 info->attr.mq_msgsize);
406 user->mq_bytes -= mq_bytes;
408 * get_ns_from_inode() ensures that the
409 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
410 * to which we now hold a reference, or it is NULL.
411 * We can't put it here under mq_lock, though.
414 ipc_ns->mq_queues_count--;
415 spin_unlock(&mq_lock);
422 static int mqueue_create(struct inode *dir, struct dentry *dentry,
423 umode_t mode, bool excl)
426 struct mq_attr *attr = dentry->d_fsdata;
428 struct ipc_namespace *ipc_ns;
431 ipc_ns = __get_ns_from_inode(dir);
436 if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
437 (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
438 !capable(CAP_SYS_RESOURCE))) {
442 ipc_ns->mq_queues_count++;
443 spin_unlock(&mq_lock);
445 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
447 error = PTR_ERR(inode);
449 ipc_ns->mq_queues_count--;
454 dir->i_size += DIRENT_SIZE;
455 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
457 d_instantiate(dentry, inode);
461 spin_unlock(&mq_lock);
467 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
469 struct inode *inode = dentry->d_inode;
471 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
472 dir->i_size -= DIRENT_SIZE;
479 * This is routine for system read from queue file.
480 * To avoid mess with doing here some sort of mq_receive we allow
481 * to read only queue size & notification info (the only values
482 * that are interesting from user point of view and aren't accessible
483 * through std routines)
485 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
486 size_t count, loff_t *off)
488 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
489 char buffer[FILENT_SIZE];
492 spin_lock(&info->lock);
493 snprintf(buffer, sizeof(buffer),
494 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
496 info->notify_owner ? info->notify.sigev_notify : 0,
497 (info->notify_owner &&
498 info->notify.sigev_notify == SIGEV_SIGNAL) ?
499 info->notify.sigev_signo : 0,
500 pid_vnr(info->notify_owner));
501 spin_unlock(&info->lock);
502 buffer[sizeof(buffer)-1] = '\0';
504 ret = simple_read_from_buffer(u_data, count, off, buffer,
509 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
513 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
515 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
517 spin_lock(&info->lock);
518 if (task_tgid(current) == info->notify_owner)
519 remove_notification(info);
521 spin_unlock(&info->lock);
525 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
527 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
530 poll_wait(filp, &info->wait_q, poll_tab);
532 spin_lock(&info->lock);
533 if (info->attr.mq_curmsgs)
534 retval = POLLIN | POLLRDNORM;
536 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
537 retval |= POLLOUT | POLLWRNORM;
538 spin_unlock(&info->lock);
543 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
544 static void wq_add(struct mqueue_inode_info *info, int sr,
545 struct ext_wait_queue *ewp)
547 struct ext_wait_queue *walk;
551 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
552 if (walk->task->static_prio <= current->static_prio) {
553 list_add_tail(&ewp->list, &walk->list);
557 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
561 * Puts current task to sleep. Caller must hold queue lock. After return
565 static int wq_sleep(struct mqueue_inode_info *info, int sr,
566 ktime_t *timeout, struct ext_wait_queue *ewp)
571 wq_add(info, sr, ewp);
574 set_current_state(TASK_INTERRUPTIBLE);
576 spin_unlock(&info->lock);
577 time = schedule_hrtimeout_range_clock(timeout, 0,
578 HRTIMER_MODE_ABS, CLOCK_REALTIME);
580 while (ewp->state == STATE_PENDING)
583 if (ewp->state == STATE_READY) {
587 spin_lock(&info->lock);
588 if (ewp->state == STATE_READY) {
592 if (signal_pending(current)) {
593 retval = -ERESTARTSYS;
601 list_del(&ewp->list);
603 spin_unlock(&info->lock);
609 * Returns waiting task that should be serviced first or NULL if none exists
611 static struct ext_wait_queue *wq_get_first_waiter(
612 struct mqueue_inode_info *info, int sr)
614 struct list_head *ptr;
616 ptr = info->e_wait_q[sr].list.prev;
617 if (ptr == &info->e_wait_q[sr].list)
619 return list_entry(ptr, struct ext_wait_queue, list);
623 static inline void set_cookie(struct sk_buff *skb, char code)
625 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
629 * The next function is only to split too long sys_mq_timedsend
631 static void __do_notify(struct mqueue_inode_info *info)
634 * invoked when there is registered process and there isn't process
635 * waiting synchronously for message AND state of queue changed from
636 * empty to not empty. Here we are sure that no one is waiting
638 if (info->notify_owner &&
639 info->attr.mq_curmsgs == 1) {
640 struct siginfo sig_i;
641 switch (info->notify.sigev_notify) {
647 sig_i.si_signo = info->notify.sigev_signo;
649 sig_i.si_code = SI_MESGQ;
650 sig_i.si_value = info->notify.sigev_value;
651 /* map current pid/uid into info->owner's namespaces */
653 sig_i.si_pid = task_tgid_nr_ns(current,
654 ns_of_pid(info->notify_owner));
655 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
658 kill_pid_info(info->notify.sigev_signo,
659 &sig_i, info->notify_owner);
662 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
663 netlink_sendskb(info->notify_sock, info->notify_cookie);
666 /* after notification unregisters process */
667 put_pid(info->notify_owner);
668 put_user_ns(info->notify_user_ns);
669 info->notify_owner = NULL;
670 info->notify_user_ns = NULL;
672 wake_up(&info->wait_q);
675 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
676 ktime_t *expires, struct timespec *ts)
678 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
680 if (!timespec_valid(ts))
683 *expires = timespec_to_ktime(*ts);
687 static void remove_notification(struct mqueue_inode_info *info)
689 if (info->notify_owner != NULL &&
690 info->notify.sigev_notify == SIGEV_THREAD) {
691 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
692 netlink_sendskb(info->notify_sock, info->notify_cookie);
694 put_pid(info->notify_owner);
695 put_user_ns(info->notify_user_ns);
696 info->notify_owner = NULL;
697 info->notify_user_ns = NULL;
700 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
703 unsigned long total_size;
705 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
707 if (capable(CAP_SYS_RESOURCE)) {
708 if (attr->mq_maxmsg > HARD_MSGMAX ||
709 attr->mq_msgsize > HARD_MSGSIZEMAX)
712 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
716 /* check for overflow */
717 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
719 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 sizeof(struct posix_msg_tree_node);
722 total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 if (total_size + mq_treesize < total_size)
729 * Invoked when creating a new queue via sys_mq_open
731 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 struct path *path, int oflag, umode_t mode,
733 struct mq_attr *attr)
735 const struct cred *cred = current_cred();
739 ret = mq_attr_ok(ipc_ns, attr);
742 /* store for use during create */
743 path->dentry->d_fsdata = attr;
745 struct mq_attr def_attr;
747 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 ipc_ns->mq_msg_default);
749 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 ipc_ns->mq_msgsize_default);
751 ret = mq_attr_ok(ipc_ns, &def_attr);
756 mode &= ~current_umask();
757 ret = vfs_create(dir, path->dentry, mode, true);
758 path->dentry->d_fsdata = NULL;
761 return dentry_open(path, oflag, cred);
764 /* Opens existing queue */
765 static struct file *do_open(struct path *path, int oflag)
767 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 MAY_READ | MAY_WRITE };
770 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 return ERR_PTR(-EINVAL);
772 acc = oflag2acc[oflag & O_ACCMODE];
773 if (inode_permission(path->dentry->d_inode, acc))
774 return ERR_PTR(-EACCES);
775 return dentry_open(path, oflag, current_cred());
778 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
779 struct mq_attr __user *, u_attr)
783 struct filename *name;
786 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
787 struct vfsmount *mnt = ipc_ns->mq_mnt;
788 struct dentry *root = mnt->mnt_root;
791 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
794 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
796 if (IS_ERR(name = getname(u_name)))
797 return PTR_ERR(name);
799 fd = get_unused_fd_flags(O_CLOEXEC);
803 ro = mnt_want_write(mnt); /* we'll drop it in any case */
805 mutex_lock(&root->d_inode->i_mutex);
806 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
807 if (IS_ERR(path.dentry)) {
808 error = PTR_ERR(path.dentry);
811 path.mnt = mntget(mnt);
813 if (oflag & O_CREAT) {
814 if (path.dentry->d_inode) { /* entry already exists */
815 audit_inode(name, path.dentry, 0);
816 if (oflag & O_EXCL) {
820 filp = do_open(&path, oflag);
826 audit_inode_parent_hidden(name, root);
827 filp = do_create(ipc_ns, root->d_inode,
829 u_attr ? &attr : NULL);
832 if (!path.dentry->d_inode) {
836 audit_inode(name, path.dentry, 0);
837 filp = do_open(&path, oflag);
841 fd_install(fd, filp);
843 error = PTR_ERR(filp);
851 mutex_unlock(&root->d_inode->i_mutex);
859 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
862 struct filename *name;
863 struct dentry *dentry;
864 struct inode *inode = NULL;
865 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
866 struct vfsmount *mnt = ipc_ns->mq_mnt;
868 name = getname(u_name);
870 return PTR_ERR(name);
872 audit_inode_parent_hidden(name, mnt->mnt_root);
873 err = mnt_want_write(mnt);
876 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
877 dentry = lookup_one_len(name->name, mnt->mnt_root,
879 if (IS_ERR(dentry)) {
880 err = PTR_ERR(dentry);
884 inode = dentry->d_inode;
889 err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
894 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
904 /* Pipelined send and receive functions.
906 * If a receiver finds no waiting message, then it registers itself in the
907 * list of waiting receivers. A sender checks that list before adding the new
908 * message into the message array. If there is a waiting receiver, then it
909 * bypasses the message array and directly hands the message over to the
911 * The receiver accepts the message and returns without grabbing the queue
912 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
913 * are necessary. The same algorithm is used for sysv semaphores, see
914 * ipc/sem.c for more details.
916 * The same algorithm is used for senders.
919 /* pipelined_send() - send a message directly to the task waiting in
920 * sys_mq_timedreceive() (without inserting message into a queue).
922 static inline void pipelined_send(struct mqueue_inode_info *info,
923 struct msg_msg *message,
924 struct ext_wait_queue *receiver)
926 receiver->msg = message;
927 list_del(&receiver->list);
928 receiver->state = STATE_PENDING;
929 wake_up_process(receiver->task);
931 receiver->state = STATE_READY;
934 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
935 * gets its message and put to the queue (we have one free place for sure). */
936 static inline void pipelined_receive(struct mqueue_inode_info *info)
938 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
942 wake_up_interruptible(&info->wait_q);
945 if (msg_insert(sender->msg, info))
947 list_del(&sender->list);
948 sender->state = STATE_PENDING;
949 wake_up_process(sender->task);
951 sender->state = STATE_READY;
954 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
955 size_t, msg_len, unsigned int, msg_prio,
956 const struct timespec __user *, u_abs_timeout)
960 struct ext_wait_queue wait;
961 struct ext_wait_queue *receiver;
962 struct msg_msg *msg_ptr;
963 struct mqueue_inode_info *info;
964 ktime_t expires, *timeout = NULL;
966 struct posix_msg_tree_node *new_leaf = NULL;
970 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
976 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
979 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
982 if (unlikely(!f.file)) {
987 inode = file_inode(f.file);
988 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
992 info = MQUEUE_I(inode);
993 audit_inode(NULL, f.file->f_path.dentry, 0);
995 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1000 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1005 /* First try to allocate memory, before doing anything with
1006 * existing queues. */
1007 msg_ptr = load_msg(u_msg_ptr, msg_len);
1008 if (IS_ERR(msg_ptr)) {
1009 ret = PTR_ERR(msg_ptr);
1012 msg_ptr->m_ts = msg_len;
1013 msg_ptr->m_type = msg_prio;
1016 * msg_insert really wants us to have a valid, spare node struct so
1017 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1018 * fall back to that if necessary.
1020 if (!info->node_cache)
1021 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1023 spin_lock(&info->lock);
1025 if (!info->node_cache && new_leaf) {
1026 /* Save our speculative allocation into the cache */
1027 INIT_LIST_HEAD(&new_leaf->msg_list);
1028 info->node_cache = new_leaf;
1029 info->qsize += sizeof(*new_leaf);
1035 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1036 if (f.file->f_flags & O_NONBLOCK) {
1039 wait.task = current;
1040 wait.msg = (void *) msg_ptr;
1041 wait.state = STATE_NONE;
1042 ret = wq_sleep(info, SEND, timeout, &wait);
1044 * wq_sleep must be called with info->lock held, and
1045 * returns with the lock released
1050 receiver = wq_get_first_waiter(info, RECV);
1052 pipelined_send(info, msg_ptr, receiver);
1054 /* adds message to the queue */
1055 ret = msg_insert(msg_ptr, info);
1060 inode->i_atime = inode->i_mtime = inode->i_ctime =
1064 spin_unlock(&info->lock);
1074 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1075 size_t, msg_len, unsigned int __user *, u_msg_prio,
1076 const struct timespec __user *, u_abs_timeout)
1079 struct msg_msg *msg_ptr;
1081 struct inode *inode;
1082 struct mqueue_inode_info *info;
1083 struct ext_wait_queue wait;
1084 ktime_t expires, *timeout = NULL;
1086 struct posix_msg_tree_node *new_leaf = NULL;
1088 if (u_abs_timeout) {
1089 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1095 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1098 if (unlikely(!f.file)) {
1103 inode = file_inode(f.file);
1104 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1108 info = MQUEUE_I(inode);
1109 audit_inode(NULL, f.file->f_path.dentry, 0);
1111 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1116 /* checks if buffer is big enough */
1117 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1123 * msg_insert really wants us to have a valid, spare node struct so
1124 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125 * fall back to that if necessary.
1127 if (!info->node_cache)
1128 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1130 spin_lock(&info->lock);
1132 if (!info->node_cache && new_leaf) {
1133 /* Save our speculative allocation into the cache */
1134 INIT_LIST_HEAD(&new_leaf->msg_list);
1135 info->node_cache = new_leaf;
1136 info->qsize += sizeof(*new_leaf);
1141 if (info->attr.mq_curmsgs == 0) {
1142 if (f.file->f_flags & O_NONBLOCK) {
1143 spin_unlock(&info->lock);
1146 wait.task = current;
1147 wait.state = STATE_NONE;
1148 ret = wq_sleep(info, RECV, timeout, &wait);
1152 msg_ptr = msg_get(info);
1154 inode->i_atime = inode->i_mtime = inode->i_ctime =
1157 /* There is now free space in queue. */
1158 pipelined_receive(info);
1159 spin_unlock(&info->lock);
1163 ret = msg_ptr->m_ts;
1165 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1166 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1178 * Notes: the case when user wants us to deregister (with NULL as pointer)
1179 * and he isn't currently owner of notification, will be silently discarded.
1180 * It isn't explicitly defined in the POSIX.
1182 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1183 const struct sigevent __user *, u_notification)
1188 struct inode *inode;
1189 struct sigevent notification;
1190 struct mqueue_inode_info *info;
1193 if (u_notification) {
1194 if (copy_from_user(¬ification, u_notification,
1195 sizeof(struct sigevent)))
1199 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1203 if (u_notification != NULL) {
1204 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1205 notification.sigev_notify != SIGEV_SIGNAL &&
1206 notification.sigev_notify != SIGEV_THREAD))
1208 if (notification.sigev_notify == SIGEV_SIGNAL &&
1209 !valid_signal(notification.sigev_signo)) {
1212 if (notification.sigev_notify == SIGEV_THREAD) {
1215 /* create the notify skb */
1216 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1221 if (copy_from_user(nc->data,
1222 notification.sigev_value.sival_ptr,
1223 NOTIFY_COOKIE_LEN)) {
1228 /* TODO: add a header? */
1229 skb_put(nc, NOTIFY_COOKIE_LEN);
1230 /* and attach it to the socket */
1232 f = fdget(notification.sigev_signo);
1237 sock = netlink_getsockbyfilp(f.file);
1240 ret = PTR_ERR(sock);
1245 timeo = MAX_SCHEDULE_TIMEOUT;
1246 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1263 inode = file_inode(f.file);
1264 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1268 info = MQUEUE_I(inode);
1271 spin_lock(&info->lock);
1272 if (u_notification == NULL) {
1273 if (info->notify_owner == task_tgid(current)) {
1274 remove_notification(info);
1275 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1277 } else if (info->notify_owner != NULL) {
1280 switch (notification.sigev_notify) {
1282 info->notify.sigev_notify = SIGEV_NONE;
1285 info->notify_sock = sock;
1286 info->notify_cookie = nc;
1289 info->notify.sigev_notify = SIGEV_THREAD;
1292 info->notify.sigev_signo = notification.sigev_signo;
1293 info->notify.sigev_value = notification.sigev_value;
1294 info->notify.sigev_notify = SIGEV_SIGNAL;
1298 info->notify_owner = get_pid(task_tgid(current));
1299 info->notify_user_ns = get_user_ns(current_user_ns());
1300 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1302 spin_unlock(&info->lock);
1307 netlink_detachskb(sock, nc);
1314 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1315 const struct mq_attr __user *, u_mqstat,
1316 struct mq_attr __user *, u_omqstat)
1319 struct mq_attr mqstat, omqstat;
1321 struct inode *inode;
1322 struct mqueue_inode_info *info;
1324 if (u_mqstat != NULL) {
1325 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1327 if (mqstat.mq_flags & (~O_NONBLOCK))
1337 inode = file_inode(f.file);
1338 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1342 info = MQUEUE_I(inode);
1344 spin_lock(&info->lock);
1346 omqstat = info->attr;
1347 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1349 audit_mq_getsetattr(mqdes, &mqstat);
1350 spin_lock(&f.file->f_lock);
1351 if (mqstat.mq_flags & O_NONBLOCK)
1352 f.file->f_flags |= O_NONBLOCK;
1354 f.file->f_flags &= ~O_NONBLOCK;
1355 spin_unlock(&f.file->f_lock);
1357 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1360 spin_unlock(&info->lock);
1363 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1364 sizeof(struct mq_attr)))
1373 static const struct inode_operations mqueue_dir_inode_operations = {
1374 .lookup = simple_lookup,
1375 .create = mqueue_create,
1376 .unlink = mqueue_unlink,
1379 static const struct file_operations mqueue_file_operations = {
1380 .flush = mqueue_flush_file,
1381 .poll = mqueue_poll_file,
1382 .read = mqueue_read_file,
1383 .llseek = default_llseek,
1386 static const struct super_operations mqueue_super_ops = {
1387 .alloc_inode = mqueue_alloc_inode,
1388 .destroy_inode = mqueue_destroy_inode,
1389 .evict_inode = mqueue_evict_inode,
1390 .statfs = simple_statfs,
1393 static struct file_system_type mqueue_fs_type = {
1395 .mount = mqueue_mount,
1396 .kill_sb = kill_litter_super,
1397 .fs_flags = FS_USERNS_MOUNT,
1400 int mq_init_ns(struct ipc_namespace *ns)
1402 ns->mq_queues_count = 0;
1403 ns->mq_queues_max = DFLT_QUEUESMAX;
1404 ns->mq_msg_max = DFLT_MSGMAX;
1405 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1406 ns->mq_msg_default = DFLT_MSG;
1407 ns->mq_msgsize_default = DFLT_MSGSIZE;
1409 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1410 if (IS_ERR(ns->mq_mnt)) {
1411 int err = PTR_ERR(ns->mq_mnt);
1418 void mq_clear_sbinfo(struct ipc_namespace *ns)
1420 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1423 void mq_put_mnt(struct ipc_namespace *ns)
1425 kern_unmount(ns->mq_mnt);
1428 static int __init init_mqueue_fs(void)
1432 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1433 sizeof(struct mqueue_inode_info), 0,
1434 SLAB_HWCACHE_ALIGN, init_once);
1435 if (mqueue_inode_cachep == NULL)
1438 /* ignore failures - they are not fatal */
1439 mq_sysctl_table = mq_register_sysctl_table();
1441 error = register_filesystem(&mqueue_fs_type);
1445 spin_lock_init(&mq_lock);
1447 error = mq_init_ns(&init_ipc_ns);
1449 goto out_filesystem;
1454 unregister_filesystem(&mqueue_fs_type);
1456 if (mq_sysctl_table)
1457 unregister_sysctl_table(mq_sysctl_table);
1458 kmem_cache_destroy(mqueue_inode_cachep);
1462 __initcall(init_mqueue_fs);