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
52 struct posix_msg_tree_node {
53 struct rb_node rb_node;
54 struct list_head msg_list;
58 struct ext_wait_queue { /* queue of sleeping tasks */
59 struct task_struct *task;
60 struct list_head list;
61 struct msg_msg *msg; /* ptr of loaded message */
62 int state; /* one of STATE_* values */
65 struct mqueue_inode_info {
67 struct inode vfs_inode;
68 wait_queue_head_t wait_q;
70 struct rb_root msg_tree;
71 struct posix_msg_tree_node *node_cache;
74 struct sigevent notify;
75 struct pid *notify_owner;
76 struct user_namespace *notify_user_ns;
77 struct user_struct *user; /* user who created, for accounting */
78 struct sock *notify_sock;
79 struct sk_buff *notify_cookie;
81 /* for tasks waiting for free space and messages, respectively */
82 struct ext_wait_queue e_wait_q[2];
84 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
87 static const struct inode_operations mqueue_dir_inode_operations;
88 static const struct file_operations mqueue_file_operations;
89 static const struct super_operations mqueue_super_ops;
90 static void remove_notification(struct mqueue_inode_info *info);
92 static struct kmem_cache *mqueue_inode_cachep;
94 static struct ctl_table_header *mq_sysctl_table;
96 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102 * This routine should be called with the mq_lock held.
104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106 return get_ipc_ns(inode->i_sb->s_fs_info);
109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111 struct ipc_namespace *ns;
114 ns = __get_ns_from_inode(inode);
115 spin_unlock(&mq_lock);
119 /* Auxiliary functions to manipulate messages' list */
120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122 struct rb_node **p, *parent = NULL;
123 struct posix_msg_tree_node *leaf;
125 p = &info->msg_tree.rb_node;
128 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130 if (likely(leaf->priority == msg->m_type))
132 else if (msg->m_type < leaf->priority)
137 if (info->node_cache) {
138 leaf = info->node_cache;
139 info->node_cache = NULL;
141 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
144 INIT_LIST_HEAD(&leaf->msg_list);
146 leaf->priority = msg->m_type;
147 rb_link_node(&leaf->rb_node, parent, p);
148 rb_insert_color(&leaf->rb_node, &info->msg_tree);
150 info->attr.mq_curmsgs++;
151 info->qsize += msg->m_ts;
152 list_add_tail(&msg->m_list, &leaf->msg_list);
156 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
158 struct rb_node **p, *parent = NULL;
159 struct posix_msg_tree_node *leaf;
163 p = &info->msg_tree.rb_node;
167 * During insert, low priorities go to the left and high to the
168 * right. On receive, we want the highest priorities first, so
169 * walk all the way to the right.
174 if (info->attr.mq_curmsgs) {
175 pr_warn_once("Inconsistency in POSIX message queue, "
176 "no tree element, but supposedly messages "
178 info->attr.mq_curmsgs = 0;
182 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
183 if (unlikely(list_empty(&leaf->msg_list))) {
184 pr_warn_once("Inconsistency in POSIX message queue, "
185 "empty leaf node but we haven't implemented "
186 "lazy leaf delete!\n");
187 rb_erase(&leaf->rb_node, &info->msg_tree);
188 if (info->node_cache) {
191 info->node_cache = leaf;
195 msg = list_first_entry(&leaf->msg_list,
196 struct msg_msg, m_list);
197 list_del(&msg->m_list);
198 if (list_empty(&leaf->msg_list)) {
199 rb_erase(&leaf->rb_node, &info->msg_tree);
200 if (info->node_cache) {
203 info->node_cache = leaf;
207 info->attr.mq_curmsgs--;
208 info->qsize -= msg->m_ts;
212 static struct inode *mqueue_get_inode(struct super_block *sb,
213 struct ipc_namespace *ipc_ns, umode_t mode,
214 struct mq_attr *attr)
216 struct user_struct *u = current_user();
220 inode = new_inode(sb);
224 inode->i_ino = get_next_ino();
225 inode->i_mode = mode;
226 inode->i_uid = current_fsuid();
227 inode->i_gid = current_fsgid();
228 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
231 struct mqueue_inode_info *info;
232 unsigned long mq_bytes, mq_treesize;
234 inode->i_fop = &mqueue_file_operations;
235 inode->i_size = FILENT_SIZE;
236 /* mqueue specific info */
237 info = MQUEUE_I(inode);
238 spin_lock_init(&info->lock);
239 init_waitqueue_head(&info->wait_q);
240 INIT_LIST_HEAD(&info->e_wait_q[0].list);
241 INIT_LIST_HEAD(&info->e_wait_q[1].list);
242 info->notify_owner = NULL;
243 info->notify_user_ns = NULL;
245 info->user = NULL; /* set when all is ok */
246 info->msg_tree = RB_ROOT;
247 info->node_cache = NULL;
248 memset(&info->attr, 0, sizeof(info->attr));
249 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
250 ipc_ns->mq_msg_default);
251 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
252 ipc_ns->mq_msgsize_default);
254 info->attr.mq_maxmsg = attr->mq_maxmsg;
255 info->attr.mq_msgsize = attr->mq_msgsize;
258 * We used to allocate a static array of pointers and account
259 * the size of that array as well as one msg_msg struct per
260 * possible message into the queue size. That's no longer
261 * accurate as the queue is now an rbtree and will grow and
262 * shrink depending on usage patterns. We can, however, still
263 * account one msg_msg struct per message, but the nodes are
264 * allocated depending on priority usage, and most programs
265 * only use one, or a handful, of priorities. However, since
266 * this is pinned memory, we need to assume worst case, so
267 * that means the min(mq_maxmsg, max_priorities) * struct
268 * posix_msg_tree_node.
270 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
271 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
272 sizeof(struct posix_msg_tree_node);
274 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 info->attr.mq_msgsize);
278 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
279 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
280 spin_unlock(&mq_lock);
281 /* mqueue_evict_inode() releases info->messages */
285 u->mq_bytes += mq_bytes;
286 spin_unlock(&mq_lock);
289 info->user = get_uid(u);
290 } else if (S_ISDIR(mode)) {
292 /* Some things misbehave if size == 0 on a directory */
293 inode->i_size = 2 * DIRENT_SIZE;
294 inode->i_op = &mqueue_dir_inode_operations;
295 inode->i_fop = &simple_dir_operations;
305 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
308 struct ipc_namespace *ns = data;
310 sb->s_blocksize = PAGE_CACHE_SIZE;
311 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
312 sb->s_magic = MQUEUE_MAGIC;
313 sb->s_op = &mqueue_super_ops;
315 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317 return PTR_ERR(inode);
319 sb->s_root = d_make_root(inode);
325 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
326 int flags, const char *dev_name,
329 if (!(flags & MS_KERNMOUNT)) {
330 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
331 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
332 * over the ipc namespace.
334 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
335 return ERR_PTR(-EPERM);
339 return mount_ns(fs_type, flags, data, mqueue_fill_super);
342 static void init_once(void *foo)
344 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
346 inode_init_once(&p->vfs_inode);
349 static struct inode *mqueue_alloc_inode(struct super_block *sb)
351 struct mqueue_inode_info *ei;
353 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
356 return &ei->vfs_inode;
359 static void mqueue_i_callback(struct rcu_head *head)
361 struct inode *inode = container_of(head, struct inode, i_rcu);
362 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
365 static void mqueue_destroy_inode(struct inode *inode)
367 call_rcu(&inode->i_rcu, mqueue_i_callback);
370 static void mqueue_evict_inode(struct inode *inode)
372 struct mqueue_inode_info *info;
373 struct user_struct *user;
374 struct ipc_namespace *ipc_ns;
375 struct msg_msg *msg, *nmsg;
380 if (S_ISDIR(inode->i_mode))
383 ipc_ns = get_ns_from_inode(inode);
384 info = MQUEUE_I(inode);
385 spin_lock(&info->lock);
386 while ((msg = msg_get(info)) != NULL)
387 list_add_tail(&msg->m_list, &tmp_msg);
388 kfree(info->node_cache);
389 spin_unlock(&info->lock);
391 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
392 list_del(&msg->m_list);
398 unsigned long mq_bytes, mq_treesize;
400 /* Total amount of bytes accounted for the mqueue */
401 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
402 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
403 sizeof(struct posix_msg_tree_node);
405 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
406 info->attr.mq_msgsize);
409 user->mq_bytes -= mq_bytes;
411 * get_ns_from_inode() ensures that the
412 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
413 * to which we now hold a reference, or it is NULL.
414 * We can't put it here under mq_lock, though.
417 ipc_ns->mq_queues_count--;
418 spin_unlock(&mq_lock);
425 static int mqueue_create(struct inode *dir, struct dentry *dentry,
426 umode_t mode, bool excl)
429 struct mq_attr *attr = dentry->d_fsdata;
431 struct ipc_namespace *ipc_ns;
434 ipc_ns = __get_ns_from_inode(dir);
440 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
441 !capable(CAP_SYS_RESOURCE)) {
445 ipc_ns->mq_queues_count++;
446 spin_unlock(&mq_lock);
448 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
450 error = PTR_ERR(inode);
452 ipc_ns->mq_queues_count--;
457 dir->i_size += DIRENT_SIZE;
458 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
460 d_instantiate(dentry, inode);
464 spin_unlock(&mq_lock);
470 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
472 struct inode *inode = d_inode(dentry);
474 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
475 dir->i_size -= DIRENT_SIZE;
482 * This is routine for system read from queue file.
483 * To avoid mess with doing here some sort of mq_receive we allow
484 * to read only queue size & notification info (the only values
485 * that are interesting from user point of view and aren't accessible
486 * through std routines)
488 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
489 size_t count, loff_t *off)
491 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
492 char buffer[FILENT_SIZE];
495 spin_lock(&info->lock);
496 snprintf(buffer, sizeof(buffer),
497 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
499 info->notify_owner ? info->notify.sigev_notify : 0,
500 (info->notify_owner &&
501 info->notify.sigev_notify == SIGEV_SIGNAL) ?
502 info->notify.sigev_signo : 0,
503 pid_vnr(info->notify_owner));
504 spin_unlock(&info->lock);
505 buffer[sizeof(buffer)-1] = '\0';
507 ret = simple_read_from_buffer(u_data, count, off, buffer,
512 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
516 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
518 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
520 spin_lock(&info->lock);
521 if (task_tgid(current) == info->notify_owner)
522 remove_notification(info);
524 spin_unlock(&info->lock);
528 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
530 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
533 poll_wait(filp, &info->wait_q, poll_tab);
535 spin_lock(&info->lock);
536 if (info->attr.mq_curmsgs)
537 retval = POLLIN | POLLRDNORM;
539 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
540 retval |= POLLOUT | POLLWRNORM;
541 spin_unlock(&info->lock);
546 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
547 static void wq_add(struct mqueue_inode_info *info, int sr,
548 struct ext_wait_queue *ewp)
550 struct ext_wait_queue *walk;
554 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
555 if (walk->task->static_prio <= current->static_prio) {
556 list_add_tail(&ewp->list, &walk->list);
560 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
564 * Puts current task to sleep. Caller must hold queue lock. After return
568 static int wq_sleep(struct mqueue_inode_info *info, int sr,
569 ktime_t *timeout, struct ext_wait_queue *ewp)
574 wq_add(info, sr, ewp);
577 __set_current_state(TASK_INTERRUPTIBLE);
579 spin_unlock(&info->lock);
580 time = schedule_hrtimeout_range_clock(timeout, 0,
581 HRTIMER_MODE_ABS, CLOCK_REALTIME);
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(d_inode(path->dentry), 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(&d_inode(root)->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 (d_really_is_positive(path.dentry)) { /* 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, d_inode(root),
829 u_attr ? &attr : NULL);
832 if (d_really_is_negative(path.dentry)) {
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(&d_inode(root)->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(&d_inode(mnt->mnt_root)->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 = d_inode(dentry);
889 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
894 mutex_unlock(&d_inode(mnt->mnt_root)->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
910 * receiver. The receiver accepts the message and returns without grabbing the
913 * - Set pointer to message.
914 * - Queue the receiver task for later wakeup (without the info->lock).
915 * - Update its state to STATE_READY. Now the receiver can continue.
916 * - Wake up the process after the lock is dropped. Should the process wake up
917 * before this wakeup (due to a timeout or a signal) it will either see
918 * STATE_READY and continue or acquire the lock to check the state again.
920 * The same algorithm is used for senders.
923 /* pipelined_send() - send a message directly to the task waiting in
924 * sys_mq_timedreceive() (without inserting message into a queue).
926 static inline void pipelined_send(struct wake_q_head *wake_q,
927 struct mqueue_inode_info *info,
928 struct msg_msg *message,
929 struct ext_wait_queue *receiver)
931 receiver->msg = message;
932 list_del(&receiver->list);
933 wake_q_add(wake_q, receiver->task);
935 * Rely on the implicit cmpxchg barrier from wake_q_add such
936 * that we can ensure that updating receiver->state is the last
937 * write operation: As once set, the receiver can continue,
938 * and if we don't have the reference count from the wake_q,
939 * yet, at that point we can later have a use-after-free
940 * condition and bogus wakeup.
942 receiver->state = STATE_READY;
945 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
946 * gets its message and put to the queue (we have one free place for sure). */
947 static inline void pipelined_receive(struct wake_q_head *wake_q,
948 struct mqueue_inode_info *info)
950 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
954 wake_up_interruptible(&info->wait_q);
957 if (msg_insert(sender->msg, info))
960 list_del(&sender->list);
961 wake_q_add(wake_q, sender->task);
962 sender->state = STATE_READY;
965 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
966 size_t, msg_len, unsigned int, msg_prio,
967 const struct timespec __user *, u_abs_timeout)
971 struct ext_wait_queue wait;
972 struct ext_wait_queue *receiver;
973 struct msg_msg *msg_ptr;
974 struct mqueue_inode_info *info;
975 ktime_t expires, *timeout = NULL;
977 struct posix_msg_tree_node *new_leaf = NULL;
982 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
988 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
991 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
994 if (unlikely(!f.file)) {
999 inode = file_inode(f.file);
1000 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1004 info = MQUEUE_I(inode);
1007 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1012 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1017 /* First try to allocate memory, before doing anything with
1018 * existing queues. */
1019 msg_ptr = load_msg(u_msg_ptr, msg_len);
1020 if (IS_ERR(msg_ptr)) {
1021 ret = PTR_ERR(msg_ptr);
1024 msg_ptr->m_ts = msg_len;
1025 msg_ptr->m_type = msg_prio;
1028 * msg_insert really wants us to have a valid, spare node struct so
1029 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1030 * fall back to that if necessary.
1032 if (!info->node_cache)
1033 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1035 spin_lock(&info->lock);
1037 if (!info->node_cache && new_leaf) {
1038 /* Save our speculative allocation into the cache */
1039 INIT_LIST_HEAD(&new_leaf->msg_list);
1040 info->node_cache = new_leaf;
1046 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1047 if (f.file->f_flags & O_NONBLOCK) {
1050 wait.task = current;
1051 wait.msg = (void *) msg_ptr;
1052 wait.state = STATE_NONE;
1053 ret = wq_sleep(info, SEND, timeout, &wait);
1055 * wq_sleep must be called with info->lock held, and
1056 * returns with the lock released
1061 receiver = wq_get_first_waiter(info, RECV);
1063 pipelined_send(&wake_q, info, msg_ptr, receiver);
1065 /* adds message to the queue */
1066 ret = msg_insert(msg_ptr, info);
1071 inode->i_atime = inode->i_mtime = inode->i_ctime =
1075 spin_unlock(&info->lock);
1086 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1087 size_t, msg_len, unsigned int __user *, u_msg_prio,
1088 const struct timespec __user *, u_abs_timeout)
1091 struct msg_msg *msg_ptr;
1093 struct inode *inode;
1094 struct mqueue_inode_info *info;
1095 struct ext_wait_queue wait;
1096 ktime_t expires, *timeout = NULL;
1098 struct posix_msg_tree_node *new_leaf = NULL;
1100 if (u_abs_timeout) {
1101 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1107 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1110 if (unlikely(!f.file)) {
1115 inode = file_inode(f.file);
1116 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1120 info = MQUEUE_I(inode);
1123 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1128 /* checks if buffer is big enough */
1129 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1135 * msg_insert really wants us to have a valid, spare node struct so
1136 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1137 * fall back to that if necessary.
1139 if (!info->node_cache)
1140 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1142 spin_lock(&info->lock);
1144 if (!info->node_cache && new_leaf) {
1145 /* Save our speculative allocation into the cache */
1146 INIT_LIST_HEAD(&new_leaf->msg_list);
1147 info->node_cache = new_leaf;
1152 if (info->attr.mq_curmsgs == 0) {
1153 if (f.file->f_flags & O_NONBLOCK) {
1154 spin_unlock(&info->lock);
1157 wait.task = current;
1158 wait.state = STATE_NONE;
1159 ret = wq_sleep(info, RECV, timeout, &wait);
1165 msg_ptr = msg_get(info);
1167 inode->i_atime = inode->i_mtime = inode->i_ctime =
1170 /* There is now free space in queue. */
1171 pipelined_receive(&wake_q, info);
1172 spin_unlock(&info->lock);
1177 ret = msg_ptr->m_ts;
1179 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1180 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1192 * Notes: the case when user wants us to deregister (with NULL as pointer)
1193 * and he isn't currently owner of notification, will be silently discarded.
1194 * It isn't explicitly defined in the POSIX.
1196 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1197 const struct sigevent __user *, u_notification)
1202 struct inode *inode;
1203 struct sigevent notification;
1204 struct mqueue_inode_info *info;
1207 if (u_notification) {
1208 if (copy_from_user(¬ification, u_notification,
1209 sizeof(struct sigevent)))
1213 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1217 if (u_notification != NULL) {
1218 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1219 notification.sigev_notify != SIGEV_SIGNAL &&
1220 notification.sigev_notify != SIGEV_THREAD))
1222 if (notification.sigev_notify == SIGEV_SIGNAL &&
1223 !valid_signal(notification.sigev_signo)) {
1226 if (notification.sigev_notify == SIGEV_THREAD) {
1229 /* create the notify skb */
1230 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1235 if (copy_from_user(nc->data,
1236 notification.sigev_value.sival_ptr,
1237 NOTIFY_COOKIE_LEN)) {
1242 /* TODO: add a header? */
1243 skb_put(nc, NOTIFY_COOKIE_LEN);
1244 /* and attach it to the socket */
1246 f = fdget(notification.sigev_signo);
1251 sock = netlink_getsockbyfilp(f.file);
1254 ret = PTR_ERR(sock);
1259 timeo = MAX_SCHEDULE_TIMEOUT;
1260 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1279 inode = file_inode(f.file);
1280 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1284 info = MQUEUE_I(inode);
1287 spin_lock(&info->lock);
1288 if (u_notification == NULL) {
1289 if (info->notify_owner == task_tgid(current)) {
1290 remove_notification(info);
1291 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1293 } else if (info->notify_owner != NULL) {
1296 switch (notification.sigev_notify) {
1298 info->notify.sigev_notify = SIGEV_NONE;
1301 info->notify_sock = sock;
1302 info->notify_cookie = nc;
1305 info->notify.sigev_notify = SIGEV_THREAD;
1308 info->notify.sigev_signo = notification.sigev_signo;
1309 info->notify.sigev_value = notification.sigev_value;
1310 info->notify.sigev_notify = SIGEV_SIGNAL;
1314 info->notify_owner = get_pid(task_tgid(current));
1315 info->notify_user_ns = get_user_ns(current_user_ns());
1316 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1318 spin_unlock(&info->lock);
1323 netlink_detachskb(sock, nc);
1330 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1331 const struct mq_attr __user *, u_mqstat,
1332 struct mq_attr __user *, u_omqstat)
1335 struct mq_attr mqstat, omqstat;
1337 struct inode *inode;
1338 struct mqueue_inode_info *info;
1340 if (u_mqstat != NULL) {
1341 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1343 if (mqstat.mq_flags & (~O_NONBLOCK))
1353 inode = file_inode(f.file);
1354 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1358 info = MQUEUE_I(inode);
1360 spin_lock(&info->lock);
1362 omqstat = info->attr;
1363 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1365 audit_mq_getsetattr(mqdes, &mqstat);
1366 spin_lock(&f.file->f_lock);
1367 if (mqstat.mq_flags & O_NONBLOCK)
1368 f.file->f_flags |= O_NONBLOCK;
1370 f.file->f_flags &= ~O_NONBLOCK;
1371 spin_unlock(&f.file->f_lock);
1373 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1376 spin_unlock(&info->lock);
1379 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1380 sizeof(struct mq_attr)))
1389 static const struct inode_operations mqueue_dir_inode_operations = {
1390 .lookup = simple_lookup,
1391 .create = mqueue_create,
1392 .unlink = mqueue_unlink,
1395 static const struct file_operations mqueue_file_operations = {
1396 .flush = mqueue_flush_file,
1397 .poll = mqueue_poll_file,
1398 .read = mqueue_read_file,
1399 .llseek = default_llseek,
1402 static const struct super_operations mqueue_super_ops = {
1403 .alloc_inode = mqueue_alloc_inode,
1404 .destroy_inode = mqueue_destroy_inode,
1405 .evict_inode = mqueue_evict_inode,
1406 .statfs = simple_statfs,
1409 static struct file_system_type mqueue_fs_type = {
1411 .mount = mqueue_mount,
1412 .kill_sb = kill_litter_super,
1413 .fs_flags = FS_USERNS_MOUNT,
1416 int mq_init_ns(struct ipc_namespace *ns)
1418 ns->mq_queues_count = 0;
1419 ns->mq_queues_max = DFLT_QUEUESMAX;
1420 ns->mq_msg_max = DFLT_MSGMAX;
1421 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1422 ns->mq_msg_default = DFLT_MSG;
1423 ns->mq_msgsize_default = DFLT_MSGSIZE;
1425 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1426 if (IS_ERR(ns->mq_mnt)) {
1427 int err = PTR_ERR(ns->mq_mnt);
1434 void mq_clear_sbinfo(struct ipc_namespace *ns)
1436 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1439 void mq_put_mnt(struct ipc_namespace *ns)
1441 kern_unmount(ns->mq_mnt);
1444 static int __init init_mqueue_fs(void)
1448 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1449 sizeof(struct mqueue_inode_info), 0,
1450 SLAB_HWCACHE_ALIGN, init_once);
1451 if (mqueue_inode_cachep == NULL)
1454 /* ignore failures - they are not fatal */
1455 mq_sysctl_table = mq_register_sysctl_table();
1457 error = register_filesystem(&mqueue_fs_type);
1461 spin_lock_init(&mq_lock);
1463 error = mq_init_ns(&init_ipc_ns);
1465 goto out_filesystem;
1470 unregister_filesystem(&mqueue_fs_type);
1472 if (mq_sysctl_table)
1473 unregister_sysctl_table(mq_sysctl_table);
1474 kmem_cache_destroy(mqueue_inode_cachep);
1478 device_initcall(init_mqueue_fs);